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"
27 #include "xfs_health.h"
34 * dquot->q_qlock (xfs_dqlock() and friends)
35 * dquot->q_flush (xfs_dqflock() and friends)
38 * If two dquots need to be locked the order is user before group/project,
39 * otherwise by the lowest id first, see xfs_dqlock2.
42 struct kmem_cache
*xfs_dqtrx_cache
;
43 static struct kmem_cache
*xfs_dquot_cache
;
45 static struct lock_class_key xfs_dquot_group_class
;
46 static struct lock_class_key xfs_dquot_project_class
;
48 /* Record observations of quota corruption with the health tracking system. */
51 struct xfs_dquot
*dqp
)
53 struct xfs_mount
*mp
= dqp
->q_mount
;
55 switch (dqp
->q_type
) {
57 xfs_fs_mark_sick(mp
, XFS_SICK_FS_UQUOTA
);
59 case XFS_DQTYPE_GROUP
:
60 xfs_fs_mark_sick(mp
, XFS_SICK_FS_GQUOTA
);
63 xfs_fs_mark_sick(mp
, XFS_SICK_FS_PQUOTA
);
72 * This is called to free all the memory associated with a dquot
76 struct xfs_dquot
*dqp
)
78 ASSERT(list_empty(&dqp
->q_lru
));
80 kvfree(dqp
->q_logitem
.qli_item
.li_lv_shadow
);
81 mutex_destroy(&dqp
->q_qlock
);
83 XFS_STATS_DEC(dqp
->q_mount
, xs_qm_dquot
);
84 kmem_cache_free(xfs_dquot_cache
, dqp
);
88 * If default limits are in force, push them into the dquot now.
89 * We overwrite the dquot limits only if they are zero and this
90 * is not the root dquot.
93 xfs_qm_adjust_dqlimits(
96 struct xfs_mount
*mp
= dq
->q_mount
;
97 struct xfs_quotainfo
*q
= mp
->m_quotainfo
;
98 struct xfs_def_quota
*defq
;
102 defq
= xfs_get_defquota(q
, xfs_dquot_type(dq
));
104 if (!dq
->q_blk
.softlimit
) {
105 dq
->q_blk
.softlimit
= defq
->blk
.soft
;
108 if (!dq
->q_blk
.hardlimit
) {
109 dq
->q_blk
.hardlimit
= defq
->blk
.hard
;
112 if (!dq
->q_ino
.softlimit
)
113 dq
->q_ino
.softlimit
= defq
->ino
.soft
;
114 if (!dq
->q_ino
.hardlimit
)
115 dq
->q_ino
.hardlimit
= defq
->ino
.hard
;
116 if (!dq
->q_rtb
.softlimit
)
117 dq
->q_rtb
.softlimit
= defq
->rtb
.soft
;
118 if (!dq
->q_rtb
.hardlimit
)
119 dq
->q_rtb
.hardlimit
= defq
->rtb
.hard
;
122 xfs_dquot_set_prealloc_limits(dq
);
125 /* Set the expiration time of a quota's grace period. */
127 xfs_dquot_set_timeout(
128 struct xfs_mount
*mp
,
131 struct xfs_quotainfo
*qi
= mp
->m_quotainfo
;
133 return clamp_t(time64_t
, timeout
, qi
->qi_expiry_min
,
137 /* Set the length of the default grace period. */
139 xfs_dquot_set_grace_period(
142 return clamp_t(time64_t
, grace
, XFS_DQ_GRACE_MIN
, XFS_DQ_GRACE_MAX
);
146 * Determine if this quota counter is over either limit and set the quota
147 * timers as appropriate.
150 xfs_qm_adjust_res_timer(
151 struct xfs_mount
*mp
,
152 struct xfs_dquot_res
*res
,
153 struct xfs_quota_limits
*qlim
)
155 ASSERT(res
->hardlimit
== 0 || res
->softlimit
<= res
->hardlimit
);
157 if ((res
->softlimit
&& res
->count
> res
->softlimit
) ||
158 (res
->hardlimit
&& res
->count
> res
->hardlimit
)) {
160 res
->timer
= xfs_dquot_set_timeout(mp
,
161 ktime_get_real_seconds() + qlim
->time
);
168 * Check the limits and timers of a dquot and start or reset timers
170 * This gets called even when quota enforcement is OFF, which makes our
171 * life a little less complicated. (We just don't reject any quota
172 * reservations in that case, when enforcement is off).
173 * We also return 0 as the values of the timers in Q_GETQUOTA calls, when
175 * In contrast, warnings are a little different in that they don't
176 * 'automatically' get started when limits get exceeded. They do
177 * get reset to zero, however, when we find the count to be under
178 * the soft limit (they are only ever set non-zero via userspace).
181 xfs_qm_adjust_dqtimers(
182 struct xfs_dquot
*dq
)
184 struct xfs_mount
*mp
= dq
->q_mount
;
185 struct xfs_quotainfo
*qi
= mp
->m_quotainfo
;
186 struct xfs_def_quota
*defq
;
189 defq
= xfs_get_defquota(qi
, xfs_dquot_type(dq
));
191 xfs_qm_adjust_res_timer(dq
->q_mount
, &dq
->q_blk
, &defq
->blk
);
192 xfs_qm_adjust_res_timer(dq
->q_mount
, &dq
->q_ino
, &defq
->ino
);
193 xfs_qm_adjust_res_timer(dq
->q_mount
, &dq
->q_rtb
, &defq
->rtb
);
197 * initialize a buffer full of dquots and log the whole thing
200 xfs_qm_init_dquot_blk(
201 struct xfs_trans
*tp
,
206 struct xfs_mount
*mp
= tp
->t_mountp
;
207 struct xfs_quotainfo
*q
= mp
->m_quotainfo
;
211 unsigned int blftype
;
215 ASSERT(xfs_buf_islocked(bp
));
218 case XFS_DQTYPE_USER
:
219 qflag
= XFS_UQUOTA_CHKD
;
220 blftype
= XFS_BLF_UDQUOT_BUF
;
222 case XFS_DQTYPE_PROJ
:
223 qflag
= XFS_PQUOTA_CHKD
;
224 blftype
= XFS_BLF_PDQUOT_BUF
;
226 case XFS_DQTYPE_GROUP
:
227 qflag
= XFS_GQUOTA_CHKD
;
228 blftype
= XFS_BLF_GDQUOT_BUF
;
238 * ID of the first dquot in the block - id's are zero based.
240 curid
= id
- (id
% q
->qi_dqperchunk
);
241 memset(d
, 0, BBTOB(q
->qi_dqchunklen
));
242 for (i
= 0; i
< q
->qi_dqperchunk
; i
++, d
++, curid
++) {
243 d
->dd_diskdq
.d_magic
= cpu_to_be16(XFS_DQUOT_MAGIC
);
244 d
->dd_diskdq
.d_version
= XFS_DQUOT_VERSION
;
245 d
->dd_diskdq
.d_id
= cpu_to_be32(curid
);
246 d
->dd_diskdq
.d_type
= type
;
247 if (curid
> 0 && xfs_has_bigtime(mp
))
248 d
->dd_diskdq
.d_type
|= XFS_DQTYPE_BIGTIME
;
249 if (xfs_has_crc(mp
)) {
250 uuid_copy(&d
->dd_uuid
, &mp
->m_sb
.sb_meta_uuid
);
251 xfs_update_cksum((char *)d
, sizeof(struct xfs_dqblk
),
256 xfs_trans_dquot_buf(tp
, bp
, blftype
);
259 * quotacheck uses delayed writes to update all the dquots on disk in an
260 * efficient manner instead of logging the individual dquot changes as
261 * they are made. However if we log the buffer allocated here and crash
262 * after quotacheck while the logged initialisation is still in the
263 * active region of the log, log recovery can replay the dquot buffer
264 * initialisation over the top of the checked dquots and corrupt quota
267 * To avoid this problem, quotacheck cannot log the initialised buffer.
268 * We must still dirty the buffer and write it back before the
269 * allocation transaction clears the log. Therefore, mark the buffer as
270 * ordered instead of logging it directly. This is safe for quotacheck
271 * because it detects and repairs allocated but initialized dquot blocks
272 * in the quota inodes.
274 if (!(mp
->m_qflags
& qflag
))
275 xfs_trans_ordered_buf(tp
, bp
);
277 xfs_trans_log_buf(tp
, bp
, 0, BBTOB(q
->qi_dqchunklen
) - 1);
281 xfs_dquot_set_prealloc(
282 struct xfs_dquot_pre
*pre
,
283 const struct xfs_dquot_res
*res
)
287 pre
->q_prealloc_hi_wmark
= res
->hardlimit
;
288 pre
->q_prealloc_lo_wmark
= res
->softlimit
;
290 space
= div_u64(pre
->q_prealloc_hi_wmark
, 100);
291 if (!pre
->q_prealloc_lo_wmark
)
292 pre
->q_prealloc_lo_wmark
= space
* 95;
294 pre
->q_low_space
[XFS_QLOWSP_1_PCNT
] = space
;
295 pre
->q_low_space
[XFS_QLOWSP_3_PCNT
] = space
* 3;
296 pre
->q_low_space
[XFS_QLOWSP_5_PCNT
] = space
* 5;
300 * Initialize the dynamic speculative preallocation thresholds. The lo/hi
301 * watermarks correspond to the soft and hard limits by default. If a soft limit
302 * is not specified, we use 95% of the hard limit.
305 xfs_dquot_set_prealloc_limits(struct xfs_dquot
*dqp
)
307 xfs_dquot_set_prealloc(&dqp
->q_blk_prealloc
, &dqp
->q_blk
);
308 xfs_dquot_set_prealloc(&dqp
->q_rtb_prealloc
, &dqp
->q_rtb
);
312 * Ensure that the given in-core dquot has a buffer on disk backing it, and
313 * return the buffer locked and held. This is called when the bmapi finds a
317 xfs_dquot_disk_alloc(
318 struct xfs_dquot
*dqp
,
319 struct xfs_buf
**bpp
)
321 struct xfs_bmbt_irec map
;
322 struct xfs_trans
*tp
;
323 struct xfs_mount
*mp
= dqp
->q_mount
;
325 xfs_dqtype_t qtype
= xfs_dquot_type(dqp
);
326 struct xfs_inode
*quotip
= xfs_quota_inode(mp
, qtype
);
330 trace_xfs_dqalloc(dqp
);
332 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_qm_dqalloc
,
333 XFS_QM_DQALLOC_SPACE_RES(mp
), 0, 0, &tp
);
337 xfs_ilock(quotip
, XFS_ILOCK_EXCL
);
338 xfs_trans_ijoin(tp
, quotip
, 0);
340 if (!xfs_this_quota_on(dqp
->q_mount
, qtype
)) {
342 * Return if this type of quotas is turned off while we didn't
349 error
= xfs_iext_count_extend(tp
, quotip
, XFS_DATA_FORK
,
350 XFS_IEXT_ADD_NOSPLIT_CNT
);
354 /* Create the block mapping. */
355 error
= xfs_bmapi_write(tp
, quotip
, dqp
->q_fileoffset
,
356 XFS_DQUOT_CLUSTER_SIZE_FSB
, XFS_BMAPI_METADATA
, 0, &map
,
361 ASSERT(map
.br_blockcount
== XFS_DQUOT_CLUSTER_SIZE_FSB
);
362 ASSERT((map
.br_startblock
!= DELAYSTARTBLOCK
) &&
363 (map
.br_startblock
!= HOLESTARTBLOCK
));
366 * Keep track of the blkno to save a lookup later
368 dqp
->q_blkno
= XFS_FSB_TO_DADDR(mp
, map
.br_startblock
);
370 /* now we can just get the buffer (there's nothing to read yet) */
371 error
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, dqp
->q_blkno
,
372 mp
->m_quotainfo
->qi_dqchunklen
, 0, &bp
);
375 bp
->b_ops
= &xfs_dquot_buf_ops
;
378 * Make a chunk of dquots out of this buffer and log
381 xfs_qm_init_dquot_blk(tp
, dqp
->q_id
, qtype
, bp
);
382 xfs_buf_set_ref(bp
, XFS_DQUOT_REF
);
385 * Hold the buffer and join it to the dfops so that we'll still own
386 * the buffer when we return to the caller. The buffer disposal on
387 * error must be paid attention to very carefully, as it has been
388 * broken since commit efa092f3d4c6 "[XFS] Fixes a bug in the quota
389 * code when allocating a new dquot record" in 2005, and the later
390 * conversion to xfs_defer_ops in commit 310a75a3c6c747 failed to keep
391 * the buffer locked across the _defer_finish call. We can now do
392 * this correctly with xfs_defer_bjoin.
394 * Above, we allocated a disk block for the dquot information and used
395 * get_buf to initialize the dquot. If the _defer_finish fails, the old
396 * transaction is gone but the new buffer is not joined or held to any
397 * transaction, so we must _buf_relse it.
399 * If everything succeeds, the caller of this function is returned a
400 * buffer that is locked and held to the transaction. The caller
401 * is responsible for unlocking any buffer passed back, either
402 * manually or by committing the transaction. On error, the buffer is
403 * released and not passed back.
405 * Keep the quota inode ILOCKed until after the transaction commit to
406 * maintain the atomicity of bmap/rmap updates.
408 xfs_trans_bhold(tp
, bp
);
409 error
= xfs_trans_commit(tp
);
410 xfs_iunlock(quotip
, XFS_ILOCK_EXCL
);
420 xfs_trans_cancel(tp
);
421 xfs_iunlock(quotip
, XFS_ILOCK_EXCL
);
426 * Read in the in-core dquot's on-disk metadata and return the buffer.
427 * Returns ENOENT to signal a hole.
431 struct xfs_mount
*mp
,
432 struct xfs_dquot
*dqp
,
433 struct xfs_buf
**bpp
)
435 struct xfs_bmbt_irec map
;
437 xfs_dqtype_t qtype
= xfs_dquot_type(dqp
);
438 struct xfs_inode
*quotip
= xfs_quota_inode(mp
, qtype
);
443 lock_mode
= xfs_ilock_data_map_shared(quotip
);
444 if (!xfs_this_quota_on(mp
, qtype
)) {
446 * Return if this type of quotas is turned off while we
447 * didn't have the quota inode lock.
449 xfs_iunlock(quotip
, lock_mode
);
454 * Find the block map; no allocations yet
456 error
= xfs_bmapi_read(quotip
, dqp
->q_fileoffset
,
457 XFS_DQUOT_CLUSTER_SIZE_FSB
, &map
, &nmaps
, 0);
458 xfs_iunlock(quotip
, lock_mode
);
463 ASSERT(map
.br_blockcount
>= 1);
464 ASSERT(map
.br_startblock
!= DELAYSTARTBLOCK
);
465 if (map
.br_startblock
== HOLESTARTBLOCK
)
468 trace_xfs_dqtobp_read(dqp
);
471 * store the blkno etc so that we don't have to do the
472 * mapping all the time
474 dqp
->q_blkno
= XFS_FSB_TO_DADDR(mp
, map
.br_startblock
);
476 error
= xfs_trans_read_buf(mp
, NULL
, mp
->m_ddev_targp
, dqp
->q_blkno
,
477 mp
->m_quotainfo
->qi_dqchunklen
, 0, &bp
,
479 if (xfs_metadata_is_sick(error
))
480 xfs_dquot_mark_sick(dqp
);
486 ASSERT(xfs_buf_islocked(bp
));
487 xfs_buf_set_ref(bp
, XFS_DQUOT_REF
);
493 /* Allocate and initialize everything we need for an incore dquot. */
494 STATIC
struct xfs_dquot
*
496 struct xfs_mount
*mp
,
500 struct xfs_dquot
*dqp
;
502 dqp
= kmem_cache_zalloc(xfs_dquot_cache
, GFP_KERNEL
| __GFP_NOFAIL
);
507 INIT_LIST_HEAD(&dqp
->q_lru
);
508 mutex_init(&dqp
->q_qlock
);
509 init_waitqueue_head(&dqp
->q_pinwait
);
510 dqp
->q_fileoffset
= (xfs_fileoff_t
)id
/ mp
->m_quotainfo
->qi_dqperchunk
;
512 * Offset of dquot in the (fixed sized) dquot chunk.
514 dqp
->q_bufoffset
= (id
% mp
->m_quotainfo
->qi_dqperchunk
) *
515 sizeof(struct xfs_dqblk
);
518 * Because we want to use a counting completion, complete
519 * the flush completion once to allow a single access to
520 * the flush completion without blocking.
522 init_completion(&dqp
->q_flush
);
523 complete(&dqp
->q_flush
);
526 * Make sure group quotas have a different lock class than user
530 case XFS_DQTYPE_USER
:
531 /* uses the default lock class */
533 case XFS_DQTYPE_GROUP
:
534 lockdep_set_class(&dqp
->q_qlock
, &xfs_dquot_group_class
);
536 case XFS_DQTYPE_PROJ
:
537 lockdep_set_class(&dqp
->q_qlock
, &xfs_dquot_project_class
);
544 xfs_qm_dquot_logitem_init(dqp
);
546 XFS_STATS_INC(mp
, xs_qm_dquot
);
550 /* Check the ondisk dquot's id and type match what the incore dquot expects. */
552 xfs_dquot_check_type(
553 struct xfs_dquot
*dqp
,
554 struct xfs_disk_dquot
*ddqp
)
559 ddqp_type
= ddqp
->d_type
& XFS_DQTYPE_REC_MASK
;
560 dqp_type
= xfs_dquot_type(dqp
);
562 if (be32_to_cpu(ddqp
->d_id
) != dqp
->q_id
)
566 * V5 filesystems always expect an exact type match. V4 filesystems
567 * expect an exact match for user dquots and for non-root group and
570 if (xfs_has_crc(dqp
->q_mount
) ||
571 dqp_type
== XFS_DQTYPE_USER
|| dqp
->q_id
!= 0)
572 return ddqp_type
== dqp_type
;
575 * V4 filesystems support either group or project quotas, but not both
576 * at the same time. The non-user quota file can be switched between
577 * group and project quota uses depending on the mount options, which
578 * means that we can encounter the other type when we try to load quota
579 * defaults. Quotacheck will soon reset the entire quota file
580 * (including the root dquot) anyway, but don't log scary corruption
583 return ddqp_type
== XFS_DQTYPE_GROUP
|| ddqp_type
== XFS_DQTYPE_PROJ
;
586 /* Copy the in-core quota fields in from the on-disk buffer. */
589 struct xfs_dquot
*dqp
,
592 struct xfs_dqblk
*dqb
= xfs_buf_offset(bp
, dqp
->q_bufoffset
);
593 struct xfs_disk_dquot
*ddqp
= &dqb
->dd_diskdq
;
596 * Ensure that we got the type and ID we were looking for.
597 * Everything else was checked by the dquot buffer verifier.
599 if (!xfs_dquot_check_type(dqp
, ddqp
)) {
600 xfs_alert_tag(bp
->b_mount
, XFS_PTAG_VERIFIER_ERROR
,
601 "Metadata corruption detected at %pS, quota %u",
602 __this_address
, dqp
->q_id
);
603 xfs_alert(bp
->b_mount
, "Unmount and run xfs_repair");
604 xfs_dquot_mark_sick(dqp
);
605 return -EFSCORRUPTED
;
608 /* copy everything from disk dquot to the incore dquot */
609 dqp
->q_type
= ddqp
->d_type
;
610 dqp
->q_blk
.hardlimit
= be64_to_cpu(ddqp
->d_blk_hardlimit
);
611 dqp
->q_blk
.softlimit
= be64_to_cpu(ddqp
->d_blk_softlimit
);
612 dqp
->q_ino
.hardlimit
= be64_to_cpu(ddqp
->d_ino_hardlimit
);
613 dqp
->q_ino
.softlimit
= be64_to_cpu(ddqp
->d_ino_softlimit
);
614 dqp
->q_rtb
.hardlimit
= be64_to_cpu(ddqp
->d_rtb_hardlimit
);
615 dqp
->q_rtb
.softlimit
= be64_to_cpu(ddqp
->d_rtb_softlimit
);
617 dqp
->q_blk
.count
= be64_to_cpu(ddqp
->d_bcount
);
618 dqp
->q_ino
.count
= be64_to_cpu(ddqp
->d_icount
);
619 dqp
->q_rtb
.count
= be64_to_cpu(ddqp
->d_rtbcount
);
621 dqp
->q_blk
.timer
= xfs_dquot_from_disk_ts(ddqp
, ddqp
->d_btimer
);
622 dqp
->q_ino
.timer
= xfs_dquot_from_disk_ts(ddqp
, ddqp
->d_itimer
);
623 dqp
->q_rtb
.timer
= xfs_dquot_from_disk_ts(ddqp
, ddqp
->d_rtbtimer
);
626 * Reservation counters are defined as reservation plus current usage
627 * to avoid having to add every time.
629 dqp
->q_blk
.reserved
= dqp
->q_blk
.count
;
630 dqp
->q_ino
.reserved
= dqp
->q_ino
.count
;
631 dqp
->q_rtb
.reserved
= dqp
->q_rtb
.count
;
633 /* initialize the dquot speculative prealloc thresholds */
634 xfs_dquot_set_prealloc_limits(dqp
);
638 /* Copy the in-core quota fields into the on-disk buffer. */
641 struct xfs_disk_dquot
*ddqp
,
642 struct xfs_dquot
*dqp
)
644 ddqp
->d_magic
= cpu_to_be16(XFS_DQUOT_MAGIC
);
645 ddqp
->d_version
= XFS_DQUOT_VERSION
;
646 ddqp
->d_type
= dqp
->q_type
;
647 ddqp
->d_id
= cpu_to_be32(dqp
->q_id
);
651 ddqp
->d_blk_hardlimit
= cpu_to_be64(dqp
->q_blk
.hardlimit
);
652 ddqp
->d_blk_softlimit
= cpu_to_be64(dqp
->q_blk
.softlimit
);
653 ddqp
->d_ino_hardlimit
= cpu_to_be64(dqp
->q_ino
.hardlimit
);
654 ddqp
->d_ino_softlimit
= cpu_to_be64(dqp
->q_ino
.softlimit
);
655 ddqp
->d_rtb_hardlimit
= cpu_to_be64(dqp
->q_rtb
.hardlimit
);
656 ddqp
->d_rtb_softlimit
= cpu_to_be64(dqp
->q_rtb
.softlimit
);
658 ddqp
->d_bcount
= cpu_to_be64(dqp
->q_blk
.count
);
659 ddqp
->d_icount
= cpu_to_be64(dqp
->q_ino
.count
);
660 ddqp
->d_rtbcount
= cpu_to_be64(dqp
->q_rtb
.count
);
664 ddqp
->d_rtbwarns
= 0;
666 ddqp
->d_btimer
= xfs_dquot_to_disk_ts(dqp
, dqp
->q_blk
.timer
);
667 ddqp
->d_itimer
= xfs_dquot_to_disk_ts(dqp
, dqp
->q_ino
.timer
);
668 ddqp
->d_rtbtimer
= xfs_dquot_to_disk_ts(dqp
, dqp
->q_rtb
.timer
);
672 * Read in the ondisk dquot using dqtobp() then copy it to an incore version,
673 * and release the buffer immediately. If @can_alloc is true, fill any
674 * holes in the on-disk metadata.
678 struct xfs_mount
*mp
,
682 struct xfs_dquot
**dqpp
)
684 struct xfs_dquot
*dqp
;
688 dqp
= xfs_dquot_alloc(mp
, id
, type
);
689 trace_xfs_dqread(dqp
);
691 /* Try to read the buffer, allocating if necessary. */
692 error
= xfs_dquot_disk_read(mp
, dqp
, &bp
);
693 if (error
== -ENOENT
&& can_alloc
)
694 error
= xfs_dquot_disk_alloc(dqp
, &bp
);
699 * At this point we should have a clean locked buffer. Copy the data
700 * to the incore dquot and release the buffer since the incore dquot
701 * has its own locking protocol so we needn't tie up the buffer any
704 ASSERT(xfs_buf_islocked(bp
));
705 error
= xfs_dquot_from_disk(dqp
, bp
);
714 trace_xfs_dqread_fail(dqp
);
715 xfs_qm_dqdestroy(dqp
);
721 * Advance to the next id in the current chunk, or if at the
722 * end of the chunk, skip ahead to first id in next allocated chunk
723 * using the SEEK_DATA interface.
727 struct xfs_mount
*mp
,
731 struct xfs_inode
*quotip
= xfs_quota_inode(mp
, type
);
732 xfs_dqid_t next_id
= *id
+ 1; /* simple advance */
734 struct xfs_bmbt_irec got
;
735 struct xfs_iext_cursor cur
;
739 /* If we'd wrap past the max ID, stop */
743 /* If new ID is within the current chunk, advancing it sufficed */
744 if (next_id
% mp
->m_quotainfo
->qi_dqperchunk
) {
749 /* Nope, next_id is now past the current chunk, so find the next one */
750 start
= (xfs_fsblock_t
)next_id
/ mp
->m_quotainfo
->qi_dqperchunk
;
752 lock_flags
= xfs_ilock_data_map_shared(quotip
);
753 error
= xfs_iread_extents(NULL
, quotip
, XFS_DATA_FORK
);
757 if (xfs_iext_lookup_extent(quotip
, "ip
->i_df
, start
, &cur
, &got
)) {
758 /* contiguous chunk, bump startoff for the id calculation */
759 if (got
.br_startoff
< start
)
760 got
.br_startoff
= start
;
761 *id
= got
.br_startoff
* mp
->m_quotainfo
->qi_dqperchunk
;
766 xfs_iunlock(quotip
, lock_flags
);
772 * Look up the dquot in the in-core cache. If found, the dquot is returned
773 * locked and ready to go.
775 static struct xfs_dquot
*
776 xfs_qm_dqget_cache_lookup(
777 struct xfs_mount
*mp
,
778 struct xfs_quotainfo
*qi
,
779 struct radix_tree_root
*tree
,
782 struct xfs_dquot
*dqp
;
785 mutex_lock(&qi
->qi_tree_lock
);
786 dqp
= radix_tree_lookup(tree
, id
);
788 mutex_unlock(&qi
->qi_tree_lock
);
789 XFS_STATS_INC(mp
, xs_qm_dqcachemisses
);
794 if (dqp
->q_flags
& XFS_DQFLAG_FREEING
) {
796 mutex_unlock(&qi
->qi_tree_lock
);
797 trace_xfs_dqget_freeing(dqp
);
803 mutex_unlock(&qi
->qi_tree_lock
);
805 trace_xfs_dqget_hit(dqp
);
806 XFS_STATS_INC(mp
, xs_qm_dqcachehits
);
811 * Try to insert a new dquot into the in-core cache. If an error occurs the
812 * caller should throw away the dquot and start over. Otherwise, the dquot
813 * is returned locked (and held by the cache) as if there had been a cache
816 * The insert needs to be done under memalloc_nofs context because the radix
817 * tree can do memory allocation during insert. The qi->qi_tree_lock is taken in
818 * memory reclaim when freeing unused dquots, so we cannot have the radix tree
819 * node allocation recursing into filesystem reclaim whilst we hold the
823 xfs_qm_dqget_cache_insert(
824 struct xfs_mount
*mp
,
825 struct xfs_quotainfo
*qi
,
826 struct radix_tree_root
*tree
,
828 struct xfs_dquot
*dqp
)
830 unsigned int nofs_flags
;
833 nofs_flags
= memalloc_nofs_save();
834 mutex_lock(&qi
->qi_tree_lock
);
835 error
= radix_tree_insert(tree
, id
, dqp
);
836 if (unlikely(error
)) {
837 /* Duplicate found! Caller must try again. */
838 trace_xfs_dqget_dup(dqp
);
842 /* Return a locked dquot to the caller, with a reference taken. */
848 mutex_unlock(&qi
->qi_tree_lock
);
849 memalloc_nofs_restore(nofs_flags
);
853 /* Check our input parameters. */
856 struct xfs_mount
*mp
,
860 case XFS_DQTYPE_USER
:
861 if (!XFS_IS_UQUOTA_ON(mp
))
864 case XFS_DQTYPE_GROUP
:
865 if (!XFS_IS_GQUOTA_ON(mp
))
868 case XFS_DQTYPE_PROJ
:
869 if (!XFS_IS_PQUOTA_ON(mp
))
879 * Given the file system, id, and type (UDQUOT/GDQUOT/PDQUOT), return a
880 * locked dquot, doing an allocation (if requested) as needed.
884 struct xfs_mount
*mp
,
888 struct xfs_dquot
**O_dqpp
)
890 struct xfs_quotainfo
*qi
= mp
->m_quotainfo
;
891 struct radix_tree_root
*tree
= xfs_dquot_tree(qi
, type
);
892 struct xfs_dquot
*dqp
;
895 error
= xfs_qm_dqget_checks(mp
, type
);
900 dqp
= xfs_qm_dqget_cache_lookup(mp
, qi
, tree
, id
);
906 error
= xfs_qm_dqread(mp
, id
, type
, can_alloc
, &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
);
921 trace_xfs_dqget_miss(dqp
);
927 * Given a dquot id and type, read and initialize a dquot from the on-disk
928 * metadata. This function is only for use during quota initialization so
929 * it ignores the dquot cache assuming that the dquot shrinker isn't set up.
930 * The caller is responsible for _qm_dqdestroy'ing the returned dquot.
933 xfs_qm_dqget_uncached(
934 struct xfs_mount
*mp
,
937 struct xfs_dquot
**dqpp
)
941 error
= xfs_qm_dqget_checks(mp
, type
);
945 return xfs_qm_dqread(mp
, id
, type
, 0, dqpp
);
948 /* Return the quota id for a given inode and type. */
950 xfs_qm_id_for_quotatype(
951 struct xfs_inode
*ip
,
955 case XFS_DQTYPE_USER
:
956 return i_uid_read(VFS_I(ip
));
957 case XFS_DQTYPE_GROUP
:
958 return i_gid_read(VFS_I(ip
));
959 case XFS_DQTYPE_PROJ
:
967 * Return the dquot for a given inode and type. If @can_alloc is true, then
968 * allocate blocks if needed. The inode's ILOCK must be held and it must not
969 * have already had an inode attached.
973 struct xfs_inode
*ip
,
976 struct xfs_dquot
**O_dqpp
)
978 struct xfs_mount
*mp
= ip
->i_mount
;
979 struct xfs_quotainfo
*qi
= mp
->m_quotainfo
;
980 struct radix_tree_root
*tree
= xfs_dquot_tree(qi
, type
);
981 struct xfs_dquot
*dqp
;
985 error
= xfs_qm_dqget_checks(mp
, type
);
989 xfs_assert_ilocked(ip
, XFS_ILOCK_EXCL
);
990 ASSERT(xfs_inode_dquot(ip
, type
) == NULL
);
991 ASSERT(!xfs_is_metadir_inode(ip
));
993 id
= xfs_qm_id_for_quotatype(ip
, type
);
996 dqp
= xfs_qm_dqget_cache_lookup(mp
, qi
, tree
, id
);
1003 * Dquot cache miss. We don't want to keep the inode lock across
1004 * a (potential) disk read. Also we don't want to deal with the lock
1005 * ordering between quotainode and this inode. OTOH, dropping the inode
1006 * lock here means dealing with a chown that can happen before
1007 * we re-acquire the lock.
1009 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1010 error
= xfs_qm_dqread(mp
, id
, type
, can_alloc
, &dqp
);
1011 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1016 * A dquot could be attached to this inode by now, since we had
1017 * dropped the ilock.
1019 if (xfs_this_quota_on(mp
, type
)) {
1020 struct xfs_dquot
*dqp1
;
1022 dqp1
= xfs_inode_dquot(ip
, type
);
1024 xfs_qm_dqdestroy(dqp
);
1030 /* inode stays locked on return */
1031 xfs_qm_dqdestroy(dqp
);
1035 error
= xfs_qm_dqget_cache_insert(mp
, qi
, tree
, id
, dqp
);
1038 * Duplicate found. Just throw away the new dquot and start
1041 xfs_qm_dqdestroy(dqp
);
1042 XFS_STATS_INC(mp
, xs_qm_dquot_dups
);
1047 xfs_assert_ilocked(ip
, XFS_ILOCK_EXCL
);
1048 trace_xfs_dqget_miss(dqp
);
1054 * Starting at @id and progressing upwards, look for an initialized incore
1055 * dquot, lock it, and return it.
1059 struct xfs_mount
*mp
,
1062 struct xfs_dquot
**dqpp
)
1064 struct xfs_dquot
*dqp
;
1068 for (; !error
; error
= xfs_dq_get_next_id(mp
, type
, &id
)) {
1069 error
= xfs_qm_dqget(mp
, id
, type
, false, &dqp
);
1070 if (error
== -ENOENT
)
1072 else if (error
!= 0)
1075 if (!XFS_IS_DQUOT_UNINITIALIZED(dqp
)) {
1087 * Release a reference to the dquot (decrement ref-count) and unlock it.
1089 * If there is a group quota attached to this dquot, carefully release that
1090 * too without tripping over deadlocks'n'stuff.
1094 struct xfs_dquot
*dqp
)
1096 ASSERT(dqp
->q_nrefs
> 0);
1097 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
1099 trace_xfs_dqput(dqp
);
1101 if (--dqp
->q_nrefs
== 0) {
1102 struct xfs_quotainfo
*qi
= dqp
->q_mount
->m_quotainfo
;
1103 trace_xfs_dqput_free(dqp
);
1105 if (list_lru_add_obj(&qi
->qi_lru
, &dqp
->q_lru
))
1106 XFS_STATS_INC(dqp
->q_mount
, xs_qm_dquot_unused
);
1112 * Release a dquot. Flush it if dirty, then dqput() it.
1113 * dquot must not be locked.
1117 struct xfs_dquot
*dqp
)
1122 trace_xfs_dqrele(dqp
);
1126 * We don't care to flush it if the dquot is dirty here.
1127 * That will create stutters that we want to avoid.
1128 * Instead we do a delayed write when we try to reclaim
1129 * a dirty dquot. Also xfs_sync will take part of the burden...
1135 * This is the dquot flushing I/O completion routine. It is called
1136 * from interrupt level when the buffer containing the dquot is
1137 * flushed to disk. It is responsible for removing the dquot logitem
1138 * from the AIL if it has not been re-logged, and unlocking the dquot's
1139 * flush lock. This behavior is very similar to that of inodes..
1142 xfs_qm_dqflush_done(
1143 struct xfs_log_item
*lip
)
1145 struct xfs_dq_logitem
*qip
= (struct xfs_dq_logitem
*)lip
;
1146 struct xfs_dquot
*dqp
= qip
->qli_dquot
;
1147 struct xfs_ail
*ailp
= lip
->li_ailp
;
1151 * We only want to pull the item from the AIL if its
1152 * location in the log has not changed since we started the flush.
1153 * Thus, we only bother if the dquot's lsn has
1154 * not changed. First we check the lsn outside the lock
1155 * since it's cheaper, and then we recheck while
1156 * holding the lock before removing the dquot from the AIL.
1158 if (test_bit(XFS_LI_IN_AIL
, &lip
->li_flags
) &&
1159 ((lip
->li_lsn
== qip
->qli_flush_lsn
) ||
1160 test_bit(XFS_LI_FAILED
, &lip
->li_flags
))) {
1162 spin_lock(&ailp
->ail_lock
);
1163 xfs_clear_li_failed(lip
);
1164 if (lip
->li_lsn
== qip
->qli_flush_lsn
) {
1165 /* xfs_ail_update_finish() drops the AIL lock */
1166 tail_lsn
= xfs_ail_delete_one(ailp
, lip
);
1167 xfs_ail_update_finish(ailp
, tail_lsn
);
1169 spin_unlock(&ailp
->ail_lock
);
1174 * Release the dq's flush lock since we're done with it.
1180 xfs_buf_dquot_iodone(
1183 struct xfs_log_item
*lip
, *n
;
1185 list_for_each_entry_safe(lip
, n
, &bp
->b_li_list
, li_bio_list
) {
1186 list_del_init(&lip
->li_bio_list
);
1187 xfs_qm_dqflush_done(lip
);
1192 xfs_buf_dquot_io_fail(
1195 struct xfs_log_item
*lip
;
1197 spin_lock(&bp
->b_mount
->m_ail
->ail_lock
);
1198 list_for_each_entry(lip
, &bp
->b_li_list
, li_bio_list
)
1199 xfs_set_li_failed(lip
, bp
);
1200 spin_unlock(&bp
->b_mount
->m_ail
->ail_lock
);
1203 /* Check incore dquot for errors before we flush. */
1204 static xfs_failaddr_t
1205 xfs_qm_dqflush_check(
1206 struct xfs_dquot
*dqp
)
1208 xfs_dqtype_t type
= xfs_dquot_type(dqp
);
1210 if (type
!= XFS_DQTYPE_USER
&&
1211 type
!= XFS_DQTYPE_GROUP
&&
1212 type
!= XFS_DQTYPE_PROJ
)
1213 return __this_address
;
1218 if (dqp
->q_blk
.softlimit
&& dqp
->q_blk
.count
> dqp
->q_blk
.softlimit
&&
1220 return __this_address
;
1222 if (dqp
->q_ino
.softlimit
&& dqp
->q_ino
.count
> dqp
->q_ino
.softlimit
&&
1224 return __this_address
;
1226 if (dqp
->q_rtb
.softlimit
&& dqp
->q_rtb
.count
> dqp
->q_rtb
.softlimit
&&
1228 return __this_address
;
1230 /* bigtime flag should never be set on root dquots */
1231 if (dqp
->q_type
& XFS_DQTYPE_BIGTIME
) {
1232 if (!xfs_has_bigtime(dqp
->q_mount
))
1233 return __this_address
;
1235 return __this_address
;
1242 * Write a modified dquot to disk.
1243 * The dquot must be locked and the flush lock too taken by caller.
1244 * The flush lock will not be unlocked until the dquot reaches the disk,
1245 * but the dquot is free to be unlocked and modified by the caller
1246 * in the interim. Dquot is still locked on return. This behavior is
1247 * identical to that of inodes.
1251 struct xfs_dquot
*dqp
,
1252 struct xfs_buf
**bpp
)
1254 struct xfs_mount
*mp
= dqp
->q_mount
;
1255 struct xfs_log_item
*lip
= &dqp
->q_logitem
.qli_item
;
1257 struct xfs_dqblk
*dqblk
;
1261 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
1262 ASSERT(!completion_done(&dqp
->q_flush
));
1264 trace_xfs_dqflush(dqp
);
1268 xfs_qm_dqunpin_wait(dqp
);
1271 * Get the buffer containing the on-disk dquot
1273 error
= xfs_trans_read_buf(mp
, NULL
, mp
->m_ddev_targp
, dqp
->q_blkno
,
1274 mp
->m_quotainfo
->qi_dqchunklen
, XBF_TRYLOCK
,
1275 &bp
, &xfs_dquot_buf_ops
);
1276 if (error
== -EAGAIN
)
1278 if (xfs_metadata_is_sick(error
))
1279 xfs_dquot_mark_sick(dqp
);
1283 fa
= xfs_qm_dqflush_check(dqp
);
1285 xfs_alert(mp
, "corrupt dquot ID 0x%x in memory at %pS",
1288 xfs_dquot_mark_sick(dqp
);
1289 error
= -EFSCORRUPTED
;
1293 /* Flush the incore dquot to the ondisk buffer. */
1294 dqblk
= xfs_buf_offset(bp
, dqp
->q_bufoffset
);
1295 xfs_dquot_to_disk(&dqblk
->dd_diskdq
, dqp
);
1298 * Clear the dirty field and remember the flush lsn for later use.
1300 dqp
->q_flags
&= ~XFS_DQFLAG_DIRTY
;
1302 xfs_trans_ail_copy_lsn(mp
->m_ail
, &dqp
->q_logitem
.qli_flush_lsn
,
1303 &dqp
->q_logitem
.qli_item
.li_lsn
);
1306 * copy the lsn into the on-disk dquot now while we have the in memory
1307 * dquot here. This can't be done later in the write verifier as we
1308 * can't get access to the log item at that point in time.
1310 * We also calculate the CRC here so that the on-disk dquot in the
1311 * buffer always has a valid CRC. This ensures there is no possibility
1312 * of a dquot without an up-to-date CRC getting to disk.
1314 if (xfs_has_crc(mp
)) {
1315 dqblk
->dd_lsn
= cpu_to_be64(dqp
->q_logitem
.qli_item
.li_lsn
);
1316 xfs_update_cksum((char *)dqblk
, sizeof(struct xfs_dqblk
),
1321 * Attach the dquot to the buffer so that we can remove this dquot from
1322 * the AIL and release the flush lock once the dquot is synced to disk.
1324 bp
->b_flags
|= _XBF_DQUOTS
;
1325 list_add_tail(&dqp
->q_logitem
.qli_item
.li_bio_list
, &bp
->b_li_list
);
1328 * If the buffer is pinned then push on the log so we won't
1329 * get stuck waiting in the write for too long.
1331 if (xfs_buf_ispinned(bp
)) {
1332 trace_xfs_dqflush_force(dqp
);
1333 xfs_log_force(mp
, 0);
1336 trace_xfs_dqflush_done(dqp
);
1341 dqp
->q_flags
&= ~XFS_DQFLAG_DIRTY
;
1342 xfs_trans_ail_delete(lip
, 0);
1343 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
1350 * Lock two xfs_dquot structures.
1352 * To avoid deadlocks we always lock the quota structure with
1353 * the lowerd id first.
1357 struct xfs_dquot
*d1
,
1358 struct xfs_dquot
*d2
)
1362 if (d1
->q_id
> d2
->q_id
) {
1363 mutex_lock(&d2
->q_qlock
);
1364 mutex_lock_nested(&d1
->q_qlock
, XFS_QLOCK_NESTED
);
1366 mutex_lock(&d1
->q_qlock
);
1367 mutex_lock_nested(&d2
->q_qlock
, XFS_QLOCK_NESTED
);
1370 mutex_lock(&d1
->q_qlock
);
1372 mutex_lock(&d2
->q_qlock
);
1381 const struct xfs_dqtrx
*qa
= a
;
1382 const struct xfs_dqtrx
*qb
= b
;
1384 if (qa
->qt_dquot
->q_id
> qb
->qt_dquot
->q_id
)
1386 if (qa
->qt_dquot
->q_id
< qb
->qt_dquot
->q_id
)
1393 struct xfs_dqtrx
*q
)
1397 BUILD_BUG_ON(XFS_QM_TRANS_MAXDQS
> MAX_LOCKDEP_SUBCLASSES
);
1399 /* Sort in order of dquot id, do not allow duplicates */
1400 for (i
= 0; i
< XFS_QM_TRANS_MAXDQS
&& q
[i
].qt_dquot
!= NULL
; i
++) {
1403 for (j
= 0; j
< i
; j
++)
1404 ASSERT(q
[i
].qt_dquot
!= q
[j
].qt_dquot
);
1409 sort(q
, i
, sizeof(struct xfs_dqtrx
), xfs_dqtrx_cmp
, NULL
);
1411 mutex_lock(&q
[0].qt_dquot
->q_qlock
);
1412 for (i
= 1; i
< XFS_QM_TRANS_MAXDQS
&& q
[i
].qt_dquot
!= NULL
; i
++)
1413 mutex_lock_nested(&q
[i
].qt_dquot
->q_qlock
,
1414 XFS_QLOCK_NESTED
+ i
- 1);
1420 xfs_dquot_cache
= kmem_cache_create("xfs_dquot",
1421 sizeof(struct xfs_dquot
),
1423 if (!xfs_dquot_cache
)
1426 xfs_dqtrx_cache
= kmem_cache_create("xfs_dqtrx",
1427 sizeof(struct xfs_dquot_acct
),
1429 if (!xfs_dqtrx_cache
)
1430 goto out_free_dquot_cache
;
1434 out_free_dquot_cache
:
1435 kmem_cache_destroy(xfs_dquot_cache
);
1443 kmem_cache_destroy(xfs_dqtrx_cache
);
1444 kmem_cache_destroy(xfs_dquot_cache
);