1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (c) 2000-2005 Silicon Graphics, Inc. 4 * All Rights Reserved. 5 */ 6 #include "xfs.h" 7 #include "xfs_fs.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_bit.h" 13 #include "xfs_sb.h" 14 #include "xfs_mount.h" 15 #include "xfs_ialloc.h" 16 #include "xfs_alloc.h" 17 #include "xfs_error.h" 18 #include "xfs_trans.h" 19 #include "xfs_buf_item.h" 20 #include "xfs_bmap_btree.h" 21 #include "xfs_alloc_btree.h" 22 #include "xfs_log.h" 23 #include "xfs_rmap_btree.h" 24 #include "xfs_refcount_btree.h" 25 #include "xfs_da_format.h" 26 #include "xfs_health.h" 27 #include "xfs_ag.h" 28 #include "xfs_rtbitmap.h" 29 #include "xfs_exchrange.h" 30 31 /* 32 * Physical superblock buffer manipulations. Shared with libxfs in userspace. 33 */ 34 35 /* 36 * Check that all the V4 feature bits that the V5 filesystem format requires are 37 * correctly set. 38 */ 39 static bool 40 xfs_sb_validate_v5_features( 41 struct xfs_sb *sbp) 42 { 43 /* We must not have any unknown V4 feature bits set */ 44 if (sbp->sb_versionnum & ~XFS_SB_VERSION_OKBITS) 45 return false; 46 47 /* 48 * The CRC bit is considered an invalid V4 flag, so we have to add it 49 * manually to the OKBITS mask. 50 */ 51 if (sbp->sb_features2 & ~(XFS_SB_VERSION2_OKBITS | 52 XFS_SB_VERSION2_CRCBIT)) 53 return false; 54 55 /* Now check all the required V4 feature flags are set. */ 56 57 #define V5_VERS_FLAGS (XFS_SB_VERSION_NLINKBIT | \ 58 XFS_SB_VERSION_ALIGNBIT | \ 59 XFS_SB_VERSION_LOGV2BIT | \ 60 XFS_SB_VERSION_EXTFLGBIT | \ 61 XFS_SB_VERSION_DIRV2BIT | \ 62 XFS_SB_VERSION_MOREBITSBIT) 63 64 #define V5_FEAT_FLAGS (XFS_SB_VERSION2_LAZYSBCOUNTBIT | \ 65 XFS_SB_VERSION2_ATTR2BIT | \ 66 XFS_SB_VERSION2_PROJID32BIT | \ 67 XFS_SB_VERSION2_CRCBIT) 68 69 if ((sbp->sb_versionnum & V5_VERS_FLAGS) != V5_VERS_FLAGS) 70 return false; 71 if ((sbp->sb_features2 & V5_FEAT_FLAGS) != V5_FEAT_FLAGS) 72 return false; 73 return true; 74 } 75 76 /* 77 * We current support XFS v5 formats with known features and v4 superblocks with 78 * at least V2 directories. 79 */ 80 bool 81 xfs_sb_good_version( 82 struct xfs_sb *sbp) 83 { 84 /* 85 * All v5 filesystems are supported, but we must check that all the 86 * required v4 feature flags are enabled correctly as the code checks 87 * those flags and not for v5 support. 88 */ 89 if (xfs_sb_is_v5(sbp)) 90 return xfs_sb_validate_v5_features(sbp); 91 92 /* versions prior to v4 are not supported */ 93 if (XFS_SB_VERSION_NUM(sbp) != XFS_SB_VERSION_4) 94 return false; 95 96 /* We must not have any unknown v4 feature bits set */ 97 if ((sbp->sb_versionnum & ~XFS_SB_VERSION_OKBITS) || 98 ((sbp->sb_versionnum & XFS_SB_VERSION_MOREBITSBIT) && 99 (sbp->sb_features2 & ~XFS_SB_VERSION2_OKBITS))) 100 return false; 101 102 /* V4 filesystems need v2 directories and unwritten extents */ 103 if (!(sbp->sb_versionnum & XFS_SB_VERSION_DIRV2BIT)) 104 return false; 105 if (!(sbp->sb_versionnum & XFS_SB_VERSION_EXTFLGBIT)) 106 return false; 107 108 /* It's a supported v4 filesystem */ 109 return true; 110 } 111 112 uint64_t 113 xfs_sb_version_to_features( 114 struct xfs_sb *sbp) 115 { 116 uint64_t features = 0; 117 118 /* optional V4 features */ 119 if (sbp->sb_rblocks > 0) 120 features |= XFS_FEAT_REALTIME; 121 if (sbp->sb_versionnum & XFS_SB_VERSION_NLINKBIT) 122 features |= XFS_FEAT_NLINK; 123 if (sbp->sb_versionnum & XFS_SB_VERSION_ATTRBIT) 124 features |= XFS_FEAT_ATTR; 125 if (sbp->sb_versionnum & XFS_SB_VERSION_QUOTABIT) 126 features |= XFS_FEAT_QUOTA; 127 if (sbp->sb_versionnum & XFS_SB_VERSION_ALIGNBIT) 128 features |= XFS_FEAT_ALIGN; 129 if (sbp->sb_versionnum & XFS_SB_VERSION_LOGV2BIT) 130 features |= XFS_FEAT_LOGV2; 131 if (sbp->sb_versionnum & XFS_SB_VERSION_DALIGNBIT) 132 features |= XFS_FEAT_DALIGN; 133 if (sbp->sb_versionnum & XFS_SB_VERSION_EXTFLGBIT) 134 features |= XFS_FEAT_EXTFLG; 135 if (sbp->sb_versionnum & XFS_SB_VERSION_SECTORBIT) 136 features |= XFS_FEAT_SECTOR; 137 if (sbp->sb_versionnum & XFS_SB_VERSION_BORGBIT) 138 features |= XFS_FEAT_ASCIICI; 139 if (sbp->sb_versionnum & XFS_SB_VERSION_MOREBITSBIT) { 140 if (sbp->sb_features2 & XFS_SB_VERSION2_LAZYSBCOUNTBIT) 141 features |= XFS_FEAT_LAZYSBCOUNT; 142 if (sbp->sb_features2 & XFS_SB_VERSION2_ATTR2BIT) 143 features |= XFS_FEAT_ATTR2; 144 if (sbp->sb_features2 & XFS_SB_VERSION2_PROJID32BIT) 145 features |= XFS_FEAT_PROJID32; 146 if (sbp->sb_features2 & XFS_SB_VERSION2_FTYPE) 147 features |= XFS_FEAT_FTYPE; 148 } 149 150 if (!xfs_sb_is_v5(sbp)) 151 return features; 152 153 /* Always on V5 features */ 154 features |= XFS_FEAT_ALIGN | XFS_FEAT_LOGV2 | XFS_FEAT_EXTFLG | 155 XFS_FEAT_LAZYSBCOUNT | XFS_FEAT_ATTR2 | XFS_FEAT_PROJID32 | 156 XFS_FEAT_V3INODES | XFS_FEAT_CRC | XFS_FEAT_PQUOTINO; 157 158 /* Optional V5 features */ 159 if (sbp->sb_features_ro_compat & XFS_SB_FEAT_RO_COMPAT_FINOBT) 160 features |= XFS_FEAT_FINOBT; 161 if (sbp->sb_features_ro_compat & XFS_SB_FEAT_RO_COMPAT_RMAPBT) 162 features |= XFS_FEAT_RMAPBT; 163 if (sbp->sb_features_ro_compat & XFS_SB_FEAT_RO_COMPAT_REFLINK) 164 features |= XFS_FEAT_REFLINK; 165 if (sbp->sb_features_ro_compat & XFS_SB_FEAT_RO_COMPAT_INOBTCNT) 166 features |= XFS_FEAT_INOBTCNT; 167 if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_FTYPE) 168 features |= XFS_FEAT_FTYPE; 169 if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_SPINODES) 170 features |= XFS_FEAT_SPINODES; 171 if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_META_UUID) 172 features |= XFS_FEAT_META_UUID; 173 if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_BIGTIME) 174 features |= XFS_FEAT_BIGTIME; 175 if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_NEEDSREPAIR) 176 features |= XFS_FEAT_NEEDSREPAIR; 177 if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_NREXT64) 178 features |= XFS_FEAT_NREXT64; 179 if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_EXCHRANGE) 180 features |= XFS_FEAT_EXCHANGE_RANGE; 181 if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_PARENT) 182 features |= XFS_FEAT_PARENT; 183 184 return features; 185 } 186 187 /* Check all the superblock fields we care about when reading one in. */ 188 STATIC int 189 xfs_validate_sb_read( 190 struct xfs_mount *mp, 191 struct xfs_sb *sbp) 192 { 193 if (!xfs_sb_is_v5(sbp)) 194 return 0; 195 196 /* 197 * Version 5 superblock feature mask validation. Reject combinations 198 * the kernel cannot support up front before checking anything else. 199 */ 200 if (xfs_sb_has_compat_feature(sbp, XFS_SB_FEAT_COMPAT_UNKNOWN)) { 201 xfs_warn(mp, 202 "Superblock has unknown compatible features (0x%x) enabled.", 203 (sbp->sb_features_compat & XFS_SB_FEAT_COMPAT_UNKNOWN)); 204 xfs_warn(mp, 205 "Using a more recent kernel is recommended."); 206 } 207 208 if (xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) { 209 xfs_alert(mp, 210 "Superblock has unknown read-only compatible features (0x%x) enabled.", 211 (sbp->sb_features_ro_compat & 212 XFS_SB_FEAT_RO_COMPAT_UNKNOWN)); 213 if (!xfs_is_readonly(mp)) { 214 xfs_warn(mp, 215 "Attempted to mount read-only compatible filesystem read-write."); 216 xfs_warn(mp, 217 "Filesystem can only be safely mounted read only."); 218 219 return -EINVAL; 220 } 221 } 222 if (xfs_sb_has_incompat_feature(sbp, XFS_SB_FEAT_INCOMPAT_UNKNOWN)) { 223 xfs_warn(mp, 224 "Superblock has unknown incompatible features (0x%x) enabled.", 225 (sbp->sb_features_incompat & 226 XFS_SB_FEAT_INCOMPAT_UNKNOWN)); 227 xfs_warn(mp, 228 "Filesystem cannot be safely mounted by this kernel."); 229 return -EINVAL; 230 } 231 232 return 0; 233 } 234 235 static uint64_t 236 xfs_sb_calc_rbmblocks( 237 struct xfs_sb *sbp) 238 { 239 return howmany_64(sbp->sb_rextents, NBBY * sbp->sb_blocksize); 240 } 241 242 /* Validate the realtime geometry */ 243 bool 244 xfs_validate_rt_geometry( 245 struct xfs_sb *sbp) 246 { 247 if (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE || 248 sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) 249 return false; 250 251 if (sbp->sb_rblocks == 0) { 252 if (sbp->sb_rextents != 0 || sbp->sb_rbmblocks != 0 || 253 sbp->sb_rextslog != 0 || sbp->sb_frextents != 0) 254 return false; 255 return true; 256 } 257 258 if (sbp->sb_rextents == 0 || 259 sbp->sb_rextents != div_u64(sbp->sb_rblocks, sbp->sb_rextsize) || 260 sbp->sb_rextslog != xfs_compute_rextslog(sbp->sb_rextents) || 261 sbp->sb_rbmblocks != xfs_sb_calc_rbmblocks(sbp)) 262 return false; 263 264 return true; 265 } 266 267 /* Check all the superblock fields we care about when writing one out. */ 268 STATIC int 269 xfs_validate_sb_write( 270 struct xfs_mount *mp, 271 struct xfs_buf *bp, 272 struct xfs_sb *sbp) 273 { 274 /* 275 * Carry out additional sb summary counter sanity checks when we write 276 * the superblock. We skip this in the read validator because there 277 * could be newer superblocks in the log and if the values are garbage 278 * even after replay we'll recalculate them at the end of log mount. 279 * 280 * mkfs has traditionally written zeroed counters to inprogress and 281 * secondary superblocks, so allow this usage to continue because 282 * we never read counters from such superblocks. 283 */ 284 if (xfs_buf_daddr(bp) == XFS_SB_DADDR && !sbp->sb_inprogress && 285 (sbp->sb_fdblocks > sbp->sb_dblocks || 286 !xfs_verify_icount(mp, sbp->sb_icount) || 287 sbp->sb_ifree > sbp->sb_icount)) { 288 xfs_warn(mp, "SB summary counter sanity check failed"); 289 return -EFSCORRUPTED; 290 } 291 292 if (!xfs_sb_is_v5(sbp)) 293 return 0; 294 295 /* 296 * Version 5 superblock feature mask validation. Reject combinations 297 * the kernel cannot support since we checked for unsupported bits in 298 * the read verifier, which means that memory is corrupt. 299 */ 300 if (xfs_sb_has_compat_feature(sbp, XFS_SB_FEAT_COMPAT_UNKNOWN)) { 301 xfs_warn(mp, 302 "Corruption detected in superblock compatible features (0x%x)!", 303 (sbp->sb_features_compat & XFS_SB_FEAT_COMPAT_UNKNOWN)); 304 return -EFSCORRUPTED; 305 } 306 307 if (!xfs_is_readonly(mp) && 308 xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) { 309 xfs_alert(mp, 310 "Corruption detected in superblock read-only compatible features (0x%x)!", 311 (sbp->sb_features_ro_compat & 312 XFS_SB_FEAT_RO_COMPAT_UNKNOWN)); 313 return -EFSCORRUPTED; 314 } 315 if (xfs_sb_has_incompat_feature(sbp, XFS_SB_FEAT_INCOMPAT_UNKNOWN)) { 316 xfs_warn(mp, 317 "Corruption detected in superblock incompatible features (0x%x)!", 318 (sbp->sb_features_incompat & 319 XFS_SB_FEAT_INCOMPAT_UNKNOWN)); 320 return -EFSCORRUPTED; 321 } 322 if (xfs_sb_has_incompat_log_feature(sbp, 323 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) { 324 xfs_warn(mp, 325 "Corruption detected in superblock incompatible log features (0x%x)!", 326 (sbp->sb_features_log_incompat & 327 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)); 328 return -EFSCORRUPTED; 329 } 330 331 /* 332 * We can't read verify the sb LSN because the read verifier is called 333 * before the log is allocated and processed. We know the log is set up 334 * before write verifier calls, so check it here. 335 */ 336 if (!xfs_log_check_lsn(mp, sbp->sb_lsn)) 337 return -EFSCORRUPTED; 338 339 return 0; 340 } 341 342 /* Check the validity of the SB. */ 343 STATIC int 344 xfs_validate_sb_common( 345 struct xfs_mount *mp, 346 struct xfs_buf *bp, 347 struct xfs_sb *sbp) 348 { 349 struct xfs_dsb *dsb = bp->b_addr; 350 uint32_t agcount = 0; 351 uint32_t rem; 352 bool has_dalign; 353 354 if (!xfs_verify_magic(bp, dsb->sb_magicnum)) { 355 xfs_warn(mp, 356 "Superblock has bad magic number 0x%x. Not an XFS filesystem?", 357 be32_to_cpu(dsb->sb_magicnum)); 358 return -EWRONGFS; 359 } 360 361 if (!xfs_sb_good_version(sbp)) { 362 xfs_warn(mp, 363 "Superblock has unknown features enabled or corrupted feature masks."); 364 return -EWRONGFS; 365 } 366 367 /* 368 * Validate feature flags and state 369 */ 370 if (xfs_sb_is_v5(sbp)) { 371 if (sbp->sb_blocksize < XFS_MIN_CRC_BLOCKSIZE) { 372 xfs_notice(mp, 373 "Block size (%u bytes) too small for Version 5 superblock (minimum %d bytes)", 374 sbp->sb_blocksize, XFS_MIN_CRC_BLOCKSIZE); 375 return -EFSCORRUPTED; 376 } 377 378 /* V5 has a separate project quota inode */ 379 if (sbp->sb_qflags & (XFS_OQUOTA_ENFD | XFS_OQUOTA_CHKD)) { 380 xfs_notice(mp, 381 "Version 5 of Super block has XFS_OQUOTA bits."); 382 return -EFSCORRUPTED; 383 } 384 385 /* 386 * Full inode chunks must be aligned to inode chunk size when 387 * sparse inodes are enabled to support the sparse chunk 388 * allocation algorithm and prevent overlapping inode records. 389 */ 390 if (sbp->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_SPINODES) { 391 uint32_t align; 392 393 align = XFS_INODES_PER_CHUNK * sbp->sb_inodesize 394 >> sbp->sb_blocklog; 395 if (sbp->sb_inoalignmt != align) { 396 xfs_warn(mp, 397 "Inode block alignment (%u) must match chunk size (%u) for sparse inodes.", 398 sbp->sb_inoalignmt, align); 399 return -EINVAL; 400 } 401 } 402 } else if (sbp->sb_qflags & (XFS_PQUOTA_ENFD | XFS_GQUOTA_ENFD | 403 XFS_PQUOTA_CHKD | XFS_GQUOTA_CHKD)) { 404 xfs_notice(mp, 405 "Superblock earlier than Version 5 has XFS_{P|G}QUOTA_{ENFD|CHKD} bits."); 406 return -EFSCORRUPTED; 407 } 408 409 if (unlikely( 410 sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) { 411 xfs_warn(mp, 412 "filesystem is marked as having an external log; " 413 "specify logdev on the mount command line."); 414 return -EINVAL; 415 } 416 417 if (unlikely( 418 sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) { 419 xfs_warn(mp, 420 "filesystem is marked as having an internal log; " 421 "do not specify logdev on the mount command line."); 422 return -EINVAL; 423 } 424 425 /* Compute agcount for this number of dblocks and agblocks */ 426 if (sbp->sb_agblocks) { 427 agcount = div_u64_rem(sbp->sb_dblocks, sbp->sb_agblocks, &rem); 428 if (rem) 429 agcount++; 430 } 431 432 /* 433 * More sanity checking. Most of these were stolen directly from 434 * xfs_repair. 435 */ 436 if (unlikely( 437 sbp->sb_agcount <= 0 || 438 sbp->sb_sectsize < XFS_MIN_SECTORSIZE || 439 sbp->sb_sectsize > XFS_MAX_SECTORSIZE || 440 sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG || 441 sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG || 442 sbp->sb_sectsize != (1 << sbp->sb_sectlog) || 443 sbp->sb_blocksize < XFS_MIN_BLOCKSIZE || 444 sbp->sb_blocksize > XFS_MAX_BLOCKSIZE || 445 sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG || 446 sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG || 447 sbp->sb_blocksize != (1 << sbp->sb_blocklog) || 448 sbp->sb_dirblklog + sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG || 449 sbp->sb_inodesize < XFS_DINODE_MIN_SIZE || 450 sbp->sb_inodesize > XFS_DINODE_MAX_SIZE || 451 sbp->sb_inodelog < XFS_DINODE_MIN_LOG || 452 sbp->sb_inodelog > XFS_DINODE_MAX_LOG || 453 sbp->sb_inodesize != (1 << sbp->sb_inodelog) || 454 sbp->sb_inopblock != howmany(sbp->sb_blocksize,sbp->sb_inodesize) || 455 XFS_FSB_TO_B(mp, sbp->sb_agblocks) < XFS_MIN_AG_BYTES || 456 XFS_FSB_TO_B(mp, sbp->sb_agblocks) > XFS_MAX_AG_BYTES || 457 sbp->sb_agblklog != xfs_highbit32(sbp->sb_agblocks - 1) + 1 || 458 agcount == 0 || agcount != sbp->sb_agcount || 459 (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog) || 460 (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE) || 461 (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) || 462 (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */) || 463 sbp->sb_dblocks == 0 || 464 sbp->sb_dblocks > XFS_MAX_DBLOCKS(sbp) || 465 sbp->sb_dblocks < XFS_MIN_DBLOCKS(sbp) || 466 sbp->sb_shared_vn != 0)) { 467 xfs_notice(mp, "SB sanity check failed"); 468 return -EFSCORRUPTED; 469 } 470 471 /* 472 * Logs that are too large are not supported at all. Reject them 473 * outright. Logs that are too small are tolerated on v4 filesystems, 474 * but we can only check that when mounting the log. Hence we skip 475 * those checks here. 476 */ 477 if (sbp->sb_logblocks > XFS_MAX_LOG_BLOCKS) { 478 xfs_notice(mp, 479 "Log size 0x%x blocks too large, maximum size is 0x%llx blocks", 480 sbp->sb_logblocks, XFS_MAX_LOG_BLOCKS); 481 return -EFSCORRUPTED; 482 } 483 484 if (XFS_FSB_TO_B(mp, sbp->sb_logblocks) > XFS_MAX_LOG_BYTES) { 485 xfs_warn(mp, 486 "log size 0x%llx bytes too large, maximum size is 0x%llx bytes", 487 XFS_FSB_TO_B(mp, sbp->sb_logblocks), 488 XFS_MAX_LOG_BYTES); 489 return -EFSCORRUPTED; 490 } 491 492 /* 493 * Do not allow filesystems with corrupted log sector or stripe units to 494 * be mounted. We cannot safely size the iclogs or write to the log if 495 * the log stripe unit is not valid. 496 */ 497 if (sbp->sb_versionnum & XFS_SB_VERSION_SECTORBIT) { 498 if (sbp->sb_logsectsize != (1U << sbp->sb_logsectlog)) { 499 xfs_notice(mp, 500 "log sector size in bytes/log2 (0x%x/0x%x) must match", 501 sbp->sb_logsectsize, 1U << sbp->sb_logsectlog); 502 return -EFSCORRUPTED; 503 } 504 } else if (sbp->sb_logsectsize || sbp->sb_logsectlog) { 505 xfs_notice(mp, 506 "log sector size in bytes/log2 (0x%x/0x%x) are not zero", 507 sbp->sb_logsectsize, sbp->sb_logsectlog); 508 return -EFSCORRUPTED; 509 } 510 511 if (sbp->sb_logsunit > 1) { 512 if (sbp->sb_logsunit % sbp->sb_blocksize) { 513 xfs_notice(mp, 514 "log stripe unit 0x%x bytes must be a multiple of block size", 515 sbp->sb_logsunit); 516 return -EFSCORRUPTED; 517 } 518 if (sbp->sb_logsunit > XLOG_MAX_RECORD_BSIZE) { 519 xfs_notice(mp, 520 "log stripe unit 0x%x bytes over maximum size (0x%x bytes)", 521 sbp->sb_logsunit, XLOG_MAX_RECORD_BSIZE); 522 return -EFSCORRUPTED; 523 } 524 } 525 526 if (!xfs_validate_rt_geometry(sbp)) { 527 xfs_notice(mp, 528 "realtime %sgeometry check failed", 529 sbp->sb_rblocks ? "" : "zeroed "); 530 return -EFSCORRUPTED; 531 } 532 533 /* 534 * Either (sb_unit and !hasdalign) or (!sb_unit and hasdalign) 535 * would imply the image is corrupted. 536 */ 537 has_dalign = sbp->sb_versionnum & XFS_SB_VERSION_DALIGNBIT; 538 if (!!sbp->sb_unit ^ has_dalign) { 539 xfs_notice(mp, "SB stripe alignment sanity check failed"); 540 return -EFSCORRUPTED; 541 } 542 543 if (!xfs_validate_stripe_geometry(mp, XFS_FSB_TO_B(mp, sbp->sb_unit), 544 XFS_FSB_TO_B(mp, sbp->sb_width), 0, 545 xfs_buf_daddr(bp) == XFS_SB_DADDR, false)) 546 return -EFSCORRUPTED; 547 548 /* 549 * Currently only very few inode sizes are supported. 550 */ 551 switch (sbp->sb_inodesize) { 552 case 256: 553 case 512: 554 case 1024: 555 case 2048: 556 break; 557 default: 558 xfs_warn(mp, "inode size of %d bytes not supported", 559 sbp->sb_inodesize); 560 return -ENOSYS; 561 } 562 563 return 0; 564 } 565 566 void 567 xfs_sb_quota_from_disk(struct xfs_sb *sbp) 568 { 569 /* 570 * older mkfs doesn't initialize quota inodes to NULLFSINO. This 571 * leads to in-core values having two different values for a quota 572 * inode to be invalid: 0 and NULLFSINO. Change it to a single value 573 * NULLFSINO. 574 * 575 * Note that this change affect only the in-core values. These 576 * values are not written back to disk unless any quota information 577 * is written to the disk. Even in that case, sb_pquotino field is 578 * not written to disk unless the superblock supports pquotino. 579 */ 580 if (sbp->sb_uquotino == 0) 581 sbp->sb_uquotino = NULLFSINO; 582 if (sbp->sb_gquotino == 0) 583 sbp->sb_gquotino = NULLFSINO; 584 if (sbp->sb_pquotino == 0) 585 sbp->sb_pquotino = NULLFSINO; 586 587 /* 588 * We need to do these manipilations only if we are working 589 * with an older version of on-disk superblock. 590 */ 591 if (xfs_sb_is_v5(sbp)) 592 return; 593 594 if (sbp->sb_qflags & XFS_OQUOTA_ENFD) 595 sbp->sb_qflags |= (sbp->sb_qflags & XFS_PQUOTA_ACCT) ? 596 XFS_PQUOTA_ENFD : XFS_GQUOTA_ENFD; 597 if (sbp->sb_qflags & XFS_OQUOTA_CHKD) 598 sbp->sb_qflags |= (sbp->sb_qflags & XFS_PQUOTA_ACCT) ? 599 XFS_PQUOTA_CHKD : XFS_GQUOTA_CHKD; 600 sbp->sb_qflags &= ~(XFS_OQUOTA_ENFD | XFS_OQUOTA_CHKD); 601 602 if (sbp->sb_qflags & XFS_PQUOTA_ACCT && 603 sbp->sb_gquotino != NULLFSINO) { 604 /* 605 * In older version of superblock, on-disk superblock only 606 * has sb_gquotino, and in-core superblock has both sb_gquotino 607 * and sb_pquotino. But, only one of them is supported at any 608 * point of time. So, if PQUOTA is set in disk superblock, 609 * copy over sb_gquotino to sb_pquotino. The NULLFSINO test 610 * above is to make sure we don't do this twice and wipe them 611 * both out! 612 */ 613 sbp->sb_pquotino = sbp->sb_gquotino; 614 sbp->sb_gquotino = NULLFSINO; 615 } 616 } 617 618 static void 619 __xfs_sb_from_disk( 620 struct xfs_sb *to, 621 struct xfs_dsb *from, 622 bool convert_xquota) 623 { 624 to->sb_magicnum = be32_to_cpu(from->sb_magicnum); 625 to->sb_blocksize = be32_to_cpu(from->sb_blocksize); 626 to->sb_dblocks = be64_to_cpu(from->sb_dblocks); 627 to->sb_rblocks = be64_to_cpu(from->sb_rblocks); 628 to->sb_rextents = be64_to_cpu(from->sb_rextents); 629 memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid)); 630 to->sb_logstart = be64_to_cpu(from->sb_logstart); 631 to->sb_rootino = be64_to_cpu(from->sb_rootino); 632 to->sb_rbmino = be64_to_cpu(from->sb_rbmino); 633 to->sb_rsumino = be64_to_cpu(from->sb_rsumino); 634 to->sb_rextsize = be32_to_cpu(from->sb_rextsize); 635 to->sb_agblocks = be32_to_cpu(from->sb_agblocks); 636 to->sb_agcount = be32_to_cpu(from->sb_agcount); 637 to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks); 638 to->sb_logblocks = be32_to_cpu(from->sb_logblocks); 639 to->sb_versionnum = be16_to_cpu(from->sb_versionnum); 640 to->sb_sectsize = be16_to_cpu(from->sb_sectsize); 641 to->sb_inodesize = be16_to_cpu(from->sb_inodesize); 642 to->sb_inopblock = be16_to_cpu(from->sb_inopblock); 643 memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname)); 644 to->sb_blocklog = from->sb_blocklog; 645 to->sb_sectlog = from->sb_sectlog; 646 to->sb_inodelog = from->sb_inodelog; 647 to->sb_inopblog = from->sb_inopblog; 648 to->sb_agblklog = from->sb_agblklog; 649 to->sb_rextslog = from->sb_rextslog; 650 to->sb_inprogress = from->sb_inprogress; 651 to->sb_imax_pct = from->sb_imax_pct; 652 to->sb_icount = be64_to_cpu(from->sb_icount); 653 to->sb_ifree = be64_to_cpu(from->sb_ifree); 654 to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks); 655 to->sb_frextents = be64_to_cpu(from->sb_frextents); 656 to->sb_uquotino = be64_to_cpu(from->sb_uquotino); 657 to->sb_gquotino = be64_to_cpu(from->sb_gquotino); 658 to->sb_qflags = be16_to_cpu(from->sb_qflags); 659 to->sb_flags = from->sb_flags; 660 to->sb_shared_vn = from->sb_shared_vn; 661 to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt); 662 to->sb_unit = be32_to_cpu(from->sb_unit); 663 to->sb_width = be32_to_cpu(from->sb_width); 664 to->sb_dirblklog = from->sb_dirblklog; 665 to->sb_logsectlog = from->sb_logsectlog; 666 to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize); 667 to->sb_logsunit = be32_to_cpu(from->sb_logsunit); 668 to->sb_features2 = be32_to_cpu(from->sb_features2); 669 to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2); 670 to->sb_features_compat = be32_to_cpu(from->sb_features_compat); 671 to->sb_features_ro_compat = be32_to_cpu(from->sb_features_ro_compat); 672 to->sb_features_incompat = be32_to_cpu(from->sb_features_incompat); 673 to->sb_features_log_incompat = 674 be32_to_cpu(from->sb_features_log_incompat); 675 /* crc is only used on disk, not in memory; just init to 0 here. */ 676 to->sb_crc = 0; 677 to->sb_spino_align = be32_to_cpu(from->sb_spino_align); 678 to->sb_pquotino = be64_to_cpu(from->sb_pquotino); 679 to->sb_lsn = be64_to_cpu(from->sb_lsn); 680 /* 681 * sb_meta_uuid is only on disk if it differs from sb_uuid and the 682 * feature flag is set; if not set we keep it only in memory. 683 */ 684 if (xfs_sb_is_v5(to) && 685 (to->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_META_UUID)) 686 uuid_copy(&to->sb_meta_uuid, &from->sb_meta_uuid); 687 else 688 uuid_copy(&to->sb_meta_uuid, &from->sb_uuid); 689 /* Convert on-disk flags to in-memory flags? */ 690 if (convert_xquota) 691 xfs_sb_quota_from_disk(to); 692 } 693 694 void 695 xfs_sb_from_disk( 696 struct xfs_sb *to, 697 struct xfs_dsb *from) 698 { 699 __xfs_sb_from_disk(to, from, true); 700 } 701 702 static void 703 xfs_sb_quota_to_disk( 704 struct xfs_dsb *to, 705 struct xfs_sb *from) 706 { 707 uint16_t qflags = from->sb_qflags; 708 709 to->sb_uquotino = cpu_to_be64(from->sb_uquotino); 710 711 /* 712 * The in-memory superblock quota state matches the v5 on-disk format so 713 * just write them out and return 714 */ 715 if (xfs_sb_is_v5(from)) { 716 to->sb_qflags = cpu_to_be16(from->sb_qflags); 717 to->sb_gquotino = cpu_to_be64(from->sb_gquotino); 718 to->sb_pquotino = cpu_to_be64(from->sb_pquotino); 719 return; 720 } 721 722 /* 723 * For older superblocks (v4), the in-core version of sb_qflags do not 724 * have XFS_OQUOTA_* flags, whereas the on-disk version does. So, 725 * convert incore XFS_{PG}QUOTA_* flags to on-disk XFS_OQUOTA_* flags. 726 */ 727 qflags &= ~(XFS_PQUOTA_ENFD | XFS_PQUOTA_CHKD | 728 XFS_GQUOTA_ENFD | XFS_GQUOTA_CHKD); 729 730 if (from->sb_qflags & 731 (XFS_PQUOTA_ENFD | XFS_GQUOTA_ENFD)) 732 qflags |= XFS_OQUOTA_ENFD; 733 if (from->sb_qflags & 734 (XFS_PQUOTA_CHKD | XFS_GQUOTA_CHKD)) 735 qflags |= XFS_OQUOTA_CHKD; 736 to->sb_qflags = cpu_to_be16(qflags); 737 738 /* 739 * GQUOTINO and PQUOTINO cannot be used together in versions 740 * of superblock that do not have pquotino. from->sb_flags 741 * tells us which quota is active and should be copied to 742 * disk. If neither are active, we should NULL the inode. 743 * 744 * In all cases, the separate pquotino must remain 0 because it 745 * is beyond the "end" of the valid non-pquotino superblock. 746 */ 747 if (from->sb_qflags & XFS_GQUOTA_ACCT) 748 to->sb_gquotino = cpu_to_be64(from->sb_gquotino); 749 else if (from->sb_qflags & XFS_PQUOTA_ACCT) 750 to->sb_gquotino = cpu_to_be64(from->sb_pquotino); 751 else { 752 /* 753 * We can't rely on just the fields being logged to tell us 754 * that it is safe to write NULLFSINO - we should only do that 755 * if quotas are not actually enabled. Hence only write 756 * NULLFSINO if both in-core quota inodes are NULL. 757 */ 758 if (from->sb_gquotino == NULLFSINO && 759 from->sb_pquotino == NULLFSINO) 760 to->sb_gquotino = cpu_to_be64(NULLFSINO); 761 } 762 763 to->sb_pquotino = 0; 764 } 765 766 void 767 xfs_sb_to_disk( 768 struct xfs_dsb *to, 769 struct xfs_sb *from) 770 { 771 xfs_sb_quota_to_disk(to, from); 772 773 to->sb_magicnum = cpu_to_be32(from->sb_magicnum); 774 to->sb_blocksize = cpu_to_be32(from->sb_blocksize); 775 to->sb_dblocks = cpu_to_be64(from->sb_dblocks); 776 to->sb_rblocks = cpu_to_be64(from->sb_rblocks); 777 to->sb_rextents = cpu_to_be64(from->sb_rextents); 778 memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid)); 779 to->sb_logstart = cpu_to_be64(from->sb_logstart); 780 to->sb_rootino = cpu_to_be64(from->sb_rootino); 781 to->sb_rbmino = cpu_to_be64(from->sb_rbmino); 782 to->sb_rsumino = cpu_to_be64(from->sb_rsumino); 783 to->sb_rextsize = cpu_to_be32(from->sb_rextsize); 784 to->sb_agblocks = cpu_to_be32(from->sb_agblocks); 785 to->sb_agcount = cpu_to_be32(from->sb_agcount); 786 to->sb_rbmblocks = cpu_to_be32(from->sb_rbmblocks); 787 to->sb_logblocks = cpu_to_be32(from->sb_logblocks); 788 to->sb_versionnum = cpu_to_be16(from->sb_versionnum); 789 to->sb_sectsize = cpu_to_be16(from->sb_sectsize); 790 to->sb_inodesize = cpu_to_be16(from->sb_inodesize); 791 to->sb_inopblock = cpu_to_be16(from->sb_inopblock); 792 memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname)); 793 to->sb_blocklog = from->sb_blocklog; 794 to->sb_sectlog = from->sb_sectlog; 795 to->sb_inodelog = from->sb_inodelog; 796 to->sb_inopblog = from->sb_inopblog; 797 to->sb_agblklog = from->sb_agblklog; 798 to->sb_rextslog = from->sb_rextslog; 799 to->sb_inprogress = from->sb_inprogress; 800 to->sb_imax_pct = from->sb_imax_pct; 801 to->sb_icount = cpu_to_be64(from->sb_icount); 802 to->sb_ifree = cpu_to_be64(from->sb_ifree); 803 to->sb_fdblocks = cpu_to_be64(from->sb_fdblocks); 804 to->sb_frextents = cpu_to_be64(from->sb_frextents); 805 806 to->sb_flags = from->sb_flags; 807 to->sb_shared_vn = from->sb_shared_vn; 808 to->sb_inoalignmt = cpu_to_be32(from->sb_inoalignmt); 809 to->sb_unit = cpu_to_be32(from->sb_unit); 810 to->sb_width = cpu_to_be32(from->sb_width); 811 to->sb_dirblklog = from->sb_dirblklog; 812 to->sb_logsectlog = from->sb_logsectlog; 813 to->sb_logsectsize = cpu_to_be16(from->sb_logsectsize); 814 to->sb_logsunit = cpu_to_be32(from->sb_logsunit); 815 816 /* 817 * We need to ensure that bad_features2 always matches features2. 818 * Hence we enforce that here rather than having to remember to do it 819 * everywhere else that updates features2. 820 */ 821 from->sb_bad_features2 = from->sb_features2; 822 to->sb_features2 = cpu_to_be32(from->sb_features2); 823 to->sb_bad_features2 = cpu_to_be32(from->sb_bad_features2); 824 825 if (!xfs_sb_is_v5(from)) 826 return; 827 828 to->sb_features_compat = cpu_to_be32(from->sb_features_compat); 829 to->sb_features_ro_compat = 830 cpu_to_be32(from->sb_features_ro_compat); 831 to->sb_features_incompat = 832 cpu_to_be32(from->sb_features_incompat); 833 to->sb_features_log_incompat = 834 cpu_to_be32(from->sb_features_log_incompat); 835 to->sb_spino_align = cpu_to_be32(from->sb_spino_align); 836 to->sb_lsn = cpu_to_be64(from->sb_lsn); 837 if (from->sb_features_incompat & XFS_SB_FEAT_INCOMPAT_META_UUID) 838 uuid_copy(&to->sb_meta_uuid, &from->sb_meta_uuid); 839 } 840 841 /* 842 * If the superblock has the CRC feature bit set or the CRC field is non-null, 843 * check that the CRC is valid. We check the CRC field is non-null because a 844 * single bit error could clear the feature bit and unused parts of the 845 * superblock are supposed to be zero. Hence a non-null crc field indicates that 846 * we've potentially lost a feature bit and we should check it anyway. 847 * 848 * However, past bugs (i.e. in growfs) left non-zeroed regions beyond the 849 * last field in V4 secondary superblocks. So for secondary superblocks, 850 * we are more forgiving, and ignore CRC failures if the primary doesn't 851 * indicate that the fs version is V5. 852 */ 853 static void 854 xfs_sb_read_verify( 855 struct xfs_buf *bp) 856 { 857 struct xfs_sb sb; 858 struct xfs_mount *mp = bp->b_mount; 859 struct xfs_dsb *dsb = bp->b_addr; 860 int error; 861 862 /* 863 * open code the version check to avoid needing to convert the entire 864 * superblock from disk order just to check the version number 865 */ 866 if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC) && 867 (((be16_to_cpu(dsb->sb_versionnum) & XFS_SB_VERSION_NUMBITS) == 868 XFS_SB_VERSION_5) || 869 dsb->sb_crc != 0)) { 870 871 if (!xfs_buf_verify_cksum(bp, XFS_SB_CRC_OFF)) { 872 /* Only fail bad secondaries on a known V5 filesystem */ 873 if (xfs_buf_daddr(bp) == XFS_SB_DADDR || 874 xfs_has_crc(mp)) { 875 error = -EFSBADCRC; 876 goto out_error; 877 } 878 } 879 } 880 881 /* 882 * Check all the superblock fields. Don't byteswap the xquota flags 883 * because _verify_common checks the on-disk values. 884 */ 885 __xfs_sb_from_disk(&sb, dsb, false); 886 error = xfs_validate_sb_common(mp, bp, &sb); 887 if (error) 888 goto out_error; 889 error = xfs_validate_sb_read(mp, &sb); 890 891 out_error: 892 if (error == -EFSCORRUPTED || error == -EFSBADCRC) 893 xfs_verifier_error(bp, error, __this_address); 894 else if (error) 895 xfs_buf_ioerror(bp, error); 896 } 897 898 /* 899 * We may be probed for a filesystem match, so we may not want to emit 900 * messages when the superblock buffer is not actually an XFS superblock. 901 * If we find an XFS superblock, then run a normal, noisy mount because we are 902 * really going to mount it and want to know about errors. 903 */ 904 static void 905 xfs_sb_quiet_read_verify( 906 struct xfs_buf *bp) 907 { 908 struct xfs_dsb *dsb = bp->b_addr; 909 910 if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC)) { 911 /* XFS filesystem, verify noisily! */ 912 xfs_sb_read_verify(bp); 913 return; 914 } 915 /* quietly fail */ 916 xfs_buf_ioerror(bp, -EWRONGFS); 917 } 918 919 static void 920 xfs_sb_write_verify( 921 struct xfs_buf *bp) 922 { 923 struct xfs_sb sb; 924 struct xfs_mount *mp = bp->b_mount; 925 struct xfs_buf_log_item *bip = bp->b_log_item; 926 struct xfs_dsb *dsb = bp->b_addr; 927 int error; 928 929 /* 930 * Check all the superblock fields. Don't byteswap the xquota flags 931 * because _verify_common checks the on-disk values. 932 */ 933 __xfs_sb_from_disk(&sb, dsb, false); 934 error = xfs_validate_sb_common(mp, bp, &sb); 935 if (error) 936 goto out_error; 937 error = xfs_validate_sb_write(mp, bp, &sb); 938 if (error) 939 goto out_error; 940 941 if (!xfs_sb_is_v5(&sb)) 942 return; 943 944 if (bip) 945 dsb->sb_lsn = cpu_to_be64(bip->bli_item.li_lsn); 946 947 xfs_buf_update_cksum(bp, XFS_SB_CRC_OFF); 948 return; 949 950 out_error: 951 xfs_verifier_error(bp, error, __this_address); 952 } 953 954 const struct xfs_buf_ops xfs_sb_buf_ops = { 955 .name = "xfs_sb", 956 .magic = { cpu_to_be32(XFS_SB_MAGIC), cpu_to_be32(XFS_SB_MAGIC) }, 957 .verify_read = xfs_sb_read_verify, 958 .verify_write = xfs_sb_write_verify, 959 }; 960 961 const struct xfs_buf_ops xfs_sb_quiet_buf_ops = { 962 .name = "xfs_sb_quiet", 963 .magic = { cpu_to_be32(XFS_SB_MAGIC), cpu_to_be32(XFS_SB_MAGIC) }, 964 .verify_read = xfs_sb_quiet_read_verify, 965 .verify_write = xfs_sb_write_verify, 966 }; 967 968 void 969 xfs_mount_sb_set_rextsize( 970 struct xfs_mount *mp, 971 struct xfs_sb *sbp) 972 { 973 mp->m_rtxblklog = log2_if_power2(sbp->sb_rextsize); 974 mp->m_rtxblkmask = mask64_if_power2(sbp->sb_rextsize); 975 } 976 977 /* 978 * xfs_mount_common 979 * 980 * Mount initialization code establishing various mount 981 * fields from the superblock associated with the given 982 * mount structure. 983 * 984 * Inode geometry are calculated in xfs_ialloc_setup_geometry. 985 */ 986 void 987 xfs_sb_mount_common( 988 struct xfs_mount *mp, 989 struct xfs_sb *sbp) 990 { 991 mp->m_agfrotor = 0; 992 atomic_set(&mp->m_agirotor, 0); 993 mp->m_maxagi = mp->m_sb.sb_agcount; 994 mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG; 995 mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT; 996 mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT; 997 mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1; 998 mp->m_blockmask = sbp->sb_blocksize - 1; 999 mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG; 1000 mp->m_blockwmask = mp->m_blockwsize - 1; 1001 xfs_mount_sb_set_rextsize(mp, sbp); 1002 1003 mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, true); 1004 mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, false); 1005 mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2; 1006 mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2; 1007 1008 mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, true); 1009 mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, false); 1010 mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2; 1011 mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2; 1012 1013 mp->m_rmap_mxr[0] = xfs_rmapbt_maxrecs(mp, sbp->sb_blocksize, true); 1014 mp->m_rmap_mxr[1] = xfs_rmapbt_maxrecs(mp, sbp->sb_blocksize, false); 1015 mp->m_rmap_mnr[0] = mp->m_rmap_mxr[0] / 2; 1016 mp->m_rmap_mnr[1] = mp->m_rmap_mxr[1] / 2; 1017 1018 mp->m_refc_mxr[0] = xfs_refcountbt_maxrecs(mp, sbp->sb_blocksize, true); 1019 mp->m_refc_mxr[1] = xfs_refcountbt_maxrecs(mp, sbp->sb_blocksize, false); 1020 mp->m_refc_mnr[0] = mp->m_refc_mxr[0] / 2; 1021 mp->m_refc_mnr[1] = mp->m_refc_mxr[1] / 2; 1022 1023 mp->m_bsize = XFS_FSB_TO_BB(mp, 1); 1024 mp->m_alloc_set_aside = xfs_alloc_set_aside(mp); 1025 mp->m_ag_max_usable = xfs_alloc_ag_max_usable(mp); 1026 } 1027 1028 /* 1029 * xfs_log_sb() can be used to copy arbitrary changes to the in-core superblock 1030 * into the superblock buffer to be logged. It does not provide the higher 1031 * level of locking that is needed to protect the in-core superblock from 1032 * concurrent access. 1033 */ 1034 void 1035 xfs_log_sb( 1036 struct xfs_trans *tp) 1037 { 1038 struct xfs_mount *mp = tp->t_mountp; 1039 struct xfs_buf *bp = xfs_trans_getsb(tp); 1040 1041 /* 1042 * Lazy sb counters don't update the in-core superblock so do that now. 1043 * If this is at unmount, the counters will be exactly correct, but at 1044 * any other time they will only be ballpark correct because of 1045 * reservations that have been taken out percpu counters. If we have an 1046 * unclean shutdown, this will be corrected by log recovery rebuilding 1047 * the counters from the AGF block counts. 1048 * 1049 * Do not update sb_frextents here because it is not part of the lazy 1050 * sb counters, despite having a percpu counter. It is always kept 1051 * consistent with the ondisk rtbitmap by xfs_trans_apply_sb_deltas() 1052 * and hence we don't need have to update it here. 1053 */ 1054 if (xfs_has_lazysbcount(mp)) { 1055 mp->m_sb.sb_icount = percpu_counter_sum_positive(&mp->m_icount); 1056 mp->m_sb.sb_ifree = min_t(uint64_t, 1057 percpu_counter_sum_positive(&mp->m_ifree), 1058 mp->m_sb.sb_icount); 1059 mp->m_sb.sb_fdblocks = 1060 percpu_counter_sum_positive(&mp->m_fdblocks); 1061 } 1062 1063 xfs_sb_to_disk(bp->b_addr, &mp->m_sb); 1064 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF); 1065 xfs_trans_log_buf(tp, bp, 0, sizeof(struct xfs_dsb) - 1); 1066 } 1067 1068 /* 1069 * xfs_sync_sb 1070 * 1071 * Sync the superblock to disk. 1072 * 1073 * Note that the caller is responsible for checking the frozen state of the 1074 * filesystem. This procedure uses the non-blocking transaction allocator and 1075 * thus will allow modifications to a frozen fs. This is required because this 1076 * code can be called during the process of freezing where use of the high-level 1077 * allocator would deadlock. 1078 */ 1079 int 1080 xfs_sync_sb( 1081 struct xfs_mount *mp, 1082 bool wait) 1083 { 1084 struct xfs_trans *tp; 1085 int error; 1086 1087 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_sb, 0, 0, 1088 XFS_TRANS_NO_WRITECOUNT, &tp); 1089 if (error) 1090 return error; 1091 1092 xfs_log_sb(tp); 1093 if (wait) 1094 xfs_trans_set_sync(tp); 1095 return xfs_trans_commit(tp); 1096 } 1097 1098 /* 1099 * Update all the secondary superblocks to match the new state of the primary. 1100 * Because we are completely overwriting all the existing fields in the 1101 * secondary superblock buffers, there is no need to read them in from disk. 1102 * Just get a new buffer, stamp it and write it. 1103 * 1104 * The sb buffers need to be cached here so that we serialise against other 1105 * operations that access the secondary superblocks, but we don't want to keep 1106 * them in memory once it is written so we mark it as a one-shot buffer. 1107 */ 1108 int 1109 xfs_update_secondary_sbs( 1110 struct xfs_mount *mp) 1111 { 1112 struct xfs_perag *pag; 1113 xfs_agnumber_t agno = 1; 1114 int saved_error = 0; 1115 int error = 0; 1116 LIST_HEAD (buffer_list); 1117 1118 /* update secondary superblocks. */ 1119 for_each_perag_from(mp, agno, pag) { 1120 struct xfs_buf *bp; 1121 1122 error = xfs_buf_get(mp->m_ddev_targp, 1123 XFS_AG_DADDR(mp, pag->pag_agno, XFS_SB_DADDR), 1124 XFS_FSS_TO_BB(mp, 1), &bp); 1125 /* 1126 * If we get an error reading or writing alternate superblocks, 1127 * continue. xfs_repair chooses the "best" superblock based 1128 * on most matches; if we break early, we'll leave more 1129 * superblocks un-updated than updated, and xfs_repair may 1130 * pick them over the properly-updated primary. 1131 */ 1132 if (error) { 1133 xfs_warn(mp, 1134 "error allocating secondary superblock for ag %d", 1135 pag->pag_agno); 1136 if (!saved_error) 1137 saved_error = error; 1138 continue; 1139 } 1140 1141 bp->b_ops = &xfs_sb_buf_ops; 1142 xfs_buf_oneshot(bp); 1143 xfs_buf_zero(bp, 0, BBTOB(bp->b_length)); 1144 xfs_sb_to_disk(bp->b_addr, &mp->m_sb); 1145 xfs_buf_delwri_queue(bp, &buffer_list); 1146 xfs_buf_relse(bp); 1147 1148 /* don't hold too many buffers at once */ 1149 if (agno % 16) 1150 continue; 1151 1152 error = xfs_buf_delwri_submit(&buffer_list); 1153 if (error) { 1154 xfs_warn(mp, 1155 "write error %d updating a secondary superblock near ag %d", 1156 error, pag->pag_agno); 1157 if (!saved_error) 1158 saved_error = error; 1159 continue; 1160 } 1161 } 1162 error = xfs_buf_delwri_submit(&buffer_list); 1163 if (error) { 1164 xfs_warn(mp, 1165 "write error %d updating a secondary superblock near ag %d", 1166 error, agno); 1167 } 1168 1169 return saved_error ? saved_error : error; 1170 } 1171 1172 /* 1173 * Same behavior as xfs_sync_sb, except that it is always synchronous and it 1174 * also writes the superblock buffer to disk sector 0 immediately. 1175 */ 1176 int 1177 xfs_sync_sb_buf( 1178 struct xfs_mount *mp) 1179 { 1180 struct xfs_trans *tp; 1181 struct xfs_buf *bp; 1182 int error; 1183 1184 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_sb, 0, 0, 0, &tp); 1185 if (error) 1186 return error; 1187 1188 bp = xfs_trans_getsb(tp); 1189 xfs_log_sb(tp); 1190 xfs_trans_bhold(tp, bp); 1191 xfs_trans_set_sync(tp); 1192 error = xfs_trans_commit(tp); 1193 if (error) 1194 goto out; 1195 /* 1196 * write out the sb buffer to get the changes to disk 1197 */ 1198 error = xfs_bwrite(bp); 1199 out: 1200 xfs_buf_relse(bp); 1201 return error; 1202 } 1203 1204 void 1205 xfs_fs_geometry( 1206 struct xfs_mount *mp, 1207 struct xfs_fsop_geom *geo, 1208 int struct_version) 1209 { 1210 struct xfs_sb *sbp = &mp->m_sb; 1211 1212 memset(geo, 0, sizeof(struct xfs_fsop_geom)); 1213 1214 geo->blocksize = sbp->sb_blocksize; 1215 geo->rtextsize = sbp->sb_rextsize; 1216 geo->agblocks = sbp->sb_agblocks; 1217 geo->agcount = sbp->sb_agcount; 1218 geo->logblocks = sbp->sb_logblocks; 1219 geo->sectsize = sbp->sb_sectsize; 1220 geo->inodesize = sbp->sb_inodesize; 1221 geo->imaxpct = sbp->sb_imax_pct; 1222 geo->datablocks = sbp->sb_dblocks; 1223 geo->rtblocks = sbp->sb_rblocks; 1224 geo->rtextents = sbp->sb_rextents; 1225 geo->logstart = sbp->sb_logstart; 1226 BUILD_BUG_ON(sizeof(geo->uuid) != sizeof(sbp->sb_uuid)); 1227 memcpy(geo->uuid, &sbp->sb_uuid, sizeof(sbp->sb_uuid)); 1228 1229 if (struct_version < 2) 1230 return; 1231 1232 geo->sunit = sbp->sb_unit; 1233 geo->swidth = sbp->sb_width; 1234 1235 if (struct_version < 3) 1236 return; 1237 1238 geo->version = XFS_FSOP_GEOM_VERSION; 1239 geo->flags = XFS_FSOP_GEOM_FLAGS_NLINK | 1240 XFS_FSOP_GEOM_FLAGS_DIRV2 | 1241 XFS_FSOP_GEOM_FLAGS_EXTFLG; 1242 if (xfs_has_attr(mp)) 1243 geo->flags |= XFS_FSOP_GEOM_FLAGS_ATTR; 1244 if (xfs_has_quota(mp)) 1245 geo->flags |= XFS_FSOP_GEOM_FLAGS_QUOTA; 1246 if (xfs_has_align(mp)) 1247 geo->flags |= XFS_FSOP_GEOM_FLAGS_IALIGN; 1248 if (xfs_has_dalign(mp)) 1249 geo->flags |= XFS_FSOP_GEOM_FLAGS_DALIGN; 1250 if (xfs_has_asciici(mp)) 1251 geo->flags |= XFS_FSOP_GEOM_FLAGS_DIRV2CI; 1252 if (xfs_has_lazysbcount(mp)) 1253 geo->flags |= XFS_FSOP_GEOM_FLAGS_LAZYSB; 1254 if (xfs_has_attr2(mp)) 1255 geo->flags |= XFS_FSOP_GEOM_FLAGS_ATTR2; 1256 if (xfs_has_projid32(mp)) 1257 geo->flags |= XFS_FSOP_GEOM_FLAGS_PROJID32; 1258 if (xfs_has_crc(mp)) 1259 geo->flags |= XFS_FSOP_GEOM_FLAGS_V5SB; 1260 if (xfs_has_ftype(mp)) 1261 geo->flags |= XFS_FSOP_GEOM_FLAGS_FTYPE; 1262 if (xfs_has_finobt(mp)) 1263 geo->flags |= XFS_FSOP_GEOM_FLAGS_FINOBT; 1264 if (xfs_has_sparseinodes(mp)) 1265 geo->flags |= XFS_FSOP_GEOM_FLAGS_SPINODES; 1266 if (xfs_has_rmapbt(mp)) 1267 geo->flags |= XFS_FSOP_GEOM_FLAGS_RMAPBT; 1268 if (xfs_has_reflink(mp)) 1269 geo->flags |= XFS_FSOP_GEOM_FLAGS_REFLINK; 1270 if (xfs_has_bigtime(mp)) 1271 geo->flags |= XFS_FSOP_GEOM_FLAGS_BIGTIME; 1272 if (xfs_has_inobtcounts(mp)) 1273 geo->flags |= XFS_FSOP_GEOM_FLAGS_INOBTCNT; 1274 if (xfs_has_parent(mp)) 1275 geo->flags |= XFS_FSOP_GEOM_FLAGS_PARENT; 1276 if (xfs_has_sector(mp)) { 1277 geo->flags |= XFS_FSOP_GEOM_FLAGS_SECTOR; 1278 geo->logsectsize = sbp->sb_logsectsize; 1279 } else { 1280 geo->logsectsize = BBSIZE; 1281 } 1282 if (xfs_has_large_extent_counts(mp)) 1283 geo->flags |= XFS_FSOP_GEOM_FLAGS_NREXT64; 1284 if (xfs_has_exchange_range(mp)) 1285 geo->flags |= XFS_FSOP_GEOM_FLAGS_EXCHANGE_RANGE; 1286 geo->rtsectsize = sbp->sb_blocksize; 1287 geo->dirblocksize = xfs_dir2_dirblock_bytes(sbp); 1288 1289 if (struct_version < 4) 1290 return; 1291 1292 if (xfs_has_logv2(mp)) 1293 geo->flags |= XFS_FSOP_GEOM_FLAGS_LOGV2; 1294 1295 geo->logsunit = sbp->sb_logsunit; 1296 1297 if (struct_version < 5) 1298 return; 1299 1300 geo->version = XFS_FSOP_GEOM_VERSION_V5; 1301 } 1302 1303 /* Read a secondary superblock. */ 1304 int 1305 xfs_sb_read_secondary( 1306 struct xfs_mount *mp, 1307 struct xfs_trans *tp, 1308 xfs_agnumber_t agno, 1309 struct xfs_buf **bpp) 1310 { 1311 struct xfs_buf *bp; 1312 int error; 1313 1314 ASSERT(agno != 0 && agno != NULLAGNUMBER); 1315 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, 1316 XFS_AG_DADDR(mp, agno, XFS_SB_BLOCK(mp)), 1317 XFS_FSS_TO_BB(mp, 1), 0, &bp, &xfs_sb_buf_ops); 1318 if (xfs_metadata_is_sick(error)) 1319 xfs_agno_mark_sick(mp, agno, XFS_SICK_AG_SB); 1320 if (error) 1321 return error; 1322 xfs_buf_set_ref(bp, XFS_SSB_REF); 1323 *bpp = bp; 1324 return 0; 1325 } 1326 1327 /* Get an uninitialised secondary superblock buffer. */ 1328 int 1329 xfs_sb_get_secondary( 1330 struct xfs_mount *mp, 1331 struct xfs_trans *tp, 1332 xfs_agnumber_t agno, 1333 struct xfs_buf **bpp) 1334 { 1335 struct xfs_buf *bp; 1336 int error; 1337 1338 ASSERT(agno != 0 && agno != NULLAGNUMBER); 1339 error = xfs_trans_get_buf(tp, mp->m_ddev_targp, 1340 XFS_AG_DADDR(mp, agno, XFS_SB_BLOCK(mp)), 1341 XFS_FSS_TO_BB(mp, 1), 0, &bp); 1342 if (error) 1343 return error; 1344 bp->b_ops = &xfs_sb_buf_ops; 1345 xfs_buf_oneshot(bp); 1346 *bpp = bp; 1347 return 0; 1348 } 1349 1350 /* 1351 * sunit, swidth, sectorsize(optional with 0) should be all in bytes, so users 1352 * won't be confused by values in error messages. This function returns false 1353 * if the stripe geometry is invalid and the caller is unable to repair the 1354 * stripe configuration later in the mount process. 1355 */ 1356 bool 1357 xfs_validate_stripe_geometry( 1358 struct xfs_mount *mp, 1359 __s64 sunit, 1360 __s64 swidth, 1361 int sectorsize, 1362 bool may_repair, 1363 bool silent) 1364 { 1365 if (swidth > INT_MAX) { 1366 if (!silent) 1367 xfs_notice(mp, 1368 "stripe width (%lld) is too large", swidth); 1369 goto check_override; 1370 } 1371 1372 if (sunit > swidth) { 1373 if (!silent) 1374 xfs_notice(mp, 1375 "stripe unit (%lld) is larger than the stripe width (%lld)", sunit, swidth); 1376 goto check_override; 1377 } 1378 1379 if (sectorsize && (int)sunit % sectorsize) { 1380 if (!silent) 1381 xfs_notice(mp, 1382 "stripe unit (%lld) must be a multiple of the sector size (%d)", 1383 sunit, sectorsize); 1384 goto check_override; 1385 } 1386 1387 if (sunit && !swidth) { 1388 if (!silent) 1389 xfs_notice(mp, 1390 "invalid stripe unit (%lld) and stripe width of 0", sunit); 1391 goto check_override; 1392 } 1393 1394 if (!sunit && swidth) { 1395 if (!silent) 1396 xfs_notice(mp, 1397 "invalid stripe width (%lld) and stripe unit of 0", swidth); 1398 goto check_override; 1399 } 1400 1401 if (sunit && (int)swidth % (int)sunit) { 1402 if (!silent) 1403 xfs_notice(mp, 1404 "stripe width (%lld) must be a multiple of the stripe unit (%lld)", 1405 swidth, sunit); 1406 goto check_override; 1407 } 1408 return true; 1409 1410 check_override: 1411 if (!may_repair) 1412 return false; 1413 /* 1414 * During mount, mp->m_dalign will not be set unless the sunit mount 1415 * option was set. If it was set, ignore the bad stripe alignment values 1416 * and allow the validation and overwrite later in the mount process to 1417 * attempt to overwrite the bad stripe alignment values with the values 1418 * supplied by mount options. 1419 */ 1420 if (!mp->m_dalign) 1421 return false; 1422 if (!silent) 1423 xfs_notice(mp, 1424 "Will try to correct with specified mount options sunit (%d) and swidth (%d)", 1425 BBTOB(mp->m_dalign), BBTOB(mp->m_swidth)); 1426 return true; 1427 } 1428 1429 /* 1430 * Compute the maximum level number of the realtime summary file, as defined by 1431 * mkfs. The historic use of highbit32 on a 64-bit quantity prohibited correct 1432 * use of rt volumes with more than 2^32 extents. 1433 */ 1434 uint8_t 1435 xfs_compute_rextslog( 1436 xfs_rtbxlen_t rtextents) 1437 { 1438 if (!rtextents) 1439 return 0; 1440 return xfs_highbit64(rtextents); 1441 } 1442