xref: /linux/fs/xfs/xfs_dquot.c (revision 1c4b5ecb7ea190fa3e9f9d6891e6c90b60e04f24)
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_dquot	*dqp,
293 	struct xfs_buf		**bpp)
294 {
295 	struct xfs_bmbt_irec	map;
296 	struct xfs_trans	*tp;
297 	struct xfs_mount	*mp = dqp->q_mount;
298 	struct xfs_buf		*bp;
299 	xfs_dqtype_t		qtype = xfs_dquot_type(dqp);
300 	struct xfs_inode	*quotip = xfs_quota_inode(mp, qtype);
301 	int			nmaps = 1;
302 	int			error;
303 
304 	trace_xfs_dqalloc(dqp);
305 
306 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_qm_dqalloc,
307 			XFS_QM_DQALLOC_SPACE_RES(mp), 0, 0, &tp);
308 	if (error)
309 		return error;
310 
311 	xfs_ilock(quotip, XFS_ILOCK_EXCL);
312 	xfs_trans_ijoin(tp, quotip, 0);
313 
314 	if (!xfs_this_quota_on(dqp->q_mount, qtype)) {
315 		/*
316 		 * Return if this type of quotas is turned off while we didn't
317 		 * have an inode lock
318 		 */
319 		error = -ESRCH;
320 		goto err_cancel;
321 	}
322 
323 	error = xfs_iext_count_may_overflow(quotip, XFS_DATA_FORK,
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 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.warnings = be16_to_cpu(ddqp->d_bwarns);
593 	dqp->q_ino.warnings = be16_to_cpu(ddqp->d_iwarns);
594 	dqp->q_rtb.warnings = be16_to_cpu(ddqp->d_rtbwarns);
595 
596 	dqp->q_blk.timer = xfs_dquot_from_disk_ts(ddqp, ddqp->d_btimer);
597 	dqp->q_ino.timer = xfs_dquot_from_disk_ts(ddqp, ddqp->d_itimer);
598 	dqp->q_rtb.timer = xfs_dquot_from_disk_ts(ddqp, ddqp->d_rtbtimer);
599 
600 	/*
601 	 * Reservation counters are defined as reservation plus current usage
602 	 * to avoid having to add every time.
603 	 */
604 	dqp->q_blk.reserved = dqp->q_blk.count;
605 	dqp->q_ino.reserved = dqp->q_ino.count;
606 	dqp->q_rtb.reserved = dqp->q_rtb.count;
607 
608 	/* initialize the dquot speculative prealloc thresholds */
609 	xfs_dquot_set_prealloc_limits(dqp);
610 	return 0;
611 }
612 
613 /* Copy the in-core quota fields into the on-disk buffer. */
614 void
615 xfs_dquot_to_disk(
616 	struct xfs_disk_dquot	*ddqp,
617 	struct xfs_dquot	*dqp)
618 {
619 	ddqp->d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
620 	ddqp->d_version = XFS_DQUOT_VERSION;
621 	ddqp->d_type = dqp->q_type;
622 	ddqp->d_id = cpu_to_be32(dqp->q_id);
623 	ddqp->d_pad0 = 0;
624 	ddqp->d_pad = 0;
625 
626 	ddqp->d_blk_hardlimit = cpu_to_be64(dqp->q_blk.hardlimit);
627 	ddqp->d_blk_softlimit = cpu_to_be64(dqp->q_blk.softlimit);
628 	ddqp->d_ino_hardlimit = cpu_to_be64(dqp->q_ino.hardlimit);
629 	ddqp->d_ino_softlimit = cpu_to_be64(dqp->q_ino.softlimit);
630 	ddqp->d_rtb_hardlimit = cpu_to_be64(dqp->q_rtb.hardlimit);
631 	ddqp->d_rtb_softlimit = cpu_to_be64(dqp->q_rtb.softlimit);
632 
633 	ddqp->d_bcount = cpu_to_be64(dqp->q_blk.count);
634 	ddqp->d_icount = cpu_to_be64(dqp->q_ino.count);
635 	ddqp->d_rtbcount = cpu_to_be64(dqp->q_rtb.count);
636 
637 	ddqp->d_bwarns = cpu_to_be16(dqp->q_blk.warnings);
638 	ddqp->d_iwarns = cpu_to_be16(dqp->q_ino.warnings);
639 	ddqp->d_rtbwarns = cpu_to_be16(dqp->q_rtb.warnings);
640 
641 	ddqp->d_btimer = xfs_dquot_to_disk_ts(dqp, dqp->q_blk.timer);
642 	ddqp->d_itimer = xfs_dquot_to_disk_ts(dqp, dqp->q_ino.timer);
643 	ddqp->d_rtbtimer = xfs_dquot_to_disk_ts(dqp, dqp->q_rtb.timer);
644 }
645 
646 /*
647  * Read in the ondisk dquot using dqtobp() then copy it to an incore version,
648  * and release the buffer immediately.  If @can_alloc is true, fill any
649  * holes in the on-disk metadata.
650  */
651 static int
652 xfs_qm_dqread(
653 	struct xfs_mount	*mp,
654 	xfs_dqid_t		id,
655 	xfs_dqtype_t		type,
656 	bool			can_alloc,
657 	struct xfs_dquot	**dqpp)
658 {
659 	struct xfs_dquot	*dqp;
660 	struct xfs_buf		*bp;
661 	int			error;
662 
663 	dqp = xfs_dquot_alloc(mp, id, type);
664 	trace_xfs_dqread(dqp);
665 
666 	/* Try to read the buffer, allocating if necessary. */
667 	error = xfs_dquot_disk_read(mp, dqp, &bp);
668 	if (error == -ENOENT && can_alloc)
669 		error = xfs_dquot_disk_alloc(dqp, &bp);
670 	if (error)
671 		goto err;
672 
673 	/*
674 	 * At this point we should have a clean locked buffer.  Copy the data
675 	 * to the incore dquot and release the buffer since the incore dquot
676 	 * has its own locking protocol so we needn't tie up the buffer any
677 	 * further.
678 	 */
679 	ASSERT(xfs_buf_islocked(bp));
680 	error = xfs_dquot_from_disk(dqp, bp);
681 	xfs_buf_relse(bp);
682 	if (error)
683 		goto err;
684 
685 	*dqpp = dqp;
686 	return error;
687 
688 err:
689 	trace_xfs_dqread_fail(dqp);
690 	xfs_qm_dqdestroy(dqp);
691 	*dqpp = NULL;
692 	return error;
693 }
694 
695 /*
696  * Advance to the next id in the current chunk, or if at the
697  * end of the chunk, skip ahead to first id in next allocated chunk
698  * using the SEEK_DATA interface.
699  */
700 static int
701 xfs_dq_get_next_id(
702 	struct xfs_mount	*mp,
703 	xfs_dqtype_t		type,
704 	xfs_dqid_t		*id)
705 {
706 	struct xfs_inode	*quotip = xfs_quota_inode(mp, type);
707 	xfs_dqid_t		next_id = *id + 1; /* simple advance */
708 	uint			lock_flags;
709 	struct xfs_bmbt_irec	got;
710 	struct xfs_iext_cursor	cur;
711 	xfs_fsblock_t		start;
712 	int			error = 0;
713 
714 	/* If we'd wrap past the max ID, stop */
715 	if (next_id < *id)
716 		return -ENOENT;
717 
718 	/* If new ID is within the current chunk, advancing it sufficed */
719 	if (next_id % mp->m_quotainfo->qi_dqperchunk) {
720 		*id = next_id;
721 		return 0;
722 	}
723 
724 	/* Nope, next_id is now past the current chunk, so find the next one */
725 	start = (xfs_fsblock_t)next_id / mp->m_quotainfo->qi_dqperchunk;
726 
727 	lock_flags = xfs_ilock_data_map_shared(quotip);
728 	error = xfs_iread_extents(NULL, quotip, XFS_DATA_FORK);
729 	if (error)
730 		return error;
731 
732 	if (xfs_iext_lookup_extent(quotip, &quotip->i_df, start, &cur, &got)) {
733 		/* contiguous chunk, bump startoff for the id calculation */
734 		if (got.br_startoff < start)
735 			got.br_startoff = start;
736 		*id = got.br_startoff * mp->m_quotainfo->qi_dqperchunk;
737 	} else {
738 		error = -ENOENT;
739 	}
740 
741 	xfs_iunlock(quotip, lock_flags);
742 
743 	return error;
744 }
745 
746 /*
747  * Look up the dquot in the in-core cache.  If found, the dquot is returned
748  * locked and ready to go.
749  */
750 static struct xfs_dquot *
751 xfs_qm_dqget_cache_lookup(
752 	struct xfs_mount	*mp,
753 	struct xfs_quotainfo	*qi,
754 	struct radix_tree_root	*tree,
755 	xfs_dqid_t		id)
756 {
757 	struct xfs_dquot	*dqp;
758 
759 restart:
760 	mutex_lock(&qi->qi_tree_lock);
761 	dqp = radix_tree_lookup(tree, id);
762 	if (!dqp) {
763 		mutex_unlock(&qi->qi_tree_lock);
764 		XFS_STATS_INC(mp, xs_qm_dqcachemisses);
765 		return NULL;
766 	}
767 
768 	xfs_dqlock(dqp);
769 	if (dqp->q_flags & XFS_DQFLAG_FREEING) {
770 		xfs_dqunlock(dqp);
771 		mutex_unlock(&qi->qi_tree_lock);
772 		trace_xfs_dqget_freeing(dqp);
773 		delay(1);
774 		goto restart;
775 	}
776 
777 	dqp->q_nrefs++;
778 	mutex_unlock(&qi->qi_tree_lock);
779 
780 	trace_xfs_dqget_hit(dqp);
781 	XFS_STATS_INC(mp, xs_qm_dqcachehits);
782 	return dqp;
783 }
784 
785 /*
786  * Try to insert a new dquot into the in-core cache.  If an error occurs the
787  * caller should throw away the dquot and start over.  Otherwise, the dquot
788  * is returned locked (and held by the cache) as if there had been a cache
789  * hit.
790  */
791 static int
792 xfs_qm_dqget_cache_insert(
793 	struct xfs_mount	*mp,
794 	struct xfs_quotainfo	*qi,
795 	struct radix_tree_root	*tree,
796 	xfs_dqid_t		id,
797 	struct xfs_dquot	*dqp)
798 {
799 	int			error;
800 
801 	mutex_lock(&qi->qi_tree_lock);
802 	error = radix_tree_insert(tree, id, dqp);
803 	if (unlikely(error)) {
804 		/* Duplicate found!  Caller must try again. */
805 		WARN_ON(error != -EEXIST);
806 		mutex_unlock(&qi->qi_tree_lock);
807 		trace_xfs_dqget_dup(dqp);
808 		return error;
809 	}
810 
811 	/* Return a locked dquot to the caller, with a reference taken. */
812 	xfs_dqlock(dqp);
813 	dqp->q_nrefs = 1;
814 
815 	qi->qi_dquots++;
816 	mutex_unlock(&qi->qi_tree_lock);
817 
818 	return 0;
819 }
820 
821 /* Check our input parameters. */
822 static int
823 xfs_qm_dqget_checks(
824 	struct xfs_mount	*mp,
825 	xfs_dqtype_t		type)
826 {
827 	switch (type) {
828 	case XFS_DQTYPE_USER:
829 		if (!XFS_IS_UQUOTA_ON(mp))
830 			return -ESRCH;
831 		return 0;
832 	case XFS_DQTYPE_GROUP:
833 		if (!XFS_IS_GQUOTA_ON(mp))
834 			return -ESRCH;
835 		return 0;
836 	case XFS_DQTYPE_PROJ:
837 		if (!XFS_IS_PQUOTA_ON(mp))
838 			return -ESRCH;
839 		return 0;
840 	default:
841 		WARN_ON_ONCE(0);
842 		return -EINVAL;
843 	}
844 }
845 
846 /*
847  * Given the file system, id, and type (UDQUOT/GDQUOT/PDQUOT), return a
848  * locked dquot, doing an allocation (if requested) as needed.
849  */
850 int
851 xfs_qm_dqget(
852 	struct xfs_mount	*mp,
853 	xfs_dqid_t		id,
854 	xfs_dqtype_t		type,
855 	bool			can_alloc,
856 	struct xfs_dquot	**O_dqpp)
857 {
858 	struct xfs_quotainfo	*qi = mp->m_quotainfo;
859 	struct radix_tree_root	*tree = xfs_dquot_tree(qi, type);
860 	struct xfs_dquot	*dqp;
861 	int			error;
862 
863 	error = xfs_qm_dqget_checks(mp, type);
864 	if (error)
865 		return error;
866 
867 restart:
868 	dqp = xfs_qm_dqget_cache_lookup(mp, qi, tree, id);
869 	if (dqp) {
870 		*O_dqpp = dqp;
871 		return 0;
872 	}
873 
874 	error = xfs_qm_dqread(mp, id, type, can_alloc, &dqp);
875 	if (error)
876 		return error;
877 
878 	error = xfs_qm_dqget_cache_insert(mp, qi, tree, id, dqp);
879 	if (error) {
880 		/*
881 		 * Duplicate found. Just throw away the new dquot and start
882 		 * over.
883 		 */
884 		xfs_qm_dqdestroy(dqp);
885 		XFS_STATS_INC(mp, xs_qm_dquot_dups);
886 		goto restart;
887 	}
888 
889 	trace_xfs_dqget_miss(dqp);
890 	*O_dqpp = dqp;
891 	return 0;
892 }
893 
894 /*
895  * Given a dquot id and type, read and initialize a dquot from the on-disk
896  * metadata.  This function is only for use during quota initialization so
897  * it ignores the dquot cache assuming that the dquot shrinker isn't set up.
898  * The caller is responsible for _qm_dqdestroy'ing the returned dquot.
899  */
900 int
901 xfs_qm_dqget_uncached(
902 	struct xfs_mount	*mp,
903 	xfs_dqid_t		id,
904 	xfs_dqtype_t		type,
905 	struct xfs_dquot	**dqpp)
906 {
907 	int			error;
908 
909 	error = xfs_qm_dqget_checks(mp, type);
910 	if (error)
911 		return error;
912 
913 	return xfs_qm_dqread(mp, id, type, 0, dqpp);
914 }
915 
916 /* Return the quota id for a given inode and type. */
917 xfs_dqid_t
918 xfs_qm_id_for_quotatype(
919 	struct xfs_inode	*ip,
920 	xfs_dqtype_t		type)
921 {
922 	switch (type) {
923 	case XFS_DQTYPE_USER:
924 		return i_uid_read(VFS_I(ip));
925 	case XFS_DQTYPE_GROUP:
926 		return i_gid_read(VFS_I(ip));
927 	case XFS_DQTYPE_PROJ:
928 		return ip->i_projid;
929 	}
930 	ASSERT(0);
931 	return 0;
932 }
933 
934 /*
935  * Return the dquot for a given inode and type.  If @can_alloc is true, then
936  * allocate blocks if needed.  The inode's ILOCK must be held and it must not
937  * have already had an inode attached.
938  */
939 int
940 xfs_qm_dqget_inode(
941 	struct xfs_inode	*ip,
942 	xfs_dqtype_t		type,
943 	bool			can_alloc,
944 	struct xfs_dquot	**O_dqpp)
945 {
946 	struct xfs_mount	*mp = ip->i_mount;
947 	struct xfs_quotainfo	*qi = mp->m_quotainfo;
948 	struct radix_tree_root	*tree = xfs_dquot_tree(qi, type);
949 	struct xfs_dquot	*dqp;
950 	xfs_dqid_t		id;
951 	int			error;
952 
953 	error = xfs_qm_dqget_checks(mp, type);
954 	if (error)
955 		return error;
956 
957 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
958 	ASSERT(xfs_inode_dquot(ip, type) == NULL);
959 
960 	id = xfs_qm_id_for_quotatype(ip, type);
961 
962 restart:
963 	dqp = xfs_qm_dqget_cache_lookup(mp, qi, tree, id);
964 	if (dqp) {
965 		*O_dqpp = dqp;
966 		return 0;
967 	}
968 
969 	/*
970 	 * Dquot cache miss. We don't want to keep the inode lock across
971 	 * a (potential) disk read. Also we don't want to deal with the lock
972 	 * ordering between quotainode and this inode. OTOH, dropping the inode
973 	 * lock here means dealing with a chown that can happen before
974 	 * we re-acquire the lock.
975 	 */
976 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
977 	error = xfs_qm_dqread(mp, id, type, can_alloc, &dqp);
978 	xfs_ilock(ip, XFS_ILOCK_EXCL);
979 	if (error)
980 		return error;
981 
982 	/*
983 	 * A dquot could be attached to this inode by now, since we had
984 	 * dropped the ilock.
985 	 */
986 	if (xfs_this_quota_on(mp, type)) {
987 		struct xfs_dquot	*dqp1;
988 
989 		dqp1 = xfs_inode_dquot(ip, type);
990 		if (dqp1) {
991 			xfs_qm_dqdestroy(dqp);
992 			dqp = dqp1;
993 			xfs_dqlock(dqp);
994 			goto dqret;
995 		}
996 	} else {
997 		/* inode stays locked on return */
998 		xfs_qm_dqdestroy(dqp);
999 		return -ESRCH;
1000 	}
1001 
1002 	error = xfs_qm_dqget_cache_insert(mp, qi, tree, id, dqp);
1003 	if (error) {
1004 		/*
1005 		 * Duplicate found. Just throw away the new dquot and start
1006 		 * over.
1007 		 */
1008 		xfs_qm_dqdestroy(dqp);
1009 		XFS_STATS_INC(mp, xs_qm_dquot_dups);
1010 		goto restart;
1011 	}
1012 
1013 dqret:
1014 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1015 	trace_xfs_dqget_miss(dqp);
1016 	*O_dqpp = dqp;
1017 	return 0;
1018 }
1019 
1020 /*
1021  * Starting at @id and progressing upwards, look for an initialized incore
1022  * dquot, lock it, and return it.
1023  */
1024 int
1025 xfs_qm_dqget_next(
1026 	struct xfs_mount	*mp,
1027 	xfs_dqid_t		id,
1028 	xfs_dqtype_t		type,
1029 	struct xfs_dquot	**dqpp)
1030 {
1031 	struct xfs_dquot	*dqp;
1032 	int			error = 0;
1033 
1034 	*dqpp = NULL;
1035 	for (; !error; error = xfs_dq_get_next_id(mp, type, &id)) {
1036 		error = xfs_qm_dqget(mp, id, type, false, &dqp);
1037 		if (error == -ENOENT)
1038 			continue;
1039 		else if (error != 0)
1040 			break;
1041 
1042 		if (!XFS_IS_DQUOT_UNINITIALIZED(dqp)) {
1043 			*dqpp = dqp;
1044 			return 0;
1045 		}
1046 
1047 		xfs_qm_dqput(dqp);
1048 	}
1049 
1050 	return error;
1051 }
1052 
1053 /*
1054  * Release a reference to the dquot (decrement ref-count) and unlock it.
1055  *
1056  * If there is a group quota attached to this dquot, carefully release that
1057  * too without tripping over deadlocks'n'stuff.
1058  */
1059 void
1060 xfs_qm_dqput(
1061 	struct xfs_dquot	*dqp)
1062 {
1063 	ASSERT(dqp->q_nrefs > 0);
1064 	ASSERT(XFS_DQ_IS_LOCKED(dqp));
1065 
1066 	trace_xfs_dqput(dqp);
1067 
1068 	if (--dqp->q_nrefs == 0) {
1069 		struct xfs_quotainfo	*qi = dqp->q_mount->m_quotainfo;
1070 		trace_xfs_dqput_free(dqp);
1071 
1072 		if (list_lru_add(&qi->qi_lru, &dqp->q_lru))
1073 			XFS_STATS_INC(dqp->q_mount, xs_qm_dquot_unused);
1074 	}
1075 	xfs_dqunlock(dqp);
1076 }
1077 
1078 /*
1079  * Release a dquot. Flush it if dirty, then dqput() it.
1080  * dquot must not be locked.
1081  */
1082 void
1083 xfs_qm_dqrele(
1084 	struct xfs_dquot	*dqp)
1085 {
1086 	if (!dqp)
1087 		return;
1088 
1089 	trace_xfs_dqrele(dqp);
1090 
1091 	xfs_dqlock(dqp);
1092 	/*
1093 	 * We don't care to flush it if the dquot is dirty here.
1094 	 * That will create stutters that we want to avoid.
1095 	 * Instead we do a delayed write when we try to reclaim
1096 	 * a dirty dquot. Also xfs_sync will take part of the burden...
1097 	 */
1098 	xfs_qm_dqput(dqp);
1099 }
1100 
1101 /*
1102  * This is the dquot flushing I/O completion routine.  It is called
1103  * from interrupt level when the buffer containing the dquot is
1104  * flushed to disk.  It is responsible for removing the dquot logitem
1105  * from the AIL if it has not been re-logged, and unlocking the dquot's
1106  * flush lock. This behavior is very similar to that of inodes..
1107  */
1108 static void
1109 xfs_qm_dqflush_done(
1110 	struct xfs_log_item	*lip)
1111 {
1112 	struct xfs_dq_logitem	*qip = (struct xfs_dq_logitem *)lip;
1113 	struct xfs_dquot	*dqp = qip->qli_dquot;
1114 	struct xfs_ail		*ailp = lip->li_ailp;
1115 	xfs_lsn_t		tail_lsn;
1116 
1117 	/*
1118 	 * We only want to pull the item from the AIL if its
1119 	 * location in the log has not changed since we started the flush.
1120 	 * Thus, we only bother if the dquot's lsn has
1121 	 * not changed. First we check the lsn outside the lock
1122 	 * since it's cheaper, and then we recheck while
1123 	 * holding the lock before removing the dquot from the AIL.
1124 	 */
1125 	if (test_bit(XFS_LI_IN_AIL, &lip->li_flags) &&
1126 	    ((lip->li_lsn == qip->qli_flush_lsn) ||
1127 	     test_bit(XFS_LI_FAILED, &lip->li_flags))) {
1128 
1129 		spin_lock(&ailp->ail_lock);
1130 		xfs_clear_li_failed(lip);
1131 		if (lip->li_lsn == qip->qli_flush_lsn) {
1132 			/* xfs_ail_update_finish() drops the AIL lock */
1133 			tail_lsn = xfs_ail_delete_one(ailp, lip);
1134 			xfs_ail_update_finish(ailp, tail_lsn);
1135 		} else {
1136 			spin_unlock(&ailp->ail_lock);
1137 		}
1138 	}
1139 
1140 	/*
1141 	 * Release the dq's flush lock since we're done with it.
1142 	 */
1143 	xfs_dqfunlock(dqp);
1144 }
1145 
1146 void
1147 xfs_buf_dquot_iodone(
1148 	struct xfs_buf		*bp)
1149 {
1150 	struct xfs_log_item	*lip, *n;
1151 
1152 	list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) {
1153 		list_del_init(&lip->li_bio_list);
1154 		xfs_qm_dqflush_done(lip);
1155 	}
1156 }
1157 
1158 void
1159 xfs_buf_dquot_io_fail(
1160 	struct xfs_buf		*bp)
1161 {
1162 	struct xfs_log_item	*lip;
1163 
1164 	spin_lock(&bp->b_mount->m_ail->ail_lock);
1165 	list_for_each_entry(lip, &bp->b_li_list, li_bio_list)
1166 		xfs_set_li_failed(lip, bp);
1167 	spin_unlock(&bp->b_mount->m_ail->ail_lock);
1168 }
1169 
1170 /* Check incore dquot for errors before we flush. */
1171 static xfs_failaddr_t
1172 xfs_qm_dqflush_check(
1173 	struct xfs_dquot	*dqp)
1174 {
1175 	xfs_dqtype_t		type = xfs_dquot_type(dqp);
1176 
1177 	if (type != XFS_DQTYPE_USER &&
1178 	    type != XFS_DQTYPE_GROUP &&
1179 	    type != XFS_DQTYPE_PROJ)
1180 		return __this_address;
1181 
1182 	if (dqp->q_id == 0)
1183 		return NULL;
1184 
1185 	if (dqp->q_blk.softlimit && dqp->q_blk.count > dqp->q_blk.softlimit &&
1186 	    !dqp->q_blk.timer)
1187 		return __this_address;
1188 
1189 	if (dqp->q_ino.softlimit && dqp->q_ino.count > dqp->q_ino.softlimit &&
1190 	    !dqp->q_ino.timer)
1191 		return __this_address;
1192 
1193 	if (dqp->q_rtb.softlimit && dqp->q_rtb.count > dqp->q_rtb.softlimit &&
1194 	    !dqp->q_rtb.timer)
1195 		return __this_address;
1196 
1197 	/* bigtime flag should never be set on root dquots */
1198 	if (dqp->q_type & XFS_DQTYPE_BIGTIME) {
1199 		if (!xfs_has_bigtime(dqp->q_mount))
1200 			return __this_address;
1201 		if (dqp->q_id == 0)
1202 			return __this_address;
1203 	}
1204 
1205 	return NULL;
1206 }
1207 
1208 /*
1209  * Write a modified dquot to disk.
1210  * The dquot must be locked and the flush lock too taken by caller.
1211  * The flush lock will not be unlocked until the dquot reaches the disk,
1212  * but the dquot is free to be unlocked and modified by the caller
1213  * in the interim. Dquot is still locked on return. This behavior is
1214  * identical to that of inodes.
1215  */
1216 int
1217 xfs_qm_dqflush(
1218 	struct xfs_dquot	*dqp,
1219 	struct xfs_buf		**bpp)
1220 {
1221 	struct xfs_mount	*mp = dqp->q_mount;
1222 	struct xfs_log_item	*lip = &dqp->q_logitem.qli_item;
1223 	struct xfs_buf		*bp;
1224 	struct xfs_dqblk	*dqblk;
1225 	xfs_failaddr_t		fa;
1226 	int			error;
1227 
1228 	ASSERT(XFS_DQ_IS_LOCKED(dqp));
1229 	ASSERT(!completion_done(&dqp->q_flush));
1230 
1231 	trace_xfs_dqflush(dqp);
1232 
1233 	*bpp = NULL;
1234 
1235 	xfs_qm_dqunpin_wait(dqp);
1236 
1237 	/*
1238 	 * Get the buffer containing the on-disk dquot
1239 	 */
1240 	error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
1241 				   mp->m_quotainfo->qi_dqchunklen, XBF_TRYLOCK,
1242 				   &bp, &xfs_dquot_buf_ops);
1243 	if (error == -EAGAIN)
1244 		goto out_unlock;
1245 	if (error)
1246 		goto out_abort;
1247 
1248 	fa = xfs_qm_dqflush_check(dqp);
1249 	if (fa) {
1250 		xfs_alert(mp, "corrupt dquot ID 0x%x in memory at %pS",
1251 				dqp->q_id, fa);
1252 		xfs_buf_relse(bp);
1253 		error = -EFSCORRUPTED;
1254 		goto out_abort;
1255 	}
1256 
1257 	/* Flush the incore dquot to the ondisk buffer. */
1258 	dqblk = bp->b_addr + dqp->q_bufoffset;
1259 	xfs_dquot_to_disk(&dqblk->dd_diskdq, dqp);
1260 
1261 	/*
1262 	 * Clear the dirty field and remember the flush lsn for later use.
1263 	 */
1264 	dqp->q_flags &= ~XFS_DQFLAG_DIRTY;
1265 
1266 	xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn,
1267 					&dqp->q_logitem.qli_item.li_lsn);
1268 
1269 	/*
1270 	 * copy the lsn into the on-disk dquot now while we have the in memory
1271 	 * dquot here. This can't be done later in the write verifier as we
1272 	 * can't get access to the log item at that point in time.
1273 	 *
1274 	 * We also calculate the CRC here so that the on-disk dquot in the
1275 	 * buffer always has a valid CRC. This ensures there is no possibility
1276 	 * of a dquot without an up-to-date CRC getting to disk.
1277 	 */
1278 	if (xfs_has_crc(mp)) {
1279 		dqblk->dd_lsn = cpu_to_be64(dqp->q_logitem.qli_item.li_lsn);
1280 		xfs_update_cksum((char *)dqblk, sizeof(struct xfs_dqblk),
1281 				 XFS_DQUOT_CRC_OFF);
1282 	}
1283 
1284 	/*
1285 	 * Attach the dquot to the buffer so that we can remove this dquot from
1286 	 * the AIL and release the flush lock once the dquot is synced to disk.
1287 	 */
1288 	bp->b_flags |= _XBF_DQUOTS;
1289 	list_add_tail(&dqp->q_logitem.qli_item.li_bio_list, &bp->b_li_list);
1290 
1291 	/*
1292 	 * If the buffer is pinned then push on the log so we won't
1293 	 * get stuck waiting in the write for too long.
1294 	 */
1295 	if (xfs_buf_ispinned(bp)) {
1296 		trace_xfs_dqflush_force(dqp);
1297 		xfs_log_force(mp, 0);
1298 	}
1299 
1300 	trace_xfs_dqflush_done(dqp);
1301 	*bpp = bp;
1302 	return 0;
1303 
1304 out_abort:
1305 	dqp->q_flags &= ~XFS_DQFLAG_DIRTY;
1306 	xfs_trans_ail_delete(lip, 0);
1307 	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1308 out_unlock:
1309 	xfs_dqfunlock(dqp);
1310 	return error;
1311 }
1312 
1313 /*
1314  * Lock two xfs_dquot structures.
1315  *
1316  * To avoid deadlocks we always lock the quota structure with
1317  * the lowerd id first.
1318  */
1319 void
1320 xfs_dqlock2(
1321 	struct xfs_dquot	*d1,
1322 	struct xfs_dquot	*d2)
1323 {
1324 	if (d1 && d2) {
1325 		ASSERT(d1 != d2);
1326 		if (d1->q_id > d2->q_id) {
1327 			mutex_lock(&d2->q_qlock);
1328 			mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED);
1329 		} else {
1330 			mutex_lock(&d1->q_qlock);
1331 			mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED);
1332 		}
1333 	} else if (d1) {
1334 		mutex_lock(&d1->q_qlock);
1335 	} else if (d2) {
1336 		mutex_lock(&d2->q_qlock);
1337 	}
1338 }
1339 
1340 int __init
1341 xfs_qm_init(void)
1342 {
1343 	xfs_dquot_cache = kmem_cache_create("xfs_dquot",
1344 					  sizeof(struct xfs_dquot),
1345 					  0, 0, NULL);
1346 	if (!xfs_dquot_cache)
1347 		goto out;
1348 
1349 	xfs_dqtrx_cache = kmem_cache_create("xfs_dqtrx",
1350 					     sizeof(struct xfs_dquot_acct),
1351 					     0, 0, NULL);
1352 	if (!xfs_dqtrx_cache)
1353 		goto out_free_dquot_cache;
1354 
1355 	return 0;
1356 
1357 out_free_dquot_cache:
1358 	kmem_cache_destroy(xfs_dquot_cache);
1359 out:
1360 	return -ENOMEM;
1361 }
1362 
1363 void
1364 xfs_qm_exit(void)
1365 {
1366 	kmem_cache_destroy(xfs_dqtrx_cache);
1367 	kmem_cache_destroy(xfs_dquot_cache);
1368 }
1369 
1370 /*
1371  * Iterate every dquot of a particular type.  The caller must ensure that the
1372  * particular quota type is active.  iter_fn can return negative error codes,
1373  * or -ECANCELED to indicate that it wants to stop iterating.
1374  */
1375 int
1376 xfs_qm_dqiterate(
1377 	struct xfs_mount	*mp,
1378 	xfs_dqtype_t		type,
1379 	xfs_qm_dqiterate_fn	iter_fn,
1380 	void			*priv)
1381 {
1382 	struct xfs_dquot	*dq;
1383 	xfs_dqid_t		id = 0;
1384 	int			error;
1385 
1386 	do {
1387 		error = xfs_qm_dqget_next(mp, id, type, &dq);
1388 		if (error == -ENOENT)
1389 			return 0;
1390 		if (error)
1391 			return error;
1392 
1393 		error = iter_fn(dq, type, priv);
1394 		id = dq->q_id;
1395 		xfs_qm_dqput(dq);
1396 	} while (error == 0 && id != 0);
1397 
1398 	return error;
1399 }
1400