xref: /linux/fs/xfs/xfs_dquot.c (revision 962fad301c33dec69324dc2d9320fd84a119a24c)
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_zone		*xfs_qm_dqtrxzone;
42 static struct kmem_zone		*xfs_qm_dqzone;
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_qm_dqzone, 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 /*
102  * Determine if this quota counter is over either limit and set the quota
103  * timers as appropriate.
104  */
105 static inline void
106 xfs_qm_adjust_res_timer(
107 	struct xfs_dquot_res	*res,
108 	struct xfs_quota_limits	*qlim)
109 {
110 	ASSERT(res->hardlimit == 0 || res->softlimit <= res->hardlimit);
111 
112 	if ((res->softlimit && res->count > res->softlimit) ||
113 	    (res->hardlimit && res->count > res->hardlimit)) {
114 		if (res->timer == 0)
115 			res->timer = ktime_get_real_seconds() + qlim->time;
116 	} else {
117 		if (res->timer == 0)
118 			res->warnings = 0;
119 		else
120 			res->timer = 0;
121 	}
122 }
123 
124 /*
125  * Check the limits and timers of a dquot and start or reset timers
126  * if necessary.
127  * This gets called even when quota enforcement is OFF, which makes our
128  * life a little less complicated. (We just don't reject any quota
129  * reservations in that case, when enforcement is off).
130  * We also return 0 as the values of the timers in Q_GETQUOTA calls, when
131  * enforcement's off.
132  * In contrast, warnings are a little different in that they don't
133  * 'automatically' get started when limits get exceeded.  They do
134  * get reset to zero, however, when we find the count to be under
135  * the soft limit (they are only ever set non-zero via userspace).
136  */
137 void
138 xfs_qm_adjust_dqtimers(
139 	struct xfs_dquot	*dq)
140 {
141 	struct xfs_mount	*mp = dq->q_mount;
142 	struct xfs_quotainfo	*qi = mp->m_quotainfo;
143 	struct xfs_def_quota	*defq;
144 
145 	ASSERT(dq->q_id);
146 	defq = xfs_get_defquota(qi, xfs_dquot_type(dq));
147 
148 	xfs_qm_adjust_res_timer(&dq->q_blk, &defq->blk);
149 	xfs_qm_adjust_res_timer(&dq->q_ino, &defq->ino);
150 	xfs_qm_adjust_res_timer(&dq->q_rtb, &defq->rtb);
151 }
152 
153 /*
154  * initialize a buffer full of dquots and log the whole thing
155  */
156 STATIC void
157 xfs_qm_init_dquot_blk(
158 	struct xfs_trans	*tp,
159 	struct xfs_mount	*mp,
160 	xfs_dqid_t		id,
161 	xfs_dqtype_t		type,
162 	struct xfs_buf		*bp)
163 {
164 	struct xfs_quotainfo	*q = mp->m_quotainfo;
165 	struct xfs_dqblk	*d;
166 	xfs_dqid_t		curid;
167 	unsigned int		qflag;
168 	unsigned int		blftype;
169 	int			i;
170 
171 	ASSERT(tp);
172 	ASSERT(xfs_buf_islocked(bp));
173 
174 	switch (type) {
175 	case XFS_DQTYPE_USER:
176 		qflag = XFS_UQUOTA_CHKD;
177 		blftype = XFS_BLF_UDQUOT_BUF;
178 		break;
179 	case XFS_DQTYPE_PROJ:
180 		qflag = XFS_PQUOTA_CHKD;
181 		blftype = XFS_BLF_PDQUOT_BUF;
182 		break;
183 	case XFS_DQTYPE_GROUP:
184 		qflag = XFS_GQUOTA_CHKD;
185 		blftype = XFS_BLF_GDQUOT_BUF;
186 		break;
187 	default:
188 		ASSERT(0);
189 		return;
190 	}
191 
192 	d = bp->b_addr;
193 
194 	/*
195 	 * ID of the first dquot in the block - id's are zero based.
196 	 */
197 	curid = id - (id % q->qi_dqperchunk);
198 	memset(d, 0, BBTOB(q->qi_dqchunklen));
199 	for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++) {
200 		d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
201 		d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
202 		d->dd_diskdq.d_id = cpu_to_be32(curid);
203 		d->dd_diskdq.d_type = type;
204 		if (xfs_sb_version_hascrc(&mp->m_sb)) {
205 			uuid_copy(&d->dd_uuid, &mp->m_sb.sb_meta_uuid);
206 			xfs_update_cksum((char *)d, sizeof(struct xfs_dqblk),
207 					 XFS_DQUOT_CRC_OFF);
208 		}
209 	}
210 
211 	xfs_trans_dquot_buf(tp, bp, blftype);
212 
213 	/*
214 	 * quotacheck uses delayed writes to update all the dquots on disk in an
215 	 * efficient manner instead of logging the individual dquot changes as
216 	 * they are made. However if we log the buffer allocated here and crash
217 	 * after quotacheck while the logged initialisation is still in the
218 	 * active region of the log, log recovery can replay the dquot buffer
219 	 * initialisation over the top of the checked dquots and corrupt quota
220 	 * accounting.
221 	 *
222 	 * To avoid this problem, quotacheck cannot log the initialised buffer.
223 	 * We must still dirty the buffer and write it back before the
224 	 * allocation transaction clears the log. Therefore, mark the buffer as
225 	 * ordered instead of logging it directly. This is safe for quotacheck
226 	 * because it detects and repairs allocated but initialized dquot blocks
227 	 * in the quota inodes.
228 	 */
229 	if (!(mp->m_qflags & qflag))
230 		xfs_trans_ordered_buf(tp, bp);
231 	else
232 		xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1);
233 }
234 
235 /*
236  * Initialize the dynamic speculative preallocation thresholds. The lo/hi
237  * watermarks correspond to the soft and hard limits by default. If a soft limit
238  * is not specified, we use 95% of the hard limit.
239  */
240 void
241 xfs_dquot_set_prealloc_limits(struct xfs_dquot *dqp)
242 {
243 	uint64_t space;
244 
245 	dqp->q_prealloc_hi_wmark = dqp->q_blk.hardlimit;
246 	dqp->q_prealloc_lo_wmark = dqp->q_blk.softlimit;
247 	if (!dqp->q_prealloc_lo_wmark) {
248 		dqp->q_prealloc_lo_wmark = dqp->q_prealloc_hi_wmark;
249 		do_div(dqp->q_prealloc_lo_wmark, 100);
250 		dqp->q_prealloc_lo_wmark *= 95;
251 	}
252 
253 	space = dqp->q_prealloc_hi_wmark;
254 
255 	do_div(space, 100);
256 	dqp->q_low_space[XFS_QLOWSP_1_PCNT] = space;
257 	dqp->q_low_space[XFS_QLOWSP_3_PCNT] = space * 3;
258 	dqp->q_low_space[XFS_QLOWSP_5_PCNT] = space * 5;
259 }
260 
261 /*
262  * Ensure that the given in-core dquot has a buffer on disk backing it, and
263  * return the buffer locked and held. This is called when the bmapi finds a
264  * hole.
265  */
266 STATIC int
267 xfs_dquot_disk_alloc(
268 	struct xfs_trans	**tpp,
269 	struct xfs_dquot	*dqp,
270 	struct xfs_buf		**bpp)
271 {
272 	struct xfs_bmbt_irec	map;
273 	struct xfs_trans	*tp = *tpp;
274 	struct xfs_mount	*mp = tp->t_mountp;
275 	struct xfs_buf		*bp;
276 	xfs_dqtype_t		qtype = xfs_dquot_type(dqp);
277 	struct xfs_inode	*quotip = xfs_quota_inode(mp, qtype);
278 	int			nmaps = 1;
279 	int			error;
280 
281 	trace_xfs_dqalloc(dqp);
282 
283 	xfs_ilock(quotip, XFS_ILOCK_EXCL);
284 	if (!xfs_this_quota_on(dqp->q_mount, qtype)) {
285 		/*
286 		 * Return if this type of quotas is turned off while we didn't
287 		 * have an inode lock
288 		 */
289 		xfs_iunlock(quotip, XFS_ILOCK_EXCL);
290 		return -ESRCH;
291 	}
292 
293 	/* Create the block mapping. */
294 	xfs_trans_ijoin(tp, quotip, XFS_ILOCK_EXCL);
295 	error = xfs_bmapi_write(tp, quotip, dqp->q_fileoffset,
296 			XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA, 0, &map,
297 			&nmaps);
298 	if (error)
299 		return error;
300 	ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB);
301 	ASSERT(nmaps == 1);
302 	ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
303 	       (map.br_startblock != HOLESTARTBLOCK));
304 
305 	/*
306 	 * Keep track of the blkno to save a lookup later
307 	 */
308 	dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
309 
310 	/* now we can just get the buffer (there's nothing to read yet) */
311 	error = xfs_trans_get_buf(tp, mp->m_ddev_targp, dqp->q_blkno,
312 			mp->m_quotainfo->qi_dqchunklen, 0, &bp);
313 	if (error)
314 		return error;
315 	bp->b_ops = &xfs_dquot_buf_ops;
316 
317 	/*
318 	 * Make a chunk of dquots out of this buffer and log
319 	 * the entire thing.
320 	 */
321 	xfs_qm_init_dquot_blk(tp, mp, dqp->q_id, qtype, bp);
322 	xfs_buf_set_ref(bp, XFS_DQUOT_REF);
323 
324 	/*
325 	 * Hold the buffer and join it to the dfops so that we'll still own
326 	 * the buffer when we return to the caller.  The buffer disposal on
327 	 * error must be paid attention to very carefully, as it has been
328 	 * broken since commit efa092f3d4c6 "[XFS] Fixes a bug in the quota
329 	 * code when allocating a new dquot record" in 2005, and the later
330 	 * conversion to xfs_defer_ops in commit 310a75a3c6c747 failed to keep
331 	 * the buffer locked across the _defer_finish call.  We can now do
332 	 * this correctly with xfs_defer_bjoin.
333 	 *
334 	 * Above, we allocated a disk block for the dquot information and used
335 	 * get_buf to initialize the dquot. If the _defer_finish fails, the old
336 	 * transaction is gone but the new buffer is not joined or held to any
337 	 * transaction, so we must _buf_relse it.
338 	 *
339 	 * If everything succeeds, the caller of this function is returned a
340 	 * buffer that is locked and held to the transaction.  The caller
341 	 * is responsible for unlocking any buffer passed back, either
342 	 * manually or by committing the transaction.  On error, the buffer is
343 	 * released and not passed back.
344 	 */
345 	xfs_trans_bhold(tp, bp);
346 	error = xfs_defer_finish(tpp);
347 	if (error) {
348 		xfs_trans_bhold_release(*tpp, bp);
349 		xfs_trans_brelse(*tpp, bp);
350 		return error;
351 	}
352 	*bpp = bp;
353 	return 0;
354 }
355 
356 /*
357  * Read in the in-core dquot's on-disk metadata and return the buffer.
358  * Returns ENOENT to signal a hole.
359  */
360 STATIC int
361 xfs_dquot_disk_read(
362 	struct xfs_mount	*mp,
363 	struct xfs_dquot	*dqp,
364 	struct xfs_buf		**bpp)
365 {
366 	struct xfs_bmbt_irec	map;
367 	struct xfs_buf		*bp;
368 	xfs_dqtype_t		qtype = xfs_dquot_type(dqp);
369 	struct xfs_inode	*quotip = xfs_quota_inode(mp, qtype);
370 	uint			lock_mode;
371 	int			nmaps = 1;
372 	int			error;
373 
374 	lock_mode = xfs_ilock_data_map_shared(quotip);
375 	if (!xfs_this_quota_on(mp, qtype)) {
376 		/*
377 		 * Return if this type of quotas is turned off while we
378 		 * didn't have the quota inode lock.
379 		 */
380 		xfs_iunlock(quotip, lock_mode);
381 		return -ESRCH;
382 	}
383 
384 	/*
385 	 * Find the block map; no allocations yet
386 	 */
387 	error = xfs_bmapi_read(quotip, dqp->q_fileoffset,
388 			XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0);
389 	xfs_iunlock(quotip, lock_mode);
390 	if (error)
391 		return error;
392 
393 	ASSERT(nmaps == 1);
394 	ASSERT(map.br_blockcount >= 1);
395 	ASSERT(map.br_startblock != DELAYSTARTBLOCK);
396 	if (map.br_startblock == HOLESTARTBLOCK)
397 		return -ENOENT;
398 
399 	trace_xfs_dqtobp_read(dqp);
400 
401 	/*
402 	 * store the blkno etc so that we don't have to do the
403 	 * mapping all the time
404 	 */
405 	dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
406 
407 	error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
408 			mp->m_quotainfo->qi_dqchunklen, 0, &bp,
409 			&xfs_dquot_buf_ops);
410 	if (error) {
411 		ASSERT(bp == NULL);
412 		return error;
413 	}
414 
415 	ASSERT(xfs_buf_islocked(bp));
416 	xfs_buf_set_ref(bp, XFS_DQUOT_REF);
417 	*bpp = bp;
418 
419 	return 0;
420 }
421 
422 /* Allocate and initialize everything we need for an incore dquot. */
423 STATIC struct xfs_dquot *
424 xfs_dquot_alloc(
425 	struct xfs_mount	*mp,
426 	xfs_dqid_t		id,
427 	xfs_dqtype_t		type)
428 {
429 	struct xfs_dquot	*dqp;
430 
431 	dqp = kmem_cache_zalloc(xfs_qm_dqzone, GFP_KERNEL | __GFP_NOFAIL);
432 
433 	dqp->q_type = type;
434 	dqp->q_id = id;
435 	dqp->q_mount = mp;
436 	INIT_LIST_HEAD(&dqp->q_lru);
437 	mutex_init(&dqp->q_qlock);
438 	init_waitqueue_head(&dqp->q_pinwait);
439 	dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk;
440 	/*
441 	 * Offset of dquot in the (fixed sized) dquot chunk.
442 	 */
443 	dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) *
444 			sizeof(xfs_dqblk_t);
445 
446 	/*
447 	 * Because we want to use a counting completion, complete
448 	 * the flush completion once to allow a single access to
449 	 * the flush completion without blocking.
450 	 */
451 	init_completion(&dqp->q_flush);
452 	complete(&dqp->q_flush);
453 
454 	/*
455 	 * Make sure group quotas have a different lock class than user
456 	 * quotas.
457 	 */
458 	switch (type) {
459 	case XFS_DQTYPE_USER:
460 		/* uses the default lock class */
461 		break;
462 	case XFS_DQTYPE_GROUP:
463 		lockdep_set_class(&dqp->q_qlock, &xfs_dquot_group_class);
464 		break;
465 	case XFS_DQTYPE_PROJ:
466 		lockdep_set_class(&dqp->q_qlock, &xfs_dquot_project_class);
467 		break;
468 	default:
469 		ASSERT(0);
470 		break;
471 	}
472 
473 	xfs_qm_dquot_logitem_init(dqp);
474 
475 	XFS_STATS_INC(mp, xs_qm_dquot);
476 	return dqp;
477 }
478 
479 /* Copy the in-core quota fields in from the on-disk buffer. */
480 STATIC int
481 xfs_dquot_from_disk(
482 	struct xfs_dquot	*dqp,
483 	struct xfs_buf		*bp)
484 {
485 	struct xfs_disk_dquot	*ddqp = bp->b_addr + dqp->q_bufoffset;
486 
487 	/*
488 	 * Ensure that we got the type and ID we were looking for.
489 	 * Everything else was checked by the dquot buffer verifier.
490 	 */
491 	if ((ddqp->d_type & XFS_DQTYPE_REC_MASK) != xfs_dquot_type(dqp) ||
492 	    be32_to_cpu(ddqp->d_id) != dqp->q_id) {
493 		xfs_alert_tag(bp->b_mount, XFS_PTAG_VERIFIER_ERROR,
494 			  "Metadata corruption detected at %pS, quota %u",
495 			  __this_address, dqp->q_id);
496 		xfs_alert(bp->b_mount, "Unmount and run xfs_repair");
497 		return -EFSCORRUPTED;
498 	}
499 
500 	/* copy everything from disk dquot to the incore dquot */
501 	dqp->q_type = ddqp->d_type;
502 	dqp->q_blk.hardlimit = be64_to_cpu(ddqp->d_blk_hardlimit);
503 	dqp->q_blk.softlimit = be64_to_cpu(ddqp->d_blk_softlimit);
504 	dqp->q_ino.hardlimit = be64_to_cpu(ddqp->d_ino_hardlimit);
505 	dqp->q_ino.softlimit = be64_to_cpu(ddqp->d_ino_softlimit);
506 	dqp->q_rtb.hardlimit = be64_to_cpu(ddqp->d_rtb_hardlimit);
507 	dqp->q_rtb.softlimit = be64_to_cpu(ddqp->d_rtb_softlimit);
508 
509 	dqp->q_blk.count = be64_to_cpu(ddqp->d_bcount);
510 	dqp->q_ino.count = be64_to_cpu(ddqp->d_icount);
511 	dqp->q_rtb.count = be64_to_cpu(ddqp->d_rtbcount);
512 
513 	dqp->q_blk.warnings = be16_to_cpu(ddqp->d_bwarns);
514 	dqp->q_ino.warnings = be16_to_cpu(ddqp->d_iwarns);
515 	dqp->q_rtb.warnings = be16_to_cpu(ddqp->d_rtbwarns);
516 
517 	dqp->q_blk.timer = be32_to_cpu(ddqp->d_btimer);
518 	dqp->q_ino.timer = be32_to_cpu(ddqp->d_itimer);
519 	dqp->q_rtb.timer = be32_to_cpu(ddqp->d_rtbtimer);
520 
521 	/*
522 	 * Reservation counters are defined as reservation plus current usage
523 	 * to avoid having to add every time.
524 	 */
525 	dqp->q_blk.reserved = dqp->q_blk.count;
526 	dqp->q_ino.reserved = dqp->q_ino.count;
527 	dqp->q_rtb.reserved = dqp->q_rtb.count;
528 
529 	/* initialize the dquot speculative prealloc thresholds */
530 	xfs_dquot_set_prealloc_limits(dqp);
531 	return 0;
532 }
533 
534 /* Copy the in-core quota fields into the on-disk buffer. */
535 void
536 xfs_dquot_to_disk(
537 	struct xfs_disk_dquot	*ddqp,
538 	struct xfs_dquot	*dqp)
539 {
540 	ddqp->d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
541 	ddqp->d_version = XFS_DQUOT_VERSION;
542 	ddqp->d_type = dqp->q_type;
543 	ddqp->d_id = cpu_to_be32(dqp->q_id);
544 	ddqp->d_pad0 = 0;
545 	ddqp->d_pad = 0;
546 
547 	ddqp->d_blk_hardlimit = cpu_to_be64(dqp->q_blk.hardlimit);
548 	ddqp->d_blk_softlimit = cpu_to_be64(dqp->q_blk.softlimit);
549 	ddqp->d_ino_hardlimit = cpu_to_be64(dqp->q_ino.hardlimit);
550 	ddqp->d_ino_softlimit = cpu_to_be64(dqp->q_ino.softlimit);
551 	ddqp->d_rtb_hardlimit = cpu_to_be64(dqp->q_rtb.hardlimit);
552 	ddqp->d_rtb_softlimit = cpu_to_be64(dqp->q_rtb.softlimit);
553 
554 	ddqp->d_bcount = cpu_to_be64(dqp->q_blk.count);
555 	ddqp->d_icount = cpu_to_be64(dqp->q_ino.count);
556 	ddqp->d_rtbcount = cpu_to_be64(dqp->q_rtb.count);
557 
558 	ddqp->d_bwarns = cpu_to_be16(dqp->q_blk.warnings);
559 	ddqp->d_iwarns = cpu_to_be16(dqp->q_ino.warnings);
560 	ddqp->d_rtbwarns = cpu_to_be16(dqp->q_rtb.warnings);
561 
562 	ddqp->d_btimer = cpu_to_be32(dqp->q_blk.timer);
563 	ddqp->d_itimer = cpu_to_be32(dqp->q_ino.timer);
564 	ddqp->d_rtbtimer = cpu_to_be32(dqp->q_rtb.timer);
565 }
566 
567 /* Allocate and initialize the dquot buffer for this in-core dquot. */
568 static int
569 xfs_qm_dqread_alloc(
570 	struct xfs_mount	*mp,
571 	struct xfs_dquot	*dqp,
572 	struct xfs_buf		**bpp)
573 {
574 	struct xfs_trans	*tp;
575 	int			error;
576 
577 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_qm_dqalloc,
578 			XFS_QM_DQALLOC_SPACE_RES(mp), 0, 0, &tp);
579 	if (error)
580 		goto err;
581 
582 	error = xfs_dquot_disk_alloc(&tp, dqp, bpp);
583 	if (error)
584 		goto err_cancel;
585 
586 	error = xfs_trans_commit(tp);
587 	if (error) {
588 		/*
589 		 * Buffer was held to the transaction, so we have to unlock it
590 		 * manually here because we're not passing it back.
591 		 */
592 		xfs_buf_relse(*bpp);
593 		*bpp = NULL;
594 		goto err;
595 	}
596 	return 0;
597 
598 err_cancel:
599 	xfs_trans_cancel(tp);
600 err:
601 	return error;
602 }
603 
604 /*
605  * Read in the ondisk dquot using dqtobp() then copy it to an incore version,
606  * and release the buffer immediately.  If @can_alloc is true, fill any
607  * holes in the on-disk metadata.
608  */
609 static int
610 xfs_qm_dqread(
611 	struct xfs_mount	*mp,
612 	xfs_dqid_t		id,
613 	xfs_dqtype_t		type,
614 	bool			can_alloc,
615 	struct xfs_dquot	**dqpp)
616 {
617 	struct xfs_dquot	*dqp;
618 	struct xfs_buf		*bp;
619 	int			error;
620 
621 	dqp = xfs_dquot_alloc(mp, id, type);
622 	trace_xfs_dqread(dqp);
623 
624 	/* Try to read the buffer, allocating if necessary. */
625 	error = xfs_dquot_disk_read(mp, dqp, &bp);
626 	if (error == -ENOENT && can_alloc)
627 		error = xfs_qm_dqread_alloc(mp, dqp, &bp);
628 	if (error)
629 		goto err;
630 
631 	/*
632 	 * At this point we should have a clean locked buffer.  Copy the data
633 	 * to the incore dquot and release the buffer since the incore dquot
634 	 * has its own locking protocol so we needn't tie up the buffer any
635 	 * further.
636 	 */
637 	ASSERT(xfs_buf_islocked(bp));
638 	error = xfs_dquot_from_disk(dqp, bp);
639 	xfs_buf_relse(bp);
640 	if (error)
641 		goto err;
642 
643 	*dqpp = dqp;
644 	return error;
645 
646 err:
647 	trace_xfs_dqread_fail(dqp);
648 	xfs_qm_dqdestroy(dqp);
649 	*dqpp = NULL;
650 	return error;
651 }
652 
653 /*
654  * Advance to the next id in the current chunk, or if at the
655  * end of the chunk, skip ahead to first id in next allocated chunk
656  * using the SEEK_DATA interface.
657  */
658 static int
659 xfs_dq_get_next_id(
660 	struct xfs_mount	*mp,
661 	xfs_dqtype_t		type,
662 	xfs_dqid_t		*id)
663 {
664 	struct xfs_inode	*quotip = xfs_quota_inode(mp, type);
665 	xfs_dqid_t		next_id = *id + 1; /* simple advance */
666 	uint			lock_flags;
667 	struct xfs_bmbt_irec	got;
668 	struct xfs_iext_cursor	cur;
669 	xfs_fsblock_t		start;
670 	int			error = 0;
671 
672 	/* If we'd wrap past the max ID, stop */
673 	if (next_id < *id)
674 		return -ENOENT;
675 
676 	/* If new ID is within the current chunk, advancing it sufficed */
677 	if (next_id % mp->m_quotainfo->qi_dqperchunk) {
678 		*id = next_id;
679 		return 0;
680 	}
681 
682 	/* Nope, next_id is now past the current chunk, so find the next one */
683 	start = (xfs_fsblock_t)next_id / mp->m_quotainfo->qi_dqperchunk;
684 
685 	lock_flags = xfs_ilock_data_map_shared(quotip);
686 	if (!(quotip->i_df.if_flags & XFS_IFEXTENTS)) {
687 		error = xfs_iread_extents(NULL, quotip, XFS_DATA_FORK);
688 		if (error)
689 			return error;
690 	}
691 
692 	if (xfs_iext_lookup_extent(quotip, &quotip->i_df, start, &cur, &got)) {
693 		/* contiguous chunk, bump startoff for the id calculation */
694 		if (got.br_startoff < start)
695 			got.br_startoff = start;
696 		*id = got.br_startoff * mp->m_quotainfo->qi_dqperchunk;
697 	} else {
698 		error = -ENOENT;
699 	}
700 
701 	xfs_iunlock(quotip, lock_flags);
702 
703 	return error;
704 }
705 
706 /*
707  * Look up the dquot in the in-core cache.  If found, the dquot is returned
708  * locked and ready to go.
709  */
710 static struct xfs_dquot *
711 xfs_qm_dqget_cache_lookup(
712 	struct xfs_mount	*mp,
713 	struct xfs_quotainfo	*qi,
714 	struct radix_tree_root	*tree,
715 	xfs_dqid_t		id)
716 {
717 	struct xfs_dquot	*dqp;
718 
719 restart:
720 	mutex_lock(&qi->qi_tree_lock);
721 	dqp = radix_tree_lookup(tree, id);
722 	if (!dqp) {
723 		mutex_unlock(&qi->qi_tree_lock);
724 		XFS_STATS_INC(mp, xs_qm_dqcachemisses);
725 		return NULL;
726 	}
727 
728 	xfs_dqlock(dqp);
729 	if (dqp->q_flags & XFS_DQFLAG_FREEING) {
730 		xfs_dqunlock(dqp);
731 		mutex_unlock(&qi->qi_tree_lock);
732 		trace_xfs_dqget_freeing(dqp);
733 		delay(1);
734 		goto restart;
735 	}
736 
737 	dqp->q_nrefs++;
738 	mutex_unlock(&qi->qi_tree_lock);
739 
740 	trace_xfs_dqget_hit(dqp);
741 	XFS_STATS_INC(mp, xs_qm_dqcachehits);
742 	return dqp;
743 }
744 
745 /*
746  * Try to insert a new dquot into the in-core cache.  If an error occurs the
747  * caller should throw away the dquot and start over.  Otherwise, the dquot
748  * is returned locked (and held by the cache) as if there had been a cache
749  * hit.
750  */
751 static int
752 xfs_qm_dqget_cache_insert(
753 	struct xfs_mount	*mp,
754 	struct xfs_quotainfo	*qi,
755 	struct radix_tree_root	*tree,
756 	xfs_dqid_t		id,
757 	struct xfs_dquot	*dqp)
758 {
759 	int			error;
760 
761 	mutex_lock(&qi->qi_tree_lock);
762 	error = radix_tree_insert(tree, id, dqp);
763 	if (unlikely(error)) {
764 		/* Duplicate found!  Caller must try again. */
765 		WARN_ON(error != -EEXIST);
766 		mutex_unlock(&qi->qi_tree_lock);
767 		trace_xfs_dqget_dup(dqp);
768 		return error;
769 	}
770 
771 	/* Return a locked dquot to the caller, with a reference taken. */
772 	xfs_dqlock(dqp);
773 	dqp->q_nrefs = 1;
774 
775 	qi->qi_dquots++;
776 	mutex_unlock(&qi->qi_tree_lock);
777 
778 	return 0;
779 }
780 
781 /* Check our input parameters. */
782 static int
783 xfs_qm_dqget_checks(
784 	struct xfs_mount	*mp,
785 	xfs_dqtype_t		type)
786 {
787 	if (WARN_ON_ONCE(!XFS_IS_QUOTA_RUNNING(mp)))
788 		return -ESRCH;
789 
790 	switch (type) {
791 	case XFS_DQTYPE_USER:
792 		if (!XFS_IS_UQUOTA_ON(mp))
793 			return -ESRCH;
794 		return 0;
795 	case XFS_DQTYPE_GROUP:
796 		if (!XFS_IS_GQUOTA_ON(mp))
797 			return -ESRCH;
798 		return 0;
799 	case XFS_DQTYPE_PROJ:
800 		if (!XFS_IS_PQUOTA_ON(mp))
801 			return -ESRCH;
802 		return 0;
803 	default:
804 		WARN_ON_ONCE(0);
805 		return -EINVAL;
806 	}
807 }
808 
809 /*
810  * Given the file system, id, and type (UDQUOT/GDQUOT), return a locked
811  * dquot, doing an allocation (if requested) as needed.
812  */
813 int
814 xfs_qm_dqget(
815 	struct xfs_mount	*mp,
816 	xfs_dqid_t		id,
817 	xfs_dqtype_t		type,
818 	bool			can_alloc,
819 	struct xfs_dquot	**O_dqpp)
820 {
821 	struct xfs_quotainfo	*qi = mp->m_quotainfo;
822 	struct radix_tree_root	*tree = xfs_dquot_tree(qi, type);
823 	struct xfs_dquot	*dqp;
824 	int			error;
825 
826 	error = xfs_qm_dqget_checks(mp, type);
827 	if (error)
828 		return error;
829 
830 restart:
831 	dqp = xfs_qm_dqget_cache_lookup(mp, qi, tree, id);
832 	if (dqp) {
833 		*O_dqpp = dqp;
834 		return 0;
835 	}
836 
837 	error = xfs_qm_dqread(mp, id, type, can_alloc, &dqp);
838 	if (error)
839 		return error;
840 
841 	error = xfs_qm_dqget_cache_insert(mp, qi, tree, id, dqp);
842 	if (error) {
843 		/*
844 		 * Duplicate found. Just throw away the new dquot and start
845 		 * over.
846 		 */
847 		xfs_qm_dqdestroy(dqp);
848 		XFS_STATS_INC(mp, xs_qm_dquot_dups);
849 		goto restart;
850 	}
851 
852 	trace_xfs_dqget_miss(dqp);
853 	*O_dqpp = dqp;
854 	return 0;
855 }
856 
857 /*
858  * Given a dquot id and type, read and initialize a dquot from the on-disk
859  * metadata.  This function is only for use during quota initialization so
860  * it ignores the dquot cache assuming that the dquot shrinker isn't set up.
861  * The caller is responsible for _qm_dqdestroy'ing the returned dquot.
862  */
863 int
864 xfs_qm_dqget_uncached(
865 	struct xfs_mount	*mp,
866 	xfs_dqid_t		id,
867 	xfs_dqtype_t		type,
868 	struct xfs_dquot	**dqpp)
869 {
870 	int			error;
871 
872 	error = xfs_qm_dqget_checks(mp, type);
873 	if (error)
874 		return error;
875 
876 	return xfs_qm_dqread(mp, id, type, 0, dqpp);
877 }
878 
879 /* Return the quota id for a given inode and type. */
880 xfs_dqid_t
881 xfs_qm_id_for_quotatype(
882 	struct xfs_inode	*ip,
883 	xfs_dqtype_t		type)
884 {
885 	switch (type) {
886 	case XFS_DQTYPE_USER:
887 		return i_uid_read(VFS_I(ip));
888 	case XFS_DQTYPE_GROUP:
889 		return i_gid_read(VFS_I(ip));
890 	case XFS_DQTYPE_PROJ:
891 		return ip->i_d.di_projid;
892 	}
893 	ASSERT(0);
894 	return 0;
895 }
896 
897 /*
898  * Return the dquot for a given inode and type.  If @can_alloc is true, then
899  * allocate blocks if needed.  The inode's ILOCK must be held and it must not
900  * have already had an inode attached.
901  */
902 int
903 xfs_qm_dqget_inode(
904 	struct xfs_inode	*ip,
905 	xfs_dqtype_t		type,
906 	bool			can_alloc,
907 	struct xfs_dquot	**O_dqpp)
908 {
909 	struct xfs_mount	*mp = ip->i_mount;
910 	struct xfs_quotainfo	*qi = mp->m_quotainfo;
911 	struct radix_tree_root	*tree = xfs_dquot_tree(qi, type);
912 	struct xfs_dquot	*dqp;
913 	xfs_dqid_t		id;
914 	int			error;
915 
916 	error = xfs_qm_dqget_checks(mp, type);
917 	if (error)
918 		return error;
919 
920 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
921 	ASSERT(xfs_inode_dquot(ip, type) == NULL);
922 
923 	id = xfs_qm_id_for_quotatype(ip, type);
924 
925 restart:
926 	dqp = xfs_qm_dqget_cache_lookup(mp, qi, tree, id);
927 	if (dqp) {
928 		*O_dqpp = dqp;
929 		return 0;
930 	}
931 
932 	/*
933 	 * Dquot cache miss. We don't want to keep the inode lock across
934 	 * a (potential) disk read. Also we don't want to deal with the lock
935 	 * ordering between quotainode and this inode. OTOH, dropping the inode
936 	 * lock here means dealing with a chown that can happen before
937 	 * we re-acquire the lock.
938 	 */
939 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
940 	error = xfs_qm_dqread(mp, id, type, can_alloc, &dqp);
941 	xfs_ilock(ip, XFS_ILOCK_EXCL);
942 	if (error)
943 		return error;
944 
945 	/*
946 	 * A dquot could be attached to this inode by now, since we had
947 	 * dropped the ilock.
948 	 */
949 	if (xfs_this_quota_on(mp, type)) {
950 		struct xfs_dquot	*dqp1;
951 
952 		dqp1 = xfs_inode_dquot(ip, type);
953 		if (dqp1) {
954 			xfs_qm_dqdestroy(dqp);
955 			dqp = dqp1;
956 			xfs_dqlock(dqp);
957 			goto dqret;
958 		}
959 	} else {
960 		/* inode stays locked on return */
961 		xfs_qm_dqdestroy(dqp);
962 		return -ESRCH;
963 	}
964 
965 	error = xfs_qm_dqget_cache_insert(mp, qi, tree, id, dqp);
966 	if (error) {
967 		/*
968 		 * Duplicate found. Just throw away the new dquot and start
969 		 * over.
970 		 */
971 		xfs_qm_dqdestroy(dqp);
972 		XFS_STATS_INC(mp, xs_qm_dquot_dups);
973 		goto restart;
974 	}
975 
976 dqret:
977 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
978 	trace_xfs_dqget_miss(dqp);
979 	*O_dqpp = dqp;
980 	return 0;
981 }
982 
983 /*
984  * Starting at @id and progressing upwards, look for an initialized incore
985  * dquot, lock it, and return it.
986  */
987 int
988 xfs_qm_dqget_next(
989 	struct xfs_mount	*mp,
990 	xfs_dqid_t		id,
991 	xfs_dqtype_t		type,
992 	struct xfs_dquot	**dqpp)
993 {
994 	struct xfs_dquot	*dqp;
995 	int			error = 0;
996 
997 	*dqpp = NULL;
998 	for (; !error; error = xfs_dq_get_next_id(mp, type, &id)) {
999 		error = xfs_qm_dqget(mp, id, type, false, &dqp);
1000 		if (error == -ENOENT)
1001 			continue;
1002 		else if (error != 0)
1003 			break;
1004 
1005 		if (!XFS_IS_DQUOT_UNINITIALIZED(dqp)) {
1006 			*dqpp = dqp;
1007 			return 0;
1008 		}
1009 
1010 		xfs_qm_dqput(dqp);
1011 	}
1012 
1013 	return error;
1014 }
1015 
1016 /*
1017  * Release a reference to the dquot (decrement ref-count) and unlock it.
1018  *
1019  * If there is a group quota attached to this dquot, carefully release that
1020  * too without tripping over deadlocks'n'stuff.
1021  */
1022 void
1023 xfs_qm_dqput(
1024 	struct xfs_dquot	*dqp)
1025 {
1026 	ASSERT(dqp->q_nrefs > 0);
1027 	ASSERT(XFS_DQ_IS_LOCKED(dqp));
1028 
1029 	trace_xfs_dqput(dqp);
1030 
1031 	if (--dqp->q_nrefs == 0) {
1032 		struct xfs_quotainfo	*qi = dqp->q_mount->m_quotainfo;
1033 		trace_xfs_dqput_free(dqp);
1034 
1035 		if (list_lru_add(&qi->qi_lru, &dqp->q_lru))
1036 			XFS_STATS_INC(dqp->q_mount, xs_qm_dquot_unused);
1037 	}
1038 	xfs_dqunlock(dqp);
1039 }
1040 
1041 /*
1042  * Release a dquot. Flush it if dirty, then dqput() it.
1043  * dquot must not be locked.
1044  */
1045 void
1046 xfs_qm_dqrele(
1047 	struct xfs_dquot	*dqp)
1048 {
1049 	if (!dqp)
1050 		return;
1051 
1052 	trace_xfs_dqrele(dqp);
1053 
1054 	xfs_dqlock(dqp);
1055 	/*
1056 	 * We don't care to flush it if the dquot is dirty here.
1057 	 * That will create stutters that we want to avoid.
1058 	 * Instead we do a delayed write when we try to reclaim
1059 	 * a dirty dquot. Also xfs_sync will take part of the burden...
1060 	 */
1061 	xfs_qm_dqput(dqp);
1062 }
1063 
1064 /*
1065  * This is the dquot flushing I/O completion routine.  It is called
1066  * from interrupt level when the buffer containing the dquot is
1067  * flushed to disk.  It is responsible for removing the dquot logitem
1068  * from the AIL if it has not been re-logged, and unlocking the dquot's
1069  * flush lock. This behavior is very similar to that of inodes..
1070  */
1071 static void
1072 xfs_qm_dqflush_done(
1073 	struct xfs_log_item	*lip)
1074 {
1075 	struct xfs_dq_logitem	*qip = (struct xfs_dq_logitem *)lip;
1076 	struct xfs_dquot	*dqp = qip->qli_dquot;
1077 	struct xfs_ail		*ailp = lip->li_ailp;
1078 	xfs_lsn_t		tail_lsn;
1079 
1080 	/*
1081 	 * We only want to pull the item from the AIL if its
1082 	 * location in the log has not changed since we started the flush.
1083 	 * Thus, we only bother if the dquot's lsn has
1084 	 * not changed. First we check the lsn outside the lock
1085 	 * since it's cheaper, and then we recheck while
1086 	 * holding the lock before removing the dquot from the AIL.
1087 	 */
1088 	if (test_bit(XFS_LI_IN_AIL, &lip->li_flags) &&
1089 	    ((lip->li_lsn == qip->qli_flush_lsn) ||
1090 	     test_bit(XFS_LI_FAILED, &lip->li_flags))) {
1091 
1092 		spin_lock(&ailp->ail_lock);
1093 		xfs_clear_li_failed(lip);
1094 		if (lip->li_lsn == qip->qli_flush_lsn) {
1095 			/* xfs_ail_update_finish() drops the AIL lock */
1096 			tail_lsn = xfs_ail_delete_one(ailp, lip);
1097 			xfs_ail_update_finish(ailp, tail_lsn);
1098 		} else {
1099 			spin_unlock(&ailp->ail_lock);
1100 		}
1101 	}
1102 
1103 	/*
1104 	 * Release the dq's flush lock since we're done with it.
1105 	 */
1106 	xfs_dqfunlock(dqp);
1107 }
1108 
1109 void
1110 xfs_dquot_done(
1111 	struct xfs_buf		*bp)
1112 {
1113 	struct xfs_log_item	*lip, *n;
1114 
1115 	list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) {
1116 		list_del_init(&lip->li_bio_list);
1117 		xfs_qm_dqflush_done(lip);
1118 	}
1119 }
1120 
1121 /* Check incore dquot for errors before we flush. */
1122 static xfs_failaddr_t
1123 xfs_qm_dqflush_check(
1124 	struct xfs_dquot	*dqp)
1125 {
1126 	xfs_dqtype_t		type = xfs_dquot_type(dqp);
1127 
1128 	if (type != XFS_DQTYPE_USER &&
1129 	    type != XFS_DQTYPE_GROUP &&
1130 	    type != XFS_DQTYPE_PROJ)
1131 		return __this_address;
1132 
1133 	if (dqp->q_id == 0)
1134 		return NULL;
1135 
1136 	if (dqp->q_blk.softlimit && dqp->q_blk.count > dqp->q_blk.softlimit &&
1137 	    !dqp->q_blk.timer)
1138 		return __this_address;
1139 
1140 	if (dqp->q_ino.softlimit && dqp->q_ino.count > dqp->q_ino.softlimit &&
1141 	    !dqp->q_ino.timer)
1142 		return __this_address;
1143 
1144 	if (dqp->q_rtb.softlimit && dqp->q_rtb.count > dqp->q_rtb.softlimit &&
1145 	    !dqp->q_rtb.timer)
1146 		return __this_address;
1147 
1148 	return NULL;
1149 }
1150 
1151 /*
1152  * Write a modified dquot to disk.
1153  * The dquot must be locked and the flush lock too taken by caller.
1154  * The flush lock will not be unlocked until the dquot reaches the disk,
1155  * but the dquot is free to be unlocked and modified by the caller
1156  * in the interim. Dquot is still locked on return. This behavior is
1157  * identical to that of inodes.
1158  */
1159 int
1160 xfs_qm_dqflush(
1161 	struct xfs_dquot	*dqp,
1162 	struct xfs_buf		**bpp)
1163 {
1164 	struct xfs_mount	*mp = dqp->q_mount;
1165 	struct xfs_log_item	*lip = &dqp->q_logitem.qli_item;
1166 	struct xfs_buf		*bp;
1167 	struct xfs_dqblk	*dqblk;
1168 	xfs_failaddr_t		fa;
1169 	int			error;
1170 
1171 	ASSERT(XFS_DQ_IS_LOCKED(dqp));
1172 	ASSERT(!completion_done(&dqp->q_flush));
1173 
1174 	trace_xfs_dqflush(dqp);
1175 
1176 	*bpp = NULL;
1177 
1178 	xfs_qm_dqunpin_wait(dqp);
1179 
1180 	/*
1181 	 * Get the buffer containing the on-disk dquot
1182 	 */
1183 	error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
1184 				   mp->m_quotainfo->qi_dqchunklen, XBF_TRYLOCK,
1185 				   &bp, &xfs_dquot_buf_ops);
1186 	if (error == -EAGAIN)
1187 		goto out_unlock;
1188 	if (error)
1189 		goto out_abort;
1190 
1191 	fa = xfs_qm_dqflush_check(dqp);
1192 	if (fa) {
1193 		xfs_alert(mp, "corrupt dquot ID 0x%x in memory at %pS",
1194 				dqp->q_id, fa);
1195 		xfs_buf_relse(bp);
1196 		error = -EFSCORRUPTED;
1197 		goto out_abort;
1198 	}
1199 
1200 	/* Flush the incore dquot to the ondisk buffer. */
1201 	dqblk = bp->b_addr + dqp->q_bufoffset;
1202 	xfs_dquot_to_disk(&dqblk->dd_diskdq, dqp);
1203 
1204 	/*
1205 	 * Clear the dirty field and remember the flush lsn for later use.
1206 	 */
1207 	dqp->q_flags &= ~XFS_DQFLAG_DIRTY;
1208 
1209 	xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn,
1210 					&dqp->q_logitem.qli_item.li_lsn);
1211 
1212 	/*
1213 	 * copy the lsn into the on-disk dquot now while we have the in memory
1214 	 * dquot here. This can't be done later in the write verifier as we
1215 	 * can't get access to the log item at that point in time.
1216 	 *
1217 	 * We also calculate the CRC here so that the on-disk dquot in the
1218 	 * buffer always has a valid CRC. This ensures there is no possibility
1219 	 * of a dquot without an up-to-date CRC getting to disk.
1220 	 */
1221 	if (xfs_sb_version_hascrc(&mp->m_sb)) {
1222 		dqblk->dd_lsn = cpu_to_be64(dqp->q_logitem.qli_item.li_lsn);
1223 		xfs_update_cksum((char *)dqblk, sizeof(struct xfs_dqblk),
1224 				 XFS_DQUOT_CRC_OFF);
1225 	}
1226 
1227 	/*
1228 	 * Attach the dquot to the buffer so that we can remove this dquot from
1229 	 * the AIL and release the flush lock once the dquot is synced to disk.
1230 	 */
1231 	bp->b_flags |= _XBF_DQUOTS;
1232 	list_add_tail(&dqp->q_logitem.qli_item.li_bio_list, &bp->b_li_list);
1233 
1234 	/*
1235 	 * If the buffer is pinned then push on the log so we won't
1236 	 * get stuck waiting in the write for too long.
1237 	 */
1238 	if (xfs_buf_ispinned(bp)) {
1239 		trace_xfs_dqflush_force(dqp);
1240 		xfs_log_force(mp, 0);
1241 	}
1242 
1243 	trace_xfs_dqflush_done(dqp);
1244 	*bpp = bp;
1245 	return 0;
1246 
1247 out_abort:
1248 	dqp->q_flags &= ~XFS_DQFLAG_DIRTY;
1249 	xfs_trans_ail_delete(lip, 0);
1250 	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1251 out_unlock:
1252 	xfs_dqfunlock(dqp);
1253 	return error;
1254 }
1255 
1256 /*
1257  * Lock two xfs_dquot structures.
1258  *
1259  * To avoid deadlocks we always lock the quota structure with
1260  * the lowerd id first.
1261  */
1262 void
1263 xfs_dqlock2(
1264 	struct xfs_dquot	*d1,
1265 	struct xfs_dquot	*d2)
1266 {
1267 	if (d1 && d2) {
1268 		ASSERT(d1 != d2);
1269 		if (d1->q_id > d2->q_id) {
1270 			mutex_lock(&d2->q_qlock);
1271 			mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED);
1272 		} else {
1273 			mutex_lock(&d1->q_qlock);
1274 			mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED);
1275 		}
1276 	} else if (d1) {
1277 		mutex_lock(&d1->q_qlock);
1278 	} else if (d2) {
1279 		mutex_lock(&d2->q_qlock);
1280 	}
1281 }
1282 
1283 int __init
1284 xfs_qm_init(void)
1285 {
1286 	xfs_qm_dqzone = kmem_cache_create("xfs_dquot",
1287 					  sizeof(struct xfs_dquot),
1288 					  0, 0, NULL);
1289 	if (!xfs_qm_dqzone)
1290 		goto out;
1291 
1292 	xfs_qm_dqtrxzone = kmem_cache_create("xfs_dqtrx",
1293 					     sizeof(struct xfs_dquot_acct),
1294 					     0, 0, NULL);
1295 	if (!xfs_qm_dqtrxzone)
1296 		goto out_free_dqzone;
1297 
1298 	return 0;
1299 
1300 out_free_dqzone:
1301 	kmem_cache_destroy(xfs_qm_dqzone);
1302 out:
1303 	return -ENOMEM;
1304 }
1305 
1306 void
1307 xfs_qm_exit(void)
1308 {
1309 	kmem_cache_destroy(xfs_qm_dqtrxzone);
1310 	kmem_cache_destroy(xfs_qm_dqzone);
1311 }
1312 
1313 /*
1314  * Iterate every dquot of a particular type.  The caller must ensure that the
1315  * particular quota type is active.  iter_fn can return negative error codes,
1316  * or -ECANCELED to indicate that it wants to stop iterating.
1317  */
1318 int
1319 xfs_qm_dqiterate(
1320 	struct xfs_mount	*mp,
1321 	xfs_dqtype_t		type,
1322 	xfs_qm_dqiterate_fn	iter_fn,
1323 	void			*priv)
1324 {
1325 	struct xfs_dquot	*dq;
1326 	xfs_dqid_t		id = 0;
1327 	int			error;
1328 
1329 	do {
1330 		error = xfs_qm_dqget_next(mp, id, type, &dq);
1331 		if (error == -ENOENT)
1332 			return 0;
1333 		if (error)
1334 			return error;
1335 
1336 		error = iter_fn(dq, type, priv);
1337 		id = dq->q_id;
1338 		xfs_qm_dqput(dq);
1339 	} while (error == 0 && id != 0);
1340 
1341 	return error;
1342 }
1343