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"
32 * dquot->q_qlock (xfs_dqlock() and friends)
33 * dquot->q_flush (xfs_dqflock() and friends)
36 * If two dquots need to be locked the order is user before group/project,
37 * otherwise by the lowest id first, see xfs_dqlock2.
40 struct kmem_zone *xfs_qm_dqtrxzone;
41 static struct kmem_zone *xfs_qm_dqzone;
43 static struct lock_class_key xfs_dquot_group_class;
44 static struct lock_class_key xfs_dquot_project_class;
47 * This is called to free all the memory associated with a dquot
53 ASSERT(list_empty(&dqp->q_lru));
55 kmem_free(dqp->q_logitem.qli_item.li_lv_shadow);
56 mutex_destroy(&dqp->q_qlock);
58 XFS_STATS_DEC(dqp->q_mount, xs_qm_dquot);
59 kmem_zone_free(xfs_qm_dqzone, dqp);
63 * If default limits are in force, push them into the dquot now.
64 * We overwrite the dquot limits only if they are zero and this
65 * is not the root dquot.
68 xfs_qm_adjust_dqlimits(
72 struct xfs_quotainfo *q = mp->m_quotainfo;
73 struct xfs_disk_dquot *d = &dq->q_core;
74 struct xfs_def_quota *defq;
78 defq = xfs_get_defquota(dq, q);
80 if (defq->bsoftlimit && !d->d_blk_softlimit) {
81 d->d_blk_softlimit = cpu_to_be64(defq->bsoftlimit);
84 if (defq->bhardlimit && !d->d_blk_hardlimit) {
85 d->d_blk_hardlimit = cpu_to_be64(defq->bhardlimit);
88 if (defq->isoftlimit && !d->d_ino_softlimit)
89 d->d_ino_softlimit = cpu_to_be64(defq->isoftlimit);
90 if (defq->ihardlimit && !d->d_ino_hardlimit)
91 d->d_ino_hardlimit = cpu_to_be64(defq->ihardlimit);
92 if (defq->rtbsoftlimit && !d->d_rtb_softlimit)
93 d->d_rtb_softlimit = cpu_to_be64(defq->rtbsoftlimit);
94 if (defq->rtbhardlimit && !d->d_rtb_hardlimit)
95 d->d_rtb_hardlimit = cpu_to_be64(defq->rtbhardlimit);
98 xfs_dquot_set_prealloc_limits(dq);
102 * Check the limits and timers of a dquot and start or reset timers
104 * This gets called even when quota enforcement is OFF, which makes our
105 * life a little less complicated. (We just don't reject any quota
106 * reservations in that case, when enforcement is off).
107 * We also return 0 as the values of the timers in Q_GETQUOTA calls, when
109 * In contrast, warnings are a little different in that they don't
110 * 'automatically' get started when limits get exceeded. They do
111 * get reset to zero, however, when we find the count to be under
112 * the soft limit (they are only ever set non-zero via userspace).
115 xfs_qm_adjust_dqtimers(
122 if (d->d_blk_hardlimit)
123 ASSERT(be64_to_cpu(d->d_blk_softlimit) <=
124 be64_to_cpu(d->d_blk_hardlimit));
125 if (d->d_ino_hardlimit)
126 ASSERT(be64_to_cpu(d->d_ino_softlimit) <=
127 be64_to_cpu(d->d_ino_hardlimit));
128 if (d->d_rtb_hardlimit)
129 ASSERT(be64_to_cpu(d->d_rtb_softlimit) <=
130 be64_to_cpu(d->d_rtb_hardlimit));
134 if ((d->d_blk_softlimit &&
135 (be64_to_cpu(d->d_bcount) >
136 be64_to_cpu(d->d_blk_softlimit))) ||
137 (d->d_blk_hardlimit &&
138 (be64_to_cpu(d->d_bcount) >
139 be64_to_cpu(d->d_blk_hardlimit)))) {
140 d->d_btimer = cpu_to_be32(get_seconds() +
141 mp->m_quotainfo->qi_btimelimit);
146 if ((!d->d_blk_softlimit ||
147 (be64_to_cpu(d->d_bcount) <=
148 be64_to_cpu(d->d_blk_softlimit))) &&
149 (!d->d_blk_hardlimit ||
150 (be64_to_cpu(d->d_bcount) <=
151 be64_to_cpu(d->d_blk_hardlimit)))) {
157 if ((d->d_ino_softlimit &&
158 (be64_to_cpu(d->d_icount) >
159 be64_to_cpu(d->d_ino_softlimit))) ||
160 (d->d_ino_hardlimit &&
161 (be64_to_cpu(d->d_icount) >
162 be64_to_cpu(d->d_ino_hardlimit)))) {
163 d->d_itimer = cpu_to_be32(get_seconds() +
164 mp->m_quotainfo->qi_itimelimit);
169 if ((!d->d_ino_softlimit ||
170 (be64_to_cpu(d->d_icount) <=
171 be64_to_cpu(d->d_ino_softlimit))) &&
172 (!d->d_ino_hardlimit ||
173 (be64_to_cpu(d->d_icount) <=
174 be64_to_cpu(d->d_ino_hardlimit)))) {
179 if (!d->d_rtbtimer) {
180 if ((d->d_rtb_softlimit &&
181 (be64_to_cpu(d->d_rtbcount) >
182 be64_to_cpu(d->d_rtb_softlimit))) ||
183 (d->d_rtb_hardlimit &&
184 (be64_to_cpu(d->d_rtbcount) >
185 be64_to_cpu(d->d_rtb_hardlimit)))) {
186 d->d_rtbtimer = cpu_to_be32(get_seconds() +
187 mp->m_quotainfo->qi_rtbtimelimit);
192 if ((!d->d_rtb_softlimit ||
193 (be64_to_cpu(d->d_rtbcount) <=
194 be64_to_cpu(d->d_rtb_softlimit))) &&
195 (!d->d_rtb_hardlimit ||
196 (be64_to_cpu(d->d_rtbcount) <=
197 be64_to_cpu(d->d_rtb_hardlimit)))) {
204 * initialize a buffer full of dquots and log the whole thing
207 xfs_qm_init_dquot_blk(
214 struct xfs_quotainfo *q = mp->m_quotainfo;
220 ASSERT(xfs_buf_islocked(bp));
225 * ID of the first dquot in the block - id's are zero based.
227 curid = id - (id % q->qi_dqperchunk);
228 memset(d, 0, BBTOB(q->qi_dqchunklen));
229 for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++) {
230 d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
231 d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
232 d->dd_diskdq.d_id = cpu_to_be32(curid);
233 d->dd_diskdq.d_flags = type;
234 if (xfs_sb_version_hascrc(&mp->m_sb)) {
235 uuid_copy(&d->dd_uuid, &mp->m_sb.sb_meta_uuid);
236 xfs_update_cksum((char *)d, sizeof(struct xfs_dqblk),
241 xfs_trans_dquot_buf(tp, bp,
242 (type & XFS_DQ_USER ? XFS_BLF_UDQUOT_BUF :
243 ((type & XFS_DQ_PROJ) ? XFS_BLF_PDQUOT_BUF :
244 XFS_BLF_GDQUOT_BUF)));
245 xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1);
249 * Initialize the dynamic speculative preallocation thresholds. The lo/hi
250 * watermarks correspond to the soft and hard limits by default. If a soft limit
251 * is not specified, we use 95% of the hard limit.
254 xfs_dquot_set_prealloc_limits(struct xfs_dquot *dqp)
258 dqp->q_prealloc_hi_wmark = be64_to_cpu(dqp->q_core.d_blk_hardlimit);
259 dqp->q_prealloc_lo_wmark = be64_to_cpu(dqp->q_core.d_blk_softlimit);
260 if (!dqp->q_prealloc_lo_wmark) {
261 dqp->q_prealloc_lo_wmark = dqp->q_prealloc_hi_wmark;
262 do_div(dqp->q_prealloc_lo_wmark, 100);
263 dqp->q_prealloc_lo_wmark *= 95;
266 space = dqp->q_prealloc_hi_wmark;
269 dqp->q_low_space[XFS_QLOWSP_1_PCNT] = space;
270 dqp->q_low_space[XFS_QLOWSP_3_PCNT] = space * 3;
271 dqp->q_low_space[XFS_QLOWSP_5_PCNT] = space * 5;
275 * Ensure that the given in-core dquot has a buffer on disk backing it, and
276 * return the buffer locked and held. This is called when the bmapi finds a
280 xfs_dquot_disk_alloc(
281 struct xfs_trans **tpp,
282 struct xfs_dquot *dqp,
283 struct xfs_buf **bpp)
285 struct xfs_bmbt_irec map;
286 struct xfs_trans *tp = *tpp;
287 struct xfs_mount *mp = tp->t_mountp;
289 struct xfs_inode *quotip = xfs_quota_inode(mp, dqp->dq_flags);
293 trace_xfs_dqalloc(dqp);
295 xfs_ilock(quotip, XFS_ILOCK_EXCL);
296 if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) {
298 * Return if this type of quotas is turned off while we didn't
301 xfs_iunlock(quotip, XFS_ILOCK_EXCL);
305 /* Create the block mapping. */
306 xfs_trans_ijoin(tp, quotip, XFS_ILOCK_EXCL);
307 error = xfs_bmapi_write(tp, quotip, dqp->q_fileoffset,
308 XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA,
309 XFS_QM_DQALLOC_SPACE_RES(mp), &map, &nmaps);
312 ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB);
314 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
315 (map.br_startblock != HOLESTARTBLOCK));
318 * Keep track of the blkno to save a lookup later
320 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
322 /* now we can just get the buffer (there's nothing to read yet) */
323 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, dqp->q_blkno,
324 mp->m_quotainfo->qi_dqchunklen, 0);
327 bp->b_ops = &xfs_dquot_buf_ops;
330 * Make a chunk of dquots out of this buffer and log
333 xfs_qm_init_dquot_blk(tp, mp, be32_to_cpu(dqp->q_core.d_id),
334 dqp->dq_flags & XFS_DQ_ALLTYPES, bp);
335 xfs_buf_set_ref(bp, XFS_DQUOT_REF);
338 * Hold the buffer and join it to the dfops so that we'll still own
339 * the buffer when we return to the caller. The buffer disposal on
340 * error must be paid attention to very carefully, as it has been
341 * broken since commit efa092f3d4c6 "[XFS] Fixes a bug in the quota
342 * code when allocating a new dquot record" in 2005, and the later
343 * conversion to xfs_defer_ops in commit 310a75a3c6c747 failed to keep
344 * the buffer locked across the _defer_finish call. We can now do
345 * this correctly with xfs_defer_bjoin.
347 * Above, we allocated a disk block for the dquot information and used
348 * get_buf to initialize the dquot. If the _defer_finish fails, the old
349 * transaction is gone but the new buffer is not joined or held to any
350 * transaction, so we must _buf_relse it.
352 * If everything succeeds, the caller of this function is returned a
353 * buffer that is locked and held to the transaction. The caller
354 * is responsible for unlocking any buffer passed back, either
355 * manually or by committing the transaction. On error, the buffer is
356 * released and not passed back.
358 xfs_trans_bhold(tp, bp);
359 error = xfs_defer_finish(tpp);
361 xfs_trans_bhold_release(*tpp, bp);
362 xfs_trans_brelse(*tpp, bp);
370 * Read in the in-core dquot's on-disk metadata and return the buffer.
371 * Returns ENOENT to signal a hole.
375 struct xfs_mount *mp,
376 struct xfs_dquot *dqp,
377 struct xfs_buf **bpp)
379 struct xfs_bmbt_irec map;
381 struct xfs_inode *quotip = xfs_quota_inode(mp, dqp->dq_flags);
386 lock_mode = xfs_ilock_data_map_shared(quotip);
387 if (!xfs_this_quota_on(mp, dqp->dq_flags)) {
389 * Return if this type of quotas is turned off while we
390 * didn't have the quota inode lock.
392 xfs_iunlock(quotip, lock_mode);
397 * Find the block map; no allocations yet
399 error = xfs_bmapi_read(quotip, dqp->q_fileoffset,
400 XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0);
401 xfs_iunlock(quotip, lock_mode);
406 ASSERT(map.br_blockcount >= 1);
407 ASSERT(map.br_startblock != DELAYSTARTBLOCK);
408 if (map.br_startblock == HOLESTARTBLOCK)
411 trace_xfs_dqtobp_read(dqp);
414 * store the blkno etc so that we don't have to do the
415 * mapping all the time
417 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
419 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
420 mp->m_quotainfo->qi_dqchunklen, 0, &bp,
427 ASSERT(xfs_buf_islocked(bp));
428 xfs_buf_set_ref(bp, XFS_DQUOT_REF);
434 /* Allocate and initialize everything we need for an incore dquot. */
435 STATIC struct xfs_dquot *
437 struct xfs_mount *mp,
441 struct xfs_dquot *dqp;
443 dqp = kmem_zone_zalloc(xfs_qm_dqzone, 0);
445 dqp->dq_flags = type;
446 dqp->q_core.d_id = cpu_to_be32(id);
448 INIT_LIST_HEAD(&dqp->q_lru);
449 mutex_init(&dqp->q_qlock);
450 init_waitqueue_head(&dqp->q_pinwait);
451 dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk;
453 * Offset of dquot in the (fixed sized) dquot chunk.
455 dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) *
459 * Because we want to use a counting completion, complete
460 * the flush completion once to allow a single access to
461 * the flush completion without blocking.
463 init_completion(&dqp->q_flush);
464 complete(&dqp->q_flush);
467 * Make sure group quotas have a different lock class than user
472 /* uses the default lock class */
475 lockdep_set_class(&dqp->q_qlock, &xfs_dquot_group_class);
478 lockdep_set_class(&dqp->q_qlock, &xfs_dquot_project_class);
485 xfs_qm_dquot_logitem_init(dqp);
487 XFS_STATS_INC(mp, xs_qm_dquot);
491 /* Copy the in-core quota fields in from the on-disk buffer. */
494 struct xfs_dquot *dqp,
497 struct xfs_disk_dquot *ddqp = bp->b_addr + dqp->q_bufoffset;
499 /* copy everything from disk dquot to the incore dquot */
500 memcpy(&dqp->q_core, ddqp, sizeof(xfs_disk_dquot_t));
503 * Reservation counters are defined as reservation plus current usage
504 * to avoid having to add every time.
506 dqp->q_res_bcount = be64_to_cpu(ddqp->d_bcount);
507 dqp->q_res_icount = be64_to_cpu(ddqp->d_icount);
508 dqp->q_res_rtbcount = be64_to_cpu(ddqp->d_rtbcount);
510 /* initialize the dquot speculative prealloc thresholds */
511 xfs_dquot_set_prealloc_limits(dqp);
514 /* Allocate and initialize the dquot buffer for this in-core dquot. */
517 struct xfs_mount *mp,
518 struct xfs_dquot *dqp,
519 struct xfs_buf **bpp)
521 struct xfs_trans *tp;
524 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_qm_dqalloc,
525 XFS_QM_DQALLOC_SPACE_RES(mp), 0, 0, &tp);
529 error = xfs_dquot_disk_alloc(&tp, dqp, bpp);
533 error = xfs_trans_commit(tp);
536 * Buffer was held to the transaction, so we have to unlock it
537 * manually here because we're not passing it back.
546 xfs_trans_cancel(tp);
552 * Read in the ondisk dquot using dqtobp() then copy it to an incore version,
553 * and release the buffer immediately. If @can_alloc is true, fill any
554 * holes in the on-disk metadata.
558 struct xfs_mount *mp,
562 struct xfs_dquot **dqpp)
564 struct xfs_dquot *dqp;
568 dqp = xfs_dquot_alloc(mp, id, type);
569 trace_xfs_dqread(dqp);
571 /* Try to read the buffer, allocating if necessary. */
572 error = xfs_dquot_disk_read(mp, dqp, &bp);
573 if (error == -ENOENT && can_alloc)
574 error = xfs_qm_dqread_alloc(mp, dqp, &bp);
579 * At this point we should have a clean locked buffer. Copy the data
580 * to the incore dquot and release the buffer since the incore dquot
581 * has its own locking protocol so we needn't tie up the buffer any
584 ASSERT(xfs_buf_islocked(bp));
585 xfs_dquot_from_disk(dqp, bp);
592 trace_xfs_dqread_fail(dqp);
593 xfs_qm_dqdestroy(dqp);
599 * Advance to the next id in the current chunk, or if at the
600 * end of the chunk, skip ahead to first id in next allocated chunk
601 * using the SEEK_DATA interface.
605 struct xfs_mount *mp,
609 struct xfs_inode *quotip = xfs_quota_inode(mp, type);
610 xfs_dqid_t next_id = *id + 1; /* simple advance */
612 struct xfs_bmbt_irec got;
613 struct xfs_iext_cursor cur;
617 /* If we'd wrap past the max ID, stop */
621 /* If new ID is within the current chunk, advancing it sufficed */
622 if (next_id % mp->m_quotainfo->qi_dqperchunk) {
627 /* Nope, next_id is now past the current chunk, so find the next one */
628 start = (xfs_fsblock_t)next_id / mp->m_quotainfo->qi_dqperchunk;
630 lock_flags = xfs_ilock_data_map_shared(quotip);
631 if (!(quotip->i_df.if_flags & XFS_IFEXTENTS)) {
632 error = xfs_iread_extents(NULL, quotip, XFS_DATA_FORK);
637 if (xfs_iext_lookup_extent(quotip, "ip->i_df, start, &cur, &got)) {
638 /* contiguous chunk, bump startoff for the id calculation */
639 if (got.br_startoff < start)
640 got.br_startoff = start;
641 *id = got.br_startoff * mp->m_quotainfo->qi_dqperchunk;
646 xfs_iunlock(quotip, lock_flags);
652 * Look up the dquot in the in-core cache. If found, the dquot is returned
653 * locked and ready to go.
655 static struct xfs_dquot *
656 xfs_qm_dqget_cache_lookup(
657 struct xfs_mount *mp,
658 struct xfs_quotainfo *qi,
659 struct radix_tree_root *tree,
662 struct xfs_dquot *dqp;
665 mutex_lock(&qi->qi_tree_lock);
666 dqp = radix_tree_lookup(tree, id);
668 mutex_unlock(&qi->qi_tree_lock);
669 XFS_STATS_INC(mp, xs_qm_dqcachemisses);
674 if (dqp->dq_flags & XFS_DQ_FREEING) {
676 mutex_unlock(&qi->qi_tree_lock);
677 trace_xfs_dqget_freeing(dqp);
683 mutex_unlock(&qi->qi_tree_lock);
685 trace_xfs_dqget_hit(dqp);
686 XFS_STATS_INC(mp, xs_qm_dqcachehits);
691 * Try to insert a new dquot into the in-core cache. If an error occurs the
692 * caller should throw away the dquot and start over. Otherwise, the dquot
693 * is returned locked (and held by the cache) as if there had been a cache
697 xfs_qm_dqget_cache_insert(
698 struct xfs_mount *mp,
699 struct xfs_quotainfo *qi,
700 struct radix_tree_root *tree,
702 struct xfs_dquot *dqp)
706 mutex_lock(&qi->qi_tree_lock);
707 error = radix_tree_insert(tree, id, dqp);
708 if (unlikely(error)) {
709 /* Duplicate found! Caller must try again. */
710 WARN_ON(error != -EEXIST);
711 mutex_unlock(&qi->qi_tree_lock);
712 trace_xfs_dqget_dup(dqp);
716 /* Return a locked dquot to the caller, with a reference taken. */
721 mutex_unlock(&qi->qi_tree_lock);
726 /* Check our input parameters. */
729 struct xfs_mount *mp,
732 if (WARN_ON_ONCE(!XFS_IS_QUOTA_RUNNING(mp)))
737 if (!XFS_IS_UQUOTA_ON(mp))
741 if (!XFS_IS_GQUOTA_ON(mp))
745 if (!XFS_IS_PQUOTA_ON(mp))
755 * Given the file system, id, and type (UDQUOT/GDQUOT), return a a locked
756 * dquot, doing an allocation (if requested) as needed.
760 struct xfs_mount *mp,
764 struct xfs_dquot **O_dqpp)
766 struct xfs_quotainfo *qi = mp->m_quotainfo;
767 struct radix_tree_root *tree = xfs_dquot_tree(qi, type);
768 struct xfs_dquot *dqp;
771 error = xfs_qm_dqget_checks(mp, type);
776 dqp = xfs_qm_dqget_cache_lookup(mp, qi, tree, id);
782 error = xfs_qm_dqread(mp, id, type, can_alloc, &dqp);
786 error = xfs_qm_dqget_cache_insert(mp, qi, tree, id, dqp);
789 * Duplicate found. Just throw away the new dquot and start
792 xfs_qm_dqdestroy(dqp);
793 XFS_STATS_INC(mp, xs_qm_dquot_dups);
797 trace_xfs_dqget_miss(dqp);
803 * Given a dquot id and type, read and initialize a dquot from the on-disk
804 * metadata. This function is only for use during quota initialization so
805 * it ignores the dquot cache assuming that the dquot shrinker isn't set up.
806 * The caller is responsible for _qm_dqdestroy'ing the returned dquot.
809 xfs_qm_dqget_uncached(
810 struct xfs_mount *mp,
813 struct xfs_dquot **dqpp)
817 error = xfs_qm_dqget_checks(mp, type);
821 return xfs_qm_dqread(mp, id, type, 0, dqpp);
824 /* Return the quota id for a given inode and type. */
826 xfs_qm_id_for_quotatype(
827 struct xfs_inode *ip,
832 return ip->i_d.di_uid;
834 return ip->i_d.di_gid;
836 return xfs_get_projid(ip);
843 * Return the dquot for a given inode and type. If @can_alloc is true, then
844 * allocate blocks if needed. The inode's ILOCK must be held and it must not
845 * have already had an inode attached.
849 struct xfs_inode *ip,
852 struct xfs_dquot **O_dqpp)
854 struct xfs_mount *mp = ip->i_mount;
855 struct xfs_quotainfo *qi = mp->m_quotainfo;
856 struct radix_tree_root *tree = xfs_dquot_tree(qi, type);
857 struct xfs_dquot *dqp;
861 error = xfs_qm_dqget_checks(mp, type);
865 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
866 ASSERT(xfs_inode_dquot(ip, type) == NULL);
868 id = xfs_qm_id_for_quotatype(ip, type);
871 dqp = xfs_qm_dqget_cache_lookup(mp, qi, tree, id);
878 * Dquot cache miss. We don't want to keep the inode lock across
879 * a (potential) disk read. Also we don't want to deal with the lock
880 * ordering between quotainode and this inode. OTOH, dropping the inode
881 * lock here means dealing with a chown that can happen before
882 * we re-acquire the lock.
884 xfs_iunlock(ip, XFS_ILOCK_EXCL);
885 error = xfs_qm_dqread(mp, id, type, can_alloc, &dqp);
886 xfs_ilock(ip, XFS_ILOCK_EXCL);
891 * A dquot could be attached to this inode by now, since we had
894 if (xfs_this_quota_on(mp, type)) {
895 struct xfs_dquot *dqp1;
897 dqp1 = xfs_inode_dquot(ip, type);
899 xfs_qm_dqdestroy(dqp);
905 /* inode stays locked on return */
906 xfs_qm_dqdestroy(dqp);
910 error = xfs_qm_dqget_cache_insert(mp, qi, tree, id, dqp);
913 * Duplicate found. Just throw away the new dquot and start
916 xfs_qm_dqdestroy(dqp);
917 XFS_STATS_INC(mp, xs_qm_dquot_dups);
922 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
923 trace_xfs_dqget_miss(dqp);
929 * Starting at @id and progressing upwards, look for an initialized incore
930 * dquot, lock it, and return it.
934 struct xfs_mount *mp,
937 struct xfs_dquot **dqpp)
939 struct xfs_dquot *dqp;
943 for (; !error; error = xfs_dq_get_next_id(mp, type, &id)) {
944 error = xfs_qm_dqget(mp, id, type, false, &dqp);
945 if (error == -ENOENT)
950 if (!XFS_IS_DQUOT_UNINITIALIZED(dqp)) {
962 * Release a reference to the dquot (decrement ref-count) and unlock it.
964 * If there is a group quota attached to this dquot, carefully release that
965 * too without tripping over deadlocks'n'stuff.
969 struct xfs_dquot *dqp)
971 ASSERT(dqp->q_nrefs > 0);
972 ASSERT(XFS_DQ_IS_LOCKED(dqp));
974 trace_xfs_dqput(dqp);
976 if (--dqp->q_nrefs == 0) {
977 struct xfs_quotainfo *qi = dqp->q_mount->m_quotainfo;
978 trace_xfs_dqput_free(dqp);
980 if (list_lru_add(&qi->qi_lru, &dqp->q_lru))
981 XFS_STATS_INC(dqp->q_mount, xs_qm_dquot_unused);
987 * Release a dquot. Flush it if dirty, then dqput() it.
988 * dquot must not be locked.
997 trace_xfs_dqrele(dqp);
1001 * We don't care to flush it if the dquot is dirty here.
1002 * That will create stutters that we want to avoid.
1003 * Instead we do a delayed write when we try to reclaim
1004 * a dirty dquot. Also xfs_sync will take part of the burden...
1010 * This is the dquot flushing I/O completion routine. It is called
1011 * from interrupt level when the buffer containing the dquot is
1012 * flushed to disk. It is responsible for removing the dquot logitem
1013 * from the AIL if it has not been re-logged, and unlocking the dquot's
1014 * flush lock. This behavior is very similar to that of inodes..
1017 xfs_qm_dqflush_done(
1019 struct xfs_log_item *lip)
1021 xfs_dq_logitem_t *qip = (struct xfs_dq_logitem *)lip;
1022 xfs_dquot_t *dqp = qip->qli_dquot;
1023 struct xfs_ail *ailp = lip->li_ailp;
1026 * We only want to pull the item from the AIL if its
1027 * location in the log has not changed since we started the flush.
1028 * Thus, we only bother if the dquot's lsn has
1029 * not changed. First we check the lsn outside the lock
1030 * since it's cheaper, and then we recheck while
1031 * holding the lock before removing the dquot from the AIL.
1033 if (test_bit(XFS_LI_IN_AIL, &lip->li_flags) &&
1034 ((lip->li_lsn == qip->qli_flush_lsn) ||
1035 test_bit(XFS_LI_FAILED, &lip->li_flags))) {
1037 /* xfs_trans_ail_delete() drops the AIL lock. */
1038 spin_lock(&ailp->ail_lock);
1039 if (lip->li_lsn == qip->qli_flush_lsn) {
1040 xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE);
1043 * Clear the failed state since we are about to drop the
1046 xfs_clear_li_failed(lip);
1047 spin_unlock(&ailp->ail_lock);
1052 * Release the dq's flush lock since we're done with it.
1058 * Write a modified dquot to disk.
1059 * The dquot must be locked and the flush lock too taken by caller.
1060 * The flush lock will not be unlocked until the dquot reaches the disk,
1061 * but the dquot is free to be unlocked and modified by the caller
1062 * in the interim. Dquot is still locked on return. This behavior is
1063 * identical to that of inodes.
1067 struct xfs_dquot *dqp,
1068 struct xfs_buf **bpp)
1070 struct xfs_mount *mp = dqp->q_mount;
1072 struct xfs_dqblk *dqb;
1073 struct xfs_disk_dquot *ddqp;
1077 ASSERT(XFS_DQ_IS_LOCKED(dqp));
1078 ASSERT(!completion_done(&dqp->q_flush));
1080 trace_xfs_dqflush(dqp);
1084 xfs_qm_dqunpin_wait(dqp);
1087 * This may have been unpinned because the filesystem is shutting
1088 * down forcibly. If that's the case we must not write this dquot
1089 * to disk, because the log record didn't make it to disk.
1091 * We also have to remove the log item from the AIL in this case,
1092 * as we wait for an emptry AIL as part of the unmount process.
1094 if (XFS_FORCED_SHUTDOWN(mp)) {
1095 struct xfs_log_item *lip = &dqp->q_logitem.qli_item;
1096 dqp->dq_flags &= ~XFS_DQ_DIRTY;
1098 xfs_trans_ail_remove(lip, SHUTDOWN_CORRUPT_INCORE);
1105 * Get the buffer containing the on-disk dquot
1107 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
1108 mp->m_quotainfo->qi_dqchunklen, 0, &bp,
1109 &xfs_dquot_buf_ops);
1114 * Calculate the location of the dquot inside the buffer.
1116 dqb = bp->b_addr + dqp->q_bufoffset;
1117 ddqp = &dqb->dd_diskdq;
1120 * A simple sanity check in case we got a corrupted dquot.
1122 fa = xfs_dqblk_verify(mp, dqb, be32_to_cpu(ddqp->d_id), 0);
1124 xfs_alert(mp, "corrupt dquot ID 0x%x in memory at %pS",
1125 be32_to_cpu(ddqp->d_id), fa);
1128 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1132 /* This is the only portion of data that needs to persist */
1133 memcpy(ddqp, &dqp->q_core, sizeof(xfs_disk_dquot_t));
1136 * Clear the dirty field and remember the flush lsn for later use.
1138 dqp->dq_flags &= ~XFS_DQ_DIRTY;
1140 xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn,
1141 &dqp->q_logitem.qli_item.li_lsn);
1144 * copy the lsn into the on-disk dquot now while we have the in memory
1145 * dquot here. This can't be done later in the write verifier as we
1146 * can't get access to the log item at that point in time.
1148 * We also calculate the CRC here so that the on-disk dquot in the
1149 * buffer always has a valid CRC. This ensures there is no possibility
1150 * of a dquot without an up-to-date CRC getting to disk.
1152 if (xfs_sb_version_hascrc(&mp->m_sb)) {
1153 dqb->dd_lsn = cpu_to_be64(dqp->q_logitem.qli_item.li_lsn);
1154 xfs_update_cksum((char *)dqb, sizeof(struct xfs_dqblk),
1159 * Attach an iodone routine so that we can remove this dquot from the
1160 * AIL and release the flush lock once the dquot is synced to disk.
1162 xfs_buf_attach_iodone(bp, xfs_qm_dqflush_done,
1163 &dqp->q_logitem.qli_item);
1166 * If the buffer is pinned then push on the log so we won't
1167 * get stuck waiting in the write for too long.
1169 if (xfs_buf_ispinned(bp)) {
1170 trace_xfs_dqflush_force(dqp);
1171 xfs_log_force(mp, 0);
1174 trace_xfs_dqflush_done(dqp);
1184 * Lock two xfs_dquot structures.
1186 * To avoid deadlocks we always lock the quota structure with
1187 * the lowerd id first.
1196 if (be32_to_cpu(d1->q_core.d_id) >
1197 be32_to_cpu(d2->q_core.d_id)) {
1198 mutex_lock(&d2->q_qlock);
1199 mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED);
1201 mutex_lock(&d1->q_qlock);
1202 mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED);
1205 mutex_lock(&d1->q_qlock);
1207 mutex_lock(&d2->q_qlock);
1215 kmem_zone_init(sizeof(struct xfs_dquot), "xfs_dquot");
1220 kmem_zone_init(sizeof(struct xfs_dquot_acct), "xfs_dqtrx");
1221 if (!xfs_qm_dqtrxzone)
1222 goto out_free_dqzone;
1227 kmem_zone_destroy(xfs_qm_dqzone);
1235 kmem_zone_destroy(xfs_qm_dqtrxzone);
1236 kmem_zone_destroy(xfs_qm_dqzone);
1240 * Iterate every dquot of a particular type. The caller must ensure that the
1241 * particular quota type is active. iter_fn can return negative error codes,
1242 * or -ECANCELED to indicate that it wants to stop iterating.
1246 struct xfs_mount *mp,
1248 xfs_qm_dqiterate_fn iter_fn,
1251 struct xfs_dquot *dq;
1256 error = xfs_qm_dqget_next(mp, id, dqtype, &dq);
1257 if (error == -ENOENT)
1262 error = iter_fn(dq, dqtype, priv);
1263 id = be32_to_cpu(dq->q_core.d_id);
1266 } while (error == 0 && id != 0);