xref: /openbmc/linux/fs/xfs/xfs_dquot.c (revision c4a11bf4)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2000-2003 Silicon Graphics, Inc.
4  * All Rights Reserved.
5  */
6 #include "xfs.h"
7 #include "xfs_fs.h"
8 #include "xfs_format.h"
9 #include "xfs_log_format.h"
10 #include "xfs_shared.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_bit.h"
13 #include "xfs_mount.h"
14 #include "xfs_defer.h"
15 #include "xfs_inode.h"
16 #include "xfs_bmap.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"
22 #include "xfs_qm.h"
23 #include "xfs_trace.h"
24 #include "xfs_log.h"
25 #include "xfs_bmap_btree.h"
26 #include "xfs_error.h"
27 
28 /*
29  * Lock order:
30  *
31  * ip->i_lock
32  *   qi->qi_tree_lock
33  *     dquot->q_qlock (xfs_dqlock() and friends)
34  *       dquot->q_flush (xfs_dqflock() and friends)
35  *       qi->qi_lru_lock
36  *
37  * If two dquots need to be locked the order is user before group/project,
38  * otherwise by the lowest id first, see xfs_dqlock2.
39  */
40 
41 struct kmem_cache		*xfs_dqtrx_cache;
42 static struct kmem_cache	*xfs_dquot_cache;
43 
44 static struct lock_class_key xfs_dquot_group_class;
45 static struct lock_class_key xfs_dquot_project_class;
46 
47 /*
48  * This is called to free all the memory associated with a dquot
49  */
50 void
51 xfs_qm_dqdestroy(
52 	struct xfs_dquot	*dqp)
53 {
54 	ASSERT(list_empty(&dqp->q_lru));
55 
56 	kmem_free(dqp->q_logitem.qli_item.li_lv_shadow);
57 	mutex_destroy(&dqp->q_qlock);
58 
59 	XFS_STATS_DEC(dqp->q_mount, xs_qm_dquot);
60 	kmem_cache_free(xfs_dquot_cache, dqp);
61 }
62 
63 /*
64  * If default limits are in force, push them into the dquot now.
65  * We overwrite the dquot limits only if they are zero and this
66  * is not the root dquot.
67  */
68 void
69 xfs_qm_adjust_dqlimits(
70 	struct xfs_dquot	*dq)
71 {
72 	struct xfs_mount	*mp = dq->q_mount;
73 	struct xfs_quotainfo	*q = mp->m_quotainfo;
74 	struct xfs_def_quota	*defq;
75 	int			prealloc = 0;
76 
77 	ASSERT(dq->q_id);
78 	defq = xfs_get_defquota(q, xfs_dquot_type(dq));
79 
80 	if (!dq->q_blk.softlimit) {
81 		dq->q_blk.softlimit = defq->blk.soft;
82 		prealloc = 1;
83 	}
84 	if (!dq->q_blk.hardlimit) {
85 		dq->q_blk.hardlimit = defq->blk.hard;
86 		prealloc = 1;
87 	}
88 	if (!dq->q_ino.softlimit)
89 		dq->q_ino.softlimit = defq->ino.soft;
90 	if (!dq->q_ino.hardlimit)
91 		dq->q_ino.hardlimit = defq->ino.hard;
92 	if (!dq->q_rtb.softlimit)
93 		dq->q_rtb.softlimit = defq->rtb.soft;
94 	if (!dq->q_rtb.hardlimit)
95 		dq->q_rtb.hardlimit = defq->rtb.hard;
96 
97 	if (prealloc)
98 		xfs_dquot_set_prealloc_limits(dq);
99 }
100 
101 /* Set the expiration time of a quota's grace period. */
102 time64_t
103 xfs_dquot_set_timeout(
104 	struct xfs_mount	*mp,
105 	time64_t		timeout)
106 {
107 	struct xfs_quotainfo	*qi = mp->m_quotainfo;
108 
109 	return clamp_t(time64_t, timeout, qi->qi_expiry_min,
110 					  qi->qi_expiry_max);
111 }
112 
113 /* Set the length of the default grace period. */
114 time64_t
115 xfs_dquot_set_grace_period(
116 	time64_t		grace)
117 {
118 	return clamp_t(time64_t, grace, XFS_DQ_GRACE_MIN, XFS_DQ_GRACE_MAX);
119 }
120 
121 /*
122  * Determine if this quota counter is over either limit and set the quota
123  * timers as appropriate.
124  */
125 static inline void
126 xfs_qm_adjust_res_timer(
127 	struct xfs_mount	*mp,
128 	struct xfs_dquot_res	*res,
129 	struct xfs_quota_limits	*qlim)
130 {
131 	ASSERT(res->hardlimit == 0 || res->softlimit <= res->hardlimit);
132 
133 	if ((res->softlimit && res->count > res->softlimit) ||
134 	    (res->hardlimit && res->count > res->hardlimit)) {
135 		if (res->timer == 0)
136 			res->timer = xfs_dquot_set_timeout(mp,
137 					ktime_get_real_seconds() + qlim->time);
138 	} else {
139 		if (res->timer == 0)
140 			res->warnings = 0;
141 		else
142 			res->timer = 0;
143 	}
144 }
145 
146 /*
147  * Check the limits and timers of a dquot and start or reset timers
148  * if necessary.
149  * This gets called even when quota enforcement is OFF, which makes our
150  * life a little less complicated. (We just don't reject any quota
151  * reservations in that case, when enforcement is off).
152  * We also return 0 as the values of the timers in Q_GETQUOTA calls, when
153  * enforcement's off.
154  * In contrast, warnings are a little different in that they don't
155  * 'automatically' get started when limits get exceeded.  They do
156  * get reset to zero, however, when we find the count to be under
157  * the soft limit (they are only ever set non-zero via userspace).
158  */
159 void
160 xfs_qm_adjust_dqtimers(
161 	struct xfs_dquot	*dq)
162 {
163 	struct xfs_mount	*mp = dq->q_mount;
164 	struct xfs_quotainfo	*qi = mp->m_quotainfo;
165 	struct xfs_def_quota	*defq;
166 
167 	ASSERT(dq->q_id);
168 	defq = xfs_get_defquota(qi, xfs_dquot_type(dq));
169 
170 	xfs_qm_adjust_res_timer(dq->q_mount, &dq->q_blk, &defq->blk);
171 	xfs_qm_adjust_res_timer(dq->q_mount, &dq->q_ino, &defq->ino);
172 	xfs_qm_adjust_res_timer(dq->q_mount, &dq->q_rtb, &defq->rtb);
173 }
174 
175 /*
176  * initialize a buffer full of dquots and log the whole thing
177  */
178 STATIC void
179 xfs_qm_init_dquot_blk(
180 	struct xfs_trans	*tp,
181 	struct xfs_mount	*mp,
182 	xfs_dqid_t		id,
183 	xfs_dqtype_t		type,
184 	struct xfs_buf		*bp)
185 {
186 	struct xfs_quotainfo	*q = mp->m_quotainfo;
187 	struct xfs_dqblk	*d;
188 	xfs_dqid_t		curid;
189 	unsigned int		qflag;
190 	unsigned int		blftype;
191 	int			i;
192 
193 	ASSERT(tp);
194 	ASSERT(xfs_buf_islocked(bp));
195 
196 	switch (type) {
197 	case XFS_DQTYPE_USER:
198 		qflag = XFS_UQUOTA_CHKD;
199 		blftype = XFS_BLF_UDQUOT_BUF;
200 		break;
201 	case XFS_DQTYPE_PROJ:
202 		qflag = XFS_PQUOTA_CHKD;
203 		blftype = XFS_BLF_PDQUOT_BUF;
204 		break;
205 	case XFS_DQTYPE_GROUP:
206 		qflag = XFS_GQUOTA_CHKD;
207 		blftype = XFS_BLF_GDQUOT_BUF;
208 		break;
209 	default:
210 		ASSERT(0);
211 		return;
212 	}
213 
214 	d = bp->b_addr;
215 
216 	/*
217 	 * ID of the first dquot in the block - id's are zero based.
218 	 */
219 	curid = id - (id % q->qi_dqperchunk);
220 	memset(d, 0, BBTOB(q->qi_dqchunklen));
221 	for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++) {
222 		d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
223 		d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
224 		d->dd_diskdq.d_id = cpu_to_be32(curid);
225 		d->dd_diskdq.d_type = type;
226 		if (curid > 0 && xfs_has_bigtime(mp))
227 			d->dd_diskdq.d_type |= XFS_DQTYPE_BIGTIME;
228 		if (xfs_has_crc(mp)) {
229 			uuid_copy(&d->dd_uuid, &mp->m_sb.sb_meta_uuid);
230 			xfs_update_cksum((char *)d, sizeof(struct xfs_dqblk),
231 					 XFS_DQUOT_CRC_OFF);
232 		}
233 	}
234 
235 	xfs_trans_dquot_buf(tp, bp, blftype);
236 
237 	/*
238 	 * quotacheck uses delayed writes to update all the dquots on disk in an
239 	 * efficient manner instead of logging the individual dquot changes as
240 	 * they are made. However if we log the buffer allocated here and crash
241 	 * after quotacheck while the logged initialisation is still in the
242 	 * active region of the log, log recovery can replay the dquot buffer
243 	 * initialisation over the top of the checked dquots and corrupt quota
244 	 * accounting.
245 	 *
246 	 * To avoid this problem, quotacheck cannot log the initialised buffer.
247 	 * We must still dirty the buffer and write it back before the
248 	 * allocation transaction clears the log. Therefore, mark the buffer as
249 	 * ordered instead of logging it directly. This is safe for quotacheck
250 	 * because it detects and repairs allocated but initialized dquot blocks
251 	 * in the quota inodes.
252 	 */
253 	if (!(mp->m_qflags & qflag))
254 		xfs_trans_ordered_buf(tp, bp);
255 	else
256 		xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1);
257 }
258 
259 /*
260  * Initialize the dynamic speculative preallocation thresholds. The lo/hi
261  * watermarks correspond to the soft and hard limits by default. If a soft limit
262  * is not specified, we use 95% of the hard limit.
263  */
264 void
265 xfs_dquot_set_prealloc_limits(struct xfs_dquot *dqp)
266 {
267 	uint64_t space;
268 
269 	dqp->q_prealloc_hi_wmark = dqp->q_blk.hardlimit;
270 	dqp->q_prealloc_lo_wmark = dqp->q_blk.softlimit;
271 	if (!dqp->q_prealloc_lo_wmark) {
272 		dqp->q_prealloc_lo_wmark = dqp->q_prealloc_hi_wmark;
273 		do_div(dqp->q_prealloc_lo_wmark, 100);
274 		dqp->q_prealloc_lo_wmark *= 95;
275 	}
276 
277 	space = dqp->q_prealloc_hi_wmark;
278 
279 	do_div(space, 100);
280 	dqp->q_low_space[XFS_QLOWSP_1_PCNT] = space;
281 	dqp->q_low_space[XFS_QLOWSP_3_PCNT] = space * 3;
282 	dqp->q_low_space[XFS_QLOWSP_5_PCNT] = space * 5;
283 }
284 
285 /*
286  * Ensure that the given in-core dquot has a buffer on disk backing it, and
287  * return the buffer locked and held. This is called when the bmapi finds a
288  * hole.
289  */
290 STATIC int
291 xfs_dquot_disk_alloc(
292 	struct xfs_trans	**tpp,
293 	struct xfs_dquot	*dqp,
294 	struct xfs_buf		**bpp)
295 {
296 	struct xfs_bmbt_irec	map;
297 	struct xfs_trans	*tp = *tpp;
298 	struct xfs_mount	*mp = tp->t_mountp;
299 	struct xfs_buf		*bp;
300 	xfs_dqtype_t		qtype = xfs_dquot_type(dqp);
301 	struct xfs_inode	*quotip = xfs_quota_inode(mp, qtype);
302 	int			nmaps = 1;
303 	int			error;
304 
305 	trace_xfs_dqalloc(dqp);
306 
307 	xfs_ilock(quotip, XFS_ILOCK_EXCL);
308 	if (!xfs_this_quota_on(dqp->q_mount, qtype)) {
309 		/*
310 		 * Return if this type of quotas is turned off while we didn't
311 		 * have an inode lock
312 		 */
313 		xfs_iunlock(quotip, XFS_ILOCK_EXCL);
314 		return -ESRCH;
315 	}
316 
317 	xfs_trans_ijoin(tp, quotip, XFS_ILOCK_EXCL);
318 
319 	error = xfs_iext_count_may_overflow(quotip, XFS_DATA_FORK,
320 			XFS_IEXT_ADD_NOSPLIT_CNT);
321 	if (error)
322 		return error;
323 
324 	/* Create the block mapping. */
325 	error = xfs_bmapi_write(tp, quotip, dqp->q_fileoffset,
326 			XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA, 0, &map,
327 			&nmaps);
328 	if (error)
329 		return error;
330 	ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB);
331 	ASSERT(nmaps == 1);
332 	ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
333 	       (map.br_startblock != HOLESTARTBLOCK));
334 
335 	/*
336 	 * Keep track of the blkno to save a lookup later
337 	 */
338 	dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
339 
340 	/* now we can just get the buffer (there's nothing to read yet) */
341 	error = xfs_trans_get_buf(tp, mp->m_ddev_targp, dqp->q_blkno,
342 			mp->m_quotainfo->qi_dqchunklen, 0, &bp);
343 	if (error)
344 		return error;
345 	bp->b_ops = &xfs_dquot_buf_ops;
346 
347 	/*
348 	 * Make a chunk of dquots out of this buffer and log
349 	 * the entire thing.
350 	 */
351 	xfs_qm_init_dquot_blk(tp, mp, dqp->q_id, qtype, bp);
352 	xfs_buf_set_ref(bp, XFS_DQUOT_REF);
353 
354 	/*
355 	 * Hold the buffer and join it to the dfops so that we'll still own
356 	 * the buffer when we return to the caller.  The buffer disposal on
357 	 * error must be paid attention to very carefully, as it has been
358 	 * broken since commit efa092f3d4c6 "[XFS] Fixes a bug in the quota
359 	 * code when allocating a new dquot record" in 2005, and the later
360 	 * conversion to xfs_defer_ops in commit 310a75a3c6c747 failed to keep
361 	 * the buffer locked across the _defer_finish call.  We can now do
362 	 * this correctly with xfs_defer_bjoin.
363 	 *
364 	 * Above, we allocated a disk block for the dquot information and used
365 	 * get_buf to initialize the dquot. If the _defer_finish fails, the old
366 	 * transaction is gone but the new buffer is not joined or held to any
367 	 * transaction, so we must _buf_relse it.
368 	 *
369 	 * If everything succeeds, the caller of this function is returned a
370 	 * buffer that is locked and held to the transaction.  The caller
371 	 * is responsible for unlocking any buffer passed back, either
372 	 * manually or by committing the transaction.  On error, the buffer is
373 	 * released and not passed back.
374 	 */
375 	xfs_trans_bhold(tp, bp);
376 	error = xfs_defer_finish(tpp);
377 	if (error) {
378 		xfs_trans_bhold_release(*tpp, bp);
379 		xfs_trans_brelse(*tpp, bp);
380 		return error;
381 	}
382 	*bpp = bp;
383 	return 0;
384 }
385 
386 /*
387  * Read in the in-core dquot's on-disk metadata and return the buffer.
388  * Returns ENOENT to signal a hole.
389  */
390 STATIC int
391 xfs_dquot_disk_read(
392 	struct xfs_mount	*mp,
393 	struct xfs_dquot	*dqp,
394 	struct xfs_buf		**bpp)
395 {
396 	struct xfs_bmbt_irec	map;
397 	struct xfs_buf		*bp;
398 	xfs_dqtype_t		qtype = xfs_dquot_type(dqp);
399 	struct xfs_inode	*quotip = xfs_quota_inode(mp, qtype);
400 	uint			lock_mode;
401 	int			nmaps = 1;
402 	int			error;
403 
404 	lock_mode = xfs_ilock_data_map_shared(quotip);
405 	if (!xfs_this_quota_on(mp, qtype)) {
406 		/*
407 		 * Return if this type of quotas is turned off while we
408 		 * didn't have the quota inode lock.
409 		 */
410 		xfs_iunlock(quotip, lock_mode);
411 		return -ESRCH;
412 	}
413 
414 	/*
415 	 * Find the block map; no allocations yet
416 	 */
417 	error = xfs_bmapi_read(quotip, dqp->q_fileoffset,
418 			XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0);
419 	xfs_iunlock(quotip, lock_mode);
420 	if (error)
421 		return error;
422 
423 	ASSERT(nmaps == 1);
424 	ASSERT(map.br_blockcount >= 1);
425 	ASSERT(map.br_startblock != DELAYSTARTBLOCK);
426 	if (map.br_startblock == HOLESTARTBLOCK)
427 		return -ENOENT;
428 
429 	trace_xfs_dqtobp_read(dqp);
430 
431 	/*
432 	 * store the blkno etc so that we don't have to do the
433 	 * mapping all the time
434 	 */
435 	dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
436 
437 	error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
438 			mp->m_quotainfo->qi_dqchunklen, 0, &bp,
439 			&xfs_dquot_buf_ops);
440 	if (error) {
441 		ASSERT(bp == NULL);
442 		return error;
443 	}
444 
445 	ASSERT(xfs_buf_islocked(bp));
446 	xfs_buf_set_ref(bp, XFS_DQUOT_REF);
447 	*bpp = bp;
448 
449 	return 0;
450 }
451 
452 /* Allocate and initialize everything we need for an incore dquot. */
453 STATIC struct xfs_dquot *
454 xfs_dquot_alloc(
455 	struct xfs_mount	*mp,
456 	xfs_dqid_t		id,
457 	xfs_dqtype_t		type)
458 {
459 	struct xfs_dquot	*dqp;
460 
461 	dqp = kmem_cache_zalloc(xfs_dquot_cache, GFP_KERNEL | __GFP_NOFAIL);
462 
463 	dqp->q_type = type;
464 	dqp->q_id = id;
465 	dqp->q_mount = mp;
466 	INIT_LIST_HEAD(&dqp->q_lru);
467 	mutex_init(&dqp->q_qlock);
468 	init_waitqueue_head(&dqp->q_pinwait);
469 	dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk;
470 	/*
471 	 * Offset of dquot in the (fixed sized) dquot chunk.
472 	 */
473 	dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) *
474 			sizeof(struct xfs_dqblk);
475 
476 	/*
477 	 * Because we want to use a counting completion, complete
478 	 * the flush completion once to allow a single access to
479 	 * the flush completion without blocking.
480 	 */
481 	init_completion(&dqp->q_flush);
482 	complete(&dqp->q_flush);
483 
484 	/*
485 	 * Make sure group quotas have a different lock class than user
486 	 * quotas.
487 	 */
488 	switch (type) {
489 	case XFS_DQTYPE_USER:
490 		/* uses the default lock class */
491 		break;
492 	case XFS_DQTYPE_GROUP:
493 		lockdep_set_class(&dqp->q_qlock, &xfs_dquot_group_class);
494 		break;
495 	case XFS_DQTYPE_PROJ:
496 		lockdep_set_class(&dqp->q_qlock, &xfs_dquot_project_class);
497 		break;
498 	default:
499 		ASSERT(0);
500 		break;
501 	}
502 
503 	xfs_qm_dquot_logitem_init(dqp);
504 
505 	XFS_STATS_INC(mp, xs_qm_dquot);
506 	return dqp;
507 }
508 
509 /* Check the ondisk dquot's id and type match what the incore dquot expects. */
510 static bool
511 xfs_dquot_check_type(
512 	struct xfs_dquot	*dqp,
513 	struct xfs_disk_dquot	*ddqp)
514 {
515 	uint8_t			ddqp_type;
516 	uint8_t			dqp_type;
517 
518 	ddqp_type = ddqp->d_type & XFS_DQTYPE_REC_MASK;
519 	dqp_type = xfs_dquot_type(dqp);
520 
521 	if (be32_to_cpu(ddqp->d_id) != dqp->q_id)
522 		return false;
523 
524 	/*
525 	 * V5 filesystems always expect an exact type match.  V4 filesystems
526 	 * expect an exact match for user dquots and for non-root group and
527 	 * project dquots.
528 	 */
529 	if (xfs_has_crc(dqp->q_mount) ||
530 	    dqp_type == XFS_DQTYPE_USER || dqp->q_id != 0)
531 		return ddqp_type == dqp_type;
532 
533 	/*
534 	 * V4 filesystems support either group or project quotas, but not both
535 	 * at the same time.  The non-user quota file can be switched between
536 	 * group and project quota uses depending on the mount options, which
537 	 * means that we can encounter the other type when we try to load quota
538 	 * defaults.  Quotacheck will soon reset the the entire quota file
539 	 * (including the root dquot) anyway, but don't log scary corruption
540 	 * reports to dmesg.
541 	 */
542 	return ddqp_type == XFS_DQTYPE_GROUP || ddqp_type == XFS_DQTYPE_PROJ;
543 }
544 
545 /* Copy the in-core quota fields in from the on-disk buffer. */
546 STATIC int
547 xfs_dquot_from_disk(
548 	struct xfs_dquot	*dqp,
549 	struct xfs_buf		*bp)
550 {
551 	struct xfs_disk_dquot	*ddqp = bp->b_addr + dqp->q_bufoffset;
552 
553 	/*
554 	 * Ensure that we got the type and ID we were looking for.
555 	 * Everything else was checked by the dquot buffer verifier.
556 	 */
557 	if (!xfs_dquot_check_type(dqp, ddqp)) {
558 		xfs_alert_tag(bp->b_mount, XFS_PTAG_VERIFIER_ERROR,
559 			  "Metadata corruption detected at %pS, quota %u",
560 			  __this_address, dqp->q_id);
561 		xfs_alert(bp->b_mount, "Unmount and run xfs_repair");
562 		return -EFSCORRUPTED;
563 	}
564 
565 	/* copy everything from disk dquot to the incore dquot */
566 	dqp->q_type = ddqp->d_type;
567 	dqp->q_blk.hardlimit = be64_to_cpu(ddqp->d_blk_hardlimit);
568 	dqp->q_blk.softlimit = be64_to_cpu(ddqp->d_blk_softlimit);
569 	dqp->q_ino.hardlimit = be64_to_cpu(ddqp->d_ino_hardlimit);
570 	dqp->q_ino.softlimit = be64_to_cpu(ddqp->d_ino_softlimit);
571 	dqp->q_rtb.hardlimit = be64_to_cpu(ddqp->d_rtb_hardlimit);
572 	dqp->q_rtb.softlimit = be64_to_cpu(ddqp->d_rtb_softlimit);
573 
574 	dqp->q_blk.count = be64_to_cpu(ddqp->d_bcount);
575 	dqp->q_ino.count = be64_to_cpu(ddqp->d_icount);
576 	dqp->q_rtb.count = be64_to_cpu(ddqp->d_rtbcount);
577 
578 	dqp->q_blk.warnings = be16_to_cpu(ddqp->d_bwarns);
579 	dqp->q_ino.warnings = be16_to_cpu(ddqp->d_iwarns);
580 	dqp->q_rtb.warnings = be16_to_cpu(ddqp->d_rtbwarns);
581 
582 	dqp->q_blk.timer = xfs_dquot_from_disk_ts(ddqp, ddqp->d_btimer);
583 	dqp->q_ino.timer = xfs_dquot_from_disk_ts(ddqp, ddqp->d_itimer);
584 	dqp->q_rtb.timer = xfs_dquot_from_disk_ts(ddqp, ddqp->d_rtbtimer);
585 
586 	/*
587 	 * Reservation counters are defined as reservation plus current usage
588 	 * to avoid having to add every time.
589 	 */
590 	dqp->q_blk.reserved = dqp->q_blk.count;
591 	dqp->q_ino.reserved = dqp->q_ino.count;
592 	dqp->q_rtb.reserved = dqp->q_rtb.count;
593 
594 	/* initialize the dquot speculative prealloc thresholds */
595 	xfs_dquot_set_prealloc_limits(dqp);
596 	return 0;
597 }
598 
599 /* Copy the in-core quota fields into the on-disk buffer. */
600 void
601 xfs_dquot_to_disk(
602 	struct xfs_disk_dquot	*ddqp,
603 	struct xfs_dquot	*dqp)
604 {
605 	ddqp->d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
606 	ddqp->d_version = XFS_DQUOT_VERSION;
607 	ddqp->d_type = dqp->q_type;
608 	ddqp->d_id = cpu_to_be32(dqp->q_id);
609 	ddqp->d_pad0 = 0;
610 	ddqp->d_pad = 0;
611 
612 	ddqp->d_blk_hardlimit = cpu_to_be64(dqp->q_blk.hardlimit);
613 	ddqp->d_blk_softlimit = cpu_to_be64(dqp->q_blk.softlimit);
614 	ddqp->d_ino_hardlimit = cpu_to_be64(dqp->q_ino.hardlimit);
615 	ddqp->d_ino_softlimit = cpu_to_be64(dqp->q_ino.softlimit);
616 	ddqp->d_rtb_hardlimit = cpu_to_be64(dqp->q_rtb.hardlimit);
617 	ddqp->d_rtb_softlimit = cpu_to_be64(dqp->q_rtb.softlimit);
618 
619 	ddqp->d_bcount = cpu_to_be64(dqp->q_blk.count);
620 	ddqp->d_icount = cpu_to_be64(dqp->q_ino.count);
621 	ddqp->d_rtbcount = cpu_to_be64(dqp->q_rtb.count);
622 
623 	ddqp->d_bwarns = cpu_to_be16(dqp->q_blk.warnings);
624 	ddqp->d_iwarns = cpu_to_be16(dqp->q_ino.warnings);
625 	ddqp->d_rtbwarns = cpu_to_be16(dqp->q_rtb.warnings);
626 
627 	ddqp->d_btimer = xfs_dquot_to_disk_ts(dqp, dqp->q_blk.timer);
628 	ddqp->d_itimer = xfs_dquot_to_disk_ts(dqp, dqp->q_ino.timer);
629 	ddqp->d_rtbtimer = xfs_dquot_to_disk_ts(dqp, dqp->q_rtb.timer);
630 }
631 
632 /* Allocate and initialize the dquot buffer for this in-core dquot. */
633 static int
634 xfs_qm_dqread_alloc(
635 	struct xfs_mount	*mp,
636 	struct xfs_dquot	*dqp,
637 	struct xfs_buf		**bpp)
638 {
639 	struct xfs_trans	*tp;
640 	int			error;
641 
642 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_qm_dqalloc,
643 			XFS_QM_DQALLOC_SPACE_RES(mp), 0, 0, &tp);
644 	if (error)
645 		goto err;
646 
647 	error = xfs_dquot_disk_alloc(&tp, dqp, bpp);
648 	if (error)
649 		goto err_cancel;
650 
651 	error = xfs_trans_commit(tp);
652 	if (error) {
653 		/*
654 		 * Buffer was held to the transaction, so we have to unlock it
655 		 * manually here because we're not passing it back.
656 		 */
657 		xfs_buf_relse(*bpp);
658 		*bpp = NULL;
659 		goto err;
660 	}
661 	return 0;
662 
663 err_cancel:
664 	xfs_trans_cancel(tp);
665 err:
666 	return error;
667 }
668 
669 /*
670  * Read in the ondisk dquot using dqtobp() then copy it to an incore version,
671  * and release the buffer immediately.  If @can_alloc is true, fill any
672  * holes in the on-disk metadata.
673  */
674 static int
675 xfs_qm_dqread(
676 	struct xfs_mount	*mp,
677 	xfs_dqid_t		id,
678 	xfs_dqtype_t		type,
679 	bool			can_alloc,
680 	struct xfs_dquot	**dqpp)
681 {
682 	struct xfs_dquot	*dqp;
683 	struct xfs_buf		*bp;
684 	int			error;
685 
686 	dqp = xfs_dquot_alloc(mp, id, type);
687 	trace_xfs_dqread(dqp);
688 
689 	/* Try to read the buffer, allocating if necessary. */
690 	error = xfs_dquot_disk_read(mp, dqp, &bp);
691 	if (error == -ENOENT && can_alloc)
692 		error = xfs_qm_dqread_alloc(mp, dqp, &bp);
693 	if (error)
694 		goto err;
695 
696 	/*
697 	 * At this point we should have a clean locked buffer.  Copy the data
698 	 * to the incore dquot and release the buffer since the incore dquot
699 	 * has its own locking protocol so we needn't tie up the buffer any
700 	 * further.
701 	 */
702 	ASSERT(xfs_buf_islocked(bp));
703 	error = xfs_dquot_from_disk(dqp, bp);
704 	xfs_buf_relse(bp);
705 	if (error)
706 		goto err;
707 
708 	*dqpp = dqp;
709 	return error;
710 
711 err:
712 	trace_xfs_dqread_fail(dqp);
713 	xfs_qm_dqdestroy(dqp);
714 	*dqpp = NULL;
715 	return error;
716 }
717 
718 /*
719  * Advance to the next id in the current chunk, or if at the
720  * end of the chunk, skip ahead to first id in next allocated chunk
721  * using the SEEK_DATA interface.
722  */
723 static int
724 xfs_dq_get_next_id(
725 	struct xfs_mount	*mp,
726 	xfs_dqtype_t		type,
727 	xfs_dqid_t		*id)
728 {
729 	struct xfs_inode	*quotip = xfs_quota_inode(mp, type);
730 	xfs_dqid_t		next_id = *id + 1; /* simple advance */
731 	uint			lock_flags;
732 	struct xfs_bmbt_irec	got;
733 	struct xfs_iext_cursor	cur;
734 	xfs_fsblock_t		start;
735 	int			error = 0;
736 
737 	/* If we'd wrap past the max ID, stop */
738 	if (next_id < *id)
739 		return -ENOENT;
740 
741 	/* If new ID is within the current chunk, advancing it sufficed */
742 	if (next_id % mp->m_quotainfo->qi_dqperchunk) {
743 		*id = next_id;
744 		return 0;
745 	}
746 
747 	/* Nope, next_id is now past the current chunk, so find the next one */
748 	start = (xfs_fsblock_t)next_id / mp->m_quotainfo->qi_dqperchunk;
749 
750 	lock_flags = xfs_ilock_data_map_shared(quotip);
751 	error = xfs_iread_extents(NULL, quotip, XFS_DATA_FORK);
752 	if (error)
753 		return error;
754 
755 	if (xfs_iext_lookup_extent(quotip, &quotip->i_df, start, &cur, &got)) {
756 		/* contiguous chunk, bump startoff for the id calculation */
757 		if (got.br_startoff < start)
758 			got.br_startoff = start;
759 		*id = got.br_startoff * mp->m_quotainfo->qi_dqperchunk;
760 	} else {
761 		error = -ENOENT;
762 	}
763 
764 	xfs_iunlock(quotip, lock_flags);
765 
766 	return error;
767 }
768 
769 /*
770  * Look up the dquot in the in-core cache.  If found, the dquot is returned
771  * locked and ready to go.
772  */
773 static struct xfs_dquot *
774 xfs_qm_dqget_cache_lookup(
775 	struct xfs_mount	*mp,
776 	struct xfs_quotainfo	*qi,
777 	struct radix_tree_root	*tree,
778 	xfs_dqid_t		id)
779 {
780 	struct xfs_dquot	*dqp;
781 
782 restart:
783 	mutex_lock(&qi->qi_tree_lock);
784 	dqp = radix_tree_lookup(tree, id);
785 	if (!dqp) {
786 		mutex_unlock(&qi->qi_tree_lock);
787 		XFS_STATS_INC(mp, xs_qm_dqcachemisses);
788 		return NULL;
789 	}
790 
791 	xfs_dqlock(dqp);
792 	if (dqp->q_flags & XFS_DQFLAG_FREEING) {
793 		xfs_dqunlock(dqp);
794 		mutex_unlock(&qi->qi_tree_lock);
795 		trace_xfs_dqget_freeing(dqp);
796 		delay(1);
797 		goto restart;
798 	}
799 
800 	dqp->q_nrefs++;
801 	mutex_unlock(&qi->qi_tree_lock);
802 
803 	trace_xfs_dqget_hit(dqp);
804 	XFS_STATS_INC(mp, xs_qm_dqcachehits);
805 	return dqp;
806 }
807 
808 /*
809  * Try to insert a new dquot into the in-core cache.  If an error occurs the
810  * caller should throw away the dquot and start over.  Otherwise, the dquot
811  * is returned locked (and held by the cache) as if there had been a cache
812  * hit.
813  */
814 static int
815 xfs_qm_dqget_cache_insert(
816 	struct xfs_mount	*mp,
817 	struct xfs_quotainfo	*qi,
818 	struct radix_tree_root	*tree,
819 	xfs_dqid_t		id,
820 	struct xfs_dquot	*dqp)
821 {
822 	int			error;
823 
824 	mutex_lock(&qi->qi_tree_lock);
825 	error = radix_tree_insert(tree, id, dqp);
826 	if (unlikely(error)) {
827 		/* Duplicate found!  Caller must try again. */
828 		WARN_ON(error != -EEXIST);
829 		mutex_unlock(&qi->qi_tree_lock);
830 		trace_xfs_dqget_dup(dqp);
831 		return error;
832 	}
833 
834 	/* Return a locked dquot to the caller, with a reference taken. */
835 	xfs_dqlock(dqp);
836 	dqp->q_nrefs = 1;
837 
838 	qi->qi_dquots++;
839 	mutex_unlock(&qi->qi_tree_lock);
840 
841 	return 0;
842 }
843 
844 /* Check our input parameters. */
845 static int
846 xfs_qm_dqget_checks(
847 	struct xfs_mount	*mp,
848 	xfs_dqtype_t		type)
849 {
850 	switch (type) {
851 	case XFS_DQTYPE_USER:
852 		if (!XFS_IS_UQUOTA_ON(mp))
853 			return -ESRCH;
854 		return 0;
855 	case XFS_DQTYPE_GROUP:
856 		if (!XFS_IS_GQUOTA_ON(mp))
857 			return -ESRCH;
858 		return 0;
859 	case XFS_DQTYPE_PROJ:
860 		if (!XFS_IS_PQUOTA_ON(mp))
861 			return -ESRCH;
862 		return 0;
863 	default:
864 		WARN_ON_ONCE(0);
865 		return -EINVAL;
866 	}
867 }
868 
869 /*
870  * Given the file system, id, and type (UDQUOT/GDQUOT/PDQUOT), return a
871  * locked dquot, doing an allocation (if requested) as needed.
872  */
873 int
874 xfs_qm_dqget(
875 	struct xfs_mount	*mp,
876 	xfs_dqid_t		id,
877 	xfs_dqtype_t		type,
878 	bool			can_alloc,
879 	struct xfs_dquot	**O_dqpp)
880 {
881 	struct xfs_quotainfo	*qi = mp->m_quotainfo;
882 	struct radix_tree_root	*tree = xfs_dquot_tree(qi, type);
883 	struct xfs_dquot	*dqp;
884 	int			error;
885 
886 	error = xfs_qm_dqget_checks(mp, type);
887 	if (error)
888 		return error;
889 
890 restart:
891 	dqp = xfs_qm_dqget_cache_lookup(mp, qi, tree, id);
892 	if (dqp) {
893 		*O_dqpp = dqp;
894 		return 0;
895 	}
896 
897 	error = xfs_qm_dqread(mp, id, type, can_alloc, &dqp);
898 	if (error)
899 		return error;
900 
901 	error = xfs_qm_dqget_cache_insert(mp, qi, tree, id, dqp);
902 	if (error) {
903 		/*
904 		 * Duplicate found. Just throw away the new dquot and start
905 		 * over.
906 		 */
907 		xfs_qm_dqdestroy(dqp);
908 		XFS_STATS_INC(mp, xs_qm_dquot_dups);
909 		goto restart;
910 	}
911 
912 	trace_xfs_dqget_miss(dqp);
913 	*O_dqpp = dqp;
914 	return 0;
915 }
916 
917 /*
918  * Given a dquot id and type, read and initialize a dquot from the on-disk
919  * metadata.  This function is only for use during quota initialization so
920  * it ignores the dquot cache assuming that the dquot shrinker isn't set up.
921  * The caller is responsible for _qm_dqdestroy'ing the returned dquot.
922  */
923 int
924 xfs_qm_dqget_uncached(
925 	struct xfs_mount	*mp,
926 	xfs_dqid_t		id,
927 	xfs_dqtype_t		type,
928 	struct xfs_dquot	**dqpp)
929 {
930 	int			error;
931 
932 	error = xfs_qm_dqget_checks(mp, type);
933 	if (error)
934 		return error;
935 
936 	return xfs_qm_dqread(mp, id, type, 0, dqpp);
937 }
938 
939 /* Return the quota id for a given inode and type. */
940 xfs_dqid_t
941 xfs_qm_id_for_quotatype(
942 	struct xfs_inode	*ip,
943 	xfs_dqtype_t		type)
944 {
945 	switch (type) {
946 	case XFS_DQTYPE_USER:
947 		return i_uid_read(VFS_I(ip));
948 	case XFS_DQTYPE_GROUP:
949 		return i_gid_read(VFS_I(ip));
950 	case XFS_DQTYPE_PROJ:
951 		return ip->i_projid;
952 	}
953 	ASSERT(0);
954 	return 0;
955 }
956 
957 /*
958  * Return the dquot for a given inode and type.  If @can_alloc is true, then
959  * allocate blocks if needed.  The inode's ILOCK must be held and it must not
960  * have already had an inode attached.
961  */
962 int
963 xfs_qm_dqget_inode(
964 	struct xfs_inode	*ip,
965 	xfs_dqtype_t		type,
966 	bool			can_alloc,
967 	struct xfs_dquot	**O_dqpp)
968 {
969 	struct xfs_mount	*mp = ip->i_mount;
970 	struct xfs_quotainfo	*qi = mp->m_quotainfo;
971 	struct radix_tree_root	*tree = xfs_dquot_tree(qi, type);
972 	struct xfs_dquot	*dqp;
973 	xfs_dqid_t		id;
974 	int			error;
975 
976 	error = xfs_qm_dqget_checks(mp, type);
977 	if (error)
978 		return error;
979 
980 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
981 	ASSERT(xfs_inode_dquot(ip, type) == NULL);
982 
983 	id = xfs_qm_id_for_quotatype(ip, type);
984 
985 restart:
986 	dqp = xfs_qm_dqget_cache_lookup(mp, qi, tree, id);
987 	if (dqp) {
988 		*O_dqpp = dqp;
989 		return 0;
990 	}
991 
992 	/*
993 	 * Dquot cache miss. We don't want to keep the inode lock across
994 	 * a (potential) disk read. Also we don't want to deal with the lock
995 	 * ordering between quotainode and this inode. OTOH, dropping the inode
996 	 * lock here means dealing with a chown that can happen before
997 	 * we re-acquire the lock.
998 	 */
999 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1000 	error = xfs_qm_dqread(mp, id, type, can_alloc, &dqp);
1001 	xfs_ilock(ip, XFS_ILOCK_EXCL);
1002 	if (error)
1003 		return error;
1004 
1005 	/*
1006 	 * A dquot could be attached to this inode by now, since we had
1007 	 * dropped the ilock.
1008 	 */
1009 	if (xfs_this_quota_on(mp, type)) {
1010 		struct xfs_dquot	*dqp1;
1011 
1012 		dqp1 = xfs_inode_dquot(ip, type);
1013 		if (dqp1) {
1014 			xfs_qm_dqdestroy(dqp);
1015 			dqp = dqp1;
1016 			xfs_dqlock(dqp);
1017 			goto dqret;
1018 		}
1019 	} else {
1020 		/* inode stays locked on return */
1021 		xfs_qm_dqdestroy(dqp);
1022 		return -ESRCH;
1023 	}
1024 
1025 	error = xfs_qm_dqget_cache_insert(mp, qi, tree, id, dqp);
1026 	if (error) {
1027 		/*
1028 		 * Duplicate found. Just throw away the new dquot and start
1029 		 * over.
1030 		 */
1031 		xfs_qm_dqdestroy(dqp);
1032 		XFS_STATS_INC(mp, xs_qm_dquot_dups);
1033 		goto restart;
1034 	}
1035 
1036 dqret:
1037 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1038 	trace_xfs_dqget_miss(dqp);
1039 	*O_dqpp = dqp;
1040 	return 0;
1041 }
1042 
1043 /*
1044  * Starting at @id and progressing upwards, look for an initialized incore
1045  * dquot, lock it, and return it.
1046  */
1047 int
1048 xfs_qm_dqget_next(
1049 	struct xfs_mount	*mp,
1050 	xfs_dqid_t		id,
1051 	xfs_dqtype_t		type,
1052 	struct xfs_dquot	**dqpp)
1053 {
1054 	struct xfs_dquot	*dqp;
1055 	int			error = 0;
1056 
1057 	*dqpp = NULL;
1058 	for (; !error; error = xfs_dq_get_next_id(mp, type, &id)) {
1059 		error = xfs_qm_dqget(mp, id, type, false, &dqp);
1060 		if (error == -ENOENT)
1061 			continue;
1062 		else if (error != 0)
1063 			break;
1064 
1065 		if (!XFS_IS_DQUOT_UNINITIALIZED(dqp)) {
1066 			*dqpp = dqp;
1067 			return 0;
1068 		}
1069 
1070 		xfs_qm_dqput(dqp);
1071 	}
1072 
1073 	return error;
1074 }
1075 
1076 /*
1077  * Release a reference to the dquot (decrement ref-count) and unlock it.
1078  *
1079  * If there is a group quota attached to this dquot, carefully release that
1080  * too without tripping over deadlocks'n'stuff.
1081  */
1082 void
1083 xfs_qm_dqput(
1084 	struct xfs_dquot	*dqp)
1085 {
1086 	ASSERT(dqp->q_nrefs > 0);
1087 	ASSERT(XFS_DQ_IS_LOCKED(dqp));
1088 
1089 	trace_xfs_dqput(dqp);
1090 
1091 	if (--dqp->q_nrefs == 0) {
1092 		struct xfs_quotainfo	*qi = dqp->q_mount->m_quotainfo;
1093 		trace_xfs_dqput_free(dqp);
1094 
1095 		if (list_lru_add(&qi->qi_lru, &dqp->q_lru))
1096 			XFS_STATS_INC(dqp->q_mount, xs_qm_dquot_unused);
1097 	}
1098 	xfs_dqunlock(dqp);
1099 }
1100 
1101 /*
1102  * Release a dquot. Flush it if dirty, then dqput() it.
1103  * dquot must not be locked.
1104  */
1105 void
1106 xfs_qm_dqrele(
1107 	struct xfs_dquot	*dqp)
1108 {
1109 	if (!dqp)
1110 		return;
1111 
1112 	trace_xfs_dqrele(dqp);
1113 
1114 	xfs_dqlock(dqp);
1115 	/*
1116 	 * We don't care to flush it if the dquot is dirty here.
1117 	 * That will create stutters that we want to avoid.
1118 	 * Instead we do a delayed write when we try to reclaim
1119 	 * a dirty dquot. Also xfs_sync will take part of the burden...
1120 	 */
1121 	xfs_qm_dqput(dqp);
1122 }
1123 
1124 /*
1125  * This is the dquot flushing I/O completion routine.  It is called
1126  * from interrupt level when the buffer containing the dquot is
1127  * flushed to disk.  It is responsible for removing the dquot logitem
1128  * from the AIL if it has not been re-logged, and unlocking the dquot's
1129  * flush lock. This behavior is very similar to that of inodes..
1130  */
1131 static void
1132 xfs_qm_dqflush_done(
1133 	struct xfs_log_item	*lip)
1134 {
1135 	struct xfs_dq_logitem	*qip = (struct xfs_dq_logitem *)lip;
1136 	struct xfs_dquot	*dqp = qip->qli_dquot;
1137 	struct xfs_ail		*ailp = lip->li_ailp;
1138 	xfs_lsn_t		tail_lsn;
1139 
1140 	/*
1141 	 * We only want to pull the item from the AIL if its
1142 	 * location in the log has not changed since we started the flush.
1143 	 * Thus, we only bother if the dquot's lsn has
1144 	 * not changed. First we check the lsn outside the lock
1145 	 * since it's cheaper, and then we recheck while
1146 	 * holding the lock before removing the dquot from the AIL.
1147 	 */
1148 	if (test_bit(XFS_LI_IN_AIL, &lip->li_flags) &&
1149 	    ((lip->li_lsn == qip->qli_flush_lsn) ||
1150 	     test_bit(XFS_LI_FAILED, &lip->li_flags))) {
1151 
1152 		spin_lock(&ailp->ail_lock);
1153 		xfs_clear_li_failed(lip);
1154 		if (lip->li_lsn == qip->qli_flush_lsn) {
1155 			/* xfs_ail_update_finish() drops the AIL lock */
1156 			tail_lsn = xfs_ail_delete_one(ailp, lip);
1157 			xfs_ail_update_finish(ailp, tail_lsn);
1158 		} else {
1159 			spin_unlock(&ailp->ail_lock);
1160 		}
1161 	}
1162 
1163 	/*
1164 	 * Release the dq's flush lock since we're done with it.
1165 	 */
1166 	xfs_dqfunlock(dqp);
1167 }
1168 
1169 void
1170 xfs_buf_dquot_iodone(
1171 	struct xfs_buf		*bp)
1172 {
1173 	struct xfs_log_item	*lip, *n;
1174 
1175 	list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) {
1176 		list_del_init(&lip->li_bio_list);
1177 		xfs_qm_dqflush_done(lip);
1178 	}
1179 }
1180 
1181 void
1182 xfs_buf_dquot_io_fail(
1183 	struct xfs_buf		*bp)
1184 {
1185 	struct xfs_log_item	*lip;
1186 
1187 	spin_lock(&bp->b_mount->m_ail->ail_lock);
1188 	list_for_each_entry(lip, &bp->b_li_list, li_bio_list)
1189 		xfs_set_li_failed(lip, bp);
1190 	spin_unlock(&bp->b_mount->m_ail->ail_lock);
1191 }
1192 
1193 /* Check incore dquot for errors before we flush. */
1194 static xfs_failaddr_t
1195 xfs_qm_dqflush_check(
1196 	struct xfs_dquot	*dqp)
1197 {
1198 	xfs_dqtype_t		type = xfs_dquot_type(dqp);
1199 
1200 	if (type != XFS_DQTYPE_USER &&
1201 	    type != XFS_DQTYPE_GROUP &&
1202 	    type != XFS_DQTYPE_PROJ)
1203 		return __this_address;
1204 
1205 	if (dqp->q_id == 0)
1206 		return NULL;
1207 
1208 	if (dqp->q_blk.softlimit && dqp->q_blk.count > dqp->q_blk.softlimit &&
1209 	    !dqp->q_blk.timer)
1210 		return __this_address;
1211 
1212 	if (dqp->q_ino.softlimit && dqp->q_ino.count > dqp->q_ino.softlimit &&
1213 	    !dqp->q_ino.timer)
1214 		return __this_address;
1215 
1216 	if (dqp->q_rtb.softlimit && dqp->q_rtb.count > dqp->q_rtb.softlimit &&
1217 	    !dqp->q_rtb.timer)
1218 		return __this_address;
1219 
1220 	/* bigtime flag should never be set on root dquots */
1221 	if (dqp->q_type & XFS_DQTYPE_BIGTIME) {
1222 		if (!xfs_has_bigtime(dqp->q_mount))
1223 			return __this_address;
1224 		if (dqp->q_id == 0)
1225 			return __this_address;
1226 	}
1227 
1228 	return NULL;
1229 }
1230 
1231 /*
1232  * Write a modified dquot to disk.
1233  * The dquot must be locked and the flush lock too taken by caller.
1234  * The flush lock will not be unlocked until the dquot reaches the disk,
1235  * but the dquot is free to be unlocked and modified by the caller
1236  * in the interim. Dquot is still locked on return. This behavior is
1237  * identical to that of inodes.
1238  */
1239 int
1240 xfs_qm_dqflush(
1241 	struct xfs_dquot	*dqp,
1242 	struct xfs_buf		**bpp)
1243 {
1244 	struct xfs_mount	*mp = dqp->q_mount;
1245 	struct xfs_log_item	*lip = &dqp->q_logitem.qli_item;
1246 	struct xfs_buf		*bp;
1247 	struct xfs_dqblk	*dqblk;
1248 	xfs_failaddr_t		fa;
1249 	int			error;
1250 
1251 	ASSERT(XFS_DQ_IS_LOCKED(dqp));
1252 	ASSERT(!completion_done(&dqp->q_flush));
1253 
1254 	trace_xfs_dqflush(dqp);
1255 
1256 	*bpp = NULL;
1257 
1258 	xfs_qm_dqunpin_wait(dqp);
1259 
1260 	/*
1261 	 * Get the buffer containing the on-disk dquot
1262 	 */
1263 	error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
1264 				   mp->m_quotainfo->qi_dqchunklen, XBF_TRYLOCK,
1265 				   &bp, &xfs_dquot_buf_ops);
1266 	if (error == -EAGAIN)
1267 		goto out_unlock;
1268 	if (error)
1269 		goto out_abort;
1270 
1271 	fa = xfs_qm_dqflush_check(dqp);
1272 	if (fa) {
1273 		xfs_alert(mp, "corrupt dquot ID 0x%x in memory at %pS",
1274 				dqp->q_id, fa);
1275 		xfs_buf_relse(bp);
1276 		error = -EFSCORRUPTED;
1277 		goto out_abort;
1278 	}
1279 
1280 	/* Flush the incore dquot to the ondisk buffer. */
1281 	dqblk = bp->b_addr + dqp->q_bufoffset;
1282 	xfs_dquot_to_disk(&dqblk->dd_diskdq, dqp);
1283 
1284 	/*
1285 	 * Clear the dirty field and remember the flush lsn for later use.
1286 	 */
1287 	dqp->q_flags &= ~XFS_DQFLAG_DIRTY;
1288 
1289 	xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn,
1290 					&dqp->q_logitem.qli_item.li_lsn);
1291 
1292 	/*
1293 	 * copy the lsn into the on-disk dquot now while we have the in memory
1294 	 * dquot here. This can't be done later in the write verifier as we
1295 	 * can't get access to the log item at that point in time.
1296 	 *
1297 	 * We also calculate the CRC here so that the on-disk dquot in the
1298 	 * buffer always has a valid CRC. This ensures there is no possibility
1299 	 * of a dquot without an up-to-date CRC getting to disk.
1300 	 */
1301 	if (xfs_has_crc(mp)) {
1302 		dqblk->dd_lsn = cpu_to_be64(dqp->q_logitem.qli_item.li_lsn);
1303 		xfs_update_cksum((char *)dqblk, sizeof(struct xfs_dqblk),
1304 				 XFS_DQUOT_CRC_OFF);
1305 	}
1306 
1307 	/*
1308 	 * Attach the dquot to the buffer so that we can remove this dquot from
1309 	 * the AIL and release the flush lock once the dquot is synced to disk.
1310 	 */
1311 	bp->b_flags |= _XBF_DQUOTS;
1312 	list_add_tail(&dqp->q_logitem.qli_item.li_bio_list, &bp->b_li_list);
1313 
1314 	/*
1315 	 * If the buffer is pinned then push on the log so we won't
1316 	 * get stuck waiting in the write for too long.
1317 	 */
1318 	if (xfs_buf_ispinned(bp)) {
1319 		trace_xfs_dqflush_force(dqp);
1320 		xfs_log_force(mp, 0);
1321 	}
1322 
1323 	trace_xfs_dqflush_done(dqp);
1324 	*bpp = bp;
1325 	return 0;
1326 
1327 out_abort:
1328 	dqp->q_flags &= ~XFS_DQFLAG_DIRTY;
1329 	xfs_trans_ail_delete(lip, 0);
1330 	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1331 out_unlock:
1332 	xfs_dqfunlock(dqp);
1333 	return error;
1334 }
1335 
1336 /*
1337  * Lock two xfs_dquot structures.
1338  *
1339  * To avoid deadlocks we always lock the quota structure with
1340  * the lowerd id first.
1341  */
1342 void
1343 xfs_dqlock2(
1344 	struct xfs_dquot	*d1,
1345 	struct xfs_dquot	*d2)
1346 {
1347 	if (d1 && d2) {
1348 		ASSERT(d1 != d2);
1349 		if (d1->q_id > d2->q_id) {
1350 			mutex_lock(&d2->q_qlock);
1351 			mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED);
1352 		} else {
1353 			mutex_lock(&d1->q_qlock);
1354 			mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED);
1355 		}
1356 	} else if (d1) {
1357 		mutex_lock(&d1->q_qlock);
1358 	} else if (d2) {
1359 		mutex_lock(&d2->q_qlock);
1360 	}
1361 }
1362 
1363 int __init
1364 xfs_qm_init(void)
1365 {
1366 	xfs_dquot_cache = kmem_cache_create("xfs_dquot",
1367 					  sizeof(struct xfs_dquot),
1368 					  0, 0, NULL);
1369 	if (!xfs_dquot_cache)
1370 		goto out;
1371 
1372 	xfs_dqtrx_cache = kmem_cache_create("xfs_dqtrx",
1373 					     sizeof(struct xfs_dquot_acct),
1374 					     0, 0, NULL);
1375 	if (!xfs_dqtrx_cache)
1376 		goto out_free_dquot_cache;
1377 
1378 	return 0;
1379 
1380 out_free_dquot_cache:
1381 	kmem_cache_destroy(xfs_dquot_cache);
1382 out:
1383 	return -ENOMEM;
1384 }
1385 
1386 void
1387 xfs_qm_exit(void)
1388 {
1389 	kmem_cache_destroy(xfs_dqtrx_cache);
1390 	kmem_cache_destroy(xfs_dquot_cache);
1391 }
1392 
1393 /*
1394  * Iterate every dquot of a particular type.  The caller must ensure that the
1395  * particular quota type is active.  iter_fn can return negative error codes,
1396  * or -ECANCELED to indicate that it wants to stop iterating.
1397  */
1398 int
1399 xfs_qm_dqiterate(
1400 	struct xfs_mount	*mp,
1401 	xfs_dqtype_t		type,
1402 	xfs_qm_dqiterate_fn	iter_fn,
1403 	void			*priv)
1404 {
1405 	struct xfs_dquot	*dq;
1406 	xfs_dqid_t		id = 0;
1407 	int			error;
1408 
1409 	do {
1410 		error = xfs_qm_dqget_next(mp, id, type, &dq);
1411 		if (error == -ENOENT)
1412 			return 0;
1413 		if (error)
1414 			return error;
1415 
1416 		error = iter_fn(dq, type, priv);
1417 		id = dq->q_id;
1418 		xfs_qm_dqput(dq);
1419 	} while (error == 0 && id != 0);
1420 
1421 	return error;
1422 }
1423