fs/btrfs/ioctl.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/fsnotify.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/highmem.h>
  27 #include <linux/time.h>
  28 #include <linux/init.h>
  29 #include <linux/string.h>
  30 #include <linux/backing-dev.h>
  31 #include <linux/mount.h>
  32 #include <linux/mpage.h>
  33 #include <linux/namei.h>
  34 #include <linux/swap.h>
  35 #include <linux/writeback.h>
  36 #include <linux/statfs.h>
  37 #include <linux/compat.h>
  38 #include <linux/bit_spinlock.h>
  39 #include <linux/security.h>
  40 #include <linux/xattr.h>
  41 #include <linux/vmalloc.h>
  42 #include <linux/slab.h>
  43 #include <linux/blkdev.h>
  44 #include <linux/uuid.h>
  45 #include <linux/btrfs.h>
  46 #include <linux/uaccess.h>
  47 #include "ctree.h"
  48 #include "disk-io.h"
  49 #include "transaction.h"
  50 #include "btrfs_inode.h"
  51 #include "print-tree.h"
  52 #include "volumes.h"
  53 #include "locking.h"
  54 #include "inode-map.h"
  55 #include "backref.h"
  56 #include "rcu-string.h"
  57 #include "send.h"
  58 #include "dev-replace.h"
  59 #include "props.h"
  60 #include "sysfs.h"
  61 #include "qgroup.h"
  62 #include "tree-log.h"
  63 #include "compression.h"
  64
  65 #ifdef CONFIG_64BIT
  66 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
  67  * structures are incorrect, as the timespec structure from userspace
  68  * is 4 bytes too small. We define these alternatives here to teach
  69  * the kernel about the 32-bit struct packing.
  70  */
  71 struct btrfs_ioctl_timespec_32 {
  72         __u64 sec;
  73         __u32 nsec;
  74 } __attribute__ ((__packed__));
  75
  76 struct btrfs_ioctl_received_subvol_args_32 {
  77         char    uuid[BTRFS_UUID_SIZE];  /* in */
  78         __u64   stransid;               /* in */
  79         __u64   rtransid;               /* out */
  80         struct btrfs_ioctl_timespec_32 stime; /* in */
  81         struct btrfs_ioctl_timespec_32 rtime; /* out */
  82         __u64   flags;                  /* in */
  83         __u64   reserved[16];           /* in */
  84 } __attribute__ ((__packed__));
  85
  86 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
  87                                 struct btrfs_ioctl_received_subvol_args_32)
  88 #endif
  89
  90
  91 static int btrfs_clone(struct inode *src, struct inode *inode,
  92                        u64 off, u64 olen, u64 olen_aligned, u64 destoff,
  93                        int no_time_update);
  94
  95 /* Mask out flags that are inappropriate for the given type of inode. */
  96 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
  97 {
  98         if (S_ISDIR(mode))
  99                 return flags;
 100         else if (S_ISREG(mode))
 101                 return flags & ~FS_DIRSYNC_FL;
 102         else
 103                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
 104 }
 105
 106 /*
 107  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
 108  */
 109 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
 110 {
 111         unsigned int iflags = 0;
 112
 113         if (flags & BTRFS_INODE_SYNC)
 114                 iflags |= FS_SYNC_FL;
 115         if (flags & BTRFS_INODE_IMMUTABLE)
 116                 iflags |= FS_IMMUTABLE_FL;
 117         if (flags & BTRFS_INODE_APPEND)
 118                 iflags |= FS_APPEND_FL;
 119         if (flags & BTRFS_INODE_NODUMP)
 120                 iflags |= FS_NODUMP_FL;
 121         if (flags & BTRFS_INODE_NOATIME)
 122                 iflags |= FS_NOATIME_FL;
 123         if (flags & BTRFS_INODE_DIRSYNC)
 124                 iflags |= FS_DIRSYNC_FL;
 125         if (flags & BTRFS_INODE_NODATACOW)
 126                 iflags |= FS_NOCOW_FL;
 127
 128         if (flags & BTRFS_INODE_NOCOMPRESS)
 129                 iflags |= FS_NOCOMP_FL;
 130         else if (flags & BTRFS_INODE_COMPRESS)
 131                 iflags |= FS_COMPR_FL;
 132
 133         return iflags;
 134 }
 135
 136 /*
 137  * Update inode->i_flags based on the btrfs internal flags.
 138  */
 139 void btrfs_update_iflags(struct inode *inode)
 140 {
 141         struct btrfs_inode *ip = BTRFS_I(inode);
 142         unsigned int new_fl = 0;
 143
 144         if (ip->flags & BTRFS_INODE_SYNC)
 145                 new_fl |= S_SYNC;
 146         if (ip->flags & BTRFS_INODE_IMMUTABLE)
 147                 new_fl |= S_IMMUTABLE;
 148         if (ip->flags & BTRFS_INODE_APPEND)
 149                 new_fl |= S_APPEND;
 150         if (ip->flags & BTRFS_INODE_NOATIME)
 151                 new_fl |= S_NOATIME;
 152         if (ip->flags & BTRFS_INODE_DIRSYNC)
 153                 new_fl |= S_DIRSYNC;
 154
 155         set_mask_bits(&inode->i_flags,
 156                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC,
 157                       new_fl);
 158 }
 159
 160 /*
 161  * Inherit flags from the parent inode.
 162  *
 163  * Currently only the compression flags and the cow flags are inherited.
 164  */
 165 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
 166 {
 167         unsigned int flags;
 168
 169         if (!dir)
 170                 return;
 171
 172         flags = BTRFS_I(dir)->flags;
 173
 174         if (flags & BTRFS_INODE_NOCOMPRESS) {
 175                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
 176                 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 177         } else if (flags & BTRFS_INODE_COMPRESS) {
 178                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
 179                 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
 180         }
 181
 182         if (flags & BTRFS_INODE_NODATACOW) {
 183                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
 184                 if (S_ISREG(inode->i_mode))
 185                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
 186         }
 187
 188         btrfs_update_iflags(inode);
 189 }
 190
 191 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 192 {
 193         struct btrfs_inode *ip = BTRFS_I(file_inode(file));
 194         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
 195
 196         if (copy_to_user(arg, &flags, sizeof(flags)))
 197                 return -EFAULT;
 198         return 0;
 199 }
 200
 201 static int check_flags(unsigned int flags)
 202 {
 203         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 204                       FS_NOATIME_FL | FS_NODUMP_FL | \
 205                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 206                       FS_NOCOMP_FL | FS_COMPR_FL |
 207                       FS_NOCOW_FL))
 208                 return -EOPNOTSUPP;
 209
 210         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 211                 return -EINVAL;
 212
 213         return 0;
 214 }
 215
 216 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 217 {
 218         struct inode *inode = file_inode(file);
 219         struct btrfs_inode *ip = BTRFS_I(inode);
 220         struct btrfs_root *root = ip->root;
 221         struct btrfs_trans_handle *trans;
 222         unsigned int flags, oldflags;
 223         int ret;
 224         u64 ip_oldflags;
 225         unsigned int i_oldflags;
 226         umode_t mode;
 227
 228         if (!inode_owner_or_capable(inode))
 229                 return -EPERM;
 230
 231         if (btrfs_root_readonly(root))
 232                 return -EROFS;
 233
 234         if (copy_from_user(&flags, arg, sizeof(flags)))
 235                 return -EFAULT;
 236
 237         ret = check_flags(flags);
 238         if (ret)
 239                 return ret;
 240
 241         ret = mnt_want_write_file(file);
 242         if (ret)
 243                 return ret;
 244
 245         inode_lock(inode);
 246
 247         ip_oldflags = ip->flags;
 248         i_oldflags = inode->i_flags;
 249         mode = inode->i_mode;
 250
 251         flags = btrfs_mask_flags(inode->i_mode, flags);
 252         oldflags = btrfs_flags_to_ioctl(ip->flags);
 253         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 254                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 255                         ret = -EPERM;
 256                         goto out_unlock;
 257                 }
 258         }
 259
 260         if (flags & FS_SYNC_FL)
 261                 ip->flags |= BTRFS_INODE_SYNC;
 262         else
 263                 ip->flags &= ~BTRFS_INODE_SYNC;
 264         if (flags & FS_IMMUTABLE_FL)
 265                 ip->flags |= BTRFS_INODE_IMMUTABLE;
 266         else
 267                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
 268         if (flags & FS_APPEND_FL)
 269                 ip->flags |= BTRFS_INODE_APPEND;
 270         else
 271                 ip->flags &= ~BTRFS_INODE_APPEND;
 272         if (flags & FS_NODUMP_FL)
 273                 ip->flags |= BTRFS_INODE_NODUMP;
 274         else
 275                 ip->flags &= ~BTRFS_INODE_NODUMP;
 276         if (flags & FS_NOATIME_FL)
 277                 ip->flags |= BTRFS_INODE_NOATIME;
 278         else
 279                 ip->flags &= ~BTRFS_INODE_NOATIME;
 280         if (flags & FS_DIRSYNC_FL)
 281                 ip->flags |= BTRFS_INODE_DIRSYNC;
 282         else
 283                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
 284         if (flags & FS_NOCOW_FL) {
 285                 if (S_ISREG(mode)) {
 286                         /*
 287                          * It's safe to turn csums off here, no extents exist.
 288                          * Otherwise we want the flag to reflect the real COW
 289                          * status of the file and will not set it.
 290                          */
 291                         if (inode->i_size == 0)
 292                                 ip->flags |= BTRFS_INODE_NODATACOW
 293                                            | BTRFS_INODE_NODATASUM;
 294                 } else {
 295                         ip->flags |= BTRFS_INODE_NODATACOW;
 296                 }
 297         } else {
 298                 /*
 299                  * Revert back under same assumptions as above
 300                  */
 301                 if (S_ISREG(mode)) {
 302                         if (inode->i_size == 0)
 303                                 ip->flags &= ~(BTRFS_INODE_NODATACOW
 304                                              | BTRFS_INODE_NODATASUM);
 305                 } else {
 306                         ip->flags &= ~BTRFS_INODE_NODATACOW;
 307                 }
 308         }
 309
 310         /*
 311          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 312          * flag may be changed automatically if compression code won't make
 313          * things smaller.
 314          */
 315         if (flags & FS_NOCOMP_FL) {
 316                 ip->flags &= ~BTRFS_INODE_COMPRESS;
 317                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
 318
 319                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 320                 if (ret && ret != -ENODATA)
 321                         goto out_drop;
 322         } else if (flags & FS_COMPR_FL) {
 323                 const char *comp;
 324
 325                 ip->flags |= BTRFS_INODE_COMPRESS;
 326                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
 327
 328                 if (root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
 329                         comp = "lzo";
 330                 else
 331                         comp = "zlib";
 332                 ret = btrfs_set_prop(inode, "btrfs.compression",
 333                                      comp, strlen(comp), 0);
 334                 if (ret)
 335                         goto out_drop;
 336
 337         } else {
 338                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 339                 if (ret && ret != -ENODATA)
 340                         goto out_drop;
 341                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 342         }
 343
 344         trans = btrfs_start_transaction(root, 1);
 345         if (IS_ERR(trans)) {
 346                 ret = PTR_ERR(trans);
 347                 goto out_drop;
 348         }
 349
 350         btrfs_update_iflags(inode);
 351         inode_inc_iversion(inode);
 352         inode->i_ctime = current_fs_time(inode->i_sb);
 353         ret = btrfs_update_inode(trans, root, inode);
 354
 355         btrfs_end_transaction(trans, root);
 356  out_drop:
 357         if (ret) {
 358                 ip->flags = ip_oldflags;
 359                 inode->i_flags = i_oldflags;
 360         }
 361
 362  out_unlock:
 363         inode_unlock(inode);
 364         mnt_drop_write_file(file);
 365         return ret;
 366 }
 367
 368 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 369 {
 370         struct inode *inode = file_inode(file);
 371
 372         return put_user(inode->i_generation, arg);
 373 }
 374
 375 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
 376 {
 377         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
 378         struct btrfs_device *device;
 379         struct request_queue *q;
 380         struct fstrim_range range;
 381         u64 minlen = ULLONG_MAX;
 382         u64 num_devices = 0;
 383         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
 384         int ret;
 385
 386         if (!capable(CAP_SYS_ADMIN))
 387                 return -EPERM;
 388
 389         rcu_read_lock();
 390         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 391                                 dev_list) {
 392                 if (!device->bdev)
 393                         continue;
 394                 q = bdev_get_queue(device->bdev);
 395                 if (blk_queue_discard(q)) {
 396                         num_devices++;
 397                         minlen = min((u64)q->limits.discard_granularity,
 398                                      minlen);
 399                 }
 400         }
 401         rcu_read_unlock();
 402
 403         if (!num_devices)
 404                 return -EOPNOTSUPP;
 405         if (copy_from_user(&range, arg, sizeof(range)))
 406                 return -EFAULT;
 407         if (range.start > total_bytes ||
 408             range.len < fs_info->sb->s_blocksize)
 409                 return -EINVAL;
 410
 411         range.len = min(range.len, total_bytes - range.start);
 412         range.minlen = max(range.minlen, minlen);
 413         ret = btrfs_trim_fs(fs_info->tree_root, &range);
 414         if (ret < 0)
 415                 return ret;
 416
 417         if (copy_to_user(arg, &range, sizeof(range)))
 418                 return -EFAULT;
 419
 420         return 0;
 421 }
 422
 423 int btrfs_is_empty_uuid(u8 *uuid)
 424 {
 425         int i;
 426
 427         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 428                 if (uuid[i])
 429                         return 0;
 430         }
 431         return 1;
 432 }
 433
 434 static noinline int create_subvol(struct inode *dir,
 435                                   struct dentry *dentry,
 436                                   char *name, int namelen,
 437                                   u64 *async_transid,
 438                                   struct btrfs_qgroup_inherit *inherit)
 439 {
 440         struct btrfs_trans_handle *trans;
 441         struct btrfs_key key;
 442         struct btrfs_root_item *root_item;
 443         struct btrfs_inode_item *inode_item;
 444         struct extent_buffer *leaf;
 445         struct btrfs_root *root = BTRFS_I(dir)->root;
 446         struct btrfs_root *new_root;
 447         struct btrfs_block_rsv block_rsv;
 448         struct timespec cur_time = current_fs_time(dir->i_sb);
 449         struct inode *inode;
 450         int ret;
 451         int err;
 452         u64 objectid;
 453         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 454         u64 index = 0;
 455         u64 qgroup_reserved;
 456         uuid_le new_uuid;
 457
 458         root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
 459         if (!root_item)
 460                 return -ENOMEM;
 461
 462         ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid);
 463         if (ret)
 464                 goto fail_free;
 465
 466         /*
 467          * Don't create subvolume whose level is not zero. Or qgroup will be
 468          * screwed up since it assumes subvolume qgroup's level to be 0.
 469          */
 470         if (btrfs_qgroup_level(objectid)) {
 471                 ret = -ENOSPC;
 472                 goto fail_free;
 473         }
 474
 475         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 476         /*
 477          * The same as the snapshot creation, please see the comment
 478          * of create_snapshot().
 479          */
 480         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
 481                                                8, &qgroup_reserved, false);
 482         if (ret)
 483                 goto fail_free;
 484
 485         trans = btrfs_start_transaction(root, 0);
 486         if (IS_ERR(trans)) {
 487                 ret = PTR_ERR(trans);
 488                 btrfs_subvolume_release_metadata(root, &block_rsv,
 489                                                  qgroup_reserved);
 490                 goto fail_free;
 491         }
 492         trans->block_rsv = &block_rsv;
 493         trans->bytes_reserved = block_rsv.size;
 494
 495         ret = btrfs_qgroup_inherit(trans, root->fs_info, 0, objectid, inherit);
 496         if (ret)
 497                 goto fail;
 498
 499         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
 500         if (IS_ERR(leaf)) {
 501                 ret = PTR_ERR(leaf);
 502                 goto fail;
 503         }
 504
 505         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
 506         btrfs_set_header_bytenr(leaf, leaf->start);
 507         btrfs_set_header_generation(leaf, trans->transid);
 508         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 509         btrfs_set_header_owner(leaf, objectid);
 510
 511         write_extent_buffer(leaf, root->fs_info->fsid, btrfs_header_fsid(),
 512                             BTRFS_FSID_SIZE);
 513         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 514                             btrfs_header_chunk_tree_uuid(leaf),
 515                             BTRFS_UUID_SIZE);
 516         btrfs_mark_buffer_dirty(leaf);
 517
 518         inode_item = &root_item->inode;
 519         btrfs_set_stack_inode_generation(inode_item, 1);
 520         btrfs_set_stack_inode_size(inode_item, 3);
 521         btrfs_set_stack_inode_nlink(inode_item, 1);
 522         btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
 523         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 524
 525         btrfs_set_root_flags(root_item, 0);
 526         btrfs_set_root_limit(root_item, 0);
 527         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 528
 529         btrfs_set_root_bytenr(root_item, leaf->start);
 530         btrfs_set_root_generation(root_item, trans->transid);
 531         btrfs_set_root_level(root_item, 0);
 532         btrfs_set_root_refs(root_item, 1);
 533         btrfs_set_root_used(root_item, leaf->len);
 534         btrfs_set_root_last_snapshot(root_item, 0);
 535
 536         btrfs_set_root_generation_v2(root_item,
 537                         btrfs_root_generation(root_item));
 538         uuid_le_gen(&new_uuid);
 539         memcpy(root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
 540         btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
 541         btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
 542         root_item->ctime = root_item->otime;
 543         btrfs_set_root_ctransid(root_item, trans->transid);
 544         btrfs_set_root_otransid(root_item, trans->transid);
 545
 546         btrfs_tree_unlock(leaf);
 547         free_extent_buffer(leaf);
 548         leaf = NULL;
 549
 550         btrfs_set_root_dirid(root_item, new_dirid);
 551
 552         key.objectid = objectid;
 553         key.offset = 0;
 554         key.type = BTRFS_ROOT_ITEM_KEY;
 555         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 556                                 root_item);
 557         if (ret)
 558                 goto fail;
 559
 560         key.offset = (u64)-1;
 561         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
 562         if (IS_ERR(new_root)) {
 563                 ret = PTR_ERR(new_root);
 564                 btrfs_abort_transaction(trans, root, ret);
 565                 goto fail;
 566         }
 567
 568         btrfs_record_root_in_trans(trans, new_root);
 569
 570         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
 571         if (ret) {
 572                 /* We potentially lose an unused inode item here */
 573                 btrfs_abort_transaction(trans, root, ret);
 574                 goto fail;
 575         }
 576
 577         mutex_lock(&new_root->objectid_mutex);
 578         new_root->highest_objectid = new_dirid;
 579         mutex_unlock(&new_root->objectid_mutex);
 580
 581         /*
 582          * insert the directory item
 583          */
 584         ret = btrfs_set_inode_index(dir, &index);
 585         if (ret) {
 586                 btrfs_abort_transaction(trans, root, ret);
 587                 goto fail;
 588         }
 589
 590         ret = btrfs_insert_dir_item(trans, root,
 591                                     name, namelen, dir, &key,
 592                                     BTRFS_FT_DIR, index);
 593         if (ret) {
 594                 btrfs_abort_transaction(trans, root, ret);
 595                 goto fail;
 596         }
 597
 598         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
 599         ret = btrfs_update_inode(trans, root, dir);
 600         BUG_ON(ret);
 601
 602         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
 603                                  objectid, root->root_key.objectid,
 604                                  btrfs_ino(dir), index, name, namelen);
 605         BUG_ON(ret);
 606
 607         ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
 608                                   root_item->uuid, BTRFS_UUID_KEY_SUBVOL,
 609                                   objectid);
 610         if (ret)
 611                 btrfs_abort_transaction(trans, root, ret);
 612
 613 fail:
 614         kfree(root_item);
 615         trans->block_rsv = NULL;
 616         trans->bytes_reserved = 0;
 617         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
 618
 619         if (async_transid) {
 620                 *async_transid = trans->transid;
 621                 err = btrfs_commit_transaction_async(trans, root, 1);
 622                 if (err)
 623                         err = btrfs_commit_transaction(trans, root);
 624         } else {
 625                 err = btrfs_commit_transaction(trans, root);
 626         }
 627         if (err && !ret)
 628                 ret = err;
 629
 630         if (!ret) {
 631                 inode = btrfs_lookup_dentry(dir, dentry);
 632                 if (IS_ERR(inode))
 633                         return PTR_ERR(inode);
 634                 d_instantiate(dentry, inode);
 635         }
 636         return ret;
 637
 638 fail_free:
 639         kfree(root_item);
 640         return ret;
 641 }
 642
 643 static void btrfs_wait_for_no_snapshoting_writes(struct btrfs_root *root)
 644 {
 645         s64 writers;
 646         DEFINE_WAIT(wait);
 647
 648         do {
 649                 prepare_to_wait(&root->subv_writers->wait, &wait,
 650                                 TASK_UNINTERRUPTIBLE);
 651
 652                 writers = percpu_counter_sum(&root->subv_writers->counter);
 653                 if (writers)
 654                         schedule();
 655
 656                 finish_wait(&root->subv_writers->wait, &wait);
 657         } while (writers);
 658 }
 659
 660 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 661                            struct dentry *dentry, char *name, int namelen,
 662                            u64 *async_transid, bool readonly,
 663                            struct btrfs_qgroup_inherit *inherit)
 664 {
 665         struct inode *inode;
 666         struct btrfs_pending_snapshot *pending_snapshot;
 667         struct btrfs_trans_handle *trans;
 668         int ret;
 669
 670         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 671                 return -EINVAL;
 672
 673         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
 674         if (!pending_snapshot)
 675                 return -ENOMEM;
 676
 677         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 678                         GFP_NOFS);
 679         pending_snapshot->path = btrfs_alloc_path();
 680         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 681                 ret = -ENOMEM;
 682                 goto free_pending;
 683         }
 684
 685         atomic_inc(&root->will_be_snapshoted);
 686         smp_mb__after_atomic();
 687         btrfs_wait_for_no_snapshoting_writes(root);
 688
 689         ret = btrfs_start_delalloc_inodes(root, 0);
 690         if (ret)
 691                 goto dec_and_free;
 692
 693         btrfs_wait_ordered_extents(root, -1, 0, (u64)-1);
 694
 695         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 696                              BTRFS_BLOCK_RSV_TEMP);
 697         /*
 698          * 1 - parent dir inode
 699          * 2 - dir entries
 700          * 1 - root item
 701          * 2 - root ref/backref
 702          * 1 - root of snapshot
 703          * 1 - UUID item
 704          */
 705         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 706                                         &pending_snapshot->block_rsv, 8,
 707                                         &pending_snapshot->qgroup_reserved,
 708                                         false);
 709         if (ret)
 710                 goto dec_and_free;
 711
 712         pending_snapshot->dentry = dentry;
 713         pending_snapshot->root = root;
 714         pending_snapshot->readonly = readonly;
 715         pending_snapshot->dir = dir;
 716         pending_snapshot->inherit = inherit;
 717
 718         trans = btrfs_start_transaction(root, 0);
 719         if (IS_ERR(trans)) {
 720                 ret = PTR_ERR(trans);
 721                 goto fail;
 722         }
 723
 724         spin_lock(&root->fs_info->trans_lock);
 725         list_add(&pending_snapshot->list,
 726                  &trans->transaction->pending_snapshots);
 727         spin_unlock(&root->fs_info->trans_lock);
 728         if (async_transid) {
 729                 *async_transid = trans->transid;
 730                 ret = btrfs_commit_transaction_async(trans,
 731                                      root->fs_info->extent_root, 1);
 732                 if (ret)
 733                         ret = btrfs_commit_transaction(trans, root);
 734         } else {
 735                 ret = btrfs_commit_transaction(trans,
 736                                                root->fs_info->extent_root);
 737         }
 738         if (ret)
 739                 goto fail;
 740
 741         ret = pending_snapshot->error;
 742         if (ret)
 743                 goto fail;
 744
 745         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 746         if (ret)
 747                 goto fail;
 748
 749         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 750         if (IS_ERR(inode)) {
 751                 ret = PTR_ERR(inode);
 752                 goto fail;
 753         }
 754
 755         d_instantiate(dentry, inode);
 756         ret = 0;
 757 fail:
 758         btrfs_subvolume_release_metadata(BTRFS_I(dir)->root,
 759                                          &pending_snapshot->block_rsv,
 760                                          pending_snapshot->qgroup_reserved);
 761 dec_and_free:
 762         if (atomic_dec_and_test(&root->will_be_snapshoted))
 763                 wake_up_atomic_t(&root->will_be_snapshoted);
 764 free_pending:
 765         kfree(pending_snapshot->root_item);
 766         btrfs_free_path(pending_snapshot->path);
 767         kfree(pending_snapshot);
 768
 769         return ret;
 770 }
 771
 772 /*  copy of may_delete in fs/namei.c()
 773  *      Check whether we can remove a link victim from directory dir, check
 774  *  whether the type of victim is right.
 775  *  1. We can't do it if dir is read-only (done in permission())
 776  *  2. We should have write and exec permissions on dir
 777  *  3. We can't remove anything from append-only dir
 778  *  4. We can't do anything with immutable dir (done in permission())
 779  *  5. If the sticky bit on dir is set we should either
 780  *      a. be owner of dir, or
 781  *      b. be owner of victim, or
 782  *      c. have CAP_FOWNER capability
 783  *  6. If the victim is append-only or immutable we can't do anything with
 784  *     links pointing to it.
 785  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 786  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 787  *  9. We can't remove a root or mountpoint.
 788  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 789  *     nfs_async_unlink().
 790  */
 791
 792 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 793 {
 794         int error;
 795
 796         if (d_really_is_negative(victim))
 797                 return -ENOENT;
 798
 799         BUG_ON(d_inode(victim->d_parent) != dir);
 800         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 801
 802         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 803         if (error)
 804                 return error;
 805         if (IS_APPEND(dir))
 806                 return -EPERM;
 807         if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
 808             IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
 809                 return -EPERM;
 810         if (isdir) {
 811                 if (!d_is_dir(victim))
 812                         return -ENOTDIR;
 813                 if (IS_ROOT(victim))
 814                         return -EBUSY;
 815         } else if (d_is_dir(victim))
 816                 return -EISDIR;
 817         if (IS_DEADDIR(dir))
 818                 return -ENOENT;
 819         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 820                 return -EBUSY;
 821         return 0;
 822 }
 823
 824 /* copy of may_create in fs/namei.c() */
 825 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 826 {
 827         if (d_really_is_positive(child))
 828                 return -EEXIST;
 829         if (IS_DEADDIR(dir))
 830                 return -ENOENT;
 831         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 832 }
 833
 834 /*
 835  * Create a new subvolume below @parent.  This is largely modeled after
 836  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 837  * inside this filesystem so it's quite a bit simpler.
 838  */
 839 static noinline int btrfs_mksubvol(struct path *parent,
 840                                    char *name, int namelen,
 841                                    struct btrfs_root *snap_src,
 842                                    u64 *async_transid, bool readonly,
 843                                    struct btrfs_qgroup_inherit *inherit)
 844 {
 845         struct inode *dir  = d_inode(parent->dentry);
 846         struct dentry *dentry;
 847         int error;
 848
 849         inode_lock_nested(dir, I_MUTEX_PARENT);
 850         // XXX: should've been
 851         // mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
 852         // if (error == -EINTR)
 853         //      return error;
 854
 855         dentry = lookup_one_len(name, parent->dentry, namelen);
 856         error = PTR_ERR(dentry);
 857         if (IS_ERR(dentry))
 858                 goto out_unlock;
 859
 860         error = btrfs_may_create(dir, dentry);
 861         if (error)
 862                 goto out_dput;
 863
 864         /*
 865          * even if this name doesn't exist, we may get hash collisions.
 866          * check for them now when we can safely fail
 867          */
 868         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 869                                                dir->i_ino, name,
 870                                                namelen);
 871         if (error)
 872                 goto out_dput;
 873
 874         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 875
 876         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 877                 goto out_up_read;
 878
 879         if (snap_src) {
 880                 error = create_snapshot(snap_src, dir, dentry, name, namelen,
 881                                         async_transid, readonly, inherit);
 882         } else {
 883                 error = create_subvol(dir, dentry, name, namelen,
 884                                       async_transid, inherit);
 885         }
 886         if (!error)
 887                 fsnotify_mkdir(dir, dentry);
 888 out_up_read:
 889         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 890 out_dput:
 891         dput(dentry);
 892 out_unlock:
 893         inode_unlock(dir);
 894         return error;
 895 }
 896
 897 /*
 898  * When we're defragging a range, we don't want to kick it off again
 899  * if it is really just waiting for delalloc to send it down.
 900  * If we find a nice big extent or delalloc range for the bytes in the
 901  * file you want to defrag, we return 0 to let you know to skip this
 902  * part of the file
 903  */
 904 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
 905 {
 906         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 907         struct extent_map *em = NULL;
 908         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 909         u64 end;
 910
 911         read_lock(&em_tree->lock);
 912         em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
 913         read_unlock(&em_tree->lock);
 914
 915         if (em) {
 916                 end = extent_map_end(em);
 917                 free_extent_map(em);
 918                 if (end - offset > thresh)
 919                         return 0;
 920         }
 921         /* if we already have a nice delalloc here, just stop */
 922         thresh /= 2;
 923         end = count_range_bits(io_tree, &offset, offset + thresh,
 924                                thresh, EXTENT_DELALLOC, 1);
 925         if (end >= thresh)
 926                 return 0;
 927         return 1;
 928 }
 929
 930 /*
 931  * helper function to walk through a file and find extents
 932  * newer than a specific transid, and smaller than thresh.
 933  *
 934  * This is used by the defragging code to find new and small
 935  * extents
 936  */
 937 static int find_new_extents(struct btrfs_root *root,
 938                             struct inode *inode, u64 newer_than,
 939                             u64 *off, u32 thresh)
 940 {
 941         struct btrfs_path *path;
 942         struct btrfs_key min_key;
 943         struct extent_buffer *leaf;
 944         struct btrfs_file_extent_item *extent;
 945         int type;
 946         int ret;
 947         u64 ino = btrfs_ino(inode);
 948
 949         path = btrfs_alloc_path();
 950         if (!path)
 951                 return -ENOMEM;
 952
 953         min_key.objectid = ino;
 954         min_key.type = BTRFS_EXTENT_DATA_KEY;
 955         min_key.offset = *off;
 956
 957         while (1) {
 958                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
 959                 if (ret != 0)
 960                         goto none;
 961 process_slot:
 962                 if (min_key.objectid != ino)
 963                         goto none;
 964                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
 965                         goto none;
 966
 967                 leaf = path->nodes[0];
 968                 extent = btrfs_item_ptr(leaf, path->slots[0],
 969                                         struct btrfs_file_extent_item);
 970
 971                 type = btrfs_file_extent_type(leaf, extent);
 972                 if (type == BTRFS_FILE_EXTENT_REG &&
 973                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
 974                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
 975                         *off = min_key.offset;
 976                         btrfs_free_path(path);
 977                         return 0;
 978                 }
 979
 980                 path->slots[0]++;
 981                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
 982                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
 983                         goto process_slot;
 984                 }
 985
 986                 if (min_key.offset == (u64)-1)
 987                         goto none;
 988
 989                 min_key.offset++;
 990                 btrfs_release_path(path);
 991         }
 992 none:
 993         btrfs_free_path(path);
 994         return -ENOENT;
 995 }
 996
 997 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
 998 {
 999         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1000         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1001         struct extent_map *em;
1002         u64 len = PAGE_SIZE;
1003
1004         /*
1005          * hopefully we have this extent in the tree already, try without
1006          * the full extent lock
1007          */
1008         read_lock(&em_tree->lock);
1009         em = lookup_extent_mapping(em_tree, start, len);
1010         read_unlock(&em_tree->lock);
1011
1012         if (!em) {
1013                 struct extent_state *cached = NULL;
1014                 u64 end = start + len - 1;
1015
1016                 /* get the big lock and read metadata off disk */
1017                 lock_extent_bits(io_tree, start, end, &cached);
1018                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
1019                 unlock_extent_cached(io_tree, start, end, &cached, GFP_NOFS);
1020
1021                 if (IS_ERR(em))
1022                         return NULL;
1023         }
1024
1025         return em;
1026 }
1027
1028 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1029 {
1030         struct extent_map *next;
1031         bool ret = true;
1032
1033         /* this is the last extent */
1034         if (em->start + em->len >= i_size_read(inode))
1035                 return false;
1036
1037         next = defrag_lookup_extent(inode, em->start + em->len);
1038         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1039                 ret = false;
1040         else if ((em->block_start + em->block_len == next->block_start) &&
1041                  (em->block_len > SZ_128K && next->block_len > SZ_128K))
1042                 ret = false;
1043
1044         free_extent_map(next);
1045         return ret;
1046 }
1047
1048 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1049                                u64 *last_len, u64 *skip, u64 *defrag_end,
1050                                int compress)
1051 {
1052         struct extent_map *em;
1053         int ret = 1;
1054         bool next_mergeable = true;
1055         bool prev_mergeable = true;
1056
1057         /*
1058          * make sure that once we start defragging an extent, we keep on
1059          * defragging it
1060          */
1061         if (start < *defrag_end)
1062                 return 1;
1063
1064         *skip = 0;
1065
1066         em = defrag_lookup_extent(inode, start);
1067         if (!em)
1068                 return 0;
1069
1070         /* this will cover holes, and inline extents */
1071         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1072                 ret = 0;
1073                 goto out;
1074         }
1075
1076         if (!*defrag_end)
1077                 prev_mergeable = false;
1078
1079         next_mergeable = defrag_check_next_extent(inode, em);
1080         /*
1081          * we hit a real extent, if it is big or the next extent is not a
1082          * real extent, don't bother defragging it
1083          */
1084         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1085             (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1086                 ret = 0;
1087 out:
1088         /*
1089          * last_len ends up being a counter of how many bytes we've defragged.
1090          * every time we choose not to defrag an extent, we reset *last_len
1091          * so that the next tiny extent will force a defrag.
1092          *
1093          * The end result of this is that tiny extents before a single big
1094          * extent will force at least part of that big extent to be defragged.
1095          */
1096         if (ret) {
1097                 *defrag_end = extent_map_end(em);
1098         } else {
1099                 *last_len = 0;
1100                 *skip = extent_map_end(em);
1101                 *defrag_end = 0;
1102         }
1103
1104         free_extent_map(em);
1105         return ret;
1106 }
1107
1108 /*
1109  * it doesn't do much good to defrag one or two pages
1110  * at a time.  This pulls in a nice chunk of pages
1111  * to COW and defrag.
1112  *
1113  * It also makes sure the delalloc code has enough
1114  * dirty data to avoid making new small extents as part
1115  * of the defrag
1116  *
1117  * It's a good idea to start RA on this range
1118  * before calling this.
1119  */
1120 static int cluster_pages_for_defrag(struct inode *inode,
1121                                     struct page **pages,
1122                                     unsigned long start_index,
1123                                     unsigned long num_pages)
1124 {
1125         unsigned long file_end;
1126         u64 isize = i_size_read(inode);
1127         u64 page_start;
1128         u64 page_end;
1129         u64 page_cnt;
1130         int ret;
1131         int i;
1132         int i_done;
1133         struct btrfs_ordered_extent *ordered;
1134         struct extent_state *cached_state = NULL;
1135         struct extent_io_tree *tree;
1136         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1137
1138         file_end = (isize - 1) >> PAGE_SHIFT;
1139         if (!isize || start_index > file_end)
1140                 return 0;
1141
1142         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1143
1144         ret = btrfs_delalloc_reserve_space(inode,
1145                         start_index << PAGE_SHIFT,
1146                         page_cnt << PAGE_SHIFT);
1147         if (ret)
1148                 return ret;
1149         i_done = 0;
1150         tree = &BTRFS_I(inode)->io_tree;
1151
1152         /* step one, lock all the pages */
1153         for (i = 0; i < page_cnt; i++) {
1154                 struct page *page;
1155 again:
1156                 page = find_or_create_page(inode->i_mapping,
1157                                            start_index + i, mask);
1158                 if (!page)
1159                         break;
1160
1161                 page_start = page_offset(page);
1162                 page_end = page_start + PAGE_SIZE - 1;
1163                 while (1) {
1164                         lock_extent_bits(tree, page_start, page_end,
1165                                          &cached_state);
1166                         ordered = btrfs_lookup_ordered_extent(inode,
1167                                                               page_start);
1168                         unlock_extent_cached(tree, page_start, page_end,
1169                                              &cached_state, GFP_NOFS);
1170                         if (!ordered)
1171                                 break;
1172
1173                         unlock_page(page);
1174                         btrfs_start_ordered_extent(inode, ordered, 1);
1175                         btrfs_put_ordered_extent(ordered);
1176                         lock_page(page);
1177                         /*
1178                          * we unlocked the page above, so we need check if
1179                          * it was released or not.
1180                          */
1181                         if (page->mapping != inode->i_mapping) {
1182                                 unlock_page(page);
1183                                 put_page(page);
1184                                 goto again;
1185                         }
1186                 }
1187
1188                 if (!PageUptodate(page)) {
1189                         btrfs_readpage(NULL, page);
1190                         lock_page(page);
1191                         if (!PageUptodate(page)) {
1192                                 unlock_page(page);
1193                                 put_page(page);
1194                                 ret = -EIO;
1195                                 break;
1196                         }
1197                 }
1198
1199                 if (page->mapping != inode->i_mapping) {
1200                         unlock_page(page);
1201                         put_page(page);
1202                         goto again;
1203                 }
1204
1205                 pages[i] = page;
1206                 i_done++;
1207         }
1208         if (!i_done || ret)
1209                 goto out;
1210
1211         if (!(inode->i_sb->s_flags & MS_ACTIVE))
1212                 goto out;
1213
1214         /*
1215          * so now we have a nice long stream of locked
1216          * and up to date pages, lets wait on them
1217          */
1218         for (i = 0; i < i_done; i++)
1219                 wait_on_page_writeback(pages[i]);
1220
1221         page_start = page_offset(pages[0]);
1222         page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1223
1224         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1225                          page_start, page_end - 1, &cached_state);
1226         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1227                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1228                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1229                           &cached_state, GFP_NOFS);
1230
1231         if (i_done != page_cnt) {
1232                 spin_lock(&BTRFS_I(inode)->lock);
1233                 BTRFS_I(inode)->outstanding_extents++;
1234                 spin_unlock(&BTRFS_I(inode)->lock);
1235                 btrfs_delalloc_release_space(inode,
1236                                 start_index << PAGE_SHIFT,
1237                                 (page_cnt - i_done) << PAGE_SHIFT);
1238         }
1239
1240
1241         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1242                           &cached_state);
1243
1244         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1245                              page_start, page_end - 1, &cached_state,
1246                              GFP_NOFS);
1247
1248         for (i = 0; i < i_done; i++) {
1249                 clear_page_dirty_for_io(pages[i]);
1250                 ClearPageChecked(pages[i]);
1251                 set_page_extent_mapped(pages[i]);
1252                 set_page_dirty(pages[i]);
1253                 unlock_page(pages[i]);
1254                 put_page(pages[i]);
1255         }
1256         return i_done;
1257 out:
1258         for (i = 0; i < i_done; i++) {
1259                 unlock_page(pages[i]);
1260                 put_page(pages[i]);
1261         }
1262         btrfs_delalloc_release_space(inode,
1263                         start_index << PAGE_SHIFT,
1264                         page_cnt << PAGE_SHIFT);
1265         return ret;
1266
1267 }
1268
1269 int btrfs_defrag_file(struct inode *inode, struct file *file,
1270                       struct btrfs_ioctl_defrag_range_args *range,
1271                       u64 newer_than, unsigned long max_to_defrag)
1272 {
1273         struct btrfs_root *root = BTRFS_I(inode)->root;
1274         struct file_ra_state *ra = NULL;
1275         unsigned long last_index;
1276         u64 isize = i_size_read(inode);
1277         u64 last_len = 0;
1278         u64 skip = 0;
1279         u64 defrag_end = 0;
1280         u64 newer_off = range->start;
1281         unsigned long i;
1282         unsigned long ra_index = 0;
1283         int ret;
1284         int defrag_count = 0;
1285         int compress_type = BTRFS_COMPRESS_ZLIB;
1286         u32 extent_thresh = range->extent_thresh;
1287         unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1288         unsigned long cluster = max_cluster;
1289         u64 new_align = ~((u64)SZ_128K - 1);
1290         struct page **pages = NULL;
1291
1292         if (isize == 0)
1293                 return 0;
1294
1295         if (range->start >= isize)
1296                 return -EINVAL;
1297
1298         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1299                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1300                         return -EINVAL;
1301                 if (range->compress_type)
1302                         compress_type = range->compress_type;
1303         }
1304
1305         if (extent_thresh == 0)
1306                 extent_thresh = SZ_256K;
1307
1308         /*
1309          * if we were not given a file, allocate a readahead
1310          * context
1311          */
1312         if (!file) {
1313                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1314                 if (!ra)
1315                         return -ENOMEM;
1316                 file_ra_state_init(ra, inode->i_mapping);
1317         } else {
1318                 ra = &file->f_ra;
1319         }
1320
1321         pages = kmalloc_array(max_cluster, sizeof(struct page *),
1322                         GFP_NOFS);
1323         if (!pages) {
1324                 ret = -ENOMEM;
1325                 goto out_ra;
1326         }
1327
1328         /* find the last page to defrag */
1329         if (range->start + range->len > range->start) {
1330                 last_index = min_t(u64, isize - 1,
1331                          range->start + range->len - 1) >> PAGE_SHIFT;
1332         } else {
1333                 last_index = (isize - 1) >> PAGE_SHIFT;
1334         }
1335
1336         if (newer_than) {
1337                 ret = find_new_extents(root, inode, newer_than,
1338                                        &newer_off, SZ_64K);
1339                 if (!ret) {
1340                         range->start = newer_off;
1341                         /*
1342                          * we always align our defrag to help keep
1343                          * the extents in the file evenly spaced
1344                          */
1345                         i = (newer_off & new_align) >> PAGE_SHIFT;
1346                 } else
1347                         goto out_ra;
1348         } else {
1349                 i = range->start >> PAGE_SHIFT;
1350         }
1351         if (!max_to_defrag)
1352                 max_to_defrag = last_index - i + 1;
1353
1354         /*
1355          * make writeback starts from i, so the defrag range can be
1356          * written sequentially.
1357          */
1358         if (i < inode->i_mapping->writeback_index)
1359                 inode->i_mapping->writeback_index = i;
1360
1361         while (i <= last_index && defrag_count < max_to_defrag &&
1362                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1363                 /*
1364                  * make sure we stop running if someone unmounts
1365                  * the FS
1366                  */
1367                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1368                         break;
1369
1370                 if (btrfs_defrag_cancelled(root->fs_info)) {
1371                         btrfs_debug(root->fs_info, "defrag_file cancelled");
1372                         ret = -EAGAIN;
1373                         break;
1374                 }
1375
1376                 if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1377                                          extent_thresh, &last_len, &skip,
1378                                          &defrag_end, range->flags &
1379                                          BTRFS_DEFRAG_RANGE_COMPRESS)) {
1380                         unsigned long next;
1381                         /*
1382                          * the should_defrag function tells us how much to skip
1383                          * bump our counter by the suggested amount
1384                          */
1385                         next = DIV_ROUND_UP(skip, PAGE_SIZE);
1386                         i = max(i + 1, next);
1387                         continue;
1388                 }
1389
1390                 if (!newer_than) {
1391                         cluster = (PAGE_ALIGN(defrag_end) >>
1392                                    PAGE_SHIFT) - i;
1393                         cluster = min(cluster, max_cluster);
1394                 } else {
1395                         cluster = max_cluster;
1396                 }
1397
1398                 if (i + cluster > ra_index) {
1399                         ra_index = max(i, ra_index);
1400                         btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1401                                        cluster);
1402                         ra_index += cluster;
1403                 }
1404
1405                 inode_lock(inode);
1406                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1407                         BTRFS_I(inode)->force_compress = compress_type;
1408                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1409                 if (ret < 0) {
1410                         inode_unlock(inode);
1411                         goto out_ra;
1412                 }
1413
1414                 defrag_count += ret;
1415                 balance_dirty_pages_ratelimited(inode->i_mapping);
1416                 inode_unlock(inode);
1417
1418                 if (newer_than) {
1419                         if (newer_off == (u64)-1)
1420                                 break;
1421
1422                         if (ret > 0)
1423                                 i += ret;
1424
1425                         newer_off = max(newer_off + 1,
1426                                         (u64)i << PAGE_SHIFT);
1427
1428                         ret = find_new_extents(root, inode, newer_than,
1429                                                &newer_off, SZ_64K);
1430                         if (!ret) {
1431                                 range->start = newer_off;
1432                                 i = (newer_off & new_align) >> PAGE_SHIFT;
1433                         } else {
1434                                 break;
1435                         }
1436                 } else {
1437                         if (ret > 0) {
1438                                 i += ret;
1439                                 last_len += ret << PAGE_SHIFT;
1440                         } else {
1441                                 i++;
1442                                 last_len = 0;
1443                         }
1444                 }
1445         }
1446
1447         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1448                 filemap_flush(inode->i_mapping);
1449                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1450                              &BTRFS_I(inode)->runtime_flags))
1451                         filemap_flush(inode->i_mapping);
1452         }
1453
1454         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1455                 /* the filemap_flush will queue IO into the worker threads, but
1456                  * we have to make sure the IO is actually started and that
1457                  * ordered extents get created before we return
1458                  */
1459                 atomic_inc(&root->fs_info->async_submit_draining);
1460                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1461                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1462                         wait_event(root->fs_info->async_submit_wait,
1463                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1464                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1465                 }
1466                 atomic_dec(&root->fs_info->async_submit_draining);
1467         }
1468
1469         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1470                 btrfs_set_fs_incompat(root->fs_info, COMPRESS_LZO);
1471         }
1472
1473         ret = defrag_count;
1474
1475 out_ra:
1476         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1477                 inode_lock(inode);
1478                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1479                 inode_unlock(inode);
1480         }
1481         if (!file)
1482                 kfree(ra);
1483         kfree(pages);
1484         return ret;
1485 }
1486
1487 static noinline int btrfs_ioctl_resize(struct file *file,
1488                                         void __user *arg)
1489 {
1490         u64 new_size;
1491         u64 old_size;
1492         u64 devid = 1;
1493         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
1494         struct btrfs_ioctl_vol_args *vol_args;
1495         struct btrfs_trans_handle *trans;
1496         struct btrfs_device *device = NULL;
1497         char *sizestr;
1498         char *retptr;
1499         char *devstr = NULL;
1500         int ret = 0;
1501         int mod = 0;
1502
1503         if (!capable(CAP_SYS_ADMIN))
1504                 return -EPERM;
1505
1506         ret = mnt_want_write_file(file);
1507         if (ret)
1508                 return ret;
1509
1510         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
1511                         1)) {
1512                 mnt_drop_write_file(file);
1513                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1514         }
1515
1516         mutex_lock(&root->fs_info->volume_mutex);
1517         vol_args = memdup_user(arg, sizeof(*vol_args));
1518         if (IS_ERR(vol_args)) {
1519                 ret = PTR_ERR(vol_args);
1520                 goto out;
1521         }
1522
1523         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1524
1525         sizestr = vol_args->name;
1526         devstr = strchr(sizestr, ':');
1527         if (devstr) {
1528                 sizestr = devstr + 1;
1529                 *devstr = '\0';
1530                 devstr = vol_args->name;
1531                 ret = kstrtoull(devstr, 10, &devid);
1532                 if (ret)
1533                         goto out_free;
1534                 if (!devid) {
1535                         ret = -EINVAL;
1536                         goto out_free;
1537                 }
1538                 btrfs_info(root->fs_info, "resizing devid %llu", devid);
1539         }
1540
1541         device = btrfs_find_device(root->fs_info, devid, NULL, NULL);
1542         if (!device) {
1543                 btrfs_info(root->fs_info, "resizer unable to find device %llu",
1544                        devid);
1545                 ret = -ENODEV;
1546                 goto out_free;
1547         }
1548
1549         if (!device->writeable) {
1550                 btrfs_info(root->fs_info,
1551                            "resizer unable to apply on readonly device %llu",
1552                        devid);
1553                 ret = -EPERM;
1554                 goto out_free;
1555         }
1556
1557         if (!strcmp(sizestr, "max"))
1558                 new_size = device->bdev->bd_inode->i_size;
1559         else {
1560                 if (sizestr[0] == '-') {
1561                         mod = -1;
1562                         sizestr++;
1563                 } else if (sizestr[0] == '+') {
1564                         mod = 1;
1565                         sizestr++;
1566                 }
1567                 new_size = memparse(sizestr, &retptr);
1568                 if (*retptr != '\0' || new_size == 0) {
1569                         ret = -EINVAL;
1570                         goto out_free;
1571                 }
1572         }
1573
1574         if (device->is_tgtdev_for_dev_replace) {
1575                 ret = -EPERM;
1576                 goto out_free;
1577         }
1578
1579         old_size = btrfs_device_get_total_bytes(device);
1580
1581         if (mod < 0) {
1582                 if (new_size > old_size) {
1583                         ret = -EINVAL;
1584                         goto out_free;
1585                 }
1586                 new_size = old_size - new_size;
1587         } else if (mod > 0) {
1588                 if (new_size > ULLONG_MAX - old_size) {
1589                         ret = -ERANGE;
1590                         goto out_free;
1591                 }
1592                 new_size = old_size + new_size;
1593         }
1594
1595         if (new_size < SZ_256M) {
1596                 ret = -EINVAL;
1597                 goto out_free;
1598         }
1599         if (new_size > device->bdev->bd_inode->i_size) {
1600                 ret = -EFBIG;
1601                 goto out_free;
1602         }
1603
1604         new_size = div_u64(new_size, root->sectorsize);
1605         new_size *= root->sectorsize;
1606
1607         btrfs_info_in_rcu(root->fs_info, "new size for %s is %llu",
1608                       rcu_str_deref(device->name), new_size);
1609
1610         if (new_size > old_size) {
1611                 trans = btrfs_start_transaction(root, 0);
1612                 if (IS_ERR(trans)) {
1613                         ret = PTR_ERR(trans);
1614                         goto out_free;
1615                 }
1616                 ret = btrfs_grow_device(trans, device, new_size);
1617                 btrfs_commit_transaction(trans, root);
1618         } else if (new_size < old_size) {
1619                 ret = btrfs_shrink_device(device, new_size);
1620         } /* equal, nothing need to do */
1621
1622 out_free:
1623         kfree(vol_args);
1624 out:
1625         mutex_unlock(&root->fs_info->volume_mutex);
1626         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
1627         mnt_drop_write_file(file);
1628         return ret;
1629 }
1630
1631 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1632                                 char *name, unsigned long fd, int subvol,
1633                                 u64 *transid, bool readonly,
1634                                 struct btrfs_qgroup_inherit *inherit)
1635 {
1636         int namelen;
1637         int ret = 0;
1638
1639         ret = mnt_want_write_file(file);
1640         if (ret)
1641                 goto out;
1642
1643         namelen = strlen(name);
1644         if (strchr(name, '/')) {
1645                 ret = -EINVAL;
1646                 goto out_drop_write;
1647         }
1648
1649         if (name[0] == '.' &&
1650            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1651                 ret = -EEXIST;
1652                 goto out_drop_write;
1653         }
1654
1655         if (subvol) {
1656                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1657                                      NULL, transid, readonly, inherit);
1658         } else {
1659                 struct fd src = fdget(fd);
1660                 struct inode *src_inode;
1661                 if (!src.file) {
1662                         ret = -EINVAL;
1663                         goto out_drop_write;
1664                 }
1665
1666                 src_inode = file_inode(src.file);
1667                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1668                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1669                                    "Snapshot src from another FS");
1670                         ret = -EXDEV;
1671                 } else if (!inode_owner_or_capable(src_inode)) {
1672                         /*
1673                          * Subvolume creation is not restricted, but snapshots
1674                          * are limited to own subvolumes only
1675                          */
1676                         ret = -EPERM;
1677                 } else {
1678                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1679                                              BTRFS_I(src_inode)->root,
1680                                              transid, readonly, inherit);
1681                 }
1682                 fdput(src);
1683         }
1684 out_drop_write:
1685         mnt_drop_write_file(file);
1686 out:
1687         return ret;
1688 }
1689
1690 static noinline int btrfs_ioctl_snap_create(struct file *file,
1691                                             void __user *arg, int subvol)
1692 {
1693         struct btrfs_ioctl_vol_args *vol_args;
1694         int ret;
1695
1696         vol_args = memdup_user(arg, sizeof(*vol_args));
1697         if (IS_ERR(vol_args))
1698                 return PTR_ERR(vol_args);
1699         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1700
1701         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1702                                               vol_args->fd, subvol,
1703                                               NULL, false, NULL);
1704
1705         kfree(vol_args);
1706         return ret;
1707 }
1708
1709 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1710                                                void __user *arg, int subvol)
1711 {
1712         struct btrfs_ioctl_vol_args_v2 *vol_args;
1713         int ret;
1714         u64 transid = 0;
1715         u64 *ptr = NULL;
1716         bool readonly = false;
1717         struct btrfs_qgroup_inherit *inherit = NULL;
1718
1719         vol_args = memdup_user(arg, sizeof(*vol_args));
1720         if (IS_ERR(vol_args))
1721                 return PTR_ERR(vol_args);
1722         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1723
1724         if (vol_args->flags &
1725             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1726               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1727                 ret = -EOPNOTSUPP;
1728                 goto free_args;
1729         }
1730
1731         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1732                 ptr = &transid;
1733         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1734                 readonly = true;
1735         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1736                 if (vol_args->size > PAGE_SIZE) {
1737                         ret = -EINVAL;
1738                         goto free_args;
1739                 }
1740                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1741                 if (IS_ERR(inherit)) {
1742                         ret = PTR_ERR(inherit);
1743                         goto free_args;
1744                 }
1745         }
1746
1747         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1748                                               vol_args->fd, subvol, ptr,
1749                                               readonly, inherit);
1750         if (ret)
1751                 goto free_inherit;
1752
1753         if (ptr && copy_to_user(arg +
1754                                 offsetof(struct btrfs_ioctl_vol_args_v2,
1755                                         transid),
1756                                 ptr, sizeof(*ptr)))
1757                 ret = -EFAULT;
1758
1759 free_inherit:
1760         kfree(inherit);
1761 free_args:
1762         kfree(vol_args);
1763         return ret;
1764 }
1765
1766 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1767                                                 void __user *arg)
1768 {
1769         struct inode *inode = file_inode(file);
1770         struct btrfs_root *root = BTRFS_I(inode)->root;
1771         int ret = 0;
1772         u64 flags = 0;
1773
1774         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1775                 return -EINVAL;
1776
1777         down_read(&root->fs_info->subvol_sem);
1778         if (btrfs_root_readonly(root))
1779                 flags |= BTRFS_SUBVOL_RDONLY;
1780         up_read(&root->fs_info->subvol_sem);
1781
1782         if (copy_to_user(arg, &flags, sizeof(flags)))
1783                 ret = -EFAULT;
1784
1785         return ret;
1786 }
1787
1788 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1789                                               void __user *arg)
1790 {
1791         struct inode *inode = file_inode(file);
1792         struct btrfs_root *root = BTRFS_I(inode)->root;
1793         struct btrfs_trans_handle *trans;
1794         u64 root_flags;
1795         u64 flags;
1796         int ret = 0;
1797
1798         if (!inode_owner_or_capable(inode))
1799                 return -EPERM;
1800
1801         ret = mnt_want_write_file(file);
1802         if (ret)
1803                 goto out;
1804
1805         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1806                 ret = -EINVAL;
1807                 goto out_drop_write;
1808         }
1809
1810         if (copy_from_user(&flags, arg, sizeof(flags))) {
1811                 ret = -EFAULT;
1812                 goto out_drop_write;
1813         }
1814
1815         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1816                 ret = -EINVAL;
1817                 goto out_drop_write;
1818         }
1819
1820         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1821                 ret = -EOPNOTSUPP;
1822                 goto out_drop_write;
1823         }
1824
1825         down_write(&root->fs_info->subvol_sem);
1826
1827         /* nothing to do */
1828         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1829                 goto out_drop_sem;
1830
1831         root_flags = btrfs_root_flags(&root->root_item);
1832         if (flags & BTRFS_SUBVOL_RDONLY) {
1833                 btrfs_set_root_flags(&root->root_item,
1834                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1835         } else {
1836                 /*
1837                  * Block RO -> RW transition if this subvolume is involved in
1838                  * send
1839                  */
1840                 spin_lock(&root->root_item_lock);
1841                 if (root->send_in_progress == 0) {
1842                         btrfs_set_root_flags(&root->root_item,
1843                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1844                         spin_unlock(&root->root_item_lock);
1845                 } else {
1846                         spin_unlock(&root->root_item_lock);
1847                         btrfs_warn(root->fs_info,
1848                         "Attempt to set subvolume %llu read-write during send",
1849                                         root->root_key.objectid);
1850                         ret = -EPERM;
1851                         goto out_drop_sem;
1852                 }
1853         }
1854
1855         trans = btrfs_start_transaction(root, 1);
1856         if (IS_ERR(trans)) {
1857                 ret = PTR_ERR(trans);
1858                 goto out_reset;
1859         }
1860
1861         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1862                                 &root->root_key, &root->root_item);
1863
1864         btrfs_commit_transaction(trans, root);
1865 out_reset:
1866         if (ret)
1867                 btrfs_set_root_flags(&root->root_item, root_flags);
1868 out_drop_sem:
1869         up_write(&root->fs_info->subvol_sem);
1870 out_drop_write:
1871         mnt_drop_write_file(file);
1872 out:
1873         return ret;
1874 }
1875
1876 /*
1877  * helper to check if the subvolume references other subvolumes
1878  */
1879 static noinline int may_destroy_subvol(struct btrfs_root *root)
1880 {
1881         struct btrfs_path *path;
1882         struct btrfs_dir_item *di;
1883         struct btrfs_key key;
1884         u64 dir_id;
1885         int ret;
1886
1887         path = btrfs_alloc_path();
1888         if (!path)
1889                 return -ENOMEM;
1890
1891         /* Make sure this root isn't set as the default subvol */
1892         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
1893         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root, path,
1894                                    dir_id, "default", 7, 0);
1895         if (di && !IS_ERR(di)) {
1896                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1897                 if (key.objectid == root->root_key.objectid) {
1898                         ret = -EPERM;
1899                         btrfs_err(root->fs_info, "deleting default subvolume "
1900                                   "%llu is not allowed", key.objectid);
1901                         goto out;
1902                 }
1903                 btrfs_release_path(path);
1904         }
1905
1906         key.objectid = root->root_key.objectid;
1907         key.type = BTRFS_ROOT_REF_KEY;
1908         key.offset = (u64)-1;
1909
1910         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1911                                 &key, path, 0, 0);
1912         if (ret < 0)
1913                 goto out;
1914         BUG_ON(ret == 0);
1915
1916         ret = 0;
1917         if (path->slots[0] > 0) {
1918                 path->slots[0]--;
1919                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1920                 if (key.objectid == root->root_key.objectid &&
1921                     key.type == BTRFS_ROOT_REF_KEY)
1922                         ret = -ENOTEMPTY;
1923         }
1924 out:
1925         btrfs_free_path(path);
1926         return ret;
1927 }
1928
1929 static noinline int key_in_sk(struct btrfs_key *key,
1930                               struct btrfs_ioctl_search_key *sk)
1931 {
1932         struct btrfs_key test;
1933         int ret;
1934
1935         test.objectid = sk->min_objectid;
1936         test.type = sk->min_type;
1937         test.offset = sk->min_offset;
1938
1939         ret = btrfs_comp_cpu_keys(key, &test);
1940         if (ret < 0)
1941                 return 0;
1942
1943         test.objectid = sk->max_objectid;
1944         test.type = sk->max_type;
1945         test.offset = sk->max_offset;
1946
1947         ret = btrfs_comp_cpu_keys(key, &test);
1948         if (ret > 0)
1949                 return 0;
1950         return 1;
1951 }
1952
1953 static noinline int copy_to_sk(struct btrfs_root *root,
1954                                struct btrfs_path *path,
1955                                struct btrfs_key *key,
1956                                struct btrfs_ioctl_search_key *sk,
1957                                size_t *buf_size,
1958                                char __user *ubuf,
1959                                unsigned long *sk_offset,
1960                                int *num_found)
1961 {
1962         u64 found_transid;
1963         struct extent_buffer *leaf;
1964         struct btrfs_ioctl_search_header sh;
1965         struct btrfs_key test;
1966         unsigned long item_off;
1967         unsigned long item_len;
1968         int nritems;
1969         int i;
1970         int slot;
1971         int ret = 0;
1972
1973         leaf = path->nodes[0];
1974         slot = path->slots[0];
1975         nritems = btrfs_header_nritems(leaf);
1976
1977         if (btrfs_header_generation(leaf) > sk->max_transid) {
1978                 i = nritems;
1979                 goto advance_key;
1980         }
1981         found_transid = btrfs_header_generation(leaf);
1982
1983         for (i = slot; i < nritems; i++) {
1984                 item_off = btrfs_item_ptr_offset(leaf, i);
1985                 item_len = btrfs_item_size_nr(leaf, i);
1986
1987                 btrfs_item_key_to_cpu(leaf, key, i);
1988                 if (!key_in_sk(key, sk))
1989                         continue;
1990
1991                 if (sizeof(sh) + item_len > *buf_size) {
1992                         if (*num_found) {
1993                                 ret = 1;
1994                                 goto out;
1995                         }
1996
1997                         /*
1998                          * return one empty item back for v1, which does not
1999                          * handle -EOVERFLOW
2000                          */
2001
2002                         *buf_size = sizeof(sh) + item_len;
2003                         item_len = 0;
2004                         ret = -EOVERFLOW;
2005                 }
2006
2007                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
2008                         ret = 1;
2009                         goto out;
2010                 }
2011
2012                 sh.objectid = key->objectid;
2013                 sh.offset = key->offset;
2014                 sh.type = key->type;
2015                 sh.len = item_len;
2016                 sh.transid = found_transid;
2017
2018                 /* copy search result header */
2019                 if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
2020                         ret = -EFAULT;
2021                         goto out;
2022                 }
2023
2024                 *sk_offset += sizeof(sh);
2025
2026                 if (item_len) {
2027                         char __user *up = ubuf + *sk_offset;
2028                         /* copy the item */
2029                         if (read_extent_buffer_to_user(leaf, up,
2030                                                        item_off, item_len)) {
2031                                 ret = -EFAULT;
2032                                 goto out;
2033                         }
2034
2035                         *sk_offset += item_len;
2036                 }
2037                 (*num_found)++;
2038
2039                 if (ret) /* -EOVERFLOW from above */
2040                         goto out;
2041
2042                 if (*num_found >= sk->nr_items) {
2043                         ret = 1;
2044                         goto out;
2045                 }
2046         }
2047 advance_key:
2048         ret = 0;
2049         test.objectid = sk->max_objectid;
2050         test.type = sk->max_type;
2051         test.offset = sk->max_offset;
2052         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2053                 ret = 1;
2054         else if (key->offset < (u64)-1)
2055                 key->offset++;
2056         else if (key->type < (u8)-1) {
2057                 key->offset = 0;
2058                 key->type++;
2059         } else if (key->objectid < (u64)-1) {
2060                 key->offset = 0;
2061                 key->type = 0;
2062                 key->objectid++;
2063         } else
2064                 ret = 1;
2065 out:
2066         /*
2067          *  0: all items from this leaf copied, continue with next
2068          *  1: * more items can be copied, but unused buffer is too small
2069          *     * all items were found
2070          *     Either way, it will stops the loop which iterates to the next
2071          *     leaf
2072          *  -EOVERFLOW: item was to large for buffer
2073          *  -EFAULT: could not copy extent buffer back to userspace
2074          */
2075         return ret;
2076 }
2077
2078 static noinline int search_ioctl(struct inode *inode,
2079                                  struct btrfs_ioctl_search_key *sk,
2080                                  size_t *buf_size,
2081                                  char __user *ubuf)
2082 {
2083         struct btrfs_root *root;
2084         struct btrfs_key key;
2085         struct btrfs_path *path;
2086         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
2087         int ret;
2088         int num_found = 0;
2089         unsigned long sk_offset = 0;
2090
2091         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2092                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2093                 return -EOVERFLOW;
2094         }
2095
2096         path = btrfs_alloc_path();
2097         if (!path)
2098                 return -ENOMEM;
2099
2100         if (sk->tree_id == 0) {
2101                 /* search the root of the inode that was passed */
2102                 root = BTRFS_I(inode)->root;
2103         } else {
2104                 key.objectid = sk->tree_id;
2105                 key.type = BTRFS_ROOT_ITEM_KEY;
2106                 key.offset = (u64)-1;
2107                 root = btrfs_read_fs_root_no_name(info, &key);
2108                 if (IS_ERR(root)) {
2109                         btrfs_free_path(path);
2110                         return -ENOENT;
2111                 }
2112         }
2113
2114         key.objectid = sk->min_objectid;
2115         key.type = sk->min_type;
2116         key.offset = sk->min_offset;
2117
2118         while (1) {
2119                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2120                 if (ret != 0) {
2121                         if (ret > 0)
2122                                 ret = 0;
2123                         goto err;
2124                 }
2125                 ret = copy_to_sk(root, path, &key, sk, buf_size, ubuf,
2126                                  &sk_offset, &num_found);
2127                 btrfs_release_path(path);
2128                 if (ret)
2129                         break;
2130
2131         }
2132         if (ret > 0)
2133                 ret = 0;
2134 err:
2135         sk->nr_items = num_found;
2136         btrfs_free_path(path);
2137         return ret;
2138 }
2139
2140 static noinline int btrfs_ioctl_tree_search(struct file *file,
2141                                            void __user *argp)
2142 {
2143         struct btrfs_ioctl_search_args __user *uargs;
2144         struct btrfs_ioctl_search_key sk;
2145         struct inode *inode;
2146         int ret;
2147         size_t buf_size;
2148
2149         if (!capable(CAP_SYS_ADMIN))
2150                 return -EPERM;
2151
2152         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2153
2154         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2155                 return -EFAULT;
2156
2157         buf_size = sizeof(uargs->buf);
2158
2159         inode = file_inode(file);
2160         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2161
2162         /*
2163          * In the origin implementation an overflow is handled by returning a
2164          * search header with a len of zero, so reset ret.
2165          */
2166         if (ret == -EOVERFLOW)
2167                 ret = 0;
2168
2169         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2170                 ret = -EFAULT;
2171         return ret;
2172 }
2173
2174 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2175                                                void __user *argp)
2176 {
2177         struct btrfs_ioctl_search_args_v2 __user *uarg;
2178         struct btrfs_ioctl_search_args_v2 args;
2179         struct inode *inode;
2180         int ret;
2181         size_t buf_size;
2182         const size_t buf_limit = SZ_16M;
2183
2184         if (!capable(CAP_SYS_ADMIN))
2185                 return -EPERM;
2186
2187         /* copy search header and buffer size */
2188         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2189         if (copy_from_user(&args, uarg, sizeof(args)))
2190                 return -EFAULT;
2191
2192         buf_size = args.buf_size;
2193
2194         if (buf_size < sizeof(struct btrfs_ioctl_search_header))
2195                 return -EOVERFLOW;
2196
2197         /* limit result size to 16MB */
2198         if (buf_size > buf_limit)
2199                 buf_size = buf_limit;
2200
2201         inode = file_inode(file);
2202         ret = search_ioctl(inode, &args.key, &buf_size,
2203                            (char *)(&uarg->buf[0]));
2204         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2205                 ret = -EFAULT;
2206         else if (ret == -EOVERFLOW &&
2207                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2208                 ret = -EFAULT;
2209
2210         return ret;
2211 }
2212
2213 /*
2214  * Search INODE_REFs to identify path name of 'dirid' directory
2215  * in a 'tree_id' tree. and sets path name to 'name'.
2216  */
2217 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2218                                 u64 tree_id, u64 dirid, char *name)
2219 {
2220         struct btrfs_root *root;
2221         struct btrfs_key key;
2222         char *ptr;
2223         int ret = -1;
2224         int slot;
2225         int len;
2226         int total_len = 0;
2227         struct btrfs_inode_ref *iref;
2228         struct extent_buffer *l;
2229         struct btrfs_path *path;
2230
2231         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2232                 name[0]='\0';
2233                 return 0;
2234         }
2235
2236         path = btrfs_alloc_path();
2237         if (!path)
2238                 return -ENOMEM;
2239
2240         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
2241
2242         key.objectid = tree_id;
2243         key.type = BTRFS_ROOT_ITEM_KEY;
2244         key.offset = (u64)-1;
2245         root = btrfs_read_fs_root_no_name(info, &key);
2246         if (IS_ERR(root)) {
2247                 btrfs_err(info, "could not find root %llu", tree_id);
2248                 ret = -ENOENT;
2249                 goto out;
2250         }
2251
2252         key.objectid = dirid;
2253         key.type = BTRFS_INODE_REF_KEY;
2254         key.offset = (u64)-1;
2255
2256         while (1) {
2257                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2258                 if (ret < 0)
2259                         goto out;
2260                 else if (ret > 0) {
2261                         ret = btrfs_previous_item(root, path, dirid,
2262                                                   BTRFS_INODE_REF_KEY);
2263                         if (ret < 0)
2264                                 goto out;
2265                         else if (ret > 0) {
2266                                 ret = -ENOENT;
2267                                 goto out;
2268                         }
2269                 }
2270
2271                 l = path->nodes[0];
2272                 slot = path->slots[0];
2273                 btrfs_item_key_to_cpu(l, &key, slot);
2274
2275                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2276                 len = btrfs_inode_ref_name_len(l, iref);
2277                 ptr -= len + 1;
2278                 total_len += len + 1;
2279                 if (ptr < name) {
2280                         ret = -ENAMETOOLONG;
2281                         goto out;
2282                 }
2283
2284                 *(ptr + len) = '/';
2285                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2286
2287                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2288                         break;
2289
2290                 btrfs_release_path(path);
2291                 key.objectid = key.offset;
2292                 key.offset = (u64)-1;
2293                 dirid = key.objectid;
2294         }
2295         memmove(name, ptr, total_len);
2296         name[total_len] = '\0';
2297         ret = 0;
2298 out:
2299         btrfs_free_path(path);
2300         return ret;
2301 }
2302
2303 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2304                                            void __user *argp)
2305 {
2306          struct btrfs_ioctl_ino_lookup_args *args;
2307          struct inode *inode;
2308         int ret = 0;
2309
2310         args = memdup_user(argp, sizeof(*args));
2311         if (IS_ERR(args))
2312                 return PTR_ERR(args);
2313
2314         inode = file_inode(file);
2315
2316         /*
2317          * Unprivileged query to obtain the containing subvolume root id. The
2318          * path is reset so it's consistent with btrfs_search_path_in_tree.
2319          */
2320         if (args->treeid == 0)
2321                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2322
2323         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2324                 args->name[0] = 0;
2325                 goto out;
2326         }
2327
2328         if (!capable(CAP_SYS_ADMIN)) {
2329                 ret = -EPERM;
2330                 goto out;
2331         }
2332
2333         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2334                                         args->treeid, args->objectid,
2335                                         args->name);
2336
2337 out:
2338         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2339                 ret = -EFAULT;
2340
2341         kfree(args);
2342         return ret;
2343 }
2344
2345 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2346                                              void __user *arg)
2347 {
2348         struct dentry *parent = file->f_path.dentry;
2349         struct dentry *dentry;
2350         struct inode *dir = d_inode(parent);
2351         struct inode *inode;
2352         struct btrfs_root *root = BTRFS_I(dir)->root;
2353         struct btrfs_root *dest = NULL;
2354         struct btrfs_ioctl_vol_args *vol_args;
2355         struct btrfs_trans_handle *trans;
2356         struct btrfs_block_rsv block_rsv;
2357         u64 root_flags;
2358         u64 qgroup_reserved;
2359         int namelen;
2360         int ret;
2361         int err = 0;
2362
2363         vol_args = memdup_user(arg, sizeof(*vol_args));
2364         if (IS_ERR(vol_args))
2365                 return PTR_ERR(vol_args);
2366
2367         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2368         namelen = strlen(vol_args->name);
2369         if (strchr(vol_args->name, '/') ||
2370             strncmp(vol_args->name, "..", namelen) == 0) {
2371                 err = -EINVAL;
2372                 goto out;
2373         }
2374
2375         err = mnt_want_write_file(file);
2376         if (err)
2377                 goto out;
2378
2379
2380         inode_lock_nested(dir, I_MUTEX_PARENT);
2381         // XXX: should've been
2382         // err = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
2383         // if (err == -EINTR)
2384         //      goto out_drop_write;
2385         dentry = lookup_one_len(vol_args->name, parent, namelen);
2386         if (IS_ERR(dentry)) {
2387                 err = PTR_ERR(dentry);
2388                 goto out_unlock_dir;
2389         }
2390
2391         if (d_really_is_negative(dentry)) {
2392                 err = -ENOENT;
2393                 goto out_dput;
2394         }
2395
2396         inode = d_inode(dentry);
2397         dest = BTRFS_I(inode)->root;
2398         if (!capable(CAP_SYS_ADMIN)) {
2399                 /*
2400                  * Regular user.  Only allow this with a special mount
2401                  * option, when the user has write+exec access to the
2402                  * subvol root, and when rmdir(2) would have been
2403                  * allowed.
2404                  *
2405                  * Note that this is _not_ check that the subvol is
2406                  * empty or doesn't contain data that we wouldn't
2407                  * otherwise be able to delete.
2408                  *
2409                  * Users who want to delete empty subvols should try
2410                  * rmdir(2).
2411                  */
2412                 err = -EPERM;
2413                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
2414                         goto out_dput;
2415
2416                 /*
2417                  * Do not allow deletion if the parent dir is the same
2418                  * as the dir to be deleted.  That means the ioctl
2419                  * must be called on the dentry referencing the root
2420                  * of the subvol, not a random directory contained
2421                  * within it.
2422                  */
2423                 err = -EINVAL;
2424                 if (root == dest)
2425                         goto out_dput;
2426
2427                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2428                 if (err)
2429                         goto out_dput;
2430         }
2431
2432         /* check if subvolume may be deleted by a user */
2433         err = btrfs_may_delete(dir, dentry, 1);
2434         if (err)
2435                 goto out_dput;
2436
2437         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
2438                 err = -EINVAL;
2439                 goto out_dput;
2440         }
2441
2442         inode_lock(inode);
2443
2444         /*
2445          * Don't allow to delete a subvolume with send in progress. This is
2446          * inside the i_mutex so the error handling that has to drop the bit
2447          * again is not run concurrently.
2448          */
2449         spin_lock(&dest->root_item_lock);
2450         root_flags = btrfs_root_flags(&dest->root_item);
2451         if (dest->send_in_progress == 0) {
2452                 btrfs_set_root_flags(&dest->root_item,
2453                                 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
2454                 spin_unlock(&dest->root_item_lock);
2455         } else {
2456                 spin_unlock(&dest->root_item_lock);
2457                 btrfs_warn(root->fs_info,
2458                         "Attempt to delete subvolume %llu during send",
2459                         dest->root_key.objectid);
2460                 err = -EPERM;
2461                 goto out_unlock_inode;
2462         }
2463
2464         down_write(&root->fs_info->subvol_sem);
2465
2466         err = may_destroy_subvol(dest);
2467         if (err)
2468                 goto out_up_write;
2469
2470         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2471         /*
2472          * One for dir inode, two for dir entries, two for root
2473          * ref/backref.
2474          */
2475         err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2476                                                5, &qgroup_reserved, true);
2477         if (err)
2478                 goto out_up_write;
2479
2480         trans = btrfs_start_transaction(root, 0);
2481         if (IS_ERR(trans)) {
2482                 err = PTR_ERR(trans);
2483                 goto out_release;
2484         }
2485         trans->block_rsv = &block_rsv;
2486         trans->bytes_reserved = block_rsv.size;
2487
2488         btrfs_record_snapshot_destroy(trans, dir);
2489
2490         ret = btrfs_unlink_subvol(trans, root, dir,
2491                                 dest->root_key.objectid,
2492                                 dentry->d_name.name,
2493                                 dentry->d_name.len);
2494         if (ret) {
2495                 err = ret;
2496                 btrfs_abort_transaction(trans, root, ret);
2497                 goto out_end_trans;
2498         }
2499
2500         btrfs_record_root_in_trans(trans, dest);
2501
2502         memset(&dest->root_item.drop_progress, 0,
2503                 sizeof(dest->root_item.drop_progress));
2504         dest->root_item.drop_level = 0;
2505         btrfs_set_root_refs(&dest->root_item, 0);
2506
2507         if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
2508                 ret = btrfs_insert_orphan_item(trans,
2509                                         root->fs_info->tree_root,
2510                                         dest->root_key.objectid);
2511                 if (ret) {
2512                         btrfs_abort_transaction(trans, root, ret);
2513                         err = ret;
2514                         goto out_end_trans;
2515                 }
2516         }
2517
2518         ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2519                                   dest->root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
2520                                   dest->root_key.objectid);
2521         if (ret && ret != -ENOENT) {
2522                 btrfs_abort_transaction(trans, root, ret);
2523                 err = ret;
2524                 goto out_end_trans;
2525         }
2526         if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2527                 ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2528                                           dest->root_item.received_uuid,
2529                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2530                                           dest->root_key.objectid);
2531                 if (ret && ret != -ENOENT) {
2532                         btrfs_abort_transaction(trans, root, ret);
2533                         err = ret;
2534                         goto out_end_trans;
2535                 }
2536         }
2537
2538 out_end_trans:
2539         trans->block_rsv = NULL;
2540         trans->bytes_reserved = 0;
2541         ret = btrfs_end_transaction(trans, root);
2542         if (ret && !err)
2543                 err = ret;
2544         inode->i_flags |= S_DEAD;
2545 out_release:
2546         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
2547 out_up_write:
2548         up_write(&root->fs_info->subvol_sem);
2549         if (err) {
2550                 spin_lock(&dest->root_item_lock);
2551                 root_flags = btrfs_root_flags(&dest->root_item);
2552                 btrfs_set_root_flags(&dest->root_item,
2553                                 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
2554                 spin_unlock(&dest->root_item_lock);
2555         }
2556 out_unlock_inode:
2557         inode_unlock(inode);
2558         if (!err) {
2559                 d_invalidate(dentry);
2560                 btrfs_invalidate_inodes(dest);
2561                 d_delete(dentry);
2562                 ASSERT(dest->send_in_progress == 0);
2563
2564                 /* the last ref */
2565                 if (dest->ino_cache_inode) {
2566                         iput(dest->ino_cache_inode);
2567                         dest->ino_cache_inode = NULL;
2568                 }
2569         }
2570 out_dput:
2571         dput(dentry);
2572 out_unlock_dir:
2573         inode_unlock(dir);
2574 //out_drop_write:
2575         mnt_drop_write_file(file);
2576 out:
2577         kfree(vol_args);
2578         return err;
2579 }
2580
2581 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2582 {
2583         struct inode *inode = file_inode(file);
2584         struct btrfs_root *root = BTRFS_I(inode)->root;
2585         struct btrfs_ioctl_defrag_range_args *range;
2586         int ret;
2587
2588         ret = mnt_want_write_file(file);
2589         if (ret)
2590                 return ret;
2591
2592         if (btrfs_root_readonly(root)) {
2593                 ret = -EROFS;
2594                 goto out;
2595         }
2596
2597         switch (inode->i_mode & S_IFMT) {
2598         case S_IFDIR:
2599                 if (!capable(CAP_SYS_ADMIN)) {
2600                         ret = -EPERM;
2601                         goto out;
2602                 }
2603                 ret = btrfs_defrag_root(root);
2604                 if (ret)
2605                         goto out;
2606                 ret = btrfs_defrag_root(root->fs_info->extent_root);
2607                 break;
2608         case S_IFREG:
2609                 if (!(file->f_mode & FMODE_WRITE)) {
2610                         ret = -EINVAL;
2611                         goto out;
2612                 }
2613
2614                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2615                 if (!range) {
2616                         ret = -ENOMEM;
2617                         goto out;
2618                 }
2619
2620                 if (argp) {
2621                         if (copy_from_user(range, argp,
2622                                            sizeof(*range))) {
2623                                 ret = -EFAULT;
2624                                 kfree(range);
2625                                 goto out;
2626                         }
2627                         /* compression requires us to start the IO */
2628                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2629                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2630                                 range->extent_thresh = (u32)-1;
2631                         }
2632                 } else {
2633                         /* the rest are all set to zero by kzalloc */
2634                         range->len = (u64)-1;
2635                 }
2636                 ret = btrfs_defrag_file(file_inode(file), file,
2637                                         range, 0, 0);
2638                 if (ret > 0)
2639                         ret = 0;
2640                 kfree(range);
2641                 break;
2642         default:
2643                 ret = -EINVAL;
2644         }
2645 out:
2646         mnt_drop_write_file(file);
2647         return ret;
2648 }
2649
2650 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2651 {
2652         struct btrfs_ioctl_vol_args *vol_args;
2653         int ret;
2654
2655         if (!capable(CAP_SYS_ADMIN))
2656                 return -EPERM;
2657
2658         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2659                         1)) {
2660                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2661         }
2662
2663         mutex_lock(&root->fs_info->volume_mutex);
2664         vol_args = memdup_user(arg, sizeof(*vol_args));
2665         if (IS_ERR(vol_args)) {
2666                 ret = PTR_ERR(vol_args);
2667                 goto out;
2668         }
2669
2670         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2671         ret = btrfs_init_new_device(root, vol_args->name);
2672
2673         if (!ret)
2674                 btrfs_info(root->fs_info, "disk added %s",vol_args->name);
2675
2676         kfree(vol_args);
2677 out:
2678         mutex_unlock(&root->fs_info->volume_mutex);
2679         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2680         return ret;
2681 }
2682
2683 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
2684 {
2685         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
2686         struct btrfs_ioctl_vol_args_v2 *vol_args;
2687         int ret;
2688
2689         if (!capable(CAP_SYS_ADMIN))
2690                 return -EPERM;
2691
2692         ret = mnt_want_write_file(file);
2693         if (ret)
2694                 return ret;
2695
2696         vol_args = memdup_user(arg, sizeof(*vol_args));
2697         if (IS_ERR(vol_args)) {
2698                 ret = PTR_ERR(vol_args);
2699                 goto err_drop;
2700         }
2701
2702         /* Check for compatibility reject unknown flags */
2703         if (vol_args->flags & ~BTRFS_VOL_ARG_V2_FLAGS_SUPPORTED)
2704                 return -EOPNOTSUPP;
2705
2706         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2707                         1)) {
2708                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2709                 goto out;
2710         }
2711
2712         mutex_lock(&root->fs_info->volume_mutex);
2713         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
2714                 ret = btrfs_rm_device(root, NULL, vol_args->devid);
2715         } else {
2716                 vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
2717                 ret = btrfs_rm_device(root, vol_args->name, 0);
2718         }
2719         mutex_unlock(&root->fs_info->volume_mutex);
2720         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2721
2722         if (!ret) {
2723                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
2724                         btrfs_info(root->fs_info, "device deleted: id %llu",
2725                                         vol_args->devid);
2726                 else
2727                         btrfs_info(root->fs_info, "device deleted: %s",
2728                                         vol_args->name);
2729         }
2730 out:
2731         kfree(vol_args);
2732 err_drop:
2733         mnt_drop_write_file(file);
2734         return ret;
2735 }
2736
2737 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2738 {
2739         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
2740         struct btrfs_ioctl_vol_args *vol_args;
2741         int ret;
2742
2743         if (!capable(CAP_SYS_ADMIN))
2744                 return -EPERM;
2745
2746         ret = mnt_want_write_file(file);
2747         if (ret)
2748                 return ret;
2749
2750         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2751                         1)) {
2752                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2753                 goto out_drop_write;
2754         }
2755
2756         vol_args = memdup_user(arg, sizeof(*vol_args));
2757         if (IS_ERR(vol_args)) {
2758                 ret = PTR_ERR(vol_args);
2759                 goto out;
2760         }
2761
2762         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2763         mutex_lock(&root->fs_info->volume_mutex);
2764         ret = btrfs_rm_device(root, vol_args->name, 0);
2765         mutex_unlock(&root->fs_info->volume_mutex);
2766
2767         if (!ret)
2768                 btrfs_info(root->fs_info, "disk deleted %s",vol_args->name);
2769         kfree(vol_args);
2770 out:
2771         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2772 out_drop_write:
2773         mnt_drop_write_file(file);
2774
2775         return ret;
2776 }
2777
2778 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2779 {
2780         struct btrfs_ioctl_fs_info_args *fi_args;
2781         struct btrfs_device *device;
2782         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2783         int ret = 0;
2784
2785         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2786         if (!fi_args)
2787                 return -ENOMEM;
2788
2789         mutex_lock(&fs_devices->device_list_mutex);
2790         fi_args->num_devices = fs_devices->num_devices;
2791         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2792
2793         list_for_each_entry(device, &fs_devices->devices, dev_list) {
2794                 if (device->devid > fi_args->max_id)
2795                         fi_args->max_id = device->devid;
2796         }
2797         mutex_unlock(&fs_devices->device_list_mutex);
2798
2799         fi_args->nodesize = root->fs_info->super_copy->nodesize;
2800         fi_args->sectorsize = root->fs_info->super_copy->sectorsize;
2801         fi_args->clone_alignment = root->fs_info->super_copy->sectorsize;
2802
2803         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2804                 ret = -EFAULT;
2805
2806         kfree(fi_args);
2807         return ret;
2808 }
2809
2810 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2811 {
2812         struct btrfs_ioctl_dev_info_args *di_args;
2813         struct btrfs_device *dev;
2814         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2815         int ret = 0;
2816         char *s_uuid = NULL;
2817
2818         di_args = memdup_user(arg, sizeof(*di_args));
2819         if (IS_ERR(di_args))
2820                 return PTR_ERR(di_args);
2821
2822         if (!btrfs_is_empty_uuid(di_args->uuid))
2823                 s_uuid = di_args->uuid;
2824
2825         mutex_lock(&fs_devices->device_list_mutex);
2826         dev = btrfs_find_device(root->fs_info, di_args->devid, s_uuid, NULL);
2827
2828         if (!dev) {
2829                 ret = -ENODEV;
2830                 goto out;
2831         }
2832
2833         di_args->devid = dev->devid;
2834         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
2835         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
2836         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2837         if (dev->name) {
2838                 struct rcu_string *name;
2839
2840                 rcu_read_lock();
2841                 name = rcu_dereference(dev->name);
2842                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2843                 rcu_read_unlock();
2844                 di_args->path[sizeof(di_args->path) - 1] = 0;
2845         } else {
2846                 di_args->path[0] = '\0';
2847         }
2848
2849 out:
2850         mutex_unlock(&fs_devices->device_list_mutex);
2851         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2852                 ret = -EFAULT;
2853
2854         kfree(di_args);
2855         return ret;
2856 }
2857
2858 static struct page *extent_same_get_page(struct inode *inode, pgoff_t index)
2859 {
2860         struct page *page;
2861
2862         page = grab_cache_page(inode->i_mapping, index);
2863         if (!page)
2864                 return ERR_PTR(-ENOMEM);
2865
2866         if (!PageUptodate(page)) {
2867                 int ret;
2868
2869                 ret = btrfs_readpage(NULL, page);
2870                 if (ret)
2871                         return ERR_PTR(ret);
2872                 lock_page(page);
2873                 if (!PageUptodate(page)) {
2874                         unlock_page(page);
2875                         put_page(page);
2876                         return ERR_PTR(-EIO);
2877                 }
2878                 if (page->mapping != inode->i_mapping) {
2879                         unlock_page(page);
2880                         put_page(page);
2881                         return ERR_PTR(-EAGAIN);
2882                 }
2883         }
2884
2885         return page;
2886 }
2887
2888 static int gather_extent_pages(struct inode *inode, struct page **pages,
2889                                int num_pages, u64 off)
2890 {
2891         int i;
2892         pgoff_t index = off >> PAGE_SHIFT;
2893
2894         for (i = 0; i < num_pages; i++) {
2895 again:
2896                 pages[i] = extent_same_get_page(inode, index + i);
2897                 if (IS_ERR(pages[i])) {
2898                         int err = PTR_ERR(pages[i]);
2899
2900                         if (err == -EAGAIN)
2901                                 goto again;
2902                         pages[i] = NULL;
2903                         return err;
2904                 }
2905         }
2906         return 0;
2907 }
2908
2909 static int lock_extent_range(struct inode *inode, u64 off, u64 len,
2910                              bool retry_range_locking)
2911 {
2912         /*
2913          * Do any pending delalloc/csum calculations on inode, one way or
2914          * another, and lock file content.
2915          * The locking order is:
2916          *
2917          *   1) pages
2918          *   2) range in the inode's io tree
2919          */
2920         while (1) {
2921                 struct btrfs_ordered_extent *ordered;
2922                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2923                 ordered = btrfs_lookup_first_ordered_extent(inode,
2924                                                             off + len - 1);
2925                 if ((!ordered ||
2926                      ordered->file_offset + ordered->len <= off ||
2927                      ordered->file_offset >= off + len) &&
2928                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2929                                     off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
2930                         if (ordered)
2931                                 btrfs_put_ordered_extent(ordered);
2932                         break;
2933                 }
2934                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2935                 if (ordered)
2936                         btrfs_put_ordered_extent(ordered);
2937                 if (!retry_range_locking)
2938                         return -EAGAIN;
2939                 btrfs_wait_ordered_range(inode, off, len);
2940         }
2941         return 0;
2942 }
2943
2944 static void btrfs_double_inode_unlock(struct inode *inode1, struct inode *inode2)
2945 {
2946         inode_unlock(inode1);
2947         inode_unlock(inode2);
2948 }
2949
2950 static void btrfs_double_inode_lock(struct inode *inode1, struct inode *inode2)
2951 {
2952         if (inode1 < inode2)
2953                 swap(inode1, inode2);
2954
2955         inode_lock_nested(inode1, I_MUTEX_PARENT);
2956         inode_lock_nested(inode2, I_MUTEX_CHILD);
2957 }
2958
2959 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
2960                                       struct inode *inode2, u64 loff2, u64 len)
2961 {
2962         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2963         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2964 }
2965
2966 static int btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
2967                                     struct inode *inode2, u64 loff2, u64 len,
2968                                     bool retry_range_locking)
2969 {
2970         int ret;
2971
2972         if (inode1 < inode2) {
2973                 swap(inode1, inode2);
2974                 swap(loff1, loff2);
2975         }
2976         ret = lock_extent_range(inode1, loff1, len, retry_range_locking);
2977         if (ret)
2978                 return ret;
2979         ret = lock_extent_range(inode2, loff2, len, retry_range_locking);
2980         if (ret)
2981                 unlock_extent(&BTRFS_I(inode1)->io_tree, loff1,
2982                               loff1 + len - 1);
2983         return ret;
2984 }
2985
2986 struct cmp_pages {
2987         int             num_pages;
2988         struct page     **src_pages;
2989         struct page     **dst_pages;
2990 };
2991
2992 static void btrfs_cmp_data_free(struct cmp_pages *cmp)
2993 {
2994         int i;
2995         struct page *pg;
2996
2997         for (i = 0; i < cmp->num_pages; i++) {
2998                 pg = cmp->src_pages[i];
2999                 if (pg) {
3000                         unlock_page(pg);
3001                         put_page(pg);
3002                 }
3003                 pg = cmp->dst_pages[i];
3004                 if (pg) {
3005                         unlock_page(pg);
3006                         put_page(pg);
3007                 }
3008         }
3009         kfree(cmp->src_pages);
3010         kfree(cmp->dst_pages);
3011 }
3012
3013 static int btrfs_cmp_data_prepare(struct inode *src, u64 loff,
3014                                   struct inode *dst, u64 dst_loff,
3015                                   u64 len, struct cmp_pages *cmp)
3016 {
3017         int ret;
3018         int num_pages = PAGE_ALIGN(len) >> PAGE_SHIFT;
3019         struct page **src_pgarr, **dst_pgarr;
3020
3021         /*
3022          * We must gather up all the pages before we initiate our
3023          * extent locking. We use an array for the page pointers. Size
3024          * of the array is bounded by len, which is in turn bounded by
3025          * BTRFS_MAX_DEDUPE_LEN.
3026          */
3027         src_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
3028         dst_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
3029         if (!src_pgarr || !dst_pgarr) {
3030                 kfree(src_pgarr);
3031                 kfree(dst_pgarr);
3032                 return -ENOMEM;
3033         }
3034         cmp->num_pages = num_pages;
3035         cmp->src_pages = src_pgarr;
3036         cmp->dst_pages = dst_pgarr;
3037
3038         ret = gather_extent_pages(src, cmp->src_pages, cmp->num_pages, loff);
3039         if (ret)
3040                 goto out;
3041
3042         ret = gather_extent_pages(dst, cmp->dst_pages, cmp->num_pages, dst_loff);
3043
3044 out:
3045         if (ret)
3046                 btrfs_cmp_data_free(cmp);
3047         return 0;
3048 }
3049
3050 static int btrfs_cmp_data(struct inode *src, u64 loff, struct inode *dst,
3051                           u64 dst_loff, u64 len, struct cmp_pages *cmp)
3052 {
3053         int ret = 0;
3054         int i;
3055         struct page *src_page, *dst_page;
3056         unsigned int cmp_len = PAGE_SIZE;
3057         void *addr, *dst_addr;
3058
3059         i = 0;
3060         while (len) {
3061                 if (len < PAGE_SIZE)
3062                         cmp_len = len;
3063
3064                 BUG_ON(i >= cmp->num_pages);
3065
3066                 src_page = cmp->src_pages[i];
3067                 dst_page = cmp->dst_pages[i];
3068                 ASSERT(PageLocked(src_page));
3069                 ASSERT(PageLocked(dst_page));
3070
3071                 addr = kmap_atomic(src_page);
3072                 dst_addr = kmap_atomic(dst_page);
3073
3074                 flush_dcache_page(src_page);
3075                 flush_dcache_page(dst_page);
3076
3077                 if (memcmp(addr, dst_addr, cmp_len))
3078                         ret = -EBADE;
3079
3080                 kunmap_atomic(addr);
3081                 kunmap_atomic(dst_addr);
3082
3083                 if (ret)
3084                         break;
3085
3086                 len -= cmp_len;
3087                 i++;
3088         }
3089
3090         return ret;
3091 }
3092
3093 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 *plen,
3094                                      u64 olen)
3095 {
3096         u64 len = *plen;
3097         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
3098
3099         if (off + olen > inode->i_size || off + olen < off)
3100                 return -EINVAL;
3101
3102         /* if we extend to eof, continue to block boundary */
3103         if (off + len == inode->i_size)
3104                 *plen = len = ALIGN(inode->i_size, bs) - off;
3105
3106         /* Check that we are block aligned - btrfs_clone() requires this */
3107         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
3108                 return -EINVAL;
3109
3110         return 0;
3111 }
3112
3113 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
3114                              struct inode *dst, u64 dst_loff)
3115 {
3116         int ret;
3117         u64 len = olen;
3118         struct cmp_pages cmp;
3119         int same_inode = 0;
3120         u64 same_lock_start = 0;
3121         u64 same_lock_len = 0;
3122
3123         if (src == dst)
3124                 same_inode = 1;
3125
3126         if (len == 0)
3127                 return 0;
3128
3129         if (same_inode) {
3130                 inode_lock(src);
3131
3132                 ret = extent_same_check_offsets(src, loff, &len, olen);
3133                 if (ret)
3134                         goto out_unlock;
3135                 ret = extent_same_check_offsets(src, dst_loff, &len, olen);
3136                 if (ret)
3137                         goto out_unlock;
3138
3139                 /*
3140                  * Single inode case wants the same checks, except we
3141                  * don't want our length pushed out past i_size as
3142                  * comparing that data range makes no sense.
3143                  *
3144                  * extent_same_check_offsets() will do this for an
3145                  * unaligned length at i_size, so catch it here and
3146                  * reject the request.
3147                  *
3148                  * This effectively means we require aligned extents
3149                  * for the single-inode case, whereas the other cases
3150                  * allow an unaligned length so long as it ends at
3151                  * i_size.
3152                  */
3153                 if (len != olen) {
3154                         ret = -EINVAL;
3155                         goto out_unlock;
3156                 }
3157
3158                 /* Check for overlapping ranges */
3159                 if (dst_loff + len > loff && dst_loff < loff + len) {
3160                         ret = -EINVAL;
3161                         goto out_unlock;
3162                 }
3163
3164                 same_lock_start = min_t(u64, loff, dst_loff);
3165                 same_lock_len = max_t(u64, loff, dst_loff) + len - same_lock_start;
3166         } else {
3167                 btrfs_double_inode_lock(src, dst);
3168
3169                 ret = extent_same_check_offsets(src, loff, &len, olen);
3170                 if (ret)
3171                         goto out_unlock;
3172
3173                 ret = extent_same_check_offsets(dst, dst_loff, &len, olen);
3174                 if (ret)
3175                         goto out_unlock;
3176         }
3177
3178         /* don't make the dst file partly checksummed */
3179         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3180             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
3181                 ret = -EINVAL;
3182                 goto out_unlock;
3183         }
3184
3185 again:
3186         ret = btrfs_cmp_data_prepare(src, loff, dst, dst_loff, olen, &cmp);
3187         if (ret)
3188                 goto out_unlock;
3189
3190         if (same_inode)
3191                 ret = lock_extent_range(src, same_lock_start, same_lock_len,
3192                                         false);
3193         else
3194                 ret = btrfs_double_extent_lock(src, loff, dst, dst_loff, len,
3195                                                false);
3196         /*
3197          * If one of the inodes has dirty pages in the respective range or
3198          * ordered extents, we need to flush dellaloc and wait for all ordered
3199          * extents in the range. We must unlock the pages and the ranges in the
3200          * io trees to avoid deadlocks when flushing delalloc (requires locking
3201          * pages) and when waiting for ordered extents to complete (they require
3202          * range locking).
3203          */
3204         if (ret == -EAGAIN) {
3205                 /*
3206                  * Ranges in the io trees already unlocked. Now unlock all
3207                  * pages before waiting for all IO to complete.
3208                  */
3209                 btrfs_cmp_data_free(&cmp);
3210                 if (same_inode) {
3211                         btrfs_wait_ordered_range(src, same_lock_start,
3212                                                  same_lock_len);
3213                 } else {
3214                         btrfs_wait_ordered_range(src, loff, len);
3215                         btrfs_wait_ordered_range(dst, dst_loff, len);
3216                 }
3217                 goto again;
3218         }
3219         ASSERT(ret == 0);
3220         if (WARN_ON(ret)) {
3221                 /* ranges in the io trees already unlocked */
3222                 btrfs_cmp_data_free(&cmp);
3223                 return ret;
3224         }
3225
3226         /* pass original length for comparison so we stay within i_size */
3227         ret = btrfs_cmp_data(src, loff, dst, dst_loff, olen, &cmp);
3228         if (ret == 0)
3229                 ret = btrfs_clone(src, dst, loff, olen, len, dst_loff, 1);
3230
3231         if (same_inode)
3232                 unlock_extent(&BTRFS_I(src)->io_tree, same_lock_start,
3233                               same_lock_start + same_lock_len - 1);
3234         else
3235                 btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
3236
3237         btrfs_cmp_data_free(&cmp);
3238 out_unlock:
3239         if (same_inode)
3240                 inode_unlock(src);
3241         else
3242                 btrfs_double_inode_unlock(src, dst);
3243
3244         return ret;
3245 }
3246
3247 #define BTRFS_MAX_DEDUPE_LEN    SZ_16M
3248
3249 ssize_t btrfs_dedupe_file_range(struct file *src_file, u64 loff, u64 olen,
3250                                 struct file *dst_file, u64 dst_loff)
3251 {
3252         struct inode *src = file_inode(src_file);
3253         struct inode *dst = file_inode(dst_file);
3254         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
3255         ssize_t res;
3256
3257         if (olen > BTRFS_MAX_DEDUPE_LEN)
3258                 olen = BTRFS_MAX_DEDUPE_LEN;
3259
3260         if (WARN_ON_ONCE(bs < PAGE_SIZE)) {
3261                 /*
3262                  * Btrfs does not support blocksize < page_size. As a
3263                  * result, btrfs_cmp_data() won't correctly handle
3264                  * this situation without an update.
3265                  */
3266                 return -EINVAL;
3267         }
3268
3269         res = btrfs_extent_same(src, loff, olen, dst, dst_loff);
3270         if (res)
3271                 return res;
3272         return olen;
3273 }
3274
3275 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3276                                      struct inode *inode,
3277                                      u64 endoff,
3278                                      const u64 destoff,
3279                                      const u64 olen,
3280                                      int no_time_update)
3281 {
3282         struct btrfs_root *root = BTRFS_I(inode)->root;
3283         int ret;
3284
3285         inode_inc_iversion(inode);
3286         if (!no_time_update)
3287                 inode->i_mtime = inode->i_ctime = current_fs_time(inode->i_sb);
3288         /*
3289          * We round up to the block size at eof when determining which
3290          * extents to clone above, but shouldn't round up the file size.
3291          */
3292         if (endoff > destoff + olen)
3293                 endoff = destoff + olen;
3294         if (endoff > inode->i_size)
3295                 btrfs_i_size_write(inode, endoff);
3296
3297         ret = btrfs_update_inode(trans, root, inode);
3298         if (ret) {
3299                 btrfs_abort_transaction(trans, root, ret);
3300                 btrfs_end_transaction(trans, root);
3301                 goto out;
3302         }
3303         ret = btrfs_end_transaction(trans, root);
3304 out:
3305         return ret;
3306 }
3307
3308 static void clone_update_extent_map(struct inode *inode,
3309                                     const struct btrfs_trans_handle *trans,
3310                                     const struct btrfs_path *path,
3311                                     const u64 hole_offset,
3312                                     const u64 hole_len)
3313 {
3314         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3315         struct extent_map *em;
3316         int ret;
3317
3318         em = alloc_extent_map();
3319         if (!em) {
3320                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3321                         &BTRFS_I(inode)->runtime_flags);
3322                 return;
3323         }
3324
3325         if (path) {
3326                 struct btrfs_file_extent_item *fi;
3327
3328                 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
3329                                     struct btrfs_file_extent_item);
3330                 btrfs_extent_item_to_extent_map(inode, path, fi, false, em);
3331                 em->generation = -1;
3332                 if (btrfs_file_extent_type(path->nodes[0], fi) ==
3333                     BTRFS_FILE_EXTENT_INLINE)
3334                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3335                                 &BTRFS_I(inode)->runtime_flags);
3336         } else {
3337                 em->start = hole_offset;
3338                 em->len = hole_len;
3339                 em->ram_bytes = em->len;
3340                 em->orig_start = hole_offset;
3341                 em->block_start = EXTENT_MAP_HOLE;
3342                 em->block_len = 0;
3343                 em->orig_block_len = 0;
3344                 em->compress_type = BTRFS_COMPRESS_NONE;
3345                 em->generation = trans->transid;
3346         }
3347
3348         while (1) {
3349                 write_lock(&em_tree->lock);
3350                 ret = add_extent_mapping(em_tree, em, 1);
3351                 write_unlock(&em_tree->lock);
3352                 if (ret != -EEXIST) {
3353                         free_extent_map(em);
3354                         break;
3355                 }
3356                 btrfs_drop_extent_cache(inode, em->start,
3357                                         em->start + em->len - 1, 0);
3358         }
3359
3360         if (ret)
3361                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3362                         &BTRFS_I(inode)->runtime_flags);
3363 }
3364
3365 /*
3366  * Make sure we do not end up inserting an inline extent into a file that has
3367  * already other (non-inline) extents. If a file has an inline extent it can
3368  * not have any other extents and the (single) inline extent must start at the
3369  * file offset 0. Failing to respect these rules will lead to file corruption,
3370  * resulting in EIO errors on read/write operations, hitting BUG_ON's in mm, etc
3371  *
3372  * We can have extents that have been already written to disk or we can have
3373  * dirty ranges still in delalloc, in which case the extent maps and items are
3374  * created only when we run delalloc, and the delalloc ranges might fall outside
3375  * the range we are currently locking in the inode's io tree. So we check the
3376  * inode's i_size because of that (i_size updates are done while holding the
3377  * i_mutex, which we are holding here).
3378  * We also check to see if the inode has a size not greater than "datal" but has
3379  * extents beyond it, due to an fallocate with FALLOC_FL_KEEP_SIZE (and we are
3380  * protected against such concurrent fallocate calls by the i_mutex).
3381  *
3382  * If the file has no extents but a size greater than datal, do not allow the
3383  * copy because we would need turn the inline extent into a non-inline one (even
3384  * with NO_HOLES enabled). If we find our destination inode only has one inline
3385  * extent, just overwrite it with the source inline extent if its size is less
3386  * than the source extent's size, or we could copy the source inline extent's
3387  * data into the destination inode's inline extent if the later is greater then
3388  * the former.
3389  */
3390 static int clone_copy_inline_extent(struct inode *src,
3391                                     struct inode *dst,
3392                                     struct btrfs_trans_handle *trans,
3393                                     struct btrfs_path *path,
3394                                     struct btrfs_key *new_key,
3395                                     const u64 drop_start,
3396                                     const u64 datal,
3397                                     const u64 skip,
3398                                     const u64 size,
3399                                     char *inline_data)
3400 {
3401         struct btrfs_root *root = BTRFS_I(dst)->root;
3402         const u64 aligned_end = ALIGN(new_key->offset + datal,
3403                                       root->sectorsize);
3404         int ret;
3405         struct btrfs_key key;
3406
3407         if (new_key->offset > 0)
3408                 return -EOPNOTSUPP;
3409
3410         key.objectid = btrfs_ino(dst);
3411         key.type = BTRFS_EXTENT_DATA_KEY;
3412         key.offset = 0;
3413         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3414         if (ret < 0) {
3415                 return ret;
3416         } else if (ret > 0) {
3417                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3418                         ret = btrfs_next_leaf(root, path);
3419                         if (ret < 0)
3420                                 return ret;
3421                         else if (ret > 0)
3422                                 goto copy_inline_extent;
3423                 }
3424                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3425                 if (key.objectid == btrfs_ino(dst) &&
3426                     key.type == BTRFS_EXTENT_DATA_KEY) {
3427                         ASSERT(key.offset > 0);
3428                         return -EOPNOTSUPP;
3429                 }
3430         } else if (i_size_read(dst) <= datal) {
3431                 struct btrfs_file_extent_item *ei;
3432                 u64 ext_len;
3433
3434                 /*
3435                  * If the file size is <= datal, make sure there are no other
3436                  * extents following (can happen do to an fallocate call with
3437                  * the flag FALLOC_FL_KEEP_SIZE).
3438                  */
3439                 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3440                                     struct btrfs_file_extent_item);
3441                 /*
3442                  * If it's an inline extent, it can not have other extents
3443                  * following it.
3444                  */
3445                 if (btrfs_file_extent_type(path->nodes[0], ei) ==
3446                     BTRFS_FILE_EXTENT_INLINE)
3447                         goto copy_inline_extent;
3448
3449                 ext_len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
3450                 if (ext_len > aligned_end)
3451                         return -EOPNOTSUPP;
3452
3453                 ret = btrfs_next_item(root, path);
3454                 if (ret < 0) {
3455                         return ret;
3456                 } else if (ret == 0) {
3457                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3458                                               path->slots[0]);
3459                         if (key.objectid == btrfs_ino(dst) &&
3460                             key.type == BTRFS_EXTENT_DATA_KEY)
3461                                 return -EOPNOTSUPP;
3462                 }
3463         }
3464
3465 copy_inline_extent:
3466         /*
3467          * We have no extent items, or we have an extent at offset 0 which may
3468          * or may not be inlined. All these cases are dealt the same way.
3469          */
3470         if (i_size_read(dst) > datal) {
3471                 /*
3472                  * If the destination inode has an inline extent...
3473                  * This would require copying the data from the source inline
3474                  * extent into the beginning of the destination's inline extent.
3475                  * But this is really complex, both extents can be compressed
3476                  * or just one of them, which would require decompressing and
3477                  * re-compressing data (which could increase the new compressed
3478                  * size, not allowing the compressed data to fit anymore in an
3479                  * inline extent).
3480                  * So just don't support this case for now (it should be rare,
3481                  * we are not really saving space when cloning inline extents).
3482                  */
3483                 return -EOPNOTSUPP;
3484         }
3485
3486         btrfs_release_path(path);
3487         ret = btrfs_drop_extents(trans, root, dst, drop_start, aligned_end, 1);
3488         if (ret)
3489                 return ret;
3490         ret = btrfs_insert_empty_item(trans, root, path, new_key, size);
3491         if (ret)
3492                 return ret;
3493
3494         if (skip) {
3495                 const u32 start = btrfs_file_extent_calc_inline_size(0);
3496
3497                 memmove(inline_data + start, inline_data + start + skip, datal);
3498         }
3499
3500         write_extent_buffer(path->nodes[0], inline_data,
3501                             btrfs_item_ptr_offset(path->nodes[0],
3502                                                   path->slots[0]),
3503                             size);
3504         inode_add_bytes(dst, datal);
3505
3506         return 0;
3507 }
3508
3509 /**
3510  * btrfs_clone() - clone a range from inode file to another
3511  *
3512  * @src: Inode to clone from
3513  * @inode: Inode to clone to
3514  * @off: Offset within source to start clone from
3515  * @olen: Original length, passed by user, of range to clone
3516  * @olen_aligned: Block-aligned value of olen
3517  * @destoff: Offset within @inode to start clone
3518  * @no_time_update: Whether to update mtime/ctime on the target inode
3519  */
3520 static int btrfs_clone(struct inode *src, struct inode *inode,
3521                        const u64 off, const u64 olen, const u64 olen_aligned,
3522                        const u64 destoff, int no_time_update)
3523 {
3524         struct btrfs_root *root = BTRFS_I(inode)->root;
3525         struct btrfs_path *path = NULL;
3526         struct extent_buffer *leaf;
3527         struct btrfs_trans_handle *trans;
3528         char *buf = NULL;
3529         struct btrfs_key key;
3530         u32 nritems;
3531         int slot;
3532         int ret;
3533         const u64 len = olen_aligned;
3534         u64 last_dest_end = destoff;
3535
3536         ret = -ENOMEM;
3537         buf = kmalloc(root->nodesize, GFP_KERNEL | __GFP_NOWARN);
3538         if (!buf) {
3539                 buf = vmalloc(root->nodesize);
3540                 if (!buf)
3541                         return ret;
3542         }
3543
3544         path = btrfs_alloc_path();
3545         if (!path) {
3546                 kvfree(buf);
3547                 return ret;
3548         }
3549
3550         path->reada = READA_FORWARD;
3551         /* clone data */
3552         key.objectid = btrfs_ino(src);
3553         key.type = BTRFS_EXTENT_DATA_KEY;
3554         key.offset = off;
3555
3556         while (1) {
3557                 u64 next_key_min_offset = key.offset + 1;
3558
3559                 /*
3560                  * note the key will change type as we walk through the
3561                  * tree.
3562                  */
3563                 path->leave_spinning = 1;
3564                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
3565                                 0, 0);
3566                 if (ret < 0)
3567                         goto out;
3568                 /*
3569                  * First search, if no extent item that starts at offset off was
3570                  * found but the previous item is an extent item, it's possible
3571                  * it might overlap our target range, therefore process it.
3572                  */
3573                 if (key.offset == off && ret > 0 && path->slots[0] > 0) {
3574                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3575                                               path->slots[0] - 1);
3576                         if (key.type == BTRFS_EXTENT_DATA_KEY)
3577                                 path->slots[0]--;
3578                 }
3579
3580                 nritems = btrfs_header_nritems(path->nodes[0]);
3581 process_slot:
3582                 if (path->slots[0] >= nritems) {
3583                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
3584                         if (ret < 0)
3585                                 goto out;
3586                         if (ret > 0)
3587                                 break;
3588                         nritems = btrfs_header_nritems(path->nodes[0]);
3589                 }
3590                 leaf = path->nodes[0];
3591                 slot = path->slots[0];
3592
3593                 btrfs_item_key_to_cpu(leaf, &key, slot);
3594                 if (key.type > BTRFS_EXTENT_DATA_KEY ||
3595                     key.objectid != btrfs_ino(src))
3596                         break;
3597
3598                 if (key.type == BTRFS_EXTENT_DATA_KEY) {
3599                         struct btrfs_file_extent_item *extent;
3600                         int type;
3601                         u32 size;
3602                         struct btrfs_key new_key;
3603                         u64 disko = 0, diskl = 0;
3604                         u64 datao = 0, datal = 0;
3605                         u8 comp;
3606                         u64 drop_start;
3607
3608                         extent = btrfs_item_ptr(leaf, slot,
3609                                                 struct btrfs_file_extent_item);
3610                         comp = btrfs_file_extent_compression(leaf, extent);
3611                         type = btrfs_file_extent_type(leaf, extent);
3612                         if (type == BTRFS_FILE_EXTENT_REG ||
3613                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3614                                 disko = btrfs_file_extent_disk_bytenr(leaf,
3615                                                                       extent);
3616                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
3617                                                                  extent);
3618                                 datao = btrfs_file_extent_offset(leaf, extent);
3619                                 datal = btrfs_file_extent_num_bytes(leaf,
3620                                                                     extent);
3621                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3622                                 /* take upper bound, may be compressed */
3623                                 datal = btrfs_file_extent_ram_bytes(leaf,
3624                                                                     extent);
3625                         }
3626
3627                         /*
3628                          * The first search might have left us at an extent
3629                          * item that ends before our target range's start, can
3630                          * happen if we have holes and NO_HOLES feature enabled.
3631                          */
3632                         if (key.offset + datal <= off) {
3633                                 path->slots[0]++;
3634                                 goto process_slot;
3635                         } else if (key.offset >= off + len) {
3636                                 break;
3637                         }
3638                         next_key_min_offset = key.offset + datal;
3639                         size = btrfs_item_size_nr(leaf, slot);
3640                         read_extent_buffer(leaf, buf,
3641                                            btrfs_item_ptr_offset(leaf, slot),
3642                                            size);
3643
3644                         btrfs_release_path(path);
3645                         path->leave_spinning = 0;
3646
3647                         memcpy(&new_key, &key, sizeof(new_key));
3648                         new_key.objectid = btrfs_ino(inode);
3649                         if (off <= key.offset)
3650                                 new_key.offset = key.offset + destoff - off;
3651                         else
3652                                 new_key.offset = destoff;
3653
3654                         /*
3655                          * Deal with a hole that doesn't have an extent item
3656                          * that represents it (NO_HOLES feature enabled).
3657                          * This hole is either in the middle of the cloning
3658                          * range or at the beginning (fully overlaps it or
3659                          * partially overlaps it).
3660                          */
3661                         if (new_key.offset != last_dest_end)
3662                                 drop_start = last_dest_end;
3663                         else
3664                                 drop_start = new_key.offset;
3665
3666                         /*
3667                          * 1 - adjusting old extent (we may have to split it)
3668                          * 1 - add new extent
3669                          * 1 - inode update
3670                          */
3671                         trans = btrfs_start_transaction(root, 3);
3672                         if (IS_ERR(trans)) {
3673                                 ret = PTR_ERR(trans);
3674                                 goto out;
3675                         }
3676
3677                         if (type == BTRFS_FILE_EXTENT_REG ||
3678                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3679                                 /*
3680                                  *    a  | --- range to clone ---|  b
3681                                  * | ------------- extent ------------- |
3682                                  */
3683
3684                                 /* subtract range b */
3685                                 if (key.offset + datal > off + len)
3686                                         datal = off + len - key.offset;
3687
3688                                 /* subtract range a */
3689                                 if (off > key.offset) {
3690                                         datao += off - key.offset;
3691                                         datal -= off - key.offset;
3692                                 }
3693
3694                                 ret = btrfs_drop_extents(trans, root, inode,
3695                                                          drop_start,
3696                                                          new_key.offset + datal,
3697                                                          1);
3698                                 if (ret) {
3699                                         if (ret != -EOPNOTSUPP)
3700                                                 btrfs_abort_transaction(trans,
3701                                                                 root, ret);
3702                                         btrfs_end_transaction(trans, root);
3703                                         goto out;
3704                                 }
3705
3706                                 ret = btrfs_insert_empty_item(trans, root, path,
3707                                                               &new_key, size);
3708                                 if (ret) {
3709                                         btrfs_abort_transaction(trans, root,
3710                                                                 ret);
3711                                         btrfs_end_transaction(trans, root);
3712                                         goto out;
3713                                 }
3714
3715                                 leaf = path->nodes[0];
3716                                 slot = path->slots[0];
3717                                 write_extent_buffer(leaf, buf,
3718                                             btrfs_item_ptr_offset(leaf, slot),
3719                                             size);
3720
3721                                 extent = btrfs_item_ptr(leaf, slot,
3722                                                 struct btrfs_file_extent_item);
3723
3724                                 /* disko == 0 means it's a hole */
3725                                 if (!disko)
3726                                         datao = 0;
3727
3728                                 btrfs_set_file_extent_offset(leaf, extent,
3729                                                              datao);
3730                                 btrfs_set_file_extent_num_bytes(leaf, extent,
3731                                                                 datal);
3732
3733                                 if (disko) {
3734                                         inode_add_bytes(inode, datal);
3735                                         ret = btrfs_inc_extent_ref(trans, root,
3736                                                         disko, diskl, 0,
3737                                                         root->root_key.objectid,
3738                                                         btrfs_ino(inode),
3739                                                         new_key.offset - datao);
3740                                         if (ret) {
3741                                                 btrfs_abort_transaction(trans,
3742                                                                         root,
3743                                                                         ret);
3744                                                 btrfs_end_transaction(trans,
3745                                                                       root);
3746                                                 goto out;
3747
3748                                         }
3749                                 }
3750                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3751                                 u64 skip = 0;
3752                                 u64 trim = 0;
3753
3754                                 if (off > key.offset) {
3755                                         skip = off - key.offset;
3756                                         new_key.offset += skip;
3757                                 }
3758
3759                                 if (key.offset + datal > off + len)
3760                                         trim = key.offset + datal - (off + len);
3761
3762                                 if (comp && (skip || trim)) {
3763                                         ret = -EINVAL;
3764                                         btrfs_end_transaction(trans, root);
3765                                         goto out;
3766                                 }
3767                                 size -= skip + trim;
3768                                 datal -= skip + trim;
3769
3770                                 ret = clone_copy_inline_extent(src, inode,
3771                                                                trans, path,
3772                                                                &new_key,
3773                                                                drop_start,
3774                                                                datal,
3775                                                                skip, size, buf);
3776                                 if (ret) {
3777                                         if (ret != -EOPNOTSUPP)
3778                                                 btrfs_abort_transaction(trans,
3779                                                                         root,
3780                                                                         ret);
3781                                         btrfs_end_transaction(trans, root);
3782                                         goto out;
3783                                 }
3784                                 leaf = path->nodes[0];
3785                                 slot = path->slots[0];
3786                         }
3787
3788                         /* If we have an implicit hole (NO_HOLES feature). */
3789                         if (drop_start < new_key.offset)
3790                                 clone_update_extent_map(inode, trans,
3791                                                 NULL, drop_start,
3792                                                 new_key.offset - drop_start);
3793
3794                         clone_update_extent_map(inode, trans, path, 0, 0);
3795
3796                         btrfs_mark_buffer_dirty(leaf);
3797                         btrfs_release_path(path);
3798
3799                         last_dest_end = ALIGN(new_key.offset + datal,
3800                                               root->sectorsize);
3801                         ret = clone_finish_inode_update(trans, inode,
3802                                                         last_dest_end,
3803                                                         destoff, olen,
3804                                                         no_time_update);
3805                         if (ret)
3806                                 goto out;
3807                         if (new_key.offset + datal >= destoff + len)
3808                                 break;
3809                 }
3810                 btrfs_release_path(path);
3811                 key.offset = next_key_min_offset;
3812         }
3813         ret = 0;
3814
3815         if (last_dest_end < destoff + len) {
3816                 /*
3817                  * We have an implicit hole (NO_HOLES feature is enabled) that
3818                  * fully or partially overlaps our cloning range at its end.
3819                  */
3820                 btrfs_release_path(path);
3821
3822                 /*
3823                  * 1 - remove extent(s)
3824                  * 1 - inode update
3825                  */
3826                 trans = btrfs_start_transaction(root, 2);
3827                 if (IS_ERR(trans)) {
3828                         ret = PTR_ERR(trans);
3829                         goto out;
3830                 }
3831                 ret = btrfs_drop_extents(trans, root, inode,
3832                                          last_dest_end, destoff + len, 1);
3833                 if (ret) {
3834                         if (ret != -EOPNOTSUPP)
3835                                 btrfs_abort_transaction(trans, root, ret);
3836                         btrfs_end_transaction(trans, root);
3837                         goto out;
3838                 }
3839                 clone_update_extent_map(inode, trans, NULL, last_dest_end,
3840                                         destoff + len - last_dest_end);
3841                 ret = clone_finish_inode_update(trans, inode, destoff + len,
3842                                                 destoff, olen, no_time_update);
3843         }
3844
3845 out:
3846         btrfs_free_path(path);
3847         kvfree(buf);
3848         return ret;
3849 }
3850
3851 static noinline int btrfs_clone_files(struct file *file, struct file *file_src,
3852                                         u64 off, u64 olen, u64 destoff)
3853 {
3854         struct inode *inode = file_inode(file);
3855         struct inode *src = file_inode(file_src);
3856         struct btrfs_root *root = BTRFS_I(inode)->root;
3857         int ret;
3858         u64 len = olen;
3859         u64 bs = root->fs_info->sb->s_blocksize;
3860         int same_inode = src == inode;
3861
3862         /*
3863          * TODO:
3864          * - split compressed inline extents.  annoying: we need to
3865          *   decompress into destination's address_space (the file offset
3866          *   may change, so source mapping won't do), then recompress (or
3867          *   otherwise reinsert) a subrange.
3868          *
3869          * - split destination inode's inline extents.  The inline extents can
3870          *   be either compressed or non-compressed.
3871          */
3872
3873         if (btrfs_root_readonly(root))
3874                 return -EROFS;
3875
3876         if (file_src->f_path.mnt != file->f_path.mnt ||
3877             src->i_sb != inode->i_sb)
3878                 return -EXDEV;
3879
3880         /* don't make the dst file partly checksummed */
3881         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3882             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
3883                 return -EINVAL;
3884
3885         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
3886                 return -EISDIR;
3887
3888         if (!same_inode) {
3889                 btrfs_double_inode_lock(src, inode);
3890         } else {
3891                 inode_lock(src);
3892         }
3893
3894         /* determine range to clone */
3895         ret = -EINVAL;
3896         if (off + len > src->i_size || off + len < off)
3897                 goto out_unlock;
3898         if (len == 0)
3899                 olen = len = src->i_size - off;
3900         /* if we extend to eof, continue to block boundary */
3901         if (off + len == src->i_size)
3902                 len = ALIGN(src->i_size, bs) - off;
3903
3904         if (len == 0) {
3905                 ret = 0;
3906                 goto out_unlock;
3907         }
3908
3909         /* verify the end result is block aligned */
3910         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
3911             !IS_ALIGNED(destoff, bs))
3912                 goto out_unlock;
3913
3914         /* verify if ranges are overlapped within the same file */
3915         if (same_inode) {
3916                 if (destoff + len > off && destoff < off + len)
3917                         goto out_unlock;
3918         }
3919
3920         if (destoff > inode->i_size) {
3921                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3922                 if (ret)
3923                         goto out_unlock;
3924         }
3925
3926         /*
3927          * Lock the target range too. Right after we replace the file extent
3928          * items in the fs tree (which now point to the cloned data), we might
3929          * have a worker replace them with extent items relative to a write
3930          * operation that was issued before this clone operation (i.e. confront
3931          * with inode.c:btrfs_finish_ordered_io).
3932          */
3933         if (same_inode) {
3934                 u64 lock_start = min_t(u64, off, destoff);
3935                 u64 lock_len = max_t(u64, off, destoff) + len - lock_start;
3936
3937                 ret = lock_extent_range(src, lock_start, lock_len, true);
3938         } else {
3939                 ret = btrfs_double_extent_lock(src, off, inode, destoff, len,
3940                                                true);
3941         }
3942         ASSERT(ret == 0);
3943         if (WARN_ON(ret)) {
3944                 /* ranges in the io trees already unlocked */
3945                 goto out_unlock;
3946         }
3947
3948         ret = btrfs_clone(src, inode, off, olen, len, destoff, 0);
3949
3950         if (same_inode) {
3951                 u64 lock_start = min_t(u64, off, destoff);
3952                 u64 lock_end = max_t(u64, off, destoff) + len - 1;
3953
3954                 unlock_extent(&BTRFS_I(src)->io_tree, lock_start, lock_end);
3955         } else {
3956                 btrfs_double_extent_unlock(src, off, inode, destoff, len);
3957         }
3958         /*
3959          * Truncate page cache pages so that future reads will see the cloned
3960          * data immediately and not the previous data.
3961          */
3962         truncate_inode_pages_range(&inode->i_data,
3963                                 round_down(destoff, PAGE_SIZE),
3964                                 round_up(destoff + len, PAGE_SIZE) - 1);
3965 out_unlock:
3966         if (!same_inode)
3967                 btrfs_double_inode_unlock(src, inode);
3968         else
3969                 inode_unlock(src);
3970         return ret;
3971 }
3972
3973 ssize_t btrfs_copy_file_range(struct file *file_in, loff_t pos_in,
3974                               struct file *file_out, loff_t pos_out,
3975                               size_t len, unsigned int flags)
3976 {
3977         ssize_t ret;
3978
3979         ret = btrfs_clone_files(file_out, file_in, pos_in, len, pos_out);
3980         if (ret == 0)
3981                 ret = len;
3982         return ret;
3983 }
3984
3985 int btrfs_clone_file_range(struct file *src_file, loff_t off,
3986                 struct file *dst_file, loff_t destoff, u64 len)
3987 {
3988         return btrfs_clone_files(dst_file, src_file, off, len, destoff);
3989 }
3990
3991 /*
3992  * there are many ways the trans_start and trans_end ioctls can lead
3993  * to deadlocks.  They should only be used by applications that
3994  * basically own the machine, and have a very in depth understanding
3995  * of all the possible deadlocks and enospc problems.
3996  */
3997 static long btrfs_ioctl_trans_start(struct file *file)
3998 {
3999         struct inode *inode = file_inode(file);
4000         struct btrfs_root *root = BTRFS_I(inode)->root;
4001         struct btrfs_trans_handle *trans;
4002         int ret;
4003
4004         ret = -EPERM;
4005         if (!capable(CAP_SYS_ADMIN))
4006                 goto out;
4007
4008         ret = -EINPROGRESS;
4009         if (file->private_data)
4010                 goto out;
4011
4012         ret = -EROFS;
4013         if (btrfs_root_readonly(root))
4014                 goto out;
4015
4016         ret = mnt_want_write_file(file);
4017         if (ret)
4018                 goto out;
4019
4020         atomic_inc(&root->fs_info->open_ioctl_trans);
4021
4022         ret = -ENOMEM;
4023         trans = btrfs_start_ioctl_transaction(root);
4024         if (IS_ERR(trans))
4025                 goto out_drop;
4026
4027         file->private_data = trans;
4028         return 0;
4029
4030 out_drop:
4031         atomic_dec(&root->fs_info->open_ioctl_trans);
4032         mnt_drop_write_file(file);
4033 out:
4034         return ret;
4035 }
4036
4037 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
4038 {
4039         struct inode *inode = file_inode(file);
4040         struct btrfs_root *root = BTRFS_I(inode)->root;
4041         struct btrfs_root *new_root;
4042         struct btrfs_dir_item *di;
4043         struct btrfs_trans_handle *trans;
4044         struct btrfs_path *path;
4045         struct btrfs_key location;
4046         struct btrfs_disk_key disk_key;
4047         u64 objectid = 0;
4048         u64 dir_id;
4049         int ret;
4050
4051         if (!capable(CAP_SYS_ADMIN))
4052                 return -EPERM;
4053
4054         ret = mnt_want_write_file(file);
4055         if (ret)
4056                 return ret;
4057
4058         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
4059                 ret = -EFAULT;
4060                 goto out;
4061         }
4062
4063         if (!objectid)
4064                 objectid = BTRFS_FS_TREE_OBJECTID;
4065
4066         location.objectid = objectid;
4067         location.type = BTRFS_ROOT_ITEM_KEY;
4068         location.offset = (u64)-1;
4069
4070         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
4071         if (IS_ERR(new_root)) {
4072                 ret = PTR_ERR(new_root);
4073                 goto out;
4074         }
4075
4076         path = btrfs_alloc_path();
4077         if (!path) {
4078                 ret = -ENOMEM;
4079                 goto out;
4080         }
4081         path->leave_spinning = 1;
4082
4083         trans = btrfs_start_transaction(root, 1);
4084         if (IS_ERR(trans)) {
4085                 btrfs_free_path(path);
4086                 ret = PTR_ERR(trans);
4087                 goto out;
4088         }
4089
4090         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
4091         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
4092                                    dir_id, "default", 7, 1);
4093         if (IS_ERR_OR_NULL(di)) {
4094                 btrfs_free_path(path);
4095                 btrfs_end_transaction(trans, root);
4096                 btrfs_err(new_root->fs_info, "Umm, you don't have the default dir"
4097                            "item, this isn't going to work");
4098                 ret = -ENOENT;
4099                 goto out;
4100         }
4101
4102         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
4103         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
4104         btrfs_mark_buffer_dirty(path->nodes[0]);
4105         btrfs_free_path(path);
4106
4107         btrfs_set_fs_incompat(root->fs_info, DEFAULT_SUBVOL);
4108         btrfs_end_transaction(trans, root);
4109 out:
4110         mnt_drop_write_file(file);
4111         return ret;
4112 }
4113
4114 void btrfs_get_block_group_info(struct list_head *groups_list,
4115                                 struct btrfs_ioctl_space_info *space)
4116 {
4117         struct btrfs_block_group_cache *block_group;
4118
4119         space->total_bytes = 0;
4120         space->used_bytes = 0;
4121         space->flags = 0;
4122         list_for_each_entry(block_group, groups_list, list) {
4123                 space->flags = block_group->flags;
4124                 space->total_bytes += block_group->key.offset;
4125                 space->used_bytes +=
4126                         btrfs_block_group_used(&block_group->item);
4127         }
4128 }
4129
4130 static long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
4131 {
4132         struct btrfs_ioctl_space_args space_args;
4133         struct btrfs_ioctl_space_info space;
4134         struct btrfs_ioctl_space_info *dest;
4135         struct btrfs_ioctl_space_info *dest_orig;
4136         struct btrfs_ioctl_space_info __user *user_dest;
4137         struct btrfs_space_info *info;
4138         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
4139                        BTRFS_BLOCK_GROUP_SYSTEM,
4140                        BTRFS_BLOCK_GROUP_METADATA,
4141                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
4142         int num_types = 4;
4143         int alloc_size;
4144         int ret = 0;
4145         u64 slot_count = 0;
4146         int i, c;
4147
4148         if (copy_from_user(&space_args,
4149                            (struct btrfs_ioctl_space_args __user *)arg,
4150                            sizeof(space_args)))
4151                 return -EFAULT;
4152
4153         for (i = 0; i < num_types; i++) {
4154                 struct btrfs_space_info *tmp;
4155
4156                 info = NULL;
4157                 rcu_read_lock();
4158                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
4159                                         list) {
4160                         if (tmp->flags == types[i]) {
4161                                 info = tmp;
4162                                 break;
4163                         }
4164                 }
4165                 rcu_read_unlock();
4166
4167                 if (!info)
4168                         continue;
4169
4170                 down_read(&info->groups_sem);
4171                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4172                         if (!list_empty(&info->block_groups[c]))
4173                                 slot_count++;
4174                 }
4175                 up_read(&info->groups_sem);
4176         }
4177
4178         /*
4179          * Global block reserve, exported as a space_info
4180          */
4181         slot_count++;
4182
4183         /* space_slots == 0 means they are asking for a count */
4184         if (space_args.space_slots == 0) {
4185                 space_args.total_spaces = slot_count;
4186                 goto out;
4187         }
4188
4189         slot_count = min_t(u64, space_args.space_slots, slot_count);
4190
4191         alloc_size = sizeof(*dest) * slot_count;
4192
4193         /* we generally have at most 6 or so space infos, one for each raid
4194          * level.  So, a whole page should be more than enough for everyone
4195          */
4196         if (alloc_size > PAGE_SIZE)
4197                 return -ENOMEM;
4198
4199         space_args.total_spaces = 0;
4200         dest = kmalloc(alloc_size, GFP_KERNEL);
4201         if (!dest)
4202                 return -ENOMEM;
4203         dest_orig = dest;
4204
4205         /* now we have a buffer to copy into */
4206         for (i = 0; i < num_types; i++) {
4207                 struct btrfs_space_info *tmp;
4208
4209                 if (!slot_count)
4210                         break;
4211
4212                 info = NULL;
4213                 rcu_read_lock();
4214                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
4215                                         list) {
4216                         if (tmp->flags == types[i]) {
4217                                 info = tmp;
4218                                 break;
4219                         }
4220                 }
4221                 rcu_read_unlock();
4222
4223                 if (!info)
4224                         continue;
4225                 down_read(&info->groups_sem);
4226                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4227                         if (!list_empty(&info->block_groups[c])) {
4228                                 btrfs_get_block_group_info(
4229                                         &info->block_groups[c], &space);
4230                                 memcpy(dest, &space, sizeof(space));
4231                                 dest++;
4232                                 space_args.total_spaces++;
4233                                 slot_count--;
4234                         }
4235                         if (!slot_count)
4236                                 break;
4237                 }
4238                 up_read(&info->groups_sem);
4239         }
4240
4241         /*
4242          * Add global block reserve
4243          */
4244         if (slot_count) {
4245                 struct btrfs_block_rsv *block_rsv = &root->fs_info->global_block_rsv;
4246
4247                 spin_lock(&block_rsv->lock);
4248                 space.total_bytes = block_rsv->size;
4249                 space.used_bytes = block_rsv->size - block_rsv->reserved;
4250                 spin_unlock(&block_rsv->lock);
4251                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
4252                 memcpy(dest, &space, sizeof(space));
4253                 space_args.total_spaces++;
4254         }
4255
4256         user_dest = (struct btrfs_ioctl_space_info __user *)
4257                 (arg + sizeof(struct btrfs_ioctl_space_args));
4258
4259         if (copy_to_user(user_dest, dest_orig, alloc_size))
4260                 ret = -EFAULT;
4261
4262         kfree(dest_orig);
4263 out:
4264         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
4265                 ret = -EFAULT;
4266
4267         return ret;
4268 }
4269
4270 /*
4271  * there are many ways the trans_start and trans_end ioctls can lead
4272  * to deadlocks.  They should only be used by applications that
4273  * basically own the machine, and have a very in depth understanding
4274  * of all the possible deadlocks and enospc problems.
4275  */
4276 long btrfs_ioctl_trans_end(struct file *file)
4277 {
4278         struct inode *inode = file_inode(file);
4279         struct btrfs_root *root = BTRFS_I(inode)->root;
4280         struct btrfs_trans_handle *trans;
4281
4282         trans = file->private_data;
4283         if (!trans)
4284                 return -EINVAL;
4285         file->private_data = NULL;
4286
4287         btrfs_end_transaction(trans, root);
4288
4289         atomic_dec(&root->fs_info->open_ioctl_trans);
4290
4291         mnt_drop_write_file(file);
4292         return 0;
4293 }
4294
4295 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
4296                                             void __user *argp)
4297 {
4298         struct btrfs_trans_handle *trans;
4299         u64 transid;
4300         int ret;
4301
4302         trans = btrfs_attach_transaction_barrier(root);
4303         if (IS_ERR(trans)) {
4304                 if (PTR_ERR(trans) != -ENOENT)
4305                         return PTR_ERR(trans);
4306
4307                 /* No running transaction, don't bother */
4308                 transid = root->fs_info->last_trans_committed;
4309                 goto out;
4310         }
4311         transid = trans->transid;
4312         ret = btrfs_commit_transaction_async(trans, root, 0);
4313         if (ret) {
4314                 btrfs_end_transaction(trans, root);
4315                 return ret;
4316         }
4317 out:
4318         if (argp)
4319                 if (copy_to_user(argp, &transid, sizeof(transid)))
4320                         return -EFAULT;
4321         return 0;
4322 }
4323
4324 static noinline long btrfs_ioctl_wait_sync(struct btrfs_root *root,
4325                                            void __user *argp)
4326 {
4327         u64 transid;
4328
4329         if (argp) {
4330                 if (copy_from_user(&transid, argp, sizeof(transid)))
4331                         return -EFAULT;
4332         } else {
4333                 transid = 0;  /* current trans */
4334         }
4335         return btrfs_wait_for_commit(root, transid);
4336 }
4337
4338 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
4339 {
4340         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4341         struct btrfs_ioctl_scrub_args *sa;
4342         int ret;
4343
4344         if (!capable(CAP_SYS_ADMIN))
4345                 return -EPERM;
4346
4347         sa = memdup_user(arg, sizeof(*sa));
4348         if (IS_ERR(sa))
4349                 return PTR_ERR(sa);
4350
4351         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
4352                 ret = mnt_want_write_file(file);
4353                 if (ret)
4354                         goto out;
4355         }
4356
4357         ret = btrfs_scrub_dev(root->fs_info, sa->devid, sa->start, sa->end,
4358                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4359                               0);
4360
4361         if (copy_to_user(arg, sa, sizeof(*sa)))
4362                 ret = -EFAULT;
4363
4364         if (!(sa->flags & BTRFS_SCRUB_READONLY))
4365                 mnt_drop_write_file(file);
4366 out:
4367         kfree(sa);
4368         return ret;
4369 }
4370
4371 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
4372 {
4373         if (!capable(CAP_SYS_ADMIN))
4374                 return -EPERM;
4375
4376         return btrfs_scrub_cancel(root->fs_info);
4377 }
4378
4379 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
4380                                        void __user *arg)
4381 {
4382         struct btrfs_ioctl_scrub_args *sa;
4383         int ret;
4384
4385         if (!capable(CAP_SYS_ADMIN))
4386                 return -EPERM;
4387
4388         sa = memdup_user(arg, sizeof(*sa));
4389         if (IS_ERR(sa))
4390                 return PTR_ERR(sa);
4391
4392         ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
4393
4394         if (copy_to_user(arg, sa, sizeof(*sa)))
4395                 ret = -EFAULT;
4396
4397         kfree(sa);
4398         return ret;
4399 }
4400
4401 static long btrfs_ioctl_get_dev_stats(struct btrfs_root *root,
4402                                       void __user *arg)
4403 {
4404         struct btrfs_ioctl_get_dev_stats *sa;
4405         int ret;
4406
4407         sa = memdup_user(arg, sizeof(*sa));
4408         if (IS_ERR(sa))
4409                 return PTR_ERR(sa);
4410
4411         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4412                 kfree(sa);
4413                 return -EPERM;
4414         }
4415
4416         ret = btrfs_get_dev_stats(root, sa);
4417
4418         if (copy_to_user(arg, sa, sizeof(*sa)))
4419                 ret = -EFAULT;
4420
4421         kfree(sa);
4422         return ret;
4423 }
4424
4425 static long btrfs_ioctl_dev_replace(struct btrfs_root *root, void __user *arg)
4426 {
4427         struct btrfs_ioctl_dev_replace_args *p;
4428         int ret;
4429
4430         if (!capable(CAP_SYS_ADMIN))
4431                 return -EPERM;
4432
4433         p = memdup_user(arg, sizeof(*p));
4434         if (IS_ERR(p))
4435                 return PTR_ERR(p);
4436
4437         switch (p->cmd) {
4438         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4439                 if (root->fs_info->sb->s_flags & MS_RDONLY) {
4440                         ret = -EROFS;
4441                         goto out;
4442                 }
4443                 if (atomic_xchg(
4444                         &root->fs_info->mutually_exclusive_operation_running,
4445                         1)) {
4446                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4447                 } else {
4448                         ret = btrfs_dev_replace_by_ioctl(root, p);
4449                         atomic_set(
4450                          &root->fs_info->mutually_exclusive_operation_running,
4451                          0);
4452                 }
4453                 break;
4454         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4455                 btrfs_dev_replace_status(root->fs_info, p);
4456                 ret = 0;
4457                 break;
4458         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4459                 ret = btrfs_dev_replace_cancel(root->fs_info, p);
4460                 break;
4461         default:
4462                 ret = -EINVAL;
4463                 break;
4464         }
4465
4466         if (copy_to_user(arg, p, sizeof(*p)))
4467                 ret = -EFAULT;
4468 out:
4469         kfree(p);
4470         return ret;
4471 }
4472
4473 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4474 {
4475         int ret = 0;
4476         int i;
4477         u64 rel_ptr;
4478         int size;
4479         struct btrfs_ioctl_ino_path_args *ipa = NULL;
4480         struct inode_fs_paths *ipath = NULL;
4481         struct btrfs_path *path;
4482
4483         if (!capable(CAP_DAC_READ_SEARCH))
4484                 return -EPERM;
4485
4486         path = btrfs_alloc_path();
4487         if (!path) {
4488                 ret = -ENOMEM;
4489                 goto out;
4490         }
4491
4492         ipa = memdup_user(arg, sizeof(*ipa));
4493         if (IS_ERR(ipa)) {
4494                 ret = PTR_ERR(ipa);
4495                 ipa = NULL;
4496                 goto out;
4497         }
4498
4499         size = min_t(u32, ipa->size, 4096);
4500         ipath = init_ipath(size, root, path);
4501         if (IS_ERR(ipath)) {
4502                 ret = PTR_ERR(ipath);
4503                 ipath = NULL;
4504                 goto out;
4505         }
4506
4507         ret = paths_from_inode(ipa->inum, ipath);
4508         if (ret < 0)
4509                 goto out;
4510
4511         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4512                 rel_ptr = ipath->fspath->val[i] -
4513                           (u64)(unsigned long)ipath->fspath->val;
4514                 ipath->fspath->val[i] = rel_ptr;
4515         }
4516
4517         ret = copy_to_user((void *)(unsigned long)ipa->fspath,
4518                            (void *)(unsigned long)ipath->fspath, size);
4519         if (ret) {
4520                 ret = -EFAULT;
4521                 goto out;
4522         }
4523
4524 out:
4525         btrfs_free_path(path);
4526         free_ipath(ipath);
4527         kfree(ipa);
4528
4529         return ret;
4530 }
4531
4532 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4533 {
4534         struct btrfs_data_container *inodes = ctx;
4535         const size_t c = 3 * sizeof(u64);
4536
4537         if (inodes->bytes_left >= c) {
4538                 inodes->bytes_left -= c;
4539                 inodes->val[inodes->elem_cnt] = inum;
4540                 inodes->val[inodes->elem_cnt + 1] = offset;
4541                 inodes->val[inodes->elem_cnt + 2] = root;
4542                 inodes->elem_cnt += 3;
4543         } else {
4544                 inodes->bytes_missing += c - inodes->bytes_left;
4545                 inodes->bytes_left = 0;
4546                 inodes->elem_missed += 3;
4547         }
4548
4549         return 0;
4550 }
4551
4552 static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
4553                                         void __user *arg)
4554 {
4555         int ret = 0;
4556         int size;
4557         struct btrfs_ioctl_logical_ino_args *loi;
4558         struct btrfs_data_container *inodes = NULL;
4559         struct btrfs_path *path = NULL;
4560
4561         if (!capable(CAP_SYS_ADMIN))
4562                 return -EPERM;
4563
4564         loi = memdup_user(arg, sizeof(*loi));
4565         if (IS_ERR(loi)) {
4566                 ret = PTR_ERR(loi);
4567                 loi = NULL;
4568                 goto out;
4569         }
4570
4571         path = btrfs_alloc_path();
4572         if (!path) {
4573                 ret = -ENOMEM;
4574                 goto out;
4575         }
4576
4577         size = min_t(u32, loi->size, SZ_64K);
4578         inodes = init_data_container(size);
4579         if (IS_ERR(inodes)) {
4580                 ret = PTR_ERR(inodes);
4581                 inodes = NULL;
4582                 goto out;
4583         }
4584
4585         ret = iterate_inodes_from_logical(loi->logical, root->fs_info, path,
4586                                           build_ino_list, inodes);
4587         if (ret == -EINVAL)
4588                 ret = -ENOENT;
4589         if (ret < 0)
4590                 goto out;
4591
4592         ret = copy_to_user((void *)(unsigned long)loi->inodes,
4593                            (void *)(unsigned long)inodes, size);
4594         if (ret)
4595                 ret = -EFAULT;
4596
4597 out:
4598         btrfs_free_path(path);
4599         vfree(inodes);
4600         kfree(loi);
4601
4602         return ret;
4603 }
4604
4605 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
4606                                struct btrfs_ioctl_balance_args *bargs)
4607 {
4608         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4609
4610         bargs->flags = bctl->flags;
4611
4612         if (atomic_read(&fs_info->balance_running))
4613                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4614         if (atomic_read(&fs_info->balance_pause_req))
4615                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4616         if (atomic_read(&fs_info->balance_cancel_req))
4617                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4618
4619         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4620         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4621         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4622
4623         if (lock) {
4624                 spin_lock(&fs_info->balance_lock);
4625                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4626                 spin_unlock(&fs_info->balance_lock);
4627         } else {
4628                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4629         }
4630 }
4631
4632 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4633 {
4634         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4635         struct btrfs_fs_info *fs_info = root->fs_info;
4636         struct btrfs_ioctl_balance_args *bargs;
4637         struct btrfs_balance_control *bctl;
4638         bool need_unlock; /* for mut. excl. ops lock */
4639         int ret;
4640
4641         if (!capable(CAP_SYS_ADMIN))
4642                 return -EPERM;
4643
4644         ret = mnt_want_write_file(file);
4645         if (ret)
4646                 return ret;
4647
4648 again:
4649         if (!atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)) {
4650                 mutex_lock(&fs_info->volume_mutex);
4651                 mutex_lock(&fs_info->balance_mutex);
4652                 need_unlock = true;
4653                 goto locked;
4654         }
4655
4656         /*
4657          * mut. excl. ops lock is locked.  Three possibilities:
4658          *   (1) some other op is running
4659          *   (2) balance is running
4660          *   (3) balance is paused -- special case (think resume)
4661          */
4662         mutex_lock(&fs_info->balance_mutex);
4663         if (fs_info->balance_ctl) {
4664                 /* this is either (2) or (3) */
4665                 if (!atomic_read(&fs_info->balance_running)) {
4666                         mutex_unlock(&fs_info->balance_mutex);
4667                         if (!mutex_trylock(&fs_info->volume_mutex))
4668                                 goto again;
4669                         mutex_lock(&fs_info->balance_mutex);
4670
4671                         if (fs_info->balance_ctl &&
4672                             !atomic_read(&fs_info->balance_running)) {
4673                                 /* this is (3) */
4674                                 need_unlock = false;
4675                                 goto locked;
4676                         }
4677
4678                         mutex_unlock(&fs_info->balance_mutex);
4679                         mutex_unlock(&fs_info->volume_mutex);
4680                         goto again;
4681                 } else {
4682                         /* this is (2) */
4683                         mutex_unlock(&fs_info->balance_mutex);
4684                         ret = -EINPROGRESS;
4685                         goto out;
4686                 }
4687         } else {
4688                 /* this is (1) */
4689                 mutex_unlock(&fs_info->balance_mutex);
4690                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4691                 goto out;
4692         }
4693
4694 locked:
4695         BUG_ON(!atomic_read(&fs_info->mutually_exclusive_operation_running));
4696
4697         if (arg) {
4698                 bargs = memdup_user(arg, sizeof(*bargs));
4699                 if (IS_ERR(bargs)) {
4700                         ret = PTR_ERR(bargs);
4701                         goto out_unlock;
4702                 }
4703
4704                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4705                         if (!fs_info->balance_ctl) {
4706                                 ret = -ENOTCONN;
4707                                 goto out_bargs;
4708                         }
4709
4710                         bctl = fs_info->balance_ctl;
4711                         spin_lock(&fs_info->balance_lock);
4712                         bctl->flags |= BTRFS_BALANCE_RESUME;
4713                         spin_unlock(&fs_info->balance_lock);
4714
4715                         goto do_balance;
4716                 }
4717         } else {
4718                 bargs = NULL;
4719         }
4720
4721         if (fs_info->balance_ctl) {
4722                 ret = -EINPROGRESS;
4723                 goto out_bargs;
4724         }
4725
4726         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4727         if (!bctl) {
4728                 ret = -ENOMEM;
4729                 goto out_bargs;
4730         }
4731
4732         bctl->fs_info = fs_info;
4733         if (arg) {
4734                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4735                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4736                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4737
4738                 bctl->flags = bargs->flags;
4739         } else {
4740                 /* balance everything - no filters */
4741                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4742         }
4743
4744         if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4745                 ret = -EINVAL;
4746                 goto out_bctl;
4747         }
4748
4749 do_balance:
4750         /*
4751          * Ownership of bctl and mutually_exclusive_operation_running
4752          * goes to to btrfs_balance.  bctl is freed in __cancel_balance,
4753          * or, if restriper was paused all the way until unmount, in
4754          * free_fs_info.  mutually_exclusive_operation_running is
4755          * cleared in __cancel_balance.
4756          */
4757         need_unlock = false;
4758
4759         ret = btrfs_balance(bctl, bargs);
4760         bctl = NULL;
4761
4762         if (arg) {
4763                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4764                         ret = -EFAULT;
4765         }
4766
4767 out_bctl:
4768         kfree(bctl);
4769 out_bargs:
4770         kfree(bargs);
4771 out_unlock:
4772         mutex_unlock(&fs_info->balance_mutex);
4773         mutex_unlock(&fs_info->volume_mutex);
4774         if (need_unlock)
4775                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
4776 out:
4777         mnt_drop_write_file(file);
4778         return ret;
4779 }
4780
4781 static long btrfs_ioctl_balance_ctl(struct btrfs_root *root, int cmd)
4782 {
4783         if (!capable(CAP_SYS_ADMIN))
4784                 return -EPERM;
4785
4786         switch (cmd) {
4787         case BTRFS_BALANCE_CTL_PAUSE:
4788                 return btrfs_pause_balance(root->fs_info);
4789         case BTRFS_BALANCE_CTL_CANCEL:
4790                 return btrfs_cancel_balance(root->fs_info);
4791         }
4792
4793         return -EINVAL;
4794 }
4795
4796 static long btrfs_ioctl_balance_progress(struct btrfs_root *root,
4797                                          void __user *arg)
4798 {
4799         struct btrfs_fs_info *fs_info = root->fs_info;
4800         struct btrfs_ioctl_balance_args *bargs;
4801         int ret = 0;
4802
4803         if (!capable(CAP_SYS_ADMIN))
4804                 return -EPERM;
4805
4806         mutex_lock(&fs_info->balance_mutex);
4807         if (!fs_info->balance_ctl) {
4808                 ret = -ENOTCONN;
4809                 goto out;
4810         }
4811
4812         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4813         if (!bargs) {
4814                 ret = -ENOMEM;
4815                 goto out;
4816         }
4817
4818         update_ioctl_balance_args(fs_info, 1, bargs);
4819
4820         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4821                 ret = -EFAULT;
4822
4823         kfree(bargs);
4824 out:
4825         mutex_unlock(&fs_info->balance_mutex);
4826         return ret;
4827 }
4828
4829 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4830 {
4831         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4832         struct btrfs_ioctl_quota_ctl_args *sa;
4833         struct btrfs_trans_handle *trans = NULL;
4834         int ret;
4835         int err;
4836
4837         if (!capable(CAP_SYS_ADMIN))
4838                 return -EPERM;
4839
4840         ret = mnt_want_write_file(file);
4841         if (ret)
4842                 return ret;
4843
4844         sa = memdup_user(arg, sizeof(*sa));
4845         if (IS_ERR(sa)) {
4846                 ret = PTR_ERR(sa);
4847                 goto drop_write;
4848         }
4849
4850         down_write(&root->fs_info->subvol_sem);
4851         trans = btrfs_start_transaction(root->fs_info->tree_root, 2);
4852         if (IS_ERR(trans)) {
4853                 ret = PTR_ERR(trans);
4854                 goto out;
4855         }
4856
4857         switch (sa->cmd) {
4858         case BTRFS_QUOTA_CTL_ENABLE:
4859                 ret = btrfs_quota_enable(trans, root->fs_info);
4860                 break;
4861         case BTRFS_QUOTA_CTL_DISABLE:
4862                 ret = btrfs_quota_disable(trans, root->fs_info);
4863                 break;
4864         default:
4865                 ret = -EINVAL;
4866                 break;
4867         }
4868
4869         err = btrfs_commit_transaction(trans, root->fs_info->tree_root);
4870         if (err && !ret)
4871                 ret = err;
4872 out:
4873         kfree(sa);
4874         up_write(&root->fs_info->subvol_sem);
4875 drop_write:
4876         mnt_drop_write_file(file);
4877         return ret;
4878 }
4879
4880 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4881 {
4882         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4883         struct btrfs_ioctl_qgroup_assign_args *sa;
4884         struct btrfs_trans_handle *trans;
4885         int ret;
4886         int err;
4887
4888         if (!capable(CAP_SYS_ADMIN))
4889                 return -EPERM;
4890
4891         ret = mnt_want_write_file(file);
4892         if (ret)
4893                 return ret;
4894
4895         sa = memdup_user(arg, sizeof(*sa));
4896         if (IS_ERR(sa)) {
4897                 ret = PTR_ERR(sa);
4898                 goto drop_write;
4899         }
4900
4901         trans = btrfs_join_transaction(root);
4902         if (IS_ERR(trans)) {
4903                 ret = PTR_ERR(trans);
4904                 goto out;
4905         }
4906
4907         /* FIXME: check if the IDs really exist */
4908         if (sa->assign) {
4909                 ret = btrfs_add_qgroup_relation(trans, root->fs_info,
4910                                                 sa->src, sa->dst);
4911         } else {
4912                 ret = btrfs_del_qgroup_relation(trans, root->fs_info,
4913                                                 sa->src, sa->dst);
4914         }
4915
4916         /* update qgroup status and info */
4917         err = btrfs_run_qgroups(trans, root->fs_info);
4918         if (err < 0)
4919                 btrfs_handle_fs_error(root->fs_info, err,
4920                             "failed to update qgroup status and info");
4921         err = btrfs_end_transaction(trans, root);
4922         if (err && !ret)
4923                 ret = err;
4924
4925 out:
4926         kfree(sa);
4927 drop_write:
4928         mnt_drop_write_file(file);
4929         return ret;
4930 }
4931
4932 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4933 {
4934         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4935         struct btrfs_ioctl_qgroup_create_args *sa;
4936         struct btrfs_trans_handle *trans;
4937         int ret;
4938         int err;
4939
4940         if (!capable(CAP_SYS_ADMIN))
4941                 return -EPERM;
4942
4943         ret = mnt_want_write_file(file);
4944         if (ret)
4945                 return ret;
4946
4947         sa = memdup_user(arg, sizeof(*sa));
4948         if (IS_ERR(sa)) {
4949                 ret = PTR_ERR(sa);
4950                 goto drop_write;
4951         }
4952
4953         if (!sa->qgroupid) {
4954                 ret = -EINVAL;
4955                 goto out;
4956         }
4957
4958         trans = btrfs_join_transaction(root);
4959         if (IS_ERR(trans)) {
4960                 ret = PTR_ERR(trans);
4961                 goto out;
4962         }
4963
4964         /* FIXME: check if the IDs really exist */
4965         if (sa->create) {
4966                 ret = btrfs_create_qgroup(trans, root->fs_info, sa->qgroupid);
4967         } else {
4968                 ret = btrfs_remove_qgroup(trans, root->fs_info, sa->qgroupid);
4969         }
4970
4971         err = btrfs_end_transaction(trans, root);
4972         if (err && !ret)
4973                 ret = err;
4974
4975 out:
4976         kfree(sa);
4977 drop_write:
4978         mnt_drop_write_file(file);
4979         return ret;
4980 }
4981
4982 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4983 {
4984         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4985         struct btrfs_ioctl_qgroup_limit_args *sa;
4986         struct btrfs_trans_handle *trans;
4987         int ret;
4988         int err;
4989         u64 qgroupid;
4990
4991         if (!capable(CAP_SYS_ADMIN))
4992                 return -EPERM;
4993
4994         ret = mnt_want_write_file(file);
4995         if (ret)
4996                 return ret;
4997
4998         sa = memdup_user(arg, sizeof(*sa));
4999         if (IS_ERR(sa)) {
5000                 ret = PTR_ERR(sa);
5001                 goto drop_write;
5002         }
5003
5004         trans = btrfs_join_transaction(root);
5005         if (IS_ERR(trans)) {
5006                 ret = PTR_ERR(trans);
5007                 goto out;
5008         }
5009
5010         qgroupid = sa->qgroupid;
5011         if (!qgroupid) {
5012                 /* take the current subvol as qgroup */
5013                 qgroupid = root->root_key.objectid;
5014         }
5015
5016         /* FIXME: check if the IDs really exist */
5017         ret = btrfs_limit_qgroup(trans, root->fs_info, qgroupid, &sa->lim);
5018
5019         err = btrfs_end_transaction(trans, root);
5020         if (err && !ret)
5021                 ret = err;
5022
5023 out:
5024         kfree(sa);
5025 drop_write:
5026         mnt_drop_write_file(file);
5027         return ret;
5028 }
5029
5030 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
5031 {
5032         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5033         struct btrfs_ioctl_quota_rescan_args *qsa;
5034         int ret;
5035
5036         if (!capable(CAP_SYS_ADMIN))
5037                 return -EPERM;
5038
5039         ret = mnt_want_write_file(file);
5040         if (ret)
5041                 return ret;
5042
5043         qsa = memdup_user(arg, sizeof(*qsa));
5044         if (IS_ERR(qsa)) {
5045                 ret = PTR_ERR(qsa);
5046                 goto drop_write;
5047         }
5048
5049         if (qsa->flags) {
5050                 ret = -EINVAL;
5051                 goto out;
5052         }
5053
5054         ret = btrfs_qgroup_rescan(root->fs_info);
5055
5056 out:
5057         kfree(qsa);
5058 drop_write:
5059         mnt_drop_write_file(file);
5060         return ret;
5061 }
5062
5063 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
5064 {
5065         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5066         struct btrfs_ioctl_quota_rescan_args *qsa;
5067         int ret = 0;
5068
5069         if (!capable(CAP_SYS_ADMIN))
5070                 return -EPERM;
5071
5072         qsa = kzalloc(sizeof(*qsa), GFP_KERNEL);
5073         if (!qsa)
5074                 return -ENOMEM;
5075
5076         if (root->fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
5077                 qsa->flags = 1;
5078                 qsa->progress = root->fs_info->qgroup_rescan_progress.objectid;
5079         }
5080
5081         if (copy_to_user(arg, qsa, sizeof(*qsa)))
5082                 ret = -EFAULT;
5083
5084         kfree(qsa);
5085         return ret;
5086 }
5087
5088 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
5089 {
5090         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5091
5092         if (!capable(CAP_SYS_ADMIN))
5093                 return -EPERM;
5094
5095         return btrfs_qgroup_wait_for_completion(root->fs_info);
5096 }
5097
5098 static long _btrfs_ioctl_set_received_subvol(struct file *file,
5099                                             struct btrfs_ioctl_received_subvol_args *sa)
5100 {
5101         struct inode *inode = file_inode(file);
5102         struct btrfs_root *root = BTRFS_I(inode)->root;
5103         struct btrfs_root_item *root_item = &root->root_item;
5104         struct btrfs_trans_handle *trans;
5105         struct timespec ct = current_fs_time(inode->i_sb);
5106         int ret = 0;
5107         int received_uuid_changed;
5108
5109         if (!inode_owner_or_capable(inode))
5110                 return -EPERM;
5111
5112         ret = mnt_want_write_file(file);
5113         if (ret < 0)
5114                 return ret;
5115
5116         down_write(&root->fs_info->subvol_sem);
5117
5118         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
5119                 ret = -EINVAL;
5120                 goto out;
5121         }
5122
5123         if (btrfs_root_readonly(root)) {
5124                 ret = -EROFS;
5125                 goto out;
5126         }
5127
5128         /*
5129          * 1 - root item
5130          * 2 - uuid items (received uuid + subvol uuid)
5131          */
5132         trans = btrfs_start_transaction(root, 3);
5133         if (IS_ERR(trans)) {
5134                 ret = PTR_ERR(trans);
5135                 trans = NULL;
5136                 goto out;
5137         }
5138
5139         sa->rtransid = trans->transid;
5140         sa->rtime.sec = ct.tv_sec;
5141         sa->rtime.nsec = ct.tv_nsec;
5142
5143         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
5144                                        BTRFS_UUID_SIZE);
5145         if (received_uuid_changed &&
5146             !btrfs_is_empty_uuid(root_item->received_uuid))
5147                 btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
5148                                     root_item->received_uuid,
5149                                     BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5150                                     root->root_key.objectid);
5151         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
5152         btrfs_set_root_stransid(root_item, sa->stransid);
5153         btrfs_set_root_rtransid(root_item, sa->rtransid);
5154         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
5155         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
5156         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
5157         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
5158
5159         ret = btrfs_update_root(trans, root->fs_info->tree_root,
5160                                 &root->root_key, &root->root_item);
5161         if (ret < 0) {
5162                 btrfs_end_transaction(trans, root);
5163                 goto out;
5164         }
5165         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
5166                 ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
5167                                           sa->uuid,
5168                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5169                                           root->root_key.objectid);
5170                 if (ret < 0 && ret != -EEXIST) {
5171                         btrfs_abort_transaction(trans, root, ret);
5172                         goto out;
5173                 }
5174         }
5175         ret = btrfs_commit_transaction(trans, root);
5176         if (ret < 0) {
5177                 btrfs_abort_transaction(trans, root, ret);
5178                 goto out;
5179         }
5180
5181 out:
5182         up_write(&root->fs_info->subvol_sem);
5183         mnt_drop_write_file(file);
5184         return ret;
5185 }
5186
5187 #ifdef CONFIG_64BIT
5188 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
5189                                                 void __user *arg)
5190 {
5191         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
5192         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
5193         int ret = 0;
5194
5195         args32 = memdup_user(arg, sizeof(*args32));
5196         if (IS_ERR(args32)) {
5197                 ret = PTR_ERR(args32);
5198                 args32 = NULL;
5199                 goto out;
5200         }
5201
5202         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
5203         if (!args64) {
5204                 ret = -ENOMEM;
5205                 goto out;
5206         }
5207
5208         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
5209         args64->stransid = args32->stransid;
5210         args64->rtransid = args32->rtransid;
5211         args64->stime.sec = args32->stime.sec;
5212         args64->stime.nsec = args32->stime.nsec;
5213         args64->rtime.sec = args32->rtime.sec;
5214         args64->rtime.nsec = args32->rtime.nsec;
5215         args64->flags = args32->flags;
5216
5217         ret = _btrfs_ioctl_set_received_subvol(file, args64);
5218         if (ret)
5219                 goto out;
5220
5221         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
5222         args32->stransid = args64->stransid;
5223         args32->rtransid = args64->rtransid;
5224         args32->stime.sec = args64->stime.sec;
5225         args32->stime.nsec = args64->stime.nsec;
5226         args32->rtime.sec = args64->rtime.sec;
5227         args32->rtime.nsec = args64->rtime.nsec;
5228         args32->flags = args64->flags;
5229
5230         ret = copy_to_user(arg, args32, sizeof(*args32));
5231         if (ret)
5232                 ret = -EFAULT;
5233
5234 out:
5235         kfree(args32);
5236         kfree(args64);
5237         return ret;
5238 }
5239 #endif
5240
5241 static long btrfs_ioctl_set_received_subvol(struct file *file,
5242                                             void __user *arg)
5243 {
5244         struct btrfs_ioctl_received_subvol_args *sa = NULL;
5245         int ret = 0;
5246
5247         sa = memdup_user(arg, sizeof(*sa));
5248         if (IS_ERR(sa)) {
5249                 ret = PTR_ERR(sa);
5250                 sa = NULL;
5251                 goto out;
5252         }
5253
5254         ret = _btrfs_ioctl_set_received_subvol(file, sa);
5255
5256         if (ret)
5257                 goto out;
5258
5259         ret = copy_to_user(arg, sa, sizeof(*sa));
5260         if (ret)
5261                 ret = -EFAULT;
5262
5263 out:
5264         kfree(sa);
5265         return ret;
5266 }
5267
5268 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
5269 {
5270         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5271         size_t len;
5272         int ret;
5273         char label[BTRFS_LABEL_SIZE];
5274
5275         spin_lock(&root->fs_info->super_lock);
5276         memcpy(label, root->fs_info->super_copy->label, BTRFS_LABEL_SIZE);
5277         spin_unlock(&root->fs_info->super_lock);
5278
5279         len = strnlen(label, BTRFS_LABEL_SIZE);
5280
5281         if (len == BTRFS_LABEL_SIZE) {
5282                 btrfs_warn(root->fs_info,
5283                         "label is too long, return the first %zu bytes", --len);
5284         }
5285
5286         ret = copy_to_user(arg, label, len);
5287
5288         return ret ? -EFAULT : 0;
5289 }
5290
5291 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
5292 {
5293         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5294         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5295         struct btrfs_trans_handle *trans;
5296         char label[BTRFS_LABEL_SIZE];
5297         int ret;
5298
5299         if (!capable(CAP_SYS_ADMIN))
5300                 return -EPERM;
5301
5302         if (copy_from_user(label, arg, sizeof(label)))
5303                 return -EFAULT;
5304
5305         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
5306                 btrfs_err(root->fs_info, "unable to set label with more than %d bytes",
5307                        BTRFS_LABEL_SIZE - 1);
5308                 return -EINVAL;
5309         }
5310
5311         ret = mnt_want_write_file(file);
5312         if (ret)
5313                 return ret;
5314
5315         trans = btrfs_start_transaction(root, 0);
5316         if (IS_ERR(trans)) {
5317                 ret = PTR_ERR(trans);
5318                 goto out_unlock;
5319         }
5320
5321         spin_lock(&root->fs_info->super_lock);
5322         strcpy(super_block->label, label);
5323         spin_unlock(&root->fs_info->super_lock);
5324         ret = btrfs_commit_transaction(trans, root);
5325
5326 out_unlock:
5327         mnt_drop_write_file(file);
5328         return ret;
5329 }
5330
5331 #define INIT_FEATURE_FLAGS(suffix) \
5332         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
5333           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
5334           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
5335
5336 int btrfs_ioctl_get_supported_features(void __user *arg)
5337 {
5338         static const struct btrfs_ioctl_feature_flags features[3] = {
5339                 INIT_FEATURE_FLAGS(SUPP),
5340                 INIT_FEATURE_FLAGS(SAFE_SET),
5341                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
5342         };
5343
5344         if (copy_to_user(arg, &features, sizeof(features)))
5345                 return -EFAULT;
5346
5347         return 0;
5348 }
5349
5350 static int btrfs_ioctl_get_features(struct file *file, void __user *arg)
5351 {
5352         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5353         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5354         struct btrfs_ioctl_feature_flags features;
5355
5356         features.compat_flags = btrfs_super_compat_flags(super_block);
5357         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5358         features.incompat_flags = btrfs_super_incompat_flags(super_block);
5359
5360         if (copy_to_user(arg, &features, sizeof(features)))
5361                 return -EFAULT;
5362
5363         return 0;
5364 }
5365
5366 static int check_feature_bits(struct btrfs_root *root,
5367                               enum btrfs_feature_set set,
5368                               u64 change_mask, u64 flags, u64 supported_flags,
5369                               u64 safe_set, u64 safe_clear)
5370 {
5371         const char *type = btrfs_feature_set_names[set];
5372         char *names;
5373         u64 disallowed, unsupported;
5374         u64 set_mask = flags & change_mask;
5375         u64 clear_mask = ~flags & change_mask;
5376
5377         unsupported = set_mask & ~supported_flags;
5378         if (unsupported) {
5379                 names = btrfs_printable_features(set, unsupported);
5380                 if (names) {
5381                         btrfs_warn(root->fs_info,
5382                            "this kernel does not support the %s feature bit%s",
5383                            names, strchr(names, ',') ? "s" : "");
5384                         kfree(names);
5385                 } else
5386                         btrfs_warn(root->fs_info,
5387                            "this kernel does not support %s bits 0x%llx",
5388                            type, unsupported);
5389                 return -EOPNOTSUPP;
5390         }
5391
5392         disallowed = set_mask & ~safe_set;
5393         if (disallowed) {
5394                 names = btrfs_printable_features(set, disallowed);
5395                 if (names) {
5396                         btrfs_warn(root->fs_info,
5397                            "can't set the %s feature bit%s while mounted",
5398                            names, strchr(names, ',') ? "s" : "");
5399                         kfree(names);
5400                 } else
5401                         btrfs_warn(root->fs_info,
5402                            "can't set %s bits 0x%llx while mounted",
5403                            type, disallowed);
5404                 return -EPERM;
5405         }
5406
5407         disallowed = clear_mask & ~safe_clear;
5408         if (disallowed) {
5409                 names = btrfs_printable_features(set, disallowed);
5410                 if (names) {
5411                         btrfs_warn(root->fs_info,
5412                            "can't clear the %s feature bit%s while mounted",
5413                            names, strchr(names, ',') ? "s" : "");
5414                         kfree(names);
5415                 } else
5416                         btrfs_warn(root->fs_info,
5417                            "can't clear %s bits 0x%llx while mounted",
5418                            type, disallowed);
5419                 return -EPERM;
5420         }
5421
5422         return 0;
5423 }
5424
5425 #define check_feature(root, change_mask, flags, mask_base)      \
5426 check_feature_bits(root, FEAT_##mask_base, change_mask, flags,  \
5427                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
5428                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
5429                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5430
5431 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5432 {
5433         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5434         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5435         struct btrfs_ioctl_feature_flags flags[2];
5436         struct btrfs_trans_handle *trans;
5437         u64 newflags;
5438         int ret;
5439
5440         if (!capable(CAP_SYS_ADMIN))
5441                 return -EPERM;
5442
5443         if (copy_from_user(flags, arg, sizeof(flags)))
5444                 return -EFAULT;
5445
5446         /* Nothing to do */
5447         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5448             !flags[0].incompat_flags)
5449                 return 0;
5450
5451         ret = check_feature(root, flags[0].compat_flags,
5452                             flags[1].compat_flags, COMPAT);
5453         if (ret)
5454                 return ret;
5455
5456         ret = check_feature(root, flags[0].compat_ro_flags,
5457                             flags[1].compat_ro_flags, COMPAT_RO);
5458         if (ret)
5459                 return ret;
5460
5461         ret = check_feature(root, flags[0].incompat_flags,
5462                             flags[1].incompat_flags, INCOMPAT);
5463         if (ret)
5464                 return ret;
5465
5466         ret = mnt_want_write_file(file);
5467         if (ret)
5468                 return ret;
5469
5470         trans = btrfs_start_transaction(root, 0);
5471         if (IS_ERR(trans)) {
5472                 ret = PTR_ERR(trans);
5473                 goto out_drop_write;
5474         }
5475
5476         spin_lock(&root->fs_info->super_lock);
5477         newflags = btrfs_super_compat_flags(super_block);
5478         newflags |= flags[0].compat_flags & flags[1].compat_flags;
5479         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5480         btrfs_set_super_compat_flags(super_block, newflags);
5481
5482         newflags = btrfs_super_compat_ro_flags(super_block);
5483         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5484         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5485         btrfs_set_super_compat_ro_flags(super_block, newflags);
5486
5487         newflags = btrfs_super_incompat_flags(super_block);
5488         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5489         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5490         btrfs_set_super_incompat_flags(super_block, newflags);
5491         spin_unlock(&root->fs_info->super_lock);
5492
5493         ret = btrfs_commit_transaction(trans, root);
5494 out_drop_write:
5495         mnt_drop_write_file(file);
5496
5497         return ret;
5498 }
5499
5500 long btrfs_ioctl(struct file *file, unsigned int
5501                 cmd, unsigned long arg)
5502 {
5503         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5504         void __user *argp = (void __user *)arg;
5505
5506         switch (cmd) {
5507         case FS_IOC_GETFLAGS:
5508                 return btrfs_ioctl_getflags(file, argp);
5509         case FS_IOC_SETFLAGS:
5510                 return btrfs_ioctl_setflags(file, argp);
5511         case FS_IOC_GETVERSION:
5512                 return btrfs_ioctl_getversion(file, argp);
5513         case FITRIM:
5514                 return btrfs_ioctl_fitrim(file, argp);
5515         case BTRFS_IOC_SNAP_CREATE:
5516                 return btrfs_ioctl_snap_create(file, argp, 0);
5517         case BTRFS_IOC_SNAP_CREATE_V2:
5518                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
5519         case BTRFS_IOC_SUBVOL_CREATE:
5520                 return btrfs_ioctl_snap_create(file, argp, 1);
5521         case BTRFS_IOC_SUBVOL_CREATE_V2:
5522                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
5523         case BTRFS_IOC_SNAP_DESTROY:
5524                 return btrfs_ioctl_snap_destroy(file, argp);
5525         case BTRFS_IOC_SUBVOL_GETFLAGS:
5526                 return btrfs_ioctl_subvol_getflags(file, argp);
5527         case BTRFS_IOC_SUBVOL_SETFLAGS:
5528                 return btrfs_ioctl_subvol_setflags(file, argp);
5529         case BTRFS_IOC_DEFAULT_SUBVOL:
5530                 return btrfs_ioctl_default_subvol(file, argp);
5531         case BTRFS_IOC_DEFRAG:
5532                 return btrfs_ioctl_defrag(file, NULL);
5533         case BTRFS_IOC_DEFRAG_RANGE:
5534                 return btrfs_ioctl_defrag(file, argp);
5535         case BTRFS_IOC_RESIZE:
5536                 return btrfs_ioctl_resize(file, argp);
5537         case BTRFS_IOC_ADD_DEV:
5538                 return btrfs_ioctl_add_dev(root, argp);
5539         case BTRFS_IOC_RM_DEV:
5540                 return btrfs_ioctl_rm_dev(file, argp);
5541         case BTRFS_IOC_RM_DEV_V2:
5542                 return btrfs_ioctl_rm_dev_v2(file, argp);
5543         case BTRFS_IOC_FS_INFO:
5544                 return btrfs_ioctl_fs_info(root, argp);
5545         case BTRFS_IOC_DEV_INFO:
5546                 return btrfs_ioctl_dev_info(root, argp);
5547         case BTRFS_IOC_BALANCE:
5548                 return btrfs_ioctl_balance(file, NULL);
5549         case BTRFS_IOC_TRANS_START:
5550                 return btrfs_ioctl_trans_start(file);
5551         case BTRFS_IOC_TRANS_END:
5552                 return btrfs_ioctl_trans_end(file);
5553         case BTRFS_IOC_TREE_SEARCH:
5554                 return btrfs_ioctl_tree_search(file, argp);
5555         case BTRFS_IOC_TREE_SEARCH_V2:
5556                 return btrfs_ioctl_tree_search_v2(file, argp);
5557         case BTRFS_IOC_INO_LOOKUP:
5558                 return btrfs_ioctl_ino_lookup(file, argp);
5559         case BTRFS_IOC_INO_PATHS:
5560                 return btrfs_ioctl_ino_to_path(root, argp);
5561         case BTRFS_IOC_LOGICAL_INO:
5562                 return btrfs_ioctl_logical_to_ino(root, argp);
5563         case BTRFS_IOC_SPACE_INFO:
5564                 return btrfs_ioctl_space_info(root, argp);
5565         case BTRFS_IOC_SYNC: {
5566                 int ret;
5567
5568                 ret = btrfs_start_delalloc_roots(root->fs_info, 0, -1);
5569                 if (ret)
5570                         return ret;
5571                 ret = btrfs_sync_fs(file_inode(file)->i_sb, 1);
5572                 /*
5573                  * The transaction thread may want to do more work,
5574                  * namely it pokes the cleaner kthread that will start
5575                  * processing uncleaned subvols.
5576                  */
5577                 wake_up_process(root->fs_info->transaction_kthread);
5578                 return ret;
5579         }
5580         case BTRFS_IOC_START_SYNC:
5581                 return btrfs_ioctl_start_sync(root, argp);
5582         case BTRFS_IOC_WAIT_SYNC:
5583                 return btrfs_ioctl_wait_sync(root, argp);
5584         case BTRFS_IOC_SCRUB:
5585                 return btrfs_ioctl_scrub(file, argp);
5586         case BTRFS_IOC_SCRUB_CANCEL:
5587                 return btrfs_ioctl_scrub_cancel(root, argp);
5588         case BTRFS_IOC_SCRUB_PROGRESS:
5589                 return btrfs_ioctl_scrub_progress(root, argp);
5590         case BTRFS_IOC_BALANCE_V2:
5591                 return btrfs_ioctl_balance(file, argp);
5592         case BTRFS_IOC_BALANCE_CTL:
5593                 return btrfs_ioctl_balance_ctl(root, arg);
5594         case BTRFS_IOC_BALANCE_PROGRESS:
5595                 return btrfs_ioctl_balance_progress(root, argp);
5596         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5597                 return btrfs_ioctl_set_received_subvol(file, argp);
5598 #ifdef CONFIG_64BIT
5599         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5600                 return btrfs_ioctl_set_received_subvol_32(file, argp);
5601 #endif
5602         case BTRFS_IOC_SEND:
5603                 return btrfs_ioctl_send(file, argp);
5604         case BTRFS_IOC_GET_DEV_STATS:
5605                 return btrfs_ioctl_get_dev_stats(root, argp);
5606         case BTRFS_IOC_QUOTA_CTL:
5607                 return btrfs_ioctl_quota_ctl(file, argp);
5608         case BTRFS_IOC_QGROUP_ASSIGN:
5609                 return btrfs_ioctl_qgroup_assign(file, argp);
5610         case BTRFS_IOC_QGROUP_CREATE:
5611                 return btrfs_ioctl_qgroup_create(file, argp);
5612         case BTRFS_IOC_QGROUP_LIMIT:
5613                 return btrfs_ioctl_qgroup_limit(file, argp);
5614         case BTRFS_IOC_QUOTA_RESCAN:
5615                 return btrfs_ioctl_quota_rescan(file, argp);
5616         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5617                 return btrfs_ioctl_quota_rescan_status(file, argp);
5618         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5619                 return btrfs_ioctl_quota_rescan_wait(file, argp);
5620         case BTRFS_IOC_DEV_REPLACE:
5621                 return btrfs_ioctl_dev_replace(root, argp);
5622         case BTRFS_IOC_GET_FSLABEL:
5623                 return btrfs_ioctl_get_fslabel(file, argp);
5624         case BTRFS_IOC_SET_FSLABEL:
5625                 return btrfs_ioctl_set_fslabel(file, argp);
5626         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5627                 return btrfs_ioctl_get_supported_features(argp);
5628         case BTRFS_IOC_GET_FEATURES:
5629                 return btrfs_ioctl_get_features(file, argp);
5630         case BTRFS_IOC_SET_FEATURES:
5631                 return btrfs_ioctl_set_features(file, argp);
5632         }
5633
5634         return -ENOTTY;
5635 }
5636
5637 #ifdef CONFIG_COMPAT
5638 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
5639 {
5640         switch (cmd) {
5641         case FS_IOC32_GETFLAGS:
5642                 cmd = FS_IOC_GETFLAGS;
5643                 break;
5644         case FS_IOC32_SETFLAGS:
5645                 cmd = FS_IOC_SETFLAGS;
5646                 break;
5647         case FS_IOC32_GETVERSION:
5648                 cmd = FS_IOC_GETVERSION;
5649                 break;
5650         default:
5651                 return -ENOIOCTLCMD;
5652         }
5653
5654         return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
5655 }
5656 #endif