quota: Add callbacks for allocating and destroying dquot structures
[safe/jmp/linux-2.6] / fs / dquot.c
1 /*
2  * Implementation of the diskquota system for the LINUX operating system. QUOTA
3  * is implemented using the BSD system call interface as the means of
4  * communication with the user level. This file contains the generic routines
5  * called by the different filesystems on allocation of an inode or block.
6  * These routines take care of the administration needed to have a consistent
7  * diskquota tracking system. The ideas of both user and group quotas are based
8  * on the Melbourne quota system as used on BSD derived systems. The internal
9  * implementation is based on one of the several variants of the LINUX
10  * inode-subsystem with added complexity of the diskquota system.
11  * 
12  * Author:      Marco van Wieringen <mvw@planets.elm.net>
13  *
14  * Fixes:   Dmitry Gorodchanin <pgmdsg@ibi.com>, 11 Feb 96
15  *
16  *              Revised list management to avoid races
17  *              -- Bill Hawes, <whawes@star.net>, 9/98
18  *
19  *              Fixed races in dquot_transfer(), dqget() and dquot_alloc_...().
20  *              As the consequence the locking was moved from dquot_decr_...(),
21  *              dquot_incr_...() to calling functions.
22  *              invalidate_dquots() now writes modified dquots.
23  *              Serialized quota_off() and quota_on() for mount point.
24  *              Fixed a few bugs in grow_dquots().
25  *              Fixed deadlock in write_dquot() - we no longer account quotas on
26  *              quota files
27  *              remove_dquot_ref() moved to inode.c - it now traverses through inodes
28  *              add_dquot_ref() restarts after blocking
29  *              Added check for bogus uid and fixed check for group in quotactl.
30  *              Jan Kara, <jack@suse.cz>, sponsored by SuSE CR, 10-11/99
31  *
32  *              Used struct list_head instead of own list struct
33  *              Invalidation of referenced dquots is no longer possible
34  *              Improved free_dquots list management
35  *              Quota and i_blocks are now updated in one place to avoid races
36  *              Warnings are now delayed so we won't block in critical section
37  *              Write updated not to require dquot lock
38  *              Jan Kara, <jack@suse.cz>, 9/2000
39  *
40  *              Added dynamic quota structure allocation
41  *              Jan Kara <jack@suse.cz> 12/2000
42  *
43  *              Rewritten quota interface. Implemented new quota format and
44  *              formats registering.
45  *              Jan Kara, <jack@suse.cz>, 2001,2002
46  *
47  *              New SMP locking.
48  *              Jan Kara, <jack@suse.cz>, 10/2002
49  *
50  *              Added journalled quota support, fix lock inversion problems
51  *              Jan Kara, <jack@suse.cz>, 2003,2004
52  *
53  * (C) Copyright 1994 - 1997 Marco van Wieringen 
54  */
55
56 #include <linux/errno.h>
57 #include <linux/kernel.h>
58 #include <linux/fs.h>
59 #include <linux/mount.h>
60 #include <linux/mm.h>
61 #include <linux/time.h>
62 #include <linux/types.h>
63 #include <linux/string.h>
64 #include <linux/fcntl.h>
65 #include <linux/stat.h>
66 #include <linux/tty.h>
67 #include <linux/file.h>
68 #include <linux/slab.h>
69 #include <linux/sysctl.h>
70 #include <linux/init.h>
71 #include <linux/module.h>
72 #include <linux/proc_fs.h>
73 #include <linux/security.h>
74 #include <linux/kmod.h>
75 #include <linux/namei.h>
76 #include <linux/buffer_head.h>
77 #include <linux/capability.h>
78 #include <linux/quotaops.h>
79 #include <linux/writeback.h> /* for inode_lock, oddly enough.. */
80 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
81 #include <net/netlink.h>
82 #include <net/genetlink.h>
83 #endif
84
85 #include <asm/uaccess.h>
86
87 #define __DQUOT_PARANOIA
88
89 /*
90  * There are two quota SMP locks. dq_list_lock protects all lists with quotas
91  * and quota formats and also dqstats structure containing statistics about the
92  * lists. dq_data_lock protects data from dq_dqb and also mem_dqinfo structures
93  * and also guards consistency of dquot->dq_dqb with inode->i_blocks, i_bytes.
94  * i_blocks and i_bytes updates itself are guarded by i_lock acquired directly
95  * in inode_add_bytes() and inode_sub_bytes().
96  *
97  * The spinlock ordering is hence: dq_data_lock > dq_list_lock > i_lock
98  *
99  * Note that some things (eg. sb pointer, type, id) doesn't change during
100  * the life of the dquot structure and so needn't to be protected by a lock
101  *
102  * Any operation working on dquots via inode pointers must hold dqptr_sem.  If
103  * operation is just reading pointers from inode (or not using them at all) the
104  * read lock is enough. If pointers are altered function must hold write lock
105  * (these locking rules also apply for S_NOQUOTA flag in the inode - note that
106  * for altering the flag i_mutex is also needed).  If operation is holding
107  * reference to dquot in other way (e.g. quotactl ops) it must be guarded by
108  * dqonoff_mutex.
109  * This locking assures that:
110  *   a) update/access to dquot pointers in inode is serialized
111  *   b) everyone is guarded against invalidate_dquots()
112  *
113  * Each dquot has its dq_lock mutex. Locked dquots might not be referenced
114  * from inodes (dquot_alloc_space() and such don't check the dq_lock).
115  * Currently dquot is locked only when it is being read to memory (or space for
116  * it is being allocated) on the first dqget() and when it is being released on
117  * the last dqput(). The allocation and release oparations are serialized by
118  * the dq_lock and by checking the use count in dquot_release().  Write
119  * operations on dquots don't hold dq_lock as they copy data under dq_data_lock
120  * spinlock to internal buffers before writing.
121  *
122  * Lock ordering (including related VFS locks) is the following:
123  *   i_mutex > dqonoff_sem > journal_lock > dqptr_sem > dquot->dq_lock >
124  *   dqio_mutex
125  * i_mutex on quota files is special (it's below dqio_mutex)
126  */
127
128 static DEFINE_SPINLOCK(dq_list_lock);
129 DEFINE_SPINLOCK(dq_data_lock);
130
131 static char *quotatypes[] = INITQFNAMES;
132 static struct quota_format_type *quota_formats; /* List of registered formats */
133 static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
134
135 /* SLAB cache for dquot structures */
136 static struct kmem_cache *dquot_cachep;
137
138 int register_quota_format(struct quota_format_type *fmt)
139 {
140         spin_lock(&dq_list_lock);
141         fmt->qf_next = quota_formats;
142         quota_formats = fmt;
143         spin_unlock(&dq_list_lock);
144         return 0;
145 }
146
147 void unregister_quota_format(struct quota_format_type *fmt)
148 {
149         struct quota_format_type **actqf;
150
151         spin_lock(&dq_list_lock);
152         for (actqf = &quota_formats; *actqf && *actqf != fmt; actqf = &(*actqf)->qf_next);
153         if (*actqf)
154                 *actqf = (*actqf)->qf_next;
155         spin_unlock(&dq_list_lock);
156 }
157
158 static struct quota_format_type *find_quota_format(int id)
159 {
160         struct quota_format_type *actqf;
161
162         spin_lock(&dq_list_lock);
163         for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id; actqf = actqf->qf_next);
164         if (!actqf || !try_module_get(actqf->qf_owner)) {
165                 int qm;
166
167                 spin_unlock(&dq_list_lock);
168                 
169                 for (qm = 0; module_names[qm].qm_fmt_id && module_names[qm].qm_fmt_id != id; qm++);
170                 if (!module_names[qm].qm_fmt_id || request_module(module_names[qm].qm_mod_name))
171                         return NULL;
172
173                 spin_lock(&dq_list_lock);
174                 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id; actqf = actqf->qf_next);
175                 if (actqf && !try_module_get(actqf->qf_owner))
176                         actqf = NULL;
177         }
178         spin_unlock(&dq_list_lock);
179         return actqf;
180 }
181
182 static void put_quota_format(struct quota_format_type *fmt)
183 {
184         module_put(fmt->qf_owner);
185 }
186
187 /*
188  * Dquot List Management:
189  * The quota code uses three lists for dquot management: the inuse_list,
190  * free_dquots, and dquot_hash[] array. A single dquot structure may be
191  * on all three lists, depending on its current state.
192  *
193  * All dquots are placed to the end of inuse_list when first created, and this
194  * list is used for invalidate operation, which must look at every dquot.
195  *
196  * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
197  * and this list is searched whenever we need an available dquot.  Dquots are
198  * removed from the list as soon as they are used again, and
199  * dqstats.free_dquots gives the number of dquots on the list. When
200  * dquot is invalidated it's completely released from memory.
201  *
202  * Dquots with a specific identity (device, type and id) are placed on
203  * one of the dquot_hash[] hash chains. The provides an efficient search
204  * mechanism to locate a specific dquot.
205  */
206
207 static LIST_HEAD(inuse_list);
208 static LIST_HEAD(free_dquots);
209 static unsigned int dq_hash_bits, dq_hash_mask;
210 static struct hlist_head *dquot_hash;
211
212 struct dqstats dqstats;
213
214 static void dqput(struct dquot *dquot);
215
216 static inline unsigned int
217 hashfn(const struct super_block *sb, unsigned int id, int type)
218 {
219         unsigned long tmp;
220
221         tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
222         return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
223 }
224
225 /*
226  * Following list functions expect dq_list_lock to be held
227  */
228 static inline void insert_dquot_hash(struct dquot *dquot)
229 {
230         struct hlist_head *head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id, dquot->dq_type);
231         hlist_add_head(&dquot->dq_hash, head);
232 }
233
234 static inline void remove_dquot_hash(struct dquot *dquot)
235 {
236         hlist_del_init(&dquot->dq_hash);
237 }
238
239 static inline struct dquot *find_dquot(unsigned int hashent, struct super_block *sb, unsigned int id, int type)
240 {
241         struct hlist_node *node;
242         struct dquot *dquot;
243
244         hlist_for_each (node, dquot_hash+hashent) {
245                 dquot = hlist_entry(node, struct dquot, dq_hash);
246                 if (dquot->dq_sb == sb && dquot->dq_id == id && dquot->dq_type == type)
247                         return dquot;
248         }
249         return NODQUOT;
250 }
251
252 /* Add a dquot to the tail of the free list */
253 static inline void put_dquot_last(struct dquot *dquot)
254 {
255         list_add_tail(&dquot->dq_free, &free_dquots);
256         dqstats.free_dquots++;
257 }
258
259 static inline void remove_free_dquot(struct dquot *dquot)
260 {
261         if (list_empty(&dquot->dq_free))
262                 return;
263         list_del_init(&dquot->dq_free);
264         dqstats.free_dquots--;
265 }
266
267 static inline void put_inuse(struct dquot *dquot)
268 {
269         /* We add to the back of inuse list so we don't have to restart
270          * when traversing this list and we block */
271         list_add_tail(&dquot->dq_inuse, &inuse_list);
272         dqstats.allocated_dquots++;
273 }
274
275 static inline void remove_inuse(struct dquot *dquot)
276 {
277         dqstats.allocated_dquots--;
278         list_del(&dquot->dq_inuse);
279 }
280 /*
281  * End of list functions needing dq_list_lock
282  */
283
284 static void wait_on_dquot(struct dquot *dquot)
285 {
286         mutex_lock(&dquot->dq_lock);
287         mutex_unlock(&dquot->dq_lock);
288 }
289
290 static inline int dquot_dirty(struct dquot *dquot)
291 {
292         return test_bit(DQ_MOD_B, &dquot->dq_flags);
293 }
294
295 static inline int mark_dquot_dirty(struct dquot *dquot)
296 {
297         return dquot->dq_sb->dq_op->mark_dirty(dquot);
298 }
299
300 int dquot_mark_dquot_dirty(struct dquot *dquot)
301 {
302         spin_lock(&dq_list_lock);
303         if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags))
304                 list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
305                                 info[dquot->dq_type].dqi_dirty_list);
306         spin_unlock(&dq_list_lock);
307         return 0;
308 }
309
310 /* This function needs dq_list_lock */
311 static inline int clear_dquot_dirty(struct dquot *dquot)
312 {
313         if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags))
314                 return 0;
315         list_del_init(&dquot->dq_dirty);
316         return 1;
317 }
318
319 void mark_info_dirty(struct super_block *sb, int type)
320 {
321         set_bit(DQF_INFO_DIRTY_B, &sb_dqopt(sb)->info[type].dqi_flags);
322 }
323 EXPORT_SYMBOL(mark_info_dirty);
324
325 /*
326  *      Read dquot from disk and alloc space for it
327  */
328
329 int dquot_acquire(struct dquot *dquot)
330 {
331         int ret = 0, ret2 = 0;
332         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
333
334         mutex_lock(&dquot->dq_lock);
335         mutex_lock(&dqopt->dqio_mutex);
336         if (!test_bit(DQ_READ_B, &dquot->dq_flags))
337                 ret = dqopt->ops[dquot->dq_type]->read_dqblk(dquot);
338         if (ret < 0)
339                 goto out_iolock;
340         set_bit(DQ_READ_B, &dquot->dq_flags);
341         /* Instantiate dquot if needed */
342         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
343                 ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
344                 /* Write the info if needed */
345                 if (info_dirty(&dqopt->info[dquot->dq_type]))
346                         ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
347                 if (ret < 0)
348                         goto out_iolock;
349                 if (ret2 < 0) {
350                         ret = ret2;
351                         goto out_iolock;
352                 }
353         }
354         set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
355 out_iolock:
356         mutex_unlock(&dqopt->dqio_mutex);
357         mutex_unlock(&dquot->dq_lock);
358         return ret;
359 }
360
361 /*
362  *      Write dquot to disk
363  */
364 int dquot_commit(struct dquot *dquot)
365 {
366         int ret = 0, ret2 = 0;
367         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
368
369         mutex_lock(&dqopt->dqio_mutex);
370         spin_lock(&dq_list_lock);
371         if (!clear_dquot_dirty(dquot)) {
372                 spin_unlock(&dq_list_lock);
373                 goto out_sem;
374         }
375         spin_unlock(&dq_list_lock);
376         /* Inactive dquot can be only if there was error during read/init
377          * => we have better not writing it */
378         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
379                 ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
380                 if (info_dirty(&dqopt->info[dquot->dq_type]))
381                         ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
382                 if (ret >= 0)
383                         ret = ret2;
384         }
385 out_sem:
386         mutex_unlock(&dqopt->dqio_mutex);
387         return ret;
388 }
389
390 /*
391  *      Release dquot
392  */
393 int dquot_release(struct dquot *dquot)
394 {
395         int ret = 0, ret2 = 0;
396         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
397
398         mutex_lock(&dquot->dq_lock);
399         /* Check whether we are not racing with some other dqget() */
400         if (atomic_read(&dquot->dq_count) > 1)
401                 goto out_dqlock;
402         mutex_lock(&dqopt->dqio_mutex);
403         if (dqopt->ops[dquot->dq_type]->release_dqblk) {
404                 ret = dqopt->ops[dquot->dq_type]->release_dqblk(dquot);
405                 /* Write the info */
406                 if (info_dirty(&dqopt->info[dquot->dq_type]))
407                         ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
408                 if (ret >= 0)
409                         ret = ret2;
410         }
411         clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
412         mutex_unlock(&dqopt->dqio_mutex);
413 out_dqlock:
414         mutex_unlock(&dquot->dq_lock);
415         return ret;
416 }
417
418 static void dquot_destroy(struct dquot *dquot)
419 {
420         kmem_cache_free(dquot_cachep, dquot);
421 }
422
423 static inline void do_destroy_dquot(struct dquot *dquot)
424 {
425         dquot->dq_sb->dq_op->destroy_dquot(dquot);
426 }
427
428 /* Invalidate all dquots on the list. Note that this function is called after
429  * quota is disabled and pointers from inodes removed so there cannot be new
430  * quota users. There can still be some users of quotas due to inodes being
431  * just deleted or pruned by prune_icache() (those are not attached to any
432  * list). We have to wait for such users.
433  */
434 static void invalidate_dquots(struct super_block *sb, int type)
435 {
436         struct dquot *dquot, *tmp;
437
438 restart:
439         spin_lock(&dq_list_lock);
440         list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
441                 if (dquot->dq_sb != sb)
442                         continue;
443                 if (dquot->dq_type != type)
444                         continue;
445                 /* Wait for dquot users */
446                 if (atomic_read(&dquot->dq_count)) {
447                         DEFINE_WAIT(wait);
448
449                         atomic_inc(&dquot->dq_count);
450                         prepare_to_wait(&dquot->dq_wait_unused, &wait,
451                                         TASK_UNINTERRUPTIBLE);
452                         spin_unlock(&dq_list_lock);
453                         /* Once dqput() wakes us up, we know it's time to free
454                          * the dquot.
455                          * IMPORTANT: we rely on the fact that there is always
456                          * at most one process waiting for dquot to free.
457                          * Otherwise dq_count would be > 1 and we would never
458                          * wake up.
459                          */
460                         if (atomic_read(&dquot->dq_count) > 1)
461                                 schedule();
462                         finish_wait(&dquot->dq_wait_unused, &wait);
463                         dqput(dquot);
464                         /* At this moment dquot() need not exist (it could be
465                          * reclaimed by prune_dqcache(). Hence we must
466                          * restart. */
467                         goto restart;
468                 }
469                 /*
470                  * Quota now has no users and it has been written on last
471                  * dqput()
472                  */
473                 remove_dquot_hash(dquot);
474                 remove_free_dquot(dquot);
475                 remove_inuse(dquot);
476                 do_destroy_dquot(dquot);
477         }
478         spin_unlock(&dq_list_lock);
479 }
480
481 int vfs_quota_sync(struct super_block *sb, int type)
482 {
483         struct list_head *dirty;
484         struct dquot *dquot;
485         struct quota_info *dqopt = sb_dqopt(sb);
486         int cnt;
487
488         mutex_lock(&dqopt->dqonoff_mutex);
489         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
490                 if (type != -1 && cnt != type)
491                         continue;
492                 if (!sb_has_quota_enabled(sb, cnt))
493                         continue;
494                 spin_lock(&dq_list_lock);
495                 dirty = &dqopt->info[cnt].dqi_dirty_list;
496                 while (!list_empty(dirty)) {
497                         dquot = list_first_entry(dirty, struct dquot, dq_dirty);
498                         /* Dirty and inactive can be only bad dquot... */
499                         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
500                                 clear_dquot_dirty(dquot);
501                                 continue;
502                         }
503                         /* Now we have active dquot from which someone is
504                          * holding reference so we can safely just increase
505                          * use count */
506                         atomic_inc(&dquot->dq_count);
507                         dqstats.lookups++;
508                         spin_unlock(&dq_list_lock);
509                         sb->dq_op->write_dquot(dquot);
510                         dqput(dquot);
511                         spin_lock(&dq_list_lock);
512                 }
513                 spin_unlock(&dq_list_lock);
514         }
515
516         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
517                 if ((cnt == type || type == -1) && sb_has_quota_enabled(sb, cnt)
518                         && info_dirty(&dqopt->info[cnt]))
519                         sb->dq_op->write_info(sb, cnt);
520         spin_lock(&dq_list_lock);
521         dqstats.syncs++;
522         spin_unlock(&dq_list_lock);
523         mutex_unlock(&dqopt->dqonoff_mutex);
524
525         return 0;
526 }
527
528 /* Free unused dquots from cache */
529 static void prune_dqcache(int count)
530 {
531         struct list_head *head;
532         struct dquot *dquot;
533
534         head = free_dquots.prev;
535         while (head != &free_dquots && count) {
536                 dquot = list_entry(head, struct dquot, dq_free);
537                 remove_dquot_hash(dquot);
538                 remove_free_dquot(dquot);
539                 remove_inuse(dquot);
540                 do_destroy_dquot(dquot);
541                 count--;
542                 head = free_dquots.prev;
543         }
544 }
545
546 /*
547  * This is called from kswapd when we think we need some
548  * more memory
549  */
550
551 static int shrink_dqcache_memory(int nr, gfp_t gfp_mask)
552 {
553         if (nr) {
554                 spin_lock(&dq_list_lock);
555                 prune_dqcache(nr);
556                 spin_unlock(&dq_list_lock);
557         }
558         return (dqstats.free_dquots / 100) * sysctl_vfs_cache_pressure;
559 }
560
561 static struct shrinker dqcache_shrinker = {
562         .shrink = shrink_dqcache_memory,
563         .seeks = DEFAULT_SEEKS,
564 };
565
566 /*
567  * Put reference to dquot
568  * NOTE: If you change this function please check whether dqput_blocks() works right...
569  * MUST be called with either dqptr_sem or dqonoff_mutex held
570  */
571 static void dqput(struct dquot *dquot)
572 {
573         int ret;
574
575         if (!dquot)
576                 return;
577 #ifdef __DQUOT_PARANOIA
578         if (!atomic_read(&dquot->dq_count)) {
579                 printk("VFS: dqput: trying to free free dquot\n");
580                 printk("VFS: device %s, dquot of %s %d\n",
581                         dquot->dq_sb->s_id,
582                         quotatypes[dquot->dq_type],
583                         dquot->dq_id);
584                 BUG();
585         }
586 #endif
587         
588         spin_lock(&dq_list_lock);
589         dqstats.drops++;
590         spin_unlock(&dq_list_lock);
591 we_slept:
592         spin_lock(&dq_list_lock);
593         if (atomic_read(&dquot->dq_count) > 1) {
594                 /* We have more than one user... nothing to do */
595                 atomic_dec(&dquot->dq_count);
596                 /* Releasing dquot during quotaoff phase? */
597                 if (!sb_has_quota_enabled(dquot->dq_sb, dquot->dq_type) &&
598                     atomic_read(&dquot->dq_count) == 1)
599                         wake_up(&dquot->dq_wait_unused);
600                 spin_unlock(&dq_list_lock);
601                 return;
602         }
603         /* Need to release dquot? */
604         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && dquot_dirty(dquot)) {
605                 spin_unlock(&dq_list_lock);
606                 /* Commit dquot before releasing */
607                 ret = dquot->dq_sb->dq_op->write_dquot(dquot);
608                 if (ret < 0) {
609                         printk(KERN_ERR "VFS: cannot write quota structure on "
610                                 "device %s (error %d). Quota may get out of "
611                                 "sync!\n", dquot->dq_sb->s_id, ret);
612                         /*
613                          * We clear dirty bit anyway, so that we avoid
614                          * infinite loop here
615                          */
616                         spin_lock(&dq_list_lock);
617                         clear_dquot_dirty(dquot);
618                         spin_unlock(&dq_list_lock);
619                 }
620                 goto we_slept;
621         }
622         /* Clear flag in case dquot was inactive (something bad happened) */
623         clear_dquot_dirty(dquot);
624         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
625                 spin_unlock(&dq_list_lock);
626                 dquot->dq_sb->dq_op->release_dquot(dquot);
627                 goto we_slept;
628         }
629         atomic_dec(&dquot->dq_count);
630 #ifdef __DQUOT_PARANOIA
631         /* sanity check */
632         BUG_ON(!list_empty(&dquot->dq_free));
633 #endif
634         put_dquot_last(dquot);
635         spin_unlock(&dq_list_lock);
636 }
637
638 static struct dquot *dquot_alloc(struct super_block *sb, int type)
639 {
640         return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
641 }
642
643 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
644 {
645         struct dquot *dquot;
646
647         dquot = sb->dq_op->alloc_dquot(sb, type);
648         if(!dquot)
649                 return NODQUOT;
650
651         mutex_init(&dquot->dq_lock);
652         INIT_LIST_HEAD(&dquot->dq_free);
653         INIT_LIST_HEAD(&dquot->dq_inuse);
654         INIT_HLIST_NODE(&dquot->dq_hash);
655         INIT_LIST_HEAD(&dquot->dq_dirty);
656         init_waitqueue_head(&dquot->dq_wait_unused);
657         dquot->dq_sb = sb;
658         dquot->dq_type = type;
659         atomic_set(&dquot->dq_count, 1);
660
661         return dquot;
662 }
663
664 /*
665  * Get reference to dquot
666  * MUST be called with either dqptr_sem or dqonoff_mutex held
667  */
668 static struct dquot *dqget(struct super_block *sb, unsigned int id, int type)
669 {
670         unsigned int hashent = hashfn(sb, id, type);
671         struct dquot *dquot, *empty = NODQUOT;
672
673         if (!sb_has_quota_enabled(sb, type))
674                 return NODQUOT;
675 we_slept:
676         spin_lock(&dq_list_lock);
677         if ((dquot = find_dquot(hashent, sb, id, type)) == NODQUOT) {
678                 if (empty == NODQUOT) {
679                         spin_unlock(&dq_list_lock);
680                         if ((empty = get_empty_dquot(sb, type)) == NODQUOT)
681                                 schedule();     /* Try to wait for a moment... */
682                         goto we_slept;
683                 }
684                 dquot = empty;
685                 dquot->dq_id = id;
686                 /* all dquots go on the inuse_list */
687                 put_inuse(dquot);
688                 /* hash it first so it can be found */
689                 insert_dquot_hash(dquot);
690                 dqstats.lookups++;
691                 spin_unlock(&dq_list_lock);
692         } else {
693                 if (!atomic_read(&dquot->dq_count))
694                         remove_free_dquot(dquot);
695                 atomic_inc(&dquot->dq_count);
696                 dqstats.cache_hits++;
697                 dqstats.lookups++;
698                 spin_unlock(&dq_list_lock);
699                 if (empty)
700                         do_destroy_dquot(empty);
701         }
702         /* Wait for dq_lock - after this we know that either dquot_release() is already
703          * finished or it will be canceled due to dq_count > 1 test */
704         wait_on_dquot(dquot);
705         /* Read the dquot and instantiate it (everything done only if needed) */
706         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && sb->dq_op->acquire_dquot(dquot) < 0) {
707                 dqput(dquot);
708                 return NODQUOT;
709         }
710 #ifdef __DQUOT_PARANOIA
711         BUG_ON(!dquot->dq_sb);  /* Has somebody invalidated entry under us? */
712 #endif
713
714         return dquot;
715 }
716
717 static int dqinit_needed(struct inode *inode, int type)
718 {
719         int cnt;
720
721         if (IS_NOQUOTA(inode))
722                 return 0;
723         if (type != -1)
724                 return inode->i_dquot[type] == NODQUOT;
725         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
726                 if (inode->i_dquot[cnt] == NODQUOT)
727                         return 1;
728         return 0;
729 }
730
731 /* This routine is guarded by dqonoff_mutex mutex */
732 static void add_dquot_ref(struct super_block *sb, int type)
733 {
734         struct inode *inode, *old_inode = NULL;
735
736         spin_lock(&inode_lock);
737         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
738                 if (!atomic_read(&inode->i_writecount))
739                         continue;
740                 if (!dqinit_needed(inode, type))
741                         continue;
742                 if (inode->i_state & (I_FREEING|I_WILL_FREE))
743                         continue;
744
745                 __iget(inode);
746                 spin_unlock(&inode_lock);
747
748                 iput(old_inode);
749                 sb->dq_op->initialize(inode, type);
750                 /* We hold a reference to 'inode' so it couldn't have been
751                  * removed from s_inodes list while we dropped the inode_lock.
752                  * We cannot iput the inode now as we can be holding the last
753                  * reference and we cannot iput it under inode_lock. So we
754                  * keep the reference and iput it later. */
755                 old_inode = inode;
756                 spin_lock(&inode_lock);
757         }
758         spin_unlock(&inode_lock);
759         iput(old_inode);
760 }
761
762 /* Return 0 if dqput() won't block (note that 1 doesn't necessarily mean blocking) */
763 static inline int dqput_blocks(struct dquot *dquot)
764 {
765         if (atomic_read(&dquot->dq_count) <= 1)
766                 return 1;
767         return 0;
768 }
769
770 /* Remove references to dquots from inode - add dquot to list for freeing if needed */
771 /* We can't race with anybody because we hold dqptr_sem for writing... */
772 static int remove_inode_dquot_ref(struct inode *inode, int type,
773                                   struct list_head *tofree_head)
774 {
775         struct dquot *dquot = inode->i_dquot[type];
776
777         inode->i_dquot[type] = NODQUOT;
778         if (dquot != NODQUOT) {
779                 if (dqput_blocks(dquot)) {
780 #ifdef __DQUOT_PARANOIA
781                         if (atomic_read(&dquot->dq_count) != 1)
782                                 printk(KERN_WARNING "VFS: Adding dquot with dq_count %d to dispose list.\n", atomic_read(&dquot->dq_count));
783 #endif
784                         spin_lock(&dq_list_lock);
785                         list_add(&dquot->dq_free, tofree_head); /* As dquot must have currently users it can't be on the free list... */
786                         spin_unlock(&dq_list_lock);
787                         return 1;
788                 }
789                 else
790                         dqput(dquot);   /* We have guaranteed we won't block */
791         }
792         return 0;
793 }
794
795 /* Free list of dquots - called from inode.c */
796 /* dquots are removed from inodes, no new references can be got so we are the only ones holding reference */
797 static void put_dquot_list(struct list_head *tofree_head)
798 {
799         struct list_head *act_head;
800         struct dquot *dquot;
801
802         act_head = tofree_head->next;
803         /* So now we have dquots on the list... Just free them */
804         while (act_head != tofree_head) {
805                 dquot = list_entry(act_head, struct dquot, dq_free);
806                 act_head = act_head->next;
807                 list_del_init(&dquot->dq_free); /* Remove dquot from the list so we won't have problems... */
808                 dqput(dquot);
809         }
810 }
811
812 static void remove_dquot_ref(struct super_block *sb, int type,
813                 struct list_head *tofree_head)
814 {
815         struct inode *inode;
816
817         spin_lock(&inode_lock);
818         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
819                 if (!IS_NOQUOTA(inode))
820                         remove_inode_dquot_ref(inode, type, tofree_head);
821         }
822         spin_unlock(&inode_lock);
823 }
824
825 /* Gather all references from inodes and drop them */
826 static void drop_dquot_ref(struct super_block *sb, int type)
827 {
828         LIST_HEAD(tofree_head);
829
830         if (sb->dq_op) {
831                 down_write(&sb_dqopt(sb)->dqptr_sem);
832                 remove_dquot_ref(sb, type, &tofree_head);
833                 up_write(&sb_dqopt(sb)->dqptr_sem);
834                 put_dquot_list(&tofree_head);
835         }
836 }
837
838 static inline void dquot_incr_inodes(struct dquot *dquot, unsigned long number)
839 {
840         dquot->dq_dqb.dqb_curinodes += number;
841 }
842
843 static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
844 {
845         dquot->dq_dqb.dqb_curspace += number;
846 }
847
848 static inline void dquot_decr_inodes(struct dquot *dquot, unsigned long number)
849 {
850         if (dquot->dq_dqb.dqb_curinodes > number)
851                 dquot->dq_dqb.dqb_curinodes -= number;
852         else
853                 dquot->dq_dqb.dqb_curinodes = 0;
854         if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
855                 dquot->dq_dqb.dqb_itime = (time_t) 0;
856         clear_bit(DQ_INODES_B, &dquot->dq_flags);
857 }
858
859 static inline void dquot_decr_space(struct dquot *dquot, qsize_t number)
860 {
861         if (dquot->dq_dqb.dqb_curspace > number)
862                 dquot->dq_dqb.dqb_curspace -= number;
863         else
864                 dquot->dq_dqb.dqb_curspace = 0;
865         if (toqb(dquot->dq_dqb.dqb_curspace) <= dquot->dq_dqb.dqb_bsoftlimit)
866                 dquot->dq_dqb.dqb_btime = (time_t) 0;
867         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
868 }
869
870 static int warning_issued(struct dquot *dquot, const int warntype)
871 {
872         int flag = (warntype == QUOTA_NL_BHARDWARN ||
873                 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
874                 ((warntype == QUOTA_NL_IHARDWARN ||
875                 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
876
877         if (!flag)
878                 return 0;
879         return test_and_set_bit(flag, &dquot->dq_flags);
880 }
881
882 #ifdef CONFIG_PRINT_QUOTA_WARNING
883 static int flag_print_warnings = 1;
884
885 static inline int need_print_warning(struct dquot *dquot)
886 {
887         if (!flag_print_warnings)
888                 return 0;
889
890         switch (dquot->dq_type) {
891                 case USRQUOTA:
892                         return current_fsuid() == dquot->dq_id;
893                 case GRPQUOTA:
894                         return in_group_p(dquot->dq_id);
895         }
896         return 0;
897 }
898
899 /* Print warning to user which exceeded quota */
900 static void print_warning(struct dquot *dquot, const int warntype)
901 {
902         char *msg = NULL;
903         struct tty_struct *tty;
904
905         if (warntype == QUOTA_NL_IHARDBELOW ||
906             warntype == QUOTA_NL_ISOFTBELOW ||
907             warntype == QUOTA_NL_BHARDBELOW ||
908             warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(dquot))
909                 return;
910
911         tty = get_current_tty();
912         if (!tty)
913                 return;
914         tty_write_message(tty, dquot->dq_sb->s_id);
915         if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
916                 tty_write_message(tty, ": warning, ");
917         else
918                 tty_write_message(tty, ": write failed, ");
919         tty_write_message(tty, quotatypes[dquot->dq_type]);
920         switch (warntype) {
921                 case QUOTA_NL_IHARDWARN:
922                         msg = " file limit reached.\r\n";
923                         break;
924                 case QUOTA_NL_ISOFTLONGWARN:
925                         msg = " file quota exceeded too long.\r\n";
926                         break;
927                 case QUOTA_NL_ISOFTWARN:
928                         msg = " file quota exceeded.\r\n";
929                         break;
930                 case QUOTA_NL_BHARDWARN:
931                         msg = " block limit reached.\r\n";
932                         break;
933                 case QUOTA_NL_BSOFTLONGWARN:
934                         msg = " block quota exceeded too long.\r\n";
935                         break;
936                 case QUOTA_NL_BSOFTWARN:
937                         msg = " block quota exceeded.\r\n";
938                         break;
939         }
940         tty_write_message(tty, msg);
941         tty_kref_put(tty);
942 }
943 #endif
944
945 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
946
947 /* Netlink family structure for quota */
948 static struct genl_family quota_genl_family = {
949         .id = GENL_ID_GENERATE,
950         .hdrsize = 0,
951         .name = "VFS_DQUOT",
952         .version = 1,
953         .maxattr = QUOTA_NL_A_MAX,
954 };
955
956 /* Send warning to userspace about user which exceeded quota */
957 static void send_warning(const struct dquot *dquot, const char warntype)
958 {
959         static atomic_t seq;
960         struct sk_buff *skb;
961         void *msg_head;
962         int ret;
963         int msg_size = 4 * nla_total_size(sizeof(u32)) +
964                        2 * nla_total_size(sizeof(u64));
965
966         /* We have to allocate using GFP_NOFS as we are called from a
967          * filesystem performing write and thus further recursion into
968          * the fs to free some data could cause deadlocks. */
969         skb = genlmsg_new(msg_size, GFP_NOFS);
970         if (!skb) {
971                 printk(KERN_ERR
972                   "VFS: Not enough memory to send quota warning.\n");
973                 return;
974         }
975         msg_head = genlmsg_put(skb, 0, atomic_add_return(1, &seq),
976                         &quota_genl_family, 0, QUOTA_NL_C_WARNING);
977         if (!msg_head) {
978                 printk(KERN_ERR
979                   "VFS: Cannot store netlink header in quota warning.\n");
980                 goto err_out;
981         }
982         ret = nla_put_u32(skb, QUOTA_NL_A_QTYPE, dquot->dq_type);
983         if (ret)
984                 goto attr_err_out;
985         ret = nla_put_u64(skb, QUOTA_NL_A_EXCESS_ID, dquot->dq_id);
986         if (ret)
987                 goto attr_err_out;
988         ret = nla_put_u32(skb, QUOTA_NL_A_WARNING, warntype);
989         if (ret)
990                 goto attr_err_out;
991         ret = nla_put_u32(skb, QUOTA_NL_A_DEV_MAJOR,
992                 MAJOR(dquot->dq_sb->s_dev));
993         if (ret)
994                 goto attr_err_out;
995         ret = nla_put_u32(skb, QUOTA_NL_A_DEV_MINOR,
996                 MINOR(dquot->dq_sb->s_dev));
997         if (ret)
998                 goto attr_err_out;
999         ret = nla_put_u64(skb, QUOTA_NL_A_CAUSED_ID, current_uid());
1000         if (ret)
1001                 goto attr_err_out;
1002         genlmsg_end(skb, msg_head);
1003
1004         ret = genlmsg_multicast(skb, 0, quota_genl_family.id, GFP_NOFS);
1005         if (ret < 0 && ret != -ESRCH)
1006                 printk(KERN_ERR
1007                         "VFS: Failed to send notification message: %d\n", ret);
1008         return;
1009 attr_err_out:
1010         printk(KERN_ERR "VFS: Not enough space to compose quota message!\n");
1011 err_out:
1012         kfree_skb(skb);
1013 }
1014 #endif
1015
1016 static inline void flush_warnings(struct dquot * const *dquots, char *warntype)
1017 {
1018         int i;
1019
1020         for (i = 0; i < MAXQUOTAS; i++)
1021                 if (dquots[i] != NODQUOT && warntype[i] != QUOTA_NL_NOWARN &&
1022                     !warning_issued(dquots[i], warntype[i])) {
1023 #ifdef CONFIG_PRINT_QUOTA_WARNING
1024                         print_warning(dquots[i], warntype[i]);
1025 #endif
1026 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
1027                         send_warning(dquots[i], warntype[i]);
1028 #endif
1029                 }
1030 }
1031
1032 static inline char ignore_hardlimit(struct dquot *dquot)
1033 {
1034         struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_type];
1035
1036         return capable(CAP_SYS_RESOURCE) &&
1037             (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD || !(info->dqi_flags & V1_DQF_RSQUASH));
1038 }
1039
1040 /* needs dq_data_lock */
1041 static int check_idq(struct dquot *dquot, ulong inodes, char *warntype)
1042 {
1043         *warntype = QUOTA_NL_NOWARN;
1044         if (inodes <= 0 || test_bit(DQ_FAKE_B, &dquot->dq_flags))
1045                 return QUOTA_OK;
1046
1047         if (dquot->dq_dqb.dqb_ihardlimit &&
1048            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_ihardlimit &&
1049             !ignore_hardlimit(dquot)) {
1050                 *warntype = QUOTA_NL_IHARDWARN;
1051                 return NO_QUOTA;
1052         }
1053
1054         if (dquot->dq_dqb.dqb_isoftlimit &&
1055            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
1056             dquot->dq_dqb.dqb_itime && get_seconds() >= dquot->dq_dqb.dqb_itime &&
1057             !ignore_hardlimit(dquot)) {
1058                 *warntype = QUOTA_NL_ISOFTLONGWARN;
1059                 return NO_QUOTA;
1060         }
1061
1062         if (dquot->dq_dqb.dqb_isoftlimit &&
1063            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
1064             dquot->dq_dqb.dqb_itime == 0) {
1065                 *warntype = QUOTA_NL_ISOFTWARN;
1066                 dquot->dq_dqb.dqb_itime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
1067         }
1068
1069         return QUOTA_OK;
1070 }
1071
1072 /* needs dq_data_lock */
1073 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc, char *warntype)
1074 {
1075         *warntype = QUOTA_NL_NOWARN;
1076         if (space <= 0 || test_bit(DQ_FAKE_B, &dquot->dq_flags))
1077                 return QUOTA_OK;
1078
1079         if (dquot->dq_dqb.dqb_bhardlimit &&
1080            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bhardlimit &&
1081             !ignore_hardlimit(dquot)) {
1082                 if (!prealloc)
1083                         *warntype = QUOTA_NL_BHARDWARN;
1084                 return NO_QUOTA;
1085         }
1086
1087         if (dquot->dq_dqb.dqb_bsoftlimit &&
1088            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bsoftlimit &&
1089             dquot->dq_dqb.dqb_btime && get_seconds() >= dquot->dq_dqb.dqb_btime &&
1090             !ignore_hardlimit(dquot)) {
1091                 if (!prealloc)
1092                         *warntype = QUOTA_NL_BSOFTLONGWARN;
1093                 return NO_QUOTA;
1094         }
1095
1096         if (dquot->dq_dqb.dqb_bsoftlimit &&
1097            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bsoftlimit &&
1098             dquot->dq_dqb.dqb_btime == 0) {
1099                 if (!prealloc) {
1100                         *warntype = QUOTA_NL_BSOFTWARN;
1101                         dquot->dq_dqb.dqb_btime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_bgrace;
1102                 }
1103                 else
1104                         /*
1105                          * We don't allow preallocation to exceed softlimit so exceeding will
1106                          * be always printed
1107                          */
1108                         return NO_QUOTA;
1109         }
1110
1111         return QUOTA_OK;
1112 }
1113
1114 static int info_idq_free(struct dquot *dquot, ulong inodes)
1115 {
1116         if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1117             dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1118                 return QUOTA_NL_NOWARN;
1119
1120         if (dquot->dq_dqb.dqb_curinodes - inodes <= dquot->dq_dqb.dqb_isoftlimit)
1121                 return QUOTA_NL_ISOFTBELOW;
1122         if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1123             dquot->dq_dqb.dqb_curinodes - inodes < dquot->dq_dqb.dqb_ihardlimit)
1124                 return QUOTA_NL_IHARDBELOW;
1125         return QUOTA_NL_NOWARN;
1126 }
1127
1128 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1129 {
1130         if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1131             toqb(dquot->dq_dqb.dqb_curspace) <= dquot->dq_dqb.dqb_bsoftlimit)
1132                 return QUOTA_NL_NOWARN;
1133
1134         if (toqb(dquot->dq_dqb.dqb_curspace - space) <=
1135             dquot->dq_dqb.dqb_bsoftlimit)
1136                 return QUOTA_NL_BSOFTBELOW;
1137         if (toqb(dquot->dq_dqb.dqb_curspace) >= dquot->dq_dqb.dqb_bhardlimit &&
1138             toqb(dquot->dq_dqb.dqb_curspace - space) <
1139                                                 dquot->dq_dqb.dqb_bhardlimit)
1140                 return QUOTA_NL_BHARDBELOW;
1141         return QUOTA_NL_NOWARN;
1142 }
1143 /*
1144  *      Initialize quota pointers in inode
1145  *      Transaction must be started at entry
1146  */
1147 int dquot_initialize(struct inode *inode, int type)
1148 {
1149         unsigned int id = 0;
1150         int cnt, ret = 0;
1151
1152         /* First test before acquiring mutex - solves deadlocks when we
1153          * re-enter the quota code and are already holding the mutex */
1154         if (IS_NOQUOTA(inode))
1155                 return 0;
1156         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1157         /* Having dqptr_sem we know NOQUOTA flags can't be altered... */
1158         if (IS_NOQUOTA(inode))
1159                 goto out_err;
1160         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1161                 if (type != -1 && cnt != type)
1162                         continue;
1163                 if (inode->i_dquot[cnt] == NODQUOT) {
1164                         switch (cnt) {
1165                                 case USRQUOTA:
1166                                         id = inode->i_uid;
1167                                         break;
1168                                 case GRPQUOTA:
1169                                         id = inode->i_gid;
1170                                         break;
1171                         }
1172                         inode->i_dquot[cnt] = dqget(inode->i_sb, id, cnt);
1173                 }
1174         }
1175 out_err:
1176         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1177         return ret;
1178 }
1179
1180 /*
1181  *      Release all quotas referenced by inode
1182  *      Transaction must be started at an entry
1183  */
1184 int dquot_drop(struct inode *inode)
1185 {
1186         int cnt;
1187
1188         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1189         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1190                 if (inode->i_dquot[cnt] != NODQUOT) {
1191                         dqput(inode->i_dquot[cnt]);
1192                         inode->i_dquot[cnt] = NODQUOT;
1193                 }
1194         }
1195         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1196         return 0;
1197 }
1198
1199 /* Wrapper to remove references to quota structures from inode */
1200 void vfs_dq_drop(struct inode *inode)
1201 {
1202         /* Here we can get arbitrary inode from clear_inode() so we have
1203          * to be careful. OTOH we don't need locking as quota operations
1204          * are allowed to change only at mount time */
1205         if (!IS_NOQUOTA(inode) && inode->i_sb && inode->i_sb->dq_op
1206             && inode->i_sb->dq_op->drop) {
1207                 int cnt;
1208                 /* Test before calling to rule out calls from proc and such
1209                  * where we are not allowed to block. Note that this is
1210                  * actually reliable test even without the lock - the caller
1211                  * must assure that nobody can come after the DQUOT_DROP and
1212                  * add quota pointers back anyway */
1213                 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1214                         if (inode->i_dquot[cnt] != NODQUOT)
1215                                 break;
1216                 if (cnt < MAXQUOTAS)
1217                         inode->i_sb->dq_op->drop(inode);
1218         }
1219 }
1220
1221 /*
1222  * Following four functions update i_blocks+i_bytes fields and
1223  * quota information (together with appropriate checks)
1224  * NOTE: We absolutely rely on the fact that caller dirties
1225  * the inode (usually macros in quotaops.h care about this) and
1226  * holds a handle for the current transaction so that dquot write and
1227  * inode write go into the same transaction.
1228  */
1229
1230 /*
1231  * This operation can block, but only after everything is updated
1232  */
1233 int dquot_alloc_space(struct inode *inode, qsize_t number, int warn)
1234 {
1235         int cnt, ret = NO_QUOTA;
1236         char warntype[MAXQUOTAS];
1237
1238         /* First test before acquiring mutex - solves deadlocks when we
1239          * re-enter the quota code and are already holding the mutex */
1240         if (IS_NOQUOTA(inode)) {
1241 out_add:
1242                 inode_add_bytes(inode, number);
1243                 return QUOTA_OK;
1244         }
1245         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1246                 warntype[cnt] = QUOTA_NL_NOWARN;
1247
1248         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1249         if (IS_NOQUOTA(inode)) {        /* Now we can do reliable test... */
1250                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1251                 goto out_add;
1252         }
1253         spin_lock(&dq_data_lock);
1254         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1255                 if (inode->i_dquot[cnt] == NODQUOT)
1256                         continue;
1257                 if (check_bdq(inode->i_dquot[cnt], number, warn, warntype+cnt) == NO_QUOTA)
1258                         goto warn_put_all;
1259         }
1260         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1261                 if (inode->i_dquot[cnt] == NODQUOT)
1262                         continue;
1263                 dquot_incr_space(inode->i_dquot[cnt], number);
1264         }
1265         inode_add_bytes(inode, number);
1266         ret = QUOTA_OK;
1267 warn_put_all:
1268         spin_unlock(&dq_data_lock);
1269         if (ret == QUOTA_OK)
1270                 /* Dirtify all the dquots - this can block when journalling */
1271                 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1272                         if (inode->i_dquot[cnt])
1273                                 mark_dquot_dirty(inode->i_dquot[cnt]);
1274         flush_warnings(inode->i_dquot, warntype);
1275         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1276         return ret;
1277 }
1278
1279 /*
1280  * This operation can block, but only after everything is updated
1281  */
1282 int dquot_alloc_inode(const struct inode *inode, unsigned long number)
1283 {
1284         int cnt, ret = NO_QUOTA;
1285         char warntype[MAXQUOTAS];
1286
1287         /* First test before acquiring mutex - solves deadlocks when we
1288          * re-enter the quota code and are already holding the mutex */
1289         if (IS_NOQUOTA(inode))
1290                 return QUOTA_OK;
1291         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1292                 warntype[cnt] = QUOTA_NL_NOWARN;
1293         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1294         if (IS_NOQUOTA(inode)) {
1295                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1296                 return QUOTA_OK;
1297         }
1298         spin_lock(&dq_data_lock);
1299         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1300                 if (inode->i_dquot[cnt] == NODQUOT)
1301                         continue;
1302                 if (check_idq(inode->i_dquot[cnt], number, warntype+cnt) == NO_QUOTA)
1303                         goto warn_put_all;
1304         }
1305
1306         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1307                 if (inode->i_dquot[cnt] == NODQUOT)
1308                         continue;
1309                 dquot_incr_inodes(inode->i_dquot[cnt], number);
1310         }
1311         ret = QUOTA_OK;
1312 warn_put_all:
1313         spin_unlock(&dq_data_lock);
1314         if (ret == QUOTA_OK)
1315                 /* Dirtify all the dquots - this can block when journalling */
1316                 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1317                         if (inode->i_dquot[cnt])
1318                                 mark_dquot_dirty(inode->i_dquot[cnt]);
1319         flush_warnings(inode->i_dquot, warntype);
1320         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1321         return ret;
1322 }
1323
1324 /*
1325  * This operation can block, but only after everything is updated
1326  */
1327 int dquot_free_space(struct inode *inode, qsize_t number)
1328 {
1329         unsigned int cnt;
1330         char warntype[MAXQUOTAS];
1331
1332         /* First test before acquiring mutex - solves deadlocks when we
1333          * re-enter the quota code and are already holding the mutex */
1334         if (IS_NOQUOTA(inode)) {
1335 out_sub:
1336                 inode_sub_bytes(inode, number);
1337                 return QUOTA_OK;
1338         }
1339
1340         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1341         /* Now recheck reliably when holding dqptr_sem */
1342         if (IS_NOQUOTA(inode)) {
1343                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1344                 goto out_sub;
1345         }
1346         spin_lock(&dq_data_lock);
1347         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1348                 if (inode->i_dquot[cnt] == NODQUOT)
1349                         continue;
1350                 warntype[cnt] = info_bdq_free(inode->i_dquot[cnt], number);
1351                 dquot_decr_space(inode->i_dquot[cnt], number);
1352         }
1353         inode_sub_bytes(inode, number);
1354         spin_unlock(&dq_data_lock);
1355         /* Dirtify all the dquots - this can block when journalling */
1356         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1357                 if (inode->i_dquot[cnt])
1358                         mark_dquot_dirty(inode->i_dquot[cnt]);
1359         flush_warnings(inode->i_dquot, warntype);
1360         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1361         return QUOTA_OK;
1362 }
1363
1364 /*
1365  * This operation can block, but only after everything is updated
1366  */
1367 int dquot_free_inode(const struct inode *inode, unsigned long number)
1368 {
1369         unsigned int cnt;
1370         char warntype[MAXQUOTAS];
1371
1372         /* First test before acquiring mutex - solves deadlocks when we
1373          * re-enter the quota code and are already holding the mutex */
1374         if (IS_NOQUOTA(inode))
1375                 return QUOTA_OK;
1376
1377         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1378         /* Now recheck reliably when holding dqptr_sem */
1379         if (IS_NOQUOTA(inode)) {
1380                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1381                 return QUOTA_OK;
1382         }
1383         spin_lock(&dq_data_lock);
1384         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1385                 if (inode->i_dquot[cnt] == NODQUOT)
1386                         continue;
1387                 warntype[cnt] = info_idq_free(inode->i_dquot[cnt], number);
1388                 dquot_decr_inodes(inode->i_dquot[cnt], number);
1389         }
1390         spin_unlock(&dq_data_lock);
1391         /* Dirtify all the dquots - this can block when journalling */
1392         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1393                 if (inode->i_dquot[cnt])
1394                         mark_dquot_dirty(inode->i_dquot[cnt]);
1395         flush_warnings(inode->i_dquot, warntype);
1396         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1397         return QUOTA_OK;
1398 }
1399
1400 /*
1401  * Transfer the number of inode and blocks from one diskquota to an other.
1402  *
1403  * This operation can block, but only after everything is updated
1404  * A transaction must be started when entering this function.
1405  */
1406 int dquot_transfer(struct inode *inode, struct iattr *iattr)
1407 {
1408         qsize_t space;
1409         struct dquot *transfer_from[MAXQUOTAS];
1410         struct dquot *transfer_to[MAXQUOTAS];
1411         int cnt, ret = NO_QUOTA, chuid = (iattr->ia_valid & ATTR_UID) && inode->i_uid != iattr->ia_uid,
1412             chgid = (iattr->ia_valid & ATTR_GID) && inode->i_gid != iattr->ia_gid;
1413         char warntype_to[MAXQUOTAS];
1414         char warntype_from_inodes[MAXQUOTAS], warntype_from_space[MAXQUOTAS];
1415
1416         /* First test before acquiring mutex - solves deadlocks when we
1417          * re-enter the quota code and are already holding the mutex */
1418         if (IS_NOQUOTA(inode))
1419                 return QUOTA_OK;
1420         /* Clear the arrays */
1421         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1422                 transfer_to[cnt] = transfer_from[cnt] = NODQUOT;
1423                 warntype_to[cnt] = QUOTA_NL_NOWARN;
1424         }
1425         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1426         /* Now recheck reliably when holding dqptr_sem */
1427         if (IS_NOQUOTA(inode)) {        /* File without quota accounting? */
1428                 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1429                 return QUOTA_OK;
1430         }
1431         /* First build the transfer_to list - here we can block on
1432          * reading/instantiating of dquots.  We know that the transaction for
1433          * us was already started so we don't violate lock ranking here */
1434         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1435                 switch (cnt) {
1436                         case USRQUOTA:
1437                                 if (!chuid)
1438                                         continue;
1439                                 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_uid, cnt);
1440                                 break;
1441                         case GRPQUOTA:
1442                                 if (!chgid)
1443                                         continue;
1444                                 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_gid, cnt);
1445                                 break;
1446                 }
1447         }
1448         spin_lock(&dq_data_lock);
1449         space = inode_get_bytes(inode);
1450         /* Build the transfer_from list and check the limits */
1451         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1452                 if (transfer_to[cnt] == NODQUOT)
1453                         continue;
1454                 transfer_from[cnt] = inode->i_dquot[cnt];
1455                 if (check_idq(transfer_to[cnt], 1, warntype_to + cnt) ==
1456                     NO_QUOTA || check_bdq(transfer_to[cnt], space, 0,
1457                     warntype_to + cnt) == NO_QUOTA)
1458                         goto warn_put_all;
1459         }
1460
1461         /*
1462          * Finally perform the needed transfer from transfer_from to transfer_to
1463          */
1464         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1465                 /*
1466                  * Skip changes for same uid or gid or for turned off quota-type.
1467                  */
1468                 if (transfer_to[cnt] == NODQUOT)
1469                         continue;
1470
1471                 /* Due to IO error we might not have transfer_from[] structure */
1472                 if (transfer_from[cnt]) {
1473                         warntype_from_inodes[cnt] =
1474                                 info_idq_free(transfer_from[cnt], 1);
1475                         warntype_from_space[cnt] =
1476                                 info_bdq_free(transfer_from[cnt], space);
1477                         dquot_decr_inodes(transfer_from[cnt], 1);
1478                         dquot_decr_space(transfer_from[cnt], space);
1479                 }
1480
1481                 dquot_incr_inodes(transfer_to[cnt], 1);
1482                 dquot_incr_space(transfer_to[cnt], space);
1483
1484                 inode->i_dquot[cnt] = transfer_to[cnt];
1485         }
1486         ret = QUOTA_OK;
1487 warn_put_all:
1488         spin_unlock(&dq_data_lock);
1489         /* Dirtify all the dquots - this can block when journalling */
1490         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1491                 if (transfer_from[cnt])
1492                         mark_dquot_dirty(transfer_from[cnt]);
1493                 if (transfer_to[cnt])
1494                         mark_dquot_dirty(transfer_to[cnt]);
1495         }
1496         flush_warnings(transfer_to, warntype_to);
1497         flush_warnings(transfer_from, warntype_from_inodes);
1498         flush_warnings(transfer_from, warntype_from_space);
1499         
1500         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1501                 if (ret == QUOTA_OK && transfer_from[cnt] != NODQUOT)
1502                         dqput(transfer_from[cnt]);
1503                 if (ret == NO_QUOTA && transfer_to[cnt] != NODQUOT)
1504                         dqput(transfer_to[cnt]);
1505         }
1506         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1507         return ret;
1508 }
1509
1510 /* Wrapper for transferring ownership of an inode */
1511 int vfs_dq_transfer(struct inode *inode, struct iattr *iattr)
1512 {
1513         if (sb_any_quota_enabled(inode->i_sb) && !IS_NOQUOTA(inode)) {
1514                 vfs_dq_init(inode);
1515                 if (inode->i_sb->dq_op->transfer(inode, iattr) == NO_QUOTA)
1516                         return 1;
1517         }
1518         return 0;
1519 }
1520
1521
1522 /*
1523  * Write info of quota file to disk
1524  */
1525 int dquot_commit_info(struct super_block *sb, int type)
1526 {
1527         int ret;
1528         struct quota_info *dqopt = sb_dqopt(sb);
1529
1530         mutex_lock(&dqopt->dqio_mutex);
1531         ret = dqopt->ops[type]->write_file_info(sb, type);
1532         mutex_unlock(&dqopt->dqio_mutex);
1533         return ret;
1534 }
1535
1536 /*
1537  * Definitions of diskquota operations.
1538  */
1539 struct dquot_operations dquot_operations = {
1540         .initialize     = dquot_initialize,
1541         .drop           = dquot_drop,
1542         .alloc_space    = dquot_alloc_space,
1543         .alloc_inode    = dquot_alloc_inode,
1544         .free_space     = dquot_free_space,
1545         .free_inode     = dquot_free_inode,
1546         .transfer       = dquot_transfer,
1547         .write_dquot    = dquot_commit,
1548         .acquire_dquot  = dquot_acquire,
1549         .release_dquot  = dquot_release,
1550         .mark_dirty     = dquot_mark_dquot_dirty,
1551         .write_info     = dquot_commit_info,
1552         .alloc_dquot    = dquot_alloc,
1553         .destroy_dquot  = dquot_destroy,
1554 };
1555
1556 static inline void set_enable_flags(struct quota_info *dqopt, int type)
1557 {
1558         switch (type) {
1559                 case USRQUOTA:
1560                         dqopt->flags |= DQUOT_USR_ENABLED;
1561                         dqopt->flags &= ~DQUOT_USR_SUSPENDED;
1562                         break;
1563                 case GRPQUOTA:
1564                         dqopt->flags |= DQUOT_GRP_ENABLED;
1565                         dqopt->flags &= ~DQUOT_GRP_SUSPENDED;
1566                         break;
1567         }
1568 }
1569
1570 static inline void reset_enable_flags(struct quota_info *dqopt, int type,
1571                                       int remount)
1572 {
1573         switch (type) {
1574                 case USRQUOTA:
1575                         dqopt->flags &= ~DQUOT_USR_ENABLED;
1576                         if (remount)
1577                                 dqopt->flags |= DQUOT_USR_SUSPENDED;
1578                         else
1579                                 dqopt->flags &= ~DQUOT_USR_SUSPENDED;
1580                         break;
1581                 case GRPQUOTA:
1582                         dqopt->flags &= ~DQUOT_GRP_ENABLED;
1583                         if (remount)
1584                                 dqopt->flags |= DQUOT_GRP_SUSPENDED;
1585                         else
1586                                 dqopt->flags &= ~DQUOT_GRP_SUSPENDED;
1587                         break;
1588         }
1589 }
1590
1591
1592 /*
1593  * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
1594  */
1595 int vfs_quota_off(struct super_block *sb, int type, int remount)
1596 {
1597         int cnt, ret = 0;
1598         struct quota_info *dqopt = sb_dqopt(sb);
1599         struct inode *toputinode[MAXQUOTAS];
1600
1601         /* We need to serialize quota_off() for device */
1602         mutex_lock(&dqopt->dqonoff_mutex);
1603
1604         /*
1605          * Skip everything if there's nothing to do. We have to do this because
1606          * sometimes we are called when fill_super() failed and calling
1607          * sync_fs() in such cases does no good.
1608          */
1609         if (!sb_any_quota_enabled(sb) && !sb_any_quota_suspended(sb)) {
1610                 mutex_unlock(&dqopt->dqonoff_mutex);
1611                 return 0;
1612         }
1613         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1614                 toputinode[cnt] = NULL;
1615                 if (type != -1 && cnt != type)
1616                         continue;
1617                 /* If we keep inodes of quota files after remount and quotaoff
1618                  * is called, drop kept inodes. */
1619                 if (!remount && sb_has_quota_suspended(sb, cnt)) {
1620                         iput(dqopt->files[cnt]);
1621                         dqopt->files[cnt] = NULL;
1622                         reset_enable_flags(dqopt, cnt, 0);
1623                         continue;
1624                 }
1625                 if (!sb_has_quota_enabled(sb, cnt))
1626                         continue;
1627                 reset_enable_flags(dqopt, cnt, remount);
1628
1629                 /* Note: these are blocking operations */
1630                 drop_dquot_ref(sb, cnt);
1631                 invalidate_dquots(sb, cnt);
1632                 /*
1633                  * Now all dquots should be invalidated, all writes done so we should be only
1634                  * users of the info. No locks needed.
1635                  */
1636                 if (info_dirty(&dqopt->info[cnt]))
1637                         sb->dq_op->write_info(sb, cnt);
1638                 if (dqopt->ops[cnt]->free_file_info)
1639                         dqopt->ops[cnt]->free_file_info(sb, cnt);
1640                 put_quota_format(dqopt->info[cnt].dqi_format);
1641
1642                 toputinode[cnt] = dqopt->files[cnt];
1643                 if (!remount)
1644                         dqopt->files[cnt] = NULL;
1645                 dqopt->info[cnt].dqi_flags = 0;
1646                 dqopt->info[cnt].dqi_igrace = 0;
1647                 dqopt->info[cnt].dqi_bgrace = 0;
1648                 dqopt->ops[cnt] = NULL;
1649         }
1650         mutex_unlock(&dqopt->dqonoff_mutex);
1651         /* Sync the superblock so that buffers with quota data are written to
1652          * disk (and so userspace sees correct data afterwards). */
1653         if (sb->s_op->sync_fs)
1654                 sb->s_op->sync_fs(sb, 1);
1655         sync_blockdev(sb->s_bdev);
1656         /* Now the quota files are just ordinary files and we can set the
1657          * inode flags back. Moreover we discard the pagecache so that
1658          * userspace sees the writes we did bypassing the pagecache. We
1659          * must also discard the blockdev buffers so that we see the
1660          * changes done by userspace on the next quotaon() */
1661         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1662                 if (toputinode[cnt]) {
1663                         mutex_lock(&dqopt->dqonoff_mutex);
1664                         /* If quota was reenabled in the meantime, we have
1665                          * nothing to do */
1666                         if (!sb_has_quota_enabled(sb, cnt)) {
1667                                 mutex_lock_nested(&toputinode[cnt]->i_mutex, I_MUTEX_QUOTA);
1668                                 toputinode[cnt]->i_flags &= ~(S_IMMUTABLE |
1669                                   S_NOATIME | S_NOQUOTA);
1670                                 truncate_inode_pages(&toputinode[cnt]->i_data, 0);
1671                                 mutex_unlock(&toputinode[cnt]->i_mutex);
1672                                 mark_inode_dirty(toputinode[cnt]);
1673                         }
1674                         mutex_unlock(&dqopt->dqonoff_mutex);
1675                         /* On remount RO, we keep the inode pointer so that we
1676                          * can reenable quota on the subsequent remount RW.
1677                          * But we have better not keep inode pointer when there
1678                          * is pending delete on the quota file... */
1679                         if (!remount)
1680                                 iput(toputinode[cnt]);
1681                         else if (!toputinode[cnt]->i_nlink)
1682                                 ret = -EBUSY;
1683                 }
1684         if (sb->s_bdev)
1685                 invalidate_bdev(sb->s_bdev);
1686         return ret;
1687 }
1688
1689 /*
1690  *      Turn quotas on on a device
1691  */
1692
1693 /* Helper function when we already have the inode */
1694 static int vfs_quota_on_inode(struct inode *inode, int type, int format_id)
1695 {
1696         struct quota_format_type *fmt = find_quota_format(format_id);
1697         struct super_block *sb = inode->i_sb;
1698         struct quota_info *dqopt = sb_dqopt(sb);
1699         int error;
1700         int oldflags = -1;
1701
1702         if (!fmt)
1703                 return -ESRCH;
1704         if (!S_ISREG(inode->i_mode)) {
1705                 error = -EACCES;
1706                 goto out_fmt;
1707         }
1708         if (IS_RDONLY(inode)) {
1709                 error = -EROFS;
1710                 goto out_fmt;
1711         }
1712         if (!sb->s_op->quota_write || !sb->s_op->quota_read) {
1713                 error = -EINVAL;
1714                 goto out_fmt;
1715         }
1716
1717         /* As we bypass the pagecache we must now flush the inode so that
1718          * we see all the changes from userspace... */
1719         write_inode_now(inode, 1);
1720         /* And now flush the block cache so that kernel sees the changes */
1721         invalidate_bdev(sb->s_bdev);
1722         mutex_lock(&inode->i_mutex);
1723         mutex_lock(&dqopt->dqonoff_mutex);
1724         if (sb_has_quota_enabled(sb, type) ||
1725                         sb_has_quota_suspended(sb, type)) {
1726                 error = -EBUSY;
1727                 goto out_lock;
1728         }
1729         /* We don't want quota and atime on quota files (deadlocks possible)
1730          * Also nobody should write to the file - we use special IO operations
1731          * which ignore the immutable bit. */
1732         down_write(&dqopt->dqptr_sem);
1733         oldflags = inode->i_flags & (S_NOATIME | S_IMMUTABLE | S_NOQUOTA);
1734         inode->i_flags |= S_NOQUOTA | S_NOATIME | S_IMMUTABLE;
1735         up_write(&dqopt->dqptr_sem);
1736         sb->dq_op->drop(inode);
1737
1738         error = -EIO;
1739         dqopt->files[type] = igrab(inode);
1740         if (!dqopt->files[type])
1741                 goto out_lock;
1742         error = -EINVAL;
1743         if (!fmt->qf_ops->check_quota_file(sb, type))
1744                 goto out_file_init;
1745
1746         dqopt->ops[type] = fmt->qf_ops;
1747         dqopt->info[type].dqi_format = fmt;
1748         dqopt->info[type].dqi_fmt_id = format_id;
1749         INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
1750         mutex_lock(&dqopt->dqio_mutex);
1751         if ((error = dqopt->ops[type]->read_file_info(sb, type)) < 0) {
1752                 mutex_unlock(&dqopt->dqio_mutex);
1753                 goto out_file_init;
1754         }
1755         mutex_unlock(&dqopt->dqio_mutex);
1756         mutex_unlock(&inode->i_mutex);
1757         set_enable_flags(dqopt, type);
1758
1759         add_dquot_ref(sb, type);
1760         mutex_unlock(&dqopt->dqonoff_mutex);
1761
1762         return 0;
1763
1764 out_file_init:
1765         dqopt->files[type] = NULL;
1766         iput(inode);
1767 out_lock:
1768         mutex_unlock(&dqopt->dqonoff_mutex);
1769         if (oldflags != -1) {
1770                 down_write(&dqopt->dqptr_sem);
1771                 /* Set the flags back (in the case of accidental quotaon()
1772                  * on a wrong file we don't want to mess up the flags) */
1773                 inode->i_flags &= ~(S_NOATIME | S_NOQUOTA | S_IMMUTABLE);
1774                 inode->i_flags |= oldflags;
1775                 up_write(&dqopt->dqptr_sem);
1776         }
1777         mutex_unlock(&inode->i_mutex);
1778 out_fmt:
1779         put_quota_format(fmt);
1780
1781         return error; 
1782 }
1783
1784 /* Reenable quotas on remount RW */
1785 static int vfs_quota_on_remount(struct super_block *sb, int type)
1786 {
1787         struct quota_info *dqopt = sb_dqopt(sb);
1788         struct inode *inode;
1789         int ret;
1790
1791         mutex_lock(&dqopt->dqonoff_mutex);
1792         if (!sb_has_quota_suspended(sb, type)) {
1793                 mutex_unlock(&dqopt->dqonoff_mutex);
1794                 return 0;
1795         }
1796         BUG_ON(sb_has_quota_enabled(sb, type));
1797
1798         inode = dqopt->files[type];
1799         dqopt->files[type] = NULL;
1800         reset_enable_flags(dqopt, type, 0);
1801         mutex_unlock(&dqopt->dqonoff_mutex);
1802
1803         ret = vfs_quota_on_inode(inode, type, dqopt->info[type].dqi_fmt_id);
1804         iput(inode);
1805
1806         return ret;
1807 }
1808
1809 int vfs_quota_on_path(struct super_block *sb, int type, int format_id,
1810                       struct path *path)
1811 {
1812         int error = security_quota_on(path->dentry);
1813         if (error)
1814                 return error;
1815         /* Quota file not on the same filesystem? */
1816         if (path->mnt->mnt_sb != sb)
1817                 error = -EXDEV;
1818         else
1819                 error = vfs_quota_on_inode(path->dentry->d_inode, type,
1820                                            format_id);
1821         return error;
1822 }
1823
1824 /* Actual function called from quotactl() */
1825 int vfs_quota_on(struct super_block *sb, int type, int format_id, char *name,
1826                  int remount)
1827 {
1828         struct path path;
1829         int error;
1830
1831         if (remount)
1832                 return vfs_quota_on_remount(sb, type);
1833
1834         error = kern_path(name, LOOKUP_FOLLOW, &path);
1835         if (!error) {
1836                 error = vfs_quota_on_path(sb, type, format_id, &path);
1837                 path_put(&path);
1838         }
1839         return error;
1840 }
1841
1842 /*
1843  * This function is used when filesystem needs to initialize quotas
1844  * during mount time.
1845  */
1846 int vfs_quota_on_mount(struct super_block *sb, char *qf_name,
1847                 int format_id, int type)
1848 {
1849         struct dentry *dentry;
1850         int error;
1851
1852         dentry = lookup_one_len(qf_name, sb->s_root, strlen(qf_name));
1853         if (IS_ERR(dentry))
1854                 return PTR_ERR(dentry);
1855
1856         if (!dentry->d_inode) {
1857                 error = -ENOENT;
1858                 goto out;
1859         }
1860
1861         error = security_quota_on(dentry);
1862         if (!error)
1863                 error = vfs_quota_on_inode(dentry->d_inode, type, format_id);
1864
1865 out:
1866         dput(dentry);
1867         return error;
1868 }
1869
1870 /* Wrapper to turn on quotas when remounting rw */
1871 int vfs_dq_quota_on_remount(struct super_block *sb)
1872 {
1873         int cnt;
1874         int ret = 0, err;
1875
1876         if (!sb->s_qcop || !sb->s_qcop->quota_on)
1877                 return -ENOSYS;
1878         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1879                 err = sb->s_qcop->quota_on(sb, cnt, 0, NULL, 1);
1880                 if (err < 0 && !ret)
1881                         ret = err;
1882         }
1883         return ret;
1884 }
1885
1886 /* Generic routine for getting common part of quota structure */
1887 static void do_get_dqblk(struct dquot *dquot, struct if_dqblk *di)
1888 {
1889         struct mem_dqblk *dm = &dquot->dq_dqb;
1890
1891         spin_lock(&dq_data_lock);
1892         di->dqb_bhardlimit = dm->dqb_bhardlimit;
1893         di->dqb_bsoftlimit = dm->dqb_bsoftlimit;
1894         di->dqb_curspace = dm->dqb_curspace;
1895         di->dqb_ihardlimit = dm->dqb_ihardlimit;
1896         di->dqb_isoftlimit = dm->dqb_isoftlimit;
1897         di->dqb_curinodes = dm->dqb_curinodes;
1898         di->dqb_btime = dm->dqb_btime;
1899         di->dqb_itime = dm->dqb_itime;
1900         di->dqb_valid = QIF_ALL;
1901         spin_unlock(&dq_data_lock);
1902 }
1903
1904 int vfs_get_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
1905 {
1906         struct dquot *dquot;
1907
1908         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1909         if (!(dquot = dqget(sb, id, type))) {
1910                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1911                 return -ESRCH;
1912         }
1913         do_get_dqblk(dquot, di);
1914         dqput(dquot);
1915         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1916         return 0;
1917 }
1918
1919 /* Generic routine for setting common part of quota structure */
1920 static int do_set_dqblk(struct dquot *dquot, struct if_dqblk *di)
1921 {
1922         struct mem_dqblk *dm = &dquot->dq_dqb;
1923         int check_blim = 0, check_ilim = 0;
1924         struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_type];
1925
1926         if ((di->dqb_valid & QIF_BLIMITS &&
1927              (di->dqb_bhardlimit > dqi->dqi_maxblimit ||
1928               di->dqb_bsoftlimit > dqi->dqi_maxblimit)) ||
1929             (di->dqb_valid & QIF_ILIMITS &&
1930              (di->dqb_ihardlimit > dqi->dqi_maxilimit ||
1931               di->dqb_isoftlimit > dqi->dqi_maxilimit)))
1932                 return -ERANGE;
1933
1934         spin_lock(&dq_data_lock);
1935         if (di->dqb_valid & QIF_SPACE) {
1936                 dm->dqb_curspace = di->dqb_curspace;
1937                 check_blim = 1;
1938         }
1939         if (di->dqb_valid & QIF_BLIMITS) {
1940                 dm->dqb_bsoftlimit = di->dqb_bsoftlimit;
1941                 dm->dqb_bhardlimit = di->dqb_bhardlimit;
1942                 check_blim = 1;
1943         }
1944         if (di->dqb_valid & QIF_INODES) {
1945                 dm->dqb_curinodes = di->dqb_curinodes;
1946                 check_ilim = 1;
1947         }
1948         if (di->dqb_valid & QIF_ILIMITS) {
1949                 dm->dqb_isoftlimit = di->dqb_isoftlimit;
1950                 dm->dqb_ihardlimit = di->dqb_ihardlimit;
1951                 check_ilim = 1;
1952         }
1953         if (di->dqb_valid & QIF_BTIME)
1954                 dm->dqb_btime = di->dqb_btime;
1955         if (di->dqb_valid & QIF_ITIME)
1956                 dm->dqb_itime = di->dqb_itime;
1957
1958         if (check_blim) {
1959                 if (!dm->dqb_bsoftlimit || toqb(dm->dqb_curspace) < dm->dqb_bsoftlimit) {
1960                         dm->dqb_btime = 0;
1961                         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1962                 }
1963                 else if (!(di->dqb_valid & QIF_BTIME))  /* Set grace only if user hasn't provided his own... */
1964                         dm->dqb_btime = get_seconds() + dqi->dqi_bgrace;
1965         }
1966         if (check_ilim) {
1967                 if (!dm->dqb_isoftlimit || dm->dqb_curinodes < dm->dqb_isoftlimit) {
1968                         dm->dqb_itime = 0;
1969                         clear_bit(DQ_INODES_B, &dquot->dq_flags);
1970                 }
1971                 else if (!(di->dqb_valid & QIF_ITIME))  /* Set grace only if user hasn't provided his own... */
1972                         dm->dqb_itime = get_seconds() + dqi->dqi_igrace;
1973         }
1974         if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit || dm->dqb_isoftlimit)
1975                 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
1976         else
1977                 set_bit(DQ_FAKE_B, &dquot->dq_flags);
1978         spin_unlock(&dq_data_lock);
1979         mark_dquot_dirty(dquot);
1980
1981         return 0;
1982 }
1983
1984 int vfs_set_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
1985 {
1986         struct dquot *dquot;
1987         int rc;
1988
1989         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1990         if (!(dquot = dqget(sb, id, type))) {
1991                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1992                 return -ESRCH;
1993         }
1994         rc = do_set_dqblk(dquot, di);
1995         dqput(dquot);
1996         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1997         return rc;
1998 }
1999
2000 /* Generic routine for getting common part of quota file information */
2001 int vfs_get_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
2002 {
2003         struct mem_dqinfo *mi;
2004   
2005         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
2006         if (!sb_has_quota_enabled(sb, type)) {
2007                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2008                 return -ESRCH;
2009         }
2010         mi = sb_dqopt(sb)->info + type;
2011         spin_lock(&dq_data_lock);
2012         ii->dqi_bgrace = mi->dqi_bgrace;
2013         ii->dqi_igrace = mi->dqi_igrace;
2014         ii->dqi_flags = mi->dqi_flags & DQF_MASK;
2015         ii->dqi_valid = IIF_ALL;
2016         spin_unlock(&dq_data_lock);
2017         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2018         return 0;
2019 }
2020
2021 /* Generic routine for setting common part of quota file information */
2022 int vfs_set_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
2023 {
2024         struct mem_dqinfo *mi;
2025
2026         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
2027         if (!sb_has_quota_enabled(sb, type)) {
2028                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2029                 return -ESRCH;
2030         }
2031         mi = sb_dqopt(sb)->info + type;
2032         spin_lock(&dq_data_lock);
2033         if (ii->dqi_valid & IIF_BGRACE)
2034                 mi->dqi_bgrace = ii->dqi_bgrace;
2035         if (ii->dqi_valid & IIF_IGRACE)
2036                 mi->dqi_igrace = ii->dqi_igrace;
2037         if (ii->dqi_valid & IIF_FLAGS)
2038                 mi->dqi_flags = (mi->dqi_flags & ~DQF_MASK) | (ii->dqi_flags & DQF_MASK);
2039         spin_unlock(&dq_data_lock);
2040         mark_info_dirty(sb, type);
2041         /* Force write to disk */
2042         sb->dq_op->write_info(sb, type);
2043         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2044         return 0;
2045 }
2046
2047 struct quotactl_ops vfs_quotactl_ops = {
2048         .quota_on       = vfs_quota_on,
2049         .quota_off      = vfs_quota_off,
2050         .quota_sync     = vfs_quota_sync,
2051         .get_info       = vfs_get_dqinfo,
2052         .set_info       = vfs_set_dqinfo,
2053         .get_dqblk      = vfs_get_dqblk,
2054         .set_dqblk      = vfs_set_dqblk
2055 };
2056
2057 static ctl_table fs_dqstats_table[] = {
2058         {
2059                 .ctl_name       = FS_DQ_LOOKUPS,
2060                 .procname       = "lookups",
2061                 .data           = &dqstats.lookups,
2062                 .maxlen         = sizeof(int),
2063                 .mode           = 0444,
2064                 .proc_handler   = &proc_dointvec,
2065         },
2066         {
2067                 .ctl_name       = FS_DQ_DROPS,
2068                 .procname       = "drops",
2069                 .data           = &dqstats.drops,
2070                 .maxlen         = sizeof(int),
2071                 .mode           = 0444,
2072                 .proc_handler   = &proc_dointvec,
2073         },
2074         {
2075                 .ctl_name       = FS_DQ_READS,
2076                 .procname       = "reads",
2077                 .data           = &dqstats.reads,
2078                 .maxlen         = sizeof(int),
2079                 .mode           = 0444,
2080                 .proc_handler   = &proc_dointvec,
2081         },
2082         {
2083                 .ctl_name       = FS_DQ_WRITES,
2084                 .procname       = "writes",
2085                 .data           = &dqstats.writes,
2086                 .maxlen         = sizeof(int),
2087                 .mode           = 0444,
2088                 .proc_handler   = &proc_dointvec,
2089         },
2090         {
2091                 .ctl_name       = FS_DQ_CACHE_HITS,
2092                 .procname       = "cache_hits",
2093                 .data           = &dqstats.cache_hits,
2094                 .maxlen         = sizeof(int),
2095                 .mode           = 0444,
2096                 .proc_handler   = &proc_dointvec,
2097         },
2098         {
2099                 .ctl_name       = FS_DQ_ALLOCATED,
2100                 .procname       = "allocated_dquots",
2101                 .data           = &dqstats.allocated_dquots,
2102                 .maxlen         = sizeof(int),
2103                 .mode           = 0444,
2104                 .proc_handler   = &proc_dointvec,
2105         },
2106         {
2107                 .ctl_name       = FS_DQ_FREE,
2108                 .procname       = "free_dquots",
2109                 .data           = &dqstats.free_dquots,
2110                 .maxlen         = sizeof(int),
2111                 .mode           = 0444,
2112                 .proc_handler   = &proc_dointvec,
2113         },
2114         {
2115                 .ctl_name       = FS_DQ_SYNCS,
2116                 .procname       = "syncs",
2117                 .data           = &dqstats.syncs,
2118                 .maxlen         = sizeof(int),
2119                 .mode           = 0444,
2120                 .proc_handler   = &proc_dointvec,
2121         },
2122 #ifdef CONFIG_PRINT_QUOTA_WARNING
2123         {
2124                 .ctl_name       = FS_DQ_WARNINGS,
2125                 .procname       = "warnings",
2126                 .data           = &flag_print_warnings,
2127                 .maxlen         = sizeof(int),
2128                 .mode           = 0644,
2129                 .proc_handler   = &proc_dointvec,
2130         },
2131 #endif
2132         { .ctl_name = 0 },
2133 };
2134
2135 static ctl_table fs_table[] = {
2136         {
2137                 .ctl_name       = FS_DQSTATS,
2138                 .procname       = "quota",
2139                 .mode           = 0555,
2140                 .child          = fs_dqstats_table,
2141         },
2142         { .ctl_name = 0 },
2143 };
2144
2145 static ctl_table sys_table[] = {
2146         {
2147                 .ctl_name       = CTL_FS,
2148                 .procname       = "fs",
2149                 .mode           = 0555,
2150                 .child          = fs_table,
2151         },
2152         { .ctl_name = 0 },
2153 };
2154
2155 static int __init dquot_init(void)
2156 {
2157         int i;
2158         unsigned long nr_hash, order;
2159
2160         printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2161
2162         register_sysctl_table(sys_table);
2163
2164         dquot_cachep = kmem_cache_create("dquot",
2165                         sizeof(struct dquot), sizeof(unsigned long) * 4,
2166                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2167                                 SLAB_MEM_SPREAD|SLAB_PANIC),
2168                         NULL);
2169
2170         order = 0;
2171         dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
2172         if (!dquot_hash)
2173                 panic("Cannot create dquot hash table");
2174
2175         /* Find power-of-two hlist_heads which can fit into allocation */
2176         nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2177         dq_hash_bits = 0;
2178         do {
2179                 dq_hash_bits++;
2180         } while (nr_hash >> dq_hash_bits);
2181         dq_hash_bits--;
2182
2183         nr_hash = 1UL << dq_hash_bits;
2184         dq_hash_mask = nr_hash - 1;
2185         for (i = 0; i < nr_hash; i++)
2186                 INIT_HLIST_HEAD(dquot_hash + i);
2187
2188         printk("Dquot-cache hash table entries: %ld (order %ld, %ld bytes)\n",
2189                         nr_hash, order, (PAGE_SIZE << order));
2190
2191         register_shrinker(&dqcache_shrinker);
2192
2193 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
2194         if (genl_register_family(&quota_genl_family) != 0)
2195                 printk(KERN_ERR "VFS: Failed to create quota netlink interface.\n");
2196 #endif
2197
2198         return 0;
2199 }
2200 module_init(dquot_init);
2201
2202 EXPORT_SYMBOL(register_quota_format);
2203 EXPORT_SYMBOL(unregister_quota_format);
2204 EXPORT_SYMBOL(dqstats);
2205 EXPORT_SYMBOL(dq_data_lock);
2206 EXPORT_SYMBOL(vfs_quota_on);
2207 EXPORT_SYMBOL(vfs_quota_on_path);
2208 EXPORT_SYMBOL(vfs_quota_on_mount);
2209 EXPORT_SYMBOL(vfs_quota_off);
2210 EXPORT_SYMBOL(vfs_quota_sync);
2211 EXPORT_SYMBOL(vfs_get_dqinfo);
2212 EXPORT_SYMBOL(vfs_set_dqinfo);
2213 EXPORT_SYMBOL(vfs_get_dqblk);
2214 EXPORT_SYMBOL(vfs_set_dqblk);
2215 EXPORT_SYMBOL(dquot_commit);
2216 EXPORT_SYMBOL(dquot_commit_info);
2217 EXPORT_SYMBOL(dquot_acquire);
2218 EXPORT_SYMBOL(dquot_release);
2219 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
2220 EXPORT_SYMBOL(dquot_initialize);
2221 EXPORT_SYMBOL(dquot_drop);
2222 EXPORT_SYMBOL(vfs_dq_drop);
2223 EXPORT_SYMBOL(dquot_alloc_space);
2224 EXPORT_SYMBOL(dquot_alloc_inode);
2225 EXPORT_SYMBOL(dquot_free_space);
2226 EXPORT_SYMBOL(dquot_free_inode);
2227 EXPORT_SYMBOL(dquot_transfer);
2228 EXPORT_SYMBOL(vfs_dq_transfer);
2229 EXPORT_SYMBOL(vfs_dq_quota_on_remount);