Merge branch 'for-2.6.25' of master.kernel.org:/pub/scm/linux/kernel/git/galak/powerp...
[linux-2.6] / fs / dquot.c
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.
11  * 
12  * Version: $Id: dquot.c,v 6.3 1996/11/17 18:35:34 mvw Exp mvw $
13  * 
14  * Author:      Marco van Wieringen <mvw@planets.elm.net>
15  *
16  * Fixes:   Dmitry Gorodchanin <pgmdsg@ibi.com>, 11 Feb 96
17  *
18  *              Revised list management to avoid races
19  *              -- Bill Hawes, <whawes@star.net>, 9/98
20  *
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
33  *
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
41  *
42  *              Added dynamic quota structure allocation
43  *              Jan Kara <jack@suse.cz> 12/2000
44  *
45  *              Rewritten quota interface. Implemented new quota format and
46  *              formats registering.
47  *              Jan Kara, <jack@suse.cz>, 2001,2002
48  *
49  *              New SMP locking.
50  *              Jan Kara, <jack@suse.cz>, 10/2002
51  *
52  *              Added journalled quota support, fix lock inversion problems
53  *              Jan Kara, <jack@suse.cz>, 2003,2004
54  *
55  * (C) Copyright 1994 - 1997 Marco van Wieringen 
56  */
57
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/init.h>
73 #include <linux/module.h>
74 #include <linux/proc_fs.h>
75 #include <linux/security.h>
76 #include <linux/kmod.h>
77 #include <linux/namei.h>
78 #include <linux/buffer_head.h>
79 #include <linux/capability.h>
80 #include <linux/quotaops.h>
81 #include <linux/writeback.h> /* for inode_lock, oddly enough.. */
82 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
83 #include <net/netlink.h>
84 #include <net/genetlink.h>
85 #endif
86
87 #include <asm/uaccess.h>
88
89 #define __DQUOT_PARANOIA
90
91 /*
92  * There are two quota SMP locks. dq_list_lock protects all lists with quotas
93  * and quota formats and also dqstats structure containing statistics about the
94  * lists. dq_data_lock protects data from dq_dqb and also mem_dqinfo structures
95  * and also guards consistency of dquot->dq_dqb with inode->i_blocks, i_bytes.
96  * i_blocks and i_bytes updates itself are guarded by i_lock acquired directly
97  * in inode_add_bytes() and inode_sub_bytes().
98  *
99  * The spinlock ordering is hence: dq_data_lock > dq_list_lock > i_lock
100  *
101  * Note that some things (eg. sb pointer, type, id) doesn't change during
102  * the life of the dquot structure and so needn't to be protected by a lock
103  *
104  * Any operation working on dquots via inode pointers must hold dqptr_sem.  If
105  * operation is just reading pointers from inode (or not using them at all) the
106  * read lock is enough. If pointers are altered function must hold write lock
107  * (these locking rules also apply for S_NOQUOTA flag in the inode - note that
108  * for altering the flag i_mutex is also needed).  If operation is holding
109  * reference to dquot in other way (e.g. quotactl ops) it must be guarded by
110  * dqonoff_mutex.
111  * This locking assures that:
112  *   a) update/access to dquot pointers in inode is serialized
113  *   b) everyone is guarded against invalidate_dquots()
114  *
115  * Each dquot has its dq_lock mutex. Locked dquots might not be referenced
116  * from inodes (dquot_alloc_space() and such don't check the dq_lock).
117  * Currently dquot is locked only when it is being read to memory (or space for
118  * it is being allocated) on the first dqget() and when it is being released on
119  * the last dqput(). The allocation and release oparations are serialized by
120  * the dq_lock and by checking the use count in dquot_release().  Write
121  * operations on dquots don't hold dq_lock as they copy data under dq_data_lock
122  * spinlock to internal buffers before writing.
123  *
124  * Lock ordering (including related VFS locks) is the following:
125  *   i_mutex > dqonoff_sem > journal_lock > dqptr_sem > dquot->dq_lock >
126  *   dqio_mutex
127  * i_mutex on quota files is special (it's below dqio_mutex)
128  */
129
130 static DEFINE_SPINLOCK(dq_list_lock);
131 DEFINE_SPINLOCK(dq_data_lock);
132
133 static char *quotatypes[] = INITQFNAMES;
134 static struct quota_format_type *quota_formats; /* List of registered formats */
135 static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
136
137 /* SLAB cache for dquot structures */
138 static struct kmem_cache *dquot_cachep;
139
140 int register_quota_format(struct quota_format_type *fmt)
141 {
142         spin_lock(&dq_list_lock);
143         fmt->qf_next = quota_formats;
144         quota_formats = fmt;
145         spin_unlock(&dq_list_lock);
146         return 0;
147 }
148
149 void unregister_quota_format(struct quota_format_type *fmt)
150 {
151         struct quota_format_type **actqf;
152
153         spin_lock(&dq_list_lock);
154         for (actqf = &quota_formats; *actqf && *actqf != fmt; actqf = &(*actqf)->qf_next);
155         if (*actqf)
156                 *actqf = (*actqf)->qf_next;
157         spin_unlock(&dq_list_lock);
158 }
159
160 static struct quota_format_type *find_quota_format(int id)
161 {
162         struct quota_format_type *actqf;
163
164         spin_lock(&dq_list_lock);
165         for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id; actqf = actqf->qf_next);
166         if (!actqf || !try_module_get(actqf->qf_owner)) {
167                 int qm;
168
169                 spin_unlock(&dq_list_lock);
170                 
171                 for (qm = 0; module_names[qm].qm_fmt_id && module_names[qm].qm_fmt_id != id; qm++);
172                 if (!module_names[qm].qm_fmt_id || request_module(module_names[qm].qm_mod_name))
173                         return NULL;
174
175                 spin_lock(&dq_list_lock);
176                 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id; actqf = actqf->qf_next);
177                 if (actqf && !try_module_get(actqf->qf_owner))
178                         actqf = NULL;
179         }
180         spin_unlock(&dq_list_lock);
181         return actqf;
182 }
183
184 static void put_quota_format(struct quota_format_type *fmt)
185 {
186         module_put(fmt->qf_owner);
187 }
188
189 /*
190  * Dquot List Management:
191  * The quota code uses three lists for dquot management: the inuse_list,
192  * free_dquots, and dquot_hash[] array. A single dquot structure may be
193  * on all three lists, depending on its current state.
194  *
195  * All dquots are placed to the end of inuse_list when first created, and this
196  * list is used for invalidate operation, which must look at every dquot.
197  *
198  * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
199  * and this list is searched whenever we need an available dquot.  Dquots are
200  * removed from the list as soon as they are used again, and
201  * dqstats.free_dquots gives the number of dquots on the list. When
202  * dquot is invalidated it's completely released from memory.
203  *
204  * Dquots with a specific identity (device, type and id) are placed on
205  * one of the dquot_hash[] hash chains. The provides an efficient search
206  * mechanism to locate a specific dquot.
207  */
208
209 static LIST_HEAD(inuse_list);
210 static LIST_HEAD(free_dquots);
211 static unsigned int dq_hash_bits, dq_hash_mask;
212 static struct hlist_head *dquot_hash;
213
214 struct dqstats dqstats;
215
216 static void dqput(struct dquot *dquot);
217
218 static inline unsigned int
219 hashfn(const struct super_block *sb, unsigned int id, int type)
220 {
221         unsigned long tmp;
222
223         tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
224         return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
225 }
226
227 /*
228  * Following list functions expect dq_list_lock to be held
229  */
230 static inline void insert_dquot_hash(struct dquot *dquot)
231 {
232         struct hlist_head *head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id, dquot->dq_type);
233         hlist_add_head(&dquot->dq_hash, head);
234 }
235
236 static inline void remove_dquot_hash(struct dquot *dquot)
237 {
238         hlist_del_init(&dquot->dq_hash);
239 }
240
241 static inline struct dquot *find_dquot(unsigned int hashent, struct super_block *sb, unsigned int id, int type)
242 {
243         struct hlist_node *node;
244         struct dquot *dquot;
245
246         hlist_for_each (node, dquot_hash+hashent) {
247                 dquot = hlist_entry(node, struct dquot, dq_hash);
248                 if (dquot->dq_sb == sb && dquot->dq_id == id && dquot->dq_type == type)
249                         return dquot;
250         }
251         return NODQUOT;
252 }
253
254 /* Add a dquot to the tail of the free list */
255 static inline void put_dquot_last(struct dquot *dquot)
256 {
257         list_add_tail(&dquot->dq_free, &free_dquots);
258         dqstats.free_dquots++;
259 }
260
261 static inline void remove_free_dquot(struct dquot *dquot)
262 {
263         if (list_empty(&dquot->dq_free))
264                 return;
265         list_del_init(&dquot->dq_free);
266         dqstats.free_dquots--;
267 }
268
269 static inline void put_inuse(struct dquot *dquot)
270 {
271         /* We add to the back of inuse list so we don't have to restart
272          * when traversing this list and we block */
273         list_add_tail(&dquot->dq_inuse, &inuse_list);
274         dqstats.allocated_dquots++;
275 }
276
277 static inline void remove_inuse(struct dquot *dquot)
278 {
279         dqstats.allocated_dquots--;
280         list_del(&dquot->dq_inuse);
281 }
282 /*
283  * End of list functions needing dq_list_lock
284  */
285
286 static void wait_on_dquot(struct dquot *dquot)
287 {
288         mutex_lock(&dquot->dq_lock);
289         mutex_unlock(&dquot->dq_lock);
290 }
291
292 #define mark_dquot_dirty(dquot) ((dquot)->dq_sb->dq_op->mark_dirty(dquot))
293
294 int dquot_mark_dquot_dirty(struct dquot *dquot)
295 {
296         spin_lock(&dq_list_lock);
297         if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags))
298                 list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
299                                 info[dquot->dq_type].dqi_dirty_list);
300         spin_unlock(&dq_list_lock);
301         return 0;
302 }
303
304 /* This function needs dq_list_lock */
305 static inline int clear_dquot_dirty(struct dquot *dquot)
306 {
307         if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags))
308                 return 0;
309         list_del_init(&dquot->dq_dirty);
310         return 1;
311 }
312
313 void mark_info_dirty(struct super_block *sb, int type)
314 {
315         set_bit(DQF_INFO_DIRTY_B, &sb_dqopt(sb)->info[type].dqi_flags);
316 }
317 EXPORT_SYMBOL(mark_info_dirty);
318
319 /*
320  *      Read dquot from disk and alloc space for it
321  */
322
323 int dquot_acquire(struct dquot *dquot)
324 {
325         int ret = 0, ret2 = 0;
326         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
327
328         mutex_lock(&dquot->dq_lock);
329         mutex_lock(&dqopt->dqio_mutex);
330         if (!test_bit(DQ_READ_B, &dquot->dq_flags))
331                 ret = dqopt->ops[dquot->dq_type]->read_dqblk(dquot);
332         if (ret < 0)
333                 goto out_iolock;
334         set_bit(DQ_READ_B, &dquot->dq_flags);
335         /* Instantiate dquot if needed */
336         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
337                 ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
338                 /* Write the info if needed */
339                 if (info_dirty(&dqopt->info[dquot->dq_type]))
340                         ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
341                 if (ret < 0)
342                         goto out_iolock;
343                 if (ret2 < 0) {
344                         ret = ret2;
345                         goto out_iolock;
346                 }
347         }
348         set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
349 out_iolock:
350         mutex_unlock(&dqopt->dqio_mutex);
351         mutex_unlock(&dquot->dq_lock);
352         return ret;
353 }
354
355 /*
356  *      Write dquot to disk
357  */
358 int dquot_commit(struct dquot *dquot)
359 {
360         int ret = 0, ret2 = 0;
361         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
362
363         mutex_lock(&dqopt->dqio_mutex);
364         spin_lock(&dq_list_lock);
365         if (!clear_dquot_dirty(dquot)) {
366                 spin_unlock(&dq_list_lock);
367                 goto out_sem;
368         }
369         spin_unlock(&dq_list_lock);
370         /* Inactive dquot can be only if there was error during read/init
371          * => we have better not writing it */
372         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
373                 ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
374                 if (info_dirty(&dqopt->info[dquot->dq_type]))
375                         ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
376                 if (ret >= 0)
377                         ret = ret2;
378         }
379 out_sem:
380         mutex_unlock(&dqopt->dqio_mutex);
381         return ret;
382 }
383
384 /*
385  *      Release dquot
386  */
387 int dquot_release(struct dquot *dquot)
388 {
389         int ret = 0, ret2 = 0;
390         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
391
392         mutex_lock(&dquot->dq_lock);
393         /* Check whether we are not racing with some other dqget() */
394         if (atomic_read(&dquot->dq_count) > 1)
395                 goto out_dqlock;
396         mutex_lock(&dqopt->dqio_mutex);
397         if (dqopt->ops[dquot->dq_type]->release_dqblk) {
398                 ret = dqopt->ops[dquot->dq_type]->release_dqblk(dquot);
399                 /* Write the info */
400                 if (info_dirty(&dqopt->info[dquot->dq_type]))
401                         ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
402                 if (ret >= 0)
403                         ret = ret2;
404         }
405         clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
406         mutex_unlock(&dqopt->dqio_mutex);
407 out_dqlock:
408         mutex_unlock(&dquot->dq_lock);
409         return ret;
410 }
411
412 /* Invalidate all dquots on the list. Note that this function is called after
413  * quota is disabled and pointers from inodes removed so there cannot be new
414  * quota users. There can still be some users of quotas due to inodes being
415  * just deleted or pruned by prune_icache() (those are not attached to any
416  * list). We have to wait for such users.
417  */
418 static void invalidate_dquots(struct super_block *sb, int type)
419 {
420         struct dquot *dquot, *tmp;
421
422 restart:
423         spin_lock(&dq_list_lock);
424         list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
425                 if (dquot->dq_sb != sb)
426                         continue;
427                 if (dquot->dq_type != type)
428                         continue;
429                 /* Wait for dquot users */
430                 if (atomic_read(&dquot->dq_count)) {
431                         DEFINE_WAIT(wait);
432
433                         atomic_inc(&dquot->dq_count);
434                         prepare_to_wait(&dquot->dq_wait_unused, &wait,
435                                         TASK_UNINTERRUPTIBLE);
436                         spin_unlock(&dq_list_lock);
437                         /* Once dqput() wakes us up, we know it's time to free
438                          * the dquot.
439                          * IMPORTANT: we rely on the fact that there is always
440                          * at most one process waiting for dquot to free.
441                          * Otherwise dq_count would be > 1 and we would never
442                          * wake up.
443                          */
444                         if (atomic_read(&dquot->dq_count) > 1)
445                                 schedule();
446                         finish_wait(&dquot->dq_wait_unused, &wait);
447                         dqput(dquot);
448                         /* At this moment dquot() need not exist (it could be
449                          * reclaimed by prune_dqcache(). Hence we must
450                          * restart. */
451                         goto restart;
452                 }
453                 /*
454                  * Quota now has no users and it has been written on last
455                  * dqput()
456                  */
457                 remove_dquot_hash(dquot);
458                 remove_free_dquot(dquot);
459                 remove_inuse(dquot);
460                 kmem_cache_free(dquot_cachep, dquot);
461         }
462         spin_unlock(&dq_list_lock);
463 }
464
465 int vfs_quota_sync(struct super_block *sb, int type)
466 {
467         struct list_head *dirty;
468         struct dquot *dquot;
469         struct quota_info *dqopt = sb_dqopt(sb);
470         int cnt;
471
472         mutex_lock(&dqopt->dqonoff_mutex);
473         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
474                 if (type != -1 && cnt != type)
475                         continue;
476                 if (!sb_has_quota_enabled(sb, cnt))
477                         continue;
478                 spin_lock(&dq_list_lock);
479                 dirty = &dqopt->info[cnt].dqi_dirty_list;
480                 while (!list_empty(dirty)) {
481                         dquot = list_first_entry(dirty, struct dquot, dq_dirty);
482                         /* Dirty and inactive can be only bad dquot... */
483                         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
484                                 clear_dquot_dirty(dquot);
485                                 continue;
486                         }
487                         /* Now we have active dquot from which someone is
488                          * holding reference so we can safely just increase
489                          * use count */
490                         atomic_inc(&dquot->dq_count);
491                         dqstats.lookups++;
492                         spin_unlock(&dq_list_lock);
493                         sb->dq_op->write_dquot(dquot);
494                         dqput(dquot);
495                         spin_lock(&dq_list_lock);
496                 }
497                 spin_unlock(&dq_list_lock);
498         }
499
500         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
501                 if ((cnt == type || type == -1) && sb_has_quota_enabled(sb, cnt)
502                         && info_dirty(&dqopt->info[cnt]))
503                         sb->dq_op->write_info(sb, cnt);
504         spin_lock(&dq_list_lock);
505         dqstats.syncs++;
506         spin_unlock(&dq_list_lock);
507         mutex_unlock(&dqopt->dqonoff_mutex);
508
509         return 0;
510 }
511
512 /* Free unused dquots from cache */
513 static void prune_dqcache(int count)
514 {
515         struct list_head *head;
516         struct dquot *dquot;
517
518         head = free_dquots.prev;
519         while (head != &free_dquots && count) {
520                 dquot = list_entry(head, struct dquot, dq_free);
521                 remove_dquot_hash(dquot);
522                 remove_free_dquot(dquot);
523                 remove_inuse(dquot);
524                 kmem_cache_free(dquot_cachep, dquot);
525                 count--;
526                 head = free_dquots.prev;
527         }
528 }
529
530 /*
531  * This is called from kswapd when we think we need some
532  * more memory
533  */
534
535 static int shrink_dqcache_memory(int nr, gfp_t gfp_mask)
536 {
537         if (nr) {
538                 spin_lock(&dq_list_lock);
539                 prune_dqcache(nr);
540                 spin_unlock(&dq_list_lock);
541         }
542         return (dqstats.free_dquots / 100) * sysctl_vfs_cache_pressure;
543 }
544
545 static struct shrinker dqcache_shrinker = {
546         .shrink = shrink_dqcache_memory,
547         .seeks = DEFAULT_SEEKS,
548 };
549
550 /*
551  * Put reference to dquot
552  * NOTE: If you change this function please check whether dqput_blocks() works right...
553  * MUST be called with either dqptr_sem or dqonoff_mutex held
554  */
555 static void dqput(struct dquot *dquot)
556 {
557         if (!dquot)
558                 return;
559 #ifdef __DQUOT_PARANOIA
560         if (!atomic_read(&dquot->dq_count)) {
561                 printk("VFS: dqput: trying to free free dquot\n");
562                 printk("VFS: device %s, dquot of %s %d\n",
563                         dquot->dq_sb->s_id,
564                         quotatypes[dquot->dq_type],
565                         dquot->dq_id);
566                 BUG();
567         }
568 #endif
569         
570         spin_lock(&dq_list_lock);
571         dqstats.drops++;
572         spin_unlock(&dq_list_lock);
573 we_slept:
574         spin_lock(&dq_list_lock);
575         if (atomic_read(&dquot->dq_count) > 1) {
576                 /* We have more than one user... nothing to do */
577                 atomic_dec(&dquot->dq_count);
578                 /* Releasing dquot during quotaoff phase? */
579                 if (!sb_has_quota_enabled(dquot->dq_sb, dquot->dq_type) &&
580                     atomic_read(&dquot->dq_count) == 1)
581                         wake_up(&dquot->dq_wait_unused);
582                 spin_unlock(&dq_list_lock);
583                 return;
584         }
585         /* Need to release dquot? */
586         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && dquot_dirty(dquot)) {
587                 spin_unlock(&dq_list_lock);
588                 /* Commit dquot before releasing */
589                 dquot->dq_sb->dq_op->write_dquot(dquot);
590                 goto we_slept;
591         }
592         /* Clear flag in case dquot was inactive (something bad happened) */
593         clear_dquot_dirty(dquot);
594         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
595                 spin_unlock(&dq_list_lock);
596                 dquot->dq_sb->dq_op->release_dquot(dquot);
597                 goto we_slept;
598         }
599         atomic_dec(&dquot->dq_count);
600 #ifdef __DQUOT_PARANOIA
601         /* sanity check */
602         BUG_ON(!list_empty(&dquot->dq_free));
603 #endif
604         put_dquot_last(dquot);
605         spin_unlock(&dq_list_lock);
606 }
607
608 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
609 {
610         struct dquot *dquot;
611
612         dquot = kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
613         if(!dquot)
614                 return NODQUOT;
615
616         mutex_init(&dquot->dq_lock);
617         INIT_LIST_HEAD(&dquot->dq_free);
618         INIT_LIST_HEAD(&dquot->dq_inuse);
619         INIT_HLIST_NODE(&dquot->dq_hash);
620         INIT_LIST_HEAD(&dquot->dq_dirty);
621         init_waitqueue_head(&dquot->dq_wait_unused);
622         dquot->dq_sb = sb;
623         dquot->dq_type = type;
624         atomic_set(&dquot->dq_count, 1);
625
626         return dquot;
627 }
628
629 /*
630  * Get reference to dquot
631  * MUST be called with either dqptr_sem or dqonoff_mutex held
632  */
633 static struct dquot *dqget(struct super_block *sb, unsigned int id, int type)
634 {
635         unsigned int hashent = hashfn(sb, id, type);
636         struct dquot *dquot, *empty = NODQUOT;
637
638         if (!sb_has_quota_enabled(sb, type))
639                 return NODQUOT;
640 we_slept:
641         spin_lock(&dq_list_lock);
642         if ((dquot = find_dquot(hashent, sb, id, type)) == NODQUOT) {
643                 if (empty == NODQUOT) {
644                         spin_unlock(&dq_list_lock);
645                         if ((empty = get_empty_dquot(sb, type)) == NODQUOT)
646                                 schedule();     /* Try to wait for a moment... */
647                         goto we_slept;
648                 }
649                 dquot = empty;
650                 dquot->dq_id = id;
651                 /* all dquots go on the inuse_list */
652                 put_inuse(dquot);
653                 /* hash it first so it can be found */
654                 insert_dquot_hash(dquot);
655                 dqstats.lookups++;
656                 spin_unlock(&dq_list_lock);
657         } else {
658                 if (!atomic_read(&dquot->dq_count))
659                         remove_free_dquot(dquot);
660                 atomic_inc(&dquot->dq_count);
661                 dqstats.cache_hits++;
662                 dqstats.lookups++;
663                 spin_unlock(&dq_list_lock);
664                 if (empty)
665                         kmem_cache_free(dquot_cachep, empty);
666         }
667         /* Wait for dq_lock - after this we know that either dquot_release() is already
668          * finished or it will be canceled due to dq_count > 1 test */
669         wait_on_dquot(dquot);
670         /* Read the dquot and instantiate it (everything done only if needed) */
671         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && sb->dq_op->acquire_dquot(dquot) < 0) {
672                 dqput(dquot);
673                 return NODQUOT;
674         }
675 #ifdef __DQUOT_PARANOIA
676         BUG_ON(!dquot->dq_sb);  /* Has somebody invalidated entry under us? */
677 #endif
678
679         return dquot;
680 }
681
682 static int dqinit_needed(struct inode *inode, int type)
683 {
684         int cnt;
685
686         if (IS_NOQUOTA(inode))
687                 return 0;
688         if (type != -1)
689                 return inode->i_dquot[type] == NODQUOT;
690         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
691                 if (inode->i_dquot[cnt] == NODQUOT)
692                         return 1;
693         return 0;
694 }
695
696 /* This routine is guarded by dqonoff_mutex mutex */
697 static void add_dquot_ref(struct super_block *sb, int type)
698 {
699         struct inode *inode;
700
701 restart:
702         spin_lock(&inode_lock);
703         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
704                 if (!atomic_read(&inode->i_writecount))
705                         continue;
706                 if (!dqinit_needed(inode, type))
707                         continue;
708                 if (inode->i_state & (I_FREEING|I_WILL_FREE))
709                         continue;
710
711                 __iget(inode);
712                 spin_unlock(&inode_lock);
713
714                 sb->dq_op->initialize(inode, type);
715                 iput(inode);
716                 /* As we may have blocked we had better restart... */
717                 goto restart;
718         }
719         spin_unlock(&inode_lock);
720 }
721
722 /* Return 0 if dqput() won't block (note that 1 doesn't necessarily mean blocking) */
723 static inline int dqput_blocks(struct dquot *dquot)
724 {
725         if (atomic_read(&dquot->dq_count) <= 1)
726                 return 1;
727         return 0;
728 }
729
730 /* Remove references to dquots from inode - add dquot to list for freeing if needed */
731 /* We can't race with anybody because we hold dqptr_sem for writing... */
732 static int remove_inode_dquot_ref(struct inode *inode, int type,
733                                   struct list_head *tofree_head)
734 {
735         struct dquot *dquot = inode->i_dquot[type];
736
737         inode->i_dquot[type] = NODQUOT;
738         if (dquot != NODQUOT) {
739                 if (dqput_blocks(dquot)) {
740 #ifdef __DQUOT_PARANOIA
741                         if (atomic_read(&dquot->dq_count) != 1)
742                                 printk(KERN_WARNING "VFS: Adding dquot with dq_count %d to dispose list.\n", atomic_read(&dquot->dq_count));
743 #endif
744                         spin_lock(&dq_list_lock);
745                         list_add(&dquot->dq_free, tofree_head); /* As dquot must have currently users it can't be on the free list... */
746                         spin_unlock(&dq_list_lock);
747                         return 1;
748                 }
749                 else
750                         dqput(dquot);   /* We have guaranteed we won't block */
751         }
752         return 0;
753 }
754
755 /* Free list of dquots - called from inode.c */
756 /* dquots are removed from inodes, no new references can be got so we are the only ones holding reference */
757 static void put_dquot_list(struct list_head *tofree_head)
758 {
759         struct list_head *act_head;
760         struct dquot *dquot;
761
762         act_head = tofree_head->next;
763         /* So now we have dquots on the list... Just free them */
764         while (act_head != tofree_head) {
765                 dquot = list_entry(act_head, struct dquot, dq_free);
766                 act_head = act_head->next;
767                 list_del_init(&dquot->dq_free); /* Remove dquot from the list so we won't have problems... */
768                 dqput(dquot);
769         }
770 }
771
772 static void remove_dquot_ref(struct super_block *sb, int type,
773                 struct list_head *tofree_head)
774 {
775         struct inode *inode;
776
777         spin_lock(&inode_lock);
778         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
779                 if (!IS_NOQUOTA(inode))
780                         remove_inode_dquot_ref(inode, type, tofree_head);
781         }
782         spin_unlock(&inode_lock);
783 }
784
785 /* Gather all references from inodes and drop them */
786 static void drop_dquot_ref(struct super_block *sb, int type)
787 {
788         LIST_HEAD(tofree_head);
789
790         if (sb->dq_op) {
791                 down_write(&sb_dqopt(sb)->dqptr_sem);
792                 remove_dquot_ref(sb, type, &tofree_head);
793                 up_write(&sb_dqopt(sb)->dqptr_sem);
794                 put_dquot_list(&tofree_head);
795         }
796 }
797
798 static inline void dquot_incr_inodes(struct dquot *dquot, unsigned long number)
799 {
800         dquot->dq_dqb.dqb_curinodes += number;
801 }
802
803 static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
804 {
805         dquot->dq_dqb.dqb_curspace += number;
806 }
807
808 static inline void dquot_decr_inodes(struct dquot *dquot, unsigned long number)
809 {
810         if (dquot->dq_dqb.dqb_curinodes > number)
811                 dquot->dq_dqb.dqb_curinodes -= number;
812         else
813                 dquot->dq_dqb.dqb_curinodes = 0;
814         if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
815                 dquot->dq_dqb.dqb_itime = (time_t) 0;
816         clear_bit(DQ_INODES_B, &dquot->dq_flags);
817 }
818
819 static inline void dquot_decr_space(struct dquot *dquot, qsize_t number)
820 {
821         if (dquot->dq_dqb.dqb_curspace > number)
822                 dquot->dq_dqb.dqb_curspace -= number;
823         else
824                 dquot->dq_dqb.dqb_curspace = 0;
825         if (toqb(dquot->dq_dqb.dqb_curspace) <= dquot->dq_dqb.dqb_bsoftlimit)
826                 dquot->dq_dqb.dqb_btime = (time_t) 0;
827         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
828 }
829
830 static int warning_issued(struct dquot *dquot, const int warntype)
831 {
832         int flag = (warntype == QUOTA_NL_BHARDWARN ||
833                 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
834                 ((warntype == QUOTA_NL_IHARDWARN ||
835                 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
836
837         if (!flag)
838                 return 0;
839         return test_and_set_bit(flag, &dquot->dq_flags);
840 }
841
842 #ifdef CONFIG_PRINT_QUOTA_WARNING
843 static int flag_print_warnings = 1;
844
845 static inline int need_print_warning(struct dquot *dquot)
846 {
847         if (!flag_print_warnings)
848                 return 0;
849
850         switch (dquot->dq_type) {
851                 case USRQUOTA:
852                         return current->fsuid == dquot->dq_id;
853                 case GRPQUOTA:
854                         return in_group_p(dquot->dq_id);
855         }
856         return 0;
857 }
858
859 /* Print warning to user which exceeded quota */
860 static void print_warning(struct dquot *dquot, const int warntype)
861 {
862         char *msg = NULL;
863         struct tty_struct *tty;
864
865         if (!need_print_warning(dquot))
866                 return;
867
868         mutex_lock(&tty_mutex);
869         tty = get_current_tty();
870         if (!tty)
871                 goto out_lock;
872         tty_write_message(tty, dquot->dq_sb->s_id);
873         if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
874                 tty_write_message(tty, ": warning, ");
875         else
876                 tty_write_message(tty, ": write failed, ");
877         tty_write_message(tty, quotatypes[dquot->dq_type]);
878         switch (warntype) {
879                 case QUOTA_NL_IHARDWARN:
880                         msg = " file limit reached.\r\n";
881                         break;
882                 case QUOTA_NL_ISOFTLONGWARN:
883                         msg = " file quota exceeded too long.\r\n";
884                         break;
885                 case QUOTA_NL_ISOFTWARN:
886                         msg = " file quota exceeded.\r\n";
887                         break;
888                 case QUOTA_NL_BHARDWARN:
889                         msg = " block limit reached.\r\n";
890                         break;
891                 case QUOTA_NL_BSOFTLONGWARN:
892                         msg = " block quota exceeded too long.\r\n";
893                         break;
894                 case QUOTA_NL_BSOFTWARN:
895                         msg = " block quota exceeded.\r\n";
896                         break;
897         }
898         tty_write_message(tty, msg);
899 out_lock:
900         mutex_unlock(&tty_mutex);
901 }
902 #endif
903
904 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
905
906 /* Netlink family structure for quota */
907 static struct genl_family quota_genl_family = {
908         .id = GENL_ID_GENERATE,
909         .hdrsize = 0,
910         .name = "VFS_DQUOT",
911         .version = 1,
912         .maxattr = QUOTA_NL_A_MAX,
913 };
914
915 /* Send warning to userspace about user which exceeded quota */
916 static void send_warning(const struct dquot *dquot, const char warntype)
917 {
918         static atomic_t seq;
919         struct sk_buff *skb;
920         void *msg_head;
921         int ret;
922         int msg_size = 4 * nla_total_size(sizeof(u32)) +
923                        2 * nla_total_size(sizeof(u64));
924
925         /* We have to allocate using GFP_NOFS as we are called from a
926          * filesystem performing write and thus further recursion into
927          * the fs to free some data could cause deadlocks. */
928         skb = genlmsg_new(msg_size, GFP_NOFS);
929         if (!skb) {
930                 printk(KERN_ERR
931                   "VFS: Not enough memory to send quota warning.\n");
932                 return;
933         }
934         msg_head = genlmsg_put(skb, 0, atomic_add_return(1, &seq),
935                         &quota_genl_family, 0, QUOTA_NL_C_WARNING);
936         if (!msg_head) {
937                 printk(KERN_ERR
938                   "VFS: Cannot store netlink header in quota warning.\n");
939                 goto err_out;
940         }
941         ret = nla_put_u32(skb, QUOTA_NL_A_QTYPE, dquot->dq_type);
942         if (ret)
943                 goto attr_err_out;
944         ret = nla_put_u64(skb, QUOTA_NL_A_EXCESS_ID, dquot->dq_id);
945         if (ret)
946                 goto attr_err_out;
947         ret = nla_put_u32(skb, QUOTA_NL_A_WARNING, warntype);
948         if (ret)
949                 goto attr_err_out;
950         ret = nla_put_u32(skb, QUOTA_NL_A_DEV_MAJOR,
951                 MAJOR(dquot->dq_sb->s_dev));
952         if (ret)
953                 goto attr_err_out;
954         ret = nla_put_u32(skb, QUOTA_NL_A_DEV_MINOR,
955                 MINOR(dquot->dq_sb->s_dev));
956         if (ret)
957                 goto attr_err_out;
958         ret = nla_put_u64(skb, QUOTA_NL_A_CAUSED_ID, current->user->uid);
959         if (ret)
960                 goto attr_err_out;
961         genlmsg_end(skb, msg_head);
962
963         ret = genlmsg_multicast(skb, 0, quota_genl_family.id, GFP_NOFS);
964         if (ret < 0 && ret != -ESRCH)
965                 printk(KERN_ERR
966                         "VFS: Failed to send notification message: %d\n", ret);
967         return;
968 attr_err_out:
969         printk(KERN_ERR "VFS: Not enough space to compose quota message!\n");
970 err_out:
971         kfree_skb(skb);
972 }
973 #endif
974
975 static inline void flush_warnings(struct dquot * const *dquots, char *warntype)
976 {
977         int i;
978
979         for (i = 0; i < MAXQUOTAS; i++)
980                 if (dquots[i] != NODQUOT && warntype[i] != QUOTA_NL_NOWARN &&
981                     !warning_issued(dquots[i], warntype[i])) {
982 #ifdef CONFIG_PRINT_QUOTA_WARNING
983                         print_warning(dquots[i], warntype[i]);
984 #endif
985 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
986                         send_warning(dquots[i], warntype[i]);
987 #endif
988                 }
989 }
990
991 static inline char ignore_hardlimit(struct dquot *dquot)
992 {
993         struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_type];
994
995         return capable(CAP_SYS_RESOURCE) &&
996             (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD || !(info->dqi_flags & V1_DQF_RSQUASH));
997 }
998
999 /* needs dq_data_lock */
1000 static int check_idq(struct dquot *dquot, ulong inodes, char *warntype)
1001 {
1002         *warntype = QUOTA_NL_NOWARN;
1003         if (inodes <= 0 || test_bit(DQ_FAKE_B, &dquot->dq_flags))
1004                 return QUOTA_OK;
1005
1006         if (dquot->dq_dqb.dqb_ihardlimit &&
1007            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_ihardlimit &&
1008             !ignore_hardlimit(dquot)) {
1009                 *warntype = QUOTA_NL_IHARDWARN;
1010                 return NO_QUOTA;
1011         }
1012
1013         if (dquot->dq_dqb.dqb_isoftlimit &&
1014            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
1015             dquot->dq_dqb.dqb_itime && get_seconds() >= dquot->dq_dqb.dqb_itime &&
1016             !ignore_hardlimit(dquot)) {
1017                 *warntype = QUOTA_NL_ISOFTLONGWARN;
1018                 return NO_QUOTA;
1019         }
1020
1021         if (dquot->dq_dqb.dqb_isoftlimit &&
1022            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
1023             dquot->dq_dqb.dqb_itime == 0) {
1024                 *warntype = QUOTA_NL_ISOFTWARN;
1025                 dquot->dq_dqb.dqb_itime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
1026         }
1027
1028         return QUOTA_OK;
1029 }
1030
1031 /* needs dq_data_lock */
1032 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc, char *warntype)
1033 {
1034         *warntype = QUOTA_NL_NOWARN;
1035         if (space <= 0 || test_bit(DQ_FAKE_B, &dquot->dq_flags))
1036                 return QUOTA_OK;
1037
1038         if (dquot->dq_dqb.dqb_bhardlimit &&
1039            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bhardlimit &&
1040             !ignore_hardlimit(dquot)) {
1041                 if (!prealloc)
1042                         *warntype = QUOTA_NL_BHARDWARN;
1043                 return NO_QUOTA;
1044         }
1045
1046         if (dquot->dq_dqb.dqb_bsoftlimit &&
1047            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bsoftlimit &&
1048             dquot->dq_dqb.dqb_btime && get_seconds() >= dquot->dq_dqb.dqb_btime &&
1049             !ignore_hardlimit(dquot)) {
1050                 if (!prealloc)
1051                         *warntype = QUOTA_NL_BSOFTLONGWARN;
1052                 return NO_QUOTA;
1053         }
1054
1055         if (dquot->dq_dqb.dqb_bsoftlimit &&
1056            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bsoftlimit &&
1057             dquot->dq_dqb.dqb_btime == 0) {
1058                 if (!prealloc) {
1059                         *warntype = QUOTA_NL_BSOFTWARN;
1060                         dquot->dq_dqb.dqb_btime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_bgrace;
1061                 }
1062                 else
1063                         /*
1064                          * We don't allow preallocation to exceed softlimit so exceeding will
1065                          * be always printed
1066                          */
1067                         return NO_QUOTA;
1068         }
1069
1070         return QUOTA_OK;
1071 }
1072
1073 /*
1074  *      Initialize quota pointers in inode
1075  *      Transaction must be started at entry
1076  */
1077 int dquot_initialize(struct inode *inode, int type)
1078 {
1079         unsigned int id = 0;
1080         int cnt, ret = 0;
1081
1082         /* First test before acquiring mutex - solves deadlocks when we
1083          * re-enter the quota code and are already holding the mutex */
1084         if (IS_NOQUOTA(inode))
1085                 return 0;
1086         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1087         /* Having dqptr_sem we know NOQUOTA flags can't be altered... */
1088         if (IS_NOQUOTA(inode))
1089                 goto out_err;
1090         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1091                 if (type != -1 && cnt != type)
1092                         continue;
1093                 if (inode->i_dquot[cnt] == NODQUOT) {
1094                         switch (cnt) {
1095                                 case USRQUOTA:
1096                                         id = inode->i_uid;
1097                                         break;
1098                                 case GRPQUOTA:
1099                                         id = inode->i_gid;
1100                                         break;
1101                         }
1102                         inode->i_dquot[cnt] = dqget(inode->i_sb, id, cnt);
1103                 }
1104         }
1105 out_err:
1106         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1107         return ret;
1108 }
1109
1110 /*
1111  *      Release all quotas referenced by inode
1112  *      Transaction must be started at an entry
1113  */
1114 int dquot_drop(struct inode *inode)
1115 {
1116         int cnt;
1117
1118         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1119         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1120                 if (inode->i_dquot[cnt] != NODQUOT) {
1121                         dqput(inode->i_dquot[cnt]);
1122                         inode->i_dquot[cnt] = NODQUOT;
1123                 }
1124         }
1125         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1126         return 0;
1127 }
1128
1129 /*
1130  * Following four functions update i_blocks+i_bytes fields and
1131  * quota information (together with appropriate checks)
1132  * NOTE: We absolutely rely on the fact that caller dirties
1133  * the inode (usually macros in quotaops.h care about this) and
1134  * holds a handle for the current transaction so that dquot write and
1135  * inode write go into the same transaction.
1136  */
1137
1138 /*
1139  * This operation can block, but only after everything is updated
1140  */
1141 int dquot_alloc_space(struct inode *inode, qsize_t number, int warn)
1142 {
1143         int cnt, ret = NO_QUOTA;
1144         char warntype[MAXQUOTAS];
1145
1146         /* First test before acquiring mutex - solves deadlocks when we
1147          * re-enter the quota code and are already holding the mutex */
1148         if (IS_NOQUOTA(inode)) {
1149 out_add:
1150                 inode_add_bytes(inode, number);
1151                 return QUOTA_OK;
1152         }
1153         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1154                 warntype[cnt] = QUOTA_NL_NOWARN;
1155
1156         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1157         if (IS_NOQUOTA(inode)) {        /* Now we can do reliable test... */
1158                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1159                 goto out_add;
1160         }
1161         spin_lock(&dq_data_lock);
1162         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1163                 if (inode->i_dquot[cnt] == NODQUOT)
1164                         continue;
1165                 if (check_bdq(inode->i_dquot[cnt], number, warn, warntype+cnt) == NO_QUOTA)
1166                         goto warn_put_all;
1167         }
1168         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1169                 if (inode->i_dquot[cnt] == NODQUOT)
1170                         continue;
1171                 dquot_incr_space(inode->i_dquot[cnt], number);
1172         }
1173         inode_add_bytes(inode, number);
1174         ret = QUOTA_OK;
1175 warn_put_all:
1176         spin_unlock(&dq_data_lock);
1177         if (ret == QUOTA_OK)
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         flush_warnings(inode->i_dquot, warntype);
1183         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1184         return ret;
1185 }
1186
1187 /*
1188  * This operation can block, but only after everything is updated
1189  */
1190 int dquot_alloc_inode(const struct inode *inode, unsigned long number)
1191 {
1192         int cnt, ret = NO_QUOTA;
1193         char warntype[MAXQUOTAS];
1194
1195         /* First test before acquiring mutex - solves deadlocks when we
1196          * re-enter the quota code and are already holding the mutex */
1197         if (IS_NOQUOTA(inode))
1198                 return QUOTA_OK;
1199         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1200                 warntype[cnt] = QUOTA_NL_NOWARN;
1201         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1202         if (IS_NOQUOTA(inode)) {
1203                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1204                 return QUOTA_OK;
1205         }
1206         spin_lock(&dq_data_lock);
1207         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1208                 if (inode->i_dquot[cnt] == NODQUOT)
1209                         continue;
1210                 if (check_idq(inode->i_dquot[cnt], number, warntype+cnt) == NO_QUOTA)
1211                         goto warn_put_all;
1212         }
1213
1214         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1215                 if (inode->i_dquot[cnt] == NODQUOT)
1216                         continue;
1217                 dquot_incr_inodes(inode->i_dquot[cnt], number);
1218         }
1219         ret = QUOTA_OK;
1220 warn_put_all:
1221         spin_unlock(&dq_data_lock);
1222         if (ret == QUOTA_OK)
1223                 /* Dirtify all the dquots - this can block when journalling */
1224                 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1225                         if (inode->i_dquot[cnt])
1226                                 mark_dquot_dirty(inode->i_dquot[cnt]);
1227         flush_warnings(inode->i_dquot, warntype);
1228         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1229         return ret;
1230 }
1231
1232 /*
1233  * This operation can block, but only after everything is updated
1234  */
1235 int dquot_free_space(struct inode *inode, qsize_t number)
1236 {
1237         unsigned int cnt;
1238
1239         /* First test before acquiring mutex - solves deadlocks when we
1240          * re-enter the quota code and are already holding the mutex */
1241         if (IS_NOQUOTA(inode)) {
1242 out_sub:
1243                 inode_sub_bytes(inode, number);
1244                 return QUOTA_OK;
1245         }
1246         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1247         /* Now recheck reliably when holding dqptr_sem */
1248         if (IS_NOQUOTA(inode)) {
1249                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1250                 goto out_sub;
1251         }
1252         spin_lock(&dq_data_lock);
1253         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1254                 if (inode->i_dquot[cnt] == NODQUOT)
1255                         continue;
1256                 dquot_decr_space(inode->i_dquot[cnt], number);
1257         }
1258         inode_sub_bytes(inode, number);
1259         spin_unlock(&dq_data_lock);
1260         /* Dirtify all the dquots - this can block when journalling */
1261         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1262                 if (inode->i_dquot[cnt])
1263                         mark_dquot_dirty(inode->i_dquot[cnt]);
1264         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1265         return QUOTA_OK;
1266 }
1267
1268 /*
1269  * This operation can block, but only after everything is updated
1270  */
1271 int dquot_free_inode(const struct inode *inode, unsigned long number)
1272 {
1273         unsigned int cnt;
1274
1275         /* First test before acquiring mutex - solves deadlocks when we
1276          * re-enter the quota code and are already holding the mutex */
1277         if (IS_NOQUOTA(inode))
1278                 return QUOTA_OK;
1279         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1280         /* Now recheck reliably when holding dqptr_sem */
1281         if (IS_NOQUOTA(inode)) {
1282                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1283                 return QUOTA_OK;
1284         }
1285         spin_lock(&dq_data_lock);
1286         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1287                 if (inode->i_dquot[cnt] == NODQUOT)
1288                         continue;
1289                 dquot_decr_inodes(inode->i_dquot[cnt], number);
1290         }
1291         spin_unlock(&dq_data_lock);
1292         /* Dirtify all the dquots - this can block when journalling */
1293         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1294                 if (inode->i_dquot[cnt])
1295                         mark_dquot_dirty(inode->i_dquot[cnt]);
1296         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1297         return QUOTA_OK;
1298 }
1299
1300 /*
1301  * Transfer the number of inode and blocks from one diskquota to an other.
1302  *
1303  * This operation can block, but only after everything is updated
1304  * A transaction must be started when entering this function.
1305  */
1306 int dquot_transfer(struct inode *inode, struct iattr *iattr)
1307 {
1308         qsize_t space;
1309         struct dquot *transfer_from[MAXQUOTAS];
1310         struct dquot *transfer_to[MAXQUOTAS];
1311         int cnt, ret = NO_QUOTA, chuid = (iattr->ia_valid & ATTR_UID) && inode->i_uid != iattr->ia_uid,
1312             chgid = (iattr->ia_valid & ATTR_GID) && inode->i_gid != iattr->ia_gid;
1313         char warntype[MAXQUOTAS];
1314
1315         /* First test before acquiring mutex - solves deadlocks when we
1316          * re-enter the quota code and are already holding the mutex */
1317         if (IS_NOQUOTA(inode))
1318                 return QUOTA_OK;
1319         /* Clear the arrays */
1320         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1321                 transfer_to[cnt] = transfer_from[cnt] = NODQUOT;
1322                 warntype[cnt] = QUOTA_NL_NOWARN;
1323         }
1324         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1325         /* Now recheck reliably when holding dqptr_sem */
1326         if (IS_NOQUOTA(inode)) {        /* File without quota accounting? */
1327                 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1328                 return QUOTA_OK;
1329         }
1330         /* First build the transfer_to list - here we can block on
1331          * reading/instantiating of dquots.  We know that the transaction for
1332          * us was already started so we don't violate lock ranking here */
1333         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1334                 switch (cnt) {
1335                         case USRQUOTA:
1336                                 if (!chuid)
1337                                         continue;
1338                                 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_uid, cnt);
1339                                 break;
1340                         case GRPQUOTA:
1341                                 if (!chgid)
1342                                         continue;
1343                                 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_gid, cnt);
1344                                 break;
1345                 }
1346         }
1347         spin_lock(&dq_data_lock);
1348         space = inode_get_bytes(inode);
1349         /* Build the transfer_from list and check the limits */
1350         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1351                 if (transfer_to[cnt] == NODQUOT)
1352                         continue;
1353                 transfer_from[cnt] = inode->i_dquot[cnt];
1354                 if (check_idq(transfer_to[cnt], 1, warntype+cnt) == NO_QUOTA ||
1355                     check_bdq(transfer_to[cnt], space, 0, warntype+cnt) == NO_QUOTA)
1356                         goto warn_put_all;
1357         }
1358
1359         /*
1360          * Finally perform the needed transfer from transfer_from to transfer_to
1361          */
1362         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1363                 /*
1364                  * Skip changes for same uid or gid or for turned off quota-type.
1365                  */
1366                 if (transfer_to[cnt] == NODQUOT)
1367                         continue;
1368
1369                 /* Due to IO error we might not have transfer_from[] structure */
1370                 if (transfer_from[cnt]) {
1371                         dquot_decr_inodes(transfer_from[cnt], 1);
1372                         dquot_decr_space(transfer_from[cnt], space);
1373                 }
1374
1375                 dquot_incr_inodes(transfer_to[cnt], 1);
1376                 dquot_incr_space(transfer_to[cnt], space);
1377
1378                 inode->i_dquot[cnt] = transfer_to[cnt];
1379         }
1380         ret = QUOTA_OK;
1381 warn_put_all:
1382         spin_unlock(&dq_data_lock);
1383         /* Dirtify all the dquots - this can block when journalling */
1384         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1385                 if (transfer_from[cnt])
1386                         mark_dquot_dirty(transfer_from[cnt]);
1387                 if (transfer_to[cnt])
1388                         mark_dquot_dirty(transfer_to[cnt]);
1389         }
1390         flush_warnings(transfer_to, warntype);
1391         
1392         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1393                 if (ret == QUOTA_OK && transfer_from[cnt] != NODQUOT)
1394                         dqput(transfer_from[cnt]);
1395                 if (ret == NO_QUOTA && transfer_to[cnt] != NODQUOT)
1396                         dqput(transfer_to[cnt]);
1397         }
1398         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1399         return ret;
1400 }
1401
1402 /*
1403  * Write info of quota file to disk
1404  */
1405 int dquot_commit_info(struct super_block *sb, int type)
1406 {
1407         int ret;
1408         struct quota_info *dqopt = sb_dqopt(sb);
1409
1410         mutex_lock(&dqopt->dqio_mutex);
1411         ret = dqopt->ops[type]->write_file_info(sb, type);
1412         mutex_unlock(&dqopt->dqio_mutex);
1413         return ret;
1414 }
1415
1416 /*
1417  * Definitions of diskquota operations.
1418  */
1419 struct dquot_operations dquot_operations = {
1420         .initialize     = dquot_initialize,
1421         .drop           = dquot_drop,
1422         .alloc_space    = dquot_alloc_space,
1423         .alloc_inode    = dquot_alloc_inode,
1424         .free_space     = dquot_free_space,
1425         .free_inode     = dquot_free_inode,
1426         .transfer       = dquot_transfer,
1427         .write_dquot    = dquot_commit,
1428         .acquire_dquot  = dquot_acquire,
1429         .release_dquot  = dquot_release,
1430         .mark_dirty     = dquot_mark_dquot_dirty,
1431         .write_info     = dquot_commit_info
1432 };
1433
1434 static inline void set_enable_flags(struct quota_info *dqopt, int type)
1435 {
1436         switch (type) {
1437                 case USRQUOTA:
1438                         dqopt->flags |= DQUOT_USR_ENABLED;
1439                         break;
1440                 case GRPQUOTA:
1441                         dqopt->flags |= DQUOT_GRP_ENABLED;
1442                         break;
1443         }
1444 }
1445
1446 static inline void reset_enable_flags(struct quota_info *dqopt, int type)
1447 {
1448         switch (type) {
1449                 case USRQUOTA:
1450                         dqopt->flags &= ~DQUOT_USR_ENABLED;
1451                         break;
1452                 case GRPQUOTA:
1453                         dqopt->flags &= ~DQUOT_GRP_ENABLED;
1454                         break;
1455         }
1456 }
1457
1458 /*
1459  * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
1460  */
1461 int vfs_quota_off(struct super_block *sb, int type)
1462 {
1463         int cnt;
1464         struct quota_info *dqopt = sb_dqopt(sb);
1465         struct inode *toputinode[MAXQUOTAS];
1466
1467         /* We need to serialize quota_off() for device */
1468         mutex_lock(&dqopt->dqonoff_mutex);
1469         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1470                 toputinode[cnt] = NULL;
1471                 if (type != -1 && cnt != type)
1472                         continue;
1473                 if (!sb_has_quota_enabled(sb, cnt))
1474                         continue;
1475                 reset_enable_flags(dqopt, cnt);
1476
1477                 /* Note: these are blocking operations */
1478                 drop_dquot_ref(sb, cnt);
1479                 invalidate_dquots(sb, cnt);
1480                 /*
1481                  * Now all dquots should be invalidated, all writes done so we should be only
1482                  * users of the info. No locks needed.
1483                  */
1484                 if (info_dirty(&dqopt->info[cnt]))
1485                         sb->dq_op->write_info(sb, cnt);
1486                 if (dqopt->ops[cnt]->free_file_info)
1487                         dqopt->ops[cnt]->free_file_info(sb, cnt);
1488                 put_quota_format(dqopt->info[cnt].dqi_format);
1489
1490                 toputinode[cnt] = dqopt->files[cnt];
1491                 dqopt->files[cnt] = NULL;
1492                 dqopt->info[cnt].dqi_flags = 0;
1493                 dqopt->info[cnt].dqi_igrace = 0;
1494                 dqopt->info[cnt].dqi_bgrace = 0;
1495                 dqopt->ops[cnt] = NULL;
1496         }
1497         mutex_unlock(&dqopt->dqonoff_mutex);
1498         /* Sync the superblock so that buffers with quota data are written to
1499          * disk (and so userspace sees correct data afterwards). */
1500         if (sb->s_op->sync_fs)
1501                 sb->s_op->sync_fs(sb, 1);
1502         sync_blockdev(sb->s_bdev);
1503         /* Now the quota files are just ordinary files and we can set the
1504          * inode flags back. Moreover we discard the pagecache so that
1505          * userspace sees the writes we did bypassing the pagecache. We
1506          * must also discard the blockdev buffers so that we see the
1507          * changes done by userspace on the next quotaon() */
1508         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1509                 if (toputinode[cnt]) {
1510                         mutex_lock(&dqopt->dqonoff_mutex);
1511                         /* If quota was reenabled in the meantime, we have
1512                          * nothing to do */
1513                         if (!sb_has_quota_enabled(sb, cnt)) {
1514                                 mutex_lock_nested(&toputinode[cnt]->i_mutex, I_MUTEX_QUOTA);
1515                                 toputinode[cnt]->i_flags &= ~(S_IMMUTABLE |
1516                                   S_NOATIME | S_NOQUOTA);
1517                                 truncate_inode_pages(&toputinode[cnt]->i_data, 0);
1518                                 mutex_unlock(&toputinode[cnt]->i_mutex);
1519                                 mark_inode_dirty(toputinode[cnt]);
1520                                 iput(toputinode[cnt]);
1521                         }
1522                         mutex_unlock(&dqopt->dqonoff_mutex);
1523                 }
1524         if (sb->s_bdev)
1525                 invalidate_bdev(sb->s_bdev);
1526         return 0;
1527 }
1528
1529 /*
1530  *      Turn quotas on on a device
1531  */
1532
1533 /* Helper function when we already have the inode */
1534 static int vfs_quota_on_inode(struct inode *inode, int type, int format_id)
1535 {
1536         struct quota_format_type *fmt = find_quota_format(format_id);
1537         struct super_block *sb = inode->i_sb;
1538         struct quota_info *dqopt = sb_dqopt(sb);
1539         int error;
1540         int oldflags = -1;
1541
1542         if (!fmt)
1543                 return -ESRCH;
1544         if (!S_ISREG(inode->i_mode)) {
1545                 error = -EACCES;
1546                 goto out_fmt;
1547         }
1548         if (IS_RDONLY(inode)) {
1549                 error = -EROFS;
1550                 goto out_fmt;
1551         }
1552         if (!sb->s_op->quota_write || !sb->s_op->quota_read) {
1553                 error = -EINVAL;
1554                 goto out_fmt;
1555         }
1556
1557         /* As we bypass the pagecache we must now flush the inode so that
1558          * we see all the changes from userspace... */
1559         write_inode_now(inode, 1);
1560         /* And now flush the block cache so that kernel sees the changes */
1561         invalidate_bdev(sb->s_bdev);
1562         mutex_lock(&inode->i_mutex);
1563         mutex_lock(&dqopt->dqonoff_mutex);
1564         if (sb_has_quota_enabled(sb, type)) {
1565                 error = -EBUSY;
1566                 goto out_lock;
1567         }
1568         /* We don't want quota and atime on quota files (deadlocks possible)
1569          * Also nobody should write to the file - we use special IO operations
1570          * which ignore the immutable bit. */
1571         down_write(&dqopt->dqptr_sem);
1572         oldflags = inode->i_flags & (S_NOATIME | S_IMMUTABLE | S_NOQUOTA);
1573         inode->i_flags |= S_NOQUOTA | S_NOATIME | S_IMMUTABLE;
1574         up_write(&dqopt->dqptr_sem);
1575         sb->dq_op->drop(inode);
1576
1577         error = -EIO;
1578         dqopt->files[type] = igrab(inode);
1579         if (!dqopt->files[type])
1580                 goto out_lock;
1581         error = -EINVAL;
1582         if (!fmt->qf_ops->check_quota_file(sb, type))
1583                 goto out_file_init;
1584
1585         dqopt->ops[type] = fmt->qf_ops;
1586         dqopt->info[type].dqi_format = fmt;
1587         INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
1588         mutex_lock(&dqopt->dqio_mutex);
1589         if ((error = dqopt->ops[type]->read_file_info(sb, type)) < 0) {
1590                 mutex_unlock(&dqopt->dqio_mutex);
1591                 goto out_file_init;
1592         }
1593         mutex_unlock(&dqopt->dqio_mutex);
1594         mutex_unlock(&inode->i_mutex);
1595         set_enable_flags(dqopt, type);
1596
1597         add_dquot_ref(sb, type);
1598         mutex_unlock(&dqopt->dqonoff_mutex);
1599
1600         return 0;
1601
1602 out_file_init:
1603         dqopt->files[type] = NULL;
1604         iput(inode);
1605 out_lock:
1606         mutex_unlock(&dqopt->dqonoff_mutex);
1607         if (oldflags != -1) {
1608                 down_write(&dqopt->dqptr_sem);
1609                 /* Set the flags back (in the case of accidental quotaon()
1610                  * on a wrong file we don't want to mess up the flags) */
1611                 inode->i_flags &= ~(S_NOATIME | S_NOQUOTA | S_IMMUTABLE);
1612                 inode->i_flags |= oldflags;
1613                 up_write(&dqopt->dqptr_sem);
1614         }
1615         mutex_unlock(&inode->i_mutex);
1616 out_fmt:
1617         put_quota_format(fmt);
1618
1619         return error; 
1620 }
1621
1622 /* Actual function called from quotactl() */
1623 int vfs_quota_on(struct super_block *sb, int type, int format_id, char *path)
1624 {
1625         struct nameidata nd;
1626         int error;
1627
1628         error = path_lookup(path, LOOKUP_FOLLOW, &nd);
1629         if (error < 0)
1630                 return error;
1631         error = security_quota_on(nd.dentry);
1632         if (error)
1633                 goto out_path;
1634         /* Quota file not on the same filesystem? */
1635         if (nd.mnt->mnt_sb != sb)
1636                 error = -EXDEV;
1637         else
1638                 error = vfs_quota_on_inode(nd.dentry->d_inode, type, format_id);
1639 out_path:
1640         path_release(&nd);
1641         return error;
1642 }
1643
1644 /*
1645  * This function is used when filesystem needs to initialize quotas
1646  * during mount time.
1647  */
1648 int vfs_quota_on_mount(struct super_block *sb, char *qf_name,
1649                 int format_id, int type)
1650 {
1651         struct dentry *dentry;
1652         int error;
1653
1654         dentry = lookup_one_len(qf_name, sb->s_root, strlen(qf_name));
1655         if (IS_ERR(dentry))
1656                 return PTR_ERR(dentry);
1657
1658         if (!dentry->d_inode) {
1659                 error = -ENOENT;
1660                 goto out;
1661         }
1662
1663         error = security_quota_on(dentry);
1664         if (!error)
1665                 error = vfs_quota_on_inode(dentry->d_inode, type, format_id);
1666
1667 out:
1668         dput(dentry);
1669         return error;
1670 }
1671
1672 /* Generic routine for getting common part of quota structure */
1673 static void do_get_dqblk(struct dquot *dquot, struct if_dqblk *di)
1674 {
1675         struct mem_dqblk *dm = &dquot->dq_dqb;
1676
1677         spin_lock(&dq_data_lock);
1678         di->dqb_bhardlimit = dm->dqb_bhardlimit;
1679         di->dqb_bsoftlimit = dm->dqb_bsoftlimit;
1680         di->dqb_curspace = dm->dqb_curspace;
1681         di->dqb_ihardlimit = dm->dqb_ihardlimit;
1682         di->dqb_isoftlimit = dm->dqb_isoftlimit;
1683         di->dqb_curinodes = dm->dqb_curinodes;
1684         di->dqb_btime = dm->dqb_btime;
1685         di->dqb_itime = dm->dqb_itime;
1686         di->dqb_valid = QIF_ALL;
1687         spin_unlock(&dq_data_lock);
1688 }
1689
1690 int vfs_get_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
1691 {
1692         struct dquot *dquot;
1693
1694         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1695         if (!(dquot = dqget(sb, id, type))) {
1696                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1697                 return -ESRCH;
1698         }
1699         do_get_dqblk(dquot, di);
1700         dqput(dquot);
1701         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1702         return 0;
1703 }
1704
1705 /* Generic routine for setting common part of quota structure */
1706 static void do_set_dqblk(struct dquot *dquot, struct if_dqblk *di)
1707 {
1708         struct mem_dqblk *dm = &dquot->dq_dqb;
1709         int check_blim = 0, check_ilim = 0;
1710
1711         spin_lock(&dq_data_lock);
1712         if (di->dqb_valid & QIF_SPACE) {
1713                 dm->dqb_curspace = di->dqb_curspace;
1714                 check_blim = 1;
1715         }
1716         if (di->dqb_valid & QIF_BLIMITS) {
1717                 dm->dqb_bsoftlimit = di->dqb_bsoftlimit;
1718                 dm->dqb_bhardlimit = di->dqb_bhardlimit;
1719                 check_blim = 1;
1720         }
1721         if (di->dqb_valid & QIF_INODES) {
1722                 dm->dqb_curinodes = di->dqb_curinodes;
1723                 check_ilim = 1;
1724         }
1725         if (di->dqb_valid & QIF_ILIMITS) {
1726                 dm->dqb_isoftlimit = di->dqb_isoftlimit;
1727                 dm->dqb_ihardlimit = di->dqb_ihardlimit;
1728                 check_ilim = 1;
1729         }
1730         if (di->dqb_valid & QIF_BTIME)
1731                 dm->dqb_btime = di->dqb_btime;
1732         if (di->dqb_valid & QIF_ITIME)
1733                 dm->dqb_itime = di->dqb_itime;
1734
1735         if (check_blim) {
1736                 if (!dm->dqb_bsoftlimit || toqb(dm->dqb_curspace) < dm->dqb_bsoftlimit) {
1737                         dm->dqb_btime = 0;
1738                         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1739                 }
1740                 else if (!(di->dqb_valid & QIF_BTIME))  /* Set grace only if user hasn't provided his own... */
1741                         dm->dqb_btime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_bgrace;
1742         }
1743         if (check_ilim) {
1744                 if (!dm->dqb_isoftlimit || dm->dqb_curinodes < dm->dqb_isoftlimit) {
1745                         dm->dqb_itime = 0;
1746                         clear_bit(DQ_INODES_B, &dquot->dq_flags);
1747                 }
1748                 else if (!(di->dqb_valid & QIF_ITIME))  /* Set grace only if user hasn't provided his own... */
1749                         dm->dqb_itime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
1750         }
1751         if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit || dm->dqb_isoftlimit)
1752                 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
1753         else
1754                 set_bit(DQ_FAKE_B, &dquot->dq_flags);
1755         spin_unlock(&dq_data_lock);
1756         mark_dquot_dirty(dquot);
1757 }
1758
1759 int vfs_set_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
1760 {
1761         struct dquot *dquot;
1762
1763         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1764         if (!(dquot = dqget(sb, id, type))) {
1765                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1766                 return -ESRCH;
1767         }
1768         do_set_dqblk(dquot, di);
1769         dqput(dquot);
1770         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1771         return 0;
1772 }
1773
1774 /* Generic routine for getting common part of quota file information */
1775 int vfs_get_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
1776 {
1777         struct mem_dqinfo *mi;
1778   
1779         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1780         if (!sb_has_quota_enabled(sb, type)) {
1781                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1782                 return -ESRCH;
1783         }
1784         mi = sb_dqopt(sb)->info + type;
1785         spin_lock(&dq_data_lock);
1786         ii->dqi_bgrace = mi->dqi_bgrace;
1787         ii->dqi_igrace = mi->dqi_igrace;
1788         ii->dqi_flags = mi->dqi_flags & DQF_MASK;
1789         ii->dqi_valid = IIF_ALL;
1790         spin_unlock(&dq_data_lock);
1791         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1792         return 0;
1793 }
1794
1795 /* Generic routine for setting common part of quota file information */
1796 int vfs_set_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
1797 {
1798         struct mem_dqinfo *mi;
1799
1800         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1801         if (!sb_has_quota_enabled(sb, type)) {
1802                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1803                 return -ESRCH;
1804         }
1805         mi = sb_dqopt(sb)->info + type;
1806         spin_lock(&dq_data_lock);
1807         if (ii->dqi_valid & IIF_BGRACE)
1808                 mi->dqi_bgrace = ii->dqi_bgrace;
1809         if (ii->dqi_valid & IIF_IGRACE)
1810                 mi->dqi_igrace = ii->dqi_igrace;
1811         if (ii->dqi_valid & IIF_FLAGS)
1812                 mi->dqi_flags = (mi->dqi_flags & ~DQF_MASK) | (ii->dqi_flags & DQF_MASK);
1813         spin_unlock(&dq_data_lock);
1814         mark_info_dirty(sb, type);
1815         /* Force write to disk */
1816         sb->dq_op->write_info(sb, type);
1817         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1818         return 0;
1819 }
1820
1821 struct quotactl_ops vfs_quotactl_ops = {
1822         .quota_on       = vfs_quota_on,
1823         .quota_off      = vfs_quota_off,
1824         .quota_sync     = vfs_quota_sync,
1825         .get_info       = vfs_get_dqinfo,
1826         .set_info       = vfs_set_dqinfo,
1827         .get_dqblk      = vfs_get_dqblk,
1828         .set_dqblk      = vfs_set_dqblk
1829 };
1830
1831 static ctl_table fs_dqstats_table[] = {
1832         {
1833                 .ctl_name       = FS_DQ_LOOKUPS,
1834                 .procname       = "lookups",
1835                 .data           = &dqstats.lookups,
1836                 .maxlen         = sizeof(int),
1837                 .mode           = 0444,
1838                 .proc_handler   = &proc_dointvec,
1839         },
1840         {
1841                 .ctl_name       = FS_DQ_DROPS,
1842                 .procname       = "drops",
1843                 .data           = &dqstats.drops,
1844                 .maxlen         = sizeof(int),
1845                 .mode           = 0444,
1846                 .proc_handler   = &proc_dointvec,
1847         },
1848         {
1849                 .ctl_name       = FS_DQ_READS,
1850                 .procname       = "reads",
1851                 .data           = &dqstats.reads,
1852                 .maxlen         = sizeof(int),
1853                 .mode           = 0444,
1854                 .proc_handler   = &proc_dointvec,
1855         },
1856         {
1857                 .ctl_name       = FS_DQ_WRITES,
1858                 .procname       = "writes",
1859                 .data           = &dqstats.writes,
1860                 .maxlen         = sizeof(int),
1861                 .mode           = 0444,
1862                 .proc_handler   = &proc_dointvec,
1863         },
1864         {
1865                 .ctl_name       = FS_DQ_CACHE_HITS,
1866                 .procname       = "cache_hits",
1867                 .data           = &dqstats.cache_hits,
1868                 .maxlen         = sizeof(int),
1869                 .mode           = 0444,
1870                 .proc_handler   = &proc_dointvec,
1871         },
1872         {
1873                 .ctl_name       = FS_DQ_ALLOCATED,
1874                 .procname       = "allocated_dquots",
1875                 .data           = &dqstats.allocated_dquots,
1876                 .maxlen         = sizeof(int),
1877                 .mode           = 0444,
1878                 .proc_handler   = &proc_dointvec,
1879         },
1880         {
1881                 .ctl_name       = FS_DQ_FREE,
1882                 .procname       = "free_dquots",
1883                 .data           = &dqstats.free_dquots,
1884                 .maxlen         = sizeof(int),
1885                 .mode           = 0444,
1886                 .proc_handler   = &proc_dointvec,
1887         },
1888         {
1889                 .ctl_name       = FS_DQ_SYNCS,
1890                 .procname       = "syncs",
1891                 .data           = &dqstats.syncs,
1892                 .maxlen         = sizeof(int),
1893                 .mode           = 0444,
1894                 .proc_handler   = &proc_dointvec,
1895         },
1896 #ifdef CONFIG_PRINT_QUOTA_WARNING
1897         {
1898                 .ctl_name       = FS_DQ_WARNINGS,
1899                 .procname       = "warnings",
1900                 .data           = &flag_print_warnings,
1901                 .maxlen         = sizeof(int),
1902                 .mode           = 0644,
1903                 .proc_handler   = &proc_dointvec,
1904         },
1905 #endif
1906         { .ctl_name = 0 },
1907 };
1908
1909 static ctl_table fs_table[] = {
1910         {
1911                 .ctl_name       = FS_DQSTATS,
1912                 .procname       = "quota",
1913                 .mode           = 0555,
1914                 .child          = fs_dqstats_table,
1915         },
1916         { .ctl_name = 0 },
1917 };
1918
1919 static ctl_table sys_table[] = {
1920         {
1921                 .ctl_name       = CTL_FS,
1922                 .procname       = "fs",
1923                 .mode           = 0555,
1924                 .child          = fs_table,
1925         },
1926         { .ctl_name = 0 },
1927 };
1928
1929 static int __init dquot_init(void)
1930 {
1931         int i;
1932         unsigned long nr_hash, order;
1933
1934         printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
1935
1936         register_sysctl_table(sys_table);
1937
1938         dquot_cachep = kmem_cache_create("dquot",
1939                         sizeof(struct dquot), sizeof(unsigned long) * 4,
1940                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1941                                 SLAB_MEM_SPREAD|SLAB_PANIC),
1942                         NULL);
1943
1944         order = 0;
1945         dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
1946         if (!dquot_hash)
1947                 panic("Cannot create dquot hash table");
1948
1949         /* Find power-of-two hlist_heads which can fit into allocation */
1950         nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
1951         dq_hash_bits = 0;
1952         do {
1953                 dq_hash_bits++;
1954         } while (nr_hash >> dq_hash_bits);
1955         dq_hash_bits--;
1956
1957         nr_hash = 1UL << dq_hash_bits;
1958         dq_hash_mask = nr_hash - 1;
1959         for (i = 0; i < nr_hash; i++)
1960                 INIT_HLIST_HEAD(dquot_hash + i);
1961
1962         printk("Dquot-cache hash table entries: %ld (order %ld, %ld bytes)\n",
1963                         nr_hash, order, (PAGE_SIZE << order));
1964
1965         register_shrinker(&dqcache_shrinker);
1966
1967 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
1968         if (genl_register_family(&quota_genl_family) != 0)
1969                 printk(KERN_ERR "VFS: Failed to create quota netlink interface.\n");
1970 #endif
1971
1972         return 0;
1973 }
1974 module_init(dquot_init);
1975
1976 EXPORT_SYMBOL(register_quota_format);
1977 EXPORT_SYMBOL(unregister_quota_format);
1978 EXPORT_SYMBOL(dqstats);
1979 EXPORT_SYMBOL(dq_data_lock);
1980 EXPORT_SYMBOL(vfs_quota_on);
1981 EXPORT_SYMBOL(vfs_quota_on_mount);
1982 EXPORT_SYMBOL(vfs_quota_off);
1983 EXPORT_SYMBOL(vfs_quota_sync);
1984 EXPORT_SYMBOL(vfs_get_dqinfo);
1985 EXPORT_SYMBOL(vfs_set_dqinfo);
1986 EXPORT_SYMBOL(vfs_get_dqblk);
1987 EXPORT_SYMBOL(vfs_set_dqblk);
1988 EXPORT_SYMBOL(dquot_commit);
1989 EXPORT_SYMBOL(dquot_commit_info);
1990 EXPORT_SYMBOL(dquot_acquire);
1991 EXPORT_SYMBOL(dquot_release);
1992 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
1993 EXPORT_SYMBOL(dquot_initialize);
1994 EXPORT_SYMBOL(dquot_drop);
1995 EXPORT_SYMBOL(dquot_alloc_space);
1996 EXPORT_SYMBOL(dquot_alloc_inode);
1997 EXPORT_SYMBOL(dquot_free_space);
1998 EXPORT_SYMBOL(dquot_free_inode);
1999 EXPORT_SYMBOL(dquot_transfer);