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 32787870 : xfs_qm_dqdestroy(
76 : struct xfs_dquot *dqp)
77 : {
78 32787870 : ASSERT(list_empty(&dqp->q_lru));
79 :
80 32787870 : kmem_free(dqp->q_logitem.qli_item.li_lv_shadow);
81 32787802 : mutex_destroy(&dqp->q_qlock);
82 :
83 32787831 : XFS_STATS_DEC(dqp->q_mount, xs_qm_dquot);
84 32787891 : kmem_cache_free(xfs_dquot_cache, dqp);
85 32787974 : }
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 571985930 : xfs_qm_adjust_dqlimits(
94 : struct xfs_dquot *dq)
95 : {
96 571985930 : struct xfs_mount *mp = dq->q_mount;
97 571985930 : struct xfs_quotainfo *q = mp->m_quotainfo;
98 571985930 : struct xfs_def_quota *defq;
99 571985930 : int prealloc = 0;
100 :
101 571985930 : ASSERT(dq->q_id);
102 571985930 : defq = xfs_get_defquota(q, xfs_dquot_type(dq));
103 :
104 571979445 : if (!dq->q_blk.softlimit) {
105 571002819 : dq->q_blk.softlimit = defq->blk.soft;
106 571002819 : prealloc = 1;
107 : }
108 571979445 : if (!dq->q_blk.hardlimit) {
109 570965964 : dq->q_blk.hardlimit = defq->blk.hard;
110 570965964 : prealloc = 1;
111 : }
112 571979445 : if (!dq->q_ino.softlimit)
113 571043770 : dq->q_ino.softlimit = defq->ino.soft;
114 571979445 : if (!dq->q_ino.hardlimit)
115 571043951 : dq->q_ino.hardlimit = defq->ino.hard;
116 571979445 : if (!dq->q_rtb.softlimit)
117 571964130 : dq->q_rtb.softlimit = defq->rtb.soft;
118 571979445 : if (!dq->q_rtb.hardlimit)
119 571964409 : dq->q_rtb.hardlimit = defq->rtb.hard;
120 :
121 571979445 : if (prealloc)
122 571002806 : xfs_dquot_set_prealloc_limits(dq);
123 571967555 : }
124 :
125 : /* Set the expiration time of a quota's grace period. */
126 : time64_t
127 412 : xfs_dquot_set_timeout(
128 : struct xfs_mount *mp,
129 : time64_t timeout)
130 : {
131 1678 : struct xfs_quotainfo *qi = mp->m_quotainfo;
132 :
133 1678 : 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 30127 : xfs_dquot_set_grace_period(
140 : time64_t grace)
141 : {
142 30127 : 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 1716145399 : xfs_qm_adjust_res_timer(
151 : struct xfs_mount *mp,
152 : struct xfs_dquot_res *res,
153 : struct xfs_quota_limits *qlim)
154 : {
155 1716145399 : ASSERT(res->hardlimit == 0 || res->softlimit <= res->hardlimit);
156 :
157 1716145399 : if ((res->softlimit && res->count > res->softlimit) ||
158 1716091959 : (res->hardlimit && res->count > res->hardlimit)) {
159 55388 : if (res->timer == 0)
160 1266 : res->timer = xfs_dquot_set_timeout(mp,
161 1266 : ktime_get_real_seconds() + qlim->time);
162 : } else {
163 1716090011 : res->timer = 0;
164 : }
165 1716145399 : }
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 572079491 : xfs_qm_adjust_dqtimers(
182 : struct xfs_dquot *dq)
183 : {
184 572079491 : struct xfs_mount *mp = dq->q_mount;
185 572079491 : struct xfs_quotainfo *qi = mp->m_quotainfo;
186 572079491 : struct xfs_def_quota *defq;
187 :
188 572079491 : ASSERT(dq->q_id);
189 572079491 : defq = xfs_get_defquota(qi, xfs_dquot_type(dq));
190 :
191 572062627 : xfs_qm_adjust_res_timer(dq->q_mount, &dq->q_blk, &defq->blk);
192 572057381 : xfs_qm_adjust_res_timer(dq->q_mount, &dq->q_ino, &defq->ino);
193 572059083 : xfs_qm_adjust_res_timer(dq->q_mount, &dq->q_rtb, &defq->rtb);
194 572061568 : }
195 :
196 : /*
197 : * initialize a buffer full of dquots and log the whole thing
198 : */
199 : STATIC void
200 6018971 : 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 6018971 : struct xfs_quotainfo *q = mp->m_quotainfo;
208 6018971 : struct xfs_dqblk *d;
209 6018971 : xfs_dqid_t curid;
210 6018971 : unsigned int qflag;
211 6018971 : unsigned int blftype;
212 6018971 : int i;
213 :
214 6018971 : ASSERT(tp);
215 6018971 : ASSERT(xfs_buf_islocked(bp));
216 :
217 6018971 : switch (type) {
218 : case XFS_DQTYPE_USER:
219 : qflag = XFS_UQUOTA_CHKD;
220 : blftype = XFS_BLF_UDQUOT_BUF;
221 : break;
222 133746 : case XFS_DQTYPE_PROJ:
223 133746 : qflag = XFS_PQUOTA_CHKD;
224 133746 : blftype = XFS_BLF_PDQUOT_BUF;
225 133746 : break;
226 2938610 : case XFS_DQTYPE_GROUP:
227 2938610 : qflag = XFS_GQUOTA_CHKD;
228 2938610 : blftype = XFS_BLF_GDQUOT_BUF;
229 2938610 : break;
230 0 : default:
231 0 : ASSERT(0);
232 0 : return;
233 : }
234 :
235 6018971 : d = bp->b_addr;
236 :
237 : /*
238 : * ID of the first dquot in the block - id's are zero based.
239 : */
240 6018971 : curid = id - (id % q->qi_dqperchunk);
241 6018971 : memset(d, 0, BBTOB(q->qi_dqchunklen));
242 186555429 : for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++) {
243 180536592 : d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
244 180536592 : d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
245 180536592 : d->dd_diskdq.d_id = cpu_to_be32(curid);
246 180536592 : d->dd_diskdq.d_type = type;
247 180536592 : if (curid > 0 && xfs_has_bigtime(mp))
248 180085011 : d->dd_diskdq.d_type |= XFS_DQTYPE_BIGTIME;
249 180536592 : if (xfs_has_crc(mp)) {
250 180530418 : uuid_copy(&d->dd_uuid, &mp->m_sb.sb_meta_uuid);
251 180526539 : xfs_update_cksum((char *)d, sizeof(struct xfs_dqblk),
252 : XFS_DQUOT_CRC_OFF);
253 : }
254 : }
255 :
256 6018837 : 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 6017858 : if (!(mp->m_qflags & qflag))
275 76127 : xfs_trans_ordered_buf(tp, bp);
276 : else
277 5941731 : 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 603355936 : xfs_dquot_set_prealloc_limits(struct xfs_dquot *dqp)
287 : {
288 603355936 : uint64_t space;
289 :
290 603355936 : dqp->q_prealloc_hi_wmark = dqp->q_blk.hardlimit;
291 603355936 : dqp->q_prealloc_lo_wmark = dqp->q_blk.softlimit;
292 603355936 : if (!dqp->q_prealloc_lo_wmark) {
293 603291536 : dqp->q_prealloc_lo_wmark = dqp->q_prealloc_hi_wmark;
294 603291536 : do_div(dqp->q_prealloc_lo_wmark, 100);
295 603291536 : dqp->q_prealloc_lo_wmark *= 95;
296 : }
297 :
298 603355936 : space = dqp->q_prealloc_hi_wmark;
299 :
300 603355936 : do_div(space, 100);
301 603355936 : dqp->q_low_space[XFS_QLOWSP_1_PCNT] = space;
302 603355936 : dqp->q_low_space[XFS_QLOWSP_3_PCNT] = space * 3;
303 603355936 : dqp->q_low_space[XFS_QLOWSP_5_PCNT] = space * 5;
304 603355936 : }
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 6066146 : xfs_dquot_disk_alloc(
313 : struct xfs_dquot *dqp,
314 : struct xfs_buf **bpp)
315 : {
316 6066146 : struct xfs_bmbt_irec map;
317 6066146 : struct xfs_trans *tp;
318 6066146 : struct xfs_mount *mp = dqp->q_mount;
319 6066146 : struct xfs_buf *bp;
320 6066146 : xfs_dqtype_t qtype = xfs_dquot_type(dqp);
321 6066146 : struct xfs_inode *quotip = xfs_quota_inode(mp, qtype);
322 6066065 : int nmaps = 1;
323 6066065 : int error;
324 :
325 6066065 : trace_xfs_dqalloc(dqp);
326 :
327 6065959 : error = xfs_trans_alloc(mp, &M_RES(mp)->tr_qm_dqalloc,
328 6065959 : XFS_QM_DQALLOC_SPACE_RES(mp), 0, 0, &tp);
329 6065292 : if (error)
330 : return error;
331 :
332 6018407 : xfs_ilock(quotip, XFS_ILOCK_EXCL);
333 6020079 : xfs_trans_ijoin(tp, quotip, 0);
334 :
335 6020281 : 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 6020281 : error = xfs_iext_count_may_overflow(quotip, XFS_DATA_FORK,
345 : XFS_IEXT_ADD_NOSPLIT_CNT);
346 6020126 : if (error == -EFBIG)
347 10 : error = xfs_iext_count_upgrade(tp, quotip,
348 : XFS_IEXT_ADD_NOSPLIT_CNT);
349 6020126 : if (error)
350 10 : goto err_cancel;
351 :
352 : /* Create the block mapping. */
353 6020116 : error = xfs_bmapi_write(tp, quotip, dqp->q_fileoffset,
354 : XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA, 0, &map,
355 : &nmaps);
356 6020016 : if (error)
357 1471 : goto err_cancel;
358 :
359 6018545 : ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB);
360 6018545 : ASSERT(nmaps == 1);
361 6018545 : 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 6018545 : 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 6017948 : error = xfs_trans_get_buf(tp, mp->m_ddev_targp, dqp->q_blkno,
371 6017948 : mp->m_quotainfo->qi_dqchunklen, 0, &bp);
372 6018171 : if (error)
373 0 : goto err_cancel;
374 6018171 : 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 6018171 : xfs_qm_init_dquot_blk(tp, mp, dqp->q_id, qtype, bp);
381 6017797 : 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 6017760 : xfs_trans_bhold(tp, bp);
408 6017264 : error = xfs_trans_commit(tp);
409 6018580 : xfs_iunlock(quotip, XFS_ILOCK_EXCL);
410 6018548 : if (error) {
411 268 : xfs_buf_relse(bp);
412 268 : return error;
413 : }
414 :
415 6018280 : *bpp = bp;
416 6018280 : return 0;
417 :
418 1481 : err_cancel:
419 1481 : xfs_trans_cancel(tp);
420 1481 : xfs_iunlock(quotip, XFS_ILOCK_EXCL);
421 1481 : 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 32787309 : xfs_dquot_disk_read(
430 : struct xfs_mount *mp,
431 : struct xfs_dquot *dqp,
432 : struct xfs_buf **bpp)
433 : {
434 32787309 : struct xfs_bmbt_irec map;
435 32787309 : struct xfs_buf *bp;
436 32787309 : xfs_dqtype_t qtype = xfs_dquot_type(dqp);
437 32787309 : struct xfs_inode *quotip = xfs_quota_inode(mp, qtype);
438 32786909 : uint lock_mode;
439 32786909 : int nmaps = 1;
440 32786909 : int error;
441 :
442 32786909 : lock_mode = xfs_ilock_data_map_shared(quotip);
443 32787056 : 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 32787056 : error = xfs_bmapi_read(quotip, dqp->q_fileoffset,
456 : XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0);
457 32786730 : xfs_iunlock(quotip, lock_mode);
458 32786851 : if (error)
459 : return error;
460 :
461 32786155 : ASSERT(nmaps == 1);
462 32786155 : ASSERT(map.br_blockcount >= 1);
463 32786155 : ASSERT(map.br_startblock != DELAYSTARTBLOCK);
464 32786155 : if (map.br_startblock == HOLESTARTBLOCK)
465 : return -ENOENT;
466 :
467 26224294 : 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 26224125 : dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
474 :
475 26224120 : error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
476 26224120 : mp->m_quotainfo->qi_dqchunklen, 0, &bp,
477 : &xfs_dquot_buf_ops);
478 26224303 : if (xfs_metadata_is_sick(error))
479 6 : xfs_dquot_mark_sick(dqp);
480 26224303 : if (error) {
481 4998 : ASSERT(bp == NULL);
482 4998 : return error;
483 : }
484 :
485 26219305 : ASSERT(xfs_buf_islocked(bp));
486 26219305 : xfs_buf_set_ref(bp, XFS_DQUOT_REF);
487 26219202 : *bpp = bp;
488 :
489 26219202 : return 0;
490 : }
491 :
492 : /* Allocate and initialize everything we need for an incore dquot. */
493 : STATIC struct xfs_dquot *
494 32787959 : xfs_dquot_alloc(
495 : struct xfs_mount *mp,
496 : xfs_dqid_t id,
497 : xfs_dqtype_t type)
498 : {
499 32787959 : struct xfs_dquot *dqp;
500 :
501 32787959 : dqp = kmem_cache_zalloc(xfs_dquot_cache, GFP_KERNEL | __GFP_NOFAIL);
502 :
503 32787910 : dqp->q_type = type;
504 32787910 : dqp->q_id = id;
505 32787910 : dqp->q_mount = mp;
506 32787910 : INIT_LIST_HEAD(&dqp->q_lru);
507 32787910 : mutex_init(&dqp->q_qlock);
508 32787775 : init_waitqueue_head(&dqp->q_pinwait);
509 32787782 : 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 32787782 : 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 32787782 : init_completion(&dqp->q_flush);
522 32787623 : complete(&dqp->q_flush);
523 :
524 : /*
525 : * Make sure group quotas have a different lock class than user
526 : * quotas.
527 : */
528 32787538 : 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 32787538 : xfs_qm_dquot_logitem_init(dqp);
544 :
545 32787278 : XFS_STATS_INC(mp, xs_qm_dquot);
546 32787426 : return dqp;
547 : }
548 :
549 : /* Check the ondisk dquot's id and type match what the incore dquot expects. */
550 : static bool
551 32237219 : xfs_dquot_check_type(
552 : struct xfs_dquot *dqp,
553 : struct xfs_disk_dquot *ddqp)
554 : {
555 32237219 : uint8_t ddqp_type;
556 32237219 : uint8_t dqp_type;
557 :
558 32237219 : ddqp_type = ddqp->d_type & XFS_DQTYPE_REC_MASK;
559 32237219 : dqp_type = xfs_dquot_type(dqp);
560 :
561 32237219 : 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 32237503 : if (xfs_has_crc(dqp->q_mount) ||
570 170 : dqp_type == XFS_DQTYPE_USER || dqp->q_id != 0)
571 32237333 : 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 : */
582 170 : 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 32237400 : xfs_dquot_from_disk(
588 : struct xfs_dquot *dqp,
589 : struct xfs_buf *bp)
590 : {
591 32237400 : 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 32237400 : 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 32237400 : dqp->q_type = ddqp->d_type;
608 32237400 : dqp->q_blk.hardlimit = be64_to_cpu(ddqp->d_blk_hardlimit);
609 32237400 : dqp->q_blk.softlimit = be64_to_cpu(ddqp->d_blk_softlimit);
610 32237400 : dqp->q_ino.hardlimit = be64_to_cpu(ddqp->d_ino_hardlimit);
611 32237400 : dqp->q_ino.softlimit = be64_to_cpu(ddqp->d_ino_softlimit);
612 32237400 : dqp->q_rtb.hardlimit = be64_to_cpu(ddqp->d_rtb_hardlimit);
613 32237400 : dqp->q_rtb.softlimit = be64_to_cpu(ddqp->d_rtb_softlimit);
614 :
615 32237400 : dqp->q_blk.count = be64_to_cpu(ddqp->d_bcount);
616 32237400 : dqp->q_ino.count = be64_to_cpu(ddqp->d_icount);
617 32237400 : dqp->q_rtb.count = be64_to_cpu(ddqp->d_rtbcount);
618 :
619 32237400 : dqp->q_blk.timer = xfs_dquot_from_disk_ts(ddqp, ddqp->d_btimer);
620 32237306 : dqp->q_ino.timer = xfs_dquot_from_disk_ts(ddqp, ddqp->d_itimer);
621 32237445 : 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 32237511 : dqp->q_blk.reserved = dqp->q_blk.count;
628 32237511 : dqp->q_ino.reserved = dqp->q_ino.count;
629 32237511 : dqp->q_rtb.reserved = dqp->q_rtb.count;
630 :
631 : /* initialize the dquot speculative prealloc thresholds */
632 32237511 : xfs_dquot_set_prealloc_limits(dqp);
633 32237511 : return 0;
634 : }
635 :
636 : /* Copy the in-core quota fields into the on-disk buffer. */
637 : void
638 7785643693 : xfs_dquot_to_disk(
639 : struct xfs_disk_dquot *ddqp,
640 : struct xfs_dquot *dqp)
641 : {
642 7785643693 : ddqp->d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
643 7785643693 : ddqp->d_version = XFS_DQUOT_VERSION;
644 7785643693 : ddqp->d_type = dqp->q_type;
645 7785643693 : ddqp->d_id = cpu_to_be32(dqp->q_id);
646 7785643693 : ddqp->d_pad0 = 0;
647 7785643693 : ddqp->d_pad = 0;
648 :
649 7785643693 : ddqp->d_blk_hardlimit = cpu_to_be64(dqp->q_blk.hardlimit);
650 7785643693 : ddqp->d_blk_softlimit = cpu_to_be64(dqp->q_blk.softlimit);
651 7785643693 : ddqp->d_ino_hardlimit = cpu_to_be64(dqp->q_ino.hardlimit);
652 7785643693 : ddqp->d_ino_softlimit = cpu_to_be64(dqp->q_ino.softlimit);
653 7785643693 : ddqp->d_rtb_hardlimit = cpu_to_be64(dqp->q_rtb.hardlimit);
654 7785643693 : ddqp->d_rtb_softlimit = cpu_to_be64(dqp->q_rtb.softlimit);
655 :
656 7785643693 : ddqp->d_bcount = cpu_to_be64(dqp->q_blk.count);
657 7785643693 : ddqp->d_icount = cpu_to_be64(dqp->q_ino.count);
658 7785643693 : ddqp->d_rtbcount = cpu_to_be64(dqp->q_rtb.count);
659 :
660 7785643693 : ddqp->d_bwarns = 0;
661 7785643693 : ddqp->d_iwarns = 0;
662 7785643693 : ddqp->d_rtbwarns = 0;
663 :
664 7785643693 : ddqp->d_btimer = xfs_dquot_to_disk_ts(dqp, dqp->q_blk.timer);
665 7785626555 : ddqp->d_itimer = xfs_dquot_to_disk_ts(dqp, dqp->q_ino.timer);
666 7785658769 : ddqp->d_rtbtimer = xfs_dquot_to_disk_ts(dqp, dqp->q_rtb.timer);
667 7785665176 : }
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 32787961 : 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 32787961 : struct xfs_dquot *dqp;
683 32787961 : struct xfs_buf *bp;
684 32787961 : int error;
685 :
686 32787961 : dqp = xfs_dquot_alloc(mp, id, type);
687 32787500 : trace_xfs_dqread(dqp);
688 :
689 : /* Try to read the buffer, allocating if necessary. */
690 32787385 : error = xfs_dquot_disk_read(mp, dqp, &bp);
691 32786285 : if (error == -ENOENT && can_alloc)
692 6066433 : error = xfs_dquot_disk_alloc(dqp, &bp);
693 32786805 : if (error)
694 549541 : 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 32237264 : ASSERT(xfs_buf_islocked(bp));
703 32237264 : error = xfs_dquot_from_disk(dqp, bp);
704 32237214 : xfs_buf_relse(bp);
705 32238119 : if (error)
706 0 : goto err;
707 :
708 32238119 : *dqpp = dqp;
709 32238119 : return error;
710 :
711 549541 : err:
712 549541 : trace_xfs_dqread_fail(dqp);
713 549473 : xfs_qm_dqdestroy(dqp);
714 549637 : *dqpp = NULL;
715 549637 : 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 27232391 : xfs_dq_get_next_id(
725 : struct xfs_mount *mp,
726 : xfs_dqtype_t type,
727 : xfs_dqid_t *id)
728 : {
729 27232391 : struct xfs_inode *quotip = xfs_quota_inode(mp, type);
730 27232391 : xfs_dqid_t next_id = *id + 1; /* simple advance */
731 27232391 : uint lock_flags;
732 27232391 : struct xfs_bmbt_irec got;
733 27232391 : struct xfs_iext_cursor cur;
734 27232391 : xfs_fsblock_t start;
735 27232391 : int error = 0;
736 :
737 : /* If we'd wrap past the max ID, stop */
738 27232391 : if (next_id < *id)
739 : return -ENOENT;
740 :
741 : /* If new ID is within the current chunk, advancing it sufficed */
742 27232318 : if (next_id % mp->m_quotainfo->qi_dqperchunk) {
743 26324320 : *id = next_id;
744 26324320 : return 0;
745 : }
746 :
747 : /* Nope, next_id is now past the current chunk, so find the next one */
748 907998 : start = (xfs_fsblock_t)next_id / mp->m_quotainfo->qi_dqperchunk;
749 :
750 907998 : lock_flags = xfs_ilock_data_map_shared(quotip);
751 907998 : error = xfs_iread_extents(NULL, quotip, XFS_DATA_FORK);
752 907998 : if (error)
753 : return error;
754 :
755 907998 : if (xfs_iext_lookup_extent(quotip, "ip->i_df, start, &cur, &got)) {
756 : /* contiguous chunk, bump startoff for the id calculation */
757 902488 : if (got.br_startoff < start)
758 1635 : got.br_startoff = start;
759 902488 : *id = got.br_startoff * mp->m_quotainfo->qi_dqperchunk;
760 : } else {
761 : error = -ENOENT;
762 : }
763 :
764 907998 : xfs_iunlock(quotip, lock_flags);
765 :
766 907998 : 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 244396457 : 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 244396490 : struct xfs_dquot *dqp;
781 :
782 244396490 : restart:
783 244396490 : mutex_lock(&qi->qi_tree_lock);
784 244525048 : dqp = radix_tree_lookup(tree, id);
785 244525048 : if (!dqp) {
786 32451052 : mutex_unlock(&qi->qi_tree_lock);
787 32451052 : XFS_STATS_INC(mp, xs_qm_dqcachemisses);
788 32451049 : return NULL;
789 : }
790 :
791 212073996 : xfs_dqlock(dqp);
792 212073996 : if (dqp->q_flags & XFS_DQFLAG_FREEING) {
793 33 : xfs_dqunlock(dqp);
794 33 : mutex_unlock(&qi->qi_tree_lock);
795 33 : trace_xfs_dqget_freeing(dqp);
796 33 : delay(1);
797 33 : goto restart;
798 : }
799 :
800 212073963 : dqp->q_nrefs++;
801 212073963 : mutex_unlock(&qi->qi_tree_lock);
802 :
803 212073943 : trace_xfs_dqget_hit(dqp);
804 212073915 : XFS_STATS_INC(mp, xs_qm_dqcachehits);
805 212073889 : 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 31979280 : 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 31979280 : int error;
823 :
824 31979280 : mutex_lock(&qi->qi_tree_lock);
825 31980072 : error = radix_tree_insert(tree, id, dqp);
826 31980072 : if (unlikely(error)) {
827 : /* Duplicate found! Caller must try again. */
828 615 : mutex_unlock(&qi->qi_tree_lock);
829 615 : trace_xfs_dqget_dup(dqp);
830 615 : return error;
831 : }
832 :
833 : /* Return a locked dquot to the caller, with a reference taken. */
834 31979457 : xfs_dqlock(dqp);
835 31979457 : dqp->q_nrefs = 1;
836 :
837 31979457 : qi->qi_dquots++;
838 31979457 : mutex_unlock(&qi->qi_tree_lock);
839 :
840 31979457 : return 0;
841 : }
842 :
843 : /* Check our input parameters. */
844 : static int
845 244744814 : xfs_qm_dqget_checks(
846 : struct xfs_mount *mp,
847 : xfs_dqtype_t type)
848 : {
849 244744814 : switch (type) {
850 109302272 : case XFS_DQTYPE_USER:
851 109302272 : if (!XFS_IS_UQUOTA_ON(mp))
852 1798 : return -ESRCH;
853 : return 0;
854 108745734 : case XFS_DQTYPE_GROUP:
855 108745734 : if (!XFS_IS_GQUOTA_ON(mp))
856 5117 : return -ESRCH;
857 : return 0;
858 26696808 : case XFS_DQTYPE_PROJ:
859 26696808 : if (!XFS_IS_PQUOTA_ON(mp))
860 1905 : 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 185523614 : 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 185523614 : struct xfs_quotainfo *qi = mp->m_quotainfo;
881 185523614 : struct radix_tree_root *tree = xfs_dquot_tree(qi, type);
882 185467952 : struct xfs_dquot *dqp;
883 185467952 : int error;
884 :
885 185467952 : error = xfs_qm_dqget_checks(mp, type);
886 185467952 : if (error)
887 : return error;
888 :
889 185463321 : restart:
890 185463652 : dqp = xfs_qm_dqget_cache_lookup(mp, qi, tree, id);
891 185600898 : if (dqp) {
892 154290717 : *O_dqpp = dqp;
893 154290717 : return 0;
894 : }
895 :
896 31310181 : error = xfs_qm_dqread(mp, id, type, can_alloc, &dqp);
897 31309736 : if (error)
898 467373 : return error;
899 :
900 30842363 : error = xfs_qm_dqget_cache_insert(mp, qi, tree, id, dqp);
901 30842717 : if (error) {
902 : /*
903 : * Duplicate found. Just throw away the new dquot and start
904 : * over.
905 : */
906 331 : xfs_qm_dqdestroy(dqp);
907 331 : XFS_STATS_INC(mp, xs_qm_dquot_dups);
908 331 : goto restart;
909 : }
910 :
911 30842386 : trace_xfs_dqget_miss(dqp);
912 30842369 : *O_dqpp = dqp;
913 30842369 : 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 341123 : 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 341123 : int error;
930 :
931 341123 : error = xfs_qm_dqget_checks(mp, type);
932 341123 : if (error)
933 : return error;
934 :
935 336934 : 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 1206972906 : xfs_qm_id_for_quotatype(
941 : struct xfs_inode *ip,
942 : xfs_dqtype_t type)
943 : {
944 1206972906 : switch (type) {
945 : case XFS_DQTYPE_USER:
946 402327402 : return i_uid_read(VFS_I(ip));
947 : case XFS_DQTYPE_GROUP:
948 402322631 : return i_gid_read(VFS_I(ip));
949 402322873 : case XFS_DQTYPE_PROJ:
950 402322873 : 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 58922831 : 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 58922831 : struct xfs_mount *mp = ip->i_mount;
969 58922831 : struct xfs_quotainfo *qi = mp->m_quotainfo;
970 58922831 : struct radix_tree_root *tree = xfs_dquot_tree(qi, type);
971 58922402 : struct xfs_dquot *dqp;
972 58922402 : xfs_dqid_t id;
973 58922402 : int error;
974 :
975 58922402 : error = xfs_qm_dqget_checks(mp, type);
976 58922402 : if (error)
977 : return error;
978 :
979 58921732 : ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
980 117843458 : ASSERT(xfs_inode_dquot(ip, type) == NULL);
981 58921729 : ASSERT(!xfs_is_metadir_inode(ip));
982 :
983 58921729 : id = xfs_qm_id_for_quotatype(ip, type);
984 :
985 58921115 : restart:
986 58921115 : dqp = xfs_qm_dqget_cache_lookup(mp, qi, tree, id);
987 58923942 : if (dqp) {
988 57783074 : *O_dqpp = dqp;
989 57783074 : return 0;
990 : }
991 :
992 : /*
993 : * Dquot cache miss. We don't want to keep the inode lock across
994 : * a (potential) disk read. Also we don't want to deal with the lock
995 : * ordering between quotainode and this inode. OTOH, dropping the inode
996 : * lock here means dealing with a chown that can happen before
997 : * we re-acquire the lock.
998 : */
999 1140868 : xfs_iunlock(ip, XFS_ILOCK_EXCL);
1000 1140857 : error = xfs_qm_dqread(mp, id, type, can_alloc, &dqp);
1001 1140856 : xfs_ilock(ip, XFS_ILOCK_EXCL);
1002 1140815 : if (error)
1003 3511 : return error;
1004 :
1005 : /*
1006 : * A dquot could be attached to this inode by now, since we had
1007 : * dropped the ilock.
1008 : */
1009 1137304 : if (xfs_this_quota_on(mp, type)) {
1010 1137304 : struct xfs_dquot *dqp1;
1011 :
1012 1137304 : dqp1 = xfs_inode_dquot(ip, type);
1013 1137304 : if (dqp1) {
1014 3 : xfs_qm_dqdestroy(dqp);
1015 3 : dqp = dqp1;
1016 3 : xfs_dqlock(dqp);
1017 3 : goto dqret;
1018 : }
1019 : } else {
1020 : /* inode stays locked on return */
1021 0 : xfs_qm_dqdestroy(dqp);
1022 0 : return -ESRCH;
1023 : }
1024 :
1025 1137301 : error = xfs_qm_dqget_cache_insert(mp, qi, tree, id, dqp);
1026 1137355 : if (error) {
1027 : /*
1028 : * Duplicate found. Just throw away the new dquot and start
1029 : * over.
1030 : */
1031 284 : xfs_qm_dqdestroy(dqp);
1032 284 : XFS_STATS_INC(mp, xs_qm_dquot_dups);
1033 284 : goto restart;
1034 : }
1035 :
1036 1137071 : dqret:
1037 1137074 : ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1038 1137074 : trace_xfs_dqget_miss(dqp);
1039 1137074 : *O_dqpp = dqp;
1040 1137074 : return 0;
1041 : }
1042 :
1043 : /*
1044 : * Starting at @id and progressing upwards, look for an initialized incore
1045 : * dquot, lock it, and return it.
1046 : */
1047 : int
1048 1013991 : xfs_qm_dqget_next(
1049 : struct xfs_mount *mp,
1050 : xfs_dqid_t id,
1051 : xfs_dqtype_t type,
1052 : struct xfs_dquot **dqpp)
1053 : {
1054 1013991 : struct xfs_dquot *dqp;
1055 1013991 : int error = 0;
1056 :
1057 1013991 : *dqpp = NULL;
1058 28246382 : for (; !error; error = xfs_dq_get_next_id(mp, type, &id)) {
1059 28240799 : error = xfs_qm_dqget(mp, id, type, false, &dqp);
1060 28240789 : if (error == -ENOENT)
1061 371793 : continue;
1062 27868996 : else if (error != 0)
1063 : break;
1064 :
1065 27868842 : if (!XFS_IS_DQUOT_UNINITIALIZED(dqp)) {
1066 1008244 : *dqpp = dqp;
1067 1008244 : return 0;
1068 : }
1069 :
1070 26860598 : xfs_qm_dqput(dqp);
1071 : }
1072 :
1073 : return error;
1074 : }
1075 :
1076 : /*
1077 : * Release a reference to the dquot (decrement ref-count) and unlock it.
1078 : *
1079 : * If there is a group quota attached to this dquot, carefully release that
1080 : * too without tripping over deadlocks'n'stuff.
1081 : */
1082 : void
1083 1012987604 : xfs_qm_dqput(
1084 : struct xfs_dquot *dqp)
1085 : {
1086 1012987604 : ASSERT(dqp->q_nrefs > 0);
1087 1012987604 : ASSERT(XFS_DQ_IS_LOCKED(dqp));
1088 :
1089 1013074771 : trace_xfs_dqput(dqp);
1090 :
1091 1013047830 : if (--dqp->q_nrefs == 0) {
1092 49512356 : struct xfs_quotainfo *qi = dqp->q_mount->m_quotainfo;
1093 49512356 : trace_xfs_dqput_free(dqp);
1094 :
1095 49512331 : if (list_lru_add(&qi->qi_lru, &dqp->q_lru))
1096 32020727 : XFS_STATS_INC(dqp->q_mount, xs_qm_dquot_unused);
1097 : }
1098 1013048123 : xfs_dqunlock(dqp);
1099 1013611396 : }
1100 :
1101 : /*
1102 : * Release a dquot. Flush it if dirty, then dqput() it.
1103 : * dquot must not be locked.
1104 : */
1105 : void
1106 984890949 : xfs_qm_dqrele(
1107 : struct xfs_dquot *dqp)
1108 : {
1109 984890949 : if (!dqp)
1110 : return;
1111 :
1112 869490422 : trace_xfs_dqrele(dqp);
1113 :
1114 869105668 : xfs_dqlock(dqp);
1115 : /*
1116 : * We don't care to flush it if the dquot is dirty here.
1117 : * That will create stutters that we want to avoid.
1118 : * Instead we do a delayed write when we try to reclaim
1119 : * a dirty dquot. Also xfs_sync will take part of the burden...
1120 : */
1121 869403947 : xfs_qm_dqput(dqp);
1122 : }
1123 :
1124 : /*
1125 : * This is the dquot flushing I/O completion routine. It is called
1126 : * from interrupt level when the buffer containing the dquot is
1127 : * flushed to disk. It is responsible for removing the dquot logitem
1128 : * from the AIL if it has not been re-logged, and unlocking the dquot's
1129 : * flush lock. This behavior is very similar to that of inodes..
1130 : */
1131 : static void
1132 34113447 : xfs_qm_dqflush_done(
1133 : struct xfs_log_item *lip)
1134 : {
1135 34113447 : struct xfs_dq_logitem *qip = (struct xfs_dq_logitem *)lip;
1136 34113447 : struct xfs_dquot *dqp = qip->qli_dquot;
1137 34113447 : struct xfs_ail *ailp = lip->li_ailp;
1138 34113447 : xfs_lsn_t tail_lsn;
1139 :
1140 : /*
1141 : * We only want to pull the item from the AIL if its
1142 : * location in the log has not changed since we started the flush.
1143 : * Thus, we only bother if the dquot's lsn has
1144 : * not changed. First we check the lsn outside the lock
1145 : * since it's cheaper, and then we recheck while
1146 : * holding the lock before removing the dquot from the AIL.
1147 : */
1148 68239001 : if (test_bit(XFS_LI_IN_AIL, &lip->li_flags) &&
1149 34061285 : ((lip->li_lsn == qip->qli_flush_lsn) ||
1150 0 : test_bit(XFS_LI_FAILED, &lip->li_flags))) {
1151 :
1152 34037071 : spin_lock(&ailp->ail_lock);
1153 34037071 : xfs_clear_li_failed(lip);
1154 34037071 : if (lip->li_lsn == qip->qli_flush_lsn) {
1155 : /* xfs_ail_update_finish() drops the AIL lock */
1156 34037071 : tail_lsn = xfs_ail_delete_one(ailp, lip);
1157 34037071 : xfs_ail_update_finish(ailp, tail_lsn);
1158 : } else {
1159 0 : spin_unlock(&ailp->ail_lock);
1160 : }
1161 : }
1162 :
1163 : /*
1164 : * Release the dq's flush lock since we're done with it.
1165 : */
1166 34113447 : xfs_dqfunlock(dqp);
1167 34113447 : }
1168 :
1169 : void
1170 29861491 : xfs_buf_dquot_iodone(
1171 : struct xfs_buf *bp)
1172 : {
1173 29861491 : struct xfs_log_item *lip, *n;
1174 :
1175 63974938 : list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) {
1176 34113447 : list_del_init(&lip->li_bio_list);
1177 34113447 : xfs_qm_dqflush_done(lip);
1178 : }
1179 29861491 : }
1180 :
1181 : void
1182 131 : xfs_buf_dquot_io_fail(
1183 : struct xfs_buf *bp)
1184 : {
1185 131 : struct xfs_log_item *lip;
1186 :
1187 131 : spin_lock(&bp->b_mount->m_ail->ail_lock);
1188 262 : list_for_each_entry(lip, &bp->b_li_list, li_bio_list)
1189 131 : xfs_set_li_failed(lip, bp);
1190 131 : spin_unlock(&bp->b_mount->m_ail->ail_lock);
1191 131 : }
1192 :
1193 : /* Check incore dquot for errors before we flush. */
1194 : static xfs_failaddr_t
1195 34113447 : xfs_qm_dqflush_check(
1196 : struct xfs_dquot *dqp)
1197 : {
1198 34113447 : xfs_dqtype_t type = xfs_dquot_type(dqp);
1199 :
1200 34113447 : if (type != XFS_DQTYPE_USER &&
1201 34113447 : type != XFS_DQTYPE_GROUP &&
1202 : type != XFS_DQTYPE_PROJ)
1203 0 : return __this_address;
1204 :
1205 34113447 : if (dqp->q_id == 0)
1206 : return NULL;
1207 :
1208 33752321 : if (dqp->q_blk.softlimit && dqp->q_blk.count > dqp->q_blk.softlimit &&
1209 741 : !dqp->q_blk.timer)
1210 0 : return __this_address;
1211 :
1212 33752321 : if (dqp->q_ino.softlimit && dqp->q_ino.count > dqp->q_ino.softlimit &&
1213 900 : !dqp->q_ino.timer)
1214 0 : return __this_address;
1215 :
1216 33752321 : if (dqp->q_rtb.softlimit && dqp->q_rtb.count > dqp->q_rtb.softlimit &&
1217 0 : !dqp->q_rtb.timer)
1218 0 : return __this_address;
1219 :
1220 : /* bigtime flag should never be set on root dquots */
1221 33752321 : if (dqp->q_type & XFS_DQTYPE_BIGTIME) {
1222 33733589 : if (!xfs_has_bigtime(dqp->q_mount))
1223 0 : return __this_address;
1224 : if (dqp->q_id == 0)
1225 0 : return __this_address;
1226 : }
1227 :
1228 : return NULL;
1229 : }
1230 :
1231 : /*
1232 : * Write a modified dquot to disk.
1233 : * The dquot must be locked and the flush lock too taken by caller.
1234 : * The flush lock will not be unlocked until the dquot reaches the disk,
1235 : * but the dquot is free to be unlocked and modified by the caller
1236 : * in the interim. Dquot is still locked on return. This behavior is
1237 : * identical to that of inodes.
1238 : */
1239 : int
1240 35710223 : xfs_qm_dqflush(
1241 : struct xfs_dquot *dqp,
1242 : struct xfs_buf **bpp)
1243 : {
1244 35710223 : struct xfs_mount *mp = dqp->q_mount;
1245 35710223 : struct xfs_log_item *lip = &dqp->q_logitem.qli_item;
1246 35710223 : struct xfs_buf *bp;
1247 35710223 : struct xfs_dqblk *dqblk;
1248 35710223 : xfs_failaddr_t fa;
1249 35710223 : int error;
1250 :
1251 35710223 : ASSERT(XFS_DQ_IS_LOCKED(dqp));
1252 35710223 : ASSERT(!completion_done(&dqp->q_flush));
1253 :
1254 35710223 : trace_xfs_dqflush(dqp);
1255 :
1256 35710223 : *bpp = NULL;
1257 :
1258 35710223 : xfs_qm_dqunpin_wait(dqp);
1259 :
1260 : /*
1261 : * Get the buffer containing the on-disk dquot
1262 : */
1263 35710223 : error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
1264 35710223 : mp->m_quotainfo->qi_dqchunklen, XBF_TRYLOCK,
1265 : &bp, &xfs_dquot_buf_ops);
1266 35710223 : if (error == -EAGAIN)
1267 6593 : goto out_unlock;
1268 35703630 : if (xfs_metadata_is_sick(error))
1269 0 : xfs_dquot_mark_sick(dqp);
1270 35703630 : if (error)
1271 1590183 : goto out_abort;
1272 :
1273 34113447 : fa = xfs_qm_dqflush_check(dqp);
1274 34113447 : if (fa) {
1275 0 : xfs_alert(mp, "corrupt dquot ID 0x%x in memory at %pS",
1276 : dqp->q_id, fa);
1277 0 : xfs_buf_relse(bp);
1278 0 : xfs_dquot_mark_sick(dqp);
1279 0 : error = -EFSCORRUPTED;
1280 0 : goto out_abort;
1281 : }
1282 :
1283 : /* Flush the incore dquot to the ondisk buffer. */
1284 34113447 : dqblk = bp->b_addr + dqp->q_bufoffset;
1285 34113447 : xfs_dquot_to_disk(&dqblk->dd_diskdq, dqp);
1286 :
1287 : /*
1288 : * Clear the dirty field and remember the flush lsn for later use.
1289 : */
1290 34113446 : dqp->q_flags &= ~XFS_DQFLAG_DIRTY;
1291 :
1292 34113446 : xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn,
1293 : &dqp->q_logitem.qli_item.li_lsn);
1294 :
1295 : /*
1296 : * copy the lsn into the on-disk dquot now while we have the in memory
1297 : * dquot here. This can't be done later in the write verifier as we
1298 : * can't get access to the log item at that point in time.
1299 : *
1300 : * We also calculate the CRC here so that the on-disk dquot in the
1301 : * buffer always has a valid CRC. This ensures there is no possibility
1302 : * of a dquot without an up-to-date CRC getting to disk.
1303 : */
1304 34113446 : if (xfs_has_crc(mp)) {
1305 34113289 : dqblk->dd_lsn = cpu_to_be64(dqp->q_logitem.qli_item.li_lsn);
1306 34113289 : xfs_update_cksum((char *)dqblk, sizeof(struct xfs_dqblk),
1307 : XFS_DQUOT_CRC_OFF);
1308 : }
1309 :
1310 : /*
1311 : * Attach the dquot to the buffer so that we can remove this dquot from
1312 : * the AIL and release the flush lock once the dquot is synced to disk.
1313 : */
1314 34113447 : bp->b_flags |= _XBF_DQUOTS;
1315 34113447 : list_add_tail(&dqp->q_logitem.qli_item.li_bio_list, &bp->b_li_list);
1316 :
1317 : /*
1318 : * If the buffer is pinned then push on the log so we won't
1319 : * get stuck waiting in the write for too long.
1320 : */
1321 34113447 : if (xfs_buf_ispinned(bp)) {
1322 50463 : trace_xfs_dqflush_force(dqp);
1323 50463 : xfs_log_force(mp, 0);
1324 : }
1325 :
1326 34113447 : trace_xfs_dqflush_done(dqp);
1327 34113446 : *bpp = bp;
1328 34113446 : return 0;
1329 :
1330 1590183 : out_abort:
1331 1590183 : dqp->q_flags &= ~XFS_DQFLAG_DIRTY;
1332 1590183 : xfs_trans_ail_delete(lip, 0);
1333 1590183 : xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1334 1596776 : out_unlock:
1335 1596776 : xfs_dqfunlock(dqp);
1336 1596776 : return error;
1337 : }
1338 :
1339 : /*
1340 : * Lock two xfs_dquot structures.
1341 : *
1342 : * To avoid deadlocks we always lock the quota structure with
1343 : * the lowerd id first.
1344 : */
1345 : void
1346 35952091 : xfs_dqlock2(
1347 : struct xfs_dquot *d1,
1348 : struct xfs_dquot *d2)
1349 : {
1350 35952091 : if (d1 && d2) {
1351 35952091 : ASSERT(d1 != d2);
1352 35952091 : if (d1->q_id > d2->q_id) {
1353 29828139 : mutex_lock(&d2->q_qlock);
1354 29912488 : mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED);
1355 : } else {
1356 6123952 : mutex_lock(&d1->q_qlock);
1357 6124035 : mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED);
1358 : }
1359 0 : } else if (d1) {
1360 0 : mutex_lock(&d1->q_qlock);
1361 0 : } else if (d2) {
1362 0 : mutex_lock(&d2->q_qlock);
1363 : }
1364 36036564 : }
1365 :
1366 : static int
1367 18705 : xfs_dqtrx_cmp(
1368 : const void *a,
1369 : const void *b)
1370 : {
1371 18705 : const struct xfs_dqtrx *qa = a;
1372 18705 : const struct xfs_dqtrx *qb = b;
1373 :
1374 18705 : if (qa->qt_dquot->q_id > qb->qt_dquot->q_id)
1375 : return 1;
1376 11403 : if (qa->qt_dquot->q_id < qb->qt_dquot->q_id)
1377 11403 : return -1;
1378 : return 0;
1379 : }
1380 :
1381 : void
1382 6235 : xfs_dqlockn(
1383 : struct xfs_dqtrx *q)
1384 : {
1385 6235 : unsigned int i;
1386 :
1387 : /* Sort in order of dquot id, do not allow duplicates */
1388 24940 : for (i = 0; i < XFS_QM_TRANS_MAXDQS && q[i].qt_dquot != NULL; i++) {
1389 : unsigned int j;
1390 :
1391 37410 : for (j = 0; j < i; j++)
1392 18705 : ASSERT(q[i].qt_dquot != q[j].qt_dquot);
1393 : }
1394 6235 : if (i == 0)
1395 : return;
1396 :
1397 6235 : sort(q, i, sizeof(struct xfs_dqtrx), xfs_dqtrx_cmp, NULL);
1398 :
1399 6235 : mutex_lock(&q[0].qt_dquot->q_qlock);
1400 24940 : for (i = 1; i < XFS_QM_TRANS_MAXDQS && q[i].qt_dquot != NULL; i++)
1401 12470 : mutex_lock_nested(&q[i].qt_dquot->q_qlock, XFS_QLOCK_NESTED);
1402 : }
1403 :
1404 : int __init
1405 59 : xfs_qm_init(void)
1406 : {
1407 59 : xfs_dquot_cache = kmem_cache_create("xfs_dquot",
1408 : sizeof(struct xfs_dquot),
1409 : 0, 0, NULL);
1410 59 : if (!xfs_dquot_cache)
1411 0 : goto out;
1412 :
1413 59 : xfs_dqtrx_cache = kmem_cache_create("xfs_dqtrx",
1414 : sizeof(struct xfs_dquot_acct),
1415 : 0, 0, NULL);
1416 59 : if (!xfs_dqtrx_cache)
1417 0 : goto out_free_dquot_cache;
1418 :
1419 : return 0;
1420 :
1421 : out_free_dquot_cache:
1422 0 : kmem_cache_destroy(xfs_dquot_cache);
1423 : out:
1424 : return -ENOMEM;
1425 : }
1426 :
1427 : void
1428 58 : xfs_qm_exit(void)
1429 : {
1430 58 : kmem_cache_destroy(xfs_dqtrx_cache);
1431 58 : kmem_cache_destroy(xfs_dquot_cache);
1432 58 : }
1433 :
1434 : /*
1435 : * Iterate every dquot of a particular type. The caller must ensure that the
1436 : * particular quota type is active. iter_fn can return negative error codes,
1437 : * or -ECANCELED to indicate that it wants to stop iterating.
1438 : */
1439 : int
1440 405153 : xfs_qm_dqiterate(
1441 : struct xfs_mount *mp,
1442 : xfs_dqtype_t type,
1443 : xfs_qm_dqiterate_fn iter_fn,
1444 : void *priv)
1445 : {
1446 405153 : struct xfs_dquot *dq;
1447 405153 : xfs_dqid_t id = 0;
1448 405153 : int error;
1449 :
1450 405153 : do {
1451 405153 : error = xfs_qm_dqget_next(mp, id, type, &dq);
1452 405138 : if (error == -ENOENT)
1453 : return 0;
1454 405138 : if (error)
1455 0 : return error;
1456 :
1457 405138 : error = iter_fn(dq, type, priv);
1458 405152 : id = dq->q_id;
1459 405152 : xfs_qm_dqput(dq);
1460 405153 : } while (error == 0 && id != 0);
1461 :
1462 : return error;
1463 : }
|
