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
51 struct xfs_dquot *dqp)
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(
116 struct xfs_mount *mp,
117 struct xfs_disk_dquot *d)
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(
208 struct xfs_trans *tp,
209 struct xfs_mount *mp,
214 struct xfs_quotainfo *q = mp->m_quotainfo;
218 unsigned int blftype;
222 ASSERT(xfs_buf_islocked(bp));
227 * ID of the first dquot in the block - id's are zero based.
229 curid = id - (id % q->qi_dqperchunk);
230 memset(d, 0, BBTOB(q->qi_dqchunklen));
231 for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++) {
232 d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
233 d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
234 d->dd_diskdq.d_id = cpu_to_be32(curid);
235 d->dd_diskdq.d_flags = type;
236 if (xfs_sb_version_hascrc(&mp->m_sb)) {
237 uuid_copy(&d->dd_uuid, &mp->m_sb.sb_meta_uuid);
238 xfs_update_cksum((char *)d, sizeof(struct xfs_dqblk),
243 if (type & XFS_DQ_USER) {
244 qflag = XFS_UQUOTA_CHKD;
245 blftype = XFS_BLF_UDQUOT_BUF;
246 } else if (type & XFS_DQ_PROJ) {
247 qflag = XFS_PQUOTA_CHKD;
248 blftype = XFS_BLF_PDQUOT_BUF;
250 qflag = XFS_GQUOTA_CHKD;
251 blftype = XFS_BLF_GDQUOT_BUF;
254 xfs_trans_dquot_buf(tp, bp, blftype);
257 * quotacheck uses delayed writes to update all the dquots on disk in an
258 * efficient manner instead of logging the individual dquot changes as
259 * they are made. However if we log the buffer allocated here and crash
260 * after quotacheck while the logged initialisation is still in the
261 * active region of the log, log recovery can replay the dquot buffer
262 * initialisation over the top of the checked dquots and corrupt quota
265 * To avoid this problem, quotacheck cannot log the initialised buffer.
266 * We must still dirty the buffer and write it back before the
267 * allocation transaction clears the log. Therefore, mark the buffer as
268 * ordered instead of logging it directly. This is safe for quotacheck
269 * because it detects and repairs allocated but initialized dquot blocks
270 * in the quota inodes.
272 if (!(mp->m_qflags & qflag))
273 xfs_trans_ordered_buf(tp, bp);
275 xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1);
279 * Initialize the dynamic speculative preallocation thresholds. The lo/hi
280 * watermarks correspond to the soft and hard limits by default. If a soft limit
281 * is not specified, we use 95% of the hard limit.
284 xfs_dquot_set_prealloc_limits(struct xfs_dquot *dqp)
288 dqp->q_prealloc_hi_wmark = be64_to_cpu(dqp->q_core.d_blk_hardlimit);
289 dqp->q_prealloc_lo_wmark = be64_to_cpu(dqp->q_core.d_blk_softlimit);
290 if (!dqp->q_prealloc_lo_wmark) {
291 dqp->q_prealloc_lo_wmark = dqp->q_prealloc_hi_wmark;
292 do_div(dqp->q_prealloc_lo_wmark, 100);
293 dqp->q_prealloc_lo_wmark *= 95;
296 space = dqp->q_prealloc_hi_wmark;
299 dqp->q_low_space[XFS_QLOWSP_1_PCNT] = space;
300 dqp->q_low_space[XFS_QLOWSP_3_PCNT] = space * 3;
301 dqp->q_low_space[XFS_QLOWSP_5_PCNT] = space * 5;
305 * Ensure that the given in-core dquot has a buffer on disk backing it, and
306 * return the buffer locked and held. This is called when the bmapi finds a
310 xfs_dquot_disk_alloc(
311 struct xfs_trans **tpp,
312 struct xfs_dquot *dqp,
313 struct xfs_buf **bpp)
315 struct xfs_bmbt_irec map;
316 struct xfs_trans *tp = *tpp;
317 struct xfs_mount *mp = tp->t_mountp;
319 struct xfs_inode *quotip = xfs_quota_inode(mp, dqp->dq_flags);
323 trace_xfs_dqalloc(dqp);
325 xfs_ilock(quotip, XFS_ILOCK_EXCL);
326 if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) {
328 * Return if this type of quotas is turned off while we didn't
331 xfs_iunlock(quotip, XFS_ILOCK_EXCL);
335 /* Create the block mapping. */
336 xfs_trans_ijoin(tp, quotip, XFS_ILOCK_EXCL);
337 error = xfs_bmapi_write(tp, quotip, dqp->q_fileoffset,
338 XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA,
339 XFS_QM_DQALLOC_SPACE_RES(mp), &map, &nmaps);
342 ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB);
344 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
345 (map.br_startblock != HOLESTARTBLOCK));
348 * Keep track of the blkno to save a lookup later
350 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
352 /* now we can just get the buffer (there's nothing to read yet) */
353 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, dqp->q_blkno,
354 mp->m_quotainfo->qi_dqchunklen, 0);
357 bp->b_ops = &xfs_dquot_buf_ops;
360 * Make a chunk of dquots out of this buffer and log
363 xfs_qm_init_dquot_blk(tp, mp, be32_to_cpu(dqp->q_core.d_id),
364 dqp->dq_flags & XFS_DQ_ALLTYPES, bp);
365 xfs_buf_set_ref(bp, XFS_DQUOT_REF);
368 * Hold the buffer and join it to the dfops so that we'll still own
369 * the buffer when we return to the caller. The buffer disposal on
370 * error must be paid attention to very carefully, as it has been
371 * broken since commit efa092f3d4c6 "[XFS] Fixes a bug in the quota
372 * code when allocating a new dquot record" in 2005, and the later
373 * conversion to xfs_defer_ops in commit 310a75a3c6c747 failed to keep
374 * the buffer locked across the _defer_finish call. We can now do
375 * this correctly with xfs_defer_bjoin.
377 * Above, we allocated a disk block for the dquot information and used
378 * get_buf to initialize the dquot. If the _defer_finish fails, the old
379 * transaction is gone but the new buffer is not joined or held to any
380 * transaction, so we must _buf_relse it.
382 * If everything succeeds, the caller of this function is returned a
383 * buffer that is locked and held to the transaction. The caller
384 * is responsible for unlocking any buffer passed back, either
385 * manually or by committing the transaction. On error, the buffer is
386 * released and not passed back.
388 xfs_trans_bhold(tp, bp);
389 error = xfs_defer_finish(tpp);
391 xfs_trans_bhold_release(*tpp, bp);
392 xfs_trans_brelse(*tpp, bp);
400 * Read in the in-core dquot's on-disk metadata and return the buffer.
401 * Returns ENOENT to signal a hole.
405 struct xfs_mount *mp,
406 struct xfs_dquot *dqp,
407 struct xfs_buf **bpp)
409 struct xfs_bmbt_irec map;
411 struct xfs_inode *quotip = xfs_quota_inode(mp, dqp->dq_flags);
416 lock_mode = xfs_ilock_data_map_shared(quotip);
417 if (!xfs_this_quota_on(mp, dqp->dq_flags)) {
419 * Return if this type of quotas is turned off while we
420 * didn't have the quota inode lock.
422 xfs_iunlock(quotip, lock_mode);
427 * Find the block map; no allocations yet
429 error = xfs_bmapi_read(quotip, dqp->q_fileoffset,
430 XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0);
431 xfs_iunlock(quotip, lock_mode);
436 ASSERT(map.br_blockcount >= 1);
437 ASSERT(map.br_startblock != DELAYSTARTBLOCK);
438 if (map.br_startblock == HOLESTARTBLOCK)
441 trace_xfs_dqtobp_read(dqp);
444 * store the blkno etc so that we don't have to do the
445 * mapping all the time
447 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
449 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
450 mp->m_quotainfo->qi_dqchunklen, 0, &bp,
457 ASSERT(xfs_buf_islocked(bp));
458 xfs_buf_set_ref(bp, XFS_DQUOT_REF);
464 /* Allocate and initialize everything we need for an incore dquot. */
465 STATIC struct xfs_dquot *
467 struct xfs_mount *mp,
471 struct xfs_dquot *dqp;
473 dqp = kmem_zone_zalloc(xfs_qm_dqzone, 0);
475 dqp->dq_flags = type;
476 dqp->q_core.d_id = cpu_to_be32(id);
478 INIT_LIST_HEAD(&dqp->q_lru);
479 mutex_init(&dqp->q_qlock);
480 init_waitqueue_head(&dqp->q_pinwait);
481 dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk;
483 * Offset of dquot in the (fixed sized) dquot chunk.
485 dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) *
489 * Because we want to use a counting completion, complete
490 * the flush completion once to allow a single access to
491 * the flush completion without blocking.
493 init_completion(&dqp->q_flush);
494 complete(&dqp->q_flush);
497 * Make sure group quotas have a different lock class than user
502 /* uses the default lock class */
505 lockdep_set_class(&dqp->q_qlock, &xfs_dquot_group_class);
508 lockdep_set_class(&dqp->q_qlock, &xfs_dquot_project_class);
515 xfs_qm_dquot_logitem_init(dqp);
517 XFS_STATS_INC(mp, xs_qm_dquot);
521 /* Copy the in-core quota fields in from the on-disk buffer. */
524 struct xfs_dquot *dqp,
527 struct xfs_disk_dquot *ddqp = bp->b_addr + dqp->q_bufoffset;
529 /* copy everything from disk dquot to the incore dquot */
530 memcpy(&dqp->q_core, ddqp, sizeof(struct xfs_disk_dquot));
533 * Reservation counters are defined as reservation plus current usage
534 * to avoid having to add every time.
536 dqp->q_res_bcount = be64_to_cpu(ddqp->d_bcount);
537 dqp->q_res_icount = be64_to_cpu(ddqp->d_icount);
538 dqp->q_res_rtbcount = be64_to_cpu(ddqp->d_rtbcount);
540 /* initialize the dquot speculative prealloc thresholds */
541 xfs_dquot_set_prealloc_limits(dqp);
544 /* Allocate and initialize the dquot buffer for this in-core dquot. */
547 struct xfs_mount *mp,
548 struct xfs_dquot *dqp,
549 struct xfs_buf **bpp)
551 struct xfs_trans *tp;
554 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_qm_dqalloc,
555 XFS_QM_DQALLOC_SPACE_RES(mp), 0, 0, &tp);
559 error = xfs_dquot_disk_alloc(&tp, dqp, bpp);
563 error = xfs_trans_commit(tp);
566 * Buffer was held to the transaction, so we have to unlock it
567 * manually here because we're not passing it back.
576 xfs_trans_cancel(tp);
582 * Read in the ondisk dquot using dqtobp() then copy it to an incore version,
583 * and release the buffer immediately. If @can_alloc is true, fill any
584 * holes in the on-disk metadata.
588 struct xfs_mount *mp,
592 struct xfs_dquot **dqpp)
594 struct xfs_dquot *dqp;
598 dqp = xfs_dquot_alloc(mp, id, type);
599 trace_xfs_dqread(dqp);
601 /* Try to read the buffer, allocating if necessary. */
602 error = xfs_dquot_disk_read(mp, dqp, &bp);
603 if (error == -ENOENT && can_alloc)
604 error = xfs_qm_dqread_alloc(mp, dqp, &bp);
609 * At this point we should have a clean locked buffer. Copy the data
610 * to the incore dquot and release the buffer since the incore dquot
611 * has its own locking protocol so we needn't tie up the buffer any
614 ASSERT(xfs_buf_islocked(bp));
615 xfs_dquot_from_disk(dqp, bp);
622 trace_xfs_dqread_fail(dqp);
623 xfs_qm_dqdestroy(dqp);
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.
635 struct xfs_mount *mp,
639 struct xfs_inode *quotip = xfs_quota_inode(mp, type);
640 xfs_dqid_t next_id = *id + 1; /* simple advance */
642 struct xfs_bmbt_irec got;
643 struct xfs_iext_cursor cur;
647 /* If we'd wrap past the max ID, stop */
651 /* If new ID is within the current chunk, advancing it sufficed */
652 if (next_id % mp->m_quotainfo->qi_dqperchunk) {
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;
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);
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;
676 xfs_iunlock(quotip, lock_flags);
682 * Look up the dquot in the in-core cache. If found, the dquot is returned
683 * locked and ready to go.
685 static struct xfs_dquot *
686 xfs_qm_dqget_cache_lookup(
687 struct xfs_mount *mp,
688 struct xfs_quotainfo *qi,
689 struct radix_tree_root *tree,
692 struct xfs_dquot *dqp;
695 mutex_lock(&qi->qi_tree_lock);
696 dqp = radix_tree_lookup(tree, id);
698 mutex_unlock(&qi->qi_tree_lock);
699 XFS_STATS_INC(mp, xs_qm_dqcachemisses);
704 if (dqp->dq_flags & XFS_DQ_FREEING) {
706 mutex_unlock(&qi->qi_tree_lock);
707 trace_xfs_dqget_freeing(dqp);
713 mutex_unlock(&qi->qi_tree_lock);
715 trace_xfs_dqget_hit(dqp);
716 XFS_STATS_INC(mp, xs_qm_dqcachehits);
721 * Try to insert a new dquot into the in-core cache. If an error occurs the
722 * caller should throw away the dquot and start over. Otherwise, the dquot
723 * is returned locked (and held by the cache) as if there had been a cache
727 xfs_qm_dqget_cache_insert(
728 struct xfs_mount *mp,
729 struct xfs_quotainfo *qi,
730 struct radix_tree_root *tree,
732 struct xfs_dquot *dqp)
736 mutex_lock(&qi->qi_tree_lock);
737 error = radix_tree_insert(tree, id, dqp);
738 if (unlikely(error)) {
739 /* Duplicate found! Caller must try again. */
740 WARN_ON(error != -EEXIST);
741 mutex_unlock(&qi->qi_tree_lock);
742 trace_xfs_dqget_dup(dqp);
746 /* Return a locked dquot to the caller, with a reference taken. */
751 mutex_unlock(&qi->qi_tree_lock);
756 /* Check our input parameters. */
759 struct xfs_mount *mp,
762 if (WARN_ON_ONCE(!XFS_IS_QUOTA_RUNNING(mp)))
767 if (!XFS_IS_UQUOTA_ON(mp))
771 if (!XFS_IS_GQUOTA_ON(mp))
775 if (!XFS_IS_PQUOTA_ON(mp))
785 * Given the file system, id, and type (UDQUOT/GDQUOT), return a a locked
786 * dquot, doing an allocation (if requested) as needed.
790 struct xfs_mount *mp,
794 struct xfs_dquot **O_dqpp)
796 struct xfs_quotainfo *qi = mp->m_quotainfo;
797 struct radix_tree_root *tree = xfs_dquot_tree(qi, type);
798 struct xfs_dquot *dqp;
801 error = xfs_qm_dqget_checks(mp, type);
806 dqp = xfs_qm_dqget_cache_lookup(mp, qi, tree, id);
812 error = xfs_qm_dqread(mp, id, type, can_alloc, &dqp);
816 error = xfs_qm_dqget_cache_insert(mp, qi, tree, id, dqp);
819 * Duplicate found. Just throw away the new dquot and start
822 xfs_qm_dqdestroy(dqp);
823 XFS_STATS_INC(mp, xs_qm_dquot_dups);
827 trace_xfs_dqget_miss(dqp);
833 * Given a dquot id and type, read and initialize a dquot from the on-disk
834 * metadata. This function is only for use during quota initialization so
835 * it ignores the dquot cache assuming that the dquot shrinker isn't set up.
836 * The caller is responsible for _qm_dqdestroy'ing the returned dquot.
839 xfs_qm_dqget_uncached(
840 struct xfs_mount *mp,
843 struct xfs_dquot **dqpp)
847 error = xfs_qm_dqget_checks(mp, type);
851 return xfs_qm_dqread(mp, id, type, 0, dqpp);
854 /* Return the quota id for a given inode and type. */
856 xfs_qm_id_for_quotatype(
857 struct xfs_inode *ip,
862 return i_uid_read(VFS_I(ip));
864 return i_gid_read(VFS_I(ip));
866 return ip->i_d.di_projid;
873 * Return the dquot for a given inode and type. If @can_alloc is true, then
874 * allocate blocks if needed. The inode's ILOCK must be held and it must not
875 * have already had an inode attached.
879 struct xfs_inode *ip,
882 struct xfs_dquot **O_dqpp)
884 struct xfs_mount *mp = ip->i_mount;
885 struct xfs_quotainfo *qi = mp->m_quotainfo;
886 struct radix_tree_root *tree = xfs_dquot_tree(qi, type);
887 struct xfs_dquot *dqp;
891 error = xfs_qm_dqget_checks(mp, type);
895 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
896 ASSERT(xfs_inode_dquot(ip, type) == NULL);
898 id = xfs_qm_id_for_quotatype(ip, type);
901 dqp = xfs_qm_dqget_cache_lookup(mp, qi, tree, id);
908 * Dquot cache miss. We don't want to keep the inode lock across
909 * a (potential) disk read. Also we don't want to deal with the lock
910 * ordering between quotainode and this inode. OTOH, dropping the inode
911 * lock here means dealing with a chown that can happen before
912 * we re-acquire the lock.
914 xfs_iunlock(ip, XFS_ILOCK_EXCL);
915 error = xfs_qm_dqread(mp, id, type, can_alloc, &dqp);
916 xfs_ilock(ip, XFS_ILOCK_EXCL);
921 * A dquot could be attached to this inode by now, since we had
924 if (xfs_this_quota_on(mp, type)) {
925 struct xfs_dquot *dqp1;
927 dqp1 = xfs_inode_dquot(ip, type);
929 xfs_qm_dqdestroy(dqp);
935 /* inode stays locked on return */
936 xfs_qm_dqdestroy(dqp);
940 error = xfs_qm_dqget_cache_insert(mp, qi, tree, id, dqp);
943 * Duplicate found. Just throw away the new dquot and start
946 xfs_qm_dqdestroy(dqp);
947 XFS_STATS_INC(mp, xs_qm_dquot_dups);
952 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
953 trace_xfs_dqget_miss(dqp);
959 * Starting at @id and progressing upwards, look for an initialized incore
960 * dquot, lock it, and return it.
964 struct xfs_mount *mp,
967 struct xfs_dquot **dqpp)
969 struct xfs_dquot *dqp;
973 for (; !error; error = xfs_dq_get_next_id(mp, type, &id)) {
974 error = xfs_qm_dqget(mp, id, type, false, &dqp);
975 if (error == -ENOENT)
980 if (!XFS_IS_DQUOT_UNINITIALIZED(dqp)) {
992 * Release a reference to the dquot (decrement ref-count) and unlock it.
994 * If there is a group quota attached to this dquot, carefully release that
995 * too without tripping over deadlocks'n'stuff.
999 struct xfs_dquot *dqp)
1001 ASSERT(dqp->q_nrefs > 0);
1002 ASSERT(XFS_DQ_IS_LOCKED(dqp));
1004 trace_xfs_dqput(dqp);
1006 if (--dqp->q_nrefs == 0) {
1007 struct xfs_quotainfo *qi = dqp->q_mount->m_quotainfo;
1008 trace_xfs_dqput_free(dqp);
1010 if (list_lru_add(&qi->qi_lru, &dqp->q_lru))
1011 XFS_STATS_INC(dqp->q_mount, xs_qm_dquot_unused);
1017 * Release a dquot. Flush it if dirty, then dqput() it.
1018 * dquot must not be locked.
1022 struct xfs_dquot *dqp)
1027 trace_xfs_dqrele(dqp);
1031 * We don't care to flush it if the dquot is dirty here.
1032 * That will create stutters that we want to avoid.
1033 * Instead we do a delayed write when we try to reclaim
1034 * a dirty dquot. Also xfs_sync will take part of the burden...
1040 * This is the dquot flushing I/O completion routine. It is called
1041 * from interrupt level when the buffer containing the dquot is
1042 * flushed to disk. It is responsible for removing the dquot logitem
1043 * from the AIL if it has not been re-logged, and unlocking the dquot's
1044 * flush lock. This behavior is very similar to that of inodes..
1047 xfs_qm_dqflush_done(
1049 struct xfs_log_item *lip)
1051 struct xfs_dq_logitem *qip = (struct xfs_dq_logitem *)lip;
1052 struct xfs_dquot *dqp = qip->qli_dquot;
1053 struct xfs_ail *ailp = lip->li_ailp;
1056 * We only want to pull the item from the AIL if its
1057 * location in the log has not changed since we started the flush.
1058 * Thus, we only bother if the dquot's lsn has
1059 * not changed. First we check the lsn outside the lock
1060 * since it's cheaper, and then we recheck while
1061 * holding the lock before removing the dquot from the AIL.
1063 if (test_bit(XFS_LI_IN_AIL, &lip->li_flags) &&
1064 ((lip->li_lsn == qip->qli_flush_lsn) ||
1065 test_bit(XFS_LI_FAILED, &lip->li_flags))) {
1067 /* xfs_trans_ail_delete() drops the AIL lock. */
1068 spin_lock(&ailp->ail_lock);
1069 if (lip->li_lsn == qip->qli_flush_lsn) {
1070 xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE);
1073 * Clear the failed state since we are about to drop the
1076 xfs_clear_li_failed(lip);
1077 spin_unlock(&ailp->ail_lock);
1082 * Release the dq's flush lock since we're done with it.
1088 * Write a modified dquot to disk.
1089 * The dquot must be locked and the flush lock too taken by caller.
1090 * The flush lock will not be unlocked until the dquot reaches the disk,
1091 * but the dquot is free to be unlocked and modified by the caller
1092 * in the interim. Dquot is still locked on return. This behavior is
1093 * identical to that of inodes.
1097 struct xfs_dquot *dqp,
1098 struct xfs_buf **bpp)
1100 struct xfs_mount *mp = dqp->q_mount;
1102 struct xfs_dqblk *dqb;
1103 struct xfs_disk_dquot *ddqp;
1107 ASSERT(XFS_DQ_IS_LOCKED(dqp));
1108 ASSERT(!completion_done(&dqp->q_flush));
1110 trace_xfs_dqflush(dqp);
1114 xfs_qm_dqunpin_wait(dqp);
1117 * This may have been unpinned because the filesystem is shutting
1118 * down forcibly. If that's the case we must not write this dquot
1119 * to disk, because the log record didn't make it to disk.
1121 * We also have to remove the log item from the AIL in this case,
1122 * as we wait for an emptry AIL as part of the unmount process.
1124 if (XFS_FORCED_SHUTDOWN(mp)) {
1125 struct xfs_log_item *lip = &dqp->q_logitem.qli_item;
1126 dqp->dq_flags &= ~XFS_DQ_DIRTY;
1128 xfs_trans_ail_remove(lip, SHUTDOWN_CORRUPT_INCORE);
1135 * Get the buffer containing the on-disk dquot
1137 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
1138 mp->m_quotainfo->qi_dqchunklen, XBF_TRYLOCK,
1139 &bp, &xfs_dquot_buf_ops);
1144 * Calculate the location of the dquot inside the buffer.
1146 dqb = bp->b_addr + dqp->q_bufoffset;
1147 ddqp = &dqb->dd_diskdq;
1149 /* sanity check the in-core structure before we flush */
1150 fa = xfs_dquot_verify(mp, &dqp->q_core, be32_to_cpu(dqp->q_core.d_id),
1153 xfs_alert(mp, "corrupt dquot ID 0x%x in memory at %pS",
1154 be32_to_cpu(dqp->q_core.d_id), fa);
1157 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1158 return -EFSCORRUPTED;
1161 /* This is the only portion of data that needs to persist */
1162 memcpy(ddqp, &dqp->q_core, sizeof(struct xfs_disk_dquot));
1165 * Clear the dirty field and remember the flush lsn for later use.
1167 dqp->dq_flags &= ~XFS_DQ_DIRTY;
1169 xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn,
1170 &dqp->q_logitem.qli_item.li_lsn);
1173 * copy the lsn into the on-disk dquot now while we have the in memory
1174 * dquot here. This can't be done later in the write verifier as we
1175 * can't get access to the log item at that point in time.
1177 * We also calculate the CRC here so that the on-disk dquot in the
1178 * buffer always has a valid CRC. This ensures there is no possibility
1179 * of a dquot without an up-to-date CRC getting to disk.
1181 if (xfs_sb_version_hascrc(&mp->m_sb)) {
1182 dqb->dd_lsn = cpu_to_be64(dqp->q_logitem.qli_item.li_lsn);
1183 xfs_update_cksum((char *)dqb, sizeof(struct xfs_dqblk),
1188 * Attach an iodone routine so that we can remove this dquot from the
1189 * AIL and release the flush lock once the dquot is synced to disk.
1191 xfs_buf_attach_iodone(bp, xfs_qm_dqflush_done,
1192 &dqp->q_logitem.qli_item);
1195 * If the buffer is pinned then push on the log so we won't
1196 * get stuck waiting in the write for too long.
1198 if (xfs_buf_ispinned(bp)) {
1199 trace_xfs_dqflush_force(dqp);
1200 xfs_log_force(mp, 0);
1203 trace_xfs_dqflush_done(dqp);
1213 * Lock two xfs_dquot structures.
1215 * To avoid deadlocks we always lock the quota structure with
1216 * the lowerd id first.
1220 struct xfs_dquot *d1,
1221 struct xfs_dquot *d2)
1225 if (be32_to_cpu(d1->q_core.d_id) >
1226 be32_to_cpu(d2->q_core.d_id)) {
1227 mutex_lock(&d2->q_qlock);
1228 mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED);
1230 mutex_lock(&d1->q_qlock);
1231 mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED);
1234 mutex_lock(&d1->q_qlock);
1236 mutex_lock(&d2->q_qlock);
1244 kmem_zone_init(sizeof(struct xfs_dquot), "xfs_dquot");
1249 kmem_zone_init(sizeof(struct xfs_dquot_acct), "xfs_dqtrx");
1250 if (!xfs_qm_dqtrxzone)
1251 goto out_free_dqzone;
1256 kmem_zone_destroy(xfs_qm_dqzone);
1264 kmem_zone_destroy(xfs_qm_dqtrxzone);
1265 kmem_zone_destroy(xfs_qm_dqzone);
1269 * Iterate every dquot of a particular type. The caller must ensure that the
1270 * particular quota type is active. iter_fn can return negative error codes,
1271 * or -ECANCELED to indicate that it wants to stop iterating.
1275 struct xfs_mount *mp,
1277 xfs_qm_dqiterate_fn iter_fn,
1280 struct xfs_dquot *dq;
1285 error = xfs_qm_dqget_next(mp, id, dqtype, &dq);
1286 if (error == -ENOENT)
1291 error = iter_fn(dq, dqtype, priv);
1292 id = be32_to_cpu(dq->q_core.d_id);
1295 } while (error == 0 && id != 0);