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/statfs.h>
  37 #include <linux/compat.h>
  38 #include <linux/bit_spinlock.h>
  39 #include <linux/security.h>
  40 #include <linux/xattr.h>
  41 #include <linux/vmalloc.h>
  42 #include <linux/slab.h>
  43 #include <linux/blkdev.h>
  44 #include <linux/uuid.h>
  45 #include <linux/btrfs.h>
  46 #include <linux/uaccess.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
  91 static int btrfs_clone(struct inode *src, struct inode *inode,
  92                        u64 off, u64 olen, u64 olen_aligned, u64 destoff,
  93                        int no_time_update);
  94
  95 /* Mask out flags that are inappropriate for the given type of inode. */
  96 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
  97 {
  98         if (S_ISDIR(mode))
  99                 return flags;
 100         else if (S_ISREG(mode))
 101                 return flags & ~FS_DIRSYNC_FL;
 102         else
 103                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
 104 }
 105
 106 /*
 107  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
 108  */
 109 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
 110 {
 111         unsigned int iflags = 0;
 112
 113         if (flags & BTRFS_INODE_SYNC)
 114                 iflags |= FS_SYNC_FL;
 115         if (flags & BTRFS_INODE_IMMUTABLE)
 116                 iflags |= FS_IMMUTABLE_FL;
 117         if (flags & BTRFS_INODE_APPEND)
 118                 iflags |= FS_APPEND_FL;
 119         if (flags & BTRFS_INODE_NODUMP)
 120                 iflags |= FS_NODUMP_FL;
 121         if (flags & BTRFS_INODE_NOATIME)
 122                 iflags |= FS_NOATIME_FL;
 123         if (flags & BTRFS_INODE_DIRSYNC)
 124                 iflags |= FS_DIRSYNC_FL;
 125         if (flags & BTRFS_INODE_NODATACOW)
 126                 iflags |= FS_NOCOW_FL;
 127
 128         if (flags & BTRFS_INODE_NOCOMPRESS)
 129                 iflags |= FS_NOCOMP_FL;
 130         else if (flags & BTRFS_INODE_COMPRESS)
 131                 iflags |= FS_COMPR_FL;
 132
 133         return iflags;
 134 }
 135
 136 /*
 137  * Update inode->i_flags based on the btrfs internal flags.
 138  */
 139 void btrfs_update_iflags(struct inode *inode)
 140 {
 141         struct btrfs_inode *ip = BTRFS_I(inode);
 142         unsigned int new_fl = 0;
 143
 144         if (ip->flags & BTRFS_INODE_SYNC)
 145                 new_fl |= S_SYNC;
 146         if (ip->flags & BTRFS_INODE_IMMUTABLE)
 147                 new_fl |= S_IMMUTABLE;
 148         if (ip->flags & BTRFS_INODE_APPEND)
 149                 new_fl |= S_APPEND;
 150         if (ip->flags & BTRFS_INODE_NOATIME)
 151                 new_fl |= S_NOATIME;
 152         if (ip->flags & BTRFS_INODE_DIRSYNC)
 153                 new_fl |= S_DIRSYNC;
 154
 155         set_mask_bits(&inode->i_flags,
 156                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC,
 157                       new_fl);
 158 }
 159
 160 /*
 161  * Inherit flags from the parent inode.
 162  *
 163  * Currently only the compression flags and the cow flags are inherited.
 164  */
 165 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
 166 {
 167         unsigned int flags;
 168
 169         if (!dir)
 170                 return;
 171
 172         flags = BTRFS_I(dir)->flags;
 173
 174         if (flags & BTRFS_INODE_NOCOMPRESS) {
 175                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
 176                 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 177         } else if (flags & BTRFS_INODE_COMPRESS) {
 178                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
 179                 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
 180         }
 181
 182         if (flags & BTRFS_INODE_NODATACOW) {
 183                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
 184                 if (S_ISREG(inode->i_mode))
 185                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
 186         }
 187
 188         btrfs_update_iflags(inode);
 189 }
 190
 191 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 192 {
 193         struct btrfs_inode *ip = BTRFS_I(file_inode(file));
 194         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
 195
 196         if (copy_to_user(arg, &flags, sizeof(flags)))
 197                 return -EFAULT;
 198         return 0;
 199 }
 200
 201 static int check_flags(unsigned int flags)
 202 {
 203         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 204                       FS_NOATIME_FL | FS_NODUMP_FL | \
 205                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 206                       FS_NOCOMP_FL | FS_COMPR_FL |
 207                       FS_NOCOW_FL))
 208                 return -EOPNOTSUPP;
 209
 210         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 211                 return -EINVAL;
 212
 213         return 0;
 214 }
 215
 216 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 217 {
 218         struct inode *inode = file_inode(file);
 219         struct btrfs_inode *ip = BTRFS_I(inode);
 220         struct btrfs_root *root = ip->root;
 221         struct btrfs_trans_handle *trans;
 222         unsigned int flags, oldflags;
 223         int ret;
 224         u64 ip_oldflags;
 225         unsigned int i_oldflags;
 226         umode_t mode;
 227
 228         if (!inode_owner_or_capable(inode))
 229                 return -EPERM;
 230
 231         if (btrfs_root_readonly(root))
 232                 return -EROFS;
 233
 234         if (copy_from_user(&flags, arg, sizeof(flags)))
 235                 return -EFAULT;
 236
 237         ret = check_flags(flags);
 238         if (ret)
 239                 return ret;
 240
 241         ret = mnt_want_write_file(file);
 242         if (ret)
 243                 return ret;
 244
 245         inode_lock(inode);
 246
 247         ip_oldflags = ip->flags;
 248         i_oldflags = inode->i_flags;
 249         mode = inode->i_mode;
 250
 251         flags = btrfs_mask_flags(inode->i_mode, flags);
 252         oldflags = btrfs_flags_to_ioctl(ip->flags);
 253         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 254                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 255                         ret = -EPERM;
 256                         goto out_unlock;
 257                 }
 258         }
 259
 260         if (flags & FS_SYNC_FL)
 261                 ip->flags |= BTRFS_INODE_SYNC;
 262         else
 263                 ip->flags &= ~BTRFS_INODE_SYNC;
 264         if (flags & FS_IMMUTABLE_FL)
 265                 ip->flags |= BTRFS_INODE_IMMUTABLE;
 266         else
 267                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
 268         if (flags & FS_APPEND_FL)
 269                 ip->flags |= BTRFS_INODE_APPEND;
 270         else
 271                 ip->flags &= ~BTRFS_INODE_APPEND;
 272         if (flags & FS_NODUMP_FL)
 273                 ip->flags |= BTRFS_INODE_NODUMP;
 274         else
 275                 ip->flags &= ~BTRFS_INODE_NODUMP;
 276         if (flags & FS_NOATIME_FL)
 277                 ip->flags |= BTRFS_INODE_NOATIME;
 278         else
 279                 ip->flags &= ~BTRFS_INODE_NOATIME;
 280         if (flags & FS_DIRSYNC_FL)
 281                 ip->flags |= BTRFS_INODE_DIRSYNC;
 282         else
 283                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
 284         if (flags & FS_NOCOW_FL) {
 285                 if (S_ISREG(mode)) {
 286                         /*
 287                          * It's safe to turn csums off here, no extents exist.
 288                          * Otherwise we want the flag to reflect the real COW
 289                          * status of the file and will not set it.
 290                          */
 291                         if (inode->i_size == 0)
 292                                 ip->flags |= BTRFS_INODE_NODATACOW
 293                                            | BTRFS_INODE_NODATASUM;
 294                 } else {
 295                         ip->flags |= BTRFS_INODE_NODATACOW;
 296                 }
 297         } else {
 298                 /*
 299                  * Revert back under same assumptions as above
 300                  */
 301                 if (S_ISREG(mode)) {
 302                         if (inode->i_size == 0)
 303                                 ip->flags &= ~(BTRFS_INODE_NODATACOW
 304                                              | BTRFS_INODE_NODATASUM);
 305                 } else {
 306                         ip->flags &= ~BTRFS_INODE_NODATACOW;
 307                 }
 308         }
 309
 310         /*
 311          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 312          * flag may be changed automatically if compression code won't make
 313          * things smaller.
 314          */
 315         if (flags & FS_NOCOMP_FL) {
 316                 ip->flags &= ~BTRFS_INODE_COMPRESS;
 317                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
 318
 319                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 320                 if (ret && ret != -ENODATA)
 321                         goto out_drop;
 322         } else if (flags & FS_COMPR_FL) {
 323                 const char *comp;
 324
 325                 ip->flags |= BTRFS_INODE_COMPRESS;
 326                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
 327
 328                 if (root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
 329                         comp = "lzo";
 330                 else
 331                         comp = "zlib";
 332                 ret = btrfs_set_prop(inode, "btrfs.compression",
 333                                      comp, strlen(comp), 0);
 334                 if (ret)
 335                         goto out_drop;
 336
 337         } else {
 338                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 339                 if (ret && ret != -ENODATA)
 340                         goto out_drop;
 341                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 342         }
 343
 344         trans = btrfs_start_transaction(root, 1);
 345         if (IS_ERR(trans)) {
 346                 ret = PTR_ERR(trans);
 347                 goto out_drop;
 348         }
 349
 350         btrfs_update_iflags(inode);
 351         inode_inc_iversion(inode);
 352         inode->i_ctime = current_fs_time(inode->i_sb);
 353         ret = btrfs_update_inode(trans, root, inode);
 354
 355         btrfs_end_transaction(trans, root);
 356  out_drop:
 357         if (ret) {
 358                 ip->flags = ip_oldflags;
 359                 inode->i_flags = i_oldflags;
 360         }
 361
 362  out_unlock:
 363         inode_unlock(inode);
 364         mnt_drop_write_file(file);
 365         return ret;
 366 }
 367
 368 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 369 {
 370         struct inode *inode = file_inode(file);
 371
 372         return put_user(inode->i_generation, arg);
 373 }
 374
 375 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
 376 {
 377         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
 378         struct btrfs_device *device;
 379         struct request_queue *q;
 380         struct fstrim_range range;
 381         u64 minlen = ULLONG_MAX;
 382         u64 num_devices = 0;
 383         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
 384         int ret;
 385
 386         if (!capable(CAP_SYS_ADMIN))
 387                 return -EPERM;
 388
 389         rcu_read_lock();
 390         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 391                                 dev_list) {
 392                 if (!device->bdev)
 393                         continue;
 394                 q = bdev_get_queue(device->bdev);
 395                 if (blk_queue_discard(q)) {
 396                         num_devices++;
 397                         minlen = min((u64)q->limits.discard_granularity,
 398                                      minlen);
 399                 }
 400         }
 401         rcu_read_unlock();
 402
 403         if (!num_devices)
 404                 return -EOPNOTSUPP;
 405         if (copy_from_user(&range, arg, sizeof(range)))
 406                 return -EFAULT;
 407         if (range.start > total_bytes ||
 408             range.len < fs_info->sb->s_blocksize)
 409                 return -EINVAL;
 410
 411         range.len = min(range.len, total_bytes - range.start);
 412         range.minlen = max(range.minlen, minlen);
 413         ret = btrfs_trim_fs(fs_info->tree_root, &range);
 414         if (ret < 0)
 415                 return ret;
 416
 417         if (copy_to_user(arg, &range, sizeof(range)))
 418                 return -EFAULT;
 419
 420         return 0;
 421 }
 422
 423 int btrfs_is_empty_uuid(u8 *uuid)
 424 {
 425         int i;
 426
 427         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 428                 if (uuid[i])
 429                         return 0;
 430         }
 431         return 1;
 432 }
 433
 434 static noinline int create_subvol(struct inode *dir,
 435                                   struct dentry *dentry,
 436                                   char *name, int namelen,
 437                                   u64 *async_transid,
 438                                   struct btrfs_qgroup_inherit *inherit)
 439 {
 440         struct btrfs_trans_handle *trans;
 441         struct btrfs_key key;
 442         struct btrfs_root_item *root_item;
 443         struct btrfs_inode_item *inode_item;
 444         struct extent_buffer *leaf;
 445         struct btrfs_root *root = BTRFS_I(dir)->root;
 446         struct btrfs_root *new_root;
 447         struct btrfs_block_rsv block_rsv;
 448         struct timespec cur_time = current_fs_time(dir->i_sb);
 449         struct inode *inode;
 450         int ret;
 451         int err;
 452         u64 objectid;
 453         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 454         u64 index = 0;
 455         u64 qgroup_reserved;
 456         uuid_le new_uuid;
 457
 458         root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
 459         if (!root_item)
 460                 return -ENOMEM;
 461
 462         ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid);
 463         if (ret)
 464                 goto fail_free;
 465
 466         /*
 467          * Don't create subvolume whose level is not zero. Or qgroup will be
 468          * screwed up since it assumes subvolume qgroup's level to be 0.
 469          */
 470         if (btrfs_qgroup_level(objectid)) {
 471                 ret = -ENOSPC;
 472                 goto fail_free;
 473         }
 474
 475         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 476         /*
 477          * The same as the snapshot creation, please see the comment
 478          * of create_snapshot().
 479          */
 480         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
 481                                                8, &qgroup_reserved, false);
 482         if (ret)
 483                 goto fail_free;
 484
 485         trans = btrfs_start_transaction(root, 0);
 486         if (IS_ERR(trans)) {
 487                 ret = PTR_ERR(trans);
 488                 btrfs_subvolume_release_metadata(root, &block_rsv,
 489                                                  qgroup_reserved);
 490                 goto fail_free;
 491         }
 492         trans->block_rsv = &block_rsv;
 493         trans->bytes_reserved = block_rsv.size;
 494
 495         ret = btrfs_qgroup_inherit(trans, root->fs_info, 0, objectid, inherit);
 496         if (ret)
 497                 goto fail;
 498
 499         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
 500         if (IS_ERR(leaf)) {
 501                 ret = PTR_ERR(leaf);
 502                 goto fail;
 503         }
 504
 505         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
 506         btrfs_set_header_bytenr(leaf, leaf->start);
 507         btrfs_set_header_generation(leaf, trans->transid);
 508         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 509         btrfs_set_header_owner(leaf, objectid);
 510
 511         write_extent_buffer(leaf, root->fs_info->fsid, btrfs_header_fsid(),
 512                             BTRFS_FSID_SIZE);
 513         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 514                             btrfs_header_chunk_tree_uuid(leaf),
 515                             BTRFS_UUID_SIZE);
 516         btrfs_mark_buffer_dirty(leaf);
 517
 518         inode_item = &root_item->inode;
 519         btrfs_set_stack_inode_generation(inode_item, 1);
 520         btrfs_set_stack_inode_size(inode_item, 3);
 521         btrfs_set_stack_inode_nlink(inode_item, 1);
 522         btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
 523         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 524
 525         btrfs_set_root_flags(root_item, 0);
 526         btrfs_set_root_limit(root_item, 0);
 527         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 528
 529         btrfs_set_root_bytenr(root_item, leaf->start);
 530         btrfs_set_root_generation(root_item, trans->transid);
 531         btrfs_set_root_level(root_item, 0);
 532         btrfs_set_root_refs(root_item, 1);
 533         btrfs_set_root_used(root_item, leaf->len);
 534         btrfs_set_root_last_snapshot(root_item, 0);
 535
 536         btrfs_set_root_generation_v2(root_item,
 537                         btrfs_root_generation(root_item));
 538         uuid_le_gen(&new_uuid);
 539         memcpy(root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
 540         btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
 541         btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
 542         root_item->ctime = root_item->otime;
 543         btrfs_set_root_ctransid(root_item, trans->transid);
 544         btrfs_set_root_otransid(root_item, trans->transid);
 545
 546         btrfs_tree_unlock(leaf);
 547         free_extent_buffer(leaf);
 548         leaf = NULL;
 549
 550         btrfs_set_root_dirid(root_item, new_dirid);
 551
 552         key.objectid = objectid;
 553         key.offset = 0;
 554         key.type = BTRFS_ROOT_ITEM_KEY;
 555         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 556                                 root_item);
 557         if (ret)
 558                 goto fail;
 559
 560         key.offset = (u64)-1;
 561         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
 562         if (IS_ERR(new_root)) {
 563                 ret = PTR_ERR(new_root);
 564                 btrfs_abort_transaction(trans, ret);
 565                 goto fail;
 566         }
 567
 568         btrfs_record_root_in_trans(trans, new_root);
 569
 570         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
 571         if (ret) {
 572                 /* We potentially lose an unused inode item here */
 573                 btrfs_abort_transaction(trans, ret);
 574                 goto fail;
 575         }
 576
 577         mutex_lock(&new_root->objectid_mutex);
 578         new_root->highest_objectid = new_dirid;
 579         mutex_unlock(&new_root->objectid_mutex);
 580
 581         /*
 582          * insert the directory item
 583          */
 584         ret = btrfs_set_inode_index(dir, &index);
 585         if (ret) {
 586                 btrfs_abort_transaction(trans, ret);
 587                 goto fail;
 588         }
 589
 590         ret = btrfs_insert_dir_item(trans, root,
 591                                     name, namelen, dir, &key,
 592                                     BTRFS_FT_DIR, index);
 593         if (ret) {
 594                 btrfs_abort_transaction(trans, ret);
 595                 goto fail;
 596         }
 597
 598         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
 599         ret = btrfs_update_inode(trans, root, dir);
 600         BUG_ON(ret);
 601
 602         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
 603                                  objectid, root->root_key.objectid,
 604                                  btrfs_ino(dir), index, name, namelen);
 605         BUG_ON(ret);
 606
 607         ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
 608                                   root_item->uuid, BTRFS_UUID_KEY_SUBVOL,
 609                                   objectid);
 610         if (ret)
 611                 btrfs_abort_transaction(trans, ret);
 612
 613 fail:
 614         kfree(root_item);
 615         trans->block_rsv = NULL;
 616         trans->bytes_reserved = 0;
 617         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
 618
 619         if (async_transid) {
 620                 *async_transid = trans->transid;
 621                 err = btrfs_commit_transaction_async(trans, root, 1);
 622                 if (err)
 623                         err = btrfs_commit_transaction(trans, root);
 624         } else {
 625                 err = btrfs_commit_transaction(trans, root);
 626         }
 627         if (err && !ret)
 628                 ret = err;
 629
 630         if (!ret) {
 631                 inode = btrfs_lookup_dentry(dir, dentry);
 632                 if (IS_ERR(inode))
 633                         return PTR_ERR(inode);
 634                 d_instantiate(dentry, inode);
 635         }
 636         return ret;
 637
 638 fail_free:
 639         kfree(root_item);
 640         return ret;
 641 }
 642
 643 static void btrfs_wait_for_no_snapshoting_writes(struct btrfs_root *root)
 644 {
 645         s64 writers;
 646         DEFINE_WAIT(wait);
 647
 648         do {
 649                 prepare_to_wait(&root->subv_writers->wait, &wait,
 650                                 TASK_UNINTERRUPTIBLE);
 651
 652                 writers = percpu_counter_sum(&root->subv_writers->counter);
 653                 if (writers)
 654                         schedule();
 655
 656                 finish_wait(&root->subv_writers->wait, &wait);
 657         } while (writers);
 658 }
 659
 660 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 661                            struct dentry *dentry, char *name, int namelen,
 662                            u64 *async_transid, bool readonly,
 663                            struct btrfs_qgroup_inherit *inherit)
 664 {
 665         struct inode *inode;
 666         struct btrfs_pending_snapshot *pending_snapshot;
 667         struct btrfs_trans_handle *trans;
 668         int ret;
 669
 670         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 671                 return -EINVAL;
 672
 673         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
 674         if (!pending_snapshot)
 675                 return -ENOMEM;
 676
 677         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 678                         GFP_NOFS);
 679         pending_snapshot->path = btrfs_alloc_path();
 680         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 681                 ret = -ENOMEM;
 682                 goto free_pending;
 683         }
 684
 685         atomic_inc(&root->will_be_snapshoted);
 686         smp_mb__after_atomic();
 687         btrfs_wait_for_no_snapshoting_writes(root);
 688
 689         ret = btrfs_start_delalloc_inodes(root, 0);
 690         if (ret)
 691                 goto dec_and_free;
 692
 693         btrfs_wait_ordered_extents(root, -1, 0, (u64)-1);
 694
 695         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 696                              BTRFS_BLOCK_RSV_TEMP);
 697         /*
 698          * 1 - parent dir inode
 699          * 2 - dir entries
 700          * 1 - root item
 701          * 2 - root ref/backref
 702          * 1 - root of snapshot
 703          * 1 - UUID item
 704          */
 705         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 706                                         &pending_snapshot->block_rsv, 8,
 707                                         &pending_snapshot->qgroup_reserved,
 708                                         false);
 709         if (ret)
 710                 goto dec_and_free;
 711
 712         pending_snapshot->dentry = dentry;
 713         pending_snapshot->root = root;
 714         pending_snapshot->readonly = readonly;
 715         pending_snapshot->dir = dir;
 716         pending_snapshot->inherit = inherit;
 717
 718         trans = btrfs_start_transaction(root, 0);
 719         if (IS_ERR(trans)) {
 720                 ret = PTR_ERR(trans);
 721                 goto fail;
 722         }
 723
 724         spin_lock(&root->fs_info->trans_lock);
 725         list_add(&pending_snapshot->list,
 726                  &trans->transaction->pending_snapshots);
 727         spin_unlock(&root->fs_info->trans_lock);
 728         if (async_transid) {
 729                 *async_transid = trans->transid;
 730                 ret = btrfs_commit_transaction_async(trans,
 731                                      root->fs_info->extent_root, 1);
 732                 if (ret)
 733                         ret = btrfs_commit_transaction(trans, root);
 734         } else {
 735                 ret = btrfs_commit_transaction(trans,
 736                                                root->fs_info->extent_root);
 737         }
 738         if (ret)
 739                 goto fail;
 740
 741         ret = pending_snapshot->error;
 742         if (ret)
 743                 goto fail;
 744
 745         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 746         if (ret)
 747                 goto fail;
 748
 749         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 750         if (IS_ERR(inode)) {
 751                 ret = PTR_ERR(inode);
 752                 goto fail;
 753         }
 754
 755         d_instantiate(dentry, inode);
 756         ret = 0;
 757 fail:
 758         btrfs_subvolume_release_metadata(BTRFS_I(dir)->root,
 759                                          &pending_snapshot->block_rsv,
 760                                          pending_snapshot->qgroup_reserved);
 761 dec_and_free:
 762         if (atomic_dec_and_test(&root->will_be_snapshoted))
 763                 wake_up_atomic_t(&root->will_be_snapshoted);
 764 free_pending:
 765         kfree(pending_snapshot->root_item);
 766         btrfs_free_path(pending_snapshot->path);
 767         kfree(pending_snapshot);
 768
 769         return ret;
 770 }
 771
 772 /*  copy of may_delete in fs/namei.c()
 773  *      Check whether we can remove a link victim from directory dir, check
 774  *  whether the type of victim is right.
 775  *  1. We can't do it if dir is read-only (done in permission())
 776  *  2. We should have write and exec permissions on dir
 777  *  3. We can't remove anything from append-only dir
 778  *  4. We can't do anything with immutable dir (done in permission())
 779  *  5. If the sticky bit on dir is set we should either
 780  *      a. be owner of dir, or
 781  *      b. be owner of victim, or
 782  *      c. have CAP_FOWNER capability
 783  *  6. If the victim is append-only or immutable we can't do anything with
 784  *     links pointing to it.
 785  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 786  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 787  *  9. We can't remove a root or mountpoint.
 788  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 789  *     nfs_async_unlink().
 790  */
 791
 792 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 793 {
 794         int error;
 795
 796         if (d_really_is_negative(victim))
 797                 return -ENOENT;
 798
 799         BUG_ON(d_inode(victim->d_parent) != dir);
 800         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 801
 802         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 803         if (error)
 804                 return error;
 805         if (IS_APPEND(dir))
 806                 return -EPERM;
 807         if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
 808             IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
 809                 return -EPERM;
 810         if (isdir) {
 811                 if (!d_is_dir(victim))
 812                         return -ENOTDIR;
 813                 if (IS_ROOT(victim))
 814                         return -EBUSY;
 815         } else if (d_is_dir(victim))
 816                 return -EISDIR;
 817         if (IS_DEADDIR(dir))
 818                 return -ENOENT;
 819         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 820                 return -EBUSY;
 821         return 0;
 822 }
 823
 824 /* copy of may_create in fs/namei.c() */
 825 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 826 {
 827         if (d_really_is_positive(child))
 828                 return -EEXIST;
 829         if (IS_DEADDIR(dir))
 830                 return -ENOENT;
 831         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 832 }
 833
 834 /*
 835  * Create a new subvolume below @parent.  This is largely modeled after
 836  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 837  * inside this filesystem so it's quite a bit simpler.
 838  */
 839 static noinline int btrfs_mksubvol(struct path *parent,
 840                                    char *name, int namelen,
 841                                    struct btrfs_root *snap_src,
 842                                    u64 *async_transid, bool readonly,
 843                                    struct btrfs_qgroup_inherit *inherit)
 844 {
 845         struct inode *dir  = d_inode(parent->dentry);
 846         struct dentry *dentry;
 847         int error;
 848
 849         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
 850         if (error == -EINTR)
 851                 return error;
 852
 853         dentry = lookup_one_len(name, parent->dentry, namelen);
 854         error = PTR_ERR(dentry);
 855         if (IS_ERR(dentry))
 856                 goto out_unlock;
 857
 858         error = btrfs_may_create(dir, dentry);
 859         if (error)
 860                 goto out_dput;
 861
 862         /*
 863          * even if this name doesn't exist, we may get hash collisions.
 864          * check for them now when we can safely fail
 865          */
 866         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 867                                                dir->i_ino, name,
 868                                                namelen);
 869         if (error)
 870                 goto out_dput;
 871
 872         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 873
 874         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 875                 goto out_up_read;
 876
 877         if (snap_src) {
 878                 error = create_snapshot(snap_src, dir, dentry, name, namelen,
 879                                         async_transid, readonly, inherit);
 880         } else {
 881                 error = create_subvol(dir, dentry, name, namelen,
 882                                       async_transid, inherit);
 883         }
 884         if (!error)
 885                 fsnotify_mkdir(dir, dentry);
 886 out_up_read:
 887         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 888 out_dput:
 889         dput(dentry);
 890 out_unlock:
 891         inode_unlock(dir);
 892         return error;
 893 }
 894
 895 /*
 896  * When we're defragging a range, we don't want to kick it off again
 897  * if it is really just waiting for delalloc to send it down.
 898  * If we find a nice big extent or delalloc range for the bytes in the
 899  * file you want to defrag, we return 0 to let you know to skip this
 900  * part of the file
 901  */
 902 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
 903 {
 904         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 905         struct extent_map *em = NULL;
 906         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 907         u64 end;
 908
 909         read_lock(&em_tree->lock);
 910         em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
 911         read_unlock(&em_tree->lock);
 912
 913         if (em) {
 914                 end = extent_map_end(em);
 915                 free_extent_map(em);
 916                 if (end - offset > thresh)
 917                         return 0;
 918         }
 919         /* if we already have a nice delalloc here, just stop */
 920         thresh /= 2;
 921         end = count_range_bits(io_tree, &offset, offset + thresh,
 922                                thresh, EXTENT_DELALLOC, 1);
 923         if (end >= thresh)
 924                 return 0;
 925         return 1;
 926 }
 927
 928 /*
 929  * helper function to walk through a file and find extents
 930  * newer than a specific transid, and smaller than thresh.
 931  *
 932  * This is used by the defragging code to find new and small
 933  * extents
 934  */
 935 static int find_new_extents(struct btrfs_root *root,
 936                             struct inode *inode, u64 newer_than,
 937                             u64 *off, u32 thresh)
 938 {
 939         struct btrfs_path *path;
 940         struct btrfs_key min_key;
 941         struct extent_buffer *leaf;
 942         struct btrfs_file_extent_item *extent;
 943         int type;
 944         int ret;
 945         u64 ino = btrfs_ino(inode);
 946
 947         path = btrfs_alloc_path();
 948         if (!path)
 949                 return -ENOMEM;
 950
 951         min_key.objectid = ino;
 952         min_key.type = BTRFS_EXTENT_DATA_KEY;
 953         min_key.offset = *off;
 954
 955         while (1) {
 956                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
 957                 if (ret != 0)
 958                         goto none;
 959 process_slot:
 960                 if (min_key.objectid != ino)
 961                         goto none;
 962                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
 963                         goto none;
 964
 965                 leaf = path->nodes[0];
 966                 extent = btrfs_item_ptr(leaf, path->slots[0],
 967                                         struct btrfs_file_extent_item);
 968
 969                 type = btrfs_file_extent_type(leaf, extent);
 970                 if (type == BTRFS_FILE_EXTENT_REG &&
 971                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
 972                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
 973                         *off = min_key.offset;
 974                         btrfs_free_path(path);
 975                         return 0;
 976                 }
 977
 978                 path->slots[0]++;
 979                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
 980                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
 981                         goto process_slot;
 982                 }
 983
 984                 if (min_key.offset == (u64)-1)
 985                         goto none;
 986
 987                 min_key.offset++;
 988                 btrfs_release_path(path);
 989         }
 990 none:
 991         btrfs_free_path(path);
 992         return -ENOENT;
 993 }
 994
 995 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
 996 {
 997         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 998         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 999         struct extent_map *em;
1000         u64 len = PAGE_SIZE;
1001
1002         /*
1003          * hopefully we have this extent in the tree already, try without
1004          * the full extent lock
1005          */
1006         read_lock(&em_tree->lock);
1007         em = lookup_extent_mapping(em_tree, start, len);
1008         read_unlock(&em_tree->lock);
1009
1010         if (!em) {
1011                 struct extent_state *cached = NULL;
1012                 u64 end = start + len - 1;
1013
1014                 /* get the big lock and read metadata off disk */
1015                 lock_extent_bits(io_tree, start, end, &cached);
1016                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
1017                 unlock_extent_cached(io_tree, start, end, &cached, GFP_NOFS);
1018
1019                 if (IS_ERR(em))
1020                         return NULL;
1021         }
1022
1023         return em;
1024 }
1025
1026 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1027 {
1028         struct extent_map *next;
1029         bool ret = true;
1030
1031         /* this is the last extent */
1032         if (em->start + em->len >= i_size_read(inode))
1033                 return false;
1034
1035         next = defrag_lookup_extent(inode, em->start + em->len);
1036         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1037                 ret = false;
1038         else if ((em->block_start + em->block_len == next->block_start) &&
1039                  (em->block_len > SZ_128K && next->block_len > SZ_128K))
1040                 ret = false;
1041
1042         free_extent_map(next);
1043         return ret;
1044 }
1045
1046 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1047                                u64 *last_len, u64 *skip, u64 *defrag_end,
1048                                int compress)
1049 {
1050         struct extent_map *em;
1051         int ret = 1;
1052         bool next_mergeable = true;
1053         bool prev_mergeable = true;
1054
1055         /*
1056          * make sure that once we start defragging an extent, we keep on
1057          * defragging it
1058          */
1059         if (start < *defrag_end)
1060                 return 1;
1061
1062         *skip = 0;
1063
1064         em = defrag_lookup_extent(inode, start);
1065         if (!em)
1066                 return 0;
1067
1068         /* this will cover holes, and inline extents */
1069         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1070                 ret = 0;
1071                 goto out;
1072         }
1073
1074         if (!*defrag_end)
1075                 prev_mergeable = false;
1076
1077         next_mergeable = defrag_check_next_extent(inode, em);
1078         /*
1079          * we hit a real extent, if it is big or the next extent is not a
1080          * real extent, don't bother defragging it
1081          */
1082         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1083             (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1084                 ret = 0;
1085 out:
1086         /*
1087          * last_len ends up being a counter of how many bytes we've defragged.
1088          * every time we choose not to defrag an extent, we reset *last_len
1089          * so that the next tiny extent will force a defrag.
1090          *
1091          * The end result of this is that tiny extents before a single big
1092          * extent will force at least part of that big extent to be defragged.
1093          */
1094         if (ret) {
1095                 *defrag_end = extent_map_end(em);
1096         } else {
1097                 *last_len = 0;
1098                 *skip = extent_map_end(em);
1099                 *defrag_end = 0;
1100         }
1101
1102         free_extent_map(em);
1103         return ret;
1104 }
1105
1106 /*
1107  * it doesn't do much good to defrag one or two pages
1108  * at a time.  This pulls in a nice chunk of pages
1109  * to COW and defrag.
1110  *
1111  * It also makes sure the delalloc code has enough
1112  * dirty data to avoid making new small extents as part
1113  * of the defrag
1114  *
1115  * It's a good idea to start RA on this range
1116  * before calling this.
1117  */
1118 static int cluster_pages_for_defrag(struct inode *inode,
1119                                     struct page **pages,
1120                                     unsigned long start_index,
1121                                     unsigned long num_pages)
1122 {
1123         unsigned long file_end;
1124         u64 isize = i_size_read(inode);
1125         u64 page_start;
1126         u64 page_end;
1127         u64 page_cnt;
1128         int ret;
1129         int i;
1130         int i_done;
1131         struct btrfs_ordered_extent *ordered;
1132         struct extent_state *cached_state = NULL;
1133         struct extent_io_tree *tree;
1134         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1135
1136         file_end = (isize - 1) >> PAGE_SHIFT;
1137         if (!isize || start_index > file_end)
1138                 return 0;
1139
1140         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1141
1142         ret = btrfs_delalloc_reserve_space(inode,
1143                         start_index << PAGE_SHIFT,
1144                         page_cnt << PAGE_SHIFT);
1145         if (ret)
1146                 return ret;
1147         i_done = 0;
1148         tree = &BTRFS_I(inode)->io_tree;
1149
1150         /* step one, lock all the pages */
1151         for (i = 0; i < page_cnt; i++) {
1152                 struct page *page;
1153 again:
1154                 page = find_or_create_page(inode->i_mapping,
1155                                            start_index + i, mask);
1156                 if (!page)
1157                         break;
1158
1159                 page_start = page_offset(page);
1160                 page_end = page_start + PAGE_SIZE - 1;
1161                 while (1) {
1162                         lock_extent_bits(tree, page_start, page_end,
1163                                          &cached_state);
1164                         ordered = btrfs_lookup_ordered_extent(inode,
1165                                                               page_start);
1166                         unlock_extent_cached(tree, page_start, page_end,
1167                                              &cached_state, GFP_NOFS);
1168                         if (!ordered)
1169                                 break;
1170
1171                         unlock_page(page);
1172                         btrfs_start_ordered_extent(inode, ordered, 1);
1173                         btrfs_put_ordered_extent(ordered);
1174                         lock_page(page);
1175                         /*
1176                          * we unlocked the page above, so we need check if
1177                          * it was released or not.
1178                          */
1179                         if (page->mapping != inode->i_mapping) {
1180                                 unlock_page(page);
1181                                 put_page(page);
1182                                 goto again;
1183                         }
1184                 }
1185
1186                 if (!PageUptodate(page)) {
1187                         btrfs_readpage(NULL, page);
1188                         lock_page(page);
1189                         if (!PageUptodate(page)) {
1190                                 unlock_page(page);
1191                                 put_page(page);
1192                                 ret = -EIO;
1193                                 break;
1194                         }
1195                 }
1196
1197                 if (page->mapping != inode->i_mapping) {
1198                         unlock_page(page);
1199                         put_page(page);
1200                         goto again;
1201                 }
1202
1203                 pages[i] = page;
1204                 i_done++;
1205         }
1206         if (!i_done || ret)
1207                 goto out;
1208
1209         if (!(inode->i_sb->s_flags & MS_ACTIVE))
1210                 goto out;
1211
1212         /*
1213          * so now we have a nice long stream of locked
1214          * and up to date pages, lets wait on them
1215          */
1216         for (i = 0; i < i_done; i++)
1217                 wait_on_page_writeback(pages[i]);
1218
1219         page_start = page_offset(pages[0]);
1220         page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1221
1222         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1223                          page_start, page_end - 1, &cached_state);
1224         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1225                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1226                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1227                           &cached_state, GFP_NOFS);
1228
1229         if (i_done != page_cnt) {
1230                 spin_lock(&BTRFS_I(inode)->lock);
1231                 BTRFS_I(inode)->outstanding_extents++;
1232                 spin_unlock(&BTRFS_I(inode)->lock);
1233                 btrfs_delalloc_release_space(inode,
1234                                 start_index << PAGE_SHIFT,
1235                                 (page_cnt - i_done) << PAGE_SHIFT);
1236         }
1237
1238
1239         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1240                           &cached_state);
1241
1242         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1243                              page_start, page_end - 1, &cached_state,
1244                              GFP_NOFS);
1245
1246         for (i = 0; i < i_done; i++) {
1247                 clear_page_dirty_for_io(pages[i]);
1248                 ClearPageChecked(pages[i]);
1249                 set_page_extent_mapped(pages[i]);
1250                 set_page_dirty(pages[i]);
1251                 unlock_page(pages[i]);
1252                 put_page(pages[i]);
1253         }
1254         return i_done;
1255 out:
1256         for (i = 0; i < i_done; i++) {
1257                 unlock_page(pages[i]);
1258                 put_page(pages[i]);
1259         }
1260         btrfs_delalloc_release_space(inode,
1261                         start_index << PAGE_SHIFT,
1262                         page_cnt << PAGE_SHIFT);
1263         return ret;
1264
1265 }
1266
1267 int btrfs_defrag_file(struct inode *inode, struct file *file,
1268                       struct btrfs_ioctl_defrag_range_args *range,
1269                       u64 newer_than, unsigned long max_to_defrag)
1270 {
1271         struct btrfs_root *root = BTRFS_I(inode)->root;
1272         struct file_ra_state *ra = NULL;
1273         unsigned long last_index;
1274         u64 isize = i_size_read(inode);
1275         u64 last_len = 0;
1276         u64 skip = 0;
1277         u64 defrag_end = 0;
1278         u64 newer_off = range->start;
1279         unsigned long i;
1280         unsigned long ra_index = 0;
1281         int ret;
1282         int defrag_count = 0;
1283         int compress_type = BTRFS_COMPRESS_ZLIB;
1284         u32 extent_thresh = range->extent_thresh;
1285         unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1286         unsigned long cluster = max_cluster;
1287         u64 new_align = ~((u64)SZ_128K - 1);
1288         struct page **pages = NULL;
1289
1290         if (isize == 0)
1291                 return 0;
1292
1293         if (range->start >= isize)
1294                 return -EINVAL;
1295
1296         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1297                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1298                         return -EINVAL;
1299                 if (range->compress_type)
1300                         compress_type = range->compress_type;
1301         }
1302
1303         if (extent_thresh == 0)
1304                 extent_thresh = SZ_256K;
1305
1306         /*
1307          * if we were not given a file, allocate a readahead
1308          * context
1309          */
1310         if (!file) {
1311                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1312                 if (!ra)
1313                         return -ENOMEM;
1314                 file_ra_state_init(ra, inode->i_mapping);
1315         } else {
1316                 ra = &file->f_ra;
1317         }
1318
1319         pages = kmalloc_array(max_cluster, sizeof(struct page *),
1320                         GFP_NOFS);
1321         if (!pages) {
1322                 ret = -ENOMEM;
1323                 goto out_ra;
1324         }
1325
1326         /* find the last page to defrag */
1327         if (range->start + range->len > range->start) {
1328                 last_index = min_t(u64, isize - 1,
1329                          range->start + range->len - 1) >> PAGE_SHIFT;
1330         } else {
1331                 last_index = (isize - 1) >> PAGE_SHIFT;
1332         }
1333
1334         if (newer_than) {
1335                 ret = find_new_extents(root, inode, newer_than,
1336                                        &newer_off, SZ_64K);
1337                 if (!ret) {
1338                         range->start = newer_off;
1339                         /*
1340                          * we always align our defrag to help keep
1341                          * the extents in the file evenly spaced
1342                          */
1343                         i = (newer_off & new_align) >> PAGE_SHIFT;
1344                 } else
1345                         goto out_ra;
1346         } else {
1347                 i = range->start >> PAGE_SHIFT;
1348         }
1349         if (!max_to_defrag)
1350                 max_to_defrag = last_index - i + 1;
1351
1352         /*
1353          * make writeback starts from i, so the defrag range can be
1354          * written sequentially.
1355          */
1356         if (i < inode->i_mapping->writeback_index)
1357                 inode->i_mapping->writeback_index = i;
1358
1359         while (i <= last_index && defrag_count < max_to_defrag &&
1360                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1361                 /*
1362                  * make sure we stop running if someone unmounts
1363                  * the FS
1364                  */
1365                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1366                         break;
1367
1368                 if (btrfs_defrag_cancelled(root->fs_info)) {
1369                         btrfs_debug(root->fs_info, "defrag_file cancelled");
1370                         ret = -EAGAIN;
1371                         break;
1372                 }
1373
1374                 if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1375                                          extent_thresh, &last_len, &skip,
1376                                          &defrag_end, range->flags &
1377                                          BTRFS_DEFRAG_RANGE_COMPRESS)) {
1378                         unsigned long next;
1379                         /*
1380                          * the should_defrag function tells us how much to skip
1381                          * bump our counter by the suggested amount
1382                          */
1383                         next = DIV_ROUND_UP(skip, PAGE_SIZE);
1384                         i = max(i + 1, next);
1385                         continue;
1386                 }
1387
1388                 if (!newer_than) {
1389                         cluster = (PAGE_ALIGN(defrag_end) >>
1390                                    PAGE_SHIFT) - i;
1391                         cluster = min(cluster, max_cluster);
1392                 } else {
1393                         cluster = max_cluster;
1394                 }
1395
1396                 if (i + cluster > ra_index) {
1397                         ra_index = max(i, ra_index);
1398                         btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1399                                        cluster);
1400                         ra_index += cluster;
1401                 }
1402
1403                 inode_lock(inode);
1404                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1405                         BTRFS_I(inode)->force_compress = compress_type;
1406                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1407                 if (ret < 0) {
1408                         inode_unlock(inode);
1409                         goto out_ra;
1410                 }
1411
1412                 defrag_count += ret;
1413                 balance_dirty_pages_ratelimited(inode->i_mapping);
1414                 inode_unlock(inode);
1415
1416                 if (newer_than) {
1417                         if (newer_off == (u64)-1)
1418                                 break;
1419
1420                         if (ret > 0)
1421                                 i += ret;
1422
1423                         newer_off = max(newer_off + 1,
1424                                         (u64)i << PAGE_SHIFT);
1425
1426                         ret = find_new_extents(root, inode, newer_than,
1427                                                &newer_off, SZ_64K);
1428                         if (!ret) {
1429                                 range->start = newer_off;
1430                                 i = (newer_off & new_align) >> PAGE_SHIFT;
1431                         } else {
1432                                 break;
1433                         }
1434                 } else {
1435                         if (ret > 0) {
1436                                 i += ret;
1437                                 last_len += ret << PAGE_SHIFT;
1438                         } else {
1439                                 i++;
1440                                 last_len = 0;
1441                         }
1442                 }
1443         }
1444
1445         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1446                 filemap_flush(inode->i_mapping);
1447                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1448                              &BTRFS_I(inode)->runtime_flags))
1449                         filemap_flush(inode->i_mapping);
1450         }
1451
1452         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1453                 /* the filemap_flush will queue IO into the worker threads, but
1454                  * we have to make sure the IO is actually started and that
1455                  * ordered extents get created before we return
1456                  */
1457                 atomic_inc(&root->fs_info->async_submit_draining);
1458                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1459                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1460                         wait_event(root->fs_info->async_submit_wait,
1461                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1462                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1463                 }
1464                 atomic_dec(&root->fs_info->async_submit_draining);
1465         }
1466
1467         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1468                 btrfs_set_fs_incompat(root->fs_info, COMPRESS_LZO);
1469         }
1470
1471         ret = defrag_count;
1472
1473 out_ra:
1474         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1475                 inode_lock(inode);
1476                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1477                 inode_unlock(inode);
1478         }
1479         if (!file)
1480                 kfree(ra);
1481         kfree(pages);
1482         return ret;
1483 }
1484
1485 static noinline int btrfs_ioctl_resize(struct file *file,
1486                                         void __user *arg)
1487 {
1488         u64 new_size;
1489         u64 old_size;
1490         u64 devid = 1;
1491         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
1492         struct btrfs_ioctl_vol_args *vol_args;
1493         struct btrfs_trans_handle *trans;
1494         struct btrfs_device *device = NULL;
1495         char *sizestr;
1496         char *retptr;
1497         char *devstr = NULL;
1498         int ret = 0;
1499         int mod = 0;
1500
1501         if (!capable(CAP_SYS_ADMIN))
1502                 return -EPERM;
1503
1504         ret = mnt_want_write_file(file);
1505         if (ret)
1506                 return ret;
1507
1508         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
1509                         1)) {
1510                 mnt_drop_write_file(file);
1511                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1512         }
1513
1514         mutex_lock(&root->fs_info->volume_mutex);
1515         vol_args = memdup_user(arg, sizeof(*vol_args));
1516         if (IS_ERR(vol_args)) {
1517                 ret = PTR_ERR(vol_args);
1518                 goto out;
1519         }
1520
1521         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1522
1523         sizestr = vol_args->name;
1524         devstr = strchr(sizestr, ':');
1525         if (devstr) {
1526                 sizestr = devstr + 1;
1527                 *devstr = '\0';
1528                 devstr = vol_args->name;
1529                 ret = kstrtoull(devstr, 10, &devid);
1530                 if (ret)
1531                         goto out_free;
1532                 if (!devid) {
1533                         ret = -EINVAL;
1534                         goto out_free;
1535                 }
1536                 btrfs_info(root->fs_info, "resizing devid %llu", devid);
1537         }
1538
1539         device = btrfs_find_device(root->fs_info, devid, NULL, NULL);
1540         if (!device) {
1541                 btrfs_info(root->fs_info, "resizer unable to find device %llu",
1542                        devid);
1543                 ret = -ENODEV;
1544                 goto out_free;
1545         }
1546
1547         if (!device->writeable) {
1548                 btrfs_info(root->fs_info,
1549                            "resizer unable to apply on readonly device %llu",
1550                        devid);
1551                 ret = -EPERM;
1552                 goto out_free;
1553         }
1554
1555         if (!strcmp(sizestr, "max"))
1556                 new_size = device->bdev->bd_inode->i_size;
1557         else {
1558                 if (sizestr[0] == '-') {
1559                         mod = -1;
1560                         sizestr++;
1561                 } else if (sizestr[0] == '+') {
1562                         mod = 1;
1563                         sizestr++;
1564                 }
1565                 new_size = memparse(sizestr, &retptr);
1566                 if (*retptr != '\0' || new_size == 0) {
1567                         ret = -EINVAL;
1568                         goto out_free;
1569                 }
1570         }
1571
1572         if (device->is_tgtdev_for_dev_replace) {
1573                 ret = -EPERM;
1574                 goto out_free;
1575         }
1576
1577         old_size = btrfs_device_get_total_bytes(device);
1578
1579         if (mod < 0) {
1580                 if (new_size > old_size) {
1581                         ret = -EINVAL;
1582                         goto out_free;
1583                 }
1584                 new_size = old_size - new_size;
1585         } else if (mod > 0) {
1586                 if (new_size > ULLONG_MAX - old_size) {
1587                         ret = -ERANGE;
1588                         goto out_free;
1589                 }
1590                 new_size = old_size + new_size;
1591         }
1592
1593         if (new_size < SZ_256M) {
1594                 ret = -EINVAL;
1595                 goto out_free;
1596         }
1597         if (new_size > device->bdev->bd_inode->i_size) {
1598                 ret = -EFBIG;
1599                 goto out_free;
1600         }
1601
1602         new_size = div_u64(new_size, root->sectorsize);
1603         new_size *= root->sectorsize;
1604
1605         btrfs_info_in_rcu(root->fs_info, "new size for %s is %llu",
1606                       rcu_str_deref(device->name), new_size);
1607
1608         if (new_size > old_size) {
1609                 trans = btrfs_start_transaction(root, 0);
1610                 if (IS_ERR(trans)) {
1611                         ret = PTR_ERR(trans);
1612                         goto out_free;
1613                 }
1614                 ret = btrfs_grow_device(trans, device, new_size);
1615                 btrfs_commit_transaction(trans, root);
1616         } else if (new_size < old_size) {
1617                 ret = btrfs_shrink_device(device, new_size);
1618         } /* equal, nothing need to do */
1619
1620 out_free:
1621         kfree(vol_args);
1622 out:
1623         mutex_unlock(&root->fs_info->volume_mutex);
1624         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
1625         mnt_drop_write_file(file);
1626         return ret;
1627 }
1628
1629 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1630                                 char *name, unsigned long fd, int subvol,
1631                                 u64 *transid, bool readonly,
1632                                 struct btrfs_qgroup_inherit *inherit)
1633 {
1634         int namelen;
1635         int ret = 0;
1636
1637         ret = mnt_want_write_file(file);
1638         if (ret)
1639                 goto out;
1640
1641         namelen = strlen(name);
1642         if (strchr(name, '/')) {
1643                 ret = -EINVAL;
1644                 goto out_drop_write;
1645         }
1646
1647         if (name[0] == '.' &&
1648            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1649                 ret = -EEXIST;
1650                 goto out_drop_write;
1651         }
1652
1653         if (subvol) {
1654                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1655                                      NULL, transid, readonly, inherit);
1656         } else {
1657                 struct fd src = fdget(fd);
1658                 struct inode *src_inode;
1659                 if (!src.file) {
1660                         ret = -EINVAL;
1661                         goto out_drop_write;
1662                 }
1663
1664                 src_inode = file_inode(src.file);
1665                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1666                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1667                                    "Snapshot src from another FS");
1668                         ret = -EXDEV;
1669                 } else if (!inode_owner_or_capable(src_inode)) {
1670                         /*
1671                          * Subvolume creation is not restricted, but snapshots
1672                          * are limited to own subvolumes only
1673                          */
1674                         ret = -EPERM;
1675                 } else {
1676                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1677                                              BTRFS_I(src_inode)->root,
1678                                              transid, readonly, inherit);
1679                 }
1680                 fdput(src);
1681         }
1682 out_drop_write:
1683         mnt_drop_write_file(file);
1684 out:
1685         return ret;
1686 }
1687
1688 static noinline int btrfs_ioctl_snap_create(struct file *file,
1689                                             void __user *arg, int subvol)
1690 {
1691         struct btrfs_ioctl_vol_args *vol_args;
1692         int ret;
1693
1694         vol_args = memdup_user(arg, sizeof(*vol_args));
1695         if (IS_ERR(vol_args))
1696                 return PTR_ERR(vol_args);
1697         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1698
1699         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1700                                               vol_args->fd, subvol,
1701                                               NULL, false, NULL);
1702
1703         kfree(vol_args);
1704         return ret;
1705 }
1706
1707 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1708                                                void __user *arg, int subvol)
1709 {
1710         struct btrfs_ioctl_vol_args_v2 *vol_args;
1711         int ret;
1712         u64 transid = 0;
1713         u64 *ptr = NULL;
1714         bool readonly = false;
1715         struct btrfs_qgroup_inherit *inherit = NULL;
1716
1717         vol_args = memdup_user(arg, sizeof(*vol_args));
1718         if (IS_ERR(vol_args))
1719                 return PTR_ERR(vol_args);
1720         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1721
1722         if (vol_args->flags &
1723             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1724               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1725                 ret = -EOPNOTSUPP;
1726                 goto free_args;
1727         }
1728
1729         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1730                 ptr = &transid;
1731         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1732                 readonly = true;
1733         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1734                 if (vol_args->size > PAGE_SIZE) {
1735                         ret = -EINVAL;
1736                         goto free_args;
1737                 }
1738                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1739                 if (IS_ERR(inherit)) {
1740                         ret = PTR_ERR(inherit);
1741                         goto free_args;
1742                 }
1743         }
1744
1745         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1746                                               vol_args->fd, subvol, ptr,
1747                                               readonly, inherit);
1748         if (ret)
1749                 goto free_inherit;
1750
1751         if (ptr && copy_to_user(arg +
1752                                 offsetof(struct btrfs_ioctl_vol_args_v2,
1753                                         transid),
1754                                 ptr, sizeof(*ptr)))
1755                 ret = -EFAULT;
1756
1757 free_inherit:
1758         kfree(inherit);
1759 free_args:
1760         kfree(vol_args);
1761         return ret;
1762 }
1763
1764 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1765                                                 void __user *arg)
1766 {
1767         struct inode *inode = file_inode(file);
1768         struct btrfs_root *root = BTRFS_I(inode)->root;
1769         int ret = 0;
1770         u64 flags = 0;
1771
1772         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1773                 return -EINVAL;
1774
1775         down_read(&root->fs_info->subvol_sem);
1776         if (btrfs_root_readonly(root))
1777                 flags |= BTRFS_SUBVOL_RDONLY;
1778         up_read(&root->fs_info->subvol_sem);
1779
1780         if (copy_to_user(arg, &flags, sizeof(flags)))
1781                 ret = -EFAULT;
1782
1783         return ret;
1784 }
1785
1786 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1787                                               void __user *arg)
1788 {
1789         struct inode *inode = file_inode(file);
1790         struct btrfs_root *root = BTRFS_I(inode)->root;
1791         struct btrfs_trans_handle *trans;
1792         u64 root_flags;
1793         u64 flags;
1794         int ret = 0;
1795
1796         if (!inode_owner_or_capable(inode))
1797                 return -EPERM;
1798
1799         ret = mnt_want_write_file(file);
1800         if (ret)
1801                 goto out;
1802
1803         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1804                 ret = -EINVAL;
1805                 goto out_drop_write;
1806         }
1807
1808         if (copy_from_user(&flags, arg, sizeof(flags))) {
1809                 ret = -EFAULT;
1810                 goto out_drop_write;
1811         }
1812
1813         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1814                 ret = -EINVAL;
1815                 goto out_drop_write;
1816         }
1817
1818         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1819                 ret = -EOPNOTSUPP;
1820                 goto out_drop_write;
1821         }
1822
1823         down_write(&root->fs_info->subvol_sem);
1824
1825         /* nothing to do */
1826         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1827                 goto out_drop_sem;
1828
1829         root_flags = btrfs_root_flags(&root->root_item);
1830         if (flags & BTRFS_SUBVOL_RDONLY) {
1831                 btrfs_set_root_flags(&root->root_item,
1832                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1833         } else {
1834                 /*
1835                  * Block RO -> RW transition if this subvolume is involved in
1836                  * send
1837                  */
1838                 spin_lock(&root->root_item_lock);
1839                 if (root->send_in_progress == 0) {
1840                         btrfs_set_root_flags(&root->root_item,
1841                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1842                         spin_unlock(&root->root_item_lock);
1843                 } else {
1844                         spin_unlock(&root->root_item_lock);
1845                         btrfs_warn(root->fs_info,
1846                         "Attempt to set subvolume %llu read-write during send",
1847                                         root->root_key.objectid);
1848                         ret = -EPERM;
1849                         goto out_drop_sem;
1850                 }
1851         }
1852
1853         trans = btrfs_start_transaction(root, 1);
1854         if (IS_ERR(trans)) {
1855                 ret = PTR_ERR(trans);
1856                 goto out_reset;
1857         }
1858
1859         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1860                                 &root->root_key, &root->root_item);
1861
1862         btrfs_commit_transaction(trans, root);
1863 out_reset:
1864         if (ret)
1865                 btrfs_set_root_flags(&root->root_item, root_flags);
1866 out_drop_sem:
1867         up_write(&root->fs_info->subvol_sem);
1868 out_drop_write:
1869         mnt_drop_write_file(file);
1870 out:
1871         return ret;
1872 }
1873
1874 /*
1875  * helper to check if the subvolume references other subvolumes
1876  */
1877 static noinline int may_destroy_subvol(struct btrfs_root *root)
1878 {
1879         struct btrfs_path *path;
1880         struct btrfs_dir_item *di;
1881         struct btrfs_key key;
1882         u64 dir_id;
1883         int ret;
1884
1885         path = btrfs_alloc_path();
1886         if (!path)
1887                 return -ENOMEM;
1888
1889         /* Make sure this root isn't set as the default subvol */
1890         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
1891         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root, path,
1892                                    dir_id, "default", 7, 0);
1893         if (di && !IS_ERR(di)) {
1894                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1895                 if (key.objectid == root->root_key.objectid) {
1896                         ret = -EPERM;
1897                         btrfs_err(root->fs_info, "deleting default subvolume "
1898                                   "%llu is not allowed", key.objectid);
1899                         goto out;
1900                 }
1901                 btrfs_release_path(path);
1902         }
1903
1904         key.objectid = root->root_key.objectid;
1905         key.type = BTRFS_ROOT_REF_KEY;
1906         key.offset = (u64)-1;
1907
1908         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1909                                 &key, path, 0, 0);
1910         if (ret < 0)
1911                 goto out;
1912         BUG_ON(ret == 0);
1913
1914         ret = 0;
1915         if (path->slots[0] > 0) {
1916                 path->slots[0]--;
1917                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1918                 if (key.objectid == root->root_key.objectid &&
1919                     key.type == BTRFS_ROOT_REF_KEY)
1920                         ret = -ENOTEMPTY;
1921         }
1922 out:
1923         btrfs_free_path(path);
1924         return ret;
1925 }
1926
1927 static noinline int key_in_sk(struct btrfs_key *key,
1928                               struct btrfs_ioctl_search_key *sk)
1929 {
1930         struct btrfs_key test;
1931         int ret;
1932
1933         test.objectid = sk->min_objectid;
1934         test.type = sk->min_type;
1935         test.offset = sk->min_offset;
1936
1937         ret = btrfs_comp_cpu_keys(key, &test);
1938         if (ret < 0)
1939                 return 0;
1940
1941         test.objectid = sk->max_objectid;
1942         test.type = sk->max_type;
1943         test.offset = sk->max_offset;
1944
1945         ret = btrfs_comp_cpu_keys(key, &test);
1946         if (ret > 0)
1947                 return 0;
1948         return 1;
1949 }
1950
1951 static noinline int copy_to_sk(struct btrfs_path *path,
1952                                struct btrfs_key *key,
1953                                struct btrfs_ioctl_search_key *sk,
1954                                size_t *buf_size,
1955                                char __user *ubuf,
1956                                unsigned long *sk_offset,
1957                                int *num_found)
1958 {
1959         u64 found_transid;
1960         struct extent_buffer *leaf;
1961         struct btrfs_ioctl_search_header sh;
1962         struct btrfs_key test;
1963         unsigned long item_off;
1964         unsigned long item_len;
1965         int nritems;
1966         int i;
1967         int slot;
1968         int ret = 0;
1969
1970         leaf = path->nodes[0];
1971         slot = path->slots[0];
1972         nritems = btrfs_header_nritems(leaf);
1973
1974         if (btrfs_header_generation(leaf) > sk->max_transid) {
1975                 i = nritems;
1976                 goto advance_key;
1977         }
1978         found_transid = btrfs_header_generation(leaf);
1979
1980         for (i = slot; i < nritems; i++) {
1981                 item_off = btrfs_item_ptr_offset(leaf, i);
1982                 item_len = btrfs_item_size_nr(leaf, i);
1983
1984                 btrfs_item_key_to_cpu(leaf, key, i);
1985                 if (!key_in_sk(key, sk))
1986                         continue;
1987
1988                 if (sizeof(sh) + item_len > *buf_size) {
1989                         if (*num_found) {
1990                                 ret = 1;
1991                                 goto out;
1992                         }
1993
1994                         /*
1995                          * return one empty item back for v1, which does not
1996                          * handle -EOVERFLOW
1997                          */
1998
1999                         *buf_size = sizeof(sh) + item_len;
2000                         item_len = 0;
2001                         ret = -EOVERFLOW;
2002                 }
2003
2004                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
2005                         ret = 1;
2006                         goto out;
2007                 }
2008
2009                 sh.objectid = key->objectid;
2010                 sh.offset = key->offset;
2011                 sh.type = key->type;
2012                 sh.len = item_len;
2013                 sh.transid = found_transid;
2014
2015                 /* copy search result header */
2016                 if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
2017                         ret = -EFAULT;
2018                         goto out;
2019                 }
2020
2021                 *sk_offset += sizeof(sh);
2022
2023                 if (item_len) {
2024                         char __user *up = ubuf + *sk_offset;
2025                         /* copy the item */
2026                         if (read_extent_buffer_to_user(leaf, up,
2027                                                        item_off, item_len)) {
2028                                 ret = -EFAULT;
2029                                 goto out;
2030                         }
2031
2032                         *sk_offset += item_len;
2033                 }
2034                 (*num_found)++;
2035
2036                 if (ret) /* -EOVERFLOW from above */
2037                         goto out;
2038
2039                 if (*num_found >= sk->nr_items) {
2040                         ret = 1;
2041                         goto out;
2042                 }
2043         }
2044 advance_key:
2045         ret = 0;
2046         test.objectid = sk->max_objectid;
2047         test.type = sk->max_type;
2048         test.offset = sk->max_offset;
2049         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2050                 ret = 1;
2051         else if (key->offset < (u64)-1)
2052                 key->offset++;
2053         else if (key->type < (u8)-1) {
2054                 key->offset = 0;
2055                 key->type++;
2056         } else if (key->objectid < (u64)-1) {
2057                 key->offset = 0;
2058                 key->type = 0;
2059                 key->objectid++;
2060         } else
2061                 ret = 1;
2062 out:
2063         /*
2064          *  0: all items from this leaf copied, continue with next
2065          *  1: * more items can be copied, but unused buffer is too small
2066          *     * all items were found
2067          *     Either way, it will stops the loop which iterates to the next
2068          *     leaf
2069          *  -EOVERFLOW: item was to large for buffer
2070          *  -EFAULT: could not copy extent buffer back to userspace
2071          */
2072         return ret;
2073 }
2074
2075 static noinline int search_ioctl(struct inode *inode,
2076                                  struct btrfs_ioctl_search_key *sk,
2077                                  size_t *buf_size,
2078                                  char __user *ubuf)
2079 {
2080         struct btrfs_root *root;
2081         struct btrfs_key key;
2082         struct btrfs_path *path;
2083         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
2084         int ret;
2085         int num_found = 0;
2086         unsigned long sk_offset = 0;
2087
2088         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2089                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2090                 return -EOVERFLOW;
2091         }
2092
2093         path = btrfs_alloc_path();
2094         if (!path)
2095                 return -ENOMEM;
2096
2097         if (sk->tree_id == 0) {
2098                 /* search the root of the inode that was passed */
2099                 root = BTRFS_I(inode)->root;
2100         } else {
2101                 key.objectid = sk->tree_id;
2102                 key.type = BTRFS_ROOT_ITEM_KEY;
2103                 key.offset = (u64)-1;
2104                 root = btrfs_read_fs_root_no_name(info, &key);
2105                 if (IS_ERR(root)) {
2106                         btrfs_free_path(path);
2107                         return -ENOENT;
2108                 }
2109         }
2110
2111         key.objectid = sk->min_objectid;
2112         key.type = sk->min_type;
2113         key.offset = sk->min_offset;
2114
2115         while (1) {
2116                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2117                 if (ret != 0) {
2118                         if (ret > 0)
2119                                 ret = 0;
2120                         goto err;
2121                 }
2122                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2123                                  &sk_offset, &num_found);
2124                 btrfs_release_path(path);
2125                 if (ret)
2126                         break;
2127
2128         }
2129         if (ret > 0)
2130                 ret = 0;
2131 err:
2132         sk->nr_items = num_found;
2133         btrfs_free_path(path);
2134         return ret;
2135 }
2136
2137 static noinline int btrfs_ioctl_tree_search(struct file *file,
2138                                            void __user *argp)
2139 {
2140         struct btrfs_ioctl_search_args __user *uargs;
2141         struct btrfs_ioctl_search_key sk;
2142         struct inode *inode;
2143         int ret;
2144         size_t buf_size;
2145
2146         if (!capable(CAP_SYS_ADMIN))
2147                 return -EPERM;
2148
2149         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2150
2151         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2152                 return -EFAULT;
2153
2154         buf_size = sizeof(uargs->buf);
2155
2156         inode = file_inode(file);
2157         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2158
2159         /*
2160          * In the origin implementation an overflow is handled by returning a
2161          * search header with a len of zero, so reset ret.
2162          */
2163         if (ret == -EOVERFLOW)
2164                 ret = 0;
2165
2166         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2167                 ret = -EFAULT;
2168         return ret;
2169 }
2170
2171 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2172                                                void __user *argp)
2173 {
2174         struct btrfs_ioctl_search_args_v2 __user *uarg;
2175         struct btrfs_ioctl_search_args_v2 args;
2176         struct inode *inode;
2177         int ret;
2178         size_t buf_size;
2179         const size_t buf_limit = SZ_16M;
2180
2181         if (!capable(CAP_SYS_ADMIN))
2182                 return -EPERM;
2183
2184         /* copy search header and buffer size */
2185         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2186         if (copy_from_user(&args, uarg, sizeof(args)))
2187                 return -EFAULT;
2188
2189         buf_size = args.buf_size;
2190
2191         if (buf_size < sizeof(struct btrfs_ioctl_search_header))
2192                 return -EOVERFLOW;
2193
2194         /* limit result size to 16MB */
2195         if (buf_size > buf_limit)
2196                 buf_size = buf_limit;
2197
2198         inode = file_inode(file);
2199         ret = search_ioctl(inode, &args.key, &buf_size,
2200                            (char *)(&uarg->buf[0]));
2201         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2202                 ret = -EFAULT;
2203         else if (ret == -EOVERFLOW &&
2204                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2205                 ret = -EFAULT;
2206
2207         return ret;
2208 }
2209
2210 /*
2211  * Search INODE_REFs to identify path name of 'dirid' directory
2212  * in a 'tree_id' tree. and sets path name to 'name'.
2213  */
2214 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2215                                 u64 tree_id, u64 dirid, char *name)
2216 {
2217         struct btrfs_root *root;
2218         struct btrfs_key key;
2219         char *ptr;
2220         int ret = -1;
2221         int slot;
2222         int len;
2223         int total_len = 0;
2224         struct btrfs_inode_ref *iref;
2225         struct extent_buffer *l;
2226         struct btrfs_path *path;
2227
2228         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2229                 name[0]='\0';
2230                 return 0;
2231         }
2232
2233         path = btrfs_alloc_path();
2234         if (!path)
2235                 return -ENOMEM;
2236
2237         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
2238
2239         key.objectid = tree_id;
2240         key.type = BTRFS_ROOT_ITEM_KEY;
2241         key.offset = (u64)-1;
2242         root = btrfs_read_fs_root_no_name(info, &key);
2243         if (IS_ERR(root)) {
2244                 btrfs_err(info, "could not find root %llu", tree_id);
2245                 ret = -ENOENT;
2246                 goto out;
2247         }
2248
2249         key.objectid = dirid;
2250         key.type = BTRFS_INODE_REF_KEY;
2251         key.offset = (u64)-1;
2252
2253         while (1) {
2254                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2255                 if (ret < 0)
2256                         goto out;
2257                 else if (ret > 0) {
2258                         ret = btrfs_previous_item(root, path, dirid,
2259                                                   BTRFS_INODE_REF_KEY);
2260                         if (ret < 0)
2261                                 goto out;
2262                         else if (ret > 0) {
2263                                 ret = -ENOENT;
2264                                 goto out;
2265                         }
2266                 }
2267
2268                 l = path->nodes[0];
2269                 slot = path->slots[0];
2270                 btrfs_item_key_to_cpu(l, &key, slot);
2271
2272                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2273                 len = btrfs_inode_ref_name_len(l, iref);
2274                 ptr -= len + 1;
2275                 total_len += len + 1;
2276                 if (ptr < name) {
2277                         ret = -ENAMETOOLONG;
2278                         goto out;
2279                 }
2280
2281                 *(ptr + len) = '/';
2282                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2283
2284                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2285                         break;
2286
2287                 btrfs_release_path(path);
2288                 key.objectid = key.offset;
2289                 key.offset = (u64)-1;
2290                 dirid = key.objectid;
2291         }
2292         memmove(name, ptr, total_len);
2293         name[total_len] = '\0';
2294         ret = 0;
2295 out:
2296         btrfs_free_path(path);
2297         return ret;
2298 }
2299
2300 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2301                                            void __user *argp)
2302 {
2303          struct btrfs_ioctl_ino_lookup_args *args;
2304          struct inode *inode;
2305         int ret = 0;
2306
2307         args = memdup_user(argp, sizeof(*args));
2308         if (IS_ERR(args))
2309                 return PTR_ERR(args);
2310
2311         inode = file_inode(file);
2312
2313         /*
2314          * Unprivileged query to obtain the containing subvolume root id. The
2315          * path is reset so it's consistent with btrfs_search_path_in_tree.
2316          */
2317         if (args->treeid == 0)
2318                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2319
2320         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2321                 args->name[0] = 0;
2322                 goto out;
2323         }
2324
2325         if (!capable(CAP_SYS_ADMIN)) {
2326                 ret = -EPERM;
2327                 goto out;
2328         }
2329
2330         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2331                                         args->treeid, args->objectid,
2332                                         args->name);
2333
2334 out:
2335         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2336                 ret = -EFAULT;
2337
2338         kfree(args);
2339         return ret;
2340 }
2341
2342 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2343                                              void __user *arg)
2344 {
2345         struct dentry *parent = file->f_path.dentry;
2346         struct dentry *dentry;
2347         struct inode *dir = d_inode(parent);
2348         struct inode *inode;
2349         struct btrfs_root *root = BTRFS_I(dir)->root;
2350         struct btrfs_root *dest = NULL;
2351         struct btrfs_ioctl_vol_args *vol_args;
2352         struct btrfs_trans_handle *trans;
2353         struct btrfs_block_rsv block_rsv;
2354         u64 root_flags;
2355         u64 qgroup_reserved;
2356         int namelen;
2357         int ret;
2358         int err = 0;
2359
2360         vol_args = memdup_user(arg, sizeof(*vol_args));
2361         if (IS_ERR(vol_args))
2362                 return PTR_ERR(vol_args);
2363
2364         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2365         namelen = strlen(vol_args->name);
2366         if (strchr(vol_args->name, '/') ||
2367             strncmp(vol_args->name, "..", namelen) == 0) {
2368                 err = -EINVAL;
2369                 goto out;
2370         }
2371
2372         err = mnt_want_write_file(file);
2373         if (err)
2374                 goto out;
2375
2376
2377         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
2378         if (err == -EINTR)
2379                 goto out_drop_write;
2380         dentry = lookup_one_len(vol_args->name, parent, namelen);
2381         if (IS_ERR(dentry)) {
2382                 err = PTR_ERR(dentry);
2383                 goto out_unlock_dir;
2384         }
2385
2386         if (d_really_is_negative(dentry)) {
2387                 err = -ENOENT;
2388                 goto out_dput;
2389         }
2390
2391         inode = d_inode(dentry);
2392         dest = BTRFS_I(inode)->root;
2393         if (!capable(CAP_SYS_ADMIN)) {
2394                 /*
2395                  * Regular user.  Only allow this with a special mount
2396                  * option, when the user has write+exec access to the
2397                  * subvol root, and when rmdir(2) would have been
2398                  * allowed.
2399                  *
2400                  * Note that this is _not_ check that the subvol is
2401                  * empty or doesn't contain data that we wouldn't
2402                  * otherwise be able to delete.
2403                  *
2404                  * Users who want to delete empty subvols should try
2405                  * rmdir(2).
2406                  */
2407                 err = -EPERM;
2408                 if (!btrfs_test_opt(root->fs_info, USER_SUBVOL_RM_ALLOWED))
2409                         goto out_dput;
2410
2411                 /*
2412                  * Do not allow deletion if the parent dir is the same
2413                  * as the dir to be deleted.  That means the ioctl
2414                  * must be called on the dentry referencing the root
2415                  * of the subvol, not a random directory contained
2416                  * within it.
2417                  */
2418                 err = -EINVAL;
2419                 if (root == dest)
2420                         goto out_dput;
2421
2422                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2423                 if (err)
2424                         goto out_dput;
2425         }
2426
2427         /* check if subvolume may be deleted by a user */
2428         err = btrfs_may_delete(dir, dentry, 1);
2429         if (err)
2430                 goto out_dput;
2431
2432         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
2433                 err = -EINVAL;
2434                 goto out_dput;
2435         }
2436
2437         inode_lock(inode);
2438
2439         /*
2440          * Don't allow to delete a subvolume with send in progress. This is
2441          * inside the i_mutex so the error handling that has to drop the bit
2442          * again is not run concurrently.
2443          */
2444         spin_lock(&dest->root_item_lock);
2445         root_flags = btrfs_root_flags(&dest->root_item);
2446         if (dest->send_in_progress == 0) {
2447                 btrfs_set_root_flags(&dest->root_item,
2448                                 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
2449                 spin_unlock(&dest->root_item_lock);
2450         } else {
2451                 spin_unlock(&dest->root_item_lock);
2452                 btrfs_warn(root->fs_info,
2453                         "Attempt to delete subvolume %llu during send",
2454                         dest->root_key.objectid);
2455                 err = -EPERM;
2456                 goto out_unlock_inode;
2457         }
2458
2459         down_write(&root->fs_info->subvol_sem);
2460
2461         err = may_destroy_subvol(dest);
2462         if (err)
2463                 goto out_up_write;
2464
2465         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2466         /*
2467          * One for dir inode, two for dir entries, two for root
2468          * ref/backref.
2469          */
2470         err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2471                                                5, &qgroup_reserved, true);
2472         if (err)
2473                 goto out_up_write;
2474
2475         trans = btrfs_start_transaction(root, 0);
2476         if (IS_ERR(trans)) {
2477                 err = PTR_ERR(trans);
2478                 goto out_release;
2479         }
2480         trans->block_rsv = &block_rsv;
2481         trans->bytes_reserved = block_rsv.size;
2482
2483         btrfs_record_snapshot_destroy(trans, dir);
2484
2485         ret = btrfs_unlink_subvol(trans, root, dir,
2486                                 dest->root_key.objectid,
2487                                 dentry->d_name.name,
2488                                 dentry->d_name.len);
2489         if (ret) {
2490                 err = ret;
2491                 btrfs_abort_transaction(trans, ret);
2492                 goto out_end_trans;
2493         }
2494
2495         btrfs_record_root_in_trans(trans, dest);
2496
2497         memset(&dest->root_item.drop_progress, 0,
2498                 sizeof(dest->root_item.drop_progress));
2499         dest->root_item.drop_level = 0;
2500         btrfs_set_root_refs(&dest->root_item, 0);
2501
2502         if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
2503                 ret = btrfs_insert_orphan_item(trans,
2504                                         root->fs_info->tree_root,
2505                                         dest->root_key.objectid);
2506                 if (ret) {
2507                         btrfs_abort_transaction(trans, ret);
2508                         err = ret;
2509                         goto out_end_trans;
2510                 }
2511         }
2512
2513         ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2514                                   dest->root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
2515                                   dest->root_key.objectid);
2516         if (ret && ret != -ENOENT) {
2517                 btrfs_abort_transaction(trans, ret);
2518                 err = ret;
2519                 goto out_end_trans;
2520         }
2521         if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2522                 ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2523                                           dest->root_item.received_uuid,
2524                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2525                                           dest->root_key.objectid);
2526                 if (ret && ret != -ENOENT) {
2527                         btrfs_abort_transaction(trans, ret);
2528                         err = ret;
2529                         goto out_end_trans;
2530                 }
2531         }
2532
2533 out_end_trans:
2534         trans->block_rsv = NULL;
2535         trans->bytes_reserved = 0;
2536         ret = btrfs_end_transaction(trans, root);
2537         if (ret && !err)
2538                 err = ret;
2539         inode->i_flags |= S_DEAD;
2540 out_release:
2541         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
2542 out_up_write:
2543         up_write(&root->fs_info->subvol_sem);
2544         if (err) {
2545                 spin_lock(&dest->root_item_lock);
2546                 root_flags = btrfs_root_flags(&dest->root_item);
2547                 btrfs_set_root_flags(&dest->root_item,
2548                                 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
2549                 spin_unlock(&dest->root_item_lock);
2550         }
2551 out_unlock_inode:
2552         inode_unlock(inode);
2553         if (!err) {
2554                 d_invalidate(dentry);
2555                 btrfs_invalidate_inodes(dest);
2556                 d_delete(dentry);
2557                 ASSERT(dest->send_in_progress == 0);
2558
2559                 /* the last ref */
2560                 if (dest->ino_cache_inode) {
2561                         iput(dest->ino_cache_inode);
2562                         dest->ino_cache_inode = NULL;
2563                 }
2564         }
2565 out_dput:
2566         dput(dentry);
2567 out_unlock_dir:
2568         inode_unlock(dir);
2569 out_drop_write:
2570         mnt_drop_write_file(file);
2571 out:
2572         kfree(vol_args);
2573         return err;
2574 }
2575
2576 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2577 {
2578         struct inode *inode = file_inode(file);
2579         struct btrfs_root *root = BTRFS_I(inode)->root;
2580         struct btrfs_ioctl_defrag_range_args *range;
2581         int ret;
2582
2583         ret = mnt_want_write_file(file);
2584         if (ret)
2585                 return ret;
2586
2587         if (btrfs_root_readonly(root)) {
2588                 ret = -EROFS;
2589                 goto out;
2590         }
2591
2592         switch (inode->i_mode & S_IFMT) {
2593         case S_IFDIR:
2594                 if (!capable(CAP_SYS_ADMIN)) {
2595                         ret = -EPERM;
2596                         goto out;
2597                 }
2598                 ret = btrfs_defrag_root(root);
2599                 if (ret)
2600                         goto out;
2601                 ret = btrfs_defrag_root(root->fs_info->extent_root);
2602                 break;
2603         case S_IFREG:
2604                 if (!(file->f_mode & FMODE_WRITE)) {
2605                         ret = -EINVAL;
2606                         goto out;
2607                 }
2608
2609                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2610                 if (!range) {
2611                         ret = -ENOMEM;
2612                         goto out;
2613                 }
2614
2615                 if (argp) {
2616                         if (copy_from_user(range, argp,
2617                                            sizeof(*range))) {
2618                                 ret = -EFAULT;
2619                                 kfree(range);
2620                                 goto out;
2621                         }
2622                         /* compression requires us to start the IO */
2623                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2624                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2625                                 range->extent_thresh = (u32)-1;
2626                         }
2627                 } else {
2628                         /* the rest are all set to zero by kzalloc */
2629                         range->len = (u64)-1;
2630                 }
2631                 ret = btrfs_defrag_file(file_inode(file), file,
2632                                         range, 0, 0);
2633                 if (ret > 0)
2634                         ret = 0;
2635                 kfree(range);
2636                 break;
2637         default:
2638                 ret = -EINVAL;
2639         }
2640 out:
2641         mnt_drop_write_file(file);
2642         return ret;
2643 }
2644
2645 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2646 {
2647         struct btrfs_ioctl_vol_args *vol_args;
2648         int ret;
2649
2650         if (!capable(CAP_SYS_ADMIN))
2651                 return -EPERM;
2652
2653         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2654                         1)) {
2655                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2656         }
2657
2658         mutex_lock(&root->fs_info->volume_mutex);
2659         vol_args = memdup_user(arg, sizeof(*vol_args));
2660         if (IS_ERR(vol_args)) {
2661                 ret = PTR_ERR(vol_args);
2662                 goto out;
2663         }
2664
2665         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2666         ret = btrfs_init_new_device(root, vol_args->name);
2667
2668         if (!ret)
2669                 btrfs_info(root->fs_info, "disk added %s",vol_args->name);
2670
2671         kfree(vol_args);
2672 out:
2673         mutex_unlock(&root->fs_info->volume_mutex);
2674         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2675         return ret;
2676 }
2677
2678 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
2679 {
2680         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
2681         struct btrfs_ioctl_vol_args_v2 *vol_args;
2682         int ret;
2683
2684         if (!capable(CAP_SYS_ADMIN))
2685                 return -EPERM;
2686
2687         ret = mnt_want_write_file(file);
2688         if (ret)
2689                 return ret;
2690
2691         vol_args = memdup_user(arg, sizeof(*vol_args));
2692         if (IS_ERR(vol_args)) {
2693                 ret = PTR_ERR(vol_args);
2694                 goto err_drop;
2695         }
2696
2697         /* Check for compatibility reject unknown flags */
2698         if (vol_args->flags & ~BTRFS_VOL_ARG_V2_FLAGS_SUPPORTED)
2699                 return -EOPNOTSUPP;
2700
2701         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2702                         1)) {
2703                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2704                 goto out;
2705         }
2706
2707         mutex_lock(&root->fs_info->volume_mutex);
2708         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
2709                 ret = btrfs_rm_device(root, NULL, vol_args->devid);
2710         } else {
2711                 vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
2712                 ret = btrfs_rm_device(root, vol_args->name, 0);
2713         }
2714         mutex_unlock(&root->fs_info->volume_mutex);
2715         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2716
2717         if (!ret) {
2718                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
2719                         btrfs_info(root->fs_info, "device deleted: id %llu",
2720                                         vol_args->devid);
2721                 else
2722                         btrfs_info(root->fs_info, "device deleted: %s",
2723                                         vol_args->name);
2724         }
2725 out:
2726         kfree(vol_args);
2727 err_drop:
2728         mnt_drop_write_file(file);
2729         return ret;
2730 }
2731
2732 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2733 {
2734         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
2735         struct btrfs_ioctl_vol_args *vol_args;
2736         int ret;
2737
2738         if (!capable(CAP_SYS_ADMIN))
2739                 return -EPERM;
2740
2741         ret = mnt_want_write_file(file);
2742         if (ret)
2743                 return ret;
2744
2745         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2746                         1)) {
2747                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2748                 goto out_drop_write;
2749         }
2750
2751         vol_args = memdup_user(arg, sizeof(*vol_args));
2752         if (IS_ERR(vol_args)) {
2753                 ret = PTR_ERR(vol_args);
2754                 goto out;
2755         }
2756
2757         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2758         mutex_lock(&root->fs_info->volume_mutex);
2759         ret = btrfs_rm_device(root, vol_args->name, 0);
2760         mutex_unlock(&root->fs_info->volume_mutex);
2761
2762         if (!ret)
2763                 btrfs_info(root->fs_info, "disk deleted %s",vol_args->name);
2764         kfree(vol_args);
2765 out:
2766         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2767 out_drop_write:
2768         mnt_drop_write_file(file);
2769
2770         return ret;
2771 }
2772
2773 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2774 {
2775         struct btrfs_ioctl_fs_info_args *fi_args;
2776         struct btrfs_device *device;
2777         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2778         int ret = 0;
2779
2780         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2781         if (!fi_args)
2782                 return -ENOMEM;
2783
2784         mutex_lock(&fs_devices->device_list_mutex);
2785         fi_args->num_devices = fs_devices->num_devices;
2786         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2787
2788         list_for_each_entry(device, &fs_devices->devices, dev_list) {
2789                 if (device->devid > fi_args->max_id)
2790                         fi_args->max_id = device->devid;
2791         }
2792         mutex_unlock(&fs_devices->device_list_mutex);
2793
2794         fi_args->nodesize = root->fs_info->super_copy->nodesize;
2795         fi_args->sectorsize = root->fs_info->super_copy->sectorsize;
2796         fi_args->clone_alignment = root->fs_info->super_copy->sectorsize;
2797
2798         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2799                 ret = -EFAULT;
2800
2801         kfree(fi_args);
2802         return ret;
2803 }
2804
2805 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2806 {
2807         struct btrfs_ioctl_dev_info_args *di_args;
2808         struct btrfs_device *dev;
2809         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2810         int ret = 0;
2811         char *s_uuid = NULL;
2812
2813         di_args = memdup_user(arg, sizeof(*di_args));
2814         if (IS_ERR(di_args))
2815                 return PTR_ERR(di_args);
2816
2817         if (!btrfs_is_empty_uuid(di_args->uuid))
2818                 s_uuid = di_args->uuid;
2819
2820         mutex_lock(&fs_devices->device_list_mutex);
2821         dev = btrfs_find_device(root->fs_info, di_args->devid, s_uuid, NULL);
2822
2823         if (!dev) {
2824                 ret = -ENODEV;
2825                 goto out;
2826         }
2827
2828         di_args->devid = dev->devid;
2829         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
2830         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
2831         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2832         if (dev->name) {
2833                 struct rcu_string *name;
2834
2835                 rcu_read_lock();
2836                 name = rcu_dereference(dev->name);
2837                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2838                 rcu_read_unlock();
2839                 di_args->path[sizeof(di_args->path) - 1] = 0;
2840         } else {
2841                 di_args->path[0] = '\0';
2842         }
2843
2844 out:
2845         mutex_unlock(&fs_devices->device_list_mutex);
2846         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2847                 ret = -EFAULT;
2848
2849         kfree(di_args);
2850         return ret;
2851 }
2852
2853 static struct page *extent_same_get_page(struct inode *inode, pgoff_t index)
2854 {
2855         struct page *page;
2856
2857         page = grab_cache_page(inode->i_mapping, index);
2858         if (!page)
2859                 return ERR_PTR(-ENOMEM);
2860
2861         if (!PageUptodate(page)) {
2862                 int ret;
2863
2864                 ret = btrfs_readpage(NULL, page);
2865                 if (ret)
2866                         return ERR_PTR(ret);
2867                 lock_page(page);
2868                 if (!PageUptodate(page)) {
2869                         unlock_page(page);
2870                         put_page(page);
2871                         return ERR_PTR(-EIO);
2872                 }
2873                 if (page->mapping != inode->i_mapping) {
2874                         unlock_page(page);
2875                         put_page(page);
2876                         return ERR_PTR(-EAGAIN);
2877                 }
2878         }
2879
2880         return page;
2881 }
2882
2883 static int gather_extent_pages(struct inode *inode, struct page **pages,
2884                                int num_pages, u64 off)
2885 {
2886         int i;
2887         pgoff_t index = off >> PAGE_SHIFT;
2888
2889         for (i = 0; i < num_pages; i++) {
2890 again:
2891                 pages[i] = extent_same_get_page(inode, index + i);
2892                 if (IS_ERR(pages[i])) {
2893                         int err = PTR_ERR(pages[i]);
2894
2895                         if (err == -EAGAIN)
2896                                 goto again;
2897                         pages[i] = NULL;
2898                         return err;
2899                 }
2900         }
2901         return 0;
2902 }
2903
2904 static int lock_extent_range(struct inode *inode, u64 off, u64 len,
2905                              bool retry_range_locking)
2906 {
2907         /*
2908          * Do any pending delalloc/csum calculations on inode, one way or
2909          * another, and lock file content.
2910          * The locking order is:
2911          *
2912          *   1) pages
2913          *   2) range in the inode's io tree
2914          */
2915         while (1) {
2916                 struct btrfs_ordered_extent *ordered;
2917                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2918                 ordered = btrfs_lookup_first_ordered_extent(inode,
2919                                                             off + len - 1);
2920                 if ((!ordered ||
2921                      ordered->file_offset + ordered->len <= off ||
2922                      ordered->file_offset >= off + len) &&
2923                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2924                                     off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
2925                         if (ordered)
2926                                 btrfs_put_ordered_extent(ordered);
2927                         break;
2928                 }
2929                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2930                 if (ordered)
2931                         btrfs_put_ordered_extent(ordered);
2932                 if (!retry_range_locking)
2933                         return -EAGAIN;
2934                 btrfs_wait_ordered_range(inode, off, len);
2935         }
2936         return 0;
2937 }
2938
2939 static void btrfs_double_inode_unlock(struct inode *inode1, struct inode *inode2)
2940 {
2941         inode_unlock(inode1);
2942         inode_unlock(inode2);
2943 }
2944
2945 static void btrfs_double_inode_lock(struct inode *inode1, struct inode *inode2)
2946 {
2947         if (inode1 < inode2)
2948                 swap(inode1, inode2);
2949
2950         inode_lock_nested(inode1, I_MUTEX_PARENT);
2951         inode_lock_nested(inode2, I_MUTEX_CHILD);
2952 }
2953
2954 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
2955                                       struct inode *inode2, u64 loff2, u64 len)
2956 {
2957         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2958         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2959 }
2960
2961 static int btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
2962                                     struct inode *inode2, u64 loff2, u64 len,
2963                                     bool retry_range_locking)
2964 {
2965         int ret;
2966
2967         if (inode1 < inode2) {
2968                 swap(inode1, inode2);
2969                 swap(loff1, loff2);
2970         }
2971         ret = lock_extent_range(inode1, loff1, len, retry_range_locking);
2972         if (ret)
2973                 return ret;
2974         ret = lock_extent_range(inode2, loff2, len, retry_range_locking);
2975         if (ret)
2976                 unlock_extent(&BTRFS_I(inode1)->io_tree, loff1,
2977                               loff1 + len - 1);
2978         return ret;
2979 }
2980
2981 struct cmp_pages {
2982         int             num_pages;
2983         struct page     **src_pages;
2984         struct page     **dst_pages;
2985 };
2986
2987 static void btrfs_cmp_data_free(struct cmp_pages *cmp)
2988 {
2989         int i;
2990         struct page *pg;
2991
2992         for (i = 0; i < cmp->num_pages; i++) {
2993                 pg = cmp->src_pages[i];
2994                 if (pg) {
2995                         unlock_page(pg);
2996                         put_page(pg);
2997                 }
2998                 pg = cmp->dst_pages[i];
2999                 if (pg) {
3000                         unlock_page(pg);
3001                         put_page(pg);
3002                 }
3003         }
3004         kfree(cmp->src_pages);
3005         kfree(cmp->dst_pages);
3006 }
3007
3008 static int btrfs_cmp_data_prepare(struct inode *src, u64 loff,
3009                                   struct inode *dst, u64 dst_loff,
3010                                   u64 len, struct cmp_pages *cmp)
3011 {
3012         int ret;
3013         int num_pages = PAGE_ALIGN(len) >> PAGE_SHIFT;
3014         struct page **src_pgarr, **dst_pgarr;
3015
3016         /*
3017          * We must gather up all the pages before we initiate our
3018          * extent locking. We use an array for the page pointers. Size
3019          * of the array is bounded by len, which is in turn bounded by
3020          * BTRFS_MAX_DEDUPE_LEN.
3021          */
3022         src_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
3023         dst_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
3024         if (!src_pgarr || !dst_pgarr) {
3025                 kfree(src_pgarr);
3026                 kfree(dst_pgarr);
3027                 return -ENOMEM;
3028         }
3029         cmp->num_pages = num_pages;
3030         cmp->src_pages = src_pgarr;
3031         cmp->dst_pages = dst_pgarr;
3032
3033         ret = gather_extent_pages(src, cmp->src_pages, cmp->num_pages, loff);
3034         if (ret)
3035                 goto out;
3036
3037         ret = gather_extent_pages(dst, cmp->dst_pages, cmp->num_pages, dst_loff);
3038
3039 out:
3040         if (ret)
3041                 btrfs_cmp_data_free(cmp);
3042         return 0;
3043 }
3044
3045 static int btrfs_cmp_data(struct inode *src, u64 loff, struct inode *dst,
3046                           u64 dst_loff, u64 len, struct cmp_pages *cmp)
3047 {
3048         int ret = 0;
3049         int i;
3050         struct page *src_page, *dst_page;
3051         unsigned int cmp_len = PAGE_SIZE;
3052         void *addr, *dst_addr;
3053
3054         i = 0;
3055         while (len) {
3056                 if (len < PAGE_SIZE)
3057                         cmp_len = len;
3058
3059                 BUG_ON(i >= cmp->num_pages);
3060
3061                 src_page = cmp->src_pages[i];
3062                 dst_page = cmp->dst_pages[i];
3063                 ASSERT(PageLocked(src_page));
3064                 ASSERT(PageLocked(dst_page));
3065
3066                 addr = kmap_atomic(src_page);
3067                 dst_addr = kmap_atomic(dst_page);
3068
3069                 flush_dcache_page(src_page);
3070                 flush_dcache_page(dst_page);
3071
3072                 if (memcmp(addr, dst_addr, cmp_len))
3073                         ret = -EBADE;
3074
3075                 kunmap_atomic(addr);
3076                 kunmap_atomic(dst_addr);
3077
3078                 if (ret)
3079                         break;
3080
3081                 len -= cmp_len;
3082                 i++;
3083         }
3084
3085         return ret;
3086 }
3087
3088 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 *plen,
3089                                      u64 olen)
3090 {
3091         u64 len = *plen;
3092         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
3093
3094         if (off + olen > inode->i_size || off + olen < off)
3095                 return -EINVAL;
3096
3097         /* if we extend to eof, continue to block boundary */
3098         if (off + len == inode->i_size)
3099                 *plen = len = ALIGN(inode->i_size, bs) - off;
3100
3101         /* Check that we are block aligned - btrfs_clone() requires this */
3102         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
3103                 return -EINVAL;
3104
3105         return 0;
3106 }
3107
3108 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
3109                              struct inode *dst, u64 dst_loff)
3110 {
3111         int ret;
3112         u64 len = olen;
3113         struct cmp_pages cmp;
3114         int same_inode = 0;
3115         u64 same_lock_start = 0;
3116         u64 same_lock_len = 0;
3117
3118         if (src == dst)
3119                 same_inode = 1;
3120
3121         if (len == 0)
3122                 return 0;
3123
3124         if (same_inode) {
3125                 inode_lock(src);
3126
3127                 ret = extent_same_check_offsets(src, loff, &len, olen);
3128                 if (ret)
3129                         goto out_unlock;
3130                 ret = extent_same_check_offsets(src, dst_loff, &len, olen);
3131                 if (ret)
3132                         goto out_unlock;
3133
3134                 /*
3135                  * Single inode case wants the same checks, except we
3136                  * don't want our length pushed out past i_size as
3137                  * comparing that data range makes no sense.
3138                  *
3139                  * extent_same_check_offsets() will do this for an
3140                  * unaligned length at i_size, so catch it here and
3141                  * reject the request.
3142                  *
3143                  * This effectively means we require aligned extents
3144                  * for the single-inode case, whereas the other cases
3145                  * allow an unaligned length so long as it ends at
3146                  * i_size.
3147                  */
3148                 if (len != olen) {
3149                         ret = -EINVAL;
3150                         goto out_unlock;
3151                 }
3152
3153                 /* Check for overlapping ranges */
3154                 if (dst_loff + len > loff && dst_loff < loff + len) {
3155                         ret = -EINVAL;
3156                         goto out_unlock;
3157                 }
3158
3159                 same_lock_start = min_t(u64, loff, dst_loff);
3160                 same_lock_len = max_t(u64, loff, dst_loff) + len - same_lock_start;
3161         } else {
3162                 btrfs_double_inode_lock(src, dst);
3163
3164                 ret = extent_same_check_offsets(src, loff, &len, olen);
3165                 if (ret)
3166                         goto out_unlock;
3167
3168                 ret = extent_same_check_offsets(dst, dst_loff, &len, olen);
3169                 if (ret)
3170                         goto out_unlock;
3171         }
3172
3173         /* don't make the dst file partly checksummed */
3174         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3175             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
3176                 ret = -EINVAL;
3177                 goto out_unlock;
3178         }
3179
3180 again:
3181         ret = btrfs_cmp_data_prepare(src, loff, dst, dst_loff, olen, &cmp);
3182         if (ret)
3183                 goto out_unlock;
3184
3185         if (same_inode)
3186                 ret = lock_extent_range(src, same_lock_start, same_lock_len,
3187                                         false);
3188         else
3189                 ret = btrfs_double_extent_lock(src, loff, dst, dst_loff, len,
3190                                                false);
3191         /*
3192          * If one of the inodes has dirty pages in the respective range or
3193          * ordered extents, we need to flush dellaloc and wait for all ordered
3194          * extents in the range. We must unlock the pages and the ranges in the
3195          * io trees to avoid deadlocks when flushing delalloc (requires locking
3196          * pages) and when waiting for ordered extents to complete (they require
3197          * range locking).
3198          */
3199         if (ret == -EAGAIN) {
3200                 /*
3201                  * Ranges in the io trees already unlocked. Now unlock all
3202                  * pages before waiting for all IO to complete.
3203                  */
3204                 btrfs_cmp_data_free(&cmp);
3205                 if (same_inode) {
3206                         btrfs_wait_ordered_range(src, same_lock_start,
3207                                                  same_lock_len);
3208                 } else {
3209                         btrfs_wait_ordered_range(src, loff, len);
3210                         btrfs_wait_ordered_range(dst, dst_loff, len);
3211                 }
3212                 goto again;
3213         }
3214         ASSERT(ret == 0);
3215         if (WARN_ON(ret)) {
3216                 /* ranges in the io trees already unlocked */
3217                 btrfs_cmp_data_free(&cmp);
3218                 return ret;
3219         }
3220
3221         /* pass original length for comparison so we stay within i_size */
3222         ret = btrfs_cmp_data(src, loff, dst, dst_loff, olen, &cmp);
3223         if (ret == 0)
3224                 ret = btrfs_clone(src, dst, loff, olen, len, dst_loff, 1);
3225
3226         if (same_inode)
3227                 unlock_extent(&BTRFS_I(src)->io_tree, same_lock_start,
3228                               same_lock_start + same_lock_len - 1);
3229         else
3230                 btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
3231
3232         btrfs_cmp_data_free(&cmp);
3233 out_unlock:
3234         if (same_inode)
3235                 inode_unlock(src);
3236         else
3237                 btrfs_double_inode_unlock(src, dst);
3238
3239         return ret;
3240 }
3241
3242 #define BTRFS_MAX_DEDUPE_LEN    SZ_16M
3243
3244 ssize_t btrfs_dedupe_file_range(struct file *src_file, u64 loff, u64 olen,
3245                                 struct file *dst_file, u64 dst_loff)
3246 {
3247         struct inode *src = file_inode(src_file);
3248         struct inode *dst = file_inode(dst_file);
3249         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
3250         ssize_t res;
3251
3252         if (olen > BTRFS_MAX_DEDUPE_LEN)
3253                 olen = BTRFS_MAX_DEDUPE_LEN;
3254
3255         if (WARN_ON_ONCE(bs < PAGE_SIZE)) {
3256                 /*