1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2003 Silicon Graphics, Inc.
8 #include "xfs_format.h"
9 #include "xfs_log_format.h"
10 #include "xfs_shared.h"
11 #include "xfs_trans_resv.h"
13 #include "xfs_mount.h"
14 #include "xfs_defer.h"
15 #include "xfs_inode.h"
17 #include "xfs_quota.h"
18 #include "xfs_trans.h"
19 #include "xfs_buf_item.h"
20 #include "xfs_trans_space.h"
21 #include "xfs_trans_priv.h"
23 #include "xfs_trace.h"
25 #include "xfs_bmap_btree.h"
26 #include "xfs_error.h"
33 * dquot->q_qlock (xfs_dqlock() and friends)
34 * dquot->q_flush (xfs_dqflock() and friends)
37 * If two dquots need to be locked the order is user before group/project,
38 * otherwise by the lowest id first, see xfs_dqlock2.
41 struct kmem_zone *xfs_qm_dqtrxzone;
42 static struct kmem_zone *xfs_qm_dqzone;
44 static struct lock_class_key xfs_dquot_group_class;
45 static struct lock_class_key xfs_dquot_project_class;
48 * This is called to free all the memory associated with a dquot
52 struct xfs_dquot *dqp)
54 ASSERT(list_empty(&dqp->q_lru));
56 kmem_free(dqp->q_logitem.qli_item.li_lv_shadow);
57 mutex_destroy(&dqp->q_qlock);
59 XFS_STATS_DEC(dqp->q_mount, xs_qm_dquot);
60 kmem_cache_free(xfs_qm_dqzone, dqp);
64 * If default limits are in force, push them into the dquot now.
65 * We overwrite the dquot limits only if they are zero and this
66 * is not the root dquot.
69 xfs_qm_adjust_dqlimits(
72 struct xfs_mount *mp = dq->q_mount;
73 struct xfs_quotainfo *q = mp->m_quotainfo;
74 struct xfs_def_quota *defq;
78 defq = xfs_get_defquota(q, xfs_dquot_type(dq));
80 if (!dq->q_blk.softlimit) {
81 dq->q_blk.softlimit = defq->blk.soft;
84 if (!dq->q_blk.hardlimit) {
85 dq->q_blk.hardlimit = defq->blk.hard;
88 if (!dq->q_ino.softlimit)
89 dq->q_ino.softlimit = defq->ino.soft;
90 if (!dq->q_ino.hardlimit)
91 dq->q_ino.hardlimit = defq->ino.hard;
92 if (!dq->q_rtb.softlimit)
93 dq->q_rtb.softlimit = defq->rtb.soft;
94 if (!dq->q_rtb.hardlimit)
95 dq->q_rtb.hardlimit = defq->rtb.hard;
98 xfs_dquot_set_prealloc_limits(dq);
101 /* Set the expiration time of a quota's grace period. */
103 xfs_dquot_set_timeout(
104 struct xfs_mount *mp,
107 struct xfs_quotainfo *qi = mp->m_quotainfo;
109 return clamp_t(time64_t, timeout, qi->qi_expiry_min,
113 /* Set the length of the default grace period. */
115 xfs_dquot_set_grace_period(
118 return clamp_t(time64_t, grace, XFS_DQ_GRACE_MIN, XFS_DQ_GRACE_MAX);
122 * Determine if this quota counter is over either limit and set the quota
123 * timers as appropriate.
126 xfs_qm_adjust_res_timer(
127 struct xfs_mount *mp,
128 struct xfs_dquot_res *res,
129 struct xfs_quota_limits *qlim)
131 ASSERT(res->hardlimit == 0 || res->softlimit <= res->hardlimit);
133 if ((res->softlimit && res->count > res->softlimit) ||
134 (res->hardlimit && res->count > res->hardlimit)) {
136 res->timer = xfs_dquot_set_timeout(mp,
137 ktime_get_real_seconds() + qlim->time);
147 * Check the limits and timers of a dquot and start or reset timers
149 * This gets called even when quota enforcement is OFF, which makes our
150 * life a little less complicated. (We just don't reject any quota
151 * reservations in that case, when enforcement is off).
152 * We also return 0 as the values of the timers in Q_GETQUOTA calls, when
154 * In contrast, warnings are a little different in that they don't
155 * 'automatically' get started when limits get exceeded. They do
156 * get reset to zero, however, when we find the count to be under
157 * the soft limit (they are only ever set non-zero via userspace).
160 xfs_qm_adjust_dqtimers(
161 struct xfs_dquot *dq)
163 struct xfs_mount *mp = dq->q_mount;
164 struct xfs_quotainfo *qi = mp->m_quotainfo;
165 struct xfs_def_quota *defq;
168 defq = xfs_get_defquota(qi, xfs_dquot_type(dq));
170 xfs_qm_adjust_res_timer(dq->q_mount, &dq->q_blk, &defq->blk);
171 xfs_qm_adjust_res_timer(dq->q_mount, &dq->q_ino, &defq->ino);
172 xfs_qm_adjust_res_timer(dq->q_mount, &dq->q_rtb, &defq->rtb);
176 * initialize a buffer full of dquots and log the whole thing
179 xfs_qm_init_dquot_blk(
180 struct xfs_trans *tp,
181 struct xfs_mount *mp,
186 struct xfs_quotainfo *q = mp->m_quotainfo;
190 unsigned int blftype;
194 ASSERT(xfs_buf_islocked(bp));
197 case XFS_DQTYPE_USER:
198 qflag = XFS_UQUOTA_CHKD;
199 blftype = XFS_BLF_UDQUOT_BUF;
201 case XFS_DQTYPE_PROJ:
202 qflag = XFS_PQUOTA_CHKD;
203 blftype = XFS_BLF_PDQUOT_BUF;
205 case XFS_DQTYPE_GROUP:
206 qflag = XFS_GQUOTA_CHKD;
207 blftype = XFS_BLF_GDQUOT_BUF;
217 * ID of the first dquot in the block - id's are zero based.
219 curid = id - (id % q->qi_dqperchunk);
220 memset(d, 0, BBTOB(q->qi_dqchunklen));
221 for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++) {
222 d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
223 d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
224 d->dd_diskdq.d_id = cpu_to_be32(curid);
225 d->dd_diskdq.d_type = type;
226 if (curid > 0 && xfs_sb_version_hasbigtime(&mp->m_sb))
227 d->dd_diskdq.d_type |= XFS_DQTYPE_BIGTIME;
228 if (xfs_sb_version_hascrc(&mp->m_sb)) {
229 uuid_copy(&d->dd_uuid, &mp->m_sb.sb_meta_uuid);
230 xfs_update_cksum((char *)d, sizeof(struct xfs_dqblk),
235 xfs_trans_dquot_buf(tp, bp, blftype);
238 * quotacheck uses delayed writes to update all the dquots on disk in an
239 * efficient manner instead of logging the individual dquot changes as
240 * they are made. However if we log the buffer allocated here and crash
241 * after quotacheck while the logged initialisation is still in the
242 * active region of the log, log recovery can replay the dquot buffer
243 * initialisation over the top of the checked dquots and corrupt quota
246 * To avoid this problem, quotacheck cannot log the initialised buffer.
247 * We must still dirty the buffer and write it back before the
248 * allocation transaction clears the log. Therefore, mark the buffer as
249 * ordered instead of logging it directly. This is safe for quotacheck
250 * because it detects and repairs allocated but initialized dquot blocks
251 * in the quota inodes.
253 if (!(mp->m_qflags & qflag))
254 xfs_trans_ordered_buf(tp, bp);
256 xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1);
260 * Initialize the dynamic speculative preallocation thresholds. The lo/hi
261 * watermarks correspond to the soft and hard limits by default. If a soft limit
262 * is not specified, we use 95% of the hard limit.
265 xfs_dquot_set_prealloc_limits(struct xfs_dquot *dqp)
269 dqp->q_prealloc_hi_wmark = dqp->q_blk.hardlimit;
270 dqp->q_prealloc_lo_wmark = dqp->q_blk.softlimit;
271 if (!dqp->q_prealloc_lo_wmark) {
272 dqp->q_prealloc_lo_wmark = dqp->q_prealloc_hi_wmark;
273 do_div(dqp->q_prealloc_lo_wmark, 100);
274 dqp->q_prealloc_lo_wmark *= 95;
277 space = dqp->q_prealloc_hi_wmark;
280 dqp->q_low_space[XFS_QLOWSP_1_PCNT] = space;
281 dqp->q_low_space[XFS_QLOWSP_3_PCNT] = space * 3;
282 dqp->q_low_space[XFS_QLOWSP_5_PCNT] = space * 5;
286 * Ensure that the given in-core dquot has a buffer on disk backing it, and
287 * return the buffer locked and held. This is called when the bmapi finds a
291 xfs_dquot_disk_alloc(
292 struct xfs_trans **tpp,
293 struct xfs_dquot *dqp,
294 struct xfs_buf **bpp)
296 struct xfs_bmbt_irec map;
297 struct xfs_trans *tp = *tpp;
298 struct xfs_mount *mp = tp->t_mountp;
300 xfs_dqtype_t qtype = xfs_dquot_type(dqp);
301 struct xfs_inode *quotip = xfs_quota_inode(mp, qtype);
305 trace_xfs_dqalloc(dqp);
307 xfs_ilock(quotip, XFS_ILOCK_EXCL);
308 if (!xfs_this_quota_on(dqp->q_mount, qtype)) {
310 * Return if this type of quotas is turned off while we didn't
313 xfs_iunlock(quotip, XFS_ILOCK_EXCL);
317 /* Create the block mapping. */
318 xfs_trans_ijoin(tp, quotip, XFS_ILOCK_EXCL);
319 error = xfs_bmapi_write(tp, quotip, dqp->q_fileoffset,
320 XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA, 0, &map,
324 ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB);
326 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
327 (map.br_startblock != HOLESTARTBLOCK));
330 * Keep track of the blkno to save a lookup later
332 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
334 /* now we can just get the buffer (there's nothing to read yet) */
335 error = xfs_trans_get_buf(tp, mp->m_ddev_targp, dqp->q_blkno,
336 mp->m_quotainfo->qi_dqchunklen, 0, &bp);
339 bp->b_ops = &xfs_dquot_buf_ops;
342 * Make a chunk of dquots out of this buffer and log
345 xfs_qm_init_dquot_blk(tp, mp, dqp->q_id, qtype, bp);
346 xfs_buf_set_ref(bp, XFS_DQUOT_REF);
349 * Hold the buffer and join it to the dfops so that we'll still own
350 * the buffer when we return to the caller. The buffer disposal on
351 * error must be paid attention to very carefully, as it has been
352 * broken since commit efa092f3d4c6 "[XFS] Fixes a bug in the quota
353 * code when allocating a new dquot record" in 2005, and the later
354 * conversion to xfs_defer_ops in commit 310a75a3c6c747 failed to keep
355 * the buffer locked across the _defer_finish call. We can now do
356 * this correctly with xfs_defer_bjoin.
358 * Above, we allocated a disk block for the dquot information and used
359 * get_buf to initialize the dquot. If the _defer_finish fails, the old
360 * transaction is gone but the new buffer is not joined or held to any
361 * transaction, so we must _buf_relse it.
363 * If everything succeeds, the caller of this function is returned a
364 * buffer that is locked and held to the transaction. The caller
365 * is responsible for unlocking any buffer passed back, either
366 * manually or by committing the transaction. On error, the buffer is
367 * released and not passed back.
369 xfs_trans_bhold(tp, bp);
370 error = xfs_defer_finish(tpp);
372 xfs_trans_bhold_release(*tpp, bp);
373 xfs_trans_brelse(*tpp, bp);
381 * Read in the in-core dquot's on-disk metadata and return the buffer.
382 * Returns ENOENT to signal a hole.
386 struct xfs_mount *mp,
387 struct xfs_dquot *dqp,
388 struct xfs_buf **bpp)
390 struct xfs_bmbt_irec map;
392 xfs_dqtype_t qtype = xfs_dquot_type(dqp);
393 struct xfs_inode *quotip = xfs_quota_inode(mp, qtype);
398 lock_mode = xfs_ilock_data_map_shared(quotip);
399 if (!xfs_this_quota_on(mp, qtype)) {
401 * Return if this type of quotas is turned off while we
402 * didn't have the quota inode lock.
404 xfs_iunlock(quotip, lock_mode);
409 * Find the block map; no allocations yet
411 error = xfs_bmapi_read(quotip, dqp->q_fileoffset,
412 XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0);
413 xfs_iunlock(quotip, lock_mode);
418 ASSERT(map.br_blockcount >= 1);
419 ASSERT(map.br_startblock != DELAYSTARTBLOCK);
420 if (map.br_startblock == HOLESTARTBLOCK)
423 trace_xfs_dqtobp_read(dqp);
426 * store the blkno etc so that we don't have to do the
427 * mapping all the time
429 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
431 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
432 mp->m_quotainfo->qi_dqchunklen, 0, &bp,
439 ASSERT(xfs_buf_islocked(bp));
440 xfs_buf_set_ref(bp, XFS_DQUOT_REF);
446 /* Allocate and initialize everything we need for an incore dquot. */
447 STATIC struct xfs_dquot *
449 struct xfs_mount *mp,
453 struct xfs_dquot *dqp;
455 dqp = kmem_cache_zalloc(xfs_qm_dqzone, GFP_KERNEL | __GFP_NOFAIL);
460 INIT_LIST_HEAD(&dqp->q_lru);
461 mutex_init(&dqp->q_qlock);
462 init_waitqueue_head(&dqp->q_pinwait);
463 dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk;
465 * Offset of dquot in the (fixed sized) dquot chunk.
467 dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) *
471 * Because we want to use a counting completion, complete
472 * the flush completion once to allow a single access to
473 * the flush completion without blocking.
475 init_completion(&dqp->q_flush);
476 complete(&dqp->q_flush);
479 * Make sure group quotas have a different lock class than user
483 case XFS_DQTYPE_USER:
484 /* uses the default lock class */
486 case XFS_DQTYPE_GROUP:
487 lockdep_set_class(&dqp->q_qlock, &xfs_dquot_group_class);
489 case XFS_DQTYPE_PROJ:
490 lockdep_set_class(&dqp->q_qlock, &xfs_dquot_project_class);
497 xfs_qm_dquot_logitem_init(dqp);
499 XFS_STATS_INC(mp, xs_qm_dquot);
503 /* Check the ondisk dquot's id and type match what the incore dquot expects. */
505 xfs_dquot_check_type(
506 struct xfs_dquot *dqp,
507 struct xfs_disk_dquot *ddqp)
512 ddqp_type = ddqp->d_type & XFS_DQTYPE_REC_MASK;
513 dqp_type = xfs_dquot_type(dqp);
515 if (be32_to_cpu(ddqp->d_id) != dqp->q_id)
519 * V5 filesystems always expect an exact type match. V4 filesystems
520 * expect an exact match for user dquots and for non-root group and
523 if (xfs_sb_version_hascrc(&dqp->q_mount->m_sb) ||
524 dqp_type == XFS_DQTYPE_USER || dqp->q_id != 0)
525 return ddqp_type == dqp_type;
528 * V4 filesystems support either group or project quotas, but not both
529 * at the same time. The non-user quota file can be switched between
530 * group and project quota uses depending on the mount options, which
531 * means that we can encounter the other type when we try to load quota
532 * defaults. Quotacheck will soon reset the the entire quota file
533 * (including the root dquot) anyway, but don't log scary corruption
536 return ddqp_type == XFS_DQTYPE_GROUP || ddqp_type == XFS_DQTYPE_PROJ;
539 /* Copy the in-core quota fields in from the on-disk buffer. */
542 struct xfs_dquot *dqp,
545 struct xfs_disk_dquot *ddqp = bp->b_addr + dqp->q_bufoffset;
548 * Ensure that we got the type and ID we were looking for.
549 * Everything else was checked by the dquot buffer verifier.
551 if (!xfs_dquot_check_type(dqp, ddqp)) {
552 xfs_alert_tag(bp->b_mount, XFS_PTAG_VERIFIER_ERROR,
553 "Metadata corruption detected at %pS, quota %u",
554 __this_address, dqp->q_id);
555 xfs_alert(bp->b_mount, "Unmount and run xfs_repair");
556 return -EFSCORRUPTED;
559 /* copy everything from disk dquot to the incore dquot */
560 dqp->q_type = ddqp->d_type;
561 dqp->q_blk.hardlimit = be64_to_cpu(ddqp->d_blk_hardlimit);
562 dqp->q_blk.softlimit = be64_to_cpu(ddqp->d_blk_softlimit);
563 dqp->q_ino.hardlimit = be64_to_cpu(ddqp->d_ino_hardlimit);
564 dqp->q_ino.softlimit = be64_to_cpu(ddqp->d_ino_softlimit);
565 dqp->q_rtb.hardlimit = be64_to_cpu(ddqp->d_rtb_hardlimit);
566 dqp->q_rtb.softlimit = be64_to_cpu(ddqp->d_rtb_softlimit);
568 dqp->q_blk.count = be64_to_cpu(ddqp->d_bcount);
569 dqp->q_ino.count = be64_to_cpu(ddqp->d_icount);
570 dqp->q_rtb.count = be64_to_cpu(ddqp->d_rtbcount);
572 dqp->q_blk.warnings = be16_to_cpu(ddqp->d_bwarns);
573 dqp->q_ino.warnings = be16_to_cpu(ddqp->d_iwarns);
574 dqp->q_rtb.warnings = be16_to_cpu(ddqp->d_rtbwarns);
576 dqp->q_blk.timer = xfs_dquot_from_disk_ts(ddqp, ddqp->d_btimer);
577 dqp->q_ino.timer = xfs_dquot_from_disk_ts(ddqp, ddqp->d_itimer);
578 dqp->q_rtb.timer = xfs_dquot_from_disk_ts(ddqp, ddqp->d_rtbtimer);
581 * Reservation counters are defined as reservation plus current usage
582 * to avoid having to add every time.
584 dqp->q_blk.reserved = dqp->q_blk.count;
585 dqp->q_ino.reserved = dqp->q_ino.count;
586 dqp->q_rtb.reserved = dqp->q_rtb.count;
588 /* initialize the dquot speculative prealloc thresholds */
589 xfs_dquot_set_prealloc_limits(dqp);
593 /* Copy the in-core quota fields into the on-disk buffer. */
596 struct xfs_disk_dquot *ddqp,
597 struct xfs_dquot *dqp)
599 ddqp->d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
600 ddqp->d_version = XFS_DQUOT_VERSION;
601 ddqp->d_type = dqp->q_type;
602 ddqp->d_id = cpu_to_be32(dqp->q_id);
606 ddqp->d_blk_hardlimit = cpu_to_be64(dqp->q_blk.hardlimit);
607 ddqp->d_blk_softlimit = cpu_to_be64(dqp->q_blk.softlimit);
608 ddqp->d_ino_hardlimit = cpu_to_be64(dqp->q_ino.hardlimit);
609 ddqp->d_ino_softlimit = cpu_to_be64(dqp->q_ino.softlimit);
610 ddqp->d_rtb_hardlimit = cpu_to_be64(dqp->q_rtb.hardlimit);
611 ddqp->d_rtb_softlimit = cpu_to_be64(dqp->q_rtb.softlimit);
613 ddqp->d_bcount = cpu_to_be64(dqp->q_blk.count);
614 ddqp->d_icount = cpu_to_be64(dqp->q_ino.count);
615 ddqp->d_rtbcount = cpu_to_be64(dqp->q_rtb.count);
617 ddqp->d_bwarns = cpu_to_be16(dqp->q_blk.warnings);
618 ddqp->d_iwarns = cpu_to_be16(dqp->q_ino.warnings);
619 ddqp->d_rtbwarns = cpu_to_be16(dqp->q_rtb.warnings);
621 ddqp->d_btimer = xfs_dquot_to_disk_ts(dqp, dqp->q_blk.timer);
622 ddqp->d_itimer = xfs_dquot_to_disk_ts(dqp, dqp->q_ino.timer);
623 ddqp->d_rtbtimer = xfs_dquot_to_disk_ts(dqp, dqp->q_rtb.timer);
626 /* Allocate and initialize the dquot buffer for this in-core dquot. */
629 struct xfs_mount *mp,
630 struct xfs_dquot *dqp,
631 struct xfs_buf **bpp)
633 struct xfs_trans *tp;
636 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_qm_dqalloc,
637 XFS_QM_DQALLOC_SPACE_RES(mp), 0, 0, &tp);
641 error = xfs_dquot_disk_alloc(&tp, dqp, bpp);
645 error = xfs_trans_commit(tp);
648 * Buffer was held to the transaction, so we have to unlock it
649 * manually here because we're not passing it back.
658 xfs_trans_cancel(tp);
664 * Read in the ondisk dquot using dqtobp() then copy it to an incore version,
665 * and release the buffer immediately. If @can_alloc is true, fill any
666 * holes in the on-disk metadata.
670 struct xfs_mount *mp,
674 struct xfs_dquot **dqpp)
676 struct xfs_dquot *dqp;
680 dqp = xfs_dquot_alloc(mp, id, type);
681 trace_xfs_dqread(dqp);
683 /* Try to read the buffer, allocating if necessary. */
684 error = xfs_dquot_disk_read(mp, dqp, &bp);
685 if (error == -ENOENT && can_alloc)
686 error = xfs_qm_dqread_alloc(mp, dqp, &bp);
691 * At this point we should have a clean locked buffer. Copy the data
692 * to the incore dquot and release the buffer since the incore dquot
693 * has its own locking protocol so we needn't tie up the buffer any
696 ASSERT(xfs_buf_islocked(bp));
697 error = xfs_dquot_from_disk(dqp, bp);
706 trace_xfs_dqread_fail(dqp);
707 xfs_qm_dqdestroy(dqp);
713 * Advance to the next id in the current chunk, or if at the
714 * end of the chunk, skip ahead to first id in next allocated chunk
715 * using the SEEK_DATA interface.
719 struct xfs_mount *mp,
723 struct xfs_inode *quotip = xfs_quota_inode(mp, type);
724 xfs_dqid_t next_id = *id + 1; /* simple advance */
726 struct xfs_bmbt_irec got;
727 struct xfs_iext_cursor cur;
731 /* If we'd wrap past the max ID, stop */
735 /* If new ID is within the current chunk, advancing it sufficed */
736 if (next_id % mp->m_quotainfo->qi_dqperchunk) {
741 /* Nope, next_id is now past the current chunk, so find the next one */
742 start = (xfs_fsblock_t)next_id / mp->m_quotainfo->qi_dqperchunk;
744 lock_flags = xfs_ilock_data_map_shared(quotip);
745 if (!(quotip->i_df.if_flags & XFS_IFEXTENTS)) {
746 error = xfs_iread_extents(NULL, quotip, XFS_DATA_FORK);
751 if (xfs_iext_lookup_extent(quotip, "ip->i_df, start, &cur, &got)) {
752 /* contiguous chunk, bump startoff for the id calculation */
753 if (got.br_startoff < start)
754 got.br_startoff = start;
755 *id = got.br_startoff * mp->m_quotainfo->qi_dqperchunk;
760 xfs_iunlock(quotip, lock_flags);
766 * Look up the dquot in the in-core cache. If found, the dquot is returned
767 * locked and ready to go.
769 static struct xfs_dquot *
770 xfs_qm_dqget_cache_lookup(
771 struct xfs_mount *mp,
772 struct xfs_quotainfo *qi,
773 struct radix_tree_root *tree,
776 struct xfs_dquot *dqp;
779 mutex_lock(&qi->qi_tree_lock);
780 dqp = radix_tree_lookup(tree, id);
782 mutex_unlock(&qi->qi_tree_lock);
783 XFS_STATS_INC(mp, xs_qm_dqcachemisses);
788 if (dqp->q_flags & XFS_DQFLAG_FREEING) {
790 mutex_unlock(&qi->qi_tree_lock);
791 trace_xfs_dqget_freeing(dqp);
797 mutex_unlock(&qi->qi_tree_lock);
799 trace_xfs_dqget_hit(dqp);
800 XFS_STATS_INC(mp, xs_qm_dqcachehits);
805 * Try to insert a new dquot into the in-core cache. If an error occurs the
806 * caller should throw away the dquot and start over. Otherwise, the dquot
807 * is returned locked (and held by the cache) as if there had been a cache
811 xfs_qm_dqget_cache_insert(
812 struct xfs_mount *mp,
813 struct xfs_quotainfo *qi,
814 struct radix_tree_root *tree,
816 struct xfs_dquot *dqp)
820 mutex_lock(&qi->qi_tree_lock);
821 error = radix_tree_insert(tree, id, dqp);
822 if (unlikely(error)) {
823 /* Duplicate found! Caller must try again. */
824 WARN_ON(error != -EEXIST);
825 mutex_unlock(&qi->qi_tree_lock);
826 trace_xfs_dqget_dup(dqp);
830 /* Return a locked dquot to the caller, with a reference taken. */
835 mutex_unlock(&qi->qi_tree_lock);
840 /* Check our input parameters. */
843 struct xfs_mount *mp,
846 if (WARN_ON_ONCE(!XFS_IS_QUOTA_RUNNING(mp)))
850 case XFS_DQTYPE_USER:
851 if (!XFS_IS_UQUOTA_ON(mp))
854 case XFS_DQTYPE_GROUP:
855 if (!XFS_IS_GQUOTA_ON(mp))
858 case XFS_DQTYPE_PROJ:
859 if (!XFS_IS_PQUOTA_ON(mp))
869 * Given the file system, id, and type (UDQUOT/GDQUOT/PDQUOT), return a
870 * locked dquot, doing an allocation (if requested) as needed.
874 struct xfs_mount *mp,
878 struct xfs_dquot **O_dqpp)
880 struct xfs_quotainfo *qi = mp->m_quotainfo;
881 struct radix_tree_root *tree = xfs_dquot_tree(qi, type);
882 struct xfs_dquot *dqp;
885 error = xfs_qm_dqget_checks(mp, type);
890 dqp = xfs_qm_dqget_cache_lookup(mp, qi, tree, id);
896 error = xfs_qm_dqread(mp, id, type, can_alloc, &dqp);
900 error = xfs_qm_dqget_cache_insert(mp, qi, tree, id, dqp);
903 * Duplicate found. Just throw away the new dquot and start
906 xfs_qm_dqdestroy(dqp);
907 XFS_STATS_INC(mp, xs_qm_dquot_dups);
911 trace_xfs_dqget_miss(dqp);
917 * Given a dquot id and type, read and initialize a dquot from the on-disk
918 * metadata. This function is only for use during quota initialization so
919 * it ignores the dquot cache assuming that the dquot shrinker isn't set up.
920 * The caller is responsible for _qm_dqdestroy'ing the returned dquot.
923 xfs_qm_dqget_uncached(
924 struct xfs_mount *mp,
927 struct xfs_dquot **dqpp)
931 error = xfs_qm_dqget_checks(mp, type);
935 return xfs_qm_dqread(mp, id, type, 0, dqpp);
938 /* Return the quota id for a given inode and type. */
940 xfs_qm_id_for_quotatype(
941 struct xfs_inode *ip,
945 case XFS_DQTYPE_USER:
946 return i_uid_read(VFS_I(ip));
947 case XFS_DQTYPE_GROUP:
948 return i_gid_read(VFS_I(ip));
949 case XFS_DQTYPE_PROJ:
950 return ip->i_d.di_projid;
957 * Return the dquot for a given inode and type. If @can_alloc is true, then
958 * allocate blocks if needed. The inode's ILOCK must be held and it must not
959 * have already had an inode attached.
963 struct xfs_inode *ip,
966 struct xfs_dquot **O_dqpp)
968 struct xfs_mount *mp = ip->i_mount;
969 struct xfs_quotainfo *qi = mp->m_quotainfo;
970 struct radix_tree_root *tree = xfs_dquot_tree(qi, type);
971 struct xfs_dquot *dqp;
975 error = xfs_qm_dqget_checks(mp, type);
979 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
980 ASSERT(xfs_inode_dquot(ip, type) == NULL);
982 id = xfs_qm_id_for_quotatype(ip, type);
985 dqp = xfs_qm_dqget_cache_lookup(mp, qi, tree, id);
992 * Dquot cache miss. We don't want to keep the inode lock across
993 * a (potential) disk read. Also we don't want to deal with the lock
994 * ordering between quotainode and this inode. OTOH, dropping the inode
995 * lock here means dealing with a chown that can happen before
996 * we re-acquire the lock.
998 xfs_iunlock(ip, XFS_ILOCK_EXCL);
999 error = xfs_qm_dqread(mp, id, type, can_alloc, &dqp);
1000 xfs_ilock(ip, XFS_ILOCK_EXCL);
1005 * A dquot could be attached to this inode by now, since we had
1006 * dropped the ilock.
1008 if (xfs_this_quota_on(mp, type)) {
1009 struct xfs_dquot *dqp1;
1011 dqp1 = xfs_inode_dquot(ip, type);
1013 xfs_qm_dqdestroy(dqp);
1019 /* inode stays locked on return */
1020 xfs_qm_dqdestroy(dqp);
1024 error = xfs_qm_dqget_cache_insert(mp, qi, tree, id, dqp);
1027 * Duplicate found. Just throw away the new dquot and start
1030 xfs_qm_dqdestroy(dqp);
1031 XFS_STATS_INC(mp, xs_qm_dquot_dups);
1036 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1037 trace_xfs_dqget_miss(dqp);
1043 * Starting at @id and progressing upwards, look for an initialized incore
1044 * dquot, lock it, and return it.
1048 struct xfs_mount *mp,
1051 struct xfs_dquot **dqpp)
1053 struct xfs_dquot *dqp;
1057 for (; !error; error = xfs_dq_get_next_id(mp, type, &id)) {
1058 error = xfs_qm_dqget(mp, id, type, false, &dqp);
1059 if (error == -ENOENT)
1061 else if (error != 0)
1064 if (!XFS_IS_DQUOT_UNINITIALIZED(dqp)) {
1076 * Release a reference to the dquot (decrement ref-count) and unlock it.
1078 * If there is a group quota attached to this dquot, carefully release that
1079 * too without tripping over deadlocks'n'stuff.
1083 struct xfs_dquot *dqp)
1085 ASSERT(dqp->q_nrefs > 0);
1086 ASSERT(XFS_DQ_IS_LOCKED(dqp));
1088 trace_xfs_dqput(dqp);
1090 if (--dqp->q_nrefs == 0) {
1091 struct xfs_quotainfo *qi = dqp->q_mount->m_quotainfo;
1092 trace_xfs_dqput_free(dqp);
1094 if (list_lru_add(&qi->qi_lru, &dqp->q_lru))
1095 XFS_STATS_INC(dqp->q_mount, xs_qm_dquot_unused);
1101 * Release a dquot. Flush it if dirty, then dqput() it.
1102 * dquot must not be locked.
1106 struct xfs_dquot *dqp)
1111 trace_xfs_dqrele(dqp);
1115 * We don't care to flush it if the dquot is dirty here.
1116 * That will create stutters that we want to avoid.
1117 * Instead we do a delayed write when we try to reclaim
1118 * a dirty dquot. Also xfs_sync will take part of the burden...
1124 * This is the dquot flushing I/O completion routine. It is called
1125 * from interrupt level when the buffer containing the dquot is
1126 * flushed to disk. It is responsible for removing the dquot logitem
1127 * from the AIL if it has not been re-logged, and unlocking the dquot's
1128 * flush lock. This behavior is very similar to that of inodes..
1131 xfs_qm_dqflush_done(
1132 struct xfs_log_item *lip)
1134 struct xfs_dq_logitem *qip = (struct xfs_dq_logitem *)lip;
1135 struct xfs_dquot *dqp = qip->qli_dquot;
1136 struct xfs_ail *ailp = lip->li_ailp;
1140 * We only want to pull the item from the AIL if its
1141 * location in the log has not changed since we started the flush.
1142 * Thus, we only bother if the dquot's lsn has
1143 * not changed. First we check the lsn outside the lock
1144 * since it's cheaper, and then we recheck while
1145 * holding the lock before removing the dquot from the AIL.
1147 if (test_bit(XFS_LI_IN_AIL, &lip->li_flags) &&
1148 ((lip->li_lsn == qip->qli_flush_lsn) ||
1149 test_bit(XFS_LI_FAILED, &lip->li_flags))) {
1151 spin_lock(&ailp->ail_lock);
1152 xfs_clear_li_failed(lip);
1153 if (lip->li_lsn == qip->qli_flush_lsn) {
1154 /* xfs_ail_update_finish() drops the AIL lock */
1155 tail_lsn = xfs_ail_delete_one(ailp, lip);
1156 xfs_ail_update_finish(ailp, tail_lsn);
1158 spin_unlock(&ailp->ail_lock);
1163 * Release the dq's flush lock since we're done with it.
1169 xfs_buf_dquot_iodone(
1172 struct xfs_log_item *lip, *n;
1174 list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) {
1175 list_del_init(&lip->li_bio_list);
1176 xfs_qm_dqflush_done(lip);
1181 xfs_buf_dquot_io_fail(
1184 struct xfs_log_item *lip;
1186 spin_lock(&bp->b_mount->m_ail->ail_lock);
1187 list_for_each_entry(lip, &bp->b_li_list, li_bio_list)
1188 xfs_set_li_failed(lip, bp);
1189 spin_unlock(&bp->b_mount->m_ail->ail_lock);
1192 /* Check incore dquot for errors before we flush. */
1193 static xfs_failaddr_t
1194 xfs_qm_dqflush_check(
1195 struct xfs_dquot *dqp)
1197 xfs_dqtype_t type = xfs_dquot_type(dqp);
1199 if (type != XFS_DQTYPE_USER &&
1200 type != XFS_DQTYPE_GROUP &&
1201 type != XFS_DQTYPE_PROJ)
1202 return __this_address;
1207 if (dqp->q_blk.softlimit && dqp->q_blk.count > dqp->q_blk.softlimit &&
1209 return __this_address;
1211 if (dqp->q_ino.softlimit && dqp->q_ino.count > dqp->q_ino.softlimit &&
1213 return __this_address;
1215 if (dqp->q_rtb.softlimit && dqp->q_rtb.count > dqp->q_rtb.softlimit &&
1217 return __this_address;
1219 /* bigtime flag should never be set on root dquots */
1220 if (dqp->q_type & XFS_DQTYPE_BIGTIME) {
1221 if (!xfs_sb_version_hasbigtime(&dqp->q_mount->m_sb))
1222 return __this_address;
1224 return __this_address;
1231 * Write a modified dquot to disk.
1232 * The dquot must be locked and the flush lock too taken by caller.
1233 * The flush lock will not be unlocked until the dquot reaches the disk,
1234 * but the dquot is free to be unlocked and modified by the caller
1235 * in the interim. Dquot is still locked on return. This behavior is
1236 * identical to that of inodes.
1240 struct xfs_dquot *dqp,
1241 struct xfs_buf **bpp)
1243 struct xfs_mount *mp = dqp->q_mount;
1244 struct xfs_log_item *lip = &dqp->q_logitem.qli_item;
1246 struct xfs_dqblk *dqblk;
1250 ASSERT(XFS_DQ_IS_LOCKED(dqp));
1251 ASSERT(!completion_done(&dqp->q_flush));
1253 trace_xfs_dqflush(dqp);
1257 xfs_qm_dqunpin_wait(dqp);
1260 * Get the buffer containing the on-disk dquot
1262 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
1263 mp->m_quotainfo->qi_dqchunklen, XBF_TRYLOCK,
1264 &bp, &xfs_dquot_buf_ops);
1265 if (error == -EAGAIN)
1270 fa = xfs_qm_dqflush_check(dqp);
1272 xfs_alert(mp, "corrupt dquot ID 0x%x in memory at %pS",
1275 error = -EFSCORRUPTED;
1279 /* Flush the incore dquot to the ondisk buffer. */
1280 dqblk = bp->b_addr + dqp->q_bufoffset;
1281 xfs_dquot_to_disk(&dqblk->dd_diskdq, dqp);
1284 * Clear the dirty field and remember the flush lsn for later use.
1286 dqp->q_flags &= ~XFS_DQFLAG_DIRTY;
1288 xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn,
1289 &dqp->q_logitem.qli_item.li_lsn);
1292 * copy the lsn into the on-disk dquot now while we have the in memory
1293 * dquot here. This can't be done later in the write verifier as we
1294 * can't get access to the log item at that point in time.
1296 * We also calculate the CRC here so that the on-disk dquot in the
1297 * buffer always has a valid CRC. This ensures there is no possibility
1298 * of a dquot without an up-to-date CRC getting to disk.
1300 if (xfs_sb_version_hascrc(&mp->m_sb)) {
1301 dqblk->dd_lsn = cpu_to_be64(dqp->q_logitem.qli_item.li_lsn);
1302 xfs_update_cksum((char *)dqblk, sizeof(struct xfs_dqblk),
1307 * Attach the dquot to the buffer so that we can remove this dquot from
1308 * the AIL and release the flush lock once the dquot is synced to disk.
1310 bp->b_flags |= _XBF_DQUOTS;
1311 list_add_tail(&dqp->q_logitem.qli_item.li_bio_list, &bp->b_li_list);
1314 * If the buffer is pinned then push on the log so we won't
1315 * get stuck waiting in the write for too long.
1317 if (xfs_buf_ispinned(bp)) {
1318 trace_xfs_dqflush_force(dqp);
1319 xfs_log_force(mp, 0);
1322 trace_xfs_dqflush_done(dqp);
1327 dqp->q_flags &= ~XFS_DQFLAG_DIRTY;
1328 xfs_trans_ail_delete(lip, 0);
1329 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1336 * Lock two xfs_dquot structures.
1338 * To avoid deadlocks we always lock the quota structure with
1339 * the lowerd id first.
1343 struct xfs_dquot *d1,
1344 struct xfs_dquot *d2)
1348 if (d1->q_id > d2->q_id) {
1349 mutex_lock(&d2->q_qlock);
1350 mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED);
1352 mutex_lock(&d1->q_qlock);
1353 mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED);
1356 mutex_lock(&d1->q_qlock);
1358 mutex_lock(&d2->q_qlock);
1365 xfs_qm_dqzone = kmem_cache_create("xfs_dquot",
1366 sizeof(struct xfs_dquot),
1371 xfs_qm_dqtrxzone = kmem_cache_create("xfs_dqtrx",
1372 sizeof(struct xfs_dquot_acct),
1374 if (!xfs_qm_dqtrxzone)
1375 goto out_free_dqzone;
1380 kmem_cache_destroy(xfs_qm_dqzone);
1388 kmem_cache_destroy(xfs_qm_dqtrxzone);
1389 kmem_cache_destroy(xfs_qm_dqzone);
1393 * Iterate every dquot of a particular type. The caller must ensure that the
1394 * particular quota type is active. iter_fn can return negative error codes,
1395 * or -ECANCELED to indicate that it wants to stop iterating.
1399 struct xfs_mount *mp,
1401 xfs_qm_dqiterate_fn iter_fn,
1404 struct xfs_dquot *dq;
1409 error = xfs_qm_dqget_next(mp, id, type, &dq);
1410 if (error == -ENOENT)
1415 error = iter_fn(dq, type, priv);
1418 } while (error == 0 && id != 0);