ACPI: fix memory hotplug range length handling
[linux-2.6/s3c2410-cpufreq.git] / fs / dquot.c
blob6b388692093976c1661546b619678cc40d37e000
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 * Version: $Id: dquot.c,v 6.3 1996/11/17 18:35:34 mvw Exp mvw $
14 * Author: Marco van Wieringen <mvw@planets.elm.net>
16 * Fixes: Dmitry Gorodchanin <pgmdsg@ibi.com>, 11 Feb 96
18 * Revised list management to avoid races
19 * -- Bill Hawes, <whawes@star.net>, 9/98
21 * Fixed races in dquot_transfer(), dqget() and dquot_alloc_...().
22 * As the consequence the locking was moved from dquot_decr_...(),
23 * dquot_incr_...() to calling functions.
24 * invalidate_dquots() now writes modified dquots.
25 * Serialized quota_off() and quota_on() for mount point.
26 * Fixed a few bugs in grow_dquots().
27 * Fixed deadlock in write_dquot() - we no longer account quotas on
28 * quota files
29 * remove_dquot_ref() moved to inode.c - it now traverses through inodes
30 * add_dquot_ref() restarts after blocking
31 * Added check for bogus uid and fixed check for group in quotactl.
32 * Jan Kara, <jack@suse.cz>, sponsored by SuSE CR, 10-11/99
34 * Used struct list_head instead of own list struct
35 * Invalidation of referenced dquots is no longer possible
36 * Improved free_dquots list management
37 * Quota and i_blocks are now updated in one place to avoid races
38 * Warnings are now delayed so we won't block in critical section
39 * Write updated not to require dquot lock
40 * Jan Kara, <jack@suse.cz>, 9/2000
42 * Added dynamic quota structure allocation
43 * Jan Kara <jack@suse.cz> 12/2000
45 * Rewritten quota interface. Implemented new quota format and
46 * formats registering.
47 * Jan Kara, <jack@suse.cz>, 2001,2002
49 * New SMP locking.
50 * Jan Kara, <jack@suse.cz>, 10/2002
52 * Added journalled quota support, fix lock inversion problems
53 * Jan Kara, <jack@suse.cz>, 2003,2004
55 * (C) Copyright 1994 - 1997 Marco van Wieringen
58 #include <linux/errno.h>
59 #include <linux/kernel.h>
60 #include <linux/fs.h>
61 #include <linux/mount.h>
62 #include <linux/mm.h>
63 #include <linux/time.h>
64 #include <linux/types.h>
65 #include <linux/string.h>
66 #include <linux/fcntl.h>
67 #include <linux/stat.h>
68 #include <linux/tty.h>
69 #include <linux/file.h>
70 #include <linux/slab.h>
71 #include <linux/sysctl.h>
72 #include <linux/smp_lock.h>
73 #include <linux/init.h>
74 #include <linux/module.h>
75 #include <linux/proc_fs.h>
76 #include <linux/security.h>
77 #include <linux/kmod.h>
78 #include <linux/namei.h>
79 #include <linux/buffer_head.h>
80 #include <linux/capability.h>
81 #include <linux/quotaops.h>
83 #include <asm/uaccess.h>
85 #define __DQUOT_PARANOIA
88 * There are two quota SMP locks. dq_list_lock protects all lists with quotas
89 * and quota formats and also dqstats structure containing statistics about the
90 * lists. dq_data_lock protects data from dq_dqb and also mem_dqinfo structures
91 * and also guards consistency of dquot->dq_dqb with inode->i_blocks, i_bytes.
92 * i_blocks and i_bytes updates itself are guarded by i_lock acquired directly
93 * in inode_add_bytes() and inode_sub_bytes().
95 * The spinlock ordering is hence: dq_data_lock > dq_list_lock > i_lock
97 * Note that some things (eg. sb pointer, type, id) doesn't change during
98 * the life of the dquot structure and so needn't to be protected by a lock
100 * Any operation working on dquots via inode pointers must hold dqptr_sem. If
101 * operation is just reading pointers from inode (or not using them at all) the
102 * read lock is enough. If pointers are altered function must hold write lock
103 * (these locking rules also apply for S_NOQUOTA flag in the inode - note that
104 * for altering the flag i_mutex is also needed). If operation is holding
105 * reference to dquot in other way (e.g. quotactl ops) it must be guarded by
106 * dqonoff_mutex.
107 * This locking assures that:
108 * a) update/access to dquot pointers in inode is serialized
109 * b) everyone is guarded against invalidate_dquots()
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 * i_mutex on quota files is special (it's below dqio_mutex)
126 static DEFINE_SPINLOCK(dq_list_lock);
127 DEFINE_SPINLOCK(dq_data_lock);
129 static char *quotatypes[] = INITQFNAMES;
130 static struct quota_format_type *quota_formats; /* List of registered formats */
131 static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
133 /* SLAB cache for dquot structures */
134 static kmem_cache_t *dquot_cachep;
136 int register_quota_format(struct quota_format_type *fmt)
138 spin_lock(&dq_list_lock);
139 fmt->qf_next = quota_formats;
140 quota_formats = fmt;
141 spin_unlock(&dq_list_lock);
142 return 0;
145 void unregister_quota_format(struct quota_format_type *fmt)
147 struct quota_format_type **actqf;
149 spin_lock(&dq_list_lock);
150 for (actqf = &quota_formats; *actqf && *actqf != fmt; actqf = &(*actqf)->qf_next);
151 if (*actqf)
152 *actqf = (*actqf)->qf_next;
153 spin_unlock(&dq_list_lock);
156 static struct quota_format_type *find_quota_format(int id)
158 struct quota_format_type *actqf;
160 spin_lock(&dq_list_lock);
161 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id; actqf = actqf->qf_next);
162 if (!actqf || !try_module_get(actqf->qf_owner)) {
163 int qm;
165 spin_unlock(&dq_list_lock);
167 for (qm = 0; module_names[qm].qm_fmt_id && module_names[qm].qm_fmt_id != id; qm++);
168 if (!module_names[qm].qm_fmt_id || request_module(module_names[qm].qm_mod_name))
169 return NULL;
171 spin_lock(&dq_list_lock);
172 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id; actqf = actqf->qf_next);
173 if (actqf && !try_module_get(actqf->qf_owner))
174 actqf = NULL;
176 spin_unlock(&dq_list_lock);
177 return actqf;
180 static void put_quota_format(struct quota_format_type *fmt)
182 module_put(fmt->qf_owner);
186 * Dquot List Management:
187 * The quota code uses three lists for dquot management: the inuse_list,
188 * free_dquots, and dquot_hash[] array. A single dquot structure may be
189 * on all three lists, depending on its current state.
191 * All dquots are placed to the end of inuse_list when first created, and this
192 * list is used for invalidate operation, which must look at every dquot.
194 * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
195 * and this list is searched whenever we need an available dquot. Dquots are
196 * removed from the list as soon as they are used again, and
197 * dqstats.free_dquots gives the number of dquots on the list. When
198 * dquot is invalidated it's completely released from memory.
200 * Dquots with a specific identity (device, type and id) are placed on
201 * one of the dquot_hash[] hash chains. The provides an efficient search
202 * mechanism to locate a specific dquot.
205 static LIST_HEAD(inuse_list);
206 static LIST_HEAD(free_dquots);
207 static unsigned int dq_hash_bits, dq_hash_mask;
208 static struct hlist_head *dquot_hash;
210 struct dqstats dqstats;
212 static void dqput(struct dquot *dquot);
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(&dquot->dq_free, free_dquots.prev);
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(&dquot->dq_inuse, inuse_list.prev);
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 #define mark_dquot_dirty(dquot) ((dquot)->dq_sb->dq_op->mark_dirty(dquot))
290 int dquot_mark_dquot_dirty(struct dquot *dquot)
292 spin_lock(&dq_list_lock);
293 if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags))
294 list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
295 info[dquot->dq_type].dqi_dirty_list);
296 spin_unlock(&dq_list_lock);
297 return 0;
300 /* This function needs dq_list_lock */
301 static inline int clear_dquot_dirty(struct dquot *dquot)
303 if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags))
304 return 0;
305 list_del_init(&dquot->dq_dirty);
306 return 1;
309 void mark_info_dirty(struct super_block *sb, int type)
311 set_bit(DQF_INFO_DIRTY_B, &sb_dqopt(sb)->info[type].dqi_flags);
313 EXPORT_SYMBOL(mark_info_dirty);
316 * Read dquot from disk and alloc space for it
319 int dquot_acquire(struct dquot *dquot)
321 int ret = 0, ret2 = 0;
322 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
324 mutex_lock(&dquot->dq_lock);
325 mutex_lock(&dqopt->dqio_mutex);
326 if (!test_bit(DQ_READ_B, &dquot->dq_flags))
327 ret = dqopt->ops[dquot->dq_type]->read_dqblk(dquot);
328 if (ret < 0)
329 goto out_iolock;
330 set_bit(DQ_READ_B, &dquot->dq_flags);
331 /* Instantiate dquot if needed */
332 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
333 ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
334 /* Write the info if needed */
335 if (info_dirty(&dqopt->info[dquot->dq_type]))
336 ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
337 if (ret < 0)
338 goto out_iolock;
339 if (ret2 < 0) {
340 ret = ret2;
341 goto out_iolock;
344 set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
345 out_iolock:
346 mutex_unlock(&dqopt->dqio_mutex);
347 mutex_unlock(&dquot->dq_lock);
348 return ret;
352 * Write dquot to disk
354 int dquot_commit(struct dquot *dquot)
356 int ret = 0, ret2 = 0;
357 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
359 mutex_lock(&dqopt->dqio_mutex);
360 spin_lock(&dq_list_lock);
361 if (!clear_dquot_dirty(dquot)) {
362 spin_unlock(&dq_list_lock);
363 goto out_sem;
365 spin_unlock(&dq_list_lock);
366 /* Inactive dquot can be only if there was error during read/init
367 * => we have better not writing it */
368 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
369 ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
370 if (info_dirty(&dqopt->info[dquot->dq_type]))
371 ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
372 if (ret >= 0)
373 ret = ret2;
375 out_sem:
376 mutex_unlock(&dqopt->dqio_mutex);
377 return ret;
381 * Release dquot
383 int dquot_release(struct dquot *dquot)
385 int ret = 0, ret2 = 0;
386 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
388 mutex_lock(&dquot->dq_lock);
389 /* Check whether we are not racing with some other dqget() */
390 if (atomic_read(&dquot->dq_count) > 1)
391 goto out_dqlock;
392 mutex_lock(&dqopt->dqio_mutex);
393 if (dqopt->ops[dquot->dq_type]->release_dqblk) {
394 ret = dqopt->ops[dquot->dq_type]->release_dqblk(dquot);
395 /* Write the info */
396 if (info_dirty(&dqopt->info[dquot->dq_type]))
397 ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
398 if (ret >= 0)
399 ret = ret2;
401 clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
402 mutex_unlock(&dqopt->dqio_mutex);
403 out_dqlock:
404 mutex_unlock(&dquot->dq_lock);
405 return ret;
408 /* Invalidate all dquots on the list. Note that this function is called after
409 * quota is disabled and pointers from inodes removed so there cannot be new
410 * quota users. There can still be some users of quotas due to inodes being
411 * just deleted or pruned by prune_icache() (those are not attached to any
412 * list). We have to wait for such users.
414 static void invalidate_dquots(struct super_block *sb, int type)
416 struct dquot *dquot, *tmp;
418 restart:
419 spin_lock(&dq_list_lock);
420 list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
421 if (dquot->dq_sb != sb)
422 continue;
423 if (dquot->dq_type != type)
424 continue;
425 /* Wait for dquot users */
426 if (atomic_read(&dquot->dq_count)) {
427 DEFINE_WAIT(wait);
429 atomic_inc(&dquot->dq_count);
430 prepare_to_wait(&dquot->dq_wait_unused, &wait,
431 TASK_UNINTERRUPTIBLE);
432 spin_unlock(&dq_list_lock);
433 /* Once dqput() wakes us up, we know it's time to free
434 * the dquot.
435 * IMPORTANT: we rely on the fact that there is always
436 * at most one process waiting for dquot to free.
437 * Otherwise dq_count would be > 1 and we would never
438 * wake up.
440 if (atomic_read(&dquot->dq_count) > 1)
441 schedule();
442 finish_wait(&dquot->dq_wait_unused, &wait);
443 dqput(dquot);
444 /* At this moment dquot() need not exist (it could be
445 * reclaimed by prune_dqcache(). Hence we must
446 * restart. */
447 goto restart;
450 * Quota now has no users and it has been written on last
451 * dqput()
453 remove_dquot_hash(dquot);
454 remove_free_dquot(dquot);
455 remove_inuse(dquot);
456 kmem_cache_free(dquot_cachep, dquot);
458 spin_unlock(&dq_list_lock);
461 int vfs_quota_sync(struct super_block *sb, int type)
463 struct list_head *dirty;
464 struct dquot *dquot;
465 struct quota_info *dqopt = sb_dqopt(sb);
466 int cnt;
468 mutex_lock(&dqopt->dqonoff_mutex);
469 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
470 if (type != -1 && cnt != type)
471 continue;
472 if (!sb_has_quota_enabled(sb, cnt))
473 continue;
474 spin_lock(&dq_list_lock);
475 dirty = &dqopt->info[cnt].dqi_dirty_list;
476 while (!list_empty(dirty)) {
477 dquot = list_entry(dirty->next, struct dquot, dq_dirty);
478 /* Dirty and inactive can be only bad dquot... */
479 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
480 clear_dquot_dirty(dquot);
481 continue;
483 /* Now we have active dquot from which someone is
484 * holding reference so we can safely just increase
485 * use count */
486 atomic_inc(&dquot->dq_count);
487 dqstats.lookups++;
488 spin_unlock(&dq_list_lock);
489 sb->dq_op->write_dquot(dquot);
490 dqput(dquot);
491 spin_lock(&dq_list_lock);
493 spin_unlock(&dq_list_lock);
496 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
497 if ((cnt == type || type == -1) && sb_has_quota_enabled(sb, cnt)
498 && info_dirty(&dqopt->info[cnt]))
499 sb->dq_op->write_info(sb, cnt);
500 spin_lock(&dq_list_lock);
501 dqstats.syncs++;
502 spin_unlock(&dq_list_lock);
503 mutex_unlock(&dqopt->dqonoff_mutex);
505 return 0;
508 /* Free unused dquots from cache */
509 static void prune_dqcache(int count)
511 struct list_head *head;
512 struct dquot *dquot;
514 head = free_dquots.prev;
515 while (head != &free_dquots && count) {
516 dquot = list_entry(head, struct dquot, dq_free);
517 remove_dquot_hash(dquot);
518 remove_free_dquot(dquot);
519 remove_inuse(dquot);
520 kmem_cache_free(dquot_cachep, dquot);
521 count--;
522 head = free_dquots.prev;
527 * This is called from kswapd when we think we need some
528 * more memory
531 static int shrink_dqcache_memory(int nr, gfp_t gfp_mask)
533 if (nr) {
534 spin_lock(&dq_list_lock);
535 prune_dqcache(nr);
536 spin_unlock(&dq_list_lock);
538 return (dqstats.free_dquots / 100) * sysctl_vfs_cache_pressure;
542 * Put reference to dquot
543 * NOTE: If you change this function please check whether dqput_blocks() works right...
544 * MUST be called with either dqptr_sem or dqonoff_mutex held
546 static void dqput(struct dquot *dquot)
548 if (!dquot)
549 return;
550 #ifdef __DQUOT_PARANOIA
551 if (!atomic_read(&dquot->dq_count)) {
552 printk("VFS: dqput: trying to free free dquot\n");
553 printk("VFS: device %s, dquot of %s %d\n",
554 dquot->dq_sb->s_id,
555 quotatypes[dquot->dq_type],
556 dquot->dq_id);
557 BUG();
559 #endif
561 spin_lock(&dq_list_lock);
562 dqstats.drops++;
563 spin_unlock(&dq_list_lock);
564 we_slept:
565 spin_lock(&dq_list_lock);
566 if (atomic_read(&dquot->dq_count) > 1) {
567 /* We have more than one user... nothing to do */
568 atomic_dec(&dquot->dq_count);
569 /* Releasing dquot during quotaoff phase? */
570 if (!sb_has_quota_enabled(dquot->dq_sb, dquot->dq_type) &&
571 atomic_read(&dquot->dq_count) == 1)
572 wake_up(&dquot->dq_wait_unused);
573 spin_unlock(&dq_list_lock);
574 return;
576 /* Need to release dquot? */
577 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && dquot_dirty(dquot)) {
578 spin_unlock(&dq_list_lock);
579 /* Commit dquot before releasing */
580 dquot->dq_sb->dq_op->write_dquot(dquot);
581 goto we_slept;
583 /* Clear flag in case dquot was inactive (something bad happened) */
584 clear_dquot_dirty(dquot);
585 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
586 spin_unlock(&dq_list_lock);
587 dquot->dq_sb->dq_op->release_dquot(dquot);
588 goto we_slept;
590 atomic_dec(&dquot->dq_count);
591 #ifdef __DQUOT_PARANOIA
592 /* sanity check */
593 if (!list_empty(&dquot->dq_free))
594 BUG();
595 #endif
596 put_dquot_last(dquot);
597 spin_unlock(&dq_list_lock);
600 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
602 struct dquot *dquot;
604 dquot = kmem_cache_alloc(dquot_cachep, SLAB_NOFS);
605 if(!dquot)
606 return NODQUOT;
608 memset((caddr_t)dquot, 0, sizeof(struct dquot));
609 mutex_init(&dquot->dq_lock);
610 INIT_LIST_HEAD(&dquot->dq_free);
611 INIT_LIST_HEAD(&dquot->dq_inuse);
612 INIT_HLIST_NODE(&dquot->dq_hash);
613 INIT_LIST_HEAD(&dquot->dq_dirty);
614 init_waitqueue_head(&dquot->dq_wait_unused);
615 dquot->dq_sb = sb;
616 dquot->dq_type = type;
617 atomic_set(&dquot->dq_count, 1);
619 return dquot;
623 * Get reference to dquot
624 * MUST be called with either dqptr_sem or dqonoff_mutex held
626 static struct dquot *dqget(struct super_block *sb, unsigned int id, int type)
628 unsigned int hashent = hashfn(sb, id, type);
629 struct dquot *dquot, *empty = NODQUOT;
631 if (!sb_has_quota_enabled(sb, type))
632 return NODQUOT;
633 we_slept:
634 spin_lock(&dq_list_lock);
635 if ((dquot = find_dquot(hashent, sb, id, type)) == NODQUOT) {
636 if (empty == NODQUOT) {
637 spin_unlock(&dq_list_lock);
638 if ((empty = get_empty_dquot(sb, type)) == NODQUOT)
639 schedule(); /* Try to wait for a moment... */
640 goto we_slept;
642 dquot = empty;
643 dquot->dq_id = id;
644 /* all dquots go on the inuse_list */
645 put_inuse(dquot);
646 /* hash it first so it can be found */
647 insert_dquot_hash(dquot);
648 dqstats.lookups++;
649 spin_unlock(&dq_list_lock);
650 } else {
651 if (!atomic_read(&dquot->dq_count))
652 remove_free_dquot(dquot);
653 atomic_inc(&dquot->dq_count);
654 dqstats.cache_hits++;
655 dqstats.lookups++;
656 spin_unlock(&dq_list_lock);
657 if (empty)
658 kmem_cache_free(dquot_cachep, empty);
660 /* Wait for dq_lock - after this we know that either dquot_release() is already
661 * finished or it will be canceled due to dq_count > 1 test */
662 wait_on_dquot(dquot);
663 /* Read the dquot and instantiate it (everything done only if needed) */
664 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && sb->dq_op->acquire_dquot(dquot) < 0) {
665 dqput(dquot);
666 return NODQUOT;
668 #ifdef __DQUOT_PARANOIA
669 if (!dquot->dq_sb) /* Has somebody invalidated entry under us? */
670 BUG();
671 #endif
673 return dquot;
676 static int dqinit_needed(struct inode *inode, int type)
678 int cnt;
680 if (IS_NOQUOTA(inode))
681 return 0;
682 if (type != -1)
683 return inode->i_dquot[type] == NODQUOT;
684 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
685 if (inode->i_dquot[cnt] == NODQUOT)
686 return 1;
687 return 0;
690 /* This routine is guarded by dqonoff_mutex mutex */
691 static void add_dquot_ref(struct super_block *sb, int type)
693 struct list_head *p;
695 restart:
696 file_list_lock();
697 list_for_each(p, &sb->s_files) {
698 struct file *filp = list_entry(p, struct file, f_u.fu_list);
699 struct inode *inode = filp->f_dentry->d_inode;
700 if (filp->f_mode & FMODE_WRITE && dqinit_needed(inode, type)) {
701 struct dentry *dentry = dget(filp->f_dentry);
702 file_list_unlock();
703 sb->dq_op->initialize(inode, type);
704 dput(dentry);
705 /* As we may have blocked we had better restart... */
706 goto restart;
709 file_list_unlock();
712 /* Return 0 if dqput() won't block (note that 1 doesn't necessarily mean blocking) */
713 static inline int dqput_blocks(struct dquot *dquot)
715 if (atomic_read(&dquot->dq_count) <= 1)
716 return 1;
717 return 0;
720 /* Remove references to dquots from inode - add dquot to list for freeing if needed */
721 /* We can't race with anybody because we hold dqptr_sem for writing... */
722 int remove_inode_dquot_ref(struct inode *inode, int type, struct list_head *tofree_head)
724 struct dquot *dquot = inode->i_dquot[type];
726 inode->i_dquot[type] = NODQUOT;
727 if (dquot != NODQUOT) {
728 if (dqput_blocks(dquot)) {
729 #ifdef __DQUOT_PARANOIA
730 if (atomic_read(&dquot->dq_count) != 1)
731 printk(KERN_WARNING "VFS: Adding dquot with dq_count %d to dispose list.\n", atomic_read(&dquot->dq_count));
732 #endif
733 spin_lock(&dq_list_lock);
734 list_add(&dquot->dq_free, tofree_head); /* As dquot must have currently users it can't be on the free list... */
735 spin_unlock(&dq_list_lock);
736 return 1;
738 else
739 dqput(dquot); /* We have guaranteed we won't block */
741 return 0;
744 /* Free list of dquots - called from inode.c */
745 /* dquots are removed from inodes, no new references can be got so we are the only ones holding reference */
746 static void put_dquot_list(struct list_head *tofree_head)
748 struct list_head *act_head;
749 struct dquot *dquot;
751 act_head = tofree_head->next;
752 /* So now we have dquots on the list... Just free them */
753 while (act_head != tofree_head) {
754 dquot = list_entry(act_head, struct dquot, dq_free);
755 act_head = act_head->next;
756 list_del_init(&dquot->dq_free); /* Remove dquot from the list so we won't have problems... */
757 dqput(dquot);
761 /* Gather all references from inodes and drop them */
762 static void drop_dquot_ref(struct super_block *sb, int type)
764 LIST_HEAD(tofree_head);
766 down_write(&sb_dqopt(sb)->dqptr_sem);
767 remove_dquot_ref(sb, type, &tofree_head);
768 up_write(&sb_dqopt(sb)->dqptr_sem);
769 put_dquot_list(&tofree_head);
772 static inline void dquot_incr_inodes(struct dquot *dquot, unsigned long number)
774 dquot->dq_dqb.dqb_curinodes += number;
777 static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
779 dquot->dq_dqb.dqb_curspace += number;
782 static inline void dquot_decr_inodes(struct dquot *dquot, unsigned long number)
784 if (dquot->dq_dqb.dqb_curinodes > number)
785 dquot->dq_dqb.dqb_curinodes -= number;
786 else
787 dquot->dq_dqb.dqb_curinodes = 0;
788 if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
789 dquot->dq_dqb.dqb_itime = (time_t) 0;
790 clear_bit(DQ_INODES_B, &dquot->dq_flags);
793 static inline void dquot_decr_space(struct dquot *dquot, qsize_t number)
795 if (dquot->dq_dqb.dqb_curspace > number)
796 dquot->dq_dqb.dqb_curspace -= number;
797 else
798 dquot->dq_dqb.dqb_curspace = 0;
799 if (toqb(dquot->dq_dqb.dqb_curspace) <= dquot->dq_dqb.dqb_bsoftlimit)
800 dquot->dq_dqb.dqb_btime = (time_t) 0;
801 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
804 static int flag_print_warnings = 1;
806 static inline int need_print_warning(struct dquot *dquot)
808 if (!flag_print_warnings)
809 return 0;
811 switch (dquot->dq_type) {
812 case USRQUOTA:
813 return current->fsuid == dquot->dq_id;
814 case GRPQUOTA:
815 return in_group_p(dquot->dq_id);
817 return 0;
820 /* Values of warnings */
821 #define NOWARN 0
822 #define IHARDWARN 1
823 #define ISOFTLONGWARN 2
824 #define ISOFTWARN 3
825 #define BHARDWARN 4
826 #define BSOFTLONGWARN 5
827 #define BSOFTWARN 6
829 /* Print warning to user which exceeded quota */
830 static void print_warning(struct dquot *dquot, const char warntype)
832 char *msg = NULL;
833 int flag = (warntype == BHARDWARN || warntype == BSOFTLONGWARN) ? DQ_BLKS_B :
834 ((warntype == IHARDWARN || warntype == ISOFTLONGWARN) ? DQ_INODES_B : 0);
836 if (!need_print_warning(dquot) || (flag && test_and_set_bit(flag, &dquot->dq_flags)))
837 return;
839 tty_write_message(current->signal->tty, dquot->dq_sb->s_id);
840 if (warntype == ISOFTWARN || warntype == BSOFTWARN)
841 tty_write_message(current->signal->tty, ": warning, ");
842 else
843 tty_write_message(current->signal->tty, ": write failed, ");
844 tty_write_message(current->signal->tty, quotatypes[dquot->dq_type]);
845 switch (warntype) {
846 case IHARDWARN:
847 msg = " file limit reached.\r\n";
848 break;
849 case ISOFTLONGWARN:
850 msg = " file quota exceeded too long.\r\n";
851 break;
852 case ISOFTWARN:
853 msg = " file quota exceeded.\r\n";
854 break;
855 case BHARDWARN:
856 msg = " block limit reached.\r\n";
857 break;
858 case BSOFTLONGWARN:
859 msg = " block quota exceeded too long.\r\n";
860 break;
861 case BSOFTWARN:
862 msg = " block quota exceeded.\r\n";
863 break;
865 tty_write_message(current->signal->tty, msg);
868 static inline void flush_warnings(struct dquot **dquots, char *warntype)
870 int i;
872 for (i = 0; i < MAXQUOTAS; i++)
873 if (dquots[i] != NODQUOT && warntype[i] != NOWARN)
874 print_warning(dquots[i], warntype[i]);
877 static inline char ignore_hardlimit(struct dquot *dquot)
879 struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_type];
881 return capable(CAP_SYS_RESOURCE) &&
882 (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD || !(info->dqi_flags & V1_DQF_RSQUASH));
885 /* needs dq_data_lock */
886 static int check_idq(struct dquot *dquot, ulong inodes, char *warntype)
888 *warntype = NOWARN;
889 if (inodes <= 0 || test_bit(DQ_FAKE_B, &dquot->dq_flags))
890 return QUOTA_OK;
892 if (dquot->dq_dqb.dqb_ihardlimit &&
893 (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_ihardlimit &&
894 !ignore_hardlimit(dquot)) {
895 *warntype = IHARDWARN;
896 return NO_QUOTA;
899 if (dquot->dq_dqb.dqb_isoftlimit &&
900 (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
901 dquot->dq_dqb.dqb_itime && get_seconds() >= dquot->dq_dqb.dqb_itime &&
902 !ignore_hardlimit(dquot)) {
903 *warntype = ISOFTLONGWARN;
904 return NO_QUOTA;
907 if (dquot->dq_dqb.dqb_isoftlimit &&
908 (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
909 dquot->dq_dqb.dqb_itime == 0) {
910 *warntype = ISOFTWARN;
911 dquot->dq_dqb.dqb_itime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
914 return QUOTA_OK;
917 /* needs dq_data_lock */
918 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc, char *warntype)
920 *warntype = 0;
921 if (space <= 0 || test_bit(DQ_FAKE_B, &dquot->dq_flags))
922 return QUOTA_OK;
924 if (dquot->dq_dqb.dqb_bhardlimit &&
925 toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bhardlimit &&
926 !ignore_hardlimit(dquot)) {
927 if (!prealloc)
928 *warntype = BHARDWARN;
929 return NO_QUOTA;
932 if (dquot->dq_dqb.dqb_bsoftlimit &&
933 toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bsoftlimit &&
934 dquot->dq_dqb.dqb_btime && get_seconds() >= dquot->dq_dqb.dqb_btime &&
935 !ignore_hardlimit(dquot)) {
936 if (!prealloc)
937 *warntype = BSOFTLONGWARN;
938 return NO_QUOTA;
941 if (dquot->dq_dqb.dqb_bsoftlimit &&
942 toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bsoftlimit &&
943 dquot->dq_dqb.dqb_btime == 0) {
944 if (!prealloc) {
945 *warntype = BSOFTWARN;
946 dquot->dq_dqb.dqb_btime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_bgrace;
948 else
950 * We don't allow preallocation to exceed softlimit so exceeding will
951 * be always printed
953 return NO_QUOTA;
956 return QUOTA_OK;
960 * Initialize quota pointers in inode
961 * Transaction must be started at entry
963 int dquot_initialize(struct inode *inode, int type)
965 unsigned int id = 0;
966 int cnt, ret = 0;
968 /* First test before acquiring mutex - solves deadlocks when we
969 * re-enter the quota code and are already holding the mutex */
970 if (IS_NOQUOTA(inode))
971 return 0;
972 down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
973 /* Having dqptr_sem we know NOQUOTA flags can't be altered... */
974 if (IS_NOQUOTA(inode))
975 goto out_err;
976 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
977 if (type != -1 && cnt != type)
978 continue;
979 if (inode->i_dquot[cnt] == NODQUOT) {
980 switch (cnt) {
981 case USRQUOTA:
982 id = inode->i_uid;
983 break;
984 case GRPQUOTA:
985 id = inode->i_gid;
986 break;
988 inode->i_dquot[cnt] = dqget(inode->i_sb, id, cnt);
991 out_err:
992 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
993 return ret;
997 * Release all quotas referenced by inode
998 * Transaction must be started at an entry
1000 int dquot_drop(struct inode *inode)
1002 int cnt;
1004 down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1005 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1006 if (inode->i_dquot[cnt] != NODQUOT) {
1007 dqput(inode->i_dquot[cnt]);
1008 inode->i_dquot[cnt] = NODQUOT;
1011 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1012 return 0;
1016 * Following four functions update i_blocks+i_bytes fields and
1017 * quota information (together with appropriate checks)
1018 * NOTE: We absolutely rely on the fact that caller dirties
1019 * the inode (usually macros in quotaops.h care about this) and
1020 * holds a handle for the current transaction so that dquot write and
1021 * inode write go into the same transaction.
1025 * This operation can block, but only after everything is updated
1027 int dquot_alloc_space(struct inode *inode, qsize_t number, int warn)
1029 int cnt, ret = NO_QUOTA;
1030 char warntype[MAXQUOTAS];
1032 /* First test before acquiring mutex - solves deadlocks when we
1033 * re-enter the quota code and are already holding the mutex */
1034 if (IS_NOQUOTA(inode)) {
1035 out_add:
1036 inode_add_bytes(inode, number);
1037 return QUOTA_OK;
1039 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1040 warntype[cnt] = NOWARN;
1042 down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1043 if (IS_NOQUOTA(inode)) { /* Now we can do reliable test... */
1044 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1045 goto out_add;
1047 spin_lock(&dq_data_lock);
1048 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1049 if (inode->i_dquot[cnt] == NODQUOT)
1050 continue;
1051 if (check_bdq(inode->i_dquot[cnt], number, warn, warntype+cnt) == NO_QUOTA)
1052 goto warn_put_all;
1054 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1055 if (inode->i_dquot[cnt] == NODQUOT)
1056 continue;
1057 dquot_incr_space(inode->i_dquot[cnt], number);
1059 inode_add_bytes(inode, number);
1060 ret = QUOTA_OK;
1061 warn_put_all:
1062 spin_unlock(&dq_data_lock);
1063 if (ret == QUOTA_OK)
1064 /* Dirtify all the dquots - this can block when journalling */
1065 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1066 if (inode->i_dquot[cnt])
1067 mark_dquot_dirty(inode->i_dquot[cnt]);
1068 flush_warnings(inode->i_dquot, warntype);
1069 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1070 return ret;
1074 * This operation can block, but only after everything is updated
1076 int dquot_alloc_inode(const struct inode *inode, unsigned long number)
1078 int cnt, ret = NO_QUOTA;
1079 char warntype[MAXQUOTAS];
1081 /* First test before acquiring mutex - solves deadlocks when we
1082 * re-enter the quota code and are already holding the mutex */
1083 if (IS_NOQUOTA(inode))
1084 return QUOTA_OK;
1085 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1086 warntype[cnt] = NOWARN;
1087 down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1088 if (IS_NOQUOTA(inode)) {
1089 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1090 return QUOTA_OK;
1092 spin_lock(&dq_data_lock);
1093 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1094 if (inode->i_dquot[cnt] == NODQUOT)
1095 continue;
1096 if (check_idq(inode->i_dquot[cnt], number, warntype+cnt) == NO_QUOTA)
1097 goto warn_put_all;
1100 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1101 if (inode->i_dquot[cnt] == NODQUOT)
1102 continue;
1103 dquot_incr_inodes(inode->i_dquot[cnt], number);
1105 ret = QUOTA_OK;
1106 warn_put_all:
1107 spin_unlock(&dq_data_lock);
1108 if (ret == QUOTA_OK)
1109 /* Dirtify all the dquots - this can block when journalling */
1110 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1111 if (inode->i_dquot[cnt])
1112 mark_dquot_dirty(inode->i_dquot[cnt]);
1113 flush_warnings((struct dquot **)inode->i_dquot, warntype);
1114 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1115 return ret;
1119 * This operation can block, but only after everything is updated
1121 int dquot_free_space(struct inode *inode, qsize_t number)
1123 unsigned int cnt;
1125 /* First test before acquiring mutex - solves deadlocks when we
1126 * re-enter the quota code and are already holding the mutex */
1127 if (IS_NOQUOTA(inode)) {
1128 out_sub:
1129 inode_sub_bytes(inode, number);
1130 return QUOTA_OK;
1132 down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1133 /* Now recheck reliably when holding dqptr_sem */
1134 if (IS_NOQUOTA(inode)) {
1135 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1136 goto out_sub;
1138 spin_lock(&dq_data_lock);
1139 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1140 if (inode->i_dquot[cnt] == NODQUOT)
1141 continue;
1142 dquot_decr_space(inode->i_dquot[cnt], number);
1144 inode_sub_bytes(inode, number);
1145 spin_unlock(&dq_data_lock);
1146 /* Dirtify all the dquots - this can block when journalling */
1147 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1148 if (inode->i_dquot[cnt])
1149 mark_dquot_dirty(inode->i_dquot[cnt]);
1150 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1151 return QUOTA_OK;
1155 * This operation can block, but only after everything is updated
1157 int dquot_free_inode(const struct inode *inode, unsigned long number)
1159 unsigned int cnt;
1161 /* First test before acquiring mutex - solves deadlocks when we
1162 * re-enter the quota code and are already holding the mutex */
1163 if (IS_NOQUOTA(inode))
1164 return QUOTA_OK;
1165 down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1166 /* Now recheck reliably when holding dqptr_sem */
1167 if (IS_NOQUOTA(inode)) {
1168 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1169 return QUOTA_OK;
1171 spin_lock(&dq_data_lock);
1172 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1173 if (inode->i_dquot[cnt] == NODQUOT)
1174 continue;
1175 dquot_decr_inodes(inode->i_dquot[cnt], number);
1177 spin_unlock(&dq_data_lock);
1178 /* Dirtify all the dquots - this can block when journalling */
1179 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1180 if (inode->i_dquot[cnt])
1181 mark_dquot_dirty(inode->i_dquot[cnt]);
1182 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1183 return QUOTA_OK;
1187 * Transfer the number of inode and blocks from one diskquota to an other.
1189 * This operation can block, but only after everything is updated
1190 * A transaction must be started when entering this function.
1192 int dquot_transfer(struct inode *inode, struct iattr *iattr)
1194 qsize_t space;
1195 struct dquot *transfer_from[MAXQUOTAS];
1196 struct dquot *transfer_to[MAXQUOTAS];
1197 int cnt, ret = NO_QUOTA, chuid = (iattr->ia_valid & ATTR_UID) && inode->i_uid != iattr->ia_uid,
1198 chgid = (iattr->ia_valid & ATTR_GID) && inode->i_gid != iattr->ia_gid;
1199 char warntype[MAXQUOTAS];
1201 /* First test before acquiring mutex - solves deadlocks when we
1202 * re-enter the quota code and are already holding the mutex */
1203 if (IS_NOQUOTA(inode))
1204 return QUOTA_OK;
1205 /* Clear the arrays */
1206 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1207 transfer_to[cnt] = transfer_from[cnt] = NODQUOT;
1208 warntype[cnt] = NOWARN;
1210 down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1211 /* Now recheck reliably when holding dqptr_sem */
1212 if (IS_NOQUOTA(inode)) { /* File without quota accounting? */
1213 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1214 return QUOTA_OK;
1216 /* First build the transfer_to list - here we can block on
1217 * reading/instantiating of dquots. We know that the transaction for
1218 * us was already started so we don't violate lock ranking here */
1219 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1220 switch (cnt) {
1221 case USRQUOTA:
1222 if (!chuid)
1223 continue;
1224 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_uid, cnt);
1225 break;
1226 case GRPQUOTA:
1227 if (!chgid)
1228 continue;
1229 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_gid, cnt);
1230 break;
1233 spin_lock(&dq_data_lock);
1234 space = inode_get_bytes(inode);
1235 /* Build the transfer_from list and check the limits */
1236 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1237 if (transfer_to[cnt] == NODQUOT)
1238 continue;
1239 transfer_from[cnt] = inode->i_dquot[cnt];
1240 if (check_idq(transfer_to[cnt], 1, warntype+cnt) == NO_QUOTA ||
1241 check_bdq(transfer_to[cnt], space, 0, warntype+cnt) == NO_QUOTA)
1242 goto warn_put_all;
1246 * Finally perform the needed transfer from transfer_from to transfer_to
1248 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1250 * Skip changes for same uid or gid or for turned off quota-type.
1252 if (transfer_to[cnt] == NODQUOT)
1253 continue;
1255 /* Due to IO error we might not have transfer_from[] structure */
1256 if (transfer_from[cnt]) {
1257 dquot_decr_inodes(transfer_from[cnt], 1);
1258 dquot_decr_space(transfer_from[cnt], space);
1261 dquot_incr_inodes(transfer_to[cnt], 1);
1262 dquot_incr_space(transfer_to[cnt], space);
1264 inode->i_dquot[cnt] = transfer_to[cnt];
1266 ret = QUOTA_OK;
1267 warn_put_all:
1268 spin_unlock(&dq_data_lock);
1269 /* Dirtify all the dquots - this can block when journalling */
1270 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1271 if (transfer_from[cnt])
1272 mark_dquot_dirty(transfer_from[cnt]);
1273 if (transfer_to[cnt])
1274 mark_dquot_dirty(transfer_to[cnt]);
1276 flush_warnings(transfer_to, warntype);
1278 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1279 if (ret == QUOTA_OK && transfer_from[cnt] != NODQUOT)
1280 dqput(transfer_from[cnt]);
1281 if (ret == NO_QUOTA && transfer_to[cnt] != NODQUOT)
1282 dqput(transfer_to[cnt]);
1284 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1285 return ret;
1289 * Write info of quota file to disk
1291 int dquot_commit_info(struct super_block *sb, int type)
1293 int ret;
1294 struct quota_info *dqopt = sb_dqopt(sb);
1296 mutex_lock(&dqopt->dqio_mutex);
1297 ret = dqopt->ops[type]->write_file_info(sb, type);
1298 mutex_unlock(&dqopt->dqio_mutex);
1299 return ret;
1303 * Definitions of diskquota operations.
1305 struct dquot_operations dquot_operations = {
1306 .initialize = dquot_initialize,
1307 .drop = dquot_drop,
1308 .alloc_space = dquot_alloc_space,
1309 .alloc_inode = dquot_alloc_inode,
1310 .free_space = dquot_free_space,
1311 .free_inode = dquot_free_inode,
1312 .transfer = dquot_transfer,
1313 .write_dquot = dquot_commit,
1314 .acquire_dquot = dquot_acquire,
1315 .release_dquot = dquot_release,
1316 .mark_dirty = dquot_mark_dquot_dirty,
1317 .write_info = dquot_commit_info
1320 static inline void set_enable_flags(struct quota_info *dqopt, int type)
1322 switch (type) {
1323 case USRQUOTA:
1324 dqopt->flags |= DQUOT_USR_ENABLED;
1325 break;
1326 case GRPQUOTA:
1327 dqopt->flags |= DQUOT_GRP_ENABLED;
1328 break;
1332 static inline void reset_enable_flags(struct quota_info *dqopt, int type)
1334 switch (type) {
1335 case USRQUOTA:
1336 dqopt->flags &= ~DQUOT_USR_ENABLED;
1337 break;
1338 case GRPQUOTA:
1339 dqopt->flags &= ~DQUOT_GRP_ENABLED;
1340 break;
1345 * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
1347 int vfs_quota_off(struct super_block *sb, int type)
1349 int cnt;
1350 struct quota_info *dqopt = sb_dqopt(sb);
1351 struct inode *toputinode[MAXQUOTAS];
1353 /* We need to serialize quota_off() for device */
1354 mutex_lock(&dqopt->dqonoff_mutex);
1355 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1356 toputinode[cnt] = NULL;
1357 if (type != -1 && cnt != type)
1358 continue;
1359 if (!sb_has_quota_enabled(sb, cnt))
1360 continue;
1361 reset_enable_flags(dqopt, cnt);
1363 /* Note: these are blocking operations */
1364 drop_dquot_ref(sb, cnt);
1365 invalidate_dquots(sb, cnt);
1367 * Now all dquots should be invalidated, all writes done so we should be only
1368 * users of the info. No locks needed.
1370 if (info_dirty(&dqopt->info[cnt]))
1371 sb->dq_op->write_info(sb, cnt);
1372 if (dqopt->ops[cnt]->free_file_info)
1373 dqopt->ops[cnt]->free_file_info(sb, cnt);
1374 put_quota_format(dqopt->info[cnt].dqi_format);
1376 toputinode[cnt] = dqopt->files[cnt];
1377 dqopt->files[cnt] = NULL;
1378 dqopt->info[cnt].dqi_flags = 0;
1379 dqopt->info[cnt].dqi_igrace = 0;
1380 dqopt->info[cnt].dqi_bgrace = 0;
1381 dqopt->ops[cnt] = NULL;
1383 mutex_unlock(&dqopt->dqonoff_mutex);
1384 /* Sync the superblock so that buffers with quota data are written to
1385 * disk (and so userspace sees correct data afterwards). */
1386 if (sb->s_op->sync_fs)
1387 sb->s_op->sync_fs(sb, 1);
1388 sync_blockdev(sb->s_bdev);
1389 /* Now the quota files are just ordinary files and we can set the
1390 * inode flags back. Moreover we discard the pagecache so that
1391 * userspace sees the writes we did bypassing the pagecache. We
1392 * must also discard the blockdev buffers so that we see the
1393 * changes done by userspace on the next quotaon() */
1394 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1395 if (toputinode[cnt]) {
1396 mutex_lock(&dqopt->dqonoff_mutex);
1397 /* If quota was reenabled in the meantime, we have
1398 * nothing to do */
1399 if (!sb_has_quota_enabled(sb, cnt)) {
1400 mutex_lock(&toputinode[cnt]->i_mutex);
1401 toputinode[cnt]->i_flags &= ~(S_IMMUTABLE |
1402 S_NOATIME | S_NOQUOTA);
1403 truncate_inode_pages(&toputinode[cnt]->i_data, 0);
1404 mutex_unlock(&toputinode[cnt]->i_mutex);
1405 mark_inode_dirty(toputinode[cnt]);
1406 iput(toputinode[cnt]);
1408 mutex_unlock(&dqopt->dqonoff_mutex);
1410 if (sb->s_bdev)
1411 invalidate_bdev(sb->s_bdev, 0);
1412 return 0;
1416 * Turn quotas on on a device
1419 /* Helper function when we already have the inode */
1420 static int vfs_quota_on_inode(struct inode *inode, int type, int format_id)
1422 struct quota_format_type *fmt = find_quota_format(format_id);
1423 struct super_block *sb = inode->i_sb;
1424 struct quota_info *dqopt = sb_dqopt(sb);
1425 int error;
1426 int oldflags = -1;
1428 if (!fmt)
1429 return -ESRCH;
1430 if (!S_ISREG(inode->i_mode)) {
1431 error = -EACCES;
1432 goto out_fmt;
1434 if (IS_RDONLY(inode)) {
1435 error = -EROFS;
1436 goto out_fmt;
1438 if (!sb->s_op->quota_write || !sb->s_op->quota_read) {
1439 error = -EINVAL;
1440 goto out_fmt;
1443 /* As we bypass the pagecache we must now flush the inode so that
1444 * we see all the changes from userspace... */
1445 write_inode_now(inode, 1);
1446 /* And now flush the block cache so that kernel sees the changes */
1447 invalidate_bdev(sb->s_bdev, 0);
1448 mutex_lock(&inode->i_mutex);
1449 mutex_lock(&dqopt->dqonoff_mutex);
1450 if (sb_has_quota_enabled(sb, type)) {
1451 error = -EBUSY;
1452 goto out_lock;
1454 /* We don't want quota and atime on quota files (deadlocks possible)
1455 * Also nobody should write to the file - we use special IO operations
1456 * which ignore the immutable bit. */
1457 down_write(&dqopt->dqptr_sem);
1458 oldflags = inode->i_flags & (S_NOATIME | S_IMMUTABLE | S_NOQUOTA);
1459 inode->i_flags |= S_NOQUOTA | S_NOATIME | S_IMMUTABLE;
1460 up_write(&dqopt->dqptr_sem);
1461 sb->dq_op->drop(inode);
1463 error = -EIO;
1464 dqopt->files[type] = igrab(inode);
1465 if (!dqopt->files[type])
1466 goto out_lock;
1467 error = -EINVAL;
1468 if (!fmt->qf_ops->check_quota_file(sb, type))
1469 goto out_file_init;
1471 dqopt->ops[type] = fmt->qf_ops;
1472 dqopt->info[type].dqi_format = fmt;
1473 INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
1474 mutex_lock(&dqopt->dqio_mutex);
1475 if ((error = dqopt->ops[type]->read_file_info(sb, type)) < 0) {
1476 mutex_unlock(&dqopt->dqio_mutex);
1477 goto out_file_init;
1479 mutex_unlock(&dqopt->dqio_mutex);
1480 mutex_unlock(&inode->i_mutex);
1481 set_enable_flags(dqopt, type);
1483 add_dquot_ref(sb, type);
1484 mutex_unlock(&dqopt->dqonoff_mutex);
1486 return 0;
1488 out_file_init:
1489 dqopt->files[type] = NULL;
1490 iput(inode);
1491 out_lock:
1492 mutex_unlock(&dqopt->dqonoff_mutex);
1493 if (oldflags != -1) {
1494 down_write(&dqopt->dqptr_sem);
1495 /* Set the flags back (in the case of accidental quotaon()
1496 * on a wrong file we don't want to mess up the flags) */
1497 inode->i_flags &= ~(S_NOATIME | S_NOQUOTA | S_IMMUTABLE);
1498 inode->i_flags |= oldflags;
1499 up_write(&dqopt->dqptr_sem);
1501 mutex_unlock(&inode->i_mutex);
1502 out_fmt:
1503 put_quota_format(fmt);
1505 return error;
1508 /* Actual function called from quotactl() */
1509 int vfs_quota_on(struct super_block *sb, int type, int format_id, char *path)
1511 struct nameidata nd;
1512 int error;
1514 error = path_lookup(path, LOOKUP_FOLLOW, &nd);
1515 if (error < 0)
1516 return error;
1517 error = security_quota_on(nd.dentry);
1518 if (error)
1519 goto out_path;
1520 /* Quota file not on the same filesystem? */
1521 if (nd.mnt->mnt_sb != sb)
1522 error = -EXDEV;
1523 else
1524 error = vfs_quota_on_inode(nd.dentry->d_inode, type, format_id);
1525 out_path:
1526 path_release(&nd);
1527 return error;
1531 * This function is used when filesystem needs to initialize quotas
1532 * during mount time.
1534 int vfs_quota_on_mount(struct super_block *sb, char *qf_name,
1535 int format_id, int type)
1537 struct dentry *dentry;
1538 int error;
1540 dentry = lookup_one_len(qf_name, sb->s_root, strlen(qf_name));
1541 if (IS_ERR(dentry))
1542 return PTR_ERR(dentry);
1544 if (!dentry->d_inode) {
1545 error = -ENOENT;
1546 goto out;
1549 error = security_quota_on(dentry);
1550 if (!error)
1551 error = vfs_quota_on_inode(dentry->d_inode, type, format_id);
1553 out:
1554 dput(dentry);
1555 return error;
1558 /* Generic routine for getting common part of quota structure */
1559 static void do_get_dqblk(struct dquot *dquot, struct if_dqblk *di)
1561 struct mem_dqblk *dm = &dquot->dq_dqb;
1563 spin_lock(&dq_data_lock);
1564 di->dqb_bhardlimit = dm->dqb_bhardlimit;
1565 di->dqb_bsoftlimit = dm->dqb_bsoftlimit;
1566 di->dqb_curspace = dm->dqb_curspace;
1567 di->dqb_ihardlimit = dm->dqb_ihardlimit;
1568 di->dqb_isoftlimit = dm->dqb_isoftlimit;
1569 di->dqb_curinodes = dm->dqb_curinodes;
1570 di->dqb_btime = dm->dqb_btime;
1571 di->dqb_itime = dm->dqb_itime;
1572 di->dqb_valid = QIF_ALL;
1573 spin_unlock(&dq_data_lock);
1576 int vfs_get_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
1578 struct dquot *dquot;
1580 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1581 if (!(dquot = dqget(sb, id, type))) {
1582 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1583 return -ESRCH;
1585 do_get_dqblk(dquot, di);
1586 dqput(dquot);
1587 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1588 return 0;
1591 /* Generic routine for setting common part of quota structure */
1592 static void do_set_dqblk(struct dquot *dquot, struct if_dqblk *di)
1594 struct mem_dqblk *dm = &dquot->dq_dqb;
1595 int check_blim = 0, check_ilim = 0;
1597 spin_lock(&dq_data_lock);
1598 if (di->dqb_valid & QIF_SPACE) {
1599 dm->dqb_curspace = di->dqb_curspace;
1600 check_blim = 1;
1602 if (di->dqb_valid & QIF_BLIMITS) {
1603 dm->dqb_bsoftlimit = di->dqb_bsoftlimit;
1604 dm->dqb_bhardlimit = di->dqb_bhardlimit;
1605 check_blim = 1;
1607 if (di->dqb_valid & QIF_INODES) {
1608 dm->dqb_curinodes = di->dqb_curinodes;
1609 check_ilim = 1;
1611 if (di->dqb_valid & QIF_ILIMITS) {
1612 dm->dqb_isoftlimit = di->dqb_isoftlimit;
1613 dm->dqb_ihardlimit = di->dqb_ihardlimit;
1614 check_ilim = 1;
1616 if (di->dqb_valid & QIF_BTIME)
1617 dm->dqb_btime = di->dqb_btime;
1618 if (di->dqb_valid & QIF_ITIME)
1619 dm->dqb_itime = di->dqb_itime;
1621 if (check_blim) {
1622 if (!dm->dqb_bsoftlimit || toqb(dm->dqb_curspace) < dm->dqb_bsoftlimit) {
1623 dm->dqb_btime = 0;
1624 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1626 else if (!(di->dqb_valid & QIF_BTIME)) /* Set grace only if user hasn't provided his own... */
1627 dm->dqb_btime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_bgrace;
1629 if (check_ilim) {
1630 if (!dm->dqb_isoftlimit || dm->dqb_curinodes < dm->dqb_isoftlimit) {
1631 dm->dqb_itime = 0;
1632 clear_bit(DQ_INODES_B, &dquot->dq_flags);
1634 else if (!(di->dqb_valid & QIF_ITIME)) /* Set grace only if user hasn't provided his own... */
1635 dm->dqb_itime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
1637 if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit || dm->dqb_isoftlimit)
1638 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
1639 else
1640 set_bit(DQ_FAKE_B, &dquot->dq_flags);
1641 spin_unlock(&dq_data_lock);
1642 mark_dquot_dirty(dquot);
1645 int vfs_set_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
1647 struct dquot *dquot;
1649 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1650 if (!(dquot = dqget(sb, id, type))) {
1651 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1652 return -ESRCH;
1654 do_set_dqblk(dquot, di);
1655 dqput(dquot);
1656 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1657 return 0;
1660 /* Generic routine for getting common part of quota file information */
1661 int vfs_get_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
1663 struct mem_dqinfo *mi;
1665 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1666 if (!sb_has_quota_enabled(sb, type)) {
1667 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1668 return -ESRCH;
1670 mi = sb_dqopt(sb)->info + type;
1671 spin_lock(&dq_data_lock);
1672 ii->dqi_bgrace = mi->dqi_bgrace;
1673 ii->dqi_igrace = mi->dqi_igrace;
1674 ii->dqi_flags = mi->dqi_flags & DQF_MASK;
1675 ii->dqi_valid = IIF_ALL;
1676 spin_unlock(&dq_data_lock);
1677 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1678 return 0;
1681 /* Generic routine for setting common part of quota file information */
1682 int vfs_set_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
1684 struct mem_dqinfo *mi;
1686 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1687 if (!sb_has_quota_enabled(sb, type)) {
1688 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1689 return -ESRCH;
1691 mi = sb_dqopt(sb)->info + type;
1692 spin_lock(&dq_data_lock);
1693 if (ii->dqi_valid & IIF_BGRACE)
1694 mi->dqi_bgrace = ii->dqi_bgrace;
1695 if (ii->dqi_valid & IIF_IGRACE)
1696 mi->dqi_igrace = ii->dqi_igrace;
1697 if (ii->dqi_valid & IIF_FLAGS)
1698 mi->dqi_flags = (mi->dqi_flags & ~DQF_MASK) | (ii->dqi_flags & DQF_MASK);
1699 spin_unlock(&dq_data_lock);
1700 mark_info_dirty(sb, type);
1701 /* Force write to disk */
1702 sb->dq_op->write_info(sb, type);
1703 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1704 return 0;
1707 struct quotactl_ops vfs_quotactl_ops = {
1708 .quota_on = vfs_quota_on,
1709 .quota_off = vfs_quota_off,
1710 .quota_sync = vfs_quota_sync,
1711 .get_info = vfs_get_dqinfo,
1712 .set_info = vfs_set_dqinfo,
1713 .get_dqblk = vfs_get_dqblk,
1714 .set_dqblk = vfs_set_dqblk
1717 static ctl_table fs_dqstats_table[] = {
1719 .ctl_name = FS_DQ_LOOKUPS,
1720 .procname = "lookups",
1721 .data = &dqstats.lookups,
1722 .maxlen = sizeof(int),
1723 .mode = 0444,
1724 .proc_handler = &proc_dointvec,
1727 .ctl_name = FS_DQ_DROPS,
1728 .procname = "drops",
1729 .data = &dqstats.drops,
1730 .maxlen = sizeof(int),
1731 .mode = 0444,
1732 .proc_handler = &proc_dointvec,
1735 .ctl_name = FS_DQ_READS,
1736 .procname = "reads",
1737 .data = &dqstats.reads,
1738 .maxlen = sizeof(int),
1739 .mode = 0444,
1740 .proc_handler = &proc_dointvec,
1743 .ctl_name = FS_DQ_WRITES,
1744 .procname = "writes",
1745 .data = &dqstats.writes,
1746 .maxlen = sizeof(int),
1747 .mode = 0444,
1748 .proc_handler = &proc_dointvec,
1751 .ctl_name = FS_DQ_CACHE_HITS,
1752 .procname = "cache_hits",
1753 .data = &dqstats.cache_hits,
1754 .maxlen = sizeof(int),
1755 .mode = 0444,
1756 .proc_handler = &proc_dointvec,
1759 .ctl_name = FS_DQ_ALLOCATED,
1760 .procname = "allocated_dquots",
1761 .data = &dqstats.allocated_dquots,
1762 .maxlen = sizeof(int),
1763 .mode = 0444,
1764 .proc_handler = &proc_dointvec,
1767 .ctl_name = FS_DQ_FREE,
1768 .procname = "free_dquots",
1769 .data = &dqstats.free_dquots,
1770 .maxlen = sizeof(int),
1771 .mode = 0444,
1772 .proc_handler = &proc_dointvec,
1775 .ctl_name = FS_DQ_SYNCS,
1776 .procname = "syncs",
1777 .data = &dqstats.syncs,
1778 .maxlen = sizeof(int),
1779 .mode = 0444,
1780 .proc_handler = &proc_dointvec,
1783 .ctl_name = FS_DQ_WARNINGS,
1784 .procname = "warnings",
1785 .data = &flag_print_warnings,
1786 .maxlen = sizeof(int),
1787 .mode = 0644,
1788 .proc_handler = &proc_dointvec,
1790 { .ctl_name = 0 },
1793 static ctl_table fs_table[] = {
1795 .ctl_name = FS_DQSTATS,
1796 .procname = "quota",
1797 .mode = 0555,
1798 .child = fs_dqstats_table,
1800 { .ctl_name = 0 },
1803 static ctl_table sys_table[] = {
1805 .ctl_name = CTL_FS,
1806 .procname = "fs",
1807 .mode = 0555,
1808 .child = fs_table,
1810 { .ctl_name = 0 },
1813 static int __init dquot_init(void)
1815 int i;
1816 unsigned long nr_hash, order;
1818 printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
1820 register_sysctl_table(sys_table, 0);
1822 dquot_cachep = kmem_cache_create("dquot",
1823 sizeof(struct dquot), sizeof(unsigned long) * 4,
1824 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1825 SLAB_MEM_SPREAD|SLAB_PANIC),
1826 NULL, NULL);
1828 order = 0;
1829 dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
1830 if (!dquot_hash)
1831 panic("Cannot create dquot hash table");
1833 /* Find power-of-two hlist_heads which can fit into allocation */
1834 nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
1835 dq_hash_bits = 0;
1836 do {
1837 dq_hash_bits++;
1838 } while (nr_hash >> dq_hash_bits);
1839 dq_hash_bits--;
1841 nr_hash = 1UL << dq_hash_bits;
1842 dq_hash_mask = nr_hash - 1;
1843 for (i = 0; i < nr_hash; i++)
1844 INIT_HLIST_HEAD(dquot_hash + i);
1846 printk("Dquot-cache hash table entries: %ld (order %ld, %ld bytes)\n",
1847 nr_hash, order, (PAGE_SIZE << order));
1849 set_shrinker(DEFAULT_SEEKS, shrink_dqcache_memory);
1851 return 0;
1853 module_init(dquot_init);
1855 EXPORT_SYMBOL(register_quota_format);
1856 EXPORT_SYMBOL(unregister_quota_format);
1857 EXPORT_SYMBOL(dqstats);
1858 EXPORT_SYMBOL(dq_data_lock);
1859 EXPORT_SYMBOL(vfs_quota_on);
1860 EXPORT_SYMBOL(vfs_quota_on_mount);
1861 EXPORT_SYMBOL(vfs_quota_off);
1862 EXPORT_SYMBOL(vfs_quota_sync);
1863 EXPORT_SYMBOL(vfs_get_dqinfo);
1864 EXPORT_SYMBOL(vfs_set_dqinfo);
1865 EXPORT_SYMBOL(vfs_get_dqblk);
1866 EXPORT_SYMBOL(vfs_set_dqblk);
1867 EXPORT_SYMBOL(dquot_commit);
1868 EXPORT_SYMBOL(dquot_commit_info);
1869 EXPORT_SYMBOL(dquot_acquire);
1870 EXPORT_SYMBOL(dquot_release);
1871 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
1872 EXPORT_SYMBOL(dquot_initialize);
1873 EXPORT_SYMBOL(dquot_drop);
1874 EXPORT_SYMBOL(dquot_alloc_space);
1875 EXPORT_SYMBOL(dquot_alloc_inode);
1876 EXPORT_SYMBOL(dquot_free_space);
1877 EXPORT_SYMBOL(dquot_free_inode);
1878 EXPORT_SYMBOL(dquot_transfer);