fs/btrfs/ioctl.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/fsnotify.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/highmem.h>
  27 #include <linux/time.h>
  28 #include <linux/init.h>
  29 #include <linux/string.h>
  30 #include <linux/backing-dev.h>
  31 #include <linux/mount.h>
  32 #include <linux/mpage.h>
  33 #include <linux/namei.h>
  34 #include <linux/swap.h>
  35 #include <linux/writeback.h>
  36 #include <linux/compat.h>
  37 #include <linux/bit_spinlock.h>
  38 #include <linux/security.h>
  39 #include <linux/xattr.h>
  40 #include <linux/mm.h>
  41 #include <linux/slab.h>
  42 #include <linux/blkdev.h>
  43 #include <linux/uuid.h>
  44 #include <linux/btrfs.h>
  45 #include <linux/uaccess.h>
  46 #include <linux/iversion.h>
  47 #include "ctree.h"
  48 #include "disk-io.h"
  49 #include "transaction.h"
  50 #include "btrfs_inode.h"
  51 #include "print-tree.h"
  52 #include "volumes.h"
  53 #include "locking.h"
  54 #include "inode-map.h"
  55 #include "backref.h"
  56 #include "rcu-string.h"
  57 #include "send.h"
  58 #include "dev-replace.h"
  59 #include "props.h"
  60 #include "sysfs.h"
  61 #include "qgroup.h"
  62 #include "tree-log.h"
  63 #include "compression.h"
  64
  65 #ifdef CONFIG_64BIT
  66 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
  67  * structures are incorrect, as the timespec structure from userspace
  68  * is 4 bytes too small. We define these alternatives here to teach
  69  * the kernel about the 32-bit struct packing.
  70  */
  71 struct btrfs_ioctl_timespec_32 {
  72         __u64 sec;
  73         __u32 nsec;
  74 } __attribute__ ((__packed__));
  75
  76 struct btrfs_ioctl_received_subvol_args_32 {
  77         char    uuid[BTRFS_UUID_SIZE];  /* in */
  78         __u64   stransid;               /* in */
  79         __u64   rtransid;               /* out */
  80         struct btrfs_ioctl_timespec_32 stime; /* in */
  81         struct btrfs_ioctl_timespec_32 rtime; /* out */
  82         __u64   flags;                  /* in */
  83         __u64   reserved[16];           /* in */
  84 } __attribute__ ((__packed__));
  85
  86 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
  87                                 struct btrfs_ioctl_received_subvol_args_32)
  88 #endif
  89
  90 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
  91 struct btrfs_ioctl_send_args_32 {
  92         __s64 send_fd;                  /* in */
  93         __u64 clone_sources_count;      /* in */
  94         compat_uptr_t clone_sources;    /* in */
  95         __u64 parent_root;              /* in */
  96         __u64 flags;                    /* in */
  97         __u64 reserved[4];              /* in */
  98 } __attribute__ ((__packed__));
  99
 100 #define BTRFS_IOC_SEND_32 _IOW(BTRFS_IOCTL_MAGIC, 38, \
 101                                struct btrfs_ioctl_send_args_32)
 102 #endif
 103
 104 static int btrfs_clone(struct inode *src, struct inode *inode,
 105                        u64 off, u64 olen, u64 olen_aligned, u64 destoff,
 106                        int no_time_update);
 107
 108 /* Mask out flags that are inappropriate for the given type of inode. */
 109 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
 110 {
 111         if (S_ISDIR(mode))
 112                 return flags;
 113         else if (S_ISREG(mode))
 114                 return flags & ~FS_DIRSYNC_FL;
 115         else
 116                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
 117 }
 118
 119 /*
 120  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
 121  */
 122 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
 123 {
 124         unsigned int iflags = 0;
 125
 126         if (flags & BTRFS_INODE_SYNC)
 127                 iflags |= FS_SYNC_FL;
 128         if (flags & BTRFS_INODE_IMMUTABLE)
 129                 iflags |= FS_IMMUTABLE_FL;
 130         if (flags & BTRFS_INODE_APPEND)
 131                 iflags |= FS_APPEND_FL;
 132         if (flags & BTRFS_INODE_NODUMP)
 133                 iflags |= FS_NODUMP_FL;
 134         if (flags & BTRFS_INODE_NOATIME)
 135                 iflags |= FS_NOATIME_FL;
 136         if (flags & BTRFS_INODE_DIRSYNC)
 137                 iflags |= FS_DIRSYNC_FL;
 138         if (flags & BTRFS_INODE_NODATACOW)
 139                 iflags |= FS_NOCOW_FL;
 140
 141         if (flags & BTRFS_INODE_NOCOMPRESS)
 142                 iflags |= FS_NOCOMP_FL;
 143         else if (flags & BTRFS_INODE_COMPRESS)
 144                 iflags |= FS_COMPR_FL;
 145
 146         return iflags;
 147 }
 148
 149 /*
 150  * Update inode->i_flags based on the btrfs internal flags.
 151  */
 152 void btrfs_update_iflags(struct inode *inode)
 153 {
 154         struct btrfs_inode *ip = BTRFS_I(inode);
 155         unsigned int new_fl = 0;
 156
 157         if (ip->flags & BTRFS_INODE_SYNC)
 158                 new_fl |= S_SYNC;
 159         if (ip->flags & BTRFS_INODE_IMMUTABLE)
 160                 new_fl |= S_IMMUTABLE;
 161         if (ip->flags & BTRFS_INODE_APPEND)
 162                 new_fl |= S_APPEND;
 163         if (ip->flags & BTRFS_INODE_NOATIME)
 164                 new_fl |= S_NOATIME;
 165         if (ip->flags & BTRFS_INODE_DIRSYNC)
 166                 new_fl |= S_DIRSYNC;
 167
 168         set_mask_bits(&inode->i_flags,
 169                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC,
 170                       new_fl);
 171 }
 172
 173 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 174 {
 175         struct btrfs_inode *ip = BTRFS_I(file_inode(file));
 176         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
 177
 178         if (copy_to_user(arg, &flags, sizeof(flags)))
 179                 return -EFAULT;
 180         return 0;
 181 }
 182
 183 static int check_flags(unsigned int flags)
 184 {
 185         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 186                       FS_NOATIME_FL | FS_NODUMP_FL | \
 187                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 188                       FS_NOCOMP_FL | FS_COMPR_FL |
 189                       FS_NOCOW_FL))
 190                 return -EOPNOTSUPP;
 191
 192         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 193                 return -EINVAL;
 194
 195         return 0;
 196 }
 197
 198 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 199 {
 200         struct inode *inode = file_inode(file);
 201         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 202         struct btrfs_inode *ip = BTRFS_I(inode);
 203         struct btrfs_root *root = ip->root;
 204         struct btrfs_trans_handle *trans;
 205         unsigned int flags, oldflags;
 206         int ret;
 207         u64 ip_oldflags;
 208         unsigned int i_oldflags;
 209         umode_t mode;
 210
 211         if (!inode_owner_or_capable(inode))
 212                 return -EPERM;
 213
 214         if (btrfs_root_readonly(root))
 215                 return -EROFS;
 216
 217         if (copy_from_user(&flags, arg, sizeof(flags)))
 218                 return -EFAULT;
 219
 220         ret = check_flags(flags);
 221         if (ret)
 222                 return ret;
 223
 224         ret = mnt_want_write_file(file);
 225         if (ret)
 226                 return ret;
 227
 228         inode_lock(inode);
 229
 230         ip_oldflags = ip->flags;
 231         i_oldflags = inode->i_flags;
 232         mode = inode->i_mode;
 233
 234         flags = btrfs_mask_flags(inode->i_mode, flags);
 235         oldflags = btrfs_flags_to_ioctl(ip->flags);
 236         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 237                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 238                         ret = -EPERM;
 239                         goto out_unlock;
 240                 }
 241         }
 242
 243         if (flags & FS_SYNC_FL)
 244                 ip->flags |= BTRFS_INODE_SYNC;
 245         else
 246                 ip->flags &= ~BTRFS_INODE_SYNC;
 247         if (flags & FS_IMMUTABLE_FL)
 248                 ip->flags |= BTRFS_INODE_IMMUTABLE;
 249         else
 250                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
 251         if (flags & FS_APPEND_FL)
 252                 ip->flags |= BTRFS_INODE_APPEND;
 253         else
 254                 ip->flags &= ~BTRFS_INODE_APPEND;
 255         if (flags & FS_NODUMP_FL)
 256                 ip->flags |= BTRFS_INODE_NODUMP;
 257         else
 258                 ip->flags &= ~BTRFS_INODE_NODUMP;
 259         if (flags & FS_NOATIME_FL)
 260                 ip->flags |= BTRFS_INODE_NOATIME;
 261         else
 262                 ip->flags &= ~BTRFS_INODE_NOATIME;
 263         if (flags & FS_DIRSYNC_FL)
 264                 ip->flags |= BTRFS_INODE_DIRSYNC;
 265         else
 266                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
 267         if (flags & FS_NOCOW_FL) {
 268                 if (S_ISREG(mode)) {
 269                         /*
 270                          * It's safe to turn csums off here, no extents exist.
 271                          * Otherwise we want the flag to reflect the real COW
 272                          * status of the file and will not set it.
 273                          */
 274                         if (inode->i_size == 0)
 275                                 ip->flags |= BTRFS_INODE_NODATACOW
 276                                            | BTRFS_INODE_NODATASUM;
 277                 } else {
 278                         ip->flags |= BTRFS_INODE_NODATACOW;
 279                 }
 280         } else {
 281                 /*
 282                  * Revert back under same assumptions as above
 283                  */
 284                 if (S_ISREG(mode)) {
 285                         if (inode->i_size == 0)
 286                                 ip->flags &= ~(BTRFS_INODE_NODATACOW
 287                                              | BTRFS_INODE_NODATASUM);
 288                 } else {
 289                         ip->flags &= ~BTRFS_INODE_NODATACOW;
 290                 }
 291         }
 292
 293         /*
 294          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 295          * flag may be changed automatically if compression code won't make
 296          * things smaller.
 297          */
 298         if (flags & FS_NOCOMP_FL) {
 299                 ip->flags &= ~BTRFS_INODE_COMPRESS;
 300                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
 301
 302                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 303                 if (ret && ret != -ENODATA)
 304                         goto out_drop;
 305         } else if (flags & FS_COMPR_FL) {
 306                 const char *comp;
 307
 308                 ip->flags |= BTRFS_INODE_COMPRESS;
 309                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
 310
 311                 comp = btrfs_compress_type2str(fs_info->compress_type);
 312                 if (!comp || comp[0] == 0)
 313                         comp = btrfs_compress_type2str(BTRFS_COMPRESS_ZLIB);
 314
 315                 ret = btrfs_set_prop(inode, "btrfs.compression",
 316                                      comp, strlen(comp), 0);
 317                 if (ret)
 318                         goto out_drop;
 319
 320         } else {
 321                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 322                 if (ret && ret != -ENODATA)
 323                         goto out_drop;
 324                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 325         }
 326
 327         trans = btrfs_start_transaction(root, 1);
 328         if (IS_ERR(trans)) {
 329                 ret = PTR_ERR(trans);
 330                 goto out_drop;
 331         }
 332
 333         btrfs_update_iflags(inode);
 334         inode_inc_iversion(inode);
 335         inode->i_ctime = current_time(inode);
 336         ret = btrfs_update_inode(trans, root, inode);
 337
 338         btrfs_end_transaction(trans);
 339  out_drop:
 340         if (ret) {
 341                 ip->flags = ip_oldflags;
 342                 inode->i_flags = i_oldflags;
 343         }
 344
 345  out_unlock:
 346         inode_unlock(inode);
 347         mnt_drop_write_file(file);
 348         return ret;
 349 }
 350
 351 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 352 {
 353         struct inode *inode = file_inode(file);
 354
 355         return put_user(inode->i_generation, arg);
 356 }
 357
 358 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
 359 {
 360         struct inode *inode = file_inode(file);
 361         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 362         struct btrfs_device *device;
 363         struct request_queue *q;
 364         struct fstrim_range range;
 365         u64 minlen = ULLONG_MAX;
 366         u64 num_devices = 0;
 367         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
 368         int ret;
 369
 370         if (!capable(CAP_SYS_ADMIN))
 371                 return -EPERM;
 372
 373         rcu_read_lock();
 374         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 375                                 dev_list) {
 376                 if (!device->bdev)
 377                         continue;
 378                 q = bdev_get_queue(device->bdev);
 379                 if (blk_queue_discard(q)) {
 380                         num_devices++;
 381                         minlen = min_t(u64, q->limits.discard_granularity,
 382                                      minlen);
 383                 }
 384         }
 385         rcu_read_unlock();
 386
 387         if (!num_devices)
 388                 return -EOPNOTSUPP;
 389         if (copy_from_user(&range, arg, sizeof(range)))
 390                 return -EFAULT;
 391         if (range.start > total_bytes ||
 392             range.len < fs_info->sb->s_blocksize)
 393                 return -EINVAL;
 394
 395         range.len = min(range.len, total_bytes - range.start);
 396         range.minlen = max(range.minlen, minlen);
 397         ret = btrfs_trim_fs(fs_info, &range);
 398         if (ret < 0)
 399                 return ret;
 400
 401         if (copy_to_user(arg, &range, sizeof(range)))
 402                 return -EFAULT;
 403
 404         return 0;
 405 }
 406
 407 int btrfs_is_empty_uuid(u8 *uuid)
 408 {
 409         int i;
 410
 411         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 412                 if (uuid[i])
 413                         return 0;
 414         }
 415         return 1;
 416 }
 417
 418 static noinline int create_subvol(struct inode *dir,
 419                                   struct dentry *dentry,
 420                                   const char *name, int namelen,
 421                                   u64 *async_transid,
 422                                   struct btrfs_qgroup_inherit *inherit)
 423 {
 424         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 425         struct btrfs_trans_handle *trans;
 426         struct btrfs_key key;
 427         struct btrfs_root_item *root_item;
 428         struct btrfs_inode_item *inode_item;
 429         struct extent_buffer *leaf;
 430         struct btrfs_root *root = BTRFS_I(dir)->root;
 431         struct btrfs_root *new_root;
 432         struct btrfs_block_rsv block_rsv;
 433         struct timespec cur_time = current_time(dir);
 434         struct inode *inode;
 435         int ret;
 436         int err;
 437         u64 objectid;
 438         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 439         u64 index = 0;
 440         u64 qgroup_reserved;
 441         uuid_le new_uuid;
 442
 443         root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
 444         if (!root_item)
 445                 return -ENOMEM;
 446
 447         ret = btrfs_find_free_objectid(fs_info->tree_root, &objectid);
 448         if (ret)
 449                 goto fail_free;
 450
 451         /*
 452          * Don't create subvolume whose level is not zero. Or qgroup will be
 453          * screwed up since it assumes subvolume qgroup's level to be 0.
 454          */
 455         if (btrfs_qgroup_level(objectid)) {
 456                 ret = -ENOSPC;
 457                 goto fail_free;
 458         }
 459
 460         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 461         /*
 462          * The same as the snapshot creation, please see the comment
 463          * of create_snapshot().
 464          */
 465         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
 466                                                8, &qgroup_reserved, false);
 467         if (ret)
 468                 goto fail_free;
 469
 470         trans = btrfs_start_transaction(root, 0);
 471         if (IS_ERR(trans)) {
 472                 ret = PTR_ERR(trans);
 473                 btrfs_subvolume_release_metadata(fs_info, &block_rsv);
 474                 goto fail_free;
 475         }
 476         trans->block_rsv = &block_rsv;
 477         trans->bytes_reserved = block_rsv.size;
 478
 479         ret = btrfs_qgroup_inherit(trans, fs_info, 0, objectid, inherit);
 480         if (ret)
 481                 goto fail;
 482
 483         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
 484         if (IS_ERR(leaf)) {
 485                 ret = PTR_ERR(leaf);
 486                 goto fail;
 487         }
 488
 489         memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
 490         btrfs_set_header_bytenr(leaf, leaf->start);
 491         btrfs_set_header_generation(leaf, trans->transid);
 492         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 493         btrfs_set_header_owner(leaf, objectid);
 494
 495         write_extent_buffer_fsid(leaf, fs_info->fsid);
 496         write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
 497         btrfs_mark_buffer_dirty(leaf);
 498
 499         inode_item = &root_item->inode;
 500         btrfs_set_stack_inode_generation(inode_item, 1);
 501         btrfs_set_stack_inode_size(inode_item, 3);
 502         btrfs_set_stack_inode_nlink(inode_item, 1);
 503         btrfs_set_stack_inode_nbytes(inode_item,
 504                                      fs_info->nodesize);
 505         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 506
 507         btrfs_set_root_flags(root_item, 0);
 508         btrfs_set_root_limit(root_item, 0);
 509         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 510
 511         btrfs_set_root_bytenr(root_item, leaf->start);
 512         btrfs_set_root_generation(root_item, trans->transid);
 513         btrfs_set_root_level(root_item, 0);
 514         btrfs_set_root_refs(root_item, 1);
 515         btrfs_set_root_used(root_item, leaf->len);
 516         btrfs_set_root_last_snapshot(root_item, 0);
 517
 518         btrfs_set_root_generation_v2(root_item,
 519                         btrfs_root_generation(root_item));
 520         uuid_le_gen(&new_uuid);
 521         memcpy(root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
 522         btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
 523         btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
 524         root_item->ctime = root_item->otime;
 525         btrfs_set_root_ctransid(root_item, trans->transid);
 526         btrfs_set_root_otransid(root_item, trans->transid);
 527
 528         btrfs_tree_unlock(leaf);
 529         free_extent_buffer(leaf);
 530         leaf = NULL;
 531
 532         btrfs_set_root_dirid(root_item, new_dirid);
 533
 534         key.objectid = objectid;
 535         key.offset = 0;
 536         key.type = BTRFS_ROOT_ITEM_KEY;
 537         ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
 538                                 root_item);
 539         if (ret)
 540                 goto fail;
 541
 542         key.offset = (u64)-1;
 543         new_root = btrfs_read_fs_root_no_name(fs_info, &key);
 544         if (IS_ERR(new_root)) {
 545                 ret = PTR_ERR(new_root);
 546                 btrfs_abort_transaction(trans, ret);
 547                 goto fail;
 548         }
 549
 550         btrfs_record_root_in_trans(trans, new_root);
 551
 552         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
 553         if (ret) {
 554                 /* We potentially lose an unused inode item here */
 555                 btrfs_abort_transaction(trans, ret);
 556                 goto fail;
 557         }
 558
 559         mutex_lock(&new_root->objectid_mutex);
 560         new_root->highest_objectid = new_dirid;
 561         mutex_unlock(&new_root->objectid_mutex);
 562
 563         /*
 564          * insert the directory item
 565          */
 566         ret = btrfs_set_inode_index(BTRFS_I(dir), &index);
 567         if (ret) {
 568                 btrfs_abort_transaction(trans, ret);
 569                 goto fail;
 570         }
 571
 572         ret = btrfs_insert_dir_item(trans, root,
 573                                     name, namelen, BTRFS_I(dir), &key,
 574                                     BTRFS_FT_DIR, index);
 575         if (ret) {
 576                 btrfs_abort_transaction(trans, ret);
 577                 goto fail;
 578         }
 579
 580         btrfs_i_size_write(BTRFS_I(dir), dir->i_size + namelen * 2);
 581         ret = btrfs_update_inode(trans, root, dir);
 582         BUG_ON(ret);
 583
 584         ret = btrfs_add_root_ref(trans, fs_info,
 585                                  objectid, root->root_key.objectid,
 586                                  btrfs_ino(BTRFS_I(dir)), index, name, namelen);
 587         BUG_ON(ret);
 588
 589         ret = btrfs_uuid_tree_add(trans, fs_info, root_item->uuid,
 590                                   BTRFS_UUID_KEY_SUBVOL, objectid);
 591         if (ret)
 592                 btrfs_abort_transaction(trans, ret);
 593
 594 fail:
 595         kfree(root_item);
 596         trans->block_rsv = NULL;
 597         trans->bytes_reserved = 0;
 598         btrfs_subvolume_release_metadata(fs_info, &block_rsv);
 599
 600         if (async_transid) {
 601                 *async_transid = trans->transid;
 602                 err = btrfs_commit_transaction_async(trans, 1);
 603                 if (err)
 604                         err = btrfs_commit_transaction(trans);
 605         } else {
 606                 err = btrfs_commit_transaction(trans);
 607         }
 608         if (err && !ret)
 609                 ret = err;
 610
 611         if (!ret) {
 612                 inode = btrfs_lookup_dentry(dir, dentry);
 613                 if (IS_ERR(inode))
 614                         return PTR_ERR(inode);
 615                 d_instantiate(dentry, inode);
 616         }
 617         return ret;
 618
 619 fail_free:
 620         kfree(root_item);
 621         return ret;
 622 }
 623
 624 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 625                            struct dentry *dentry,
 626                            u64 *async_transid, bool readonly,
 627                            struct btrfs_qgroup_inherit *inherit)
 628 {
 629         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 630         struct inode *inode;
 631         struct btrfs_pending_snapshot *pending_snapshot;
 632         struct btrfs_trans_handle *trans;
 633         int ret;
 634
 635         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 636                 return -EINVAL;
 637
 638         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
 639         if (!pending_snapshot)
 640                 return -ENOMEM;
 641
 642         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 643                         GFP_KERNEL);
 644         pending_snapshot->path = btrfs_alloc_path();
 645         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 646                 ret = -ENOMEM;
 647                 goto free_pending;
 648         }
 649
 650         atomic_inc(&root->will_be_snapshotted);
 651         smp_mb__after_atomic();
 652         /* wait for no snapshot writes */
 653         wait_event(root->subv_writers->wait,
 654                    percpu_counter_sum(&root->subv_writers->counter) == 0);
 655
 656         ret = btrfs_start_delalloc_inodes(root, 0);
 657         if (ret)
 658                 goto dec_and_free;
 659
 660         btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
 661
 662         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 663                              BTRFS_BLOCK_RSV_TEMP);
 664         /*
 665          * 1 - parent dir inode
 666          * 2 - dir entries
 667          * 1 - root item
 668          * 2 - root ref/backref
 669          * 1 - root of snapshot
 670          * 1 - UUID item
 671          */
 672         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 673                                         &pending_snapshot->block_rsv, 8,
 674                                         &pending_snapshot->qgroup_reserved,
 675                                         false);
 676         if (ret)
 677                 goto dec_and_free;
 678
 679         pending_snapshot->dentry = dentry;
 680         pending_snapshot->root = root;
 681         pending_snapshot->readonly = readonly;
 682         pending_snapshot->dir = dir;
 683         pending_snapshot->inherit = inherit;
 684
 685         trans = btrfs_start_transaction(root, 0);
 686         if (IS_ERR(trans)) {
 687                 ret = PTR_ERR(trans);
 688                 goto fail;
 689         }
 690
 691         spin_lock(&fs_info->trans_lock);
 692         list_add(&pending_snapshot->list,
 693                  &trans->transaction->pending_snapshots);
 694         spin_unlock(&fs_info->trans_lock);
 695         if (async_transid) {
 696                 *async_transid = trans->transid;
 697                 ret = btrfs_commit_transaction_async(trans, 1);
 698                 if (ret)
 699                         ret = btrfs_commit_transaction(trans);
 700         } else {
 701                 ret = btrfs_commit_transaction(trans);
 702         }
 703         if (ret)
 704                 goto fail;
 705
 706         ret = pending_snapshot->error;
 707         if (ret)
 708                 goto fail;
 709
 710         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 711         if (ret)
 712                 goto fail;
 713
 714         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 715         if (IS_ERR(inode)) {
 716                 ret = PTR_ERR(inode);
 717                 goto fail;
 718         }
 719
 720         d_instantiate(dentry, inode);
 721         ret = 0;
 722 fail:
 723         btrfs_subvolume_release_metadata(fs_info, &pending_snapshot->block_rsv);
 724 dec_and_free:
 725         if (atomic_dec_and_test(&root->will_be_snapshotted))
 726                 wake_up_atomic_t(&root->will_be_snapshotted);
 727 free_pending:
 728         kfree(pending_snapshot->root_item);
 729         btrfs_free_path(pending_snapshot->path);
 730         kfree(pending_snapshot);
 731
 732         return ret;
 733 }
 734
 735 /*  copy of may_delete in fs/namei.c()
 736  *      Check whether we can remove a link victim from directory dir, check
 737  *  whether the type of victim is right.
 738  *  1. We can't do it if dir is read-only (done in permission())
 739  *  2. We should have write and exec permissions on dir
 740  *  3. We can't remove anything from append-only dir
 741  *  4. We can't do anything with immutable dir (done in permission())
 742  *  5. If the sticky bit on dir is set we should either
 743  *      a. be owner of dir, or
 744  *      b. be owner of victim, or
 745  *      c. have CAP_FOWNER capability
 746  *  6. If the victim is append-only or immutable we can't do anything with
 747  *     links pointing to it.
 748  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 749  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 750  *  9. We can't remove a root or mountpoint.
 751  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 752  *     nfs_async_unlink().
 753  */
 754
 755 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 756 {
 757         int error;
 758
 759         if (d_really_is_negative(victim))
 760                 return -ENOENT;
 761
 762         BUG_ON(d_inode(victim->d_parent) != dir);
 763         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 764
 765         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 766         if (error)
 767                 return error;
 768         if (IS_APPEND(dir))
 769                 return -EPERM;
 770         if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
 771             IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
 772                 return -EPERM;
 773         if (isdir) {
 774                 if (!d_is_dir(victim))
 775                         return -ENOTDIR;
 776                 if (IS_ROOT(victim))
 777                         return -EBUSY;
 778         } else if (d_is_dir(victim))
 779                 return -EISDIR;
 780         if (IS_DEADDIR(dir))
 781                 return -ENOENT;
 782         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 783                 return -EBUSY;
 784         return 0;
 785 }
 786
 787 /* copy of may_create in fs/namei.c() */
 788 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 789 {
 790         if (d_really_is_positive(child))
 791                 return -EEXIST;
 792         if (IS_DEADDIR(dir))
 793                 return -ENOENT;
 794         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 795 }
 796
 797 /*
 798  * Create a new subvolume below @parent.  This is largely modeled after
 799  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 800  * inside this filesystem so it's quite a bit simpler.
 801  */
 802 static noinline int btrfs_mksubvol(const struct path *parent,
 803                                    const char *name, int namelen,
 804                                    struct btrfs_root *snap_src,
 805                                    u64 *async_transid, bool readonly,
 806                                    struct btrfs_qgroup_inherit *inherit)
 807 {
 808         struct inode *dir = d_inode(parent->dentry);
 809         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 810         struct dentry *dentry;
 811         int error;
 812
 813         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
 814         if (error == -EINTR)
 815                 return error;
 816
 817         dentry = lookup_one_len(name, parent->dentry, namelen);
 818         error = PTR_ERR(dentry);
 819         if (IS_ERR(dentry))
 820                 goto out_unlock;
 821
 822         error = btrfs_may_create(dir, dentry);
 823         if (error)
 824                 goto out_dput;
 825
 826         /*
 827          * even if this name doesn't exist, we may get hash collisions.
 828          * check for them now when we can safely fail
 829          */
 830         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 831                                                dir->i_ino, name,
 832                                                namelen);
 833         if (error)
 834                 goto out_dput;
 835
 836         down_read(&fs_info->subvol_sem);
 837
 838         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 839                 goto out_up_read;
 840
 841         if (snap_src) {
 842                 error = create_snapshot(snap_src, dir, dentry,
 843                                         async_transid, readonly, inherit);
 844         } else {
 845                 error = create_subvol(dir, dentry, name, namelen,
 846                                       async_transid, inherit);
 847         }
 848         if (!error)
 849                 fsnotify_mkdir(dir, dentry);
 850 out_up_read:
 851         up_read(&fs_info->subvol_sem);
 852 out_dput:
 853         dput(dentry);
 854 out_unlock:
 855         inode_unlock(dir);
 856         return error;
 857 }
 858
 859 /*
 860  * When we're defragging a range, we don't want to kick it off again
 861  * if it is really just waiting for delalloc to send it down.
 862  * If we find a nice big extent or delalloc range for the bytes in the
 863  * file you want to defrag, we return 0 to let you know to skip this
 864  * part of the file
 865  */
 866 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
 867 {
 868         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 869         struct extent_map *em = NULL;
 870         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 871         u64 end;
 872
 873         read_lock(&em_tree->lock);
 874         em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
 875         read_unlock(&em_tree->lock);
 876
 877         if (em) {
 878                 end = extent_map_end(em);
 879                 free_extent_map(em);
 880                 if (end - offset > thresh)
 881                         return 0;
 882         }
 883         /* if we already have a nice delalloc here, just stop */
 884         thresh /= 2;
 885         end = count_range_bits(io_tree, &offset, offset + thresh,
 886                                thresh, EXTENT_DELALLOC, 1);
 887         if (end >= thresh)
 888                 return 0;
 889         return 1;
 890 }
 891
 892 /*
 893  * helper function to walk through a file and find extents
 894  * newer than a specific transid, and smaller than thresh.
 895  *
 896  * This is used by the defragging code to find new and small
 897  * extents
 898  */
 899 static int find_new_extents(struct btrfs_root *root,
 900                             struct inode *inode, u64 newer_than,
 901                             u64 *off, u32 thresh)
 902 {
 903         struct btrfs_path *path;
 904         struct btrfs_key min_key;
 905         struct extent_buffer *leaf;
 906         struct btrfs_file_extent_item *extent;
 907         int type;
 908         int ret;
 909         u64 ino = btrfs_ino(BTRFS_I(inode));
 910
 911         path = btrfs_alloc_path();
 912         if (!path)
 913                 return -ENOMEM;
 914
 915         min_key.objectid = ino;
 916         min_key.type = BTRFS_EXTENT_DATA_KEY;
 917         min_key.offset = *off;
 918
 919         while (1) {
 920                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
 921                 if (ret != 0)
 922                         goto none;
 923 process_slot:
 924                 if (min_key.objectid != ino)
 925                         goto none;
 926                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
 927                         goto none;
 928
 929                 leaf = path->nodes[0];
 930                 extent = btrfs_item_ptr(leaf, path->slots[0],
 931                                         struct btrfs_file_extent_item);
 932
 933                 type = btrfs_file_extent_type(leaf, extent);
 934                 if (type == BTRFS_FILE_EXTENT_REG &&
 935                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
 936                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
 937                         *off = min_key.offset;
 938                         btrfs_free_path(path);
 939                         return 0;
 940                 }
 941
 942                 path->slots[0]++;
 943                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
 944                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
 945                         goto process_slot;
 946                 }
 947
 948                 if (min_key.offset == (u64)-1)
 949                         goto none;
 950
 951                 min_key.offset++;
 952                 btrfs_release_path(path);
 953         }
 954 none:
 955         btrfs_free_path(path);
 956         return -ENOENT;
 957 }
 958
 959 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
 960 {
 961         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 962         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 963         struct extent_map *em;
 964         u64 len = PAGE_SIZE;
 965
 966         /*
 967          * hopefully we have this extent in the tree already, try without
 968          * the full extent lock
 969          */
 970         read_lock(&em_tree->lock);
 971         em = lookup_extent_mapping(em_tree, start, len);
 972         read_unlock(&em_tree->lock);
 973
 974         if (!em) {
 975                 struct extent_state *cached = NULL;
 976                 u64 end = start + len - 1;
 977
 978                 /* get the big lock and read metadata off disk */
 979                 lock_extent_bits(io_tree, start, end, &cached);
 980                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
 981                 unlock_extent_cached(io_tree, start, end, &cached);
 982
 983                 if (IS_ERR(em))
 984                         return NULL;
 985         }
 986
 987         return em;
 988 }
 989
 990 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
 991 {
 992         struct extent_map *next;
 993         bool ret = true;
 994
 995         /* this is the last extent */
 996         if (em->start + em->len >= i_size_read(inode))
 997                 return false;
 998
 999         next = defrag_lookup_extent(inode, em->start + em->len);
1000         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1001                 ret = false;
1002         else if ((em->block_start + em->block_len == next->block_start) &&
1003                  (em->block_len > SZ_128K && next->block_len > SZ_128K))
1004                 ret = false;
1005
1006         free_extent_map(next);
1007         return ret;
1008 }
1009
1010 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1011                                u64 *last_len, u64 *skip, u64 *defrag_end,
1012                                int compress)
1013 {
1014         struct extent_map *em;
1015         int ret = 1;
1016         bool next_mergeable = true;
1017         bool prev_mergeable = true;
1018
1019         /*
1020          * make sure that once we start defragging an extent, we keep on
1021          * defragging it
1022          */
1023         if (start < *defrag_end)
1024                 return 1;
1025
1026         *skip = 0;
1027
1028         em = defrag_lookup_extent(inode, start);
1029         if (!em)
1030                 return 0;
1031
1032         /* this will cover holes, and inline extents */
1033         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1034                 ret = 0;
1035                 goto out;
1036         }
1037
1038         if (!*defrag_end)
1039                 prev_mergeable = false;
1040
1041         next_mergeable = defrag_check_next_extent(inode, em);
1042         /*
1043          * we hit a real extent, if it is big or the next extent is not a
1044          * real extent, don't bother defragging it
1045          */
1046         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1047             (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1048                 ret = 0;
1049 out:
1050         /*
1051          * last_len ends up being a counter of how many bytes we've defragged.
1052          * every time we choose not to defrag an extent, we reset *last_len
1053          * so that the next tiny extent will force a defrag.
1054          *
1055          * The end result of this is that tiny extents before a single big
1056          * extent will force at least part of that big extent to be defragged.
1057          */
1058         if (ret) {
1059                 *defrag_end = extent_map_end(em);
1060         } else {
1061                 *last_len = 0;
1062                 *skip = extent_map_end(em);
1063                 *defrag_end = 0;
1064         }
1065
1066         free_extent_map(em);
1067         return ret;
1068 }
1069
1070 /*
1071  * it doesn't do much good to defrag one or two pages
1072  * at a time.  This pulls in a nice chunk of pages
1073  * to COW and defrag.
1074  *
1075  * It also makes sure the delalloc code has enough
1076  * dirty data to avoid making new small extents as part
1077  * of the defrag
1078  *
1079  * It's a good idea to start RA on this range
1080  * before calling this.
1081  */
1082 static int cluster_pages_for_defrag(struct inode *inode,
1083                                     struct page **pages,
1084                                     unsigned long start_index,
1085                                     unsigned long num_pages)
1086 {
1087         unsigned long file_end;
1088         u64 isize = i_size_read(inode);
1089         u64 page_start;
1090         u64 page_end;
1091         u64 page_cnt;
1092         int ret;
1093         int i;
1094         int i_done;
1095         struct btrfs_ordered_extent *ordered;
1096         struct extent_state *cached_state = NULL;
1097         struct extent_io_tree *tree;
1098         struct extent_changeset *data_reserved = NULL;
1099         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1100
1101         file_end = (isize - 1) >> PAGE_SHIFT;
1102         if (!isize || start_index > file_end)
1103                 return 0;
1104
1105         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1106
1107         ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
1108                         start_index << PAGE_SHIFT,
1109                         page_cnt << PAGE_SHIFT);
1110         if (ret)
1111                 return ret;
1112         i_done = 0;
1113         tree = &BTRFS_I(inode)->io_tree;
1114
1115         /* step one, lock all the pages */
1116         for (i = 0; i < page_cnt; i++) {
1117                 struct page *page;
1118 again:
1119                 page = find_or_create_page(inode->i_mapping,
1120                                            start_index + i, mask);
1121                 if (!page)
1122                         break;
1123
1124                 page_start = page_offset(page);
1125                 page_end = page_start + PAGE_SIZE - 1;
1126                 while (1) {
1127                         lock_extent_bits(tree, page_start, page_end,
1128                                          &cached_state);
1129                         ordered = btrfs_lookup_ordered_extent(inode,
1130                                                               page_start);
1131                         unlock_extent_cached(tree, page_start, page_end,
1132                                              &cached_state);
1133                         if (!ordered)
1134                                 break;
1135
1136                         unlock_page(page);
1137                         btrfs_start_ordered_extent(inode, ordered, 1);
1138                         btrfs_put_ordered_extent(ordered);
1139                         lock_page(page);
1140                         /*
1141                          * we unlocked the page above, so we need check if
1142                          * it was released or not.
1143                          */
1144                         if (page->mapping != inode->i_mapping) {
1145                                 unlock_page(page);
1146                                 put_page(page);
1147                                 goto again;
1148                         }
1149                 }
1150
1151                 if (!PageUptodate(page)) {
1152                         btrfs_readpage(NULL, page);
1153                         lock_page(page);
1154                         if (!PageUptodate(page)) {
1155                                 unlock_page(page);
1156                                 put_page(page);
1157                                 ret = -EIO;
1158                                 break;
1159                         }
1160                 }
1161
1162                 if (page->mapping != inode->i_mapping) {
1163                         unlock_page(page);
1164                         put_page(page);
1165                         goto again;
1166                 }
1167
1168                 pages[i] = page;
1169                 i_done++;
1170         }
1171         if (!i_done || ret)
1172                 goto out;
1173
1174         if (!(inode->i_sb->s_flags & SB_ACTIVE))
1175                 goto out;
1176
1177         /*
1178          * so now we have a nice long stream of locked
1179          * and up to date pages, lets wait on them
1180          */
1181         for (i = 0; i < i_done; i++)
1182                 wait_on_page_writeback(pages[i]);
1183
1184         page_start = page_offset(pages[0]);
1185         page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1186
1187         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1188                          page_start, page_end - 1, &cached_state);
1189         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1190                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1191                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1192                           &cached_state);
1193
1194         if (i_done != page_cnt) {
1195                 spin_lock(&BTRFS_I(inode)->lock);
1196                 BTRFS_I(inode)->outstanding_extents++;
1197                 spin_unlock(&BTRFS_I(inode)->lock);
1198                 btrfs_delalloc_release_space(inode, data_reserved,
1199                                 start_index << PAGE_SHIFT,
1200                                 (page_cnt - i_done) << PAGE_SHIFT);
1201         }
1202
1203
1204         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1205                           &cached_state);
1206
1207         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1208                              page_start, page_end - 1, &cached_state);
1209
1210         for (i = 0; i < i_done; i++) {
1211                 clear_page_dirty_for_io(pages[i]);
1212                 ClearPageChecked(pages[i]);
1213                 set_page_extent_mapped(pages[i]);
1214                 set_page_dirty(pages[i]);
1215                 unlock_page(pages[i]);
1216                 put_page(pages[i]);
1217         }
1218         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
1219         extent_changeset_free(data_reserved);
1220         return i_done;
1221 out:
1222         for (i = 0; i < i_done; i++) {
1223                 unlock_page(pages[i]);
1224                 put_page(pages[i]);
1225         }
1226         btrfs_delalloc_release_space(inode, data_reserved,
1227                         start_index << PAGE_SHIFT,
1228                         page_cnt << PAGE_SHIFT);
1229         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
1230         extent_changeset_free(data_reserved);
1231         return ret;
1232
1233 }
1234
1235 int btrfs_defrag_file(struct inode *inode, struct file *file,
1236                       struct btrfs_ioctl_defrag_range_args *range,
1237                       u64 newer_than, unsigned long max_to_defrag)
1238 {
1239         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1240         struct btrfs_root *root = BTRFS_I(inode)->root;
1241         struct file_ra_state *ra = NULL;
1242         unsigned long last_index;
1243         u64 isize = i_size_read(inode);
1244         u64 last_len = 0;
1245         u64 skip = 0;
1246         u64 defrag_end = 0;
1247         u64 newer_off = range->start;
1248         unsigned long i;
1249         unsigned long ra_index = 0;
1250         int ret;
1251         int defrag_count = 0;
1252         int compress_type = BTRFS_COMPRESS_ZLIB;
1253         u32 extent_thresh = range->extent_thresh;
1254         unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1255         unsigned long cluster = max_cluster;
1256         u64 new_align = ~((u64)SZ_128K - 1);
1257         struct page **pages = NULL;
1258         bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
1259
1260         if (isize == 0)
1261                 return 0;
1262
1263         if (range->start >= isize)
1264                 return -EINVAL;
1265
1266         if (do_compress) {
1267                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1268                         return -EINVAL;
1269                 if (range->compress_type)
1270                         compress_type = range->compress_type;
1271         }
1272
1273         if (extent_thresh == 0)
1274                 extent_thresh = SZ_256K;
1275
1276         /*
1277          * If we were not given a file, allocate a readahead context. As
1278          * readahead is just an optimization, defrag will work without it so
1279          * we don't error out.
1280          */
1281         if (!file) {
1282                 ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1283                 if (ra)
1284                         file_ra_state_init(ra, inode->i_mapping);
1285         } else {
1286                 ra = &file->f_ra;
1287         }
1288
1289         pages = kmalloc_array(max_cluster, sizeof(struct page *), GFP_KERNEL);
1290         if (!pages) {
1291                 ret = -ENOMEM;
1292                 goto out_ra;
1293         }
1294
1295         /* find the last page to defrag */
1296         if (range->start + range->len > range->start) {
1297                 last_index = min_t(u64, isize - 1,
1298                          range->start + range->len - 1) >> PAGE_SHIFT;
1299         } else {
1300                 last_index = (isize - 1) >> PAGE_SHIFT;
1301         }
1302
1303         if (newer_than) {
1304                 ret = find_new_extents(root, inode, newer_than,
1305                                        &newer_off, SZ_64K);
1306                 if (!ret) {
1307                         range->start = newer_off;
1308                         /*
1309                          * we always align our defrag to help keep
1310                          * the extents in the file evenly spaced
1311                          */
1312                         i = (newer_off & new_align) >> PAGE_SHIFT;
1313                 } else
1314                         goto out_ra;
1315         } else {
1316                 i = range->start >> PAGE_SHIFT;
1317         }
1318         if (!max_to_defrag)
1319                 max_to_defrag = last_index - i + 1;
1320
1321         /*
1322          * make writeback starts from i, so the defrag range can be
1323          * written sequentially.
1324          */
1325         if (i < inode->i_mapping->writeback_index)
1326                 inode->i_mapping->writeback_index = i;
1327
1328         while (i <= last_index && defrag_count < max_to_defrag &&
1329                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1330                 /*
1331                  * make sure we stop running if someone unmounts
1332                  * the FS
1333                  */
1334                 if (!(inode->i_sb->s_flags & SB_ACTIVE))
1335                         break;
1336
1337                 if (btrfs_defrag_cancelled(fs_info)) {
1338                         btrfs_debug(fs_info, "defrag_file cancelled");
1339                         ret = -EAGAIN;
1340                         break;
1341                 }
1342
1343                 if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1344                                          extent_thresh, &last_len, &skip,
1345                                          &defrag_end, do_compress)){
1346                         unsigned long next;
1347                         /*
1348                          * the should_defrag function tells us how much to skip
1349                          * bump our counter by the suggested amount
1350                          */
1351                         next = DIV_ROUND_UP(skip, PAGE_SIZE);
1352                         i = max(i + 1, next);
1353                         continue;
1354                 }
1355
1356                 if (!newer_than) {
1357                         cluster = (PAGE_ALIGN(defrag_end) >>
1358                                    PAGE_SHIFT) - i;
1359                         cluster = min(cluster, max_cluster);
1360                 } else {
1361                         cluster = max_cluster;
1362                 }
1363
1364                 if (i + cluster > ra_index) {
1365                         ra_index = max(i, ra_index);
1366                         if (ra)
1367                                 page_cache_sync_readahead(inode->i_mapping, ra,
1368                                                 file, ra_index, cluster);
1369                         ra_index += cluster;
1370                 }
1371
1372                 inode_lock(inode);
1373                 if (do_compress)
1374                         BTRFS_I(inode)->defrag_compress = compress_type;
1375                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1376                 if (ret < 0) {
1377                         inode_unlock(inode);
1378                         goto out_ra;
1379                 }
1380
1381                 defrag_count += ret;
1382                 balance_dirty_pages_ratelimited(inode->i_mapping);
1383                 inode_unlock(inode);
1384
1385                 if (newer_than) {
1386                         if (newer_off == (u64)-1)
1387                                 break;
1388
1389                         if (ret > 0)
1390                                 i += ret;
1391
1392                         newer_off = max(newer_off + 1,
1393                                         (u64)i << PAGE_SHIFT);
1394
1395                         ret = find_new_extents(root, inode, newer_than,
1396                                                &newer_off, SZ_64K);
1397                         if (!ret) {
1398                                 range->start = newer_off;
1399                                 i = (newer_off & new_align) >> PAGE_SHIFT;
1400                         } else {
1401                                 break;
1402                         }
1403                 } else {
1404                         if (ret > 0) {
1405                                 i += ret;
1406                                 last_len += ret << PAGE_SHIFT;
1407                         } else {
1408                                 i++;
1409                                 last_len = 0;
1410                         }
1411                 }
1412         }
1413
1414         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1415                 filemap_flush(inode->i_mapping);
1416                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1417                              &BTRFS_I(inode)->runtime_flags))
1418                         filemap_flush(inode->i_mapping);
1419         }
1420
1421         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1422                 btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1423         } else if (range->compress_type == BTRFS_COMPRESS_ZSTD) {
1424                 btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
1425         }
1426
1427         ret = defrag_count;
1428
1429 out_ra:
1430         if (do_compress) {
1431                 inode_lock(inode);
1432                 BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1433                 inode_unlock(inode);
1434         }
1435         if (!file)
1436                 kfree(ra);
1437         kfree(pages);
1438         return ret;
1439 }
1440
1441 static noinline int btrfs_ioctl_resize(struct file *file,
1442                                         void __user *arg)
1443 {
1444         struct inode *inode = file_inode(file);
1445         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1446         u64 new_size;
1447         u64 old_size;
1448         u64 devid = 1;
1449         struct btrfs_root *root = BTRFS_I(inode)->root;
1450         struct btrfs_ioctl_vol_args *vol_args;
1451         struct btrfs_trans_handle *trans;
1452         struct btrfs_device *device = NULL;
1453         char *sizestr;
1454         char *retptr;
1455         char *devstr = NULL;
1456         int ret = 0;
1457         int mod = 0;
1458
1459         if (!capable(CAP_SYS_ADMIN))
1460                 return -EPERM;
1461
1462         ret = mnt_want_write_file(file);
1463         if (ret)
1464                 return ret;
1465
1466         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
1467                 mnt_drop_write_file(file);
1468                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1469         }
1470
1471         mutex_lock(&fs_info->volume_mutex);
1472         vol_args = memdup_user(arg, sizeof(*vol_args));
1473         if (IS_ERR(vol_args)) {
1474                 ret = PTR_ERR(vol_args);
1475                 goto out;
1476         }
1477
1478         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1479
1480         sizestr = vol_args->name;
1481         devstr = strchr(sizestr, ':');
1482         if (devstr) {
1483                 sizestr = devstr + 1;
1484                 *devstr = '\0';
1485                 devstr = vol_args->name;
1486                 ret = kstrtoull(devstr, 10, &devid);
1487                 if (ret)
1488                         goto out_free;
1489                 if (!devid) {
1490                         ret = -EINVAL;
1491                         goto out_free;
1492                 }
1493                 btrfs_info(fs_info, "resizing devid %llu", devid);
1494         }
1495
1496         device = btrfs_find_device(fs_info, devid, NULL, NULL);
1497         if (!device) {
1498                 btrfs_info(fs_info, "resizer unable to find device %llu",
1499                            devid);
1500                 ret = -ENODEV;
1501                 goto out_free;
1502         }
1503
1504         if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1505                 btrfs_info(fs_info,
1506                            "resizer unable to apply on readonly device %llu",
1507                        devid);
1508                 ret = -EPERM;
1509                 goto out_free;
1510         }
1511
1512         if (!strcmp(sizestr, "max"))
1513                 new_size = device->bdev->bd_inode->i_size;
1514         else {
1515                 if (sizestr[0] == '-') {
1516                         mod = -1;
1517                         sizestr++;
1518                 } else if (sizestr[0] == '+') {
1519                         mod = 1;
1520                         sizestr++;
1521                 }
1522                 new_size = memparse(sizestr, &retptr);
1523                 if (*retptr != '\0' || new_size == 0) {
1524                         ret = -EINVAL;
1525                         goto out_free;
1526                 }
1527         }
1528
1529         if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1530                 ret = -EPERM;
1531                 goto out_free;
1532         }
1533
1534         old_size = btrfs_device_get_total_bytes(device);
1535
1536         if (mod < 0) {
1537                 if (new_size > old_size) {
1538                         ret = -EINVAL;
1539                         goto out_free;
1540                 }
1541                 new_size = old_size - new_size;
1542         } else if (mod > 0) {
1543                 if (new_size > ULLONG_MAX - old_size) {
1544                         ret = -ERANGE;
1545                         goto out_free;
1546                 }
1547                 new_size = old_size + new_size;
1548         }
1549
1550         if (new_size < SZ_256M) {
1551                 ret = -EINVAL;
1552                 goto out_free;
1553         }
1554         if (new_size > device->bdev->bd_inode->i_size) {
1555                 ret = -EFBIG;
1556                 goto out_free;
1557         }
1558
1559         new_size = round_down(new_size, fs_info->sectorsize);
1560
1561         btrfs_info_in_rcu(fs_info, "new size for %s is %llu",
1562                           rcu_str_deref(device->name), new_size);
1563
1564         if (new_size > old_size) {
1565                 trans = btrfs_start_transaction(root, 0);
1566                 if (IS_ERR(trans)) {
1567                         ret = PTR_ERR(trans);
1568                         goto out_free;
1569                 }
1570                 ret = btrfs_grow_device(trans, device, new_size);
1571                 btrfs_commit_transaction(trans);
1572         } else if (new_size < old_size) {
1573                 ret = btrfs_shrink_device(device, new_size);
1574         } /* equal, nothing need to do */
1575
1576 out_free:
1577         kfree(vol_args);
1578 out:
1579         mutex_unlock(&fs_info->volume_mutex);
1580         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
1581         mnt_drop_write_file(file);
1582         return ret;
1583 }
1584
1585 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1586                                 const char *name, unsigned long fd, int subvol,
1587                                 u64 *transid, bool readonly,
1588                                 struct btrfs_qgroup_inherit *inherit)
1589 {
1590         int namelen;
1591         int ret = 0;
1592
1593         if (!S_ISDIR(file_inode(file)->i_mode))
1594                 return -ENOTDIR;
1595
1596         ret = mnt_want_write_file(file);
1597         if (ret)
1598                 goto out;
1599
1600         namelen = strlen(name);
1601         if (strchr(name, '/')) {
1602                 ret = -EINVAL;
1603                 goto out_drop_write;
1604         }
1605
1606         if (name[0] == '.' &&
1607            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1608                 ret = -EEXIST;
1609                 goto out_drop_write;
1610         }
1611
1612         if (subvol) {
1613                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1614                                      NULL, transid, readonly, inherit);
1615         } else {
1616                 struct fd src = fdget(fd);
1617                 struct inode *src_inode;
1618                 if (!src.file) {
1619                         ret = -EINVAL;
1620                         goto out_drop_write;
1621                 }
1622
1623                 src_inode = file_inode(src.file);
1624                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1625                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1626                                    "Snapshot src from another FS");
1627                         ret = -EXDEV;
1628                 } else if (!inode_owner_or_capable(src_inode)) {
1629                         /*
1630                          * Subvolume creation is not restricted, but snapshots
1631                          * are limited to own subvolumes only
1632                          */
1633                         ret = -EPERM;
1634                 } else {
1635                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1636                                              BTRFS_I(src_inode)->root,
1637                                              transid, readonly, inherit);
1638                 }
1639                 fdput(src);
1640         }
1641 out_drop_write:
1642         mnt_drop_write_file(file);
1643 out:
1644         return ret;
1645 }
1646
1647 static noinline int btrfs_ioctl_snap_create(struct file *file,
1648                                             void __user *arg, int subvol)
1649 {
1650         struct btrfs_ioctl_vol_args *vol_args;
1651         int ret;
1652
1653         if (!S_ISDIR(file_inode(file)->i_mode))
1654                 return -ENOTDIR;
1655
1656         vol_args = memdup_user(arg, sizeof(*vol_args));
1657         if (IS_ERR(vol_args))
1658                 return PTR_ERR(vol_args);
1659         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1660
1661         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1662                                               vol_args->fd, subvol,
1663                                               NULL, false, NULL);
1664
1665         kfree(vol_args);
1666         return ret;
1667 }
1668
1669 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1670                                                void __user *arg, int subvol)
1671 {
1672         struct btrfs_ioctl_vol_args_v2 *vol_args;
1673         int ret;
1674         u64 transid = 0;
1675         u64 *ptr = NULL;
1676         bool readonly = false;
1677         struct btrfs_qgroup_inherit *inherit = NULL;
1678
1679         if (!S_ISDIR(file_inode(file)->i_mode))
1680                 return -ENOTDIR;
1681
1682         vol_args = memdup_user(arg, sizeof(*vol_args));
1683         if (IS_ERR(vol_args))
1684                 return PTR_ERR(vol_args);
1685         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1686
1687         if (vol_args->flags &
1688             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1689               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1690                 ret = -EOPNOTSUPP;
1691                 goto free_args;
1692         }
1693
1694         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1695                 ptr = &transid;
1696         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1697                 readonly = true;
1698         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1699                 if (vol_args->size > PAGE_SIZE) {
1700                         ret = -EINVAL;
1701                         goto free_args;
1702                 }
1703                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1704                 if (IS_ERR(inherit)) {
1705                         ret = PTR_ERR(inherit);
1706                         goto free_args;
1707                 }
1708         }
1709
1710         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1711                                               vol_args->fd, subvol, ptr,
1712                                               readonly, inherit);
1713         if (ret)
1714                 goto free_inherit;
1715
1716         if (ptr && copy_to_user(arg +
1717                                 offsetof(struct btrfs_ioctl_vol_args_v2,
1718                                         transid),
1719                                 ptr, sizeof(*ptr)))
1720                 ret = -EFAULT;
1721
1722 free_inherit:
1723         kfree(inherit);
1724 free_args:
1725         kfree(vol_args);
1726         return ret;
1727 }
1728
1729 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1730                                                 void __user *arg)
1731 {
1732         struct inode *inode = file_inode(file);
1733         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1734         struct btrfs_root *root = BTRFS_I(inode)->root;
1735         int ret = 0;
1736         u64 flags = 0;
1737
1738         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
1739                 return -EINVAL;
1740
1741         down_read(&fs_info->subvol_sem);
1742         if (btrfs_root_readonly(root))
1743                 flags |= BTRFS_SUBVOL_RDONLY;
1744         up_read(&fs_info->subvol_sem);
1745
1746         if (copy_to_user(arg, &flags, sizeof(flags)))
1747                 ret = -EFAULT;
1748
1749         return ret;
1750 }
1751
1752 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1753                                               void __user *arg)
1754 {
1755         struct inode *inode = file_inode(file);
1756         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1757         struct btrfs_root *root = BTRFS_I(inode)->root;
1758         struct btrfs_trans_handle *trans;
1759         u64 root_flags;
1760         u64 flags;
1761         int ret = 0;
1762
1763         if (!inode_owner_or_capable(inode))
1764                 return -EPERM;
1765
1766         ret = mnt_want_write_file(file);
1767         if (ret)
1768                 goto out;
1769
1770         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
1771                 ret = -EINVAL;
1772                 goto out_drop_write;
1773         }
1774
1775         if (copy_from_user(&flags, arg, sizeof(flags))) {
1776                 ret = -EFAULT;
1777                 goto out_drop_write;
1778         }
1779
1780         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1781                 ret = -EINVAL;
1782                 goto out_drop_write;
1783         }
1784
1785         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1786                 ret = -EOPNOTSUPP;
1787                 goto out_drop_write;
1788         }
1789
1790         down_write(&fs_info->subvol_sem);
1791
1792         /* nothing to do */
1793         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1794                 goto out_drop_sem;
1795
1796         root_flags = btrfs_root_flags(&root->root_item);
1797         if (flags & BTRFS_SUBVOL_RDONLY) {
1798                 btrfs_set_root_flags(&root->root_item,
1799                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1800         } else {
1801                 /*
1802                  * Block RO -> RW transition if this subvolume is involved in
1803                  * send
1804                  */
1805                 spin_lock(&root->root_item_lock);
1806                 if (root->send_in_progress == 0) {
1807                         btrfs_set_root_flags(&root->root_item,
1808                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1809                         spin_unlock(&root->root_item_lock);
1810                 } else {
1811                         spin_unlock(&root->root_item_lock);
1812                         btrfs_warn(fs_info,
1813                                    "Attempt to set subvolume %llu read-write during send",
1814                                    root->root_key.objectid);
1815                         ret = -EPERM;
1816                         goto out_drop_sem;
1817                 }
1818         }
1819
1820         trans = btrfs_start_transaction(root, 1);
1821         if (IS_ERR(trans)) {
1822                 ret = PTR_ERR(trans);
1823                 goto out_reset;
1824         }
1825
1826         ret = btrfs_update_root(trans, fs_info->tree_root,
1827                                 &root->root_key, &root->root_item);
1828         if (ret < 0) {
1829                 btrfs_end_transaction(trans);
1830                 goto out_reset;
1831         }
1832
1833         ret = btrfs_commit_transaction(trans);
1834
1835 out_reset:
1836         if (ret)
1837                 btrfs_set_root_flags(&root->root_item, root_flags);
1838 out_drop_sem:
1839         up_write(&fs_info->subvol_sem);
1840 out_drop_write:
1841         mnt_drop_write_file(file);
1842 out:
1843         return ret;
1844 }
1845
1846 /*
1847  * helper to check if the subvolume references other subvolumes
1848  */
1849 static noinline int may_destroy_subvol(struct btrfs_root *root)
1850 {
1851         struct btrfs_fs_info *fs_info = root->fs_info;
1852         struct btrfs_path *path;
1853         struct btrfs_dir_item *di;
1854         struct btrfs_key key;
1855         u64 dir_id;
1856         int ret;
1857
1858         path = btrfs_alloc_path();
1859         if (!path)
1860                 return -ENOMEM;
1861
1862         /* Make sure this root isn't set as the default subvol */
1863         dir_id = btrfs_super_root_dir(fs_info->super_copy);
1864         di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
1865                                    dir_id, "default", 7, 0);
1866         if (di && !IS_ERR(di)) {
1867                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1868                 if (key.objectid == root->root_key.objectid) {
1869                         ret = -EPERM;
1870                         btrfs_err(fs_info,
1871                                   "deleting default subvolume %llu is not allowed",
1872                                   key.objectid);
1873                         goto out;
1874                 }
1875                 btrfs_release_path(path);
1876         }
1877
1878         key.objectid = root->root_key.objectid;
1879         key.type = BTRFS_ROOT_REF_KEY;
1880         key.offset = (u64)-1;
1881
1882         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1883         if (ret < 0)
1884                 goto out;
1885         BUG_ON(ret == 0);
1886
1887         ret = 0;
1888         if (path->slots[0] > 0) {
1889                 path->slots[0]--;
1890                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1891                 if (key.objectid == root->root_key.objectid &&
1892                     key.type == BTRFS_ROOT_REF_KEY)
1893                         ret = -ENOTEMPTY;
1894         }
1895 out:
1896         btrfs_free_path(path);
1897         return ret;
1898 }
1899
1900 static noinline int key_in_sk(struct btrfs_key *key,
1901                               struct btrfs_ioctl_search_key *sk)
1902 {
1903         struct btrfs_key test;
1904         int ret;
1905
1906         test.objectid = sk->min_objectid;
1907         test.type = sk->min_type;
1908         test.offset = sk->min_offset;
1909
1910         ret = btrfs_comp_cpu_keys(key, &test);
1911         if (ret < 0)
1912                 return 0;
1913
1914         test.objectid = sk->max_objectid;
1915         test.type = sk->max_type;
1916         test.offset = sk->max_offset;
1917
1918         ret = btrfs_comp_cpu_keys(key, &test);
1919         if (ret > 0)
1920                 return 0;
1921         return 1;
1922 }
1923
1924 static noinline int copy_to_sk(struct btrfs_path *path,
1925                                struct btrfs_key *key,
1926                                struct btrfs_ioctl_search_key *sk,
1927                                size_t *buf_size,
1928                                char __user *ubuf,
1929                                unsigned long *sk_offset,
1930                                int *num_found)
1931 {
1932         u64 found_transid;
1933         struct extent_buffer *leaf;
1934         struct btrfs_ioctl_search_header sh;
1935         struct btrfs_key test;
1936         unsigned long item_off;
1937         unsigned long item_len;
1938         int nritems;
1939         int i;
1940         int slot;
1941         int ret = 0;
1942
1943         leaf = path->nodes[0];
1944         slot = path->slots[0];
1945         nritems = btrfs_header_nritems(leaf);
1946
1947         if (btrfs_header_generation(leaf) > sk->max_transid) {
1948                 i = nritems;
1949                 goto advance_key;
1950         }
1951         found_transid = btrfs_header_generation(leaf);
1952
1953         for (i = slot; i < nritems; i++) {
1954                 item_off = btrfs_item_ptr_offset(leaf, i);
1955                 item_len = btrfs_item_size_nr(leaf, i);
1956
1957                 btrfs_item_key_to_cpu(leaf, key, i);
1958                 if (!key_in_sk(key, sk))
1959                         continue;
1960
1961                 if (sizeof(sh) + item_len > *buf_size) {
1962                         if (*num_found) {
1963                                 ret = 1;
1964                                 goto out;
1965                         }
1966
1967                         /*
1968                          * return one empty item back for v1, which does not
1969                          * handle -EOVERFLOW
1970                          */
1971
1972                         *buf_size = sizeof(sh) + item_len;
1973                         item_len = 0;
1974                         ret = -EOVERFLOW;
1975                 }
1976
1977                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
1978                         ret = 1;
1979                         goto out;
1980                 }
1981
1982                 sh.objectid = key->objectid;
1983                 sh.offset = key->offset;
1984                 sh.type = key->type;
1985                 sh.len = item_len;
1986                 sh.transid = found_transid;
1987
1988                 /* copy search result header */
1989                 if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
1990                         ret = -EFAULT;
1991                         goto out;
1992                 }
1993
1994                 *sk_offset += sizeof(sh);
1995
1996                 if (item_len) {
1997                         char __user *up = ubuf + *sk_offset;
1998                         /* copy the item */
1999                         if (read_extent_buffer_to_user(leaf, up,
2000                                                        item_off, item_len)) {
2001                                 ret = -EFAULT;
2002                                 goto out;
2003                         }
2004
2005                         *sk_offset += item_len;
2006                 }
2007                 (*num_found)++;
2008
2009                 if (ret) /* -EOVERFLOW from above */
2010                         goto out;
2011
2012                 if (*num_found >= sk->nr_items) {
2013                         ret = 1;
2014                         goto out;
2015                 }
2016         }
2017 advance_key:
2018         ret = 0;
2019         test.objectid = sk->max_objectid;
2020         test.type = sk->max_type;
2021         test.offset = sk->max_offset;
2022         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2023                 ret = 1;
2024         else if (key->offset < (u64)-1)
2025                 key->offset++;
2026         else if (key->type < (u8)-1) {
2027                 key->offset = 0;
2028                 key->type++;
2029         } else if (key->objectid < (u64)-1) {
2030                 key->offset = 0;
2031                 key->type = 0;
2032                 key->objectid++;
2033         } else
2034                 ret = 1;
2035 out:
2036         /*
2037          *  0: all items from this leaf copied, continue with next
2038          *  1: * more items can be copied, but unused buffer is too small
2039          *     * all items were found
2040          *     Either way, it will stops the loop which iterates to the next
2041          *     leaf
2042          *  -EOVERFLOW: item was to large for buffer
2043          *  -EFAULT: could not copy extent buffer back to userspace
2044          */
2045         return ret;
2046 }
2047
2048 static noinline int search_ioctl(struct inode *inode,
2049                                  struct btrfs_ioctl_search_key *sk,
2050                                  size_t *buf_size,
2051                                  char __user *ubuf)
2052 {
2053         struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2054         struct btrfs_root *root;
2055         struct btrfs_key key;
2056         struct btrfs_path *path;
2057         int ret;
2058         int num_found = 0;
2059         unsigned long sk_offset = 0;
2060
2061         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2062                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2063                 return -EOVERFLOW;
2064         }
2065
2066         path = btrfs_alloc_path();
2067         if (!path)
2068                 return -ENOMEM;
2069
2070         if (sk->tree_id == 0) {
2071                 /* search the root of the inode that was passed */
2072                 root = BTRFS_I(inode)->root;
2073         } else {
2074                 key.objectid = sk->tree_id;
2075                 key.type = BTRFS_ROOT_ITEM_KEY;
2076                 key.offset = (u64)-1;
2077                 root = btrfs_read_fs_root_no_name(info, &key);
2078                 if (IS_ERR(root)) {
2079                         btrfs_free_path(path);
2080                         return -ENOENT;
2081                 }
2082         }
2083
2084         key.objectid = sk->min_objectid;
2085         key.type = sk->min_type;
2086         key.offset = sk->min_offset;
2087
2088         while (1) {
2089                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2090                 if (ret != 0) {
2091                         if (ret > 0)
2092                                 ret = 0;
2093                         goto err;
2094                 }
2095                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2096                                  &sk_offset, &num_found);
2097                 btrfs_release_path(path);
2098                 if (ret)
2099                         break;
2100
2101         }
2102         if (ret > 0)
2103                 ret = 0;
2104 err:
2105         sk->nr_items = num_found;
2106         btrfs_free_path(path);
2107         return ret;
2108 }
2109
2110 static noinline int btrfs_ioctl_tree_search(struct file *file,
2111                                            void __user *argp)
2112 {
2113         struct btrfs_ioctl_search_args __user *uargs;
2114         struct btrfs_ioctl_search_key sk;
2115         struct inode *inode;
2116         int ret;
2117         size_t buf_size;
2118
2119         if (!capable(CAP_SYS_ADMIN))
2120                 return -EPERM;
2121
2122         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2123
2124         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2125                 return -EFAULT;
2126
2127         buf_size = sizeof(uargs->buf);
2128
2129         inode = file_inode(file);
2130         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2131
2132         /*
2133          * In the origin implementation an overflow is handled by returning a
2134          * search header with a len of zero, so reset ret.
2135          */
2136         if (ret == -EOVERFLOW)
2137                 ret = 0;
2138
2139         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2140                 ret = -EFAULT;
2141         return ret;
2142 }
2143
2144 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2145                                                void __user *argp)
2146 {
2147         struct btrfs_ioctl_search_args_v2 __user *uarg;
2148         struct btrfs_ioctl_search_args_v2 args;
2149         struct inode *inode;
2150         int ret;
2151         size_t buf_size;
2152         const size_t buf_limit = SZ_16M;
2153
2154         if (!capable(CAP_SYS_ADMIN))
2155                 return -EPERM;
2156
2157         /* copy search header and buffer size */
2158         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2159         if (copy_from_user(&args, uarg, sizeof(args)))
2160                 return -EFAULT;
2161
2162         buf_size = args.buf_size;
2163
2164         /* limit result size to 16MB */
2165         if (buf_size > buf_limit)
2166                 buf_size = buf_limit;
2167
2168         inode = file_inode(file);
2169         ret = search_ioctl(inode, &args.key, &buf_size,
2170                            (char __user *)(&uarg->buf[0]));
2171         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2172                 ret = -EFAULT;
2173         else if (ret == -EOVERFLOW &&
2174                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2175                 ret = -EFAULT;
2176
2177         return ret;
2178 }
2179
2180 /*
2181  * Search INODE_REFs to identify path name of 'dirid' directory
2182  * in a 'tree_id' tree. and sets path name to 'name'.
2183  */
2184 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2185                                 u64 tree_id, u64 dirid, char *name)
2186 {
2187         struct btrfs_root *root;
2188         struct btrfs_key key;
2189         char *ptr;
2190         int ret = -1;
2191         int slot;
2192         int len;
2193         int total_len = 0;
2194         struct btrfs_inode_ref *iref;
2195         struct extent_buffer *l;
2196         struct btrfs_path *path;
2197
2198         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2199                 name[0]='\0';
2200                 return 0;
2201         }
2202
2203         path = btrfs_alloc_path();
2204         if (!path)
2205                 return -ENOMEM;
2206
2207         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1];
2208
2209         key.objectid = tree_id;
2210         key.type = BTRFS_ROOT_ITEM_KEY;
2211         key.offset = (u64)-1;
2212         root = btrfs_read_fs_root_no_name(info, &key);
2213         if (IS_ERR(root)) {
2214                 btrfs_err(info, "could not find root %llu", tree_id);
2215                 ret = -ENOENT;
2216                 goto out;
2217         }
2218
2219         key.objectid = dirid;
2220         key.type = BTRFS_INODE_REF_KEY;
2221         key.offset = (u64)-1;
2222
2223         while (1) {
2224                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2225                 if (ret < 0)
2226                         goto out;
2227                 else if (ret > 0) {
2228                         ret = btrfs_previous_item(root, path, dirid,
2229                                                   BTRFS_INODE_REF_KEY);
2230                         if (ret < 0)
2231                                 goto out;
2232                         else if (ret > 0) {
2233                                 ret = -ENOENT;
2234                                 goto out;
2235                         }
2236                 }
2237
2238                 l = path->nodes[0];
2239                 slot = path->slots[0];
2240                 btrfs_item_key_to_cpu(l, &key, slot);
2241
2242                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2243                 len = btrfs_inode_ref_name_len(l, iref);
2244                 ptr -= len + 1;
2245                 total_len += len + 1;
2246                 if (ptr < name) {
2247                         ret = -ENAMETOOLONG;
2248                         goto out;
2249                 }
2250
2251                 *(ptr + len) = '/';
2252                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2253
2254                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2255                         break;
2256
2257                 btrfs_release_path(path);
2258                 key.objectid = key.offset;
2259                 key.offset = (u64)-1;
2260                 dirid = key.objectid;
2261         }
2262         memmove(name, ptr, total_len);
2263         name[total_len] = '\0';
2264         ret = 0;
2265 out:
2266         btrfs_free_path(path);
2267         return ret;
2268 }
2269
2270 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2271                                            void __user *argp)
2272 {
2273          struct btrfs_ioctl_ino_lookup_args *args;
2274          struct inode *inode;
2275         int ret = 0;
2276
2277         args = memdup_user(argp, sizeof(*args));
2278         if (IS_ERR(args))
2279                 return PTR_ERR(args);
2280
2281         inode = file_inode(file);
2282
2283         /*
2284          * Unprivileged query to obtain the containing subvolume root id. The
2285          * path is reset so it's consistent with btrfs_search_path_in_tree.
2286          */
2287         if (args->treeid == 0)
2288                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2289
2290         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2291                 args->name[0] = 0;
2292                 goto out;
2293         }
2294
2295         if (!capable(CAP_SYS_ADMIN)) {
2296                 ret = -EPERM;
2297                 goto out;
2298         }
2299
2300         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2301                                         args->treeid, args->objectid,
2302                                         args->name);
2303
2304 out:
2305         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2306                 ret = -EFAULT;
2307
2308         kfree(args);
2309         return ret;
2310 }
2311
2312 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2313                                              void __user *arg)
2314 {
2315         struct dentry *parent = file->f_path.dentry;
2316         struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
2317         struct dentry *dentry;
2318         struct inode *dir = d_inode(parent);
2319         struct inode *inode;
2320         struct btrfs_root *root = BTRFS_I(dir)->root;
2321         struct btrfs_root *dest = NULL;
2322         struct btrfs_ioctl_vol_args *vol_args;
2323         struct btrfs_trans_handle *trans;
2324         struct btrfs_block_rsv block_rsv;
2325         u64 root_flags;
2326         u64 qgroup_reserved;
2327         int namelen;
2328         int ret;
2329         int err = 0;
2330
2331         if (!S_ISDIR(dir->i_mode))
2332                 return -ENOTDIR;
2333
2334         vol_args = memdup_user(arg, sizeof(*vol_args));
2335         if (IS_ERR(vol_args))
2336                 return PTR_ERR(vol_args);
2337
2338         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2339         namelen = strlen(vol_args->name);
2340         if (strchr(vol_args->name, '/') ||
2341             strncmp(vol_args->name, "..", namelen) == 0) {
2342                 err = -EINVAL;
2343                 goto out;
2344         }
2345
2346         err = mnt_want_write_file(file);
2347         if (err)
2348                 goto out;
2349
2350
2351         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
2352         if (err == -EINTR)
2353                 goto out_drop_write;
2354         dentry = lookup_one_len(vol_args->name, parent, namelen);
2355         if (IS_ERR(dentry)) {
2356                 err = PTR_ERR(dentry);
2357                 goto out_unlock_dir;
2358         }
2359
2360         if (d_really_is_negative(dentry)) {
2361                 err = -ENOENT;
2362                 goto out_dput;
2363         }
2364
2365         inode = d_inode(dentry);
2366         dest = BTRFS_I(inode)->root;
2367         if (!capable(CAP_SYS_ADMIN)) {
2368                 /*
2369                  * Regular user.  Only allow this with a special mount
2370                  * option, when the user has write+exec access to the
2371                  * subvol root, and when rmdir(2) would have been
2372                  * allowed.
2373                  *
2374                  * Note that this is _not_ check that the subvol is
2375                  * empty or doesn't contain data that we wouldn't
2376                  * otherwise be able to delete.
2377                  *
2378                  * Users who want to delete empty subvols should try
2379                  * rmdir(2).
2380                  */
2381                 err = -EPERM;
2382                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
2383                         goto out_dput;
2384
2385                 /*
2386                  * Do not allow deletion if the parent dir is the same
2387                  * as the dir to be deleted.  That means the ioctl
2388                  * must be called on the dentry referencing the root
2389                  * of the subvol, not a random directory contained
2390                  * within it.
2391                  */
2392                 err = -EINVAL;
2393                 if (root == dest)
2394                         goto out_dput;
2395
2396                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2397                 if (err)
2398                         goto out_dput;
2399         }
2400
2401         /* check if subvolume may be deleted by a user */
2402         err = btrfs_may_delete(dir, dentry, 1);
2403         if (err)
2404                 goto out_dput;
2405
2406         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
2407                 err = -EINVAL;
2408                 goto out_dput;
2409         }
2410
2411         inode_lock(inode);
2412
2413         /*
2414          * Don't allow to delete a subvolume with send in progress. This is
2415          * inside the i_mutex so the error handling that has to drop the bit
2416          * again is not run concurrently.
2417          */
2418         spin_lock(&dest->root_item_lock);
2419         root_flags = btrfs_root_flags(&dest->root_item);
2420         if (dest->send_in_progress == 0) {
2421                 btrfs_set_root_flags(&dest->root_item,
2422                                 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
2423                 spin_unlock(&dest->root_item_lock);
2424         } else {
2425                 spin_unlock(&dest->root_item_lock);
2426                 btrfs_warn(fs_info,
2427                            "Attempt to delete subvolume %llu during send",
2428                            dest->root_key.objectid);
2429                 err = -EPERM;
2430                 goto out_unlock_inode;
2431         }
2432
2433         down_write(&fs_info->subvol_sem);
2434
2435         err = may_destroy_subvol(dest);
2436         if (err)
2437                 goto out_up_write;
2438
2439         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2440         /*
2441          * One for dir inode, two for dir entries, two for root
2442          * ref/backref.
2443          */
2444         err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2445                                                5, &qgroup_reserved, true);
2446         if (err)
2447                 goto out_up_write;
2448
2449         trans = btrfs_start_transaction(root, 0);
2450         if (IS_ERR(trans)) {
2451                 err = PTR_ERR(trans);
2452                 goto out_release;
2453         }
2454         trans->block_rsv = &block_rsv;
2455         trans->bytes_reserved = block_rsv.size;
2456
2457         btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
2458
2459         ret = btrfs_unlink_subvol(trans, root, dir,
2460                                 dest->root_key.objectid,
2461                                 dentry->d_name.name,
2462                                 dentry->d_name.len);
2463         if (ret) {
2464                 err = ret;
2465                 btrfs_abort_transaction(trans, ret);
2466                 goto out_end_trans;
2467         }
2468
2469         btrfs_record_root_in_trans(trans, dest);
2470
2471         memset(&dest->root_item.drop_progress, 0,
2472                 sizeof(dest->root_item.drop_progress));
2473         dest->root_item.drop_level = 0;
2474         btrfs_set_root_refs(&dest->root_item, 0);
2475
2476         if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
2477                 ret = btrfs_insert_orphan_item(trans,
2478                                         fs_info->tree_root,
2479                                         dest->root_key.objectid);
2480                 if (ret) {
2481                         btrfs_abort_transaction(trans, ret);
2482                         err = ret;
2483                         goto out_end_trans;
2484                 }
2485         }
2486
2487         ret = btrfs_uuid_tree_rem(trans, fs_info, dest->root_item.uuid,
2488                                   BTRFS_UUID_KEY_SUBVOL,
2489                                   dest->root_key.objectid);
2490         if (ret && ret != -ENOENT) {
2491                 btrfs_abort_transaction(trans, ret);
2492                 err = ret;
2493                 goto out_end_trans;
2494         }
2495         if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2496                 ret = btrfs_uuid_tree_rem(trans, fs_info,
2497                                           dest->root_item.received_uuid,
2498                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2499                                           dest->root_key.objectid);
2500                 if (ret && ret != -ENOENT) {
2501                         btrfs_abort_transaction(trans, ret);
2502                         err = ret;
2503                         goto out_end_trans;
2504                 }
2505         }
2506
2507 out_end_trans:
2508         trans->block_rsv = NULL;
2509         trans->bytes_reserved = 0;
2510         ret = btrfs_end_transaction(trans);
2511         if (ret && !err)
2512                 err = ret;
2513         inode->i_flags |= S_DEAD;
2514 out_release:
2515         btrfs_subvolume_release_metadata(fs_info, &block_rsv);
2516 out_up_write:
2517         up_write(&fs_info->subvol_sem);
2518         if (err) {
2519                 spin_lock(&dest->root_item_lock);
2520                 root_flags = btrfs_root_flags(&dest->root_item);
2521                 btrfs_set_root_flags(&dest->root_item,
2522                                 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
2523                 spin_unlock(&dest->root_item_lock);
2524         }
2525 out_unlock_inode:
2526         inode_unlock(inode);
2527         if (!err) {
2528                 d_invalidate(dentry);
2529                 btrfs_invalidate_inodes(dest);
2530                 d_delete(dentry);
2531                 ASSERT(dest->send_in_progress == 0);
2532
2533                 /* the last ref */
2534                 if (dest->ino_cache_inode) {
2535                         iput(dest->ino_cache_inode);
2536                         dest->ino_cache_inode = NULL;
2537                 }
2538         }
2539 out_dput:
2540         dput(dentry);
2541 out_unlock_dir:
2542         inode_unlock(dir);
2543 out_drop_write:
2544         mnt_drop_write_file(file);
2545 out:
2546         kfree(vol_args);
2547         return err;
2548 }
2549
2550 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2551 {
2552         struct inode *inode = file_inode(file);
2553         struct btrfs_root *root = BTRFS_I(inode)->root;
2554         struct btrfs_ioctl_defrag_range_args *range;
2555         int ret;
2556
2557         ret = mnt_want_write_file(file);
2558         if (ret)
2559                 return ret;
2560
2561         if (btrfs_root_readonly(root)) {
2562                 ret = -EROFS;
2563                 goto out;
2564         }
2565
2566         switch (inode->i_mode & S_IFMT) {
2567         case S_IFDIR:
2568                 if (!capable(CAP_SYS_ADMIN)) {
2569                         ret = -EPERM;
2570                         goto out;
2571                 }
2572                 ret = btrfs_defrag_root(root);
2573                 break;
2574         case S_IFREG:
2575                 if (!(file->f_mode & FMODE_WRITE)) {
2576                         ret = -EINVAL;
2577                         goto out;
2578                 }
2579
2580                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2581                 if (!range) {
2582                         ret = -ENOMEM;
2583                         goto out;
2584                 }
2585
2586                 if (argp) {
2587                         if (copy_from_user(range, argp,
2588                                            sizeof(*range))) {
2589                                 ret = -EFAULT;
2590                                 kfree(range);
2591                                 goto out;
2592                         }
2593                         /* compression requires us to start the IO */
2594                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2595                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2596                                 range->extent_thresh = (u32)-1;
2597                         }
2598                 } else {
2599                         /* the rest are all set to zero by kzalloc */
2600                         range->len = (u64)-1;
2601                 }
2602                 ret = btrfs_defrag_file(file_inode(file), file,
2603                                         range, BTRFS_OLDEST_GENERATION, 0);
2604                 if (ret > 0)
2605                         ret = 0;
2606                 kfree(range);
2607                 break;
2608         default:
2609                 ret = -EINVAL;
2610         }
2611 out:
2612         mnt_drop_write_file(file);
2613         return ret;
2614 }
2615
2616 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
2617 {
2618         struct btrfs_ioctl_vol_args *vol_args;
2619         int ret;
2620
2621         if (!capable(CAP_SYS_ADMIN))
2622                 return -EPERM;
2623
2624         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
2625                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2626
2627         mutex_lock(&fs_info->volume_mutex);
2628         vol_args = memdup_user(arg, sizeof(*vol_args));
2629         if (IS_ERR(vol_args)) {
2630                 ret = PTR_ERR(vol_args);
2631                 goto out;
2632         }
2633
2634         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2635         ret = btrfs_init_new_device(fs_info, vol_args->name);
2636
2637         if (!ret)
2638                 btrfs_info(fs_info, "disk added %s", vol_args->name);
2639
2640         kfree(vol_args);
2641 out:
2642         mutex_unlock(&fs_info->volume_mutex);
2643         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
2644         return ret;
2645 }
2646
2647 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
2648 {
2649         struct inode *inode = file_inode(file);
2650         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2651         struct btrfs_ioctl_vol_args_v2 *vol_args;
2652         int ret;
2653
2654         if (!capable(CAP_SYS_ADMIN))
2655                 return -EPERM;
2656
2657         ret = mnt_want_write_file(file);
2658         if (ret)
2659                 return ret;
2660
2661         vol_args = memdup_user(arg, sizeof(*vol_args));
2662         if (IS_ERR(vol_args)) {
2663                 ret = PTR_ERR(vol_args);
2664                 goto err_drop;
2665         }
2666
2667         /* Check for compatibility reject unknown flags */
2668         if (vol_args->flags & ~BTRFS_VOL_ARG_V2_FLAGS_SUPPORTED)
2669                 return -EOPNOTSUPP;
2670
2671         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
2672                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2673                 goto out;
2674         }
2675
2676         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
2677                 ret = btrfs_rm_device(fs_info, NULL, vol_args->devid);
2678         } else {
2679                 vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
2680                 ret = btrfs_rm_device(fs_info, vol_args->name, 0);
2681         }
2682         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
2683
2684         if (!ret) {
2685                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
2686                         btrfs_info(fs_info, "device deleted: id %llu",
2687                                         vol_args->devid);
2688                 else
2689                         btrfs_info(fs_info, "device deleted: %s",
2690                                         vol_args->name);
2691         }
2692 out:
2693         kfree(vol_args);
2694 err_drop:
2695         mnt_drop_write_file(file);
2696         return ret;
2697 }
2698
2699 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2700 {
2701         struct inode *inode = file_inode(file);
2702         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2703         struct btrfs_ioctl_vol_args *vol_args;
2704         int ret;
2705
2706         if (!capable(CAP_SYS_ADMIN))
2707                 return -EPERM;
2708
2709         ret = mnt_want_write_file(file);
2710         if (ret)
2711                 return ret;
2712
2713         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
2714                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2715                 goto out_drop_write;
2716         }
2717
2718         vol_args = memdup_user(arg, sizeof(*vol_args));
2719         if (IS_ERR(vol_args)) {
2720                 ret = PTR_ERR(vol_args);
2721                 goto out;
2722         }
2723
2724         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2725         ret = btrfs_rm_device(fs_info, vol_args->name, 0);
2726
2727         if (!ret)
2728                 btrfs_info(fs_info, "disk deleted %s", vol_args->name);
2729         kfree(vol_args);
2730 out:
2731         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
2732 out_drop_write:
2733         mnt_drop_write_file(file);
2734
2735         return ret;
2736 }
2737
2738 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
2739                                 void __user *arg)
2740 {
2741         struct btrfs_ioctl_fs_info_args *fi_args;
2742         struct btrfs_device *device;
2743         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2744         int ret = 0;
2745
2746         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2747         if (!fi_args)
2748                 return -ENOMEM;
2749
2750         rcu_read_lock();
2751         fi_args->num_devices = fs_devices->num_devices;
2752
2753         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2754                 if (device->devid > fi_args->max_id)
2755                         fi_args->max_id = device->devid;
2756         }
2757         rcu_read_unlock();
2758
2759         memcpy(&fi_args->fsid, fs_info->fsid, sizeof(fi_args->fsid));
2760         fi_args->nodesize = fs_info->nodesize;
2761         fi_args->sectorsize = fs_info->sectorsize;
2762         fi_args->clone_alignment = fs_info->sectorsize;
2763
2764         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2765                 ret = -EFAULT;
2766
2767         kfree(fi_args);
2768         return ret;
2769 }
2770
2771 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
2772                                  void __user *arg)
2773 {
2774         struct btrfs_ioctl_dev_info_args *di_args;
2775         struct btrfs_device *dev;
2776         int ret = 0;
2777         char *s_uuid = NULL;
2778
2779         di_args = memdup_user(arg, sizeof(*di_args));
2780         if (IS_ERR(di_args))
2781                 return PTR_ERR(di_args);
2782
2783         if (!btrfs_is_empty_uuid(di_args->uuid))
2784                 s_uuid = di_args->uuid;
2785
2786         rcu_read_lock();
2787         dev = btrfs_find_device(fs_info, di_args->devid, s_uuid, NULL);
2788
2789         if (!dev) {
2790                 ret = -ENODEV;
2791                 goto out;
2792         }
2793
2794         di_args->devid = dev->devid;
2795         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
2796         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
2797         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2798         if (dev->name) {
2799                 struct rcu_string *name;
2800
2801                 name = rcu_dereference(dev->name);
2802                 strncpy(di_args->path, name->str, sizeof(di_args->path) - 1);
2803                 di_args->path[sizeof(di_args->path) - 1] = 0;
2804         } else {
2805                 di_args->path[0] = '\0';
2806         }
2807
2808 out:
2809         rcu_read_unlock();
2810         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2811                 ret = -EFAULT;
2812
2813         kfree(di_args);
2814         return ret;
2815 }
2816
2817 static struct page *extent_same_get_page(struct inode *inode, pgoff_t index)
2818 {
2819         struct page *page;
2820
2821         page = grab_cache_page(inode->i_mapping, index);
2822         if (!page)
2823                 return ERR_PTR(-ENOMEM);
2824
2825         if (!PageUptodate(page)) {
2826                 int ret;
2827
2828                 ret = btrfs_readpage(NULL, page);
2829                 if (ret)
2830                         return ERR_PTR(ret);
2831                 lock_page(page);
2832                 if (!PageUptodate(page)) {
2833                         unlock_page(page);
2834                         put_page(page);
2835                         return ERR_PTR(-EIO);
2836                 }
2837                 if (page->mapping != inode->i_mapping) {
2838                         unlock_page(page);
2839                         put_page(page);
2840                         return ERR_PTR(-EAGAIN);
2841                 }
2842         }
2843
2844         return page;
2845 }
2846
2847 static int gather_extent_pages(struct inode *inode, struct page **pages,
2848                                int num_pages, u64 off)
2849 {
2850         int i;
2851         pgoff_t index = off >> PAGE_SHIFT;
2852
2853         for (i = 0; i < num_pages; i++) {
2854 again:
2855                 pages[i] = extent_same_get_page(inode, index + i);
2856                 if (IS_ERR(pages[i])) {
2857                         int err = PTR_ERR(pages[i]);
2858
2859                         if (err == -EAGAIN)
2860                                 goto again;
2861                         pages[i] = NULL;
2862                         return err;
2863                 }
2864         }
2865         return 0;
2866 }
2867
2868 static int lock_extent_range(struct inode *inode, u64 off, u64 len,
2869                              bool retry_range_locking)
2870 {
2871         /*
2872          * Do any pending delalloc/csum calculations on inode, one way or
2873          * another, and lock file content.
2874          * The locking order is:
2875          *
2876          *   1) pages
2877          *   2) range in the inode's io tree
2878          */
2879         while (1) {
2880                 struct btrfs_ordered_extent *ordered;
2881                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2882                 ordered = btrfs_lookup_first_ordered_extent(inode,
2883                                                             off + len - 1);
2884                 if ((!ordered ||
2885                      ordered->file_offset + ordered->len <= off ||
2886                      ordered->file_offset >= off + len) &&
2887                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2888                                     off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
2889                         if (ordered)
2890                                 btrfs_put_ordered_extent(ordered);
2891                         break;
2892                 }
2893                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2894                 if (ordered)
2895                         btrfs_put_ordered_extent(ordered);
2896                 if (!retry_range_locking)
2897                         return -EAGAIN;
2898                 btrfs_wait_ordered_range(inode, off, len);
2899         }
2900         return 0;
2901 }
2902
2903 static void btrfs_double_inode_unlock(struct inode *inode1, struct inode *inode2)
2904 {
2905         inode_unlock(inode1);
2906         inode_unlock(inode2);
2907 }
2908
2909 static void btrfs_double_inode_lock(struct inode *inode1, struct inode *inode2)
2910 {
2911         if (inode1 < inode2)
2912                 swap(inode1, inode2);
2913
2914         inode_lock_nested(inode1, I_MUTEX_PARENT);
2915         inode_lock_nested(inode2, I_MUTEX_CHILD);
2916 }
2917
2918 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
2919                                       struct inode *inode2, u64 loff2, u64 len)
2920 {
2921         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2922         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2923 }
2924
2925 static int btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
2926                                     struct inode *inode2, u64 loff2, u64 len,
2927                                     bool retry_range_locking)
2928 {
2929         int ret;
2930
2931         if (inode1 < inode2) {
2932                 swap(inode1, inode2);
2933                 swap(loff1, loff2);
2934         }
2935         ret = lock_extent_range(inode1, loff1, len, retry_range_locking);
2936         if (ret)
2937                 return ret;
2938         ret = lock_extent_range(inode2, loff2, len, retry_range_locking);
2939         if (ret)
2940                 unlock_extent(&BTRFS_I(inode1)->io_tree, loff1,
2941                               loff1 + len - 1);
2942         return ret;
2943 }
2944
2945 struct cmp_pages {
2946         int             num_pages;
2947         struct page     **src_pages;
2948         struct page     **dst_pages;
2949 };
2950
2951 static void btrfs_cmp_data_free(struct cmp_pages *cmp)
2952 {
2953         int i;
2954         struct page *pg;
2955
2956         for (i = 0; i < cmp->num_pages; i++) {
2957                 pg = cmp->src_pages[i];
2958                 if (pg) {
2959                         unlock_page(pg);
2960                         put_page(pg);
2961                 }
2962                 pg = cmp->dst_pages[i];
2963                 if (pg) {
2964                         unlock_page(pg);
2965                         put_page(pg);
2966                 }
2967         }
2968         kfree(cmp->src_pages);
2969         kfree(cmp->dst_pages);
2970 }
2971
2972 static int btrfs_cmp_data_prepare(struct inode *src, u64 loff,
2973                                   struct inode *dst, u64 dst_loff,
2974                                   u64 len, struct cmp_pages *cmp)
2975 {
2976         int ret;
2977         int num_pages = PAGE_ALIGN(len) >> PAGE_SHIFT;
2978         struct page **src_pgarr, **dst_pgarr;
2979
2980         /*
2981          * We must gather up all the pages before we initiate our
2982          * extent locking. We use an array for the page pointers. Size
2983          * of the array is bounded by len, which is in turn bounded by
2984          * BTRFS_MAX_DEDUPE_LEN.
2985          */
2986         src_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
2987         dst_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
2988         if (!src_pgarr || !dst_pgarr) {
2989                 kfree(src_pgarr);
2990                 kfree(dst_pgarr);
2991                 return -ENOMEM;
2992         }
2993         cmp->num_pages = num_pages;
2994         cmp->src_pages = src_pgarr;
2995         cmp->dst_pages = dst_pgarr;
2996
2997         /*
2998          * If deduping ranges in the same inode, locking rules make it mandatory
2999          * to always lock pages in ascending order to avoid deadlocks with
3000          * concurrent tasks (such as starting writeback/delalloc).
3001          */
3002         if (src == dst && dst_loff < loff) {
3003                 swap(src_pgarr, dst_pgarr);
3004                 swap(loff, dst_loff);
3005         }
3006
3007         ret = gather_extent_pages(src, src_pgarr, cmp->num_pages, loff);
3008         if (ret)
3009                 goto out;
3010
3011         ret = gather_extent_pages(dst, dst_pgarr, cmp->num_pages, dst_loff);
3012
3013 out:
3014         if (ret)
3015                 btrfs_cmp_data_free(cmp);
3016         return ret;
3017 }
3018
3019 static int btrfs_cmp_data(u64 len, struct cmp_pages *cmp)
3020 {
3021         int ret = 0;
3022         int i;
3023         struct page *src_page, *dst_page;
3024         unsigned int cmp_len = PAGE_SIZE;
3025         void *addr, *dst_addr;
3026
3027         i = 0;
3028         while (len) {
3029                 if (len < PAGE_SIZE)
3030                         cmp_len = len;
3031
3032                 BUG_ON(i >= cmp->num_pages);
3033
3034                 src_page = cmp->src_pages[i];
3035                 dst_page = cmp->dst_pages[i];
3036                 ASSERT(PageLocked(src_page));
3037                 ASSERT(PageLocked(dst_page));
3038
3039                 addr = kmap_atomic(src_page);
3040                 dst_addr = kmap_atomic(dst_page);
3041
3042                 flush_dcache_page(src_page);
3043                 flush_dcache_page(dst_page);
3044
3045                 if (memcmp(addr, dst_addr, cmp_len))
3046                         ret = -EBADE;
3047
3048                 kunmap_atomic(addr);
3049                 kunmap_atomic(dst_addr);
3050
3051                 if (ret)
3052                         break;
3053
3054                 len -= cmp_len;
3055                 i++;
3056         }
3057
3058         return ret;
3059 }
3060
3061 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 *plen,
3062                                      u64 olen)
3063 {
3064         u64 len = *plen;
3065         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
3066
3067         if (off + olen > inode->i_size || off + olen < off)
3068                 return -EINVAL;
3069
3070         /* if we extend to eof, continue to block boundary */
3071         if (off + len == inode->i_size)
3072                 *plen = len = ALIGN(inode->i_size, bs) - off;
3073
3074         /* Check that we are block aligned - btrfs_clone() requires this */
3075         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
3076                 return -EINVAL;
3077
3078         return 0;
3079 }
3080
3081 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
3082                              struct inode *dst, u64 dst_loff)
3083 {
3084         int ret;
3085         u64 len = olen;
3086         struct cmp_pages cmp;
3087         bool same_inode = (src == dst);
3088         u64 same_lock_start = 0;
3089         u64 same_lock_len = 0;
3090
3091         if (len == 0)
3092                 return 0;
3093
3094         if (same_inode)
3095                 inode_lock(src);
3096         else
3097                 btrfs_double_inode_lock(src, dst);
3098
3099         ret = extent_same_check_offsets(src, loff, &len, olen);
3100         if (ret)
3101                 goto out_unlock;
3102
3103         ret = extent_same_check_offsets(dst, dst_loff, &len, olen);
3104         if (ret)
3105                 goto out_unlock;
3106
3107         if (same_inode) {
3108                 /*
3109                  * Single inode case wants the same checks, except we
3110                  * don't want our length pushed out past i_size as
3111                  * comparing that data range makes no sense.
3112                  *
3113                  * extent_same_check_offsets() will do this for an
3114                  * unaligned length at i_size, so catch it here and
3115                  * reject the request.
3116                  *
3117                  * This effectively means we require aligned extents
3118                  * for the single-inode case, whereas the other cases
3119                  * allow an unaligned length so long as it ends at
3120                  * i_size.
3121                  */
3122                 if (len != olen) {
3123                         ret = -EINVAL;
3124                         goto out_unlock;
3125                 }
3126
3127                 /* Check for overlapping ranges */
3128                 if (dst_loff + len > loff && dst_loff < loff + len) {
3129                         ret = -EINVAL;
3130                         goto out_unlock;
3131                 }
3132
3133                 same_lock_start = min_t(u64, loff, dst_loff);
3134                 same_lock_len = max_t(u64, loff, dst_loff) + len - same_lock_start;
3135         }
3136
3137         /* don't make the dst file partly checksummed */
3138         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3139             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
3140                 ret = -EINVAL;
3141                 goto out_unlock;
3142         }
3143
3144 again:
3145         ret = btrfs_cmp_data_prepare(src, loff, dst, dst_loff, olen, &cmp);
3146         if (ret)
3147                 goto out_unlock;
3148
3149         if (same_inode)
3150                 ret = lock_extent_range(src, same_lock_start, same_lock_len,
3151                                         false);
3152         else
3153                 ret = btrfs_double_extent_lock(src, loff, dst, dst_loff, len,
3154                                                false);
3155         /*
3156          * If one of the inodes has dirty pages in the respective range or
3157          * ordered extents, we need to flush dellaloc and wait for all ordered
3158          * extents in the range. We must unlock the pages and the ranges in the
3159          * io trees to avoid deadlocks when flushing delalloc (requires locking
3160          * pages) and when waiting for ordered extents to complete (they require
3161          * range locking).
3162          */
3163         if (ret == -EAGAIN) {
3164                 /*
3165                  * Ranges in the io trees already unlocked. Now unlock all
3166                  * pages before waiting for all IO to complete.
3167                  */
3168                 btrfs_cmp_data_free(&cmp);
3169                 if (same_inode) {
3170                         btrfs_wait_ordered_range(src, same_lock_start,
3171                                                  same_lock_len);
3172                 } else {
3173                         btrfs_wait_ordered_range(src, loff, len);
3174                         btrfs_wait_ordered_range(dst, dst_loff, len);
3175                 }
3176                 goto again;
3177         }
3178         ASSERT(ret == 0);
3179         if (WARN_ON(ret)) {
3180                 /* ranges in the io trees already unlocked */
3181                 btrfs_cmp_data_free(&cmp);
3182                 return ret;
3183         }
3184
3185         /* pass original length for comparison so we stay within i_size */
3186         ret = btrfs_cmp_data(olen, &cmp);
3187         if (ret == 0)
3188                 ret = btrfs_clone(src, dst, loff, olen, len, dst_loff, 1);
3189
3190         if (same_inode)
3191                 unlock_extent(&BTRFS_I(src)->io_tree, same_lock_start,
3192                               same_lock_start + same_lock_len - 1);
3193         else
3194                 btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
3195
3196         btrfs_cmp_data_free(&cmp);
3197 out_unlock:
3198         if (same_inode)
3199                 inode_unlock(src);
3200         else
3201                 btrfs_double_inode_unlock(src, dst);
3202
3203         return ret;
3204 }
3205
3206 #define BTRFS_MAX_DEDUPE_LEN    SZ_16M
3207
3208 ssize_t btrfs_dedupe_file_range(struct file *src_file, u64 loff, u64 olen,
3209                                 struct file *dst_file, u64 dst_loff)
3210 {
3211         struct inode *src = file_inode(src_file);
3212         struct inode *dst = file_inode(dst_file);
3213         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
3214         ssize_t res;
3215
3216         if (olen > BTRFS_MAX_DEDUPE_LEN)
3217                 olen = BTRFS_MAX_DEDUPE_LEN;
3218
3219         if (WARN_ON_ONCE(bs < PAGE_SIZE)) {
3220                 /*
3221                  * Btrfs does not support blocksize < page_size. As a
3222                  * result, btrfs_cmp_data() won't correctly handle
3223                  * this situation without an update.
3224                  */
3225                 return -EINVAL;
3226         }
3227
3228         res = btrfs_extent_same(src, loff, olen, dst, dst_loff);
3229         if (res)
3230                 return res;
3231         return olen;
3232 }
3233
3234 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3235                                      struct inode *inode,
3236                                      u64 endoff,
3237                                      const u64 destoff,
3238                                      const u64 olen,
3239                                      int no_time_update)
3240 {
3241         struct btrfs_root *root = BTRFS_I(inode)->root;
3242         int ret;
3243
3244         inode_inc_iversion(inode);
3245         if (!no_time_update)
3246                 inode->i_mtime = inode->i_ctime = current_time(inode);
3247         /*
3248          * We round up to the block size at eof when determining which
3249          * extents to clone above, but shouldn't round up the file size.
3250          */
3251         if (endoff > destoff + olen)
3252                 endoff = destoff + olen;
3253         if (endoff > inode->i_size)
3254                 btrfs_i_size_write(BTRFS_I(inode), endoff);
3255
3256         ret = btrfs_update_inode(trans, root, inode);
3257         if (ret) {
3258                 btrfs_abort_transaction(trans, ret);
3259                 btrfs_end_transaction(trans);
3260                 goto out;
3261         }
3262         ret = btrfs_end_transaction(trans);
3263 out:
3264         return ret;
3265 }
3266
3267 static void clone_update_extent_map(struct btrfs_inode *inode,
3268                                     const struct btrfs_trans_handle *trans,
3269                                     const struct btrfs_path *path,
3270                                     const u64 hole_offset,
3271                                     const u64 hole_len)
3272 {
3273         struct extent_map_tree *em_tree = &inode->extent_tree;
3274         struct extent_map *em;
3275         int ret;
3276
3277         em = alloc_extent_map();
3278         if (!em) {
3279                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
3280                 return;
3281         }
3282
3283         if (path) {
3284                 struct btrfs_file_extent_item *fi;
3285
3286                 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
3287                                     struct btrfs_file_extent_item);
3288                 btrfs_extent_item_to_extent_map(inode, path, fi, false, em);
3289                 em->generation = -1;
3290                 if (btrfs_file_extent_type(path->nodes[0], fi) ==
3291                     BTRFS_FILE_EXTENT_INLINE)
3292                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3293                                         &inode->runtime_flags);
3294         } else {
3295                 em->start = hole_offset;
3296                 em->len = hole_len;
3297                 em->ram_bytes = em->len;
3298                 em->orig_start = hole_offset;
3299                 em->block_start = EXTENT_MAP_HOLE;
3300                 em->block_len = 0;
3301                 em->orig_block_len = 0;
3302                 em->compress_type = BTRFS_COMPRESS_NONE;
3303                 em->generation = trans->transid;
3304         }
3305
3306         while (1) {
3307                 write_lock(&em_tree->lock);
3308                 ret = add_extent_mapping(em_tree, em, 1);
3309                 write_unlock(&em_tree->lock);
3310                 if (ret != -EEXIST) {
3311                         free_extent_map(em);
3312                         break;
3313                 }
3314                 btrfs_drop_extent_cache(inode, em->start,
3315                                         em->start + em->len - 1, 0);
3316         }
3317
3318         if (ret)
3319                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
3320 }
3321
3322 /*
3323  * Make sure we do not end up inserting an inline extent into a file that has
3324  * already other (non-inline) extents. If a file has an inline extent it can
3325  * not have any other extents and the (single) inline extent must start at the
3326  * file offset 0. Failing to respect these rules will lead to file corruption,
3327  * resulting in EIO errors on read/write operations, hitting BUG_ON's in mm, etc
3328  *
3329  * We can have extents that have been already written to disk or we can have
3330  * dirty ranges still in delalloc, in which case the extent maps and items are
3331  * created only when we run delalloc, and the delalloc ranges might fall outside
3332  * the range we are currently locking in the inode's io tree. So we check the
3333  * inode's i_size because of that (i_size updates are done while holding the
3334  * i_mutex, which we are holding here).
3335  * We also check to see if the inode has a size not greater than "datal" but has
3336  * extents beyond it, due to an fallocate with FALLOC_FL_KEEP_SIZE (and we are
3337  * protected against such concurrent fallocate calls by the i_mutex).
3338  *
3339  * If the file has no extents but a size greater than datal, do not allow the
3340  * copy because we would need turn the inline extent into a non-inline one (even
3341  * with NO_HOLES enabled). If we find our destination inode only has one inline
3342  * extent, just overwrite it with the source inline extent if its size is less
3343  * than the source extent's size, or we could copy the source inline extent's
3344  * data into the destination inode's inline extent if the later is greater then
3345  * the former.
3346  */
3347 static int clone_copy_inline_extent(struct inode *dst,
3348                                     struct btrfs_trans_handle *trans,
3349                                     struct btrfs_path *path,
3350                                     struct btrfs_key *new_key,
3351                                     const u64 drop_start,
3352                                     const u64 datal,
3353                                     const u64 skip,
3354                                     const u64 size,
3355                                     char *inline_data)
3356 {
3357         struct btrfs_fs_info *fs_info = btrfs_sb(dst->i_sb);
3358         struct btrfs_root *root = BTRFS_I(dst)->root;
3359         const u64 aligned_end = ALIGN(new_key->offset + datal,
3360                                       fs_info->sectorsize);
3361         int ret;
3362         struct btrfs_key key;
3363
3364         if (new_key->offset > 0)
3365                 return -EOPNOTSUPP;
3366
3367         key.objectid = btrfs_ino(BTRFS_I(dst));
3368         key.type = BTRFS_EXTENT_DATA_KEY;
3369         key.offset = 0;
3370         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3371         if (ret < 0) {
3372                 return ret;
3373         } else if (ret > 0) {
3374                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3375                         ret = btrfs_next_leaf(root, path);
3376                         if (ret < 0)
3377                                 return ret;
3378                         else if (ret > 0)
3379                                 goto copy_inline_extent;
3380                 }
3381                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3382                 if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
3383                     key.type == BTRFS_EXTENT_DATA_KEY) {
3384                         ASSERT(key.offset > 0);
3385                         return -EOPNOTSUPP;
3386                 }
3387         } else if (i_size_read(dst) <= datal) {
3388                 struct btrfs_file_extent_item *ei;
3389                 u64 ext_len;
3390
3391                 /*
3392                  * If the file size is <= datal, make sure there are no other
3393                  * extents following (can happen do to an fallocate call with
3394                  * the flag FALLOC_FL_KEEP_SIZE).
3395                  */
3396                 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3397                                     struct btrfs_file_extent_item);
3398                 /*
3399                  * If it's an inline extent, it can not have other extents
3400                  * following it.
3401                  */
3402                 if (btrfs_file_extent_type(path->nodes[0], ei) ==
3403                     BTRFS_FILE_EXTENT_INLINE)
3404                         goto copy_inline_extent;
3405
3406                 ext_len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
3407                 if (ext_len > aligned_end)
3408                         return -EOPNOTSUPP;
3409
3410                 ret = btrfs_next_item(root, path);
3411                 if (ret < 0) {
3412                         return ret;
3413                 } else if (ret == 0) {
3414                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3415                                               path->slots[0]);
3416                         if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
3417                             key.type == BTRFS_EXTENT_DATA_KEY)
3418                                 return -EOPNOTSUPP;
3419                 }
3420         }
3421
3422 copy_inline_extent:
3423         /*
3424          * We have no extent items, or we have an extent at offset 0 which may
3425          * or may not be inlined. All these cases are dealt the same way.
3426          */
3427         if (i_size_read(dst) > datal) {
3428                 /*
3429                  * If the destination inode has an inline extent...
3430                  * This would require copying the data from the source inline
3431                  * extent into the beginning of the destination's inline extent.
3432                  * But this is really complex, both extents can be compressed
3433                  * or just one of them, which would require decompressing and
3434                  * re-compressing data (which could increase the new compressed
3435                  * size, not allowing the compressed data to fit anymore in an
3436                  * inline extent).
3437                  * So just don't support this case for now (it should be rare,
3438                  * we are not really saving space when cloning inline extents).
3439                  */
3440                 return -EOPNOTSUPP;
3441         }
3442
3443         btrfs_release_path(path);
3444         ret = btrfs_drop_extents(trans, root, dst, drop_start, aligned_end, 1);
3445         if (ret)
3446                 return ret;
3447         ret = btrfs_insert_empty_item(trans, root, path, new_key, size);
3448         if (ret)
3449                 return ret;
3450
3451         if (skip) {
3452                 const u32 start = btrfs_file_extent_calc_inline_size(0);
3453
3454                 memmove(inline_data + start, inline_data + start + skip, datal);
3455         }
3456
3457         write_extent_buffer(path->nodes[0], inline_data,
3458                             btrfs_item_ptr_offset(path->nodes[0],
3459                                                   path->slots[0]),
3460                             size);
3461         inode_add_bytes(dst, datal);
3462
3463         return 0;
3464 }
3465
3466 /**
3467  * btrfs_clone() - clone a range from inode file to another
3468  *
3469  * @src: Inode to clone from
3470  * @inode: Inode to clone to
3471  * @off: Offset within source to start clone from
3472  * @olen: Original length, passed by user, of range to clone
3473  * @olen_aligned: Block-aligned value of olen
3474  * @destoff: Offset within @inode to start clone
3475  * @no_time_update: Whether to update mtime/ctime on the target inode
3476  */
3477 static int btrfs_clone(struct inode *src, struct inode *inode,
3478                        const u64 off, const u64 olen, const u64 olen_aligned,
3479                        const u64 destoff, int no_time_update)
3480 {
3481         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3482         struct btrfs_root *root = BTRFS_I(inode)->root;
3483         struct btrfs_path *path = NULL;
3484         struct extent_buffer *leaf;
3485         struct btrfs_trans_handle *trans;
3486         char *buf = NULL;
3487         struct btrfs_key key;
3488         u32 nritems;
3489         int slot;
3490         int ret;
3491         const u64 len = olen_aligned;
3492         u64 last_dest_end = destoff;
3493
3494         ret = -ENOMEM;
3495         buf = kvmalloc(fs_info->nodesize, GFP_KERNEL);
3496         if (!buf)
3497                 return ret;
3498
3499         path = btrfs_alloc_path();
3500         if (!path) {
3501                 kvfree(buf);
3502                 return ret;
3503         }
3504
3505         path->reada = READA_FORWARD;
3506         /* clone data */
3507         key.objectid = btrfs_ino(BTRFS_I(src));
3508         key.type = BTRFS_EXTENT_DATA_KEY;
3509         key.offset = off;
3510
3511         while (1) {
3512                 u64 next_key_min_offset = key.offset + 1;
3513
3514                 /*
3515                  * note the key will change type as we walk through the
3516                  * tree.
3517                  */
3518                 path->leave_spinning = 1;
3519                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
3520                                 0, 0);
3521                 if (ret < 0)
3522                         goto out;
3523                 /*
3524                  * First search, if no extent item that starts at offset off was
3525                  * found but the previous item is an extent item, it's possible
3526                  * it might overlap our target range, therefore process it.
3527                  */
3528                 if (key.offset == off && ret > 0 && path->slots[0] > 0) {
3529                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3530                                               path->slots[0] - 1);
3531                         if (key.type == BTRFS_EXTENT_DATA_KEY)
3532                                 path->slots[0]--;
3533                 }
3534
3535                 nritems = btrfs_header_nritems(path->nodes[0]);
3536 process_slot:
3537                 if (path->slots[0] >= nritems) {
3538                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
3539                         if (ret < 0)
3540                                 goto out;
3541                         if (ret > 0)
3542                                 break;
3543                         nritems = btrfs_header_nritems(path->nodes[0]);
3544                 }
3545                 leaf = path->nodes[0];
3546                 slot = path->slots[0];
3547
3548                 btrfs_item_key_to_cpu(leaf, &key, slot);
3549                 if (key.type > BTRFS_EXTENT_DATA_KEY ||
3550                     key.objectid != btrfs_ino(BTRFS_I(src)))
3551                         break;
3552
3553                 if (key.type == BTRFS_EXTENT_DATA_KEY) {
3554                         struct btrfs_file_extent_item *extent;
3555                         int type;
3556                         u32 size;
3557                         struct btrfs_key new_key;
3558                         u64 disko = 0, diskl = 0;
3559                         u64 datao = 0, datal = 0;
3560                         u8 comp;
3561                         u64 drop_start;
3562
3563                         extent = btrfs_item_ptr(leaf, slot,
3564                                                 struct btrfs_file_extent_item);
3565                         comp = btrfs_file_extent_compression(leaf, extent);
3566                         type = btrfs_file_extent_type(leaf, extent);
3567                         if (type == BTRFS_FILE_EXTENT_REG ||
3568                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3569                                 disko = btrfs_file_extent_disk_bytenr(leaf,
3570                                                                       extent);
3571                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
3572                                                                  extent);
3573                                 datao = btrfs_file_extent_offset(leaf, extent);
3574                                 datal = btrfs_file_extent_num_bytes(leaf,
3575                                                                     extent);
3576                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3577                                 /* take upper bound, may be compressed */
3578                                 datal = btrfs_file_extent_ram_bytes(leaf,
3579                                                                     extent);
3580                         }
3581
3582                         /*
3583                          * The first search might have left us at an extent
3584                          * item that ends before our target range's start, can
3585                          * happen if we have holes and NO_HOLES feature enabled.
3586                          */
3587                         if (key.offset + datal <= off) {
3588                                 path->slots[0]++;
3589                                 goto process_slot;
3590                         } else if (key.offset >= off + len) {
3591                                 break;
3592                         }
3593                         next_key_min_offset = key.offset + datal;
3594                         size = btrfs_item_size_nr(leaf, slot);
3595                         read_extent_buffer(leaf, buf,
3596                                            btrfs_item_ptr_offset(leaf, slot),
3597                                            size);
3598
3599                         btrfs_release_path(path);
3600                         path->leave_spinning = 0;
3601
3602                         memcpy(&new_key, &key, sizeof(new_key));
3603                         new_key.objectid = btrfs_ino(BTRFS_I(inode));
3604                         if (off <= key.offset)
3605                                 new_key.offset = key.offset + destoff - off;
3606                         else
3607                                 new_key.offset = destoff;
3608
3609                         /*
3610                          * Deal with a hole that doesn't have an extent item
3611                          * that represents it (NO_HOLES feature enabled).
3612                          * This hole is either in the middle of the cloning
3613                          * range or at the beginning (fully overlaps it or
3614                          * partially overlaps it).
3615                          */
3616                         if (new_key.offset != last_dest_end)
3617                                 drop_start = last_dest_end;
3618                         else
3619                                 drop_start = new_key.offset;
3620
3621                         /*
3622                          * 1 - adjusting old extent (we may have to split it)
3623                          * 1 - add new extent
3624                          * 1 - inode update
3625                          */
3626                         trans = btrfs_start_transaction(root, 3);
3627                         if (IS_ERR(trans)) {
3628                                 ret = PTR_ERR(trans);
3629                                 goto out;
3630                         }
3631
3632                         if (type == BTRFS_FILE_EXTENT_REG ||
3633                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3634                                 /*
3635                                  *    a  | --- range to clone ---|  b
3636                                  * | ------------- extent ------------- |
3637                                  */
3638
3639                                 /* subtract range b */
3640                                 if (key.offset + datal > off + len)
3641                                         datal = off + len - key.offset;
3642
3643                                 /* subtract range a */
3644                                 if (off > key.offset) {
3645                                         datao += off - key.offset;
3646                                         datal -= off - key.offset;
3647                                 }
3648
3649                                 ret = btrfs_drop_extents(trans, root, inode,
3650                                                          drop_start,
3651                                                          new_key.offset + datal,
3652                                                          1);
3653                                 if (ret) {
3654                                         if (ret != -EOPNOTSUPP)
3655                                                 btrfs_abort_transaction(trans,
3656                                                                         ret);
3657                                         btrfs_end_transaction(trans);
3658                                         goto out;
3659                                 }
3660
3661                                 ret = btrfs_insert_empty_item(trans, root, path,
3662                                                               &new_key, size);
3663                                 if (ret) {
3664                                         btrfs_abort_transaction(trans, ret);
3665                                         btrfs_end_transaction(trans);
3666                                         goto out;
3667                                 }
3668
3669                                 leaf = path->nodes[0];
3670                                 slot = path->slots[0];
3671                                 write_extent_buffer(leaf, buf,
3672                                             btrfs_item_ptr_offset(leaf, slot),
3673                                             size);
3674
3675                                 extent = btrfs_item_ptr(leaf, slot,
3676                                                 struct btrfs_file_extent_item);
3677
3678                                 /* disko == 0 means it's a hole */
3679                                 if (!disko)
3680                                         datao = 0;
3681
3682                                 btrfs_set_file_extent_offset(leaf, extent,
3683                                                              datao);
3684                                 btrfs_set_file_extent_num_bytes(leaf, extent,
3685                                                                 datal);
3686
3687                                 if (disko) {
3688                                         inode_add_bytes(inode, datal);
3689                                         ret = btrfs_inc_extent_ref(trans,
3690                                                         root,
3691                                                         disko, diskl, 0,
3692                                                         root->root_key.objectid,
3693                                                         btrfs_ino(BTRFS_I(inode)),
3694                                                         new_key.offset - datao);
3695                                         if (ret) {
3696                                                 btrfs_abort_transaction(trans,
3697                                                                         ret);