xref: /openbmc/linux/fs/xfs/xfs_dquot.c (revision 612a462a)
1 /*
2  * Copyright (c) 2000-2003 Silicon Graphics, Inc.
3  * All Rights Reserved.
4  *
5  * This program is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU General Public License as
7  * published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it would be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write the Free Software Foundation,
16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
17  */
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_format.h"
21 #include "xfs_log_format.h"
22 #include "xfs_shared.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_bit.h"
25 #include "xfs_mount.h"
26 #include "xfs_defer.h"
27 #include "xfs_inode.h"
28 #include "xfs_bmap.h"
29 #include "xfs_bmap_util.h"
30 #include "xfs_alloc.h"
31 #include "xfs_quota.h"
32 #include "xfs_error.h"
33 #include "xfs_trans.h"
34 #include "xfs_buf_item.h"
35 #include "xfs_trans_space.h"
36 #include "xfs_trans_priv.h"
37 #include "xfs_qm.h"
38 #include "xfs_cksum.h"
39 #include "xfs_trace.h"
40 #include "xfs_log.h"
41 #include "xfs_bmap_btree.h"
42 
43 /*
44  * Lock order:
45  *
46  * ip->i_lock
47  *   qi->qi_tree_lock
48  *     dquot->q_qlock (xfs_dqlock() and friends)
49  *       dquot->q_flush (xfs_dqflock() and friends)
50  *       qi->qi_lru_lock
51  *
52  * If two dquots need to be locked the order is user before group/project,
53  * otherwise by the lowest id first, see xfs_dqlock2.
54  */
55 
56 #ifdef DEBUG
57 xfs_buftarg_t *xfs_dqerror_target;
58 int xfs_do_dqerror;
59 int xfs_dqreq_num;
60 int xfs_dqerror_mod = 33;
61 #endif
62 
63 struct kmem_zone		*xfs_qm_dqtrxzone;
64 static struct kmem_zone		*xfs_qm_dqzone;
65 
66 static struct lock_class_key xfs_dquot_group_class;
67 static struct lock_class_key xfs_dquot_project_class;
68 
69 /*
70  * This is called to free all the memory associated with a dquot
71  */
72 void
73 xfs_qm_dqdestroy(
74 	xfs_dquot_t	*dqp)
75 {
76 	ASSERT(list_empty(&dqp->q_lru));
77 
78 	kmem_free(dqp->q_logitem.qli_item.li_lv_shadow);
79 	mutex_destroy(&dqp->q_qlock);
80 
81 	XFS_STATS_DEC(dqp->q_mount, xs_qm_dquot);
82 	kmem_zone_free(xfs_qm_dqzone, dqp);
83 }
84 
85 /*
86  * If default limits are in force, push them into the dquot now.
87  * We overwrite the dquot limits only if they are zero and this
88  * is not the root dquot.
89  */
90 void
91 xfs_qm_adjust_dqlimits(
92 	struct xfs_mount	*mp,
93 	struct xfs_dquot	*dq)
94 {
95 	struct xfs_quotainfo	*q = mp->m_quotainfo;
96 	struct xfs_disk_dquot	*d = &dq->q_core;
97 	struct xfs_def_quota	*defq;
98 	int			prealloc = 0;
99 
100 	ASSERT(d->d_id);
101 	defq = xfs_get_defquota(dq, q);
102 
103 	if (defq->bsoftlimit && !d->d_blk_softlimit) {
104 		d->d_blk_softlimit = cpu_to_be64(defq->bsoftlimit);
105 		prealloc = 1;
106 	}
107 	if (defq->bhardlimit && !d->d_blk_hardlimit) {
108 		d->d_blk_hardlimit = cpu_to_be64(defq->bhardlimit);
109 		prealloc = 1;
110 	}
111 	if (defq->isoftlimit && !d->d_ino_softlimit)
112 		d->d_ino_softlimit = cpu_to_be64(defq->isoftlimit);
113 	if (defq->ihardlimit && !d->d_ino_hardlimit)
114 		d->d_ino_hardlimit = cpu_to_be64(defq->ihardlimit);
115 	if (defq->rtbsoftlimit && !d->d_rtb_softlimit)
116 		d->d_rtb_softlimit = cpu_to_be64(defq->rtbsoftlimit);
117 	if (defq->rtbhardlimit && !d->d_rtb_hardlimit)
118 		d->d_rtb_hardlimit = cpu_to_be64(defq->rtbhardlimit);
119 
120 	if (prealloc)
121 		xfs_dquot_set_prealloc_limits(dq);
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 	xfs_mount_t		*mp,
140 	xfs_disk_dquot_t	*d)
141 {
142 	ASSERT(d->d_id);
143 
144 #ifdef DEBUG
145 	if (d->d_blk_hardlimit)
146 		ASSERT(be64_to_cpu(d->d_blk_softlimit) <=
147 		       be64_to_cpu(d->d_blk_hardlimit));
148 	if (d->d_ino_hardlimit)
149 		ASSERT(be64_to_cpu(d->d_ino_softlimit) <=
150 		       be64_to_cpu(d->d_ino_hardlimit));
151 	if (d->d_rtb_hardlimit)
152 		ASSERT(be64_to_cpu(d->d_rtb_softlimit) <=
153 		       be64_to_cpu(d->d_rtb_hardlimit));
154 #endif
155 
156 	if (!d->d_btimer) {
157 		if ((d->d_blk_softlimit &&
158 		     (be64_to_cpu(d->d_bcount) >
159 		      be64_to_cpu(d->d_blk_softlimit))) ||
160 		    (d->d_blk_hardlimit &&
161 		     (be64_to_cpu(d->d_bcount) >
162 		      be64_to_cpu(d->d_blk_hardlimit)))) {
163 			d->d_btimer = cpu_to_be32(get_seconds() +
164 					mp->m_quotainfo->qi_btimelimit);
165 		} else {
166 			d->d_bwarns = 0;
167 		}
168 	} else {
169 		if ((!d->d_blk_softlimit ||
170 		     (be64_to_cpu(d->d_bcount) <=
171 		      be64_to_cpu(d->d_blk_softlimit))) &&
172 		    (!d->d_blk_hardlimit ||
173 		    (be64_to_cpu(d->d_bcount) <=
174 		     be64_to_cpu(d->d_blk_hardlimit)))) {
175 			d->d_btimer = 0;
176 		}
177 	}
178 
179 	if (!d->d_itimer) {
180 		if ((d->d_ino_softlimit &&
181 		     (be64_to_cpu(d->d_icount) >
182 		      be64_to_cpu(d->d_ino_softlimit))) ||
183 		    (d->d_ino_hardlimit &&
184 		     (be64_to_cpu(d->d_icount) >
185 		      be64_to_cpu(d->d_ino_hardlimit)))) {
186 			d->d_itimer = cpu_to_be32(get_seconds() +
187 					mp->m_quotainfo->qi_itimelimit);
188 		} else {
189 			d->d_iwarns = 0;
190 		}
191 	} else {
192 		if ((!d->d_ino_softlimit ||
193 		     (be64_to_cpu(d->d_icount) <=
194 		      be64_to_cpu(d->d_ino_softlimit)))  &&
195 		    (!d->d_ino_hardlimit ||
196 		     (be64_to_cpu(d->d_icount) <=
197 		      be64_to_cpu(d->d_ino_hardlimit)))) {
198 			d->d_itimer = 0;
199 		}
200 	}
201 
202 	if (!d->d_rtbtimer) {
203 		if ((d->d_rtb_softlimit &&
204 		     (be64_to_cpu(d->d_rtbcount) >
205 		      be64_to_cpu(d->d_rtb_softlimit))) ||
206 		    (d->d_rtb_hardlimit &&
207 		     (be64_to_cpu(d->d_rtbcount) >
208 		      be64_to_cpu(d->d_rtb_hardlimit)))) {
209 			d->d_rtbtimer = cpu_to_be32(get_seconds() +
210 					mp->m_quotainfo->qi_rtbtimelimit);
211 		} else {
212 			d->d_rtbwarns = 0;
213 		}
214 	} else {
215 		if ((!d->d_rtb_softlimit ||
216 		     (be64_to_cpu(d->d_rtbcount) <=
217 		      be64_to_cpu(d->d_rtb_softlimit))) &&
218 		    (!d->d_rtb_hardlimit ||
219 		     (be64_to_cpu(d->d_rtbcount) <=
220 		      be64_to_cpu(d->d_rtb_hardlimit)))) {
221 			d->d_rtbtimer = 0;
222 		}
223 	}
224 }
225 
226 /*
227  * initialize a buffer full of dquots and log the whole thing
228  */
229 STATIC void
230 xfs_qm_init_dquot_blk(
231 	xfs_trans_t	*tp,
232 	xfs_mount_t	*mp,
233 	xfs_dqid_t	id,
234 	uint		type,
235 	xfs_buf_t	*bp)
236 {
237 	struct xfs_quotainfo	*q = mp->m_quotainfo;
238 	xfs_dqblk_t	*d;
239 	xfs_dqid_t	curid;
240 	int		i;
241 
242 	ASSERT(tp);
243 	ASSERT(xfs_buf_islocked(bp));
244 
245 	d = bp->b_addr;
246 
247 	/*
248 	 * ID of the first dquot in the block - id's are zero based.
249 	 */
250 	curid = id - (id % q->qi_dqperchunk);
251 	memset(d, 0, BBTOB(q->qi_dqchunklen));
252 	for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++) {
253 		d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
254 		d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
255 		d->dd_diskdq.d_id = cpu_to_be32(curid);
256 		d->dd_diskdq.d_flags = type;
257 		if (xfs_sb_version_hascrc(&mp->m_sb)) {
258 			uuid_copy(&d->dd_uuid, &mp->m_sb.sb_meta_uuid);
259 			xfs_update_cksum((char *)d, sizeof(struct xfs_dqblk),
260 					 XFS_DQUOT_CRC_OFF);
261 		}
262 	}
263 
264 	xfs_trans_dquot_buf(tp, bp,
265 			    (type & XFS_DQ_USER ? XFS_BLF_UDQUOT_BUF :
266 			    ((type & XFS_DQ_PROJ) ? XFS_BLF_PDQUOT_BUF :
267 			     XFS_BLF_GDQUOT_BUF)));
268 	xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1);
269 }
270 
271 /*
272  * Initialize the dynamic speculative preallocation thresholds. The lo/hi
273  * watermarks correspond to the soft and hard limits by default. If a soft limit
274  * is not specified, we use 95% of the hard limit.
275  */
276 void
277 xfs_dquot_set_prealloc_limits(struct xfs_dquot *dqp)
278 {
279 	uint64_t space;
280 
281 	dqp->q_prealloc_hi_wmark = be64_to_cpu(dqp->q_core.d_blk_hardlimit);
282 	dqp->q_prealloc_lo_wmark = be64_to_cpu(dqp->q_core.d_blk_softlimit);
283 	if (!dqp->q_prealloc_lo_wmark) {
284 		dqp->q_prealloc_lo_wmark = dqp->q_prealloc_hi_wmark;
285 		do_div(dqp->q_prealloc_lo_wmark, 100);
286 		dqp->q_prealloc_lo_wmark *= 95;
287 	}
288 
289 	space = dqp->q_prealloc_hi_wmark;
290 
291 	do_div(space, 100);
292 	dqp->q_low_space[XFS_QLOWSP_1_PCNT] = space;
293 	dqp->q_low_space[XFS_QLOWSP_3_PCNT] = space * 3;
294 	dqp->q_low_space[XFS_QLOWSP_5_PCNT] = space * 5;
295 }
296 
297 /*
298  * Allocate a block and fill it with dquots.
299  * This is called when the bmapi finds a hole.
300  */
301 STATIC int
302 xfs_qm_dqalloc(
303 	xfs_trans_t	**tpp,
304 	xfs_mount_t	*mp,
305 	xfs_dquot_t	*dqp,
306 	xfs_inode_t	*quotip,
307 	xfs_fileoff_t	offset_fsb,
308 	xfs_buf_t	**O_bpp)
309 {
310 	xfs_fsblock_t	firstblock;
311 	struct xfs_defer_ops dfops;
312 	xfs_bmbt_irec_t map;
313 	int		nmaps, error;
314 	xfs_buf_t	*bp;
315 	xfs_trans_t	*tp = *tpp;
316 
317 	ASSERT(tp != NULL);
318 
319 	trace_xfs_dqalloc(dqp);
320 
321 	/*
322 	 * Initialize the bmap freelist prior to calling bmapi code.
323 	 */
324 	xfs_defer_init(&dfops, &firstblock);
325 	xfs_ilock(quotip, XFS_ILOCK_EXCL);
326 	/*
327 	 * Return if this type of quotas is turned off while we didn't
328 	 * have an inode lock
329 	 */
330 	if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) {
331 		xfs_iunlock(quotip, XFS_ILOCK_EXCL);
332 		return -ESRCH;
333 	}
334 
335 	xfs_trans_ijoin(tp, quotip, XFS_ILOCK_EXCL);
336 	nmaps = 1;
337 	error = xfs_bmapi_write(tp, quotip, offset_fsb,
338 				XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA,
339 				&firstblock, XFS_QM_DQALLOC_SPACE_RES(mp),
340 				&map, &nmaps, &dfops);
341 	if (error)
342 		goto error0;
343 	ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB);
344 	ASSERT(nmaps == 1);
345 	ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
346 	       (map.br_startblock != HOLESTARTBLOCK));
347 
348 	/*
349 	 * Keep track of the blkno to save a lookup later
350 	 */
351 	dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
352 
353 	/* now we can just get the buffer (there's nothing to read yet) */
354 	bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
355 			       dqp->q_blkno,
356 			       mp->m_quotainfo->qi_dqchunklen,
357 			       0);
358 	if (!bp) {
359 		error = -ENOMEM;
360 		goto error1;
361 	}
362 	bp->b_ops = &xfs_dquot_buf_ops;
363 
364 	/*
365 	 * Make a chunk of dquots out of this buffer and log
366 	 * the entire thing.
367 	 */
368 	xfs_qm_init_dquot_blk(tp, mp, be32_to_cpu(dqp->q_core.d_id),
369 			      dqp->dq_flags & XFS_DQ_ALLTYPES, bp);
370 
371 	/*
372 	 * xfs_defer_finish() may commit the current transaction and
373 	 * start a second transaction if the freelist is not empty.
374 	 *
375 	 * Since we still want to modify this buffer, we need to
376 	 * ensure that the buffer is not released on commit of
377 	 * the first transaction and ensure the buffer is added to the
378 	 * second transaction.
379 	 *
380 	 * If there is only one transaction then don't stop the buffer
381 	 * from being released when it commits later on.
382 	 */
383 
384 	xfs_trans_bhold(tp, bp);
385 
386 	error = xfs_defer_finish(tpp, &dfops);
387 	if (error)
388 		goto error1;
389 
390 	/* Transaction was committed? */
391 	if (*tpp != tp) {
392 		tp = *tpp;
393 		xfs_trans_bjoin(tp, bp);
394 	} else {
395 		xfs_trans_bhold_release(tp, bp);
396 	}
397 
398 	*O_bpp = bp;
399 	return 0;
400 
401 error1:
402 	xfs_defer_cancel(&dfops);
403 error0:
404 	xfs_iunlock(quotip, XFS_ILOCK_EXCL);
405 
406 	return error;
407 }
408 
409 STATIC int
410 xfs_qm_dqrepair(
411 	struct xfs_mount	*mp,
412 	struct xfs_trans	*tp,
413 	struct xfs_dquot	*dqp,
414 	xfs_dqid_t		firstid,
415 	struct xfs_buf		**bpp)
416 {
417 	int			error;
418 	struct xfs_disk_dquot	*ddq;
419 	struct xfs_dqblk	*d;
420 	int			i;
421 
422 	/*
423 	 * Read the buffer without verification so we get the corrupted
424 	 * buffer returned to us. make sure we verify it on write, though.
425 	 */
426 	error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, dqp->q_blkno,
427 				   mp->m_quotainfo->qi_dqchunklen,
428 				   0, bpp, NULL);
429 
430 	if (error) {
431 		ASSERT(*bpp == NULL);
432 		return error;
433 	}
434 	(*bpp)->b_ops = &xfs_dquot_buf_ops;
435 
436 	ASSERT(xfs_buf_islocked(*bpp));
437 	d = (struct xfs_dqblk *)(*bpp)->b_addr;
438 
439 	/* Do the actual repair of dquots in this buffer */
440 	for (i = 0; i < mp->m_quotainfo->qi_dqperchunk; i++) {
441 		ddq = &d[i].dd_diskdq;
442 		error = xfs_dqcheck(mp, ddq, firstid + i,
443 				       dqp->dq_flags & XFS_DQ_ALLTYPES,
444 				       XFS_QMOPT_DQREPAIR, "xfs_qm_dqrepair");
445 		if (error) {
446 			/* repair failed, we're screwed */
447 			xfs_trans_brelse(tp, *bpp);
448 			return -EIO;
449 		}
450 	}
451 
452 	return 0;
453 }
454 
455 /*
456  * Maps a dquot to the buffer containing its on-disk version.
457  * This returns a ptr to the buffer containing the on-disk dquot
458  * in the bpp param, and a ptr to the on-disk dquot within that buffer
459  */
460 STATIC int
461 xfs_qm_dqtobp(
462 	xfs_trans_t		**tpp,
463 	xfs_dquot_t		*dqp,
464 	xfs_disk_dquot_t	**O_ddpp,
465 	xfs_buf_t		**O_bpp,
466 	uint			flags)
467 {
468 	struct xfs_bmbt_irec	map;
469 	int			nmaps = 1, error;
470 	struct xfs_buf		*bp;
471 	struct xfs_inode	*quotip;
472 	struct xfs_mount	*mp = dqp->q_mount;
473 	xfs_dqid_t		id = be32_to_cpu(dqp->q_core.d_id);
474 	struct xfs_trans	*tp = (tpp ? *tpp : NULL);
475 	uint			lock_mode;
476 
477 	quotip = xfs_quota_inode(dqp->q_mount, dqp->dq_flags);
478 	dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk;
479 
480 	lock_mode = xfs_ilock_data_map_shared(quotip);
481 	if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) {
482 		/*
483 		 * Return if this type of quotas is turned off while we
484 		 * didn't have the quota inode lock.
485 		 */
486 		xfs_iunlock(quotip, lock_mode);
487 		return -ESRCH;
488 	}
489 
490 	/*
491 	 * Find the block map; no allocations yet
492 	 */
493 	error = xfs_bmapi_read(quotip, dqp->q_fileoffset,
494 			       XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0);
495 
496 	xfs_iunlock(quotip, lock_mode);
497 	if (error)
498 		return error;
499 
500 	ASSERT(nmaps == 1);
501 	ASSERT(map.br_blockcount == 1);
502 
503 	/*
504 	 * Offset of dquot in the (fixed sized) dquot chunk.
505 	 */
506 	dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) *
507 		sizeof(xfs_dqblk_t);
508 
509 	ASSERT(map.br_startblock != DELAYSTARTBLOCK);
510 	if (map.br_startblock == HOLESTARTBLOCK) {
511 		/*
512 		 * We don't allocate unless we're asked to
513 		 */
514 		if (!(flags & XFS_QMOPT_DQALLOC))
515 			return -ENOENT;
516 
517 		ASSERT(tp);
518 		error = xfs_qm_dqalloc(tpp, mp, dqp, quotip,
519 					dqp->q_fileoffset, &bp);
520 		if (error)
521 			return error;
522 		tp = *tpp;
523 	} else {
524 		trace_xfs_dqtobp_read(dqp);
525 
526 		/*
527 		 * store the blkno etc so that we don't have to do the
528 		 * mapping all the time
529 		 */
530 		dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
531 
532 		error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
533 					   dqp->q_blkno,
534 					   mp->m_quotainfo->qi_dqchunklen,
535 					   0, &bp, &xfs_dquot_buf_ops);
536 
537 		if (error == -EFSCORRUPTED && (flags & XFS_QMOPT_DQREPAIR)) {
538 			xfs_dqid_t firstid = (xfs_dqid_t)map.br_startoff *
539 						mp->m_quotainfo->qi_dqperchunk;
540 			ASSERT(bp == NULL);
541 			error = xfs_qm_dqrepair(mp, tp, dqp, firstid, &bp);
542 		}
543 
544 		if (error) {
545 			ASSERT(bp == NULL);
546 			return error;
547 		}
548 	}
549 
550 	ASSERT(xfs_buf_islocked(bp));
551 	*O_bpp = bp;
552 	*O_ddpp = bp->b_addr + dqp->q_bufoffset;
553 
554 	return 0;
555 }
556 
557 
558 /*
559  * Read in the ondisk dquot using dqtobp() then copy it to an incore version,
560  * and release the buffer immediately.
561  *
562  * If XFS_QMOPT_DQALLOC is set, allocate a dquot on disk if it needed.
563  */
564 int
565 xfs_qm_dqread(
566 	struct xfs_mount	*mp,
567 	xfs_dqid_t		id,
568 	uint			type,
569 	uint			flags,
570 	struct xfs_dquot	**O_dqpp)
571 {
572 	struct xfs_dquot	*dqp;
573 	struct xfs_disk_dquot	*ddqp;
574 	struct xfs_buf		*bp;
575 	struct xfs_trans	*tp = NULL;
576 	int			error;
577 
578 	dqp = kmem_zone_zalloc(xfs_qm_dqzone, KM_SLEEP);
579 
580 	dqp->dq_flags = type;
581 	dqp->q_core.d_id = cpu_to_be32(id);
582 	dqp->q_mount = mp;
583 	INIT_LIST_HEAD(&dqp->q_lru);
584 	mutex_init(&dqp->q_qlock);
585 	init_waitqueue_head(&dqp->q_pinwait);
586 
587 	/*
588 	 * Because we want to use a counting completion, complete
589 	 * the flush completion once to allow a single access to
590 	 * the flush completion without blocking.
591 	 */
592 	init_completion(&dqp->q_flush);
593 	complete(&dqp->q_flush);
594 
595 	/*
596 	 * Make sure group quotas have a different lock class than user
597 	 * quotas.
598 	 */
599 	switch (type) {
600 	case XFS_DQ_USER:
601 		/* uses the default lock class */
602 		break;
603 	case XFS_DQ_GROUP:
604 		lockdep_set_class(&dqp->q_qlock, &xfs_dquot_group_class);
605 		break;
606 	case XFS_DQ_PROJ:
607 		lockdep_set_class(&dqp->q_qlock, &xfs_dquot_project_class);
608 		break;
609 	default:
610 		ASSERT(0);
611 		break;
612 	}
613 
614 	XFS_STATS_INC(mp, xs_qm_dquot);
615 
616 	trace_xfs_dqread(dqp);
617 
618 	if (flags & XFS_QMOPT_DQALLOC) {
619 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_qm_dqalloc,
620 				XFS_QM_DQALLOC_SPACE_RES(mp), 0, 0, &tp);
621 		if (error)
622 			goto error0;
623 	}
624 
625 	/*
626 	 * get a pointer to the on-disk dquot and the buffer containing it
627 	 * dqp already knows its own type (GROUP/USER).
628 	 */
629 	error = xfs_qm_dqtobp(&tp, dqp, &ddqp, &bp, flags);
630 	if (error) {
631 		/*
632 		 * This can happen if quotas got turned off (ESRCH),
633 		 * or if the dquot didn't exist on disk and we ask to
634 		 * allocate (ENOENT).
635 		 */
636 		trace_xfs_dqread_fail(dqp);
637 		goto error1;
638 	}
639 
640 	/* copy everything from disk dquot to the incore dquot */
641 	memcpy(&dqp->q_core, ddqp, sizeof(xfs_disk_dquot_t));
642 	xfs_qm_dquot_logitem_init(dqp);
643 
644 	/*
645 	 * Reservation counters are defined as reservation plus current usage
646 	 * to avoid having to add every time.
647 	 */
648 	dqp->q_res_bcount = be64_to_cpu(ddqp->d_bcount);
649 	dqp->q_res_icount = be64_to_cpu(ddqp->d_icount);
650 	dqp->q_res_rtbcount = be64_to_cpu(ddqp->d_rtbcount);
651 
652 	/* initialize the dquot speculative prealloc thresholds */
653 	xfs_dquot_set_prealloc_limits(dqp);
654 
655 	/* Mark the buf so that this will stay incore a little longer */
656 	xfs_buf_set_ref(bp, XFS_DQUOT_REF);
657 
658 	/*
659 	 * We got the buffer with a xfs_trans_read_buf() (in dqtobp())
660 	 * So we need to release with xfs_trans_brelse().
661 	 * The strategy here is identical to that of inodes; we lock
662 	 * the dquot in xfs_qm_dqget() before making it accessible to
663 	 * others. This is because dquots, like inodes, need a good level of
664 	 * concurrency, and we don't want to take locks on the entire buffers
665 	 * for dquot accesses.
666 	 * Note also that the dquot buffer may even be dirty at this point, if
667 	 * this particular dquot was repaired. We still aren't afraid to
668 	 * brelse it because we have the changes incore.
669 	 */
670 	ASSERT(xfs_buf_islocked(bp));
671 	xfs_trans_brelse(tp, bp);
672 
673 	if (tp) {
674 		error = xfs_trans_commit(tp);
675 		if (error)
676 			goto error0;
677 	}
678 
679 	*O_dqpp = dqp;
680 	return error;
681 
682 error1:
683 	if (tp)
684 		xfs_trans_cancel(tp);
685 error0:
686 	xfs_qm_dqdestroy(dqp);
687 	*O_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 	uint			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 	xfs_extnum_t		idx;
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 	if (!(quotip->i_df.if_flags & XFS_IFEXTENTS)) {
725 		error = xfs_iread_extents(NULL, quotip, XFS_DATA_FORK);
726 		if (error)
727 			return error;
728 	}
729 
730 	if (xfs_iext_lookup_extent(quotip, &quotip->i_df, start, &idx, &got)) {
731 		/* contiguous chunk, bump startoff for the id calculation */
732 		if (got.br_startoff < start)
733 			got.br_startoff = start;
734 		*id = got.br_startoff * mp->m_quotainfo->qi_dqperchunk;
735 	} else {
736 		error = -ENOENT;
737 	}
738 
739 	xfs_iunlock(quotip, lock_flags);
740 
741 	return error;
742 }
743 
744 /*
745  * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a
746  * a locked dquot, doing an allocation (if requested) as needed.
747  * When both an inode and an id are given, the inode's id takes precedence.
748  * That is, if the id changes while we don't hold the ilock inside this
749  * function, the new dquot is returned, not necessarily the one requested
750  * in the id argument.
751  */
752 int
753 xfs_qm_dqget(
754 	xfs_mount_t	*mp,
755 	xfs_inode_t	*ip,	  /* locked inode (optional) */
756 	xfs_dqid_t	id,	  /* uid/projid/gid depending on type */
757 	uint		type,	  /* XFS_DQ_USER/XFS_DQ_PROJ/XFS_DQ_GROUP */
758 	uint		flags,	  /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */
759 	xfs_dquot_t	**O_dqpp) /* OUT : locked incore dquot */
760 {
761 	struct xfs_quotainfo	*qi = mp->m_quotainfo;
762 	struct radix_tree_root *tree = xfs_dquot_tree(qi, type);
763 	struct xfs_dquot	*dqp;
764 	int			error;
765 
766 	ASSERT(XFS_IS_QUOTA_RUNNING(mp));
767 	if ((! XFS_IS_UQUOTA_ON(mp) && type == XFS_DQ_USER) ||
768 	    (! XFS_IS_PQUOTA_ON(mp) && type == XFS_DQ_PROJ) ||
769 	    (! XFS_IS_GQUOTA_ON(mp) && type == XFS_DQ_GROUP)) {
770 		return -ESRCH;
771 	}
772 
773 #ifdef DEBUG
774 	if (xfs_do_dqerror) {
775 		if ((xfs_dqerror_target == mp->m_ddev_targp) &&
776 		    (xfs_dqreq_num++ % xfs_dqerror_mod) == 0) {
777 			xfs_debug(mp, "Returning error in dqget");
778 			return -EIO;
779 		}
780 	}
781 
782 	ASSERT(type == XFS_DQ_USER ||
783 	       type == XFS_DQ_PROJ ||
784 	       type == XFS_DQ_GROUP);
785 	if (ip) {
786 		ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
787 		ASSERT(xfs_inode_dquot(ip, type) == NULL);
788 	}
789 #endif
790 
791 restart:
792 	mutex_lock(&qi->qi_tree_lock);
793 	dqp = radix_tree_lookup(tree, id);
794 	if (dqp) {
795 		xfs_dqlock(dqp);
796 		if (dqp->dq_flags & XFS_DQ_FREEING) {
797 			xfs_dqunlock(dqp);
798 			mutex_unlock(&qi->qi_tree_lock);
799 			trace_xfs_dqget_freeing(dqp);
800 			delay(1);
801 			goto restart;
802 		}
803 
804 		/* uninit / unused quota found in radix tree, keep looking  */
805 		if (flags & XFS_QMOPT_DQNEXT) {
806 			if (XFS_IS_DQUOT_UNINITIALIZED(dqp)) {
807 				xfs_dqunlock(dqp);
808 				mutex_unlock(&qi->qi_tree_lock);
809 				error = xfs_dq_get_next_id(mp, type, &id);
810 				if (error)
811 					return error;
812 				goto restart;
813 			}
814 		}
815 
816 		dqp->q_nrefs++;
817 		mutex_unlock(&qi->qi_tree_lock);
818 
819 		trace_xfs_dqget_hit(dqp);
820 		XFS_STATS_INC(mp, xs_qm_dqcachehits);
821 		*O_dqpp = dqp;
822 		return 0;
823 	}
824 	mutex_unlock(&qi->qi_tree_lock);
825 	XFS_STATS_INC(mp, xs_qm_dqcachemisses);
826 
827 	/*
828 	 * Dquot cache miss. We don't want to keep the inode lock across
829 	 * a (potential) disk read. Also we don't want to deal with the lock
830 	 * ordering between quotainode and this inode. OTOH, dropping the inode
831 	 * lock here means dealing with a chown that can happen before
832 	 * we re-acquire the lock.
833 	 */
834 	if (ip)
835 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
836 
837 	error = xfs_qm_dqread(mp, id, type, flags, &dqp);
838 
839 	if (ip)
840 		xfs_ilock(ip, XFS_ILOCK_EXCL);
841 
842 	/* If we are asked to find next active id, keep looking */
843 	if (error == -ENOENT && (flags & XFS_QMOPT_DQNEXT)) {
844 		error = xfs_dq_get_next_id(mp, type, &id);
845 		if (!error)
846 			goto restart;
847 	}
848 
849 	if (error)
850 		return error;
851 
852 	if (ip) {
853 		/*
854 		 * A dquot could be attached to this inode by now, since
855 		 * we had dropped the ilock.
856 		 */
857 		if (xfs_this_quota_on(mp, type)) {
858 			struct xfs_dquot	*dqp1;
859 
860 			dqp1 = xfs_inode_dquot(ip, type);
861 			if (dqp1) {
862 				xfs_qm_dqdestroy(dqp);
863 				dqp = dqp1;
864 				xfs_dqlock(dqp);
865 				goto dqret;
866 			}
867 		} else {
868 			/* inode stays locked on return */
869 			xfs_qm_dqdestroy(dqp);
870 			return -ESRCH;
871 		}
872 	}
873 
874 	mutex_lock(&qi->qi_tree_lock);
875 	error = radix_tree_insert(tree, id, dqp);
876 	if (unlikely(error)) {
877 		WARN_ON(error != -EEXIST);
878 
879 		/*
880 		 * Duplicate found. Just throw away the new dquot and start
881 		 * over.
882 		 */
883 		mutex_unlock(&qi->qi_tree_lock);
884 		trace_xfs_dqget_dup(dqp);
885 		xfs_qm_dqdestroy(dqp);
886 		XFS_STATS_INC(mp, xs_qm_dquot_dups);
887 		goto restart;
888 	}
889 
890 	/*
891 	 * We return a locked dquot to the caller, with a reference taken
892 	 */
893 	xfs_dqlock(dqp);
894 	dqp->q_nrefs = 1;
895 
896 	qi->qi_dquots++;
897 	mutex_unlock(&qi->qi_tree_lock);
898 
899 	/* If we are asked to find next active id, keep looking */
900 	if (flags & XFS_QMOPT_DQNEXT) {
901 		if (XFS_IS_DQUOT_UNINITIALIZED(dqp)) {
902 			xfs_qm_dqput(dqp);
903 			error = xfs_dq_get_next_id(mp, type, &id);
904 			if (error)
905 				return error;
906 			goto restart;
907 		}
908 	}
909 
910  dqret:
911 	ASSERT((ip == NULL) || xfs_isilocked(ip, XFS_ILOCK_EXCL));
912 	trace_xfs_dqget_miss(dqp);
913 	*O_dqpp = dqp;
914 	return 0;
915 }
916 
917 /*
918  * Release a reference to the dquot (decrement ref-count) and unlock it.
919  *
920  * If there is a group quota attached to this dquot, carefully release that
921  * too without tripping over deadlocks'n'stuff.
922  */
923 void
924 xfs_qm_dqput(
925 	struct xfs_dquot	*dqp)
926 {
927 	ASSERT(dqp->q_nrefs > 0);
928 	ASSERT(XFS_DQ_IS_LOCKED(dqp));
929 
930 	trace_xfs_dqput(dqp);
931 
932 	if (--dqp->q_nrefs == 0) {
933 		struct xfs_quotainfo	*qi = dqp->q_mount->m_quotainfo;
934 		trace_xfs_dqput_free(dqp);
935 
936 		if (list_lru_add(&qi->qi_lru, &dqp->q_lru))
937 			XFS_STATS_INC(dqp->q_mount, xs_qm_dquot_unused);
938 	}
939 	xfs_dqunlock(dqp);
940 }
941 
942 /*
943  * Release a dquot. Flush it if dirty, then dqput() it.
944  * dquot must not be locked.
945  */
946 void
947 xfs_qm_dqrele(
948 	xfs_dquot_t	*dqp)
949 {
950 	if (!dqp)
951 		return;
952 
953 	trace_xfs_dqrele(dqp);
954 
955 	xfs_dqlock(dqp);
956 	/*
957 	 * We don't care to flush it if the dquot is dirty here.
958 	 * That will create stutters that we want to avoid.
959 	 * Instead we do a delayed write when we try to reclaim
960 	 * a dirty dquot. Also xfs_sync will take part of the burden...
961 	 */
962 	xfs_qm_dqput(dqp);
963 }
964 
965 /*
966  * This is the dquot flushing I/O completion routine.  It is called
967  * from interrupt level when the buffer containing the dquot is
968  * flushed to disk.  It is responsible for removing the dquot logitem
969  * from the AIL if it has not been re-logged, and unlocking the dquot's
970  * flush lock. This behavior is very similar to that of inodes..
971  */
972 STATIC void
973 xfs_qm_dqflush_done(
974 	struct xfs_buf		*bp,
975 	struct xfs_log_item	*lip)
976 {
977 	xfs_dq_logitem_t	*qip = (struct xfs_dq_logitem *)lip;
978 	xfs_dquot_t		*dqp = qip->qli_dquot;
979 	struct xfs_ail		*ailp = lip->li_ailp;
980 
981 	/*
982 	 * We only want to pull the item from the AIL if its
983 	 * location in the log has not changed since we started the flush.
984 	 * Thus, we only bother if the dquot's lsn has
985 	 * not changed. First we check the lsn outside the lock
986 	 * since it's cheaper, and then we recheck while
987 	 * holding the lock before removing the dquot from the AIL.
988 	 */
989 	if ((lip->li_flags & XFS_LI_IN_AIL) &&
990 	    lip->li_lsn == qip->qli_flush_lsn) {
991 
992 		/* xfs_trans_ail_delete() drops the AIL lock. */
993 		spin_lock(&ailp->xa_lock);
994 		if (lip->li_lsn == qip->qli_flush_lsn)
995 			xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE);
996 		else
997 			spin_unlock(&ailp->xa_lock);
998 	}
999 
1000 	/*
1001 	 * Release the dq's flush lock since we're done with it.
1002 	 */
1003 	xfs_dqfunlock(dqp);
1004 }
1005 
1006 /*
1007  * Write a modified dquot to disk.
1008  * The dquot must be locked and the flush lock too taken by caller.
1009  * The flush lock will not be unlocked until the dquot reaches the disk,
1010  * but the dquot is free to be unlocked and modified by the caller
1011  * in the interim. Dquot is still locked on return. This behavior is
1012  * identical to that of inodes.
1013  */
1014 int
1015 xfs_qm_dqflush(
1016 	struct xfs_dquot	*dqp,
1017 	struct xfs_buf		**bpp)
1018 {
1019 	struct xfs_mount	*mp = dqp->q_mount;
1020 	struct xfs_buf		*bp;
1021 	struct xfs_disk_dquot	*ddqp;
1022 	int			error;
1023 
1024 	ASSERT(XFS_DQ_IS_LOCKED(dqp));
1025 	ASSERT(!completion_done(&dqp->q_flush));
1026 
1027 	trace_xfs_dqflush(dqp);
1028 
1029 	*bpp = NULL;
1030 
1031 	xfs_qm_dqunpin_wait(dqp);
1032 
1033 	/*
1034 	 * This may have been unpinned because the filesystem is shutting
1035 	 * down forcibly. If that's the case we must not write this dquot
1036 	 * to disk, because the log record didn't make it to disk.
1037 	 *
1038 	 * We also have to remove the log item from the AIL in this case,
1039 	 * as we wait for an emptry AIL as part of the unmount process.
1040 	 */
1041 	if (XFS_FORCED_SHUTDOWN(mp)) {
1042 		struct xfs_log_item	*lip = &dqp->q_logitem.qli_item;
1043 		dqp->dq_flags &= ~XFS_DQ_DIRTY;
1044 
1045 		xfs_trans_ail_remove(lip, SHUTDOWN_CORRUPT_INCORE);
1046 
1047 		error = -EIO;
1048 		goto out_unlock;
1049 	}
1050 
1051 	/*
1052 	 * Get the buffer containing the on-disk dquot
1053 	 */
1054 	error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
1055 				   mp->m_quotainfo->qi_dqchunklen, 0, &bp,
1056 				   &xfs_dquot_buf_ops);
1057 	if (error)
1058 		goto out_unlock;
1059 
1060 	/*
1061 	 * Calculate the location of the dquot inside the buffer.
1062 	 */
1063 	ddqp = bp->b_addr + dqp->q_bufoffset;
1064 
1065 	/*
1066 	 * A simple sanity check in case we got a corrupted dquot..
1067 	 */
1068 	error = xfs_dqcheck(mp, &dqp->q_core, be32_to_cpu(ddqp->d_id), 0,
1069 			   XFS_QMOPT_DOWARN, "dqflush (incore copy)");
1070 	if (error) {
1071 		xfs_buf_relse(bp);
1072 		xfs_dqfunlock(dqp);
1073 		xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1074 		return -EIO;
1075 	}
1076 
1077 	/* This is the only portion of data that needs to persist */
1078 	memcpy(ddqp, &dqp->q_core, sizeof(xfs_disk_dquot_t));
1079 
1080 	/*
1081 	 * Clear the dirty field and remember the flush lsn for later use.
1082 	 */
1083 	dqp->dq_flags &= ~XFS_DQ_DIRTY;
1084 
1085 	xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn,
1086 					&dqp->q_logitem.qli_item.li_lsn);
1087 
1088 	/*
1089 	 * copy the lsn into the on-disk dquot now while we have the in memory
1090 	 * dquot here. This can't be done later in the write verifier as we
1091 	 * can't get access to the log item at that point in time.
1092 	 *
1093 	 * We also calculate the CRC here so that the on-disk dquot in the
1094 	 * buffer always has a valid CRC. This ensures there is no possibility
1095 	 * of a dquot without an up-to-date CRC getting to disk.
1096 	 */
1097 	if (xfs_sb_version_hascrc(&mp->m_sb)) {
1098 		struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddqp;
1099 
1100 		dqb->dd_lsn = cpu_to_be64(dqp->q_logitem.qli_item.li_lsn);
1101 		xfs_update_cksum((char *)dqb, sizeof(struct xfs_dqblk),
1102 				 XFS_DQUOT_CRC_OFF);
1103 	}
1104 
1105 	/*
1106 	 * Attach an iodone routine so that we can remove this dquot from the
1107 	 * AIL and release the flush lock once the dquot is synced to disk.
1108 	 */
1109 	xfs_buf_attach_iodone(bp, xfs_qm_dqflush_done,
1110 				  &dqp->q_logitem.qli_item);
1111 
1112 	/*
1113 	 * If the buffer is pinned then push on the log so we won't
1114 	 * get stuck waiting in the write for too long.
1115 	 */
1116 	if (xfs_buf_ispinned(bp)) {
1117 		trace_xfs_dqflush_force(dqp);
1118 		xfs_log_force(mp, 0);
1119 	}
1120 
1121 	trace_xfs_dqflush_done(dqp);
1122 	*bpp = bp;
1123 	return 0;
1124 
1125 out_unlock:
1126 	xfs_dqfunlock(dqp);
1127 	return -EIO;
1128 }
1129 
1130 /*
1131  * Lock two xfs_dquot structures.
1132  *
1133  * To avoid deadlocks we always lock the quota structure with
1134  * the lowerd id first.
1135  */
1136 void
1137 xfs_dqlock2(
1138 	xfs_dquot_t	*d1,
1139 	xfs_dquot_t	*d2)
1140 {
1141 	if (d1 && d2) {
1142 		ASSERT(d1 != d2);
1143 		if (be32_to_cpu(d1->q_core.d_id) >
1144 		    be32_to_cpu(d2->q_core.d_id)) {
1145 			mutex_lock(&d2->q_qlock);
1146 			mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED);
1147 		} else {
1148 			mutex_lock(&d1->q_qlock);
1149 			mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED);
1150 		}
1151 	} else if (d1) {
1152 		mutex_lock(&d1->q_qlock);
1153 	} else if (d2) {
1154 		mutex_lock(&d2->q_qlock);
1155 	}
1156 }
1157 
1158 int __init
1159 xfs_qm_init(void)
1160 {
1161 	xfs_qm_dqzone =
1162 		kmem_zone_init(sizeof(struct xfs_dquot), "xfs_dquot");
1163 	if (!xfs_qm_dqzone)
1164 		goto out;
1165 
1166 	xfs_qm_dqtrxzone =
1167 		kmem_zone_init(sizeof(struct xfs_dquot_acct), "xfs_dqtrx");
1168 	if (!xfs_qm_dqtrxzone)
1169 		goto out_free_dqzone;
1170 
1171 	return 0;
1172 
1173 out_free_dqzone:
1174 	kmem_zone_destroy(xfs_qm_dqzone);
1175 out:
1176 	return -ENOMEM;
1177 }
1178 
1179 void
1180 xfs_qm_exit(void)
1181 {
1182 	kmem_zone_destroy(xfs_qm_dqtrxzone);
1183 	kmem_zone_destroy(xfs_qm_dqzone);
1184 }
1185