mfd: Don't allow WM8350 to be built modular
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / quota / dquot.c
blobe0b870f4749f2ad01abab80d56e66759359f43c2
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>
83 #define __DQUOT_PARANOIA
86 * There are three quota SMP locks. dq_list_lock protects all lists with quotas
87 * and quota formats, dqstats structure containing statistics about the lists
88 * dq_data_lock protects data from dq_dqb and also mem_dqinfo structures and
89 * also guards consistency of dquot->dq_dqb with inode->i_blocks, i_bytes.
90 * i_blocks and i_bytes updates itself are guarded by i_lock acquired directly
91 * in inode_add_bytes() and inode_sub_bytes(). dq_state_lock protects
92 * modifications of quota state (on quotaon and quotaoff) and readers who care
93 * about latest values take it as well.
95 * The spinlock ordering is hence: dq_data_lock > dq_list_lock > i_lock,
96 * dq_list_lock > dq_state_lock
98 * Note that some things (eg. sb pointer, type, id) doesn't change during
99 * the life of the dquot structure and so needn't to be protected by a lock
101 * Any operation working on dquots via inode pointers must hold dqptr_sem. If
102 * operation is just reading pointers from inode (or not using them at all) the
103 * read lock is enough. If pointers are altered function must hold write lock.
104 * Special care needs to be taken about S_NOQUOTA inode flag (marking that
105 * inode is a quota file). Functions adding pointers from inode to dquots have
106 * to check this flag under dqptr_sem and then (if S_NOQUOTA is not set) they
107 * have to do all pointer modifications before dropping dqptr_sem. This makes
108 * sure they cannot race with quotaon which first sets S_NOQUOTA flag and
109 * then drops all pointers to dquots from an inode.
111 * Each dquot has its dq_lock mutex. Locked dquots might not be referenced
112 * from inodes (dquot_alloc_space() and such don't check the dq_lock).
113 * Currently dquot is locked only when it is being read to memory (or space for
114 * it is being allocated) on the first dqget() and when it is being released on
115 * the last dqput(). The allocation and release oparations are serialized by
116 * the dq_lock and by checking the use count in dquot_release(). Write
117 * operations on dquots don't hold dq_lock as they copy data under dq_data_lock
118 * spinlock to internal buffers before writing.
120 * Lock ordering (including related VFS locks) is the following:
121 * i_mutex > dqonoff_sem > journal_lock > dqptr_sem > dquot->dq_lock >
122 * dqio_mutex
123 * The lock ordering of dqptr_sem imposed by quota code is only dqonoff_sem >
124 * dqptr_sem. But filesystem has to count with the fact that functions such as
125 * dquot_alloc_space() acquire dqptr_sem and they usually have to be called
126 * from inside a transaction to keep filesystem consistency after a crash. Also
127 * filesystems usually want to do some IO on dquot from ->mark_dirty which is
128 * called with dqptr_sem held.
129 * i_mutex on quota files is special (it's below dqio_mutex)
132 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_list_lock);
133 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_state_lock);
134 __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_data_lock);
135 EXPORT_SYMBOL(dq_data_lock);
137 static char *quotatypes[] = INITQFNAMES;
138 static struct quota_format_type *quota_formats; /* List of registered formats */
139 static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
141 /* SLAB cache for dquot structures */
142 static struct kmem_cache *dquot_cachep;
144 int register_quota_format(struct quota_format_type *fmt)
146 spin_lock(&dq_list_lock);
147 fmt->qf_next = quota_formats;
148 quota_formats = fmt;
149 spin_unlock(&dq_list_lock);
150 return 0;
152 EXPORT_SYMBOL(register_quota_format);
154 void unregister_quota_format(struct quota_format_type *fmt)
156 struct quota_format_type **actqf;
158 spin_lock(&dq_list_lock);
159 for (actqf = &quota_formats; *actqf && *actqf != fmt;
160 actqf = &(*actqf)->qf_next)
162 if (*actqf)
163 *actqf = (*actqf)->qf_next;
164 spin_unlock(&dq_list_lock);
166 EXPORT_SYMBOL(unregister_quota_format);
168 static struct quota_format_type *find_quota_format(int id)
170 struct quota_format_type *actqf;
172 spin_lock(&dq_list_lock);
173 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
174 actqf = actqf->qf_next)
176 if (!actqf || !try_module_get(actqf->qf_owner)) {
177 int qm;
179 spin_unlock(&dq_list_lock);
181 for (qm = 0; module_names[qm].qm_fmt_id &&
182 module_names[qm].qm_fmt_id != id; qm++)
184 if (!module_names[qm].qm_fmt_id ||
185 request_module(module_names[qm].qm_mod_name))
186 return NULL;
188 spin_lock(&dq_list_lock);
189 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
190 actqf = actqf->qf_next)
192 if (actqf && !try_module_get(actqf->qf_owner))
193 actqf = NULL;
195 spin_unlock(&dq_list_lock);
196 return actqf;
199 static void put_quota_format(struct quota_format_type *fmt)
201 module_put(fmt->qf_owner);
205 * Dquot List Management:
206 * The quota code uses three lists for dquot management: the inuse_list,
207 * free_dquots, and dquot_hash[] array. A single dquot structure may be
208 * on all three lists, depending on its current state.
210 * All dquots are placed to the end of inuse_list when first created, and this
211 * list is used for invalidate operation, which must look at every dquot.
213 * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
214 * and this list is searched whenever we need an available dquot. Dquots are
215 * removed from the list as soon as they are used again, and
216 * dqstats.free_dquots gives the number of dquots on the list. When
217 * dquot is invalidated it's completely released from memory.
219 * Dquots with a specific identity (device, type and id) are placed on
220 * one of the dquot_hash[] hash chains. The provides an efficient search
221 * mechanism to locate a specific dquot.
224 static LIST_HEAD(inuse_list);
225 static LIST_HEAD(free_dquots);
226 static unsigned int dq_hash_bits, dq_hash_mask;
227 static struct hlist_head *dquot_hash;
229 struct dqstats dqstats;
230 EXPORT_SYMBOL(dqstats);
232 static qsize_t inode_get_rsv_space(struct inode *inode);
233 static void __dquot_initialize(struct inode *inode, int type);
235 static inline unsigned int
236 hashfn(const struct super_block *sb, unsigned int id, int type)
238 unsigned long tmp;
240 tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
241 return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
245 * Following list functions expect dq_list_lock to be held
247 static inline void insert_dquot_hash(struct dquot *dquot)
249 struct hlist_head *head;
250 head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id, dquot->dq_type);
251 hlist_add_head(&dquot->dq_hash, head);
254 static inline void remove_dquot_hash(struct dquot *dquot)
256 hlist_del_init(&dquot->dq_hash);
259 static struct dquot *find_dquot(unsigned int hashent, struct super_block *sb,
260 unsigned int id, int type)
262 struct hlist_node *node;
263 struct dquot *dquot;
265 hlist_for_each (node, dquot_hash+hashent) {
266 dquot = hlist_entry(node, struct dquot, dq_hash);
267 if (dquot->dq_sb == sb && dquot->dq_id == id &&
268 dquot->dq_type == type)
269 return dquot;
271 return NULL;
274 /* Add a dquot to the tail of the free list */
275 static inline void put_dquot_last(struct dquot *dquot)
277 list_add_tail(&dquot->dq_free, &free_dquots);
278 dqstats.free_dquots++;
281 static inline void remove_free_dquot(struct dquot *dquot)
283 if (list_empty(&dquot->dq_free))
284 return;
285 list_del_init(&dquot->dq_free);
286 dqstats.free_dquots--;
289 static inline void put_inuse(struct dquot *dquot)
291 /* We add to the back of inuse list so we don't have to restart
292 * when traversing this list and we block */
293 list_add_tail(&dquot->dq_inuse, &inuse_list);
294 dqstats.allocated_dquots++;
297 static inline void remove_inuse(struct dquot *dquot)
299 dqstats.allocated_dquots--;
300 list_del(&dquot->dq_inuse);
303 * End of list functions needing dq_list_lock
306 static void wait_on_dquot(struct dquot *dquot)
308 mutex_lock(&dquot->dq_lock);
309 mutex_unlock(&dquot->dq_lock);
312 static inline int dquot_dirty(struct dquot *dquot)
314 return test_bit(DQ_MOD_B, &dquot->dq_flags);
317 static inline int mark_dquot_dirty(struct dquot *dquot)
319 return dquot->dq_sb->dq_op->mark_dirty(dquot);
322 int dquot_mark_dquot_dirty(struct dquot *dquot)
324 spin_lock(&dq_list_lock);
325 if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags))
326 list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
327 info[dquot->dq_type].dqi_dirty_list);
328 spin_unlock(&dq_list_lock);
329 return 0;
331 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
333 /* Dirtify all the dquots - this can block when journalling */
334 static inline int mark_all_dquot_dirty(struct dquot * const *dquot)
336 int ret, err, cnt;
338 ret = err = 0;
339 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
340 if (dquot[cnt])
341 /* Even in case of error we have to continue */
342 ret = mark_dquot_dirty(dquot[cnt]);
343 if (!err)
344 err = ret;
346 return err;
349 static inline void dqput_all(struct dquot **dquot)
351 unsigned int cnt;
353 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
354 dqput(dquot[cnt]);
357 /* This function needs dq_list_lock */
358 static inline int clear_dquot_dirty(struct dquot *dquot)
360 if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags))
361 return 0;
362 list_del_init(&dquot->dq_dirty);
363 return 1;
366 void mark_info_dirty(struct super_block *sb, int type)
368 set_bit(DQF_INFO_DIRTY_B, &sb_dqopt(sb)->info[type].dqi_flags);
370 EXPORT_SYMBOL(mark_info_dirty);
373 * Read dquot from disk and alloc space for it
376 int dquot_acquire(struct dquot *dquot)
378 int ret = 0, ret2 = 0;
379 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
381 mutex_lock(&dquot->dq_lock);
382 mutex_lock(&dqopt->dqio_mutex);
383 if (!test_bit(DQ_READ_B, &dquot->dq_flags))
384 ret = dqopt->ops[dquot->dq_type]->read_dqblk(dquot);
385 if (ret < 0)
386 goto out_iolock;
387 set_bit(DQ_READ_B, &dquot->dq_flags);
388 /* Instantiate dquot if needed */
389 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
390 ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
391 /* Write the info if needed */
392 if (info_dirty(&dqopt->info[dquot->dq_type])) {
393 ret2 = dqopt->ops[dquot->dq_type]->write_file_info(
394 dquot->dq_sb, dquot->dq_type);
396 if (ret < 0)
397 goto out_iolock;
398 if (ret2 < 0) {
399 ret = ret2;
400 goto out_iolock;
403 set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
404 out_iolock:
405 mutex_unlock(&dqopt->dqio_mutex);
406 mutex_unlock(&dquot->dq_lock);
407 return ret;
409 EXPORT_SYMBOL(dquot_acquire);
412 * Write dquot to disk
414 int dquot_commit(struct dquot *dquot)
416 int ret = 0, ret2 = 0;
417 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
419 mutex_lock(&dqopt->dqio_mutex);
420 spin_lock(&dq_list_lock);
421 if (!clear_dquot_dirty(dquot)) {
422 spin_unlock(&dq_list_lock);
423 goto out_sem;
425 spin_unlock(&dq_list_lock);
426 /* Inactive dquot can be only if there was error during read/init
427 * => we have better not writing it */
428 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
429 ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
430 if (info_dirty(&dqopt->info[dquot->dq_type])) {
431 ret2 = dqopt->ops[dquot->dq_type]->write_file_info(
432 dquot->dq_sb, dquot->dq_type);
434 if (ret >= 0)
435 ret = ret2;
437 out_sem:
438 mutex_unlock(&dqopt->dqio_mutex);
439 return ret;
441 EXPORT_SYMBOL(dquot_commit);
444 * Release dquot
446 int dquot_release(struct dquot *dquot)
448 int ret = 0, ret2 = 0;
449 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
451 mutex_lock(&dquot->dq_lock);
452 /* Check whether we are not racing with some other dqget() */
453 if (atomic_read(&dquot->dq_count) > 1)
454 goto out_dqlock;
455 mutex_lock(&dqopt->dqio_mutex);
456 if (dqopt->ops[dquot->dq_type]->release_dqblk) {
457 ret = dqopt->ops[dquot->dq_type]->release_dqblk(dquot);
458 /* Write the info */
459 if (info_dirty(&dqopt->info[dquot->dq_type])) {
460 ret2 = dqopt->ops[dquot->dq_type]->write_file_info(
461 dquot->dq_sb, dquot->dq_type);
463 if (ret >= 0)
464 ret = ret2;
466 clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
467 mutex_unlock(&dqopt->dqio_mutex);
468 out_dqlock:
469 mutex_unlock(&dquot->dq_lock);
470 return ret;
472 EXPORT_SYMBOL(dquot_release);
474 void dquot_destroy(struct dquot *dquot)
476 kmem_cache_free(dquot_cachep, dquot);
478 EXPORT_SYMBOL(dquot_destroy);
480 static inline void do_destroy_dquot(struct dquot *dquot)
482 dquot->dq_sb->dq_op->destroy_dquot(dquot);
485 /* Invalidate all dquots on the list. Note that this function is called after
486 * quota is disabled and pointers from inodes removed so there cannot be new
487 * quota users. There can still be some users of quotas due to inodes being
488 * just deleted or pruned by prune_icache() (those are not attached to any
489 * list) or parallel quotactl call. We have to wait for such users.
491 static void invalidate_dquots(struct super_block *sb, int type)
493 struct dquot *dquot, *tmp;
495 restart:
496 spin_lock(&dq_list_lock);
497 list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
498 if (dquot->dq_sb != sb)
499 continue;
500 if (dquot->dq_type != type)
501 continue;
502 /* Wait for dquot users */
503 if (atomic_read(&dquot->dq_count)) {
504 DEFINE_WAIT(wait);
506 atomic_inc(&dquot->dq_count);
507 prepare_to_wait(&dquot->dq_wait_unused, &wait,
508 TASK_UNINTERRUPTIBLE);
509 spin_unlock(&dq_list_lock);
510 /* Once dqput() wakes us up, we know it's time to free
511 * the dquot.
512 * IMPORTANT: we rely on the fact that there is always
513 * at most one process waiting for dquot to free.
514 * Otherwise dq_count would be > 1 and we would never
515 * wake up.
517 if (atomic_read(&dquot->dq_count) > 1)
518 schedule();
519 finish_wait(&dquot->dq_wait_unused, &wait);
520 dqput(dquot);
521 /* At this moment dquot() need not exist (it could be
522 * reclaimed by prune_dqcache(). Hence we must
523 * restart. */
524 goto restart;
527 * Quota now has no users and it has been written on last
528 * dqput()
530 remove_dquot_hash(dquot);
531 remove_free_dquot(dquot);
532 remove_inuse(dquot);
533 do_destroy_dquot(dquot);
535 spin_unlock(&dq_list_lock);
538 /* Call callback for every active dquot on given filesystem */
539 int dquot_scan_active(struct super_block *sb,
540 int (*fn)(struct dquot *dquot, unsigned long priv),
541 unsigned long priv)
543 struct dquot *dquot, *old_dquot = NULL;
544 int ret = 0;
546 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
547 spin_lock(&dq_list_lock);
548 list_for_each_entry(dquot, &inuse_list, dq_inuse) {
549 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
550 continue;
551 if (dquot->dq_sb != sb)
552 continue;
553 /* Now we have active dquot so we can just increase use count */
554 atomic_inc(&dquot->dq_count);
555 dqstats.lookups++;
556 spin_unlock(&dq_list_lock);
557 dqput(old_dquot);
558 old_dquot = dquot;
559 ret = fn(dquot, priv);
560 if (ret < 0)
561 goto out;
562 spin_lock(&dq_list_lock);
563 /* We are safe to continue now because our dquot could not
564 * be moved out of the inuse list while we hold the reference */
566 spin_unlock(&dq_list_lock);
567 out:
568 dqput(old_dquot);
569 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
570 return ret;
572 EXPORT_SYMBOL(dquot_scan_active);
574 int vfs_quota_sync(struct super_block *sb, int type, int wait)
576 struct list_head *dirty;
577 struct dquot *dquot;
578 struct quota_info *dqopt = sb_dqopt(sb);
579 int cnt;
581 mutex_lock(&dqopt->dqonoff_mutex);
582 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
583 if (type != -1 && cnt != type)
584 continue;
585 if (!sb_has_quota_active(sb, cnt))
586 continue;
587 spin_lock(&dq_list_lock);
588 dirty = &dqopt->info[cnt].dqi_dirty_list;
589 while (!list_empty(dirty)) {
590 dquot = list_first_entry(dirty, struct dquot,
591 dq_dirty);
592 /* Dirty and inactive can be only bad dquot... */
593 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
594 clear_dquot_dirty(dquot);
595 continue;
597 /* Now we have active dquot from which someone is
598 * holding reference so we can safely just increase
599 * use count */
600 atomic_inc(&dquot->dq_count);
601 dqstats.lookups++;
602 spin_unlock(&dq_list_lock);
603 sb->dq_op->write_dquot(dquot);
604 dqput(dquot);
605 spin_lock(&dq_list_lock);
607 spin_unlock(&dq_list_lock);
610 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
611 if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
612 && info_dirty(&dqopt->info[cnt]))
613 sb->dq_op->write_info(sb, cnt);
614 spin_lock(&dq_list_lock);
615 dqstats.syncs++;
616 spin_unlock(&dq_list_lock);
617 mutex_unlock(&dqopt->dqonoff_mutex);
619 if (!wait || (sb_dqopt(sb)->flags & DQUOT_QUOTA_SYS_FILE))
620 return 0;
622 /* This is not very clever (and fast) but currently I don't know about
623 * any other simple way of getting quota data to disk and we must get
624 * them there for userspace to be visible... */
625 if (sb->s_op->sync_fs)
626 sb->s_op->sync_fs(sb, 1);
627 sync_blockdev(sb->s_bdev);
630 * Now when everything is written we can discard the pagecache so
631 * that userspace sees the changes.
633 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
634 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
635 if (type != -1 && cnt != type)
636 continue;
637 if (!sb_has_quota_active(sb, cnt))
638 continue;
639 mutex_lock_nested(&sb_dqopt(sb)->files[cnt]->i_mutex,
640 I_MUTEX_QUOTA);
641 truncate_inode_pages(&sb_dqopt(sb)->files[cnt]->i_data, 0);
642 mutex_unlock(&sb_dqopt(sb)->files[cnt]->i_mutex);
644 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
646 return 0;
648 EXPORT_SYMBOL(vfs_quota_sync);
650 /* Free unused dquots from cache */
651 static void prune_dqcache(int count)
653 struct list_head *head;
654 struct dquot *dquot;
656 head = free_dquots.prev;
657 while (head != &free_dquots && count) {
658 dquot = list_entry(head, struct dquot, dq_free);
659 remove_dquot_hash(dquot);
660 remove_free_dquot(dquot);
661 remove_inuse(dquot);
662 do_destroy_dquot(dquot);
663 count--;
664 head = free_dquots.prev;
669 * This is called from kswapd when we think we need some
670 * more memory
673 static int shrink_dqcache_memory(int nr, gfp_t gfp_mask)
675 if (nr) {
676 spin_lock(&dq_list_lock);
677 prune_dqcache(nr);
678 spin_unlock(&dq_list_lock);
680 return (dqstats.free_dquots / 100) * sysctl_vfs_cache_pressure;
683 static struct shrinker dqcache_shrinker = {
684 .shrink = shrink_dqcache_memory,
685 .seeks = DEFAULT_SEEKS,
689 * Put reference to dquot
690 * NOTE: If you change this function please check whether dqput_blocks() works right...
692 void dqput(struct dquot *dquot)
694 int ret;
696 if (!dquot)
697 return;
698 #ifdef __DQUOT_PARANOIA
699 if (!atomic_read(&dquot->dq_count)) {
700 printk("VFS: dqput: trying to free free dquot\n");
701 printk("VFS: device %s, dquot of %s %d\n",
702 dquot->dq_sb->s_id,
703 quotatypes[dquot->dq_type],
704 dquot->dq_id);
705 BUG();
707 #endif
709 spin_lock(&dq_list_lock);
710 dqstats.drops++;
711 spin_unlock(&dq_list_lock);
712 we_slept:
713 spin_lock(&dq_list_lock);
714 if (atomic_read(&dquot->dq_count) > 1) {
715 /* We have more than one user... nothing to do */
716 atomic_dec(&dquot->dq_count);
717 /* Releasing dquot during quotaoff phase? */
718 if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_type) &&
719 atomic_read(&dquot->dq_count) == 1)
720 wake_up(&dquot->dq_wait_unused);
721 spin_unlock(&dq_list_lock);
722 return;
724 /* Need to release dquot? */
725 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && dquot_dirty(dquot)) {
726 spin_unlock(&dq_list_lock);
727 /* Commit dquot before releasing */
728 ret = dquot->dq_sb->dq_op->write_dquot(dquot);
729 if (ret < 0) {
730 printk(KERN_ERR "VFS: cannot write quota structure on "
731 "device %s (error %d). Quota may get out of "
732 "sync!\n", dquot->dq_sb->s_id, ret);
734 * We clear dirty bit anyway, so that we avoid
735 * infinite loop here
737 spin_lock(&dq_list_lock);
738 clear_dquot_dirty(dquot);
739 spin_unlock(&dq_list_lock);
741 goto we_slept;
743 /* Clear flag in case dquot was inactive (something bad happened) */
744 clear_dquot_dirty(dquot);
745 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
746 spin_unlock(&dq_list_lock);
747 dquot->dq_sb->dq_op->release_dquot(dquot);
748 goto we_slept;
750 atomic_dec(&dquot->dq_count);
751 #ifdef __DQUOT_PARANOIA
752 /* sanity check */
753 BUG_ON(!list_empty(&dquot->dq_free));
754 #endif
755 put_dquot_last(dquot);
756 spin_unlock(&dq_list_lock);
758 EXPORT_SYMBOL(dqput);
760 struct dquot *dquot_alloc(struct super_block *sb, int type)
762 return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
764 EXPORT_SYMBOL(dquot_alloc);
766 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
768 struct dquot *dquot;
770 dquot = sb->dq_op->alloc_dquot(sb, type);
771 if(!dquot)
772 return NULL;
774 mutex_init(&dquot->dq_lock);
775 INIT_LIST_HEAD(&dquot->dq_free);
776 INIT_LIST_HEAD(&dquot->dq_inuse);
777 INIT_HLIST_NODE(&dquot->dq_hash);
778 INIT_LIST_HEAD(&dquot->dq_dirty);
779 init_waitqueue_head(&dquot->dq_wait_unused);
780 dquot->dq_sb = sb;
781 dquot->dq_type = type;
782 atomic_set(&dquot->dq_count, 1);
784 return dquot;
788 * Get reference to dquot
790 * Locking is slightly tricky here. We are guarded from parallel quotaoff()
791 * destroying our dquot by:
792 * a) checking for quota flags under dq_list_lock and
793 * b) getting a reference to dquot before we release dq_list_lock
795 struct dquot *dqget(struct super_block *sb, unsigned int id, int type)
797 unsigned int hashent = hashfn(sb, id, type);
798 struct dquot *dquot = NULL, *empty = NULL;
800 if (!sb_has_quota_active(sb, type))
801 return NULL;
802 we_slept:
803 spin_lock(&dq_list_lock);
804 spin_lock(&dq_state_lock);
805 if (!sb_has_quota_active(sb, type)) {
806 spin_unlock(&dq_state_lock);
807 spin_unlock(&dq_list_lock);
808 goto out;
810 spin_unlock(&dq_state_lock);
812 dquot = find_dquot(hashent, sb, id, type);
813 if (!dquot) {
814 if (!empty) {
815 spin_unlock(&dq_list_lock);
816 empty = get_empty_dquot(sb, type);
817 if (!empty)
818 schedule(); /* Try to wait for a moment... */
819 goto we_slept;
821 dquot = empty;
822 empty = NULL;
823 dquot->dq_id = id;
824 /* all dquots go on the inuse_list */
825 put_inuse(dquot);
826 /* hash it first so it can be found */
827 insert_dquot_hash(dquot);
828 dqstats.lookups++;
829 spin_unlock(&dq_list_lock);
830 } else {
831 if (!atomic_read(&dquot->dq_count))
832 remove_free_dquot(dquot);
833 atomic_inc(&dquot->dq_count);
834 dqstats.cache_hits++;
835 dqstats.lookups++;
836 spin_unlock(&dq_list_lock);
838 /* Wait for dq_lock - after this we know that either dquot_release() is
839 * already finished or it will be canceled due to dq_count > 1 test */
840 wait_on_dquot(dquot);
841 /* Read the dquot / allocate space in quota file */
842 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) &&
843 sb->dq_op->acquire_dquot(dquot) < 0) {
844 dqput(dquot);
845 dquot = NULL;
846 goto out;
848 #ifdef __DQUOT_PARANOIA
849 BUG_ON(!dquot->dq_sb); /* Has somebody invalidated entry under us? */
850 #endif
851 out:
852 if (empty)
853 do_destroy_dquot(empty);
855 return dquot;
857 EXPORT_SYMBOL(dqget);
859 static int dqinit_needed(struct inode *inode, int type)
861 int cnt;
863 if (IS_NOQUOTA(inode))
864 return 0;
865 if (type != -1)
866 return !inode->i_dquot[type];
867 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
868 if (!inode->i_dquot[cnt])
869 return 1;
870 return 0;
873 /* This routine is guarded by dqonoff_mutex mutex */
874 static void add_dquot_ref(struct super_block *sb, int type)
876 struct inode *inode, *old_inode = NULL;
877 int reserved = 0;
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 if (unlikely(inode_get_rsv_space(inode) > 0))
884 reserved = 1;
885 if (!atomic_read(&inode->i_writecount))
886 continue;
887 if (!dqinit_needed(inode, type))
888 continue;
890 __iget(inode);
891 spin_unlock(&inode_lock);
893 iput(old_inode);
894 __dquot_initialize(inode, type);
895 /* We hold a reference to 'inode' so it couldn't have been
896 * removed from s_inodes list while we dropped the inode_lock.
897 * We cannot iput the inode now as we can be holding the last
898 * reference and we cannot iput it under inode_lock. So we
899 * keep the reference and iput it later. */
900 old_inode = inode;
901 spin_lock(&inode_lock);
903 spin_unlock(&inode_lock);
904 iput(old_inode);
906 if (reserved) {
907 printk(KERN_WARNING "VFS (%s): Writes happened before quota"
908 " was turned on thus quota information is probably "
909 "inconsistent. Please run quotacheck(8).\n", sb->s_id);
914 * Return 0 if dqput() won't block.
915 * (note that 1 doesn't necessarily mean blocking)
917 static inline int dqput_blocks(struct dquot *dquot)
919 if (atomic_read(&dquot->dq_count) <= 1)
920 return 1;
921 return 0;
925 * Remove references to dquots from inode and add dquot to list for freeing
926 * if we have the last referece to dquot
927 * We can't race with anybody because we hold dqptr_sem for writing...
929 static int remove_inode_dquot_ref(struct inode *inode, int type,
930 struct list_head *tofree_head)
932 struct dquot *dquot = inode->i_dquot[type];
934 inode->i_dquot[type] = NULL;
935 if (dquot) {
936 if (dqput_blocks(dquot)) {
937 #ifdef __DQUOT_PARANOIA
938 if (atomic_read(&dquot->dq_count) != 1)
939 printk(KERN_WARNING "VFS: Adding dquot with dq_count %d to dispose list.\n", atomic_read(&dquot->dq_count));
940 #endif
941 spin_lock(&dq_list_lock);
942 /* As dquot must have currently users it can't be on
943 * the free list... */
944 list_add(&dquot->dq_free, tofree_head);
945 spin_unlock(&dq_list_lock);
946 return 1;
948 else
949 dqput(dquot); /* We have guaranteed we won't block */
951 return 0;
955 * Free list of dquots
956 * Dquots are removed from inodes and no new references can be got so we are
957 * the only ones holding reference
959 static void put_dquot_list(struct list_head *tofree_head)
961 struct list_head *act_head;
962 struct dquot *dquot;
964 act_head = tofree_head->next;
965 while (act_head != tofree_head) {
966 dquot = list_entry(act_head, struct dquot, dq_free);
967 act_head = act_head->next;
968 /* Remove dquot from the list so we won't have problems... */
969 list_del_init(&dquot->dq_free);
970 dqput(dquot);
974 static void remove_dquot_ref(struct super_block *sb, int type,
975 struct list_head *tofree_head)
977 struct inode *inode;
979 spin_lock(&inode_lock);
980 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
982 * We have to scan also I_NEW inodes because they can already
983 * have quota pointer initialized. Luckily, we need to touch
984 * only quota pointers and these have separate locking
985 * (dqptr_sem).
987 if (!IS_NOQUOTA(inode))
988 remove_inode_dquot_ref(inode, type, tofree_head);
990 spin_unlock(&inode_lock);
993 /* Gather all references from inodes and drop them */
994 static void drop_dquot_ref(struct super_block *sb, int type)
996 LIST_HEAD(tofree_head);
998 if (sb->dq_op) {
999 down_write(&sb_dqopt(sb)->dqptr_sem);
1000 remove_dquot_ref(sb, type, &tofree_head);
1001 up_write(&sb_dqopt(sb)->dqptr_sem);
1002 put_dquot_list(&tofree_head);
1006 static inline void dquot_incr_inodes(struct dquot *dquot, qsize_t number)
1008 dquot->dq_dqb.dqb_curinodes += number;
1011 static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
1013 dquot->dq_dqb.dqb_curspace += number;
1016 static inline void dquot_resv_space(struct dquot *dquot, qsize_t number)
1018 dquot->dq_dqb.dqb_rsvspace += number;
1022 * Claim reserved quota space
1024 static void dquot_claim_reserved_space(struct dquot *dquot, qsize_t number)
1026 if (dquot->dq_dqb.dqb_rsvspace < number) {
1027 WARN_ON_ONCE(1);
1028 number = dquot->dq_dqb.dqb_rsvspace;
1030 dquot->dq_dqb.dqb_curspace += number;
1031 dquot->dq_dqb.dqb_rsvspace -= number;
1034 static inline
1035 void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1037 if (dquot->dq_dqb.dqb_rsvspace >= number)
1038 dquot->dq_dqb.dqb_rsvspace -= number;
1039 else {
1040 WARN_ON_ONCE(1);
1041 dquot->dq_dqb.dqb_rsvspace = 0;
1045 static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1047 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1048 dquot->dq_dqb.dqb_curinodes >= number)
1049 dquot->dq_dqb.dqb_curinodes -= number;
1050 else
1051 dquot->dq_dqb.dqb_curinodes = 0;
1052 if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1053 dquot->dq_dqb.dqb_itime = (time_t) 0;
1054 clear_bit(DQ_INODES_B, &dquot->dq_flags);
1057 static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1059 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1060 dquot->dq_dqb.dqb_curspace >= number)
1061 dquot->dq_dqb.dqb_curspace -= number;
1062 else
1063 dquot->dq_dqb.dqb_curspace = 0;
1064 if (dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1065 dquot->dq_dqb.dqb_btime = (time_t) 0;
1066 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1069 static int warning_issued(struct dquot *dquot, const int warntype)
1071 int flag = (warntype == QUOTA_NL_BHARDWARN ||
1072 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1073 ((warntype == QUOTA_NL_IHARDWARN ||
1074 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1076 if (!flag)
1077 return 0;
1078 return test_and_set_bit(flag, &dquot->dq_flags);
1081 #ifdef CONFIG_PRINT_QUOTA_WARNING
1082 static int flag_print_warnings = 1;
1084 static int need_print_warning(struct dquot *dquot)
1086 if (!flag_print_warnings)
1087 return 0;
1089 switch (dquot->dq_type) {
1090 case USRQUOTA:
1091 return current_fsuid() == dquot->dq_id;
1092 case GRPQUOTA:
1093 return in_group_p(dquot->dq_id);
1095 return 0;
1098 /* Print warning to user which exceeded quota */
1099 static void print_warning(struct dquot *dquot, const int warntype)
1101 char *msg = NULL;
1102 struct tty_struct *tty;
1104 if (warntype == QUOTA_NL_IHARDBELOW ||
1105 warntype == QUOTA_NL_ISOFTBELOW ||
1106 warntype == QUOTA_NL_BHARDBELOW ||
1107 warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(dquot))
1108 return;
1110 tty = get_current_tty();
1111 if (!tty)
1112 return;
1113 tty_write_message(tty, dquot->dq_sb->s_id);
1114 if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1115 tty_write_message(tty, ": warning, ");
1116 else
1117 tty_write_message(tty, ": write failed, ");
1118 tty_write_message(tty, quotatypes[dquot->dq_type]);
1119 switch (warntype) {
1120 case QUOTA_NL_IHARDWARN:
1121 msg = " file limit reached.\r\n";
1122 break;
1123 case QUOTA_NL_ISOFTLONGWARN:
1124 msg = " file quota exceeded too long.\r\n";
1125 break;
1126 case QUOTA_NL_ISOFTWARN:
1127 msg = " file quota exceeded.\r\n";
1128 break;
1129 case QUOTA_NL_BHARDWARN:
1130 msg = " block limit reached.\r\n";
1131 break;
1132 case QUOTA_NL_BSOFTLONGWARN:
1133 msg = " block quota exceeded too long.\r\n";
1134 break;
1135 case QUOTA_NL_BSOFTWARN:
1136 msg = " block quota exceeded.\r\n";
1137 break;
1139 tty_write_message(tty, msg);
1140 tty_kref_put(tty);
1142 #endif
1145 * Write warnings to the console and send warning messages over netlink.
1147 * Note that this function can sleep.
1149 static void flush_warnings(struct dquot *const *dquots, char *warntype)
1151 struct dquot *dq;
1152 int i;
1154 for (i = 0; i < MAXQUOTAS; i++) {
1155 dq = dquots[i];
1156 if (dq && warntype[i] != QUOTA_NL_NOWARN &&
1157 !warning_issued(dq, warntype[i])) {
1158 #ifdef CONFIG_PRINT_QUOTA_WARNING
1159 print_warning(dq, warntype[i]);
1160 #endif
1161 quota_send_warning(dq->dq_type, dq->dq_id,
1162 dq->dq_sb->s_dev, warntype[i]);
1167 static int ignore_hardlimit(struct dquot *dquot)
1169 struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_type];
1171 return capable(CAP_SYS_RESOURCE) &&
1172 (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1173 !(info->dqi_flags & V1_DQF_RSQUASH));
1176 /* needs dq_data_lock */
1177 static int check_idq(struct dquot *dquot, qsize_t inodes, char *warntype)
1179 qsize_t newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1181 *warntype = QUOTA_NL_NOWARN;
1182 if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_type) ||
1183 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1184 return 0;
1186 if (dquot->dq_dqb.dqb_ihardlimit &&
1187 newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1188 !ignore_hardlimit(dquot)) {
1189 *warntype = QUOTA_NL_IHARDWARN;
1190 return -EDQUOT;
1193 if (dquot->dq_dqb.dqb_isoftlimit &&
1194 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1195 dquot->dq_dqb.dqb_itime &&
1196 get_seconds() >= dquot->dq_dqb.dqb_itime &&
1197 !ignore_hardlimit(dquot)) {
1198 *warntype = QUOTA_NL_ISOFTLONGWARN;
1199 return -EDQUOT;
1202 if (dquot->dq_dqb.dqb_isoftlimit &&
1203 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1204 dquot->dq_dqb.dqb_itime == 0) {
1205 *warntype = QUOTA_NL_ISOFTWARN;
1206 dquot->dq_dqb.dqb_itime = get_seconds() +
1207 sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
1210 return 0;
1213 /* needs dq_data_lock */
1214 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc, char *warntype)
1216 qsize_t tspace;
1217 struct super_block *sb = dquot->dq_sb;
1219 *warntype = QUOTA_NL_NOWARN;
1220 if (!sb_has_quota_limits_enabled(sb, dquot->dq_type) ||
1221 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1222 return 0;
1224 tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1225 + space;
1227 if (dquot->dq_dqb.dqb_bhardlimit &&
1228 tspace > dquot->dq_dqb.dqb_bhardlimit &&
1229 !ignore_hardlimit(dquot)) {
1230 if (!prealloc)
1231 *warntype = QUOTA_NL_BHARDWARN;
1232 return -EDQUOT;
1235 if (dquot->dq_dqb.dqb_bsoftlimit &&
1236 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1237 dquot->dq_dqb.dqb_btime &&
1238 get_seconds() >= dquot->dq_dqb.dqb_btime &&
1239 !ignore_hardlimit(dquot)) {
1240 if (!prealloc)
1241 *warntype = QUOTA_NL_BSOFTLONGWARN;
1242 return -EDQUOT;
1245 if (dquot->dq_dqb.dqb_bsoftlimit &&
1246 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1247 dquot->dq_dqb.dqb_btime == 0) {
1248 if (!prealloc) {
1249 *warntype = QUOTA_NL_BSOFTWARN;
1250 dquot->dq_dqb.dqb_btime = get_seconds() +
1251 sb_dqopt(sb)->info[dquot->dq_type].dqi_bgrace;
1253 else
1255 * We don't allow preallocation to exceed softlimit so exceeding will
1256 * be always printed
1258 return -EDQUOT;
1261 return 0;
1264 static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1266 qsize_t newinodes;
1268 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1269 dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1270 !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_type))
1271 return QUOTA_NL_NOWARN;
1273 newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1274 if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1275 return QUOTA_NL_ISOFTBELOW;
1276 if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1277 newinodes < dquot->dq_dqb.dqb_ihardlimit)
1278 return QUOTA_NL_IHARDBELOW;
1279 return QUOTA_NL_NOWARN;
1282 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1284 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1285 dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1286 return QUOTA_NL_NOWARN;
1288 if (dquot->dq_dqb.dqb_curspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1289 return QUOTA_NL_BSOFTBELOW;
1290 if (dquot->dq_dqb.dqb_curspace >= dquot->dq_dqb.dqb_bhardlimit &&
1291 dquot->dq_dqb.dqb_curspace - space < dquot->dq_dqb.dqb_bhardlimit)
1292 return QUOTA_NL_BHARDBELOW;
1293 return QUOTA_NL_NOWARN;
1297 * Initialize quota pointers in inode
1299 * We do things in a bit complicated way but by that we avoid calling
1300 * dqget() and thus filesystem callbacks under dqptr_sem.
1302 * It is better to call this function outside of any transaction as it
1303 * might need a lot of space in journal for dquot structure allocation.
1305 static void __dquot_initialize(struct inode *inode, int type)
1307 unsigned int id = 0;
1308 int cnt;
1309 struct dquot *got[MAXQUOTAS];
1310 struct super_block *sb = inode->i_sb;
1311 qsize_t rsv;
1313 /* First test before acquiring mutex - solves deadlocks when we
1314 * re-enter the quota code and are already holding the mutex */
1315 if (!sb_any_quota_active(inode->i_sb) || IS_NOQUOTA(inode))
1316 return;
1318 /* First get references to structures we might need. */
1319 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1320 got[cnt] = NULL;
1321 if (type != -1 && cnt != type)
1322 continue;
1323 switch (cnt) {
1324 case USRQUOTA:
1325 id = inode->i_uid;
1326 break;
1327 case GRPQUOTA:
1328 id = inode->i_gid;
1329 break;
1331 got[cnt] = dqget(sb, id, cnt);
1334 down_write(&sb_dqopt(sb)->dqptr_sem);
1335 if (IS_NOQUOTA(inode))
1336 goto out_err;
1337 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1338 if (type != -1 && cnt != type)
1339 continue;
1340 /* Avoid races with quotaoff() */
1341 if (!sb_has_quota_active(sb, cnt))
1342 continue;
1343 if (!inode->i_dquot[cnt]) {
1344 inode->i_dquot[cnt] = got[cnt];
1345 got[cnt] = NULL;
1347 * Make quota reservation system happy if someone
1348 * did a write before quota was turned on
1350 rsv = inode_get_rsv_space(inode);
1351 if (unlikely(rsv))
1352 dquot_resv_space(inode->i_dquot[cnt], rsv);
1355 out_err:
1356 up_write(&sb_dqopt(sb)->dqptr_sem);
1357 /* Drop unused references */
1358 dqput_all(got);
1361 void dquot_initialize(struct inode *inode)
1363 __dquot_initialize(inode, -1);
1365 EXPORT_SYMBOL(dquot_initialize);
1368 * Release all quotas referenced by inode
1370 static void __dquot_drop(struct inode *inode)
1372 int cnt;
1373 struct dquot *put[MAXQUOTAS];
1375 down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1376 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1377 put[cnt] = inode->i_dquot[cnt];
1378 inode->i_dquot[cnt] = NULL;
1380 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1381 dqput_all(put);
1384 void dquot_drop(struct inode *inode)
1386 int cnt;
1388 if (IS_NOQUOTA(inode))
1389 return;
1392 * Test before calling to rule out calls from proc and such
1393 * where we are not allowed to block. Note that this is
1394 * actually reliable test even without the lock - the caller
1395 * must assure that nobody can come after the DQUOT_DROP and
1396 * add quota pointers back anyway.
1398 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1399 if (inode->i_dquot[cnt])
1400 break;
1403 if (cnt < MAXQUOTAS)
1404 __dquot_drop(inode);
1406 EXPORT_SYMBOL(dquot_drop);
1409 * inode_reserved_space is managed internally by quota, and protected by
1410 * i_lock similar to i_blocks+i_bytes.
1412 static qsize_t *inode_reserved_space(struct inode * inode)
1414 /* Filesystem must explicitly define it's own method in order to use
1415 * quota reservation interface */
1416 BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1417 return inode->i_sb->dq_op->get_reserved_space(inode);
1420 void inode_add_rsv_space(struct inode *inode, qsize_t number)
1422 spin_lock(&inode->i_lock);
1423 *inode_reserved_space(inode) += number;
1424 spin_unlock(&inode->i_lock);
1426 EXPORT_SYMBOL(inode_add_rsv_space);
1428 void inode_claim_rsv_space(struct inode *inode, qsize_t number)
1430 spin_lock(&inode->i_lock);
1431 *inode_reserved_space(inode) -= number;
1432 __inode_add_bytes(inode, number);
1433 spin_unlock(&inode->i_lock);
1435 EXPORT_SYMBOL(inode_claim_rsv_space);
1437 void inode_sub_rsv_space(struct inode *inode, qsize_t number)
1439 spin_lock(&inode->i_lock);
1440 *inode_reserved_space(inode) -= number;
1441 spin_unlock(&inode->i_lock);
1443 EXPORT_SYMBOL(inode_sub_rsv_space);
1445 static qsize_t inode_get_rsv_space(struct inode *inode)
1447 qsize_t ret;
1449 if (!inode->i_sb->dq_op->get_reserved_space)
1450 return 0;
1451 spin_lock(&inode->i_lock);
1452 ret = *inode_reserved_space(inode);
1453 spin_unlock(&inode->i_lock);
1454 return ret;
1457 static void inode_incr_space(struct inode *inode, qsize_t number,
1458 int reserve)
1460 if (reserve)
1461 inode_add_rsv_space(inode, number);
1462 else
1463 inode_add_bytes(inode, number);
1466 static void inode_decr_space(struct inode *inode, qsize_t number, int reserve)
1468 if (reserve)
1469 inode_sub_rsv_space(inode, number);
1470 else
1471 inode_sub_bytes(inode, number);
1475 * This functions updates i_blocks+i_bytes fields and quota information
1476 * (together with appropriate checks).
1478 * NOTE: We absolutely rely on the fact that caller dirties the inode
1479 * (usually helpers in quotaops.h care about this) and holds a handle for
1480 * the current transaction so that dquot write and inode write go into the
1481 * same transaction.
1485 * This operation can block, but only after everything is updated
1487 int __dquot_alloc_space(struct inode *inode, qsize_t number,
1488 int warn, int reserve)
1490 int cnt, ret = 0;
1491 char warntype[MAXQUOTAS];
1494 * First test before acquiring mutex - solves deadlocks when we
1495 * re-enter the quota code and are already holding the mutex
1497 if (!sb_any_quota_active(inode->i_sb) || IS_NOQUOTA(inode)) {
1498 inode_incr_space(inode, number, reserve);
1499 goto out;
1502 down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1503 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1504 warntype[cnt] = QUOTA_NL_NOWARN;
1506 spin_lock(&dq_data_lock);
1507 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1508 if (!inode->i_dquot[cnt])
1509 continue;
1510 ret = check_bdq(inode->i_dquot[cnt], number, !warn,
1511 warntype+cnt);
1512 if (ret) {
1513 spin_unlock(&dq_data_lock);
1514 goto out_flush_warn;
1517 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1518 if (!inode->i_dquot[cnt])
1519 continue;
1520 if (reserve)
1521 dquot_resv_space(inode->i_dquot[cnt], number);
1522 else
1523 dquot_incr_space(inode->i_dquot[cnt], number);
1525 inode_incr_space(inode, number, reserve);
1526 spin_unlock(&dq_data_lock);
1528 if (reserve)
1529 goto out_flush_warn;
1530 mark_all_dquot_dirty(inode->i_dquot);
1531 out_flush_warn:
1532 flush_warnings(inode->i_dquot, warntype);
1533 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1534 out:
1535 return ret;
1537 EXPORT_SYMBOL(__dquot_alloc_space);
1540 * This operation can block, but only after everything is updated
1542 int dquot_alloc_inode(const struct inode *inode)
1544 int cnt, ret = 0;
1545 char warntype[MAXQUOTAS];
1547 /* First test before acquiring mutex - solves deadlocks when we
1548 * re-enter the quota code and are already holding the mutex */
1549 if (!sb_any_quota_active(inode->i_sb) || IS_NOQUOTA(inode))
1550 return 0;
1551 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1552 warntype[cnt] = QUOTA_NL_NOWARN;
1553 down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1554 spin_lock(&dq_data_lock);
1555 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1556 if (!inode->i_dquot[cnt])
1557 continue;
1558 ret = check_idq(inode->i_dquot[cnt], 1, warntype + cnt);
1559 if (ret)
1560 goto warn_put_all;
1563 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1564 if (!inode->i_dquot[cnt])
1565 continue;
1566 dquot_incr_inodes(inode->i_dquot[cnt], 1);
1569 warn_put_all:
1570 spin_unlock(&dq_data_lock);
1571 if (ret == 0)
1572 mark_all_dquot_dirty(inode->i_dquot);
1573 flush_warnings(inode->i_dquot, warntype);
1574 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1575 return ret;
1577 EXPORT_SYMBOL(dquot_alloc_inode);
1580 * Convert in-memory reserved quotas to real consumed quotas
1582 int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1584 int cnt;
1586 if (!sb_any_quota_active(inode->i_sb) || IS_NOQUOTA(inode)) {
1587 inode_claim_rsv_space(inode, number);
1588 return 0;
1591 down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1592 spin_lock(&dq_data_lock);
1593 /* Claim reserved quotas to allocated quotas */
1594 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1595 if (inode->i_dquot[cnt])
1596 dquot_claim_reserved_space(inode->i_dquot[cnt],
1597 number);
1599 /* Update inode bytes */
1600 inode_claim_rsv_space(inode, number);
1601 spin_unlock(&dq_data_lock);
1602 mark_all_dquot_dirty(inode->i_dquot);
1603 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1604 return 0;
1606 EXPORT_SYMBOL(dquot_claim_space_nodirty);
1609 * This operation can block, but only after everything is updated
1611 void __dquot_free_space(struct inode *inode, qsize_t number, int reserve)
1613 unsigned int cnt;
1614 char warntype[MAXQUOTAS];
1616 /* First test before acquiring mutex - solves deadlocks when we
1617 * re-enter the quota code and are already holding the mutex */
1618 if (!sb_any_quota_active(inode->i_sb) || IS_NOQUOTA(inode)) {
1619 inode_decr_space(inode, number, reserve);
1620 return;
1623 down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1624 spin_lock(&dq_data_lock);
1625 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1626 if (!inode->i_dquot[cnt])
1627 continue;
1628 warntype[cnt] = info_bdq_free(inode->i_dquot[cnt], number);
1629 if (reserve)
1630 dquot_free_reserved_space(inode->i_dquot[cnt], number);
1631 else
1632 dquot_decr_space(inode->i_dquot[cnt], number);
1634 inode_decr_space(inode, number, reserve);
1635 spin_unlock(&dq_data_lock);
1637 if (reserve)
1638 goto out_unlock;
1639 mark_all_dquot_dirty(inode->i_dquot);
1640 out_unlock:
1641 flush_warnings(inode->i_dquot, warntype);
1642 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1644 EXPORT_SYMBOL(__dquot_free_space);
1647 * This operation can block, but only after everything is updated
1649 void dquot_free_inode(const struct inode *inode)
1651 unsigned int cnt;
1652 char warntype[MAXQUOTAS];
1654 /* First test before acquiring mutex - solves deadlocks when we
1655 * re-enter the quota code and are already holding the mutex */
1656 if (!sb_any_quota_active(inode->i_sb) || IS_NOQUOTA(inode))
1657 return;
1659 down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1660 spin_lock(&dq_data_lock);
1661 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1662 if (!inode->i_dquot[cnt])
1663 continue;
1664 warntype[cnt] = info_idq_free(inode->i_dquot[cnt], 1);
1665 dquot_decr_inodes(inode->i_dquot[cnt], 1);
1667 spin_unlock(&dq_data_lock);
1668 mark_all_dquot_dirty(inode->i_dquot);
1669 flush_warnings(inode->i_dquot, warntype);
1670 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1672 EXPORT_SYMBOL(dquot_free_inode);
1675 * Transfer the number of inode and blocks from one diskquota to an other.
1677 * This operation can block, but only after everything is updated
1678 * A transaction must be started when entering this function.
1680 static int __dquot_transfer(struct inode *inode, qid_t *chid, unsigned long mask)
1682 qsize_t space, cur_space;
1683 qsize_t rsv_space = 0;
1684 struct dquot *transfer_from[MAXQUOTAS];
1685 struct dquot *transfer_to[MAXQUOTAS];
1686 int cnt, ret = 0;
1687 char warntype_to[MAXQUOTAS];
1688 char warntype_from_inodes[MAXQUOTAS], warntype_from_space[MAXQUOTAS];
1690 /* First test before acquiring mutex - solves deadlocks when we
1691 * re-enter the quota code and are already holding the mutex */
1692 if (IS_NOQUOTA(inode))
1693 return 0;
1694 /* Initialize the arrays */
1695 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1696 transfer_from[cnt] = NULL;
1697 transfer_to[cnt] = NULL;
1698 warntype_to[cnt] = QUOTA_NL_NOWARN;
1700 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1701 if (mask & (1 << cnt))
1702 transfer_to[cnt] = dqget(inode->i_sb, chid[cnt], cnt);
1704 down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1705 if (IS_NOQUOTA(inode)) { /* File without quota accounting? */
1706 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1707 goto put_all;
1709 spin_lock(&dq_data_lock);
1710 cur_space = inode_get_bytes(inode);
1711 rsv_space = inode_get_rsv_space(inode);
1712 space = cur_space + rsv_space;
1713 /* Build the transfer_from list and check the limits */
1714 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1715 if (!transfer_to[cnt])
1716 continue;
1717 transfer_from[cnt] = inode->i_dquot[cnt];
1718 ret = check_idq(transfer_to[cnt], 1, warntype_to + cnt);
1719 if (ret)
1720 goto over_quota;
1721 ret = check_bdq(transfer_to[cnt], space, 0, warntype_to + cnt);
1722 if (ret)
1723 goto over_quota;
1727 * Finally perform the needed transfer from transfer_from to transfer_to
1729 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1731 * Skip changes for same uid or gid or for turned off quota-type.
1733 if (!transfer_to[cnt])
1734 continue;
1736 /* Due to IO error we might not have transfer_from[] structure */
1737 if (transfer_from[cnt]) {
1738 warntype_from_inodes[cnt] =
1739 info_idq_free(transfer_from[cnt], 1);
1740 warntype_from_space[cnt] =
1741 info_bdq_free(transfer_from[cnt], space);
1742 dquot_decr_inodes(transfer_from[cnt], 1);
1743 dquot_decr_space(transfer_from[cnt], cur_space);
1744 dquot_free_reserved_space(transfer_from[cnt],
1745 rsv_space);
1748 dquot_incr_inodes(transfer_to[cnt], 1);
1749 dquot_incr_space(transfer_to[cnt], cur_space);
1750 dquot_resv_space(transfer_to[cnt], rsv_space);
1752 inode->i_dquot[cnt] = transfer_to[cnt];
1754 spin_unlock(&dq_data_lock);
1755 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1757 mark_all_dquot_dirty(transfer_from);
1758 mark_all_dquot_dirty(transfer_to);
1759 /* The reference we got is transferred to the inode */
1760 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1761 transfer_to[cnt] = NULL;
1762 warn_put_all:
1763 flush_warnings(transfer_to, warntype_to);
1764 flush_warnings(transfer_from, warntype_from_inodes);
1765 flush_warnings(transfer_from, warntype_from_space);
1766 put_all:
1767 dqput_all(transfer_from);
1768 dqput_all(transfer_to);
1769 return ret;
1770 over_quota:
1771 spin_unlock(&dq_data_lock);
1772 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1773 /* Clear dquot pointers we don't want to dqput() */
1774 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1775 transfer_from[cnt] = NULL;
1776 goto warn_put_all;
1779 /* Wrapper for transferring ownership of an inode for uid/gid only
1780 * Called from FSXXX_setattr()
1782 int dquot_transfer(struct inode *inode, struct iattr *iattr)
1784 qid_t chid[MAXQUOTAS];
1785 unsigned long mask = 0;
1787 if (iattr->ia_valid & ATTR_UID && iattr->ia_uid != inode->i_uid) {
1788 mask |= 1 << USRQUOTA;
1789 chid[USRQUOTA] = iattr->ia_uid;
1791 if (iattr->ia_valid & ATTR_GID && iattr->ia_gid != inode->i_gid) {
1792 mask |= 1 << GRPQUOTA;
1793 chid[GRPQUOTA] = iattr->ia_gid;
1795 if (sb_any_quota_active(inode->i_sb) && !IS_NOQUOTA(inode)) {
1796 dquot_initialize(inode);
1797 return __dquot_transfer(inode, chid, mask);
1799 return 0;
1801 EXPORT_SYMBOL(dquot_transfer);
1804 * Write info of quota file to disk
1806 int dquot_commit_info(struct super_block *sb, int type)
1808 int ret;
1809 struct quota_info *dqopt = sb_dqopt(sb);
1811 mutex_lock(&dqopt->dqio_mutex);
1812 ret = dqopt->ops[type]->write_file_info(sb, type);
1813 mutex_unlock(&dqopt->dqio_mutex);
1814 return ret;
1816 EXPORT_SYMBOL(dquot_commit_info);
1819 * Definitions of diskquota operations.
1821 const struct dquot_operations dquot_operations = {
1822 .write_dquot = dquot_commit,
1823 .acquire_dquot = dquot_acquire,
1824 .release_dquot = dquot_release,
1825 .mark_dirty = dquot_mark_dquot_dirty,
1826 .write_info = dquot_commit_info,
1827 .alloc_dquot = dquot_alloc,
1828 .destroy_dquot = dquot_destroy,
1832 * Generic helper for ->open on filesystems supporting disk quotas.
1834 int dquot_file_open(struct inode *inode, struct file *file)
1836 int error;
1838 error = generic_file_open(inode, file);
1839 if (!error && (file->f_mode & FMODE_WRITE))
1840 dquot_initialize(inode);
1841 return error;
1843 EXPORT_SYMBOL(dquot_file_open);
1846 * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
1848 int vfs_quota_disable(struct super_block *sb, int type, unsigned int flags)
1850 int cnt, ret = 0;
1851 struct quota_info *dqopt = sb_dqopt(sb);
1852 struct inode *toputinode[MAXQUOTAS];
1854 /* Cannot turn off usage accounting without turning off limits, or
1855 * suspend quotas and simultaneously turn quotas off. */
1856 if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
1857 || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
1858 DQUOT_USAGE_ENABLED)))
1859 return -EINVAL;
1861 /* We need to serialize quota_off() for device */
1862 mutex_lock(&dqopt->dqonoff_mutex);
1865 * Skip everything if there's nothing to do. We have to do this because
1866 * sometimes we are called when fill_super() failed and calling
1867 * sync_fs() in such cases does no good.
1869 if (!sb_any_quota_loaded(sb)) {
1870 mutex_unlock(&dqopt->dqonoff_mutex);
1871 return 0;
1873 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1874 toputinode[cnt] = NULL;
1875 if (type != -1 && cnt != type)
1876 continue;
1877 if (!sb_has_quota_loaded(sb, cnt))
1878 continue;
1880 if (flags & DQUOT_SUSPENDED) {
1881 spin_lock(&dq_state_lock);
1882 dqopt->flags |=
1883 dquot_state_flag(DQUOT_SUSPENDED, cnt);
1884 spin_unlock(&dq_state_lock);
1885 } else {
1886 spin_lock(&dq_state_lock);
1887 dqopt->flags &= ~dquot_state_flag(flags, cnt);
1888 /* Turning off suspended quotas? */
1889 if (!sb_has_quota_loaded(sb, cnt) &&
1890 sb_has_quota_suspended(sb, cnt)) {
1891 dqopt->flags &= ~dquot_state_flag(
1892 DQUOT_SUSPENDED, cnt);
1893 spin_unlock(&dq_state_lock);
1894 iput(dqopt->files[cnt]);
1895 dqopt->files[cnt] = NULL;
1896 continue;
1898 spin_unlock(&dq_state_lock);
1901 /* We still have to keep quota loaded? */
1902 if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
1903 continue;
1905 /* Note: these are blocking operations */
1906 drop_dquot_ref(sb, cnt);
1907 invalidate_dquots(sb, cnt);
1909 * Now all dquots should be invalidated, all writes done so we
1910 * should be only users of the info. No locks needed.
1912 if (info_dirty(&dqopt->info[cnt]))
1913 sb->dq_op->write_info(sb, cnt);
1914 if (dqopt->ops[cnt]->free_file_info)
1915 dqopt->ops[cnt]->free_file_info(sb, cnt);
1916 put_quota_format(dqopt->info[cnt].dqi_format);
1918 toputinode[cnt] = dqopt->files[cnt];
1919 if (!sb_has_quota_loaded(sb, cnt))
1920 dqopt->files[cnt] = NULL;
1921 dqopt->info[cnt].dqi_flags = 0;
1922 dqopt->info[cnt].dqi_igrace = 0;
1923 dqopt->info[cnt].dqi_bgrace = 0;
1924 dqopt->ops[cnt] = NULL;
1926 mutex_unlock(&dqopt->dqonoff_mutex);
1928 /* Skip syncing and setting flags if quota files are hidden */
1929 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
1930 goto put_inodes;
1932 /* Sync the superblock so that buffers with quota data are written to
1933 * disk (and so userspace sees correct data afterwards). */
1934 if (sb->s_op->sync_fs)
1935 sb->s_op->sync_fs(sb, 1);
1936 sync_blockdev(sb->s_bdev);
1937 /* Now the quota files are just ordinary files and we can set the
1938 * inode flags back. Moreover we discard the pagecache so that
1939 * userspace sees the writes we did bypassing the pagecache. We
1940 * must also discard the blockdev buffers so that we see the
1941 * changes done by userspace on the next quotaon() */
1942 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1943 if (toputinode[cnt]) {
1944 mutex_lock(&dqopt->dqonoff_mutex);
1945 /* If quota was reenabled in the meantime, we have
1946 * nothing to do */
1947 if (!sb_has_quota_loaded(sb, cnt)) {
1948 mutex_lock_nested(&toputinode[cnt]->i_mutex,
1949 I_MUTEX_QUOTA);
1950 toputinode[cnt]->i_flags &= ~(S_IMMUTABLE |
1951 S_NOATIME | S_NOQUOTA);
1952 truncate_inode_pages(&toputinode[cnt]->i_data,
1954 mutex_unlock(&toputinode[cnt]->i_mutex);
1955 mark_inode_dirty(toputinode[cnt]);
1957 mutex_unlock(&dqopt->dqonoff_mutex);
1959 if (sb->s_bdev)
1960 invalidate_bdev(sb->s_bdev);
1961 put_inodes:
1962 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1963 if (toputinode[cnt]) {
1964 /* On remount RO, we keep the inode pointer so that we
1965 * can reenable quota on the subsequent remount RW. We
1966 * have to check 'flags' variable and not use sb_has_
1967 * function because another quotaon / quotaoff could
1968 * change global state before we got here. We refuse
1969 * to suspend quotas when there is pending delete on
1970 * the quota file... */
1971 if (!(flags & DQUOT_SUSPENDED))
1972 iput(toputinode[cnt]);
1973 else if (!toputinode[cnt]->i_nlink)
1974 ret = -EBUSY;
1976 return ret;
1978 EXPORT_SYMBOL(vfs_quota_disable);
1980 int vfs_quota_off(struct super_block *sb, int type, int remount)
1982 return vfs_quota_disable(sb, type, remount ? DQUOT_SUSPENDED :
1983 (DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED));
1985 EXPORT_SYMBOL(vfs_quota_off);
1987 * Turn quotas on on a device
1991 * Helper function to turn quotas on when we already have the inode of
1992 * quota file and no quota information is loaded.
1994 static int vfs_load_quota_inode(struct inode *inode, int type, int format_id,
1995 unsigned int flags)
1997 struct quota_format_type *fmt = find_quota_format(format_id);
1998 struct super_block *sb = inode->i_sb;
1999 struct quota_info *dqopt = sb_dqopt(sb);
2000 int error;
2001 int oldflags = -1;
2003 if (!fmt)
2004 return -ESRCH;
2005 if (!S_ISREG(inode->i_mode)) {
2006 error = -EACCES;
2007 goto out_fmt;
2009 if (IS_RDONLY(inode)) {
2010 error = -EROFS;
2011 goto out_fmt;
2013 if (!sb->s_op->quota_write || !sb->s_op->quota_read) {
2014 error = -EINVAL;
2015 goto out_fmt;
2017 /* Usage always has to be set... */
2018 if (!(flags & DQUOT_USAGE_ENABLED)) {
2019 error = -EINVAL;
2020 goto out_fmt;
2023 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2024 /* As we bypass the pagecache we must now flush all the
2025 * dirty data and invalidate caches so that kernel sees
2026 * changes from userspace. It is not enough to just flush
2027 * the quota file since if blocksize < pagesize, invalidation
2028 * of the cache could fail because of other unrelated dirty
2029 * data */
2030 sync_filesystem(sb);
2031 invalidate_bdev(sb->s_bdev);
2033 mutex_lock(&dqopt->dqonoff_mutex);
2034 if (sb_has_quota_loaded(sb, type)) {
2035 error = -EBUSY;
2036 goto out_lock;
2039 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2040 /* We don't want quota and atime on quota files (deadlocks
2041 * possible) Also nobody should write to the file - we use
2042 * special IO operations which ignore the immutable bit. */
2043 mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
2044 oldflags = inode->i_flags & (S_NOATIME | S_IMMUTABLE |
2045 S_NOQUOTA);
2046 inode->i_flags |= S_NOQUOTA | S_NOATIME | S_IMMUTABLE;
2047 mutex_unlock(&inode->i_mutex);
2049 * When S_NOQUOTA is set, remove dquot references as no more
2050 * references can be added
2052 __dquot_drop(inode);
2055 error = -EIO;
2056 dqopt->files[type] = igrab(inode);
2057 if (!dqopt->files[type])
2058 goto out_lock;
2059 error = -EINVAL;
2060 if (!fmt->qf_ops->check_quota_file(sb, type))
2061 goto out_file_init;
2063 dqopt->ops[type] = fmt->qf_ops;
2064 dqopt->info[type].dqi_format = fmt;
2065 dqopt->info[type].dqi_fmt_id = format_id;
2066 INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2067 mutex_lock(&dqopt->dqio_mutex);
2068 error = dqopt->ops[type]->read_file_info(sb, type);
2069 if (error < 0) {
2070 mutex_unlock(&dqopt->dqio_mutex);
2071 goto out_file_init;
2073 mutex_unlock(&dqopt->dqio_mutex);
2074 spin_lock(&dq_state_lock);
2075 dqopt->flags |= dquot_state_flag(flags, type);
2076 spin_unlock(&dq_state_lock);
2078 add_dquot_ref(sb, type);
2079 mutex_unlock(&dqopt->dqonoff_mutex);
2081 return 0;
2083 out_file_init:
2084 dqopt->files[type] = NULL;
2085 iput(inode);
2086 out_lock:
2087 if (oldflags != -1) {
2088 mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
2089 /* Set the flags back (in the case of accidental quotaon()
2090 * on a wrong file we don't want to mess up the flags) */
2091 inode->i_flags &= ~(S_NOATIME | S_NOQUOTA | S_IMMUTABLE);
2092 inode->i_flags |= oldflags;
2093 mutex_unlock(&inode->i_mutex);
2095 mutex_unlock(&dqopt->dqonoff_mutex);
2096 out_fmt:
2097 put_quota_format(fmt);
2099 return error;
2102 /* Reenable quotas on remount RW */
2103 static int vfs_quota_on_remount(struct super_block *sb, int type)
2105 struct quota_info *dqopt = sb_dqopt(sb);
2106 struct inode *inode;
2107 int ret;
2108 unsigned int flags;
2110 mutex_lock(&dqopt->dqonoff_mutex);
2111 if (!sb_has_quota_suspended(sb, type)) {
2112 mutex_unlock(&dqopt->dqonoff_mutex);
2113 return 0;
2115 inode = dqopt->files[type];
2116 dqopt->files[type] = NULL;
2117 spin_lock(&dq_state_lock);
2118 flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2119 DQUOT_LIMITS_ENABLED, type);
2120 dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, type);
2121 spin_unlock(&dq_state_lock);
2122 mutex_unlock(&dqopt->dqonoff_mutex);
2124 flags = dquot_generic_flag(flags, type);
2125 ret = vfs_load_quota_inode(inode, type, dqopt->info[type].dqi_fmt_id,
2126 flags);
2127 iput(inode);
2129 return ret;
2132 int vfs_quota_on_path(struct super_block *sb, int type, int format_id,
2133 struct path *path)
2135 int error = security_quota_on(path->dentry);
2136 if (error)
2137 return error;
2138 /* Quota file not on the same filesystem? */
2139 if (path->mnt->mnt_sb != sb)
2140 error = -EXDEV;
2141 else
2142 error = vfs_load_quota_inode(path->dentry->d_inode, type,
2143 format_id, DQUOT_USAGE_ENABLED |
2144 DQUOT_LIMITS_ENABLED);
2145 return error;
2147 EXPORT_SYMBOL(vfs_quota_on_path);
2149 int vfs_quota_on(struct super_block *sb, int type, int format_id, char *name,
2150 int remount)
2152 struct path path;
2153 int error;
2155 if (remount)
2156 return vfs_quota_on_remount(sb, type);
2158 error = kern_path(name, LOOKUP_FOLLOW, &path);
2159 if (!error) {
2160 error = vfs_quota_on_path(sb, type, format_id, &path);
2161 path_put(&path);
2163 return error;
2165 EXPORT_SYMBOL(vfs_quota_on);
2168 * More powerful function for turning on quotas allowing setting
2169 * of individual quota flags
2171 int vfs_quota_enable(struct inode *inode, int type, int format_id,
2172 unsigned int flags)
2174 int ret = 0;
2175 struct super_block *sb = inode->i_sb;
2176 struct quota_info *dqopt = sb_dqopt(sb);
2178 /* Just unsuspend quotas? */
2179 if (flags & DQUOT_SUSPENDED)
2180 return vfs_quota_on_remount(sb, type);
2181 if (!flags)
2182 return 0;
2183 /* Just updating flags needed? */
2184 if (sb_has_quota_loaded(sb, type)) {
2185 mutex_lock(&dqopt->dqonoff_mutex);
2186 /* Now do a reliable test... */
2187 if (!sb_has_quota_loaded(sb, type)) {
2188 mutex_unlock(&dqopt->dqonoff_mutex);
2189 goto load_quota;
2191 if (flags & DQUOT_USAGE_ENABLED &&
2192 sb_has_quota_usage_enabled(sb, type)) {
2193 ret = -EBUSY;
2194 goto out_lock;
2196 if (flags & DQUOT_LIMITS_ENABLED &&
2197 sb_has_quota_limits_enabled(sb, type)) {
2198 ret = -EBUSY;
2199 goto out_lock;
2201 spin_lock(&dq_state_lock);
2202 sb_dqopt(sb)->flags |= dquot_state_flag(flags, type);
2203 spin_unlock(&dq_state_lock);
2204 out_lock:
2205 mutex_unlock(&dqopt->dqonoff_mutex);
2206 return ret;
2209 load_quota:
2210 return vfs_load_quota_inode(inode, type, format_id, flags);
2212 EXPORT_SYMBOL(vfs_quota_enable);
2215 * This function is used when filesystem needs to initialize quotas
2216 * during mount time.
2218 int vfs_quota_on_mount(struct super_block *sb, char *qf_name,
2219 int format_id, int type)
2221 struct dentry *dentry;
2222 int error;
2224 mutex_lock(&sb->s_root->d_inode->i_mutex);
2225 dentry = lookup_one_len(qf_name, sb->s_root, strlen(qf_name));
2226 mutex_unlock(&sb->s_root->d_inode->i_mutex);
2227 if (IS_ERR(dentry))
2228 return PTR_ERR(dentry);
2230 if (!dentry->d_inode) {
2231 error = -ENOENT;
2232 goto out;
2235 error = security_quota_on(dentry);
2236 if (!error)
2237 error = vfs_load_quota_inode(dentry->d_inode, type, format_id,
2238 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2240 out:
2241 dput(dentry);
2242 return error;
2244 EXPORT_SYMBOL(vfs_quota_on_mount);
2246 /* Wrapper to turn on quotas when remounting rw */
2247 int vfs_dq_quota_on_remount(struct super_block *sb)
2249 int cnt;
2250 int ret = 0, err;
2252 if (!sb->s_qcop || !sb->s_qcop->quota_on)
2253 return -ENOSYS;
2254 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2255 err = sb->s_qcop->quota_on(sb, cnt, 0, NULL, 1);
2256 if (err < 0 && !ret)
2257 ret = err;
2259 return ret;
2261 EXPORT_SYMBOL(vfs_dq_quota_on_remount);
2263 static inline qsize_t qbtos(qsize_t blocks)
2265 return blocks << QIF_DQBLKSIZE_BITS;
2268 static inline qsize_t stoqb(qsize_t space)
2270 return (space + QIF_DQBLKSIZE - 1) >> QIF_DQBLKSIZE_BITS;
2273 /* Generic routine for getting common part of quota structure */
2274 static void do_get_dqblk(struct dquot *dquot, struct if_dqblk *di)
2276 struct mem_dqblk *dm = &dquot->dq_dqb;
2278 spin_lock(&dq_data_lock);
2279 di->dqb_bhardlimit = stoqb(dm->dqb_bhardlimit);
2280 di->dqb_bsoftlimit = stoqb(dm->dqb_bsoftlimit);
2281 di->dqb_curspace = dm->dqb_curspace + dm->dqb_rsvspace;
2282 di->dqb_ihardlimit = dm->dqb_ihardlimit;
2283 di->dqb_isoftlimit = dm->dqb_isoftlimit;
2284 di->dqb_curinodes = dm->dqb_curinodes;
2285 di->dqb_btime = dm->dqb_btime;
2286 di->dqb_itime = dm->dqb_itime;
2287 di->dqb_valid = QIF_ALL;
2288 spin_unlock(&dq_data_lock);
2291 int vfs_get_dqblk(struct super_block *sb, int type, qid_t id,
2292 struct if_dqblk *di)
2294 struct dquot *dquot;
2296 dquot = dqget(sb, id, type);
2297 if (!dquot)
2298 return -ESRCH;
2299 do_get_dqblk(dquot, di);
2300 dqput(dquot);
2302 return 0;
2304 EXPORT_SYMBOL(vfs_get_dqblk);
2306 /* Generic routine for setting common part of quota structure */
2307 static int do_set_dqblk(struct dquot *dquot, struct if_dqblk *di)
2309 struct mem_dqblk *dm = &dquot->dq_dqb;
2310 int check_blim = 0, check_ilim = 0;
2311 struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_type];
2313 if ((di->dqb_valid & QIF_BLIMITS &&
2314 (di->dqb_bhardlimit > dqi->dqi_maxblimit ||
2315 di->dqb_bsoftlimit > dqi->dqi_maxblimit)) ||
2316 (di->dqb_valid & QIF_ILIMITS &&
2317 (di->dqb_ihardlimit > dqi->dqi_maxilimit ||
2318 di->dqb_isoftlimit > dqi->dqi_maxilimit)))
2319 return -ERANGE;
2321 spin_lock(&dq_data_lock);
2322 if (di->dqb_valid & QIF_SPACE) {
2323 dm->dqb_curspace = di->dqb_curspace - dm->dqb_rsvspace;
2324 check_blim = 1;
2325 __set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2327 if (di->dqb_valid & QIF_BLIMITS) {
2328 dm->dqb_bsoftlimit = qbtos(di->dqb_bsoftlimit);
2329 dm->dqb_bhardlimit = qbtos(di->dqb_bhardlimit);
2330 check_blim = 1;
2331 __set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2333 if (di->dqb_valid & QIF_INODES) {
2334 dm->dqb_curinodes = di->dqb_curinodes;
2335 check_ilim = 1;
2336 __set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2338 if (di->dqb_valid & QIF_ILIMITS) {
2339 dm->dqb_isoftlimit = di->dqb_isoftlimit;
2340 dm->dqb_ihardlimit = di->dqb_ihardlimit;
2341 check_ilim = 1;
2342 __set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2344 if (di->dqb_valid & QIF_BTIME) {
2345 dm->dqb_btime = di->dqb_btime;
2346 check_blim = 1;
2347 __set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2349 if (di->dqb_valid & QIF_ITIME) {
2350 dm->dqb_itime = di->dqb_itime;
2351 check_ilim = 1;
2352 __set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2355 if (check_blim) {
2356 if (!dm->dqb_bsoftlimit ||
2357 dm->dqb_curspace < dm->dqb_bsoftlimit) {
2358 dm->dqb_btime = 0;
2359 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2360 } else if (!(di->dqb_valid & QIF_BTIME))
2361 /* Set grace only if user hasn't provided his own... */
2362 dm->dqb_btime = get_seconds() + dqi->dqi_bgrace;
2364 if (check_ilim) {
2365 if (!dm->dqb_isoftlimit ||
2366 dm->dqb_curinodes < dm->dqb_isoftlimit) {
2367 dm->dqb_itime = 0;
2368 clear_bit(DQ_INODES_B, &dquot->dq_flags);
2369 } else if (!(di->dqb_valid & QIF_ITIME))
2370 /* Set grace only if user hasn't provided his own... */
2371 dm->dqb_itime = get_seconds() + dqi->dqi_igrace;
2373 if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2374 dm->dqb_isoftlimit)
2375 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2376 else
2377 set_bit(DQ_FAKE_B, &dquot->dq_flags);
2378 spin_unlock(&dq_data_lock);
2379 mark_dquot_dirty(dquot);
2381 return 0;
2384 int vfs_set_dqblk(struct super_block *sb, int type, qid_t id,
2385 struct if_dqblk *di)
2387 struct dquot *dquot;
2388 int rc;
2390 dquot = dqget(sb, id, type);
2391 if (!dquot) {
2392 rc = -ESRCH;
2393 goto out;
2395 rc = do_set_dqblk(dquot, di);
2396 dqput(dquot);
2397 out:
2398 return rc;
2400 EXPORT_SYMBOL(vfs_set_dqblk);
2402 /* Generic routine for getting common part of quota file information */
2403 int vfs_get_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
2405 struct mem_dqinfo *mi;
2407 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
2408 if (!sb_has_quota_active(sb, type)) {
2409 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2410 return -ESRCH;
2412 mi = sb_dqopt(sb)->info + type;
2413 spin_lock(&dq_data_lock);
2414 ii->dqi_bgrace = mi->dqi_bgrace;
2415 ii->dqi_igrace = mi->dqi_igrace;
2416 ii->dqi_flags = mi->dqi_flags & DQF_MASK;
2417 ii->dqi_valid = IIF_ALL;
2418 spin_unlock(&dq_data_lock);
2419 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2420 return 0;
2422 EXPORT_SYMBOL(vfs_get_dqinfo);
2424 /* Generic routine for setting common part of quota file information */
2425 int vfs_set_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
2427 struct mem_dqinfo *mi;
2428 int err = 0;
2430 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
2431 if (!sb_has_quota_active(sb, type)) {
2432 err = -ESRCH;
2433 goto out;
2435 mi = sb_dqopt(sb)->info + type;
2436 spin_lock(&dq_data_lock);
2437 if (ii->dqi_valid & IIF_BGRACE)
2438 mi->dqi_bgrace = ii->dqi_bgrace;
2439 if (ii->dqi_valid & IIF_IGRACE)
2440 mi->dqi_igrace = ii->dqi_igrace;
2441 if (ii->dqi_valid & IIF_FLAGS)
2442 mi->dqi_flags = (mi->dqi_flags & ~DQF_MASK) |
2443 (ii->dqi_flags & DQF_MASK);
2444 spin_unlock(&dq_data_lock);
2445 mark_info_dirty(sb, type);
2446 /* Force write to disk */
2447 sb->dq_op->write_info(sb, type);
2448 out:
2449 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2450 return err;
2452 EXPORT_SYMBOL(vfs_set_dqinfo);
2454 const struct quotactl_ops vfs_quotactl_ops = {
2455 .quota_on = vfs_quota_on,
2456 .quota_off = vfs_quota_off,
2457 .quota_sync = vfs_quota_sync,
2458 .get_info = vfs_get_dqinfo,
2459 .set_info = vfs_set_dqinfo,
2460 .get_dqblk = vfs_get_dqblk,
2461 .set_dqblk = vfs_set_dqblk
2464 static ctl_table fs_dqstats_table[] = {
2466 .procname = "lookups",
2467 .data = &dqstats.lookups,
2468 .maxlen = sizeof(int),
2469 .mode = 0444,
2470 .proc_handler = proc_dointvec,
2473 .procname = "drops",
2474 .data = &dqstats.drops,
2475 .maxlen = sizeof(int),
2476 .mode = 0444,
2477 .proc_handler = proc_dointvec,
2480 .procname = "reads",
2481 .data = &dqstats.reads,
2482 .maxlen = sizeof(int),
2483 .mode = 0444,
2484 .proc_handler = proc_dointvec,
2487 .procname = "writes",
2488 .data = &dqstats.writes,
2489 .maxlen = sizeof(int),
2490 .mode = 0444,
2491 .proc_handler = proc_dointvec,
2494 .procname = "cache_hits",
2495 .data = &dqstats.cache_hits,
2496 .maxlen = sizeof(int),
2497 .mode = 0444,
2498 .proc_handler = proc_dointvec,
2501 .procname = "allocated_dquots",
2502 .data = &dqstats.allocated_dquots,
2503 .maxlen = sizeof(int),
2504 .mode = 0444,
2505 .proc_handler = proc_dointvec,
2508 .procname = "free_dquots",
2509 .data = &dqstats.free_dquots,
2510 .maxlen = sizeof(int),
2511 .mode = 0444,
2512 .proc_handler = proc_dointvec,
2515 .procname = "syncs",
2516 .data = &dqstats.syncs,
2517 .maxlen = sizeof(int),
2518 .mode = 0444,
2519 .proc_handler = proc_dointvec,
2521 #ifdef CONFIG_PRINT_QUOTA_WARNING
2523 .procname = "warnings",
2524 .data = &flag_print_warnings,
2525 .maxlen = sizeof(int),
2526 .mode = 0644,
2527 .proc_handler = proc_dointvec,
2529 #endif
2530 { },
2533 static ctl_table fs_table[] = {
2535 .procname = "quota",
2536 .mode = 0555,
2537 .child = fs_dqstats_table,
2539 { },
2542 static ctl_table sys_table[] = {
2544 .procname = "fs",
2545 .mode = 0555,
2546 .child = fs_table,
2548 { },
2551 static int __init dquot_init(void)
2553 int i;
2554 unsigned long nr_hash, order;
2556 printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2558 register_sysctl_table(sys_table);
2560 dquot_cachep = kmem_cache_create("dquot",
2561 sizeof(struct dquot), sizeof(unsigned long) * 4,
2562 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2563 SLAB_MEM_SPREAD|SLAB_PANIC),
2564 NULL);
2566 order = 0;
2567 dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
2568 if (!dquot_hash)
2569 panic("Cannot create dquot hash table");
2571 /* Find power-of-two hlist_heads which can fit into allocation */
2572 nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2573 dq_hash_bits = 0;
2574 do {
2575 dq_hash_bits++;
2576 } while (nr_hash >> dq_hash_bits);
2577 dq_hash_bits--;
2579 nr_hash = 1UL << dq_hash_bits;
2580 dq_hash_mask = nr_hash - 1;
2581 for (i = 0; i < nr_hash; i++)
2582 INIT_HLIST_HEAD(dquot_hash + i);
2584 printk("Dquot-cache hash table entries: %ld (order %ld, %ld bytes)\n",
2585 nr_hash, order, (PAGE_SIZE << order));
2587 register_shrinker(&dqcache_shrinker);
2589 return 0;
2591 module_init(dquot_init);