fs/btrfs/ioctl.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/fsnotify.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/highmem.h>
  27 #include <linux/time.h>
  28 #include <linux/init.h>
  29 #include <linux/string.h>
  30 #include <linux/backing-dev.h>
  31 #include <linux/mount.h>
  32 #include <linux/mpage.h>
  33 #include <linux/namei.h>
  34 #include <linux/swap.h>
  35 #include <linux/writeback.h>
  36 #include <linux/compat.h>
  37 #include <linux/bit_spinlock.h>
  38 #include <linux/security.h>
  39 #include <linux/xattr.h>
  40 #include <linux/mm.h>
  41 #include <linux/slab.h>
  42 #include <linux/blkdev.h>
  43 #include <linux/uuid.h>
  44 #include <linux/btrfs.h>
  45 #include <linux/uaccess.h>
  46 #include "ctree.h"
  47 #include "disk-io.h"
  48 #include "transaction.h"
  49 #include "btrfs_inode.h"
  50 #include "print-tree.h"
  51 #include "volumes.h"
  52 #include "locking.h"
  53 #include "inode-map.h"
  54 #include "backref.h"
  55 #include "rcu-string.h"
  56 #include "send.h"
  57 #include "dev-replace.h"
  58 #include "props.h"
  59 #include "sysfs.h"
  60 #include "qgroup.h"
  61 #include "tree-log.h"
  62 #include "compression.h"
  63
  64 #ifdef CONFIG_64BIT
  65 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
  66  * structures are incorrect, as the timespec structure from userspace
  67  * is 4 bytes too small. We define these alternatives here to teach
  68  * the kernel about the 32-bit struct packing.
  69  */
  70 struct btrfs_ioctl_timespec_32 {
  71         __u64 sec;
  72         __u32 nsec;
  73 } __attribute__ ((__packed__));
  74
  75 struct btrfs_ioctl_received_subvol_args_32 {
  76         char    uuid[BTRFS_UUID_SIZE];  /* in */
  77         __u64   stransid;               /* in */
  78         __u64   rtransid;               /* out */
  79         struct btrfs_ioctl_timespec_32 stime; /* in */
  80         struct btrfs_ioctl_timespec_32 rtime; /* out */
  81         __u64   flags;                  /* in */
  82         __u64   reserved[16];           /* in */
  83 } __attribute__ ((__packed__));
  84
  85 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
  86                                 struct btrfs_ioctl_received_subvol_args_32)
  87 #endif
  88
  89
  90 static int btrfs_clone(struct inode *src, struct inode *inode,
  91                        u64 off, u64 olen, u64 olen_aligned, u64 destoff,
  92                        int no_time_update);
  93
  94 /* Mask out flags that are inappropriate for the given type of inode. */
  95 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
  96 {
  97         if (S_ISDIR(mode))
  98                 return flags;
  99         else if (S_ISREG(mode))
 100                 return flags & ~FS_DIRSYNC_FL;
 101         else
 102                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
 103 }
 104
 105 /*
 106  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
 107  */
 108 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
 109 {
 110         unsigned int iflags = 0;
 111
 112         if (flags & BTRFS_INODE_SYNC)
 113                 iflags |= FS_SYNC_FL;
 114         if (flags & BTRFS_INODE_IMMUTABLE)
 115                 iflags |= FS_IMMUTABLE_FL;
 116         if (flags & BTRFS_INODE_APPEND)
 117                 iflags |= FS_APPEND_FL;
 118         if (flags & BTRFS_INODE_NODUMP)
 119                 iflags |= FS_NODUMP_FL;
 120         if (flags & BTRFS_INODE_NOATIME)
 121                 iflags |= FS_NOATIME_FL;
 122         if (flags & BTRFS_INODE_DIRSYNC)
 123                 iflags |= FS_DIRSYNC_FL;
 124         if (flags & BTRFS_INODE_NODATACOW)
 125                 iflags |= FS_NOCOW_FL;
 126
 127         if (flags & BTRFS_INODE_NOCOMPRESS)
 128                 iflags |= FS_NOCOMP_FL;
 129         else if (flags & BTRFS_INODE_COMPRESS)
 130                 iflags |= FS_COMPR_FL;
 131
 132         return iflags;
 133 }
 134
 135 /*
 136  * Update inode->i_flags based on the btrfs internal flags.
 137  */
 138 void btrfs_update_iflags(struct inode *inode)
 139 {
 140         struct btrfs_inode *ip = BTRFS_I(inode);
 141         unsigned int new_fl = 0;
 142
 143         if (ip->flags & BTRFS_INODE_SYNC)
 144                 new_fl |= S_SYNC;
 145         if (ip->flags & BTRFS_INODE_IMMUTABLE)
 146                 new_fl |= S_IMMUTABLE;
 147         if (ip->flags & BTRFS_INODE_APPEND)
 148                 new_fl |= S_APPEND;
 149         if (ip->flags & BTRFS_INODE_NOATIME)
 150                 new_fl |= S_NOATIME;
 151         if (ip->flags & BTRFS_INODE_DIRSYNC)
 152                 new_fl |= S_DIRSYNC;
 153
 154         set_mask_bits(&inode->i_flags,
 155                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC,
 156                       new_fl);
 157 }
 158
 159 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 160 {
 161         struct btrfs_inode *ip = BTRFS_I(file_inode(file));
 162         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
 163
 164         if (copy_to_user(arg, &flags, sizeof(flags)))
 165                 return -EFAULT;
 166         return 0;
 167 }
 168
 169 static int check_flags(unsigned int flags)
 170 {
 171         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 172                       FS_NOATIME_FL | FS_NODUMP_FL | \
 173                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 174                       FS_NOCOMP_FL | FS_COMPR_FL |
 175                       FS_NOCOW_FL))
 176                 return -EOPNOTSUPP;
 177
 178         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 179                 return -EINVAL;
 180
 181         return 0;
 182 }
 183
 184 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 185 {
 186         struct inode *inode = file_inode(file);
 187         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 188         struct btrfs_inode *ip = BTRFS_I(inode);
 189         struct btrfs_root *root = ip->root;
 190         struct btrfs_trans_handle *trans;
 191         unsigned int flags, oldflags;
 192         int ret;
 193         u64 ip_oldflags;
 194         unsigned int i_oldflags;
 195         umode_t mode;
 196
 197         if (!inode_owner_or_capable(inode))
 198                 return -EPERM;
 199
 200         if (btrfs_root_readonly(root))
 201                 return -EROFS;
 202
 203         if (copy_from_user(&flags, arg, sizeof(flags)))
 204                 return -EFAULT;
 205
 206         ret = check_flags(flags);
 207         if (ret)
 208                 return ret;
 209
 210         ret = mnt_want_write_file(file);
 211         if (ret)
 212                 return ret;
 213
 214         inode_lock(inode);
 215
 216         ip_oldflags = ip->flags;
 217         i_oldflags = inode->i_flags;
 218         mode = inode->i_mode;
 219
 220         flags = btrfs_mask_flags(inode->i_mode, flags);
 221         oldflags = btrfs_flags_to_ioctl(ip->flags);
 222         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 223                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 224                         ret = -EPERM;
 225                         goto out_unlock;
 226                 }
 227         }
 228
 229         if (flags & FS_SYNC_FL)
 230                 ip->flags |= BTRFS_INODE_SYNC;
 231         else
 232                 ip->flags &= ~BTRFS_INODE_SYNC;
 233         if (flags & FS_IMMUTABLE_FL)
 234                 ip->flags |= BTRFS_INODE_IMMUTABLE;
 235         else
 236                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
 237         if (flags & FS_APPEND_FL)
 238                 ip->flags |= BTRFS_INODE_APPEND;
 239         else
 240                 ip->flags &= ~BTRFS_INODE_APPEND;
 241         if (flags & FS_NODUMP_FL)
 242                 ip->flags |= BTRFS_INODE_NODUMP;
 243         else
 244                 ip->flags &= ~BTRFS_INODE_NODUMP;
 245         if (flags & FS_NOATIME_FL)
 246                 ip->flags |= BTRFS_INODE_NOATIME;
 247         else
 248                 ip->flags &= ~BTRFS_INODE_NOATIME;
 249         if (flags & FS_DIRSYNC_FL)
 250                 ip->flags |= BTRFS_INODE_DIRSYNC;
 251         else
 252                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
 253         if (flags & FS_NOCOW_FL) {
 254                 if (S_ISREG(mode)) {
 255                         /*
 256                          * It's safe to turn csums off here, no extents exist.
 257                          * Otherwise we want the flag to reflect the real COW
 258                          * status of the file and will not set it.
 259                          */
 260                         if (inode->i_size == 0)
 261                                 ip->flags |= BTRFS_INODE_NODATACOW
 262                                            | BTRFS_INODE_NODATASUM;
 263                 } else {
 264                         ip->flags |= BTRFS_INODE_NODATACOW;
 265                 }
 266         } else {
 267                 /*
 268                  * Revert back under same assumptions as above
 269                  */
 270                 if (S_ISREG(mode)) {
 271                         if (inode->i_size == 0)
 272                                 ip->flags &= ~(BTRFS_INODE_NODATACOW
 273                                              | BTRFS_INODE_NODATASUM);
 274                 } else {
 275                         ip->flags &= ~BTRFS_INODE_NODATACOW;
 276                 }
 277         }
 278
 279         /*
 280          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 281          * flag may be changed automatically if compression code won't make
 282          * things smaller.
 283          */
 284         if (flags & FS_NOCOMP_FL) {
 285                 ip->flags &= ~BTRFS_INODE_COMPRESS;
 286                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
 287
 288                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 289                 if (ret && ret != -ENODATA)
 290                         goto out_drop;
 291         } else if (flags & FS_COMPR_FL) {
 292                 const char *comp;
 293
 294                 ip->flags |= BTRFS_INODE_COMPRESS;
 295                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
 296
 297                 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
 298                         comp = "lzo";
 299                 else
 300                         comp = "zlib";
 301                 ret = btrfs_set_prop(inode, "btrfs.compression",
 302                                      comp, strlen(comp), 0);
 303                 if (ret)
 304                         goto out_drop;
 305
 306         } else {
 307                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 308                 if (ret && ret != -ENODATA)
 309                         goto out_drop;
 310                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 311         }
 312
 313         trans = btrfs_start_transaction(root, 1);
 314         if (IS_ERR(trans)) {
 315                 ret = PTR_ERR(trans);
 316                 goto out_drop;
 317         }
 318
 319         btrfs_update_iflags(inode);
 320         inode_inc_iversion(inode);
 321         inode->i_ctime = current_time(inode);
 322         ret = btrfs_update_inode(trans, root, inode);
 323
 324         btrfs_end_transaction(trans);
 325  out_drop:
 326         if (ret) {
 327                 ip->flags = ip_oldflags;
 328                 inode->i_flags = i_oldflags;
 329         }
 330
 331  out_unlock:
 332         inode_unlock(inode);
 333         mnt_drop_write_file(file);
 334         return ret;
 335 }
 336
 337 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 338 {
 339         struct inode *inode = file_inode(file);
 340
 341         return put_user(inode->i_generation, arg);
 342 }
 343
 344 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
 345 {
 346         struct inode *inode = file_inode(file);
 347         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 348         struct btrfs_device *device;
 349         struct request_queue *q;
 350         struct fstrim_range range;
 351         u64 minlen = ULLONG_MAX;
 352         u64 num_devices = 0;
 353         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
 354         int ret;
 355
 356         if (!capable(CAP_SYS_ADMIN))
 357                 return -EPERM;
 358
 359         rcu_read_lock();
 360         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 361                                 dev_list) {
 362                 if (!device->bdev)
 363                         continue;
 364                 q = bdev_get_queue(device->bdev);
 365                 if (blk_queue_discard(q)) {
 366                         num_devices++;
 367                         minlen = min_t(u64, q->limits.discard_granularity,
 368                                      minlen);
 369                 }
 370         }
 371         rcu_read_unlock();
 372
 373         if (!num_devices)
 374                 return -EOPNOTSUPP;
 375         if (copy_from_user(&range, arg, sizeof(range)))
 376                 return -EFAULT;
 377         if (range.start > total_bytes ||
 378             range.len < fs_info->sb->s_blocksize)
 379                 return -EINVAL;
 380
 381         range.len = min(range.len, total_bytes - range.start);
 382         range.minlen = max(range.minlen, minlen);
 383         ret = btrfs_trim_fs(fs_info, &range);
 384         if (ret < 0)
 385                 return ret;
 386
 387         if (copy_to_user(arg, &range, sizeof(range)))
 388                 return -EFAULT;
 389
 390         return 0;
 391 }
 392
 393 int btrfs_is_empty_uuid(u8 *uuid)
 394 {
 395         int i;
 396
 397         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 398                 if (uuid[i])
 399                         return 0;
 400         }
 401         return 1;
 402 }
 403
 404 static noinline int create_subvol(struct inode *dir,
 405                                   struct dentry *dentry,
 406                                   const char *name, int namelen,
 407                                   u64 *async_transid,
 408                                   struct btrfs_qgroup_inherit *inherit)
 409 {
 410         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 411         struct btrfs_trans_handle *trans;
 412         struct btrfs_key key;
 413         struct btrfs_root_item *root_item;
 414         struct btrfs_inode_item *inode_item;
 415         struct extent_buffer *leaf;
 416         struct btrfs_root *root = BTRFS_I(dir)->root;
 417         struct btrfs_root *new_root;
 418         struct btrfs_block_rsv block_rsv;
 419         struct timespec cur_time = current_time(dir);
 420         struct inode *inode;
 421         int ret;
 422         int err;
 423         u64 objectid;
 424         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 425         u64 index = 0;
 426         u64 qgroup_reserved;
 427         uuid_le new_uuid;
 428
 429         root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
 430         if (!root_item)
 431                 return -ENOMEM;
 432
 433         ret = btrfs_find_free_objectid(fs_info->tree_root, &objectid);
 434         if (ret)
 435                 goto fail_free;
 436
 437         /*
 438          * Don't create subvolume whose level is not zero. Or qgroup will be
 439          * screwed up since it assumes subvolume qgroup's level to be 0.
 440          */
 441         if (btrfs_qgroup_level(objectid)) {
 442                 ret = -ENOSPC;
 443                 goto fail_free;
 444         }
 445
 446         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 447         /*
 448          * The same as the snapshot creation, please see the comment
 449          * of create_snapshot().
 450          */
 451         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
 452                                                8, &qgroup_reserved, false);
 453         if (ret)
 454                 goto fail_free;
 455
 456         trans = btrfs_start_transaction(root, 0);
 457         if (IS_ERR(trans)) {
 458                 ret = PTR_ERR(trans);
 459                 btrfs_subvolume_release_metadata(fs_info, &block_rsv);
 460                 goto fail_free;
 461         }
 462         trans->block_rsv = &block_rsv;
 463         trans->bytes_reserved = block_rsv.size;
 464
 465         ret = btrfs_qgroup_inherit(trans, fs_info, 0, objectid, inherit);
 466         if (ret)
 467                 goto fail;
 468
 469         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
 470         if (IS_ERR(leaf)) {
 471                 ret = PTR_ERR(leaf);
 472                 goto fail;
 473         }
 474
 475         memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
 476         btrfs_set_header_bytenr(leaf, leaf->start);
 477         btrfs_set_header_generation(leaf, trans->transid);
 478         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 479         btrfs_set_header_owner(leaf, objectid);
 480
 481         write_extent_buffer_fsid(leaf, fs_info->fsid);
 482         write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
 483         btrfs_mark_buffer_dirty(leaf);
 484
 485         inode_item = &root_item->inode;
 486         btrfs_set_stack_inode_generation(inode_item, 1);
 487         btrfs_set_stack_inode_size(inode_item, 3);
 488         btrfs_set_stack_inode_nlink(inode_item, 1);
 489         btrfs_set_stack_inode_nbytes(inode_item,
 490                                      fs_info->nodesize);
 491         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 492
 493         btrfs_set_root_flags(root_item, 0);
 494         btrfs_set_root_limit(root_item, 0);
 495         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 496
 497         btrfs_set_root_bytenr(root_item, leaf->start);
 498         btrfs_set_root_generation(root_item, trans->transid);
 499         btrfs_set_root_level(root_item, 0);
 500         btrfs_set_root_refs(root_item, 1);
 501         btrfs_set_root_used(root_item, leaf->len);
 502         btrfs_set_root_last_snapshot(root_item, 0);
 503
 504         btrfs_set_root_generation_v2(root_item,
 505                         btrfs_root_generation(root_item));
 506         uuid_le_gen(&new_uuid);
 507         memcpy(root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
 508         btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
 509         btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
 510         root_item->ctime = root_item->otime;
 511         btrfs_set_root_ctransid(root_item, trans->transid);
 512         btrfs_set_root_otransid(root_item, trans->transid);
 513
 514         btrfs_tree_unlock(leaf);
 515         free_extent_buffer(leaf);
 516         leaf = NULL;
 517
 518         btrfs_set_root_dirid(root_item, new_dirid);
 519
 520         key.objectid = objectid;
 521         key.offset = 0;
 522         key.type = BTRFS_ROOT_ITEM_KEY;
 523         ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
 524                                 root_item);
 525         if (ret)
 526                 goto fail;
 527
 528         key.offset = (u64)-1;
 529         new_root = btrfs_read_fs_root_no_name(fs_info, &key);
 530         if (IS_ERR(new_root)) {
 531                 ret = PTR_ERR(new_root);
 532                 btrfs_abort_transaction(trans, ret);
 533                 goto fail;
 534         }
 535
 536         btrfs_record_root_in_trans(trans, new_root);
 537
 538         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
 539         if (ret) {
 540                 /* We potentially lose an unused inode item here */
 541                 btrfs_abort_transaction(trans, ret);
 542                 goto fail;
 543         }
 544
 545         mutex_lock(&new_root->objectid_mutex);
 546         new_root->highest_objectid = new_dirid;
 547         mutex_unlock(&new_root->objectid_mutex);
 548
 549         /*
 550          * insert the directory item
 551          */
 552         ret = btrfs_set_inode_index(BTRFS_I(dir), &index);
 553         if (ret) {
 554                 btrfs_abort_transaction(trans, ret);
 555                 goto fail;
 556         }
 557
 558         ret = btrfs_insert_dir_item(trans, root,
 559                                     name, namelen, BTRFS_I(dir), &key,
 560                                     BTRFS_FT_DIR, index);
 561         if (ret) {
 562                 btrfs_abort_transaction(trans, ret);
 563                 goto fail;
 564         }
 565
 566         btrfs_i_size_write(BTRFS_I(dir), dir->i_size + namelen * 2);
 567         ret = btrfs_update_inode(trans, root, dir);
 568         BUG_ON(ret);
 569
 570         ret = btrfs_add_root_ref(trans, fs_info,
 571                                  objectid, root->root_key.objectid,
 572                                  btrfs_ino(BTRFS_I(dir)), index, name, namelen);
 573         BUG_ON(ret);
 574
 575         ret = btrfs_uuid_tree_add(trans, fs_info, root_item->uuid,
 576                                   BTRFS_UUID_KEY_SUBVOL, objectid);
 577         if (ret)
 578                 btrfs_abort_transaction(trans, ret);
 579
 580 fail:
 581         kfree(root_item);
 582         trans->block_rsv = NULL;
 583         trans->bytes_reserved = 0;
 584         btrfs_subvolume_release_metadata(fs_info, &block_rsv);
 585
 586         if (async_transid) {
 587                 *async_transid = trans->transid;
 588                 err = btrfs_commit_transaction_async(trans, 1);
 589                 if (err)
 590                         err = btrfs_commit_transaction(trans);
 591         } else {
 592                 err = btrfs_commit_transaction(trans);
 593         }
 594         if (err && !ret)
 595                 ret = err;
 596
 597         if (!ret) {
 598                 inode = btrfs_lookup_dentry(dir, dentry);
 599                 if (IS_ERR(inode))
 600                         return PTR_ERR(inode);
 601                 d_instantiate(dentry, inode);
 602         }
 603         return ret;
 604
 605 fail_free:
 606         kfree(root_item);
 607         return ret;
 608 }
 609
 610 static void btrfs_wait_for_no_snapshotting_writes(struct btrfs_root *root)
 611 {
 612         s64 writers;
 613         DEFINE_WAIT(wait);
 614
 615         do {
 616                 prepare_to_wait(&root->subv_writers->wait, &wait,
 617                                 TASK_UNINTERRUPTIBLE);
 618
 619                 writers = percpu_counter_sum(&root->subv_writers->counter);
 620                 if (writers)
 621                         schedule();
 622
 623                 finish_wait(&root->subv_writers->wait, &wait);
 624         } while (writers);
 625 }
 626
 627 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 628                            struct dentry *dentry,
 629                            u64 *async_transid, bool readonly,
 630                            struct btrfs_qgroup_inherit *inherit)
 631 {
 632         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 633         struct inode *inode;
 634         struct btrfs_pending_snapshot *pending_snapshot;
 635         struct btrfs_trans_handle *trans;
 636         int ret;
 637
 638         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 639                 return -EINVAL;
 640
 641         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
 642         if (!pending_snapshot)
 643                 return -ENOMEM;
 644
 645         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 646                         GFP_KERNEL);
 647         pending_snapshot->path = btrfs_alloc_path();
 648         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 649                 ret = -ENOMEM;
 650                 goto free_pending;
 651         }
 652
 653         atomic_inc(&root->will_be_snapshotted);
 654         smp_mb__after_atomic();
 655         btrfs_wait_for_no_snapshotting_writes(root);
 656
 657         ret = btrfs_start_delalloc_inodes(root, 0);
 658         if (ret)
 659                 goto dec_and_free;
 660
 661         btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
 662
 663         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 664                              BTRFS_BLOCK_RSV_TEMP);
 665         /*
 666          * 1 - parent dir inode
 667          * 2 - dir entries
 668          * 1 - root item
 669          * 2 - root ref/backref
 670          * 1 - root of snapshot
 671          * 1 - UUID item
 672          */
 673         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 674                                         &pending_snapshot->block_rsv, 8,
 675                                         &pending_snapshot->qgroup_reserved,
 676                                         false);
 677         if (ret)
 678                 goto dec_and_free;
 679
 680         pending_snapshot->dentry = dentry;
 681         pending_snapshot->root = root;
 682         pending_snapshot->readonly = readonly;
 683         pending_snapshot->dir = dir;
 684         pending_snapshot->inherit = inherit;
 685
 686         trans = btrfs_start_transaction(root, 0);
 687         if (IS_ERR(trans)) {
 688                 ret = PTR_ERR(trans);
 689                 goto fail;
 690         }
 691
 692         spin_lock(&fs_info->trans_lock);
 693         list_add(&pending_snapshot->list,
 694                  &trans->transaction->pending_snapshots);
 695         spin_unlock(&fs_info->trans_lock);
 696         if (async_transid) {
 697                 *async_transid = trans->transid;
 698                 ret = btrfs_commit_transaction_async(trans, 1);
 699                 if (ret)
 700                         ret = btrfs_commit_transaction(trans);
 701         } else {
 702                 ret = btrfs_commit_transaction(trans);
 703         }
 704         if (ret)
 705                 goto fail;
 706
 707         ret = pending_snapshot->error;
 708         if (ret)
 709                 goto fail;
 710
 711         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 712         if (ret)
 713                 goto fail;
 714
 715         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 716         if (IS_ERR(inode)) {
 717                 ret = PTR_ERR(inode);
 718                 goto fail;
 719         }
 720
 721         d_instantiate(dentry, inode);
 722         ret = 0;
 723 fail:
 724         btrfs_subvolume_release_metadata(fs_info, &pending_snapshot->block_rsv);
 725 dec_and_free:
 726         if (atomic_dec_and_test(&root->will_be_snapshotted))
 727                 wake_up_atomic_t(&root->will_be_snapshotted);
 728 free_pending:
 729         kfree(pending_snapshot->root_item);
 730         btrfs_free_path(pending_snapshot->path);
 731         kfree(pending_snapshot);
 732
 733         return ret;
 734 }
 735
 736 /*  copy of may_delete in fs/namei.c()
 737  *      Check whether we can remove a link victim from directory dir, check
 738  *  whether the type of victim is right.
 739  *  1. We can't do it if dir is read-only (done in permission())
 740  *  2. We should have write and exec permissions on dir
 741  *  3. We can't remove anything from append-only dir
 742  *  4. We can't do anything with immutable dir (done in permission())
 743  *  5. If the sticky bit on dir is set we should either
 744  *      a. be owner of dir, or
 745  *      b. be owner of victim, or
 746  *      c. have CAP_FOWNER capability
 747  *  6. If the victim is append-only or immutable we can't do anything with
 748  *     links pointing to it.
 749  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 750  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 751  *  9. We can't remove a root or mountpoint.
 752  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 753  *     nfs_async_unlink().
 754  */
 755
 756 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 757 {
 758         int error;
 759
 760         if (d_really_is_negative(victim))
 761                 return -ENOENT;
 762
 763         BUG_ON(d_inode(victim->d_parent) != dir);
 764         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 765
 766         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 767         if (error)
 768                 return error;
 769         if (IS_APPEND(dir))
 770                 return -EPERM;
 771         if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
 772             IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
 773                 return -EPERM;
 774         if (isdir) {
 775                 if (!d_is_dir(victim))
 776                         return -ENOTDIR;
 777                 if (IS_ROOT(victim))
 778                         return -EBUSY;
 779         } else if (d_is_dir(victim))
 780                 return -EISDIR;
 781         if (IS_DEADDIR(dir))
 782                 return -ENOENT;
 783         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 784                 return -EBUSY;
 785         return 0;
 786 }
 787
 788 /* copy of may_create in fs/namei.c() */
 789 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 790 {
 791         if (d_really_is_positive(child))
 792                 return -EEXIST;
 793         if (IS_DEADDIR(dir))
 794                 return -ENOENT;
 795         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 796 }
 797
 798 /*
 799  * Create a new subvolume below @parent.  This is largely modeled after
 800  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 801  * inside this filesystem so it's quite a bit simpler.
 802  */
 803 static noinline int btrfs_mksubvol(const struct path *parent,
 804                                    const char *name, int namelen,
 805                                    struct btrfs_root *snap_src,
 806                                    u64 *async_transid, bool readonly,
 807                                    struct btrfs_qgroup_inherit *inherit)
 808 {
 809         struct inode *dir = d_inode(parent->dentry);
 810         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 811         struct dentry *dentry;
 812         int error;
 813
 814         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
 815         if (error == -EINTR)
 816                 return error;
 817
 818         dentry = lookup_one_len(name, parent->dentry, namelen);
 819         error = PTR_ERR(dentry);
 820         if (IS_ERR(dentry))
 821                 goto out_unlock;
 822
 823         error = btrfs_may_create(dir, dentry);
 824         if (error)
 825                 goto out_dput;
 826
 827         /*
 828          * even if this name doesn't exist, we may get hash collisions.
 829          * check for them now when we can safely fail
 830          */
 831         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 832                                                dir->i_ino, name,
 833                                                namelen);
 834         if (error)
 835                 goto out_dput;
 836
 837         down_read(&fs_info->subvol_sem);
 838
 839         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 840                 goto out_up_read;
 841
 842         if (snap_src) {
 843                 error = create_snapshot(snap_src, dir, dentry,
 844                                         async_transid, readonly, inherit);
 845         } else {
 846                 error = create_subvol(dir, dentry, name, namelen,
 847                                       async_transid, inherit);
 848         }
 849         if (!error)
 850                 fsnotify_mkdir(dir, dentry);
 851 out_up_read:
 852         up_read(&fs_info->subvol_sem);
 853 out_dput:
 854         dput(dentry);
 855 out_unlock:
 856         inode_unlock(dir);
 857         return error;
 858 }
 859
 860 /*
 861  * When we're defragging a range, we don't want to kick it off again
 862  * if it is really just waiting for delalloc to send it down.
 863  * If we find a nice big extent or delalloc range for the bytes in the
 864  * file you want to defrag, we return 0 to let you know to skip this
 865  * part of the file
 866  */
 867 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
 868 {
 869         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 870         struct extent_map *em = NULL;
 871         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 872         u64 end;
 873
 874         read_lock(&em_tree->lock);
 875         em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
 876         read_unlock(&em_tree->lock);
 877
 878         if (em) {
 879                 end = extent_map_end(em);
 880                 free_extent_map(em);
 881                 if (end - offset > thresh)
 882                         return 0;
 883         }
 884         /* if we already have a nice delalloc here, just stop */
 885         thresh /= 2;
 886         end = count_range_bits(io_tree, &offset, offset + thresh,
 887                                thresh, EXTENT_DELALLOC, 1);
 888         if (end >= thresh)
 889                 return 0;
 890         return 1;
 891 }
 892
 893 /*
 894  * helper function to walk through a file and find extents
 895  * newer than a specific transid, and smaller than thresh.
 896  *
 897  * This is used by the defragging code to find new and small
 898  * extents
 899  */
 900 static int find_new_extents(struct btrfs_root *root,
 901                             struct inode *inode, u64 newer_than,
 902                             u64 *off, u32 thresh)
 903 {
 904         struct btrfs_path *path;
 905         struct btrfs_key min_key;
 906         struct extent_buffer *leaf;
 907         struct btrfs_file_extent_item *extent;
 908         int type;
 909         int ret;
 910         u64 ino = btrfs_ino(BTRFS_I(inode));
 911
 912         path = btrfs_alloc_path();
 913         if (!path)
 914                 return -ENOMEM;
 915
 916         min_key.objectid = ino;
 917         min_key.type = BTRFS_EXTENT_DATA_KEY;
 918         min_key.offset = *off;
 919
 920         while (1) {
 921                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
 922                 if (ret != 0)
 923                         goto none;
 924 process_slot:
 925                 if (min_key.objectid != ino)
 926                         goto none;
 927                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
 928                         goto none;
 929
 930                 leaf = path->nodes[0];
 931                 extent = btrfs_item_ptr(leaf, path->slots[0],
 932                                         struct btrfs_file_extent_item);
 933
 934                 type = btrfs_file_extent_type(leaf, extent);
 935                 if (type == BTRFS_FILE_EXTENT_REG &&
 936                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
 937                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
 938                         *off = min_key.offset;
 939                         btrfs_free_path(path);
 940                         return 0;
 941                 }
 942
 943                 path->slots[0]++;
 944                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
 945                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
 946                         goto process_slot;
 947                 }
 948
 949                 if (min_key.offset == (u64)-1)
 950                         goto none;
 951
 952                 min_key.offset++;
 953                 btrfs_release_path(path);
 954         }
 955 none:
 956         btrfs_free_path(path);
 957         return -ENOENT;
 958 }
 959
 960 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
 961 {
 962         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 963         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 964         struct extent_map *em;
 965         u64 len = PAGE_SIZE;
 966
 967         /*
 968          * hopefully we have this extent in the tree already, try without
 969          * the full extent lock
 970          */
 971         read_lock(&em_tree->lock);
 972         em = lookup_extent_mapping(em_tree, start, len);
 973         read_unlock(&em_tree->lock);
 974
 975         if (!em) {
 976                 struct extent_state *cached = NULL;
 977                 u64 end = start + len - 1;
 978
 979                 /* get the big lock and read metadata off disk */
 980                 lock_extent_bits(io_tree, start, end, &cached);
 981                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
 982                 unlock_extent_cached(io_tree, start, end, &cached, GFP_NOFS);
 983
 984                 if (IS_ERR(em))
 985                         return NULL;
 986         }
 987
 988         return em;
 989 }
 990
 991 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
 992 {
 993         struct extent_map *next;
 994         bool ret = true;
 995
 996         /* this is the last extent */
 997         if (em->start + em->len >= i_size_read(inode))
 998                 return false;
 999
1000         next = defrag_lookup_extent(inode, em->start + em->len);
1001         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1002                 ret = false;
1003         else if ((em->block_start + em->block_len == next->block_start) &&
1004                  (em->block_len > SZ_128K && next->block_len > SZ_128K))
1005                 ret = false;
1006
1007         free_extent_map(next);
1008         return ret;
1009 }
1010
1011 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1012                                u64 *last_len, u64 *skip, u64 *defrag_end,
1013                                int compress)
1014 {
1015         struct extent_map *em;
1016         int ret = 1;
1017         bool next_mergeable = true;
1018         bool prev_mergeable = true;
1019
1020         /*
1021          * make sure that once we start defragging an extent, we keep on
1022          * defragging it
1023          */
1024         if (start < *defrag_end)
1025                 return 1;
1026
1027         *skip = 0;
1028
1029         em = defrag_lookup_extent(inode, start);
1030         if (!em)
1031                 return 0;
1032
1033         /* this will cover holes, and inline extents */
1034         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1035                 ret = 0;
1036                 goto out;
1037         }
1038
1039         if (!*defrag_end)
1040                 prev_mergeable = false;
1041
1042         next_mergeable = defrag_check_next_extent(inode, em);
1043         /*
1044          * we hit a real extent, if it is big or the next extent is not a
1045          * real extent, don't bother defragging it
1046          */
1047         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1048             (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1049                 ret = 0;
1050 out:
1051         /*
1052          * last_len ends up being a counter of how many bytes we've defragged.
1053          * every time we choose not to defrag an extent, we reset *last_len
1054          * so that the next tiny extent will force a defrag.
1055          *
1056          * The end result of this is that tiny extents before a single big
1057          * extent will force at least part of that big extent to be defragged.
1058          */
1059         if (ret) {
1060                 *defrag_end = extent_map_end(em);
1061         } else {
1062                 *last_len = 0;
1063                 *skip = extent_map_end(em);
1064                 *defrag_end = 0;
1065         }
1066
1067         free_extent_map(em);
1068         return ret;
1069 }
1070
1071 /*
1072  * it doesn't do much good to defrag one or two pages
1073  * at a time.  This pulls in a nice chunk of pages
1074  * to COW and defrag.
1075  *
1076  * It also makes sure the delalloc code has enough
1077  * dirty data to avoid making new small extents as part
1078  * of the defrag
1079  *
1080  * It's a good idea to start RA on this range
1081  * before calling this.
1082  */
1083 static int cluster_pages_for_defrag(struct inode *inode,
1084                                     struct page **pages,
1085                                     unsigned long start_index,
1086                                     unsigned long num_pages)
1087 {
1088         unsigned long file_end;
1089         u64 isize = i_size_read(inode);
1090         u64 page_start;
1091         u64 page_end;
1092         u64 page_cnt;
1093         int ret;
1094         int i;
1095         int i_done;
1096         struct btrfs_ordered_extent *ordered;
1097         struct extent_state *cached_state = NULL;
1098         struct extent_io_tree *tree;
1099         struct extent_changeset *data_reserved = NULL;
1100         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1101
1102         file_end = (isize - 1) >> PAGE_SHIFT;
1103         if (!isize || start_index > file_end)
1104                 return 0;
1105
1106         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1107
1108         ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
1109                         start_index << PAGE_SHIFT,
1110                         page_cnt << PAGE_SHIFT);
1111         if (ret)
1112                 return ret;
1113         i_done = 0;
1114         tree = &BTRFS_I(inode)->io_tree;
1115
1116         /* step one, lock all the pages */
1117         for (i = 0; i < page_cnt; i++) {
1118                 struct page *page;
1119 again:
1120                 page = find_or_create_page(inode->i_mapping,
1121                                            start_index + i, mask);
1122                 if (!page)
1123                         break;
1124
1125                 page_start = page_offset(page);
1126                 page_end = page_start + PAGE_SIZE - 1;
1127                 while (1) {
1128                         lock_extent_bits(tree, page_start, page_end,
1129                                          &cached_state);
1130                         ordered = btrfs_lookup_ordered_extent(inode,
1131                                                               page_start);
1132                         unlock_extent_cached(tree, page_start, page_end,
1133                                              &cached_state, GFP_NOFS);
1134                         if (!ordered)
1135                                 break;
1136
1137                         unlock_page(page);
1138                         btrfs_start_ordered_extent(inode, ordered, 1);
1139                         btrfs_put_ordered_extent(ordered);
1140                         lock_page(page);
1141                         /*
1142                          * we unlocked the page above, so we need check if
1143                          * it was released or not.
1144                          */
1145                         if (page->mapping != inode->i_mapping) {
1146                                 unlock_page(page);
1147                                 put_page(page);
1148                                 goto again;
1149                         }
1150                 }
1151
1152                 if (!PageUptodate(page)) {
1153                         btrfs_readpage(NULL, page);
1154                         lock_page(page);
1155                         if (!PageUptodate(page)) {
1156                                 unlock_page(page);
1157                                 put_page(page);
1158                                 ret = -EIO;
1159                                 break;
1160                         }
1161                 }
1162
1163                 if (page->mapping != inode->i_mapping) {
1164                         unlock_page(page);
1165                         put_page(page);
1166                         goto again;
1167                 }
1168
1169                 pages[i] = page;
1170                 i_done++;
1171         }
1172         if (!i_done || ret)
1173                 goto out;
1174
1175         if (!(inode->i_sb->s_flags & MS_ACTIVE))
1176                 goto out;
1177
1178         /*
1179          * so now we have a nice long stream of locked
1180          * and up to date pages, lets wait on them
1181          */
1182         for (i = 0; i < i_done; i++)
1183                 wait_on_page_writeback(pages[i]);
1184
1185         page_start = page_offset(pages[0]);
1186         page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1187
1188         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1189                          page_start, page_end - 1, &cached_state);
1190         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1191                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1192                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1193                           &cached_state, GFP_NOFS);
1194
1195         if (i_done != page_cnt) {
1196                 spin_lock(&BTRFS_I(inode)->lock);
1197                 BTRFS_I(inode)->outstanding_extents++;
1198                 spin_unlock(&BTRFS_I(inode)->lock);
1199                 btrfs_delalloc_release_space(inode, data_reserved,
1200                                 start_index << PAGE_SHIFT,
1201                                 (page_cnt - i_done) << PAGE_SHIFT);
1202         }
1203
1204
1205         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1206                           &cached_state);
1207
1208         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1209                              page_start, page_end - 1, &cached_state,
1210                              GFP_NOFS);
1211
1212         for (i = 0; i < i_done; i++) {
1213                 clear_page_dirty_for_io(pages[i]);
1214                 ClearPageChecked(pages[i]);
1215                 set_page_extent_mapped(pages[i]);
1216                 set_page_dirty(pages[i]);
1217                 unlock_page(pages[i]);
1218                 put_page(pages[i]);
1219         }
1220         extent_changeset_free(data_reserved);
1221         return i_done;
1222 out:
1223         for (i = 0; i < i_done; i++) {
1224                 unlock_page(pages[i]);
1225                 put_page(pages[i]);
1226         }
1227         btrfs_delalloc_release_space(inode, data_reserved,
1228                         start_index << PAGE_SHIFT,
1229                         page_cnt << PAGE_SHIFT);
1230         extent_changeset_free(data_reserved);
1231         return ret;
1232
1233 }
1234
1235 int btrfs_defrag_file(struct inode *inode, struct file *file,
1236                       struct btrfs_ioctl_defrag_range_args *range,
1237                       u64 newer_than, unsigned long max_to_defrag)
1238 {
1239         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1240         struct btrfs_root *root = BTRFS_I(inode)->root;
1241         struct file_ra_state *ra = NULL;
1242         unsigned long last_index;
1243         u64 isize = i_size_read(inode);
1244         u64 last_len = 0;
1245         u64 skip = 0;
1246         u64 defrag_end = 0;
1247         u64 newer_off = range->start;
1248         unsigned long i;
1249         unsigned long ra_index = 0;
1250         int ret;
1251         int defrag_count = 0;
1252         int compress_type = BTRFS_COMPRESS_ZLIB;
1253         u32 extent_thresh = range->extent_thresh;
1254         unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1255         unsigned long cluster = max_cluster;
1256         u64 new_align = ~((u64)SZ_128K - 1);
1257         struct page **pages = NULL;
1258
1259         if (isize == 0)
1260                 return 0;
1261
1262         if (range->start >= isize)
1263                 return -EINVAL;
1264
1265         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1266                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1267                         return -EINVAL;
1268                 if (range->compress_type)
1269                         compress_type = range->compress_type;
1270         }
1271
1272         if (extent_thresh == 0)
1273                 extent_thresh = SZ_256K;
1274
1275         /*
1276          * If we were not given a file, allocate a readahead context. As
1277          * readahead is just an optimization, defrag will work without it so
1278          * we don't error out.
1279          */
1280         if (!file) {
1281                 ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1282                 if (ra)
1283                         file_ra_state_init(ra, inode->i_mapping);
1284         } else {
1285                 ra = &file->f_ra;
1286         }
1287
1288         pages = kmalloc_array(max_cluster, sizeof(struct page *), GFP_KERNEL);
1289         if (!pages) {
1290                 ret = -ENOMEM;
1291                 goto out_ra;
1292         }
1293
1294         /* find the last page to defrag */
1295         if (range->start + range->len > range->start) {
1296                 last_index = min_t(u64, isize - 1,
1297                          range->start + range->len - 1) >> PAGE_SHIFT;
1298         } else {
1299                 last_index = (isize - 1) >> PAGE_SHIFT;
1300         }
1301
1302         if (newer_than) {
1303                 ret = find_new_extents(root, inode, newer_than,
1304                                        &newer_off, SZ_64K);
1305                 if (!ret) {
1306                         range->start = newer_off;
1307                         /*
1308                          * we always align our defrag to help keep
1309                          * the extents in the file evenly spaced
1310                          */
1311                         i = (newer_off & new_align) >> PAGE_SHIFT;
1312                 } else
1313                         goto out_ra;
1314         } else {
1315                 i = range->start >> PAGE_SHIFT;
1316         }
1317         if (!max_to_defrag)
1318                 max_to_defrag = last_index - i + 1;
1319
1320         /*
1321          * make writeback starts from i, so the defrag range can be
1322          * written sequentially.
1323          */
1324         if (i < inode->i_mapping->writeback_index)
1325                 inode->i_mapping->writeback_index = i;
1326
1327         while (i <= last_index && defrag_count < max_to_defrag &&
1328                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1329                 /*
1330                  * make sure we stop running if someone unmounts
1331                  * the FS
1332                  */
1333                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1334                         break;
1335
1336                 if (btrfs_defrag_cancelled(fs_info)) {
1337                         btrfs_debug(fs_info, "defrag_file cancelled");
1338                         ret = -EAGAIN;
1339                         break;
1340                 }
1341
1342                 if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1343                                          extent_thresh, &last_len, &skip,
1344                                          &defrag_end, range->flags &
1345                                          BTRFS_DEFRAG_RANGE_COMPRESS)) {
1346                         unsigned long next;
1347                         /*
1348                          * the should_defrag function tells us how much to skip
1349                          * bump our counter by the suggested amount
1350                          */
1351                         next = DIV_ROUND_UP(skip, PAGE_SIZE);
1352                         i = max(i + 1, next);
1353                         continue;
1354                 }
1355
1356                 if (!newer_than) {
1357                         cluster = (PAGE_ALIGN(defrag_end) >>
1358                                    PAGE_SHIFT) - i;
1359                         cluster = min(cluster, max_cluster);
1360                 } else {
1361                         cluster = max_cluster;
1362                 }
1363
1364                 if (i + cluster > ra_index) {
1365                         ra_index = max(i, ra_index);
1366                         if (ra)
1367                                 page_cache_sync_readahead(inode->i_mapping, ra,
1368                                                 file, ra_index, cluster);
1369                         ra_index += cluster;
1370                 }
1371
1372                 inode_lock(inode);
1373                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1374                         BTRFS_I(inode)->defrag_compress = compress_type;
1375                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1376                 if (ret < 0) {
1377                         inode_unlock(inode);
1378                         goto out_ra;
1379                 }
1380
1381                 defrag_count += ret;
1382                 balance_dirty_pages_ratelimited(inode->i_mapping);
1383                 inode_unlock(inode);
1384
1385                 if (newer_than) {
1386                         if (newer_off == (u64)-1)
1387                                 break;
1388
1389                         if (ret > 0)
1390                                 i += ret;
1391
1392                         newer_off = max(newer_off + 1,
1393                                         (u64)i << PAGE_SHIFT);
1394
1395                         ret = find_new_extents(root, inode, newer_than,
1396                                                &newer_off, SZ_64K);
1397                         if (!ret) {
1398                                 range->start = newer_off;
1399                                 i = (newer_off & new_align) >> PAGE_SHIFT;
1400                         } else {
1401                                 break;
1402                         }
1403                 } else {
1404                         if (ret > 0) {
1405                                 i += ret;
1406                                 last_len += ret << PAGE_SHIFT;
1407                         } else {
1408                                 i++;
1409                                 last_len = 0;
1410                         }
1411                 }
1412         }
1413
1414         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1415                 filemap_flush(inode->i_mapping);
1416                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1417                              &BTRFS_I(inode)->runtime_flags))
1418                         filemap_flush(inode->i_mapping);
1419         }
1420
1421         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1422                 /* the filemap_flush will queue IO into the worker threads, but
1423                  * we have to make sure the IO is actually started and that
1424                  * ordered extents get created before we return
1425                  */
1426                 atomic_inc(&fs_info->async_submit_draining);
1427                 while (atomic_read(&fs_info->nr_async_submits) ||
1428                        atomic_read(&fs_info->async_delalloc_pages)) {
1429                         wait_event(fs_info->async_submit_wait,
1430                                    (atomic_read(&fs_info->nr_async_submits) == 0 &&
1431                                     atomic_read(&fs_info->async_delalloc_pages) == 0));
1432                 }
1433                 atomic_dec(&fs_info->async_submit_draining);
1434         }
1435
1436         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1437                 btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1438         }
1439
1440         ret = defrag_count;
1441
1442 out_ra:
1443         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1444                 inode_lock(inode);
1445                 BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1446                 inode_unlock(inode);
1447         }
1448         if (!file)
1449                 kfree(ra);
1450         kfree(pages);
1451         return ret;
1452 }
1453
1454 static noinline int btrfs_ioctl_resize(struct file *file,
1455                                         void __user *arg)
1456 {
1457         struct inode *inode = file_inode(file);
1458         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1459         u64 new_size;
1460         u64 old_size;
1461         u64 devid = 1;
1462         struct btrfs_root *root = BTRFS_I(inode)->root;
1463         struct btrfs_ioctl_vol_args *vol_args;
1464         struct btrfs_trans_handle *trans;
1465         struct btrfs_device *device = NULL;
1466         char *sizestr;
1467         char *retptr;
1468         char *devstr = NULL;
1469         int ret = 0;
1470         int mod = 0;
1471
1472         if (!capable(CAP_SYS_ADMIN))
1473                 return -EPERM;
1474
1475         ret = mnt_want_write_file(file);
1476         if (ret)
1477                 return ret;
1478
1479         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
1480                 mnt_drop_write_file(file);
1481                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1482         }
1483
1484         mutex_lock(&fs_info->volume_mutex);
1485         vol_args = memdup_user(arg, sizeof(*vol_args));
1486         if (IS_ERR(vol_args)) {
1487                 ret = PTR_ERR(vol_args);
1488                 goto out;
1489         }
1490
1491         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1492
1493         sizestr = vol_args->name;
1494         devstr = strchr(sizestr, ':');
1495         if (devstr) {
1496                 sizestr = devstr + 1;
1497                 *devstr = '\0';
1498                 devstr = vol_args->name;
1499                 ret = kstrtoull(devstr, 10, &devid);
1500                 if (ret)
1501                         goto out_free;
1502                 if (!devid) {
1503                         ret = -EINVAL;
1504                         goto out_free;
1505                 }
1506                 btrfs_info(fs_info, "resizing devid %llu", devid);
1507         }
1508
1509         device = btrfs_find_device(fs_info, devid, NULL, NULL);
1510         if (!device) {
1511                 btrfs_info(fs_info, "resizer unable to find device %llu",
1512                            devid);
1513                 ret = -ENODEV;
1514                 goto out_free;
1515         }
1516
1517         if (!device->writeable) {
1518                 btrfs_info(fs_info,
1519                            "resizer unable to apply on readonly device %llu",
1520                        devid);
1521                 ret = -EPERM;
1522                 goto out_free;
1523         }
1524
1525         if (!strcmp(sizestr, "max"))
1526                 new_size = device->bdev->bd_inode->i_size;
1527         else {
1528                 if (sizestr[0] == '-') {
1529                         mod = -1;
1530                         sizestr++;
1531                 } else if (sizestr[0] == '+') {
1532                         mod = 1;
1533                         sizestr++;
1534                 }
1535                 new_size = memparse(sizestr, &retptr);
1536                 if (*retptr != '\0' || new_size == 0) {
1537                         ret = -EINVAL;
1538                         goto out_free;
1539                 }
1540         }
1541
1542         if (device->is_tgtdev_for_dev_replace) {
1543                 ret = -EPERM;
1544                 goto out_free;
1545         }
1546
1547         old_size = btrfs_device_get_total_bytes(device);
1548
1549         if (mod < 0) {
1550                 if (new_size > old_size) {
1551                         ret = -EINVAL;
1552                         goto out_free;
1553                 }
1554                 new_size = old_size - new_size;
1555         } else if (mod > 0) {
1556                 if (new_size > ULLONG_MAX - old_size) {
1557                         ret = -ERANGE;
1558                         goto out_free;
1559                 }
1560                 new_size = old_size + new_size;
1561         }
1562
1563         if (new_size < SZ_256M) {
1564                 ret = -EINVAL;
1565                 goto out_free;
1566         }
1567         if (new_size > device->bdev->bd_inode->i_size) {
1568                 ret = -EFBIG;
1569                 goto out_free;
1570         }
1571
1572         new_size = round_down(new_size, fs_info->sectorsize);
1573
1574         btrfs_info_in_rcu(fs_info, "new size for %s is %llu",
1575                           rcu_str_deref(device->name), new_size);
1576
1577         if (new_size > old_size) {
1578                 trans = btrfs_start_transaction(root, 0);
1579                 if (IS_ERR(trans)) {
1580                         ret = PTR_ERR(trans);
1581                         goto out_free;
1582                 }
1583                 ret = btrfs_grow_device(trans, device, new_size);
1584                 btrfs_commit_transaction(trans);
1585         } else if (new_size < old_size) {
1586                 ret = btrfs_shrink_device(device, new_size);
1587         } /* equal, nothing need to do */
1588
1589 out_free:
1590         kfree(vol_args);
1591 out:
1592         mutex_unlock(&fs_info->volume_mutex);
1593         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
1594         mnt_drop_write_file(file);
1595         return ret;
1596 }
1597
1598 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1599                                 const char *name, unsigned long fd, int subvol,
1600                                 u64 *transid, bool readonly,
1601                                 struct btrfs_qgroup_inherit *inherit)
1602 {
1603         int namelen;
1604         int ret = 0;
1605
1606         if (!S_ISDIR(file_inode(file)->i_mode))
1607                 return -ENOTDIR;
1608
1609         ret = mnt_want_write_file(file);
1610         if (ret)
1611                 goto out;
1612
1613         namelen = strlen(name);
1614         if (strchr(name, '/')) {
1615                 ret = -EINVAL;
1616                 goto out_drop_write;
1617         }
1618
1619         if (name[0] == '.' &&
1620            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1621                 ret = -EEXIST;
1622                 goto out_drop_write;
1623         }
1624
1625         if (subvol) {
1626                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1627                                      NULL, transid, readonly, inherit);
1628         } else {
1629                 struct fd src = fdget(fd);
1630                 struct inode *src_inode;
1631                 if (!src.file) {
1632                         ret = -EINVAL;
1633                         goto out_drop_write;
1634                 }
1635
1636                 src_inode = file_inode(src.file);
1637                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1638                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1639                                    "Snapshot src from another FS");
1640                         ret = -EXDEV;
1641                 } else if (!inode_owner_or_capable(src_inode)) {
1642                         /*
1643                          * Subvolume creation is not restricted, but snapshots
1644                          * are limited to own subvolumes only
1645                          */
1646                         ret = -EPERM;
1647                 } else {
1648                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1649                                              BTRFS_I(src_inode)->root,
1650                                              transid, readonly, inherit);
1651                 }
1652                 fdput(src);
1653         }
1654 out_drop_write:
1655         mnt_drop_write_file(file);
1656 out:
1657         return ret;
1658 }
1659
1660 static noinline int btrfs_ioctl_snap_create(struct file *file,
1661                                             void __user *arg, int subvol)
1662 {
1663         struct btrfs_ioctl_vol_args *vol_args;
1664         int ret;
1665
1666         if (!S_ISDIR(file_inode(file)->i_mode))
1667                 return -ENOTDIR;
1668
1669         vol_args = memdup_user(arg, sizeof(*vol_args));
1670         if (IS_ERR(vol_args))
1671                 return PTR_ERR(vol_args);
1672         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1673
1674         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1675                                               vol_args->fd, subvol,
1676                                               NULL, false, NULL);
1677
1678         kfree(vol_args);
1679         return ret;
1680 }
1681
1682 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1683                                                void __user *arg, int subvol)
1684 {
1685         struct btrfs_ioctl_vol_args_v2 *vol_args;
1686         int ret;
1687         u64 transid = 0;
1688         u64 *ptr = NULL;
1689         bool readonly = false;
1690         struct btrfs_qgroup_inherit *inherit = NULL;
1691
1692         if (!S_ISDIR(file_inode(file)->i_mode))
1693                 return -ENOTDIR;
1694
1695         vol_args = memdup_user(arg, sizeof(*vol_args));
1696         if (IS_ERR(vol_args))
1697                 return PTR_ERR(vol_args);
1698         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1699
1700         if (vol_args->flags &
1701             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1702               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1703                 ret = -EOPNOTSUPP;
1704                 goto free_args;
1705         }
1706
1707         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1708                 ptr = &transid;
1709         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1710                 readonly = true;
1711         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1712                 if (vol_args->size > PAGE_SIZE) {
1713                         ret = -EINVAL;
1714                         goto free_args;
1715                 }
1716                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1717                 if (IS_ERR(inherit)) {
1718                         ret = PTR_ERR(inherit);
1719                         goto free_args;
1720                 }
1721         }
1722
1723         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1724                                               vol_args->fd, subvol, ptr,
1725                                               readonly, inherit);
1726         if (ret)
1727                 goto free_inherit;
1728
1729         if (ptr && copy_to_user(arg +
1730                                 offsetof(struct btrfs_ioctl_vol_args_v2,
1731                                         transid),
1732                                 ptr, sizeof(*ptr)))
1733                 ret = -EFAULT;
1734
1735 free_inherit:
1736         kfree(inherit);
1737 free_args:
1738         kfree(vol_args);
1739         return ret;
1740 }
1741
1742 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1743                                                 void __user *arg)
1744 {
1745         struct inode *inode = file_inode(file);
1746         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1747         struct btrfs_root *root = BTRFS_I(inode)->root;
1748         int ret = 0;
1749         u64 flags = 0;
1750
1751         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
1752                 return -EINVAL;
1753
1754         down_read(&fs_info->subvol_sem);
1755         if (btrfs_root_readonly(root))
1756                 flags |= BTRFS_SUBVOL_RDONLY;
1757         up_read(&fs_info->subvol_sem);
1758
1759         if (copy_to_user(arg, &flags, sizeof(flags)))
1760                 ret = -EFAULT;
1761
1762         return ret;
1763 }
1764
1765 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1766                                               void __user *arg)
1767 {
1768         struct inode *inode = file_inode(file);
1769         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1770         struct btrfs_root *root = BTRFS_I(inode)->root;
1771         struct btrfs_trans_handle *trans;
1772         u64 root_flags;
1773         u64 flags;
1774         int ret = 0;
1775
1776         if (!inode_owner_or_capable(inode))
1777                 return -EPERM;
1778
1779         ret = mnt_want_write_file(file);
1780         if (ret)
1781                 goto out;
1782
1783         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
1784                 ret = -EINVAL;
1785                 goto out_drop_write;
1786         }
1787
1788         if (copy_from_user(&flags, arg, sizeof(flags))) {
1789                 ret = -EFAULT;
1790                 goto out_drop_write;
1791         }
1792
1793         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1794                 ret = -EINVAL;
1795                 goto out_drop_write;
1796         }
1797
1798         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1799                 ret = -EOPNOTSUPP;
1800                 goto out_drop_write;
1801         }
1802
1803         down_write(&fs_info->subvol_sem);
1804
1805         /* nothing to do */
1806         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1807                 goto out_drop_sem;
1808
1809         root_flags = btrfs_root_flags(&root->root_item);
1810         if (flags & BTRFS_SUBVOL_RDONLY) {
1811                 btrfs_set_root_flags(&root->root_item,
1812                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1813         } else {
1814                 /*
1815                  * Block RO -> RW transition if this subvolume is involved in
1816                  * send
1817                  */
1818                 spin_lock(&root->root_item_lock);
1819                 if (root->send_in_progress == 0) {
1820                         btrfs_set_root_flags(&root->root_item,
1821                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1822                         spin_unlock(&root->root_item_lock);
1823                 } else {
1824                         spin_unlock(&root->root_item_lock);
1825                         btrfs_warn(fs_info,
1826                                    "Attempt to set subvolume %llu read-write during send",
1827                                    root->root_key.objectid);
1828                         ret = -EPERM;
1829                         goto out_drop_sem;
1830                 }
1831         }
1832
1833         trans = btrfs_start_transaction(root, 1);
1834         if (IS_ERR(trans)) {
1835                 ret = PTR_ERR(trans);
1836                 goto out_reset;
1837         }
1838
1839         ret = btrfs_update_root(trans, fs_info->tree_root,
1840                                 &root->root_key, &root->root_item);
1841
1842         btrfs_commit_transaction(trans);
1843 out_reset:
1844         if (ret)
1845                 btrfs_set_root_flags(&root->root_item, root_flags);
1846 out_drop_sem:
1847         up_write(&fs_info->subvol_sem);
1848 out_drop_write:
1849         mnt_drop_write_file(file);
1850 out:
1851         return ret;
1852 }
1853
1854 /*
1855  * helper to check if the subvolume references other subvolumes
1856  */
1857 static noinline int may_destroy_subvol(struct btrfs_root *root)
1858 {
1859         struct btrfs_fs_info *fs_info = root->fs_info;
1860         struct btrfs_path *path;
1861         struct btrfs_dir_item *di;
1862         struct btrfs_key key;
1863         u64 dir_id;
1864         int ret;
1865
1866         path = btrfs_alloc_path();
1867         if (!path)
1868                 return -ENOMEM;
1869
1870         /* Make sure this root isn't set as the default subvol */
1871         dir_id = btrfs_super_root_dir(fs_info->super_copy);
1872         di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
1873                                    dir_id, "default", 7, 0);
1874         if (di && !IS_ERR(di)) {
1875                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1876                 if (key.objectid == root->root_key.objectid) {
1877                         ret = -EPERM;
1878                         btrfs_err(fs_info,
1879                                   "deleting default subvolume %llu is not allowed",
1880                                   key.objectid);
1881                         goto out;
1882                 }
1883                 btrfs_release_path(path);
1884         }
1885
1886         key.objectid = root->root_key.objectid;
1887         key.type = BTRFS_ROOT_REF_KEY;
1888         key.offset = (u64)-1;
1889
1890         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1891         if (ret < 0)
1892                 goto out;
1893         BUG_ON(ret == 0);
1894
1895         ret = 0;
1896         if (path->slots[0] > 0) {
1897                 path->slots[0]--;
1898                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1899                 if (key.objectid == root->root_key.objectid &&
1900                     key.type == BTRFS_ROOT_REF_KEY)
1901                         ret = -ENOTEMPTY;
1902         }
1903 out:
1904         btrfs_free_path(path);
1905         return ret;
1906 }
1907
1908 static noinline int key_in_sk(struct btrfs_key *key,
1909                               struct btrfs_ioctl_search_key *sk)
1910 {
1911         struct btrfs_key test;
1912         int ret;
1913
1914         test.objectid = sk->min_objectid;
1915         test.type = sk->min_type;
1916         test.offset = sk->min_offset;
1917
1918         ret = btrfs_comp_cpu_keys(key, &test);
1919         if (ret < 0)
1920                 return 0;
1921
1922         test.objectid = sk->max_objectid;
1923         test.type = sk->max_type;
1924         test.offset = sk->max_offset;
1925
1926         ret = btrfs_comp_cpu_keys(key, &test);
1927         if (ret > 0)
1928                 return 0;
1929         return 1;
1930 }
1931
1932 static noinline int copy_to_sk(struct btrfs_path *path,
1933                                struct btrfs_key *key,
1934                                struct btrfs_ioctl_search_key *sk,
1935                                size_t *buf_size,
1936                                char __user *ubuf,
1937                                unsigned long *sk_offset,
1938                                int *num_found)
1939 {
1940         u64 found_transid;
1941         struct extent_buffer *leaf;
1942         struct btrfs_ioctl_search_header sh;
1943         struct btrfs_key test;
1944         unsigned long item_off;
1945         unsigned long item_len;
1946         int nritems;
1947         int i;
1948         int slot;
1949         int ret = 0;
1950
1951         leaf = path->nodes[0];
1952         slot = path->slots[0];
1953         nritems = btrfs_header_nritems(leaf);
1954
1955         if (btrfs_header_generation(leaf) > sk->max_transid) {
1956                 i = nritems;
1957                 goto advance_key;
1958         }
1959         found_transid = btrfs_header_generation(leaf);
1960
1961         for (i = slot; i < nritems; i++) {
1962                 item_off = btrfs_item_ptr_offset(leaf, i);
1963                 item_len = btrfs_item_size_nr(leaf, i);
1964
1965                 btrfs_item_key_to_cpu(leaf, key, i);
1966                 if (!key_in_sk(key, sk))
1967                         continue;
1968
1969                 if (sizeof(sh) + item_len > *buf_size) {
1970                         if (*num_found) {
1971                                 ret = 1;
1972                                 goto out;
1973                         }
1974
1975                         /*
1976                          * return one empty item back for v1, which does not
1977                          * handle -EOVERFLOW
1978                          */
1979
1980                         *buf_size = sizeof(sh) + item_len;
1981                         item_len = 0;
1982                         ret = -EOVERFLOW;
1983                 }
1984
1985                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
1986                         ret = 1;
1987                         goto out;
1988                 }
1989
1990                 sh.objectid = key->objectid;
1991                 sh.offset = key->offset;
1992                 sh.type = key->type;
1993                 sh.len = item_len;
1994                 sh.transid = found_transid;
1995
1996                 /* copy search result header */
1997                 if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
1998                         ret = -EFAULT;
1999                         goto out;
2000                 }
2001
2002                 *sk_offset += sizeof(sh);
2003
2004                 if (item_len) {
2005                         char __user *up = ubuf + *sk_offset;
2006                         /* copy the item */
2007                         if (read_extent_buffer_to_user(leaf, up,
2008                                                        item_off, item_len)) {
2009                                 ret = -EFAULT;
2010                                 goto out;
2011                         }
2012
2013                         *sk_offset += item_len;
2014                 }
2015                 (*num_found)++;
2016
2017                 if (ret) /* -EOVERFLOW from above */
2018                         goto out;
2019
2020                 if (*num_found >= sk->nr_items) {
2021                         ret = 1;
2022                         goto out;
2023                 }
2024         }
2025 advance_key:
2026         ret = 0;
2027         test.objectid = sk->max_objectid;
2028         test.type = sk->max_type;
2029         test.offset = sk->max_offset;
2030         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2031                 ret = 1;
2032         else if (key->offset < (u64)-1)
2033                 key->offset++;
2034         else if (key->type < (u8)-1) {
2035                 key->offset = 0;
2036                 key->type++;
2037         } else if (key->objectid < (u64)-1) {
2038                 key->offset = 0;
2039                 key->type = 0;
2040                 key->objectid++;
2041         } else
2042                 ret = 1;
2043 out:
2044         /*
2045          *  0: all items from this leaf copied, continue with next
2046          *  1: * more items can be copied, but unused buffer is too small
2047          *     * all items were found
2048          *     Either way, it will stops the loop which iterates to the next
2049          *     leaf
2050          *  -EOVERFLOW: item was to large for buffer
2051          *  -EFAULT: could not copy extent buffer back to userspace
2052          */
2053         return ret;
2054 }
2055
2056 static noinline int search_ioctl(struct inode *inode,
2057                                  struct btrfs_ioctl_search_key *sk,
2058                                  size_t *buf_size,
2059                                  char __user *ubuf)
2060 {
2061         struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2062         struct btrfs_root *root;
2063         struct btrfs_key key;
2064         struct btrfs_path *path;
2065         int ret;
2066         int num_found = 0;
2067         unsigned long sk_offset = 0;
2068
2069         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2070                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2071                 return -EOVERFLOW;
2072         }
2073
2074         path = btrfs_alloc_path();
2075         if (!path)
2076                 return -ENOMEM;
2077
2078         if (sk->tree_id == 0) {
2079                 /* search the root of the inode that was passed */
2080                 root = BTRFS_I(inode)->root;
2081         } else {
2082                 key.objectid = sk->tree_id;
2083                 key.type = BTRFS_ROOT_ITEM_KEY;
2084                 key.offset = (u64)-1;
2085                 root = btrfs_read_fs_root_no_name(info, &key);
2086                 if (IS_ERR(root)) {
2087                         btrfs_free_path(path);
2088                         return -ENOENT;
2089                 }
2090         }
2091
2092         key.objectid = sk->min_objectid;
2093         key.type = sk->min_type;
2094         key.offset = sk->min_offset;
2095
2096         while (1) {
2097                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2098                 if (ret != 0) {
2099                         if (ret > 0)
2100                                 ret = 0;
2101                         goto err;
2102                 }
2103                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2104                                  &sk_offset, &num_found);
2105                 btrfs_release_path(path);
2106                 if (ret)
2107                         break;
2108
2109         }
2110         if (ret > 0)
2111                 ret = 0;
2112 err:
2113         sk->nr_items = num_found;
2114         btrfs_free_path(path);
2115         return ret;
2116 }
2117
2118 static noinline int btrfs_ioctl_tree_search(struct file *file,
2119                                            void __user *argp)
2120 {
2121         struct btrfs_ioctl_search_args __user *uargs;
2122         struct btrfs_ioctl_search_key sk;
2123         struct inode *inode;
2124         int ret;
2125         size_t buf_size;
2126
2127         if (!capable(CAP_SYS_ADMIN))
2128                 return -EPERM;
2129
2130         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2131
2132         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2133                 return -EFAULT;
2134
2135         buf_size = sizeof(uargs->buf);
2136
2137         inode = file_inode(file);
2138         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2139
2140         /*
2141          * In the origin implementation an overflow is handled by returning a
2142          * search header with a len of zero, so reset ret.
2143          */
2144         if (ret == -EOVERFLOW)
2145                 ret = 0;
2146
2147         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2148                 ret = -EFAULT;
2149         return ret;
2150 }
2151
2152 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2153                                                void __user *argp)
2154 {
2155         struct btrfs_ioctl_search_args_v2 __user *uarg;
2156         struct btrfs_ioctl_search_args_v2 args;
2157         struct inode *inode;
2158         int ret;
2159         size_t buf_size;
2160         const size_t buf_limit = SZ_16M;
2161
2162         if (!capable(CAP_SYS_ADMIN))
2163                 return -EPERM;
2164
2165         /* copy search header and buffer size */
2166         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2167         if (copy_from_user(&args, uarg, sizeof(args)))
2168                 return -EFAULT;
2169
2170         buf_size = args.buf_size;
2171
2172         if (buf_size < sizeof(struct btrfs_ioctl_search_header))
2173                 return -EOVERFLOW;
2174
2175         /* limit result size to 16MB */
2176         if (buf_size > buf_limit)
2177                 buf_size = buf_limit;
2178
2179         inode = file_inode(file);
2180         ret = search_ioctl(inode, &args.key, &buf_size,
2181                            (char *)(&uarg->buf[0]));
2182         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2183                 ret = -EFAULT;
2184         else if (ret == -EOVERFLOW &&
2185                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2186                 ret = -EFAULT;
2187
2188         return ret;
2189 }
2190
2191 /*
2192  * Search INODE_REFs to identify path name of 'dirid' directory
2193  * in a 'tree_id' tree. and sets path name to 'name'.
2194  */
2195 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2196                                 u64 tree_id, u64 dirid, char *name)
2197 {
2198         struct btrfs_root *root;
2199         struct btrfs_key key;
2200         char *ptr;
2201         int ret = -1;
2202         int slot;
2203         int len;
2204         int total_len = 0;
2205         struct btrfs_inode_ref *iref;
2206         struct extent_buffer *l;
2207         struct btrfs_path *path;
2208
2209         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2210                 name[0]='\0';
2211                 return 0;
2212         }
2213
2214         path = btrfs_alloc_path();
2215         if (!path)
2216                 return -ENOMEM;
2217
2218         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
2219
2220         key.objectid = tree_id;
2221         key.type = BTRFS_ROOT_ITEM_KEY;
2222         key.offset = (u64)-1;
2223         root = btrfs_read_fs_root_no_name(info, &key);
2224         if (IS_ERR(root)) {
2225                 btrfs_err(info, "could not find root %llu", tree_id);
2226                 ret = -ENOENT;
2227                 goto out;
2228         }
2229
2230         key.objectid = dirid;
2231         key.type = BTRFS_INODE_REF_KEY;
2232         key.offset = (u64)-1;
2233
2234         while (1) {
2235                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2236                 if (ret < 0)
2237                         goto out;
2238                 else if (ret > 0) {
2239                         ret = btrfs_previous_item(root, path, dirid,
2240                                                   BTRFS_INODE_REF_KEY);
2241                         if (ret < 0)
2242                                 goto out;
2243                         else if (ret > 0) {
2244                                 ret = -ENOENT;
2245                                 goto out;
2246                         }
2247                 }
2248
2249                 l = path->nodes[0];
2250                 slot = path->slots[0];
2251                 btrfs_item_key_to_cpu(l, &key, slot);
2252
2253                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2254                 len = btrfs_inode_ref_name_len(l, iref);
2255                 ptr -= len + 1;
2256                 total_len += len + 1;
2257                 if (ptr < name) {
2258                         ret = -ENAMETOOLONG;
2259                         goto out;
2260                 }
2261
2262                 *(ptr + len) = '/';
2263                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2264
2265                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2266                         break;
2267
2268                 btrfs_release_path(path);
2269                 key.objectid = key.offset;
2270                 key.offset = (u64)-1;
2271                 dirid = key.objectid;
2272         }
2273         memmove(name, ptr, total_len);
2274         name[total_len] = '\0';
2275         ret = 0;
2276 out:
2277         btrfs_free_path(path);
2278         return ret;
2279 }
2280
2281 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2282                                            void __user *argp)
2283 {
2284          struct btrfs_ioctl_ino_lookup_args *args;
2285          struct inode *inode;
2286         int ret = 0;
2287
2288         args = memdup_user(argp, sizeof(*args));
2289         if (IS_ERR(args))
2290                 return PTR_ERR(args);
2291
2292         inode = file_inode(file);
2293
2294         /*
2295          * Unprivileged query to obtain the containing subvolume root id. The
2296          * path is reset so it's consistent with btrfs_search_path_in_tree.
2297          */
2298         if (args->treeid == 0)
2299                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2300
2301         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2302                 args->name[0] = 0;
2303                 goto out;
2304         }
2305
2306         if (!capable(CAP_SYS_ADMIN)) {
2307                 ret = -EPERM;
2308                 goto out;
2309         }
2310
2311         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2312                                         args->treeid, args->objectid,
2313                                         args->name);
2314
2315 out:
2316         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2317                 ret = -EFAULT;
2318
2319         kfree(args);
2320         return ret;
2321 }
2322
2323 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2324                                              void __user *arg)
2325 {
2326         struct dentry *parent = file->f_path.dentry;
2327         struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
2328         struct dentry *dentry;
2329         struct inode *dir = d_inode(parent);
2330         struct inode *inode;
2331         struct btrfs_root *root = BTRFS_I(dir)->root;
2332         struct btrfs_root *dest = NULL;
2333         struct btrfs_ioctl_vol_args *vol_args;
2334         struct btrfs_trans_handle *trans;
2335         struct btrfs_block_rsv block_rsv;
2336         u64 root_flags;
2337         u64 qgroup_reserved;
2338         int namelen;
2339         int ret;
2340         int err = 0;
2341
2342         if (!S_ISDIR(dir->i_mode))
2343                 return -ENOTDIR;
2344
2345         vol_args = memdup_user(arg, sizeof(*vol_args));
2346         if (IS_ERR(vol_args))
2347                 return PTR_ERR(vol_args);
2348
2349         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2350         namelen = strlen(vol_args->name);
2351         if (strchr(vol_args->name, '/') ||
2352             strncmp(vol_args->name, "..", namelen) == 0) {
2353                 err = -EINVAL;
2354                 goto out;
2355         }
2356
2357         err = mnt_want_write_file(file);
2358         if (err)
2359                 goto out;
2360
2361
2362         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
2363         if (err == -EINTR)
2364                 goto out_drop_write;
2365         dentry = lookup_one_len(vol_args->name, parent, namelen);
2366         if (IS_ERR(dentry)) {
2367                 err = PTR_ERR(dentry);
2368                 goto out_unlock_dir;
2369         }
2370
2371         if (d_really_is_negative(dentry)) {
2372                 err = -ENOENT;
2373                 goto out_dput;
2374         }
2375
2376         inode = d_inode(dentry);
2377         dest = BTRFS_I(inode)->root;
2378         if (!capable(CAP_SYS_ADMIN)) {
2379                 /*
2380                  * Regular user.  Only allow this with a special mount
2381                  * option, when the user has write+exec access to the
2382                  * subvol root, and when rmdir(2) would have been
2383                  * allowed.
2384                  *
2385                  * Note that this is _not_ check that the subvol is
2386                  * empty or doesn't contain data that we wouldn't
2387                  * otherwise be able to delete.
2388                  *
2389                  * Users who want to delete empty subvols should try
2390                  * rmdir(2).
2391                  */
2392                 err = -EPERM;
2393                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
2394                         goto out_dput;
2395
2396                 /*
2397                  * Do not allow deletion if the parent dir is the same
2398                  * as the dir to be deleted.  That means the ioctl
2399                  * must be called on the dentry referencing the root
2400                  * of the subvol, not a random directory contained
2401                  * within it.
2402                  */
2403                 err = -EINVAL;
2404                 if (root == dest)
2405                         goto out_dput;
2406
2407                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2408                 if (err)
2409                         goto out_dput;
2410         }
2411
2412         /* check if subvolume may be deleted by a user */
2413         err = btrfs_may_delete(dir, dentry, 1);
2414         if (err)
2415                 goto out_dput;
2416
2417         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
2418                 err = -EINVAL;
2419                 goto out_dput;
2420         }
2421
2422         inode_lock(inode);
2423
2424         /*
2425          * Don't allow to delete a subvolume with send in progress. This is
2426          * inside the i_mutex so the error handling that has to drop the bit
2427          * again is not run concurrently.
2428          */
2429         spin_lock(&dest->root_item_lock);
2430         root_flags = btrfs_root_flags(&dest->root_item);
2431         if (dest->send_in_progress == 0) {
2432                 btrfs_set_root_flags(&dest->root_item,
2433                                 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
2434                 spin_unlock(&dest->root_item_lock);
2435         } else {
2436                 spin_unlock(&dest->root_item_lock);
2437                 btrfs_warn(fs_info,
2438                            "Attempt to delete subvolume %llu during send",
2439                            dest->root_key.objectid);
2440                 err = -EPERM;
2441                 goto out_unlock_inode;
2442         }
2443
2444         down_write(&fs_info->subvol_sem);
2445
2446         err = may_destroy_subvol(dest);
2447         if (err)
2448                 goto out_up_write;
2449
2450         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2451         /*
2452          * One for dir inode, two for dir entries, two for root
2453          * ref/backref.
2454          */
2455         err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2456                                                5, &qgroup_reserved, true);
2457         if (err)
2458                 goto out_up_write;
2459
2460         trans = btrfs_start_transaction(root, 0);
2461         if (IS_ERR(trans)) {
2462                 err = PTR_ERR(trans);
2463                 goto out_release;
2464         }
2465         trans->block_rsv = &block_rsv;
2466         trans->bytes_reserved = block_rsv.size;
2467
2468         btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
2469
2470         ret = btrfs_unlink_subvol(trans, root, dir,
2471                                 dest->root_key.objectid,
2472                                 dentry->d_name.name,
2473                                 dentry->d_name.len);
2474         if (ret) {
2475                 err = ret;
2476                 btrfs_abort_transaction(trans, ret);
2477                 goto out_end_trans;
2478         }
2479
2480         btrfs_record_root_in_trans(trans, dest);
2481
2482         memset(&dest->root_item.drop_progress, 0,
2483                 sizeof(dest->root_item.drop_progress));
2484         dest->root_item.drop_level = 0;
2485         btrfs_set_root_refs(&dest->root_item, 0);
2486
2487         if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
2488                 ret = btrfs_insert_orphan_item(trans,
2489                                         fs_info->tree_root,
2490                                         dest->root_key.objectid);
2491                 if (ret) {
2492                         btrfs_abort_transaction(trans, ret);
2493                         err = ret;
2494                         goto out_end_trans;
2495                 }
2496         }
2497
2498         ret = btrfs_uuid_tree_rem(trans, fs_info, dest->root_item.uuid,
2499                                   BTRFS_UUID_KEY_SUBVOL,
2500                                   dest->root_key.objectid);
2501         if (ret && ret != -ENOENT) {
2502                 btrfs_abort_transaction(trans, ret);
2503                 err = ret;
2504                 goto out_end_trans;
2505         }
2506         if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2507                 ret = btrfs_uuid_tree_rem(trans, fs_info,
2508                                           dest->root_item.received_uuid,
2509                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2510                                           dest->root_key.objectid);
2511                 if (ret && ret != -ENOENT) {
2512                         btrfs_abort_transaction(trans, ret);
2513                         err = ret;
2514                         goto out_end_trans;
2515                 }
2516         }
2517
2518 out_end_trans:
2519         trans->block_rsv = NULL;
2520         trans->bytes_reserved = 0;
2521         ret = btrfs_end_transaction(trans);
2522         if (ret && !err)
2523                 err = ret;
2524         inode->i_flags |= S_DEAD;
2525 out_release:
2526         btrfs_subvolume_release_metadata(fs_info, &block_rsv);
2527 out_up_write:
2528         up_write(&fs_info->subvol_sem);
2529         if (err) {
2530                 spin_lock(&dest->root_item_lock);
2531                 root_flags = btrfs_root_flags(&dest->root_item);
2532                 btrfs_set_root_flags(&dest->root_item,
2533                                 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
2534                 spin_unlock(&dest->root_item_lock);
2535         }
2536 out_unlock_inode:
2537         inode_unlock(inode);
2538         if (!err) {
2539                 d_invalidate(dentry);
2540                 btrfs_invalidate_inodes(dest);
2541                 d_delete(dentry);
2542                 ASSERT(dest->send_in_progress == 0);
2543
2544                 /* the last ref */
2545                 if (dest->ino_cache_inode) {
2546                         iput(dest->ino_cache_inode);
2547                         dest->ino_cache_inode = NULL;
2548                 }
2549         }
2550 out_dput:
2551         dput(dentry);
2552 out_unlock_dir:
2553         inode_unlock(dir);
2554 out_drop_write:
2555         mnt_drop_write_file(file);
2556 out:
2557         kfree(vol_args);
2558         return err;
2559 }
2560
2561 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2562 {
2563         struct inode *inode = file_inode(file);
2564         struct btrfs_root *root = BTRFS_I(inode)->root;
2565         struct btrfs_ioctl_defrag_range_args *range;
2566         int ret;
2567
2568         ret = mnt_want_write_file(file);
2569         if (ret)
2570                 return ret;
2571
2572         if (btrfs_root_readonly(root)) {
2573                 ret = -EROFS;
2574                 goto out;
2575         }
2576
2577         switch (inode->i_mode & S_IFMT) {
2578         case S_IFDIR:
2579                 if (!capable(CAP_SYS_ADMIN)) {
2580                         ret = -EPERM;
2581                         goto out;
2582                 }
2583                 ret = btrfs_defrag_root(root);
2584                 break;
2585         case S_IFREG:
2586                 if (!(file->f_mode & FMODE_WRITE)) {
2587                         ret = -EINVAL;
2588                         goto out;
2589                 }
2590
2591                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2592                 if (!range) {
2593                         ret = -ENOMEM;
2594                         goto out;
2595                 }
2596
2597                 if (argp) {
2598                         if (copy_from_user(range, argp,
2599                                            sizeof(*range))) {
2600                                 ret = -EFAULT;
2601                                 kfree(range);
2602                                 goto out;
2603                         }
2604                         /* compression requires us to start the IO */
2605                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2606                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2607                                 range->extent_thresh = (u32)-1;
2608                         }
2609                 } else {
2610                         /* the rest are all set to zero by kzalloc */
2611                         range->len = (u64)-1;
2612                 }
2613                 ret = btrfs_defrag_file(file_inode(file), file,
2614                                         range, 0, 0);
2615                 if (ret > 0)
2616                         ret = 0;
2617                 kfree(range);
2618                 break;
2619         default:
2620                 ret = -EINVAL;
2621         }
2622 out:
2623         mnt_drop_write_file(file);
2624         return ret;
2625 }
2626
2627 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
2628 {
2629         struct btrfs_ioctl_vol_args *vol_args;
2630         int ret;
2631
2632         if (!capable(CAP_SYS_ADMIN))
2633                 return -EPERM;
2634
2635         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
2636                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2637
2638         mutex_lock(&fs_info->volume_mutex);
2639         vol_args = memdup_user(arg, sizeof(*vol_args));
2640         if (IS_ERR(vol_args)) {
2641                 ret = PTR_ERR(vol_args);
2642                 goto out;
2643         }
2644
2645         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2646         ret = btrfs_init_new_device(fs_info, vol_args->name);
2647
2648         if (!ret)
2649                 btrfs_info(fs_info, "disk added %s", vol_args->name);
2650
2651         kfree(vol_args);
2652 out:
2653         mutex_unlock(&fs_info->volume_mutex);
2654         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
2655         return ret;
2656 }
2657
2658 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
2659 {
2660         struct inode *inode = file_inode(file);
2661         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2662         struct btrfs_ioctl_vol_args_v2 *vol_args;
2663         int ret;
2664
2665         if (!capable(CAP_SYS_ADMIN))
2666                 return -EPERM;
2667
2668         ret = mnt_want_write_file(file);
2669         if (ret)
2670                 return ret;
2671
2672         vol_args = memdup_user(arg, sizeof(*vol_args));
2673         if (IS_ERR(vol_args)) {
2674                 ret = PTR_ERR(vol_args);
2675                 goto err_drop;
2676         }
2677
2678         /* Check for compatibility reject unknown flags */
2679         if (vol_args->flags & ~BTRFS_VOL_ARG_V2_FLAGS_SUPPORTED)
2680                 return -EOPNOTSUPP;
2681
2682         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
2683                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2684                 goto out;
2685         }
2686
2687         mutex_lock(&fs_info->volume_mutex);
2688         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
2689                 ret = btrfs_rm_device(fs_info, NULL, vol_args->devid);
2690         } else {
2691                 vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
2692                 ret = btrfs_rm_device(fs_info, vol_args->name, 0);
2693         }
2694         mutex_unlock(&fs_info->volume_mutex);
2695         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
2696
2697         if (!ret) {
2698                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
2699                         btrfs_info(fs_info, "device deleted: id %llu",
2700                                         vol_args->devid);
2701                 else
2702                         btrfs_info(fs_info, "device deleted: %s",
2703                                         vol_args->name);
2704         }
2705 out:
2706         kfree(vol_args);
2707 err_drop:
2708         mnt_drop_write_file(file);
2709         return ret;
2710 }
2711
2712 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2713 {
2714         struct inode *inode = file_inode(file);
2715         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2716         struct btrfs_ioctl_vol_args *vol_args;
2717         int ret;
2718
2719         if (!capable(CAP_SYS_ADMIN))
2720                 return -EPERM;
2721
2722         ret = mnt_want_write_file(file);
2723         if (ret)
2724                 return ret;
2725
2726         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
2727                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2728                 goto out_drop_write;
2729         }
2730
2731         vol_args = memdup_user(arg, sizeof(*vol_args));
2732         if (IS_ERR(vol_args)) {
2733                 ret = PTR_ERR(vol_args);
2734                 goto out;
2735         }
2736
2737         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2738         mutex_lock(&fs_info->volume_mutex);
2739         ret = btrfs_rm_device(fs_info, vol_args->name, 0);
2740         mutex_unlock(&fs_info->volume_mutex);
2741
2742         if (!ret)
2743                 btrfs_info(fs_info, "disk deleted %s", vol_args->name);
2744         kfree(vol_args);
2745 out:
2746         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
2747 out_drop_write:
2748         mnt_drop_write_file(file);
2749
2750         return ret;
2751 }
2752
2753 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
2754                                 void __user *arg)
2755 {
2756         struct btrfs_ioctl_fs_info_args *fi_args;
2757         struct btrfs_device *device;
2758         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2759         int ret = 0;
2760
2761         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2762         if (!fi_args)
2763                 return -ENOMEM;
2764
2765         mutex_lock(&fs_devices->device_list_mutex);
2766         fi_args->num_devices = fs_devices->num_devices;
2767         memcpy(&fi_args->fsid, fs_info->fsid, sizeof(fi_args->fsid));
2768
2769         list_for_each_entry(device, &fs_devices->devices, dev_list) {
2770                 if (device->devid > fi_args->max_id)
2771                         fi_args->max_id = device->devid;
2772         }
2773         mutex_unlock(&fs_devices->device_list_mutex);
2774
2775         fi_args->nodesize = fs_info->super_copy->nodesize;
2776         fi_args->sectorsize = fs_info->super_copy->sectorsize;
2777         fi_args->clone_alignment = fs_info->super_copy->sectorsize;
2778
2779         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2780                 ret = -EFAULT;
2781
2782         kfree(fi_args);
2783         return ret;
2784 }
2785
2786 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
2787                                  void __user *arg)
2788 {
2789         struct btrfs_ioctl_dev_info_args *di_args;
2790         struct btrfs_device *dev;
2791         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2792         int ret = 0;
2793         char *s_uuid = NULL;
2794
2795         di_args = memdup_user(arg, sizeof(*di_args));
2796         if (IS_ERR(di_args))
2797                 return PTR_ERR(di_args);
2798
2799         if (!btrfs_is_empty_uuid(di_args->uuid))
2800                 s_uuid = di_args->uuid;
2801
2802         mutex_lock(&fs_devices->device_list_mutex);
2803         dev = btrfs_find_device(fs_info, di_args->devid, s_uuid, NULL);
2804
2805         if (!dev) {
2806                 ret = -ENODEV;
2807                 goto out;
2808         }
2809
2810         di_args->devid = dev->devid;
2811         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
2812         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
2813         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2814         if (dev->name) {
2815                 struct rcu_string *name;
2816
2817                 rcu_read_lock();
2818                 name = rcu_dereference(dev->name);
2819                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2820                 rcu_read_unlock();
2821                 di_args->path[sizeof(di_args->path) - 1] = 0;
2822         } else {
2823                 di_args->path[0] = '\0';
2824         }
2825
2826 out:
2827         mutex_unlock(&fs_devices->device_list_mutex);
2828         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2829                 ret = -EFAULT;
2830
2831         kfree(di_args);
2832         return ret;
2833 }
2834
2835 static struct page *extent_same_get_page(struct inode *inode, pgoff_t index)
2836 {
2837         struct page *page;
2838
2839         page = grab_cache_page(inode->i_mapping, index);
2840         if (!page)
2841                 return ERR_PTR(-ENOMEM);
2842
2843         if (!PageUptodate(page)) {
2844                 int ret;
2845
2846                 ret = btrfs_readpage(NULL, page);
2847                 if (ret)
2848                         return ERR_PTR(ret);
2849                 lock_page(page);
2850                 if (!PageUptodate(page)) {
2851                         unlock_page(page);
2852                         put_page(page);
2853                         return ERR_PTR(-EIO);
2854                 }
2855                 if (page->mapping != inode->i_mapping) {
2856                         unlock_page(page);
2857                         put_page(page);
2858                         return ERR_PTR(-EAGAIN);
2859                 }
2860         }
2861
2862         return page;
2863 }
2864
2865 static int gather_extent_pages(struct inode *inode, struct page **pages,
2866                                int num_pages, u64 off)
2867 {
2868         int i;
2869         pgoff_t index = off >> PAGE_SHIFT;
2870
2871         for (i = 0; i < num_pages; i++) {
2872 again:
2873                 pages[i] = extent_same_get_page(inode, index + i);
2874                 if (IS_ERR(pages[i])) {
2875                         int err = PTR_ERR(pages[i]);
2876
2877                         if (err == -EAGAIN)
2878                                 goto again;
2879                         pages[i] = NULL;
2880                         return err;
2881                 }
2882         }
2883         return 0;
2884 }
2885
2886 static int lock_extent_range(struct inode *inode, u64 off, u64 len,
2887                              bool retry_range_locking)
2888 {
2889         /*
2890          * Do any pending delalloc/csum calculations on inode, one way or
2891          * another, and lock file content.
2892          * The locking order is:
2893          *
2894          *   1) pages
2895          *   2) range in the inode's io tree
2896          */
2897         while (1) {
2898                 struct btrfs_ordered_extent *ordered;
2899                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2900                 ordered = btrfs_lookup_first_ordered_extent(inode,
2901                                                             off + len - 1);
2902                 if ((!ordered ||
2903                      ordered->file_offset + ordered->len <= off ||
2904                      ordered->file_offset >= off + len) &&
2905                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2906                                     off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
2907                         if (ordered)
2908                                 btrfs_put_ordered_extent(ordered);
2909                         break;
2910                 }
2911                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2912                 if (ordered)
2913                         btrfs_put_ordered_extent(ordered);
2914                 if (!retry_range_locking)
2915                         return -EAGAIN;
2916                 btrfs_wait_ordered_range(inode, off, len);
2917         }
2918         return 0;
2919 }
2920
2921 static void btrfs_double_inode_unlock(struct inode *inode1, struct inode *inode2)
2922 {
2923         inode_unlock(inode1);
2924         inode_unlock(inode2);
2925 }
2926
2927 static void btrfs_double_inode_lock(struct inode *inode1, struct inode *inode2)
2928 {
2929         if (inode1 < inode2)
2930                 swap(inode1, inode2);
2931
2932         inode_lock_nested(inode1, I_MUTEX_PARENT);
2933         inode_lock_nested(inode2, I_MUTEX_CHILD);
2934 }
2935
2936 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
2937                                       struct inode *inode2, u64 loff2, u64 len)
2938 {
2939         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2940         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2941 }
2942
2943 static int btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
2944                                     struct inode *inode2, u64 loff2, u64 len,
2945                                     bool retry_range_locking)
2946 {
2947         int ret;
2948
2949         if (inode1 < inode2) {
2950                 swap(inode1, inode2);
2951                 swap(loff1, loff2);
2952         }
2953         ret = lock_extent_range(inode1, loff1, len, retry_range_locking);
2954         if (ret)
2955                 return ret;
2956         ret = lock_extent_range(inode2, loff2, len, retry_range_locking);
2957         if (ret)
2958                 unlock_extent(&BTRFS_I(inode1)->io_tree, loff1,
2959                               loff1 + len - 1);
2960         return ret;
2961 }
2962
2963 struct cmp_pages {
2964         int             num_pages;
2965         struct page     **src_pages;
2966         struct page     **dst_pages;
2967 };
2968
2969 static void btrfs_cmp_data_free(struct cmp_pages *cmp)
2970 {
2971         int i;
2972         struct page *pg;
2973
2974         for (i = 0; i < cmp->num_pages; i++) {
2975                 pg = cmp->src_pages[i];
2976                 if (pg) {
2977                         unlock_page(pg);
2978                         put_page(pg);
2979                 }
2980                 pg = cmp->dst_pages[i];
2981                 if (pg) {
2982                         unlock_page(pg);
2983                         put_page(pg);
2984                 }
2985         }
2986         kfree(cmp->src_pages);
2987         kfree(cmp->dst_pages);
2988 }
2989
2990 static int btrfs_cmp_data_prepare(struct inode *src, u64 loff,
2991                                   struct inode *dst, u64 dst_loff,
2992                                   u64 len, struct cmp_pages *cmp)
2993 {
2994         int ret;
2995         int num_pages = PAGE_ALIGN(len) >> PAGE_SHIFT;
2996         struct page **src_pgarr, **dst_pgarr;
2997
2998         /*
2999          * We must gather up all the pages before we initiate our
3000          * extent locking. We use an array for the page pointers. Size
3001          * of the array is bounded by len, which is in turn bounded by
3002          * BTRFS_MAX_DEDUPE_LEN.
3003          */
3004         src_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
3005         dst_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
3006         if (!src_pgarr || !dst_pgarr) {
3007                 kfree(src_pgarr);
3008                 kfree(dst_pgarr);
3009                 return -ENOMEM;
3010         }
3011         cmp->num_pages = num_pages;
3012         cmp->src_pages = src_pgarr;
3013         cmp->dst_pages = dst_pgarr;
3014
3015         /*
3016          * If deduping ranges in the same inode, locking rules make it mandatory
3017          * to always lock pages in ascending order to avoid deadlocks with
3018          * concurrent tasks (such as starting writeback/delalloc).
3019          */
3020         if (src == dst && dst_loff < loff) {
3021                 swap(src_pgarr, dst_pgarr);
3022                 swap(loff, dst_loff);
3023         }
3024
3025         ret = gather_extent_pages(src, src_pgarr, cmp->num_pages, loff);
3026         if (ret)
3027                 goto out;
3028
3029         ret = gather_extent_pages(dst, dst_pgarr, cmp->num_pages, dst_loff);
3030
3031 out:
3032         if (ret)
3033                 btrfs_cmp_data_free(cmp);
3034         return 0;
3035 }
3036
3037 static int btrfs_cmp_data(u64 len, struct cmp_pages *cmp)
3038 {
3039         int ret = 0;
3040         int i;
3041         struct page *src_page, *dst_page;
3042         unsigned int cmp_len = PAGE_SIZE;
3043         void *addr, *dst_addr;
3044
3045         i = 0;
3046         while (len) {
3047                 if (len < PAGE_SIZE)
3048                         cmp_len = len;
3049
3050                 BUG_ON(i >= cmp->num_pages);
3051
3052                 src_page = cmp->src_pages[i];
3053                 dst_page = cmp->dst_pages[i];
3054                 ASSERT(PageLocked(src_page));
3055                 ASSERT(PageLocked(dst_page));
3056
3057                 addr = kmap_atomic(src_page);
3058                 dst_addr = kmap_atomic(dst_page);
3059
3060                 flush_dcache_page(src_page);
3061                 flush_dcache_page(dst_page);
3062
3063                 if (memcmp(addr, dst_addr, cmp_len))
3064                         ret = -EBADE;
3065
3066                 kunmap_atomic(addr);
3067                 kunmap_atomic(dst_addr);
3068
3069                 if (ret)
3070                         break;
3071
3072                 len -= cmp_len;
3073                 i++;
3074         }
3075
3076         return ret;
3077 }
3078
3079 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 *plen,
3080                                      u64 olen)
3081 {
3082         u64 len = *plen;
3083         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
3084
3085         if (off + olen > inode->i_size || off + olen < off)
3086                 return -EINVAL;
3087
3088         /* if we extend to eof, continue to block boundary */
3089         if (off + len == inode->i_size)
3090                 *plen = len = ALIGN(inode->i_size, bs) - off;
3091
3092         /* Check that we are block aligned - btrfs_clone() requires this */
3093         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
3094                 return -EINVAL;
3095
3096         return 0;
3097 }
3098
3099 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
3100                              struct inode *dst, u64 dst_loff)
3101 {
3102         int ret;
3103         u64 len = olen;
3104         struct cmp_pages cmp;
3105         bool same_inode = (src == dst);
3106         u64 same_lock_start = 0;
3107         u64 same_lock_len = 0;
3108
3109         if (len == 0)
3110                 return 0;
3111
3112         if (same_inode)
3113                 inode_lock(src);
3114         else
3115                 btrfs_double_inode_lock(src, dst);
3116
3117         ret = extent_same_check_offsets(src, loff, &len, olen);
3118         if (ret)
3119                 goto out_unlock;
3120
3121         ret = extent_same_check_offsets(dst, dst_loff, &len, olen);
3122         if (ret)
3123                 goto out_unlock;
3124
3125         if (same_inode) {
3126                 /*
3127                  * Single inode case wants the same checks, except we
3128                  * don't want our length pushed out past i_size as
3129                  * comparing that data range makes no sense.
3130                  *
3131                  * extent_same_check_offsets() will do this for an
3132                  * unaligned length at i_size, so catch it here and
3133                  * reject the request.
3134                  *
3135                  * This effectively means we require aligned extents
3136                  * for the single-inode case, whereas the other cases
3137                  * allow an unaligned length so long as it ends at
3138                  * i_size.
3139                  */
3140                 if (len != olen) {
3141                         ret = -EINVAL;
3142                         goto out_unlock;
3143                 }
3144
3145                 /* Check for overlapping ranges */
3146                 if (dst_loff + len > loff && dst_loff < loff + len) {
3147                         ret = -EINVAL;
3148                         goto out_unlock;
3149                 }
3150
3151                 same_lock_start = min_t(u64, loff, dst_loff);
3152                 same_lock_len = max_t(u64, loff, dst_loff) + len - same_lock_start;
3153         }
3154
3155         /* don't make the dst file partly checksummed */
3156         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3157             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
3158                 ret = -EINVAL;
3159                 goto out_unlock;
3160         }
3161
3162 again:
3163         ret = btrfs_cmp_data_prepare(src, loff, dst, dst_loff, olen, &cmp);
3164         if (ret)
3165                 goto out_unlock;
3166
3167         if (same_inode)
3168                 ret = lock_extent_range(src, same_lock_start, same_lock_len,
3169                                         false);
3170         else
3171                 ret = btrfs_double_extent_lock(src, loff, dst, dst_loff, len,
3172                                                false);
3173         /*
3174          * If one of the inodes has dirty pages in the respective range or
3175          * ordered extents, we need to flush dellaloc and wait for all ordered
3176          * extents in the range. We must unlock the pages and the ranges in the
3177          * io trees to avoid deadlocks when flushing delalloc (requires locking
3178          * pages) and when waiting for ordered extents to complete (they require
3179          * range locking).
3180          */
3181         if (ret == -EAGAIN) {
3182                 /*
3183                  * Ranges in the io trees already unlocked. Now unlock all
3184                  * pages before waiting for all IO to complete.
3185                  */
3186                 btrfs_cmp_data_free(&cmp);
3187                 if (same_inode) {
3188                         btrfs_wait_ordered_range(src, same_lock_start,
3189                                                  same_lock_len);
3190                 } else {
3191                         btrfs_wait_ordered_range(src, loff, len);
3192                         btrfs_wait_ordered_range(dst, dst_loff, len);
3193                 }
3194                 goto again;
3195         }
3196         ASSERT(ret == 0);
3197         if (WARN_ON(ret)) {
3198                 /* ranges in the io trees already unlocked */
3199                 btrfs_cmp_data_free(&cmp);
3200                 return ret;
3201         }
3202
3203         /* pass original length for comparison so we stay within i_size */
3204         ret = btrfs_cmp_data(olen, &cmp);
3205         if (ret == 0)
3206                 ret = btrfs_clone(src, dst, loff, olen, len, dst_loff, 1);
3207
3208         if (same_inode)
3209                 unlock_extent(&BTRFS_I(src)->io_tree, same_lock_start,
3210                               same_lock_start + same_lock_len - 1);
3211         else
3212                 btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
3213
3214         btrfs_cmp_data_free(&cmp);
3215 out_unlock:
3216         if (same_inode)
3217                 inode_unlock(src);
3218         else
3219                 btrfs_double_inode_unlock(src, dst);
3220
3221         return ret;
3222 }
3223
3224 #define BTRFS_MAX_DEDUPE_LEN    SZ_16M
3225
3226 ssize_t btrfs_dedupe_file_range(struct file *src_file, u64 loff, u64 olen,
3227                                 struct file *dst_file, u64 dst_loff)
3228 {
3229         struct inode *src = file_inode(src_file);
3230         struct inode *dst = file_inode(dst_file);
3231         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
3232         ssize_t res;
3233
3234         if (olen > BTRFS_MAX_DEDUPE_LEN)
3235                 olen = BTRFS_MAX_DEDUPE_LEN;
3236
3237         if (WARN_ON_ONCE(bs < PAGE_SIZE)) {
3238                 /*
3239                  * Btrfs does not support blocksize < page_size. As a
3240                  * result, btrfs_cmp_data() won't correctly handle
3241                  * this situation without an update.
3242                  */
3243                 return -EINVAL;
3244         }
3245
3246         res = btrfs_extent_same(src, loff, olen, dst, dst_loff);
3247         if (res)
3248                 return res;
3249         return olen;
3250 }
3251
3252 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3253                                      struct inode *inode,
3254                                      u64 endoff,
3255                                      const u64 destoff,
3256                                      const u64 olen,
3257                                      int no_time_update)
3258 {
3259         struct btrfs_root *root = BTRFS_I(inode)->root;
3260         int ret;
3261
3262         inode_inc_iversion(inode);
3263         if (!no_time_update)
3264                 inode->i_mtime = inode->i_ctime = current_time(inode);
3265         /*
3266          * We round up to the block size at eof when determining which
3267          * extents to clone above, but shouldn't round up the file size.
3268          */
3269         if (endoff > destoff + olen)
3270                 endoff = destoff + olen;
3271         if (endoff > inode->i_size)
3272                 btrfs_i_size_write(BTRFS_I(inode), endoff);
3273
3274         ret = btrfs_update_inode(trans, root, inode);
3275         if (ret) {
3276                 btrfs_abort_transaction(trans, ret);
3277                 btrfs_end_transaction(trans);
3278                 goto out;
3279         }
3280         ret = btrfs_end_transaction(trans);
3281 out:
3282         return ret;
3283 }
3284
3285 static void clone_update_extent_map(struct btrfs_inode *inode,
3286                                     const struct btrfs_trans_handle *trans,
3287                                     const struct btrfs_path *path,
3288                                     const u64 hole_offset,
3289                                     const u64 hole_len)
3290 {
3291         struct extent_map_tree *em_tree = &inode->extent_tree;
3292         struct extent_map *em;
3293         int ret;
3294
3295         em = alloc_extent_map();
3296         if (!em) {
3297                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
3298                 return;
3299         }
3300
3301         if (path) {
3302                 struct btrfs_file_extent_item *fi;
3303
3304                 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
3305                                     struct btrfs_file_extent_item);
3306                 btrfs_extent_item_to_extent_map(inode, path, fi, false, em);
3307                 em->generation = -1;
3308                 if (btrfs_file_extent_type(path->nodes[0], fi) ==
3309                     BTRFS_FILE_EXTENT_INLINE)
3310                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3311                                         &inode->runtime_flags);
3312         } else {
3313                 em->start = hole_offset;
3314                 em->len = hole_len;
3315                 em->ram_bytes = em->len;
3316                 em->orig_start = hole_offset;
3317                 em->block_start = EXTENT_MAP_HOLE;
3318                 em->block_len = 0;
3319                 em->orig_block_len = 0;
3320                 em->compress_type = BTRFS_COMPRESS_NONE;
3321                 em->generation = trans->transid;
3322         }
3323
3324         while (1) {
3325                 write_lock(&em_tree->lock);
3326                 ret = add_extent_mapping(em_tree, em, 1);
3327                 write_unlock(&em_tree->lock);
3328                 if (ret != -EEXIST) {
3329                         free_extent_map(em);
3330                         break;
3331                 }
3332                 btrfs_drop_extent_cache(inode, em->start,
3333                                         em->start + em->len - 1, 0);
3334         }
3335
3336         if (ret)
3337                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
3338 }
3339
3340 /*
3341  * Make sure we do not end up inserting an inline extent into a file that has
3342  * already other (non-inline) extents. If a file has an inline extent it can
3343  * not have any other extents and the (single) inline extent must start at the
3344  * file offset 0. Failing to respect these rules will lead to file corruption,
3345  * resulting in EIO errors on read/write operations, hitting BUG_ON's in mm, etc
3346  *
3347  * We can have extents that have been already written to disk or we can have
3348  * dirty ranges still in delalloc, in which case the extent maps and items are
3349  * created only when we run delalloc, and the delalloc ranges might fall outside
3350  * the range we are currently locking in the inode's io tree. So we check the
3351  * inode's i_size because of that (i_size updates are done while holding the
3352  * i_mutex, which we are holding here).
3353  * We also check to see if the inode has a size not greater than "datal" but has
3354  * extents beyond it, due to an fallocate with FALLOC_FL_KEEP_SIZE (and we are
3355  * protected against such concurrent fallocate calls by the i_mutex).
3356  *
3357  * If the file has no extents but a size greater than datal, do not allow the
3358  * copy because we would need turn the inline extent into a non-inline one (even
3359  * with NO_HOLES enabled). If we find our destination inode only has one inline
3360  * extent, just overwrite it with the source inline extent if its size is less
3361  * than the source extent's size, or we could copy the source inline extent's
3362  * data into the destination inode's inline extent if the later is greater then
3363  * the former.
3364  */
3365 static int clone_copy_inline_extent(struct inode *dst,
3366                                     struct btrfs_trans_handle *trans,
3367                                     struct btrfs_path *path,
3368                                     struct btrfs_key *new_key,
3369                                     const u64 drop_start,
3370                                     const u64 datal,
3371                                     const u64 skip,
3372                                     const u64 size,
3373                                     char *inline_data)
3374 {
3375         struct btrfs_fs_info *fs_info = btrfs_sb(dst->i_sb);
3376         struct btrfs_root *root = BTRFS_I(dst)->root;
3377         const u64 aligned_end = ALIGN(new_key->offset + datal,
3378                                       fs_info->sectorsize);
3379         int ret;
3380         struct btrfs_key key;
3381
3382         if (new_key->offset > 0)
3383                 return -EOPNOTSUPP;
3384
3385         key.objectid = btrfs_ino(BTRFS_I(dst));
3386         key.type = BTRFS_EXTENT_DATA_KEY;
3387         key.offset = 0;
3388         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3389         if (ret < 0) {
3390                 return ret;
3391         } else if (ret > 0) {
3392                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3393                         ret = btrfs_next_leaf(root, path);
3394                         if (ret < 0)
3395                                 return ret;
3396                         else if (ret > 0)
3397                                 goto copy_inline_extent;
3398                 }
3399                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3400                 if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
3401                     key.type == BTRFS_EXTENT_DATA_KEY) {
3402                         ASSERT(key.offset > 0);
3403                         return -EOPNOTSUPP;
3404                 }
3405         } else if (i_size_read(dst) <= datal) {
3406                 struct btrfs_file_extent_item *ei;
3407                 u64 ext_len;
3408
3409                 /*
3410                  * If the file size is <= datal, make sure there are no other
3411                  * extents following (can happen do to an fallocate call with
3412                  * the flag FALLOC_FL_KEEP_SIZE).
3413                  */
3414                 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3415                                     struct btrfs_file_extent_item);
3416                 /*
3417                  * If it's an inline extent, it can not have other extents
3418                  * following it.
3419                  */
3420                 if (btrfs_file_extent_type(path->nodes[0], ei) ==
3421                     BTRFS_FILE_EXTENT_INLINE)
3422                         goto copy_inline_extent;
3423
3424                 ext_len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
3425                 if (ext_len > aligned_end)
3426                         return -EOPNOTSUPP;
3427
3428                 ret = btrfs_next_item(root, path);
3429                 if (ret < 0) {
3430                         return ret;
3431                 } else if (ret == 0) {
3432                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3433                                               path->slots[0]);
3434                         if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
3435                             key.type == BTRFS_EXTENT_DATA_KEY)
3436                                 return -EOPNOTSUPP;
3437                 }
3438         }
3439
3440 copy_inline_extent:
3441         /*
3442          * We have no extent items, or we have an extent at offset 0 which may
3443          * or may not be inlined. All these cases are dealt the same way.
3444          */
3445         if (i_size_read(dst) > datal) {
3446                 /*
3447                  * If the destination inode has an inline extent...
3448                  * This would require copying the data from the source inline
3449                  * extent into the beginning of the destination's inline extent.
3450                  * But this is really complex, both extents can be compressed
3451                  * or just one of them, which would require decompressing and
3452                  * re-compressing data (which could increase the new compressed
3453                  * size, not allowing the compressed data to fit anymore in an
3454                  * inline extent).
3455                  * So just don't support this case for now (it should be rare,
3456                  * we are not really saving space when cloning inline extents).
3457                  */
3458                 return -EOPNOTSUPP;
3459         }
3460
3461         btrfs_release_path(path);
3462         ret = btrfs_drop_extents(trans, root, dst, drop_start, aligned_end, 1);
3463         if (ret)
3464                 return ret;
3465         ret = btrfs_insert_empty_item(trans, root, path, new_key, size);
3466         if (ret)
3467                 return ret;
3468
3469         if (skip) {
3470                 const u32 start = btrfs_file_extent_calc_inline_size(0);
3471
3472                 memmove(inline_data + start, inline_data + start + skip, datal);
3473         }
3474
3475         write_extent_buffer(path->nodes[0], inline_data,
3476                             btrfs_item_ptr_offset(path->nodes[0],
3477                                                   path->slots[0]),
3478                             size);
3479         inode_add_bytes(dst, datal);
3480
3481         return 0;
3482 }
3483
3484 /**
3485  * btrfs_clone() - clone a range from inode file to another
3486  *
3487  * @src: Inode to clone from
3488  * @inode: Inode to clone to
3489  * @off: Offset within source to start clone from
3490  * @olen: Original length, passed by user, of range to clone
3491  * @olen_aligned: Block-aligned value of olen
3492  * @destoff: Offset within @inode to start clone
3493  * @no_time_update: Whether to update mtime/ctime on the target inode
3494  */
3495 static int btrfs_clone(struct inode *src, struct inode *inode,
3496                        const u64 off, const u64 olen, const u64 olen_aligned,
3497                        const u64 destoff, int no_time_update)
3498 {
3499         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3500         struct btrfs_root *root = BTRFS_I(inode)->root;
3501         struct btrfs_path *path = NULL;
3502         struct extent_buffer *leaf;
3503         struct btrfs_trans_handle *trans;
3504         char *buf = NULL;
3505         struct btrfs_key key;
3506         u32 nritems;
3507         int slot;
3508         int ret;
3509         const u64 len = olen_aligned;
3510         u64 last_dest_end = destoff;
3511
3512         ret = -ENOMEM;
3513         buf = kvmalloc(fs_info->nodesize, GFP_KERNEL);
3514         if (!buf)
3515                 return ret;
3516
3517         path = btrfs_alloc_path();
3518         if (!path) {
3519                 kvfree(buf);
3520                 return ret;
3521         }
3522
3523         path->reada = READA_FORWARD;
3524         /* clone data */
3525         key.objectid = btrfs_ino(BTRFS_I(src));
3526         key.type = BTRFS_EXTENT_DATA_KEY;
3527         key.offset = off;
3528
3529         while (1) {
3530                 u64 next_key_min_offset = key.offset + 1;
3531
3532                 /*
3533                  * note the key will change type as we walk through the
3534                  * tree.
3535                  */
3536                 path->leave_spinning = 1;
3537                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
3538                                 0, 0);
3539                 if (ret < 0)
3540                         goto out;
3541                 /*
3542                  * First search, if no extent item that starts at offset off was
3543                  * found but the previous item is an extent item, it's possible
3544                  * it might overlap our target range, therefore process it.
3545                  */
3546                 if (key.offset == off && ret > 0 && path->slots[0] > 0) {
3547                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3548                                               path->slots[0] - 1);
3549                         if (key.type == BTRFS_EXTENT_DATA_KEY)
3550                                 path->slots[0]--;
3551                 }
3552
3553                 nritems = btrfs_header_nritems(path->nodes[0]);
3554 process_slot:
3555                 if (path->slots[0] >= nritems) {
3556                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
3557                         if (ret < 0)
3558                                 goto out;
3559                         if (ret > 0)
3560                                 break;
3561                         nritems = btrfs_header_nritems(path->nodes[0]);
3562                 }
3563                 leaf = path->nodes[0];
3564                 slot = path->slots[0];
3565
3566                 btrfs_item_key_to_cpu(leaf, &key, slot);
3567                 if (key.type > BTRFS_EXTENT_DATA_KEY ||
3568                     key.objectid != btrfs_ino(BTRFS_I(src)))
3569                         break;
3570
3571                 if (key.type == BTRFS_EXTENT_DATA_KEY) {
3572                         struct btrfs_file_extent_item *extent;
3573                         int type;
3574                         u32 size;
3575                         struct btrfs_key new_key;
3576                         u64 disko = 0, diskl = 0;
3577                         u64 datao = 0, datal = 0;
3578                         u8 comp;
3579                         u64 drop_start;
3580
3581                         extent = btrfs_item_ptr(leaf, slot,
3582                                                 struct btrfs_file_extent_item);
3583                         comp = btrfs_file_extent_compression(leaf, extent);
3584                         type = btrfs_file_extent_type(leaf, extent);
3585                         if (type == BTRFS_FILE_EXTENT_REG ||
3586                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3587                                 disko = btrfs_file_extent_disk_bytenr(leaf,
3588                                                                       extent);
3589                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
3590                                                                  extent);
3591                                 datao = btrfs_file_extent_offset(leaf, extent);
3592                                 datal = btrfs_file_extent_num_bytes(leaf,
3593                                                                     extent);
3594                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3595                                 /* take upper bound, may be compressed */
3596                                 datal = btrfs_file_extent_ram_bytes(leaf,
3597                                                                     extent);
3598                         }
3599
3600                         /*
3601                          * The first search might have left us at an extent
3602                          * item that ends before our target range's start, can
3603                          * happen if we have holes and NO_HOLES feature enabled.
3604                          */
3605                         if (key.offset + datal <= off) {
3606                                 path->slots[0]++;
3607                                 goto process_slot;
3608                         } else if (key.offset >= off + len) {
3609                                 break;
3610                         }
3611                         next_key_min_offset = key.offset + datal;
3612                         size = btrfs_item_size_nr(leaf, slot);
3613                         read_extent_buffer(leaf, buf,
3614                                            btrfs_item_ptr_offset(leaf, slot),
3615                                            size);
3616
3617                         btrfs_release_path(path);
3618                         path->leave_spinning = 0;
3619
3620                         memcpy(&new_key, &key, sizeof(new_key));
3621                         new_key.objectid = btrfs_ino(BTRFS_I(inode));
3622                         if (off <= key.offset)
3623                                 new_key.offset = key.offset + destoff - off;
3624                         else
3625                                 new_key.offset = destoff;
3626
3627                         /*
3628                          * Deal with a hole that doesn't have an extent item
3629                          * that represents it (NO_HOLES feature enabled).
3630                          * This hole is either in the middle of the cloning
3631                          * range or at the beginning (fully overlaps it or
3632                          * partially overlaps it).
3633                          */
3634                         if (new_key.offset != last_dest_end)
3635                                 drop_start = last_dest_end;
3636                         else
3637                                 drop_start = new_key.offset;
3638
3639                         /*
3640                          * 1 - adjusting old extent (we may have to split it)
3641                          * 1 - add new extent
3642                          * 1 - inode update
3643                          */
3644                         trans = btrfs_start_transaction(root, 3);
3645                         if (IS_ERR(trans)) {
3646                                 ret = PTR_ERR(trans);
3647                                 goto out;
3648                         }
3649
3650                         if (type == BTRFS_FILE_EXTENT_REG ||
3651                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3652                                 /*
3653                                  *    a  | --- range to clone ---|  b
3654                                  * | ------------- extent ------------- |
3655                                  */
3656
3657                                 /* subtract range b */
3658                                 if (key.offset + datal > off + len)
3659                                         datal = off + len - key.offset;
3660
3661                                 /* subtract range a */
3662                                 if (off > key.offset) {
3663                                         datao += off - key.offset;
3664                                         datal -= off - key.offset;
3665                                 }
3666
3667                                 ret = btrfs_drop_extents(trans, root, inode,
3668                                                          drop_start,
3669                                                          new_key.offset + datal,
3670                                                          1);
3671                                 if (ret) {
3672                                         if (ret != -EOPNOTSUPP)
3673                                                 btrfs_abort_transaction(trans,
3674                                                                         ret);
3675                                         btrfs_end_transaction(trans);
3676                                         goto out;
3677                                 }
3678
3679                                 ret = btrfs_insert_empty_item(trans, root, path,
3680                                                               &new_key, size);
3681                                 if (ret) {
3682                                         btrfs_abort_transaction(trans, ret);
3683                                         btrfs_end_transaction(trans);
3684                                         goto out;
3685                                 }
3686
3687                                 leaf = path->nodes[0];
3688                                 slot = path->slots[0];
3689                                 write_extent_buffer(leaf, buf,
3690                                             btrfs_item_ptr_offset(leaf, slot),
3691                                             size);
3692
3693                                 extent = btrfs_item_ptr(leaf, slot,
3694                                                 struct btrfs_file_extent_item);
3695
3696                                 /* disko == 0 means it's a hole */
3697                                 if (!disko)
3698                                         datao = 0;
3699
3700                                 btrfs_set_file_extent_offset(leaf, extent,
3701                                                              datao);
3702                                 btrfs_set_file_extent_num_bytes(leaf, extent,
3703                                                                 datal);
3704
3705                                 if (disko) {
3706                                         inode_add_bytes(inode, datal);
3707                                         ret = btrfs_inc_extent_ref(trans,
3708                                                         fs_info,
3709                                                         disko, diskl, 0,
3710                                                         root->root_key.objectid,
3711                                                         btrfs_ino(BTRFS_I(inode)),
3712                                                         new_key.offset - datao);
3713                                         if (ret) {
3714                                                 btrfs_abort_transaction(trans,
3715                                                                         ret);
3716                                                 btrfs_end_transaction(trans);
3717                                                 goto out;
3718
3719                                         }
3720                                 }
3721                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3722                                 u64 skip = 0;
3723                                 u64 trim = 0;
3724
3725                                 if (off > key.offset) {
3726                                         skip = off - key.offset;
3727                                         new_key.offset += skip;
3728                                 }
3729
3730                                 if (key.offset + datal > off + len)
3731                                         trim = key.offset + datal - (off + len);
3732
3733                                 if (comp && (skip || trim)) {
3734                                         ret = -EINVAL;
3735                                         btrfs_end_transaction(trans);
3736                                         goto out;
3737                                 }
3738                                 size -= skip + trim;
3739                                 datal -= skip + trim;
3740
3741                                 ret = clone_copy_inline_extent(inode,
3742                                                                trans, path,
3743                                                                &new_key,
3744                                                                drop_start,
3745                                                                datal,
3746                                                                skip, size, buf);
3747                                 if (ret) {
3748                                         if (ret != -EOPNOTSUPP)
3749                                                 btrfs_abort_transaction(trans,
3750                                                                         ret);
3751                                         btrfs_end_transaction(trans);
3752                                         goto out;
3753                                 }
3754                                 leaf = path->nodes[0];
3755                                 slot = path->slots[0];
3756                         }
3757
3758                         /* If we have an implicit hole (NO_HOLES feature). */
3759                         if (drop_start < new_key.offset)
3760                                 clone_update_extent_map(BTRFS_I(inode), trans,
3761                                                 NULL, drop_start,
3762                                                 new_key.offset - drop_start);
3763
3764                         clone_update_extent_map(BTRFS_I(inode), trans,
3765                                         path, 0, 0);
3766
3767                         btrfs_mark_buffer_dirty(leaf);
3768                         btrfs_release_path(path);
3769
3770                         last_dest_end = ALIGN(new_key.offset + datal,
3771                                               fs_info->sectorsize);
3772                         ret = clone_finish_inode_update(trans, inode,
3773                                                         last_dest_end,
3774                                                         destoff, olen,
3775                                                         no_time_update);
3776                         if (ret)
3777                                 goto out;
3778                         if (new_key.offset + datal >= destoff + len)
3779                                 break;
3780                 }
3781                 btrfs_release_path(path);
3782                 key.offset = next_key_min_offset;
3783
3784                 if (fatal_signal_pending(current)) {
3785                         ret = -EINTR;
3786                         goto out;
3787                 }
3788         }
3789         ret = 0;
3790
3791         if (last_dest_end < destoff + len) {
3792                 /*
3793                  * We have an implicit hole (NO_HOLES feature is enabled) that
3794                  * fully or partially overlaps our cloning range at its end.
3795                  */
3796                 btrfs_release_path(path);
3797
3798                 /*
3799                  * 1 - remove extent(s)
3800                  * 1 - inode update
3801                  */
3802                 trans = btrfs_start_transaction(root, 2);
3803                 if (IS_ERR(trans)) {
3804                         ret = PTR_ERR(trans);
3805                         goto out;
3806                 }
3807                 ret = btrfs_drop_extents(trans, root, inode,
3808                                          last_dest_end, destoff + len, 1);
3809                 if (ret) {
3810                         if (ret != -EOPNOTSUPP)
3811                                 btrfs_abort_transaction(trans, ret);
3812                         btrfs_end_transaction(trans);
3813                         goto out;
3814                 }
3815                 clone_update_extent_map(BTRFS_I(inode), trans, NULL,
3816                                 last_dest_end,
3817                                 destoff + len - last_dest_end);
3818                 ret = clone_finish_inode_update(trans, inode, destoff + len,
3819                                                 destoff, olen, no_time_update);
3820         }
3821
3822 out:
3823         btrfs_free_path(path);
3824         kvfree(buf);
3825         return ret;
3826 }
3827
3828 static noinline int btrfs_clone_files(struct file *file, struct file *file_src,
3829                                         u64 off, u64 olen, u64 destoff)
3830 {
3831         struct inode *inode = file_inode(file);
3832         struct inode *src = file_inode(file_src);
3833         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3834         struct btrfs_root *root = BTRFS_I(inode)->root;
3835         int ret;
3836         u64 len = olen;
3837         u64 bs = fs_info->sb->s_blocksize;
3838         int same_inode = src == inode;
3839
3840         /*
3841          * TODO:
3842          * - split compressed inline extents.  annoying: we need to
3843          *   decompress into destination's address_space (the file offset
3844          *   may change, so source mapping won't do), then recompress (or
3845          *   otherwise reinsert) a subrange.
3846          *
3847          * - split destination inode's inline extents.  The inline extents can
3848          *   be either compressed or non-compressed.
3849          */
3850
3851         if (btrfs_root_readonly(root))
3852                 return -EROFS;
3853
3854         if (file_src->f_path.mnt != file->f_path.mnt ||
3855             src->i_sb != inode->i_sb)
3856                 return -EXDEV;
3857
3858         /* don't make the dst file partly checksummed */
3859         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3860             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
3861                 return -EINVAL;
3862
3863         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
3864                 return -EISDIR;
3865
3866         if (!same_inode) {
3867                 btrfs_double_inode_lock(src, inode);
3868         } else {
3869                 inode_lock(src);
3870         }
3871
3872         /* determine range to clone */
3873         ret = -EINVAL;
3874         if (off + len > src->i_size || off + len < off)
3875                 goto out_unlock;
3876         if (len == 0)
3877                 olen = len = src->i_size - off;
3878         /* if we extend to eof, continue to block boundary */
3879         if (off + len == src->i_size)
3880                 len = ALIGN(src->i_size, bs) - off;
3881
3882         if (len == 0) {
3883                 ret = 0;
3884                 goto out_unlock;
3885         }
3886
3887         /* verify the end result is block aligned */
3888         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
3889             !IS_ALIGNED(destoff, bs))
3890                 goto out_unlock;
3891
3892         /* verify if ranges are overlapped within the same file */
3893         if (same_inode) {
3894                 if (destoff + len > off && destoff < off + len)
3895                         goto out_unlock;
3896         }
3897
3898         if (destoff > inode->i_size) {
3899                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3900                 if (ret)
3901                         goto out_unlock;
3902         }
3903
3904         /*
3905          * Lock the target range too. Right after we replace the file extent
3906          * items in the fs tree (which now point to the cloned data), we might
3907          * have a worker replace them with extent items relative to a write
3908          * operation that was issued before this clone operation (i.e. confront
3909          * with inode.c:btrfs_finish_ordered_io).
3910          */
3911         if (same_inode) {
3912                 u64 lock_start = min_t(u64, off, destoff);
3913                 u64 lock_len = max_t(u64, off, destoff) + len - lock_start;
3914
3915                 ret = lock_extent_range(src, lock_start, lock_len, true);
3916         } else {
3917                 ret = btrfs_double_extent_lock(src, off, inode, destoff, len,
3918                                                true);
3919         }
3920         ASSERT(ret == 0);
3921         if (WARN_ON(ret)) {
3922                 /* ranges in the io trees already unlocked */
3923                 goto out_unlock;
3924         }
3925
3926         ret = btrfs_clone(src, inode, off, olen, len, destoff, 0);
3927
3928         if (same_inode) {
3929                 u64 lock_start = min_t(u64, off, destoff);
3930                 u64 lock_end = max_t(u64, off, destoff) + len - 1;
3931
3932                 unlock_extent(&BTRFS_I(src)->io_tree, lock_start, lock_end);
3933         } else {
3934                 btrfs_double_extent_unlock(src, off, inode, destoff, len);
3935         }
3936         /*
3937          * Truncate page cache pages so that future reads will see the cloned
3938          * data immediately and not the previous data.
3939          */
3940         truncate_inode_pages_range(&inode->i_data,
3941                                 round_down(destoff, PAGE_SIZE),
3942                                 round_up(destoff + len, PAGE_SIZE) - 1);
3943 out_unlock:
3944         if (!same_inode)
3945                 btrfs_double_inode_unlock(src, inode);
3946         else
3947                 inode_unlock(src);
3948         return ret;
3949 }
3950
3951 int btrfs_clone_file_range(struct file *src_file, loff_t off,
3952                 struct file *dst_file, loff_t destoff, u64 len)
3953 {
3954         return btrfs_clone_files(dst_file, src_file, off, len, destoff);
3955 }
3956
3957 /*
3958  * there are many ways the trans_start and trans_end ioctls can lead
3959  * to deadlocks.  They should only be used by applications that
3960  * basically own the machine, and have a very in depth understanding
3961  * of all the possible deadlocks and enospc problems.
3962  */
3963 static long btrfs_ioctl_trans_start(struct file *file)
3964 {
3965         struct inode *inode = file_inode(file);
3966         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3967         struct btrfs_root *root = BTRFS_I(inode)->root;
3968         struct btrfs_trans_handle *trans;
3969         int ret;
3970         static bool warned = false;
3971
3972         ret = -EPERM;
3973         if (!capable(CAP_SYS_ADMIN))
3974                 goto out;
3975
3976         if (!warned) {
3977                 btrfs_warn(fs_info,
3978                         "Userspace transaction mechanism is considered "
3979                         "deprecated and slated to be removed in 4.17. "
3980                         "If you have a valid use case please "
3981                         "speak up on the mailing list");
3982                 WARN_ON(1);
3983                 warned = true;
3984         }
3985
3986         ret = -EINPROGRESS;
3987         if (file->private_data)
3988                 goto out;
3989
3990         ret = -EROFS;
3991         if (btrfs_root_readonly(root))
3992                 goto out;
3993
3994         ret = mnt_want_write_file(file);
3995         if (ret)
3996                 goto out;
3997
3998         atomic_inc(&fs_info->open_ioctl_trans);
3999
4000         ret = -ENOMEM;
4001         trans = btrfs_start_ioctl_transaction(root);
4002         if (IS_ERR(trans))
4003                 goto out_drop;
4004
4005         file->private_data = trans;
4006         return 0;
4007
4008 out_drop:
4009         atomic_dec(&fs_info->open_ioctl_trans);
4010         mnt_drop_write_file(file);
4011 out:
4012         return ret;
4013 }
4014
4015 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
4016 {
4017         struct inode *inode = file_inode(file);
4018         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4019         struct btrfs_root *root = BTRFS_I(inode)->root;
4020         struct btrfs_root *new_root;
4021         struct btrfs_dir_item *di;
4022         struct btrfs_trans_handle *trans;
4023         struct btrfs_path *path;
4024         struct btrfs_key location;
4025         struct btrfs_disk_key disk_key;
4026         u64 objectid = 0;
4027         u64 dir_id;
4028         int ret;
4029
4030         if (!capable(CAP_SYS_ADMIN))
4031                 return -EPERM;
4032
4033         ret = mnt_want_write_file(file);
4034         if (ret)
4035                 return ret;
4036
4037         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
4038                 ret = -EFAULT;
4039                 goto out;
4040         }
4041
4042         if (!objectid)
4043                 objectid = BTRFS_FS_TREE_OBJECTID;
4044
4045         location.objectid = objectid;
4046         location.type = BTRFS_ROOT_ITEM_KEY;
4047         location.offset = (u64)-1;
4048
4049         new_root = btrfs_read_fs_root_no_name(fs_info, &location);
4050         if (IS_ERR(new_root)) {
4051                 ret = PTR_ERR(new_root);
4052                 goto out;
4053         }
4054
4055         path = btrfs_alloc_path();
4056         if (!path) {
4057                 ret = -ENOMEM;
4058                 goto out;
4059         }
4060         path->leave_spinning = 1;
4061
4062         trans = btrfs_start_transaction(root, 1);
4063         if (IS_ERR(trans)) {
4064                 btrfs_free_path(path);
4065                 ret = PTR_ERR(trans);
4066                 goto out;
4067         }
4068
4069         dir_id = btrfs_super_root_dir(fs_info->super_copy);
4070         di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
4071                                    dir_id, "default", 7, 1);
4072         if (IS_ERR_OR_NULL(di)) {
4073                 btrfs_free_path(path);
4074                 btrfs_end_transaction(trans);
4075                 btrfs_err(fs_info,
4076                           "Umm, you don't have the default diritem, this isn't going to work");
4077                 ret = -ENOENT;
4078                 goto out;
4079         }
4080
4081         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
4082         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
4083         btrfs_mark_buffer_dirty(path->nodes[0]);
4084         btrfs_free_path(path);
4085
4086         btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
4087         btrfs_end_transaction(trans);
4088 out:
4089         mnt_drop_write_file(file);
4090         return ret;
4091 }
4092
4093 void btrfs_get_block_group_info(struct list_head *groups_list,
4094                                 struct btrfs_ioctl_space_info *space)
4095 {
4096         struct btrfs_block_group_cache *block_group;
4097
4098         space->total_bytes = 0;
4099         space->used_bytes = 0;
4100         space->flags = 0;
4101         list_for_each_entry(block_group, groups_list, list) {
4102                 space->flags = block_group->flags;
4103                 space->total_bytes += block_group->key.offset;
4104                 space->used_bytes +=
4105                         btrfs_block_group_used(&block_group->item);
4106         }
4107 }
4108
4109 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
4110                                    void __user *arg)
4111 {
4112         struct btrfs_ioctl_space_args space_args;
4113         struct btrfs_ioctl_space_info space;
4114         struct btrfs_ioctl_space_info *dest;
4115         struct btrfs_ioctl_space_info *dest_orig;
4116         struct btrfs_ioctl_space_info __user *user_dest;
4117         struct btrfs_space_info *info;
4118         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
4119                        BTRFS_BLOCK_GROUP_SYSTEM,
4120                        BTRFS_BLOCK_GROUP_METADATA,
4121                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
4122         int num_types = 4;
4123         int alloc_size;
4124         int ret = 0;
4125         u64 slot_count = 0;
4126         int i, c;
4127
4128         if (copy_from_user(&space_args,
4129                            (struct btrfs_ioctl_space_args __user *)arg,
4130                            sizeof(space_args)))
4131                 return -EFAULT;
4132
4133         for (i = 0; i < num_types; i++) {
4134                 struct btrfs_space_info *tmp;
4135
4136                 info = NULL;
4137                 rcu_read_lock();
4138                 list_for_each_entry_rcu(tmp, &fs_info->space_info,
4139                                         list) {
4140                         if (tmp->flags == types[i]) {
4141                                 info = tmp;
4142                                 break;
4143                         }
4144                 }
4145                 rcu_read_unlock();
4146
4147                 if (!info)
4148                         continue;
4149
4150                 down_read(&info->groups_sem);
4151                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4152                         if (!list_empty(&info->block_groups[c]))
4153                                 slot_count++;
4154                 }
4155                 up_read(&info->groups_sem);
4156         }
4157
4158         /*
4159          * Global block reserve, exported as a space_info
4160          */
4161         slot_count++;
4162
4163         /* space_slots == 0 means they are asking for a count */
4164         if (space_args.space_slots == 0) {
4165                 space_args.total_spaces = slot_count;
4166                 goto out;
4167         }
4168
4169         slot_count = min_t(u64, space_args.space_slots, slot_count);
4170
4171         alloc_size = sizeof(*dest) * slot_count;
4172
4173         /* we generally have at most 6 or so space infos, one for each raid
4174          * level.  So, a whole page should be more than enough for everyone
4175          */
4176         if (alloc_size > PAGE_SIZE)
4177                 return -ENOMEM;
4178
4179         space_args.total_spaces = 0;
4180         dest = kmalloc(alloc_size, GFP_KERNEL);
4181         if (!dest)
4182                 return -ENOMEM;
4183         dest_orig = dest;
4184
4185         /* now we have a buffer to copy into */
4186         for (i = 0; i < num_types; i++) {
4187                 struct btrfs_space_info *tmp;
4188
4189                 if (!slot_count)
4190                         break;
4191
4192                 info = NULL;
4193                 rcu_read_lock();
4194                 list_for_each_entry_rcu(tmp, &fs_info->space_info,
4195                                         list) {
4196                         if (tmp->flags == types[i]) {
4197                                 info = tmp;
4198                                 break;
4199                         }
4200                 }
4201                 rcu_read_unlock();
4202
4203                 if (!info)
4204                         continue;
4205                 down_read(&info->groups_sem);
4206                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4207                         if (!list_empty(&info->block_groups[c])) {
4208                                 btrfs_get_block_group_info(
4209                                         &info->block_groups[c], &space);
4210                                 memcpy(dest, &space, sizeof(space));
4211                                 dest++;
4212                                 space_args.total_spaces++;
4213                                 slot_count--;
4214                         }
4215                         if (!slot_count)
4216                                 break;
4217                 }
4218                 up_read(&info->groups_sem);
4219         }
4220
4221         /*
4222          * Add global block reserve
4223          */
4224         if (slot_count) {
4225                 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4226
4227                 spin_lock(&block_rsv->lock);
4228                 space.total_bytes = block_rsv->size;
4229                 space.used_bytes = block_rsv->size - block_rsv->reserved;
4230                 spin_unlock(&block_rsv->lock);
4231                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
4232                 memcpy(dest, &space, sizeof(space));
4233                 space_args.total_spaces++;
4234         }
4235
4236         user_dest = (struct btrfs_ioctl_space_info __user *)
4237                 (arg + sizeof(struct btrfs_ioctl_space_args));
4238
4239         if (copy_to_user(user_dest, dest_orig, alloc_size))
4240                 ret = -EFAULT;
4241
4242         kfree(dest_orig);
4243 out:
4244         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
4245                 ret = -EFAULT;
4246
4247         return ret;
4248 }
4249
4250 /*
4251  * there are many ways the trans_start and trans_end ioctls can lead
4252  * to deadlocks.  They should only be used by applications that
4253  * basically own the machine, and have a very in depth understanding
4254  * of all the possible deadlocks and enospc problems.
4255  */
4256 long btrfs_ioctl_trans_end(struct file *file)
4257 {
4258         struct inode *inode = file_inode(file);
4259         struct btrfs_root *root = BTRFS_I(inode)->root;
4260         struct btrfs_trans_handle *trans;
4261
4262         trans = file->private_data;
4263         if (!trans)
4264                 return -EINVAL;
4265         file->private_data = NULL;
4266
4267         btrfs_end_transaction(trans);
4268
4269         atomic_dec(&root->fs_info->open_ioctl_trans);
4270
4271         mnt_drop_write_file(file);
4272         return 0;
4273 }
4274
4275 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
4276                                             void __user *argp)
4277 {
4278         struct btrfs_trans_handle *trans;
4279         u64 transid;
4280         int ret;
4281
4282         trans = btrfs_attach_transaction_barrier(root);
4283         if (IS_ERR(trans)) {
4284                 if (PTR_ERR(trans) != -ENOENT)
4285                         return PTR_ERR(trans);
4286
4287                 /* No running transaction, don't bother */
4288                 transid = root->fs_info->last_trans_committed;
4289                 goto out;
4290         }
4291         transid = trans->transid;
4292         ret = btrfs_commit_transaction_async(trans, 0);
4293         if (ret) {
4294                 btrfs_end_transaction(trans);
4295                 return ret;
4296         }
4297 out:
4298         if (argp)
4299                 if (copy_to_user(argp, &transid, sizeof(transid)))
4300                         return -EFAULT;
4301         return 0;
4302 }
4303
4304 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
4305                                            void __user *argp)
4306 {
4307         u64 transid;
4308
4309         if (argp) {
4310                 if (copy_from_user(&transid, argp, sizeof(transid)))
4311                         return -EFAULT;
4312         } else {
4313                 transid = 0;  /* current trans */
4314         }
4315         return btrfs_wait_for_commit(fs_info, transid);
4316 }
4317
4318 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
4319 {
4320         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
4321         struct btrfs_ioctl_scrub_args *sa;
4322         int ret;
4323
4324         if (!capable(CAP_SYS_ADMIN))
4325                 return -EPERM;
4326
4327         sa = memdup_user(arg, sizeof(*sa));
4328         if (IS_ERR(sa))
4329                 return PTR_ERR(sa);
4330
4331         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
4332                 ret = mnt_want_write_file(file);
4333                 if (ret)
4334                         goto out;
4335         }
4336
4337         ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
4338                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4339                               0);
4340
4341         if (copy_to_user(arg, sa, sizeof(*sa)))
4342                 ret = -EFAULT;
4343
4344         if (!(sa->flags & BTRFS_SCRUB_READONLY))
4345                 mnt_drop_write_file(file);
4346 out:
4347         kfree(sa);
4348         return ret;
4349 }
4350
4351 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
4352 {
4353         if (!capable(CAP_SYS_ADMIN))
4354                 return -EPERM;
4355
4356         return btrfs_scrub_cancel(fs_info);
4357 }
4358
4359 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
4360                                        void __user *arg)
4361 {
4362         struct btrfs_ioctl_scrub_args *sa;
4363         int ret;
4364
4365         if (!capable(CAP_SYS_ADMIN))
4366                 return -EPERM;
4367
4368         sa = memdup_user(arg, sizeof(*sa));
4369         if (IS_ERR(sa))
4370                 return PTR_ERR(sa);
4371
4372         ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
4373
4374         if (copy_to_user(arg, sa, sizeof(*sa)))
4375                 ret = -EFAULT;
4376
4377         kfree(sa);
4378         return ret;
4379 }
4380
4381 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
4382                                       void __user *arg)
4383 {
4384         struct btrfs_ioctl_get_dev_stats *sa;
4385         int ret;
4386
4387         sa = memdup_user(arg, sizeof(*sa));
4388         if (IS_ERR(sa))
4389                 return PTR_ERR(sa);
4390
4391         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4392                 kfree(sa);
4393                 return -EPERM;
4394         }
4395
4396         ret = btrfs_get_dev_stats(fs_info, sa);
4397
4398         if (copy_to_user(arg, sa, sizeof(*sa)))
4399                 ret = -EFAULT;
4400
4401         kfree(sa);
4402         return ret;
4403 }
4404
4405 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
4406                                     void __user *arg)
4407 {
4408         struct btrfs_ioctl_dev_replace_args *p;
4409         int ret;
4410
4411         if (!capable(CAP_SYS_ADMIN))
4412                 return -EPERM;
4413
4414         p = memdup_user(arg, sizeof(*p));
4415         if (IS_ERR(p))
4416                 return PTR_ERR(p);
4417
4418         switch (p->cmd) {
4419         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4420                 if (fs_info->sb->s_flags & MS_RDONLY) {
4421                         ret = -EROFS;
4422                         goto out;
4423                 }
4424                 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
4425                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4426                 } else {
4427                         ret = btrfs_dev_replace_by_ioctl(fs_info, p);
4428                         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4429                 }
4430                 break;
4431         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4432                 btrfs_dev_replace_status(fs_info, p);
4433                 ret = 0;
4434                 break;
4435         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4436                 ret = btrfs_dev_replace_cancel(fs_info, p);
4437                 break;
4438         default:
4439                 ret = -EINVAL;
4440                 break;
4441         }
4442
4443         if (copy_to_user(arg, p, sizeof(*p)))
4444                 ret = -EFAULT;
4445 out:
4446         kfree(p);
4447         return ret;
4448 }
4449
4450 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4451 {
4452         int ret = 0;
4453         int i;
4454         u64 rel_ptr;
4455         int size;
4456         struct btrfs_ioctl_ino_path_args *ipa = NULL;
4457         struct inode_fs_paths *ipath = NULL;
4458         struct btrfs_path *path;
4459
4460         if (!capable(CAP_DAC_READ_SEARCH))
4461                 return -EPERM;
4462
4463         path = btrfs_alloc_path();
4464         if (!path) {
4465                 ret = -ENOMEM;
4466                 goto out;
4467         }
4468
4469         ipa = memdup_user(arg, sizeof(*ipa));
4470         if (IS_ERR(ipa)) {
4471                 ret = PTR_ERR(ipa);
4472                 ipa = NULL;
4473                 goto out;
4474         }
4475
4476         size = min_t(u32, ipa->size, 4096);
4477         ipath = init_ipath(size, root, path);
4478         if (IS_ERR(ipath)) {
4479                 ret = PTR_ERR(ipath);
4480                 ipath = NULL;
4481                 goto out;
4482         }
4483
4484         ret = paths_from_inode(ipa->inum, ipath);
4485         if (ret < 0)
4486                 goto out;
4487
4488         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4489                 rel_ptr = ipath->fspath->val[i] -
4490                           (u64)(unsigned long)ipath->fspath->val;
4491                 ipath->fspath->val[i] = rel_ptr;
4492         }
4493
4494         ret = copy_to_user((void *)(unsigned long)ipa->fspath,
4495                            (void *)(unsigned long)ipath->fspath, size);
4496         if (ret) {
4497                 ret = -EFAULT;
4498                 goto out;
4499         }
4500
4501 out:
4502         btrfs_free_path(path);
4503         free_ipath(ipath);
4504         kfree(ipa);
4505
4506         return ret;
4507 }
4508
4509 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4510 {
4511         struct btrfs_data_container *inodes = ctx;
4512         const size_t c = 3 * sizeof(u64);
4513
4514         if (inodes->bytes_left >= c) {
4515                 inodes->bytes_left -= c;
4516                 inodes->val[inodes->elem_cnt] = inum;
4517                 inodes->val[inodes->elem_cnt + 1] = offset;
4518                 inodes->val[inodes->elem_cnt + 2] = root;
4519                 inodes->elem_cnt += 3;
4520         } else {
4521                 inodes->bytes_missing += c - inodes->bytes_left;
4522                 inodes->bytes_left = 0;
4523                 inodes->elem_missed += 3;
4524         }
4525
4526         return 0;
4527 }
4528
4529 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
4530                                         void __user *arg)
4531 {
4532         int ret = 0;
4533         int size;
4534         struct btrfs_ioctl_logical_ino_args *loi;
4535         struct btrfs_data_container *inodes = NULL;
4536         struct btrfs_path *path = NULL;
4537
4538         if (!capable(CAP_SYS_ADMIN))
4539                 return -EPERM;
4540
4541         loi = memdup_user(arg, sizeof(*loi));
4542         if (IS_ERR(loi))
4543                 return PTR_ERR(loi);
4544
4545         path = btrfs_alloc_path();
4546         if (!path) {
4547                 ret = -ENOMEM;
4548                 goto out;
4549         }
4550
4551         size = min_t(u32, loi->size, SZ_64K);
4552         inodes = init_data_container(size);
4553         if (IS_ERR(inodes)) {
4554                 ret = PTR_ERR(inodes);
4555                 inodes = NULL;
4556                 goto out;
4557         }
4558
4559         ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
4560                                           build_ino_list, inodes);
4561         if (ret == -EINVAL)
4562                 ret = -ENOENT;
4563         if (ret < 0)
4564                 goto out;
4565
4566         ret = copy_to_user((void *)(unsigned long)loi->inodes,
4567                            (void *)(unsigned long)inodes, size);
4568         if (ret)
4569                 ret = -EFAULT;
4570
4571 out:
4572         btrfs_free_path(path);
4573         kvfree(inodes);
4574         kfree(loi);
4575
4576         return ret;
4577 }
4578
4579 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
4580                                struct btrfs_ioctl_balance_args *bargs)
4581 {
4582         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4583
4584         bargs->flags = bctl->flags;
4585
4586         if (atomic_read(&fs_info->balance_running))
4587                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4588         if (atomic_read(&fs_info->balance_pause_req))
4589                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4590         if (atomic_read(&fs_info->balance_cancel_req))
4591                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4592
4593         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4594         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4595         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4596
4597         if (lock) {
4598                 spin_lock(&fs_info->balance_lock);
4599                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4600                 spin_unlock(&fs_info->balance_lock);
4601         } else {
4602                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4603         }
4604 }
4605
4606 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4607 {
4608         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4609         struct btrfs_fs_info *fs_info = root->fs_info;
4610         struct btrfs_ioctl_balance_args *bargs;
4611         struct btrfs_balance_control *bctl;
4612         bool need_unlock; /* for mut. excl. ops lock */
4613         int ret;
4614
4615         if (!capable(CAP_SYS_ADMIN))
4616                 return -EPERM;
4617
4618         ret = mnt_want_write_file(file);
4619         if (ret)
4620                 return ret;
4621
4622 again:
4623         if (!test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
4624                 mutex_lock(&fs_info->volume_mutex);
4625                 mutex_lock(&fs_info->balance_mutex);
4626                 need_unlock = true;
4627                 goto locked;
4628         }
4629
4630         /*
4631          * mut. excl. ops lock is locked.  Three possibilities:
4632          *   (1) some other op is running
4633          *   (2) balance is running
4634          *   (3) balance is paused -- special case (think resume)
4635          */
4636         mutex_lock(&fs_info->balance_mutex);
4637         if (fs_info->balance_ctl) {
4638                 /* this is either (2) or (3) */
4639                 if (!atomic_read(&fs_info->balance_running)) {
4640                         mutex_unlock(&fs_info->balance_mutex);
4641                         if (!mutex_trylock(&fs_info->volume_mutex))
4642                                 goto again;
4643                         mutex_lock(&fs_info->balance_mutex);
4644
4645                         if (fs_info->balance_ctl &&
4646                             !atomic_read(&fs_info->balance_running)) {
4647                                 /* this is (3) */
4648                                 need_unlock = false;
4649                                 goto locked;
4650                         }
4651
4652                         mutex_unlock(&fs_info->balance_mutex);
4653                         mutex_unlock(&fs_info->volume_mutex);
4654                         goto again;
4655                 } else {
4656                         /* this is (2) */
4657                         mutex_unlock(&fs_info->balance_mutex);
4658                         ret = -EINPROGRESS;
4659                         goto out;
4660                 }
4661         } else {
4662                 /* this is (1) */
4663                 mutex_unlock(&fs_info->balance_mutex);
4664                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4665                 goto out;
4666         }
4667
4668 locked:
4669         BUG_ON(!test_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
4670
4671         if (arg) {
4672                 bargs = memdup_user(arg, sizeof(*bargs));
4673                 if (IS_ERR(bargs)) {
4674                         ret = PTR_ERR(bargs);
4675                         goto out_unlock;
4676                 }
4677
4678                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4679                         if (!fs_info->balance_ctl) {
4680                                 ret = -ENOTCONN;
4681                                 goto out_bargs;
4682                         }
4683
4684                         bctl = fs_info->balance_ctl;
4685                         spin_lock(&fs_info->balance_lock);
4686                         bctl->flags |= BTRFS_BALANCE_RESUME;
4687                         spin_unlock(&fs_info->balance_lock);
4688
4689                         goto do_balance;
4690                 }
4691         } else {
4692                 bargs = NULL;
4693         }
4694
4695         if (fs_info->balance_ctl) {
4696                 ret = -EINPROGRESS;
4697                 goto out_bargs;
4698         }
4699
4700         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4701         if (!bctl) {
4702                 ret = -ENOMEM;
4703                 goto out_bargs;
4704         }
4705
4706         bctl->fs_info = fs_info;
4707         if (arg) {
4708                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4709                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4710                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4711
4712                 bctl->flags = bargs->flags;
4713         } else {
4714                 /* balance everything - no filters */
4715                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4716         }
4717
4718         if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4719                 ret = -EINVAL;
4720                 goto out_bctl;
4721         }
4722
4723 do_balance:
4724         /*
4725          * Ownership of bctl and filesystem flag BTRFS_FS_EXCL_OP
4726          * goes to to btrfs_balance.  bctl is freed in __cancel_balance,
4727          * or, if restriper was paused all the way until unmount, in
4728          * free_fs_info.  The flag is cleared in __cancel_balance.
4729          */
4730         need_unlock = false;
4731
4732         ret = btrfs_balance(bctl, bargs);
4733         bctl = NULL;
4734
4735         if (arg) {
4736                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4737                         ret = -EFAULT;
4738         }
4739
4740 out_bctl:
4741         kfree(bctl);
4742 out_bargs:
4743         kfree(bargs);
4744 out_unlock:
4745         mutex_unlock(&fs_info->balance_mutex);
4746         mutex_unlock(&fs_info->volume_mutex);
4747         if (need_unlock)
4748                 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4749 out:
4750         mnt_drop_write_file(file);
4751         return ret;
4752 }
4753
4754 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
4755 {
4756         if (!capable(CAP_SYS_ADMIN))
4757                 return -EPERM;
4758
4759         switch (cmd) {
4760         case BTRFS_BALANCE_CTL_PAUSE:
4761                 return btrfs_pause_balance(fs_info);
4762         case BTRFS_BALANCE_CTL_CANCEL:
4763                 return btrfs_cancel_balance(fs_info);
4764         }
4765
4766         return -EINVAL;
4767 }
4768
4769 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
4770                                          void __user *arg)
4771 {
4772         struct btrfs_ioctl_balance_args *bargs;
4773         int ret = 0;
4774
4775         if (!capable(CAP_SYS_ADMIN))
4776                 return -EPERM;
4777
4778         mutex_lock(&fs_info->balance_mutex);
4779         if (!fs_info->balance_ctl) {
4780                 ret = -ENOTCONN;
4781                 goto out;
4782         }
4783
4784         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4785         if (!bargs) {
4786                 ret = -ENOMEM;
4787                 goto out;
4788         }
4789
4790         update_ioctl_balance_args(fs_info, 1, bargs);
4791
4792         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4793                 ret = -EFAULT;
4794
4795         kfree(bargs);
4796 out:
4797         mutex_unlock(&fs_info->balance_mutex);
4798         return ret;
4799 }
4800
4801 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4802 {
4803         struct inode *inode = file_inode(file);
4804         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4805         struct btrfs_ioctl_quota_ctl_args *sa;
4806         struct btrfs_trans_handle *trans = NULL;
4807         int ret;
4808         int err;
4809
4810         if (!capable(CAP_SYS_ADMIN))
4811                 return -EPERM;
4812
4813         ret = mnt_want_write_file(file);
4814         if (ret)
4815                 return ret;
4816
4817         sa = memdup_user(arg, sizeof(*sa));
4818         if (IS_ERR(sa)) {
4819                 ret = PTR_ERR(sa);
4820                 goto drop_write;
4821         }
4822
4823         down_write(&fs_info->subvol_sem);
4824         trans = btrfs_start_transaction(fs_info->tree_root, 2);
4825         if (IS_ERR(trans)) {
4826                 ret = PTR_ERR(trans);
4827                 goto out;
4828         }
4829
4830         switch (sa->cmd) {
4831         case BTRFS_QUOTA_CTL_ENABLE:
4832                 ret = btrfs_quota_enable(trans, fs_info);
4833                 break;
4834         case BTRFS_QUOTA_CTL_DISABLE:
4835                 ret = btrfs_quota_disable(trans, fs_info);
4836                 break;
4837         default:
4838                 ret = -EINVAL;
4839                 break;
4840         }
4841
4842         err = btrfs_commit_transaction(trans);
4843         if (err && !ret)
4844                 ret = err;
4845 out:
4846         kfree(sa);
4847         up_write(&fs_info->subvol_sem);
4848 drop_write:
4849         mnt_drop_write_file(file);
4850         return ret;
4851 }
4852
4853 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4854 {
4855         struct inode *inode = file_inode(file);
4856         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4857         struct btrfs_root *root = BTRFS_I(inode)->root;
4858         struct btrfs_ioctl_qgroup_assign_args *sa;
4859         struct btrfs_trans_handle *trans;
4860         int ret;
4861         int err;
4862
4863         if (!capable(CAP_SYS_ADMIN))
4864                 return -EPERM;
4865
4866         ret = mnt_want_write_file(file);
4867         if (ret)
4868                 return ret;
4869
4870         sa = memdup_user(arg, sizeof(*sa));
4871         if (IS_ERR(sa)) {
4872                 ret = PTR_ERR(sa);
4873                 goto drop_write;
4874         }
4875
4876         trans = btrfs_join_transaction(root);
4877         if (IS_ERR(trans)) {
4878                 ret = PTR_ERR(trans);
4879                 goto out;
4880         }
4881
4882         if (sa->assign) {
4883                 ret = btrfs_add_qgroup_relation(trans, fs_info,
4884                                                 sa->src, sa->dst);
4885         } else {
4886                 ret = btrfs_del_qgroup_relation(trans, fs_info,
4887                                                 sa->src, sa->dst);
4888         }
4889
4890         /* update qgroup status and info */
4891         err = btrfs_run_qgroups(trans, fs_info);
4892         if (err < 0)
4893                 btrfs_handle_fs_error(fs_info, err,
4894                                       "failed to update qgroup status and info");
4895         err = btrfs_end_transaction(trans);
4896         if (err && !ret)
4897                 ret = err;
4898
4899 out:
4900         kfree(sa);
4901 drop_write:
4902         mnt_drop_write_file(file);
4903         return ret;
4904 }
4905
4906 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4907 {
4908         struct inode *inode = file_inode(file);
4909         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4910         struct btrfs_root *root = BTRFS_I(inode)->root;
4911         struct btrfs_ioctl_qgroup_create_args *sa;
4912         struct btrfs_trans_handle *trans;
4913         int ret;
4914         int err;
4915
4916         if (!capable(CAP_SYS_ADMIN))
4917                 return -EPERM;
4918
4919         ret = mnt_want_write_file(file);
4920         if (ret)
4921                 return ret;
4922
4923         sa = memdup_user(arg, sizeof(*sa));
4924         if (IS_ERR(sa)) {
4925                 ret = PTR_ERR(sa);
4926                 goto drop_write;
4927         }
4928
4929         if (!sa->qgroupid) {
4930                 ret = -EINVAL;
4931                 goto out;
4932         }
4933
4934         trans = btrfs_join_transaction(root);
4935         if (IS_ERR(trans)) {
4936                 ret = PTR_ERR(trans);
4937                 goto out;
4938         }
4939
4940         if (sa->create) {
4941                 ret = btrfs_create_qgroup(trans, fs_info, sa->qgroupid);
4942         } else {
4943                 ret = btrfs_remove_qgroup(trans, fs_info, sa->qgroupid);
4944         }
4945
4946         err = btrfs_end_transaction(trans);
4947         if (err && !ret)
4948                 ret = err;
4949
4950 out:
4951         kfree(sa);
4952 drop_write:
4953         mnt_drop_write_file(file);
4954         return ret;
4955 }
4956
4957 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4958 {
4959         struct inode *inode = file_inode(file);
4960         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4961         struct btrfs_root *root = BTRFS_I(inode)->root;
4962         struct btrfs_ioctl_qgroup_limit_args *sa;
4963         struct btrfs_trans_handle *trans;
4964         int ret;
4965         int err;
4966         u64 qgroupid;
4967
4968         if (!capable(CAP_SYS_ADMIN))
4969                 return -EPERM;
4970
4971         ret = mnt_want_write_file(file);
4972         if (ret)
4973                 return ret;
4974
4975         sa = memdup_user(arg, sizeof(*sa));
4976         if (IS_ERR(sa)) {
4977                 ret = PTR_ERR(sa);
4978                 goto drop_write;
4979         }
4980
4981         trans = btrfs_join_transaction(root);
4982         if (IS_ERR(trans)) {
4983                 ret = PTR_ERR(trans);
4984                 goto out;
4985         }
4986
4987         qgroupid = sa->qgroupid;
4988         if (!qgroupid) {
4989                 /* take the current subvol as qgroup */
4990                 qgroupid = root->root_key.objectid;
4991         }
4992
4993         ret = btrfs_limit_qgroup(trans, fs_info, qgroupid, &sa->lim);
4994
4995         err = btrfs_end_transaction(trans);
4996         if (err && !ret)
4997                 ret = err;
4998
4999 out:
5000         kfree(sa);
5001 drop_write:
5002         mnt_drop_write_file(file);
5003         return ret;
5004 }
5005
5006 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
5007 {
5008         struct inode *inode = file_inode(file);
5009         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5010         struct btrfs_ioctl_quota_rescan_args *qsa;
5011         int ret;
5012
5013         if (!capable(CAP_SYS_ADMIN))
5014                 return -EPERM;
5015
5016         ret = mnt_want_write_file(file);
5017         if (ret)
5018                 return ret;
5019
5020         qsa = memdup_user(arg, sizeof(*qsa));
5021         if (IS_ERR(qsa)) {
5022                 ret = PTR_ERR(qsa);
5023                 goto drop_write;
5024         }
5025
5026         if (qsa->flags) {
5027                 ret = -EINVAL;
5028                 goto out;
5029         }
5030
5031         ret = btrfs_qgroup_rescan(fs_info);
5032
5033 out:
5034         kfree(qsa);
5035 drop_write:
5036         mnt_drop_write_file(file);
5037         return ret;
5038 }
5039
5040 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
5041 {
5042         struct inode *inode = file_inode(file);
5043         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5044         struct btrfs_ioctl_quota_rescan_args *qsa;
5045         int ret = 0;
5046
5047         if (!capable(CAP_SYS_ADMIN))
5048                 return -EPERM;
5049
5050         qsa = kzalloc(sizeof(*qsa), GFP_KERNEL);
5051         if (!qsa)
5052                 return -ENOMEM;
5053
5054         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
5055                 qsa->flags = 1;
5056                 qsa->progress = fs_info->qgroup_rescan_progress.objectid;
5057         }
5058
5059         if (copy_to_user(arg, qsa, sizeof(*qsa)))
5060                 ret = -EFAULT;
5061
5062         kfree(qsa);
5063         return ret;
5064 }
5065
5066 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
5067 {
5068         struct inode *inode = file_inode(file);
5069         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5070
5071         if (!capable(CAP_SYS_ADMIN))
5072                 return -EPERM;
5073
5074         return btrfs_qgroup_wait_for_completion(fs_info, true);
5075 }
5076
5077 static long _btrfs_ioctl_set_received_subvol(struct file *file,
5078                                             struct btrfs_ioctl_received_subvol_args *sa)
5079 {
5080         struct inode *inode = file_inode(file);
5081         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5082         struct btrfs_root *root = BTRFS_I(inode)->root;
5083         struct btrfs_root_item *root_item = &root->root_item;
5084         struct btrfs_trans_handle *trans;
5085         struct timespec ct = current_time(inode);
5086         int ret = 0;
5087         int received_uuid_changed;
5088
5089         if (!inode_owner_or_capable(inode))
5090                 return -EPERM;
5091
5092         ret = mnt_want_write_file(file);
5093         if (ret < 0)
5094                 return ret;
5095
5096         down_write(&fs_info->subvol_sem);
5097
5098         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
5099                 ret = -EINVAL;
5100                 goto out;
5101         }
5102
5103         if (btrfs_root_readonly(root)) {
5104                 ret = -EROFS;
5105                 goto out;
5106         }
5107
5108         /*
5109          * 1 - root item
5110          * 2 - uuid items (received uuid + subvol uuid)
5111          */
5112         trans = btrfs_start_transaction(root, 3);
5113         if (IS_ERR(trans)) {
5114                 ret = PTR_ERR(trans);
5115                 trans = NULL;
5116                 goto out;
5117         }
5118
5119         sa->rtransid = trans->transid;
5120         sa->rtime.sec = ct.tv_sec;
5121         sa->rtime.nsec = ct.tv_nsec;
5122
5123         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
5124                                        BTRFS_UUID_SIZE);
5125         if (received_uuid_changed &&
5126             !btrfs_is_empty_uuid(root_item->received_uuid))
5127                 btrfs_uuid_tree_rem(trans, fs_info, root_item->received_uuid,
5128                                     BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5129                                     root->root_key.objectid);
5130         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
5131         btrfs_set_root_stransid(root_item, sa->stransid);
5132         btrfs_set_root_rtransid(root_item, sa->rtransid);
5133         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
5134         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
5135         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
5136         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
5137
5138         ret = btrfs_update_root(trans, fs_info->tree_root,
5139                                 &root->root_key, &root->root_item);
5140         if (ret < 0) {
5141                 btrfs_end_transaction(trans);
5142                 goto out;
5143         }
5144         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
5145                 ret = btrfs_uuid_tree_add(trans, fs_info, sa->uuid,
5146                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5147                                           root->root_key.objectid);
5148                 if (ret < 0 && ret != -EEXIST) {
5149                         btrfs_abort_transaction(trans, ret);
5150                         goto out;
5151                 }
5152         }
5153         ret = btrfs_commit_transaction(trans);
5154         if (ret < 0) {
5155                 btrfs_abort_transaction(trans, ret);
5156                 goto out;
5157         }
5158
5159 out:
5160         up_write(&fs_info->subvol_sem);
5161         mnt_drop_write_file(file);
5162         return ret;
5163 }
5164
5165 #ifdef CONFIG_64BIT
5166 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
5167                                                 void __user *arg)
5168 {
5169         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
5170         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
5171         int ret = 0;
5172
5173         args32 = memdup_user(arg, sizeof(*args32));
5174         if (IS_ERR(args32))
5175                 return PTR_ERR(args32);
5176
5177         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
5178         if (!args64) {
5179                 ret = -ENOMEM;
5180                 goto out;
5181         }
5182
5183         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
5184         args64->stransid = args32->stransid;
5185         args64->rtransid = args32->rtransid;
5186         args64->stime.sec = args32->stime.sec;
5187         args64->stime.nsec = args32->stime.nsec;
5188         args64->rtime.sec = args32->rtime.sec;
5189         args64->rtime.nsec = args32->rtime.nsec;
5190         args64->flags = args32->flags;
5191
5192         ret = _btrfs_ioctl_set_received_subvol(file, args64);
5193         if (ret)
5194                 goto out;
5195
5196         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
5197         args32->stransid = args64->stransid;
5198         args32->rtransid = args64->rtransid;
5199         args32->stime.sec = args64->stime.sec;
5200         args32->stime.nsec = args64->stime.nsec;
5201         args32->rtime.sec = args64->rtime.sec;
5202         args32->rtime.nsec = args64->rtime.nsec;
5203         args32->flags = args64->flags;
5204
5205         ret = copy_to_user(arg, args32, sizeof(*args32));
5206         if (ret)
5207                 ret = -EFAULT;
5208
5209 out:
5210         kfree(args32);
5211         kfree(args64);
5212         return ret;
5213 }
5214 #endif
5215
5216 static long btrfs_ioctl_set_received_subvol(struct file *file,
5217                                             void __user *arg)
5218 {
5219         struct btrfs_ioctl_received_subvol_args *sa = NULL;
5220         int ret = 0;
5221
5222         sa = memdup_user(arg, sizeof(*sa));
5223         if (IS_ERR(sa))
5224                 return PTR_ERR(sa);
5225
5226         ret = _btrfs_ioctl_set_received_subvol(file, sa);
5227
5228         if (ret)
5229                 goto out;
5230
5231         ret = copy_to_user(arg, sa, sizeof(*sa));
5232         if (ret)
5233                 ret = -EFAULT;
5234
5235 out:
5236         kfree(sa);
5237         return ret;
5238 }
5239
5240 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
5241 {
5242         struct inode *inode = file_inode(file);
5243         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5244         size_t len;
5245         int ret;
5246         char label[BTRFS_LABEL_SIZE];
5247
5248         spin_lock(&fs_info->super_lock);
5249         memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
5250         spin_unlock(&fs_info->super_lock);
5251
5252         len = strnlen(label, BTRFS_LABEL_SIZE);
5253
5254         if (len == BTRFS_LABEL_SIZE) {
5255                 btrfs_warn(fs_info,
5256                            "label is too long, return the first %zu bytes",
5257                            --len);
5258         }
5259
5260         ret = copy_to_user(arg, label, len);
5261
5262         return ret ? -EFAULT : 0;
5263 }
5264
5265 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
5266 {
5267         struct inode *inode = file_inode(file);
5268         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5269         struct btrfs_root *root = BTRFS_I(inode)->root;
5270         struct btrfs_super_block *super_block = fs_info->super_copy;
5271         struct btrfs_trans_handle *trans;
5272         char label[BTRFS_LABEL_SIZE];
5273         int ret;
5274
5275         if (!capable(CAP_SYS_ADMIN))
5276                 return -EPERM;
5277
5278         if (copy_from_user(label, arg, sizeof(label)))
5279                 return -EFAULT;
5280
5281         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
5282                 btrfs_err(fs_info,
5283                           "unable to set label with more than %d bytes",
5284                           BTRFS_LABEL_SIZE - 1);
5285                 return -EINVAL;
5286         }
5287
5288         ret = mnt_want_write_file(file);
5289         if (ret)
5290                 return ret;
5291
5292         trans = btrfs_start_transaction(root, 0);
5293         if (IS_ERR(trans)) {
5294                 ret = PTR_ERR(trans);
5295                 goto out_unlock;
5296         }
5297
5298         spin_lock(&fs_info->super_lock);
5299         strcpy(super_block->label, label);
5300         spin_unlock(&fs_info->super_lock);
5301         ret = btrfs_commit_transaction(trans);
5302
5303 out_unlock:
5304         mnt_drop_write_file(file);
5305         return ret;
5306 }
5307
5308 #define INIT_FEATURE_FLAGS(suffix) \
5309         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
5310           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
5311           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
5312
5313 int btrfs_ioctl_get_supported_features(void __user *arg)
5314 {
5315         static const struct btrfs_ioctl_feature_flags features[3] = {
5316                 INIT_FEATURE_FLAGS(SUPP),
5317                 INIT_FEATURE_FLAGS(SAFE_SET),
5318                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
5319         };
5320
5321         if (copy_to_user(arg, &features, sizeof(features)))
5322                 return -EFAULT;
5323
5324         return 0;
5325 }
5326
5327 static int btrfs_ioctl_get_features(struct file *file, void __user *arg)
5328 {
5329         struct inode *inode = file_inode(file);
5330         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5331         struct btrfs_super_block *super_block = fs_info->super_copy;
5332         struct btrfs_ioctl_feature_flags features;
5333
5334         features.compat_flags = btrfs_super_compat_flags(super_block);
5335         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5336         features.incompat_flags = btrfs_super_incompat_flags(super_block);
5337
5338         if (copy_to_user(arg, &features, sizeof(features)))
5339                 return -EFAULT;
5340
5341         return 0;
5342 }
5343
5344 static int check_feature_bits(struct btrfs_fs_info *fs_info,
5345                               enum btrfs_feature_set set,
5346                               u64 change_mask, u64 flags, u64 supported_flags,
5347                               u64 safe_set, u64 safe_clear)
5348 {
5349         const char *type = btrfs_feature_set_names[set];
5350         char *names;
5351         u64 disallowed, unsupported;
5352         u64 set_mask = flags & change_mask;
5353         u64 clear_mask = ~flags & change_mask;
5354
5355         unsupported = set_mask & ~supported_flags;
5356         if (unsupported) {
5357                 names = btrfs_printable_features(set, unsupported);
5358                 if (names) {
5359                         btrfs_warn(fs_info,
5360                                    "this kernel does not support the %s feature bit%s",
5361                                    names, strchr(names, ',') ? "s" : "");
5362                         kfree(names);
5363                 } else
5364                         btrfs_warn(fs_info,
5365                                    "this kernel does not support %s bits 0x%llx",
5366                                    type, unsupported);
5367                 return -EOPNOTSUPP;
5368         }
5369
5370         disallowed = set_mask & ~safe_set;
5371         if (disallowed) {
5372                 names = btrfs_printable_features(set, disallowed);
5373                 if (names) {
5374                         btrfs_warn(fs_info,
5375                                    "can't set the %s feature bit%s while mounted",
5376                                    names, strchr(names, ',') ? "s" : "");
5377                         kfree(names);
5378                 } else
5379                         btrfs_warn(fs_info,
5380                                    "can't set %s bits 0x%llx while mounted",
5381                                    type, disallowed);
5382                 return -EPERM;
5383         }
5384
5385         disallowed = clear_mask & ~safe_clear;
5386         if (disallowed) {
5387                 names = btrfs_printable_features(set, disallowed);
5388                 if (names) {
5389                         btrfs_warn(fs_info,
5390                                    "can't clear the %s feature bit%s while mounted",
5391                                    names, strchr(names, ',') ? "s" : "");
5392                         kfree(names);
5393                 } else
5394                         btrfs_warn(fs_info,
5395                                    "can't clear %s bits 0x%llx while mounted",
5396                                    type, disallowed);
5397                 return -EPERM;
5398         }
5399
5400         return 0;
5401 }
5402
5403 #define check_feature(fs_info, change_mask, flags, mask_base)   \
5404 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,       \
5405                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
5406                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
5407                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5408
5409 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5410 {
5411         struct inode *inode = file_inode(file);
5412         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5413         struct btrfs_root *root = BTRFS_I(inode)->root;
5414         struct btrfs_super_block *super_block = fs_info->super_copy;
5415         struct btrfs_ioctl_feature_flags flags[2];
5416         struct btrfs_trans_handle *trans;
5417         u64 newflags;
5418         int ret;
5419
5420         if (!capable(CAP_SYS_ADMIN))
5421                 return -EPERM;
5422
5423         if (copy_from_user(flags, arg, sizeof(flags)))
5424                 return -EFAULT;
5425
5426         /* Nothing to do */
5427         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5428             !flags[0].incompat_flags)
5429                 return 0;
5430
5431         ret = check_feature(fs_info, flags[0].compat_flags,
5432                             flags[1].compat_flags, COMPAT);
5433         if (ret)
5434                 return ret;
5435
5436         ret = check_feature(fs_info, flags[0].compat_ro_flags,
5437                             flags[1].compat_ro_flags, COMPAT_RO);
5438         if (ret)
5439                 return ret;
5440
5441         ret = check_feature(fs_info, flags[0].incompat_flags,
5442                             flags[1].incompat_flags, INCOMPAT);
5443         if (ret)
5444                 return ret;
5445
5446         ret = mnt_want_write_file(file);
5447         if (ret)
5448                 return ret;
5449
5450         trans = btrfs_start_transaction(root, 0);
5451         if (IS_ERR(trans)) {
5452                 ret = PTR_ERR(trans);
5453                 goto out_drop_write;
5454         }
5455
5456         spin_lock(&fs_info->super_lock);
5457         newflags = btrfs_super_compat_flags(super_block);
5458         newflags |= flags[0].compat_flags & flags[1].compat_flags;
5459         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5460         btrfs_set_super_compat_flags(super_block, newflags);
5461
5462         newflags = btrfs_super_compat_ro_flags(super_block);
5463         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5464         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5465         btrfs_set_super_compat_ro_flags(super_block, newflags);
5466
5467         newflags = btrfs_super_incompat_flags(super_block);
5468         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5469         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5470         btrfs_set_super_incompat_flags(super_block, newflags);
5471         spin_unlock(&fs_info->super_lock);
5472
5473         ret = btrfs_commit_transaction(trans);
5474 out_drop_write:
5475         mnt_drop_write_file(file);
5476
5477         return ret;
5478 }
5479
5480 long btrfs_ioctl(struct file *file, unsigned int
5481                 cmd, unsigned long arg)
5482 {
5483         struct inode *inode = file_inode(file);
5484         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5485         struct btrfs_root *root = BTRFS_I(inode)->root;
5486         void __user *argp = (void __user *)arg;
5487
5488         switch (cmd) {
5489         case FS_IOC_GETFLAGS:
5490                 return btrfs_ioctl_getflags(file, argp);
5491         case FS_IOC_SETFLAGS:
5492                 return btrfs_ioctl_setflags(file, argp);
5493         case FS_IOC_GETVERSION:
5494                 return btrfs_ioctl_getversion(file, argp);
5495         case FITRIM:
5496                 return btrfs_ioctl_fitrim(file, argp);
5497         case BTRFS_IOC_SNAP_CREATE:
5498                 return btrfs_ioctl_snap_create(file, argp, 0);
5499         case BTRFS_IOC_SNAP_CREATE_V2:
5500                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
5501         case BTRFS_IOC_SUBVOL_CREATE:
5502                 return btrfs_ioctl_snap_create(file, argp, 1);
5503         case BTRFS_IOC_SUBVOL_CREATE_V2:
5504                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
5505         case BTRFS_IOC_SNAP_DESTROY:
5506                 return btrfs_ioctl_snap_destroy(file, argp);
5507         case BTRFS_IOC_SUBVOL_GETFLAGS:
5508                 return btrfs_ioctl_subvol_getflags(file, argp);
5509         case BTRFS_IOC_SUBVOL_SETFLAGS:
5510                 return btrfs_ioctl_subvol_setflags(file, argp);
5511         case BTRFS_IOC_DEFAULT_SUBVOL:
5512                 return btrfs_ioctl_default_subvol(file, argp);
5513         case BTRFS_IOC_DEFRAG:
5514                 return btrfs_ioctl_defrag(file, NULL);
5515         case BTRFS_IOC_DEFRAG_RANGE:
5516                 return btrfs_ioctl_defrag(file, argp);
5517         case BTRFS_IOC_RESIZE:
5518                 return btrfs_ioctl_resize(file, argp);
5519         case BTRFS_IOC_ADD_DEV:
5520                 return btrfs_ioctl_add_dev(fs_info, argp);
5521         case BTRFS_IOC_RM_DEV:
5522                 return btrfs_ioctl_rm_dev(file, argp);
5523         case BTRFS_IOC_RM_DEV_V2:
5524                 return btrfs_ioctl_rm_dev_v2(file, argp);
5525         case BTRFS_IOC_FS_INFO:
5526                 return btrfs_ioctl_fs_info(fs_info, argp);
5527         case BTRFS_IOC_DEV_INFO:
5528                 return btrfs_ioctl_dev_info(fs_info, argp);
5529         case BTRFS_IOC_BALANCE:
5530                 return btrfs_ioctl_balance(file, NULL);
5531         case BTRFS_IOC_TRANS_START:
5532                 return btrfs_ioctl_trans_start(file);
5533         case BTRFS_IOC_TRANS_END:
5534                 return btrfs_ioctl_trans_end(file);
5535         case BTRFS_IOC_TREE_SEARCH:
5536                 return btrfs_ioctl_tree_search(file, argp);
5537         case BTRFS_IOC_TREE_SEARCH_V2:
5538                 return btrfs_ioctl_tree_search_v2(file, argp);
5539         case BTRFS_IOC_INO_LOOKUP:
5540                 return btrfs_ioctl_ino_lookup(file, argp);
5541         case BTRFS_IOC_INO_PATHS:
5542                 return btrfs_ioctl_ino_to_path(root, argp);
5543         case BTRFS_IOC_LOGICAL_INO:
5544                 return btrfs_ioctl_logical_to_ino(fs_info, argp);
5545         case BTRFS_IOC_SPACE_INFO:
5546                 return btrfs_ioctl_space_info(fs_info, argp);
5547         case BTRFS_IOC_SYNC: {
5548                 int ret;
5549
5550                 ret = btrfs_start_delalloc_roots(fs_info, 0, -1);
5551                 if (ret)
5552                         return ret;
5553                 ret = btrfs_sync_fs(inode->i_sb, 1);
5554                 /*
5555                  * The transaction thread may want to do more work,
5556                  * namely it pokes the cleaner kthread that will start
5557                  * processing uncleaned subvols.
5558                  */
5559                 wake_up_process(fs_info->transaction_kthread);
5560                 return ret;
5561         }
5562         case BTRFS_IOC_START_SYNC:
5563                 return btrfs_ioctl_start_sync(root, argp);
5564         case BTRFS_IOC_WAIT_SYNC:
5565                 return btrfs_ioctl_wait_sync(fs_info, argp);
5566         case BTRFS_IOC_SCRUB:
5567                 return btrfs_ioctl_scrub(file, argp);
5568         case BTRFS_IOC_SCRUB_CANCEL:
5569                 return btrfs_ioctl_scrub_cancel(fs_info);
5570         case BTRFS_IOC_SCRUB_PROGRESS:
5571                 return btrfs_ioctl_scrub_progress(fs_info, argp);
5572         case BTRFS_IOC_BALANCE_V2:
5573                 return btrfs_ioctl_balance(file, argp);
5574         case BTRFS_IOC_BALANCE_CTL:
5575                 return btrfs_ioctl_balance_ctl(fs_info, arg);
5576         case BTRFS_IOC_BALANCE_PROGRESS:
5577                 return btrfs_ioctl_balance_progress(fs_info, argp);
5578         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5579                 return btrfs_ioctl_set_received_subvol(file, argp);
5580 #ifdef CONFIG_64BIT
5581         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5582                 return btrfs_ioctl_set_received_subvol_32(file, argp);
5583 #endif
5584         case BTRFS_IOC_SEND:
5585                 return btrfs_ioctl_send(file, argp);
5586         case BTRFS_IOC_GET_DEV_STATS:
5587                 return btrfs_ioctl_get_dev_stats(fs_info, argp);
5588         case BTRFS_IOC_QUOTA_CTL:
5589                 return btrfs_ioctl_quota_ctl(file, argp);
5590         case BTRFS_IOC_QGROUP_ASSIGN:
5591                 return btrfs_ioctl_qgroup_assign(file, argp);
5592         case BTRFS_IOC_QGROUP_CREATE:
5593                 return btrfs_ioctl_qgroup_create(file, argp);
5594         case BTRFS_IOC_QGROUP_LIMIT:
5595                 return btrfs_ioctl_qgroup_limit(file, argp);
5596         case BTRFS_IOC_QUOTA_RESCAN:
5597                 return btrfs_ioctl_quota_rescan(file, argp);
5598         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5599                 return btrfs_ioctl_quota_rescan_status(file, argp);
5600         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5601                 return btrfs_ioctl_quota_rescan_wait(file, argp);
5602         case BTRFS_IOC_DEV_REPLACE:
5603                 return btrfs_ioctl_dev_replace(fs_info, argp);
5604         case BTRFS_IOC_GET_FSLABEL:
5605                 return btrfs_ioctl_get_fslabel(file, argp);
5606         case BTRFS_IOC_SET_FSLABEL:
5607                 return btrfs_ioctl_set_fslabel(file, argp);
5608         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5609                 return btrfs_ioctl_get_supported_features(argp);
5610         case BTRFS_IOC_GET_FEATURES:
5611                 return btrfs_ioctl_get_features(file, argp);
5612         case BTRFS_IOC_SET_FEATURES:
5613                 return btrfs_ioctl_set_features(file, argp);
5614         }
5615
5616         return -ENOTTY;
5617 }
5618
5619 #ifdef CONFIG_COMPAT
5620 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
5621 {
5622         /*
5623          * These all access 32-bit values anyway so no further
5624          * handling is necessary.
5625          */
5626         switch (cmd) {
5627         case FS_IOC32_GETFLAGS:
5628                 cmd = FS_IOC_GETFLAGS;
5629                 break;
5630         case FS_IOC32_SETFLAGS:
5631                 cmd = FS_IOC_SETFLAGS;
5632                 break;
5633         case FS_IOC32_GETVERSION:
5634                 cmd = FS_IOC_GETVERSION;
5635                 break;
5636         }
5637
5638         return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
5639 }
5640 #endif