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. They do
290 * get reset to zero, however, when we find the count to be under
291 * the soft limit (they are only ever set non-zero via userspace).
294 xfs_qm_adjust_dqtimers(
301 if (INT_GET(d->d_blk_hardlimit, ARCH_CONVERT))
302 ASSERT(INT_GET(d->d_blk_softlimit, ARCH_CONVERT) <=
303 INT_GET(d->d_blk_hardlimit, ARCH_CONVERT));
304 if (INT_GET(d->d_ino_hardlimit, ARCH_CONVERT))
305 ASSERT(INT_GET(d->d_ino_softlimit, ARCH_CONVERT) <=
306 INT_GET(d->d_ino_hardlimit, ARCH_CONVERT));
307 if (INT_GET(d->d_rtb_hardlimit, ARCH_CONVERT))
308 ASSERT(INT_GET(d->d_rtb_softlimit, ARCH_CONVERT) <=
309 INT_GET(d->d_rtb_hardlimit, ARCH_CONVERT));
312 if ((INT_GET(d->d_blk_softlimit, ARCH_CONVERT) &&
313 (INT_GET(d->d_bcount, ARCH_CONVERT) >=
314 INT_GET(d->d_blk_softlimit, ARCH_CONVERT))) ||
315 (INT_GET(d->d_blk_hardlimit, ARCH_CONVERT) &&
316 (INT_GET(d->d_bcount, ARCH_CONVERT) >=
317 INT_GET(d->d_blk_hardlimit, ARCH_CONVERT)))) {
318 INT_SET(d->d_btimer, ARCH_CONVERT,
319 get_seconds() + XFS_QI_BTIMELIMIT(mp));
324 if ((!d->d_blk_softlimit ||
325 (INT_GET(d->d_bcount, ARCH_CONVERT) <
326 INT_GET(d->d_blk_softlimit, ARCH_CONVERT))) &&
327 (!d->d_blk_hardlimit ||
328 (INT_GET(d->d_bcount, ARCH_CONVERT) <
329 INT_GET(d->d_blk_hardlimit, ARCH_CONVERT)))) {
335 if ((INT_GET(d->d_ino_softlimit, ARCH_CONVERT) &&
336 (INT_GET(d->d_icount, ARCH_CONVERT) >=
337 INT_GET(d->d_ino_softlimit, ARCH_CONVERT))) ||
338 (INT_GET(d->d_ino_hardlimit, ARCH_CONVERT) &&
339 (INT_GET(d->d_icount, ARCH_CONVERT) >=
340 INT_GET(d->d_ino_hardlimit, ARCH_CONVERT)))) {
341 INT_SET(d->d_itimer, ARCH_CONVERT,
342 get_seconds() + XFS_QI_ITIMELIMIT(mp));
347 if ((!d->d_ino_softlimit ||
348 (INT_GET(d->d_icount, ARCH_CONVERT) <
349 INT_GET(d->d_ino_softlimit, ARCH_CONVERT))) &&
350 (!d->d_ino_hardlimit ||
351 (INT_GET(d->d_icount, ARCH_CONVERT) <
352 INT_GET(d->d_ino_hardlimit, ARCH_CONVERT)))) {
357 if (!d->d_rtbtimer) {
358 if ((INT_GET(d->d_rtb_softlimit, ARCH_CONVERT) &&
359 (INT_GET(d->d_rtbcount, ARCH_CONVERT) >=
360 INT_GET(d->d_rtb_softlimit, ARCH_CONVERT))) ||
361 (INT_GET(d->d_rtb_hardlimit, ARCH_CONVERT) &&
362 (INT_GET(d->d_rtbcount, ARCH_CONVERT) >=
363 INT_GET(d->d_rtb_hardlimit, ARCH_CONVERT)))) {
364 INT_SET(d->d_rtbtimer, ARCH_CONVERT,
365 get_seconds() + XFS_QI_RTBTIMELIMIT(mp));
370 if ((!d->d_rtb_softlimit ||
371 (INT_GET(d->d_rtbcount, ARCH_CONVERT) <
372 INT_GET(d->d_rtb_softlimit, ARCH_CONVERT))) &&
373 (!d->d_rtb_hardlimit ||
374 (INT_GET(d->d_rtbcount, ARCH_CONVERT) <
375 INT_GET(d->d_rtb_hardlimit, ARCH_CONVERT)))) {
382 * initialize a buffer full of dquots and log the whole thing
385 xfs_qm_init_dquot_blk(
396 ASSERT(XFS_BUF_ISBUSY(bp));
397 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
399 d = (xfs_dqblk_t *)XFS_BUF_PTR(bp);
402 * ID of the first dquot in the block - id's are zero based.
404 curid = id - (id % XFS_QM_DQPERBLK(mp));
406 memset(d, 0, BBTOB(XFS_QI_DQCHUNKLEN(mp)));
407 for (i = 0; i < XFS_QM_DQPERBLK(mp); i++, d++, curid++)
408 xfs_qm_dqinit_core(curid, type, d);
409 xfs_trans_dquot_buf(tp, bp,
410 (type & XFS_DQ_USER ? XFS_BLI_UDQUOT_BUF :
411 ((type & XFS_DQ_PROJ) ? XFS_BLI_PDQUOT_BUF :
412 XFS_BLI_GDQUOT_BUF)));
413 xfs_trans_log_buf(tp, bp, 0, BBTOB(XFS_QI_DQCHUNKLEN(mp)) - 1);
419 * Allocate a block and fill it with dquots.
420 * This is called when the bmapi finds a hole.
428 xfs_fileoff_t offset_fsb,
431 xfs_fsblock_t firstblock;
432 xfs_bmap_free_t flist;
434 int nmaps, error, committed;
438 xfs_dqtrace_entry(dqp, "DQALLOC");
441 * Initialize the bmap freelist prior to calling bmapi code.
443 XFS_BMAP_INIT(&flist, &firstblock);
444 xfs_ilock(quotip, XFS_ILOCK_EXCL);
446 * Return if this type of quotas is turned off while we didn't
449 if (XFS_IS_THIS_QUOTA_OFF(dqp)) {
450 xfs_iunlock(quotip, XFS_ILOCK_EXCL);
455 * xfs_trans_commit normally decrements the vnode ref count
456 * when it unlocks the inode. Since we want to keep the quota
457 * inode around, we bump the vnode ref count now.
459 VN_HOLD(XFS_ITOV(quotip));
461 xfs_trans_ijoin(tp, quotip, XFS_ILOCK_EXCL);
463 if ((error = xfs_bmapi(tp, quotip,
464 offset_fsb, XFS_DQUOT_CLUSTER_SIZE_FSB,
465 XFS_BMAPI_METADATA | XFS_BMAPI_WRITE,
467 XFS_QM_DQALLOC_SPACE_RES(mp),
468 &map, &nmaps, &flist))) {
471 ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB);
473 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
474 (map.br_startblock != HOLESTARTBLOCK));
477 * Keep track of the blkno to save a lookup later
479 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
481 /* now we can just get the buffer (there's nothing to read yet) */
482 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
484 XFS_QI_DQCHUNKLEN(mp),
486 if (!bp || (error = XFS_BUF_GETERROR(bp)))
489 * Make a chunk of dquots out of this buffer and log
492 xfs_qm_init_dquot_blk(tp, mp, INT_GET(dqp->q_core.d_id, ARCH_CONVERT),
493 dqp->dq_flags & XFS_DQ_ALLTYPES, bp);
495 if ((error = xfs_bmap_finish(&tp, &flist, firstblock, &committed))) {
503 xfs_bmap_cancel(&flist);
505 xfs_iunlock(quotip, XFS_ILOCK_EXCL);
511 * Maps a dquot to the buffer containing its on-disk version.
512 * This returns a ptr to the buffer containing the on-disk dquot
513 * in the bpp param, and a ptr to the on-disk dquot within that buffer
519 xfs_disk_dquot_t **O_ddpp,
528 xfs_disk_dquot_t *ddq;
533 id = INT_GET(dqp->q_core.d_id, ARCH_CONVERT);
538 * If we don't know where the dquot lives, find out.
540 if (dqp->q_blkno == (xfs_daddr_t) 0) {
541 /* We use the id as an index */
542 dqp->q_fileoffset = (xfs_fileoff_t) ((uint)id /
543 XFS_QM_DQPERBLK(mp));
545 quotip = XFS_DQ_TO_QIP(dqp);
546 xfs_ilock(quotip, XFS_ILOCK_SHARED);
548 * Return if this type of quotas is turned off while we didn't
551 if (XFS_IS_THIS_QUOTA_OFF(dqp)) {
552 xfs_iunlock(quotip, XFS_ILOCK_SHARED);
556 * Find the block map; no allocations yet
558 error = xfs_bmapi(NULL, quotip, dqp->q_fileoffset,
559 XFS_DQUOT_CLUSTER_SIZE_FSB,
561 NULL, 0, &map, &nmaps, NULL);
563 xfs_iunlock(quotip, XFS_ILOCK_SHARED);
567 ASSERT(map.br_blockcount == 1);
570 * offset of dquot in the (fixed sized) dquot chunk.
572 dqp->q_bufoffset = (id % XFS_QM_DQPERBLK(mp)) *
574 if (map.br_startblock == HOLESTARTBLOCK) {
576 * We don't allocate unless we're asked to
578 if (!(flags & XFS_QMOPT_DQALLOC))
582 if ((error = xfs_qm_dqalloc(tp, mp, dqp, quotip,
583 dqp->q_fileoffset, &bp)))
588 * store the blkno etc so that we don't have to do the
589 * mapping all the time
591 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
594 ASSERT(dqp->q_blkno != DELAYSTARTBLOCK);
595 ASSERT(dqp->q_blkno != HOLESTARTBLOCK);
598 * Read in the buffer, unless we've just done the allocation
599 * (in which case we already have the buf).
602 xfs_dqtrace_entry(dqp, "DQTOBP READBUF");
603 if ((error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
605 XFS_QI_DQCHUNKLEN(mp),
610 return XFS_ERROR(error);
612 ASSERT(XFS_BUF_ISBUSY(bp));
613 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
616 * calculate the location of the dquot inside the buffer.
618 ddq = (xfs_disk_dquot_t *)((char *)XFS_BUF_PTR(bp) + dqp->q_bufoffset);
621 * A simple sanity check in case we got a corrupted dquot...
623 if (xfs_qm_dqcheck(ddq, id, dqp->dq_flags & XFS_DQ_ALLTYPES,
624 flags & (XFS_QMOPT_DQREPAIR|XFS_QMOPT_DOWARN),
626 if (!(flags & XFS_QMOPT_DQREPAIR)) {
627 xfs_trans_brelse(tp, bp);
628 return XFS_ERROR(EIO);
630 XFS_BUF_BUSY(bp); /* We dirtied this */
641 * Read in the ondisk dquot using dqtobp() then copy it to an incore version,
642 * and release the buffer immediately.
650 xfs_dquot_t *dqp, /* dquot to get filled in */
653 xfs_disk_dquot_t *ddqp;
658 * get a pointer to the on-disk dquot and the buffer containing it
659 * dqp already knows its own type (GROUP/USER).
661 xfs_dqtrace_entry(dqp, "DQREAD");
662 if ((error = xfs_qm_dqtobp(tp, dqp, &ddqp, &bp, flags))) {
666 /* copy everything from disk dquot to the incore dquot */
667 memcpy(&dqp->q_core, ddqp, sizeof(xfs_disk_dquot_t));
668 ASSERT(INT_GET(dqp->q_core.d_id, ARCH_CONVERT) == id);
669 xfs_qm_dquot_logitem_init(dqp);
672 * Reservation counters are defined as reservation plus current usage
673 * to avoid having to add everytime.
675 dqp->q_res_bcount = INT_GET(ddqp->d_bcount, ARCH_CONVERT);
676 dqp->q_res_icount = INT_GET(ddqp->d_icount, ARCH_CONVERT);
677 dqp->q_res_rtbcount = INT_GET(ddqp->d_rtbcount, ARCH_CONVERT);
679 /* Mark the buf so that this will stay incore a little longer */
680 XFS_BUF_SET_VTYPE_REF(bp, B_FS_DQUOT, XFS_DQUOT_REF);
683 * We got the buffer with a xfs_trans_read_buf() (in dqtobp())
684 * So we need to release with xfs_trans_brelse().
685 * The strategy here is identical to that of inodes; we lock
686 * the dquot in xfs_qm_dqget() before making it accessible to
687 * others. This is because dquots, like inodes, need a good level of
688 * concurrency, and we don't want to take locks on the entire buffers
689 * for dquot accesses.
690 * Note also that the dquot buffer may even be dirty at this point, if
691 * this particular dquot was repaired. We still aren't afraid to
692 * brelse it because we have the changes incore.
694 ASSERT(XFS_BUF_ISBUSY(bp));
695 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
696 xfs_trans_brelse(tp, bp);
703 * allocate an incore dquot from the kernel heap,
704 * and fill its core with quota information kept on disk.
705 * If XFS_QMOPT_DQALLOC is set, it'll allocate a dquot on disk
706 * if it wasn't already allocated.
711 xfs_dqid_t id, /* gid or uid, depending on type */
712 uint type, /* UDQUOT or GDQUOT */
713 uint flags, /* DQALLOC, DQREPAIR */
714 xfs_dquot_t **O_dqpp)/* OUT : incore dquot, not locked */
721 dqp = xfs_qm_dqinit(mp, id, type);
723 if (flags & XFS_QMOPT_DQALLOC) {
724 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_DQALLOC);
725 if ((error = xfs_trans_reserve(tp,
726 XFS_QM_DQALLOC_SPACE_RES(mp),
727 XFS_WRITE_LOG_RES(mp) +
728 BBTOB(XFS_QI_DQCHUNKLEN(mp)) - 1 +
731 XFS_TRANS_PERM_LOG_RES,
732 XFS_WRITE_LOG_COUNT))) {
736 cancelflags = XFS_TRANS_RELEASE_LOG_RES;
740 * Read it from disk; xfs_dqread() takes care of
741 * all the necessary initialization of dquot's fields (locks, etc)
743 if ((error = xfs_qm_dqread(tp, id, dqp, flags))) {
745 * This can happen if quotas got turned off (ESRCH),
746 * or if the dquot didn't exist on disk and we ask to
749 xfs_dqtrace_entry(dqp, "DQREAD FAIL");
750 cancelflags |= XFS_TRANS_ABORT;
754 if ((error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES,
765 xfs_trans_cancel(tp, cancelflags);
767 xfs_qm_dqdestroy(dqp);
773 * Lookup a dquot in the incore dquot hashtable. We keep two separate
774 * hashtables for user and group dquots; and, these are global tables
775 * inside the XQM, not per-filesystem tables.
776 * The hash chain must be locked by caller, and it is left locked
777 * on return. Returning dquot is locked.
784 xfs_dquot_t **O_dqpp)
790 ASSERT(XFS_DQ_IS_HASH_LOCKED(qh));
792 flist_locked = B_FALSE;
795 * Traverse the hashchain looking for a match
797 for (dqp = qh->qh_next; dqp != NULL; dqp = dqp->HL_NEXT) {
799 * We already have the hashlock. We don't need the
800 * dqlock to look at the id field of the dquot, since the
801 * id can't be modified without the hashlock anyway.
803 if (INT_GET(dqp->q_core.d_id, ARCH_CONVERT) == id && dqp->q_mount == mp) {
804 xfs_dqtrace_entry(dqp, "DQFOUND BY LOOKUP");
806 * All in core dquots must be on the dqlist of mp
808 ASSERT(dqp->MPL_PREVP != NULL);
811 if (dqp->q_nrefs == 0) {
812 ASSERT (XFS_DQ_IS_ON_FREELIST(dqp));
813 if (! xfs_qm_freelist_lock_nowait(xfs_Gqm)) {
814 xfs_dqtrace_entry(dqp, "DQLOOKUP: WANT");
817 * We may have raced with dqreclaim_one()
818 * (and lost). So, flag that we don't
819 * want the dquot to be reclaimed.
821 dqp->dq_flags |= XFS_DQ_WANT;
823 xfs_qm_freelist_lock(xfs_Gqm);
825 dqp->dq_flags &= ~(XFS_DQ_WANT);
827 flist_locked = B_TRUE;
831 * id couldn't have changed; we had the hashlock all
834 ASSERT(INT_GET(dqp->q_core.d_id, ARCH_CONVERT) == id);
837 if (dqp->q_nrefs != 0) {
838 xfs_qm_freelist_unlock(xfs_Gqm);
839 flist_locked = B_FALSE;
842 * take it off the freelist
844 xfs_dqtrace_entry(dqp,
845 "DQLOOKUP: TAKEOFF FL");
846 XQM_FREELIST_REMOVE(dqp);
847 /* xfs_qm_freelist_print(&(xfs_Gqm->
859 xfs_qm_freelist_unlock(xfs_Gqm);
861 * move the dquot to the front of the hashchain
863 ASSERT(XFS_DQ_IS_HASH_LOCKED(qh));
864 if (dqp->HL_PREVP != &qh->qh_next) {
865 xfs_dqtrace_entry(dqp,
866 "DQLOOKUP: HASH MOVETOFRONT");
867 if ((d = dqp->HL_NEXT))
868 d->HL_PREVP = dqp->HL_PREVP;
869 *(dqp->HL_PREVP) = d;
871 d->HL_PREVP = &dqp->HL_NEXT;
873 dqp->HL_PREVP = &qh->qh_next;
876 xfs_dqtrace_entry(dqp, "LOOKUP END");
878 ASSERT(XFS_DQ_IS_HASH_LOCKED(qh));
884 ASSERT(XFS_DQ_IS_HASH_LOCKED(qh));
889 * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a
890 * a locked dquot, doing an allocation (if requested) as needed.
891 * When both an inode and an id are given, the inode's id takes precedence.
892 * That is, if the id changes while we don't hold the ilock inside this
893 * function, the new dquot is returned, not necessarily the one requested
894 * in the id argument.
899 xfs_inode_t *ip, /* locked inode (optional) */
900 xfs_dqid_t id, /* uid/projid/gid depending on type */
901 uint type, /* XFS_DQ_USER/XFS_DQ_PROJ/XFS_DQ_GROUP */
902 uint flags, /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */
903 xfs_dquot_t **O_dqpp) /* OUT : locked incore dquot */
910 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
911 if ((! XFS_IS_UQUOTA_ON(mp) && type == XFS_DQ_USER) ||
912 (! XFS_IS_PQUOTA_ON(mp) && type == XFS_DQ_PROJ) ||
913 (! XFS_IS_GQUOTA_ON(mp) && type == XFS_DQ_GROUP)) {
916 h = XFS_DQ_HASH(mp, id, type);
919 if (xfs_do_dqerror) {
920 if ((xfs_dqerror_target == mp->m_ddev_targp) &&
921 (xfs_dqreq_num++ % xfs_dqerror_mod) == 0) {
922 cmn_err(CE_DEBUG, "Returning error in dqget");
931 ASSERT(type == XFS_DQ_USER ||
932 type == XFS_DQ_PROJ ||
933 type == XFS_DQ_GROUP);
935 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip));
936 if (type == XFS_DQ_USER)
937 ASSERT(ip->i_udquot == NULL);
939 ASSERT(ip->i_gdquot == NULL);
945 * Look in the cache (hashtable).
946 * The chain is kept locked during lookup.
948 if (xfs_qm_dqlookup(mp, id, h, O_dqpp) == 0) {
949 XQM_STATS_INC(xqmstats.xs_qm_dqcachehits);
951 * The dquot was found, moved to the front of the chain,
952 * taken off the freelist if it was on it, and locked
953 * at this point. Just unlock the hashchain and return.
956 ASSERT(XFS_DQ_IS_LOCKED(*O_dqpp));
957 XFS_DQ_HASH_UNLOCK(h);
958 xfs_dqtrace_entry(*O_dqpp, "DQGET DONE (FROM CACHE)");
959 return (0); /* success */
961 XQM_STATS_INC(xqmstats.xs_qm_dqcachemisses);
964 * Dquot cache miss. We don't want to keep the inode lock across
965 * a (potential) disk read. Also we don't want to deal with the lock
966 * ordering between quotainode and this inode. OTOH, dropping the inode
967 * lock here means dealing with a chown that can happen before
968 * we re-acquire the lock.
971 xfs_iunlock(ip, XFS_ILOCK_EXCL);
973 * Save the hashchain version stamp, and unlock the chain, so that
974 * we don't keep the lock across a disk read
976 version = h->qh_version;
977 XFS_DQ_HASH_UNLOCK(h);
980 * Allocate the dquot on the kernel heap, and read the ondisk
981 * portion off the disk. Also, do all the necessary initialization
982 * This can return ENOENT if dquot didn't exist on disk and we didn't
983 * ask it to allocate; ESRCH if quotas got turned off suddenly.
985 if ((error = xfs_qm_idtodq(mp, id, type,
986 flags & (XFS_QMOPT_DQALLOC|XFS_QMOPT_DQREPAIR|
990 xfs_ilock(ip, XFS_ILOCK_EXCL);
995 * See if this is mount code calling to look at the overall quota limits
996 * which are stored in the id == 0 user or group's dquot.
997 * Since we may not have done a quotacheck by this point, just return
998 * the dquot without attaching it to any hashtables, lists, etc, or even
999 * taking a reference.
1000 * The caller must dqdestroy this once done.
1002 if (flags & XFS_QMOPT_DQSUSER) {
1009 * Dquot lock comes after hashlock in the lock ordering
1012 xfs_ilock(ip, XFS_ILOCK_EXCL);
1013 if (! XFS_IS_DQTYPE_ON(mp, type)) {
1014 /* inode stays locked on return */
1015 xfs_qm_dqdestroy(dqp);
1016 return XFS_ERROR(ESRCH);
1019 * A dquot could be attached to this inode by now, since
1020 * we had dropped the ilock.
1022 if (type == XFS_DQ_USER) {
1024 xfs_qm_dqdestroy(dqp);
1031 xfs_qm_dqdestroy(dqp);
1040 * Hashlock comes after ilock in lock order
1042 XFS_DQ_HASH_LOCK(h);
1043 if (version != h->qh_version) {
1044 xfs_dquot_t *tmpdqp;
1046 * Now, see if somebody else put the dquot in the
1047 * hashtable before us. This can happen because we didn't
1048 * keep the hashchain lock. We don't have to worry about
1049 * lock order between the two dquots here since dqp isn't
1050 * on any findable lists yet.
1052 if (xfs_qm_dqlookup(mp, id, h, &tmpdqp) == 0) {
1054 * Duplicate found. Just throw away the new dquot
1057 xfs_qm_dqput(tmpdqp);
1058 XFS_DQ_HASH_UNLOCK(h);
1059 xfs_qm_dqdestroy(dqp);
1060 XQM_STATS_INC(xqmstats.xs_qm_dquot_dups);
1066 * Put the dquot at the beginning of the hash-chain and mp's list
1067 * LOCK ORDER: hashlock, freelistlock, mplistlock, udqlock, gdqlock ..
1069 ASSERT(XFS_DQ_IS_HASH_LOCKED(h));
1071 XQM_HASHLIST_INSERT(h, dqp);
1074 * Attach this dquot to this filesystem's list of all dquots,
1075 * kept inside the mount structure in m_quotainfo field
1077 xfs_qm_mplist_lock(mp);
1080 * We return a locked dquot to the caller, with a reference taken
1085 XQM_MPLIST_INSERT(&(XFS_QI_MPL_LIST(mp)), dqp);
1087 xfs_qm_mplist_unlock(mp);
1088 XFS_DQ_HASH_UNLOCK(h);
1090 ASSERT((ip == NULL) || XFS_ISLOCKED_INODE_EXCL(ip));
1091 xfs_dqtrace_entry(dqp, "DQGET DONE");
1098 * Release a reference to the dquot (decrement ref-count)
1099 * and unlock it. If there is a group quota attached to this
1100 * dquot, carefully release that too without tripping over
1101 * deadlocks'n'stuff.
1109 ASSERT(dqp->q_nrefs > 0);
1110 ASSERT(XFS_DQ_IS_LOCKED(dqp));
1111 xfs_dqtrace_entry(dqp, "DQPUT");
1113 if (dqp->q_nrefs != 1) {
1120 * drop the dqlock and acquire the freelist and dqlock
1121 * in the right order; but try to get it out-of-order first
1123 if (! xfs_qm_freelist_lock_nowait(xfs_Gqm)) {
1124 xfs_dqtrace_entry(dqp, "DQPUT: FLLOCK-WAIT");
1126 xfs_qm_freelist_lock(xfs_Gqm);
1133 /* We can't depend on nrefs being == 1 here */
1134 if (--dqp->q_nrefs == 0) {
1135 xfs_dqtrace_entry(dqp, "DQPUT: ON FREELIST");
1137 * insert at end of the freelist.
1139 XQM_FREELIST_INSERT(&(xfs_Gqm->qm_dqfreelist), dqp);
1142 * If we just added a udquot to the freelist, then
1143 * we want to release the gdquot reference that
1144 * it (probably) has. Otherwise it'll keep the
1145 * gdquot from getting reclaimed.
1147 if ((gdqp = dqp->q_gdquot)) {
1149 * Avoid a recursive dqput call
1152 dqp->q_gdquot = NULL;
1155 /* xfs_qm_freelist_print(&(xfs_Gqm->qm_dqfreelist),
1156 "@@@@@++ Free list (after append) @@@@@+");
1162 * If we had a group quota inside the user quota as a hint,
1169 xfs_qm_freelist_unlock(xfs_Gqm);
1173 * Release a dquot. Flush it if dirty, then dqput() it.
1174 * dquot must not be locked.
1181 xfs_dqtrace_entry(dqp, "DQRELE");
1185 * We don't care to flush it if the dquot is dirty here.
1186 * That will create stutters that we want to avoid.
1187 * Instead we do a delayed write when we try to reclaim
1188 * a dirty dquot. Also xfs_sync will take part of the burden...
1195 * Write a modified dquot to disk.
1196 * The dquot must be locked and the flush lock too taken by caller.
1197 * The flush lock will not be unlocked until the dquot reaches the disk,
1198 * but the dquot is free to be unlocked and modified by the caller
1199 * in the interim. Dquot is still locked on return. This behavior is
1200 * identical to that of inodes.
1209 xfs_disk_dquot_t *ddqp;
1213 ASSERT(XFS_DQ_IS_LOCKED(dqp));
1214 ASSERT(XFS_DQ_IS_FLUSH_LOCKED(dqp));
1215 xfs_dqtrace_entry(dqp, "DQFLUSH");
1218 * If not dirty, nada.
1220 if (!XFS_DQ_IS_DIRTY(dqp)) {
1226 * Cant flush a pinned dquot. Wait for it.
1228 xfs_qm_dqunpin_wait(dqp);
1231 * This may have been unpinned because the filesystem is shutting
1232 * down forcibly. If that's the case we must not write this dquot
1233 * to disk, because the log record didn't make it to disk!
1235 if (XFS_FORCED_SHUTDOWN(dqp->q_mount)) {
1236 dqp->dq_flags &= ~(XFS_DQ_DIRTY);
1238 return XFS_ERROR(EIO);
1242 * Get the buffer containing the on-disk dquot
1243 * We don't need a transaction envelope because we know that the
1244 * the ondisk-dquot has already been allocated for.
1246 if ((error = xfs_qm_dqtobp(NULL, dqp, &ddqp, &bp, XFS_QMOPT_DOWARN))) {
1247 xfs_dqtrace_entry(dqp, "DQTOBP FAIL");
1248 ASSERT(error != ENOENT);
1250 * Quotas could have gotten turned off (ESRCH)
1256 if (xfs_qm_dqcheck(&dqp->q_core, INT_GET(ddqp->d_id, ARCH_CONVERT),
1257 0, XFS_QMOPT_DOWARN, "dqflush (incore copy)")) {
1258 xfs_force_shutdown(dqp->q_mount, XFS_CORRUPT_INCORE);
1259 return XFS_ERROR(EIO);
1262 /* This is the only portion of data that needs to persist */
1263 memcpy(ddqp, &(dqp->q_core), sizeof(xfs_disk_dquot_t));
1266 * Clear the dirty field and remember the flush lsn for later use.
1268 dqp->dq_flags &= ~(XFS_DQ_DIRTY);
1271 /* lsn is 64 bits */
1273 dqp->q_logitem.qli_flush_lsn = dqp->q_logitem.qli_item.li_lsn;
1277 * Attach an iodone routine so that we can remove this dquot from the
1278 * AIL and release the flush lock once the dquot is synced to disk.
1280 xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t *, xfs_log_item_t *))
1281 xfs_qm_dqflush_done, &(dqp->q_logitem.qli_item));
1283 * If the buffer is pinned then push on the log so we won't
1284 * get stuck waiting in the write for too long.
1286 if (XFS_BUF_ISPINNED(bp)) {
1287 xfs_dqtrace_entry(dqp, "DQFLUSH LOG FORCE");
1288 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
1291 if (flags & XFS_QMOPT_DELWRI) {
1292 xfs_bdwrite(mp, bp);
1293 } else if (flags & XFS_QMOPT_ASYNC) {
1294 xfs_bawrite(mp, bp);
1296 error = xfs_bwrite(mp, bp);
1298 xfs_dqtrace_entry(dqp, "DQFLUSH END");
1300 * dqp is still locked, but caller is free to unlock it now.
1307 * This is the dquot flushing I/O completion routine. It is called
1308 * from interrupt level when the buffer containing the dquot is
1309 * flushed to disk. It is responsible for removing the dquot logitem
1310 * from the AIL if it has not been re-logged, and unlocking the dquot's
1311 * flush lock. This behavior is very similar to that of inodes..
1315 xfs_qm_dqflush_done(
1317 xfs_dq_logitem_t *qip)
1322 dqp = qip->qli_dquot;
1325 * We only want to pull the item from the AIL if its
1326 * location in the log has not changed since we started the flush.
1327 * Thus, we only bother if the dquot's lsn has
1328 * not changed. First we check the lsn outside the lock
1329 * since it's cheaper, and then we recheck while
1330 * holding the lock before removing the dquot from the AIL.
1332 if ((qip->qli_item.li_flags & XFS_LI_IN_AIL) &&
1333 qip->qli_item.li_lsn == qip->qli_flush_lsn) {
1335 AIL_LOCK(dqp->q_mount, s);
1337 * xfs_trans_delete_ail() drops the AIL lock.
1339 if (qip->qli_item.li_lsn == qip->qli_flush_lsn)
1340 xfs_trans_delete_ail(dqp->q_mount,
1341 (xfs_log_item_t*)qip, s);
1343 AIL_UNLOCK(dqp->q_mount, s);
1347 * Release the dq's flush lock since we're done with it.
1354 xfs_qm_dqflock_nowait(
1359 locked = cpsema(&((dqp)->q_flock));
1361 /* XXX ifdef these out */
1363 (dqp)->dq_flags |= XFS_DQ_FLOCKED;
1369 xfs_qm_dqlock_nowait(
1372 return (mutex_trylock(&((dqp)->q_qlock)));
1379 mutex_lock(&(dqp->q_qlock), PINOD);
1386 mutex_unlock(&(dqp->q_qlock));
1387 if (dqp->q_logitem.qli_dquot == dqp) {
1388 /* Once was dqp->q_mount, but might just have been cleared */
1389 xfs_trans_unlocked_item(dqp->q_logitem.qli_item.li_mountp,
1390 (xfs_log_item_t*)&(dqp->q_logitem));
1396 xfs_dqunlock_nonotify(
1399 mutex_unlock(&(dqp->q_qlock));
1409 if (INT_GET(d1->q_core.d_id, ARCH_CONVERT) >
1410 INT_GET(d2->q_core.d_id, ARCH_CONVERT)) {
1428 * Take a dquot out of the mount's dqlist as well as the hashlist.
1429 * This is called via unmount as well as quotaoff, and the purge
1430 * will always succeed unless there are soft (temp) references
1433 * This returns 0 if it was purged, 1 if it wasn't. It's not an error code
1434 * that we're returning! XXXsup - not cool.
1442 xfs_dqhash_t *thishash;
1447 ASSERT(XFS_QM_IS_MPLIST_LOCKED(mp));
1448 ASSERT(XFS_DQ_IS_HASH_LOCKED(dqp->q_hash));
1452 * We really can't afford to purge a dquot that is
1453 * referenced, because these are hard refs.
1454 * It shouldn't happen in general because we went thru _all_ inodes in
1455 * dqrele_all_inodes before calling this and didn't let the mountlock go.
1456 * However it is possible that we have dquots with temporary
1457 * references that are not attached to an inode. e.g. see xfs_setattr().
1459 if (dqp->q_nrefs != 0) {
1461 XFS_DQ_HASH_UNLOCK(dqp->q_hash);
1465 ASSERT(XFS_DQ_IS_ON_FREELIST(dqp));
1468 * If we're turning off quotas, we have to make sure that, for
1469 * example, we don't delete quota disk blocks while dquots are
1470 * in the process of getting written to those disk blocks.
1471 * This dquot might well be on AIL, and we can't leave it there
1472 * if we're turning off quotas. Basically, we need this flush
1473 * lock, and are willing to block on it.
1475 if (! xfs_qm_dqflock_nowait(dqp)) {
1477 * Block on the flush lock after nudging dquot buffer,
1480 xfs_qm_dqflock_pushbuf_wait(dqp);
1484 * XXXIf we're turning this type of quotas off, we don't care
1485 * about the dirty metadata sitting in this dquot. OTOH, if
1486 * we're unmounting, we do care, so we flush it and wait.
1488 if (XFS_DQ_IS_DIRTY(dqp)) {
1489 xfs_dqtrace_entry(dqp, "DQPURGE ->DQFLUSH: DQDIRTY");
1490 /* dqflush unlocks dqflock */
1492 * Given that dqpurge is a very rare occurrence, it is OK
1493 * that we're holding the hashlist and mplist locks
1494 * across the disk write. But, ... XXXsup
1496 * We don't care about getting disk errors here. We need
1497 * to purge this dquot anyway, so we go ahead regardless.
1499 (void) xfs_qm_dqflush(dqp, XFS_QMOPT_SYNC);
1502 ASSERT(dqp->q_pincount == 0);
1503 ASSERT(XFS_FORCED_SHUTDOWN(mp) ||
1504 !(dqp->q_logitem.qli_item.li_flags & XFS_LI_IN_AIL));
1506 thishash = dqp->q_hash;
1507 XQM_HASHLIST_REMOVE(thishash, dqp);
1508 XQM_MPLIST_REMOVE(&(XFS_QI_MPL_LIST(mp)), dqp);
1510 * XXX Move this to the front of the freelist, if we can get the
1513 ASSERT(XFS_DQ_IS_ON_FREELIST(dqp));
1515 dqp->q_mount = NULL;
1517 dqp->dq_flags = XFS_DQ_INACTIVE;
1518 memset(&dqp->q_core, 0, sizeof(dqp->q_core));
1521 XFS_DQ_HASH_UNLOCK(thishash);
1528 xfs_qm_dqprint(xfs_dquot_t *dqp)
1530 cmn_err(CE_DEBUG, "-----------KERNEL DQUOT----------------");
1531 cmn_err(CE_DEBUG, "---- dquotID = %d",
1532 (int)INT_GET(dqp->q_core.d_id, ARCH_CONVERT));
1533 cmn_err(CE_DEBUG, "---- type = %s", DQFLAGTO_TYPESTR(dqp));
1534 cmn_err(CE_DEBUG, "---- fs = 0x%p", dqp->q_mount);
1535 cmn_err(CE_DEBUG, "---- blkno = 0x%x", (int) dqp->q_blkno);
1536 cmn_err(CE_DEBUG, "---- boffset = 0x%x", (int) dqp->q_bufoffset);
1537 cmn_err(CE_DEBUG, "---- blkhlimit = %Lu (0x%x)",
1538 INT_GET(dqp->q_core.d_blk_hardlimit, ARCH_CONVERT),
1539 (int) INT_GET(dqp->q_core.d_blk_hardlimit, ARCH_CONVERT));
1540 cmn_err(CE_DEBUG, "---- blkslimit = %Lu (0x%x)",
1541 INT_GET(dqp->q_core.d_blk_softlimit, ARCH_CONVERT),
1542 (int)INT_GET(dqp->q_core.d_blk_softlimit, ARCH_CONVERT));
1543 cmn_err(CE_DEBUG, "---- inohlimit = %Lu (0x%x)",
1544 INT_GET(dqp->q_core.d_ino_hardlimit, ARCH_CONVERT),
1545 (int)INT_GET(dqp->q_core.d_ino_hardlimit, ARCH_CONVERT));
1546 cmn_err(CE_DEBUG, "---- inoslimit = %Lu (0x%x)",
1547 INT_GET(dqp->q_core.d_ino_softlimit, ARCH_CONVERT),
1548 (int)INT_GET(dqp->q_core.d_ino_softlimit, ARCH_CONVERT));
1549 cmn_err(CE_DEBUG, "---- bcount = %Lu (0x%x)",
1550 INT_GET(dqp->q_core.d_bcount, ARCH_CONVERT),
1551 (int)INT_GET(dqp->q_core.d_bcount, ARCH_CONVERT));
1552 cmn_err(CE_DEBUG, "---- icount = %Lu (0x%x)",
1553 INT_GET(dqp->q_core.d_icount, ARCH_CONVERT),
1554 (int)INT_GET(dqp->q_core.d_icount, ARCH_CONVERT));
1555 cmn_err(CE_DEBUG, "---- btimer = %d",
1556 (int)INT_GET(dqp->q_core.d_btimer, ARCH_CONVERT));
1557 cmn_err(CE_DEBUG, "---- itimer = %d",
1558 (int)INT_GET(dqp->q_core.d_itimer, ARCH_CONVERT));
1559 cmn_err(CE_DEBUG, "---------------------------");
1564 * Give the buffer a little push if it is incore and
1565 * wait on the flush lock.
1568 xfs_qm_dqflock_pushbuf_wait(
1574 * Check to see if the dquot has been flushed delayed
1575 * write. If so, grab its buffer and send it
1576 * out immediately. We'll be able to acquire
1577 * the flush lock when the I/O completes.
1579 bp = xfs_incore(dqp->q_mount->m_ddev_targp, dqp->q_blkno,
1580 XFS_QI_DQCHUNKLEN(dqp->q_mount),
1581 XFS_INCORE_TRYLOCK);
1583 if (XFS_BUF_ISDELAYWRITE(bp)) {
1584 if (XFS_BUF_ISPINNED(bp)) {
1585 xfs_log_force(dqp->q_mount,
1589 xfs_bawrite(dqp->q_mount, bp);