c6ae6c0e0cfc53aa70f7b4679d6f84295c69c10f
[linux-3.10.git] / 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/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>
82
83 #include <asm/uaccess.h>
84
85 #define __DQUOT_PARANOIA
86
87 /*
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().
94  *
95  * The spinlock ordering is hence: dq_data_lock > dq_list_lock > i_lock
96  *
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
99  *
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()
110  *
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.
119  *
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)
124  */
125
126 static DEFINE_SPINLOCK(dq_list_lock);
127 DEFINE_SPINLOCK(dq_data_lock);
128
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;
132
133 /* SLAB cache for dquot structures */
134 static kmem_cache_t *dquot_cachep;
135
136 int register_quota_format(struct quota_format_type *fmt)
137 {
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;
143 }
144
145 void unregister_quota_format(struct quota_format_type *fmt)
146 {
147         struct quota_format_type **actqf;
148
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);
154 }
155
156 static struct quota_format_type *find_quota_format(int id)
157 {
158         struct quota_format_type *actqf;
159
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;
164
165                 spin_unlock(&dq_list_lock);
166                 
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;
170
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;
175         }
176         spin_unlock(&dq_list_lock);
177         return actqf;
178 }
179
180 static void put_quota_format(struct quota_format_type *fmt)
181 {
182         module_put(fmt->qf_owner);
183 }
184
185 /*
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.
190  *
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.
193  *
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.
199  *
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.
203  */
204
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;
209
210 struct dqstats dqstats;
211
212 static void dqput(struct dquot *dquot);
213
214 static inline unsigned int
215 hashfn(const struct super_block *sb, unsigned int id, int type)
216 {
217         unsigned long tmp;
218
219         tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
220         return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
221 }
222
223 /*
224  * Following list functions expect dq_list_lock to be held
225  */
226 static inline void insert_dquot_hash(struct dquot *dquot)
227 {
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);
230 }
231
232 static inline void remove_dquot_hash(struct dquot *dquot)
233 {
234         hlist_del_init(&dquot->dq_hash);
235 }
236
237 static inline struct dquot *find_dquot(unsigned int hashent, struct super_block *sb, unsigned int id, int type)
238 {
239         struct hlist_node *node;
240         struct dquot *dquot;
241
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;
246         }
247         return NODQUOT;
248 }
249
250 /* Add a dquot to the tail of the free list */
251 static inline void put_dquot_last(struct dquot *dquot)
252 {
253         list_add_tail(&dquot->dq_free, &free_dquots);
254         dqstats.free_dquots++;
255 }
256
257 static inline void remove_free_dquot(struct dquot *dquot)
258 {
259         if (list_empty(&dquot->dq_free))
260                 return;
261         list_del_init(&dquot->dq_free);
262         dqstats.free_dquots--;
263 }
264
265 static inline void put_inuse(struct dquot *dquot)
266 {
267         /* We add to the back of inuse list so we don't have to restart
268          * when traversing this list and we block */
269         list_add_tail(&dquot->dq_inuse, &inuse_list);
270         dqstats.allocated_dquots++;
271 }
272
273 static inline void remove_inuse(struct dquot *dquot)
274 {
275         dqstats.allocated_dquots--;
276         list_del(&dquot->dq_inuse);
277 }
278 /*
279  * End of list functions needing dq_list_lock
280  */
281
282 static void wait_on_dquot(struct dquot *dquot)
283 {
284         mutex_lock(&dquot->dq_lock);
285         mutex_unlock(&dquot->dq_lock);
286 }
287
288 #define mark_dquot_dirty(dquot) ((dquot)->dq_sb->dq_op->mark_dirty(dquot))
289
290 int dquot_mark_dquot_dirty(struct dquot *dquot)
291 {
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;
298 }
299
300 /* This function needs dq_list_lock */
301 static inline int clear_dquot_dirty(struct dquot *dquot)
302 {
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;
307 }
308
309 void mark_info_dirty(struct super_block *sb, int type)
310 {
311         set_bit(DQF_INFO_DIRTY_B, &sb_dqopt(sb)->info[type].dqi_flags);
312 }
313 EXPORT_SYMBOL(mark_info_dirty);
314
315 /*
316  *      Read dquot from disk and alloc space for it
317  */
318
319 int dquot_acquire(struct dquot *dquot)
320 {
321         int ret = 0, ret2 = 0;
322         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
323
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;
342                 }
343         }
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;
349 }
350
351 /*
352  *      Write dquot to disk
353  */
354 int dquot_commit(struct dquot *dquot)
355 {
356         int ret = 0, ret2 = 0;
357         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
358
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;
364         }
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;
374         }
375 out_sem:
376         mutex_unlock(&dqopt->dqio_mutex);
377         return ret;
378 }
379
380 /*
381  *      Release dquot
382  */
383 int dquot_release(struct dquot *dquot)
384 {
385         int ret = 0, ret2 = 0;
386         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
387
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;
400         }
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;
406 }
407
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.
413  */
414 static void invalidate_dquots(struct super_block *sb, int type)
415 {
416         struct dquot *dquot, *tmp;
417
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);
428
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.
439                          */
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;
448                 }
449                 /*
450                  * Quota now has no users and it has been written on last
451                  * dqput()
452                  */
453                 remove_dquot_hash(dquot);
454                 remove_free_dquot(dquot);
455                 remove_inuse(dquot);
456                 kmem_cache_free(dquot_cachep, dquot);
457         }
458         spin_unlock(&dq_list_lock);
459 }
460
461 int vfs_quota_sync(struct super_block *sb, int type)
462 {
463         struct list_head *dirty;
464         struct dquot *dquot;
465         struct quota_info *dqopt = sb_dqopt(sb);
466         int cnt;
467
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;
482                         }
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);
492                 }
493                 spin_unlock(&dq_list_lock);
494         }
495
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);
504
505         return 0;
506 }
507
508 /* Free unused dquots from cache */
509 static void prune_dqcache(int count)
510 {
511         struct list_head *head;
512         struct dquot *dquot;
513
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;
523         }
524 }
525
526 /*
527  * This is called from kswapd when we think we need some
528  * more memory
529  */
530
531 static int shrink_dqcache_memory(int nr, gfp_t gfp_mask)
532 {
533         if (nr) {
534                 spin_lock(&dq_list_lock);
535                 prune_dqcache(nr);
536                 spin_unlock(&dq_list_lock);
537         }
538         return (dqstats.free_dquots / 100) * sysctl_vfs_cache_pressure;
539 }
540
541 /*
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
545  */
546 static void dqput(struct dquot *dquot)
547 {
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();
558         }
559 #endif
560         
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;
575         }
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;
582         }
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;
589         }
590         atomic_dec(&dquot->dq_count);
591 #ifdef __DQUOT_PARANOIA
592         /* sanity check */
593         BUG_ON(!list_empty(&dquot->dq_free));
594 #endif
595         put_dquot_last(dquot);
596         spin_unlock(&dq_list_lock);
597 }
598
599 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
600 {
601         struct dquot *dquot;
602
603         dquot = kmem_cache_alloc(dquot_cachep, GFP_NOFS);
604         if(!dquot)
605                 return NODQUOT;
606
607         memset((caddr_t)dquot, 0, sizeof(struct dquot));
608         mutex_init(&dquot->dq_lock);
609         INIT_LIST_HEAD(&dquot->dq_free);
610         INIT_LIST_HEAD(&dquot->dq_inuse);
611         INIT_HLIST_NODE(&dquot->dq_hash);
612         INIT_LIST_HEAD(&dquot->dq_dirty);
613         init_waitqueue_head(&dquot->dq_wait_unused);
614         dquot->dq_sb = sb;
615         dquot->dq_type = type;
616         atomic_set(&dquot->dq_count, 1);
617
618         return dquot;
619 }
620
621 /*
622  * Get reference to dquot
623  * MUST be called with either dqptr_sem or dqonoff_mutex held
624  */
625 static struct dquot *dqget(struct super_block *sb, unsigned int id, int type)
626 {
627         unsigned int hashent = hashfn(sb, id, type);
628         struct dquot *dquot, *empty = NODQUOT;
629
630         if (!sb_has_quota_enabled(sb, type))
631                 return NODQUOT;
632 we_slept:
633         spin_lock(&dq_list_lock);
634         if ((dquot = find_dquot(hashent, sb, id, type)) == NODQUOT) {
635                 if (empty == NODQUOT) {
636                         spin_unlock(&dq_list_lock);
637                         if ((empty = get_empty_dquot(sb, type)) == NODQUOT)
638                                 schedule();     /* Try to wait for a moment... */
639                         goto we_slept;
640                 }
641                 dquot = empty;
642                 dquot->dq_id = id;
643                 /* all dquots go on the inuse_list */
644                 put_inuse(dquot);
645                 /* hash it first so it can be found */
646                 insert_dquot_hash(dquot);
647                 dqstats.lookups++;
648                 spin_unlock(&dq_list_lock);
649         } else {
650                 if (!atomic_read(&dquot->dq_count))
651                         remove_free_dquot(dquot);
652                 atomic_inc(&dquot->dq_count);
653                 dqstats.cache_hits++;
654                 dqstats.lookups++;
655                 spin_unlock(&dq_list_lock);
656                 if (empty)
657                         kmem_cache_free(dquot_cachep, empty);
658         }
659         /* Wait for dq_lock - after this we know that either dquot_release() is already
660          * finished or it will be canceled due to dq_count > 1 test */
661         wait_on_dquot(dquot);
662         /* Read the dquot and instantiate it (everything done only if needed) */
663         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && sb->dq_op->acquire_dquot(dquot) < 0) {
664                 dqput(dquot);
665                 return NODQUOT;
666         }
667 #ifdef __DQUOT_PARANOIA
668         BUG_ON(!dquot->dq_sb);  /* Has somebody invalidated entry under us? */
669 #endif
670
671         return dquot;
672 }
673
674 static int dqinit_needed(struct inode *inode, int type)
675 {
676         int cnt;
677
678         if (IS_NOQUOTA(inode))
679                 return 0;
680         if (type != -1)
681                 return inode->i_dquot[type] == NODQUOT;
682         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
683                 if (inode->i_dquot[cnt] == NODQUOT)
684                         return 1;
685         return 0;
686 }
687
688 /* This routine is guarded by dqonoff_mutex mutex */
689 static void add_dquot_ref(struct super_block *sb, int type)
690 {
691         struct list_head *p;
692
693 restart:
694         file_list_lock();
695         list_for_each(p, &sb->s_files) {
696                 struct file *filp = list_entry(p, struct file, f_u.fu_list);
697                 struct inode *inode = filp->f_dentry->d_inode;
698                 if (filp->f_mode & FMODE_WRITE && dqinit_needed(inode, type)) {
699                         struct dentry *dentry = dget(filp->f_dentry);
700                         file_list_unlock();
701                         sb->dq_op->initialize(inode, type);
702                         dput(dentry);
703                         /* As we may have blocked we had better restart... */
704                         goto restart;
705                 }
706         }
707         file_list_unlock();
708 }
709
710 /* Return 0 if dqput() won't block (note that 1 doesn't necessarily mean blocking) */
711 static inline int dqput_blocks(struct dquot *dquot)
712 {
713         if (atomic_read(&dquot->dq_count) <= 1)
714                 return 1;
715         return 0;
716 }
717
718 /* Remove references to dquots from inode - add dquot to list for freeing if needed */
719 /* We can't race with anybody because we hold dqptr_sem for writing... */
720 int remove_inode_dquot_ref(struct inode *inode, int type, struct list_head *tofree_head)
721 {
722         struct dquot *dquot = inode->i_dquot[type];
723
724         inode->i_dquot[type] = NODQUOT;
725         if (dquot != NODQUOT) {
726                 if (dqput_blocks(dquot)) {
727 #ifdef __DQUOT_PARANOIA
728                         if (atomic_read(&dquot->dq_count) != 1)
729                                 printk(KERN_WARNING "VFS: Adding dquot with dq_count %d to dispose list.\n", atomic_read(&dquot->dq_count));
730 #endif
731                         spin_lock(&dq_list_lock);
732                         list_add(&dquot->dq_free, tofree_head); /* As dquot must have currently users it can't be on the free list... */
733                         spin_unlock(&dq_list_lock);
734                         return 1;
735                 }
736                 else
737                         dqput(dquot);   /* We have guaranteed we won't block */
738         }
739         return 0;
740 }
741
742 /* Free list of dquots - called from inode.c */
743 /* dquots are removed from inodes, no new references can be got so we are the only ones holding reference */
744 static void put_dquot_list(struct list_head *tofree_head)
745 {
746         struct list_head *act_head;
747         struct dquot *dquot;
748
749         act_head = tofree_head->next;
750         /* So now we have dquots on the list... Just free them */
751         while (act_head != tofree_head) {
752                 dquot = list_entry(act_head, struct dquot, dq_free);
753                 act_head = act_head->next;
754                 list_del_init(&dquot->dq_free); /* Remove dquot from the list so we won't have problems... */
755                 dqput(dquot);
756         }
757 }
758
759 /* Gather all references from inodes and drop them */
760 static void drop_dquot_ref(struct super_block *sb, int type)
761 {
762         LIST_HEAD(tofree_head);
763
764         down_write(&sb_dqopt(sb)->dqptr_sem);
765         remove_dquot_ref(sb, type, &tofree_head);
766         up_write(&sb_dqopt(sb)->dqptr_sem);
767         put_dquot_list(&tofree_head);
768 }
769
770 static inline void dquot_incr_inodes(struct dquot *dquot, unsigned long number)
771 {
772         dquot->dq_dqb.dqb_curinodes += number;
773 }
774
775 static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
776 {
777         dquot->dq_dqb.dqb_curspace += number;
778 }
779
780 static inline void dquot_decr_inodes(struct dquot *dquot, unsigned long number)
781 {
782         if (dquot->dq_dqb.dqb_curinodes > number)
783                 dquot->dq_dqb.dqb_curinodes -= number;
784         else
785                 dquot->dq_dqb.dqb_curinodes = 0;
786         if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
787                 dquot->dq_dqb.dqb_itime = (time_t) 0;
788         clear_bit(DQ_INODES_B, &dquot->dq_flags);
789 }
790
791 static inline void dquot_decr_space(struct dquot *dquot, qsize_t number)
792 {
793         if (dquot->dq_dqb.dqb_curspace > number)
794                 dquot->dq_dqb.dqb_curspace -= number;
795         else
796                 dquot->dq_dqb.dqb_curspace = 0;
797         if (toqb(dquot->dq_dqb.dqb_curspace) <= dquot->dq_dqb.dqb_bsoftlimit)
798                 dquot->dq_dqb.dqb_btime = (time_t) 0;
799         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
800 }
801
802 static int flag_print_warnings = 1;
803
804 static inline int need_print_warning(struct dquot *dquot)
805 {
806         if (!flag_print_warnings)
807                 return 0;
808
809         switch (dquot->dq_type) {
810                 case USRQUOTA:
811                         return current->fsuid == dquot->dq_id;
812                 case GRPQUOTA:
813                         return in_group_p(dquot->dq_id);
814         }
815         return 0;
816 }
817
818 /* Values of warnings */
819 #define NOWARN 0
820 #define IHARDWARN 1
821 #define ISOFTLONGWARN 2
822 #define ISOFTWARN 3
823 #define BHARDWARN 4
824 #define BSOFTLONGWARN 5
825 #define BSOFTWARN 6
826
827 /* Print warning to user which exceeded quota */
828 static void print_warning(struct dquot *dquot, const char warntype)
829 {
830         char *msg = NULL;
831         int flag = (warntype == BHARDWARN || warntype == BSOFTLONGWARN) ? DQ_BLKS_B :
832           ((warntype == IHARDWARN || warntype == ISOFTLONGWARN) ? DQ_INODES_B : 0);
833
834         if (!need_print_warning(dquot) || (flag && test_and_set_bit(flag, &dquot->dq_flags)))
835                 return;
836
837         mutex_lock(&tty_mutex);
838         if (!current->signal->tty)
839                 goto out_lock;
840         tty_write_message(current->signal->tty, dquot->dq_sb->s_id);
841         if (warntype == ISOFTWARN || warntype == BSOFTWARN)
842                 tty_write_message(current->signal->tty, ": warning, ");
843         else
844                 tty_write_message(current->signal->tty, ": write failed, ");
845         tty_write_message(current->signal->tty, quotatypes[dquot->dq_type]);
846         switch (warntype) {
847                 case IHARDWARN:
848                         msg = " file limit reached.\r\n";
849                         break;
850                 case ISOFTLONGWARN:
851                         msg = " file quota exceeded too long.\r\n";
852                         break;
853                 case ISOFTWARN:
854                         msg = " file quota exceeded.\r\n";
855                         break;
856                 case BHARDWARN:
857                         msg = " block limit reached.\r\n";
858                         break;
859                 case BSOFTLONGWARN:
860                         msg = " block quota exceeded too long.\r\n";
861                         break;
862                 case BSOFTWARN:
863                         msg = " block quota exceeded.\r\n";
864                         break;
865         }
866         tty_write_message(current->signal->tty, msg);
867 out_lock:
868         mutex_unlock(&tty_mutex);
869 }
870
871 static inline void flush_warnings(struct dquot **dquots, char *warntype)
872 {
873         int i;
874
875         for (i = 0; i < MAXQUOTAS; i++)
876                 if (dquots[i] != NODQUOT && warntype[i] != NOWARN)
877                         print_warning(dquots[i], warntype[i]);
878 }
879
880 static inline char ignore_hardlimit(struct dquot *dquot)
881 {
882         struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_type];
883
884         return capable(CAP_SYS_RESOURCE) &&
885             (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD || !(info->dqi_flags & V1_DQF_RSQUASH));
886 }
887
888 /* needs dq_data_lock */
889 static int check_idq(struct dquot *dquot, ulong inodes, char *warntype)
890 {
891         *warntype = NOWARN;
892         if (inodes <= 0 || test_bit(DQ_FAKE_B, &dquot->dq_flags))
893                 return QUOTA_OK;
894
895         if (dquot->dq_dqb.dqb_ihardlimit &&
896            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_ihardlimit &&
897             !ignore_hardlimit(dquot)) {
898                 *warntype = IHARDWARN;
899                 return NO_QUOTA;
900         }
901
902         if (dquot->dq_dqb.dqb_isoftlimit &&
903            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
904             dquot->dq_dqb.dqb_itime && get_seconds() >= dquot->dq_dqb.dqb_itime &&
905             !ignore_hardlimit(dquot)) {
906                 *warntype = ISOFTLONGWARN;
907                 return NO_QUOTA;
908         }
909
910         if (dquot->dq_dqb.dqb_isoftlimit &&
911            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
912             dquot->dq_dqb.dqb_itime == 0) {
913                 *warntype = ISOFTWARN;
914                 dquot->dq_dqb.dqb_itime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
915         }
916
917         return QUOTA_OK;
918 }
919
920 /* needs dq_data_lock */
921 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc, char *warntype)
922 {
923         *warntype = 0;
924         if (space <= 0 || test_bit(DQ_FAKE_B, &dquot->dq_flags))
925                 return QUOTA_OK;
926
927         if (dquot->dq_dqb.dqb_bhardlimit &&
928            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bhardlimit &&
929             !ignore_hardlimit(dquot)) {
930                 if (!prealloc)
931                         *warntype = BHARDWARN;
932                 return NO_QUOTA;
933         }
934
935         if (dquot->dq_dqb.dqb_bsoftlimit &&
936            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bsoftlimit &&
937             dquot->dq_dqb.dqb_btime && get_seconds() >= dquot->dq_dqb.dqb_btime &&
938             !ignore_hardlimit(dquot)) {
939                 if (!prealloc)
940                         *warntype = BSOFTLONGWARN;
941                 return NO_QUOTA;
942         }
943
944         if (dquot->dq_dqb.dqb_bsoftlimit &&
945            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bsoftlimit &&
946             dquot->dq_dqb.dqb_btime == 0) {
947                 if (!prealloc) {
948                         *warntype = BSOFTWARN;
949                         dquot->dq_dqb.dqb_btime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_bgrace;
950                 }
951                 else
952                         /*
953                          * We don't allow preallocation to exceed softlimit so exceeding will
954                          * be always printed
955                          */
956                         return NO_QUOTA;
957         }
958
959         return QUOTA_OK;
960 }
961
962 /*
963  *      Initialize quota pointers in inode
964  *      Transaction must be started at entry
965  */
966 int dquot_initialize(struct inode *inode, int type)
967 {
968         unsigned int id = 0;
969         int cnt, ret = 0;
970
971         /* First test before acquiring mutex - solves deadlocks when we
972          * re-enter the quota code and are already holding the mutex */
973         if (IS_NOQUOTA(inode))
974                 return 0;
975         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
976         /* Having dqptr_sem we know NOQUOTA flags can't be altered... */
977         if (IS_NOQUOTA(inode))
978                 goto out_err;
979         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
980                 if (type != -1 && cnt != type)
981                         continue;
982                 if (inode->i_dquot[cnt] == NODQUOT) {
983                         switch (cnt) {
984                                 case USRQUOTA:
985                                         id = inode->i_uid;
986                                         break;
987                                 case GRPQUOTA:
988                                         id = inode->i_gid;
989                                         break;
990                         }
991                         inode->i_dquot[cnt] = dqget(inode->i_sb, id, cnt);
992                 }
993         }
994 out_err:
995         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
996         return ret;
997 }
998
999 /*
1000  *      Release all quotas referenced by inode
1001  *      Transaction must be started at an entry
1002  */
1003 int dquot_drop(struct inode *inode)
1004 {
1005         int cnt;
1006
1007         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1008         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1009                 if (inode->i_dquot[cnt] != NODQUOT) {
1010                         dqput(inode->i_dquot[cnt]);
1011                         inode->i_dquot[cnt] = NODQUOT;
1012                 }
1013         }
1014         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1015         return 0;
1016 }
1017
1018 /*
1019  * Following four functions update i_blocks+i_bytes fields and
1020  * quota information (together with appropriate checks)
1021  * NOTE: We absolutely rely on the fact that caller dirties
1022  * the inode (usually macros in quotaops.h care about this) and
1023  * holds a handle for the current transaction so that dquot write and
1024  * inode write go into the same transaction.
1025  */
1026
1027 /*
1028  * This operation can block, but only after everything is updated
1029  */
1030 int dquot_alloc_space(struct inode *inode, qsize_t number, int warn)
1031 {
1032         int cnt, ret = NO_QUOTA;
1033         char warntype[MAXQUOTAS];
1034
1035         /* First test before acquiring mutex - solves deadlocks when we
1036          * re-enter the quota code and are already holding the mutex */
1037         if (IS_NOQUOTA(inode)) {
1038 out_add:
1039                 inode_add_bytes(inode, number);
1040                 return QUOTA_OK;
1041         }
1042         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1043                 warntype[cnt] = NOWARN;
1044
1045         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1046         if (IS_NOQUOTA(inode)) {        /* Now we can do reliable test... */
1047                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1048                 goto out_add;
1049         }
1050         spin_lock(&dq_data_lock);
1051         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1052                 if (inode->i_dquot[cnt] == NODQUOT)
1053                         continue;
1054                 if (check_bdq(inode->i_dquot[cnt], number, warn, warntype+cnt) == NO_QUOTA)
1055                         goto warn_put_all;
1056         }
1057         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1058                 if (inode->i_dquot[cnt] == NODQUOT)
1059                         continue;
1060                 dquot_incr_space(inode->i_dquot[cnt], number);
1061         }
1062         inode_add_bytes(inode, number);
1063         ret = QUOTA_OK;
1064 warn_put_all:
1065         spin_unlock(&dq_data_lock);
1066         if (ret == QUOTA_OK)
1067                 /* Dirtify all the dquots - this can block when journalling */
1068                 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1069                         if (inode->i_dquot[cnt])
1070                                 mark_dquot_dirty(inode->i_dquot[cnt]);
1071         flush_warnings(inode->i_dquot, warntype);
1072         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1073         return ret;
1074 }
1075
1076 /*
1077  * This operation can block, but only after everything is updated
1078  */
1079 int dquot_alloc_inode(const struct inode *inode, unsigned long number)
1080 {
1081         int cnt, ret = NO_QUOTA;
1082         char warntype[MAXQUOTAS];
1083
1084         /* First test before acquiring mutex - solves deadlocks when we
1085          * re-enter the quota code and are already holding the mutex */
1086         if (IS_NOQUOTA(inode))
1087                 return QUOTA_OK;
1088         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1089                 warntype[cnt] = NOWARN;
1090         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1091         if (IS_NOQUOTA(inode)) {
1092                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1093                 return QUOTA_OK;
1094         }
1095         spin_lock(&dq_data_lock);
1096         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1097                 if (inode->i_dquot[cnt] == NODQUOT)
1098                         continue;
1099                 if (check_idq(inode->i_dquot[cnt], number, warntype+cnt) == NO_QUOTA)
1100                         goto warn_put_all;
1101         }
1102
1103         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1104                 if (inode->i_dquot[cnt] == NODQUOT)
1105                         continue;
1106                 dquot_incr_inodes(inode->i_dquot[cnt], number);
1107         }
1108         ret = QUOTA_OK;
1109 warn_put_all:
1110         spin_unlock(&dq_data_lock);
1111         if (ret == QUOTA_OK)
1112                 /* Dirtify all the dquots - this can block when journalling */
1113                 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1114                         if (inode->i_dquot[cnt])
1115                                 mark_dquot_dirty(inode->i_dquot[cnt]);
1116         flush_warnings((struct dquot **)inode->i_dquot, warntype);
1117         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1118         return ret;
1119 }
1120
1121 /*
1122  * This operation can block, but only after everything is updated
1123  */
1124 int dquot_free_space(struct inode *inode, qsize_t number)
1125 {
1126         unsigned int cnt;
1127
1128         /* First test before acquiring mutex - solves deadlocks when we
1129          * re-enter the quota code and are already holding the mutex */
1130         if (IS_NOQUOTA(inode)) {
1131 out_sub:
1132                 inode_sub_bytes(inode, number);
1133                 return QUOTA_OK;
1134         }
1135         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1136         /* Now recheck reliably when holding dqptr_sem */
1137         if (IS_NOQUOTA(inode)) {
1138                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1139                 goto out_sub;
1140         }
1141         spin_lock(&dq_data_lock);
1142         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1143                 if (inode->i_dquot[cnt] == NODQUOT)
1144                         continue;
1145                 dquot_decr_space(inode->i_dquot[cnt], number);
1146         }
1147         inode_sub_bytes(inode, number);
1148         spin_unlock(&dq_data_lock);
1149         /* Dirtify all the dquots - this can block when journalling */
1150         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1151                 if (inode->i_dquot[cnt])
1152                         mark_dquot_dirty(inode->i_dquot[cnt]);
1153         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1154         return QUOTA_OK;
1155 }
1156
1157 /*
1158  * This operation can block, but only after everything is updated
1159  */
1160 int dquot_free_inode(const struct inode *inode, unsigned long number)
1161 {
1162         unsigned int cnt;
1163
1164         /* First test before acquiring mutex - solves deadlocks when we
1165          * re-enter the quota code and are already holding the mutex */
1166         if (IS_NOQUOTA(inode))
1167                 return QUOTA_OK;
1168         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1169         /* Now recheck reliably when holding dqptr_sem */
1170         if (IS_NOQUOTA(inode)) {
1171                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1172                 return QUOTA_OK;
1173         }
1174         spin_lock(&dq_data_lock);
1175         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1176                 if (inode->i_dquot[cnt] == NODQUOT)
1177                         continue;
1178                 dquot_decr_inodes(inode->i_dquot[cnt], number);
1179         }
1180         spin_unlock(&dq_data_lock);
1181         /* Dirtify all the dquots - this can block when journalling */
1182         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1183                 if (inode->i_dquot[cnt])
1184                         mark_dquot_dirty(inode->i_dquot[cnt]);
1185         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1186         return QUOTA_OK;
1187 }
1188
1189 /*
1190  * Transfer the number of inode and blocks from one diskquota to an other.
1191  *
1192  * This operation can block, but only after everything is updated
1193  * A transaction must be started when entering this function.
1194  */
1195 int dquot_transfer(struct inode *inode, struct iattr *iattr)
1196 {
1197         qsize_t space;
1198         struct dquot *transfer_from[MAXQUOTAS];
1199         struct dquot *transfer_to[MAXQUOTAS];
1200         int cnt, ret = NO_QUOTA, chuid = (iattr->ia_valid & ATTR_UID) && inode->i_uid != iattr->ia_uid,
1201             chgid = (iattr->ia_valid & ATTR_GID) && inode->i_gid != iattr->ia_gid;
1202         char warntype[MAXQUOTAS];
1203
1204         /* First test before acquiring mutex - solves deadlocks when we
1205          * re-enter the quota code and are already holding the mutex */
1206         if (IS_NOQUOTA(inode))
1207                 return QUOTA_OK;
1208         /* Clear the arrays */
1209         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1210                 transfer_to[cnt] = transfer_from[cnt] = NODQUOT;
1211                 warntype[cnt] = NOWARN;
1212         }
1213         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1214         /* Now recheck reliably when holding dqptr_sem */
1215         if (IS_NOQUOTA(inode)) {        /* File without quota accounting? */
1216                 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1217                 return QUOTA_OK;
1218         }
1219         /* First build the transfer_to list - here we can block on
1220          * reading/instantiating of dquots.  We know that the transaction for
1221          * us was already started so we don't violate lock ranking here */
1222         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1223                 switch (cnt) {
1224                         case USRQUOTA:
1225                                 if (!chuid)
1226                                         continue;
1227                                 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_uid, cnt);
1228                                 break;
1229                         case GRPQUOTA:
1230                                 if (!chgid)
1231                                         continue;
1232                                 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_gid, cnt);
1233                                 break;
1234                 }
1235         }
1236         spin_lock(&dq_data_lock);
1237         space = inode_get_bytes(inode);
1238         /* Build the transfer_from list and check the limits */
1239         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1240                 if (transfer_to[cnt] == NODQUOT)
1241                         continue;
1242                 transfer_from[cnt] = inode->i_dquot[cnt];
1243                 if (check_idq(transfer_to[cnt], 1, warntype+cnt) == NO_QUOTA ||
1244                     check_bdq(transfer_to[cnt], space, 0, warntype+cnt) == NO_QUOTA)
1245                         goto warn_put_all;
1246         }
1247
1248         /*
1249          * Finally perform the needed transfer from transfer_from to transfer_to
1250          */
1251         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1252                 /*
1253                  * Skip changes for same uid or gid or for turned off quota-type.
1254                  */
1255                 if (transfer_to[cnt] == NODQUOT)
1256                         continue;
1257
1258                 /* Due to IO error we might not have transfer_from[] structure */
1259                 if (transfer_from[cnt]) {
1260                         dquot_decr_inodes(transfer_from[cnt], 1);
1261                         dquot_decr_space(transfer_from[cnt], space);
1262                 }
1263
1264                 dquot_incr_inodes(transfer_to[cnt], 1);
1265                 dquot_incr_space(transfer_to[cnt], space);
1266
1267                 inode->i_dquot[cnt] = transfer_to[cnt];
1268         }
1269         ret = QUOTA_OK;
1270 warn_put_all:
1271         spin_unlock(&dq_data_lock);
1272         /* Dirtify all the dquots - this can block when journalling */
1273         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1274                 if (transfer_from[cnt])
1275                         mark_dquot_dirty(transfer_from[cnt]);
1276                 if (transfer_to[cnt])
1277                         mark_dquot_dirty(transfer_to[cnt]);
1278         }
1279         flush_warnings(transfer_to, warntype);
1280         
1281         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1282                 if (ret == QUOTA_OK && transfer_from[cnt] != NODQUOT)
1283                         dqput(transfer_from[cnt]);
1284                 if (ret == NO_QUOTA && transfer_to[cnt] != NODQUOT)
1285                         dqput(transfer_to[cnt]);
1286         }
1287         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1288         return ret;
1289 }
1290
1291 /*
1292  * Write info of quota file to disk
1293  */
1294 int dquot_commit_info(struct super_block *sb, int type)
1295 {
1296         int ret;
1297         struct quota_info *dqopt = sb_dqopt(sb);
1298
1299         mutex_lock(&dqopt->dqio_mutex);
1300         ret = dqopt->ops[type]->write_file_info(sb, type);
1301         mutex_unlock(&dqopt->dqio_mutex);
1302         return ret;
1303 }
1304
1305 /*
1306  * Definitions of diskquota operations.
1307  */
1308 struct dquot_operations dquot_operations = {
1309         .initialize     = dquot_initialize,
1310         .drop           = dquot_drop,
1311         .alloc_space    = dquot_alloc_space,
1312         .alloc_inode    = dquot_alloc_inode,
1313         .free_space     = dquot_free_space,
1314         .free_inode     = dquot_free_inode,
1315         .transfer       = dquot_transfer,
1316         .write_dquot    = dquot_commit,
1317         .acquire_dquot  = dquot_acquire,
1318         .release_dquot  = dquot_release,
1319         .mark_dirty     = dquot_mark_dquot_dirty,
1320         .write_info     = dquot_commit_info
1321 };
1322
1323 static inline void set_enable_flags(struct quota_info *dqopt, int type)
1324 {
1325         switch (type) {
1326                 case USRQUOTA:
1327                         dqopt->flags |= DQUOT_USR_ENABLED;
1328                         break;
1329                 case GRPQUOTA:
1330                         dqopt->flags |= DQUOT_GRP_ENABLED;
1331                         break;
1332         }
1333 }
1334
1335 static inline void reset_enable_flags(struct quota_info *dqopt, int type)
1336 {
1337         switch (type) {
1338                 case USRQUOTA:
1339                         dqopt->flags &= ~DQUOT_USR_ENABLED;
1340                         break;
1341                 case GRPQUOTA:
1342                         dqopt->flags &= ~DQUOT_GRP_ENABLED;
1343                         break;
1344         }
1345 }
1346
1347 /*
1348  * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
1349  */
1350 int vfs_quota_off(struct super_block *sb, int type)
1351 {
1352         int cnt;
1353         struct quota_info *dqopt = sb_dqopt(sb);
1354         struct inode *toputinode[MAXQUOTAS];
1355
1356         /* We need to serialize quota_off() for device */
1357         mutex_lock(&dqopt->dqonoff_mutex);
1358         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1359                 toputinode[cnt] = NULL;
1360                 if (type != -1 && cnt != type)
1361                         continue;
1362                 if (!sb_has_quota_enabled(sb, cnt))
1363                         continue;
1364                 reset_enable_flags(dqopt, cnt);
1365
1366                 /* Note: these are blocking operations */
1367                 drop_dquot_ref(sb, cnt);
1368                 invalidate_dquots(sb, cnt);
1369                 /*
1370                  * Now all dquots should be invalidated, all writes done so we should be only
1371                  * users of the info. No locks needed.
1372                  */
1373                 if (info_dirty(&dqopt->info[cnt]))
1374                         sb->dq_op->write_info(sb, cnt);
1375                 if (dqopt->ops[cnt]->free_file_info)
1376                         dqopt->ops[cnt]->free_file_info(sb, cnt);
1377                 put_quota_format(dqopt->info[cnt].dqi_format);
1378
1379                 toputinode[cnt] = dqopt->files[cnt];
1380                 dqopt->files[cnt] = NULL;
1381                 dqopt->info[cnt].dqi_flags = 0;
1382                 dqopt->info[cnt].dqi_igrace = 0;
1383                 dqopt->info[cnt].dqi_bgrace = 0;
1384                 dqopt->ops[cnt] = NULL;
1385         }
1386         mutex_unlock(&dqopt->dqonoff_mutex);
1387         /* Sync the superblock so that buffers with quota data are written to
1388          * disk (and so userspace sees correct data afterwards). */
1389         if (sb->s_op->sync_fs)
1390                 sb->s_op->sync_fs(sb, 1);
1391         sync_blockdev(sb->s_bdev);
1392         /* Now the quota files are just ordinary files and we can set the
1393          * inode flags back. Moreover we discard the pagecache so that
1394          * userspace sees the writes we did bypassing the pagecache. We
1395          * must also discard the blockdev buffers so that we see the
1396          * changes done by userspace on the next quotaon() */
1397         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1398                 if (toputinode[cnt]) {
1399                         mutex_lock(&dqopt->dqonoff_mutex);
1400                         /* If quota was reenabled in the meantime, we have
1401                          * nothing to do */
1402                         if (!sb_has_quota_enabled(sb, cnt)) {
1403                                 mutex_lock(&toputinode[cnt]->i_mutex);
1404                                 toputinode[cnt]->i_flags &= ~(S_IMMUTABLE |
1405                                   S_NOATIME | S_NOQUOTA);
1406                                 truncate_inode_pages(&toputinode[cnt]->i_data, 0);
1407                                 mutex_unlock(&toputinode[cnt]->i_mutex);
1408                                 mark_inode_dirty(toputinode[cnt]);
1409                                 iput(toputinode[cnt]);
1410                         }
1411                         mutex_unlock(&dqopt->dqonoff_mutex);
1412                 }
1413         if (sb->s_bdev)
1414                 invalidate_bdev(sb->s_bdev, 0);
1415         return 0;
1416 }
1417
1418 /*
1419  *      Turn quotas on on a device
1420  */
1421
1422 /* Helper function when we already have the inode */
1423 static int vfs_quota_on_inode(struct inode *inode, int type, int format_id)
1424 {
1425         struct quota_format_type *fmt = find_quota_format(format_id);
1426         struct super_block *sb = inode->i_sb;
1427         struct quota_info *dqopt = sb_dqopt(sb);
1428         int error;
1429         int oldflags = -1;
1430
1431         if (!fmt)
1432                 return -ESRCH;
1433         if (!S_ISREG(inode->i_mode)) {
1434                 error = -EACCES;
1435                 goto out_fmt;
1436         }
1437         if (IS_RDONLY(inode)) {
1438                 error = -EROFS;
1439                 goto out_fmt;
1440         }
1441         if (!sb->s_op->quota_write || !sb->s_op->quota_read) {
1442                 error = -EINVAL;
1443                 goto out_fmt;
1444         }
1445
1446         /* As we bypass the pagecache we must now flush the inode so that
1447          * we see all the changes from userspace... */
1448         write_inode_now(inode, 1);
1449         /* And now flush the block cache so that kernel sees the changes */
1450         invalidate_bdev(sb->s_bdev, 0);
1451         mutex_lock(&inode->i_mutex);
1452         mutex_lock(&dqopt->dqonoff_mutex);
1453         if (sb_has_quota_enabled(sb, type)) {
1454                 error = -EBUSY;
1455                 goto out_lock;
1456         }
1457         /* We don't want quota and atime on quota files (deadlocks possible)
1458          * Also nobody should write to the file - we use special IO operations
1459          * which ignore the immutable bit. */
1460         down_write(&dqopt->dqptr_sem);
1461         oldflags = inode->i_flags & (S_NOATIME | S_IMMUTABLE | S_NOQUOTA);
1462         inode->i_flags |= S_NOQUOTA | S_NOATIME | S_IMMUTABLE;
1463         up_write(&dqopt->dqptr_sem);
1464         sb->dq_op->drop(inode);
1465
1466         error = -EIO;
1467         dqopt->files[type] = igrab(inode);
1468         if (!dqopt->files[type])
1469                 goto out_lock;
1470         error = -EINVAL;
1471         if (!fmt->qf_ops->check_quota_file(sb, type))
1472                 goto out_file_init;
1473
1474         dqopt->ops[type] = fmt->qf_ops;
1475         dqopt->info[type].dqi_format = fmt;
1476         INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
1477         mutex_lock(&dqopt->dqio_mutex);
1478         if ((error = dqopt->ops[type]->read_file_info(sb, type)) < 0) {
1479                 mutex_unlock(&dqopt->dqio_mutex);
1480                 goto out_file_init;
1481         }
1482         mutex_unlock(&dqopt->dqio_mutex);
1483         mutex_unlock(&inode->i_mutex);
1484         set_enable_flags(dqopt, type);
1485
1486         add_dquot_ref(sb, type);
1487         mutex_unlock(&dqopt->dqonoff_mutex);
1488
1489         return 0;
1490
1491 out_file_init:
1492         dqopt->files[type] = NULL;
1493         iput(inode);
1494 out_lock:
1495         mutex_unlock(&dqopt->dqonoff_mutex);
1496         if (oldflags != -1) {
1497                 down_write(&dqopt->dqptr_sem);
1498                 /* Set the flags back (in the case of accidental quotaon()
1499                  * on a wrong file we don't want to mess up the flags) */
1500                 inode->i_flags &= ~(S_NOATIME | S_NOQUOTA | S_IMMUTABLE);
1501                 inode->i_flags |= oldflags;
1502                 up_write(&dqopt->dqptr_sem);
1503         }
1504         mutex_unlock(&inode->i_mutex);
1505 out_fmt:
1506         put_quota_format(fmt);
1507
1508         return error; 
1509 }
1510
1511 /* Actual function called from quotactl() */
1512 int vfs_quota_on(struct super_block *sb, int type, int format_id, char *path)
1513 {
1514         struct nameidata nd;
1515         int error;
1516
1517         error = path_lookup(path, LOOKUP_FOLLOW, &nd);
1518         if (error < 0)
1519                 return error;
1520         error = security_quota_on(nd.dentry);
1521         if (error)
1522                 goto out_path;
1523         /* Quota file not on the same filesystem? */
1524         if (nd.mnt->mnt_sb != sb)
1525                 error = -EXDEV;
1526         else
1527                 error = vfs_quota_on_inode(nd.dentry->d_inode, type, format_id);
1528 out_path:
1529         path_release(&nd);
1530         return error;
1531 }
1532
1533 /*
1534  * This function is used when filesystem needs to initialize quotas
1535  * during mount time.
1536  */
1537 int vfs_quota_on_mount(struct super_block *sb, char *qf_name,
1538                 int format_id, int type)
1539 {
1540         struct dentry *dentry;
1541         int error;
1542
1543         dentry = lookup_one_len(qf_name, sb->s_root, strlen(qf_name));
1544         if (IS_ERR(dentry))
1545                 return PTR_ERR(dentry);
1546
1547         if (!dentry->d_inode) {
1548                 error = -ENOENT;
1549                 goto out;
1550         }
1551
1552         error = security_quota_on(dentry);
1553         if (!error)
1554                 error = vfs_quota_on_inode(dentry->d_inode, type, format_id);
1555
1556 out:
1557         dput(dentry);
1558         return error;
1559 }
1560
1561 /* Generic routine for getting common part of quota structure */
1562 static void do_get_dqblk(struct dquot *dquot, struct if_dqblk *di)
1563 {
1564         struct mem_dqblk *dm = &dquot->dq_dqb;
1565
1566         spin_lock(&dq_data_lock);
1567         di->dqb_bhardlimit = dm->dqb_bhardlimit;
1568         di->dqb_bsoftlimit = dm->dqb_bsoftlimit;
1569         di->dqb_curspace = dm->dqb_curspace;
1570         di->dqb_ihardlimit = dm->dqb_ihardlimit;
1571         di->dqb_isoftlimit = dm->dqb_isoftlimit;
1572         di->dqb_curinodes = dm->dqb_curinodes;
1573         di->dqb_btime = dm->dqb_btime;
1574         di->dqb_itime = dm->dqb_itime;
1575         di->dqb_valid = QIF_ALL;
1576         spin_unlock(&dq_data_lock);
1577 }
1578
1579 int vfs_get_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
1580 {
1581         struct dquot *dquot;
1582
1583         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1584         if (!(dquot = dqget(sb, id, type))) {
1585                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1586                 return -ESRCH;
1587         }
1588         do_get_dqblk(dquot, di);
1589         dqput(dquot);
1590         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1591         return 0;
1592 }
1593
1594 /* Generic routine for setting common part of quota structure */
1595 static void do_set_dqblk(struct dquot *dquot, struct if_dqblk *di)
1596 {
1597         struct mem_dqblk *dm = &dquot->dq_dqb;
1598         int check_blim = 0, check_ilim = 0;
1599
1600         spin_lock(&dq_data_lock);
1601         if (di->dqb_valid & QIF_SPACE) {
1602                 dm->dqb_curspace = di->dqb_curspace;
1603                 check_blim = 1;
1604         }
1605         if (di->dqb_valid & QIF_BLIMITS) {
1606                 dm->dqb_bsoftlimit = di->dqb_bsoftlimit;
1607                 dm->dqb_bhardlimit = di->dqb_bhardlimit;
1608                 check_blim = 1;
1609         }
1610         if (di->dqb_valid & QIF_INODES) {
1611                 dm->dqb_curinodes = di->dqb_curinodes;
1612                 check_ilim = 1;
1613         }
1614         if (di->dqb_valid & QIF_ILIMITS) {
1615                 dm->dqb_isoftlimit = di->dqb_isoftlimit;
1616                 dm->dqb_ihardlimit = di->dqb_ihardlimit;
1617                 check_ilim = 1;
1618         }
1619         if (di->dqb_valid & QIF_BTIME)
1620                 dm->dqb_btime = di->dqb_btime;
1621         if (di->dqb_valid & QIF_ITIME)
1622                 dm->dqb_itime = di->dqb_itime;
1623
1624         if (check_blim) {
1625                 if (!dm->dqb_bsoftlimit || toqb(dm->dqb_curspace) < dm->dqb_bsoftlimit) {
1626                         dm->dqb_btime = 0;
1627                         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1628                 }
1629                 else if (!(di->dqb_valid & QIF_BTIME))  /* Set grace only if user hasn't provided his own... */
1630                         dm->dqb_btime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_bgrace;
1631         }
1632         if (check_ilim) {
1633                 if (!dm->dqb_isoftlimit || dm->dqb_curinodes < dm->dqb_isoftlimit) {
1634                         dm->dqb_itime = 0;
1635                         clear_bit(DQ_INODES_B, &dquot->dq_flags);
1636                 }
1637                 else if (!(di->dqb_valid & QIF_ITIME))  /* Set grace only if user hasn't provided his own... */
1638                         dm->dqb_itime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
1639         }
1640         if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit || dm->dqb_isoftlimit)
1641                 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
1642         else
1643                 set_bit(DQ_FAKE_B, &dquot->dq_flags);
1644         spin_unlock(&dq_data_lock);
1645         mark_dquot_dirty(dquot);
1646 }
1647
1648 int vfs_set_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
1649 {
1650         struct dquot *dquot;
1651
1652         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1653         if (!(dquot = dqget(sb, id, type))) {
1654                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1655                 return -ESRCH;
1656         }
1657         do_set_dqblk(dquot, di);
1658         dqput(dquot);
1659         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1660         return 0;
1661 }
1662
1663 /* Generic routine for getting common part of quota file information */
1664 int vfs_get_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
1665 {
1666         struct mem_dqinfo *mi;
1667   
1668         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1669         if (!sb_has_quota_enabled(sb, type)) {
1670                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1671                 return -ESRCH;
1672         }
1673         mi = sb_dqopt(sb)->info + type;
1674         spin_lock(&dq_data_lock);
1675         ii->dqi_bgrace = mi->dqi_bgrace;
1676         ii->dqi_igrace = mi->dqi_igrace;
1677         ii->dqi_flags = mi->dqi_flags & DQF_MASK;
1678         ii->dqi_valid = IIF_ALL;
1679         spin_unlock(&dq_data_lock);
1680         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1681         return 0;
1682 }
1683
1684 /* Generic routine for setting common part of quota file information */
1685 int vfs_set_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
1686 {
1687         struct mem_dqinfo *mi;
1688
1689         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1690         if (!sb_has_quota_enabled(sb, type)) {
1691                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1692                 return -ESRCH;
1693         }
1694         mi = sb_dqopt(sb)->info + type;
1695         spin_lock(&dq_data_lock);
1696         if (ii->dqi_valid & IIF_BGRACE)
1697                 mi->dqi_bgrace = ii->dqi_bgrace;
1698         if (ii->dqi_valid & IIF_IGRACE)
1699                 mi->dqi_igrace = ii->dqi_igrace;
1700         if (ii->dqi_valid & IIF_FLAGS)
1701                 mi->dqi_flags = (mi->dqi_flags & ~DQF_MASK) | (ii->dqi_flags & DQF_MASK);
1702         spin_unlock(&dq_data_lock);
1703         mark_info_dirty(sb, type);
1704         /* Force write to disk */
1705         sb->dq_op->write_info(sb, type);
1706         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1707         return 0;
1708 }
1709
1710 struct quotactl_ops vfs_quotactl_ops = {
1711         .quota_on       = vfs_quota_on,
1712         .quota_off      = vfs_quota_off,
1713         .quota_sync     = vfs_quota_sync,
1714         .get_info       = vfs_get_dqinfo,
1715         .set_info       = vfs_set_dqinfo,
1716         .get_dqblk      = vfs_get_dqblk,
1717         .set_dqblk      = vfs_set_dqblk
1718 };
1719
1720 static ctl_table fs_dqstats_table[] = {
1721         {
1722                 .ctl_name       = FS_DQ_LOOKUPS,
1723                 .procname       = "lookups",
1724                 .data           = &dqstats.lookups,
1725                 .maxlen         = sizeof(int),
1726                 .mode           = 0444,
1727                 .proc_handler   = &proc_dointvec,
1728         },
1729         {
1730                 .ctl_name       = FS_DQ_DROPS,
1731                 .procname       = "drops",
1732                 .data           = &dqstats.drops,
1733                 .maxlen         = sizeof(int),
1734                 .mode           = 0444,
1735                 .proc_handler   = &proc_dointvec,
1736         },
1737         {
1738                 .ctl_name       = FS_DQ_READS,
1739                 .procname       = "reads",
1740                 .data           = &dqstats.reads,
1741                 .maxlen         = sizeof(int),
1742                 .mode           = 0444,
1743                 .proc_handler   = &proc_dointvec,
1744         },
1745         {
1746                 .ctl_name       = FS_DQ_WRITES,
1747                 .procname       = "writes",
1748                 .data           = &dqstats.writes,
1749                 .maxlen         = sizeof(int),
1750                 .mode           = 0444,
1751                 .proc_handler   = &proc_dointvec,
1752         },
1753         {
1754                 .ctl_name       = FS_DQ_CACHE_HITS,
1755                 .procname       = "cache_hits",
1756                 .data           = &dqstats.cache_hits,
1757                 .maxlen         = sizeof(int),
1758                 .mode           = 0444,
1759                 .proc_handler   = &proc_dointvec,
1760         },
1761         {
1762                 .ctl_name       = FS_DQ_ALLOCATED,
1763                 .procname       = "allocated_dquots",
1764                 .data           = &dqstats.allocated_dquots,
1765                 .maxlen         = sizeof(int),
1766                 .mode           = 0444,
1767                 .proc_handler   = &proc_dointvec,
1768         },
1769         {
1770                 .ctl_name       = FS_DQ_FREE,
1771                 .procname       = "free_dquots",
1772                 .data           = &dqstats.free_dquots,
1773                 .maxlen         = sizeof(int),
1774                 .mode           = 0444,
1775                 .proc_handler   = &proc_dointvec,
1776         },
1777         {
1778                 .ctl_name       = FS_DQ_SYNCS,
1779                 .procname       = "syncs",
1780                 .data           = &dqstats.syncs,
1781                 .maxlen         = sizeof(int),
1782                 .mode           = 0444,
1783                 .proc_handler   = &proc_dointvec,
1784         },
1785         {
1786                 .ctl_name       = FS_DQ_WARNINGS,
1787                 .procname       = "warnings",
1788                 .data           = &flag_print_warnings,
1789                 .maxlen         = sizeof(int),
1790                 .mode           = 0644,
1791                 .proc_handler   = &proc_dointvec,
1792         },
1793         { .ctl_name = 0 },
1794 };
1795
1796 static ctl_table fs_table[] = {
1797         {
1798                 .ctl_name       = FS_DQSTATS,
1799                 .procname       = "quota",
1800                 .mode           = 0555,
1801                 .child          = fs_dqstats_table,
1802         },
1803         { .ctl_name = 0 },
1804 };
1805
1806 static ctl_table sys_table[] = {
1807         {
1808                 .ctl_name       = CTL_FS,
1809                 .procname       = "fs",
1810                 .mode           = 0555,
1811                 .child          = fs_table,
1812         },
1813         { .ctl_name = 0 },
1814 };
1815
1816 static int __init dquot_init(void)
1817 {
1818         int i;
1819         unsigned long nr_hash, order;
1820
1821         printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
1822
1823         register_sysctl_table(sys_table, 0);
1824
1825         dquot_cachep = kmem_cache_create("dquot", 
1826                         sizeof(struct dquot), sizeof(unsigned long) * 4,
1827                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1828                                 SLAB_MEM_SPREAD|SLAB_PANIC),
1829                         NULL, NULL);
1830
1831         order = 0;
1832         dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
1833         if (!dquot_hash)
1834                 panic("Cannot create dquot hash table");
1835
1836         /* Find power-of-two hlist_heads which can fit into allocation */
1837         nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
1838         dq_hash_bits = 0;
1839         do {
1840                 dq_hash_bits++;
1841         } while (nr_hash >> dq_hash_bits);
1842         dq_hash_bits--;
1843
1844         nr_hash = 1UL << dq_hash_bits;
1845         dq_hash_mask = nr_hash - 1;
1846         for (i = 0; i < nr_hash; i++)
1847                 INIT_HLIST_HEAD(dquot_hash + i);
1848
1849         printk("Dquot-cache hash table entries: %ld (order %ld, %ld bytes)\n",
1850                         nr_hash, order, (PAGE_SIZE << order));
1851
1852         set_shrinker(DEFAULT_SEEKS, shrink_dqcache_memory);
1853
1854         return 0;
1855 }
1856 module_init(dquot_init);
1857
1858 EXPORT_SYMBOL(register_quota_format);
1859 EXPORT_SYMBOL(unregister_quota_format);
1860 EXPORT_SYMBOL(dqstats);
1861 EXPORT_SYMBOL(dq_data_lock);
1862 EXPORT_SYMBOL(vfs_quota_on);
1863 EXPORT_SYMBOL(vfs_quota_on_mount);
1864 EXPORT_SYMBOL(vfs_quota_off);
1865 EXPORT_SYMBOL(vfs_quota_sync);
1866 EXPORT_SYMBOL(vfs_get_dqinfo);
1867 EXPORT_SYMBOL(vfs_set_dqinfo);
1868 EXPORT_SYMBOL(vfs_get_dqblk);
1869 EXPORT_SYMBOL(vfs_set_dqblk);
1870 EXPORT_SYMBOL(dquot_commit);
1871 EXPORT_SYMBOL(dquot_commit_info);
1872 EXPORT_SYMBOL(dquot_acquire);
1873 EXPORT_SYMBOL(dquot_release);
1874 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
1875 EXPORT_SYMBOL(dquot_initialize);
1876 EXPORT_SYMBOL(dquot_drop);
1877 EXPORT_SYMBOL(dquot_alloc_space);
1878 EXPORT_SYMBOL(dquot_alloc_inode);
1879 EXPORT_SYMBOL(dquot_free_space);
1880 EXPORT_SYMBOL(dquot_free_inode);
1881 EXPORT_SYMBOL(dquot_transfer);