x86/paravirt: implement PVOP_CALL macros for callee-save functions
[linux-2.6/kvm.git] / fs / dquot.c
blobbca3cac4bee76e831b71f6b59c13641b91e48103
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.. */
80 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
81 #include <net/netlink.h>
82 #include <net/genetlink.h>
83 #endif
85 #include <asm/uaccess.h>
87 #define __DQUOT_PARANOIA
90 * There are three quota SMP locks. dq_list_lock protects all lists with quotas
91 * and quota formats, dqstats structure containing statistics about the lists
92 * dq_data_lock protects data from dq_dqb and also mem_dqinfo structures and
93 * also guards consistency of dquot->dq_dqb with inode->i_blocks, i_bytes.
94 * i_blocks and i_bytes updates itself are guarded by i_lock acquired directly
95 * in inode_add_bytes() and inode_sub_bytes(). dq_state_lock protects
96 * modifications of quota state (on quotaon and quotaoff) and readers who care
97 * about latest values take it as well.
99 * The spinlock ordering is hence: dq_data_lock > dq_list_lock > i_lock,
100 * dq_list_lock > dq_state_lock
102 * Note that some things (eg. sb pointer, type, id) doesn't change during
103 * the life of the dquot structure and so needn't to be protected by a lock
105 * Any operation working on dquots via inode pointers must hold dqptr_sem. If
106 * operation is just reading pointers from inode (or not using them at all) the
107 * read lock is enough. If pointers are altered function must hold write lock
108 * (these locking rules also apply for S_NOQUOTA flag in the inode - note that
109 * for altering the flag i_mutex is also needed).
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 DEFINE_SPINLOCK(dq_list_lock);
133 static DEFINE_SPINLOCK(dq_state_lock);
134 DEFINE_SPINLOCK(dq_data_lock);
136 static char *quotatypes[] = INITQFNAMES;
137 static struct quota_format_type *quota_formats; /* List of registered formats */
138 static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
140 /* SLAB cache for dquot structures */
141 static struct kmem_cache *dquot_cachep;
143 int register_quota_format(struct quota_format_type *fmt)
145 spin_lock(&dq_list_lock);
146 fmt->qf_next = quota_formats;
147 quota_formats = fmt;
148 spin_unlock(&dq_list_lock);
149 return 0;
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; actqf = &(*actqf)->qf_next);
158 if (*actqf)
159 *actqf = (*actqf)->qf_next;
160 spin_unlock(&dq_list_lock);
163 static struct quota_format_type *find_quota_format(int id)
165 struct quota_format_type *actqf;
167 spin_lock(&dq_list_lock);
168 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id; actqf = actqf->qf_next);
169 if (!actqf || !try_module_get(actqf->qf_owner)) {
170 int qm;
172 spin_unlock(&dq_list_lock);
174 for (qm = 0; module_names[qm].qm_fmt_id && module_names[qm].qm_fmt_id != id; qm++);
175 if (!module_names[qm].qm_fmt_id || request_module(module_names[qm].qm_mod_name))
176 return NULL;
178 spin_lock(&dq_list_lock);
179 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id; actqf = actqf->qf_next);
180 if (actqf && !try_module_get(actqf->qf_owner))
181 actqf = NULL;
183 spin_unlock(&dq_list_lock);
184 return actqf;
187 static void put_quota_format(struct quota_format_type *fmt)
189 module_put(fmt->qf_owner);
193 * Dquot List Management:
194 * The quota code uses three lists for dquot management: the inuse_list,
195 * free_dquots, and dquot_hash[] array. A single dquot structure may be
196 * on all three lists, depending on its current state.
198 * All dquots are placed to the end of inuse_list when first created, and this
199 * list is used for invalidate operation, which must look at every dquot.
201 * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
202 * and this list is searched whenever we need an available dquot. Dquots are
203 * removed from the list as soon as they are used again, and
204 * dqstats.free_dquots gives the number of dquots on the list. When
205 * dquot is invalidated it's completely released from memory.
207 * Dquots with a specific identity (device, type and id) are placed on
208 * one of the dquot_hash[] hash chains. The provides an efficient search
209 * mechanism to locate a specific dquot.
212 static LIST_HEAD(inuse_list);
213 static LIST_HEAD(free_dquots);
214 static unsigned int dq_hash_bits, dq_hash_mask;
215 static struct hlist_head *dquot_hash;
217 struct dqstats dqstats;
219 static inline unsigned int
220 hashfn(const struct super_block *sb, unsigned int id, int type)
222 unsigned long tmp;
224 tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
225 return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
229 * Following list functions expect dq_list_lock to be held
231 static inline void insert_dquot_hash(struct dquot *dquot)
233 struct hlist_head *head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id, dquot->dq_type);
234 hlist_add_head(&dquot->dq_hash, head);
237 static inline void remove_dquot_hash(struct dquot *dquot)
239 hlist_del_init(&dquot->dq_hash);
242 static inline struct dquot *find_dquot(unsigned int hashent, struct super_block *sb, unsigned int id, int type)
244 struct hlist_node *node;
245 struct dquot *dquot;
247 hlist_for_each (node, dquot_hash+hashent) {
248 dquot = hlist_entry(node, struct dquot, dq_hash);
249 if (dquot->dq_sb == sb && dquot->dq_id == id && dquot->dq_type == type)
250 return dquot;
252 return NODQUOT;
255 /* Add a dquot to the tail of the free list */
256 static inline void put_dquot_last(struct dquot *dquot)
258 list_add_tail(&dquot->dq_free, &free_dquots);
259 dqstats.free_dquots++;
262 static inline void remove_free_dquot(struct dquot *dquot)
264 if (list_empty(&dquot->dq_free))
265 return;
266 list_del_init(&dquot->dq_free);
267 dqstats.free_dquots--;
270 static inline void put_inuse(struct dquot *dquot)
272 /* We add to the back of inuse list so we don't have to restart
273 * when traversing this list and we block */
274 list_add_tail(&dquot->dq_inuse, &inuse_list);
275 dqstats.allocated_dquots++;
278 static inline void remove_inuse(struct dquot *dquot)
280 dqstats.allocated_dquots--;
281 list_del(&dquot->dq_inuse);
284 * End of list functions needing dq_list_lock
287 static void wait_on_dquot(struct dquot *dquot)
289 mutex_lock(&dquot->dq_lock);
290 mutex_unlock(&dquot->dq_lock);
293 static inline int dquot_dirty(struct dquot *dquot)
295 return test_bit(DQ_MOD_B, &dquot->dq_flags);
298 static inline int mark_dquot_dirty(struct dquot *dquot)
300 return dquot->dq_sb->dq_op->mark_dirty(dquot);
303 int dquot_mark_dquot_dirty(struct dquot *dquot)
305 spin_lock(&dq_list_lock);
306 if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags))
307 list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
308 info[dquot->dq_type].dqi_dirty_list);
309 spin_unlock(&dq_list_lock);
310 return 0;
313 /* This function needs dq_list_lock */
314 static inline int clear_dquot_dirty(struct dquot *dquot)
316 if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags))
317 return 0;
318 list_del_init(&dquot->dq_dirty);
319 return 1;
322 void mark_info_dirty(struct super_block *sb, int type)
324 set_bit(DQF_INFO_DIRTY_B, &sb_dqopt(sb)->info[type].dqi_flags);
326 EXPORT_SYMBOL(mark_info_dirty);
329 * Read dquot from disk and alloc space for it
332 int dquot_acquire(struct dquot *dquot)
334 int ret = 0, ret2 = 0;
335 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
337 mutex_lock(&dquot->dq_lock);
338 mutex_lock(&dqopt->dqio_mutex);
339 if (!test_bit(DQ_READ_B, &dquot->dq_flags))
340 ret = dqopt->ops[dquot->dq_type]->read_dqblk(dquot);
341 if (ret < 0)
342 goto out_iolock;
343 set_bit(DQ_READ_B, &dquot->dq_flags);
344 /* Instantiate dquot if needed */
345 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
346 ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
347 /* Write the info if needed */
348 if (info_dirty(&dqopt->info[dquot->dq_type]))
349 ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
350 if (ret < 0)
351 goto out_iolock;
352 if (ret2 < 0) {
353 ret = ret2;
354 goto out_iolock;
357 set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
358 out_iolock:
359 mutex_unlock(&dqopt->dqio_mutex);
360 mutex_unlock(&dquot->dq_lock);
361 return ret;
365 * Write dquot to disk
367 int dquot_commit(struct dquot *dquot)
369 int ret = 0, ret2 = 0;
370 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
372 mutex_lock(&dqopt->dqio_mutex);
373 spin_lock(&dq_list_lock);
374 if (!clear_dquot_dirty(dquot)) {
375 spin_unlock(&dq_list_lock);
376 goto out_sem;
378 spin_unlock(&dq_list_lock);
379 /* Inactive dquot can be only if there was error during read/init
380 * => we have better not writing it */
381 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
382 ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
383 if (info_dirty(&dqopt->info[dquot->dq_type]))
384 ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
385 if (ret >= 0)
386 ret = ret2;
388 out_sem:
389 mutex_unlock(&dqopt->dqio_mutex);
390 return ret;
394 * Release dquot
396 int dquot_release(struct dquot *dquot)
398 int ret = 0, ret2 = 0;
399 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
401 mutex_lock(&dquot->dq_lock);
402 /* Check whether we are not racing with some other dqget() */
403 if (atomic_read(&dquot->dq_count) > 1)
404 goto out_dqlock;
405 mutex_lock(&dqopt->dqio_mutex);
406 if (dqopt->ops[dquot->dq_type]->release_dqblk) {
407 ret = dqopt->ops[dquot->dq_type]->release_dqblk(dquot);
408 /* Write the info */
409 if (info_dirty(&dqopt->info[dquot->dq_type]))
410 ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
411 if (ret >= 0)
412 ret = ret2;
414 clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
415 mutex_unlock(&dqopt->dqio_mutex);
416 out_dqlock:
417 mutex_unlock(&dquot->dq_lock);
418 return ret;
421 void dquot_destroy(struct dquot *dquot)
423 kmem_cache_free(dquot_cachep, dquot);
425 EXPORT_SYMBOL(dquot_destroy);
427 static inline void do_destroy_dquot(struct dquot *dquot)
429 dquot->dq_sb->dq_op->destroy_dquot(dquot);
432 /* Invalidate all dquots on the list. Note that this function is called after
433 * quota is disabled and pointers from inodes removed so there cannot be new
434 * quota users. There can still be some users of quotas due to inodes being
435 * just deleted or pruned by prune_icache() (those are not attached to any
436 * list) or parallel quotactl call. We have to wait for such users.
438 static void invalidate_dquots(struct super_block *sb, int type)
440 struct dquot *dquot, *tmp;
442 restart:
443 spin_lock(&dq_list_lock);
444 list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
445 if (dquot->dq_sb != sb)
446 continue;
447 if (dquot->dq_type != type)
448 continue;
449 /* Wait for dquot users */
450 if (atomic_read(&dquot->dq_count)) {
451 DEFINE_WAIT(wait);
453 atomic_inc(&dquot->dq_count);
454 prepare_to_wait(&dquot->dq_wait_unused, &wait,
455 TASK_UNINTERRUPTIBLE);
456 spin_unlock(&dq_list_lock);
457 /* Once dqput() wakes us up, we know it's time to free
458 * the dquot.
459 * IMPORTANT: we rely on the fact that there is always
460 * at most one process waiting for dquot to free.
461 * Otherwise dq_count would be > 1 and we would never
462 * wake up.
464 if (atomic_read(&dquot->dq_count) > 1)
465 schedule();
466 finish_wait(&dquot->dq_wait_unused, &wait);
467 dqput(dquot);
468 /* At this moment dquot() need not exist (it could be
469 * reclaimed by prune_dqcache(). Hence we must
470 * restart. */
471 goto restart;
474 * Quota now has no users and it has been written on last
475 * dqput()
477 remove_dquot_hash(dquot);
478 remove_free_dquot(dquot);
479 remove_inuse(dquot);
480 do_destroy_dquot(dquot);
482 spin_unlock(&dq_list_lock);
485 /* Call callback for every active dquot on given filesystem */
486 int dquot_scan_active(struct super_block *sb,
487 int (*fn)(struct dquot *dquot, unsigned long priv),
488 unsigned long priv)
490 struct dquot *dquot, *old_dquot = NULL;
491 int ret = 0;
493 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
494 spin_lock(&dq_list_lock);
495 list_for_each_entry(dquot, &inuse_list, dq_inuse) {
496 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
497 continue;
498 if (dquot->dq_sb != sb)
499 continue;
500 /* Now we have active dquot so we can just increase use count */
501 atomic_inc(&dquot->dq_count);
502 dqstats.lookups++;
503 spin_unlock(&dq_list_lock);
504 dqput(old_dquot);
505 old_dquot = dquot;
506 ret = fn(dquot, priv);
507 if (ret < 0)
508 goto out;
509 spin_lock(&dq_list_lock);
510 /* We are safe to continue now because our dquot could not
511 * be moved out of the inuse list while we hold the reference */
513 spin_unlock(&dq_list_lock);
514 out:
515 dqput(old_dquot);
516 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
517 return ret;
520 int vfs_quota_sync(struct super_block *sb, int type)
522 struct list_head *dirty;
523 struct dquot *dquot;
524 struct quota_info *dqopt = sb_dqopt(sb);
525 int cnt;
527 mutex_lock(&dqopt->dqonoff_mutex);
528 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
529 if (type != -1 && cnt != type)
530 continue;
531 if (!sb_has_quota_active(sb, cnt))
532 continue;
533 spin_lock(&dq_list_lock);
534 dirty = &dqopt->info[cnt].dqi_dirty_list;
535 while (!list_empty(dirty)) {
536 dquot = list_first_entry(dirty, struct dquot, dq_dirty);
537 /* Dirty and inactive can be only bad dquot... */
538 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
539 clear_dquot_dirty(dquot);
540 continue;
542 /* Now we have active dquot from which someone is
543 * holding reference so we can safely just increase
544 * use count */
545 atomic_inc(&dquot->dq_count);
546 dqstats.lookups++;
547 spin_unlock(&dq_list_lock);
548 sb->dq_op->write_dquot(dquot);
549 dqput(dquot);
550 spin_lock(&dq_list_lock);
552 spin_unlock(&dq_list_lock);
555 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
556 if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
557 && info_dirty(&dqopt->info[cnt]))
558 sb->dq_op->write_info(sb, cnt);
559 spin_lock(&dq_list_lock);
560 dqstats.syncs++;
561 spin_unlock(&dq_list_lock);
562 mutex_unlock(&dqopt->dqonoff_mutex);
564 return 0;
567 /* Free unused dquots from cache */
568 static void prune_dqcache(int count)
570 struct list_head *head;
571 struct dquot *dquot;
573 head = free_dquots.prev;
574 while (head != &free_dquots && count) {
575 dquot = list_entry(head, struct dquot, dq_free);
576 remove_dquot_hash(dquot);
577 remove_free_dquot(dquot);
578 remove_inuse(dquot);
579 do_destroy_dquot(dquot);
580 count--;
581 head = free_dquots.prev;
586 * This is called from kswapd when we think we need some
587 * more memory
590 static int shrink_dqcache_memory(int nr, gfp_t gfp_mask)
592 if (nr) {
593 spin_lock(&dq_list_lock);
594 prune_dqcache(nr);
595 spin_unlock(&dq_list_lock);
597 return (dqstats.free_dquots / 100) * sysctl_vfs_cache_pressure;
600 static struct shrinker dqcache_shrinker = {
601 .shrink = shrink_dqcache_memory,
602 .seeks = DEFAULT_SEEKS,
606 * Put reference to dquot
607 * NOTE: If you change this function please check whether dqput_blocks() works right...
609 void dqput(struct dquot *dquot)
611 int ret;
613 if (!dquot)
614 return;
615 #ifdef __DQUOT_PARANOIA
616 if (!atomic_read(&dquot->dq_count)) {
617 printk("VFS: dqput: trying to free free dquot\n");
618 printk("VFS: device %s, dquot of %s %d\n",
619 dquot->dq_sb->s_id,
620 quotatypes[dquot->dq_type],
621 dquot->dq_id);
622 BUG();
624 #endif
626 spin_lock(&dq_list_lock);
627 dqstats.drops++;
628 spin_unlock(&dq_list_lock);
629 we_slept:
630 spin_lock(&dq_list_lock);
631 if (atomic_read(&dquot->dq_count) > 1) {
632 /* We have more than one user... nothing to do */
633 atomic_dec(&dquot->dq_count);
634 /* Releasing dquot during quotaoff phase? */
635 if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_type) &&
636 atomic_read(&dquot->dq_count) == 1)
637 wake_up(&dquot->dq_wait_unused);
638 spin_unlock(&dq_list_lock);
639 return;
641 /* Need to release dquot? */
642 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && dquot_dirty(dquot)) {
643 spin_unlock(&dq_list_lock);
644 /* Commit dquot before releasing */
645 ret = dquot->dq_sb->dq_op->write_dquot(dquot);
646 if (ret < 0) {
647 printk(KERN_ERR "VFS: cannot write quota structure on "
648 "device %s (error %d). Quota may get out of "
649 "sync!\n", dquot->dq_sb->s_id, ret);
651 * We clear dirty bit anyway, so that we avoid
652 * infinite loop here
654 spin_lock(&dq_list_lock);
655 clear_dquot_dirty(dquot);
656 spin_unlock(&dq_list_lock);
658 goto we_slept;
660 /* Clear flag in case dquot was inactive (something bad happened) */
661 clear_dquot_dirty(dquot);
662 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
663 spin_unlock(&dq_list_lock);
664 dquot->dq_sb->dq_op->release_dquot(dquot);
665 goto we_slept;
667 atomic_dec(&dquot->dq_count);
668 #ifdef __DQUOT_PARANOIA
669 /* sanity check */
670 BUG_ON(!list_empty(&dquot->dq_free));
671 #endif
672 put_dquot_last(dquot);
673 spin_unlock(&dq_list_lock);
676 struct dquot *dquot_alloc(struct super_block *sb, int type)
678 return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
680 EXPORT_SYMBOL(dquot_alloc);
682 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
684 struct dquot *dquot;
686 dquot = sb->dq_op->alloc_dquot(sb, type);
687 if(!dquot)
688 return NODQUOT;
690 mutex_init(&dquot->dq_lock);
691 INIT_LIST_HEAD(&dquot->dq_free);
692 INIT_LIST_HEAD(&dquot->dq_inuse);
693 INIT_HLIST_NODE(&dquot->dq_hash);
694 INIT_LIST_HEAD(&dquot->dq_dirty);
695 init_waitqueue_head(&dquot->dq_wait_unused);
696 dquot->dq_sb = sb;
697 dquot->dq_type = type;
698 atomic_set(&dquot->dq_count, 1);
700 return dquot;
704 * Get reference to dquot
706 * Locking is slightly tricky here. We are guarded from parallel quotaoff()
707 * destroying our dquot by:
708 * a) checking for quota flags under dq_list_lock and
709 * b) getting a reference to dquot before we release dq_list_lock
711 struct dquot *dqget(struct super_block *sb, unsigned int id, int type)
713 unsigned int hashent = hashfn(sb, id, type);
714 struct dquot *dquot = NODQUOT, *empty = NODQUOT;
716 if (!sb_has_quota_active(sb, type))
717 return NODQUOT;
718 we_slept:
719 spin_lock(&dq_list_lock);
720 spin_lock(&dq_state_lock);
721 if (!sb_has_quota_active(sb, type)) {
722 spin_unlock(&dq_state_lock);
723 spin_unlock(&dq_list_lock);
724 goto out;
726 spin_unlock(&dq_state_lock);
728 if ((dquot = find_dquot(hashent, sb, id, type)) == NODQUOT) {
729 if (empty == NODQUOT) {
730 spin_unlock(&dq_list_lock);
731 if ((empty = get_empty_dquot(sb, type)) == NODQUOT)
732 schedule(); /* Try to wait for a moment... */
733 goto we_slept;
735 dquot = empty;
736 empty = NODQUOT;
737 dquot->dq_id = id;
738 /* all dquots go on the inuse_list */
739 put_inuse(dquot);
740 /* hash it first so it can be found */
741 insert_dquot_hash(dquot);
742 dqstats.lookups++;
743 spin_unlock(&dq_list_lock);
744 } else {
745 if (!atomic_read(&dquot->dq_count))
746 remove_free_dquot(dquot);
747 atomic_inc(&dquot->dq_count);
748 dqstats.cache_hits++;
749 dqstats.lookups++;
750 spin_unlock(&dq_list_lock);
752 /* Wait for dq_lock - after this we know that either dquot_release() is already
753 * finished or it will be canceled due to dq_count > 1 test */
754 wait_on_dquot(dquot);
755 /* Read the dquot and instantiate it (everything done only if needed) */
756 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && sb->dq_op->acquire_dquot(dquot) < 0) {
757 dqput(dquot);
758 dquot = NODQUOT;
759 goto out;
761 #ifdef __DQUOT_PARANOIA
762 BUG_ON(!dquot->dq_sb); /* Has somebody invalidated entry under us? */
763 #endif
764 out:
765 if (empty)
766 do_destroy_dquot(empty);
768 return dquot;
771 static int dqinit_needed(struct inode *inode, int type)
773 int cnt;
775 if (IS_NOQUOTA(inode))
776 return 0;
777 if (type != -1)
778 return inode->i_dquot[type] == NODQUOT;
779 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
780 if (inode->i_dquot[cnt] == NODQUOT)
781 return 1;
782 return 0;
785 /* This routine is guarded by dqonoff_mutex mutex */
786 static void add_dquot_ref(struct super_block *sb, int type)
788 struct inode *inode, *old_inode = NULL;
790 spin_lock(&inode_lock);
791 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
792 if (!atomic_read(&inode->i_writecount))
793 continue;
794 if (!dqinit_needed(inode, type))
795 continue;
796 if (inode->i_state & (I_FREEING|I_WILL_FREE))
797 continue;
799 __iget(inode);
800 spin_unlock(&inode_lock);
802 iput(old_inode);
803 sb->dq_op->initialize(inode, type);
804 /* We hold a reference to 'inode' so it couldn't have been
805 * removed from s_inodes list while we dropped the inode_lock.
806 * We cannot iput the inode now as we can be holding the last
807 * reference and we cannot iput it under inode_lock. So we
808 * keep the reference and iput it later. */
809 old_inode = inode;
810 spin_lock(&inode_lock);
812 spin_unlock(&inode_lock);
813 iput(old_inode);
816 /* Return 0 if dqput() won't block (note that 1 doesn't necessarily mean blocking) */
817 static inline int dqput_blocks(struct dquot *dquot)
819 if (atomic_read(&dquot->dq_count) <= 1)
820 return 1;
821 return 0;
824 /* Remove references to dquots from inode - add dquot to list for freeing if needed */
825 /* We can't race with anybody because we hold dqptr_sem for writing... */
826 static int remove_inode_dquot_ref(struct inode *inode, int type,
827 struct list_head *tofree_head)
829 struct dquot *dquot = inode->i_dquot[type];
831 inode->i_dquot[type] = NODQUOT;
832 if (dquot != NODQUOT) {
833 if (dqput_blocks(dquot)) {
834 #ifdef __DQUOT_PARANOIA
835 if (atomic_read(&dquot->dq_count) != 1)
836 printk(KERN_WARNING "VFS: Adding dquot with dq_count %d to dispose list.\n", atomic_read(&dquot->dq_count));
837 #endif
838 spin_lock(&dq_list_lock);
839 list_add(&dquot->dq_free, tofree_head); /* As dquot must have currently users it can't be on the free list... */
840 spin_unlock(&dq_list_lock);
841 return 1;
843 else
844 dqput(dquot); /* We have guaranteed we won't block */
846 return 0;
849 /* Free list of dquots - called from inode.c */
850 /* dquots are removed from inodes, no new references can be got so we are the only ones holding reference */
851 static void put_dquot_list(struct list_head *tofree_head)
853 struct list_head *act_head;
854 struct dquot *dquot;
856 act_head = tofree_head->next;
857 /* So now we have dquots on the list... Just free them */
858 while (act_head != tofree_head) {
859 dquot = list_entry(act_head, struct dquot, dq_free);
860 act_head = act_head->next;
861 list_del_init(&dquot->dq_free); /* Remove dquot from the list so we won't have problems... */
862 dqput(dquot);
866 static void remove_dquot_ref(struct super_block *sb, int type,
867 struct list_head *tofree_head)
869 struct inode *inode;
871 spin_lock(&inode_lock);
872 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
873 if (!IS_NOQUOTA(inode))
874 remove_inode_dquot_ref(inode, type, tofree_head);
876 spin_unlock(&inode_lock);
879 /* Gather all references from inodes and drop them */
880 static void drop_dquot_ref(struct super_block *sb, int type)
882 LIST_HEAD(tofree_head);
884 if (sb->dq_op) {
885 down_write(&sb_dqopt(sb)->dqptr_sem);
886 remove_dquot_ref(sb, type, &tofree_head);
887 up_write(&sb_dqopt(sb)->dqptr_sem);
888 put_dquot_list(&tofree_head);
892 static inline void dquot_incr_inodes(struct dquot *dquot, qsize_t number)
894 dquot->dq_dqb.dqb_curinodes += number;
897 static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
899 dquot->dq_dqb.dqb_curspace += number;
902 static inline void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
904 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
905 dquot->dq_dqb.dqb_curinodes >= number)
906 dquot->dq_dqb.dqb_curinodes -= number;
907 else
908 dquot->dq_dqb.dqb_curinodes = 0;
909 if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
910 dquot->dq_dqb.dqb_itime = (time_t) 0;
911 clear_bit(DQ_INODES_B, &dquot->dq_flags);
914 static inline void dquot_decr_space(struct dquot *dquot, qsize_t number)
916 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
917 dquot->dq_dqb.dqb_curspace >= number)
918 dquot->dq_dqb.dqb_curspace -= number;
919 else
920 dquot->dq_dqb.dqb_curspace = 0;
921 if (dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
922 dquot->dq_dqb.dqb_btime = (time_t) 0;
923 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
926 static int warning_issued(struct dquot *dquot, const int warntype)
928 int flag = (warntype == QUOTA_NL_BHARDWARN ||
929 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
930 ((warntype == QUOTA_NL_IHARDWARN ||
931 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
933 if (!flag)
934 return 0;
935 return test_and_set_bit(flag, &dquot->dq_flags);
938 #ifdef CONFIG_PRINT_QUOTA_WARNING
939 static int flag_print_warnings = 1;
941 static inline int need_print_warning(struct dquot *dquot)
943 if (!flag_print_warnings)
944 return 0;
946 switch (dquot->dq_type) {
947 case USRQUOTA:
948 return current_fsuid() == dquot->dq_id;
949 case GRPQUOTA:
950 return in_group_p(dquot->dq_id);
952 return 0;
955 /* Print warning to user which exceeded quota */
956 static void print_warning(struct dquot *dquot, const int warntype)
958 char *msg = NULL;
959 struct tty_struct *tty;
961 if (warntype == QUOTA_NL_IHARDBELOW ||
962 warntype == QUOTA_NL_ISOFTBELOW ||
963 warntype == QUOTA_NL_BHARDBELOW ||
964 warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(dquot))
965 return;
967 tty = get_current_tty();
968 if (!tty)
969 return;
970 tty_write_message(tty, dquot->dq_sb->s_id);
971 if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
972 tty_write_message(tty, ": warning, ");
973 else
974 tty_write_message(tty, ": write failed, ");
975 tty_write_message(tty, quotatypes[dquot->dq_type]);
976 switch (warntype) {
977 case QUOTA_NL_IHARDWARN:
978 msg = " file limit reached.\r\n";
979 break;
980 case QUOTA_NL_ISOFTLONGWARN:
981 msg = " file quota exceeded too long.\r\n";
982 break;
983 case QUOTA_NL_ISOFTWARN:
984 msg = " file quota exceeded.\r\n";
985 break;
986 case QUOTA_NL_BHARDWARN:
987 msg = " block limit reached.\r\n";
988 break;
989 case QUOTA_NL_BSOFTLONGWARN:
990 msg = " block quota exceeded too long.\r\n";
991 break;
992 case QUOTA_NL_BSOFTWARN:
993 msg = " block quota exceeded.\r\n";
994 break;
996 tty_write_message(tty, msg);
997 tty_kref_put(tty);
999 #endif
1001 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
1003 /* Netlink family structure for quota */
1004 static struct genl_family quota_genl_family = {
1005 .id = GENL_ID_GENERATE,
1006 .hdrsize = 0,
1007 .name = "VFS_DQUOT",
1008 .version = 1,
1009 .maxattr = QUOTA_NL_A_MAX,
1012 /* Send warning to userspace about user which exceeded quota */
1013 static void send_warning(const struct dquot *dquot, const char warntype)
1015 static atomic_t seq;
1016 struct sk_buff *skb;
1017 void *msg_head;
1018 int ret;
1019 int msg_size = 4 * nla_total_size(sizeof(u32)) +
1020 2 * nla_total_size(sizeof(u64));
1022 /* We have to allocate using GFP_NOFS as we are called from a
1023 * filesystem performing write and thus further recursion into
1024 * the fs to free some data could cause deadlocks. */
1025 skb = genlmsg_new(msg_size, GFP_NOFS);
1026 if (!skb) {
1027 printk(KERN_ERR
1028 "VFS: Not enough memory to send quota warning.\n");
1029 return;
1031 msg_head = genlmsg_put(skb, 0, atomic_add_return(1, &seq),
1032 &quota_genl_family, 0, QUOTA_NL_C_WARNING);
1033 if (!msg_head) {
1034 printk(KERN_ERR
1035 "VFS: Cannot store netlink header in quota warning.\n");
1036 goto err_out;
1038 ret = nla_put_u32(skb, QUOTA_NL_A_QTYPE, dquot->dq_type);
1039 if (ret)
1040 goto attr_err_out;
1041 ret = nla_put_u64(skb, QUOTA_NL_A_EXCESS_ID, dquot->dq_id);
1042 if (ret)
1043 goto attr_err_out;
1044 ret = nla_put_u32(skb, QUOTA_NL_A_WARNING, warntype);
1045 if (ret)
1046 goto attr_err_out;
1047 ret = nla_put_u32(skb, QUOTA_NL_A_DEV_MAJOR,
1048 MAJOR(dquot->dq_sb->s_dev));
1049 if (ret)
1050 goto attr_err_out;
1051 ret = nla_put_u32(skb, QUOTA_NL_A_DEV_MINOR,
1052 MINOR(dquot->dq_sb->s_dev));
1053 if (ret)
1054 goto attr_err_out;
1055 ret = nla_put_u64(skb, QUOTA_NL_A_CAUSED_ID, current_uid());
1056 if (ret)
1057 goto attr_err_out;
1058 genlmsg_end(skb, msg_head);
1060 ret = genlmsg_multicast(skb, 0, quota_genl_family.id, GFP_NOFS);
1061 if (ret < 0 && ret != -ESRCH)
1062 printk(KERN_ERR
1063 "VFS: Failed to send notification message: %d\n", ret);
1064 return;
1065 attr_err_out:
1066 printk(KERN_ERR "VFS: Not enough space to compose quota message!\n");
1067 err_out:
1068 kfree_skb(skb);
1070 #endif
1072 static inline void flush_warnings(struct dquot * const *dquots, char *warntype)
1074 int i;
1076 for (i = 0; i < MAXQUOTAS; i++)
1077 if (dquots[i] != NODQUOT && warntype[i] != QUOTA_NL_NOWARN &&
1078 !warning_issued(dquots[i], warntype[i])) {
1079 #ifdef CONFIG_PRINT_QUOTA_WARNING
1080 print_warning(dquots[i], warntype[i]);
1081 #endif
1082 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
1083 send_warning(dquots[i], warntype[i]);
1084 #endif
1088 static inline char ignore_hardlimit(struct dquot *dquot)
1090 struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_type];
1092 return capable(CAP_SYS_RESOURCE) &&
1093 (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD || !(info->dqi_flags & V1_DQF_RSQUASH));
1096 /* needs dq_data_lock */
1097 static int check_idq(struct dquot *dquot, qsize_t inodes, char *warntype)
1099 *warntype = QUOTA_NL_NOWARN;
1100 if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_type) ||
1101 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1102 return QUOTA_OK;
1104 if (dquot->dq_dqb.dqb_ihardlimit &&
1105 (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_ihardlimit &&
1106 !ignore_hardlimit(dquot)) {
1107 *warntype = QUOTA_NL_IHARDWARN;
1108 return NO_QUOTA;
1111 if (dquot->dq_dqb.dqb_isoftlimit &&
1112 (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
1113 dquot->dq_dqb.dqb_itime && get_seconds() >= dquot->dq_dqb.dqb_itime &&
1114 !ignore_hardlimit(dquot)) {
1115 *warntype = QUOTA_NL_ISOFTLONGWARN;
1116 return NO_QUOTA;
1119 if (dquot->dq_dqb.dqb_isoftlimit &&
1120 (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
1121 dquot->dq_dqb.dqb_itime == 0) {
1122 *warntype = QUOTA_NL_ISOFTWARN;
1123 dquot->dq_dqb.dqb_itime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
1126 return QUOTA_OK;
1129 /* needs dq_data_lock */
1130 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc, char *warntype)
1132 *warntype = QUOTA_NL_NOWARN;
1133 if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_type) ||
1134 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1135 return QUOTA_OK;
1137 if (dquot->dq_dqb.dqb_bhardlimit &&
1138 dquot->dq_dqb.dqb_curspace + space > dquot->dq_dqb.dqb_bhardlimit &&
1139 !ignore_hardlimit(dquot)) {
1140 if (!prealloc)
1141 *warntype = QUOTA_NL_BHARDWARN;
1142 return NO_QUOTA;
1145 if (dquot->dq_dqb.dqb_bsoftlimit &&
1146 dquot->dq_dqb.dqb_curspace + space > dquot->dq_dqb.dqb_bsoftlimit &&
1147 dquot->dq_dqb.dqb_btime && get_seconds() >= dquot->dq_dqb.dqb_btime &&
1148 !ignore_hardlimit(dquot)) {
1149 if (!prealloc)
1150 *warntype = QUOTA_NL_BSOFTLONGWARN;
1151 return NO_QUOTA;
1154 if (dquot->dq_dqb.dqb_bsoftlimit &&
1155 dquot->dq_dqb.dqb_curspace + space > dquot->dq_dqb.dqb_bsoftlimit &&
1156 dquot->dq_dqb.dqb_btime == 0) {
1157 if (!prealloc) {
1158 *warntype = QUOTA_NL_BSOFTWARN;
1159 dquot->dq_dqb.dqb_btime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_bgrace;
1161 else
1163 * We don't allow preallocation to exceed softlimit so exceeding will
1164 * be always printed
1166 return NO_QUOTA;
1169 return QUOTA_OK;
1172 static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1174 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1175 dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1176 !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_type))
1177 return QUOTA_NL_NOWARN;
1179 if (dquot->dq_dqb.dqb_curinodes - inodes <= dquot->dq_dqb.dqb_isoftlimit)
1180 return QUOTA_NL_ISOFTBELOW;
1181 if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1182 dquot->dq_dqb.dqb_curinodes - inodes < dquot->dq_dqb.dqb_ihardlimit)
1183 return QUOTA_NL_IHARDBELOW;
1184 return QUOTA_NL_NOWARN;
1187 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1189 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1190 dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1191 return QUOTA_NL_NOWARN;
1193 if (dquot->dq_dqb.dqb_curspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1194 return QUOTA_NL_BSOFTBELOW;
1195 if (dquot->dq_dqb.dqb_curspace >= dquot->dq_dqb.dqb_bhardlimit &&
1196 dquot->dq_dqb.dqb_curspace - space < dquot->dq_dqb.dqb_bhardlimit)
1197 return QUOTA_NL_BHARDBELOW;
1198 return QUOTA_NL_NOWARN;
1201 * Initialize quota pointers in inode
1202 * We do things in a bit complicated way but by that we avoid calling
1203 * dqget() and thus filesystem callbacks under dqptr_sem.
1205 int dquot_initialize(struct inode *inode, int type)
1207 unsigned int id = 0;
1208 int cnt, ret = 0;
1209 struct dquot *got[MAXQUOTAS] = { NODQUOT, NODQUOT };
1210 struct super_block *sb = inode->i_sb;
1212 /* First test before acquiring mutex - solves deadlocks when we
1213 * re-enter the quota code and are already holding the mutex */
1214 if (IS_NOQUOTA(inode))
1215 return 0;
1217 /* First get references to structures we might need. */
1218 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1219 if (type != -1 && cnt != type)
1220 continue;
1221 switch (cnt) {
1222 case USRQUOTA:
1223 id = inode->i_uid;
1224 break;
1225 case GRPQUOTA:
1226 id = inode->i_gid;
1227 break;
1229 got[cnt] = dqget(sb, id, cnt);
1232 down_write(&sb_dqopt(sb)->dqptr_sem);
1233 /* Having dqptr_sem we know NOQUOTA flags can't be altered... */
1234 if (IS_NOQUOTA(inode))
1235 goto out_err;
1236 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1237 if (type != -1 && cnt != type)
1238 continue;
1239 /* Avoid races with quotaoff() */
1240 if (!sb_has_quota_active(sb, cnt))
1241 continue;
1242 if (inode->i_dquot[cnt] == NODQUOT) {
1243 inode->i_dquot[cnt] = got[cnt];
1244 got[cnt] = NODQUOT;
1247 out_err:
1248 up_write(&sb_dqopt(sb)->dqptr_sem);
1249 /* Drop unused references */
1250 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1251 dqput(got[cnt]);
1252 return ret;
1256 * Release all quotas referenced by inode
1258 int dquot_drop(struct inode *inode)
1260 int cnt;
1261 struct dquot *put[MAXQUOTAS];
1263 down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1264 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1265 put[cnt] = inode->i_dquot[cnt];
1266 inode->i_dquot[cnt] = NODQUOT;
1268 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1270 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1271 dqput(put[cnt]);
1272 return 0;
1275 /* Wrapper to remove references to quota structures from inode */
1276 void vfs_dq_drop(struct inode *inode)
1278 /* Here we can get arbitrary inode from clear_inode() so we have
1279 * to be careful. OTOH we don't need locking as quota operations
1280 * are allowed to change only at mount time */
1281 if (!IS_NOQUOTA(inode) && inode->i_sb && inode->i_sb->dq_op
1282 && inode->i_sb->dq_op->drop) {
1283 int cnt;
1284 /* Test before calling to rule out calls from proc and such
1285 * where we are not allowed to block. Note that this is
1286 * actually reliable test even without the lock - the caller
1287 * must assure that nobody can come after the DQUOT_DROP and
1288 * add quota pointers back anyway */
1289 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1290 if (inode->i_dquot[cnt] != NODQUOT)
1291 break;
1292 if (cnt < MAXQUOTAS)
1293 inode->i_sb->dq_op->drop(inode);
1298 * Following four functions update i_blocks+i_bytes fields and
1299 * quota information (together with appropriate checks)
1300 * NOTE: We absolutely rely on the fact that caller dirties
1301 * the inode (usually macros in quotaops.h care about this) and
1302 * holds a handle for the current transaction so that dquot write and
1303 * inode write go into the same transaction.
1307 * This operation can block, but only after everything is updated
1309 int dquot_alloc_space(struct inode *inode, qsize_t number, int warn)
1311 int cnt, ret = NO_QUOTA;
1312 char warntype[MAXQUOTAS];
1314 /* First test before acquiring mutex - solves deadlocks when we
1315 * re-enter the quota code and are already holding the mutex */
1316 if (IS_NOQUOTA(inode)) {
1317 out_add:
1318 inode_add_bytes(inode, number);
1319 return QUOTA_OK;
1321 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1322 warntype[cnt] = QUOTA_NL_NOWARN;
1324 down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1325 if (IS_NOQUOTA(inode)) { /* Now we can do reliable test... */
1326 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1327 goto out_add;
1329 spin_lock(&dq_data_lock);
1330 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1331 if (inode->i_dquot[cnt] == NODQUOT)
1332 continue;
1333 if (check_bdq(inode->i_dquot[cnt], number, warn, warntype+cnt) == NO_QUOTA)
1334 goto warn_put_all;
1336 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1337 if (inode->i_dquot[cnt] == NODQUOT)
1338 continue;
1339 dquot_incr_space(inode->i_dquot[cnt], number);
1341 inode_add_bytes(inode, number);
1342 ret = QUOTA_OK;
1343 warn_put_all:
1344 spin_unlock(&dq_data_lock);
1345 if (ret == QUOTA_OK)
1346 /* Dirtify all the dquots - this can block when journalling */
1347 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1348 if (inode->i_dquot[cnt])
1349 mark_dquot_dirty(inode->i_dquot[cnt]);
1350 flush_warnings(inode->i_dquot, warntype);
1351 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1352 return ret;
1356 * This operation can block, but only after everything is updated
1358 int dquot_alloc_inode(const struct inode *inode, qsize_t number)
1360 int cnt, ret = NO_QUOTA;
1361 char warntype[MAXQUOTAS];
1363 /* First test before acquiring mutex - solves deadlocks when we
1364 * re-enter the quota code and are already holding the mutex */
1365 if (IS_NOQUOTA(inode))
1366 return QUOTA_OK;
1367 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1368 warntype[cnt] = QUOTA_NL_NOWARN;
1369 down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1370 if (IS_NOQUOTA(inode)) {
1371 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1372 return QUOTA_OK;
1374 spin_lock(&dq_data_lock);
1375 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1376 if (inode->i_dquot[cnt] == NODQUOT)
1377 continue;
1378 if (check_idq(inode->i_dquot[cnt], number, warntype+cnt) == NO_QUOTA)
1379 goto warn_put_all;
1382 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1383 if (inode->i_dquot[cnt] == NODQUOT)
1384 continue;
1385 dquot_incr_inodes(inode->i_dquot[cnt], number);
1387 ret = QUOTA_OK;
1388 warn_put_all:
1389 spin_unlock(&dq_data_lock);
1390 if (ret == QUOTA_OK)
1391 /* Dirtify all the dquots - this can block when journalling */
1392 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1393 if (inode->i_dquot[cnt])
1394 mark_dquot_dirty(inode->i_dquot[cnt]);
1395 flush_warnings(inode->i_dquot, warntype);
1396 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1397 return ret;
1401 * This operation can block, but only after everything is updated
1403 int dquot_free_space(struct inode *inode, qsize_t number)
1405 unsigned int cnt;
1406 char warntype[MAXQUOTAS];
1408 /* First test before acquiring mutex - solves deadlocks when we
1409 * re-enter the quota code and are already holding the mutex */
1410 if (IS_NOQUOTA(inode)) {
1411 out_sub:
1412 inode_sub_bytes(inode, number);
1413 return QUOTA_OK;
1416 down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1417 /* Now recheck reliably when holding dqptr_sem */
1418 if (IS_NOQUOTA(inode)) {
1419 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1420 goto out_sub;
1422 spin_lock(&dq_data_lock);
1423 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1424 if (inode->i_dquot[cnt] == NODQUOT)
1425 continue;
1426 warntype[cnt] = info_bdq_free(inode->i_dquot[cnt], number);
1427 dquot_decr_space(inode->i_dquot[cnt], number);
1429 inode_sub_bytes(inode, number);
1430 spin_unlock(&dq_data_lock);
1431 /* Dirtify all the dquots - this can block when journalling */
1432 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1433 if (inode->i_dquot[cnt])
1434 mark_dquot_dirty(inode->i_dquot[cnt]);
1435 flush_warnings(inode->i_dquot, warntype);
1436 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1437 return QUOTA_OK;
1441 * This operation can block, but only after everything is updated
1443 int dquot_free_inode(const struct inode *inode, qsize_t number)
1445 unsigned int cnt;
1446 char warntype[MAXQUOTAS];
1448 /* First test before acquiring mutex - solves deadlocks when we
1449 * re-enter the quota code and are already holding the mutex */
1450 if (IS_NOQUOTA(inode))
1451 return QUOTA_OK;
1453 down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1454 /* Now recheck reliably when holding dqptr_sem */
1455 if (IS_NOQUOTA(inode)) {
1456 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1457 return QUOTA_OK;
1459 spin_lock(&dq_data_lock);
1460 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1461 if (inode->i_dquot[cnt] == NODQUOT)
1462 continue;
1463 warntype[cnt] = info_idq_free(inode->i_dquot[cnt], number);
1464 dquot_decr_inodes(inode->i_dquot[cnt], number);
1466 spin_unlock(&dq_data_lock);
1467 /* Dirtify all the dquots - this can block when journalling */
1468 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1469 if (inode->i_dquot[cnt])
1470 mark_dquot_dirty(inode->i_dquot[cnt]);
1471 flush_warnings(inode->i_dquot, warntype);
1472 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1473 return QUOTA_OK;
1477 * Transfer the number of inode and blocks from one diskquota to an other.
1479 * This operation can block, but only after everything is updated
1480 * A transaction must be started when entering this function.
1482 int dquot_transfer(struct inode *inode, struct iattr *iattr)
1484 qsize_t space;
1485 struct dquot *transfer_from[MAXQUOTAS];
1486 struct dquot *transfer_to[MAXQUOTAS];
1487 int cnt, ret = QUOTA_OK;
1488 int chuid = iattr->ia_valid & ATTR_UID && inode->i_uid != iattr->ia_uid,
1489 chgid = iattr->ia_valid & ATTR_GID && inode->i_gid != iattr->ia_gid;
1490 char warntype_to[MAXQUOTAS];
1491 char warntype_from_inodes[MAXQUOTAS], warntype_from_space[MAXQUOTAS];
1493 /* First test before acquiring mutex - solves deadlocks when we
1494 * re-enter the quota code and are already holding the mutex */
1495 if (IS_NOQUOTA(inode))
1496 return QUOTA_OK;
1497 /* Initialize the arrays */
1498 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1499 transfer_from[cnt] = NODQUOT;
1500 transfer_to[cnt] = NODQUOT;
1501 warntype_to[cnt] = QUOTA_NL_NOWARN;
1502 switch (cnt) {
1503 case USRQUOTA:
1504 if (!chuid)
1505 continue;
1506 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_uid, cnt);
1507 break;
1508 case GRPQUOTA:
1509 if (!chgid)
1510 continue;
1511 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_gid, cnt);
1512 break;
1516 down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1517 /* Now recheck reliably when holding dqptr_sem */
1518 if (IS_NOQUOTA(inode)) { /* File without quota accounting? */
1519 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1520 goto put_all;
1522 spin_lock(&dq_data_lock);
1523 space = inode_get_bytes(inode);
1524 /* Build the transfer_from list and check the limits */
1525 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1526 if (transfer_to[cnt] == NODQUOT)
1527 continue;
1528 transfer_from[cnt] = inode->i_dquot[cnt];
1529 if (check_idq(transfer_to[cnt], 1, warntype_to + cnt) ==
1530 NO_QUOTA || check_bdq(transfer_to[cnt], space, 0,
1531 warntype_to + cnt) == NO_QUOTA)
1532 goto over_quota;
1536 * Finally perform the needed transfer from transfer_from to transfer_to
1538 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1540 * Skip changes for same uid or gid or for turned off quota-type.
1542 if (transfer_to[cnt] == NODQUOT)
1543 continue;
1545 /* Due to IO error we might not have transfer_from[] structure */
1546 if (transfer_from[cnt]) {
1547 warntype_from_inodes[cnt] =
1548 info_idq_free(transfer_from[cnt], 1);
1549 warntype_from_space[cnt] =
1550 info_bdq_free(transfer_from[cnt], space);
1551 dquot_decr_inodes(transfer_from[cnt], 1);
1552 dquot_decr_space(transfer_from[cnt], space);
1555 dquot_incr_inodes(transfer_to[cnt], 1);
1556 dquot_incr_space(transfer_to[cnt], space);
1558 inode->i_dquot[cnt] = transfer_to[cnt];
1560 spin_unlock(&dq_data_lock);
1561 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1563 /* Dirtify all the dquots - this can block when journalling */
1564 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1565 if (transfer_from[cnt])
1566 mark_dquot_dirty(transfer_from[cnt]);
1567 if (transfer_to[cnt]) {
1568 mark_dquot_dirty(transfer_to[cnt]);
1569 /* The reference we got is transferred to the inode */
1570 transfer_to[cnt] = NODQUOT;
1573 warn_put_all:
1574 flush_warnings(transfer_to, warntype_to);
1575 flush_warnings(transfer_from, warntype_from_inodes);
1576 flush_warnings(transfer_from, warntype_from_space);
1577 put_all:
1578 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1579 dqput(transfer_from[cnt]);
1580 dqput(transfer_to[cnt]);
1582 return ret;
1583 over_quota:
1584 spin_unlock(&dq_data_lock);
1585 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1586 /* Clear dquot pointers we don't want to dqput() */
1587 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1588 transfer_from[cnt] = NODQUOT;
1589 ret = NO_QUOTA;
1590 goto warn_put_all;
1593 /* Wrapper for transferring ownership of an inode */
1594 int vfs_dq_transfer(struct inode *inode, struct iattr *iattr)
1596 if (sb_any_quota_active(inode->i_sb) && !IS_NOQUOTA(inode)) {
1597 vfs_dq_init(inode);
1598 if (inode->i_sb->dq_op->transfer(inode, iattr) == NO_QUOTA)
1599 return 1;
1601 return 0;
1606 * Write info of quota file to disk
1608 int dquot_commit_info(struct super_block *sb, int type)
1610 int ret;
1611 struct quota_info *dqopt = sb_dqopt(sb);
1613 mutex_lock(&dqopt->dqio_mutex);
1614 ret = dqopt->ops[type]->write_file_info(sb, type);
1615 mutex_unlock(&dqopt->dqio_mutex);
1616 return ret;
1620 * Definitions of diskquota operations.
1622 struct dquot_operations dquot_operations = {
1623 .initialize = dquot_initialize,
1624 .drop = dquot_drop,
1625 .alloc_space = dquot_alloc_space,
1626 .alloc_inode = dquot_alloc_inode,
1627 .free_space = dquot_free_space,
1628 .free_inode = dquot_free_inode,
1629 .transfer = dquot_transfer,
1630 .write_dquot = dquot_commit,
1631 .acquire_dquot = dquot_acquire,
1632 .release_dquot = dquot_release,
1633 .mark_dirty = dquot_mark_dquot_dirty,
1634 .write_info = dquot_commit_info,
1635 .alloc_dquot = dquot_alloc,
1636 .destroy_dquot = dquot_destroy,
1640 * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
1642 int vfs_quota_disable(struct super_block *sb, int type, unsigned int flags)
1644 int cnt, ret = 0;
1645 struct quota_info *dqopt = sb_dqopt(sb);
1646 struct inode *toputinode[MAXQUOTAS];
1648 /* Cannot turn off usage accounting without turning off limits, or
1649 * suspend quotas and simultaneously turn quotas off. */
1650 if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
1651 || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
1652 DQUOT_USAGE_ENABLED)))
1653 return -EINVAL;
1655 /* We need to serialize quota_off() for device */
1656 mutex_lock(&dqopt->dqonoff_mutex);
1659 * Skip everything if there's nothing to do. We have to do this because
1660 * sometimes we are called when fill_super() failed and calling
1661 * sync_fs() in such cases does no good.
1663 if (!sb_any_quota_loaded(sb)) {
1664 mutex_unlock(&dqopt->dqonoff_mutex);
1665 return 0;
1667 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1668 toputinode[cnt] = NULL;
1669 if (type != -1 && cnt != type)
1670 continue;
1671 if (!sb_has_quota_loaded(sb, cnt))
1672 continue;
1674 if (flags & DQUOT_SUSPENDED) {
1675 spin_lock(&dq_state_lock);
1676 dqopt->flags |=
1677 dquot_state_flag(DQUOT_SUSPENDED, cnt);
1678 spin_unlock(&dq_state_lock);
1679 } else {
1680 spin_lock(&dq_state_lock);
1681 dqopt->flags &= ~dquot_state_flag(flags, cnt);
1682 /* Turning off suspended quotas? */
1683 if (!sb_has_quota_loaded(sb, cnt) &&
1684 sb_has_quota_suspended(sb, cnt)) {
1685 dqopt->flags &= ~dquot_state_flag(
1686 DQUOT_SUSPENDED, cnt);
1687 spin_unlock(&dq_state_lock);
1688 iput(dqopt->files[cnt]);
1689 dqopt->files[cnt] = NULL;
1690 continue;
1692 spin_unlock(&dq_state_lock);
1695 /* We still have to keep quota loaded? */
1696 if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
1697 continue;
1699 /* Note: these are blocking operations */
1700 drop_dquot_ref(sb, cnt);
1701 invalidate_dquots(sb, cnt);
1703 * Now all dquots should be invalidated, all writes done so we should be only
1704 * users of the info. No locks needed.
1706 if (info_dirty(&dqopt->info[cnt]))
1707 sb->dq_op->write_info(sb, cnt);
1708 if (dqopt->ops[cnt]->free_file_info)
1709 dqopt->ops[cnt]->free_file_info(sb, cnt);
1710 put_quota_format(dqopt->info[cnt].dqi_format);
1712 toputinode[cnt] = dqopt->files[cnt];
1713 if (!sb_has_quota_loaded(sb, cnt))
1714 dqopt->files[cnt] = NULL;
1715 dqopt->info[cnt].dqi_flags = 0;
1716 dqopt->info[cnt].dqi_igrace = 0;
1717 dqopt->info[cnt].dqi_bgrace = 0;
1718 dqopt->ops[cnt] = NULL;
1720 mutex_unlock(&dqopt->dqonoff_mutex);
1722 /* Skip syncing and setting flags if quota files are hidden */
1723 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
1724 goto put_inodes;
1726 /* Sync the superblock so that buffers with quota data are written to
1727 * disk (and so userspace sees correct data afterwards). */
1728 if (sb->s_op->sync_fs)
1729 sb->s_op->sync_fs(sb, 1);
1730 sync_blockdev(sb->s_bdev);
1731 /* Now the quota files are just ordinary files and we can set the
1732 * inode flags back. Moreover we discard the pagecache so that
1733 * userspace sees the writes we did bypassing the pagecache. We
1734 * must also discard the blockdev buffers so that we see the
1735 * changes done by userspace on the next quotaon() */
1736 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1737 if (toputinode[cnt]) {
1738 mutex_lock(&dqopt->dqonoff_mutex);
1739 /* If quota was reenabled in the meantime, we have
1740 * nothing to do */
1741 if (!sb_has_quota_loaded(sb, cnt)) {
1742 mutex_lock_nested(&toputinode[cnt]->i_mutex, I_MUTEX_QUOTA);
1743 toputinode[cnt]->i_flags &= ~(S_IMMUTABLE |
1744 S_NOATIME | S_NOQUOTA);
1745 truncate_inode_pages(&toputinode[cnt]->i_data, 0);
1746 mutex_unlock(&toputinode[cnt]->i_mutex);
1747 mark_inode_dirty(toputinode[cnt]);
1749 mutex_unlock(&dqopt->dqonoff_mutex);
1751 if (sb->s_bdev)
1752 invalidate_bdev(sb->s_bdev);
1753 put_inodes:
1754 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1755 if (toputinode[cnt]) {
1756 /* On remount RO, we keep the inode pointer so that we
1757 * can reenable quota on the subsequent remount RW. We
1758 * have to check 'flags' variable and not use sb_has_
1759 * function because another quotaon / quotaoff could
1760 * change global state before we got here. We refuse
1761 * to suspend quotas when there is pending delete on
1762 * the quota file... */
1763 if (!(flags & DQUOT_SUSPENDED))
1764 iput(toputinode[cnt]);
1765 else if (!toputinode[cnt]->i_nlink)
1766 ret = -EBUSY;
1768 return ret;
1771 int vfs_quota_off(struct super_block *sb, int type, int remount)
1773 return vfs_quota_disable(sb, type, remount ? DQUOT_SUSPENDED :
1774 (DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED));
1778 * Turn quotas on on a device
1782 * Helper function to turn quotas on when we already have the inode of
1783 * quota file and no quota information is loaded.
1785 static int vfs_load_quota_inode(struct inode *inode, int type, int format_id,
1786 unsigned int flags)
1788 struct quota_format_type *fmt = find_quota_format(format_id);
1789 struct super_block *sb = inode->i_sb;
1790 struct quota_info *dqopt = sb_dqopt(sb);
1791 int error;
1792 int oldflags = -1;
1794 if (!fmt)
1795 return -ESRCH;
1796 if (!S_ISREG(inode->i_mode)) {
1797 error = -EACCES;
1798 goto out_fmt;
1800 if (IS_RDONLY(inode)) {
1801 error = -EROFS;
1802 goto out_fmt;
1804 if (!sb->s_op->quota_write || !sb->s_op->quota_read) {
1805 error = -EINVAL;
1806 goto out_fmt;
1808 /* Usage always has to be set... */
1809 if (!(flags & DQUOT_USAGE_ENABLED)) {
1810 error = -EINVAL;
1811 goto out_fmt;
1814 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
1815 /* As we bypass the pagecache we must now flush the inode so
1816 * that we see all the changes from userspace... */
1817 write_inode_now(inode, 1);
1818 /* And now flush the block cache so that kernel sees the
1819 * changes */
1820 invalidate_bdev(sb->s_bdev);
1822 mutex_lock(&inode->i_mutex);
1823 mutex_lock(&dqopt->dqonoff_mutex);
1824 if (sb_has_quota_loaded(sb, type)) {
1825 error = -EBUSY;
1826 goto out_lock;
1829 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
1830 /* We don't want quota and atime on quota files (deadlocks
1831 * possible) Also nobody should write to the file - we use
1832 * special IO operations which ignore the immutable bit. */
1833 down_write(&dqopt->dqptr_sem);
1834 oldflags = inode->i_flags & (S_NOATIME | S_IMMUTABLE | S_NOQUOTA);
1835 inode->i_flags |= S_NOQUOTA | S_NOATIME | S_IMMUTABLE;
1836 up_write(&dqopt->dqptr_sem);
1837 sb->dq_op->drop(inode);
1840 error = -EIO;
1841 dqopt->files[type] = igrab(inode);
1842 if (!dqopt->files[type])
1843 goto out_lock;
1844 error = -EINVAL;
1845 if (!fmt->qf_ops->check_quota_file(sb, type))
1846 goto out_file_init;
1848 dqopt->ops[type] = fmt->qf_ops;
1849 dqopt->info[type].dqi_format = fmt;
1850 dqopt->info[type].dqi_fmt_id = format_id;
1851 INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
1852 mutex_lock(&dqopt->dqio_mutex);
1853 if ((error = dqopt->ops[type]->read_file_info(sb, type)) < 0) {
1854 mutex_unlock(&dqopt->dqio_mutex);
1855 goto out_file_init;
1857 mutex_unlock(&dqopt->dqio_mutex);
1858 mutex_unlock(&inode->i_mutex);
1859 spin_lock(&dq_state_lock);
1860 dqopt->flags |= dquot_state_flag(flags, type);
1861 spin_unlock(&dq_state_lock);
1863 add_dquot_ref(sb, type);
1864 mutex_unlock(&dqopt->dqonoff_mutex);
1866 return 0;
1868 out_file_init:
1869 dqopt->files[type] = NULL;
1870 iput(inode);
1871 out_lock:
1872 mutex_unlock(&dqopt->dqonoff_mutex);
1873 if (oldflags != -1) {
1874 down_write(&dqopt->dqptr_sem);
1875 /* Set the flags back (in the case of accidental quotaon()
1876 * on a wrong file we don't want to mess up the flags) */
1877 inode->i_flags &= ~(S_NOATIME | S_NOQUOTA | S_IMMUTABLE);
1878 inode->i_flags |= oldflags;
1879 up_write(&dqopt->dqptr_sem);
1881 mutex_unlock(&inode->i_mutex);
1882 out_fmt:
1883 put_quota_format(fmt);
1885 return error;
1888 /* Reenable quotas on remount RW */
1889 static int vfs_quota_on_remount(struct super_block *sb, int type)
1891 struct quota_info *dqopt = sb_dqopt(sb);
1892 struct inode *inode;
1893 int ret;
1894 unsigned int flags;
1896 mutex_lock(&dqopt->dqonoff_mutex);
1897 if (!sb_has_quota_suspended(sb, type)) {
1898 mutex_unlock(&dqopt->dqonoff_mutex);
1899 return 0;
1901 inode = dqopt->files[type];
1902 dqopt->files[type] = NULL;
1903 spin_lock(&dq_state_lock);
1904 flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
1905 DQUOT_LIMITS_ENABLED, type);
1906 dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, type);
1907 spin_unlock(&dq_state_lock);
1908 mutex_unlock(&dqopt->dqonoff_mutex);
1910 flags = dquot_generic_flag(flags, type);
1911 ret = vfs_load_quota_inode(inode, type, dqopt->info[type].dqi_fmt_id,
1912 flags);
1913 iput(inode);
1915 return ret;
1918 int vfs_quota_on_path(struct super_block *sb, int type, int format_id,
1919 struct path *path)
1921 int error = security_quota_on(path->dentry);
1922 if (error)
1923 return error;
1924 /* Quota file not on the same filesystem? */
1925 if (path->mnt->mnt_sb != sb)
1926 error = -EXDEV;
1927 else
1928 error = vfs_load_quota_inode(path->dentry->d_inode, type,
1929 format_id, DQUOT_USAGE_ENABLED |
1930 DQUOT_LIMITS_ENABLED);
1931 return error;
1934 int vfs_quota_on(struct super_block *sb, int type, int format_id, char *name,
1935 int remount)
1937 struct path path;
1938 int error;
1940 if (remount)
1941 return vfs_quota_on_remount(sb, type);
1943 error = kern_path(name, LOOKUP_FOLLOW, &path);
1944 if (!error) {
1945 error = vfs_quota_on_path(sb, type, format_id, &path);
1946 path_put(&path);
1948 return error;
1952 * More powerful function for turning on quotas allowing setting
1953 * of individual quota flags
1955 int vfs_quota_enable(struct inode *inode, int type, int format_id,
1956 unsigned int flags)
1958 int ret = 0;
1959 struct super_block *sb = inode->i_sb;
1960 struct quota_info *dqopt = sb_dqopt(sb);
1962 /* Just unsuspend quotas? */
1963 if (flags & DQUOT_SUSPENDED)
1964 return vfs_quota_on_remount(sb, type);
1965 if (!flags)
1966 return 0;
1967 /* Just updating flags needed? */
1968 if (sb_has_quota_loaded(sb, type)) {
1969 mutex_lock(&dqopt->dqonoff_mutex);
1970 /* Now do a reliable test... */
1971 if (!sb_has_quota_loaded(sb, type)) {
1972 mutex_unlock(&dqopt->dqonoff_mutex);
1973 goto load_quota;
1975 if (flags & DQUOT_USAGE_ENABLED &&
1976 sb_has_quota_usage_enabled(sb, type)) {
1977 ret = -EBUSY;
1978 goto out_lock;
1980 if (flags & DQUOT_LIMITS_ENABLED &&
1981 sb_has_quota_limits_enabled(sb, type)) {
1982 ret = -EBUSY;
1983 goto out_lock;
1985 spin_lock(&dq_state_lock);
1986 sb_dqopt(sb)->flags |= dquot_state_flag(flags, type);
1987 spin_unlock(&dq_state_lock);
1988 out_lock:
1989 mutex_unlock(&dqopt->dqonoff_mutex);
1990 return ret;
1993 load_quota:
1994 return vfs_load_quota_inode(inode, type, format_id, flags);
1998 * This function is used when filesystem needs to initialize quotas
1999 * during mount time.
2001 int vfs_quota_on_mount(struct super_block *sb, char *qf_name,
2002 int format_id, int type)
2004 struct dentry *dentry;
2005 int error;
2007 dentry = lookup_one_len(qf_name, sb->s_root, strlen(qf_name));
2008 if (IS_ERR(dentry))
2009 return PTR_ERR(dentry);
2011 if (!dentry->d_inode) {
2012 error = -ENOENT;
2013 goto out;
2016 error = security_quota_on(dentry);
2017 if (!error)
2018 error = vfs_load_quota_inode(dentry->d_inode, type, format_id,
2019 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2021 out:
2022 dput(dentry);
2023 return error;
2026 /* Wrapper to turn on quotas when remounting rw */
2027 int vfs_dq_quota_on_remount(struct super_block *sb)
2029 int cnt;
2030 int ret = 0, err;
2032 if (!sb->s_qcop || !sb->s_qcop->quota_on)
2033 return -ENOSYS;
2034 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2035 err = sb->s_qcop->quota_on(sb, cnt, 0, NULL, 1);
2036 if (err < 0 && !ret)
2037 ret = err;
2039 return ret;
2042 static inline qsize_t qbtos(qsize_t blocks)
2044 return blocks << QIF_DQBLKSIZE_BITS;
2047 static inline qsize_t stoqb(qsize_t space)
2049 return (space + QIF_DQBLKSIZE - 1) >> QIF_DQBLKSIZE_BITS;
2052 /* Generic routine for getting common part of quota structure */
2053 static void do_get_dqblk(struct dquot *dquot, struct if_dqblk *di)
2055 struct mem_dqblk *dm = &dquot->dq_dqb;
2057 spin_lock(&dq_data_lock);
2058 di->dqb_bhardlimit = stoqb(dm->dqb_bhardlimit);
2059 di->dqb_bsoftlimit = stoqb(dm->dqb_bsoftlimit);
2060 di->dqb_curspace = dm->dqb_curspace;
2061 di->dqb_ihardlimit = dm->dqb_ihardlimit;
2062 di->dqb_isoftlimit = dm->dqb_isoftlimit;
2063 di->dqb_curinodes = dm->dqb_curinodes;
2064 di->dqb_btime = dm->dqb_btime;
2065 di->dqb_itime = dm->dqb_itime;
2066 di->dqb_valid = QIF_ALL;
2067 spin_unlock(&dq_data_lock);
2070 int vfs_get_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
2072 struct dquot *dquot;
2074 dquot = dqget(sb, id, type);
2075 if (dquot == NODQUOT)
2076 return -ESRCH;
2077 do_get_dqblk(dquot, di);
2078 dqput(dquot);
2080 return 0;
2083 /* Generic routine for setting common part of quota structure */
2084 static int do_set_dqblk(struct dquot *dquot, struct if_dqblk *di)
2086 struct mem_dqblk *dm = &dquot->dq_dqb;
2087 int check_blim = 0, check_ilim = 0;
2088 struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_type];
2090 if ((di->dqb_valid & QIF_BLIMITS &&
2091 (di->dqb_bhardlimit > dqi->dqi_maxblimit ||
2092 di->dqb_bsoftlimit > dqi->dqi_maxblimit)) ||
2093 (di->dqb_valid & QIF_ILIMITS &&
2094 (di->dqb_ihardlimit > dqi->dqi_maxilimit ||
2095 di->dqb_isoftlimit > dqi->dqi_maxilimit)))
2096 return -ERANGE;
2098 spin_lock(&dq_data_lock);
2099 if (di->dqb_valid & QIF_SPACE) {
2100 dm->dqb_curspace = di->dqb_curspace;
2101 check_blim = 1;
2102 __set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2104 if (di->dqb_valid & QIF_BLIMITS) {
2105 dm->dqb_bsoftlimit = qbtos(di->dqb_bsoftlimit);
2106 dm->dqb_bhardlimit = qbtos(di->dqb_bhardlimit);
2107 check_blim = 1;
2108 __set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2110 if (di->dqb_valid & QIF_INODES) {
2111 dm->dqb_curinodes = di->dqb_curinodes;
2112 check_ilim = 1;
2113 __set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2115 if (di->dqb_valid & QIF_ILIMITS) {
2116 dm->dqb_isoftlimit = di->dqb_isoftlimit;
2117 dm->dqb_ihardlimit = di->dqb_ihardlimit;
2118 check_ilim = 1;
2119 __set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2121 if (di->dqb_valid & QIF_BTIME) {
2122 dm->dqb_btime = di->dqb_btime;
2123 check_blim = 1;
2124 __set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2126 if (di->dqb_valid & QIF_ITIME) {
2127 dm->dqb_itime = di->dqb_itime;
2128 check_ilim = 1;
2129 __set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2132 if (check_blim) {
2133 if (!dm->dqb_bsoftlimit || dm->dqb_curspace < dm->dqb_bsoftlimit) {
2134 dm->dqb_btime = 0;
2135 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2137 else if (!(di->dqb_valid & QIF_BTIME)) /* Set grace only if user hasn't provided his own... */
2138 dm->dqb_btime = get_seconds() + dqi->dqi_bgrace;
2140 if (check_ilim) {
2141 if (!dm->dqb_isoftlimit || dm->dqb_curinodes < dm->dqb_isoftlimit) {
2142 dm->dqb_itime = 0;
2143 clear_bit(DQ_INODES_B, &dquot->dq_flags);
2145 else if (!(di->dqb_valid & QIF_ITIME)) /* Set grace only if user hasn't provided his own... */
2146 dm->dqb_itime = get_seconds() + dqi->dqi_igrace;
2148 if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit || dm->dqb_isoftlimit)
2149 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2150 else
2151 set_bit(DQ_FAKE_B, &dquot->dq_flags);
2152 spin_unlock(&dq_data_lock);
2153 mark_dquot_dirty(dquot);
2155 return 0;
2158 int vfs_set_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
2160 struct dquot *dquot;
2161 int rc;
2163 dquot = dqget(sb, id, type);
2164 if (!dquot) {
2165 rc = -ESRCH;
2166 goto out;
2168 rc = do_set_dqblk(dquot, di);
2169 dqput(dquot);
2170 out:
2171 return rc;
2174 /* Generic routine for getting common part of quota file information */
2175 int vfs_get_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
2177 struct mem_dqinfo *mi;
2179 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
2180 if (!sb_has_quota_active(sb, type)) {
2181 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2182 return -ESRCH;
2184 mi = sb_dqopt(sb)->info + type;
2185 spin_lock(&dq_data_lock);
2186 ii->dqi_bgrace = mi->dqi_bgrace;
2187 ii->dqi_igrace = mi->dqi_igrace;
2188 ii->dqi_flags = mi->dqi_flags & DQF_MASK;
2189 ii->dqi_valid = IIF_ALL;
2190 spin_unlock(&dq_data_lock);
2191 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2192 return 0;
2195 /* Generic routine for setting common part of quota file information */
2196 int vfs_set_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
2198 struct mem_dqinfo *mi;
2199 int err = 0;
2201 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
2202 if (!sb_has_quota_active(sb, type)) {
2203 err = -ESRCH;
2204 goto out;
2206 mi = sb_dqopt(sb)->info + type;
2207 spin_lock(&dq_data_lock);
2208 if (ii->dqi_valid & IIF_BGRACE)
2209 mi->dqi_bgrace = ii->dqi_bgrace;
2210 if (ii->dqi_valid & IIF_IGRACE)
2211 mi->dqi_igrace = ii->dqi_igrace;
2212 if (ii->dqi_valid & IIF_FLAGS)
2213 mi->dqi_flags = (mi->dqi_flags & ~DQF_MASK) | (ii->dqi_flags & DQF_MASK);
2214 spin_unlock(&dq_data_lock);
2215 mark_info_dirty(sb, type);
2216 /* Force write to disk */
2217 sb->dq_op->write_info(sb, type);
2218 out:
2219 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2220 return err;
2223 struct quotactl_ops vfs_quotactl_ops = {
2224 .quota_on = vfs_quota_on,
2225 .quota_off = vfs_quota_off,
2226 .quota_sync = vfs_quota_sync,
2227 .get_info = vfs_get_dqinfo,
2228 .set_info = vfs_set_dqinfo,
2229 .get_dqblk = vfs_get_dqblk,
2230 .set_dqblk = vfs_set_dqblk
2233 static ctl_table fs_dqstats_table[] = {
2235 .ctl_name = FS_DQ_LOOKUPS,
2236 .procname = "lookups",
2237 .data = &dqstats.lookups,
2238 .maxlen = sizeof(int),
2239 .mode = 0444,
2240 .proc_handler = &proc_dointvec,
2243 .ctl_name = FS_DQ_DROPS,
2244 .procname = "drops",
2245 .data = &dqstats.drops,
2246 .maxlen = sizeof(int),
2247 .mode = 0444,
2248 .proc_handler = &proc_dointvec,
2251 .ctl_name = FS_DQ_READS,
2252 .procname = "reads",
2253 .data = &dqstats.reads,
2254 .maxlen = sizeof(int),
2255 .mode = 0444,
2256 .proc_handler = &proc_dointvec,
2259 .ctl_name = FS_DQ_WRITES,
2260 .procname = "writes",
2261 .data = &dqstats.writes,
2262 .maxlen = sizeof(int),
2263 .mode = 0444,
2264 .proc_handler = &proc_dointvec,
2267 .ctl_name = FS_DQ_CACHE_HITS,
2268 .procname = "cache_hits",
2269 .data = &dqstats.cache_hits,
2270 .maxlen = sizeof(int),
2271 .mode = 0444,
2272 .proc_handler = &proc_dointvec,
2275 .ctl_name = FS_DQ_ALLOCATED,
2276 .procname = "allocated_dquots",
2277 .data = &dqstats.allocated_dquots,
2278 .maxlen = sizeof(int),
2279 .mode = 0444,
2280 .proc_handler = &proc_dointvec,
2283 .ctl_name = FS_DQ_FREE,
2284 .procname = "free_dquots",
2285 .data = &dqstats.free_dquots,
2286 .maxlen = sizeof(int),
2287 .mode = 0444,
2288 .proc_handler = &proc_dointvec,
2291 .ctl_name = FS_DQ_SYNCS,
2292 .procname = "syncs",
2293 .data = &dqstats.syncs,
2294 .maxlen = sizeof(int),
2295 .mode = 0444,
2296 .proc_handler = &proc_dointvec,
2298 #ifdef CONFIG_PRINT_QUOTA_WARNING
2300 .ctl_name = FS_DQ_WARNINGS,
2301 .procname = "warnings",
2302 .data = &flag_print_warnings,
2303 .maxlen = sizeof(int),
2304 .mode = 0644,
2305 .proc_handler = &proc_dointvec,
2307 #endif
2308 { .ctl_name = 0 },
2311 static ctl_table fs_table[] = {
2313 .ctl_name = FS_DQSTATS,
2314 .procname = "quota",
2315 .mode = 0555,
2316 .child = fs_dqstats_table,
2318 { .ctl_name = 0 },
2321 static ctl_table sys_table[] = {
2323 .ctl_name = CTL_FS,
2324 .procname = "fs",
2325 .mode = 0555,
2326 .child = fs_table,
2328 { .ctl_name = 0 },
2331 static int __init dquot_init(void)
2333 int i;
2334 unsigned long nr_hash, order;
2336 printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2338 register_sysctl_table(sys_table);
2340 dquot_cachep = kmem_cache_create("dquot",
2341 sizeof(struct dquot), sizeof(unsigned long) * 4,
2342 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2343 SLAB_MEM_SPREAD|SLAB_PANIC),
2344 NULL);
2346 order = 0;
2347 dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
2348 if (!dquot_hash)
2349 panic("Cannot create dquot hash table");
2351 /* Find power-of-two hlist_heads which can fit into allocation */
2352 nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2353 dq_hash_bits = 0;
2354 do {
2355 dq_hash_bits++;
2356 } while (nr_hash >> dq_hash_bits);
2357 dq_hash_bits--;
2359 nr_hash = 1UL << dq_hash_bits;
2360 dq_hash_mask = nr_hash - 1;
2361 for (i = 0; i < nr_hash; i++)
2362 INIT_HLIST_HEAD(dquot_hash + i);
2364 printk("Dquot-cache hash table entries: %ld (order %ld, %ld bytes)\n",
2365 nr_hash, order, (PAGE_SIZE << order));
2367 register_shrinker(&dqcache_shrinker);
2369 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
2370 if (genl_register_family(&quota_genl_family) != 0)
2371 printk(KERN_ERR "VFS: Failed to create quota netlink interface.\n");
2372 #endif
2374 return 0;
2376 module_init(dquot_init);
2378 EXPORT_SYMBOL(register_quota_format);
2379 EXPORT_SYMBOL(unregister_quota_format);
2380 EXPORT_SYMBOL(dqstats);
2381 EXPORT_SYMBOL(dq_data_lock);
2382 EXPORT_SYMBOL(vfs_quota_enable);
2383 EXPORT_SYMBOL(vfs_quota_on);
2384 EXPORT_SYMBOL(vfs_quota_on_path);
2385 EXPORT_SYMBOL(vfs_quota_on_mount);
2386 EXPORT_SYMBOL(vfs_quota_disable);
2387 EXPORT_SYMBOL(vfs_quota_off);
2388 EXPORT_SYMBOL(dquot_scan_active);
2389 EXPORT_SYMBOL(vfs_quota_sync);
2390 EXPORT_SYMBOL(vfs_get_dqinfo);
2391 EXPORT_SYMBOL(vfs_set_dqinfo);
2392 EXPORT_SYMBOL(vfs_get_dqblk);
2393 EXPORT_SYMBOL(vfs_set_dqblk);
2394 EXPORT_SYMBOL(dquot_commit);
2395 EXPORT_SYMBOL(dquot_commit_info);
2396 EXPORT_SYMBOL(dquot_acquire);
2397 EXPORT_SYMBOL(dquot_release);
2398 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
2399 EXPORT_SYMBOL(dquot_initialize);
2400 EXPORT_SYMBOL(dquot_drop);
2401 EXPORT_SYMBOL(vfs_dq_drop);
2402 EXPORT_SYMBOL(dqget);
2403 EXPORT_SYMBOL(dqput);
2404 EXPORT_SYMBOL(dquot_alloc_space);
2405 EXPORT_SYMBOL(dquot_alloc_inode);
2406 EXPORT_SYMBOL(dquot_free_space);
2407 EXPORT_SYMBOL(dquot_free_inode);
2408 EXPORT_SYMBOL(dquot_transfer);
2409 EXPORT_SYMBOL(vfs_dq_transfer);
2410 EXPORT_SYMBOL(vfs_dq_quota_on_remount);