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_ag.h" 21 #include "xfs_ag_resv.h" 22 23 /* 24 * growfs operations 25 */ 26 static int 27 xfs_growfs_data_private( 28 struct xfs_mount *mp, /* mount point for filesystem */ 29 struct xfs_growfs_data *in) /* growfs data input struct */ 30 { 31 struct xfs_buf *bp; 32 int error; 33 xfs_agnumber_t nagcount; 34 xfs_agnumber_t nagimax = 0; 35 xfs_rfsblock_t nb, nb_div, nb_mod; 36 xfs_rfsblock_t delta; 37 xfs_agnumber_t oagcount; 38 struct xfs_trans *tp; 39 struct aghdr_init_data id = {}; 40 41 nb = in->newblocks; 42 if (nb < mp->m_sb.sb_dblocks) 43 return -EINVAL; 44 if ((error = xfs_sb_validate_fsb_count(&mp->m_sb, nb))) 45 return error; 46 error = xfs_buf_read_uncached(mp->m_ddev_targp, 47 XFS_FSB_TO_BB(mp, nb) - XFS_FSS_TO_BB(mp, 1), 48 XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL); 49 if (error) 50 return error; 51 xfs_buf_relse(bp); 52 53 nb_div = nb; 54 nb_mod = do_div(nb_div, mp->m_sb.sb_agblocks); 55 nagcount = nb_div + (nb_mod != 0); 56 if (nb_mod && nb_mod < XFS_MIN_AG_BLOCKS) { 57 nagcount--; 58 nb = (xfs_rfsblock_t)nagcount * mp->m_sb.sb_agblocks; 59 if (nb < mp->m_sb.sb_dblocks) 60 return -EINVAL; 61 } 62 delta = nb - mp->m_sb.sb_dblocks; 63 oagcount = mp->m_sb.sb_agcount; 64 65 /* allocate the new per-ag structures */ 66 if (nagcount > oagcount) { 67 error = xfs_initialize_perag(mp, nagcount, &nagimax); 68 if (error) 69 return error; 70 } 71 72 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata, 73 XFS_GROWFS_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, &tp); 74 if (error) 75 return error; 76 77 /* 78 * Write new AG headers to disk. Non-transactional, but need to be 79 * written and completed prior to the growfs transaction being logged. 80 * To do this, we use a delayed write buffer list and wait for 81 * submission and IO completion of the list as a whole. This allows the 82 * IO subsystem to merge all the AG headers in a single AG into a single 83 * IO and hide most of the latency of the IO from us. 84 * 85 * This also means that if we get an error whilst building the buffer 86 * list to write, we can cancel the entire list without having written 87 * anything. 88 */ 89 INIT_LIST_HEAD(&id.buffer_list); 90 for (id.agno = nagcount - 1; 91 id.agno >= oagcount; 92 id.agno--, delta -= id.agsize) { 93 94 if (id.agno == nagcount - 1) 95 id.agsize = nb - 96 (id.agno * (xfs_rfsblock_t)mp->m_sb.sb_agblocks); 97 else 98 id.agsize = mp->m_sb.sb_agblocks; 99 100 error = xfs_ag_init_headers(mp, &id); 101 if (error) { 102 xfs_buf_delwri_cancel(&id.buffer_list); 103 goto out_trans_cancel; 104 } 105 } 106 error = xfs_buf_delwri_submit(&id.buffer_list); 107 if (error) 108 goto out_trans_cancel; 109 110 xfs_trans_agblocks_delta(tp, id.nfree); 111 112 /* If there are new blocks in the old last AG, extend it. */ 113 if (delta) { 114 error = xfs_ag_extend_space(mp, tp, &id, delta); 115 if (error) 116 goto out_trans_cancel; 117 } 118 119 /* 120 * Update changed superblock fields transactionally. These are not 121 * seen by the rest of the world until the transaction commit applies 122 * them atomically to the superblock. 123 */ 124 if (nagcount > oagcount) 125 xfs_trans_mod_sb(tp, XFS_TRANS_SB_AGCOUNT, nagcount - oagcount); 126 if (nb > mp->m_sb.sb_dblocks) 127 xfs_trans_mod_sb(tp, XFS_TRANS_SB_DBLOCKS, 128 nb - mp->m_sb.sb_dblocks); 129 if (id.nfree) 130 xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, id.nfree); 131 xfs_trans_set_sync(tp); 132 error = xfs_trans_commit(tp); 133 if (error) 134 return error; 135 136 /* New allocation groups fully initialized, so update mount struct */ 137 if (nagimax) 138 mp->m_maxagi = nagimax; 139 xfs_set_low_space_thresholds(mp); 140 mp->m_alloc_set_aside = xfs_alloc_set_aside(mp); 141 142 /* 143 * If we expanded the last AG, free the per-AG reservation 144 * so we can reinitialize it with the new size. 145 */ 146 if (delta) { 147 struct xfs_perag *pag; 148 149 pag = xfs_perag_get(mp, id.agno); 150 error = xfs_ag_resv_free(pag); 151 xfs_perag_put(pag); 152 if (error) 153 return error; 154 } 155 156 /* 157 * Reserve AG metadata blocks. ENOSPC here does not mean there was a 158 * growfs failure, just that there still isn't space for new user data 159 * after the grow has been run. 160 */ 161 error = xfs_fs_reserve_ag_blocks(mp); 162 if (error == -ENOSPC) 163 error = 0; 164 return error; 165 166 out_trans_cancel: 167 xfs_trans_cancel(tp); 168 return error; 169 } 170 171 static int 172 xfs_growfs_log_private( 173 struct xfs_mount *mp, /* mount point for filesystem */ 174 struct xfs_growfs_log *in) /* growfs log input struct */ 175 { 176 xfs_extlen_t nb; 177 178 nb = in->newblocks; 179 if (nb < XFS_MIN_LOG_BLOCKS || nb < XFS_B_TO_FSB(mp, XFS_MIN_LOG_BYTES)) 180 return -EINVAL; 181 if (nb == mp->m_sb.sb_logblocks && 182 in->isint == (mp->m_sb.sb_logstart != 0)) 183 return -EINVAL; 184 /* 185 * Moving the log is hard, need new interfaces to sync 186 * the log first, hold off all activity while moving it. 187 * Can have shorter or longer log in the same space, 188 * or transform internal to external log or vice versa. 189 */ 190 return -ENOSYS; 191 } 192 193 static int 194 xfs_growfs_imaxpct( 195 struct xfs_mount *mp, 196 __u32 imaxpct) 197 { 198 struct xfs_trans *tp; 199 int dpct; 200 int error; 201 202 if (imaxpct > 100) 203 return -EINVAL; 204 205 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata, 206 XFS_GROWFS_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, &tp); 207 if (error) 208 return error; 209 210 dpct = imaxpct - mp->m_sb.sb_imax_pct; 211 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IMAXPCT, dpct); 212 xfs_trans_set_sync(tp); 213 return xfs_trans_commit(tp); 214 } 215 216 /* 217 * protected versions of growfs function acquire and release locks on the mount 218 * point - exported through ioctls: XFS_IOC_FSGROWFSDATA, XFS_IOC_FSGROWFSLOG, 219 * XFS_IOC_FSGROWFSRT 220 */ 221 int 222 xfs_growfs_data( 223 struct xfs_mount *mp, 224 struct xfs_growfs_data *in) 225 { 226 int error = 0; 227 228 if (!capable(CAP_SYS_ADMIN)) 229 return -EPERM; 230 if (!mutex_trylock(&mp->m_growlock)) 231 return -EWOULDBLOCK; 232 233 /* update imaxpct separately to the physical grow of the filesystem */ 234 if (in->imaxpct != mp->m_sb.sb_imax_pct) { 235 error = xfs_growfs_imaxpct(mp, in->imaxpct); 236 if (error) 237 goto out_error; 238 } 239 240 if (in->newblocks != mp->m_sb.sb_dblocks) { 241 error = xfs_growfs_data_private(mp, in); 242 if (error) 243 goto out_error; 244 } 245 246 /* Post growfs calculations needed to reflect new state in operations */ 247 if (mp->m_sb.sb_imax_pct) { 248 uint64_t icount = mp->m_sb.sb_dblocks * mp->m_sb.sb_imax_pct; 249 do_div(icount, 100); 250 M_IGEO(mp)->maxicount = XFS_FSB_TO_INO(mp, icount); 251 } else 252 M_IGEO(mp)->maxicount = 0; 253 254 /* Update secondary superblocks now the physical grow has completed */ 255 error = xfs_update_secondary_sbs(mp); 256 257 out_error: 258 /* 259 * Increment the generation unconditionally, the error could be from 260 * updating the secondary superblocks, in which case the new size 261 * is live already. 262 */ 263 mp->m_generation++; 264 mutex_unlock(&mp->m_growlock); 265 return error; 266 } 267 268 int 269 xfs_growfs_log( 270 xfs_mount_t *mp, 271 struct xfs_growfs_log *in) 272 { 273 int error; 274 275 if (!capable(CAP_SYS_ADMIN)) 276 return -EPERM; 277 if (!mutex_trylock(&mp->m_growlock)) 278 return -EWOULDBLOCK; 279 error = xfs_growfs_log_private(mp, in); 280 mutex_unlock(&mp->m_growlock); 281 return error; 282 } 283 284 /* 285 * exported through ioctl XFS_IOC_FSCOUNTS 286 */ 287 288 void 289 xfs_fs_counts( 290 xfs_mount_t *mp, 291 xfs_fsop_counts_t *cnt) 292 { 293 cnt->allocino = percpu_counter_read_positive(&mp->m_icount); 294 cnt->freeino = percpu_counter_read_positive(&mp->m_ifree); 295 cnt->freedata = percpu_counter_read_positive(&mp->m_fdblocks) - 296 mp->m_alloc_set_aside; 297 298 spin_lock(&mp->m_sb_lock); 299 cnt->freertx = mp->m_sb.sb_frextents; 300 spin_unlock(&mp->m_sb_lock); 301 } 302 303 /* 304 * exported through ioctl XFS_IOC_SET_RESBLKS & XFS_IOC_GET_RESBLKS 305 * 306 * xfs_reserve_blocks is called to set m_resblks 307 * in the in-core mount table. The number of unused reserved blocks 308 * is kept in m_resblks_avail. 309 * 310 * Reserve the requested number of blocks if available. Otherwise return 311 * as many as possible to satisfy the request. The actual number 312 * reserved are returned in outval 313 * 314 * A null inval pointer indicates that only the current reserved blocks 315 * available should be returned no settings are changed. 316 */ 317 318 int 319 xfs_reserve_blocks( 320 xfs_mount_t *mp, 321 uint64_t *inval, 322 xfs_fsop_resblks_t *outval) 323 { 324 int64_t lcounter, delta; 325 int64_t fdblks_delta = 0; 326 uint64_t request; 327 int64_t free; 328 int error = 0; 329 330 /* If inval is null, report current values and return */ 331 if (inval == (uint64_t *)NULL) { 332 if (!outval) 333 return -EINVAL; 334 outval->resblks = mp->m_resblks; 335 outval->resblks_avail = mp->m_resblks_avail; 336 return 0; 337 } 338 339 request = *inval; 340 341 /* 342 * With per-cpu counters, this becomes an interesting problem. we need 343 * to work out if we are freeing or allocation blocks first, then we can 344 * do the modification as necessary. 345 * 346 * We do this under the m_sb_lock so that if we are near ENOSPC, we will 347 * hold out any changes while we work out what to do. This means that 348 * the amount of free space can change while we do this, so we need to 349 * retry if we end up trying to reserve more space than is available. 350 */ 351 spin_lock(&mp->m_sb_lock); 352 353 /* 354 * If our previous reservation was larger than the current value, 355 * then move any unused blocks back to the free pool. Modify the resblks 356 * counters directly since we shouldn't have any problems unreserving 357 * space. 358 */ 359 if (mp->m_resblks > request) { 360 lcounter = mp->m_resblks_avail - request; 361 if (lcounter > 0) { /* release unused blocks */ 362 fdblks_delta = lcounter; 363 mp->m_resblks_avail -= lcounter; 364 } 365 mp->m_resblks = request; 366 if (fdblks_delta) { 367 spin_unlock(&mp->m_sb_lock); 368 error = xfs_mod_fdblocks(mp, fdblks_delta, 0); 369 spin_lock(&mp->m_sb_lock); 370 } 371 372 goto out; 373 } 374 375 /* 376 * If the request is larger than the current reservation, reserve the 377 * blocks before we update the reserve counters. Sample m_fdblocks and 378 * perform a partial reservation if the request exceeds free space. 379 */ 380 error = -ENOSPC; 381 do { 382 free = percpu_counter_sum(&mp->m_fdblocks) - 383 mp->m_alloc_set_aside; 384 if (free <= 0) 385 break; 386 387 delta = request - mp->m_resblks; 388 lcounter = free - delta; 389 if (lcounter < 0) 390 /* We can't satisfy the request, just get what we can */ 391 fdblks_delta = free; 392 else 393 fdblks_delta = delta; 394 395 /* 396 * We'll either succeed in getting space from the free block 397 * count or we'll get an ENOSPC. If we get a ENOSPC, it means 398 * things changed while we were calculating fdblks_delta and so 399 * we should try again to see if there is anything left to 400 * reserve. 401 * 402 * Don't set the reserved flag here - we don't want to reserve 403 * the extra reserve blocks from the reserve..... 404 */ 405 spin_unlock(&mp->m_sb_lock); 406 error = xfs_mod_fdblocks(mp, -fdblks_delta, 0); 407 spin_lock(&mp->m_sb_lock); 408 } while (error == -ENOSPC); 409 410 /* 411 * Update the reserve counters if blocks have been successfully 412 * allocated. 413 */ 414 if (!error && fdblks_delta) { 415 mp->m_resblks += fdblks_delta; 416 mp->m_resblks_avail += fdblks_delta; 417 } 418 419 out: 420 if (outval) { 421 outval->resblks = mp->m_resblks; 422 outval->resblks_avail = mp->m_resblks_avail; 423 } 424 425 spin_unlock(&mp->m_sb_lock); 426 return error; 427 } 428 429 int 430 xfs_fs_goingdown( 431 xfs_mount_t *mp, 432 uint32_t inflags) 433 { 434 switch (inflags) { 435 case XFS_FSOP_GOING_FLAGS_DEFAULT: { 436 if (!freeze_bdev(mp->m_super->s_bdev)) { 437 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT); 438 thaw_bdev(mp->m_super->s_bdev); 439 } 440 break; 441 } 442 case XFS_FSOP_GOING_FLAGS_LOGFLUSH: 443 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT); 444 break; 445 case XFS_FSOP_GOING_FLAGS_NOLOGFLUSH: 446 xfs_force_shutdown(mp, 447 SHUTDOWN_FORCE_UMOUNT | SHUTDOWN_LOG_IO_ERROR); 448 break; 449 default: 450 return -EINVAL; 451 } 452 453 return 0; 454 } 455 456 /* 457 * Force a shutdown of the filesystem instantly while keeping the filesystem 458 * consistent. We don't do an unmount here; just shutdown the shop, make sure 459 * that absolutely nothing persistent happens to this filesystem after this 460 * point. 461 */ 462 void 463 xfs_do_force_shutdown( 464 struct xfs_mount *mp, 465 int flags, 466 char *fname, 467 int lnnum) 468 { 469 bool logerror = flags & SHUTDOWN_LOG_IO_ERROR; 470 471 /* 472 * No need to duplicate efforts. 473 */ 474 if (XFS_FORCED_SHUTDOWN(mp) && !logerror) 475 return; 476 477 /* 478 * This flags XFS_MOUNT_FS_SHUTDOWN, makes sure that we don't 479 * queue up anybody new on the log reservations, and wakes up 480 * everybody who's sleeping on log reservations to tell them 481 * the bad news. 482 */ 483 if (xfs_log_force_umount(mp, logerror)) 484 return; 485 486 if (flags & SHUTDOWN_FORCE_UMOUNT) { 487 xfs_alert(mp, 488 "User initiated shutdown received. Shutting down filesystem"); 489 return; 490 } 491 492 xfs_notice(mp, 493 "%s(0x%x) called from line %d of file %s. Return address = "PTR_FMT, 494 __func__, flags, lnnum, fname, __return_address); 495 496 if (flags & SHUTDOWN_CORRUPT_INCORE) { 497 xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_CORRUPT, 498 "Corruption of in-memory data detected. Shutting down filesystem"); 499 if (XFS_ERRLEVEL_HIGH <= xfs_error_level) 500 xfs_stack_trace(); 501 } else if (logerror) { 502 xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_LOGERROR, 503 "Log I/O Error Detected. Shutting down filesystem"); 504 } else { 505 xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR, 506 "I/O Error Detected. Shutting down filesystem"); 507 } 508 509 xfs_alert(mp, 510 "Please unmount the filesystem and rectify the problem(s)"); 511 } 512 513 /* 514 * Reserve free space for per-AG metadata. 515 */ 516 int 517 xfs_fs_reserve_ag_blocks( 518 struct xfs_mount *mp) 519 { 520 xfs_agnumber_t agno; 521 struct xfs_perag *pag; 522 int error = 0; 523 int err2; 524 525 mp->m_finobt_nores = false; 526 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { 527 pag = xfs_perag_get(mp, agno); 528 err2 = xfs_ag_resv_init(pag, NULL); 529 xfs_perag_put(pag); 530 if (err2 && !error) 531 error = err2; 532 } 533 534 if (error && error != -ENOSPC) { 535 xfs_warn(mp, 536 "Error %d reserving per-AG metadata reserve pool.", error); 537 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); 538 } 539 540 return error; 541 } 542 543 /* 544 * Free space reserved for per-AG metadata. 545 */ 546 int 547 xfs_fs_unreserve_ag_blocks( 548 struct xfs_mount *mp) 549 { 550 xfs_agnumber_t agno; 551 struct xfs_perag *pag; 552 int error = 0; 553 int err2; 554 555 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { 556 pag = xfs_perag_get(mp, agno); 557 err2 = xfs_ag_resv_free(pag); 558 xfs_perag_put(pag); 559 if (err2 && !error) 560 error = err2; 561 } 562 563 if (error) 564 xfs_warn(mp, 565 "Error %d freeing per-AG metadata reserve pool.", error); 566 567 return error; 568 } 569