x86, ioapic: Simplify probe_nr_irqs_gsi.
[linux-2.6/kvm.git] / fs / quota / dquot.c
blob788b5802a7ce5dc043066d9fa20278aa0e9ef2bf
1 /*
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.
12 * Author: Marco van Wieringen <mvw@planets.elm.net>
14 * Fixes: Dmitry Gorodchanin <pgmdsg@ibi.com>, 11 Feb 96
16 * Revised list management to avoid races
17 * -- Bill Hawes, <whawes@star.net>, 9/98
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
26 * quota files
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.
30 * Jan Kara, <jack@suse.cz>, sponsored by SuSE CR, 10-11/99
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
38 * Jan Kara, <jack@suse.cz>, 9/2000
40 * Added dynamic quota structure allocation
41 * Jan Kara <jack@suse.cz> 12/2000
43 * Rewritten quota interface. Implemented new quota format and
44 * formats registering.
45 * Jan Kara, <jack@suse.cz>, 2001,2002
47 * New SMP locking.
48 * Jan Kara, <jack@suse.cz>, 10/2002
50 * Added journalled quota support, fix lock inversion problems
51 * Jan Kara, <jack@suse.cz>, 2003,2004
53 * (C) Copyright 1994 - 1997 Marco van Wieringen
56 #include <linux/errno.h>
57 #include <linux/kernel.h>
58 #include <linux/fs.h>
59 #include <linux/mount.h>
60 #include <linux/mm.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/kmod.h>
75 #include <linux/namei.h>
76 #include <linux/buffer_head.h>
77 #include <linux/capability.h>
78 #include <linux/quotaops.h>
79 #include <linux/writeback.h> /* for inode_lock, oddly enough.. */
81 #include <asm/uaccess.h>
84 * There are three quota SMP locks. dq_list_lock protects all lists with quotas
85 * and quota formats, dqstats structure containing statistics about the lists
86 * dq_data_lock protects data from dq_dqb and also mem_dqinfo structures and
87 * also guards consistency of dquot->dq_dqb with inode->i_blocks, i_bytes.
88 * i_blocks and i_bytes updates itself are guarded by i_lock acquired directly
89 * in inode_add_bytes() and inode_sub_bytes(). dq_state_lock protects
90 * modifications of quota state (on quotaon and quotaoff) and readers who care
91 * about latest values take it as well.
93 * The spinlock ordering is hence: dq_data_lock > dq_list_lock > i_lock,
94 * dq_list_lock > dq_state_lock
96 * Note that some things (eg. sb pointer, type, id) doesn't change during
97 * the life of the dquot structure and so needn't to be protected by a lock
99 * Any operation working on dquots via inode pointers must hold dqptr_sem. If
100 * operation is just reading pointers from inode (or not using them at all) the
101 * read lock is enough. If pointers are altered function must hold write lock.
102 * Special care needs to be taken about S_NOQUOTA inode flag (marking that
103 * inode is a quota file). Functions adding pointers from inode to dquots have
104 * to check this flag under dqptr_sem and then (if S_NOQUOTA is not set) they
105 * have to do all pointer modifications before dropping dqptr_sem. This makes
106 * sure they cannot race with quotaon which first sets S_NOQUOTA flag and
107 * then drops all pointers to dquots from an inode.
109 * Each dquot has its dq_lock mutex. Locked dquots might not be referenced
110 * from inodes (dquot_alloc_space() and such don't check the dq_lock).
111 * Currently dquot is locked only when it is being read to memory (or space for
112 * it is being allocated) on the first dqget() and when it is being released on
113 * the last dqput(). The allocation and release oparations are serialized by
114 * the dq_lock and by checking the use count in dquot_release(). Write
115 * operations on dquots don't hold dq_lock as they copy data under dq_data_lock
116 * spinlock to internal buffers before writing.
118 * Lock ordering (including related VFS locks) is the following:
119 * i_mutex > dqonoff_sem > journal_lock > dqptr_sem > dquot->dq_lock >
120 * dqio_mutex
121 * The lock ordering of dqptr_sem imposed by quota code is only dqonoff_sem >
122 * dqptr_sem. But filesystem has to count with the fact that functions such as
123 * dquot_alloc_space() acquire dqptr_sem and they usually have to be called
124 * from inside a transaction to keep filesystem consistency after a crash. Also
125 * filesystems usually want to do some IO on dquot from ->mark_dirty which is
126 * called with dqptr_sem held.
127 * i_mutex on quota files is special (it's below dqio_mutex)
130 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_list_lock);
131 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_state_lock);
132 __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_data_lock);
133 EXPORT_SYMBOL(dq_data_lock);
135 static char *quotatypes[] = INITQFNAMES;
136 static struct quota_format_type *quota_formats; /* List of registered formats */
137 static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
139 /* SLAB cache for dquot structures */
140 static struct kmem_cache *dquot_cachep;
142 int register_quota_format(struct quota_format_type *fmt)
144 spin_lock(&dq_list_lock);
145 fmt->qf_next = quota_formats;
146 quota_formats = fmt;
147 spin_unlock(&dq_list_lock);
148 return 0;
150 EXPORT_SYMBOL(register_quota_format);
152 void unregister_quota_format(struct quota_format_type *fmt)
154 struct quota_format_type **actqf;
156 spin_lock(&dq_list_lock);
157 for (actqf = &quota_formats; *actqf && *actqf != fmt;
158 actqf = &(*actqf)->qf_next)
160 if (*actqf)
161 *actqf = (*actqf)->qf_next;
162 spin_unlock(&dq_list_lock);
164 EXPORT_SYMBOL(unregister_quota_format);
166 static struct quota_format_type *find_quota_format(int id)
168 struct quota_format_type *actqf;
170 spin_lock(&dq_list_lock);
171 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
172 actqf = actqf->qf_next)
174 if (!actqf || !try_module_get(actqf->qf_owner)) {
175 int qm;
177 spin_unlock(&dq_list_lock);
179 for (qm = 0; module_names[qm].qm_fmt_id &&
180 module_names[qm].qm_fmt_id != id; qm++)
182 if (!module_names[qm].qm_fmt_id ||
183 request_module(module_names[qm].qm_mod_name))
184 return NULL;
186 spin_lock(&dq_list_lock);
187 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
188 actqf = actqf->qf_next)
190 if (actqf && !try_module_get(actqf->qf_owner))
191 actqf = NULL;
193 spin_unlock(&dq_list_lock);
194 return actqf;
197 static void put_quota_format(struct quota_format_type *fmt)
199 module_put(fmt->qf_owner);
203 * Dquot List Management:
204 * The quota code uses three lists for dquot management: the inuse_list,
205 * free_dquots, and dquot_hash[] array. A single dquot structure may be
206 * on all three lists, depending on its current state.
208 * All dquots are placed to the end of inuse_list when first created, and this
209 * list is used for invalidate operation, which must look at every dquot.
211 * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
212 * and this list is searched whenever we need an available dquot. Dquots are
213 * removed from the list as soon as they are used again, and
214 * dqstats.free_dquots gives the number of dquots on the list. When
215 * dquot is invalidated it's completely released from memory.
217 * Dquots with a specific identity (device, type and id) are placed on
218 * one of the dquot_hash[] hash chains. The provides an efficient search
219 * mechanism to locate a specific dquot.
222 static LIST_HEAD(inuse_list);
223 static LIST_HEAD(free_dquots);
224 static unsigned int dq_hash_bits, dq_hash_mask;
225 static struct hlist_head *dquot_hash;
227 struct dqstats dqstats;
228 EXPORT_SYMBOL(dqstats);
230 static qsize_t inode_get_rsv_space(struct inode *inode);
231 static void __dquot_initialize(struct inode *inode, int type);
233 static inline unsigned int
234 hashfn(const struct super_block *sb, unsigned int id, int type)
236 unsigned long tmp;
238 tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
239 return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
243 * Following list functions expect dq_list_lock to be held
245 static inline void insert_dquot_hash(struct dquot *dquot)
247 struct hlist_head *head;
248 head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id, dquot->dq_type);
249 hlist_add_head(&dquot->dq_hash, head);
252 static inline void remove_dquot_hash(struct dquot *dquot)
254 hlist_del_init(&dquot->dq_hash);
257 static struct dquot *find_dquot(unsigned int hashent, struct super_block *sb,
258 unsigned int id, int type)
260 struct hlist_node *node;
261 struct dquot *dquot;
263 hlist_for_each (node, dquot_hash+hashent) {
264 dquot = hlist_entry(node, struct dquot, dq_hash);
265 if (dquot->dq_sb == sb && dquot->dq_id == id &&
266 dquot->dq_type == type)
267 return dquot;
269 return NULL;
272 /* Add a dquot to the tail of the free list */
273 static inline void put_dquot_last(struct dquot *dquot)
275 list_add_tail(&dquot->dq_free, &free_dquots);
276 dqstats.free_dquots++;
279 static inline void remove_free_dquot(struct dquot *dquot)
281 if (list_empty(&dquot->dq_free))
282 return;
283 list_del_init(&dquot->dq_free);
284 dqstats.free_dquots--;
287 static inline void put_inuse(struct dquot *dquot)
289 /* We add to the back of inuse list so we don't have to restart
290 * when traversing this list and we block */
291 list_add_tail(&dquot->dq_inuse, &inuse_list);
292 dqstats.allocated_dquots++;
295 static inline void remove_inuse(struct dquot *dquot)
297 dqstats.allocated_dquots--;
298 list_del(&dquot->dq_inuse);
301 * End of list functions needing dq_list_lock
304 static void wait_on_dquot(struct dquot *dquot)
306 mutex_lock(&dquot->dq_lock);
307 mutex_unlock(&dquot->dq_lock);
310 static inline int dquot_dirty(struct dquot *dquot)
312 return test_bit(DQ_MOD_B, &dquot->dq_flags);
315 static inline int mark_dquot_dirty(struct dquot *dquot)
317 return dquot->dq_sb->dq_op->mark_dirty(dquot);
320 int dquot_mark_dquot_dirty(struct dquot *dquot)
322 spin_lock(&dq_list_lock);
323 if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags))
324 list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
325 info[dquot->dq_type].dqi_dirty_list);
326 spin_unlock(&dq_list_lock);
327 return 0;
329 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
331 /* Dirtify all the dquots - this can block when journalling */
332 static inline int mark_all_dquot_dirty(struct dquot * const *dquot)
334 int ret, err, cnt;
336 ret = err = 0;
337 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
338 if (dquot[cnt])
339 /* Even in case of error we have to continue */
340 ret = mark_dquot_dirty(dquot[cnt]);
341 if (!err)
342 err = ret;
344 return err;
347 static inline void dqput_all(struct dquot **dquot)
349 unsigned int cnt;
351 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
352 dqput(dquot[cnt]);
355 /* This function needs dq_list_lock */
356 static inline int clear_dquot_dirty(struct dquot *dquot)
358 if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags))
359 return 0;
360 list_del_init(&dquot->dq_dirty);
361 return 1;
364 void mark_info_dirty(struct super_block *sb, int type)
366 set_bit(DQF_INFO_DIRTY_B, &sb_dqopt(sb)->info[type].dqi_flags);
368 EXPORT_SYMBOL(mark_info_dirty);
371 * Read dquot from disk and alloc space for it
374 int dquot_acquire(struct dquot *dquot)
376 int ret = 0, ret2 = 0;
377 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
379 mutex_lock(&dquot->dq_lock);
380 mutex_lock(&dqopt->dqio_mutex);
381 if (!test_bit(DQ_READ_B, &dquot->dq_flags))
382 ret = dqopt->ops[dquot->dq_type]->read_dqblk(dquot);
383 if (ret < 0)
384 goto out_iolock;
385 set_bit(DQ_READ_B, &dquot->dq_flags);
386 /* Instantiate dquot if needed */
387 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
388 ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
389 /* Write the info if needed */
390 if (info_dirty(&dqopt->info[dquot->dq_type])) {
391 ret2 = dqopt->ops[dquot->dq_type]->write_file_info(
392 dquot->dq_sb, dquot->dq_type);
394 if (ret < 0)
395 goto out_iolock;
396 if (ret2 < 0) {
397 ret = ret2;
398 goto out_iolock;
401 set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
402 out_iolock:
403 mutex_unlock(&dqopt->dqio_mutex);
404 mutex_unlock(&dquot->dq_lock);
405 return ret;
407 EXPORT_SYMBOL(dquot_acquire);
410 * Write dquot to disk
412 int dquot_commit(struct dquot *dquot)
414 int ret = 0, ret2 = 0;
415 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
417 mutex_lock(&dqopt->dqio_mutex);
418 spin_lock(&dq_list_lock);
419 if (!clear_dquot_dirty(dquot)) {
420 spin_unlock(&dq_list_lock);
421 goto out_sem;
423 spin_unlock(&dq_list_lock);
424 /* Inactive dquot can be only if there was error during read/init
425 * => we have better not writing it */
426 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
427 ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
428 if (info_dirty(&dqopt->info[dquot->dq_type])) {
429 ret2 = dqopt->ops[dquot->dq_type]->write_file_info(
430 dquot->dq_sb, dquot->dq_type);
432 if (ret >= 0)
433 ret = ret2;
435 out_sem:
436 mutex_unlock(&dqopt->dqio_mutex);
437 return ret;
439 EXPORT_SYMBOL(dquot_commit);
442 * Release dquot
444 int dquot_release(struct dquot *dquot)
446 int ret = 0, ret2 = 0;
447 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
449 mutex_lock(&dquot->dq_lock);
450 /* Check whether we are not racing with some other dqget() */
451 if (atomic_read(&dquot->dq_count) > 1)
452 goto out_dqlock;
453 mutex_lock(&dqopt->dqio_mutex);
454 if (dqopt->ops[dquot->dq_type]->release_dqblk) {
455 ret = dqopt->ops[dquot->dq_type]->release_dqblk(dquot);
456 /* Write the info */
457 if (info_dirty(&dqopt->info[dquot->dq_type])) {
458 ret2 = dqopt->ops[dquot->dq_type]->write_file_info(
459 dquot->dq_sb, dquot->dq_type);
461 if (ret >= 0)
462 ret = ret2;
464 clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
465 mutex_unlock(&dqopt->dqio_mutex);
466 out_dqlock:
467 mutex_unlock(&dquot->dq_lock);
468 return ret;
470 EXPORT_SYMBOL(dquot_release);
472 void dquot_destroy(struct dquot *dquot)
474 kmem_cache_free(dquot_cachep, dquot);
476 EXPORT_SYMBOL(dquot_destroy);
478 static inline void do_destroy_dquot(struct dquot *dquot)
480 dquot->dq_sb->dq_op->destroy_dquot(dquot);
483 /* Invalidate all dquots on the list. Note that this function is called after
484 * quota is disabled and pointers from inodes removed so there cannot be new
485 * quota users. There can still be some users of quotas due to inodes being
486 * just deleted or pruned by prune_icache() (those are not attached to any
487 * list) or parallel quotactl call. We have to wait for such users.
489 static void invalidate_dquots(struct super_block *sb, int type)
491 struct dquot *dquot, *tmp;
493 restart:
494 spin_lock(&dq_list_lock);
495 list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
496 if (dquot->dq_sb != sb)
497 continue;
498 if (dquot->dq_type != type)
499 continue;
500 /* Wait for dquot users */
501 if (atomic_read(&dquot->dq_count)) {
502 DEFINE_WAIT(wait);
504 atomic_inc(&dquot->dq_count);
505 prepare_to_wait(&dquot->dq_wait_unused, &wait,
506 TASK_UNINTERRUPTIBLE);
507 spin_unlock(&dq_list_lock);
508 /* Once dqput() wakes us up, we know it's time to free
509 * the dquot.
510 * IMPORTANT: we rely on the fact that there is always
511 * at most one process waiting for dquot to free.
512 * Otherwise dq_count would be > 1 and we would never
513 * wake up.
515 if (atomic_read(&dquot->dq_count) > 1)
516 schedule();
517 finish_wait(&dquot->dq_wait_unused, &wait);
518 dqput(dquot);
519 /* At this moment dquot() need not exist (it could be
520 * reclaimed by prune_dqcache(). Hence we must
521 * restart. */
522 goto restart;
525 * Quota now has no users and it has been written on last
526 * dqput()
528 remove_dquot_hash(dquot);
529 remove_free_dquot(dquot);
530 remove_inuse(dquot);
531 do_destroy_dquot(dquot);
533 spin_unlock(&dq_list_lock);
536 /* Call callback for every active dquot on given filesystem */
537 int dquot_scan_active(struct super_block *sb,
538 int (*fn)(struct dquot *dquot, unsigned long priv),
539 unsigned long priv)
541 struct dquot *dquot, *old_dquot = NULL;
542 int ret = 0;
544 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
545 spin_lock(&dq_list_lock);
546 list_for_each_entry(dquot, &inuse_list, dq_inuse) {
547 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
548 continue;
549 if (dquot->dq_sb != sb)
550 continue;
551 /* Now we have active dquot so we can just increase use count */
552 atomic_inc(&dquot->dq_count);
553 dqstats.lookups++;
554 spin_unlock(&dq_list_lock);
555 dqput(old_dquot);
556 old_dquot = dquot;
557 ret = fn(dquot, priv);
558 if (ret < 0)
559 goto out;
560 spin_lock(&dq_list_lock);
561 /* We are safe to continue now because our dquot could not
562 * be moved out of the inuse list while we hold the reference */
564 spin_unlock(&dq_list_lock);
565 out:
566 dqput(old_dquot);
567 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
568 return ret;
570 EXPORT_SYMBOL(dquot_scan_active);
572 int vfs_quota_sync(struct super_block *sb, int type, int wait)
574 struct list_head *dirty;
575 struct dquot *dquot;
576 struct quota_info *dqopt = sb_dqopt(sb);
577 int cnt;
579 mutex_lock(&dqopt->dqonoff_mutex);
580 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
581 if (type != -1 && cnt != type)
582 continue;
583 if (!sb_has_quota_active(sb, cnt))
584 continue;
585 spin_lock(&dq_list_lock);
586 dirty = &dqopt->info[cnt].dqi_dirty_list;
587 while (!list_empty(dirty)) {
588 dquot = list_first_entry(dirty, struct dquot,
589 dq_dirty);
590 /* Dirty and inactive can be only bad dquot... */
591 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
592 clear_dquot_dirty(dquot);
593 continue;
595 /* Now we have active dquot from which someone is
596 * holding reference so we can safely just increase
597 * use count */
598 atomic_inc(&dquot->dq_count);
599 dqstats.lookups++;
600 spin_unlock(&dq_list_lock);
601 sb->dq_op->write_dquot(dquot);
602 dqput(dquot);
603 spin_lock(&dq_list_lock);
605 spin_unlock(&dq_list_lock);
608 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
609 if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
610 && info_dirty(&dqopt->info[cnt]))
611 sb->dq_op->write_info(sb, cnt);
612 spin_lock(&dq_list_lock);
613 dqstats.syncs++;
614 spin_unlock(&dq_list_lock);
615 mutex_unlock(&dqopt->dqonoff_mutex);
617 if (!wait || (sb_dqopt(sb)->flags & DQUOT_QUOTA_SYS_FILE))
618 return 0;
620 /* This is not very clever (and fast) but currently I don't know about
621 * any other simple way of getting quota data to disk and we must get
622 * them there for userspace to be visible... */
623 if (sb->s_op->sync_fs)
624 sb->s_op->sync_fs(sb, 1);
625 sync_blockdev(sb->s_bdev);
628 * Now when everything is written we can discard the pagecache so
629 * that userspace sees the changes.
631 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
632 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
633 if (type != -1 && cnt != type)
634 continue;
635 if (!sb_has_quota_active(sb, cnt))
636 continue;
637 mutex_lock_nested(&sb_dqopt(sb)->files[cnt]->i_mutex,
638 I_MUTEX_QUOTA);
639 truncate_inode_pages(&sb_dqopt(sb)->files[cnt]->i_data, 0);
640 mutex_unlock(&sb_dqopt(sb)->files[cnt]->i_mutex);
642 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
644 return 0;
646 EXPORT_SYMBOL(vfs_quota_sync);
648 /* Free unused dquots from cache */
649 static void prune_dqcache(int count)
651 struct list_head *head;
652 struct dquot *dquot;
654 head = free_dquots.prev;
655 while (head != &free_dquots && count) {
656 dquot = list_entry(head, struct dquot, dq_free);
657 remove_dquot_hash(dquot);
658 remove_free_dquot(dquot);
659 remove_inuse(dquot);
660 do_destroy_dquot(dquot);
661 count--;
662 head = free_dquots.prev;
667 * This is called from kswapd when we think we need some
668 * more memory
671 static int shrink_dqcache_memory(int nr, gfp_t gfp_mask)
673 if (nr) {
674 spin_lock(&dq_list_lock);
675 prune_dqcache(nr);
676 spin_unlock(&dq_list_lock);
678 return (dqstats.free_dquots / 100) * sysctl_vfs_cache_pressure;
681 static struct shrinker dqcache_shrinker = {
682 .shrink = shrink_dqcache_memory,
683 .seeks = DEFAULT_SEEKS,
687 * Put reference to dquot
688 * NOTE: If you change this function please check whether dqput_blocks() works right...
690 void dqput(struct dquot *dquot)
692 int ret;
694 if (!dquot)
695 return;
696 #ifdef CONFIG_QUOTA_DEBUG
697 if (!atomic_read(&dquot->dq_count)) {
698 printk("VFS: dqput: trying to free free dquot\n");
699 printk("VFS: device %s, dquot of %s %d\n",
700 dquot->dq_sb->s_id,
701 quotatypes[dquot->dq_type],
702 dquot->dq_id);
703 BUG();
705 #endif
707 spin_lock(&dq_list_lock);
708 dqstats.drops++;
709 spin_unlock(&dq_list_lock);
710 we_slept:
711 spin_lock(&dq_list_lock);
712 if (atomic_read(&dquot->dq_count) > 1) {
713 /* We have more than one user... nothing to do */
714 atomic_dec(&dquot->dq_count);
715 /* Releasing dquot during quotaoff phase? */
716 if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_type) &&
717 atomic_read(&dquot->dq_count) == 1)
718 wake_up(&dquot->dq_wait_unused);
719 spin_unlock(&dq_list_lock);
720 return;
722 /* Need to release dquot? */
723 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && dquot_dirty(dquot)) {
724 spin_unlock(&dq_list_lock);
725 /* Commit dquot before releasing */
726 ret = dquot->dq_sb->dq_op->write_dquot(dquot);
727 if (ret < 0) {
728 printk(KERN_ERR "VFS: cannot write quota structure on "
729 "device %s (error %d). Quota may get out of "
730 "sync!\n", dquot->dq_sb->s_id, ret);
732 * We clear dirty bit anyway, so that we avoid
733 * infinite loop here
735 spin_lock(&dq_list_lock);
736 clear_dquot_dirty(dquot);
737 spin_unlock(&dq_list_lock);
739 goto we_slept;
741 /* Clear flag in case dquot was inactive (something bad happened) */
742 clear_dquot_dirty(dquot);
743 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
744 spin_unlock(&dq_list_lock);
745 dquot->dq_sb->dq_op->release_dquot(dquot);
746 goto we_slept;
748 atomic_dec(&dquot->dq_count);
749 #ifdef CONFIG_QUOTA_DEBUG
750 /* sanity check */
751 BUG_ON(!list_empty(&dquot->dq_free));
752 #endif
753 put_dquot_last(dquot);
754 spin_unlock(&dq_list_lock);
756 EXPORT_SYMBOL(dqput);
758 struct dquot *dquot_alloc(struct super_block *sb, int type)
760 return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
762 EXPORT_SYMBOL(dquot_alloc);
764 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
766 struct dquot *dquot;
768 dquot = sb->dq_op->alloc_dquot(sb, type);
769 if(!dquot)
770 return NULL;
772 mutex_init(&dquot->dq_lock);
773 INIT_LIST_HEAD(&dquot->dq_free);
774 INIT_LIST_HEAD(&dquot->dq_inuse);
775 INIT_HLIST_NODE(&dquot->dq_hash);
776 INIT_LIST_HEAD(&dquot->dq_dirty);
777 init_waitqueue_head(&dquot->dq_wait_unused);
778 dquot->dq_sb = sb;
779 dquot->dq_type = type;
780 atomic_set(&dquot->dq_count, 1);
782 return dquot;
786 * Get reference to dquot
788 * Locking is slightly tricky here. We are guarded from parallel quotaoff()
789 * destroying our dquot by:
790 * a) checking for quota flags under dq_list_lock and
791 * b) getting a reference to dquot before we release dq_list_lock
793 struct dquot *dqget(struct super_block *sb, unsigned int id, int type)
795 unsigned int hashent = hashfn(sb, id, type);
796 struct dquot *dquot = NULL, *empty = NULL;
798 if (!sb_has_quota_active(sb, type))
799 return NULL;
800 we_slept:
801 spin_lock(&dq_list_lock);
802 spin_lock(&dq_state_lock);
803 if (!sb_has_quota_active(sb, type)) {
804 spin_unlock(&dq_state_lock);
805 spin_unlock(&dq_list_lock);
806 goto out;
808 spin_unlock(&dq_state_lock);
810 dquot = find_dquot(hashent, sb, id, type);
811 if (!dquot) {
812 if (!empty) {
813 spin_unlock(&dq_list_lock);
814 empty = get_empty_dquot(sb, type);
815 if (!empty)
816 schedule(); /* Try to wait for a moment... */
817 goto we_slept;
819 dquot = empty;
820 empty = NULL;
821 dquot->dq_id = id;
822 /* all dquots go on the inuse_list */
823 put_inuse(dquot);
824 /* hash it first so it can be found */
825 insert_dquot_hash(dquot);
826 dqstats.lookups++;
827 spin_unlock(&dq_list_lock);
828 } else {
829 if (!atomic_read(&dquot->dq_count))
830 remove_free_dquot(dquot);
831 atomic_inc(&dquot->dq_count);
832 dqstats.cache_hits++;
833 dqstats.lookups++;
834 spin_unlock(&dq_list_lock);
836 /* Wait for dq_lock - after this we know that either dquot_release() is
837 * already finished or it will be canceled due to dq_count > 1 test */
838 wait_on_dquot(dquot);
839 /* Read the dquot / allocate space in quota file */
840 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) &&
841 sb->dq_op->acquire_dquot(dquot) < 0) {
842 dqput(dquot);
843 dquot = NULL;
844 goto out;
846 #ifdef CONFIG_QUOTA_DEBUG
847 BUG_ON(!dquot->dq_sb); /* Has somebody invalidated entry under us? */
848 #endif
849 out:
850 if (empty)
851 do_destroy_dquot(empty);
853 return dquot;
855 EXPORT_SYMBOL(dqget);
857 static int dqinit_needed(struct inode *inode, int type)
859 int cnt;
861 if (IS_NOQUOTA(inode))
862 return 0;
863 if (type != -1)
864 return !inode->i_dquot[type];
865 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
866 if (!inode->i_dquot[cnt])
867 return 1;
868 return 0;
871 /* This routine is guarded by dqonoff_mutex mutex */
872 static void add_dquot_ref(struct super_block *sb, int type)
874 struct inode *inode, *old_inode = NULL;
875 #ifdef CONFIG_QUOTA_DEBUG
876 int reserved = 0;
877 #endif
879 spin_lock(&inode_lock);
880 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
881 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE|I_NEW))
882 continue;
883 #ifdef CONFIG_QUOTA_DEBUG
884 if (unlikely(inode_get_rsv_space(inode) > 0))
885 reserved = 1;
886 #endif
887 if (!atomic_read(&inode->i_writecount))
888 continue;
889 if (!dqinit_needed(inode, type))
890 continue;
892 __iget(inode);
893 spin_unlock(&inode_lock);
895 iput(old_inode);
896 __dquot_initialize(inode, type);
897 /* We hold a reference to 'inode' so it couldn't have been
898 * removed from s_inodes list while we dropped the inode_lock.
899 * We cannot iput the inode now as we can be holding the last
900 * reference and we cannot iput it under inode_lock. So we
901 * keep the reference and iput it later. */
902 old_inode = inode;
903 spin_lock(&inode_lock);
905 spin_unlock(&inode_lock);
906 iput(old_inode);
908 #ifdef CONFIG_QUOTA_DEBUG
909 if (reserved) {
910 printk(KERN_WARNING "VFS (%s): Writes happened before quota"
911 " was turned on thus quota information is probably "
912 "inconsistent. Please run quotacheck(8).\n", sb->s_id);
914 #endif
918 * Return 0 if dqput() won't block.
919 * (note that 1 doesn't necessarily mean blocking)
921 static inline int dqput_blocks(struct dquot *dquot)
923 if (atomic_read(&dquot->dq_count) <= 1)
924 return 1;
925 return 0;
929 * Remove references to dquots from inode and add dquot to list for freeing
930 * if we have the last referece to dquot
931 * We can't race with anybody because we hold dqptr_sem for writing...
933 static int remove_inode_dquot_ref(struct inode *inode, int type,
934 struct list_head *tofree_head)
936 struct dquot *dquot = inode->i_dquot[type];
938 inode->i_dquot[type] = NULL;
939 if (dquot) {
940 if (dqput_blocks(dquot)) {
941 #ifdef CONFIG_QUOTA_DEBUG
942 if (atomic_read(&dquot->dq_count) != 1)
943 printk(KERN_WARNING "VFS: Adding dquot with dq_count %d to dispose list.\n", atomic_read(&dquot->dq_count));
944 #endif
945 spin_lock(&dq_list_lock);
946 /* As dquot must have currently users it can't be on
947 * the free list... */
948 list_add(&dquot->dq_free, tofree_head);
949 spin_unlock(&dq_list_lock);
950 return 1;
952 else
953 dqput(dquot); /* We have guaranteed we won't block */
955 return 0;
959 * Free list of dquots
960 * Dquots are removed from inodes and no new references can be got so we are
961 * the only ones holding reference
963 static void put_dquot_list(struct list_head *tofree_head)
965 struct list_head *act_head;
966 struct dquot *dquot;
968 act_head = tofree_head->next;
969 while (act_head != tofree_head) {
970 dquot = list_entry(act_head, struct dquot, dq_free);
971 act_head = act_head->next;
972 /* Remove dquot from the list so we won't have problems... */
973 list_del_init(&dquot->dq_free);
974 dqput(dquot);
978 static void remove_dquot_ref(struct super_block *sb, int type,
979 struct list_head *tofree_head)
981 struct inode *inode;
983 spin_lock(&inode_lock);
984 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
986 * We have to scan also I_NEW inodes because they can already
987 * have quota pointer initialized. Luckily, we need to touch
988 * only quota pointers and these have separate locking
989 * (dqptr_sem).
991 if (!IS_NOQUOTA(inode))
992 remove_inode_dquot_ref(inode, type, tofree_head);
994 spin_unlock(&inode_lock);
997 /* Gather all references from inodes and drop them */
998 static void drop_dquot_ref(struct super_block *sb, int type)
1000 LIST_HEAD(tofree_head);
1002 if (sb->dq_op) {
1003 down_write(&sb_dqopt(sb)->dqptr_sem);
1004 remove_dquot_ref(sb, type, &tofree_head);
1005 up_write(&sb_dqopt(sb)->dqptr_sem);
1006 put_dquot_list(&tofree_head);
1010 static inline void dquot_incr_inodes(struct dquot *dquot, qsize_t number)
1012 dquot->dq_dqb.dqb_curinodes += number;
1015 static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
1017 dquot->dq_dqb.dqb_curspace += number;
1020 static inline void dquot_resv_space(struct dquot *dquot, qsize_t number)
1022 dquot->dq_dqb.dqb_rsvspace += number;
1026 * Claim reserved quota space
1028 static void dquot_claim_reserved_space(struct dquot *dquot, qsize_t number)
1030 if (dquot->dq_dqb.dqb_rsvspace < number) {
1031 WARN_ON_ONCE(1);
1032 number = dquot->dq_dqb.dqb_rsvspace;
1034 dquot->dq_dqb.dqb_curspace += number;
1035 dquot->dq_dqb.dqb_rsvspace -= number;
1038 static inline
1039 void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1041 if (dquot->dq_dqb.dqb_rsvspace >= number)
1042 dquot->dq_dqb.dqb_rsvspace -= number;
1043 else {
1044 WARN_ON_ONCE(1);
1045 dquot->dq_dqb.dqb_rsvspace = 0;
1049 static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1051 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1052 dquot->dq_dqb.dqb_curinodes >= number)
1053 dquot->dq_dqb.dqb_curinodes -= number;
1054 else
1055 dquot->dq_dqb.dqb_curinodes = 0;
1056 if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1057 dquot->dq_dqb.dqb_itime = (time_t) 0;
1058 clear_bit(DQ_INODES_B, &dquot->dq_flags);
1061 static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1063 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1064 dquot->dq_dqb.dqb_curspace >= number)
1065 dquot->dq_dqb.dqb_curspace -= number;
1066 else
1067 dquot->dq_dqb.dqb_curspace = 0;
1068 if (dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1069 dquot->dq_dqb.dqb_btime = (time_t) 0;
1070 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1073 static int warning_issued(struct dquot *dquot, const int warntype)
1075 int flag = (warntype == QUOTA_NL_BHARDWARN ||
1076 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1077 ((warntype == QUOTA_NL_IHARDWARN ||
1078 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1080 if (!flag)
1081 return 0;
1082 return test_and_set_bit(flag, &dquot->dq_flags);
1085 #ifdef CONFIG_PRINT_QUOTA_WARNING
1086 static int flag_print_warnings = 1;
1088 static int need_print_warning(struct dquot *dquot)
1090 if (!flag_print_warnings)
1091 return 0;
1093 switch (dquot->dq_type) {
1094 case USRQUOTA:
1095 return current_fsuid() == dquot->dq_id;
1096 case GRPQUOTA:
1097 return in_group_p(dquot->dq_id);
1099 return 0;
1102 /* Print warning to user which exceeded quota */
1103 static void print_warning(struct dquot *dquot, const int warntype)
1105 char *msg = NULL;
1106 struct tty_struct *tty;
1108 if (warntype == QUOTA_NL_IHARDBELOW ||
1109 warntype == QUOTA_NL_ISOFTBELOW ||
1110 warntype == QUOTA_NL_BHARDBELOW ||
1111 warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(dquot))
1112 return;
1114 tty = get_current_tty();
1115 if (!tty)
1116 return;
1117 tty_write_message(tty, dquot->dq_sb->s_id);
1118 if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1119 tty_write_message(tty, ": warning, ");
1120 else
1121 tty_write_message(tty, ": write failed, ");
1122 tty_write_message(tty, quotatypes[dquot->dq_type]);
1123 switch (warntype) {
1124 case QUOTA_NL_IHARDWARN:
1125 msg = " file limit reached.\r\n";
1126 break;
1127 case QUOTA_NL_ISOFTLONGWARN:
1128 msg = " file quota exceeded too long.\r\n";
1129 break;
1130 case QUOTA_NL_ISOFTWARN:
1131 msg = " file quota exceeded.\r\n";
1132 break;
1133 case QUOTA_NL_BHARDWARN:
1134 msg = " block limit reached.\r\n";
1135 break;
1136 case QUOTA_NL_BSOFTLONGWARN:
1137 msg = " block quota exceeded too long.\r\n";
1138 break;
1139 case QUOTA_NL_BSOFTWARN:
1140 msg = " block quota exceeded.\r\n";
1141 break;
1143 tty_write_message(tty, msg);
1144 tty_kref_put(tty);
1146 #endif
1149 * Write warnings to the console and send warning messages over netlink.
1151 * Note that this function can sleep.
1153 static void flush_warnings(struct dquot *const *dquots, char *warntype)
1155 struct dquot *dq;
1156 int i;
1158 for (i = 0; i < MAXQUOTAS; i++) {
1159 dq = dquots[i];
1160 if (dq && warntype[i] != QUOTA_NL_NOWARN &&
1161 !warning_issued(dq, warntype[i])) {
1162 #ifdef CONFIG_PRINT_QUOTA_WARNING
1163 print_warning(dq, warntype[i]);
1164 #endif
1165 quota_send_warning(dq->dq_type, dq->dq_id,
1166 dq->dq_sb->s_dev, warntype[i]);
1171 static int ignore_hardlimit(struct dquot *dquot)
1173 struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_type];
1175 return capable(CAP_SYS_RESOURCE) &&
1176 (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1177 !(info->dqi_flags & V1_DQF_RSQUASH));
1180 /* needs dq_data_lock */
1181 static int check_idq(struct dquot *dquot, qsize_t inodes, char *warntype)
1183 qsize_t newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1185 *warntype = QUOTA_NL_NOWARN;
1186 if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_type) ||
1187 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1188 return 0;
1190 if (dquot->dq_dqb.dqb_ihardlimit &&
1191 newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1192 !ignore_hardlimit(dquot)) {
1193 *warntype = QUOTA_NL_IHARDWARN;
1194 return -EDQUOT;
1197 if (dquot->dq_dqb.dqb_isoftlimit &&
1198 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1199 dquot->dq_dqb.dqb_itime &&
1200 get_seconds() >= dquot->dq_dqb.dqb_itime &&
1201 !ignore_hardlimit(dquot)) {
1202 *warntype = QUOTA_NL_ISOFTLONGWARN;
1203 return -EDQUOT;
1206 if (dquot->dq_dqb.dqb_isoftlimit &&
1207 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1208 dquot->dq_dqb.dqb_itime == 0) {
1209 *warntype = QUOTA_NL_ISOFTWARN;
1210 dquot->dq_dqb.dqb_itime = get_seconds() +
1211 sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
1214 return 0;
1217 /* needs dq_data_lock */
1218 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc, char *warntype)
1220 qsize_t tspace;
1221 struct super_block *sb = dquot->dq_sb;
1223 *warntype = QUOTA_NL_NOWARN;
1224 if (!sb_has_quota_limits_enabled(sb, dquot->dq_type) ||
1225 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1226 return 0;
1228 tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1229 + space;
1231 if (dquot->dq_dqb.dqb_bhardlimit &&
1232 tspace > dquot->dq_dqb.dqb_bhardlimit &&
1233 !ignore_hardlimit(dquot)) {
1234 if (!prealloc)
1235 *warntype = QUOTA_NL_BHARDWARN;
1236 return -EDQUOT;
1239 if (dquot->dq_dqb.dqb_bsoftlimit &&
1240 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1241 dquot->dq_dqb.dqb_btime &&
1242 get_seconds() >= dquot->dq_dqb.dqb_btime &&
1243 !ignore_hardlimit(dquot)) {
1244 if (!prealloc)
1245 *warntype = QUOTA_NL_BSOFTLONGWARN;
1246 return -EDQUOT;
1249 if (dquot->dq_dqb.dqb_bsoftlimit &&
1250 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1251 dquot->dq_dqb.dqb_btime == 0) {
1252 if (!prealloc) {
1253 *warntype = QUOTA_NL_BSOFTWARN;
1254 dquot->dq_dqb.dqb_btime = get_seconds() +
1255 sb_dqopt(sb)->info[dquot->dq_type].dqi_bgrace;
1257 else
1259 * We don't allow preallocation to exceed softlimit so exceeding will
1260 * be always printed
1262 return -EDQUOT;
1265 return 0;
1268 static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1270 qsize_t newinodes;
1272 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1273 dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1274 !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_type))
1275 return QUOTA_NL_NOWARN;
1277 newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1278 if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1279 return QUOTA_NL_ISOFTBELOW;
1280 if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1281 newinodes < dquot->dq_dqb.dqb_ihardlimit)
1282 return QUOTA_NL_IHARDBELOW;
1283 return QUOTA_NL_NOWARN;
1286 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1288 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1289 dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1290 return QUOTA_NL_NOWARN;
1292 if (dquot->dq_dqb.dqb_curspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1293 return QUOTA_NL_BSOFTBELOW;
1294 if (dquot->dq_dqb.dqb_curspace >= dquot->dq_dqb.dqb_bhardlimit &&
1295 dquot->dq_dqb.dqb_curspace - space < dquot->dq_dqb.dqb_bhardlimit)
1296 return QUOTA_NL_BHARDBELOW;
1297 return QUOTA_NL_NOWARN;
1301 * Initialize quota pointers in inode
1303 * We do things in a bit complicated way but by that we avoid calling
1304 * dqget() and thus filesystem callbacks under dqptr_sem.
1306 * It is better to call this function outside of any transaction as it
1307 * might need a lot of space in journal for dquot structure allocation.
1309 static void __dquot_initialize(struct inode *inode, int type)
1311 unsigned int id = 0;
1312 int cnt;
1313 struct dquot *got[MAXQUOTAS];
1314 struct super_block *sb = inode->i_sb;
1315 qsize_t rsv;
1317 /* First test before acquiring mutex - solves deadlocks when we
1318 * re-enter the quota code and are already holding the mutex */
1319 if (!sb_any_quota_active(inode->i_sb) || IS_NOQUOTA(inode))
1320 return;
1322 /* First get references to structures we might need. */
1323 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1324 got[cnt] = NULL;
1325 if (type != -1 && cnt != type)
1326 continue;
1327 switch (cnt) {
1328 case USRQUOTA:
1329 id = inode->i_uid;
1330 break;
1331 case GRPQUOTA:
1332 id = inode->i_gid;
1333 break;
1335 got[cnt] = dqget(sb, id, cnt);
1338 down_write(&sb_dqopt(sb)->dqptr_sem);
1339 if (IS_NOQUOTA(inode))
1340 goto out_err;
1341 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1342 if (type != -1 && cnt != type)
1343 continue;
1344 /* Avoid races with quotaoff() */
1345 if (!sb_has_quota_active(sb, cnt))
1346 continue;
1347 if (!inode->i_dquot[cnt]) {
1348 inode->i_dquot[cnt] = got[cnt];
1349 got[cnt] = NULL;
1351 * Make quota reservation system happy if someone
1352 * did a write before quota was turned on
1354 rsv = inode_get_rsv_space(inode);
1355 if (unlikely(rsv))
1356 dquot_resv_space(inode->i_dquot[cnt], rsv);
1359 out_err:
1360 up_write(&sb_dqopt(sb)->dqptr_sem);
1361 /* Drop unused references */
1362 dqput_all(got);
1365 void dquot_initialize(struct inode *inode)
1367 __dquot_initialize(inode, -1);
1369 EXPORT_SYMBOL(dquot_initialize);
1372 * Release all quotas referenced by inode
1374 static void __dquot_drop(struct inode *inode)
1376 int cnt;
1377 struct dquot *put[MAXQUOTAS];
1379 down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1380 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1381 put[cnt] = inode->i_dquot[cnt];
1382 inode->i_dquot[cnt] = NULL;
1384 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1385 dqput_all(put);
1388 void dquot_drop(struct inode *inode)
1390 int cnt;
1392 if (IS_NOQUOTA(inode))
1393 return;
1396 * Test before calling to rule out calls from proc and such
1397 * where we are not allowed to block. Note that this is
1398 * actually reliable test even without the lock - the caller
1399 * must assure that nobody can come after the DQUOT_DROP and
1400 * add quota pointers back anyway.
1402 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1403 if (inode->i_dquot[cnt])
1404 break;
1407 if (cnt < MAXQUOTAS)
1408 __dquot_drop(inode);
1410 EXPORT_SYMBOL(dquot_drop);
1413 * inode_reserved_space is managed internally by quota, and protected by
1414 * i_lock similar to i_blocks+i_bytes.
1416 static qsize_t *inode_reserved_space(struct inode * inode)
1418 /* Filesystem must explicitly define it's own method in order to use
1419 * quota reservation interface */
1420 BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1421 return inode->i_sb->dq_op->get_reserved_space(inode);
1424 void inode_add_rsv_space(struct inode *inode, qsize_t number)
1426 spin_lock(&inode->i_lock);
1427 *inode_reserved_space(inode) += number;
1428 spin_unlock(&inode->i_lock);
1430 EXPORT_SYMBOL(inode_add_rsv_space);
1432 void inode_claim_rsv_space(struct inode *inode, qsize_t number)
1434 spin_lock(&inode->i_lock);
1435 *inode_reserved_space(inode) -= number;
1436 __inode_add_bytes(inode, number);
1437 spin_unlock(&inode->i_lock);
1439 EXPORT_SYMBOL(inode_claim_rsv_space);
1441 void inode_sub_rsv_space(struct inode *inode, qsize_t number)
1443 spin_lock(&inode->i_lock);
1444 *inode_reserved_space(inode) -= number;
1445 spin_unlock(&inode->i_lock);
1447 EXPORT_SYMBOL(inode_sub_rsv_space);
1449 static qsize_t inode_get_rsv_space(struct inode *inode)
1451 qsize_t ret;
1453 if (!inode->i_sb->dq_op->get_reserved_space)
1454 return 0;
1455 spin_lock(&inode->i_lock);
1456 ret = *inode_reserved_space(inode);
1457 spin_unlock(&inode->i_lock);
1458 return ret;
1461 static void inode_incr_space(struct inode *inode, qsize_t number,
1462 int reserve)
1464 if (reserve)
1465 inode_add_rsv_space(inode, number);
1466 else
1467 inode_add_bytes(inode, number);
1470 static void inode_decr_space(struct inode *inode, qsize_t number, int reserve)
1472 if (reserve)
1473 inode_sub_rsv_space(inode, number);
1474 else
1475 inode_sub_bytes(inode, number);
1479 * This functions updates i_blocks+i_bytes fields and quota information
1480 * (together with appropriate checks).
1482 * NOTE: We absolutely rely on the fact that caller dirties the inode
1483 * (usually helpers in quotaops.h care about this) and holds a handle for
1484 * the current transaction so that dquot write and inode write go into the
1485 * same transaction.
1489 * This operation can block, but only after everything is updated
1491 int __dquot_alloc_space(struct inode *inode, qsize_t number,
1492 int warn, int reserve)
1494 int cnt, ret = 0;
1495 char warntype[MAXQUOTAS];
1498 * First test before acquiring mutex - solves deadlocks when we
1499 * re-enter the quota code and are already holding the mutex
1501 if (!sb_any_quota_active(inode->i_sb) || IS_NOQUOTA(inode)) {
1502 inode_incr_space(inode, number, reserve);
1503 goto out;
1506 down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1507 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1508 warntype[cnt] = QUOTA_NL_NOWARN;
1510 spin_lock(&dq_data_lock);
1511 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1512 if (!inode->i_dquot[cnt])
1513 continue;
1514 ret = check_bdq(inode->i_dquot[cnt], number, !warn,
1515 warntype+cnt);
1516 if (ret) {
1517 spin_unlock(&dq_data_lock);
1518 goto out_flush_warn;
1521 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1522 if (!inode->i_dquot[cnt])
1523 continue;
1524 if (reserve)
1525 dquot_resv_space(inode->i_dquot[cnt], number);
1526 else
1527 dquot_incr_space(inode->i_dquot[cnt], number);
1529 inode_incr_space(inode, number, reserve);
1530 spin_unlock(&dq_data_lock);
1532 if (reserve)
1533 goto out_flush_warn;
1534 mark_all_dquot_dirty(inode->i_dquot);
1535 out_flush_warn:
1536 flush_warnings(inode->i_dquot, warntype);
1537 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1538 out:
1539 return ret;
1541 EXPORT_SYMBOL(__dquot_alloc_space);
1544 * This operation can block, but only after everything is updated
1546 int dquot_alloc_inode(const struct inode *inode)
1548 int cnt, ret = 0;
1549 char warntype[MAXQUOTAS];
1551 /* First test before acquiring mutex - solves deadlocks when we
1552 * re-enter the quota code and are already holding the mutex */
1553 if (!sb_any_quota_active(inode->i_sb) || IS_NOQUOTA(inode))
1554 return 0;
1555 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1556 warntype[cnt] = QUOTA_NL_NOWARN;
1557 down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1558 spin_lock(&dq_data_lock);
1559 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1560 if (!inode->i_dquot[cnt])
1561 continue;
1562 ret = check_idq(inode->i_dquot[cnt], 1, warntype + cnt);
1563 if (ret)
1564 goto warn_put_all;
1567 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1568 if (!inode->i_dquot[cnt])
1569 continue;
1570 dquot_incr_inodes(inode->i_dquot[cnt], 1);
1573 warn_put_all:
1574 spin_unlock(&dq_data_lock);
1575 if (ret == 0)
1576 mark_all_dquot_dirty(inode->i_dquot);
1577 flush_warnings(inode->i_dquot, warntype);
1578 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1579 return ret;
1581 EXPORT_SYMBOL(dquot_alloc_inode);
1584 * Convert in-memory reserved quotas to real consumed quotas
1586 int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1588 int cnt;
1590 if (!sb_any_quota_active(inode->i_sb) || IS_NOQUOTA(inode)) {
1591 inode_claim_rsv_space(inode, number);
1592 return 0;
1595 down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1596 spin_lock(&dq_data_lock);
1597 /* Claim reserved quotas to allocated quotas */
1598 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1599 if (inode->i_dquot[cnt])
1600 dquot_claim_reserved_space(inode->i_dquot[cnt],
1601 number);
1603 /* Update inode bytes */
1604 inode_claim_rsv_space(inode, number);
1605 spin_unlock(&dq_data_lock);
1606 mark_all_dquot_dirty(inode->i_dquot);
1607 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1608 return 0;
1610 EXPORT_SYMBOL(dquot_claim_space_nodirty);
1613 * This operation can block, but only after everything is updated
1615 void __dquot_free_space(struct inode *inode, qsize_t number, int reserve)
1617 unsigned int cnt;
1618 char warntype[MAXQUOTAS];
1620 /* First test before acquiring mutex - solves deadlocks when we
1621 * re-enter the quota code and are already holding the mutex */
1622 if (!sb_any_quota_active(inode->i_sb) || IS_NOQUOTA(inode)) {
1623 inode_decr_space(inode, number, reserve);
1624 return;
1627 down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1628 spin_lock(&dq_data_lock);
1629 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1630 if (!inode->i_dquot[cnt])
1631 continue;
1632 warntype[cnt] = info_bdq_free(inode->i_dquot[cnt], number);
1633 if (reserve)
1634 dquot_free_reserved_space(inode->i_dquot[cnt], number);
1635 else
1636 dquot_decr_space(inode->i_dquot[cnt], number);
1638 inode_decr_space(inode, number, reserve);
1639 spin_unlock(&dq_data_lock);
1641 if (reserve)
1642 goto out_unlock;
1643 mark_all_dquot_dirty(inode->i_dquot);
1644 out_unlock:
1645 flush_warnings(inode->i_dquot, warntype);
1646 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1648 EXPORT_SYMBOL(__dquot_free_space);
1651 * This operation can block, but only after everything is updated
1653 void dquot_free_inode(const struct inode *inode)
1655 unsigned int cnt;
1656 char warntype[MAXQUOTAS];
1658 /* First test before acquiring mutex - solves deadlocks when we
1659 * re-enter the quota code and are already holding the mutex */
1660 if (!sb_any_quota_active(inode->i_sb) || IS_NOQUOTA(inode))
1661 return;
1663 down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1664 spin_lock(&dq_data_lock);
1665 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1666 if (!inode->i_dquot[cnt])
1667 continue;
1668 warntype[cnt] = info_idq_free(inode->i_dquot[cnt], 1);
1669 dquot_decr_inodes(inode->i_dquot[cnt], 1);
1671 spin_unlock(&dq_data_lock);
1672 mark_all_dquot_dirty(inode->i_dquot);
1673 flush_warnings(inode->i_dquot, warntype);
1674 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1676 EXPORT_SYMBOL(dquot_free_inode);
1679 * Transfer the number of inode and blocks from one diskquota to an other.
1681 * This operation can block, but only after everything is updated
1682 * A transaction must be started when entering this function.
1684 static int __dquot_transfer(struct inode *inode, qid_t *chid, unsigned long mask)
1686 qsize_t space, cur_space;
1687 qsize_t rsv_space = 0;
1688 struct dquot *transfer_from[MAXQUOTAS];
1689 struct dquot *transfer_to[MAXQUOTAS];
1690 int cnt, ret = 0;
1691 char warntype_to[MAXQUOTAS];
1692 char warntype_from_inodes[MAXQUOTAS], warntype_from_space[MAXQUOTAS];
1694 /* First test before acquiring mutex - solves deadlocks when we
1695 * re-enter the quota code and are already holding the mutex */
1696 if (IS_NOQUOTA(inode))
1697 return 0;
1698 /* Initialize the arrays */
1699 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1700 transfer_from[cnt] = NULL;
1701 transfer_to[cnt] = NULL;
1702 warntype_to[cnt] = QUOTA_NL_NOWARN;
1704 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1705 if (mask & (1 << cnt))
1706 transfer_to[cnt] = dqget(inode->i_sb, chid[cnt], cnt);
1708 down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1709 if (IS_NOQUOTA(inode)) { /* File without quota accounting? */
1710 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1711 goto put_all;
1713 spin_lock(&dq_data_lock);
1714 cur_space = inode_get_bytes(inode);
1715 rsv_space = inode_get_rsv_space(inode);
1716 space = cur_space + rsv_space;
1717 /* Build the transfer_from list and check the limits */
1718 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1719 if (!transfer_to[cnt])
1720 continue;
1721 transfer_from[cnt] = inode->i_dquot[cnt];
1722 ret = check_idq(transfer_to[cnt], 1, warntype_to + cnt);
1723 if (ret)
1724 goto over_quota;
1725 ret = check_bdq(transfer_to[cnt], space, 0, warntype_to + cnt);
1726 if (ret)
1727 goto over_quota;
1731 * Finally perform the needed transfer from transfer_from to transfer_to
1733 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1735 * Skip changes for same uid or gid or for turned off quota-type.
1737 if (!transfer_to[cnt])
1738 continue;
1740 /* Due to IO error we might not have transfer_from[] structure */
1741 if (transfer_from[cnt]) {
1742 warntype_from_inodes[cnt] =
1743 info_idq_free(transfer_from[cnt], 1);
1744 warntype_from_space[cnt] =
1745 info_bdq_free(transfer_from[cnt], space);
1746 dquot_decr_inodes(transfer_from[cnt], 1);
1747 dquot_decr_space(transfer_from[cnt], cur_space);
1748 dquot_free_reserved_space(transfer_from[cnt],
1749 rsv_space);
1752 dquot_incr_inodes(transfer_to[cnt], 1);
1753 dquot_incr_space(transfer_to[cnt], cur_space);
1754 dquot_resv_space(transfer_to[cnt], rsv_space);
1756 inode->i_dquot[cnt] = transfer_to[cnt];
1758 spin_unlock(&dq_data_lock);
1759 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1761 mark_all_dquot_dirty(transfer_from);
1762 mark_all_dquot_dirty(transfer_to);
1763 /* The reference we got is transferred to the inode */
1764 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1765 transfer_to[cnt] = NULL;
1766 warn_put_all:
1767 flush_warnings(transfer_to, warntype_to);
1768 flush_warnings(transfer_from, warntype_from_inodes);
1769 flush_warnings(transfer_from, warntype_from_space);
1770 put_all:
1771 dqput_all(transfer_from);
1772 dqput_all(transfer_to);
1773 return ret;
1774 over_quota:
1775 spin_unlock(&dq_data_lock);
1776 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1777 /* Clear dquot pointers we don't want to dqput() */
1778 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1779 transfer_from[cnt] = NULL;
1780 goto warn_put_all;
1783 /* Wrapper for transferring ownership of an inode for uid/gid only
1784 * Called from FSXXX_setattr()
1786 int dquot_transfer(struct inode *inode, struct iattr *iattr)
1788 qid_t chid[MAXQUOTAS];
1789 unsigned long mask = 0;
1791 if (iattr->ia_valid & ATTR_UID && iattr->ia_uid != inode->i_uid) {
1792 mask |= 1 << USRQUOTA;
1793 chid[USRQUOTA] = iattr->ia_uid;
1795 if (iattr->ia_valid & ATTR_GID && iattr->ia_gid != inode->i_gid) {
1796 mask |= 1 << GRPQUOTA;
1797 chid[GRPQUOTA] = iattr->ia_gid;
1799 if (sb_any_quota_active(inode->i_sb) && !IS_NOQUOTA(inode)) {
1800 dquot_initialize(inode);
1801 return __dquot_transfer(inode, chid, mask);
1803 return 0;
1805 EXPORT_SYMBOL(dquot_transfer);
1808 * Write info of quota file to disk
1810 int dquot_commit_info(struct super_block *sb, int type)
1812 int ret;
1813 struct quota_info *dqopt = sb_dqopt(sb);
1815 mutex_lock(&dqopt->dqio_mutex);
1816 ret = dqopt->ops[type]->write_file_info(sb, type);
1817 mutex_unlock(&dqopt->dqio_mutex);
1818 return ret;
1820 EXPORT_SYMBOL(dquot_commit_info);
1823 * Definitions of diskquota operations.
1825 const struct dquot_operations dquot_operations = {
1826 .write_dquot = dquot_commit,
1827 .acquire_dquot = dquot_acquire,
1828 .release_dquot = dquot_release,
1829 .mark_dirty = dquot_mark_dquot_dirty,
1830 .write_info = dquot_commit_info,
1831 .alloc_dquot = dquot_alloc,
1832 .destroy_dquot = dquot_destroy,
1836 * Generic helper for ->open on filesystems supporting disk quotas.
1838 int dquot_file_open(struct inode *inode, struct file *file)
1840 int error;
1842 error = generic_file_open(inode, file);
1843 if (!error && (file->f_mode & FMODE_WRITE))
1844 dquot_initialize(inode);
1845 return error;
1847 EXPORT_SYMBOL(dquot_file_open);
1850 * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
1852 int vfs_quota_disable(struct super_block *sb, int type, unsigned int flags)
1854 int cnt, ret = 0;
1855 struct quota_info *dqopt = sb_dqopt(sb);
1856 struct inode *toputinode[MAXQUOTAS];
1858 /* Cannot turn off usage accounting without turning off limits, or
1859 * suspend quotas and simultaneously turn quotas off. */
1860 if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
1861 || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
1862 DQUOT_USAGE_ENABLED)))
1863 return -EINVAL;
1865 /* We need to serialize quota_off() for device */
1866 mutex_lock(&dqopt->dqonoff_mutex);
1869 * Skip everything if there's nothing to do. We have to do this because
1870 * sometimes we are called when fill_super() failed and calling
1871 * sync_fs() in such cases does no good.
1873 if (!sb_any_quota_loaded(sb)) {
1874 mutex_unlock(&dqopt->dqonoff_mutex);
1875 return 0;
1877 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1878 toputinode[cnt] = NULL;
1879 if (type != -1 && cnt != type)
1880 continue;
1881 if (!sb_has_quota_loaded(sb, cnt))
1882 continue;
1884 if (flags & DQUOT_SUSPENDED) {
1885 spin_lock(&dq_state_lock);
1886 dqopt->flags |=
1887 dquot_state_flag(DQUOT_SUSPENDED, cnt);
1888 spin_unlock(&dq_state_lock);
1889 } else {
1890 spin_lock(&dq_state_lock);
1891 dqopt->flags &= ~dquot_state_flag(flags, cnt);
1892 /* Turning off suspended quotas? */
1893 if (!sb_has_quota_loaded(sb, cnt) &&
1894 sb_has_quota_suspended(sb, cnt)) {
1895 dqopt->flags &= ~dquot_state_flag(
1896 DQUOT_SUSPENDED, cnt);
1897 spin_unlock(&dq_state_lock);
1898 iput(dqopt->files[cnt]);
1899 dqopt->files[cnt] = NULL;
1900 continue;
1902 spin_unlock(&dq_state_lock);
1905 /* We still have to keep quota loaded? */
1906 if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
1907 continue;
1909 /* Note: these are blocking operations */
1910 drop_dquot_ref(sb, cnt);
1911 invalidate_dquots(sb, cnt);
1913 * Now all dquots should be invalidated, all writes done so we
1914 * should be only users of the info. No locks needed.
1916 if (info_dirty(&dqopt->info[cnt]))
1917 sb->dq_op->write_info(sb, cnt);
1918 if (dqopt->ops[cnt]->free_file_info)
1919 dqopt->ops[cnt]->free_file_info(sb, cnt);
1920 put_quota_format(dqopt->info[cnt].dqi_format);
1922 toputinode[cnt] = dqopt->files[cnt];
1923 if (!sb_has_quota_loaded(sb, cnt))
1924 dqopt->files[cnt] = NULL;
1925 dqopt->info[cnt].dqi_flags = 0;
1926 dqopt->info[cnt].dqi_igrace = 0;
1927 dqopt->info[cnt].dqi_bgrace = 0;
1928 dqopt->ops[cnt] = NULL;
1930 mutex_unlock(&dqopt->dqonoff_mutex);
1932 /* Skip syncing and setting flags if quota files are hidden */
1933 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
1934 goto put_inodes;
1936 /* Sync the superblock so that buffers with quota data are written to
1937 * disk (and so userspace sees correct data afterwards). */
1938 if (sb->s_op->sync_fs)
1939 sb->s_op->sync_fs(sb, 1);
1940 sync_blockdev(sb->s_bdev);
1941 /* Now the quota files are just ordinary files and we can set the
1942 * inode flags back. Moreover we discard the pagecache so that
1943 * userspace sees the writes we did bypassing the pagecache. We
1944 * must also discard the blockdev buffers so that we see the
1945 * changes done by userspace on the next quotaon() */
1946 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1947 if (toputinode[cnt]) {
1948 mutex_lock(&dqopt->dqonoff_mutex);
1949 /* If quota was reenabled in the meantime, we have
1950 * nothing to do */
1951 if (!sb_has_quota_loaded(sb, cnt)) {
1952 mutex_lock_nested(&toputinode[cnt]->i_mutex,
1953 I_MUTEX_QUOTA);
1954 toputinode[cnt]->i_flags &= ~(S_IMMUTABLE |
1955 S_NOATIME | S_NOQUOTA);
1956 truncate_inode_pages(&toputinode[cnt]->i_data,
1958 mutex_unlock(&toputinode[cnt]->i_mutex);
1959 mark_inode_dirty(toputinode[cnt]);
1961 mutex_unlock(&dqopt->dqonoff_mutex);
1963 if (sb->s_bdev)
1964 invalidate_bdev(sb->s_bdev);
1965 put_inodes:
1966 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1967 if (toputinode[cnt]) {
1968 /* On remount RO, we keep the inode pointer so that we
1969 * can reenable quota on the subsequent remount RW. We
1970 * have to check 'flags' variable and not use sb_has_
1971 * function because another quotaon / quotaoff could
1972 * change global state before we got here. We refuse
1973 * to suspend quotas when there is pending delete on
1974 * the quota file... */
1975 if (!(flags & DQUOT_SUSPENDED))
1976 iput(toputinode[cnt]);
1977 else if (!toputinode[cnt]->i_nlink)
1978 ret = -EBUSY;
1980 return ret;
1982 EXPORT_SYMBOL(vfs_quota_disable);
1984 int vfs_quota_off(struct super_block *sb, int type, int remount)
1986 return vfs_quota_disable(sb, type, remount ? DQUOT_SUSPENDED :
1987 (DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED));
1989 EXPORT_SYMBOL(vfs_quota_off);
1991 * Turn quotas on on a device
1995 * Helper function to turn quotas on when we already have the inode of
1996 * quota file and no quota information is loaded.
1998 static int vfs_load_quota_inode(struct inode *inode, int type, int format_id,
1999 unsigned int flags)
2001 struct quota_format_type *fmt = find_quota_format(format_id);
2002 struct super_block *sb = inode->i_sb;
2003 struct quota_info *dqopt = sb_dqopt(sb);
2004 int error;
2005 int oldflags = -1;
2007 if (!fmt)
2008 return -ESRCH;
2009 if (!S_ISREG(inode->i_mode)) {
2010 error = -EACCES;
2011 goto out_fmt;
2013 if (IS_RDONLY(inode)) {
2014 error = -EROFS;
2015 goto out_fmt;
2017 if (!sb->s_op->quota_write || !sb->s_op->quota_read) {
2018 error = -EINVAL;
2019 goto out_fmt;
2021 /* Usage always has to be set... */
2022 if (!(flags & DQUOT_USAGE_ENABLED)) {
2023 error = -EINVAL;
2024 goto out_fmt;
2027 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2028 /* As we bypass the pagecache we must now flush all the
2029 * dirty data and invalidate caches so that kernel sees
2030 * changes from userspace. It is not enough to just flush
2031 * the quota file since if blocksize < pagesize, invalidation
2032 * of the cache could fail because of other unrelated dirty
2033 * data */
2034 sync_filesystem(sb);
2035 invalidate_bdev(sb->s_bdev);
2037 mutex_lock(&dqopt->dqonoff_mutex);
2038 if (sb_has_quota_loaded(sb, type)) {
2039 error = -EBUSY;
2040 goto out_lock;
2043 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2044 /* We don't want quota and atime on quota files (deadlocks
2045 * possible) Also nobody should write to the file - we use
2046 * special IO operations which ignore the immutable bit. */
2047 mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
2048 oldflags = inode->i_flags & (S_NOATIME | S_IMMUTABLE |
2049 S_NOQUOTA);
2050 inode->i_flags |= S_NOQUOTA | S_NOATIME | S_IMMUTABLE;
2051 mutex_unlock(&inode->i_mutex);
2053 * When S_NOQUOTA is set, remove dquot references as no more
2054 * references can be added
2056 __dquot_drop(inode);
2059 error = -EIO;
2060 dqopt->files[type] = igrab(inode);
2061 if (!dqopt->files[type])
2062 goto out_lock;
2063 error = -EINVAL;
2064 if (!fmt->qf_ops->check_quota_file(sb, type))
2065 goto out_file_init;
2067 dqopt->ops[type] = fmt->qf_ops;
2068 dqopt->info[type].dqi_format = fmt;
2069 dqopt->info[type].dqi_fmt_id = format_id;
2070 INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2071 mutex_lock(&dqopt->dqio_mutex);
2072 error = dqopt->ops[type]->read_file_info(sb, type);
2073 if (error < 0) {
2074 mutex_unlock(&dqopt->dqio_mutex);
2075 goto out_file_init;
2077 mutex_unlock(&dqopt->dqio_mutex);
2078 spin_lock(&dq_state_lock);
2079 dqopt->flags |= dquot_state_flag(flags, type);
2080 spin_unlock(&dq_state_lock);
2082 add_dquot_ref(sb, type);
2083 mutex_unlock(&dqopt->dqonoff_mutex);
2085 return 0;
2087 out_file_init:
2088 dqopt->files[type] = NULL;
2089 iput(inode);
2090 out_lock:
2091 if (oldflags != -1) {
2092 mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
2093 /* Set the flags back (in the case of accidental quotaon()
2094 * on a wrong file we don't want to mess up the flags) */
2095 inode->i_flags &= ~(S_NOATIME | S_NOQUOTA | S_IMMUTABLE);
2096 inode->i_flags |= oldflags;
2097 mutex_unlock(&inode->i_mutex);
2099 mutex_unlock(&dqopt->dqonoff_mutex);
2100 out_fmt:
2101 put_quota_format(fmt);
2103 return error;
2106 /* Reenable quotas on remount RW */
2107 static int vfs_quota_on_remount(struct super_block *sb, int type)
2109 struct quota_info *dqopt = sb_dqopt(sb);
2110 struct inode *inode;
2111 int ret;
2112 unsigned int flags;
2114 mutex_lock(&dqopt->dqonoff_mutex);
2115 if (!sb_has_quota_suspended(sb, type)) {
2116 mutex_unlock(&dqopt->dqonoff_mutex);
2117 return 0;
2119 inode = dqopt->files[type];
2120 dqopt->files[type] = NULL;
2121 spin_lock(&dq_state_lock);
2122 flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2123 DQUOT_LIMITS_ENABLED, type);
2124 dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, type);
2125 spin_unlock(&dq_state_lock);
2126 mutex_unlock(&dqopt->dqonoff_mutex);
2128 flags = dquot_generic_flag(flags, type);
2129 ret = vfs_load_quota_inode(inode, type, dqopt->info[type].dqi_fmt_id,
2130 flags);
2131 iput(inode);
2133 return ret;
2136 int vfs_quota_on_path(struct super_block *sb, int type, int format_id,
2137 struct path *path)
2139 int error = security_quota_on(path->dentry);
2140 if (error)
2141 return error;
2142 /* Quota file not on the same filesystem? */
2143 if (path->mnt->mnt_sb != sb)
2144 error = -EXDEV;
2145 else
2146 error = vfs_load_quota_inode(path->dentry->d_inode, type,
2147 format_id, DQUOT_USAGE_ENABLED |
2148 DQUOT_LIMITS_ENABLED);
2149 return error;
2151 EXPORT_SYMBOL(vfs_quota_on_path);
2153 int vfs_quota_on(struct super_block *sb, int type, int format_id, char *name,
2154 int remount)
2156 struct path path;
2157 int error;
2159 if (remount)
2160 return vfs_quota_on_remount(sb, type);
2162 error = kern_path(name, LOOKUP_FOLLOW, &path);
2163 if (!error) {
2164 error = vfs_quota_on_path(sb, type, format_id, &path);
2165 path_put(&path);
2167 return error;
2169 EXPORT_SYMBOL(vfs_quota_on);
2172 * More powerful function for turning on quotas allowing setting
2173 * of individual quota flags
2175 int vfs_quota_enable(struct inode *inode, int type, int format_id,
2176 unsigned int flags)
2178 int ret = 0;
2179 struct super_block *sb = inode->i_sb;
2180 struct quota_info *dqopt = sb_dqopt(sb);
2182 /* Just unsuspend quotas? */
2183 if (flags & DQUOT_SUSPENDED)
2184 return vfs_quota_on_remount(sb, type);
2185 if (!flags)
2186 return 0;
2187 /* Just updating flags needed? */
2188 if (sb_has_quota_loaded(sb, type)) {
2189 mutex_lock(&dqopt->dqonoff_mutex);
2190 /* Now do a reliable test... */
2191 if (!sb_has_quota_loaded(sb, type)) {
2192 mutex_unlock(&dqopt->dqonoff_mutex);
2193 goto load_quota;
2195 if (flags & DQUOT_USAGE_ENABLED &&
2196 sb_has_quota_usage_enabled(sb, type)) {
2197 ret = -EBUSY;
2198 goto out_lock;
2200 if (flags & DQUOT_LIMITS_ENABLED &&
2201 sb_has_quota_limits_enabled(sb, type)) {
2202 ret = -EBUSY;
2203 goto out_lock;
2205 spin_lock(&dq_state_lock);
2206 sb_dqopt(sb)->flags |= dquot_state_flag(flags, type);
2207 spin_unlock(&dq_state_lock);
2208 out_lock:
2209 mutex_unlock(&dqopt->dqonoff_mutex);
2210 return ret;
2213 load_quota:
2214 return vfs_load_quota_inode(inode, type, format_id, flags);
2216 EXPORT_SYMBOL(vfs_quota_enable);
2219 * This function is used when filesystem needs to initialize quotas
2220 * during mount time.
2222 int vfs_quota_on_mount(struct super_block *sb, char *qf_name,
2223 int format_id, int type)
2225 struct dentry *dentry;
2226 int error;
2228 mutex_lock(&sb->s_root->d_inode->i_mutex);
2229 dentry = lookup_one_len(qf_name, sb->s_root, strlen(qf_name));
2230 mutex_unlock(&sb->s_root->d_inode->i_mutex);
2231 if (IS_ERR(dentry))
2232 return PTR_ERR(dentry);
2234 if (!dentry->d_inode) {
2235 error = -ENOENT;
2236 goto out;
2239 error = security_quota_on(dentry);
2240 if (!error)
2241 error = vfs_load_quota_inode(dentry->d_inode, type, format_id,
2242 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2244 out:
2245 dput(dentry);
2246 return error;
2248 EXPORT_SYMBOL(vfs_quota_on_mount);
2250 /* Wrapper to turn on quotas when remounting rw */
2251 int vfs_dq_quota_on_remount(struct super_block *sb)
2253 int cnt;
2254 int ret = 0, err;
2256 if (!sb->s_qcop || !sb->s_qcop->quota_on)
2257 return -ENOSYS;
2258 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2259 err = sb->s_qcop->quota_on(sb, cnt, 0, NULL, 1);
2260 if (err < 0 && !ret)
2261 ret = err;
2263 return ret;
2265 EXPORT_SYMBOL(vfs_dq_quota_on_remount);
2267 static inline qsize_t qbtos(qsize_t blocks)
2269 return blocks << QIF_DQBLKSIZE_BITS;
2272 static inline qsize_t stoqb(qsize_t space)
2274 return (space + QIF_DQBLKSIZE - 1) >> QIF_DQBLKSIZE_BITS;
2277 /* Generic routine for getting common part of quota structure */
2278 static void do_get_dqblk(struct dquot *dquot, struct if_dqblk *di)
2280 struct mem_dqblk *dm = &dquot->dq_dqb;
2282 spin_lock(&dq_data_lock);
2283 di->dqb_bhardlimit = stoqb(dm->dqb_bhardlimit);
2284 di->dqb_bsoftlimit = stoqb(dm->dqb_bsoftlimit);
2285 di->dqb_curspace = dm->dqb_curspace + dm->dqb_rsvspace;
2286 di->dqb_ihardlimit = dm->dqb_ihardlimit;
2287 di->dqb_isoftlimit = dm->dqb_isoftlimit;
2288 di->dqb_curinodes = dm->dqb_curinodes;
2289 di->dqb_btime = dm->dqb_btime;
2290 di->dqb_itime = dm->dqb_itime;
2291 di->dqb_valid = QIF_ALL;
2292 spin_unlock(&dq_data_lock);
2295 int vfs_get_dqblk(struct super_block *sb, int type, qid_t id,
2296 struct if_dqblk *di)
2298 struct dquot *dquot;
2300 dquot = dqget(sb, id, type);
2301 if (!dquot)
2302 return -ESRCH;
2303 do_get_dqblk(dquot, di);
2304 dqput(dquot);
2306 return 0;
2308 EXPORT_SYMBOL(vfs_get_dqblk);
2310 /* Generic routine for setting common part of quota structure */
2311 static int do_set_dqblk(struct dquot *dquot, struct if_dqblk *di)
2313 struct mem_dqblk *dm = &dquot->dq_dqb;
2314 int check_blim = 0, check_ilim = 0;
2315 struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_type];
2317 if ((di->dqb_valid & QIF_BLIMITS &&
2318 (di->dqb_bhardlimit > dqi->dqi_maxblimit ||
2319 di->dqb_bsoftlimit > dqi->dqi_maxblimit)) ||
2320 (di->dqb_valid & QIF_ILIMITS &&
2321 (di->dqb_ihardlimit > dqi->dqi_maxilimit ||
2322 di->dqb_isoftlimit > dqi->dqi_maxilimit)))
2323 return -ERANGE;
2325 spin_lock(&dq_data_lock);
2326 if (di->dqb_valid & QIF_SPACE) {
2327 dm->dqb_curspace = di->dqb_curspace - dm->dqb_rsvspace;
2328 check_blim = 1;
2329 set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2331 if (di->dqb_valid & QIF_BLIMITS) {
2332 dm->dqb_bsoftlimit = qbtos(di->dqb_bsoftlimit);
2333 dm->dqb_bhardlimit = qbtos(di->dqb_bhardlimit);
2334 check_blim = 1;
2335 set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2337 if (di->dqb_valid & QIF_INODES) {
2338 dm->dqb_curinodes = di->dqb_curinodes;
2339 check_ilim = 1;
2340 set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2342 if (di->dqb_valid & QIF_ILIMITS) {
2343 dm->dqb_isoftlimit = di->dqb_isoftlimit;
2344 dm->dqb_ihardlimit = di->dqb_ihardlimit;
2345 check_ilim = 1;
2346 set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2348 if (di->dqb_valid & QIF_BTIME) {
2349 dm->dqb_btime = di->dqb_btime;
2350 check_blim = 1;
2351 set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2353 if (di->dqb_valid & QIF_ITIME) {
2354 dm->dqb_itime = di->dqb_itime;
2355 check_ilim = 1;
2356 set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2359 if (check_blim) {
2360 if (!dm->dqb_bsoftlimit ||
2361 dm->dqb_curspace < dm->dqb_bsoftlimit) {
2362 dm->dqb_btime = 0;
2363 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2364 } else if (!(di->dqb_valid & QIF_BTIME))
2365 /* Set grace only if user hasn't provided his own... */
2366 dm->dqb_btime = get_seconds() + dqi->dqi_bgrace;
2368 if (check_ilim) {
2369 if (!dm->dqb_isoftlimit ||
2370 dm->dqb_curinodes < dm->dqb_isoftlimit) {
2371 dm->dqb_itime = 0;
2372 clear_bit(DQ_INODES_B, &dquot->dq_flags);
2373 } else if (!(di->dqb_valid & QIF_ITIME))
2374 /* Set grace only if user hasn't provided his own... */
2375 dm->dqb_itime = get_seconds() + dqi->dqi_igrace;
2377 if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2378 dm->dqb_isoftlimit)
2379 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2380 else
2381 set_bit(DQ_FAKE_B, &dquot->dq_flags);
2382 spin_unlock(&dq_data_lock);
2383 mark_dquot_dirty(dquot);
2385 return 0;
2388 int vfs_set_dqblk(struct super_block *sb, int type, qid_t id,
2389 struct if_dqblk *di)
2391 struct dquot *dquot;
2392 int rc;
2394 dquot = dqget(sb, id, type);
2395 if (!dquot) {
2396 rc = -ESRCH;
2397 goto out;
2399 rc = do_set_dqblk(dquot, di);
2400 dqput(dquot);
2401 out:
2402 return rc;
2404 EXPORT_SYMBOL(vfs_set_dqblk);
2406 /* Generic routine for getting common part of quota file information */
2407 int vfs_get_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
2409 struct mem_dqinfo *mi;
2411 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
2412 if (!sb_has_quota_active(sb, type)) {
2413 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2414 return -ESRCH;
2416 mi = sb_dqopt(sb)->info + type;
2417 spin_lock(&dq_data_lock);
2418 ii->dqi_bgrace = mi->dqi_bgrace;
2419 ii->dqi_igrace = mi->dqi_igrace;
2420 ii->dqi_flags = mi->dqi_flags & DQF_MASK;
2421 ii->dqi_valid = IIF_ALL;
2422 spin_unlock(&dq_data_lock);
2423 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2424 return 0;
2426 EXPORT_SYMBOL(vfs_get_dqinfo);
2428 /* Generic routine for setting common part of quota file information */
2429 int vfs_set_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
2431 struct mem_dqinfo *mi;
2432 int err = 0;
2434 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
2435 if (!sb_has_quota_active(sb, type)) {
2436 err = -ESRCH;
2437 goto out;
2439 mi = sb_dqopt(sb)->info + type;
2440 spin_lock(&dq_data_lock);
2441 if (ii->dqi_valid & IIF_BGRACE)
2442 mi->dqi_bgrace = ii->dqi_bgrace;
2443 if (ii->dqi_valid & IIF_IGRACE)
2444 mi->dqi_igrace = ii->dqi_igrace;
2445 if (ii->dqi_valid & IIF_FLAGS)
2446 mi->dqi_flags = (mi->dqi_flags & ~DQF_MASK) |
2447 (ii->dqi_flags & DQF_MASK);
2448 spin_unlock(&dq_data_lock);
2449 mark_info_dirty(sb, type);
2450 /* Force write to disk */
2451 sb->dq_op->write_info(sb, type);
2452 out:
2453 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2454 return err;
2456 EXPORT_SYMBOL(vfs_set_dqinfo);
2458 const struct quotactl_ops vfs_quotactl_ops = {
2459 .quota_on = vfs_quota_on,
2460 .quota_off = vfs_quota_off,
2461 .quota_sync = vfs_quota_sync,
2462 .get_info = vfs_get_dqinfo,
2463 .set_info = vfs_set_dqinfo,
2464 .get_dqblk = vfs_get_dqblk,
2465 .set_dqblk = vfs_set_dqblk
2468 static ctl_table fs_dqstats_table[] = {
2470 .procname = "lookups",
2471 .data = &dqstats.lookups,
2472 .maxlen = sizeof(int),
2473 .mode = 0444,
2474 .proc_handler = proc_dointvec,
2477 .procname = "drops",
2478 .data = &dqstats.drops,
2479 .maxlen = sizeof(int),
2480 .mode = 0444,
2481 .proc_handler = proc_dointvec,
2484 .procname = "reads",
2485 .data = &dqstats.reads,
2486 .maxlen = sizeof(int),
2487 .mode = 0444,
2488 .proc_handler = proc_dointvec,
2491 .procname = "writes",
2492 .data = &dqstats.writes,
2493 .maxlen = sizeof(int),
2494 .mode = 0444,
2495 .proc_handler = proc_dointvec,
2498 .procname = "cache_hits",
2499 .data = &dqstats.cache_hits,
2500 .maxlen = sizeof(int),
2501 .mode = 0444,
2502 .proc_handler = proc_dointvec,
2505 .procname = "allocated_dquots",
2506 .data = &dqstats.allocated_dquots,
2507 .maxlen = sizeof(int),
2508 .mode = 0444,
2509 .proc_handler = proc_dointvec,
2512 .procname = "free_dquots",
2513 .data = &dqstats.free_dquots,
2514 .maxlen = sizeof(int),
2515 .mode = 0444,
2516 .proc_handler = proc_dointvec,
2519 .procname = "syncs",
2520 .data = &dqstats.syncs,
2521 .maxlen = sizeof(int),
2522 .mode = 0444,
2523 .proc_handler = proc_dointvec,
2525 #ifdef CONFIG_PRINT_QUOTA_WARNING
2527 .procname = "warnings",
2528 .data = &flag_print_warnings,
2529 .maxlen = sizeof(int),
2530 .mode = 0644,
2531 .proc_handler = proc_dointvec,
2533 #endif
2534 { },
2537 static ctl_table fs_table[] = {
2539 .procname = "quota",
2540 .mode = 0555,
2541 .child = fs_dqstats_table,
2543 { },
2546 static ctl_table sys_table[] = {
2548 .procname = "fs",
2549 .mode = 0555,
2550 .child = fs_table,
2552 { },
2555 static int __init dquot_init(void)
2557 int i;
2558 unsigned long nr_hash, order;
2560 printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2562 register_sysctl_table(sys_table);
2564 dquot_cachep = kmem_cache_create("dquot",
2565 sizeof(struct dquot), sizeof(unsigned long) * 4,
2566 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2567 SLAB_MEM_SPREAD|SLAB_PANIC),
2568 NULL);
2570 order = 0;
2571 dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
2572 if (!dquot_hash)
2573 panic("Cannot create dquot hash table");
2575 /* Find power-of-two hlist_heads which can fit into allocation */
2576 nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2577 dq_hash_bits = 0;
2578 do {
2579 dq_hash_bits++;
2580 } while (nr_hash >> dq_hash_bits);
2581 dq_hash_bits--;
2583 nr_hash = 1UL << dq_hash_bits;
2584 dq_hash_mask = nr_hash - 1;
2585 for (i = 0; i < nr_hash; i++)
2586 INIT_HLIST_HEAD(dquot_hash + i);
2588 printk("Dquot-cache hash table entries: %ld (order %ld, %ld bytes)\n",
2589 nr_hash, order, (PAGE_SIZE << order));
2591 register_shrinker(&dqcache_shrinker);
2593 return 0;
2595 module_init(dquot_init);