quota: Implement function for scanning active dquots
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / dquot.c
blob075dc76904e71931ecffa8d40473854bcb68d808
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 two quota SMP locks. dq_list_lock protects all lists with quotas
91 * and quota formats and also dqstats structure containing statistics about the
92 * lists. dq_data_lock protects data from dq_dqb and also mem_dqinfo structures
93 * and 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().
97 * The spinlock ordering is hence: dq_data_lock > dq_list_lock > i_lock
99 * Note that some things (eg. sb pointer, type, id) doesn't change during
100 * the life of the dquot structure and so needn't to be protected by a lock
102 * Any operation working on dquots via inode pointers must hold dqptr_sem. If
103 * operation is just reading pointers from inode (or not using them at all) the
104 * read lock is enough. If pointers are altered function must hold write lock
105 * (these locking rules also apply for S_NOQUOTA flag in the inode - note that
106 * for altering the flag i_mutex is also needed). If operation is holding
107 * reference to dquot in other way (e.g. quotactl ops) it must be guarded by
108 * dqonoff_mutex.
109 * This locking assures that:
110 * a) update/access to dquot pointers in inode is serialized
111 * b) everyone is guarded against invalidate_dquots()
113 * Each dquot has its dq_lock mutex. Locked dquots might not be referenced
114 * from inodes (dquot_alloc_space() and such don't check the dq_lock).
115 * Currently dquot is locked only when it is being read to memory (or space for
116 * it is being allocated) on the first dqget() and when it is being released on
117 * the last dqput(). The allocation and release oparations are serialized by
118 * the dq_lock and by checking the use count in dquot_release(). Write
119 * operations on dquots don't hold dq_lock as they copy data under dq_data_lock
120 * spinlock to internal buffers before writing.
122 * Lock ordering (including related VFS locks) is the following:
123 * i_mutex > dqonoff_sem > journal_lock > dqptr_sem > dquot->dq_lock >
124 * dqio_mutex
125 * i_mutex on quota files is special (it's below dqio_mutex)
128 static DEFINE_SPINLOCK(dq_list_lock);
129 DEFINE_SPINLOCK(dq_data_lock);
131 static char *quotatypes[] = INITQFNAMES;
132 static struct quota_format_type *quota_formats; /* List of registered formats */
133 static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
135 /* SLAB cache for dquot structures */
136 static struct kmem_cache *dquot_cachep;
138 int register_quota_format(struct quota_format_type *fmt)
140 spin_lock(&dq_list_lock);
141 fmt->qf_next = quota_formats;
142 quota_formats = fmt;
143 spin_unlock(&dq_list_lock);
144 return 0;
147 void unregister_quota_format(struct quota_format_type *fmt)
149 struct quota_format_type **actqf;
151 spin_lock(&dq_list_lock);
152 for (actqf = &quota_formats; *actqf && *actqf != fmt; actqf = &(*actqf)->qf_next);
153 if (*actqf)
154 *actqf = (*actqf)->qf_next;
155 spin_unlock(&dq_list_lock);
158 static struct quota_format_type *find_quota_format(int id)
160 struct quota_format_type *actqf;
162 spin_lock(&dq_list_lock);
163 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id; actqf = actqf->qf_next);
164 if (!actqf || !try_module_get(actqf->qf_owner)) {
165 int qm;
167 spin_unlock(&dq_list_lock);
169 for (qm = 0; module_names[qm].qm_fmt_id && module_names[qm].qm_fmt_id != id; qm++);
170 if (!module_names[qm].qm_fmt_id || request_module(module_names[qm].qm_mod_name))
171 return NULL;
173 spin_lock(&dq_list_lock);
174 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id; actqf = actqf->qf_next);
175 if (actqf && !try_module_get(actqf->qf_owner))
176 actqf = NULL;
178 spin_unlock(&dq_list_lock);
179 return actqf;
182 static void put_quota_format(struct quota_format_type *fmt)
184 module_put(fmt->qf_owner);
188 * Dquot List Management:
189 * The quota code uses three lists for dquot management: the inuse_list,
190 * free_dquots, and dquot_hash[] array. A single dquot structure may be
191 * on all three lists, depending on its current state.
193 * All dquots are placed to the end of inuse_list when first created, and this
194 * list is used for invalidate operation, which must look at every dquot.
196 * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
197 * and this list is searched whenever we need an available dquot. Dquots are
198 * removed from the list as soon as they are used again, and
199 * dqstats.free_dquots gives the number of dquots on the list. When
200 * dquot is invalidated it's completely released from memory.
202 * Dquots with a specific identity (device, type and id) are placed on
203 * one of the dquot_hash[] hash chains. The provides an efficient search
204 * mechanism to locate a specific dquot.
207 static LIST_HEAD(inuse_list);
208 static LIST_HEAD(free_dquots);
209 static unsigned int dq_hash_bits, dq_hash_mask;
210 static struct hlist_head *dquot_hash;
212 struct dqstats dqstats;
214 static inline unsigned int
215 hashfn(const struct super_block *sb, unsigned int id, int type)
217 unsigned long tmp;
219 tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
220 return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
224 * Following list functions expect dq_list_lock to be held
226 static inline void insert_dquot_hash(struct dquot *dquot)
228 struct hlist_head *head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id, dquot->dq_type);
229 hlist_add_head(&dquot->dq_hash, head);
232 static inline void remove_dquot_hash(struct dquot *dquot)
234 hlist_del_init(&dquot->dq_hash);
237 static inline struct dquot *find_dquot(unsigned int hashent, struct super_block *sb, unsigned int id, int type)
239 struct hlist_node *node;
240 struct dquot *dquot;
242 hlist_for_each (node, dquot_hash+hashent) {
243 dquot = hlist_entry(node, struct dquot, dq_hash);
244 if (dquot->dq_sb == sb && dquot->dq_id == id && dquot->dq_type == type)
245 return dquot;
247 return NODQUOT;
250 /* Add a dquot to the tail of the free list */
251 static inline void put_dquot_last(struct dquot *dquot)
253 list_add_tail(&dquot->dq_free, &free_dquots);
254 dqstats.free_dquots++;
257 static inline void remove_free_dquot(struct dquot *dquot)
259 if (list_empty(&dquot->dq_free))
260 return;
261 list_del_init(&dquot->dq_free);
262 dqstats.free_dquots--;
265 static inline void put_inuse(struct dquot *dquot)
267 /* We add to the back of inuse list so we don't have to restart
268 * when traversing this list and we block */
269 list_add_tail(&dquot->dq_inuse, &inuse_list);
270 dqstats.allocated_dquots++;
273 static inline void remove_inuse(struct dquot *dquot)
275 dqstats.allocated_dquots--;
276 list_del(&dquot->dq_inuse);
279 * End of list functions needing dq_list_lock
282 static void wait_on_dquot(struct dquot *dquot)
284 mutex_lock(&dquot->dq_lock);
285 mutex_unlock(&dquot->dq_lock);
288 static inline int dquot_dirty(struct dquot *dquot)
290 return test_bit(DQ_MOD_B, &dquot->dq_flags);
293 static inline int mark_dquot_dirty(struct dquot *dquot)
295 return dquot->dq_sb->dq_op->mark_dirty(dquot);
298 int dquot_mark_dquot_dirty(struct dquot *dquot)
300 spin_lock(&dq_list_lock);
301 if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags))
302 list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
303 info[dquot->dq_type].dqi_dirty_list);
304 spin_unlock(&dq_list_lock);
305 return 0;
308 /* This function needs dq_list_lock */
309 static inline int clear_dquot_dirty(struct dquot *dquot)
311 if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags))
312 return 0;
313 list_del_init(&dquot->dq_dirty);
314 return 1;
317 void mark_info_dirty(struct super_block *sb, int type)
319 set_bit(DQF_INFO_DIRTY_B, &sb_dqopt(sb)->info[type].dqi_flags);
321 EXPORT_SYMBOL(mark_info_dirty);
324 * Read dquot from disk and alloc space for it
327 int dquot_acquire(struct dquot *dquot)
329 int ret = 0, ret2 = 0;
330 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
332 mutex_lock(&dquot->dq_lock);
333 mutex_lock(&dqopt->dqio_mutex);
334 if (!test_bit(DQ_READ_B, &dquot->dq_flags))
335 ret = dqopt->ops[dquot->dq_type]->read_dqblk(dquot);
336 if (ret < 0)
337 goto out_iolock;
338 set_bit(DQ_READ_B, &dquot->dq_flags);
339 /* Instantiate dquot if needed */
340 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
341 ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
342 /* Write the info if needed */
343 if (info_dirty(&dqopt->info[dquot->dq_type]))
344 ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
345 if (ret < 0)
346 goto out_iolock;
347 if (ret2 < 0) {
348 ret = ret2;
349 goto out_iolock;
352 set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
353 out_iolock:
354 mutex_unlock(&dqopt->dqio_mutex);
355 mutex_unlock(&dquot->dq_lock);
356 return ret;
360 * Write dquot to disk
362 int dquot_commit(struct dquot *dquot)
364 int ret = 0, ret2 = 0;
365 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
367 mutex_lock(&dqopt->dqio_mutex);
368 spin_lock(&dq_list_lock);
369 if (!clear_dquot_dirty(dquot)) {
370 spin_unlock(&dq_list_lock);
371 goto out_sem;
373 spin_unlock(&dq_list_lock);
374 /* Inactive dquot can be only if there was error during read/init
375 * => we have better not writing it */
376 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
377 ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
378 if (info_dirty(&dqopt->info[dquot->dq_type]))
379 ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
380 if (ret >= 0)
381 ret = ret2;
383 out_sem:
384 mutex_unlock(&dqopt->dqio_mutex);
385 return ret;
389 * Release dquot
391 int dquot_release(struct dquot *dquot)
393 int ret = 0, ret2 = 0;
394 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
396 mutex_lock(&dquot->dq_lock);
397 /* Check whether we are not racing with some other dqget() */
398 if (atomic_read(&dquot->dq_count) > 1)
399 goto out_dqlock;
400 mutex_lock(&dqopt->dqio_mutex);
401 if (dqopt->ops[dquot->dq_type]->release_dqblk) {
402 ret = dqopt->ops[dquot->dq_type]->release_dqblk(dquot);
403 /* Write the info */
404 if (info_dirty(&dqopt->info[dquot->dq_type]))
405 ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
406 if (ret >= 0)
407 ret = ret2;
409 clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
410 mutex_unlock(&dqopt->dqio_mutex);
411 out_dqlock:
412 mutex_unlock(&dquot->dq_lock);
413 return ret;
416 static void dquot_destroy(struct dquot *dquot)
418 kmem_cache_free(dquot_cachep, dquot);
421 static inline void do_destroy_dquot(struct dquot *dquot)
423 dquot->dq_sb->dq_op->destroy_dquot(dquot);
426 /* Invalidate all dquots on the list. Note that this function is called after
427 * quota is disabled and pointers from inodes removed so there cannot be new
428 * quota users. There can still be some users of quotas due to inodes being
429 * just deleted or pruned by prune_icache() (those are not attached to any
430 * list). We have to wait for such users.
432 static void invalidate_dquots(struct super_block *sb, int type)
434 struct dquot *dquot, *tmp;
436 restart:
437 spin_lock(&dq_list_lock);
438 list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
439 if (dquot->dq_sb != sb)
440 continue;
441 if (dquot->dq_type != type)
442 continue;
443 /* Wait for dquot users */
444 if (atomic_read(&dquot->dq_count)) {
445 DEFINE_WAIT(wait);
447 atomic_inc(&dquot->dq_count);
448 prepare_to_wait(&dquot->dq_wait_unused, &wait,
449 TASK_UNINTERRUPTIBLE);
450 spin_unlock(&dq_list_lock);
451 /* Once dqput() wakes us up, we know it's time to free
452 * the dquot.
453 * IMPORTANT: we rely on the fact that there is always
454 * at most one process waiting for dquot to free.
455 * Otherwise dq_count would be > 1 and we would never
456 * wake up.
458 if (atomic_read(&dquot->dq_count) > 1)
459 schedule();
460 finish_wait(&dquot->dq_wait_unused, &wait);
461 dqput(dquot);
462 /* At this moment dquot() need not exist (it could be
463 * reclaimed by prune_dqcache(). Hence we must
464 * restart. */
465 goto restart;
468 * Quota now has no users and it has been written on last
469 * dqput()
471 remove_dquot_hash(dquot);
472 remove_free_dquot(dquot);
473 remove_inuse(dquot);
474 do_destroy_dquot(dquot);
476 spin_unlock(&dq_list_lock);
479 /* Call callback for every active dquot on given filesystem */
480 int dquot_scan_active(struct super_block *sb,
481 int (*fn)(struct dquot *dquot, unsigned long priv),
482 unsigned long priv)
484 struct dquot *dquot, *old_dquot = NULL;
485 int ret = 0;
487 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
488 spin_lock(&dq_list_lock);
489 list_for_each_entry(dquot, &inuse_list, dq_inuse) {
490 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
491 continue;
492 if (dquot->dq_sb != sb)
493 continue;
494 /* Now we have active dquot so we can just increase use count */
495 atomic_inc(&dquot->dq_count);
496 dqstats.lookups++;
497 spin_unlock(&dq_list_lock);
498 dqput(old_dquot);
499 old_dquot = dquot;
500 ret = fn(dquot, priv);
501 if (ret < 0)
502 goto out;
503 spin_lock(&dq_list_lock);
504 /* We are safe to continue now because our dquot could not
505 * be moved out of the inuse list while we hold the reference */
507 spin_unlock(&dq_list_lock);
508 out:
509 dqput(old_dquot);
510 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
511 return ret;
514 int vfs_quota_sync(struct super_block *sb, int type)
516 struct list_head *dirty;
517 struct dquot *dquot;
518 struct quota_info *dqopt = sb_dqopt(sb);
519 int cnt;
521 mutex_lock(&dqopt->dqonoff_mutex);
522 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
523 if (type != -1 && cnt != type)
524 continue;
525 if (!sb_has_quota_active(sb, cnt))
526 continue;
527 spin_lock(&dq_list_lock);
528 dirty = &dqopt->info[cnt].dqi_dirty_list;
529 while (!list_empty(dirty)) {
530 dquot = list_first_entry(dirty, struct dquot, dq_dirty);
531 /* Dirty and inactive can be only bad dquot... */
532 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
533 clear_dquot_dirty(dquot);
534 continue;
536 /* Now we have active dquot from which someone is
537 * holding reference so we can safely just increase
538 * use count */
539 atomic_inc(&dquot->dq_count);
540 dqstats.lookups++;
541 spin_unlock(&dq_list_lock);
542 sb->dq_op->write_dquot(dquot);
543 dqput(dquot);
544 spin_lock(&dq_list_lock);
546 spin_unlock(&dq_list_lock);
549 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
550 if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
551 && info_dirty(&dqopt->info[cnt]))
552 sb->dq_op->write_info(sb, cnt);
553 spin_lock(&dq_list_lock);
554 dqstats.syncs++;
555 spin_unlock(&dq_list_lock);
556 mutex_unlock(&dqopt->dqonoff_mutex);
558 return 0;
561 /* Free unused dquots from cache */
562 static void prune_dqcache(int count)
564 struct list_head *head;
565 struct dquot *dquot;
567 head = free_dquots.prev;
568 while (head != &free_dquots && count) {
569 dquot = list_entry(head, struct dquot, dq_free);
570 remove_dquot_hash(dquot);
571 remove_free_dquot(dquot);
572 remove_inuse(dquot);
573 do_destroy_dquot(dquot);
574 count--;
575 head = free_dquots.prev;
580 * This is called from kswapd when we think we need some
581 * more memory
584 static int shrink_dqcache_memory(int nr, gfp_t gfp_mask)
586 if (nr) {
587 spin_lock(&dq_list_lock);
588 prune_dqcache(nr);
589 spin_unlock(&dq_list_lock);
591 return (dqstats.free_dquots / 100) * sysctl_vfs_cache_pressure;
594 static struct shrinker dqcache_shrinker = {
595 .shrink = shrink_dqcache_memory,
596 .seeks = DEFAULT_SEEKS,
600 * Put reference to dquot
601 * NOTE: If you change this function please check whether dqput_blocks() works right...
602 * MUST be called with either dqptr_sem or dqonoff_mutex held
604 void dqput(struct dquot *dquot)
606 int ret;
608 if (!dquot)
609 return;
610 #ifdef __DQUOT_PARANOIA
611 if (!atomic_read(&dquot->dq_count)) {
612 printk("VFS: dqput: trying to free free dquot\n");
613 printk("VFS: device %s, dquot of %s %d\n",
614 dquot->dq_sb->s_id,
615 quotatypes[dquot->dq_type],
616 dquot->dq_id);
617 BUG();
619 #endif
621 spin_lock(&dq_list_lock);
622 dqstats.drops++;
623 spin_unlock(&dq_list_lock);
624 we_slept:
625 spin_lock(&dq_list_lock);
626 if (atomic_read(&dquot->dq_count) > 1) {
627 /* We have more than one user... nothing to do */
628 atomic_dec(&dquot->dq_count);
629 /* Releasing dquot during quotaoff phase? */
630 if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_type) &&
631 atomic_read(&dquot->dq_count) == 1)
632 wake_up(&dquot->dq_wait_unused);
633 spin_unlock(&dq_list_lock);
634 return;
636 /* Need to release dquot? */
637 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && dquot_dirty(dquot)) {
638 spin_unlock(&dq_list_lock);
639 /* Commit dquot before releasing */
640 ret = dquot->dq_sb->dq_op->write_dquot(dquot);
641 if (ret < 0) {
642 printk(KERN_ERR "VFS: cannot write quota structure on "
643 "device %s (error %d). Quota may get out of "
644 "sync!\n", dquot->dq_sb->s_id, ret);
646 * We clear dirty bit anyway, so that we avoid
647 * infinite loop here
649 spin_lock(&dq_list_lock);
650 clear_dquot_dirty(dquot);
651 spin_unlock(&dq_list_lock);
653 goto we_slept;
655 /* Clear flag in case dquot was inactive (something bad happened) */
656 clear_dquot_dirty(dquot);
657 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
658 spin_unlock(&dq_list_lock);
659 dquot->dq_sb->dq_op->release_dquot(dquot);
660 goto we_slept;
662 atomic_dec(&dquot->dq_count);
663 #ifdef __DQUOT_PARANOIA
664 /* sanity check */
665 BUG_ON(!list_empty(&dquot->dq_free));
666 #endif
667 put_dquot_last(dquot);
668 spin_unlock(&dq_list_lock);
671 static struct dquot *dquot_alloc(struct super_block *sb, int type)
673 return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
676 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
678 struct dquot *dquot;
680 dquot = sb->dq_op->alloc_dquot(sb, type);
681 if(!dquot)
682 return NODQUOT;
684 mutex_init(&dquot->dq_lock);
685 INIT_LIST_HEAD(&dquot->dq_free);
686 INIT_LIST_HEAD(&dquot->dq_inuse);
687 INIT_HLIST_NODE(&dquot->dq_hash);
688 INIT_LIST_HEAD(&dquot->dq_dirty);
689 init_waitqueue_head(&dquot->dq_wait_unused);
690 dquot->dq_sb = sb;
691 dquot->dq_type = type;
692 atomic_set(&dquot->dq_count, 1);
694 return dquot;
698 * Check whether dquot is in memory.
699 * MUST be called with either dqptr_sem or dqonoff_mutex held
701 int dquot_is_cached(struct super_block *sb, unsigned int id, int type)
703 unsigned int hashent = hashfn(sb, id, type);
704 int ret = 0;
706 if (!sb_has_quota_active(sb, type))
707 return 0;
708 spin_lock(&dq_list_lock);
709 if (find_dquot(hashent, sb, id, type) != NODQUOT)
710 ret = 1;
711 spin_unlock(&dq_list_lock);
712 return ret;
716 * Get reference to dquot
717 * MUST be called with either dqptr_sem or dqonoff_mutex held
719 struct dquot *dqget(struct super_block *sb, unsigned int id, int type)
721 unsigned int hashent = hashfn(sb, id, type);
722 struct dquot *dquot, *empty = NODQUOT;
724 if (!sb_has_quota_active(sb, type))
725 return NODQUOT;
726 we_slept:
727 spin_lock(&dq_list_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 dquot->dq_id = id;
737 /* all dquots go on the inuse_list */
738 put_inuse(dquot);
739 /* hash it first so it can be found */
740 insert_dquot_hash(dquot);
741 dqstats.lookups++;
742 spin_unlock(&dq_list_lock);
743 } else {
744 if (!atomic_read(&dquot->dq_count))
745 remove_free_dquot(dquot);
746 atomic_inc(&dquot->dq_count);
747 dqstats.cache_hits++;
748 dqstats.lookups++;
749 spin_unlock(&dq_list_lock);
750 if (empty)
751 do_destroy_dquot(empty);
753 /* Wait for dq_lock - after this we know that either dquot_release() is already
754 * finished or it will be canceled due to dq_count > 1 test */
755 wait_on_dquot(dquot);
756 /* Read the dquot and instantiate it (everything done only if needed) */
757 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && sb->dq_op->acquire_dquot(dquot) < 0) {
758 dqput(dquot);
759 return NODQUOT;
761 #ifdef __DQUOT_PARANOIA
762 BUG_ON(!dquot->dq_sb); /* Has somebody invalidated entry under us? */
763 #endif
765 return dquot;
768 static int dqinit_needed(struct inode *inode, int type)
770 int cnt;
772 if (IS_NOQUOTA(inode))
773 return 0;
774 if (type != -1)
775 return inode->i_dquot[type] == NODQUOT;
776 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
777 if (inode->i_dquot[cnt] == NODQUOT)
778 return 1;
779 return 0;
782 /* This routine is guarded by dqonoff_mutex mutex */
783 static void add_dquot_ref(struct super_block *sb, int type)
785 struct inode *inode, *old_inode = NULL;
787 spin_lock(&inode_lock);
788 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
789 if (!atomic_read(&inode->i_writecount))
790 continue;
791 if (!dqinit_needed(inode, type))
792 continue;
793 if (inode->i_state & (I_FREEING|I_WILL_FREE))
794 continue;
796 __iget(inode);
797 spin_unlock(&inode_lock);
799 iput(old_inode);
800 sb->dq_op->initialize(inode, type);
801 /* We hold a reference to 'inode' so it couldn't have been
802 * removed from s_inodes list while we dropped the inode_lock.
803 * We cannot iput the inode now as we can be holding the last
804 * reference and we cannot iput it under inode_lock. So we
805 * keep the reference and iput it later. */
806 old_inode = inode;
807 spin_lock(&inode_lock);
809 spin_unlock(&inode_lock);
810 iput(old_inode);
813 /* Return 0 if dqput() won't block (note that 1 doesn't necessarily mean blocking) */
814 static inline int dqput_blocks(struct dquot *dquot)
816 if (atomic_read(&dquot->dq_count) <= 1)
817 return 1;
818 return 0;
821 /* Remove references to dquots from inode - add dquot to list for freeing if needed */
822 /* We can't race with anybody because we hold dqptr_sem for writing... */
823 static int remove_inode_dquot_ref(struct inode *inode, int type,
824 struct list_head *tofree_head)
826 struct dquot *dquot = inode->i_dquot[type];
828 inode->i_dquot[type] = NODQUOT;
829 if (dquot != NODQUOT) {
830 if (dqput_blocks(dquot)) {
831 #ifdef __DQUOT_PARANOIA
832 if (atomic_read(&dquot->dq_count) != 1)
833 printk(KERN_WARNING "VFS: Adding dquot with dq_count %d to dispose list.\n", atomic_read(&dquot->dq_count));
834 #endif
835 spin_lock(&dq_list_lock);
836 list_add(&dquot->dq_free, tofree_head); /* As dquot must have currently users it can't be on the free list... */
837 spin_unlock(&dq_list_lock);
838 return 1;
840 else
841 dqput(dquot); /* We have guaranteed we won't block */
843 return 0;
846 /* Free list of dquots - called from inode.c */
847 /* dquots are removed from inodes, no new references can be got so we are the only ones holding reference */
848 static void put_dquot_list(struct list_head *tofree_head)
850 struct list_head *act_head;
851 struct dquot *dquot;
853 act_head = tofree_head->next;
854 /* So now we have dquots on the list... Just free them */
855 while (act_head != tofree_head) {
856 dquot = list_entry(act_head, struct dquot, dq_free);
857 act_head = act_head->next;
858 list_del_init(&dquot->dq_free); /* Remove dquot from the list so we won't have problems... */
859 dqput(dquot);
863 static void remove_dquot_ref(struct super_block *sb, int type,
864 struct list_head *tofree_head)
866 struct inode *inode;
868 spin_lock(&inode_lock);
869 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
870 if (!IS_NOQUOTA(inode))
871 remove_inode_dquot_ref(inode, type, tofree_head);
873 spin_unlock(&inode_lock);
876 /* Gather all references from inodes and drop them */
877 static void drop_dquot_ref(struct super_block *sb, int type)
879 LIST_HEAD(tofree_head);
881 if (sb->dq_op) {
882 down_write(&sb_dqopt(sb)->dqptr_sem);
883 remove_dquot_ref(sb, type, &tofree_head);
884 up_write(&sb_dqopt(sb)->dqptr_sem);
885 put_dquot_list(&tofree_head);
889 static inline void dquot_incr_inodes(struct dquot *dquot, qsize_t number)
891 dquot->dq_dqb.dqb_curinodes += number;
894 static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
896 dquot->dq_dqb.dqb_curspace += number;
899 static inline void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
901 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
902 dquot->dq_dqb.dqb_curinodes >= number)
903 dquot->dq_dqb.dqb_curinodes -= number;
904 else
905 dquot->dq_dqb.dqb_curinodes = 0;
906 if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
907 dquot->dq_dqb.dqb_itime = (time_t) 0;
908 clear_bit(DQ_INODES_B, &dquot->dq_flags);
911 static inline void dquot_decr_space(struct dquot *dquot, qsize_t number)
913 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
914 dquot->dq_dqb.dqb_curspace >= number)
915 dquot->dq_dqb.dqb_curspace -= number;
916 else
917 dquot->dq_dqb.dqb_curspace = 0;
918 if (dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
919 dquot->dq_dqb.dqb_btime = (time_t) 0;
920 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
923 static int warning_issued(struct dquot *dquot, const int warntype)
925 int flag = (warntype == QUOTA_NL_BHARDWARN ||
926 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
927 ((warntype == QUOTA_NL_IHARDWARN ||
928 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
930 if (!flag)
931 return 0;
932 return test_and_set_bit(flag, &dquot->dq_flags);
935 #ifdef CONFIG_PRINT_QUOTA_WARNING
936 static int flag_print_warnings = 1;
938 static inline int need_print_warning(struct dquot *dquot)
940 if (!flag_print_warnings)
941 return 0;
943 switch (dquot->dq_type) {
944 case USRQUOTA:
945 return current_fsuid() == dquot->dq_id;
946 case GRPQUOTA:
947 return in_group_p(dquot->dq_id);
949 return 0;
952 /* Print warning to user which exceeded quota */
953 static void print_warning(struct dquot *dquot, const int warntype)
955 char *msg = NULL;
956 struct tty_struct *tty;
958 if (warntype == QUOTA_NL_IHARDBELOW ||
959 warntype == QUOTA_NL_ISOFTBELOW ||
960 warntype == QUOTA_NL_BHARDBELOW ||
961 warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(dquot))
962 return;
964 tty = get_current_tty();
965 if (!tty)
966 return;
967 tty_write_message(tty, dquot->dq_sb->s_id);
968 if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
969 tty_write_message(tty, ": warning, ");
970 else
971 tty_write_message(tty, ": write failed, ");
972 tty_write_message(tty, quotatypes[dquot->dq_type]);
973 switch (warntype) {
974 case QUOTA_NL_IHARDWARN:
975 msg = " file limit reached.\r\n";
976 break;
977 case QUOTA_NL_ISOFTLONGWARN:
978 msg = " file quota exceeded too long.\r\n";
979 break;
980 case QUOTA_NL_ISOFTWARN:
981 msg = " file quota exceeded.\r\n";
982 break;
983 case QUOTA_NL_BHARDWARN:
984 msg = " block limit reached.\r\n";
985 break;
986 case QUOTA_NL_BSOFTLONGWARN:
987 msg = " block quota exceeded too long.\r\n";
988 break;
989 case QUOTA_NL_BSOFTWARN:
990 msg = " block quota exceeded.\r\n";
991 break;
993 tty_write_message(tty, msg);
994 tty_kref_put(tty);
996 #endif
998 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
1000 /* Netlink family structure for quota */
1001 static struct genl_family quota_genl_family = {
1002 .id = GENL_ID_GENERATE,
1003 .hdrsize = 0,
1004 .name = "VFS_DQUOT",
1005 .version = 1,
1006 .maxattr = QUOTA_NL_A_MAX,
1009 /* Send warning to userspace about user which exceeded quota */
1010 static void send_warning(const struct dquot *dquot, const char warntype)
1012 static atomic_t seq;
1013 struct sk_buff *skb;
1014 void *msg_head;
1015 int ret;
1016 int msg_size = 4 * nla_total_size(sizeof(u32)) +
1017 2 * nla_total_size(sizeof(u64));
1019 /* We have to allocate using GFP_NOFS as we are called from a
1020 * filesystem performing write and thus further recursion into
1021 * the fs to free some data could cause deadlocks. */
1022 skb = genlmsg_new(msg_size, GFP_NOFS);
1023 if (!skb) {
1024 printk(KERN_ERR
1025 "VFS: Not enough memory to send quota warning.\n");
1026 return;
1028 msg_head = genlmsg_put(skb, 0, atomic_add_return(1, &seq),
1029 &quota_genl_family, 0, QUOTA_NL_C_WARNING);
1030 if (!msg_head) {
1031 printk(KERN_ERR
1032 "VFS: Cannot store netlink header in quota warning.\n");
1033 goto err_out;
1035 ret = nla_put_u32(skb, QUOTA_NL_A_QTYPE, dquot->dq_type);
1036 if (ret)
1037 goto attr_err_out;
1038 ret = nla_put_u64(skb, QUOTA_NL_A_EXCESS_ID, dquot->dq_id);
1039 if (ret)
1040 goto attr_err_out;
1041 ret = nla_put_u32(skb, QUOTA_NL_A_WARNING, warntype);
1042 if (ret)
1043 goto attr_err_out;
1044 ret = nla_put_u32(skb, QUOTA_NL_A_DEV_MAJOR,
1045 MAJOR(dquot->dq_sb->s_dev));
1046 if (ret)
1047 goto attr_err_out;
1048 ret = nla_put_u32(skb, QUOTA_NL_A_DEV_MINOR,
1049 MINOR(dquot->dq_sb->s_dev));
1050 if (ret)
1051 goto attr_err_out;
1052 ret = nla_put_u64(skb, QUOTA_NL_A_CAUSED_ID, current_uid());
1053 if (ret)
1054 goto attr_err_out;
1055 genlmsg_end(skb, msg_head);
1057 ret = genlmsg_multicast(skb, 0, quota_genl_family.id, GFP_NOFS);
1058 if (ret < 0 && ret != -ESRCH)
1059 printk(KERN_ERR
1060 "VFS: Failed to send notification message: %d\n", ret);
1061 return;
1062 attr_err_out:
1063 printk(KERN_ERR "VFS: Not enough space to compose quota message!\n");
1064 err_out:
1065 kfree_skb(skb);
1067 #endif
1069 static inline void flush_warnings(struct dquot * const *dquots, char *warntype)
1071 int i;
1073 for (i = 0; i < MAXQUOTAS; i++)
1074 if (dquots[i] != NODQUOT && warntype[i] != QUOTA_NL_NOWARN &&
1075 !warning_issued(dquots[i], warntype[i])) {
1076 #ifdef CONFIG_PRINT_QUOTA_WARNING
1077 print_warning(dquots[i], warntype[i]);
1078 #endif
1079 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
1080 send_warning(dquots[i], warntype[i]);
1081 #endif
1085 static inline char ignore_hardlimit(struct dquot *dquot)
1087 struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_type];
1089 return capable(CAP_SYS_RESOURCE) &&
1090 (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD || !(info->dqi_flags & V1_DQF_RSQUASH));
1093 /* needs dq_data_lock */
1094 static int check_idq(struct dquot *dquot, qsize_t inodes, char *warntype)
1096 *warntype = QUOTA_NL_NOWARN;
1097 if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_type) ||
1098 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1099 return QUOTA_OK;
1101 if (dquot->dq_dqb.dqb_ihardlimit &&
1102 (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_ihardlimit &&
1103 !ignore_hardlimit(dquot)) {
1104 *warntype = QUOTA_NL_IHARDWARN;
1105 return NO_QUOTA;
1108 if (dquot->dq_dqb.dqb_isoftlimit &&
1109 (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
1110 dquot->dq_dqb.dqb_itime && get_seconds() >= dquot->dq_dqb.dqb_itime &&
1111 !ignore_hardlimit(dquot)) {
1112 *warntype = QUOTA_NL_ISOFTLONGWARN;
1113 return NO_QUOTA;
1116 if (dquot->dq_dqb.dqb_isoftlimit &&
1117 (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
1118 dquot->dq_dqb.dqb_itime == 0) {
1119 *warntype = QUOTA_NL_ISOFTWARN;
1120 dquot->dq_dqb.dqb_itime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
1123 return QUOTA_OK;
1126 /* needs dq_data_lock */
1127 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc, char *warntype)
1129 *warntype = QUOTA_NL_NOWARN;
1130 if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_type) ||
1131 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1132 return QUOTA_OK;
1134 if (dquot->dq_dqb.dqb_bhardlimit &&
1135 dquot->dq_dqb.dqb_curspace + space > dquot->dq_dqb.dqb_bhardlimit &&
1136 !ignore_hardlimit(dquot)) {
1137 if (!prealloc)
1138 *warntype = QUOTA_NL_BHARDWARN;
1139 return NO_QUOTA;
1142 if (dquot->dq_dqb.dqb_bsoftlimit &&
1143 dquot->dq_dqb.dqb_curspace + space > dquot->dq_dqb.dqb_bsoftlimit &&
1144 dquot->dq_dqb.dqb_btime && get_seconds() >= dquot->dq_dqb.dqb_btime &&
1145 !ignore_hardlimit(dquot)) {
1146 if (!prealloc)
1147 *warntype = QUOTA_NL_BSOFTLONGWARN;
1148 return NO_QUOTA;
1151 if (dquot->dq_dqb.dqb_bsoftlimit &&
1152 dquot->dq_dqb.dqb_curspace + space > dquot->dq_dqb.dqb_bsoftlimit &&
1153 dquot->dq_dqb.dqb_btime == 0) {
1154 if (!prealloc) {
1155 *warntype = QUOTA_NL_BSOFTWARN;
1156 dquot->dq_dqb.dqb_btime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_bgrace;
1158 else
1160 * We don't allow preallocation to exceed softlimit so exceeding will
1161 * be always printed
1163 return NO_QUOTA;
1166 return QUOTA_OK;
1169 static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1171 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1172 dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1173 !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_type))
1174 return QUOTA_NL_NOWARN;
1176 if (dquot->dq_dqb.dqb_curinodes - inodes <= dquot->dq_dqb.dqb_isoftlimit)
1177 return QUOTA_NL_ISOFTBELOW;
1178 if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1179 dquot->dq_dqb.dqb_curinodes - inodes < dquot->dq_dqb.dqb_ihardlimit)
1180 return QUOTA_NL_IHARDBELOW;
1181 return QUOTA_NL_NOWARN;
1184 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1186 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1187 dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1188 return QUOTA_NL_NOWARN;
1190 if (dquot->dq_dqb.dqb_curspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1191 return QUOTA_NL_BSOFTBELOW;
1192 if (dquot->dq_dqb.dqb_curspace >= dquot->dq_dqb.dqb_bhardlimit &&
1193 dquot->dq_dqb.dqb_curspace - space < dquot->dq_dqb.dqb_bhardlimit)
1194 return QUOTA_NL_BHARDBELOW;
1195 return QUOTA_NL_NOWARN;
1198 * Initialize quota pointers in inode
1199 * Transaction must be started at entry
1201 int dquot_initialize(struct inode *inode, int type)
1203 unsigned int id = 0;
1204 int cnt, ret = 0;
1206 /* First test before acquiring mutex - solves deadlocks when we
1207 * re-enter the quota code and are already holding the mutex */
1208 if (IS_NOQUOTA(inode))
1209 return 0;
1210 down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1211 /* Having dqptr_sem we know NOQUOTA flags can't be altered... */
1212 if (IS_NOQUOTA(inode))
1213 goto out_err;
1214 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1215 if (type != -1 && cnt != type)
1216 continue;
1217 if (inode->i_dquot[cnt] == NODQUOT) {
1218 switch (cnt) {
1219 case USRQUOTA:
1220 id = inode->i_uid;
1221 break;
1222 case GRPQUOTA:
1223 id = inode->i_gid;
1224 break;
1226 inode->i_dquot[cnt] = dqget(inode->i_sb, id, cnt);
1229 out_err:
1230 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1231 return ret;
1235 * Release all quotas referenced by inode
1236 * Transaction must be started at an entry
1238 int dquot_drop_locked(struct inode *inode)
1240 int cnt;
1242 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1243 if (inode->i_dquot[cnt] != NODQUOT) {
1244 dqput(inode->i_dquot[cnt]);
1245 inode->i_dquot[cnt] = NODQUOT;
1248 return 0;
1251 int dquot_drop(struct inode *inode)
1253 down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1254 dquot_drop_locked(inode);
1255 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1256 return 0;
1259 /* Wrapper to remove references to quota structures from inode */
1260 void vfs_dq_drop(struct inode *inode)
1262 /* Here we can get arbitrary inode from clear_inode() so we have
1263 * to be careful. OTOH we don't need locking as quota operations
1264 * are allowed to change only at mount time */
1265 if (!IS_NOQUOTA(inode) && inode->i_sb && inode->i_sb->dq_op
1266 && inode->i_sb->dq_op->drop) {
1267 int cnt;
1268 /* Test before calling to rule out calls from proc and such
1269 * where we are not allowed to block. Note that this is
1270 * actually reliable test even without the lock - the caller
1271 * must assure that nobody can come after the DQUOT_DROP and
1272 * add quota pointers back anyway */
1273 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1274 if (inode->i_dquot[cnt] != NODQUOT)
1275 break;
1276 if (cnt < MAXQUOTAS)
1277 inode->i_sb->dq_op->drop(inode);
1282 * Following four functions update i_blocks+i_bytes fields and
1283 * quota information (together with appropriate checks)
1284 * NOTE: We absolutely rely on the fact that caller dirties
1285 * the inode (usually macros in quotaops.h care about this) and
1286 * holds a handle for the current transaction so that dquot write and
1287 * inode write go into the same transaction.
1291 * This operation can block, but only after everything is updated
1293 int dquot_alloc_space(struct inode *inode, qsize_t number, int warn)
1295 int cnt, ret = NO_QUOTA;
1296 char warntype[MAXQUOTAS];
1298 /* First test before acquiring mutex - solves deadlocks when we
1299 * re-enter the quota code and are already holding the mutex */
1300 if (IS_NOQUOTA(inode)) {
1301 out_add:
1302 inode_add_bytes(inode, number);
1303 return QUOTA_OK;
1305 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1306 warntype[cnt] = QUOTA_NL_NOWARN;
1308 down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1309 if (IS_NOQUOTA(inode)) { /* Now we can do reliable test... */
1310 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1311 goto out_add;
1313 spin_lock(&dq_data_lock);
1314 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1315 if (inode->i_dquot[cnt] == NODQUOT)
1316 continue;
1317 if (check_bdq(inode->i_dquot[cnt], number, warn, warntype+cnt) == NO_QUOTA)
1318 goto warn_put_all;
1320 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1321 if (inode->i_dquot[cnt] == NODQUOT)
1322 continue;
1323 dquot_incr_space(inode->i_dquot[cnt], number);
1325 inode_add_bytes(inode, number);
1326 ret = QUOTA_OK;
1327 warn_put_all:
1328 spin_unlock(&dq_data_lock);
1329 if (ret == QUOTA_OK)
1330 /* Dirtify all the dquots - this can block when journalling */
1331 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1332 if (inode->i_dquot[cnt])
1333 mark_dquot_dirty(inode->i_dquot[cnt]);
1334 flush_warnings(inode->i_dquot, warntype);
1335 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1336 return ret;
1340 * This operation can block, but only after everything is updated
1342 int dquot_alloc_inode(const struct inode *inode, qsize_t number)
1344 int cnt, ret = NO_QUOTA;
1345 char warntype[MAXQUOTAS];
1347 /* First test before acquiring mutex - solves deadlocks when we
1348 * re-enter the quota code and are already holding the mutex */
1349 if (IS_NOQUOTA(inode))
1350 return QUOTA_OK;
1351 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1352 warntype[cnt] = QUOTA_NL_NOWARN;
1353 down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1354 if (IS_NOQUOTA(inode)) {
1355 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1356 return QUOTA_OK;
1358 spin_lock(&dq_data_lock);
1359 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1360 if (inode->i_dquot[cnt] == NODQUOT)
1361 continue;
1362 if (check_idq(inode->i_dquot[cnt], number, warntype+cnt) == NO_QUOTA)
1363 goto warn_put_all;
1366 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1367 if (inode->i_dquot[cnt] == NODQUOT)
1368 continue;
1369 dquot_incr_inodes(inode->i_dquot[cnt], number);
1371 ret = QUOTA_OK;
1372 warn_put_all:
1373 spin_unlock(&dq_data_lock);
1374 if (ret == QUOTA_OK)
1375 /* Dirtify all the dquots - this can block when journalling */
1376 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1377 if (inode->i_dquot[cnt])
1378 mark_dquot_dirty(inode->i_dquot[cnt]);
1379 flush_warnings(inode->i_dquot, warntype);
1380 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1381 return ret;
1385 * This operation can block, but only after everything is updated
1387 int dquot_free_space(struct inode *inode, qsize_t number)
1389 unsigned int cnt;
1390 char warntype[MAXQUOTAS];
1392 /* First test before acquiring mutex - solves deadlocks when we
1393 * re-enter the quota code and are already holding the mutex */
1394 if (IS_NOQUOTA(inode)) {
1395 out_sub:
1396 inode_sub_bytes(inode, number);
1397 return QUOTA_OK;
1400 down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1401 /* Now recheck reliably when holding dqptr_sem */
1402 if (IS_NOQUOTA(inode)) {
1403 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1404 goto out_sub;
1406 spin_lock(&dq_data_lock);
1407 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1408 if (inode->i_dquot[cnt] == NODQUOT)
1409 continue;
1410 warntype[cnt] = info_bdq_free(inode->i_dquot[cnt], number);
1411 dquot_decr_space(inode->i_dquot[cnt], number);
1413 inode_sub_bytes(inode, number);
1414 spin_unlock(&dq_data_lock);
1415 /* Dirtify all the dquots - this can block when journalling */
1416 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1417 if (inode->i_dquot[cnt])
1418 mark_dquot_dirty(inode->i_dquot[cnt]);
1419 flush_warnings(inode->i_dquot, warntype);
1420 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1421 return QUOTA_OK;
1425 * This operation can block, but only after everything is updated
1427 int dquot_free_inode(const struct inode *inode, qsize_t number)
1429 unsigned int cnt;
1430 char warntype[MAXQUOTAS];
1432 /* First test before acquiring mutex - solves deadlocks when we
1433 * re-enter the quota code and are already holding the mutex */
1434 if (IS_NOQUOTA(inode))
1435 return QUOTA_OK;
1437 down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1438 /* Now recheck reliably when holding dqptr_sem */
1439 if (IS_NOQUOTA(inode)) {
1440 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1441 return QUOTA_OK;
1443 spin_lock(&dq_data_lock);
1444 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1445 if (inode->i_dquot[cnt] == NODQUOT)
1446 continue;
1447 warntype[cnt] = info_idq_free(inode->i_dquot[cnt], number);
1448 dquot_decr_inodes(inode->i_dquot[cnt], number);
1450 spin_unlock(&dq_data_lock);
1451 /* Dirtify all the dquots - this can block when journalling */
1452 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1453 if (inode->i_dquot[cnt])
1454 mark_dquot_dirty(inode->i_dquot[cnt]);
1455 flush_warnings(inode->i_dquot, warntype);
1456 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1457 return QUOTA_OK;
1461 * Transfer the number of inode and blocks from one diskquota to an other.
1463 * This operation can block, but only after everything is updated
1464 * A transaction must be started when entering this function.
1466 int dquot_transfer(struct inode *inode, struct iattr *iattr)
1468 qsize_t space;
1469 struct dquot *transfer_from[MAXQUOTAS];
1470 struct dquot *transfer_to[MAXQUOTAS];
1471 int cnt, ret = NO_QUOTA, chuid = (iattr->ia_valid & ATTR_UID) && inode->i_uid != iattr->ia_uid,
1472 chgid = (iattr->ia_valid & ATTR_GID) && inode->i_gid != iattr->ia_gid;
1473 char warntype_to[MAXQUOTAS];
1474 char warntype_from_inodes[MAXQUOTAS], warntype_from_space[MAXQUOTAS];
1476 /* First test before acquiring mutex - solves deadlocks when we
1477 * re-enter the quota code and are already holding the mutex */
1478 if (IS_NOQUOTA(inode))
1479 return QUOTA_OK;
1480 /* Clear the arrays */
1481 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1482 transfer_to[cnt] = transfer_from[cnt] = NODQUOT;
1483 warntype_to[cnt] = QUOTA_NL_NOWARN;
1485 down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1486 /* Now recheck reliably when holding dqptr_sem */
1487 if (IS_NOQUOTA(inode)) { /* File without quota accounting? */
1488 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1489 return QUOTA_OK;
1491 /* First build the transfer_to list - here we can block on
1492 * reading/instantiating of dquots. We know that the transaction for
1493 * us was already started so we don't violate lock ranking here */
1494 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1495 switch (cnt) {
1496 case USRQUOTA:
1497 if (!chuid)
1498 continue;
1499 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_uid, cnt);
1500 break;
1501 case GRPQUOTA:
1502 if (!chgid)
1503 continue;
1504 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_gid, cnt);
1505 break;
1508 spin_lock(&dq_data_lock);
1509 space = inode_get_bytes(inode);
1510 /* Build the transfer_from list and check the limits */
1511 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1512 if (transfer_to[cnt] == NODQUOT)
1513 continue;
1514 transfer_from[cnt] = inode->i_dquot[cnt];
1515 if (check_idq(transfer_to[cnt], 1, warntype_to + cnt) ==
1516 NO_QUOTA || check_bdq(transfer_to[cnt], space, 0,
1517 warntype_to + cnt) == NO_QUOTA)
1518 goto warn_put_all;
1522 * Finally perform the needed transfer from transfer_from to transfer_to
1524 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1526 * Skip changes for same uid or gid or for turned off quota-type.
1528 if (transfer_to[cnt] == NODQUOT)
1529 continue;
1531 /* Due to IO error we might not have transfer_from[] structure */
1532 if (transfer_from[cnt]) {
1533 warntype_from_inodes[cnt] =
1534 info_idq_free(transfer_from[cnt], 1);
1535 warntype_from_space[cnt] =
1536 info_bdq_free(transfer_from[cnt], space);
1537 dquot_decr_inodes(transfer_from[cnt], 1);
1538 dquot_decr_space(transfer_from[cnt], space);
1541 dquot_incr_inodes(transfer_to[cnt], 1);
1542 dquot_incr_space(transfer_to[cnt], space);
1544 inode->i_dquot[cnt] = transfer_to[cnt];
1546 ret = QUOTA_OK;
1547 warn_put_all:
1548 spin_unlock(&dq_data_lock);
1549 /* Dirtify all the dquots - this can block when journalling */
1550 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1551 if (transfer_from[cnt])
1552 mark_dquot_dirty(transfer_from[cnt]);
1553 if (transfer_to[cnt])
1554 mark_dquot_dirty(transfer_to[cnt]);
1556 flush_warnings(transfer_to, warntype_to);
1557 flush_warnings(transfer_from, warntype_from_inodes);
1558 flush_warnings(transfer_from, warntype_from_space);
1560 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1561 if (ret == QUOTA_OK && transfer_from[cnt] != NODQUOT)
1562 dqput(transfer_from[cnt]);
1563 if (ret == NO_QUOTA && transfer_to[cnt] != NODQUOT)
1564 dqput(transfer_to[cnt]);
1566 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1567 return ret;
1570 /* Wrapper for transferring ownership of an inode */
1571 int vfs_dq_transfer(struct inode *inode, struct iattr *iattr)
1573 if (sb_any_quota_active(inode->i_sb) && !IS_NOQUOTA(inode)) {
1574 vfs_dq_init(inode);
1575 if (inode->i_sb->dq_op->transfer(inode, iattr) == NO_QUOTA)
1576 return 1;
1578 return 0;
1583 * Write info of quota file to disk
1585 int dquot_commit_info(struct super_block *sb, int type)
1587 int ret;
1588 struct quota_info *dqopt = sb_dqopt(sb);
1590 mutex_lock(&dqopt->dqio_mutex);
1591 ret = dqopt->ops[type]->write_file_info(sb, type);
1592 mutex_unlock(&dqopt->dqio_mutex);
1593 return ret;
1597 * Definitions of diskquota operations.
1599 struct dquot_operations dquot_operations = {
1600 .initialize = dquot_initialize,
1601 .drop = dquot_drop,
1602 .alloc_space = dquot_alloc_space,
1603 .alloc_inode = dquot_alloc_inode,
1604 .free_space = dquot_free_space,
1605 .free_inode = dquot_free_inode,
1606 .transfer = dquot_transfer,
1607 .write_dquot = dquot_commit,
1608 .acquire_dquot = dquot_acquire,
1609 .release_dquot = dquot_release,
1610 .mark_dirty = dquot_mark_dquot_dirty,
1611 .write_info = dquot_commit_info,
1612 .alloc_dquot = dquot_alloc,
1613 .destroy_dquot = dquot_destroy,
1617 * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
1619 int vfs_quota_disable(struct super_block *sb, int type, unsigned int flags)
1621 int cnt, ret = 0;
1622 struct quota_info *dqopt = sb_dqopt(sb);
1623 struct inode *toputinode[MAXQUOTAS];
1625 /* Cannot turn off usage accounting without turning off limits, or
1626 * suspend quotas and simultaneously turn quotas off. */
1627 if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
1628 || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
1629 DQUOT_USAGE_ENABLED)))
1630 return -EINVAL;
1632 /* We need to serialize quota_off() for device */
1633 mutex_lock(&dqopt->dqonoff_mutex);
1636 * Skip everything if there's nothing to do. We have to do this because
1637 * sometimes we are called when fill_super() failed and calling
1638 * sync_fs() in such cases does no good.
1640 if (!sb_any_quota_loaded(sb)) {
1641 mutex_unlock(&dqopt->dqonoff_mutex);
1642 return 0;
1644 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1645 toputinode[cnt] = NULL;
1646 if (type != -1 && cnt != type)
1647 continue;
1648 if (!sb_has_quota_loaded(sb, cnt))
1649 continue;
1651 if (flags & DQUOT_SUSPENDED) {
1652 dqopt->flags |=
1653 dquot_state_flag(DQUOT_SUSPENDED, cnt);
1654 } else {
1655 dqopt->flags &= ~dquot_state_flag(flags, cnt);
1656 /* Turning off suspended quotas? */
1657 if (!sb_has_quota_loaded(sb, cnt) &&
1658 sb_has_quota_suspended(sb, cnt)) {
1659 dqopt->flags &= ~dquot_state_flag(
1660 DQUOT_SUSPENDED, cnt);
1661 iput(dqopt->files[cnt]);
1662 dqopt->files[cnt] = NULL;
1663 continue;
1667 /* We still have to keep quota loaded? */
1668 if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
1669 continue;
1671 /* Note: these are blocking operations */
1672 drop_dquot_ref(sb, cnt);
1673 invalidate_dquots(sb, cnt);
1675 * Now all dquots should be invalidated, all writes done so we should be only
1676 * users of the info. No locks needed.
1678 if (info_dirty(&dqopt->info[cnt]))
1679 sb->dq_op->write_info(sb, cnt);
1680 if (dqopt->ops[cnt]->free_file_info)
1681 dqopt->ops[cnt]->free_file_info(sb, cnt);
1682 put_quota_format(dqopt->info[cnt].dqi_format);
1684 toputinode[cnt] = dqopt->files[cnt];
1685 if (!sb_has_quota_loaded(sb, cnt))
1686 dqopt->files[cnt] = NULL;
1687 dqopt->info[cnt].dqi_flags = 0;
1688 dqopt->info[cnt].dqi_igrace = 0;
1689 dqopt->info[cnt].dqi_bgrace = 0;
1690 dqopt->ops[cnt] = NULL;
1692 mutex_unlock(&dqopt->dqonoff_mutex);
1694 /* Skip syncing and setting flags if quota files are hidden */
1695 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
1696 goto put_inodes;
1698 /* Sync the superblock so that buffers with quota data are written to
1699 * disk (and so userspace sees correct data afterwards). */
1700 if (sb->s_op->sync_fs)
1701 sb->s_op->sync_fs(sb, 1);
1702 sync_blockdev(sb->s_bdev);
1703 /* Now the quota files are just ordinary files and we can set the
1704 * inode flags back. Moreover we discard the pagecache so that
1705 * userspace sees the writes we did bypassing the pagecache. We
1706 * must also discard the blockdev buffers so that we see the
1707 * changes done by userspace on the next quotaon() */
1708 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1709 if (toputinode[cnt]) {
1710 mutex_lock(&dqopt->dqonoff_mutex);
1711 /* If quota was reenabled in the meantime, we have
1712 * nothing to do */
1713 if (!sb_has_quota_loaded(sb, cnt)) {
1714 mutex_lock_nested(&toputinode[cnt]->i_mutex, I_MUTEX_QUOTA);
1715 toputinode[cnt]->i_flags &= ~(S_IMMUTABLE |
1716 S_NOATIME | S_NOQUOTA);
1717 truncate_inode_pages(&toputinode[cnt]->i_data, 0);
1718 mutex_unlock(&toputinode[cnt]->i_mutex);
1719 mark_inode_dirty(toputinode[cnt]);
1721 mutex_unlock(&dqopt->dqonoff_mutex);
1723 if (sb->s_bdev)
1724 invalidate_bdev(sb->s_bdev);
1725 put_inodes:
1726 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1727 if (toputinode[cnt]) {
1728 /* On remount RO, we keep the inode pointer so that we
1729 * can reenable quota on the subsequent remount RW. We
1730 * have to check 'flags' variable and not use sb_has_
1731 * function because another quotaon / quotaoff could
1732 * change global state before we got here. We refuse
1733 * to suspend quotas when there is pending delete on
1734 * the quota file... */
1735 if (!(flags & DQUOT_SUSPENDED))
1736 iput(toputinode[cnt]);
1737 else if (!toputinode[cnt]->i_nlink)
1738 ret = -EBUSY;
1740 return ret;
1743 int vfs_quota_off(struct super_block *sb, int type, int remount)
1745 return vfs_quota_disable(sb, type, remount ? DQUOT_SUSPENDED :
1746 (DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED));
1750 * Turn quotas on on a device
1754 * Helper function to turn quotas on when we already have the inode of
1755 * quota file and no quota information is loaded.
1757 static int vfs_load_quota_inode(struct inode *inode, int type, int format_id,
1758 unsigned int flags)
1760 struct quota_format_type *fmt = find_quota_format(format_id);
1761 struct super_block *sb = inode->i_sb;
1762 struct quota_info *dqopt = sb_dqopt(sb);
1763 int error;
1764 int oldflags = -1;
1766 if (!fmt)
1767 return -ESRCH;
1768 if (!S_ISREG(inode->i_mode)) {
1769 error = -EACCES;
1770 goto out_fmt;
1772 if (IS_RDONLY(inode)) {
1773 error = -EROFS;
1774 goto out_fmt;
1776 if (!sb->s_op->quota_write || !sb->s_op->quota_read) {
1777 error = -EINVAL;
1778 goto out_fmt;
1780 /* Usage always has to be set... */
1781 if (!(flags & DQUOT_USAGE_ENABLED)) {
1782 error = -EINVAL;
1783 goto out_fmt;
1786 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
1787 /* As we bypass the pagecache we must now flush the inode so
1788 * that we see all the changes from userspace... */
1789 write_inode_now(inode, 1);
1790 /* And now flush the block cache so that kernel sees the
1791 * changes */
1792 invalidate_bdev(sb->s_bdev);
1794 mutex_lock(&inode->i_mutex);
1795 mutex_lock(&dqopt->dqonoff_mutex);
1796 if (sb_has_quota_loaded(sb, type)) {
1797 error = -EBUSY;
1798 goto out_lock;
1801 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
1802 /* We don't want quota and atime on quota files (deadlocks
1803 * possible) Also nobody should write to the file - we use
1804 * special IO operations which ignore the immutable bit. */
1805 down_write(&dqopt->dqptr_sem);
1806 oldflags = inode->i_flags & (S_NOATIME | S_IMMUTABLE | S_NOQUOTA);
1807 inode->i_flags |= S_NOQUOTA | S_NOATIME | S_IMMUTABLE;
1808 up_write(&dqopt->dqptr_sem);
1809 sb->dq_op->drop(inode);
1812 error = -EIO;
1813 dqopt->files[type] = igrab(inode);
1814 if (!dqopt->files[type])
1815 goto out_lock;
1816 error = -EINVAL;
1817 if (!fmt->qf_ops->check_quota_file(sb, type))
1818 goto out_file_init;
1820 dqopt->ops[type] = fmt->qf_ops;
1821 dqopt->info[type].dqi_format = fmt;
1822 dqopt->info[type].dqi_fmt_id = format_id;
1823 INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
1824 mutex_lock(&dqopt->dqio_mutex);
1825 if ((error = dqopt->ops[type]->read_file_info(sb, type)) < 0) {
1826 mutex_unlock(&dqopt->dqio_mutex);
1827 goto out_file_init;
1829 mutex_unlock(&dqopt->dqio_mutex);
1830 mutex_unlock(&inode->i_mutex);
1831 dqopt->flags |= dquot_state_flag(flags, type);
1833 add_dquot_ref(sb, type);
1834 mutex_unlock(&dqopt->dqonoff_mutex);
1836 return 0;
1838 out_file_init:
1839 dqopt->files[type] = NULL;
1840 iput(inode);
1841 out_lock:
1842 mutex_unlock(&dqopt->dqonoff_mutex);
1843 if (oldflags != -1) {
1844 down_write(&dqopt->dqptr_sem);
1845 /* Set the flags back (in the case of accidental quotaon()
1846 * on a wrong file we don't want to mess up the flags) */
1847 inode->i_flags &= ~(S_NOATIME | S_NOQUOTA | S_IMMUTABLE);
1848 inode->i_flags |= oldflags;
1849 up_write(&dqopt->dqptr_sem);
1851 mutex_unlock(&inode->i_mutex);
1852 out_fmt:
1853 put_quota_format(fmt);
1855 return error;
1858 /* Reenable quotas on remount RW */
1859 static int vfs_quota_on_remount(struct super_block *sb, int type)
1861 struct quota_info *dqopt = sb_dqopt(sb);
1862 struct inode *inode;
1863 int ret;
1864 unsigned int flags;
1866 mutex_lock(&dqopt->dqonoff_mutex);
1867 if (!sb_has_quota_suspended(sb, type)) {
1868 mutex_unlock(&dqopt->dqonoff_mutex);
1869 return 0;
1871 inode = dqopt->files[type];
1872 dqopt->files[type] = NULL;
1873 flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
1874 DQUOT_LIMITS_ENABLED, type);
1875 dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, type);
1876 mutex_unlock(&dqopt->dqonoff_mutex);
1878 flags = dquot_generic_flag(flags, type);
1879 ret = vfs_load_quota_inode(inode, type, dqopt->info[type].dqi_fmt_id,
1880 flags);
1881 iput(inode);
1883 return ret;
1886 int vfs_quota_on_path(struct super_block *sb, int type, int format_id,
1887 struct path *path)
1889 int error = security_quota_on(path->dentry);
1890 if (error)
1891 return error;
1892 /* Quota file not on the same filesystem? */
1893 if (path->mnt->mnt_sb != sb)
1894 error = -EXDEV;
1895 else
1896 error = vfs_load_quota_inode(path->dentry->d_inode, type,
1897 format_id, DQUOT_USAGE_ENABLED |
1898 DQUOT_LIMITS_ENABLED);
1899 return error;
1902 int vfs_quota_on(struct super_block *sb, int type, int format_id, char *name,
1903 int remount)
1905 struct path path;
1906 int error;
1908 if (remount)
1909 return vfs_quota_on_remount(sb, type);
1911 error = kern_path(name, LOOKUP_FOLLOW, &path);
1912 if (!error) {
1913 error = vfs_quota_on_path(sb, type, format_id, &path);
1914 path_put(&path);
1916 return error;
1920 * More powerful function for turning on quotas allowing setting
1921 * of individual quota flags
1923 int vfs_quota_enable(struct inode *inode, int type, int format_id,
1924 unsigned int flags)
1926 int ret = 0;
1927 struct super_block *sb = inode->i_sb;
1928 struct quota_info *dqopt = sb_dqopt(sb);
1930 /* Just unsuspend quotas? */
1931 if (flags & DQUOT_SUSPENDED)
1932 return vfs_quota_on_remount(sb, type);
1933 if (!flags)
1934 return 0;
1935 /* Just updating flags needed? */
1936 if (sb_has_quota_loaded(sb, type)) {
1937 mutex_lock(&dqopt->dqonoff_mutex);
1938 /* Now do a reliable test... */
1939 if (!sb_has_quota_loaded(sb, type)) {
1940 mutex_unlock(&dqopt->dqonoff_mutex);
1941 goto load_quota;
1943 if (flags & DQUOT_USAGE_ENABLED &&
1944 sb_has_quota_usage_enabled(sb, type)) {
1945 ret = -EBUSY;
1946 goto out_lock;
1948 if (flags & DQUOT_LIMITS_ENABLED &&
1949 sb_has_quota_limits_enabled(sb, type)) {
1950 ret = -EBUSY;
1951 goto out_lock;
1953 sb_dqopt(sb)->flags |= dquot_state_flag(flags, type);
1954 out_lock:
1955 mutex_unlock(&dqopt->dqonoff_mutex);
1956 return ret;
1959 load_quota:
1960 return vfs_load_quota_inode(inode, type, format_id, flags);
1964 * This function is used when filesystem needs to initialize quotas
1965 * during mount time.
1967 int vfs_quota_on_mount(struct super_block *sb, char *qf_name,
1968 int format_id, int type)
1970 struct dentry *dentry;
1971 int error;
1973 dentry = lookup_one_len(qf_name, sb->s_root, strlen(qf_name));
1974 if (IS_ERR(dentry))
1975 return PTR_ERR(dentry);
1977 if (!dentry->d_inode) {
1978 error = -ENOENT;
1979 goto out;
1982 error = security_quota_on(dentry);
1983 if (!error)
1984 error = vfs_load_quota_inode(dentry->d_inode, type, format_id,
1985 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
1987 out:
1988 dput(dentry);
1989 return error;
1992 /* Wrapper to turn on quotas when remounting rw */
1993 int vfs_dq_quota_on_remount(struct super_block *sb)
1995 int cnt;
1996 int ret = 0, err;
1998 if (!sb->s_qcop || !sb->s_qcop->quota_on)
1999 return -ENOSYS;
2000 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2001 err = sb->s_qcop->quota_on(sb, cnt, 0, NULL, 1);
2002 if (err < 0 && !ret)
2003 ret = err;
2005 return ret;
2008 static inline qsize_t qbtos(qsize_t blocks)
2010 return blocks << QIF_DQBLKSIZE_BITS;
2013 static inline qsize_t stoqb(qsize_t space)
2015 return (space + QIF_DQBLKSIZE - 1) >> QIF_DQBLKSIZE_BITS;
2018 /* Generic routine for getting common part of quota structure */
2019 static void do_get_dqblk(struct dquot *dquot, struct if_dqblk *di)
2021 struct mem_dqblk *dm = &dquot->dq_dqb;
2023 spin_lock(&dq_data_lock);
2024 di->dqb_bhardlimit = stoqb(dm->dqb_bhardlimit);
2025 di->dqb_bsoftlimit = stoqb(dm->dqb_bsoftlimit);
2026 di->dqb_curspace = dm->dqb_curspace;
2027 di->dqb_ihardlimit = dm->dqb_ihardlimit;
2028 di->dqb_isoftlimit = dm->dqb_isoftlimit;
2029 di->dqb_curinodes = dm->dqb_curinodes;
2030 di->dqb_btime = dm->dqb_btime;
2031 di->dqb_itime = dm->dqb_itime;
2032 di->dqb_valid = QIF_ALL;
2033 spin_unlock(&dq_data_lock);
2036 int vfs_get_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
2038 struct dquot *dquot;
2040 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
2041 if (!(dquot = dqget(sb, id, type))) {
2042 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2043 return -ESRCH;
2045 do_get_dqblk(dquot, di);
2046 dqput(dquot);
2047 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2048 return 0;
2051 /* Generic routine for setting common part of quota structure */
2052 static int do_set_dqblk(struct dquot *dquot, struct if_dqblk *di)
2054 struct mem_dqblk *dm = &dquot->dq_dqb;
2055 int check_blim = 0, check_ilim = 0;
2056 struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_type];
2058 if ((di->dqb_valid & QIF_BLIMITS &&
2059 (di->dqb_bhardlimit > dqi->dqi_maxblimit ||
2060 di->dqb_bsoftlimit > dqi->dqi_maxblimit)) ||
2061 (di->dqb_valid & QIF_ILIMITS &&
2062 (di->dqb_ihardlimit > dqi->dqi_maxilimit ||
2063 di->dqb_isoftlimit > dqi->dqi_maxilimit)))
2064 return -ERANGE;
2066 spin_lock(&dq_data_lock);
2067 if (di->dqb_valid & QIF_SPACE) {
2068 dm->dqb_curspace = di->dqb_curspace;
2069 check_blim = 1;
2070 __set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2072 if (di->dqb_valid & QIF_BLIMITS) {
2073 dm->dqb_bsoftlimit = qbtos(di->dqb_bsoftlimit);
2074 dm->dqb_bhardlimit = qbtos(di->dqb_bhardlimit);
2075 check_blim = 1;
2076 __set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2078 if (di->dqb_valid & QIF_INODES) {
2079 dm->dqb_curinodes = di->dqb_curinodes;
2080 check_ilim = 1;
2081 __set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2083 if (di->dqb_valid & QIF_ILIMITS) {
2084 dm->dqb_isoftlimit = di->dqb_isoftlimit;
2085 dm->dqb_ihardlimit = di->dqb_ihardlimit;
2086 check_ilim = 1;
2087 __set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2089 if (di->dqb_valid & QIF_BTIME) {
2090 dm->dqb_btime = di->dqb_btime;
2091 __set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2093 if (di->dqb_valid & QIF_ITIME) {
2094 dm->dqb_itime = di->dqb_itime;
2095 __set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2098 if (check_blim) {
2099 if (!dm->dqb_bsoftlimit || dm->dqb_curspace < dm->dqb_bsoftlimit) {
2100 dm->dqb_btime = 0;
2101 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2103 else if (!(di->dqb_valid & QIF_BTIME)) /* Set grace only if user hasn't provided his own... */
2104 dm->dqb_btime = get_seconds() + dqi->dqi_bgrace;
2106 if (check_ilim) {
2107 if (!dm->dqb_isoftlimit || dm->dqb_curinodes < dm->dqb_isoftlimit) {
2108 dm->dqb_itime = 0;
2109 clear_bit(DQ_INODES_B, &dquot->dq_flags);
2111 else if (!(di->dqb_valid & QIF_ITIME)) /* Set grace only if user hasn't provided his own... */
2112 dm->dqb_itime = get_seconds() + dqi->dqi_igrace;
2114 if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit || dm->dqb_isoftlimit)
2115 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2116 else
2117 set_bit(DQ_FAKE_B, &dquot->dq_flags);
2118 spin_unlock(&dq_data_lock);
2119 mark_dquot_dirty(dquot);
2121 return 0;
2124 int vfs_set_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
2126 struct dquot *dquot;
2127 int rc;
2129 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
2130 dquot = dqget(sb, id, type);
2131 if (!dquot) {
2132 rc = -ESRCH;
2133 goto out;
2135 rc = do_set_dqblk(dquot, di);
2136 dqput(dquot);
2137 out:
2138 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2139 return rc;
2142 /* Generic routine for getting common part of quota file information */
2143 int vfs_get_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
2145 struct mem_dqinfo *mi;
2147 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
2148 if (!sb_has_quota_active(sb, type)) {
2149 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2150 return -ESRCH;
2152 mi = sb_dqopt(sb)->info + type;
2153 spin_lock(&dq_data_lock);
2154 ii->dqi_bgrace = mi->dqi_bgrace;
2155 ii->dqi_igrace = mi->dqi_igrace;
2156 ii->dqi_flags = mi->dqi_flags & DQF_MASK;
2157 ii->dqi_valid = IIF_ALL;
2158 spin_unlock(&dq_data_lock);
2159 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2160 return 0;
2163 /* Generic routine for setting common part of quota file information */
2164 int vfs_set_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
2166 struct mem_dqinfo *mi;
2167 int err = 0;
2169 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
2170 if (!sb_has_quota_active(sb, type)) {
2171 err = -ESRCH;
2172 goto out;
2174 mi = sb_dqopt(sb)->info + type;
2175 spin_lock(&dq_data_lock);
2176 if (ii->dqi_valid & IIF_BGRACE)
2177 mi->dqi_bgrace = ii->dqi_bgrace;
2178 if (ii->dqi_valid & IIF_IGRACE)
2179 mi->dqi_igrace = ii->dqi_igrace;
2180 if (ii->dqi_valid & IIF_FLAGS)
2181 mi->dqi_flags = (mi->dqi_flags & ~DQF_MASK) | (ii->dqi_flags & DQF_MASK);
2182 spin_unlock(&dq_data_lock);
2183 mark_info_dirty(sb, type);
2184 /* Force write to disk */
2185 sb->dq_op->write_info(sb, type);
2186 out:
2187 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2188 return err;
2191 struct quotactl_ops vfs_quotactl_ops = {
2192 .quota_on = vfs_quota_on,
2193 .quota_off = vfs_quota_off,
2194 .quota_sync = vfs_quota_sync,
2195 .get_info = vfs_get_dqinfo,
2196 .set_info = vfs_set_dqinfo,
2197 .get_dqblk = vfs_get_dqblk,
2198 .set_dqblk = vfs_set_dqblk
2201 static ctl_table fs_dqstats_table[] = {
2203 .ctl_name = FS_DQ_LOOKUPS,
2204 .procname = "lookups",
2205 .data = &dqstats.lookups,
2206 .maxlen = sizeof(int),
2207 .mode = 0444,
2208 .proc_handler = &proc_dointvec,
2211 .ctl_name = FS_DQ_DROPS,
2212 .procname = "drops",
2213 .data = &dqstats.drops,
2214 .maxlen = sizeof(int),
2215 .mode = 0444,
2216 .proc_handler = &proc_dointvec,
2219 .ctl_name = FS_DQ_READS,
2220 .procname = "reads",
2221 .data = &dqstats.reads,
2222 .maxlen = sizeof(int),
2223 .mode = 0444,
2224 .proc_handler = &proc_dointvec,
2227 .ctl_name = FS_DQ_WRITES,
2228 .procname = "writes",
2229 .data = &dqstats.writes,
2230 .maxlen = sizeof(int),
2231 .mode = 0444,
2232 .proc_handler = &proc_dointvec,
2235 .ctl_name = FS_DQ_CACHE_HITS,
2236 .procname = "cache_hits",
2237 .data = &dqstats.cache_hits,
2238 .maxlen = sizeof(int),
2239 .mode = 0444,
2240 .proc_handler = &proc_dointvec,
2243 .ctl_name = FS_DQ_ALLOCATED,
2244 .procname = "allocated_dquots",
2245 .data = &dqstats.allocated_dquots,
2246 .maxlen = sizeof(int),
2247 .mode = 0444,
2248 .proc_handler = &proc_dointvec,
2251 .ctl_name = FS_DQ_FREE,
2252 .procname = "free_dquots",
2253 .data = &dqstats.free_dquots,
2254 .maxlen = sizeof(int),
2255 .mode = 0444,
2256 .proc_handler = &proc_dointvec,
2259 .ctl_name = FS_DQ_SYNCS,
2260 .procname = "syncs",
2261 .data = &dqstats.syncs,
2262 .maxlen = sizeof(int),
2263 .mode = 0444,
2264 .proc_handler = &proc_dointvec,
2266 #ifdef CONFIG_PRINT_QUOTA_WARNING
2268 .ctl_name = FS_DQ_WARNINGS,
2269 .procname = "warnings",
2270 .data = &flag_print_warnings,
2271 .maxlen = sizeof(int),
2272 .mode = 0644,
2273 .proc_handler = &proc_dointvec,
2275 #endif
2276 { .ctl_name = 0 },
2279 static ctl_table fs_table[] = {
2281 .ctl_name = FS_DQSTATS,
2282 .procname = "quota",
2283 .mode = 0555,
2284 .child = fs_dqstats_table,
2286 { .ctl_name = 0 },
2289 static ctl_table sys_table[] = {
2291 .ctl_name = CTL_FS,
2292 .procname = "fs",
2293 .mode = 0555,
2294 .child = fs_table,
2296 { .ctl_name = 0 },
2299 static int __init dquot_init(void)
2301 int i;
2302 unsigned long nr_hash, order;
2304 printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2306 register_sysctl_table(sys_table);
2308 dquot_cachep = kmem_cache_create("dquot",
2309 sizeof(struct dquot), sizeof(unsigned long) * 4,
2310 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2311 SLAB_MEM_SPREAD|SLAB_PANIC),
2312 NULL);
2314 order = 0;
2315 dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
2316 if (!dquot_hash)
2317 panic("Cannot create dquot hash table");
2319 /* Find power-of-two hlist_heads which can fit into allocation */
2320 nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2321 dq_hash_bits = 0;
2322 do {
2323 dq_hash_bits++;
2324 } while (nr_hash >> dq_hash_bits);
2325 dq_hash_bits--;
2327 nr_hash = 1UL << dq_hash_bits;
2328 dq_hash_mask = nr_hash - 1;
2329 for (i = 0; i < nr_hash; i++)
2330 INIT_HLIST_HEAD(dquot_hash + i);
2332 printk("Dquot-cache hash table entries: %ld (order %ld, %ld bytes)\n",
2333 nr_hash, order, (PAGE_SIZE << order));
2335 register_shrinker(&dqcache_shrinker);
2337 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
2338 if (genl_register_family(&quota_genl_family) != 0)
2339 printk(KERN_ERR "VFS: Failed to create quota netlink interface.\n");
2340 #endif
2342 return 0;
2344 module_init(dquot_init);
2346 EXPORT_SYMBOL(register_quota_format);
2347 EXPORT_SYMBOL(unregister_quota_format);
2348 EXPORT_SYMBOL(dqstats);
2349 EXPORT_SYMBOL(dq_data_lock);
2350 EXPORT_SYMBOL(vfs_quota_enable);
2351 EXPORT_SYMBOL(vfs_quota_on);
2352 EXPORT_SYMBOL(vfs_quota_on_path);
2353 EXPORT_SYMBOL(vfs_quota_on_mount);
2354 EXPORT_SYMBOL(vfs_quota_disable);
2355 EXPORT_SYMBOL(vfs_quota_off);
2356 EXPORT_SYMBOL(dquot_scan_active);
2357 EXPORT_SYMBOL(vfs_quota_sync);
2358 EXPORT_SYMBOL(vfs_get_dqinfo);
2359 EXPORT_SYMBOL(vfs_set_dqinfo);
2360 EXPORT_SYMBOL(vfs_get_dqblk);
2361 EXPORT_SYMBOL(vfs_set_dqblk);
2362 EXPORT_SYMBOL(dquot_commit);
2363 EXPORT_SYMBOL(dquot_commit_info);
2364 EXPORT_SYMBOL(dquot_acquire);
2365 EXPORT_SYMBOL(dquot_release);
2366 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
2367 EXPORT_SYMBOL(dquot_initialize);
2368 EXPORT_SYMBOL(dquot_drop);
2369 EXPORT_SYMBOL(dquot_drop_locked);
2370 EXPORT_SYMBOL(vfs_dq_drop);
2371 EXPORT_SYMBOL(dqget);
2372 EXPORT_SYMBOL(dqput);
2373 EXPORT_SYMBOL(dquot_is_cached);
2374 EXPORT_SYMBOL(dquot_alloc_space);
2375 EXPORT_SYMBOL(dquot_alloc_inode);
2376 EXPORT_SYMBOL(dquot_free_space);
2377 EXPORT_SYMBOL(dquot_free_inode);
2378 EXPORT_SYMBOL(dquot_transfer);
2379 EXPORT_SYMBOL(vfs_dq_transfer);
2380 EXPORT_SYMBOL(vfs_dq_quota_on_remount);