2 * Implementation of the diskquota system for the LINUX operating system. QUOTA
3 * is implemented using the BSD system call interface as the means of
4 * communication with the user level. This file contains the generic routines
5 * called by the different filesystems on allocation of an inode or block.
6 * These routines take care of the administration needed to have a consistent
7 * diskquota tracking system. The ideas of both user and group quotas are based
8 * on the Melbourne quota system as used on BSD derived systems. The internal
9 * implementation is based on one of the several variants of the LINUX
10 * inode-subsystem with added complexity of the diskquota system.
16 * Revised list management to avoid races
19 * Fixed races in dquot_transfer(), dqget() and dquot_alloc_...().
20 * As the consequence the locking was moved from dquot_decr_...(),
21 * dquot_incr_...() to calling functions.
22 * invalidate_dquots() now writes modified dquots.
23 * Serialized quota_off() and quota_on() for mount point.
24 * Fixed a few bugs in grow_dquots().
25 * Fixed deadlock in write_dquot() - we no longer account quotas on
27 * remove_dquot_ref() moved to inode.c - it now traverses through inodes
28 * add_dquot_ref() restarts after blocking
29 * Added check for bogus uid and fixed check for group in quotactl.
32 * Used struct list_head instead of own list struct
33 * Invalidation of referenced dquots is no longer possible
34 * Improved free_dquots list management
35 * Quota and i_blocks are now updated in one place to avoid races
36 * Warnings are now delayed so we won't block in critical section
37 * Write updated not to require dquot lock
40 * Added dynamic quota structure allocation
43 * Rewritten quota interface. Implemented new quota format and
44 * formats registering.
50 * Added journalled quota support, fix lock inversion problems
53 * (C) Copyright 1994 - 1997 Marco van Wieringen
56 #include <linux/errno.h>
57 #include <linux/kernel.h>
59 #include <linux/mount.h>
61 #include <linux/time.h>
62 #include <linux/types.h>
63 #include <linux/string.h>
64 #include <linux/fcntl.h>
65 #include <linux/stat.h>
66 #include <linux/tty.h>
67 #include <linux/file.h>
68 #include <linux/slab.h>
69 #include <linux/sysctl.h>
70 #include <linux/init.h>
71 #include <linux/module.h>
72 #include <linux/proc_fs.h>
73 #include <linux/security.h>
74 #include <linux/sched.h>
75 #include <linux/cred.h>
76 #include <linux/kmod.h>
77 #include <linux/namei.h>
78 #include <linux/capability.h>
79 #include <linux/quotaops.h>
80 #include "../internal.h" /* ugh */
82 #include <linux/uaccess.h>
85 * There are five quota SMP locks:
86 * * dq_list_lock protects all lists with quotas and quota formats.
87 * * dquot->dq_dqb_lock protects data from dq_dqb
88 * * inode->i_lock protects inode->i_blocks, i_bytes and also guards
89 * consistency of dquot->dq_dqb with inode->i_blocks, i_bytes so that
90 * dquot_transfer() can stabilize amount it transfers
91 * * dq_data_lock protects mem_dqinfo structures and modifications of dquot
92 * pointers in the inode
93 * * dq_state_lock protects modifications of quota state (on quotaon and
94 * quotaoff) and readers who care about latest values take it as well.
96 * The spinlock ordering is hence:
97 * dq_data_lock > dq_list_lock > i_lock > dquot->dq_dqb_lock,
98 * dq_list_lock > dq_state_lock
100 * Note that some things (eg. sb pointer, type, id) doesn't change during
101 * the life of the dquot structure and so needn't to be protected by a lock
103 * Operation accessing dquots via inode pointers are protected by dquot_srcu.
104 * Operation of reading pointer needs srcu_read_lock(&dquot_srcu), and
105 * synchronize_srcu(&dquot_srcu) is called after clearing pointers from
106 * inode and before dropping dquot references to avoid use of dquots after
107 * they are freed. dq_data_lock is used to serialize the pointer setting and
108 * clearing operations.
109 * Special care needs to be taken about S_NOQUOTA inode flag (marking that
110 * inode is a quota file). Functions adding pointers from inode to dquots have
111 * to check this flag under dq_data_lock and then (if S_NOQUOTA is not set) they
112 * have to do all pointer modifications before dropping dq_data_lock. This makes
113 * sure they cannot race with quotaon which first sets S_NOQUOTA flag and
114 * then drops all pointers to dquots from an inode.
116 * Each dquot has its dq_lock mutex. Dquot is locked when it is being read to
117 * memory (or space for it is being allocated) on the first dqget(), when it is
118 * being written out, and when it is being released on the last dqput(). The
119 * allocation and release operations are serialized by the dq_lock and by
120 * checking the use count in dquot_release().
122 * Lock ordering (including related VFS locks) is the following:
123 * s_umount > i_mutex > journal_lock > dquot->dq_lock > dqio_sem
126 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_list_lock);
127 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_state_lock);
128 __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_data_lock);
129 EXPORT_SYMBOL(dq_data_lock);
130 DEFINE_STATIC_SRCU(dquot_srcu);
132 static DECLARE_WAIT_QUEUE_HEAD(dquot_ref_wq);
134 void __quota_error(struct super_block *sb, const char *func,
135 const char *fmt, ...)
137 if (printk_ratelimit()) {
139 struct va_format vaf;
146 printk(KERN_ERR "Quota error (device %s): %s: %pV\n",
147 sb->s_id, func, &vaf);
152 EXPORT_SYMBOL(__quota_error);
154 #if defined(CONFIG_QUOTA_DEBUG) || defined(CONFIG_PRINT_QUOTA_WARNING)
155 static char *quotatypes[] = INITQFNAMES;
157 static struct quota_format_type *quota_formats; /* List of registered formats */
158 static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
160 /* SLAB cache for dquot structures */
161 static struct kmem_cache *dquot_cachep;
163 int register_quota_format(struct quota_format_type *fmt)
165 spin_lock(&dq_list_lock);
166 fmt->qf_next = quota_formats;
168 spin_unlock(&dq_list_lock);
171 EXPORT_SYMBOL(register_quota_format);
173 void unregister_quota_format(struct quota_format_type *fmt)
175 struct quota_format_type **actqf;
177 spin_lock(&dq_list_lock);
178 for (actqf = "a_formats; *actqf && *actqf != fmt;
179 actqf = &(*actqf)->qf_next)
182 *actqf = (*actqf)->qf_next;
183 spin_unlock(&dq_list_lock);
185 EXPORT_SYMBOL(unregister_quota_format);
187 static struct quota_format_type *find_quota_format(int id)
189 struct quota_format_type *actqf;
191 spin_lock(&dq_list_lock);
192 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
193 actqf = actqf->qf_next)
195 if (!actqf || !try_module_get(actqf->qf_owner)) {
198 spin_unlock(&dq_list_lock);
200 for (qm = 0; module_names[qm].qm_fmt_id &&
201 module_names[qm].qm_fmt_id != id; qm++)
203 if (!module_names[qm].qm_fmt_id ||
204 request_module(module_names[qm].qm_mod_name))
207 spin_lock(&dq_list_lock);
208 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
209 actqf = actqf->qf_next)
211 if (actqf && !try_module_get(actqf->qf_owner))
214 spin_unlock(&dq_list_lock);
218 static void put_quota_format(struct quota_format_type *fmt)
220 module_put(fmt->qf_owner);
224 * Dquot List Management:
225 * The quota code uses three lists for dquot management: the inuse_list,
226 * free_dquots, and dquot_hash[] array. A single dquot structure may be
227 * on all three lists, depending on its current state.
229 * All dquots are placed to the end of inuse_list when first created, and this
230 * list is used for invalidate operation, which must look at every dquot.
232 * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
233 * and this list is searched whenever we need an available dquot. Dquots are
234 * removed from the list as soon as they are used again, and
235 * dqstats.free_dquots gives the number of dquots on the list. When
236 * dquot is invalidated it's completely released from memory.
238 * Dquots with a specific identity (device, type and id) are placed on
239 * one of the dquot_hash[] hash chains. The provides an efficient search
240 * mechanism to locate a specific dquot.
243 static LIST_HEAD(inuse_list);
244 static LIST_HEAD(free_dquots);
245 static unsigned int dq_hash_bits, dq_hash_mask;
246 static struct hlist_head *dquot_hash;
248 struct dqstats dqstats;
249 EXPORT_SYMBOL(dqstats);
251 static qsize_t inode_get_rsv_space(struct inode *inode);
252 static qsize_t __inode_get_rsv_space(struct inode *inode);
253 static int __dquot_initialize(struct inode *inode, int type);
255 static inline unsigned int
256 hashfn(const struct super_block *sb, struct kqid qid)
258 unsigned int id = from_kqid(&init_user_ns, qid);
262 tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
263 return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
267 * Following list functions expect dq_list_lock to be held
269 static inline void insert_dquot_hash(struct dquot *dquot)
271 struct hlist_head *head;
272 head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id);
273 hlist_add_head(&dquot->dq_hash, head);
276 static inline void remove_dquot_hash(struct dquot *dquot)
278 hlist_del_init(&dquot->dq_hash);
281 static struct dquot *find_dquot(unsigned int hashent, struct super_block *sb,
284 struct hlist_node *node;
287 hlist_for_each (node, dquot_hash+hashent) {
288 dquot = hlist_entry(node, struct dquot, dq_hash);
289 if (dquot->dq_sb == sb && qid_eq(dquot->dq_id, qid))
295 /* Add a dquot to the tail of the free list */
296 static inline void put_dquot_last(struct dquot *dquot)
298 list_add_tail(&dquot->dq_free, &free_dquots);
299 dqstats_inc(DQST_FREE_DQUOTS);
302 static inline void remove_free_dquot(struct dquot *dquot)
304 if (list_empty(&dquot->dq_free))
306 list_del_init(&dquot->dq_free);
307 dqstats_dec(DQST_FREE_DQUOTS);
310 static inline void put_inuse(struct dquot *dquot)
312 /* We add to the back of inuse list so we don't have to restart
313 * when traversing this list and we block */
314 list_add_tail(&dquot->dq_inuse, &inuse_list);
315 dqstats_inc(DQST_ALLOC_DQUOTS);
318 static inline void remove_inuse(struct dquot *dquot)
320 dqstats_dec(DQST_ALLOC_DQUOTS);
321 list_del(&dquot->dq_inuse);
324 * End of list functions needing dq_list_lock
327 static void wait_on_dquot(struct dquot *dquot)
329 mutex_lock(&dquot->dq_lock);
330 mutex_unlock(&dquot->dq_lock);
333 static inline int dquot_dirty(struct dquot *dquot)
335 return test_bit(DQ_MOD_B, &dquot->dq_flags);
338 static inline int mark_dquot_dirty(struct dquot *dquot)
340 return dquot->dq_sb->dq_op->mark_dirty(dquot);
343 /* Mark dquot dirty in atomic manner, and return it's old dirty flag state */
344 int dquot_mark_dquot_dirty(struct dquot *dquot)
348 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
351 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NOLIST_DIRTY)
352 return test_and_set_bit(DQ_MOD_B, &dquot->dq_flags);
354 /* If quota is dirty already, we don't have to acquire dq_list_lock */
355 if (test_bit(DQ_MOD_B, &dquot->dq_flags))
358 spin_lock(&dq_list_lock);
359 if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags)) {
360 list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
361 info[dquot->dq_id.type].dqi_dirty_list);
364 spin_unlock(&dq_list_lock);
367 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
369 /* Dirtify all the dquots - this can block when journalling */
370 static inline int mark_all_dquot_dirty(struct dquot * const *dquot)
375 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
377 /* Even in case of error we have to continue */
378 ret = mark_dquot_dirty(dquot[cnt]);
385 static inline void dqput_all(struct dquot **dquot)
389 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
393 static inline int clear_dquot_dirty(struct dquot *dquot)
395 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NOLIST_DIRTY)
396 return test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags);
398 spin_lock(&dq_list_lock);
399 if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags)) {
400 spin_unlock(&dq_list_lock);
403 list_del_init(&dquot->dq_dirty);
404 spin_unlock(&dq_list_lock);
408 void mark_info_dirty(struct super_block *sb, int type)
410 spin_lock(&dq_data_lock);
411 sb_dqopt(sb)->info[type].dqi_flags |= DQF_INFO_DIRTY;
412 spin_unlock(&dq_data_lock);
414 EXPORT_SYMBOL(mark_info_dirty);
417 * Read dquot from disk and alloc space for it
420 int dquot_acquire(struct dquot *dquot)
422 int ret = 0, ret2 = 0;
423 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
425 mutex_lock(&dquot->dq_lock);
426 if (!test_bit(DQ_READ_B, &dquot->dq_flags))
427 ret = dqopt->ops[dquot->dq_id.type]->read_dqblk(dquot);
430 /* Make sure flags update is visible after dquot has been filled */
431 smp_mb__before_atomic();
432 set_bit(DQ_READ_B, &dquot->dq_flags);
433 /* Instantiate dquot if needed */
434 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
435 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
436 /* Write the info if needed */
437 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
438 ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
439 dquot->dq_sb, dquot->dq_id.type);
449 * Make sure flags update is visible after on-disk struct has been
450 * allocated. Paired with smp_rmb() in dqget().
452 smp_mb__before_atomic();
453 set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
455 mutex_unlock(&dquot->dq_lock);
458 EXPORT_SYMBOL(dquot_acquire);
461 * Write dquot to disk
463 int dquot_commit(struct dquot *dquot)
466 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
468 mutex_lock(&dquot->dq_lock);
469 if (!clear_dquot_dirty(dquot))
471 /* Inactive dquot can be only if there was error during read/init
472 * => we have better not writing it */
473 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
474 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
478 mutex_unlock(&dquot->dq_lock);
481 EXPORT_SYMBOL(dquot_commit);
486 int dquot_release(struct dquot *dquot)
488 int ret = 0, ret2 = 0;
489 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
491 mutex_lock(&dquot->dq_lock);
492 /* Check whether we are not racing with some other dqget() */
493 if (atomic_read(&dquot->dq_count) > 1)
495 if (dqopt->ops[dquot->dq_id.type]->release_dqblk) {
496 ret = dqopt->ops[dquot->dq_id.type]->release_dqblk(dquot);
498 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
499 ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
500 dquot->dq_sb, dquot->dq_id.type);
505 clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
507 mutex_unlock(&dquot->dq_lock);
510 EXPORT_SYMBOL(dquot_release);
512 void dquot_destroy(struct dquot *dquot)
514 kmem_cache_free(dquot_cachep, dquot);
516 EXPORT_SYMBOL(dquot_destroy);
518 static inline void do_destroy_dquot(struct dquot *dquot)
520 dquot->dq_sb->dq_op->destroy_dquot(dquot);
523 /* Invalidate all dquots on the list. Note that this function is called after
524 * quota is disabled and pointers from inodes removed so there cannot be new
525 * quota users. There can still be some users of quotas due to inodes being
526 * just deleted or pruned by prune_icache() (those are not attached to any
527 * list) or parallel quotactl call. We have to wait for such users.
529 static void invalidate_dquots(struct super_block *sb, int type)
531 struct dquot *dquot, *tmp;
534 spin_lock(&dq_list_lock);
535 list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
536 if (dquot->dq_sb != sb)
538 if (dquot->dq_id.type != type)
540 /* Wait for dquot users */
541 if (atomic_read(&dquot->dq_count)) {
543 spin_unlock(&dq_list_lock);
545 * Once dqput() wakes us up, we know it's time to free
547 * IMPORTANT: we rely on the fact that there is always
548 * at most one process waiting for dquot to free.
549 * Otherwise dq_count would be > 1 and we would never
552 wait_event(dquot_ref_wq,
553 atomic_read(&dquot->dq_count) == 1);
555 /* At this moment dquot() need not exist (it could be
556 * reclaimed by prune_dqcache(). Hence we must
561 * Quota now has no users and it has been written on last
564 remove_dquot_hash(dquot);
565 remove_free_dquot(dquot);
567 do_destroy_dquot(dquot);
569 spin_unlock(&dq_list_lock);
572 /* Call callback for every active dquot on given filesystem */
573 int dquot_scan_active(struct super_block *sb,
574 int (*fn)(struct dquot *dquot, unsigned long priv),
577 struct dquot *dquot, *old_dquot = NULL;
580 WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
582 spin_lock(&dq_list_lock);
583 list_for_each_entry(dquot, &inuse_list, dq_inuse) {
584 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
586 if (dquot->dq_sb != sb)
588 /* Now we have active dquot so we can just increase use count */
589 atomic_inc(&dquot->dq_count);
590 spin_unlock(&dq_list_lock);
591 dqstats_inc(DQST_LOOKUPS);
595 * ->release_dquot() can be racing with us. Our reference
596 * protects us from new calls to it so just wait for any
597 * outstanding call and recheck the DQ_ACTIVE_B after that.
599 wait_on_dquot(dquot);
600 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
601 ret = fn(dquot, priv);
605 spin_lock(&dq_list_lock);
606 /* We are safe to continue now because our dquot could not
607 * be moved out of the inuse list while we hold the reference */
609 spin_unlock(&dq_list_lock);
614 EXPORT_SYMBOL(dquot_scan_active);
616 /* Write all dquot structures to quota files */
617 int dquot_writeback_dquots(struct super_block *sb, int type)
619 struct list_head *dirty;
621 struct quota_info *dqopt = sb_dqopt(sb);
625 WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
627 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
628 if (type != -1 && cnt != type)
630 if (!sb_has_quota_active(sb, cnt))
632 spin_lock(&dq_list_lock);
633 dirty = &dqopt->info[cnt].dqi_dirty_list;
634 while (!list_empty(dirty)) {
635 dquot = list_first_entry(dirty, struct dquot,
638 WARN_ON(!test_bit(DQ_ACTIVE_B, &dquot->dq_flags));
640 /* Now we have active dquot from which someone is
641 * holding reference so we can safely just increase
644 spin_unlock(&dq_list_lock);
645 dqstats_inc(DQST_LOOKUPS);
646 err = sb->dq_op->write_dquot(dquot);
650 spin_lock(&dq_list_lock);
652 spin_unlock(&dq_list_lock);
655 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
656 if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
657 && info_dirty(&dqopt->info[cnt]))
658 sb->dq_op->write_info(sb, cnt);
659 dqstats_inc(DQST_SYNCS);
663 EXPORT_SYMBOL(dquot_writeback_dquots);
665 /* Write all dquot structures to disk and make them visible from userspace */
666 int dquot_quota_sync(struct super_block *sb, int type)
668 struct quota_info *dqopt = sb_dqopt(sb);
672 ret = dquot_writeback_dquots(sb, type);
675 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
678 /* This is not very clever (and fast) but currently I don't know about
679 * any other simple way of getting quota data to disk and we must get
680 * them there for userspace to be visible... */
681 if (sb->s_op->sync_fs)
682 sb->s_op->sync_fs(sb, 1);
683 sync_blockdev(sb->s_bdev);
686 * Now when everything is written we can discard the pagecache so
687 * that userspace sees the changes.
689 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
690 if (type != -1 && cnt != type)
692 if (!sb_has_quota_active(sb, cnt))
694 inode_lock(dqopt->files[cnt]);
695 truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
696 inode_unlock(dqopt->files[cnt]);
701 EXPORT_SYMBOL(dquot_quota_sync);
704 dqcache_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
706 struct list_head *head;
708 unsigned long freed = 0;
710 spin_lock(&dq_list_lock);
711 head = free_dquots.prev;
712 while (head != &free_dquots && sc->nr_to_scan) {
713 dquot = list_entry(head, struct dquot, dq_free);
714 remove_dquot_hash(dquot);
715 remove_free_dquot(dquot);
717 do_destroy_dquot(dquot);
720 head = free_dquots.prev;
722 spin_unlock(&dq_list_lock);
727 dqcache_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
729 return vfs_pressure_ratio(
730 percpu_counter_read_positive(&dqstats.counter[DQST_FREE_DQUOTS]));
733 static struct shrinker dqcache_shrinker = {
734 .count_objects = dqcache_shrink_count,
735 .scan_objects = dqcache_shrink_scan,
736 .seeks = DEFAULT_SEEKS,
740 * Put reference to dquot
742 void dqput(struct dquot *dquot)
748 #ifdef CONFIG_QUOTA_DEBUG
749 if (!atomic_read(&dquot->dq_count)) {
750 quota_error(dquot->dq_sb, "trying to free free dquot of %s %d",
751 quotatypes[dquot->dq_id.type],
752 from_kqid(&init_user_ns, dquot->dq_id));
756 dqstats_inc(DQST_DROPS);
758 spin_lock(&dq_list_lock);
759 if (atomic_read(&dquot->dq_count) > 1) {
760 /* We have more than one user... nothing to do */
761 atomic_dec(&dquot->dq_count);
762 /* Releasing dquot during quotaoff phase? */
763 if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_id.type) &&
764 atomic_read(&dquot->dq_count) == 1)
765 wake_up(&dquot_ref_wq);
766 spin_unlock(&dq_list_lock);
769 /* Need to release dquot? */
770 if (dquot_dirty(dquot)) {
771 spin_unlock(&dq_list_lock);
772 /* Commit dquot before releasing */
773 ret = dquot->dq_sb->dq_op->write_dquot(dquot);
775 quota_error(dquot->dq_sb, "Can't write quota structure"
776 " (error %d). Quota may get out of sync!",
779 * We clear dirty bit anyway, so that we avoid
782 clear_dquot_dirty(dquot);
786 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
787 spin_unlock(&dq_list_lock);
788 dquot->dq_sb->dq_op->release_dquot(dquot);
791 atomic_dec(&dquot->dq_count);
792 #ifdef CONFIG_QUOTA_DEBUG
794 BUG_ON(!list_empty(&dquot->dq_free));
796 put_dquot_last(dquot);
797 spin_unlock(&dq_list_lock);
799 EXPORT_SYMBOL(dqput);
801 struct dquot *dquot_alloc(struct super_block *sb, int type)
803 return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
805 EXPORT_SYMBOL(dquot_alloc);
807 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
811 dquot = sb->dq_op->alloc_dquot(sb, type);
815 mutex_init(&dquot->dq_lock);
816 INIT_LIST_HEAD(&dquot->dq_free);
817 INIT_LIST_HEAD(&dquot->dq_inuse);
818 INIT_HLIST_NODE(&dquot->dq_hash);
819 INIT_LIST_HEAD(&dquot->dq_dirty);
821 dquot->dq_id = make_kqid_invalid(type);
822 atomic_set(&dquot->dq_count, 1);
823 spin_lock_init(&dquot->dq_dqb_lock);
829 * Get reference to dquot
831 * Locking is slightly tricky here. We are guarded from parallel quotaoff()
832 * destroying our dquot by:
833 * a) checking for quota flags under dq_list_lock and
834 * b) getting a reference to dquot before we release dq_list_lock
836 struct dquot *dqget(struct super_block *sb, struct kqid qid)
838 unsigned int hashent = hashfn(sb, qid);
839 struct dquot *dquot, *empty = NULL;
841 if (!qid_has_mapping(sb->s_user_ns, qid))
842 return ERR_PTR(-EINVAL);
844 if (!sb_has_quota_active(sb, qid.type))
845 return ERR_PTR(-ESRCH);
847 spin_lock(&dq_list_lock);
848 spin_lock(&dq_state_lock);
849 if (!sb_has_quota_active(sb, qid.type)) {
850 spin_unlock(&dq_state_lock);
851 spin_unlock(&dq_list_lock);
852 dquot = ERR_PTR(-ESRCH);
855 spin_unlock(&dq_state_lock);
857 dquot = find_dquot(hashent, sb, qid);
860 spin_unlock(&dq_list_lock);
861 empty = get_empty_dquot(sb, qid.type);
863 schedule(); /* Try to wait for a moment... */
869 /* all dquots go on the inuse_list */
871 /* hash it first so it can be found */
872 insert_dquot_hash(dquot);
873 spin_unlock(&dq_list_lock);
874 dqstats_inc(DQST_LOOKUPS);
876 if (!atomic_read(&dquot->dq_count))
877 remove_free_dquot(dquot);
878 atomic_inc(&dquot->dq_count);
879 spin_unlock(&dq_list_lock);
880 dqstats_inc(DQST_CACHE_HITS);
881 dqstats_inc(DQST_LOOKUPS);
883 /* Wait for dq_lock - after this we know that either dquot_release() is
884 * already finished or it will be canceled due to dq_count > 1 test */
885 wait_on_dquot(dquot);
886 /* Read the dquot / allocate space in quota file */
887 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
890 err = sb->dq_op->acquire_dquot(dquot);
893 dquot = ERR_PTR(err);
898 * Make sure following reads see filled structure - paired with
899 * smp_mb__before_atomic() in dquot_acquire().
902 #ifdef CONFIG_QUOTA_DEBUG
903 BUG_ON(!dquot->dq_sb); /* Has somebody invalidated entry under us? */
907 do_destroy_dquot(empty);
911 EXPORT_SYMBOL(dqget);
913 static inline struct dquot **i_dquot(struct inode *inode)
915 return inode->i_sb->s_op->get_dquots(inode);
918 static int dqinit_needed(struct inode *inode, int type)
920 struct dquot * const *dquots;
923 if (IS_NOQUOTA(inode))
926 dquots = i_dquot(inode);
928 return !dquots[type];
929 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
935 /* This routine is guarded by s_umount semaphore */
936 static void add_dquot_ref(struct super_block *sb, int type)
938 struct inode *inode, *old_inode = NULL;
939 #ifdef CONFIG_QUOTA_DEBUG
943 spin_lock(&sb->s_inode_list_lock);
944 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
945 spin_lock(&inode->i_lock);
946 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
947 !atomic_read(&inode->i_writecount) ||
948 !dqinit_needed(inode, type)) {
949 spin_unlock(&inode->i_lock);
953 spin_unlock(&inode->i_lock);
954 spin_unlock(&sb->s_inode_list_lock);
956 #ifdef CONFIG_QUOTA_DEBUG
957 if (unlikely(inode_get_rsv_space(inode) > 0))
961 __dquot_initialize(inode, type);
964 * We hold a reference to 'inode' so it couldn't have been
965 * removed from s_inodes list while we dropped the
966 * s_inode_list_lock. We cannot iput the inode now as we can be
967 * holding the last reference and we cannot iput it under
968 * s_inode_list_lock. So we keep the reference and iput it
972 spin_lock(&sb->s_inode_list_lock);
974 spin_unlock(&sb->s_inode_list_lock);
977 #ifdef CONFIG_QUOTA_DEBUG
979 quota_error(sb, "Writes happened before quota was turned on "
980 "thus quota information is probably inconsistent. "
981 "Please run quotacheck(8)");
987 * Remove references to dquots from inode and add dquot to list for freeing
988 * if we have the last reference to dquot
990 static void remove_inode_dquot_ref(struct inode *inode, int type,
991 struct list_head *tofree_head)
993 struct dquot **dquots = i_dquot(inode);
994 struct dquot *dquot = dquots[type];
1000 if (list_empty(&dquot->dq_free)) {
1002 * The inode still has reference to dquot so it can't be in the
1005 spin_lock(&dq_list_lock);
1006 list_add(&dquot->dq_free, tofree_head);
1007 spin_unlock(&dq_list_lock);
1010 * Dquot is already in a list to put so we won't drop the last
1018 * Free list of dquots
1019 * Dquots are removed from inodes and no new references can be got so we are
1020 * the only ones holding reference
1022 static void put_dquot_list(struct list_head *tofree_head)
1024 struct list_head *act_head;
1025 struct dquot *dquot;
1027 act_head = tofree_head->next;
1028 while (act_head != tofree_head) {
1029 dquot = list_entry(act_head, struct dquot, dq_free);
1030 act_head = act_head->next;
1031 /* Remove dquot from the list so we won't have problems... */
1032 list_del_init(&dquot->dq_free);
1037 static void remove_dquot_ref(struct super_block *sb, int type,
1038 struct list_head *tofree_head)
1040 struct inode *inode;
1043 spin_lock(&sb->s_inode_list_lock);
1044 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1046 * We have to scan also I_NEW inodes because they can already
1047 * have quota pointer initialized. Luckily, we need to touch
1048 * only quota pointers and these have separate locking
1051 spin_lock(&dq_data_lock);
1052 if (!IS_NOQUOTA(inode)) {
1053 if (unlikely(inode_get_rsv_space(inode) > 0))
1055 remove_inode_dquot_ref(inode, type, tofree_head);
1057 spin_unlock(&dq_data_lock);
1059 spin_unlock(&sb->s_inode_list_lock);
1060 #ifdef CONFIG_QUOTA_DEBUG
1062 printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1063 " was disabled thus quota information is probably "
1064 "inconsistent. Please run quotacheck(8).\n", sb->s_id);
1069 /* Gather all references from inodes and drop them */
1070 static void drop_dquot_ref(struct super_block *sb, int type)
1072 LIST_HEAD(tofree_head);
1075 remove_dquot_ref(sb, type, &tofree_head);
1076 synchronize_srcu(&dquot_srcu);
1077 put_dquot_list(&tofree_head);
1082 void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1084 if (dquot->dq_dqb.dqb_rsvspace >= number)
1085 dquot->dq_dqb.dqb_rsvspace -= number;
1088 dquot->dq_dqb.dqb_rsvspace = 0;
1090 if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1091 dquot->dq_dqb.dqb_bsoftlimit)
1092 dquot->dq_dqb.dqb_btime = (time64_t) 0;
1093 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1096 static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1098 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1099 dquot->dq_dqb.dqb_curinodes >= number)
1100 dquot->dq_dqb.dqb_curinodes -= number;
1102 dquot->dq_dqb.dqb_curinodes = 0;
1103 if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1104 dquot->dq_dqb.dqb_itime = (time64_t) 0;
1105 clear_bit(DQ_INODES_B, &dquot->dq_flags);
1108 static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1110 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1111 dquot->dq_dqb.dqb_curspace >= number)
1112 dquot->dq_dqb.dqb_curspace -= number;
1114 dquot->dq_dqb.dqb_curspace = 0;
1115 if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1116 dquot->dq_dqb.dqb_bsoftlimit)
1117 dquot->dq_dqb.dqb_btime = (time64_t) 0;
1118 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1122 struct super_block *w_sb;
1123 struct kqid w_dq_id;
1127 static int warning_issued(struct dquot *dquot, const int warntype)
1129 int flag = (warntype == QUOTA_NL_BHARDWARN ||
1130 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1131 ((warntype == QUOTA_NL_IHARDWARN ||
1132 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1136 return test_and_set_bit(flag, &dquot->dq_flags);
1139 #ifdef CONFIG_PRINT_QUOTA_WARNING
1140 static int flag_print_warnings = 1;
1142 static int need_print_warning(struct dquot_warn *warn)
1144 if (!flag_print_warnings)
1147 switch (warn->w_dq_id.type) {
1149 return uid_eq(current_fsuid(), warn->w_dq_id.uid);
1151 return in_group_p(warn->w_dq_id.gid);
1158 /* Print warning to user which exceeded quota */
1159 static void print_warning(struct dquot_warn *warn)
1162 struct tty_struct *tty;
1163 int warntype = warn->w_type;
1165 if (warntype == QUOTA_NL_IHARDBELOW ||
1166 warntype == QUOTA_NL_ISOFTBELOW ||
1167 warntype == QUOTA_NL_BHARDBELOW ||
1168 warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(warn))
1171 tty = get_current_tty();
1174 tty_write_message(tty, warn->w_sb->s_id);
1175 if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1176 tty_write_message(tty, ": warning, ");
1178 tty_write_message(tty, ": write failed, ");
1179 tty_write_message(tty, quotatypes[warn->w_dq_id.type]);
1181 case QUOTA_NL_IHARDWARN:
1182 msg = " file limit reached.\r\n";
1184 case QUOTA_NL_ISOFTLONGWARN:
1185 msg = " file quota exceeded too long.\r\n";
1187 case QUOTA_NL_ISOFTWARN:
1188 msg = " file quota exceeded.\r\n";
1190 case QUOTA_NL_BHARDWARN:
1191 msg = " block limit reached.\r\n";
1193 case QUOTA_NL_BSOFTLONGWARN:
1194 msg = " block quota exceeded too long.\r\n";
1196 case QUOTA_NL_BSOFTWARN:
1197 msg = " block quota exceeded.\r\n";
1200 tty_write_message(tty, msg);
1205 static void prepare_warning(struct dquot_warn *warn, struct dquot *dquot,
1208 if (warning_issued(dquot, warntype))
1210 warn->w_type = warntype;
1211 warn->w_sb = dquot->dq_sb;
1212 warn->w_dq_id = dquot->dq_id;
1216 * Write warnings to the console and send warning messages over netlink.
1218 * Note that this function can call into tty and networking code.
1220 static void flush_warnings(struct dquot_warn *warn)
1224 for (i = 0; i < MAXQUOTAS; i++) {
1225 if (warn[i].w_type == QUOTA_NL_NOWARN)
1227 #ifdef CONFIG_PRINT_QUOTA_WARNING
1228 print_warning(&warn[i]);
1230 quota_send_warning(warn[i].w_dq_id,
1231 warn[i].w_sb->s_dev, warn[i].w_type);
1235 static int ignore_hardlimit(struct dquot *dquot)
1237 struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
1239 return capable(CAP_SYS_RESOURCE) &&
1240 (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1241 !(info->dqi_flags & DQF_ROOT_SQUASH));
1244 static int dquot_add_inodes(struct dquot *dquot, qsize_t inodes,
1245 struct dquot_warn *warn)
1250 spin_lock(&dquot->dq_dqb_lock);
1251 newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1252 if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type) ||
1253 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1256 if (dquot->dq_dqb.dqb_ihardlimit &&
1257 newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1258 !ignore_hardlimit(dquot)) {
1259 prepare_warning(warn, dquot, QUOTA_NL_IHARDWARN);
1264 if (dquot->dq_dqb.dqb_isoftlimit &&
1265 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1266 dquot->dq_dqb.dqb_itime &&
1267 ktime_get_real_seconds() >= dquot->dq_dqb.dqb_itime &&
1268 !ignore_hardlimit(dquot)) {
1269 prepare_warning(warn, dquot, QUOTA_NL_ISOFTLONGWARN);
1274 if (dquot->dq_dqb.dqb_isoftlimit &&
1275 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1276 dquot->dq_dqb.dqb_itime == 0) {
1277 prepare_warning(warn, dquot, QUOTA_NL_ISOFTWARN);
1278 dquot->dq_dqb.dqb_itime = ktime_get_real_seconds() +
1279 sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type].dqi_igrace;
1282 dquot->dq_dqb.dqb_curinodes = newinodes;
1285 spin_unlock(&dquot->dq_dqb_lock);
1289 static int dquot_add_space(struct dquot *dquot, qsize_t space,
1290 qsize_t rsv_space, unsigned int flags,
1291 struct dquot_warn *warn)
1294 struct super_block *sb = dquot->dq_sb;
1297 spin_lock(&dquot->dq_dqb_lock);
1298 if (!sb_has_quota_limits_enabled(sb, dquot->dq_id.type) ||
1299 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1302 tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1303 + space + rsv_space;
1305 if (flags & DQUOT_SPACE_NOFAIL)
1308 if (dquot->dq_dqb.dqb_bhardlimit &&
1309 tspace > dquot->dq_dqb.dqb_bhardlimit &&
1310 !ignore_hardlimit(dquot)) {
1311 if (flags & DQUOT_SPACE_WARN)
1312 prepare_warning(warn, dquot, QUOTA_NL_BHARDWARN);
1317 if (dquot->dq_dqb.dqb_bsoftlimit &&
1318 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1319 dquot->dq_dqb.dqb_btime &&
1320 ktime_get_real_seconds() >= dquot->dq_dqb.dqb_btime &&
1321 !ignore_hardlimit(dquot)) {
1322 if (flags & DQUOT_SPACE_WARN)
1323 prepare_warning(warn, dquot, QUOTA_NL_BSOFTLONGWARN);
1328 if (dquot->dq_dqb.dqb_bsoftlimit &&
1329 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1330 dquot->dq_dqb.dqb_btime == 0) {
1331 if (flags & DQUOT_SPACE_WARN) {
1332 prepare_warning(warn, dquot, QUOTA_NL_BSOFTWARN);
1333 dquot->dq_dqb.dqb_btime = ktime_get_real_seconds() +
1334 sb_dqopt(sb)->info[dquot->dq_id.type].dqi_bgrace;
1337 * We don't allow preallocation to exceed softlimit so exceeding will
1345 dquot->dq_dqb.dqb_rsvspace += rsv_space;
1346 dquot->dq_dqb.dqb_curspace += space;
1348 spin_unlock(&dquot->dq_dqb_lock);
1352 static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1356 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1357 dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1358 !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type))
1359 return QUOTA_NL_NOWARN;
1361 newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1362 if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1363 return QUOTA_NL_ISOFTBELOW;
1364 if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1365 newinodes < dquot->dq_dqb.dqb_ihardlimit)
1366 return QUOTA_NL_IHARDBELOW;
1367 return QUOTA_NL_NOWARN;
1370 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1374 tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace;
1376 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1377 tspace <= dquot->dq_dqb.dqb_bsoftlimit)
1378 return QUOTA_NL_NOWARN;
1380 if (tspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1381 return QUOTA_NL_BSOFTBELOW;
1382 if (tspace >= dquot->dq_dqb.dqb_bhardlimit &&
1383 tspace - space < dquot->dq_dqb.dqb_bhardlimit)
1384 return QUOTA_NL_BHARDBELOW;
1385 return QUOTA_NL_NOWARN;
1388 static int dquot_active(const struct inode *inode)
1390 struct super_block *sb = inode->i_sb;
1392 if (IS_NOQUOTA(inode))
1394 return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1398 * Initialize quota pointers in inode
1400 * It is better to call this function outside of any transaction as it
1401 * might need a lot of space in journal for dquot structure allocation.
1403 static int __dquot_initialize(struct inode *inode, int type)
1405 int cnt, init_needed = 0;
1406 struct dquot **dquots, *got[MAXQUOTAS] = {};
1407 struct super_block *sb = inode->i_sb;
1411 if (!dquot_active(inode))
1414 dquots = i_dquot(inode);
1416 /* First get references to structures we might need. */
1417 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1421 struct dquot *dquot;
1423 if (type != -1 && cnt != type)
1426 * The i_dquot should have been initialized in most cases,
1427 * we check it without locking here to avoid unnecessary
1428 * dqget()/dqput() calls.
1433 if (!sb_has_quota_active(sb, cnt))
1440 qid = make_kqid_uid(inode->i_uid);
1443 qid = make_kqid_gid(inode->i_gid);
1446 rc = inode->i_sb->dq_op->get_projid(inode, &projid);
1449 qid = make_kqid_projid(projid);
1452 dquot = dqget(sb, qid);
1453 if (IS_ERR(dquot)) {
1454 /* We raced with somebody turning quotas off... */
1455 if (PTR_ERR(dquot) != -ESRCH) {
1456 ret = PTR_ERR(dquot);
1464 /* All required i_dquot has been initialized */
1468 spin_lock(&dq_data_lock);
1469 if (IS_NOQUOTA(inode))
1471 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1472 if (type != -1 && cnt != type)
1474 /* Avoid races with quotaoff() */
1475 if (!sb_has_quota_active(sb, cnt))
1477 /* We could race with quotaon or dqget() could have failed */
1481 dquots[cnt] = got[cnt];
1484 * Make quota reservation system happy if someone
1485 * did a write before quota was turned on
1487 rsv = inode_get_rsv_space(inode);
1488 if (unlikely(rsv)) {
1489 spin_lock(&inode->i_lock);
1490 /* Get reservation again under proper lock */
1491 rsv = __inode_get_rsv_space(inode);
1492 spin_lock(&dquots[cnt]->dq_dqb_lock);
1493 dquots[cnt]->dq_dqb.dqb_rsvspace += rsv;
1494 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1495 spin_unlock(&inode->i_lock);
1500 spin_unlock(&dq_data_lock);
1502 /* Drop unused references */
1508 int dquot_initialize(struct inode *inode)
1510 return __dquot_initialize(inode, -1);
1512 EXPORT_SYMBOL(dquot_initialize);
1514 bool dquot_initialize_needed(struct inode *inode)
1516 struct dquot **dquots;
1519 if (!dquot_active(inode))
1522 dquots = i_dquot(inode);
1523 for (i = 0; i < MAXQUOTAS; i++)
1524 if (!dquots[i] && sb_has_quota_active(inode->i_sb, i))
1528 EXPORT_SYMBOL(dquot_initialize_needed);
1531 * Release all quotas referenced by inode.
1533 * This function only be called on inode free or converting
1534 * a file to quota file, no other users for the i_dquot in
1535 * both cases, so we needn't call synchronize_srcu() after
1538 static void __dquot_drop(struct inode *inode)
1541 struct dquot **dquots = i_dquot(inode);
1542 struct dquot *put[MAXQUOTAS];
1544 spin_lock(&dq_data_lock);
1545 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1546 put[cnt] = dquots[cnt];
1549 spin_unlock(&dq_data_lock);
1553 void dquot_drop(struct inode *inode)
1555 struct dquot * const *dquots;
1558 if (IS_NOQUOTA(inode))
1562 * Test before calling to rule out calls from proc and such
1563 * where we are not allowed to block. Note that this is
1564 * actually reliable test even without the lock - the caller
1565 * must assure that nobody can come after the DQUOT_DROP and
1566 * add quota pointers back anyway.
1568 dquots = i_dquot(inode);
1569 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1574 if (cnt < MAXQUOTAS)
1575 __dquot_drop(inode);
1577 EXPORT_SYMBOL(dquot_drop);
1580 * inode_reserved_space is managed internally by quota, and protected by
1581 * i_lock similar to i_blocks+i_bytes.
1583 static qsize_t *inode_reserved_space(struct inode * inode)
1585 /* Filesystem must explicitly define it's own method in order to use
1586 * quota reservation interface */
1587 BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1588 return inode->i_sb->dq_op->get_reserved_space(inode);
1591 static qsize_t __inode_get_rsv_space(struct inode *inode)
1593 if (!inode->i_sb->dq_op->get_reserved_space)
1595 return *inode_reserved_space(inode);
1598 static qsize_t inode_get_rsv_space(struct inode *inode)
1602 if (!inode->i_sb->dq_op->get_reserved_space)
1604 spin_lock(&inode->i_lock);
1605 ret = __inode_get_rsv_space(inode);
1606 spin_unlock(&inode->i_lock);
1611 * This functions updates i_blocks+i_bytes fields and quota information
1612 * (together with appropriate checks).
1614 * NOTE: We absolutely rely on the fact that caller dirties the inode
1615 * (usually helpers in quotaops.h care about this) and holds a handle for
1616 * the current transaction so that dquot write and inode write go into the
1621 * This operation can block, but only after everything is updated
1623 int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
1625 int cnt, ret = 0, index;
1626 struct dquot_warn warn[MAXQUOTAS];
1627 int reserve = flags & DQUOT_SPACE_RESERVE;
1628 struct dquot **dquots;
1630 if (!dquot_active(inode)) {
1632 spin_lock(&inode->i_lock);
1633 *inode_reserved_space(inode) += number;
1634 spin_unlock(&inode->i_lock);
1636 inode_add_bytes(inode, number);
1641 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1642 warn[cnt].w_type = QUOTA_NL_NOWARN;
1644 dquots = i_dquot(inode);
1645 index = srcu_read_lock(&dquot_srcu);
1646 spin_lock(&inode->i_lock);
1647 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1650 if (flags & DQUOT_SPACE_RESERVE) {
1651 ret = dquot_add_space(dquots[cnt], 0, number, flags,
1654 ret = dquot_add_space(dquots[cnt], number, 0, flags,
1658 /* Back out changes we already did */
1659 for (cnt--; cnt >= 0; cnt--) {
1662 spin_lock(&dquots[cnt]->dq_dqb_lock);
1663 if (flags & DQUOT_SPACE_RESERVE) {
1664 dquots[cnt]->dq_dqb.dqb_rsvspace -=
1667 dquots[cnt]->dq_dqb.dqb_curspace -=
1670 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1672 spin_unlock(&inode->i_lock);
1673 goto out_flush_warn;
1677 *inode_reserved_space(inode) += number;
1679 __inode_add_bytes(inode, number);
1680 spin_unlock(&inode->i_lock);
1683 goto out_flush_warn;
1684 mark_all_dquot_dirty(dquots);
1686 srcu_read_unlock(&dquot_srcu, index);
1687 flush_warnings(warn);
1691 EXPORT_SYMBOL(__dquot_alloc_space);
1694 * This operation can block, but only after everything is updated
1696 int dquot_alloc_inode(struct inode *inode)
1698 int cnt, ret = 0, index;
1699 struct dquot_warn warn[MAXQUOTAS];
1700 struct dquot * const *dquots;
1702 if (!dquot_active(inode))
1704 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1705 warn[cnt].w_type = QUOTA_NL_NOWARN;
1707 dquots = i_dquot(inode);
1708 index = srcu_read_lock(&dquot_srcu);
1709 spin_lock(&inode->i_lock);
1710 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1713 ret = dquot_add_inodes(dquots[cnt], 1, &warn[cnt]);
1715 for (cnt--; cnt >= 0; cnt--) {
1718 /* Back out changes we already did */
1719 spin_lock(&dquots[cnt]->dq_dqb_lock);
1720 dquots[cnt]->dq_dqb.dqb_curinodes--;
1721 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1728 spin_unlock(&inode->i_lock);
1730 mark_all_dquot_dirty(dquots);
1731 srcu_read_unlock(&dquot_srcu, index);
1732 flush_warnings(warn);
1735 EXPORT_SYMBOL(dquot_alloc_inode);
1738 * Convert in-memory reserved quotas to real consumed quotas
1740 int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1742 struct dquot **dquots;
1745 if (!dquot_active(inode)) {
1746 spin_lock(&inode->i_lock);
1747 *inode_reserved_space(inode) -= number;
1748 __inode_add_bytes(inode, number);
1749 spin_unlock(&inode->i_lock);
1753 dquots = i_dquot(inode);
1754 index = srcu_read_lock(&dquot_srcu);
1755 spin_lock(&inode->i_lock);
1756 /* Claim reserved quotas to allocated quotas */
1757 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1759 struct dquot *dquot = dquots[cnt];
1761 spin_lock(&dquot->dq_dqb_lock);
1762 if (WARN_ON_ONCE(dquot->dq_dqb.dqb_rsvspace < number))
1763 number = dquot->dq_dqb.dqb_rsvspace;
1764 dquot->dq_dqb.dqb_curspace += number;
1765 dquot->dq_dqb.dqb_rsvspace -= number;
1766 spin_unlock(&dquot->dq_dqb_lock);
1769 /* Update inode bytes */
1770 *inode_reserved_space(inode) -= number;
1771 __inode_add_bytes(inode, number);
1772 spin_unlock(&inode->i_lock);
1773 mark_all_dquot_dirty(dquots);
1774 srcu_read_unlock(&dquot_srcu, index);
1777 EXPORT_SYMBOL(dquot_claim_space_nodirty);
1780 * Convert allocated space back to in-memory reserved quotas
1782 void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
1784 struct dquot **dquots;
1787 if (!dquot_active(inode)) {
1788 spin_lock(&inode->i_lock);
1789 *inode_reserved_space(inode) += number;
1790 __inode_sub_bytes(inode, number);
1791 spin_unlock(&inode->i_lock);
1795 dquots = i_dquot(inode);
1796 index = srcu_read_lock(&dquot_srcu);
1797 spin_lock(&inode->i_lock);
1798 /* Claim reserved quotas to allocated quotas */
1799 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1801 struct dquot *dquot = dquots[cnt];
1803 spin_lock(&dquot->dq_dqb_lock);
1804 if (WARN_ON_ONCE(dquot->dq_dqb.dqb_curspace < number))
1805 number = dquot->dq_dqb.dqb_curspace;
1806 dquot->dq_dqb.dqb_rsvspace += number;
1807 dquot->dq_dqb.dqb_curspace -= number;
1808 spin_unlock(&dquot->dq_dqb_lock);
1811 /* Update inode bytes */
1812 *inode_reserved_space(inode) += number;
1813 __inode_sub_bytes(inode, number);
1814 spin_unlock(&inode->i_lock);
1815 mark_all_dquot_dirty(dquots);
1816 srcu_read_unlock(&dquot_srcu, index);
1819 EXPORT_SYMBOL(dquot_reclaim_space_nodirty);
1822 * This operation can block, but only after everything is updated
1824 void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
1827 struct dquot_warn warn[MAXQUOTAS];
1828 struct dquot **dquots;
1829 int reserve = flags & DQUOT_SPACE_RESERVE, index;
1831 if (!dquot_active(inode)) {
1833 spin_lock(&inode->i_lock);
1834 *inode_reserved_space(inode) -= number;
1835 spin_unlock(&inode->i_lock);
1837 inode_sub_bytes(inode, number);
1842 dquots = i_dquot(inode);
1843 index = srcu_read_lock(&dquot_srcu);
1844 spin_lock(&inode->i_lock);
1845 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1848 warn[cnt].w_type = QUOTA_NL_NOWARN;
1851 spin_lock(&dquots[cnt]->dq_dqb_lock);
1852 wtype = info_bdq_free(dquots[cnt], number);
1853 if (wtype != QUOTA_NL_NOWARN)
1854 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1856 dquot_free_reserved_space(dquots[cnt], number);
1858 dquot_decr_space(dquots[cnt], number);
1859 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1862 *inode_reserved_space(inode) -= number;
1864 __inode_sub_bytes(inode, number);
1865 spin_unlock(&inode->i_lock);
1869 mark_all_dquot_dirty(dquots);
1871 srcu_read_unlock(&dquot_srcu, index);
1872 flush_warnings(warn);
1874 EXPORT_SYMBOL(__dquot_free_space);
1877 * This operation can block, but only after everything is updated
1879 void dquot_free_inode(struct inode *inode)
1882 struct dquot_warn warn[MAXQUOTAS];
1883 struct dquot * const *dquots;
1886 if (!dquot_active(inode))
1889 dquots = i_dquot(inode);
1890 index = srcu_read_lock(&dquot_srcu);
1891 spin_lock(&inode->i_lock);
1892 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1895 warn[cnt].w_type = QUOTA_NL_NOWARN;
1898 spin_lock(&dquots[cnt]->dq_dqb_lock);
1899 wtype = info_idq_free(dquots[cnt], 1);
1900 if (wtype != QUOTA_NL_NOWARN)
1901 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1902 dquot_decr_inodes(dquots[cnt], 1);
1903 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1905 spin_unlock(&inode->i_lock);
1906 mark_all_dquot_dirty(dquots);
1907 srcu_read_unlock(&dquot_srcu, index);
1908 flush_warnings(warn);
1910 EXPORT_SYMBOL(dquot_free_inode);
1913 * Transfer the number of inode and blocks from one diskquota to an other.
1914 * On success, dquot references in transfer_to are consumed and references
1915 * to original dquots that need to be released are placed there. On failure,
1916 * references are kept untouched.
1918 * This operation can block, but only after everything is updated
1919 * A transaction must be started when entering this function.
1921 * We are holding reference on transfer_from & transfer_to, no need to
1922 * protect them by srcu_read_lock().
1924 int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
1927 qsize_t rsv_space = 0;
1928 qsize_t inode_usage = 1;
1929 struct dquot *transfer_from[MAXQUOTAS] = {};
1931 char is_valid[MAXQUOTAS] = {};
1932 struct dquot_warn warn_to[MAXQUOTAS];
1933 struct dquot_warn warn_from_inodes[MAXQUOTAS];
1934 struct dquot_warn warn_from_space[MAXQUOTAS];
1936 if (IS_NOQUOTA(inode))
1939 if (inode->i_sb->dq_op->get_inode_usage) {
1940 ret = inode->i_sb->dq_op->get_inode_usage(inode, &inode_usage);
1945 /* Initialize the arrays */
1946 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1947 warn_to[cnt].w_type = QUOTA_NL_NOWARN;
1948 warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
1949 warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
1952 spin_lock(&dq_data_lock);
1953 spin_lock(&inode->i_lock);
1954 if (IS_NOQUOTA(inode)) { /* File without quota accounting? */
1955 spin_unlock(&inode->i_lock);
1956 spin_unlock(&dq_data_lock);
1959 cur_space = __inode_get_bytes(inode);
1960 rsv_space = __inode_get_rsv_space(inode);
1962 * Build the transfer_from list, check limits, and update usage in
1963 * the target structures.
1965 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1967 * Skip changes for same uid or gid or for turned off quota-type.
1969 if (!transfer_to[cnt])
1971 /* Avoid races with quotaoff() */
1972 if (!sb_has_quota_active(inode->i_sb, cnt))
1975 transfer_from[cnt] = i_dquot(inode)[cnt];
1976 ret = dquot_add_inodes(transfer_to[cnt], inode_usage,
1980 ret = dquot_add_space(transfer_to[cnt], cur_space, rsv_space, 0,
1983 spin_lock(&transfer_to[cnt]->dq_dqb_lock);
1984 dquot_decr_inodes(transfer_to[cnt], inode_usage);
1985 spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
1990 /* Decrease usage for source structures and update quota pointers */
1991 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1994 /* Due to IO error we might not have transfer_from[] structure */
1995 if (transfer_from[cnt]) {
1998 spin_lock(&transfer_from[cnt]->dq_dqb_lock);
1999 wtype = info_idq_free(transfer_from[cnt], inode_usage);
2000 if (wtype != QUOTA_NL_NOWARN)
2001 prepare_warning(&warn_from_inodes[cnt],
2002 transfer_from[cnt], wtype);
2003 wtype = info_bdq_free(transfer_from[cnt],
2004 cur_space + rsv_space);
2005 if (wtype != QUOTA_NL_NOWARN)
2006 prepare_warning(&warn_from_space[cnt],
2007 transfer_from[cnt], wtype);
2008 dquot_decr_inodes(transfer_from[cnt], inode_usage);
2009 dquot_decr_space(transfer_from[cnt], cur_space);
2010 dquot_free_reserved_space(transfer_from[cnt],
2012 spin_unlock(&transfer_from[cnt]->dq_dqb_lock);
2014 i_dquot(inode)[cnt] = transfer_to[cnt];
2016 spin_unlock(&inode->i_lock);
2017 spin_unlock(&dq_data_lock);
2019 mark_all_dquot_dirty(transfer_from);
2020 mark_all_dquot_dirty(transfer_to);
2021 flush_warnings(warn_to);
2022 flush_warnings(warn_from_inodes);
2023 flush_warnings(warn_from_space);
2024 /* Pass back references to put */
2025 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2027 transfer_to[cnt] = transfer_from[cnt];
2030 /* Back out changes we already did */
2031 for (cnt--; cnt >= 0; cnt--) {
2034 spin_lock(&transfer_to[cnt]->dq_dqb_lock);
2035 dquot_decr_inodes(transfer_to[cnt], inode_usage);
2036 dquot_decr_space(transfer_to[cnt], cur_space);
2037 dquot_free_reserved_space(transfer_to[cnt], rsv_space);
2038 spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2040 spin_unlock(&inode->i_lock);
2041 spin_unlock(&dq_data_lock);
2042 flush_warnings(warn_to);
2045 EXPORT_SYMBOL(__dquot_transfer);
2047 /* Wrapper for transferring ownership of an inode for uid/gid only
2048 * Called from FSXXX_setattr()
2050 int dquot_transfer(struct inode *inode, struct iattr *iattr)
2052 struct dquot *transfer_to[MAXQUOTAS] = {};
2053 struct dquot *dquot;
2054 struct super_block *sb = inode->i_sb;
2057 if (!dquot_active(inode))
2060 if (iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)){
2061 dquot = dqget(sb, make_kqid_uid(iattr->ia_uid));
2062 if (IS_ERR(dquot)) {
2063 if (PTR_ERR(dquot) != -ESRCH) {
2064 ret = PTR_ERR(dquot);
2069 transfer_to[USRQUOTA] = dquot;
2071 if (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid)){
2072 dquot = dqget(sb, make_kqid_gid(iattr->ia_gid));
2073 if (IS_ERR(dquot)) {
2074 if (PTR_ERR(dquot) != -ESRCH) {
2075 ret = PTR_ERR(dquot);
2080 transfer_to[GRPQUOTA] = dquot;
2082 ret = __dquot_transfer(inode, transfer_to);
2084 dqput_all(transfer_to);
2087 EXPORT_SYMBOL(dquot_transfer);
2090 * Write info of quota file to disk
2092 int dquot_commit_info(struct super_block *sb, int type)
2094 struct quota_info *dqopt = sb_dqopt(sb);
2096 return dqopt->ops[type]->write_file_info(sb, type);
2098 EXPORT_SYMBOL(dquot_commit_info);
2100 int dquot_get_next_id(struct super_block *sb, struct kqid *qid)
2102 struct quota_info *dqopt = sb_dqopt(sb);
2104 if (!sb_has_quota_active(sb, qid->type))
2106 if (!dqopt->ops[qid->type]->get_next_id)
2108 return dqopt->ops[qid->type]->get_next_id(sb, qid);
2110 EXPORT_SYMBOL(dquot_get_next_id);
2113 * Definitions of diskquota operations.
2115 const struct dquot_operations dquot_operations = {
2116 .write_dquot = dquot_commit,
2117 .acquire_dquot = dquot_acquire,
2118 .release_dquot = dquot_release,
2119 .mark_dirty = dquot_mark_dquot_dirty,
2120 .write_info = dquot_commit_info,
2121 .alloc_dquot = dquot_alloc,
2122 .destroy_dquot = dquot_destroy,
2123 .get_next_id = dquot_get_next_id,
2125 EXPORT_SYMBOL(dquot_operations);
2128 * Generic helper for ->open on filesystems supporting disk quotas.
2130 int dquot_file_open(struct inode *inode, struct file *file)
2134 error = generic_file_open(inode, file);
2135 if (!error && (file->f_mode & FMODE_WRITE))
2136 dquot_initialize(inode);
2139 EXPORT_SYMBOL(dquot_file_open);
2142 * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
2144 int dquot_disable(struct super_block *sb, int type, unsigned int flags)
2147 struct quota_info *dqopt = sb_dqopt(sb);
2148 struct inode *toputinode[MAXQUOTAS];
2150 /* s_umount should be held in exclusive mode */
2151 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2152 up_read(&sb->s_umount);
2154 /* Cannot turn off usage accounting without turning off limits, or
2155 * suspend quotas and simultaneously turn quotas off. */
2156 if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
2157 || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
2158 DQUOT_USAGE_ENABLED)))
2162 * Skip everything if there's nothing to do. We have to do this because
2163 * sometimes we are called when fill_super() failed and calling
2164 * sync_fs() in such cases does no good.
2166 if (!sb_any_quota_loaded(sb))
2169 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2170 toputinode[cnt] = NULL;
2171 if (type != -1 && cnt != type)
2173 if (!sb_has_quota_loaded(sb, cnt))
2176 if (flags & DQUOT_SUSPENDED) {
2177 spin_lock(&dq_state_lock);
2179 dquot_state_flag(DQUOT_SUSPENDED, cnt);
2180 spin_unlock(&dq_state_lock);
2182 spin_lock(&dq_state_lock);
2183 dqopt->flags &= ~dquot_state_flag(flags, cnt);
2184 /* Turning off suspended quotas? */
2185 if (!sb_has_quota_loaded(sb, cnt) &&
2186 sb_has_quota_suspended(sb, cnt)) {
2187 dqopt->flags &= ~dquot_state_flag(
2188 DQUOT_SUSPENDED, cnt);
2189 spin_unlock(&dq_state_lock);
2190 iput(dqopt->files[cnt]);
2191 dqopt->files[cnt] = NULL;
2194 spin_unlock(&dq_state_lock);
2197 /* We still have to keep quota loaded? */
2198 if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
2201 /* Note: these are blocking operations */
2202 drop_dquot_ref(sb, cnt);
2203 invalidate_dquots(sb, cnt);
2205 * Now all dquots should be invalidated, all writes done so we
2206 * should be only users of the info. No locks needed.
2208 if (info_dirty(&dqopt->info[cnt]))
2209 sb->dq_op->write_info(sb, cnt);
2210 if (dqopt->ops[cnt]->free_file_info)
2211 dqopt->ops[cnt]->free_file_info(sb, cnt);
2212 put_quota_format(dqopt->info[cnt].dqi_format);
2214 toputinode[cnt] = dqopt->files[cnt];
2215 if (!sb_has_quota_loaded(sb, cnt))
2216 dqopt->files[cnt] = NULL;
2217 dqopt->info[cnt].dqi_flags = 0;
2218 dqopt->info[cnt].dqi_igrace = 0;
2219 dqopt->info[cnt].dqi_bgrace = 0;
2220 dqopt->ops[cnt] = NULL;
2223 /* Skip syncing and setting flags if quota files are hidden */
2224 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2227 /* Sync the superblock so that buffers with quota data are written to
2228 * disk (and so userspace sees correct data afterwards). */
2229 if (sb->s_op->sync_fs)
2230 sb->s_op->sync_fs(sb, 1);
2231 sync_blockdev(sb->s_bdev);
2232 /* Now the quota files are just ordinary files and we can set the
2233 * inode flags back. Moreover we discard the pagecache so that
2234 * userspace sees the writes we did bypassing the pagecache. We
2235 * must also discard the blockdev buffers so that we see the
2236 * changes done by userspace on the next quotaon() */
2237 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2238 /* This can happen when suspending quotas on remount-ro... */
2239 if (toputinode[cnt] && !sb_has_quota_loaded(sb, cnt)) {
2240 inode_lock(toputinode[cnt]);
2241 toputinode[cnt]->i_flags &= ~S_NOQUOTA;
2242 truncate_inode_pages(&toputinode[cnt]->i_data, 0);
2243 inode_unlock(toputinode[cnt]);
2244 mark_inode_dirty_sync(toputinode[cnt]);
2247 invalidate_bdev(sb->s_bdev);
2249 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2250 if (toputinode[cnt]) {
2251 /* On remount RO, we keep the inode pointer so that we
2252 * can reenable quota on the subsequent remount RW. We
2253 * have to check 'flags' variable and not use sb_has_
2254 * function because another quotaon / quotaoff could
2255 * change global state before we got here. We refuse
2256 * to suspend quotas when there is pending delete on
2257 * the quota file... */
2258 if (!(flags & DQUOT_SUSPENDED))
2259 iput(toputinode[cnt]);
2260 else if (!toputinode[cnt]->i_nlink)
2265 EXPORT_SYMBOL(dquot_disable);
2267 int dquot_quota_off(struct super_block *sb, int type)
2269 return dquot_disable(sb, type,
2270 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2272 EXPORT_SYMBOL(dquot_quota_off);
2275 * Turn quotas on on a device
2279 * Helper function to turn quotas on when we already have the inode of
2280 * quota file and no quota information is loaded.
2282 static int vfs_load_quota_inode(struct inode *inode, int type, int format_id,
2285 struct quota_format_type *fmt = find_quota_format(format_id);
2286 struct super_block *sb = inode->i_sb;
2287 struct quota_info *dqopt = sb_dqopt(sb);
2292 if (!S_ISREG(inode->i_mode)) {
2296 if (IS_RDONLY(inode)) {
2300 if (!sb->s_op->quota_write || !sb->s_op->quota_read ||
2301 (type == PRJQUOTA && sb->dq_op->get_projid == NULL)) {
2305 /* Filesystems outside of init_user_ns not yet supported */
2306 if (sb->s_user_ns != &init_user_ns) {
2310 /* Usage always has to be set... */
2311 if (!(flags & DQUOT_USAGE_ENABLED)) {
2315 if (sb_has_quota_loaded(sb, type)) {
2320 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2321 /* As we bypass the pagecache we must now flush all the
2322 * dirty data and invalidate caches so that kernel sees
2323 * changes from userspace. It is not enough to just flush
2324 * the quota file since if blocksize < pagesize, invalidation
2325 * of the cache could fail because of other unrelated dirty
2327 sync_filesystem(sb);
2328 invalidate_bdev(sb->s_bdev);
2331 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2332 /* We don't want quota and atime on quota files (deadlocks
2333 * possible) Also nobody should write to the file - we use
2334 * special IO operations which ignore the immutable bit. */
2336 inode->i_flags |= S_NOQUOTA;
2337 inode_unlock(inode);
2339 * When S_NOQUOTA is set, remove dquot references as no more
2340 * references can be added
2342 __dquot_drop(inode);
2346 dqopt->files[type] = igrab(inode);
2347 if (!dqopt->files[type])
2348 goto out_file_flags;
2350 if (!fmt->qf_ops->check_quota_file(sb, type))
2353 dqopt->ops[type] = fmt->qf_ops;
2354 dqopt->info[type].dqi_format = fmt;
2355 dqopt->info[type].dqi_fmt_id = format_id;
2356 INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2357 error = dqopt->ops[type]->read_file_info(sb, type);
2360 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE) {
2361 spin_lock(&dq_data_lock);
2362 dqopt->info[type].dqi_flags |= DQF_SYS_FILE;
2363 spin_unlock(&dq_data_lock);
2365 spin_lock(&dq_state_lock);
2366 dqopt->flags |= dquot_state_flag(flags, type);
2367 spin_unlock(&dq_state_lock);
2369 add_dquot_ref(sb, type);
2374 dqopt->files[type] = NULL;
2378 inode->i_flags &= ~S_NOQUOTA;
2379 inode_unlock(inode);
2381 put_quota_format(fmt);
2386 /* Reenable quotas on remount RW */
2387 int dquot_resume(struct super_block *sb, int type)
2389 struct quota_info *dqopt = sb_dqopt(sb);
2390 struct inode *inode;
2394 /* s_umount should be held in exclusive mode */
2395 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2396 up_read(&sb->s_umount);
2398 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2399 if (type != -1 && cnt != type)
2401 if (!sb_has_quota_suspended(sb, cnt))
2404 inode = dqopt->files[cnt];
2405 dqopt->files[cnt] = NULL;
2406 spin_lock(&dq_state_lock);
2407 flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2408 DQUOT_LIMITS_ENABLED,
2410 dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2411 spin_unlock(&dq_state_lock);
2413 flags = dquot_generic_flag(flags, cnt);
2414 ret = vfs_load_quota_inode(inode, cnt,
2415 dqopt->info[cnt].dqi_fmt_id, flags);
2421 EXPORT_SYMBOL(dquot_resume);
2423 int dquot_quota_on(struct super_block *sb, int type, int format_id,
2424 const struct path *path)
2426 int error = security_quota_on(path->dentry);
2429 /* Quota file not on the same filesystem? */
2430 if (path->dentry->d_sb != sb)
2433 error = vfs_load_quota_inode(d_inode(path->dentry), type,
2434 format_id, DQUOT_USAGE_ENABLED |
2435 DQUOT_LIMITS_ENABLED);
2438 EXPORT_SYMBOL(dquot_quota_on);
2441 * More powerful function for turning on quotas allowing setting
2442 * of individual quota flags
2444 int dquot_enable(struct inode *inode, int type, int format_id,
2447 struct super_block *sb = inode->i_sb;
2449 /* Just unsuspend quotas? */
2450 BUG_ON(flags & DQUOT_SUSPENDED);
2451 /* s_umount should be held in exclusive mode */
2452 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2453 up_read(&sb->s_umount);
2457 /* Just updating flags needed? */
2458 if (sb_has_quota_loaded(sb, type)) {
2459 if (flags & DQUOT_USAGE_ENABLED &&
2460 sb_has_quota_usage_enabled(sb, type))
2462 if (flags & DQUOT_LIMITS_ENABLED &&
2463 sb_has_quota_limits_enabled(sb, type))
2465 spin_lock(&dq_state_lock);
2466 sb_dqopt(sb)->flags |= dquot_state_flag(flags, type);
2467 spin_unlock(&dq_state_lock);
2471 return vfs_load_quota_inode(inode, type, format_id, flags);
2473 EXPORT_SYMBOL(dquot_enable);
2476 * This function is used when filesystem needs to initialize quotas
2477 * during mount time.
2479 int dquot_quota_on_mount(struct super_block *sb, char *qf_name,
2480 int format_id, int type)
2482 struct dentry *dentry;
2485 dentry = lookup_one_len_unlocked(qf_name, sb->s_root, strlen(qf_name));
2487 return PTR_ERR(dentry);
2489 if (d_really_is_negative(dentry)) {
2494 error = security_quota_on(dentry);
2496 error = vfs_load_quota_inode(d_inode(dentry), type, format_id,
2497 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2503 EXPORT_SYMBOL(dquot_quota_on_mount);
2505 static int dquot_quota_enable(struct super_block *sb, unsigned int flags)
2509 struct quota_info *dqopt = sb_dqopt(sb);
2511 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2513 /* Accounting cannot be turned on while fs is mounted */
2514 flags &= ~(FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT);
2517 for (type = 0; type < MAXQUOTAS; type++) {
2518 if (!(flags & qtype_enforce_flag(type)))
2520 /* Can't enforce without accounting */
2521 if (!sb_has_quota_usage_enabled(sb, type))
2523 ret = dquot_enable(dqopt->files[type], type,
2524 dqopt->info[type].dqi_fmt_id,
2525 DQUOT_LIMITS_ENABLED);
2531 /* Backout enforcement enablement we already did */
2532 for (type--; type >= 0; type--) {
2533 if (flags & qtype_enforce_flag(type))
2534 dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2536 /* Error code translation for better compatibility with XFS */
2542 static int dquot_quota_disable(struct super_block *sb, unsigned int flags)
2546 struct quota_info *dqopt = sb_dqopt(sb);
2548 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2551 * We don't support turning off accounting via quotactl. In principle
2552 * quota infrastructure can do this but filesystems don't expect
2553 * userspace to be able to do it.
2556 (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT))
2559 /* Filter out limits not enabled */
2560 for (type = 0; type < MAXQUOTAS; type++)
2561 if (!sb_has_quota_limits_enabled(sb, type))
2562 flags &= ~qtype_enforce_flag(type);
2566 for (type = 0; type < MAXQUOTAS; type++) {
2567 if (flags & qtype_enforce_flag(type)) {
2568 ret = dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2575 /* Backout enforcement disabling we already did */
2576 for (type--; type >= 0; type--) {
2577 if (flags & qtype_enforce_flag(type))
2578 dquot_enable(dqopt->files[type], type,
2579 dqopt->info[type].dqi_fmt_id,
2580 DQUOT_LIMITS_ENABLED);
2585 /* Generic routine for getting common part of quota structure */
2586 static void do_get_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2588 struct mem_dqblk *dm = &dquot->dq_dqb;
2590 memset(di, 0, sizeof(*di));
2591 spin_lock(&dquot->dq_dqb_lock);
2592 di->d_spc_hardlimit = dm->dqb_bhardlimit;
2593 di->d_spc_softlimit = dm->dqb_bsoftlimit;
2594 di->d_ino_hardlimit = dm->dqb_ihardlimit;
2595 di->d_ino_softlimit = dm->dqb_isoftlimit;
2596 di->d_space = dm->dqb_curspace + dm->dqb_rsvspace;
2597 di->d_ino_count = dm->dqb_curinodes;
2598 di->d_spc_timer = dm->dqb_btime;
2599 di->d_ino_timer = dm->dqb_itime;
2600 spin_unlock(&dquot->dq_dqb_lock);
2603 int dquot_get_dqblk(struct super_block *sb, struct kqid qid,
2604 struct qc_dqblk *di)
2606 struct dquot *dquot;
2608 dquot = dqget(sb, qid);
2610 return PTR_ERR(dquot);
2611 do_get_dqblk(dquot, di);
2616 EXPORT_SYMBOL(dquot_get_dqblk);
2618 int dquot_get_next_dqblk(struct super_block *sb, struct kqid *qid,
2619 struct qc_dqblk *di)
2621 struct dquot *dquot;
2624 if (!sb->dq_op->get_next_id)
2626 err = sb->dq_op->get_next_id(sb, qid);
2629 dquot = dqget(sb, *qid);
2631 return PTR_ERR(dquot);
2632 do_get_dqblk(dquot, di);
2637 EXPORT_SYMBOL(dquot_get_next_dqblk);
2639 #define VFS_QC_MASK \
2640 (QC_SPACE | QC_SPC_SOFT | QC_SPC_HARD | \
2641 QC_INO_COUNT | QC_INO_SOFT | QC_INO_HARD | \
2642 QC_SPC_TIMER | QC_INO_TIMER)
2644 /* Generic routine for setting common part of quota structure */
2645 static int do_set_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2647 struct mem_dqblk *dm = &dquot->dq_dqb;
2648 int check_blim = 0, check_ilim = 0;
2649 struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
2651 if (di->d_fieldmask & ~VFS_QC_MASK)
2654 if (((di->d_fieldmask & QC_SPC_SOFT) &&
2655 di->d_spc_softlimit > dqi->dqi_max_spc_limit) ||
2656 ((di->d_fieldmask & QC_SPC_HARD) &&
2657 di->d_spc_hardlimit > dqi->dqi_max_spc_limit) ||
2658 ((di->d_fieldmask & QC_INO_SOFT) &&
2659 (di->d_ino_softlimit > dqi->dqi_max_ino_limit)) ||
2660 ((di->d_fieldmask & QC_INO_HARD) &&
2661 (di->d_ino_hardlimit > dqi->dqi_max_ino_limit)))
2664 spin_lock(&dquot->dq_dqb_lock);
2665 if (di->d_fieldmask & QC_SPACE) {
2666 dm->dqb_curspace = di->d_space - dm->dqb_rsvspace;
2668 set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2671 if (di->d_fieldmask & QC_SPC_SOFT)
2672 dm->dqb_bsoftlimit = di->d_spc_softlimit;
2673 if (di->d_fieldmask & QC_SPC_HARD)
2674 dm->dqb_bhardlimit = di->d_spc_hardlimit;
2675 if (di->d_fieldmask & (QC_SPC_SOFT | QC_SPC_HARD)) {
2677 set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2680 if (di->d_fieldmask & QC_INO_COUNT) {
2681 dm->dqb_curinodes = di->d_ino_count;
2683 set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2686 if (di->d_fieldmask & QC_INO_SOFT)
2687 dm->dqb_isoftlimit = di->d_ino_softlimit;
2688 if (di->d_fieldmask & QC_INO_HARD)
2689 dm->dqb_ihardlimit = di->d_ino_hardlimit;
2690 if (di->d_fieldmask & (QC_INO_SOFT | QC_INO_HARD)) {
2692 set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2695 if (di->d_fieldmask & QC_SPC_TIMER) {
2696 dm->dqb_btime = di->d_spc_timer;
2698 set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2701 if (di->d_fieldmask & QC_INO_TIMER) {
2702 dm->dqb_itime = di->d_ino_timer;
2704 set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2708 if (!dm->dqb_bsoftlimit ||
2709 dm->dqb_curspace + dm->dqb_rsvspace < dm->dqb_bsoftlimit) {
2711 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2712 } else if (!(di->d_fieldmask & QC_SPC_TIMER))
2713 /* Set grace only if user hasn't provided his own... */
2714 dm->dqb_btime = ktime_get_real_seconds() + dqi->dqi_bgrace;
2717 if (!dm->dqb_isoftlimit ||
2718 dm->dqb_curinodes < dm->dqb_isoftlimit) {
2720 clear_bit(DQ_INODES_B, &dquot->dq_flags);
2721 } else if (!(di->d_fieldmask & QC_INO_TIMER))
2722 /* Set grace only if user hasn't provided his own... */
2723 dm->dqb_itime = ktime_get_real_seconds() + dqi->dqi_igrace;
2725 if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2727 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2729 set_bit(DQ_FAKE_B, &dquot->dq_flags);
2730 spin_unlock(&dquot->dq_dqb_lock);
2731 mark_dquot_dirty(dquot);
2736 int dquot_set_dqblk(struct super_block *sb, struct kqid qid,
2737 struct qc_dqblk *di)
2739 struct dquot *dquot;
2742 dquot = dqget(sb, qid);
2743 if (IS_ERR(dquot)) {
2744 rc = PTR_ERR(dquot);
2747 rc = do_set_dqblk(dquot, di);
2752 EXPORT_SYMBOL(dquot_set_dqblk);
2754 /* Generic routine for getting common part of quota file information */
2755 int dquot_get_state(struct super_block *sb, struct qc_state *state)
2757 struct mem_dqinfo *mi;
2758 struct qc_type_state *tstate;
2759 struct quota_info *dqopt = sb_dqopt(sb);
2762 memset(state, 0, sizeof(*state));
2763 for (type = 0; type < MAXQUOTAS; type++) {
2764 if (!sb_has_quota_active(sb, type))
2766 tstate = state->s_state + type;
2767 mi = sb_dqopt(sb)->info + type;
2768 tstate->flags = QCI_ACCT_ENABLED;
2769 spin_lock(&dq_data_lock);
2770 if (mi->dqi_flags & DQF_SYS_FILE)
2771 tstate->flags |= QCI_SYSFILE;
2772 if (mi->dqi_flags & DQF_ROOT_SQUASH)
2773 tstate->flags |= QCI_ROOT_SQUASH;
2774 if (sb_has_quota_limits_enabled(sb, type))
2775 tstate->flags |= QCI_LIMITS_ENFORCED;
2776 tstate->spc_timelimit = mi->dqi_bgrace;
2777 tstate->ino_timelimit = mi->dqi_igrace;
2778 tstate->ino = dqopt->files[type]->i_ino;
2779 tstate->blocks = dqopt->files[type]->i_blocks;
2780 tstate->nextents = 1; /* We don't know... */
2781 spin_unlock(&dq_data_lock);
2785 EXPORT_SYMBOL(dquot_get_state);
2787 /* Generic routine for setting common part of quota file information */
2788 int dquot_set_dqinfo(struct super_block *sb, int type, struct qc_info *ii)
2790 struct mem_dqinfo *mi;
2793 if ((ii->i_fieldmask & QC_WARNS_MASK) ||
2794 (ii->i_fieldmask & QC_RT_SPC_TIMER))
2796 if (!sb_has_quota_active(sb, type))
2798 mi = sb_dqopt(sb)->info + type;
2799 if (ii->i_fieldmask & QC_FLAGS) {
2800 if ((ii->i_flags & QCI_ROOT_SQUASH &&
2801 mi->dqi_format->qf_fmt_id != QFMT_VFS_OLD))
2804 spin_lock(&dq_data_lock);
2805 if (ii->i_fieldmask & QC_SPC_TIMER)
2806 mi->dqi_bgrace = ii->i_spc_timelimit;
2807 if (ii->i_fieldmask & QC_INO_TIMER)
2808 mi->dqi_igrace = ii->i_ino_timelimit;
2809 if (ii->i_fieldmask & QC_FLAGS) {
2810 if (ii->i_flags & QCI_ROOT_SQUASH)
2811 mi->dqi_flags |= DQF_ROOT_SQUASH;
2813 mi->dqi_flags &= ~DQF_ROOT_SQUASH;
2815 spin_unlock(&dq_data_lock);
2816 mark_info_dirty(sb, type);
2817 /* Force write to disk */
2818 sb->dq_op->write_info(sb, type);
2821 EXPORT_SYMBOL(dquot_set_dqinfo);
2823 const struct quotactl_ops dquot_quotactl_sysfile_ops = {
2824 .quota_enable = dquot_quota_enable,
2825 .quota_disable = dquot_quota_disable,
2826 .quota_sync = dquot_quota_sync,
2827 .get_state = dquot_get_state,
2828 .set_info = dquot_set_dqinfo,
2829 .get_dqblk = dquot_get_dqblk,
2830 .get_nextdqblk = dquot_get_next_dqblk,
2831 .set_dqblk = dquot_set_dqblk
2833 EXPORT_SYMBOL(dquot_quotactl_sysfile_ops);
2835 static int do_proc_dqstats(struct ctl_table *table, int write,
2836 void __user *buffer, size_t *lenp, loff_t *ppos)
2838 unsigned int type = (int *)table->data - dqstats.stat;
2840 /* Update global table */
2841 dqstats.stat[type] =
2842 percpu_counter_sum_positive(&dqstats.counter[type]);
2843 return proc_dointvec(table, write, buffer, lenp, ppos);
2846 static struct ctl_table fs_dqstats_table[] = {
2848 .procname = "lookups",
2849 .data = &dqstats.stat[DQST_LOOKUPS],
2850 .maxlen = sizeof(int),
2852 .proc_handler = do_proc_dqstats,
2855 .procname = "drops",
2856 .data = &dqstats.stat[DQST_DROPS],
2857 .maxlen = sizeof(int),
2859 .proc_handler = do_proc_dqstats,
2862 .procname = "reads",
2863 .data = &dqstats.stat[DQST_READS],
2864 .maxlen = sizeof(int),
2866 .proc_handler = do_proc_dqstats,
2869 .procname = "writes",
2870 .data = &dqstats.stat[DQST_WRITES],
2871 .maxlen = sizeof(int),
2873 .proc_handler = do_proc_dqstats,
2876 .procname = "cache_hits",
2877 .data = &dqstats.stat[DQST_CACHE_HITS],
2878 .maxlen = sizeof(int),
2880 .proc_handler = do_proc_dqstats,
2883 .procname = "allocated_dquots",
2884 .data = &dqstats.stat[DQST_ALLOC_DQUOTS],
2885 .maxlen = sizeof(int),
2887 .proc_handler = do_proc_dqstats,
2890 .procname = "free_dquots",
2891 .data = &dqstats.stat[DQST_FREE_DQUOTS],
2892 .maxlen = sizeof(int),
2894 .proc_handler = do_proc_dqstats,
2897 .procname = "syncs",
2898 .data = &dqstats.stat[DQST_SYNCS],
2899 .maxlen = sizeof(int),
2901 .proc_handler = do_proc_dqstats,
2903 #ifdef CONFIG_PRINT_QUOTA_WARNING
2905 .procname = "warnings",
2906 .data = &flag_print_warnings,
2907 .maxlen = sizeof(int),
2909 .proc_handler = proc_dointvec,
2915 static struct ctl_table fs_table[] = {
2917 .procname = "quota",
2919 .child = fs_dqstats_table,
2924 static struct ctl_table sys_table[] = {
2933 static int __init dquot_init(void)
2936 unsigned long nr_hash, order;
2938 printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2940 register_sysctl_table(sys_table);
2942 dquot_cachep = kmem_cache_create("dquot",
2943 sizeof(struct dquot), sizeof(unsigned long) * 4,
2944 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2945 SLAB_MEM_SPREAD|SLAB_PANIC),
2949 dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
2951 panic("Cannot create dquot hash table");
2953 for (i = 0; i < _DQST_DQSTAT_LAST; i++) {
2954 ret = percpu_counter_init(&dqstats.counter[i], 0, GFP_KERNEL);
2956 panic("Cannot create dquot stat counters");
2959 /* Find power-of-two hlist_heads which can fit into allocation */
2960 nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2964 } while (nr_hash >> dq_hash_bits);
2967 nr_hash = 1UL << dq_hash_bits;
2968 dq_hash_mask = nr_hash - 1;
2969 for (i = 0; i < nr_hash; i++)
2970 INIT_HLIST_HEAD(dquot_hash + i);
2972 pr_info("VFS: Dquot-cache hash table entries: %ld (order %ld,"
2973 " %ld bytes)\n", nr_hash, order, (PAGE_SIZE << order));
2975 register_shrinker(&dqcache_shrinker);
2979 fs_initcall(dquot_init);