1/* 2 * Copyright (c) 2000-2005 Silicon Graphics, Inc. 3 * All Rights Reserved. 4 * 5 * This program is free software; you can redistribute it and/or 6 * modify it under the terms of the GNU General Public License as 7 * published by the Free Software Foundation. 8 * 9 * This program is distributed in the hope that it would be useful, 10 * but WITHOUT ANY WARRANTY; without even the implied warranty of 11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 12 * GNU General Public License for more details. 13 * 14 * You should have received a copy of the GNU General Public License 15 * along with this program; if not, write the Free Software Foundation, 16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA 17 */ 18#include "xfs.h" 19#include "xfs_fs.h" 20#include "xfs_shared.h" 21#include "xfs_format.h" 22#include "xfs_log_format.h" 23#include "xfs_trans_resv.h" 24#include "xfs_bit.h" 25#include "xfs_sb.h" 26#include "xfs_mount.h" 27#include "xfs_inode.h" 28#include "xfs_ialloc.h" 29#include "xfs_itable.h" 30#include "xfs_quota.h" 31#include "xfs_error.h" 32#include "xfs_bmap.h" 33#include "xfs_bmap_btree.h" 34#include "xfs_trans.h" 35#include "xfs_trans_space.h" 36#include "xfs_qm.h" 37#include "xfs_trace.h" 38#include "xfs_icache.h" 39#include "xfs_cksum.h" 40 41/* 42 * The global quota manager. There is only one of these for the entire 43 * system, _not_ one per file system. XQM keeps track of the overall 44 * quota functionality, including maintaining the freelist and hash 45 * tables of dquots. 46 */ 47STATIC int xfs_qm_init_quotainos(xfs_mount_t *); 48STATIC int xfs_qm_init_quotainfo(xfs_mount_t *); 49 50 51STATIC void xfs_qm_dqfree_one(struct xfs_dquot *dqp); 52/* 53 * We use the batch lookup interface to iterate over the dquots as it 54 * currently is the only interface into the radix tree code that allows 55 * fuzzy lookups instead of exact matches. Holding the lock over multiple 56 * operations is fine as all callers are used either during mount/umount 57 * or quotaoff. 58 */ 59#define XFS_DQ_LOOKUP_BATCH 32 60 61STATIC int 62xfs_qm_dquot_walk( 63 struct xfs_mount *mp, 64 int type, 65 int (*execute)(struct xfs_dquot *dqp, void *data), 66 void *data) 67{ 68 struct xfs_quotainfo *qi = mp->m_quotainfo; 69 struct radix_tree_root *tree = xfs_dquot_tree(qi, type); 70 uint32_t next_index; 71 int last_error = 0; 72 int skipped; 73 int nr_found; 74 75restart: 76 skipped = 0; 77 next_index = 0; 78 nr_found = 0; 79 80 while (1) { 81 struct xfs_dquot *batch[XFS_DQ_LOOKUP_BATCH]; 82 int error = 0; 83 int i; 84 85 mutex_lock(&qi->qi_tree_lock); 86 nr_found = radix_tree_gang_lookup(tree, (void **)batch, 87 next_index, XFS_DQ_LOOKUP_BATCH); 88 if (!nr_found) { 89 mutex_unlock(&qi->qi_tree_lock); 90 break; 91 } 92 93 for (i = 0; i < nr_found; i++) { 94 struct xfs_dquot *dqp = batch[i]; 95 96 next_index = be32_to_cpu(dqp->q_core.d_id) + 1; 97 98 error = execute(batch[i], data); 99 if (error == -EAGAIN) { 100 skipped++; 101 continue; 102 } 103 if (error && last_error != -EFSCORRUPTED) 104 last_error = error; 105 } 106 107 mutex_unlock(&qi->qi_tree_lock); 108 109 /* bail out if the filesystem is corrupted. */ 110 if (last_error == -EFSCORRUPTED) { 111 skipped = 0; 112 break; 113 } 114 } 115 116 if (skipped) { 117 delay(1); 118 goto restart; 119 } 120 121 return last_error; 122} 123 124 125/* 126 * Purge a dquot from all tracking data structures and free it. 127 */ 128STATIC int 129xfs_qm_dqpurge( 130 struct xfs_dquot *dqp, 131 void *data) 132{ 133 struct xfs_mount *mp = dqp->q_mount; 134 struct xfs_quotainfo *qi = mp->m_quotainfo; 135 136 xfs_dqlock(dqp); 137 if ((dqp->dq_flags & XFS_DQ_FREEING) || dqp->q_nrefs != 0) { 138 xfs_dqunlock(dqp); 139 return -EAGAIN; 140 } 141 142 dqp->dq_flags |= XFS_DQ_FREEING; 143 144 xfs_dqflock(dqp); 145 146 /* 147 * If we are turning this type of quotas off, we don't care 148 * about the dirty metadata sitting in this dquot. OTOH, if 149 * we're unmounting, we do care, so we flush it and wait. 150 */ 151 if (XFS_DQ_IS_DIRTY(dqp)) { 152 struct xfs_buf *bp = NULL; 153 int error; 154 155 /* 156 * We don't care about getting disk errors here. We need 157 * to purge this dquot anyway, so we go ahead regardless. 158 */ 159 error = xfs_qm_dqflush(dqp, &bp); 160 if (error) { 161 xfs_warn(mp, "%s: dquot %p flush failed", 162 __func__, dqp); 163 } else { 164 error = xfs_bwrite(bp); 165 xfs_buf_relse(bp); 166 } 167 xfs_dqflock(dqp); 168 } 169 170 ASSERT(atomic_read(&dqp->q_pincount) == 0); 171 ASSERT(XFS_FORCED_SHUTDOWN(mp) || 172 !(dqp->q_logitem.qli_item.li_flags & XFS_LI_IN_AIL)); 173 174 xfs_dqfunlock(dqp); 175 xfs_dqunlock(dqp); 176 177 radix_tree_delete(xfs_dquot_tree(qi, dqp->q_core.d_flags), 178 be32_to_cpu(dqp->q_core.d_id)); 179 qi->qi_dquots--; 180 181 /* 182 * We move dquots to the freelist as soon as their reference count 183 * hits zero, so it really should be on the freelist here. 184 */ 185 ASSERT(!list_empty(&dqp->q_lru)); 186 list_lru_del(&qi->qi_lru, &dqp->q_lru); 187 XFS_STATS_DEC(xs_qm_dquot_unused); 188 189 xfs_qm_dqdestroy(dqp); 190 return 0; 191} 192 193/* 194 * Purge the dquot cache. 195 */ 196void 197xfs_qm_dqpurge_all( 198 struct xfs_mount *mp, 199 uint flags) 200{ 201 if (flags & XFS_QMOPT_UQUOTA) 202 xfs_qm_dquot_walk(mp, XFS_DQ_USER, xfs_qm_dqpurge, NULL); 203 if (flags & XFS_QMOPT_GQUOTA) 204 xfs_qm_dquot_walk(mp, XFS_DQ_GROUP, xfs_qm_dqpurge, NULL); 205 if (flags & XFS_QMOPT_PQUOTA) 206 xfs_qm_dquot_walk(mp, XFS_DQ_PROJ, xfs_qm_dqpurge, NULL); 207} 208 209/* 210 * Just destroy the quotainfo structure. 211 */ 212void 213xfs_qm_unmount( 214 struct xfs_mount *mp) 215{ 216 if (mp->m_quotainfo) { 217 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL); 218 xfs_qm_destroy_quotainfo(mp); 219 } 220} 221 222/* 223 * Called from the vfsops layer. 224 */ 225void 226xfs_qm_unmount_quotas( 227 xfs_mount_t *mp) 228{ 229 /* 230 * Release the dquots that root inode, et al might be holding, 231 * before we flush quotas and blow away the quotainfo structure. 232 */ 233 ASSERT(mp->m_rootip); 234 xfs_qm_dqdetach(mp->m_rootip); 235 if (mp->m_rbmip) 236 xfs_qm_dqdetach(mp->m_rbmip); 237 if (mp->m_rsumip) 238 xfs_qm_dqdetach(mp->m_rsumip); 239 240 /* 241 * Release the quota inodes. 242 */ 243 if (mp->m_quotainfo) { 244 if (mp->m_quotainfo->qi_uquotaip) { 245 IRELE(mp->m_quotainfo->qi_uquotaip); 246 mp->m_quotainfo->qi_uquotaip = NULL; 247 } 248 if (mp->m_quotainfo->qi_gquotaip) { 249 IRELE(mp->m_quotainfo->qi_gquotaip); 250 mp->m_quotainfo->qi_gquotaip = NULL; 251 } 252 if (mp->m_quotainfo->qi_pquotaip) { 253 IRELE(mp->m_quotainfo->qi_pquotaip); 254 mp->m_quotainfo->qi_pquotaip = NULL; 255 } 256 } 257} 258 259STATIC int 260xfs_qm_dqattach_one( 261 xfs_inode_t *ip, 262 xfs_dqid_t id, 263 uint type, 264 uint doalloc, 265 xfs_dquot_t **IO_idqpp) 266{ 267 xfs_dquot_t *dqp; 268 int error; 269 270 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); 271 error = 0; 272 273 /* 274 * See if we already have it in the inode itself. IO_idqpp is &i_udquot 275 * or &i_gdquot. This made the code look weird, but made the logic a lot 276 * simpler. 277 */ 278 dqp = *IO_idqpp; 279 if (dqp) { 280 trace_xfs_dqattach_found(dqp); 281 return 0; 282 } 283 284 /* 285 * Find the dquot from somewhere. This bumps the reference count of 286 * dquot and returns it locked. This can return ENOENT if dquot didn't 287 * exist on disk and we didn't ask it to allocate; ESRCH if quotas got 288 * turned off suddenly. 289 */ 290 error = xfs_qm_dqget(ip->i_mount, ip, id, type, 291 doalloc | XFS_QMOPT_DOWARN, &dqp); 292 if (error) 293 return error; 294 295 trace_xfs_dqattach_get(dqp); 296 297 /* 298 * dqget may have dropped and re-acquired the ilock, but it guarantees 299 * that the dquot returned is the one that should go in the inode. 300 */ 301 *IO_idqpp = dqp; 302 xfs_dqunlock(dqp); 303 return 0; 304} 305 306static bool 307xfs_qm_need_dqattach( 308 struct xfs_inode *ip) 309{ 310 struct xfs_mount *mp = ip->i_mount; 311 312 if (!XFS_IS_QUOTA_RUNNING(mp)) 313 return false; 314 if (!XFS_IS_QUOTA_ON(mp)) 315 return false; 316 if (!XFS_NOT_DQATTACHED(mp, ip)) 317 return false; 318 if (xfs_is_quota_inode(&mp->m_sb, ip->i_ino)) 319 return false; 320 return true; 321} 322 323/* 324 * Given a locked inode, attach dquot(s) to it, taking U/G/P-QUOTAON 325 * into account. 326 * If XFS_QMOPT_DQALLOC, the dquot(s) will be allocated if needed. 327 * Inode may get unlocked and relocked in here, and the caller must deal with 328 * the consequences. 329 */ 330int 331xfs_qm_dqattach_locked( 332 xfs_inode_t *ip, 333 uint flags) 334{ 335 xfs_mount_t *mp = ip->i_mount; 336 int error = 0; 337 338 if (!xfs_qm_need_dqattach(ip)) 339 return 0; 340 341 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); 342 343 if (XFS_IS_UQUOTA_ON(mp) && !ip->i_udquot) { 344 error = xfs_qm_dqattach_one(ip, ip->i_d.di_uid, XFS_DQ_USER, 345 flags & XFS_QMOPT_DQALLOC, 346 &ip->i_udquot); 347 if (error) 348 goto done; 349 ASSERT(ip->i_udquot); 350 } 351 352 if (XFS_IS_GQUOTA_ON(mp) && !ip->i_gdquot) { 353 error = xfs_qm_dqattach_one(ip, ip->i_d.di_gid, XFS_DQ_GROUP, 354 flags & XFS_QMOPT_DQALLOC, 355 &ip->i_gdquot); 356 if (error) 357 goto done; 358 ASSERT(ip->i_gdquot); 359 } 360 361 if (XFS_IS_PQUOTA_ON(mp) && !ip->i_pdquot) { 362 error = xfs_qm_dqattach_one(ip, xfs_get_projid(ip), XFS_DQ_PROJ, 363 flags & XFS_QMOPT_DQALLOC, 364 &ip->i_pdquot); 365 if (error) 366 goto done; 367 ASSERT(ip->i_pdquot); 368 } 369 370done: 371 /* 372 * Don't worry about the dquots that we may have attached before any 373 * error - they'll get detached later if it has not already been done. 374 */ 375 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); 376 return error; 377} 378 379int 380xfs_qm_dqattach( 381 struct xfs_inode *ip, 382 uint flags) 383{ 384 int error; 385 386 if (!xfs_qm_need_dqattach(ip)) 387 return 0; 388 389 xfs_ilock(ip, XFS_ILOCK_EXCL); 390 error = xfs_qm_dqattach_locked(ip, flags); 391 xfs_iunlock(ip, XFS_ILOCK_EXCL); 392 393 return error; 394} 395 396/* 397 * Release dquots (and their references) if any. 398 * The inode should be locked EXCL except when this's called by 399 * xfs_ireclaim. 400 */ 401void 402xfs_qm_dqdetach( 403 xfs_inode_t *ip) 404{ 405 if (!(ip->i_udquot || ip->i_gdquot || ip->i_pdquot)) 406 return; 407 408 trace_xfs_dquot_dqdetach(ip); 409 410 ASSERT(!xfs_is_quota_inode(&ip->i_mount->m_sb, ip->i_ino)); 411 if (ip->i_udquot) { 412 xfs_qm_dqrele(ip->i_udquot); 413 ip->i_udquot = NULL; 414 } 415 if (ip->i_gdquot) { 416 xfs_qm_dqrele(ip->i_gdquot); 417 ip->i_gdquot = NULL; 418 } 419 if (ip->i_pdquot) { 420 xfs_qm_dqrele(ip->i_pdquot); 421 ip->i_pdquot = NULL; 422 } 423} 424 425struct xfs_qm_isolate { 426 struct list_head buffers; 427 struct list_head dispose; 428}; 429 430static enum lru_status 431xfs_qm_dquot_isolate( 432 struct list_head *item, 433 struct list_lru_one *lru, 434 spinlock_t *lru_lock, 435 void *arg) 436 __releases(lru_lock) __acquires(lru_lock) 437{ 438 struct xfs_dquot *dqp = container_of(item, 439 struct xfs_dquot, q_lru); 440 struct xfs_qm_isolate *isol = arg; 441 442 if (!xfs_dqlock_nowait(dqp)) 443 goto out_miss_busy; 444 445 /* 446 * This dquot has acquired a reference in the meantime remove it from 447 * the freelist and try again. 448 */ 449 if (dqp->q_nrefs) { 450 xfs_dqunlock(dqp); 451 XFS_STATS_INC(xs_qm_dqwants); 452 453 trace_xfs_dqreclaim_want(dqp); 454 list_lru_isolate(lru, &dqp->q_lru); 455 XFS_STATS_DEC(xs_qm_dquot_unused); 456 return LRU_REMOVED; 457 } 458 459 /* 460 * If the dquot is dirty, flush it. If it's already being flushed, just 461 * skip it so there is time for the IO to complete before we try to 462 * reclaim it again on the next LRU pass. 463 */ 464 if (!xfs_dqflock_nowait(dqp)) { 465 xfs_dqunlock(dqp); 466 goto out_miss_busy; 467 } 468 469 if (XFS_DQ_IS_DIRTY(dqp)) { 470 struct xfs_buf *bp = NULL; 471 int error; 472 473 trace_xfs_dqreclaim_dirty(dqp); 474 475 /* we have to drop the LRU lock to flush the dquot */ 476 spin_unlock(lru_lock); 477 478 error = xfs_qm_dqflush(dqp, &bp); 479 if (error) { 480 xfs_warn(dqp->q_mount, "%s: dquot %p flush failed", 481 __func__, dqp); 482 goto out_unlock_dirty; 483 } 484 485 xfs_buf_delwri_queue(bp, &isol->buffers); 486 xfs_buf_relse(bp); 487 goto out_unlock_dirty; 488 } 489 xfs_dqfunlock(dqp); 490 491 /* 492 * Prevent lookups now that we are past the point of no return. 493 */ 494 dqp->dq_flags |= XFS_DQ_FREEING; 495 xfs_dqunlock(dqp); 496 497 ASSERT(dqp->q_nrefs == 0); 498 list_lru_isolate_move(lru, &dqp->q_lru, &isol->dispose); 499 XFS_STATS_DEC(xs_qm_dquot_unused); 500 trace_xfs_dqreclaim_done(dqp); 501 XFS_STATS_INC(xs_qm_dqreclaims); 502 return LRU_REMOVED; 503 504out_miss_busy: 505 trace_xfs_dqreclaim_busy(dqp); 506 XFS_STATS_INC(xs_qm_dqreclaim_misses); 507 return LRU_SKIP; 508 509out_unlock_dirty: 510 trace_xfs_dqreclaim_busy(dqp); 511 XFS_STATS_INC(xs_qm_dqreclaim_misses); 512 xfs_dqunlock(dqp); 513 spin_lock(lru_lock); 514 return LRU_RETRY; 515} 516 517static unsigned long 518xfs_qm_shrink_scan( 519 struct shrinker *shrink, 520 struct shrink_control *sc) 521{ 522 struct xfs_quotainfo *qi = container_of(shrink, 523 struct xfs_quotainfo, qi_shrinker); 524 struct xfs_qm_isolate isol; 525 unsigned long freed; 526 int error; 527 528 if ((sc->gfp_mask & (__GFP_FS|__GFP_WAIT)) != (__GFP_FS|__GFP_WAIT)) 529 return 0; 530 531 INIT_LIST_HEAD(&isol.buffers); 532 INIT_LIST_HEAD(&isol.dispose); 533 534 freed = list_lru_shrink_walk(&qi->qi_lru, sc, 535 xfs_qm_dquot_isolate, &isol); 536 537 error = xfs_buf_delwri_submit(&isol.buffers); 538 if (error) 539 xfs_warn(NULL, "%s: dquot reclaim failed", __func__); 540 541 while (!list_empty(&isol.dispose)) { 542 struct xfs_dquot *dqp; 543 544 dqp = list_first_entry(&isol.dispose, struct xfs_dquot, q_lru); 545 list_del_init(&dqp->q_lru); 546 xfs_qm_dqfree_one(dqp); 547 } 548 549 return freed; 550} 551 552static unsigned long 553xfs_qm_shrink_count( 554 struct shrinker *shrink, 555 struct shrink_control *sc) 556{ 557 struct xfs_quotainfo *qi = container_of(shrink, 558 struct xfs_quotainfo, qi_shrinker); 559 560 return list_lru_shrink_count(&qi->qi_lru, sc); 561} 562 563/* 564 * This initializes all the quota information that's kept in the 565 * mount structure 566 */ 567STATIC int 568xfs_qm_init_quotainfo( 569 xfs_mount_t *mp) 570{ 571 xfs_quotainfo_t *qinf; 572 int error; 573 xfs_dquot_t *dqp; 574 575 ASSERT(XFS_IS_QUOTA_RUNNING(mp)); 576 577 qinf = mp->m_quotainfo = kmem_zalloc(sizeof(xfs_quotainfo_t), KM_SLEEP); 578 579 error = list_lru_init(&qinf->qi_lru); 580 if (error) 581 goto out_free_qinf; 582 583 /* 584 * See if quotainodes are setup, and if not, allocate them, 585 * and change the superblock accordingly. 586 */ 587 error = xfs_qm_init_quotainos(mp); 588 if (error) 589 goto out_free_lru; 590 591 INIT_RADIX_TREE(&qinf->qi_uquota_tree, GFP_NOFS); 592 INIT_RADIX_TREE(&qinf->qi_gquota_tree, GFP_NOFS); 593 INIT_RADIX_TREE(&qinf->qi_pquota_tree, GFP_NOFS); 594 mutex_init(&qinf->qi_tree_lock); 595 596 /* mutex used to serialize quotaoffs */ 597 mutex_init(&qinf->qi_quotaofflock); 598 599 /* Precalc some constants */ 600 qinf->qi_dqchunklen = XFS_FSB_TO_BB(mp, XFS_DQUOT_CLUSTER_SIZE_FSB); 601 qinf->qi_dqperchunk = xfs_calc_dquots_per_chunk(qinf->qi_dqchunklen); 602 603 mp->m_qflags |= (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_CHKD); 604 605 /* 606 * We try to get the limits from the superuser's limits fields. 607 * This is quite hacky, but it is standard quota practice. 608 * 609 * We look at the USR dquot with id == 0 first, but if user quotas 610 * are not enabled we goto the GRP dquot with id == 0. 611 * We don't really care to keep separate default limits for user 612 * and group quotas, at least not at this point. 613 * 614 * Since we may not have done a quotacheck by this point, just read 615 * the dquot without attaching it to any hashtables or lists. 616 */ 617 error = xfs_qm_dqread(mp, 0, 618 XFS_IS_UQUOTA_RUNNING(mp) ? XFS_DQ_USER : 619 (XFS_IS_GQUOTA_RUNNING(mp) ? XFS_DQ_GROUP : 620 XFS_DQ_PROJ), 621 XFS_QMOPT_DOWARN, &dqp); 622 if (!error) { 623 xfs_disk_dquot_t *ddqp = &dqp->q_core; 624 625 /* 626 * The warnings and timers set the grace period given to 627 * a user or group before he or she can not perform any 628 * more writing. If it is zero, a default is used. 629 */ 630 qinf->qi_btimelimit = ddqp->d_btimer ? 631 be32_to_cpu(ddqp->d_btimer) : XFS_QM_BTIMELIMIT; 632 qinf->qi_itimelimit = ddqp->d_itimer ? 633 be32_to_cpu(ddqp->d_itimer) : XFS_QM_ITIMELIMIT; 634 qinf->qi_rtbtimelimit = ddqp->d_rtbtimer ? 635 be32_to_cpu(ddqp->d_rtbtimer) : XFS_QM_RTBTIMELIMIT; 636 qinf->qi_bwarnlimit = ddqp->d_bwarns ? 637 be16_to_cpu(ddqp->d_bwarns) : XFS_QM_BWARNLIMIT; 638 qinf->qi_iwarnlimit = ddqp->d_iwarns ? 639 be16_to_cpu(ddqp->d_iwarns) : XFS_QM_IWARNLIMIT; 640 qinf->qi_rtbwarnlimit = ddqp->d_rtbwarns ? 641 be16_to_cpu(ddqp->d_rtbwarns) : XFS_QM_RTBWARNLIMIT; 642 qinf->qi_bhardlimit = be64_to_cpu(ddqp->d_blk_hardlimit); 643 qinf->qi_bsoftlimit = be64_to_cpu(ddqp->d_blk_softlimit); 644 qinf->qi_ihardlimit = be64_to_cpu(ddqp->d_ino_hardlimit); 645 qinf->qi_isoftlimit = be64_to_cpu(ddqp->d_ino_softlimit); 646 qinf->qi_rtbhardlimit = be64_to_cpu(ddqp->d_rtb_hardlimit); 647 qinf->qi_rtbsoftlimit = be64_to_cpu(ddqp->d_rtb_softlimit); 648 649 xfs_qm_dqdestroy(dqp); 650 } else { 651 qinf->qi_btimelimit = XFS_QM_BTIMELIMIT; 652 qinf->qi_itimelimit = XFS_QM_ITIMELIMIT; 653 qinf->qi_rtbtimelimit = XFS_QM_RTBTIMELIMIT; 654 qinf->qi_bwarnlimit = XFS_QM_BWARNLIMIT; 655 qinf->qi_iwarnlimit = XFS_QM_IWARNLIMIT; 656 qinf->qi_rtbwarnlimit = XFS_QM_RTBWARNLIMIT; 657 } 658 659 qinf->qi_shrinker.count_objects = xfs_qm_shrink_count; 660 qinf->qi_shrinker.scan_objects = xfs_qm_shrink_scan; 661 qinf->qi_shrinker.seeks = DEFAULT_SEEKS; 662 qinf->qi_shrinker.flags = SHRINKER_NUMA_AWARE; 663 register_shrinker(&qinf->qi_shrinker); 664 return 0; 665 666out_free_lru: 667 list_lru_destroy(&qinf->qi_lru); 668out_free_qinf: 669 kmem_free(qinf); 670 mp->m_quotainfo = NULL; 671 return error; 672} 673 674 675/* 676 * Gets called when unmounting a filesystem or when all quotas get 677 * turned off. 678 * This purges the quota inodes, destroys locks and frees itself. 679 */ 680void 681xfs_qm_destroy_quotainfo( 682 xfs_mount_t *mp) 683{ 684 xfs_quotainfo_t *qi; 685 686 qi = mp->m_quotainfo; 687 ASSERT(qi != NULL); 688 689 unregister_shrinker(&qi->qi_shrinker); 690 list_lru_destroy(&qi->qi_lru); 691 692 if (qi->qi_uquotaip) { 693 IRELE(qi->qi_uquotaip); 694 qi->qi_uquotaip = NULL; /* paranoia */ 695 } 696 if (qi->qi_gquotaip) { 697 IRELE(qi->qi_gquotaip); 698 qi->qi_gquotaip = NULL; 699 } 700 if (qi->qi_pquotaip) { 701 IRELE(qi->qi_pquotaip); 702 qi->qi_pquotaip = NULL; 703 } 704 mutex_destroy(&qi->qi_quotaofflock); 705 kmem_free(qi); 706 mp->m_quotainfo = NULL; 707} 708 709/* 710 * Create an inode and return with a reference already taken, but unlocked 711 * This is how we create quota inodes 712 */ 713STATIC int 714xfs_qm_qino_alloc( 715 xfs_mount_t *mp, 716 xfs_inode_t **ip, 717 uint flags) 718{ 719 xfs_trans_t *tp; 720 int error; 721 int committed; 722 bool need_alloc = true; 723 724 *ip = NULL; 725 /* 726 * With superblock that doesn't have separate pquotino, we 727 * share an inode between gquota and pquota. If the on-disk 728 * superblock has GQUOTA and the filesystem is now mounted 729 * with PQUOTA, just use sb_gquotino for sb_pquotino and 730 * vice-versa. 731 */ 732 if (!xfs_sb_version_has_pquotino(&mp->m_sb) && 733 (flags & (XFS_QMOPT_PQUOTA|XFS_QMOPT_GQUOTA))) { 734 xfs_ino_t ino = NULLFSINO; 735 736 if ((flags & XFS_QMOPT_PQUOTA) && 737 (mp->m_sb.sb_gquotino != NULLFSINO)) { 738 ino = mp->m_sb.sb_gquotino; 739 ASSERT(mp->m_sb.sb_pquotino == NULLFSINO); 740 } else if ((flags & XFS_QMOPT_GQUOTA) && 741 (mp->m_sb.sb_pquotino != NULLFSINO)) { 742 ino = mp->m_sb.sb_pquotino; 743 ASSERT(mp->m_sb.sb_gquotino == NULLFSINO); 744 } 745 if (ino != NULLFSINO) { 746 error = xfs_iget(mp, NULL, ino, 0, 0, ip); 747 if (error) 748 return error; 749 mp->m_sb.sb_gquotino = NULLFSINO; 750 mp->m_sb.sb_pquotino = NULLFSINO; 751 need_alloc = false; 752 } 753 } 754 755 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_QINOCREATE); 756 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_create, 757 XFS_QM_QINOCREATE_SPACE_RES(mp), 0); 758 if (error) { 759 xfs_trans_cancel(tp, 0); 760 return error; 761 } 762 763 if (need_alloc) { 764 error = xfs_dir_ialloc(&tp, NULL, S_IFREG, 1, 0, 0, 1, ip, 765 &committed); 766 if (error) { 767 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | 768 XFS_TRANS_ABORT); 769 return error; 770 } 771 } 772 773 /* 774 * Make the changes in the superblock, and log those too. 775 * sbfields arg may contain fields other than *QUOTINO; 776 * VERSIONNUM for example. 777 */ 778 spin_lock(&mp->m_sb_lock); 779 if (flags & XFS_QMOPT_SBVERSION) { 780 ASSERT(!xfs_sb_version_hasquota(&mp->m_sb)); 781 782 xfs_sb_version_addquota(&mp->m_sb); 783 mp->m_sb.sb_uquotino = NULLFSINO; 784 mp->m_sb.sb_gquotino = NULLFSINO; 785 mp->m_sb.sb_pquotino = NULLFSINO; 786 787 /* qflags will get updated fully _after_ quotacheck */ 788 mp->m_sb.sb_qflags = mp->m_qflags & XFS_ALL_QUOTA_ACCT; 789 } 790 if (flags & XFS_QMOPT_UQUOTA) 791 mp->m_sb.sb_uquotino = (*ip)->i_ino; 792 else if (flags & XFS_QMOPT_GQUOTA) 793 mp->m_sb.sb_gquotino = (*ip)->i_ino; 794 else 795 mp->m_sb.sb_pquotino = (*ip)->i_ino; 796 spin_unlock(&mp->m_sb_lock); 797 xfs_log_sb(tp); 798 799 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); 800 if (error) { 801 ASSERT(XFS_FORCED_SHUTDOWN(mp)); 802 xfs_alert(mp, "%s failed (error %d)!", __func__, error); 803 } 804 if (need_alloc) 805 xfs_finish_inode_setup(*ip); 806 return error; 807} 808 809 810STATIC void 811xfs_qm_reset_dqcounts( 812 xfs_mount_t *mp, 813 xfs_buf_t *bp, 814 xfs_dqid_t id, 815 uint type) 816{ 817 struct xfs_dqblk *dqb; 818 int j; 819 820 trace_xfs_reset_dqcounts(bp, _RET_IP_); 821 822 /* 823 * Reset all counters and timers. They'll be 824 * started afresh by xfs_qm_quotacheck. 825 */ 826#ifdef DEBUG 827 j = XFS_FSB_TO_B(mp, XFS_DQUOT_CLUSTER_SIZE_FSB); 828 do_div(j, sizeof(xfs_dqblk_t)); 829 ASSERT(mp->m_quotainfo->qi_dqperchunk == j); 830#endif 831 dqb = bp->b_addr; 832 for (j = 0; j < mp->m_quotainfo->qi_dqperchunk; j++) { 833 struct xfs_disk_dquot *ddq; 834 835 ddq = (struct xfs_disk_dquot *)&dqb[j]; 836 837 /* 838 * Do a sanity check, and if needed, repair the dqblk. Don't 839 * output any warnings because it's perfectly possible to 840 * find uninitialised dquot blks. See comment in xfs_dqcheck. 841 */ 842 xfs_dqcheck(mp, ddq, id+j, type, XFS_QMOPT_DQREPAIR, 843 "xfs_quotacheck"); 844 /* 845 * Reset type in case we are reusing group quota file for 846 * project quotas or vice versa 847 */ 848 ddq->d_flags = type; 849 ddq->d_bcount = 0; 850 ddq->d_icount = 0; 851 ddq->d_rtbcount = 0; 852 ddq->d_btimer = 0; 853 ddq->d_itimer = 0; 854 ddq->d_rtbtimer = 0; 855 ddq->d_bwarns = 0; 856 ddq->d_iwarns = 0; 857 ddq->d_rtbwarns = 0; 858 859 if (xfs_sb_version_hascrc(&mp->m_sb)) { 860 xfs_update_cksum((char *)&dqb[j], 861 sizeof(struct xfs_dqblk), 862 XFS_DQUOT_CRC_OFF); 863 } 864 } 865} 866 867STATIC int 868xfs_qm_dqiter_bufs( 869 struct xfs_mount *mp, 870 xfs_dqid_t firstid, 871 xfs_fsblock_t bno, 872 xfs_filblks_t blkcnt, 873 uint flags, 874 struct list_head *buffer_list) 875{ 876 struct xfs_buf *bp; 877 int error; 878 int type; 879 880 ASSERT(blkcnt > 0); 881 type = flags & XFS_QMOPT_UQUOTA ? XFS_DQ_USER : 882 (flags & XFS_QMOPT_PQUOTA ? XFS_DQ_PROJ : XFS_DQ_GROUP); 883 error = 0; 884 885 /* 886 * Blkcnt arg can be a very big number, and might even be 887 * larger than the log itself. So, we have to break it up into 888 * manageable-sized transactions. 889 * Note that we don't start a permanent transaction here; we might 890 * not be able to get a log reservation for the whole thing up front, 891 * and we don't really care to either, because we just discard 892 * everything if we were to crash in the middle of this loop. 893 */ 894 while (blkcnt--) { 895 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, 896 XFS_FSB_TO_DADDR(mp, bno), 897 mp->m_quotainfo->qi_dqchunklen, 0, &bp, 898 &xfs_dquot_buf_ops); 899 900 /* 901 * CRC and validation errors will return a EFSCORRUPTED here. If 902 * this occurs, re-read without CRC validation so that we can 903 * repair the damage via xfs_qm_reset_dqcounts(). This process 904 * will leave a trace in the log indicating corruption has 905 * been detected. 906 */ 907 if (error == -EFSCORRUPTED) { 908 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, 909 XFS_FSB_TO_DADDR(mp, bno), 910 mp->m_quotainfo->qi_dqchunklen, 0, &bp, 911 NULL); 912 } 913 914 if (error) 915 break; 916 917 /* 918 * A corrupt buffer might not have a verifier attached, so 919 * make sure we have the correct one attached before writeback 920 * occurs. 921 */ 922 bp->b_ops = &xfs_dquot_buf_ops; 923 xfs_qm_reset_dqcounts(mp, bp, firstid, type); 924 xfs_buf_delwri_queue(bp, buffer_list); 925 xfs_buf_relse(bp); 926 927 /* goto the next block. */ 928 bno++; 929 firstid += mp->m_quotainfo->qi_dqperchunk; 930 } 931 932 return error; 933} 934 935/* 936 * Iterate over all allocated USR/GRP/PRJ dquots in the system, calling a 937 * caller supplied function for every chunk of dquots that we find. 938 */ 939STATIC int 940xfs_qm_dqiterate( 941 struct xfs_mount *mp, 942 struct xfs_inode *qip, 943 uint flags, 944 struct list_head *buffer_list) 945{ 946 struct xfs_bmbt_irec *map; 947 int i, nmaps; /* number of map entries */ 948 int error; /* return value */ 949 xfs_fileoff_t lblkno; 950 xfs_filblks_t maxlblkcnt; 951 xfs_dqid_t firstid; 952 xfs_fsblock_t rablkno; 953 xfs_filblks_t rablkcnt; 954 955 error = 0; 956 /* 957 * This looks racy, but we can't keep an inode lock across a 958 * trans_reserve. But, this gets called during quotacheck, and that 959 * happens only at mount time which is single threaded. 960 */ 961 if (qip->i_d.di_nblocks == 0) 962 return 0; 963 964 map = kmem_alloc(XFS_DQITER_MAP_SIZE * sizeof(*map), KM_SLEEP); 965 966 lblkno = 0; 967 maxlblkcnt = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); 968 do { 969 uint lock_mode; 970 971 nmaps = XFS_DQITER_MAP_SIZE; 972 /* 973 * We aren't changing the inode itself. Just changing 974 * some of its data. No new blocks are added here, and 975 * the inode is never added to the transaction. 976 */ 977 lock_mode = xfs_ilock_data_map_shared(qip); 978 error = xfs_bmapi_read(qip, lblkno, maxlblkcnt - lblkno, 979 map, &nmaps, 0); 980 xfs_iunlock(qip, lock_mode); 981 if (error) 982 break; 983 984 ASSERT(nmaps <= XFS_DQITER_MAP_SIZE); 985 for (i = 0; i < nmaps; i++) { 986 ASSERT(map[i].br_startblock != DELAYSTARTBLOCK); 987 ASSERT(map[i].br_blockcount); 988 989 990 lblkno += map[i].br_blockcount; 991 992 if (map[i].br_startblock == HOLESTARTBLOCK) 993 continue; 994 995 firstid = (xfs_dqid_t) map[i].br_startoff * 996 mp->m_quotainfo->qi_dqperchunk; 997 /* 998 * Do a read-ahead on the next extent. 999 */ 1000 if ((i+1 < nmaps) && 1001 (map[i+1].br_startblock != HOLESTARTBLOCK)) { 1002 rablkcnt = map[i+1].br_blockcount; 1003 rablkno = map[i+1].br_startblock; 1004 while (rablkcnt--) { 1005 xfs_buf_readahead(mp->m_ddev_targp, 1006 XFS_FSB_TO_DADDR(mp, rablkno), 1007 mp->m_quotainfo->qi_dqchunklen, 1008 &xfs_dquot_buf_ops); 1009 rablkno++; 1010 } 1011 } 1012 /* 1013 * Iterate thru all the blks in the extent and 1014 * reset the counters of all the dquots inside them. 1015 */ 1016 error = xfs_qm_dqiter_bufs(mp, firstid, 1017 map[i].br_startblock, 1018 map[i].br_blockcount, 1019 flags, buffer_list); 1020 if (error) 1021 goto out; 1022 } 1023 } while (nmaps > 0); 1024 1025out: 1026 kmem_free(map); 1027 return error; 1028} 1029 1030/* 1031 * Called by dqusage_adjust in doing a quotacheck. 1032 * 1033 * Given the inode, and a dquot id this updates both the incore dqout as well 1034 * as the buffer copy. This is so that once the quotacheck is done, we can 1035 * just log all the buffers, as opposed to logging numerous updates to 1036 * individual dquots. 1037 */ 1038STATIC int 1039xfs_qm_quotacheck_dqadjust( 1040 struct xfs_inode *ip, 1041 xfs_dqid_t id, 1042 uint type, 1043 xfs_qcnt_t nblks, 1044 xfs_qcnt_t rtblks) 1045{ 1046 struct xfs_mount *mp = ip->i_mount; 1047 struct xfs_dquot *dqp; 1048 int error; 1049 1050 error = xfs_qm_dqget(mp, ip, id, type, 1051 XFS_QMOPT_DQALLOC | XFS_QMOPT_DOWARN, &dqp); 1052 if (error) { 1053 /* 1054 * Shouldn't be able to turn off quotas here. 1055 */ 1056 ASSERT(error != -ESRCH); 1057 ASSERT(error != -ENOENT); 1058 return error; 1059 } 1060 1061 trace_xfs_dqadjust(dqp); 1062 1063 /* 1064 * Adjust the inode count and the block count to reflect this inode's 1065 * resource usage. 1066 */ 1067 be64_add_cpu(&dqp->q_core.d_icount, 1); 1068 dqp->q_res_icount++; 1069 if (nblks) { 1070 be64_add_cpu(&dqp->q_core.d_bcount, nblks); 1071 dqp->q_res_bcount += nblks; 1072 } 1073 if (rtblks) { 1074 be64_add_cpu(&dqp->q_core.d_rtbcount, rtblks); 1075 dqp->q_res_rtbcount += rtblks; 1076 } 1077 1078 /* 1079 * Set default limits, adjust timers (since we changed usages) 1080 * 1081 * There are no timers for the default values set in the root dquot. 1082 */ 1083 if (dqp->q_core.d_id) { 1084 xfs_qm_adjust_dqlimits(mp, dqp); 1085 xfs_qm_adjust_dqtimers(mp, &dqp->q_core); 1086 } 1087 1088 dqp->dq_flags |= XFS_DQ_DIRTY; 1089 xfs_qm_dqput(dqp); 1090 return 0; 1091} 1092 1093STATIC int 1094xfs_qm_get_rtblks( 1095 xfs_inode_t *ip, 1096 xfs_qcnt_t *O_rtblks) 1097{ 1098 xfs_filblks_t rtblks; /* total rt blks */ 1099 xfs_extnum_t idx; /* extent record index */ 1100 xfs_ifork_t *ifp; /* inode fork pointer */ 1101 xfs_extnum_t nextents; /* number of extent entries */ 1102 int error; 1103 1104 ASSERT(XFS_IS_REALTIME_INODE(ip)); 1105 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK); 1106 if (!(ifp->if_flags & XFS_IFEXTENTS)) { 1107 if ((error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK))) 1108 return error; 1109 } 1110 rtblks = 0; 1111 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); 1112 for (idx = 0; idx < nextents; idx++) 1113 rtblks += xfs_bmbt_get_blockcount(xfs_iext_get_ext(ifp, idx)); 1114 *O_rtblks = (xfs_qcnt_t)rtblks; 1115 return 0; 1116} 1117 1118/* 1119 * callback routine supplied to bulkstat(). Given an inumber, find its 1120 * dquots and update them to account for resources taken by that inode. 1121 */ 1122/* ARGSUSED */ 1123STATIC int 1124xfs_qm_dqusage_adjust( 1125 xfs_mount_t *mp, /* mount point for filesystem */ 1126 xfs_ino_t ino, /* inode number to get data for */ 1127 void __user *buffer, /* not used */ 1128 int ubsize, /* not used */ 1129 int *ubused, /* not used */ 1130 int *res) /* result code value */ 1131{ 1132 xfs_inode_t *ip; 1133 xfs_qcnt_t nblks, rtblks = 0; 1134 int error; 1135 1136 ASSERT(XFS_IS_QUOTA_RUNNING(mp)); 1137 1138 /* 1139 * rootino must have its resources accounted for, not so with the quota 1140 * inodes. 1141 */ 1142 if (xfs_is_quota_inode(&mp->m_sb, ino)) { 1143 *res = BULKSTAT_RV_NOTHING; 1144 return -EINVAL; 1145 } 1146 1147 /* 1148 * We don't _need_ to take the ilock EXCL. However, the xfs_qm_dqget 1149 * interface expects the inode to be exclusively locked because that's 1150 * the case in all other instances. It's OK that we do this because 1151 * quotacheck is done only at mount time. 1152 */ 1153 error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_EXCL, &ip); 1154 if (error) { 1155 *res = BULKSTAT_RV_NOTHING; 1156 return error; 1157 } 1158 1159 ASSERT(ip->i_delayed_blks == 0); 1160 1161 if (XFS_IS_REALTIME_INODE(ip)) { 1162 /* 1163 * Walk thru the extent list and count the realtime blocks. 1164 */ 1165 error = xfs_qm_get_rtblks(ip, &rtblks); 1166 if (error) 1167 goto error0; 1168 } 1169 1170 nblks = (xfs_qcnt_t)ip->i_d.di_nblocks - rtblks; 1171 1172 /* 1173 * Add the (disk blocks and inode) resources occupied by this 1174 * inode to its dquots. We do this adjustment in the incore dquot, 1175 * and also copy the changes to its buffer. 1176 * We don't care about putting these changes in a transaction 1177 * envelope because if we crash in the middle of a 'quotacheck' 1178 * we have to start from the beginning anyway. 1179 * Once we're done, we'll log all the dquot bufs. 1180 * 1181 * The *QUOTA_ON checks below may look pretty racy, but quotachecks 1182 * and quotaoffs don't race. (Quotachecks happen at mount time only). 1183 */ 1184 if (XFS_IS_UQUOTA_ON(mp)) { 1185 error = xfs_qm_quotacheck_dqadjust(ip, ip->i_d.di_uid, 1186 XFS_DQ_USER, nblks, rtblks); 1187 if (error) 1188 goto error0; 1189 } 1190 1191 if (XFS_IS_GQUOTA_ON(mp)) { 1192 error = xfs_qm_quotacheck_dqadjust(ip, ip->i_d.di_gid, 1193 XFS_DQ_GROUP, nblks, rtblks); 1194 if (error) 1195 goto error0; 1196 } 1197 1198 if (XFS_IS_PQUOTA_ON(mp)) { 1199 error = xfs_qm_quotacheck_dqadjust(ip, xfs_get_projid(ip), 1200 XFS_DQ_PROJ, nblks, rtblks); 1201 if (error) 1202 goto error0; 1203 } 1204 1205 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1206 IRELE(ip); 1207 *res = BULKSTAT_RV_DIDONE; 1208 return 0; 1209 1210error0: 1211 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1212 IRELE(ip); 1213 *res = BULKSTAT_RV_GIVEUP; 1214 return error; 1215} 1216 1217STATIC int 1218xfs_qm_flush_one( 1219 struct xfs_dquot *dqp, 1220 void *data) 1221{ 1222 struct list_head *buffer_list = data; 1223 struct xfs_buf *bp = NULL; 1224 int error = 0; 1225 1226 xfs_dqlock(dqp); 1227 if (dqp->dq_flags & XFS_DQ_FREEING) 1228 goto out_unlock; 1229 if (!XFS_DQ_IS_DIRTY(dqp)) 1230 goto out_unlock; 1231 1232 xfs_dqflock(dqp); 1233 error = xfs_qm_dqflush(dqp, &bp); 1234 if (error) 1235 goto out_unlock; 1236 1237 xfs_buf_delwri_queue(bp, buffer_list); 1238 xfs_buf_relse(bp); 1239out_unlock: 1240 xfs_dqunlock(dqp); 1241 return error; 1242} 1243 1244/* 1245 * Walk thru all the filesystem inodes and construct a consistent view 1246 * of the disk quota world. If the quotacheck fails, disable quotas. 1247 */ 1248STATIC int 1249xfs_qm_quotacheck( 1250 xfs_mount_t *mp) 1251{ 1252 int done, count, error, error2; 1253 xfs_ino_t lastino; 1254 size_t structsz; 1255 uint flags; 1256 LIST_HEAD (buffer_list); 1257 struct xfs_inode *uip = mp->m_quotainfo->qi_uquotaip; 1258 struct xfs_inode *gip = mp->m_quotainfo->qi_gquotaip; 1259 struct xfs_inode *pip = mp->m_quotainfo->qi_pquotaip; 1260 1261 count = INT_MAX; 1262 structsz = 1; 1263 lastino = 0; 1264 flags = 0; 1265 1266 ASSERT(uip || gip || pip); 1267 ASSERT(XFS_IS_QUOTA_RUNNING(mp)); 1268 1269 xfs_notice(mp, "Quotacheck needed: Please wait."); 1270 1271 /* 1272 * First we go thru all the dquots on disk, USR and GRP/PRJ, and reset 1273 * their counters to zero. We need a clean slate. 1274 * We don't log our changes till later. 1275 */ 1276 if (uip) { 1277 error = xfs_qm_dqiterate(mp, uip, XFS_QMOPT_UQUOTA, 1278 &buffer_list); 1279 if (error) 1280 goto error_return; 1281 flags |= XFS_UQUOTA_CHKD; 1282 } 1283 1284 if (gip) { 1285 error = xfs_qm_dqiterate(mp, gip, XFS_QMOPT_GQUOTA, 1286 &buffer_list); 1287 if (error) 1288 goto error_return; 1289 flags |= XFS_GQUOTA_CHKD; 1290 } 1291 1292 if (pip) { 1293 error = xfs_qm_dqiterate(mp, pip, XFS_QMOPT_PQUOTA, 1294 &buffer_list); 1295 if (error) 1296 goto error_return; 1297 flags |= XFS_PQUOTA_CHKD; 1298 } 1299 1300 do { 1301 /* 1302 * Iterate thru all the inodes in the file system, 1303 * adjusting the corresponding dquot counters in core. 1304 */ 1305 error = xfs_bulkstat(mp, &lastino, &count, 1306 xfs_qm_dqusage_adjust, 1307 structsz, NULL, &done); 1308 if (error) 1309 break; 1310 1311 } while (!done); 1312 1313 /* 1314 * We've made all the changes that we need to make incore. Flush them 1315 * down to disk buffers if everything was updated successfully. 1316 */ 1317 if (XFS_IS_UQUOTA_ON(mp)) { 1318 error = xfs_qm_dquot_walk(mp, XFS_DQ_USER, xfs_qm_flush_one, 1319 &buffer_list); 1320 } 1321 if (XFS_IS_GQUOTA_ON(mp)) { 1322 error2 = xfs_qm_dquot_walk(mp, XFS_DQ_GROUP, xfs_qm_flush_one, 1323 &buffer_list); 1324 if (!error) 1325 error = error2; 1326 } 1327 if (XFS_IS_PQUOTA_ON(mp)) { 1328 error2 = xfs_qm_dquot_walk(mp, XFS_DQ_PROJ, xfs_qm_flush_one, 1329 &buffer_list); 1330 if (!error) 1331 error = error2; 1332 } 1333 1334 error2 = xfs_buf_delwri_submit(&buffer_list); 1335 if (!error) 1336 error = error2; 1337 1338 /* 1339 * We can get this error if we couldn't do a dquot allocation inside 1340 * xfs_qm_dqusage_adjust (via bulkstat). We don't care about the 1341 * dirty dquots that might be cached, we just want to get rid of them 1342 * and turn quotaoff. The dquots won't be attached to any of the inodes 1343 * at this point (because we intentionally didn't in dqget_noattach). 1344 */ 1345 if (error) { 1346 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL); 1347 goto error_return; 1348 } 1349 1350 /* 1351 * If one type of quotas is off, then it will lose its 1352 * quotachecked status, since we won't be doing accounting for 1353 * that type anymore. 1354 */ 1355 mp->m_qflags &= ~XFS_ALL_QUOTA_CHKD; 1356 mp->m_qflags |= flags; 1357 1358 error_return: 1359 while (!list_empty(&buffer_list)) { 1360 struct xfs_buf *bp = 1361 list_first_entry(&buffer_list, struct xfs_buf, b_list); 1362 list_del_init(&bp->b_list); 1363 xfs_buf_relse(bp); 1364 } 1365 1366 if (error) { 1367 xfs_warn(mp, 1368 "Quotacheck: Unsuccessful (Error %d): Disabling quotas.", 1369 error); 1370 /* 1371 * We must turn off quotas. 1372 */ 1373 ASSERT(mp->m_quotainfo != NULL); 1374 xfs_qm_destroy_quotainfo(mp); 1375 if (xfs_mount_reset_sbqflags(mp)) { 1376 xfs_warn(mp, 1377 "Quotacheck: Failed to reset quota flags."); 1378 } 1379 } else 1380 xfs_notice(mp, "Quotacheck: Done."); 1381 return error; 1382} 1383 1384/* 1385 * This is called from xfs_mountfs to start quotas and initialize all 1386 * necessary data structures like quotainfo. This is also responsible for 1387 * running a quotacheck as necessary. We are guaranteed that the superblock 1388 * is consistently read in at this point. 1389 * 1390 * If we fail here, the mount will continue with quota turned off. We don't 1391 * need to inidicate success or failure at all. 1392 */ 1393void 1394xfs_qm_mount_quotas( 1395 struct xfs_mount *mp) 1396{ 1397 int error = 0; 1398 uint sbf; 1399 1400 /* 1401 * If quotas on realtime volumes is not supported, we disable 1402 * quotas immediately. 1403 */ 1404 if (mp->m_sb.sb_rextents) { 1405 xfs_notice(mp, "Cannot turn on quotas for realtime filesystem"); 1406 mp->m_qflags = 0; 1407 goto write_changes; 1408 } 1409 1410 ASSERT(XFS_IS_QUOTA_RUNNING(mp)); 1411 1412 /* 1413 * Allocate the quotainfo structure inside the mount struct, and 1414 * create quotainode(s), and change/rev superblock if necessary. 1415 */ 1416 error = xfs_qm_init_quotainfo(mp); 1417 if (error) { 1418 /* 1419 * We must turn off quotas. 1420 */ 1421 ASSERT(mp->m_quotainfo == NULL); 1422 mp->m_qflags = 0; 1423 goto write_changes; 1424 } 1425 /* 1426 * If any of the quotas are not consistent, do a quotacheck. 1427 */ 1428 if (XFS_QM_NEED_QUOTACHECK(mp)) { 1429 error = xfs_qm_quotacheck(mp); 1430 if (error) { 1431 /* Quotacheck failed and disabled quotas. */ 1432 return; 1433 } 1434 } 1435 /* 1436 * If one type of quotas is off, then it will lose its 1437 * quotachecked status, since we won't be doing accounting for 1438 * that type anymore. 1439 */ 1440 if (!XFS_IS_UQUOTA_ON(mp)) 1441 mp->m_qflags &= ~XFS_UQUOTA_CHKD; 1442 if (!XFS_IS_GQUOTA_ON(mp)) 1443 mp->m_qflags &= ~XFS_GQUOTA_CHKD; 1444 if (!XFS_IS_PQUOTA_ON(mp)) 1445 mp->m_qflags &= ~XFS_PQUOTA_CHKD; 1446 1447 write_changes: 1448 /* 1449 * We actually don't have to acquire the m_sb_lock at all. 1450 * This can only be called from mount, and that's single threaded. XXX 1451 */ 1452 spin_lock(&mp->m_sb_lock); 1453 sbf = mp->m_sb.sb_qflags; 1454 mp->m_sb.sb_qflags = mp->m_qflags & XFS_MOUNT_QUOTA_ALL; 1455 spin_unlock(&mp->m_sb_lock); 1456 1457 if (sbf != (mp->m_qflags & XFS_MOUNT_QUOTA_ALL)) { 1458 if (xfs_sync_sb(mp, false)) { 1459 /* 1460 * We could only have been turning quotas off. 1461 * We aren't in very good shape actually because 1462 * the incore structures are convinced that quotas are 1463 * off, but the on disk superblock doesn't know that ! 1464 */ 1465 ASSERT(!(XFS_IS_QUOTA_RUNNING(mp))); 1466 xfs_alert(mp, "%s: Superblock update failed!", 1467 __func__); 1468 } 1469 } 1470 1471 if (error) { 1472 xfs_warn(mp, "Failed to initialize disk quotas."); 1473 return; 1474 } 1475} 1476 1477/* 1478 * This is called after the superblock has been read in and we're ready to 1479 * iget the quota inodes. 1480 */ 1481STATIC int 1482xfs_qm_init_quotainos( 1483 xfs_mount_t *mp) 1484{ 1485 struct xfs_inode *uip = NULL; 1486 struct xfs_inode *gip = NULL; 1487 struct xfs_inode *pip = NULL; 1488 int error; 1489 uint flags = 0; 1490 1491 ASSERT(mp->m_quotainfo); 1492 1493 /* 1494 * Get the uquota and gquota inodes 1495 */ 1496 if (xfs_sb_version_hasquota(&mp->m_sb)) { 1497 if (XFS_IS_UQUOTA_ON(mp) && 1498 mp->m_sb.sb_uquotino != NULLFSINO) { 1499 ASSERT(mp->m_sb.sb_uquotino > 0); 1500 error = xfs_iget(mp, NULL, mp->m_sb.sb_uquotino, 1501 0, 0, &uip); 1502 if (error) 1503 return error; 1504 } 1505 if (XFS_IS_GQUOTA_ON(mp) && 1506 mp->m_sb.sb_gquotino != NULLFSINO) { 1507 ASSERT(mp->m_sb.sb_gquotino > 0); 1508 error = xfs_iget(mp, NULL, mp->m_sb.sb_gquotino, 1509 0, 0, &gip); 1510 if (error) 1511 goto error_rele; 1512 } 1513 if (XFS_IS_PQUOTA_ON(mp) && 1514 mp->m_sb.sb_pquotino != NULLFSINO) { 1515 ASSERT(mp->m_sb.sb_pquotino > 0); 1516 error = xfs_iget(mp, NULL, mp->m_sb.sb_pquotino, 1517 0, 0, &pip); 1518 if (error) 1519 goto error_rele; 1520 } 1521 } else { 1522 flags |= XFS_QMOPT_SBVERSION; 1523 } 1524 1525 /* 1526 * Create the three inodes, if they don't exist already. The changes 1527 * made above will get added to a transaction and logged in one of 1528 * the qino_alloc calls below. If the device is readonly, 1529 * temporarily switch to read-write to do this. 1530 */ 1531 if (XFS_IS_UQUOTA_ON(mp) && uip == NULL) { 1532 error = xfs_qm_qino_alloc(mp, &uip, 1533 flags | XFS_QMOPT_UQUOTA); 1534 if (error) 1535 goto error_rele; 1536 1537 flags &= ~XFS_QMOPT_SBVERSION; 1538 } 1539 if (XFS_IS_GQUOTA_ON(mp) && gip == NULL) { 1540 error = xfs_qm_qino_alloc(mp, &gip, 1541 flags | XFS_QMOPT_GQUOTA); 1542 if (error) 1543 goto error_rele; 1544 1545 flags &= ~XFS_QMOPT_SBVERSION; 1546 } 1547 if (XFS_IS_PQUOTA_ON(mp) && pip == NULL) { 1548 error = xfs_qm_qino_alloc(mp, &pip, 1549 flags | XFS_QMOPT_PQUOTA); 1550 if (error) 1551 goto error_rele; 1552 } 1553 1554 mp->m_quotainfo->qi_uquotaip = uip; 1555 mp->m_quotainfo->qi_gquotaip = gip; 1556 mp->m_quotainfo->qi_pquotaip = pip; 1557 1558 return 0; 1559 1560error_rele: 1561 if (uip) 1562 IRELE(uip); 1563 if (gip) 1564 IRELE(gip); 1565 if (pip) 1566 IRELE(pip); 1567 return error; 1568} 1569 1570STATIC void 1571xfs_qm_dqfree_one( 1572 struct xfs_dquot *dqp) 1573{ 1574 struct xfs_mount *mp = dqp->q_mount; 1575 struct xfs_quotainfo *qi = mp->m_quotainfo; 1576 1577 mutex_lock(&qi->qi_tree_lock); 1578 radix_tree_delete(xfs_dquot_tree(qi, dqp->q_core.d_flags), 1579 be32_to_cpu(dqp->q_core.d_id)); 1580 1581 qi->qi_dquots--; 1582 mutex_unlock(&qi->qi_tree_lock); 1583 1584 xfs_qm_dqdestroy(dqp); 1585} 1586 1587/* --------------- utility functions for vnodeops ---------------- */ 1588 1589 1590/* 1591 * Given an inode, a uid, gid and prid make sure that we have 1592 * allocated relevant dquot(s) on disk, and that we won't exceed inode 1593 * quotas by creating this file. 1594 * This also attaches dquot(s) to the given inode after locking it, 1595 * and returns the dquots corresponding to the uid and/or gid. 1596 * 1597 * in : inode (unlocked) 1598 * out : udquot, gdquot with references taken and unlocked 1599 */ 1600int 1601xfs_qm_vop_dqalloc( 1602 struct xfs_inode *ip, 1603 xfs_dqid_t uid, 1604 xfs_dqid_t gid, 1605 prid_t prid, 1606 uint flags, 1607 struct xfs_dquot **O_udqpp, 1608 struct xfs_dquot **O_gdqpp, 1609 struct xfs_dquot **O_pdqpp) 1610{ 1611 struct xfs_mount *mp = ip->i_mount; 1612 struct xfs_dquot *uq = NULL; 1613 struct xfs_dquot *gq = NULL; 1614 struct xfs_dquot *pq = NULL; 1615 int error; 1616 uint lockflags; 1617 1618 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp)) 1619 return 0; 1620 1621 lockflags = XFS_ILOCK_EXCL; 1622 xfs_ilock(ip, lockflags); 1623 1624 if ((flags & XFS_QMOPT_INHERIT) && XFS_INHERIT_GID(ip)) 1625 gid = ip->i_d.di_gid; 1626 1627 /* 1628 * Attach the dquot(s) to this inode, doing a dquot allocation 1629 * if necessary. The dquot(s) will not be locked. 1630 */ 1631 if (XFS_NOT_DQATTACHED(mp, ip)) { 1632 error = xfs_qm_dqattach_locked(ip, XFS_QMOPT_DQALLOC); 1633 if (error) { 1634 xfs_iunlock(ip, lockflags); 1635 return error; 1636 } 1637 } 1638 1639 if ((flags & XFS_QMOPT_UQUOTA) && XFS_IS_UQUOTA_ON(mp)) { 1640 if (ip->i_d.di_uid != uid) { 1641 /* 1642 * What we need is the dquot that has this uid, and 1643 * if we send the inode to dqget, the uid of the inode 1644 * takes priority over what's sent in the uid argument. 1645 * We must unlock inode here before calling dqget if 1646 * we're not sending the inode, because otherwise 1647 * we'll deadlock by doing trans_reserve while 1648 * holding ilock. 1649 */ 1650 xfs_iunlock(ip, lockflags); 1651 error = xfs_qm_dqget(mp, NULL, uid, 1652 XFS_DQ_USER, 1653 XFS_QMOPT_DQALLOC | 1654 XFS_QMOPT_DOWARN, 1655 &uq); 1656 if (error) { 1657 ASSERT(error != -ENOENT); 1658 return error; 1659 } 1660 /* 1661 * Get the ilock in the right order. 1662 */ 1663 xfs_dqunlock(uq); 1664 lockflags = XFS_ILOCK_SHARED; 1665 xfs_ilock(ip, lockflags); 1666 } else { 1667 /* 1668 * Take an extra reference, because we'll return 1669 * this to caller 1670 */ 1671 ASSERT(ip->i_udquot); 1672 uq = xfs_qm_dqhold(ip->i_udquot); 1673 } 1674 } 1675 if ((flags & XFS_QMOPT_GQUOTA) && XFS_IS_GQUOTA_ON(mp)) { 1676 if (ip->i_d.di_gid != gid) { 1677 xfs_iunlock(ip, lockflags); 1678 error = xfs_qm_dqget(mp, NULL, gid, 1679 XFS_DQ_GROUP, 1680 XFS_QMOPT_DQALLOC | 1681 XFS_QMOPT_DOWARN, 1682 &gq); 1683 if (error) { 1684 ASSERT(error != -ENOENT); 1685 goto error_rele; 1686 } 1687 xfs_dqunlock(gq); 1688 lockflags = XFS_ILOCK_SHARED; 1689 xfs_ilock(ip, lockflags); 1690 } else { 1691 ASSERT(ip->i_gdquot); 1692 gq = xfs_qm_dqhold(ip->i_gdquot); 1693 } 1694 } 1695 if ((flags & XFS_QMOPT_PQUOTA) && XFS_IS_PQUOTA_ON(mp)) { 1696 if (xfs_get_projid(ip) != prid) { 1697 xfs_iunlock(ip, lockflags); 1698 error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)prid, 1699 XFS_DQ_PROJ, 1700 XFS_QMOPT_DQALLOC | 1701 XFS_QMOPT_DOWARN, 1702 &pq); 1703 if (error) { 1704 ASSERT(error != -ENOENT); 1705 goto error_rele; 1706 } 1707 xfs_dqunlock(pq); 1708 lockflags = XFS_ILOCK_SHARED; 1709 xfs_ilock(ip, lockflags); 1710 } else { 1711 ASSERT(ip->i_pdquot); 1712 pq = xfs_qm_dqhold(ip->i_pdquot); 1713 } 1714 } 1715 if (uq) 1716 trace_xfs_dquot_dqalloc(ip); 1717 1718 xfs_iunlock(ip, lockflags); 1719 if (O_udqpp) 1720 *O_udqpp = uq; 1721 else 1722 xfs_qm_dqrele(uq); 1723 if (O_gdqpp) 1724 *O_gdqpp = gq; 1725 else 1726 xfs_qm_dqrele(gq); 1727 if (O_pdqpp) 1728 *O_pdqpp = pq; 1729 else 1730 xfs_qm_dqrele(pq); 1731 return 0; 1732 1733error_rele: 1734 xfs_qm_dqrele(gq); 1735 xfs_qm_dqrele(uq); 1736 return error; 1737} 1738 1739/* 1740 * Actually transfer ownership, and do dquot modifications. 1741 * These were already reserved. 1742 */ 1743xfs_dquot_t * 1744xfs_qm_vop_chown( 1745 xfs_trans_t *tp, 1746 xfs_inode_t *ip, 1747 xfs_dquot_t **IO_olddq, 1748 xfs_dquot_t *newdq) 1749{ 1750 xfs_dquot_t *prevdq; 1751 uint bfield = XFS_IS_REALTIME_INODE(ip) ? 1752 XFS_TRANS_DQ_RTBCOUNT : XFS_TRANS_DQ_BCOUNT; 1753 1754 1755 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); 1756 ASSERT(XFS_IS_QUOTA_RUNNING(ip->i_mount)); 1757 1758 /* old dquot */ 1759 prevdq = *IO_olddq; 1760 ASSERT(prevdq); 1761 ASSERT(prevdq != newdq); 1762 1763 xfs_trans_mod_dquot(tp, prevdq, bfield, -(ip->i_d.di_nblocks)); 1764 xfs_trans_mod_dquot(tp, prevdq, XFS_TRANS_DQ_ICOUNT, -1); 1765 1766 /* the sparkling new dquot */ 1767 xfs_trans_mod_dquot(tp, newdq, bfield, ip->i_d.di_nblocks); 1768 xfs_trans_mod_dquot(tp, newdq, XFS_TRANS_DQ_ICOUNT, 1); 1769 1770 /* 1771 * Take an extra reference, because the inode is going to keep 1772 * this dquot pointer even after the trans_commit. 1773 */ 1774 *IO_olddq = xfs_qm_dqhold(newdq); 1775 1776 return prevdq; 1777} 1778 1779/* 1780 * Quota reservations for setattr(AT_UID|AT_GID|AT_PROJID). 1781 */ 1782int 1783xfs_qm_vop_chown_reserve( 1784 struct xfs_trans *tp, 1785 struct xfs_inode *ip, 1786 struct xfs_dquot *udqp, 1787 struct xfs_dquot *gdqp, 1788 struct xfs_dquot *pdqp, 1789 uint flags) 1790{ 1791 struct xfs_mount *mp = ip->i_mount; 1792 uint delblks, blkflags, prjflags = 0; 1793 struct xfs_dquot *udq_unres = NULL; 1794 struct xfs_dquot *gdq_unres = NULL; 1795 struct xfs_dquot *pdq_unres = NULL; 1796 struct xfs_dquot *udq_delblks = NULL; 1797 struct xfs_dquot *gdq_delblks = NULL; 1798 struct xfs_dquot *pdq_delblks = NULL; 1799 int error; 1800 1801 1802 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); 1803 ASSERT(XFS_IS_QUOTA_RUNNING(mp)); 1804 1805 delblks = ip->i_delayed_blks; 1806 blkflags = XFS_IS_REALTIME_INODE(ip) ? 1807 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS; 1808 1809 if (XFS_IS_UQUOTA_ON(mp) && udqp && 1810 ip->i_d.di_uid != be32_to_cpu(udqp->q_core.d_id)) { 1811 udq_delblks = udqp; 1812 /* 1813 * If there are delayed allocation blocks, then we have to 1814 * unreserve those from the old dquot, and add them to the 1815 * new dquot. 1816 */ 1817 if (delblks) { 1818 ASSERT(ip->i_udquot); 1819 udq_unres = ip->i_udquot; 1820 } 1821 } 1822 if (XFS_IS_GQUOTA_ON(ip->i_mount) && gdqp && 1823 ip->i_d.di_gid != be32_to_cpu(gdqp->q_core.d_id)) { 1824 gdq_delblks = gdqp; 1825 if (delblks) { 1826 ASSERT(ip->i_gdquot); 1827 gdq_unres = ip->i_gdquot; 1828 } 1829 } 1830 1831 if (XFS_IS_PQUOTA_ON(ip->i_mount) && pdqp && 1832 xfs_get_projid(ip) != be32_to_cpu(pdqp->q_core.d_id)) { 1833 prjflags = XFS_QMOPT_ENOSPC; 1834 pdq_delblks = pdqp; 1835 if (delblks) { 1836 ASSERT(ip->i_pdquot); 1837 pdq_unres = ip->i_pdquot; 1838 } 1839 } 1840 1841 error = xfs_trans_reserve_quota_bydquots(tp, ip->i_mount, 1842 udq_delblks, gdq_delblks, pdq_delblks, 1843 ip->i_d.di_nblocks, 1, 1844 flags | blkflags | prjflags); 1845 if (error) 1846 return error; 1847 1848 /* 1849 * Do the delayed blks reservations/unreservations now. Since, these 1850 * are done without the help of a transaction, if a reservation fails 1851 * its previous reservations won't be automatically undone by trans 1852 * code. So, we have to do it manually here. 1853 */ 1854 if (delblks) { 1855 /* 1856 * Do the reservations first. Unreservation can't fail. 1857 */ 1858 ASSERT(udq_delblks || gdq_delblks || pdq_delblks); 1859 ASSERT(udq_unres || gdq_unres || pdq_unres); 1860 error = xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount, 1861 udq_delblks, gdq_delblks, pdq_delblks, 1862 (xfs_qcnt_t)delblks, 0, 1863 flags | blkflags | prjflags); 1864 if (error) 1865 return error; 1866 xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount, 1867 udq_unres, gdq_unres, pdq_unres, 1868 -((xfs_qcnt_t)delblks), 0, blkflags); 1869 } 1870 1871 return 0; 1872} 1873 1874int 1875xfs_qm_vop_rename_dqattach( 1876 struct xfs_inode **i_tab) 1877{ 1878 struct xfs_mount *mp = i_tab[0]->i_mount; 1879 int i; 1880 1881 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp)) 1882 return 0; 1883 1884 for (i = 0; (i < 4 && i_tab[i]); i++) { 1885 struct xfs_inode *ip = i_tab[i]; 1886 int error; 1887 1888 /* 1889 * Watch out for duplicate entries in the table. 1890 */ 1891 if (i == 0 || ip != i_tab[i-1]) { 1892 if (XFS_NOT_DQATTACHED(mp, ip)) { 1893 error = xfs_qm_dqattach(ip, 0); 1894 if (error) 1895 return error; 1896 } 1897 } 1898 } 1899 return 0; 1900} 1901 1902void 1903xfs_qm_vop_create_dqattach( 1904 struct xfs_trans *tp, 1905 struct xfs_inode *ip, 1906 struct xfs_dquot *udqp, 1907 struct xfs_dquot *gdqp, 1908 struct xfs_dquot *pdqp) 1909{ 1910 struct xfs_mount *mp = tp->t_mountp; 1911 1912 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp)) 1913 return; 1914 1915 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); 1916 ASSERT(XFS_IS_QUOTA_RUNNING(mp)); 1917 1918 if (udqp && XFS_IS_UQUOTA_ON(mp)) { 1919 ASSERT(ip->i_udquot == NULL); 1920 ASSERT(ip->i_d.di_uid == be32_to_cpu(udqp->q_core.d_id)); 1921 1922 ip->i_udquot = xfs_qm_dqhold(udqp); 1923 xfs_trans_mod_dquot(tp, udqp, XFS_TRANS_DQ_ICOUNT, 1); 1924 } 1925 if (gdqp && XFS_IS_GQUOTA_ON(mp)) { 1926 ASSERT(ip->i_gdquot == NULL); 1927 ASSERT(ip->i_d.di_gid == be32_to_cpu(gdqp->q_core.d_id)); 1928 ip->i_gdquot = xfs_qm_dqhold(gdqp); 1929 xfs_trans_mod_dquot(tp, gdqp, XFS_TRANS_DQ_ICOUNT, 1); 1930 } 1931 if (pdqp && XFS_IS_PQUOTA_ON(mp)) { 1932 ASSERT(ip->i_pdquot == NULL); 1933 ASSERT(xfs_get_projid(ip) == be32_to_cpu(pdqp->q_core.d_id)); 1934 1935 ip->i_pdquot = xfs_qm_dqhold(pdqp); 1936 xfs_trans_mod_dquot(tp, pdqp, XFS_TRANS_DQ_ICOUNT, 1); 1937 } 1938} 1939 1940