Line data Source code
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 : #include "xfs_health.h"
28 :
29 : /*
30 : * Lock order:
31 : *
32 : * ip->i_lock
33 : * qi->qi_tree_lock
34 : * dquot->q_qlock (xfs_dqlock() and friends)
35 : * dquot->q_flush (xfs_dqflock() and friends)
36 : * qi->qi_lru_lock
37 : *
38 : * If two dquots need to be locked the order is user before group/project,
39 : * otherwise by the lowest id first, see xfs_dqlock2.
40 : */
41 :
42 : struct kmem_cache *xfs_dqtrx_cache;
43 : static struct kmem_cache *xfs_dquot_cache;
44 :
45 : static struct lock_class_key xfs_dquot_group_class;
46 : static struct lock_class_key xfs_dquot_project_class;
47 :
48 : /* Record observations of quota corruption with the health tracking system. */
49 : static void
50 6 : xfs_dquot_mark_sick(
51 : struct xfs_dquot *dqp)
52 : {
53 6 : struct xfs_mount *mp = dqp->q_mount;
54 :
55 6 : switch (dqp->q_type) {
56 2 : case XFS_DQTYPE_USER:
57 2 : xfs_fs_mark_sick(mp, XFS_SICK_FS_UQUOTA);
58 2 : break;
59 2 : case XFS_DQTYPE_GROUP:
60 2 : xfs_fs_mark_sick(mp, XFS_SICK_FS_GQUOTA);
61 2 : break;
62 2 : case XFS_DQTYPE_PROJ:
63 2 : xfs_fs_mark_sick(mp, XFS_SICK_FS_PQUOTA);
64 2 : break;
65 0 : default:
66 0 : ASSERT(0);
67 0 : break;
68 : }
69 6 : }
70 :
71 : /*
72 : * This is called to free all the memory associated with a dquot
73 : */
74 : void
75 16171656 : xfs_qm_dqdestroy(
76 : struct xfs_dquot *dqp)
77 : {
78 16171656 : ASSERT(list_empty(&dqp->q_lru));
79 :
80 16171656 : kmem_free(dqp->q_logitem.qli_item.li_lv_shadow);
81 16171633 : mutex_destroy(&dqp->q_qlock);
82 :
83 16171642 : XFS_STATS_DEC(dqp->q_mount, xs_qm_dquot);
84 16171657 : kmem_cache_free(xfs_dquot_cache, dqp);
85 16171699 : }
86 :
87 : /*
88 : * If default limits are in force, push them into the dquot now.
89 : * We overwrite the dquot limits only if they are zero and this
90 : * is not the root dquot.
91 : */
92 : void
93 256458281 : xfs_qm_adjust_dqlimits(
94 : struct xfs_dquot *dq)
95 : {
96 256458281 : struct xfs_mount *mp = dq->q_mount;
97 256458281 : struct xfs_quotainfo *q = mp->m_quotainfo;
98 256458281 : struct xfs_def_quota *defq;
99 256458281 : int prealloc = 0;
100 :
101 256458281 : ASSERT(dq->q_id);
102 256458281 : defq = xfs_get_defquota(q, xfs_dquot_type(dq));
103 :
104 256454608 : if (!dq->q_blk.softlimit) {
105 256022323 : dq->q_blk.softlimit = defq->blk.soft;
106 256022323 : prealloc = 1;
107 : }
108 256454608 : if (!dq->q_blk.hardlimit) {
109 256008315 : dq->q_blk.hardlimit = defq->blk.hard;
110 256008315 : prealloc = 1;
111 : }
112 256454608 : if (!dq->q_ino.softlimit)
113 256040182 : dq->q_ino.softlimit = defq->ino.soft;
114 256454608 : if (!dq->q_ino.hardlimit)
115 256040124 : dq->q_ino.hardlimit = defq->ino.hard;
116 256454608 : if (!dq->q_rtb.softlimit)
117 256454958 : dq->q_rtb.softlimit = defq->rtb.soft;
118 256454608 : if (!dq->q_rtb.hardlimit)
119 256454724 : dq->q_rtb.hardlimit = defq->rtb.hard;
120 :
121 256454608 : if (prealloc)
122 256021057 : xfs_dquot_set_prealloc_limits(dq);
123 256449409 : }
124 :
125 : /* Set the expiration time of a quota's grace period. */
126 : time64_t
127 222 : xfs_dquot_set_timeout(
128 : struct xfs_mount *mp,
129 : time64_t timeout)
130 : {
131 906 : struct xfs_quotainfo *qi = mp->m_quotainfo;
132 :
133 906 : return clamp_t(time64_t, timeout, qi->qi_expiry_min,
134 : qi->qi_expiry_max);
135 : }
136 :
137 : /* Set the length of the default grace period. */
138 : time64_t
139 16896 : xfs_dquot_set_grace_period(
140 : time64_t grace)
141 : {
142 16896 : return clamp_t(time64_t, grace, XFS_DQ_GRACE_MIN, XFS_DQ_GRACE_MAX);
143 : }
144 :
145 : /*
146 : * Determine if this quota counter is over either limit and set the quota
147 : * timers as appropriate.
148 : */
149 : static inline void
150 769484020 : xfs_qm_adjust_res_timer(
151 : struct xfs_mount *mp,
152 : struct xfs_dquot_res *res,
153 : struct xfs_quota_limits *qlim)
154 : {
155 769484020 : ASSERT(res->hardlimit == 0 || res->softlimit <= res->hardlimit);
156 :
157 769484020 : if ((res->softlimit && res->count > res->softlimit) ||
158 769456643 : (res->hardlimit && res->count > res->hardlimit)) {
159 28373 : if (res->timer == 0)
160 684 : res->timer = xfs_dquot_set_timeout(mp,
161 684 : ktime_get_real_seconds() + qlim->time);
162 : } else {
163 769455647 : res->timer = 0;
164 : }
165 769484020 : }
166 :
167 : /*
168 : * Check the limits and timers of a dquot and start or reset timers
169 : * if necessary.
170 : * This gets called even when quota enforcement is OFF, which makes our
171 : * life a little less complicated. (We just don't reject any quota
172 : * reservations in that case, when enforcement is off).
173 : * We also return 0 as the values of the timers in Q_GETQUOTA calls, when
174 : * enforcement's off.
175 : * In contrast, warnings are a little different in that they don't
176 : * 'automatically' get started when limits get exceeded. They do
177 : * get reset to zero, however, when we find the count to be under
178 : * the soft limit (they are only ever set non-zero via userspace).
179 : */
180 : void
181 256509405 : xfs_qm_adjust_dqtimers(
182 : struct xfs_dquot *dq)
183 : {
184 256509405 : struct xfs_mount *mp = dq->q_mount;
185 256509405 : struct xfs_quotainfo *qi = mp->m_quotainfo;
186 256509405 : struct xfs_def_quota *defq;
187 :
188 256509405 : ASSERT(dq->q_id);
189 256509405 : defq = xfs_get_defquota(qi, xfs_dquot_type(dq));
190 :
191 256500948 : xfs_qm_adjust_res_timer(dq->q_mount, &dq->q_blk, &defq->blk);
192 256497971 : xfs_qm_adjust_res_timer(dq->q_mount, &dq->q_ino, &defq->ino);
193 256499584 : xfs_qm_adjust_res_timer(dq->q_mount, &dq->q_rtb, &defq->rtb);
194 256500536 : }
195 :
196 : /*
197 : * initialize a buffer full of dquots and log the whole thing
198 : */
199 : STATIC void
200 2927324 : xfs_qm_init_dquot_blk(
201 : struct xfs_trans *tp,
202 : struct xfs_mount *mp,
203 : xfs_dqid_t id,
204 : xfs_dqtype_t type,
205 : struct xfs_buf *bp)
206 : {
207 2927324 : struct xfs_quotainfo *q = mp->m_quotainfo;
208 2927324 : struct xfs_dqblk *d;
209 2927324 : xfs_dqid_t curid;
210 2927324 : unsigned int qflag;
211 2927324 : unsigned int blftype;
212 2927324 : int i;
213 :
214 2927324 : ASSERT(tp);
215 2927324 : ASSERT(xfs_buf_islocked(bp));
216 :
217 2927324 : switch (type) {
218 : case XFS_DQTYPE_USER:
219 : qflag = XFS_UQUOTA_CHKD;
220 : blftype = XFS_BLF_UDQUOT_BUF;
221 : break;
222 61343 : case XFS_DQTYPE_PROJ:
223 61343 : qflag = XFS_PQUOTA_CHKD;
224 61343 : blftype = XFS_BLF_PDQUOT_BUF;
225 61343 : break;
226 1430648 : case XFS_DQTYPE_GROUP:
227 1430648 : qflag = XFS_GQUOTA_CHKD;
228 1430648 : blftype = XFS_BLF_GDQUOT_BUF;
229 1430648 : break;
230 0 : default:
231 0 : ASSERT(0);
232 0 : return;
233 : }
234 :
235 2927324 : d = bp->b_addr;
236 :
237 : /*
238 : * ID of the first dquot in the block - id's are zero based.
239 : */
240 2927324 : curid = id - (id % q->qi_dqperchunk);
241 2927324 : memset(d, 0, BBTOB(q->qi_dqchunklen));
242 90728804 : for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++) {
243 87801556 : d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
244 87801556 : d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
245 87801556 : d->dd_diskdq.d_id = cpu_to_be32(curid);
246 87801556 : d->dd_diskdq.d_type = type;
247 87801556 : if (curid > 0 && xfs_has_bigtime(mp))
248 87516615 : d->dd_diskdq.d_type |= XFS_DQTYPE_BIGTIME;
249 87801556 : if (xfs_has_crc(mp)) {
250 87798556 : uuid_copy(&d->dd_uuid, &mp->m_sb.sb_meta_uuid);
251 87796576 : xfs_update_cksum((char *)d, sizeof(struct xfs_dqblk),
252 : XFS_DQUOT_CRC_OFF);
253 : }
254 : }
255 :
256 2927248 : xfs_trans_dquot_buf(tp, bp, blftype);
257 :
258 : /*
259 : * quotacheck uses delayed writes to update all the dquots on disk in an
260 : * efficient manner instead of logging the individual dquot changes as
261 : * they are made. However if we log the buffer allocated here and crash
262 : * after quotacheck while the logged initialisation is still in the
263 : * active region of the log, log recovery can replay the dquot buffer
264 : * initialisation over the top of the checked dquots and corrupt quota
265 : * accounting.
266 : *
267 : * To avoid this problem, quotacheck cannot log the initialised buffer.
268 : * We must still dirty the buffer and write it back before the
269 : * allocation transaction clears the log. Therefore, mark the buffer as
270 : * ordered instead of logging it directly. This is safe for quotacheck
271 : * because it detects and repairs allocated but initialized dquot blocks
272 : * in the quota inodes.
273 : */
274 2926720 : if (!(mp->m_qflags & qflag))
275 37029 : xfs_trans_ordered_buf(tp, bp);
276 : else
277 2889691 : xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1);
278 : }
279 :
280 : /*
281 : * Initialize the dynamic speculative preallocation thresholds. The lo/hi
282 : * watermarks correspond to the soft and hard limits by default. If a soft limit
283 : * is not specified, we use 95% of the hard limit.
284 : */
285 : void
286 271994346 : xfs_dquot_set_prealloc_limits(struct xfs_dquot *dqp)
287 : {
288 271994346 : uint64_t space;
289 :
290 271994346 : dqp->q_prealloc_hi_wmark = dqp->q_blk.hardlimit;
291 271994346 : dqp->q_prealloc_lo_wmark = dqp->q_blk.softlimit;
292 271994346 : if (!dqp->q_prealloc_lo_wmark) {
293 271959973 : dqp->q_prealloc_lo_wmark = dqp->q_prealloc_hi_wmark;
294 271959973 : do_div(dqp->q_prealloc_lo_wmark, 100);
295 271959973 : dqp->q_prealloc_lo_wmark *= 95;
296 : }
297 :
298 271994346 : space = dqp->q_prealloc_hi_wmark;
299 :
300 271994346 : do_div(space, 100);
301 271994346 : dqp->q_low_space[XFS_QLOWSP_1_PCNT] = space;
302 271994346 : dqp->q_low_space[XFS_QLOWSP_3_PCNT] = space * 3;
303 271994346 : dqp->q_low_space[XFS_QLOWSP_5_PCNT] = space * 5;
304 271994346 : }
305 :
306 : /*
307 : * Ensure that the given in-core dquot has a buffer on disk backing it, and
308 : * return the buffer locked and held. This is called when the bmapi finds a
309 : * hole.
310 : */
311 : STATIC int
312 2955101 : xfs_dquot_disk_alloc(
313 : struct xfs_dquot *dqp,
314 : struct xfs_buf **bpp)
315 : {
316 2955101 : struct xfs_bmbt_irec map;
317 2955101 : struct xfs_trans *tp;
318 2955101 : struct xfs_mount *mp = dqp->q_mount;
319 2955101 : struct xfs_buf *bp;
320 2955101 : xfs_dqtype_t qtype = xfs_dquot_type(dqp);
321 2955101 : struct xfs_inode *quotip = xfs_quota_inode(mp, qtype);
322 2955051 : int nmaps = 1;
323 2955051 : int error;
324 :
325 2955051 : trace_xfs_dqalloc(dqp);
326 :
327 2955001 : error = xfs_trans_alloc(mp, &M_RES(mp)->tr_qm_dqalloc,
328 2955001 : XFS_QM_DQALLOC_SPACE_RES(mp), 0, 0, &tp);
329 2954697 : if (error)
330 : return error;
331 :
332 2928182 : xfs_ilock(quotip, XFS_ILOCK_EXCL);
333 2928774 : xfs_trans_ijoin(tp, quotip, 0);
334 :
335 2928851 : if (!xfs_this_quota_on(dqp->q_mount, qtype)) {
336 : /*
337 : * Return if this type of quotas is turned off while we didn't
338 : * have an inode lock
339 : */
340 0 : error = -ESRCH;
341 0 : goto err_cancel;
342 : }
343 :
344 2928851 : error = xfs_iext_count_may_overflow(quotip, XFS_DATA_FORK,
345 : XFS_IEXT_ADD_NOSPLIT_CNT);
346 2928748 : if (error == -EFBIG)
347 5 : error = xfs_iext_count_upgrade(tp, quotip,
348 : XFS_IEXT_ADD_NOSPLIT_CNT);
349 2928748 : if (error)
350 5 : goto err_cancel;
351 :
352 : /* Create the block mapping. */
353 2928743 : error = xfs_bmapi_write(tp, quotip, dqp->q_fileoffset,
354 : XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA, 0, &map,
355 : &nmaps);
356 2928669 : if (error)
357 1606 : goto err_cancel;
358 :
359 2927063 : ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB);
360 2927063 : ASSERT(nmaps == 1);
361 2927063 : ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
362 : (map.br_startblock != HOLESTARTBLOCK));
363 :
364 : /*
365 : * Keep track of the blkno to save a lookup later
366 : */
367 2927063 : dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
368 :
369 : /* now we can just get the buffer (there's nothing to read yet) */
370 2926857 : error = xfs_trans_get_buf(tp, mp->m_ddev_targp, dqp->q_blkno,
371 2926857 : mp->m_quotainfo->qi_dqchunklen, 0, &bp);
372 2926954 : if (error)
373 0 : goto err_cancel;
374 2926954 : bp->b_ops = &xfs_dquot_buf_ops;
375 :
376 : /*
377 : * Make a chunk of dquots out of this buffer and log
378 : * the entire thing.
379 : */
380 2926954 : xfs_qm_init_dquot_blk(tp, mp, dqp->q_id, qtype, bp);
381 2926762 : xfs_buf_set_ref(bp, XFS_DQUOT_REF);
382 :
383 : /*
384 : * Hold the buffer and join it to the dfops so that we'll still own
385 : * the buffer when we return to the caller. The buffer disposal on
386 : * error must be paid attention to very carefully, as it has been
387 : * broken since commit efa092f3d4c6 "[XFS] Fixes a bug in the quota
388 : * code when allocating a new dquot record" in 2005, and the later
389 : * conversion to xfs_defer_ops in commit 310a75a3c6c747 failed to keep
390 : * the buffer locked across the _defer_finish call. We can now do
391 : * this correctly with xfs_defer_bjoin.
392 : *
393 : * Above, we allocated a disk block for the dquot information and used
394 : * get_buf to initialize the dquot. If the _defer_finish fails, the old
395 : * transaction is gone but the new buffer is not joined or held to any
396 : * transaction, so we must _buf_relse it.
397 : *
398 : * If everything succeeds, the caller of this function is returned a
399 : * buffer that is locked and held to the transaction. The caller
400 : * is responsible for unlocking any buffer passed back, either
401 : * manually or by committing the transaction. On error, the buffer is
402 : * released and not passed back.
403 : *
404 : * Keep the quota inode ILOCKed until after the transaction commit to
405 : * maintain the atomicity of bmap/rmap updates.
406 : */
407 2926761 : xfs_trans_bhold(tp, bp);
408 2926413 : error = xfs_trans_commit(tp);
409 2927081 : xfs_iunlock(quotip, XFS_ILOCK_EXCL);
410 2927086 : if (error) {
411 221 : xfs_buf_relse(bp);
412 221 : return error;
413 : }
414 :
415 2926865 : *bpp = bp;
416 2926865 : return 0;
417 :
418 1611 : err_cancel:
419 1611 : xfs_trans_cancel(tp);
420 1611 : xfs_iunlock(quotip, XFS_ILOCK_EXCL);
421 1611 : return error;
422 : }
423 :
424 : /*
425 : * Read in the in-core dquot's on-disk metadata and return the buffer.
426 : * Returns ENOENT to signal a hole.
427 : */
428 : STATIC int
429 16171293 : xfs_dquot_disk_read(
430 : struct xfs_mount *mp,
431 : struct xfs_dquot *dqp,
432 : struct xfs_buf **bpp)
433 : {
434 16171293 : struct xfs_bmbt_irec map;
435 16171293 : struct xfs_buf *bp;
436 16171293 : xfs_dqtype_t qtype = xfs_dquot_type(dqp);
437 16171293 : struct xfs_inode *quotip = xfs_quota_inode(mp, qtype);
438 16171047 : uint lock_mode;
439 16171047 : int nmaps = 1;
440 16171047 : int error;
441 :
442 16171047 : lock_mode = xfs_ilock_data_map_shared(quotip);
443 16171184 : if (!xfs_this_quota_on(mp, qtype)) {
444 : /*
445 : * Return if this type of quotas is turned off while we
446 : * didn't have the quota inode lock.
447 : */
448 0 : xfs_iunlock(quotip, lock_mode);
449 0 : return -ESRCH;
450 : }
451 :
452 : /*
453 : * Find the block map; no allocations yet
454 : */
455 16171184 : error = xfs_bmapi_read(quotip, dqp->q_fileoffset,
456 : XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0);
457 16170978 : xfs_iunlock(quotip, lock_mode);
458 16171202 : if (error)
459 : return error;
460 :
461 16170588 : ASSERT(nmaps == 1);
462 16170588 : ASSERT(map.br_blockcount >= 1);
463 16170588 : ASSERT(map.br_startblock != DELAYSTARTBLOCK);
464 16170588 : if (map.br_startblock == HOLESTARTBLOCK)
465 : return -ENOENT;
466 :
467 12989784 : trace_xfs_dqtobp_read(dqp);
468 :
469 : /*
470 : * store the blkno etc so that we don't have to do the
471 : * mapping all the time
472 : */
473 12989708 : dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
474 :
475 12989716 : error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
476 12989716 : mp->m_quotainfo->qi_dqchunklen, 0, &bp,
477 : &xfs_dquot_buf_ops);
478 12989846 : if (xfs_metadata_is_sick(error))
479 6 : xfs_dquot_mark_sick(dqp);
480 12989846 : if (error) {
481 5718 : ASSERT(bp == NULL);
482 5718 : return error;
483 : }
484 :
485 12984128 : ASSERT(xfs_buf_islocked(bp));
486 12984128 : xfs_buf_set_ref(bp, XFS_DQUOT_REF);
487 12984059 : *bpp = bp;
488 :
489 12984059 : return 0;
490 : }
491 :
492 : /* Allocate and initialize everything we need for an incore dquot. */
493 : STATIC struct xfs_dquot *
494 16171673 : xfs_dquot_alloc(
495 : struct xfs_mount *mp,
496 : xfs_dqid_t id,
497 : xfs_dqtype_t type)
498 : {
499 16171673 : struct xfs_dquot *dqp;
500 :
501 16171673 : dqp = kmem_cache_zalloc(xfs_dquot_cache, GFP_KERNEL | __GFP_NOFAIL);
502 :
503 16171657 : dqp->q_type = type;
504 16171657 : dqp->q_id = id;
505 16171657 : dqp->q_mount = mp;
506 16171657 : INIT_LIST_HEAD(&dqp->q_lru);
507 16171657 : mutex_init(&dqp->q_qlock);
508 16171569 : init_waitqueue_head(&dqp->q_pinwait);
509 16171555 : dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk;
510 : /*
511 : * Offset of dquot in the (fixed sized) dquot chunk.
512 : */
513 16171555 : dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) *
514 : sizeof(struct xfs_dqblk);
515 :
516 : /*
517 : * Because we want to use a counting completion, complete
518 : * the flush completion once to allow a single access to
519 : * the flush completion without blocking.
520 : */
521 16171555 : init_completion(&dqp->q_flush);
522 16171459 : complete(&dqp->q_flush);
523 :
524 : /*
525 : * Make sure group quotas have a different lock class than user
526 : * quotas.
527 : */
528 16171402 : switch (type) {
529 : case XFS_DQTYPE_USER:
530 : /* uses the default lock class */
531 : break;
532 : case XFS_DQTYPE_GROUP:
533 : lockdep_set_class(&dqp->q_qlock, &xfs_dquot_group_class);
534 : break;
535 : case XFS_DQTYPE_PROJ:
536 : lockdep_set_class(&dqp->q_qlock, &xfs_dquot_project_class);
537 : break;
538 0 : default:
539 0 : ASSERT(0);
540 0 : break;
541 : }
542 :
543 16171402 : xfs_qm_dquot_logitem_init(dqp);
544 :
545 16171279 : XFS_STATS_INC(mp, xs_qm_dquot);
546 16171357 : return dqp;
547 : }
548 :
549 : /* Check the ondisk dquot's id and type match what the incore dquot expects. */
550 : static bool
551 15910807 : xfs_dquot_check_type(
552 : struct xfs_dquot *dqp,
553 : struct xfs_disk_dquot *ddqp)
554 : {
555 15910807 : uint8_t ddqp_type;
556 15910807 : uint8_t dqp_type;
557 :
558 15910807 : ddqp_type = ddqp->d_type & XFS_DQTYPE_REC_MASK;
559 15910807 : dqp_type = xfs_dquot_type(dqp);
560 :
561 15910807 : if (be32_to_cpu(ddqp->d_id) != dqp->q_id)
562 : return false;
563 :
564 : /*
565 : * V5 filesystems always expect an exact type match. V4 filesystems
566 : * expect an exact match for user dquots and for non-root group and
567 : * project dquots.
568 : */
569 15910947 : if (xfs_has_crc(dqp->q_mount) ||
570 85 : dqp_type == XFS_DQTYPE_USER || dqp->q_id != 0)
571 15910862 : return ddqp_type == dqp_type;
572 :
573 : /*
574 : * V4 filesystems support either group or project quotas, but not both
575 : * at the same time. The non-user quota file can be switched between
576 : * group and project quota uses depending on the mount options, which
577 : * means that we can encounter the other type when we try to load quota
578 : * defaults. Quotacheck will soon reset the entire quota file
579 : * (including the root dquot) anyway, but don't log scary corruption
580 : * reports to dmesg.
581 : */
582 85 : return ddqp_type == XFS_DQTYPE_GROUP || ddqp_type == XFS_DQTYPE_PROJ;
583 : }
584 :
585 : /* Copy the in-core quota fields in from the on-disk buffer. */
586 : STATIC int
587 15910904 : xfs_dquot_from_disk(
588 : struct xfs_dquot *dqp,
589 : struct xfs_buf *bp)
590 : {
591 15910904 : struct xfs_disk_dquot *ddqp = bp->b_addr + dqp->q_bufoffset;
592 :
593 : /*
594 : * Ensure that we got the type and ID we were looking for.
595 : * Everything else was checked by the dquot buffer verifier.
596 : */
597 15910904 : if (!xfs_dquot_check_type(dqp, ddqp)) {
598 0 : xfs_alert_tag(bp->b_mount, XFS_PTAG_VERIFIER_ERROR,
599 : "Metadata corruption detected at %pS, quota %u",
600 : __this_address, dqp->q_id);
601 0 : xfs_alert(bp->b_mount, "Unmount and run xfs_repair");
602 0 : xfs_dquot_mark_sick(dqp);
603 0 : return -EFSCORRUPTED;
604 : }
605 :
606 : /* copy everything from disk dquot to the incore dquot */
607 15910904 : dqp->q_type = ddqp->d_type;
608 15910904 : dqp->q_blk.hardlimit = be64_to_cpu(ddqp->d_blk_hardlimit);
609 15910904 : dqp->q_blk.softlimit = be64_to_cpu(ddqp->d_blk_softlimit);
610 15910904 : dqp->q_ino.hardlimit = be64_to_cpu(ddqp->d_ino_hardlimit);
611 15910904 : dqp->q_ino.softlimit = be64_to_cpu(ddqp->d_ino_softlimit);
612 15910904 : dqp->q_rtb.hardlimit = be64_to_cpu(ddqp->d_rtb_hardlimit);
613 15910904 : dqp->q_rtb.softlimit = be64_to_cpu(ddqp->d_rtb_softlimit);
614 :
615 15910904 : dqp->q_blk.count = be64_to_cpu(ddqp->d_bcount);
616 15910904 : dqp->q_ino.count = be64_to_cpu(ddqp->d_icount);
617 15910904 : dqp->q_rtb.count = be64_to_cpu(ddqp->d_rtbcount);
618 :
619 15910904 : dqp->q_blk.timer = xfs_dquot_from_disk_ts(ddqp, ddqp->d_btimer);
620 15910824 : dqp->q_ino.timer = xfs_dquot_from_disk_ts(ddqp, ddqp->d_itimer);
621 15910863 : dqp->q_rtb.timer = xfs_dquot_from_disk_ts(ddqp, ddqp->d_rtbtimer);
622 :
623 : /*
624 : * Reservation counters are defined as reservation plus current usage
625 : * to avoid having to add every time.
626 : */
627 15910947 : dqp->q_blk.reserved = dqp->q_blk.count;
628 15910947 : dqp->q_ino.reserved = dqp->q_ino.count;
629 15910947 : dqp->q_rtb.reserved = dqp->q_rtb.count;
630 :
631 : /* initialize the dquot speculative prealloc thresholds */
632 15910947 : xfs_dquot_set_prealloc_limits(dqp);
633 15910947 : return 0;
634 : }
635 :
636 : /* Copy the in-core quota fields into the on-disk buffer. */
637 : void
638 4005211771 : xfs_dquot_to_disk(
639 : struct xfs_disk_dquot *ddqp,
640 : struct xfs_dquot *dqp)
641 : {
642 4005211771 : ddqp->d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
643 4005211771 : ddqp->d_version = XFS_DQUOT_VERSION;
644 4005211771 : ddqp->d_type = dqp->q_type;
645 4005211771 : ddqp->d_id = cpu_to_be32(dqp->q_id);
646 4005211771 : ddqp->d_pad0 = 0;
647 4005211771 : ddqp->d_pad = 0;
648 :
649 4005211771 : ddqp->d_blk_hardlimit = cpu_to_be64(dqp->q_blk.hardlimit);
650 4005211771 : ddqp->d_blk_softlimit = cpu_to_be64(dqp->q_blk.softlimit);
651 4005211771 : ddqp->d_ino_hardlimit = cpu_to_be64(dqp->q_ino.hardlimit);
652 4005211771 : ddqp->d_ino_softlimit = cpu_to_be64(dqp->q_ino.softlimit);
653 4005211771 : ddqp->d_rtb_hardlimit = cpu_to_be64(dqp->q_rtb.hardlimit);
654 4005211771 : ddqp->d_rtb_softlimit = cpu_to_be64(dqp->q_rtb.softlimit);
655 :
656 4005211771 : ddqp->d_bcount = cpu_to_be64(dqp->q_blk.count);
657 4005211771 : ddqp->d_icount = cpu_to_be64(dqp->q_ino.count);
658 4005211771 : ddqp->d_rtbcount = cpu_to_be64(dqp->q_rtb.count);
659 :
660 4005211771 : ddqp->d_bwarns = 0;
661 4005211771 : ddqp->d_iwarns = 0;
662 4005211771 : ddqp->d_rtbwarns = 0;
663 :
664 4005211771 : ddqp->d_btimer = xfs_dquot_to_disk_ts(dqp, dqp->q_blk.timer);
665 4005202465 : ddqp->d_itimer = xfs_dquot_to_disk_ts(dqp, dqp->q_ino.timer);
666 4005217513 : ddqp->d_rtbtimer = xfs_dquot_to_disk_ts(dqp, dqp->q_rtb.timer);
667 4005220283 : }
668 :
669 : /*
670 : * Read in the ondisk dquot using dqtobp() then copy it to an incore version,
671 : * and release the buffer immediately. If @can_alloc is true, fill any
672 : * holes in the on-disk metadata.
673 : */
674 : static int
675 16171672 : xfs_qm_dqread(
676 : struct xfs_mount *mp,
677 : xfs_dqid_t id,
678 : xfs_dqtype_t type,
679 : bool can_alloc,
680 : struct xfs_dquot **dqpp)
681 : {
682 16171672 : struct xfs_dquot *dqp;
683 16171672 : struct xfs_buf *bp;
684 16171672 : int error;
685 :
686 16171672 : dqp = xfs_dquot_alloc(mp, id, type);
687 16171422 : trace_xfs_dqread(dqp);
688 :
689 : /* Try to read the buffer, allocating if necessary. */
690 16171363 : error = xfs_dquot_disk_read(mp, dqp, &bp);
691 16171037 : if (error == -ENOENT && can_alloc)
692 2955180 : error = xfs_dquot_disk_alloc(dqp, &bp);
693 16171077 : if (error)
694 260209 : goto err;
695 :
696 : /*
697 : * At this point we should have a clean locked buffer. Copy the data
698 : * to the incore dquot and release the buffer since the incore dquot
699 : * has its own locking protocol so we needn't tie up the buffer any
700 : * further.
701 : */
702 15910868 : ASSERT(xfs_buf_islocked(bp));
703 15910868 : error = xfs_dquot_from_disk(dqp, bp);
704 15910679 : xfs_buf_relse(bp);
705 15911313 : if (error)
706 0 : goto err;
707 :
708 15911313 : *dqpp = dqp;
709 15911313 : return error;
710 :
711 260209 : err:
712 260209 : trace_xfs_dqread_fail(dqp);
713 260185 : xfs_qm_dqdestroy(dqp);
714 260244 : *dqpp = NULL;
715 260244 : return error;
716 : }
717 :
718 : /*
719 : * Advance to the next id in the current chunk, or if at the
720 : * end of the chunk, skip ahead to first id in next allocated chunk
721 : * using the SEEK_DATA interface.
722 : */
723 : static int
724 12935383 : xfs_dq_get_next_id(
725 : struct xfs_mount *mp,
726 : xfs_dqtype_t type,
727 : xfs_dqid_t *id)
728 : {
729 12935383 : struct xfs_inode *quotip = xfs_quota_inode(mp, type);
730 12935383 : xfs_dqid_t next_id = *id + 1; /* simple advance */
731 12935383 : uint lock_flags;
732 12935383 : struct xfs_bmbt_irec got;
733 12935383 : struct xfs_iext_cursor cur;
734 12935383 : xfs_fsblock_t start;
735 12935383 : int error = 0;
736 :
737 : /* If we'd wrap past the max ID, stop */
738 12935383 : if (next_id < *id)
739 : return -ENOENT;
740 :
741 : /* If new ID is within the current chunk, advancing it sufficed */
742 12935343 : if (next_id % mp->m_quotainfo->qi_dqperchunk) {
743 12503882 : *id = next_id;
744 12503882 : return 0;
745 : }
746 :
747 : /* Nope, next_id is now past the current chunk, so find the next one */
748 431461 : start = (xfs_fsblock_t)next_id / mp->m_quotainfo->qi_dqperchunk;
749 :
750 431461 : lock_flags = xfs_ilock_data_map_shared(quotip);
751 431461 : error = xfs_iread_extents(NULL, quotip, XFS_DATA_FORK);
752 431461 : if (error)
753 : return error;
754 :
755 431461 : if (xfs_iext_lookup_extent(quotip, "ip->i_df, start, &cur, &got)) {
756 : /* contiguous chunk, bump startoff for the id calculation */
757 428508 : if (got.br_startoff < start)
758 641 : got.br_startoff = start;
759 428508 : *id = got.br_startoff * mp->m_quotainfo->qi_dqperchunk;
760 : } else {
761 : error = -ENOENT;
762 : }
763 :
764 431461 : xfs_iunlock(quotip, lock_flags);
765 :
766 431461 : return error;
767 : }
768 :
769 : /*
770 : * Look up the dquot in the in-core cache. If found, the dquot is returned
771 : * locked and ready to go.
772 : */
773 : static struct xfs_dquot *
774 155204669 : xfs_qm_dqget_cache_lookup(
775 : struct xfs_mount *mp,
776 : struct xfs_quotainfo *qi,
777 : struct radix_tree_root *tree,
778 : xfs_dqid_t id)
779 : {
780 155204673 : struct xfs_dquot *dqp;
781 :
782 155204673 : restart:
783 155204673 : mutex_lock(&qi->qi_tree_lock);
784 155320305 : dqp = radix_tree_lookup(tree, id);
785 155320305 : if (!dqp) {
786 15959644 : mutex_unlock(&qi->qi_tree_lock);
787 15959643 : XFS_STATS_INC(mp, xs_qm_dqcachemisses);
788 15959641 : return NULL;
789 : }
790 :
791 139360661 : xfs_dqlock(dqp);
792 139360661 : if (dqp->q_flags & XFS_DQFLAG_FREEING) {
793 4 : xfs_dqunlock(dqp);
794 4 : mutex_unlock(&qi->qi_tree_lock);
795 4 : trace_xfs_dqget_freeing(dqp);
796 4 : delay(1);
797 4 : goto restart;
798 : }
799 :
800 139360657 : dqp->q_nrefs++;
801 139360657 : mutex_unlock(&qi->qi_tree_lock);
802 :
803 139360640 : trace_xfs_dqget_hit(dqp);
804 139360624 : XFS_STATS_INC(mp, xs_qm_dqcachehits);
805 139360598 : return dqp;
806 : }
807 :
808 : /*
809 : * Try to insert a new dquot into the in-core cache. If an error occurs the
810 : * caller should throw away the dquot and start over. Otherwise, the dquot
811 : * is returned locked (and held by the cache) as if there had been a cache
812 : * hit.
813 : */
814 : static int
815 15739538 : xfs_qm_dqget_cache_insert(
816 : struct xfs_mount *mp,
817 : struct xfs_quotainfo *qi,
818 : struct radix_tree_root *tree,
819 : xfs_dqid_t id,
820 : struct xfs_dquot *dqp)
821 : {
822 15739538 : int error;
823 :
824 15739538 : mutex_lock(&qi->qi_tree_lock);
825 15739957 : error = radix_tree_insert(tree, id, dqp);
826 15739957 : if (unlikely(error)) {
827 : /* Duplicate found! Caller must try again. */
828 385 : mutex_unlock(&qi->qi_tree_lock);
829 385 : trace_xfs_dqget_dup(dqp);
830 385 : return error;
831 : }
832 :
833 : /* Return a locked dquot to the caller, with a reference taken. */
834 15739572 : xfs_dqlock(dqp);
835 15739572 : dqp->q_nrefs = 1;
836 :
837 15739572 : qi->qi_dquots++;
838 15739572 : mutex_unlock(&qi->qi_tree_lock);
839 :
840 15739572 : return 0;
841 : }
842 :
843 : /* Check our input parameters. */
844 : static int
845 155411186 : xfs_qm_dqget_checks(
846 : struct xfs_mount *mp,
847 : xfs_dqtype_t type)
848 : {
849 155411186 : switch (type) {
850 66990468 : case XFS_DQTYPE_USER:
851 66990468 : if (!XFS_IS_UQUOTA_ON(mp))
852 1008 : return -ESRCH;
853 : return 0;
854 66745374 : case XFS_DQTYPE_GROUP:
855 66745374 : if (!XFS_IS_GQUOTA_ON(mp))
856 2822 : return -ESRCH;
857 : return 0;
858 21675344 : case XFS_DQTYPE_PROJ:
859 21675344 : if (!XFS_IS_PQUOTA_ON(mp))
860 1095 : return -ESRCH;
861 : return 0;
862 : default:
863 : WARN_ON_ONCE(0);
864 : return -EINVAL;
865 : }
866 : }
867 :
868 : /*
869 : * Given the file system, id, and type (UDQUOT/GDQUOT/PDQUOT), return a
870 : * locked dquot, doing an allocation (if requested) as needed.
871 : */
872 : int
873 102482299 : xfs_qm_dqget(
874 : struct xfs_mount *mp,
875 : xfs_dqid_t id,
876 : xfs_dqtype_t type,
877 : bool can_alloc,
878 : struct xfs_dquot **O_dqpp)
879 : {
880 102482299 : struct xfs_quotainfo *qi = mp->m_quotainfo;
881 102482299 : struct radix_tree_root *tree = xfs_dquot_tree(qi, type);
882 102406592 : struct xfs_dquot *dqp;
883 102406592 : int error;
884 :
885 102406592 : error = xfs_qm_dqget_checks(mp, type);
886 102406592 : if (error)
887 : return error;
888 :
889 102404066 : restart:
890 102404304 : dqp = xfs_qm_dqget_cache_lookup(mp, qi, tree, id);
891 102550022 : if (dqp) {
892 87711016 : *O_dqpp = dqp;
893 87711016 : return 0;
894 : }
895 :
896 14839006 : error = xfs_qm_dqread(mp, id, type, can_alloc, &dqp);
897 14838817 : if (error)
898 215587 : return error;
899 :
900 14623230 : error = xfs_qm_dqget_cache_insert(mp, qi, tree, id, dqp);
901 14623392 : if (error) {
902 : /*
903 : * Duplicate found. Just throw away the new dquot and start
904 : * over.
905 : */
906 238 : xfs_qm_dqdestroy(dqp);
907 238 : XFS_STATS_INC(mp, xs_qm_dquot_dups);
908 238 : goto restart;
909 : }
910 :
911 14623154 : trace_xfs_dqget_miss(dqp);
912 14623150 : *O_dqpp = dqp;
913 14623150 : return 0;
914 : }
915 :
916 : /*
917 : * Given a dquot id and type, read and initialize a dquot from the on-disk
918 : * metadata. This function is only for use during quota initialization so
919 : * it ignores the dquot cache assuming that the dquot shrinker isn't set up.
920 : * The caller is responsible for _qm_dqdestroy'ing the returned dquot.
921 : */
922 : int
923 214453 : xfs_qm_dqget_uncached(
924 : struct xfs_mount *mp,
925 : xfs_dqid_t id,
926 : xfs_dqtype_t type,
927 : struct xfs_dquot **dqpp)
928 : {
929 214453 : int error;
930 :
931 214453 : error = xfs_qm_dqget_checks(mp, type);
932 214453 : if (error)
933 : return error;
934 :
935 212054 : return xfs_qm_dqread(mp, id, type, 0, dqpp);
936 : }
937 :
938 : /* Return the quota id for a given inode and type. */
939 : xfs_dqid_t
940 741859425 : xfs_qm_id_for_quotatype(
941 : struct xfs_inode *ip,
942 : xfs_dqtype_t type)
943 : {
944 741859425 : switch (type) {
945 : case XFS_DQTYPE_USER:
946 247289332 : return i_uid_read(VFS_I(ip));
947 : case XFS_DQTYPE_GROUP:
948 247285490 : return i_gid_read(VFS_I(ip));
949 247284603 : case XFS_DQTYPE_PROJ:
950 247284603 : return ip->i_projid;
951 : }
952 0 : ASSERT(0);
953 0 : return 0;
954 : }
955 :
956 : /*
957 : * Return the dquot for a given inode and type. If @can_alloc is true, then
958 : * allocate blocks if needed. The inode's ILOCK must be held and it must not
959 : * have already had an inode attached.
960 : */
961 : int
962 52769475 : xfs_qm_dqget_inode(
963 : struct xfs_inode *ip,
964 : xfs_dqtype_t type,
965 : bool can_alloc,
966 : struct xfs_dquot **O_dqpp)
967 : {
968 52769475 : struct xfs_mount *mp = ip->i_mount;
969 52769475 : struct xfs_quotainfo *qi = mp->m_quotainfo;
970 52769475 : struct radix_tree_root *tree = xfs_dquot_tree(qi, type);
971 52769112 : struct xfs_dquot *dqp;
972 52769112 : xfs_dqid_t id;
973 52769112 : int error;
974 :
975 52769112 : error = xfs_qm_dqget_checks(mp, type);
976 52769112 : if (error)
977 : return error;
978 :
979 52768886 : ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
980 105537512 : ASSERT(xfs_inode_dquot(ip, type) == NULL);
981 :
982 52768756 : id = xfs_qm_id_for_quotatype(ip, type);
983 :
984 52768376 : restart:
985 52768376 : dqp = xfs_qm_dqget_cache_lookup(mp, qi, tree, id);
986 52770169 : if (dqp) {
987 51649537 : *O_dqpp = dqp;
988 51649537 : return 0;
989 : }
990 :
991 : /*
992 : * Dquot cache miss. We don't want to keep the inode lock across
993 : * a (potential) disk read. Also we don't want to deal with the lock
994 : * ordering between quotainode and this inode. OTOH, dropping the inode
995 : * lock here means dealing with a chown that can happen before
996 : * we re-acquire the lock.
997 : */
998 1120632 : xfs_iunlock(ip, XFS_ILOCK_EXCL);
999 1120619 : error = xfs_qm_dqread(mp, id, type, can_alloc, &dqp);
1000 1120600 : xfs_ilock(ip, XFS_ILOCK_EXCL);
1001 1120585 : if (error)
1002 4071 : return error;
1003 :
1004 : /*
1005 : * A dquot could be attached to this inode by now, since we had
1006 : * dropped the ilock.
1007 : */
1008 1116514 : if (xfs_this_quota_on(mp, type)) {
1009 1116514 : struct xfs_dquot *dqp1;
1010 :
1011 1116514 : dqp1 = xfs_inode_dquot(ip, type);
1012 1116514 : if (dqp1) {
1013 0 : xfs_qm_dqdestroy(dqp);
1014 0 : dqp = dqp1;
1015 0 : xfs_dqlock(dqp);
1016 0 : goto dqret;
1017 : }
1018 : } else {
1019 : /* inode stays locked on return */
1020 0 : xfs_qm_dqdestroy(dqp);
1021 0 : return -ESRCH;
1022 : }
1023 :
1024 1116514 : error = xfs_qm_dqget_cache_insert(mp, qi, tree, id, dqp);
1025 1116565 : if (error) {
1026 : /*
1027 : * Duplicate found. Just throw away the new dquot and start
1028 : * over.
1029 : */
1030 147 : xfs_qm_dqdestroy(dqp);
1031 147 : XFS_STATS_INC(mp, xs_qm_dquot_dups);
1032 147 : goto restart;
1033 : }
1034 :
1035 1116418 : dqret:
1036 1116418 : ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1037 1116418 : trace_xfs_dqget_miss(dqp);
1038 1116418 : *O_dqpp = dqp;
1039 1116418 : return 0;
1040 : }
1041 :
1042 : /*
1043 : * Starting at @id and progressing upwards, look for an initialized incore
1044 : * dquot, lock it, and return it.
1045 : */
1046 : int
1047 416323 : xfs_qm_dqget_next(
1048 : struct xfs_mount *mp,
1049 : xfs_dqid_t id,
1050 : xfs_dqtype_t type,
1051 : struct xfs_dquot **dqpp)
1052 : {
1053 416323 : struct xfs_dquot *dqp;
1054 416323 : int error = 0;
1055 :
1056 416323 : *dqpp = NULL;
1057 13351706 : for (; !error; error = xfs_dq_get_next_id(mp, type, &id)) {
1058 13348713 : error = xfs_qm_dqget(mp, id, type, false, &dqp);
1059 13348709 : if (error == -ENOENT)
1060 160668 : continue;
1061 13188041 : else if (error != 0)
1062 : break;
1063 :
1064 13187957 : if (!XFS_IS_DQUOT_UNINITIALIZED(dqp)) {
1065 413242 : *dqpp = dqp;
1066 413242 : return 0;
1067 : }
1068 :
1069 12774715 : xfs_qm_dqput(dqp);
1070 : }
1071 :
1072 : return error;
1073 : }
1074 :
1075 : /*
1076 : * Release a reference to the dquot (decrement ref-count) and unlock it.
1077 : *
1078 : * If there is a group quota attached to this dquot, carefully release that
1079 : * too without tripping over deadlocks'n'stuff.
1080 : */
1081 : void
1082 576205481 : xfs_qm_dqput(
1083 : struct xfs_dquot *dqp)
1084 : {
1085 576205481 : ASSERT(dqp->q_nrefs > 0);
1086 576205481 : ASSERT(XFS_DQ_IS_LOCKED(dqp));
1087 :
1088 576254014 : trace_xfs_dqput(dqp);
1089 :
1090 576231712 : if (--dqp->q_nrefs == 0) {
1091 29397476 : struct xfs_quotainfo *qi = dqp->q_mount->m_quotainfo;
1092 29397476 : trace_xfs_dqput_free(dqp);
1093 :
1094 29397456 : if (list_lru_add(&qi->qi_lru, &dqp->q_lru))
1095 15787215 : XFS_STATS_INC(dqp->q_mount, xs_qm_dquot_unused);
1096 : }
1097 576231830 : xfs_dqunlock(dqp);
1098 576566080 : }
1099 :
1100 : /*
1101 : * Release a dquot. Flush it if dirty, then dqput() it.
1102 : * dquot must not be locked.
1103 : */
1104 : void
1105 847894491 : xfs_qm_dqrele(
1106 : struct xfs_dquot *dqp)
1107 : {
1108 847894491 : if (!dqp)
1109 : return;
1110 :
1111 494670082 : trace_xfs_dqrele(dqp);
1112 :
1113 494482242 : xfs_dqlock(dqp);
1114 : /*
1115 : * We don't care to flush it if the dquot is dirty here.
1116 : * That will create stutters that we want to avoid.
1117 : * Instead we do a delayed write when we try to reclaim
1118 : * a dirty dquot. Also xfs_sync will take part of the burden...
1119 : */
1120 494700536 : xfs_qm_dqput(dqp);
1121 : }
1122 :
1123 : /*
1124 : * This is the dquot flushing I/O completion routine. It is called
1125 : * from interrupt level when the buffer containing the dquot is
1126 : * flushed to disk. It is responsible for removing the dquot logitem
1127 : * from the AIL if it has not been re-logged, and unlocking the dquot's
1128 : * flush lock. This behavior is very similar to that of inodes..
1129 : */
1130 : static void
1131 16581762 : xfs_qm_dqflush_done(
1132 : struct xfs_log_item *lip)
1133 : {
1134 16581762 : struct xfs_dq_logitem *qip = (struct xfs_dq_logitem *)lip;
1135 16581762 : struct xfs_dquot *dqp = qip->qli_dquot;
1136 16581762 : struct xfs_ail *ailp = lip->li_ailp;
1137 16581762 : xfs_lsn_t tail_lsn;
1138 :
1139 : /*
1140 : * We only want to pull the item from the AIL if its
1141 : * location in the log has not changed since we started the flush.
1142 : * Thus, we only bother if the dquot's lsn has
1143 : * not changed. First we check the lsn outside the lock
1144 : * since it's cheaper, and then we recheck while
1145 : * holding the lock before removing the dquot from the AIL.
1146 : */
1147 33170043 : if (test_bit(XFS_LI_IN_AIL, &lip->li_flags) &&
1148 16560140 : ((lip->li_lsn == qip->qli_flush_lsn) ||
1149 0 : test_bit(XFS_LI_FAILED, &lip->li_flags))) {
1150 :
1151 16547102 : spin_lock(&ailp->ail_lock);
1152 16547102 : xfs_clear_li_failed(lip);
1153 16547102 : if (lip->li_lsn == qip->qli_flush_lsn) {
1154 : /* xfs_ail_update_finish() drops the AIL lock */
1155 16547102 : tail_lsn = xfs_ail_delete_one(ailp, lip);
1156 16547102 : xfs_ail_update_finish(ailp, tail_lsn);
1157 : } else {
1158 0 : spin_unlock(&ailp->ail_lock);
1159 : }
1160 : }
1161 :
1162 : /*
1163 : * Release the dq's flush lock since we're done with it.
1164 : */
1165 16581762 : xfs_dqfunlock(dqp);
1166 16581762 : }
1167 :
1168 : void
1169 14679302 : xfs_buf_dquot_iodone(
1170 : struct xfs_buf *bp)
1171 : {
1172 14679302 : struct xfs_log_item *lip, *n;
1173 :
1174 31261064 : list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) {
1175 16581762 : list_del_init(&lip->li_bio_list);
1176 16581762 : xfs_qm_dqflush_done(lip);
1177 : }
1178 14679302 : }
1179 :
1180 : void
1181 107 : xfs_buf_dquot_io_fail(
1182 : struct xfs_buf *bp)
1183 : {
1184 107 : struct xfs_log_item *lip;
1185 :
1186 107 : spin_lock(&bp->b_mount->m_ail->ail_lock);
1187 214 : list_for_each_entry(lip, &bp->b_li_list, li_bio_list)
1188 107 : xfs_set_li_failed(lip, bp);
1189 107 : spin_unlock(&bp->b_mount->m_ail->ail_lock);
1190 107 : }
1191 :
1192 : /* Check incore dquot for errors before we flush. */
1193 : static xfs_failaddr_t
1194 16581762 : xfs_qm_dqflush_check(
1195 : struct xfs_dquot *dqp)
1196 : {
1197 16581762 : xfs_dqtype_t type = xfs_dquot_type(dqp);
1198 :
1199 16581762 : if (type != XFS_DQTYPE_USER &&
1200 16581762 : type != XFS_DQTYPE_GROUP &&
1201 : type != XFS_DQTYPE_PROJ)
1202 0 : return __this_address;
1203 :
1204 16581762 : if (dqp->q_id == 0)
1205 : return NULL;
1206 :
1207 16399708 : if (dqp->q_blk.softlimit && dqp->q_blk.count > dqp->q_blk.softlimit &&
1208 408 : !dqp->q_blk.timer)
1209 0 : return __this_address;
1210 :
1211 16399708 : if (dqp->q_ino.softlimit && dqp->q_ino.count > dqp->q_ino.softlimit &&
1212 498 : !dqp->q_ino.timer)
1213 0 : return __this_address;
1214 :
1215 16399708 : if (dqp->q_rtb.softlimit && dqp->q_rtb.count > dqp->q_rtb.softlimit &&
1216 0 : !dqp->q_rtb.timer)
1217 0 : return __this_address;
1218 :
1219 : /* bigtime flag should never be set on root dquots */
1220 16399708 : if (dqp->q_type & XFS_DQTYPE_BIGTIME) {
1221 16387018 : if (!xfs_has_bigtime(dqp->q_mount))
1222 0 : return __this_address;
1223 : if (dqp->q_id == 0)
1224 0 : return __this_address;
1225 : }
1226 :
1227 : return NULL;
1228 : }
1229 :
1230 : /*
1231 : * Write a modified dquot to disk.
1232 : * The dquot must be locked and the flush lock too taken by caller.
1233 : * The flush lock will not be unlocked until the dquot reaches the disk,
1234 : * but the dquot is free to be unlocked and modified by the caller
1235 : * in the interim. Dquot is still locked on return. This behavior is
1236 : * identical to that of inodes.
1237 : */
1238 : int
1239 18251818 : xfs_qm_dqflush(
1240 : struct xfs_dquot *dqp,
1241 : struct xfs_buf **bpp)
1242 : {
1243 18251818 : struct xfs_mount *mp = dqp->q_mount;
1244 18251818 : struct xfs_log_item *lip = &dqp->q_logitem.qli_item;
1245 18251818 : struct xfs_buf *bp;
1246 18251818 : struct xfs_dqblk *dqblk;
1247 18251818 : xfs_failaddr_t fa;
1248 18251818 : int error;
1249 :
1250 18251818 : ASSERT(XFS_DQ_IS_LOCKED(dqp));
1251 18251818 : ASSERT(!completion_done(&dqp->q_flush));
1252 :
1253 18251818 : trace_xfs_dqflush(dqp);
1254 :
1255 18251818 : *bpp = NULL;
1256 :
1257 18251818 : xfs_qm_dqunpin_wait(dqp);
1258 :
1259 : /*
1260 : * Get the buffer containing the on-disk dquot
1261 : */
1262 18251818 : error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
1263 18251818 : mp->m_quotainfo->qi_dqchunklen, XBF_TRYLOCK,
1264 : &bp, &xfs_dquot_buf_ops);
1265 18251818 : if (error == -EAGAIN)
1266 4569 : goto out_unlock;
1267 18247249 : if (xfs_metadata_is_sick(error))
1268 0 : xfs_dquot_mark_sick(dqp);
1269 18247249 : if (error)
1270 1665487 : goto out_abort;
1271 :
1272 16581762 : fa = xfs_qm_dqflush_check(dqp);
1273 16581762 : if (fa) {
1274 0 : xfs_alert(mp, "corrupt dquot ID 0x%x in memory at %pS",
1275 : dqp->q_id, fa);
1276 0 : xfs_buf_relse(bp);
1277 0 : xfs_dquot_mark_sick(dqp);
1278 0 : error = -EFSCORRUPTED;
1279 0 : goto out_abort;
1280 : }
1281 :
1282 : /* Flush the incore dquot to the ondisk buffer. */
1283 16581762 : dqblk = bp->b_addr + dqp->q_bufoffset;
1284 16581762 : xfs_dquot_to_disk(&dqblk->dd_diskdq, dqp);
1285 :
1286 : /*
1287 : * Clear the dirty field and remember the flush lsn for later use.
1288 : */
1289 16581762 : dqp->q_flags &= ~XFS_DQFLAG_DIRTY;
1290 :
1291 16581762 : xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn,
1292 : &dqp->q_logitem.qli_item.li_lsn);
1293 :
1294 : /*
1295 : * copy the lsn into the on-disk dquot now while we have the in memory
1296 : * dquot here. This can't be done later in the write verifier as we
1297 : * can't get access to the log item at that point in time.
1298 : *
1299 : * We also calculate the CRC here so that the on-disk dquot in the
1300 : * buffer always has a valid CRC. This ensures there is no possibility
1301 : * of a dquot without an up-to-date CRC getting to disk.
1302 : */
1303 16581762 : if (xfs_has_crc(mp)) {
1304 16581680 : dqblk->dd_lsn = cpu_to_be64(dqp->q_logitem.qli_item.li_lsn);
1305 16581680 : xfs_update_cksum((char *)dqblk, sizeof(struct xfs_dqblk),
1306 : XFS_DQUOT_CRC_OFF);
1307 : }
1308 :
1309 : /*
1310 : * Attach the dquot to the buffer so that we can remove this dquot from
1311 : * the AIL and release the flush lock once the dquot is synced to disk.
1312 : */
1313 16581762 : bp->b_flags |= _XBF_DQUOTS;
1314 16581762 : list_add_tail(&dqp->q_logitem.qli_item.li_bio_list, &bp->b_li_list);
1315 :
1316 : /*
1317 : * If the buffer is pinned then push on the log so we won't
1318 : * get stuck waiting in the write for too long.
1319 : */
1320 16581762 : if (xfs_buf_ispinned(bp)) {
1321 22631 : trace_xfs_dqflush_force(dqp);
1322 22631 : xfs_log_force(mp, 0);
1323 : }
1324 :
1325 16581762 : trace_xfs_dqflush_done(dqp);
1326 16581762 : *bpp = bp;
1327 16581762 : return 0;
1328 :
1329 1665487 : out_abort:
1330 1665487 : dqp->q_flags &= ~XFS_DQFLAG_DIRTY;
1331 1665487 : xfs_trans_ail_delete(lip, 0);
1332 1665487 : xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1333 1670056 : out_unlock:
1334 1670056 : xfs_dqfunlock(dqp);
1335 1670056 : return error;
1336 : }
1337 :
1338 : /*
1339 : * Lock two xfs_dquot structures.
1340 : *
1341 : * To avoid deadlocks we always lock the quota structure with
1342 : * the lowerd id first.
1343 : */
1344 : void
1345 17969083 : xfs_dqlock2(
1346 : struct xfs_dquot *d1,
1347 : struct xfs_dquot *d2)
1348 : {
1349 17969083 : if (d1 && d2) {
1350 17969083 : ASSERT(d1 != d2);
1351 17969083 : if (d1->q_id > d2->q_id) {
1352 14871863 : mutex_lock(&d2->q_qlock);
1353 14954430 : mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED);
1354 : } else {
1355 3097220 : mutex_lock(&d1->q_qlock);
1356 3097290 : mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED);
1357 : }
1358 0 : } else if (d1) {
1359 0 : mutex_lock(&d1->q_qlock);
1360 0 : } else if (d2) {
1361 0 : mutex_lock(&d2->q_qlock);
1362 : }
1363 18051743 : }
1364 :
1365 : static int
1366 8799 : xfs_dqtrx_cmp(
1367 : const void *a,
1368 : const void *b)
1369 : {
1370 8799 : const struct xfs_dqtrx *qa = a;
1371 8799 : const struct xfs_dqtrx *qb = b;
1372 :
1373 8799 : if (qa->qt_dquot->q_id > qb->qt_dquot->q_id)
1374 : return 1;
1375 5520 : if (qa->qt_dquot->q_id < qb->qt_dquot->q_id)
1376 5520 : return -1;
1377 : return 0;
1378 : }
1379 :
1380 : void
1381 2933 : xfs_dqlockn(
1382 : struct xfs_dqtrx *q)
1383 : {
1384 2933 : unsigned int i;
1385 :
1386 : /* Sort in order of dquot id, do not allow duplicates */
1387 11732 : for (i = 0; i < XFS_QM_TRANS_MAXDQS && q[i].qt_dquot != NULL; i++) {
1388 : unsigned int j;
1389 :
1390 17598 : for (j = 0; j < i; j++)
1391 8799 : ASSERT(q[i].qt_dquot != q[j].qt_dquot);
1392 : }
1393 2933 : if (i == 0)
1394 : return;
1395 :
1396 2933 : sort(q, i, sizeof(struct xfs_dqtrx), xfs_dqtrx_cmp, NULL);
1397 :
1398 2933 : mutex_lock(&q[0].qt_dquot->q_qlock);
1399 11732 : for (i = 1; i < XFS_QM_TRANS_MAXDQS && q[i].qt_dquot != NULL; i++)
1400 5866 : mutex_lock_nested(&q[i].qt_dquot->q_qlock, XFS_QLOCK_NESTED);
1401 : }
1402 :
1403 : int __init
1404 50 : xfs_qm_init(void)
1405 : {
1406 50 : xfs_dquot_cache = kmem_cache_create("xfs_dquot",
1407 : sizeof(struct xfs_dquot),
1408 : 0, 0, NULL);
1409 50 : if (!xfs_dquot_cache)
1410 0 : goto out;
1411 :
1412 50 : xfs_dqtrx_cache = kmem_cache_create("xfs_dqtrx",
1413 : sizeof(struct xfs_dquot_acct),
1414 : 0, 0, NULL);
1415 50 : if (!xfs_dqtrx_cache)
1416 0 : goto out_free_dquot_cache;
1417 :
1418 : return 0;
1419 :
1420 : out_free_dquot_cache:
1421 0 : kmem_cache_destroy(xfs_dquot_cache);
1422 : out:
1423 : return -ENOMEM;
1424 : }
1425 :
1426 : void
1427 49 : xfs_qm_exit(void)
1428 : {
1429 49 : kmem_cache_destroy(xfs_dqtrx_cache);
1430 49 : kmem_cache_destroy(xfs_dquot_cache);
1431 49 : }
1432 :
1433 : /*
1434 : * Iterate every dquot of a particular type. The caller must ensure that the
1435 : * particular quota type is active. iter_fn can return negative error codes,
1436 : * or -ECANCELED to indicate that it wants to stop iterating.
1437 : */
1438 : int
1439 144310 : xfs_qm_dqiterate(
1440 : struct xfs_mount *mp,
1441 : xfs_dqtype_t type,
1442 : xfs_qm_dqiterate_fn iter_fn,
1443 : void *priv)
1444 : {
1445 144310 : struct xfs_dquot *dq;
1446 144310 : xfs_dqid_t id = 0;
1447 144310 : int error;
1448 :
1449 144310 : do {
1450 144310 : error = xfs_qm_dqget_next(mp, id, type, &dq);
1451 144303 : if (error == -ENOENT)
1452 : return 0;
1453 144303 : if (error)
1454 0 : return error;
1455 :
1456 144303 : error = iter_fn(dq, type, priv);
1457 144309 : id = dq->q_id;
1458 144309 : xfs_qm_dqput(dq);
1459 144310 : } while (error == 0 && id != 0);
1460 :
1461 : return error;
1462 : }
|
