1 // SPDX-License-Identifier: GPL-2.0
3 * Implementation of the diskquota system for the LINUX operating system. QUOTA
4 * is implemented using the BSD system call interface as the means of
5 * communication with the user level. This file contains the generic routines
6 * called by the different filesystems on allocation of an inode or block.
7 * These routines take care of the administration needed to have a consistent
8 * diskquota tracking system. The ideas of both user and group quotas are based
9 * on the Melbourne quota system as used on BSD derived systems. The internal
10 * implementation is based on one of the several variants of the LINUX
11 * inode-subsystem with added complexity of the diskquota system.
13 * Author: Marco van Wieringen <mvw@planets.elm.net>
15 * Fixes: Dmitry Gorodchanin <pgmdsg@ibi.com>, 11 Feb 96
17 * Revised list management to avoid races
18 * -- Bill Hawes, <whawes@star.net>, 9/98
20 * Fixed races in dquot_transfer(), dqget() and dquot_alloc_...().
21 * As the consequence the locking was moved from dquot_decr_...(),
22 * dquot_incr_...() to calling functions.
23 * invalidate_dquots() now writes modified dquots.
24 * Serialized quota_off() and quota_on() for mount point.
25 * Fixed a few bugs in grow_dquots().
26 * Fixed deadlock in write_dquot() - we no longer account quotas on
28 * remove_dquot_ref() moved to inode.c - it now traverses through inodes
29 * add_dquot_ref() restarts after blocking
30 * Added check for bogus uid and fixed check for group in quotactl.
31 * Jan Kara, <jack@suse.cz>, sponsored by SuSE CR, 10-11/99
33 * Used struct list_head instead of own list struct
34 * Invalidation of referenced dquots is no longer possible
35 * Improved free_dquots list management
36 * Quota and i_blocks are now updated in one place to avoid races
37 * Warnings are now delayed so we won't block in critical section
38 * Write updated not to require dquot lock
39 * Jan Kara, <jack@suse.cz>, 9/2000
41 * Added dynamic quota structure allocation
42 * Jan Kara <jack@suse.cz> 12/2000
44 * Rewritten quota interface. Implemented new quota format and
45 * formats registering.
46 * Jan Kara, <jack@suse.cz>, 2001,2002
49 * Jan Kara, <jack@suse.cz>, 10/2002
51 * Added journalled quota support, fix lock inversion problems
52 * Jan Kara, <jack@suse.cz>, 2003,2004
54 * (C) Copyright 1994 - 1997 Marco van Wieringen
57 #include <linux/errno.h>
58 #include <linux/kernel.h>
60 #include <linux/mount.h>
62 #include <linux/time.h>
63 #include <linux/types.h>
64 #include <linux/string.h>
65 #include <linux/fcntl.h>
66 #include <linux/stat.h>
67 #include <linux/tty.h>
68 #include <linux/file.h>
69 #include <linux/slab.h>
70 #include <linux/sysctl.h>
71 #include <linux/init.h>
72 #include <linux/module.h>
73 #include <linux/proc_fs.h>
74 #include <linux/security.h>
75 #include <linux/sched.h>
76 #include <linux/cred.h>
77 #include <linux/kmod.h>
78 #include <linux/namei.h>
79 #include <linux/capability.h>
80 #include <linux/quotaops.h>
81 #include "../internal.h" /* ugh */
83 #include <linux/uaccess.h>
86 * There are five quota SMP locks:
87 * * dq_list_lock protects all lists with quotas and quota formats.
88 * * dquot->dq_dqb_lock protects data from dq_dqb
89 * * inode->i_lock protects inode->i_blocks, i_bytes and also guards
90 * consistency of dquot->dq_dqb with inode->i_blocks, i_bytes so that
91 * dquot_transfer() can stabilize amount it transfers
92 * * dq_data_lock protects mem_dqinfo structures and modifications of dquot
93 * pointers in the inode
94 * * dq_state_lock protects modifications of quota state (on quotaon and
95 * quotaoff) and readers who care about latest values take it as well.
97 * The spinlock ordering is hence:
98 * dq_data_lock > dq_list_lock > i_lock > dquot->dq_dqb_lock,
99 * dq_list_lock > dq_state_lock
101 * Note that some things (eg. sb pointer, type, id) doesn't change during
102 * the life of the dquot structure and so needn't to be protected by a lock
104 * Operation accessing dquots via inode pointers are protected by dquot_srcu.
105 * Operation of reading pointer needs srcu_read_lock(&dquot_srcu), and
106 * synchronize_srcu(&dquot_srcu) is called after clearing pointers from
107 * inode and before dropping dquot references to avoid use of dquots after
108 * they are freed. dq_data_lock is used to serialize the pointer setting and
109 * clearing operations.
110 * Special care needs to be taken about S_NOQUOTA inode flag (marking that
111 * inode is a quota file). Functions adding pointers from inode to dquots have
112 * to check this flag under dq_data_lock and then (if S_NOQUOTA is not set) they
113 * have to do all pointer modifications before dropping dq_data_lock. This makes
114 * sure they cannot race with quotaon which first sets S_NOQUOTA flag and
115 * then drops all pointers to dquots from an inode.
117 * Each dquot has its dq_lock mutex. Dquot is locked when it is being read to
118 * memory (or space for it is being allocated) on the first dqget(), when it is
119 * being written out, and when it is being released on the last dqput(). The
120 * allocation and release operations are serialized by the dq_lock and by
121 * checking the use count in dquot_release().
123 * Lock ordering (including related VFS locks) is the following:
124 * s_umount > i_mutex > journal_lock > dquot->dq_lock > dqio_sem
127 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_list_lock);
128 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_state_lock);
129 __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_data_lock);
130 EXPORT_SYMBOL(dq_data_lock);
131 DEFINE_STATIC_SRCU(dquot_srcu);
133 static DECLARE_WAIT_QUEUE_HEAD(dquot_ref_wq);
135 void __quota_error(struct super_block *sb, const char *func,
136 const char *fmt, ...)
138 if (printk_ratelimit()) {
140 struct va_format vaf;
147 printk(KERN_ERR "Quota error (device %s): %s: %pV\n",
148 sb->s_id, func, &vaf);
153 EXPORT_SYMBOL(__quota_error);
155 #if defined(CONFIG_QUOTA_DEBUG) || defined(CONFIG_PRINT_QUOTA_WARNING)
156 static char *quotatypes[] = INITQFNAMES;
158 static struct quota_format_type *quota_formats; /* List of registered formats */
159 static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
161 /* SLAB cache for dquot structures */
162 static struct kmem_cache *dquot_cachep;
164 int register_quota_format(struct quota_format_type *fmt)
166 spin_lock(&dq_list_lock);
167 fmt->qf_next = quota_formats;
169 spin_unlock(&dq_list_lock);
172 EXPORT_SYMBOL(register_quota_format);
174 void unregister_quota_format(struct quota_format_type *fmt)
176 struct quota_format_type **actqf;
178 spin_lock(&dq_list_lock);
179 for (actqf = "a_formats; *actqf && *actqf != fmt;
180 actqf = &(*actqf)->qf_next)
183 *actqf = (*actqf)->qf_next;
184 spin_unlock(&dq_list_lock);
186 EXPORT_SYMBOL(unregister_quota_format);
188 static struct quota_format_type *find_quota_format(int id)
190 struct quota_format_type *actqf;
192 spin_lock(&dq_list_lock);
193 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
194 actqf = actqf->qf_next)
196 if (!actqf || !try_module_get(actqf->qf_owner)) {
199 spin_unlock(&dq_list_lock);
201 for (qm = 0; module_names[qm].qm_fmt_id &&
202 module_names[qm].qm_fmt_id != id; qm++)
204 if (!module_names[qm].qm_fmt_id ||
205 request_module(module_names[qm].qm_mod_name))
208 spin_lock(&dq_list_lock);
209 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
210 actqf = actqf->qf_next)
212 if (actqf && !try_module_get(actqf->qf_owner))
215 spin_unlock(&dq_list_lock);
219 static void put_quota_format(struct quota_format_type *fmt)
221 module_put(fmt->qf_owner);
225 * Dquot List Management:
226 * The quota code uses three lists for dquot management: the inuse_list,
227 * free_dquots, and dquot_hash[] array. A single dquot structure may be
228 * on all three lists, depending on its current state.
230 * All dquots are placed to the end of inuse_list when first created, and this
231 * list is used for invalidate operation, which must look at every dquot.
233 * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
234 * and this list is searched whenever we need an available dquot. Dquots are
235 * removed from the list as soon as they are used again, and
236 * dqstats.free_dquots gives the number of dquots on the list. When
237 * dquot is invalidated it's completely released from memory.
239 * Dquots with a specific identity (device, type and id) are placed on
240 * one of the dquot_hash[] hash chains. The provides an efficient search
241 * mechanism to locate a specific dquot.
244 static LIST_HEAD(inuse_list);
245 static LIST_HEAD(free_dquots);
246 static unsigned int dq_hash_bits, dq_hash_mask;
247 static struct hlist_head *dquot_hash;
249 struct dqstats dqstats;
250 EXPORT_SYMBOL(dqstats);
252 static qsize_t inode_get_rsv_space(struct inode *inode);
253 static qsize_t __inode_get_rsv_space(struct inode *inode);
254 static int __dquot_initialize(struct inode *inode, int type);
256 static inline unsigned int
257 hashfn(const struct super_block *sb, struct kqid qid)
259 unsigned int id = from_kqid(&init_user_ns, qid);
263 tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
264 return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
268 * Following list functions expect dq_list_lock to be held
270 static inline void insert_dquot_hash(struct dquot *dquot)
272 struct hlist_head *head;
273 head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id);
274 hlist_add_head(&dquot->dq_hash, head);
277 static inline void remove_dquot_hash(struct dquot *dquot)
279 hlist_del_init(&dquot->dq_hash);
282 static struct dquot *find_dquot(unsigned int hashent, struct super_block *sb,
285 struct hlist_node *node;
288 hlist_for_each (node, dquot_hash+hashent) {
289 dquot = hlist_entry(node, struct dquot, dq_hash);
290 if (dquot->dq_sb == sb && qid_eq(dquot->dq_id, qid))
296 /* Add a dquot to the tail of the free list */
297 static inline void put_dquot_last(struct dquot *dquot)
299 list_add_tail(&dquot->dq_free, &free_dquots);
300 dqstats_inc(DQST_FREE_DQUOTS);
303 static inline void remove_free_dquot(struct dquot *dquot)
305 if (list_empty(&dquot->dq_free))
307 list_del_init(&dquot->dq_free);
308 dqstats_dec(DQST_FREE_DQUOTS);
311 static inline void put_inuse(struct dquot *dquot)
313 /* We add to the back of inuse list so we don't have to restart
314 * when traversing this list and we block */
315 list_add_tail(&dquot->dq_inuse, &inuse_list);
316 dqstats_inc(DQST_ALLOC_DQUOTS);
319 static inline void remove_inuse(struct dquot *dquot)
321 dqstats_dec(DQST_ALLOC_DQUOTS);
322 list_del(&dquot->dq_inuse);
325 * End of list functions needing dq_list_lock
328 static void wait_on_dquot(struct dquot *dquot)
330 mutex_lock(&dquot->dq_lock);
331 mutex_unlock(&dquot->dq_lock);
334 static inline int dquot_dirty(struct dquot *dquot)
336 return test_bit(DQ_MOD_B, &dquot->dq_flags);
339 static inline int mark_dquot_dirty(struct dquot *dquot)
341 return dquot->dq_sb->dq_op->mark_dirty(dquot);
344 /* Mark dquot dirty in atomic manner, and return it's old dirty flag state */
345 int dquot_mark_dquot_dirty(struct dquot *dquot)
349 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
352 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NOLIST_DIRTY)
353 return test_and_set_bit(DQ_MOD_B, &dquot->dq_flags);
355 /* If quota is dirty already, we don't have to acquire dq_list_lock */
356 if (test_bit(DQ_MOD_B, &dquot->dq_flags))
359 spin_lock(&dq_list_lock);
360 if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags)) {
361 list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
362 info[dquot->dq_id.type].dqi_dirty_list);
365 spin_unlock(&dq_list_lock);
368 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
370 /* Dirtify all the dquots - this can block when journalling */
371 static inline int mark_all_dquot_dirty(struct dquot * const *dquot)
376 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
378 /* Even in case of error we have to continue */
379 ret = mark_dquot_dirty(dquot[cnt]);
386 static inline void dqput_all(struct dquot **dquot)
390 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
394 static inline int clear_dquot_dirty(struct dquot *dquot)
396 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NOLIST_DIRTY)
397 return test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags);
399 spin_lock(&dq_list_lock);
400 if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags)) {
401 spin_unlock(&dq_list_lock);
404 list_del_init(&dquot->dq_dirty);
405 spin_unlock(&dq_list_lock);
409 void mark_info_dirty(struct super_block *sb, int type)
411 spin_lock(&dq_data_lock);
412 sb_dqopt(sb)->info[type].dqi_flags |= DQF_INFO_DIRTY;
413 spin_unlock(&dq_data_lock);
415 EXPORT_SYMBOL(mark_info_dirty);
418 * Read dquot from disk and alloc space for it
421 int dquot_acquire(struct dquot *dquot)
423 int ret = 0, ret2 = 0;
424 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
426 mutex_lock(&dquot->dq_lock);
427 if (!test_bit(DQ_READ_B, &dquot->dq_flags))
428 ret = dqopt->ops[dquot->dq_id.type]->read_dqblk(dquot);
431 /* Make sure flags update is visible after dquot has been filled */
432 smp_mb__before_atomic();
433 set_bit(DQ_READ_B, &dquot->dq_flags);
434 /* Instantiate dquot if needed */
435 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
436 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
437 /* Write the info if needed */
438 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
439 ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
440 dquot->dq_sb, dquot->dq_id.type);
450 * Make sure flags update is visible after on-disk struct has been
451 * allocated. Paired with smp_rmb() in dqget().
453 smp_mb__before_atomic();
454 set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
456 mutex_unlock(&dquot->dq_lock);
459 EXPORT_SYMBOL(dquot_acquire);
462 * Write dquot to disk
464 int dquot_commit(struct dquot *dquot)
467 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
469 mutex_lock(&dquot->dq_lock);
470 if (!clear_dquot_dirty(dquot))
472 /* Inactive dquot can be only if there was error during read/init
473 * => we have better not writing it */
474 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
475 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
479 mutex_unlock(&dquot->dq_lock);
482 EXPORT_SYMBOL(dquot_commit);
487 int dquot_release(struct dquot *dquot)
489 int ret = 0, ret2 = 0;
490 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
492 mutex_lock(&dquot->dq_lock);
493 /* Check whether we are not racing with some other dqget() */
494 if (dquot_is_busy(dquot))
496 if (dqopt->ops[dquot->dq_id.type]->release_dqblk) {
497 ret = dqopt->ops[dquot->dq_id.type]->release_dqblk(dquot);
499 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
500 ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
501 dquot->dq_sb, dquot->dq_id.type);
506 clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
508 mutex_unlock(&dquot->dq_lock);
511 EXPORT_SYMBOL(dquot_release);
513 void dquot_destroy(struct dquot *dquot)
515 kmem_cache_free(dquot_cachep, dquot);
517 EXPORT_SYMBOL(dquot_destroy);
519 static inline void do_destroy_dquot(struct dquot *dquot)
521 dquot->dq_sb->dq_op->destroy_dquot(dquot);
524 /* Invalidate all dquots on the list. Note that this function is called after
525 * quota is disabled and pointers from inodes removed so there cannot be new
526 * quota users. There can still be some users of quotas due to inodes being
527 * just deleted or pruned by prune_icache() (those are not attached to any
528 * list) or parallel quotactl call. We have to wait for such users.
530 static void invalidate_dquots(struct super_block *sb, int type)
532 struct dquot *dquot, *tmp;
535 spin_lock(&dq_list_lock);
536 list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
537 if (dquot->dq_sb != sb)
539 if (dquot->dq_id.type != type)
541 /* Wait for dquot users */
542 if (atomic_read(&dquot->dq_count)) {
544 spin_unlock(&dq_list_lock);
546 * Once dqput() wakes us up, we know it's time to free
548 * IMPORTANT: we rely on the fact that there is always
549 * at most one process waiting for dquot to free.
550 * Otherwise dq_count would be > 1 and we would never
553 wait_event(dquot_ref_wq,
554 atomic_read(&dquot->dq_count) == 1);
556 /* At this moment dquot() need not exist (it could be
557 * reclaimed by prune_dqcache(). Hence we must
562 * Quota now has no users and it has been written on last
565 remove_dquot_hash(dquot);
566 remove_free_dquot(dquot);
568 do_destroy_dquot(dquot);
570 spin_unlock(&dq_list_lock);
573 /* Call callback for every active dquot on given filesystem */
574 int dquot_scan_active(struct super_block *sb,
575 int (*fn)(struct dquot *dquot, unsigned long priv),
578 struct dquot *dquot, *old_dquot = NULL;
581 WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
583 spin_lock(&dq_list_lock);
584 list_for_each_entry(dquot, &inuse_list, dq_inuse) {
585 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
587 if (dquot->dq_sb != sb)
589 /* Now we have active dquot so we can just increase use count */
590 atomic_inc(&dquot->dq_count);
591 spin_unlock(&dq_list_lock);
592 dqstats_inc(DQST_LOOKUPS);
596 * ->release_dquot() can be racing with us. Our reference
597 * protects us from new calls to it so just wait for any
598 * outstanding call and recheck the DQ_ACTIVE_B after that.
600 wait_on_dquot(dquot);
601 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
602 ret = fn(dquot, priv);
606 spin_lock(&dq_list_lock);
607 /* We are safe to continue now because our dquot could not
608 * be moved out of the inuse list while we hold the reference */
610 spin_unlock(&dq_list_lock);
615 EXPORT_SYMBOL(dquot_scan_active);
617 /* Write all dquot structures to quota files */
618 int dquot_writeback_dquots(struct super_block *sb, int type)
620 struct list_head dirty;
622 struct quota_info *dqopt = sb_dqopt(sb);
626 WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
628 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
629 if (type != -1 && cnt != type)
631 if (!sb_has_quota_active(sb, cnt))
633 spin_lock(&dq_list_lock);
634 /* Move list away to avoid livelock. */
635 list_replace_init(&dqopt->info[cnt].dqi_dirty_list, &dirty);
636 while (!list_empty(&dirty)) {
637 dquot = list_first_entry(&dirty, struct dquot,
640 WARN_ON(!test_bit(DQ_ACTIVE_B, &dquot->dq_flags));
642 /* Now we have active dquot from which someone is
643 * holding reference so we can safely just increase
646 spin_unlock(&dq_list_lock);
647 dqstats_inc(DQST_LOOKUPS);
648 err = sb->dq_op->write_dquot(dquot);
652 spin_lock(&dq_list_lock);
654 spin_unlock(&dq_list_lock);
657 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
658 if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
659 && info_dirty(&dqopt->info[cnt]))
660 sb->dq_op->write_info(sb, cnt);
661 dqstats_inc(DQST_SYNCS);
665 EXPORT_SYMBOL(dquot_writeback_dquots);
667 /* Write all dquot structures to disk and make them visible from userspace */
668 int dquot_quota_sync(struct super_block *sb, int type)
670 struct quota_info *dqopt = sb_dqopt(sb);
674 ret = dquot_writeback_dquots(sb, type);
677 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
680 /* This is not very clever (and fast) but currently I don't know about
681 * any other simple way of getting quota data to disk and we must get
682 * them there for userspace to be visible... */
683 if (sb->s_op->sync_fs)
684 sb->s_op->sync_fs(sb, 1);
685 sync_blockdev(sb->s_bdev);
688 * Now when everything is written we can discard the pagecache so
689 * that userspace sees the changes.
691 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
692 if (type != -1 && cnt != type)
694 if (!sb_has_quota_active(sb, cnt))
696 inode_lock(dqopt->files[cnt]);
697 truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
698 inode_unlock(dqopt->files[cnt]);
703 EXPORT_SYMBOL(dquot_quota_sync);
706 dqcache_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
708 struct list_head *head;
710 unsigned long freed = 0;
712 spin_lock(&dq_list_lock);
713 head = free_dquots.prev;
714 while (head != &free_dquots && sc->nr_to_scan) {
715 dquot = list_entry(head, struct dquot, dq_free);
716 remove_dquot_hash(dquot);
717 remove_free_dquot(dquot);
719 do_destroy_dquot(dquot);
722 head = free_dquots.prev;
724 spin_unlock(&dq_list_lock);
729 dqcache_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
731 return vfs_pressure_ratio(
732 percpu_counter_read_positive(&dqstats.counter[DQST_FREE_DQUOTS]));
735 static struct shrinker dqcache_shrinker = {
736 .count_objects = dqcache_shrink_count,
737 .scan_objects = dqcache_shrink_scan,
738 .seeks = DEFAULT_SEEKS,
742 * Put reference to dquot
744 void dqput(struct dquot *dquot)
750 #ifdef CONFIG_QUOTA_DEBUG
751 if (!atomic_read(&dquot->dq_count)) {
752 quota_error(dquot->dq_sb, "trying to free free dquot of %s %d",
753 quotatypes[dquot->dq_id.type],
754 from_kqid(&init_user_ns, dquot->dq_id));
758 dqstats_inc(DQST_DROPS);
760 spin_lock(&dq_list_lock);
761 if (atomic_read(&dquot->dq_count) > 1) {
762 /* We have more than one user... nothing to do */
763 atomic_dec(&dquot->dq_count);
764 /* Releasing dquot during quotaoff phase? */
765 if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_id.type) &&
766 atomic_read(&dquot->dq_count) == 1)
767 wake_up(&dquot_ref_wq);
768 spin_unlock(&dq_list_lock);
771 /* Need to release dquot? */
772 if (dquot_dirty(dquot)) {
773 spin_unlock(&dq_list_lock);
774 /* Commit dquot before releasing */
775 ret = dquot->dq_sb->dq_op->write_dquot(dquot);
777 quota_error(dquot->dq_sb, "Can't write quota structure"
778 " (error %d). Quota may get out of sync!",
781 * We clear dirty bit anyway, so that we avoid
784 clear_dquot_dirty(dquot);
788 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
789 spin_unlock(&dq_list_lock);
790 dquot->dq_sb->dq_op->release_dquot(dquot);
793 atomic_dec(&dquot->dq_count);
794 #ifdef CONFIG_QUOTA_DEBUG
796 BUG_ON(!list_empty(&dquot->dq_free));
798 put_dquot_last(dquot);
799 spin_unlock(&dq_list_lock);
801 EXPORT_SYMBOL(dqput);
803 struct dquot *dquot_alloc(struct super_block *sb, int type)
805 return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
807 EXPORT_SYMBOL(dquot_alloc);
809 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
813 dquot = sb->dq_op->alloc_dquot(sb, type);
817 mutex_init(&dquot->dq_lock);
818 INIT_LIST_HEAD(&dquot->dq_free);
819 INIT_LIST_HEAD(&dquot->dq_inuse);
820 INIT_HLIST_NODE(&dquot->dq_hash);
821 INIT_LIST_HEAD(&dquot->dq_dirty);
823 dquot->dq_id = make_kqid_invalid(type);
824 atomic_set(&dquot->dq_count, 1);
825 spin_lock_init(&dquot->dq_dqb_lock);
831 * Get reference to dquot
833 * Locking is slightly tricky here. We are guarded from parallel quotaoff()
834 * destroying our dquot by:
835 * a) checking for quota flags under dq_list_lock and
836 * b) getting a reference to dquot before we release dq_list_lock
838 struct dquot *dqget(struct super_block *sb, struct kqid qid)
840 unsigned int hashent = hashfn(sb, qid);
841 struct dquot *dquot, *empty = NULL;
843 if (!qid_has_mapping(sb->s_user_ns, qid))
844 return ERR_PTR(-EINVAL);
846 if (!sb_has_quota_active(sb, qid.type))
847 return ERR_PTR(-ESRCH);
849 spin_lock(&dq_list_lock);
850 spin_lock(&dq_state_lock);
851 if (!sb_has_quota_active(sb, qid.type)) {
852 spin_unlock(&dq_state_lock);
853 spin_unlock(&dq_list_lock);
854 dquot = ERR_PTR(-ESRCH);
857 spin_unlock(&dq_state_lock);
859 dquot = find_dquot(hashent, sb, qid);
862 spin_unlock(&dq_list_lock);
863 empty = get_empty_dquot(sb, qid.type);
865 schedule(); /* Try to wait for a moment... */
871 /* all dquots go on the inuse_list */
873 /* hash it first so it can be found */
874 insert_dquot_hash(dquot);
875 spin_unlock(&dq_list_lock);
876 dqstats_inc(DQST_LOOKUPS);
878 if (!atomic_read(&dquot->dq_count))
879 remove_free_dquot(dquot);
880 atomic_inc(&dquot->dq_count);
881 spin_unlock(&dq_list_lock);
882 dqstats_inc(DQST_CACHE_HITS);
883 dqstats_inc(DQST_LOOKUPS);
885 /* Wait for dq_lock - after this we know that either dquot_release() is
886 * already finished or it will be canceled due to dq_count > 1 test */
887 wait_on_dquot(dquot);
888 /* Read the dquot / allocate space in quota file */
889 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
892 err = sb->dq_op->acquire_dquot(dquot);
895 dquot = ERR_PTR(err);
900 * Make sure following reads see filled structure - paired with
901 * smp_mb__before_atomic() in dquot_acquire().
904 #ifdef CONFIG_QUOTA_DEBUG
905 BUG_ON(!dquot->dq_sb); /* Has somebody invalidated entry under us? */
909 do_destroy_dquot(empty);
913 EXPORT_SYMBOL(dqget);
915 static inline struct dquot **i_dquot(struct inode *inode)
917 return inode->i_sb->s_op->get_dquots(inode);
920 static int dqinit_needed(struct inode *inode, int type)
922 struct dquot * const *dquots;
925 if (IS_NOQUOTA(inode))
928 dquots = i_dquot(inode);
930 return !dquots[type];
931 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
937 /* This routine is guarded by s_umount semaphore */
938 static int add_dquot_ref(struct super_block *sb, int type)
940 struct inode *inode, *old_inode = NULL;
941 #ifdef CONFIG_QUOTA_DEBUG
946 spin_lock(&sb->s_inode_list_lock);
947 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
948 spin_lock(&inode->i_lock);
949 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
950 !atomic_read(&inode->i_writecount) ||
951 !dqinit_needed(inode, type)) {
952 spin_unlock(&inode->i_lock);
956 spin_unlock(&inode->i_lock);
957 spin_unlock(&sb->s_inode_list_lock);
959 #ifdef CONFIG_QUOTA_DEBUG
960 if (unlikely(inode_get_rsv_space(inode) > 0))
964 err = __dquot_initialize(inode, type);
971 * We hold a reference to 'inode' so it couldn't have been
972 * removed from s_inodes list while we dropped the
973 * s_inode_list_lock. We cannot iput the inode now as we can be
974 * holding the last reference and we cannot iput it under
975 * s_inode_list_lock. So we keep the reference and iput it
980 spin_lock(&sb->s_inode_list_lock);
982 spin_unlock(&sb->s_inode_list_lock);
985 #ifdef CONFIG_QUOTA_DEBUG
987 quota_error(sb, "Writes happened before quota was turned on "
988 "thus quota information is probably inconsistent. "
989 "Please run quotacheck(8)");
996 * Remove references to dquots from inode and add dquot to list for freeing
997 * if we have the last reference to dquot
999 static void remove_inode_dquot_ref(struct inode *inode, int type,
1000 struct list_head *tofree_head)
1002 struct dquot **dquots = i_dquot(inode);
1003 struct dquot *dquot = dquots[type];
1008 dquots[type] = NULL;
1009 if (list_empty(&dquot->dq_free)) {
1011 * The inode still has reference to dquot so it can't be in the
1014 spin_lock(&dq_list_lock);
1015 list_add(&dquot->dq_free, tofree_head);
1016 spin_unlock(&dq_list_lock);
1019 * Dquot is already in a list to put so we won't drop the last
1027 * Free list of dquots
1028 * Dquots are removed from inodes and no new references can be got so we are
1029 * the only ones holding reference
1031 static void put_dquot_list(struct list_head *tofree_head)
1033 struct list_head *act_head;
1034 struct dquot *dquot;
1036 act_head = tofree_head->next;
1037 while (act_head != tofree_head) {
1038 dquot = list_entry(act_head, struct dquot, dq_free);
1039 act_head = act_head->next;
1040 /* Remove dquot from the list so we won't have problems... */
1041 list_del_init(&dquot->dq_free);
1046 static void remove_dquot_ref(struct super_block *sb, int type,
1047 struct list_head *tofree_head)
1049 struct inode *inode;
1052 spin_lock(&sb->s_inode_list_lock);
1053 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1055 * We have to scan also I_NEW inodes because they can already
1056 * have quota pointer initialized. Luckily, we need to touch
1057 * only quota pointers and these have separate locking
1060 spin_lock(&dq_data_lock);
1061 if (!IS_NOQUOTA(inode)) {
1062 if (unlikely(inode_get_rsv_space(inode) > 0))
1064 remove_inode_dquot_ref(inode, type, tofree_head);
1066 spin_unlock(&dq_data_lock);
1068 spin_unlock(&sb->s_inode_list_lock);
1069 #ifdef CONFIG_QUOTA_DEBUG
1071 printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1072 " was disabled thus quota information is probably "
1073 "inconsistent. Please run quotacheck(8).\n", sb->s_id);
1078 /* Gather all references from inodes and drop them */
1079 static void drop_dquot_ref(struct super_block *sb, int type)
1081 LIST_HEAD(tofree_head);
1084 remove_dquot_ref(sb, type, &tofree_head);
1085 synchronize_srcu(&dquot_srcu);
1086 put_dquot_list(&tofree_head);
1091 void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1093 if (dquot->dq_dqb.dqb_rsvspace >= number)
1094 dquot->dq_dqb.dqb_rsvspace -= number;
1097 dquot->dq_dqb.dqb_rsvspace = 0;
1099 if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1100 dquot->dq_dqb.dqb_bsoftlimit)
1101 dquot->dq_dqb.dqb_btime = (time64_t) 0;
1102 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1105 static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1107 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1108 dquot->dq_dqb.dqb_curinodes >= number)
1109 dquot->dq_dqb.dqb_curinodes -= number;
1111 dquot->dq_dqb.dqb_curinodes = 0;
1112 if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1113 dquot->dq_dqb.dqb_itime = (time64_t) 0;
1114 clear_bit(DQ_INODES_B, &dquot->dq_flags);
1117 static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1119 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1120 dquot->dq_dqb.dqb_curspace >= number)
1121 dquot->dq_dqb.dqb_curspace -= number;
1123 dquot->dq_dqb.dqb_curspace = 0;
1124 if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1125 dquot->dq_dqb.dqb_bsoftlimit)
1126 dquot->dq_dqb.dqb_btime = (time64_t) 0;
1127 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1131 struct super_block *w_sb;
1132 struct kqid w_dq_id;
1136 static int warning_issued(struct dquot *dquot, const int warntype)
1138 int flag = (warntype == QUOTA_NL_BHARDWARN ||
1139 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1140 ((warntype == QUOTA_NL_IHARDWARN ||
1141 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1145 return test_and_set_bit(flag, &dquot->dq_flags);
1148 #ifdef CONFIG_PRINT_QUOTA_WARNING
1149 static int flag_print_warnings = 1;
1151 static int need_print_warning(struct dquot_warn *warn)
1153 if (!flag_print_warnings)
1156 switch (warn->w_dq_id.type) {
1158 return uid_eq(current_fsuid(), warn->w_dq_id.uid);
1160 return in_group_p(warn->w_dq_id.gid);
1167 /* Print warning to user which exceeded quota */
1168 static void print_warning(struct dquot_warn *warn)
1171 struct tty_struct *tty;
1172 int warntype = warn->w_type;
1174 if (warntype == QUOTA_NL_IHARDBELOW ||
1175 warntype == QUOTA_NL_ISOFTBELOW ||
1176 warntype == QUOTA_NL_BHARDBELOW ||
1177 warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(warn))
1180 tty = get_current_tty();
1183 tty_write_message(tty, warn->w_sb->s_id);
1184 if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1185 tty_write_message(tty, ": warning, ");
1187 tty_write_message(tty, ": write failed, ");
1188 tty_write_message(tty, quotatypes[warn->w_dq_id.type]);
1190 case QUOTA_NL_IHARDWARN:
1191 msg = " file limit reached.\r\n";
1193 case QUOTA_NL_ISOFTLONGWARN:
1194 msg = " file quota exceeded too long.\r\n";
1196 case QUOTA_NL_ISOFTWARN:
1197 msg = " file quota exceeded.\r\n";
1199 case QUOTA_NL_BHARDWARN:
1200 msg = " block limit reached.\r\n";
1202 case QUOTA_NL_BSOFTLONGWARN:
1203 msg = " block quota exceeded too long.\r\n";
1205 case QUOTA_NL_BSOFTWARN:
1206 msg = " block quota exceeded.\r\n";
1209 tty_write_message(tty, msg);
1214 static void prepare_warning(struct dquot_warn *warn, struct dquot *dquot,
1217 if (warning_issued(dquot, warntype))
1219 warn->w_type = warntype;
1220 warn->w_sb = dquot->dq_sb;
1221 warn->w_dq_id = dquot->dq_id;
1225 * Write warnings to the console and send warning messages over netlink.
1227 * Note that this function can call into tty and networking code.
1229 static void flush_warnings(struct dquot_warn *warn)
1233 for (i = 0; i < MAXQUOTAS; i++) {
1234 if (warn[i].w_type == QUOTA_NL_NOWARN)
1236 #ifdef CONFIG_PRINT_QUOTA_WARNING
1237 print_warning(&warn[i]);
1239 quota_send_warning(warn[i].w_dq_id,
1240 warn[i].w_sb->s_dev, warn[i].w_type);
1244 static int ignore_hardlimit(struct dquot *dquot)
1246 struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
1248 return capable(CAP_SYS_RESOURCE) &&
1249 (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1250 !(info->dqi_flags & DQF_ROOT_SQUASH));
1253 static int dquot_add_inodes(struct dquot *dquot, qsize_t inodes,
1254 struct dquot_warn *warn)
1259 spin_lock(&dquot->dq_dqb_lock);
1260 newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1261 if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type) ||
1262 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1265 if (dquot->dq_dqb.dqb_ihardlimit &&
1266 newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1267 !ignore_hardlimit(dquot)) {
1268 prepare_warning(warn, dquot, QUOTA_NL_IHARDWARN);
1273 if (dquot->dq_dqb.dqb_isoftlimit &&
1274 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1275 dquot->dq_dqb.dqb_itime &&
1276 ktime_get_real_seconds() >= dquot->dq_dqb.dqb_itime &&
1277 !ignore_hardlimit(dquot)) {
1278 prepare_warning(warn, dquot, QUOTA_NL_ISOFTLONGWARN);
1283 if (dquot->dq_dqb.dqb_isoftlimit &&
1284 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1285 dquot->dq_dqb.dqb_itime == 0) {
1286 prepare_warning(warn, dquot, QUOTA_NL_ISOFTWARN);
1287 dquot->dq_dqb.dqb_itime = ktime_get_real_seconds() +
1288 sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type].dqi_igrace;
1291 dquot->dq_dqb.dqb_curinodes = newinodes;
1294 spin_unlock(&dquot->dq_dqb_lock);
1298 static int dquot_add_space(struct dquot *dquot, qsize_t space,
1299 qsize_t rsv_space, unsigned int flags,
1300 struct dquot_warn *warn)
1303 struct super_block *sb = dquot->dq_sb;
1306 spin_lock(&dquot->dq_dqb_lock);
1307 if (!sb_has_quota_limits_enabled(sb, dquot->dq_id.type) ||
1308 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1311 tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1312 + space + rsv_space;
1314 if (dquot->dq_dqb.dqb_bhardlimit &&
1315 tspace > dquot->dq_dqb.dqb_bhardlimit &&
1316 !ignore_hardlimit(dquot)) {
1317 if (flags & DQUOT_SPACE_WARN)
1318 prepare_warning(warn, dquot, QUOTA_NL_BHARDWARN);
1323 if (dquot->dq_dqb.dqb_bsoftlimit &&
1324 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1325 dquot->dq_dqb.dqb_btime &&
1326 ktime_get_real_seconds() >= dquot->dq_dqb.dqb_btime &&
1327 !ignore_hardlimit(dquot)) {
1328 if (flags & DQUOT_SPACE_WARN)
1329 prepare_warning(warn, dquot, QUOTA_NL_BSOFTLONGWARN);
1334 if (dquot->dq_dqb.dqb_bsoftlimit &&
1335 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1336 dquot->dq_dqb.dqb_btime == 0) {
1337 if (flags & DQUOT_SPACE_WARN) {
1338 prepare_warning(warn, dquot, QUOTA_NL_BSOFTWARN);
1339 dquot->dq_dqb.dqb_btime = ktime_get_real_seconds() +
1340 sb_dqopt(sb)->info[dquot->dq_id.type].dqi_bgrace;
1343 * We don't allow preallocation to exceed softlimit so exceeding will
1352 * We have to be careful and go through warning generation & grace time
1353 * setting even if DQUOT_SPACE_NOFAIL is set. That's why we check it
1356 if (flags & DQUOT_SPACE_NOFAIL)
1359 dquot->dq_dqb.dqb_rsvspace += rsv_space;
1360 dquot->dq_dqb.dqb_curspace += space;
1362 spin_unlock(&dquot->dq_dqb_lock);
1366 static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1370 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1371 dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1372 !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type))
1373 return QUOTA_NL_NOWARN;
1375 newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1376 if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1377 return QUOTA_NL_ISOFTBELOW;
1378 if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1379 newinodes < dquot->dq_dqb.dqb_ihardlimit)
1380 return QUOTA_NL_IHARDBELOW;
1381 return QUOTA_NL_NOWARN;
1384 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1388 tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace;
1390 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1391 tspace <= dquot->dq_dqb.dqb_bsoftlimit)
1392 return QUOTA_NL_NOWARN;
1394 if (tspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1395 return QUOTA_NL_BSOFTBELOW;
1396 if (tspace >= dquot->dq_dqb.dqb_bhardlimit &&
1397 tspace - space < dquot->dq_dqb.dqb_bhardlimit)
1398 return QUOTA_NL_BHARDBELOW;
1399 return QUOTA_NL_NOWARN;
1402 static int dquot_active(const struct inode *inode)
1404 struct super_block *sb = inode->i_sb;
1406 if (IS_NOQUOTA(inode))
1408 return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1412 * Initialize quota pointers in inode
1414 * It is better to call this function outside of any transaction as it
1415 * might need a lot of space in journal for dquot structure allocation.
1417 static int __dquot_initialize(struct inode *inode, int type)
1419 int cnt, init_needed = 0;
1420 struct dquot **dquots, *got[MAXQUOTAS] = {};
1421 struct super_block *sb = inode->i_sb;
1425 if (!dquot_active(inode))
1428 dquots = i_dquot(inode);
1430 /* First get references to structures we might need. */
1431 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1435 struct dquot *dquot;
1437 if (type != -1 && cnt != type)
1440 * The i_dquot should have been initialized in most cases,
1441 * we check it without locking here to avoid unnecessary
1442 * dqget()/dqput() calls.
1447 if (!sb_has_quota_active(sb, cnt))
1454 qid = make_kqid_uid(inode->i_uid);
1457 qid = make_kqid_gid(inode->i_gid);
1460 rc = inode->i_sb->dq_op->get_projid(inode, &projid);
1463 qid = make_kqid_projid(projid);
1466 dquot = dqget(sb, qid);
1467 if (IS_ERR(dquot)) {
1468 /* We raced with somebody turning quotas off... */
1469 if (PTR_ERR(dquot) != -ESRCH) {
1470 ret = PTR_ERR(dquot);
1478 /* All required i_dquot has been initialized */
1482 spin_lock(&dq_data_lock);
1483 if (IS_NOQUOTA(inode))
1485 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1486 if (type != -1 && cnt != type)
1488 /* Avoid races with quotaoff() */
1489 if (!sb_has_quota_active(sb, cnt))
1491 /* We could race with quotaon or dqget() could have failed */
1495 dquots[cnt] = got[cnt];
1498 * Make quota reservation system happy if someone
1499 * did a write before quota was turned on
1501 rsv = inode_get_rsv_space(inode);
1502 if (unlikely(rsv)) {
1503 spin_lock(&inode->i_lock);
1504 /* Get reservation again under proper lock */
1505 rsv = __inode_get_rsv_space(inode);
1506 spin_lock(&dquots[cnt]->dq_dqb_lock);
1507 dquots[cnt]->dq_dqb.dqb_rsvspace += rsv;
1508 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1509 spin_unlock(&inode->i_lock);
1514 spin_unlock(&dq_data_lock);
1516 /* Drop unused references */
1522 int dquot_initialize(struct inode *inode)
1524 return __dquot_initialize(inode, -1);
1526 EXPORT_SYMBOL(dquot_initialize);
1528 bool dquot_initialize_needed(struct inode *inode)
1530 struct dquot **dquots;
1533 if (!dquot_active(inode))
1536 dquots = i_dquot(inode);
1537 for (i = 0; i < MAXQUOTAS; i++)
1538 if (!dquots[i] && sb_has_quota_active(inode->i_sb, i))
1542 EXPORT_SYMBOL(dquot_initialize_needed);
1545 * Release all quotas referenced by inode.
1547 * This function only be called on inode free or converting
1548 * a file to quota file, no other users for the i_dquot in
1549 * both cases, so we needn't call synchronize_srcu() after
1552 static void __dquot_drop(struct inode *inode)
1555 struct dquot **dquots = i_dquot(inode);
1556 struct dquot *put[MAXQUOTAS];
1558 spin_lock(&dq_data_lock);
1559 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1560 put[cnt] = dquots[cnt];
1563 spin_unlock(&dq_data_lock);
1567 void dquot_drop(struct inode *inode)
1569 struct dquot * const *dquots;
1572 if (IS_NOQUOTA(inode))
1576 * Test before calling to rule out calls from proc and such
1577 * where we are not allowed to block. Note that this is
1578 * actually reliable test even without the lock - the caller
1579 * must assure that nobody can come after the DQUOT_DROP and
1580 * add quota pointers back anyway.
1582 dquots = i_dquot(inode);
1583 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1588 if (cnt < MAXQUOTAS)
1589 __dquot_drop(inode);
1591 EXPORT_SYMBOL(dquot_drop);
1594 * inode_reserved_space is managed internally by quota, and protected by
1595 * i_lock similar to i_blocks+i_bytes.
1597 static qsize_t *inode_reserved_space(struct inode * inode)
1599 /* Filesystem must explicitly define it's own method in order to use
1600 * quota reservation interface */
1601 BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1602 return inode->i_sb->dq_op->get_reserved_space(inode);
1605 static qsize_t __inode_get_rsv_space(struct inode *inode)
1607 if (!inode->i_sb->dq_op->get_reserved_space)
1609 return *inode_reserved_space(inode);
1612 static qsize_t inode_get_rsv_space(struct inode *inode)
1616 if (!inode->i_sb->dq_op->get_reserved_space)
1618 spin_lock(&inode->i_lock);
1619 ret = __inode_get_rsv_space(inode);
1620 spin_unlock(&inode->i_lock);
1625 * This functions updates i_blocks+i_bytes fields and quota information
1626 * (together with appropriate checks).
1628 * NOTE: We absolutely rely on the fact that caller dirties the inode
1629 * (usually helpers in quotaops.h care about this) and holds a handle for
1630 * the current transaction so that dquot write and inode write go into the
1635 * This operation can block, but only after everything is updated
1637 int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
1639 int cnt, ret = 0, index;
1640 struct dquot_warn warn[MAXQUOTAS];
1641 int reserve = flags & DQUOT_SPACE_RESERVE;
1642 struct dquot **dquots;
1644 if (!dquot_active(inode)) {
1646 spin_lock(&inode->i_lock);
1647 *inode_reserved_space(inode) += number;
1648 spin_unlock(&inode->i_lock);
1650 inode_add_bytes(inode, number);
1655 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1656 warn[cnt].w_type = QUOTA_NL_NOWARN;
1658 dquots = i_dquot(inode);
1659 index = srcu_read_lock(&dquot_srcu);
1660 spin_lock(&inode->i_lock);
1661 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1664 if (flags & DQUOT_SPACE_RESERVE) {
1665 ret = dquot_add_space(dquots[cnt], 0, number, flags,
1668 ret = dquot_add_space(dquots[cnt], number, 0, flags,
1672 /* Back out changes we already did */
1673 for (cnt--; cnt >= 0; cnt--) {
1676 spin_lock(&dquots[cnt]->dq_dqb_lock);
1677 if (flags & DQUOT_SPACE_RESERVE) {
1678 dquots[cnt]->dq_dqb.dqb_rsvspace -=
1681 dquots[cnt]->dq_dqb.dqb_curspace -=
1684 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1686 spin_unlock(&inode->i_lock);
1687 goto out_flush_warn;
1691 *inode_reserved_space(inode) += number;
1693 __inode_add_bytes(inode, number);
1694 spin_unlock(&inode->i_lock);
1697 goto out_flush_warn;
1698 mark_all_dquot_dirty(dquots);
1700 srcu_read_unlock(&dquot_srcu, index);
1701 flush_warnings(warn);
1705 EXPORT_SYMBOL(__dquot_alloc_space);
1708 * This operation can block, but only after everything is updated
1710 int dquot_alloc_inode(struct inode *inode)
1712 int cnt, ret = 0, index;
1713 struct dquot_warn warn[MAXQUOTAS];
1714 struct dquot * const *dquots;
1716 if (!dquot_active(inode))
1718 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1719 warn[cnt].w_type = QUOTA_NL_NOWARN;
1721 dquots = i_dquot(inode);
1722 index = srcu_read_lock(&dquot_srcu);
1723 spin_lock(&inode->i_lock);
1724 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1727 ret = dquot_add_inodes(dquots[cnt], 1, &warn[cnt]);
1729 for (cnt--; cnt >= 0; cnt--) {
1732 /* Back out changes we already did */
1733 spin_lock(&dquots[cnt]->dq_dqb_lock);
1734 dquots[cnt]->dq_dqb.dqb_curinodes--;
1735 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1742 spin_unlock(&inode->i_lock);
1744 mark_all_dquot_dirty(dquots);
1745 srcu_read_unlock(&dquot_srcu, index);
1746 flush_warnings(warn);
1749 EXPORT_SYMBOL(dquot_alloc_inode);
1752 * Convert in-memory reserved quotas to real consumed quotas
1754 int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1756 struct dquot **dquots;
1759 if (!dquot_active(inode)) {
1760 spin_lock(&inode->i_lock);
1761 *inode_reserved_space(inode) -= number;
1762 __inode_add_bytes(inode, number);
1763 spin_unlock(&inode->i_lock);
1767 dquots = i_dquot(inode);
1768 index = srcu_read_lock(&dquot_srcu);
1769 spin_lock(&inode->i_lock);
1770 /* Claim reserved quotas to allocated quotas */
1771 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1773 struct dquot *dquot = dquots[cnt];
1775 spin_lock(&dquot->dq_dqb_lock);
1776 if (WARN_ON_ONCE(dquot->dq_dqb.dqb_rsvspace < number))
1777 number = dquot->dq_dqb.dqb_rsvspace;
1778 dquot->dq_dqb.dqb_curspace += number;
1779 dquot->dq_dqb.dqb_rsvspace -= number;
1780 spin_unlock(&dquot->dq_dqb_lock);
1783 /* Update inode bytes */
1784 *inode_reserved_space(inode) -= number;
1785 __inode_add_bytes(inode, number);
1786 spin_unlock(&inode->i_lock);
1787 mark_all_dquot_dirty(dquots);
1788 srcu_read_unlock(&dquot_srcu, index);
1791 EXPORT_SYMBOL(dquot_claim_space_nodirty);
1794 * Convert allocated space back to in-memory reserved quotas
1796 void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
1798 struct dquot **dquots;
1801 if (!dquot_active(inode)) {
1802 spin_lock(&inode->i_lock);
1803 *inode_reserved_space(inode) += number;
1804 __inode_sub_bytes(inode, number);
1805 spin_unlock(&inode->i_lock);
1809 dquots = i_dquot(inode);
1810 index = srcu_read_lock(&dquot_srcu);
1811 spin_lock(&inode->i_lock);
1812 /* Claim reserved quotas to allocated quotas */
1813 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1815 struct dquot *dquot = dquots[cnt];
1817 spin_lock(&dquot->dq_dqb_lock);
1818 if (WARN_ON_ONCE(dquot->dq_dqb.dqb_curspace < number))
1819 number = dquot->dq_dqb.dqb_curspace;
1820 dquot->dq_dqb.dqb_rsvspace += number;
1821 dquot->dq_dqb.dqb_curspace -= number;
1822 spin_unlock(&dquot->dq_dqb_lock);
1825 /* Update inode bytes */
1826 *inode_reserved_space(inode) += number;
1827 __inode_sub_bytes(inode, number);
1828 spin_unlock(&inode->i_lock);
1829 mark_all_dquot_dirty(dquots);
1830 srcu_read_unlock(&dquot_srcu, index);
1833 EXPORT_SYMBOL(dquot_reclaim_space_nodirty);
1836 * This operation can block, but only after everything is updated
1838 void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
1841 struct dquot_warn warn[MAXQUOTAS];
1842 struct dquot **dquots;
1843 int reserve = flags & DQUOT_SPACE_RESERVE, index;
1845 if (!dquot_active(inode)) {
1847 spin_lock(&inode->i_lock);
1848 *inode_reserved_space(inode) -= number;
1849 spin_unlock(&inode->i_lock);
1851 inode_sub_bytes(inode, number);
1856 dquots = i_dquot(inode);
1857 index = srcu_read_lock(&dquot_srcu);
1858 spin_lock(&inode->i_lock);
1859 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1862 warn[cnt].w_type = QUOTA_NL_NOWARN;
1865 spin_lock(&dquots[cnt]->dq_dqb_lock);
1866 wtype = info_bdq_free(dquots[cnt], number);
1867 if (wtype != QUOTA_NL_NOWARN)
1868 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1870 dquot_free_reserved_space(dquots[cnt], number);
1872 dquot_decr_space(dquots[cnt], number);
1873 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1876 *inode_reserved_space(inode) -= number;
1878 __inode_sub_bytes(inode, number);
1879 spin_unlock(&inode->i_lock);
1883 mark_all_dquot_dirty(dquots);
1885 srcu_read_unlock(&dquot_srcu, index);
1886 flush_warnings(warn);
1888 EXPORT_SYMBOL(__dquot_free_space);
1891 * This operation can block, but only after everything is updated
1893 void dquot_free_inode(struct inode *inode)
1896 struct dquot_warn warn[MAXQUOTAS];
1897 struct dquot * const *dquots;
1900 if (!dquot_active(inode))
1903 dquots = i_dquot(inode);
1904 index = srcu_read_lock(&dquot_srcu);
1905 spin_lock(&inode->i_lock);
1906 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1909 warn[cnt].w_type = QUOTA_NL_NOWARN;
1912 spin_lock(&dquots[cnt]->dq_dqb_lock);
1913 wtype = info_idq_free(dquots[cnt], 1);
1914 if (wtype != QUOTA_NL_NOWARN)
1915 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1916 dquot_decr_inodes(dquots[cnt], 1);
1917 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1919 spin_unlock(&inode->i_lock);
1920 mark_all_dquot_dirty(dquots);
1921 srcu_read_unlock(&dquot_srcu, index);
1922 flush_warnings(warn);
1924 EXPORT_SYMBOL(dquot_free_inode);
1927 * Transfer the number of inode and blocks from one diskquota to an other.
1928 * On success, dquot references in transfer_to are consumed and references
1929 * to original dquots that need to be released are placed there. On failure,
1930 * references are kept untouched.
1932 * This operation can block, but only after everything is updated
1933 * A transaction must be started when entering this function.
1935 * We are holding reference on transfer_from & transfer_to, no need to
1936 * protect them by srcu_read_lock().
1938 int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
1941 qsize_t rsv_space = 0;
1942 qsize_t inode_usage = 1;
1943 struct dquot *transfer_from[MAXQUOTAS] = {};
1945 char is_valid[MAXQUOTAS] = {};
1946 struct dquot_warn warn_to[MAXQUOTAS];
1947 struct dquot_warn warn_from_inodes[MAXQUOTAS];
1948 struct dquot_warn warn_from_space[MAXQUOTAS];
1950 if (IS_NOQUOTA(inode))
1953 if (inode->i_sb->dq_op->get_inode_usage) {
1954 ret = inode->i_sb->dq_op->get_inode_usage(inode, &inode_usage);
1959 /* Initialize the arrays */
1960 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1961 warn_to[cnt].w_type = QUOTA_NL_NOWARN;
1962 warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
1963 warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
1966 spin_lock(&dq_data_lock);
1967 spin_lock(&inode->i_lock);
1968 if (IS_NOQUOTA(inode)) { /* File without quota accounting? */
1969 spin_unlock(&inode->i_lock);
1970 spin_unlock(&dq_data_lock);
1973 cur_space = __inode_get_bytes(inode);
1974 rsv_space = __inode_get_rsv_space(inode);
1976 * Build the transfer_from list, check limits, and update usage in
1977 * the target structures.
1979 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1981 * Skip changes for same uid or gid or for turned off quota-type.
1983 if (!transfer_to[cnt])
1985 /* Avoid races with quotaoff() */
1986 if (!sb_has_quota_active(inode->i_sb, cnt))
1989 transfer_from[cnt] = i_dquot(inode)[cnt];
1990 ret = dquot_add_inodes(transfer_to[cnt], inode_usage,
1994 ret = dquot_add_space(transfer_to[cnt], cur_space, rsv_space,
1995 DQUOT_SPACE_WARN, &warn_to[cnt]);
1997 spin_lock(&transfer_to[cnt]->dq_dqb_lock);
1998 dquot_decr_inodes(transfer_to[cnt], inode_usage);
1999 spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2004 /* Decrease usage for source structures and update quota pointers */
2005 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2008 /* Due to IO error we might not have transfer_from[] structure */
2009 if (transfer_from[cnt]) {
2012 spin_lock(&transfer_from[cnt]->dq_dqb_lock);
2013 wtype = info_idq_free(transfer_from[cnt], inode_usage);
2014 if (wtype != QUOTA_NL_NOWARN)
2015 prepare_warning(&warn_from_inodes[cnt],
2016 transfer_from[cnt], wtype);
2017 wtype = info_bdq_free(transfer_from[cnt],
2018 cur_space + rsv_space);
2019 if (wtype != QUOTA_NL_NOWARN)
2020 prepare_warning(&warn_from_space[cnt],
2021 transfer_from[cnt], wtype);
2022 dquot_decr_inodes(transfer_from[cnt], inode_usage);
2023 dquot_decr_space(transfer_from[cnt], cur_space);
2024 dquot_free_reserved_space(transfer_from[cnt],
2026 spin_unlock(&transfer_from[cnt]->dq_dqb_lock);
2028 i_dquot(inode)[cnt] = transfer_to[cnt];
2030 spin_unlock(&inode->i_lock);
2031 spin_unlock(&dq_data_lock);
2033 mark_all_dquot_dirty(transfer_from);
2034 mark_all_dquot_dirty(transfer_to);
2035 flush_warnings(warn_to);
2036 flush_warnings(warn_from_inodes);
2037 flush_warnings(warn_from_space);
2038 /* Pass back references to put */
2039 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2041 transfer_to[cnt] = transfer_from[cnt];
2044 /* Back out changes we already did */
2045 for (cnt--; cnt >= 0; cnt--) {
2048 spin_lock(&transfer_to[cnt]->dq_dqb_lock);
2049 dquot_decr_inodes(transfer_to[cnt], inode_usage);
2050 dquot_decr_space(transfer_to[cnt], cur_space);
2051 dquot_free_reserved_space(transfer_to[cnt], rsv_space);
2052 spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2054 spin_unlock(&inode->i_lock);
2055 spin_unlock(&dq_data_lock);
2056 flush_warnings(warn_to);
2059 EXPORT_SYMBOL(__dquot_transfer);
2061 /* Wrapper for transferring ownership of an inode for uid/gid only
2062 * Called from FSXXX_setattr()
2064 int dquot_transfer(struct inode *inode, struct iattr *iattr)
2066 struct dquot *transfer_to[MAXQUOTAS] = {};
2067 struct dquot *dquot;
2068 struct super_block *sb = inode->i_sb;
2071 if (!dquot_active(inode))
2074 if (iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)){
2075 dquot = dqget(sb, make_kqid_uid(iattr->ia_uid));
2076 if (IS_ERR(dquot)) {
2077 if (PTR_ERR(dquot) != -ESRCH) {
2078 ret = PTR_ERR(dquot);
2083 transfer_to[USRQUOTA] = dquot;
2085 if (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid)){
2086 dquot = dqget(sb, make_kqid_gid(iattr->ia_gid));
2087 if (IS_ERR(dquot)) {
2088 if (PTR_ERR(dquot) != -ESRCH) {
2089 ret = PTR_ERR(dquot);
2094 transfer_to[GRPQUOTA] = dquot;
2096 ret = __dquot_transfer(inode, transfer_to);
2098 dqput_all(transfer_to);
2101 EXPORT_SYMBOL(dquot_transfer);
2104 * Write info of quota file to disk
2106 int dquot_commit_info(struct super_block *sb, int type)
2108 struct quota_info *dqopt = sb_dqopt(sb);
2110 return dqopt->ops[type]->write_file_info(sb, type);
2112 EXPORT_SYMBOL(dquot_commit_info);
2114 int dquot_get_next_id(struct super_block *sb, struct kqid *qid)
2116 struct quota_info *dqopt = sb_dqopt(sb);
2118 if (!sb_has_quota_active(sb, qid->type))
2120 if (!dqopt->ops[qid->type]->get_next_id)
2122 return dqopt->ops[qid->type]->get_next_id(sb, qid);
2124 EXPORT_SYMBOL(dquot_get_next_id);
2127 * Definitions of diskquota operations.
2129 const struct dquot_operations dquot_operations = {
2130 .write_dquot = dquot_commit,
2131 .acquire_dquot = dquot_acquire,
2132 .release_dquot = dquot_release,
2133 .mark_dirty = dquot_mark_dquot_dirty,
2134 .write_info = dquot_commit_info,
2135 .alloc_dquot = dquot_alloc,
2136 .destroy_dquot = dquot_destroy,
2137 .get_next_id = dquot_get_next_id,
2139 EXPORT_SYMBOL(dquot_operations);
2142 * Generic helper for ->open on filesystems supporting disk quotas.
2144 int dquot_file_open(struct inode *inode, struct file *file)
2148 error = generic_file_open(inode, file);
2149 if (!error && (file->f_mode & FMODE_WRITE))
2150 dquot_initialize(inode);
2153 EXPORT_SYMBOL(dquot_file_open);
2156 * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
2158 int dquot_disable(struct super_block *sb, int type, unsigned int flags)
2161 struct quota_info *dqopt = sb_dqopt(sb);
2162 struct inode *toputinode[MAXQUOTAS];
2164 /* s_umount should be held in exclusive mode */
2165 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2166 up_read(&sb->s_umount);
2168 /* Cannot turn off usage accounting without turning off limits, or
2169 * suspend quotas and simultaneously turn quotas off. */
2170 if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
2171 || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
2172 DQUOT_USAGE_ENABLED)))
2176 * Skip everything if there's nothing to do. We have to do this because
2177 * sometimes we are called when fill_super() failed and calling
2178 * sync_fs() in such cases does no good.
2180 if (!sb_any_quota_loaded(sb))
2183 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2184 toputinode[cnt] = NULL;
2185 if (type != -1 && cnt != type)
2187 if (!sb_has_quota_loaded(sb, cnt))
2190 if (flags & DQUOT_SUSPENDED) {
2191 spin_lock(&dq_state_lock);
2193 dquot_state_flag(DQUOT_SUSPENDED, cnt);
2194 spin_unlock(&dq_state_lock);
2196 spin_lock(&dq_state_lock);
2197 dqopt->flags &= ~dquot_state_flag(flags, cnt);
2198 /* Turning off suspended quotas? */
2199 if (!sb_has_quota_loaded(sb, cnt) &&
2200 sb_has_quota_suspended(sb, cnt)) {
2201 dqopt->flags &= ~dquot_state_flag(
2202 DQUOT_SUSPENDED, cnt);
2203 spin_unlock(&dq_state_lock);
2204 iput(dqopt->files[cnt]);
2205 dqopt->files[cnt] = NULL;
2208 spin_unlock(&dq_state_lock);
2211 /* We still have to keep quota loaded? */
2212 if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
2215 /* Note: these are blocking operations */
2216 drop_dquot_ref(sb, cnt);
2217 invalidate_dquots(sb, cnt);
2219 * Now all dquots should be invalidated, all writes done so we
2220 * should be only users of the info. No locks needed.
2222 if (info_dirty(&dqopt->info[cnt]))
2223 sb->dq_op->write_info(sb, cnt);
2224 if (dqopt->ops[cnt]->free_file_info)
2225 dqopt->ops[cnt]->free_file_info(sb, cnt);
2226 put_quota_format(dqopt->info[cnt].dqi_format);
2228 toputinode[cnt] = dqopt->files[cnt];
2229 if (!sb_has_quota_loaded(sb, cnt))
2230 dqopt->files[cnt] = NULL;
2231 dqopt->info[cnt].dqi_flags = 0;
2232 dqopt->info[cnt].dqi_igrace = 0;
2233 dqopt->info[cnt].dqi_bgrace = 0;
2234 dqopt->ops[cnt] = NULL;
2237 /* Skip syncing and setting flags if quota files are hidden */
2238 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2241 /* Sync the superblock so that buffers with quota data are written to
2242 * disk (and so userspace sees correct data afterwards). */
2243 if (sb->s_op->sync_fs)
2244 sb->s_op->sync_fs(sb, 1);
2245 sync_blockdev(sb->s_bdev);
2246 /* Now the quota files are just ordinary files and we can set the
2247 * inode flags back. Moreover we discard the pagecache so that
2248 * userspace sees the writes we did bypassing the pagecache. We
2249 * must also discard the blockdev buffers so that we see the
2250 * changes done by userspace on the next quotaon() */
2251 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2252 /* This can happen when suspending quotas on remount-ro... */
2253 if (toputinode[cnt] && !sb_has_quota_loaded(sb, cnt)) {
2254 inode_lock(toputinode[cnt]);
2255 toputinode[cnt]->i_flags &= ~S_NOQUOTA;
2256 truncate_inode_pages(&toputinode[cnt]->i_data, 0);
2257 inode_unlock(toputinode[cnt]);
2258 mark_inode_dirty_sync(toputinode[cnt]);
2261 invalidate_bdev(sb->s_bdev);
2263 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2264 if (toputinode[cnt]) {
2265 /* On remount RO, we keep the inode pointer so that we
2266 * can reenable quota on the subsequent remount RW. We
2267 * have to check 'flags' variable and not use sb_has_
2268 * function because another quotaon / quotaoff could
2269 * change global state before we got here. We refuse
2270 * to suspend quotas when there is pending delete on
2271 * the quota file... */
2272 if (!(flags & DQUOT_SUSPENDED))
2273 iput(toputinode[cnt]);
2274 else if (!toputinode[cnt]->i_nlink)
2279 EXPORT_SYMBOL(dquot_disable);
2281 int dquot_quota_off(struct super_block *sb, int type)
2283 return dquot_disable(sb, type,
2284 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2286 EXPORT_SYMBOL(dquot_quota_off);
2289 * Turn quotas on on a device
2293 * Helper function to turn quotas on when we already have the inode of
2294 * quota file and no quota information is loaded.
2296 static int vfs_load_quota_inode(struct inode *inode, int type, int format_id,
2299 struct quota_format_type *fmt = find_quota_format(format_id);
2300 struct super_block *sb = inode->i_sb;
2301 struct quota_info *dqopt = sb_dqopt(sb);
2306 if (!S_ISREG(inode->i_mode)) {
2310 if (IS_RDONLY(inode)) {
2314 if (!sb->s_op->quota_write || !sb->s_op->quota_read ||
2315 (type == PRJQUOTA && sb->dq_op->get_projid == NULL)) {
2319 /* Filesystems outside of init_user_ns not yet supported */
2320 if (sb->s_user_ns != &init_user_ns) {
2324 /* Usage always has to be set... */
2325 if (!(flags & DQUOT_USAGE_ENABLED)) {
2329 if (sb_has_quota_loaded(sb, type)) {
2334 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2335 /* As we bypass the pagecache we must now flush all the
2336 * dirty data and invalidate caches so that kernel sees
2337 * changes from userspace. It is not enough to just flush
2338 * the quota file since if blocksize < pagesize, invalidation
2339 * of the cache could fail because of other unrelated dirty
2341 sync_filesystem(sb);
2342 invalidate_bdev(sb->s_bdev);
2345 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2346 /* We don't want quota and atime on quota files (deadlocks
2347 * possible) Also nobody should write to the file - we use
2348 * special IO operations which ignore the immutable bit. */
2350 inode->i_flags |= S_NOQUOTA;
2351 inode_unlock(inode);
2353 * When S_NOQUOTA is set, remove dquot references as no more
2354 * references can be added
2356 __dquot_drop(inode);
2360 dqopt->files[type] = igrab(inode);
2361 if (!dqopt->files[type])
2362 goto out_file_flags;
2364 if (!fmt->qf_ops->check_quota_file(sb, type))
2367 dqopt->ops[type] = fmt->qf_ops;
2368 dqopt->info[type].dqi_format = fmt;
2369 dqopt->info[type].dqi_fmt_id = format_id;
2370 INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2371 error = dqopt->ops[type]->read_file_info(sb, type);
2374 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE) {
2375 spin_lock(&dq_data_lock);
2376 dqopt->info[type].dqi_flags |= DQF_SYS_FILE;
2377 spin_unlock(&dq_data_lock);
2379 spin_lock(&dq_state_lock);
2380 dqopt->flags |= dquot_state_flag(flags, type);
2381 spin_unlock(&dq_state_lock);
2383 error = add_dquot_ref(sb, type);
2385 dquot_disable(sb, type, flags);
2389 dqopt->files[type] = NULL;
2393 inode->i_flags &= ~S_NOQUOTA;
2394 inode_unlock(inode);
2396 put_quota_format(fmt);
2401 /* Reenable quotas on remount RW */
2402 int dquot_resume(struct super_block *sb, int type)
2404 struct quota_info *dqopt = sb_dqopt(sb);
2405 struct inode *inode;
2409 /* s_umount should be held in exclusive mode */
2410 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2411 up_read(&sb->s_umount);
2413 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2414 if (type != -1 && cnt != type)
2416 if (!sb_has_quota_suspended(sb, cnt))
2419 inode = dqopt->files[cnt];
2420 dqopt->files[cnt] = NULL;
2421 spin_lock(&dq_state_lock);
2422 flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2423 DQUOT_LIMITS_ENABLED,
2425 dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2426 spin_unlock(&dq_state_lock);
2428 flags = dquot_generic_flag(flags, cnt);
2429 ret = vfs_load_quota_inode(inode, cnt,
2430 dqopt->info[cnt].dqi_fmt_id, flags);
2436 EXPORT_SYMBOL(dquot_resume);
2438 int dquot_quota_on(struct super_block *sb, int type, int format_id,
2439 const struct path *path)
2441 int error = security_quota_on(path->dentry);
2444 /* Quota file not on the same filesystem? */
2445 if (path->dentry->d_sb != sb)
2448 error = vfs_load_quota_inode(d_inode(path->dentry), type,
2449 format_id, DQUOT_USAGE_ENABLED |
2450 DQUOT_LIMITS_ENABLED);
2453 EXPORT_SYMBOL(dquot_quota_on);
2456 * More powerful function for turning on quotas allowing setting
2457 * of individual quota flags
2459 int dquot_enable(struct inode *inode, int type, int format_id,
2462 struct super_block *sb = inode->i_sb;
2464 /* Just unsuspend quotas? */
2465 BUG_ON(flags & DQUOT_SUSPENDED);
2466 /* s_umount should be held in exclusive mode */
2467 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2468 up_read(&sb->s_umount);
2472 /* Just updating flags needed? */
2473 if (sb_has_quota_loaded(sb, type)) {
2474 if (flags & DQUOT_USAGE_ENABLED &&
2475 sb_has_quota_usage_enabled(sb, type))
2477 if (flags & DQUOT_LIMITS_ENABLED &&
2478 sb_has_quota_limits_enabled(sb, type))
2480 spin_lock(&dq_state_lock);
2481 sb_dqopt(sb)->flags |= dquot_state_flag(flags, type);
2482 spin_unlock(&dq_state_lock);
2486 return vfs_load_quota_inode(inode, type, format_id, flags);
2488 EXPORT_SYMBOL(dquot_enable);
2491 * This function is used when filesystem needs to initialize quotas
2492 * during mount time.
2494 int dquot_quota_on_mount(struct super_block *sb, char *qf_name,
2495 int format_id, int type)
2497 struct dentry *dentry;
2500 dentry = lookup_one_len_unlocked(qf_name, sb->s_root, strlen(qf_name));
2502 return PTR_ERR(dentry);
2504 if (d_really_is_negative(dentry)) {
2509 error = security_quota_on(dentry);
2511 error = vfs_load_quota_inode(d_inode(dentry), type, format_id,
2512 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2518 EXPORT_SYMBOL(dquot_quota_on_mount);
2520 static int dquot_quota_enable(struct super_block *sb, unsigned int flags)
2524 struct quota_info *dqopt = sb_dqopt(sb);
2526 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2528 /* Accounting cannot be turned on while fs is mounted */
2529 flags &= ~(FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT);
2532 for (type = 0; type < MAXQUOTAS; type++) {
2533 if (!(flags & qtype_enforce_flag(type)))
2535 /* Can't enforce without accounting */
2536 if (!sb_has_quota_usage_enabled(sb, type))
2538 ret = dquot_enable(dqopt->files[type], type,
2539 dqopt->info[type].dqi_fmt_id,
2540 DQUOT_LIMITS_ENABLED);
2546 /* Backout enforcement enablement we already did */
2547 for (type--; type >= 0; type--) {
2548 if (flags & qtype_enforce_flag(type))
2549 dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2551 /* Error code translation for better compatibility with XFS */
2557 static int dquot_quota_disable(struct super_block *sb, unsigned int flags)
2561 struct quota_info *dqopt = sb_dqopt(sb);
2563 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2566 * We don't support turning off accounting via quotactl. In principle
2567 * quota infrastructure can do this but filesystems don't expect
2568 * userspace to be able to do it.
2571 (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT))
2574 /* Filter out limits not enabled */
2575 for (type = 0; type < MAXQUOTAS; type++)
2576 if (!sb_has_quota_limits_enabled(sb, type))
2577 flags &= ~qtype_enforce_flag(type);
2581 for (type = 0; type < MAXQUOTAS; type++) {
2582 if (flags & qtype_enforce_flag(type)) {
2583 ret = dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2590 /* Backout enforcement disabling we already did */
2591 for (type--; type >= 0; type--) {
2592 if (flags & qtype_enforce_flag(type))
2593 dquot_enable(dqopt->files[type], type,
2594 dqopt->info[type].dqi_fmt_id,
2595 DQUOT_LIMITS_ENABLED);
2600 /* Generic routine for getting common part of quota structure */
2601 static void do_get_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2603 struct mem_dqblk *dm = &dquot->dq_dqb;
2605 memset(di, 0, sizeof(*di));
2606 spin_lock(&dquot->dq_dqb_lock);
2607 di->d_spc_hardlimit = dm->dqb_bhardlimit;
2608 di->d_spc_softlimit = dm->dqb_bsoftlimit;
2609 di->d_ino_hardlimit = dm->dqb_ihardlimit;
2610 di->d_ino_softlimit = dm->dqb_isoftlimit;
2611 di->d_space = dm->dqb_curspace + dm->dqb_rsvspace;
2612 di->d_ino_count = dm->dqb_curinodes;
2613 di->d_spc_timer = dm->dqb_btime;
2614 di->d_ino_timer = dm->dqb_itime;
2615 spin_unlock(&dquot->dq_dqb_lock);
2618 int dquot_get_dqblk(struct super_block *sb, struct kqid qid,
2619 struct qc_dqblk *di)
2621 struct dquot *dquot;
2623 dquot = dqget(sb, qid);
2625 return PTR_ERR(dquot);
2626 do_get_dqblk(dquot, di);
2631 EXPORT_SYMBOL(dquot_get_dqblk);
2633 int dquot_get_next_dqblk(struct super_block *sb, struct kqid *qid,
2634 struct qc_dqblk *di)
2636 struct dquot *dquot;
2639 if (!sb->dq_op->get_next_id)
2641 err = sb->dq_op->get_next_id(sb, qid);
2644 dquot = dqget(sb, *qid);
2646 return PTR_ERR(dquot);
2647 do_get_dqblk(dquot, di);
2652 EXPORT_SYMBOL(dquot_get_next_dqblk);
2654 #define VFS_QC_MASK \
2655 (QC_SPACE | QC_SPC_SOFT | QC_SPC_HARD | \
2656 QC_INO_COUNT | QC_INO_SOFT | QC_INO_HARD | \
2657 QC_SPC_TIMER | QC_INO_TIMER)
2659 /* Generic routine for setting common part of quota structure */
2660 static int do_set_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2662 struct mem_dqblk *dm = &dquot->dq_dqb;
2663 int check_blim = 0, check_ilim = 0;
2664 struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
2666 if (di->d_fieldmask & ~VFS_QC_MASK)
2669 if (((di->d_fieldmask & QC_SPC_SOFT) &&
2670 di->d_spc_softlimit > dqi->dqi_max_spc_limit) ||
2671 ((di->d_fieldmask & QC_SPC_HARD) &&
2672 di->d_spc_hardlimit > dqi->dqi_max_spc_limit) ||
2673 ((di->d_fieldmask & QC_INO_SOFT) &&
2674 (di->d_ino_softlimit > dqi->dqi_max_ino_limit)) ||
2675 ((di->d_fieldmask & QC_INO_HARD) &&
2676 (di->d_ino_hardlimit > dqi->dqi_max_ino_limit)))
2679 spin_lock(&dquot->dq_dqb_lock);
2680 if (di->d_fieldmask & QC_SPACE) {
2681 dm->dqb_curspace = di->d_space - dm->dqb_rsvspace;
2683 set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2686 if (di->d_fieldmask & QC_SPC_SOFT)
2687 dm->dqb_bsoftlimit = di->d_spc_softlimit;
2688 if (di->d_fieldmask & QC_SPC_HARD)
2689 dm->dqb_bhardlimit = di->d_spc_hardlimit;
2690 if (di->d_fieldmask & (QC_SPC_SOFT | QC_SPC_HARD)) {
2692 set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2695 if (di->d_fieldmask & QC_INO_COUNT) {
2696 dm->dqb_curinodes = di->d_ino_count;
2698 set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2701 if (di->d_fieldmask & QC_INO_SOFT)
2702 dm->dqb_isoftlimit = di->d_ino_softlimit;
2703 if (di->d_fieldmask & QC_INO_HARD)
2704 dm->dqb_ihardlimit = di->d_ino_hardlimit;
2705 if (di->d_fieldmask & (QC_INO_SOFT | QC_INO_HARD)) {
2707 set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2710 if (di->d_fieldmask & QC_SPC_TIMER) {
2711 dm->dqb_btime = di->d_spc_timer;
2713 set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2716 if (di->d_fieldmask & QC_INO_TIMER) {
2717 dm->dqb_itime = di->d_ino_timer;
2719 set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2723 if (!dm->dqb_bsoftlimit ||
2724 dm->dqb_curspace + dm->dqb_rsvspace < dm->dqb_bsoftlimit) {
2726 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2727 } else if (!(di->d_fieldmask & QC_SPC_TIMER))
2728 /* Set grace only if user hasn't provided his own... */
2729 dm->dqb_btime = ktime_get_real_seconds() + dqi->dqi_bgrace;
2732 if (!dm->dqb_isoftlimit ||
2733 dm->dqb_curinodes < dm->dqb_isoftlimit) {
2735 clear_bit(DQ_INODES_B, &dquot->dq_flags);
2736 } else if (!(di->d_fieldmask & QC_INO_TIMER))
2737 /* Set grace only if user hasn't provided his own... */
2738 dm->dqb_itime = ktime_get_real_seconds() + dqi->dqi_igrace;
2740 if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2742 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2744 set_bit(DQ_FAKE_B, &dquot->dq_flags);
2745 spin_unlock(&dquot->dq_dqb_lock);
2746 mark_dquot_dirty(dquot);
2751 int dquot_set_dqblk(struct super_block *sb, struct kqid qid,
2752 struct qc_dqblk *di)
2754 struct dquot *dquot;
2757 dquot = dqget(sb, qid);
2758 if (IS_ERR(dquot)) {
2759 rc = PTR_ERR(dquot);
2762 rc = do_set_dqblk(dquot, di);
2767 EXPORT_SYMBOL(dquot_set_dqblk);
2769 /* Generic routine for getting common part of quota file information */
2770 int dquot_get_state(struct super_block *sb, struct qc_state *state)
2772 struct mem_dqinfo *mi;
2773 struct qc_type_state *tstate;
2774 struct quota_info *dqopt = sb_dqopt(sb);
2777 memset(state, 0, sizeof(*state));
2778 for (type = 0; type < MAXQUOTAS; type++) {
2779 if (!sb_has_quota_active(sb, type))
2781 tstate = state->s_state + type;
2782 mi = sb_dqopt(sb)->info + type;
2783 tstate->flags = QCI_ACCT_ENABLED;
2784 spin_lock(&dq_data_lock);
2785 if (mi->dqi_flags & DQF_SYS_FILE)
2786 tstate->flags |= QCI_SYSFILE;
2787 if (mi->dqi_flags & DQF_ROOT_SQUASH)
2788 tstate->flags |= QCI_ROOT_SQUASH;
2789 if (sb_has_quota_limits_enabled(sb, type))
2790 tstate->flags |= QCI_LIMITS_ENFORCED;
2791 tstate->spc_timelimit = mi->dqi_bgrace;
2792 tstate->ino_timelimit = mi->dqi_igrace;
2793 tstate->ino = dqopt->files[type]->i_ino;
2794 tstate->blocks = dqopt->files[type]->i_blocks;
2795 tstate->nextents = 1; /* We don't know... */
2796 spin_unlock(&dq_data_lock);
2800 EXPORT_SYMBOL(dquot_get_state);
2802 /* Generic routine for setting common part of quota file information */
2803 int dquot_set_dqinfo(struct super_block *sb, int type, struct qc_info *ii)
2805 struct mem_dqinfo *mi;
2808 if ((ii->i_fieldmask & QC_WARNS_MASK) ||
2809 (ii->i_fieldmask & QC_RT_SPC_TIMER))
2811 if (!sb_has_quota_active(sb, type))
2813 mi = sb_dqopt(sb)->info + type;
2814 if (ii->i_fieldmask & QC_FLAGS) {
2815 if ((ii->i_flags & QCI_ROOT_SQUASH &&
2816 mi->dqi_format->qf_fmt_id != QFMT_VFS_OLD))
2819 spin_lock(&dq_data_lock);
2820 if (ii->i_fieldmask & QC_SPC_TIMER)
2821 mi->dqi_bgrace = ii->i_spc_timelimit;
2822 if (ii->i_fieldmask & QC_INO_TIMER)
2823 mi->dqi_igrace = ii->i_ino_timelimit;
2824 if (ii->i_fieldmask & QC_FLAGS) {
2825 if (ii->i_flags & QCI_ROOT_SQUASH)
2826 mi->dqi_flags |= DQF_ROOT_SQUASH;
2828 mi->dqi_flags &= ~DQF_ROOT_SQUASH;
2830 spin_unlock(&dq_data_lock);
2831 mark_info_dirty(sb, type);
2832 /* Force write to disk */
2833 sb->dq_op->write_info(sb, type);
2836 EXPORT_SYMBOL(dquot_set_dqinfo);
2838 const struct quotactl_ops dquot_quotactl_sysfile_ops = {
2839 .quota_enable = dquot_quota_enable,
2840 .quota_disable = dquot_quota_disable,
2841 .quota_sync = dquot_quota_sync,
2842 .get_state = dquot_get_state,
2843 .set_info = dquot_set_dqinfo,
2844 .get_dqblk = dquot_get_dqblk,
2845 .get_nextdqblk = dquot_get_next_dqblk,
2846 .set_dqblk = dquot_set_dqblk
2848 EXPORT_SYMBOL(dquot_quotactl_sysfile_ops);
2850 static int do_proc_dqstats(struct ctl_table *table, int write,
2851 void __user *buffer, size_t *lenp, loff_t *ppos)
2853 unsigned int type = (unsigned long *)table->data - dqstats.stat;
2854 s64 value = percpu_counter_sum(&dqstats.counter[type]);
2856 /* Filter negative values for non-monotonic counters */
2857 if (value < 0 && (type == DQST_ALLOC_DQUOTS ||
2858 type == DQST_FREE_DQUOTS))
2861 /* Update global table */
2862 dqstats.stat[type] = value;
2863 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
2866 static struct ctl_table fs_dqstats_table[] = {
2868 .procname = "lookups",
2869 .data = &dqstats.stat[DQST_LOOKUPS],
2870 .maxlen = sizeof(unsigned long),
2872 .proc_handler = do_proc_dqstats,
2875 .procname = "drops",
2876 .data = &dqstats.stat[DQST_DROPS],
2877 .maxlen = sizeof(unsigned long),
2879 .proc_handler = do_proc_dqstats,
2882 .procname = "reads",
2883 .data = &dqstats.stat[DQST_READS],
2884 .maxlen = sizeof(unsigned long),
2886 .proc_handler = do_proc_dqstats,
2889 .procname = "writes",
2890 .data = &dqstats.stat[DQST_WRITES],
2891 .maxlen = sizeof(unsigned long),
2893 .proc_handler = do_proc_dqstats,
2896 .procname = "cache_hits",
2897 .data = &dqstats.stat[DQST_CACHE_HITS],
2898 .maxlen = sizeof(unsigned long),
2900 .proc_handler = do_proc_dqstats,
2903 .procname = "allocated_dquots",
2904 .data = &dqstats.stat[DQST_ALLOC_DQUOTS],
2905 .maxlen = sizeof(unsigned long),
2907 .proc_handler = do_proc_dqstats,
2910 .procname = "free_dquots",
2911 .data = &dqstats.stat[DQST_FREE_DQUOTS],
2912 .maxlen = sizeof(unsigned long),
2914 .proc_handler = do_proc_dqstats,
2917 .procname = "syncs",
2918 .data = &dqstats.stat[DQST_SYNCS],
2919 .maxlen = sizeof(unsigned long),
2921 .proc_handler = do_proc_dqstats,
2923 #ifdef CONFIG_PRINT_QUOTA_WARNING
2925 .procname = "warnings",
2926 .data = &flag_print_warnings,
2927 .maxlen = sizeof(int),
2929 .proc_handler = proc_dointvec,
2935 static struct ctl_table fs_table[] = {
2937 .procname = "quota",
2939 .child = fs_dqstats_table,
2944 static struct ctl_table sys_table[] = {
2953 static int __init dquot_init(void)
2956 unsigned long nr_hash, order;
2958 printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2960 register_sysctl_table(sys_table);
2962 dquot_cachep = kmem_cache_create("dquot",
2963 sizeof(struct dquot), sizeof(unsigned long) * 4,
2964 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2965 SLAB_MEM_SPREAD|SLAB_PANIC),
2969 dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
2971 panic("Cannot create dquot hash table");
2973 for (i = 0; i < _DQST_DQSTAT_LAST; i++) {
2974 ret = percpu_counter_init(&dqstats.counter[i], 0, GFP_KERNEL);
2976 panic("Cannot create dquot stat counters");
2979 /* Find power-of-two hlist_heads which can fit into allocation */
2980 nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2984 } while (nr_hash >> dq_hash_bits);
2987 nr_hash = 1UL << dq_hash_bits;
2988 dq_hash_mask = nr_hash - 1;
2989 for (i = 0; i < nr_hash; i++)
2990 INIT_HLIST_HEAD(dquot_hash + i);
2992 pr_info("VFS: Dquot-cache hash table entries: %ld (order %ld,"
2993 " %ld bytes)\n", nr_hash, order, (PAGE_SIZE << order));
2995 if (register_shrinker(&dqcache_shrinker))
2996 panic("Cannot register dquot shrinker");
3000 fs_initcall(dquot_init);