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_sb.h" 13 #include "xfs_mount.h" 14 #include "xfs_trans.h" 15 #include "xfs_error.h" 16 #include "xfs_alloc.h" 17 #include "xfs_fsops.h" 18 #include "xfs_trans_space.h" 19 #include "xfs_log.h" 20 #include "xfs_log_priv.h" 21 #include "xfs_ag.h" 22 #include "xfs_ag_resv.h" 23 #include "xfs_trace.h" 24 25 /* 26 * Write new AG headers to disk. Non-transactional, but need to be 27 * written and completed prior to the growfs transaction being logged. 28 * To do this, we use a delayed write buffer list and wait for 29 * submission and IO completion of the list as a whole. This allows the 30 * IO subsystem to merge all the AG headers in a single AG into a single 31 * IO and hide most of the latency of the IO from us. 32 * 33 * This also means that if we get an error whilst building the buffer 34 * list to write, we can cancel the entire list without having written 35 * anything. 36 */ 37 static int 38 xfs_resizefs_init_new_ags( 39 struct xfs_trans *tp, 40 struct aghdr_init_data *id, 41 xfs_agnumber_t oagcount, 42 xfs_agnumber_t nagcount, 43 xfs_rfsblock_t delta, 44 struct xfs_perag *last_pag, 45 bool *lastag_extended) 46 { 47 struct xfs_mount *mp = tp->t_mountp; 48 xfs_rfsblock_t nb = mp->m_sb.sb_dblocks + delta; 49 int error; 50 51 *lastag_extended = false; 52 53 INIT_LIST_HEAD(&id->buffer_list); 54 for (id->agno = nagcount - 1; 55 id->agno >= oagcount; 56 id->agno--, delta -= id->agsize) { 57 58 if (id->agno == nagcount - 1) 59 id->agsize = nb - (id->agno * 60 (xfs_rfsblock_t)mp->m_sb.sb_agblocks); 61 else 62 id->agsize = mp->m_sb.sb_agblocks; 63 64 error = xfs_ag_init_headers(mp, id); 65 if (error) { 66 xfs_buf_delwri_cancel(&id->buffer_list); 67 return error; 68 } 69 } 70 71 error = xfs_buf_delwri_submit(&id->buffer_list); 72 if (error) 73 return error; 74 75 if (delta) { 76 *lastag_extended = true; 77 error = xfs_ag_extend_space(last_pag, tp, delta); 78 } 79 return error; 80 } 81 82 /* 83 * growfs operations 84 */ 85 static int 86 xfs_growfs_data_private( 87 struct xfs_mount *mp, /* mount point for filesystem */ 88 struct xfs_growfs_data *in) /* growfs data input struct */ 89 { 90 xfs_agnumber_t oagcount = mp->m_sb.sb_agcount; 91 struct xfs_buf *bp; 92 int error; 93 xfs_agnumber_t nagcount; 94 xfs_agnumber_t nagimax = 0; 95 xfs_rfsblock_t nb, nb_div, nb_mod; 96 int64_t delta; 97 bool lastag_extended = false; 98 struct xfs_trans *tp; 99 struct aghdr_init_data id = {}; 100 struct xfs_perag *last_pag; 101 102 nb = in->newblocks; 103 error = xfs_sb_validate_fsb_count(&mp->m_sb, nb); 104 if (error) 105 return error; 106 107 if (nb > mp->m_sb.sb_dblocks) { 108 error = xfs_buf_read_uncached(mp->m_ddev_targp, 109 XFS_FSB_TO_BB(mp, nb) - XFS_FSS_TO_BB(mp, 1), 110 XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL); 111 if (error) 112 return error; 113 xfs_buf_relse(bp); 114 } 115 116 nb_div = nb; 117 nb_mod = do_div(nb_div, mp->m_sb.sb_agblocks); 118 if (nb_mod && nb_mod >= XFS_MIN_AG_BLOCKS) 119 nb_div++; 120 else if (nb_mod) 121 nb = nb_div * mp->m_sb.sb_agblocks; 122 123 if (nb_div > XFS_MAX_AGNUMBER + 1) { 124 nb_div = XFS_MAX_AGNUMBER + 1; 125 nb = nb_div * mp->m_sb.sb_agblocks; 126 } 127 nagcount = nb_div; 128 delta = nb - mp->m_sb.sb_dblocks; 129 /* 130 * Reject filesystems with a single AG because they are not 131 * supported, and reject a shrink operation that would cause a 132 * filesystem to become unsupported. 133 */ 134 if (delta < 0 && nagcount < 2) 135 return -EINVAL; 136 137 /* No work to do */ 138 if (delta == 0) 139 return 0; 140 141 /* TODO: shrinking the entire AGs hasn't yet completed */ 142 if (nagcount < oagcount) 143 return -EINVAL; 144 145 /* allocate the new per-ag structures */ 146 error = xfs_initialize_perag(mp, oagcount, nagcount, nb, &nagimax); 147 if (error) 148 return error; 149 150 if (delta > 0) 151 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata, 152 XFS_GROWFS_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, 153 &tp); 154 else 155 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata, -delta, 0, 156 0, &tp); 157 if (error) 158 goto out_free_unused_perag; 159 160 last_pag = xfs_perag_get(mp, oagcount - 1); 161 if (delta > 0) { 162 error = xfs_resizefs_init_new_ags(tp, &id, oagcount, nagcount, 163 delta, last_pag, &lastag_extended); 164 } else { 165 xfs_warn_experimental(mp, XFS_EXPERIMENTAL_SHRINK); 166 error = xfs_ag_shrink_space(last_pag, &tp, -delta); 167 } 168 xfs_perag_put(last_pag); 169 if (error) 170 goto out_trans_cancel; 171 172 /* 173 * Update changed superblock fields transactionally. These are not 174 * seen by the rest of the world until the transaction commit applies 175 * them atomically to the superblock. 176 */ 177 if (nagcount > oagcount) 178 xfs_trans_mod_sb(tp, XFS_TRANS_SB_AGCOUNT, nagcount - oagcount); 179 if (delta) 180 xfs_trans_mod_sb(tp, XFS_TRANS_SB_DBLOCKS, delta); 181 if (id.nfree) 182 xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, id.nfree); 183 184 /* 185 * Sync sb counters now to reflect the updated values. This is 186 * particularly important for shrink because the write verifier 187 * will fail if sb_fdblocks is ever larger than sb_dblocks. 188 */ 189 if (xfs_has_lazysbcount(mp)) 190 xfs_log_sb(tp); 191 192 xfs_trans_set_sync(tp); 193 error = xfs_trans_commit(tp); 194 if (error) 195 return error; 196 197 /* New allocation groups fully initialized, so update mount struct */ 198 if (nagimax) 199 mp->m_maxagi = nagimax; 200 xfs_set_low_space_thresholds(mp); 201 mp->m_alloc_set_aside = xfs_alloc_set_aside(mp); 202 203 if (delta > 0) { 204 /* 205 * If we expanded the last AG, free the per-AG reservation 206 * so we can reinitialize it with the new size. 207 */ 208 if (lastag_extended) { 209 struct xfs_perag *pag; 210 211 pag = xfs_perag_get(mp, id.agno); 212 xfs_ag_resv_free(pag); 213 xfs_perag_put(pag); 214 } 215 /* 216 * Reserve AG metadata blocks. ENOSPC here does not mean there 217 * was a growfs failure, just that there still isn't space for 218 * new user data after the grow has been run. 219 */ 220 error = xfs_fs_reserve_ag_blocks(mp); 221 if (error == -ENOSPC) 222 error = 0; 223 } 224 return error; 225 226 out_trans_cancel: 227 xfs_trans_cancel(tp); 228 out_free_unused_perag: 229 if (nagcount > oagcount) 230 xfs_free_perag_range(mp, oagcount, nagcount); 231 return error; 232 } 233 234 static int 235 xfs_growfs_log_private( 236 struct xfs_mount *mp, /* mount point for filesystem */ 237 struct xfs_growfs_log *in) /* growfs log input struct */ 238 { 239 xfs_extlen_t nb; 240 241 nb = in->newblocks; 242 if (nb < XFS_MIN_LOG_BLOCKS || nb < XFS_B_TO_FSB(mp, XFS_MIN_LOG_BYTES)) 243 return -EINVAL; 244 if (nb == mp->m_sb.sb_logblocks && 245 in->isint == (mp->m_sb.sb_logstart != 0)) 246 return -EINVAL; 247 /* 248 * Moving the log is hard, need new interfaces to sync 249 * the log first, hold off all activity while moving it. 250 * Can have shorter or longer log in the same space, 251 * or transform internal to external log or vice versa. 252 */ 253 return -ENOSYS; 254 } 255 256 static int 257 xfs_growfs_imaxpct( 258 struct xfs_mount *mp, 259 __u32 imaxpct) 260 { 261 struct xfs_trans *tp; 262 int dpct; 263 int error; 264 265 if (imaxpct > 100) 266 return -EINVAL; 267 268 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata, 269 XFS_GROWFS_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, &tp); 270 if (error) 271 return error; 272 273 dpct = imaxpct - mp->m_sb.sb_imax_pct; 274 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IMAXPCT, dpct); 275 xfs_trans_set_sync(tp); 276 return xfs_trans_commit(tp); 277 } 278 279 /* 280 * protected versions of growfs function acquire and release locks on the mount 281 * point - exported through ioctls: XFS_IOC_FSGROWFSDATA, XFS_IOC_FSGROWFSLOG, 282 * XFS_IOC_FSGROWFSRT 283 */ 284 int 285 xfs_growfs_data( 286 struct xfs_mount *mp, 287 struct xfs_growfs_data *in) 288 { 289 int error = 0; 290 291 if (!capable(CAP_SYS_ADMIN)) 292 return -EPERM; 293 if (!mutex_trylock(&mp->m_growlock)) 294 return -EWOULDBLOCK; 295 296 /* update imaxpct separately to the physical grow of the filesystem */ 297 if (in->imaxpct != mp->m_sb.sb_imax_pct) { 298 error = xfs_growfs_imaxpct(mp, in->imaxpct); 299 if (error) 300 goto out_error; 301 } 302 303 if (in->newblocks != mp->m_sb.sb_dblocks) { 304 error = xfs_growfs_data_private(mp, in); 305 if (error) 306 goto out_error; 307 } 308 309 /* Post growfs calculations needed to reflect new state in operations */ 310 if (mp->m_sb.sb_imax_pct) { 311 uint64_t icount = mp->m_sb.sb_dblocks * mp->m_sb.sb_imax_pct; 312 do_div(icount, 100); 313 M_IGEO(mp)->maxicount = XFS_FSB_TO_INO(mp, icount); 314 } else 315 M_IGEO(mp)->maxicount = 0; 316 317 /* Update secondary superblocks now the physical grow has completed */ 318 error = xfs_update_secondary_sbs(mp); 319 320 out_error: 321 /* 322 * Increment the generation unconditionally, the error could be from 323 * updating the secondary superblocks, in which case the new size 324 * is live already. 325 */ 326 mp->m_generation++; 327 mutex_unlock(&mp->m_growlock); 328 return error; 329 } 330 331 int 332 xfs_growfs_log( 333 xfs_mount_t *mp, 334 struct xfs_growfs_log *in) 335 { 336 int error; 337 338 if (!capable(CAP_SYS_ADMIN)) 339 return -EPERM; 340 if (!mutex_trylock(&mp->m_growlock)) 341 return -EWOULDBLOCK; 342 error = xfs_growfs_log_private(mp, in); 343 mutex_unlock(&mp->m_growlock); 344 return error; 345 } 346 347 /* 348 * Reserve the requested number of blocks if available. Otherwise return 349 * as many as possible to satisfy the request. The actual number 350 * reserved are returned in outval. 351 */ 352 int 353 xfs_reserve_blocks( 354 struct xfs_mount *mp, 355 uint64_t request) 356 { 357 int64_t lcounter, delta; 358 int64_t fdblks_delta = 0; 359 int64_t free; 360 int error = 0; 361 362 /* 363 * With per-cpu counters, this becomes an interesting problem. we need 364 * to work out if we are freeing or allocation blocks first, then we can 365 * do the modification as necessary. 366 * 367 * We do this under the m_sb_lock so that if we are near ENOSPC, we will 368 * hold out any changes while we work out what to do. This means that 369 * the amount of free space can change while we do this, so we need to 370 * retry if we end up trying to reserve more space than is available. 371 */ 372 spin_lock(&mp->m_sb_lock); 373 374 /* 375 * If our previous reservation was larger than the current value, 376 * then move any unused blocks back to the free pool. Modify the resblks 377 * counters directly since we shouldn't have any problems unreserving 378 * space. 379 */ 380 if (mp->m_resblks > request) { 381 lcounter = mp->m_resblks_avail - request; 382 if (lcounter > 0) { /* release unused blocks */ 383 fdblks_delta = lcounter; 384 mp->m_resblks_avail -= lcounter; 385 } 386 mp->m_resblks = request; 387 if (fdblks_delta) { 388 spin_unlock(&mp->m_sb_lock); 389 xfs_add_fdblocks(mp, fdblks_delta); 390 spin_lock(&mp->m_sb_lock); 391 } 392 393 goto out; 394 } 395 396 /* 397 * If the request is larger than the current reservation, reserve the 398 * blocks before we update the reserve counters. Sample m_fdblocks and 399 * perform a partial reservation if the request exceeds free space. 400 * 401 * The code below estimates how many blocks it can request from 402 * fdblocks to stash in the reserve pool. This is a classic TOCTOU 403 * race since fdblocks updates are not always coordinated via 404 * m_sb_lock. Set the reserve size even if there's not enough free 405 * space to fill it because mod_fdblocks will refill an undersized 406 * reserve when it can. 407 */ 408 free = percpu_counter_sum(&mp->m_fdblocks) - 409 xfs_fdblocks_unavailable(mp); 410 delta = request - mp->m_resblks; 411 mp->m_resblks = request; 412 if (delta > 0 && free > 0) { 413 /* 414 * We'll either succeed in getting space from the free block 415 * count or we'll get an ENOSPC. Don't set the reserved flag 416 * here - we don't want to reserve the extra reserve blocks 417 * from the reserve. 418 * 419 * The desired reserve size can change after we drop the lock. 420 * Use mod_fdblocks to put the space into the reserve or into 421 * fdblocks as appropriate. 422 */ 423 fdblks_delta = min(free, delta); 424 spin_unlock(&mp->m_sb_lock); 425 error = xfs_dec_fdblocks(mp, fdblks_delta, 0); 426 if (!error) 427 xfs_add_fdblocks(mp, fdblks_delta); 428 spin_lock(&mp->m_sb_lock); 429 } 430 out: 431 spin_unlock(&mp->m_sb_lock); 432 return error; 433 } 434 435 int 436 xfs_fs_goingdown( 437 xfs_mount_t *mp, 438 uint32_t inflags) 439 { 440 switch (inflags) { 441 case XFS_FSOP_GOING_FLAGS_DEFAULT: { 442 if (!bdev_freeze(mp->m_super->s_bdev)) { 443 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT); 444 bdev_thaw(mp->m_super->s_bdev); 445 } 446 break; 447 } 448 case XFS_FSOP_GOING_FLAGS_LOGFLUSH: 449 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT); 450 break; 451 case XFS_FSOP_GOING_FLAGS_NOLOGFLUSH: 452 xfs_force_shutdown(mp, 453 SHUTDOWN_FORCE_UMOUNT | SHUTDOWN_LOG_IO_ERROR); 454 break; 455 default: 456 return -EINVAL; 457 } 458 459 return 0; 460 } 461 462 /* 463 * Force a shutdown of the filesystem instantly while keeping the filesystem 464 * consistent. We don't do an unmount here; just shutdown the shop, make sure 465 * that absolutely nothing persistent happens to this filesystem after this 466 * point. 467 * 468 * The shutdown state change is atomic, resulting in the first and only the 469 * first shutdown call processing the shutdown. This means we only shutdown the 470 * log once as it requires, and we don't spam the logs when multiple concurrent 471 * shutdowns race to set the shutdown flags. 472 */ 473 void 474 xfs_do_force_shutdown( 475 struct xfs_mount *mp, 476 uint32_t flags, 477 char *fname, 478 int lnnum) 479 { 480 int tag; 481 const char *why; 482 483 484 if (xfs_set_shutdown(mp)) { 485 xlog_shutdown_wait(mp->m_log); 486 return; 487 } 488 if (mp->m_sb_bp) 489 mp->m_sb_bp->b_flags |= XBF_DONE; 490 491 if (flags & SHUTDOWN_FORCE_UMOUNT) 492 xfs_alert(mp, "User initiated shutdown received."); 493 494 if (xlog_force_shutdown(mp->m_log, flags)) { 495 tag = XFS_PTAG_SHUTDOWN_LOGERROR; 496 why = "Log I/O Error"; 497 } else if (flags & SHUTDOWN_CORRUPT_INCORE) { 498 tag = XFS_PTAG_SHUTDOWN_CORRUPT; 499 why = "Corruption of in-memory data"; 500 } else if (flags & SHUTDOWN_CORRUPT_ONDISK) { 501 tag = XFS_PTAG_SHUTDOWN_CORRUPT; 502 why = "Corruption of on-disk metadata"; 503 } else if (flags & SHUTDOWN_DEVICE_REMOVED) { 504 tag = XFS_PTAG_SHUTDOWN_IOERROR; 505 why = "Block device removal"; 506 } else { 507 tag = XFS_PTAG_SHUTDOWN_IOERROR; 508 why = "Metadata I/O Error"; 509 } 510 511 trace_xfs_force_shutdown(mp, tag, flags, fname, lnnum); 512 513 xfs_alert_tag(mp, tag, 514 "%s (0x%x) detected at %pS (%s:%d). Shutting down filesystem.", 515 why, flags, __return_address, fname, lnnum); 516 xfs_alert(mp, 517 "Please unmount the filesystem and rectify the problem(s)"); 518 if (xfs_error_level >= XFS_ERRLEVEL_HIGH) 519 xfs_stack_trace(); 520 } 521 522 /* 523 * Reserve free space for per-AG metadata. 524 */ 525 int 526 xfs_fs_reserve_ag_blocks( 527 struct xfs_mount *mp) 528 { 529 struct xfs_perag *pag = NULL; 530 int error = 0; 531 int err2; 532 533 mp->m_finobt_nores = false; 534 while ((pag = xfs_perag_next(mp, pag))) { 535 err2 = xfs_ag_resv_init(pag, NULL); 536 if (err2 && !error) 537 error = err2; 538 } 539 540 if (error && error != -ENOSPC) { 541 xfs_warn(mp, 542 "Error %d reserving per-AG metadata reserve pool.", error); 543 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); 544 } 545 546 return error; 547 } 548 549 /* 550 * Free space reserved for per-AG metadata. 551 */ 552 void 553 xfs_fs_unreserve_ag_blocks( 554 struct xfs_mount *mp) 555 { 556 struct xfs_perag *pag = NULL; 557 558 while ((pag = xfs_perag_next(mp, pag))) 559 xfs_ag_resv_free(pag); 560 } 561