2 * Copyright (c) 2000-2003 Silicon Graphics, Inc. All Rights Reserved.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms of version 2 of the GNU General Public License as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it would be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
12 * Further, this software is distributed without any warranty that it is
13 * free of the rightful claim of any third person regarding infringement
14 * or the like. Any license provided herein, whether implied or
15 * otherwise, applies only to this software file. Patent licenses, if
16 * any, provided herein do not apply to combinations of this program with
17 * other software, or any other product whatsoever.
19 * You should have received a copy of the GNU General Public License along
20 * with this program; if not, write the Free Software Foundation, Inc., 59
21 * Temple Place - Suite 330, Boston MA 02111-1307, USA.
23 * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
24 * Mountain View, CA 94043, or:
28 * For further information regarding this notice, see:
30 * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
37 #include "xfs_trans.h"
42 #include "xfs_alloc.h"
43 #include "xfs_dmapi.h"
44 #include "xfs_quota.h"
45 #include "xfs_mount.h"
46 #include "xfs_alloc_btree.h"
47 #include "xfs_bmap_btree.h"
48 #include "xfs_ialloc_btree.h"
49 #include "xfs_btree.h"
50 #include "xfs_ialloc.h"
51 #include "xfs_attr_sf.h"
52 #include "xfs_dir_sf.h"
53 #include "xfs_dir2_sf.h"
54 #include "xfs_dinode.h"
55 #include "xfs_inode.h"
58 #include "xfs_rtalloc.h"
59 #include "xfs_error.h"
60 #include "xfs_itable.h"
66 #include "xfs_buf_item.h"
67 #include "xfs_trans_space.h"
68 #include "xfs_trans_priv.h"
77 dquot hash-chain lock (hashlock)
78 xqm dquot freelist lock (freelistlock
79 mount's dquot list lock (mplistlock)
80 user dquot lock - lock ordering among dquots is based on the uid or gid
81 group dquot lock - similar to udquots. Between the two dquots, the udquot
82 has to be locked first.
83 pin lock - the dquot lock must be held to take this lock.
87 STATIC void xfs_qm_dqflush_done(xfs_buf_t *, xfs_dq_logitem_t *);
90 xfs_buftarg_t *xfs_dqerror_target;
93 int xfs_dqerror_mod = 33;
97 * Allocate and initialize a dquot. We don't always allocate fresh memory;
98 * we try to reclaim a free dquot if the number of incore dquots are above
100 * The only field inside the core that gets initialized at this point
101 * is the d_id field. The idea is to fill in the entire q_core
102 * when we read in the on disk dquot.
111 boolean_t brandnewdquot;
113 brandnewdquot = xfs_qm_dqalloc_incore(&dqp);
114 dqp->dq_flags = type;
115 INT_SET(dqp->q_core.d_id, ARCH_CONVERT, id);
119 * No need to re-initialize these if this is a reclaimed dquot.
122 dqp->dq_flnext = dqp->dq_flprev = dqp;
123 mutex_init(&dqp->q_qlock, MUTEX_DEFAULT, "xdq");
124 initnsema(&dqp->q_flock, 1, "fdq");
125 sv_init(&dqp->q_pinwait, SV_DEFAULT, "pdq");
127 #ifdef XFS_DQUOT_TRACE
128 dqp->q_trace = ktrace_alloc(DQUOT_TRACE_SIZE, KM_SLEEP);
129 xfs_dqtrace_entry(dqp, "DQINIT");
133 * Only the q_core portion was zeroed in dqreclaim_one().
134 * So, we need to reset others.
138 dqp->MPL_NEXT = dqp->HL_NEXT = NULL;
139 dqp->HL_PREVP = dqp->MPL_PREVP = NULL;
140 dqp->q_bufoffset = 0;
141 dqp->q_fileoffset = 0;
142 dqp->q_transp = NULL;
143 dqp->q_gdquot = NULL;
144 dqp->q_res_bcount = 0;
145 dqp->q_res_icount = 0;
146 dqp->q_res_rtbcount = 0;
149 ASSERT(dqp->dq_flnext == dqp->dq_flprev);
151 #ifdef XFS_DQUOT_TRACE
152 ASSERT(dqp->q_trace);
153 xfs_dqtrace_entry(dqp, "DQRECLAIMED_INIT");
158 * log item gets initialized later
164 * This is called to free all the memory associated with a dquot
170 ASSERT(! XFS_DQ_IS_ON_FREELIST(dqp));
172 mutex_destroy(&dqp->q_qlock);
173 freesema(&dqp->q_flock);
174 sv_destroy(&dqp->q_pinwait);
176 #ifdef XFS_DQUOT_TRACE
178 ktrace_free(dqp->q_trace);
181 kmem_zone_free(xfs_Gqm->qm_dqzone, dqp);
182 atomic_dec(&xfs_Gqm->qm_totaldquots);
186 * This is what a 'fresh' dquot inside a dquot chunk looks like on disk.
195 * Caller has zero'd the entire dquot 'chunk' already.
197 INT_SET(d->dd_diskdq.d_magic, ARCH_CONVERT, XFS_DQUOT_MAGIC);
198 INT_SET(d->dd_diskdq.d_version, ARCH_CONVERT, XFS_DQUOT_VERSION);
199 INT_SET(d->dd_diskdq.d_id, ARCH_CONVERT, id);
200 INT_SET(d->dd_diskdq.d_flags, ARCH_CONVERT, type);
204 #ifdef XFS_DQUOT_TRACE
206 * Dquot tracing for debugging.
216 xfs_dquot_t *udqp = NULL;
219 ASSERT(dqp->q_trace);
224 ktrace_enter(dqp->q_trace,
225 (void *)(__psint_t)DQUOT_KTRACE_ENTRY,
227 (void *)(__psint_t)dqp->q_nrefs,
228 (void *)(__psint_t)dqp->dq_flags,
229 (void *)(__psint_t)dqp->q_res_bcount,
230 (void *)(__psint_t)INT_GET(dqp->q_core.d_bcount,
232 (void *)(__psint_t)INT_GET(dqp->q_core.d_icount,
234 (void *)(__psint_t)INT_GET(dqp->q_core.d_blk_hardlimit,
236 (void *)(__psint_t)INT_GET(dqp->q_core.d_blk_softlimit,
238 (void *)(__psint_t)INT_GET(dqp->q_core.d_ino_hardlimit,
240 (void *)(__psint_t)INT_GET(dqp->q_core.d_ino_softlimit,
242 (void *)(__psint_t)INT_GET(dqp->q_core.d_id, ARCH_CONVERT),
243 (void *)(__psint_t)current_pid(),
244 (void *)(__psint_t)ino,
245 (void *)(__psint_t)retaddr,
246 (void *)(__psint_t)udqp);
253 * If default limits are in force, push them into the dquot now.
254 * We overwrite the dquot limits only if they are zero and this
255 * is not the root dquot.
258 xfs_qm_adjust_dqlimits(
262 xfs_quotainfo_t *q = mp->m_quotainfo;
266 if (q->qi_bsoftlimit && !d->d_blk_softlimit)
267 INT_SET(d->d_blk_softlimit, ARCH_CONVERT, q->qi_bsoftlimit);
268 if (q->qi_bhardlimit && !d->d_blk_hardlimit)
269 INT_SET(d->d_blk_hardlimit, ARCH_CONVERT, q->qi_bhardlimit);
270 if (q->qi_isoftlimit && !d->d_ino_softlimit)
271 INT_SET(d->d_ino_softlimit, ARCH_CONVERT, q->qi_isoftlimit);
272 if (q->qi_ihardlimit && !d->d_ino_hardlimit)
273 INT_SET(d->d_ino_hardlimit, ARCH_CONVERT, q->qi_ihardlimit);
274 if (q->qi_rtbsoftlimit && !d->d_rtb_softlimit)
275 INT_SET(d->d_rtb_softlimit, ARCH_CONVERT, q->qi_rtbsoftlimit);
276 if (q->qi_rtbhardlimit && !d->d_rtb_hardlimit)
277 INT_SET(d->d_rtb_hardlimit, ARCH_CONVERT, q->qi_rtbhardlimit);
281 * Check the limits and timers of a dquot and start or reset timers
283 * This gets called even when quota enforcement is OFF, which makes our
284 * life a little less complicated. (We just don't reject any quota
285 * reservations in that case, when enforcement is off).
286 * We also return 0 as the values of the timers in Q_GETQUOTA calls, when
288 * In contrast, warnings are a little different in that they don't
289 * 'automatically' get started when limits get exceeded.
292 xfs_qm_adjust_dqtimers(
299 if (INT_GET(d->d_blk_hardlimit, ARCH_CONVERT))
300 ASSERT(INT_GET(d->d_blk_softlimit, ARCH_CONVERT) <=
301 INT_GET(d->d_blk_hardlimit, ARCH_CONVERT));
302 if (INT_GET(d->d_ino_hardlimit, ARCH_CONVERT))
303 ASSERT(INT_GET(d->d_ino_softlimit, ARCH_CONVERT) <=
304 INT_GET(d->d_ino_hardlimit, ARCH_CONVERT));
305 if (INT_GET(d->d_rtb_hardlimit, ARCH_CONVERT))
306 ASSERT(INT_GET(d->d_rtb_softlimit, ARCH_CONVERT) <=
307 INT_GET(d->d_rtb_hardlimit, ARCH_CONVERT));
310 if ((INT_GET(d->d_blk_softlimit, ARCH_CONVERT) &&
311 (INT_GET(d->d_bcount, ARCH_CONVERT) >=
312 INT_GET(d->d_blk_softlimit, ARCH_CONVERT))) ||
313 (INT_GET(d->d_blk_hardlimit, ARCH_CONVERT) &&
314 (INT_GET(d->d_bcount, ARCH_CONVERT) >=
315 INT_GET(d->d_blk_hardlimit, ARCH_CONVERT)))) {
316 INT_SET(d->d_btimer, ARCH_CONVERT,
317 get_seconds() + XFS_QI_BTIMELIMIT(mp));
320 if ((!d->d_blk_softlimit ||
321 (INT_GET(d->d_bcount, ARCH_CONVERT) <
322 INT_GET(d->d_blk_softlimit, ARCH_CONVERT))) &&
323 (!d->d_blk_hardlimit ||
324 (INT_GET(d->d_bcount, ARCH_CONVERT) <
325 INT_GET(d->d_blk_hardlimit, ARCH_CONVERT)))) {
331 if ((INT_GET(d->d_ino_softlimit, ARCH_CONVERT) &&
332 (INT_GET(d->d_icount, ARCH_CONVERT) >=
333 INT_GET(d->d_ino_softlimit, ARCH_CONVERT))) ||
334 (INT_GET(d->d_ino_hardlimit, ARCH_CONVERT) &&
335 (INT_GET(d->d_icount, ARCH_CONVERT) >=
336 INT_GET(d->d_ino_hardlimit, ARCH_CONVERT)))) {
337 INT_SET(d->d_itimer, ARCH_CONVERT,
338 get_seconds() + XFS_QI_ITIMELIMIT(mp));
341 if ((!d->d_ino_softlimit ||
342 (INT_GET(d->d_icount, ARCH_CONVERT) <
343 INT_GET(d->d_ino_softlimit, ARCH_CONVERT))) &&
344 (!d->d_ino_hardlimit ||
345 (INT_GET(d->d_icount, ARCH_CONVERT) <
346 INT_GET(d->d_ino_hardlimit, ARCH_CONVERT)))) {
351 if (!d->d_rtbtimer) {
352 if ((INT_GET(d->d_rtb_softlimit, ARCH_CONVERT) &&
353 (INT_GET(d->d_rtbcount, ARCH_CONVERT) >=
354 INT_GET(d->d_rtb_softlimit, ARCH_CONVERT))) ||
355 (INT_GET(d->d_rtb_hardlimit, ARCH_CONVERT) &&
356 (INT_GET(d->d_rtbcount, ARCH_CONVERT) >=
357 INT_GET(d->d_rtb_hardlimit, ARCH_CONVERT)))) {
358 INT_SET(d->d_rtbtimer, ARCH_CONVERT,
359 get_seconds() + XFS_QI_RTBTIMELIMIT(mp));
362 if ((!d->d_rtb_softlimit ||
363 (INT_GET(d->d_rtbcount, ARCH_CONVERT) <
364 INT_GET(d->d_rtb_softlimit, ARCH_CONVERT))) &&
365 (!d->d_rtb_hardlimit ||
366 (INT_GET(d->d_rtbcount, ARCH_CONVERT) <
367 INT_GET(d->d_rtb_hardlimit, ARCH_CONVERT)))) {
374 * initialize a buffer full of dquots and log the whole thing
377 xfs_qm_init_dquot_blk(
388 ASSERT(XFS_BUF_ISBUSY(bp));
389 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
391 d = (xfs_dqblk_t *)XFS_BUF_PTR(bp);
394 * ID of the first dquot in the block - id's are zero based.
396 curid = id - (id % XFS_QM_DQPERBLK(mp));
398 memset(d, 0, BBTOB(XFS_QI_DQCHUNKLEN(mp)));
399 for (i = 0; i < XFS_QM_DQPERBLK(mp); i++, d++, curid++)
400 xfs_qm_dqinit_core(curid, type, d);
401 xfs_trans_dquot_buf(tp, bp,
405 xfs_trans_log_buf(tp, bp, 0, BBTOB(XFS_QI_DQCHUNKLEN(mp)) - 1);
411 * Allocate a block and fill it with dquots.
412 * This is called when the bmapi finds a hole.
420 xfs_fileoff_t offset_fsb,
423 xfs_fsblock_t firstblock;
424 xfs_bmap_free_t flist;
426 int nmaps, error, committed;
430 xfs_dqtrace_entry(dqp, "DQALLOC");
433 * Initialize the bmap freelist prior to calling bmapi code.
435 XFS_BMAP_INIT(&flist, &firstblock);
436 xfs_ilock(quotip, XFS_ILOCK_EXCL);
438 * Return if this type of quotas is turned off while we didn't
441 if (XFS_IS_THIS_QUOTA_OFF(dqp)) {
442 xfs_iunlock(quotip, XFS_ILOCK_EXCL);
447 * xfs_trans_commit normally decrements the vnode ref count
448 * when it unlocks the inode. Since we want to keep the quota
449 * inode around, we bump the vnode ref count now.
451 VN_HOLD(XFS_ITOV(quotip));
453 xfs_trans_ijoin(tp, quotip, XFS_ILOCK_EXCL);
455 if ((error = xfs_bmapi(tp, quotip,
456 offset_fsb, XFS_DQUOT_CLUSTER_SIZE_FSB,
457 XFS_BMAPI_METADATA | XFS_BMAPI_WRITE,
459 XFS_QM_DQALLOC_SPACE_RES(mp),
460 &map, &nmaps, &flist))) {
463 ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB);
465 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
466 (map.br_startblock != HOLESTARTBLOCK));
469 * Keep track of the blkno to save a lookup later
471 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
473 /* now we can just get the buffer (there's nothing to read yet) */
474 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
476 XFS_QI_DQCHUNKLEN(mp),
478 if (!bp || (error = XFS_BUF_GETERROR(bp)))
481 * Make a chunk of dquots out of this buffer and log
484 xfs_qm_init_dquot_blk(tp, mp, INT_GET(dqp->q_core.d_id, ARCH_CONVERT),
485 dqp->dq_flags & (XFS_DQ_USER|XFS_DQ_GROUP),
488 if ((error = xfs_bmap_finish(&tp, &flist, firstblock, &committed))) {
496 xfs_bmap_cancel(&flist);
498 xfs_iunlock(quotip, XFS_ILOCK_EXCL);
504 * Maps a dquot to the buffer containing its on-disk version.
505 * This returns a ptr to the buffer containing the on-disk dquot
506 * in the bpp param, and a ptr to the on-disk dquot within that buffer
512 xfs_disk_dquot_t **O_ddpp,
521 xfs_disk_dquot_t *ddq;
526 id = INT_GET(dqp->q_core.d_id, ARCH_CONVERT);
531 * If we don't know where the dquot lives, find out.
533 if (dqp->q_blkno == (xfs_daddr_t) 0) {
534 /* We use the id as an index */
535 dqp->q_fileoffset = (xfs_fileoff_t) ((uint)id /
536 XFS_QM_DQPERBLK(mp));
538 quotip = XFS_DQ_TO_QIP(dqp);
539 xfs_ilock(quotip, XFS_ILOCK_SHARED);
541 * Return if this type of quotas is turned off while we didn't
544 if (XFS_IS_THIS_QUOTA_OFF(dqp)) {
545 xfs_iunlock(quotip, XFS_ILOCK_SHARED);
549 * Find the block map; no allocations yet
551 error = xfs_bmapi(NULL, quotip, dqp->q_fileoffset,
552 XFS_DQUOT_CLUSTER_SIZE_FSB,
554 NULL, 0, &map, &nmaps, NULL);
556 xfs_iunlock(quotip, XFS_ILOCK_SHARED);
560 ASSERT(map.br_blockcount == 1);
563 * offset of dquot in the (fixed sized) dquot chunk.
565 dqp->q_bufoffset = (id % XFS_QM_DQPERBLK(mp)) *
567 if (map.br_startblock == HOLESTARTBLOCK) {
569 * We don't allocate unless we're asked to
571 if (!(flags & XFS_QMOPT_DQALLOC))
575 if ((error = xfs_qm_dqalloc(tp, mp, dqp, quotip,
576 dqp->q_fileoffset, &bp)))
581 * store the blkno etc so that we don't have to do the
582 * mapping all the time
584 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
587 ASSERT(dqp->q_blkno != DELAYSTARTBLOCK);
588 ASSERT(dqp->q_blkno != HOLESTARTBLOCK);
591 * Read in the buffer, unless we've just done the allocation
592 * (in which case we already have the buf).
595 xfs_dqtrace_entry(dqp, "DQTOBP READBUF");
596 if ((error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
598 XFS_QI_DQCHUNKLEN(mp),
603 return XFS_ERROR(error);
605 ASSERT(XFS_BUF_ISBUSY(bp));
606 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
609 * calculate the location of the dquot inside the buffer.
611 ddq = (xfs_disk_dquot_t *)((char *)XFS_BUF_PTR(bp) + dqp->q_bufoffset);
614 * A simple sanity check in case we got a corrupted dquot...
616 if (xfs_qm_dqcheck(ddq, id,
617 dqp->dq_flags & (XFS_DQ_USER|XFS_DQ_GROUP),
618 flags & (XFS_QMOPT_DQREPAIR|XFS_QMOPT_DOWARN),
620 if (!(flags & XFS_QMOPT_DQREPAIR)) {
621 xfs_trans_brelse(tp, bp);
622 return XFS_ERROR(EIO);
624 XFS_BUF_BUSY(bp); /* We dirtied this */
635 * Read in the ondisk dquot using dqtobp() then copy it to an incore version,
636 * and release the buffer immediately.
644 xfs_dquot_t *dqp, /* dquot to get filled in */
647 xfs_disk_dquot_t *ddqp;
652 * get a pointer to the on-disk dquot and the buffer containing it
653 * dqp already knows its own type (GROUP/USER).
655 xfs_dqtrace_entry(dqp, "DQREAD");
656 if ((error = xfs_qm_dqtobp(tp, dqp, &ddqp, &bp, flags))) {
660 /* copy everything from disk dquot to the incore dquot */
661 memcpy(&dqp->q_core, ddqp, sizeof(xfs_disk_dquot_t));
662 ASSERT(INT_GET(dqp->q_core.d_id, ARCH_CONVERT) == id);
663 xfs_qm_dquot_logitem_init(dqp);
666 * Reservation counters are defined as reservation plus current usage
667 * to avoid having to add everytime.
669 dqp->q_res_bcount = INT_GET(ddqp->d_bcount, ARCH_CONVERT);
670 dqp->q_res_icount = INT_GET(ddqp->d_icount, ARCH_CONVERT);
671 dqp->q_res_rtbcount = INT_GET(ddqp->d_rtbcount, ARCH_CONVERT);
673 /* Mark the buf so that this will stay incore a little longer */
674 XFS_BUF_SET_VTYPE_REF(bp, B_FS_DQUOT, XFS_DQUOT_REF);
677 * We got the buffer with a xfs_trans_read_buf() (in dqtobp())
678 * So we need to release with xfs_trans_brelse().
679 * The strategy here is identical to that of inodes; we lock
680 * the dquot in xfs_qm_dqget() before making it accessible to
681 * others. This is because dquots, like inodes, need a good level of
682 * concurrency, and we don't want to take locks on the entire buffers
683 * for dquot accesses.
684 * Note also that the dquot buffer may even be dirty at this point, if
685 * this particular dquot was repaired. We still aren't afraid to
686 * brelse it because we have the changes incore.
688 ASSERT(XFS_BUF_ISBUSY(bp));
689 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
690 xfs_trans_brelse(tp, bp);
697 * allocate an incore dquot from the kernel heap,
698 * and fill its core with quota information kept on disk.
699 * If XFS_QMOPT_DQALLOC is set, it'll allocate a dquot on disk
700 * if it wasn't already allocated.
705 xfs_dqid_t id, /* gid or uid, depending on type */
706 uint type, /* UDQUOT or GDQUOT */
707 uint flags, /* DQALLOC, DQREPAIR */
708 xfs_dquot_t **O_dqpp)/* OUT : incore dquot, not locked */
715 dqp = xfs_qm_dqinit(mp, id, type);
717 if (flags & XFS_QMOPT_DQALLOC) {
718 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_DQALLOC);
719 if ((error = xfs_trans_reserve(tp,
720 XFS_QM_DQALLOC_SPACE_RES(mp),
721 XFS_WRITE_LOG_RES(mp) +
722 BBTOB(XFS_QI_DQCHUNKLEN(mp)) - 1 +
725 XFS_TRANS_PERM_LOG_RES,
726 XFS_WRITE_LOG_COUNT))) {
730 cancelflags = XFS_TRANS_RELEASE_LOG_RES;
734 * Read it from disk; xfs_dqread() takes care of
735 * all the necessary initialization of dquot's fields (locks, etc)
737 if ((error = xfs_qm_dqread(tp, id, dqp, flags))) {
739 * This can happen if quotas got turned off (ESRCH),
740 * or if the dquot didn't exist on disk and we ask to
743 xfs_dqtrace_entry(dqp, "DQREAD FAIL");
744 cancelflags |= XFS_TRANS_ABORT;
748 if ((error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES,
759 xfs_trans_cancel(tp, cancelflags);
761 xfs_qm_dqdestroy(dqp);
767 * Lookup a dquot in the incore dquot hashtable. We keep two separate
768 * hashtables for user and group dquots; and, these are global tables
769 * inside the XQM, not per-filesystem tables.
770 * The hash chain must be locked by caller, and it is left locked
771 * on return. Returning dquot is locked.
778 xfs_dquot_t **O_dqpp)
784 ASSERT(XFS_DQ_IS_HASH_LOCKED(qh));
786 flist_locked = B_FALSE;
789 * Traverse the hashchain looking for a match
791 for (dqp = qh->qh_next; dqp != NULL; dqp = dqp->HL_NEXT) {
793 * We already have the hashlock. We don't need the
794 * dqlock to look at the id field of the dquot, since the
795 * id can't be modified without the hashlock anyway.
797 if (INT_GET(dqp->q_core.d_id, ARCH_CONVERT) == id && dqp->q_mount == mp) {
798 xfs_dqtrace_entry(dqp, "DQFOUND BY LOOKUP");
800 * All in core dquots must be on the dqlist of mp
802 ASSERT(dqp->MPL_PREVP != NULL);
805 if (dqp->q_nrefs == 0) {
806 ASSERT (XFS_DQ_IS_ON_FREELIST(dqp));
807 if (! xfs_qm_freelist_lock_nowait(xfs_Gqm)) {
808 xfs_dqtrace_entry(dqp, "DQLOOKUP: WANT");
811 * We may have raced with dqreclaim_one()
812 * (and lost). So, flag that we don't
813 * want the dquot to be reclaimed.
815 dqp->dq_flags |= XFS_DQ_WANT;
817 xfs_qm_freelist_lock(xfs_Gqm);
819 dqp->dq_flags &= ~(XFS_DQ_WANT);
821 flist_locked = B_TRUE;
825 * id couldn't have changed; we had the hashlock all
828 ASSERT(INT_GET(dqp->q_core.d_id, ARCH_CONVERT) == id);
831 if (dqp->q_nrefs != 0) {
832 xfs_qm_freelist_unlock(xfs_Gqm);
833 flist_locked = B_FALSE;
836 * take it off the freelist
838 xfs_dqtrace_entry(dqp,
839 "DQLOOKUP: TAKEOFF FL");
840 XQM_FREELIST_REMOVE(dqp);
841 /* xfs_qm_freelist_print(&(xfs_Gqm->
853 xfs_qm_freelist_unlock(xfs_Gqm);
855 * move the dquot to the front of the hashchain
857 ASSERT(XFS_DQ_IS_HASH_LOCKED(qh));
858 if (dqp->HL_PREVP != &qh->qh_next) {
859 xfs_dqtrace_entry(dqp,
860 "DQLOOKUP: HASH MOVETOFRONT");
861 if ((d = dqp->HL_NEXT))
862 d->HL_PREVP = dqp->HL_PREVP;
863 *(dqp->HL_PREVP) = d;
865 d->HL_PREVP = &dqp->HL_NEXT;
867 dqp->HL_PREVP = &qh->qh_next;
870 xfs_dqtrace_entry(dqp, "LOOKUP END");
872 ASSERT(XFS_DQ_IS_HASH_LOCKED(qh));
878 ASSERT(XFS_DQ_IS_HASH_LOCKED(qh));
883 * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a
884 * a locked dquot, doing an allocation (if requested) as needed.
885 * When both an inode and an id are given, the inode's id takes precedence.
886 * That is, if the id changes while we don't hold the ilock inside this
887 * function, the new dquot is returned, not necessarily the one requested
888 * in the id argument.
893 xfs_inode_t *ip, /* locked inode (optional) */
894 xfs_dqid_t id, /* gid or uid, depending on type */
895 uint type, /* UDQUOT or GDQUOT */
896 uint flags, /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */
897 xfs_dquot_t **O_dqpp) /* OUT : locked incore dquot */
904 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
905 if ((! XFS_IS_UQUOTA_ON(mp) && type == XFS_DQ_USER) ||
906 (! XFS_IS_GQUOTA_ON(mp) && type == XFS_DQ_GROUP)) {
909 h = XFS_DQ_HASH(mp, id, type);
912 if (xfs_do_dqerror) {
913 if ((xfs_dqerror_target == mp->m_ddev_targp) &&
914 (xfs_dqreq_num++ % xfs_dqerror_mod) == 0) {
915 cmn_err(CE_DEBUG, "Returning error in dqget");
924 ASSERT(type == XFS_DQ_USER || type == XFS_DQ_GROUP);
926 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip));
927 if (type == XFS_DQ_USER)
928 ASSERT(ip->i_udquot == NULL);
930 ASSERT(ip->i_gdquot == NULL);
936 * Look in the cache (hashtable).
937 * The chain is kept locked during lookup.
939 if (xfs_qm_dqlookup(mp, id, h, O_dqpp) == 0) {
940 XQM_STATS_INC(xqmstats.xs_qm_dqcachehits);
942 * The dquot was found, moved to the front of the chain,
943 * taken off the freelist if it was on it, and locked
944 * at this point. Just unlock the hashchain and return.
947 ASSERT(XFS_DQ_IS_LOCKED(*O_dqpp));
948 XFS_DQ_HASH_UNLOCK(h);
949 xfs_dqtrace_entry(*O_dqpp, "DQGET DONE (FROM CACHE)");
950 return (0); /* success */
952 XQM_STATS_INC(xqmstats.xs_qm_dqcachemisses);
955 * Dquot cache miss. We don't want to keep the inode lock across
956 * a (potential) disk read. Also we don't want to deal with the lock
957 * ordering between quotainode and this inode. OTOH, dropping the inode
958 * lock here means dealing with a chown that can happen before
959 * we re-acquire the lock.
962 xfs_iunlock(ip, XFS_ILOCK_EXCL);
964 * Save the hashchain version stamp, and unlock the chain, so that
965 * we don't keep the lock across a disk read
967 version = h->qh_version;
968 XFS_DQ_HASH_UNLOCK(h);
971 * Allocate the dquot on the kernel heap, and read the ondisk
972 * portion off the disk. Also, do all the necessary initialization
973 * This can return ENOENT if dquot didn't exist on disk and we didn't
974 * ask it to allocate; ESRCH if quotas got turned off suddenly.
976 if ((error = xfs_qm_idtodq(mp, id, type,
977 flags & (XFS_QMOPT_DQALLOC|XFS_QMOPT_DQREPAIR|
981 xfs_ilock(ip, XFS_ILOCK_EXCL);
986 * See if this is mount code calling to look at the overall quota limits
987 * which are stored in the id == 0 user or group's dquot.
988 * Since we may not have done a quotacheck by this point, just return
989 * the dquot without attaching it to any hashtables, lists, etc, or even
990 * taking a reference.
991 * The caller must dqdestroy this once done.
993 if (flags & XFS_QMOPT_DQSUSER) {
1000 * Dquot lock comes after hashlock in the lock ordering
1003 xfs_ilock(ip, XFS_ILOCK_EXCL);
1004 if (! XFS_IS_DQTYPE_ON(mp, type)) {
1005 /* inode stays locked on return */
1006 xfs_qm_dqdestroy(dqp);
1007 return XFS_ERROR(ESRCH);
1010 * A dquot could be attached to this inode by now, since
1011 * we had dropped the ilock.
1013 if (type == XFS_DQ_USER) {
1015 xfs_qm_dqdestroy(dqp);
1022 xfs_qm_dqdestroy(dqp);
1031 * Hashlock comes after ilock in lock order
1033 XFS_DQ_HASH_LOCK(h);
1034 if (version != h->qh_version) {
1035 xfs_dquot_t *tmpdqp;
1037 * Now, see if somebody else put the dquot in the
1038 * hashtable before us. This can happen because we didn't
1039 * keep the hashchain lock. We don't have to worry about
1040 * lock order between the two dquots here since dqp isn't
1041 * on any findable lists yet.
1043 if (xfs_qm_dqlookup(mp, id, h, &tmpdqp) == 0) {
1045 * Duplicate found. Just throw away the new dquot
1048 xfs_qm_dqput(tmpdqp);
1049 XFS_DQ_HASH_UNLOCK(h);
1050 xfs_qm_dqdestroy(dqp);
1051 XQM_STATS_INC(xqmstats.xs_qm_dquot_dups);
1057 * Put the dquot at the beginning of the hash-chain and mp's list
1058 * LOCK ORDER: hashlock, freelistlock, mplistlock, udqlock, gdqlock ..
1060 ASSERT(XFS_DQ_IS_HASH_LOCKED(h));
1062 XQM_HASHLIST_INSERT(h, dqp);
1065 * Attach this dquot to this filesystem's list of all dquots,
1066 * kept inside the mount structure in m_quotainfo field
1068 xfs_qm_mplist_lock(mp);
1071 * We return a locked dquot to the caller, with a reference taken
1076 XQM_MPLIST_INSERT(&(XFS_QI_MPL_LIST(mp)), dqp);
1078 xfs_qm_mplist_unlock(mp);
1079 XFS_DQ_HASH_UNLOCK(h);
1081 ASSERT((ip == NULL) || XFS_ISLOCKED_INODE_EXCL(ip));
1082 xfs_dqtrace_entry(dqp, "DQGET DONE");
1089 * Release a reference to the dquot (decrement ref-count)
1090 * and unlock it. If there is a group quota attached to this
1091 * dquot, carefully release that too without tripping over
1092 * deadlocks'n'stuff.
1100 ASSERT(dqp->q_nrefs > 0);
1101 ASSERT(XFS_DQ_IS_LOCKED(dqp));
1102 xfs_dqtrace_entry(dqp, "DQPUT");
1104 if (dqp->q_nrefs != 1) {
1111 * drop the dqlock and acquire the freelist and dqlock
1112 * in the right order; but try to get it out-of-order first
1114 if (! xfs_qm_freelist_lock_nowait(xfs_Gqm)) {
1115 xfs_dqtrace_entry(dqp, "DQPUT: FLLOCK-WAIT");
1117 xfs_qm_freelist_lock(xfs_Gqm);
1124 /* We can't depend on nrefs being == 1 here */
1125 if (--dqp->q_nrefs == 0) {
1126 xfs_dqtrace_entry(dqp, "DQPUT: ON FREELIST");
1128 * insert at end of the freelist.
1130 XQM_FREELIST_INSERT(&(xfs_Gqm->qm_dqfreelist), dqp);
1133 * If we just added a udquot to the freelist, then
1134 * we want to release the gdquot reference that
1135 * it (probably) has. Otherwise it'll keep the
1136 * gdquot from getting reclaimed.
1138 if ((gdqp = dqp->q_gdquot)) {
1140 * Avoid a recursive dqput call
1143 dqp->q_gdquot = NULL;
1146 /* xfs_qm_freelist_print(&(xfs_Gqm->qm_dqfreelist),
1147 "@@@@@++ Free list (after append) @@@@@+");
1153 * If we had a group quota inside the user quota as a hint,
1160 xfs_qm_freelist_unlock(xfs_Gqm);
1164 * Release a dquot. Flush it if dirty, then dqput() it.
1165 * dquot must not be locked.
1172 xfs_dqtrace_entry(dqp, "DQRELE");
1176 * We don't care to flush it if the dquot is dirty here.
1177 * That will create stutters that we want to avoid.
1178 * Instead we do a delayed write when we try to reclaim
1179 * a dirty dquot. Also xfs_sync will take part of the burden...
1186 * Write a modified dquot to disk.
1187 * The dquot must be locked and the flush lock too taken by caller.
1188 * The flush lock will not be unlocked until the dquot reaches the disk,
1189 * but the dquot is free to be unlocked and modified by the caller
1190 * in the interim. Dquot is still locked on return. This behavior is
1191 * identical to that of inodes.
1200 xfs_disk_dquot_t *ddqp;
1204 ASSERT(XFS_DQ_IS_LOCKED(dqp));
1205 ASSERT(XFS_DQ_IS_FLUSH_LOCKED(dqp));
1206 xfs_dqtrace_entry(dqp, "DQFLUSH");
1209 * If not dirty, nada.
1211 if (!XFS_DQ_IS_DIRTY(dqp)) {
1217 * Cant flush a pinned dquot. Wait for it.
1219 xfs_qm_dqunpin_wait(dqp);
1222 * This may have been unpinned because the filesystem is shutting
1223 * down forcibly. If that's the case we must not write this dquot
1224 * to disk, because the log record didn't make it to disk!
1226 if (XFS_FORCED_SHUTDOWN(dqp->q_mount)) {
1227 dqp->dq_flags &= ~(XFS_DQ_DIRTY);
1229 return XFS_ERROR(EIO);
1233 * Get the buffer containing the on-disk dquot
1234 * We don't need a transaction envelope because we know that the
1235 * the ondisk-dquot has already been allocated for.
1237 if ((error = xfs_qm_dqtobp(NULL, dqp, &ddqp, &bp, XFS_QMOPT_DOWARN))) {
1238 xfs_dqtrace_entry(dqp, "DQTOBP FAIL");
1239 ASSERT(error != ENOENT);
1241 * Quotas could have gotten turned off (ESRCH)
1247 if (xfs_qm_dqcheck(&dqp->q_core, INT_GET(ddqp->d_id, ARCH_CONVERT), 0, XFS_QMOPT_DOWARN,
1248 "dqflush (incore copy)")) {
1249 xfs_force_shutdown(dqp->q_mount, XFS_CORRUPT_INCORE);
1250 return XFS_ERROR(EIO);
1253 /* This is the only portion of data that needs to persist */
1254 memcpy(ddqp, &(dqp->q_core), sizeof(xfs_disk_dquot_t));
1257 * Clear the dirty field and remember the flush lsn for later use.
1259 dqp->dq_flags &= ~(XFS_DQ_DIRTY);
1262 /* lsn is 64 bits */
1264 dqp->q_logitem.qli_flush_lsn = dqp->q_logitem.qli_item.li_lsn;
1268 * Attach an iodone routine so that we can remove this dquot from the
1269 * AIL and release the flush lock once the dquot is synced to disk.
1271 xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t *, xfs_log_item_t *))
1272 xfs_qm_dqflush_done, &(dqp->q_logitem.qli_item));
1274 * If the buffer is pinned then push on the log so we won't
1275 * get stuck waiting in the write for too long.
1277 if (XFS_BUF_ISPINNED(bp)) {
1278 xfs_dqtrace_entry(dqp, "DQFLUSH LOG FORCE");
1279 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
1282 if (flags & XFS_QMOPT_DELWRI) {
1283 xfs_bdwrite(mp, bp);
1284 } else if (flags & XFS_QMOPT_ASYNC) {
1285 xfs_bawrite(mp, bp);
1287 error = xfs_bwrite(mp, bp);
1289 xfs_dqtrace_entry(dqp, "DQFLUSH END");
1291 * dqp is still locked, but caller is free to unlock it now.
1298 * This is the dquot flushing I/O completion routine. It is called
1299 * from interrupt level when the buffer containing the dquot is
1300 * flushed to disk. It is responsible for removing the dquot logitem
1301 * from the AIL if it has not been re-logged, and unlocking the dquot's
1302 * flush lock. This behavior is very similar to that of inodes..
1306 xfs_qm_dqflush_done(
1308 xfs_dq_logitem_t *qip)
1313 dqp = qip->qli_dquot;
1316 * We only want to pull the item from the AIL if its
1317 * location in the log has not changed since we started the flush.
1318 * Thus, we only bother if the dquot's lsn has
1319 * not changed. First we check the lsn outside the lock
1320 * since it's cheaper, and then we recheck while
1321 * holding the lock before removing the dquot from the AIL.
1323 if ((qip->qli_item.li_flags & XFS_LI_IN_AIL) &&
1324 qip->qli_item.li_lsn == qip->qli_flush_lsn) {
1326 AIL_LOCK(dqp->q_mount, s);
1328 * xfs_trans_delete_ail() drops the AIL lock.
1330 if (qip->qli_item.li_lsn == qip->qli_flush_lsn)
1331 xfs_trans_delete_ail(dqp->q_mount,
1332 (xfs_log_item_t*)qip, s);
1334 AIL_UNLOCK(dqp->q_mount, s);
1338 * Release the dq's flush lock since we're done with it.
1345 xfs_qm_dqflock_nowait(
1350 locked = cpsema(&((dqp)->q_flock));
1352 /* XXX ifdef these out */
1354 (dqp)->dq_flags |= XFS_DQ_FLOCKED;
1360 xfs_qm_dqlock_nowait(
1363 return (mutex_trylock(&((dqp)->q_qlock)));
1370 mutex_lock(&(dqp->q_qlock), PINOD);
1377 mutex_unlock(&(dqp->q_qlock));
1378 if (dqp->q_logitem.qli_dquot == dqp) {
1379 /* Once was dqp->q_mount, but might just have been cleared */
1380 xfs_trans_unlocked_item(dqp->q_logitem.qli_item.li_mountp,
1381 (xfs_log_item_t*)&(dqp->q_logitem));
1387 xfs_dqunlock_nonotify(
1390 mutex_unlock(&(dqp->q_qlock));
1400 if (INT_GET(d1->q_core.d_id, ARCH_CONVERT) > INT_GET(d2->q_core.d_id, ARCH_CONVERT)) {
1418 * Take a dquot out of the mount's dqlist as well as the hashlist.
1419 * This is called via unmount as well as quotaoff, and the purge
1420 * will always succeed unless there are soft (temp) references
1423 * This returns 0 if it was purged, 1 if it wasn't. It's not an error code
1424 * that we're returning! XXXsup - not cool.
1432 xfs_dqhash_t *thishash;
1437 ASSERT(XFS_QM_IS_MPLIST_LOCKED(mp));
1438 ASSERT(XFS_DQ_IS_HASH_LOCKED(dqp->q_hash));
1442 * We really can't afford to purge a dquot that is
1443 * referenced, because these are hard refs.
1444 * It shouldn't happen in general because we went thru _all_ inodes in
1445 * dqrele_all_inodes before calling this and didn't let the mountlock go.
1446 * However it is possible that we have dquots with temporary
1447 * references that are not attached to an inode. e.g. see xfs_setattr().
1449 if (dqp->q_nrefs != 0) {
1451 XFS_DQ_HASH_UNLOCK(dqp->q_hash);
1455 ASSERT(XFS_DQ_IS_ON_FREELIST(dqp));
1458 * If we're turning off quotas, we have to make sure that, for
1459 * example, we don't delete quota disk blocks while dquots are
1460 * in the process of getting written to those disk blocks.
1461 * This dquot might well be on AIL, and we can't leave it there
1462 * if we're turning off quotas. Basically, we need this flush
1463 * lock, and are willing to block on it.
1465 if (! xfs_qm_dqflock_nowait(dqp)) {
1467 * Block on the flush lock after nudging dquot buffer,
1470 xfs_qm_dqflock_pushbuf_wait(dqp);
1474 * XXXIf we're turning this type of quotas off, we don't care
1475 * about the dirty metadata sitting in this dquot. OTOH, if
1476 * we're unmounting, we do care, so we flush it and wait.
1478 if (XFS_DQ_IS_DIRTY(dqp)) {
1479 xfs_dqtrace_entry(dqp, "DQPURGE ->DQFLUSH: DQDIRTY");
1480 /* dqflush unlocks dqflock */
1482 * Given that dqpurge is a very rare occurrence, it is OK
1483 * that we're holding the hashlist and mplist locks
1484 * across the disk write. But, ... XXXsup
1486 * We don't care about getting disk errors here. We need
1487 * to purge this dquot anyway, so we go ahead regardless.
1489 (void) xfs_qm_dqflush(dqp, XFS_QMOPT_SYNC);
1492 ASSERT(dqp->q_pincount == 0);
1493 ASSERT(XFS_FORCED_SHUTDOWN(mp) ||
1494 !(dqp->q_logitem.qli_item.li_flags & XFS_LI_IN_AIL));
1496 thishash = dqp->q_hash;
1497 XQM_HASHLIST_REMOVE(thishash, dqp);
1498 XQM_MPLIST_REMOVE(&(XFS_QI_MPL_LIST(mp)), dqp);
1500 * XXX Move this to the front of the freelist, if we can get the
1503 ASSERT(XFS_DQ_IS_ON_FREELIST(dqp));
1505 dqp->q_mount = NULL;
1507 dqp->dq_flags = XFS_DQ_INACTIVE;
1508 memset(&dqp->q_core, 0, sizeof(dqp->q_core));
1511 XFS_DQ_HASH_UNLOCK(thishash);
1518 xfs_qm_dqprint(xfs_dquot_t *dqp)
1520 cmn_err(CE_DEBUG, "-----------KERNEL DQUOT----------------");
1521 cmn_err(CE_DEBUG, "---- dquotID = %d",
1522 (int)INT_GET(dqp->q_core.d_id, ARCH_CONVERT));
1523 cmn_err(CE_DEBUG, "---- type = %s",
1524 XFS_QM_ISUDQ(dqp) ? "USR" : "GRP");
1525 cmn_err(CE_DEBUG, "---- fs = 0x%p", dqp->q_mount);
1526 cmn_err(CE_DEBUG, "---- blkno = 0x%x", (int) dqp->q_blkno);
1527 cmn_err(CE_DEBUG, "---- boffset = 0x%x", (int) dqp->q_bufoffset);
1528 cmn_err(CE_DEBUG, "---- blkhlimit = %Lu (0x%x)",
1529 INT_GET(dqp->q_core.d_blk_hardlimit, ARCH_CONVERT),
1530 (int) INT_GET(dqp->q_core.d_blk_hardlimit, ARCH_CONVERT));
1531 cmn_err(CE_DEBUG, "---- blkslimit = %Lu (0x%x)",
1532 INT_GET(dqp->q_core.d_blk_softlimit, ARCH_CONVERT),
1533 (int)INT_GET(dqp->q_core.d_blk_softlimit, ARCH_CONVERT));
1534 cmn_err(CE_DEBUG, "---- inohlimit = %Lu (0x%x)",
1535 INT_GET(dqp->q_core.d_ino_hardlimit, ARCH_CONVERT),
1536 (int)INT_GET(dqp->q_core.d_ino_hardlimit, ARCH_CONVERT));
1537 cmn_err(CE_DEBUG, "---- inoslimit = %Lu (0x%x)",
1538 INT_GET(dqp->q_core.d_ino_softlimit, ARCH_CONVERT),
1539 (int)INT_GET(dqp->q_core.d_ino_softlimit, ARCH_CONVERT));
1540 cmn_err(CE_DEBUG, "---- bcount = %Lu (0x%x)",
1541 INT_GET(dqp->q_core.d_bcount, ARCH_CONVERT),
1542 (int)INT_GET(dqp->q_core.d_bcount, ARCH_CONVERT));
1543 cmn_err(CE_DEBUG, "---- icount = %Lu (0x%x)",
1544 INT_GET(dqp->q_core.d_icount, ARCH_CONVERT),
1545 (int)INT_GET(dqp->q_core.d_icount, ARCH_CONVERT));
1546 cmn_err(CE_DEBUG, "---- btimer = %d",
1547 (int)INT_GET(dqp->q_core.d_btimer, ARCH_CONVERT));
1548 cmn_err(CE_DEBUG, "---- itimer = %d",
1549 (int)INT_GET(dqp->q_core.d_itimer, ARCH_CONVERT));
1550 cmn_err(CE_DEBUG, "---------------------------");
1555 * Give the buffer a little push if it is incore and
1556 * wait on the flush lock.
1559 xfs_qm_dqflock_pushbuf_wait(
1565 * Check to see if the dquot has been flushed delayed
1566 * write. If so, grab its buffer and send it
1567 * out immediately. We'll be able to acquire
1568 * the flush lock when the I/O completes.
1570 bp = xfs_incore(dqp->q_mount->m_ddev_targp, dqp->q_blkno,
1571 XFS_QI_DQCHUNKLEN(dqp->q_mount),
1572 XFS_INCORE_TRYLOCK);
1574 if (XFS_BUF_ISDELAYWRITE(bp)) {
1575 if (XFS_BUF_ISPINNED(bp)) {
1576 xfs_log_force(dqp->q_mount,
1580 xfs_bawrite(dqp->q_mount, bp);