1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (c) 2000-2006 Silicon Graphics, Inc. 4 * All Rights Reserved. 5 */ 6 7 #include "xfs.h" 8 #include "xfs_shared.h" 9 #include "xfs_format.h" 10 #include "xfs_log_format.h" 11 #include "xfs_trans_resv.h" 12 #include "xfs_sb.h" 13 #include "xfs_mount.h" 14 #include "xfs_inode.h" 15 #include "xfs_btree.h" 16 #include "xfs_bmap.h" 17 #include "xfs_alloc.h" 18 #include "xfs_fsops.h" 19 #include "xfs_trans.h" 20 #include "xfs_buf_item.h" 21 #include "xfs_log.h" 22 #include "xfs_log_priv.h" 23 #include "xfs_dir2.h" 24 #include "xfs_extfree_item.h" 25 #include "xfs_mru_cache.h" 26 #include "xfs_inode_item.h" 27 #include "xfs_icache.h" 28 #include "xfs_trace.h" 29 #include "xfs_icreate_item.h" 30 #include "xfs_filestream.h" 31 #include "xfs_quota.h" 32 #include "xfs_sysfs.h" 33 #include "xfs_ondisk.h" 34 #include "xfs_rmap_item.h" 35 #include "xfs_refcount_item.h" 36 #include "xfs_bmap_item.h" 37 #include "xfs_reflink.h" 38 #include "xfs_pwork.h" 39 #include "xfs_ag.h" 40 #include "xfs_defer.h" 41 #include "xfs_attr_item.h" 42 #include "xfs_xattr.h" 43 #include "xfs_iunlink_item.h" 44 #include "xfs_dahash_test.h" 45 #include "xfs_rtbitmap.h" 46 #include "xfs_exchmaps_item.h" 47 #include "xfs_parent.h" 48 #include "xfs_rtalloc.h" 49 #include "xfs_zone_alloc.h" 50 #include "scrub/stats.h" 51 #include "scrub/rcbag_btree.h" 52 53 #include <linux/magic.h> 54 #include <linux/fs_context.h> 55 #include <linux/fs_parser.h> 56 57 static const struct super_operations xfs_super_operations; 58 59 static struct dentry *xfs_debugfs; /* top-level xfs debugfs dir */ 60 static struct kset *xfs_kset; /* top-level xfs sysfs dir */ 61 #ifdef DEBUG 62 static struct xfs_kobj xfs_dbg_kobj; /* global debug sysfs attrs */ 63 #endif 64 65 enum xfs_dax_mode { 66 XFS_DAX_INODE = 0, 67 XFS_DAX_ALWAYS = 1, 68 XFS_DAX_NEVER = 2, 69 }; 70 71 /* Were quota mount options provided? Must use the upper 16 bits of qflags. */ 72 #define XFS_QFLAGS_MNTOPTS (1U << 31) 73 74 static void 75 xfs_mount_set_dax_mode( 76 struct xfs_mount *mp, 77 enum xfs_dax_mode mode) 78 { 79 switch (mode) { 80 case XFS_DAX_INODE: 81 mp->m_features &= ~(XFS_FEAT_DAX_ALWAYS | XFS_FEAT_DAX_NEVER); 82 break; 83 case XFS_DAX_ALWAYS: 84 mp->m_features |= XFS_FEAT_DAX_ALWAYS; 85 mp->m_features &= ~XFS_FEAT_DAX_NEVER; 86 break; 87 case XFS_DAX_NEVER: 88 mp->m_features |= XFS_FEAT_DAX_NEVER; 89 mp->m_features &= ~XFS_FEAT_DAX_ALWAYS; 90 break; 91 } 92 } 93 94 static const struct constant_table dax_param_enums[] = { 95 {"inode", XFS_DAX_INODE }, 96 {"always", XFS_DAX_ALWAYS }, 97 {"never", XFS_DAX_NEVER }, 98 {} 99 }; 100 101 /* 102 * Table driven mount option parser. 103 */ 104 enum { 105 Op_deprecated, Opt_logbufs, Opt_logbsize, Opt_logdev, Opt_rtdev, 106 Opt_wsync, Opt_noalign, Opt_swalloc, Opt_sunit, Opt_swidth, Opt_nouuid, 107 Opt_grpid, Opt_nogrpid, Opt_bsdgroups, Opt_sysvgroups, 108 Opt_allocsize, Opt_norecovery, Opt_inode64, Opt_inode32, 109 Opt_largeio, Opt_nolargeio, 110 Opt_filestreams, Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota, 111 Opt_prjquota, Opt_uquota, Opt_gquota, Opt_pquota, 112 Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce, 113 Opt_discard, Opt_nodiscard, Opt_dax, Opt_dax_enum, Opt_max_open_zones, 114 Opt_lifetime, Opt_nolifetime, Opt_max_atomic_write, 115 }; 116 117 #define fsparam_dead(NAME) \ 118 __fsparam(NULL, (NAME), Op_deprecated, fs_param_deprecated, NULL) 119 120 static const struct fs_parameter_spec xfs_fs_parameters[] = { 121 /* 122 * These mount options were supposed to be deprecated in September 2025 123 * but the deprecation warning was buggy, so not all users were 124 * notified. The deprecation is now obnoxiously loud and postponed to 125 * September 2030. 126 */ 127 fsparam_dead("attr2"), 128 fsparam_dead("noattr2"), 129 fsparam_dead("ikeep"), 130 fsparam_dead("noikeep"), 131 132 fsparam_u32("logbufs", Opt_logbufs), 133 fsparam_string("logbsize", Opt_logbsize), 134 fsparam_string("logdev", Opt_logdev), 135 fsparam_string("rtdev", Opt_rtdev), 136 fsparam_flag("wsync", Opt_wsync), 137 fsparam_flag("noalign", Opt_noalign), 138 fsparam_flag("swalloc", Opt_swalloc), 139 fsparam_u32("sunit", Opt_sunit), 140 fsparam_u32("swidth", Opt_swidth), 141 fsparam_flag("nouuid", Opt_nouuid), 142 fsparam_flag("grpid", Opt_grpid), 143 fsparam_flag("nogrpid", Opt_nogrpid), 144 fsparam_flag("bsdgroups", Opt_bsdgroups), 145 fsparam_flag("sysvgroups", Opt_sysvgroups), 146 fsparam_string("allocsize", Opt_allocsize), 147 fsparam_flag("norecovery", Opt_norecovery), 148 fsparam_flag("inode64", Opt_inode64), 149 fsparam_flag("inode32", Opt_inode32), 150 fsparam_flag("largeio", Opt_largeio), 151 fsparam_flag("nolargeio", Opt_nolargeio), 152 fsparam_flag("filestreams", Opt_filestreams), 153 fsparam_flag("quota", Opt_quota), 154 fsparam_flag("noquota", Opt_noquota), 155 fsparam_flag("usrquota", Opt_usrquota), 156 fsparam_flag("grpquota", Opt_grpquota), 157 fsparam_flag("prjquota", Opt_prjquota), 158 fsparam_flag("uquota", Opt_uquota), 159 fsparam_flag("gquota", Opt_gquota), 160 fsparam_flag("pquota", Opt_pquota), 161 fsparam_flag("uqnoenforce", Opt_uqnoenforce), 162 fsparam_flag("gqnoenforce", Opt_gqnoenforce), 163 fsparam_flag("pqnoenforce", Opt_pqnoenforce), 164 fsparam_flag("qnoenforce", Opt_qnoenforce), 165 fsparam_flag("discard", Opt_discard), 166 fsparam_flag("nodiscard", Opt_nodiscard), 167 fsparam_flag("dax", Opt_dax), 168 fsparam_enum("dax", Opt_dax_enum, dax_param_enums), 169 fsparam_u32("max_open_zones", Opt_max_open_zones), 170 fsparam_flag("lifetime", Opt_lifetime), 171 fsparam_flag("nolifetime", Opt_nolifetime), 172 fsparam_string("max_atomic_write", Opt_max_atomic_write), 173 {} 174 }; 175 176 struct proc_xfs_info { 177 uint64_t flag; 178 char *str; 179 }; 180 181 static int 182 xfs_fs_show_options( 183 struct seq_file *m, 184 struct dentry *root) 185 { 186 static struct proc_xfs_info xfs_info_set[] = { 187 /* the few simple ones we can get from the mount struct */ 188 { XFS_FEAT_WSYNC, ",wsync" }, 189 { XFS_FEAT_NOALIGN, ",noalign" }, 190 { XFS_FEAT_SWALLOC, ",swalloc" }, 191 { XFS_FEAT_NOUUID, ",nouuid" }, 192 { XFS_FEAT_NORECOVERY, ",norecovery" }, 193 { XFS_FEAT_FILESTREAMS, ",filestreams" }, 194 { XFS_FEAT_GRPID, ",grpid" }, 195 { XFS_FEAT_DISCARD, ",discard" }, 196 { XFS_FEAT_LARGE_IOSIZE, ",largeio" }, 197 { XFS_FEAT_DAX_ALWAYS, ",dax=always" }, 198 { XFS_FEAT_DAX_NEVER, ",dax=never" }, 199 { XFS_FEAT_NOLIFETIME, ",nolifetime" }, 200 { 0, NULL } 201 }; 202 struct xfs_mount *mp = XFS_M(root->d_sb); 203 struct proc_xfs_info *xfs_infop; 204 205 for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) { 206 if (mp->m_features & xfs_infop->flag) 207 seq_puts(m, xfs_infop->str); 208 } 209 210 seq_printf(m, ",inode%d", xfs_has_small_inums(mp) ? 32 : 64); 211 212 if (xfs_has_allocsize(mp)) 213 seq_printf(m, ",allocsize=%dk", 214 (1 << mp->m_allocsize_log) >> 10); 215 216 if (mp->m_logbufs > 0) 217 seq_printf(m, ",logbufs=%d", mp->m_logbufs); 218 if (mp->m_logbsize > 0) 219 seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10); 220 221 if (mp->m_logname) 222 seq_show_option(m, "logdev", mp->m_logname); 223 if (mp->m_rtname) 224 seq_show_option(m, "rtdev", mp->m_rtname); 225 226 if (mp->m_dalign > 0) 227 seq_printf(m, ",sunit=%d", 228 (int)XFS_FSB_TO_BB(mp, mp->m_dalign)); 229 if (mp->m_swidth > 0) 230 seq_printf(m, ",swidth=%d", 231 (int)XFS_FSB_TO_BB(mp, mp->m_swidth)); 232 233 if (mp->m_qflags & XFS_UQUOTA_ENFD) 234 seq_puts(m, ",usrquota"); 235 else if (mp->m_qflags & XFS_UQUOTA_ACCT) 236 seq_puts(m, ",uqnoenforce"); 237 238 if (mp->m_qflags & XFS_PQUOTA_ENFD) 239 seq_puts(m, ",prjquota"); 240 else if (mp->m_qflags & XFS_PQUOTA_ACCT) 241 seq_puts(m, ",pqnoenforce"); 242 243 if (mp->m_qflags & XFS_GQUOTA_ENFD) 244 seq_puts(m, ",grpquota"); 245 else if (mp->m_qflags & XFS_GQUOTA_ACCT) 246 seq_puts(m, ",gqnoenforce"); 247 248 if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT)) 249 seq_puts(m, ",noquota"); 250 251 if (mp->m_max_open_zones) 252 seq_printf(m, ",max_open_zones=%u", mp->m_max_open_zones); 253 if (mp->m_awu_max_bytes) 254 seq_printf(m, ",max_atomic_write=%lluk", 255 mp->m_awu_max_bytes >> 10); 256 257 return 0; 258 } 259 260 static bool 261 xfs_set_inode_alloc_perag( 262 struct xfs_perag *pag, 263 xfs_ino_t ino, 264 xfs_agnumber_t max_metadata) 265 { 266 if (!xfs_is_inode32(pag_mount(pag))) { 267 set_bit(XFS_AGSTATE_ALLOWS_INODES, &pag->pag_opstate); 268 clear_bit(XFS_AGSTATE_PREFERS_METADATA, &pag->pag_opstate); 269 return false; 270 } 271 272 if (ino > XFS_MAXINUMBER_32) { 273 clear_bit(XFS_AGSTATE_ALLOWS_INODES, &pag->pag_opstate); 274 clear_bit(XFS_AGSTATE_PREFERS_METADATA, &pag->pag_opstate); 275 return false; 276 } 277 278 set_bit(XFS_AGSTATE_ALLOWS_INODES, &pag->pag_opstate); 279 if (pag_agno(pag) < max_metadata) 280 set_bit(XFS_AGSTATE_PREFERS_METADATA, &pag->pag_opstate); 281 else 282 clear_bit(XFS_AGSTATE_PREFERS_METADATA, &pag->pag_opstate); 283 return true; 284 } 285 286 /* 287 * Set parameters for inode allocation heuristics, taking into account 288 * filesystem size and inode32/inode64 mount options; i.e. specifically 289 * whether or not XFS_FEAT_SMALL_INUMS is set. 290 * 291 * Inode allocation patterns are altered only if inode32 is requested 292 * (XFS_FEAT_SMALL_INUMS), and the filesystem is sufficiently large. 293 * If altered, XFS_OPSTATE_INODE32 is set as well. 294 * 295 * An agcount independent of that in the mount structure is provided 296 * because in the growfs case, mp->m_sb.sb_agcount is not yet updated 297 * to the potentially higher ag count. 298 * 299 * Returns the maximum AG index which may contain inodes. 300 */ 301 xfs_agnumber_t 302 xfs_set_inode_alloc( 303 struct xfs_mount *mp, 304 xfs_agnumber_t agcount) 305 { 306 xfs_agnumber_t index; 307 xfs_agnumber_t maxagi = 0; 308 xfs_sb_t *sbp = &mp->m_sb; 309 xfs_agnumber_t max_metadata; 310 xfs_agino_t agino; 311 xfs_ino_t ino; 312 313 /* 314 * Calculate how much should be reserved for inodes to meet 315 * the max inode percentage. Used only for inode32. 316 */ 317 if (M_IGEO(mp)->maxicount) { 318 uint64_t icount; 319 320 icount = sbp->sb_dblocks * sbp->sb_imax_pct; 321 do_div(icount, 100); 322 icount += sbp->sb_agblocks - 1; 323 do_div(icount, sbp->sb_agblocks); 324 max_metadata = icount; 325 } else { 326 max_metadata = agcount; 327 } 328 329 /* Get the last possible inode in the filesystem */ 330 agino = XFS_AGB_TO_AGINO(mp, sbp->sb_agblocks - 1); 331 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino); 332 333 /* 334 * If user asked for no more than 32-bit inodes, and the fs is 335 * sufficiently large, set XFS_OPSTATE_INODE32 if we must alter 336 * the allocator to accommodate the request. 337 */ 338 if (xfs_has_small_inums(mp) && ino > XFS_MAXINUMBER_32) 339 xfs_set_inode32(mp); 340 else 341 xfs_clear_inode32(mp); 342 343 for (index = 0; index < agcount; index++) { 344 struct xfs_perag *pag; 345 346 ino = XFS_AGINO_TO_INO(mp, index, agino); 347 348 pag = xfs_perag_get(mp, index); 349 if (xfs_set_inode_alloc_perag(pag, ino, max_metadata)) 350 maxagi++; 351 xfs_perag_put(pag); 352 } 353 354 return xfs_is_inode32(mp) ? maxagi : agcount; 355 } 356 357 static int 358 xfs_setup_dax_always( 359 struct xfs_mount *mp) 360 { 361 if (!mp->m_ddev_targp->bt_daxdev && 362 (!mp->m_rtdev_targp || !mp->m_rtdev_targp->bt_daxdev)) { 363 xfs_alert(mp, 364 "DAX unsupported by block device. Turning off DAX."); 365 goto disable_dax; 366 } 367 368 if (mp->m_super->s_blocksize != PAGE_SIZE) { 369 xfs_alert(mp, 370 "DAX not supported for blocksize. Turning off DAX."); 371 goto disable_dax; 372 } 373 374 if (xfs_has_reflink(mp) && 375 bdev_is_partition(mp->m_ddev_targp->bt_bdev)) { 376 xfs_alert(mp, 377 "DAX and reflink cannot work with multi-partitions!"); 378 return -EINVAL; 379 } 380 381 return 0; 382 383 disable_dax: 384 xfs_mount_set_dax_mode(mp, XFS_DAX_NEVER); 385 return 0; 386 } 387 388 STATIC int 389 xfs_blkdev_get( 390 xfs_mount_t *mp, 391 const char *name, 392 struct file **bdev_filep) 393 { 394 int error = 0; 395 blk_mode_t mode; 396 397 mode = sb_open_mode(mp->m_super->s_flags); 398 *bdev_filep = bdev_file_open_by_path(name, mode, 399 mp->m_super, &fs_holder_ops); 400 if (IS_ERR(*bdev_filep)) { 401 error = PTR_ERR(*bdev_filep); 402 *bdev_filep = NULL; 403 xfs_warn(mp, "Invalid device [%s], error=%d", name, error); 404 } 405 406 return error; 407 } 408 409 STATIC void 410 xfs_shutdown_devices( 411 struct xfs_mount *mp) 412 { 413 /* 414 * Udev is triggered whenever anyone closes a block device or unmounts 415 * a file systemm on a block device. 416 * The default udev rules invoke blkid to read the fs super and create 417 * symlinks to the bdev under /dev/disk. For this, it uses buffered 418 * reads through the page cache. 419 * 420 * xfs_db also uses buffered reads to examine metadata. There is no 421 * coordination between xfs_db and udev, which means that they can run 422 * concurrently. Note there is no coordination between the kernel and 423 * blkid either. 424 * 425 * On a system with 64k pages, the page cache can cache the superblock 426 * and the root inode (and hence the root directory) with the same 64k 427 * page. If udev spawns blkid after the mkfs and the system is busy 428 * enough that it is still running when xfs_db starts up, they'll both 429 * read from the same page in the pagecache. 430 * 431 * The unmount writes updated inode metadata to disk directly. The XFS 432 * buffer cache does not use the bdev pagecache, so it needs to 433 * invalidate that pagecache on unmount. If the above scenario occurs, 434 * the pagecache no longer reflects what's on disk, xfs_db reads the 435 * stale metadata, and fails to find /a. Most of the time this succeeds 436 * because closing a bdev invalidates the page cache, but when processes 437 * race, everyone loses. 438 */ 439 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) { 440 blkdev_issue_flush(mp->m_logdev_targp->bt_bdev); 441 invalidate_bdev(mp->m_logdev_targp->bt_bdev); 442 } 443 if (mp->m_rtdev_targp) { 444 blkdev_issue_flush(mp->m_rtdev_targp->bt_bdev); 445 invalidate_bdev(mp->m_rtdev_targp->bt_bdev); 446 } 447 blkdev_issue_flush(mp->m_ddev_targp->bt_bdev); 448 invalidate_bdev(mp->m_ddev_targp->bt_bdev); 449 } 450 451 /* 452 * The file system configurations are: 453 * (1) device (partition) with data and internal log 454 * (2) logical volume with data and log subvolumes. 455 * (3) logical volume with data, log, and realtime subvolumes. 456 * 457 * We only have to handle opening the log and realtime volumes here if 458 * they are present. The data subvolume has already been opened by 459 * get_sb_bdev() and is stored in sb->s_bdev. 460 */ 461 STATIC int 462 xfs_open_devices( 463 struct xfs_mount *mp) 464 { 465 struct super_block *sb = mp->m_super; 466 struct block_device *ddev = sb->s_bdev; 467 struct file *logdev_file = NULL, *rtdev_file = NULL; 468 int error; 469 470 /* 471 * Open real time and log devices - order is important. 472 */ 473 if (mp->m_logname) { 474 error = xfs_blkdev_get(mp, mp->m_logname, &logdev_file); 475 if (error) 476 return error; 477 } 478 479 if (mp->m_rtname) { 480 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev_file); 481 if (error) 482 goto out_close_logdev; 483 484 if (file_bdev(rtdev_file) == ddev || 485 (logdev_file && 486 file_bdev(rtdev_file) == file_bdev(logdev_file))) { 487 xfs_warn(mp, 488 "Cannot mount filesystem with identical rtdev and ddev/logdev."); 489 error = -EINVAL; 490 goto out_close_rtdev; 491 } 492 } 493 494 /* 495 * Setup xfs_mount buffer target pointers 496 */ 497 mp->m_ddev_targp = xfs_alloc_buftarg(mp, sb->s_bdev_file); 498 if (IS_ERR(mp->m_ddev_targp)) { 499 error = PTR_ERR(mp->m_ddev_targp); 500 mp->m_ddev_targp = NULL; 501 goto out_close_rtdev; 502 } 503 504 if (rtdev_file) { 505 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev_file); 506 if (IS_ERR(mp->m_rtdev_targp)) { 507 error = PTR_ERR(mp->m_rtdev_targp); 508 mp->m_rtdev_targp = NULL; 509 goto out_free_ddev_targ; 510 } 511 } 512 513 if (logdev_file && file_bdev(logdev_file) != ddev) { 514 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev_file); 515 if (IS_ERR(mp->m_logdev_targp)) { 516 error = PTR_ERR(mp->m_logdev_targp); 517 mp->m_logdev_targp = NULL; 518 goto out_free_rtdev_targ; 519 } 520 } else { 521 mp->m_logdev_targp = mp->m_ddev_targp; 522 /* Handle won't be used, drop it */ 523 if (logdev_file) 524 bdev_fput(logdev_file); 525 } 526 527 return 0; 528 529 out_free_rtdev_targ: 530 if (mp->m_rtdev_targp) 531 xfs_free_buftarg(mp->m_rtdev_targp); 532 out_free_ddev_targ: 533 xfs_free_buftarg(mp->m_ddev_targp); 534 out_close_rtdev: 535 if (rtdev_file) 536 bdev_fput(rtdev_file); 537 out_close_logdev: 538 if (logdev_file) 539 bdev_fput(logdev_file); 540 return error; 541 } 542 543 /* 544 * Setup xfs_mount buffer target pointers based on superblock 545 */ 546 STATIC int 547 xfs_setup_devices( 548 struct xfs_mount *mp) 549 { 550 int error; 551 552 error = xfs_configure_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize, 553 mp->m_sb.sb_dblocks); 554 if (error) 555 return error; 556 557 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) { 558 unsigned int log_sector_size = BBSIZE; 559 560 if (xfs_has_sector(mp)) 561 log_sector_size = mp->m_sb.sb_logsectsize; 562 error = xfs_configure_buftarg(mp->m_logdev_targp, 563 log_sector_size, mp->m_sb.sb_logblocks); 564 if (error) 565 return error; 566 } 567 568 if (mp->m_sb.sb_rtstart) { 569 if (mp->m_rtdev_targp) { 570 xfs_warn(mp, 571 "can't use internal and external rtdev at the same time"); 572 return -EINVAL; 573 } 574 mp->m_rtdev_targp = mp->m_ddev_targp; 575 } else if (mp->m_rtname) { 576 error = xfs_configure_buftarg(mp->m_rtdev_targp, 577 mp->m_sb.sb_sectsize, mp->m_sb.sb_rblocks); 578 if (error) 579 return error; 580 } 581 582 return 0; 583 } 584 585 STATIC int 586 xfs_init_mount_workqueues( 587 struct xfs_mount *mp) 588 { 589 mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s", 590 XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM | WQ_PERCPU), 591 1, mp->m_super->s_id); 592 if (!mp->m_buf_workqueue) 593 goto out; 594 595 mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s", 596 XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM | WQ_PERCPU), 597 0, mp->m_super->s_id); 598 if (!mp->m_unwritten_workqueue) 599 goto out_destroy_buf; 600 601 mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s", 602 XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM | WQ_PERCPU), 603 0, mp->m_super->s_id); 604 if (!mp->m_reclaim_workqueue) 605 goto out_destroy_unwritten; 606 607 mp->m_blockgc_wq = alloc_workqueue("xfs-blockgc/%s", 608 XFS_WQFLAGS(WQ_UNBOUND | WQ_FREEZABLE | WQ_MEM_RECLAIM), 609 0, mp->m_super->s_id); 610 if (!mp->m_blockgc_wq) 611 goto out_destroy_reclaim; 612 613 mp->m_inodegc_wq = alloc_workqueue("xfs-inodegc/%s", 614 XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM | WQ_PERCPU), 615 1, mp->m_super->s_id); 616 if (!mp->m_inodegc_wq) 617 goto out_destroy_blockgc; 618 619 mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s", 620 XFS_WQFLAGS(WQ_FREEZABLE | WQ_PERCPU), 0, 621 mp->m_super->s_id); 622 if (!mp->m_sync_workqueue) 623 goto out_destroy_inodegc; 624 625 return 0; 626 627 out_destroy_inodegc: 628 destroy_workqueue(mp->m_inodegc_wq); 629 out_destroy_blockgc: 630 destroy_workqueue(mp->m_blockgc_wq); 631 out_destroy_reclaim: 632 destroy_workqueue(mp->m_reclaim_workqueue); 633 out_destroy_unwritten: 634 destroy_workqueue(mp->m_unwritten_workqueue); 635 out_destroy_buf: 636 destroy_workqueue(mp->m_buf_workqueue); 637 out: 638 return -ENOMEM; 639 } 640 641 STATIC void 642 xfs_destroy_mount_workqueues( 643 struct xfs_mount *mp) 644 { 645 destroy_workqueue(mp->m_sync_workqueue); 646 destroy_workqueue(mp->m_blockgc_wq); 647 destroy_workqueue(mp->m_inodegc_wq); 648 destroy_workqueue(mp->m_reclaim_workqueue); 649 destroy_workqueue(mp->m_unwritten_workqueue); 650 destroy_workqueue(mp->m_buf_workqueue); 651 } 652 653 static void 654 xfs_flush_inodes_worker( 655 struct work_struct *work) 656 { 657 struct xfs_mount *mp = container_of(work, struct xfs_mount, 658 m_flush_inodes_work); 659 struct super_block *sb = mp->m_super; 660 661 if (down_read_trylock(&sb->s_umount)) { 662 sync_inodes_sb(sb); 663 up_read(&sb->s_umount); 664 } 665 } 666 667 /* 668 * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK 669 * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting 670 * for IO to complete so that we effectively throttle multiple callers to the 671 * rate at which IO is completing. 672 */ 673 void 674 xfs_flush_inodes( 675 struct xfs_mount *mp) 676 { 677 /* 678 * If flush_work() returns true then that means we waited for a flush 679 * which was already in progress. Don't bother running another scan. 680 */ 681 if (flush_work(&mp->m_flush_inodes_work)) 682 return; 683 684 queue_work(mp->m_sync_workqueue, &mp->m_flush_inodes_work); 685 flush_work(&mp->m_flush_inodes_work); 686 } 687 688 /* Catch misguided souls that try to use this interface on XFS */ 689 STATIC struct inode * 690 xfs_fs_alloc_inode( 691 struct super_block *sb) 692 { 693 BUG(); 694 return NULL; 695 } 696 697 /* 698 * Now that the generic code is guaranteed not to be accessing 699 * the linux inode, we can inactivate and reclaim the inode. 700 */ 701 STATIC void 702 xfs_fs_destroy_inode( 703 struct inode *inode) 704 { 705 struct xfs_inode *ip = XFS_I(inode); 706 707 trace_xfs_destroy_inode(ip); 708 709 ASSERT(!rwsem_is_locked(&inode->i_rwsem)); 710 XFS_STATS_INC(ip->i_mount, vn_rele); 711 XFS_STATS_INC(ip->i_mount, vn_remove); 712 xfs_inode_mark_reclaimable(ip); 713 } 714 715 static void 716 xfs_fs_dirty_inode( 717 struct inode *inode, 718 int flags) 719 { 720 struct xfs_inode *ip = XFS_I(inode); 721 struct xfs_mount *mp = ip->i_mount; 722 struct xfs_trans *tp; 723 724 if (!(inode->i_sb->s_flags & SB_LAZYTIME)) 725 return; 726 727 /* 728 * Only do the timestamp update if the inode is dirty (I_DIRTY_SYNC) 729 * and has dirty timestamp (I_DIRTY_TIME). I_DIRTY_TIME can be passed 730 * in flags possibly together with I_DIRTY_SYNC. 731 */ 732 if ((flags & ~I_DIRTY_TIME) != I_DIRTY_SYNC || !(flags & I_DIRTY_TIME)) 733 return; 734 735 if (xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp)) 736 return; 737 xfs_ilock(ip, XFS_ILOCK_EXCL); 738 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); 739 xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP); 740 xfs_trans_commit(tp); 741 } 742 743 /* 744 * Slab object creation initialisation for the XFS inode. 745 * This covers only the idempotent fields in the XFS inode; 746 * all other fields need to be initialised on allocation 747 * from the slab. This avoids the need to repeatedly initialise 748 * fields in the xfs inode that left in the initialise state 749 * when freeing the inode. 750 */ 751 STATIC void 752 xfs_fs_inode_init_once( 753 void *inode) 754 { 755 struct xfs_inode *ip = inode; 756 757 memset(ip, 0, sizeof(struct xfs_inode)); 758 759 /* vfs inode */ 760 inode_init_once(VFS_I(ip)); 761 762 /* xfs inode */ 763 atomic_set(&ip->i_pincount, 0); 764 spin_lock_init(&ip->i_flags_lock); 765 init_rwsem(&ip->i_lock); 766 } 767 768 /* 769 * We do an unlocked check for XFS_IDONTCACHE here because we are already 770 * serialised against cache hits here via the inode->i_lock and igrab() in 771 * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be 772 * racing with us, and it avoids needing to grab a spinlock here for every inode 773 * we drop the final reference on. 774 */ 775 STATIC int 776 xfs_fs_drop_inode( 777 struct inode *inode) 778 { 779 struct xfs_inode *ip = XFS_I(inode); 780 781 /* 782 * If this unlinked inode is in the middle of recovery, don't 783 * drop the inode just yet; log recovery will take care of 784 * that. See the comment for this inode flag. 785 */ 786 if (ip->i_flags & XFS_IRECOVERY) { 787 ASSERT(xlog_recovery_needed(ip->i_mount->m_log)); 788 return 0; 789 } 790 791 return inode_generic_drop(inode); 792 } 793 794 STATIC void 795 xfs_fs_evict_inode( 796 struct inode *inode) 797 { 798 if (IS_DAX(inode)) 799 dax_break_layout_final(inode); 800 801 truncate_inode_pages_final(&inode->i_data); 802 clear_inode(inode); 803 804 if (IS_ENABLED(CONFIG_XFS_RT) && 805 S_ISREG(inode->i_mode) && inode->i_private) { 806 xfs_open_zone_put(inode->i_private); 807 inode->i_private = NULL; 808 } 809 } 810 811 static void 812 xfs_mount_free( 813 struct xfs_mount *mp) 814 { 815 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) 816 xfs_free_buftarg(mp->m_logdev_targp); 817 if (mp->m_rtdev_targp && mp->m_rtdev_targp != mp->m_ddev_targp) 818 xfs_free_buftarg(mp->m_rtdev_targp); 819 if (mp->m_ddev_targp) 820 xfs_free_buftarg(mp->m_ddev_targp); 821 822 debugfs_remove(mp->m_debugfs); 823 kfree(mp->m_rtname); 824 kfree(mp->m_logname); 825 kfree(mp); 826 } 827 828 STATIC int 829 xfs_fs_sync_fs( 830 struct super_block *sb, 831 int wait) 832 { 833 struct xfs_mount *mp = XFS_M(sb); 834 int error; 835 836 trace_xfs_fs_sync_fs(mp, __return_address); 837 838 /* 839 * Doing anything during the async pass would be counterproductive. 840 */ 841 if (!wait) 842 return 0; 843 844 error = xfs_log_force(mp, XFS_LOG_SYNC); 845 if (error) 846 return error; 847 848 if (laptop_mode) { 849 /* 850 * The disk must be active because we're syncing. 851 * We schedule log work now (now that the disk is 852 * active) instead of later (when it might not be). 853 */ 854 flush_delayed_work(&mp->m_log->l_work); 855 } 856 857 /* 858 * If we are called with page faults frozen out, it means we are about 859 * to freeze the transaction subsystem. Take the opportunity to shut 860 * down inodegc because once SB_FREEZE_FS is set it's too late to 861 * prevent inactivation races with freeze. The fs doesn't get called 862 * again by the freezing process until after SB_FREEZE_FS has been set, 863 * so it's now or never. Same logic applies to speculative allocation 864 * garbage collection. 865 * 866 * We don't care if this is a normal syncfs call that does this or 867 * freeze that does this - we can run this multiple times without issue 868 * and we won't race with a restart because a restart can only occur 869 * when the state is either SB_FREEZE_FS or SB_FREEZE_COMPLETE. 870 */ 871 if (sb->s_writers.frozen == SB_FREEZE_PAGEFAULT) { 872 xfs_inodegc_stop(mp); 873 xfs_blockgc_stop(mp); 874 xfs_zone_gc_stop(mp); 875 } 876 877 return 0; 878 } 879 880 static xfs_extlen_t 881 xfs_internal_log_size( 882 struct xfs_mount *mp) 883 { 884 if (!mp->m_sb.sb_logstart) 885 return 0; 886 return mp->m_sb.sb_logblocks; 887 } 888 889 static void 890 xfs_statfs_data( 891 struct xfs_mount *mp, 892 struct kstatfs *st) 893 { 894 int64_t fdblocks = 895 xfs_sum_freecounter(mp, XC_FREE_BLOCKS); 896 897 /* make sure st->f_bfree does not underflow */ 898 st->f_bfree = max(0LL, 899 fdblocks - xfs_freecounter_unavailable(mp, XC_FREE_BLOCKS)); 900 901 /* 902 * sb_dblocks can change during growfs, but nothing cares about reporting 903 * the old or new value during growfs. 904 */ 905 st->f_blocks = mp->m_sb.sb_dblocks - xfs_internal_log_size(mp); 906 } 907 908 /* 909 * When stat(v)fs is called on a file with the realtime bit set or a directory 910 * with the rtinherit bit, report freespace information for the RT device 911 * instead of the main data device. 912 */ 913 static void 914 xfs_statfs_rt( 915 struct xfs_mount *mp, 916 struct kstatfs *st) 917 { 918 st->f_bfree = xfs_rtbxlen_to_blen(mp, 919 xfs_sum_freecounter(mp, XC_FREE_RTEXTENTS)); 920 st->f_blocks = mp->m_sb.sb_rblocks - xfs_rtbxlen_to_blen(mp, 921 mp->m_free[XC_FREE_RTEXTENTS].res_total); 922 } 923 924 static void 925 xfs_statfs_inodes( 926 struct xfs_mount *mp, 927 struct kstatfs *st) 928 { 929 uint64_t icount = percpu_counter_sum(&mp->m_icount); 930 uint64_t ifree = percpu_counter_sum(&mp->m_ifree); 931 uint64_t fakeinos = XFS_FSB_TO_INO(mp, st->f_bfree); 932 933 st->f_files = min(icount + fakeinos, (uint64_t)XFS_MAXINUMBER); 934 if (M_IGEO(mp)->maxicount) 935 st->f_files = min_t(typeof(st->f_files), st->f_files, 936 M_IGEO(mp)->maxicount); 937 938 /* If sb_icount overshot maxicount, report actual allocation */ 939 st->f_files = max_t(typeof(st->f_files), st->f_files, 940 mp->m_sb.sb_icount); 941 942 /* Make sure st->f_ffree does not underflow */ 943 st->f_ffree = max_t(int64_t, 0, st->f_files - (icount - ifree)); 944 } 945 946 STATIC int 947 xfs_fs_statfs( 948 struct dentry *dentry, 949 struct kstatfs *st) 950 { 951 struct xfs_mount *mp = XFS_M(dentry->d_sb); 952 struct xfs_inode *ip = XFS_I(d_inode(dentry)); 953 954 /* 955 * Expedite background inodegc but don't wait. We do not want to block 956 * here waiting hours for a billion extent file to be truncated. 957 */ 958 xfs_inodegc_push(mp); 959 960 st->f_type = XFS_SUPER_MAGIC; 961 st->f_namelen = MAXNAMELEN - 1; 962 st->f_bsize = mp->m_sb.sb_blocksize; 963 st->f_fsid = u64_to_fsid(huge_encode_dev(mp->m_ddev_targp->bt_dev)); 964 965 xfs_statfs_data(mp, st); 966 xfs_statfs_inodes(mp, st); 967 968 if (XFS_IS_REALTIME_MOUNT(mp) && 969 (ip->i_diflags & (XFS_DIFLAG_RTINHERIT | XFS_DIFLAG_REALTIME))) 970 xfs_statfs_rt(mp, st); 971 972 if ((ip->i_diflags & XFS_DIFLAG_PROJINHERIT) && 973 ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) == 974 (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD)) 975 xfs_qm_statvfs(ip, st); 976 977 /* 978 * XFS does not distinguish between blocks available to privileged and 979 * unprivileged users. 980 */ 981 st->f_bavail = st->f_bfree; 982 return 0; 983 } 984 985 STATIC void 986 xfs_save_resvblks( 987 struct xfs_mount *mp) 988 { 989 enum xfs_free_counter i; 990 991 for (i = 0; i < XC_FREE_NR; i++) { 992 mp->m_free[i].res_saved = mp->m_free[i].res_total; 993 xfs_reserve_blocks(mp, i, 0); 994 } 995 } 996 997 STATIC void 998 xfs_restore_resvblks( 999 struct xfs_mount *mp) 1000 { 1001 uint64_t resblks; 1002 enum xfs_free_counter i; 1003 1004 for (i = 0; i < XC_FREE_NR; i++) { 1005 if (mp->m_free[i].res_saved) { 1006 resblks = mp->m_free[i].res_saved; 1007 mp->m_free[i].res_saved = 0; 1008 } else 1009 resblks = xfs_default_resblks(mp, i); 1010 xfs_reserve_blocks(mp, i, resblks); 1011 } 1012 } 1013 1014 /* 1015 * Second stage of a freeze. The data is already frozen so we only 1016 * need to take care of the metadata. Once that's done sync the superblock 1017 * to the log to dirty it in case of a crash while frozen. This ensures that we 1018 * will recover the unlinked inode lists on the next mount. 1019 */ 1020 STATIC int 1021 xfs_fs_freeze( 1022 struct super_block *sb) 1023 { 1024 struct xfs_mount *mp = XFS_M(sb); 1025 unsigned int flags; 1026 int ret; 1027 1028 /* 1029 * The filesystem is now frozen far enough that memory reclaim 1030 * cannot safely operate on the filesystem. Hence we need to 1031 * set a GFP_NOFS context here to avoid recursion deadlocks. 1032 */ 1033 flags = memalloc_nofs_save(); 1034 xfs_save_resvblks(mp); 1035 ret = xfs_log_quiesce(mp); 1036 memalloc_nofs_restore(flags); 1037 1038 /* 1039 * For read-write filesystems, we need to restart the inodegc on error 1040 * because we stopped it at SB_FREEZE_PAGEFAULT level and a thaw is not 1041 * going to be run to restart it now. We are at SB_FREEZE_FS level 1042 * here, so we can restart safely without racing with a stop in 1043 * xfs_fs_sync_fs(). 1044 */ 1045 if (ret && !xfs_is_readonly(mp)) { 1046 xfs_blockgc_start(mp); 1047 xfs_inodegc_start(mp); 1048 xfs_zone_gc_start(mp); 1049 } 1050 1051 return ret; 1052 } 1053 1054 STATIC int 1055 xfs_fs_unfreeze( 1056 struct super_block *sb) 1057 { 1058 struct xfs_mount *mp = XFS_M(sb); 1059 1060 xfs_restore_resvblks(mp); 1061 xfs_log_work_queue(mp); 1062 1063 /* 1064 * Don't reactivate the inodegc worker on a readonly filesystem because 1065 * inodes are sent directly to reclaim. Don't reactivate the blockgc 1066 * worker because there are no speculative preallocations on a readonly 1067 * filesystem. 1068 */ 1069 if (!xfs_is_readonly(mp)) { 1070 xfs_zone_gc_start(mp); 1071 xfs_blockgc_start(mp); 1072 xfs_inodegc_start(mp); 1073 } 1074 1075 return 0; 1076 } 1077 1078 /* 1079 * This function fills in xfs_mount_t fields based on mount args. 1080 * Note: the superblock _has_ now been read in. 1081 */ 1082 STATIC int 1083 xfs_finish_flags( 1084 struct xfs_mount *mp) 1085 { 1086 /* Fail a mount where the logbuf is smaller than the log stripe */ 1087 if (xfs_has_logv2(mp)) { 1088 if (mp->m_logbsize <= 0 && 1089 mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) { 1090 mp->m_logbsize = mp->m_sb.sb_logsunit; 1091 } else if (mp->m_logbsize > 0 && 1092 mp->m_logbsize < mp->m_sb.sb_logsunit) { 1093 xfs_warn(mp, 1094 "logbuf size must be greater than or equal to log stripe size"); 1095 return -EINVAL; 1096 } 1097 } else { 1098 /* Fail a mount if the logbuf is larger than 32K */ 1099 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) { 1100 xfs_warn(mp, 1101 "logbuf size for version 1 logs must be 16K or 32K"); 1102 return -EINVAL; 1103 } 1104 } 1105 1106 /* 1107 * prohibit r/w mounts of read-only filesystems 1108 */ 1109 if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !xfs_is_readonly(mp)) { 1110 xfs_warn(mp, 1111 "cannot mount a read-only filesystem as read-write"); 1112 return -EROFS; 1113 } 1114 1115 if ((mp->m_qflags & XFS_GQUOTA_ACCT) && 1116 (mp->m_qflags & XFS_PQUOTA_ACCT) && 1117 !xfs_has_pquotino(mp)) { 1118 xfs_warn(mp, 1119 "Super block does not support project and group quota together"); 1120 return -EINVAL; 1121 } 1122 1123 if (!xfs_has_zoned(mp)) { 1124 if (mp->m_max_open_zones) { 1125 xfs_warn(mp, 1126 "max_open_zones mount option only supported on zoned file systems."); 1127 return -EINVAL; 1128 } 1129 if (mp->m_features & XFS_FEAT_NOLIFETIME) { 1130 xfs_warn(mp, 1131 "nolifetime mount option only supported on zoned file systems."); 1132 return -EINVAL; 1133 } 1134 } 1135 1136 return 0; 1137 } 1138 1139 static int 1140 xfs_init_percpu_counters( 1141 struct xfs_mount *mp) 1142 { 1143 int error; 1144 int i; 1145 1146 error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL); 1147 if (error) 1148 return -ENOMEM; 1149 1150 error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL); 1151 if (error) 1152 goto free_icount; 1153 1154 error = percpu_counter_init(&mp->m_delalloc_blks, 0, GFP_KERNEL); 1155 if (error) 1156 goto free_ifree; 1157 1158 error = percpu_counter_init(&mp->m_delalloc_rtextents, 0, GFP_KERNEL); 1159 if (error) 1160 goto free_delalloc; 1161 1162 for (i = 0; i < XC_FREE_NR; i++) { 1163 error = percpu_counter_init(&mp->m_free[i].count, 0, 1164 GFP_KERNEL); 1165 if (error) 1166 goto free_freecounters; 1167 } 1168 1169 return 0; 1170 1171 free_freecounters: 1172 while (--i >= 0) 1173 percpu_counter_destroy(&mp->m_free[i].count); 1174 percpu_counter_destroy(&mp->m_delalloc_rtextents); 1175 free_delalloc: 1176 percpu_counter_destroy(&mp->m_delalloc_blks); 1177 free_ifree: 1178 percpu_counter_destroy(&mp->m_ifree); 1179 free_icount: 1180 percpu_counter_destroy(&mp->m_icount); 1181 return -ENOMEM; 1182 } 1183 1184 void 1185 xfs_reinit_percpu_counters( 1186 struct xfs_mount *mp) 1187 { 1188 percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount); 1189 percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree); 1190 xfs_set_freecounter(mp, XC_FREE_BLOCKS, mp->m_sb.sb_fdblocks); 1191 if (!xfs_has_zoned(mp)) 1192 xfs_set_freecounter(mp, XC_FREE_RTEXTENTS, 1193 mp->m_sb.sb_frextents); 1194 } 1195 1196 static void 1197 xfs_destroy_percpu_counters( 1198 struct xfs_mount *mp) 1199 { 1200 enum xfs_free_counter i; 1201 1202 for (i = 0; i < XC_FREE_NR; i++) 1203 percpu_counter_destroy(&mp->m_free[i].count); 1204 percpu_counter_destroy(&mp->m_icount); 1205 percpu_counter_destroy(&mp->m_ifree); 1206 ASSERT(xfs_is_shutdown(mp) || 1207 percpu_counter_sum(&mp->m_delalloc_rtextents) == 0); 1208 percpu_counter_destroy(&mp->m_delalloc_rtextents); 1209 ASSERT(xfs_is_shutdown(mp) || 1210 percpu_counter_sum(&mp->m_delalloc_blks) == 0); 1211 percpu_counter_destroy(&mp->m_delalloc_blks); 1212 } 1213 1214 static int 1215 xfs_inodegc_init_percpu( 1216 struct xfs_mount *mp) 1217 { 1218 struct xfs_inodegc *gc; 1219 int cpu; 1220 1221 mp->m_inodegc = alloc_percpu(struct xfs_inodegc); 1222 if (!mp->m_inodegc) 1223 return -ENOMEM; 1224 1225 for_each_possible_cpu(cpu) { 1226 gc = per_cpu_ptr(mp->m_inodegc, cpu); 1227 gc->cpu = cpu; 1228 gc->mp = mp; 1229 init_llist_head(&gc->list); 1230 gc->items = 0; 1231 gc->error = 0; 1232 INIT_DELAYED_WORK(&gc->work, xfs_inodegc_worker); 1233 } 1234 return 0; 1235 } 1236 1237 static void 1238 xfs_inodegc_free_percpu( 1239 struct xfs_mount *mp) 1240 { 1241 if (!mp->m_inodegc) 1242 return; 1243 free_percpu(mp->m_inodegc); 1244 } 1245 1246 static void 1247 xfs_fs_put_super( 1248 struct super_block *sb) 1249 { 1250 struct xfs_mount *mp = XFS_M(sb); 1251 1252 xfs_notice(mp, "Unmounting Filesystem %pU", &mp->m_sb.sb_uuid); 1253 xfs_filestream_unmount(mp); 1254 xfs_unmountfs(mp); 1255 1256 xfs_rtmount_freesb(mp); 1257 xfs_freesb(mp); 1258 xchk_mount_stats_free(mp); 1259 free_percpu(mp->m_stats.xs_stats); 1260 xfs_inodegc_free_percpu(mp); 1261 xfs_destroy_percpu_counters(mp); 1262 xfs_destroy_mount_workqueues(mp); 1263 xfs_shutdown_devices(mp); 1264 } 1265 1266 static long 1267 xfs_fs_nr_cached_objects( 1268 struct super_block *sb, 1269 struct shrink_control *sc) 1270 { 1271 /* Paranoia: catch incorrect calls during mount setup or teardown */ 1272 if (WARN_ON_ONCE(!sb->s_fs_info)) 1273 return 0; 1274 return xfs_reclaim_inodes_count(XFS_M(sb)); 1275 } 1276 1277 static long 1278 xfs_fs_free_cached_objects( 1279 struct super_block *sb, 1280 struct shrink_control *sc) 1281 { 1282 return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan); 1283 } 1284 1285 static void 1286 xfs_fs_shutdown( 1287 struct super_block *sb) 1288 { 1289 xfs_force_shutdown(XFS_M(sb), SHUTDOWN_DEVICE_REMOVED); 1290 } 1291 1292 static int 1293 xfs_fs_show_stats( 1294 struct seq_file *m, 1295 struct dentry *root) 1296 { 1297 struct xfs_mount *mp = XFS_M(root->d_sb); 1298 1299 if (xfs_has_zoned(mp) && IS_ENABLED(CONFIG_XFS_RT)) 1300 xfs_zoned_show_stats(m, mp); 1301 return 0; 1302 } 1303 1304 static const struct super_operations xfs_super_operations = { 1305 .alloc_inode = xfs_fs_alloc_inode, 1306 .destroy_inode = xfs_fs_destroy_inode, 1307 .dirty_inode = xfs_fs_dirty_inode, 1308 .drop_inode = xfs_fs_drop_inode, 1309 .evict_inode = xfs_fs_evict_inode, 1310 .put_super = xfs_fs_put_super, 1311 .sync_fs = xfs_fs_sync_fs, 1312 .freeze_fs = xfs_fs_freeze, 1313 .unfreeze_fs = xfs_fs_unfreeze, 1314 .statfs = xfs_fs_statfs, 1315 .show_options = xfs_fs_show_options, 1316 .nr_cached_objects = xfs_fs_nr_cached_objects, 1317 .free_cached_objects = xfs_fs_free_cached_objects, 1318 .shutdown = xfs_fs_shutdown, 1319 .show_stats = xfs_fs_show_stats, 1320 }; 1321 1322 static int 1323 suffix_kstrtoint( 1324 const char *s, 1325 unsigned int base, 1326 int *res) 1327 { 1328 int last, shift_left_factor = 0, _res; 1329 char *value; 1330 int ret = 0; 1331 1332 value = kstrdup(s, GFP_KERNEL); 1333 if (!value) 1334 return -ENOMEM; 1335 1336 last = strlen(value) - 1; 1337 if (value[last] == 'K' || value[last] == 'k') { 1338 shift_left_factor = 10; 1339 value[last] = '\0'; 1340 } 1341 if (value[last] == 'M' || value[last] == 'm') { 1342 shift_left_factor = 20; 1343 value[last] = '\0'; 1344 } 1345 if (value[last] == 'G' || value[last] == 'g') { 1346 shift_left_factor = 30; 1347 value[last] = '\0'; 1348 } 1349 1350 if (kstrtoint(value, base, &_res)) 1351 ret = -EINVAL; 1352 kfree(value); 1353 *res = _res << shift_left_factor; 1354 return ret; 1355 } 1356 1357 static int 1358 suffix_kstrtoull( 1359 const char *s, 1360 unsigned int base, 1361 unsigned long long *res) 1362 { 1363 int last, shift_left_factor = 0; 1364 unsigned long long _res; 1365 char *value; 1366 int ret = 0; 1367 1368 value = kstrdup(s, GFP_KERNEL); 1369 if (!value) 1370 return -ENOMEM; 1371 1372 last = strlen(value) - 1; 1373 if (value[last] == 'K' || value[last] == 'k') { 1374 shift_left_factor = 10; 1375 value[last] = '\0'; 1376 } 1377 if (value[last] == 'M' || value[last] == 'm') { 1378 shift_left_factor = 20; 1379 value[last] = '\0'; 1380 } 1381 if (value[last] == 'G' || value[last] == 'g') { 1382 shift_left_factor = 30; 1383 value[last] = '\0'; 1384 } 1385 1386 if (kstrtoull(value, base, &_res)) 1387 ret = -EINVAL; 1388 kfree(value); 1389 *res = _res << shift_left_factor; 1390 return ret; 1391 } 1392 1393 static inline void 1394 xfs_fs_warn_deprecated( 1395 struct fs_context *fc, 1396 struct fs_parameter *param) 1397 { 1398 /* 1399 * Always warn about someone passing in a deprecated mount option. 1400 * Previously we wouldn't print the warning if we were reconfiguring 1401 * and current mount point already had the flag set, but that was not 1402 * the right thing to do. 1403 * 1404 * Many distributions mount the root filesystem with no options in the 1405 * initramfs and rely on mount -a to remount the root fs with the 1406 * options in fstab. However, the old behavior meant that there would 1407 * never be a warning about deprecated mount options for the root fs in 1408 * /etc/fstab. On a single-fs system, that means no warning at all. 1409 * 1410 * Compounding this problem are distribution scripts that copy 1411 * /proc/mounts to fstab, which means that we can't remove mount 1412 * options unless we're 100% sure they have only ever been advertised 1413 * in /proc/mounts in response to explicitly provided mount options. 1414 */ 1415 xfs_warn(fc->s_fs_info, "%s mount option is deprecated.", param->key); 1416 } 1417 1418 /* 1419 * Set mount state from a mount option. 1420 * 1421 * NOTE: mp->m_super is NULL here! 1422 */ 1423 static int 1424 xfs_fs_parse_param( 1425 struct fs_context *fc, 1426 struct fs_parameter *param) 1427 { 1428 struct xfs_mount *parsing_mp = fc->s_fs_info; 1429 struct fs_parse_result result; 1430 int size = 0; 1431 int opt; 1432 1433 BUILD_BUG_ON(XFS_QFLAGS_MNTOPTS & XFS_MOUNT_QUOTA_ALL); 1434 1435 opt = fs_parse(fc, xfs_fs_parameters, param, &result); 1436 if (opt < 0) 1437 return opt; 1438 1439 switch (opt) { 1440 case Op_deprecated: 1441 xfs_fs_warn_deprecated(fc, param); 1442 return 0; 1443 case Opt_logbufs: 1444 parsing_mp->m_logbufs = result.uint_32; 1445 return 0; 1446 case Opt_logbsize: 1447 if (suffix_kstrtoint(param->string, 10, &parsing_mp->m_logbsize)) 1448 return -EINVAL; 1449 return 0; 1450 case Opt_logdev: 1451 kfree(parsing_mp->m_logname); 1452 parsing_mp->m_logname = kstrdup(param->string, GFP_KERNEL); 1453 if (!parsing_mp->m_logname) 1454 return -ENOMEM; 1455 return 0; 1456 case Opt_rtdev: 1457 kfree(parsing_mp->m_rtname); 1458 parsing_mp->m_rtname = kstrdup(param->string, GFP_KERNEL); 1459 if (!parsing_mp->m_rtname) 1460 return -ENOMEM; 1461 return 0; 1462 case Opt_allocsize: 1463 if (suffix_kstrtoint(param->string, 10, &size)) 1464 return -EINVAL; 1465 parsing_mp->m_allocsize_log = ffs(size) - 1; 1466 parsing_mp->m_features |= XFS_FEAT_ALLOCSIZE; 1467 return 0; 1468 case Opt_grpid: 1469 case Opt_bsdgroups: 1470 parsing_mp->m_features |= XFS_FEAT_GRPID; 1471 return 0; 1472 case Opt_nogrpid: 1473 case Opt_sysvgroups: 1474 parsing_mp->m_features &= ~XFS_FEAT_GRPID; 1475 return 0; 1476 case Opt_wsync: 1477 parsing_mp->m_features |= XFS_FEAT_WSYNC; 1478 return 0; 1479 case Opt_norecovery: 1480 parsing_mp->m_features |= XFS_FEAT_NORECOVERY; 1481 return 0; 1482 case Opt_noalign: 1483 parsing_mp->m_features |= XFS_FEAT_NOALIGN; 1484 return 0; 1485 case Opt_swalloc: 1486 parsing_mp->m_features |= XFS_FEAT_SWALLOC; 1487 return 0; 1488 case Opt_sunit: 1489 parsing_mp->m_dalign = result.uint_32; 1490 return 0; 1491 case Opt_swidth: 1492 parsing_mp->m_swidth = result.uint_32; 1493 return 0; 1494 case Opt_inode32: 1495 parsing_mp->m_features |= XFS_FEAT_SMALL_INUMS; 1496 return 0; 1497 case Opt_inode64: 1498 parsing_mp->m_features &= ~XFS_FEAT_SMALL_INUMS; 1499 return 0; 1500 case Opt_nouuid: 1501 parsing_mp->m_features |= XFS_FEAT_NOUUID; 1502 return 0; 1503 case Opt_largeio: 1504 parsing_mp->m_features |= XFS_FEAT_LARGE_IOSIZE; 1505 return 0; 1506 case Opt_nolargeio: 1507 parsing_mp->m_features &= ~XFS_FEAT_LARGE_IOSIZE; 1508 return 0; 1509 case Opt_filestreams: 1510 parsing_mp->m_features |= XFS_FEAT_FILESTREAMS; 1511 return 0; 1512 case Opt_noquota: 1513 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT; 1514 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD; 1515 parsing_mp->m_qflags |= XFS_QFLAGS_MNTOPTS; 1516 return 0; 1517 case Opt_quota: 1518 case Opt_uquota: 1519 case Opt_usrquota: 1520 parsing_mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ENFD); 1521 parsing_mp->m_qflags |= XFS_QFLAGS_MNTOPTS; 1522 return 0; 1523 case Opt_qnoenforce: 1524 case Opt_uqnoenforce: 1525 parsing_mp->m_qflags |= XFS_UQUOTA_ACCT; 1526 parsing_mp->m_qflags &= ~XFS_UQUOTA_ENFD; 1527 parsing_mp->m_qflags |= XFS_QFLAGS_MNTOPTS; 1528 return 0; 1529 case Opt_pquota: 1530 case Opt_prjquota: 1531 parsing_mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ENFD); 1532 parsing_mp->m_qflags |= XFS_QFLAGS_MNTOPTS; 1533 return 0; 1534 case Opt_pqnoenforce: 1535 parsing_mp->m_qflags |= XFS_PQUOTA_ACCT; 1536 parsing_mp->m_qflags &= ~XFS_PQUOTA_ENFD; 1537 parsing_mp->m_qflags |= XFS_QFLAGS_MNTOPTS; 1538 return 0; 1539 case Opt_gquota: 1540 case Opt_grpquota: 1541 parsing_mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ENFD); 1542 parsing_mp->m_qflags |= XFS_QFLAGS_MNTOPTS; 1543 return 0; 1544 case Opt_gqnoenforce: 1545 parsing_mp->m_qflags |= XFS_GQUOTA_ACCT; 1546 parsing_mp->m_qflags &= ~XFS_GQUOTA_ENFD; 1547 parsing_mp->m_qflags |= XFS_QFLAGS_MNTOPTS; 1548 return 0; 1549 case Opt_discard: 1550 parsing_mp->m_features |= XFS_FEAT_DISCARD; 1551 return 0; 1552 case Opt_nodiscard: 1553 parsing_mp->m_features &= ~XFS_FEAT_DISCARD; 1554 return 0; 1555 #ifdef CONFIG_FS_DAX 1556 case Opt_dax: 1557 xfs_mount_set_dax_mode(parsing_mp, XFS_DAX_ALWAYS); 1558 return 0; 1559 case Opt_dax_enum: 1560 xfs_mount_set_dax_mode(parsing_mp, result.uint_32); 1561 return 0; 1562 #endif 1563 case Opt_max_open_zones: 1564 parsing_mp->m_max_open_zones = result.uint_32; 1565 return 0; 1566 case Opt_lifetime: 1567 parsing_mp->m_features &= ~XFS_FEAT_NOLIFETIME; 1568 return 0; 1569 case Opt_nolifetime: 1570 parsing_mp->m_features |= XFS_FEAT_NOLIFETIME; 1571 return 0; 1572 case Opt_max_atomic_write: 1573 if (suffix_kstrtoull(param->string, 10, 1574 &parsing_mp->m_awu_max_bytes)) { 1575 xfs_warn(parsing_mp, 1576 "max atomic write size must be positive integer"); 1577 return -EINVAL; 1578 } 1579 return 0; 1580 default: 1581 xfs_warn(parsing_mp, "unknown mount option [%s].", param->key); 1582 return -EINVAL; 1583 } 1584 1585 return 0; 1586 } 1587 1588 static int 1589 xfs_fs_validate_params( 1590 struct xfs_mount *mp) 1591 { 1592 /* No recovery flag requires a read-only mount */ 1593 if (xfs_has_norecovery(mp) && !xfs_is_readonly(mp)) { 1594 xfs_warn(mp, "no-recovery mounts must be read-only."); 1595 return -EINVAL; 1596 } 1597 1598 if (xfs_has_noalign(mp) && (mp->m_dalign || mp->m_swidth)) { 1599 xfs_warn(mp, 1600 "sunit and swidth options incompatible with the noalign option"); 1601 return -EINVAL; 1602 } 1603 1604 if (!IS_ENABLED(CONFIG_XFS_QUOTA) && 1605 (mp->m_qflags & ~XFS_QFLAGS_MNTOPTS)) { 1606 xfs_warn(mp, "quota support not available in this kernel."); 1607 return -EINVAL; 1608 } 1609 1610 if ((mp->m_dalign && !mp->m_swidth) || 1611 (!mp->m_dalign && mp->m_swidth)) { 1612 xfs_warn(mp, "sunit and swidth must be specified together"); 1613 return -EINVAL; 1614 } 1615 1616 if (mp->m_dalign && (mp->m_swidth % mp->m_dalign != 0)) { 1617 xfs_warn(mp, 1618 "stripe width (%d) must be a multiple of the stripe unit (%d)", 1619 mp->m_swidth, mp->m_dalign); 1620 return -EINVAL; 1621 } 1622 1623 if (mp->m_logbufs != -1 && 1624 mp->m_logbufs != 0 && 1625 (mp->m_logbufs < XLOG_MIN_ICLOGS || 1626 mp->m_logbufs > XLOG_MAX_ICLOGS)) { 1627 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]", 1628 mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS); 1629 return -EINVAL; 1630 } 1631 1632 if (mp->m_logbsize != -1 && 1633 mp->m_logbsize != 0 && 1634 (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE || 1635 mp->m_logbsize > XLOG_MAX_RECORD_BSIZE || 1636 !is_power_of_2(mp->m_logbsize))) { 1637 xfs_warn(mp, 1638 "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]", 1639 mp->m_logbsize); 1640 return -EINVAL; 1641 } 1642 1643 if (xfs_has_allocsize(mp) && 1644 (mp->m_allocsize_log > XFS_MAX_IO_LOG || 1645 mp->m_allocsize_log < XFS_MIN_IO_LOG)) { 1646 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]", 1647 mp->m_allocsize_log, XFS_MIN_IO_LOG, XFS_MAX_IO_LOG); 1648 return -EINVAL; 1649 } 1650 1651 return 0; 1652 } 1653 1654 struct dentry * 1655 xfs_debugfs_mkdir( 1656 const char *name, 1657 struct dentry *parent) 1658 { 1659 struct dentry *child; 1660 1661 /* Apparently we're expected to ignore error returns?? */ 1662 child = debugfs_create_dir(name, parent); 1663 if (IS_ERR(child)) 1664 return NULL; 1665 1666 return child; 1667 } 1668 1669 static int 1670 xfs_fs_fill_super( 1671 struct super_block *sb, 1672 struct fs_context *fc) 1673 { 1674 struct xfs_mount *mp = sb->s_fs_info; 1675 struct inode *root; 1676 int flags = 0, error; 1677 1678 mp->m_super = sb; 1679 1680 /* 1681 * Copy VFS mount flags from the context now that all parameter parsing 1682 * is guaranteed to have been completed by either the old mount API or 1683 * the newer fsopen/fsconfig API. 1684 */ 1685 if (fc->sb_flags & SB_RDONLY) 1686 xfs_set_readonly(mp); 1687 if (fc->sb_flags & SB_DIRSYNC) 1688 mp->m_features |= XFS_FEAT_DIRSYNC; 1689 if (fc->sb_flags & SB_SYNCHRONOUS) 1690 mp->m_features |= XFS_FEAT_WSYNC; 1691 1692 error = xfs_fs_validate_params(mp); 1693 if (error) 1694 return error; 1695 1696 sb_min_blocksize(sb, BBSIZE); 1697 sb->s_xattr = xfs_xattr_handlers; 1698 sb->s_export_op = &xfs_export_operations; 1699 #ifdef CONFIG_XFS_QUOTA 1700 sb->s_qcop = &xfs_quotactl_operations; 1701 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ; 1702 #endif 1703 sb->s_op = &xfs_super_operations; 1704 1705 /* 1706 * Delay mount work if the debug hook is set. This is debug 1707 * instrumention to coordinate simulation of xfs mount failures with 1708 * VFS superblock operations 1709 */ 1710 if (xfs_globals.mount_delay) { 1711 xfs_notice(mp, "Delaying mount for %d seconds.", 1712 xfs_globals.mount_delay); 1713 msleep(xfs_globals.mount_delay * 1000); 1714 } 1715 1716 if (fc->sb_flags & SB_SILENT) 1717 flags |= XFS_MFSI_QUIET; 1718 1719 error = xfs_open_devices(mp); 1720 if (error) 1721 return error; 1722 1723 if (xfs_debugfs) { 1724 mp->m_debugfs = xfs_debugfs_mkdir(mp->m_super->s_id, 1725 xfs_debugfs); 1726 } else { 1727 mp->m_debugfs = NULL; 1728 } 1729 1730 error = xfs_init_mount_workqueues(mp); 1731 if (error) 1732 goto out_shutdown_devices; 1733 1734 error = xfs_init_percpu_counters(mp); 1735 if (error) 1736 goto out_destroy_workqueues; 1737 1738 error = xfs_inodegc_init_percpu(mp); 1739 if (error) 1740 goto out_destroy_counters; 1741 1742 /* Allocate stats memory before we do operations that might use it */ 1743 mp->m_stats.xs_stats = alloc_percpu(struct xfsstats); 1744 if (!mp->m_stats.xs_stats) { 1745 error = -ENOMEM; 1746 goto out_destroy_inodegc; 1747 } 1748 1749 error = xchk_mount_stats_alloc(mp); 1750 if (error) 1751 goto out_free_stats; 1752 1753 error = xfs_readsb(mp, flags); 1754 if (error) 1755 goto out_free_scrub_stats; 1756 1757 error = xfs_finish_flags(mp); 1758 if (error) 1759 goto out_free_sb; 1760 1761 error = xfs_setup_devices(mp); 1762 if (error) 1763 goto out_free_sb; 1764 1765 /* 1766 * V4 support is undergoing deprecation. 1767 * 1768 * Note: this has to use an open coded m_features check as xfs_has_crc 1769 * always returns false for !CONFIG_XFS_SUPPORT_V4. 1770 */ 1771 if (!(mp->m_features & XFS_FEAT_CRC)) { 1772 if (!IS_ENABLED(CONFIG_XFS_SUPPORT_V4)) { 1773 xfs_warn(mp, 1774 "Deprecated V4 format (crc=0) not supported by kernel."); 1775 error = -EINVAL; 1776 goto out_free_sb; 1777 } 1778 xfs_warn_once(mp, 1779 "Deprecated V4 format (crc=0) will not be supported after September 2030."); 1780 } 1781 1782 /* ASCII case insensitivity is undergoing deprecation. */ 1783 if (xfs_has_asciici(mp)) { 1784 #ifdef CONFIG_XFS_SUPPORT_ASCII_CI 1785 xfs_warn_once(mp, 1786 "Deprecated ASCII case-insensitivity feature (ascii-ci=1) will not be supported after September 2030."); 1787 #else 1788 xfs_warn(mp, 1789 "Deprecated ASCII case-insensitivity feature (ascii-ci=1) not supported by kernel."); 1790 error = -EINVAL; 1791 goto out_free_sb; 1792 #endif 1793 } 1794 1795 /* 1796 * Filesystem claims it needs repair, so refuse the mount unless 1797 * norecovery is also specified, in which case the filesystem can 1798 * be mounted with no risk of further damage. 1799 */ 1800 if (xfs_has_needsrepair(mp) && !xfs_has_norecovery(mp)) { 1801 xfs_warn(mp, "Filesystem needs repair. Please run xfs_repair."); 1802 error = -EFSCORRUPTED; 1803 goto out_free_sb; 1804 } 1805 1806 /* 1807 * Don't touch the filesystem if a user tool thinks it owns the primary 1808 * superblock. mkfs doesn't clear the flag from secondary supers, so 1809 * we don't check them at all. 1810 */ 1811 if (mp->m_sb.sb_inprogress) { 1812 xfs_warn(mp, "Offline file system operation in progress!"); 1813 error = -EFSCORRUPTED; 1814 goto out_free_sb; 1815 } 1816 1817 if (mp->m_sb.sb_blocksize > PAGE_SIZE) { 1818 size_t max_folio_size = mapping_max_folio_size_supported(); 1819 1820 if (!xfs_has_crc(mp)) { 1821 xfs_warn(mp, 1822 "V4 Filesystem with blocksize %d bytes. Only pagesize (%ld) or less is supported.", 1823 mp->m_sb.sb_blocksize, PAGE_SIZE); 1824 error = -ENOSYS; 1825 goto out_free_sb; 1826 } 1827 1828 if (mp->m_sb.sb_blocksize > max_folio_size) { 1829 xfs_warn(mp, 1830 "block size (%u bytes) not supported; Only block size (%zu) or less is supported", 1831 mp->m_sb.sb_blocksize, max_folio_size); 1832 error = -ENOSYS; 1833 goto out_free_sb; 1834 } 1835 1836 xfs_warn_experimental(mp, XFS_EXPERIMENTAL_LBS); 1837 } 1838 1839 /* Ensure this filesystem fits in the page cache limits */ 1840 if (xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_dblocks) || 1841 xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_rblocks)) { 1842 xfs_warn(mp, 1843 "file system too large to be mounted on this system."); 1844 error = -EFBIG; 1845 goto out_free_sb; 1846 } 1847 1848 /* 1849 * XFS block mappings use 54 bits to store the logical block offset. 1850 * This should suffice to handle the maximum file size that the VFS 1851 * supports (currently 2^63 bytes on 64-bit and ULONG_MAX << PAGE_SHIFT 1852 * bytes on 32-bit), but as XFS and VFS have gotten the s_maxbytes 1853 * calculation wrong on 32-bit kernels in the past, we'll add a WARN_ON 1854 * to check this assertion. 1855 * 1856 * Avoid integer overflow by comparing the maximum bmbt offset to the 1857 * maximum pagecache offset in units of fs blocks. 1858 */ 1859 if (!xfs_verify_fileoff(mp, XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE))) { 1860 xfs_warn(mp, 1861 "MAX_LFS_FILESIZE block offset (%llu) exceeds extent map maximum (%llu)!", 1862 XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE), 1863 XFS_MAX_FILEOFF); 1864 error = -EINVAL; 1865 goto out_free_sb; 1866 } 1867 1868 error = xfs_rtmount_readsb(mp); 1869 if (error) 1870 goto out_free_sb; 1871 1872 error = xfs_filestream_mount(mp); 1873 if (error) 1874 goto out_free_rtsb; 1875 1876 /* 1877 * we must configure the block size in the superblock before we run the 1878 * full mount process as the mount process can lookup and cache inodes. 1879 */ 1880 sb->s_magic = XFS_SUPER_MAGIC; 1881 sb->s_blocksize = mp->m_sb.sb_blocksize; 1882 sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1; 1883 sb->s_maxbytes = MAX_LFS_FILESIZE; 1884 sb->s_max_links = XFS_MAXLINK; 1885 sb->s_time_gran = 1; 1886 if (xfs_has_bigtime(mp)) { 1887 sb->s_time_min = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MIN); 1888 sb->s_time_max = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MAX); 1889 } else { 1890 sb->s_time_min = XFS_LEGACY_TIME_MIN; 1891 sb->s_time_max = XFS_LEGACY_TIME_MAX; 1892 } 1893 trace_xfs_inode_timestamp_range(mp, sb->s_time_min, sb->s_time_max); 1894 sb->s_iflags |= SB_I_CGROUPWB | SB_I_ALLOW_HSM; 1895 1896 set_posix_acl_flag(sb); 1897 1898 /* version 5 superblocks support inode version counters. */ 1899 if (xfs_has_crc(mp)) 1900 sb->s_flags |= SB_I_VERSION; 1901 1902 if (xfs_has_dax_always(mp)) { 1903 error = xfs_setup_dax_always(mp); 1904 if (error) 1905 goto out_filestream_unmount; 1906 } 1907 1908 if (xfs_has_discard(mp) && !bdev_max_discard_sectors(sb->s_bdev)) { 1909 xfs_warn(mp, 1910 "mounting with \"discard\" option, but the device does not support discard"); 1911 mp->m_features &= ~XFS_FEAT_DISCARD; 1912 } 1913 1914 if (xfs_has_zoned(mp)) { 1915 if (!xfs_has_metadir(mp)) { 1916 xfs_alert(mp, 1917 "metadir feature required for zoned realtime devices."); 1918 error = -EINVAL; 1919 goto out_filestream_unmount; 1920 } 1921 xfs_warn_experimental(mp, XFS_EXPERIMENTAL_ZONED); 1922 } else if (xfs_has_metadir(mp)) { 1923 xfs_warn_experimental(mp, XFS_EXPERIMENTAL_METADIR); 1924 } 1925 1926 if (xfs_has_reflink(mp)) { 1927 if (xfs_has_realtime(mp) && 1928 !xfs_reflink_supports_rextsize(mp, mp->m_sb.sb_rextsize)) { 1929 xfs_alert(mp, 1930 "reflink not compatible with realtime extent size %u!", 1931 mp->m_sb.sb_rextsize); 1932 error = -EINVAL; 1933 goto out_filestream_unmount; 1934 } 1935 1936 if (xfs_has_zoned(mp)) { 1937 xfs_alert(mp, 1938 "reflink not compatible with zoned RT device!"); 1939 error = -EINVAL; 1940 goto out_filestream_unmount; 1941 } 1942 1943 if (xfs_globals.always_cow) { 1944 xfs_info(mp, "using DEBUG-only always_cow mode."); 1945 mp->m_always_cow = true; 1946 } 1947 } 1948 1949 /* 1950 * If no quota mount options were provided, maybe we'll try to pick 1951 * up the quota accounting and enforcement flags from the ondisk sb. 1952 */ 1953 if (!(mp->m_qflags & XFS_QFLAGS_MNTOPTS)) 1954 xfs_set_resuming_quotaon(mp); 1955 mp->m_qflags &= ~XFS_QFLAGS_MNTOPTS; 1956 1957 error = xfs_mountfs(mp); 1958 if (error) 1959 goto out_filestream_unmount; 1960 1961 root = igrab(VFS_I(mp->m_rootip)); 1962 if (!root) { 1963 error = -ENOENT; 1964 goto out_unmount; 1965 } 1966 sb->s_root = d_make_root(root); 1967 if (!sb->s_root) { 1968 error = -ENOMEM; 1969 goto out_unmount; 1970 } 1971 1972 return 0; 1973 1974 out_filestream_unmount: 1975 xfs_filestream_unmount(mp); 1976 out_free_rtsb: 1977 xfs_rtmount_freesb(mp); 1978 out_free_sb: 1979 xfs_freesb(mp); 1980 out_free_scrub_stats: 1981 xchk_mount_stats_free(mp); 1982 out_free_stats: 1983 free_percpu(mp->m_stats.xs_stats); 1984 out_destroy_inodegc: 1985 xfs_inodegc_free_percpu(mp); 1986 out_destroy_counters: 1987 xfs_destroy_percpu_counters(mp); 1988 out_destroy_workqueues: 1989 xfs_destroy_mount_workqueues(mp); 1990 out_shutdown_devices: 1991 xfs_shutdown_devices(mp); 1992 return error; 1993 1994 out_unmount: 1995 xfs_filestream_unmount(mp); 1996 xfs_unmountfs(mp); 1997 goto out_free_rtsb; 1998 } 1999 2000 static int 2001 xfs_fs_get_tree( 2002 struct fs_context *fc) 2003 { 2004 return get_tree_bdev(fc, xfs_fs_fill_super); 2005 } 2006 2007 static int 2008 xfs_remount_rw( 2009 struct xfs_mount *mp) 2010 { 2011 struct xfs_sb *sbp = &mp->m_sb; 2012 int error; 2013 2014 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp && 2015 xfs_readonly_buftarg(mp->m_logdev_targp)) { 2016 xfs_warn(mp, 2017 "ro->rw transition prohibited by read-only logdev"); 2018 return -EACCES; 2019 } 2020 2021 if (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp)) { 2022 xfs_warn(mp, 2023 "ro->rw transition prohibited by read-only rtdev"); 2024 return -EACCES; 2025 } 2026 2027 if (xfs_has_norecovery(mp)) { 2028 xfs_warn(mp, 2029 "ro->rw transition prohibited on norecovery mount"); 2030 return -EINVAL; 2031 } 2032 2033 if (xfs_sb_is_v5(sbp) && 2034 xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) { 2035 xfs_warn(mp, 2036 "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem", 2037 (sbp->sb_features_ro_compat & 2038 XFS_SB_FEAT_RO_COMPAT_UNKNOWN)); 2039 return -EINVAL; 2040 } 2041 2042 xfs_clear_readonly(mp); 2043 2044 /* 2045 * If this is the first remount to writeable state we might have some 2046 * superblock changes to update. 2047 */ 2048 if (mp->m_update_sb) { 2049 error = xfs_sync_sb(mp, false); 2050 if (error) { 2051 xfs_warn(mp, "failed to write sb changes"); 2052 return error; 2053 } 2054 mp->m_update_sb = false; 2055 } 2056 2057 /* 2058 * Fill out the reserve pool if it is empty. Use the stashed value if 2059 * it is non-zero, otherwise go with the default. 2060 */ 2061 xfs_restore_resvblks(mp); 2062 xfs_log_work_queue(mp); 2063 xfs_blockgc_start(mp); 2064 2065 /* Create the per-AG metadata reservation pool .*/ 2066 error = xfs_fs_reserve_ag_blocks(mp); 2067 if (error && error != -ENOSPC) 2068 return error; 2069 2070 /* Re-enable the background inode inactivation worker. */ 2071 xfs_inodegc_start(mp); 2072 2073 /* Restart zone reclaim */ 2074 xfs_zone_gc_start(mp); 2075 2076 return 0; 2077 } 2078 2079 static int 2080 xfs_remount_ro( 2081 struct xfs_mount *mp) 2082 { 2083 struct xfs_icwalk icw = { 2084 .icw_flags = XFS_ICWALK_FLAG_SYNC, 2085 }; 2086 int error; 2087 2088 /* Flush all the dirty data to disk. */ 2089 error = sync_filesystem(mp->m_super); 2090 if (error) 2091 return error; 2092 2093 /* 2094 * Cancel background eofb scanning so it cannot race with the final 2095 * log force+buftarg wait and deadlock the remount. 2096 */ 2097 xfs_blockgc_stop(mp); 2098 2099 /* 2100 * Clear out all remaining COW staging extents and speculative post-EOF 2101 * preallocations so that we don't leave inodes requiring inactivation 2102 * cleanups during reclaim on a read-only mount. We must process every 2103 * cached inode, so this requires a synchronous cache scan. 2104 */ 2105 error = xfs_blockgc_free_space(mp, &icw); 2106 if (error) { 2107 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); 2108 return error; 2109 } 2110 2111 /* 2112 * Stop the inodegc background worker. xfs_fs_reconfigure already 2113 * flushed all pending inodegc work when it sync'd the filesystem. 2114 * The VFS holds s_umount, so we know that inodes cannot enter 2115 * xfs_fs_destroy_inode during a remount operation. In readonly mode 2116 * we send inodes straight to reclaim, so no inodes will be queued. 2117 */ 2118 xfs_inodegc_stop(mp); 2119 2120 /* Stop zone reclaim */ 2121 xfs_zone_gc_stop(mp); 2122 2123 /* Free the per-AG metadata reservation pool. */ 2124 xfs_fs_unreserve_ag_blocks(mp); 2125 2126 /* 2127 * Before we sync the metadata, we need to free up the reserve block 2128 * pool so that the used block count in the superblock on disk is 2129 * correct at the end of the remount. Stash the current* reserve pool 2130 * size so that if we get remounted rw, we can return it to the same 2131 * size. 2132 */ 2133 xfs_save_resvblks(mp); 2134 2135 xfs_log_clean(mp); 2136 xfs_set_readonly(mp); 2137 2138 return 0; 2139 } 2140 2141 /* 2142 * Logically we would return an error here to prevent users from believing 2143 * they might have changed mount options using remount which can't be changed. 2144 * 2145 * But unfortunately mount(8) adds all options from mtab and fstab to the mount 2146 * arguments in some cases so we can't blindly reject options, but have to 2147 * check for each specified option if it actually differs from the currently 2148 * set option and only reject it if that's the case. 2149 * 2150 * Until that is implemented we return success for every remount request, and 2151 * silently ignore all options that we can't actually change. 2152 */ 2153 static int 2154 xfs_fs_reconfigure( 2155 struct fs_context *fc) 2156 { 2157 struct xfs_mount *mp = XFS_M(fc->root->d_sb); 2158 struct xfs_mount *new_mp = fc->s_fs_info; 2159 int flags = fc->sb_flags; 2160 int error; 2161 2162 new_mp->m_qflags &= ~XFS_QFLAGS_MNTOPTS; 2163 2164 /* version 5 superblocks always support version counters. */ 2165 if (xfs_has_crc(mp)) 2166 fc->sb_flags |= SB_I_VERSION; 2167 2168 error = xfs_fs_validate_params(new_mp); 2169 if (error) 2170 return error; 2171 2172 /* Validate new max_atomic_write option before making other changes */ 2173 if (mp->m_awu_max_bytes != new_mp->m_awu_max_bytes) { 2174 error = xfs_set_max_atomic_write_opt(mp, 2175 new_mp->m_awu_max_bytes); 2176 if (error) 2177 return error; 2178 } 2179 2180 /* inode32 -> inode64 */ 2181 if (xfs_has_small_inums(mp) && !xfs_has_small_inums(new_mp)) { 2182 mp->m_features &= ~XFS_FEAT_SMALL_INUMS; 2183 mp->m_maxagi = xfs_set_inode_alloc(mp, mp->m_sb.sb_agcount); 2184 } 2185 2186 /* inode64 -> inode32 */ 2187 if (!xfs_has_small_inums(mp) && xfs_has_small_inums(new_mp)) { 2188 mp->m_features |= XFS_FEAT_SMALL_INUMS; 2189 mp->m_maxagi = xfs_set_inode_alloc(mp, mp->m_sb.sb_agcount); 2190 } 2191 2192 /* 2193 * Now that mp has been modified according to the remount options, we 2194 * do a final option validation with xfs_finish_flags() just like it is 2195 * just like it is done during mount. We cannot use 2196 * done during mount. We cannot use xfs_finish_flags() on new_mp as it 2197 * contains only the user given options. 2198 */ 2199 error = xfs_finish_flags(mp); 2200 if (error) 2201 return error; 2202 2203 /* ro -> rw */ 2204 if (xfs_is_readonly(mp) && !(flags & SB_RDONLY)) { 2205 error = xfs_remount_rw(mp); 2206 if (error) 2207 return error; 2208 } 2209 2210 /* rw -> ro */ 2211 if (!xfs_is_readonly(mp) && (flags & SB_RDONLY)) { 2212 error = xfs_remount_ro(mp); 2213 if (error) 2214 return error; 2215 } 2216 2217 return 0; 2218 } 2219 2220 static void 2221 xfs_fs_free( 2222 struct fs_context *fc) 2223 { 2224 struct xfs_mount *mp = fc->s_fs_info; 2225 2226 /* 2227 * mp is stored in the fs_context when it is initialized. 2228 * mp is transferred to the superblock on a successful mount, 2229 * but if an error occurs before the transfer we have to free 2230 * it here. 2231 */ 2232 if (mp) 2233 xfs_mount_free(mp); 2234 } 2235 2236 static const struct fs_context_operations xfs_context_ops = { 2237 .parse_param = xfs_fs_parse_param, 2238 .get_tree = xfs_fs_get_tree, 2239 .reconfigure = xfs_fs_reconfigure, 2240 .free = xfs_fs_free, 2241 }; 2242 2243 /* 2244 * WARNING: do not initialise any parameters in this function that depend on 2245 * mount option parsing having already been performed as this can be called from 2246 * fsopen() before any parameters have been set. 2247 */ 2248 static int 2249 xfs_init_fs_context( 2250 struct fs_context *fc) 2251 { 2252 struct xfs_mount *mp; 2253 int i; 2254 2255 mp = kzalloc(sizeof(struct xfs_mount), GFP_KERNEL); 2256 if (!mp) 2257 return -ENOMEM; 2258 2259 spin_lock_init(&mp->m_sb_lock); 2260 for (i = 0; i < XG_TYPE_MAX; i++) 2261 xa_init(&mp->m_groups[i].xa); 2262 mutex_init(&mp->m_growlock); 2263 mutex_init(&mp->m_metafile_resv_lock); 2264 INIT_WORK(&mp->m_flush_inodes_work, xfs_flush_inodes_worker); 2265 INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker); 2266 mp->m_kobj.kobject.kset = xfs_kset; 2267 /* 2268 * We don't create the finobt per-ag space reservation until after log 2269 * recovery, so we must set this to true so that an ifree transaction 2270 * started during log recovery will not depend on space reservations 2271 * for finobt expansion. 2272 */ 2273 mp->m_finobt_nores = true; 2274 2275 /* 2276 * These can be overridden by the mount option parsing. 2277 */ 2278 mp->m_logbufs = -1; 2279 mp->m_logbsize = -1; 2280 mp->m_allocsize_log = 16; /* 64k */ 2281 2282 xfs_hooks_init(&mp->m_dir_update_hooks); 2283 2284 fc->s_fs_info = mp; 2285 fc->ops = &xfs_context_ops; 2286 2287 return 0; 2288 } 2289 2290 static void 2291 xfs_kill_sb( 2292 struct super_block *sb) 2293 { 2294 kill_block_super(sb); 2295 xfs_mount_free(XFS_M(sb)); 2296 } 2297 2298 static struct file_system_type xfs_fs_type = { 2299 .owner = THIS_MODULE, 2300 .name = "xfs", 2301 .init_fs_context = xfs_init_fs_context, 2302 .parameters = xfs_fs_parameters, 2303 .kill_sb = xfs_kill_sb, 2304 .fs_flags = FS_REQUIRES_DEV | FS_ALLOW_IDMAP | FS_MGTIME | 2305 FS_LBS, 2306 }; 2307 MODULE_ALIAS_FS("xfs"); 2308 2309 STATIC int __init 2310 xfs_init_caches(void) 2311 { 2312 int error; 2313 2314 xfs_buf_cache = kmem_cache_create("xfs_buf", sizeof(struct xfs_buf), 0, 2315 SLAB_HWCACHE_ALIGN | 2316 SLAB_RECLAIM_ACCOUNT, 2317 NULL); 2318 if (!xfs_buf_cache) 2319 goto out; 2320 2321 xfs_log_ticket_cache = kmem_cache_create("xfs_log_ticket", 2322 sizeof(struct xlog_ticket), 2323 0, 0, NULL); 2324 if (!xfs_log_ticket_cache) 2325 goto out_destroy_buf_cache; 2326 2327 error = xfs_btree_init_cur_caches(); 2328 if (error) 2329 goto out_destroy_log_ticket_cache; 2330 2331 error = rcbagbt_init_cur_cache(); 2332 if (error) 2333 goto out_destroy_btree_cur_cache; 2334 2335 error = xfs_defer_init_item_caches(); 2336 if (error) 2337 goto out_destroy_rcbagbt_cur_cache; 2338 2339 xfs_da_state_cache = kmem_cache_create("xfs_da_state", 2340 sizeof(struct xfs_da_state), 2341 0, 0, NULL); 2342 if (!xfs_da_state_cache) 2343 goto out_destroy_defer_item_cache; 2344 2345 xfs_ifork_cache = kmem_cache_create("xfs_ifork", 2346 sizeof(struct xfs_ifork), 2347 0, 0, NULL); 2348 if (!xfs_ifork_cache) 2349 goto out_destroy_da_state_cache; 2350 2351 xfs_trans_cache = kmem_cache_create("xfs_trans", 2352 sizeof(struct xfs_trans), 2353 0, 0, NULL); 2354 if (!xfs_trans_cache) 2355 goto out_destroy_ifork_cache; 2356 2357 2358 /* 2359 * The size of the cache-allocated buf log item is the maximum 2360 * size possible under XFS. This wastes a little bit of memory, 2361 * but it is much faster. 2362 */ 2363 xfs_buf_item_cache = kmem_cache_create("xfs_buf_item", 2364 sizeof(struct xfs_buf_log_item), 2365 0, 0, NULL); 2366 if (!xfs_buf_item_cache) 2367 goto out_destroy_trans_cache; 2368 2369 xfs_efd_cache = kmem_cache_create("xfs_efd_item", 2370 xfs_efd_log_item_sizeof(XFS_EFD_MAX_FAST_EXTENTS), 2371 0, 0, NULL); 2372 if (!xfs_efd_cache) 2373 goto out_destroy_buf_item_cache; 2374 2375 xfs_efi_cache = kmem_cache_create("xfs_efi_item", 2376 xfs_efi_log_item_sizeof(XFS_EFI_MAX_FAST_EXTENTS), 2377 0, 0, NULL); 2378 if (!xfs_efi_cache) 2379 goto out_destroy_efd_cache; 2380 2381 xfs_inode_cache = kmem_cache_create("xfs_inode", 2382 sizeof(struct xfs_inode), 0, 2383 (SLAB_HWCACHE_ALIGN | 2384 SLAB_RECLAIM_ACCOUNT | 2385 SLAB_ACCOUNT), 2386 xfs_fs_inode_init_once); 2387 if (!xfs_inode_cache) 2388 goto out_destroy_efi_cache; 2389 2390 xfs_ili_cache = kmem_cache_create("xfs_ili", 2391 sizeof(struct xfs_inode_log_item), 0, 2392 SLAB_RECLAIM_ACCOUNT, 2393 NULL); 2394 if (!xfs_ili_cache) 2395 goto out_destroy_inode_cache; 2396 2397 xfs_icreate_cache = kmem_cache_create("xfs_icr", 2398 sizeof(struct xfs_icreate_item), 2399 0, 0, NULL); 2400 if (!xfs_icreate_cache) 2401 goto out_destroy_ili_cache; 2402 2403 xfs_rud_cache = kmem_cache_create("xfs_rud_item", 2404 sizeof(struct xfs_rud_log_item), 2405 0, 0, NULL); 2406 if (!xfs_rud_cache) 2407 goto out_destroy_icreate_cache; 2408 2409 xfs_rui_cache = kmem_cache_create("xfs_rui_item", 2410 xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS), 2411 0, 0, NULL); 2412 if (!xfs_rui_cache) 2413 goto out_destroy_rud_cache; 2414 2415 xfs_cud_cache = kmem_cache_create("xfs_cud_item", 2416 sizeof(struct xfs_cud_log_item), 2417 0, 0, NULL); 2418 if (!xfs_cud_cache) 2419 goto out_destroy_rui_cache; 2420 2421 xfs_cui_cache = kmem_cache_create("xfs_cui_item", 2422 xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS), 2423 0, 0, NULL); 2424 if (!xfs_cui_cache) 2425 goto out_destroy_cud_cache; 2426 2427 xfs_bud_cache = kmem_cache_create("xfs_bud_item", 2428 sizeof(struct xfs_bud_log_item), 2429 0, 0, NULL); 2430 if (!xfs_bud_cache) 2431 goto out_destroy_cui_cache; 2432 2433 xfs_bui_cache = kmem_cache_create("xfs_bui_item", 2434 xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS), 2435 0, 0, NULL); 2436 if (!xfs_bui_cache) 2437 goto out_destroy_bud_cache; 2438 2439 xfs_attrd_cache = kmem_cache_create("xfs_attrd_item", 2440 sizeof(struct xfs_attrd_log_item), 2441 0, 0, NULL); 2442 if (!xfs_attrd_cache) 2443 goto out_destroy_bui_cache; 2444 2445 xfs_attri_cache = kmem_cache_create("xfs_attri_item", 2446 sizeof(struct xfs_attri_log_item), 2447 0, 0, NULL); 2448 if (!xfs_attri_cache) 2449 goto out_destroy_attrd_cache; 2450 2451 xfs_iunlink_cache = kmem_cache_create("xfs_iul_item", 2452 sizeof(struct xfs_iunlink_item), 2453 0, 0, NULL); 2454 if (!xfs_iunlink_cache) 2455 goto out_destroy_attri_cache; 2456 2457 xfs_xmd_cache = kmem_cache_create("xfs_xmd_item", 2458 sizeof(struct xfs_xmd_log_item), 2459 0, 0, NULL); 2460 if (!xfs_xmd_cache) 2461 goto out_destroy_iul_cache; 2462 2463 xfs_xmi_cache = kmem_cache_create("xfs_xmi_item", 2464 sizeof(struct xfs_xmi_log_item), 2465 0, 0, NULL); 2466 if (!xfs_xmi_cache) 2467 goto out_destroy_xmd_cache; 2468 2469 xfs_parent_args_cache = kmem_cache_create("xfs_parent_args", 2470 sizeof(struct xfs_parent_args), 2471 0, 0, NULL); 2472 if (!xfs_parent_args_cache) 2473 goto out_destroy_xmi_cache; 2474 2475 return 0; 2476 2477 out_destroy_xmi_cache: 2478 kmem_cache_destroy(xfs_xmi_cache); 2479 out_destroy_xmd_cache: 2480 kmem_cache_destroy(xfs_xmd_cache); 2481 out_destroy_iul_cache: 2482 kmem_cache_destroy(xfs_iunlink_cache); 2483 out_destroy_attri_cache: 2484 kmem_cache_destroy(xfs_attri_cache); 2485 out_destroy_attrd_cache: 2486 kmem_cache_destroy(xfs_attrd_cache); 2487 out_destroy_bui_cache: 2488 kmem_cache_destroy(xfs_bui_cache); 2489 out_destroy_bud_cache: 2490 kmem_cache_destroy(xfs_bud_cache); 2491 out_destroy_cui_cache: 2492 kmem_cache_destroy(xfs_cui_cache); 2493 out_destroy_cud_cache: 2494 kmem_cache_destroy(xfs_cud_cache); 2495 out_destroy_rui_cache: 2496 kmem_cache_destroy(xfs_rui_cache); 2497 out_destroy_rud_cache: 2498 kmem_cache_destroy(xfs_rud_cache); 2499 out_destroy_icreate_cache: 2500 kmem_cache_destroy(xfs_icreate_cache); 2501 out_destroy_ili_cache: 2502 kmem_cache_destroy(xfs_ili_cache); 2503 out_destroy_inode_cache: 2504 kmem_cache_destroy(xfs_inode_cache); 2505 out_destroy_efi_cache: 2506 kmem_cache_destroy(xfs_efi_cache); 2507 out_destroy_efd_cache: 2508 kmem_cache_destroy(xfs_efd_cache); 2509 out_destroy_buf_item_cache: 2510 kmem_cache_destroy(xfs_buf_item_cache); 2511 out_destroy_trans_cache: 2512 kmem_cache_destroy(xfs_trans_cache); 2513 out_destroy_ifork_cache: 2514 kmem_cache_destroy(xfs_ifork_cache); 2515 out_destroy_da_state_cache: 2516 kmem_cache_destroy(xfs_da_state_cache); 2517 out_destroy_defer_item_cache: 2518 xfs_defer_destroy_item_caches(); 2519 out_destroy_rcbagbt_cur_cache: 2520 rcbagbt_destroy_cur_cache(); 2521 out_destroy_btree_cur_cache: 2522 xfs_btree_destroy_cur_caches(); 2523 out_destroy_log_ticket_cache: 2524 kmem_cache_destroy(xfs_log_ticket_cache); 2525 out_destroy_buf_cache: 2526 kmem_cache_destroy(xfs_buf_cache); 2527 out: 2528 return -ENOMEM; 2529 } 2530 2531 STATIC void 2532 xfs_destroy_caches(void) 2533 { 2534 /* 2535 * Make sure all delayed rcu free are flushed before we 2536 * destroy caches. 2537 */ 2538 rcu_barrier(); 2539 kmem_cache_destroy(xfs_parent_args_cache); 2540 kmem_cache_destroy(xfs_xmd_cache); 2541 kmem_cache_destroy(xfs_xmi_cache); 2542 kmem_cache_destroy(xfs_iunlink_cache); 2543 kmem_cache_destroy(xfs_attri_cache); 2544 kmem_cache_destroy(xfs_attrd_cache); 2545 kmem_cache_destroy(xfs_bui_cache); 2546 kmem_cache_destroy(xfs_bud_cache); 2547 kmem_cache_destroy(xfs_cui_cache); 2548 kmem_cache_destroy(xfs_cud_cache); 2549 kmem_cache_destroy(xfs_rui_cache); 2550 kmem_cache_destroy(xfs_rud_cache); 2551 kmem_cache_destroy(xfs_icreate_cache); 2552 kmem_cache_destroy(xfs_ili_cache); 2553 kmem_cache_destroy(xfs_inode_cache); 2554 kmem_cache_destroy(xfs_efi_cache); 2555 kmem_cache_destroy(xfs_efd_cache); 2556 kmem_cache_destroy(xfs_buf_item_cache); 2557 kmem_cache_destroy(xfs_trans_cache); 2558 kmem_cache_destroy(xfs_ifork_cache); 2559 kmem_cache_destroy(xfs_da_state_cache); 2560 xfs_defer_destroy_item_caches(); 2561 rcbagbt_destroy_cur_cache(); 2562 xfs_btree_destroy_cur_caches(); 2563 kmem_cache_destroy(xfs_log_ticket_cache); 2564 kmem_cache_destroy(xfs_buf_cache); 2565 } 2566 2567 STATIC int __init 2568 xfs_init_workqueues(void) 2569 { 2570 /* 2571 * The allocation workqueue can be used in memory reclaim situations 2572 * (writepage path), and parallelism is only limited by the number of 2573 * AGs in all the filesystems mounted. Hence use the default large 2574 * max_active value for this workqueue. 2575 */ 2576 xfs_alloc_wq = alloc_workqueue("xfsalloc", XFS_WQFLAGS(WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_PERCPU), 2577 0); 2578 if (!xfs_alloc_wq) 2579 return -ENOMEM; 2580 2581 xfs_discard_wq = alloc_workqueue("xfsdiscard", XFS_WQFLAGS(WQ_UNBOUND), 2582 0); 2583 if (!xfs_discard_wq) 2584 goto out_free_alloc_wq; 2585 2586 return 0; 2587 out_free_alloc_wq: 2588 destroy_workqueue(xfs_alloc_wq); 2589 return -ENOMEM; 2590 } 2591 2592 STATIC void 2593 xfs_destroy_workqueues(void) 2594 { 2595 destroy_workqueue(xfs_discard_wq); 2596 destroy_workqueue(xfs_alloc_wq); 2597 } 2598 2599 STATIC int __init 2600 init_xfs_fs(void) 2601 { 2602 int error; 2603 2604 xfs_check_ondisk_structs(); 2605 2606 error = xfs_dahash_test(); 2607 if (error) 2608 return error; 2609 2610 printk(KERN_INFO XFS_VERSION_STRING " with " 2611 XFS_BUILD_OPTIONS " enabled\n"); 2612 2613 xfs_dir_startup(); 2614 2615 error = xfs_init_caches(); 2616 if (error) 2617 goto out; 2618 2619 error = xfs_init_workqueues(); 2620 if (error) 2621 goto out_destroy_caches; 2622 2623 error = xfs_mru_cache_init(); 2624 if (error) 2625 goto out_destroy_wq; 2626 2627 error = xfs_init_procfs(); 2628 if (error) 2629 goto out_mru_cache_uninit; 2630 2631 error = xfs_sysctl_register(); 2632 if (error) 2633 goto out_cleanup_procfs; 2634 2635 xfs_debugfs = xfs_debugfs_mkdir("xfs", NULL); 2636 2637 xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj); 2638 if (!xfs_kset) { 2639 error = -ENOMEM; 2640 goto out_debugfs_unregister; 2641 } 2642 2643 xfsstats.xs_kobj.kobject.kset = xfs_kset; 2644 2645 xfsstats.xs_stats = alloc_percpu(struct xfsstats); 2646 if (!xfsstats.xs_stats) { 2647 error = -ENOMEM; 2648 goto out_kset_unregister; 2649 } 2650 2651 error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL, 2652 "stats"); 2653 if (error) 2654 goto out_free_stats; 2655 2656 error = xchk_global_stats_setup(xfs_debugfs); 2657 if (error) 2658 goto out_remove_stats_kobj; 2659 2660 #ifdef DEBUG 2661 xfs_dbg_kobj.kobject.kset = xfs_kset; 2662 error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug"); 2663 if (error) 2664 goto out_remove_scrub_stats; 2665 #endif 2666 2667 error = xfs_qm_init(); 2668 if (error) 2669 goto out_remove_dbg_kobj; 2670 2671 error = register_filesystem(&xfs_fs_type); 2672 if (error) 2673 goto out_qm_exit; 2674 return 0; 2675 2676 out_qm_exit: 2677 xfs_qm_exit(); 2678 out_remove_dbg_kobj: 2679 #ifdef DEBUG 2680 xfs_sysfs_del(&xfs_dbg_kobj); 2681 out_remove_scrub_stats: 2682 #endif 2683 xchk_global_stats_teardown(); 2684 out_remove_stats_kobj: 2685 xfs_sysfs_del(&xfsstats.xs_kobj); 2686 out_free_stats: 2687 free_percpu(xfsstats.xs_stats); 2688 out_kset_unregister: 2689 kset_unregister(xfs_kset); 2690 out_debugfs_unregister: 2691 debugfs_remove(xfs_debugfs); 2692 xfs_sysctl_unregister(); 2693 out_cleanup_procfs: 2694 xfs_cleanup_procfs(); 2695 out_mru_cache_uninit: 2696 xfs_mru_cache_uninit(); 2697 out_destroy_wq: 2698 xfs_destroy_workqueues(); 2699 out_destroy_caches: 2700 xfs_destroy_caches(); 2701 out: 2702 return error; 2703 } 2704 2705 STATIC void __exit 2706 exit_xfs_fs(void) 2707 { 2708 xfs_qm_exit(); 2709 unregister_filesystem(&xfs_fs_type); 2710 #ifdef DEBUG 2711 xfs_sysfs_del(&xfs_dbg_kobj); 2712 #endif 2713 xchk_global_stats_teardown(); 2714 xfs_sysfs_del(&xfsstats.xs_kobj); 2715 free_percpu(xfsstats.xs_stats); 2716 kset_unregister(xfs_kset); 2717 debugfs_remove(xfs_debugfs); 2718 xfs_sysctl_unregister(); 2719 xfs_cleanup_procfs(); 2720 xfs_mru_cache_uninit(); 2721 xfs_destroy_workqueues(); 2722 xfs_destroy_caches(); 2723 xfs_uuid_table_free(); 2724 } 2725 2726 module_init(init_xfs_fs); 2727 module_exit(exit_xfs_fs); 2728 2729 MODULE_AUTHOR("Silicon Graphics, Inc."); 2730 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled"); 2731 MODULE_LICENSE("GPL"); 2732