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);
651 * Clear dirty bit anyway to avoid infinite
654 clear_dquot_dirty(dquot);
659 spin_lock(&dq_list_lock);
661 spin_unlock(&dq_list_lock);
664 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
665 if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
666 && info_dirty(&dqopt->info[cnt]))
667 sb->dq_op->write_info(sb, cnt);
668 dqstats_inc(DQST_SYNCS);
672 EXPORT_SYMBOL(dquot_writeback_dquots);
674 /* Write all dquot structures to disk and make them visible from userspace */
675 int dquot_quota_sync(struct super_block *sb, int type)
677 struct quota_info *dqopt = sb_dqopt(sb);
681 ret = dquot_writeback_dquots(sb, type);
684 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
687 /* This is not very clever (and fast) but currently I don't know about
688 * any other simple way of getting quota data to disk and we must get
689 * them there for userspace to be visible... */
690 if (sb->s_op->sync_fs) {
691 ret = sb->s_op->sync_fs(sb, 1);
695 ret = sync_blockdev(sb->s_bdev);
700 * Now when everything is written we can discard the pagecache so
701 * that userspace sees the changes.
703 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
704 if (type != -1 && cnt != type)
706 if (!sb_has_quota_active(sb, cnt))
708 inode_lock(dqopt->files[cnt]);
709 truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
710 inode_unlock(dqopt->files[cnt]);
715 EXPORT_SYMBOL(dquot_quota_sync);
718 dqcache_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
721 unsigned long freed = 0;
723 spin_lock(&dq_list_lock);
724 while (!list_empty(&free_dquots) && sc->nr_to_scan) {
725 dquot = list_first_entry(&free_dquots, struct dquot, dq_free);
726 remove_dquot_hash(dquot);
727 remove_free_dquot(dquot);
729 do_destroy_dquot(dquot);
733 spin_unlock(&dq_list_lock);
738 dqcache_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
740 return vfs_pressure_ratio(
741 percpu_counter_read_positive(&dqstats.counter[DQST_FREE_DQUOTS]));
744 static struct shrinker dqcache_shrinker = {
745 .count_objects = dqcache_shrink_count,
746 .scan_objects = dqcache_shrink_scan,
747 .seeks = DEFAULT_SEEKS,
751 * Put reference to dquot
753 void dqput(struct dquot *dquot)
759 #ifdef CONFIG_QUOTA_DEBUG
760 if (!atomic_read(&dquot->dq_count)) {
761 quota_error(dquot->dq_sb, "trying to free free dquot of %s %d",
762 quotatypes[dquot->dq_id.type],
763 from_kqid(&init_user_ns, dquot->dq_id));
767 dqstats_inc(DQST_DROPS);
769 spin_lock(&dq_list_lock);
770 if (atomic_read(&dquot->dq_count) > 1) {
771 /* We have more than one user... nothing to do */
772 atomic_dec(&dquot->dq_count);
773 /* Releasing dquot during quotaoff phase? */
774 if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_id.type) &&
775 atomic_read(&dquot->dq_count) == 1)
776 wake_up(&dquot_ref_wq);
777 spin_unlock(&dq_list_lock);
780 /* Need to release dquot? */
781 if (dquot_dirty(dquot)) {
782 spin_unlock(&dq_list_lock);
783 /* Commit dquot before releasing */
784 ret = dquot->dq_sb->dq_op->write_dquot(dquot);
786 quota_error(dquot->dq_sb, "Can't write quota structure"
787 " (error %d). Quota may get out of sync!",
790 * We clear dirty bit anyway, so that we avoid
793 clear_dquot_dirty(dquot);
797 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
798 spin_unlock(&dq_list_lock);
799 dquot->dq_sb->dq_op->release_dquot(dquot);
802 atomic_dec(&dquot->dq_count);
803 #ifdef CONFIG_QUOTA_DEBUG
805 BUG_ON(!list_empty(&dquot->dq_free));
807 put_dquot_last(dquot);
808 spin_unlock(&dq_list_lock);
810 EXPORT_SYMBOL(dqput);
812 struct dquot *dquot_alloc(struct super_block *sb, int type)
814 return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
816 EXPORT_SYMBOL(dquot_alloc);
818 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
822 dquot = sb->dq_op->alloc_dquot(sb, type);
826 mutex_init(&dquot->dq_lock);
827 INIT_LIST_HEAD(&dquot->dq_free);
828 INIT_LIST_HEAD(&dquot->dq_inuse);
829 INIT_HLIST_NODE(&dquot->dq_hash);
830 INIT_LIST_HEAD(&dquot->dq_dirty);
832 dquot->dq_id = make_kqid_invalid(type);
833 atomic_set(&dquot->dq_count, 1);
834 spin_lock_init(&dquot->dq_dqb_lock);
840 * Get reference to dquot
842 * Locking is slightly tricky here. We are guarded from parallel quotaoff()
843 * destroying our dquot by:
844 * a) checking for quota flags under dq_list_lock and
845 * b) getting a reference to dquot before we release dq_list_lock
847 struct dquot *dqget(struct super_block *sb, struct kqid qid)
849 unsigned int hashent = hashfn(sb, qid);
850 struct dquot *dquot, *empty = NULL;
852 if (!qid_has_mapping(sb->s_user_ns, qid))
853 return ERR_PTR(-EINVAL);
855 if (!sb_has_quota_active(sb, qid.type))
856 return ERR_PTR(-ESRCH);
858 spin_lock(&dq_list_lock);
859 spin_lock(&dq_state_lock);
860 if (!sb_has_quota_active(sb, qid.type)) {
861 spin_unlock(&dq_state_lock);
862 spin_unlock(&dq_list_lock);
863 dquot = ERR_PTR(-ESRCH);
866 spin_unlock(&dq_state_lock);
868 dquot = find_dquot(hashent, sb, qid);
871 spin_unlock(&dq_list_lock);
872 empty = get_empty_dquot(sb, qid.type);
874 schedule(); /* Try to wait for a moment... */
880 /* all dquots go on the inuse_list */
882 /* hash it first so it can be found */
883 insert_dquot_hash(dquot);
884 spin_unlock(&dq_list_lock);
885 dqstats_inc(DQST_LOOKUPS);
887 if (!atomic_read(&dquot->dq_count))
888 remove_free_dquot(dquot);
889 atomic_inc(&dquot->dq_count);
890 spin_unlock(&dq_list_lock);
891 dqstats_inc(DQST_CACHE_HITS);
892 dqstats_inc(DQST_LOOKUPS);
894 /* Wait for dq_lock - after this we know that either dquot_release() is
895 * already finished or it will be canceled due to dq_count > 1 test */
896 wait_on_dquot(dquot);
897 /* Read the dquot / allocate space in quota file */
898 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
901 err = sb->dq_op->acquire_dquot(dquot);
904 dquot = ERR_PTR(err);
909 * Make sure following reads see filled structure - paired with
910 * smp_mb__before_atomic() in dquot_acquire().
913 #ifdef CONFIG_QUOTA_DEBUG
914 BUG_ON(!dquot->dq_sb); /* Has somebody invalidated entry under us? */
918 do_destroy_dquot(empty);
922 EXPORT_SYMBOL(dqget);
924 static inline struct dquot **i_dquot(struct inode *inode)
926 return inode->i_sb->s_op->get_dquots(inode);
929 static int dqinit_needed(struct inode *inode, int type)
931 struct dquot * const *dquots;
934 if (IS_NOQUOTA(inode))
937 dquots = i_dquot(inode);
939 return !dquots[type];
940 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
946 /* This routine is guarded by s_umount semaphore */
947 static int add_dquot_ref(struct super_block *sb, int type)
949 struct inode *inode, *old_inode = NULL;
950 #ifdef CONFIG_QUOTA_DEBUG
955 spin_lock(&sb->s_inode_list_lock);
956 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
957 spin_lock(&inode->i_lock);
958 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
959 !atomic_read(&inode->i_writecount) ||
960 !dqinit_needed(inode, type)) {
961 spin_unlock(&inode->i_lock);
965 spin_unlock(&inode->i_lock);
966 spin_unlock(&sb->s_inode_list_lock);
968 #ifdef CONFIG_QUOTA_DEBUG
969 if (unlikely(inode_get_rsv_space(inode) > 0))
973 err = __dquot_initialize(inode, type);
980 * We hold a reference to 'inode' so it couldn't have been
981 * removed from s_inodes list while we dropped the
982 * s_inode_list_lock. We cannot iput the inode now as we can be
983 * holding the last reference and we cannot iput it under
984 * s_inode_list_lock. So we keep the reference and iput it
989 spin_lock(&sb->s_inode_list_lock);
991 spin_unlock(&sb->s_inode_list_lock);
994 #ifdef CONFIG_QUOTA_DEBUG
996 quota_error(sb, "Writes happened before quota was turned on "
997 "thus quota information is probably inconsistent. "
998 "Please run quotacheck(8)");
1005 * Remove references to dquots from inode and add dquot to list for freeing
1006 * if we have the last reference to dquot
1008 static void remove_inode_dquot_ref(struct inode *inode, int type,
1009 struct list_head *tofree_head)
1011 struct dquot **dquots = i_dquot(inode);
1012 struct dquot *dquot = dquots[type];
1017 dquots[type] = NULL;
1018 if (list_empty(&dquot->dq_free)) {
1020 * The inode still has reference to dquot so it can't be in the
1023 spin_lock(&dq_list_lock);
1024 list_add(&dquot->dq_free, tofree_head);
1025 spin_unlock(&dq_list_lock);
1028 * Dquot is already in a list to put so we won't drop the last
1036 * Free list of dquots
1037 * Dquots are removed from inodes and no new references can be got so we are
1038 * the only ones holding reference
1040 static void put_dquot_list(struct list_head *tofree_head)
1042 struct list_head *act_head;
1043 struct dquot *dquot;
1045 act_head = tofree_head->next;
1046 while (act_head != tofree_head) {
1047 dquot = list_entry(act_head, struct dquot, dq_free);
1048 act_head = act_head->next;
1049 /* Remove dquot from the list so we won't have problems... */
1050 list_del_init(&dquot->dq_free);
1055 static void remove_dquot_ref(struct super_block *sb, int type,
1056 struct list_head *tofree_head)
1058 struct inode *inode;
1061 spin_lock(&sb->s_inode_list_lock);
1062 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1064 * We have to scan also I_NEW inodes because they can already
1065 * have quota pointer initialized. Luckily, we need to touch
1066 * only quota pointers and these have separate locking
1069 spin_lock(&dq_data_lock);
1070 if (!IS_NOQUOTA(inode)) {
1071 if (unlikely(inode_get_rsv_space(inode) > 0))
1073 remove_inode_dquot_ref(inode, type, tofree_head);
1075 spin_unlock(&dq_data_lock);
1077 spin_unlock(&sb->s_inode_list_lock);
1078 #ifdef CONFIG_QUOTA_DEBUG
1080 printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1081 " was disabled thus quota information is probably "
1082 "inconsistent. Please run quotacheck(8).\n", sb->s_id);
1087 /* Gather all references from inodes and drop them */
1088 static void drop_dquot_ref(struct super_block *sb, int type)
1090 LIST_HEAD(tofree_head);
1093 remove_dquot_ref(sb, type, &tofree_head);
1094 synchronize_srcu(&dquot_srcu);
1095 put_dquot_list(&tofree_head);
1100 void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1102 if (dquot->dq_dqb.dqb_rsvspace >= number)
1103 dquot->dq_dqb.dqb_rsvspace -= number;
1106 dquot->dq_dqb.dqb_rsvspace = 0;
1108 if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1109 dquot->dq_dqb.dqb_bsoftlimit)
1110 dquot->dq_dqb.dqb_btime = (time64_t) 0;
1111 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1114 static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1116 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1117 dquot->dq_dqb.dqb_curinodes >= number)
1118 dquot->dq_dqb.dqb_curinodes -= number;
1120 dquot->dq_dqb.dqb_curinodes = 0;
1121 if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1122 dquot->dq_dqb.dqb_itime = (time64_t) 0;
1123 clear_bit(DQ_INODES_B, &dquot->dq_flags);
1126 static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1128 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1129 dquot->dq_dqb.dqb_curspace >= number)
1130 dquot->dq_dqb.dqb_curspace -= number;
1132 dquot->dq_dqb.dqb_curspace = 0;
1133 if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1134 dquot->dq_dqb.dqb_bsoftlimit)
1135 dquot->dq_dqb.dqb_btime = (time64_t) 0;
1136 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1140 struct super_block *w_sb;
1141 struct kqid w_dq_id;
1145 static int warning_issued(struct dquot *dquot, const int warntype)
1147 int flag = (warntype == QUOTA_NL_BHARDWARN ||
1148 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1149 ((warntype == QUOTA_NL_IHARDWARN ||
1150 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1154 return test_and_set_bit(flag, &dquot->dq_flags);
1157 #ifdef CONFIG_PRINT_QUOTA_WARNING
1158 static int flag_print_warnings = 1;
1160 static int need_print_warning(struct dquot_warn *warn)
1162 if (!flag_print_warnings)
1165 switch (warn->w_dq_id.type) {
1167 return uid_eq(current_fsuid(), warn->w_dq_id.uid);
1169 return in_group_p(warn->w_dq_id.gid);
1176 /* Print warning to user which exceeded quota */
1177 static void print_warning(struct dquot_warn *warn)
1180 struct tty_struct *tty;
1181 int warntype = warn->w_type;
1183 if (warntype == QUOTA_NL_IHARDBELOW ||
1184 warntype == QUOTA_NL_ISOFTBELOW ||
1185 warntype == QUOTA_NL_BHARDBELOW ||
1186 warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(warn))
1189 tty = get_current_tty();
1192 tty_write_message(tty, warn->w_sb->s_id);
1193 if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1194 tty_write_message(tty, ": warning, ");
1196 tty_write_message(tty, ": write failed, ");
1197 tty_write_message(tty, quotatypes[warn->w_dq_id.type]);
1199 case QUOTA_NL_IHARDWARN:
1200 msg = " file limit reached.\r\n";
1202 case QUOTA_NL_ISOFTLONGWARN:
1203 msg = " file quota exceeded too long.\r\n";
1205 case QUOTA_NL_ISOFTWARN:
1206 msg = " file quota exceeded.\r\n";
1208 case QUOTA_NL_BHARDWARN:
1209 msg = " block limit reached.\r\n";
1211 case QUOTA_NL_BSOFTLONGWARN:
1212 msg = " block quota exceeded too long.\r\n";
1214 case QUOTA_NL_BSOFTWARN:
1215 msg = " block quota exceeded.\r\n";
1218 tty_write_message(tty, msg);
1223 static void prepare_warning(struct dquot_warn *warn, struct dquot *dquot,
1226 if (warning_issued(dquot, warntype))
1228 warn->w_type = warntype;
1229 warn->w_sb = dquot->dq_sb;
1230 warn->w_dq_id = dquot->dq_id;
1234 * Write warnings to the console and send warning messages over netlink.
1236 * Note that this function can call into tty and networking code.
1238 static void flush_warnings(struct dquot_warn *warn)
1242 for (i = 0; i < MAXQUOTAS; i++) {
1243 if (warn[i].w_type == QUOTA_NL_NOWARN)
1245 #ifdef CONFIG_PRINT_QUOTA_WARNING
1246 print_warning(&warn[i]);
1248 quota_send_warning(warn[i].w_dq_id,
1249 warn[i].w_sb->s_dev, warn[i].w_type);
1253 static int ignore_hardlimit(struct dquot *dquot)
1255 struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
1257 return capable(CAP_SYS_RESOURCE) &&
1258 (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1259 !(info->dqi_flags & DQF_ROOT_SQUASH));
1262 static int dquot_add_inodes(struct dquot *dquot, qsize_t inodes,
1263 struct dquot_warn *warn)
1268 spin_lock(&dquot->dq_dqb_lock);
1269 newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1270 if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type) ||
1271 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1274 if (dquot->dq_dqb.dqb_ihardlimit &&
1275 newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1276 !ignore_hardlimit(dquot)) {
1277 prepare_warning(warn, dquot, QUOTA_NL_IHARDWARN);
1282 if (dquot->dq_dqb.dqb_isoftlimit &&
1283 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1284 dquot->dq_dqb.dqb_itime &&
1285 ktime_get_real_seconds() >= dquot->dq_dqb.dqb_itime &&
1286 !ignore_hardlimit(dquot)) {
1287 prepare_warning(warn, dquot, QUOTA_NL_ISOFTLONGWARN);
1292 if (dquot->dq_dqb.dqb_isoftlimit &&
1293 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1294 dquot->dq_dqb.dqb_itime == 0) {
1295 prepare_warning(warn, dquot, QUOTA_NL_ISOFTWARN);
1296 dquot->dq_dqb.dqb_itime = ktime_get_real_seconds() +
1297 sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type].dqi_igrace;
1300 dquot->dq_dqb.dqb_curinodes = newinodes;
1303 spin_unlock(&dquot->dq_dqb_lock);
1307 static int dquot_add_space(struct dquot *dquot, qsize_t space,
1308 qsize_t rsv_space, unsigned int flags,
1309 struct dquot_warn *warn)
1312 struct super_block *sb = dquot->dq_sb;
1315 spin_lock(&dquot->dq_dqb_lock);
1316 if (!sb_has_quota_limits_enabled(sb, dquot->dq_id.type) ||
1317 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1320 tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1321 + space + rsv_space;
1323 if (dquot->dq_dqb.dqb_bhardlimit &&
1324 tspace > dquot->dq_dqb.dqb_bhardlimit &&
1325 !ignore_hardlimit(dquot)) {
1326 if (flags & DQUOT_SPACE_WARN)
1327 prepare_warning(warn, dquot, QUOTA_NL_BHARDWARN);
1332 if (dquot->dq_dqb.dqb_bsoftlimit &&
1333 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1334 dquot->dq_dqb.dqb_btime &&
1335 ktime_get_real_seconds() >= dquot->dq_dqb.dqb_btime &&
1336 !ignore_hardlimit(dquot)) {
1337 if (flags & DQUOT_SPACE_WARN)
1338 prepare_warning(warn, dquot, QUOTA_NL_BSOFTLONGWARN);
1343 if (dquot->dq_dqb.dqb_bsoftlimit &&
1344 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1345 dquot->dq_dqb.dqb_btime == 0) {
1346 if (flags & DQUOT_SPACE_WARN) {
1347 prepare_warning(warn, dquot, QUOTA_NL_BSOFTWARN);
1348 dquot->dq_dqb.dqb_btime = ktime_get_real_seconds() +
1349 sb_dqopt(sb)->info[dquot->dq_id.type].dqi_bgrace;
1352 * We don't allow preallocation to exceed softlimit so exceeding will
1361 * We have to be careful and go through warning generation & grace time
1362 * setting even if DQUOT_SPACE_NOFAIL is set. That's why we check it
1365 if (flags & DQUOT_SPACE_NOFAIL)
1368 dquot->dq_dqb.dqb_rsvspace += rsv_space;
1369 dquot->dq_dqb.dqb_curspace += space;
1371 spin_unlock(&dquot->dq_dqb_lock);
1375 static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1379 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1380 dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1381 !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type))
1382 return QUOTA_NL_NOWARN;
1384 newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1385 if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1386 return QUOTA_NL_ISOFTBELOW;
1387 if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1388 newinodes < dquot->dq_dqb.dqb_ihardlimit)
1389 return QUOTA_NL_IHARDBELOW;
1390 return QUOTA_NL_NOWARN;
1393 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1397 tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace;
1399 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1400 tspace <= dquot->dq_dqb.dqb_bsoftlimit)
1401 return QUOTA_NL_NOWARN;
1403 if (tspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1404 return QUOTA_NL_BSOFTBELOW;
1405 if (tspace >= dquot->dq_dqb.dqb_bhardlimit &&
1406 tspace - space < dquot->dq_dqb.dqb_bhardlimit)
1407 return QUOTA_NL_BHARDBELOW;
1408 return QUOTA_NL_NOWARN;
1411 static int dquot_active(const struct inode *inode)
1413 struct super_block *sb = inode->i_sb;
1415 if (IS_NOQUOTA(inode))
1417 return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1421 * Initialize quota pointers in inode
1423 * It is better to call this function outside of any transaction as it
1424 * might need a lot of space in journal for dquot structure allocation.
1426 static int __dquot_initialize(struct inode *inode, int type)
1428 int cnt, init_needed = 0;
1429 struct dquot **dquots, *got[MAXQUOTAS] = {};
1430 struct super_block *sb = inode->i_sb;
1434 if (!dquot_active(inode))
1437 dquots = i_dquot(inode);
1439 /* First get references to structures we might need. */
1440 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1444 struct dquot *dquot;
1446 if (type != -1 && cnt != type)
1449 * The i_dquot should have been initialized in most cases,
1450 * we check it without locking here to avoid unnecessary
1451 * dqget()/dqput() calls.
1456 if (!sb_has_quota_active(sb, cnt))
1463 qid = make_kqid_uid(inode->i_uid);
1466 qid = make_kqid_gid(inode->i_gid);
1469 rc = inode->i_sb->dq_op->get_projid(inode, &projid);
1472 qid = make_kqid_projid(projid);
1475 dquot = dqget(sb, qid);
1476 if (IS_ERR(dquot)) {
1477 /* We raced with somebody turning quotas off... */
1478 if (PTR_ERR(dquot) != -ESRCH) {
1479 ret = PTR_ERR(dquot);
1487 /* All required i_dquot has been initialized */
1491 spin_lock(&dq_data_lock);
1492 if (IS_NOQUOTA(inode))
1494 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1495 if (type != -1 && cnt != type)
1497 /* Avoid races with quotaoff() */
1498 if (!sb_has_quota_active(sb, cnt))
1500 /* We could race with quotaon or dqget() could have failed */
1504 dquots[cnt] = got[cnt];
1507 * Make quota reservation system happy if someone
1508 * did a write before quota was turned on
1510 rsv = inode_get_rsv_space(inode);
1511 if (unlikely(rsv)) {
1512 spin_lock(&inode->i_lock);
1513 /* Get reservation again under proper lock */
1514 rsv = __inode_get_rsv_space(inode);
1515 spin_lock(&dquots[cnt]->dq_dqb_lock);
1516 dquots[cnt]->dq_dqb.dqb_rsvspace += rsv;
1517 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1518 spin_unlock(&inode->i_lock);
1523 spin_unlock(&dq_data_lock);
1525 /* Drop unused references */
1531 int dquot_initialize(struct inode *inode)
1533 return __dquot_initialize(inode, -1);
1535 EXPORT_SYMBOL(dquot_initialize);
1537 bool dquot_initialize_needed(struct inode *inode)
1539 struct dquot **dquots;
1542 if (!dquot_active(inode))
1545 dquots = i_dquot(inode);
1546 for (i = 0; i < MAXQUOTAS; i++)
1547 if (!dquots[i] && sb_has_quota_active(inode->i_sb, i))
1551 EXPORT_SYMBOL(dquot_initialize_needed);
1554 * Release all quotas referenced by inode.
1556 * This function only be called on inode free or converting
1557 * a file to quota file, no other users for the i_dquot in
1558 * both cases, so we needn't call synchronize_srcu() after
1561 static void __dquot_drop(struct inode *inode)
1564 struct dquot **dquots = i_dquot(inode);
1565 struct dquot *put[MAXQUOTAS];
1567 spin_lock(&dq_data_lock);
1568 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1569 put[cnt] = dquots[cnt];
1572 spin_unlock(&dq_data_lock);
1576 void dquot_drop(struct inode *inode)
1578 struct dquot * const *dquots;
1581 if (IS_NOQUOTA(inode))
1585 * Test before calling to rule out calls from proc and such
1586 * where we are not allowed to block. Note that this is
1587 * actually reliable test even without the lock - the caller
1588 * must assure that nobody can come after the DQUOT_DROP and
1589 * add quota pointers back anyway.
1591 dquots = i_dquot(inode);
1592 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1597 if (cnt < MAXQUOTAS)
1598 __dquot_drop(inode);
1600 EXPORT_SYMBOL(dquot_drop);
1603 * inode_reserved_space is managed internally by quota, and protected by
1604 * i_lock similar to i_blocks+i_bytes.
1606 static qsize_t *inode_reserved_space(struct inode * inode)
1608 /* Filesystem must explicitly define it's own method in order to use
1609 * quota reservation interface */
1610 BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1611 return inode->i_sb->dq_op->get_reserved_space(inode);
1614 static qsize_t __inode_get_rsv_space(struct inode *inode)
1616 if (!inode->i_sb->dq_op->get_reserved_space)
1618 return *inode_reserved_space(inode);
1621 static qsize_t inode_get_rsv_space(struct inode *inode)
1625 if (!inode->i_sb->dq_op->get_reserved_space)
1627 spin_lock(&inode->i_lock);
1628 ret = __inode_get_rsv_space(inode);
1629 spin_unlock(&inode->i_lock);
1634 * This functions updates i_blocks+i_bytes fields and quota information
1635 * (together with appropriate checks).
1637 * NOTE: We absolutely rely on the fact that caller dirties the inode
1638 * (usually helpers in quotaops.h care about this) and holds a handle for
1639 * the current transaction so that dquot write and inode write go into the
1644 * This operation can block, but only after everything is updated
1646 int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
1648 int cnt, ret = 0, index;
1649 struct dquot_warn warn[MAXQUOTAS];
1650 int reserve = flags & DQUOT_SPACE_RESERVE;
1651 struct dquot **dquots;
1653 if (!dquot_active(inode)) {
1655 spin_lock(&inode->i_lock);
1656 *inode_reserved_space(inode) += number;
1657 spin_unlock(&inode->i_lock);
1659 inode_add_bytes(inode, number);
1664 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1665 warn[cnt].w_type = QUOTA_NL_NOWARN;
1667 dquots = i_dquot(inode);
1668 index = srcu_read_lock(&dquot_srcu);
1669 spin_lock(&inode->i_lock);
1670 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1673 if (flags & DQUOT_SPACE_RESERVE) {
1674 ret = dquot_add_space(dquots[cnt], 0, number, flags,
1677 ret = dquot_add_space(dquots[cnt], number, 0, flags,
1681 /* Back out changes we already did */
1682 for (cnt--; cnt >= 0; cnt--) {
1685 spin_lock(&dquots[cnt]->dq_dqb_lock);
1686 if (flags & DQUOT_SPACE_RESERVE) {
1687 dquots[cnt]->dq_dqb.dqb_rsvspace -=
1690 dquots[cnt]->dq_dqb.dqb_curspace -=
1693 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1695 spin_unlock(&inode->i_lock);
1696 goto out_flush_warn;
1700 *inode_reserved_space(inode) += number;
1702 __inode_add_bytes(inode, number);
1703 spin_unlock(&inode->i_lock);
1706 goto out_flush_warn;
1707 mark_all_dquot_dirty(dquots);
1709 srcu_read_unlock(&dquot_srcu, index);
1710 flush_warnings(warn);
1714 EXPORT_SYMBOL(__dquot_alloc_space);
1717 * This operation can block, but only after everything is updated
1719 int dquot_alloc_inode(struct inode *inode)
1721 int cnt, ret = 0, index;
1722 struct dquot_warn warn[MAXQUOTAS];
1723 struct dquot * const *dquots;
1725 if (!dquot_active(inode))
1727 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1728 warn[cnt].w_type = QUOTA_NL_NOWARN;
1730 dquots = i_dquot(inode);
1731 index = srcu_read_lock(&dquot_srcu);
1732 spin_lock(&inode->i_lock);
1733 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1736 ret = dquot_add_inodes(dquots[cnt], 1, &warn[cnt]);
1738 for (cnt--; cnt >= 0; cnt--) {
1741 /* Back out changes we already did */
1742 spin_lock(&dquots[cnt]->dq_dqb_lock);
1743 dquots[cnt]->dq_dqb.dqb_curinodes--;
1744 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1751 spin_unlock(&inode->i_lock);
1753 mark_all_dquot_dirty(dquots);
1754 srcu_read_unlock(&dquot_srcu, index);
1755 flush_warnings(warn);
1758 EXPORT_SYMBOL(dquot_alloc_inode);
1761 * Convert in-memory reserved quotas to real consumed quotas
1763 int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1765 struct dquot **dquots;
1768 if (!dquot_active(inode)) {
1769 spin_lock(&inode->i_lock);
1770 *inode_reserved_space(inode) -= number;
1771 __inode_add_bytes(inode, number);
1772 spin_unlock(&inode->i_lock);
1776 dquots = i_dquot(inode);
1777 index = srcu_read_lock(&dquot_srcu);
1778 spin_lock(&inode->i_lock);
1779 /* Claim reserved quotas to allocated quotas */
1780 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1782 struct dquot *dquot = dquots[cnt];
1784 spin_lock(&dquot->dq_dqb_lock);
1785 if (WARN_ON_ONCE(dquot->dq_dqb.dqb_rsvspace < number))
1786 number = dquot->dq_dqb.dqb_rsvspace;
1787 dquot->dq_dqb.dqb_curspace += number;
1788 dquot->dq_dqb.dqb_rsvspace -= number;
1789 spin_unlock(&dquot->dq_dqb_lock);
1792 /* Update inode bytes */
1793 *inode_reserved_space(inode) -= number;
1794 __inode_add_bytes(inode, number);
1795 spin_unlock(&inode->i_lock);
1796 mark_all_dquot_dirty(dquots);
1797 srcu_read_unlock(&dquot_srcu, index);
1800 EXPORT_SYMBOL(dquot_claim_space_nodirty);
1803 * Convert allocated space back to in-memory reserved quotas
1805 void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
1807 struct dquot **dquots;
1810 if (!dquot_active(inode)) {
1811 spin_lock(&inode->i_lock);
1812 *inode_reserved_space(inode) += number;
1813 __inode_sub_bytes(inode, number);
1814 spin_unlock(&inode->i_lock);
1818 dquots = i_dquot(inode);
1819 index = srcu_read_lock(&dquot_srcu);
1820 spin_lock(&inode->i_lock);
1821 /* Claim reserved quotas to allocated quotas */
1822 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1824 struct dquot *dquot = dquots[cnt];
1826 spin_lock(&dquot->dq_dqb_lock);
1827 if (WARN_ON_ONCE(dquot->dq_dqb.dqb_curspace < number))
1828 number = dquot->dq_dqb.dqb_curspace;
1829 dquot->dq_dqb.dqb_rsvspace += number;
1830 dquot->dq_dqb.dqb_curspace -= number;
1831 spin_unlock(&dquot->dq_dqb_lock);
1834 /* Update inode bytes */
1835 *inode_reserved_space(inode) += number;
1836 __inode_sub_bytes(inode, number);
1837 spin_unlock(&inode->i_lock);
1838 mark_all_dquot_dirty(dquots);
1839 srcu_read_unlock(&dquot_srcu, index);
1842 EXPORT_SYMBOL(dquot_reclaim_space_nodirty);
1845 * This operation can block, but only after everything is updated
1847 void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
1850 struct dquot_warn warn[MAXQUOTAS];
1851 struct dquot **dquots;
1852 int reserve = flags & DQUOT_SPACE_RESERVE, index;
1854 if (!dquot_active(inode)) {
1856 spin_lock(&inode->i_lock);
1857 *inode_reserved_space(inode) -= number;
1858 spin_unlock(&inode->i_lock);
1860 inode_sub_bytes(inode, number);
1865 dquots = i_dquot(inode);
1866 index = srcu_read_lock(&dquot_srcu);
1867 spin_lock(&inode->i_lock);
1868 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1871 warn[cnt].w_type = QUOTA_NL_NOWARN;
1874 spin_lock(&dquots[cnt]->dq_dqb_lock);
1875 wtype = info_bdq_free(dquots[cnt], number);
1876 if (wtype != QUOTA_NL_NOWARN)
1877 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1879 dquot_free_reserved_space(dquots[cnt], number);
1881 dquot_decr_space(dquots[cnt], number);
1882 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1885 *inode_reserved_space(inode) -= number;
1887 __inode_sub_bytes(inode, number);
1888 spin_unlock(&inode->i_lock);
1892 mark_all_dquot_dirty(dquots);
1894 srcu_read_unlock(&dquot_srcu, index);
1895 flush_warnings(warn);
1897 EXPORT_SYMBOL(__dquot_free_space);
1900 * This operation can block, but only after everything is updated
1902 void dquot_free_inode(struct inode *inode)
1905 struct dquot_warn warn[MAXQUOTAS];
1906 struct dquot * const *dquots;
1909 if (!dquot_active(inode))
1912 dquots = i_dquot(inode);
1913 index = srcu_read_lock(&dquot_srcu);
1914 spin_lock(&inode->i_lock);
1915 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1918 warn[cnt].w_type = QUOTA_NL_NOWARN;
1921 spin_lock(&dquots[cnt]->dq_dqb_lock);
1922 wtype = info_idq_free(dquots[cnt], 1);
1923 if (wtype != QUOTA_NL_NOWARN)
1924 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1925 dquot_decr_inodes(dquots[cnt], 1);
1926 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1928 spin_unlock(&inode->i_lock);
1929 mark_all_dquot_dirty(dquots);
1930 srcu_read_unlock(&dquot_srcu, index);
1931 flush_warnings(warn);
1933 EXPORT_SYMBOL(dquot_free_inode);
1936 * Transfer the number of inode and blocks from one diskquota to an other.
1937 * On success, dquot references in transfer_to are consumed and references
1938 * to original dquots that need to be released are placed there. On failure,
1939 * references are kept untouched.
1941 * This operation can block, but only after everything is updated
1942 * A transaction must be started when entering this function.
1944 * We are holding reference on transfer_from & transfer_to, no need to
1945 * protect them by srcu_read_lock().
1947 int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
1950 qsize_t rsv_space = 0;
1951 qsize_t inode_usage = 1;
1952 struct dquot *transfer_from[MAXQUOTAS] = {};
1954 char is_valid[MAXQUOTAS] = {};
1955 struct dquot_warn warn_to[MAXQUOTAS];
1956 struct dquot_warn warn_from_inodes[MAXQUOTAS];
1957 struct dquot_warn warn_from_space[MAXQUOTAS];
1959 if (IS_NOQUOTA(inode))
1962 if (inode->i_sb->dq_op->get_inode_usage) {
1963 ret = inode->i_sb->dq_op->get_inode_usage(inode, &inode_usage);
1968 /* Initialize the arrays */
1969 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1970 warn_to[cnt].w_type = QUOTA_NL_NOWARN;
1971 warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
1972 warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
1975 spin_lock(&dq_data_lock);
1976 spin_lock(&inode->i_lock);
1977 if (IS_NOQUOTA(inode)) { /* File without quota accounting? */
1978 spin_unlock(&inode->i_lock);
1979 spin_unlock(&dq_data_lock);
1982 cur_space = __inode_get_bytes(inode);
1983 rsv_space = __inode_get_rsv_space(inode);
1985 * Build the transfer_from list, check limits, and update usage in
1986 * the target structures.
1988 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1990 * Skip changes for same uid or gid or for turned off quota-type.
1992 if (!transfer_to[cnt])
1994 /* Avoid races with quotaoff() */
1995 if (!sb_has_quota_active(inode->i_sb, cnt))
1998 transfer_from[cnt] = i_dquot(inode)[cnt];
1999 ret = dquot_add_inodes(transfer_to[cnt], inode_usage,
2003 ret = dquot_add_space(transfer_to[cnt], cur_space, rsv_space,
2004 DQUOT_SPACE_WARN, &warn_to[cnt]);
2006 spin_lock(&transfer_to[cnt]->dq_dqb_lock);
2007 dquot_decr_inodes(transfer_to[cnt], inode_usage);
2008 spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2013 /* Decrease usage for source structures and update quota pointers */
2014 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2017 /* Due to IO error we might not have transfer_from[] structure */
2018 if (transfer_from[cnt]) {
2021 spin_lock(&transfer_from[cnt]->dq_dqb_lock);
2022 wtype = info_idq_free(transfer_from[cnt], inode_usage);
2023 if (wtype != QUOTA_NL_NOWARN)
2024 prepare_warning(&warn_from_inodes[cnt],
2025 transfer_from[cnt], wtype);
2026 wtype = info_bdq_free(transfer_from[cnt],
2027 cur_space + rsv_space);
2028 if (wtype != QUOTA_NL_NOWARN)
2029 prepare_warning(&warn_from_space[cnt],
2030 transfer_from[cnt], wtype);
2031 dquot_decr_inodes(transfer_from[cnt], inode_usage);
2032 dquot_decr_space(transfer_from[cnt], cur_space);
2033 dquot_free_reserved_space(transfer_from[cnt],
2035 spin_unlock(&transfer_from[cnt]->dq_dqb_lock);
2037 i_dquot(inode)[cnt] = transfer_to[cnt];
2039 spin_unlock(&inode->i_lock);
2040 spin_unlock(&dq_data_lock);
2042 mark_all_dquot_dirty(transfer_from);
2043 mark_all_dquot_dirty(transfer_to);
2044 flush_warnings(warn_to);
2045 flush_warnings(warn_from_inodes);
2046 flush_warnings(warn_from_space);
2047 /* Pass back references to put */
2048 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2050 transfer_to[cnt] = transfer_from[cnt];
2053 /* Back out changes we already did */
2054 for (cnt--; cnt >= 0; cnt--) {
2057 spin_lock(&transfer_to[cnt]->dq_dqb_lock);
2058 dquot_decr_inodes(transfer_to[cnt], inode_usage);
2059 dquot_decr_space(transfer_to[cnt], cur_space);
2060 dquot_free_reserved_space(transfer_to[cnt], rsv_space);
2061 spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2063 spin_unlock(&inode->i_lock);
2064 spin_unlock(&dq_data_lock);
2065 flush_warnings(warn_to);
2068 EXPORT_SYMBOL(__dquot_transfer);
2070 /* Wrapper for transferring ownership of an inode for uid/gid only
2071 * Called from FSXXX_setattr()
2073 int dquot_transfer(struct inode *inode, struct iattr *iattr)
2075 struct dquot *transfer_to[MAXQUOTAS] = {};
2076 struct dquot *dquot;
2077 struct super_block *sb = inode->i_sb;
2080 if (!dquot_active(inode))
2083 if (iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)){
2084 dquot = dqget(sb, make_kqid_uid(iattr->ia_uid));
2085 if (IS_ERR(dquot)) {
2086 if (PTR_ERR(dquot) != -ESRCH) {
2087 ret = PTR_ERR(dquot);
2092 transfer_to[USRQUOTA] = dquot;
2094 if (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid)){
2095 dquot = dqget(sb, make_kqid_gid(iattr->ia_gid));
2096 if (IS_ERR(dquot)) {
2097 if (PTR_ERR(dquot) != -ESRCH) {
2098 ret = PTR_ERR(dquot);
2103 transfer_to[GRPQUOTA] = dquot;
2105 ret = __dquot_transfer(inode, transfer_to);
2107 dqput_all(transfer_to);
2110 EXPORT_SYMBOL(dquot_transfer);
2113 * Write info of quota file to disk
2115 int dquot_commit_info(struct super_block *sb, int type)
2117 struct quota_info *dqopt = sb_dqopt(sb);
2119 return dqopt->ops[type]->write_file_info(sb, type);
2121 EXPORT_SYMBOL(dquot_commit_info);
2123 int dquot_get_next_id(struct super_block *sb, struct kqid *qid)
2125 struct quota_info *dqopt = sb_dqopt(sb);
2127 if (!sb_has_quota_active(sb, qid->type))
2129 if (!dqopt->ops[qid->type]->get_next_id)
2131 return dqopt->ops[qid->type]->get_next_id(sb, qid);
2133 EXPORT_SYMBOL(dquot_get_next_id);
2136 * Definitions of diskquota operations.
2138 const struct dquot_operations dquot_operations = {
2139 .write_dquot = dquot_commit,
2140 .acquire_dquot = dquot_acquire,
2141 .release_dquot = dquot_release,
2142 .mark_dirty = dquot_mark_dquot_dirty,
2143 .write_info = dquot_commit_info,
2144 .alloc_dquot = dquot_alloc,
2145 .destroy_dquot = dquot_destroy,
2146 .get_next_id = dquot_get_next_id,
2148 EXPORT_SYMBOL(dquot_operations);
2151 * Generic helper for ->open on filesystems supporting disk quotas.
2153 int dquot_file_open(struct inode *inode, struct file *file)
2157 error = generic_file_open(inode, file);
2158 if (!error && (file->f_mode & FMODE_WRITE))
2159 error = dquot_initialize(inode);
2162 EXPORT_SYMBOL(dquot_file_open);
2165 * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
2167 int dquot_disable(struct super_block *sb, int type, unsigned int flags)
2170 struct quota_info *dqopt = sb_dqopt(sb);
2171 struct inode *toputinode[MAXQUOTAS];
2173 /* s_umount should be held in exclusive mode */
2174 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2175 up_read(&sb->s_umount);
2177 /* Cannot turn off usage accounting without turning off limits, or
2178 * suspend quotas and simultaneously turn quotas off. */
2179 if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
2180 || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
2181 DQUOT_USAGE_ENABLED)))
2185 * Skip everything if there's nothing to do. We have to do this because
2186 * sometimes we are called when fill_super() failed and calling
2187 * sync_fs() in such cases does no good.
2189 if (!sb_any_quota_loaded(sb))
2192 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2193 toputinode[cnt] = NULL;
2194 if (type != -1 && cnt != type)
2196 if (!sb_has_quota_loaded(sb, cnt))
2199 if (flags & DQUOT_SUSPENDED) {
2200 spin_lock(&dq_state_lock);
2202 dquot_state_flag(DQUOT_SUSPENDED, cnt);
2203 spin_unlock(&dq_state_lock);
2205 spin_lock(&dq_state_lock);
2206 dqopt->flags &= ~dquot_state_flag(flags, cnt);
2207 /* Turning off suspended quotas? */
2208 if (!sb_has_quota_loaded(sb, cnt) &&
2209 sb_has_quota_suspended(sb, cnt)) {
2210 dqopt->flags &= ~dquot_state_flag(
2211 DQUOT_SUSPENDED, cnt);
2212 spin_unlock(&dq_state_lock);
2213 iput(dqopt->files[cnt]);
2214 dqopt->files[cnt] = NULL;
2217 spin_unlock(&dq_state_lock);
2220 /* We still have to keep quota loaded? */
2221 if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
2224 /* Note: these are blocking operations */
2225 drop_dquot_ref(sb, cnt);
2226 invalidate_dquots(sb, cnt);
2228 * Now all dquots should be invalidated, all writes done so we
2229 * should be only users of the info. No locks needed.
2231 if (info_dirty(&dqopt->info[cnt]))
2232 sb->dq_op->write_info(sb, cnt);
2233 if (dqopt->ops[cnt]->free_file_info)
2234 dqopt->ops[cnt]->free_file_info(sb, cnt);
2235 put_quota_format(dqopt->info[cnt].dqi_format);
2237 toputinode[cnt] = dqopt->files[cnt];
2238 if (!sb_has_quota_loaded(sb, cnt))
2239 dqopt->files[cnt] = NULL;
2240 dqopt->info[cnt].dqi_flags = 0;
2241 dqopt->info[cnt].dqi_igrace = 0;
2242 dqopt->info[cnt].dqi_bgrace = 0;
2243 dqopt->ops[cnt] = NULL;
2246 /* Skip syncing and setting flags if quota files are hidden */
2247 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2250 /* Sync the superblock so that buffers with quota data are written to
2251 * disk (and so userspace sees correct data afterwards). */
2252 if (sb->s_op->sync_fs)
2253 sb->s_op->sync_fs(sb, 1);
2254 sync_blockdev(sb->s_bdev);
2255 /* Now the quota files are just ordinary files and we can set the
2256 * inode flags back. Moreover we discard the pagecache so that
2257 * userspace sees the writes we did bypassing the pagecache. We
2258 * must also discard the blockdev buffers so that we see the
2259 * changes done by userspace on the next quotaon() */
2260 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2261 /* This can happen when suspending quotas on remount-ro... */
2262 if (toputinode[cnt] && !sb_has_quota_loaded(sb, cnt)) {
2263 inode_lock(toputinode[cnt]);
2264 toputinode[cnt]->i_flags &= ~S_NOQUOTA;
2265 truncate_inode_pages(&toputinode[cnt]->i_data, 0);
2266 inode_unlock(toputinode[cnt]);
2267 mark_inode_dirty_sync(toputinode[cnt]);
2270 invalidate_bdev(sb->s_bdev);
2272 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2273 if (toputinode[cnt]) {
2274 /* On remount RO, we keep the inode pointer so that we
2275 * can reenable quota on the subsequent remount RW. We
2276 * have to check 'flags' variable and not use sb_has_
2277 * function because another quotaon / quotaoff could
2278 * change global state before we got here. We refuse
2279 * to suspend quotas when there is pending delete on
2280 * the quota file... */
2281 if (!(flags & DQUOT_SUSPENDED))
2282 iput(toputinode[cnt]);
2283 else if (!toputinode[cnt]->i_nlink)
2288 EXPORT_SYMBOL(dquot_disable);
2290 int dquot_quota_off(struct super_block *sb, int type)
2292 return dquot_disable(sb, type,
2293 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2295 EXPORT_SYMBOL(dquot_quota_off);
2298 * Turn quotas on on a device
2302 * Helper function to turn quotas on when we already have the inode of
2303 * quota file and no quota information is loaded.
2305 static int vfs_load_quota_inode(struct inode *inode, int type, int format_id,
2308 struct quota_format_type *fmt = find_quota_format(format_id);
2309 struct super_block *sb = inode->i_sb;
2310 struct quota_info *dqopt = sb_dqopt(sb);
2315 if (!S_ISREG(inode->i_mode)) {
2319 if (IS_RDONLY(inode)) {
2323 if (!sb->s_op->quota_write || !sb->s_op->quota_read ||
2324 (type == PRJQUOTA && sb->dq_op->get_projid == NULL)) {
2328 /* Filesystems outside of init_user_ns not yet supported */
2329 if (sb->s_user_ns != &init_user_ns) {
2333 /* Usage always has to be set... */
2334 if (!(flags & DQUOT_USAGE_ENABLED)) {
2338 if (sb_has_quota_loaded(sb, type)) {
2343 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2344 /* As we bypass the pagecache we must now flush all the
2345 * dirty data and invalidate caches so that kernel sees
2346 * changes from userspace. It is not enough to just flush
2347 * the quota file since if blocksize < pagesize, invalidation
2348 * of the cache could fail because of other unrelated dirty
2350 sync_filesystem(sb);
2351 invalidate_bdev(sb->s_bdev);
2354 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2355 /* We don't want quota and atime on quota files (deadlocks
2356 * possible) Also nobody should write to the file - we use
2357 * special IO operations which ignore the immutable bit. */
2359 inode->i_flags |= S_NOQUOTA;
2360 inode_unlock(inode);
2362 * When S_NOQUOTA is set, remove dquot references as no more
2363 * references can be added
2365 __dquot_drop(inode);
2369 dqopt->files[type] = igrab(inode);
2370 if (!dqopt->files[type])
2371 goto out_file_flags;
2373 if (!fmt->qf_ops->check_quota_file(sb, type))
2376 dqopt->ops[type] = fmt->qf_ops;
2377 dqopt->info[type].dqi_format = fmt;
2378 dqopt->info[type].dqi_fmt_id = format_id;
2379 INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2380 error = dqopt->ops[type]->read_file_info(sb, type);
2383 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE) {
2384 spin_lock(&dq_data_lock);
2385 dqopt->info[type].dqi_flags |= DQF_SYS_FILE;
2386 spin_unlock(&dq_data_lock);
2388 spin_lock(&dq_state_lock);
2389 dqopt->flags |= dquot_state_flag(flags, type);
2390 spin_unlock(&dq_state_lock);
2392 error = add_dquot_ref(sb, type);
2394 dquot_disable(sb, type, flags);
2398 dqopt->files[type] = NULL;
2402 inode->i_flags &= ~S_NOQUOTA;
2403 inode_unlock(inode);
2405 put_quota_format(fmt);
2410 /* Reenable quotas on remount RW */
2411 int dquot_resume(struct super_block *sb, int type)
2413 struct quota_info *dqopt = sb_dqopt(sb);
2414 struct inode *inode;
2418 /* s_umount should be held in exclusive mode */
2419 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2420 up_read(&sb->s_umount);
2422 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2423 if (type != -1 && cnt != type)
2425 if (!sb_has_quota_suspended(sb, cnt))
2428 inode = dqopt->files[cnt];
2429 dqopt->files[cnt] = NULL;
2430 spin_lock(&dq_state_lock);
2431 flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2432 DQUOT_LIMITS_ENABLED,
2434 dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2435 spin_unlock(&dq_state_lock);
2437 flags = dquot_generic_flag(flags, cnt);
2438 ret = vfs_load_quota_inode(inode, cnt,
2439 dqopt->info[cnt].dqi_fmt_id, flags);
2445 EXPORT_SYMBOL(dquot_resume);
2447 int dquot_quota_on(struct super_block *sb, int type, int format_id,
2448 const struct path *path)
2450 int error = security_quota_on(path->dentry);
2453 /* Quota file not on the same filesystem? */
2454 if (path->dentry->d_sb != sb)
2457 error = vfs_load_quota_inode(d_inode(path->dentry), type,
2458 format_id, DQUOT_USAGE_ENABLED |
2459 DQUOT_LIMITS_ENABLED);
2462 EXPORT_SYMBOL(dquot_quota_on);
2465 * More powerful function for turning on quotas allowing setting
2466 * of individual quota flags
2468 int dquot_enable(struct inode *inode, int type, int format_id,
2471 struct super_block *sb = inode->i_sb;
2473 /* Just unsuspend quotas? */
2474 BUG_ON(flags & DQUOT_SUSPENDED);
2475 /* s_umount should be held in exclusive mode */
2476 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2477 up_read(&sb->s_umount);
2481 /* Just updating flags needed? */
2482 if (sb_has_quota_loaded(sb, type)) {
2483 if (flags & DQUOT_USAGE_ENABLED &&
2484 sb_has_quota_usage_enabled(sb, type))
2486 if (flags & DQUOT_LIMITS_ENABLED &&
2487 sb_has_quota_limits_enabled(sb, type))
2489 spin_lock(&dq_state_lock);
2490 sb_dqopt(sb)->flags |= dquot_state_flag(flags, type);
2491 spin_unlock(&dq_state_lock);
2495 return vfs_load_quota_inode(inode, type, format_id, flags);
2497 EXPORT_SYMBOL(dquot_enable);
2500 * This function is used when filesystem needs to initialize quotas
2501 * during mount time.
2503 int dquot_quota_on_mount(struct super_block *sb, char *qf_name,
2504 int format_id, int type)
2506 struct dentry *dentry;
2509 dentry = lookup_one_len_unlocked(qf_name, sb->s_root, strlen(qf_name));
2511 return PTR_ERR(dentry);
2513 if (d_really_is_negative(dentry)) {
2518 error = security_quota_on(dentry);
2520 error = vfs_load_quota_inode(d_inode(dentry), type, format_id,
2521 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2527 EXPORT_SYMBOL(dquot_quota_on_mount);
2529 static int dquot_quota_enable(struct super_block *sb, unsigned int flags)
2533 struct quota_info *dqopt = sb_dqopt(sb);
2535 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2537 /* Accounting cannot be turned on while fs is mounted */
2538 flags &= ~(FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT);
2541 for (type = 0; type < MAXQUOTAS; type++) {
2542 if (!(flags & qtype_enforce_flag(type)))
2544 /* Can't enforce without accounting */
2545 if (!sb_has_quota_usage_enabled(sb, type))
2547 ret = dquot_enable(dqopt->files[type], type,
2548 dqopt->info[type].dqi_fmt_id,
2549 DQUOT_LIMITS_ENABLED);
2555 /* Backout enforcement enablement we already did */
2556 for (type--; type >= 0; type--) {
2557 if (flags & qtype_enforce_flag(type))
2558 dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2560 /* Error code translation for better compatibility with XFS */
2566 static int dquot_quota_disable(struct super_block *sb, unsigned int flags)
2570 struct quota_info *dqopt = sb_dqopt(sb);
2572 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2575 * We don't support turning off accounting via quotactl. In principle
2576 * quota infrastructure can do this but filesystems don't expect
2577 * userspace to be able to do it.
2580 (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT))
2583 /* Filter out limits not enabled */
2584 for (type = 0; type < MAXQUOTAS; type++)
2585 if (!sb_has_quota_limits_enabled(sb, type))
2586 flags &= ~qtype_enforce_flag(type);
2590 for (type = 0; type < MAXQUOTAS; type++) {
2591 if (flags & qtype_enforce_flag(type)) {
2592 ret = dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2599 /* Backout enforcement disabling we already did */
2600 for (type--; type >= 0; type--) {
2601 if (flags & qtype_enforce_flag(type))
2602 dquot_enable(dqopt->files[type], type,
2603 dqopt->info[type].dqi_fmt_id,
2604 DQUOT_LIMITS_ENABLED);
2609 /* Generic routine for getting common part of quota structure */
2610 static void do_get_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2612 struct mem_dqblk *dm = &dquot->dq_dqb;
2614 memset(di, 0, sizeof(*di));
2615 spin_lock(&dquot->dq_dqb_lock);
2616 di->d_spc_hardlimit = dm->dqb_bhardlimit;
2617 di->d_spc_softlimit = dm->dqb_bsoftlimit;
2618 di->d_ino_hardlimit = dm->dqb_ihardlimit;
2619 di->d_ino_softlimit = dm->dqb_isoftlimit;
2620 di->d_space = dm->dqb_curspace + dm->dqb_rsvspace;
2621 di->d_ino_count = dm->dqb_curinodes;
2622 di->d_spc_timer = dm->dqb_btime;
2623 di->d_ino_timer = dm->dqb_itime;
2624 spin_unlock(&dquot->dq_dqb_lock);
2627 int dquot_get_dqblk(struct super_block *sb, struct kqid qid,
2628 struct qc_dqblk *di)
2630 struct dquot *dquot;
2632 dquot = dqget(sb, qid);
2634 return PTR_ERR(dquot);
2635 do_get_dqblk(dquot, di);
2640 EXPORT_SYMBOL(dquot_get_dqblk);
2642 int dquot_get_next_dqblk(struct super_block *sb, struct kqid *qid,
2643 struct qc_dqblk *di)
2645 struct dquot *dquot;
2648 if (!sb->dq_op->get_next_id)
2650 err = sb->dq_op->get_next_id(sb, qid);
2653 dquot = dqget(sb, *qid);
2655 return PTR_ERR(dquot);
2656 do_get_dqblk(dquot, di);
2661 EXPORT_SYMBOL(dquot_get_next_dqblk);
2663 #define VFS_QC_MASK \
2664 (QC_SPACE | QC_SPC_SOFT | QC_SPC_HARD | \
2665 QC_INO_COUNT | QC_INO_SOFT | QC_INO_HARD | \
2666 QC_SPC_TIMER | QC_INO_TIMER)
2668 /* Generic routine for setting common part of quota structure */
2669 static int do_set_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2671 struct mem_dqblk *dm = &dquot->dq_dqb;
2672 int check_blim = 0, check_ilim = 0;
2673 struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
2675 if (di->d_fieldmask & ~VFS_QC_MASK)
2678 if (((di->d_fieldmask & QC_SPC_SOFT) &&
2679 di->d_spc_softlimit > dqi->dqi_max_spc_limit) ||
2680 ((di->d_fieldmask & QC_SPC_HARD) &&
2681 di->d_spc_hardlimit > dqi->dqi_max_spc_limit) ||
2682 ((di->d_fieldmask & QC_INO_SOFT) &&
2683 (di->d_ino_softlimit > dqi->dqi_max_ino_limit)) ||
2684 ((di->d_fieldmask & QC_INO_HARD) &&
2685 (di->d_ino_hardlimit > dqi->dqi_max_ino_limit)))
2688 spin_lock(&dquot->dq_dqb_lock);
2689 if (di->d_fieldmask & QC_SPACE) {
2690 dm->dqb_curspace = di->d_space - dm->dqb_rsvspace;
2692 set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2695 if (di->d_fieldmask & QC_SPC_SOFT)
2696 dm->dqb_bsoftlimit = di->d_spc_softlimit;
2697 if (di->d_fieldmask & QC_SPC_HARD)
2698 dm->dqb_bhardlimit = di->d_spc_hardlimit;
2699 if (di->d_fieldmask & (QC_SPC_SOFT | QC_SPC_HARD)) {
2701 set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2704 if (di->d_fieldmask & QC_INO_COUNT) {
2705 dm->dqb_curinodes = di->d_ino_count;
2707 set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2710 if (di->d_fieldmask & QC_INO_SOFT)
2711 dm->dqb_isoftlimit = di->d_ino_softlimit;
2712 if (di->d_fieldmask & QC_INO_HARD)
2713 dm->dqb_ihardlimit = di->d_ino_hardlimit;
2714 if (di->d_fieldmask & (QC_INO_SOFT | QC_INO_HARD)) {
2716 set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2719 if (di->d_fieldmask & QC_SPC_TIMER) {
2720 dm->dqb_btime = di->d_spc_timer;
2722 set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2725 if (di->d_fieldmask & QC_INO_TIMER) {
2726 dm->dqb_itime = di->d_ino_timer;
2728 set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2732 if (!dm->dqb_bsoftlimit ||
2733 dm->dqb_curspace + dm->dqb_rsvspace < dm->dqb_bsoftlimit) {
2735 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2736 } else if (!(di->d_fieldmask & QC_SPC_TIMER))
2737 /* Set grace only if user hasn't provided his own... */
2738 dm->dqb_btime = ktime_get_real_seconds() + dqi->dqi_bgrace;
2741 if (!dm->dqb_isoftlimit ||
2742 dm->dqb_curinodes < dm->dqb_isoftlimit) {
2744 clear_bit(DQ_INODES_B, &dquot->dq_flags);
2745 } else if (!(di->d_fieldmask & QC_INO_TIMER))
2746 /* Set grace only if user hasn't provided his own... */
2747 dm->dqb_itime = ktime_get_real_seconds() + dqi->dqi_igrace;
2749 if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2751 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2753 set_bit(DQ_FAKE_B, &dquot->dq_flags);
2754 spin_unlock(&dquot->dq_dqb_lock);
2755 mark_dquot_dirty(dquot);
2760 int dquot_set_dqblk(struct super_block *sb, struct kqid qid,
2761 struct qc_dqblk *di)
2763 struct dquot *dquot;
2766 dquot = dqget(sb, qid);
2767 if (IS_ERR(dquot)) {
2768 rc = PTR_ERR(dquot);
2771 rc = do_set_dqblk(dquot, di);
2776 EXPORT_SYMBOL(dquot_set_dqblk);
2778 /* Generic routine for getting common part of quota file information */
2779 int dquot_get_state(struct super_block *sb, struct qc_state *state)
2781 struct mem_dqinfo *mi;
2782 struct qc_type_state *tstate;
2783 struct quota_info *dqopt = sb_dqopt(sb);
2786 memset(state, 0, sizeof(*state));
2787 for (type = 0; type < MAXQUOTAS; type++) {
2788 if (!sb_has_quota_active(sb, type))
2790 tstate = state->s_state + type;
2791 mi = sb_dqopt(sb)->info + type;
2792 tstate->flags = QCI_ACCT_ENABLED;
2793 spin_lock(&dq_data_lock);
2794 if (mi->dqi_flags & DQF_SYS_FILE)
2795 tstate->flags |= QCI_SYSFILE;
2796 if (mi->dqi_flags & DQF_ROOT_SQUASH)
2797 tstate->flags |= QCI_ROOT_SQUASH;
2798 if (sb_has_quota_limits_enabled(sb, type))
2799 tstate->flags |= QCI_LIMITS_ENFORCED;
2800 tstate->spc_timelimit = mi->dqi_bgrace;
2801 tstate->ino_timelimit = mi->dqi_igrace;
2802 tstate->ino = dqopt->files[type]->i_ino;
2803 tstate->blocks = dqopt->files[type]->i_blocks;
2804 tstate->nextents = 1; /* We don't know... */
2805 spin_unlock(&dq_data_lock);
2809 EXPORT_SYMBOL(dquot_get_state);
2811 /* Generic routine for setting common part of quota file information */
2812 int dquot_set_dqinfo(struct super_block *sb, int type, struct qc_info *ii)
2814 struct mem_dqinfo *mi;
2817 if ((ii->i_fieldmask & QC_WARNS_MASK) ||
2818 (ii->i_fieldmask & QC_RT_SPC_TIMER))
2820 if (!sb_has_quota_active(sb, type))
2822 mi = sb_dqopt(sb)->info + type;
2823 if (ii->i_fieldmask & QC_FLAGS) {
2824 if ((ii->i_flags & QCI_ROOT_SQUASH &&
2825 mi->dqi_format->qf_fmt_id != QFMT_VFS_OLD))
2828 spin_lock(&dq_data_lock);
2829 if (ii->i_fieldmask & QC_SPC_TIMER)
2830 mi->dqi_bgrace = ii->i_spc_timelimit;
2831 if (ii->i_fieldmask & QC_INO_TIMER)
2832 mi->dqi_igrace = ii->i_ino_timelimit;
2833 if (ii->i_fieldmask & QC_FLAGS) {
2834 if (ii->i_flags & QCI_ROOT_SQUASH)
2835 mi->dqi_flags |= DQF_ROOT_SQUASH;
2837 mi->dqi_flags &= ~DQF_ROOT_SQUASH;
2839 spin_unlock(&dq_data_lock);
2840 mark_info_dirty(sb, type);
2841 /* Force write to disk */
2842 sb->dq_op->write_info(sb, type);
2845 EXPORT_SYMBOL(dquot_set_dqinfo);
2847 const struct quotactl_ops dquot_quotactl_sysfile_ops = {
2848 .quota_enable = dquot_quota_enable,
2849 .quota_disable = dquot_quota_disable,
2850 .quota_sync = dquot_quota_sync,
2851 .get_state = dquot_get_state,
2852 .set_info = dquot_set_dqinfo,
2853 .get_dqblk = dquot_get_dqblk,
2854 .get_nextdqblk = dquot_get_next_dqblk,
2855 .set_dqblk = dquot_set_dqblk
2857 EXPORT_SYMBOL(dquot_quotactl_sysfile_ops);
2859 static int do_proc_dqstats(struct ctl_table *table, int write,
2860 void __user *buffer, size_t *lenp, loff_t *ppos)
2862 unsigned int type = (unsigned long *)table->data - dqstats.stat;
2863 s64 value = percpu_counter_sum(&dqstats.counter[type]);
2865 /* Filter negative values for non-monotonic counters */
2866 if (value < 0 && (type == DQST_ALLOC_DQUOTS ||
2867 type == DQST_FREE_DQUOTS))
2870 /* Update global table */
2871 dqstats.stat[type] = value;
2872 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
2875 static struct ctl_table fs_dqstats_table[] = {
2877 .procname = "lookups",
2878 .data = &dqstats.stat[DQST_LOOKUPS],
2879 .maxlen = sizeof(unsigned long),
2881 .proc_handler = do_proc_dqstats,
2884 .procname = "drops",
2885 .data = &dqstats.stat[DQST_DROPS],
2886 .maxlen = sizeof(unsigned long),
2888 .proc_handler = do_proc_dqstats,
2891 .procname = "reads",
2892 .data = &dqstats.stat[DQST_READS],
2893 .maxlen = sizeof(unsigned long),
2895 .proc_handler = do_proc_dqstats,
2898 .procname = "writes",
2899 .data = &dqstats.stat[DQST_WRITES],
2900 .maxlen = sizeof(unsigned long),
2902 .proc_handler = do_proc_dqstats,
2905 .procname = "cache_hits",
2906 .data = &dqstats.stat[DQST_CACHE_HITS],
2907 .maxlen = sizeof(unsigned long),
2909 .proc_handler = do_proc_dqstats,
2912 .procname = "allocated_dquots",
2913 .data = &dqstats.stat[DQST_ALLOC_DQUOTS],
2914 .maxlen = sizeof(unsigned long),
2916 .proc_handler = do_proc_dqstats,
2919 .procname = "free_dquots",
2920 .data = &dqstats.stat[DQST_FREE_DQUOTS],
2921 .maxlen = sizeof(unsigned long),
2923 .proc_handler = do_proc_dqstats,
2926 .procname = "syncs",
2927 .data = &dqstats.stat[DQST_SYNCS],
2928 .maxlen = sizeof(unsigned long),
2930 .proc_handler = do_proc_dqstats,
2932 #ifdef CONFIG_PRINT_QUOTA_WARNING
2934 .procname = "warnings",
2935 .data = &flag_print_warnings,
2936 .maxlen = sizeof(int),
2938 .proc_handler = proc_dointvec,
2944 static struct ctl_table fs_table[] = {
2946 .procname = "quota",
2948 .child = fs_dqstats_table,
2953 static struct ctl_table sys_table[] = {
2962 static int __init dquot_init(void)
2965 unsigned long nr_hash, order;
2967 printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2969 register_sysctl_table(sys_table);
2971 dquot_cachep = kmem_cache_create("dquot",
2972 sizeof(struct dquot), sizeof(unsigned long) * 4,
2973 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2974 SLAB_MEM_SPREAD|SLAB_PANIC),
2978 dquot_hash = (struct hlist_head *)__get_free_pages(GFP_KERNEL, order);
2980 panic("Cannot create dquot hash table");
2982 for (i = 0; i < _DQST_DQSTAT_LAST; i++) {
2983 ret = percpu_counter_init(&dqstats.counter[i], 0, GFP_KERNEL);
2985 panic("Cannot create dquot stat counters");
2988 /* Find power-of-two hlist_heads which can fit into allocation */
2989 nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2993 } while (nr_hash >> dq_hash_bits);
2996 nr_hash = 1UL << dq_hash_bits;
2997 dq_hash_mask = nr_hash - 1;
2998 for (i = 0; i < nr_hash; i++)
2999 INIT_HLIST_HEAD(dquot_hash + i);
3001 pr_info("VFS: Dquot-cache hash table entries: %ld (order %ld,"
3002 " %ld bytes)\n", nr_hash, order, (PAGE_SIZE << order));
3004 if (register_shrinker(&dqcache_shrinker))
3005 panic("Cannot register dquot shrinker");
3009 fs_initcall(dquot_init);