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 17057391 : xfs_qm_dqdestroy(
76 : struct xfs_dquot *dqp)
77 : {
78 17057391 : ASSERT(list_empty(&dqp->q_lru));
79 :
80 17057391 : kmem_free(dqp->q_logitem.qli_item.li_lv_shadow);
81 17057365 : mutex_destroy(&dqp->q_qlock);
82 :
83 17057363 : XFS_STATS_DEC(dqp->q_mount, xs_qm_dquot);
84 17057363 : kmem_cache_free(xfs_dquot_cache, dqp);
85 17057394 : }
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 291299092 : xfs_qm_adjust_dqlimits(
94 : struct xfs_dquot *dq)
95 : {
96 291299092 : struct xfs_mount *mp = dq->q_mount;
97 291299092 : struct xfs_quotainfo *q = mp->m_quotainfo;
98 291299092 : struct xfs_def_quota *defq;
99 291299092 : int prealloc = 0;
100 :
101 291299092 : ASSERT(dq->q_id);
102 291299092 : defq = xfs_get_defquota(q, xfs_dquot_type(dq));
103 :
104 291300252 : if (!dq->q_blk.softlimit) {
105 291145601 : dq->q_blk.softlimit = defq->blk.soft;
106 291145601 : prealloc = 1;
107 : }
108 291300252 : if (!dq->q_blk.hardlimit) {
109 291144402 : dq->q_blk.hardlimit = defq->blk.hard;
110 291144402 : prealloc = 1;
111 : }
112 291300252 : if (!dq->q_ino.softlimit)
113 291150911 : dq->q_ino.softlimit = defq->ino.soft;
114 291300252 : if (!dq->q_ino.hardlimit)
115 291152088 : dq->q_ino.hardlimit = defq->ino.hard;
116 291300252 : if (!dq->q_rtb.softlimit)
117 291300518 : dq->q_rtb.softlimit = defq->rtb.soft;
118 291300252 : if (!dq->q_rtb.hardlimit)
119 291301413 : dq->q_rtb.hardlimit = defq->rtb.hard;
120 :
121 291300252 : if (prealloc)
122 291146787 : xfs_dquot_set_prealloc_limits(dq);
123 291300604 : }
124 :
125 : /* Set the expiration time of a quota's grace period. */
126 : time64_t
127 74 : xfs_dquot_set_timeout(
128 : struct xfs_mount *mp,
129 : time64_t timeout)
130 : {
131 305 : struct xfs_quotainfo *qi = mp->m_quotainfo;
132 :
133 305 : 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 5575 : xfs_dquot_set_grace_period(
140 : time64_t grace)
141 : {
142 5575 : 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 873956789 : xfs_qm_adjust_res_timer(
151 : struct xfs_mount *mp,
152 : struct xfs_dquot_res *res,
153 : struct xfs_quota_limits *qlim)
154 : {
155 873956789 : ASSERT(res->hardlimit == 0 || res->softlimit <= res->hardlimit);
156 :
157 873956789 : if ((res->softlimit && res->count > res->softlimit) ||
158 873947535 : (res->hardlimit && res->count > res->hardlimit)) {
159 9410 : if (res->timer == 0)
160 231 : res->timer = xfs_dquot_set_timeout(mp,
161 231 : ktime_get_real_seconds() + qlim->time);
162 : } else {
163 873947379 : res->timer = 0;
164 : }
165 873956789 : }
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 291318902 : xfs_qm_adjust_dqtimers(
182 : struct xfs_dquot *dq)
183 : {
184 291318902 : struct xfs_mount *mp = dq->q_mount;
185 291318902 : struct xfs_quotainfo *qi = mp->m_quotainfo;
186 291318902 : struct xfs_def_quota *defq;
187 :
188 291318902 : ASSERT(dq->q_id);
189 291318902 : defq = xfs_get_defquota(qi, xfs_dquot_type(dq));
190 :
191 291320906 : xfs_qm_adjust_res_timer(dq->q_mount, &dq->q_blk, &defq->blk);
192 291321769 : xfs_qm_adjust_res_timer(dq->q_mount, &dq->q_ino, &defq->ino);
193 291321805 : xfs_qm_adjust_res_timer(dq->q_mount, &dq->q_rtb, &defq->rtb);
194 291321218 : }
195 :
196 : /*
197 : * initialize a buffer full of dquots and log the whole thing
198 : */
199 : STATIC void
200 3731942 : 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 3731942 : struct xfs_quotainfo *q = mp->m_quotainfo;
208 3731942 : struct xfs_dqblk *d;
209 3731942 : xfs_dqid_t curid;
210 3731942 : unsigned int qflag;
211 3731942 : unsigned int blftype;
212 3731942 : int i;
213 :
214 3731942 : ASSERT(tp);
215 3731942 : ASSERT(xfs_buf_islocked(bp));
216 :
217 3731942 : switch (type) {
218 : case XFS_DQTYPE_USER:
219 : qflag = XFS_UQUOTA_CHKD;
220 : blftype = XFS_BLF_UDQUOT_BUF;
221 : break;
222 65149 : case XFS_DQTYPE_PROJ:
223 65149 : qflag = XFS_PQUOTA_CHKD;
224 65149 : blftype = XFS_BLF_PDQUOT_BUF;
225 65149 : break;
226 1831557 : case XFS_DQTYPE_GROUP:
227 1831557 : qflag = XFS_GQUOTA_CHKD;
228 1831557 : blftype = XFS_BLF_GDQUOT_BUF;
229 1831557 : break;
230 0 : default:
231 0 : ASSERT(0);
232 0 : return;
233 : }
234 :
235 3731942 : d = bp->b_addr;
236 :
237 : /*
238 : * ID of the first dquot in the block - id's are zero based.
239 : */
240 3731942 : curid = id - (id % q->qi_dqperchunk);
241 3731942 : memset(d, 0, BBTOB(q->qi_dqchunklen));
242 124759881 : for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++) {
243 121027952 : d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
244 121027952 : d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
245 121027952 : d->dd_diskdq.d_id = cpu_to_be32(curid);
246 121027952 : d->dd_diskdq.d_type = type;
247 121027952 : if (curid > 0 && xfs_has_bigtime(mp))
248 120881363 : d->dd_diskdq.d_type |= XFS_DQTYPE_BIGTIME;
249 121027952 : if (xfs_has_crc(mp)) {
250 121026391 : uuid_copy(&d->dd_uuid, &mp->m_sb.sb_meta_uuid);
251 121026393 : xfs_update_cksum((char *)d, sizeof(struct xfs_dqblk),
252 : XFS_DQUOT_CRC_OFF);
253 : }
254 : }
255 :
256 3731929 : 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 3731918 : if (!(mp->m_qflags & qflag))
275 29093 : xfs_trans_ordered_buf(tp, bp);
276 : else
277 3702825 : 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 308128821 : xfs_dquot_set_prealloc_limits(struct xfs_dquot *dqp)
287 : {
288 308128821 : uint64_t space;
289 :
290 308128821 : dqp->q_prealloc_hi_wmark = dqp->q_blk.hardlimit;
291 308128821 : dqp->q_prealloc_lo_wmark = dqp->q_blk.softlimit;
292 308128821 : if (!dqp->q_prealloc_lo_wmark) {
293 308119403 : dqp->q_prealloc_lo_wmark = dqp->q_prealloc_hi_wmark;
294 308119403 : do_div(dqp->q_prealloc_lo_wmark, 100);
295 308119403 : dqp->q_prealloc_lo_wmark *= 95;
296 : }
297 :
298 308128821 : space = dqp->q_prealloc_hi_wmark;
299 :
300 308128821 : do_div(space, 100);
301 308128821 : dqp->q_low_space[XFS_QLOWSP_1_PCNT] = space;
302 308128821 : dqp->q_low_space[XFS_QLOWSP_3_PCNT] = space * 3;
303 308128821 : dqp->q_low_space[XFS_QLOWSP_5_PCNT] = space * 5;
304 308128821 : }
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 3759827 : xfs_dquot_disk_alloc(
313 : struct xfs_dquot *dqp,
314 : struct xfs_buf **bpp)
315 : {
316 3759827 : struct xfs_bmbt_irec map;
317 3759827 : struct xfs_trans *tp;
318 3759827 : struct xfs_mount *mp = dqp->q_mount;
319 3759827 : struct xfs_buf *bp;
320 3759827 : xfs_dqtype_t qtype = xfs_dquot_type(dqp);
321 3759827 : struct xfs_inode *quotip = xfs_quota_inode(mp, qtype);
322 3759826 : int nmaps = 1;
323 3759826 : int error;
324 :
325 3759826 : trace_xfs_dqalloc(dqp);
326 :
327 3759805 : error = xfs_trans_alloc(mp, &M_RES(mp)->tr_qm_dqalloc,
328 3759805 : XFS_QM_DQALLOC_SPACE_RES(mp), 0, 0, &tp);
329 3759821 : if (error)
330 : return error;
331 :
332 3733585 : xfs_ilock(quotip, XFS_ILOCK_EXCL);
333 3733605 : xfs_trans_ijoin(tp, quotip, 0);
334 :
335 3733556 : 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 3733556 : error = xfs_iext_count_may_overflow(quotip, XFS_DATA_FORK,
345 : XFS_IEXT_ADD_NOSPLIT_CNT);
346 3733557 : if (error == -EFBIG)
347 2 : error = xfs_iext_count_upgrade(tp, quotip,
348 : XFS_IEXT_ADD_NOSPLIT_CNT);
349 3733557 : if (error)
350 2 : goto err_cancel;
351 :
352 : /* Create the block mapping. */
353 3733555 : error = xfs_bmapi_write(tp, quotip, dqp->q_fileoffset,
354 : XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA, 0, &map,
355 : &nmaps);
356 3733610 : if (error)
357 1681 : goto err_cancel;
358 :
359 3731929 : ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB);
360 3731929 : ASSERT(nmaps == 1);
361 3731929 : 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 3731929 : 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 3731929 : error = xfs_trans_get_buf(tp, mp->m_ddev_targp, dqp->q_blkno,
371 3731929 : mp->m_quotainfo->qi_dqchunklen, 0, &bp);
372 3731937 : if (error)
373 0 : goto err_cancel;
374 3731937 : 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 3731937 : xfs_qm_init_dquot_blk(tp, mp, dqp->q_id, qtype, bp);
381 3731930 : 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 3731933 : xfs_trans_bhold(tp, bp);
408 3731912 : error = xfs_trans_commit(tp);
409 3731941 : xfs_iunlock(quotip, XFS_ILOCK_EXCL);
410 3731931 : if (error) {
411 196 : xfs_buf_relse(bp);
412 196 : return error;
413 : }
414 :
415 3731735 : *bpp = bp;
416 3731735 : return 0;
417 :
418 1683 : err_cancel:
419 1683 : xfs_trans_cancel(tp);
420 1683 : xfs_iunlock(quotip, XFS_ILOCK_EXCL);
421 1683 : 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 17057305 : xfs_dquot_disk_read(
430 : struct xfs_mount *mp,
431 : struct xfs_dquot *dqp,
432 : struct xfs_buf **bpp)
433 : {
434 17057305 : struct xfs_bmbt_irec map;
435 17057305 : struct xfs_buf *bp;
436 17057305 : xfs_dqtype_t qtype = xfs_dquot_type(dqp);
437 17057305 : struct xfs_inode *quotip = xfs_quota_inode(mp, qtype);
438 17057257 : uint lock_mode;
439 17057257 : int nmaps = 1;
440 17057257 : int error;
441 :
442 17057257 : lock_mode = xfs_ilock_data_map_shared(quotip);
443 17056965 : 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 17056965 : error = xfs_bmapi_read(quotip, dqp->q_fileoffset,
456 : XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0);
457 17057031 : xfs_iunlock(quotip, lock_mode);
458 17057253 : if (error)
459 : return error;
460 :
461 17056386 : ASSERT(nmaps == 1);
462 17056386 : ASSERT(map.br_blockcount >= 1);
463 17056386 : ASSERT(map.br_startblock != DELAYSTARTBLOCK);
464 17056386 : if (map.br_startblock == HOLESTARTBLOCK)
465 : return -ENOENT;
466 :
467 13240482 : 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 13240488 : dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
474 :
475 13240488 : error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
476 13240488 : mp->m_quotainfo->qi_dqchunklen, 0, &bp,
477 : &xfs_dquot_buf_ops);
478 13240426 : if (xfs_metadata_is_sick(error))
479 6 : xfs_dquot_mark_sick(dqp);
480 13240426 : if (error) {
481 6979 : ASSERT(bp == NULL);
482 6979 : return error;
483 : }
484 :
485 13233447 : ASSERT(xfs_buf_islocked(bp));
486 13233447 : xfs_buf_set_ref(bp, XFS_DQUOT_REF);
487 13233336 : *bpp = bp;
488 :
489 13233336 : return 0;
490 : }
491 :
492 : /* Allocate and initialize everything we need for an incore dquot. */
493 : STATIC struct xfs_dquot *
494 17057338 : xfs_dquot_alloc(
495 : struct xfs_mount *mp,
496 : xfs_dqid_t id,
497 : xfs_dqtype_t type)
498 : {
499 17057338 : struct xfs_dquot *dqp;
500 :
501 17057338 : dqp = kmem_cache_zalloc(xfs_dquot_cache, GFP_KERNEL | __GFP_NOFAIL);
502 :
503 17057313 : dqp->q_type = type;
504 17057313 : dqp->q_id = id;
505 17057313 : dqp->q_mount = mp;
506 17057313 : INIT_LIST_HEAD(&dqp->q_lru);
507 17057313 : mutex_init(&dqp->q_qlock);
508 17057331 : init_waitqueue_head(&dqp->q_pinwait);
509 17057323 : 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 17057323 : 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 17057323 : init_completion(&dqp->q_flush);
522 17057351 : complete(&dqp->q_flush);
523 :
524 : /*
525 : * Make sure group quotas have a different lock class than user
526 : * quotas.
527 : */
528 17057316 : 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 17057316 : xfs_qm_dquot_logitem_init(dqp);
544 :
545 17057307 : XFS_STATS_INC(mp, xs_qm_dquot);
546 17057307 : return dqp;
547 : }
548 :
549 : /* Check the ondisk dquot's id and type match what the incore dquot expects. */
550 : static bool
551 16964858 : xfs_dquot_check_type(
552 : struct xfs_dquot *dqp,
553 : struct xfs_disk_dquot *ddqp)
554 : {
555 16964858 : uint8_t ddqp_type;
556 16964858 : uint8_t dqp_type;
557 :
558 16964858 : ddqp_type = ddqp->d_type & XFS_DQTYPE_REC_MASK;
559 16964858 : dqp_type = xfs_dquot_type(dqp);
560 :
561 33929716 : 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 16965179 : if (xfs_has_crc(dqp->q_mount) ||
570 32 : dqp_type == XFS_DQTYPE_USER || dqp->q_id != 0)
571 16965147 : 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 32 : 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 16965200 : xfs_dquot_from_disk(
588 : struct xfs_dquot *dqp,
589 : struct xfs_buf *bp)
590 : {
591 16965200 : 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 16965200 : 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 16965200 : dqp->q_type = ddqp->d_type;
608 16965200 : dqp->q_blk.hardlimit = be64_to_cpu(ddqp->d_blk_hardlimit);
609 16965200 : dqp->q_blk.softlimit = be64_to_cpu(ddqp->d_blk_softlimit);
610 16965200 : dqp->q_ino.hardlimit = be64_to_cpu(ddqp->d_ino_hardlimit);
611 16965200 : dqp->q_ino.softlimit = be64_to_cpu(ddqp->d_ino_softlimit);
612 16965200 : dqp->q_rtb.hardlimit = be64_to_cpu(ddqp->d_rtb_hardlimit);
613 16965200 : dqp->q_rtb.softlimit = be64_to_cpu(ddqp->d_rtb_softlimit);
614 :
615 16965200 : dqp->q_blk.count = be64_to_cpu(ddqp->d_bcount);
616 16965200 : dqp->q_ino.count = be64_to_cpu(ddqp->d_icount);
617 16965200 : dqp->q_rtb.count = be64_to_cpu(ddqp->d_rtbcount);
618 :
619 16965200 : dqp->q_blk.timer = xfs_dquot_from_disk_ts(ddqp, ddqp->d_btimer);
620 16965144 : dqp->q_ino.timer = xfs_dquot_from_disk_ts(ddqp, ddqp->d_itimer);
621 16965144 : 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 16965127 : dqp->q_blk.reserved = dqp->q_blk.count;
628 16965127 : dqp->q_ino.reserved = dqp->q_ino.count;
629 16965127 : dqp->q_rtb.reserved = dqp->q_rtb.count;
630 :
631 : /* initialize the dquot speculative prealloc thresholds */
632 16965127 : xfs_dquot_set_prealloc_limits(dqp);
633 16965127 : return 0;
634 : }
635 :
636 : /* Copy the in-core quota fields into the on-disk buffer. */
637 : void
638 4595968902 : xfs_dquot_to_disk(
639 : struct xfs_disk_dquot *ddqp,
640 : struct xfs_dquot *dqp)
641 : {
642 4595968902 : ddqp->d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
643 4595968902 : ddqp->d_version = XFS_DQUOT_VERSION;
644 4595968902 : ddqp->d_type = dqp->q_type;
645 4595968902 : ddqp->d_id = cpu_to_be32(dqp->q_id);
646 4595968902 : ddqp->d_pad0 = 0;
647 4595968902 : ddqp->d_pad = 0;
648 :
649 4595968902 : ddqp->d_blk_hardlimit = cpu_to_be64(dqp->q_blk.hardlimit);
650 4595968902 : ddqp->d_blk_softlimit = cpu_to_be64(dqp->q_blk.softlimit);
651 4595968902 : ddqp->d_ino_hardlimit = cpu_to_be64(dqp->q_ino.hardlimit);
652 4595968902 : ddqp->d_ino_softlimit = cpu_to_be64(dqp->q_ino.softlimit);
653 4595968902 : ddqp->d_rtb_hardlimit = cpu_to_be64(dqp->q_rtb.hardlimit);
654 4595968902 : ddqp->d_rtb_softlimit = cpu_to_be64(dqp->q_rtb.softlimit);
655 :
656 4595968902 : ddqp->d_bcount = cpu_to_be64(dqp->q_blk.count);
657 4595968902 : ddqp->d_icount = cpu_to_be64(dqp->q_ino.count);
658 4595968902 : ddqp->d_rtbcount = cpu_to_be64(dqp->q_rtb.count);
659 :
660 4595968902 : ddqp->d_bwarns = 0;
661 4595968902 : ddqp->d_iwarns = 0;
662 4595968902 : ddqp->d_rtbwarns = 0;
663 :
664 4595968902 : ddqp->d_btimer = xfs_dquot_to_disk_ts(dqp, dqp->q_blk.timer);
665 4595966752 : ddqp->d_itimer = xfs_dquot_to_disk_ts(dqp, dqp->q_ino.timer);
666 4595974653 : ddqp->d_rtbtimer = xfs_dquot_to_disk_ts(dqp, dqp->q_rtb.timer);
667 4595985700 : }
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 17057363 : 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 17057363 : struct xfs_dquot *dqp;
683 17057363 : struct xfs_buf *bp;
684 17057363 : int error;
685 :
686 17057363 : dqp = xfs_dquot_alloc(mp, id, type);
687 17057302 : trace_xfs_dqread(dqp);
688 :
689 : /* Try to read the buffer, allocating if necessary. */
690 17057324 : error = xfs_dquot_disk_read(mp, dqp, &bp);
691 17057103 : if (error == -ENOENT && can_alloc)
692 3759829 : error = xfs_dquot_disk_alloc(dqp, &bp);
693 17057137 : if (error)
694 92049 : 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 16965088 : ASSERT(xfs_buf_islocked(bp));
703 16965088 : error = xfs_dquot_from_disk(dqp, bp);
704 16965099 : xfs_buf_relse(bp);
705 16965278 : if (error)
706 0 : goto err;
707 :
708 16965278 : *dqpp = dqp;
709 16965278 : return error;
710 :
711 92049 : err:
712 92049 : trace_xfs_dqread_fail(dqp);
713 92046 : xfs_qm_dqdestroy(dqp);
714 92030 : *dqpp = NULL;
715 92030 : 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 5813121 : xfs_dq_get_next_id(
725 : struct xfs_mount *mp,
726 : xfs_dqtype_t type,
727 : xfs_dqid_t *id)
728 : {
729 5813121 : struct xfs_inode *quotip = xfs_quota_inode(mp, type);
730 5813121 : xfs_dqid_t next_id = *id + 1; /* simple advance */
731 5813121 : uint lock_flags;
732 5813121 : struct xfs_bmbt_irec got;
733 5813121 : struct xfs_iext_cursor cur;
734 5813121 : xfs_fsblock_t start;
735 5813121 : int error = 0;
736 :
737 : /* If we'd wrap past the max ID, stop */
738 5813121 : if (next_id < *id)
739 : return -ENOENT;
740 :
741 : /* If new ID is within the current chunk, advancing it sufficed */
742 5813113 : if (next_id % mp->m_quotainfo->qi_dqperchunk) {
743 5681010 : *id = next_id;
744 5681010 : return 0;
745 : }
746 :
747 : /* Nope, next_id is now past the current chunk, so find the next one */
748 132103 : start = (xfs_fsblock_t)next_id / mp->m_quotainfo->qi_dqperchunk;
749 :
750 132103 : lock_flags = xfs_ilock_data_map_shared(quotip);
751 132103 : error = xfs_iread_extents(NULL, quotip, XFS_DATA_FORK);
752 132103 : if (error)
753 : return error;
754 :
755 132103 : if (xfs_iext_lookup_extent(quotip, "ip->i_df, start, &cur, &got)) {
756 : /* contiguous chunk, bump startoff for the id calculation */
757 131043 : if (got.br_startoff < start)
758 180 : got.br_startoff = start;
759 131043 : *id = got.br_startoff * mp->m_quotainfo->qi_dqperchunk;
760 : } else {
761 : error = -ENOENT;
762 : }
763 :
764 132103 : xfs_iunlock(quotip, lock_flags);
765 :
766 132103 : 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 129946044 : 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 129947000 : struct xfs_dquot *dqp;
781 :
782 129947000 : restart:
783 129947000 : mutex_lock(&qi->qi_tree_lock);
784 129951510 : dqp = radix_tree_lookup(tree, id);
785 129951510 : if (!dqp) {
786 16925087 : mutex_unlock(&qi->qi_tree_lock);
787 16925087 : XFS_STATS_INC(mp, xs_qm_dqcachemisses);
788 16925087 : return NULL;
789 : }
790 :
791 113026423 : xfs_dqlock(dqp);
792 113026423 : if (dqp->q_flags & XFS_DQFLAG_FREEING) {
793 956 : xfs_dqunlock(dqp);
794 956 : mutex_unlock(&qi->qi_tree_lock);
795 956 : trace_xfs_dqget_freeing(dqp);
796 956 : delay(1);
797 956 : goto restart;
798 : }
799 :
800 113025467 : dqp->q_nrefs++;
801 113025467 : mutex_unlock(&qi->qi_tree_lock);
802 :
803 113025426 : trace_xfs_dqget_hit(dqp);
804 113025461 : XFS_STATS_INC(mp, xs_qm_dqcachehits);
805 113025461 : 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 16850139 : 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 16850139 : int error;
823 :
824 16850139 : mutex_lock(&qi->qi_tree_lock);
825 16850208 : error = radix_tree_insert(tree, id, dqp);
826 16850208 : if (unlikely(error)) {
827 : /* Duplicate found! Caller must try again. */
828 179 : mutex_unlock(&qi->qi_tree_lock);
829 179 : trace_xfs_dqget_dup(dqp);
830 179 : return error;
831 : }
832 :
833 : /* Return a locked dquot to the caller, with a reference taken. */
834 16850029 : xfs_dqlock(dqp);
835 16850029 : dqp->q_nrefs = 1;
836 :
837 16850029 : qi->qi_dquots++;
838 16850029 : mutex_unlock(&qi->qi_tree_lock);
839 :
840 16850029 : return 0;
841 : }
842 :
843 : /* Check our input parameters. */
844 : static int
845 130083125 : xfs_qm_dqget_checks(
846 : struct xfs_mount *mp,
847 : xfs_dqtype_t type)
848 : {
849 130083125 : switch (type) {
850 49938250 : case XFS_DQTYPE_USER:
851 49938250 : if (!XFS_IS_UQUOTA_ON(mp))
852 328 : return -ESRCH;
853 : return 0;
854 49637495 : case XFS_DQTYPE_GROUP:
855 49637495 : if (!XFS_IS_GQUOTA_ON(mp))
856 934 : return -ESRCH;
857 : return 0;
858 30507380 : case XFS_DQTYPE_PROJ:
859 30507380 : if (!XFS_IS_PQUOTA_ON(mp))
860 340 : 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 43716535 : 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 43716535 : struct xfs_quotainfo *qi = mp->m_quotainfo;
881 43716535 : struct radix_tree_root *tree = xfs_dquot_tree(qi, type);
882 43716223 : struct xfs_dquot *dqp;
883 43716223 : int error;
884 :
885 43716223 : error = xfs_qm_dqget_checks(mp, type);
886 43716223 : if (error)
887 : return error;
888 :
889 43715381 : restart:
890 43715501 : dqp = xfs_qm_dqget_cache_lookup(mp, qi, tree, id);
891 43716952 : if (dqp) {
892 31007700 : *O_dqpp = dqp;
893 31007700 : return 0;
894 : }
895 :
896 12709252 : error = xfs_qm_dqread(mp, id, type, can_alloc, &dqp);
897 12709219 : if (error)
898 69275 : return error;
899 :
900 12639944 : error = xfs_qm_dqget_cache_insert(mp, qi, tree, id, dqp);
901 12639954 : if (error) {
902 : /*
903 : * Duplicate found. Just throw away the new dquot and start
904 : * over.
905 : */
906 120 : xfs_qm_dqdestroy(dqp);
907 120 : XFS_STATS_INC(mp, xs_qm_dquot_dups);
908 120 : goto restart;
909 : }
910 :
911 12639834 : trace_xfs_dqget_miss(dqp);
912 12639833 : *O_dqpp = dqp;
913 12639833 : 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 133052 : 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 133052 : int error;
930 :
931 133052 : error = xfs_qm_dqget_checks(mp, type);
932 133052 : if (error)
933 : return error;
934 :
935 132292 : 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 328960694 : xfs_qm_id_for_quotatype(
941 : struct xfs_inode *ip,
942 : xfs_dqtype_t type)
943 : {
944 328960694 : switch (type) {
945 : case XFS_DQTYPE_USER:
946 109656702 : return i_uid_read(VFS_I(ip));
947 : case XFS_DQTYPE_GROUP:
948 109652349 : return i_gid_read(VFS_I(ip));
949 109651643 : case XFS_DQTYPE_PROJ:
950 109651643 : 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 86231886 : 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 86231886 : struct xfs_mount *mp = ip->i_mount;
969 86231886 : struct xfs_quotainfo *qi = mp->m_quotainfo;
970 86231886 : struct radix_tree_root *tree = xfs_dquot_tree(qi, type);
971 86232702 : struct xfs_dquot *dqp;
972 86232702 : xfs_dqid_t id;
973 86232702 : int error;
974 :
975 86232702 : error = xfs_qm_dqget_checks(mp, type);
976 86232702 : if (error)
977 : return error;
978 :
979 86232499 : ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
980 172466442 : ASSERT(xfs_inode_dquot(ip, type) == NULL);
981 86233221 : ASSERT(!xfs_is_metadir_inode(ip));
982 :
983 86233221 : id = xfs_qm_id_for_quotatype(ip, type);
984 :
985 86232229 : restart:
986 86232229 : dqp = xfs_qm_dqget_cache_lookup(mp, qi, tree, id);
987 86233590 : if (dqp) {
988 82017758 : *O_dqpp = dqp;
989 82017758 : 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 4215832 : xfs_iunlock(ip, XFS_ILOCK_EXCL);
1000 4215832 : error = xfs_qm_dqread(mp, id, type, can_alloc, &dqp);
1001 4215805 : xfs_ilock(ip, XFS_ILOCK_EXCL);
1002 4215810 : if (error)
1003 5577 : return error;
1004 :
1005 : /*
1006 : * A dquot could be attached to this inode by now, since we had
1007 : * dropped the ilock.
1008 : */
1009 4210233 : if (xfs_this_quota_on(mp, type)) {
1010 4210233 : struct xfs_dquot *dqp1;
1011 :
1012 4210233 : dqp1 = xfs_inode_dquot(ip, type);
1013 4210233 : if (dqp1) {
1014 0 : xfs_qm_dqdestroy(dqp);
1015 0 : dqp = dqp1;
1016 0 : xfs_dqlock(dqp);
1017 0 : goto dqret;
1018 : }
1019 : } else {
1020 : /* inode stays locked on return */
1021 0 : xfs_qm_dqdestroy(dqp);
1022 0 : return -ESRCH;
1023 : }
1024 :
1025 4210233 : error = xfs_qm_dqget_cache_insert(mp, qi, tree, id, dqp);
1026 4210254 : if (error) {
1027 : /*
1028 : * Duplicate found. Just throw away the new dquot and start
1029 : * over.
1030 : */
1031 59 : xfs_qm_dqdestroy(dqp);
1032 59 : XFS_STATS_INC(mp, xs_qm_dquot_dups);
1033 59 : goto restart;
1034 : }
1035 :
1036 4210195 : dqret:
1037 4210195 : ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1038 4210195 : trace_xfs_dqget_miss(dqp);
1039 4210190 : *O_dqpp = dqp;
1040 4210190 : 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 273999 : 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 273999 : struct xfs_dquot *dqp;
1055 273999 : int error = 0;
1056 :
1057 273999 : *dqpp = NULL;
1058 6087120 : for (; !error; error = xfs_dq_get_next_id(mp, type, &id)) {
1059 6086052 : error = xfs_qm_dqget(mp, id, type, false, &dqp);
1060 6086052 : if (error == -ENOENT)
1061 31326 : continue;
1062 6054726 : else if (error != 0)
1063 : break;
1064 :
1065 6054698 : if (!XFS_IS_DQUOT_UNINITIALIZED(dqp)) {
1066 272903 : *dqpp = dqp;
1067 272903 : return 0;
1068 : }
1069 :
1070 5781795 : 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 523936114 : xfs_qm_dqput(
1084 : struct xfs_dquot *dqp)
1085 : {
1086 523936114 : ASSERT(dqp->q_nrefs > 0);
1087 523936114 : ASSERT(XFS_DQ_IS_LOCKED(dqp));
1088 :
1089 523938567 : trace_xfs_dqput(dqp);
1090 :
1091 523940086 : if (--dqp->q_nrefs == 0) {
1092 27988599 : struct xfs_quotainfo *qi = dqp->q_mount->m_quotainfo;
1093 27988599 : trace_xfs_dqput_free(dqp);
1094 :
1095 27988601 : if (list_lru_add(&qi->qi_lru, &dqp->q_lru))
1096 19029473 : XFS_STATS_INC(dqp->q_mount, xs_qm_dquot_unused);
1097 : }
1098 523940243 : xfs_dqunlock(dqp);
1099 523850164 : }
1100 :
1101 : /*
1102 : * Release a dquot. Flush it if dirty, then dqput() it.
1103 : * dquot must not be locked.
1104 : */
1105 : void
1106 519416892 : xfs_qm_dqrele(
1107 : struct xfs_dquot *dqp)
1108 : {
1109 519416892 : if (!dqp)
1110 : return;
1111 :
1112 498515448 : trace_xfs_dqrele(dqp);
1113 :
1114 498532248 : 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 498417779 : 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 25042864 : xfs_qm_dqflush_done(
1133 : struct xfs_log_item *lip)
1134 : {
1135 25042864 : struct xfs_dq_logitem *qip = (struct xfs_dq_logitem *)lip;
1136 25042864 : struct xfs_dquot *dqp = qip->qli_dquot;
1137 25042864 : struct xfs_ail *ailp = lip->li_ailp;
1138 25042864 : 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 50117424 : if (test_bit(XFS_LI_IN_AIL, &lip->li_flags) &&
1149 25060881 : ((lip->li_lsn == qip->qli_flush_lsn) ||
1150 0 : test_bit(XFS_LI_FAILED, &lip->li_flags))) {
1151 :
1152 24997489 : spin_lock(&ailp->ail_lock);
1153 24997489 : xfs_clear_li_failed(lip);
1154 24997489 : if (lip->li_lsn == qip->qli_flush_lsn) {
1155 : /* xfs_ail_update_finish() drops the AIL lock */
1156 24997489 : tail_lsn = xfs_ail_delete_one(ailp, lip);
1157 24997489 : 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 25042864 : xfs_dqfunlock(dqp);
1167 25042864 : }
1168 :
1169 : void
1170 21011796 : xfs_buf_dquot_iodone(
1171 : struct xfs_buf *bp)
1172 : {
1173 21011796 : struct xfs_log_item *lip, *n;
1174 :
1175 46054660 : list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) {
1176 25042864 : list_del_init(&lip->li_bio_list);
1177 25042864 : xfs_qm_dqflush_done(lip);
1178 : }
1179 21011796 : }
1180 :
1181 : void
1182 42 : xfs_buf_dquot_io_fail(
1183 : struct xfs_buf *bp)
1184 : {
1185 42 : struct xfs_log_item *lip;
1186 :
1187 42 : spin_lock(&bp->b_mount->m_ail->ail_lock);
1188 84 : list_for_each_entry(lip, &bp->b_li_list, li_bio_list)
1189 42 : xfs_set_li_failed(lip, bp);
1190 42 : spin_unlock(&bp->b_mount->m_ail->ail_lock);
1191 42 : }
1192 :
1193 : /* Check incore dquot for errors before we flush. */
1194 : static xfs_failaddr_t
1195 25042807 : xfs_qm_dqflush_check(
1196 : struct xfs_dquot *dqp)
1197 : {
1198 25042807 : xfs_dqtype_t type = xfs_dquot_type(dqp);
1199 :
1200 25042807 : if (type != XFS_DQTYPE_USER &&
1201 25042807 : type != XFS_DQTYPE_GROUP &&
1202 : type != XFS_DQTYPE_PROJ)
1203 0 : return __this_address;
1204 :
1205 25042807 : if (dqp->q_id == 0)
1206 : return NULL;
1207 :
1208 24823194 : if (dqp->q_blk.softlimit && dqp->q_blk.count > dqp->q_blk.softlimit &&
1209 137 : !dqp->q_blk.timer)
1210 0 : return __this_address;
1211 :
1212 24823194 : if (dqp->q_ino.softlimit && dqp->q_ino.count > dqp->q_ino.softlimit &&
1213 166 : !dqp->q_ino.timer)
1214 0 : return __this_address;
1215 :
1216 24823194 : 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 24823194 : if (dqp->q_type & XFS_DQTYPE_BIGTIME) {
1222 24819123 : 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 26497346 : xfs_qm_dqflush(
1241 : struct xfs_dquot *dqp,
1242 : struct xfs_buf **bpp)
1243 : {
1244 26497346 : struct xfs_mount *mp = dqp->q_mount;
1245 26497346 : struct xfs_log_item *lip = &dqp->q_logitem.qli_item;
1246 26497346 : struct xfs_buf *bp;
1247 26497346 : struct xfs_dqblk *dqblk;
1248 26497346 : xfs_failaddr_t fa;
1249 26497346 : int error;
1250 :
1251 26497346 : ASSERT(XFS_DQ_IS_LOCKED(dqp));
1252 26497350 : ASSERT(!completion_done(&dqp->q_flush));
1253 :
1254 26497357 : trace_xfs_dqflush(dqp);
1255 :
1256 26497344 : *bpp = NULL;
1257 :
1258 26497344 : xfs_qm_dqunpin_wait(dqp);
1259 :
1260 : /*
1261 : * Get the buffer containing the on-disk dquot
1262 : */
1263 26497340 : error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
1264 26497340 : mp->m_quotainfo->qi_dqchunklen, XBF_TRYLOCK,
1265 : &bp, &xfs_dquot_buf_ops);
1266 26497278 : if (error == -EAGAIN)
1267 5327 : goto out_unlock;
1268 26491951 : if (xfs_metadata_is_sick(error))
1269 0 : xfs_dquot_mark_sick(dqp);
1270 26491951 : if (error)
1271 1449187 : goto out_abort;
1272 :
1273 25042764 : fa = xfs_qm_dqflush_check(dqp);
1274 25042798 : 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 25042798 : dqblk = bp->b_addr + dqp->q_bufoffset;
1285 25042798 : 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 25042817 : dqp->q_flags &= ~XFS_DQFLAG_DIRTY;
1291 :
1292 25042817 : 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 25042817 : if (xfs_has_crc(mp)) {
1305 25042776 : dqblk->dd_lsn = cpu_to_be64(dqp->q_logitem.qli_item.li_lsn);
1306 25042776 : 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 25042856 : bp->b_flags |= _XBF_DQUOTS;
1315 25042856 : 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 25042833 : if (xfs_buf_ispinned(bp)) {
1322 11000 : trace_xfs_dqflush_force(dqp);
1323 11000 : xfs_log_force(mp, 0);
1324 : }
1325 :
1326 25042833 : trace_xfs_dqflush_done(dqp);
1327 25042807 : *bpp = bp;
1328 25042807 : return 0;
1329 :
1330 1449187 : out_abort:
1331 1449187 : dqp->q_flags &= ~XFS_DQFLAG_DIRTY;
1332 1449187 : xfs_trans_ail_delete(lip, 0);
1333 1449187 : xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1334 1454514 : out_unlock:
1335 1454514 : xfs_dqfunlock(dqp);
1336 1454514 : 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 14070734 : xfs_dqlock2(
1347 : struct xfs_dquot *d1,
1348 : struct xfs_dquot *d2)
1349 : {
1350 14070734 : if (d1 && d2) {
1351 14070734 : ASSERT(d1 != d2);
1352 14070734 : if (d1->q_id > d2->q_id) {
1353 11123282 : mutex_lock(&d2->q_qlock);
1354 11123704 : mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED);
1355 : } else {
1356 2947452 : mutex_lock(&d1->q_qlock);
1357 2947457 : 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 14071158 : }
1365 :
1366 : static int
1367 13620 : xfs_dqtrx_cmp(
1368 : const void *a,
1369 : const void *b)
1370 : {
1371 13620 : const struct xfs_dqtrx *qa = a;
1372 13620 : const struct xfs_dqtrx *qb = b;
1373 :
1374 13620 : if (qa->qt_dquot->q_id > qb->qt_dquot->q_id)
1375 : return 1;
1376 7930 : if (qa->qt_dquot->q_id < qb->qt_dquot->q_id)
1377 7930 : return -1;
1378 : return 0;
1379 : }
1380 :
1381 : void
1382 4540 : xfs_dqlockn(
1383 : struct xfs_dqtrx *q)
1384 : {
1385 4540 : unsigned int i;
1386 :
1387 : /* Sort in order of dquot id, do not allow duplicates */
1388 18160 : for (i = 0; i < XFS_QM_TRANS_MAXDQS && q[i].qt_dquot != NULL; i++) {
1389 : unsigned int j;
1390 :
1391 27240 : for (j = 0; j < i; j++)
1392 13620 : ASSERT(q[i].qt_dquot != q[j].qt_dquot);
1393 : }
1394 4540 : if (i == 0)
1395 : return;
1396 :
1397 4540 : sort(q, i, sizeof(struct xfs_dqtrx), xfs_dqtrx_cmp, NULL);
1398 :
1399 4540 : mutex_lock(&q[0].qt_dquot->q_qlock);
1400 18160 : for (i = 1; i < XFS_QM_TRANS_MAXDQS && q[i].qt_dquot != NULL; i++)
1401 9080 : mutex_lock_nested(&q[i].qt_dquot->q_qlock, XFS_QLOCK_NESTED);
1402 : }
1403 :
1404 : int __init
1405 12 : xfs_qm_init(void)
1406 : {
1407 12 : xfs_dquot_cache = kmem_cache_create("xfs_dquot",
1408 : sizeof(struct xfs_dquot),
1409 : 0, 0, NULL);
1410 12 : if (!xfs_dquot_cache)
1411 0 : goto out;
1412 :
1413 12 : xfs_dqtrx_cache = kmem_cache_create("xfs_dqtrx",
1414 : sizeof(struct xfs_dquot_acct),
1415 : 0, 0, NULL);
1416 12 : 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 12 : xfs_qm_exit(void)
1429 : {
1430 12 : kmem_cache_destroy(xfs_dqtrx_cache);
1431 12 : kmem_cache_destroy(xfs_dquot_cache);
1432 12 : }
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 211368 : 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 211368 : struct xfs_dquot *dq;
1447 211368 : xfs_dqid_t id = 0;
1448 211368 : int error;
1449 :
1450 211368 : do {
1451 211368 : error = xfs_qm_dqget_next(mp, id, type, &dq);
1452 211368 : if (error == -ENOENT)
1453 : return 0;
1454 211368 : if (error)
1455 0 : return error;
1456 :
1457 211368 : error = iter_fn(dq, type, priv);
1458 211368 : id = dq->q_id;
1459 211368 : xfs_qm_dqput(dq);
1460 211368 : } while (error == 0 && id != 0);
1461 :
1462 : return error;
1463 : }
|
