xref: /openbmc/linux/fs/xfs/xfs_dquot.c (revision 18afb028)
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 		res->timer = 0;
140 	}
141 }
142 
143 /*
144  * Check the limits and timers of a dquot and start or reset timers
145  * if necessary.
146  * This gets called even when quota enforcement is OFF, which makes our
147  * life a little less complicated. (We just don't reject any quota
148  * reservations in that case, when enforcement is off).
149  * We also return 0 as the values of the timers in Q_GETQUOTA calls, when
150  * enforcement's off.
151  * In contrast, warnings are a little different in that they don't
152  * 'automatically' get started when limits get exceeded.  They do
153  * get reset to zero, however, when we find the count to be under
154  * the soft limit (they are only ever set non-zero via userspace).
155  */
156 void
157 xfs_qm_adjust_dqtimers(
158 	struct xfs_dquot	*dq)
159 {
160 	struct xfs_mount	*mp = dq->q_mount;
161 	struct xfs_quotainfo	*qi = mp->m_quotainfo;
162 	struct xfs_def_quota	*defq;
163 
164 	ASSERT(dq->q_id);
165 	defq = xfs_get_defquota(qi, xfs_dquot_type(dq));
166 
167 	xfs_qm_adjust_res_timer(dq->q_mount, &dq->q_blk, &defq->blk);
168 	xfs_qm_adjust_res_timer(dq->q_mount, &dq->q_ino, &defq->ino);
169 	xfs_qm_adjust_res_timer(dq->q_mount, &dq->q_rtb, &defq->rtb);
170 }
171 
172 /*
173  * initialize a buffer full of dquots and log the whole thing
174  */
175 STATIC void
176 xfs_qm_init_dquot_blk(
177 	struct xfs_trans	*tp,
178 	struct xfs_mount	*mp,
179 	xfs_dqid_t		id,
180 	xfs_dqtype_t		type,
181 	struct xfs_buf		*bp)
182 {
183 	struct xfs_quotainfo	*q = mp->m_quotainfo;
184 	struct xfs_dqblk	*d;
185 	xfs_dqid_t		curid;
186 	unsigned int		qflag;
187 	unsigned int		blftype;
188 	int			i;
189 
190 	ASSERT(tp);
191 	ASSERT(xfs_buf_islocked(bp));
192 
193 	switch (type) {
194 	case XFS_DQTYPE_USER:
195 		qflag = XFS_UQUOTA_CHKD;
196 		blftype = XFS_BLF_UDQUOT_BUF;
197 		break;
198 	case XFS_DQTYPE_PROJ:
199 		qflag = XFS_PQUOTA_CHKD;
200 		blftype = XFS_BLF_PDQUOT_BUF;
201 		break;
202 	case XFS_DQTYPE_GROUP:
203 		qflag = XFS_GQUOTA_CHKD;
204 		blftype = XFS_BLF_GDQUOT_BUF;
205 		break;
206 	default:
207 		ASSERT(0);
208 		return;
209 	}
210 
211 	d = bp->b_addr;
212 
213 	/*
214 	 * ID of the first dquot in the block - id's are zero based.
215 	 */
216 	curid = id - (id % q->qi_dqperchunk);
217 	memset(d, 0, BBTOB(q->qi_dqchunklen));
218 	for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++) {
219 		d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
220 		d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
221 		d->dd_diskdq.d_id = cpu_to_be32(curid);
222 		d->dd_diskdq.d_type = type;
223 		if (curid > 0 && xfs_has_bigtime(mp))
224 			d->dd_diskdq.d_type |= XFS_DQTYPE_BIGTIME;
225 		if (xfs_has_crc(mp)) {
226 			uuid_copy(&d->dd_uuid, &mp->m_sb.sb_meta_uuid);
227 			xfs_update_cksum((char *)d, sizeof(struct xfs_dqblk),
228 					 XFS_DQUOT_CRC_OFF);
229 		}
230 	}
231 
232 	xfs_trans_dquot_buf(tp, bp, blftype);
233 
234 	/*
235 	 * quotacheck uses delayed writes to update all the dquots on disk in an
236 	 * efficient manner instead of logging the individual dquot changes as
237 	 * they are made. However if we log the buffer allocated here and crash
238 	 * after quotacheck while the logged initialisation is still in the
239 	 * active region of the log, log recovery can replay the dquot buffer
240 	 * initialisation over the top of the checked dquots and corrupt quota
241 	 * accounting.
242 	 *
243 	 * To avoid this problem, quotacheck cannot log the initialised buffer.
244 	 * We must still dirty the buffer and write it back before the
245 	 * allocation transaction clears the log. Therefore, mark the buffer as
246 	 * ordered instead of logging it directly. This is safe for quotacheck
247 	 * because it detects and repairs allocated but initialized dquot blocks
248 	 * in the quota inodes.
249 	 */
250 	if (!(mp->m_qflags & qflag))
251 		xfs_trans_ordered_buf(tp, bp);
252 	else
253 		xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1);
254 }
255 
256 /*
257  * Initialize the dynamic speculative preallocation thresholds. The lo/hi
258  * watermarks correspond to the soft and hard limits by default. If a soft limit
259  * is not specified, we use 95% of the hard limit.
260  */
261 void
262 xfs_dquot_set_prealloc_limits(struct xfs_dquot *dqp)
263 {
264 	uint64_t space;
265 
266 	dqp->q_prealloc_hi_wmark = dqp->q_blk.hardlimit;
267 	dqp->q_prealloc_lo_wmark = dqp->q_blk.softlimit;
268 	if (!dqp->q_prealloc_lo_wmark) {
269 		dqp->q_prealloc_lo_wmark = dqp->q_prealloc_hi_wmark;
270 		do_div(dqp->q_prealloc_lo_wmark, 100);
271 		dqp->q_prealloc_lo_wmark *= 95;
272 	}
273 
274 	space = dqp->q_prealloc_hi_wmark;
275 
276 	do_div(space, 100);
277 	dqp->q_low_space[XFS_QLOWSP_1_PCNT] = space;
278 	dqp->q_low_space[XFS_QLOWSP_3_PCNT] = space * 3;
279 	dqp->q_low_space[XFS_QLOWSP_5_PCNT] = space * 5;
280 }
281 
282 /*
283  * Ensure that the given in-core dquot has a buffer on disk backing it, and
284  * return the buffer locked and held. This is called when the bmapi finds a
285  * hole.
286  */
287 STATIC int
288 xfs_dquot_disk_alloc(
289 	struct xfs_dquot	*dqp,
290 	struct xfs_buf		**bpp)
291 {
292 	struct xfs_bmbt_irec	map;
293 	struct xfs_trans	*tp;
294 	struct xfs_mount	*mp = dqp->q_mount;
295 	struct xfs_buf		*bp;
296 	xfs_dqtype_t		qtype = xfs_dquot_type(dqp);
297 	struct xfs_inode	*quotip = xfs_quota_inode(mp, qtype);
298 	int			nmaps = 1;
299 	int			error;
300 
301 	trace_xfs_dqalloc(dqp);
302 
303 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_qm_dqalloc,
304 			XFS_QM_DQALLOC_SPACE_RES(mp), 0, 0, &tp);
305 	if (error)
306 		return error;
307 
308 	xfs_ilock(quotip, XFS_ILOCK_EXCL);
309 	xfs_trans_ijoin(tp, quotip, 0);
310 
311 	if (!xfs_this_quota_on(dqp->q_mount, qtype)) {
312 		/*
313 		 * Return if this type of quotas is turned off while we didn't
314 		 * have an inode lock
315 		 */
316 		error = -ESRCH;
317 		goto err_cancel;
318 	}
319 
320 	error = xfs_iext_count_may_overflow(quotip, XFS_DATA_FORK,
321 			XFS_IEXT_ADD_NOSPLIT_CNT);
322 	if (error == -EFBIG)
323 		error = xfs_iext_count_upgrade(tp, quotip,
324 				XFS_IEXT_ADD_NOSPLIT_CNT);
325 	if (error)
326 		goto err_cancel;
327 
328 	/* Create the block mapping. */
329 	error = xfs_bmapi_write(tp, quotip, dqp->q_fileoffset,
330 			XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA, 0, &map,
331 			&nmaps);
332 	if (error)
333 		goto err_cancel;
334 
335 	ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB);
336 	ASSERT(nmaps == 1);
337 	ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
338 	       (map.br_startblock != HOLESTARTBLOCK));
339 
340 	/*
341 	 * Keep track of the blkno to save a lookup later
342 	 */
343 	dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
344 
345 	/* now we can just get the buffer (there's nothing to read yet) */
346 	error = xfs_trans_get_buf(tp, mp->m_ddev_targp, dqp->q_blkno,
347 			mp->m_quotainfo->qi_dqchunklen, 0, &bp);
348 	if (error)
349 		goto err_cancel;
350 	bp->b_ops = &xfs_dquot_buf_ops;
351 
352 	/*
353 	 * Make a chunk of dquots out of this buffer and log
354 	 * the entire thing.
355 	 */
356 	xfs_qm_init_dquot_blk(tp, mp, dqp->q_id, qtype, bp);
357 	xfs_buf_set_ref(bp, XFS_DQUOT_REF);
358 
359 	/*
360 	 * Hold the buffer and join it to the dfops so that we'll still own
361 	 * the buffer when we return to the caller.  The buffer disposal on
362 	 * error must be paid attention to very carefully, as it has been
363 	 * broken since commit efa092f3d4c6 "[XFS] Fixes a bug in the quota
364 	 * code when allocating a new dquot record" in 2005, and the later
365 	 * conversion to xfs_defer_ops in commit 310a75a3c6c747 failed to keep
366 	 * the buffer locked across the _defer_finish call.  We can now do
367 	 * this correctly with xfs_defer_bjoin.
368 	 *
369 	 * Above, we allocated a disk block for the dquot information and used
370 	 * get_buf to initialize the dquot. If the _defer_finish fails, the old
371 	 * transaction is gone but the new buffer is not joined or held to any
372 	 * transaction, so we must _buf_relse it.
373 	 *
374 	 * If everything succeeds, the caller of this function is returned a
375 	 * buffer that is locked and held to the transaction.  The caller
376 	 * is responsible for unlocking any buffer passed back, either
377 	 * manually or by committing the transaction.  On error, the buffer is
378 	 * released and not passed back.
379 	 *
380 	 * Keep the quota inode ILOCKed until after the transaction commit to
381 	 * maintain the atomicity of bmap/rmap updates.
382 	 */
383 	xfs_trans_bhold(tp, bp);
384 	error = xfs_trans_commit(tp);
385 	xfs_iunlock(quotip, XFS_ILOCK_EXCL);
386 	if (error) {
387 		xfs_buf_relse(bp);
388 		return error;
389 	}
390 
391 	*bpp = bp;
392 	return 0;
393 
394 err_cancel:
395 	xfs_trans_cancel(tp);
396 	xfs_iunlock(quotip, XFS_ILOCK_EXCL);
397 	return error;
398 }
399 
400 /*
401  * Read in the in-core dquot's on-disk metadata and return the buffer.
402  * Returns ENOENT to signal a hole.
403  */
404 STATIC int
405 xfs_dquot_disk_read(
406 	struct xfs_mount	*mp,
407 	struct xfs_dquot	*dqp,
408 	struct xfs_buf		**bpp)
409 {
410 	struct xfs_bmbt_irec	map;
411 	struct xfs_buf		*bp;
412 	xfs_dqtype_t		qtype = xfs_dquot_type(dqp);
413 	struct xfs_inode	*quotip = xfs_quota_inode(mp, qtype);
414 	uint			lock_mode;
415 	int			nmaps = 1;
416 	int			error;
417 
418 	lock_mode = xfs_ilock_data_map_shared(quotip);
419 	if (!xfs_this_quota_on(mp, qtype)) {
420 		/*
421 		 * Return if this type of quotas is turned off while we
422 		 * didn't have the quota inode lock.
423 		 */
424 		xfs_iunlock(quotip, lock_mode);
425 		return -ESRCH;
426 	}
427 
428 	/*
429 	 * Find the block map; no allocations yet
430 	 */
431 	error = xfs_bmapi_read(quotip, dqp->q_fileoffset,
432 			XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0);
433 	xfs_iunlock(quotip, lock_mode);
434 	if (error)
435 		return error;
436 
437 	ASSERT(nmaps == 1);
438 	ASSERT(map.br_blockcount >= 1);
439 	ASSERT(map.br_startblock != DELAYSTARTBLOCK);
440 	if (map.br_startblock == HOLESTARTBLOCK)
441 		return -ENOENT;
442 
443 	trace_xfs_dqtobp_read(dqp);
444 
445 	/*
446 	 * store the blkno etc so that we don't have to do the
447 	 * mapping all the time
448 	 */
449 	dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
450 
451 	error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
452 			mp->m_quotainfo->qi_dqchunklen, 0, &bp,
453 			&xfs_dquot_buf_ops);
454 	if (error) {
455 		ASSERT(bp == NULL);
456 		return error;
457 	}
458 
459 	ASSERT(xfs_buf_islocked(bp));
460 	xfs_buf_set_ref(bp, XFS_DQUOT_REF);
461 	*bpp = bp;
462 
463 	return 0;
464 }
465 
466 /* Allocate and initialize everything we need for an incore dquot. */
467 STATIC struct xfs_dquot *
468 xfs_dquot_alloc(
469 	struct xfs_mount	*mp,
470 	xfs_dqid_t		id,
471 	xfs_dqtype_t		type)
472 {
473 	struct xfs_dquot	*dqp;
474 
475 	dqp = kmem_cache_zalloc(xfs_dquot_cache, GFP_KERNEL | __GFP_NOFAIL);
476 
477 	dqp->q_type = type;
478 	dqp->q_id = id;
479 	dqp->q_mount = mp;
480 	INIT_LIST_HEAD(&dqp->q_lru);
481 	mutex_init(&dqp->q_qlock);
482 	init_waitqueue_head(&dqp->q_pinwait);
483 	dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk;
484 	/*
485 	 * Offset of dquot in the (fixed sized) dquot chunk.
486 	 */
487 	dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) *
488 			sizeof(struct xfs_dqblk);
489 
490 	/*
491 	 * Because we want to use a counting completion, complete
492 	 * the flush completion once to allow a single access to
493 	 * the flush completion without blocking.
494 	 */
495 	init_completion(&dqp->q_flush);
496 	complete(&dqp->q_flush);
497 
498 	/*
499 	 * Make sure group quotas have a different lock class than user
500 	 * quotas.
501 	 */
502 	switch (type) {
503 	case XFS_DQTYPE_USER:
504 		/* uses the default lock class */
505 		break;
506 	case XFS_DQTYPE_GROUP:
507 		lockdep_set_class(&dqp->q_qlock, &xfs_dquot_group_class);
508 		break;
509 	case XFS_DQTYPE_PROJ:
510 		lockdep_set_class(&dqp->q_qlock, &xfs_dquot_project_class);
511 		break;
512 	default:
513 		ASSERT(0);
514 		break;
515 	}
516 
517 	xfs_qm_dquot_logitem_init(dqp);
518 
519 	XFS_STATS_INC(mp, xs_qm_dquot);
520 	return dqp;
521 }
522 
523 /* Check the ondisk dquot's id and type match what the incore dquot expects. */
524 static bool
525 xfs_dquot_check_type(
526 	struct xfs_dquot	*dqp,
527 	struct xfs_disk_dquot	*ddqp)
528 {
529 	uint8_t			ddqp_type;
530 	uint8_t			dqp_type;
531 
532 	ddqp_type = ddqp->d_type & XFS_DQTYPE_REC_MASK;
533 	dqp_type = xfs_dquot_type(dqp);
534 
535 	if (be32_to_cpu(ddqp->d_id) != dqp->q_id)
536 		return false;
537 
538 	/*
539 	 * V5 filesystems always expect an exact type match.  V4 filesystems
540 	 * expect an exact match for user dquots and for non-root group and
541 	 * project dquots.
542 	 */
543 	if (xfs_has_crc(dqp->q_mount) ||
544 	    dqp_type == XFS_DQTYPE_USER || dqp->q_id != 0)
545 		return ddqp_type == dqp_type;
546 
547 	/*
548 	 * V4 filesystems support either group or project quotas, but not both
549 	 * at the same time.  The non-user quota file can be switched between
550 	 * group and project quota uses depending on the mount options, which
551 	 * means that we can encounter the other type when we try to load quota
552 	 * defaults.  Quotacheck will soon reset the entire quota file
553 	 * (including the root dquot) anyway, but don't log scary corruption
554 	 * reports to dmesg.
555 	 */
556 	return ddqp_type == XFS_DQTYPE_GROUP || ddqp_type == XFS_DQTYPE_PROJ;
557 }
558 
559 /* Copy the in-core quota fields in from the on-disk buffer. */
560 STATIC int
561 xfs_dquot_from_disk(
562 	struct xfs_dquot	*dqp,
563 	struct xfs_buf		*bp)
564 {
565 	struct xfs_disk_dquot	*ddqp = bp->b_addr + dqp->q_bufoffset;
566 
567 	/*
568 	 * Ensure that we got the type and ID we were looking for.
569 	 * Everything else was checked by the dquot buffer verifier.
570 	 */
571 	if (!xfs_dquot_check_type(dqp, ddqp)) {
572 		xfs_alert_tag(bp->b_mount, XFS_PTAG_VERIFIER_ERROR,
573 			  "Metadata corruption detected at %pS, quota %u",
574 			  __this_address, dqp->q_id);
575 		xfs_alert(bp->b_mount, "Unmount and run xfs_repair");
576 		return -EFSCORRUPTED;
577 	}
578 
579 	/* copy everything from disk dquot to the incore dquot */
580 	dqp->q_type = ddqp->d_type;
581 	dqp->q_blk.hardlimit = be64_to_cpu(ddqp->d_blk_hardlimit);
582 	dqp->q_blk.softlimit = be64_to_cpu(ddqp->d_blk_softlimit);
583 	dqp->q_ino.hardlimit = be64_to_cpu(ddqp->d_ino_hardlimit);
584 	dqp->q_ino.softlimit = be64_to_cpu(ddqp->d_ino_softlimit);
585 	dqp->q_rtb.hardlimit = be64_to_cpu(ddqp->d_rtb_hardlimit);
586 	dqp->q_rtb.softlimit = be64_to_cpu(ddqp->d_rtb_softlimit);
587 
588 	dqp->q_blk.count = be64_to_cpu(ddqp->d_bcount);
589 	dqp->q_ino.count = be64_to_cpu(ddqp->d_icount);
590 	dqp->q_rtb.count = be64_to_cpu(ddqp->d_rtbcount);
591 
592 	dqp->q_blk.timer = xfs_dquot_from_disk_ts(ddqp, ddqp->d_btimer);
593 	dqp->q_ino.timer = xfs_dquot_from_disk_ts(ddqp, ddqp->d_itimer);
594 	dqp->q_rtb.timer = xfs_dquot_from_disk_ts(ddqp, ddqp->d_rtbtimer);
595 
596 	/*
597 	 * Reservation counters are defined as reservation plus current usage
598 	 * to avoid having to add every time.
599 	 */
600 	dqp->q_blk.reserved = dqp->q_blk.count;
601 	dqp->q_ino.reserved = dqp->q_ino.count;
602 	dqp->q_rtb.reserved = dqp->q_rtb.count;
603 
604 	/* initialize the dquot speculative prealloc thresholds */
605 	xfs_dquot_set_prealloc_limits(dqp);
606 	return 0;
607 }
608 
609 /* Copy the in-core quota fields into the on-disk buffer. */
610 void
611 xfs_dquot_to_disk(
612 	struct xfs_disk_dquot	*ddqp,
613 	struct xfs_dquot	*dqp)
614 {
615 	ddqp->d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
616 	ddqp->d_version = XFS_DQUOT_VERSION;
617 	ddqp->d_type = dqp->q_type;
618 	ddqp->d_id = cpu_to_be32(dqp->q_id);
619 	ddqp->d_pad0 = 0;
620 	ddqp->d_pad = 0;
621 
622 	ddqp->d_blk_hardlimit = cpu_to_be64(dqp->q_blk.hardlimit);
623 	ddqp->d_blk_softlimit = cpu_to_be64(dqp->q_blk.softlimit);
624 	ddqp->d_ino_hardlimit = cpu_to_be64(dqp->q_ino.hardlimit);
625 	ddqp->d_ino_softlimit = cpu_to_be64(dqp->q_ino.softlimit);
626 	ddqp->d_rtb_hardlimit = cpu_to_be64(dqp->q_rtb.hardlimit);
627 	ddqp->d_rtb_softlimit = cpu_to_be64(dqp->q_rtb.softlimit);
628 
629 	ddqp->d_bcount = cpu_to_be64(dqp->q_blk.count);
630 	ddqp->d_icount = cpu_to_be64(dqp->q_ino.count);
631 	ddqp->d_rtbcount = cpu_to_be64(dqp->q_rtb.count);
632 
633 	ddqp->d_bwarns = 0;
634 	ddqp->d_iwarns = 0;
635 	ddqp->d_rtbwarns = 0;
636 
637 	ddqp->d_btimer = xfs_dquot_to_disk_ts(dqp, dqp->q_blk.timer);
638 	ddqp->d_itimer = xfs_dquot_to_disk_ts(dqp, dqp->q_ino.timer);
639 	ddqp->d_rtbtimer = xfs_dquot_to_disk_ts(dqp, dqp->q_rtb.timer);
640 }
641 
642 /*
643  * Read in the ondisk dquot using dqtobp() then copy it to an incore version,
644  * and release the buffer immediately.  If @can_alloc is true, fill any
645  * holes in the on-disk metadata.
646  */
647 static int
648 xfs_qm_dqread(
649 	struct xfs_mount	*mp,
650 	xfs_dqid_t		id,
651 	xfs_dqtype_t		type,
652 	bool			can_alloc,
653 	struct xfs_dquot	**dqpp)
654 {
655 	struct xfs_dquot	*dqp;
656 	struct xfs_buf		*bp;
657 	int			error;
658 
659 	dqp = xfs_dquot_alloc(mp, id, type);
660 	trace_xfs_dqread(dqp);
661 
662 	/* Try to read the buffer, allocating if necessary. */
663 	error = xfs_dquot_disk_read(mp, dqp, &bp);
664 	if (error == -ENOENT && can_alloc)
665 		error = xfs_dquot_disk_alloc(dqp, &bp);
666 	if (error)
667 		goto err;
668 
669 	/*
670 	 * At this point we should have a clean locked buffer.  Copy the data
671 	 * to the incore dquot and release the buffer since the incore dquot
672 	 * has its own locking protocol so we needn't tie up the buffer any
673 	 * further.
674 	 */
675 	ASSERT(xfs_buf_islocked(bp));
676 	error = xfs_dquot_from_disk(dqp, bp);
677 	xfs_buf_relse(bp);
678 	if (error)
679 		goto err;
680 
681 	*dqpp = dqp;
682 	return error;
683 
684 err:
685 	trace_xfs_dqread_fail(dqp);
686 	xfs_qm_dqdestroy(dqp);
687 	*dqpp = NULL;
688 	return error;
689 }
690 
691 /*
692  * Advance to the next id in the current chunk, or if at the
693  * end of the chunk, skip ahead to first id in next allocated chunk
694  * using the SEEK_DATA interface.
695  */
696 static int
697 xfs_dq_get_next_id(
698 	struct xfs_mount	*mp,
699 	xfs_dqtype_t		type,
700 	xfs_dqid_t		*id)
701 {
702 	struct xfs_inode	*quotip = xfs_quota_inode(mp, type);
703 	xfs_dqid_t		next_id = *id + 1; /* simple advance */
704 	uint			lock_flags;
705 	struct xfs_bmbt_irec	got;
706 	struct xfs_iext_cursor	cur;
707 	xfs_fsblock_t		start;
708 	int			error = 0;
709 
710 	/* If we'd wrap past the max ID, stop */
711 	if (next_id < *id)
712 		return -ENOENT;
713 
714 	/* If new ID is within the current chunk, advancing it sufficed */
715 	if (next_id % mp->m_quotainfo->qi_dqperchunk) {
716 		*id = next_id;
717 		return 0;
718 	}
719 
720 	/* Nope, next_id is now past the current chunk, so find the next one */
721 	start = (xfs_fsblock_t)next_id / mp->m_quotainfo->qi_dqperchunk;
722 
723 	lock_flags = xfs_ilock_data_map_shared(quotip);
724 	error = xfs_iread_extents(NULL, quotip, XFS_DATA_FORK);
725 	if (error)
726 		return error;
727 
728 	if (xfs_iext_lookup_extent(quotip, &quotip->i_df, start, &cur, &got)) {
729 		/* contiguous chunk, bump startoff for the id calculation */
730 		if (got.br_startoff < start)
731 			got.br_startoff = start;
732 		*id = got.br_startoff * mp->m_quotainfo->qi_dqperchunk;
733 	} else {
734 		error = -ENOENT;
735 	}
736 
737 	xfs_iunlock(quotip, lock_flags);
738 
739 	return error;
740 }
741 
742 /*
743  * Look up the dquot in the in-core cache.  If found, the dquot is returned
744  * locked and ready to go.
745  */
746 static struct xfs_dquot *
747 xfs_qm_dqget_cache_lookup(
748 	struct xfs_mount	*mp,
749 	struct xfs_quotainfo	*qi,
750 	struct radix_tree_root	*tree,
751 	xfs_dqid_t		id)
752 {
753 	struct xfs_dquot	*dqp;
754 
755 restart:
756 	mutex_lock(&qi->qi_tree_lock);
757 	dqp = radix_tree_lookup(tree, id);
758 	if (!dqp) {
759 		mutex_unlock(&qi->qi_tree_lock);
760 		XFS_STATS_INC(mp, xs_qm_dqcachemisses);
761 		return NULL;
762 	}
763 
764 	xfs_dqlock(dqp);
765 	if (dqp->q_flags & XFS_DQFLAG_FREEING) {
766 		xfs_dqunlock(dqp);
767 		mutex_unlock(&qi->qi_tree_lock);
768 		trace_xfs_dqget_freeing(dqp);
769 		delay(1);
770 		goto restart;
771 	}
772 
773 	dqp->q_nrefs++;
774 	mutex_unlock(&qi->qi_tree_lock);
775 
776 	trace_xfs_dqget_hit(dqp);
777 	XFS_STATS_INC(mp, xs_qm_dqcachehits);
778 	return dqp;
779 }
780 
781 /*
782  * Try to insert a new dquot into the in-core cache.  If an error occurs the
783  * caller should throw away the dquot and start over.  Otherwise, the dquot
784  * is returned locked (and held by the cache) as if there had been a cache
785  * hit.
786  */
787 static int
788 xfs_qm_dqget_cache_insert(
789 	struct xfs_mount	*mp,
790 	struct xfs_quotainfo	*qi,
791 	struct radix_tree_root	*tree,
792 	xfs_dqid_t		id,
793 	struct xfs_dquot	*dqp)
794 {
795 	int			error;
796 
797 	mutex_lock(&qi->qi_tree_lock);
798 	error = radix_tree_insert(tree, id, dqp);
799 	if (unlikely(error)) {
800 		/* Duplicate found!  Caller must try again. */
801 		mutex_unlock(&qi->qi_tree_lock);
802 		trace_xfs_dqget_dup(dqp);
803 		return error;
804 	}
805 
806 	/* Return a locked dquot to the caller, with a reference taken. */
807 	xfs_dqlock(dqp);
808 	dqp->q_nrefs = 1;
809 
810 	qi->qi_dquots++;
811 	mutex_unlock(&qi->qi_tree_lock);
812 
813 	return 0;
814 }
815 
816 /* Check our input parameters. */
817 static int
818 xfs_qm_dqget_checks(
819 	struct xfs_mount	*mp,
820 	xfs_dqtype_t		type)
821 {
822 	switch (type) {
823 	case XFS_DQTYPE_USER:
824 		if (!XFS_IS_UQUOTA_ON(mp))
825 			return -ESRCH;
826 		return 0;
827 	case XFS_DQTYPE_GROUP:
828 		if (!XFS_IS_GQUOTA_ON(mp))
829 			return -ESRCH;
830 		return 0;
831 	case XFS_DQTYPE_PROJ:
832 		if (!XFS_IS_PQUOTA_ON(mp))
833 			return -ESRCH;
834 		return 0;
835 	default:
836 		WARN_ON_ONCE(0);
837 		return -EINVAL;
838 	}
839 }
840 
841 /*
842  * Given the file system, id, and type (UDQUOT/GDQUOT/PDQUOT), return a
843  * locked dquot, doing an allocation (if requested) as needed.
844  */
845 int
846 xfs_qm_dqget(
847 	struct xfs_mount	*mp,
848 	xfs_dqid_t		id,
849 	xfs_dqtype_t		type,
850 	bool			can_alloc,
851 	struct xfs_dquot	**O_dqpp)
852 {
853 	struct xfs_quotainfo	*qi = mp->m_quotainfo;
854 	struct radix_tree_root	*tree = xfs_dquot_tree(qi, type);
855 	struct xfs_dquot	*dqp;
856 	int			error;
857 
858 	error = xfs_qm_dqget_checks(mp, type);
859 	if (error)
860 		return error;
861 
862 restart:
863 	dqp = xfs_qm_dqget_cache_lookup(mp, qi, tree, id);
864 	if (dqp) {
865 		*O_dqpp = dqp;
866 		return 0;
867 	}
868 
869 	error = xfs_qm_dqread(mp, id, type, can_alloc, &dqp);
870 	if (error)
871 		return error;
872 
873 	error = xfs_qm_dqget_cache_insert(mp, qi, tree, id, dqp);
874 	if (error) {
875 		/*
876 		 * Duplicate found. Just throw away the new dquot and start
877 		 * over.
878 		 */
879 		xfs_qm_dqdestroy(dqp);
880 		XFS_STATS_INC(mp, xs_qm_dquot_dups);
881 		goto restart;
882 	}
883 
884 	trace_xfs_dqget_miss(dqp);
885 	*O_dqpp = dqp;
886 	return 0;
887 }
888 
889 /*
890  * Given a dquot id and type, read and initialize a dquot from the on-disk
891  * metadata.  This function is only for use during quota initialization so
892  * it ignores the dquot cache assuming that the dquot shrinker isn't set up.
893  * The caller is responsible for _qm_dqdestroy'ing the returned dquot.
894  */
895 int
896 xfs_qm_dqget_uncached(
897 	struct xfs_mount	*mp,
898 	xfs_dqid_t		id,
899 	xfs_dqtype_t		type,
900 	struct xfs_dquot	**dqpp)
901 {
902 	int			error;
903 
904 	error = xfs_qm_dqget_checks(mp, type);
905 	if (error)
906 		return error;
907 
908 	return xfs_qm_dqread(mp, id, type, 0, dqpp);
909 }
910 
911 /* Return the quota id for a given inode and type. */
912 xfs_dqid_t
913 xfs_qm_id_for_quotatype(
914 	struct xfs_inode	*ip,
915 	xfs_dqtype_t		type)
916 {
917 	switch (type) {
918 	case XFS_DQTYPE_USER:
919 		return i_uid_read(VFS_I(ip));
920 	case XFS_DQTYPE_GROUP:
921 		return i_gid_read(VFS_I(ip));
922 	case XFS_DQTYPE_PROJ:
923 		return ip->i_projid;
924 	}
925 	ASSERT(0);
926 	return 0;
927 }
928 
929 /*
930  * Return the dquot for a given inode and type.  If @can_alloc is true, then
931  * allocate blocks if needed.  The inode's ILOCK must be held and it must not
932  * have already had an inode attached.
933  */
934 int
935 xfs_qm_dqget_inode(
936 	struct xfs_inode	*ip,
937 	xfs_dqtype_t		type,
938 	bool			can_alloc,
939 	struct xfs_dquot	**O_dqpp)
940 {
941 	struct xfs_mount	*mp = ip->i_mount;
942 	struct xfs_quotainfo	*qi = mp->m_quotainfo;
943 	struct radix_tree_root	*tree = xfs_dquot_tree(qi, type);
944 	struct xfs_dquot	*dqp;
945 	xfs_dqid_t		id;
946 	int			error;
947 
948 	error = xfs_qm_dqget_checks(mp, type);
949 	if (error)
950 		return error;
951 
952 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
953 	ASSERT(xfs_inode_dquot(ip, type) == NULL);
954 
955 	id = xfs_qm_id_for_quotatype(ip, type);
956 
957 restart:
958 	dqp = xfs_qm_dqget_cache_lookup(mp, qi, tree, id);
959 	if (dqp) {
960 		*O_dqpp = dqp;
961 		return 0;
962 	}
963 
964 	/*
965 	 * Dquot cache miss. We don't want to keep the inode lock across
966 	 * a (potential) disk read. Also we don't want to deal with the lock
967 	 * ordering between quotainode and this inode. OTOH, dropping the inode
968 	 * lock here means dealing with a chown that can happen before
969 	 * we re-acquire the lock.
970 	 */
971 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
972 	error = xfs_qm_dqread(mp, id, type, can_alloc, &dqp);
973 	xfs_ilock(ip, XFS_ILOCK_EXCL);
974 	if (error)
975 		return error;
976 
977 	/*
978 	 * A dquot could be attached to this inode by now, since we had
979 	 * dropped the ilock.
980 	 */
981 	if (xfs_this_quota_on(mp, type)) {
982 		struct xfs_dquot	*dqp1;
983 
984 		dqp1 = xfs_inode_dquot(ip, type);
985 		if (dqp1) {
986 			xfs_qm_dqdestroy(dqp);
987 			dqp = dqp1;
988 			xfs_dqlock(dqp);
989 			goto dqret;
990 		}
991 	} else {
992 		/* inode stays locked on return */
993 		xfs_qm_dqdestroy(dqp);
994 		return -ESRCH;
995 	}
996 
997 	error = xfs_qm_dqget_cache_insert(mp, qi, tree, id, dqp);
998 	if (error) {
999 		/*
1000 		 * Duplicate found. Just throw away the new dquot and start
1001 		 * over.
1002 		 */
1003 		xfs_qm_dqdestroy(dqp);
1004 		XFS_STATS_INC(mp, xs_qm_dquot_dups);
1005 		goto restart;
1006 	}
1007 
1008 dqret:
1009 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1010 	trace_xfs_dqget_miss(dqp);
1011 	*O_dqpp = dqp;
1012 	return 0;
1013 }
1014 
1015 /*
1016  * Starting at @id and progressing upwards, look for an initialized incore
1017  * dquot, lock it, and return it.
1018  */
1019 int
1020 xfs_qm_dqget_next(
1021 	struct xfs_mount	*mp,
1022 	xfs_dqid_t		id,
1023 	xfs_dqtype_t		type,
1024 	struct xfs_dquot	**dqpp)
1025 {
1026 	struct xfs_dquot	*dqp;
1027 	int			error = 0;
1028 
1029 	*dqpp = NULL;
1030 	for (; !error; error = xfs_dq_get_next_id(mp, type, &id)) {
1031 		error = xfs_qm_dqget(mp, id, type, false, &dqp);
1032 		if (error == -ENOENT)
1033 			continue;
1034 		else if (error != 0)
1035 			break;
1036 
1037 		if (!XFS_IS_DQUOT_UNINITIALIZED(dqp)) {
1038 			*dqpp = dqp;
1039 			return 0;
1040 		}
1041 
1042 		xfs_qm_dqput(dqp);
1043 	}
1044 
1045 	return error;
1046 }
1047 
1048 /*
1049  * Release a reference to the dquot (decrement ref-count) and unlock it.
1050  *
1051  * If there is a group quota attached to this dquot, carefully release that
1052  * too without tripping over deadlocks'n'stuff.
1053  */
1054 void
1055 xfs_qm_dqput(
1056 	struct xfs_dquot	*dqp)
1057 {
1058 	ASSERT(dqp->q_nrefs > 0);
1059 	ASSERT(XFS_DQ_IS_LOCKED(dqp));
1060 
1061 	trace_xfs_dqput(dqp);
1062 
1063 	if (--dqp->q_nrefs == 0) {
1064 		struct xfs_quotainfo	*qi = dqp->q_mount->m_quotainfo;
1065 		trace_xfs_dqput_free(dqp);
1066 
1067 		if (list_lru_add(&qi->qi_lru, &dqp->q_lru))
1068 			XFS_STATS_INC(dqp->q_mount, xs_qm_dquot_unused);
1069 	}
1070 	xfs_dqunlock(dqp);
1071 }
1072 
1073 /*
1074  * Release a dquot. Flush it if dirty, then dqput() it.
1075  * dquot must not be locked.
1076  */
1077 void
1078 xfs_qm_dqrele(
1079 	struct xfs_dquot	*dqp)
1080 {
1081 	if (!dqp)
1082 		return;
1083 
1084 	trace_xfs_dqrele(dqp);
1085 
1086 	xfs_dqlock(dqp);
1087 	/*
1088 	 * We don't care to flush it if the dquot is dirty here.
1089 	 * That will create stutters that we want to avoid.
1090 	 * Instead we do a delayed write when we try to reclaim
1091 	 * a dirty dquot. Also xfs_sync will take part of the burden...
1092 	 */
1093 	xfs_qm_dqput(dqp);
1094 }
1095 
1096 /*
1097  * This is the dquot flushing I/O completion routine.  It is called
1098  * from interrupt level when the buffer containing the dquot is
1099  * flushed to disk.  It is responsible for removing the dquot logitem
1100  * from the AIL if it has not been re-logged, and unlocking the dquot's
1101  * flush lock. This behavior is very similar to that of inodes..
1102  */
1103 static void
1104 xfs_qm_dqflush_done(
1105 	struct xfs_log_item	*lip)
1106 {
1107 	struct xfs_dq_logitem	*qip = (struct xfs_dq_logitem *)lip;
1108 	struct xfs_dquot	*dqp = qip->qli_dquot;
1109 	struct xfs_ail		*ailp = lip->li_ailp;
1110 	xfs_lsn_t		tail_lsn;
1111 
1112 	/*
1113 	 * We only want to pull the item from the AIL if its
1114 	 * location in the log has not changed since we started the flush.
1115 	 * Thus, we only bother if the dquot's lsn has
1116 	 * not changed. First we check the lsn outside the lock
1117 	 * since it's cheaper, and then we recheck while
1118 	 * holding the lock before removing the dquot from the AIL.
1119 	 */
1120 	if (test_bit(XFS_LI_IN_AIL, &lip->li_flags) &&
1121 	    ((lip->li_lsn == qip->qli_flush_lsn) ||
1122 	     test_bit(XFS_LI_FAILED, &lip->li_flags))) {
1123 
1124 		spin_lock(&ailp->ail_lock);
1125 		xfs_clear_li_failed(lip);
1126 		if (lip->li_lsn == qip->qli_flush_lsn) {
1127 			/* xfs_ail_update_finish() drops the AIL lock */
1128 			tail_lsn = xfs_ail_delete_one(ailp, lip);
1129 			xfs_ail_update_finish(ailp, tail_lsn);
1130 		} else {
1131 			spin_unlock(&ailp->ail_lock);
1132 		}
1133 	}
1134 
1135 	/*
1136 	 * Release the dq's flush lock since we're done with it.
1137 	 */
1138 	xfs_dqfunlock(dqp);
1139 }
1140 
1141 void
1142 xfs_buf_dquot_iodone(
1143 	struct xfs_buf		*bp)
1144 {
1145 	struct xfs_log_item	*lip, *n;
1146 
1147 	list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) {
1148 		list_del_init(&lip->li_bio_list);
1149 		xfs_qm_dqflush_done(lip);
1150 	}
1151 }
1152 
1153 void
1154 xfs_buf_dquot_io_fail(
1155 	struct xfs_buf		*bp)
1156 {
1157 	struct xfs_log_item	*lip;
1158 
1159 	spin_lock(&bp->b_mount->m_ail->ail_lock);
1160 	list_for_each_entry(lip, &bp->b_li_list, li_bio_list)
1161 		xfs_set_li_failed(lip, bp);
1162 	spin_unlock(&bp->b_mount->m_ail->ail_lock);
1163 }
1164 
1165 /* Check incore dquot for errors before we flush. */
1166 static xfs_failaddr_t
1167 xfs_qm_dqflush_check(
1168 	struct xfs_dquot	*dqp)
1169 {
1170 	xfs_dqtype_t		type = xfs_dquot_type(dqp);
1171 
1172 	if (type != XFS_DQTYPE_USER &&
1173 	    type != XFS_DQTYPE_GROUP &&
1174 	    type != XFS_DQTYPE_PROJ)
1175 		return __this_address;
1176 
1177 	if (dqp->q_id == 0)
1178 		return NULL;
1179 
1180 	if (dqp->q_blk.softlimit && dqp->q_blk.count > dqp->q_blk.softlimit &&
1181 	    !dqp->q_blk.timer)
1182 		return __this_address;
1183 
1184 	if (dqp->q_ino.softlimit && dqp->q_ino.count > dqp->q_ino.softlimit &&
1185 	    !dqp->q_ino.timer)
1186 		return __this_address;
1187 
1188 	if (dqp->q_rtb.softlimit && dqp->q_rtb.count > dqp->q_rtb.softlimit &&
1189 	    !dqp->q_rtb.timer)
1190 		return __this_address;
1191 
1192 	/* bigtime flag should never be set on root dquots */
1193 	if (dqp->q_type & XFS_DQTYPE_BIGTIME) {
1194 		if (!xfs_has_bigtime(dqp->q_mount))
1195 			return __this_address;
1196 		if (dqp->q_id == 0)
1197 			return __this_address;
1198 	}
1199 
1200 	return NULL;
1201 }
1202 
1203 /*
1204  * Write a modified dquot to disk.
1205  * The dquot must be locked and the flush lock too taken by caller.
1206  * The flush lock will not be unlocked until the dquot reaches the disk,
1207  * but the dquot is free to be unlocked and modified by the caller
1208  * in the interim. Dquot is still locked on return. This behavior is
1209  * identical to that of inodes.
1210  */
1211 int
1212 xfs_qm_dqflush(
1213 	struct xfs_dquot	*dqp,
1214 	struct xfs_buf		**bpp)
1215 {
1216 	struct xfs_mount	*mp = dqp->q_mount;
1217 	struct xfs_log_item	*lip = &dqp->q_logitem.qli_item;
1218 	struct xfs_buf		*bp;
1219 	struct xfs_dqblk	*dqblk;
1220 	xfs_failaddr_t		fa;
1221 	int			error;
1222 
1223 	ASSERT(XFS_DQ_IS_LOCKED(dqp));
1224 	ASSERT(!completion_done(&dqp->q_flush));
1225 
1226 	trace_xfs_dqflush(dqp);
1227 
1228 	*bpp = NULL;
1229 
1230 	xfs_qm_dqunpin_wait(dqp);
1231 
1232 	/*
1233 	 * Get the buffer containing the on-disk dquot
1234 	 */
1235 	error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
1236 				   mp->m_quotainfo->qi_dqchunklen, XBF_TRYLOCK,
1237 				   &bp, &xfs_dquot_buf_ops);
1238 	if (error == -EAGAIN)
1239 		goto out_unlock;
1240 	if (error)
1241 		goto out_abort;
1242 
1243 	fa = xfs_qm_dqflush_check(dqp);
1244 	if (fa) {
1245 		xfs_alert(mp, "corrupt dquot ID 0x%x in memory at %pS",
1246 				dqp->q_id, fa);
1247 		xfs_buf_relse(bp);
1248 		error = -EFSCORRUPTED;
1249 		goto out_abort;
1250 	}
1251 
1252 	/* Flush the incore dquot to the ondisk buffer. */
1253 	dqblk = bp->b_addr + dqp->q_bufoffset;
1254 	xfs_dquot_to_disk(&dqblk->dd_diskdq, dqp);
1255 
1256 	/*
1257 	 * Clear the dirty field and remember the flush lsn for later use.
1258 	 */
1259 	dqp->q_flags &= ~XFS_DQFLAG_DIRTY;
1260 
1261 	xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn,
1262 					&dqp->q_logitem.qli_item.li_lsn);
1263 
1264 	/*
1265 	 * copy the lsn into the on-disk dquot now while we have the in memory
1266 	 * dquot here. This can't be done later in the write verifier as we
1267 	 * can't get access to the log item at that point in time.
1268 	 *
1269 	 * We also calculate the CRC here so that the on-disk dquot in the
1270 	 * buffer always has a valid CRC. This ensures there is no possibility
1271 	 * of a dquot without an up-to-date CRC getting to disk.
1272 	 */
1273 	if (xfs_has_crc(mp)) {
1274 		dqblk->dd_lsn = cpu_to_be64(dqp->q_logitem.qli_item.li_lsn);
1275 		xfs_update_cksum((char *)dqblk, sizeof(struct xfs_dqblk),
1276 				 XFS_DQUOT_CRC_OFF);
1277 	}
1278 
1279 	/*
1280 	 * Attach the dquot to the buffer so that we can remove this dquot from
1281 	 * the AIL and release the flush lock once the dquot is synced to disk.
1282 	 */
1283 	bp->b_flags |= _XBF_DQUOTS;
1284 	list_add_tail(&dqp->q_logitem.qli_item.li_bio_list, &bp->b_li_list);
1285 
1286 	/*
1287 	 * If the buffer is pinned then push on the log so we won't
1288 	 * get stuck waiting in the write for too long.
1289 	 */
1290 	if (xfs_buf_ispinned(bp)) {
1291 		trace_xfs_dqflush_force(dqp);
1292 		xfs_log_force(mp, 0);
1293 	}
1294 
1295 	trace_xfs_dqflush_done(dqp);
1296 	*bpp = bp;
1297 	return 0;
1298 
1299 out_abort:
1300 	dqp->q_flags &= ~XFS_DQFLAG_DIRTY;
1301 	xfs_trans_ail_delete(lip, 0);
1302 	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1303 out_unlock:
1304 	xfs_dqfunlock(dqp);
1305 	return error;
1306 }
1307 
1308 /*
1309  * Lock two xfs_dquot structures.
1310  *
1311  * To avoid deadlocks we always lock the quota structure with
1312  * the lowerd id first.
1313  */
1314 void
1315 xfs_dqlock2(
1316 	struct xfs_dquot	*d1,
1317 	struct xfs_dquot	*d2)
1318 {
1319 	if (d1 && d2) {
1320 		ASSERT(d1 != d2);
1321 		if (d1->q_id > d2->q_id) {
1322 			mutex_lock(&d2->q_qlock);
1323 			mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED);
1324 		} else {
1325 			mutex_lock(&d1->q_qlock);
1326 			mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED);
1327 		}
1328 	} else if (d1) {
1329 		mutex_lock(&d1->q_qlock);
1330 	} else if (d2) {
1331 		mutex_lock(&d2->q_qlock);
1332 	}
1333 }
1334 
1335 int __init
1336 xfs_qm_init(void)
1337 {
1338 	xfs_dquot_cache = kmem_cache_create("xfs_dquot",
1339 					  sizeof(struct xfs_dquot),
1340 					  0, 0, NULL);
1341 	if (!xfs_dquot_cache)
1342 		goto out;
1343 
1344 	xfs_dqtrx_cache = kmem_cache_create("xfs_dqtrx",
1345 					     sizeof(struct xfs_dquot_acct),
1346 					     0, 0, NULL);
1347 	if (!xfs_dqtrx_cache)
1348 		goto out_free_dquot_cache;
1349 
1350 	return 0;
1351 
1352 out_free_dquot_cache:
1353 	kmem_cache_destroy(xfs_dquot_cache);
1354 out:
1355 	return -ENOMEM;
1356 }
1357 
1358 void
1359 xfs_qm_exit(void)
1360 {
1361 	kmem_cache_destroy(xfs_dqtrx_cache);
1362 	kmem_cache_destroy(xfs_dquot_cache);
1363 }
1364 
1365 /*
1366  * Iterate every dquot of a particular type.  The caller must ensure that the
1367  * particular quota type is active.  iter_fn can return negative error codes,
1368  * or -ECANCELED to indicate that it wants to stop iterating.
1369  */
1370 int
1371 xfs_qm_dqiterate(
1372 	struct xfs_mount	*mp,
1373 	xfs_dqtype_t		type,
1374 	xfs_qm_dqiterate_fn	iter_fn,
1375 	void			*priv)
1376 {
1377 	struct xfs_dquot	*dq;
1378 	xfs_dqid_t		id = 0;
1379 	int			error;
1380 
1381 	do {
1382 		error = xfs_qm_dqget_next(mp, id, type, &dq);
1383 		if (error == -ENOENT)
1384 			return 0;
1385 		if (error)
1386 			return error;
1387 
1388 		error = iter_fn(dq, type, priv);
1389 		id = dq->q_id + 1;
1390 		xfs_qm_dqput(dq);
1391 	} while (error == 0 && id != 0);
1392 
1393 	return error;
1394 }
1395