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