fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mpage.h>
  31 #include <linux/swap.h>
  32 #include <linux/writeback.h>
  33 #include <linux/statfs.h>
  34 #include <linux/compat.h>
  35 #include <linux/bit_spinlock.h>
  36 #include <linux/xattr.h>
  37 #include <linux/posix_acl.h>
  38 #include <linux/falloc.h>
  39 #include <linux/slab.h>
  40 #include <linux/ratelimit.h>
  41 #include <linux/mount.h>
  42 #include "compat.h"
  43 #include "ctree.h"
  44 #include "disk-io.h"
  45 #include "transaction.h"
  46 #include "btrfs_inode.h"
  47 #include "ioctl.h"
  48 #include "print-tree.h"
  49 #include "ordered-data.h"
  50 #include "xattr.h"
  51 #include "tree-log.h"
  52 #include "volumes.h"
  53 #include "compression.h"
  54 #include "locking.h"
  55 #include "free-space-cache.h"
  56 #include "inode-map.h"
  57
  58 struct btrfs_iget_args {
  59         u64 ino;
  60         struct btrfs_root *root;
  61 };
  62
  63 static const struct inode_operations btrfs_dir_inode_operations;
  64 static const struct inode_operations btrfs_symlink_inode_operations;
  65 static const struct inode_operations btrfs_dir_ro_inode_operations;
  66 static const struct inode_operations btrfs_special_inode_operations;
  67 static const struct inode_operations btrfs_file_inode_operations;
  68 static const struct address_space_operations btrfs_aops;
  69 static const struct address_space_operations btrfs_symlink_aops;
  70 static const struct file_operations btrfs_dir_file_operations;
  71 static struct extent_io_ops btrfs_extent_io_ops;
  72
  73 static struct kmem_cache *btrfs_inode_cachep;
  74 struct kmem_cache *btrfs_trans_handle_cachep;
  75 struct kmem_cache *btrfs_transaction_cachep;
  76 struct kmem_cache *btrfs_path_cachep;
  77 struct kmem_cache *btrfs_free_space_cachep;
  78
  79 #define S_SHIFT 12
  80 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  81         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  82         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  83         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  84         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  85         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  86         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  87         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  88 };
  89
  90 static int btrfs_setsize(struct inode *inode, loff_t newsize);
  91 static int btrfs_truncate(struct inode *inode);
  92 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
  93 static noinline int cow_file_range(struct inode *inode,
  94                                    struct page *locked_page,
  95                                    u64 start, u64 end, int *page_started,
  96                                    unsigned long *nr_written, int unlock);
  97 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
  98                                 struct btrfs_root *root, struct inode *inode);
  99
 100 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 101                                      struct inode *inode,  struct inode *dir,
 102                                      const struct qstr *qstr)
 103 {
 104         int err;
 105
 106         err = btrfs_init_acl(trans, inode, dir);
 107         if (!err)
 108                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 109         return err;
 110 }
 111
 112 /*
 113  * this does all the hard work for inserting an inline extent into
 114  * the btree.  The caller should have done a btrfs_drop_extents so that
 115  * no overlapping inline items exist in the btree
 116  */
 117 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 118                                 struct btrfs_root *root, struct inode *inode,
 119                                 u64 start, size_t size, size_t compressed_size,
 120                                 int compress_type,
 121                                 struct page **compressed_pages)
 122 {
 123         struct btrfs_key key;
 124         struct btrfs_path *path;
 125         struct extent_buffer *leaf;
 126         struct page *page = NULL;
 127         char *kaddr;
 128         unsigned long ptr;
 129         struct btrfs_file_extent_item *ei;
 130         int err = 0;
 131         int ret;
 132         size_t cur_size = size;
 133         size_t datasize;
 134         unsigned long offset;
 135
 136         if (compressed_size && compressed_pages)
 137                 cur_size = compressed_size;
 138
 139         path = btrfs_alloc_path();
 140         if (!path)
 141                 return -ENOMEM;
 142
 143         path->leave_spinning = 1;
 144
 145         key.objectid = btrfs_ino(inode);
 146         key.offset = start;
 147         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 148         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 149
 150         inode_add_bytes(inode, size);
 151         ret = btrfs_insert_empty_item(trans, root, path, &key,
 152                                       datasize);
 153         BUG_ON(ret);
 154         if (ret) {
 155                 err = ret;
 156                 goto fail;
 157         }
 158         leaf = path->nodes[0];
 159         ei = btrfs_item_ptr(leaf, path->slots[0],
 160                             struct btrfs_file_extent_item);
 161         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 162         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 163         btrfs_set_file_extent_encryption(leaf, ei, 0);
 164         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 165         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 166         ptr = btrfs_file_extent_inline_start(ei);
 167
 168         if (compress_type != BTRFS_COMPRESS_NONE) {
 169                 struct page *cpage;
 170                 int i = 0;
 171                 while (compressed_size > 0) {
 172                         cpage = compressed_pages[i];
 173                         cur_size = min_t(unsigned long, compressed_size,
 174                                        PAGE_CACHE_SIZE);
 175
 176                         kaddr = kmap_atomic(cpage, KM_USER0);
 177                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 178                         kunmap_atomic(kaddr, KM_USER0);
 179
 180                         i++;
 181                         ptr += cur_size;
 182                         compressed_size -= cur_size;
 183                 }
 184                 btrfs_set_file_extent_compression(leaf, ei,
 185                                                   compress_type);
 186         } else {
 187                 page = find_get_page(inode->i_mapping,
 188                                      start >> PAGE_CACHE_SHIFT);
 189                 btrfs_set_file_extent_compression(leaf, ei, 0);
 190                 kaddr = kmap_atomic(page, KM_USER0);
 191                 offset = start & (PAGE_CACHE_SIZE - 1);
 192                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 193                 kunmap_atomic(kaddr, KM_USER0);
 194                 page_cache_release(page);
 195         }
 196         btrfs_mark_buffer_dirty(leaf);
 197         btrfs_free_path(path);
 198
 199         /*
 200          * we're an inline extent, so nobody can
 201          * extend the file past i_size without locking
 202          * a page we already have locked.
 203          *
 204          * We must do any isize and inode updates
 205          * before we unlock the pages.  Otherwise we
 206          * could end up racing with unlink.
 207          */
 208         BTRFS_I(inode)->disk_i_size = inode->i_size;
 209         btrfs_update_inode(trans, root, inode);
 210
 211         return 0;
 212 fail:
 213         btrfs_free_path(path);
 214         return err;
 215 }
 216
 217
 218 /*
 219  * conditionally insert an inline extent into the file.  This
 220  * does the checks required to make sure the data is small enough
 221  * to fit as an inline extent.
 222  */
 223 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 224                                  struct btrfs_root *root,
 225                                  struct inode *inode, u64 start, u64 end,
 226                                  size_t compressed_size, int compress_type,
 227                                  struct page **compressed_pages)
 228 {
 229         u64 isize = i_size_read(inode);
 230         u64 actual_end = min(end + 1, isize);
 231         u64 inline_len = actual_end - start;
 232         u64 aligned_end = (end + root->sectorsize - 1) &
 233                         ~((u64)root->sectorsize - 1);
 234         u64 hint_byte;
 235         u64 data_len = inline_len;
 236         int ret;
 237
 238         if (compressed_size)
 239                 data_len = compressed_size;
 240
 241         if (start > 0 ||
 242             actual_end >= PAGE_CACHE_SIZE ||
 243             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 244             (!compressed_size &&
 245             (actual_end & (root->sectorsize - 1)) == 0) ||
 246             end + 1 < isize ||
 247             data_len > root->fs_info->max_inline) {
 248                 return 1;
 249         }
 250
 251         ret = btrfs_drop_extents(trans, inode, start, aligned_end,
 252                                  &hint_byte, 1);
 253         BUG_ON(ret);
 254
 255         if (isize > actual_end)
 256                 inline_len = min_t(u64, isize, actual_end);
 257         ret = insert_inline_extent(trans, root, inode, start,
 258                                    inline_len, compressed_size,
 259                                    compress_type, compressed_pages);
 260         BUG_ON(ret);
 261         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 262         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 263         return 0;
 264 }
 265
 266 struct async_extent {
 267         u64 start;
 268         u64 ram_size;
 269         u64 compressed_size;
 270         struct page **pages;
 271         unsigned long nr_pages;
 272         int compress_type;
 273         struct list_head list;
 274 };
 275
 276 struct async_cow {
 277         struct inode *inode;
 278         struct btrfs_root *root;
 279         struct page *locked_page;
 280         u64 start;
 281         u64 end;
 282         struct list_head extents;
 283         struct btrfs_work work;
 284 };
 285
 286 static noinline int add_async_extent(struct async_cow *cow,
 287                                      u64 start, u64 ram_size,
 288                                      u64 compressed_size,
 289                                      struct page **pages,
 290                                      unsigned long nr_pages,
 291                                      int compress_type)
 292 {
 293         struct async_extent *async_extent;
 294
 295         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 296         BUG_ON(!async_extent);
 297         async_extent->start = start;
 298         async_extent->ram_size = ram_size;
 299         async_extent->compressed_size = compressed_size;
 300         async_extent->pages = pages;
 301         async_extent->nr_pages = nr_pages;
 302         async_extent->compress_type = compress_type;
 303         list_add_tail(&async_extent->list, &cow->extents);
 304         return 0;
 305 }
 306
 307 /*
 308  * we create compressed extents in two phases.  The first
 309  * phase compresses a range of pages that have already been
 310  * locked (both pages and state bits are locked).
 311  *
 312  * This is done inside an ordered work queue, and the compression
 313  * is spread across many cpus.  The actual IO submission is step
 314  * two, and the ordered work queue takes care of making sure that
 315  * happens in the same order things were put onto the queue by
 316  * writepages and friends.
 317  *
 318  * If this code finds it can't get good compression, it puts an
 319  * entry onto the work queue to write the uncompressed bytes.  This
 320  * makes sure that both compressed inodes and uncompressed inodes
 321  * are written in the same order that pdflush sent them down.
 322  */
 323 static noinline int compress_file_range(struct inode *inode,
 324                                         struct page *locked_page,
 325                                         u64 start, u64 end,
 326                                         struct async_cow *async_cow,
 327                                         int *num_added)
 328 {
 329         struct btrfs_root *root = BTRFS_I(inode)->root;
 330         struct btrfs_trans_handle *trans;
 331         u64 num_bytes;
 332         u64 blocksize = root->sectorsize;
 333         u64 actual_end;
 334         u64 isize = i_size_read(inode);
 335         int ret = 0;
 336         struct page **pages = NULL;
 337         unsigned long nr_pages;
 338         unsigned long nr_pages_ret = 0;
 339         unsigned long total_compressed = 0;
 340         unsigned long total_in = 0;
 341         unsigned long max_compressed = 128 * 1024;
 342         unsigned long max_uncompressed = 128 * 1024;
 343         int i;
 344         int will_compress;
 345         int compress_type = root->fs_info->compress_type;
 346
 347         /* if this is a small write inside eof, kick off a defragbot */
 348         if (end <= BTRFS_I(inode)->disk_i_size && (end - start + 1) < 16 * 1024)
 349                 btrfs_add_inode_defrag(NULL, inode);
 350
 351         actual_end = min_t(u64, isize, end + 1);
 352 again:
 353         will_compress = 0;
 354         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 355         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 356
 357         /*
 358          * we don't want to send crud past the end of i_size through
 359          * compression, that's just a waste of CPU time.  So, if the
 360          * end of the file is before the start of our current
 361          * requested range of bytes, we bail out to the uncompressed
 362          * cleanup code that can deal with all of this.
 363          *
 364          * It isn't really the fastest way to fix things, but this is a
 365          * very uncommon corner.
 366          */
 367         if (actual_end <= start)
 368                 goto cleanup_and_bail_uncompressed;
 369
 370         total_compressed = actual_end - start;
 371
 372         /* we want to make sure that amount of ram required to uncompress
 373          * an extent is reasonable, so we limit the total size in ram
 374          * of a compressed extent to 128k.  This is a crucial number
 375          * because it also controls how easily we can spread reads across
 376          * cpus for decompression.
 377          *
 378          * We also want to make sure the amount of IO required to do
 379          * a random read is reasonably small, so we limit the size of
 380          * a compressed extent to 128k.
 381          */
 382         total_compressed = min(total_compressed, max_uncompressed);
 383         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 384         num_bytes = max(blocksize,  num_bytes);
 385         total_in = 0;
 386         ret = 0;
 387
 388         /*
 389          * we do compression for mount -o compress and when the
 390          * inode has not been flagged as nocompress.  This flag can
 391          * change at any time if we discover bad compression ratios.
 392          */
 393         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 394             (btrfs_test_opt(root, COMPRESS) ||
 395              (BTRFS_I(inode)->force_compress) ||
 396              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 397                 WARN_ON(pages);
 398                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 399                 if (!pages) {
 400                         /* just bail out to the uncompressed code */
 401                         goto cont;
 402                 }
 403
 404                 if (BTRFS_I(inode)->force_compress)
 405                         compress_type = BTRFS_I(inode)->force_compress;
 406
 407                 ret = btrfs_compress_pages(compress_type,
 408                                            inode->i_mapping, start,
 409                                            total_compressed, pages,
 410                                            nr_pages, &nr_pages_ret,
 411                                            &total_in,
 412                                            &total_compressed,
 413                                            max_compressed);
 414
 415                 if (!ret) {
 416                         unsigned long offset = total_compressed &
 417                                 (PAGE_CACHE_SIZE - 1);
 418                         struct page *page = pages[nr_pages_ret - 1];
 419                         char *kaddr;
 420
 421                         /* zero the tail end of the last page, we might be
 422                          * sending it down to disk
 423                          */
 424                         if (offset) {
 425                                 kaddr = kmap_atomic(page, KM_USER0);
 426                                 memset(kaddr + offset, 0,
 427                                        PAGE_CACHE_SIZE - offset);
 428                                 kunmap_atomic(kaddr, KM_USER0);
 429                         }
 430                         will_compress = 1;
 431                 }
 432         }
 433 cont:
 434         if (start == 0) {
 435                 trans = btrfs_join_transaction(root);
 436                 BUG_ON(IS_ERR(trans));
 437                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 438
 439                 /* lets try to make an inline extent */
 440                 if (ret || total_in < (actual_end - start)) {
 441                         /* we didn't compress the entire range, try
 442                          * to make an uncompressed inline extent.
 443                          */
 444                         ret = cow_file_range_inline(trans, root, inode,
 445                                                     start, end, 0, 0, NULL);
 446                 } else {
 447                         /* try making a compressed inline extent */
 448                         ret = cow_file_range_inline(trans, root, inode,
 449                                                     start, end,
 450                                                     total_compressed,
 451                                                     compress_type, pages);
 452                 }
 453                 if (ret == 0) {
 454                         /*
 455                          * inline extent creation worked, we don't need
 456                          * to create any more async work items.  Unlock
 457                          * and free up our temp pages.
 458                          */
 459                         extent_clear_unlock_delalloc(inode,
 460                              &BTRFS_I(inode)->io_tree,
 461                              start, end, NULL,
 462                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 463                              EXTENT_CLEAR_DELALLOC |
 464                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 465
 466                         btrfs_end_transaction(trans, root);
 467                         goto free_pages_out;
 468                 }
 469                 btrfs_end_transaction(trans, root);
 470         }
 471
 472         if (will_compress) {
 473                 /*
 474                  * we aren't doing an inline extent round the compressed size
 475                  * up to a block size boundary so the allocator does sane
 476                  * things
 477                  */
 478                 total_compressed = (total_compressed + blocksize - 1) &
 479                         ~(blocksize - 1);
 480
 481                 /*
 482                  * one last check to make sure the compression is really a
 483                  * win, compare the page count read with the blocks on disk
 484                  */
 485                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 486                         ~(PAGE_CACHE_SIZE - 1);
 487                 if (total_compressed >= total_in) {
 488                         will_compress = 0;
 489                 } else {
 490                         num_bytes = total_in;
 491                 }
 492         }
 493         if (!will_compress && pages) {
 494                 /*
 495                  * the compression code ran but failed to make things smaller,
 496                  * free any pages it allocated and our page pointer array
 497                  */
 498                 for (i = 0; i < nr_pages_ret; i++) {
 499                         WARN_ON(pages[i]->mapping);
 500                         page_cache_release(pages[i]);
 501                 }
 502                 kfree(pages);
 503                 pages = NULL;
 504                 total_compressed = 0;
 505                 nr_pages_ret = 0;
 506
 507                 /* flag the file so we don't compress in the future */
 508                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 509                     !(BTRFS_I(inode)->force_compress)) {
 510                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 511                 }
 512         }
 513         if (will_compress) {
 514                 *num_added += 1;
 515
 516                 /* the async work queues will take care of doing actual
 517                  * allocation on disk for these compressed pages,
 518                  * and will submit them to the elevator.
 519                  */
 520                 add_async_extent(async_cow, start, num_bytes,
 521                                  total_compressed, pages, nr_pages_ret,
 522                                  compress_type);
 523
 524                 if (start + num_bytes < end) {
 525                         start += num_bytes;
 526                         pages = NULL;
 527                         cond_resched();
 528                         goto again;
 529                 }
 530         } else {
 531 cleanup_and_bail_uncompressed:
 532                 /*
 533                  * No compression, but we still need to write the pages in
 534                  * the file we've been given so far.  redirty the locked
 535                  * page if it corresponds to our extent and set things up
 536                  * for the async work queue to run cow_file_range to do
 537                  * the normal delalloc dance
 538                  */
 539                 if (page_offset(locked_page) >= start &&
 540                     page_offset(locked_page) <= end) {
 541                         __set_page_dirty_nobuffers(locked_page);
 542                         /* unlocked later on in the async handlers */
 543                 }
 544                 add_async_extent(async_cow, start, end - start + 1,
 545                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 546                 *num_added += 1;
 547         }
 548
 549 out:
 550         return 0;
 551
 552 free_pages_out:
 553         for (i = 0; i < nr_pages_ret; i++) {
 554                 WARN_ON(pages[i]->mapping);
 555                 page_cache_release(pages[i]);
 556         }
 557         kfree(pages);
 558
 559         goto out;
 560 }
 561
 562 /*
 563  * phase two of compressed writeback.  This is the ordered portion
 564  * of the code, which only gets called in the order the work was
 565  * queued.  We walk all the async extents created by compress_file_range
 566  * and send them down to the disk.
 567  */
 568 static noinline int submit_compressed_extents(struct inode *inode,
 569                                               struct async_cow *async_cow)
 570 {
 571         struct async_extent *async_extent;
 572         u64 alloc_hint = 0;
 573         struct btrfs_trans_handle *trans;
 574         struct btrfs_key ins;
 575         struct extent_map *em;
 576         struct btrfs_root *root = BTRFS_I(inode)->root;
 577         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 578         struct extent_io_tree *io_tree;
 579         int ret = 0;
 580
 581         if (list_empty(&async_cow->extents))
 582                 return 0;
 583
 584
 585         while (!list_empty(&async_cow->extents)) {
 586                 async_extent = list_entry(async_cow->extents.next,
 587                                           struct async_extent, list);
 588                 list_del(&async_extent->list);
 589
 590                 io_tree = &BTRFS_I(inode)->io_tree;
 591
 592 retry:
 593                 /* did the compression code fall back to uncompressed IO? */
 594                 if (!async_extent->pages) {
 595                         int page_started = 0;
 596                         unsigned long nr_written = 0;
 597
 598                         lock_extent(io_tree, async_extent->start,
 599                                          async_extent->start +
 600                                          async_extent->ram_size - 1, GFP_NOFS);
 601
 602                         /* allocate blocks */
 603                         ret = cow_file_range(inode, async_cow->locked_page,
 604                                              async_extent->start,
 605                                              async_extent->start +
 606                                              async_extent->ram_size - 1,
 607                                              &page_started, &nr_written, 0);
 608
 609                         /*
 610                          * if page_started, cow_file_range inserted an
 611                          * inline extent and took care of all the unlocking
 612                          * and IO for us.  Otherwise, we need to submit
 613                          * all those pages down to the drive.
 614                          */
 615                         if (!page_started && !ret)
 616                                 extent_write_locked_range(io_tree,
 617                                                   inode, async_extent->start,
 618                                                   async_extent->start +
 619                                                   async_extent->ram_size - 1,
 620                                                   btrfs_get_extent,
 621                                                   WB_SYNC_ALL);
 622                         kfree(async_extent);
 623                         cond_resched();
 624                         continue;
 625                 }
 626
 627                 lock_extent(io_tree, async_extent->start,
 628                             async_extent->start + async_extent->ram_size - 1,
 629                             GFP_NOFS);
 630
 631                 trans = btrfs_join_transaction(root);
 632                 BUG_ON(IS_ERR(trans));
 633                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 634                 ret = btrfs_reserve_extent(trans, root,
 635                                            async_extent->compressed_size,
 636                                            async_extent->compressed_size,
 637                                            0, alloc_hint,
 638                                            (u64)-1, &ins, 1);
 639                 btrfs_end_transaction(trans, root);
 640
 641                 if (ret) {
 642                         int i;
 643                         for (i = 0; i < async_extent->nr_pages; i++) {
 644                                 WARN_ON(async_extent->pages[i]->mapping);
 645                                 page_cache_release(async_extent->pages[i]);
 646                         }
 647                         kfree(async_extent->pages);
 648                         async_extent->nr_pages = 0;
 649                         async_extent->pages = NULL;
 650                         unlock_extent(io_tree, async_extent->start,
 651                                       async_extent->start +
 652                                       async_extent->ram_size - 1, GFP_NOFS);
 653                         goto retry;
 654                 }
 655
 656                 /*
 657                  * here we're doing allocation and writeback of the
 658                  * compressed pages
 659                  */
 660                 btrfs_drop_extent_cache(inode, async_extent->start,
 661                                         async_extent->start +
 662                                         async_extent->ram_size - 1, 0);
 663
 664                 em = alloc_extent_map();
 665                 BUG_ON(!em);
 666                 em->start = async_extent->start;
 667                 em->len = async_extent->ram_size;
 668                 em->orig_start = em->start;
 669
 670                 em->block_start = ins.objectid;
 671                 em->block_len = ins.offset;
 672                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 673                 em->compress_type = async_extent->compress_type;
 674                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 675                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 676
 677                 while (1) {
 678                         write_lock(&em_tree->lock);
 679                         ret = add_extent_mapping(em_tree, em);
 680                         write_unlock(&em_tree->lock);
 681                         if (ret != -EEXIST) {
 682                                 free_extent_map(em);
 683                                 break;
 684                         }
 685                         btrfs_drop_extent_cache(inode, async_extent->start,
 686                                                 async_extent->start +
 687                                                 async_extent->ram_size - 1, 0);
 688                 }
 689
 690                 ret = btrfs_add_ordered_extent_compress(inode,
 691                                                 async_extent->start,
 692                                                 ins.objectid,
 693                                                 async_extent->ram_size,
 694                                                 ins.offset,
 695                                                 BTRFS_ORDERED_COMPRESSED,
 696                                                 async_extent->compress_type);
 697                 BUG_ON(ret);
 698
 699                 /*
 700                  * clear dirty, set writeback and unlock the pages.
 701                  */
 702                 extent_clear_unlock_delalloc(inode,
 703                                 &BTRFS_I(inode)->io_tree,
 704                                 async_extent->start,
 705                                 async_extent->start +
 706                                 async_extent->ram_size - 1,
 707                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 708                                 EXTENT_CLEAR_UNLOCK |
 709                                 EXTENT_CLEAR_DELALLOC |
 710                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 711
 712                 ret = btrfs_submit_compressed_write(inode,
 713                                     async_extent->start,
 714                                     async_extent->ram_size,
 715                                     ins.objectid,
 716                                     ins.offset, async_extent->pages,
 717                                     async_extent->nr_pages);
 718
 719                 BUG_ON(ret);
 720                 alloc_hint = ins.objectid + ins.offset;
 721                 kfree(async_extent);
 722                 cond_resched();
 723         }
 724
 725         return 0;
 726 }
 727
 728 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 729                                       u64 num_bytes)
 730 {
 731         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 732         struct extent_map *em;
 733         u64 alloc_hint = 0;
 734
 735         read_lock(&em_tree->lock);
 736         em = search_extent_mapping(em_tree, start, num_bytes);
 737         if (em) {
 738                 /*
 739                  * if block start isn't an actual block number then find the
 740                  * first block in this inode and use that as a hint.  If that
 741                  * block is also bogus then just don't worry about it.
 742                  */
 743                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 744                         free_extent_map(em);
 745                         em = search_extent_mapping(em_tree, 0, 0);
 746                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 747                                 alloc_hint = em->block_start;
 748                         if (em)
 749                                 free_extent_map(em);
 750                 } else {
 751                         alloc_hint = em->block_start;
 752                         free_extent_map(em);
 753                 }
 754         }
 755         read_unlock(&em_tree->lock);
 756
 757         return alloc_hint;
 758 }
 759
 760 /*
 761  * when extent_io.c finds a delayed allocation range in the file,
 762  * the call backs end up in this code.  The basic idea is to
 763  * allocate extents on disk for the range, and create ordered data structs
 764  * in ram to track those extents.
 765  *
 766  * locked_page is the page that writepage had locked already.  We use
 767  * it to make sure we don't do extra locks or unlocks.
 768  *
 769  * *page_started is set to one if we unlock locked_page and do everything
 770  * required to start IO on it.  It may be clean and already done with
 771  * IO when we return.
 772  */
 773 static noinline int cow_file_range(struct inode *inode,
 774                                    struct page *locked_page,
 775                                    u64 start, u64 end, int *page_started,
 776                                    unsigned long *nr_written,
 777                                    int unlock)
 778 {
 779         struct btrfs_root *root = BTRFS_I(inode)->root;
 780         struct btrfs_trans_handle *trans;
 781         u64 alloc_hint = 0;
 782         u64 num_bytes;
 783         unsigned long ram_size;
 784         u64 disk_num_bytes;
 785         u64 cur_alloc_size;
 786         u64 blocksize = root->sectorsize;
 787         struct btrfs_key ins;
 788         struct extent_map *em;
 789         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 790         int ret = 0;
 791
 792         BUG_ON(btrfs_is_free_space_inode(root, inode));
 793         trans = btrfs_join_transaction(root);
 794         BUG_ON(IS_ERR(trans));
 795         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 796
 797         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 798         num_bytes = max(blocksize,  num_bytes);
 799         disk_num_bytes = num_bytes;
 800         ret = 0;
 801
 802         /* if this is a small write inside eof, kick off defrag */
 803         if (end <= BTRFS_I(inode)->disk_i_size && num_bytes < 64 * 1024)
 804                 btrfs_add_inode_defrag(trans, inode);
 805
 806         if (start == 0) {
 807                 /* lets try to make an inline extent */
 808                 ret = cow_file_range_inline(trans, root, inode,
 809                                             start, end, 0, 0, NULL);
 810                 if (ret == 0) {
 811                         extent_clear_unlock_delalloc(inode,
 812                                      &BTRFS_I(inode)->io_tree,
 813                                      start, end, NULL,
 814                                      EXTENT_CLEAR_UNLOCK_PAGE |
 815                                      EXTENT_CLEAR_UNLOCK |
 816                                      EXTENT_CLEAR_DELALLOC |
 817                                      EXTENT_CLEAR_DIRTY |
 818                                      EXTENT_SET_WRITEBACK |
 819                                      EXTENT_END_WRITEBACK);
 820
 821                         *nr_written = *nr_written +
 822                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 823                         *page_started = 1;
 824                         ret = 0;
 825                         goto out;
 826                 }
 827         }
 828
 829         BUG_ON(disk_num_bytes >
 830                btrfs_super_total_bytes(root->fs_info->super_copy));
 831
 832         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 833         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 834
 835         while (disk_num_bytes > 0) {
 836                 unsigned long op;
 837
 838                 cur_alloc_size = disk_num_bytes;
 839                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 840                                            root->sectorsize, 0, alloc_hint,
 841                                            (u64)-1, &ins, 1);
 842                 BUG_ON(ret);
 843
 844                 em = alloc_extent_map();
 845                 BUG_ON(!em);
 846                 em->start = start;
 847                 em->orig_start = em->start;
 848                 ram_size = ins.offset;
 849                 em->len = ins.offset;
 850
 851                 em->block_start = ins.objectid;
 852                 em->block_len = ins.offset;
 853                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 854                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 855
 856                 while (1) {
 857                         write_lock(&em_tree->lock);
 858                         ret = add_extent_mapping(em_tree, em);
 859                         write_unlock(&em_tree->lock);
 860                         if (ret != -EEXIST) {
 861                                 free_extent_map(em);
 862                                 break;
 863                         }
 864                         btrfs_drop_extent_cache(inode, start,
 865                                                 start + ram_size - 1, 0);
 866                 }
 867
 868                 cur_alloc_size = ins.offset;
 869                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 870                                                ram_size, cur_alloc_size, 0);
 871                 BUG_ON(ret);
 872
 873                 if (root->root_key.objectid ==
 874                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 875                         ret = btrfs_reloc_clone_csums(inode, start,
 876                                                       cur_alloc_size);
 877                         BUG_ON(ret);
 878                 }
 879
 880                 if (disk_num_bytes < cur_alloc_size)
 881                         break;
 882
 883                 /* we're not doing compressed IO, don't unlock the first
 884                  * page (which the caller expects to stay locked), don't
 885                  * clear any dirty bits and don't set any writeback bits
 886                  *
 887                  * Do set the Private2 bit so we know this page was properly
 888                  * setup for writepage
 889                  */
 890                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 891                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 892                         EXTENT_SET_PRIVATE2;
 893
 894                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 895                                              start, start + ram_size - 1,
 896                                              locked_page, op);
 897                 disk_num_bytes -= cur_alloc_size;
 898                 num_bytes -= cur_alloc_size;
 899                 alloc_hint = ins.objectid + ins.offset;
 900                 start += cur_alloc_size;
 901         }
 902 out:
 903         ret = 0;
 904         btrfs_end_transaction(trans, root);
 905
 906         return ret;
 907 }
 908
 909 /*
 910  * work queue call back to started compression on a file and pages
 911  */
 912 static noinline void async_cow_start(struct btrfs_work *work)
 913 {
 914         struct async_cow *async_cow;
 915         int num_added = 0;
 916         async_cow = container_of(work, struct async_cow, work);
 917
 918         compress_file_range(async_cow->inode, async_cow->locked_page,
 919                             async_cow->start, async_cow->end, async_cow,
 920                             &num_added);
 921         if (num_added == 0)
 922                 async_cow->inode = NULL;
 923 }
 924
 925 /*
 926  * work queue call back to submit previously compressed pages
 927  */
 928 static noinline void async_cow_submit(struct btrfs_work *work)
 929 {
 930         struct async_cow *async_cow;
 931         struct btrfs_root *root;
 932         unsigned long nr_pages;
 933
 934         async_cow = container_of(work, struct async_cow, work);
 935
 936         root = async_cow->root;
 937         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
 938                 PAGE_CACHE_SHIFT;
 939
 940         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
 941
 942         if (atomic_read(&root->fs_info->async_delalloc_pages) <
 943             5 * 1042 * 1024 &&
 944             waitqueue_active(&root->fs_info->async_submit_wait))
 945                 wake_up(&root->fs_info->async_submit_wait);
 946
 947         if (async_cow->inode)
 948                 submit_compressed_extents(async_cow->inode, async_cow);
 949 }
 950
 951 static noinline void async_cow_free(struct btrfs_work *work)
 952 {
 953         struct async_cow *async_cow;
 954         async_cow = container_of(work, struct async_cow, work);
 955         kfree(async_cow);
 956 }
 957
 958 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
 959                                 u64 start, u64 end, int *page_started,
 960                                 unsigned long *nr_written)
 961 {
 962         struct async_cow *async_cow;
 963         struct btrfs_root *root = BTRFS_I(inode)->root;
 964         unsigned long nr_pages;
 965         u64 cur_end;
 966         int limit = 10 * 1024 * 1042;
 967
 968         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
 969                          1, 0, NULL, GFP_NOFS);
 970         while (start < end) {
 971                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
 972                 BUG_ON(!async_cow);
 973                 async_cow->inode = inode;
 974                 async_cow->root = root;
 975                 async_cow->locked_page = locked_page;
 976                 async_cow->start = start;
 977
 978                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 979                         cur_end = end;
 980                 else
 981                         cur_end = min(end, start + 512 * 1024 - 1);
 982
 983                 async_cow->end = cur_end;
 984                 INIT_LIST_HEAD(&async_cow->extents);
 985
 986                 async_cow->work.func = async_cow_start;
 987                 async_cow->work.ordered_func = async_cow_submit;
 988                 async_cow->work.ordered_free = async_cow_free;
 989                 async_cow->work.flags = 0;
 990
 991                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
 992                         PAGE_CACHE_SHIFT;
 993                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
 994
 995                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
 996                                    &async_cow->work);
 997
 998                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
 999                         wait_event(root->fs_info->async_submit_wait,
1000                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1001                             limit));
1002                 }
1003
1004                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1005                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1006                         wait_event(root->fs_info->async_submit_wait,
1007                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1008                            0));
1009                 }
1010
1011                 *nr_written += nr_pages;
1012                 start = cur_end + 1;
1013         }
1014         *page_started = 1;
1015         return 0;
1016 }
1017
1018 static noinline int csum_exist_in_range(struct btrfs_root *root,
1019                                         u64 bytenr, u64 num_bytes)
1020 {
1021         int ret;
1022         struct btrfs_ordered_sum *sums;
1023         LIST_HEAD(list);
1024
1025         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1026                                        bytenr + num_bytes - 1, &list, 0);
1027         if (ret == 0 && list_empty(&list))
1028                 return 0;
1029
1030         while (!list_empty(&list)) {
1031                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1032                 list_del(&sums->list);
1033                 kfree(sums);
1034         }
1035         return 1;
1036 }
1037
1038 /*
1039  * when nowcow writeback call back.  This checks for snapshots or COW copies
1040  * of the extents that exist in the file, and COWs the file as required.
1041  *
1042  * If no cow copies or snapshots exist, we write directly to the existing
1043  * blocks on disk
1044  */
1045 static noinline int run_delalloc_nocow(struct inode *inode,
1046                                        struct page *locked_page,
1047                               u64 start, u64 end, int *page_started, int force,
1048                               unsigned long *nr_written)
1049 {
1050         struct btrfs_root *root = BTRFS_I(inode)->root;
1051         struct btrfs_trans_handle *trans;
1052         struct extent_buffer *leaf;
1053         struct btrfs_path *path;
1054         struct btrfs_file_extent_item *fi;
1055         struct btrfs_key found_key;
1056         u64 cow_start;
1057         u64 cur_offset;
1058         u64 extent_end;
1059         u64 extent_offset;
1060         u64 disk_bytenr;
1061         u64 num_bytes;
1062         int extent_type;
1063         int ret;
1064         int type;
1065         int nocow;
1066         int check_prev = 1;
1067         bool nolock;
1068         u64 ino = btrfs_ino(inode);
1069
1070         path = btrfs_alloc_path();
1071         if (!path)
1072                 return -ENOMEM;
1073
1074         nolock = btrfs_is_free_space_inode(root, inode);
1075
1076         if (nolock)
1077                 trans = btrfs_join_transaction_nolock(root);
1078         else
1079                 trans = btrfs_join_transaction(root);
1080
1081         BUG_ON(IS_ERR(trans));
1082         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1083
1084         cow_start = (u64)-1;
1085         cur_offset = start;
1086         while (1) {
1087                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1088                                                cur_offset, 0);
1089                 BUG_ON(ret < 0);
1090                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1091                         leaf = path->nodes[0];
1092                         btrfs_item_key_to_cpu(leaf, &found_key,
1093                                               path->slots[0] - 1);
1094                         if (found_key.objectid == ino &&
1095                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1096                                 path->slots[0]--;
1097                 }
1098                 check_prev = 0;
1099 next_slot:
1100                 leaf = path->nodes[0];
1101                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1102                         ret = btrfs_next_leaf(root, path);
1103                         if (ret < 0)
1104                                 BUG_ON(1);
1105                         if (ret > 0)
1106                                 break;
1107                         leaf = path->nodes[0];
1108                 }
1109
1110                 nocow = 0;
1111                 disk_bytenr = 0;
1112                 num_bytes = 0;
1113                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1114
1115                 if (found_key.objectid > ino ||
1116                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1117                     found_key.offset > end)
1118                         break;
1119
1120                 if (found_key.offset > cur_offset) {
1121                         extent_end = found_key.offset;
1122                         extent_type = 0;
1123                         goto out_check;
1124                 }
1125
1126                 fi = btrfs_item_ptr(leaf, path->slots[0],
1127                                     struct btrfs_file_extent_item);
1128                 extent_type = btrfs_file_extent_type(leaf, fi);
1129
1130                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1131                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1132                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1133                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1134                         extent_end = found_key.offset +
1135                                 btrfs_file_extent_num_bytes(leaf, fi);
1136                         if (extent_end <= start) {
1137                                 path->slots[0]++;
1138                                 goto next_slot;
1139                         }
1140                         if (disk_bytenr == 0)
1141                                 goto out_check;
1142                         if (btrfs_file_extent_compression(leaf, fi) ||
1143                             btrfs_file_extent_encryption(leaf, fi) ||
1144                             btrfs_file_extent_other_encoding(leaf, fi))
1145                                 goto out_check;
1146                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1147                                 goto out_check;
1148                         if (btrfs_extent_readonly(root, disk_bytenr))
1149                                 goto out_check;
1150                         if (btrfs_cross_ref_exist(trans, root, ino,
1151                                                   found_key.offset -
1152                                                   extent_offset, disk_bytenr))
1153                                 goto out_check;
1154                         disk_bytenr += extent_offset;
1155                         disk_bytenr += cur_offset - found_key.offset;
1156                         num_bytes = min(end + 1, extent_end) - cur_offset;
1157                         /*
1158                          * force cow if csum exists in the range.
1159                          * this ensure that csum for a given extent are
1160                          * either valid or do not exist.
1161                          */
1162                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1163                                 goto out_check;
1164                         nocow = 1;
1165                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1166                         extent_end = found_key.offset +
1167                                 btrfs_file_extent_inline_len(leaf, fi);
1168                         extent_end = ALIGN(extent_end, root->sectorsize);
1169                 } else {
1170                         BUG_ON(1);
1171                 }
1172 out_check:
1173                 if (extent_end <= start) {
1174                         path->slots[0]++;
1175                         goto next_slot;
1176                 }
1177                 if (!nocow) {
1178                         if (cow_start == (u64)-1)
1179                                 cow_start = cur_offset;
1180                         cur_offset = extent_end;
1181                         if (cur_offset > end)
1182                                 break;
1183                         path->slots[0]++;
1184                         goto next_slot;
1185                 }
1186
1187                 btrfs_release_path(path);
1188                 if (cow_start != (u64)-1) {
1189                         ret = cow_file_range(inode, locked_page, cow_start,
1190                                         found_key.offset - 1, page_started,
1191                                         nr_written, 1);
1192                         BUG_ON(ret);
1193                         cow_start = (u64)-1;
1194                 }
1195
1196                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1197                         struct extent_map *em;
1198                         struct extent_map_tree *em_tree;
1199                         em_tree = &BTRFS_I(inode)->extent_tree;
1200                         em = alloc_extent_map();
1201                         BUG_ON(!em);
1202                         em->start = cur_offset;
1203                         em->orig_start = em->start;
1204                         em->len = num_bytes;
1205                         em->block_len = num_bytes;
1206                         em->block_start = disk_bytenr;
1207                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1208                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1209                         while (1) {
1210                                 write_lock(&em_tree->lock);
1211                                 ret = add_extent_mapping(em_tree, em);
1212                                 write_unlock(&em_tree->lock);
1213                                 if (ret != -EEXIST) {
1214                                         free_extent_map(em);
1215                                         break;
1216                                 }
1217                                 btrfs_drop_extent_cache(inode, em->start,
1218                                                 em->start + em->len - 1, 0);
1219                         }
1220                         type = BTRFS_ORDERED_PREALLOC;
1221                 } else {
1222                         type = BTRFS_ORDERED_NOCOW;
1223                 }
1224
1225                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1226                                                num_bytes, num_bytes, type);
1227                 BUG_ON(ret);
1228
1229                 if (root->root_key.objectid ==
1230                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1231                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1232                                                       num_bytes);
1233                         BUG_ON(ret);
1234                 }
1235
1236                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1237                                 cur_offset, cur_offset + num_bytes - 1,
1238                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1239                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1240                                 EXTENT_SET_PRIVATE2);
1241                 cur_offset = extent_end;
1242                 if (cur_offset > end)
1243                         break;
1244         }
1245         btrfs_release_path(path);
1246
1247         if (cur_offset <= end && cow_start == (u64)-1)
1248                 cow_start = cur_offset;
1249         if (cow_start != (u64)-1) {
1250                 ret = cow_file_range(inode, locked_page, cow_start, end,
1251                                      page_started, nr_written, 1);
1252                 BUG_ON(ret);
1253         }
1254
1255         if (nolock) {
1256                 ret = btrfs_end_transaction_nolock(trans, root);
1257                 BUG_ON(ret);
1258         } else {
1259                 ret = btrfs_end_transaction(trans, root);
1260                 BUG_ON(ret);
1261         }
1262         btrfs_free_path(path);
1263         return 0;
1264 }
1265
1266 /*
1267  * extent_io.c call back to do delayed allocation processing
1268  */
1269 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1270                               u64 start, u64 end, int *page_started,
1271                               unsigned long *nr_written)
1272 {
1273         int ret;
1274         struct btrfs_root *root = BTRFS_I(inode)->root;
1275
1276         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1277                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1278                                          page_started, 1, nr_written);
1279         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1280                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1281                                          page_started, 0, nr_written);
1282         else if (!btrfs_test_opt(root, COMPRESS) &&
1283                  !(BTRFS_I(inode)->force_compress) &&
1284                  !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1285                 ret = cow_file_range(inode, locked_page, start, end,
1286                                       page_started, nr_written, 1);
1287         else
1288                 ret = cow_file_range_async(inode, locked_page, start, end,
1289                                            page_started, nr_written);
1290         return ret;
1291 }
1292
1293 static void btrfs_split_extent_hook(struct inode *inode,
1294                                     struct extent_state *orig, u64 split)
1295 {
1296         /* not delalloc, ignore it */
1297         if (!(orig->state & EXTENT_DELALLOC))
1298                 return;
1299
1300         spin_lock(&BTRFS_I(inode)->lock);
1301         BTRFS_I(inode)->outstanding_extents++;
1302         spin_unlock(&BTRFS_I(inode)->lock);
1303 }
1304
1305 /*
1306  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1307  * extents so we can keep track of new extents that are just merged onto old
1308  * extents, such as when we are doing sequential writes, so we can properly
1309  * account for the metadata space we'll need.
1310  */
1311 static void btrfs_merge_extent_hook(struct inode *inode,
1312                                     struct extent_state *new,
1313                                     struct extent_state *other)
1314 {
1315         /* not delalloc, ignore it */
1316         if (!(other->state & EXTENT_DELALLOC))
1317                 return;
1318
1319         spin_lock(&BTRFS_I(inode)->lock);
1320         BTRFS_I(inode)->outstanding_extents--;
1321         spin_unlock(&BTRFS_I(inode)->lock);
1322 }
1323
1324 /*
1325  * extent_io.c set_bit_hook, used to track delayed allocation
1326  * bytes in this file, and to maintain the list of inodes that
1327  * have pending delalloc work to be done.
1328  */
1329 static void btrfs_set_bit_hook(struct inode *inode,
1330                                struct extent_state *state, int *bits)
1331 {
1332
1333         /*
1334          * set_bit and clear bit hooks normally require _irqsave/restore
1335          * but in this case, we are only testing for the DELALLOC
1336          * bit, which is only set or cleared with irqs on
1337          */
1338         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1339                 struct btrfs_root *root = BTRFS_I(inode)->root;
1340                 u64 len = state->end + 1 - state->start;
1341                 bool do_list = !btrfs_is_free_space_inode(root, inode);
1342
1343                 if (*bits & EXTENT_FIRST_DELALLOC) {
1344                         *bits &= ~EXTENT_FIRST_DELALLOC;
1345                 } else {
1346                         spin_lock(&BTRFS_I(inode)->lock);
1347                         BTRFS_I(inode)->outstanding_extents++;
1348                         spin_unlock(&BTRFS_I(inode)->lock);
1349                 }
1350
1351                 spin_lock(&root->fs_info->delalloc_lock);
1352                 BTRFS_I(inode)->delalloc_bytes += len;
1353                 root->fs_info->delalloc_bytes += len;
1354                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1355                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1356                                       &root->fs_info->delalloc_inodes);
1357                 }
1358                 spin_unlock(&root->fs_info->delalloc_lock);
1359         }
1360 }
1361
1362 /*
1363  * extent_io.c clear_bit_hook, see set_bit_hook for why
1364  */
1365 static void btrfs_clear_bit_hook(struct inode *inode,
1366                                  struct extent_state *state, int *bits)
1367 {
1368         /*
1369          * set_bit and clear bit hooks normally require _irqsave/restore
1370          * but in this case, we are only testing for the DELALLOC
1371          * bit, which is only set or cleared with irqs on
1372          */
1373         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1374                 struct btrfs_root *root = BTRFS_I(inode)->root;
1375                 u64 len = state->end + 1 - state->start;
1376                 bool do_list = !btrfs_is_free_space_inode(root, inode);
1377
1378                 if (*bits & EXTENT_FIRST_DELALLOC) {
1379                         *bits &= ~EXTENT_FIRST_DELALLOC;
1380                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1381                         spin_lock(&BTRFS_I(inode)->lock);
1382                         BTRFS_I(inode)->outstanding_extents--;
1383                         spin_unlock(&BTRFS_I(inode)->lock);
1384                 }
1385
1386                 if (*bits & EXTENT_DO_ACCOUNTING)
1387                         btrfs_delalloc_release_metadata(inode, len);
1388
1389                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1390                     && do_list)
1391                         btrfs_free_reserved_data_space(inode, len);
1392
1393                 spin_lock(&root->fs_info->delalloc_lock);
1394                 root->fs_info->delalloc_bytes -= len;
1395                 BTRFS_I(inode)->delalloc_bytes -= len;
1396
1397                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1398                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1399                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1400                 }
1401                 spin_unlock(&root->fs_info->delalloc_lock);
1402         }
1403 }
1404
1405 /*
1406  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1407  * we don't create bios that span stripes or chunks
1408  */
1409 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1410                          size_t size, struct bio *bio,
1411                          unsigned long bio_flags)
1412 {
1413         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1414         struct btrfs_mapping_tree *map_tree;
1415         u64 logical = (u64)bio->bi_sector << 9;
1416         u64 length = 0;
1417         u64 map_length;
1418         int ret;
1419
1420         if (bio_flags & EXTENT_BIO_COMPRESSED)
1421                 return 0;
1422
1423         length = bio->bi_size;
1424         map_tree = &root->fs_info->mapping_tree;
1425         map_length = length;
1426         ret = btrfs_map_block(map_tree, READ, logical,
1427                               &map_length, NULL, 0);
1428
1429         if (map_length < length + size)
1430                 return 1;
1431         return ret;
1432 }
1433
1434 /*
1435  * in order to insert checksums into the metadata in large chunks,
1436  * we wait until bio submission time.   All the pages in the bio are
1437  * checksummed and sums are attached onto the ordered extent record.
1438  *
1439  * At IO completion time the cums attached on the ordered extent record
1440  * are inserted into the btree
1441  */
1442 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1443                                     struct bio *bio, int mirror_num,
1444                                     unsigned long bio_flags,
1445                                     u64 bio_offset)
1446 {
1447         struct btrfs_root *root = BTRFS_I(inode)->root;
1448         int ret = 0;
1449
1450         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1451         BUG_ON(ret);
1452         return 0;
1453 }
1454
1455 /*
1456  * in order to insert checksums into the metadata in large chunks,
1457  * we wait until bio submission time.   All the pages in the bio are
1458  * checksummed and sums are attached onto the ordered extent record.
1459  *
1460  * At IO completion time the cums attached on the ordered extent record
1461  * are inserted into the btree
1462  */
1463 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1464                           int mirror_num, unsigned long bio_flags,
1465                           u64 bio_offset)
1466 {
1467         struct btrfs_root *root = BTRFS_I(inode)->root;
1468         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1469 }
1470
1471 /*
1472  * extent_io.c submission hook. This does the right thing for csum calculation
1473  * on write, or reading the csums from the tree before a read
1474  */
1475 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1476                           int mirror_num, unsigned long bio_flags,
1477                           u64 bio_offset)
1478 {
1479         struct btrfs_root *root = BTRFS_I(inode)->root;
1480         int ret = 0;
1481         int skip_sum;
1482
1483         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1484
1485         if (btrfs_is_free_space_inode(root, inode))
1486                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1487         else
1488                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1489         BUG_ON(ret);
1490
1491         if (!(rw & REQ_WRITE)) {
1492                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1493                         return btrfs_submit_compressed_read(inode, bio,
1494                                                     mirror_num, bio_flags);
1495                 } else if (!skip_sum) {
1496                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1497                         if (ret)
1498                                 return ret;
1499                 }
1500                 goto mapit;
1501         } else if (!skip_sum) {
1502                 /* csum items have already been cloned */
1503                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1504                         goto mapit;
1505                 /* we're doing a write, do the async checksumming */
1506                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1507                                    inode, rw, bio, mirror_num,
1508                                    bio_flags, bio_offset,
1509                                    __btrfs_submit_bio_start,
1510                                    __btrfs_submit_bio_done);
1511         }
1512
1513 mapit:
1514         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1515 }
1516
1517 /*
1518  * given a list of ordered sums record them in the inode.  This happens
1519  * at IO completion time based on sums calculated at bio submission time.
1520  */
1521 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1522                              struct inode *inode, u64 file_offset,
1523                              struct list_head *list)
1524 {
1525         struct btrfs_ordered_sum *sum;
1526
1527         list_for_each_entry(sum, list, list) {
1528                 btrfs_csum_file_blocks(trans,
1529                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1530         }
1531         return 0;
1532 }
1533
1534 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1535                               struct extent_state **cached_state)
1536 {
1537         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1538                 WARN_ON(1);
1539         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1540                                    cached_state, GFP_NOFS);
1541 }
1542
1543 /* see btrfs_writepage_start_hook for details on why this is required */
1544 struct btrfs_writepage_fixup {
1545         struct page *page;
1546         struct btrfs_work work;
1547 };
1548
1549 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1550 {
1551         struct btrfs_writepage_fixup *fixup;
1552         struct btrfs_ordered_extent *ordered;
1553         struct extent_state *cached_state = NULL;
1554         struct page *page;
1555         struct inode *inode;
1556         u64 page_start;
1557         u64 page_end;
1558
1559         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1560         page = fixup->page;
1561 again:
1562         lock_page(page);
1563         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1564                 ClearPageChecked(page);
1565                 goto out_page;
1566         }
1567
1568         inode = page->mapping->host;
1569         page_start = page_offset(page);
1570         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1571
1572         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1573                          &cached_state, GFP_NOFS);
1574
1575         /* already ordered? We're done */
1576         if (PagePrivate2(page))
1577                 goto out;
1578
1579         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1580         if (ordered) {
1581                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1582                                      page_end, &cached_state, GFP_NOFS);
1583                 unlock_page(page);
1584                 btrfs_start_ordered_extent(inode, ordered, 1);
1585                 goto again;
1586         }
1587
1588         BUG();
1589         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1590         ClearPageChecked(page);
1591 out:
1592         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1593                              &cached_state, GFP_NOFS);
1594 out_page:
1595         unlock_page(page);
1596         page_cache_release(page);
1597         kfree(fixup);
1598 }
1599
1600 /*
1601  * There are a few paths in the higher layers of the kernel that directly
1602  * set the page dirty bit without asking the filesystem if it is a
1603  * good idea.  This causes problems because we want to make sure COW
1604  * properly happens and the data=ordered rules are followed.
1605  *
1606  * In our case any range that doesn't have the ORDERED bit set
1607  * hasn't been properly setup for IO.  We kick off an async process
1608  * to fix it up.  The async helper will wait for ordered extents, set
1609  * the delalloc bit and make it safe to write the page.
1610  */
1611 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1612 {
1613         struct inode *inode = page->mapping->host;
1614         struct btrfs_writepage_fixup *fixup;
1615         struct btrfs_root *root = BTRFS_I(inode)->root;
1616
1617         /* this page is properly in the ordered list */
1618         if (TestClearPagePrivate2(page))
1619                 return 0;
1620
1621         if (PageChecked(page))
1622                 return -EAGAIN;
1623
1624         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1625         if (!fixup)
1626                 return -EAGAIN;
1627
1628         SetPageChecked(page);
1629         page_cache_get(page);
1630         fixup->work.func = btrfs_writepage_fixup_worker;
1631         fixup->page = page;
1632         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1633         return -EAGAIN;
1634 }
1635
1636 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1637                                        struct inode *inode, u64 file_pos,
1638                                        u64 disk_bytenr, u64 disk_num_bytes,
1639                                        u64 num_bytes, u64 ram_bytes,
1640                                        u8 compression, u8 encryption,
1641                                        u16 other_encoding, int extent_type)
1642 {
1643         struct btrfs_root *root = BTRFS_I(inode)->root;
1644         struct btrfs_file_extent_item *fi;
1645         struct btrfs_path *path;
1646         struct extent_buffer *leaf;
1647         struct btrfs_key ins;
1648         u64 hint;
1649         int ret;
1650
1651         path = btrfs_alloc_path();
1652         if (!path)
1653                 return -ENOMEM;
1654
1655         path->leave_spinning = 1;
1656
1657         /*
1658          * we may be replacing one extent in the tree with another.
1659          * The new extent is pinned in the extent map, and we don't want
1660          * to drop it from the cache until it is completely in the btree.
1661          *
1662          * So, tell btrfs_drop_extents to leave this extent in the cache.
1663          * the caller is expected to unpin it and allow it to be merged
1664          * with the others.
1665          */
1666         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1667                                  &hint, 0);
1668         BUG_ON(ret);
1669
1670         ins.objectid = btrfs_ino(inode);
1671         ins.offset = file_pos;
1672         ins.type = BTRFS_EXTENT_DATA_KEY;
1673         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1674         BUG_ON(ret);
1675         leaf = path->nodes[0];
1676         fi = btrfs_item_ptr(leaf, path->slots[0],
1677                             struct btrfs_file_extent_item);
1678         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1679         btrfs_set_file_extent_type(leaf, fi, extent_type);
1680         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1681         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1682         btrfs_set_file_extent_offset(leaf, fi, 0);
1683         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1684         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1685         btrfs_set_file_extent_compression(leaf, fi, compression);
1686         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1687         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1688
1689         btrfs_unlock_up_safe(path, 1);
1690         btrfs_set_lock_blocking(leaf);
1691
1692         btrfs_mark_buffer_dirty(leaf);
1693
1694         inode_add_bytes(inode, num_bytes);
1695
1696         ins.objectid = disk_bytenr;
1697         ins.offset = disk_num_bytes;
1698         ins.type = BTRFS_EXTENT_ITEM_KEY;
1699         ret = btrfs_alloc_reserved_file_extent(trans, root,
1700                                         root->root_key.objectid,
1701                                         btrfs_ino(inode), file_pos, &ins);
1702         BUG_ON(ret);
1703         btrfs_free_path(path);
1704
1705         return 0;
1706 }
1707
1708 /*
1709  * helper function for btrfs_finish_ordered_io, this
1710  * just reads in some of the csum leaves to prime them into ram
1711  * before we start the transaction.  It limits the amount of btree
1712  * reads required while inside the transaction.
1713  */
1714 /* as ordered data IO finishes, this gets called so we can finish
1715  * an ordered extent if the range of bytes in the file it covers are
1716  * fully written.
1717  */
1718 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1719 {
1720         struct btrfs_root *root = BTRFS_I(inode)->root;
1721         struct btrfs_trans_handle *trans = NULL;
1722         struct btrfs_ordered_extent *ordered_extent = NULL;
1723         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1724         struct extent_state *cached_state = NULL;
1725         int compress_type = 0;
1726         int ret;
1727         bool nolock;
1728
1729         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1730                                              end - start + 1);
1731         if (!ret)
1732                 return 0;
1733         BUG_ON(!ordered_extent);
1734
1735         nolock = btrfs_is_free_space_inode(root, inode);
1736
1737         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1738                 BUG_ON(!list_empty(&ordered_extent->list));
1739                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1740                 if (!ret) {
1741                         if (nolock)
1742                                 trans = btrfs_join_transaction_nolock(root);
1743                         else
1744                                 trans = btrfs_join_transaction(root);
1745                         BUG_ON(IS_ERR(trans));
1746                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1747                         ret = btrfs_update_inode_fallback(trans, root, inode);
1748                         BUG_ON(ret);
1749                 }
1750                 goto out;
1751         }
1752
1753         lock_extent_bits(io_tree, ordered_extent->file_offset,
1754                          ordered_extent->file_offset + ordered_extent->len - 1,
1755                          0, &cached_state, GFP_NOFS);
1756
1757         if (nolock)
1758                 trans = btrfs_join_transaction_nolock(root);
1759         else
1760                 trans = btrfs_join_transaction(root);
1761         BUG_ON(IS_ERR(trans));
1762         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1763
1764         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1765                 compress_type = ordered_extent->compress_type;
1766         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1767                 BUG_ON(compress_type);
1768                 ret = btrfs_mark_extent_written(trans, inode,
1769                                                 ordered_extent->file_offset,
1770                                                 ordered_extent->file_offset +
1771                                                 ordered_extent->len);
1772                 BUG_ON(ret);
1773         } else {
1774                 BUG_ON(root == root->fs_info->tree_root);
1775                 ret = insert_reserved_file_extent(trans, inode,
1776                                                 ordered_extent->file_offset,
1777                                                 ordered_extent->start,
1778                                                 ordered_extent->disk_len,
1779                                                 ordered_extent->len,
1780                                                 ordered_extent->len,
1781                                                 compress_type, 0, 0,
1782                                                 BTRFS_FILE_EXTENT_REG);
1783                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1784                                    ordered_extent->file_offset,
1785                                    ordered_extent->len);
1786                 BUG_ON(ret);
1787         }
1788         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1789                              ordered_extent->file_offset +
1790                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1791
1792         add_pending_csums(trans, inode, ordered_extent->file_offset,
1793                           &ordered_extent->list);
1794
1795         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1796         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1797                 ret = btrfs_update_inode_fallback(trans, root, inode);
1798                 BUG_ON(ret);
1799         }
1800         ret = 0;
1801 out:
1802         if (root != root->fs_info->tree_root)
1803                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1804         if (trans) {
1805                 if (nolock)
1806                         btrfs_end_transaction_nolock(trans, root);
1807                 else
1808                         btrfs_end_transaction(trans, root);
1809         }
1810
1811         /* once for us */
1812         btrfs_put_ordered_extent(ordered_extent);
1813         /* once for the tree */
1814         btrfs_put_ordered_extent(ordered_extent);
1815
1816         return 0;
1817 }
1818
1819 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1820                                 struct extent_state *state, int uptodate)
1821 {
1822         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1823
1824         ClearPagePrivate2(page);
1825         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1826 }
1827
1828 /*
1829  * when reads are done, we need to check csums to verify the data is correct
1830  * if there's a match, we allow the bio to finish.  If not, the code in
1831  * extent_io.c will try to find good copies for us.
1832  */
1833 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1834                                struct extent_state *state)
1835 {
1836         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1837         struct inode *inode = page->mapping->host;
1838         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1839         char *kaddr;
1840         u64 private = ~(u32)0;
1841         int ret;
1842         struct btrfs_root *root = BTRFS_I(inode)->root;
1843         u32 csum = ~(u32)0;
1844
1845         if (PageChecked(page)) {
1846                 ClearPageChecked(page);
1847                 goto good;
1848         }
1849
1850         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1851                 goto good;
1852
1853         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1854             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1855                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1856                                   GFP_NOFS);
1857                 return 0;
1858         }
1859
1860         if (state && state->start == start) {
1861                 private = state->private;
1862                 ret = 0;
1863         } else {
1864                 ret = get_state_private(io_tree, start, &private);
1865         }
1866         kaddr = kmap_atomic(page, KM_USER0);
1867         if (ret)
1868                 goto zeroit;
1869
1870         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1871         btrfs_csum_final(csum, (char *)&csum);
1872         if (csum != private)
1873                 goto zeroit;
1874
1875         kunmap_atomic(kaddr, KM_USER0);
1876 good:
1877         return 0;
1878
1879 zeroit:
1880         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
1881                        "private %llu\n",
1882                        (unsigned long long)btrfs_ino(page->mapping->host),
1883                        (unsigned long long)start, csum,
1884                        (unsigned long long)private);
1885         memset(kaddr + offset, 1, end - start + 1);
1886         flush_dcache_page(page);
1887         kunmap_atomic(kaddr, KM_USER0);
1888         if (private == 0)
1889                 return 0;
1890         return -EIO;
1891 }
1892
1893 struct delayed_iput {
1894         struct list_head list;
1895         struct inode *inode;
1896 };
1897
1898 void btrfs_add_delayed_iput(struct inode *inode)
1899 {
1900         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1901         struct delayed_iput *delayed;
1902
1903         if (atomic_add_unless(&inode->i_count, -1, 1))
1904                 return;
1905
1906         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
1907         delayed->inode = inode;
1908
1909         spin_lock(&fs_info->delayed_iput_lock);
1910         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
1911         spin_unlock(&fs_info->delayed_iput_lock);
1912 }
1913
1914 void btrfs_run_delayed_iputs(struct btrfs_root *root)
1915 {
1916         LIST_HEAD(list);
1917         struct btrfs_fs_info *fs_info = root->fs_info;
1918         struct delayed_iput *delayed;
1919         int empty;
1920
1921         spin_lock(&fs_info->delayed_iput_lock);
1922         empty = list_empty(&fs_info->delayed_iputs);
1923         spin_unlock(&fs_info->delayed_iput_lock);
1924         if (empty)
1925                 return;
1926
1927         down_read(&root->fs_info->cleanup_work_sem);
1928         spin_lock(&fs_info->delayed_iput_lock);
1929         list_splice_init(&fs_info->delayed_iputs, &list);
1930         spin_unlock(&fs_info->delayed_iput_lock);
1931
1932         while (!list_empty(&list)) {
1933                 delayed = list_entry(list.next, struct delayed_iput, list);
1934                 list_del(&delayed->list);
1935                 iput(delayed->inode);
1936                 kfree(delayed);
1937         }
1938         up_read(&root->fs_info->cleanup_work_sem);
1939 }
1940
1941 enum btrfs_orphan_cleanup_state {
1942         ORPHAN_CLEANUP_STARTED  = 1,
1943         ORPHAN_CLEANUP_DONE     = 2,
1944 };
1945
1946 /*
1947  * This is called in transaction commit time. If there are no orphan
1948  * files in the subvolume, it removes orphan item and frees block_rsv
1949  * structure.
1950  */
1951 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
1952                               struct btrfs_root *root)
1953 {
1954         struct btrfs_block_rsv *block_rsv;
1955         int ret;
1956
1957         if (!list_empty(&root->orphan_list) ||
1958             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
1959                 return;
1960
1961         spin_lock(&root->orphan_lock);
1962         if (!list_empty(&root->orphan_list)) {
1963                 spin_unlock(&root->orphan_lock);
1964                 return;
1965         }
1966
1967         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
1968                 spin_unlock(&root->orphan_lock);
1969                 return;
1970         }
1971
1972         block_rsv = root->orphan_block_rsv;
1973         root->orphan_block_rsv = NULL;
1974         spin_unlock(&root->orphan_lock);
1975
1976         if (root->orphan_item_inserted &&
1977             btrfs_root_refs(&root->root_item) > 0) {
1978                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
1979                                             root->root_key.objectid);
1980                 BUG_ON(ret);
1981                 root->orphan_item_inserted = 0;
1982         }
1983
1984         if (block_rsv) {
1985                 WARN_ON(block_rsv->size > 0);
1986                 btrfs_free_block_rsv(root, block_rsv);
1987         }
1988 }
1989
1990 /*
1991  * This creates an orphan entry for the given inode in case something goes
1992  * wrong in the middle of an unlink/truncate.
1993  *
1994  * NOTE: caller of this function should reserve 5 units of metadata for
1995  *       this function.
1996  */
1997 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1998 {
1999         struct btrfs_root *root = BTRFS_I(inode)->root;
2000         struct btrfs_block_rsv *block_rsv = NULL;
2001         int reserve = 0;
2002         int insert = 0;
2003         int ret;
2004
2005         if (!root->orphan_block_rsv) {
2006                 block_rsv = btrfs_alloc_block_rsv(root);
2007                 if (!block_rsv)
2008                         return -ENOMEM;
2009         }
2010
2011         spin_lock(&root->orphan_lock);
2012         if (!root->orphan_block_rsv) {
2013                 root->orphan_block_rsv = block_rsv;
2014         } else if (block_rsv) {
2015                 btrfs_free_block_rsv(root, block_rsv);
2016                 block_rsv = NULL;
2017         }
2018
2019         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2020                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2021 #if 0
2022                 /*
2023                  * For proper ENOSPC handling, we should do orphan
2024                  * cleanup when mounting. But this introduces backward
2025                  * compatibility issue.
2026                  */
2027                 if (!xchg(&root->orphan_item_inserted, 1))
2028                         insert = 2;
2029                 else
2030                         insert = 1;
2031 #endif
2032                 insert = 1;
2033         }
2034
2035         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2036                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2037                 reserve = 1;
2038         }
2039         spin_unlock(&root->orphan_lock);
2040
2041         /* grab metadata reservation from transaction handle */
2042         if (reserve) {
2043                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2044                 BUG_ON(ret);
2045         }
2046
2047         /* insert an orphan item to track this unlinked/truncated file */
2048         if (insert >= 1) {
2049                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2050                 BUG_ON(ret && ret != -EEXIST);
2051         }
2052
2053         /* insert an orphan item to track subvolume contains orphan files */
2054         if (insert >= 2) {
2055                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2056                                                root->root_key.objectid);
2057                 BUG_ON(ret);
2058         }
2059         return 0;
2060 }
2061
2062 /*
2063  * We have done the truncate/delete so we can go ahead and remove the orphan
2064  * item for this particular inode.
2065  */
2066 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2067 {
2068         struct btrfs_root *root = BTRFS_I(inode)->root;
2069         int delete_item = 0;
2070         int release_rsv = 0;
2071         int ret = 0;
2072
2073         spin_lock(&root->orphan_lock);
2074         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2075                 list_del_init(&BTRFS_I(inode)->i_orphan);
2076                 delete_item = 1;
2077         }
2078
2079         if (BTRFS_I(inode)->orphan_meta_reserved) {
2080                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2081                 release_rsv = 1;
2082         }
2083         spin_unlock(&root->orphan_lock);
2084
2085         if (trans && delete_item) {
2086                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2087                 BUG_ON(ret);
2088         }
2089
2090         if (release_rsv)
2091                 btrfs_orphan_release_metadata(inode);
2092
2093         return 0;
2094 }
2095
2096 /*
2097  * this cleans up any orphans that may be left on the list from the last use
2098  * of this root.
2099  */
2100 int btrfs_orphan_cleanup(struct btrfs_root *root)
2101 {
2102         struct btrfs_path *path;
2103         struct extent_buffer *leaf;
2104         struct btrfs_key key, found_key;
2105         struct btrfs_trans_handle *trans;
2106         struct inode *inode;
2107         u64 last_objectid = 0;
2108         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2109
2110         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2111                 return 0;
2112
2113         path = btrfs_alloc_path();
2114         if (!path) {
2115                 ret = -ENOMEM;
2116                 goto out;
2117         }
2118         path->reada = -1;
2119
2120         key.objectid = BTRFS_ORPHAN_OBJECTID;
2121         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2122         key.offset = (u64)-1;
2123
2124         while (1) {
2125                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2126                 if (ret < 0)
2127                         goto out;
2128
2129                 /*
2130                  * if ret == 0 means we found what we were searching for, which
2131                  * is weird, but possible, so only screw with path if we didn't
2132                  * find the key and see if we have stuff that matches
2133                  */
2134                 if (ret > 0) {
2135                         ret = 0;
2136                         if (path->slots[0] == 0)
2137                                 break;
2138                         path->slots[0]--;
2139                 }
2140
2141                 /* pull out the item */
2142                 leaf = path->nodes[0];
2143                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2144
2145                 /* make sure the item matches what we want */
2146                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2147                         break;
2148                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2149                         break;
2150
2151                 /* release the path since we're done with it */
2152                 btrfs_release_path(path);
2153
2154                 /*
2155                  * this is where we are basically btrfs_lookup, without the
2156                  * crossing root thing.  we store the inode number in the
2157                  * offset of the orphan item.
2158                  */
2159
2160                 if (found_key.offset == last_objectid) {
2161                         printk(KERN_ERR "btrfs: Error removing orphan entry, "
2162                                "stopping orphan cleanup\n");
2163                         ret = -EINVAL;
2164                         goto out;
2165                 }
2166
2167                 last_objectid = found_key.offset;
2168
2169                 found_key.objectid = found_key.offset;
2170                 found_key.type = BTRFS_INODE_ITEM_KEY;
2171                 found_key.offset = 0;
2172                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2173                 ret = PTR_RET(inode);
2174                 if (ret && ret != -ESTALE)
2175                         goto out;
2176
2177                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2178                         struct btrfs_root *dead_root;
2179                         struct btrfs_fs_info *fs_info = root->fs_info;
2180                         int is_dead_root = 0;
2181
2182                         /*
2183                          * this is an orphan in the tree root. Currently these
2184                          * could come from 2 sources:
2185                          *  a) a snapshot deletion in progress
2186                          *  b) a free space cache inode
2187                          * We need to distinguish those two, as the snapshot
2188                          * orphan must not get deleted.
2189                          * find_dead_roots already ran before us, so if this
2190                          * is a snapshot deletion, we should find the root
2191                          * in the dead_roots list
2192                          */
2193                         spin_lock(&fs_info->trans_lock);
2194                         list_for_each_entry(dead_root, &fs_info->dead_roots,
2195                                             root_list) {
2196                                 if (dead_root->root_key.objectid ==
2197                                     found_key.objectid) {
2198                                         is_dead_root = 1;
2199                                         break;
2200                                 }
2201                         }
2202                         spin_unlock(&fs_info->trans_lock);
2203                         if (is_dead_root) {
2204                                 /* prevent this orphan from being found again */
2205                                 key.offset = found_key.objectid - 1;
2206                                 continue;
2207                         }
2208                 }
2209                 /*
2210                  * Inode is already gone but the orphan item is still there,
2211                  * kill the orphan item.
2212                  */
2213                 if (ret == -ESTALE) {
2214                         trans = btrfs_start_transaction(root, 1);
2215                         if (IS_ERR(trans)) {
2216                                 ret = PTR_ERR(trans);
2217                                 goto out;
2218                         }
2219                         ret = btrfs_del_orphan_item(trans, root,
2220                                                     found_key.objectid);
2221                         BUG_ON(ret);
2222                         btrfs_end_transaction(trans, root);
2223                         continue;
2224                 }
2225
2226                 /*
2227                  * add this inode to the orphan list so btrfs_orphan_del does
2228                  * the proper thing when we hit it
2229                  */
2230                 spin_lock(&root->orphan_lock);
2231                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2232                 spin_unlock(&root->orphan_lock);
2233
2234                 /* if we have links, this was a truncate, lets do that */
2235                 if (inode->i_nlink) {
2236                         if (!S_ISREG(inode->i_mode)) {
2237                                 WARN_ON(1);
2238                                 iput(inode);
2239                                 continue;
2240                         }
2241                         nr_truncate++;
2242                         ret = btrfs_truncate(inode);
2243                 } else {
2244                         nr_unlink++;
2245                 }
2246
2247                 /* this will do delete_inode and everything for us */
2248                 iput(inode);
2249                 if (ret)
2250                         goto out;
2251         }
2252         /* release the path since we're done with it */
2253         btrfs_release_path(path);
2254
2255         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2256
2257         if (root->orphan_block_rsv)
2258                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2259                                         (u64)-1);
2260
2261         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2262                 trans = btrfs_join_transaction(root);
2263                 if (!IS_ERR(trans))
2264                         btrfs_end_transaction(trans, root);
2265         }
2266
2267         if (nr_unlink)
2268                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2269         if (nr_truncate)
2270                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2271
2272 out:
2273         if (ret)
2274                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2275         btrfs_free_path(path);
2276         return ret;
2277 }
2278
2279 /*
2280  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2281  * don't find any xattrs, we know there can't be any acls.
2282  *
2283  * slot is the slot the inode is in, objectid is the objectid of the inode
2284  */
2285 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2286                                           int slot, u64 objectid)
2287 {
2288         u32 nritems = btrfs_header_nritems(leaf);
2289         struct btrfs_key found_key;
2290         int scanned = 0;
2291
2292         slot++;
2293         while (slot < nritems) {
2294                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2295
2296                 /* we found a different objectid, there must not be acls */
2297                 if (found_key.objectid != objectid)
2298                         return 0;
2299
2300                 /* we found an xattr, assume we've got an acl */
2301                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2302                         return 1;
2303
2304                 /*
2305                  * we found a key greater than an xattr key, there can't
2306                  * be any acls later on
2307                  */
2308                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2309                         return 0;
2310
2311                 slot++;
2312                 scanned++;
2313
2314                 /*
2315                  * it goes inode, inode backrefs, xattrs, extents,
2316                  * so if there are a ton of hard links to an inode there can
2317                  * be a lot of backrefs.  Don't waste time searching too hard,
2318                  * this is just an optimization
2319                  */
2320                 if (scanned >= 8)
2321                         break;
2322         }
2323         /* we hit the end of the leaf before we found an xattr or
2324          * something larger than an xattr.  We have to assume the inode
2325          * has acls
2326          */
2327         return 1;
2328 }
2329
2330 /*
2331  * read an inode from the btree into the in-memory inode
2332  */
2333 static void btrfs_read_locked_inode(struct inode *inode)
2334 {
2335         struct btrfs_path *path;
2336         struct extent_buffer *leaf;
2337         struct btrfs_inode_item *inode_item;
2338         struct btrfs_timespec *tspec;
2339         struct btrfs_root *root = BTRFS_I(inode)->root;
2340         struct btrfs_key location;
2341         int maybe_acls;
2342         u32 rdev;
2343         int ret;
2344         bool filled = false;
2345
2346         ret = btrfs_fill_inode(inode, &rdev);
2347         if (!ret)
2348                 filled = true;
2349
2350         path = btrfs_alloc_path();
2351         if (!path)
2352                 goto make_bad;
2353
2354         path->leave_spinning = 1;
2355         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2356
2357         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2358         if (ret)
2359                 goto make_bad;
2360
2361         leaf = path->nodes[0];
2362
2363         if (filled)
2364                 goto cache_acl;
2365
2366         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2367                                     struct btrfs_inode_item);
2368         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2369         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2370         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2371         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2372         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2373
2374         tspec = btrfs_inode_atime(inode_item);
2375         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2376         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2377
2378         tspec = btrfs_inode_mtime(inode_item);
2379         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2380         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2381
2382         tspec = btrfs_inode_ctime(inode_item);
2383         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2384         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2385
2386         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2387         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2388         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2389         inode->i_generation = BTRFS_I(inode)->generation;
2390         inode->i_rdev = 0;
2391         rdev = btrfs_inode_rdev(leaf, inode_item);
2392
2393         BTRFS_I(inode)->index_cnt = (u64)-1;
2394         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2395 cache_acl:
2396         /*
2397          * try to precache a NULL acl entry for files that don't have
2398          * any xattrs or acls
2399          */
2400         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2401                                            btrfs_ino(inode));
2402         if (!maybe_acls)
2403                 cache_no_acl(inode);
2404
2405         btrfs_free_path(path);
2406
2407         switch (inode->i_mode & S_IFMT) {
2408         case S_IFREG:
2409                 inode->i_mapping->a_ops = &btrfs_aops;
2410                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2411                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2412                 inode->i_fop = &btrfs_file_operations;
2413                 inode->i_op = &btrfs_file_inode_operations;
2414                 break;
2415         case S_IFDIR:
2416                 inode->i_fop = &btrfs_dir_file_operations;
2417                 if (root == root->fs_info->tree_root)
2418                         inode->i_op = &btrfs_dir_ro_inode_operations;
2419                 else
2420                         inode->i_op = &btrfs_dir_inode_operations;
2421                 break;
2422         case S_IFLNK:
2423                 inode->i_op = &btrfs_symlink_inode_operations;
2424                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2425                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2426                 break;
2427         default:
2428                 inode->i_op = &btrfs_special_inode_operations;
2429                 init_special_inode(inode, inode->i_mode, rdev);
2430                 break;
2431         }
2432
2433         btrfs_update_iflags(inode);
2434         return;
2435
2436 make_bad:
2437         btrfs_free_path(path);
2438         make_bad_inode(inode);
2439 }
2440
2441 /*
2442  * given a leaf and an inode, copy the inode fields into the leaf
2443  */
2444 static void fill_inode_item(struct btrfs_trans_handle *trans,
2445                             struct extent_buffer *leaf,
2446                             struct btrfs_inode_item *item,
2447                             struct inode *inode)
2448 {
2449         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2450         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2451         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2452         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2453         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2454
2455         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2456                                inode->i_atime.tv_sec);
2457         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2458                                 inode->i_atime.tv_nsec);
2459
2460         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2461                                inode->i_mtime.tv_sec);
2462         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2463                                 inode->i_mtime.tv_nsec);
2464
2465         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2466                                inode->i_ctime.tv_sec);
2467         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2468                                 inode->i_ctime.tv_nsec);
2469
2470         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2471         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2472         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2473         btrfs_set_inode_transid(leaf, item, trans->transid);
2474         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2475         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2476         btrfs_set_inode_block_group(leaf, item, 0);
2477 }
2478
2479 /*
2480  * copy everything in the in-memory inode into the btree.
2481  */
2482 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2483                                 struct btrfs_root *root, struct inode *inode)
2484 {
2485         struct btrfs_inode_item *inode_item;
2486         struct btrfs_path *path;
2487         struct extent_buffer *leaf;
2488         int ret;
2489
2490         path = btrfs_alloc_path();
2491         if (!path)
2492                 return -ENOMEM;
2493
2494         path->leave_spinning = 1;
2495         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2496                                  1);
2497         if (ret) {
2498                 if (ret > 0)
2499                         ret = -ENOENT;
2500                 goto failed;
2501         }
2502
2503         btrfs_unlock_up_safe(path, 1);
2504         leaf = path->nodes[0];
2505         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2506                                     struct btrfs_inode_item);
2507
2508         fill_inode_item(trans, leaf, inode_item, inode);
2509         btrfs_mark_buffer_dirty(leaf);
2510         btrfs_set_inode_last_trans(trans, inode);
2511         ret = 0;
2512 failed:
2513         btrfs_free_path(path);
2514         return ret;
2515 }
2516
2517 /*
2518  * copy everything in the in-memory inode into the btree.
2519  */
2520 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2521                                 struct btrfs_root *root, struct inode *inode)
2522 {
2523         int ret;
2524
2525         /*
2526          * If the inode is a free space inode, we can deadlock during commit
2527          * if we put it into the delayed code.
2528          *
2529          * The data relocation inode should also be directly updated
2530          * without delay
2531          */
2532         if (!btrfs_is_free_space_inode(root, inode)
2533             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2534                 ret = btrfs_delayed_update_inode(trans, root, inode);
2535                 if (!ret)
2536                         btrfs_set_inode_last_trans(trans, inode);
2537                 return ret;
2538         }
2539
2540         return btrfs_update_inode_item(trans, root, inode);
2541 }
2542
2543 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2544                                 struct btrfs_root *root, struct inode *inode)
2545 {
2546         int ret;
2547
2548         ret = btrfs_update_inode(trans, root, inode);
2549         if (ret == -ENOSPC)
2550                 return btrfs_update_inode_item(trans, root, inode);
2551         return ret;
2552 }
2553
2554 /*
2555  * unlink helper that gets used here in inode.c and in the tree logging
2556  * recovery code.  It remove a link in a directory with a given name, and
2557  * also drops the back refs in the inode to the directory
2558  */
2559 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2560                                 struct btrfs_root *root,
2561                                 struct inode *dir, struct inode *inode,
2562                                 const char *name, int name_len)
2563 {
2564         struct btrfs_path *path;
2565         int ret = 0;
2566         struct extent_buffer *leaf;
2567         struct btrfs_dir_item *di;
2568         struct btrfs_key key;
2569         u64 index;
2570         u64 ino = btrfs_ino(inode);
2571         u64 dir_ino = btrfs_ino(dir);
2572
2573         path = btrfs_alloc_path();
2574         if (!path) {
2575                 ret = -ENOMEM;
2576                 goto out;
2577         }
2578
2579         path->leave_spinning = 1;
2580         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2581                                     name, name_len, -1);
2582         if (IS_ERR(di)) {
2583                 ret = PTR_ERR(di);
2584                 goto err;
2585         }
2586         if (!di) {
2587                 ret = -ENOENT;
2588                 goto err;
2589         }
2590         leaf = path->nodes[0];
2591         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2592         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2593         if (ret)
2594                 goto err;
2595         btrfs_release_path(path);
2596
2597         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2598                                   dir_ino, &index);
2599         if (ret) {
2600                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2601                        "inode %llu parent %llu\n", name_len, name,
2602                        (unsigned long long)ino, (unsigned long long)dir_ino);
2603                 goto err;
2604         }
2605
2606         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2607         if (ret)
2608                 goto err;
2609
2610         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2611                                          inode, dir_ino);
2612         BUG_ON(ret != 0 && ret != -ENOENT);
2613
2614         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2615                                            dir, index);
2616         if (ret == -ENOENT)
2617                 ret = 0;
2618 err:
2619         btrfs_free_path(path);
2620         if (ret)
2621                 goto out;
2622
2623         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2624         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2625         btrfs_update_inode(trans, root, dir);
2626 out:
2627         return ret;
2628 }
2629
2630 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2631                        struct btrfs_root *root,
2632                        struct inode *dir, struct inode *inode,
2633                        const char *name, int name_len)
2634 {
2635         int ret;
2636         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2637         if (!ret) {
2638                 btrfs_drop_nlink(inode);
2639                 ret = btrfs_update_inode(trans, root, inode);
2640         }
2641         return ret;
2642 }
2643
2644
2645 /* helper to check if there is any shared block in the path */
2646 static int check_path_shared(struct btrfs_root *root,
2647                              struct btrfs_path *path)
2648 {
2649         struct extent_buffer *eb;
2650         int level;
2651         u64 refs = 1;
2652
2653         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2654                 int ret;
2655
2656                 if (!path->nodes[level])
2657                         break;
2658                 eb = path->nodes[level];
2659                 if (!btrfs_block_can_be_shared(root, eb))
2660                         continue;
2661                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2662                                                &refs, NULL);
2663                 if (refs > 1)
2664                         return 1;
2665         }
2666         return 0;
2667 }
2668
2669 /*
2670  * helper to start transaction for unlink and rmdir.
2671  *
2672  * unlink and rmdir are special in btrfs, they do not always free space.
2673  * so in enospc case, we should make sure they will free space before
2674  * allowing them to use the global metadata reservation.
2675  */
2676 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2677                                                        struct dentry *dentry)
2678 {
2679         struct btrfs_trans_handle *trans;
2680         struct btrfs_root *root = BTRFS_I(dir)->root;
2681         struct btrfs_path *path;
2682         struct btrfs_inode_ref *ref;
2683         struct btrfs_dir_item *di;
2684         struct inode *inode = dentry->d_inode;
2685         u64 index;
2686         int check_link = 1;
2687         int err = -ENOSPC;
2688         int ret;
2689         u64 ino = btrfs_ino(inode);
2690         u64 dir_ino = btrfs_ino(dir);
2691
2692         /*
2693          * 1 for the possible orphan item
2694          * 1 for the dir item
2695          * 1 for the dir index
2696          * 1 for the inode ref
2697          * 1 for the inode ref in the tree log
2698          * 2 for the dir entries in the log
2699          * 1 for the inode
2700          */
2701         trans = btrfs_start_transaction(root, 8);
2702         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2703                 return trans;
2704
2705         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2706                 return ERR_PTR(-ENOSPC);
2707
2708         /* check if there is someone else holds reference */
2709         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2710                 return ERR_PTR(-ENOSPC);
2711
2712         if (atomic_read(&inode->i_count) > 2)
2713                 return ERR_PTR(-ENOSPC);
2714
2715         if (xchg(&root->fs_info->enospc_unlink, 1))
2716                 return ERR_PTR(-ENOSPC);
2717
2718         path = btrfs_alloc_path();
2719         if (!path) {
2720                 root->fs_info->enospc_unlink = 0;
2721                 return ERR_PTR(-ENOMEM);
2722         }
2723
2724         /* 1 for the orphan item */
2725         trans = btrfs_start_transaction(root, 1);
2726         if (IS_ERR(trans)) {
2727                 btrfs_free_path(path);
2728                 root->fs_info->enospc_unlink = 0;
2729                 return trans;
2730         }
2731
2732         path->skip_locking = 1;
2733         path->search_commit_root = 1;
2734
2735         ret = btrfs_lookup_inode(trans, root, path,
2736                                 &BTRFS_I(dir)->location, 0);
2737         if (ret < 0) {
2738                 err = ret;
2739                 goto out;
2740         }
2741         if (ret == 0) {
2742                 if (check_path_shared(root, path))
2743                         goto out;
2744         } else {
2745                 check_link = 0;
2746         }
2747         btrfs_release_path(path);
2748
2749         ret = btrfs_lookup_inode(trans, root, path,
2750                                 &BTRFS_I(inode)->location, 0);
2751         if (ret < 0) {
2752                 err = ret;
2753                 goto out;
2754         }
2755         if (ret == 0) {
2756                 if (check_path_shared(root, path))
2757                         goto out;
2758         } else {
2759                 check_link = 0;
2760         }
2761         btrfs_release_path(path);
2762
2763         if (ret == 0 && S_ISREG(inode->i_mode)) {
2764                 ret = btrfs_lookup_file_extent(trans, root, path,
2765                                                ino, (u64)-1, 0);
2766                 if (ret < 0) {
2767                         err = ret;
2768                         goto out;
2769                 }
2770                 BUG_ON(ret == 0);
2771                 if (check_path_shared(root, path))
2772                         goto out;
2773                 btrfs_release_path(path);
2774         }
2775
2776         if (!check_link) {
2777                 err = 0;
2778                 goto out;
2779         }
2780
2781         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2782                                 dentry->d_name.name, dentry->d_name.len, 0);
2783         if (IS_ERR(di)) {
2784                 err = PTR_ERR(di);
2785                 goto out;
2786         }
2787         if (di) {
2788                 if (check_path_shared(root, path))
2789                         goto out;
2790         } else {
2791                 err = 0;
2792                 goto out;
2793         }
2794         btrfs_release_path(path);
2795
2796         ref = btrfs_lookup_inode_ref(trans, root, path,
2797                                 dentry->d_name.name, dentry->d_name.len,
2798                                 ino, dir_ino, 0);
2799         if (IS_ERR(ref)) {
2800                 err = PTR_ERR(ref);
2801                 goto out;
2802         }
2803         BUG_ON(!ref);
2804         if (check_path_shared(root, path))
2805                 goto out;
2806         index = btrfs_inode_ref_index(path->nodes[0], ref);
2807         btrfs_release_path(path);
2808
2809         /*
2810          * This is a commit root search, if we can lookup inode item and other
2811          * relative items in the commit root, it means the transaction of
2812          * dir/file creation has been committed, and the dir index item that we
2813          * delay to insert has also been inserted into the commit root. So
2814          * we needn't worry about the delayed insertion of the dir index item
2815          * here.
2816          */
2817         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
2818                                 dentry->d_name.name, dentry->d_name.len, 0);
2819         if (IS_ERR(di)) {
2820                 err = PTR_ERR(di);
2821                 goto out;
2822         }
2823         BUG_ON(ret == -ENOENT);
2824         if (check_path_shared(root, path))
2825                 goto out;
2826
2827         err = 0;
2828 out:
2829         btrfs_free_path(path);
2830         /* Migrate the orphan reservation over */
2831         if (!err)
2832                 err = btrfs_block_rsv_migrate(trans->block_rsv,
2833                                 &root->fs_info->global_block_rsv,
2834                                 trans->bytes_reserved);
2835
2836         if (err) {
2837                 btrfs_end_transaction(trans, root);
2838                 root->fs_info->enospc_unlink = 0;
2839                 return ERR_PTR(err);
2840         }
2841
2842         trans->block_rsv = &root->fs_info->global_block_rsv;
2843         return trans;
2844 }
2845
2846 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2847                                struct btrfs_root *root)
2848 {
2849         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2850                 btrfs_block_rsv_release(root, trans->block_rsv,
2851                                         trans->bytes_reserved);
2852                 trans->block_rsv = &root->fs_info->trans_block_rsv;
2853                 BUG_ON(!root->fs_info->enospc_unlink);
2854                 root->fs_info->enospc_unlink = 0;
2855         }
2856         btrfs_end_transaction(trans, root);
2857 }
2858
2859 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2860 {
2861         struct btrfs_root *root = BTRFS_I(dir)->root;
2862         struct btrfs_trans_handle *trans;
2863         struct inode *inode = dentry->d_inode;
2864         int ret;
2865         unsigned long nr = 0;
2866
2867         trans = __unlink_start_trans(dir, dentry);
2868         if (IS_ERR(trans))
2869                 return PTR_ERR(trans);
2870
2871         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2872
2873         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2874                                  dentry->d_name.name, dentry->d_name.len);
2875         if (ret)
2876                 goto out;
2877
2878         if (inode->i_nlink == 0) {
2879                 ret = btrfs_orphan_add(trans, inode);
2880                 if (ret)
2881                         goto out;
2882         }
2883
2884 out:
2885         nr = trans->blocks_used;
2886         __unlink_end_trans(trans, root);
2887         btrfs_btree_balance_dirty(root, nr);
2888         return ret;
2889 }
2890
2891 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2892                         struct btrfs_root *root,
2893                         struct inode *dir, u64 objectid,
2894                         const char *name, int name_len)
2895 {
2896         struct btrfs_path *path;
2897         struct extent_buffer *leaf;
2898         struct btrfs_dir_item *di;
2899         struct btrfs_key key;
2900         u64 index;
2901         int ret;
2902         u64 dir_ino = btrfs_ino(dir);
2903
2904         path = btrfs_alloc_path();
2905         if (!path)
2906                 return -ENOMEM;
2907
2908         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2909                                    name, name_len, -1);
2910         BUG_ON(IS_ERR_OR_NULL(di));
2911
2912         leaf = path->nodes[0];
2913         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2914         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2915         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2916         BUG_ON(ret);
2917         btrfs_release_path(path);
2918
2919         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2920                                  objectid, root->root_key.objectid,
2921                                  dir_ino, &index, name, name_len);
2922         if (ret < 0) {
2923                 BUG_ON(ret != -ENOENT);
2924                 di = btrfs_search_dir_index_item(root, path, dir_ino,
2925                                                  name, name_len);
2926                 BUG_ON(IS_ERR_OR_NULL(di));
2927
2928                 leaf = path->nodes[0];
2929                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2930                 btrfs_release_path(path);
2931                 index = key.offset;
2932         }
2933         btrfs_release_path(path);
2934
2935         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2936         BUG_ON(ret);
2937
2938         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2939         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2940         ret = btrfs_update_inode(trans, root, dir);
2941         BUG_ON(ret);
2942
2943         btrfs_free_path(path);
2944         return 0;
2945 }
2946
2947 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2948 {
2949         struct inode *inode = dentry->d_inode;
2950         int err = 0;
2951         struct btrfs_root *root = BTRFS_I(dir)->root;
2952         struct btrfs_trans_handle *trans;
2953         unsigned long nr = 0;
2954
2955         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2956             btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
2957                 return -ENOTEMPTY;
2958
2959         trans = __unlink_start_trans(dir, dentry);
2960         if (IS_ERR(trans))
2961                 return PTR_ERR(trans);
2962
2963         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
2964                 err = btrfs_unlink_subvol(trans, root, dir,
2965                                           BTRFS_I(inode)->location.objectid,
2966                                           dentry->d_name.name,
2967                                           dentry->d_name.len);
2968                 goto out;
2969         }
2970
2971         err = btrfs_orphan_add(trans, inode);
2972         if (err)
2973                 goto out;
2974
2975         /* now the directory is empty */
2976         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2977                                  dentry->d_name.name, dentry->d_name.len);
2978         if (!err)
2979                 btrfs_i_size_write(inode, 0);
2980 out:
2981         nr = trans->blocks_used;
2982         __unlink_end_trans(trans, root);
2983         btrfs_btree_balance_dirty(root, nr);
2984
2985         return err;
2986 }
2987
2988 /*
2989  * this can truncate away extent items, csum items and directory items.
2990  * It starts at a high offset and removes keys until it can't find
2991  * any higher than new_size
2992  *
2993  * csum items that cross the new i_size are truncated to the new size
2994  * as well.
2995  *
2996  * min_type is the minimum key type to truncate down to.  If set to 0, this
2997  * will kill all the items on this inode, including the INODE_ITEM_KEY.
2998  */
2999 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3000                                struct btrfs_root *root,
3001                                struct inode *inode,
3002                                u64 new_size, u32 min_type)
3003 {
3004         struct btrfs_path *path;
3005         struct extent_buffer *leaf;
3006         struct btrfs_file_extent_item *fi;
3007         struct btrfs_key key;
3008         struct btrfs_key found_key;
3009         u64 extent_start = 0;
3010         u64 extent_num_bytes = 0;
3011         u64 extent_offset = 0;
3012         u64 item_end = 0;
3013         u64 mask = root->sectorsize - 1;
3014         u32 found_type = (u8)-1;
3015         int found_extent;
3016         int del_item;
3017         int pending_del_nr = 0;
3018         int pending_del_slot = 0;
3019         int extent_type = -1;
3020         int ret;
3021         int err = 0;
3022         u64 ino = btrfs_ino(inode);
3023
3024         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3025
3026         path = btrfs_alloc_path();
3027         if (!path)
3028                 return -ENOMEM;
3029         path->reada = -1;
3030
3031         if (root->ref_cows || root == root->fs_info->tree_root)
3032                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3033
3034         /*
3035          * This function is also used to drop the items in the log tree before
3036          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3037          * it is used to drop the loged items. So we shouldn't kill the delayed
3038          * items.
3039          */
3040         if (min_type == 0 && root == BTRFS_I(inode)->root)
3041                 btrfs_kill_delayed_inode_items(inode);
3042
3043         key.objectid = ino;
3044         key.offset = (u64)-1;
3045         key.type = (u8)-1;
3046
3047 search_again:
3048         path->leave_spinning = 1;
3049         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3050         if (ret < 0) {
3051                 err = ret;
3052                 goto out;
3053         }
3054
3055         if (ret > 0) {
3056                 /* there are no items in the tree for us to truncate, we're
3057                  * done
3058                  */
3059                 if (path->slots[0] == 0)
3060                         goto out;
3061                 path->slots[0]--;
3062         }
3063
3064         while (1) {
3065                 fi = NULL;
3066                 leaf = path->nodes[0];
3067                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3068                 found_type = btrfs_key_type(&found_key);
3069
3070                 if (found_key.objectid != ino)
3071                         break;
3072
3073                 if (found_type < min_type)
3074                         break;
3075
3076                 item_end = found_key.offset;
3077                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3078                         fi = btrfs_item_ptr(leaf, path->slots[0],
3079                                             struct btrfs_file_extent_item);
3080                         extent_type = btrfs_file_extent_type(leaf, fi);
3081                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3082                                 item_end +=
3083                                     btrfs_file_extent_num_bytes(leaf, fi);
3084                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3085                                 item_end += btrfs_file_extent_inline_len(leaf,
3086                                                                          fi);
3087                         }
3088                         item_end--;
3089                 }
3090                 if (found_type > min_type) {
3091                         del_item = 1;
3092                 } else {
3093                         if (item_end < new_size)
3094                                 break;
3095                         if (found_key.offset >= new_size)
3096                                 del_item = 1;
3097                         else
3098                                 del_item = 0;
3099                 }
3100                 found_extent = 0;
3101                 /* FIXME, shrink the extent if the ref count is only 1 */
3102                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3103                         goto delete;
3104
3105                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3106                         u64 num_dec;
3107                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3108                         if (!del_item) {
3109                                 u64 orig_num_bytes =
3110                                         btrfs_file_extent_num_bytes(leaf, fi);
3111                                 extent_num_bytes = new_size -
3112                                         found_key.offset + root->sectorsize - 1;
3113                                 extent_num_bytes = extent_num_bytes &
3114                                         ~((u64)root->sectorsize - 1);
3115                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3116                                                          extent_num_bytes);
3117                                 num_dec = (orig_num_bytes -
3118                                            extent_num_bytes);
3119                                 if (root->ref_cows && extent_start != 0)
3120                                         inode_sub_bytes(inode, num_dec);
3121                                 btrfs_mark_buffer_dirty(leaf);
3122                         } else {
3123                                 extent_num_bytes =
3124                                         btrfs_file_extent_disk_num_bytes(leaf,
3125                                                                          fi);
3126                                 extent_offset = found_key.offset -
3127                                         btrfs_file_extent_offset(leaf, fi);
3128
3129                                 /* FIXME blocksize != 4096 */
3130                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3131                                 if (extent_start != 0) {
3132                                         found_extent = 1;
3133                                         if (root->ref_cows)
3134                                                 inode_sub_bytes(inode, num_dec);
3135                                 }
3136                         }
3137                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3138                         /*
3139                          * we can't truncate inline items that have had
3140                          * special encodings
3141                          */
3142                         if (!del_item &&
3143                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3144                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3145                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3146                                 u32 size = new_size - found_key.offset;
3147
3148                                 if (root->ref_cows) {
3149                                         inode_sub_bytes(inode, item_end + 1 -
3150                                                         new_size);
3151                                 }
3152                                 size =
3153                                     btrfs_file_extent_calc_inline_size(size);
3154                                 ret = btrfs_truncate_item(trans, root, path,
3155                                                           size, 1);
3156                         } else if (root->ref_cows) {
3157                                 inode_sub_bytes(inode, item_end + 1 -
3158                                                 found_key.offset);
3159                         }
3160                 }
3161 delete:
3162                 if (del_item) {
3163                         if (!pending_del_nr) {
3164                                 /* no pending yet, add ourselves */
3165                                 pending_del_slot = path->slots[0];
3166                                 pending_del_nr = 1;
3167                         } else if (pending_del_nr &&
3168                                    path->slots[0] + 1 == pending_del_slot) {
3169                                 /* hop on the pending chunk */
3170                                 pending_del_nr++;
3171                                 pending_del_slot = path->slots[0];
3172                         } else {
3173                                 BUG();
3174                         }
3175                 } else {
3176                         break;
3177                 }
3178                 if (found_extent && (root->ref_cows ||
3179                                      root == root->fs_info->tree_root)) {
3180                         btrfs_set_path_blocking(path);
3181                         ret = btrfs_free_extent(trans, root, extent_start,
3182                                                 extent_num_bytes, 0,
3183                                                 btrfs_header_owner(leaf),
3184                                                 ino, extent_offset, 0);
3185                         BUG_ON(ret);
3186                 }
3187
3188                 if (found_type == BTRFS_INODE_ITEM_KEY)
3189                         break;
3190
3191                 if (path->slots[0] == 0 ||
3192                     path->slots[0] != pending_del_slot) {
3193                         if (root->ref_cows &&
3194                             BTRFS_I(inode)->location.objectid !=
3195                                                 BTRFS_FREE_INO_OBJECTID) {
3196                                 err = -EAGAIN;
3197                                 goto out;
3198                         }
3199                         if (pending_del_nr) {
3200                                 ret = btrfs_del_items(trans, root, path,
3201                                                 pending_del_slot,
3202                                                 pending_del_nr);
3203                                 BUG_ON(ret);
3204                                 pending_del_nr = 0;
3205                         }
3206                         btrfs_release_path(path);
3207                         goto search_again;
3208                 } else {
3209                         path->slots[0]--;
3210                 }
3211         }
3212 out:
3213         if (pending_del_nr) {
3214                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3215                                       pending_del_nr);
3216                 BUG_ON(ret);
3217         }
3218         btrfs_free_path(path);
3219         return err;
3220 }
3221
3222 /*
3223  * taken from block_truncate_page, but does cow as it zeros out
3224  * any bytes left in the last page in the file.
3225  */
3226 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3227 {
3228         struct inode *inode = mapping->host;
3229         struct btrfs_root *root = BTRFS_I(inode)->root;
3230         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3231         struct btrfs_ordered_extent *ordered;
3232         struct extent_state *cached_state = NULL;
3233         char *kaddr;
3234         u32 blocksize = root->sectorsize;
3235         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3236         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3237         struct page *page;
3238         gfp_t mask = btrfs_alloc_write_mask(mapping);
3239         int ret = 0;
3240         u64 page_start;
3241         u64 page_end;
3242
3243         if ((offset & (blocksize - 1)) == 0)
3244                 goto out;
3245         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3246         if (ret)
3247                 goto out;
3248
3249         ret = -ENOMEM;
3250 again:
3251         page = find_or_create_page(mapping, index, mask);
3252         if (!page) {
3253                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3254                 goto out;
3255         }
3256
3257         page_start = page_offset(page);
3258         page_end = page_start + PAGE_CACHE_SIZE - 1;
3259
3260         if (!PageUptodate(page)) {
3261                 ret = btrfs_readpage(NULL, page);
3262                 lock_page(page);
3263                 if (page->mapping != mapping) {
3264                         unlock_page(page);
3265                         page_cache_release(page);
3266                         goto again;
3267                 }
3268                 if (!PageUptodate(page)) {
3269                         ret = -EIO;
3270                         goto out_unlock;
3271                 }
3272         }
3273         wait_on_page_writeback(page);
3274
3275         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3276                          GFP_NOFS);
3277         set_page_extent_mapped(page);
3278
3279         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3280         if (ordered) {
3281                 unlock_extent_cached(io_tree, page_start, page_end,
3282                                      &cached_state, GFP_NOFS);
3283                 unlock_page(page);
3284                 page_cache_release(page);
3285                 btrfs_start_ordered_extent(inode, ordered, 1);
3286                 btrfs_put_ordered_extent(ordered);
3287                 goto again;
3288         }
3289
3290         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3291                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3292                           0, 0, &cached_state, GFP_NOFS);
3293
3294         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3295                                         &cached_state);
3296         if (ret) {
3297                 unlock_extent_cached(io_tree, page_start, page_end,
3298                                      &cached_state, GFP_NOFS);
3299                 goto out_unlock;
3300         }
3301
3302         ret = 0;
3303         if (offset != PAGE_CACHE_SIZE) {
3304                 kaddr = kmap(page);
3305                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3306                 flush_dcache_page(page);
3307                 kunmap(page);
3308         }
3309         ClearPageChecked(page);
3310         set_page_dirty(page);
3311         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3312                              GFP_NOFS);
3313
3314 out_unlock:
3315         if (ret)
3316                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3317         unlock_page(page);
3318         page_cache_release(page);
3319 out:
3320         return ret;
3321 }
3322
3323 /*
3324  * This function puts in dummy file extents for the area we're creating a hole
3325  * for.  So if we are truncating this file to a larger size we need to insert
3326  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3327  * the range between oldsize and size
3328  */
3329 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3330 {
3331         struct btrfs_trans_handle *trans;
3332         struct btrfs_root *root = BTRFS_I(inode)->root;
3333         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3334         struct extent_map *em = NULL;
3335         struct extent_state *cached_state = NULL;
3336         u64 mask = root->sectorsize - 1;
3337         u64 hole_start = (oldsize + mask) & ~mask;
3338         u64 block_end = (size + mask) & ~mask;
3339         u64 last_byte;
3340         u64 cur_offset;
3341         u64 hole_size;
3342         int err = 0;
3343
3344         if (size <= hole_start)
3345                 return 0;
3346
3347         while (1) {
3348                 struct btrfs_ordered_extent *ordered;
3349                 btrfs_wait_ordered_range(inode, hole_start,
3350                                          block_end - hole_start);
3351                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3352                                  &cached_state, GFP_NOFS);
3353                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3354                 if (!ordered)
3355                         break;
3356                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3357                                      &cached_state, GFP_NOFS);
3358                 btrfs_put_ordered_extent(ordered);
3359         }
3360
3361         cur_offset = hole_start;
3362         while (1) {
3363                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3364                                 block_end - cur_offset, 0);
3365                 BUG_ON(IS_ERR_OR_NULL(em));
3366                 last_byte = min(extent_map_end(em), block_end);
3367                 last_byte = (last_byte + mask) & ~mask;
3368                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3369                         u64 hint_byte = 0;
3370                         hole_size = last_byte - cur_offset;
3371
3372                         trans = btrfs_start_transaction(root, 3);
3373                         if (IS_ERR(trans)) {
3374                                 err = PTR_ERR(trans);
3375                                 break;
3376                         }
3377
3378                         err = btrfs_drop_extents(trans, inode, cur_offset,
3379                                                  cur_offset + hole_size,
3380                                                  &hint_byte, 1);
3381                         if (err) {
3382                                 btrfs_update_inode(trans, root, inode);
3383                                 btrfs_end_transaction(trans, root);
3384                                 break;
3385                         }
3386
3387                         err = btrfs_insert_file_extent(trans, root,
3388                                         btrfs_ino(inode), cur_offset, 0,
3389                                         0, hole_size, 0, hole_size,
3390                                         0, 0, 0);
3391                         if (err) {
3392                                 btrfs_update_inode(trans, root, inode);
3393                                 btrfs_end_transaction(trans, root);
3394                                 break;
3395                         }
3396
3397                         btrfs_drop_extent_cache(inode, hole_start,
3398                                         last_byte - 1, 0);
3399
3400                         btrfs_update_inode(trans, root, inode);
3401                         btrfs_end_transaction(trans, root);
3402                 }
3403                 free_extent_map(em);
3404                 em = NULL;
3405                 cur_offset = last_byte;
3406                 if (cur_offset >= block_end)
3407                         break;
3408         }
3409
3410         free_extent_map(em);
3411         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3412                              GFP_NOFS);
3413         return err;
3414 }
3415
3416 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3417 {
3418         struct btrfs_root *root = BTRFS_I(inode)->root;
3419         struct btrfs_trans_handle *trans;
3420         loff_t oldsize = i_size_read(inode);
3421         int ret;
3422
3423         if (newsize == oldsize)
3424                 return 0;
3425
3426         if (newsize > oldsize) {
3427                 truncate_pagecache(inode, oldsize, newsize);
3428                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3429                 if (ret)
3430                         return ret;
3431
3432                 trans = btrfs_start_transaction(root, 1);
3433                 if (IS_ERR(trans))
3434                         return PTR_ERR(trans);
3435
3436                 i_size_write(inode, newsize);
3437                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3438                 ret = btrfs_update_inode(trans, root, inode);
3439                 btrfs_end_transaction(trans, root);
3440         } else {
3441
3442                 /*
3443                  * We're truncating a file that used to have good data down to
3444                  * zero. Make sure it gets into the ordered flush list so that
3445                  * any new writes get down to disk quickly.
3446                  */
3447                 if (newsize == 0)
3448                         BTRFS_I(inode)->ordered_data_close = 1;
3449
3450                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3451                 truncate_setsize(inode, newsize);
3452                 ret = btrfs_truncate(inode);
3453         }
3454
3455         return ret;
3456 }
3457
3458 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3459 {
3460         struct inode *inode = dentry->d_inode;
3461         struct btrfs_root *root = BTRFS_I(inode)->root;
3462         int err;
3463
3464         if (btrfs_root_readonly(root))
3465                 return -EROFS;
3466
3467         err = inode_change_ok(inode, attr);
3468         if (err)
3469                 return err;
3470
3471         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3472                 err = btrfs_setsize(inode, attr->ia_size);
3473                 if (err)
3474                         return err;
3475         }
3476
3477         if (attr->ia_valid) {
3478                 setattr_copy(inode, attr);
3479                 err = btrfs_dirty_inode(inode);
3480
3481                 if (!err && attr->ia_valid & ATTR_MODE)
3482                         err = btrfs_acl_chmod(inode);
3483         }
3484
3485         return err;
3486 }
3487
3488 void btrfs_evict_inode(struct inode *inode)
3489 {
3490         struct btrfs_trans_handle *trans;
3491         struct btrfs_root *root = BTRFS_I(inode)->root;
3492         struct btrfs_block_rsv *rsv, *global_rsv;
3493         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3494         unsigned long nr;
3495         int ret;
3496
3497         trace_btrfs_inode_evict(inode);
3498
3499         truncate_inode_pages(&inode->i_data, 0);
3500         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3501                                btrfs_is_free_space_inode(root, inode)))
3502                 goto no_delete;
3503
3504         if (is_bad_inode(inode)) {
3505                 btrfs_orphan_del(NULL, inode);
3506                 goto no_delete;
3507         }
3508         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3509         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3510
3511         if (root->fs_info->log_root_recovering) {
3512                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3513                 goto no_delete;
3514         }
3515
3516         if (inode->i_nlink > 0) {
3517                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3518                 goto no_delete;
3519         }
3520
3521         rsv = btrfs_alloc_block_rsv(root);
3522         if (!rsv) {
3523                 btrfs_orphan_del(NULL, inode);
3524                 goto no_delete;
3525         }
3526         rsv->size = min_size;
3527         global_rsv = &root->fs_info->global_block_rsv;
3528
3529         btrfs_i_size_write(inode, 0);
3530
3531         /*
3532          * This is a bit simpler than btrfs_truncate since
3533          *
3534          * 1) We've already reserved our space for our orphan item in the
3535          *    unlink.
3536          * 2) We're going to delete the inode item, so we don't need to update
3537          *    it at all.
3538          *
3539          * So we just need to reserve some slack space in case we add bytes when
3540          * doing the truncate.
3541          */
3542         while (1) {
3543                 ret = btrfs_block_rsv_refill_noflush(root, rsv, min_size);
3544
3545                 /*
3546                  * Try and steal from the global reserve since we will
3547                  * likely not use this space anyway, we want to try as
3548                  * hard as possible to get this to work.
3549                  */
3550                 if (ret)
3551                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3552
3553                 if (ret) {
3554                         printk(KERN_WARNING "Could not get space for a "
3555                                "delete, will truncate on mount %d\n", ret);
3556                         btrfs_orphan_del(NULL, inode);
3557                         btrfs_free_block_rsv(root, rsv);
3558                         goto no_delete;
3559                 }
3560
3561                 trans = btrfs_start_transaction(root, 0);
3562                 if (IS_ERR(trans)) {
3563                         btrfs_orphan_del(NULL, inode);
3564                         btrfs_free_block_rsv(root, rsv);
3565                         goto no_delete;
3566                 }
3567
3568                 trans->block_rsv = rsv;
3569
3570                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3571                 if (ret != -EAGAIN)
3572                         break;
3573
3574                 nr = trans->blocks_used;
3575                 btrfs_end_transaction(trans, root);
3576                 trans = NULL;
3577                 btrfs_btree_balance_dirty(root, nr);
3578         }
3579
3580         btrfs_free_block_rsv(root, rsv);
3581
3582         if (ret == 0) {
3583                 trans->block_rsv = root->orphan_block_rsv;
3584                 ret = btrfs_orphan_del(trans, inode);
3585                 BUG_ON(ret);
3586         }
3587
3588         trans->block_rsv = &root->fs_info->trans_block_rsv;
3589         if (!(root == root->fs_info->tree_root ||
3590               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3591                 btrfs_return_ino(root, btrfs_ino(inode));
3592
3593         nr = trans->blocks_used;
3594         btrfs_end_transaction(trans, root);
3595         btrfs_btree_balance_dirty(root, nr);
3596 no_delete:
3597         end_writeback(inode);
3598         return;
3599 }
3600
3601 /*
3602  * this returns the key found in the dir entry in the location pointer.
3603  * If no dir entries were found, location->objectid is 0.
3604  */
3605 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3606                                struct btrfs_key *location)
3607 {
3608         const char *name = dentry->d_name.name;
3609         int namelen = dentry->d_name.len;
3610         struct btrfs_dir_item *di;
3611         struct btrfs_path *path;
3612         struct btrfs_root *root = BTRFS_I(dir)->root;
3613         int ret = 0;
3614
3615         path = btrfs_alloc_path();
3616         if (!path)
3617                 return -ENOMEM;
3618
3619         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3620                                     namelen, 0);
3621         if (IS_ERR(di))
3622                 ret = PTR_ERR(di);
3623
3624         if (IS_ERR_OR_NULL(di))
3625                 goto out_err;
3626
3627         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3628 out:
3629         btrfs_free_path(path);
3630         return ret;
3631 out_err:
3632         location->objectid = 0;
3633         goto out;
3634 }
3635
3636 /*
3637  * when we hit a tree root in a directory, the btrfs part of the inode
3638  * needs to be changed to reflect the root directory of the tree root.  This
3639  * is kind of like crossing a mount point.
3640  */
3641 static int fixup_tree_root_location(struct btrfs_root *root,
3642                                     struct inode *dir,
3643                                     struct dentry *dentry,
3644                                     struct btrfs_key *location,
3645                                     struct btrfs_root **sub_root)
3646 {
3647         struct btrfs_path *path;
3648         struct btrfs_root *new_root;
3649         struct btrfs_root_ref *ref;
3650         struct extent_buffer *leaf;
3651         int ret;
3652         int err = 0;
3653
3654         path = btrfs_alloc_path();
3655         if (!path) {
3656                 err = -ENOMEM;
3657                 goto out;
3658         }
3659
3660         err = -ENOENT;
3661         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3662                                   BTRFS_I(dir)->root->root_key.objectid,
3663                                   location->objectid);
3664         if (ret) {
3665                 if (ret < 0)
3666                         err = ret;
3667                 goto out;
3668         }
3669
3670         leaf = path->nodes[0];
3671         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3672         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3673             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3674                 goto out;
3675
3676         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3677                                    (unsigned long)(ref + 1),
3678                                    dentry->d_name.len);
3679         if (ret)
3680                 goto out;
3681
3682         btrfs_release_path(path);
3683
3684         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3685         if (IS_ERR(new_root)) {
3686                 err = PTR_ERR(new_root);
3687                 goto out;
3688         }
3689
3690         if (btrfs_root_refs(&new_root->root_item) == 0) {
3691                 err = -ENOENT;
3692                 goto out;
3693         }
3694
3695         *sub_root = new_root;
3696         location->objectid = btrfs_root_dirid(&new_root->root_item);
3697         location->type = BTRFS_INODE_ITEM_KEY;
3698         location->offset = 0;
3699         err = 0;
3700 out:
3701         btrfs_free_path(path);
3702         return err;
3703 }
3704
3705 static void inode_tree_add(struct inode *inode)
3706 {
3707         struct btrfs_root *root = BTRFS_I(inode)->root;
3708         struct btrfs_inode *entry;
3709         struct rb_node **p;
3710         struct rb_node *parent;
3711         u64 ino = btrfs_ino(inode);
3712 again:
3713         p = &root->inode_tree.rb_node;
3714         parent = NULL;
3715
3716         if (inode_unhashed(inode))
3717                 return;
3718
3719         spin_lock(&root->inode_lock);
3720         while (*p) {
3721                 parent = *p;
3722                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3723
3724                 if (ino < btrfs_ino(&entry->vfs_inode))
3725                         p = &parent->rb_left;
3726                 else if (ino > btrfs_ino(&entry->vfs_inode))
3727                         p = &parent->rb_right;
3728                 else {
3729                         WARN_ON(!(entry->vfs_inode.i_state &
3730                                   (I_WILL_FREE | I_FREEING)));
3731                         rb_erase(parent, &root->inode_tree);
3732                         RB_CLEAR_NODE(parent);
3733                         spin_unlock(&root->inode_lock);
3734                         goto again;
3735                 }
3736         }
3737         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3738         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3739         spin_unlock(&root->inode_lock);
3740 }
3741
3742 static void inode_tree_del(struct inode *inode)
3743 {
3744         struct btrfs_root *root = BTRFS_I(inode)->root;
3745         int empty = 0;
3746
3747         spin_lock(&root->inode_lock);
3748         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3749                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3750                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3751                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3752         }
3753         spin_unlock(&root->inode_lock);
3754
3755         /*
3756          * Free space cache has inodes in the tree root, but the tree root has a
3757          * root_refs of 0, so this could end up dropping the tree root as a
3758          * snapshot, so we need the extra !root->fs_info->tree_root check to
3759          * make sure we don't drop it.
3760          */
3761         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3762             root != root->fs_info->tree_root) {
3763                 synchronize_srcu(&root->fs_info->subvol_srcu);
3764                 spin_lock(&root->inode_lock);
3765                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3766                 spin_unlock(&root->inode_lock);
3767                 if (empty)
3768                         btrfs_add_dead_root(root);
3769         }
3770 }
3771
3772 int btrfs_invalidate_inodes(struct btrfs_root *root)
3773 {
3774         struct rb_node *node;
3775         struct rb_node *prev;
3776         struct btrfs_inode *entry;
3777         struct inode *inode;
3778         u64 objectid = 0;
3779
3780         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3781
3782         spin_lock(&root->inode_lock);
3783 again:
3784         node = root->inode_tree.rb_node;
3785         prev = NULL;
3786         while (node) {
3787                 prev = node;
3788                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3789
3790                 if (objectid < btrfs_ino(&entry->vfs_inode))
3791                         node = node->rb_left;
3792                 else if (objectid > btrfs_ino(&entry->vfs_inode))
3793                         node = node->rb_right;
3794                 else
3795                         break;
3796         }
3797         if (!node) {
3798                 while (prev) {
3799                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
3800                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
3801                                 node = prev;
3802                                 break;
3803                         }
3804                         prev = rb_next(prev);
3805                 }
3806         }
3807         while (node) {
3808                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3809                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
3810                 inode = igrab(&entry->vfs_inode);
3811                 if (inode) {
3812                         spin_unlock(&root->inode_lock);
3813                         if (atomic_read(&inode->i_count) > 1)
3814                                 d_prune_aliases(inode);
3815                         /*
3816                          * btrfs_drop_inode will have it removed from
3817                          * the inode cache when its usage count
3818                          * hits zero.
3819                          */
3820                         iput(inode);
3821                         cond_resched();
3822                         spin_lock(&root->inode_lock);
3823                         goto again;
3824                 }
3825
3826                 if (cond_resched_lock(&root->inode_lock))
3827                         goto again;
3828
3829                 node = rb_next(node);
3830         }
3831         spin_unlock(&root->inode_lock);
3832         return 0;
3833 }
3834
3835 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3836 {
3837         struct btrfs_iget_args *args = p;
3838         inode->i_ino = args->ino;
3839         BTRFS_I(inode)->root = args->root;
3840         btrfs_set_inode_space_info(args->root, inode);
3841         return 0;
3842 }
3843
3844 static int btrfs_find_actor(struct inode *inode, void *opaque)
3845 {
3846         struct btrfs_iget_args *args = opaque;
3847         return args->ino == btrfs_ino(inode) &&
3848                 args->root == BTRFS_I(inode)->root;
3849 }
3850
3851 static struct inode *btrfs_iget_locked(struct super_block *s,
3852                                        u64 objectid,
3853                                        struct btrfs_root *root)
3854 {
3855         struct inode *inode;
3856         struct btrfs_iget_args args;
3857         args.ino = objectid;
3858         args.root = root;
3859
3860         inode = iget5_locked(s, objectid, btrfs_find_actor,
3861                              btrfs_init_locked_inode,
3862                              (void *)&args);
3863         return inode;
3864 }
3865
3866 /* Get an inode object given its location and corresponding root.
3867  * Returns in *is_new if the inode was read from disk
3868  */
3869 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3870                          struct btrfs_root *root, int *new)
3871 {
3872         struct inode *inode;
3873
3874         inode = btrfs_iget_locked(s, location->objectid, root);
3875         if (!inode)
3876                 return ERR_PTR(-ENOMEM);
3877
3878         if (inode->i_state & I_NEW) {
3879                 BTRFS_I(inode)->root = root;
3880                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3881                 btrfs_read_locked_inode(inode);
3882                 if (!is_bad_inode(inode)) {
3883                         inode_tree_add(inode);
3884                         unlock_new_inode(inode);
3885                         if (new)
3886                                 *new = 1;
3887                 } else {
3888                         unlock_new_inode(inode);
3889                         iput(inode);
3890                         inode = ERR_PTR(-ESTALE);
3891                 }
3892         }
3893
3894         return inode;
3895 }
3896
3897 static struct inode *new_simple_dir(struct super_block *s,
3898                                     struct btrfs_key *key,
3899                                     struct btrfs_root *root)
3900 {
3901         struct inode *inode = new_inode(s);
3902
3903         if (!inode)
3904                 return ERR_PTR(-ENOMEM);
3905
3906         BTRFS_I(inode)->root = root;
3907         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3908         BTRFS_I(inode)->dummy_inode = 1;
3909
3910         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3911         inode->i_op = &simple_dir_inode_operations;
3912         inode->i_fop = &simple_dir_operations;
3913         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3914         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3915
3916         return inode;
3917 }
3918
3919 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3920 {
3921         struct inode *inode;
3922         struct btrfs_root *root = BTRFS_I(dir)->root;
3923         struct btrfs_root *sub_root = root;
3924         struct btrfs_key location;
3925         int index;
3926         int ret = 0;
3927
3928         if (dentry->d_name.len > BTRFS_NAME_LEN)
3929                 return ERR_PTR(-ENAMETOOLONG);
3930
3931         if (unlikely(d_need_lookup(dentry))) {
3932                 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
3933                 kfree(dentry->d_fsdata);
3934                 dentry->d_fsdata = NULL;
3935                 /* This thing is hashed, drop it for now */
3936                 d_drop(dentry);
3937         } else {
3938                 ret = btrfs_inode_by_name(dir, dentry, &location);
3939         }
3940
3941         if (ret < 0)
3942                 return ERR_PTR(ret);
3943
3944         if (location.objectid == 0)
3945                 return NULL;
3946
3947         if (location.type == BTRFS_INODE_ITEM_KEY) {
3948                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
3949                 return inode;
3950         }
3951
3952         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
3953
3954         index = srcu_read_lock(&root->fs_info->subvol_srcu);
3955         ret = fixup_tree_root_location(root, dir, dentry,
3956                                        &location, &sub_root);
3957         if (ret < 0) {
3958                 if (ret != -ENOENT)
3959                         inode = ERR_PTR(ret);
3960                 else
3961                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
3962         } else {
3963                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
3964         }
3965         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
3966
3967         if (!IS_ERR(inode) && root != sub_root) {
3968                 down_read(&root->fs_info->cleanup_work_sem);
3969                 if (!(inode->i_sb->s_flags & MS_RDONLY))
3970                         ret = btrfs_orphan_cleanup(sub_root);
3971                 up_read(&root->fs_info->cleanup_work_sem);
3972                 if (ret)
3973                         inode = ERR_PTR(ret);
3974         }
3975
3976         return inode;
3977 }
3978
3979 static int btrfs_dentry_delete(const struct dentry *dentry)
3980 {
3981         struct btrfs_root *root;
3982
3983         if (!dentry->d_inode && !IS_ROOT(dentry))
3984                 dentry = dentry->d_parent;
3985
3986         if (dentry->d_inode) {
3987                 root = BTRFS_I(dentry->d_inode)->root;
3988                 if (btrfs_root_refs(&root->root_item) == 0)
3989                         return 1;
3990         }
3991         return 0;
3992 }
3993
3994 static void btrfs_dentry_release(struct dentry *dentry)
3995 {
3996         if (dentry->d_fsdata)
3997                 kfree(dentry->d_fsdata);
3998 }
3999
4000 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4001                                    struct nameidata *nd)
4002 {
4003         struct dentry *ret;
4004
4005         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4006         if (unlikely(d_need_lookup(dentry))) {
4007                 spin_lock(&dentry->d_lock);
4008                 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4009                 spin_unlock(&dentry->d_lock);
4010         }
4011         return ret;
4012 }
4013
4014 unsigned char btrfs_filetype_table[] = {
4015         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4016 };
4017
4018 static int btrfs_real_readdir(struct file *filp, void *dirent,
4019                               filldir_t filldir)
4020 {
4021         struct inode *inode = filp->f_dentry->d_inode;
4022         struct btrfs_root *root = BTRFS_I(inode)->root;
4023         struct btrfs_item *item;
4024         struct btrfs_dir_item *di;
4025         struct btrfs_key key;
4026         struct btrfs_key found_key;
4027         struct btrfs_path *path;
4028         struct list_head ins_list;
4029         struct list_head del_list;
4030         struct qstr q;
4031         int ret;
4032         struct extent_buffer *leaf;
4033         int slot;
4034         unsigned char d_type;
4035         int over = 0;
4036         u32 di_cur;
4037         u32 di_total;
4038         u32 di_len;
4039         int key_type = BTRFS_DIR_INDEX_KEY;
4040         char tmp_name[32];
4041         char *name_ptr;
4042         int name_len;
4043         int is_curr = 0;        /* filp->f_pos points to the current index? */
4044
4045         /* FIXME, use a real flag for deciding about the key type */
4046         if (root->fs_info->tree_root == root)
4047                 key_type = BTRFS_DIR_ITEM_KEY;
4048
4049         /* special case for "." */
4050         if (filp->f_pos == 0) {
4051                 over = filldir(dirent, ".", 1,
4052                                filp->f_pos, btrfs_ino(inode), DT_DIR);
4053                 if (over)
4054                         return 0;
4055                 filp->f_pos = 1;
4056         }
4057         /* special case for .., just use the back ref */
4058         if (filp->f_pos == 1) {
4059                 u64 pino = parent_ino(filp->f_path.dentry);
4060                 over = filldir(dirent, "..", 2,
4061                                filp->f_pos, pino, DT_DIR);
4062                 if (over)
4063                         return 0;
4064                 filp->f_pos = 2;
4065         }
4066         path = btrfs_alloc_path();
4067         if (!path)
4068                 return -ENOMEM;
4069
4070         path->reada = 1;
4071
4072         if (key_type == BTRFS_DIR_INDEX_KEY) {
4073                 INIT_LIST_HEAD(&ins_list);
4074                 INIT_LIST_HEAD(&del_list);
4075                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4076         }
4077
4078         btrfs_set_key_type(&key, key_type);
4079         key.offset = filp->f_pos;
4080         key.objectid = btrfs_ino(inode);
4081
4082         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4083         if (ret < 0)
4084                 goto err;
4085
4086         while (1) {
4087                 leaf = path->nodes[0];
4088                 slot = path->slots[0];
4089                 if (slot >= btrfs_header_nritems(leaf)) {
4090                         ret = btrfs_next_leaf(root, path);
4091                         if (ret < 0)
4092                                 goto err;
4093                         else if (ret > 0)
4094                                 break;
4095                         continue;
4096                 }
4097
4098                 item = btrfs_item_nr(leaf, slot);
4099                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4100
4101                 if (found_key.objectid != key.objectid)
4102                         break;
4103                 if (btrfs_key_type(&found_key) != key_type)
4104                         break;
4105                 if (found_key.offset < filp->f_pos)
4106                         goto next;
4107                 if (key_type == BTRFS_DIR_INDEX_KEY &&
4108                     btrfs_should_delete_dir_index(&del_list,
4109                                                   found_key.offset))
4110                         goto next;
4111
4112                 filp->f_pos = found_key.offset;
4113                 is_curr = 1;
4114
4115                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4116                 di_cur = 0;
4117                 di_total = btrfs_item_size(leaf, item);
4118
4119                 while (di_cur < di_total) {
4120                         struct btrfs_key location;
4121                         struct dentry *tmp;
4122
4123                         if (verify_dir_item(root, leaf, di))
4124                                 break;
4125
4126                         name_len = btrfs_dir_name_len(leaf, di);
4127                         if (name_len <= sizeof(tmp_name)) {
4128                                 name_ptr = tmp_name;
4129                         } else {
4130                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4131                                 if (!name_ptr) {
4132                                         ret = -ENOMEM;
4133                                         goto err;
4134                                 }
4135                         }
4136                         read_extent_buffer(leaf, name_ptr,
4137                                            (unsigned long)(di + 1), name_len);
4138
4139                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4140                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4141
4142                         q.name = name_ptr;
4143                         q.len = name_len;
4144                         q.hash = full_name_hash(q.name, q.len);
4145                         tmp = d_lookup(filp->f_dentry, &q);
4146                         if (!tmp) {
4147                                 struct btrfs_key *newkey;
4148
4149                                 newkey = kzalloc(sizeof(struct btrfs_key),
4150                                                  GFP_NOFS);
4151                                 if (!newkey)
4152                                         goto no_dentry;
4153                                 tmp = d_alloc(filp->f_dentry, &q);
4154                                 if (!tmp) {
4155                                         kfree(newkey);
4156                                         dput(tmp);
4157                                         goto no_dentry;
4158                                 }
4159                                 memcpy(newkey, &location,
4160                                        sizeof(struct btrfs_key));
4161                                 tmp->d_fsdata = newkey;
4162                                 tmp->d_flags |= DCACHE_NEED_LOOKUP;
4163                                 d_rehash(tmp);
4164                                 dput(tmp);
4165                         } else {
4166                                 dput(tmp);
4167                         }
4168 no_dentry:
4169                         /* is this a reference to our own snapshot? If so
4170                          * skip it
4171                          */
4172                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4173                             location.objectid == root->root_key.objectid) {
4174                                 over = 0;
4175                                 goto skip;
4176                         }
4177                         over = filldir(dirent, name_ptr, name_len,
4178                                        found_key.offset, location.objectid,
4179                                        d_type);
4180
4181 skip:
4182                         if (name_ptr != tmp_name)
4183                                 kfree(name_ptr);
4184
4185                         if (over)
4186                                 goto nopos;
4187                         di_len = btrfs_dir_name_len(leaf, di) +
4188                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4189                         di_cur += di_len;
4190                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4191                 }
4192 next:
4193                 path->slots[0]++;
4194         }
4195
4196         if (key_type == BTRFS_DIR_INDEX_KEY) {
4197                 if (is_curr)
4198                         filp->f_pos++;
4199                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4200                                                       &ins_list);
4201                 if (ret)
4202                         goto nopos;
4203         }
4204
4205         /* Reached end of directory/root. Bump pos past the last item. */
4206         if (key_type == BTRFS_DIR_INDEX_KEY)
4207                 /*
4208                  * 32-bit glibc will use getdents64, but then strtol -
4209                  * so the last number we can serve is this.
4210                  */
4211                 filp->f_pos = 0x7fffffff;
4212         else
4213                 filp->f_pos++;
4214 nopos:
4215         ret = 0;
4216 err:
4217         if (key_type == BTRFS_DIR_INDEX_KEY)
4218                 btrfs_put_delayed_items(&ins_list, &del_list);
4219         btrfs_free_path(path);
4220         return ret;
4221 }
4222
4223 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4224 {
4225         struct btrfs_root *root = BTRFS_I(inode)->root;
4226         struct btrfs_trans_handle *trans;
4227         int ret = 0;
4228         bool nolock = false;
4229
4230         if (BTRFS_I(inode)->dummy_inode)
4231                 return 0;
4232
4233         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
4234                 nolock = true;
4235
4236         if (wbc->sync_mode == WB_SYNC_ALL) {
4237                 if (nolock)
4238                         trans = btrfs_join_transaction_nolock(root);
4239                 else
4240                         trans = btrfs_join_transaction(root);
4241                 if (IS_ERR(trans))
4242                         return PTR_ERR(trans);
4243                 if (nolock)
4244                         ret = btrfs_end_transaction_nolock(trans, root);
4245                 else
4246                         ret = btrfs_commit_transaction(trans, root);
4247         }
4248         return ret;
4249 }
4250
4251 /*
4252  * This is somewhat expensive, updating the tree every time the
4253  * inode changes.  But, it is most likely to find the inode in cache.
4254  * FIXME, needs more benchmarking...there are no reasons other than performance
4255  * to keep or drop this code.
4256  */
4257 int btrfs_dirty_inode(struct inode *inode)
4258 {
4259         struct btrfs_root *root = BTRFS_I(inode)->root;
4260         struct btrfs_trans_handle *trans;
4261         int ret;
4262
4263         if (BTRFS_I(inode)->dummy_inode)
4264                 return 0;
4265
4266         trans = btrfs_join_transaction(root);
4267         if (IS_ERR(trans))
4268                 return PTR_ERR(trans);
4269
4270         ret = btrfs_update_inode(trans, root, inode);
4271         if (ret && ret == -ENOSPC) {
4272                 /* whoops, lets try again with the full transaction */
4273                 btrfs_end_transaction(trans, root);
4274                 trans = btrfs_start_transaction(root, 1);
4275                 if (IS_ERR(trans))
4276                         return PTR_ERR(trans);
4277
4278                 ret = btrfs_update_inode(trans, root, inode);
4279         }
4280         btrfs_end_transaction(trans, root);
4281         if (BTRFS_I(inode)->delayed_node)
4282                 btrfs_balance_delayed_items(root);
4283
4284         return ret;
4285 }
4286
4287 /*
4288  * This is a copy of file_update_time.  We need this so we can return error on
4289  * ENOSPC for updating the inode in the case of file write and mmap writes.
4290  */
4291 int btrfs_update_time(struct file *file)
4292 {
4293         struct inode *inode = file->f_path.dentry->d_inode;
4294         struct timespec now;
4295         int ret;
4296         enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
4297
4298         /* First try to exhaust all avenues to not sync */
4299         if (IS_NOCMTIME(inode))
4300                 return 0;
4301
4302         now = current_fs_time(inode->i_sb);
4303         if (!timespec_equal(&inode->i_mtime, &now))
4304                 sync_it = S_MTIME;
4305
4306         if (!timespec_equal(&inode->i_ctime, &now))
4307                 sync_it |= S_CTIME;
4308
4309         if (IS_I_VERSION(inode))
4310                 sync_it |= S_VERSION;
4311
4312         if (!sync_it)
4313                 return 0;
4314
4315         /* Finally allowed to write? Takes lock. */
4316         if (mnt_want_write_file(file))
4317                 return 0;
4318
4319         /* Only change inode inside the lock region */
4320         if (sync_it & S_VERSION)
4321                 inode_inc_iversion(inode);
4322         if (sync_it & S_CTIME)
4323                 inode->i_ctime = now;
4324         if (sync_it & S_MTIME)
4325                 inode->i_mtime = now;
4326         ret = btrfs_dirty_inode(inode);
4327         if (!ret)
4328                 mark_inode_dirty_sync(inode);
4329         mnt_drop_write(file->f_path.mnt);
4330         return ret;
4331 }
4332
4333 /*
4334  * find the highest existing sequence number in a directory
4335  * and then set the in-memory index_cnt variable to reflect
4336  * free sequence numbers
4337  */
4338 static int btrfs_set_inode_index_count(struct inode *inode)
4339 {
4340         struct btrfs_root *root = BTRFS_I(inode)->root;
4341         struct btrfs_key key, found_key;
4342         struct btrfs_path *path;
4343         struct extent_buffer *leaf;
4344         int ret;
4345
4346         key.objectid = btrfs_ino(inode);
4347         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4348         key.offset = (u64)-1;
4349
4350         path = btrfs_alloc_path();
4351         if (!path)
4352                 return -ENOMEM;
4353
4354         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4355         if (ret < 0)
4356                 goto out;
4357         /* FIXME: we should be able to handle this */
4358         if (ret == 0)
4359                 goto out;
4360         ret = 0;
4361
4362         /*
4363          * MAGIC NUMBER EXPLANATION:
4364          * since we search a directory based on f_pos we have to start at 2
4365          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4366          * else has to start at 2
4367          */
4368         if (path->slots[0] == 0) {
4369                 BTRFS_I(inode)->index_cnt = 2;
4370                 goto out;
4371         }
4372
4373         path->slots[0]--;
4374
4375         leaf = path->nodes[0];
4376         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4377
4378         if (found_key.objectid != btrfs_ino(inode) ||
4379             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4380                 BTRFS_I(inode)->index_cnt = 2;
4381                 goto out;
4382         }
4383
4384         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4385 out:
4386         btrfs_free_path(path);
4387         return ret;
4388 }
4389
4390 /*
4391  * helper to find a free sequence number in a given directory.  This current
4392  * code is very simple, later versions will do smarter things in the btree
4393  */
4394 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4395 {
4396         int ret = 0;
4397
4398         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4399                 ret = btrfs_inode_delayed_dir_index_count(dir);
4400                 if (ret) {
4401                         ret = btrfs_set_inode_index_count(dir);
4402                         if (ret)
4403                                 return ret;
4404                 }
4405         }
4406
4407         *index = BTRFS_I(dir)->index_cnt;
4408         BTRFS_I(dir)->index_cnt++;
4409
4410         return ret;
4411 }
4412
4413 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4414                                      struct btrfs_root *root,
4415                                      struct inode *dir,
4416                                      const char *name, int name_len,
4417                                      u64 ref_objectid, u64 objectid,
4418                                      umode_t mode, u64 *index)
4419 {
4420         struct inode *inode;
4421         struct btrfs_inode_item *inode_item;
4422         struct btrfs_key *location;
4423         struct btrfs_path *path;
4424         struct btrfs_inode_ref *ref;
4425         struct btrfs_key key[2];
4426         u32 sizes[2];
4427         unsigned long ptr;
4428         int ret;
4429         int owner;
4430
4431         path = btrfs_alloc_path();
4432         if (!path)
4433                 return ERR_PTR(-ENOMEM);
4434
4435         inode = new_inode(root->fs_info->sb);
4436         if (!inode) {
4437                 btrfs_free_path(path);
4438                 return ERR_PTR(-ENOMEM);
4439         }
4440
4441         /*
4442          * we have to initialize this early, so we can reclaim the inode
4443          * number if we fail afterwards in this function.
4444          */
4445         inode->i_ino = objectid;
4446
4447         if (dir) {
4448                 trace_btrfs_inode_request(dir);
4449
4450                 ret = btrfs_set_inode_index(dir, index);
4451                 if (ret) {
4452                         btrfs_free_path(path);
4453                         iput(inode);
4454                         return ERR_PTR(ret);
4455                 }
4456         }
4457         /*
4458          * index_cnt is ignored for everything but a dir,
4459          * btrfs_get_inode_index_count has an explanation for the magic
4460          * number
4461          */
4462         BTRFS_I(inode)->index_cnt = 2;
4463         BTRFS_I(inode)->root = root;
4464         BTRFS_I(inode)->generation = trans->transid;
4465         inode->i_generation = BTRFS_I(inode)->generation;
4466         btrfs_set_inode_space_info(root, inode);
4467
4468         if (S_ISDIR(mode))
4469                 owner = 0;
4470         else
4471                 owner = 1;
4472
4473         key[0].objectid = objectid;
4474         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4475         key[0].offset = 0;
4476
4477         key[1].objectid = objectid;
4478         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4479         key[1].offset = ref_objectid;
4480
4481         sizes[0] = sizeof(struct btrfs_inode_item);
4482         sizes[1] = name_len + sizeof(*ref);
4483
4484         path->leave_spinning = 1;
4485         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4486         if (ret != 0)
4487                 goto fail;
4488
4489         inode_init_owner(inode, dir, mode);
4490         inode_set_bytes(inode, 0);
4491         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4492         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4493                                   struct btrfs_inode_item);
4494         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4495
4496         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4497                              struct btrfs_inode_ref);
4498         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4499         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4500         ptr = (unsigned long)(ref + 1);
4501         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4502
4503         btrfs_mark_buffer_dirty(path->nodes[0]);
4504         btrfs_free_path(path);
4505
4506         location = &BTRFS_I(inode)->location;
4507         location->objectid = objectid;
4508         location->offset = 0;
4509         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4510
4511         btrfs_inherit_iflags(inode, dir);
4512
4513         if (S_ISREG(mode)) {
4514                 if (btrfs_test_opt(root, NODATASUM))
4515                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4516                 if (btrfs_test_opt(root, NODATACOW) ||
4517                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4518                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4519         }
4520
4521         insert_inode_hash(inode);
4522         inode_tree_add(inode);
4523
4524         trace_btrfs_inode_new(inode);
4525         btrfs_set_inode_last_trans(trans, inode);
4526
4527         return inode;
4528 fail:
4529         if (dir)
4530                 BTRFS_I(dir)->index_cnt--;
4531         btrfs_free_path(path);
4532         iput(inode);
4533         return ERR_PTR(ret);
4534 }
4535
4536 static inline u8 btrfs_inode_type(struct inode *inode)
4537 {
4538         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4539 }
4540
4541 /*
4542  * utility function to add 'inode' into 'parent_inode' with
4543  * a give name and a given sequence number.
4544  * if 'add_backref' is true, also insert a backref from the
4545  * inode to the parent directory.
4546  */
4547 int btrfs_add_link(struct btrfs_trans_handle *trans,
4548                    struct inode *parent_inode, struct inode *inode,
4549                    const char *name, int name_len, int add_backref, u64 index)
4550 {
4551         int ret = 0;
4552         struct btrfs_key key;
4553         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4554         u64 ino = btrfs_ino(inode);
4555         u64 parent_ino = btrfs_ino(parent_inode);
4556
4557         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4558                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4559         } else {
4560                 key.objectid = ino;
4561                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4562                 key.offset = 0;
4563         }
4564
4565         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4566                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4567                                          key.objectid, root->root_key.objectid,
4568                                          parent_ino, index, name, name_len);
4569         } else if (add_backref) {
4570                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4571                                              parent_ino, index);
4572         }
4573
4574         if (ret == 0) {
4575                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4576                                             parent_inode, &key,
4577                                             btrfs_inode_type(inode), index);
4578                 BUG_ON(ret);
4579
4580                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4581                                    name_len * 2);
4582                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4583                 ret = btrfs_update_inode(trans, root, parent_inode);
4584         }
4585         return ret;
4586 }
4587
4588 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4589                             struct inode *dir, struct dentry *dentry,
4590                             struct inode *inode, int backref, u64 index)
4591 {
4592         int err = btrfs_add_link(trans, dir, inode,
4593                                  dentry->d_name.name, dentry->d_name.len,
4594                                  backref, index);
4595         if (err > 0)
4596                 err = -EEXIST;
4597         return err;
4598 }
4599
4600 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4601                         umode_t mode, dev_t rdev)
4602 {
4603         struct btrfs_trans_handle *trans;
4604         struct btrfs_root *root = BTRFS_I(dir)->root;
4605         struct inode *inode = NULL;
4606         int err;
4607         int drop_inode = 0;
4608         u64 objectid;
4609         unsigned long nr = 0;
4610         u64 index = 0;
4611
4612         if (!new_valid_dev(rdev))
4613                 return -EINVAL;
4614
4615         /*
4616          * 2 for inode item and ref
4617          * 2 for dir items
4618          * 1 for xattr if selinux is on
4619          */
4620         trans = btrfs_start_transaction(root, 5);
4621         if (IS_ERR(trans))
4622                 return PTR_ERR(trans);
4623
4624         err = btrfs_find_free_ino(root, &objectid);
4625         if (err)
4626                 goto out_unlock;
4627
4628         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4629                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4630                                 mode, &index);
4631         if (IS_ERR(inode)) {
4632                 err = PTR_ERR(inode);
4633                 goto out_unlock;
4634         }
4635
4636         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4637         if (err) {
4638                 drop_inode = 1;
4639                 goto out_unlock;
4640         }
4641
4642         /*
4643         * If the active LSM wants to access the inode during
4644         * d_instantiate it needs these. Smack checks to see
4645         * if the filesystem supports xattrs by looking at the
4646         * ops vector.
4647         */
4648
4649         inode->i_op = &btrfs_special_inode_operations;
4650         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4651         if (err)
4652                 drop_inode = 1;
4653         else {
4654                 init_special_inode(inode, inode->i_mode, rdev);
4655                 btrfs_update_inode(trans, root, inode);
4656                 d_instantiate(dentry, inode);
4657         }
4658 out_unlock:
4659         nr = trans->blocks_used;
4660         btrfs_end_transaction(trans, root);
4661         btrfs_btree_balance_dirty(root, nr);
4662         if (drop_inode) {
4663                 inode_dec_link_count(inode);
4664                 iput(inode);
4665         }
4666         return err;
4667 }
4668
4669 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4670                         umode_t mode, struct nameidata *nd)
4671 {
4672         struct btrfs_trans_handle *trans;
4673         struct btrfs_root *root = BTRFS_I(dir)->root;
4674         struct inode *inode = NULL;
4675         int drop_inode = 0;
4676         int err;
4677         unsigned long nr = 0;
4678         u64 objectid;
4679         u64 index = 0;
4680
4681         /*
4682          * 2 for inode item and ref
4683          * 2 for dir items
4684          * 1 for xattr if selinux is on
4685          */
4686         trans = btrfs_start_transaction(root, 5);
4687         if (IS_ERR(trans))
4688                 return PTR_ERR(trans);
4689
4690         err = btrfs_find_free_ino(root, &objectid);
4691         if (err)
4692                 goto out_unlock;
4693
4694         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4695                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4696                                 mode, &index);
4697         if (IS_ERR(inode)) {
4698                 err = PTR_ERR(inode);
4699                 goto out_unlock;
4700         }
4701
4702         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4703         if (err) {
4704                 drop_inode = 1;
4705                 goto out_unlock;
4706         }
4707
4708         /*
4709         * If the active LSM wants to access the inode during
4710         * d_instantiate it needs these. Smack checks to see
4711         * if the filesystem supports xattrs by looking at the
4712         * ops vector.
4713         */
4714         inode->i_fop = &btrfs_file_operations;
4715         inode->i_op = &btrfs_file_inode_operations;
4716
4717         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4718         if (err)
4719                 drop_inode = 1;
4720         else {
4721                 inode->i_mapping->a_ops = &btrfs_aops;
4722                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4723                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4724                 d_instantiate(dentry, inode);
4725         }
4726 out_unlock:
4727         nr = trans->blocks_used;
4728         btrfs_end_transaction(trans, root);
4729         if (drop_inode) {
4730                 inode_dec_link_count(inode);
4731                 iput(inode);
4732         }
4733         btrfs_btree_balance_dirty(root, nr);
4734         return err;
4735 }
4736
4737 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4738                       struct dentry *dentry)
4739 {
4740         struct btrfs_trans_handle *trans;
4741         struct btrfs_root *root = BTRFS_I(dir)->root;
4742         struct inode *inode = old_dentry->d_inode;
4743         u64 index;
4744         unsigned long nr = 0;
4745         int err;
4746         int drop_inode = 0;
4747
4748         /* do not allow sys_link's with other subvols of the same device */
4749         if (root->objectid != BTRFS_I(inode)->root->objectid)
4750                 return -EXDEV;
4751
4752         if (inode->i_nlink == ~0U)
4753                 return -EMLINK;
4754
4755         err = btrfs_set_inode_index(dir, &index);
4756         if (err)
4757                 goto fail;
4758
4759         /*
4760          * 2 items for inode and inode ref
4761          * 2 items for dir items
4762          * 1 item for parent inode
4763          */
4764         trans = btrfs_start_transaction(root, 5);
4765         if (IS_ERR(trans)) {
4766                 err = PTR_ERR(trans);
4767                 goto fail;
4768         }
4769
4770         btrfs_inc_nlink(inode);
4771         inode->i_ctime = CURRENT_TIME;
4772         ihold(inode);
4773
4774         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4775
4776         if (err) {
4777                 drop_inode = 1;
4778         } else {
4779                 struct dentry *parent = dentry->d_parent;
4780                 err = btrfs_update_inode(trans, root, inode);
4781                 BUG_ON(err);
4782                 d_instantiate(dentry, inode);
4783                 btrfs_log_new_name(trans, inode, NULL, parent);
4784         }
4785
4786         nr = trans->blocks_used;
4787         btrfs_end_transaction(trans, root);
4788 fail:
4789         if (drop_inode) {
4790                 inode_dec_link_count(inode);
4791                 iput(inode);
4792         }
4793         btrfs_btree_balance_dirty(root, nr);
4794         return err;
4795 }
4796
4797 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
4798 {
4799         struct inode *inode = NULL;
4800         struct btrfs_trans_handle *trans;
4801         struct btrfs_root *root = BTRFS_I(dir)->root;
4802         int err = 0;
4803         int drop_on_err = 0;
4804         u64 objectid = 0;
4805         u64 index = 0;
4806         unsigned long nr = 1;
4807
4808         /*
4809          * 2 items for inode and ref
4810          * 2 items for dir items
4811          * 1 for xattr if selinux is on
4812          */
4813         trans = btrfs_start_transaction(root, 5);
4814         if (IS_ERR(trans))
4815                 return PTR_ERR(trans);
4816
4817         err = btrfs_find_free_ino(root, &objectid);
4818         if (err)
4819                 goto out_fail;
4820
4821         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4822                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4823                                 S_IFDIR | mode, &index);
4824         if (IS_ERR(inode)) {
4825                 err = PTR_ERR(inode);
4826                 goto out_fail;
4827         }
4828
4829         drop_on_err = 1;
4830
4831         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4832         if (err)
4833                 goto out_fail;
4834
4835         inode->i_op = &btrfs_dir_inode_operations;
4836         inode->i_fop = &btrfs_dir_file_operations;
4837
4838         btrfs_i_size_write(inode, 0);
4839         err = btrfs_update_inode(trans, root, inode);
4840         if (err)
4841                 goto out_fail;
4842
4843         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4844                              dentry->d_name.len, 0, index);
4845         if (err)
4846                 goto out_fail;
4847
4848         d_instantiate(dentry, inode);
4849         drop_on_err = 0;
4850
4851 out_fail:
4852         nr = trans->blocks_used;
4853         btrfs_end_transaction(trans, root);
4854         if (drop_on_err)
4855                 iput(inode);
4856         btrfs_btree_balance_dirty(root, nr);
4857         return err;
4858 }
4859
4860 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4861  * and an extent that you want to insert, deal with overlap and insert
4862  * the new extent into the tree.
4863  */
4864 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4865                                 struct extent_map *existing,
4866                                 struct extent_map *em,
4867                                 u64 map_start, u64 map_len)
4868 {
4869         u64 start_diff;
4870
4871         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4872         start_diff = map_start - em->start;
4873         em->start = map_start;
4874         em->len = map_len;
4875         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4876             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4877                 em->block_start += start_diff;
4878                 em->block_len -= start_diff;
4879         }
4880         return add_extent_mapping(em_tree, em);
4881 }
4882
4883 static noinline int uncompress_inline(struct btrfs_path *path,
4884                                       struct inode *inode, struct page *page,
4885                                       size_t pg_offset, u64 extent_offset,
4886                                       struct btrfs_file_extent_item *item)
4887 {
4888         int ret;
4889         struct extent_buffer *leaf = path->nodes[0];
4890         char *tmp;
4891         size_t max_size;
4892         unsigned long inline_size;
4893         unsigned long ptr;
4894         int compress_type;
4895
4896         WARN_ON(pg_offset != 0);
4897         compress_type = btrfs_file_extent_compression(leaf, item);
4898         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4899         inline_size = btrfs_file_extent_inline_item_len(leaf,
4900                                         btrfs_item_nr(leaf, path->slots[0]));
4901         tmp = kmalloc(inline_size, GFP_NOFS);
4902         if (!tmp)
4903                 return -ENOMEM;
4904         ptr = btrfs_file_extent_inline_start(item);
4905
4906         read_extent_buffer(leaf, tmp, ptr, inline_size);
4907
4908         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4909         ret = btrfs_decompress(compress_type, tmp, page,
4910                                extent_offset, inline_size, max_size);
4911         if (ret) {
4912                 char *kaddr = kmap_atomic(page, KM_USER0);
4913                 unsigned long copy_size = min_t(u64,
4914                                   PAGE_CACHE_SIZE - pg_offset,
4915                                   max_size - extent_offset);
4916                 memset(kaddr + pg_offset, 0, copy_size);
4917                 kunmap_atomic(kaddr, KM_USER0);
4918         }
4919         kfree(tmp);
4920         return 0;
4921 }
4922
4923 /*
4924  * a bit scary, this does extent mapping from logical file offset to the disk.
4925  * the ugly parts come from merging extents from the disk with the in-ram
4926  * representation.  This gets more complex because of the data=ordered code,
4927  * where the in-ram extents might be locked pending data=ordered completion.
4928  *
4929  * This also copies inline extents directly into the page.
4930  */
4931
4932 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4933                                     size_t pg_offset, u64 start, u64 len,
4934                                     int create)
4935 {
4936         int ret;
4937         int err = 0;
4938         u64 bytenr;
4939         u64 extent_start = 0;
4940         u64 extent_end = 0;
4941         u64 objectid = btrfs_ino(inode);
4942         u32 found_type;
4943         struct btrfs_path *path = NULL;
4944         struct btrfs_root *root = BTRFS_I(inode)->root;
4945         struct btrfs_file_extent_item *item;
4946         struct extent_buffer *leaf;
4947         struct btrfs_key found_key;
4948         struct extent_map *em = NULL;
4949         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4950         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4951         struct btrfs_trans_handle *trans = NULL;
4952         int compress_type;
4953
4954 again:
4955         read_lock(&em_tree->lock);
4956         em = lookup_extent_mapping(em_tree, start, len);
4957         if (em)
4958                 em->bdev = root->fs_info->fs_devices->latest_bdev;
4959         read_unlock(&em_tree->lock);
4960
4961         if (em) {
4962                 if (em->start > start || em->start + em->len <= start)
4963                         free_extent_map(em);
4964                 else if (em->block_start == EXTENT_MAP_INLINE && page)
4965                         free_extent_map(em);
4966                 else
4967                         goto out;
4968         }
4969         em = alloc_extent_map();
4970         if (!em) {
4971                 err = -ENOMEM;
4972                 goto out;
4973         }
4974         em->bdev = root->fs_info->fs_devices->latest_bdev;
4975         em->start = EXTENT_MAP_HOLE;
4976         em->orig_start = EXTENT_MAP_HOLE;
4977         em->len = (u64)-1;
4978         em->block_len = (u64)-1;
4979
4980         if (!path) {
4981                 path = btrfs_alloc_path();
4982                 if (!path) {
4983                         err = -ENOMEM;
4984                         goto out;
4985                 }
4986                 /*
4987                  * Chances are we'll be called again, so go ahead and do
4988                  * readahead
4989                  */
4990                 path->reada = 1;
4991         }
4992
4993         ret = btrfs_lookup_file_extent(trans, root, path,
4994                                        objectid, start, trans != NULL);
4995         if (ret < 0) {
4996                 err = ret;
4997                 goto out;
4998         }
4999
5000         if (ret != 0) {
5001                 if (path->slots[0] == 0)
5002                         goto not_found;
5003                 path->slots[0]--;
5004         }
5005
5006         leaf = path->nodes[0];
5007         item = btrfs_item_ptr(leaf, path->slots[0],
5008                               struct btrfs_file_extent_item);
5009         /* are we inside the extent that was found? */
5010         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5011         found_type = btrfs_key_type(&found_key);
5012         if (found_key.objectid != objectid ||
5013             found_type != BTRFS_EXTENT_DATA_KEY) {
5014                 goto not_found;
5015         }
5016
5017         found_type = btrfs_file_extent_type(leaf, item);
5018         extent_start = found_key.offset;
5019         compress_type = btrfs_file_extent_compression(leaf, item);
5020         if (found_type == BTRFS_FILE_EXTENT_REG ||
5021             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5022                 extent_end = extent_start +
5023                        btrfs_file_extent_num_bytes(leaf, item);
5024         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5025                 size_t size;
5026                 size = btrfs_file_extent_inline_len(leaf, item);
5027                 extent_end = (extent_start + size + root->sectorsize - 1) &
5028                         ~((u64)root->sectorsize - 1);
5029         }
5030
5031         if (start >= extent_end) {
5032                 path->slots[0]++;
5033                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5034                         ret = btrfs_next_leaf(root, path);
5035                         if (ret < 0) {
5036                                 err = ret;
5037                                 goto out;
5038                         }
5039                         if (ret > 0)
5040                                 goto not_found;
5041                         leaf = path->nodes[0];
5042                 }
5043                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5044                 if (found_key.objectid != objectid ||
5045                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5046                         goto not_found;
5047                 if (start + len <= found_key.offset)
5048                         goto not_found;
5049                 em->start = start;
5050                 em->len = found_key.offset - start;
5051                 goto not_found_em;
5052         }
5053
5054         if (found_type == BTRFS_FILE_EXTENT_REG ||
5055             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5056                 em->start = extent_start;
5057                 em->len = extent_end - extent_start;
5058                 em->orig_start = extent_start -
5059                                  btrfs_file_extent_offset(leaf, item);
5060                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5061                 if (bytenr == 0) {
5062                         em->block_start = EXTENT_MAP_HOLE;
5063                         goto insert;
5064                 }
5065                 if (compress_type != BTRFS_COMPRESS_NONE) {
5066                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5067                         em->compress_type = compress_type;
5068                         em->block_start = bytenr;
5069                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5070                                                                          item);
5071                 } else {
5072                         bytenr += btrfs_file_extent_offset(leaf, item);
5073                         em->block_start = bytenr;
5074                         em->block_len = em->len;
5075                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5076                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5077                 }
5078                 goto insert;
5079         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5080                 unsigned long ptr;
5081                 char *map;
5082                 size_t size;
5083                 size_t extent_offset;
5084                 size_t copy_size;
5085
5086                 em->block_start = EXTENT_MAP_INLINE;
5087                 if (!page || create) {
5088                         em->start = extent_start;
5089                         em->len = extent_end - extent_start;
5090                         goto out;
5091                 }
5092
5093                 size = btrfs_file_extent_inline_len(leaf, item);
5094                 extent_offset = page_offset(page) + pg_offset - extent_start;
5095                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5096                                 size - extent_offset);
5097                 em->start = extent_start + extent_offset;
5098                 em->len = (copy_size + root->sectorsize - 1) &
5099                         ~((u64)root->sectorsize - 1);
5100                 em->orig_start = EXTENT_MAP_INLINE;
5101                 if (compress_type) {
5102                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5103                         em->compress_type = compress_type;
5104                 }
5105                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5106                 if (create == 0 && !PageUptodate(page)) {
5107                         if (btrfs_file_extent_compression(leaf, item) !=
5108                             BTRFS_COMPRESS_NONE) {
5109                                 ret = uncompress_inline(path, inode, page,
5110                                                         pg_offset,
5111                                                         extent_offset, item);
5112                                 BUG_ON(ret);
5113                         } else {
5114                                 map = kmap(page);
5115                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5116                                                    copy_size);
5117                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5118                                         memset(map + pg_offset + copy_size, 0,
5119                                                PAGE_CACHE_SIZE - pg_offset -
5120                                                copy_size);
5121                                 }
5122                                 kunmap(page);
5123                         }
5124                         flush_dcache_page(page);
5125                 } else if (create && PageUptodate(page)) {
5126                         BUG();
5127                         if (!trans) {
5128                                 kunmap(page);
5129                                 free_extent_map(em);
5130                                 em = NULL;
5131
5132                                 btrfs_release_path(path);
5133                                 trans = btrfs_join_transaction(root);
5134
5135                                 if (IS_ERR(trans))
5136                                         return ERR_CAST(trans);
5137                                 goto again;
5138                         }
5139                         map = kmap(page);
5140                         write_extent_buffer(leaf, map + pg_offset, ptr,
5141                                             copy_size);
5142                         kunmap(page);
5143                         btrfs_mark_buffer_dirty(leaf);
5144                 }
5145                 set_extent_uptodate(io_tree, em->start,
5146                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5147                 goto insert;
5148         } else {
5149                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5150                 WARN_ON(1);
5151         }
5152 not_found:
5153         em->start = start;
5154         em->len = len;
5155 not_found_em:
5156         em->block_start = EXTENT_MAP_HOLE;
5157         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5158 insert:
5159         btrfs_release_path(path);
5160         if (em->start > start || extent_map_end(em) <= start) {
5161                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5162                        "[%llu %llu]\n", (unsigned long long)em->start,
5163                        (unsigned long long)em->len,
5164                        (unsigned long long)start,
5165                        (unsigned long long)len);
5166                 err = -EIO;
5167                 goto out;
5168         }
5169
5170         err = 0;
5171         write_lock(&em_tree->lock);
5172         ret = add_extent_mapping(em_tree, em);
5173         /* it is possible that someone inserted the extent into the tree
5174          * while we had the lock dropped.  It is also possible that
5175          * an overlapping map exists in the tree
5176          */
5177         if (ret == -EEXIST) {
5178                 struct extent_map *existing;
5179
5180                 ret = 0;
5181
5182                 existing = lookup_extent_mapping(em_tree, start, len);
5183                 if (existing && (existing->start > start ||
5184                     existing->start + existing->len <= start)) {
5185                         free_extent_map(existing);
5186                         existing = NULL;
5187                 }
5188                 if (!existing) {
5189                         existing = lookup_extent_mapping(em_tree, em->start,
5190                                                          em->len);
5191                         if (existing) {
5192                                 err = merge_extent_mapping(em_tree, existing,
5193                                                            em, start,
5194                                                            root->sectorsize);
5195                                 free_extent_map(existing);
5196                                 if (err) {
5197                                         free_extent_map(em);
5198                                         em = NULL;
5199                                 }
5200                         } else {
5201                                 err = -EIO;
5202                                 free_extent_map(em);
5203                                 em = NULL;
5204                         }
5205                 } else {
5206                         free_extent_map(em);
5207                         em = existing;
5208                         err = 0;
5209                 }
5210         }
5211         write_unlock(&em_tree->lock);
5212 out:
5213
5214         trace_btrfs_get_extent(root, em);
5215
5216         if (path)
5217                 btrfs_free_path(path);
5218         if (trans) {
5219                 ret = btrfs_end_transaction(trans, root);
5220                 if (!err)
5221                         err = ret;
5222         }
5223         if (err) {
5224                 free_extent_map(em);
5225                 return ERR_PTR(err);
5226         }
5227         return em;
5228 }
5229
5230 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5231                                            size_t pg_offset, u64 start, u64 len,
5232                                            int create)
5233 {
5234         struct extent_map *em;
5235         struct extent_map *hole_em = NULL;
5236         u64 range_start = start;
5237         u64 end;
5238         u64 found;
5239         u64 found_end;
5240         int err = 0;
5241
5242         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5243         if (IS_ERR(em))
5244                 return em;
5245         if (em) {
5246                 /*
5247                  * if our em maps to a hole, there might
5248                  * actually be delalloc bytes behind it
5249                  */
5250                 if (em->block_start != EXTENT_MAP_HOLE)
5251                         return em;
5252                 else
5253                         hole_em = em;
5254         }
5255
5256         /* check to see if we've wrapped (len == -1 or similar) */
5257         end = start + len;
5258         if (end < start)
5259                 end = (u64)-1;
5260         else
5261                 end -= 1;
5262
5263         em = NULL;
5264
5265         /* ok, we didn't find anything, lets look for delalloc */
5266         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5267                                  end, len, EXTENT_DELALLOC, 1);
5268         found_end = range_start + found;
5269         if (found_end < range_start)
5270                 found_end = (u64)-1;
5271
5272         /*
5273          * we didn't find anything useful, return
5274          * the original results from get_extent()
5275          */
5276         if (range_start > end || found_end <= start) {
5277                 em = hole_em;
5278                 hole_em = NULL;
5279                 goto out;
5280         }
5281
5282         /* adjust the range_start to make sure it doesn't
5283          * go backwards from the start they passed in
5284          */
5285         range_start = max(start,range_start);
5286         found = found_end - range_start;
5287
5288         if (found > 0) {
5289                 u64 hole_start = start;
5290                 u64 hole_len = len;
5291
5292                 em = alloc_extent_map();
5293                 if (!em) {
5294                         err = -ENOMEM;
5295                         goto out;
5296                 }
5297                 /*
5298                  * when btrfs_get_extent can't find anything it
5299                  * returns one huge hole
5300                  *
5301                  * make sure what it found really fits our range, and
5302                  * adjust to make sure it is based on the start from
5303                  * the caller
5304                  */
5305                 if (hole_em) {
5306                         u64 calc_end = extent_map_end(hole_em);
5307
5308                         if (calc_end <= start || (hole_em->start > end)) {
5309                                 free_extent_map(hole_em);
5310                                 hole_em = NULL;
5311                         } else {
5312                                 hole_start = max(hole_em->start, start);
5313                                 hole_len = calc_end - hole_start;
5314                         }
5315                 }
5316                 em->bdev = NULL;
5317                 if (hole_em && range_start > hole_start) {
5318                         /* our hole starts before our delalloc, so we
5319                          * have to return just the parts of the hole
5320                          * that go until  the delalloc starts
5321                          */
5322                         em->len = min(hole_len,
5323                                       range_start - hole_start);
5324                         em->start = hole_start;
5325                         em->orig_start = hole_start;
5326                         /*
5327                          * don't adjust block start at all,
5328                          * it is fixed at EXTENT_MAP_HOLE
5329                          */
5330                         em->block_start = hole_em->block_start;
5331                         em->block_len = hole_len;
5332                 } else {
5333                         em->start = range_start;
5334                         em->len = found;
5335                         em->orig_start = range_start;
5336                         em->block_start = EXTENT_MAP_DELALLOC;
5337                         em->block_len = found;
5338                 }
5339         } else if (hole_em) {
5340                 return hole_em;
5341         }
5342 out:
5343
5344         free_extent_map(hole_em);
5345         if (err) {
5346                 free_extent_map(em);
5347                 return ERR_PTR(err);
5348         }
5349         return em;
5350 }
5351
5352 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5353                                                   struct extent_map *em,
5354                                                   u64 start, u64 len)
5355 {
5356         struct btrfs_root *root = BTRFS_I(inode)->root;
5357         struct btrfs_trans_handle *trans;
5358         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5359         struct btrfs_key ins;
5360         u64 alloc_hint;
5361         int ret;
5362         bool insert = false;
5363
5364         /*
5365          * Ok if the extent map we looked up is a hole and is for the exact
5366          * range we want, there is no reason to allocate a new one, however if
5367          * it is not right then we need to free this one and drop the cache for
5368          * our range.
5369          */
5370         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5371             em->len != len) {
5372                 free_extent_map(em);
5373                 em = NULL;
5374                 insert = true;
5375                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5376         }
5377
5378         trans = btrfs_join_transaction(root);
5379         if (IS_ERR(trans))
5380                 return ERR_CAST(trans);
5381
5382         if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5383                 btrfs_add_inode_defrag(trans, inode);
5384
5385         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5386
5387         alloc_hint = get_extent_allocation_hint(inode, start, len);
5388         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5389                                    alloc_hint, (u64)-1, &ins, 1);
5390         if (ret) {
5391                 em = ERR_PTR(ret);
5392                 goto out;
5393         }
5394
5395         if (!em) {
5396                 em = alloc_extent_map();
5397                 if (!em) {
5398                         em = ERR_PTR(-ENOMEM);
5399                         goto out;
5400                 }
5401         }
5402
5403         em->start = start;
5404         em->orig_start = em->start;
5405         em->len = ins.offset;
5406
5407         em->block_start = ins.objectid;
5408         em->block_len = ins.offset;
5409         em->bdev = root->fs_info->fs_devices->latest_bdev;
5410
5411         /*
5412          * We need to do this because if we're using the original em we searched
5413          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5414          */
5415         em->flags = 0;
5416         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5417
5418         while (insert) {
5419                 write_lock(&em_tree->lock);
5420                 ret = add_extent_mapping(em_tree, em);
5421                 write_unlock(&em_tree->lock);
5422                 if (ret != -EEXIST)
5423                         break;
5424                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5425         }
5426
5427         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5428                                            ins.offset, ins.offset, 0);
5429         if (ret) {
5430                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5431                 em = ERR_PTR(ret);
5432         }
5433 out:
5434         btrfs_end_transaction(trans, root);
5435         return em;
5436 }
5437
5438 /*
5439  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5440  * block must be cow'd
5441  */
5442 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5443                                       struct inode *inode, u64 offset, u64 len)
5444 {
5445         struct btrfs_path *path;
5446         int ret;
5447         struct extent_buffer *leaf;
5448         struct btrfs_root *root = BTRFS_I(inode)->root;
5449         struct btrfs_file_extent_item *fi;
5450         struct btrfs_key key;
5451         u64 disk_bytenr;
5452         u64 backref_offset;
5453         u64 extent_end;
5454         u64 num_bytes;
5455         int slot;
5456         int found_type;
5457
5458         path = btrfs_alloc_path();
5459         if (!path)
5460                 return -ENOMEM;
5461
5462         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5463                                        offset, 0);
5464         if (ret < 0)
5465                 goto out;
5466
5467         slot = path->slots[0];
5468         if (ret == 1) {
5469                 if (slot == 0) {
5470                         /* can't find the item, must cow */
5471                         ret = 0;
5472                         goto out;
5473                 }
5474                 slot--;
5475         }
5476         ret = 0;
5477         leaf = path->nodes[0];
5478         btrfs_item_key_to_cpu(leaf, &key, slot);
5479         if (key.objectid != btrfs_ino(inode) ||
5480             key.type != BTRFS_EXTENT_DATA_KEY) {
5481                 /* not our file or wrong item type, must cow */
5482                 goto out;
5483         }
5484
5485         if (key.offset > offset) {
5486                 /* Wrong offset, must cow */
5487                 goto out;
5488         }
5489
5490         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5491         found_type = btrfs_file_extent_type(leaf, fi);
5492         if (found_type != BTRFS_FILE_EXTENT_REG &&
5493             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5494                 /* not a regular extent, must cow */
5495                 goto out;
5496         }
5497         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5498         backref_offset = btrfs_file_extent_offset(leaf, fi);
5499
5500         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5501         if (extent_end < offset + len) {
5502                 /* extent doesn't include our full range, must cow */
5503                 goto out;
5504         }
5505
5506         if (btrfs_extent_readonly(root, disk_bytenr))
5507                 goto out;
5508
5509         /*
5510          * look for other files referencing this extent, if we
5511          * find any we must cow
5512          */
5513         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5514                                   key.offset - backref_offset, disk_bytenr))
5515                 goto out;
5516
5517         /*
5518          * adjust disk_bytenr and num_bytes to cover just the bytes
5519          * in this extent we are about to write.  If there
5520          * are any csums in that range we have to cow in order
5521          * to keep the csums correct
5522          */
5523         disk_bytenr += backref_offset;
5524         disk_bytenr += offset - key.offset;
5525         num_bytes = min(offset + len, extent_end) - offset;
5526         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5527                                 goto out;
5528         /*
5529          * all of the above have passed, it is safe to overwrite this extent
5530          * without cow
5531          */
5532         ret = 1;
5533 out:
5534         btrfs_free_path(path);
5535         return ret;
5536 }
5537
5538 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5539                                    struct buffer_head *bh_result, int create)
5540 {
5541         struct extent_map *em;
5542         struct btrfs_root *root = BTRFS_I(inode)->root;
5543         u64 start = iblock << inode->i_blkbits;
5544         u64 len = bh_result->b_size;
5545         struct btrfs_trans_handle *trans;
5546
5547         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5548         if (IS_ERR(em))
5549                 return PTR_ERR(em);
5550
5551         /*
5552          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5553          * io.  INLINE is special, and we could probably kludge it in here, but
5554          * it's still buffered so for safety lets just fall back to the generic
5555          * buffered path.
5556          *
5557          * For COMPRESSED we _have_ to read the entire extent in so we can
5558          * decompress it, so there will be buffering required no matter what we
5559          * do, so go ahead and fallback to buffered.
5560          *
5561          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5562          * to buffered IO.  Don't blame me, this is the price we pay for using
5563          * the generic code.
5564          */
5565         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5566             em->block_start == EXTENT_MAP_INLINE) {
5567                 free_extent_map(em);
5568                 return -ENOTBLK;
5569         }
5570
5571         /* Just a good old fashioned hole, return */
5572         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5573                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5574                 free_extent_map(em);
5575                 /* DIO will do one hole at a time, so just unlock a sector */
5576                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5577                               start + root->sectorsize - 1, GFP_NOFS);
5578                 return 0;
5579         }
5580
5581         /*
5582          * We don't allocate a new extent in the following cases
5583          *
5584          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5585          * existing extent.
5586          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5587          * just use the extent.
5588          *
5589          */
5590         if (!create) {
5591                 len = em->len - (start - em->start);
5592                 goto map;
5593         }
5594
5595         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5596             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5597              em->block_start != EXTENT_MAP_HOLE)) {
5598                 int type;
5599                 int ret;
5600                 u64 block_start;
5601
5602                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5603                         type = BTRFS_ORDERED_PREALLOC;
5604                 else
5605                         type = BTRFS_ORDERED_NOCOW;
5606                 len = min(len, em->len - (start - em->start));
5607                 block_start = em->block_start + (start - em->start);
5608
5609                 /*
5610                  * we're not going to log anything, but we do need
5611                  * to make sure the current transaction stays open
5612                  * while we look for nocow cross refs
5613                  */
5614                 trans = btrfs_join_transaction(root);
5615                 if (IS_ERR(trans))
5616                         goto must_cow;
5617
5618                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5619                         ret = btrfs_add_ordered_extent_dio(inode, start,
5620                                            block_start, len, len, type);
5621                         btrfs_end_transaction(trans, root);
5622                         if (ret) {
5623                                 free_extent_map(em);
5624                                 return ret;
5625                         }
5626                         goto unlock;
5627                 }
5628                 btrfs_end_transaction(trans, root);
5629         }
5630 must_cow:
5631         /*
5632          * this will cow the extent, reset the len in case we changed
5633          * it above
5634          */
5635         len = bh_result->b_size;
5636         em = btrfs_new_extent_direct(inode, em, start, len);
5637         if (IS_ERR(em))
5638                 return PTR_ERR(em);
5639         len = min(len, em->len - (start - em->start));
5640 unlock:
5641         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5642                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5643                           0, NULL, GFP_NOFS);
5644 map:
5645         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5646                 inode->i_blkbits;
5647         bh_result->b_size = len;
5648         bh_result->b_bdev = em->bdev;
5649         set_buffer_mapped(bh_result);
5650         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5651                 set_buffer_new(bh_result);
5652
5653         free_extent_map(em);
5654
5655         return 0;
5656 }
5657
5658 struct btrfs_dio_private {
5659         struct inode *inode;
5660         u64 logical_offset;
5661         u64 disk_bytenr;
5662         u64 bytes;
5663         u32 *csums;
5664         void *private;
5665
5666         /* number of bios pending for this dio */
5667         atomic_t pending_bios;
5668
5669         /* IO errors */
5670         int errors;
5671
5672         struct bio *orig_bio;
5673 };
5674
5675 static void btrfs_endio_direct_read(struct bio *bio, int err)
5676 {
5677         struct btrfs_dio_private *dip = bio->bi_private;
5678         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5679         struct bio_vec *bvec = bio->bi_io_vec;
5680         struct inode *inode = dip->inode;
5681         struct btrfs_root *root = BTRFS_I(inode)->root;
5682         u64 start;
5683         u32 *private = dip->csums;
5684
5685         start = dip->logical_offset;
5686         do {
5687                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5688                         struct page *page = bvec->bv_page;
5689                         char *kaddr;
5690                         u32 csum = ~(u32)0;
5691                         unsigned long flags;
5692
5693                         local_irq_save(flags);
5694                         kaddr = kmap_atomic(page, KM_IRQ0);
5695                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5696                                                csum, bvec->bv_len);
5697                         btrfs_csum_final(csum, (char *)&csum);
5698                         kunmap_atomic(kaddr, KM_IRQ0);
5699                         local_irq_restore(flags);
5700
5701                         flush_dcache_page(bvec->bv_page);
5702                         if (csum != *private) {
5703                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
5704                                       " %llu csum %u private %u\n",
5705                                       (unsigned long long)btrfs_ino(inode),
5706                                       (unsigned long long)start,
5707                                       csum, *private);
5708                                 err = -EIO;
5709                         }
5710                 }
5711
5712                 start += bvec->bv_len;
5713                 private++;
5714                 bvec++;
5715         } while (bvec <= bvec_end);
5716
5717         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5718                       dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5719         bio->bi_private = dip->private;
5720
5721         kfree(dip->csums);
5722         kfree(dip);
5723
5724         /* If we had a csum failure make sure to clear the uptodate flag */
5725         if (err)
5726                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5727         dio_end_io(bio, err);
5728 }
5729
5730 static void btrfs_endio_direct_write(struct bio *bio, int err)
5731 {
5732         struct btrfs_dio_private *dip = bio->bi_private;
5733         struct inode *inode = dip->inode;
5734         struct btrfs_root *root = BTRFS_I(inode)->root;
5735         struct btrfs_trans_handle *trans;
5736         struct btrfs_ordered_extent *ordered = NULL;
5737         struct extent_state *cached_state = NULL;
5738         u64 ordered_offset = dip->logical_offset;
5739         u64 ordered_bytes = dip->bytes;
5740         int ret;
5741
5742         if (err)
5743                 goto out_done;
5744 again:
5745         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5746                                                    &ordered_offset,
5747                                                    ordered_bytes);
5748         if (!ret)
5749                 goto out_test;
5750
5751         BUG_ON(!ordered);
5752
5753         trans = btrfs_join_transaction(root);
5754         if (IS_ERR(trans)) {
5755                 err = -ENOMEM;
5756                 goto out;
5757         }
5758         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5759
5760         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5761                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5762                 if (!ret)
5763                         err = btrfs_update_inode_fallback(trans, root, inode);
5764                 goto out;
5765         }
5766
5767         lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5768                          ordered->file_offset + ordered->len - 1, 0,
5769                          &cached_state, GFP_NOFS);
5770
5771         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5772                 ret = btrfs_mark_extent_written(trans, inode,
5773                                                 ordered->file_offset,
5774                                                 ordered->file_offset +
5775                                                 ordered->len);
5776                 if (ret) {
5777                         err = ret;
5778                         goto out_unlock;
5779                 }
5780         } else {
5781                 ret = insert_reserved_file_extent(trans, inode,
5782                                                   ordered->file_offset,
5783                                                   ordered->start,
5784                                                   ordered->disk_len,
5785                                                   ordered->len,
5786                                                   ordered->len,
5787                                                   0, 0, 0,
5788                                                   BTRFS_FILE_EXTENT_REG);
5789                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5790                                    ordered->file_offset, ordered->len);
5791                 if (ret) {
5792                         err = ret;
5793                         WARN_ON(1);
5794                         goto out_unlock;
5795                 }
5796         }
5797
5798         add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5799         ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5800         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags))
5801                 btrfs_update_inode_fallback(trans, root, inode);
5802         ret = 0;
5803 out_unlock:
5804         unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5805                              ordered->file_offset + ordered->len - 1,
5806                              &cached_state, GFP_NOFS);
5807 out:
5808         btrfs_delalloc_release_metadata(inode, ordered->len);
5809         btrfs_end_transaction(trans, root);
5810         ordered_offset = ordered->file_offset + ordered->len;
5811         btrfs_put_ordered_extent(ordered);
5812         btrfs_put_ordered_extent(ordered);
5813
5814 out_test:
5815         /*
5816          * our bio might span multiple ordered extents.  If we haven't
5817          * completed the accounting for the whole dio, go back and try again
5818          */
5819         if (ordered_offset < dip->logical_offset + dip->bytes) {
5820                 ordered_bytes = dip->logical_offset + dip->bytes -
5821                         ordered_offset;
5822                 goto again;
5823         }
5824 out_done:
5825         bio->bi_private = dip->private;
5826
5827         kfree(dip->csums);
5828         kfree(dip);
5829
5830         /* If we had an error make sure to clear the uptodate flag */
5831         if (err)
5832                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5833         dio_end_io(bio, err);
5834 }
5835
5836 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5837                                     struct bio *bio, int mirror_num,
5838                                     unsigned long bio_flags, u64 offset)
5839 {
5840         int ret;
5841         struct btrfs_root *root = BTRFS_I(inode)->root;
5842         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5843         BUG_ON(ret);
5844         return 0;
5845 }
5846
5847 static void btrfs_end_dio_bio(struct bio *bio, int err)
5848 {
5849         struct btrfs_dio_private *dip = bio->bi_private;
5850
5851         if (err) {
5852                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
5853                       "sector %#Lx len %u err no %d\n",
5854                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
5855                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
5856                 dip->errors = 1;
5857
5858                 /*
5859                  * before atomic variable goto zero, we must make sure
5860                  * dip->errors is perceived to be set.
5861                  */
5862                 smp_mb__before_atomic_dec();
5863         }
5864
5865         /* if there are more bios still pending for this dio, just exit */
5866         if (!atomic_dec_and_test(&dip->pending_bios))
5867                 goto out;
5868
5869         if (dip->errors)
5870                 bio_io_error(dip->orig_bio);
5871         else {
5872                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5873                 bio_endio(dip->orig_bio, 0);
5874         }
5875 out:
5876         bio_put(bio);
5877 }
5878
5879 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5880                                        u64 first_sector, gfp_t gfp_flags)
5881 {
5882         int nr_vecs = bio_get_nr_vecs(bdev);
5883         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5884 }
5885
5886 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5887                                          int rw, u64 file_offset, int skip_sum,
5888                                          u32 *csums, int async_submit)
5889 {
5890         int write = rw & REQ_WRITE;
5891         struct btrfs_root *root = BTRFS_I(inode)->root;
5892         int ret;
5893
5894         bio_get(bio);
5895         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5896         if (ret)
5897                 goto err;
5898
5899         if (skip_sum)
5900                 goto map;
5901
5902         if (write && async_submit) {
5903                 ret = btrfs_wq_submit_bio(root->fs_info,
5904                                    inode, rw, bio, 0, 0,
5905                                    file_offset,
5906                                    __btrfs_submit_bio_start_direct_io,
5907                                    __btrfs_submit_bio_done);
5908                 goto err;
5909         } else if (write) {
5910                 /*
5911                  * If we aren't doing async submit, calculate the csum of the
5912                  * bio now.
5913                  */
5914                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5915                 if (ret)
5916                         goto err;
5917         } else if (!skip_sum) {
5918                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5919                                           file_offset, csums);
5920                 if (ret)
5921                         goto err;
5922         }
5923
5924 map:
5925         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
5926 err:
5927         bio_put(bio);
5928         return ret;
5929 }
5930
5931 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5932                                     int skip_sum)
5933 {
5934         struct inode *inode = dip->inode;
5935         struct btrfs_root *root = BTRFS_I(inode)->root;
5936         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5937         struct bio *bio;
5938         struct bio *orig_bio = dip->orig_bio;
5939         struct bio_vec *bvec = orig_bio->bi_io_vec;
5940         u64 start_sector = orig_bio->bi_sector;
5941         u64 file_offset = dip->logical_offset;
5942         u64 submit_len = 0;
5943         u64 map_length;
5944         int nr_pages = 0;
5945         u32 *csums = dip->csums;
5946         int ret = 0;
5947         int async_submit = 0;
5948         int write = rw & REQ_WRITE;
5949
5950         map_length = orig_bio->bi_size;
5951         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5952                               &map_length, NULL, 0);
5953         if (ret) {
5954                 bio_put(orig_bio);
5955                 return -EIO;
5956         }
5957
5958         if (map_length >= orig_bio->bi_size) {
5959                 bio = orig_bio;
5960                 goto submit;
5961         }
5962
5963         async_submit = 1;
5964         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5965         if (!bio)
5966                 return -ENOMEM;
5967         bio->bi_private = dip;
5968         bio->bi_end_io = btrfs_end_dio_bio;
5969         atomic_inc(&dip->pending_bios);
5970
5971         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5972                 if (unlikely(map_length < submit_len + bvec->bv_len ||
5973                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5974                                  bvec->bv_offset) < bvec->bv_len)) {
5975                         /*
5976                          * inc the count before we submit the bio so
5977                          * we know the end IO handler won't happen before
5978                          * we inc the count. Otherwise, the dip might get freed
5979                          * before we're done setting it up
5980                          */
5981                         atomic_inc(&dip->pending_bios);
5982                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
5983                                                      file_offset, skip_sum,
5984                                                      csums, async_submit);
5985                         if (ret) {
5986                                 bio_put(bio);
5987                                 atomic_dec(&dip->pending_bios);
5988                                 goto out_err;
5989                         }
5990
5991                         /* Write's use the ordered csums */
5992                         if (!write && !skip_sum)
5993                                 csums = csums + nr_pages;
5994                         start_sector += submit_len >> 9;
5995                         file_offset += submit_len;
5996
5997                         submit_len = 0;
5998                         nr_pages = 0;
5999
6000                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6001                                                   start_sector, GFP_NOFS);
6002                         if (!bio)
6003                                 goto out_err;
6004                         bio->bi_private = dip;
6005                         bio->bi_end_io = btrfs_end_dio_bio;
6006
6007                         map_length = orig_bio->bi_size;
6008                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6009                                               &map_length, NULL, 0);
6010                         if (ret) {
6011                                 bio_put(bio);
6012                                 goto out_err;
6013                         }
6014                 } else {
6015                         submit_len += bvec->bv_len;
6016                         nr_pages ++;
6017                         bvec++;
6018                 }
6019         }
6020
6021 submit:
6022         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6023                                      csums, async_submit);
6024         if (!ret)
6025                 return 0;
6026
6027         bio_put(bio);
6028 out_err:
6029         dip->errors = 1;
6030         /*
6031          * before atomic variable goto zero, we must
6032          * make sure dip->errors is perceived to be set.
6033          */
6034         smp_mb__before_atomic_dec();
6035         if (atomic_dec_and_test(&dip->pending_bios))
6036                 bio_io_error(dip->orig_bio);
6037
6038         /* bio_end_io() will handle error, so we needn't return it */
6039         return 0;
6040 }
6041
6042 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6043                                 loff_t file_offset)
6044 {
6045         struct btrfs_root *root = BTRFS_I(inode)->root;
6046         struct btrfs_dio_private *dip;
6047         struct bio_vec *bvec = bio->bi_io_vec;
6048         int skip_sum;
6049         int write = rw & REQ_WRITE;
6050         int ret = 0;
6051
6052         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6053
6054         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6055         if (!dip) {
6056                 ret = -ENOMEM;
6057                 goto free_ordered;
6058         }
6059         dip->csums = NULL;
6060
6061         /* Write's use the ordered csum stuff, so we don't need dip->csums */
6062         if (!write && !skip_sum) {
6063                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6064                 if (!dip->csums) {
6065                         kfree(dip);
6066                         ret = -ENOMEM;
6067                         goto free_ordered;
6068                 }
6069         }
6070
6071         dip->private = bio->bi_private;
6072         dip->inode = inode;
6073         dip->logical_offset = file_offset;
6074
6075         dip->bytes = 0;
6076         do {
6077                 dip->bytes += bvec->bv_len;
6078                 bvec++;
6079         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6080
6081         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6082         bio->bi_private = dip;
6083         dip->errors = 0;
6084         dip->orig_bio = bio;
6085         atomic_set(&dip->pending_bios, 0);
6086
6087         if (write)
6088                 bio->bi_end_io = btrfs_endio_direct_write;
6089         else
6090                 bio->bi_end_io = btrfs_endio_direct_read;
6091
6092         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6093         if (!ret)
6094                 return;
6095 free_ordered:
6096         /*
6097          * If this is a write, we need to clean up the reserved space and kill
6098          * the ordered extent.
6099          */
6100         if (write) {
6101                 struct btrfs_ordered_extent *ordered;
6102                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6103                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6104                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6105                         btrfs_free_reserved_extent(root, ordered->start,
6106                                                    ordered->disk_len);
6107                 btrfs_put_ordered_extent(ordered);
6108                 btrfs_put_ordered_extent(ordered);
6109         }
6110         bio_endio(bio, ret);
6111 }
6112
6113 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6114                         const struct iovec *iov, loff_t offset,
6115                         unsigned long nr_segs)
6116 {
6117         int seg;
6118         int i;
6119         size_t size;
6120         unsigned long addr;
6121         unsigned blocksize_mask = root->sectorsize - 1;
6122         ssize_t retval = -EINVAL;
6123         loff_t end = offset;
6124
6125         if (offset & blocksize_mask)
6126                 goto out;
6127
6128         /* Check the memory alignment.  Blocks cannot straddle pages */
6129         for (seg = 0; seg < nr_segs; seg++) {
6130                 addr = (unsigned long)iov[seg].iov_base;
6131                 size = iov[seg].iov_len;
6132                 end += size;
6133                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6134                         goto out;
6135
6136                 /* If this is a write we don't need to check anymore */
6137                 if (rw & WRITE)
6138                         continue;
6139
6140                 /*
6141                  * Check to make sure we don't have duplicate iov_base's in this
6142                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6143                  * when reading back.
6144                  */
6145                 for (i = seg + 1; i < nr_segs; i++) {
6146                         if (iov[seg].iov_base == iov[i].iov_base)
6147                                 goto out;
6148                 }
6149         }
6150         retval = 0;
6151 out:
6152         return retval;
6153 }
6154 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6155                         const struct iovec *iov, loff_t offset,
6156                         unsigned long nr_segs)
6157 {
6158         struct file *file = iocb->ki_filp;
6159         struct inode *inode = file->f_mapping->host;
6160         struct btrfs_ordered_extent *ordered;
6161         struct extent_state *cached_state = NULL;
6162         u64 lockstart, lockend;
6163         ssize_t ret;
6164         int writing = rw & WRITE;
6165         int write_bits = 0;
6166         size_t count = iov_length(iov, nr_segs);
6167
6168         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6169                             offset, nr_segs)) {
6170                 return 0;
6171         }
6172
6173         lockstart = offset;
6174         lockend = offset + count - 1;
6175
6176         if (writing) {
6177                 ret = btrfs_delalloc_reserve_space(inode, count);
6178                 if (ret)
6179                         goto out;
6180         }
6181
6182         while (1) {
6183                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6184                                  0, &cached_state, GFP_NOFS);
6185                 /*
6186                  * We're concerned with the entire range that we're going to be
6187                  * doing DIO to, so we need to make sure theres no ordered
6188                  * extents in this range.
6189                  */
6190                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6191                                                      lockend - lockstart + 1);
6192                 if (!ordered)
6193                         break;
6194                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6195                                      &cached_state, GFP_NOFS);
6196                 btrfs_start_ordered_extent(inode, ordered, 1);
6197                 btrfs_put_ordered_extent(ordered);
6198                 cond_resched();
6199         }
6200
6201         /*
6202          * we don't use btrfs_set_extent_delalloc because we don't want
6203          * the dirty or uptodate bits
6204          */
6205         if (writing) {
6206                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6207                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6208                                      EXTENT_DELALLOC, 0, NULL, &cached_state,
6209                                      GFP_NOFS);
6210                 if (ret) {
6211                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6212                                          lockend, EXTENT_LOCKED | write_bits,
6213                                          1, 0, &cached_state, GFP_NOFS);
6214                         goto out;
6215                 }
6216         }
6217
6218         free_extent_state(cached_state);
6219         cached_state = NULL;
6220
6221         ret = __blockdev_direct_IO(rw, iocb, inode,
6222                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6223                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6224                    btrfs_submit_direct, 0);
6225
6226         if (ret < 0 && ret != -EIOCBQUEUED) {
6227                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6228                               offset + iov_length(iov, nr_segs) - 1,
6229                               EXTENT_LOCKED | write_bits, 1, 0,
6230                               &cached_state, GFP_NOFS);
6231         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6232                 /*
6233                  * We're falling back to buffered, unlock the section we didn't
6234                  * do IO on.
6235                  */
6236                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6237                               offset + iov_length(iov, nr_segs) - 1,
6238                               EXTENT_LOCKED | write_bits, 1, 0,
6239                               &cached_state, GFP_NOFS);
6240         }
6241 out:
6242         free_extent_state(cached_state);
6243         return ret;
6244 }
6245
6246 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6247                 __u64 start, __u64 len)
6248 {
6249         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6250 }
6251
6252 int btrfs_readpage(struct file *file, struct page *page)
6253 {
6254         struct extent_io_tree *tree;
6255         tree = &BTRFS_I(page->mapping->host)->io_tree;
6256         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6257 }
6258
6259 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6260 {
6261         struct extent_io_tree *tree;
6262
6263
6264         if (current->flags & PF_MEMALLOC) {
6265                 redirty_page_for_writepage(wbc, page);
6266                 unlock_page(page);
6267                 return 0;
6268         }
6269         tree = &BTRFS_I(page->mapping->host)->io_tree;
6270         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6271 }
6272
6273 int btrfs_writepages(struct address_space *mapping,
6274                      struct writeback_control *wbc)
6275 {
6276         struct extent_io_tree *tree;
6277
6278         tree = &BTRFS_I(mapping->host)->io_tree;
6279         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6280 }
6281
6282 static int
6283 btrfs_readpages(struct file *file, struct address_space *mapping,
6284                 struct list_head *pages, unsigned nr_pages)
6285 {
6286         struct extent_io_tree *tree;
6287         tree = &BTRFS_I(mapping->host)->io_tree;
6288         return extent_readpages(tree, mapping, pages, nr_pages,
6289                                 btrfs_get_extent);
6290 }
6291 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6292 {
6293         struct extent_io_tree *tree;
6294         struct extent_map_tree *map;
6295         int ret;
6296
6297         tree = &BTRFS_I(page->mapping->host)->io_tree;
6298         map = &BTRFS_I(page->mapping->host)->extent_tree;
6299         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6300         if (ret == 1) {
6301                 ClearPagePrivate(page);
6302                 set_page_private(page, 0);
6303                 page_cache_release(page);
6304         }
6305         return ret;
6306 }
6307
6308 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6309 {
6310         if (PageWriteback(page) || PageDirty(page))
6311                 return 0;
6312         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6313 }
6314
6315 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6316 {
6317         struct extent_io_tree *tree;
6318         struct btrfs_ordered_extent *ordered;
6319         struct extent_state *cached_state = NULL;
6320         u64 page_start = page_offset(page);
6321         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6322
6323
6324         /*
6325          * we have the page locked, so new writeback can't start,
6326          * and the dirty bit won't be cleared while we are here.
6327          *
6328          * Wait for IO on this page so that we can safely clear
6329          * the PagePrivate2 bit and do ordered accounting
6330          */
6331         wait_on_page_writeback(page);
6332
6333         tree = &BTRFS_I(page->mapping->host)->io_tree;
6334         if (offset) {
6335                 btrfs_releasepage(page, GFP_NOFS);
6336                 return;
6337         }
6338         lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6339                          GFP_NOFS);
6340         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6341                                            page_offset(page));
6342         if (ordered) {
6343                 /*
6344                  * IO on this page will never be started, so we need
6345                  * to account for any ordered extents now
6346                  */
6347                 clear_extent_bit(tree, page_start, page_end,
6348                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6349                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6350                                  &cached_state, GFP_NOFS);
6351                 /*
6352                  * whoever cleared the private bit is responsible
6353                  * for the finish_ordered_io
6354                  */
6355                 if (TestClearPagePrivate2(page)) {
6356                         btrfs_finish_ordered_io(page->mapping->host,
6357                                                 page_start, page_end);
6358                 }
6359                 btrfs_put_ordered_extent(ordered);
6360                 cached_state = NULL;
6361                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6362                                  GFP_NOFS);
6363         }
6364         clear_extent_bit(tree, page_start, page_end,
6365                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6366                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6367         __btrfs_releasepage(page, GFP_NOFS);
6368
6369         ClearPageChecked(page);
6370         if (PagePrivate(page)) {
6371                 ClearPagePrivate(page);
6372                 set_page_private(page, 0);
6373                 page_cache_release(page);
6374         }
6375 }
6376
6377 /*
6378  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6379  * called from a page fault handler when a page is first dirtied. Hence we must
6380  * be careful to check for EOF conditions here. We set the page up correctly
6381  * for a written page which means we get ENOSPC checking when writing into
6382  * holes and correct delalloc and unwritten extent mapping on filesystems that
6383  * support these features.
6384  *
6385  * We are not allowed to take the i_mutex here so we have to play games to
6386  * protect against truncate races as the page could now be beyond EOF.  Because
6387  * vmtruncate() writes the inode size before removing pages, once we have the
6388  * page lock we can determine safely if the page is beyond EOF. If it is not
6389  * beyond EOF, then the page is guaranteed safe against truncation until we
6390  * unlock the page.
6391  */
6392 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6393 {
6394         struct page *page = vmf->page;
6395         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6396         struct btrfs_root *root = BTRFS_I(inode)->root;
6397         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6398         struct btrfs_ordered_extent *ordered;
6399         struct extent_state *cached_state = NULL;
6400         char *kaddr;
6401         unsigned long zero_start;
6402         loff_t size;
6403         int ret;
6404         int reserved = 0;
6405         u64 page_start;
6406         u64 page_end;
6407
6408         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6409         if (!ret) {
6410                 ret = btrfs_update_time(vma->vm_file);
6411                 reserved = 1;
6412         }
6413         if (ret) {
6414                 if (ret == -ENOMEM)
6415                         ret = VM_FAULT_OOM;
6416                 else /* -ENOSPC, -EIO, etc */
6417                         ret = VM_FAULT_SIGBUS;
6418                 if (reserved)
6419                         goto out;
6420                 goto out_noreserve;
6421         }
6422
6423         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6424 again:
6425         lock_page(page);
6426         size = i_size_read(inode);
6427         page_start = page_offset(page);
6428         page_end = page_start + PAGE_CACHE_SIZE - 1;
6429
6430         if ((page->mapping != inode->i_mapping) ||
6431             (page_start >= size)) {
6432                 /* page got truncated out from underneath us */
6433                 goto out_unlock;
6434         }
6435         wait_on_page_writeback(page);
6436
6437         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6438                          GFP_NOFS);
6439         set_page_extent_mapped(page);
6440
6441         /*
6442          * we can't set the delalloc bits if there are pending ordered
6443          * extents.  Drop our locks and wait for them to finish
6444          */
6445         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6446         if (ordered) {
6447                 unlock_extent_cached(io_tree, page_start, page_end,
6448                                      &cached_state, GFP_NOFS);
6449                 unlock_page(page);
6450                 btrfs_start_ordered_extent(inode, ordered, 1);
6451                 btrfs_put_ordered_extent(ordered);
6452                 goto again;
6453         }
6454
6455         /*
6456          * XXX - page_mkwrite gets called every time the page is dirtied, even
6457          * if it was already dirty, so for space accounting reasons we need to
6458          * clear any delalloc bits for the range we are fixing to save.  There
6459          * is probably a better way to do this, but for now keep consistent with
6460          * prepare_pages in the normal write path.
6461          */
6462         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6463                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6464                           0, 0, &cached_state, GFP_NOFS);
6465
6466         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6467                                         &cached_state);
6468         if (ret) {
6469                 unlock_extent_cached(io_tree, page_start, page_end,
6470                                      &cached_state, GFP_NOFS);
6471                 ret = VM_FAULT_SIGBUS;
6472                 goto out_unlock;
6473         }
6474         ret = 0;
6475
6476         /* page is wholly or partially inside EOF */
6477         if (page_start + PAGE_CACHE_SIZE > size)
6478                 zero_start = size & ~PAGE_CACHE_MASK;
6479         else
6480                 zero_start = PAGE_CACHE_SIZE;
6481
6482         if (zero_start != PAGE_CACHE_SIZE) {
6483                 kaddr = kmap(page);
6484                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6485                 flush_dcache_page(page);
6486                 kunmap(page);
6487         }
6488         ClearPageChecked(page);
6489         set_page_dirty(page);
6490         SetPageUptodate(page);
6491
6492         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6493         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6494
6495         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6496
6497 out_unlock:
6498         if (!ret)
6499                 return VM_FAULT_LOCKED;
6500         unlock_page(page);
6501 out:
6502         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6503 out_noreserve:
6504         return ret;
6505 }
6506
6507 static int btrfs_truncate(struct inode *inode)
6508 {
6509         struct btrfs_root *root = BTRFS_I(inode)->root;
6510         struct btrfs_block_rsv *rsv;
6511         int ret;
6512         int err = 0;
6513         struct btrfs_trans_handle *trans;
6514         unsigned long nr;
6515         u64 mask = root->sectorsize - 1;
6516         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6517
6518         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6519         if (ret)
6520                 return ret;
6521
6522         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6523         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6524
6525         /*
6526          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6527          * 3 things going on here
6528          *
6529          * 1) We need to reserve space for our orphan item and the space to
6530          * delete our orphan item.  Lord knows we don't want to have a dangling
6531          * orphan item because we didn't reserve space to remove it.
6532          *
6533          * 2) We need to reserve space to update our inode.
6534          *
6535          * 3) We need to have something to cache all the space that is going to
6536          * be free'd up by the truncate operation, but also have some slack
6537          * space reserved in case it uses space during the truncate (thank you
6538          * very much snapshotting).
6539          *
6540          * And we need these to all be seperate.  The fact is we can use alot of
6541          * space doing the truncate, and we have no earthly idea how much space
6542          * we will use, so we need the truncate reservation to be seperate so it
6543          * doesn't end up using space reserved for updating the inode or
6544          * removing the orphan item.  We also need to be able to stop the
6545          * transaction and start a new one, which means we need to be able to
6546          * update the inode several times, and we have no idea of knowing how
6547          * many times that will be, so we can't just reserve 1 item for the
6548          * entirety of the opration, so that has to be done seperately as well.
6549          * Then there is the orphan item, which does indeed need to be held on
6550          * to for the whole operation, and we need nobody to touch this reserved
6551          * space except the orphan code.
6552          *
6553          * So that leaves us with
6554          *
6555          * 1) root->orphan_block_rsv - for the orphan deletion.
6556          * 2) rsv - for the truncate reservation, which we will steal from the
6557          * transaction reservation.
6558          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6559          * updating the inode.
6560          */
6561         rsv = btrfs_alloc_block_rsv(root);
6562         if (!rsv)
6563                 return -ENOMEM;
6564         rsv->size = min_size;
6565
6566         /*
6567          * 1 for the truncate slack space
6568          * 1 for the orphan item we're going to add
6569          * 1 for the orphan item deletion
6570          * 1 for updating the inode.
6571          */
6572         trans = btrfs_start_transaction(root, 4);
6573         if (IS_ERR(trans)) {
6574                 err = PTR_ERR(trans);
6575                 goto out;
6576         }
6577
6578         /* Migrate the slack space for the truncate to our reserve */
6579         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6580                                       min_size);
6581         BUG_ON(ret);
6582
6583         ret = btrfs_orphan_add(trans, inode);
6584         if (ret) {
6585                 btrfs_end_transaction(trans, root);
6586                 goto out;
6587         }
6588
6589         /*
6590          * setattr is responsible for setting the ordered_data_close flag,
6591          * but that is only tested during the last file release.  That
6592          * could happen well after the next commit, leaving a great big
6593          * window where new writes may get lost if someone chooses to write
6594          * to this file after truncating to zero
6595          *
6596          * The inode doesn't have any dirty data here, and so if we commit
6597          * this is a noop.  If someone immediately starts writing to the inode
6598          * it is very likely we'll catch some of their writes in this
6599          * transaction, and the commit will find this file on the ordered
6600          * data list with good things to send down.
6601          *
6602          * This is a best effort solution, there is still a window where
6603          * using truncate to replace the contents of the file will
6604          * end up with a zero length file after a crash.
6605          */
6606         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6607                 btrfs_add_ordered_operation(trans, root, inode);
6608
6609         while (1) {
6610                 ret = btrfs_block_rsv_refill(root, rsv, min_size);
6611                 if (ret) {
6612                         /*
6613                          * This can only happen with the original transaction we
6614                          * started above, every other time we shouldn't have a
6615                          * transaction started yet.
6616                          */
6617                         if (ret == -EAGAIN)
6618                                 goto end_trans;
6619                         err = ret;
6620                         break;
6621                 }
6622
6623                 if (!trans) {
6624                         /* Just need the 1 for updating the inode */
6625                         trans = btrfs_start_transaction(root, 1);
6626                         if (IS_ERR(trans)) {
6627                                 ret = err = PTR_ERR(trans);
6628                                 trans = NULL;
6629                                 break;
6630                         }
6631                 }
6632
6633                 trans->block_rsv = rsv;
6634
6635                 ret = btrfs_truncate_inode_items(trans, root, inode,
6636                                                  inode->i_size,
6637                                                  BTRFS_EXTENT_DATA_KEY);
6638                 if (ret != -EAGAIN) {
6639                         err = ret;
6640                         break;
6641                 }
6642
6643                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6644                 ret = btrfs_update_inode(trans, root, inode);
6645                 if (ret) {
6646                         err = ret;
6647                         break;
6648                 }
6649 end_trans:
6650                 nr = trans->blocks_used;
6651                 btrfs_end_transaction(trans, root);
6652                 trans = NULL;
6653                 btrfs_btree_balance_dirty(root, nr);
6654         }
6655
6656         if (ret == 0 && inode->i_nlink > 0) {
6657                 trans->block_rsv = root->orphan_block_rsv;
6658                 ret = btrfs_orphan_del(trans, inode);
6659                 if (ret)
6660                         err = ret;
6661         } else if (ret && inode->i_nlink > 0) {
6662                 /*
6663                  * Failed to do the truncate, remove us from the in memory
6664                  * orphan list.
6665                  */
6666                 ret = btrfs_orphan_del(NULL, inode);
6667         }
6668
6669         if (trans) {
6670                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6671                 ret = btrfs_update_inode(trans, root, inode);
6672                 if (ret && !err)
6673                         err = ret;
6674
6675                 nr = trans->blocks_used;
6676                 ret = btrfs_end_transaction(trans, root);
6677                 btrfs_btree_balance_dirty(root, nr);
6678         }
6679
6680 out:
6681         btrfs_free_block_rsv(root, rsv);
6682
6683         if (ret && !err)
6684                 err = ret;
6685
6686         return err;
6687 }
6688
6689 /*
6690  * create a new subvolume directory/inode (helper for the ioctl).
6691  */
6692 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6693                              struct btrfs_root *new_root, u64 new_dirid)
6694 {
6695         struct inode *inode;
6696         int err;
6697         u64 index = 0;
6698
6699         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6700                                 new_dirid, S_IFDIR | 0700, &index);
6701         if (IS_ERR(inode))
6702                 return PTR_ERR(inode);
6703         inode->i_op = &btrfs_dir_inode_operations;
6704         inode->i_fop = &btrfs_dir_file_operations;
6705
6706         set_nlink(inode, 1);
6707         btrfs_i_size_write(inode, 0);
6708
6709         err = btrfs_update_inode(trans, new_root, inode);
6710         BUG_ON(err);
6711
6712         iput(inode);
6713         return 0;
6714 }
6715
6716 struct inode *btrfs_alloc_inode(struct super_block *sb)
6717 {
6718         struct btrfs_inode *ei;
6719         struct inode *inode;
6720
6721         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6722         if (!ei)
6723                 return NULL;
6724
6725         ei->root = NULL;
6726         ei->space_info = NULL;
6727         ei->generation = 0;
6728         ei->sequence = 0;
6729         ei->last_trans = 0;
6730         ei->last_sub_trans = 0;
6731         ei->logged_trans = 0;
6732         ei->delalloc_bytes = 0;
6733         ei->disk_i_size = 0;
6734         ei->flags = 0;
6735         ei->csum_bytes = 0;
6736         ei->index_cnt = (u64)-1;
6737         ei->last_unlink_trans = 0;
6738
6739         spin_lock_init(&ei->lock);
6740         ei->outstanding_extents = 0;
6741         ei->reserved_extents = 0;
6742
6743         ei->ordered_data_close = 0;
6744         ei->orphan_meta_reserved = 0;
6745         ei->dummy_inode = 0;
6746         ei->in_defrag = 0;
6747         ei->delalloc_meta_reserved = 0;
6748         ei->force_compress = BTRFS_COMPRESS_NONE;
6749
6750         ei->delayed_node = NULL;
6751
6752         inode = &ei->vfs_inode;
6753         extent_map_tree_init(&ei->extent_tree);
6754         extent_io_tree_init(&ei->io_tree, &inode->i_data);
6755         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6756         mutex_init(&ei->log_mutex);
6757         mutex_init(&ei->delalloc_mutex);
6758         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6759         INIT_LIST_HEAD(&ei->i_orphan);
6760         INIT_LIST_HEAD(&ei->delalloc_inodes);
6761         INIT_LIST_HEAD(&ei->ordered_operations);
6762         RB_CLEAR_NODE(&ei->rb_node);
6763
6764         return inode;
6765 }
6766
6767 static void btrfs_i_callback(struct rcu_head *head)
6768 {
6769         struct inode *inode = container_of(head, struct inode, i_rcu);
6770         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6771 }
6772
6773 void btrfs_destroy_inode(struct inode *inode)
6774 {
6775         struct btrfs_ordered_extent *ordered;
6776         struct btrfs_root *root = BTRFS_I(inode)->root;
6777
6778         WARN_ON(!list_empty(&inode->i_dentry));
6779         WARN_ON(inode->i_data.nrpages);
6780         WARN_ON(BTRFS_I(inode)->outstanding_extents);
6781         WARN_ON(BTRFS_I(inode)->reserved_extents);
6782         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
6783         WARN_ON(BTRFS_I(inode)->csum_bytes);
6784
6785         /*
6786          * This can happen where we create an inode, but somebody else also
6787          * created the same inode and we need to destroy the one we already
6788          * created.
6789          */
6790         if (!root)
6791                 goto free;
6792
6793         /*
6794          * Make sure we're properly removed from the ordered operation
6795          * lists.
6796          */
6797         smp_mb();
6798         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6799                 spin_lock(&root->fs_info->ordered_extent_lock);
6800                 list_del_init(&BTRFS_I(inode)->ordered_operations);
6801                 spin_unlock(&root->fs_info->ordered_extent_lock);
6802         }
6803
6804         spin_lock(&root->orphan_lock);
6805         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6806                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6807                        (unsigned long long)btrfs_ino(inode));
6808                 list_del_init(&BTRFS_I(inode)->i_orphan);
6809         }
6810         spin_unlock(&root->orphan_lock);
6811
6812         while (1) {
6813                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6814                 if (!ordered)
6815                         break;
6816                 else {
6817                         printk(KERN_ERR "btrfs found ordered "
6818                                "extent %llu %llu on inode cleanup\n",
6819                                (unsigned long long)ordered->file_offset,
6820                                (unsigned long long)ordered->len);
6821                         btrfs_remove_ordered_extent(inode, ordered);
6822                         btrfs_put_ordered_extent(ordered);
6823                         btrfs_put_ordered_extent(ordered);
6824                 }
6825         }
6826         inode_tree_del(inode);
6827         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6828 free:
6829         btrfs_remove_delayed_node(inode);
6830         call_rcu(&inode->i_rcu, btrfs_i_callback);
6831 }
6832
6833 int btrfs_drop_inode(struct inode *inode)
6834 {
6835         struct btrfs_root *root = BTRFS_I(inode)->root;
6836
6837         if (btrfs_root_refs(&root->root_item) == 0 &&
6838             !btrfs_is_free_space_inode(root, inode))
6839                 return 1;
6840         else
6841                 return generic_drop_inode(inode);
6842 }
6843
6844 static void init_once(void *foo)
6845 {
6846         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6847
6848         inode_init_once(&ei->vfs_inode);
6849 }
6850
6851 void btrfs_destroy_cachep(void)
6852 {
6853         if (btrfs_inode_cachep)
6854                 kmem_cache_destroy(btrfs_inode_cachep);
6855         if (btrfs_trans_handle_cachep)
6856                 kmem_cache_destroy(btrfs_trans_handle_cachep);
6857         if (btrfs_transaction_cachep)
6858                 kmem_cache_destroy(btrfs_transaction_cachep);
6859         if (btrfs_path_cachep)
6860                 kmem_cache_destroy(btrfs_path_cachep);
6861         if (btrfs_free_space_cachep)
6862                 kmem_cache_destroy(btrfs_free_space_cachep);
6863 }
6864
6865 int btrfs_init_cachep(void)
6866 {
6867         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6868                         sizeof(struct btrfs_inode), 0,
6869                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6870         if (!btrfs_inode_cachep)
6871                 goto fail;
6872
6873         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6874                         sizeof(struct btrfs_trans_handle), 0,
6875                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6876         if (!btrfs_trans_handle_cachep)
6877                 goto fail;
6878
6879         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6880                         sizeof(struct btrfs_transaction), 0,
6881                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6882         if (!btrfs_transaction_cachep)
6883                 goto fail;
6884
6885         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6886                         sizeof(struct btrfs_path), 0,
6887                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6888         if (!btrfs_path_cachep)
6889                 goto fail;
6890
6891         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6892                         sizeof(struct btrfs_free_space), 0,
6893                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6894         if (!btrfs_free_space_cachep)
6895                 goto fail;
6896
6897         return 0;
6898 fail:
6899         btrfs_destroy_cachep();
6900         return -ENOMEM;
6901 }
6902
6903 static int btrfs_getattr(struct vfsmount *mnt,
6904                          struct dentry *dentry, struct kstat *stat)
6905 {
6906         struct inode *inode = dentry->d_inode;
6907         u32 blocksize = inode->i_sb->s_blocksize;
6908
6909         generic_fillattr(inode, stat);
6910         stat->dev = BTRFS_I(inode)->root->anon_dev;
6911         stat->blksize = PAGE_CACHE_SIZE;
6912         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
6913                 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
6914         return 0;
6915 }
6916
6917 /*
6918  * If a file is moved, it will inherit the cow and compression flags of the new
6919  * directory.
6920  */
6921 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6922 {
6923         struct btrfs_inode *b_dir = BTRFS_I(dir);
6924         struct btrfs_inode *b_inode = BTRFS_I(inode);
6925
6926         if (b_dir->flags & BTRFS_INODE_NODATACOW)
6927                 b_inode->flags |= BTRFS_INODE_NODATACOW;
6928         else
6929                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6930
6931         if (b_dir->flags & BTRFS_INODE_COMPRESS)
6932                 b_inode->flags |= BTRFS_INODE_COMPRESS;
6933         else
6934                 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6935 }
6936
6937 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6938                            struct inode *new_dir, struct dentry *new_dentry)
6939 {
6940         struct btrfs_trans_handle *trans;
6941         struct btrfs_root *root = BTRFS_I(old_dir)->root;
6942         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6943         struct inode *new_inode = new_dentry->d_inode;
6944         struct inode *old_inode = old_dentry->d_inode;
6945         struct timespec ctime = CURRENT_TIME;
6946         u64 index = 0;
6947         u64 root_objectid;
6948         int ret;
6949         u64 old_ino = btrfs_ino(old_inode);
6950
6951         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6952                 return -EPERM;
6953
6954         /* we only allow rename subvolume link between subvolumes */
6955         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6956                 return -EXDEV;
6957
6958         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6959             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
6960                 return -ENOTEMPTY;
6961
6962         if (S_ISDIR(old_inode->i_mode) && new_inode &&
6963             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6964                 return -ENOTEMPTY;
6965         /*
6966          * we're using rename to replace one file with another.
6967          * and the replacement file is large.  Start IO on it now so
6968          * we don't add too much work to the end of the transaction
6969          */
6970         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6971             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6972                 filemap_flush(old_inode->i_mapping);
6973
6974         /* close the racy window with snapshot create/destroy ioctl */
6975         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
6976                 down_read(&root->fs_info->subvol_sem);
6977         /*
6978          * We want to reserve the absolute worst case amount of items.  So if
6979          * both inodes are subvols and we need to unlink them then that would
6980          * require 4 item modifications, but if they are both normal inodes it
6981          * would require 5 item modifications, so we'll assume their normal
6982          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6983          * should cover the worst case number of items we'll modify.
6984          */
6985         trans = btrfs_start_transaction(root, 20);
6986         if (IS_ERR(trans)) {
6987                 ret = PTR_ERR(trans);
6988                 goto out_notrans;
6989         }
6990
6991         if (dest != root)
6992                 btrfs_record_root_in_trans(trans, dest);
6993
6994         ret = btrfs_set_inode_index(new_dir, &index);
6995         if (ret)
6996                 goto out_fail;
6997
6998         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6999                 /* force full log commit if subvolume involved. */
7000                 root->fs_info->last_trans_log_full_commit = trans->transid;
7001         } else {
7002                 ret = btrfs_insert_inode_ref(trans, dest,
7003                                              new_dentry->d_name.name,
7004                                              new_dentry->d_name.len,
7005                                              old_ino,
7006                                              btrfs_ino(new_dir), index);
7007                 if (ret)
7008                         goto out_fail;
7009                 /*
7010                  * this is an ugly little race, but the rename is required
7011                  * to make sure that if we crash, the inode is either at the
7012                  * old name or the new one.  pinning the log transaction lets
7013                  * us make sure we don't allow a log commit to come in after
7014                  * we unlink the name but before we add the new name back in.
7015                  */
7016                 btrfs_pin_log_trans(root);
7017         }
7018         /*
7019          * make sure the inode gets flushed if it is replacing
7020          * something.
7021          */
7022         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7023                 btrfs_add_ordered_operation(trans, root, old_inode);
7024
7025         old_dir->i_ctime = old_dir->i_mtime = ctime;
7026         new_dir->i_ctime = new_dir->i_mtime = ctime;
7027         old_inode->i_ctime = ctime;
7028
7029         if (old_dentry->d_parent != new_dentry->d_parent)
7030                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7031
7032         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7033                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7034                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7035                                         old_dentry->d_name.name,
7036                                         old_dentry->d_name.len);
7037         } else {
7038                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7039                                         old_dentry->d_inode,
7040                                         old_dentry->d_name.name,
7041                                         old_dentry->d_name.len);
7042                 if (!ret)
7043                         ret = btrfs_update_inode(trans, root, old_inode);
7044         }
7045         BUG_ON(ret);
7046
7047         if (new_inode) {
7048                 new_inode->i_ctime = CURRENT_TIME;
7049                 if (unlikely(btrfs_ino(new_inode) ==
7050                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7051                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7052                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7053                                                 root_objectid,
7054                                                 new_dentry->d_name.name,
7055                                                 new_dentry->d_name.len);
7056                         BUG_ON(new_inode->i_nlink == 0);
7057                 } else {
7058                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7059                                                  new_dentry->d_inode,
7060                                                  new_dentry->d_name.name,
7061                                                  new_dentry->d_name.len);
7062                 }
7063                 BUG_ON(ret);
7064                 if (new_inode->i_nlink == 0) {
7065                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7066                         BUG_ON(ret);
7067                 }
7068         }
7069
7070         fixup_inode_flags(new_dir, old_inode);
7071
7072         ret = btrfs_add_link(trans, new_dir, old_inode,
7073                              new_dentry->d_name.name,
7074                              new_dentry->d_name.len, 0, index);
7075         BUG_ON(ret);
7076
7077         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7078                 struct dentry *parent = new_dentry->d_parent;
7079                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7080                 btrfs_end_log_trans(root);
7081         }
7082 out_fail:
7083         btrfs_end_transaction(trans, root);
7084 out_notrans:
7085         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7086                 up_read(&root->fs_info->subvol_sem);
7087
7088         return ret;
7089 }
7090
7091 /*
7092  * some fairly slow code that needs optimization. This walks the list
7093  * of all the inodes with pending delalloc and forces them to disk.
7094  */
7095 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7096 {
7097         struct list_head *head = &root->fs_info->delalloc_inodes;
7098         struct btrfs_inode *binode;
7099         struct inode *inode;
7100
7101         if (root->fs_info->sb->s_flags & MS_RDONLY)
7102                 return -EROFS;
7103
7104         spin_lock(&root->fs_info->delalloc_lock);
7105         while (!list_empty(head)) {
7106                 binode = list_entry(head->next, struct btrfs_inode,
7107                                     delalloc_inodes);
7108                 inode = igrab(&binode->vfs_inode);
7109                 if (!inode)
7110                         list_del_init(&binode->delalloc_inodes);
7111                 spin_unlock(&root->fs_info->delalloc_lock);
7112                 if (inode) {
7113                         filemap_flush(inode->i_mapping);
7114                         if (delay_iput)
7115                                 btrfs_add_delayed_iput(inode);
7116                         else
7117                                 iput(inode);
7118                 }
7119                 cond_resched();
7120                 spin_lock(&root->fs_info->delalloc_lock);
7121         }
7122         spin_unlock(&root->fs_info->delalloc_lock);
7123
7124         /* the filemap_flush will queue IO into the worker threads, but
7125          * we have to make sure the IO is actually started and that
7126          * ordered extents get created before we return
7127          */
7128         atomic_inc(&root->fs_info->async_submit_draining);
7129         while (atomic_read(&root->fs_info->nr_async_submits) ||
7130               atomic_read(&root->fs_info->async_delalloc_pages)) {
7131                 wait_event(root->fs_info->async_submit_wait,
7132                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7133                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7134         }
7135         atomic_dec(&root->fs_info->async_submit_draining);
7136         return 0;
7137 }
7138
7139 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7140                          const char *symname)
7141 {
7142         struct btrfs_trans_handle *trans;
7143         struct btrfs_root *root = BTRFS_I(dir)->root;
7144         struct btrfs_path *path;
7145         struct btrfs_key key;
7146         struct inode *inode = NULL;
7147         int err;
7148         int drop_inode = 0;
7149         u64 objectid;
7150         u64 index = 0 ;
7151         int name_len;
7152         int datasize;
7153         unsigned long ptr;
7154         struct btrfs_file_extent_item *ei;
7155         struct extent_buffer *leaf;
7156         unsigned long nr = 0;
7157
7158         name_len = strlen(symname) + 1;
7159         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7160                 return -ENAMETOOLONG;
7161
7162         /*
7163          * 2 items for inode item and ref
7164          * 2 items for dir items
7165          * 1 item for xattr if selinux is on
7166          */
7167         trans = btrfs_start_transaction(root, 5);
7168         if (IS_ERR(trans))
7169                 return PTR_ERR(trans);
7170
7171         err = btrfs_find_free_ino(root, &objectid);
7172         if (err)
7173                 goto out_unlock;
7174
7175         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7176                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7177                                 S_IFLNK|S_IRWXUGO, &index);
7178         if (IS_ERR(inode)) {
7179                 err = PTR_ERR(inode);
7180                 goto out_unlock;
7181         }
7182
7183         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7184         if (err) {
7185                 drop_inode = 1;
7186                 goto out_unlock;
7187         }
7188
7189         /*
7190         * If the active LSM wants to access the inode during
7191         * d_instantiate it needs these. Smack checks to see
7192         * if the filesystem supports xattrs by looking at the
7193         * ops vector.
7194         */
7195         inode->i_fop = &btrfs_file_operations;
7196         inode->i_op = &btrfs_file_inode_operations;
7197
7198         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7199         if (err)
7200                 drop_inode = 1;
7201         else {
7202                 inode->i_mapping->a_ops = &btrfs_aops;
7203                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7204                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7205         }
7206         if (drop_inode)
7207                 goto out_unlock;
7208
7209         path = btrfs_alloc_path();
7210         if (!path) {
7211                 err = -ENOMEM;
7212                 drop_inode = 1;
7213                 goto out_unlock;
7214         }
7215         key.objectid = btrfs_ino(inode);
7216         key.offset = 0;
7217         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7218         datasize = btrfs_file_extent_calc_inline_size(name_len);
7219         err = btrfs_insert_empty_item(trans, root, path, &key,
7220                                       datasize);
7221         if (err) {
7222                 drop_inode = 1;
7223                 btrfs_free_path(path);
7224                 goto out_unlock;
7225         }
7226         leaf = path->nodes[0];
7227         ei = btrfs_item_ptr(leaf, path->slots[0],
7228                             struct btrfs_file_extent_item);
7229         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7230         btrfs_set_file_extent_type(leaf, ei,
7231                                    BTRFS_FILE_EXTENT_INLINE);
7232         btrfs_set_file_extent_encryption(leaf, ei, 0);
7233         btrfs_set_file_extent_compression(leaf, ei, 0);
7234         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7235         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7236
7237         ptr = btrfs_file_extent_inline_start(ei);
7238         write_extent_buffer(leaf, symname, ptr, name_len);
7239         btrfs_mark_buffer_dirty(leaf);
7240         btrfs_free_path(path);
7241
7242         inode->i_op = &btrfs_symlink_inode_operations;
7243         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7244         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7245         inode_set_bytes(inode, name_len);
7246         btrfs_i_size_write(inode, name_len - 1);
7247         err = btrfs_update_inode(trans, root, inode);
7248         if (err)
7249                 drop_inode = 1;
7250
7251 out_unlock:
7252         if (!err)
7253                 d_instantiate(dentry, inode);
7254         nr = trans->blocks_used;
7255         btrfs_end_transaction(trans, root);
7256         if (drop_inode) {
7257                 inode_dec_link_count(inode);
7258                 iput(inode);
7259         }
7260         btrfs_btree_balance_dirty(root, nr);
7261         return err;
7262 }
7263
7264 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7265                                        u64 start, u64 num_bytes, u64 min_size,
7266                                        loff_t actual_len, u64 *alloc_hint,
7267                                        struct btrfs_trans_handle *trans)
7268 {
7269         struct btrfs_root *root = BTRFS_I(inode)->root;
7270         struct btrfs_key ins;
7271         u64 cur_offset = start;
7272         u64 i_size;
7273         int ret = 0;
7274         bool own_trans = true;
7275
7276         if (trans)
7277                 own_trans = false;
7278         while (num_bytes > 0) {
7279                 if (own_trans) {
7280                         trans = btrfs_start_transaction(root, 3);
7281                         if (IS_ERR(trans)) {
7282                                 ret = PTR_ERR(trans);
7283                                 break;
7284                         }
7285                 }
7286
7287                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7288                                            0, *alloc_hint, (u64)-1, &ins, 1);
7289                 if (ret) {
7290                         if (own_trans)
7291                                 btrfs_end_transaction(trans, root);
7292                         break;
7293                 }
7294
7295                 ret = insert_reserved_file_extent(trans, inode,
7296                                                   cur_offset, ins.objectid,
7297                                                   ins.offset, ins.offset,
7298                                                   ins.offset, 0, 0, 0,
7299                                                   BTRFS_FILE_EXTENT_PREALLOC);
7300                 BUG_ON(ret);
7301                 btrfs_drop_extent_cache(inode, cur_offset,
7302                                         cur_offset + ins.offset -1, 0);
7303
7304                 num_bytes -= ins.offset;
7305                 cur_offset += ins.offset;
7306                 *alloc_hint = ins.objectid + ins.offset;
7307
7308                 inode->i_ctime = CURRENT_TIME;
7309                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7310                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7311                     (actual_len > inode->i_size) &&
7312                     (cur_offset > inode->i_size)) {
7313                         if (cur_offset > actual_len)
7314                                 i_size = actual_len;
7315                         else
7316                                 i_size = cur_offset;
7317                         i_size_write(inode, i_size);
7318                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7319                 }
7320
7321                 ret = btrfs_update_inode(trans, root, inode);
7322                 BUG_ON(ret);
7323
7324                 if (own_trans)
7325                         btrfs_end_transaction(trans, root);
7326         }
7327         return ret;
7328 }
7329
7330 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7331                               u64 start, u64 num_bytes, u64 min_size,
7332                               loff_t actual_len, u64 *alloc_hint)
7333 {
7334         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7335                                            min_size, actual_len, alloc_hint,
7336                                            NULL);
7337 }
7338
7339 int btrfs_prealloc_file_range_trans(struct inode *inode,
7340                                     struct btrfs_trans_handle *trans, int mode,
7341                                     u64 start, u64 num_bytes, u64 min_size,
7342                                     loff_t actual_len, u64 *alloc_hint)
7343 {
7344         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7345                                            min_size, actual_len, alloc_hint, trans);
7346 }
7347
7348 static int btrfs_set_page_dirty(struct page *page)
7349 {
7350         return __set_page_dirty_nobuffers(page);
7351 }
7352
7353 static int btrfs_permission(struct inode *inode, int mask)
7354 {
7355         struct btrfs_root *root = BTRFS_I(inode)->root;
7356         umode_t mode = inode->i_mode;
7357
7358         if (mask & MAY_WRITE &&
7359             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7360                 if (btrfs_root_readonly(root))
7361                         return -EROFS;
7362                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7363                         return -EACCES;
7364         }
7365         return generic_permission(inode, mask);
7366 }
7367
7368 static const struct inode_operations btrfs_dir_inode_operations = {
7369         .getattr        = btrfs_getattr,
7370         .lookup         = btrfs_lookup,
7371         .create         = btrfs_create,
7372         .unlink         = btrfs_unlink,
7373         .link           = btrfs_link,
7374         .mkdir          = btrfs_mkdir,
7375         .rmdir          = btrfs_rmdir,
7376         .rename         = btrfs_rename,
7377         .symlink        = btrfs_symlink,
7378         .setattr        = btrfs_setattr,
7379         .mknod          = btrfs_mknod,
7380         .setxattr       = btrfs_setxattr,
7381         .getxattr       = btrfs_getxattr,
7382         .listxattr      = btrfs_listxattr,
7383         .removexattr    = btrfs_removexattr,
7384         .permission     = btrfs_permission,
7385         .get_acl        = btrfs_get_acl,
7386 };
7387 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7388         .lookup         = btrfs_lookup,
7389         .permission     = btrfs_permission,
7390         .get_acl        = btrfs_get_acl,
7391 };
7392
7393 static const struct file_operations btrfs_dir_file_operations = {
7394         .llseek         = generic_file_llseek,
7395         .read           = generic_read_dir,
7396         .readdir        = btrfs_real_readdir,
7397         .unlocked_ioctl = btrfs_ioctl,
7398 #ifdef CONFIG_COMPAT
7399         .compat_ioctl   = btrfs_ioctl,
7400 #endif
7401         .release        = btrfs_release_file,
7402         .fsync          = btrfs_sync_file,
7403 };
7404
7405 static struct extent_io_ops btrfs_extent_io_ops = {
7406         .fill_delalloc = run_delalloc_range,
7407         .submit_bio_hook = btrfs_submit_bio_hook,
7408         .merge_bio_hook = btrfs_merge_bio_hook,
7409         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7410         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7411         .writepage_start_hook = btrfs_writepage_start_hook,
7412         .set_bit_hook = btrfs_set_bit_hook,
7413         .clear_bit_hook = btrfs_clear_bit_hook,
7414         .merge_extent_hook = btrfs_merge_extent_hook,
7415         .split_extent_hook = btrfs_split_extent_hook,
7416 };
7417
7418 /*
7419  * btrfs doesn't support the bmap operation because swapfiles
7420  * use bmap to make a mapping of extents in the file.  They assume
7421  * these extents won't change over the life of the file and they
7422  * use the bmap result to do IO directly to the drive.
7423  *
7424  * the btrfs bmap call would return logical addresses that aren't
7425  * suitable for IO and they also will change frequently as COW
7426  * operations happen.  So, swapfile + btrfs == corruption.
7427  *
7428  * For now we're avoiding this by dropping bmap.
7429  */
7430 static const struct address_space_operations btrfs_aops = {
7431         .readpage       = btrfs_readpage,
7432         .writepage      = btrfs_writepage,
7433         .writepages     = btrfs_writepages,
7434         .readpages      = btrfs_readpages,
7435         .direct_IO      = btrfs_direct_IO,
7436         .invalidatepage = btrfs_invalidatepage,
7437         .releasepage    = btrfs_releasepage,
7438         .set_page_dirty = btrfs_set_page_dirty,
7439         .error_remove_page = generic_error_remove_page,
7440 };
7441
7442 static const struct address_space_operations btrfs_symlink_aops = {
7443         .readpage       = btrfs_readpage,
7444         .writepage      = btrfs_writepage,
7445         .invalidatepage = btrfs_invalidatepage,
7446         .releasepage    = btrfs_releasepage,
7447 };
7448
7449 static const struct inode_operations btrfs_file_inode_operations = {
7450         .getattr        = btrfs_getattr,
7451         .setattr        = btrfs_setattr,
7452         .setxattr       = btrfs_setxattr,
7453         .getxattr       = btrfs_getxattr,
7454         .listxattr      = btrfs_listxattr,
7455         .removexattr    = btrfs_removexattr,
7456         .permission     = btrfs_permission,
7457         .fiemap         = btrfs_fiemap,
7458         .get_acl        = btrfs_get_acl,
7459 };
7460 static const struct inode_operations btrfs_special_inode_operations = {
7461         .getattr        = btrfs_getattr,
7462         .setattr        = btrfs_setattr,
7463         .permission     = btrfs_permission,
7464         .setxattr       = btrfs_setxattr,
7465         .getxattr       = btrfs_getxattr,
7466         .listxattr      = btrfs_listxattr,
7467         .removexattr    = btrfs_removexattr,
7468         .get_acl        = btrfs_get_acl,
7469 };
7470 static const struct inode_operations btrfs_symlink_inode_operations = {
7471         .readlink       = generic_readlink,
7472         .follow_link    = page_follow_link_light,
7473         .put_link       = page_put_link,
7474         .getattr        = btrfs_getattr,
7475         .setattr        = btrfs_setattr,
7476         .permission     = btrfs_permission,
7477         .setxattr       = btrfs_setxattr,
7478         .getxattr       = btrfs_getxattr,
7479         .listxattr      = btrfs_listxattr,
7480         .removexattr    = btrfs_removexattr,
7481         .get_acl        = btrfs_get_acl,
7482 };
7483
7484 const struct dentry_operations btrfs_dentry_operations = {
7485         .d_delete       = btrfs_dentry_delete,
7486         .d_release      = btrfs_dentry_release,
7487 };