[PATCH] Fix oops in invalidate_dquots()
[linux-2.6.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_sem.
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 semaphore. 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 > dqio_sem
122  * i_mutex on quota files is special (it's below dqio_sem)
123  */
124
125 static DEFINE_SPINLOCK(dq_list_lock);
126 DEFINE_SPINLOCK(dq_data_lock);
127
128 static char *quotatypes[] = INITQFNAMES;
129 static struct quota_format_type *quota_formats; /* List of registered formats */
130 static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
131
132 /* SLAB cache for dquot structures */
133 static kmem_cache_t *dquot_cachep;
134
135 int register_quota_format(struct quota_format_type *fmt)
136 {
137         spin_lock(&dq_list_lock);
138         fmt->qf_next = quota_formats;
139         quota_formats = fmt;
140         spin_unlock(&dq_list_lock);
141         return 0;
142 }
143
144 void unregister_quota_format(struct quota_format_type *fmt)
145 {
146         struct quota_format_type **actqf;
147
148         spin_lock(&dq_list_lock);
149         for (actqf = &quota_formats; *actqf && *actqf != fmt; actqf = &(*actqf)->qf_next);
150         if (*actqf)
151                 *actqf = (*actqf)->qf_next;
152         spin_unlock(&dq_list_lock);
153 }
154
155 static struct quota_format_type *find_quota_format(int id)
156 {
157         struct quota_format_type *actqf;
158
159         spin_lock(&dq_list_lock);
160         for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id; actqf = actqf->qf_next);
161         if (!actqf || !try_module_get(actqf->qf_owner)) {
162                 int qm;
163
164                 spin_unlock(&dq_list_lock);
165                 
166                 for (qm = 0; module_names[qm].qm_fmt_id && module_names[qm].qm_fmt_id != id; qm++);
167                 if (!module_names[qm].qm_fmt_id || request_module(module_names[qm].qm_mod_name))
168                         return NULL;
169
170                 spin_lock(&dq_list_lock);
171                 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id; actqf = actqf->qf_next);
172                 if (actqf && !try_module_get(actqf->qf_owner))
173                         actqf = NULL;
174         }
175         spin_unlock(&dq_list_lock);
176         return actqf;
177 }
178
179 static void put_quota_format(struct quota_format_type *fmt)
180 {
181         module_put(fmt->qf_owner);
182 }
183
184 /*
185  * Dquot List Management:
186  * The quota code uses three lists for dquot management: the inuse_list,
187  * free_dquots, and dquot_hash[] array. A single dquot structure may be
188  * on all three lists, depending on its current state.
189  *
190  * All dquots are placed to the end of inuse_list when first created, and this
191  * list is used for invalidate operation, which must look at every dquot.
192  *
193  * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
194  * and this list is searched whenever we need an available dquot.  Dquots are
195  * removed from the list as soon as they are used again, and
196  * dqstats.free_dquots gives the number of dquots on the list. When
197  * dquot is invalidated it's completely released from memory.
198  *
199  * Dquots with a specific identity (device, type and id) are placed on
200  * one of the dquot_hash[] hash chains. The provides an efficient search
201  * mechanism to locate a specific dquot.
202  */
203
204 static LIST_HEAD(inuse_list);
205 static LIST_HEAD(free_dquots);
206 static unsigned int dq_hash_bits, dq_hash_mask;
207 static struct hlist_head *dquot_hash;
208
209 struct dqstats dqstats;
210
211 static void dqput(struct dquot *dquot);
212
213 static inline unsigned int
214 hashfn(const struct super_block *sb, unsigned int id, int type)
215 {
216         unsigned long tmp;
217
218         tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
219         return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
220 }
221
222 /*
223  * Following list functions expect dq_list_lock to be held
224  */
225 static inline void insert_dquot_hash(struct dquot *dquot)
226 {
227         struct hlist_head *head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id, dquot->dq_type);
228         hlist_add_head(&dquot->dq_hash, head);
229 }
230
231 static inline void remove_dquot_hash(struct dquot *dquot)
232 {
233         hlist_del_init(&dquot->dq_hash);
234 }
235
236 static inline struct dquot *find_dquot(unsigned int hashent, struct super_block *sb, unsigned int id, int type)
237 {
238         struct hlist_node *node;
239         struct dquot *dquot;
240
241         hlist_for_each (node, dquot_hash+hashent) {
242                 dquot = hlist_entry(node, struct dquot, dq_hash);
243                 if (dquot->dq_sb == sb && dquot->dq_id == id && dquot->dq_type == type)
244                         return dquot;
245         }
246         return NODQUOT;
247 }
248
249 /* Add a dquot to the tail of the free list */
250 static inline void put_dquot_last(struct dquot *dquot)
251 {
252         list_add(&dquot->dq_free, free_dquots.prev);
253         dqstats.free_dquots++;
254 }
255
256 static inline void remove_free_dquot(struct dquot *dquot)
257 {
258         if (list_empty(&dquot->dq_free))
259                 return;
260         list_del_init(&dquot->dq_free);
261         dqstats.free_dquots--;
262 }
263
264 static inline void put_inuse(struct dquot *dquot)
265 {
266         /* We add to the back of inuse list so we don't have to restart
267          * when traversing this list and we block */
268         list_add(&dquot->dq_inuse, inuse_list.prev);
269         dqstats.allocated_dquots++;
270 }
271
272 static inline void remove_inuse(struct dquot *dquot)
273 {
274         dqstats.allocated_dquots--;
275         list_del(&dquot->dq_inuse);
276 }
277 /*
278  * End of list functions needing dq_list_lock
279  */
280
281 static void wait_on_dquot(struct dquot *dquot)
282 {
283         down(&dquot->dq_lock);
284         up(&dquot->dq_lock);
285 }
286
287 #define mark_dquot_dirty(dquot) ((dquot)->dq_sb->dq_op->mark_dirty(dquot))
288
289 int dquot_mark_dquot_dirty(struct dquot *dquot)
290 {
291         spin_lock(&dq_list_lock);
292         if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags))
293                 list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
294                                 info[dquot->dq_type].dqi_dirty_list);
295         spin_unlock(&dq_list_lock);
296         return 0;
297 }
298
299 /* This function needs dq_list_lock */
300 static inline int clear_dquot_dirty(struct dquot *dquot)
301 {
302         if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags))
303                 return 0;
304         list_del_init(&dquot->dq_dirty);
305         return 1;
306 }
307
308 void mark_info_dirty(struct super_block *sb, int type)
309 {
310         set_bit(DQF_INFO_DIRTY_B, &sb_dqopt(sb)->info[type].dqi_flags);
311 }
312 EXPORT_SYMBOL(mark_info_dirty);
313
314 /*
315  *      Read dquot from disk and alloc space for it
316  */
317
318 int dquot_acquire(struct dquot *dquot)
319 {
320         int ret = 0, ret2 = 0;
321         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
322
323         down(&dquot->dq_lock);
324         down(&dqopt->dqio_sem);
325         if (!test_bit(DQ_READ_B, &dquot->dq_flags))
326                 ret = dqopt->ops[dquot->dq_type]->read_dqblk(dquot);
327         if (ret < 0)
328                 goto out_iolock;
329         set_bit(DQ_READ_B, &dquot->dq_flags);
330         /* Instantiate dquot if needed */
331         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
332                 ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
333                 /* Write the info if needed */
334                 if (info_dirty(&dqopt->info[dquot->dq_type]))
335                         ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
336                 if (ret < 0)
337                         goto out_iolock;
338                 if (ret2 < 0) {
339                         ret = ret2;
340                         goto out_iolock;
341                 }
342         }
343         set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
344 out_iolock:
345         up(&dqopt->dqio_sem);
346         up(&dquot->dq_lock);
347         return ret;
348 }
349
350 /*
351  *      Write dquot to disk
352  */
353 int dquot_commit(struct dquot *dquot)
354 {
355         int ret = 0, ret2 = 0;
356         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
357
358         down(&dqopt->dqio_sem);
359         spin_lock(&dq_list_lock);
360         if (!clear_dquot_dirty(dquot)) {
361                 spin_unlock(&dq_list_lock);
362                 goto out_sem;
363         }
364         spin_unlock(&dq_list_lock);
365         /* Inactive dquot can be only if there was error during read/init
366          * => we have better not writing it */
367         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
368                 ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
369                 if (info_dirty(&dqopt->info[dquot->dq_type]))
370                         ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
371                 if (ret >= 0)
372                         ret = ret2;
373         }
374 out_sem:
375         up(&dqopt->dqio_sem);
376         return ret;
377 }
378
379 /*
380  *      Release dquot
381  */
382 int dquot_release(struct dquot *dquot)
383 {
384         int ret = 0, ret2 = 0;
385         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
386
387         down(&dquot->dq_lock);
388         /* Check whether we are not racing with some other dqget() */
389         if (atomic_read(&dquot->dq_count) > 1)
390                 goto out_dqlock;
391         down(&dqopt->dqio_sem);
392         if (dqopt->ops[dquot->dq_type]->release_dqblk) {
393                 ret = dqopt->ops[dquot->dq_type]->release_dqblk(dquot);
394                 /* Write the info */
395                 if (info_dirty(&dqopt->info[dquot->dq_type]))
396                         ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
397                 if (ret >= 0)
398                         ret = ret2;
399         }
400         clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
401         up(&dqopt->dqio_sem);
402 out_dqlock:
403         up(&dquot->dq_lock);
404         return ret;
405 }
406
407 /* Invalidate all dquots on the list. Note that this function is called after
408  * quota is disabled and pointers from inodes removed so there cannot be new
409  * quota users. There can still be some users of quotas due to inodes being
410  * just deleted or pruned by prune_icache() (those are not attached to any
411  * list). We have to wait for such users.
412  */
413 static void invalidate_dquots(struct super_block *sb, int type)
414 {
415         struct dquot *dquot, *tmp;
416
417 restart:
418         spin_lock(&dq_list_lock);
419         list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
420                 if (dquot->dq_sb != sb)
421                         continue;
422                 if (dquot->dq_type != type)
423                         continue;
424                 /* Wait for dquot users */
425                 if (atomic_read(&dquot->dq_count)) {
426                         DEFINE_WAIT(wait);
427
428                         atomic_inc(&dquot->dq_count);
429                         prepare_to_wait(&dquot->dq_wait_unused, &wait,
430                                         TASK_UNINTERRUPTIBLE);
431                         spin_unlock(&dq_list_lock);
432                         /* Once dqput() wakes us up, we know it's time to free
433                          * the dquot.
434                          * IMPORTANT: we rely on the fact that there is always
435                          * at most one process waiting for dquot to free.
436                          * Otherwise dq_count would be > 1 and we would never
437                          * wake up.
438                          */
439                         if (atomic_read(&dquot->dq_count) > 1)
440                                 schedule();
441                         finish_wait(&dquot->dq_wait_unused, &wait);
442                         dqput(dquot);
443                         /* At this moment dquot() need not exist (it could be
444                          * reclaimed by prune_dqcache(). Hence we must
445                          * restart. */
446                         goto restart;
447                 }
448                 /*
449                  * Quota now has no users and it has been written on last
450                  * dqput()
451                  */
452                 remove_dquot_hash(dquot);
453                 remove_free_dquot(dquot);
454                 remove_inuse(dquot);
455                 kmem_cache_free(dquot_cachep, dquot);
456         }
457         spin_unlock(&dq_list_lock);
458 }
459
460 int vfs_quota_sync(struct super_block *sb, int type)
461 {
462         struct list_head *dirty;
463         struct dquot *dquot;
464         struct quota_info *dqopt = sb_dqopt(sb);
465         int cnt;
466
467         down(&dqopt->dqonoff_sem);
468         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
469                 if (type != -1 && cnt != type)
470                         continue;
471                 if (!sb_has_quota_enabled(sb, cnt))
472                         continue;
473                 spin_lock(&dq_list_lock);
474                 dirty = &dqopt->info[cnt].dqi_dirty_list;
475                 while (!list_empty(dirty)) {
476                         dquot = list_entry(dirty->next, struct dquot, dq_dirty);
477                         /* Dirty and inactive can be only bad dquot... */
478                         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
479                                 clear_dquot_dirty(dquot);
480                                 continue;
481                         }
482                         /* Now we have active dquot from which someone is
483                          * holding reference so we can safely just increase
484                          * use count */
485                         atomic_inc(&dquot->dq_count);
486                         dqstats.lookups++;
487                         spin_unlock(&dq_list_lock);
488                         sb->dq_op->write_dquot(dquot);
489                         dqput(dquot);
490                         spin_lock(&dq_list_lock);
491                 }
492                 spin_unlock(&dq_list_lock);
493         }
494
495         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
496                 if ((cnt == type || type == -1) && sb_has_quota_enabled(sb, cnt)
497                         && info_dirty(&dqopt->info[cnt]))
498                         sb->dq_op->write_info(sb, cnt);
499         spin_lock(&dq_list_lock);
500         dqstats.syncs++;
501         spin_unlock(&dq_list_lock);
502         up(&dqopt->dqonoff_sem);
503
504         return 0;
505 }
506
507 /* Free unused dquots from cache */
508 static void prune_dqcache(int count)
509 {
510         struct list_head *head;
511         struct dquot *dquot;
512
513         head = free_dquots.prev;
514         while (head != &free_dquots && count) {
515                 dquot = list_entry(head, struct dquot, dq_free);
516                 remove_dquot_hash(dquot);
517                 remove_free_dquot(dquot);
518                 remove_inuse(dquot);
519                 kmem_cache_free(dquot_cachep, dquot);
520                 count--;
521                 head = free_dquots.prev;
522         }
523 }
524
525 /*
526  * This is called from kswapd when we think we need some
527  * more memory
528  */
529
530 static int shrink_dqcache_memory(int nr, gfp_t gfp_mask)
531 {
532         if (nr) {
533                 spin_lock(&dq_list_lock);
534                 prune_dqcache(nr);
535                 spin_unlock(&dq_list_lock);
536         }
537         return (dqstats.free_dquots / 100) * sysctl_vfs_cache_pressure;
538 }
539
540 /*
541  * Put reference to dquot
542  * NOTE: If you change this function please check whether dqput_blocks() works right...
543  * MUST be called with either dqptr_sem or dqonoff_sem held
544  */
545 static void dqput(struct dquot *dquot)
546 {
547         if (!dquot)
548                 return;
549 #ifdef __DQUOT_PARANOIA
550         if (!atomic_read(&dquot->dq_count)) {
551                 printk("VFS: dqput: trying to free free dquot\n");
552                 printk("VFS: device %s, dquot of %s %d\n",
553                         dquot->dq_sb->s_id,
554                         quotatypes[dquot->dq_type],
555                         dquot->dq_id);
556                 BUG();
557         }
558 #endif
559         
560         spin_lock(&dq_list_lock);
561         dqstats.drops++;
562         spin_unlock(&dq_list_lock);
563 we_slept:
564         spin_lock(&dq_list_lock);
565         if (atomic_read(&dquot->dq_count) > 1) {
566                 /* We have more than one user... nothing to do */
567                 atomic_dec(&dquot->dq_count);
568                 /* Releasing dquot during quotaoff phase? */
569                 if (!sb_has_quota_enabled(dquot->dq_sb, dquot->dq_type) &&
570                     atomic_read(&dquot->dq_count) == 1)
571                         wake_up(&dquot->dq_wait_unused);
572                 spin_unlock(&dq_list_lock);
573                 return;
574         }
575         /* Need to release dquot? */
576         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && dquot_dirty(dquot)) {
577                 spin_unlock(&dq_list_lock);
578                 /* Commit dquot before releasing */
579                 dquot->dq_sb->dq_op->write_dquot(dquot);
580                 goto we_slept;
581         }
582         /* Clear flag in case dquot was inactive (something bad happened) */
583         clear_dquot_dirty(dquot);
584         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
585                 spin_unlock(&dq_list_lock);
586                 dquot->dq_sb->dq_op->release_dquot(dquot);
587                 goto we_slept;
588         }
589         atomic_dec(&dquot->dq_count);
590 #ifdef __DQUOT_PARANOIA
591         /* sanity check */
592         if (!list_empty(&dquot->dq_free))
593                 BUG();
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, SLAB_NOFS);
604         if(!dquot)
605                 return NODQUOT;
606
607         memset((caddr_t)dquot, 0, sizeof(struct dquot));
608         sema_init(&dquot->dq_lock, 1);
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_sem 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         if (!dquot->dq_sb)      /* Has somebody invalidated entry under us? */
669                 BUG();
670 #endif
671
672         return dquot;
673 }
674
675 static int dqinit_needed(struct inode *inode, int type)
676 {
677         int cnt;
678
679         if (IS_NOQUOTA(inode))
680                 return 0;
681         if (type != -1)
682                 return inode->i_dquot[type] == NODQUOT;
683         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
684                 if (inode->i_dquot[cnt] == NODQUOT)
685                         return 1;
686         return 0;
687 }
688
689 /* This routine is guarded by dqonoff_sem semaphore */
690 static void add_dquot_ref(struct super_block *sb, int type)
691 {
692         struct list_head *p;
693
694 restart:
695         file_list_lock();
696         list_for_each(p, &sb->s_files) {
697                 struct file *filp = list_entry(p, struct file, f_u.fu_list);
698                 struct inode *inode = filp->f_dentry->d_inode;
699                 if (filp->f_mode & FMODE_WRITE && dqinit_needed(inode, type)) {
700                         struct dentry *dentry = dget(filp->f_dentry);
701                         file_list_unlock();
702                         sb->dq_op->initialize(inode, type);
703                         dput(dentry);
704                         /* As we may have blocked we had better restart... */
705                         goto restart;
706                 }
707         }
708         file_list_unlock();
709 }
710
711 /* Return 0 if dqput() won't block (note that 1 doesn't necessarily mean blocking) */
712 static inline int dqput_blocks(struct dquot *dquot)
713 {
714         if (atomic_read(&dquot->dq_count) <= 1)
715                 return 1;
716         return 0;
717 }
718
719 /* Remove references to dquots from inode - add dquot to list for freeing if needed */
720 /* We can't race with anybody because we hold dqptr_sem for writing... */
721 int remove_inode_dquot_ref(struct inode *inode, int type, struct list_head *tofree_head)
722 {
723         struct dquot *dquot = inode->i_dquot[type];
724
725         inode->i_dquot[type] = NODQUOT;
726         if (dquot != NODQUOT) {
727                 if (dqput_blocks(dquot)) {
728 #ifdef __DQUOT_PARANOIA
729                         if (atomic_read(&dquot->dq_count) != 1)
730                                 printk(KERN_WARNING "VFS: Adding dquot with dq_count %d to dispose list.\n", atomic_read(&dquot->dq_count));
731 #endif
732                         spin_lock(&dq_list_lock);
733                         list_add(&dquot->dq_free, tofree_head); /* As dquot must have currently users it can't be on the free list... */
734                         spin_unlock(&dq_list_lock);
735                         return 1;
736                 }
737                 else
738                         dqput(dquot);   /* We have guaranteed we won't block */
739         }
740         return 0;
741 }
742
743 /* Free list of dquots - called from inode.c */
744 /* dquots are removed from inodes, no new references can be got so we are the only ones holding reference */
745 static void put_dquot_list(struct list_head *tofree_head)
746 {
747         struct list_head *act_head;
748         struct dquot *dquot;
749
750         act_head = tofree_head->next;
751         /* So now we have dquots on the list... Just free them */
752         while (act_head != tofree_head) {
753                 dquot = list_entry(act_head, struct dquot, dq_free);
754                 act_head = act_head->next;
755                 list_del_init(&dquot->dq_free); /* Remove dquot from the list so we won't have problems... */
756                 dqput(dquot);
757         }
758 }
759
760 /* Gather all references from inodes and drop them */
761 static void drop_dquot_ref(struct super_block *sb, int type)
762 {
763         LIST_HEAD(tofree_head);
764
765         down_write(&sb_dqopt(sb)->dqptr_sem);
766         remove_dquot_ref(sb, type, &tofree_head);
767         up_write(&sb_dqopt(sb)->dqptr_sem);
768         put_dquot_list(&tofree_head);
769 }
770
771 static inline void dquot_incr_inodes(struct dquot *dquot, unsigned long number)
772 {
773         dquot->dq_dqb.dqb_curinodes += number;
774 }
775
776 static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
777 {
778         dquot->dq_dqb.dqb_curspace += number;
779 }
780
781 static inline void dquot_decr_inodes(struct dquot *dquot, unsigned long number)
782 {
783         if (dquot->dq_dqb.dqb_curinodes > number)
784                 dquot->dq_dqb.dqb_curinodes -= number;
785         else
786                 dquot->dq_dqb.dqb_curinodes = 0;
787         if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
788                 dquot->dq_dqb.dqb_itime = (time_t) 0;
789         clear_bit(DQ_INODES_B, &dquot->dq_flags);
790 }
791
792 static inline void dquot_decr_space(struct dquot *dquot, qsize_t number)
793 {
794         if (dquot->dq_dqb.dqb_curspace > number)
795                 dquot->dq_dqb.dqb_curspace -= number;
796         else
797                 dquot->dq_dqb.dqb_curspace = 0;
798         if (toqb(dquot->dq_dqb.dqb_curspace) <= dquot->dq_dqb.dqb_bsoftlimit)
799                 dquot->dq_dqb.dqb_btime = (time_t) 0;
800         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
801 }
802
803 static int flag_print_warnings = 1;
804
805 static inline int need_print_warning(struct dquot *dquot)
806 {
807         if (!flag_print_warnings)
808                 return 0;
809
810         switch (dquot->dq_type) {
811                 case USRQUOTA:
812                         return current->fsuid == dquot->dq_id;
813                 case GRPQUOTA:
814                         return in_group_p(dquot->dq_id);
815         }
816         return 0;
817 }
818
819 /* Values of warnings */
820 #define NOWARN 0
821 #define IHARDWARN 1
822 #define ISOFTLONGWARN 2
823 #define ISOFTWARN 3
824 #define BHARDWARN 4
825 #define BSOFTLONGWARN 5
826 #define BSOFTWARN 6
827
828 /* Print warning to user which exceeded quota */
829 static void print_warning(struct dquot *dquot, const char warntype)
830 {
831         char *msg = NULL;
832         int flag = (warntype == BHARDWARN || warntype == BSOFTLONGWARN) ? DQ_BLKS_B :
833           ((warntype == IHARDWARN || warntype == ISOFTLONGWARN) ? DQ_INODES_B : 0);
834
835         if (!need_print_warning(dquot) || (flag && test_and_set_bit(flag, &dquot->dq_flags)))
836                 return;
837
838         tty_write_message(current->signal->tty, dquot->dq_sb->s_id);
839         if (warntype == ISOFTWARN || warntype == BSOFTWARN)
840                 tty_write_message(current->signal->tty, ": warning, ");
841         else
842                 tty_write_message(current->signal->tty, ": write failed, ");
843         tty_write_message(current->signal->tty, quotatypes[dquot->dq_type]);
844         switch (warntype) {
845                 case IHARDWARN:
846                         msg = " file limit reached.\r\n";
847                         break;
848                 case ISOFTLONGWARN:
849                         msg = " file quota exceeded too long.\r\n";
850                         break;
851                 case ISOFTWARN:
852                         msg = " file quota exceeded.\r\n";
853                         break;
854                 case BHARDWARN:
855                         msg = " block limit reached.\r\n";
856                         break;
857                 case BSOFTLONGWARN:
858                         msg = " block quota exceeded too long.\r\n";
859                         break;
860                 case BSOFTWARN:
861                         msg = " block quota exceeded.\r\n";
862                         break;
863         }
864         tty_write_message(current->signal->tty, msg);
865 }
866
867 static inline void flush_warnings(struct dquot **dquots, char *warntype)
868 {
869         int i;
870
871         for (i = 0; i < MAXQUOTAS; i++)
872                 if (dquots[i] != NODQUOT && warntype[i] != NOWARN)
873                         print_warning(dquots[i], warntype[i]);
874 }
875
876 static inline char ignore_hardlimit(struct dquot *dquot)
877 {
878         struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_type];
879
880         return capable(CAP_SYS_RESOURCE) &&
881             (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD || !(info->dqi_flags & V1_DQF_RSQUASH));
882 }
883
884 /* needs dq_data_lock */
885 static int check_idq(struct dquot *dquot, ulong inodes, char *warntype)
886 {
887         *warntype = NOWARN;
888         if (inodes <= 0 || test_bit(DQ_FAKE_B, &dquot->dq_flags))
889                 return QUOTA_OK;
890
891         if (dquot->dq_dqb.dqb_ihardlimit &&
892            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_ihardlimit &&
893             !ignore_hardlimit(dquot)) {
894                 *warntype = IHARDWARN;
895                 return NO_QUOTA;
896         }
897
898         if (dquot->dq_dqb.dqb_isoftlimit &&
899            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
900             dquot->dq_dqb.dqb_itime && get_seconds() >= dquot->dq_dqb.dqb_itime &&
901             !ignore_hardlimit(dquot)) {
902                 *warntype = ISOFTLONGWARN;
903                 return NO_QUOTA;
904         }
905
906         if (dquot->dq_dqb.dqb_isoftlimit &&
907            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
908             dquot->dq_dqb.dqb_itime == 0) {
909                 *warntype = ISOFTWARN;
910                 dquot->dq_dqb.dqb_itime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
911         }
912
913         return QUOTA_OK;
914 }
915
916 /* needs dq_data_lock */
917 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc, char *warntype)
918 {
919         *warntype = 0;
920         if (space <= 0 || test_bit(DQ_FAKE_B, &dquot->dq_flags))
921                 return QUOTA_OK;
922
923         if (dquot->dq_dqb.dqb_bhardlimit &&
924            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bhardlimit &&
925             !ignore_hardlimit(dquot)) {
926                 if (!prealloc)
927                         *warntype = BHARDWARN;
928                 return NO_QUOTA;
929         }
930
931         if (dquot->dq_dqb.dqb_bsoftlimit &&
932            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bsoftlimit &&
933             dquot->dq_dqb.dqb_btime && get_seconds() >= dquot->dq_dqb.dqb_btime &&
934             !ignore_hardlimit(dquot)) {
935                 if (!prealloc)
936                         *warntype = BSOFTLONGWARN;
937                 return NO_QUOTA;
938         }
939
940         if (dquot->dq_dqb.dqb_bsoftlimit &&
941            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bsoftlimit &&
942             dquot->dq_dqb.dqb_btime == 0) {
943                 if (!prealloc) {
944                         *warntype = BSOFTWARN;
945                         dquot->dq_dqb.dqb_btime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_bgrace;
946                 }
947                 else
948                         /*
949                          * We don't allow preallocation to exceed softlimit so exceeding will
950                          * be always printed
951                          */
952                         return NO_QUOTA;
953         }
954
955         return QUOTA_OK;
956 }
957
958 /*
959  *      Initialize quota pointers in inode
960  *      Transaction must be started at entry
961  */
962 int dquot_initialize(struct inode *inode, int type)
963 {
964         unsigned int id = 0;
965         int cnt, ret = 0;
966
967         /* First test before acquiring semaphore - solves deadlocks when we
968          * re-enter the quota code and are already holding the semaphore */
969         if (IS_NOQUOTA(inode))
970                 return 0;
971         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
972         /* Having dqptr_sem we know NOQUOTA flags can't be altered... */
973         if (IS_NOQUOTA(inode))
974                 goto out_err;
975         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
976                 if (type != -1 && cnt != type)
977                         continue;
978                 if (inode->i_dquot[cnt] == NODQUOT) {
979                         switch (cnt) {
980                                 case USRQUOTA:
981                                         id = inode->i_uid;
982                                         break;
983                                 case GRPQUOTA:
984                                         id = inode->i_gid;
985                                         break;
986                         }
987                         inode->i_dquot[cnt] = dqget(inode->i_sb, id, cnt);
988                 }
989         }
990 out_err:
991         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
992         return ret;
993 }
994
995 /*
996  *      Release all quotas referenced by inode
997  *      Transaction must be started at an entry
998  */
999 int dquot_drop(struct inode *inode)
1000 {
1001         int cnt;
1002
1003         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1004         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1005                 if (inode->i_dquot[cnt] != NODQUOT) {
1006                         dqput(inode->i_dquot[cnt]);
1007                         inode->i_dquot[cnt] = NODQUOT;
1008                 }
1009         }
1010         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1011         return 0;
1012 }
1013
1014 /*
1015  * Following four functions update i_blocks+i_bytes fields and
1016  * quota information (together with appropriate checks)
1017  * NOTE: We absolutely rely on the fact that caller dirties
1018  * the inode (usually macros in quotaops.h care about this) and
1019  * holds a handle for the current transaction so that dquot write and
1020  * inode write go into the same transaction.
1021  */
1022
1023 /*
1024  * This operation can block, but only after everything is updated
1025  */
1026 int dquot_alloc_space(struct inode *inode, qsize_t number, int warn)
1027 {
1028         int cnt, ret = NO_QUOTA;
1029         char warntype[MAXQUOTAS];
1030
1031         /* First test before acquiring semaphore - solves deadlocks when we
1032          * re-enter the quota code and are already holding the semaphore */
1033         if (IS_NOQUOTA(inode)) {
1034 out_add:
1035                 inode_add_bytes(inode, number);
1036                 return QUOTA_OK;
1037         }
1038         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1039                 warntype[cnt] = NOWARN;
1040
1041         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1042         if (IS_NOQUOTA(inode)) {        /* Now we can do reliable test... */
1043                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1044                 goto out_add;
1045         }
1046         spin_lock(&dq_data_lock);
1047         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1048                 if (inode->i_dquot[cnt] == NODQUOT)
1049                         continue;
1050                 if (check_bdq(inode->i_dquot[cnt], number, warn, warntype+cnt) == NO_QUOTA)
1051                         goto warn_put_all;
1052         }
1053         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1054                 if (inode->i_dquot[cnt] == NODQUOT)
1055                         continue;
1056                 dquot_incr_space(inode->i_dquot[cnt], number);
1057         }
1058         inode_add_bytes(inode, number);
1059         ret = QUOTA_OK;
1060 warn_put_all:
1061         spin_unlock(&dq_data_lock);
1062         if (ret == QUOTA_OK)
1063                 /* Dirtify all the dquots - this can block when journalling */
1064                 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1065                         if (inode->i_dquot[cnt])
1066                                 mark_dquot_dirty(inode->i_dquot[cnt]);
1067         flush_warnings(inode->i_dquot, warntype);
1068         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1069         return ret;
1070 }
1071
1072 /*
1073  * This operation can block, but only after everything is updated
1074  */
1075 int dquot_alloc_inode(const struct inode *inode, unsigned long number)
1076 {
1077         int cnt, ret = NO_QUOTA;
1078         char warntype[MAXQUOTAS];
1079
1080         /* First test before acquiring semaphore - solves deadlocks when we
1081          * re-enter the quota code and are already holding the semaphore */
1082         if (IS_NOQUOTA(inode))
1083                 return QUOTA_OK;
1084         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1085                 warntype[cnt] = NOWARN;
1086         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1087         if (IS_NOQUOTA(inode)) {
1088                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1089                 return QUOTA_OK;
1090         }
1091         spin_lock(&dq_data_lock);
1092         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1093                 if (inode->i_dquot[cnt] == NODQUOT)
1094                         continue;
1095                 if (check_idq(inode->i_dquot[cnt], number, warntype+cnt) == NO_QUOTA)
1096                         goto warn_put_all;
1097         }
1098
1099         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1100                 if (inode->i_dquot[cnt] == NODQUOT)
1101                         continue;
1102                 dquot_incr_inodes(inode->i_dquot[cnt], number);
1103         }
1104         ret = QUOTA_OK;
1105 warn_put_all:
1106         spin_unlock(&dq_data_lock);
1107         if (ret == QUOTA_OK)
1108                 /* Dirtify all the dquots - this can block when journalling */
1109                 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1110                         if (inode->i_dquot[cnt])
1111                                 mark_dquot_dirty(inode->i_dquot[cnt]);
1112         flush_warnings((struct dquot **)inode->i_dquot, warntype);
1113         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1114         return ret;
1115 }
1116
1117 /*
1118  * This operation can block, but only after everything is updated
1119  */
1120 int dquot_free_space(struct inode *inode, qsize_t number)
1121 {
1122         unsigned int cnt;
1123
1124         /* First test before acquiring semaphore - solves deadlocks when we
1125          * re-enter the quota code and are already holding the semaphore */
1126         if (IS_NOQUOTA(inode)) {
1127 out_sub:
1128                 inode_sub_bytes(inode, number);
1129                 return QUOTA_OK;
1130         }
1131         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1132         /* Now recheck reliably when holding dqptr_sem */
1133         if (IS_NOQUOTA(inode)) {
1134                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1135                 goto out_sub;
1136         }
1137         spin_lock(&dq_data_lock);
1138         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1139                 if (inode->i_dquot[cnt] == NODQUOT)
1140                         continue;
1141                 dquot_decr_space(inode->i_dquot[cnt], number);
1142         }
1143         inode_sub_bytes(inode, number);
1144         spin_unlock(&dq_data_lock);
1145         /* Dirtify all the dquots - this can block when journalling */
1146         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1147                 if (inode->i_dquot[cnt])
1148                         mark_dquot_dirty(inode->i_dquot[cnt]);
1149         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1150         return QUOTA_OK;
1151 }
1152
1153 /*
1154  * This operation can block, but only after everything is updated
1155  */
1156 int dquot_free_inode(const struct inode *inode, unsigned long number)
1157 {
1158         unsigned int cnt;
1159
1160         /* First test before acquiring semaphore - solves deadlocks when we
1161          * re-enter the quota code and are already holding the semaphore */
1162         if (IS_NOQUOTA(inode))
1163                 return QUOTA_OK;
1164         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1165         /* Now recheck reliably when holding dqptr_sem */
1166         if (IS_NOQUOTA(inode)) {
1167                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1168                 return QUOTA_OK;
1169         }
1170         spin_lock(&dq_data_lock);
1171         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1172                 if (inode->i_dquot[cnt] == NODQUOT)
1173                         continue;
1174                 dquot_decr_inodes(inode->i_dquot[cnt], number);
1175         }
1176         spin_unlock(&dq_data_lock);
1177         /* Dirtify all the dquots - this can block when journalling */
1178         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1179                 if (inode->i_dquot[cnt])
1180                         mark_dquot_dirty(inode->i_dquot[cnt]);
1181         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1182         return QUOTA_OK;
1183 }
1184
1185 /*
1186  * Transfer the number of inode and blocks from one diskquota to an other.
1187  *
1188  * This operation can block, but only after everything is updated
1189  * A transaction must be started when entering this function.
1190  */
1191 int dquot_transfer(struct inode *inode, struct iattr *iattr)
1192 {
1193         qsize_t space;
1194         struct dquot *transfer_from[MAXQUOTAS];
1195         struct dquot *transfer_to[MAXQUOTAS];
1196         int cnt, ret = NO_QUOTA, chuid = (iattr->ia_valid & ATTR_UID) && inode->i_uid != iattr->ia_uid,
1197             chgid = (iattr->ia_valid & ATTR_GID) && inode->i_gid != iattr->ia_gid;
1198         char warntype[MAXQUOTAS];
1199
1200         /* First test before acquiring semaphore - solves deadlocks when we
1201          * re-enter the quota code and are already holding the semaphore */
1202         if (IS_NOQUOTA(inode))
1203                 return QUOTA_OK;
1204         /* Clear the arrays */
1205         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1206                 transfer_to[cnt] = transfer_from[cnt] = NODQUOT;
1207                 warntype[cnt] = NOWARN;
1208         }
1209         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1210         /* Now recheck reliably when holding dqptr_sem */
1211         if (IS_NOQUOTA(inode)) {        /* File without quota accounting? */
1212                 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1213                 return QUOTA_OK;
1214         }
1215         /* First build the transfer_to list - here we can block on
1216          * reading/instantiating of dquots.  We know that the transaction for
1217          * us was already started so we don't violate lock ranking here */
1218         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1219                 switch (cnt) {
1220                         case USRQUOTA:
1221                                 if (!chuid)
1222                                         continue;
1223                                 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_uid, cnt);
1224                                 break;
1225                         case GRPQUOTA:
1226                                 if (!chgid)
1227                                         continue;
1228                                 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_gid, cnt);
1229                                 break;
1230                 }
1231         }
1232         spin_lock(&dq_data_lock);
1233         space = inode_get_bytes(inode);
1234         /* Build the transfer_from list and check the limits */
1235         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1236                 if (transfer_to[cnt] == NODQUOT)
1237                         continue;
1238                 transfer_from[cnt] = inode->i_dquot[cnt];
1239                 if (check_idq(transfer_to[cnt], 1, warntype+cnt) == NO_QUOTA ||
1240                     check_bdq(transfer_to[cnt], space, 0, warntype+cnt) == NO_QUOTA)
1241                         goto warn_put_all;
1242         }
1243
1244         /*
1245          * Finally perform the needed transfer from transfer_from to transfer_to
1246          */
1247         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1248                 /*
1249                  * Skip changes for same uid or gid or for turned off quota-type.
1250                  */
1251                 if (transfer_to[cnt] == NODQUOT)
1252                         continue;
1253
1254                 /* Due to IO error we might not have transfer_from[] structure */
1255                 if (transfer_from[cnt]) {
1256                         dquot_decr_inodes(transfer_from[cnt], 1);
1257                         dquot_decr_space(transfer_from[cnt], space);
1258                 }
1259
1260                 dquot_incr_inodes(transfer_to[cnt], 1);
1261                 dquot_incr_space(transfer_to[cnt], space);
1262
1263                 inode->i_dquot[cnt] = transfer_to[cnt];
1264         }
1265         ret = QUOTA_OK;
1266 warn_put_all:
1267         spin_unlock(&dq_data_lock);
1268         /* Dirtify all the dquots - this can block when journalling */
1269         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1270                 if (transfer_from[cnt])
1271                         mark_dquot_dirty(transfer_from[cnt]);
1272                 if (transfer_to[cnt])
1273                         mark_dquot_dirty(transfer_to[cnt]);
1274         }
1275         flush_warnings(transfer_to, warntype);
1276         
1277         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1278                 if (ret == QUOTA_OK && transfer_from[cnt] != NODQUOT)
1279                         dqput(transfer_from[cnt]);
1280                 if (ret == NO_QUOTA && transfer_to[cnt] != NODQUOT)
1281                         dqput(transfer_to[cnt]);
1282         }
1283         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1284         return ret;
1285 }
1286
1287 /*
1288  * Write info of quota file to disk
1289  */
1290 int dquot_commit_info(struct super_block *sb, int type)
1291 {
1292         int ret;
1293         struct quota_info *dqopt = sb_dqopt(sb);
1294
1295         down(&dqopt->dqio_sem);
1296         ret = dqopt->ops[type]->write_file_info(sb, type);
1297         up(&dqopt->dqio_sem);
1298         return ret;
1299 }
1300
1301 /*
1302  * Definitions of diskquota operations.
1303  */
1304 struct dquot_operations dquot_operations = {
1305         .initialize     = dquot_initialize,
1306         .drop           = dquot_drop,
1307         .alloc_space    = dquot_alloc_space,
1308         .alloc_inode    = dquot_alloc_inode,
1309         .free_space     = dquot_free_space,
1310         .free_inode     = dquot_free_inode,
1311         .transfer       = dquot_transfer,
1312         .write_dquot    = dquot_commit,
1313         .acquire_dquot  = dquot_acquire,
1314         .release_dquot  = dquot_release,
1315         .mark_dirty     = dquot_mark_dquot_dirty,
1316         .write_info     = dquot_commit_info
1317 };
1318
1319 static inline void set_enable_flags(struct quota_info *dqopt, int type)
1320 {
1321         switch (type) {
1322                 case USRQUOTA:
1323                         dqopt->flags |= DQUOT_USR_ENABLED;
1324                         break;
1325                 case GRPQUOTA:
1326                         dqopt->flags |= DQUOT_GRP_ENABLED;
1327                         break;
1328         }
1329 }
1330
1331 static inline void reset_enable_flags(struct quota_info *dqopt, int type)
1332 {
1333         switch (type) {
1334                 case USRQUOTA:
1335                         dqopt->flags &= ~DQUOT_USR_ENABLED;
1336                         break;
1337                 case GRPQUOTA:
1338                         dqopt->flags &= ~DQUOT_GRP_ENABLED;
1339                         break;
1340         }
1341 }
1342
1343 /*
1344  * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
1345  */
1346 int vfs_quota_off(struct super_block *sb, int type)
1347 {
1348         int cnt;
1349         struct quota_info *dqopt = sb_dqopt(sb);
1350         struct inode *toputinode[MAXQUOTAS];
1351
1352         /* We need to serialize quota_off() for device */
1353         down(&dqopt->dqonoff_sem);
1354         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1355                 toputinode[cnt] = NULL;
1356                 if (type != -1 && cnt != type)
1357                         continue;
1358                 if (!sb_has_quota_enabled(sb, cnt))
1359                         continue;
1360                 reset_enable_flags(dqopt, cnt);
1361
1362                 /* Note: these are blocking operations */
1363                 drop_dquot_ref(sb, cnt);
1364                 invalidate_dquots(sb, cnt);
1365                 /*
1366                  * Now all dquots should be invalidated, all writes done so we should be only
1367                  * users of the info. No locks needed.
1368                  */
1369                 if (info_dirty(&dqopt->info[cnt]))
1370                         sb->dq_op->write_info(sb, cnt);
1371                 if (dqopt->ops[cnt]->free_file_info)
1372                         dqopt->ops[cnt]->free_file_info(sb, cnt);
1373                 put_quota_format(dqopt->info[cnt].dqi_format);
1374
1375                 toputinode[cnt] = dqopt->files[cnt];
1376                 dqopt->files[cnt] = NULL;
1377                 dqopt->info[cnt].dqi_flags = 0;
1378                 dqopt->info[cnt].dqi_igrace = 0;
1379                 dqopt->info[cnt].dqi_bgrace = 0;
1380                 dqopt->ops[cnt] = NULL;
1381         }
1382         up(&dqopt->dqonoff_sem);
1383         /* Sync the superblock so that buffers with quota data are written to
1384          * disk (and so userspace sees correct data afterwards). */
1385         if (sb->s_op->sync_fs)
1386                 sb->s_op->sync_fs(sb, 1);
1387         sync_blockdev(sb->s_bdev);
1388         /* Now the quota files are just ordinary files and we can set the
1389          * inode flags back. Moreover we discard the pagecache so that
1390          * userspace sees the writes we did bypassing the pagecache. We
1391          * must also discard the blockdev buffers so that we see the
1392          * changes done by userspace on the next quotaon() */
1393         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1394                 if (toputinode[cnt]) {
1395                         down(&dqopt->dqonoff_sem);
1396                         /* If quota was reenabled in the meantime, we have
1397                          * nothing to do */
1398                         if (!sb_has_quota_enabled(sb, cnt)) {
1399                                 mutex_lock(&toputinode[cnt]->i_mutex);
1400                                 toputinode[cnt]->i_flags &= ~(S_IMMUTABLE |
1401                                   S_NOATIME | S_NOQUOTA);
1402                                 truncate_inode_pages(&toputinode[cnt]->i_data, 0);
1403                                 mutex_unlock(&toputinode[cnt]->i_mutex);
1404                                 mark_inode_dirty(toputinode[cnt]);
1405                                 iput(toputinode[cnt]);
1406                         }
1407                         up(&dqopt->dqonoff_sem);
1408                 }
1409         if (sb->s_bdev)
1410                 invalidate_bdev(sb->s_bdev, 0);
1411         return 0;
1412 }
1413
1414 /*
1415  *      Turn quotas on on a device
1416  */
1417
1418 /* Helper function when we already have the inode */
1419 static int vfs_quota_on_inode(struct inode *inode, int type, int format_id)
1420 {
1421         struct quota_format_type *fmt = find_quota_format(format_id);
1422         struct super_block *sb = inode->i_sb;
1423         struct quota_info *dqopt = sb_dqopt(sb);
1424         int error;
1425         int oldflags = -1;
1426
1427         if (!fmt)
1428                 return -ESRCH;
1429         if (!S_ISREG(inode->i_mode)) {
1430                 error = -EACCES;
1431                 goto out_fmt;
1432         }
1433         if (IS_RDONLY(inode)) {
1434                 error = -EROFS;
1435                 goto out_fmt;
1436         }
1437         if (!sb->s_op->quota_write || !sb->s_op->quota_read) {
1438                 error = -EINVAL;
1439                 goto out_fmt;
1440         }
1441
1442         /* As we bypass the pagecache we must now flush the inode so that
1443          * we see all the changes from userspace... */
1444         write_inode_now(inode, 1);
1445         /* And now flush the block cache so that kernel sees the changes */
1446         invalidate_bdev(sb->s_bdev, 0);
1447         mutex_lock(&inode->i_mutex);
1448         down(&dqopt->dqonoff_sem);
1449         if (sb_has_quota_enabled(sb, type)) {
1450                 error = -EBUSY;
1451                 goto out_lock;
1452         }
1453         /* We don't want quota and atime on quota files (deadlocks possible)
1454          * Also nobody should write to the file - we use special IO operations
1455          * which ignore the immutable bit. */
1456         down_write(&dqopt->dqptr_sem);
1457         oldflags = inode->i_flags & (S_NOATIME | S_IMMUTABLE | S_NOQUOTA);
1458         inode->i_flags |= S_NOQUOTA | S_NOATIME | S_IMMUTABLE;
1459         up_write(&dqopt->dqptr_sem);
1460         sb->dq_op->drop(inode);
1461
1462         error = -EIO;
1463         dqopt->files[type] = igrab(inode);
1464         if (!dqopt->files[type])
1465                 goto out_lock;
1466         error = -EINVAL;
1467         if (!fmt->qf_ops->check_quota_file(sb, type))
1468                 goto out_file_init;
1469
1470         dqopt->ops[type] = fmt->qf_ops;
1471         dqopt->info[type].dqi_format = fmt;
1472         INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
1473         down(&dqopt->dqio_sem);
1474         if ((error = dqopt->ops[type]->read_file_info(sb, type)) < 0) {
1475                 up(&dqopt->dqio_sem);
1476                 goto out_file_init;
1477         }
1478         up(&dqopt->dqio_sem);
1479         mutex_unlock(&inode->i_mutex);
1480         set_enable_flags(dqopt, type);
1481
1482         add_dquot_ref(sb, type);
1483         up(&dqopt->dqonoff_sem);
1484
1485         return 0;
1486
1487 out_file_init:
1488         dqopt->files[type] = NULL;
1489         iput(inode);
1490 out_lock:
1491         up(&dqopt->dqonoff_sem);
1492         if (oldflags != -1) {
1493                 down_write(&dqopt->dqptr_sem);
1494                 /* Set the flags back (in the case of accidental quotaon()
1495                  * on a wrong file we don't want to mess up the flags) */
1496                 inode->i_flags &= ~(S_NOATIME | S_NOQUOTA | S_IMMUTABLE);
1497                 inode->i_flags |= oldflags;
1498                 up_write(&dqopt->dqptr_sem);
1499         }
1500         mutex_unlock(&inode->i_mutex);
1501 out_fmt:
1502         put_quota_format(fmt);
1503
1504         return error; 
1505 }
1506
1507 /* Actual function called from quotactl() */
1508 int vfs_quota_on(struct super_block *sb, int type, int format_id, char *path)
1509 {
1510         struct nameidata nd;
1511         int error;
1512
1513         error = path_lookup(path, LOOKUP_FOLLOW, &nd);
1514         if (error < 0)
1515                 return error;
1516         error = security_quota_on(nd.dentry);
1517         if (error)
1518                 goto out_path;
1519         /* Quota file not on the same filesystem? */
1520         if (nd.mnt->mnt_sb != sb)
1521                 error = -EXDEV;
1522         else
1523                 error = vfs_quota_on_inode(nd.dentry->d_inode, type, format_id);
1524 out_path:
1525         path_release(&nd);
1526         return error;
1527 }
1528
1529 /*
1530  * This function is used when filesystem needs to initialize quotas
1531  * during mount time.
1532  */
1533 int vfs_quota_on_mount(struct super_block *sb, char *qf_name,
1534                 int format_id, int type)
1535 {
1536         struct dentry *dentry;
1537         int error;
1538
1539         dentry = lookup_one_len(qf_name, sb->s_root, strlen(qf_name));
1540         if (IS_ERR(dentry))
1541                 return PTR_ERR(dentry);
1542
1543         if (!dentry->d_inode) {
1544                 error = -ENOENT;
1545                 goto out;
1546         }
1547
1548         error = security_quota_on(dentry);
1549         if (!error)
1550                 error = vfs_quota_on_inode(dentry->d_inode, type, format_id);
1551
1552 out:
1553         dput(dentry);
1554         return error;
1555 }
1556
1557 /* Generic routine for getting common part of quota structure */
1558 static void do_get_dqblk(struct dquot *dquot, struct if_dqblk *di)
1559 {
1560         struct mem_dqblk *dm = &dquot->dq_dqb;
1561
1562         spin_lock(&dq_data_lock);
1563         di->dqb_bhardlimit = dm->dqb_bhardlimit;
1564         di->dqb_bsoftlimit = dm->dqb_bsoftlimit;
1565         di->dqb_curspace = dm->dqb_curspace;
1566         di->dqb_ihardlimit = dm->dqb_ihardlimit;
1567         di->dqb_isoftlimit = dm->dqb_isoftlimit;
1568         di->dqb_curinodes = dm->dqb_curinodes;
1569         di->dqb_btime = dm->dqb_btime;
1570         di->dqb_itime = dm->dqb_itime;
1571         di->dqb_valid = QIF_ALL;
1572         spin_unlock(&dq_data_lock);
1573 }
1574
1575 int vfs_get_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
1576 {
1577         struct dquot *dquot;
1578
1579         down(&sb_dqopt(sb)->dqonoff_sem);
1580         if (!(dquot = dqget(sb, id, type))) {
1581                 up(&sb_dqopt(sb)->dqonoff_sem);
1582                 return -ESRCH;
1583         }
1584         do_get_dqblk(dquot, di);
1585         dqput(dquot);
1586         up(&sb_dqopt(sb)->dqonoff_sem);
1587         return 0;
1588 }
1589
1590 /* Generic routine for setting common part of quota structure */
1591 static void do_set_dqblk(struct dquot *dquot, struct if_dqblk *di)
1592 {
1593         struct mem_dqblk *dm = &dquot->dq_dqb;
1594         int check_blim = 0, check_ilim = 0;
1595
1596         spin_lock(&dq_data_lock);
1597         if (di->dqb_valid & QIF_SPACE) {
1598                 dm->dqb_curspace = di->dqb_curspace;
1599                 check_blim = 1;
1600         }
1601         if (di->dqb_valid & QIF_BLIMITS) {
1602                 dm->dqb_bsoftlimit = di->dqb_bsoftlimit;
1603                 dm->dqb_bhardlimit = di->dqb_bhardlimit;
1604                 check_blim = 1;
1605         }
1606         if (di->dqb_valid & QIF_INODES) {
1607                 dm->dqb_curinodes = di->dqb_curinodes;
1608                 check_ilim = 1;
1609         }
1610         if (di->dqb_valid & QIF_ILIMITS) {
1611                 dm->dqb_isoftlimit = di->dqb_isoftlimit;
1612                 dm->dqb_ihardlimit = di->dqb_ihardlimit;
1613                 check_ilim = 1;
1614         }
1615         if (di->dqb_valid & QIF_BTIME)
1616                 dm->dqb_btime = di->dqb_btime;
1617         if (di->dqb_valid & QIF_ITIME)
1618                 dm->dqb_itime = di->dqb_itime;
1619
1620         if (check_blim) {
1621                 if (!dm->dqb_bsoftlimit || toqb(dm->dqb_curspace) < dm->dqb_bsoftlimit) {
1622                         dm->dqb_btime = 0;
1623                         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1624                 }
1625                 else if (!(di->dqb_valid & QIF_BTIME))  /* Set grace only if user hasn't provided his own... */
1626                         dm->dqb_btime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_bgrace;
1627         }
1628         if (check_ilim) {
1629                 if (!dm->dqb_isoftlimit || dm->dqb_curinodes < dm->dqb_isoftlimit) {
1630                         dm->dqb_itime = 0;
1631                         clear_bit(DQ_INODES_B, &dquot->dq_flags);
1632                 }
1633                 else if (!(di->dqb_valid & QIF_ITIME))  /* Set grace only if user hasn't provided his own... */
1634                         dm->dqb_itime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
1635         }
1636         if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit || dm->dqb_isoftlimit)
1637                 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
1638         else
1639                 set_bit(DQ_FAKE_B, &dquot->dq_flags);
1640         spin_unlock(&dq_data_lock);
1641         mark_dquot_dirty(dquot);
1642 }
1643
1644 int vfs_set_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
1645 {
1646         struct dquot *dquot;
1647
1648         down(&sb_dqopt(sb)->dqonoff_sem);
1649         if (!(dquot = dqget(sb, id, type))) {
1650                 up(&sb_dqopt(sb)->dqonoff_sem);
1651                 return -ESRCH;
1652         }
1653         do_set_dqblk(dquot, di);
1654         dqput(dquot);
1655         up(&sb_dqopt(sb)->dqonoff_sem);
1656         return 0;
1657 }
1658
1659 /* Generic routine for getting common part of quota file information */
1660 int vfs_get_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
1661 {
1662         struct mem_dqinfo *mi;
1663   
1664         down(&sb_dqopt(sb)->dqonoff_sem);
1665         if (!sb_has_quota_enabled(sb, type)) {
1666                 up(&sb_dqopt(sb)->dqonoff_sem);
1667                 return -ESRCH;
1668         }
1669         mi = sb_dqopt(sb)->info + type;
1670         spin_lock(&dq_data_lock);
1671         ii->dqi_bgrace = mi->dqi_bgrace;
1672         ii->dqi_igrace = mi->dqi_igrace;
1673         ii->dqi_flags = mi->dqi_flags & DQF_MASK;
1674         ii->dqi_valid = IIF_ALL;
1675         spin_unlock(&dq_data_lock);
1676         up(&sb_dqopt(sb)->dqonoff_sem);
1677         return 0;
1678 }
1679
1680 /* Generic routine for setting common part of quota file information */
1681 int vfs_set_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
1682 {
1683         struct mem_dqinfo *mi;
1684
1685         down(&sb_dqopt(sb)->dqonoff_sem);
1686         if (!sb_has_quota_enabled(sb, type)) {
1687                 up(&sb_dqopt(sb)->dqonoff_sem);
1688                 return -ESRCH;
1689         }
1690         mi = sb_dqopt(sb)->info + type;
1691         spin_lock(&dq_data_lock);
1692         if (ii->dqi_valid & IIF_BGRACE)
1693                 mi->dqi_bgrace = ii->dqi_bgrace;
1694         if (ii->dqi_valid & IIF_IGRACE)
1695                 mi->dqi_igrace = ii->dqi_igrace;
1696         if (ii->dqi_valid & IIF_FLAGS)
1697                 mi->dqi_flags = (mi->dqi_flags & ~DQF_MASK) | (ii->dqi_flags & DQF_MASK);
1698         spin_unlock(&dq_data_lock);
1699         mark_info_dirty(sb, type);
1700         /* Force write to disk */
1701         sb->dq_op->write_info(sb, type);
1702         up(&sb_dqopt(sb)->dqonoff_sem);
1703         return 0;
1704 }
1705
1706 struct quotactl_ops vfs_quotactl_ops = {
1707         .quota_on       = vfs_quota_on,
1708         .quota_off      = vfs_quota_off,
1709         .quota_sync     = vfs_quota_sync,
1710         .get_info       = vfs_get_dqinfo,
1711         .set_info       = vfs_set_dqinfo,
1712         .get_dqblk      = vfs_get_dqblk,
1713         .set_dqblk      = vfs_set_dqblk
1714 };
1715
1716 static ctl_table fs_dqstats_table[] = {
1717         {
1718                 .ctl_name       = FS_DQ_LOOKUPS,
1719                 .procname       = "lookups",
1720                 .data           = &dqstats.lookups,
1721                 .maxlen         = sizeof(int),
1722                 .mode           = 0444,
1723                 .proc_handler   = &proc_dointvec,
1724         },
1725         {
1726                 .ctl_name       = FS_DQ_DROPS,
1727                 .procname       = "drops",
1728                 .data           = &dqstats.drops,
1729                 .maxlen         = sizeof(int),
1730                 .mode           = 0444,
1731                 .proc_handler   = &proc_dointvec,
1732         },
1733         {
1734                 .ctl_name       = FS_DQ_READS,
1735                 .procname       = "reads",
1736                 .data           = &dqstats.reads,
1737                 .maxlen         = sizeof(int),
1738                 .mode           = 0444,
1739                 .proc_handler   = &proc_dointvec,
1740         },
1741         {
1742                 .ctl_name       = FS_DQ_WRITES,
1743                 .procname       = "writes",
1744                 .data           = &dqstats.writes,
1745                 .maxlen         = sizeof(int),
1746                 .mode           = 0444,
1747                 .proc_handler   = &proc_dointvec,
1748         },
1749         {
1750                 .ctl_name       = FS_DQ_CACHE_HITS,
1751                 .procname       = "cache_hits",
1752                 .data           = &dqstats.cache_hits,
1753                 .maxlen         = sizeof(int),
1754                 .mode           = 0444,
1755                 .proc_handler   = &proc_dointvec,
1756         },
1757         {
1758                 .ctl_name       = FS_DQ_ALLOCATED,
1759                 .procname       = "allocated_dquots",
1760                 .data           = &dqstats.allocated_dquots,
1761                 .maxlen         = sizeof(int),
1762                 .mode           = 0444,
1763                 .proc_handler   = &proc_dointvec,
1764         },
1765         {
1766                 .ctl_name       = FS_DQ_FREE,
1767                 .procname       = "free_dquots",
1768                 .data           = &dqstats.free_dquots,
1769                 .maxlen         = sizeof(int),
1770                 .mode           = 0444,
1771                 .proc_handler   = &proc_dointvec,
1772         },
1773         {
1774                 .ctl_name       = FS_DQ_SYNCS,
1775                 .procname       = "syncs",
1776                 .data           = &dqstats.syncs,
1777                 .maxlen         = sizeof(int),
1778                 .mode           = 0444,
1779                 .proc_handler   = &proc_dointvec,
1780         },
1781         {
1782                 .ctl_name       = FS_DQ_WARNINGS,
1783                 .procname       = "warnings",
1784                 .data           = &flag_print_warnings,
1785                 .maxlen         = sizeof(int),
1786                 .mode           = 0644,
1787                 .proc_handler   = &proc_dointvec,
1788         },
1789         { .ctl_name = 0 },
1790 };
1791
1792 static ctl_table fs_table[] = {
1793         {
1794                 .ctl_name       = FS_DQSTATS,
1795                 .procname       = "quota",
1796                 .mode           = 0555,
1797                 .child          = fs_dqstats_table,
1798         },
1799         { .ctl_name = 0 },
1800 };
1801
1802 static ctl_table sys_table[] = {
1803         {
1804                 .ctl_name       = CTL_FS,
1805                 .procname       = "fs",
1806                 .mode           = 0555,
1807                 .child          = fs_table,
1808         },
1809         { .ctl_name = 0 },
1810 };
1811
1812 static int __init dquot_init(void)
1813 {
1814         int i;
1815         unsigned long nr_hash, order;
1816
1817         printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
1818
1819         register_sysctl_table(sys_table, 0);
1820
1821         dquot_cachep = kmem_cache_create("dquot", 
1822                         sizeof(struct dquot), sizeof(unsigned long) * 4,
1823                         SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|SLAB_PANIC,
1824                         NULL, NULL);
1825
1826         order = 0;
1827         dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
1828         if (!dquot_hash)
1829                 panic("Cannot create dquot hash table");
1830
1831         /* Find power-of-two hlist_heads which can fit into allocation */
1832         nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
1833         dq_hash_bits = 0;
1834         do {
1835                 dq_hash_bits++;
1836         } while (nr_hash >> dq_hash_bits);
1837         dq_hash_bits--;
1838
1839         nr_hash = 1UL << dq_hash_bits;
1840         dq_hash_mask = nr_hash - 1;
1841         for (i = 0; i < nr_hash; i++)
1842                 INIT_HLIST_HEAD(dquot_hash + i);
1843
1844         printk("Dquot-cache hash table entries: %ld (order %ld, %ld bytes)\n",
1845                         nr_hash, order, (PAGE_SIZE << order));
1846
1847         set_shrinker(DEFAULT_SEEKS, shrink_dqcache_memory);
1848
1849         return 0;
1850 }
1851 module_init(dquot_init);
1852
1853 EXPORT_SYMBOL(register_quota_format);
1854 EXPORT_SYMBOL(unregister_quota_format);
1855 EXPORT_SYMBOL(dqstats);
1856 EXPORT_SYMBOL(dq_data_lock);
1857 EXPORT_SYMBOL(vfs_quota_on);
1858 EXPORT_SYMBOL(vfs_quota_on_mount);
1859 EXPORT_SYMBOL(vfs_quota_off);
1860 EXPORT_SYMBOL(vfs_quota_sync);
1861 EXPORT_SYMBOL(vfs_get_dqinfo);
1862 EXPORT_SYMBOL(vfs_set_dqinfo);
1863 EXPORT_SYMBOL(vfs_get_dqblk);
1864 EXPORT_SYMBOL(vfs_set_dqblk);
1865 EXPORT_SYMBOL(dquot_commit);
1866 EXPORT_SYMBOL(dquot_commit_info);
1867 EXPORT_SYMBOL(dquot_acquire);
1868 EXPORT_SYMBOL(dquot_release);
1869 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
1870 EXPORT_SYMBOL(dquot_initialize);
1871 EXPORT_SYMBOL(dquot_drop);
1872 EXPORT_SYMBOL(dquot_alloc_space);
1873 EXPORT_SYMBOL(dquot_alloc_inode);
1874 EXPORT_SYMBOL(dquot_free_space);
1875 EXPORT_SYMBOL(dquot_free_inode);
1876 EXPORT_SYMBOL(dquot_transfer);