1 /* 2 * Copyright (c) 2000-2003 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_format.h" 21 #include "xfs_log_format.h" 22 #include "xfs_shared.h" 23 #include "xfs_trans_resv.h" 24 #include "xfs_bit.h" 25 #include "xfs_mount.h" 26 #include "xfs_defer.h" 27 #include "xfs_inode.h" 28 #include "xfs_bmap.h" 29 #include "xfs_bmap_util.h" 30 #include "xfs_alloc.h" 31 #include "xfs_quota.h" 32 #include "xfs_error.h" 33 #include "xfs_trans.h" 34 #include "xfs_buf_item.h" 35 #include "xfs_trans_space.h" 36 #include "xfs_trans_priv.h" 37 #include "xfs_qm.h" 38 #include "xfs_cksum.h" 39 #include "xfs_trace.h" 40 #include "xfs_log.h" 41 #include "xfs_bmap_btree.h" 42 43 /* 44 * Lock order: 45 * 46 * ip->i_lock 47 * qi->qi_tree_lock 48 * dquot->q_qlock (xfs_dqlock() and friends) 49 * dquot->q_flush (xfs_dqflock() and friends) 50 * qi->qi_lru_lock 51 * 52 * If two dquots need to be locked the order is user before group/project, 53 * otherwise by the lowest id first, see xfs_dqlock2. 54 */ 55 56 struct kmem_zone *xfs_qm_dqtrxzone; 57 static struct kmem_zone *xfs_qm_dqzone; 58 59 static struct lock_class_key xfs_dquot_group_class; 60 static struct lock_class_key xfs_dquot_project_class; 61 62 /* 63 * This is called to free all the memory associated with a dquot 64 */ 65 void 66 xfs_qm_dqdestroy( 67 xfs_dquot_t *dqp) 68 { 69 ASSERT(list_empty(&dqp->q_lru)); 70 71 kmem_free(dqp->q_logitem.qli_item.li_lv_shadow); 72 mutex_destroy(&dqp->q_qlock); 73 74 XFS_STATS_DEC(dqp->q_mount, xs_qm_dquot); 75 kmem_zone_free(xfs_qm_dqzone, dqp); 76 } 77 78 /* 79 * If default limits are in force, push them into the dquot now. 80 * We overwrite the dquot limits only if they are zero and this 81 * is not the root dquot. 82 */ 83 void 84 xfs_qm_adjust_dqlimits( 85 struct xfs_mount *mp, 86 struct xfs_dquot *dq) 87 { 88 struct xfs_quotainfo *q = mp->m_quotainfo; 89 struct xfs_disk_dquot *d = &dq->q_core; 90 struct xfs_def_quota *defq; 91 int prealloc = 0; 92 93 ASSERT(d->d_id); 94 defq = xfs_get_defquota(dq, q); 95 96 if (defq->bsoftlimit && !d->d_blk_softlimit) { 97 d->d_blk_softlimit = cpu_to_be64(defq->bsoftlimit); 98 prealloc = 1; 99 } 100 if (defq->bhardlimit && !d->d_blk_hardlimit) { 101 d->d_blk_hardlimit = cpu_to_be64(defq->bhardlimit); 102 prealloc = 1; 103 } 104 if (defq->isoftlimit && !d->d_ino_softlimit) 105 d->d_ino_softlimit = cpu_to_be64(defq->isoftlimit); 106 if (defq->ihardlimit && !d->d_ino_hardlimit) 107 d->d_ino_hardlimit = cpu_to_be64(defq->ihardlimit); 108 if (defq->rtbsoftlimit && !d->d_rtb_softlimit) 109 d->d_rtb_softlimit = cpu_to_be64(defq->rtbsoftlimit); 110 if (defq->rtbhardlimit && !d->d_rtb_hardlimit) 111 d->d_rtb_hardlimit = cpu_to_be64(defq->rtbhardlimit); 112 113 if (prealloc) 114 xfs_dquot_set_prealloc_limits(dq); 115 } 116 117 /* 118 * Check the limits and timers of a dquot and start or reset timers 119 * if necessary. 120 * This gets called even when quota enforcement is OFF, which makes our 121 * life a little less complicated. (We just don't reject any quota 122 * reservations in that case, when enforcement is off). 123 * We also return 0 as the values of the timers in Q_GETQUOTA calls, when 124 * enforcement's off. 125 * In contrast, warnings are a little different in that they don't 126 * 'automatically' get started when limits get exceeded. They do 127 * get reset to zero, however, when we find the count to be under 128 * the soft limit (they are only ever set non-zero via userspace). 129 */ 130 void 131 xfs_qm_adjust_dqtimers( 132 xfs_mount_t *mp, 133 xfs_disk_dquot_t *d) 134 { 135 ASSERT(d->d_id); 136 137 #ifdef DEBUG 138 if (d->d_blk_hardlimit) 139 ASSERT(be64_to_cpu(d->d_blk_softlimit) <= 140 be64_to_cpu(d->d_blk_hardlimit)); 141 if (d->d_ino_hardlimit) 142 ASSERT(be64_to_cpu(d->d_ino_softlimit) <= 143 be64_to_cpu(d->d_ino_hardlimit)); 144 if (d->d_rtb_hardlimit) 145 ASSERT(be64_to_cpu(d->d_rtb_softlimit) <= 146 be64_to_cpu(d->d_rtb_hardlimit)); 147 #endif 148 149 if (!d->d_btimer) { 150 if ((d->d_blk_softlimit && 151 (be64_to_cpu(d->d_bcount) > 152 be64_to_cpu(d->d_blk_softlimit))) || 153 (d->d_blk_hardlimit && 154 (be64_to_cpu(d->d_bcount) > 155 be64_to_cpu(d->d_blk_hardlimit)))) { 156 d->d_btimer = cpu_to_be32(get_seconds() + 157 mp->m_quotainfo->qi_btimelimit); 158 } else { 159 d->d_bwarns = 0; 160 } 161 } else { 162 if ((!d->d_blk_softlimit || 163 (be64_to_cpu(d->d_bcount) <= 164 be64_to_cpu(d->d_blk_softlimit))) && 165 (!d->d_blk_hardlimit || 166 (be64_to_cpu(d->d_bcount) <= 167 be64_to_cpu(d->d_blk_hardlimit)))) { 168 d->d_btimer = 0; 169 } 170 } 171 172 if (!d->d_itimer) { 173 if ((d->d_ino_softlimit && 174 (be64_to_cpu(d->d_icount) > 175 be64_to_cpu(d->d_ino_softlimit))) || 176 (d->d_ino_hardlimit && 177 (be64_to_cpu(d->d_icount) > 178 be64_to_cpu(d->d_ino_hardlimit)))) { 179 d->d_itimer = cpu_to_be32(get_seconds() + 180 mp->m_quotainfo->qi_itimelimit); 181 } else { 182 d->d_iwarns = 0; 183 } 184 } else { 185 if ((!d->d_ino_softlimit || 186 (be64_to_cpu(d->d_icount) <= 187 be64_to_cpu(d->d_ino_softlimit))) && 188 (!d->d_ino_hardlimit || 189 (be64_to_cpu(d->d_icount) <= 190 be64_to_cpu(d->d_ino_hardlimit)))) { 191 d->d_itimer = 0; 192 } 193 } 194 195 if (!d->d_rtbtimer) { 196 if ((d->d_rtb_softlimit && 197 (be64_to_cpu(d->d_rtbcount) > 198 be64_to_cpu(d->d_rtb_softlimit))) || 199 (d->d_rtb_hardlimit && 200 (be64_to_cpu(d->d_rtbcount) > 201 be64_to_cpu(d->d_rtb_hardlimit)))) { 202 d->d_rtbtimer = cpu_to_be32(get_seconds() + 203 mp->m_quotainfo->qi_rtbtimelimit); 204 } else { 205 d->d_rtbwarns = 0; 206 } 207 } else { 208 if ((!d->d_rtb_softlimit || 209 (be64_to_cpu(d->d_rtbcount) <= 210 be64_to_cpu(d->d_rtb_softlimit))) && 211 (!d->d_rtb_hardlimit || 212 (be64_to_cpu(d->d_rtbcount) <= 213 be64_to_cpu(d->d_rtb_hardlimit)))) { 214 d->d_rtbtimer = 0; 215 } 216 } 217 } 218 219 /* 220 * initialize a buffer full of dquots and log the whole thing 221 */ 222 STATIC void 223 xfs_qm_init_dquot_blk( 224 xfs_trans_t *tp, 225 xfs_mount_t *mp, 226 xfs_dqid_t id, 227 uint type, 228 xfs_buf_t *bp) 229 { 230 struct xfs_quotainfo *q = mp->m_quotainfo; 231 xfs_dqblk_t *d; 232 xfs_dqid_t curid; 233 int i; 234 235 ASSERT(tp); 236 ASSERT(xfs_buf_islocked(bp)); 237 238 d = bp->b_addr; 239 240 /* 241 * ID of the first dquot in the block - id's are zero based. 242 */ 243 curid = id - (id % q->qi_dqperchunk); 244 memset(d, 0, BBTOB(q->qi_dqchunklen)); 245 for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++) { 246 d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC); 247 d->dd_diskdq.d_version = XFS_DQUOT_VERSION; 248 d->dd_diskdq.d_id = cpu_to_be32(curid); 249 d->dd_diskdq.d_flags = type; 250 if (xfs_sb_version_hascrc(&mp->m_sb)) { 251 uuid_copy(&d->dd_uuid, &mp->m_sb.sb_meta_uuid); 252 xfs_update_cksum((char *)d, sizeof(struct xfs_dqblk), 253 XFS_DQUOT_CRC_OFF); 254 } 255 } 256 257 xfs_trans_dquot_buf(tp, bp, 258 (type & XFS_DQ_USER ? XFS_BLF_UDQUOT_BUF : 259 ((type & XFS_DQ_PROJ) ? XFS_BLF_PDQUOT_BUF : 260 XFS_BLF_GDQUOT_BUF))); 261 xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1); 262 } 263 264 /* 265 * Initialize the dynamic speculative preallocation thresholds. The lo/hi 266 * watermarks correspond to the soft and hard limits by default. If a soft limit 267 * is not specified, we use 95% of the hard limit. 268 */ 269 void 270 xfs_dquot_set_prealloc_limits(struct xfs_dquot *dqp) 271 { 272 uint64_t space; 273 274 dqp->q_prealloc_hi_wmark = be64_to_cpu(dqp->q_core.d_blk_hardlimit); 275 dqp->q_prealloc_lo_wmark = be64_to_cpu(dqp->q_core.d_blk_softlimit); 276 if (!dqp->q_prealloc_lo_wmark) { 277 dqp->q_prealloc_lo_wmark = dqp->q_prealloc_hi_wmark; 278 do_div(dqp->q_prealloc_lo_wmark, 100); 279 dqp->q_prealloc_lo_wmark *= 95; 280 } 281 282 space = dqp->q_prealloc_hi_wmark; 283 284 do_div(space, 100); 285 dqp->q_low_space[XFS_QLOWSP_1_PCNT] = space; 286 dqp->q_low_space[XFS_QLOWSP_3_PCNT] = space * 3; 287 dqp->q_low_space[XFS_QLOWSP_5_PCNT] = space * 5; 288 } 289 290 /* 291 * Allocate a block and fill it with dquots. 292 * This is called when the bmapi finds a hole. 293 */ 294 STATIC int 295 xfs_qm_dqalloc( 296 xfs_trans_t **tpp, 297 xfs_mount_t *mp, 298 xfs_dquot_t *dqp, 299 xfs_inode_t *quotip, 300 xfs_fileoff_t offset_fsb, 301 xfs_buf_t **O_bpp) 302 { 303 xfs_fsblock_t firstblock; 304 struct xfs_defer_ops dfops; 305 xfs_bmbt_irec_t map; 306 int nmaps, error; 307 xfs_buf_t *bp; 308 xfs_trans_t *tp = *tpp; 309 310 ASSERT(tp != NULL); 311 312 trace_xfs_dqalloc(dqp); 313 314 /* 315 * Initialize the bmap freelist prior to calling bmapi code. 316 */ 317 xfs_defer_init(&dfops, &firstblock); 318 xfs_ilock(quotip, XFS_ILOCK_EXCL); 319 /* 320 * Return if this type of quotas is turned off while we didn't 321 * have an inode lock 322 */ 323 if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) { 324 xfs_iunlock(quotip, XFS_ILOCK_EXCL); 325 return -ESRCH; 326 } 327 328 xfs_trans_ijoin(tp, quotip, XFS_ILOCK_EXCL); 329 nmaps = 1; 330 error = xfs_bmapi_write(tp, quotip, offset_fsb, 331 XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA, 332 &firstblock, XFS_QM_DQALLOC_SPACE_RES(mp), 333 &map, &nmaps, &dfops); 334 if (error) 335 goto error0; 336 ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB); 337 ASSERT(nmaps == 1); 338 ASSERT((map.br_startblock != DELAYSTARTBLOCK) && 339 (map.br_startblock != HOLESTARTBLOCK)); 340 341 /* 342 * Keep track of the blkno to save a lookup later 343 */ 344 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock); 345 346 /* now we can just get the buffer (there's nothing to read yet) */ 347 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, 348 dqp->q_blkno, 349 mp->m_quotainfo->qi_dqchunklen, 350 0); 351 if (!bp) { 352 error = -ENOMEM; 353 goto error1; 354 } 355 bp->b_ops = &xfs_dquot_buf_ops; 356 357 /* 358 * Make a chunk of dquots out of this buffer and log 359 * the entire thing. 360 */ 361 xfs_qm_init_dquot_blk(tp, mp, be32_to_cpu(dqp->q_core.d_id), 362 dqp->dq_flags & XFS_DQ_ALLTYPES, bp); 363 364 /* 365 * xfs_defer_finish() may commit the current transaction and 366 * start a second transaction if the freelist is not empty. 367 * 368 * Since we still want to modify this buffer, we need to 369 * ensure that the buffer is not released on commit of 370 * the first transaction and ensure the buffer is added to the 371 * second transaction. 372 * 373 * If there is only one transaction then don't stop the buffer 374 * from being released when it commits later on. 375 */ 376 377 xfs_trans_bhold(tp, bp); 378 379 error = xfs_defer_finish(tpp, &dfops); 380 if (error) 381 goto error1; 382 383 /* Transaction was committed? */ 384 if (*tpp != tp) { 385 tp = *tpp; 386 xfs_trans_bjoin(tp, bp); 387 } else { 388 xfs_trans_bhold_release(tp, bp); 389 } 390 391 *O_bpp = bp; 392 return 0; 393 394 error1: 395 xfs_defer_cancel(&dfops); 396 error0: 397 return error; 398 } 399 400 /* 401 * Maps a dquot to the buffer containing its on-disk version. 402 * This returns a ptr to the buffer containing the on-disk dquot 403 * in the bpp param, and a ptr to the on-disk dquot within that buffer 404 */ 405 STATIC int 406 xfs_qm_dqtobp( 407 xfs_trans_t **tpp, 408 xfs_dquot_t *dqp, 409 xfs_disk_dquot_t **O_ddpp, 410 xfs_buf_t **O_bpp, 411 uint flags) 412 { 413 struct xfs_bmbt_irec map; 414 int nmaps = 1, error; 415 struct xfs_buf *bp; 416 struct xfs_inode *quotip; 417 struct xfs_mount *mp = dqp->q_mount; 418 xfs_dqid_t id = be32_to_cpu(dqp->q_core.d_id); 419 struct xfs_trans *tp = (tpp ? *tpp : NULL); 420 uint lock_mode; 421 422 quotip = xfs_quota_inode(dqp->q_mount, dqp->dq_flags); 423 dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk; 424 425 lock_mode = xfs_ilock_data_map_shared(quotip); 426 if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) { 427 /* 428 * Return if this type of quotas is turned off while we 429 * didn't have the quota inode lock. 430 */ 431 xfs_iunlock(quotip, lock_mode); 432 return -ESRCH; 433 } 434 435 /* 436 * Find the block map; no allocations yet 437 */ 438 error = xfs_bmapi_read(quotip, dqp->q_fileoffset, 439 XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0); 440 441 xfs_iunlock(quotip, lock_mode); 442 if (error) 443 return error; 444 445 ASSERT(nmaps == 1); 446 ASSERT(map.br_blockcount == 1); 447 448 /* 449 * Offset of dquot in the (fixed sized) dquot chunk. 450 */ 451 dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) * 452 sizeof(xfs_dqblk_t); 453 454 ASSERT(map.br_startblock != DELAYSTARTBLOCK); 455 if (map.br_startblock == HOLESTARTBLOCK) { 456 /* 457 * We don't allocate unless we're asked to 458 */ 459 if (!(flags & XFS_QMOPT_DQALLOC)) 460 return -ENOENT; 461 462 ASSERT(tp); 463 error = xfs_qm_dqalloc(tpp, mp, dqp, quotip, 464 dqp->q_fileoffset, &bp); 465 if (error) 466 return error; 467 tp = *tpp; 468 } else { 469 trace_xfs_dqtobp_read(dqp); 470 471 /* 472 * store the blkno etc so that we don't have to do the 473 * mapping all the time 474 */ 475 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock); 476 477 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, 478 dqp->q_blkno, 479 mp->m_quotainfo->qi_dqchunklen, 480 0, &bp, &xfs_dquot_buf_ops); 481 if (error) { 482 ASSERT(bp == NULL); 483 return error; 484 } 485 } 486 487 ASSERT(xfs_buf_islocked(bp)); 488 *O_bpp = bp; 489 *O_ddpp = bp->b_addr + dqp->q_bufoffset; 490 491 return 0; 492 } 493 494 495 /* 496 * Read in the ondisk dquot using dqtobp() then copy it to an incore version, 497 * and release the buffer immediately. 498 * 499 * If XFS_QMOPT_DQALLOC is set, allocate a dquot on disk if it needed. 500 */ 501 int 502 xfs_qm_dqread( 503 struct xfs_mount *mp, 504 xfs_dqid_t id, 505 uint type, 506 uint flags, 507 struct xfs_dquot **O_dqpp) 508 { 509 struct xfs_dquot *dqp; 510 struct xfs_disk_dquot *ddqp; 511 struct xfs_buf *bp; 512 struct xfs_trans *tp = NULL; 513 int error; 514 515 dqp = kmem_zone_zalloc(xfs_qm_dqzone, KM_SLEEP); 516 517 dqp->dq_flags = type; 518 dqp->q_core.d_id = cpu_to_be32(id); 519 dqp->q_mount = mp; 520 INIT_LIST_HEAD(&dqp->q_lru); 521 mutex_init(&dqp->q_qlock); 522 init_waitqueue_head(&dqp->q_pinwait); 523 524 /* 525 * Because we want to use a counting completion, complete 526 * the flush completion once to allow a single access to 527 * the flush completion without blocking. 528 */ 529 init_completion(&dqp->q_flush); 530 complete(&dqp->q_flush); 531 532 /* 533 * Make sure group quotas have a different lock class than user 534 * quotas. 535 */ 536 switch (type) { 537 case XFS_DQ_USER: 538 /* uses the default lock class */ 539 break; 540 case XFS_DQ_GROUP: 541 lockdep_set_class(&dqp->q_qlock, &xfs_dquot_group_class); 542 break; 543 case XFS_DQ_PROJ: 544 lockdep_set_class(&dqp->q_qlock, &xfs_dquot_project_class); 545 break; 546 default: 547 ASSERT(0); 548 break; 549 } 550 551 XFS_STATS_INC(mp, xs_qm_dquot); 552 553 trace_xfs_dqread(dqp); 554 555 if (flags & XFS_QMOPT_DQALLOC) { 556 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_qm_dqalloc, 557 XFS_QM_DQALLOC_SPACE_RES(mp), 0, 0, &tp); 558 if (error) 559 goto error0; 560 } 561 562 /* 563 * get a pointer to the on-disk dquot and the buffer containing it 564 * dqp already knows its own type (GROUP/USER). 565 */ 566 error = xfs_qm_dqtobp(&tp, dqp, &ddqp, &bp, flags); 567 if (error) { 568 /* 569 * This can happen if quotas got turned off (ESRCH), 570 * or if the dquot didn't exist on disk and we ask to 571 * allocate (ENOENT). 572 */ 573 trace_xfs_dqread_fail(dqp); 574 goto error1; 575 } 576 577 /* copy everything from disk dquot to the incore dquot */ 578 memcpy(&dqp->q_core, ddqp, sizeof(xfs_disk_dquot_t)); 579 xfs_qm_dquot_logitem_init(dqp); 580 581 /* 582 * Reservation counters are defined as reservation plus current usage 583 * to avoid having to add every time. 584 */ 585 dqp->q_res_bcount = be64_to_cpu(ddqp->d_bcount); 586 dqp->q_res_icount = be64_to_cpu(ddqp->d_icount); 587 dqp->q_res_rtbcount = be64_to_cpu(ddqp->d_rtbcount); 588 589 /* initialize the dquot speculative prealloc thresholds */ 590 xfs_dquot_set_prealloc_limits(dqp); 591 592 /* Mark the buf so that this will stay incore a little longer */ 593 xfs_buf_set_ref(bp, XFS_DQUOT_REF); 594 595 /* 596 * We got the buffer with a xfs_trans_read_buf() (in dqtobp()) 597 * So we need to release with xfs_trans_brelse(). 598 * The strategy here is identical to that of inodes; we lock 599 * the dquot in xfs_qm_dqget() before making it accessible to 600 * others. This is because dquots, like inodes, need a good level of 601 * concurrency, and we don't want to take locks on the entire buffers 602 * for dquot accesses. 603 * Note also that the dquot buffer may even be dirty at this point, if 604 * this particular dquot was repaired. We still aren't afraid to 605 * brelse it because we have the changes incore. 606 */ 607 ASSERT(xfs_buf_islocked(bp)); 608 xfs_trans_brelse(tp, bp); 609 610 if (tp) { 611 error = xfs_trans_commit(tp); 612 if (error) 613 goto error0; 614 } 615 616 *O_dqpp = dqp; 617 return error; 618 619 error1: 620 if (tp) 621 xfs_trans_cancel(tp); 622 error0: 623 xfs_qm_dqdestroy(dqp); 624 *O_dqpp = NULL; 625 return error; 626 } 627 628 /* 629 * Advance to the next id in the current chunk, or if at the 630 * end of the chunk, skip ahead to first id in next allocated chunk 631 * using the SEEK_DATA interface. 632 */ 633 static int 634 xfs_dq_get_next_id( 635 struct xfs_mount *mp, 636 uint type, 637 xfs_dqid_t *id) 638 { 639 struct xfs_inode *quotip = xfs_quota_inode(mp, type); 640 xfs_dqid_t next_id = *id + 1; /* simple advance */ 641 uint lock_flags; 642 struct xfs_bmbt_irec got; 643 struct xfs_iext_cursor cur; 644 xfs_fsblock_t start; 645 int error = 0; 646 647 /* If we'd wrap past the max ID, stop */ 648 if (next_id < *id) 649 return -ENOENT; 650 651 /* If new ID is within the current chunk, advancing it sufficed */ 652 if (next_id % mp->m_quotainfo->qi_dqperchunk) { 653 *id = next_id; 654 return 0; 655 } 656 657 /* Nope, next_id is now past the current chunk, so find the next one */ 658 start = (xfs_fsblock_t)next_id / mp->m_quotainfo->qi_dqperchunk; 659 660 lock_flags = xfs_ilock_data_map_shared(quotip); 661 if (!(quotip->i_df.if_flags & XFS_IFEXTENTS)) { 662 error = xfs_iread_extents(NULL, quotip, XFS_DATA_FORK); 663 if (error) 664 return error; 665 } 666 667 if (xfs_iext_lookup_extent(quotip, "ip->i_df, start, &cur, &got)) { 668 /* contiguous chunk, bump startoff for the id calculation */ 669 if (got.br_startoff < start) 670 got.br_startoff = start; 671 *id = got.br_startoff * mp->m_quotainfo->qi_dqperchunk; 672 } else { 673 error = -ENOENT; 674 } 675 676 xfs_iunlock(quotip, lock_flags); 677 678 return error; 679 } 680 681 /* 682 * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a 683 * a locked dquot, doing an allocation (if requested) as needed. 684 * When both an inode and an id are given, the inode's id takes precedence. 685 * That is, if the id changes while we don't hold the ilock inside this 686 * function, the new dquot is returned, not necessarily the one requested 687 * in the id argument. 688 */ 689 int 690 xfs_qm_dqget( 691 xfs_mount_t *mp, 692 xfs_inode_t *ip, /* locked inode (optional) */ 693 xfs_dqid_t id, /* uid/projid/gid depending on type */ 694 uint type, /* XFS_DQ_USER/XFS_DQ_PROJ/XFS_DQ_GROUP */ 695 uint flags, /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */ 696 xfs_dquot_t **O_dqpp) /* OUT : locked incore dquot */ 697 { 698 struct xfs_quotainfo *qi = mp->m_quotainfo; 699 struct radix_tree_root *tree = xfs_dquot_tree(qi, type); 700 struct xfs_dquot *dqp; 701 int error; 702 703 ASSERT(XFS_IS_QUOTA_RUNNING(mp)); 704 if ((! XFS_IS_UQUOTA_ON(mp) && type == XFS_DQ_USER) || 705 (! XFS_IS_PQUOTA_ON(mp) && type == XFS_DQ_PROJ) || 706 (! XFS_IS_GQUOTA_ON(mp) && type == XFS_DQ_GROUP)) { 707 return -ESRCH; 708 } 709 710 ASSERT(type == XFS_DQ_USER || 711 type == XFS_DQ_PROJ || 712 type == XFS_DQ_GROUP); 713 if (ip) { 714 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); 715 ASSERT(xfs_inode_dquot(ip, type) == NULL); 716 } 717 718 restart: 719 mutex_lock(&qi->qi_tree_lock); 720 dqp = radix_tree_lookup(tree, id); 721 if (dqp) { 722 xfs_dqlock(dqp); 723 if (dqp->dq_flags & XFS_DQ_FREEING) { 724 xfs_dqunlock(dqp); 725 mutex_unlock(&qi->qi_tree_lock); 726 trace_xfs_dqget_freeing(dqp); 727 delay(1); 728 goto restart; 729 } 730 731 /* uninit / unused quota found in radix tree, keep looking */ 732 if (flags & XFS_QMOPT_DQNEXT) { 733 if (XFS_IS_DQUOT_UNINITIALIZED(dqp)) { 734 xfs_dqunlock(dqp); 735 mutex_unlock(&qi->qi_tree_lock); 736 error = xfs_dq_get_next_id(mp, type, &id); 737 if (error) 738 return error; 739 goto restart; 740 } 741 } 742 743 dqp->q_nrefs++; 744 mutex_unlock(&qi->qi_tree_lock); 745 746 trace_xfs_dqget_hit(dqp); 747 XFS_STATS_INC(mp, xs_qm_dqcachehits); 748 *O_dqpp = dqp; 749 return 0; 750 } 751 mutex_unlock(&qi->qi_tree_lock); 752 XFS_STATS_INC(mp, xs_qm_dqcachemisses); 753 754 /* 755 * Dquot cache miss. We don't want to keep the inode lock across 756 * a (potential) disk read. Also we don't want to deal with the lock 757 * ordering between quotainode and this inode. OTOH, dropping the inode 758 * lock here means dealing with a chown that can happen before 759 * we re-acquire the lock. 760 */ 761 if (ip) 762 xfs_iunlock(ip, XFS_ILOCK_EXCL); 763 764 error = xfs_qm_dqread(mp, id, type, flags, &dqp); 765 766 if (ip) 767 xfs_ilock(ip, XFS_ILOCK_EXCL); 768 769 /* If we are asked to find next active id, keep looking */ 770 if (error == -ENOENT && (flags & XFS_QMOPT_DQNEXT)) { 771 error = xfs_dq_get_next_id(mp, type, &id); 772 if (!error) 773 goto restart; 774 } 775 776 if (error) 777 return error; 778 779 if (ip) { 780 /* 781 * A dquot could be attached to this inode by now, since 782 * we had dropped the ilock. 783 */ 784 if (xfs_this_quota_on(mp, type)) { 785 struct xfs_dquot *dqp1; 786 787 dqp1 = xfs_inode_dquot(ip, type); 788 if (dqp1) { 789 xfs_qm_dqdestroy(dqp); 790 dqp = dqp1; 791 xfs_dqlock(dqp); 792 goto dqret; 793 } 794 } else { 795 /* inode stays locked on return */ 796 xfs_qm_dqdestroy(dqp); 797 return -ESRCH; 798 } 799 } 800 801 mutex_lock(&qi->qi_tree_lock); 802 error = radix_tree_insert(tree, id, dqp); 803 if (unlikely(error)) { 804 WARN_ON(error != -EEXIST); 805 806 /* 807 * Duplicate found. Just throw away the new dquot and start 808 * over. 809 */ 810 mutex_unlock(&qi->qi_tree_lock); 811 trace_xfs_dqget_dup(dqp); 812 xfs_qm_dqdestroy(dqp); 813 XFS_STATS_INC(mp, xs_qm_dquot_dups); 814 goto restart; 815 } 816 817 /* 818 * We return a locked dquot to the caller, with a reference taken 819 */ 820 xfs_dqlock(dqp); 821 dqp->q_nrefs = 1; 822 823 qi->qi_dquots++; 824 mutex_unlock(&qi->qi_tree_lock); 825 826 /* If we are asked to find next active id, keep looking */ 827 if (flags & XFS_QMOPT_DQNEXT) { 828 if (XFS_IS_DQUOT_UNINITIALIZED(dqp)) { 829 xfs_qm_dqput(dqp); 830 error = xfs_dq_get_next_id(mp, type, &id); 831 if (error) 832 return error; 833 goto restart; 834 } 835 } 836 837 dqret: 838 ASSERT((ip == NULL) || xfs_isilocked(ip, XFS_ILOCK_EXCL)); 839 trace_xfs_dqget_miss(dqp); 840 *O_dqpp = dqp; 841 return 0; 842 } 843 844 /* 845 * Release a reference to the dquot (decrement ref-count) and unlock it. 846 * 847 * If there is a group quota attached to this dquot, carefully release that 848 * too without tripping over deadlocks'n'stuff. 849 */ 850 void 851 xfs_qm_dqput( 852 struct xfs_dquot *dqp) 853 { 854 ASSERT(dqp->q_nrefs > 0); 855 ASSERT(XFS_DQ_IS_LOCKED(dqp)); 856 857 trace_xfs_dqput(dqp); 858 859 if (--dqp->q_nrefs == 0) { 860 struct xfs_quotainfo *qi = dqp->q_mount->m_quotainfo; 861 trace_xfs_dqput_free(dqp); 862 863 if (list_lru_add(&qi->qi_lru, &dqp->q_lru)) 864 XFS_STATS_INC(dqp->q_mount, xs_qm_dquot_unused); 865 } 866 xfs_dqunlock(dqp); 867 } 868 869 /* 870 * Release a dquot. Flush it if dirty, then dqput() it. 871 * dquot must not be locked. 872 */ 873 void 874 xfs_qm_dqrele( 875 xfs_dquot_t *dqp) 876 { 877 if (!dqp) 878 return; 879 880 trace_xfs_dqrele(dqp); 881 882 xfs_dqlock(dqp); 883 /* 884 * We don't care to flush it if the dquot is dirty here. 885 * That will create stutters that we want to avoid. 886 * Instead we do a delayed write when we try to reclaim 887 * a dirty dquot. Also xfs_sync will take part of the burden... 888 */ 889 xfs_qm_dqput(dqp); 890 } 891 892 /* 893 * This is the dquot flushing I/O completion routine. It is called 894 * from interrupt level when the buffer containing the dquot is 895 * flushed to disk. It is responsible for removing the dquot logitem 896 * from the AIL if it has not been re-logged, and unlocking the dquot's 897 * flush lock. This behavior is very similar to that of inodes.. 898 */ 899 STATIC void 900 xfs_qm_dqflush_done( 901 struct xfs_buf *bp, 902 struct xfs_log_item *lip) 903 { 904 xfs_dq_logitem_t *qip = (struct xfs_dq_logitem *)lip; 905 xfs_dquot_t *dqp = qip->qli_dquot; 906 struct xfs_ail *ailp = lip->li_ailp; 907 908 /* 909 * We only want to pull the item from the AIL if its 910 * location in the log has not changed since we started the flush. 911 * Thus, we only bother if the dquot's lsn has 912 * not changed. First we check the lsn outside the lock 913 * since it's cheaper, and then we recheck while 914 * holding the lock before removing the dquot from the AIL. 915 */ 916 if ((lip->li_flags & XFS_LI_IN_AIL) && 917 ((lip->li_lsn == qip->qli_flush_lsn) || 918 (lip->li_flags & XFS_LI_FAILED))) { 919 920 /* xfs_trans_ail_delete() drops the AIL lock. */ 921 spin_lock(&ailp->ail_lock); 922 if (lip->li_lsn == qip->qli_flush_lsn) { 923 xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE); 924 } else { 925 /* 926 * Clear the failed state since we are about to drop the 927 * flush lock 928 */ 929 if (lip->li_flags & XFS_LI_FAILED) 930 xfs_clear_li_failed(lip); 931 spin_unlock(&ailp->ail_lock); 932 } 933 } 934 935 /* 936 * Release the dq's flush lock since we're done with it. 937 */ 938 xfs_dqfunlock(dqp); 939 } 940 941 /* 942 * Write a modified dquot to disk. 943 * The dquot must be locked and the flush lock too taken by caller. 944 * The flush lock will not be unlocked until the dquot reaches the disk, 945 * but the dquot is free to be unlocked and modified by the caller 946 * in the interim. Dquot is still locked on return. This behavior is 947 * identical to that of inodes. 948 */ 949 int 950 xfs_qm_dqflush( 951 struct xfs_dquot *dqp, 952 struct xfs_buf **bpp) 953 { 954 struct xfs_mount *mp = dqp->q_mount; 955 struct xfs_buf *bp; 956 struct xfs_disk_dquot *ddqp; 957 xfs_failaddr_t fa; 958 int error; 959 960 ASSERT(XFS_DQ_IS_LOCKED(dqp)); 961 ASSERT(!completion_done(&dqp->q_flush)); 962 963 trace_xfs_dqflush(dqp); 964 965 *bpp = NULL; 966 967 xfs_qm_dqunpin_wait(dqp); 968 969 /* 970 * This may have been unpinned because the filesystem is shutting 971 * down forcibly. If that's the case we must not write this dquot 972 * to disk, because the log record didn't make it to disk. 973 * 974 * We also have to remove the log item from the AIL in this case, 975 * as we wait for an emptry AIL as part of the unmount process. 976 */ 977 if (XFS_FORCED_SHUTDOWN(mp)) { 978 struct xfs_log_item *lip = &dqp->q_logitem.qli_item; 979 dqp->dq_flags &= ~XFS_DQ_DIRTY; 980 981 xfs_trans_ail_remove(lip, SHUTDOWN_CORRUPT_INCORE); 982 983 error = -EIO; 984 goto out_unlock; 985 } 986 987 /* 988 * Get the buffer containing the on-disk dquot 989 */ 990 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno, 991 mp->m_quotainfo->qi_dqchunklen, 0, &bp, 992 &xfs_dquot_buf_ops); 993 if (error) 994 goto out_unlock; 995 996 /* 997 * Calculate the location of the dquot inside the buffer. 998 */ 999 ddqp = bp->b_addr + dqp->q_bufoffset; 1000 1001 /* 1002 * A simple sanity check in case we got a corrupted dquot.. 1003 */ 1004 fa = xfs_dquot_verify(mp, &dqp->q_core, be32_to_cpu(ddqp->d_id), 0, 0); 1005 if (fa) { 1006 xfs_alert(mp, "corrupt dquot ID 0x%x in memory at %pS", 1007 be32_to_cpu(ddqp->d_id), fa); 1008 xfs_buf_relse(bp); 1009 xfs_dqfunlock(dqp); 1010 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); 1011 return -EIO; 1012 } 1013 1014 /* This is the only portion of data that needs to persist */ 1015 memcpy(ddqp, &dqp->q_core, sizeof(xfs_disk_dquot_t)); 1016 1017 /* 1018 * Clear the dirty field and remember the flush lsn for later use. 1019 */ 1020 dqp->dq_flags &= ~XFS_DQ_DIRTY; 1021 1022 xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn, 1023 &dqp->q_logitem.qli_item.li_lsn); 1024 1025 /* 1026 * copy the lsn into the on-disk dquot now while we have the in memory 1027 * dquot here. This can't be done later in the write verifier as we 1028 * can't get access to the log item at that point in time. 1029 * 1030 * We also calculate the CRC here so that the on-disk dquot in the 1031 * buffer always has a valid CRC. This ensures there is no possibility 1032 * of a dquot without an up-to-date CRC getting to disk. 1033 */ 1034 if (xfs_sb_version_hascrc(&mp->m_sb)) { 1035 struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddqp; 1036 1037 dqb->dd_lsn = cpu_to_be64(dqp->q_logitem.qli_item.li_lsn); 1038 xfs_update_cksum((char *)dqb, sizeof(struct xfs_dqblk), 1039 XFS_DQUOT_CRC_OFF); 1040 } 1041 1042 /* 1043 * Attach an iodone routine so that we can remove this dquot from the 1044 * AIL and release the flush lock once the dquot is synced to disk. 1045 */ 1046 xfs_buf_attach_iodone(bp, xfs_qm_dqflush_done, 1047 &dqp->q_logitem.qli_item); 1048 1049 /* 1050 * If the buffer is pinned then push on the log so we won't 1051 * get stuck waiting in the write for too long. 1052 */ 1053 if (xfs_buf_ispinned(bp)) { 1054 trace_xfs_dqflush_force(dqp); 1055 xfs_log_force(mp, 0); 1056 } 1057 1058 trace_xfs_dqflush_done(dqp); 1059 *bpp = bp; 1060 return 0; 1061 1062 out_unlock: 1063 xfs_dqfunlock(dqp); 1064 return -EIO; 1065 } 1066 1067 /* 1068 * Lock two xfs_dquot structures. 1069 * 1070 * To avoid deadlocks we always lock the quota structure with 1071 * the lowerd id first. 1072 */ 1073 void 1074 xfs_dqlock2( 1075 xfs_dquot_t *d1, 1076 xfs_dquot_t *d2) 1077 { 1078 if (d1 && d2) { 1079 ASSERT(d1 != d2); 1080 if (be32_to_cpu(d1->q_core.d_id) > 1081 be32_to_cpu(d2->q_core.d_id)) { 1082 mutex_lock(&d2->q_qlock); 1083 mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED); 1084 } else { 1085 mutex_lock(&d1->q_qlock); 1086 mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED); 1087 } 1088 } else if (d1) { 1089 mutex_lock(&d1->q_qlock); 1090 } else if (d2) { 1091 mutex_lock(&d2->q_qlock); 1092 } 1093 } 1094 1095 int __init 1096 xfs_qm_init(void) 1097 { 1098 xfs_qm_dqzone = 1099 kmem_zone_init(sizeof(struct xfs_dquot), "xfs_dquot"); 1100 if (!xfs_qm_dqzone) 1101 goto out; 1102 1103 xfs_qm_dqtrxzone = 1104 kmem_zone_init(sizeof(struct xfs_dquot_acct), "xfs_dqtrx"); 1105 if (!xfs_qm_dqtrxzone) 1106 goto out_free_dqzone; 1107 1108 return 0; 1109 1110 out_free_dqzone: 1111 kmem_zone_destroy(xfs_qm_dqzone); 1112 out: 1113 return -ENOMEM; 1114 } 1115 1116 void 1117 xfs_qm_exit(void) 1118 { 1119 kmem_zone_destroy(xfs_qm_dqtrxzone); 1120 kmem_zone_destroy(xfs_qm_dqzone); 1121 } 1122