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_da_format.h" 15 #include "xfs_inode.h" 16 #include "xfs_btree.h" 17 #include "xfs_bmap.h" 18 #include "xfs_alloc.h" 19 #include "xfs_error.h" 20 #include "xfs_fsops.h" 21 #include "xfs_trans.h" 22 #include "xfs_buf_item.h" 23 #include "xfs_log.h" 24 #include "xfs_log_priv.h" 25 #include "xfs_da_btree.h" 26 #include "xfs_dir2.h" 27 #include "xfs_extfree_item.h" 28 #include "xfs_mru_cache.h" 29 #include "xfs_inode_item.h" 30 #include "xfs_icache.h" 31 #include "xfs_trace.h" 32 #include "xfs_icreate_item.h" 33 #include "xfs_filestream.h" 34 #include "xfs_quota.h" 35 #include "xfs_sysfs.h" 36 #include "xfs_ondisk.h" 37 #include "xfs_rmap_item.h" 38 #include "xfs_refcount_item.h" 39 #include "xfs_bmap_item.h" 40 #include "xfs_reflink.h" 41 42 #include <linux/namei.h> 43 #include <linux/dax.h> 44 #include <linux/init.h> 45 #include <linux/slab.h> 46 #include <linux/mount.h> 47 #include <linux/mempool.h> 48 #include <linux/writeback.h> 49 #include <linux/kthread.h> 50 #include <linux/freezer.h> 51 #include <linux/parser.h> 52 53 static const struct super_operations xfs_super_operations; 54 struct bio_set xfs_ioend_bioset; 55 56 static struct kset *xfs_kset; /* top-level xfs sysfs dir */ 57 #ifdef DEBUG 58 static struct xfs_kobj xfs_dbg_kobj; /* global debug sysfs attrs */ 59 #endif 60 61 /* 62 * Table driven mount option parser. 63 */ 64 enum { 65 Opt_logbufs, Opt_logbsize, Opt_logdev, Opt_rtdev, Opt_biosize, 66 Opt_wsync, Opt_noalign, Opt_swalloc, Opt_sunit, Opt_swidth, Opt_nouuid, 67 Opt_mtpt, Opt_grpid, Opt_nogrpid, Opt_bsdgroups, Opt_sysvgroups, 68 Opt_allocsize, Opt_norecovery, Opt_barrier, Opt_nobarrier, 69 Opt_inode64, Opt_inode32, Opt_ikeep, Opt_noikeep, 70 Opt_largeio, Opt_nolargeio, Opt_attr2, Opt_noattr2, Opt_filestreams, 71 Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota, Opt_prjquota, 72 Opt_uquota, Opt_gquota, Opt_pquota, 73 Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce, 74 Opt_discard, Opt_nodiscard, Opt_dax, Opt_err, 75 }; 76 77 static const match_table_t tokens = { 78 {Opt_logbufs, "logbufs=%u"}, /* number of XFS log buffers */ 79 {Opt_logbsize, "logbsize=%s"}, /* size of XFS log buffers */ 80 {Opt_logdev, "logdev=%s"}, /* log device */ 81 {Opt_rtdev, "rtdev=%s"}, /* realtime I/O device */ 82 {Opt_biosize, "biosize=%u"}, /* log2 of preferred buffered io size */ 83 {Opt_wsync, "wsync"}, /* safe-mode nfs compatible mount */ 84 {Opt_noalign, "noalign"}, /* turn off stripe alignment */ 85 {Opt_swalloc, "swalloc"}, /* turn on stripe width allocation */ 86 {Opt_sunit, "sunit=%u"}, /* data volume stripe unit */ 87 {Opt_swidth, "swidth=%u"}, /* data volume stripe width */ 88 {Opt_nouuid, "nouuid"}, /* ignore filesystem UUID */ 89 {Opt_mtpt, "mtpt"}, /* filesystem mount point */ 90 {Opt_grpid, "grpid"}, /* group-ID from parent directory */ 91 {Opt_nogrpid, "nogrpid"}, /* group-ID from current process */ 92 {Opt_bsdgroups, "bsdgroups"}, /* group-ID from parent directory */ 93 {Opt_sysvgroups,"sysvgroups"}, /* group-ID from current process */ 94 {Opt_allocsize, "allocsize=%s"},/* preferred allocation size */ 95 {Opt_norecovery,"norecovery"}, /* don't run XFS recovery */ 96 {Opt_inode64, "inode64"}, /* inodes can be allocated anywhere */ 97 {Opt_inode32, "inode32"}, /* inode allocation limited to 98 * XFS_MAXINUMBER_32 */ 99 {Opt_ikeep, "ikeep"}, /* do not free empty inode clusters */ 100 {Opt_noikeep, "noikeep"}, /* free empty inode clusters */ 101 {Opt_largeio, "largeio"}, /* report large I/O sizes in stat() */ 102 {Opt_nolargeio, "nolargeio"}, /* do not report large I/O sizes 103 * in stat(). */ 104 {Opt_attr2, "attr2"}, /* do use attr2 attribute format */ 105 {Opt_noattr2, "noattr2"}, /* do not use attr2 attribute format */ 106 {Opt_filestreams,"filestreams"},/* use filestreams allocator */ 107 {Opt_quota, "quota"}, /* disk quotas (user) */ 108 {Opt_noquota, "noquota"}, /* no quotas */ 109 {Opt_usrquota, "usrquota"}, /* user quota enabled */ 110 {Opt_grpquota, "grpquota"}, /* group quota enabled */ 111 {Opt_prjquota, "prjquota"}, /* project quota enabled */ 112 {Opt_uquota, "uquota"}, /* user quota (IRIX variant) */ 113 {Opt_gquota, "gquota"}, /* group quota (IRIX variant) */ 114 {Opt_pquota, "pquota"}, /* project quota (IRIX variant) */ 115 {Opt_uqnoenforce,"uqnoenforce"},/* user quota limit enforcement */ 116 {Opt_gqnoenforce,"gqnoenforce"},/* group quota limit enforcement */ 117 {Opt_pqnoenforce,"pqnoenforce"},/* project quota limit enforcement */ 118 {Opt_qnoenforce, "qnoenforce"}, /* same as uqnoenforce */ 119 {Opt_discard, "discard"}, /* Discard unused blocks */ 120 {Opt_nodiscard, "nodiscard"}, /* Do not discard unused blocks */ 121 122 {Opt_dax, "dax"}, /* Enable direct access to bdev pages */ 123 124 /* Deprecated mount options scheduled for removal */ 125 {Opt_barrier, "barrier"}, /* use writer barriers for log write and 126 * unwritten extent conversion */ 127 {Opt_nobarrier, "nobarrier"}, /* .. disable */ 128 129 {Opt_err, NULL}, 130 }; 131 132 133 STATIC int 134 suffix_kstrtoint(const substring_t *s, unsigned int base, int *res) 135 { 136 int last, shift_left_factor = 0, _res; 137 char *value; 138 int ret = 0; 139 140 value = match_strdup(s); 141 if (!value) 142 return -ENOMEM; 143 144 last = strlen(value) - 1; 145 if (value[last] == 'K' || value[last] == 'k') { 146 shift_left_factor = 10; 147 value[last] = '\0'; 148 } 149 if (value[last] == 'M' || value[last] == 'm') { 150 shift_left_factor = 20; 151 value[last] = '\0'; 152 } 153 if (value[last] == 'G' || value[last] == 'g') { 154 shift_left_factor = 30; 155 value[last] = '\0'; 156 } 157 158 if (kstrtoint(value, base, &_res)) 159 ret = -EINVAL; 160 kfree(value); 161 *res = _res << shift_left_factor; 162 return ret; 163 } 164 165 /* 166 * This function fills in xfs_mount_t fields based on mount args. 167 * Note: the superblock has _not_ yet been read in. 168 * 169 * Note that this function leaks the various device name allocations on 170 * failure. The caller takes care of them. 171 * 172 * *sb is const because this is also used to test options on the remount 173 * path, and we don't want this to have any side effects at remount time. 174 * Today this function does not change *sb, but just to future-proof... 175 */ 176 STATIC int 177 xfs_parseargs( 178 struct xfs_mount *mp, 179 char *options) 180 { 181 const struct super_block *sb = mp->m_super; 182 char *p; 183 substring_t args[MAX_OPT_ARGS]; 184 int dsunit = 0; 185 int dswidth = 0; 186 int iosize = 0; 187 uint8_t iosizelog = 0; 188 189 /* 190 * set up the mount name first so all the errors will refer to the 191 * correct device. 192 */ 193 mp->m_fsname = kstrndup(sb->s_id, MAXNAMELEN, GFP_KERNEL); 194 if (!mp->m_fsname) 195 return -ENOMEM; 196 mp->m_fsname_len = strlen(mp->m_fsname) + 1; 197 198 /* 199 * Copy binary VFS mount flags we are interested in. 200 */ 201 if (sb_rdonly(sb)) 202 mp->m_flags |= XFS_MOUNT_RDONLY; 203 if (sb->s_flags & SB_DIRSYNC) 204 mp->m_flags |= XFS_MOUNT_DIRSYNC; 205 if (sb->s_flags & SB_SYNCHRONOUS) 206 mp->m_flags |= XFS_MOUNT_WSYNC; 207 208 /* 209 * Set some default flags that could be cleared by the mount option 210 * parsing. 211 */ 212 mp->m_flags |= XFS_MOUNT_BARRIER; 213 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE; 214 215 /* 216 * These can be overridden by the mount option parsing. 217 */ 218 mp->m_logbufs = -1; 219 mp->m_logbsize = -1; 220 221 if (!options) 222 goto done; 223 224 while ((p = strsep(&options, ",")) != NULL) { 225 int token; 226 227 if (!*p) 228 continue; 229 230 token = match_token(p, tokens, args); 231 switch (token) { 232 case Opt_logbufs: 233 if (match_int(args, &mp->m_logbufs)) 234 return -EINVAL; 235 break; 236 case Opt_logbsize: 237 if (suffix_kstrtoint(args, 10, &mp->m_logbsize)) 238 return -EINVAL; 239 break; 240 case Opt_logdev: 241 kfree(mp->m_logname); 242 mp->m_logname = match_strdup(args); 243 if (!mp->m_logname) 244 return -ENOMEM; 245 break; 246 case Opt_mtpt: 247 xfs_warn(mp, "%s option not allowed on this system", p); 248 return -EINVAL; 249 case Opt_rtdev: 250 kfree(mp->m_rtname); 251 mp->m_rtname = match_strdup(args); 252 if (!mp->m_rtname) 253 return -ENOMEM; 254 break; 255 case Opt_allocsize: 256 case Opt_biosize: 257 if (suffix_kstrtoint(args, 10, &iosize)) 258 return -EINVAL; 259 iosizelog = ffs(iosize) - 1; 260 break; 261 case Opt_grpid: 262 case Opt_bsdgroups: 263 mp->m_flags |= XFS_MOUNT_GRPID; 264 break; 265 case Opt_nogrpid: 266 case Opt_sysvgroups: 267 mp->m_flags &= ~XFS_MOUNT_GRPID; 268 break; 269 case Opt_wsync: 270 mp->m_flags |= XFS_MOUNT_WSYNC; 271 break; 272 case Opt_norecovery: 273 mp->m_flags |= XFS_MOUNT_NORECOVERY; 274 break; 275 case Opt_noalign: 276 mp->m_flags |= XFS_MOUNT_NOALIGN; 277 break; 278 case Opt_swalloc: 279 mp->m_flags |= XFS_MOUNT_SWALLOC; 280 break; 281 case Opt_sunit: 282 if (match_int(args, &dsunit)) 283 return -EINVAL; 284 break; 285 case Opt_swidth: 286 if (match_int(args, &dswidth)) 287 return -EINVAL; 288 break; 289 case Opt_inode32: 290 mp->m_flags |= XFS_MOUNT_SMALL_INUMS; 291 break; 292 case Opt_inode64: 293 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS; 294 break; 295 case Opt_nouuid: 296 mp->m_flags |= XFS_MOUNT_NOUUID; 297 break; 298 case Opt_ikeep: 299 mp->m_flags |= XFS_MOUNT_IKEEP; 300 break; 301 case Opt_noikeep: 302 mp->m_flags &= ~XFS_MOUNT_IKEEP; 303 break; 304 case Opt_largeio: 305 mp->m_flags &= ~XFS_MOUNT_COMPAT_IOSIZE; 306 break; 307 case Opt_nolargeio: 308 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE; 309 break; 310 case Opt_attr2: 311 mp->m_flags |= XFS_MOUNT_ATTR2; 312 break; 313 case Opt_noattr2: 314 mp->m_flags &= ~XFS_MOUNT_ATTR2; 315 mp->m_flags |= XFS_MOUNT_NOATTR2; 316 break; 317 case Opt_filestreams: 318 mp->m_flags |= XFS_MOUNT_FILESTREAMS; 319 break; 320 case Opt_noquota: 321 mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT; 322 mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD; 323 mp->m_qflags &= ~XFS_ALL_QUOTA_ACTIVE; 324 break; 325 case Opt_quota: 326 case Opt_uquota: 327 case Opt_usrquota: 328 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE | 329 XFS_UQUOTA_ENFD); 330 break; 331 case Opt_qnoenforce: 332 case Opt_uqnoenforce: 333 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE); 334 mp->m_qflags &= ~XFS_UQUOTA_ENFD; 335 break; 336 case Opt_pquota: 337 case Opt_prjquota: 338 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE | 339 XFS_PQUOTA_ENFD); 340 break; 341 case Opt_pqnoenforce: 342 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE); 343 mp->m_qflags &= ~XFS_PQUOTA_ENFD; 344 break; 345 case Opt_gquota: 346 case Opt_grpquota: 347 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE | 348 XFS_GQUOTA_ENFD); 349 break; 350 case Opt_gqnoenforce: 351 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE); 352 mp->m_qflags &= ~XFS_GQUOTA_ENFD; 353 break; 354 case Opt_discard: 355 mp->m_flags |= XFS_MOUNT_DISCARD; 356 break; 357 case Opt_nodiscard: 358 mp->m_flags &= ~XFS_MOUNT_DISCARD; 359 break; 360 #ifdef CONFIG_FS_DAX 361 case Opt_dax: 362 mp->m_flags |= XFS_MOUNT_DAX; 363 break; 364 #endif 365 case Opt_barrier: 366 xfs_warn(mp, "%s option is deprecated, ignoring.", p); 367 mp->m_flags |= XFS_MOUNT_BARRIER; 368 break; 369 case Opt_nobarrier: 370 xfs_warn(mp, "%s option is deprecated, ignoring.", p); 371 mp->m_flags &= ~XFS_MOUNT_BARRIER; 372 break; 373 default: 374 xfs_warn(mp, "unknown mount option [%s].", p); 375 return -EINVAL; 376 } 377 } 378 379 /* 380 * no recovery flag requires a read-only mount 381 */ 382 if ((mp->m_flags & XFS_MOUNT_NORECOVERY) && 383 !(mp->m_flags & XFS_MOUNT_RDONLY)) { 384 xfs_warn(mp, "no-recovery mounts must be read-only."); 385 return -EINVAL; 386 } 387 388 if ((mp->m_flags & XFS_MOUNT_NOALIGN) && (dsunit || dswidth)) { 389 xfs_warn(mp, 390 "sunit and swidth options incompatible with the noalign option"); 391 return -EINVAL; 392 } 393 394 #ifndef CONFIG_XFS_QUOTA 395 if (XFS_IS_QUOTA_RUNNING(mp)) { 396 xfs_warn(mp, "quota support not available in this kernel."); 397 return -EINVAL; 398 } 399 #endif 400 401 if ((dsunit && !dswidth) || (!dsunit && dswidth)) { 402 xfs_warn(mp, "sunit and swidth must be specified together"); 403 return -EINVAL; 404 } 405 406 if (dsunit && (dswidth % dsunit != 0)) { 407 xfs_warn(mp, 408 "stripe width (%d) must be a multiple of the stripe unit (%d)", 409 dswidth, dsunit); 410 return -EINVAL; 411 } 412 413 done: 414 if (dsunit && !(mp->m_flags & XFS_MOUNT_NOALIGN)) { 415 /* 416 * At this point the superblock has not been read 417 * in, therefore we do not know the block size. 418 * Before the mount call ends we will convert 419 * these to FSBs. 420 */ 421 mp->m_dalign = dsunit; 422 mp->m_swidth = dswidth; 423 } 424 425 if (mp->m_logbufs != -1 && 426 mp->m_logbufs != 0 && 427 (mp->m_logbufs < XLOG_MIN_ICLOGS || 428 mp->m_logbufs > XLOG_MAX_ICLOGS)) { 429 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]", 430 mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS); 431 return -EINVAL; 432 } 433 if (mp->m_logbsize != -1 && 434 mp->m_logbsize != 0 && 435 (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE || 436 mp->m_logbsize > XLOG_MAX_RECORD_BSIZE || 437 !is_power_of_2(mp->m_logbsize))) { 438 xfs_warn(mp, 439 "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]", 440 mp->m_logbsize); 441 return -EINVAL; 442 } 443 444 if (iosizelog) { 445 if (iosizelog > XFS_MAX_IO_LOG || 446 iosizelog < XFS_MIN_IO_LOG) { 447 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]", 448 iosizelog, XFS_MIN_IO_LOG, 449 XFS_MAX_IO_LOG); 450 return -EINVAL; 451 } 452 453 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE; 454 mp->m_readio_log = iosizelog; 455 mp->m_writeio_log = iosizelog; 456 } 457 458 return 0; 459 } 460 461 struct proc_xfs_info { 462 uint64_t flag; 463 char *str; 464 }; 465 466 STATIC int 467 xfs_showargs( 468 struct xfs_mount *mp, 469 struct seq_file *m) 470 { 471 static struct proc_xfs_info xfs_info_set[] = { 472 /* the few simple ones we can get from the mount struct */ 473 { XFS_MOUNT_IKEEP, ",ikeep" }, 474 { XFS_MOUNT_WSYNC, ",wsync" }, 475 { XFS_MOUNT_NOALIGN, ",noalign" }, 476 { XFS_MOUNT_SWALLOC, ",swalloc" }, 477 { XFS_MOUNT_NOUUID, ",nouuid" }, 478 { XFS_MOUNT_NORECOVERY, ",norecovery" }, 479 { XFS_MOUNT_ATTR2, ",attr2" }, 480 { XFS_MOUNT_FILESTREAMS, ",filestreams" }, 481 { XFS_MOUNT_GRPID, ",grpid" }, 482 { XFS_MOUNT_DISCARD, ",discard" }, 483 { XFS_MOUNT_SMALL_INUMS, ",inode32" }, 484 { XFS_MOUNT_DAX, ",dax" }, 485 { 0, NULL } 486 }; 487 static struct proc_xfs_info xfs_info_unset[] = { 488 /* the few simple ones we can get from the mount struct */ 489 { XFS_MOUNT_COMPAT_IOSIZE, ",largeio" }, 490 { XFS_MOUNT_BARRIER, ",nobarrier" }, 491 { XFS_MOUNT_SMALL_INUMS, ",inode64" }, 492 { 0, NULL } 493 }; 494 struct proc_xfs_info *xfs_infop; 495 496 for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) { 497 if (mp->m_flags & xfs_infop->flag) 498 seq_puts(m, xfs_infop->str); 499 } 500 for (xfs_infop = xfs_info_unset; xfs_infop->flag; xfs_infop++) { 501 if (!(mp->m_flags & xfs_infop->flag)) 502 seq_puts(m, xfs_infop->str); 503 } 504 505 if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) 506 seq_printf(m, ",allocsize=%dk", 507 (int)(1 << mp->m_writeio_log) >> 10); 508 509 if (mp->m_logbufs > 0) 510 seq_printf(m, ",logbufs=%d", mp->m_logbufs); 511 if (mp->m_logbsize > 0) 512 seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10); 513 514 if (mp->m_logname) 515 seq_show_option(m, "logdev", mp->m_logname); 516 if (mp->m_rtname) 517 seq_show_option(m, "rtdev", mp->m_rtname); 518 519 if (mp->m_dalign > 0) 520 seq_printf(m, ",sunit=%d", 521 (int)XFS_FSB_TO_BB(mp, mp->m_dalign)); 522 if (mp->m_swidth > 0) 523 seq_printf(m, ",swidth=%d", 524 (int)XFS_FSB_TO_BB(mp, mp->m_swidth)); 525 526 if (mp->m_qflags & (XFS_UQUOTA_ACCT|XFS_UQUOTA_ENFD)) 527 seq_puts(m, ",usrquota"); 528 else if (mp->m_qflags & XFS_UQUOTA_ACCT) 529 seq_puts(m, ",uqnoenforce"); 530 531 if (mp->m_qflags & XFS_PQUOTA_ACCT) { 532 if (mp->m_qflags & XFS_PQUOTA_ENFD) 533 seq_puts(m, ",prjquota"); 534 else 535 seq_puts(m, ",pqnoenforce"); 536 } 537 if (mp->m_qflags & XFS_GQUOTA_ACCT) { 538 if (mp->m_qflags & XFS_GQUOTA_ENFD) 539 seq_puts(m, ",grpquota"); 540 else 541 seq_puts(m, ",gqnoenforce"); 542 } 543 544 if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT)) 545 seq_puts(m, ",noquota"); 546 547 return 0; 548 } 549 static uint64_t 550 xfs_max_file_offset( 551 unsigned int blockshift) 552 { 553 unsigned int pagefactor = 1; 554 unsigned int bitshift = BITS_PER_LONG - 1; 555 556 /* Figure out maximum filesize, on Linux this can depend on 557 * the filesystem blocksize (on 32 bit platforms). 558 * __block_write_begin does this in an [unsigned] long... 559 * page->index << (PAGE_SHIFT - bbits) 560 * So, for page sized blocks (4K on 32 bit platforms), 561 * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is 562 * (((u64)PAGE_SIZE << (BITS_PER_LONG-1))-1) 563 * but for smaller blocksizes it is less (bbits = log2 bsize). 564 * Note1: get_block_t takes a long (implicit cast from above) 565 * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch 566 * can optionally convert the [unsigned] long from above into 567 * an [unsigned] long long. 568 */ 569 570 #if BITS_PER_LONG == 32 571 # if defined(CONFIG_LBDAF) 572 ASSERT(sizeof(sector_t) == 8); 573 pagefactor = PAGE_SIZE; 574 bitshift = BITS_PER_LONG; 575 # else 576 pagefactor = PAGE_SIZE >> (PAGE_SHIFT - blockshift); 577 # endif 578 #endif 579 580 return (((uint64_t)pagefactor) << bitshift) - 1; 581 } 582 583 /* 584 * Set parameters for inode allocation heuristics, taking into account 585 * filesystem size and inode32/inode64 mount options; i.e. specifically 586 * whether or not XFS_MOUNT_SMALL_INUMS is set. 587 * 588 * Inode allocation patterns are altered only if inode32 is requested 589 * (XFS_MOUNT_SMALL_INUMS), and the filesystem is sufficiently large. 590 * If altered, XFS_MOUNT_32BITINODES is set as well. 591 * 592 * An agcount independent of that in the mount structure is provided 593 * because in the growfs case, mp->m_sb.sb_agcount is not yet updated 594 * to the potentially higher ag count. 595 * 596 * Returns the maximum AG index which may contain inodes. 597 */ 598 xfs_agnumber_t 599 xfs_set_inode_alloc( 600 struct xfs_mount *mp, 601 xfs_agnumber_t agcount) 602 { 603 xfs_agnumber_t index; 604 xfs_agnumber_t maxagi = 0; 605 xfs_sb_t *sbp = &mp->m_sb; 606 xfs_agnumber_t max_metadata; 607 xfs_agino_t agino; 608 xfs_ino_t ino; 609 610 /* 611 * Calculate how much should be reserved for inodes to meet 612 * the max inode percentage. Used only for inode32. 613 */ 614 if (mp->m_maxicount) { 615 uint64_t icount; 616 617 icount = sbp->sb_dblocks * sbp->sb_imax_pct; 618 do_div(icount, 100); 619 icount += sbp->sb_agblocks - 1; 620 do_div(icount, sbp->sb_agblocks); 621 max_metadata = icount; 622 } else { 623 max_metadata = agcount; 624 } 625 626 /* Get the last possible inode in the filesystem */ 627 agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0); 628 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino); 629 630 /* 631 * If user asked for no more than 32-bit inodes, and the fs is 632 * sufficiently large, set XFS_MOUNT_32BITINODES if we must alter 633 * the allocator to accommodate the request. 634 */ 635 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32) 636 mp->m_flags |= XFS_MOUNT_32BITINODES; 637 else 638 mp->m_flags &= ~XFS_MOUNT_32BITINODES; 639 640 for (index = 0; index < agcount; index++) { 641 struct xfs_perag *pag; 642 643 ino = XFS_AGINO_TO_INO(mp, index, agino); 644 645 pag = xfs_perag_get(mp, index); 646 647 if (mp->m_flags & XFS_MOUNT_32BITINODES) { 648 if (ino > XFS_MAXINUMBER_32) { 649 pag->pagi_inodeok = 0; 650 pag->pagf_metadata = 0; 651 } else { 652 pag->pagi_inodeok = 1; 653 maxagi++; 654 if (index < max_metadata) 655 pag->pagf_metadata = 1; 656 else 657 pag->pagf_metadata = 0; 658 } 659 } else { 660 pag->pagi_inodeok = 1; 661 pag->pagf_metadata = 0; 662 } 663 664 xfs_perag_put(pag); 665 } 666 667 return (mp->m_flags & XFS_MOUNT_32BITINODES) ? maxagi : agcount; 668 } 669 670 STATIC int 671 xfs_blkdev_get( 672 xfs_mount_t *mp, 673 const char *name, 674 struct block_device **bdevp) 675 { 676 int error = 0; 677 678 *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL, 679 mp); 680 if (IS_ERR(*bdevp)) { 681 error = PTR_ERR(*bdevp); 682 xfs_warn(mp, "Invalid device [%s], error=%d", name, error); 683 } 684 685 return error; 686 } 687 688 STATIC void 689 xfs_blkdev_put( 690 struct block_device *bdev) 691 { 692 if (bdev) 693 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); 694 } 695 696 void 697 xfs_blkdev_issue_flush( 698 xfs_buftarg_t *buftarg) 699 { 700 blkdev_issue_flush(buftarg->bt_bdev, GFP_NOFS, NULL); 701 } 702 703 STATIC void 704 xfs_close_devices( 705 struct xfs_mount *mp) 706 { 707 struct dax_device *dax_ddev = mp->m_ddev_targp->bt_daxdev; 708 709 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) { 710 struct block_device *logdev = mp->m_logdev_targp->bt_bdev; 711 struct dax_device *dax_logdev = mp->m_logdev_targp->bt_daxdev; 712 713 xfs_free_buftarg(mp->m_logdev_targp); 714 xfs_blkdev_put(logdev); 715 fs_put_dax(dax_logdev); 716 } 717 if (mp->m_rtdev_targp) { 718 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev; 719 struct dax_device *dax_rtdev = mp->m_rtdev_targp->bt_daxdev; 720 721 xfs_free_buftarg(mp->m_rtdev_targp); 722 xfs_blkdev_put(rtdev); 723 fs_put_dax(dax_rtdev); 724 } 725 xfs_free_buftarg(mp->m_ddev_targp); 726 fs_put_dax(dax_ddev); 727 } 728 729 /* 730 * The file system configurations are: 731 * (1) device (partition) with data and internal log 732 * (2) logical volume with data and log subvolumes. 733 * (3) logical volume with data, log, and realtime subvolumes. 734 * 735 * We only have to handle opening the log and realtime volumes here if 736 * they are present. The data subvolume has already been opened by 737 * get_sb_bdev() and is stored in sb->s_bdev. 738 */ 739 STATIC int 740 xfs_open_devices( 741 struct xfs_mount *mp) 742 { 743 struct block_device *ddev = mp->m_super->s_bdev; 744 struct dax_device *dax_ddev = fs_dax_get_by_bdev(ddev); 745 struct dax_device *dax_logdev = NULL, *dax_rtdev = NULL; 746 struct block_device *logdev = NULL, *rtdev = NULL; 747 int error; 748 749 /* 750 * Open real time and log devices - order is important. 751 */ 752 if (mp->m_logname) { 753 error = xfs_blkdev_get(mp, mp->m_logname, &logdev); 754 if (error) 755 goto out; 756 dax_logdev = fs_dax_get_by_bdev(logdev); 757 } 758 759 if (mp->m_rtname) { 760 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev); 761 if (error) 762 goto out_close_logdev; 763 764 if (rtdev == ddev || rtdev == logdev) { 765 xfs_warn(mp, 766 "Cannot mount filesystem with identical rtdev and ddev/logdev."); 767 error = -EINVAL; 768 goto out_close_rtdev; 769 } 770 dax_rtdev = fs_dax_get_by_bdev(rtdev); 771 } 772 773 /* 774 * Setup xfs_mount buffer target pointers 775 */ 776 error = -ENOMEM; 777 mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev, dax_ddev); 778 if (!mp->m_ddev_targp) 779 goto out_close_rtdev; 780 781 if (rtdev) { 782 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev, dax_rtdev); 783 if (!mp->m_rtdev_targp) 784 goto out_free_ddev_targ; 785 } 786 787 if (logdev && logdev != ddev) { 788 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev, dax_logdev); 789 if (!mp->m_logdev_targp) 790 goto out_free_rtdev_targ; 791 } else { 792 mp->m_logdev_targp = mp->m_ddev_targp; 793 } 794 795 return 0; 796 797 out_free_rtdev_targ: 798 if (mp->m_rtdev_targp) 799 xfs_free_buftarg(mp->m_rtdev_targp); 800 out_free_ddev_targ: 801 xfs_free_buftarg(mp->m_ddev_targp); 802 out_close_rtdev: 803 xfs_blkdev_put(rtdev); 804 fs_put_dax(dax_rtdev); 805 out_close_logdev: 806 if (logdev && logdev != ddev) { 807 xfs_blkdev_put(logdev); 808 fs_put_dax(dax_logdev); 809 } 810 out: 811 fs_put_dax(dax_ddev); 812 return error; 813 } 814 815 /* 816 * Setup xfs_mount buffer target pointers based on superblock 817 */ 818 STATIC int 819 xfs_setup_devices( 820 struct xfs_mount *mp) 821 { 822 int error; 823 824 error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize); 825 if (error) 826 return error; 827 828 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) { 829 unsigned int log_sector_size = BBSIZE; 830 831 if (xfs_sb_version_hassector(&mp->m_sb)) 832 log_sector_size = mp->m_sb.sb_logsectsize; 833 error = xfs_setsize_buftarg(mp->m_logdev_targp, 834 log_sector_size); 835 if (error) 836 return error; 837 } 838 if (mp->m_rtdev_targp) { 839 error = xfs_setsize_buftarg(mp->m_rtdev_targp, 840 mp->m_sb.sb_sectsize); 841 if (error) 842 return error; 843 } 844 845 return 0; 846 } 847 848 STATIC int 849 xfs_init_mount_workqueues( 850 struct xfs_mount *mp) 851 { 852 mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s", 853 WQ_MEM_RECLAIM|WQ_FREEZABLE, 1, mp->m_fsname); 854 if (!mp->m_buf_workqueue) 855 goto out; 856 857 mp->m_data_workqueue = alloc_workqueue("xfs-data/%s", 858 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname); 859 if (!mp->m_data_workqueue) 860 goto out_destroy_buf; 861 862 mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s", 863 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname); 864 if (!mp->m_unwritten_workqueue) 865 goto out_destroy_data_iodone_queue; 866 867 mp->m_cil_workqueue = alloc_workqueue("xfs-cil/%s", 868 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname); 869 if (!mp->m_cil_workqueue) 870 goto out_destroy_unwritten; 871 872 mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s", 873 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname); 874 if (!mp->m_reclaim_workqueue) 875 goto out_destroy_cil; 876 877 mp->m_log_workqueue = alloc_workqueue("xfs-log/%s", 878 WQ_MEM_RECLAIM|WQ_FREEZABLE|WQ_HIGHPRI, 0, 879 mp->m_fsname); 880 if (!mp->m_log_workqueue) 881 goto out_destroy_reclaim; 882 883 mp->m_eofblocks_workqueue = alloc_workqueue("xfs-eofblocks/%s", 884 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname); 885 if (!mp->m_eofblocks_workqueue) 886 goto out_destroy_log; 887 888 mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s", WQ_FREEZABLE, 0, 889 mp->m_fsname); 890 if (!mp->m_sync_workqueue) 891 goto out_destroy_eofb; 892 893 return 0; 894 895 out_destroy_eofb: 896 destroy_workqueue(mp->m_eofblocks_workqueue); 897 out_destroy_log: 898 destroy_workqueue(mp->m_log_workqueue); 899 out_destroy_reclaim: 900 destroy_workqueue(mp->m_reclaim_workqueue); 901 out_destroy_cil: 902 destroy_workqueue(mp->m_cil_workqueue); 903 out_destroy_unwritten: 904 destroy_workqueue(mp->m_unwritten_workqueue); 905 out_destroy_data_iodone_queue: 906 destroy_workqueue(mp->m_data_workqueue); 907 out_destroy_buf: 908 destroy_workqueue(mp->m_buf_workqueue); 909 out: 910 return -ENOMEM; 911 } 912 913 STATIC void 914 xfs_destroy_mount_workqueues( 915 struct xfs_mount *mp) 916 { 917 destroy_workqueue(mp->m_sync_workqueue); 918 destroy_workqueue(mp->m_eofblocks_workqueue); 919 destroy_workqueue(mp->m_log_workqueue); 920 destroy_workqueue(mp->m_reclaim_workqueue); 921 destroy_workqueue(mp->m_cil_workqueue); 922 destroy_workqueue(mp->m_data_workqueue); 923 destroy_workqueue(mp->m_unwritten_workqueue); 924 destroy_workqueue(mp->m_buf_workqueue); 925 } 926 927 /* 928 * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK 929 * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting 930 * for IO to complete so that we effectively throttle multiple callers to the 931 * rate at which IO is completing. 932 */ 933 void 934 xfs_flush_inodes( 935 struct xfs_mount *mp) 936 { 937 struct super_block *sb = mp->m_super; 938 939 if (down_read_trylock(&sb->s_umount)) { 940 sync_inodes_sb(sb); 941 up_read(&sb->s_umount); 942 } 943 } 944 945 /* Catch misguided souls that try to use this interface on XFS */ 946 STATIC struct inode * 947 xfs_fs_alloc_inode( 948 struct super_block *sb) 949 { 950 BUG(); 951 return NULL; 952 } 953 954 /* 955 * Now that the generic code is guaranteed not to be accessing 956 * the linux inode, we can inactivate and reclaim the inode. 957 */ 958 STATIC void 959 xfs_fs_destroy_inode( 960 struct inode *inode) 961 { 962 struct xfs_inode *ip = XFS_I(inode); 963 964 trace_xfs_destroy_inode(ip); 965 966 ASSERT(!rwsem_is_locked(&inode->i_rwsem)); 967 XFS_STATS_INC(ip->i_mount, vn_rele); 968 XFS_STATS_INC(ip->i_mount, vn_remove); 969 970 xfs_inactive(ip); 971 972 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0); 973 XFS_STATS_INC(ip->i_mount, vn_reclaim); 974 975 /* 976 * We should never get here with one of the reclaim flags already set. 977 */ 978 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIMABLE)); 979 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIM)); 980 981 /* 982 * We always use background reclaim here because even if the 983 * inode is clean, it still may be under IO and hence we have 984 * to take the flush lock. The background reclaim path handles 985 * this more efficiently than we can here, so simply let background 986 * reclaim tear down all inodes. 987 */ 988 xfs_inode_set_reclaim_tag(ip); 989 } 990 991 static void 992 xfs_fs_dirty_inode( 993 struct inode *inode, 994 int flag) 995 { 996 struct xfs_inode *ip = XFS_I(inode); 997 struct xfs_mount *mp = ip->i_mount; 998 struct xfs_trans *tp; 999 1000 if (!(inode->i_sb->s_flags & SB_LAZYTIME)) 1001 return; 1002 if (flag != I_DIRTY_SYNC || !(inode->i_state & I_DIRTY_TIME)) 1003 return; 1004 1005 if (xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp)) 1006 return; 1007 xfs_ilock(ip, XFS_ILOCK_EXCL); 1008 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); 1009 xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP); 1010 xfs_trans_commit(tp); 1011 } 1012 1013 /* 1014 * Slab object creation initialisation for the XFS inode. 1015 * This covers only the idempotent fields in the XFS inode; 1016 * all other fields need to be initialised on allocation 1017 * from the slab. This avoids the need to repeatedly initialise 1018 * fields in the xfs inode that left in the initialise state 1019 * when freeing the inode. 1020 */ 1021 STATIC void 1022 xfs_fs_inode_init_once( 1023 void *inode) 1024 { 1025 struct xfs_inode *ip = inode; 1026 1027 memset(ip, 0, sizeof(struct xfs_inode)); 1028 1029 /* vfs inode */ 1030 inode_init_once(VFS_I(ip)); 1031 1032 /* xfs inode */ 1033 atomic_set(&ip->i_pincount, 0); 1034 spin_lock_init(&ip->i_flags_lock); 1035 1036 mrlock_init(&ip->i_mmaplock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER, 1037 "xfsino", ip->i_ino); 1038 mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER, 1039 "xfsino", ip->i_ino); 1040 } 1041 1042 /* 1043 * We do an unlocked check for XFS_IDONTCACHE here because we are already 1044 * serialised against cache hits here via the inode->i_lock and igrab() in 1045 * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be 1046 * racing with us, and it avoids needing to grab a spinlock here for every inode 1047 * we drop the final reference on. 1048 */ 1049 STATIC int 1050 xfs_fs_drop_inode( 1051 struct inode *inode) 1052 { 1053 struct xfs_inode *ip = XFS_I(inode); 1054 1055 /* 1056 * If this unlinked inode is in the middle of recovery, don't 1057 * drop the inode just yet; log recovery will take care of 1058 * that. See the comment for this inode flag. 1059 */ 1060 if (ip->i_flags & XFS_IRECOVERY) { 1061 ASSERT(ip->i_mount->m_log->l_flags & XLOG_RECOVERY_NEEDED); 1062 return 0; 1063 } 1064 1065 return generic_drop_inode(inode) || (ip->i_flags & XFS_IDONTCACHE); 1066 } 1067 1068 STATIC void 1069 xfs_free_fsname( 1070 struct xfs_mount *mp) 1071 { 1072 kfree(mp->m_fsname); 1073 kfree(mp->m_rtname); 1074 kfree(mp->m_logname); 1075 } 1076 1077 STATIC int 1078 xfs_fs_sync_fs( 1079 struct super_block *sb, 1080 int wait) 1081 { 1082 struct xfs_mount *mp = XFS_M(sb); 1083 1084 /* 1085 * Doing anything during the async pass would be counterproductive. 1086 */ 1087 if (!wait) 1088 return 0; 1089 1090 xfs_log_force(mp, XFS_LOG_SYNC); 1091 if (laptop_mode) { 1092 /* 1093 * The disk must be active because we're syncing. 1094 * We schedule log work now (now that the disk is 1095 * active) instead of later (when it might not be). 1096 */ 1097 flush_delayed_work(&mp->m_log->l_work); 1098 } 1099 1100 return 0; 1101 } 1102 1103 STATIC int 1104 xfs_fs_statfs( 1105 struct dentry *dentry, 1106 struct kstatfs *statp) 1107 { 1108 struct xfs_mount *mp = XFS_M(dentry->d_sb); 1109 xfs_sb_t *sbp = &mp->m_sb; 1110 struct xfs_inode *ip = XFS_I(d_inode(dentry)); 1111 uint64_t fakeinos, id; 1112 uint64_t icount; 1113 uint64_t ifree; 1114 uint64_t fdblocks; 1115 xfs_extlen_t lsize; 1116 int64_t ffree; 1117 1118 statp->f_type = XFS_SB_MAGIC; 1119 statp->f_namelen = MAXNAMELEN - 1; 1120 1121 id = huge_encode_dev(mp->m_ddev_targp->bt_dev); 1122 statp->f_fsid.val[0] = (u32)id; 1123 statp->f_fsid.val[1] = (u32)(id >> 32); 1124 1125 icount = percpu_counter_sum(&mp->m_icount); 1126 ifree = percpu_counter_sum(&mp->m_ifree); 1127 fdblocks = percpu_counter_sum(&mp->m_fdblocks); 1128 1129 spin_lock(&mp->m_sb_lock); 1130 statp->f_bsize = sbp->sb_blocksize; 1131 lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0; 1132 statp->f_blocks = sbp->sb_dblocks - lsize; 1133 spin_unlock(&mp->m_sb_lock); 1134 1135 statp->f_bfree = fdblocks - mp->m_alloc_set_aside; 1136 statp->f_bavail = statp->f_bfree; 1137 1138 fakeinos = statp->f_bfree << sbp->sb_inopblog; 1139 statp->f_files = min(icount + fakeinos, (uint64_t)XFS_MAXINUMBER); 1140 if (mp->m_maxicount) 1141 statp->f_files = min_t(typeof(statp->f_files), 1142 statp->f_files, 1143 mp->m_maxicount); 1144 1145 /* If sb_icount overshot maxicount, report actual allocation */ 1146 statp->f_files = max_t(typeof(statp->f_files), 1147 statp->f_files, 1148 sbp->sb_icount); 1149 1150 /* make sure statp->f_ffree does not underflow */ 1151 ffree = statp->f_files - (icount - ifree); 1152 statp->f_ffree = max_t(int64_t, ffree, 0); 1153 1154 1155 if ((ip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && 1156 ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) == 1157 (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD)) 1158 xfs_qm_statvfs(ip, statp); 1159 1160 if (XFS_IS_REALTIME_MOUNT(mp) && 1161 (ip->i_d.di_flags & (XFS_DIFLAG_RTINHERIT | XFS_DIFLAG_REALTIME))) { 1162 statp->f_blocks = sbp->sb_rblocks; 1163 statp->f_bavail = statp->f_bfree = 1164 sbp->sb_frextents * sbp->sb_rextsize; 1165 } 1166 1167 return 0; 1168 } 1169 1170 STATIC void 1171 xfs_save_resvblks(struct xfs_mount *mp) 1172 { 1173 uint64_t resblks = 0; 1174 1175 mp->m_resblks_save = mp->m_resblks; 1176 xfs_reserve_blocks(mp, &resblks, NULL); 1177 } 1178 1179 STATIC void 1180 xfs_restore_resvblks(struct xfs_mount *mp) 1181 { 1182 uint64_t resblks; 1183 1184 if (mp->m_resblks_save) { 1185 resblks = mp->m_resblks_save; 1186 mp->m_resblks_save = 0; 1187 } else 1188 resblks = xfs_default_resblks(mp); 1189 1190 xfs_reserve_blocks(mp, &resblks, NULL); 1191 } 1192 1193 /* 1194 * Trigger writeback of all the dirty metadata in the file system. 1195 * 1196 * This ensures that the metadata is written to their location on disk rather 1197 * than just existing in transactions in the log. This means after a quiesce 1198 * there is no log replay required to write the inodes to disk - this is the 1199 * primary difference between a sync and a quiesce. 1200 * 1201 * Note: xfs_log_quiesce() stops background log work - the callers must ensure 1202 * it is started again when appropriate. 1203 */ 1204 void 1205 xfs_quiesce_attr( 1206 struct xfs_mount *mp) 1207 { 1208 int error = 0; 1209 1210 /* wait for all modifications to complete */ 1211 while (atomic_read(&mp->m_active_trans) > 0) 1212 delay(100); 1213 1214 /* force the log to unpin objects from the now complete transactions */ 1215 xfs_log_force(mp, XFS_LOG_SYNC); 1216 1217 /* reclaim inodes to do any IO before the freeze completes */ 1218 xfs_reclaim_inodes(mp, 0); 1219 xfs_reclaim_inodes(mp, SYNC_WAIT); 1220 1221 /* Push the superblock and write an unmount record */ 1222 error = xfs_log_sbcount(mp); 1223 if (error) 1224 xfs_warn(mp, "xfs_attr_quiesce: failed to log sb changes. " 1225 "Frozen image may not be consistent."); 1226 /* 1227 * Just warn here till VFS can correctly support 1228 * read-only remount without racing. 1229 */ 1230 WARN_ON(atomic_read(&mp->m_active_trans) != 0); 1231 1232 xfs_log_quiesce(mp); 1233 } 1234 1235 STATIC int 1236 xfs_test_remount_options( 1237 struct super_block *sb, 1238 char *options) 1239 { 1240 int error = 0; 1241 struct xfs_mount *tmp_mp; 1242 1243 tmp_mp = kmem_zalloc(sizeof(*tmp_mp), KM_MAYFAIL); 1244 if (!tmp_mp) 1245 return -ENOMEM; 1246 1247 tmp_mp->m_super = sb; 1248 error = xfs_parseargs(tmp_mp, options); 1249 xfs_free_fsname(tmp_mp); 1250 kmem_free(tmp_mp); 1251 1252 return error; 1253 } 1254 1255 STATIC int 1256 xfs_fs_remount( 1257 struct super_block *sb, 1258 int *flags, 1259 char *options) 1260 { 1261 struct xfs_mount *mp = XFS_M(sb); 1262 xfs_sb_t *sbp = &mp->m_sb; 1263 substring_t args[MAX_OPT_ARGS]; 1264 char *p; 1265 int error; 1266 1267 /* First, check for complete junk; i.e. invalid options */ 1268 error = xfs_test_remount_options(sb, options); 1269 if (error) 1270 return error; 1271 1272 sync_filesystem(sb); 1273 while ((p = strsep(&options, ",")) != NULL) { 1274 int token; 1275 1276 if (!*p) 1277 continue; 1278 1279 token = match_token(p, tokens, args); 1280 switch (token) { 1281 case Opt_barrier: 1282 xfs_warn(mp, "%s option is deprecated, ignoring.", p); 1283 mp->m_flags |= XFS_MOUNT_BARRIER; 1284 break; 1285 case Opt_nobarrier: 1286 xfs_warn(mp, "%s option is deprecated, ignoring.", p); 1287 mp->m_flags &= ~XFS_MOUNT_BARRIER; 1288 break; 1289 case Opt_inode64: 1290 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS; 1291 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount); 1292 break; 1293 case Opt_inode32: 1294 mp->m_flags |= XFS_MOUNT_SMALL_INUMS; 1295 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount); 1296 break; 1297 default: 1298 /* 1299 * Logically we would return an error here to prevent 1300 * users from believing they might have changed 1301 * mount options using remount which can't be changed. 1302 * 1303 * But unfortunately mount(8) adds all options from 1304 * mtab and fstab to the mount arguments in some cases 1305 * so we can't blindly reject options, but have to 1306 * check for each specified option if it actually 1307 * differs from the currently set option and only 1308 * reject it if that's the case. 1309 * 1310 * Until that is implemented we return success for 1311 * every remount request, and silently ignore all 1312 * options that we can't actually change. 1313 */ 1314 #if 0 1315 xfs_info(mp, 1316 "mount option \"%s\" not supported for remount", p); 1317 return -EINVAL; 1318 #else 1319 break; 1320 #endif 1321 } 1322 } 1323 1324 /* ro -> rw */ 1325 if ((mp->m_flags & XFS_MOUNT_RDONLY) && !(*flags & SB_RDONLY)) { 1326 if (mp->m_flags & XFS_MOUNT_NORECOVERY) { 1327 xfs_warn(mp, 1328 "ro->rw transition prohibited on norecovery mount"); 1329 return -EINVAL; 1330 } 1331 1332 if (XFS_SB_VERSION_NUM(sbp) == XFS_SB_VERSION_5 && 1333 xfs_sb_has_ro_compat_feature(sbp, 1334 XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) { 1335 xfs_warn(mp, 1336 "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem", 1337 (sbp->sb_features_ro_compat & 1338 XFS_SB_FEAT_RO_COMPAT_UNKNOWN)); 1339 return -EINVAL; 1340 } 1341 1342 mp->m_flags &= ~XFS_MOUNT_RDONLY; 1343 1344 /* 1345 * If this is the first remount to writeable state we 1346 * might have some superblock changes to update. 1347 */ 1348 if (mp->m_update_sb) { 1349 error = xfs_sync_sb(mp, false); 1350 if (error) { 1351 xfs_warn(mp, "failed to write sb changes"); 1352 return error; 1353 } 1354 mp->m_update_sb = false; 1355 } 1356 1357 /* 1358 * Fill out the reserve pool if it is empty. Use the stashed 1359 * value if it is non-zero, otherwise go with the default. 1360 */ 1361 xfs_restore_resvblks(mp); 1362 xfs_log_work_queue(mp); 1363 1364 /* Recover any CoW blocks that never got remapped. */ 1365 error = xfs_reflink_recover_cow(mp); 1366 if (error) { 1367 xfs_err(mp, 1368 "Error %d recovering leftover CoW allocations.", error); 1369 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); 1370 return error; 1371 } 1372 xfs_icache_enable_reclaim(mp); 1373 1374 /* Create the per-AG metadata reservation pool .*/ 1375 error = xfs_fs_reserve_ag_blocks(mp); 1376 if (error && error != -ENOSPC) 1377 return error; 1378 } 1379 1380 /* rw -> ro */ 1381 if (!(mp->m_flags & XFS_MOUNT_RDONLY) && (*flags & SB_RDONLY)) { 1382 /* 1383 * Cancel background eofb scanning so it cannot race with the 1384 * final log force+buftarg wait and deadlock the remount. 1385 */ 1386 xfs_icache_disable_reclaim(mp); 1387 1388 /* Get rid of any leftover CoW reservations... */ 1389 error = xfs_icache_free_cowblocks(mp, NULL); 1390 if (error) { 1391 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); 1392 return error; 1393 } 1394 1395 /* Free the per-AG metadata reservation pool. */ 1396 error = xfs_fs_unreserve_ag_blocks(mp); 1397 if (error) { 1398 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); 1399 return error; 1400 } 1401 1402 /* 1403 * Before we sync the metadata, we need to free up the reserve 1404 * block pool so that the used block count in the superblock on 1405 * disk is correct at the end of the remount. Stash the current 1406 * reserve pool size so that if we get remounted rw, we can 1407 * return it to the same size. 1408 */ 1409 xfs_save_resvblks(mp); 1410 1411 xfs_quiesce_attr(mp); 1412 mp->m_flags |= XFS_MOUNT_RDONLY; 1413 } 1414 1415 return 0; 1416 } 1417 1418 /* 1419 * Second stage of a freeze. The data is already frozen so we only 1420 * need to take care of the metadata. Once that's done sync the superblock 1421 * to the log to dirty it in case of a crash while frozen. This ensures that we 1422 * will recover the unlinked inode lists on the next mount. 1423 */ 1424 STATIC int 1425 xfs_fs_freeze( 1426 struct super_block *sb) 1427 { 1428 struct xfs_mount *mp = XFS_M(sb); 1429 1430 xfs_icache_disable_reclaim(mp); 1431 xfs_save_resvblks(mp); 1432 xfs_quiesce_attr(mp); 1433 return xfs_sync_sb(mp, true); 1434 } 1435 1436 STATIC int 1437 xfs_fs_unfreeze( 1438 struct super_block *sb) 1439 { 1440 struct xfs_mount *mp = XFS_M(sb); 1441 1442 xfs_restore_resvblks(mp); 1443 xfs_log_work_queue(mp); 1444 xfs_icache_enable_reclaim(mp); 1445 return 0; 1446 } 1447 1448 STATIC int 1449 xfs_fs_show_options( 1450 struct seq_file *m, 1451 struct dentry *root) 1452 { 1453 return xfs_showargs(XFS_M(root->d_sb), m); 1454 } 1455 1456 /* 1457 * This function fills in xfs_mount_t fields based on mount args. 1458 * Note: the superblock _has_ now been read in. 1459 */ 1460 STATIC int 1461 xfs_finish_flags( 1462 struct xfs_mount *mp) 1463 { 1464 int ronly = (mp->m_flags & XFS_MOUNT_RDONLY); 1465 1466 /* Fail a mount where the logbuf is smaller than the log stripe */ 1467 if (xfs_sb_version_haslogv2(&mp->m_sb)) { 1468 if (mp->m_logbsize <= 0 && 1469 mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) { 1470 mp->m_logbsize = mp->m_sb.sb_logsunit; 1471 } else if (mp->m_logbsize > 0 && 1472 mp->m_logbsize < mp->m_sb.sb_logsunit) { 1473 xfs_warn(mp, 1474 "logbuf size must be greater than or equal to log stripe size"); 1475 return -EINVAL; 1476 } 1477 } else { 1478 /* Fail a mount if the logbuf is larger than 32K */ 1479 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) { 1480 xfs_warn(mp, 1481 "logbuf size for version 1 logs must be 16K or 32K"); 1482 return -EINVAL; 1483 } 1484 } 1485 1486 /* 1487 * V5 filesystems always use attr2 format for attributes. 1488 */ 1489 if (xfs_sb_version_hascrc(&mp->m_sb) && 1490 (mp->m_flags & XFS_MOUNT_NOATTR2)) { 1491 xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. " 1492 "attr2 is always enabled for V5 filesystems."); 1493 return -EINVAL; 1494 } 1495 1496 /* 1497 * mkfs'ed attr2 will turn on attr2 mount unless explicitly 1498 * told by noattr2 to turn it off 1499 */ 1500 if (xfs_sb_version_hasattr2(&mp->m_sb) && 1501 !(mp->m_flags & XFS_MOUNT_NOATTR2)) 1502 mp->m_flags |= XFS_MOUNT_ATTR2; 1503 1504 /* 1505 * prohibit r/w mounts of read-only filesystems 1506 */ 1507 if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) { 1508 xfs_warn(mp, 1509 "cannot mount a read-only filesystem as read-write"); 1510 return -EROFS; 1511 } 1512 1513 if ((mp->m_qflags & (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE)) && 1514 (mp->m_qflags & (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE)) && 1515 !xfs_sb_version_has_pquotino(&mp->m_sb)) { 1516 xfs_warn(mp, 1517 "Super block does not support project and group quota together"); 1518 return -EINVAL; 1519 } 1520 1521 return 0; 1522 } 1523 1524 static int 1525 xfs_init_percpu_counters( 1526 struct xfs_mount *mp) 1527 { 1528 int error; 1529 1530 error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL); 1531 if (error) 1532 return -ENOMEM; 1533 1534 error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL); 1535 if (error) 1536 goto free_icount; 1537 1538 error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL); 1539 if (error) 1540 goto free_ifree; 1541 1542 return 0; 1543 1544 free_ifree: 1545 percpu_counter_destroy(&mp->m_ifree); 1546 free_icount: 1547 percpu_counter_destroy(&mp->m_icount); 1548 return -ENOMEM; 1549 } 1550 1551 void 1552 xfs_reinit_percpu_counters( 1553 struct xfs_mount *mp) 1554 { 1555 percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount); 1556 percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree); 1557 percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks); 1558 } 1559 1560 static void 1561 xfs_destroy_percpu_counters( 1562 struct xfs_mount *mp) 1563 { 1564 percpu_counter_destroy(&mp->m_icount); 1565 percpu_counter_destroy(&mp->m_ifree); 1566 percpu_counter_destroy(&mp->m_fdblocks); 1567 } 1568 1569 static struct xfs_mount * 1570 xfs_mount_alloc( 1571 struct super_block *sb) 1572 { 1573 struct xfs_mount *mp; 1574 1575 mp = kzalloc(sizeof(struct xfs_mount), GFP_KERNEL); 1576 if (!mp) 1577 return NULL; 1578 1579 mp->m_super = sb; 1580 spin_lock_init(&mp->m_sb_lock); 1581 spin_lock_init(&mp->m_agirotor_lock); 1582 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC); 1583 spin_lock_init(&mp->m_perag_lock); 1584 mutex_init(&mp->m_growlock); 1585 atomic_set(&mp->m_active_trans, 0); 1586 INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker); 1587 INIT_DELAYED_WORK(&mp->m_eofblocks_work, xfs_eofblocks_worker); 1588 INIT_DELAYED_WORK(&mp->m_cowblocks_work, xfs_cowblocks_worker); 1589 mp->m_kobj.kobject.kset = xfs_kset; 1590 return mp; 1591 } 1592 1593 1594 STATIC int 1595 xfs_fs_fill_super( 1596 struct super_block *sb, 1597 void *data, 1598 int silent) 1599 { 1600 struct inode *root; 1601 struct xfs_mount *mp = NULL; 1602 int flags = 0, error = -ENOMEM; 1603 1604 /* 1605 * allocate mp and do all low-level struct initializations before we 1606 * attach it to the super 1607 */ 1608 mp = xfs_mount_alloc(sb); 1609 if (!mp) 1610 goto out; 1611 sb->s_fs_info = mp; 1612 1613 error = xfs_parseargs(mp, (char *)data); 1614 if (error) 1615 goto out_free_fsname; 1616 1617 sb_min_blocksize(sb, BBSIZE); 1618 sb->s_xattr = xfs_xattr_handlers; 1619 sb->s_export_op = &xfs_export_operations; 1620 #ifdef CONFIG_XFS_QUOTA 1621 sb->s_qcop = &xfs_quotactl_operations; 1622 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ; 1623 #endif 1624 sb->s_op = &xfs_super_operations; 1625 1626 /* 1627 * Delay mount work if the debug hook is set. This is debug 1628 * instrumention to coordinate simulation of xfs mount failures with 1629 * VFS superblock operations 1630 */ 1631 if (xfs_globals.mount_delay) { 1632 xfs_notice(mp, "Delaying mount for %d seconds.", 1633 xfs_globals.mount_delay); 1634 msleep(xfs_globals.mount_delay * 1000); 1635 } 1636 1637 if (silent) 1638 flags |= XFS_MFSI_QUIET; 1639 1640 error = xfs_open_devices(mp); 1641 if (error) 1642 goto out_free_fsname; 1643 1644 error = xfs_init_mount_workqueues(mp); 1645 if (error) 1646 goto out_close_devices; 1647 1648 error = xfs_init_percpu_counters(mp); 1649 if (error) 1650 goto out_destroy_workqueues; 1651 1652 /* Allocate stats memory before we do operations that might use it */ 1653 mp->m_stats.xs_stats = alloc_percpu(struct xfsstats); 1654 if (!mp->m_stats.xs_stats) { 1655 error = -ENOMEM; 1656 goto out_destroy_counters; 1657 } 1658 1659 error = xfs_readsb(mp, flags); 1660 if (error) 1661 goto out_free_stats; 1662 1663 error = xfs_finish_flags(mp); 1664 if (error) 1665 goto out_free_sb; 1666 1667 error = xfs_setup_devices(mp); 1668 if (error) 1669 goto out_free_sb; 1670 1671 error = xfs_filestream_mount(mp); 1672 if (error) 1673 goto out_free_sb; 1674 1675 /* 1676 * we must configure the block size in the superblock before we run the 1677 * full mount process as the mount process can lookup and cache inodes. 1678 */ 1679 sb->s_magic = XFS_SB_MAGIC; 1680 sb->s_blocksize = mp->m_sb.sb_blocksize; 1681 sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1; 1682 sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits); 1683 sb->s_max_links = XFS_MAXLINK; 1684 sb->s_time_gran = 1; 1685 set_posix_acl_flag(sb); 1686 1687 /* version 5 superblocks support inode version counters. */ 1688 if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5) 1689 sb->s_flags |= SB_I_VERSION; 1690 1691 if (mp->m_flags & XFS_MOUNT_DAX) { 1692 bool rtdev_is_dax = false, datadev_is_dax; 1693 1694 xfs_warn(mp, 1695 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk"); 1696 1697 datadev_is_dax = bdev_dax_supported(mp->m_ddev_targp->bt_bdev, 1698 sb->s_blocksize); 1699 if (mp->m_rtdev_targp) 1700 rtdev_is_dax = bdev_dax_supported( 1701 mp->m_rtdev_targp->bt_bdev, sb->s_blocksize); 1702 if (!rtdev_is_dax && !datadev_is_dax) { 1703 xfs_alert(mp, 1704 "DAX unsupported by block device. Turning off DAX."); 1705 mp->m_flags &= ~XFS_MOUNT_DAX; 1706 } 1707 if (xfs_sb_version_hasreflink(&mp->m_sb)) { 1708 xfs_alert(mp, 1709 "DAX and reflink cannot be used together!"); 1710 error = -EINVAL; 1711 goto out_filestream_unmount; 1712 } 1713 } 1714 1715 if (mp->m_flags & XFS_MOUNT_DISCARD) { 1716 struct request_queue *q = bdev_get_queue(sb->s_bdev); 1717 1718 if (!blk_queue_discard(q)) { 1719 xfs_warn(mp, "mounting with \"discard\" option, but " 1720 "the device does not support discard"); 1721 mp->m_flags &= ~XFS_MOUNT_DISCARD; 1722 } 1723 } 1724 1725 if (xfs_sb_version_hasreflink(&mp->m_sb) && mp->m_sb.sb_rblocks) { 1726 xfs_alert(mp, 1727 "reflink not compatible with realtime device!"); 1728 error = -EINVAL; 1729 goto out_filestream_unmount; 1730 } 1731 1732 if (xfs_sb_version_hasrmapbt(&mp->m_sb) && mp->m_sb.sb_rblocks) { 1733 xfs_alert(mp, 1734 "reverse mapping btree not compatible with realtime device!"); 1735 error = -EINVAL; 1736 goto out_filestream_unmount; 1737 } 1738 1739 error = xfs_mountfs(mp); 1740 if (error) 1741 goto out_filestream_unmount; 1742 1743 root = igrab(VFS_I(mp->m_rootip)); 1744 if (!root) { 1745 error = -ENOENT; 1746 goto out_unmount; 1747 } 1748 sb->s_root = d_make_root(root); 1749 if (!sb->s_root) { 1750 error = -ENOMEM; 1751 goto out_unmount; 1752 } 1753 1754 return 0; 1755 1756 out_filestream_unmount: 1757 xfs_filestream_unmount(mp); 1758 out_free_sb: 1759 xfs_freesb(mp); 1760 out_free_stats: 1761 free_percpu(mp->m_stats.xs_stats); 1762 out_destroy_counters: 1763 xfs_destroy_percpu_counters(mp); 1764 out_destroy_workqueues: 1765 xfs_destroy_mount_workqueues(mp); 1766 out_close_devices: 1767 xfs_close_devices(mp); 1768 out_free_fsname: 1769 sb->s_fs_info = NULL; 1770 xfs_free_fsname(mp); 1771 kfree(mp); 1772 out: 1773 return error; 1774 1775 out_unmount: 1776 xfs_filestream_unmount(mp); 1777 xfs_unmountfs(mp); 1778 goto out_free_sb; 1779 } 1780 1781 STATIC void 1782 xfs_fs_put_super( 1783 struct super_block *sb) 1784 { 1785 struct xfs_mount *mp = XFS_M(sb); 1786 1787 /* if ->fill_super failed, we have no mount to tear down */ 1788 if (!sb->s_fs_info) 1789 return; 1790 1791 xfs_notice(mp, "Unmounting Filesystem"); 1792 xfs_filestream_unmount(mp); 1793 xfs_unmountfs(mp); 1794 1795 xfs_freesb(mp); 1796 free_percpu(mp->m_stats.xs_stats); 1797 xfs_destroy_percpu_counters(mp); 1798 xfs_destroy_mount_workqueues(mp); 1799 xfs_close_devices(mp); 1800 1801 sb->s_fs_info = NULL; 1802 xfs_free_fsname(mp); 1803 kfree(mp); 1804 } 1805 1806 STATIC struct dentry * 1807 xfs_fs_mount( 1808 struct file_system_type *fs_type, 1809 int flags, 1810 const char *dev_name, 1811 void *data) 1812 { 1813 return mount_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super); 1814 } 1815 1816 static long 1817 xfs_fs_nr_cached_objects( 1818 struct super_block *sb, 1819 struct shrink_control *sc) 1820 { 1821 /* Paranoia: catch incorrect calls during mount setup or teardown */ 1822 if (WARN_ON_ONCE(!sb->s_fs_info)) 1823 return 0; 1824 return xfs_reclaim_inodes_count(XFS_M(sb)); 1825 } 1826 1827 static long 1828 xfs_fs_free_cached_objects( 1829 struct super_block *sb, 1830 struct shrink_control *sc) 1831 { 1832 return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan); 1833 } 1834 1835 static const struct super_operations xfs_super_operations = { 1836 .alloc_inode = xfs_fs_alloc_inode, 1837 .destroy_inode = xfs_fs_destroy_inode, 1838 .dirty_inode = xfs_fs_dirty_inode, 1839 .drop_inode = xfs_fs_drop_inode, 1840 .put_super = xfs_fs_put_super, 1841 .sync_fs = xfs_fs_sync_fs, 1842 .freeze_fs = xfs_fs_freeze, 1843 .unfreeze_fs = xfs_fs_unfreeze, 1844 .statfs = xfs_fs_statfs, 1845 .remount_fs = xfs_fs_remount, 1846 .show_options = xfs_fs_show_options, 1847 .nr_cached_objects = xfs_fs_nr_cached_objects, 1848 .free_cached_objects = xfs_fs_free_cached_objects, 1849 }; 1850 1851 static struct file_system_type xfs_fs_type = { 1852 .owner = THIS_MODULE, 1853 .name = "xfs", 1854 .mount = xfs_fs_mount, 1855 .kill_sb = kill_block_super, 1856 .fs_flags = FS_REQUIRES_DEV, 1857 }; 1858 MODULE_ALIAS_FS("xfs"); 1859 1860 STATIC int __init 1861 xfs_init_zones(void) 1862 { 1863 if (bioset_init(&xfs_ioend_bioset, 4 * MAX_BUF_PER_PAGE, 1864 offsetof(struct xfs_ioend, io_inline_bio), 1865 BIOSET_NEED_BVECS)) 1866 goto out; 1867 1868 xfs_log_ticket_zone = kmem_zone_init(sizeof(xlog_ticket_t), 1869 "xfs_log_ticket"); 1870 if (!xfs_log_ticket_zone) 1871 goto out_free_ioend_bioset; 1872 1873 xfs_bmap_free_item_zone = kmem_zone_init( 1874 sizeof(struct xfs_extent_free_item), 1875 "xfs_bmap_free_item"); 1876 if (!xfs_bmap_free_item_zone) 1877 goto out_destroy_log_ticket_zone; 1878 1879 xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t), 1880 "xfs_btree_cur"); 1881 if (!xfs_btree_cur_zone) 1882 goto out_destroy_bmap_free_item_zone; 1883 1884 xfs_da_state_zone = kmem_zone_init(sizeof(xfs_da_state_t), 1885 "xfs_da_state"); 1886 if (!xfs_da_state_zone) 1887 goto out_destroy_btree_cur_zone; 1888 1889 xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork"); 1890 if (!xfs_ifork_zone) 1891 goto out_destroy_da_state_zone; 1892 1893 xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans"); 1894 if (!xfs_trans_zone) 1895 goto out_destroy_ifork_zone; 1896 1897 1898 /* 1899 * The size of the zone allocated buf log item is the maximum 1900 * size possible under XFS. This wastes a little bit of memory, 1901 * but it is much faster. 1902 */ 1903 xfs_buf_item_zone = kmem_zone_init(sizeof(struct xfs_buf_log_item), 1904 "xfs_buf_item"); 1905 if (!xfs_buf_item_zone) 1906 goto out_destroy_trans_zone; 1907 1908 xfs_efd_zone = kmem_zone_init((sizeof(xfs_efd_log_item_t) + 1909 ((XFS_EFD_MAX_FAST_EXTENTS - 1) * 1910 sizeof(xfs_extent_t))), "xfs_efd_item"); 1911 if (!xfs_efd_zone) 1912 goto out_destroy_buf_item_zone; 1913 1914 xfs_efi_zone = kmem_zone_init((sizeof(xfs_efi_log_item_t) + 1915 ((XFS_EFI_MAX_FAST_EXTENTS - 1) * 1916 sizeof(xfs_extent_t))), "xfs_efi_item"); 1917 if (!xfs_efi_zone) 1918 goto out_destroy_efd_zone; 1919 1920 xfs_inode_zone = 1921 kmem_zone_init_flags(sizeof(xfs_inode_t), "xfs_inode", 1922 KM_ZONE_HWALIGN | KM_ZONE_RECLAIM | KM_ZONE_SPREAD | 1923 KM_ZONE_ACCOUNT, xfs_fs_inode_init_once); 1924 if (!xfs_inode_zone) 1925 goto out_destroy_efi_zone; 1926 1927 xfs_ili_zone = 1928 kmem_zone_init_flags(sizeof(xfs_inode_log_item_t), "xfs_ili", 1929 KM_ZONE_SPREAD, NULL); 1930 if (!xfs_ili_zone) 1931 goto out_destroy_inode_zone; 1932 xfs_icreate_zone = kmem_zone_init(sizeof(struct xfs_icreate_item), 1933 "xfs_icr"); 1934 if (!xfs_icreate_zone) 1935 goto out_destroy_ili_zone; 1936 1937 xfs_rud_zone = kmem_zone_init(sizeof(struct xfs_rud_log_item), 1938 "xfs_rud_item"); 1939 if (!xfs_rud_zone) 1940 goto out_destroy_icreate_zone; 1941 1942 xfs_rui_zone = kmem_zone_init( 1943 xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS), 1944 "xfs_rui_item"); 1945 if (!xfs_rui_zone) 1946 goto out_destroy_rud_zone; 1947 1948 xfs_cud_zone = kmem_zone_init(sizeof(struct xfs_cud_log_item), 1949 "xfs_cud_item"); 1950 if (!xfs_cud_zone) 1951 goto out_destroy_rui_zone; 1952 1953 xfs_cui_zone = kmem_zone_init( 1954 xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS), 1955 "xfs_cui_item"); 1956 if (!xfs_cui_zone) 1957 goto out_destroy_cud_zone; 1958 1959 xfs_bud_zone = kmem_zone_init(sizeof(struct xfs_bud_log_item), 1960 "xfs_bud_item"); 1961 if (!xfs_bud_zone) 1962 goto out_destroy_cui_zone; 1963 1964 xfs_bui_zone = kmem_zone_init( 1965 xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS), 1966 "xfs_bui_item"); 1967 if (!xfs_bui_zone) 1968 goto out_destroy_bud_zone; 1969 1970 return 0; 1971 1972 out_destroy_bud_zone: 1973 kmem_zone_destroy(xfs_bud_zone); 1974 out_destroy_cui_zone: 1975 kmem_zone_destroy(xfs_cui_zone); 1976 out_destroy_cud_zone: 1977 kmem_zone_destroy(xfs_cud_zone); 1978 out_destroy_rui_zone: 1979 kmem_zone_destroy(xfs_rui_zone); 1980 out_destroy_rud_zone: 1981 kmem_zone_destroy(xfs_rud_zone); 1982 out_destroy_icreate_zone: 1983 kmem_zone_destroy(xfs_icreate_zone); 1984 out_destroy_ili_zone: 1985 kmem_zone_destroy(xfs_ili_zone); 1986 out_destroy_inode_zone: 1987 kmem_zone_destroy(xfs_inode_zone); 1988 out_destroy_efi_zone: 1989 kmem_zone_destroy(xfs_efi_zone); 1990 out_destroy_efd_zone: 1991 kmem_zone_destroy(xfs_efd_zone); 1992 out_destroy_buf_item_zone: 1993 kmem_zone_destroy(xfs_buf_item_zone); 1994 out_destroy_trans_zone: 1995 kmem_zone_destroy(xfs_trans_zone); 1996 out_destroy_ifork_zone: 1997 kmem_zone_destroy(xfs_ifork_zone); 1998 out_destroy_da_state_zone: 1999 kmem_zone_destroy(xfs_da_state_zone); 2000 out_destroy_btree_cur_zone: 2001 kmem_zone_destroy(xfs_btree_cur_zone); 2002 out_destroy_bmap_free_item_zone: 2003 kmem_zone_destroy(xfs_bmap_free_item_zone); 2004 out_destroy_log_ticket_zone: 2005 kmem_zone_destroy(xfs_log_ticket_zone); 2006 out_free_ioend_bioset: 2007 bioset_exit(&xfs_ioend_bioset); 2008 out: 2009 return -ENOMEM; 2010 } 2011 2012 STATIC void 2013 xfs_destroy_zones(void) 2014 { 2015 /* 2016 * Make sure all delayed rcu free are flushed before we 2017 * destroy caches. 2018 */ 2019 rcu_barrier(); 2020 kmem_zone_destroy(xfs_bui_zone); 2021 kmem_zone_destroy(xfs_bud_zone); 2022 kmem_zone_destroy(xfs_cui_zone); 2023 kmem_zone_destroy(xfs_cud_zone); 2024 kmem_zone_destroy(xfs_rui_zone); 2025 kmem_zone_destroy(xfs_rud_zone); 2026 kmem_zone_destroy(xfs_icreate_zone); 2027 kmem_zone_destroy(xfs_ili_zone); 2028 kmem_zone_destroy(xfs_inode_zone); 2029 kmem_zone_destroy(xfs_efi_zone); 2030 kmem_zone_destroy(xfs_efd_zone); 2031 kmem_zone_destroy(xfs_buf_item_zone); 2032 kmem_zone_destroy(xfs_trans_zone); 2033 kmem_zone_destroy(xfs_ifork_zone); 2034 kmem_zone_destroy(xfs_da_state_zone); 2035 kmem_zone_destroy(xfs_btree_cur_zone); 2036 kmem_zone_destroy(xfs_bmap_free_item_zone); 2037 kmem_zone_destroy(xfs_log_ticket_zone); 2038 bioset_exit(&xfs_ioend_bioset); 2039 } 2040 2041 STATIC int __init 2042 xfs_init_workqueues(void) 2043 { 2044 /* 2045 * The allocation workqueue can be used in memory reclaim situations 2046 * (writepage path), and parallelism is only limited by the number of 2047 * AGs in all the filesystems mounted. Hence use the default large 2048 * max_active value for this workqueue. 2049 */ 2050 xfs_alloc_wq = alloc_workqueue("xfsalloc", 2051 WQ_MEM_RECLAIM|WQ_FREEZABLE, 0); 2052 if (!xfs_alloc_wq) 2053 return -ENOMEM; 2054 2055 xfs_discard_wq = alloc_workqueue("xfsdiscard", WQ_UNBOUND, 0); 2056 if (!xfs_discard_wq) 2057 goto out_free_alloc_wq; 2058 2059 return 0; 2060 out_free_alloc_wq: 2061 destroy_workqueue(xfs_alloc_wq); 2062 return -ENOMEM; 2063 } 2064 2065 STATIC void 2066 xfs_destroy_workqueues(void) 2067 { 2068 destroy_workqueue(xfs_discard_wq); 2069 destroy_workqueue(xfs_alloc_wq); 2070 } 2071 2072 STATIC int __init 2073 init_xfs_fs(void) 2074 { 2075 int error; 2076 2077 xfs_check_ondisk_structs(); 2078 2079 printk(KERN_INFO XFS_VERSION_STRING " with " 2080 XFS_BUILD_OPTIONS " enabled\n"); 2081 2082 xfs_extent_free_init_defer_op(); 2083 xfs_rmap_update_init_defer_op(); 2084 xfs_refcount_update_init_defer_op(); 2085 xfs_bmap_update_init_defer_op(); 2086 2087 xfs_dir_startup(); 2088 2089 error = xfs_init_zones(); 2090 if (error) 2091 goto out; 2092 2093 error = xfs_init_workqueues(); 2094 if (error) 2095 goto out_destroy_zones; 2096 2097 error = xfs_mru_cache_init(); 2098 if (error) 2099 goto out_destroy_wq; 2100 2101 error = xfs_buf_init(); 2102 if (error) 2103 goto out_mru_cache_uninit; 2104 2105 error = xfs_init_procfs(); 2106 if (error) 2107 goto out_buf_terminate; 2108 2109 error = xfs_sysctl_register(); 2110 if (error) 2111 goto out_cleanup_procfs; 2112 2113 xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj); 2114 if (!xfs_kset) { 2115 error = -ENOMEM; 2116 goto out_sysctl_unregister; 2117 } 2118 2119 xfsstats.xs_kobj.kobject.kset = xfs_kset; 2120 2121 xfsstats.xs_stats = alloc_percpu(struct xfsstats); 2122 if (!xfsstats.xs_stats) { 2123 error = -ENOMEM; 2124 goto out_kset_unregister; 2125 } 2126 2127 error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL, 2128 "stats"); 2129 if (error) 2130 goto out_free_stats; 2131 2132 #ifdef DEBUG 2133 xfs_dbg_kobj.kobject.kset = xfs_kset; 2134 error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug"); 2135 if (error) 2136 goto out_remove_stats_kobj; 2137 #endif 2138 2139 error = xfs_qm_init(); 2140 if (error) 2141 goto out_remove_dbg_kobj; 2142 2143 error = register_filesystem(&xfs_fs_type); 2144 if (error) 2145 goto out_qm_exit; 2146 return 0; 2147 2148 out_qm_exit: 2149 xfs_qm_exit(); 2150 out_remove_dbg_kobj: 2151 #ifdef DEBUG 2152 xfs_sysfs_del(&xfs_dbg_kobj); 2153 out_remove_stats_kobj: 2154 #endif 2155 xfs_sysfs_del(&xfsstats.xs_kobj); 2156 out_free_stats: 2157 free_percpu(xfsstats.xs_stats); 2158 out_kset_unregister: 2159 kset_unregister(xfs_kset); 2160 out_sysctl_unregister: 2161 xfs_sysctl_unregister(); 2162 out_cleanup_procfs: 2163 xfs_cleanup_procfs(); 2164 out_buf_terminate: 2165 xfs_buf_terminate(); 2166 out_mru_cache_uninit: 2167 xfs_mru_cache_uninit(); 2168 out_destroy_wq: 2169 xfs_destroy_workqueues(); 2170 out_destroy_zones: 2171 xfs_destroy_zones(); 2172 out: 2173 return error; 2174 } 2175 2176 STATIC void __exit 2177 exit_xfs_fs(void) 2178 { 2179 xfs_qm_exit(); 2180 unregister_filesystem(&xfs_fs_type); 2181 #ifdef DEBUG 2182 xfs_sysfs_del(&xfs_dbg_kobj); 2183 #endif 2184 xfs_sysfs_del(&xfsstats.xs_kobj); 2185 free_percpu(xfsstats.xs_stats); 2186 kset_unregister(xfs_kset); 2187 xfs_sysctl_unregister(); 2188 xfs_cleanup_procfs(); 2189 xfs_buf_terminate(); 2190 xfs_mru_cache_uninit(); 2191 xfs_destroy_workqueues(); 2192 xfs_destroy_zones(); 2193 xfs_uuid_table_free(); 2194 } 2195 2196 module_init(init_xfs_fs); 2197 module_exit(exit_xfs_fs); 2198 2199 MODULE_AUTHOR("Silicon Graphics, Inc."); 2200 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled"); 2201 MODULE_LICENSE("GPL"); 2202