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 (atomic_read(&dquot->dq_count) > 1)
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 dirty = &dqopt->info[cnt].dqi_dirty_list;
635 while (!list_empty(dirty)) {
636 dquot = list_first_entry(dirty, struct dquot,
639 WARN_ON(!test_bit(DQ_ACTIVE_B, &dquot->dq_flags));
641 /* Now we have active dquot from which someone is
642 * holding reference so we can safely just increase
645 spin_unlock(&dq_list_lock);
646 dqstats_inc(DQST_LOOKUPS);
647 err = sb->dq_op->write_dquot(dquot);
650 * Clear dirty bit anyway to avoid infinite
653 clear_dquot_dirty(dquot);
658 spin_lock(&dq_list_lock);
660 spin_unlock(&dq_list_lock);
663 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
664 if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
665 && info_dirty(&dqopt->info[cnt]))
666 sb->dq_op->write_info(sb, cnt);
667 dqstats_inc(DQST_SYNCS);
671 EXPORT_SYMBOL(dquot_writeback_dquots);
673 /* Write all dquot structures to disk and make them visible from userspace */
674 int dquot_quota_sync(struct super_block *sb, int type)
676 struct quota_info *dqopt = sb_dqopt(sb);
680 ret = dquot_writeback_dquots(sb, type);
683 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
686 /* This is not very clever (and fast) but currently I don't know about
687 * any other simple way of getting quota data to disk and we must get
688 * them there for userspace to be visible... */
689 if (sb->s_op->sync_fs)
690 sb->s_op->sync_fs(sb, 1);
691 sync_blockdev(sb->s_bdev);
694 * Now when everything is written we can discard the pagecache so
695 * that userspace sees the changes.
697 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
698 if (type != -1 && cnt != type)
700 if (!sb_has_quota_active(sb, cnt))
702 inode_lock(dqopt->files[cnt]);
703 truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
704 inode_unlock(dqopt->files[cnt]);
709 EXPORT_SYMBOL(dquot_quota_sync);
712 dqcache_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
714 struct list_head *head;
716 unsigned long freed = 0;
718 spin_lock(&dq_list_lock);
719 head = free_dquots.prev;
720 while (head != &free_dquots && sc->nr_to_scan) {
721 dquot = list_entry(head, struct dquot, dq_free);
722 remove_dquot_hash(dquot);
723 remove_free_dquot(dquot);
725 do_destroy_dquot(dquot);
728 head = free_dquots.prev;
730 spin_unlock(&dq_list_lock);
735 dqcache_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
737 return vfs_pressure_ratio(
738 percpu_counter_read_positive(&dqstats.counter[DQST_FREE_DQUOTS]));
741 static struct shrinker dqcache_shrinker = {
742 .count_objects = dqcache_shrink_count,
743 .scan_objects = dqcache_shrink_scan,
744 .seeks = DEFAULT_SEEKS,
748 * Put reference to dquot
750 void dqput(struct dquot *dquot)
756 #ifdef CONFIG_QUOTA_DEBUG
757 if (!atomic_read(&dquot->dq_count)) {
758 quota_error(dquot->dq_sb, "trying to free free dquot of %s %d",
759 quotatypes[dquot->dq_id.type],
760 from_kqid(&init_user_ns, dquot->dq_id));
764 dqstats_inc(DQST_DROPS);
766 spin_lock(&dq_list_lock);
767 if (atomic_read(&dquot->dq_count) > 1) {
768 /* We have more than one user... nothing to do */
769 atomic_dec(&dquot->dq_count);
770 /* Releasing dquot during quotaoff phase? */
771 if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_id.type) &&
772 atomic_read(&dquot->dq_count) == 1)
773 wake_up(&dquot_ref_wq);
774 spin_unlock(&dq_list_lock);
777 /* Need to release dquot? */
778 if (dquot_dirty(dquot)) {
779 spin_unlock(&dq_list_lock);
780 /* Commit dquot before releasing */
781 ret = dquot->dq_sb->dq_op->write_dquot(dquot);
783 quota_error(dquot->dq_sb, "Can't write quota structure"
784 " (error %d). Quota may get out of sync!",
787 * We clear dirty bit anyway, so that we avoid
790 clear_dquot_dirty(dquot);
794 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
795 spin_unlock(&dq_list_lock);
796 dquot->dq_sb->dq_op->release_dquot(dquot);
799 atomic_dec(&dquot->dq_count);
800 #ifdef CONFIG_QUOTA_DEBUG
802 BUG_ON(!list_empty(&dquot->dq_free));
804 put_dquot_last(dquot);
805 spin_unlock(&dq_list_lock);
807 EXPORT_SYMBOL(dqput);
809 struct dquot *dquot_alloc(struct super_block *sb, int type)
811 return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
813 EXPORT_SYMBOL(dquot_alloc);
815 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
819 dquot = sb->dq_op->alloc_dquot(sb, type);
823 mutex_init(&dquot->dq_lock);
824 INIT_LIST_HEAD(&dquot->dq_free);
825 INIT_LIST_HEAD(&dquot->dq_inuse);
826 INIT_HLIST_NODE(&dquot->dq_hash);
827 INIT_LIST_HEAD(&dquot->dq_dirty);
829 dquot->dq_id = make_kqid_invalid(type);
830 atomic_set(&dquot->dq_count, 1);
831 spin_lock_init(&dquot->dq_dqb_lock);
837 * Get reference to dquot
839 * Locking is slightly tricky here. We are guarded from parallel quotaoff()
840 * destroying our dquot by:
841 * a) checking for quota flags under dq_list_lock and
842 * b) getting a reference to dquot before we release dq_list_lock
844 struct dquot *dqget(struct super_block *sb, struct kqid qid)
846 unsigned int hashent = hashfn(sb, qid);
847 struct dquot *dquot, *empty = NULL;
849 if (!qid_has_mapping(sb->s_user_ns, qid))
850 return ERR_PTR(-EINVAL);
852 if (!sb_has_quota_active(sb, qid.type))
853 return ERR_PTR(-ESRCH);
855 spin_lock(&dq_list_lock);
856 spin_lock(&dq_state_lock);
857 if (!sb_has_quota_active(sb, qid.type)) {
858 spin_unlock(&dq_state_lock);
859 spin_unlock(&dq_list_lock);
860 dquot = ERR_PTR(-ESRCH);
863 spin_unlock(&dq_state_lock);
865 dquot = find_dquot(hashent, sb, qid);
868 spin_unlock(&dq_list_lock);
869 empty = get_empty_dquot(sb, qid.type);
871 schedule(); /* Try to wait for a moment... */
877 /* all dquots go on the inuse_list */
879 /* hash it first so it can be found */
880 insert_dquot_hash(dquot);
881 spin_unlock(&dq_list_lock);
882 dqstats_inc(DQST_LOOKUPS);
884 if (!atomic_read(&dquot->dq_count))
885 remove_free_dquot(dquot);
886 atomic_inc(&dquot->dq_count);
887 spin_unlock(&dq_list_lock);
888 dqstats_inc(DQST_CACHE_HITS);
889 dqstats_inc(DQST_LOOKUPS);
891 /* Wait for dq_lock - after this we know that either dquot_release() is
892 * already finished or it will be canceled due to dq_count > 1 test */
893 wait_on_dquot(dquot);
894 /* Read the dquot / allocate space in quota file */
895 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
898 err = sb->dq_op->acquire_dquot(dquot);
901 dquot = ERR_PTR(err);
906 * Make sure following reads see filled structure - paired with
907 * smp_mb__before_atomic() in dquot_acquire().
910 #ifdef CONFIG_QUOTA_DEBUG
911 BUG_ON(!dquot->dq_sb); /* Has somebody invalidated entry under us? */
915 do_destroy_dquot(empty);
919 EXPORT_SYMBOL(dqget);
921 static inline struct dquot **i_dquot(struct inode *inode)
923 return inode->i_sb->s_op->get_dquots(inode);
926 static int dqinit_needed(struct inode *inode, int type)
928 struct dquot * const *dquots;
931 if (IS_NOQUOTA(inode))
934 dquots = i_dquot(inode);
936 return !dquots[type];
937 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
943 /* This routine is guarded by s_umount semaphore */
944 static void add_dquot_ref(struct super_block *sb, int type)
946 struct inode *inode, *old_inode = NULL;
947 #ifdef CONFIG_QUOTA_DEBUG
951 spin_lock(&sb->s_inode_list_lock);
952 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
953 spin_lock(&inode->i_lock);
954 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
955 !atomic_read(&inode->i_writecount) ||
956 !dqinit_needed(inode, type)) {
957 spin_unlock(&inode->i_lock);
961 spin_unlock(&inode->i_lock);
962 spin_unlock(&sb->s_inode_list_lock);
964 #ifdef CONFIG_QUOTA_DEBUG
965 if (unlikely(inode_get_rsv_space(inode) > 0))
969 __dquot_initialize(inode, type);
972 * We hold a reference to 'inode' so it couldn't have been
973 * removed from s_inodes list while we dropped the
974 * s_inode_list_lock. We cannot iput the inode now as we can be
975 * holding the last reference and we cannot iput it under
976 * s_inode_list_lock. So we keep the reference and iput it
980 spin_lock(&sb->s_inode_list_lock);
982 spin_unlock(&sb->s_inode_list_lock);
985 #ifdef CONFIG_QUOTA_DEBUG
987 quota_error(sb, "Writes happened before quota was turned on "
988 "thus quota information is probably inconsistent. "
989 "Please run quotacheck(8)");
995 * Remove references to dquots from inode and add dquot to list for freeing
996 * if we have the last reference to dquot
998 static void remove_inode_dquot_ref(struct inode *inode, int type,
999 struct list_head *tofree_head)
1001 struct dquot **dquots = i_dquot(inode);
1002 struct dquot *dquot = dquots[type];
1007 dquots[type] = NULL;
1008 if (list_empty(&dquot->dq_free)) {
1010 * The inode still has reference to dquot so it can't be in the
1013 spin_lock(&dq_list_lock);
1014 list_add(&dquot->dq_free, tofree_head);
1015 spin_unlock(&dq_list_lock);
1018 * Dquot is already in a list to put so we won't drop the last
1026 * Free list of dquots
1027 * Dquots are removed from inodes and no new references can be got so we are
1028 * the only ones holding reference
1030 static void put_dquot_list(struct list_head *tofree_head)
1032 struct list_head *act_head;
1033 struct dquot *dquot;
1035 act_head = tofree_head->next;
1036 while (act_head != tofree_head) {
1037 dquot = list_entry(act_head, struct dquot, dq_free);
1038 act_head = act_head->next;
1039 /* Remove dquot from the list so we won't have problems... */
1040 list_del_init(&dquot->dq_free);
1045 static void remove_dquot_ref(struct super_block *sb, int type,
1046 struct list_head *tofree_head)
1048 struct inode *inode;
1051 spin_lock(&sb->s_inode_list_lock);
1052 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1054 * We have to scan also I_NEW inodes because they can already
1055 * have quota pointer initialized. Luckily, we need to touch
1056 * only quota pointers and these have separate locking
1059 spin_lock(&dq_data_lock);
1060 if (!IS_NOQUOTA(inode)) {
1061 if (unlikely(inode_get_rsv_space(inode) > 0))
1063 remove_inode_dquot_ref(inode, type, tofree_head);
1065 spin_unlock(&dq_data_lock);
1067 spin_unlock(&sb->s_inode_list_lock);
1068 #ifdef CONFIG_QUOTA_DEBUG
1070 printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1071 " was disabled thus quota information is probably "
1072 "inconsistent. Please run quotacheck(8).\n", sb->s_id);
1077 /* Gather all references from inodes and drop them */
1078 static void drop_dquot_ref(struct super_block *sb, int type)
1080 LIST_HEAD(tofree_head);
1083 remove_dquot_ref(sb, type, &tofree_head);
1084 synchronize_srcu(&dquot_srcu);
1085 put_dquot_list(&tofree_head);
1090 void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1092 if (dquot->dq_dqb.dqb_rsvspace >= number)
1093 dquot->dq_dqb.dqb_rsvspace -= number;
1096 dquot->dq_dqb.dqb_rsvspace = 0;
1098 if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1099 dquot->dq_dqb.dqb_bsoftlimit)
1100 dquot->dq_dqb.dqb_btime = (time64_t) 0;
1101 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1104 static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1106 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1107 dquot->dq_dqb.dqb_curinodes >= number)
1108 dquot->dq_dqb.dqb_curinodes -= number;
1110 dquot->dq_dqb.dqb_curinodes = 0;
1111 if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1112 dquot->dq_dqb.dqb_itime = (time64_t) 0;
1113 clear_bit(DQ_INODES_B, &dquot->dq_flags);
1116 static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1118 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1119 dquot->dq_dqb.dqb_curspace >= number)
1120 dquot->dq_dqb.dqb_curspace -= number;
1122 dquot->dq_dqb.dqb_curspace = 0;
1123 if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1124 dquot->dq_dqb.dqb_bsoftlimit)
1125 dquot->dq_dqb.dqb_btime = (time64_t) 0;
1126 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1130 struct super_block *w_sb;
1131 struct kqid w_dq_id;
1135 static int warning_issued(struct dquot *dquot, const int warntype)
1137 int flag = (warntype == QUOTA_NL_BHARDWARN ||
1138 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1139 ((warntype == QUOTA_NL_IHARDWARN ||
1140 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1144 return test_and_set_bit(flag, &dquot->dq_flags);
1147 #ifdef CONFIG_PRINT_QUOTA_WARNING
1148 static int flag_print_warnings = 1;
1150 static int need_print_warning(struct dquot_warn *warn)
1152 if (!flag_print_warnings)
1155 switch (warn->w_dq_id.type) {
1157 return uid_eq(current_fsuid(), warn->w_dq_id.uid);
1159 return in_group_p(warn->w_dq_id.gid);
1166 /* Print warning to user which exceeded quota */
1167 static void print_warning(struct dquot_warn *warn)
1170 struct tty_struct *tty;
1171 int warntype = warn->w_type;
1173 if (warntype == QUOTA_NL_IHARDBELOW ||
1174 warntype == QUOTA_NL_ISOFTBELOW ||
1175 warntype == QUOTA_NL_BHARDBELOW ||
1176 warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(warn))
1179 tty = get_current_tty();
1182 tty_write_message(tty, warn->w_sb->s_id);
1183 if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1184 tty_write_message(tty, ": warning, ");
1186 tty_write_message(tty, ": write failed, ");
1187 tty_write_message(tty, quotatypes[warn->w_dq_id.type]);
1189 case QUOTA_NL_IHARDWARN:
1190 msg = " file limit reached.\r\n";
1192 case QUOTA_NL_ISOFTLONGWARN:
1193 msg = " file quota exceeded too long.\r\n";
1195 case QUOTA_NL_ISOFTWARN:
1196 msg = " file quota exceeded.\r\n";
1198 case QUOTA_NL_BHARDWARN:
1199 msg = " block limit reached.\r\n";
1201 case QUOTA_NL_BSOFTLONGWARN:
1202 msg = " block quota exceeded too long.\r\n";
1204 case QUOTA_NL_BSOFTWARN:
1205 msg = " block quota exceeded.\r\n";
1208 tty_write_message(tty, msg);
1213 static void prepare_warning(struct dquot_warn *warn, struct dquot *dquot,
1216 if (warning_issued(dquot, warntype))
1218 warn->w_type = warntype;
1219 warn->w_sb = dquot->dq_sb;
1220 warn->w_dq_id = dquot->dq_id;
1224 * Write warnings to the console and send warning messages over netlink.
1226 * Note that this function can call into tty and networking code.
1228 static void flush_warnings(struct dquot_warn *warn)
1232 for (i = 0; i < MAXQUOTAS; i++) {
1233 if (warn[i].w_type == QUOTA_NL_NOWARN)
1235 #ifdef CONFIG_PRINT_QUOTA_WARNING
1236 print_warning(&warn[i]);
1238 quota_send_warning(warn[i].w_dq_id,
1239 warn[i].w_sb->s_dev, warn[i].w_type);
1243 static int ignore_hardlimit(struct dquot *dquot)
1245 struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
1247 return capable(CAP_SYS_RESOURCE) &&
1248 (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1249 !(info->dqi_flags & DQF_ROOT_SQUASH));
1252 static int dquot_add_inodes(struct dquot *dquot, qsize_t inodes,
1253 struct dquot_warn *warn)
1258 spin_lock(&dquot->dq_dqb_lock);
1259 newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1260 if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type) ||
1261 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1264 if (dquot->dq_dqb.dqb_ihardlimit &&
1265 newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1266 !ignore_hardlimit(dquot)) {
1267 prepare_warning(warn, dquot, QUOTA_NL_IHARDWARN);
1272 if (dquot->dq_dqb.dqb_isoftlimit &&
1273 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1274 dquot->dq_dqb.dqb_itime &&
1275 ktime_get_real_seconds() >= dquot->dq_dqb.dqb_itime &&
1276 !ignore_hardlimit(dquot)) {
1277 prepare_warning(warn, dquot, QUOTA_NL_ISOFTLONGWARN);
1282 if (dquot->dq_dqb.dqb_isoftlimit &&
1283 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1284 dquot->dq_dqb.dqb_itime == 0) {
1285 prepare_warning(warn, dquot, QUOTA_NL_ISOFTWARN);
1286 dquot->dq_dqb.dqb_itime = ktime_get_real_seconds() +
1287 sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type].dqi_igrace;
1290 dquot->dq_dqb.dqb_curinodes = newinodes;
1293 spin_unlock(&dquot->dq_dqb_lock);
1297 static int dquot_add_space(struct dquot *dquot, qsize_t space,
1298 qsize_t rsv_space, unsigned int flags,
1299 struct dquot_warn *warn)
1302 struct super_block *sb = dquot->dq_sb;
1305 spin_lock(&dquot->dq_dqb_lock);
1306 if (!sb_has_quota_limits_enabled(sb, dquot->dq_id.type) ||
1307 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1310 tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1311 + space + rsv_space;
1313 if (dquot->dq_dqb.dqb_bhardlimit &&
1314 tspace > dquot->dq_dqb.dqb_bhardlimit &&
1315 !ignore_hardlimit(dquot)) {
1316 if (flags & DQUOT_SPACE_WARN)
1317 prepare_warning(warn, dquot, QUOTA_NL_BHARDWARN);
1322 if (dquot->dq_dqb.dqb_bsoftlimit &&
1323 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1324 dquot->dq_dqb.dqb_btime &&
1325 ktime_get_real_seconds() >= dquot->dq_dqb.dqb_btime &&
1326 !ignore_hardlimit(dquot)) {
1327 if (flags & DQUOT_SPACE_WARN)
1328 prepare_warning(warn, dquot, QUOTA_NL_BSOFTLONGWARN);
1333 if (dquot->dq_dqb.dqb_bsoftlimit &&
1334 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1335 dquot->dq_dqb.dqb_btime == 0) {
1336 if (flags & DQUOT_SPACE_WARN) {
1337 prepare_warning(warn, dquot, QUOTA_NL_BSOFTWARN);
1338 dquot->dq_dqb.dqb_btime = ktime_get_real_seconds() +
1339 sb_dqopt(sb)->info[dquot->dq_id.type].dqi_bgrace;
1342 * We don't allow preallocation to exceed softlimit so exceeding will
1351 * We have to be careful and go through warning generation & grace time
1352 * setting even if DQUOT_SPACE_NOFAIL is set. That's why we check it
1355 if (flags & DQUOT_SPACE_NOFAIL)
1358 dquot->dq_dqb.dqb_rsvspace += rsv_space;
1359 dquot->dq_dqb.dqb_curspace += space;
1361 spin_unlock(&dquot->dq_dqb_lock);
1365 static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1369 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1370 dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1371 !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type))
1372 return QUOTA_NL_NOWARN;
1374 newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1375 if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1376 return QUOTA_NL_ISOFTBELOW;
1377 if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1378 newinodes < dquot->dq_dqb.dqb_ihardlimit)
1379 return QUOTA_NL_IHARDBELOW;
1380 return QUOTA_NL_NOWARN;
1383 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1387 tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace;
1389 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1390 tspace <= dquot->dq_dqb.dqb_bsoftlimit)
1391 return QUOTA_NL_NOWARN;
1393 if (tspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1394 return QUOTA_NL_BSOFTBELOW;
1395 if (tspace >= dquot->dq_dqb.dqb_bhardlimit &&
1396 tspace - space < dquot->dq_dqb.dqb_bhardlimit)
1397 return QUOTA_NL_BHARDBELOW;
1398 return QUOTA_NL_NOWARN;
1401 static int dquot_active(const struct inode *inode)
1403 struct super_block *sb = inode->i_sb;
1405 if (IS_NOQUOTA(inode))
1407 return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1411 * Initialize quota pointers in inode
1413 * It is better to call this function outside of any transaction as it
1414 * might need a lot of space in journal for dquot structure allocation.
1416 static int __dquot_initialize(struct inode *inode, int type)
1418 int cnt, init_needed = 0;
1419 struct dquot **dquots, *got[MAXQUOTAS] = {};
1420 struct super_block *sb = inode->i_sb;
1424 if (!dquot_active(inode))
1427 dquots = i_dquot(inode);
1429 /* First get references to structures we might need. */
1430 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1434 struct dquot *dquot;
1436 if (type != -1 && cnt != type)
1439 * The i_dquot should have been initialized in most cases,
1440 * we check it without locking here to avoid unnecessary
1441 * dqget()/dqput() calls.
1446 if (!sb_has_quota_active(sb, cnt))
1453 qid = make_kqid_uid(inode->i_uid);
1456 qid = make_kqid_gid(inode->i_gid);
1459 rc = inode->i_sb->dq_op->get_projid(inode, &projid);
1462 qid = make_kqid_projid(projid);
1465 dquot = dqget(sb, qid);
1466 if (IS_ERR(dquot)) {
1467 /* We raced with somebody turning quotas off... */
1468 if (PTR_ERR(dquot) != -ESRCH) {
1469 ret = PTR_ERR(dquot);
1477 /* All required i_dquot has been initialized */
1481 spin_lock(&dq_data_lock);
1482 if (IS_NOQUOTA(inode))
1484 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1485 if (type != -1 && cnt != type)
1487 /* Avoid races with quotaoff() */
1488 if (!sb_has_quota_active(sb, cnt))
1490 /* We could race with quotaon or dqget() could have failed */
1494 dquots[cnt] = got[cnt];
1497 * Make quota reservation system happy if someone
1498 * did a write before quota was turned on
1500 rsv = inode_get_rsv_space(inode);
1501 if (unlikely(rsv)) {
1502 spin_lock(&inode->i_lock);
1503 /* Get reservation again under proper lock */
1504 rsv = __inode_get_rsv_space(inode);
1505 spin_lock(&dquots[cnt]->dq_dqb_lock);
1506 dquots[cnt]->dq_dqb.dqb_rsvspace += rsv;
1507 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1508 spin_unlock(&inode->i_lock);
1513 spin_unlock(&dq_data_lock);
1515 /* Drop unused references */
1521 int dquot_initialize(struct inode *inode)
1523 return __dquot_initialize(inode, -1);
1525 EXPORT_SYMBOL(dquot_initialize);
1527 bool dquot_initialize_needed(struct inode *inode)
1529 struct dquot **dquots;
1532 if (!dquot_active(inode))
1535 dquots = i_dquot(inode);
1536 for (i = 0; i < MAXQUOTAS; i++)
1537 if (!dquots[i] && sb_has_quota_active(inode->i_sb, i))
1541 EXPORT_SYMBOL(dquot_initialize_needed);
1544 * Release all quotas referenced by inode.
1546 * This function only be called on inode free or converting
1547 * a file to quota file, no other users for the i_dquot in
1548 * both cases, so we needn't call synchronize_srcu() after
1551 static void __dquot_drop(struct inode *inode)
1554 struct dquot **dquots = i_dquot(inode);
1555 struct dquot *put[MAXQUOTAS];
1557 spin_lock(&dq_data_lock);
1558 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1559 put[cnt] = dquots[cnt];
1562 spin_unlock(&dq_data_lock);
1566 void dquot_drop(struct inode *inode)
1568 struct dquot * const *dquots;
1571 if (IS_NOQUOTA(inode))
1575 * Test before calling to rule out calls from proc and such
1576 * where we are not allowed to block. Note that this is
1577 * actually reliable test even without the lock - the caller
1578 * must assure that nobody can come after the DQUOT_DROP and
1579 * add quota pointers back anyway.
1581 dquots = i_dquot(inode);
1582 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1587 if (cnt < MAXQUOTAS)
1588 __dquot_drop(inode);
1590 EXPORT_SYMBOL(dquot_drop);
1593 * inode_reserved_space is managed internally by quota, and protected by
1594 * i_lock similar to i_blocks+i_bytes.
1596 static qsize_t *inode_reserved_space(struct inode * inode)
1598 /* Filesystem must explicitly define it's own method in order to use
1599 * quota reservation interface */
1600 BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1601 return inode->i_sb->dq_op->get_reserved_space(inode);
1604 static qsize_t __inode_get_rsv_space(struct inode *inode)
1606 if (!inode->i_sb->dq_op->get_reserved_space)
1608 return *inode_reserved_space(inode);
1611 static qsize_t inode_get_rsv_space(struct inode *inode)
1615 if (!inode->i_sb->dq_op->get_reserved_space)
1617 spin_lock(&inode->i_lock);
1618 ret = __inode_get_rsv_space(inode);
1619 spin_unlock(&inode->i_lock);
1624 * This functions updates i_blocks+i_bytes fields and quota information
1625 * (together with appropriate checks).
1627 * NOTE: We absolutely rely on the fact that caller dirties the inode
1628 * (usually helpers in quotaops.h care about this) and holds a handle for
1629 * the current transaction so that dquot write and inode write go into the
1634 * This operation can block, but only after everything is updated
1636 int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
1638 int cnt, ret = 0, index;
1639 struct dquot_warn warn[MAXQUOTAS];
1640 int reserve = flags & DQUOT_SPACE_RESERVE;
1641 struct dquot **dquots;
1643 if (!dquot_active(inode)) {
1645 spin_lock(&inode->i_lock);
1646 *inode_reserved_space(inode) += number;
1647 spin_unlock(&inode->i_lock);
1649 inode_add_bytes(inode, number);
1654 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1655 warn[cnt].w_type = QUOTA_NL_NOWARN;
1657 dquots = i_dquot(inode);
1658 index = srcu_read_lock(&dquot_srcu);
1659 spin_lock(&inode->i_lock);
1660 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1663 if (flags & DQUOT_SPACE_RESERVE) {
1664 ret = dquot_add_space(dquots[cnt], 0, number, flags,
1667 ret = dquot_add_space(dquots[cnt], number, 0, flags,
1671 /* Back out changes we already did */
1672 for (cnt--; cnt >= 0; cnt--) {
1675 spin_lock(&dquots[cnt]->dq_dqb_lock);
1676 if (flags & DQUOT_SPACE_RESERVE) {
1677 dquots[cnt]->dq_dqb.dqb_rsvspace -=
1680 dquots[cnt]->dq_dqb.dqb_curspace -=
1683 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1685 spin_unlock(&inode->i_lock);
1686 goto out_flush_warn;
1690 *inode_reserved_space(inode) += number;
1692 __inode_add_bytes(inode, number);
1693 spin_unlock(&inode->i_lock);
1696 goto out_flush_warn;
1697 mark_all_dquot_dirty(dquots);
1699 srcu_read_unlock(&dquot_srcu, index);
1700 flush_warnings(warn);
1704 EXPORT_SYMBOL(__dquot_alloc_space);
1707 * This operation can block, but only after everything is updated
1709 int dquot_alloc_inode(struct inode *inode)
1711 int cnt, ret = 0, index;
1712 struct dquot_warn warn[MAXQUOTAS];
1713 struct dquot * const *dquots;
1715 if (!dquot_active(inode))
1717 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1718 warn[cnt].w_type = QUOTA_NL_NOWARN;
1720 dquots = i_dquot(inode);
1721 index = srcu_read_lock(&dquot_srcu);
1722 spin_lock(&inode->i_lock);
1723 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1726 ret = dquot_add_inodes(dquots[cnt], 1, &warn[cnt]);
1728 for (cnt--; cnt >= 0; cnt--) {
1731 /* Back out changes we already did */
1732 spin_lock(&dquots[cnt]->dq_dqb_lock);
1733 dquots[cnt]->dq_dqb.dqb_curinodes--;
1734 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1741 spin_unlock(&inode->i_lock);
1743 mark_all_dquot_dirty(dquots);
1744 srcu_read_unlock(&dquot_srcu, index);
1745 flush_warnings(warn);
1748 EXPORT_SYMBOL(dquot_alloc_inode);
1751 * Convert in-memory reserved quotas to real consumed quotas
1753 int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1755 struct dquot **dquots;
1758 if (!dquot_active(inode)) {
1759 spin_lock(&inode->i_lock);
1760 *inode_reserved_space(inode) -= number;
1761 __inode_add_bytes(inode, number);
1762 spin_unlock(&inode->i_lock);
1766 dquots = i_dquot(inode);
1767 index = srcu_read_lock(&dquot_srcu);
1768 spin_lock(&inode->i_lock);
1769 /* Claim reserved quotas to allocated quotas */
1770 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1772 struct dquot *dquot = dquots[cnt];
1774 spin_lock(&dquot->dq_dqb_lock);
1775 if (WARN_ON_ONCE(dquot->dq_dqb.dqb_rsvspace < number))
1776 number = dquot->dq_dqb.dqb_rsvspace;
1777 dquot->dq_dqb.dqb_curspace += number;
1778 dquot->dq_dqb.dqb_rsvspace -= number;
1779 spin_unlock(&dquot->dq_dqb_lock);
1782 /* Update inode bytes */
1783 *inode_reserved_space(inode) -= number;
1784 __inode_add_bytes(inode, number);
1785 spin_unlock(&inode->i_lock);
1786 mark_all_dquot_dirty(dquots);
1787 srcu_read_unlock(&dquot_srcu, index);
1790 EXPORT_SYMBOL(dquot_claim_space_nodirty);
1793 * Convert allocated space back to in-memory reserved quotas
1795 void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
1797 struct dquot **dquots;
1800 if (!dquot_active(inode)) {
1801 spin_lock(&inode->i_lock);
1802 *inode_reserved_space(inode) += number;
1803 __inode_sub_bytes(inode, number);
1804 spin_unlock(&inode->i_lock);
1808 dquots = i_dquot(inode);
1809 index = srcu_read_lock(&dquot_srcu);
1810 spin_lock(&inode->i_lock);
1811 /* Claim reserved quotas to allocated quotas */
1812 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1814 struct dquot *dquot = dquots[cnt];
1816 spin_lock(&dquot->dq_dqb_lock);
1817 if (WARN_ON_ONCE(dquot->dq_dqb.dqb_curspace < number))
1818 number = dquot->dq_dqb.dqb_curspace;
1819 dquot->dq_dqb.dqb_rsvspace += number;
1820 dquot->dq_dqb.dqb_curspace -= number;
1821 spin_unlock(&dquot->dq_dqb_lock);
1824 /* Update inode bytes */
1825 *inode_reserved_space(inode) += number;
1826 __inode_sub_bytes(inode, number);
1827 spin_unlock(&inode->i_lock);
1828 mark_all_dquot_dirty(dquots);
1829 srcu_read_unlock(&dquot_srcu, index);
1832 EXPORT_SYMBOL(dquot_reclaim_space_nodirty);
1835 * This operation can block, but only after everything is updated
1837 void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
1840 struct dquot_warn warn[MAXQUOTAS];
1841 struct dquot **dquots;
1842 int reserve = flags & DQUOT_SPACE_RESERVE, index;
1844 if (!dquot_active(inode)) {
1846 spin_lock(&inode->i_lock);
1847 *inode_reserved_space(inode) -= number;
1848 spin_unlock(&inode->i_lock);
1850 inode_sub_bytes(inode, number);
1855 dquots = i_dquot(inode);
1856 index = srcu_read_lock(&dquot_srcu);
1857 spin_lock(&inode->i_lock);
1858 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1861 warn[cnt].w_type = QUOTA_NL_NOWARN;
1864 spin_lock(&dquots[cnt]->dq_dqb_lock);
1865 wtype = info_bdq_free(dquots[cnt], number);
1866 if (wtype != QUOTA_NL_NOWARN)
1867 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1869 dquot_free_reserved_space(dquots[cnt], number);
1871 dquot_decr_space(dquots[cnt], number);
1872 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1875 *inode_reserved_space(inode) -= number;
1877 __inode_sub_bytes(inode, number);
1878 spin_unlock(&inode->i_lock);
1882 mark_all_dquot_dirty(dquots);
1884 srcu_read_unlock(&dquot_srcu, index);
1885 flush_warnings(warn);
1887 EXPORT_SYMBOL(__dquot_free_space);
1890 * This operation can block, but only after everything is updated
1892 void dquot_free_inode(struct inode *inode)
1895 struct dquot_warn warn[MAXQUOTAS];
1896 struct dquot * const *dquots;
1899 if (!dquot_active(inode))
1902 dquots = i_dquot(inode);
1903 index = srcu_read_lock(&dquot_srcu);
1904 spin_lock(&inode->i_lock);
1905 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1908 warn[cnt].w_type = QUOTA_NL_NOWARN;
1911 spin_lock(&dquots[cnt]->dq_dqb_lock);
1912 wtype = info_idq_free(dquots[cnt], 1);
1913 if (wtype != QUOTA_NL_NOWARN)
1914 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1915 dquot_decr_inodes(dquots[cnt], 1);
1916 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1918 spin_unlock(&inode->i_lock);
1919 mark_all_dquot_dirty(dquots);
1920 srcu_read_unlock(&dquot_srcu, index);
1921 flush_warnings(warn);
1923 EXPORT_SYMBOL(dquot_free_inode);
1926 * Transfer the number of inode and blocks from one diskquota to an other.
1927 * On success, dquot references in transfer_to are consumed and references
1928 * to original dquots that need to be released are placed there. On failure,
1929 * references are kept untouched.
1931 * This operation can block, but only after everything is updated
1932 * A transaction must be started when entering this function.
1934 * We are holding reference on transfer_from & transfer_to, no need to
1935 * protect them by srcu_read_lock().
1937 int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
1940 qsize_t rsv_space = 0;
1941 qsize_t inode_usage = 1;
1942 struct dquot *transfer_from[MAXQUOTAS] = {};
1944 char is_valid[MAXQUOTAS] = {};
1945 struct dquot_warn warn_to[MAXQUOTAS];
1946 struct dquot_warn warn_from_inodes[MAXQUOTAS];
1947 struct dquot_warn warn_from_space[MAXQUOTAS];
1949 if (IS_NOQUOTA(inode))
1952 if (inode->i_sb->dq_op->get_inode_usage) {
1953 ret = inode->i_sb->dq_op->get_inode_usage(inode, &inode_usage);
1958 /* Initialize the arrays */
1959 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1960 warn_to[cnt].w_type = QUOTA_NL_NOWARN;
1961 warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
1962 warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
1965 spin_lock(&dq_data_lock);
1966 spin_lock(&inode->i_lock);
1967 if (IS_NOQUOTA(inode)) { /* File without quota accounting? */
1968 spin_unlock(&inode->i_lock);
1969 spin_unlock(&dq_data_lock);
1972 cur_space = __inode_get_bytes(inode);
1973 rsv_space = __inode_get_rsv_space(inode);
1975 * Build the transfer_from list, check limits, and update usage in
1976 * the target structures.
1978 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1980 * Skip changes for same uid or gid or for turned off quota-type.
1982 if (!transfer_to[cnt])
1984 /* Avoid races with quotaoff() */
1985 if (!sb_has_quota_active(inode->i_sb, cnt))
1988 transfer_from[cnt] = i_dquot(inode)[cnt];
1989 ret = dquot_add_inodes(transfer_to[cnt], inode_usage,
1993 ret = dquot_add_space(transfer_to[cnt], cur_space, rsv_space, 0,
1996 spin_lock(&transfer_to[cnt]->dq_dqb_lock);
1997 dquot_decr_inodes(transfer_to[cnt], inode_usage);
1998 spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2003 /* Decrease usage for source structures and update quota pointers */
2004 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2007 /* Due to IO error we might not have transfer_from[] structure */
2008 if (transfer_from[cnt]) {
2011 spin_lock(&transfer_from[cnt]->dq_dqb_lock);
2012 wtype = info_idq_free(transfer_from[cnt], inode_usage);
2013 if (wtype != QUOTA_NL_NOWARN)
2014 prepare_warning(&warn_from_inodes[cnt],
2015 transfer_from[cnt], wtype);
2016 wtype = info_bdq_free(transfer_from[cnt],
2017 cur_space + rsv_space);
2018 if (wtype != QUOTA_NL_NOWARN)
2019 prepare_warning(&warn_from_space[cnt],
2020 transfer_from[cnt], wtype);
2021 dquot_decr_inodes(transfer_from[cnt], inode_usage);
2022 dquot_decr_space(transfer_from[cnt], cur_space);
2023 dquot_free_reserved_space(transfer_from[cnt],
2025 spin_unlock(&transfer_from[cnt]->dq_dqb_lock);
2027 i_dquot(inode)[cnt] = transfer_to[cnt];
2029 spin_unlock(&inode->i_lock);
2030 spin_unlock(&dq_data_lock);
2032 mark_all_dquot_dirty(transfer_from);
2033 mark_all_dquot_dirty(transfer_to);
2034 flush_warnings(warn_to);
2035 flush_warnings(warn_from_inodes);
2036 flush_warnings(warn_from_space);
2037 /* Pass back references to put */
2038 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2040 transfer_to[cnt] = transfer_from[cnt];
2043 /* Back out changes we already did */
2044 for (cnt--; cnt >= 0; cnt--) {
2047 spin_lock(&transfer_to[cnt]->dq_dqb_lock);
2048 dquot_decr_inodes(transfer_to[cnt], inode_usage);
2049 dquot_decr_space(transfer_to[cnt], cur_space);
2050 dquot_free_reserved_space(transfer_to[cnt], rsv_space);
2051 spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2053 spin_unlock(&inode->i_lock);
2054 spin_unlock(&dq_data_lock);
2055 flush_warnings(warn_to);
2058 EXPORT_SYMBOL(__dquot_transfer);
2060 /* Wrapper for transferring ownership of an inode for uid/gid only
2061 * Called from FSXXX_setattr()
2063 int dquot_transfer(struct inode *inode, struct iattr *iattr)
2065 struct dquot *transfer_to[MAXQUOTAS] = {};
2066 struct dquot *dquot;
2067 struct super_block *sb = inode->i_sb;
2070 if (!dquot_active(inode))
2073 if (iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)){
2074 dquot = dqget(sb, make_kqid_uid(iattr->ia_uid));
2075 if (IS_ERR(dquot)) {
2076 if (PTR_ERR(dquot) != -ESRCH) {
2077 ret = PTR_ERR(dquot);
2082 transfer_to[USRQUOTA] = dquot;
2084 if (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid)){
2085 dquot = dqget(sb, make_kqid_gid(iattr->ia_gid));
2086 if (IS_ERR(dquot)) {
2087 if (PTR_ERR(dquot) != -ESRCH) {
2088 ret = PTR_ERR(dquot);
2093 transfer_to[GRPQUOTA] = dquot;
2095 ret = __dquot_transfer(inode, transfer_to);
2097 dqput_all(transfer_to);
2100 EXPORT_SYMBOL(dquot_transfer);
2103 * Write info of quota file to disk
2105 int dquot_commit_info(struct super_block *sb, int type)
2107 struct quota_info *dqopt = sb_dqopt(sb);
2109 return dqopt->ops[type]->write_file_info(sb, type);
2111 EXPORT_SYMBOL(dquot_commit_info);
2113 int dquot_get_next_id(struct super_block *sb, struct kqid *qid)
2115 struct quota_info *dqopt = sb_dqopt(sb);
2117 if (!sb_has_quota_active(sb, qid->type))
2119 if (!dqopt->ops[qid->type]->get_next_id)
2121 return dqopt->ops[qid->type]->get_next_id(sb, qid);
2123 EXPORT_SYMBOL(dquot_get_next_id);
2126 * Definitions of diskquota operations.
2128 const struct dquot_operations dquot_operations = {
2129 .write_dquot = dquot_commit,
2130 .acquire_dquot = dquot_acquire,
2131 .release_dquot = dquot_release,
2132 .mark_dirty = dquot_mark_dquot_dirty,
2133 .write_info = dquot_commit_info,
2134 .alloc_dquot = dquot_alloc,
2135 .destroy_dquot = dquot_destroy,
2136 .get_next_id = dquot_get_next_id,
2138 EXPORT_SYMBOL(dquot_operations);
2141 * Generic helper for ->open on filesystems supporting disk quotas.
2143 int dquot_file_open(struct inode *inode, struct file *file)
2147 error = generic_file_open(inode, file);
2148 if (!error && (file->f_mode & FMODE_WRITE))
2149 error = dquot_initialize(inode);
2152 EXPORT_SYMBOL(dquot_file_open);
2155 * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
2157 int dquot_disable(struct super_block *sb, int type, unsigned int flags)
2160 struct quota_info *dqopt = sb_dqopt(sb);
2161 struct inode *toputinode[MAXQUOTAS];
2163 /* s_umount should be held in exclusive mode */
2164 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2165 up_read(&sb->s_umount);
2167 /* Cannot turn off usage accounting without turning off limits, or
2168 * suspend quotas and simultaneously turn quotas off. */
2169 if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
2170 || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
2171 DQUOT_USAGE_ENABLED)))
2175 * Skip everything if there's nothing to do. We have to do this because
2176 * sometimes we are called when fill_super() failed and calling
2177 * sync_fs() in such cases does no good.
2179 if (!sb_any_quota_loaded(sb))
2182 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2183 toputinode[cnt] = NULL;
2184 if (type != -1 && cnt != type)
2186 if (!sb_has_quota_loaded(sb, cnt))
2189 if (flags & DQUOT_SUSPENDED) {
2190 spin_lock(&dq_state_lock);
2192 dquot_state_flag(DQUOT_SUSPENDED, cnt);
2193 spin_unlock(&dq_state_lock);
2195 spin_lock(&dq_state_lock);
2196 dqopt->flags &= ~dquot_state_flag(flags, cnt);
2197 /* Turning off suspended quotas? */
2198 if (!sb_has_quota_loaded(sb, cnt) &&
2199 sb_has_quota_suspended(sb, cnt)) {
2200 dqopt->flags &= ~dquot_state_flag(
2201 DQUOT_SUSPENDED, cnt);
2202 spin_unlock(&dq_state_lock);
2203 iput(dqopt->files[cnt]);
2204 dqopt->files[cnt] = NULL;
2207 spin_unlock(&dq_state_lock);
2210 /* We still have to keep quota loaded? */
2211 if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
2214 /* Note: these are blocking operations */
2215 drop_dquot_ref(sb, cnt);
2216 invalidate_dquots(sb, cnt);
2218 * Now all dquots should be invalidated, all writes done so we
2219 * should be only users of the info. No locks needed.
2221 if (info_dirty(&dqopt->info[cnt]))
2222 sb->dq_op->write_info(sb, cnt);
2223 if (dqopt->ops[cnt]->free_file_info)
2224 dqopt->ops[cnt]->free_file_info(sb, cnt);
2225 put_quota_format(dqopt->info[cnt].dqi_format);
2227 toputinode[cnt] = dqopt->files[cnt];
2228 if (!sb_has_quota_loaded(sb, cnt))
2229 dqopt->files[cnt] = NULL;
2230 dqopt->info[cnt].dqi_flags = 0;
2231 dqopt->info[cnt].dqi_igrace = 0;
2232 dqopt->info[cnt].dqi_bgrace = 0;
2233 dqopt->ops[cnt] = NULL;
2236 /* Skip syncing and setting flags if quota files are hidden */
2237 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2240 /* Sync the superblock so that buffers with quota data are written to
2241 * disk (and so userspace sees correct data afterwards). */
2242 if (sb->s_op->sync_fs)
2243 sb->s_op->sync_fs(sb, 1);
2244 sync_blockdev(sb->s_bdev);
2245 /* Now the quota files are just ordinary files and we can set the
2246 * inode flags back. Moreover we discard the pagecache so that
2247 * userspace sees the writes we did bypassing the pagecache. We
2248 * must also discard the blockdev buffers so that we see the
2249 * changes done by userspace on the next quotaon() */
2250 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2251 /* This can happen when suspending quotas on remount-ro... */
2252 if (toputinode[cnt] && !sb_has_quota_loaded(sb, cnt)) {
2253 inode_lock(toputinode[cnt]);
2254 toputinode[cnt]->i_flags &= ~S_NOQUOTA;
2255 truncate_inode_pages(&toputinode[cnt]->i_data, 0);
2256 inode_unlock(toputinode[cnt]);
2257 mark_inode_dirty_sync(toputinode[cnt]);
2260 invalidate_bdev(sb->s_bdev);
2262 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2263 if (toputinode[cnt]) {
2264 /* On remount RO, we keep the inode pointer so that we
2265 * can reenable quota on the subsequent remount RW. We
2266 * have to check 'flags' variable and not use sb_has_
2267 * function because another quotaon / quotaoff could
2268 * change global state before we got here. We refuse
2269 * to suspend quotas when there is pending delete on
2270 * the quota file... */
2271 if (!(flags & DQUOT_SUSPENDED))
2272 iput(toputinode[cnt]);
2273 else if (!toputinode[cnt]->i_nlink)
2278 EXPORT_SYMBOL(dquot_disable);
2280 int dquot_quota_off(struct super_block *sb, int type)
2282 return dquot_disable(sb, type,
2283 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2285 EXPORT_SYMBOL(dquot_quota_off);
2288 * Turn quotas on on a device
2292 * Helper function to turn quotas on when we already have the inode of
2293 * quota file and no quota information is loaded.
2295 static int vfs_load_quota_inode(struct inode *inode, int type, int format_id,
2298 struct quota_format_type *fmt = find_quota_format(format_id);
2299 struct super_block *sb = inode->i_sb;
2300 struct quota_info *dqopt = sb_dqopt(sb);
2305 if (!S_ISREG(inode->i_mode)) {
2309 if (IS_RDONLY(inode)) {
2313 if (!sb->s_op->quota_write || !sb->s_op->quota_read ||
2314 (type == PRJQUOTA && sb->dq_op->get_projid == NULL)) {
2318 /* Filesystems outside of init_user_ns not yet supported */
2319 if (sb->s_user_ns != &init_user_ns) {
2323 /* Usage always has to be set... */
2324 if (!(flags & DQUOT_USAGE_ENABLED)) {
2328 if (sb_has_quota_loaded(sb, type)) {
2333 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2334 /* As we bypass the pagecache we must now flush all the
2335 * dirty data and invalidate caches so that kernel sees
2336 * changes from userspace. It is not enough to just flush
2337 * the quota file since if blocksize < pagesize, invalidation
2338 * of the cache could fail because of other unrelated dirty
2340 sync_filesystem(sb);
2341 invalidate_bdev(sb->s_bdev);
2344 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2345 /* We don't want quota and atime on quota files (deadlocks
2346 * possible) Also nobody should write to the file - we use
2347 * special IO operations which ignore the immutable bit. */
2349 inode->i_flags |= S_NOQUOTA;
2350 inode_unlock(inode);
2352 * When S_NOQUOTA is set, remove dquot references as no more
2353 * references can be added
2355 __dquot_drop(inode);
2359 dqopt->files[type] = igrab(inode);
2360 if (!dqopt->files[type])
2361 goto out_file_flags;
2363 if (!fmt->qf_ops->check_quota_file(sb, type))
2366 dqopt->ops[type] = fmt->qf_ops;
2367 dqopt->info[type].dqi_format = fmt;
2368 dqopt->info[type].dqi_fmt_id = format_id;
2369 INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2370 error = dqopt->ops[type]->read_file_info(sb, type);
2373 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE) {
2374 spin_lock(&dq_data_lock);
2375 dqopt->info[type].dqi_flags |= DQF_SYS_FILE;
2376 spin_unlock(&dq_data_lock);
2378 spin_lock(&dq_state_lock);
2379 dqopt->flags |= dquot_state_flag(flags, type);
2380 spin_unlock(&dq_state_lock);
2382 add_dquot_ref(sb, type);
2387 dqopt->files[type] = NULL;
2391 inode->i_flags &= ~S_NOQUOTA;
2392 inode_unlock(inode);
2394 put_quota_format(fmt);
2399 /* Reenable quotas on remount RW */
2400 int dquot_resume(struct super_block *sb, int type)
2402 struct quota_info *dqopt = sb_dqopt(sb);
2403 struct inode *inode;
2407 /* s_umount should be held in exclusive mode */
2408 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2409 up_read(&sb->s_umount);
2411 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2412 if (type != -1 && cnt != type)
2414 if (!sb_has_quota_suspended(sb, cnt))
2417 inode = dqopt->files[cnt];
2418 dqopt->files[cnt] = NULL;
2419 spin_lock(&dq_state_lock);
2420 flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2421 DQUOT_LIMITS_ENABLED,
2423 dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2424 spin_unlock(&dq_state_lock);
2426 flags = dquot_generic_flag(flags, cnt);
2427 ret = vfs_load_quota_inode(inode, cnt,
2428 dqopt->info[cnt].dqi_fmt_id, flags);
2434 EXPORT_SYMBOL(dquot_resume);
2436 int dquot_quota_on(struct super_block *sb, int type, int format_id,
2437 const struct path *path)
2439 int error = security_quota_on(path->dentry);
2442 /* Quota file not on the same filesystem? */
2443 if (path->dentry->d_sb != sb)
2446 error = vfs_load_quota_inode(d_inode(path->dentry), type,
2447 format_id, DQUOT_USAGE_ENABLED |
2448 DQUOT_LIMITS_ENABLED);
2451 EXPORT_SYMBOL(dquot_quota_on);
2454 * More powerful function for turning on quotas allowing setting
2455 * of individual quota flags
2457 int dquot_enable(struct inode *inode, int type, int format_id,
2460 struct super_block *sb = inode->i_sb;
2462 /* Just unsuspend quotas? */
2463 BUG_ON(flags & DQUOT_SUSPENDED);
2464 /* s_umount should be held in exclusive mode */
2465 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2466 up_read(&sb->s_umount);
2470 /* Just updating flags needed? */
2471 if (sb_has_quota_loaded(sb, type)) {
2472 if (flags & DQUOT_USAGE_ENABLED &&
2473 sb_has_quota_usage_enabled(sb, type))
2475 if (flags & DQUOT_LIMITS_ENABLED &&
2476 sb_has_quota_limits_enabled(sb, type))
2478 spin_lock(&dq_state_lock);
2479 sb_dqopt(sb)->flags |= dquot_state_flag(flags, type);
2480 spin_unlock(&dq_state_lock);
2484 return vfs_load_quota_inode(inode, type, format_id, flags);
2486 EXPORT_SYMBOL(dquot_enable);
2489 * This function is used when filesystem needs to initialize quotas
2490 * during mount time.
2492 int dquot_quota_on_mount(struct super_block *sb, char *qf_name,
2493 int format_id, int type)
2495 struct dentry *dentry;
2498 dentry = lookup_one_len_unlocked(qf_name, sb->s_root, strlen(qf_name));
2500 return PTR_ERR(dentry);
2502 if (d_really_is_negative(dentry)) {
2507 error = security_quota_on(dentry);
2509 error = vfs_load_quota_inode(d_inode(dentry), type, format_id,
2510 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2516 EXPORT_SYMBOL(dquot_quota_on_mount);
2518 static int dquot_quota_enable(struct super_block *sb, unsigned int flags)
2522 struct quota_info *dqopt = sb_dqopt(sb);
2524 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2526 /* Accounting cannot be turned on while fs is mounted */
2527 flags &= ~(FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT);
2530 for (type = 0; type < MAXQUOTAS; type++) {
2531 if (!(flags & qtype_enforce_flag(type)))
2533 /* Can't enforce without accounting */
2534 if (!sb_has_quota_usage_enabled(sb, type))
2536 ret = dquot_enable(dqopt->files[type], type,
2537 dqopt->info[type].dqi_fmt_id,
2538 DQUOT_LIMITS_ENABLED);
2544 /* Backout enforcement enablement we already did */
2545 for (type--; type >= 0; type--) {
2546 if (flags & qtype_enforce_flag(type))
2547 dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2549 /* Error code translation for better compatibility with XFS */
2555 static int dquot_quota_disable(struct super_block *sb, unsigned int flags)
2559 struct quota_info *dqopt = sb_dqopt(sb);
2561 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2564 * We don't support turning off accounting via quotactl. In principle
2565 * quota infrastructure can do this but filesystems don't expect
2566 * userspace to be able to do it.
2569 (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT))
2572 /* Filter out limits not enabled */
2573 for (type = 0; type < MAXQUOTAS; type++)
2574 if (!sb_has_quota_limits_enabled(sb, type))
2575 flags &= ~qtype_enforce_flag(type);
2579 for (type = 0; type < MAXQUOTAS; type++) {
2580 if (flags & qtype_enforce_flag(type)) {
2581 ret = dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2588 /* Backout enforcement disabling we already did */
2589 for (type--; type >= 0; type--) {
2590 if (flags & qtype_enforce_flag(type))
2591 dquot_enable(dqopt->files[type], type,
2592 dqopt->info[type].dqi_fmt_id,
2593 DQUOT_LIMITS_ENABLED);
2598 /* Generic routine for getting common part of quota structure */
2599 static void do_get_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2601 struct mem_dqblk *dm = &dquot->dq_dqb;
2603 memset(di, 0, sizeof(*di));
2604 spin_lock(&dquot->dq_dqb_lock);
2605 di->d_spc_hardlimit = dm->dqb_bhardlimit;
2606 di->d_spc_softlimit = dm->dqb_bsoftlimit;
2607 di->d_ino_hardlimit = dm->dqb_ihardlimit;
2608 di->d_ino_softlimit = dm->dqb_isoftlimit;
2609 di->d_space = dm->dqb_curspace + dm->dqb_rsvspace;
2610 di->d_ino_count = dm->dqb_curinodes;
2611 di->d_spc_timer = dm->dqb_btime;
2612 di->d_ino_timer = dm->dqb_itime;
2613 spin_unlock(&dquot->dq_dqb_lock);
2616 int dquot_get_dqblk(struct super_block *sb, struct kqid qid,
2617 struct qc_dqblk *di)
2619 struct dquot *dquot;
2621 dquot = dqget(sb, qid);
2623 return PTR_ERR(dquot);
2624 do_get_dqblk(dquot, di);
2629 EXPORT_SYMBOL(dquot_get_dqblk);
2631 int dquot_get_next_dqblk(struct super_block *sb, struct kqid *qid,
2632 struct qc_dqblk *di)
2634 struct dquot *dquot;
2637 if (!sb->dq_op->get_next_id)
2639 err = sb->dq_op->get_next_id(sb, qid);
2642 dquot = dqget(sb, *qid);
2644 return PTR_ERR(dquot);
2645 do_get_dqblk(dquot, di);
2650 EXPORT_SYMBOL(dquot_get_next_dqblk);
2652 #define VFS_QC_MASK \
2653 (QC_SPACE | QC_SPC_SOFT | QC_SPC_HARD | \
2654 QC_INO_COUNT | QC_INO_SOFT | QC_INO_HARD | \
2655 QC_SPC_TIMER | QC_INO_TIMER)
2657 /* Generic routine for setting common part of quota structure */
2658 static int do_set_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2660 struct mem_dqblk *dm = &dquot->dq_dqb;
2661 int check_blim = 0, check_ilim = 0;
2662 struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
2664 if (di->d_fieldmask & ~VFS_QC_MASK)
2667 if (((di->d_fieldmask & QC_SPC_SOFT) &&
2668 di->d_spc_softlimit > dqi->dqi_max_spc_limit) ||
2669 ((di->d_fieldmask & QC_SPC_HARD) &&
2670 di->d_spc_hardlimit > dqi->dqi_max_spc_limit) ||
2671 ((di->d_fieldmask & QC_INO_SOFT) &&
2672 (di->d_ino_softlimit > dqi->dqi_max_ino_limit)) ||
2673 ((di->d_fieldmask & QC_INO_HARD) &&
2674 (di->d_ino_hardlimit > dqi->dqi_max_ino_limit)))
2677 spin_lock(&dquot->dq_dqb_lock);
2678 if (di->d_fieldmask & QC_SPACE) {
2679 dm->dqb_curspace = di->d_space - dm->dqb_rsvspace;
2681 set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2684 if (di->d_fieldmask & QC_SPC_SOFT)
2685 dm->dqb_bsoftlimit = di->d_spc_softlimit;
2686 if (di->d_fieldmask & QC_SPC_HARD)
2687 dm->dqb_bhardlimit = di->d_spc_hardlimit;
2688 if (di->d_fieldmask & (QC_SPC_SOFT | QC_SPC_HARD)) {
2690 set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2693 if (di->d_fieldmask & QC_INO_COUNT) {
2694 dm->dqb_curinodes = di->d_ino_count;
2696 set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2699 if (di->d_fieldmask & QC_INO_SOFT)
2700 dm->dqb_isoftlimit = di->d_ino_softlimit;
2701 if (di->d_fieldmask & QC_INO_HARD)
2702 dm->dqb_ihardlimit = di->d_ino_hardlimit;
2703 if (di->d_fieldmask & (QC_INO_SOFT | QC_INO_HARD)) {
2705 set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2708 if (di->d_fieldmask & QC_SPC_TIMER) {
2709 dm->dqb_btime = di->d_spc_timer;
2711 set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2714 if (di->d_fieldmask & QC_INO_TIMER) {
2715 dm->dqb_itime = di->d_ino_timer;
2717 set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2721 if (!dm->dqb_bsoftlimit ||
2722 dm->dqb_curspace + dm->dqb_rsvspace < dm->dqb_bsoftlimit) {
2724 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2725 } else if (!(di->d_fieldmask & QC_SPC_TIMER))
2726 /* Set grace only if user hasn't provided his own... */
2727 dm->dqb_btime = ktime_get_real_seconds() + dqi->dqi_bgrace;
2730 if (!dm->dqb_isoftlimit ||
2731 dm->dqb_curinodes < dm->dqb_isoftlimit) {
2733 clear_bit(DQ_INODES_B, &dquot->dq_flags);
2734 } else if (!(di->d_fieldmask & QC_INO_TIMER))
2735 /* Set grace only if user hasn't provided his own... */
2736 dm->dqb_itime = ktime_get_real_seconds() + dqi->dqi_igrace;
2738 if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2740 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2742 set_bit(DQ_FAKE_B, &dquot->dq_flags);
2743 spin_unlock(&dquot->dq_dqb_lock);
2744 mark_dquot_dirty(dquot);
2749 int dquot_set_dqblk(struct super_block *sb, struct kqid qid,
2750 struct qc_dqblk *di)
2752 struct dquot *dquot;
2755 dquot = dqget(sb, qid);
2756 if (IS_ERR(dquot)) {
2757 rc = PTR_ERR(dquot);
2760 rc = do_set_dqblk(dquot, di);
2765 EXPORT_SYMBOL(dquot_set_dqblk);
2767 /* Generic routine for getting common part of quota file information */
2768 int dquot_get_state(struct super_block *sb, struct qc_state *state)
2770 struct mem_dqinfo *mi;
2771 struct qc_type_state *tstate;
2772 struct quota_info *dqopt = sb_dqopt(sb);
2775 memset(state, 0, sizeof(*state));
2776 for (type = 0; type < MAXQUOTAS; type++) {
2777 if (!sb_has_quota_active(sb, type))
2779 tstate = state->s_state + type;
2780 mi = sb_dqopt(sb)->info + type;
2781 tstate->flags = QCI_ACCT_ENABLED;
2782 spin_lock(&dq_data_lock);
2783 if (mi->dqi_flags & DQF_SYS_FILE)
2784 tstate->flags |= QCI_SYSFILE;
2785 if (mi->dqi_flags & DQF_ROOT_SQUASH)
2786 tstate->flags |= QCI_ROOT_SQUASH;
2787 if (sb_has_quota_limits_enabled(sb, type))
2788 tstate->flags |= QCI_LIMITS_ENFORCED;
2789 tstate->spc_timelimit = mi->dqi_bgrace;
2790 tstate->ino_timelimit = mi->dqi_igrace;
2791 tstate->ino = dqopt->files[type]->i_ino;
2792 tstate->blocks = dqopt->files[type]->i_blocks;
2793 tstate->nextents = 1; /* We don't know... */
2794 spin_unlock(&dq_data_lock);
2798 EXPORT_SYMBOL(dquot_get_state);
2800 /* Generic routine for setting common part of quota file information */
2801 int dquot_set_dqinfo(struct super_block *sb, int type, struct qc_info *ii)
2803 struct mem_dqinfo *mi;
2806 if ((ii->i_fieldmask & QC_WARNS_MASK) ||
2807 (ii->i_fieldmask & QC_RT_SPC_TIMER))
2809 if (!sb_has_quota_active(sb, type))
2811 mi = sb_dqopt(sb)->info + type;
2812 if (ii->i_fieldmask & QC_FLAGS) {
2813 if ((ii->i_flags & QCI_ROOT_SQUASH &&
2814 mi->dqi_format->qf_fmt_id != QFMT_VFS_OLD))
2817 spin_lock(&dq_data_lock);
2818 if (ii->i_fieldmask & QC_SPC_TIMER)
2819 mi->dqi_bgrace = ii->i_spc_timelimit;
2820 if (ii->i_fieldmask & QC_INO_TIMER)
2821 mi->dqi_igrace = ii->i_ino_timelimit;
2822 if (ii->i_fieldmask & QC_FLAGS) {
2823 if (ii->i_flags & QCI_ROOT_SQUASH)
2824 mi->dqi_flags |= DQF_ROOT_SQUASH;
2826 mi->dqi_flags &= ~DQF_ROOT_SQUASH;
2828 spin_unlock(&dq_data_lock);
2829 mark_info_dirty(sb, type);
2830 /* Force write to disk */
2831 sb->dq_op->write_info(sb, type);
2834 EXPORT_SYMBOL(dquot_set_dqinfo);
2836 const struct quotactl_ops dquot_quotactl_sysfile_ops = {
2837 .quota_enable = dquot_quota_enable,
2838 .quota_disable = dquot_quota_disable,
2839 .quota_sync = dquot_quota_sync,
2840 .get_state = dquot_get_state,
2841 .set_info = dquot_set_dqinfo,
2842 .get_dqblk = dquot_get_dqblk,
2843 .get_nextdqblk = dquot_get_next_dqblk,
2844 .set_dqblk = dquot_set_dqblk
2846 EXPORT_SYMBOL(dquot_quotactl_sysfile_ops);
2848 static int do_proc_dqstats(struct ctl_table *table, int write,
2849 void __user *buffer, size_t *lenp, loff_t *ppos)
2851 unsigned int type = (int *)table->data - dqstats.stat;
2853 /* Update global table */
2854 dqstats.stat[type] =
2855 percpu_counter_sum_positive(&dqstats.counter[type]);
2856 return proc_dointvec(table, write, buffer, lenp, ppos);
2859 static struct ctl_table fs_dqstats_table[] = {
2861 .procname = "lookups",
2862 .data = &dqstats.stat[DQST_LOOKUPS],
2863 .maxlen = sizeof(int),
2865 .proc_handler = do_proc_dqstats,
2868 .procname = "drops",
2869 .data = &dqstats.stat[DQST_DROPS],
2870 .maxlen = sizeof(int),
2872 .proc_handler = do_proc_dqstats,
2875 .procname = "reads",
2876 .data = &dqstats.stat[DQST_READS],
2877 .maxlen = sizeof(int),
2879 .proc_handler = do_proc_dqstats,
2882 .procname = "writes",
2883 .data = &dqstats.stat[DQST_WRITES],
2884 .maxlen = sizeof(int),
2886 .proc_handler = do_proc_dqstats,
2889 .procname = "cache_hits",
2890 .data = &dqstats.stat[DQST_CACHE_HITS],
2891 .maxlen = sizeof(int),
2893 .proc_handler = do_proc_dqstats,
2896 .procname = "allocated_dquots",
2897 .data = &dqstats.stat[DQST_ALLOC_DQUOTS],
2898 .maxlen = sizeof(int),
2900 .proc_handler = do_proc_dqstats,
2903 .procname = "free_dquots",
2904 .data = &dqstats.stat[DQST_FREE_DQUOTS],
2905 .maxlen = sizeof(int),
2907 .proc_handler = do_proc_dqstats,
2910 .procname = "syncs",
2911 .data = &dqstats.stat[DQST_SYNCS],
2912 .maxlen = sizeof(int),
2914 .proc_handler = do_proc_dqstats,
2916 #ifdef CONFIG_PRINT_QUOTA_WARNING
2918 .procname = "warnings",
2919 .data = &flag_print_warnings,
2920 .maxlen = sizeof(int),
2922 .proc_handler = proc_dointvec,
2928 static struct ctl_table fs_table[] = {
2930 .procname = "quota",
2932 .child = fs_dqstats_table,
2937 static struct ctl_table sys_table[] = {
2946 static int __init dquot_init(void)
2949 unsigned long nr_hash, order;
2951 printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2953 register_sysctl_table(sys_table);
2955 dquot_cachep = kmem_cache_create("dquot",
2956 sizeof(struct dquot), sizeof(unsigned long) * 4,
2957 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2958 SLAB_MEM_SPREAD|SLAB_PANIC),
2962 dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
2964 panic("Cannot create dquot hash table");
2966 for (i = 0; i < _DQST_DQSTAT_LAST; i++) {
2967 ret = percpu_counter_init(&dqstats.counter[i], 0, GFP_KERNEL);
2969 panic("Cannot create dquot stat counters");
2972 /* Find power-of-two hlist_heads which can fit into allocation */
2973 nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2977 } while (nr_hash >> dq_hash_bits);
2980 nr_hash = 1UL << dq_hash_bits;
2981 dq_hash_mask = nr_hash - 1;
2982 for (i = 0; i < nr_hash; i++)
2983 INIT_HLIST_HEAD(dquot_hash + i);
2985 pr_info("VFS: Dquot-cache hash table entries: %ld (order %ld,"
2986 " %ld bytes)\n", nr_hash, order, (PAGE_SIZE << order));
2988 register_shrinker(&dqcache_shrinker);
2992 fs_initcall(dquot_init);