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_cache *xfs_dqtrx_cache;
42 static struct kmem_cache *xfs_dquot_cache;
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_dquot_cache, 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);
144 * Check the limits and timers of a dquot and start or reset timers
146 * This gets called even when quota enforcement is OFF, which makes our
147 * life a little less complicated. (We just don't reject any quota
148 * reservations in that case, when enforcement is off).
149 * We also return 0 as the values of the timers in Q_GETQUOTA calls, when
151 * In contrast, warnings are a little different in that they don't
152 * 'automatically' get started when limits get exceeded. They do
153 * get reset to zero, however, when we find the count to be under
154 * the soft limit (they are only ever set non-zero via userspace).
157 xfs_qm_adjust_dqtimers(
158 struct xfs_dquot *dq)
160 struct xfs_mount *mp = dq->q_mount;
161 struct xfs_quotainfo *qi = mp->m_quotainfo;
162 struct xfs_def_quota *defq;
165 defq = xfs_get_defquota(qi, xfs_dquot_type(dq));
167 xfs_qm_adjust_res_timer(dq->q_mount, &dq->q_blk, &defq->blk);
168 xfs_qm_adjust_res_timer(dq->q_mount, &dq->q_ino, &defq->ino);
169 xfs_qm_adjust_res_timer(dq->q_mount, &dq->q_rtb, &defq->rtb);
173 * initialize a buffer full of dquots and log the whole thing
176 xfs_qm_init_dquot_blk(
177 struct xfs_trans *tp,
178 struct xfs_mount *mp,
183 struct xfs_quotainfo *q = mp->m_quotainfo;
187 unsigned int blftype;
191 ASSERT(xfs_buf_islocked(bp));
194 case XFS_DQTYPE_USER:
195 qflag = XFS_UQUOTA_CHKD;
196 blftype = XFS_BLF_UDQUOT_BUF;
198 case XFS_DQTYPE_PROJ:
199 qflag = XFS_PQUOTA_CHKD;
200 blftype = XFS_BLF_PDQUOT_BUF;
202 case XFS_DQTYPE_GROUP:
203 qflag = XFS_GQUOTA_CHKD;
204 blftype = XFS_BLF_GDQUOT_BUF;
214 * ID of the first dquot in the block - id's are zero based.
216 curid = id - (id % q->qi_dqperchunk);
217 memset(d, 0, BBTOB(q->qi_dqchunklen));
218 for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++) {
219 d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
220 d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
221 d->dd_diskdq.d_id = cpu_to_be32(curid);
222 d->dd_diskdq.d_type = type;
223 if (curid > 0 && xfs_has_bigtime(mp))
224 d->dd_diskdq.d_type |= XFS_DQTYPE_BIGTIME;
225 if (xfs_has_crc(mp)) {
226 uuid_copy(&d->dd_uuid, &mp->m_sb.sb_meta_uuid);
227 xfs_update_cksum((char *)d, sizeof(struct xfs_dqblk),
232 xfs_trans_dquot_buf(tp, bp, blftype);
235 * quotacheck uses delayed writes to update all the dquots on disk in an
236 * efficient manner instead of logging the individual dquot changes as
237 * they are made. However if we log the buffer allocated here and crash
238 * after quotacheck while the logged initialisation is still in the
239 * active region of the log, log recovery can replay the dquot buffer
240 * initialisation over the top of the checked dquots and corrupt quota
243 * To avoid this problem, quotacheck cannot log the initialised buffer.
244 * We must still dirty the buffer and write it back before the
245 * allocation transaction clears the log. Therefore, mark the buffer as
246 * ordered instead of logging it directly. This is safe for quotacheck
247 * because it detects and repairs allocated but initialized dquot blocks
248 * in the quota inodes.
250 if (!(mp->m_qflags & qflag))
251 xfs_trans_ordered_buf(tp, bp);
253 xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1);
257 * Initialize the dynamic speculative preallocation thresholds. The lo/hi
258 * watermarks correspond to the soft and hard limits by default. If a soft limit
259 * is not specified, we use 95% of the hard limit.
262 xfs_dquot_set_prealloc_limits(struct xfs_dquot *dqp)
266 dqp->q_prealloc_hi_wmark = dqp->q_blk.hardlimit;
267 dqp->q_prealloc_lo_wmark = dqp->q_blk.softlimit;
268 if (!dqp->q_prealloc_lo_wmark) {
269 dqp->q_prealloc_lo_wmark = dqp->q_prealloc_hi_wmark;
270 do_div(dqp->q_prealloc_lo_wmark, 100);
271 dqp->q_prealloc_lo_wmark *= 95;
274 space = dqp->q_prealloc_hi_wmark;
277 dqp->q_low_space[XFS_QLOWSP_1_PCNT] = space;
278 dqp->q_low_space[XFS_QLOWSP_3_PCNT] = space * 3;
279 dqp->q_low_space[XFS_QLOWSP_5_PCNT] = space * 5;
283 * Ensure that the given in-core dquot has a buffer on disk backing it, and
284 * return the buffer locked and held. This is called when the bmapi finds a
288 xfs_dquot_disk_alloc(
289 struct xfs_dquot *dqp,
290 struct xfs_buf **bpp)
292 struct xfs_bmbt_irec map;
293 struct xfs_trans *tp;
294 struct xfs_mount *mp = dqp->q_mount;
296 xfs_dqtype_t qtype = xfs_dquot_type(dqp);
297 struct xfs_inode *quotip = xfs_quota_inode(mp, qtype);
301 trace_xfs_dqalloc(dqp);
303 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_qm_dqalloc,
304 XFS_QM_DQALLOC_SPACE_RES(mp), 0, 0, &tp);
308 xfs_ilock(quotip, XFS_ILOCK_EXCL);
309 xfs_trans_ijoin(tp, quotip, 0);
311 if (!xfs_this_quota_on(dqp->q_mount, qtype)) {
313 * Return if this type of quotas is turned off while we didn't
320 error = xfs_iext_count_may_overflow(quotip, XFS_DATA_FORK,
321 XFS_IEXT_ADD_NOSPLIT_CNT);
323 error = xfs_iext_count_upgrade(tp, quotip,
324 XFS_IEXT_ADD_NOSPLIT_CNT);
328 /* Create the block mapping. */
329 error = xfs_bmapi_write(tp, quotip, dqp->q_fileoffset,
330 XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA, 0, &map,
335 ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB);
337 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
338 (map.br_startblock != HOLESTARTBLOCK));
341 * Keep track of the blkno to save a lookup later
343 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
345 /* now we can just get the buffer (there's nothing to read yet) */
346 error = xfs_trans_get_buf(tp, mp->m_ddev_targp, dqp->q_blkno,
347 mp->m_quotainfo->qi_dqchunklen, 0, &bp);
350 bp->b_ops = &xfs_dquot_buf_ops;
353 * Make a chunk of dquots out of this buffer and log
356 xfs_qm_init_dquot_blk(tp, mp, dqp->q_id, qtype, bp);
357 xfs_buf_set_ref(bp, XFS_DQUOT_REF);
360 * Hold the buffer and join it to the dfops so that we'll still own
361 * the buffer when we return to the caller. The buffer disposal on
362 * error must be paid attention to very carefully, as it has been
363 * broken since commit efa092f3d4c6 "[XFS] Fixes a bug in the quota
364 * code when allocating a new dquot record" in 2005, and the later
365 * conversion to xfs_defer_ops in commit 310a75a3c6c747 failed to keep
366 * the buffer locked across the _defer_finish call. We can now do
367 * this correctly with xfs_defer_bjoin.
369 * Above, we allocated a disk block for the dquot information and used
370 * get_buf to initialize the dquot. If the _defer_finish fails, the old
371 * transaction is gone but the new buffer is not joined or held to any
372 * transaction, so we must _buf_relse it.
374 * If everything succeeds, the caller of this function is returned a
375 * buffer that is locked and held to the transaction. The caller
376 * is responsible for unlocking any buffer passed back, either
377 * manually or by committing the transaction. On error, the buffer is
378 * released and not passed back.
380 * Keep the quota inode ILOCKed until after the transaction commit to
381 * maintain the atomicity of bmap/rmap updates.
383 xfs_trans_bhold(tp, bp);
384 error = xfs_trans_commit(tp);
385 xfs_iunlock(quotip, XFS_ILOCK_EXCL);
395 xfs_trans_cancel(tp);
396 xfs_iunlock(quotip, XFS_ILOCK_EXCL);
401 * Read in the in-core dquot's on-disk metadata and return the buffer.
402 * Returns ENOENT to signal a hole.
406 struct xfs_mount *mp,
407 struct xfs_dquot *dqp,
408 struct xfs_buf **bpp)
410 struct xfs_bmbt_irec map;
412 xfs_dqtype_t qtype = xfs_dquot_type(dqp);
413 struct xfs_inode *quotip = xfs_quota_inode(mp, qtype);
418 lock_mode = xfs_ilock_data_map_shared(quotip);
419 if (!xfs_this_quota_on(mp, qtype)) {
421 * Return if this type of quotas is turned off while we
422 * didn't have the quota inode lock.
424 xfs_iunlock(quotip, lock_mode);
429 * Find the block map; no allocations yet
431 error = xfs_bmapi_read(quotip, dqp->q_fileoffset,
432 XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0);
433 xfs_iunlock(quotip, lock_mode);
438 ASSERT(map.br_blockcount >= 1);
439 ASSERT(map.br_startblock != DELAYSTARTBLOCK);
440 if (map.br_startblock == HOLESTARTBLOCK)
443 trace_xfs_dqtobp_read(dqp);
446 * store the blkno etc so that we don't have to do the
447 * mapping all the time
449 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
451 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
452 mp->m_quotainfo->qi_dqchunklen, 0, &bp,
459 ASSERT(xfs_buf_islocked(bp));
460 xfs_buf_set_ref(bp, XFS_DQUOT_REF);
466 /* Allocate and initialize everything we need for an incore dquot. */
467 STATIC struct xfs_dquot *
469 struct xfs_mount *mp,
473 struct xfs_dquot *dqp;
475 dqp = kmem_cache_zalloc(xfs_dquot_cache, GFP_KERNEL | __GFP_NOFAIL);
480 INIT_LIST_HEAD(&dqp->q_lru);
481 mutex_init(&dqp->q_qlock);
482 init_waitqueue_head(&dqp->q_pinwait);
483 dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk;
485 * Offset of dquot in the (fixed sized) dquot chunk.
487 dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) *
488 sizeof(struct xfs_dqblk);
491 * Because we want to use a counting completion, complete
492 * the flush completion once to allow a single access to
493 * the flush completion without blocking.
495 init_completion(&dqp->q_flush);
496 complete(&dqp->q_flush);
499 * Make sure group quotas have a different lock class than user
503 case XFS_DQTYPE_USER:
504 /* uses the default lock class */
506 case XFS_DQTYPE_GROUP:
507 lockdep_set_class(&dqp->q_qlock, &xfs_dquot_group_class);
509 case XFS_DQTYPE_PROJ:
510 lockdep_set_class(&dqp->q_qlock, &xfs_dquot_project_class);
517 xfs_qm_dquot_logitem_init(dqp);
519 XFS_STATS_INC(mp, xs_qm_dquot);
523 /* Check the ondisk dquot's id and type match what the incore dquot expects. */
525 xfs_dquot_check_type(
526 struct xfs_dquot *dqp,
527 struct xfs_disk_dquot *ddqp)
532 ddqp_type = ddqp->d_type & XFS_DQTYPE_REC_MASK;
533 dqp_type = xfs_dquot_type(dqp);
535 if (be32_to_cpu(ddqp->d_id) != dqp->q_id)
539 * V5 filesystems always expect an exact type match. V4 filesystems
540 * expect an exact match for user dquots and for non-root group and
543 if (xfs_has_crc(dqp->q_mount) ||
544 dqp_type == XFS_DQTYPE_USER || dqp->q_id != 0)
545 return ddqp_type == dqp_type;
548 * V4 filesystems support either group or project quotas, but not both
549 * at the same time. The non-user quota file can be switched between
550 * group and project quota uses depending on the mount options, which
551 * means that we can encounter the other type when we try to load quota
552 * defaults. Quotacheck will soon reset the entire quota file
553 * (including the root dquot) anyway, but don't log scary corruption
556 return ddqp_type == XFS_DQTYPE_GROUP || ddqp_type == XFS_DQTYPE_PROJ;
559 /* Copy the in-core quota fields in from the on-disk buffer. */
562 struct xfs_dquot *dqp,
565 struct xfs_disk_dquot *ddqp = bp->b_addr + dqp->q_bufoffset;
568 * Ensure that we got the type and ID we were looking for.
569 * Everything else was checked by the dquot buffer verifier.
571 if (!xfs_dquot_check_type(dqp, ddqp)) {
572 xfs_alert_tag(bp->b_mount, XFS_PTAG_VERIFIER_ERROR,
573 "Metadata corruption detected at %pS, quota %u",
574 __this_address, dqp->q_id);
575 xfs_alert(bp->b_mount, "Unmount and run xfs_repair");
576 return -EFSCORRUPTED;
579 /* copy everything from disk dquot to the incore dquot */
580 dqp->q_type = ddqp->d_type;
581 dqp->q_blk.hardlimit = be64_to_cpu(ddqp->d_blk_hardlimit);
582 dqp->q_blk.softlimit = be64_to_cpu(ddqp->d_blk_softlimit);
583 dqp->q_ino.hardlimit = be64_to_cpu(ddqp->d_ino_hardlimit);
584 dqp->q_ino.softlimit = be64_to_cpu(ddqp->d_ino_softlimit);
585 dqp->q_rtb.hardlimit = be64_to_cpu(ddqp->d_rtb_hardlimit);
586 dqp->q_rtb.softlimit = be64_to_cpu(ddqp->d_rtb_softlimit);
588 dqp->q_blk.count = be64_to_cpu(ddqp->d_bcount);
589 dqp->q_ino.count = be64_to_cpu(ddqp->d_icount);
590 dqp->q_rtb.count = be64_to_cpu(ddqp->d_rtbcount);
592 dqp->q_blk.timer = xfs_dquot_from_disk_ts(ddqp, ddqp->d_btimer);
593 dqp->q_ino.timer = xfs_dquot_from_disk_ts(ddqp, ddqp->d_itimer);
594 dqp->q_rtb.timer = xfs_dquot_from_disk_ts(ddqp, ddqp->d_rtbtimer);
597 * Reservation counters are defined as reservation plus current usage
598 * to avoid having to add every time.
600 dqp->q_blk.reserved = dqp->q_blk.count;
601 dqp->q_ino.reserved = dqp->q_ino.count;
602 dqp->q_rtb.reserved = dqp->q_rtb.count;
604 /* initialize the dquot speculative prealloc thresholds */
605 xfs_dquot_set_prealloc_limits(dqp);
609 /* Copy the in-core quota fields into the on-disk buffer. */
612 struct xfs_disk_dquot *ddqp,
613 struct xfs_dquot *dqp)
615 ddqp->d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
616 ddqp->d_version = XFS_DQUOT_VERSION;
617 ddqp->d_type = dqp->q_type;
618 ddqp->d_id = cpu_to_be32(dqp->q_id);
622 ddqp->d_blk_hardlimit = cpu_to_be64(dqp->q_blk.hardlimit);
623 ddqp->d_blk_softlimit = cpu_to_be64(dqp->q_blk.softlimit);
624 ddqp->d_ino_hardlimit = cpu_to_be64(dqp->q_ino.hardlimit);
625 ddqp->d_ino_softlimit = cpu_to_be64(dqp->q_ino.softlimit);
626 ddqp->d_rtb_hardlimit = cpu_to_be64(dqp->q_rtb.hardlimit);
627 ddqp->d_rtb_softlimit = cpu_to_be64(dqp->q_rtb.softlimit);
629 ddqp->d_bcount = cpu_to_be64(dqp->q_blk.count);
630 ddqp->d_icount = cpu_to_be64(dqp->q_ino.count);
631 ddqp->d_rtbcount = cpu_to_be64(dqp->q_rtb.count);
635 ddqp->d_rtbwarns = 0;
637 ddqp->d_btimer = xfs_dquot_to_disk_ts(dqp, dqp->q_blk.timer);
638 ddqp->d_itimer = xfs_dquot_to_disk_ts(dqp, dqp->q_ino.timer);
639 ddqp->d_rtbtimer = xfs_dquot_to_disk_ts(dqp, dqp->q_rtb.timer);
643 * Read in the ondisk dquot using dqtobp() then copy it to an incore version,
644 * and release the buffer immediately. If @can_alloc is true, fill any
645 * holes in the on-disk metadata.
649 struct xfs_mount *mp,
653 struct xfs_dquot **dqpp)
655 struct xfs_dquot *dqp;
659 dqp = xfs_dquot_alloc(mp, id, type);
660 trace_xfs_dqread(dqp);
662 /* Try to read the buffer, allocating if necessary. */
663 error = xfs_dquot_disk_read(mp, dqp, &bp);
664 if (error == -ENOENT && can_alloc)
665 error = xfs_dquot_disk_alloc(dqp, &bp);
670 * At this point we should have a clean locked buffer. Copy the data
671 * to the incore dquot and release the buffer since the incore dquot
672 * has its own locking protocol so we needn't tie up the buffer any
675 ASSERT(xfs_buf_islocked(bp));
676 error = xfs_dquot_from_disk(dqp, bp);
685 trace_xfs_dqread_fail(dqp);
686 xfs_qm_dqdestroy(dqp);
692 * Advance to the next id in the current chunk, or if at the
693 * end of the chunk, skip ahead to first id in next allocated chunk
694 * using the SEEK_DATA interface.
698 struct xfs_mount *mp,
702 struct xfs_inode *quotip = xfs_quota_inode(mp, type);
703 xfs_dqid_t next_id = *id + 1; /* simple advance */
705 struct xfs_bmbt_irec got;
706 struct xfs_iext_cursor cur;
710 /* If we'd wrap past the max ID, stop */
714 /* If new ID is within the current chunk, advancing it sufficed */
715 if (next_id % mp->m_quotainfo->qi_dqperchunk) {
720 /* Nope, next_id is now past the current chunk, so find the next one */
721 start = (xfs_fsblock_t)next_id / mp->m_quotainfo->qi_dqperchunk;
723 lock_flags = xfs_ilock_data_map_shared(quotip);
724 error = xfs_iread_extents(NULL, quotip, XFS_DATA_FORK);
728 if (xfs_iext_lookup_extent(quotip, "ip->i_df, start, &cur, &got)) {
729 /* contiguous chunk, bump startoff for the id calculation */
730 if (got.br_startoff < start)
731 got.br_startoff = start;
732 *id = got.br_startoff * mp->m_quotainfo->qi_dqperchunk;
737 xfs_iunlock(quotip, lock_flags);
743 * Look up the dquot in the in-core cache. If found, the dquot is returned
744 * locked and ready to go.
746 static struct xfs_dquot *
747 xfs_qm_dqget_cache_lookup(
748 struct xfs_mount *mp,
749 struct xfs_quotainfo *qi,
750 struct radix_tree_root *tree,
753 struct xfs_dquot *dqp;
756 mutex_lock(&qi->qi_tree_lock);
757 dqp = radix_tree_lookup(tree, id);
759 mutex_unlock(&qi->qi_tree_lock);
760 XFS_STATS_INC(mp, xs_qm_dqcachemisses);
765 if (dqp->q_flags & XFS_DQFLAG_FREEING) {
767 mutex_unlock(&qi->qi_tree_lock);
768 trace_xfs_dqget_freeing(dqp);
774 mutex_unlock(&qi->qi_tree_lock);
776 trace_xfs_dqget_hit(dqp);
777 XFS_STATS_INC(mp, xs_qm_dqcachehits);
782 * Try to insert a new dquot into the in-core cache. If an error occurs the
783 * caller should throw away the dquot and start over. Otherwise, the dquot
784 * is returned locked (and held by the cache) as if there had been a cache
788 xfs_qm_dqget_cache_insert(
789 struct xfs_mount *mp,
790 struct xfs_quotainfo *qi,
791 struct radix_tree_root *tree,
793 struct xfs_dquot *dqp)
797 mutex_lock(&qi->qi_tree_lock);
798 error = radix_tree_insert(tree, id, dqp);
799 if (unlikely(error)) {
800 /* Duplicate found! Caller must try again. */
801 WARN_ON(error != -EEXIST);
802 mutex_unlock(&qi->qi_tree_lock);
803 trace_xfs_dqget_dup(dqp);
807 /* Return a locked dquot to the caller, with a reference taken. */
812 mutex_unlock(&qi->qi_tree_lock);
817 /* Check our input parameters. */
820 struct xfs_mount *mp,
824 case XFS_DQTYPE_USER:
825 if (!XFS_IS_UQUOTA_ON(mp))
828 case XFS_DQTYPE_GROUP:
829 if (!XFS_IS_GQUOTA_ON(mp))
832 case XFS_DQTYPE_PROJ:
833 if (!XFS_IS_PQUOTA_ON(mp))
843 * Given the file system, id, and type (UDQUOT/GDQUOT/PDQUOT), return a
844 * locked dquot, doing an allocation (if requested) as needed.
848 struct xfs_mount *mp,
852 struct xfs_dquot **O_dqpp)
854 struct xfs_quotainfo *qi = mp->m_quotainfo;
855 struct radix_tree_root *tree = xfs_dquot_tree(qi, type);
856 struct xfs_dquot *dqp;
859 error = xfs_qm_dqget_checks(mp, type);
864 dqp = xfs_qm_dqget_cache_lookup(mp, qi, tree, id);
870 error = xfs_qm_dqread(mp, id, type, can_alloc, &dqp);
874 error = xfs_qm_dqget_cache_insert(mp, qi, tree, id, dqp);
877 * Duplicate found. Just throw away the new dquot and start
880 xfs_qm_dqdestroy(dqp);
881 XFS_STATS_INC(mp, xs_qm_dquot_dups);
885 trace_xfs_dqget_miss(dqp);
891 * Given a dquot id and type, read and initialize a dquot from the on-disk
892 * metadata. This function is only for use during quota initialization so
893 * it ignores the dquot cache assuming that the dquot shrinker isn't set up.
894 * The caller is responsible for _qm_dqdestroy'ing the returned dquot.
897 xfs_qm_dqget_uncached(
898 struct xfs_mount *mp,
901 struct xfs_dquot **dqpp)
905 error = xfs_qm_dqget_checks(mp, type);
909 return xfs_qm_dqread(mp, id, type, 0, dqpp);
912 /* Return the quota id for a given inode and type. */
914 xfs_qm_id_for_quotatype(
915 struct xfs_inode *ip,
919 case XFS_DQTYPE_USER:
920 return i_uid_read(VFS_I(ip));
921 case XFS_DQTYPE_GROUP:
922 return i_gid_read(VFS_I(ip));
923 case XFS_DQTYPE_PROJ:
931 * Return the dquot for a given inode and type. If @can_alloc is true, then
932 * allocate blocks if needed. The inode's ILOCK must be held and it must not
933 * have already had an inode attached.
937 struct xfs_inode *ip,
940 struct xfs_dquot **O_dqpp)
942 struct xfs_mount *mp = ip->i_mount;
943 struct xfs_quotainfo *qi = mp->m_quotainfo;
944 struct radix_tree_root *tree = xfs_dquot_tree(qi, type);
945 struct xfs_dquot *dqp;
949 error = xfs_qm_dqget_checks(mp, type);
953 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
954 ASSERT(xfs_inode_dquot(ip, type) == NULL);
956 id = xfs_qm_id_for_quotatype(ip, type);
959 dqp = xfs_qm_dqget_cache_lookup(mp, qi, tree, id);
966 * Dquot cache miss. We don't want to keep the inode lock across
967 * a (potential) disk read. Also we don't want to deal with the lock
968 * ordering between quotainode and this inode. OTOH, dropping the inode
969 * lock here means dealing with a chown that can happen before
970 * we re-acquire the lock.
972 xfs_iunlock(ip, XFS_ILOCK_EXCL);
973 error = xfs_qm_dqread(mp, id, type, can_alloc, &dqp);
974 xfs_ilock(ip, XFS_ILOCK_EXCL);
979 * A dquot could be attached to this inode by now, since we had
982 if (xfs_this_quota_on(mp, type)) {
983 struct xfs_dquot *dqp1;
985 dqp1 = xfs_inode_dquot(ip, type);
987 xfs_qm_dqdestroy(dqp);
993 /* inode stays locked on return */
994 xfs_qm_dqdestroy(dqp);
998 error = xfs_qm_dqget_cache_insert(mp, qi, tree, id, dqp);
1001 * Duplicate found. Just throw away the new dquot and start
1004 xfs_qm_dqdestroy(dqp);
1005 XFS_STATS_INC(mp, xs_qm_dquot_dups);
1010 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1011 trace_xfs_dqget_miss(dqp);
1017 * Starting at @id and progressing upwards, look for an initialized incore
1018 * dquot, lock it, and return it.
1022 struct xfs_mount *mp,
1025 struct xfs_dquot **dqpp)
1027 struct xfs_dquot *dqp;
1031 for (; !error; error = xfs_dq_get_next_id(mp, type, &id)) {
1032 error = xfs_qm_dqget(mp, id, type, false, &dqp);
1033 if (error == -ENOENT)
1035 else if (error != 0)
1038 if (!XFS_IS_DQUOT_UNINITIALIZED(dqp)) {
1050 * Release a reference to the dquot (decrement ref-count) and unlock it.
1052 * If there is a group quota attached to this dquot, carefully release that
1053 * too without tripping over deadlocks'n'stuff.
1057 struct xfs_dquot *dqp)
1059 ASSERT(dqp->q_nrefs > 0);
1060 ASSERT(XFS_DQ_IS_LOCKED(dqp));
1062 trace_xfs_dqput(dqp);
1064 if (--dqp->q_nrefs == 0) {
1065 struct xfs_quotainfo *qi = dqp->q_mount->m_quotainfo;
1066 trace_xfs_dqput_free(dqp);
1068 if (list_lru_add(&qi->qi_lru, &dqp->q_lru))
1069 XFS_STATS_INC(dqp->q_mount, xs_qm_dquot_unused);
1075 * Release a dquot. Flush it if dirty, then dqput() it.
1076 * dquot must not be locked.
1080 struct xfs_dquot *dqp)
1085 trace_xfs_dqrele(dqp);
1089 * We don't care to flush it if the dquot is dirty here.
1090 * That will create stutters that we want to avoid.
1091 * Instead we do a delayed write when we try to reclaim
1092 * a dirty dquot. Also xfs_sync will take part of the burden...
1098 * This is the dquot flushing I/O completion routine. It is called
1099 * from interrupt level when the buffer containing the dquot is
1100 * flushed to disk. It is responsible for removing the dquot logitem
1101 * from the AIL if it has not been re-logged, and unlocking the dquot's
1102 * flush lock. This behavior is very similar to that of inodes..
1105 xfs_qm_dqflush_done(
1106 struct xfs_log_item *lip)
1108 struct xfs_dq_logitem *qip = (struct xfs_dq_logitem *)lip;
1109 struct xfs_dquot *dqp = qip->qli_dquot;
1110 struct xfs_ail *ailp = lip->li_ailp;
1114 * We only want to pull the item from the AIL if its
1115 * location in the log has not changed since we started the flush.
1116 * Thus, we only bother if the dquot's lsn has
1117 * not changed. First we check the lsn outside the lock
1118 * since it's cheaper, and then we recheck while
1119 * holding the lock before removing the dquot from the AIL.
1121 if (test_bit(XFS_LI_IN_AIL, &lip->li_flags) &&
1122 ((lip->li_lsn == qip->qli_flush_lsn) ||
1123 test_bit(XFS_LI_FAILED, &lip->li_flags))) {
1125 spin_lock(&ailp->ail_lock);
1126 xfs_clear_li_failed(lip);
1127 if (lip->li_lsn == qip->qli_flush_lsn) {
1128 /* xfs_ail_update_finish() drops the AIL lock */
1129 tail_lsn = xfs_ail_delete_one(ailp, lip);
1130 xfs_ail_update_finish(ailp, tail_lsn);
1132 spin_unlock(&ailp->ail_lock);
1137 * Release the dq's flush lock since we're done with it.
1143 xfs_buf_dquot_iodone(
1146 struct xfs_log_item *lip, *n;
1148 list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) {
1149 list_del_init(&lip->li_bio_list);
1150 xfs_qm_dqflush_done(lip);
1155 xfs_buf_dquot_io_fail(
1158 struct xfs_log_item *lip;
1160 spin_lock(&bp->b_mount->m_ail->ail_lock);
1161 list_for_each_entry(lip, &bp->b_li_list, li_bio_list)
1162 xfs_set_li_failed(lip, bp);
1163 spin_unlock(&bp->b_mount->m_ail->ail_lock);
1166 /* Check incore dquot for errors before we flush. */
1167 static xfs_failaddr_t
1168 xfs_qm_dqflush_check(
1169 struct xfs_dquot *dqp)
1171 xfs_dqtype_t type = xfs_dquot_type(dqp);
1173 if (type != XFS_DQTYPE_USER &&
1174 type != XFS_DQTYPE_GROUP &&
1175 type != XFS_DQTYPE_PROJ)
1176 return __this_address;
1181 if (dqp->q_blk.softlimit && dqp->q_blk.count > dqp->q_blk.softlimit &&
1183 return __this_address;
1185 if (dqp->q_ino.softlimit && dqp->q_ino.count > dqp->q_ino.softlimit &&
1187 return __this_address;
1189 if (dqp->q_rtb.softlimit && dqp->q_rtb.count > dqp->q_rtb.softlimit &&
1191 return __this_address;
1193 /* bigtime flag should never be set on root dquots */
1194 if (dqp->q_type & XFS_DQTYPE_BIGTIME) {
1195 if (!xfs_has_bigtime(dqp->q_mount))
1196 return __this_address;
1198 return __this_address;
1205 * Write a modified dquot to disk.
1206 * The dquot must be locked and the flush lock too taken by caller.
1207 * The flush lock will not be unlocked until the dquot reaches the disk,
1208 * but the dquot is free to be unlocked and modified by the caller
1209 * in the interim. Dquot is still locked on return. This behavior is
1210 * identical to that of inodes.
1214 struct xfs_dquot *dqp,
1215 struct xfs_buf **bpp)
1217 struct xfs_mount *mp = dqp->q_mount;
1218 struct xfs_log_item *lip = &dqp->q_logitem.qli_item;
1220 struct xfs_dqblk *dqblk;
1224 ASSERT(XFS_DQ_IS_LOCKED(dqp));
1225 ASSERT(!completion_done(&dqp->q_flush));
1227 trace_xfs_dqflush(dqp);
1231 xfs_qm_dqunpin_wait(dqp);
1234 * Get the buffer containing the on-disk dquot
1236 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
1237 mp->m_quotainfo->qi_dqchunklen, XBF_TRYLOCK,
1238 &bp, &xfs_dquot_buf_ops);
1239 if (error == -EAGAIN)
1244 fa = xfs_qm_dqflush_check(dqp);
1246 xfs_alert(mp, "corrupt dquot ID 0x%x in memory at %pS",
1249 error = -EFSCORRUPTED;
1253 /* Flush the incore dquot to the ondisk buffer. */
1254 dqblk = bp->b_addr + dqp->q_bufoffset;
1255 xfs_dquot_to_disk(&dqblk->dd_diskdq, dqp);
1258 * Clear the dirty field and remember the flush lsn for later use.
1260 dqp->q_flags &= ~XFS_DQFLAG_DIRTY;
1262 xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn,
1263 &dqp->q_logitem.qli_item.li_lsn);
1266 * copy the lsn into the on-disk dquot now while we have the in memory
1267 * dquot here. This can't be done later in the write verifier as we
1268 * can't get access to the log item at that point in time.
1270 * We also calculate the CRC here so that the on-disk dquot in the
1271 * buffer always has a valid CRC. This ensures there is no possibility
1272 * of a dquot without an up-to-date CRC getting to disk.
1274 if (xfs_has_crc(mp)) {
1275 dqblk->dd_lsn = cpu_to_be64(dqp->q_logitem.qli_item.li_lsn);
1276 xfs_update_cksum((char *)dqblk, sizeof(struct xfs_dqblk),
1281 * Attach the dquot to the buffer so that we can remove this dquot from
1282 * the AIL and release the flush lock once the dquot is synced to disk.
1284 bp->b_flags |= _XBF_DQUOTS;
1285 list_add_tail(&dqp->q_logitem.qli_item.li_bio_list, &bp->b_li_list);
1288 * If the buffer is pinned then push on the log so we won't
1289 * get stuck waiting in the write for too long.
1291 if (xfs_buf_ispinned(bp)) {
1292 trace_xfs_dqflush_force(dqp);
1293 xfs_log_force(mp, 0);
1296 trace_xfs_dqflush_done(dqp);
1301 dqp->q_flags &= ~XFS_DQFLAG_DIRTY;
1302 xfs_trans_ail_delete(lip, 0);
1303 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1310 * Lock two xfs_dquot structures.
1312 * To avoid deadlocks we always lock the quota structure with
1313 * the lowerd id first.
1317 struct xfs_dquot *d1,
1318 struct xfs_dquot *d2)
1322 if (d1->q_id > d2->q_id) {
1323 mutex_lock(&d2->q_qlock);
1324 mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED);
1326 mutex_lock(&d1->q_qlock);
1327 mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED);
1330 mutex_lock(&d1->q_qlock);
1332 mutex_lock(&d2->q_qlock);
1339 xfs_dquot_cache = kmem_cache_create("xfs_dquot",
1340 sizeof(struct xfs_dquot),
1342 if (!xfs_dquot_cache)
1345 xfs_dqtrx_cache = kmem_cache_create("xfs_dqtrx",
1346 sizeof(struct xfs_dquot_acct),
1348 if (!xfs_dqtrx_cache)
1349 goto out_free_dquot_cache;
1353 out_free_dquot_cache:
1354 kmem_cache_destroy(xfs_dquot_cache);
1362 kmem_cache_destroy(xfs_dqtrx_cache);
1363 kmem_cache_destroy(xfs_dquot_cache);
1367 * Iterate every dquot of a particular type. The caller must ensure that the
1368 * particular quota type is active. iter_fn can return negative error codes,
1369 * or -ECANCELED to indicate that it wants to stop iterating.
1373 struct xfs_mount *mp,
1375 xfs_qm_dqiterate_fn iter_fn,
1378 struct xfs_dquot *dq;
1383 error = xfs_qm_dqget_next(mp, id, type, &dq);
1384 if (error == -ENOENT)
1389 error = iter_fn(dq, type, priv);
1392 } while (error == 0 && id != 0);