fs/btrfs/super.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2007 Oracle.  All rights reserved.
   4  */
   5
   6 #include <linux/blkdev.h>
   7 #include <linux/module.h>
   8 #include <linux/fs.h>
   9 #include <linux/pagemap.h>
  10 #include <linux/highmem.h>
  11 #include <linux/time.h>
  12 #include <linux/init.h>
  13 #include <linux/seq_file.h>
  14 #include <linux/string.h>
  15 #include <linux/backing-dev.h>
  16 #include <linux/mount.h>
  17 #include <linux/writeback.h>
  18 #include <linux/statfs.h>
  19 #include <linux/compat.h>
  20 #include <linux/parser.h>
  21 #include <linux/ctype.h>
  22 #include <linux/namei.h>
  23 #include <linux/miscdevice.h>
  24 #include <linux/magic.h>
  25 #include <linux/slab.h>
  26 #include <linux/ratelimit.h>
  27 #include <linux/crc32c.h>
  28 #include <linux/btrfs.h>
  29 #include <linux/security.h>
  30 #include <linux/fs_parser.h>
  31 #include "messages.h"
  32 #include "delayed-inode.h"
  33 #include "ctree.h"
  34 #include "disk-io.h"
  35 #include "transaction.h"
  36 #include "btrfs_inode.h"
  37 #include "direct-io.h"
  38 #include "props.h"
  39 #include "xattr.h"
  40 #include "bio.h"
  41 #include "export.h"
  42 #include "compression.h"
  43 #include "dev-replace.h"
  44 #include "free-space-cache.h"
  45 #include "backref.h"
  46 #include "space-info.h"
  47 #include "sysfs.h"
  48 #include "zoned.h"
  49 #include "tests/btrfs-tests.h"
  50 #include "block-group.h"
  51 #include "discard.h"
  52 #include "qgroup.h"
  53 #include "raid56.h"
  54 #include "fs.h"
  55 #include "accessors.h"
  56 #include "defrag.h"
  57 #include "dir-item.h"
  58 #include "ioctl.h"
  59 #include "scrub.h"
  60 #include "verity.h"
  61 #include "super.h"
  62 #include "extent-tree.h"
  63 #define CREATE_TRACE_POINTS
  64 #include <trace/events/btrfs.h>
  65
  66 static const struct super_operations btrfs_super_ops;
  67 static struct file_system_type btrfs_fs_type;
  68
  69 static void btrfs_put_super(struct super_block *sb)
  70 {
  71         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
  72
  73         btrfs_info(fs_info, "last unmount of filesystem %pU", fs_info->fs_devices->fsid);
  74         close_ctree(fs_info);
  75 }
  76
  77 /* Store the mount options related information. */
  78 struct btrfs_fs_context {
  79         char *subvol_name;
  80         u64 subvol_objectid;
  81         u64 max_inline;
  82         u32 commit_interval;
  83         u32 metadata_ratio;
  84         u32 thread_pool_size;
  85         unsigned long long mount_opt;
  86         unsigned long compress_type:4;
  87         int compress_level;
  88         refcount_t refs;
  89 };
  90
  91 enum {
  92         Opt_acl,
  93         Opt_clear_cache,
  94         Opt_commit_interval,
  95         Opt_compress,
  96         Opt_compress_force,
  97         Opt_compress_force_type,
  98         Opt_compress_type,
  99         Opt_degraded,
 100         Opt_device,
 101         Opt_fatal_errors,
 102         Opt_flushoncommit,
 103         Opt_max_inline,
 104         Opt_barrier,
 105         Opt_datacow,
 106         Opt_datasum,
 107         Opt_defrag,
 108         Opt_discard,
 109         Opt_discard_mode,
 110         Opt_ratio,
 111         Opt_rescan_uuid_tree,
 112         Opt_skip_balance,
 113         Opt_space_cache,
 114         Opt_space_cache_version,
 115         Opt_ssd,
 116         Opt_ssd_spread,
 117         Opt_subvol,
 118         Opt_subvol_empty,
 119         Opt_subvolid,
 120         Opt_thread_pool,
 121         Opt_treelog,
 122         Opt_user_subvol_rm_allowed,
 123         Opt_norecovery,
 124
 125         /* Rescue options */
 126         Opt_rescue,
 127         Opt_usebackuproot,
 128
 129         /* Debugging options */
 130         Opt_enospc_debug,
 131 #ifdef CONFIG_BTRFS_DEBUG
 132         Opt_fragment, Opt_fragment_data, Opt_fragment_metadata, Opt_fragment_all,
 133 #endif
 134 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
 135         Opt_ref_verify,
 136 #endif
 137         Opt_err,
 138 };
 139
 140 enum {
 141         Opt_fatal_errors_panic,
 142         Opt_fatal_errors_bug,
 143 };
 144
 145 static const struct constant_table btrfs_parameter_fatal_errors[] = {
 146         { "panic", Opt_fatal_errors_panic },
 147         { "bug", Opt_fatal_errors_bug },
 148         {}
 149 };
 150
 151 enum {
 152         Opt_discard_sync,
 153         Opt_discard_async,
 154 };
 155
 156 static const struct constant_table btrfs_parameter_discard[] = {
 157         { "sync", Opt_discard_sync },
 158         { "async", Opt_discard_async },
 159         {}
 160 };
 161
 162 enum {
 163         Opt_space_cache_v1,
 164         Opt_space_cache_v2,
 165 };
 166
 167 static const struct constant_table btrfs_parameter_space_cache[] = {
 168         { "v1", Opt_space_cache_v1 },
 169         { "v2", Opt_space_cache_v2 },
 170         {}
 171 };
 172
 173 enum {
 174         Opt_rescue_usebackuproot,
 175         Opt_rescue_nologreplay,
 176         Opt_rescue_ignorebadroots,
 177         Opt_rescue_ignoredatacsums,
 178         Opt_rescue_ignoremetacsums,
 179         Opt_rescue_ignoresuperflags,
 180         Opt_rescue_parameter_all,
 181 };
 182
 183 static const struct constant_table btrfs_parameter_rescue[] = {
 184         { "usebackuproot", Opt_rescue_usebackuproot },
 185         { "nologreplay", Opt_rescue_nologreplay },
 186         { "ignorebadroots", Opt_rescue_ignorebadroots },
 187         { "ibadroots", Opt_rescue_ignorebadroots },
 188         { "ignoredatacsums", Opt_rescue_ignoredatacsums },
 189         { "ignoremetacsums", Opt_rescue_ignoremetacsums},
 190         { "ignoresuperflags", Opt_rescue_ignoresuperflags},
 191         { "idatacsums", Opt_rescue_ignoredatacsums },
 192         { "imetacsums", Opt_rescue_ignoremetacsums},
 193         { "isuperflags", Opt_rescue_ignoresuperflags},
 194         { "all", Opt_rescue_parameter_all },
 195         {}
 196 };
 197
 198 #ifdef CONFIG_BTRFS_DEBUG
 199 enum {
 200         Opt_fragment_parameter_data,
 201         Opt_fragment_parameter_metadata,
 202         Opt_fragment_parameter_all,
 203 };
 204
 205 static const struct constant_table btrfs_parameter_fragment[] = {
 206         { "data", Opt_fragment_parameter_data },
 207         { "metadata", Opt_fragment_parameter_metadata },
 208         { "all", Opt_fragment_parameter_all },
 209         {}
 210 };
 211 #endif
 212
 213 static const struct fs_parameter_spec btrfs_fs_parameters[] = {
 214         fsparam_flag_no("acl", Opt_acl),
 215         fsparam_flag_no("autodefrag", Opt_defrag),
 216         fsparam_flag_no("barrier", Opt_barrier),
 217         fsparam_flag("clear_cache", Opt_clear_cache),
 218         fsparam_u32("commit", Opt_commit_interval),
 219         fsparam_flag("compress", Opt_compress),
 220         fsparam_string("compress", Opt_compress_type),
 221         fsparam_flag("compress-force", Opt_compress_force),
 222         fsparam_string("compress-force", Opt_compress_force_type),
 223         fsparam_flag_no("datacow", Opt_datacow),
 224         fsparam_flag_no("datasum", Opt_datasum),
 225         fsparam_flag("degraded", Opt_degraded),
 226         fsparam_string("device", Opt_device),
 227         fsparam_flag_no("discard", Opt_discard),
 228         fsparam_enum("discard", Opt_discard_mode, btrfs_parameter_discard),
 229         fsparam_enum("fatal_errors", Opt_fatal_errors, btrfs_parameter_fatal_errors),
 230         fsparam_flag_no("flushoncommit", Opt_flushoncommit),
 231         fsparam_string("max_inline", Opt_max_inline),
 232         fsparam_u32("metadata_ratio", Opt_ratio),
 233         fsparam_flag("rescan_uuid_tree", Opt_rescan_uuid_tree),
 234         fsparam_flag("skip_balance", Opt_skip_balance),
 235         fsparam_flag_no("space_cache", Opt_space_cache),
 236         fsparam_enum("space_cache", Opt_space_cache_version, btrfs_parameter_space_cache),
 237         fsparam_flag_no("ssd", Opt_ssd),
 238         fsparam_flag_no("ssd_spread", Opt_ssd_spread),
 239         fsparam_string("subvol", Opt_subvol),
 240         fsparam_flag("subvol=", Opt_subvol_empty),
 241         fsparam_u64("subvolid", Opt_subvolid),
 242         fsparam_u32("thread_pool", Opt_thread_pool),
 243         fsparam_flag_no("treelog", Opt_treelog),
 244         fsparam_flag("user_subvol_rm_allowed", Opt_user_subvol_rm_allowed),
 245
 246         /* Rescue options. */
 247         fsparam_enum("rescue", Opt_rescue, btrfs_parameter_rescue),
 248         /* Deprecated, with alias rescue=usebackuproot */
 249         __fsparam(NULL, "usebackuproot", Opt_usebackuproot, fs_param_deprecated, NULL),
 250         /* For compatibility only, alias for "rescue=nologreplay". */
 251         fsparam_flag("norecovery", Opt_norecovery),
 252
 253         /* Debugging options. */
 254         fsparam_flag_no("enospc_debug", Opt_enospc_debug),
 255 #ifdef CONFIG_BTRFS_DEBUG
 256         fsparam_enum("fragment", Opt_fragment, btrfs_parameter_fragment),
 257 #endif
 258 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
 259         fsparam_flag("ref_verify", Opt_ref_verify),
 260 #endif
 261         {}
 262 };
 263
 264 /* No support for restricting writes to btrfs devices yet... */
 265 static inline blk_mode_t btrfs_open_mode(struct fs_context *fc)
 266 {
 267         return sb_open_mode(fc->sb_flags) & ~BLK_OPEN_RESTRICT_WRITES;
 268 }
 269
 270 static int btrfs_parse_param(struct fs_context *fc, struct fs_parameter *param)
 271 {
 272         struct btrfs_fs_context *ctx = fc->fs_private;
 273         struct fs_parse_result result;
 274         int opt;
 275
 276         opt = fs_parse(fc, btrfs_fs_parameters, param, &result);
 277         if (opt < 0)
 278                 return opt;
 279
 280         switch (opt) {
 281         case Opt_degraded:
 282                 btrfs_set_opt(ctx->mount_opt, DEGRADED);
 283                 break;
 284         case Opt_subvol_empty:
 285                 /*
 286                  * This exists because we used to allow it on accident, so we're
 287                  * keeping it to maintain ABI.  See 37becec95ac3 ("Btrfs: allow
 288                  * empty subvol= again").
 289                  */
 290                 break;
 291         case Opt_subvol:
 292                 kfree(ctx->subvol_name);
 293                 ctx->subvol_name = kstrdup(param->string, GFP_KERNEL);
 294                 if (!ctx->subvol_name)
 295                         return -ENOMEM;
 296                 break;
 297         case Opt_subvolid:
 298                 ctx->subvol_objectid = result.uint_64;
 299
 300                 /* subvolid=0 means give me the original fs_tree. */
 301                 if (!ctx->subvol_objectid)
 302                         ctx->subvol_objectid = BTRFS_FS_TREE_OBJECTID;
 303                 break;
 304         case Opt_device: {
 305                 struct btrfs_device *device;
 306                 blk_mode_t mode = btrfs_open_mode(fc);
 307
 308                 mutex_lock(&uuid_mutex);
 309                 device = btrfs_scan_one_device(param->string, mode, false);
 310                 mutex_unlock(&uuid_mutex);
 311                 if (IS_ERR(device))
 312                         return PTR_ERR(device);
 313                 break;
 314         }
 315         case Opt_datasum:
 316                 if (result.negated) {
 317                         btrfs_set_opt(ctx->mount_opt, NODATASUM);
 318                 } else {
 319                         btrfs_clear_opt(ctx->mount_opt, NODATACOW);
 320                         btrfs_clear_opt(ctx->mount_opt, NODATASUM);
 321                 }
 322                 break;
 323         case Opt_datacow:
 324                 if (result.negated) {
 325                         btrfs_clear_opt(ctx->mount_opt, COMPRESS);
 326                         btrfs_clear_opt(ctx->mount_opt, FORCE_COMPRESS);
 327                         btrfs_set_opt(ctx->mount_opt, NODATACOW);
 328                         btrfs_set_opt(ctx->mount_opt, NODATASUM);
 329                 } else {
 330                         btrfs_clear_opt(ctx->mount_opt, NODATACOW);
 331                 }
 332                 break;
 333         case Opt_compress_force:
 334         case Opt_compress_force_type:
 335                 btrfs_set_opt(ctx->mount_opt, FORCE_COMPRESS);
 336                 fallthrough;
 337         case Opt_compress:
 338         case Opt_compress_type:
 339                 /*
 340                  * Provide the same semantics as older kernels that don't use fs
 341                  * context, specifying the "compress" option clears
 342                  * "force-compress" without the need to pass
 343                  * "compress-force=[no|none]" before specifying "compress".
 344                  */
 345                 if (opt != Opt_compress_force && opt != Opt_compress_force_type)
 346                         btrfs_clear_opt(ctx->mount_opt, FORCE_COMPRESS);
 347
 348                 if (opt == Opt_compress || opt == Opt_compress_force) {
 349                         ctx->compress_type = BTRFS_COMPRESS_ZLIB;
 350                         ctx->compress_level = BTRFS_ZLIB_DEFAULT_LEVEL;
 351                         btrfs_set_opt(ctx->mount_opt, COMPRESS);
 352                         btrfs_clear_opt(ctx->mount_opt, NODATACOW);
 353                         btrfs_clear_opt(ctx->mount_opt, NODATASUM);
 354                 } else if (strncmp(param->string, "zlib", 4) == 0) {
 355                         ctx->compress_type = BTRFS_COMPRESS_ZLIB;
 356                         ctx->compress_level =
 357                                 btrfs_compress_str2level(BTRFS_COMPRESS_ZLIB,
 358                                                          param->string + 4);
 359                         btrfs_set_opt(ctx->mount_opt, COMPRESS);
 360                         btrfs_clear_opt(ctx->mount_opt, NODATACOW);
 361                         btrfs_clear_opt(ctx->mount_opt, NODATASUM);
 362                 } else if (strncmp(param->string, "lzo", 3) == 0) {
 363                         ctx->compress_type = BTRFS_COMPRESS_LZO;
 364                         ctx->compress_level = 0;
 365                         btrfs_set_opt(ctx->mount_opt, COMPRESS);
 366                         btrfs_clear_opt(ctx->mount_opt, NODATACOW);
 367                         btrfs_clear_opt(ctx->mount_opt, NODATASUM);
 368                 } else if (strncmp(param->string, "zstd", 4) == 0) {
 369                         ctx->compress_type = BTRFS_COMPRESS_ZSTD;
 370                         ctx->compress_level =
 371                                 btrfs_compress_str2level(BTRFS_COMPRESS_ZSTD,
 372                                                          param->string + 4);
 373                         btrfs_set_opt(ctx->mount_opt, COMPRESS);
 374                         btrfs_clear_opt(ctx->mount_opt, NODATACOW);
 375                         btrfs_clear_opt(ctx->mount_opt, NODATASUM);
 376                 } else if (strncmp(param->string, "no", 2) == 0) {
 377                         ctx->compress_level = 0;
 378                         ctx->compress_type = 0;
 379                         btrfs_clear_opt(ctx->mount_opt, COMPRESS);
 380                         btrfs_clear_opt(ctx->mount_opt, FORCE_COMPRESS);
 381                 } else {
 382                         btrfs_err(NULL, "unrecognized compression value %s",
 383                                   param->string);
 384                         return -EINVAL;
 385                 }
 386                 break;
 387         case Opt_ssd:
 388                 if (result.negated) {
 389                         btrfs_set_opt(ctx->mount_opt, NOSSD);
 390                         btrfs_clear_opt(ctx->mount_opt, SSD);
 391                         btrfs_clear_opt(ctx->mount_opt, SSD_SPREAD);
 392                 } else {
 393                         btrfs_set_opt(ctx->mount_opt, SSD);
 394                         btrfs_clear_opt(ctx->mount_opt, NOSSD);
 395                 }
 396                 break;
 397         case Opt_ssd_spread:
 398                 if (result.negated) {
 399                         btrfs_clear_opt(ctx->mount_opt, SSD_SPREAD);
 400                 } else {
 401                         btrfs_set_opt(ctx->mount_opt, SSD);
 402                         btrfs_set_opt(ctx->mount_opt, SSD_SPREAD);
 403                         btrfs_clear_opt(ctx->mount_opt, NOSSD);
 404                 }
 405                 break;
 406         case Opt_barrier:
 407                 if (result.negated)
 408                         btrfs_set_opt(ctx->mount_opt, NOBARRIER);
 409                 else
 410                         btrfs_clear_opt(ctx->mount_opt, NOBARRIER);
 411                 break;
 412         case Opt_thread_pool:
 413                 if (result.uint_32 == 0) {
 414                         btrfs_err(NULL, "invalid value 0 for thread_pool");
 415                         return -EINVAL;
 416                 }
 417                 ctx->thread_pool_size = result.uint_32;
 418                 break;
 419         case Opt_max_inline:
 420                 ctx->max_inline = memparse(param->string, NULL);
 421                 break;
 422         case Opt_acl:
 423                 if (result.negated) {
 424                         fc->sb_flags &= ~SB_POSIXACL;
 425                 } else {
 426 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
 427                         fc->sb_flags |= SB_POSIXACL;
 428 #else
 429                         btrfs_err(NULL, "support for ACL not compiled in");
 430                         return -EINVAL;
 431 #endif
 432                 }
 433                 /*
 434                  * VFS limits the ability to toggle ACL on and off via remount,
 435                  * despite every file system allowing this.  This seems to be
 436                  * an oversight since we all do, but it'll fail if we're
 437                  * remounting.  So don't set the mask here, we'll check it in
 438                  * btrfs_reconfigure and do the toggling ourselves.
 439                  */
 440                 if (fc->purpose != FS_CONTEXT_FOR_RECONFIGURE)
 441                         fc->sb_flags_mask |= SB_POSIXACL;
 442                 break;
 443         case Opt_treelog:
 444                 if (result.negated)
 445                         btrfs_set_opt(ctx->mount_opt, NOTREELOG);
 446                 else
 447                         btrfs_clear_opt(ctx->mount_opt, NOTREELOG);
 448                 break;
 449         case Opt_norecovery:
 450                 btrfs_info(NULL,
 451 "'norecovery' is for compatibility only, recommended to use 'rescue=nologreplay'");
 452                 btrfs_set_opt(ctx->mount_opt, NOLOGREPLAY);
 453                 break;
 454         case Opt_flushoncommit:
 455                 if (result.negated)
 456                         btrfs_clear_opt(ctx->mount_opt, FLUSHONCOMMIT);
 457                 else
 458                         btrfs_set_opt(ctx->mount_opt, FLUSHONCOMMIT);
 459                 break;
 460         case Opt_ratio:
 461                 ctx->metadata_ratio = result.uint_32;
 462                 break;
 463         case Opt_discard:
 464                 if (result.negated) {
 465                         btrfs_clear_opt(ctx->mount_opt, DISCARD_SYNC);
 466                         btrfs_clear_opt(ctx->mount_opt, DISCARD_ASYNC);
 467                         btrfs_set_opt(ctx->mount_opt, NODISCARD);
 468                 } else {
 469                         btrfs_set_opt(ctx->mount_opt, DISCARD_SYNC);
 470                         btrfs_clear_opt(ctx->mount_opt, DISCARD_ASYNC);
 471                 }
 472                 break;
 473         case Opt_discard_mode:
 474                 switch (result.uint_32) {
 475                 case Opt_discard_sync:
 476                         btrfs_clear_opt(ctx->mount_opt, DISCARD_ASYNC);
 477                         btrfs_set_opt(ctx->mount_opt, DISCARD_SYNC);
 478                         break;
 479                 case Opt_discard_async:
 480                         btrfs_clear_opt(ctx->mount_opt, DISCARD_SYNC);
 481                         btrfs_set_opt(ctx->mount_opt, DISCARD_ASYNC);
 482                         break;
 483                 default:
 484                         btrfs_err(NULL, "unrecognized discard mode value %s",
 485                                   param->key);
 486                         return -EINVAL;
 487                 }
 488                 btrfs_clear_opt(ctx->mount_opt, NODISCARD);
 489                 break;
 490         case Opt_space_cache:
 491                 if (result.negated) {
 492                         btrfs_set_opt(ctx->mount_opt, NOSPACECACHE);
 493                         btrfs_clear_opt(ctx->mount_opt, SPACE_CACHE);
 494                         btrfs_clear_opt(ctx->mount_opt, FREE_SPACE_TREE);
 495                 } else {
 496                         btrfs_clear_opt(ctx->mount_opt, FREE_SPACE_TREE);
 497                         btrfs_set_opt(ctx->mount_opt, SPACE_CACHE);
 498                 }
 499                 break;
 500         case Opt_space_cache_version:
 501                 switch (result.uint_32) {
 502                 case Opt_space_cache_v1:
 503                         btrfs_set_opt(ctx->mount_opt, SPACE_CACHE);
 504                         btrfs_clear_opt(ctx->mount_opt, FREE_SPACE_TREE);
 505                         break;
 506                 case Opt_space_cache_v2:
 507                         btrfs_clear_opt(ctx->mount_opt, SPACE_CACHE);
 508                         btrfs_set_opt(ctx->mount_opt, FREE_SPACE_TREE);
 509                         break;
 510                 default:
 511                         btrfs_err(NULL, "unrecognized space_cache value %s",
 512                                   param->key);
 513                         return -EINVAL;
 514                 }
 515                 break;
 516         case Opt_rescan_uuid_tree:
 517                 btrfs_set_opt(ctx->mount_opt, RESCAN_UUID_TREE);
 518                 break;
 519         case Opt_clear_cache:
 520                 btrfs_set_opt(ctx->mount_opt, CLEAR_CACHE);
 521                 break;
 522         case Opt_user_subvol_rm_allowed:
 523                 btrfs_set_opt(ctx->mount_opt, USER_SUBVOL_RM_ALLOWED);
 524                 break;
 525         case Opt_enospc_debug:
 526                 if (result.negated)
 527                         btrfs_clear_opt(ctx->mount_opt, ENOSPC_DEBUG);
 528                 else
 529                         btrfs_set_opt(ctx->mount_opt, ENOSPC_DEBUG);
 530                 break;
 531         case Opt_defrag:
 532                 if (result.negated)
 533                         btrfs_clear_opt(ctx->mount_opt, AUTO_DEFRAG);
 534                 else
 535                         btrfs_set_opt(ctx->mount_opt, AUTO_DEFRAG);
 536                 break;
 537         case Opt_usebackuproot:
 538                 btrfs_warn(NULL,
 539                            "'usebackuproot' is deprecated, use 'rescue=usebackuproot' instead");
 540                 btrfs_set_opt(ctx->mount_opt, USEBACKUPROOT);
 541
 542                 /* If we're loading the backup roots we can't trust the space cache. */
 543                 btrfs_set_opt(ctx->mount_opt, CLEAR_CACHE);
 544                 break;
 545         case Opt_skip_balance:
 546                 btrfs_set_opt(ctx->mount_opt, SKIP_BALANCE);
 547                 break;
 548         case Opt_fatal_errors:
 549                 switch (result.uint_32) {
 550                 case Opt_fatal_errors_panic:
 551                         btrfs_set_opt(ctx->mount_opt, PANIC_ON_FATAL_ERROR);
 552                         break;
 553                 case Opt_fatal_errors_bug:
 554                         btrfs_clear_opt(ctx->mount_opt, PANIC_ON_FATAL_ERROR);
 555                         break;
 556                 default:
 557                         btrfs_err(NULL, "unrecognized fatal_errors value %s",
 558                                   param->key);
 559                         return -EINVAL;
 560                 }
 561                 break;
 562         case Opt_commit_interval:
 563                 ctx->commit_interval = result.uint_32;
 564                 if (ctx->commit_interval > BTRFS_WARNING_COMMIT_INTERVAL) {
 565                         btrfs_warn(NULL, "excessive commit interval %u, use with care",
 566                                    ctx->commit_interval);
 567                 }
 568                 if (ctx->commit_interval == 0)
 569                         ctx->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
 570                 break;
 571         case Opt_rescue:
 572                 switch (result.uint_32) {
 573                 case Opt_rescue_usebackuproot:
 574                         btrfs_set_opt(ctx->mount_opt, USEBACKUPROOT);
 575                         break;
 576                 case Opt_rescue_nologreplay:
 577                         btrfs_set_opt(ctx->mount_opt, NOLOGREPLAY);
 578                         break;
 579                 case Opt_rescue_ignorebadroots:
 580                         btrfs_set_opt(ctx->mount_opt, IGNOREBADROOTS);
 581                         break;
 582                 case Opt_rescue_ignoredatacsums:
 583                         btrfs_set_opt(ctx->mount_opt, IGNOREDATACSUMS);
 584                         break;
 585                 case Opt_rescue_ignoremetacsums:
 586                         btrfs_set_opt(ctx->mount_opt, IGNOREMETACSUMS);
 587                         break;
 588                 case Opt_rescue_ignoresuperflags:
 589                         btrfs_set_opt(ctx->mount_opt, IGNORESUPERFLAGS);
 590                         break;
 591                 case Opt_rescue_parameter_all:
 592                         btrfs_set_opt(ctx->mount_opt, IGNOREDATACSUMS);
 593                         btrfs_set_opt(ctx->mount_opt, IGNOREMETACSUMS);
 594                         btrfs_set_opt(ctx->mount_opt, IGNORESUPERFLAGS);
 595                         btrfs_set_opt(ctx->mount_opt, IGNOREBADROOTS);
 596                         btrfs_set_opt(ctx->mount_opt, NOLOGREPLAY);
 597                         break;
 598                 default:
 599                         btrfs_info(NULL, "unrecognized rescue option '%s'",
 600                                    param->key);
 601                         return -EINVAL;
 602                 }
 603                 break;
 604 #ifdef CONFIG_BTRFS_DEBUG
 605         case Opt_fragment:
 606                 switch (result.uint_32) {
 607                 case Opt_fragment_parameter_all:
 608                         btrfs_set_opt(ctx->mount_opt, FRAGMENT_DATA);
 609                         btrfs_set_opt(ctx->mount_opt, FRAGMENT_METADATA);
 610                         break;
 611                 case Opt_fragment_parameter_metadata:
 612                         btrfs_set_opt(ctx->mount_opt, FRAGMENT_METADATA);
 613                         break;
 614                 case Opt_fragment_parameter_data:
 615                         btrfs_set_opt(ctx->mount_opt, FRAGMENT_DATA);
 616                         break;
 617                 default:
 618                         btrfs_info(NULL, "unrecognized fragment option '%s'",
 619                                    param->key);
 620                         return -EINVAL;
 621                 }
 622                 break;
 623 #endif
 624 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
 625         case Opt_ref_verify:
 626                 btrfs_set_opt(ctx->mount_opt, REF_VERIFY);
 627                 break;
 628 #endif
 629         default:
 630                 btrfs_err(NULL, "unrecognized mount option '%s'", param->key);
 631                 return -EINVAL;
 632         }
 633
 634         return 0;
 635 }
 636
 637 /*
 638  * Some options only have meaning at mount time and shouldn't persist across
 639  * remounts, or be displayed. Clear these at the end of mount and remount code
 640  * paths.
 641  */
 642 static void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info)
 643 {
 644         btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
 645         btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE);
 646         btrfs_clear_opt(fs_info->mount_opt, NOSPACECACHE);
 647 }
 648
 649 static bool check_ro_option(const struct btrfs_fs_info *fs_info,
 650                             unsigned long long mount_opt, unsigned long long opt,
 651                             const char *opt_name)
 652 {
 653         if (mount_opt & opt) {
 654                 btrfs_err(fs_info, "%s must be used with ro mount option",
 655                           opt_name);
 656                 return true;
 657         }
 658         return false;
 659 }
 660
 661 bool btrfs_check_options(const struct btrfs_fs_info *info,
 662                          unsigned long long *mount_opt,
 663                          unsigned long flags)
 664 {
 665         bool ret = true;
 666
 667         if (!(flags & SB_RDONLY) &&
 668             (check_ro_option(info, *mount_opt, BTRFS_MOUNT_NOLOGREPLAY, "nologreplay") ||
 669              check_ro_option(info, *mount_opt, BTRFS_MOUNT_IGNOREBADROOTS, "ignorebadroots") ||
 670              check_ro_option(info, *mount_opt, BTRFS_MOUNT_IGNOREDATACSUMS, "ignoredatacsums") ||
 671              check_ro_option(info, *mount_opt, BTRFS_MOUNT_IGNOREMETACSUMS, "ignoremetacsums") ||
 672              check_ro_option(info, *mount_opt, BTRFS_MOUNT_IGNORESUPERFLAGS, "ignoresuperflags")))
 673                 ret = false;
 674
 675         if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE) &&
 676             !btrfs_raw_test_opt(*mount_opt, FREE_SPACE_TREE) &&
 677             !btrfs_raw_test_opt(*mount_opt, CLEAR_CACHE)) {
 678                 btrfs_err(info, "cannot disable free-space-tree");
 679                 ret = false;
 680         }
 681         if (btrfs_fs_compat_ro(info, BLOCK_GROUP_TREE) &&
 682              !btrfs_raw_test_opt(*mount_opt, FREE_SPACE_TREE)) {
 683                 btrfs_err(info, "cannot disable free-space-tree with block-group-tree feature");
 684                 ret = false;
 685         }
 686
 687         if (btrfs_check_mountopts_zoned(info, mount_opt))
 688                 ret = false;
 689
 690         if (!test_bit(BTRFS_FS_STATE_REMOUNTING, &info->fs_state)) {
 691                 if (btrfs_raw_test_opt(*mount_opt, SPACE_CACHE)) {
 692                         btrfs_info(info, "disk space caching is enabled");
 693                         btrfs_warn(info,
 694 "space cache v1 is being deprecated and will be removed in a future release, please use -o space_cache=v2");
 695                 }
 696                 if (btrfs_raw_test_opt(*mount_opt, FREE_SPACE_TREE))
 697                         btrfs_info(info, "using free-space-tree");
 698         }
 699
 700         return ret;
 701 }
 702
 703 /*
 704  * This is subtle, we only call this during open_ctree().  We need to pre-load
 705  * the mount options with the on-disk settings.  Before the new mount API took
 706  * effect we would do this on mount and remount.  With the new mount API we'll
 707  * only do this on the initial mount.
 708  *
 709  * This isn't a change in behavior, because we're using the current state of the
 710  * file system to set the current mount options.  If you mounted with special
 711  * options to disable these features and then remounted we wouldn't revert the
 712  * settings, because mounting without these features cleared the on-disk
 713  * settings, so this being called on re-mount is not needed.
 714  */
 715 void btrfs_set_free_space_cache_settings(struct btrfs_fs_info *fs_info)
 716 {
 717         if (fs_info->sectorsize < PAGE_SIZE) {
 718                 btrfs_clear_opt(fs_info->mount_opt, SPACE_CACHE);
 719                 if (!btrfs_test_opt(fs_info, FREE_SPACE_TREE)) {
 720                         btrfs_info(fs_info,
 721                                    "forcing free space tree for sector size %u with page size %lu",
 722                                    fs_info->sectorsize, PAGE_SIZE);
 723                         btrfs_set_opt(fs_info->mount_opt, FREE_SPACE_TREE);
 724                 }
 725         }
 726
 727         /*
 728          * At this point our mount options are populated, so we only mess with
 729          * these settings if we don't have any settings already.
 730          */
 731         if (btrfs_test_opt(fs_info, FREE_SPACE_TREE))
 732                 return;
 733
 734         if (btrfs_is_zoned(fs_info) &&
 735             btrfs_free_space_cache_v1_active(fs_info)) {
 736                 btrfs_info(fs_info, "zoned: clearing existing space cache");
 737                 btrfs_set_super_cache_generation(fs_info->super_copy, 0);
 738                 return;
 739         }
 740
 741         if (btrfs_test_opt(fs_info, SPACE_CACHE))
 742                 return;
 743
 744         if (btrfs_test_opt(fs_info, NOSPACECACHE))
 745                 return;
 746
 747         /*
 748          * At this point we don't have explicit options set by the user, set
 749          * them ourselves based on the state of the file system.
 750          */
 751         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
 752                 btrfs_set_opt(fs_info->mount_opt, FREE_SPACE_TREE);
 753         else if (btrfs_free_space_cache_v1_active(fs_info))
 754                 btrfs_set_opt(fs_info->mount_opt, SPACE_CACHE);
 755 }
 756
 757 static void set_device_specific_options(struct btrfs_fs_info *fs_info)
 758 {
 759         if (!btrfs_test_opt(fs_info, NOSSD) &&
 760             !fs_info->fs_devices->rotating)
 761                 btrfs_set_opt(fs_info->mount_opt, SSD);
 762
 763         /*
 764          * For devices supporting discard turn on discard=async automatically,
 765          * unless it's already set or disabled. This could be turned off by
 766          * nodiscard for the same mount.
 767          *
 768          * The zoned mode piggy backs on the discard functionality for
 769          * resetting a zone. There is no reason to delay the zone reset as it is
 770          * fast enough. So, do not enable async discard for zoned mode.
 771          */
 772         if (!(btrfs_test_opt(fs_info, DISCARD_SYNC) ||
 773               btrfs_test_opt(fs_info, DISCARD_ASYNC) ||
 774               btrfs_test_opt(fs_info, NODISCARD)) &&
 775             fs_info->fs_devices->discardable &&
 776             !btrfs_is_zoned(fs_info))
 777                 btrfs_set_opt(fs_info->mount_opt, DISCARD_ASYNC);
 778 }
 779
 780 char *btrfs_get_subvol_name_from_objectid(struct btrfs_fs_info *fs_info,
 781                                           u64 subvol_objectid)
 782 {
 783         struct btrfs_root *root = fs_info->tree_root;
 784         struct btrfs_root *fs_root = NULL;
 785         struct btrfs_root_ref *root_ref;
 786         struct btrfs_inode_ref *inode_ref;
 787         struct btrfs_key key;
 788         struct btrfs_path *path = NULL;
 789         char *name = NULL, *ptr;
 790         u64 dirid;
 791         int len;
 792         int ret;
 793
 794         path = btrfs_alloc_path();
 795         if (!path) {
 796                 ret = -ENOMEM;
 797                 goto err;
 798         }
 799
 800         name = kmalloc(PATH_MAX, GFP_KERNEL);
 801         if (!name) {
 802                 ret = -ENOMEM;
 803                 goto err;
 804         }
 805         ptr = name + PATH_MAX - 1;
 806         ptr[0] = '\0';
 807
 808         /*
 809          * Walk up the subvolume trees in the tree of tree roots by root
 810          * backrefs until we hit the top-level subvolume.
 811          */
 812         while (subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
 813                 key.objectid = subvol_objectid;
 814                 key.type = BTRFS_ROOT_BACKREF_KEY;
 815                 key.offset = (u64)-1;
 816
 817                 ret = btrfs_search_backwards(root, &key, path);
 818                 if (ret < 0) {
 819                         goto err;
 820                 } else if (ret > 0) {
 821                         ret = -ENOENT;
 822                         goto err;
 823                 }
 824
 825                 subvol_objectid = key.offset;
 826
 827                 root_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
 828                                           struct btrfs_root_ref);
 829                 len = btrfs_root_ref_name_len(path->nodes[0], root_ref);
 830                 ptr -= len + 1;
 831                 if (ptr < name) {
 832                         ret = -ENAMETOOLONG;
 833                         goto err;
 834                 }
 835                 read_extent_buffer(path->nodes[0], ptr + 1,
 836                                    (unsigned long)(root_ref + 1), len);
 837                 ptr[0] = '/';
 838                 dirid = btrfs_root_ref_dirid(path->nodes[0], root_ref);
 839                 btrfs_release_path(path);
 840
 841                 fs_root = btrfs_get_fs_root(fs_info, subvol_objectid, true);
 842                 if (IS_ERR(fs_root)) {
 843                         ret = PTR_ERR(fs_root);
 844                         fs_root = NULL;
 845                         goto err;
 846                 }
 847
 848                 /*
 849                  * Walk up the filesystem tree by inode refs until we hit the
 850                  * root directory.
 851                  */
 852                 while (dirid != BTRFS_FIRST_FREE_OBJECTID) {
 853                         key.objectid = dirid;
 854                         key.type = BTRFS_INODE_REF_KEY;
 855                         key.offset = (u64)-1;
 856
 857                         ret = btrfs_search_backwards(fs_root, &key, path);
 858                         if (ret < 0) {
 859                                 goto err;
 860                         } else if (ret > 0) {
 861                                 ret = -ENOENT;
 862                                 goto err;
 863                         }
 864
 865                         dirid = key.offset;
 866
 867                         inode_ref = btrfs_item_ptr(path->nodes[0],
 868                                                    path->slots[0],
 869                                                    struct btrfs_inode_ref);
 870                         len = btrfs_inode_ref_name_len(path->nodes[0],
 871                                                        inode_ref);
 872                         ptr -= len + 1;
 873                         if (ptr < name) {
 874                                 ret = -ENAMETOOLONG;
 875                                 goto err;
 876                         }
 877                         read_extent_buffer(path->nodes[0], ptr + 1,
 878                                            (unsigned long)(inode_ref + 1), len);
 879                         ptr[0] = '/';
 880                         btrfs_release_path(path);
 881                 }
 882                 btrfs_put_root(fs_root);
 883                 fs_root = NULL;
 884         }
 885
 886         btrfs_free_path(path);
 887         if (ptr == name + PATH_MAX - 1) {
 888                 name[0] = '/';
 889                 name[1] = '\0';
 890         } else {
 891                 memmove(name, ptr, name + PATH_MAX - ptr);
 892         }
 893         return name;
 894
 895 err:
 896         btrfs_put_root(fs_root);
 897         btrfs_free_path(path);
 898         kfree(name);
 899         return ERR_PTR(ret);
 900 }
 901
 902 static int get_default_subvol_objectid(struct btrfs_fs_info *fs_info, u64 *objectid)
 903 {
 904         struct btrfs_root *root = fs_info->tree_root;
 905         struct btrfs_dir_item *di;
 906         struct btrfs_path *path;
 907         struct btrfs_key location;
 908         struct fscrypt_str name = FSTR_INIT("default", 7);
 909         u64 dir_id;
 910
 911         path = btrfs_alloc_path();
 912         if (!path)
 913                 return -ENOMEM;
 914
 915         /*
 916          * Find the "default" dir item which points to the root item that we
 917          * will mount by default if we haven't been given a specific subvolume
 918          * to mount.
 919          */
 920         dir_id = btrfs_super_root_dir(fs_info->super_copy);
 921         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, &name, 0);
 922         if (IS_ERR(di)) {
 923                 btrfs_free_path(path);
 924                 return PTR_ERR(di);
 925         }
 926         if (!di) {
 927                 /*
 928                  * Ok the default dir item isn't there.  This is weird since
 929                  * it's always been there, but don't freak out, just try and
 930                  * mount the top-level subvolume.
 931                  */
 932                 btrfs_free_path(path);
 933                 *objectid = BTRFS_FS_TREE_OBJECTID;
 934                 return 0;
 935         }
 936
 937         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
 938         btrfs_free_path(path);
 939         *objectid = location.objectid;
 940         return 0;
 941 }
 942
 943 static int btrfs_fill_super(struct super_block *sb,
 944                             struct btrfs_fs_devices *fs_devices)
 945 {
 946         struct btrfs_inode *inode;
 947         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 948         int err;
 949
 950         sb->s_maxbytes = MAX_LFS_FILESIZE;
 951         sb->s_magic = BTRFS_SUPER_MAGIC;
 952         sb->s_op = &btrfs_super_ops;
 953         sb->s_d_op = &btrfs_dentry_operations;
 954         sb->s_export_op = &btrfs_export_ops;
 955 #ifdef CONFIG_FS_VERITY
 956         sb->s_vop = &btrfs_verityops;
 957 #endif
 958         sb->s_xattr = btrfs_xattr_handlers;
 959         sb->s_time_gran = 1;
 960         sb->s_iflags |= SB_I_CGROUPWB | SB_I_ALLOW_HSM;
 961
 962         err = super_setup_bdi(sb);
 963         if (err) {
 964                 btrfs_err(fs_info, "super_setup_bdi failed");
 965                 return err;
 966         }
 967
 968         err = open_ctree(sb, fs_devices);
 969         if (err) {
 970                 btrfs_err(fs_info, "open_ctree failed: %d", err);
 971                 return err;
 972         }
 973
 974         inode = btrfs_iget(BTRFS_FIRST_FREE_OBJECTID, fs_info->fs_root);
 975         if (IS_ERR(inode)) {
 976                 err = PTR_ERR(inode);
 977                 btrfs_handle_fs_error(fs_info, err, NULL);
 978                 goto fail_close;
 979         }
 980
 981         sb->s_root = d_make_root(&inode->vfs_inode);
 982         if (!sb->s_root) {
 983                 err = -ENOMEM;
 984                 goto fail_close;
 985         }
 986
 987         sb->s_flags |= SB_ACTIVE;
 988         return 0;
 989
 990 fail_close:
 991         close_ctree(fs_info);
 992         return err;
 993 }
 994
 995 int btrfs_sync_fs(struct super_block *sb, int wait)
 996 {
 997         struct btrfs_trans_handle *trans;
 998         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 999         struct btrfs_root *root = fs_info->tree_root;
1000
1001         trace_btrfs_sync_fs(fs_info, wait);
1002
1003         if (!wait) {
1004                 filemap_flush(fs_info->btree_inode->i_mapping);
1005                 return 0;
1006         }
1007
1008         btrfs_wait_ordered_roots(fs_info, U64_MAX, NULL);
1009
1010         trans = btrfs_attach_transaction_barrier(root);
1011         if (IS_ERR(trans)) {
1012                 /* no transaction, don't bother */
1013                 if (PTR_ERR(trans) == -ENOENT) {
1014                         /*
1015                          * Exit unless we have some pending changes
1016                          * that need to go through commit
1017                          */
1018                         if (!test_bit(BTRFS_FS_NEED_TRANS_COMMIT,
1019                                       &fs_info->flags))
1020                                 return 0;
1021                         /*
1022                          * A non-blocking test if the fs is frozen. We must not
1023                          * start a new transaction here otherwise a deadlock
1024                          * happens. The pending operations are delayed to the
1025                          * next commit after thawing.
1026                          */
1027                         if (sb_start_write_trylock(sb))
1028                                 sb_end_write(sb);
1029                         else
1030                                 return 0;
1031                         trans = btrfs_start_transaction(root, 0);
1032                 }
1033                 if (IS_ERR(trans))
1034                         return PTR_ERR(trans);
1035         }
1036         return btrfs_commit_transaction(trans);
1037 }
1038
1039 static void print_rescue_option(struct seq_file *seq, const char *s, bool *printed)
1040 {
1041         seq_printf(seq, "%s%s", (*printed) ? ":" : ",rescue=", s);
1042         *printed = true;
1043 }
1044
1045 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
1046 {
1047         struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
1048         const char *compress_type;
1049         const char *subvol_name;
1050         bool printed = false;
1051
1052         if (btrfs_test_opt(info, DEGRADED))
1053                 seq_puts(seq, ",degraded");
1054         if (btrfs_test_opt(info, NODATASUM))
1055                 seq_puts(seq, ",nodatasum");
1056         if (btrfs_test_opt(info, NODATACOW))
1057                 seq_puts(seq, ",nodatacow");
1058         if (btrfs_test_opt(info, NOBARRIER))
1059                 seq_puts(seq, ",nobarrier");
1060         if (info->max_inline != BTRFS_DEFAULT_MAX_INLINE)
1061                 seq_printf(seq, ",max_inline=%llu", info->max_inline);
1062         if (info->thread_pool_size !=  min_t(unsigned long,
1063                                              num_online_cpus() + 2, 8))
1064                 seq_printf(seq, ",thread_pool=%u", info->thread_pool_size);
1065         if (btrfs_test_opt(info, COMPRESS)) {
1066                 compress_type = btrfs_compress_type2str(info->compress_type);
1067                 if (btrfs_test_opt(info, FORCE_COMPRESS))
1068                         seq_printf(seq, ",compress-force=%s", compress_type);
1069                 else
1070                         seq_printf(seq, ",compress=%s", compress_type);
1071                 if (info->compress_level)
1072                         seq_printf(seq, ":%d", info->compress_level);
1073         }
1074         if (btrfs_test_opt(info, NOSSD))
1075                 seq_puts(seq, ",nossd");
1076         if (btrfs_test_opt(info, SSD_SPREAD))
1077                 seq_puts(seq, ",ssd_spread");
1078         else if (btrfs_test_opt(info, SSD))
1079                 seq_puts(seq, ",ssd");
1080         if (btrfs_test_opt(info, NOTREELOG))
1081                 seq_puts(seq, ",notreelog");
1082         if (btrfs_test_opt(info, NOLOGREPLAY))
1083                 print_rescue_option(seq, "nologreplay", &printed);
1084         if (btrfs_test_opt(info, USEBACKUPROOT))
1085                 print_rescue_option(seq, "usebackuproot", &printed);
1086         if (btrfs_test_opt(info, IGNOREBADROOTS))
1087                 print_rescue_option(seq, "ignorebadroots", &printed);
1088         if (btrfs_test_opt(info, IGNOREDATACSUMS))
1089                 print_rescue_option(seq, "ignoredatacsums", &printed);
1090         if (btrfs_test_opt(info, IGNOREMETACSUMS))
1091                 print_rescue_option(seq, "ignoremetacsums", &printed);
1092         if (btrfs_test_opt(info, IGNORESUPERFLAGS))
1093                 print_rescue_option(seq, "ignoresuperflags", &printed);
1094         if (btrfs_test_opt(info, FLUSHONCOMMIT))
1095                 seq_puts(seq, ",flushoncommit");
1096         if (btrfs_test_opt(info, DISCARD_SYNC))
1097                 seq_puts(seq, ",discard");
1098         if (btrfs_test_opt(info, DISCARD_ASYNC))
1099                 seq_puts(seq, ",discard=async");
1100         if (!(info->sb->s_flags & SB_POSIXACL))
1101                 seq_puts(seq, ",noacl");
1102         if (btrfs_free_space_cache_v1_active(info))
1103                 seq_puts(seq, ",space_cache");
1104         else if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE))
1105                 seq_puts(seq, ",space_cache=v2");
1106         else
1107                 seq_puts(seq, ",nospace_cache");
1108         if (btrfs_test_opt(info, RESCAN_UUID_TREE))
1109                 seq_puts(seq, ",rescan_uuid_tree");
1110         if (btrfs_test_opt(info, CLEAR_CACHE))
1111                 seq_puts(seq, ",clear_cache");
1112         if (btrfs_test_opt(info, USER_SUBVOL_RM_ALLOWED))
1113                 seq_puts(seq, ",user_subvol_rm_allowed");
1114         if (btrfs_test_opt(info, ENOSPC_DEBUG))
1115                 seq_puts(seq, ",enospc_debug");
1116         if (btrfs_test_opt(info, AUTO_DEFRAG))
1117                 seq_puts(seq, ",autodefrag");
1118         if (btrfs_test_opt(info, SKIP_BALANCE))
1119                 seq_puts(seq, ",skip_balance");
1120         if (info->metadata_ratio)
1121                 seq_printf(seq, ",metadata_ratio=%u", info->metadata_ratio);
1122         if (btrfs_test_opt(info, PANIC_ON_FATAL_ERROR))
1123                 seq_puts(seq, ",fatal_errors=panic");
1124         if (info->commit_interval != BTRFS_DEFAULT_COMMIT_INTERVAL)
1125                 seq_printf(seq, ",commit=%u", info->commit_interval);
1126 #ifdef CONFIG_BTRFS_DEBUG
1127         if (btrfs_test_opt(info, FRAGMENT_DATA))
1128                 seq_puts(seq, ",fragment=data");
1129         if (btrfs_test_opt(info, FRAGMENT_METADATA))
1130                 seq_puts(seq, ",fragment=metadata");
1131 #endif
1132         if (btrfs_test_opt(info, REF_VERIFY))
1133                 seq_puts(seq, ",ref_verify");
1134         seq_printf(seq, ",subvolid=%llu", btrfs_root_id(BTRFS_I(d_inode(dentry))->root));
1135         subvol_name = btrfs_get_subvol_name_from_objectid(info,
1136                         btrfs_root_id(BTRFS_I(d_inode(dentry))->root));
1137         if (!IS_ERR(subvol_name)) {
1138                 seq_show_option(seq, "subvol", subvol_name);
1139                 kfree(subvol_name);
1140         }
1141         return 0;
1142 }
1143
1144 /*
1145  * subvolumes are identified by ino 256
1146  */
1147 static inline bool is_subvolume_inode(struct inode *inode)
1148 {
1149         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
1150                 return true;
1151         return false;
1152 }
1153
1154 static struct dentry *mount_subvol(const char *subvol_name, u64 subvol_objectid,
1155                                    struct vfsmount *mnt)
1156 {
1157         struct dentry *root;
1158         int ret;
1159
1160         if (!subvol_name) {
1161                 if (!subvol_objectid) {
1162                         ret = get_default_subvol_objectid(btrfs_sb(mnt->mnt_sb),
1163                                                           &subvol_objectid);
1164                         if (ret) {
1165                                 root = ERR_PTR(ret);
1166                                 goto out;
1167                         }
1168                 }
1169                 subvol_name = btrfs_get_subvol_name_from_objectid(
1170                                         btrfs_sb(mnt->mnt_sb), subvol_objectid);
1171                 if (IS_ERR(subvol_name)) {
1172                         root = ERR_CAST(subvol_name);
1173                         subvol_name = NULL;
1174                         goto out;
1175                 }
1176
1177         }
1178
1179         root = mount_subtree(mnt, subvol_name);
1180         /* mount_subtree() drops our reference on the vfsmount. */
1181         mnt = NULL;
1182
1183         if (!IS_ERR(root)) {
1184                 struct super_block *s = root->d_sb;
1185                 struct btrfs_fs_info *fs_info = btrfs_sb(s);
1186                 struct inode *root_inode = d_inode(root);
1187                 u64 root_objectid = btrfs_root_id(BTRFS_I(root_inode)->root);
1188
1189                 ret = 0;
1190                 if (!is_subvolume_inode(root_inode)) {
1191                         btrfs_err(fs_info, "'%s' is not a valid subvolume",
1192                                subvol_name);
1193                         ret = -EINVAL;
1194                 }
1195                 if (subvol_objectid && root_objectid != subvol_objectid) {
1196                         /*
1197                          * This will also catch a race condition where a
1198                          * subvolume which was passed by ID is renamed and
1199                          * another subvolume is renamed over the old location.
1200                          */
1201                         btrfs_err(fs_info,
1202                                   "subvol '%s' does not match subvolid %llu",
1203                                   subvol_name, subvol_objectid);
1204                         ret = -EINVAL;
1205                 }
1206                 if (ret) {
1207                         dput(root);
1208                         root = ERR_PTR(ret);
1209                         deactivate_locked_super(s);
1210                 }
1211         }
1212
1213 out:
1214         mntput(mnt);
1215         kfree(subvol_name);
1216         return root;
1217 }
1218
1219 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1220                                      u32 new_pool_size, u32 old_pool_size)
1221 {
1222         if (new_pool_size == old_pool_size)
1223                 return;
1224
1225         fs_info->thread_pool_size = new_pool_size;
1226
1227         btrfs_info(fs_info, "resize thread pool %d -> %d",
1228                old_pool_size, new_pool_size);
1229
1230         btrfs_workqueue_set_max(fs_info->workers, new_pool_size);
1231         btrfs_workqueue_set_max(fs_info->delalloc_workers, new_pool_size);
1232         btrfs_workqueue_set_max(fs_info->caching_workers, new_pool_size);
1233         workqueue_set_max_active(fs_info->endio_workers, new_pool_size);
1234         workqueue_set_max_active(fs_info->endio_meta_workers, new_pool_size);
1235         btrfs_workqueue_set_max(fs_info->endio_write_workers, new_pool_size);
1236         btrfs_workqueue_set_max(fs_info->endio_freespace_worker, new_pool_size);
1237         btrfs_workqueue_set_max(fs_info->delayed_workers, new_pool_size);
1238 }
1239
1240 static inline void btrfs_remount_begin(struct btrfs_fs_info *fs_info,
1241                                        unsigned long long old_opts, int flags)
1242 {
1243         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1244             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1245              (flags & SB_RDONLY))) {
1246                 /* wait for any defraggers to finish */
1247                 wait_event(fs_info->transaction_wait,
1248                            (atomic_read(&fs_info->defrag_running) == 0));
1249                 if (flags & SB_RDONLY)
1250                         sync_filesystem(fs_info->sb);
1251         }
1252 }
1253
1254 static inline void btrfs_remount_cleanup(struct btrfs_fs_info *fs_info,
1255                                          unsigned long long old_opts)
1256 {
1257         const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
1258
1259         /*
1260          * We need to cleanup all defragable inodes if the autodefragment is
1261          * close or the filesystem is read only.
1262          */
1263         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1264             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) || sb_rdonly(fs_info->sb))) {
1265                 btrfs_cleanup_defrag_inodes(fs_info);
1266         }
1267
1268         /* If we toggled discard async */
1269         if (!btrfs_raw_test_opt(old_opts, DISCARD_ASYNC) &&
1270             btrfs_test_opt(fs_info, DISCARD_ASYNC))
1271                 btrfs_discard_resume(fs_info);
1272         else if (btrfs_raw_test_opt(old_opts, DISCARD_ASYNC) &&
1273                  !btrfs_test_opt(fs_info, DISCARD_ASYNC))
1274                 btrfs_discard_cleanup(fs_info);
1275
1276         /* If we toggled space cache */
1277         if (cache_opt != btrfs_free_space_cache_v1_active(fs_info))
1278                 btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
1279 }
1280
1281 static int btrfs_remount_rw(struct btrfs_fs_info *fs_info)
1282 {
1283         int ret;
1284
1285         if (BTRFS_FS_ERROR(fs_info)) {
1286                 btrfs_err(fs_info,
1287                           "remounting read-write after error is not allowed");
1288                 return -EINVAL;
1289         }
1290
1291         if (fs_info->fs_devices->rw_devices == 0)
1292                 return -EACCES;
1293
1294         if (!btrfs_check_rw_degradable(fs_info, NULL)) {
1295                 btrfs_warn(fs_info,
1296                            "too many missing devices, writable remount is not allowed");
1297                 return -EACCES;
1298         }
1299
1300         if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1301                 btrfs_warn(fs_info,
1302                            "mount required to replay tree-log, cannot remount read-write");
1303                 return -EINVAL;
1304         }
1305
1306         /*
1307          * NOTE: when remounting with a change that does writes, don't put it
1308          * anywhere above this point, as we are not sure to be safe to write
1309          * until we pass the above checks.
1310          */
1311         ret = btrfs_start_pre_rw_mount(fs_info);
1312         if (ret)
1313                 return ret;
1314
1315         btrfs_clear_sb_rdonly(fs_info->sb);
1316
1317         set_bit(BTRFS_FS_OPEN, &fs_info->flags);
1318
1319         /*
1320          * If we've gone from readonly -> read-write, we need to get our
1321          * sync/async discard lists in the right state.
1322          */
1323         btrfs_discard_resume(fs_info);
1324
1325         return 0;
1326 }
1327
1328 static int btrfs_remount_ro(struct btrfs_fs_info *fs_info)
1329 {
1330         /*
1331          * This also happens on 'umount -rf' or on shutdown, when the
1332          * filesystem is busy.
1333          */
1334         cancel_work_sync(&fs_info->async_reclaim_work);
1335         cancel_work_sync(&fs_info->async_data_reclaim_work);
1336
1337         btrfs_discard_cleanup(fs_info);
1338
1339         /* Wait for the uuid_scan task to finish */
1340         down(&fs_info->uuid_tree_rescan_sem);
1341         /* Avoid complains from lockdep et al. */
1342         up(&fs_info->uuid_tree_rescan_sem);
1343
1344         btrfs_set_sb_rdonly(fs_info->sb);
1345
1346         /*
1347          * Setting SB_RDONLY will put the cleaner thread to sleep at the next
1348          * loop if it's already active.  If it's already asleep, we'll leave
1349          * unused block groups on disk until we're mounted read-write again
1350          * unless we clean them up here.
1351          */
1352         btrfs_delete_unused_bgs(fs_info);
1353
1354         /*
1355          * The cleaner task could be already running before we set the flag
1356          * BTRFS_FS_STATE_RO (and SB_RDONLY in the superblock).  We must make
1357          * sure that after we finish the remount, i.e. after we call
1358          * btrfs_commit_super(), the cleaner can no longer start a transaction
1359          * - either because it was dropping a dead root, running delayed iputs
1360          *   or deleting an unused block group (the cleaner picked a block
1361          *   group from the list of unused block groups before we were able to
1362          *   in the previous call to btrfs_delete_unused_bgs()).
1363          */
1364         wait_on_bit(&fs_info->flags, BTRFS_FS_CLEANER_RUNNING, TASK_UNINTERRUPTIBLE);
1365
1366         /*
1367          * We've set the superblock to RO mode, so we might have made the
1368          * cleaner task sleep without running all pending delayed iputs. Go
1369          * through all the delayed iputs here, so that if an unmount happens
1370          * without remounting RW we don't end up at finishing close_ctree()
1371          * with a non-empty list of delayed iputs.
1372          */
1373         btrfs_run_delayed_iputs(fs_info);
1374
1375         btrfs_dev_replace_suspend_for_unmount(fs_info);
1376         btrfs_scrub_cancel(fs_info);
1377         btrfs_pause_balance(fs_info);
1378
1379         /*
1380          * Pause the qgroup rescan worker if it is running. We don't want it to
1381          * be still running after we are in RO mode, as after that, by the time
1382          * we unmount, it might have left a transaction open, so we would leak
1383          * the transaction and/or crash.
1384          */
1385         btrfs_qgroup_wait_for_completion(fs_info, false);
1386
1387         return btrfs_commit_super(fs_info);
1388 }
1389
1390 static void btrfs_ctx_to_info(struct btrfs_fs_info *fs_info, struct btrfs_fs_context *ctx)
1391 {
1392         fs_info->max_inline = ctx->max_inline;
1393         fs_info->commit_interval = ctx->commit_interval;
1394         fs_info->metadata_ratio = ctx->metadata_ratio;
1395         fs_info->thread_pool_size = ctx->thread_pool_size;
1396         fs_info->mount_opt = ctx->mount_opt;
1397         fs_info->compress_type = ctx->compress_type;
1398         fs_info->compress_level = ctx->compress_level;
1399 }
1400
1401 static void btrfs_info_to_ctx(struct btrfs_fs_info *fs_info, struct btrfs_fs_context *ctx)
1402 {
1403         ctx->max_inline = fs_info->max_inline;
1404         ctx->commit_interval = fs_info->commit_interval;
1405         ctx->metadata_ratio = fs_info->metadata_ratio;
1406         ctx->thread_pool_size = fs_info->thread_pool_size;
1407         ctx->mount_opt = fs_info->mount_opt;
1408         ctx->compress_type = fs_info->compress_type;
1409         ctx->compress_level = fs_info->compress_level;
1410 }
1411
1412 #define btrfs_info_if_set(fs_info, old_ctx, opt, fmt, args...)                  \
1413 do {                                                                            \
1414         if ((!old_ctx || !btrfs_raw_test_opt(old_ctx->mount_opt, opt)) &&       \
1415             btrfs_raw_test_opt(fs_info->mount_opt, opt))                        \
1416                 btrfs_info(fs_info, fmt, ##args);                               \
1417 } while (0)
1418
1419 #define btrfs_info_if_unset(fs_info, old_ctx, opt, fmt, args...)        \
1420 do {                                                                    \
1421         if ((old_ctx && btrfs_raw_test_opt(old_ctx->mount_opt, opt)) && \
1422             !btrfs_raw_test_opt(fs_info->mount_opt, opt))               \
1423                 btrfs_info(fs_info, fmt, ##args);                       \
1424 } while (0)
1425
1426 static void btrfs_emit_options(struct btrfs_fs_info *info,
1427                                struct btrfs_fs_context *old)
1428 {
1429         btrfs_info_if_set(info, old, NODATASUM, "setting nodatasum");
1430         btrfs_info_if_set(info, old, DEGRADED, "allowing degraded mounts");
1431         btrfs_info_if_set(info, old, NODATASUM, "setting nodatasum");
1432         btrfs_info_if_set(info, old, SSD, "enabling ssd optimizations");
1433         btrfs_info_if_set(info, old, SSD_SPREAD, "using spread ssd allocation scheme");
1434         btrfs_info_if_set(info, old, NOBARRIER, "turning off barriers");
1435         btrfs_info_if_set(info, old, NOTREELOG, "disabling tree log");
1436         btrfs_info_if_set(info, old, NOLOGREPLAY, "disabling log replay at mount time");
1437         btrfs_info_if_set(info, old, FLUSHONCOMMIT, "turning on flush-on-commit");
1438         btrfs_info_if_set(info, old, DISCARD_SYNC, "turning on sync discard");
1439         btrfs_info_if_set(info, old, DISCARD_ASYNC, "turning on async discard");
1440         btrfs_info_if_set(info, old, FREE_SPACE_TREE, "enabling free space tree");
1441         btrfs_info_if_set(info, old, SPACE_CACHE, "enabling disk space caching");
1442         btrfs_info_if_set(info, old, CLEAR_CACHE, "force clearing of disk cache");
1443         btrfs_info_if_set(info, old, AUTO_DEFRAG, "enabling auto defrag");
1444         btrfs_info_if_set(info, old, FRAGMENT_DATA, "fragmenting data");
1445         btrfs_info_if_set(info, old, FRAGMENT_METADATA, "fragmenting metadata");
1446         btrfs_info_if_set(info, old, REF_VERIFY, "doing ref verification");
1447         btrfs_info_if_set(info, old, USEBACKUPROOT, "trying to use backup root at mount time");
1448         btrfs_info_if_set(info, old, IGNOREBADROOTS, "ignoring bad roots");
1449         btrfs_info_if_set(info, old, IGNOREDATACSUMS, "ignoring data csums");
1450         btrfs_info_if_set(info, old, IGNOREMETACSUMS, "ignoring meta csums");
1451         btrfs_info_if_set(info, old, IGNORESUPERFLAGS, "ignoring unknown super block flags");
1452
1453         btrfs_info_if_unset(info, old, NODATACOW, "setting datacow");
1454         btrfs_info_if_unset(info, old, SSD, "not using ssd optimizations");
1455         btrfs_info_if_unset(info, old, SSD_SPREAD, "not using spread ssd allocation scheme");
1456         btrfs_info_if_unset(info, old, NOBARRIER, "turning off barriers");
1457         btrfs_info_if_unset(info, old, NOTREELOG, "enabling tree log");
1458         btrfs_info_if_unset(info, old, SPACE_CACHE, "disabling disk space caching");
1459         btrfs_info_if_unset(info, old, FREE_SPACE_TREE, "disabling free space tree");
1460         btrfs_info_if_unset(info, old, AUTO_DEFRAG, "disabling auto defrag");
1461         btrfs_info_if_unset(info, old, COMPRESS, "use no compression");
1462
1463         /* Did the compression settings change? */
1464         if (btrfs_test_opt(info, COMPRESS) &&
1465             (!old ||
1466              old->compress_type != info->compress_type ||
1467              old->compress_level != info->compress_level ||
1468              (!btrfs_raw_test_opt(old->mount_opt, FORCE_COMPRESS) &&
1469               btrfs_raw_test_opt(info->mount_opt, FORCE_COMPRESS)))) {
1470                 const char *compress_type = btrfs_compress_type2str(info->compress_type);
1471
1472                 btrfs_info(info, "%s %s compression, level %d",
1473                            btrfs_test_opt(info, FORCE_COMPRESS) ? "force" : "use",
1474                            compress_type, info->compress_level);
1475         }
1476
1477         if (info->max_inline != BTRFS_DEFAULT_MAX_INLINE)
1478                 btrfs_info(info, "max_inline set to %llu", info->max_inline);
1479 }
1480
1481 static int btrfs_reconfigure(struct fs_context *fc)
1482 {
1483         struct super_block *sb = fc->root->d_sb;
1484         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1485         struct btrfs_fs_context *ctx = fc->fs_private;
1486         struct btrfs_fs_context old_ctx;
1487         int ret = 0;
1488         bool mount_reconfigure = (fc->s_fs_info != NULL);
1489
1490         btrfs_info_to_ctx(fs_info, &old_ctx);
1491
1492         /*
1493          * This is our "bind mount" trick, we don't want to allow the user to do
1494          * anything other than mount a different ro/rw and a different subvol,
1495          * all of the mount options should be maintained.
1496          */
1497         if (mount_reconfigure)
1498                 ctx->mount_opt = old_ctx.mount_opt;
1499
1500         sync_filesystem(sb);
1501         set_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1502
1503         if (!btrfs_check_options(fs_info, &ctx->mount_opt, fc->sb_flags))
1504                 return -EINVAL;
1505
1506         ret = btrfs_check_features(fs_info, !(fc->sb_flags & SB_RDONLY));
1507         if (ret < 0)
1508                 return ret;
1509
1510         btrfs_ctx_to_info(fs_info, ctx);
1511         btrfs_remount_begin(fs_info, old_ctx.mount_opt, fc->sb_flags);
1512         btrfs_resize_thread_pool(fs_info, fs_info->thread_pool_size,
1513                                  old_ctx.thread_pool_size);
1514
1515         if ((bool)btrfs_test_opt(fs_info, FREE_SPACE_TREE) !=
1516             (bool)btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
1517             (!sb_rdonly(sb) || (fc->sb_flags & SB_RDONLY))) {
1518                 btrfs_warn(fs_info,
1519                 "remount supports changing free space tree only from RO to RW");
1520                 /* Make sure free space cache options match the state on disk. */
1521                 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
1522                         btrfs_set_opt(fs_info->mount_opt, FREE_SPACE_TREE);
1523                         btrfs_clear_opt(fs_info->mount_opt, SPACE_CACHE);
1524                 }
1525                 if (btrfs_free_space_cache_v1_active(fs_info)) {
1526                         btrfs_clear_opt(fs_info->mount_opt, FREE_SPACE_TREE);
1527                         btrfs_set_opt(fs_info->mount_opt, SPACE_CACHE);
1528                 }
1529         }
1530
1531         ret = 0;
1532         if (!sb_rdonly(sb) && (fc->sb_flags & SB_RDONLY))
1533                 ret = btrfs_remount_ro(fs_info);
1534         else if (sb_rdonly(sb) && !(fc->sb_flags & SB_RDONLY))
1535                 ret = btrfs_remount_rw(fs_info);
1536         if (ret)
1537                 goto restore;
1538
1539         /*
1540          * If we set the mask during the parameter parsing VFS would reject the
1541          * remount.  Here we can set the mask and the value will be updated
1542          * appropriately.
1543          */
1544         if ((fc->sb_flags & SB_POSIXACL) != (sb->s_flags & SB_POSIXACL))
1545                 fc->sb_flags_mask |= SB_POSIXACL;
1546
1547         btrfs_emit_options(fs_info, &old_ctx);
1548         wake_up_process(fs_info->transaction_kthread);
1549         btrfs_remount_cleanup(fs_info, old_ctx.mount_opt);
1550         btrfs_clear_oneshot_options(fs_info);
1551         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1552
1553         return 0;
1554 restore:
1555         btrfs_ctx_to_info(fs_info, &old_ctx);
1556         btrfs_remount_cleanup(fs_info, old_ctx.mount_opt);
1557         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1558         return ret;
1559 }
1560
1561 /* Used to sort the devices by max_avail(descending sort) */
1562 static int btrfs_cmp_device_free_bytes(const void *a, const void *b)
1563 {
1564         const struct btrfs_device_info *dev_info1 = a;
1565         const struct btrfs_device_info *dev_info2 = b;
1566
1567         if (dev_info1->max_avail > dev_info2->max_avail)
1568                 return -1;
1569         else if (dev_info1->max_avail < dev_info2->max_avail)
1570                 return 1;
1571         return 0;
1572 }
1573
1574 /*
1575  * sort the devices by max_avail, in which max free extent size of each device
1576  * is stored.(Descending Sort)
1577  */
1578 static inline void btrfs_descending_sort_devices(
1579                                         struct btrfs_device_info *devices,
1580                                         size_t nr_devices)
1581 {
1582         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1583              btrfs_cmp_device_free_bytes, NULL);
1584 }
1585
1586 /*
1587  * The helper to calc the free space on the devices that can be used to store
1588  * file data.
1589  */
1590 static inline int btrfs_calc_avail_data_space(struct btrfs_fs_info *fs_info,
1591                                               u64 *free_bytes)
1592 {
1593         struct btrfs_device_info *devices_info;
1594         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1595         struct btrfs_device *device;
1596         u64 type;
1597         u64 avail_space;
1598         u64 min_stripe_size;
1599         int num_stripes = 1;
1600         int i = 0, nr_devices;
1601         const struct btrfs_raid_attr *rattr;
1602
1603         /*
1604          * We aren't under the device list lock, so this is racy-ish, but good
1605          * enough for our purposes.
1606          */
1607         nr_devices = fs_info->fs_devices->open_devices;
1608         if (!nr_devices) {
1609                 smp_mb();
1610                 nr_devices = fs_info->fs_devices->open_devices;
1611                 ASSERT(nr_devices);
1612                 if (!nr_devices) {
1613                         *free_bytes = 0;
1614                         return 0;
1615                 }
1616         }
1617
1618         devices_info = kmalloc_array(nr_devices, sizeof(*devices_info),
1619                                GFP_KERNEL);
1620         if (!devices_info)
1621                 return -ENOMEM;
1622
1623         /* calc min stripe number for data space allocation */
1624         type = btrfs_data_alloc_profile(fs_info);
1625         rattr = &btrfs_raid_array[btrfs_bg_flags_to_raid_index(type)];
1626
1627         if (type & BTRFS_BLOCK_GROUP_RAID0)
1628                 num_stripes = nr_devices;
1629         else if (type & BTRFS_BLOCK_GROUP_RAID1_MASK)
1630                 num_stripes = rattr->ncopies;
1631         else if (type & BTRFS_BLOCK_GROUP_RAID10)
1632                 num_stripes = 4;
1633
1634         /* Adjust for more than 1 stripe per device */
1635         min_stripe_size = rattr->dev_stripes * BTRFS_STRIPE_LEN;
1636
1637         rcu_read_lock();
1638         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
1639                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
1640                                                 &device->dev_state) ||
1641                     !device->bdev ||
1642                     test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
1643                         continue;
1644
1645                 if (i >= nr_devices)
1646                         break;
1647
1648                 avail_space = device->total_bytes - device->bytes_used;
1649
1650                 /* align with stripe_len */
1651                 avail_space = rounddown(avail_space, BTRFS_STRIPE_LEN);
1652
1653                 /*
1654                  * Ensure we have at least min_stripe_size on top of the
1655                  * reserved space on the device.
1656                  */
1657                 if (avail_space <= BTRFS_DEVICE_RANGE_RESERVED + min_stripe_size)
1658                         continue;
1659
1660                 avail_space -= BTRFS_DEVICE_RANGE_RESERVED;
1661
1662                 devices_info[i].dev = device;
1663                 devices_info[i].max_avail = avail_space;
1664
1665                 i++;
1666         }
1667         rcu_read_unlock();
1668
1669         nr_devices = i;
1670
1671         btrfs_descending_sort_devices(devices_info, nr_devices);
1672
1673         i = nr_devices - 1;
1674         avail_space = 0;
1675         while (nr_devices >= rattr->devs_min) {
1676                 num_stripes = min(num_stripes, nr_devices);
1677
1678                 if (devices_info[i].max_avail >= min_stripe_size) {
1679                         int j;
1680                         u64 alloc_size;
1681
1682                         avail_space += devices_info[i].max_avail * num_stripes;
1683                         alloc_size = devices_info[i].max_avail;
1684                         for (j = i + 1 - num_stripes; j <= i; j++)
1685                                 devices_info[j].max_avail -= alloc_size;
1686                 }
1687                 i--;
1688                 nr_devices--;
1689         }
1690
1691         kfree(devices_info);
1692         *free_bytes = avail_space;
1693         return 0;
1694 }
1695
1696 /*
1697  * Calculate numbers for 'df', pessimistic in case of mixed raid profiles.
1698  *
1699  * If there's a redundant raid level at DATA block groups, use the respective
1700  * multiplier to scale the sizes.
1701  *
1702  * Unused device space usage is based on simulating the chunk allocator
1703  * algorithm that respects the device sizes and order of allocations.  This is
1704  * a close approximation of the actual use but there are other factors that may
1705  * change the result (like a new metadata chunk).
1706  *
1707  * If metadata is exhausted, f_bavail will be 0.
1708  */
1709 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
1710 {
1711         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
1712         struct btrfs_super_block *disk_super = fs_info->super_copy;
1713         struct btrfs_space_info *found;
1714         u64 total_used = 0;
1715         u64 total_free_data = 0;
1716         u64 total_free_meta = 0;
1717         u32 bits = fs_info->sectorsize_bits;
1718         __be32 *fsid = (__be32 *)fs_info->fs_devices->fsid;
1719         unsigned factor = 1;
1720         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
1721         int ret;
1722         u64 thresh = 0;
1723         int mixed = 0;
1724
1725         list_for_each_entry(found, &fs_info->space_info, list) {
1726                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
1727                         int i;
1728
1729                         total_free_data += found->disk_total - found->disk_used;
1730                         total_free_data -=
1731                                 btrfs_account_ro_block_groups_free_space(found);
1732
1733                         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1734                                 if (!list_empty(&found->block_groups[i]))
1735                                         factor = btrfs_bg_type_to_factor(
1736                                                 btrfs_raid_array[i].bg_flag);
1737                         }
1738                 }
1739
1740                 /*
1741                  * Metadata in mixed block group profiles are accounted in data
1742                  */
1743                 if (!mixed && found->flags & BTRFS_BLOCK_GROUP_METADATA) {
1744                         if (found->flags & BTRFS_BLOCK_GROUP_DATA)
1745                                 mixed = 1;
1746                         else
1747                                 total_free_meta += found->disk_total -
1748                                         found->disk_used;
1749                 }
1750
1751                 total_used += found->disk_used;
1752         }
1753
1754         buf->f_blocks = div_u64(btrfs_super_total_bytes(disk_super), factor);
1755         buf->f_blocks >>= bits;
1756         buf->f_bfree = buf->f_blocks - (div_u64(total_used, factor) >> bits);
1757
1758         /* Account global block reserve as used, it's in logical size already */
1759         spin_lock(&block_rsv->lock);
1760         /* Mixed block groups accounting is not byte-accurate, avoid overflow */
1761         if (buf->f_bfree >= block_rsv->size >> bits)
1762                 buf->f_bfree -= block_rsv->size >> bits;
1763         else
1764                 buf->f_bfree = 0;
1765         spin_unlock(&block_rsv->lock);
1766
1767         buf->f_bavail = div_u64(total_free_data, factor);
1768         ret = btrfs_calc_avail_data_space(fs_info, &total_free_data);
1769         if (ret)
1770                 return ret;
1771         buf->f_bavail += div_u64(total_free_data, factor);
1772         buf->f_bavail = buf->f_bavail >> bits;
1773
1774         /*
1775          * We calculate the remaining metadata space minus global reserve. If
1776          * this is (supposedly) smaller than zero, there's no space. But this
1777          * does not hold in practice, the exhausted state happens where's still
1778          * some positive delta. So we apply some guesswork and compare the
1779          * delta to a 4M threshold.  (Practically observed delta was ~2M.)
1780          *
1781          * We probably cannot calculate the exact threshold value because this
1782          * depends on the internal reservations requested by various
1783          * operations, so some operations that consume a few metadata will
1784          * succeed even if the Avail is zero. But this is better than the other
1785          * way around.
1786          */
1787         thresh = SZ_4M;
1788
1789         /*
1790          * We only want to claim there's no available space if we can no longer
1791          * allocate chunks for our metadata profile and our global reserve will
1792          * not fit in the free metadata space.  If we aren't ->full then we
1793          * still can allocate chunks and thus are fine using the currently
1794          * calculated f_bavail.
1795          */
1796         if (!mixed && block_rsv->space_info->full &&
1797             (total_free_meta < thresh || total_free_meta - thresh < block_rsv->size))
1798                 buf->f_bavail = 0;
1799
1800         buf->f_type = BTRFS_SUPER_MAGIC;
1801         buf->f_bsize = fs_info->sectorsize;
1802         buf->f_namelen = BTRFS_NAME_LEN;
1803
1804         /* We treat it as constant endianness (it doesn't matter _which_)
1805            because we want the fsid to come out the same whether mounted
1806            on a big-endian or little-endian host */
1807         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
1808         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
1809         /* Mask in the root object ID too, to disambiguate subvols */
1810         buf->f_fsid.val[0] ^= btrfs_root_id(BTRFS_I(d_inode(dentry))->root) >> 32;
1811         buf->f_fsid.val[1] ^= btrfs_root_id(BTRFS_I(d_inode(dentry))->root);
1812
1813         return 0;
1814 }
1815
1816 static int btrfs_fc_test_super(struct super_block *sb, struct fs_context *fc)
1817 {
1818         struct btrfs_fs_info *p = fc->s_fs_info;
1819         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1820
1821         return fs_info->fs_devices == p->fs_devices;
1822 }
1823
1824 static int btrfs_get_tree_super(struct fs_context *fc)
1825 {
1826         struct btrfs_fs_info *fs_info = fc->s_fs_info;
1827         struct btrfs_fs_context *ctx = fc->fs_private;
1828         struct btrfs_fs_devices *fs_devices = NULL;
1829         struct block_device *bdev;
1830         struct btrfs_device *device;
1831         struct super_block *sb;
1832         blk_mode_t mode = btrfs_open_mode(fc);
1833         int ret;
1834
1835         btrfs_ctx_to_info(fs_info, ctx);
1836         mutex_lock(&uuid_mutex);
1837
1838         /*
1839          * With 'true' passed to btrfs_scan_one_device() (mount time) we expect
1840          * either a valid device or an error.
1841          */
1842         device = btrfs_scan_one_device(fc->source, mode, true);
1843         ASSERT(device != NULL);
1844         if (IS_ERR(device)) {
1845                 mutex_unlock(&uuid_mutex);
1846                 return PTR_ERR(device);
1847         }
1848
1849         fs_devices = device->fs_devices;
1850         fs_info->fs_devices = fs_devices;
1851
1852         ret = btrfs_open_devices(fs_devices, mode, &btrfs_fs_type);
1853         mutex_unlock(&uuid_mutex);
1854         if (ret)
1855                 return ret;
1856
1857         if (!(fc->sb_flags & SB_RDONLY) && fs_devices->rw_devices == 0) {
1858                 ret = -EACCES;
1859                 goto error;
1860         }
1861
1862         bdev = fs_devices->latest_dev->bdev;
1863
1864         /*
1865          * From now on the error handling is not straightforward.
1866          *
1867          * If successful, this will transfer the fs_info into the super block,
1868          * and fc->s_fs_info will be NULL.  However if there's an existing
1869          * super, we'll still have fc->s_fs_info populated.  If we error
1870          * completely out it'll be cleaned up when we drop the fs_context,
1871          * otherwise it's tied to the lifetime of the super_block.
1872          */
1873         sb = sget_fc(fc, btrfs_fc_test_super, set_anon_super_fc);
1874         if (IS_ERR(sb)) {
1875                 ret = PTR_ERR(sb);
1876                 goto error;
1877         }
1878
1879         set_device_specific_options(fs_info);
1880
1881         if (sb->s_root) {
1882                 btrfs_close_devices(fs_devices);
1883                 /*
1884                  * At this stage we may have RO flag mismatch between
1885                  * fc->sb_flags and sb->s_flags.  Caller should detect such
1886                  * mismatch and reconfigure with sb->s_umount rwsem held if
1887                  * needed.
1888                  */
1889         } else {
1890                 snprintf(sb->s_id, sizeof(sb->s_id), "%pg", bdev);
1891                 shrinker_debugfs_rename(sb->s_shrink, "sb-btrfs:%s", sb->s_id);
1892                 btrfs_sb(sb)->bdev_holder = &btrfs_fs_type;
1893                 ret = btrfs_fill_super(sb, fs_devices);
1894                 if (ret) {
1895                         deactivate_locked_super(sb);
1896                         return ret;
1897                 }
1898         }
1899
1900         btrfs_clear_oneshot_options(fs_info);
1901
1902         fc->root = dget(sb->s_root);
1903         return 0;
1904
1905 error:
1906         btrfs_close_devices(fs_devices);
1907         return ret;
1908 }
1909
1910 /*
1911  * Ever since commit 0723a0473fb4 ("btrfs: allow mounting btrfs subvolumes
1912  * with different ro/rw options") the following works:
1913  *
1914  *        (i) mount /dev/sda3 -o subvol=foo,ro /mnt/foo
1915  *       (ii) mount /dev/sda3 -o subvol=bar,rw /mnt/bar
1916  *
1917  * which looks nice and innocent but is actually pretty intricate and deserves
1918  * a long comment.
1919  *
1920  * On another filesystem a subvolume mount is close to something like:
1921  *
1922  *      (iii) # create rw superblock + initial mount
1923  *            mount -t xfs /dev/sdb /opt/
1924  *
1925  *            # create ro bind mount
1926  *            mount --bind -o ro /opt/foo /mnt/foo
1927  *
1928  *            # unmount initial mount
1929  *            umount /opt
1930  *
1931  * Of course, there's some special subvolume sauce and there's the fact that the
1932  * sb->s_root dentry is really swapped after mount_subtree(). But conceptually
1933  * it's very close and will help us understand the issue.
1934  *
1935  * The old mount API didn't cleanly distinguish between a mount being made ro
1936  * and a superblock being made ro.  The only way to change the ro state of
1937  * either object was by passing ms_rdonly. If a new mount was created via
1938  * mount(2) such as:
1939  *
1940  *      mount("/dev/sdb", "/mnt", "xfs", ms_rdonly, null);
1941  *
1942  * the MS_RDONLY flag being specified had two effects:
1943  *
1944  * (1) MNT_READONLY was raised -> the resulting mount got
1945  *     @mnt->mnt_flags |= MNT_READONLY raised.
1946  *
1947  * (2) MS_RDONLY was passed to the filesystem's mount method and the filesystems
1948  *     made the superblock ro. Note, how SB_RDONLY has the same value as
1949  *     ms_rdonly and is raised whenever MS_RDONLY is passed through mount(2).
1950  *
1951  * Creating a subtree mount via (iii) ends up leaving a rw superblock with a
1952  * subtree mounted ro.
1953  *
1954  * But consider the effect on the old mount API on btrfs subvolume mounting
1955  * which combines the distinct step in (iii) into a single step.
1956  *
1957  * By issuing (i) both the mount and the superblock are turned ro. Now when (ii)
1958  * is issued the superblock is ro and thus even if the mount created for (ii) is
1959  * rw it wouldn't help. Hence, btrfs needed to transition the superblock from ro
1960  * to rw for (ii) which it did using an internal remount call.
1961  *
1962  * IOW, subvolume mounting was inherently complicated due to the ambiguity of
1963  * MS_RDONLY in mount(2). Note, this ambiguity has mount(8) always translate
1964  * "ro" to MS_RDONLY. IOW, in both (i) and (ii) "ro" becomes MS_RDONLY when
1965  * passed by mount(8) to mount(2).
1966  *
1967  * Enter the new mount API. The new mount API disambiguates making a mount ro
1968  * and making a superblock ro.
1969  *
1970  * (3) To turn a mount ro the MOUNT_ATTR_ONLY flag can be used with either
1971  *     fsmount() or mount_setattr() this is a pure VFS level change for a
1972  *     specific mount or mount tree that is never seen by the filesystem itself.
1973  *
1974  * (4) To turn a superblock ro the "ro" flag must be used with
1975  *     fsconfig(FSCONFIG_SET_FLAG, "ro"). This option is seen by the filesystem
1976  *     in fc->sb_flags.
1977  *
1978  * But, currently the util-linux mount command already utilizes the new mount
1979  * API and is still setting fsconfig(FSCONFIG_SET_FLAG, "ro") no matter if it's
1980  * btrfs or not, setting the whole super block RO.  To make per-subvolume mounting
1981  * work with different options work we need to keep backward compatibility.
1982  */
1983 static int btrfs_reconfigure_for_mount(struct fs_context *fc, struct vfsmount *mnt)
1984 {
1985         int ret = 0;
1986
1987         if (fc->sb_flags & SB_RDONLY)
1988                 return ret;
1989
1990         down_write(&mnt->mnt_sb->s_umount);
1991         if (!(fc->sb_flags & SB_RDONLY) && (mnt->mnt_sb->s_flags & SB_RDONLY))
1992                 ret = btrfs_reconfigure(fc);
1993         up_write(&mnt->mnt_sb->s_umount);
1994         return ret;
1995 }
1996
1997 static int btrfs_get_tree_subvol(struct fs_context *fc)
1998 {
1999         struct btrfs_fs_info *fs_info = NULL;
2000         struct btrfs_fs_context *ctx = fc->fs_private;
2001         struct fs_context *dup_fc;
2002         struct dentry *dentry;
2003         struct vfsmount *mnt;
2004         int ret = 0;
2005
2006         /*
2007          * Setup a dummy root and fs_info for test/set super.  This is because
2008          * we don't actually fill this stuff out until open_ctree, but we need
2009          * then open_ctree will properly initialize the file system specific
2010          * settings later.  btrfs_init_fs_info initializes the static elements
2011          * of the fs_info (locks and such) to make cleanup easier if we find a
2012          * superblock with our given fs_devices later on at sget() time.
2013          */
2014         fs_info = kvzalloc(sizeof(struct btrfs_fs_info), GFP_KERNEL);
2015         if (!fs_info)
2016                 return -ENOMEM;
2017
2018         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_KERNEL);
2019         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_KERNEL);
2020         if (!fs_info->super_copy || !fs_info->super_for_commit) {
2021                 btrfs_free_fs_info(fs_info);
2022                 return -ENOMEM;
2023         }
2024         btrfs_init_fs_info(fs_info);
2025
2026         dup_fc = vfs_dup_fs_context(fc);
2027         if (IS_ERR(dup_fc)) {
2028                 btrfs_free_fs_info(fs_info);
2029                 return PTR_ERR(dup_fc);
2030         }
2031
2032         /*
2033          * When we do the sget_fc this gets transferred to the sb, so we only
2034          * need to set it on the dup_fc as this is what creates the super block.
2035          */
2036         dup_fc->s_fs_info = fs_info;
2037
2038         /*
2039          * We'll do the security settings in our btrfs_get_tree_super() mount
2040          * loop, they were duplicated into dup_fc, we can drop the originals
2041          * here.
2042          */
2043         security_free_mnt_opts(&fc->security);
2044         fc->security = NULL;
2045
2046         mnt = fc_mount(dup_fc);
2047         if (IS_ERR(mnt)) {
2048                 put_fs_context(dup_fc);
2049                 return PTR_ERR(mnt);
2050         }
2051         ret = btrfs_reconfigure_for_mount(dup_fc, mnt);
2052         put_fs_context(dup_fc);
2053         if (ret) {
2054                 mntput(mnt);
2055                 return ret;
2056         }
2057
2058         /*
2059          * This free's ->subvol_name, because if it isn't set we have to
2060          * allocate a buffer to hold the subvol_name, so we just drop our
2061          * reference to it here.
2062          */
2063         dentry = mount_subvol(ctx->subvol_name, ctx->subvol_objectid, mnt);
2064         ctx->subvol_name = NULL;
2065         if (IS_ERR(dentry))
2066                 return PTR_ERR(dentry);
2067
2068         fc->root = dentry;
2069         return 0;
2070 }
2071
2072 static int btrfs_get_tree(struct fs_context *fc)
2073 {
2074         /*
2075          * Since we use mount_subtree to mount the default/specified subvol, we
2076          * have to do mounts in two steps.
2077          *
2078          * First pass through we call btrfs_get_tree_subvol(), this is just a
2079          * wrapper around fc_mount() to call back into here again, and this time
2080          * we'll call btrfs_get_tree_super().  This will do the open_ctree() and
2081          * everything to open the devices and file system.  Then we return back
2082          * with a fully constructed vfsmount in btrfs_get_tree_subvol(), and
2083          * from there we can do our mount_subvol() call, which will lookup
2084          * whichever subvol we're mounting and setup this fc with the
2085          * appropriate dentry for the subvol.
2086          */
2087         if (fc->s_fs_info)
2088                 return btrfs_get_tree_super(fc);
2089         return btrfs_get_tree_subvol(fc);
2090 }
2091
2092 static void btrfs_kill_super(struct super_block *sb)
2093 {
2094         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2095         kill_anon_super(sb);
2096         btrfs_free_fs_info(fs_info);
2097 }
2098
2099 static void btrfs_free_fs_context(struct fs_context *fc)
2100 {
2101         struct btrfs_fs_context *ctx = fc->fs_private;
2102         struct btrfs_fs_info *fs_info = fc->s_fs_info;
2103
2104         if (fs_info)
2105                 btrfs_free_fs_info(fs_info);
2106
2107         if (ctx && refcount_dec_and_test(&ctx->refs)) {
2108                 kfree(ctx->subvol_name);
2109                 kfree(ctx);
2110         }
2111 }
2112
2113 static int btrfs_dup_fs_context(struct fs_context *fc, struct fs_context *src_fc)
2114 {
2115         struct btrfs_fs_context *ctx = src_fc->fs_private;
2116
2117         /*
2118          * Give a ref to our ctx to this dup, as we want to keep it around for
2119          * our original fc so we can have the subvolume name or objectid.
2120          *
2121          * We unset ->source in the original fc because the dup needs it for
2122          * mounting, and then once we free the dup it'll free ->source, so we
2123          * need to make sure we're only pointing to it in one fc.
2124          */
2125         refcount_inc(&ctx->refs);
2126         fc->fs_private = ctx;
2127         fc->source = src_fc->source;
2128         src_fc->source = NULL;
2129         return 0;
2130 }
2131
2132 static const struct fs_context_operations btrfs_fs_context_ops = {
2133         .parse_param    = btrfs_parse_param,
2134         .reconfigure    = btrfs_reconfigure,
2135         .get_tree       = btrfs_get_tree,
2136         .dup            = btrfs_dup_fs_context,
2137         .free           = btrfs_free_fs_context,
2138 };
2139
2140 static int btrfs_init_fs_context(struct fs_context *fc)
2141 {
2142         struct btrfs_fs_context *ctx;
2143
2144         ctx = kzalloc(sizeof(struct btrfs_fs_context), GFP_KERNEL);
2145         if (!ctx)
2146                 return -ENOMEM;
2147
2148         refcount_set(&ctx->refs, 1);
2149         fc->fs_private = ctx;
2150         fc->ops = &btrfs_fs_context_ops;
2151
2152         if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
2153                 btrfs_info_to_ctx(btrfs_sb(fc->root->d_sb), ctx);
2154         } else {
2155                 ctx->thread_pool_size =
2156                         min_t(unsigned long, num_online_cpus() + 2, 8);
2157                 ctx->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2158                 ctx->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2159         }
2160
2161 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
2162         fc->sb_flags |= SB_POSIXACL;
2163 #endif
2164         fc->sb_flags |= SB_I_VERSION;
2165
2166         return 0;
2167 }
2168
2169 static struct file_system_type btrfs_fs_type = {
2170         .owner                  = THIS_MODULE,
2171         .name                   = "btrfs",
2172         .init_fs_context        = btrfs_init_fs_context,
2173         .parameters             = btrfs_fs_parameters,
2174         .kill_sb                = btrfs_kill_super,
2175         .fs_flags               = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA |
2176                                   FS_ALLOW_IDMAP | FS_MGTIME,
2177  };
2178
2179 MODULE_ALIAS_FS("btrfs");
2180
2181 static int btrfs_control_open(struct inode *inode, struct file *file)
2182 {
2183         /*
2184          * The control file's private_data is used to hold the
2185          * transaction when it is started and is used to keep
2186          * track of whether a transaction is already in progress.
2187          */
2188         file->private_data = NULL;
2189         return 0;
2190 }
2191
2192 /*
2193  * Used by /dev/btrfs-control for devices ioctls.
2194  */
2195 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
2196                                 unsigned long arg)
2197 {
2198         struct btrfs_ioctl_vol_args *vol;
2199         struct btrfs_device *device = NULL;
2200         dev_t devt = 0;
2201         int ret = -ENOTTY;
2202
2203         if (!capable(CAP_SYS_ADMIN))
2204                 return -EPERM;
2205
2206         vol = memdup_user((void __user *)arg, sizeof(*vol));
2207         if (IS_ERR(vol))
2208                 return PTR_ERR(vol);
2209         ret = btrfs_check_ioctl_vol_args_path(vol);
2210         if (ret < 0)
2211                 goto out;
2212
2213         switch (cmd) {
2214         case BTRFS_IOC_SCAN_DEV:
2215                 mutex_lock(&uuid_mutex);
2216                 /*
2217                  * Scanning outside of mount can return NULL which would turn
2218                  * into 0 error code.
2219                  */
2220                 device = btrfs_scan_one_device(vol->name, BLK_OPEN_READ, false);
2221                 ret = PTR_ERR_OR_ZERO(device);
2222                 mutex_unlock(&uuid_mutex);
2223                 break;
2224         case BTRFS_IOC_FORGET_DEV:
2225                 if (vol->name[0] != 0) {
2226                         ret = lookup_bdev(vol->name, &devt);
2227                         if (ret)
2228                                 break;
2229                 }
2230                 ret = btrfs_forget_devices(devt);
2231                 break;
2232         case BTRFS_IOC_DEVICES_READY:
2233                 mutex_lock(&uuid_mutex);
2234                 /*
2235                  * Scanning outside of mount can return NULL which would turn
2236                  * into 0 error code.
2237                  */
2238                 device = btrfs_scan_one_device(vol->name, BLK_OPEN_READ, false);
2239                 if (IS_ERR_OR_NULL(device)) {
2240                         mutex_unlock(&uuid_mutex);
2241                         if (IS_ERR(device))
2242                                 ret = PTR_ERR(device);
2243                         else
2244                                 ret = 0;
2245                         break;
2246                 }
2247                 ret = !(device->fs_devices->num_devices ==
2248                         device->fs_devices->total_devices);
2249                 mutex_unlock(&uuid_mutex);
2250                 break;
2251         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
2252                 ret = btrfs_ioctl_get_supported_features((void __user*)arg);
2253                 break;
2254         }
2255
2256 out:
2257         kfree(vol);
2258         return ret;
2259 }
2260
2261 static int btrfs_freeze(struct super_block *sb)
2262 {
2263         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2264
2265         set_bit(BTRFS_FS_FROZEN, &fs_info->flags);
2266         /*
2267          * We don't need a barrier here, we'll wait for any transaction that
2268          * could be in progress on other threads (and do delayed iputs that
2269          * we want to avoid on a frozen filesystem), or do the commit
2270          * ourselves.
2271          */
2272         return btrfs_commit_current_transaction(fs_info->tree_root);
2273 }
2274
2275 static int check_dev_super(struct btrfs_device *dev)
2276 {
2277         struct btrfs_fs_info *fs_info = dev->fs_info;
2278         struct btrfs_super_block *sb;
2279         u64 last_trans;
2280         u16 csum_type;
2281         int ret = 0;
2282
2283         /* This should be called with fs still frozen. */
2284         ASSERT(test_bit(BTRFS_FS_FROZEN, &fs_info->flags));
2285
2286         /* Missing dev, no need to check. */
2287         if (!dev->bdev)
2288                 return 0;
2289
2290         /* Only need to check the primary super block. */
2291         sb = btrfs_read_disk_super(dev->bdev, 0, true);
2292         if (IS_ERR(sb))
2293                 return PTR_ERR(sb);
2294
2295         /* Verify the checksum. */
2296         csum_type = btrfs_super_csum_type(sb);
2297         if (csum_type != btrfs_super_csum_type(fs_info->super_copy)) {
2298                 btrfs_err(fs_info, "csum type changed, has %u expect %u",
2299                           csum_type, btrfs_super_csum_type(fs_info->super_copy));
2300                 ret = -EUCLEAN;
2301                 goto out;
2302         }
2303
2304         if (btrfs_check_super_csum(fs_info, sb)) {
2305                 btrfs_err(fs_info, "csum for on-disk super block no longer matches");
2306                 ret = -EUCLEAN;
2307                 goto out;
2308         }
2309
2310         /* Btrfs_validate_super() includes fsid check against super->fsid. */
2311         ret = btrfs_validate_super(fs_info, sb, 0);
2312         if (ret < 0)
2313                 goto out;
2314
2315         last_trans = btrfs_get_last_trans_committed(fs_info);
2316         if (btrfs_super_generation(sb) != last_trans) {
2317                 btrfs_err(fs_info, "transid mismatch, has %llu expect %llu",
2318                           btrfs_super_generation(sb), last_trans);
2319                 ret = -EUCLEAN;
2320                 goto out;
2321         }
2322 out:
2323         btrfs_release_disk_super(sb);
2324         return ret;
2325 }
2326
2327 static int btrfs_unfreeze(struct super_block *sb)
2328 {
2329         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2330         struct btrfs_device *device;
2331         int ret = 0;
2332
2333         /*
2334          * Make sure the fs is not changed by accident (like hibernation then
2335          * modified by other OS).
2336          * If we found anything wrong, we mark the fs error immediately.
2337          *
2338          * And since the fs is frozen, no one can modify the fs yet, thus
2339          * we don't need to hold device_list_mutex.
2340          */
2341         list_for_each_entry(device, &fs_info->fs_devices->devices, dev_list) {
2342                 ret = check_dev_super(device);
2343                 if (ret < 0) {
2344                         btrfs_handle_fs_error(fs_info, ret,
2345                                 "super block on devid %llu got modified unexpectedly",
2346                                 device->devid);
2347                         break;
2348                 }
2349         }
2350         clear_bit(BTRFS_FS_FROZEN, &fs_info->flags);
2351
2352         /*
2353          * We still return 0, to allow VFS layer to unfreeze the fs even the
2354          * above checks failed. Since the fs is either fine or read-only, we're
2355          * safe to continue, without causing further damage.
2356          */
2357         return 0;
2358 }
2359
2360 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
2361 {
2362         struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
2363
2364         /*
2365          * There should be always a valid pointer in latest_dev, it may be stale
2366          * for a short moment in case it's being deleted but still valid until
2367          * the end of RCU grace period.
2368          */
2369         rcu_read_lock();
2370         seq_escape(m, btrfs_dev_name(fs_info->fs_devices->latest_dev), " \t\n\\");
2371         rcu_read_unlock();
2372
2373         return 0;
2374 }
2375
2376 static long btrfs_nr_cached_objects(struct super_block *sb, struct shrink_control *sc)
2377 {
2378         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2379         const s64 nr = percpu_counter_sum_positive(&fs_info->evictable_extent_maps);
2380
2381         trace_btrfs_extent_map_shrinker_count(fs_info, nr);
2382
2383         return nr;
2384 }
2385
2386 static long btrfs_free_cached_objects(struct super_block *sb, struct shrink_control *sc)
2387 {
2388         const long nr_to_scan = min_t(unsigned long, LONG_MAX, sc->nr_to_scan);
2389         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2390
2391         btrfs_free_extent_maps(fs_info, nr_to_scan);
2392
2393         /* The extent map shrinker runs asynchronously, so always return 0. */
2394         return 0;
2395 }
2396
2397 static const struct super_operations btrfs_super_ops = {
2398         .drop_inode     = btrfs_drop_inode,
2399         .evict_inode    = btrfs_evict_inode,
2400         .put_super      = btrfs_put_super,
2401         .sync_fs        = btrfs_sync_fs,
2402         .show_options   = btrfs_show_options,
2403         .show_devname   = btrfs_show_devname,
2404         .alloc_inode    = btrfs_alloc_inode,
2405         .destroy_inode  = btrfs_destroy_inode,
2406         .free_inode     = btrfs_free_inode,
2407         .statfs         = btrfs_statfs,
2408         .freeze_fs      = btrfs_freeze,
2409         .unfreeze_fs    = btrfs_unfreeze,
2410         .nr_cached_objects = btrfs_nr_cached_objects,
2411         .free_cached_objects = btrfs_free_cached_objects,
2412 };
2413
2414 static const struct file_operations btrfs_ctl_fops = {
2415         .open = btrfs_control_open,
2416         .unlocked_ioctl  = btrfs_control_ioctl,
2417         .compat_ioctl = compat_ptr_ioctl,
2418         .owner   = THIS_MODULE,
2419         .llseek = noop_llseek,
2420 };
2421
2422 static struct miscdevice btrfs_misc = {
2423         .minor          = BTRFS_MINOR,
2424         .name           = "btrfs-control",
2425         .fops           = &btrfs_ctl_fops
2426 };
2427
2428 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
2429 MODULE_ALIAS("devname:btrfs-control");
2430
2431 static int __init btrfs_interface_init(void)
2432 {
2433         return misc_register(&btrfs_misc);
2434 }
2435
2436 static __cold void btrfs_interface_exit(void)
2437 {
2438         misc_deregister(&btrfs_misc);
2439 }
2440
2441 static int __init btrfs_print_mod_info(void)
2442 {
2443         static const char options[] = ""
2444 #ifdef CONFIG_BTRFS_EXPERIMENTAL
2445                         ", experimental=on"
2446 #endif
2447 #ifdef CONFIG_BTRFS_DEBUG
2448                         ", debug=on"
2449 #endif
2450 #ifdef CONFIG_BTRFS_ASSERT
2451                         ", assert=on"
2452 #endif
2453 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
2454                         ", ref-verify=on"
2455 #endif
2456 #ifdef CONFIG_BLK_DEV_ZONED
2457                         ", zoned=yes"
2458 #else
2459                         ", zoned=no"
2460 #endif
2461 #ifdef CONFIG_FS_VERITY
2462                         ", fsverity=yes"
2463 #else
2464                         ", fsverity=no"
2465 #endif
2466                         ;
2467
2468 #ifdef CONFIG_BTRFS_EXPERIMENTAL
2469         if (btrfs_get_mod_read_policy() == NULL)
2470                 pr_info("Btrfs loaded%s\n", options);
2471         else
2472                 pr_info("Btrfs loaded%s, read_policy=%s\n",
2473                          options, btrfs_get_mod_read_policy());
2474 #else
2475         pr_info("Btrfs loaded%s\n", options);
2476 #endif
2477
2478         return 0;
2479 }
2480
2481 static int register_btrfs(void)
2482 {
2483         return register_filesystem(&btrfs_fs_type);
2484 }
2485
2486 static void unregister_btrfs(void)
2487 {
2488         unregister_filesystem(&btrfs_fs_type);
2489 }
2490
2491 /* Helper structure for long init/exit functions. */
2492 struct init_sequence {
2493         int (*init_func)(void);
2494         /* Can be NULL if the init_func doesn't need cleanup. */
2495         void (*exit_func)(void);
2496 };
2497
2498 static const struct init_sequence mod_init_seq[] = {
2499         {
2500                 .init_func = btrfs_props_init,
2501                 .exit_func = NULL,
2502         }, {
2503                 .init_func = btrfs_init_sysfs,
2504                 .exit_func = btrfs_exit_sysfs,
2505         }, {
2506                 .init_func = btrfs_init_compress,
2507                 .exit_func = btrfs_exit_compress,
2508         }, {
2509                 .init_func = btrfs_init_cachep,
2510                 .exit_func = btrfs_destroy_cachep,
2511         }, {
2512                 .init_func = btrfs_init_dio,
2513                 .exit_func = btrfs_destroy_dio,
2514         }, {
2515                 .init_func = btrfs_transaction_init,
2516                 .exit_func = btrfs_transaction_exit,
2517         }, {
2518                 .init_func = btrfs_ctree_init,
2519                 .exit_func = btrfs_ctree_exit,
2520         }, {
2521                 .init_func = btrfs_free_space_init,
2522                 .exit_func = btrfs_free_space_exit,
2523         }, {
2524                 .init_func = btrfs_extent_state_init_cachep,
2525                 .exit_func = btrfs_extent_state_free_cachep,
2526         }, {
2527                 .init_func = extent_buffer_init_cachep,
2528                 .exit_func = extent_buffer_free_cachep,
2529         }, {
2530                 .init_func = btrfs_bioset_init,
2531                 .exit_func = btrfs_bioset_exit,
2532         }, {
2533                 .init_func = btrfs_extent_map_init,
2534                 .exit_func = btrfs_extent_map_exit,
2535 #ifdef CONFIG_BTRFS_EXPERIMENTAL
2536         }, {
2537                 .init_func = btrfs_read_policy_init,
2538                 .exit_func = NULL,
2539 #endif
2540         }, {
2541                 .init_func = ordered_data_init,
2542                 .exit_func = ordered_data_exit,
2543         }, {
2544                 .init_func = btrfs_delayed_inode_init,
2545                 .exit_func = btrfs_delayed_inode_exit,
2546         }, {
2547                 .init_func = btrfs_auto_defrag_init,
2548                 .exit_func = btrfs_auto_defrag_exit,
2549         }, {
2550                 .init_func = btrfs_delayed_ref_init,
2551                 .exit_func = btrfs_delayed_ref_exit,
2552         }, {
2553                 .init_func = btrfs_prelim_ref_init,
2554                 .exit_func = btrfs_prelim_ref_exit,
2555         }, {
2556                 .init_func = btrfs_interface_init,
2557                 .exit_func = btrfs_interface_exit,
2558         }, {
2559                 .init_func = btrfs_print_mod_info,
2560                 .exit_func = NULL,
2561         }, {
2562                 .init_func = btrfs_run_sanity_tests,
2563                 .exit_func = NULL,
2564         }, {
2565                 .init_func = register_btrfs,
2566                 .exit_func = unregister_btrfs,
2567         }
2568 };
2569
2570 static bool mod_init_result[ARRAY_SIZE(mod_init_seq)];
2571
2572 static __always_inline void btrfs_exit_btrfs_fs(void)
2573 {
2574         int i;
2575
2576         for (i = ARRAY_SIZE(mod_init_seq) - 1; i >= 0; i--) {
2577                 if (!mod_init_result[i])
2578                         continue;
2579                 if (mod_init_seq[i].exit_func)
2580                         mod_init_seq[i].exit_func();
2581                 mod_init_result[i] = false;
2582         }
2583 }
2584
2585 static void __exit exit_btrfs_fs(void)
2586 {
2587         btrfs_exit_btrfs_fs();
2588         btrfs_cleanup_fs_uuids();
2589 }
2590
2591 static int __init init_btrfs_fs(void)
2592 {
2593         int ret;
2594         int i;
2595
2596         for (i = 0; i < ARRAY_SIZE(mod_init_seq); i++) {
2597                 ASSERT(!mod_init_result[i]);
2598                 ret = mod_init_seq[i].init_func();
2599                 if (ret < 0) {
2600                         btrfs_exit_btrfs_fs();
2601                         return ret;
2602                 }
2603                 mod_init_result[i] = true;
2604         }
2605         return 0;
2606 }
2607
2608 late_initcall(init_btrfs_fs);
2609 module_exit(exit_btrfs_fs)
2610
2611 MODULE_DESCRIPTION("B-Tree File System (BTRFS)");
2612 MODULE_LICENSE("GPL");
2613 MODULE_SOFTDEP("pre: crc32c");
2614 MODULE_SOFTDEP("pre: xxhash64");
2615 MODULE_SOFTDEP("pre: sha256");
2616 MODULE_SOFTDEP("pre: blake2b-256");