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 struct xfs_buf *bp; 91 int error; 92 xfs_agnumber_t nagcount; 93 xfs_agnumber_t nagimax = 0; 94 xfs_rfsblock_t nb, nb_div, nb_mod; 95 int64_t delta; 96 bool lastag_extended = false; 97 xfs_agnumber_t oagcount; 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 oagcount = mp->m_sb.sb_agcount; 142 /* allocate the new per-ag structures */ 143 if (nagcount > oagcount) { 144 error = xfs_initialize_perag(mp, nagcount, nb, &nagimax); 145 if (error) 146 return error; 147 } else if (nagcount < oagcount) { 148 /* TODO: shrinking the entire AGs hasn't yet completed */ 149 return -EINVAL; 150 } 151 152 if (delta > 0) 153 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata, 154 XFS_GROWFS_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, 155 &tp); 156 else 157 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata, -delta, 0, 158 0, &tp); 159 if (error) 160 goto out_free_unused_perag; 161 162 last_pag = xfs_perag_get(mp, oagcount - 1); 163 if (delta > 0) { 164 error = xfs_resizefs_init_new_ags(tp, &id, oagcount, nagcount, 165 delta, last_pag, &lastag_extended); 166 } else { 167 xfs_warn_mount(mp, XFS_OPSTATE_WARNED_SHRINK, 168 "EXPERIMENTAL online shrink feature in use. Use at your own risk!"); 169 170 error = xfs_ag_shrink_space(last_pag, &tp, -delta); 171 } 172 xfs_perag_put(last_pag); 173 if (error) 174 goto out_trans_cancel; 175 176 /* 177 * Update changed superblock fields transactionally. These are not 178 * seen by the rest of the world until the transaction commit applies 179 * them atomically to the superblock. 180 */ 181 if (nagcount > oagcount) 182 xfs_trans_mod_sb(tp, XFS_TRANS_SB_AGCOUNT, nagcount - oagcount); 183 if (delta) 184 xfs_trans_mod_sb(tp, XFS_TRANS_SB_DBLOCKS, delta); 185 if (id.nfree) 186 xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, id.nfree); 187 188 /* 189 * Sync sb counters now to reflect the updated values. This is 190 * particularly important for shrink because the write verifier 191 * will fail if sb_fdblocks is ever larger than sb_dblocks. 192 */ 193 if (xfs_has_lazysbcount(mp)) 194 xfs_log_sb(tp); 195 196 xfs_trans_set_sync(tp); 197 error = xfs_trans_commit(tp); 198 if (error) 199 return error; 200 201 /* New allocation groups fully initialized, so update mount struct */ 202 if (nagimax) 203 mp->m_maxagi = nagimax; 204 xfs_set_low_space_thresholds(mp); 205 mp->m_alloc_set_aside = xfs_alloc_set_aside(mp); 206 207 if (delta > 0) { 208 /* 209 * If we expanded the last AG, free the per-AG reservation 210 * so we can reinitialize it with the new size. 211 */ 212 if (lastag_extended) { 213 struct xfs_perag *pag; 214 215 pag = xfs_perag_get(mp, id.agno); 216 xfs_ag_resv_free(pag); 217 xfs_perag_put(pag); 218 } 219 /* 220 * Reserve AG metadata blocks. ENOSPC here does not mean there 221 * was a growfs failure, just that there still isn't space for 222 * new user data after the grow has been run. 223 */ 224 error = xfs_fs_reserve_ag_blocks(mp); 225 if (error == -ENOSPC) 226 error = 0; 227 } 228 return error; 229 230 out_trans_cancel: 231 xfs_trans_cancel(tp); 232 out_free_unused_perag: 233 if (nagcount > oagcount) 234 xfs_free_unused_perag_range(mp, oagcount, nagcount); 235 return error; 236 } 237 238 static int 239 xfs_growfs_log_private( 240 struct xfs_mount *mp, /* mount point for filesystem */ 241 struct xfs_growfs_log *in) /* growfs log input struct */ 242 { 243 xfs_extlen_t nb; 244 245 nb = in->newblocks; 246 if (nb < XFS_MIN_LOG_BLOCKS || nb < XFS_B_TO_FSB(mp, XFS_MIN_LOG_BYTES)) 247 return -EINVAL; 248 if (nb == mp->m_sb.sb_logblocks && 249 in->isint == (mp->m_sb.sb_logstart != 0)) 250 return -EINVAL; 251 /* 252 * Moving the log is hard, need new interfaces to sync 253 * the log first, hold off all activity while moving it. 254 * Can have shorter or longer log in the same space, 255 * or transform internal to external log or vice versa. 256 */ 257 return -ENOSYS; 258 } 259 260 static int 261 xfs_growfs_imaxpct( 262 struct xfs_mount *mp, 263 __u32 imaxpct) 264 { 265 struct xfs_trans *tp; 266 int dpct; 267 int error; 268 269 if (imaxpct > 100) 270 return -EINVAL; 271 272 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata, 273 XFS_GROWFS_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, &tp); 274 if (error) 275 return error; 276 277 dpct = imaxpct - mp->m_sb.sb_imax_pct; 278 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IMAXPCT, dpct); 279 xfs_trans_set_sync(tp); 280 return xfs_trans_commit(tp); 281 } 282 283 /* 284 * protected versions of growfs function acquire and release locks on the mount 285 * point - exported through ioctls: XFS_IOC_FSGROWFSDATA, XFS_IOC_FSGROWFSLOG, 286 * XFS_IOC_FSGROWFSRT 287 */ 288 int 289 xfs_growfs_data( 290 struct xfs_mount *mp, 291 struct xfs_growfs_data *in) 292 { 293 int error = 0; 294 295 if (!capable(CAP_SYS_ADMIN)) 296 return -EPERM; 297 if (!mutex_trylock(&mp->m_growlock)) 298 return -EWOULDBLOCK; 299 300 /* update imaxpct separately to the physical grow of the filesystem */ 301 if (in->imaxpct != mp->m_sb.sb_imax_pct) { 302 error = xfs_growfs_imaxpct(mp, in->imaxpct); 303 if (error) 304 goto out_error; 305 } 306 307 if (in->newblocks != mp->m_sb.sb_dblocks) { 308 error = xfs_growfs_data_private(mp, in); 309 if (error) 310 goto out_error; 311 } 312 313 /* Post growfs calculations needed to reflect new state in operations */ 314 if (mp->m_sb.sb_imax_pct) { 315 uint64_t icount = mp->m_sb.sb_dblocks * mp->m_sb.sb_imax_pct; 316 do_div(icount, 100); 317 M_IGEO(mp)->maxicount = XFS_FSB_TO_INO(mp, icount); 318 } else 319 M_IGEO(mp)->maxicount = 0; 320 321 /* Update secondary superblocks now the physical grow has completed */ 322 error = xfs_update_secondary_sbs(mp); 323 324 out_error: 325 /* 326 * Increment the generation unconditionally, the error could be from 327 * updating the secondary superblocks, in which case the new size 328 * is live already. 329 */ 330 mp->m_generation++; 331 mutex_unlock(&mp->m_growlock); 332 return error; 333 } 334 335 int 336 xfs_growfs_log( 337 xfs_mount_t *mp, 338 struct xfs_growfs_log *in) 339 { 340 int error; 341 342 if (!capable(CAP_SYS_ADMIN)) 343 return -EPERM; 344 if (!mutex_trylock(&mp->m_growlock)) 345 return -EWOULDBLOCK; 346 error = xfs_growfs_log_private(mp, in); 347 mutex_unlock(&mp->m_growlock); 348 return error; 349 } 350 351 /* 352 * Reserve the requested number of blocks if available. Otherwise return 353 * as many as possible to satisfy the request. The actual number 354 * reserved are returned in outval. 355 */ 356 int 357 xfs_reserve_blocks( 358 struct xfs_mount *mp, 359 uint64_t request) 360 { 361 int64_t lcounter, delta; 362 int64_t fdblks_delta = 0; 363 int64_t free; 364 int error = 0; 365 366 /* 367 * With per-cpu counters, this becomes an interesting problem. we need 368 * to work out if we are freeing or allocation blocks first, then we can 369 * do the modification as necessary. 370 * 371 * We do this under the m_sb_lock so that if we are near ENOSPC, we will 372 * hold out any changes while we work out what to do. This means that 373 * the amount of free space can change while we do this, so we need to 374 * retry if we end up trying to reserve more space than is available. 375 */ 376 spin_lock(&mp->m_sb_lock); 377 378 /* 379 * If our previous reservation was larger than the current value, 380 * then move any unused blocks back to the free pool. Modify the resblks 381 * counters directly since we shouldn't have any problems unreserving 382 * space. 383 */ 384 if (mp->m_resblks > request) { 385 lcounter = mp->m_resblks_avail - request; 386 if (lcounter > 0) { /* release unused blocks */ 387 fdblks_delta = lcounter; 388 mp->m_resblks_avail -= lcounter; 389 } 390 mp->m_resblks = request; 391 if (fdblks_delta) { 392 spin_unlock(&mp->m_sb_lock); 393 xfs_add_fdblocks(mp, fdblks_delta); 394 spin_lock(&mp->m_sb_lock); 395 } 396 397 goto out; 398 } 399 400 /* 401 * If the request is larger than the current reservation, reserve the 402 * blocks before we update the reserve counters. Sample m_fdblocks and 403 * perform a partial reservation if the request exceeds free space. 404 * 405 * The code below estimates how many blocks it can request from 406 * fdblocks to stash in the reserve pool. This is a classic TOCTOU 407 * race since fdblocks updates are not always coordinated via 408 * m_sb_lock. Set the reserve size even if there's not enough free 409 * space to fill it because mod_fdblocks will refill an undersized 410 * reserve when it can. 411 */ 412 free = percpu_counter_sum(&mp->m_fdblocks) - 413 xfs_fdblocks_unavailable(mp); 414 delta = request - mp->m_resblks; 415 mp->m_resblks = request; 416 if (delta > 0 && free > 0) { 417 /* 418 * We'll either succeed in getting space from the free block 419 * count or we'll get an ENOSPC. Don't set the reserved flag 420 * here - we don't want to reserve the extra reserve blocks 421 * from the reserve. 422 * 423 * The desired reserve size can change after we drop the lock. 424 * Use mod_fdblocks to put the space into the reserve or into 425 * fdblocks as appropriate. 426 */ 427 fdblks_delta = min(free, delta); 428 spin_unlock(&mp->m_sb_lock); 429 error = xfs_dec_fdblocks(mp, fdblks_delta, 0); 430 if (!error) 431 xfs_add_fdblocks(mp, fdblks_delta); 432 spin_lock(&mp->m_sb_lock); 433 } 434 out: 435 spin_unlock(&mp->m_sb_lock); 436 return error; 437 } 438 439 int 440 xfs_fs_goingdown( 441 xfs_mount_t *mp, 442 uint32_t inflags) 443 { 444 switch (inflags) { 445 case XFS_FSOP_GOING_FLAGS_DEFAULT: { 446 if (!bdev_freeze(mp->m_super->s_bdev)) { 447 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT); 448 bdev_thaw(mp->m_super->s_bdev); 449 } 450 break; 451 } 452 case XFS_FSOP_GOING_FLAGS_LOGFLUSH: 453 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT); 454 break; 455 case XFS_FSOP_GOING_FLAGS_NOLOGFLUSH: 456 xfs_force_shutdown(mp, 457 SHUTDOWN_FORCE_UMOUNT | SHUTDOWN_LOG_IO_ERROR); 458 break; 459 default: 460 return -EINVAL; 461 } 462 463 return 0; 464 } 465 466 /* 467 * Force a shutdown of the filesystem instantly while keeping the filesystem 468 * consistent. We don't do an unmount here; just shutdown the shop, make sure 469 * that absolutely nothing persistent happens to this filesystem after this 470 * point. 471 * 472 * The shutdown state change is atomic, resulting in the first and only the 473 * first shutdown call processing the shutdown. This means we only shutdown the 474 * log once as it requires, and we don't spam the logs when multiple concurrent 475 * shutdowns race to set the shutdown flags. 476 */ 477 void 478 xfs_do_force_shutdown( 479 struct xfs_mount *mp, 480 uint32_t flags, 481 char *fname, 482 int lnnum) 483 { 484 int tag; 485 const char *why; 486 487 488 if (test_and_set_bit(XFS_OPSTATE_SHUTDOWN, &mp->m_opstate)) { 489 xlog_shutdown_wait(mp->m_log); 490 return; 491 } 492 if (mp->m_sb_bp) 493 mp->m_sb_bp->b_flags |= XBF_DONE; 494 495 if (flags & SHUTDOWN_FORCE_UMOUNT) 496 xfs_alert(mp, "User initiated shutdown received."); 497 498 if (xlog_force_shutdown(mp->m_log, flags)) { 499 tag = XFS_PTAG_SHUTDOWN_LOGERROR; 500 why = "Log I/O Error"; 501 } else if (flags & SHUTDOWN_CORRUPT_INCORE) { 502 tag = XFS_PTAG_SHUTDOWN_CORRUPT; 503 why = "Corruption of in-memory data"; 504 } else if (flags & SHUTDOWN_CORRUPT_ONDISK) { 505 tag = XFS_PTAG_SHUTDOWN_CORRUPT; 506 why = "Corruption of on-disk metadata"; 507 } else if (flags & SHUTDOWN_DEVICE_REMOVED) { 508 tag = XFS_PTAG_SHUTDOWN_IOERROR; 509 why = "Block device removal"; 510 } else { 511 tag = XFS_PTAG_SHUTDOWN_IOERROR; 512 why = "Metadata I/O Error"; 513 } 514 515 trace_xfs_force_shutdown(mp, tag, flags, fname, lnnum); 516 517 xfs_alert_tag(mp, tag, 518 "%s (0x%x) detected at %pS (%s:%d). Shutting down filesystem.", 519 why, flags, __return_address, fname, lnnum); 520 xfs_alert(mp, 521 "Please unmount the filesystem and rectify the problem(s)"); 522 if (xfs_error_level >= XFS_ERRLEVEL_HIGH) 523 xfs_stack_trace(); 524 } 525 526 /* 527 * Reserve free space for per-AG metadata. 528 */ 529 int 530 xfs_fs_reserve_ag_blocks( 531 struct xfs_mount *mp) 532 { 533 xfs_agnumber_t agno; 534 struct xfs_perag *pag; 535 int error = 0; 536 int err2; 537 538 mp->m_finobt_nores = false; 539 for_each_perag(mp, agno, pag) { 540 err2 = xfs_ag_resv_init(pag, NULL); 541 if (err2 && !error) 542 error = err2; 543 } 544 545 if (error && error != -ENOSPC) { 546 xfs_warn(mp, 547 "Error %d reserving per-AG metadata reserve pool.", error); 548 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); 549 } 550 551 return error; 552 } 553 554 /* 555 * Free space reserved for per-AG metadata. 556 */ 557 void 558 xfs_fs_unreserve_ag_blocks( 559 struct xfs_mount *mp) 560 { 561 xfs_agnumber_t agno; 562 struct xfs_perag *pag; 563 564 for_each_perag(mp, agno, pag) 565 xfs_ag_resv_free(pag); 566 } 567