fs/btrfs/ioctl.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/fsnotify.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/highmem.h>
  27 #include <linux/time.h>
  28 #include <linux/init.h>
  29 #include <linux/string.h>
  30 #include <linux/backing-dev.h>
  31 #include <linux/mount.h>
  32 #include <linux/mpage.h>
  33 #include <linux/namei.h>
  34 #include <linux/swap.h>
  35 #include <linux/writeback.h>
  36 #include <linux/statfs.h>
  37 #include <linux/compat.h>
  38 #include <linux/bit_spinlock.h>
  39 #include <linux/security.h>
  40 #include <linux/xattr.h>
  41 #include <linux/vmalloc.h>
  42 #include <linux/slab.h>
  43 #include <linux/blkdev.h>
  44 #include <linux/uuid.h>
  45 #include <linux/btrfs.h>
  46 #include <linux/uaccess.h>
  47 #include "ctree.h"
  48 #include "disk-io.h"
  49 #include "transaction.h"
  50 #include "btrfs_inode.h"
  51 #include "print-tree.h"
  52 #include "volumes.h"
  53 #include "locking.h"
  54 #include "inode-map.h"
  55 #include "backref.h"
  56 #include "rcu-string.h"
  57 #include "send.h"
  58 #include "dev-replace.h"
  59 #include "props.h"
  60 #include "sysfs.h"
  61 #include "qgroup.h"
  62 #include "tree-log.h"
  63 #include "compression.h"
  64
  65 #ifdef CONFIG_64BIT
  66 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
  67  * structures are incorrect, as the timespec structure from userspace
  68  * is 4 bytes too small. We define these alternatives here to teach
  69  * the kernel about the 32-bit struct packing.
  70  */
  71 struct btrfs_ioctl_timespec_32 {
  72         __u64 sec;
  73         __u32 nsec;
  74 } __attribute__ ((__packed__));
  75
  76 struct btrfs_ioctl_received_subvol_args_32 {
  77         char    uuid[BTRFS_UUID_SIZE];  /* in */
  78         __u64   stransid;               /* in */
  79         __u64   rtransid;               /* out */
  80         struct btrfs_ioctl_timespec_32 stime; /* in */
  81         struct btrfs_ioctl_timespec_32 rtime; /* out */
  82         __u64   flags;                  /* in */
  83         __u64   reserved[16];           /* in */
  84 } __attribute__ ((__packed__));
  85
  86 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
  87                                 struct btrfs_ioctl_received_subvol_args_32)
  88 #endif
  89
  90
  91 static int btrfs_clone(struct inode *src, struct inode *inode,
  92                        u64 off, u64 olen, u64 olen_aligned, u64 destoff,
  93                        int no_time_update);
  94
  95 /* Mask out flags that are inappropriate for the given type of inode. */
  96 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
  97 {
  98         if (S_ISDIR(mode))
  99                 return flags;
 100         else if (S_ISREG(mode))
 101                 return flags & ~FS_DIRSYNC_FL;
 102         else
 103                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
 104 }
 105
 106 /*
 107  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
 108  */
 109 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
 110 {
 111         unsigned int iflags = 0;
 112
 113         if (flags & BTRFS_INODE_SYNC)
 114                 iflags |= FS_SYNC_FL;
 115         if (flags & BTRFS_INODE_IMMUTABLE)
 116                 iflags |= FS_IMMUTABLE_FL;
 117         if (flags & BTRFS_INODE_APPEND)
 118                 iflags |= FS_APPEND_FL;
 119         if (flags & BTRFS_INODE_NODUMP)
 120                 iflags |= FS_NODUMP_FL;
 121         if (flags & BTRFS_INODE_NOATIME)
 122                 iflags |= FS_NOATIME_FL;
 123         if (flags & BTRFS_INODE_DIRSYNC)
 124                 iflags |= FS_DIRSYNC_FL;
 125         if (flags & BTRFS_INODE_NODATACOW)
 126                 iflags |= FS_NOCOW_FL;
 127
 128         if ((flags & BTRFS_INODE_COMPRESS) && !(flags & BTRFS_INODE_NOCOMPRESS))
 129                 iflags |= FS_COMPR_FL;
 130         else if (flags & BTRFS_INODE_NOCOMPRESS)
 131                 iflags |= FS_NOCOMP_FL;
 132
 133         return iflags;
 134 }
 135
 136 /*
 137  * Update inode->i_flags based on the btrfs internal flags.
 138  */
 139 void btrfs_update_iflags(struct inode *inode)
 140 {
 141         struct btrfs_inode *ip = BTRFS_I(inode);
 142         unsigned int new_fl = 0;
 143
 144         if (ip->flags & BTRFS_INODE_SYNC)
 145                 new_fl |= S_SYNC;
 146         if (ip->flags & BTRFS_INODE_IMMUTABLE)
 147                 new_fl |= S_IMMUTABLE;
 148         if (ip->flags & BTRFS_INODE_APPEND)
 149                 new_fl |= S_APPEND;
 150         if (ip->flags & BTRFS_INODE_NOATIME)
 151                 new_fl |= S_NOATIME;
 152         if (ip->flags & BTRFS_INODE_DIRSYNC)
 153                 new_fl |= S_DIRSYNC;
 154
 155         set_mask_bits(&inode->i_flags,
 156                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC,
 157                       new_fl);
 158 }
 159
 160 /*
 161  * Inherit flags from the parent inode.
 162  *
 163  * Currently only the compression flags and the cow flags are inherited.
 164  */
 165 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
 166 {
 167         unsigned int flags;
 168
 169         if (!dir)
 170                 return;
 171
 172         flags = BTRFS_I(dir)->flags;
 173
 174         if (flags & BTRFS_INODE_NOCOMPRESS) {
 175                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
 176                 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 177         } else if (flags & BTRFS_INODE_COMPRESS) {
 178                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
 179                 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
 180         }
 181
 182         if (flags & BTRFS_INODE_NODATACOW) {
 183                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
 184                 if (S_ISREG(inode->i_mode))
 185                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
 186         }
 187
 188         btrfs_update_iflags(inode);
 189 }
 190
 191 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 192 {
 193         struct btrfs_inode *ip = BTRFS_I(file_inode(file));
 194         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
 195
 196         if (copy_to_user(arg, &flags, sizeof(flags)))
 197                 return -EFAULT;
 198         return 0;
 199 }
 200
 201 static int check_flags(unsigned int flags)
 202 {
 203         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 204                       FS_NOATIME_FL | FS_NODUMP_FL | \
 205                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 206                       FS_NOCOMP_FL | FS_COMPR_FL |
 207                       FS_NOCOW_FL))
 208                 return -EOPNOTSUPP;
 209
 210         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 211                 return -EINVAL;
 212
 213         return 0;
 214 }
 215
 216 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 217 {
 218         struct inode *inode = file_inode(file);
 219         struct btrfs_inode *ip = BTRFS_I(inode);
 220         struct btrfs_root *root = ip->root;
 221         struct btrfs_trans_handle *trans;
 222         unsigned int flags, oldflags;
 223         int ret;
 224         u64 ip_oldflags;
 225         unsigned int i_oldflags;
 226         umode_t mode;
 227
 228         if (!inode_owner_or_capable(inode))
 229                 return -EPERM;
 230
 231         if (btrfs_root_readonly(root))
 232                 return -EROFS;
 233
 234         if (copy_from_user(&flags, arg, sizeof(flags)))
 235                 return -EFAULT;
 236
 237         ret = check_flags(flags);
 238         if (ret)
 239                 return ret;
 240
 241         ret = mnt_want_write_file(file);
 242         if (ret)
 243                 return ret;
 244
 245         inode_lock(inode);
 246
 247         ip_oldflags = ip->flags;
 248         i_oldflags = inode->i_flags;
 249         mode = inode->i_mode;
 250
 251         flags = btrfs_mask_flags(inode->i_mode, flags);
 252         oldflags = btrfs_flags_to_ioctl(ip->flags);
 253         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 254                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 255                         ret = -EPERM;
 256                         goto out_unlock;
 257                 }
 258         }
 259
 260         if (flags & FS_SYNC_FL)
 261                 ip->flags |= BTRFS_INODE_SYNC;
 262         else
 263                 ip->flags &= ~BTRFS_INODE_SYNC;
 264         if (flags & FS_IMMUTABLE_FL)
 265                 ip->flags |= BTRFS_INODE_IMMUTABLE;
 266         else
 267                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
 268         if (flags & FS_APPEND_FL)
 269                 ip->flags |= BTRFS_INODE_APPEND;
 270         else
 271                 ip->flags &= ~BTRFS_INODE_APPEND;
 272         if (flags & FS_NODUMP_FL)
 273                 ip->flags |= BTRFS_INODE_NODUMP;
 274         else
 275                 ip->flags &= ~BTRFS_INODE_NODUMP;
 276         if (flags & FS_NOATIME_FL)
 277                 ip->flags |= BTRFS_INODE_NOATIME;
 278         else
 279                 ip->flags &= ~BTRFS_INODE_NOATIME;
 280         if (flags & FS_DIRSYNC_FL)
 281                 ip->flags |= BTRFS_INODE_DIRSYNC;
 282         else
 283                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
 284         if (flags & FS_NOCOW_FL) {
 285                 if (S_ISREG(mode)) {
 286                         /*
 287                          * It's safe to turn csums off here, no extents exist.
 288                          * Otherwise we want the flag to reflect the real COW
 289                          * status of the file and will not set it.
 290                          */
 291                         if (inode->i_size == 0)
 292                                 ip->flags |= BTRFS_INODE_NODATACOW
 293                                            | BTRFS_INODE_NODATASUM;
 294                 } else {
 295                         ip->flags |= BTRFS_INODE_NODATACOW;
 296                 }
 297         } else {
 298                 /*
 299                  * Revert back under same assuptions as above
 300                  */
 301                 if (S_ISREG(mode)) {
 302                         if (inode->i_size == 0)
 303                                 ip->flags &= ~(BTRFS_INODE_NODATACOW
 304                                              | BTRFS_INODE_NODATASUM);
 305                 } else {
 306                         ip->flags &= ~BTRFS_INODE_NODATACOW;
 307                 }
 308         }
 309
 310         /*
 311          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 312          * flag may be changed automatically if compression code won't make
 313          * things smaller.
 314          */
 315         if (flags & FS_NOCOMP_FL) {
 316                 ip->flags &= ~BTRFS_INODE_COMPRESS;
 317                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
 318
 319                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 320                 if (ret && ret != -ENODATA)
 321                         goto out_drop;
 322         } else if (flags & FS_COMPR_FL) {
 323                 const char *comp;
 324
 325                 ip->flags |= BTRFS_INODE_COMPRESS;
 326                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
 327
 328                 if (root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
 329                         comp = "lzo";
 330                 else
 331                         comp = "zlib";
 332                 ret = btrfs_set_prop(inode, "btrfs.compression",
 333                                      comp, strlen(comp), 0);
 334                 if (ret)
 335                         goto out_drop;
 336
 337         } else {
 338                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 339                 if (ret && ret != -ENODATA)
 340                         goto out_drop;
 341                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 342         }
 343
 344         trans = btrfs_start_transaction(root, 1);
 345         if (IS_ERR(trans)) {
 346                 ret = PTR_ERR(trans);
 347                 goto out_drop;
 348         }
 349
 350         btrfs_update_iflags(inode);
 351         inode_inc_iversion(inode);
 352         inode->i_ctime = current_fs_time(inode->i_sb);
 353         ret = btrfs_update_inode(trans, root, inode);
 354
 355         btrfs_end_transaction(trans, root);
 356  out_drop:
 357         if (ret) {
 358                 ip->flags = ip_oldflags;
 359                 inode->i_flags = i_oldflags;
 360         }
 361
 362  out_unlock:
 363         inode_unlock(inode);
 364         mnt_drop_write_file(file);
 365         return ret;
 366 }
 367
 368 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 369 {
 370         struct inode *inode = file_inode(file);
 371
 372         return put_user(inode->i_generation, arg);
 373 }
 374
 375 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
 376 {
 377         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
 378         struct btrfs_device *device;
 379         struct request_queue *q;
 380         struct fstrim_range range;
 381         u64 minlen = ULLONG_MAX;
 382         u64 num_devices = 0;
 383         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
 384         int ret;
 385
 386         if (!capable(CAP_SYS_ADMIN))
 387                 return -EPERM;
 388
 389         rcu_read_lock();
 390         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 391                                 dev_list) {
 392                 if (!device->bdev)
 393                         continue;
 394                 q = bdev_get_queue(device->bdev);
 395                 if (blk_queue_discard(q)) {
 396                         num_devices++;
 397                         minlen = min((u64)q->limits.discard_granularity,
 398                                      minlen);
 399                 }
 400         }
 401         rcu_read_unlock();
 402
 403         if (!num_devices)
 404                 return -EOPNOTSUPP;
 405         if (copy_from_user(&range, arg, sizeof(range)))
 406                 return -EFAULT;
 407         if (range.start > total_bytes ||
 408             range.len < fs_info->sb->s_blocksize)
 409                 return -EINVAL;
 410
 411         range.len = min(range.len, total_bytes - range.start);
 412         range.minlen = max(range.minlen, minlen);
 413         ret = btrfs_trim_fs(fs_info->tree_root, &range);
 414         if (ret < 0)
 415                 return ret;
 416
 417         if (copy_to_user(arg, &range, sizeof(range)))
 418                 return -EFAULT;
 419
 420         return 0;
 421 }
 422
 423 int btrfs_is_empty_uuid(u8 *uuid)
 424 {
 425         int i;
 426
 427         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 428                 if (uuid[i])
 429                         return 0;
 430         }
 431         return 1;
 432 }
 433
 434 static noinline int create_subvol(struct inode *dir,
 435                                   struct dentry *dentry,
 436                                   char *name, int namelen,
 437                                   u64 *async_transid,
 438                                   struct btrfs_qgroup_inherit *inherit)
 439 {
 440         struct btrfs_trans_handle *trans;
 441         struct btrfs_key key;
 442         struct btrfs_root_item root_item;
 443         struct btrfs_inode_item *inode_item;
 444         struct extent_buffer *leaf;
 445         struct btrfs_root *root = BTRFS_I(dir)->root;
 446         struct btrfs_root *new_root;
 447         struct btrfs_block_rsv block_rsv;
 448         struct timespec cur_time = current_fs_time(dir->i_sb);
 449         struct inode *inode;
 450         int ret;
 451         int err;
 452         u64 objectid;
 453         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 454         u64 index = 0;
 455         u64 qgroup_reserved;
 456         uuid_le new_uuid;
 457
 458         ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid);
 459         if (ret)
 460                 return ret;
 461
 462         /*
 463          * Don't create subvolume whose level is not zero. Or qgroup will be
 464          * screwed up since it assume subvolme qgroup's level to be 0.
 465          */
 466         if (btrfs_qgroup_level(objectid))
 467                 return -ENOSPC;
 468
 469         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 470         /*
 471          * The same as the snapshot creation, please see the comment
 472          * of create_snapshot().
 473          */
 474         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
 475                                                8, &qgroup_reserved, false);
 476         if (ret)
 477                 return ret;
 478
 479         trans = btrfs_start_transaction(root, 0);
 480         if (IS_ERR(trans)) {
 481                 ret = PTR_ERR(trans);
 482                 btrfs_subvolume_release_metadata(root, &block_rsv,
 483                                                  qgroup_reserved);
 484                 return ret;
 485         }
 486         trans->block_rsv = &block_rsv;
 487         trans->bytes_reserved = block_rsv.size;
 488
 489         ret = btrfs_qgroup_inherit(trans, root->fs_info, 0, objectid, inherit);
 490         if (ret)
 491                 goto fail;
 492
 493         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
 494         if (IS_ERR(leaf)) {
 495                 ret = PTR_ERR(leaf);
 496                 goto fail;
 497         }
 498
 499         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
 500         btrfs_set_header_bytenr(leaf, leaf->start);
 501         btrfs_set_header_generation(leaf, trans->transid);
 502         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 503         btrfs_set_header_owner(leaf, objectid);
 504
 505         write_extent_buffer(leaf, root->fs_info->fsid, btrfs_header_fsid(),
 506                             BTRFS_FSID_SIZE);
 507         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 508                             btrfs_header_chunk_tree_uuid(leaf),
 509                             BTRFS_UUID_SIZE);
 510         btrfs_mark_buffer_dirty(leaf);
 511
 512         memset(&root_item, 0, sizeof(root_item));
 513
 514         inode_item = &root_item.inode;
 515         btrfs_set_stack_inode_generation(inode_item, 1);
 516         btrfs_set_stack_inode_size(inode_item, 3);
 517         btrfs_set_stack_inode_nlink(inode_item, 1);
 518         btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
 519         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 520
 521         btrfs_set_root_flags(&root_item, 0);
 522         btrfs_set_root_limit(&root_item, 0);
 523         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 524
 525         btrfs_set_root_bytenr(&root_item, leaf->start);
 526         btrfs_set_root_generation(&root_item, trans->transid);
 527         btrfs_set_root_level(&root_item, 0);
 528         btrfs_set_root_refs(&root_item, 1);
 529         btrfs_set_root_used(&root_item, leaf->len);
 530         btrfs_set_root_last_snapshot(&root_item, 0);
 531
 532         btrfs_set_root_generation_v2(&root_item,
 533                         btrfs_root_generation(&root_item));
 534         uuid_le_gen(&new_uuid);
 535         memcpy(root_item.uuid, new_uuid.b, BTRFS_UUID_SIZE);
 536         btrfs_set_stack_timespec_sec(&root_item.otime, cur_time.tv_sec);
 537         btrfs_set_stack_timespec_nsec(&root_item.otime, cur_time.tv_nsec);
 538         root_item.ctime = root_item.otime;
 539         btrfs_set_root_ctransid(&root_item, trans->transid);
 540         btrfs_set_root_otransid(&root_item, trans->transid);
 541
 542         btrfs_tree_unlock(leaf);
 543         free_extent_buffer(leaf);
 544         leaf = NULL;
 545
 546         btrfs_set_root_dirid(&root_item, new_dirid);
 547
 548         key.objectid = objectid;
 549         key.offset = 0;
 550         key.type = BTRFS_ROOT_ITEM_KEY;
 551         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 552                                 &root_item);
 553         if (ret)
 554                 goto fail;
 555
 556         key.offset = (u64)-1;
 557         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
 558         if (IS_ERR(new_root)) {
 559                 ret = PTR_ERR(new_root);
 560                 btrfs_abort_transaction(trans, root, ret);
 561                 goto fail;
 562         }
 563
 564         btrfs_record_root_in_trans(trans, new_root);
 565
 566         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
 567         if (ret) {
 568                 /* We potentially lose an unused inode item here */
 569                 btrfs_abort_transaction(trans, root, ret);
 570                 goto fail;
 571         }
 572
 573         mutex_lock(&new_root->objectid_mutex);
 574         new_root->highest_objectid = new_dirid;
 575         mutex_unlock(&new_root->objectid_mutex);
 576
 577         /*
 578          * insert the directory item
 579          */
 580         ret = btrfs_set_inode_index(dir, &index);
 581         if (ret) {
 582                 btrfs_abort_transaction(trans, root, ret);
 583                 goto fail;
 584         }
 585
 586         ret = btrfs_insert_dir_item(trans, root,
 587                                     name, namelen, dir, &key,
 588                                     BTRFS_FT_DIR, index);
 589         if (ret) {
 590                 btrfs_abort_transaction(trans, root, ret);
 591                 goto fail;
 592         }
 593
 594         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
 595         ret = btrfs_update_inode(trans, root, dir);
 596         BUG_ON(ret);
 597
 598         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
 599                                  objectid, root->root_key.objectid,
 600                                  btrfs_ino(dir), index, name, namelen);
 601         BUG_ON(ret);
 602
 603         ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
 604                                   root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
 605                                   objectid);
 606         if (ret)
 607                 btrfs_abort_transaction(trans, root, ret);
 608
 609 fail:
 610         trans->block_rsv = NULL;
 611         trans->bytes_reserved = 0;
 612         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
 613
 614         if (async_transid) {
 615                 *async_transid = trans->transid;
 616                 err = btrfs_commit_transaction_async(trans, root, 1);
 617                 if (err)
 618                         err = btrfs_commit_transaction(trans, root);
 619         } else {
 620                 err = btrfs_commit_transaction(trans, root);
 621         }
 622         if (err && !ret)
 623                 ret = err;
 624
 625         if (!ret) {
 626                 inode = btrfs_lookup_dentry(dir, dentry);
 627                 if (IS_ERR(inode))
 628                         return PTR_ERR(inode);
 629                 d_instantiate(dentry, inode);
 630         }
 631         return ret;
 632 }
 633
 634 static void btrfs_wait_for_no_snapshoting_writes(struct btrfs_root *root)
 635 {
 636         s64 writers;
 637         DEFINE_WAIT(wait);
 638
 639         do {
 640                 prepare_to_wait(&root->subv_writers->wait, &wait,
 641                                 TASK_UNINTERRUPTIBLE);
 642
 643                 writers = percpu_counter_sum(&root->subv_writers->counter);
 644                 if (writers)
 645                         schedule();
 646
 647                 finish_wait(&root->subv_writers->wait, &wait);
 648         } while (writers);
 649 }
 650
 651 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 652                            struct dentry *dentry, char *name, int namelen,
 653                            u64 *async_transid, bool readonly,
 654                            struct btrfs_qgroup_inherit *inherit)
 655 {
 656         struct inode *inode;
 657         struct btrfs_pending_snapshot *pending_snapshot;
 658         struct btrfs_trans_handle *trans;
 659         int ret;
 660
 661         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 662                 return -EINVAL;
 663
 664         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
 665         if (!pending_snapshot)
 666                 return -ENOMEM;
 667
 668         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 669                         GFP_NOFS);
 670         pending_snapshot->path = btrfs_alloc_path();
 671         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 672                 ret = -ENOMEM;
 673                 goto free_pending;
 674         }
 675
 676         atomic_inc(&root->will_be_snapshoted);
 677         smp_mb__after_atomic();
 678         btrfs_wait_for_no_snapshoting_writes(root);
 679
 680         ret = btrfs_start_delalloc_inodes(root, 0);
 681         if (ret)
 682                 goto dec_and_free;
 683
 684         btrfs_wait_ordered_extents(root, -1);
 685
 686         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 687                              BTRFS_BLOCK_RSV_TEMP);
 688         /*
 689          * 1 - parent dir inode
 690          * 2 - dir entries
 691          * 1 - root item
 692          * 2 - root ref/backref
 693          * 1 - root of snapshot
 694          * 1 - UUID item
 695          */
 696         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 697                                         &pending_snapshot->block_rsv, 8,
 698                                         &pending_snapshot->qgroup_reserved,
 699                                         false);
 700         if (ret)
 701                 goto dec_and_free;
 702
 703         pending_snapshot->dentry = dentry;
 704         pending_snapshot->root = root;
 705         pending_snapshot->readonly = readonly;
 706         pending_snapshot->dir = dir;
 707         pending_snapshot->inherit = inherit;
 708
 709         trans = btrfs_start_transaction(root, 0);
 710         if (IS_ERR(trans)) {
 711                 ret = PTR_ERR(trans);
 712                 goto fail;
 713         }
 714
 715         spin_lock(&root->fs_info->trans_lock);
 716         list_add(&pending_snapshot->list,
 717                  &trans->transaction->pending_snapshots);
 718         spin_unlock(&root->fs_info->trans_lock);
 719         if (async_transid) {
 720                 *async_transid = trans->transid;
 721                 ret = btrfs_commit_transaction_async(trans,
 722                                      root->fs_info->extent_root, 1);
 723                 if (ret)
 724                         ret = btrfs_commit_transaction(trans, root);
 725         } else {
 726                 ret = btrfs_commit_transaction(trans,
 727                                                root->fs_info->extent_root);
 728         }
 729         if (ret)
 730                 goto fail;
 731
 732         ret = pending_snapshot->error;
 733         if (ret)
 734                 goto fail;
 735
 736         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 737         if (ret)
 738                 goto fail;
 739
 740         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 741         if (IS_ERR(inode)) {
 742                 ret = PTR_ERR(inode);
 743                 goto fail;
 744         }
 745
 746         d_instantiate(dentry, inode);
 747         ret = 0;
 748 fail:
 749         btrfs_subvolume_release_metadata(BTRFS_I(dir)->root,
 750                                          &pending_snapshot->block_rsv,
 751                                          pending_snapshot->qgroup_reserved);
 752 dec_and_free:
 753         if (atomic_dec_and_test(&root->will_be_snapshoted))
 754                 wake_up_atomic_t(&root->will_be_snapshoted);
 755 free_pending:
 756         kfree(pending_snapshot->root_item);
 757         btrfs_free_path(pending_snapshot->path);
 758         kfree(pending_snapshot);
 759
 760         return ret;
 761 }
 762
 763 /*  copy of may_delete in fs/namei.c()
 764  *      Check whether we can remove a link victim from directory dir, check
 765  *  whether the type of victim is right.
 766  *  1. We can't do it if dir is read-only (done in permission())
 767  *  2. We should have write and exec permissions on dir
 768  *  3. We can't remove anything from append-only dir
 769  *  4. We can't do anything with immutable dir (done in permission())
 770  *  5. If the sticky bit on dir is set we should either
 771  *      a. be owner of dir, or
 772  *      b. be owner of victim, or
 773  *      c. have CAP_FOWNER capability
 774  *  6. If the victim is append-only or immutable we can't do antyhing with
 775  *     links pointing to it.
 776  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 777  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 778  *  9. We can't remove a root or mountpoint.
 779  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 780  *     nfs_async_unlink().
 781  */
 782
 783 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 784 {
 785         int error;
 786
 787         if (d_really_is_negative(victim))
 788                 return -ENOENT;
 789
 790         BUG_ON(d_inode(victim->d_parent) != dir);
 791         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 792
 793         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 794         if (error)
 795                 return error;
 796         if (IS_APPEND(dir))
 797                 return -EPERM;
 798         if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
 799             IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
 800                 return -EPERM;
 801         if (isdir) {
 802                 if (!d_is_dir(victim))
 803                         return -ENOTDIR;
 804                 if (IS_ROOT(victim))
 805                         return -EBUSY;
 806         } else if (d_is_dir(victim))
 807                 return -EISDIR;
 808         if (IS_DEADDIR(dir))
 809                 return -ENOENT;
 810         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 811                 return -EBUSY;
 812         return 0;
 813 }
 814
 815 /* copy of may_create in fs/namei.c() */
 816 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 817 {
 818         if (d_really_is_positive(child))
 819                 return -EEXIST;
 820         if (IS_DEADDIR(dir))
 821                 return -ENOENT;
 822         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 823 }
 824
 825 /*
 826  * Create a new subvolume below @parent.  This is largely modeled after
 827  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 828  * inside this filesystem so it's quite a bit simpler.
 829  */
 830 static noinline int btrfs_mksubvol(struct path *parent,
 831                                    char *name, int namelen,
 832                                    struct btrfs_root *snap_src,
 833                                    u64 *async_transid, bool readonly,
 834                                    struct btrfs_qgroup_inherit *inherit)
 835 {
 836         struct inode *dir  = d_inode(parent->dentry);
 837         struct dentry *dentry;
 838         int error;
 839
 840         inode_lock_nested(dir, I_MUTEX_PARENT);
 841         // XXX: should've been
 842         // mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
 843         // if (error == -EINTR)
 844         //      return error;
 845
 846         dentry = lookup_one_len(name, parent->dentry, namelen);
 847         error = PTR_ERR(dentry);
 848         if (IS_ERR(dentry))
 849                 goto out_unlock;
 850
 851         error = btrfs_may_create(dir, dentry);
 852         if (error)
 853                 goto out_dput;
 854
 855         /*
 856          * even if this name doesn't exist, we may get hash collisions.
 857          * check for them now when we can safely fail
 858          */
 859         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 860                                                dir->i_ino, name,
 861                                                namelen);
 862         if (error)
 863                 goto out_dput;
 864
 865         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 866
 867         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 868                 goto out_up_read;
 869
 870         if (snap_src) {
 871                 error = create_snapshot(snap_src, dir, dentry, name, namelen,
 872                                         async_transid, readonly, inherit);
 873         } else {
 874                 error = create_subvol(dir, dentry, name, namelen,
 875                                       async_transid, inherit);
 876         }
 877         if (!error)
 878                 fsnotify_mkdir(dir, dentry);
 879 out_up_read:
 880         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 881 out_dput:
 882         dput(dentry);
 883 out_unlock:
 884         inode_unlock(dir);
 885         return error;
 886 }
 887
 888 /*
 889  * When we're defragging a range, we don't want to kick it off again
 890  * if it is really just waiting for delalloc to send it down.
 891  * If we find a nice big extent or delalloc range for the bytes in the
 892  * file you want to defrag, we return 0 to let you know to skip this
 893  * part of the file
 894  */
 895 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
 896 {
 897         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 898         struct extent_map *em = NULL;
 899         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 900         u64 end;
 901
 902         read_lock(&em_tree->lock);
 903         em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
 904         read_unlock(&em_tree->lock);
 905
 906         if (em) {
 907                 end = extent_map_end(em);
 908                 free_extent_map(em);
 909                 if (end - offset > thresh)
 910                         return 0;
 911         }
 912         /* if we already have a nice delalloc here, just stop */
 913         thresh /= 2;
 914         end = count_range_bits(io_tree, &offset, offset + thresh,
 915                                thresh, EXTENT_DELALLOC, 1);
 916         if (end >= thresh)
 917                 return 0;
 918         return 1;
 919 }
 920
 921 /*
 922  * helper function to walk through a file and find extents
 923  * newer than a specific transid, and smaller than thresh.
 924  *
 925  * This is used by the defragging code to find new and small
 926  * extents
 927  */
 928 static int find_new_extents(struct btrfs_root *root,
 929                             struct inode *inode, u64 newer_than,
 930                             u64 *off, u32 thresh)
 931 {
 932         struct btrfs_path *path;
 933         struct btrfs_key min_key;
 934         struct extent_buffer *leaf;
 935         struct btrfs_file_extent_item *extent;
 936         int type;
 937         int ret;
 938         u64 ino = btrfs_ino(inode);
 939
 940         path = btrfs_alloc_path();
 941         if (!path)
 942                 return -ENOMEM;
 943
 944         min_key.objectid = ino;
 945         min_key.type = BTRFS_EXTENT_DATA_KEY;
 946         min_key.offset = *off;
 947
 948         while (1) {
 949                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
 950                 if (ret != 0)
 951                         goto none;
 952 process_slot:
 953                 if (min_key.objectid != ino)
 954                         goto none;
 955                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
 956                         goto none;
 957
 958                 leaf = path->nodes[0];
 959                 extent = btrfs_item_ptr(leaf, path->slots[0],
 960                                         struct btrfs_file_extent_item);
 961
 962                 type = btrfs_file_extent_type(leaf, extent);
 963                 if (type == BTRFS_FILE_EXTENT_REG &&
 964                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
 965                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
 966                         *off = min_key.offset;
 967                         btrfs_free_path(path);
 968                         return 0;
 969                 }
 970
 971                 path->slots[0]++;
 972                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
 973                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
 974                         goto process_slot;
 975                 }
 976
 977                 if (min_key.offset == (u64)-1)
 978                         goto none;
 979
 980                 min_key.offset++;
 981                 btrfs_release_path(path);
 982         }
 983 none:
 984         btrfs_free_path(path);
 985         return -ENOENT;
 986 }
 987
 988 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
 989 {
 990         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 991         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 992         struct extent_map *em;
 993         u64 len = PAGE_SIZE;
 994
 995         /*
 996          * hopefully we have this extent in the tree already, try without
 997          * the full extent lock
 998          */
 999         read_lock(&em_tree->lock);
1000         em = lookup_extent_mapping(em_tree, start, len);
1001         read_unlock(&em_tree->lock);
1002
1003         if (!em) {
1004                 struct extent_state *cached = NULL;
1005                 u64 end = start + len - 1;
1006
1007                 /* get the big lock and read metadata off disk */
1008                 lock_extent_bits(io_tree, start, end, &cached);
1009                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
1010                 unlock_extent_cached(io_tree, start, end, &cached, GFP_NOFS);
1011
1012                 if (IS_ERR(em))
1013                         return NULL;
1014         }
1015
1016         return em;
1017 }
1018
1019 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1020 {
1021         struct extent_map *next;
1022         bool ret = true;
1023
1024         /* this is the last extent */
1025         if (em->start + em->len >= i_size_read(inode))
1026                 return false;
1027
1028         next = defrag_lookup_extent(inode, em->start + em->len);
1029         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1030                 ret = false;
1031         else if ((em->block_start + em->block_len == next->block_start) &&
1032                  (em->block_len > SZ_128K && next->block_len > SZ_128K))
1033                 ret = false;
1034
1035         free_extent_map(next);
1036         return ret;
1037 }
1038
1039 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1040                                u64 *last_len, u64 *skip, u64 *defrag_end,
1041                                int compress)
1042 {
1043         struct extent_map *em;
1044         int ret = 1;
1045         bool next_mergeable = true;
1046         bool prev_mergeable = true;
1047
1048         /*
1049          * make sure that once we start defragging an extent, we keep on
1050          * defragging it
1051          */
1052         if (start < *defrag_end)
1053                 return 1;
1054
1055         *skip = 0;
1056
1057         em = defrag_lookup_extent(inode, start);
1058         if (!em)
1059                 return 0;
1060
1061         /* this will cover holes, and inline extents */
1062         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1063                 ret = 0;
1064                 goto out;
1065         }
1066
1067         if (!*defrag_end)
1068                 prev_mergeable = false;
1069
1070         next_mergeable = defrag_check_next_extent(inode, em);
1071         /*
1072          * we hit a real extent, if it is big or the next extent is not a
1073          * real extent, don't bother defragging it
1074          */
1075         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1076             (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1077                 ret = 0;
1078 out:
1079         /*
1080          * last_len ends up being a counter of how many bytes we've defragged.
1081          * every time we choose not to defrag an extent, we reset *last_len
1082          * so that the next tiny extent will force a defrag.
1083          *
1084          * The end result of this is that tiny extents before a single big
1085          * extent will force at least part of that big extent to be defragged.
1086          */
1087         if (ret) {
1088                 *defrag_end = extent_map_end(em);
1089         } else {
1090                 *last_len = 0;
1091                 *skip = extent_map_end(em);
1092                 *defrag_end = 0;
1093         }
1094
1095         free_extent_map(em);
1096         return ret;
1097 }
1098
1099 /*
1100  * it doesn't do much good to defrag one or two pages
1101  * at a time.  This pulls in a nice chunk of pages
1102  * to COW and defrag.
1103  *
1104  * It also makes sure the delalloc code has enough
1105  * dirty data to avoid making new small extents as part
1106  * of the defrag
1107  *
1108  * It's a good idea to start RA on this range
1109  * before calling this.
1110  */
1111 static int cluster_pages_for_defrag(struct inode *inode,
1112                                     struct page **pages,
1113                                     unsigned long start_index,
1114                                     unsigned long num_pages)
1115 {
1116         unsigned long file_end;
1117         u64 isize = i_size_read(inode);
1118         u64 page_start;
1119         u64 page_end;
1120         u64 page_cnt;
1121         int ret;
1122         int i;
1123         int i_done;
1124         struct btrfs_ordered_extent *ordered;
1125         struct extent_state *cached_state = NULL;
1126         struct extent_io_tree *tree;
1127         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1128
1129         file_end = (isize - 1) >> PAGE_SHIFT;
1130         if (!isize || start_index > file_end)
1131                 return 0;
1132
1133         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1134
1135         ret = btrfs_delalloc_reserve_space(inode,
1136                         start_index << PAGE_SHIFT,
1137                         page_cnt << PAGE_SHIFT);
1138         if (ret)
1139                 return ret;
1140         i_done = 0;
1141         tree = &BTRFS_I(inode)->io_tree;
1142
1143         /* step one, lock all the pages */
1144         for (i = 0; i < page_cnt; i++) {
1145                 struct page *page;
1146 again:
1147                 page = find_or_create_page(inode->i_mapping,
1148                                            start_index + i, mask);
1149                 if (!page)
1150                         break;
1151
1152                 page_start = page_offset(page);
1153                 page_end = page_start + PAGE_SIZE - 1;
1154                 while (1) {
1155                         lock_extent_bits(tree, page_start, page_end,
1156                                          &cached_state);
1157                         ordered = btrfs_lookup_ordered_extent(inode,
1158                                                               page_start);
1159                         unlock_extent_cached(tree, page_start, page_end,
1160                                              &cached_state, GFP_NOFS);
1161                         if (!ordered)
1162                                 break;
1163
1164                         unlock_page(page);
1165                         btrfs_start_ordered_extent(inode, ordered, 1);
1166                         btrfs_put_ordered_extent(ordered);
1167                         lock_page(page);
1168                         /*
1169                          * we unlocked the page above, so we need check if
1170                          * it was released or not.
1171                          */
1172                         if (page->mapping != inode->i_mapping) {
1173                                 unlock_page(page);
1174                                 put_page(page);
1175                                 goto again;
1176                         }
1177                 }
1178
1179                 if (!PageUptodate(page)) {
1180                         btrfs_readpage(NULL, page);
1181                         lock_page(page);
1182                         if (!PageUptodate(page)) {
1183                                 unlock_page(page);
1184                                 put_page(page);
1185                                 ret = -EIO;
1186                                 break;
1187                         }
1188                 }
1189
1190                 if (page->mapping != inode->i_mapping) {
1191                         unlock_page(page);
1192                         put_page(page);
1193                         goto again;
1194                 }
1195
1196                 pages[i] = page;
1197                 i_done++;
1198         }
1199         if (!i_done || ret)
1200                 goto out;
1201
1202         if (!(inode->i_sb->s_flags & MS_ACTIVE))
1203                 goto out;
1204
1205         /*
1206          * so now we have a nice long stream of locked
1207          * and up to date pages, lets wait on them
1208          */
1209         for (i = 0; i < i_done; i++)
1210                 wait_on_page_writeback(pages[i]);
1211
1212         page_start = page_offset(pages[0]);
1213         page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1214
1215         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1216                          page_start, page_end - 1, &cached_state);
1217         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1218                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1219                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1220                           &cached_state, GFP_NOFS);
1221
1222         if (i_done != page_cnt) {
1223                 spin_lock(&BTRFS_I(inode)->lock);
1224                 BTRFS_I(inode)->outstanding_extents++;
1225                 spin_unlock(&BTRFS_I(inode)->lock);
1226                 btrfs_delalloc_release_space(inode,
1227                                 start_index << PAGE_SHIFT,
1228                                 (page_cnt - i_done) << PAGE_SHIFT);
1229         }
1230
1231
1232         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1233                           &cached_state, GFP_NOFS);
1234
1235         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1236                              page_start, page_end - 1, &cached_state,
1237                              GFP_NOFS);
1238
1239         for (i = 0; i < i_done; i++) {
1240                 clear_page_dirty_for_io(pages[i]);
1241                 ClearPageChecked(pages[i]);
1242                 set_page_extent_mapped(pages[i]);
1243                 set_page_dirty(pages[i]);
1244                 unlock_page(pages[i]);
1245                 put_page(pages[i]);
1246         }
1247         return i_done;
1248 out:
1249         for (i = 0; i < i_done; i++) {
1250                 unlock_page(pages[i]);
1251                 put_page(pages[i]);
1252         }
1253         btrfs_delalloc_release_space(inode,
1254                         start_index << PAGE_SHIFT,
1255                         page_cnt << PAGE_SHIFT);
1256         return ret;
1257
1258 }
1259
1260 int btrfs_defrag_file(struct inode *inode, struct file *file,
1261                       struct btrfs_ioctl_defrag_range_args *range,
1262                       u64 newer_than, unsigned long max_to_defrag)
1263 {
1264         struct btrfs_root *root = BTRFS_I(inode)->root;
1265         struct file_ra_state *ra = NULL;
1266         unsigned long last_index;
1267         u64 isize = i_size_read(inode);
1268         u64 last_len = 0;
1269         u64 skip = 0;
1270         u64 defrag_end = 0;
1271         u64 newer_off = range->start;
1272         unsigned long i;
1273         unsigned long ra_index = 0;
1274         int ret;
1275         int defrag_count = 0;
1276         int compress_type = BTRFS_COMPRESS_ZLIB;
1277         u32 extent_thresh = range->extent_thresh;
1278         unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1279         unsigned long cluster = max_cluster;
1280         u64 new_align = ~((u64)SZ_128K - 1);
1281         struct page **pages = NULL;
1282
1283         if (isize == 0)
1284                 return 0;
1285
1286         if (range->start >= isize)
1287                 return -EINVAL;
1288
1289         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1290                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1291                         return -EINVAL;
1292                 if (range->compress_type)
1293                         compress_type = range->compress_type;
1294         }
1295
1296         if (extent_thresh == 0)
1297                 extent_thresh = SZ_256K;
1298
1299         /*
1300          * if we were not given a file, allocate a readahead
1301          * context
1302          */
1303         if (!file) {
1304                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1305                 if (!ra)
1306                         return -ENOMEM;
1307                 file_ra_state_init(ra, inode->i_mapping);
1308         } else {
1309                 ra = &file->f_ra;
1310         }
1311
1312         pages = kmalloc_array(max_cluster, sizeof(struct page *),
1313                         GFP_NOFS);
1314         if (!pages) {
1315                 ret = -ENOMEM;
1316                 goto out_ra;
1317         }
1318
1319         /* find the last page to defrag */
1320         if (range->start + range->len > range->start) {
1321                 last_index = min_t(u64, isize - 1,
1322                          range->start + range->len - 1) >> PAGE_SHIFT;
1323         } else {
1324                 last_index = (isize - 1) >> PAGE_SHIFT;
1325         }
1326
1327         if (newer_than) {
1328                 ret = find_new_extents(root, inode, newer_than,
1329                                        &newer_off, SZ_64K);
1330                 if (!ret) {
1331                         range->start = newer_off;
1332                         /*
1333                          * we always align our defrag to help keep
1334                          * the extents in the file evenly spaced
1335                          */
1336                         i = (newer_off & new_align) >> PAGE_SHIFT;
1337                 } else
1338                         goto out_ra;
1339         } else {
1340                 i = range->start >> PAGE_SHIFT;
1341         }
1342         if (!max_to_defrag)
1343                 max_to_defrag = last_index - i + 1;
1344
1345         /*
1346          * make writeback starts from i, so the defrag range can be
1347          * written sequentially.
1348          */
1349         if (i < inode->i_mapping->writeback_index)
1350                 inode->i_mapping->writeback_index = i;
1351
1352         while (i <= last_index && defrag_count < max_to_defrag &&
1353                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1354                 /*
1355                  * make sure we stop running if someone unmounts
1356                  * the FS
1357                  */
1358                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1359                         break;
1360
1361                 if (btrfs_defrag_cancelled(root->fs_info)) {
1362                         btrfs_debug(root->fs_info, "defrag_file cancelled");
1363                         ret = -EAGAIN;
1364                         break;
1365                 }
1366
1367                 if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1368                                          extent_thresh, &last_len, &skip,
1369                                          &defrag_end, range->flags &
1370                                          BTRFS_DEFRAG_RANGE_COMPRESS)) {
1371                         unsigned long next;
1372                         /*
1373                          * the should_defrag function tells us how much to skip
1374                          * bump our counter by the suggested amount
1375                          */
1376                         next = DIV_ROUND_UP(skip, PAGE_SIZE);
1377                         i = max(i + 1, next);
1378                         continue;
1379                 }
1380
1381                 if (!newer_than) {
1382                         cluster = (PAGE_ALIGN(defrag_end) >>
1383                                    PAGE_SHIFT) - i;
1384                         cluster = min(cluster, max_cluster);
1385                 } else {
1386                         cluster = max_cluster;
1387                 }
1388
1389                 if (i + cluster > ra_index) {
1390                         ra_index = max(i, ra_index);
1391                         btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1392                                        cluster);
1393                         ra_index += cluster;
1394                 }
1395
1396                 inode_lock(inode);
1397                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1398                         BTRFS_I(inode)->force_compress = compress_type;
1399                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1400                 if (ret < 0) {
1401                         inode_unlock(inode);
1402                         goto out_ra;
1403                 }
1404
1405                 defrag_count += ret;
1406                 balance_dirty_pages_ratelimited(inode->i_mapping);
1407                 inode_unlock(inode);
1408
1409                 if (newer_than) {
1410                         if (newer_off == (u64)-1)
1411                                 break;
1412
1413                         if (ret > 0)
1414                                 i += ret;
1415
1416                         newer_off = max(newer_off + 1,
1417                                         (u64)i << PAGE_SHIFT);
1418
1419                         ret = find_new_extents(root, inode, newer_than,
1420                                                &newer_off, SZ_64K);
1421                         if (!ret) {
1422                                 range->start = newer_off;
1423                                 i = (newer_off & new_align) >> PAGE_SHIFT;
1424                         } else {
1425                                 break;
1426                         }
1427                 } else {
1428                         if (ret > 0) {
1429                                 i += ret;
1430                                 last_len += ret << PAGE_SHIFT;
1431                         } else {
1432                                 i++;
1433                                 last_len = 0;
1434                         }
1435                 }
1436         }
1437
1438         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1439                 filemap_flush(inode->i_mapping);
1440                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1441                              &BTRFS_I(inode)->runtime_flags))
1442                         filemap_flush(inode->i_mapping);
1443         }
1444
1445         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1446                 /* the filemap_flush will queue IO into the worker threads, but
1447                  * we have to make sure the IO is actually started and that
1448                  * ordered extents get created before we return
1449                  */
1450                 atomic_inc(&root->fs_info->async_submit_draining);
1451                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1452                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1453                         wait_event(root->fs_info->async_submit_wait,
1454                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1455                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1456                 }
1457                 atomic_dec(&root->fs_info->async_submit_draining);
1458         }
1459
1460         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1461                 btrfs_set_fs_incompat(root->fs_info, COMPRESS_LZO);
1462         }
1463
1464         ret = defrag_count;
1465
1466 out_ra:
1467         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1468                 inode_lock(inode);
1469                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1470                 inode_unlock(inode);
1471         }
1472         if (!file)
1473                 kfree(ra);
1474         kfree(pages);
1475         return ret;
1476 }
1477
1478 static noinline int btrfs_ioctl_resize(struct file *file,
1479                                         void __user *arg)
1480 {
1481         u64 new_size;
1482         u64 old_size;
1483         u64 devid = 1;
1484         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
1485         struct btrfs_ioctl_vol_args *vol_args;
1486         struct btrfs_trans_handle *trans;
1487         struct btrfs_device *device = NULL;
1488         char *sizestr;
1489         char *retptr;
1490         char *devstr = NULL;
1491         int ret = 0;
1492         int mod = 0;
1493
1494         if (!capable(CAP_SYS_ADMIN))
1495                 return -EPERM;
1496
1497         ret = mnt_want_write_file(file);
1498         if (ret)
1499                 return ret;
1500
1501         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
1502                         1)) {
1503                 mnt_drop_write_file(file);
1504                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1505         }
1506
1507         mutex_lock(&root->fs_info->volume_mutex);
1508         vol_args = memdup_user(arg, sizeof(*vol_args));
1509         if (IS_ERR(vol_args)) {
1510                 ret = PTR_ERR(vol_args);
1511                 goto out;
1512         }
1513
1514         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1515
1516         sizestr = vol_args->name;
1517         devstr = strchr(sizestr, ':');
1518         if (devstr) {
1519                 sizestr = devstr + 1;
1520                 *devstr = '\0';
1521                 devstr = vol_args->name;
1522                 ret = kstrtoull(devstr, 10, &devid);
1523                 if (ret)
1524                         goto out_free;
1525                 if (!devid) {
1526                         ret = -EINVAL;
1527                         goto out_free;
1528                 }
1529                 btrfs_info(root->fs_info, "resizing devid %llu", devid);
1530         }
1531
1532         device = btrfs_find_device(root->fs_info, devid, NULL, NULL);
1533         if (!device) {
1534                 btrfs_info(root->fs_info, "resizer unable to find device %llu",
1535                        devid);
1536                 ret = -ENODEV;
1537                 goto out_free;
1538         }
1539
1540         if (!device->writeable) {
1541                 btrfs_info(root->fs_info,
1542                            "resizer unable to apply on readonly device %llu",
1543                        devid);
1544                 ret = -EPERM;
1545                 goto out_free;
1546         }
1547
1548         if (!strcmp(sizestr, "max"))
1549                 new_size = device->bdev->bd_inode->i_size;
1550         else {
1551                 if (sizestr[0] == '-') {
1552                         mod = -1;
1553                         sizestr++;
1554                 } else if (sizestr[0] == '+') {
1555                         mod = 1;
1556                         sizestr++;
1557                 }
1558                 new_size = memparse(sizestr, &retptr);
1559                 if (*retptr != '\0' || new_size == 0) {
1560                         ret = -EINVAL;
1561                         goto out_free;
1562                 }
1563         }
1564
1565         if (device->is_tgtdev_for_dev_replace) {
1566                 ret = -EPERM;
1567                 goto out_free;
1568         }
1569
1570         old_size = btrfs_device_get_total_bytes(device);
1571
1572         if (mod < 0) {
1573                 if (new_size > old_size) {
1574                         ret = -EINVAL;
1575                         goto out_free;
1576                 }
1577                 new_size = old_size - new_size;
1578         } else if (mod > 0) {
1579                 if (new_size > ULLONG_MAX - old_size) {
1580                         ret = -ERANGE;
1581                         goto out_free;
1582                 }
1583                 new_size = old_size + new_size;
1584         }
1585
1586         if (new_size < SZ_256M) {
1587                 ret = -EINVAL;
1588                 goto out_free;
1589         }
1590         if (new_size > device->bdev->bd_inode->i_size) {
1591                 ret = -EFBIG;
1592                 goto out_free;
1593         }
1594
1595         new_size = div_u64(new_size, root->sectorsize);
1596         new_size *= root->sectorsize;
1597
1598         btrfs_info_in_rcu(root->fs_info, "new size for %s is %llu",
1599                       rcu_str_deref(device->name), new_size);
1600
1601         if (new_size > old_size) {
1602                 trans = btrfs_start_transaction(root, 0);
1603                 if (IS_ERR(trans)) {
1604                         ret = PTR_ERR(trans);
1605                         goto out_free;
1606                 }
1607                 ret = btrfs_grow_device(trans, device, new_size);
1608                 btrfs_commit_transaction(trans, root);
1609         } else if (new_size < old_size) {
1610                 ret = btrfs_shrink_device(device, new_size);
1611         } /* equal, nothing need to do */
1612
1613 out_free:
1614         kfree(vol_args);
1615 out:
1616         mutex_unlock(&root->fs_info->volume_mutex);
1617         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
1618         mnt_drop_write_file(file);
1619         return ret;
1620 }
1621
1622 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1623                                 char *name, unsigned long fd, int subvol,
1624                                 u64 *transid, bool readonly,
1625                                 struct btrfs_qgroup_inherit *inherit)
1626 {
1627         int namelen;
1628         int ret = 0;
1629
1630         ret = mnt_want_write_file(file);
1631         if (ret)
1632                 goto out;
1633
1634         namelen = strlen(name);
1635         if (strchr(name, '/')) {
1636                 ret = -EINVAL;
1637                 goto out_drop_write;
1638         }
1639
1640         if (name[0] == '.' &&
1641            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1642                 ret = -EEXIST;
1643                 goto out_drop_write;
1644         }
1645
1646         if (subvol) {
1647                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1648                                      NULL, transid, readonly, inherit);
1649         } else {
1650                 struct fd src = fdget(fd);
1651                 struct inode *src_inode;
1652                 if (!src.file) {
1653                         ret = -EINVAL;
1654                         goto out_drop_write;
1655                 }
1656
1657                 src_inode = file_inode(src.file);
1658                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1659                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1660                                    "Snapshot src from another FS");
1661                         ret = -EXDEV;
1662                 } else if (!inode_owner_or_capable(src_inode)) {
1663                         /*
1664                          * Subvolume creation is not restricted, but snapshots
1665                          * are limited to own subvolumes only
1666                          */
1667                         ret = -EPERM;
1668                 } else {
1669                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1670                                              BTRFS_I(src_inode)->root,
1671                                              transid, readonly, inherit);
1672                 }
1673                 fdput(src);
1674         }
1675 out_drop_write:
1676         mnt_drop_write_file(file);
1677 out:
1678         return ret;
1679 }
1680
1681 static noinline int btrfs_ioctl_snap_create(struct file *file,
1682                                             void __user *arg, int subvol)
1683 {
1684         struct btrfs_ioctl_vol_args *vol_args;
1685         int ret;
1686
1687         vol_args = memdup_user(arg, sizeof(*vol_args));
1688         if (IS_ERR(vol_args))
1689                 return PTR_ERR(vol_args);
1690         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1691
1692         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1693                                               vol_args->fd, subvol,
1694                                               NULL, false, NULL);
1695
1696         kfree(vol_args);
1697         return ret;
1698 }
1699
1700 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1701                                                void __user *arg, int subvol)
1702 {
1703         struct btrfs_ioctl_vol_args_v2 *vol_args;
1704         int ret;
1705         u64 transid = 0;
1706         u64 *ptr = NULL;
1707         bool readonly = false;
1708         struct btrfs_qgroup_inherit *inherit = NULL;
1709
1710         vol_args = memdup_user(arg, sizeof(*vol_args));
1711         if (IS_ERR(vol_args))
1712                 return PTR_ERR(vol_args);
1713         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1714
1715         if (vol_args->flags &
1716             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1717               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1718                 ret = -EOPNOTSUPP;
1719                 goto free_args;
1720         }
1721
1722         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1723                 ptr = &transid;
1724         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1725                 readonly = true;
1726         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1727                 if (vol_args->size > PAGE_SIZE) {
1728                         ret = -EINVAL;
1729                         goto free_args;
1730                 }
1731                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1732                 if (IS_ERR(inherit)) {
1733                         ret = PTR_ERR(inherit);
1734                         goto free_args;
1735                 }
1736         }
1737
1738         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1739                                               vol_args->fd, subvol, ptr,
1740                                               readonly, inherit);
1741         if (ret)
1742                 goto free_inherit;
1743
1744         if (ptr && copy_to_user(arg +
1745                                 offsetof(struct btrfs_ioctl_vol_args_v2,
1746                                         transid),
1747                                 ptr, sizeof(*ptr)))
1748                 ret = -EFAULT;
1749
1750 free_inherit:
1751         kfree(inherit);
1752 free_args:
1753         kfree(vol_args);
1754         return ret;
1755 }
1756
1757 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1758                                                 void __user *arg)
1759 {
1760         struct inode *inode = file_inode(file);
1761         struct btrfs_root *root = BTRFS_I(inode)->root;
1762         int ret = 0;
1763         u64 flags = 0;
1764
1765         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1766                 return -EINVAL;
1767
1768         down_read(&root->fs_info->subvol_sem);
1769         if (btrfs_root_readonly(root))
1770                 flags |= BTRFS_SUBVOL_RDONLY;
1771         up_read(&root->fs_info->subvol_sem);
1772
1773         if (copy_to_user(arg, &flags, sizeof(flags)))
1774                 ret = -EFAULT;
1775
1776         return ret;
1777 }
1778
1779 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1780                                               void __user *arg)
1781 {
1782         struct inode *inode = file_inode(file);
1783         struct btrfs_root *root = BTRFS_I(inode)->root;
1784         struct btrfs_trans_handle *trans;
1785         u64 root_flags;
1786         u64 flags;
1787         int ret = 0;
1788
1789         if (!inode_owner_or_capable(inode))
1790                 return -EPERM;
1791
1792         ret = mnt_want_write_file(file);
1793         if (ret)
1794                 goto out;
1795
1796         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1797                 ret = -EINVAL;
1798                 goto out_drop_write;
1799         }
1800
1801         if (copy_from_user(&flags, arg, sizeof(flags))) {
1802                 ret = -EFAULT;
1803                 goto out_drop_write;
1804         }
1805
1806         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1807                 ret = -EINVAL;
1808                 goto out_drop_write;
1809         }
1810
1811         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1812                 ret = -EOPNOTSUPP;
1813                 goto out_drop_write;
1814         }
1815
1816         down_write(&root->fs_info->subvol_sem);
1817
1818         /* nothing to do */
1819         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1820                 goto out_drop_sem;
1821
1822         root_flags = btrfs_root_flags(&root->root_item);
1823         if (flags & BTRFS_SUBVOL_RDONLY) {
1824                 btrfs_set_root_flags(&root->root_item,
1825                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1826         } else {
1827                 /*
1828                  * Block RO -> RW transition if this subvolume is involved in
1829                  * send
1830                  */
1831                 spin_lock(&root->root_item_lock);
1832                 if (root->send_in_progress == 0) {
1833                         btrfs_set_root_flags(&root->root_item,
1834                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1835                         spin_unlock(&root->root_item_lock);
1836                 } else {
1837                         spin_unlock(&root->root_item_lock);
1838                         btrfs_warn(root->fs_info,
1839                         "Attempt to set subvolume %llu read-write during send",
1840                                         root->root_key.objectid);
1841                         ret = -EPERM;
1842                         goto out_drop_sem;
1843                 }
1844         }
1845
1846         trans = btrfs_start_transaction(root, 1);
1847         if (IS_ERR(trans)) {
1848                 ret = PTR_ERR(trans);
1849                 goto out_reset;
1850         }
1851
1852         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1853                                 &root->root_key, &root->root_item);
1854
1855         btrfs_commit_transaction(trans, root);
1856 out_reset:
1857         if (ret)
1858                 btrfs_set_root_flags(&root->root_item, root_flags);
1859 out_drop_sem:
1860         up_write(&root->fs_info->subvol_sem);
1861 out_drop_write:
1862         mnt_drop_write_file(file);
1863 out:
1864         return ret;
1865 }
1866
1867 /*
1868  * helper to check if the subvolume references other subvolumes
1869  */
1870 static noinline int may_destroy_subvol(struct btrfs_root *root)
1871 {
1872         struct btrfs_path *path;
1873         struct btrfs_dir_item *di;
1874         struct btrfs_key key;
1875         u64 dir_id;
1876         int ret;
1877
1878         path = btrfs_alloc_path();
1879         if (!path)
1880                 return -ENOMEM;
1881
1882         /* Make sure this root isn't set as the default subvol */
1883         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
1884         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root, path,
1885                                    dir_id, "default", 7, 0);
1886         if (di && !IS_ERR(di)) {
1887                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1888                 if (key.objectid == root->root_key.objectid) {
1889                         ret = -EPERM;
1890                         btrfs_err(root->fs_info, "deleting default subvolume "
1891                                   "%llu is not allowed", key.objectid);
1892                         goto out;
1893                 }
1894                 btrfs_release_path(path);
1895         }
1896
1897         key.objectid = root->root_key.objectid;
1898         key.type = BTRFS_ROOT_REF_KEY;
1899         key.offset = (u64)-1;
1900
1901         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1902                                 &key, path, 0, 0);
1903         if (ret < 0)
1904                 goto out;
1905         BUG_ON(ret == 0);
1906
1907         ret = 0;
1908         if (path->slots[0] > 0) {
1909                 path->slots[0]--;
1910                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1911                 if (key.objectid == root->root_key.objectid &&
1912                     key.type == BTRFS_ROOT_REF_KEY)
1913                         ret = -ENOTEMPTY;
1914         }
1915 out:
1916         btrfs_free_path(path);
1917         return ret;
1918 }
1919
1920 static noinline int key_in_sk(struct btrfs_key *key,
1921                               struct btrfs_ioctl_search_key *sk)
1922 {
1923         struct btrfs_key test;
1924         int ret;
1925
1926         test.objectid = sk->min_objectid;
1927         test.type = sk->min_type;
1928         test.offset = sk->min_offset;
1929
1930         ret = btrfs_comp_cpu_keys(key, &test);
1931         if (ret < 0)
1932                 return 0;
1933
1934         test.objectid = sk->max_objectid;
1935         test.type = sk->max_type;
1936         test.offset = sk->max_offset;
1937
1938         ret = btrfs_comp_cpu_keys(key, &test);
1939         if (ret > 0)
1940                 return 0;
1941         return 1;
1942 }
1943
1944 static noinline int copy_to_sk(struct btrfs_root *root,
1945                                struct btrfs_path *path,
1946                                struct btrfs_key *key,
1947                                struct btrfs_ioctl_search_key *sk,
1948                                size_t *buf_size,
1949                                char __user *ubuf,
1950                                unsigned long *sk_offset,
1951                                int *num_found)
1952 {
1953         u64 found_transid;
1954         struct extent_buffer *leaf;
1955         struct btrfs_ioctl_search_header sh;
1956         struct btrfs_key test;
1957         unsigned long item_off;
1958         unsigned long item_len;
1959         int nritems;
1960         int i;
1961         int slot;
1962         int ret = 0;
1963
1964         leaf = path->nodes[0];
1965         slot = path->slots[0];
1966         nritems = btrfs_header_nritems(leaf);
1967
1968         if (btrfs_header_generation(leaf) > sk->max_transid) {
1969                 i = nritems;
1970                 goto advance_key;
1971         }
1972         found_transid = btrfs_header_generation(leaf);
1973
1974         for (i = slot; i < nritems; i++) {
1975                 item_off = btrfs_item_ptr_offset(leaf, i);
1976                 item_len = btrfs_item_size_nr(leaf, i);
1977
1978                 btrfs_item_key_to_cpu(leaf, key, i);
1979                 if (!key_in_sk(key, sk))
1980                         continue;
1981
1982                 if (sizeof(sh) + item_len > *buf_size) {
1983                         if (*num_found) {
1984                                 ret = 1;
1985                                 goto out;
1986                         }
1987
1988                         /*
1989                          * return one empty item back for v1, which does not
1990                          * handle -EOVERFLOW
1991                          */
1992
1993                         *buf_size = sizeof(sh) + item_len;
1994                         item_len = 0;
1995                         ret = -EOVERFLOW;
1996                 }
1997
1998                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
1999                         ret = 1;
2000                         goto out;
2001                 }
2002
2003                 sh.objectid = key->objectid;
2004                 sh.offset = key->offset;
2005                 sh.type = key->type;
2006                 sh.len = item_len;
2007                 sh.transid = found_transid;
2008
2009                 /* copy search result header */
2010                 if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
2011                         ret = -EFAULT;
2012                         goto out;
2013                 }
2014
2015                 *sk_offset += sizeof(sh);
2016
2017                 if (item_len) {
2018                         char __user *up = ubuf + *sk_offset;
2019                         /* copy the item */
2020                         if (read_extent_buffer_to_user(leaf, up,
2021                                                        item_off, item_len)) {
2022                                 ret = -EFAULT;
2023                                 goto out;
2024                         }
2025
2026                         *sk_offset += item_len;
2027                 }
2028                 (*num_found)++;
2029
2030                 if (ret) /* -EOVERFLOW from above */
2031                         goto out;
2032
2033                 if (*num_found >= sk->nr_items) {
2034                         ret = 1;
2035                         goto out;
2036                 }
2037         }
2038 advance_key:
2039         ret = 0;
2040         test.objectid = sk->max_objectid;
2041         test.type = sk->max_type;
2042         test.offset = sk->max_offset;
2043         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2044                 ret = 1;
2045         else if (key->offset < (u64)-1)
2046                 key->offset++;
2047         else if (key->type < (u8)-1) {
2048                 key->offset = 0;
2049                 key->type++;
2050         } else if (key->objectid < (u64)-1) {
2051                 key->offset = 0;
2052                 key->type = 0;
2053                 key->objectid++;
2054         } else
2055                 ret = 1;
2056 out:
2057         /*
2058          *  0: all items from this leaf copied, continue with next
2059          *  1: * more items can be copied, but unused buffer is too small
2060          *     * all items were found
2061          *     Either way, it will stops the loop which iterates to the next
2062          *     leaf
2063          *  -EOVERFLOW: item was to large for buffer
2064          *  -EFAULT: could not copy extent buffer back to userspace
2065          */
2066         return ret;
2067 }
2068
2069 static noinline int search_ioctl(struct inode *inode,
2070                                  struct btrfs_ioctl_search_key *sk,
2071                                  size_t *buf_size,
2072                                  char __user *ubuf)
2073 {
2074         struct btrfs_root *root;
2075         struct btrfs_key key;
2076         struct btrfs_path *path;
2077         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
2078         int ret;
2079         int num_found = 0;
2080         unsigned long sk_offset = 0;
2081
2082         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2083                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2084                 return -EOVERFLOW;
2085         }
2086
2087         path = btrfs_alloc_path();
2088         if (!path)
2089                 return -ENOMEM;
2090
2091         if (sk->tree_id == 0) {
2092                 /* search the root of the inode that was passed */
2093                 root = BTRFS_I(inode)->root;
2094         } else {
2095                 key.objectid = sk->tree_id;
2096                 key.type = BTRFS_ROOT_ITEM_KEY;
2097                 key.offset = (u64)-1;
2098                 root = btrfs_read_fs_root_no_name(info, &key);
2099                 if (IS_ERR(root)) {
2100                         btrfs_free_path(path);
2101                         return -ENOENT;
2102                 }
2103         }
2104
2105         key.objectid = sk->min_objectid;
2106         key.type = sk->min_type;
2107         key.offset = sk->min_offset;
2108
2109         while (1) {
2110                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2111                 if (ret != 0) {
2112                         if (ret > 0)
2113                                 ret = 0;
2114                         goto err;
2115                 }
2116                 ret = copy_to_sk(root, path, &key, sk, buf_size, ubuf,
2117                                  &sk_offset, &num_found);
2118                 btrfs_release_path(path);
2119                 if (ret)
2120                         break;
2121
2122         }
2123         if (ret > 0)
2124                 ret = 0;
2125 err:
2126         sk->nr_items = num_found;
2127         btrfs_free_path(path);
2128         return ret;
2129 }
2130
2131 static noinline int btrfs_ioctl_tree_search(struct file *file,
2132                                            void __user *argp)
2133 {
2134         struct btrfs_ioctl_search_args __user *uargs;
2135         struct btrfs_ioctl_search_key sk;
2136         struct inode *inode;
2137         int ret;
2138         size_t buf_size;
2139
2140         if (!capable(CAP_SYS_ADMIN))
2141                 return -EPERM;
2142
2143         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2144
2145         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2146                 return -EFAULT;
2147
2148         buf_size = sizeof(uargs->buf);
2149
2150         inode = file_inode(file);
2151         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2152
2153         /*
2154          * In the origin implementation an overflow is handled by returning a
2155          * search header with a len of zero, so reset ret.
2156          */
2157         if (ret == -EOVERFLOW)
2158                 ret = 0;
2159
2160         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2161                 ret = -EFAULT;
2162         return ret;
2163 }
2164
2165 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2166                                                void __user *argp)
2167 {
2168         struct btrfs_ioctl_search_args_v2 __user *uarg;
2169         struct btrfs_ioctl_search_args_v2 args;
2170         struct inode *inode;
2171         int ret;
2172         size_t buf_size;
2173         const size_t buf_limit = SZ_16M;
2174
2175         if (!capable(CAP_SYS_ADMIN))
2176                 return -EPERM;
2177
2178         /* copy search header and buffer size */
2179         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2180         if (copy_from_user(&args, uarg, sizeof(args)))
2181                 return -EFAULT;
2182
2183         buf_size = args.buf_size;
2184
2185         if (buf_size < sizeof(struct btrfs_ioctl_search_header))
2186                 return -EOVERFLOW;
2187
2188         /* limit result size to 16MB */
2189         if (buf_size > buf_limit)
2190                 buf_size = buf_limit;
2191
2192         inode = file_inode(file);
2193         ret = search_ioctl(inode, &args.key, &buf_size,
2194                            (char *)(&uarg->buf[0]));
2195         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2196                 ret = -EFAULT;
2197         else if (ret == -EOVERFLOW &&
2198                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2199                 ret = -EFAULT;
2200
2201         return ret;
2202 }
2203
2204 /*
2205  * Search INODE_REFs to identify path name of 'dirid' directory
2206  * in a 'tree_id' tree. and sets path name to 'name'.
2207  */
2208 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2209                                 u64 tree_id, u64 dirid, char *name)
2210 {
2211         struct btrfs_root *root;
2212         struct btrfs_key key;
2213         char *ptr;
2214         int ret = -1;
2215         int slot;
2216         int len;
2217         int total_len = 0;
2218         struct btrfs_inode_ref *iref;
2219         struct extent_buffer *l;
2220         struct btrfs_path *path;
2221
2222         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2223                 name[0]='\0';
2224                 return 0;
2225         }
2226
2227         path = btrfs_alloc_path();
2228         if (!path)
2229                 return -ENOMEM;
2230
2231         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
2232
2233         key.objectid = tree_id;
2234         key.type = BTRFS_ROOT_ITEM_KEY;
2235         key.offset = (u64)-1;
2236         root = btrfs_read_fs_root_no_name(info, &key);
2237         if (IS_ERR(root)) {
2238                 btrfs_err(info, "could not find root %llu", tree_id);
2239                 ret = -ENOENT;
2240                 goto out;
2241         }
2242
2243         key.objectid = dirid;
2244         key.type = BTRFS_INODE_REF_KEY;
2245         key.offset = (u64)-1;
2246
2247         while (1) {
2248                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2249                 if (ret < 0)
2250                         goto out;
2251                 else if (ret > 0) {
2252                         ret = btrfs_previous_item(root, path, dirid,
2253                                                   BTRFS_INODE_REF_KEY);
2254                         if (ret < 0)
2255                                 goto out;
2256                         else if (ret > 0) {
2257                                 ret = -ENOENT;
2258                                 goto out;
2259                         }
2260                 }
2261
2262                 l = path->nodes[0];
2263                 slot = path->slots[0];
2264                 btrfs_item_key_to_cpu(l, &key, slot);
2265
2266                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2267                 len = btrfs_inode_ref_name_len(l, iref);
2268                 ptr -= len + 1;
2269                 total_len += len + 1;
2270                 if (ptr < name) {
2271                         ret = -ENAMETOOLONG;
2272                         goto out;
2273                 }
2274
2275                 *(ptr + len) = '/';
2276                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2277
2278                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2279                         break;
2280
2281                 btrfs_release_path(path);
2282                 key.objectid = key.offset;
2283                 key.offset = (u64)-1;
2284                 dirid = key.objectid;
2285         }
2286         memmove(name, ptr, total_len);
2287         name[total_len] = '\0';
2288         ret = 0;
2289 out:
2290         btrfs_free_path(path);
2291         return ret;
2292 }
2293
2294 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2295                                            void __user *argp)
2296 {
2297          struct btrfs_ioctl_ino_lookup_args *args;
2298          struct inode *inode;
2299         int ret = 0;
2300
2301         args = memdup_user(argp, sizeof(*args));
2302         if (IS_ERR(args))
2303                 return PTR_ERR(args);
2304
2305         inode = file_inode(file);
2306
2307         /*
2308          * Unprivileged query to obtain the containing subvolume root id. The
2309          * path is reset so it's consistent with btrfs_search_path_in_tree.
2310          */
2311         if (args->treeid == 0)
2312                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2313
2314         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2315                 args->name[0] = 0;
2316                 goto out;
2317         }
2318
2319         if (!capable(CAP_SYS_ADMIN)) {
2320                 ret = -EPERM;
2321                 goto out;
2322         }
2323
2324         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2325                                         args->treeid, args->objectid,
2326                                         args->name);
2327
2328 out:
2329         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2330                 ret = -EFAULT;
2331
2332         kfree(args);
2333         return ret;
2334 }
2335
2336 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2337                                              void __user *arg)
2338 {
2339         struct dentry *parent = file->f_path.dentry;
2340         struct dentry *dentry;
2341         struct inode *dir = d_inode(parent);
2342         struct inode *inode;
2343         struct btrfs_root *root = BTRFS_I(dir)->root;
2344         struct btrfs_root *dest = NULL;
2345         struct btrfs_ioctl_vol_args *vol_args;
2346         struct btrfs_trans_handle *trans;
2347         struct btrfs_block_rsv block_rsv;
2348         u64 root_flags;
2349         u64 qgroup_reserved;
2350         int namelen;
2351         int ret;
2352         int err = 0;
2353
2354         vol_args = memdup_user(arg, sizeof(*vol_args));
2355         if (IS_ERR(vol_args))
2356                 return PTR_ERR(vol_args);
2357
2358         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2359         namelen = strlen(vol_args->name);
2360         if (strchr(vol_args->name, '/') ||
2361             strncmp(vol_args->name, "..", namelen) == 0) {
2362                 err = -EINVAL;
2363                 goto out;
2364         }
2365
2366         err = mnt_want_write_file(file);
2367         if (err)
2368                 goto out;
2369
2370
2371         inode_lock_nested(dir, I_MUTEX_PARENT);
2372         // XXX: should've been
2373         // err = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
2374         // if (err == -EINTR)
2375         //      goto out_drop_write;
2376         dentry = lookup_one_len(vol_args->name, parent, namelen);
2377         if (IS_ERR(dentry)) {
2378                 err = PTR_ERR(dentry);
2379                 goto out_unlock_dir;
2380         }
2381
2382         if (d_really_is_negative(dentry)) {
2383                 err = -ENOENT;
2384                 goto out_dput;
2385         }
2386
2387         inode = d_inode(dentry);
2388         dest = BTRFS_I(inode)->root;
2389         if (!capable(CAP_SYS_ADMIN)) {
2390                 /*
2391                  * Regular user.  Only allow this with a special mount
2392                  * option, when the user has write+exec access to the
2393                  * subvol root, and when rmdir(2) would have been
2394                  * allowed.
2395                  *
2396                  * Note that this is _not_ check that the subvol is
2397                  * empty or doesn't contain data that we wouldn't
2398                  * otherwise be able to delete.
2399                  *
2400                  * Users who want to delete empty subvols should try
2401                  * rmdir(2).
2402                  */
2403                 err = -EPERM;
2404                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
2405                         goto out_dput;
2406
2407                 /*
2408                  * Do not allow deletion if the parent dir is the same
2409                  * as the dir to be deleted.  That means the ioctl
2410                  * must be called on the dentry referencing the root
2411                  * of the subvol, not a random directory contained
2412                  * within it.
2413                  */
2414                 err = -EINVAL;
2415                 if (root == dest)
2416                         goto out_dput;
2417
2418                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2419                 if (err)
2420                         goto out_dput;
2421         }
2422
2423         /* check if subvolume may be deleted by a user */
2424         err = btrfs_may_delete(dir, dentry, 1);
2425         if (err)
2426                 goto out_dput;
2427
2428         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
2429                 err = -EINVAL;
2430                 goto out_dput;
2431         }
2432
2433         inode_lock(inode);
2434
2435         /*
2436          * Don't allow to delete a subvolume with send in progress. This is
2437          * inside the i_mutex so the error handling that has to drop the bit
2438          * again is not run concurrently.
2439          */
2440         spin_lock(&dest->root_item_lock);
2441         root_flags = btrfs_root_flags(&dest->root_item);
2442         if (dest->send_in_progress == 0) {
2443                 btrfs_set_root_flags(&dest->root_item,
2444                                 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
2445                 spin_unlock(&dest->root_item_lock);
2446         } else {
2447                 spin_unlock(&dest->root_item_lock);
2448                 btrfs_warn(root->fs_info,
2449                         "Attempt to delete subvolume %llu during send",
2450                         dest->root_key.objectid);
2451                 err = -EPERM;
2452                 goto out_unlock_inode;
2453         }
2454
2455         down_write(&root->fs_info->subvol_sem);
2456
2457         err = may_destroy_subvol(dest);
2458         if (err)
2459                 goto out_up_write;
2460
2461         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2462         /*
2463          * One for dir inode, two for dir entries, two for root
2464          * ref/backref.
2465          */
2466         err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2467                                                5, &qgroup_reserved, true);
2468         if (err)
2469                 goto out_up_write;
2470
2471         trans = btrfs_start_transaction(root, 0);
2472         if (IS_ERR(trans)) {
2473                 err = PTR_ERR(trans);
2474                 goto out_release;
2475         }
2476         trans->block_rsv = &block_rsv;
2477         trans->bytes_reserved = block_rsv.size;
2478
2479         btrfs_record_snapshot_destroy(trans, dir);
2480
2481         ret = btrfs_unlink_subvol(trans, root, dir,
2482                                 dest->root_key.objectid,
2483                                 dentry->d_name.name,
2484                                 dentry->d_name.len);
2485         if (ret) {
2486                 err = ret;
2487                 btrfs_abort_transaction(trans, root, ret);
2488                 goto out_end_trans;
2489         }
2490
2491         btrfs_record_root_in_trans(trans, dest);
2492
2493         memset(&dest->root_item.drop_progress, 0,
2494                 sizeof(dest->root_item.drop_progress));
2495         dest->root_item.drop_level = 0;
2496         btrfs_set_root_refs(&dest->root_item, 0);
2497
2498         if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
2499                 ret = btrfs_insert_orphan_item(trans,
2500                                         root->fs_info->tree_root,
2501                                         dest->root_key.objectid);
2502                 if (ret) {
2503                         btrfs_abort_transaction(trans, root, ret);
2504                         err = ret;
2505                         goto out_end_trans;
2506                 }
2507         }
2508
2509         ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2510                                   dest->root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
2511                                   dest->root_key.objectid);
2512         if (ret && ret != -ENOENT) {
2513                 btrfs_abort_transaction(trans, root, ret);
2514                 err = ret;
2515                 goto out_end_trans;
2516         }
2517         if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2518                 ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2519                                           dest->root_item.received_uuid,
2520                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2521                                           dest->root_key.objectid);
2522                 if (ret && ret != -ENOENT) {
2523                         btrfs_abort_transaction(trans, root, ret);
2524                         err = ret;
2525                         goto out_end_trans;
2526                 }
2527         }
2528
2529 out_end_trans:
2530         trans->block_rsv = NULL;
2531         trans->bytes_reserved = 0;
2532         ret = btrfs_end_transaction(trans, root);
2533         if (ret && !err)
2534                 err = ret;
2535         inode->i_flags |= S_DEAD;
2536 out_release:
2537         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
2538 out_up_write:
2539         up_write(&root->fs_info->subvol_sem);
2540         if (err) {
2541                 spin_lock(&dest->root_item_lock);
2542                 root_flags = btrfs_root_flags(&dest->root_item);
2543                 btrfs_set_root_flags(&dest->root_item,
2544                                 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
2545                 spin_unlock(&dest->root_item_lock);
2546         }
2547 out_unlock_inode:
2548         inode_unlock(inode);
2549         if (!err) {
2550                 d_invalidate(dentry);
2551                 btrfs_invalidate_inodes(dest);
2552                 d_delete(dentry);
2553                 ASSERT(dest->send_in_progress == 0);
2554
2555                 /* the last ref */
2556                 if (dest->ino_cache_inode) {
2557                         iput(dest->ino_cache_inode);
2558                         dest->ino_cache_inode = NULL;
2559                 }
2560         }
2561 out_dput:
2562         dput(dentry);
2563 out_unlock_dir:
2564         inode_unlock(dir);
2565 //out_drop_write:
2566         mnt_drop_write_file(file);
2567 out:
2568         kfree(vol_args);
2569         return err;
2570 }
2571
2572 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2573 {
2574         struct inode *inode = file_inode(file);
2575         struct btrfs_root *root = BTRFS_I(inode)->root;
2576         struct btrfs_ioctl_defrag_range_args *range;
2577         int ret;
2578
2579         ret = mnt_want_write_file(file);
2580         if (ret)
2581                 return ret;
2582
2583         if (btrfs_root_readonly(root)) {
2584                 ret = -EROFS;
2585                 goto out;
2586         }
2587
2588         switch (inode->i_mode & S_IFMT) {
2589         case S_IFDIR:
2590                 if (!capable(CAP_SYS_ADMIN)) {
2591                         ret = -EPERM;
2592                         goto out;
2593                 }
2594                 ret = btrfs_defrag_root(root);
2595                 if (ret)
2596                         goto out;
2597                 ret = btrfs_defrag_root(root->fs_info->extent_root);
2598                 break;
2599         case S_IFREG:
2600                 if (!(file->f_mode & FMODE_WRITE)) {
2601                         ret = -EINVAL;
2602                         goto out;
2603                 }
2604
2605                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2606                 if (!range) {
2607                         ret = -ENOMEM;
2608                         goto out;
2609                 }
2610
2611                 if (argp) {
2612                         if (copy_from_user(range, argp,
2613                                            sizeof(*range))) {
2614                                 ret = -EFAULT;
2615                                 kfree(range);
2616                                 goto out;
2617                         }
2618                         /* compression requires us to start the IO */
2619                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2620                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2621                                 range->extent_thresh = (u32)-1;
2622                         }
2623                 } else {
2624                         /* the rest are all set to zero by kzalloc */
2625                         range->len = (u64)-1;
2626                 }
2627                 ret = btrfs_defrag_file(file_inode(file), file,
2628                                         range, 0, 0);
2629                 if (ret > 0)
2630                         ret = 0;
2631                 kfree(range);
2632                 break;
2633         default:
2634                 ret = -EINVAL;
2635         }
2636 out:
2637         mnt_drop_write_file(file);
2638         return ret;
2639 }
2640
2641 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2642 {
2643         struct btrfs_ioctl_vol_args *vol_args;
2644         int ret;
2645
2646         if (!capable(CAP_SYS_ADMIN))
2647                 return -EPERM;
2648
2649         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2650                         1)) {
2651                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2652         }
2653
2654         mutex_lock(&root->fs_info->volume_mutex);
2655         vol_args = memdup_user(arg, sizeof(*vol_args));
2656         if (IS_ERR(vol_args)) {
2657                 ret = PTR_ERR(vol_args);
2658                 goto out;
2659         }
2660
2661         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2662         ret = btrfs_init_new_device(root, vol_args->name);
2663
2664         if (!ret)
2665                 btrfs_info(root->fs_info, "disk added %s",vol_args->name);
2666
2667         kfree(vol_args);
2668 out:
2669         mutex_unlock(&root->fs_info->volume_mutex);
2670         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2671         return ret;
2672 }
2673
2674 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2675 {
2676         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
2677         struct btrfs_ioctl_vol_args *vol_args;
2678         int ret;
2679
2680         if (!capable(CAP_SYS_ADMIN))
2681                 return -EPERM;
2682
2683         ret = mnt_want_write_file(file);
2684         if (ret)
2685                 return ret;
2686
2687         vol_args = memdup_user(arg, sizeof(*vol_args));
2688         if (IS_ERR(vol_args)) {
2689                 ret = PTR_ERR(vol_args);
2690                 goto err_drop;
2691         }
2692
2693         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2694
2695         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2696                         1)) {
2697                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2698                 goto out;
2699         }
2700
2701         mutex_lock(&root->fs_info->volume_mutex);
2702         ret = btrfs_rm_device(root, vol_args->name);
2703         mutex_unlock(&root->fs_info->volume_mutex);
2704         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2705
2706         if (!ret)
2707                 btrfs_info(root->fs_info, "disk deleted %s",vol_args->name);
2708
2709 out:
2710         kfree(vol_args);
2711 err_drop:
2712         mnt_drop_write_file(file);
2713         return ret;
2714 }
2715
2716 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2717 {
2718         struct btrfs_ioctl_fs_info_args *fi_args;
2719         struct btrfs_device *device;
2720         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2721         int ret = 0;
2722
2723         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2724         if (!fi_args)
2725                 return -ENOMEM;
2726
2727         mutex_lock(&fs_devices->device_list_mutex);
2728         fi_args->num_devices = fs_devices->num_devices;
2729         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2730
2731         list_for_each_entry(device, &fs_devices->devices, dev_list) {
2732                 if (device->devid > fi_args->max_id)
2733                         fi_args->max_id = device->devid;
2734         }
2735         mutex_unlock(&fs_devices->device_list_mutex);
2736
2737         fi_args->nodesize = root->fs_info->super_copy->nodesize;
2738         fi_args->sectorsize = root->fs_info->super_copy->sectorsize;
2739         fi_args->clone_alignment = root->fs_info->super_copy->sectorsize;
2740
2741         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2742                 ret = -EFAULT;
2743
2744         kfree(fi_args);
2745         return ret;
2746 }
2747
2748 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2749 {
2750         struct btrfs_ioctl_dev_info_args *di_args;
2751         struct btrfs_device *dev;
2752         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2753         int ret = 0;
2754         char *s_uuid = NULL;
2755
2756         di_args = memdup_user(arg, sizeof(*di_args));
2757         if (IS_ERR(di_args))
2758                 return PTR_ERR(di_args);
2759
2760         if (!btrfs_is_empty_uuid(di_args->uuid))
2761                 s_uuid = di_args->uuid;
2762
2763         mutex_lock(&fs_devices->device_list_mutex);
2764         dev = btrfs_find_device(root->fs_info, di_args->devid, s_uuid, NULL);
2765
2766         if (!dev) {
2767                 ret = -ENODEV;
2768                 goto out;
2769         }
2770
2771         di_args->devid = dev->devid;
2772         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
2773         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
2774         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2775         if (dev->name) {
2776                 struct rcu_string *name;
2777
2778                 rcu_read_lock();
2779                 name = rcu_dereference(dev->name);
2780                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2781                 rcu_read_unlock();
2782                 di_args->path[sizeof(di_args->path) - 1] = 0;
2783         } else {
2784                 di_args->path[0] = '\0';
2785         }
2786
2787 out:
2788         mutex_unlock(&fs_devices->device_list_mutex);
2789         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2790                 ret = -EFAULT;
2791
2792         kfree(di_args);
2793         return ret;
2794 }
2795
2796 static struct page *extent_same_get_page(struct inode *inode, pgoff_t index)
2797 {
2798         struct page *page;
2799
2800         page = grab_cache_page(inode->i_mapping, index);
2801         if (!page)
2802                 return ERR_PTR(-ENOMEM);
2803
2804         if (!PageUptodate(page)) {
2805                 int ret;
2806
2807                 ret = btrfs_readpage(NULL, page);
2808                 if (ret)
2809                         return ERR_PTR(ret);
2810                 lock_page(page);
2811                 if (!PageUptodate(page)) {
2812                         unlock_page(page);
2813                         put_page(page);
2814                         return ERR_PTR(-EIO);
2815                 }
2816                 if (page->mapping != inode->i_mapping) {
2817                         unlock_page(page);
2818                         put_page(page);
2819                         return ERR_PTR(-EAGAIN);
2820                 }
2821         }
2822
2823         return page;
2824 }
2825
2826 static int gather_extent_pages(struct inode *inode, struct page **pages,
2827                                int num_pages, u64 off)
2828 {
2829         int i;
2830         pgoff_t index = off >> PAGE_SHIFT;
2831
2832         for (i = 0; i < num_pages; i++) {
2833 again:
2834                 pages[i] = extent_same_get_page(inode, index + i);
2835                 if (IS_ERR(pages[i])) {
2836                         int err = PTR_ERR(pages[i]);
2837
2838                         if (err == -EAGAIN)
2839                                 goto again;
2840                         pages[i] = NULL;
2841                         return err;
2842                 }
2843         }
2844         return 0;
2845 }
2846
2847 static int lock_extent_range(struct inode *inode, u64 off, u64 len,
2848                              bool retry_range_locking)
2849 {
2850         /*
2851          * Do any pending delalloc/csum calculations on inode, one way or
2852          * another, and lock file content.
2853          * The locking order is:
2854          *
2855          *   1) pages
2856          *   2) range in the inode's io tree
2857          */
2858         while (1) {
2859                 struct btrfs_ordered_extent *ordered;
2860                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2861                 ordered = btrfs_lookup_first_ordered_extent(inode,
2862                                                             off + len - 1);
2863                 if ((!ordered ||
2864                      ordered->file_offset + ordered->len <= off ||
2865                      ordered->file_offset >= off + len) &&
2866                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2867                                     off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
2868                         if (ordered)
2869                                 btrfs_put_ordered_extent(ordered);
2870                         break;
2871                 }
2872                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2873                 if (ordered)
2874                         btrfs_put_ordered_extent(ordered);
2875                 if (!retry_range_locking)
2876                         return -EAGAIN;
2877                 btrfs_wait_ordered_range(inode, off, len);
2878         }
2879         return 0;
2880 }
2881
2882 static void btrfs_double_inode_unlock(struct inode *inode1, struct inode *inode2)
2883 {
2884         inode_unlock(inode1);
2885         inode_unlock(inode2);
2886 }
2887
2888 static void btrfs_double_inode_lock(struct inode *inode1, struct inode *inode2)
2889 {
2890         if (inode1 < inode2)
2891                 swap(inode1, inode2);
2892
2893         inode_lock_nested(inode1, I_MUTEX_PARENT);
2894         inode_lock_nested(inode2, I_MUTEX_CHILD);
2895 }
2896
2897 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
2898                                       struct inode *inode2, u64 loff2, u64 len)
2899 {
2900         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2901         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2902 }
2903
2904 static int btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
2905                                     struct inode *inode2, u64 loff2, u64 len,
2906                                     bool retry_range_locking)
2907 {
2908         int ret;
2909
2910         if (inode1 < inode2) {
2911                 swap(inode1, inode2);
2912                 swap(loff1, loff2);
2913         }
2914         ret = lock_extent_range(inode1, loff1, len, retry_range_locking);
2915         if (ret)
2916                 return ret;
2917         ret = lock_extent_range(inode2, loff2, len, retry_range_locking);
2918         if (ret)
2919                 unlock_extent(&BTRFS_I(inode1)->io_tree, loff1,
2920                               loff1 + len - 1);
2921         return ret;
2922 }
2923
2924 struct cmp_pages {
2925         int             num_pages;
2926         struct page     **src_pages;
2927         struct page     **dst_pages;
2928 };
2929
2930 static void btrfs_cmp_data_free(struct cmp_pages *cmp)
2931 {
2932         int i;
2933         struct page *pg;
2934
2935         for (i = 0; i < cmp->num_pages; i++) {
2936                 pg = cmp->src_pages[i];
2937                 if (pg) {
2938                         unlock_page(pg);
2939                         put_page(pg);
2940                 }
2941                 pg = cmp->dst_pages[i];
2942                 if (pg) {
2943                         unlock_page(pg);
2944                         put_page(pg);
2945                 }
2946         }
2947         kfree(cmp->src_pages);
2948         kfree(cmp->dst_pages);
2949 }
2950
2951 static int btrfs_cmp_data_prepare(struct inode *src, u64 loff,
2952                                   struct inode *dst, u64 dst_loff,
2953                                   u64 len, struct cmp_pages *cmp)
2954 {
2955         int ret;
2956         int num_pages = PAGE_ALIGN(len) >> PAGE_SHIFT;
2957         struct page **src_pgarr, **dst_pgarr;
2958
2959         /*
2960          * We must gather up all the pages before we initiate our
2961          * extent locking. We use an array for the page pointers. Size
2962          * of the array is bounded by len, which is in turn bounded by
2963          * BTRFS_MAX_DEDUPE_LEN.
2964          */
2965         src_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
2966         dst_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
2967         if (!src_pgarr || !dst_pgarr) {
2968                 kfree(src_pgarr);
2969                 kfree(dst_pgarr);
2970                 return -ENOMEM;
2971         }
2972         cmp->num_pages = num_pages;
2973         cmp->src_pages = src_pgarr;
2974         cmp->dst_pages = dst_pgarr;
2975
2976         ret = gather_extent_pages(src, cmp->src_pages, cmp->num_pages, loff);
2977         if (ret)
2978                 goto out;
2979
2980         ret = gather_extent_pages(dst, cmp->dst_pages, cmp->num_pages, dst_loff);
2981
2982 out:
2983         if (ret)
2984                 btrfs_cmp_data_free(cmp);
2985         return 0;
2986 }
2987
2988 static int btrfs_cmp_data(struct inode *src, u64 loff, struct inode *dst,
2989                           u64 dst_loff, u64 len, struct cmp_pages *cmp)
2990 {
2991         int ret = 0;
2992         int i;
2993         struct page *src_page, *dst_page;
2994         unsigned int cmp_len = PAGE_SIZE;
2995         void *addr, *dst_addr;
2996
2997         i = 0;
2998         while (len) {
2999                 if (len < PAGE_SIZE)
3000                         cmp_len = len;
3001
3002                 BUG_ON(i >= cmp->num_pages);
3003
3004                 src_page = cmp->src_pages[i];
3005                 dst_page = cmp->dst_pages[i];
3006                 ASSERT(PageLocked(src_page));
3007                 ASSERT(PageLocked(dst_page));
3008
3009                 addr = kmap_atomic(src_page);
3010                 dst_addr = kmap_atomic(dst_page);
3011
3012                 flush_dcache_page(src_page);
3013                 flush_dcache_page(dst_page);
3014
3015                 if (memcmp(addr, dst_addr, cmp_len))
3016                         ret = -EBADE;
3017
3018                 kunmap_atomic(addr);
3019                 kunmap_atomic(dst_addr);
3020
3021                 if (ret)
3022                         break;
3023
3024                 len -= cmp_len;
3025                 i++;
3026         }
3027
3028         return ret;
3029 }
3030
3031 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 *plen,
3032                                      u64 olen)
3033 {
3034         u64 len = *plen;
3035         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
3036
3037         if (off + olen > inode->i_size || off + olen < off)
3038                 return -EINVAL;
3039
3040         /* if we extend to eof, continue to block boundary */
3041         if (off + len == inode->i_size)
3042                 *plen = len = ALIGN(inode->i_size, bs) - off;
3043
3044         /* Check that we are block aligned - btrfs_clone() requires this */
3045         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
3046                 return -EINVAL;
3047
3048         return 0;
3049 }
3050
3051 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
3052                              struct inode *dst, u64 dst_loff)
3053 {
3054         int ret;
3055         u64 len = olen;
3056         struct cmp_pages cmp;
3057         int same_inode = 0;
3058         u64 same_lock_start = 0;
3059         u64 same_lock_len = 0;
3060
3061         if (src == dst)
3062                 same_inode = 1;
3063
3064         if (len == 0)
3065                 return 0;
3066
3067         if (same_inode) {
3068                 inode_lock(src);
3069
3070                 ret = extent_same_check_offsets(src, loff, &len, olen);
3071                 if (ret)
3072                         goto out_unlock;
3073                 ret = extent_same_check_offsets(src, dst_loff, &len, olen);
3074                 if (ret)
3075                         goto out_unlock;
3076
3077                 /*
3078                  * Single inode case wants the same checks, except we
3079                  * don't want our length pushed out past i_size as
3080                  * comparing that data range makes no sense.
3081                  *
3082                  * extent_same_check_offsets() will do this for an
3083                  * unaligned length at i_size, so catch it here and
3084                  * reject the request.
3085                  *
3086                  * This effectively means we require aligned extents
3087                  * for the single-inode case, whereas the other cases
3088                  * allow an unaligned length so long as it ends at
3089                  * i_size.
3090                  */
3091                 if (len != olen) {
3092                         ret = -EINVAL;
3093                         goto out_unlock;
3094                 }
3095
3096                 /* Check for overlapping ranges */
3097                 if (dst_loff + len > loff && dst_loff < loff + len) {
3098                         ret = -EINVAL;
3099                         goto out_unlock;
3100                 }
3101
3102                 same_lock_start = min_t(u64, loff, dst_loff);
3103                 same_lock_len = max_t(u64, loff, dst_loff) + len - same_lock_start;
3104         } else {
3105                 btrfs_double_inode_lock(src, dst);
3106
3107                 ret = extent_same_check_offsets(src, loff, &len, olen);
3108                 if (ret)
3109                         goto out_unlock;
3110
3111                 ret = extent_same_check_offsets(dst, dst_loff, &len, olen);
3112                 if (ret)
3113                         goto out_unlock;
3114         }
3115
3116         /* don't make the dst file partly checksummed */
3117         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3118             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
3119                 ret = -EINVAL;
3120                 goto out_unlock;
3121         }
3122
3123 again:
3124         ret = btrfs_cmp_data_prepare(src, loff, dst, dst_loff, olen, &cmp);
3125         if (ret)
3126                 goto out_unlock;
3127
3128         if (same_inode)
3129                 ret = lock_extent_range(src, same_lock_start, same_lock_len,
3130                                         false);
3131         else
3132                 ret = btrfs_double_extent_lock(src, loff, dst, dst_loff, len,
3133                                                false);
3134         /*
3135          * If one of the inodes has dirty pages in the respective range or
3136          * ordered extents, we need to flush dellaloc and wait for all ordered
3137          * extents in the range. We must unlock the pages and the ranges in the
3138          * io trees to avoid deadlocks when flushing delalloc (requires locking
3139          * pages) and when waiting for ordered extents to complete (they require
3140          * range locking).
3141          */
3142         if (ret == -EAGAIN) {
3143                 /*
3144                  * Ranges in the io trees already unlocked. Now unlock all
3145                  * pages before waiting for all IO to complete.
3146                  */
3147                 btrfs_cmp_data_free(&cmp);
3148                 if (same_inode) {
3149                         btrfs_wait_ordered_range(src, same_lock_start,
3150                                                  same_lock_len);
3151                 } else {
3152                         btrfs_wait_ordered_range(src, loff, len);
3153                         btrfs_wait_ordered_range(dst, dst_loff, len);
3154                 }
3155                 goto again;
3156         }
3157         ASSERT(ret == 0);
3158         if (WARN_ON(ret)) {
3159                 /* ranges in the io trees already unlocked */
3160                 btrfs_cmp_data_free(&cmp);
3161                 return ret;
3162         }
3163
3164         /* pass original length for comparison so we stay within i_size */
3165         ret = btrfs_cmp_data(src, loff, dst, dst_loff, olen, &cmp);
3166         if (ret == 0)
3167                 ret = btrfs_clone(src, dst, loff, olen, len, dst_loff, 1);
3168
3169         if (same_inode)
3170                 unlock_extent(&BTRFS_I(src)->io_tree, same_lock_start,
3171                               same_lock_start + same_lock_len - 1);
3172         else
3173                 btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
3174
3175         btrfs_cmp_data_free(&cmp);
3176 out_unlock:
3177         if (same_inode)
3178                 inode_unlock(src);
3179         else
3180                 btrfs_double_inode_unlock(src, dst);
3181
3182         return ret;
3183 }
3184
3185 #define BTRFS_MAX_DEDUPE_LEN    SZ_16M
3186
3187 ssize_t btrfs_dedupe_file_range(struct file *src_file, u64 loff, u64 olen,
3188                                 struct file *dst_file, u64 dst_loff)
3189 {
3190         struct inode *src = file_inode(src_file);
3191         struct inode *dst = file_inode(dst_file);
3192         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
3193         ssize_t res;
3194
3195         if (olen > BTRFS_MAX_DEDUPE_LEN)
3196                 olen = BTRFS_MAX_DEDUPE_LEN;
3197
3198         if (WARN_ON_ONCE(bs < PAGE_SIZE)) {
3199                 /*
3200                  * Btrfs does not support blocksize < page_size. As a
3201                  * result, btrfs_cmp_data() won't correctly handle
3202                  * this situation without an update.
3203                  */
3204                 return -EINVAL;
3205         }
3206
3207         res = btrfs_extent_same(src, loff, olen, dst, dst_loff);
3208         if (res)
3209                 return res;
3210         return olen;
3211 }
3212
3213 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3214                                      struct inode *inode,
3215                                      u64 endoff,
3216                                      const u64 destoff,
3217                                      const u64 olen,
3218                                      int no_time_update)
3219 {
3220         struct btrfs_root *root = BTRFS_I(inode)->root;
3221         int ret;
3222
3223         inode_inc_iversion(inode);
3224         if (!no_time_update)
3225                 inode->i_mtime = inode->i_ctime = current_fs_time(inode->i_sb);
3226         /*
3227          * We round up to the block size at eof when determining which
3228          * extents to clone above, but shouldn't round up the file size.
3229          */
3230         if (endoff > destoff + olen)
3231                 endoff = destoff + olen;
3232         if (endoff > inode->i_size)
3233                 btrfs_i_size_write(inode, endoff);
3234
3235         ret = btrfs_update_inode(trans, root, inode);
3236         if (ret) {
3237                 btrfs_abort_transaction(trans, root, ret);
3238                 btrfs_end_transaction(trans, root);
3239                 goto out;
3240         }
3241         ret = btrfs_end_transaction(trans, root);
3242 out:
3243         return ret;
3244 }
3245
3246 static void clone_update_extent_map(struct inode *inode,
3247                                     const struct btrfs_trans_handle *trans,
3248                                     const struct btrfs_path *path,
3249                                     const u64 hole_offset,
3250                                     const u64 hole_len)
3251 {
3252         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3253         struct extent_map *em;
3254         int ret;
3255
3256         em = alloc_extent_map();
3257         if (!em) {
3258                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3259                         &BTRFS_I(inode)->runtime_flags);
3260                 return;
3261         }
3262
3263         if (path) {
3264                 struct btrfs_file_extent_item *fi;
3265
3266                 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
3267                                     struct btrfs_file_extent_item);
3268                 btrfs_extent_item_to_extent_map(inode, path, fi, false, em);
3269                 em->generation = -1;
3270                 if (btrfs_file_extent_type(path->nodes[0], fi) ==
3271                     BTRFS_FILE_EXTENT_INLINE)
3272                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3273                                 &BTRFS_I(inode)->runtime_flags);
3274         } else {
3275                 em->start = hole_offset;
3276                 em->len = hole_len;
3277                 em->ram_bytes = em->len;
3278                 em->orig_start = hole_offset;
3279                 em->block_start = EXTENT_MAP_HOLE;
3280                 em->block_len = 0;
3281                 em->orig_block_len = 0;
3282                 em->compress_type = BTRFS_COMPRESS_NONE;
3283                 em->generation = trans->transid;
3284         }
3285
3286         while (1) {
3287                 write_lock(&em_tree->lock);
3288                 ret = add_extent_mapping(em_tree, em, 1);
3289                 write_unlock(&em_tree->lock);
3290                 if (ret != -EEXIST) {
3291                         free_extent_map(em);
3292                         break;
3293                 }
3294                 btrfs_drop_extent_cache(inode, em->start,
3295                                         em->start + em->len - 1, 0);
3296         }
3297
3298         if (ret)
3299                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3300                         &BTRFS_I(inode)->runtime_flags);
3301 }
3302
3303 /*
3304  * Make sure we do not end up inserting an inline extent into a file that has
3305  * already other (non-inline) extents. If a file has an inline extent it can
3306  * not have any other extents and the (single) inline extent must start at the
3307  * file offset 0. Failing to respect these rules will lead to file corruption,
3308  * resulting in EIO errors on read/write operations, hitting BUG_ON's in mm, etc
3309  *
3310  * We can have extents that have been already written to disk or we can have
3311  * dirty ranges still in delalloc, in which case the extent maps and items are
3312  * created only when we run delalloc, and the delalloc ranges might fall outside
3313  * the range we are currently locking in the inode's io tree. So we check the
3314  * inode's i_size because of that (i_size updates are done while holding the
3315  * i_mutex, which we are holding here).
3316  * We also check to see if the inode has a size not greater than "datal" but has
3317  * extents beyond it, due to an fallocate with FALLOC_FL_KEEP_SIZE (and we are
3318  * protected against such concurrent fallocate calls by the i_mutex).
3319  *
3320  * If the file has no extents but a size greater than datal, do not allow the
3321  * copy because we would need turn the inline extent into a non-inline one (even
3322  * with NO_HOLES enabled). If we find our destination inode only has one inline
3323  * extent, just overwrite it with the source inline extent if its size is less
3324  * than the source extent's size, or we could copy the source inline extent's
3325  * data into the destination inode's inline extent if the later is greater then
3326  * the former.
3327  */
3328 static int clone_copy_inline_extent(struct inode *src,
3329                                     struct inode *dst,
3330                                     struct btrfs_trans_handle *trans,
3331                                     struct btrfs_path *path,
3332                                     struct btrfs_key *new_key,
3333                                     const u64 drop_start,
3334                                     const u64 datal,
3335                                     const u64 skip,
3336                                     const u64 size,
3337                                     char *inline_data)
3338 {
3339         struct btrfs_root *root = BTRFS_I(dst)->root;
3340         const u64 aligned_end = ALIGN(new_key->offset + datal,
3341                                       root->sectorsize);
3342         int ret;
3343         struct btrfs_key key;
3344
3345         if (new_key->offset > 0)
3346                 return -EOPNOTSUPP;
3347
3348         key.objectid = btrfs_ino(dst);
3349         key.type = BTRFS_EXTENT_DATA_KEY;
3350         key.offset = 0;
3351         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3352         if (ret < 0) {
3353                 return ret;
3354         } else if (ret > 0) {
3355                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3356                         ret = btrfs_next_leaf(root, path);
3357                         if (ret < 0)
3358                                 return ret;
3359                         else if (ret > 0)
3360                                 goto copy_inline_extent;
3361                 }
3362                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3363                 if (key.objectid == btrfs_ino(dst) &&
3364                     key.type == BTRFS_EXTENT_DATA_KEY) {
3365                         ASSERT(key.offset > 0);
3366                         return -EOPNOTSUPP;
3367                 }
3368         } else if (i_size_read(dst) <= datal) {
3369                 struct btrfs_file_extent_item *ei;
3370                 u64 ext_len;
3371
3372                 /*
3373                  * If the file size is <= datal, make sure there are no other
3374                  * extents following (can happen do to an fallocate call with
3375                  * the flag FALLOC_FL_KEEP_SIZE).
3376                  */
3377                 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3378                                     struct btrfs_file_extent_item);
3379                 /*
3380                  * If it's an inline extent, it can not have other extents
3381                  * following it.
3382                  */
3383                 if (btrfs_file_extent_type(path->nodes[0], ei) ==
3384                     BTRFS_FILE_EXTENT_INLINE)
3385                         goto copy_inline_extent;
3386
3387                 ext_len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
3388                 if (ext_len > aligned_end)
3389                         return -EOPNOTSUPP;
3390
3391                 ret = btrfs_next_item(root, path);
3392                 if (ret < 0) {
3393                         return ret;
3394                 } else if (ret == 0) {
3395                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3396                                               path->slots[0]);
3397                         if (key.objectid == btrfs_ino(dst) &&
3398                             key.type == BTRFS_EXTENT_DATA_KEY)
3399                                 return -EOPNOTSUPP;
3400                 }
3401         }
3402
3403 copy_inline_extent:
3404         /*
3405          * We have no extent items, or we have an extent at offset 0 which may
3406          * or may not be inlined. All these cases are dealt the same way.
3407          */
3408         if (i_size_read(dst) > datal) {
3409                 /*
3410                  * If the destination inode has an inline extent...
3411                  * This would require copying the data from the source inline
3412                  * extent into the beginning of the destination's inline extent.
3413                  * But this is really complex, both extents can be compressed
3414                  * or just one of them, which would require decompressing and
3415                  * re-compressing data (which could increase the new compressed
3416                  * size, not allowing the compressed data to fit anymore in an
3417                  * inline extent).
3418                  * So just don't support this case for now (it should be rare,
3419                  * we are not really saving space when cloning inline extents).
3420                  */
3421                 return -EOPNOTSUPP;
3422         }
3423
3424         btrfs_release_path(path);
3425         ret = btrfs_drop_extents(trans, root, dst, drop_start, aligned_end, 1);
3426         if (ret)
3427                 return ret;
3428         ret = btrfs_insert_empty_item(trans, root, path, new_key, size);
3429         if (ret)
3430                 return ret;
3431
3432         if (skip) {
3433                 const u32 start = btrfs_file_extent_calc_inline_size(0);
3434
3435                 memmove(inline_data + start, inline_data + start + skip, datal);
3436         }
3437
3438         write_extent_buffer(path->nodes[0], inline_data,
3439                             btrfs_item_ptr_offset(path->nodes[0],
3440                                                   path->slots[0]),
3441                             size);
3442         inode_add_bytes(dst, datal);
3443
3444         return 0;
3445 }
3446
3447 /**
3448  * btrfs_clone() - clone a range from inode file to another
3449  *
3450  * @src: Inode to clone from
3451  * @inode: Inode to clone to
3452  * @off: Offset within source to start clone from
3453  * @olen: Original length, passed by user, of range to clone
3454  * @olen_aligned: Block-aligned value of olen
3455  * @destoff: Offset within @inode to start clone
3456  * @no_time_update: Whether to update mtime/ctime on the target inode
3457  */
3458 static int btrfs_clone(struct inode *src, struct inode *inode,
3459                        const u64 off, const u64 olen, const u64 olen_aligned,
3460                        const u64 destoff, int no_time_update)
3461 {
3462         struct btrfs_root *root = BTRFS_I(inode)->root;
3463         struct btrfs_path *path = NULL;
3464         struct extent_buffer *leaf;
3465         struct btrfs_trans_handle *trans;
3466         char *buf = NULL;
3467         struct btrfs_key key;
3468         u32 nritems;
3469         int slot;
3470         int ret;
3471         const u64 len = olen_aligned;
3472         u64 last_dest_end = destoff;
3473
3474         ret = -ENOMEM;
3475         buf = vmalloc(root->nodesize);
3476         if (!buf)
3477                 return ret;
3478
3479         path = btrfs_alloc_path();
3480         if (!path) {
3481                 vfree(buf);
3482                 return ret;
3483         }
3484
3485         path->reada = READA_FORWARD;
3486         /* clone data */
3487         key.objectid = btrfs_ino(src);
3488         key.type = BTRFS_EXTENT_DATA_KEY;
3489         key.offset = off;
3490
3491         while (1) {
3492                 u64 next_key_min_offset = key.offset + 1;
3493
3494                 /*
3495                  * note the key will change type as we walk through the
3496                  * tree.
3497                  */
3498                 path->leave_spinning = 1;
3499                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
3500                                 0, 0);
3501                 if (ret < 0)
3502                         goto out;
3503                 /*
3504                  * First search, if no extent item that starts at offset off was
3505                  * found but the previous item is an extent item, it's possible
3506                  * it might overlap our target range, therefore process it.
3507                  */
3508                 if (key.offset == off && ret > 0 && path->slots[0] > 0) {
3509                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3510                                               path->slots[0] - 1);
3511                         if (key.type == BTRFS_EXTENT_DATA_KEY)
3512                                 path->slots[0]--;
3513                 }
3514
3515                 nritems = btrfs_header_nritems(path->nodes[0]);
3516 process_slot:
3517                 if (path->slots[0] >= nritems) {
3518                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
3519                         if (ret < 0)
3520                                 goto out;
3521                         if (ret > 0)
3522                                 break;
3523                         nritems = btrfs_header_nritems(path->nodes[0]);
3524                 }
3525                 leaf = path->nodes[0];
3526                 slot = path->slots[0];
3527
3528                 btrfs_item_key_to_cpu(leaf, &key, slot);
3529                 if (key.type > BTRFS_EXTENT_DATA_KEY ||
3530                     key.objectid != btrfs_ino(src))
3531                         break;
3532
3533                 if (key.type == BTRFS_EXTENT_DATA_KEY) {
3534                         struct btrfs_file_extent_item *extent;
3535                         int type;
3536                         u32 size;
3537                         struct btrfs_key new_key;
3538                         u64 disko = 0, diskl = 0;
3539                         u64 datao = 0, datal = 0;
3540                         u8 comp;
3541                         u64 drop_start;
3542
3543                         extent = btrfs_item_ptr(leaf, slot,
3544                                                 struct btrfs_file_extent_item);
3545                         comp = btrfs_file_extent_compression(leaf, extent);
3546                         type = btrfs_file_extent_type(leaf, extent);
3547                         if (type == BTRFS_FILE_EXTENT_REG ||
3548                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3549                                 disko = btrfs_file_extent_disk_bytenr(leaf,
3550                                                                       extent);
3551                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
3552                                                                  extent);
3553                                 datao = btrfs_file_extent_offset(leaf, extent);
3554                                 datal = btrfs_file_extent_num_bytes(leaf,
3555                                                                     extent);
3556                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3557                                 /* take upper bound, may be compressed */
3558                                 datal = btrfs_file_extent_ram_bytes(leaf,
3559                                                                     extent);
3560                         }
3561
3562                         /*
3563                          * The first search might have left us at an extent
3564                          * item that ends before our target range's start, can
3565                          * happen if we have holes and NO_HOLES feature enabled.
3566                          */
3567                         if (key.offset + datal <= off) {
3568                                 path->slots[0]++;
3569                                 goto process_slot;
3570                         } else if (key.offset >= off + len) {
3571                                 break;
3572                         }
3573                         next_key_min_offset = key.offset + datal;
3574                         size = btrfs_item_size_nr(leaf, slot);
3575                         read_extent_buffer(leaf, buf,
3576                                            btrfs_item_ptr_offset(leaf, slot),
3577                                            size);
3578
3579                         btrfs_release_path(path);
3580                         path->leave_spinning = 0;
3581
3582                         memcpy(&new_key, &key, sizeof(new_key));
3583                         new_key.objectid = btrfs_ino(inode);
3584                         if (off <= key.offset)
3585                                 new_key.offset = key.offset + destoff - off;
3586                         else
3587                                 new_key.offset = destoff;
3588
3589                         /*
3590                          * Deal with a hole that doesn't have an extent item
3591                          * that represents it (NO_HOLES feature enabled).
3592                          * This hole is either in the middle of the cloning
3593                          * range or at the beginning (fully overlaps it or
3594                          * partially overlaps it).
3595                          */
3596                         if (new_key.offset != last_dest_end)
3597                                 drop_start = last_dest_end;
3598                         else
3599                                 drop_start = new_key.offset;
3600
3601                         /*
3602                          * 1 - adjusting old extent (we may have to split it)
3603                          * 1 - add new extent
3604                          * 1 - inode update
3605                          */
3606                         trans = btrfs_start_transaction(root, 3);
3607                         if (IS_ERR(trans)) {
3608                                 ret = PTR_ERR(trans);
3609                                 goto out;
3610                         }
3611
3612                         if (type == BTRFS_FILE_EXTENT_REG ||
3613                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3614                                 /*
3615                                  *    a  | --- range to clone ---|  b
3616                                  * | ------------- extent ------------- |
3617                                  */
3618
3619                                 /* subtract range b */
3620                                 if (key.offset + datal > off + len)
3621                                         datal = off + len - key.offset;
3622
3623                                 /* subtract range a */
3624                                 if (off > key.offset) {
3625                                         datao += off - key.offset;
3626                                         datal -= off - key.offset;
3627                                 }
3628
3629                                 ret = btrfs_drop_extents(trans, root, inode,
3630                                                          drop_start,
3631                                                          new_key.offset + datal,
3632                                                          1);
3633                                 if (ret) {
3634                                         if (ret != -EOPNOTSUPP)
3635                                                 btrfs_abort_transaction(trans,
3636                                                                 root, ret);
3637                                         btrfs_end_transaction(trans, root);
3638                                         goto out;
3639                                 }
3640
3641                                 ret = btrfs_insert_empty_item(trans, root, path,
3642                                                               &new_key, size);
3643                                 if (ret) {
3644                                         btrfs_abort_transaction(trans, root,
3645                                                                 ret);
3646                                         btrfs_end_transaction(trans, root);
3647                                         goto out;
3648                                 }
3649
3650                                 leaf = path->nodes[0];
3651                                 slot = path->slots[0];
3652                                 write_extent_buffer(leaf, buf,
3653                                             btrfs_item_ptr_offset(leaf, slot),
3654                                             size);
3655
3656                                 extent = btrfs_item_ptr(leaf, slot,
3657                                                 struct btrfs_file_extent_item);
3658
3659                                 /* disko == 0 means it's a hole */
3660                                 if (!disko)
3661                                         datao = 0;
3662
3663                                 btrfs_set_file_extent_offset(leaf, extent,
3664                                                              datao);
3665                                 btrfs_set_file_extent_num_bytes(leaf, extent,
3666                                                                 datal);
3667
3668                                 if (disko) {
3669                                         inode_add_bytes(inode, datal);
3670                                         ret = btrfs_inc_extent_ref(trans, root,
3671                                                         disko, diskl, 0,
3672                                                         root->root_key.objectid,
3673                                                         btrfs_ino(inode),
3674                                                         new_key.offset - datao);
3675                                         if (ret) {
3676                                                 btrfs_abort_transaction(trans,
3677                                                                         root,
3678                                                                         ret);
3679                                                 btrfs_end_transaction(trans,
3680                                                                       root);
3681                                                 goto out;
3682
3683                                         }
3684                                 }
3685                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3686                                 u64 skip = 0;
3687                                 u64 trim = 0;
3688
3689                                 if (off > key.offset) {
3690                                         skip = off - key.offset;
3691                                         new_key.offset += skip;
3692                                 }
3693
3694                                 if (key.offset + datal > off + len)
3695                                         trim = key.offset + datal - (off + len);
3696
3697                                 if (comp && (skip || trim)) {
3698                                         ret = -EINVAL;
3699                                         btrfs_end_transaction(trans, root);
3700                                         goto out;
3701                                 }
3702                                 size -= skip + trim;
3703                                 datal -= skip + trim;
3704
3705                                 ret = clone_copy_inline_extent(src, inode,
3706                                                                trans, path,
3707                                                                &new_key,
3708                                                                drop_start,
3709                                                                datal,
3710                                                                skip, size, buf);
3711                                 if (ret) {
3712                                         if (ret != -EOPNOTSUPP)
3713                                                 btrfs_abort_transaction(trans,
3714                                                                         root,
3715                                                                         ret);
3716                                         btrfs_end_transaction(trans, root);
3717                                         goto out;
3718                                 }
3719                                 leaf = path->nodes[0];
3720                                 slot = path->slots[0];
3721                         }
3722
3723                         /* If we have an implicit hole (NO_HOLES feature). */
3724                         if (drop_start < new_key.offset)
3725                                 clone_update_extent_map(inode, trans,
3726                                                 NULL, drop_start,
3727                                                 new_key.offset - drop_start);
3728
3729                         clone_update_extent_map(inode, trans, path, 0, 0);
3730
3731                         btrfs_mark_buffer_dirty(leaf);
3732                         btrfs_release_path(path);
3733
3734                         last_dest_end = ALIGN(new_key.offset + datal,
3735                                               root->sectorsize);
3736                         ret = clone_finish_inode_update(trans, inode,
3737                                                         last_dest_end,
3738                                                         destoff, olen,
3739                                                         no_time_update);
3740                         if (ret)
3741                                 goto out;
3742                         if (new_key.offset + datal >= destoff + len)
3743                                 break;
3744                 }
3745                 btrfs_release_path(path);
3746                 key.offset = next_key_min_offset;
3747         }
3748         ret = 0;
3749
3750         if (last_dest_end < destoff + len) {
3751                 /*
3752                  * We have an implicit hole (NO_HOLES feature is enabled) that
3753                  * fully or partially overlaps our cloning range at its end.
3754                  */
3755                 btrfs_release_path(path);
3756
3757                 /*
3758                  * 1 - remove extent(s)
3759                  * 1 - inode update
3760                  */
3761                 trans = btrfs_start_transaction(root, 2);
3762                 if (IS_ERR(trans)) {
3763                         ret = PTR_ERR(trans);
3764                         goto out;
3765                 }
3766                 ret = btrfs_drop_extents(trans, root, inode,
3767                                          last_dest_end, destoff + len, 1);
3768                 if (ret) {
3769                         if (ret != -EOPNOTSUPP)
3770                                 btrfs_abort_transaction(trans, root, ret);
3771                         btrfs_end_transaction(trans, root);
3772                         goto out;
3773                 }
3774                 clone_update_extent_map(inode, trans, NULL, last_dest_end,
3775                                         destoff + len - last_dest_end);
3776                 ret = clone_finish_inode_update(trans, inode, destoff + len,
3777                                                 destoff, olen, no_time_update);
3778         }
3779
3780 out:
3781         btrfs_free_path(path);
3782         vfree(buf);
3783         return ret;
3784 }
3785
3786 static noinline int btrfs_clone_files(struct file *file, struct file *file_src,
3787                                         u64 off, u64 olen, u64 destoff)
3788 {
3789         struct inode *inode = file_inode(file);
3790         struct inode *src = file_inode(file_src);
3791         struct btrfs_root *root = BTRFS_I(inode)->root;
3792         int ret;
3793         u64 len = olen;
3794         u64 bs = root->fs_info->sb->s_blocksize;
3795         int same_inode = src == inode;
3796
3797         /*
3798          * TODO:
3799          * - split compressed inline extents.  annoying: we need to
3800          *   decompress into destination's address_space (the file offset
3801          *   may change, so source mapping won't do), then recompress (or
3802          *   otherwise reinsert) a subrange.
3803          *
3804          * - split destination inode's inline extents.  The inline extents can
3805          *   be either compressed or non-compressed.
3806          */
3807
3808         if (btrfs_root_readonly(root))
3809                 return -EROFS;
3810
3811         if (file_src->f_path.mnt != file->f_path.mnt ||
3812             src->i_sb != inode->i_sb)
3813                 return -EXDEV;
3814
3815         /* don't make the dst file partly checksummed */
3816         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3817             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
3818                 return -EINVAL;
3819
3820         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
3821                 return -EISDIR;
3822
3823         if (!same_inode) {
3824                 btrfs_double_inode_lock(src, inode);
3825         } else {
3826                 inode_lock(src);
3827         }
3828
3829         /* determine range to clone */
3830         ret = -EINVAL;
3831         if (off + len > src->i_size || off + len < off)
3832                 goto out_unlock;
3833         if (len == 0)
3834                 olen = len = src->i_size - off;
3835         /* if we extend to eof, continue to block boundary */
3836         if (off + len == src->i_size)
3837                 len = ALIGN(src->i_size, bs) - off;
3838
3839         if (len == 0) {
3840                 ret = 0;
3841                 goto out_unlock;
3842         }
3843
3844         /* verify the end result is block aligned */
3845         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
3846             !IS_ALIGNED(destoff, bs))
3847                 goto out_unlock;
3848
3849         /* verify if ranges are overlapped within the same file */
3850         if (same_inode) {
3851                 if (destoff + len > off && destoff < off + len)
3852                         goto out_unlock;
3853         }
3854
3855         if (destoff > inode->i_size) {
3856                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3857                 if (ret)
3858                         goto out_unlock;
3859         }
3860
3861         /*
3862          * Lock the target range too. Right after we replace the file extent
3863          * items in the fs tree (which now point to the cloned data), we might
3864          * have a worker replace them with extent items relative to a write
3865          * operation that was issued before this clone operation (i.e. confront
3866          * with inode.c:btrfs_finish_ordered_io).
3867          */
3868         if (same_inode) {
3869                 u64 lock_start = min_t(u64, off, destoff);
3870                 u64 lock_len = max_t(u64, off, destoff) + len - lock_start;
3871
3872                 ret = lock_extent_range(src, lock_start, lock_len, true);
3873         } else {
3874                 ret = btrfs_double_extent_lock(src, off, inode, destoff, len,
3875                                                true);
3876         }
3877         ASSERT(ret == 0);
3878         if (WARN_ON(ret)) {
3879                 /* ranges in the io trees already unlocked */
3880                 goto out_unlock;
3881         }
3882
3883         ret = btrfs_clone(src, inode, off, olen, len, destoff, 0);
3884
3885         if (same_inode) {
3886                 u64 lock_start = min_t(u64, off, destoff);
3887                 u64 lock_end = max_t(u64, off, destoff) + len - 1;
3888
3889                 unlock_extent(&BTRFS_I(src)->io_tree, lock_start, lock_end);
3890         } else {
3891                 btrfs_double_extent_unlock(src, off, inode, destoff, len);
3892         }
3893         /*
3894          * Truncate page cache pages so that future reads will see the cloned
3895          * data immediately and not the previous data.
3896          */
3897         truncate_inode_pages_range(&inode->i_data,
3898                                 round_down(destoff, PAGE_SIZE),
3899                                 round_up(destoff + len, PAGE_SIZE) - 1);
3900 out_unlock:
3901         if (!same_inode)
3902                 btrfs_double_inode_unlock(src, inode);
3903         else
3904                 inode_unlock(src);
3905         return ret;
3906 }
3907
3908 ssize_t btrfs_copy_file_range(struct file *file_in, loff_t pos_in,
3909                               struct file *file_out, loff_t pos_out,
3910                               size_t len, unsigned int flags)
3911 {
3912         ssize_t ret;
3913
3914         ret = btrfs_clone_files(file_out, file_in, pos_in, len, pos_out);
3915         if (ret == 0)
3916                 ret = len;
3917         return ret;
3918 }
3919
3920 int btrfs_clone_file_range(struct file *src_file, loff_t off,
3921                 struct file *dst_file, loff_t destoff, u64 len)
3922 {
3923         return btrfs_clone_files(dst_file, src_file, off, len, destoff);
3924 }
3925
3926 /*
3927  * there are many ways the trans_start and trans_end ioctls can lead
3928  * to deadlocks.  They should only be used by applications that
3929  * basically own the machine, and have a very in depth understanding
3930  * of all the possible deadlocks and enospc problems.
3931  */
3932 static long btrfs_ioctl_trans_start(struct file *file)
3933 {
3934         struct inode *inode = file_inode(file);
3935         struct btrfs_root *root = BTRFS_I(inode)->root;
3936         struct btrfs_trans_handle *trans;
3937         int ret;
3938
3939         ret = -EPERM;
3940         if (!capable(CAP_SYS_ADMIN))
3941                 goto out;
3942
3943         ret = -EINPROGRESS;
3944         if (file->private_data)
3945                 goto out;
3946
3947         ret = -EROFS;
3948         if (btrfs_root_readonly(root))
3949                 goto out;
3950
3951         ret = mnt_want_write_file(file);
3952         if (ret)
3953                 goto out;
3954
3955         atomic_inc(&root->fs_info->open_ioctl_trans);
3956
3957         ret = -ENOMEM;
3958         trans = btrfs_start_ioctl_transaction(root);
3959         if (IS_ERR(trans))
3960                 goto out_drop;
3961
3962         file->private_data = trans;
3963         return 0;
3964
3965 out_drop:
3966         atomic_dec(&root->fs_info->open_ioctl_trans);
3967         mnt_drop_write_file(file);
3968 out:
3969         return ret;
3970 }
3971
3972 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3973 {
3974         struct inode *inode = file_inode(file);
3975         struct btrfs_root *root = BTRFS_I(inode)->root;
3976         struct btrfs_root *new_root;
3977         struct btrfs_dir_item *di;
3978         struct btrfs_trans_handle *trans;
3979         struct btrfs_path *path;
3980         struct btrfs_key location;
3981         struct btrfs_disk_key disk_key;
3982         u64 objectid = 0;
3983         u64 dir_id;
3984         int ret;
3985
3986         if (!capable(CAP_SYS_ADMIN))
3987                 return -EPERM;
3988
3989         ret = mnt_want_write_file(file);
3990         if (ret)
3991                 return ret;
3992
3993         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3994                 ret = -EFAULT;
3995                 goto out;
3996         }
3997
3998         if (!objectid)
3999                 objectid = BTRFS_FS_TREE_OBJECTID;
4000
4001         location.objectid = objectid;
4002         location.type = BTRFS_ROOT_ITEM_KEY;
4003         location.offset = (u64)-1;
4004
4005         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
4006         if (IS_ERR(new_root)) {
4007                 ret = PTR_ERR(new_root);
4008                 goto out;
4009         }
4010
4011         path = btrfs_alloc_path();
4012         if (!path) {
4013                 ret = -ENOMEM;
4014                 goto out;
4015         }
4016         path->leave_spinning = 1;
4017
4018         trans = btrfs_start_transaction(root, 1);
4019         if (IS_ERR(trans)) {
4020                 btrfs_free_path(path);
4021                 ret = PTR_ERR(trans);
4022                 goto out;
4023         }
4024
4025         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
4026         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
4027                                    dir_id, "default", 7, 1);
4028         if (IS_ERR_OR_NULL(di)) {
4029                 btrfs_free_path(path);
4030                 btrfs_end_transaction(trans, root);
4031                 btrfs_err(new_root->fs_info, "Umm, you don't have the default dir"
4032                            "item, this isn't going to work");
4033                 ret = -ENOENT;
4034                 goto out;
4035         }
4036
4037         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
4038         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
4039         btrfs_mark_buffer_dirty(path->nodes[0]);
4040         btrfs_free_path(path);
4041
4042         btrfs_set_fs_incompat(root->fs_info, DEFAULT_SUBVOL);
4043         btrfs_end_transaction(trans, root);
4044 out:
4045         mnt_drop_write_file(file);
4046         return ret;
4047 }
4048
4049 void btrfs_get_block_group_info(struct list_head *groups_list,
4050                                 struct btrfs_ioctl_space_info *space)
4051 {
4052         struct btrfs_block_group_cache *block_group;
4053
4054         space->total_bytes = 0;
4055         space->used_bytes = 0;
4056         space->flags = 0;
4057         list_for_each_entry(block_group, groups_list, list) {
4058                 space->flags = block_group->flags;
4059                 space->total_bytes += block_group->key.offset;
4060                 space->used_bytes +=
4061                         btrfs_block_group_used(&block_group->item);
4062         }
4063 }
4064
4065 static long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
4066 {
4067         struct btrfs_ioctl_space_args space_args;
4068         struct btrfs_ioctl_space_info space;
4069         struct btrfs_ioctl_space_info *dest;
4070         struct btrfs_ioctl_space_info *dest_orig;
4071         struct btrfs_ioctl_space_info __user *user_dest;
4072         struct btrfs_space_info *info;
4073         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
4074                        BTRFS_BLOCK_GROUP_SYSTEM,
4075                        BTRFS_BLOCK_GROUP_METADATA,
4076                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
4077         int num_types = 4;
4078         int alloc_size;
4079         int ret = 0;
4080         u64 slot_count = 0;
4081         int i, c;
4082
4083         if (copy_from_user(&space_args,
4084                            (struct btrfs_ioctl_space_args __user *)arg,
4085                            sizeof(space_args)))
4086                 return -EFAULT;
4087
4088         for (i = 0; i < num_types; i++) {
4089                 struct btrfs_space_info *tmp;
4090
4091                 info = NULL;
4092                 rcu_read_lock();
4093                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
4094                                         list) {
4095                         if (tmp->flags == types[i]) {
4096                                 info = tmp;
4097                                 break;
4098                         }
4099                 }
4100                 rcu_read_unlock();
4101
4102                 if (!info)
4103                         continue;
4104
4105                 down_read(&info->groups_sem);
4106                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4107                         if (!list_empty(&info->block_groups[c]))
4108                                 slot_count++;
4109                 }
4110                 up_read(&info->groups_sem);
4111         }
4112
4113         /*
4114          * Global block reserve, exported as a space_info
4115          */
4116         slot_count++;
4117
4118         /* space_slots == 0 means they are asking for a count */
4119         if (space_args.space_slots == 0) {
4120                 space_args.total_spaces = slot_count;
4121                 goto out;
4122         }
4123
4124         slot_count = min_t(u64, space_args.space_slots, slot_count);
4125
4126         alloc_size = sizeof(*dest) * slot_count;
4127
4128         /* we generally have at most 6 or so space infos, one for each raid
4129          * level.  So, a whole page should be more than enough for everyone
4130          */
4131         if (alloc_size > PAGE_SIZE)
4132                 return -ENOMEM;
4133
4134         space_args.total_spaces = 0;
4135         dest = kmalloc(alloc_size, GFP_KERNEL);
4136         if (!dest)
4137                 return -ENOMEM;
4138         dest_orig = dest;
4139
4140         /* now we have a buffer to copy into */
4141         for (i = 0; i < num_types; i++) {
4142                 struct btrfs_space_info *tmp;
4143
4144                 if (!slot_count)
4145                         break;
4146
4147                 info = NULL;
4148                 rcu_read_lock();
4149                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
4150                                         list) {
4151                         if (tmp->flags == types[i]) {
4152                                 info = tmp;
4153                                 break;
4154                         }
4155                 }
4156                 rcu_read_unlock();
4157
4158                 if (!info)
4159                         continue;
4160                 down_read(&info->groups_sem);
4161                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4162                         if (!list_empty(&info->block_groups[c])) {
4163                                 btrfs_get_block_group_info(
4164                                         &info->block_groups[c], &space);
4165                                 memcpy(dest, &space, sizeof(space));
4166                                 dest++;
4167                                 space_args.total_spaces++;
4168                                 slot_count--;
4169                         }
4170                         if (!slot_count)
4171                                 break;
4172                 }
4173                 up_read(&info->groups_sem);
4174         }
4175
4176         /*
4177          * Add global block reserve
4178          */
4179         if (slot_count) {
4180                 struct btrfs_block_rsv *block_rsv = &root->fs_info->global_block_rsv;
4181
4182                 spin_lock(&block_rsv->lock);
4183                 space.total_bytes = block_rsv->size;
4184                 space.used_bytes = block_rsv->size - block_rsv->reserved;
4185                 spin_unlock(&block_rsv->lock);
4186                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
4187                 memcpy(dest, &space, sizeof(space));
4188                 space_args.total_spaces++;
4189         }
4190
4191         user_dest = (struct btrfs_ioctl_space_info __user *)
4192                 (arg + sizeof(struct btrfs_ioctl_space_args));
4193
4194         if (copy_to_user(user_dest, dest_orig, alloc_size))
4195                 ret = -EFAULT;
4196
4197         kfree(dest_orig);
4198 out:
4199         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
4200                 ret = -EFAULT;
4201
4202         return ret;
4203 }
4204
4205 /*
4206  * there are many ways the trans_start and trans_end ioctls can lead
4207  * to deadlocks.  They should only be used by applications that
4208  * basically own the machine, and have a very in depth understanding
4209  * of all the possible deadlocks and enospc problems.
4210  */
4211 long btrfs_ioctl_trans_end(struct file *file)
4212 {
4213         struct inode *inode = file_inode(file);
4214         struct btrfs_root *root = BTRFS_I(inode)->root;
4215         struct btrfs_trans_handle *trans;
4216
4217         trans = file->private_data;
4218         if (!trans)
4219                 return -EINVAL;
4220         file->private_data = NULL;
4221
4222         btrfs_end_transaction(trans, root);
4223
4224         atomic_dec(&root->fs_info->open_ioctl_trans);
4225
4226         mnt_drop_write_file(file);
4227         return 0;
4228 }
4229
4230 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
4231                                             void __user *argp)
4232 {
4233         struct btrfs_trans_handle *trans;
4234         u64 transid;
4235         int ret;
4236
4237         trans = btrfs_attach_transaction_barrier(root);
4238         if (IS_ERR(trans)) {
4239                 if (PTR_ERR(trans) != -ENOENT)
4240                         return PTR_ERR(trans);
4241
4242                 /* No running transaction, don't bother */
4243                 transid = root->fs_info->last_trans_committed;
4244                 goto out;
4245         }
4246         transid = trans->transid;
4247         ret = btrfs_commit_transaction_async(trans, root, 0);
4248         if (ret) {
4249                 btrfs_end_transaction(trans, root);
4250                 return ret;
4251         }
4252 out:
4253         if (argp)
4254                 if (copy_to_user(argp, &transid, sizeof(transid)))
4255                         return -EFAULT;
4256         return 0;
4257 }
4258
4259 static noinline long btrfs_ioctl_wait_sync(struct btrfs_root *root,
4260                                            void __user *argp)
4261 {
4262         u64 transid;
4263
4264         if (argp) {
4265                 if (copy_from_user(&transid, argp, sizeof(transid)))
4266                         return -EFAULT;
4267         } else {
4268                 transid = 0;  /* current trans */
4269         }
4270         return btrfs_wait_for_commit(root, transid);
4271 }
4272
4273 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
4274 {
4275         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4276         struct btrfs_ioctl_scrub_args *sa;
4277         int ret;
4278
4279         if (!capable(CAP_SYS_ADMIN))
4280                 return -EPERM;
4281
4282         sa = memdup_user(arg, sizeof(*sa));
4283         if (IS_ERR(sa))
4284                 return PTR_ERR(sa);
4285
4286         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
4287                 ret = mnt_want_write_file(file);
4288                 if (ret)
4289                         goto out;
4290         }
4291
4292         ret = btrfs_scrub_dev(root->fs_info, sa->devid, sa->start, sa->end,
4293                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4294                               0);
4295
4296         if (copy_to_user(arg, sa, sizeof(*sa)))
4297                 ret = -EFAULT;
4298
4299         if (!(sa->flags & BTRFS_SCRUB_READONLY))
4300                 mnt_drop_write_file(file);
4301 out:
4302         kfree(sa);
4303         return ret;
4304 }
4305
4306 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
4307 {
4308         if (!capable(CAP_SYS_ADMIN))
4309                 return -EPERM;
4310
4311         return btrfs_scrub_cancel(root->fs_info);
4312 }
4313
4314 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
4315                                        void __user *arg)
4316 {
4317         struct btrfs_ioctl_scrub_args *sa;
4318         int ret;
4319
4320         if (!capable(CAP_SYS_ADMIN))
4321                 return -EPERM;
4322
4323         sa = memdup_user(arg, sizeof(*sa));
4324         if (IS_ERR(sa))
4325                 return PTR_ERR(sa);
4326
4327         ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
4328
4329         if (copy_to_user(arg, sa, sizeof(*sa)))
4330                 ret = -EFAULT;
4331
4332         kfree(sa);
4333         return ret;
4334 }
4335
4336 static long btrfs_ioctl_get_dev_stats(struct btrfs_root *root,
4337                                       void __user *arg)
4338 {
4339         struct btrfs_ioctl_get_dev_stats *sa;
4340         int ret;
4341
4342         sa = memdup_user(arg, sizeof(*sa));
4343         if (IS_ERR(sa))
4344                 return PTR_ERR(sa);
4345
4346         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4347                 kfree(sa);
4348                 return -EPERM;
4349         }
4350
4351         ret = btrfs_get_dev_stats(root, sa);
4352
4353         if (copy_to_user(arg, sa, sizeof(*sa)))
4354                 ret = -EFAULT;
4355
4356         kfree(sa);
4357         return ret;
4358 }
4359
4360 static long btrfs_ioctl_dev_replace(struct btrfs_root *root, void __user *arg)
4361 {
4362         struct btrfs_ioctl_dev_replace_args *p;
4363         int ret;
4364
4365         if (!capable(CAP_SYS_ADMIN))
4366                 return -EPERM;
4367
4368         p = memdup_user(arg, sizeof(*p));
4369         if (IS_ERR(p))
4370                 return PTR_ERR(p);
4371
4372         switch (p->cmd) {
4373         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4374                 if (root->fs_info->sb->s_flags & MS_RDONLY) {
4375                         ret = -EROFS;
4376                         goto out;
4377                 }
4378                 if (atomic_xchg(
4379                         &root->fs_info->mutually_exclusive_operation_running,
4380                         1)) {
4381                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4382                 } else {
4383                         ret = btrfs_dev_replace_start(root, p);
4384                         atomic_set(
4385                          &root->fs_info->mutually_exclusive_operation_running,
4386                          0);
4387                 }
4388                 break;
4389         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4390                 btrfs_dev_replace_status(root->fs_info, p);
4391                 ret = 0;
4392                 break;
4393         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4394                 ret = btrfs_dev_replace_cancel(root->fs_info, p);
4395                 break;
4396         default:
4397                 ret = -EINVAL;
4398                 break;
4399         }
4400
4401         if (copy_to_user(arg, p, sizeof(*p)))
4402                 ret = -EFAULT;
4403 out:
4404         kfree(p);
4405         return ret;
4406 }
4407
4408 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4409 {
4410         int ret = 0;
4411         int i;
4412         u64 rel_ptr;
4413         int size;
4414         struct btrfs_ioctl_ino_path_args *ipa = NULL;
4415         struct inode_fs_paths *ipath = NULL;
4416         struct btrfs_path *path;
4417
4418         if (!capable(CAP_DAC_READ_SEARCH))
4419                 return -EPERM;
4420
4421         path = btrfs_alloc_path();
4422         if (!path) {
4423                 ret = -ENOMEM;
4424                 goto out;
4425         }
4426
4427         ipa = memdup_user(arg, sizeof(*ipa));
4428         if (IS_ERR(ipa)) {
4429                 ret = PTR_ERR(ipa);
4430                 ipa = NULL;
4431                 goto out;
4432         }
4433
4434         size = min_t(u32, ipa->size, 4096);
4435         ipath = init_ipath(size, root, path);
4436         if (IS_ERR(ipath)) {
4437                 ret = PTR_ERR(ipath);
4438                 ipath = NULL;
4439                 goto out;
4440         }
4441
4442         ret = paths_from_inode(ipa->inum, ipath);
4443         if (ret < 0)
4444                 goto out;
4445
4446         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4447                 rel_ptr = ipath->fspath->val[i] -
4448                           (u64)(unsigned long)ipath->fspath->val;
4449                 ipath->fspath->val[i] = rel_ptr;
4450         }
4451
4452         ret = copy_to_user((void *)(unsigned long)ipa->fspath,
4453                            (void *)(unsigned long)ipath->fspath, size);
4454         if (ret) {
4455                 ret = -EFAULT;
4456                 goto out;
4457         }
4458
4459 out:
4460         btrfs_free_path(path);
4461         free_ipath(ipath);
4462         kfree(ipa);
4463
4464         return ret;
4465 }
4466
4467 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4468 {
4469         struct btrfs_data_container *inodes = ctx;
4470         const size_t c = 3 * sizeof(u64);
4471
4472         if (inodes->bytes_left >= c) {
4473                 inodes->bytes_left -= c;
4474                 inodes->val[inodes->elem_cnt] = inum;
4475                 inodes->val[inodes->elem_cnt + 1] = offset;
4476                 inodes->val[inodes->elem_cnt + 2] = root;
4477                 inodes->elem_cnt += 3;
4478         } else {
4479                 inodes->bytes_missing += c - inodes->bytes_left;
4480                 inodes->bytes_left = 0;
4481                 inodes->elem_missed += 3;
4482         }
4483
4484         return 0;
4485 }
4486
4487 static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
4488                                         void __user *arg)
4489 {
4490         int ret = 0;
4491         int size;
4492         struct btrfs_ioctl_logical_ino_args *loi;
4493         struct btrfs_data_container *inodes = NULL;
4494         struct btrfs_path *path = NULL;
4495
4496         if (!capable(CAP_SYS_ADMIN))
4497                 return -EPERM;
4498
4499         loi = memdup_user(arg, sizeof(*loi));
4500         if (IS_ERR(loi)) {
4501                 ret = PTR_ERR(loi);
4502                 loi = NULL;
4503                 goto out;
4504         }
4505
4506         path = btrfs_alloc_path();
4507         if (!path) {
4508                 ret = -ENOMEM;
4509                 goto out;
4510         }
4511
4512         size = min_t(u32, loi->size, SZ_64K);
4513         inodes = init_data_container(size);
4514         if (IS_ERR(inodes)) {
4515                 ret = PTR_ERR(inodes);
4516                 inodes = NULL;
4517                 goto out;
4518         }
4519
4520         ret = iterate_inodes_from_logical(loi->logical, root->fs_info, path,
4521                                           build_ino_list, inodes);
4522         if (ret == -EINVAL)
4523                 ret = -ENOENT;
4524         if (ret < 0)
4525                 goto out;
4526
4527         ret = copy_to_user((void *)(unsigned long)loi->inodes,
4528                            (void *)(unsigned long)inodes, size);
4529         if (ret)
4530                 ret = -EFAULT;
4531
4532 out:
4533         btrfs_free_path(path);
4534         vfree(inodes);
4535         kfree(loi);
4536
4537         return ret;
4538 }
4539
4540 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
4541                                struct btrfs_ioctl_balance_args *bargs)
4542 {
4543         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4544
4545         bargs->flags = bctl->flags;
4546
4547         if (atomic_read(&fs_info->balance_running))
4548                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4549         if (atomic_read(&fs_info->balance_pause_req))
4550                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4551         if (atomic_read(&fs_info->balance_cancel_req))
4552                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4553
4554         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4555         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4556         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4557
4558         if (lock) {
4559                 spin_lock(&fs_info->balance_lock);
4560                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4561                 spin_unlock(&fs_info->balance_lock);
4562         } else {
4563                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4564         }
4565 }
4566
4567 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4568 {
4569         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4570         struct btrfs_fs_info *fs_info = root->fs_info;
4571         struct btrfs_ioctl_balance_args *bargs;
4572         struct btrfs_balance_control *bctl;
4573         bool need_unlock; /* for mut. excl. ops lock */
4574         int ret;
4575
4576         if (!capable(CAP_SYS_ADMIN))
4577                 return -EPERM;
4578
4579         ret = mnt_want_write_file(file);
4580         if (ret)
4581                 return ret;
4582
4583 again:
4584         if (!atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)) {
4585                 mutex_lock(&fs_info->volume_mutex);
4586                 mutex_lock(&fs_info->balance_mutex);
4587                 need_unlock = true;
4588                 goto locked;
4589         }
4590
4591         /*
4592          * mut. excl. ops lock is locked.  Three possibilites:
4593          *   (1) some other op is running
4594          *   (2) balance is running
4595          *   (3) balance is paused -- special case (think resume)
4596          */
4597         mutex_lock(&fs_info->balance_mutex);
4598         if (fs_info->balance_ctl) {
4599                 /* this is either (2) or (3) */
4600                 if (!atomic_read(&fs_info->balance_running)) {
4601                         mutex_unlock(&fs_info->balance_mutex);
4602                         if (!mutex_trylock(&fs_info->volume_mutex))
4603                                 goto again;
4604                         mutex_lock(&fs_info->balance_mutex);
4605
4606                         if (fs_info->balance_ctl &&
4607                             !atomic_read(&fs_info->balance_running)) {
4608                                 /* this is (3) */
4609                                 need_unlock = false;
4610                                 goto locked;
4611                         }
4612
4613                         mutex_unlock(&fs_info->balance_mutex);
4614                         mutex_unlock(&fs_info->volume_mutex);
4615                         goto again;
4616                 } else {
4617                         /* this is (2) */
4618                         mutex_unlock(&fs_info->balance_mutex);
4619                         ret = -EINPROGRESS;
4620                         goto out;
4621                 }
4622         } else {
4623                 /* this is (1) */
4624                 mutex_unlock(&fs_info->balance_mutex);
4625                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4626                 goto out;
4627         }
4628
4629 locked:
4630         BUG_ON(!atomic_read(&fs_info->mutually_exclusive_operation_running));
4631
4632         if (arg) {
4633                 bargs = memdup_user(arg, sizeof(*bargs));
4634                 if (IS_ERR(bargs)) {
4635                         ret = PTR_ERR(bargs);
4636                         goto out_unlock;
4637                 }
4638
4639                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4640                         if (!fs_info->balance_ctl) {
4641                                 ret = -ENOTCONN;
4642                                 goto out_bargs;
4643                         }
4644
4645                         bctl = fs_info->balance_ctl;
4646                         spin_lock(&fs_info->balance_lock);
4647                         bctl->flags |= BTRFS_BALANCE_RESUME;
4648                         spin_unlock(&fs_info->balance_lock);
4649
4650                         goto do_balance;
4651                 }
4652         } else {
4653                 bargs = NULL;
4654         }
4655
4656         if (fs_info->balance_ctl) {
4657                 ret = -EINPROGRESS;
4658                 goto out_bargs;
4659         }
4660
4661         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4662         if (!bctl) {
4663                 ret = -ENOMEM;
4664                 goto out_bargs;
4665         }
4666
4667         bctl->fs_info = fs_info;
4668         if (arg) {
4669                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4670                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4671                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4672
4673                 bctl->flags = bargs->flags;
4674         } else {
4675                 /* balance everything - no filters */
4676                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4677         }
4678
4679         if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4680                 ret = -EINVAL;
4681                 goto out_bctl;
4682         }
4683
4684 do_balance:
4685         /*
4686          * Ownership of bctl and mutually_exclusive_operation_running
4687          * goes to to btrfs_balance.  bctl is freed in __cancel_balance,
4688          * or, if restriper was paused all the way until unmount, in
4689          * free_fs_info.  mutually_exclusive_operation_running is
4690          * cleared in __cancel_balance.
4691          */
4692         need_unlock = false;
4693
4694         ret = btrfs_balance(bctl, bargs);
4695         bctl = NULL;
4696
4697         if (arg) {
4698                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4699                         ret = -EFAULT;
4700         }
4701
4702 out_bctl:
4703         kfree(bctl);
4704 out_bargs:
4705         kfree(bargs);
4706 out_unlock:
4707         mutex_unlock(&fs_info->balance_mutex);
4708         mutex_unlock(&fs_info->volume_mutex);
4709         if (need_unlock)
4710                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
4711 out:
4712         mnt_drop_write_file(file);
4713         return ret;
4714 }
4715
4716 static long btrfs_ioctl_balance_ctl(struct btrfs_root *root, int cmd)
4717 {
4718         if (!capable(CAP_SYS_ADMIN))
4719                 return -EPERM;
4720
4721         switch (cmd) {
4722         case BTRFS_BALANCE_CTL_PAUSE:
4723                 return btrfs_pause_balance(root->fs_info);
4724         case BTRFS_BALANCE_CTL_CANCEL:
4725                 return btrfs_cancel_balance(root->fs_info);
4726         }
4727
4728         return -EINVAL;
4729 }
4730
4731 static long btrfs_ioctl_balance_progress(struct btrfs_root *root,
4732                                          void __user *arg)
4733 {
4734         struct btrfs_fs_info *fs_info = root->fs_info;
4735         struct btrfs_ioctl_balance_args *bargs;
4736         int ret = 0;
4737
4738         if (!capable(CAP_SYS_ADMIN))
4739                 return -EPERM;
4740
4741         mutex_lock(&fs_info->balance_mutex);
4742         if (!fs_info->balance_ctl) {
4743                 ret = -ENOTCONN;
4744                 goto out;
4745         }
4746
4747         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4748         if (!bargs) {
4749                 ret = -ENOMEM;
4750                 goto out;
4751         }
4752
4753         update_ioctl_balance_args(fs_info, 1, bargs);
4754
4755         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4756                 ret = -EFAULT;
4757
4758         kfree(bargs);
4759 out:
4760         mutex_unlock(&fs_info->balance_mutex);
4761         return ret;
4762 }
4763
4764 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4765 {
4766         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4767         struct btrfs_ioctl_quota_ctl_args *sa;
4768         struct btrfs_trans_handle *trans = NULL;
4769         int ret;
4770         int err;
4771
4772         if (!capable(CAP_SYS_ADMIN))
4773                 return -EPERM;
4774
4775         ret = mnt_want_write_file(file);
4776         if (ret)
4777                 return ret;
4778
4779         sa = memdup_user(arg, sizeof(*sa));
4780         if (IS_ERR(sa)) {
4781                 ret = PTR_ERR(sa);
4782                 goto drop_write;
4783         }
4784
4785         down_write(&root->fs_info->subvol_sem);
4786         trans = btrfs_start_transaction(root->fs_info->tree_root, 2);
4787         if (IS_ERR(trans)) {
4788                 ret = PTR_ERR(trans);
4789                 goto out;
4790         }
4791
4792         switch (sa->cmd) {
4793         case BTRFS_QUOTA_CTL_ENABLE:
4794                 ret = btrfs_quota_enable(trans, root->fs_info);
4795                 break;
4796         case BTRFS_QUOTA_CTL_DISABLE:
4797                 ret = btrfs_quota_disable(trans, root->fs_info);
4798                 break;
4799         default:
4800                 ret = -EINVAL;
4801                 break;
4802         }
4803
4804         err = btrfs_commit_transaction(trans, root->fs_info->tree_root);
4805         if (err && !ret)
4806                 ret = err;
4807 out:
4808         kfree(sa);
4809         up_write(&root->fs_info->subvol_sem);
4810 drop_write:
4811         mnt_drop_write_file(file);
4812         return ret;
4813 }
4814
4815 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4816 {
4817         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4818         struct btrfs_ioctl_qgroup_assign_args *sa;
4819         struct btrfs_trans_handle *trans;
4820         int ret;
4821         int err;
4822
4823         if (!capable(CAP_SYS_ADMIN))
4824                 return -EPERM;
4825
4826         ret = mnt_want_write_file(file);
4827         if (ret)
4828                 return ret;
4829
4830         sa = memdup_user(arg, sizeof(*sa));
4831         if (IS_ERR(sa)) {
4832                 ret = PTR_ERR(sa);
4833                 goto drop_write;
4834         }
4835
4836         trans = btrfs_join_transaction(root);
4837         if (IS_ERR(trans)) {
4838                 ret = PTR_ERR(trans);
4839                 goto out;
4840         }
4841
4842         /* FIXME: check if the IDs really exist */
4843         if (sa->assign) {
4844                 ret = btrfs_add_qgroup_relation(trans, root->fs_info,
4845                                                 sa->src, sa->dst);
4846         } else {
4847                 ret = btrfs_del_qgroup_relation(trans, root->fs_info,
4848                                                 sa->src, sa->dst);
4849         }
4850
4851         /* update qgroup status and info */
4852         err = btrfs_run_qgroups(trans, root->fs_info);
4853         if (err < 0)
4854                 btrfs_std_error(root->fs_info, ret,
4855                             "failed to update qgroup status and info\n");
4856         err = btrfs_end_transaction(trans, root);
4857         if (err && !ret)
4858                 ret = err;
4859
4860 out:
4861         kfree(sa);
4862 drop_write:
4863         mnt_drop_write_file(file);
4864         return ret;
4865 }
4866
4867 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4868 {
4869         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4870         struct btrfs_ioctl_qgroup_create_args *sa;
4871         struct btrfs_trans_handle *trans;
4872         int ret;
4873         int err;
4874
4875         if (!capable(CAP_SYS_ADMIN))
4876                 return -EPERM;
4877
4878         ret = mnt_want_write_file(file);
4879         if (ret)
4880                 return ret;
4881
4882         sa = memdup_user(arg, sizeof(*sa));
4883         if (IS_ERR(sa)) {
4884                 ret = PTR_ERR(sa);
4885                 goto drop_write;
4886         }
4887
4888         if (!sa->qgroupid) {
4889                 ret = -EINVAL;
4890                 goto out;
4891         }
4892
4893         trans = btrfs_join_transaction(root);
4894         if (IS_ERR(trans)) {
4895                 ret = PTR_ERR(trans);
4896                 goto out;
4897         }
4898
4899         /* FIXME: check if the IDs really exist */
4900         if (sa->create) {
4901                 ret = btrfs_create_qgroup(trans, root->fs_info, sa->qgroupid);
4902         } else {
4903                 ret = btrfs_remove_qgroup(trans, root->fs_info, sa->qgroupid);
4904         }
4905
4906         err = btrfs_end_transaction(trans, root);
4907         if (err && !ret)
4908                 ret = err;
4909
4910 out:
4911         kfree(sa);
4912 drop_write:
4913         mnt_drop_write_file(file);
4914         return ret;
4915 }
4916
4917 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4918 {
4919         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4920         struct btrfs_ioctl_qgroup_limit_args *sa;
4921         struct btrfs_trans_handle *trans;
4922         int ret;
4923         int err;
4924         u64 qgroupid;
4925
4926         if (!capable(CAP_SYS_ADMIN))
4927                 return -EPERM;
4928
4929         ret = mnt_want_write_file(file);
4930         if (ret)
4931                 return ret;
4932
4933         sa = memdup_user(arg, sizeof(*sa));
4934         if (IS_ERR(sa)) {
4935                 ret = PTR_ERR(sa);
4936                 goto drop_write;
4937         }
4938
4939         trans = btrfs_join_transaction(root);
4940         if (IS_ERR(trans)) {
4941                 ret = PTR_ERR(trans);
4942                 goto out;
4943         }
4944
4945         qgroupid = sa->qgroupid;
4946         if (!qgroupid) {
4947                 /* take the current subvol as qgroup */
4948                 qgroupid = root->root_key.objectid;
4949         }
4950
4951         /* FIXME: check if the IDs really exist */
4952         ret = btrfs_limit_qgroup(trans, root->fs_info, qgroupid, &sa->lim);
4953
4954         err = btrfs_end_transaction(trans, root);
4955         if (err && !ret)
4956                 ret = err;
4957
4958 out:
4959         kfree(sa);
4960 drop_write:
4961         mnt_drop_write_file(file);
4962         return ret;
4963 }
4964
4965 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4966 {
4967         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4968         struct btrfs_ioctl_quota_rescan_args *qsa;
4969         int ret;
4970
4971         if (!capable(CAP_SYS_ADMIN))
4972                 return -EPERM;
4973
4974         ret = mnt_want_write_file(file);
4975         if (ret)
4976                 return ret;
4977
4978         qsa = memdup_user(arg, sizeof(*qsa));
4979         if (IS_ERR(qsa)) {
4980                 ret = PTR_ERR(qsa);
4981                 goto drop_write;
4982         }
4983
4984         if (qsa->flags) {
4985                 ret = -EINVAL;
4986                 goto out;
4987         }
4988
4989         ret = btrfs_qgroup_rescan(root->fs_info);
4990
4991 out:
4992         kfree(qsa);
4993 drop_write:
4994         mnt_drop_write_file(file);
4995         return ret;
4996 }
4997
4998 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
4999 {
5000         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5001         struct btrfs_ioctl_quota_rescan_args *qsa;
5002         int ret = 0;
5003
5004         if (!capable(CAP_SYS_ADMIN))
5005                 return -EPERM;
5006
5007         qsa = kzalloc(sizeof(*qsa), GFP_KERNEL);
5008         if (!qsa)
5009                 return -ENOMEM;
5010
5011         if (root->fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
5012                 qsa->flags = 1;
5013                 qsa->progress = root->fs_info->qgroup_rescan_progress.objectid;
5014         }
5015
5016         if (copy_to_user(arg, qsa, sizeof(*qsa)))
5017                 ret = -EFAULT;
5018
5019         kfree(qsa);
5020         return ret;
5021 }
5022
5023 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
5024 {
5025         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5026
5027         if (!capable(CAP_SYS_ADMIN))
5028                 return -EPERM;
5029
5030         return btrfs_qgroup_wait_for_completion(root->fs_info);
5031 }
5032
5033 static long _btrfs_ioctl_set_received_subvol(struct file *file,
5034                                             struct btrfs_ioctl_received_subvol_args *sa)
5035 {
5036         struct inode *inode = file_inode(file);
5037         struct btrfs_root *root = BTRFS_I(inode)->root;
5038         struct btrfs_root_item *root_item = &root->root_item;
5039         struct btrfs_trans_handle *trans;
5040         struct timespec ct = current_fs_time(inode->i_sb);
5041         int ret = 0;
5042         int received_uuid_changed;
5043
5044         if (!inode_owner_or_capable(inode))
5045                 return -EPERM;
5046
5047         ret = mnt_want_write_file(file);
5048         if (ret < 0)
5049                 return ret;
5050
5051         down_write(&root->fs_info->subvol_sem);
5052
5053         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
5054                 ret = -EINVAL;
5055                 goto out;
5056         }
5057
5058         if (btrfs_root_readonly(root)) {
5059                 ret = -EROFS;
5060                 goto out;
5061         }
5062
5063         /*
5064          * 1 - root item
5065          * 2 - uuid items (received uuid + subvol uuid)
5066          */
5067         trans = btrfs_start_transaction(root, 3);
5068         if (IS_ERR(trans)) {
5069                 ret = PTR_ERR(trans);
5070                 trans = NULL;
5071                 goto out;
5072         }
5073
5074         sa->rtransid = trans->transid;
5075         sa->rtime.sec = ct.tv_sec;
5076         sa->rtime.nsec = ct.tv_nsec;
5077
5078         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
5079                                        BTRFS_UUID_SIZE);
5080         if (received_uuid_changed &&
5081             !btrfs_is_empty_uuid(root_item->received_uuid))
5082                 btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
5083                                     root_item->received_uuid,
5084                                     BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5085                                     root->root_key.objectid);
5086         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
5087         btrfs_set_root_stransid(root_item, sa->stransid);
5088         btrfs_set_root_rtransid(root_item, sa->rtransid);
5089         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
5090         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
5091         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
5092         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
5093
5094         ret = btrfs_update_root(trans, root->fs_info->tree_root,
5095                                 &root->root_key, &root->root_item);
5096         if (ret < 0) {
5097                 btrfs_end_transaction(trans, root);
5098                 goto out;
5099         }
5100         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
5101                 ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
5102                                           sa->uuid,
5103                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5104                                           root->root_key.objectid);
5105                 if (ret < 0 && ret != -EEXIST) {
5106                         btrfs_abort_transaction(trans, root, ret);
5107                         goto out;
5108                 }
5109         }
5110         ret = btrfs_commit_transaction(trans, root);
5111         if (ret < 0) {
5112                 btrfs_abort_transaction(trans, root, ret);
5113                 goto out;
5114         }
5115
5116 out:
5117         up_write(&root->fs_info->subvol_sem);
5118         mnt_drop_write_file(file);
5119         return ret;
5120 }
5121
5122 #ifdef CONFIG_64BIT
5123 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
5124                                                 void __user *arg)
5125 {
5126         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
5127         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
5128         int ret = 0;
5129
5130         args32 = memdup_user(arg, sizeof(*args32));
5131         if (IS_ERR(args32)) {
5132                 ret = PTR_ERR(args32);
5133                 args32 = NULL;
5134                 goto out;
5135         }
5136
5137         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
5138         if (!args64) {
5139                 ret = -ENOMEM;
5140                 goto out;
5141         }
5142
5143         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
5144         args64->stransid = args32->stransid;
5145         args64->rtransid = args32->rtransid;
5146         args64->stime.sec = args32->stime.sec;
5147         args64->stime.nsec = args32->stime.nsec;
5148         args64->rtime.sec = args32->rtime.sec;
5149         args64->rtime.nsec = args32->rtime.nsec;
5150         args64->flags = args32->flags;
5151
5152         ret = _btrfs_ioctl_set_received_subvol(file, args64);
5153         if (ret)
5154                 goto out;
5155
5156         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
5157         args32->stransid = args64->stransid;
5158         args32->rtransid = args64->rtransid;
5159         args32->stime.sec = args64->stime.sec;
5160         args32->stime.nsec = args64->stime.nsec;
5161         args32->rtime.sec = args64->rtime.sec;
5162         args32->rtime.nsec = args64->rtime.nsec;
5163         args32->flags = args64->flags;
5164
5165         ret = copy_to_user(arg, args32, sizeof(*args32));
5166         if (ret)
5167                 ret = -EFAULT;
5168
5169 out:
5170         kfree(args32);
5171         kfree(args64);
5172         return ret;
5173 }
5174 #endif
5175
5176 static long btrfs_ioctl_set_received_subvol(struct file *file,
5177                                             void __user *arg)
5178 {
5179         struct btrfs_ioctl_received_subvol_args *sa = NULL;
5180         int ret = 0;
5181
5182         sa = memdup_user(arg, sizeof(*sa));
5183         if (IS_ERR(sa)) {
5184                 ret = PTR_ERR(sa);
5185                 sa = NULL;
5186                 goto out;
5187         }
5188
5189         ret = _btrfs_ioctl_set_received_subvol(file, sa);
5190
5191         if (ret)
5192                 goto out;
5193
5194         ret = copy_to_user(arg, sa, sizeof(*sa));
5195         if (ret)
5196                 ret = -EFAULT;
5197
5198 out:
5199         kfree(sa);
5200         return ret;
5201 }
5202
5203 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
5204 {
5205         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5206         size_t len;
5207         int ret;
5208         char label[BTRFS_LABEL_SIZE];
5209
5210         spin_lock(&root->fs_info->super_lock);
5211         memcpy(label, root->fs_info->super_copy->label, BTRFS_LABEL_SIZE);
5212         spin_unlock(&root->fs_info->super_lock);
5213
5214         len = strnlen(label, BTRFS_LABEL_SIZE);
5215
5216         if (len == BTRFS_LABEL_SIZE) {
5217                 btrfs_warn(root->fs_info,
5218                         "label is too long, return the first %zu bytes", --len);
5219         }
5220
5221         ret = copy_to_user(arg, label, len);
5222
5223         return ret ? -EFAULT : 0;
5224 }
5225
5226 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
5227 {
5228         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5229         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5230         struct btrfs_trans_handle *trans;
5231         char label[BTRFS_LABEL_SIZE];
5232         int ret;
5233
5234         if (!capable(CAP_SYS_ADMIN))
5235                 return -EPERM;
5236
5237         if (copy_from_user(label, arg, sizeof(label)))
5238                 return -EFAULT;
5239
5240         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
5241                 btrfs_err(root->fs_info, "unable to set label with more than %d bytes",
5242                        BTRFS_LABEL_SIZE - 1);
5243                 return -EINVAL;
5244         }
5245
5246         ret = mnt_want_write_file(file);
5247         if (ret)
5248                 return ret;
5249
5250         trans = btrfs_start_transaction(root, 0);
5251         if (IS_ERR(trans)) {
5252                 ret = PTR_ERR(trans);
5253                 goto out_unlock;
5254         }
5255
5256         spin_lock(&root->fs_info->super_lock);
5257         strcpy(super_block->label, label);
5258         spin_unlock(&root->fs_info->super_lock);
5259         ret = btrfs_commit_transaction(trans, root);
5260
5261 out_unlock:
5262         mnt_drop_write_file(file);
5263         return ret;
5264 }
5265
5266 #define INIT_FEATURE_FLAGS(suffix) \
5267         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
5268           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
5269           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
5270
5271 int btrfs_ioctl_get_supported_features(void __user *arg)
5272 {
5273         static const struct btrfs_ioctl_feature_flags features[3] = {
5274                 INIT_FEATURE_FLAGS(SUPP),
5275                 INIT_FEATURE_FLAGS(SAFE_SET),
5276                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
5277         };
5278
5279         if (copy_to_user(arg, &features, sizeof(features)))
5280                 return -EFAULT;
5281
5282         return 0;
5283 }
5284
5285 static int btrfs_ioctl_get_features(struct file *file, void __user *arg)
5286 {
5287         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5288         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5289         struct btrfs_ioctl_feature_flags features;
5290
5291         features.compat_flags = btrfs_super_compat_flags(super_block);
5292         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5293         features.incompat_flags = btrfs_super_incompat_flags(super_block);
5294
5295         if (copy_to_user(arg, &features, sizeof(features)))
5296                 return -EFAULT;
5297
5298         return 0;
5299 }
5300
5301 static int check_feature_bits(struct btrfs_root *root,
5302                               enum btrfs_feature_set set,
5303                               u64 change_mask, u64 flags, u64 supported_flags,
5304                               u64 safe_set, u64 safe_clear)
5305 {
5306         const char *type = btrfs_feature_set_names[set];
5307         char *names;
5308         u64 disallowed, unsupported;
5309         u64 set_mask = flags & change_mask;
5310         u64 clear_mask = ~flags & change_mask;
5311
5312         unsupported = set_mask & ~supported_flags;
5313         if (unsupported) {
5314                 names = btrfs_printable_features(set, unsupported);
5315                 if (names) {
5316                         btrfs_warn(root->fs_info,
5317                            "this kernel does not support the %s feature bit%s",
5318                            names, strchr(names, ',') ? "s" : "");
5319                         kfree(names);
5320                 } else
5321                         btrfs_warn(root->fs_info,
5322                            "this kernel does not support %s bits 0x%llx",
5323                            type, unsupported);
5324                 return -EOPNOTSUPP;
5325         }
5326
5327         disallowed = set_mask & ~safe_set;
5328         if (disallowed) {
5329                 names = btrfs_printable_features(set, disallowed);
5330                 if (names) {
5331                         btrfs_warn(root->fs_info,
5332                            "can't set the %s feature bit%s while mounted",
5333                            names, strchr(names, ',') ? "s" : "");
5334                         kfree(names);
5335                 } else
5336                         btrfs_warn(root->fs_info,
5337                            "can't set %s bits 0x%llx while mounted",
5338                            type, disallowed);
5339                 return -EPERM;
5340         }
5341
5342         disallowed = clear_mask & ~safe_clear;
5343         if (disallowed) {
5344                 names = btrfs_printable_features(set, disallowed);
5345                 if (names) {
5346                         btrfs_warn(root->fs_info,
5347                            "can't clear the %s feature bit%s while mounted",
5348                            names, strchr(names, ',') ? "s" : "");
5349                         kfree(names);
5350                 } else
5351                         btrfs_warn(root->fs_info,
5352                            "can't clear %s bits 0x%llx while mounted",
5353                            type, disallowed);
5354                 return -EPERM;
5355         }
5356
5357         return 0;
5358 }
5359
5360 #define check_feature(root, change_mask, flags, mask_base)      \
5361 check_feature_bits(root, FEAT_##mask_base, change_mask, flags,  \
5362                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
5363                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
5364                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5365
5366 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5367 {
5368         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5369         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5370         struct btrfs_ioctl_feature_flags flags[2];
5371         struct btrfs_trans_handle *trans;
5372         u64 newflags;
5373         int ret;
5374
5375         if (!capable(CAP_SYS_ADMIN))
5376                 return -EPERM;
5377
5378         if (copy_from_user(flags, arg, sizeof(flags)))
5379                 return -EFAULT;
5380
5381         /* Nothing to do */
5382         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5383             !flags[0].incompat_flags)
5384                 return 0;
5385
5386         ret = check_feature(root, flags[0].compat_flags,
5387                             flags[1].compat_flags, COMPAT);
5388         if (ret)
5389                 return ret;
5390
5391         ret = check_feature(root, flags[0].compat_ro_flags,
5392                             flags[1].compat_ro_flags, COMPAT_RO);
5393         if (ret)
5394                 return ret;
5395
5396         ret = check_feature(root, flags[0].incompat_flags,
5397                             flags[1].incompat_flags, INCOMPAT);
5398         if (ret)
5399                 return ret;
5400
5401         trans = btrfs_start_transaction(root, 0);
5402         if (IS_ERR(trans))
5403                 return PTR_ERR(trans);
5404
5405         spin_lock(&root->fs_info->super_lock);
5406         newflags = btrfs_super_compat_flags(super_block);
5407         newflags |= flags[0].compat_flags & flags[1].compat_flags;
5408         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5409         btrfs_set_super_compat_flags(super_block, newflags);
5410
5411         newflags = btrfs_super_compat_ro_flags(super_block);
5412         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5413         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5414         btrfs_set_super_compat_ro_flags(super_block, newflags);
5415
5416         newflags = btrfs_super_incompat_flags(super_block);
5417         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5418         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5419         btrfs_set_super_incompat_flags(super_block, newflags);
5420         spin_unlock(&root->fs_info->super_lock);
5421
5422         return btrfs_commit_transaction(trans, root);
5423 }
5424
5425 long btrfs_ioctl(struct file *file, unsigned int
5426                 cmd, unsigned long arg)
5427 {
5428         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5429         void __user *argp = (void __user *)arg;
5430
5431         switch (cmd) {
5432         case FS_IOC_GETFLAGS:
5433                 return btrfs_ioctl_getflags(file, argp);
5434         case FS_IOC_SETFLAGS:
5435                 return btrfs_ioctl_setflags(file, argp);
5436         case FS_IOC_GETVERSION:
5437                 return btrfs_ioctl_getversion(file, argp);
5438         case FITRIM:
5439                 return btrfs_ioctl_fitrim(file, argp);
5440         case BTRFS_IOC_SNAP_CREATE:
5441                 return btrfs_ioctl_snap_create(file, argp, 0);
5442         case BTRFS_IOC_SNAP_CREATE_V2:
5443                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
5444         case BTRFS_IOC_SUBVOL_CREATE:
5445                 return btrfs_ioctl_snap_create(file, argp, 1);
5446         case BTRFS_IOC_SUBVOL_CREATE_V2:
5447                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
5448         case BTRFS_IOC_SNAP_DESTROY:
5449                 return btrfs_ioctl_snap_destroy(file, argp);
5450         case BTRFS_IOC_SUBVOL_GETFLAGS:
5451                 return btrfs_ioctl_subvol_getflags(file, argp);
5452         case BTRFS_IOC_SUBVOL_SETFLAGS:
5453                 return btrfs_ioctl_subvol_setflags(file, argp);
5454         case BTRFS_IOC_DEFAULT_SUBVOL:
5455                 return btrfs_ioctl_default_subvol(file, argp);
5456         case BTRFS_IOC_DEFRAG:
5457                 return btrfs_ioctl_defrag(file, NULL);
5458         case BTRFS_IOC_DEFRAG_RANGE:
5459                 return btrfs_ioctl_defrag(file, argp);
5460         case BTRFS_IOC_RESIZE:
5461                 return btrfs_ioctl_resize(file, argp);
5462         case BTRFS_IOC_ADD_DEV:
5463                 return btrfs_ioctl_add_dev(root, argp);
5464         case BTRFS_IOC_RM_DEV:
5465                 return btrfs_ioctl_rm_dev(file, argp);
5466         case BTRFS_IOC_FS_INFO:
5467                 return btrfs_ioctl_fs_info(root, argp);
5468         case BTRFS_IOC_DEV_INFO:
5469                 return btrfs_ioctl_dev_info(root, argp);
5470         case BTRFS_IOC_BALANCE:
5471                 return btrfs_ioctl_balance(file, NULL);
5472         case BTRFS_IOC_TRANS_START:
5473                 return btrfs_ioctl_trans_start(file);
5474         case BTRFS_IOC_TRANS_END:
5475                 return btrfs_ioctl_trans_end(file);
5476         case BTRFS_IOC_TREE_SEARCH:
5477                 return btrfs_ioctl_tree_search(file, argp);
5478         case BTRFS_IOC_TREE_SEARCH_V2:
5479                 return btrfs_ioctl_tree_search_v2(file, argp);
5480         case BTRFS_IOC_INO_LOOKUP:
5481                 return btrfs_ioctl_ino_lookup(file, argp);
5482         case BTRFS_IOC_INO_PATHS:
5483                 return btrfs_ioctl_ino_to_path(root, argp);
5484         case BTRFS_IOC_LOGICAL_INO:
5485                 return btrfs_ioctl_logical_to_ino(root, argp);
5486         case BTRFS_IOC_SPACE_INFO:
5487                 return btrfs_ioctl_space_info(root, argp);
5488         case BTRFS_IOC_SYNC: {
5489                 int ret;
5490
5491                 ret = btrfs_start_delalloc_roots(root->fs_info, 0, -1);
5492                 if (ret)
5493                         return ret;
5494                 ret = btrfs_sync_fs(file_inode(file)->i_sb, 1);
5495                 /*
5496                  * The transaction thread may want to do more work,
5497                  * namely it pokes the cleaner ktread that will start
5498                  * processing uncleaned subvols.
5499                  */
5500                 wake_up_process(root->fs_info->transaction_kthread);
5501                 return ret;
5502         }
5503         case BTRFS_IOC_START_SYNC:
5504                 return btrfs_ioctl_start_sync(root, argp);
5505         case BTRFS_IOC_WAIT_SYNC:
5506                 return btrfs_ioctl_wait_sync(root, argp);
5507         case BTRFS_IOC_SCRUB:
5508                 return btrfs_ioctl_scrub(file, argp);
5509         case BTRFS_IOC_SCRUB_CANCEL:
5510                 return btrfs_ioctl_scrub_cancel(root, argp);
5511         case BTRFS_IOC_SCRUB_PROGRESS:
5512                 return btrfs_ioctl_scrub_progress(root, argp);
5513         case BTRFS_IOC_BALANCE_V2:
5514                 return btrfs_ioctl_balance(file, argp);
5515         case BTRFS_IOC_BALANCE_CTL:
5516                 return btrfs_ioctl_balance_ctl(root, arg);
5517         case BTRFS_IOC_BALANCE_PROGRESS:
5518                 return btrfs_ioctl_balance_progress(root, argp);
5519         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5520                 return btrfs_ioctl_set_received_subvol(file, argp);
5521 #ifdef CONFIG_64BIT
5522         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5523                 return btrfs_ioctl_set_received_subvol_32(file, argp);
5524 #endif
5525         case BTRFS_IOC_SEND:
5526                 return btrfs_ioctl_send(file, argp);
5527         case BTRFS_IOC_GET_DEV_STATS:
5528                 return btrfs_ioctl_get_dev_stats(root, argp);
5529         case BTRFS_IOC_QUOTA_CTL:
5530                 return btrfs_ioctl_quota_ctl(file, argp);
5531         case BTRFS_IOC_QGROUP_ASSIGN:
5532                 return btrfs_ioctl_qgroup_assign(file, argp);
5533         case BTRFS_IOC_QGROUP_CREATE:
5534                 return btrfs_ioctl_qgroup_create(file, argp);
5535         case BTRFS_IOC_QGROUP_LIMIT:
5536                 return btrfs_ioctl_qgroup_limit(file, argp);
5537         case BTRFS_IOC_QUOTA_RESCAN:
5538                 return btrfs_ioctl_quota_rescan(file, argp);
5539         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5540                 return btrfs_ioctl_quota_rescan_status(file, argp);
5541         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5542                 return btrfs_ioctl_quota_rescan_wait(file, argp);
5543         case BTRFS_IOC_DEV_REPLACE:
5544                 return btrfs_ioctl_dev_replace(root, argp);
5545         case BTRFS_IOC_GET_FSLABEL:
5546                 return btrfs_ioctl_get_fslabel(file, argp);
5547         case BTRFS_IOC_SET_FSLABEL:
5548                 return btrfs_ioctl_set_fslabel(file, argp);
5549         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5550                 return btrfs_ioctl_get_supported_features(argp);
5551         case BTRFS_IOC_GET_FEATURES:
5552                 return btrfs_ioctl_get_features(file, argp);
5553         case BTRFS_IOC_SET_FEATURES:
5554                 return btrfs_ioctl_set_features(file, argp);
5555         }
5556
5557         return -ENOTTY;
5558 }