1 /* 2 * Copyright (c) 2000-2006 Silicon Graphics, Inc. 3 * All Rights Reserved. 4 * 5 * This program is free software; you can redistribute it and/or 6 * modify it under the terms of the GNU General Public License as 7 * published by the Free Software Foundation. 8 * 9 * This program is distributed in the hope that it would be useful, 10 * but WITHOUT ANY WARRANTY; without even the implied warranty of 11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 12 * GNU General Public License for more details. 13 * 14 * You should have received a copy of the GNU General Public License 15 * along with this program; if not, write the Free Software Foundation, 16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA 17 */ 18 19 #include "xfs.h" 20 #include "xfs_bit.h" 21 #include "xfs_log.h" 22 #include "xfs_inum.h" 23 #include "xfs_trans.h" 24 #include "xfs_sb.h" 25 #include "xfs_ag.h" 26 #include "xfs_dir2.h" 27 #include "xfs_alloc.h" 28 #include "xfs_quota.h" 29 #include "xfs_mount.h" 30 #include "xfs_bmap_btree.h" 31 #include "xfs_alloc_btree.h" 32 #include "xfs_ialloc_btree.h" 33 #include "xfs_dinode.h" 34 #include "xfs_inode.h" 35 #include "xfs_btree.h" 36 #include "xfs_ialloc.h" 37 #include "xfs_bmap.h" 38 #include "xfs_rtalloc.h" 39 #include "xfs_error.h" 40 #include "xfs_itable.h" 41 #include "xfs_fsops.h" 42 #include "xfs_attr.h" 43 #include "xfs_buf_item.h" 44 #include "xfs_utils.h" 45 #include "xfs_vnodeops.h" 46 #include "xfs_log_priv.h" 47 #include "xfs_trans_priv.h" 48 #include "xfs_filestream.h" 49 #include "xfs_da_btree.h" 50 #include "xfs_extfree_item.h" 51 #include "xfs_mru_cache.h" 52 #include "xfs_inode_item.h" 53 #include "xfs_sync.h" 54 #include "xfs_trace.h" 55 56 #include <linux/namei.h> 57 #include <linux/init.h> 58 #include <linux/slab.h> 59 #include <linux/mount.h> 60 #include <linux/mempool.h> 61 #include <linux/writeback.h> 62 #include <linux/kthread.h> 63 #include <linux/freezer.h> 64 #include <linux/parser.h> 65 66 static const struct super_operations xfs_super_operations; 67 static kmem_zone_t *xfs_ioend_zone; 68 mempool_t *xfs_ioend_pool; 69 70 #define MNTOPT_LOGBUFS "logbufs" /* number of XFS log buffers */ 71 #define MNTOPT_LOGBSIZE "logbsize" /* size of XFS log buffers */ 72 #define MNTOPT_LOGDEV "logdev" /* log device */ 73 #define MNTOPT_RTDEV "rtdev" /* realtime I/O device */ 74 #define MNTOPT_BIOSIZE "biosize" /* log2 of preferred buffered io size */ 75 #define MNTOPT_WSYNC "wsync" /* safe-mode nfs compatible mount */ 76 #define MNTOPT_NOALIGN "noalign" /* turn off stripe alignment */ 77 #define MNTOPT_SWALLOC "swalloc" /* turn on stripe width allocation */ 78 #define MNTOPT_SUNIT "sunit" /* data volume stripe unit */ 79 #define MNTOPT_SWIDTH "swidth" /* data volume stripe width */ 80 #define MNTOPT_NOUUID "nouuid" /* ignore filesystem UUID */ 81 #define MNTOPT_MTPT "mtpt" /* filesystem mount point */ 82 #define MNTOPT_GRPID "grpid" /* group-ID from parent directory */ 83 #define MNTOPT_NOGRPID "nogrpid" /* group-ID from current process */ 84 #define MNTOPT_BSDGROUPS "bsdgroups" /* group-ID from parent directory */ 85 #define MNTOPT_SYSVGROUPS "sysvgroups" /* group-ID from current process */ 86 #define MNTOPT_ALLOCSIZE "allocsize" /* preferred allocation size */ 87 #define MNTOPT_NORECOVERY "norecovery" /* don't run XFS recovery */ 88 #define MNTOPT_BARRIER "barrier" /* use writer barriers for log write and 89 * unwritten extent conversion */ 90 #define MNTOPT_NOBARRIER "nobarrier" /* .. disable */ 91 #define MNTOPT_64BITINODE "inode64" /* inodes can be allocated anywhere */ 92 #define MNTOPT_IKEEP "ikeep" /* do not free empty inode clusters */ 93 #define MNTOPT_NOIKEEP "noikeep" /* free empty inode clusters */ 94 #define MNTOPT_LARGEIO "largeio" /* report large I/O sizes in stat() */ 95 #define MNTOPT_NOLARGEIO "nolargeio" /* do not report large I/O sizes 96 * in stat(). */ 97 #define MNTOPT_ATTR2 "attr2" /* do use attr2 attribute format */ 98 #define MNTOPT_NOATTR2 "noattr2" /* do not use attr2 attribute format */ 99 #define MNTOPT_FILESTREAM "filestreams" /* use filestreams allocator */ 100 #define MNTOPT_QUOTA "quota" /* disk quotas (user) */ 101 #define MNTOPT_NOQUOTA "noquota" /* no quotas */ 102 #define MNTOPT_USRQUOTA "usrquota" /* user quota enabled */ 103 #define MNTOPT_GRPQUOTA "grpquota" /* group quota enabled */ 104 #define MNTOPT_PRJQUOTA "prjquota" /* project quota enabled */ 105 #define MNTOPT_UQUOTA "uquota" /* user quota (IRIX variant) */ 106 #define MNTOPT_GQUOTA "gquota" /* group quota (IRIX variant) */ 107 #define MNTOPT_PQUOTA "pquota" /* project quota (IRIX variant) */ 108 #define MNTOPT_UQUOTANOENF "uqnoenforce"/* user quota limit enforcement */ 109 #define MNTOPT_GQUOTANOENF "gqnoenforce"/* group quota limit enforcement */ 110 #define MNTOPT_PQUOTANOENF "pqnoenforce"/* project quota limit enforcement */ 111 #define MNTOPT_QUOTANOENF "qnoenforce" /* same as uqnoenforce */ 112 #define MNTOPT_DELAYLOG "delaylog" /* Delayed logging enabled */ 113 #define MNTOPT_NODELAYLOG "nodelaylog" /* Delayed logging disabled */ 114 #define MNTOPT_DISCARD "discard" /* Discard unused blocks */ 115 #define MNTOPT_NODISCARD "nodiscard" /* Do not discard unused blocks */ 116 117 /* 118 * Table driven mount option parser. 119 * 120 * Currently only used for remount, but it will be used for mount 121 * in the future, too. 122 */ 123 enum { 124 Opt_barrier, Opt_nobarrier, Opt_err 125 }; 126 127 static const match_table_t tokens = { 128 {Opt_barrier, "barrier"}, 129 {Opt_nobarrier, "nobarrier"}, 130 {Opt_err, NULL} 131 }; 132 133 134 STATIC unsigned long 135 suffix_strtoul(char *s, char **endp, unsigned int base) 136 { 137 int last, shift_left_factor = 0; 138 char *value = s; 139 140 last = strlen(value) - 1; 141 if (value[last] == 'K' || value[last] == 'k') { 142 shift_left_factor = 10; 143 value[last] = '\0'; 144 } 145 if (value[last] == 'M' || value[last] == 'm') { 146 shift_left_factor = 20; 147 value[last] = '\0'; 148 } 149 if (value[last] == 'G' || value[last] == 'g') { 150 shift_left_factor = 30; 151 value[last] = '\0'; 152 } 153 154 return simple_strtoul((const char *)s, endp, base) << shift_left_factor; 155 } 156 157 /* 158 * This function fills in xfs_mount_t fields based on mount args. 159 * Note: the superblock has _not_ yet been read in. 160 * 161 * Note that this function leaks the various device name allocations on 162 * failure. The caller takes care of them. 163 */ 164 STATIC int 165 xfs_parseargs( 166 struct xfs_mount *mp, 167 char *options) 168 { 169 struct super_block *sb = mp->m_super; 170 char *this_char, *value, *eov; 171 int dsunit = 0; 172 int dswidth = 0; 173 int iosize = 0; 174 __uint8_t iosizelog = 0; 175 176 /* 177 * set up the mount name first so all the errors will refer to the 178 * correct device. 179 */ 180 mp->m_fsname = kstrndup(sb->s_id, MAXNAMELEN, GFP_KERNEL); 181 if (!mp->m_fsname) 182 return ENOMEM; 183 mp->m_fsname_len = strlen(mp->m_fsname) + 1; 184 185 /* 186 * Copy binary VFS mount flags we are interested in. 187 */ 188 if (sb->s_flags & MS_RDONLY) 189 mp->m_flags |= XFS_MOUNT_RDONLY; 190 if (sb->s_flags & MS_DIRSYNC) 191 mp->m_flags |= XFS_MOUNT_DIRSYNC; 192 if (sb->s_flags & MS_SYNCHRONOUS) 193 mp->m_flags |= XFS_MOUNT_WSYNC; 194 195 /* 196 * Set some default flags that could be cleared by the mount option 197 * parsing. 198 */ 199 mp->m_flags |= XFS_MOUNT_BARRIER; 200 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE; 201 mp->m_flags |= XFS_MOUNT_SMALL_INUMS; 202 mp->m_flags |= XFS_MOUNT_DELAYLOG; 203 204 /* 205 * These can be overridden by the mount option parsing. 206 */ 207 mp->m_logbufs = -1; 208 mp->m_logbsize = -1; 209 210 if (!options) 211 goto done; 212 213 while ((this_char = strsep(&options, ",")) != NULL) { 214 if (!*this_char) 215 continue; 216 if ((value = strchr(this_char, '=')) != NULL) 217 *value++ = 0; 218 219 if (!strcmp(this_char, MNTOPT_LOGBUFS)) { 220 if (!value || !*value) { 221 xfs_warn(mp, "%s option requires an argument", 222 this_char); 223 return EINVAL; 224 } 225 mp->m_logbufs = simple_strtoul(value, &eov, 10); 226 } else if (!strcmp(this_char, MNTOPT_LOGBSIZE)) { 227 if (!value || !*value) { 228 xfs_warn(mp, "%s option requires an argument", 229 this_char); 230 return EINVAL; 231 } 232 mp->m_logbsize = suffix_strtoul(value, &eov, 10); 233 } else if (!strcmp(this_char, MNTOPT_LOGDEV)) { 234 if (!value || !*value) { 235 xfs_warn(mp, "%s option requires an argument", 236 this_char); 237 return EINVAL; 238 } 239 mp->m_logname = kstrndup(value, MAXNAMELEN, GFP_KERNEL); 240 if (!mp->m_logname) 241 return ENOMEM; 242 } else if (!strcmp(this_char, MNTOPT_MTPT)) { 243 xfs_warn(mp, "%s option not allowed on this system", 244 this_char); 245 return EINVAL; 246 } else if (!strcmp(this_char, MNTOPT_RTDEV)) { 247 if (!value || !*value) { 248 xfs_warn(mp, "%s option requires an argument", 249 this_char); 250 return EINVAL; 251 } 252 mp->m_rtname = kstrndup(value, MAXNAMELEN, GFP_KERNEL); 253 if (!mp->m_rtname) 254 return ENOMEM; 255 } else if (!strcmp(this_char, MNTOPT_BIOSIZE)) { 256 if (!value || !*value) { 257 xfs_warn(mp, "%s option requires an argument", 258 this_char); 259 return EINVAL; 260 } 261 iosize = simple_strtoul(value, &eov, 10); 262 iosizelog = ffs(iosize) - 1; 263 } else if (!strcmp(this_char, MNTOPT_ALLOCSIZE)) { 264 if (!value || !*value) { 265 xfs_warn(mp, "%s option requires an argument", 266 this_char); 267 return EINVAL; 268 } 269 iosize = suffix_strtoul(value, &eov, 10); 270 iosizelog = ffs(iosize) - 1; 271 } else if (!strcmp(this_char, MNTOPT_GRPID) || 272 !strcmp(this_char, MNTOPT_BSDGROUPS)) { 273 mp->m_flags |= XFS_MOUNT_GRPID; 274 } else if (!strcmp(this_char, MNTOPT_NOGRPID) || 275 !strcmp(this_char, MNTOPT_SYSVGROUPS)) { 276 mp->m_flags &= ~XFS_MOUNT_GRPID; 277 } else if (!strcmp(this_char, MNTOPT_WSYNC)) { 278 mp->m_flags |= XFS_MOUNT_WSYNC; 279 } else if (!strcmp(this_char, MNTOPT_NORECOVERY)) { 280 mp->m_flags |= XFS_MOUNT_NORECOVERY; 281 } else if (!strcmp(this_char, MNTOPT_NOALIGN)) { 282 mp->m_flags |= XFS_MOUNT_NOALIGN; 283 } else if (!strcmp(this_char, MNTOPT_SWALLOC)) { 284 mp->m_flags |= XFS_MOUNT_SWALLOC; 285 } else if (!strcmp(this_char, MNTOPT_SUNIT)) { 286 if (!value || !*value) { 287 xfs_warn(mp, "%s option requires an argument", 288 this_char); 289 return EINVAL; 290 } 291 dsunit = simple_strtoul(value, &eov, 10); 292 } else if (!strcmp(this_char, MNTOPT_SWIDTH)) { 293 if (!value || !*value) { 294 xfs_warn(mp, "%s option requires an argument", 295 this_char); 296 return EINVAL; 297 } 298 dswidth = simple_strtoul(value, &eov, 10); 299 } else if (!strcmp(this_char, MNTOPT_64BITINODE)) { 300 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS; 301 #if !XFS_BIG_INUMS 302 xfs_warn(mp, "%s option not allowed on this system", 303 this_char); 304 return EINVAL; 305 #endif 306 } else if (!strcmp(this_char, MNTOPT_NOUUID)) { 307 mp->m_flags |= XFS_MOUNT_NOUUID; 308 } else if (!strcmp(this_char, MNTOPT_BARRIER)) { 309 mp->m_flags |= XFS_MOUNT_BARRIER; 310 } else if (!strcmp(this_char, MNTOPT_NOBARRIER)) { 311 mp->m_flags &= ~XFS_MOUNT_BARRIER; 312 } else if (!strcmp(this_char, MNTOPT_IKEEP)) { 313 mp->m_flags |= XFS_MOUNT_IKEEP; 314 } else if (!strcmp(this_char, MNTOPT_NOIKEEP)) { 315 mp->m_flags &= ~XFS_MOUNT_IKEEP; 316 } else if (!strcmp(this_char, MNTOPT_LARGEIO)) { 317 mp->m_flags &= ~XFS_MOUNT_COMPAT_IOSIZE; 318 } else if (!strcmp(this_char, MNTOPT_NOLARGEIO)) { 319 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE; 320 } else if (!strcmp(this_char, MNTOPT_ATTR2)) { 321 mp->m_flags |= XFS_MOUNT_ATTR2; 322 } else if (!strcmp(this_char, MNTOPT_NOATTR2)) { 323 mp->m_flags &= ~XFS_MOUNT_ATTR2; 324 mp->m_flags |= XFS_MOUNT_NOATTR2; 325 } else if (!strcmp(this_char, MNTOPT_FILESTREAM)) { 326 mp->m_flags |= XFS_MOUNT_FILESTREAMS; 327 } else if (!strcmp(this_char, MNTOPT_NOQUOTA)) { 328 mp->m_qflags &= ~(XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE | 329 XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE | 330 XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE | 331 XFS_UQUOTA_ENFD | XFS_OQUOTA_ENFD); 332 } else if (!strcmp(this_char, MNTOPT_QUOTA) || 333 !strcmp(this_char, MNTOPT_UQUOTA) || 334 !strcmp(this_char, MNTOPT_USRQUOTA)) { 335 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE | 336 XFS_UQUOTA_ENFD); 337 } else if (!strcmp(this_char, MNTOPT_QUOTANOENF) || 338 !strcmp(this_char, MNTOPT_UQUOTANOENF)) { 339 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE); 340 mp->m_qflags &= ~XFS_UQUOTA_ENFD; 341 } else if (!strcmp(this_char, MNTOPT_PQUOTA) || 342 !strcmp(this_char, MNTOPT_PRJQUOTA)) { 343 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE | 344 XFS_OQUOTA_ENFD); 345 } else if (!strcmp(this_char, MNTOPT_PQUOTANOENF)) { 346 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE); 347 mp->m_qflags &= ~XFS_OQUOTA_ENFD; 348 } else if (!strcmp(this_char, MNTOPT_GQUOTA) || 349 !strcmp(this_char, MNTOPT_GRPQUOTA)) { 350 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE | 351 XFS_OQUOTA_ENFD); 352 } else if (!strcmp(this_char, MNTOPT_GQUOTANOENF)) { 353 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE); 354 mp->m_qflags &= ~XFS_OQUOTA_ENFD; 355 } else if (!strcmp(this_char, MNTOPT_DELAYLOG)) { 356 mp->m_flags |= XFS_MOUNT_DELAYLOG; 357 } else if (!strcmp(this_char, MNTOPT_NODELAYLOG)) { 358 mp->m_flags &= ~XFS_MOUNT_DELAYLOG; 359 xfs_warn(mp, 360 "nodelaylog is deprecated and will be removed in Linux 3.3"); 361 } else if (!strcmp(this_char, MNTOPT_DISCARD)) { 362 mp->m_flags |= XFS_MOUNT_DISCARD; 363 } else if (!strcmp(this_char, MNTOPT_NODISCARD)) { 364 mp->m_flags &= ~XFS_MOUNT_DISCARD; 365 } else if (!strcmp(this_char, "ihashsize")) { 366 xfs_warn(mp, 367 "ihashsize no longer used, option is deprecated."); 368 } else if (!strcmp(this_char, "osyncisdsync")) { 369 xfs_warn(mp, 370 "osyncisdsync has no effect, option is deprecated."); 371 } else if (!strcmp(this_char, "osyncisosync")) { 372 xfs_warn(mp, 373 "osyncisosync has no effect, option is deprecated."); 374 } else if (!strcmp(this_char, "irixsgid")) { 375 xfs_warn(mp, 376 "irixsgid is now a sysctl(2) variable, option is deprecated."); 377 } else { 378 xfs_warn(mp, "unknown mount option [%s].", this_char); 379 return EINVAL; 380 } 381 } 382 383 /* 384 * no recovery flag requires a read-only mount 385 */ 386 if ((mp->m_flags & XFS_MOUNT_NORECOVERY) && 387 !(mp->m_flags & XFS_MOUNT_RDONLY)) { 388 xfs_warn(mp, "no-recovery mounts must be read-only."); 389 return EINVAL; 390 } 391 392 if ((mp->m_flags & XFS_MOUNT_NOALIGN) && (dsunit || dswidth)) { 393 xfs_warn(mp, 394 "sunit and swidth options incompatible with the noalign option"); 395 return EINVAL; 396 } 397 398 if ((mp->m_flags & XFS_MOUNT_DISCARD) && 399 !(mp->m_flags & XFS_MOUNT_DELAYLOG)) { 400 xfs_warn(mp, 401 "the discard option is incompatible with the nodelaylog option"); 402 return EINVAL; 403 } 404 405 #ifndef CONFIG_XFS_QUOTA 406 if (XFS_IS_QUOTA_RUNNING(mp)) { 407 xfs_warn(mp, "quota support not available in this kernel."); 408 return EINVAL; 409 } 410 #endif 411 412 if ((mp->m_qflags & (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE)) && 413 (mp->m_qflags & (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE))) { 414 xfs_warn(mp, "cannot mount with both project and group quota"); 415 return EINVAL; 416 } 417 418 if ((dsunit && !dswidth) || (!dsunit && dswidth)) { 419 xfs_warn(mp, "sunit and swidth must be specified together"); 420 return EINVAL; 421 } 422 423 if (dsunit && (dswidth % dsunit != 0)) { 424 xfs_warn(mp, 425 "stripe width (%d) must be a multiple of the stripe unit (%d)", 426 dswidth, dsunit); 427 return EINVAL; 428 } 429 430 done: 431 if (!(mp->m_flags & XFS_MOUNT_NOALIGN)) { 432 /* 433 * At this point the superblock has not been read 434 * in, therefore we do not know the block size. 435 * Before the mount call ends we will convert 436 * these to FSBs. 437 */ 438 if (dsunit) { 439 mp->m_dalign = dsunit; 440 mp->m_flags |= XFS_MOUNT_RETERR; 441 } 442 443 if (dswidth) 444 mp->m_swidth = dswidth; 445 } 446 447 if (mp->m_logbufs != -1 && 448 mp->m_logbufs != 0 && 449 (mp->m_logbufs < XLOG_MIN_ICLOGS || 450 mp->m_logbufs > XLOG_MAX_ICLOGS)) { 451 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]", 452 mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS); 453 return XFS_ERROR(EINVAL); 454 } 455 if (mp->m_logbsize != -1 && 456 mp->m_logbsize != 0 && 457 (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE || 458 mp->m_logbsize > XLOG_MAX_RECORD_BSIZE || 459 !is_power_of_2(mp->m_logbsize))) { 460 xfs_warn(mp, 461 "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]", 462 mp->m_logbsize); 463 return XFS_ERROR(EINVAL); 464 } 465 466 if (iosizelog) { 467 if (iosizelog > XFS_MAX_IO_LOG || 468 iosizelog < XFS_MIN_IO_LOG) { 469 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]", 470 iosizelog, XFS_MIN_IO_LOG, 471 XFS_MAX_IO_LOG); 472 return XFS_ERROR(EINVAL); 473 } 474 475 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE; 476 mp->m_readio_log = iosizelog; 477 mp->m_writeio_log = iosizelog; 478 } 479 480 return 0; 481 } 482 483 struct proc_xfs_info { 484 int flag; 485 char *str; 486 }; 487 488 STATIC int 489 xfs_showargs( 490 struct xfs_mount *mp, 491 struct seq_file *m) 492 { 493 static struct proc_xfs_info xfs_info_set[] = { 494 /* the few simple ones we can get from the mount struct */ 495 { XFS_MOUNT_IKEEP, "," MNTOPT_IKEEP }, 496 { XFS_MOUNT_WSYNC, "," MNTOPT_WSYNC }, 497 { XFS_MOUNT_NOALIGN, "," MNTOPT_NOALIGN }, 498 { XFS_MOUNT_SWALLOC, "," MNTOPT_SWALLOC }, 499 { XFS_MOUNT_NOUUID, "," MNTOPT_NOUUID }, 500 { XFS_MOUNT_NORECOVERY, "," MNTOPT_NORECOVERY }, 501 { XFS_MOUNT_ATTR2, "," MNTOPT_ATTR2 }, 502 { XFS_MOUNT_FILESTREAMS, "," MNTOPT_FILESTREAM }, 503 { XFS_MOUNT_GRPID, "," MNTOPT_GRPID }, 504 { XFS_MOUNT_DELAYLOG, "," MNTOPT_DELAYLOG }, 505 { XFS_MOUNT_DISCARD, "," MNTOPT_DISCARD }, 506 { 0, NULL } 507 }; 508 static struct proc_xfs_info xfs_info_unset[] = { 509 /* the few simple ones we can get from the mount struct */ 510 { XFS_MOUNT_COMPAT_IOSIZE, "," MNTOPT_LARGEIO }, 511 { XFS_MOUNT_BARRIER, "," MNTOPT_NOBARRIER }, 512 { XFS_MOUNT_SMALL_INUMS, "," MNTOPT_64BITINODE }, 513 { 0, NULL } 514 }; 515 struct proc_xfs_info *xfs_infop; 516 517 for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) { 518 if (mp->m_flags & xfs_infop->flag) 519 seq_puts(m, xfs_infop->str); 520 } 521 for (xfs_infop = xfs_info_unset; xfs_infop->flag; xfs_infop++) { 522 if (!(mp->m_flags & xfs_infop->flag)) 523 seq_puts(m, xfs_infop->str); 524 } 525 526 if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) 527 seq_printf(m, "," MNTOPT_ALLOCSIZE "=%dk", 528 (int)(1 << mp->m_writeio_log) >> 10); 529 530 if (mp->m_logbufs > 0) 531 seq_printf(m, "," MNTOPT_LOGBUFS "=%d", mp->m_logbufs); 532 if (mp->m_logbsize > 0) 533 seq_printf(m, "," MNTOPT_LOGBSIZE "=%dk", mp->m_logbsize >> 10); 534 535 if (mp->m_logname) 536 seq_printf(m, "," MNTOPT_LOGDEV "=%s", mp->m_logname); 537 if (mp->m_rtname) 538 seq_printf(m, "," MNTOPT_RTDEV "=%s", mp->m_rtname); 539 540 if (mp->m_dalign > 0) 541 seq_printf(m, "," MNTOPT_SUNIT "=%d", 542 (int)XFS_FSB_TO_BB(mp, mp->m_dalign)); 543 if (mp->m_swidth > 0) 544 seq_printf(m, "," MNTOPT_SWIDTH "=%d", 545 (int)XFS_FSB_TO_BB(mp, mp->m_swidth)); 546 547 if (mp->m_qflags & (XFS_UQUOTA_ACCT|XFS_UQUOTA_ENFD)) 548 seq_puts(m, "," MNTOPT_USRQUOTA); 549 else if (mp->m_qflags & XFS_UQUOTA_ACCT) 550 seq_puts(m, "," MNTOPT_UQUOTANOENF); 551 552 /* Either project or group quotas can be active, not both */ 553 554 if (mp->m_qflags & XFS_PQUOTA_ACCT) { 555 if (mp->m_qflags & XFS_OQUOTA_ENFD) 556 seq_puts(m, "," MNTOPT_PRJQUOTA); 557 else 558 seq_puts(m, "," MNTOPT_PQUOTANOENF); 559 } else if (mp->m_qflags & XFS_GQUOTA_ACCT) { 560 if (mp->m_qflags & XFS_OQUOTA_ENFD) 561 seq_puts(m, "," MNTOPT_GRPQUOTA); 562 else 563 seq_puts(m, "," MNTOPT_GQUOTANOENF); 564 } 565 566 if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT)) 567 seq_puts(m, "," MNTOPT_NOQUOTA); 568 569 return 0; 570 } 571 __uint64_t 572 xfs_max_file_offset( 573 unsigned int blockshift) 574 { 575 unsigned int pagefactor = 1; 576 unsigned int bitshift = BITS_PER_LONG - 1; 577 578 /* Figure out maximum filesize, on Linux this can depend on 579 * the filesystem blocksize (on 32 bit platforms). 580 * __block_write_begin does this in an [unsigned] long... 581 * page->index << (PAGE_CACHE_SHIFT - bbits) 582 * So, for page sized blocks (4K on 32 bit platforms), 583 * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is 584 * (((u64)PAGE_CACHE_SIZE << (BITS_PER_LONG-1))-1) 585 * but for smaller blocksizes it is less (bbits = log2 bsize). 586 * Note1: get_block_t takes a long (implicit cast from above) 587 * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch 588 * can optionally convert the [unsigned] long from above into 589 * an [unsigned] long long. 590 */ 591 592 #if BITS_PER_LONG == 32 593 # if defined(CONFIG_LBDAF) 594 ASSERT(sizeof(sector_t) == 8); 595 pagefactor = PAGE_CACHE_SIZE; 596 bitshift = BITS_PER_LONG; 597 # else 598 pagefactor = PAGE_CACHE_SIZE >> (PAGE_CACHE_SHIFT - blockshift); 599 # endif 600 #endif 601 602 return (((__uint64_t)pagefactor) << bitshift) - 1; 603 } 604 605 STATIC int 606 xfs_blkdev_get( 607 xfs_mount_t *mp, 608 const char *name, 609 struct block_device **bdevp) 610 { 611 int error = 0; 612 613 *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL, 614 mp); 615 if (IS_ERR(*bdevp)) { 616 error = PTR_ERR(*bdevp); 617 xfs_warn(mp, "Invalid device [%s], error=%d\n", name, error); 618 } 619 620 return -error; 621 } 622 623 STATIC void 624 xfs_blkdev_put( 625 struct block_device *bdev) 626 { 627 if (bdev) 628 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); 629 } 630 631 void 632 xfs_blkdev_issue_flush( 633 xfs_buftarg_t *buftarg) 634 { 635 blkdev_issue_flush(buftarg->bt_bdev, GFP_KERNEL, NULL); 636 } 637 638 STATIC void 639 xfs_close_devices( 640 struct xfs_mount *mp) 641 { 642 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) { 643 struct block_device *logdev = mp->m_logdev_targp->bt_bdev; 644 xfs_free_buftarg(mp, mp->m_logdev_targp); 645 xfs_blkdev_put(logdev); 646 } 647 if (mp->m_rtdev_targp) { 648 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev; 649 xfs_free_buftarg(mp, mp->m_rtdev_targp); 650 xfs_blkdev_put(rtdev); 651 } 652 xfs_free_buftarg(mp, mp->m_ddev_targp); 653 } 654 655 /* 656 * The file system configurations are: 657 * (1) device (partition) with data and internal log 658 * (2) logical volume with data and log subvolumes. 659 * (3) logical volume with data, log, and realtime subvolumes. 660 * 661 * We only have to handle opening the log and realtime volumes here if 662 * they are present. The data subvolume has already been opened by 663 * get_sb_bdev() and is stored in sb->s_bdev. 664 */ 665 STATIC int 666 xfs_open_devices( 667 struct xfs_mount *mp) 668 { 669 struct block_device *ddev = mp->m_super->s_bdev; 670 struct block_device *logdev = NULL, *rtdev = NULL; 671 int error; 672 673 /* 674 * Open real time and log devices - order is important. 675 */ 676 if (mp->m_logname) { 677 error = xfs_blkdev_get(mp, mp->m_logname, &logdev); 678 if (error) 679 goto out; 680 } 681 682 if (mp->m_rtname) { 683 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev); 684 if (error) 685 goto out_close_logdev; 686 687 if (rtdev == ddev || rtdev == logdev) { 688 xfs_warn(mp, 689 "Cannot mount filesystem with identical rtdev and ddev/logdev."); 690 error = EINVAL; 691 goto out_close_rtdev; 692 } 693 } 694 695 /* 696 * Setup xfs_mount buffer target pointers 697 */ 698 error = ENOMEM; 699 mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev, 0, mp->m_fsname); 700 if (!mp->m_ddev_targp) 701 goto out_close_rtdev; 702 703 if (rtdev) { 704 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev, 1, 705 mp->m_fsname); 706 if (!mp->m_rtdev_targp) 707 goto out_free_ddev_targ; 708 } 709 710 if (logdev && logdev != ddev) { 711 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev, 1, 712 mp->m_fsname); 713 if (!mp->m_logdev_targp) 714 goto out_free_rtdev_targ; 715 } else { 716 mp->m_logdev_targp = mp->m_ddev_targp; 717 } 718 719 return 0; 720 721 out_free_rtdev_targ: 722 if (mp->m_rtdev_targp) 723 xfs_free_buftarg(mp, mp->m_rtdev_targp); 724 out_free_ddev_targ: 725 xfs_free_buftarg(mp, mp->m_ddev_targp); 726 out_close_rtdev: 727 if (rtdev) 728 xfs_blkdev_put(rtdev); 729 out_close_logdev: 730 if (logdev && logdev != ddev) 731 xfs_blkdev_put(logdev); 732 out: 733 return error; 734 } 735 736 /* 737 * Setup xfs_mount buffer target pointers based on superblock 738 */ 739 STATIC int 740 xfs_setup_devices( 741 struct xfs_mount *mp) 742 { 743 int error; 744 745 error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize, 746 mp->m_sb.sb_sectsize); 747 if (error) 748 return error; 749 750 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) { 751 unsigned int log_sector_size = BBSIZE; 752 753 if (xfs_sb_version_hassector(&mp->m_sb)) 754 log_sector_size = mp->m_sb.sb_logsectsize; 755 error = xfs_setsize_buftarg(mp->m_logdev_targp, 756 mp->m_sb.sb_blocksize, 757 log_sector_size); 758 if (error) 759 return error; 760 } 761 if (mp->m_rtdev_targp) { 762 error = xfs_setsize_buftarg(mp->m_rtdev_targp, 763 mp->m_sb.sb_blocksize, 764 mp->m_sb.sb_sectsize); 765 if (error) 766 return error; 767 } 768 769 return 0; 770 } 771 772 /* Catch misguided souls that try to use this interface on XFS */ 773 STATIC struct inode * 774 xfs_fs_alloc_inode( 775 struct super_block *sb) 776 { 777 BUG(); 778 return NULL; 779 } 780 781 /* 782 * Now that the generic code is guaranteed not to be accessing 783 * the linux inode, we can reclaim the inode. 784 */ 785 STATIC void 786 xfs_fs_destroy_inode( 787 struct inode *inode) 788 { 789 struct xfs_inode *ip = XFS_I(inode); 790 791 trace_xfs_destroy_inode(ip); 792 793 XFS_STATS_INC(vn_reclaim); 794 795 /* bad inode, get out here ASAP */ 796 if (is_bad_inode(inode)) 797 goto out_reclaim; 798 799 xfs_ioend_wait(ip); 800 801 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0); 802 803 /* 804 * We should never get here with one of the reclaim flags already set. 805 */ 806 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIMABLE)); 807 ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIM)); 808 809 /* 810 * We always use background reclaim here because even if the 811 * inode is clean, it still may be under IO and hence we have 812 * to take the flush lock. The background reclaim path handles 813 * this more efficiently than we can here, so simply let background 814 * reclaim tear down all inodes. 815 */ 816 out_reclaim: 817 xfs_inode_set_reclaim_tag(ip); 818 } 819 820 /* 821 * Slab object creation initialisation for the XFS inode. 822 * This covers only the idempotent fields in the XFS inode; 823 * all other fields need to be initialised on allocation 824 * from the slab. This avoids the need to repeatedly initialise 825 * fields in the xfs inode that left in the initialise state 826 * when freeing the inode. 827 */ 828 STATIC void 829 xfs_fs_inode_init_once( 830 void *inode) 831 { 832 struct xfs_inode *ip = inode; 833 834 memset(ip, 0, sizeof(struct xfs_inode)); 835 836 /* vfs inode */ 837 inode_init_once(VFS_I(ip)); 838 839 /* xfs inode */ 840 atomic_set(&ip->i_iocount, 0); 841 atomic_set(&ip->i_pincount, 0); 842 spin_lock_init(&ip->i_flags_lock); 843 init_waitqueue_head(&ip->i_ipin_wait); 844 /* 845 * Because we want to use a counting completion, complete 846 * the flush completion once to allow a single access to 847 * the flush completion without blocking. 848 */ 849 init_completion(&ip->i_flush); 850 complete(&ip->i_flush); 851 852 mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER, 853 "xfsino", ip->i_ino); 854 } 855 856 /* 857 * Dirty the XFS inode when mark_inode_dirty_sync() is called so that 858 * we catch unlogged VFS level updates to the inode. 859 * 860 * We need the barrier() to maintain correct ordering between unlogged 861 * updates and the transaction commit code that clears the i_update_core 862 * field. This requires all updates to be completed before marking the 863 * inode dirty. 864 */ 865 STATIC void 866 xfs_fs_dirty_inode( 867 struct inode *inode, 868 int flags) 869 { 870 barrier(); 871 XFS_I(inode)->i_update_core = 1; 872 } 873 874 STATIC int 875 xfs_log_inode( 876 struct xfs_inode *ip) 877 { 878 struct xfs_mount *mp = ip->i_mount; 879 struct xfs_trans *tp; 880 int error; 881 882 tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS); 883 error = xfs_trans_reserve(tp, 0, XFS_FSYNC_TS_LOG_RES(mp), 0, 0, 0); 884 if (error) { 885 xfs_trans_cancel(tp, 0); 886 return error; 887 } 888 889 xfs_ilock(ip, XFS_ILOCK_EXCL); 890 xfs_trans_ijoin_ref(tp, ip, XFS_ILOCK_EXCL); 891 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 892 return xfs_trans_commit(tp, 0); 893 } 894 895 STATIC int 896 xfs_fs_write_inode( 897 struct inode *inode, 898 struct writeback_control *wbc) 899 { 900 struct xfs_inode *ip = XFS_I(inode); 901 struct xfs_mount *mp = ip->i_mount; 902 int error = EAGAIN; 903 904 trace_xfs_write_inode(ip); 905 906 if (XFS_FORCED_SHUTDOWN(mp)) 907 return -XFS_ERROR(EIO); 908 if (!ip->i_update_core) 909 return 0; 910 911 if (wbc->sync_mode == WB_SYNC_ALL) { 912 /* 913 * Make sure the inode has made it it into the log. Instead 914 * of forcing it all the way to stable storage using a 915 * synchronous transaction we let the log force inside the 916 * ->sync_fs call do that for thus, which reduces the number 917 * of synchronous log foces dramatically. 918 */ 919 xfs_ioend_wait(ip); 920 error = xfs_log_inode(ip); 921 if (error) 922 goto out; 923 return 0; 924 } else { 925 /* 926 * We make this non-blocking if the inode is contended, return 927 * EAGAIN to indicate to the caller that they did not succeed. 928 * This prevents the flush path from blocking on inodes inside 929 * another operation right now, they get caught later by 930 * xfs_sync. 931 */ 932 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) 933 goto out; 934 935 if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) 936 goto out_unlock; 937 938 /* 939 * Now we have the flush lock and the inode is not pinned, we 940 * can check if the inode is really clean as we know that 941 * there are no pending transaction completions, it is not 942 * waiting on the delayed write queue and there is no IO in 943 * progress. 944 */ 945 if (xfs_inode_clean(ip)) { 946 xfs_ifunlock(ip); 947 error = 0; 948 goto out_unlock; 949 } 950 error = xfs_iflush(ip, SYNC_TRYLOCK); 951 } 952 953 out_unlock: 954 xfs_iunlock(ip, XFS_ILOCK_SHARED); 955 out: 956 /* 957 * if we failed to write out the inode then mark 958 * it dirty again so we'll try again later. 959 */ 960 if (error) 961 xfs_mark_inode_dirty_sync(ip); 962 return -error; 963 } 964 965 STATIC void 966 xfs_fs_evict_inode( 967 struct inode *inode) 968 { 969 xfs_inode_t *ip = XFS_I(inode); 970 971 trace_xfs_evict_inode(ip); 972 973 truncate_inode_pages(&inode->i_data, 0); 974 end_writeback(inode); 975 XFS_STATS_INC(vn_rele); 976 XFS_STATS_INC(vn_remove); 977 XFS_STATS_DEC(vn_active); 978 979 /* 980 * The iolock is used by the file system to coordinate reads, 981 * writes, and block truncates. Up to this point the lock 982 * protected concurrent accesses by users of the inode. But 983 * from here forward we're doing some final processing of the 984 * inode because we're done with it, and although we reuse the 985 * iolock for protection it is really a distinct lock class 986 * (in the lockdep sense) from before. To keep lockdep happy 987 * (and basically indicate what we are doing), we explicitly 988 * re-init the iolock here. 989 */ 990 ASSERT(!rwsem_is_locked(&ip->i_iolock.mr_lock)); 991 mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", ip->i_ino); 992 lockdep_set_class_and_name(&ip->i_iolock.mr_lock, 993 &xfs_iolock_reclaimable, "xfs_iolock_reclaimable"); 994 995 xfs_inactive(ip); 996 } 997 998 STATIC void 999 xfs_free_fsname( 1000 struct xfs_mount *mp) 1001 { 1002 kfree(mp->m_fsname); 1003 kfree(mp->m_rtname); 1004 kfree(mp->m_logname); 1005 } 1006 1007 STATIC void 1008 xfs_fs_put_super( 1009 struct super_block *sb) 1010 { 1011 struct xfs_mount *mp = XFS_M(sb); 1012 1013 xfs_syncd_stop(mp); 1014 1015 /* 1016 * Blow away any referenced inode in the filestreams cache. 1017 * This can and will cause log traffic as inodes go inactive 1018 * here. 1019 */ 1020 xfs_filestream_unmount(mp); 1021 1022 XFS_bflush(mp->m_ddev_targp); 1023 1024 xfs_unmountfs(mp); 1025 xfs_freesb(mp); 1026 xfs_icsb_destroy_counters(mp); 1027 xfs_close_devices(mp); 1028 xfs_free_fsname(mp); 1029 kfree(mp); 1030 } 1031 1032 STATIC int 1033 xfs_fs_sync_fs( 1034 struct super_block *sb, 1035 int wait) 1036 { 1037 struct xfs_mount *mp = XFS_M(sb); 1038 int error; 1039 1040 /* 1041 * Not much we can do for the first async pass. Writing out the 1042 * superblock would be counter-productive as we are going to redirty 1043 * when writing out other data and metadata (and writing out a single 1044 * block is quite fast anyway). 1045 * 1046 * Try to asynchronously kick off quota syncing at least. 1047 */ 1048 if (!wait) { 1049 xfs_qm_sync(mp, SYNC_TRYLOCK); 1050 return 0; 1051 } 1052 1053 error = xfs_quiesce_data(mp); 1054 if (error) 1055 return -error; 1056 1057 if (laptop_mode) { 1058 /* 1059 * The disk must be active because we're syncing. 1060 * We schedule xfssyncd now (now that the disk is 1061 * active) instead of later (when it might not be). 1062 */ 1063 flush_delayed_work_sync(&mp->m_sync_work); 1064 } 1065 1066 return 0; 1067 } 1068 1069 STATIC int 1070 xfs_fs_statfs( 1071 struct dentry *dentry, 1072 struct kstatfs *statp) 1073 { 1074 struct xfs_mount *mp = XFS_M(dentry->d_sb); 1075 xfs_sb_t *sbp = &mp->m_sb; 1076 struct xfs_inode *ip = XFS_I(dentry->d_inode); 1077 __uint64_t fakeinos, id; 1078 xfs_extlen_t lsize; 1079 __int64_t ffree; 1080 1081 statp->f_type = XFS_SB_MAGIC; 1082 statp->f_namelen = MAXNAMELEN - 1; 1083 1084 id = huge_encode_dev(mp->m_ddev_targp->bt_dev); 1085 statp->f_fsid.val[0] = (u32)id; 1086 statp->f_fsid.val[1] = (u32)(id >> 32); 1087 1088 xfs_icsb_sync_counters(mp, XFS_ICSB_LAZY_COUNT); 1089 1090 spin_lock(&mp->m_sb_lock); 1091 statp->f_bsize = sbp->sb_blocksize; 1092 lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0; 1093 statp->f_blocks = sbp->sb_dblocks - lsize; 1094 statp->f_bfree = statp->f_bavail = 1095 sbp->sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp); 1096 fakeinos = statp->f_bfree << sbp->sb_inopblog; 1097 statp->f_files = 1098 MIN(sbp->sb_icount + fakeinos, (__uint64_t)XFS_MAXINUMBER); 1099 if (mp->m_maxicount) 1100 statp->f_files = min_t(typeof(statp->f_files), 1101 statp->f_files, 1102 mp->m_maxicount); 1103 1104 /* make sure statp->f_ffree does not underflow */ 1105 ffree = statp->f_files - (sbp->sb_icount - sbp->sb_ifree); 1106 statp->f_ffree = max_t(__int64_t, ffree, 0); 1107 1108 spin_unlock(&mp->m_sb_lock); 1109 1110 if ((ip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) || 1111 ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_OQUOTA_ENFD))) == 1112 (XFS_PQUOTA_ACCT|XFS_OQUOTA_ENFD)) 1113 xfs_qm_statvfs(ip, statp); 1114 return 0; 1115 } 1116 1117 STATIC void 1118 xfs_save_resvblks(struct xfs_mount *mp) 1119 { 1120 __uint64_t resblks = 0; 1121 1122 mp->m_resblks_save = mp->m_resblks; 1123 xfs_reserve_blocks(mp, &resblks, NULL); 1124 } 1125 1126 STATIC void 1127 xfs_restore_resvblks(struct xfs_mount *mp) 1128 { 1129 __uint64_t resblks; 1130 1131 if (mp->m_resblks_save) { 1132 resblks = mp->m_resblks_save; 1133 mp->m_resblks_save = 0; 1134 } else 1135 resblks = xfs_default_resblks(mp); 1136 1137 xfs_reserve_blocks(mp, &resblks, NULL); 1138 } 1139 1140 STATIC int 1141 xfs_fs_remount( 1142 struct super_block *sb, 1143 int *flags, 1144 char *options) 1145 { 1146 struct xfs_mount *mp = XFS_M(sb); 1147 substring_t args[MAX_OPT_ARGS]; 1148 char *p; 1149 int error; 1150 1151 while ((p = strsep(&options, ",")) != NULL) { 1152 int token; 1153 1154 if (!*p) 1155 continue; 1156 1157 token = match_token(p, tokens, args); 1158 switch (token) { 1159 case Opt_barrier: 1160 mp->m_flags |= XFS_MOUNT_BARRIER; 1161 break; 1162 case Opt_nobarrier: 1163 mp->m_flags &= ~XFS_MOUNT_BARRIER; 1164 break; 1165 default: 1166 /* 1167 * Logically we would return an error here to prevent 1168 * users from believing they might have changed 1169 * mount options using remount which can't be changed. 1170 * 1171 * But unfortunately mount(8) adds all options from 1172 * mtab and fstab to the mount arguments in some cases 1173 * so we can't blindly reject options, but have to 1174 * check for each specified option if it actually 1175 * differs from the currently set option and only 1176 * reject it if that's the case. 1177 * 1178 * Until that is implemented we return success for 1179 * every remount request, and silently ignore all 1180 * options that we can't actually change. 1181 */ 1182 #if 0 1183 xfs_info(mp, 1184 "mount option \"%s\" not supported for remount\n", p); 1185 return -EINVAL; 1186 #else 1187 break; 1188 #endif 1189 } 1190 } 1191 1192 /* ro -> rw */ 1193 if ((mp->m_flags & XFS_MOUNT_RDONLY) && !(*flags & MS_RDONLY)) { 1194 mp->m_flags &= ~XFS_MOUNT_RDONLY; 1195 1196 /* 1197 * If this is the first remount to writeable state we 1198 * might have some superblock changes to update. 1199 */ 1200 if (mp->m_update_flags) { 1201 error = xfs_mount_log_sb(mp, mp->m_update_flags); 1202 if (error) { 1203 xfs_warn(mp, "failed to write sb changes"); 1204 return error; 1205 } 1206 mp->m_update_flags = 0; 1207 } 1208 1209 /* 1210 * Fill out the reserve pool if it is empty. Use the stashed 1211 * value if it is non-zero, otherwise go with the default. 1212 */ 1213 xfs_restore_resvblks(mp); 1214 } 1215 1216 /* rw -> ro */ 1217 if (!(mp->m_flags & XFS_MOUNT_RDONLY) && (*flags & MS_RDONLY)) { 1218 /* 1219 * After we have synced the data but before we sync the 1220 * metadata, we need to free up the reserve block pool so that 1221 * the used block count in the superblock on disk is correct at 1222 * the end of the remount. Stash the current reserve pool size 1223 * so that if we get remounted rw, we can return it to the same 1224 * size. 1225 */ 1226 1227 xfs_quiesce_data(mp); 1228 xfs_save_resvblks(mp); 1229 xfs_quiesce_attr(mp); 1230 mp->m_flags |= XFS_MOUNT_RDONLY; 1231 } 1232 1233 return 0; 1234 } 1235 1236 /* 1237 * Second stage of a freeze. The data is already frozen so we only 1238 * need to take care of the metadata. Once that's done write a dummy 1239 * record to dirty the log in case of a crash while frozen. 1240 */ 1241 STATIC int 1242 xfs_fs_freeze( 1243 struct super_block *sb) 1244 { 1245 struct xfs_mount *mp = XFS_M(sb); 1246 1247 xfs_save_resvblks(mp); 1248 xfs_quiesce_attr(mp); 1249 return -xfs_fs_log_dummy(mp); 1250 } 1251 1252 STATIC int 1253 xfs_fs_unfreeze( 1254 struct super_block *sb) 1255 { 1256 struct xfs_mount *mp = XFS_M(sb); 1257 1258 xfs_restore_resvblks(mp); 1259 return 0; 1260 } 1261 1262 STATIC int 1263 xfs_fs_show_options( 1264 struct seq_file *m, 1265 struct vfsmount *mnt) 1266 { 1267 return -xfs_showargs(XFS_M(mnt->mnt_sb), m); 1268 } 1269 1270 /* 1271 * This function fills in xfs_mount_t fields based on mount args. 1272 * Note: the superblock _has_ now been read in. 1273 */ 1274 STATIC int 1275 xfs_finish_flags( 1276 struct xfs_mount *mp) 1277 { 1278 int ronly = (mp->m_flags & XFS_MOUNT_RDONLY); 1279 1280 /* Fail a mount where the logbuf is smaller than the log stripe */ 1281 if (xfs_sb_version_haslogv2(&mp->m_sb)) { 1282 if (mp->m_logbsize <= 0 && 1283 mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) { 1284 mp->m_logbsize = mp->m_sb.sb_logsunit; 1285 } else if (mp->m_logbsize > 0 && 1286 mp->m_logbsize < mp->m_sb.sb_logsunit) { 1287 xfs_warn(mp, 1288 "logbuf size must be greater than or equal to log stripe size"); 1289 return XFS_ERROR(EINVAL); 1290 } 1291 } else { 1292 /* Fail a mount if the logbuf is larger than 32K */ 1293 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) { 1294 xfs_warn(mp, 1295 "logbuf size for version 1 logs must be 16K or 32K"); 1296 return XFS_ERROR(EINVAL); 1297 } 1298 } 1299 1300 /* 1301 * mkfs'ed attr2 will turn on attr2 mount unless explicitly 1302 * told by noattr2 to turn it off 1303 */ 1304 if (xfs_sb_version_hasattr2(&mp->m_sb) && 1305 !(mp->m_flags & XFS_MOUNT_NOATTR2)) 1306 mp->m_flags |= XFS_MOUNT_ATTR2; 1307 1308 /* 1309 * prohibit r/w mounts of read-only filesystems 1310 */ 1311 if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) { 1312 xfs_warn(mp, 1313 "cannot mount a read-only filesystem as read-write"); 1314 return XFS_ERROR(EROFS); 1315 } 1316 1317 return 0; 1318 } 1319 1320 STATIC int 1321 xfs_fs_fill_super( 1322 struct super_block *sb, 1323 void *data, 1324 int silent) 1325 { 1326 struct inode *root; 1327 struct xfs_mount *mp = NULL; 1328 int flags = 0, error = ENOMEM; 1329 1330 mp = kzalloc(sizeof(struct xfs_mount), GFP_KERNEL); 1331 if (!mp) 1332 goto out; 1333 1334 spin_lock_init(&mp->m_sb_lock); 1335 mutex_init(&mp->m_growlock); 1336 atomic_set(&mp->m_active_trans, 0); 1337 1338 mp->m_super = sb; 1339 sb->s_fs_info = mp; 1340 1341 error = xfs_parseargs(mp, (char *)data); 1342 if (error) 1343 goto out_free_fsname; 1344 1345 sb_min_blocksize(sb, BBSIZE); 1346 sb->s_xattr = xfs_xattr_handlers; 1347 sb->s_export_op = &xfs_export_operations; 1348 #ifdef CONFIG_XFS_QUOTA 1349 sb->s_qcop = &xfs_quotactl_operations; 1350 #endif 1351 sb->s_op = &xfs_super_operations; 1352 1353 if (silent) 1354 flags |= XFS_MFSI_QUIET; 1355 1356 error = xfs_open_devices(mp); 1357 if (error) 1358 goto out_free_fsname; 1359 1360 error = xfs_icsb_init_counters(mp); 1361 if (error) 1362 goto out_close_devices; 1363 1364 error = xfs_readsb(mp, flags); 1365 if (error) 1366 goto out_destroy_counters; 1367 1368 error = xfs_finish_flags(mp); 1369 if (error) 1370 goto out_free_sb; 1371 1372 error = xfs_setup_devices(mp); 1373 if (error) 1374 goto out_free_sb; 1375 1376 error = xfs_filestream_mount(mp); 1377 if (error) 1378 goto out_free_sb; 1379 1380 /* 1381 * we must configure the block size in the superblock before we run the 1382 * full mount process as the mount process can lookup and cache inodes. 1383 * For the same reason we must also initialise the syncd and register 1384 * the inode cache shrinker so that inodes can be reclaimed during 1385 * operations like a quotacheck that iterate all inodes in the 1386 * filesystem. 1387 */ 1388 sb->s_magic = XFS_SB_MAGIC; 1389 sb->s_blocksize = mp->m_sb.sb_blocksize; 1390 sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1; 1391 sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits); 1392 sb->s_time_gran = 1; 1393 set_posix_acl_flag(sb); 1394 1395 error = xfs_mountfs(mp); 1396 if (error) 1397 goto out_filestream_unmount; 1398 1399 error = xfs_syncd_init(mp); 1400 if (error) 1401 goto out_unmount; 1402 1403 root = igrab(VFS_I(mp->m_rootip)); 1404 if (!root) { 1405 error = ENOENT; 1406 goto out_syncd_stop; 1407 } 1408 if (is_bad_inode(root)) { 1409 error = EINVAL; 1410 goto out_syncd_stop; 1411 } 1412 sb->s_root = d_alloc_root(root); 1413 if (!sb->s_root) { 1414 error = ENOMEM; 1415 goto out_iput; 1416 } 1417 1418 return 0; 1419 1420 out_filestream_unmount: 1421 xfs_filestream_unmount(mp); 1422 out_free_sb: 1423 xfs_freesb(mp); 1424 out_destroy_counters: 1425 xfs_icsb_destroy_counters(mp); 1426 out_close_devices: 1427 xfs_close_devices(mp); 1428 out_free_fsname: 1429 xfs_free_fsname(mp); 1430 kfree(mp); 1431 out: 1432 return -error; 1433 1434 out_iput: 1435 iput(root); 1436 out_syncd_stop: 1437 xfs_syncd_stop(mp); 1438 out_unmount: 1439 /* 1440 * Blow away any referenced inode in the filestreams cache. 1441 * This can and will cause log traffic as inodes go inactive 1442 * here. 1443 */ 1444 xfs_filestream_unmount(mp); 1445 1446 XFS_bflush(mp->m_ddev_targp); 1447 1448 xfs_unmountfs(mp); 1449 goto out_free_sb; 1450 } 1451 1452 STATIC struct dentry * 1453 xfs_fs_mount( 1454 struct file_system_type *fs_type, 1455 int flags, 1456 const char *dev_name, 1457 void *data) 1458 { 1459 return mount_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super); 1460 } 1461 1462 static int 1463 xfs_fs_nr_cached_objects( 1464 struct super_block *sb) 1465 { 1466 return xfs_reclaim_inodes_count(XFS_M(sb)); 1467 } 1468 1469 static void 1470 xfs_fs_free_cached_objects( 1471 struct super_block *sb, 1472 int nr_to_scan) 1473 { 1474 xfs_reclaim_inodes_nr(XFS_M(sb), nr_to_scan); 1475 } 1476 1477 static const struct super_operations xfs_super_operations = { 1478 .alloc_inode = xfs_fs_alloc_inode, 1479 .destroy_inode = xfs_fs_destroy_inode, 1480 .dirty_inode = xfs_fs_dirty_inode, 1481 .write_inode = xfs_fs_write_inode, 1482 .evict_inode = xfs_fs_evict_inode, 1483 .put_super = xfs_fs_put_super, 1484 .sync_fs = xfs_fs_sync_fs, 1485 .freeze_fs = xfs_fs_freeze, 1486 .unfreeze_fs = xfs_fs_unfreeze, 1487 .statfs = xfs_fs_statfs, 1488 .remount_fs = xfs_fs_remount, 1489 .show_options = xfs_fs_show_options, 1490 .nr_cached_objects = xfs_fs_nr_cached_objects, 1491 .free_cached_objects = xfs_fs_free_cached_objects, 1492 }; 1493 1494 static struct file_system_type xfs_fs_type = { 1495 .owner = THIS_MODULE, 1496 .name = "xfs", 1497 .mount = xfs_fs_mount, 1498 .kill_sb = kill_block_super, 1499 .fs_flags = FS_REQUIRES_DEV, 1500 }; 1501 1502 STATIC int __init 1503 xfs_init_zones(void) 1504 { 1505 1506 xfs_ioend_zone = kmem_zone_init(sizeof(xfs_ioend_t), "xfs_ioend"); 1507 if (!xfs_ioend_zone) 1508 goto out; 1509 1510 xfs_ioend_pool = mempool_create_slab_pool(4 * MAX_BUF_PER_PAGE, 1511 xfs_ioend_zone); 1512 if (!xfs_ioend_pool) 1513 goto out_destroy_ioend_zone; 1514 1515 xfs_log_ticket_zone = kmem_zone_init(sizeof(xlog_ticket_t), 1516 "xfs_log_ticket"); 1517 if (!xfs_log_ticket_zone) 1518 goto out_destroy_ioend_pool; 1519 1520 xfs_bmap_free_item_zone = kmem_zone_init(sizeof(xfs_bmap_free_item_t), 1521 "xfs_bmap_free_item"); 1522 if (!xfs_bmap_free_item_zone) 1523 goto out_destroy_log_ticket_zone; 1524 1525 xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t), 1526 "xfs_btree_cur"); 1527 if (!xfs_btree_cur_zone) 1528 goto out_destroy_bmap_free_item_zone; 1529 1530 xfs_da_state_zone = kmem_zone_init(sizeof(xfs_da_state_t), 1531 "xfs_da_state"); 1532 if (!xfs_da_state_zone) 1533 goto out_destroy_btree_cur_zone; 1534 1535 xfs_dabuf_zone = kmem_zone_init(sizeof(xfs_dabuf_t), "xfs_dabuf"); 1536 if (!xfs_dabuf_zone) 1537 goto out_destroy_da_state_zone; 1538 1539 xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork"); 1540 if (!xfs_ifork_zone) 1541 goto out_destroy_dabuf_zone; 1542 1543 xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans"); 1544 if (!xfs_trans_zone) 1545 goto out_destroy_ifork_zone; 1546 1547 xfs_log_item_desc_zone = 1548 kmem_zone_init(sizeof(struct xfs_log_item_desc), 1549 "xfs_log_item_desc"); 1550 if (!xfs_log_item_desc_zone) 1551 goto out_destroy_trans_zone; 1552 1553 /* 1554 * The size of the zone allocated buf log item is the maximum 1555 * size possible under XFS. This wastes a little bit of memory, 1556 * but it is much faster. 1557 */ 1558 xfs_buf_item_zone = kmem_zone_init((sizeof(xfs_buf_log_item_t) + 1559 (((XFS_MAX_BLOCKSIZE / XFS_BLF_CHUNK) / 1560 NBWORD) * sizeof(int))), "xfs_buf_item"); 1561 if (!xfs_buf_item_zone) 1562 goto out_destroy_log_item_desc_zone; 1563 1564 xfs_efd_zone = kmem_zone_init((sizeof(xfs_efd_log_item_t) + 1565 ((XFS_EFD_MAX_FAST_EXTENTS - 1) * 1566 sizeof(xfs_extent_t))), "xfs_efd_item"); 1567 if (!xfs_efd_zone) 1568 goto out_destroy_buf_item_zone; 1569 1570 xfs_efi_zone = kmem_zone_init((sizeof(xfs_efi_log_item_t) + 1571 ((XFS_EFI_MAX_FAST_EXTENTS - 1) * 1572 sizeof(xfs_extent_t))), "xfs_efi_item"); 1573 if (!xfs_efi_zone) 1574 goto out_destroy_efd_zone; 1575 1576 xfs_inode_zone = 1577 kmem_zone_init_flags(sizeof(xfs_inode_t), "xfs_inode", 1578 KM_ZONE_HWALIGN | KM_ZONE_RECLAIM | KM_ZONE_SPREAD, 1579 xfs_fs_inode_init_once); 1580 if (!xfs_inode_zone) 1581 goto out_destroy_efi_zone; 1582 1583 xfs_ili_zone = 1584 kmem_zone_init_flags(sizeof(xfs_inode_log_item_t), "xfs_ili", 1585 KM_ZONE_SPREAD, NULL); 1586 if (!xfs_ili_zone) 1587 goto out_destroy_inode_zone; 1588 1589 return 0; 1590 1591 out_destroy_inode_zone: 1592 kmem_zone_destroy(xfs_inode_zone); 1593 out_destroy_efi_zone: 1594 kmem_zone_destroy(xfs_efi_zone); 1595 out_destroy_efd_zone: 1596 kmem_zone_destroy(xfs_efd_zone); 1597 out_destroy_buf_item_zone: 1598 kmem_zone_destroy(xfs_buf_item_zone); 1599 out_destroy_log_item_desc_zone: 1600 kmem_zone_destroy(xfs_log_item_desc_zone); 1601 out_destroy_trans_zone: 1602 kmem_zone_destroy(xfs_trans_zone); 1603 out_destroy_ifork_zone: 1604 kmem_zone_destroy(xfs_ifork_zone); 1605 out_destroy_dabuf_zone: 1606 kmem_zone_destroy(xfs_dabuf_zone); 1607 out_destroy_da_state_zone: 1608 kmem_zone_destroy(xfs_da_state_zone); 1609 out_destroy_btree_cur_zone: 1610 kmem_zone_destroy(xfs_btree_cur_zone); 1611 out_destroy_bmap_free_item_zone: 1612 kmem_zone_destroy(xfs_bmap_free_item_zone); 1613 out_destroy_log_ticket_zone: 1614 kmem_zone_destroy(xfs_log_ticket_zone); 1615 out_destroy_ioend_pool: 1616 mempool_destroy(xfs_ioend_pool); 1617 out_destroy_ioend_zone: 1618 kmem_zone_destroy(xfs_ioend_zone); 1619 out: 1620 return -ENOMEM; 1621 } 1622 1623 STATIC void 1624 xfs_destroy_zones(void) 1625 { 1626 kmem_zone_destroy(xfs_ili_zone); 1627 kmem_zone_destroy(xfs_inode_zone); 1628 kmem_zone_destroy(xfs_efi_zone); 1629 kmem_zone_destroy(xfs_efd_zone); 1630 kmem_zone_destroy(xfs_buf_item_zone); 1631 kmem_zone_destroy(xfs_log_item_desc_zone); 1632 kmem_zone_destroy(xfs_trans_zone); 1633 kmem_zone_destroy(xfs_ifork_zone); 1634 kmem_zone_destroy(xfs_dabuf_zone); 1635 kmem_zone_destroy(xfs_da_state_zone); 1636 kmem_zone_destroy(xfs_btree_cur_zone); 1637 kmem_zone_destroy(xfs_bmap_free_item_zone); 1638 kmem_zone_destroy(xfs_log_ticket_zone); 1639 mempool_destroy(xfs_ioend_pool); 1640 kmem_zone_destroy(xfs_ioend_zone); 1641 1642 } 1643 1644 STATIC int __init 1645 xfs_init_workqueues(void) 1646 { 1647 /* 1648 * max_active is set to 8 to give enough concurency to allow 1649 * multiple work operations on each CPU to run. This allows multiple 1650 * filesystems to be running sync work concurrently, and scales with 1651 * the number of CPUs in the system. 1652 */ 1653 xfs_syncd_wq = alloc_workqueue("xfssyncd", WQ_CPU_INTENSIVE, 8); 1654 if (!xfs_syncd_wq) 1655 goto out; 1656 1657 xfs_ail_wq = alloc_workqueue("xfsail", WQ_CPU_INTENSIVE, 8); 1658 if (!xfs_ail_wq) 1659 goto out_destroy_syncd; 1660 1661 return 0; 1662 1663 out_destroy_syncd: 1664 destroy_workqueue(xfs_syncd_wq); 1665 out: 1666 return -ENOMEM; 1667 } 1668 1669 STATIC void 1670 xfs_destroy_workqueues(void) 1671 { 1672 destroy_workqueue(xfs_ail_wq); 1673 destroy_workqueue(xfs_syncd_wq); 1674 } 1675 1676 STATIC int __init 1677 init_xfs_fs(void) 1678 { 1679 int error; 1680 1681 printk(KERN_INFO XFS_VERSION_STRING " with " 1682 XFS_BUILD_OPTIONS " enabled\n"); 1683 1684 xfs_ioend_init(); 1685 xfs_dir_startup(); 1686 1687 error = xfs_init_zones(); 1688 if (error) 1689 goto out; 1690 1691 error = xfs_init_workqueues(); 1692 if (error) 1693 goto out_destroy_zones; 1694 1695 error = xfs_mru_cache_init(); 1696 if (error) 1697 goto out_destroy_wq; 1698 1699 error = xfs_filestream_init(); 1700 if (error) 1701 goto out_mru_cache_uninit; 1702 1703 error = xfs_buf_init(); 1704 if (error) 1705 goto out_filestream_uninit; 1706 1707 error = xfs_init_procfs(); 1708 if (error) 1709 goto out_buf_terminate; 1710 1711 error = xfs_sysctl_register(); 1712 if (error) 1713 goto out_cleanup_procfs; 1714 1715 vfs_initquota(); 1716 1717 error = register_filesystem(&xfs_fs_type); 1718 if (error) 1719 goto out_sysctl_unregister; 1720 return 0; 1721 1722 out_sysctl_unregister: 1723 xfs_sysctl_unregister(); 1724 out_cleanup_procfs: 1725 xfs_cleanup_procfs(); 1726 out_buf_terminate: 1727 xfs_buf_terminate(); 1728 out_filestream_uninit: 1729 xfs_filestream_uninit(); 1730 out_mru_cache_uninit: 1731 xfs_mru_cache_uninit(); 1732 out_destroy_wq: 1733 xfs_destroy_workqueues(); 1734 out_destroy_zones: 1735 xfs_destroy_zones(); 1736 out: 1737 return error; 1738 } 1739 1740 STATIC void __exit 1741 exit_xfs_fs(void) 1742 { 1743 vfs_exitquota(); 1744 unregister_filesystem(&xfs_fs_type); 1745 xfs_sysctl_unregister(); 1746 xfs_cleanup_procfs(); 1747 xfs_buf_terminate(); 1748 xfs_filestream_uninit(); 1749 xfs_mru_cache_uninit(); 1750 xfs_destroy_workqueues(); 1751 xfs_destroy_zones(); 1752 } 1753 1754 module_init(init_xfs_fs); 1755 module_exit(exit_xfs_fs); 1756 1757 MODULE_AUTHOR("Silicon Graphics, Inc."); 1758 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled"); 1759 MODULE_LICENSE("GPL"); 1760