1 // SPDX-License-Identifier: GPL-2.0
3 * Implementation of operations over global quota file
5 #include <linux/spinlock.h>
7 #include <linux/slab.h>
8 #include <linux/quota.h>
9 #include <linux/quotaops.h>
10 #include <linux/dqblk_qtree.h>
11 #include <linux/jiffies.h>
12 #include <linux/writeback.h>
13 #include <linux/workqueue.h>
14 #include <linux/llist.h>
15 #include <linux/iversion.h>
17 #include <cluster/masklog.h>
22 #include "blockcheck.h"
30 #include "buffer_head_io.h"
32 #include "ocfs2_trace.h"
35 * Locking of quotas with OCFS2 is rather complex. Here are rules that
36 * should be obeyed by all the functions:
37 * - any write of quota structure (either to local or global file) is protected
38 * by dqio_sem or dquot->dq_lock.
39 * - any modification of global quota file holds inode cluster lock, i_mutex,
40 * and ip_alloc_sem of the global quota file (achieved by
41 * ocfs2_lock_global_qf). It also has to hold qinfo_lock.
42 * - an allocation of new blocks for local quota file is protected by
45 * A rough sketch of locking dependencies (lf = local file, gf = global file):
46 * Normal filesystem operation:
47 * start_trans -> dqio_sem -> write to lf
48 * Syncing of local and global file:
49 * ocfs2_lock_global_qf -> start_trans -> dqio_sem -> qinfo_lock ->
52 * Acquire dquot for the first time:
53 * dq_lock -> ocfs2_lock_global_qf -> qinfo_lock -> read from gf
54 * -> alloc space for gf
55 * -> start_trans -> qinfo_lock -> write to gf
56 * -> ip_alloc_sem of lf -> alloc space for lf
58 * Release last reference to dquot:
59 * dq_lock -> ocfs2_lock_global_qf -> start_trans -> qinfo_lock -> write to gf
61 * Note that all the above operations also hold the inode cluster lock of lf.
63 * inode cluster lock of recovered lf
64 * -> read bitmaps -> ip_alloc_sem of lf
65 * -> ocfs2_lock_global_qf -> start_trans -> dqio_sem -> qinfo_lock ->
69 static void qsync_work_fn(struct work_struct *work);
71 static void ocfs2_global_disk2memdqb(struct dquot *dquot, void *dp)
73 struct ocfs2_global_disk_dqblk *d = dp;
74 struct mem_dqblk *m = &dquot->dq_dqb;
76 /* Update from disk only entries not set by the admin */
77 if (!test_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags)) {
78 m->dqb_ihardlimit = le64_to_cpu(d->dqb_ihardlimit);
79 m->dqb_isoftlimit = le64_to_cpu(d->dqb_isoftlimit);
81 if (!test_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags))
82 m->dqb_curinodes = le64_to_cpu(d->dqb_curinodes);
83 if (!test_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags)) {
84 m->dqb_bhardlimit = le64_to_cpu(d->dqb_bhardlimit);
85 m->dqb_bsoftlimit = le64_to_cpu(d->dqb_bsoftlimit);
87 if (!test_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags))
88 m->dqb_curspace = le64_to_cpu(d->dqb_curspace);
89 if (!test_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags))
90 m->dqb_btime = le64_to_cpu(d->dqb_btime);
91 if (!test_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags))
92 m->dqb_itime = le64_to_cpu(d->dqb_itime);
93 OCFS2_DQUOT(dquot)->dq_use_count = le32_to_cpu(d->dqb_use_count);
96 static void ocfs2_global_mem2diskdqb(void *dp, struct dquot *dquot)
98 struct ocfs2_global_disk_dqblk *d = dp;
99 struct mem_dqblk *m = &dquot->dq_dqb;
101 d->dqb_id = cpu_to_le32(from_kqid(&init_user_ns, dquot->dq_id));
102 d->dqb_use_count = cpu_to_le32(OCFS2_DQUOT(dquot)->dq_use_count);
103 d->dqb_ihardlimit = cpu_to_le64(m->dqb_ihardlimit);
104 d->dqb_isoftlimit = cpu_to_le64(m->dqb_isoftlimit);
105 d->dqb_curinodes = cpu_to_le64(m->dqb_curinodes);
106 d->dqb_bhardlimit = cpu_to_le64(m->dqb_bhardlimit);
107 d->dqb_bsoftlimit = cpu_to_le64(m->dqb_bsoftlimit);
108 d->dqb_curspace = cpu_to_le64(m->dqb_curspace);
109 d->dqb_btime = cpu_to_le64(m->dqb_btime);
110 d->dqb_itime = cpu_to_le64(m->dqb_itime);
111 d->dqb_pad1 = d->dqb_pad2 = 0;
114 static int ocfs2_global_is_id(void *dp, struct dquot *dquot)
116 struct ocfs2_global_disk_dqblk *d = dp;
117 struct ocfs2_mem_dqinfo *oinfo =
118 sb_dqinfo(dquot->dq_sb, dquot->dq_id.type)->dqi_priv;
120 if (qtree_entry_unused(&oinfo->dqi_gi, dp))
123 return qid_eq(make_kqid(&init_user_ns, dquot->dq_id.type,
124 le32_to_cpu(d->dqb_id)),
128 const struct qtree_fmt_operations ocfs2_global_ops = {
129 .mem2disk_dqblk = ocfs2_global_mem2diskdqb,
130 .disk2mem_dqblk = ocfs2_global_disk2memdqb,
131 .is_id = ocfs2_global_is_id,
134 int ocfs2_validate_quota_block(struct super_block *sb, struct buffer_head *bh)
136 struct ocfs2_disk_dqtrailer *dqt =
137 ocfs2_block_dqtrailer(sb->s_blocksize, bh->b_data);
139 trace_ocfs2_validate_quota_block((unsigned long long)bh->b_blocknr);
141 BUG_ON(!buffer_uptodate(bh));
144 * If the ecc fails, we return the error but otherwise
145 * leave the filesystem running. We know any error is
146 * local to this block.
148 return ocfs2_validate_meta_ecc(sb, bh->b_data, &dqt->dq_check);
151 int ocfs2_read_quota_phys_block(struct inode *inode, u64 p_block,
152 struct buffer_head **bhp)
157 rc = ocfs2_read_blocks(INODE_CACHE(inode), p_block, 1, bhp, 0,
158 ocfs2_validate_quota_block);
164 /* Read data from global quotafile - avoid pagecache and such because we cannot
165 * afford acquiring the locks... We use quota cluster lock to serialize
166 * operations. Caller is responsible for acquiring it. */
167 ssize_t ocfs2_quota_read(struct super_block *sb, int type, char *data,
168 size_t len, loff_t off)
170 struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
171 struct inode *gqinode = oinfo->dqi_gqinode;
172 loff_t i_size = i_size_read(gqinode);
173 int offset = off & (sb->s_blocksize - 1);
174 sector_t blk = off >> sb->s_blocksize_bits;
176 struct buffer_head *bh;
177 size_t toread, tocopy;
178 u64 pblock = 0, pcount = 0;
182 if (off + len > i_size)
186 tocopy = min_t(size_t, (sb->s_blocksize - offset), toread);
188 err = ocfs2_extent_map_get_blocks(gqinode, blk, &pblock,
199 err = ocfs2_read_quota_phys_block(gqinode, pblock, &bh);
204 memcpy(data, bh->b_data + offset, tocopy);
214 /* Write to quotafile (we know the transaction is already started and has
216 ssize_t ocfs2_quota_write(struct super_block *sb, int type,
217 const char *data, size_t len, loff_t off)
219 struct mem_dqinfo *info = sb_dqinfo(sb, type);
220 struct ocfs2_mem_dqinfo *oinfo = info->dqi_priv;
221 struct inode *gqinode = oinfo->dqi_gqinode;
222 int offset = off & (sb->s_blocksize - 1);
223 sector_t blk = off >> sb->s_blocksize_bits;
224 int err = 0, new = 0, ja_type;
225 struct buffer_head *bh = NULL;
226 handle_t *handle = journal_current_handle();
230 mlog(ML_ERROR, "Quota write (off=%llu, len=%llu) cancelled "
231 "because transaction was not started.\n",
232 (unsigned long long)off, (unsigned long long)len);
235 if (len > sb->s_blocksize - OCFS2_QBLK_RESERVED_SPACE - offset) {
237 len = sb->s_blocksize - OCFS2_QBLK_RESERVED_SPACE - offset;
240 if (i_size_read(gqinode) < off + len) {
242 ocfs2_align_bytes_to_blocks(sb, off + len);
244 /* Space is already allocated in ocfs2_acquire_dquot() */
245 err = ocfs2_simple_size_update(gqinode,
252 err = ocfs2_extent_map_get_blocks(gqinode, blk, &pblock, &pcount, NULL);
257 /* Not rewriting whole block? */
258 if ((offset || len < sb->s_blocksize - OCFS2_QBLK_RESERVED_SPACE) &&
260 err = ocfs2_read_quota_phys_block(gqinode, pblock, &bh);
261 ja_type = OCFS2_JOURNAL_ACCESS_WRITE;
263 bh = sb_getblk(sb, pblock);
266 ja_type = OCFS2_JOURNAL_ACCESS_CREATE;
274 memset(bh->b_data, 0, sb->s_blocksize);
275 memcpy(bh->b_data + offset, data, len);
276 flush_dcache_page(bh->b_page);
277 set_buffer_uptodate(bh);
279 ocfs2_set_buffer_uptodate(INODE_CACHE(gqinode), bh);
280 err = ocfs2_journal_access_dq(handle, INODE_CACHE(gqinode), bh,
286 ocfs2_journal_dirty(handle, bh);
293 inode_inc_iversion(gqinode);
294 ocfs2_mark_inode_dirty(handle, gqinode, oinfo->dqi_gqi_bh);
298 int ocfs2_lock_global_qf(struct ocfs2_mem_dqinfo *oinfo, int ex)
301 struct buffer_head *bh = NULL;
303 status = ocfs2_inode_lock(oinfo->dqi_gqinode, &bh, ex);
306 spin_lock(&dq_data_lock);
307 if (!oinfo->dqi_gqi_count++)
308 oinfo->dqi_gqi_bh = bh;
310 WARN_ON(bh != oinfo->dqi_gqi_bh);
311 spin_unlock(&dq_data_lock);
313 inode_lock(oinfo->dqi_gqinode);
314 down_write(&OCFS2_I(oinfo->dqi_gqinode)->ip_alloc_sem);
316 down_read(&OCFS2_I(oinfo->dqi_gqinode)->ip_alloc_sem);
321 void ocfs2_unlock_global_qf(struct ocfs2_mem_dqinfo *oinfo, int ex)
324 up_write(&OCFS2_I(oinfo->dqi_gqinode)->ip_alloc_sem);
325 inode_unlock(oinfo->dqi_gqinode);
327 up_read(&OCFS2_I(oinfo->dqi_gqinode)->ip_alloc_sem);
329 ocfs2_inode_unlock(oinfo->dqi_gqinode, ex);
330 brelse(oinfo->dqi_gqi_bh);
331 spin_lock(&dq_data_lock);
332 if (!--oinfo->dqi_gqi_count)
333 oinfo->dqi_gqi_bh = NULL;
334 spin_unlock(&dq_data_lock);
337 /* Read information header from global quota file */
338 int ocfs2_global_read_info(struct super_block *sb, int type)
340 struct inode *gqinode = NULL;
341 unsigned int ino[OCFS2_MAXQUOTAS] = { USER_QUOTA_SYSTEM_INODE,
342 GROUP_QUOTA_SYSTEM_INODE };
343 struct ocfs2_global_disk_dqinfo dinfo;
344 struct mem_dqinfo *info = sb_dqinfo(sb, type);
345 struct ocfs2_mem_dqinfo *oinfo = info->dqi_priv;
349 /* Read global header */
350 gqinode = ocfs2_get_system_file_inode(OCFS2_SB(sb), ino[type],
353 mlog(ML_ERROR, "failed to get global quota inode (type=%d)\n",
358 oinfo->dqi_gi.dqi_sb = sb;
359 oinfo->dqi_gi.dqi_type = type;
360 oinfo->dqi_gi.dqi_entry_size = sizeof(struct ocfs2_global_disk_dqblk);
361 oinfo->dqi_gi.dqi_ops = &ocfs2_global_ops;
362 oinfo->dqi_gqi_bh = NULL;
363 oinfo->dqi_gqi_count = 0;
364 oinfo->dqi_gqinode = gqinode;
365 status = ocfs2_lock_global_qf(oinfo, 0);
371 status = ocfs2_extent_map_get_blocks(gqinode, 0, &oinfo->dqi_giblk,
376 status = ocfs2_qinfo_lock(oinfo, 0);
379 status = sb->s_op->quota_read(sb, type, (char *)&dinfo,
380 sizeof(struct ocfs2_global_disk_dqinfo),
381 OCFS2_GLOBAL_INFO_OFF);
382 ocfs2_qinfo_unlock(oinfo, 0);
383 ocfs2_unlock_global_qf(oinfo, 0);
384 if (status != sizeof(struct ocfs2_global_disk_dqinfo)) {
385 mlog(ML_ERROR, "Cannot read global quota info (%d).\n",
392 info->dqi_bgrace = le32_to_cpu(dinfo.dqi_bgrace);
393 info->dqi_igrace = le32_to_cpu(dinfo.dqi_igrace);
394 oinfo->dqi_syncms = le32_to_cpu(dinfo.dqi_syncms);
395 oinfo->dqi_gi.dqi_blocks = le32_to_cpu(dinfo.dqi_blocks);
396 oinfo->dqi_gi.dqi_free_blk = le32_to_cpu(dinfo.dqi_free_blk);
397 oinfo->dqi_gi.dqi_free_entry = le32_to_cpu(dinfo.dqi_free_entry);
398 oinfo->dqi_gi.dqi_blocksize_bits = sb->s_blocksize_bits;
399 oinfo->dqi_gi.dqi_usable_bs = sb->s_blocksize -
400 OCFS2_QBLK_RESERVED_SPACE;
401 oinfo->dqi_gi.dqi_qtree_depth = qtree_depth(&oinfo->dqi_gi);
402 INIT_DELAYED_WORK(&oinfo->dqi_sync_work, qsync_work_fn);
403 schedule_delayed_work(&oinfo->dqi_sync_work,
404 msecs_to_jiffies(oinfo->dqi_syncms));
409 ocfs2_unlock_global_qf(oinfo, 0);
414 /* Write information to global quota file. Expects exlusive lock on quota
415 * file inode and quota info */
416 static int __ocfs2_global_write_info(struct super_block *sb, int type)
418 struct mem_dqinfo *info = sb_dqinfo(sb, type);
419 struct ocfs2_mem_dqinfo *oinfo = info->dqi_priv;
420 struct ocfs2_global_disk_dqinfo dinfo;
423 spin_lock(&dq_data_lock);
424 info->dqi_flags &= ~DQF_INFO_DIRTY;
425 dinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
426 dinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
427 spin_unlock(&dq_data_lock);
428 dinfo.dqi_syncms = cpu_to_le32(oinfo->dqi_syncms);
429 dinfo.dqi_blocks = cpu_to_le32(oinfo->dqi_gi.dqi_blocks);
430 dinfo.dqi_free_blk = cpu_to_le32(oinfo->dqi_gi.dqi_free_blk);
431 dinfo.dqi_free_entry = cpu_to_le32(oinfo->dqi_gi.dqi_free_entry);
432 size = sb->s_op->quota_write(sb, type, (char *)&dinfo,
433 sizeof(struct ocfs2_global_disk_dqinfo),
434 OCFS2_GLOBAL_INFO_OFF);
435 if (size != sizeof(struct ocfs2_global_disk_dqinfo)) {
436 mlog(ML_ERROR, "Cannot write global quota info structure\n");
444 int ocfs2_global_write_info(struct super_block *sb, int type)
447 struct quota_info *dqopt = sb_dqopt(sb);
448 struct ocfs2_mem_dqinfo *info = dqopt->info[type].dqi_priv;
450 down_write(&dqopt->dqio_sem);
451 err = ocfs2_qinfo_lock(info, 1);
454 err = __ocfs2_global_write_info(sb, type);
455 ocfs2_qinfo_unlock(info, 1);
457 up_write(&dqopt->dqio_sem);
461 static int ocfs2_global_qinit_alloc(struct super_block *sb, int type)
463 struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
466 * We may need to allocate tree blocks and a leaf block but not the
469 return oinfo->dqi_gi.dqi_qtree_depth;
472 static int ocfs2_calc_global_qinit_credits(struct super_block *sb, int type)
474 /* We modify all the allocated blocks, tree root, info block and
476 return (ocfs2_global_qinit_alloc(sb, type) + 2) *
477 OCFS2_QUOTA_BLOCK_UPDATE_CREDITS + 1;
480 /* Sync local information about quota modifications with global quota file.
481 * Caller must have started the transaction and obtained exclusive lock for
482 * global quota file inode */
483 int __ocfs2_sync_dquot(struct dquot *dquot, int freeing)
486 struct super_block *sb = dquot->dq_sb;
487 int type = dquot->dq_id.type;
488 struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
489 struct ocfs2_global_disk_dqblk dqblk;
490 s64 spacechange, inodechange;
491 time64_t olditime, oldbtime;
493 err = sb->s_op->quota_read(sb, type, (char *)&dqblk,
494 sizeof(struct ocfs2_global_disk_dqblk),
496 if (err != sizeof(struct ocfs2_global_disk_dqblk)) {
498 mlog(ML_ERROR, "Short read from global quota file "
505 /* Update space and inode usage. Get also other information from
506 * global quota file so that we don't overwrite any changes there.
508 spin_lock(&dquot->dq_dqb_lock);
509 spacechange = dquot->dq_dqb.dqb_curspace -
510 OCFS2_DQUOT(dquot)->dq_origspace;
511 inodechange = dquot->dq_dqb.dqb_curinodes -
512 OCFS2_DQUOT(dquot)->dq_originodes;
513 olditime = dquot->dq_dqb.dqb_itime;
514 oldbtime = dquot->dq_dqb.dqb_btime;
515 ocfs2_global_disk2memdqb(dquot, &dqblk);
516 trace_ocfs2_sync_dquot(from_kqid(&init_user_ns, dquot->dq_id),
517 dquot->dq_dqb.dqb_curspace,
518 (long long)spacechange,
519 dquot->dq_dqb.dqb_curinodes,
520 (long long)inodechange);
521 if (!test_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags))
522 dquot->dq_dqb.dqb_curspace += spacechange;
523 if (!test_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags))
524 dquot->dq_dqb.dqb_curinodes += inodechange;
525 /* Set properly space grace time... */
526 if (dquot->dq_dqb.dqb_bsoftlimit &&
527 dquot->dq_dqb.dqb_curspace > dquot->dq_dqb.dqb_bsoftlimit) {
528 if (!test_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags) &&
530 if (dquot->dq_dqb.dqb_btime > 0)
531 dquot->dq_dqb.dqb_btime =
532 min(dquot->dq_dqb.dqb_btime, oldbtime);
534 dquot->dq_dqb.dqb_btime = oldbtime;
537 dquot->dq_dqb.dqb_btime = 0;
538 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
540 /* Set properly inode grace time... */
541 if (dquot->dq_dqb.dqb_isoftlimit &&
542 dquot->dq_dqb.dqb_curinodes > dquot->dq_dqb.dqb_isoftlimit) {
543 if (!test_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags) &&
545 if (dquot->dq_dqb.dqb_itime > 0)
546 dquot->dq_dqb.dqb_itime =
547 min(dquot->dq_dqb.dqb_itime, olditime);
549 dquot->dq_dqb.dqb_itime = olditime;
552 dquot->dq_dqb.dqb_itime = 0;
553 clear_bit(DQ_INODES_B, &dquot->dq_flags);
555 /* All information is properly updated, clear the flags */
556 __clear_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
557 __clear_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
558 __clear_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
559 __clear_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
560 __clear_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
561 __clear_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
562 OCFS2_DQUOT(dquot)->dq_origspace = dquot->dq_dqb.dqb_curspace;
563 OCFS2_DQUOT(dquot)->dq_originodes = dquot->dq_dqb.dqb_curinodes;
564 spin_unlock(&dquot->dq_dqb_lock);
565 err = ocfs2_qinfo_lock(info, freeing);
567 mlog(ML_ERROR, "Failed to lock quota info, losing quota write"
568 " (type=%d, id=%u)\n", dquot->dq_id.type,
569 (unsigned)from_kqid(&init_user_ns, dquot->dq_id));
573 OCFS2_DQUOT(dquot)->dq_use_count--;
574 err = qtree_write_dquot(&info->dqi_gi, dquot);
577 if (freeing && !OCFS2_DQUOT(dquot)->dq_use_count) {
578 err = qtree_release_dquot(&info->dqi_gi, dquot);
579 if (info_dirty(sb_dqinfo(sb, type))) {
580 err2 = __ocfs2_global_write_info(sb, type);
586 ocfs2_qinfo_unlock(info, freeing);
594 * Functions for periodic syncing of dquots with global file
596 static int ocfs2_sync_dquot_helper(struct dquot *dquot, unsigned long type)
599 struct super_block *sb = dquot->dq_sb;
600 struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
601 struct ocfs2_super *osb = OCFS2_SB(sb);
604 trace_ocfs2_sync_dquot_helper(from_kqid(&init_user_ns, dquot->dq_id),
607 if (type != dquot->dq_id.type)
609 status = ocfs2_lock_global_qf(oinfo, 1);
613 handle = ocfs2_start_trans(osb, OCFS2_QSYNC_CREDITS);
614 if (IS_ERR(handle)) {
615 status = PTR_ERR(handle);
619 down_write(&sb_dqopt(sb)->dqio_sem);
620 status = ocfs2_sync_dquot(dquot);
623 /* We have to write local structure as well... */
624 status = ocfs2_local_write_dquot(dquot);
627 up_write(&sb_dqopt(sb)->dqio_sem);
628 ocfs2_commit_trans(osb, handle);
630 ocfs2_unlock_global_qf(oinfo, 1);
635 static void qsync_work_fn(struct work_struct *work)
637 struct ocfs2_mem_dqinfo *oinfo = container_of(work,
638 struct ocfs2_mem_dqinfo,
640 struct super_block *sb = oinfo->dqi_gqinode->i_sb;
643 * We have to be careful here not to deadlock on s_umount as umount
644 * disabling quotas may be in progress and it waits for this work to
645 * complete. If trylock fails, we'll do the sync next time...
647 if (down_read_trylock(&sb->s_umount)) {
648 dquot_scan_active(sb, ocfs2_sync_dquot_helper, oinfo->dqi_type);
649 up_read(&sb->s_umount);
651 schedule_delayed_work(&oinfo->dqi_sync_work,
652 msecs_to_jiffies(oinfo->dqi_syncms));
656 * Wrappers for generic quota functions
659 static int ocfs2_write_dquot(struct dquot *dquot)
662 struct ocfs2_super *osb = OCFS2_SB(dquot->dq_sb);
665 trace_ocfs2_write_dquot(from_kqid(&init_user_ns, dquot->dq_id),
668 handle = ocfs2_start_trans(osb, OCFS2_QWRITE_CREDITS);
669 if (IS_ERR(handle)) {
670 status = PTR_ERR(handle);
674 down_write(&sb_dqopt(dquot->dq_sb)->dqio_sem);
675 status = ocfs2_local_write_dquot(dquot);
676 up_write(&sb_dqopt(dquot->dq_sb)->dqio_sem);
677 ocfs2_commit_trans(osb, handle);
682 static int ocfs2_calc_qdel_credits(struct super_block *sb, int type)
684 struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
686 * We modify tree, leaf block, global info, local chunk header,
687 * global and local inode; OCFS2_QINFO_WRITE_CREDITS already
688 * accounts for inode update
690 return (oinfo->dqi_gi.dqi_qtree_depth + 2) *
691 OCFS2_QUOTA_BLOCK_UPDATE_CREDITS +
692 OCFS2_QINFO_WRITE_CREDITS +
693 OCFS2_INODE_UPDATE_CREDITS;
696 void ocfs2_drop_dquot_refs(struct work_struct *work)
698 struct ocfs2_super *osb = container_of(work, struct ocfs2_super,
700 struct llist_node *list;
701 struct ocfs2_dquot *odquot, *next_odquot;
703 list = llist_del_all(&osb->dquot_drop_list);
704 llist_for_each_entry_safe(odquot, next_odquot, list, list) {
705 /* Drop the reference we acquired in ocfs2_dquot_release() */
706 dqput(&odquot->dq_dquot);
711 * Called when the last reference to dquot is dropped. If we are called from
712 * downconvert thread, we cannot do all the handling here because grabbing
713 * quota lock could deadlock (the node holding the quota lock could need some
714 * other cluster lock to proceed but with blocked downconvert thread we cannot
717 static int ocfs2_release_dquot(struct dquot *dquot)
720 struct ocfs2_mem_dqinfo *oinfo =
721 sb_dqinfo(dquot->dq_sb, dquot->dq_id.type)->dqi_priv;
722 struct ocfs2_super *osb = OCFS2_SB(dquot->dq_sb);
725 trace_ocfs2_release_dquot(from_kqid(&init_user_ns, dquot->dq_id),
728 mutex_lock(&dquot->dq_lock);
729 /* Check whether we are not racing with some other dqget() */
730 if (dquot_is_busy(dquot))
732 /* Running from downconvert thread? Postpone quota processing to wq */
733 if (current == osb->dc_task) {
735 * Grab our own reference to dquot and queue it for delayed
736 * dropping. Quota code rechecks after calling
737 * ->release_dquot() and won't free dquot structure.
740 /* First entry on list -> queue work */
741 if (llist_add(&OCFS2_DQUOT(dquot)->list, &osb->dquot_drop_list))
742 queue_work(osb->ocfs2_wq, &osb->dquot_drop_work);
745 status = ocfs2_lock_global_qf(oinfo, 1);
748 handle = ocfs2_start_trans(osb,
749 ocfs2_calc_qdel_credits(dquot->dq_sb, dquot->dq_id.type));
750 if (IS_ERR(handle)) {
751 status = PTR_ERR(handle);
756 status = ocfs2_global_release_dquot(dquot);
761 status = ocfs2_local_release_dquot(handle, dquot);
763 * If we fail here, we cannot do much as global structure is
764 * already released. So just complain...
769 * Clear dq_off so that we search for the structure in quota file next
770 * time we acquire it. The structure might be deleted and reallocated
771 * elsewhere by another node while our dquot structure is on freelist.
774 clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
776 ocfs2_commit_trans(osb, handle);
778 ocfs2_unlock_global_qf(oinfo, 1);
780 mutex_unlock(&dquot->dq_lock);
787 * Read global dquot structure from disk or create it if it does
788 * not exist. Also update use count of the global structure and
789 * create structure in node-local quota file.
791 static int ocfs2_acquire_dquot(struct dquot *dquot)
795 struct super_block *sb = dquot->dq_sb;
796 struct ocfs2_super *osb = OCFS2_SB(sb);
797 int type = dquot->dq_id.type;
798 struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
799 struct inode *gqinode = info->dqi_gqinode;
800 int need_alloc = ocfs2_global_qinit_alloc(sb, type);
803 trace_ocfs2_acquire_dquot(from_kqid(&init_user_ns, dquot->dq_id),
805 mutex_lock(&dquot->dq_lock);
807 * We need an exclusive lock, because we're going to update use count
808 * and instantiate possibly new dquot structure
810 status = ocfs2_lock_global_qf(info, 1);
813 status = ocfs2_qinfo_lock(info, 0);
817 * We always want to read dquot structure from disk because we don't
818 * know what happened with it while it was on freelist.
820 status = qtree_read_dquot(&info->dqi_gi, dquot);
821 ocfs2_qinfo_unlock(info, 0);
825 OCFS2_DQUOT(dquot)->dq_use_count++;
826 OCFS2_DQUOT(dquot)->dq_origspace = dquot->dq_dqb.dqb_curspace;
827 OCFS2_DQUOT(dquot)->dq_originodes = dquot->dq_dqb.dqb_curinodes;
828 if (!dquot->dq_off) { /* No real quota entry? */
831 * Add blocks to quota file before we start a transaction since
832 * locking allocators ranks above a transaction start
834 WARN_ON(journal_current_handle());
835 status = ocfs2_extend_no_holes(gqinode, NULL,
836 i_size_read(gqinode) + (need_alloc << sb->s_blocksize_bits),
837 i_size_read(gqinode));
842 handle = ocfs2_start_trans(osb,
843 ocfs2_calc_global_qinit_credits(sb, type));
844 if (IS_ERR(handle)) {
845 status = PTR_ERR(handle);
848 status = ocfs2_qinfo_lock(info, ex);
851 status = qtree_write_dquot(&info->dqi_gi, dquot);
852 if (ex && info_dirty(sb_dqinfo(sb, type))) {
853 err = __ocfs2_global_write_info(sb, type);
857 ocfs2_qinfo_unlock(info, ex);
859 ocfs2_commit_trans(osb, handle);
861 ocfs2_unlock_global_qf(info, 1);
865 status = ocfs2_create_local_dquot(dquot);
868 set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
870 mutex_unlock(&dquot->dq_lock);
876 static int ocfs2_get_next_id(struct super_block *sb, struct kqid *qid)
878 int type = qid->type;
879 struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
882 trace_ocfs2_get_next_id(from_kqid(&init_user_ns, *qid), type);
883 if (!sb_has_quota_loaded(sb, type)) {
887 status = ocfs2_lock_global_qf(info, 0);
890 status = ocfs2_qinfo_lock(info, 0);
893 status = qtree_get_next_id(&info->dqi_gi, qid);
894 ocfs2_qinfo_unlock(info, 0);
896 ocfs2_unlock_global_qf(info, 0);
899 * Avoid logging ENOENT since it just means there isn't next ID and
900 * ESRCH which means quota isn't enabled for the filesystem.
902 if (status && status != -ENOENT && status != -ESRCH)
907 static int ocfs2_mark_dquot_dirty(struct dquot *dquot)
909 unsigned long mask = (1 << (DQ_LASTSET_B + QIF_ILIMITS_B)) |
910 (1 << (DQ_LASTSET_B + QIF_BLIMITS_B)) |
911 (1 << (DQ_LASTSET_B + QIF_INODES_B)) |
912 (1 << (DQ_LASTSET_B + QIF_SPACE_B)) |
913 (1 << (DQ_LASTSET_B + QIF_BTIME_B)) |
914 (1 << (DQ_LASTSET_B + QIF_ITIME_B));
917 struct super_block *sb = dquot->dq_sb;
918 int type = dquot->dq_id.type;
919 struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
921 struct ocfs2_super *osb = OCFS2_SB(sb);
923 trace_ocfs2_mark_dquot_dirty(from_kqid(&init_user_ns, dquot->dq_id),
926 /* In case user set some limits, sync dquot immediately to global
927 * quota file so that information propagates quicker */
928 spin_lock(&dquot->dq_dqb_lock);
929 if (dquot->dq_flags & mask)
931 spin_unlock(&dquot->dq_dqb_lock);
932 /* This is a slight hack but we can't afford getting global quota
933 * lock if we already have a transaction started. */
934 if (!sync || journal_current_handle()) {
935 status = ocfs2_write_dquot(dquot);
938 status = ocfs2_lock_global_qf(oinfo, 1);
941 handle = ocfs2_start_trans(osb, OCFS2_QSYNC_CREDITS);
942 if (IS_ERR(handle)) {
943 status = PTR_ERR(handle);
947 down_write(&sb_dqopt(sb)->dqio_sem);
948 status = ocfs2_sync_dquot(dquot);
953 /* Now write updated local dquot structure */
954 status = ocfs2_local_write_dquot(dquot);
956 up_write(&sb_dqopt(sb)->dqio_sem);
957 ocfs2_commit_trans(osb, handle);
959 ocfs2_unlock_global_qf(oinfo, 1);
966 /* This should happen only after set_dqinfo(). */
967 static int ocfs2_write_info(struct super_block *sb, int type)
971 struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
973 status = ocfs2_lock_global_qf(oinfo, 1);
976 handle = ocfs2_start_trans(OCFS2_SB(sb), OCFS2_QINFO_WRITE_CREDITS);
977 if (IS_ERR(handle)) {
978 status = PTR_ERR(handle);
982 status = dquot_commit_info(sb, type);
983 ocfs2_commit_trans(OCFS2_SB(sb), handle);
985 ocfs2_unlock_global_qf(oinfo, 1);
992 static struct dquot *ocfs2_alloc_dquot(struct super_block *sb, int type)
994 struct ocfs2_dquot *dquot =
995 kmem_cache_zalloc(ocfs2_dquot_cachep, GFP_NOFS);
999 return &dquot->dq_dquot;
1002 static void ocfs2_destroy_dquot(struct dquot *dquot)
1004 kmem_cache_free(ocfs2_dquot_cachep, dquot);
1007 const struct dquot_operations ocfs2_quota_operations = {
1008 /* We never make dquot dirty so .write_dquot is never called */
1009 .acquire_dquot = ocfs2_acquire_dquot,
1010 .release_dquot = ocfs2_release_dquot,
1011 .mark_dirty = ocfs2_mark_dquot_dirty,
1012 .write_info = ocfs2_write_info,
1013 .alloc_dquot = ocfs2_alloc_dquot,
1014 .destroy_dquot = ocfs2_destroy_dquot,
1015 .get_next_id = ocfs2_get_next_id,