]> www.pilppa.org Git - linux-2.6-omap-h63xx.git/blob - fs/dquot.c
quota: Allow negative usage of space and inodes
[linux-2.6-omap-h63xx.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  * 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_active(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_active(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_active(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_active(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, qsize_t 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, qsize_t number)
849 {
850         if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
851             dquot->dq_dqb.dqb_curinodes >= number)
852                 dquot->dq_dqb.dqb_curinodes -= number;
853         else
854                 dquot->dq_dqb.dqb_curinodes = 0;
855         if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
856                 dquot->dq_dqb.dqb_itime = (time_t) 0;
857         clear_bit(DQ_INODES_B, &dquot->dq_flags);
858 }
859
860 static inline void dquot_decr_space(struct dquot *dquot, qsize_t number)
861 {
862         if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
863             dquot->dq_dqb.dqb_curspace >= number)
864                 dquot->dq_dqb.dqb_curspace -= number;
865         else
866                 dquot->dq_dqb.dqb_curspace = 0;
867         if (dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
868                 dquot->dq_dqb.dqb_btime = (time_t) 0;
869         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
870 }
871
872 static int warning_issued(struct dquot *dquot, const int warntype)
873 {
874         int flag = (warntype == QUOTA_NL_BHARDWARN ||
875                 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
876                 ((warntype == QUOTA_NL_IHARDWARN ||
877                 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
878
879         if (!flag)
880                 return 0;
881         return test_and_set_bit(flag, &dquot->dq_flags);
882 }
883
884 #ifdef CONFIG_PRINT_QUOTA_WARNING
885 static int flag_print_warnings = 1;
886
887 static inline int need_print_warning(struct dquot *dquot)
888 {
889         if (!flag_print_warnings)
890                 return 0;
891
892         switch (dquot->dq_type) {
893                 case USRQUOTA:
894                         return current_fsuid() == dquot->dq_id;
895                 case GRPQUOTA:
896                         return in_group_p(dquot->dq_id);
897         }
898         return 0;
899 }
900
901 /* Print warning to user which exceeded quota */
902 static void print_warning(struct dquot *dquot, const int warntype)
903 {
904         char *msg = NULL;
905         struct tty_struct *tty;
906
907         if (warntype == QUOTA_NL_IHARDBELOW ||
908             warntype == QUOTA_NL_ISOFTBELOW ||
909             warntype == QUOTA_NL_BHARDBELOW ||
910             warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(dquot))
911                 return;
912
913         tty = get_current_tty();
914         if (!tty)
915                 return;
916         tty_write_message(tty, dquot->dq_sb->s_id);
917         if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
918                 tty_write_message(tty, ": warning, ");
919         else
920                 tty_write_message(tty, ": write failed, ");
921         tty_write_message(tty, quotatypes[dquot->dq_type]);
922         switch (warntype) {
923                 case QUOTA_NL_IHARDWARN:
924                         msg = " file limit reached.\r\n";
925                         break;
926                 case QUOTA_NL_ISOFTLONGWARN:
927                         msg = " file quota exceeded too long.\r\n";
928                         break;
929                 case QUOTA_NL_ISOFTWARN:
930                         msg = " file quota exceeded.\r\n";
931                         break;
932                 case QUOTA_NL_BHARDWARN:
933                         msg = " block limit reached.\r\n";
934                         break;
935                 case QUOTA_NL_BSOFTLONGWARN:
936                         msg = " block quota exceeded too long.\r\n";
937                         break;
938                 case QUOTA_NL_BSOFTWARN:
939                         msg = " block quota exceeded.\r\n";
940                         break;
941         }
942         tty_write_message(tty, msg);
943         tty_kref_put(tty);
944 }
945 #endif
946
947 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
948
949 /* Netlink family structure for quota */
950 static struct genl_family quota_genl_family = {
951         .id = GENL_ID_GENERATE,
952         .hdrsize = 0,
953         .name = "VFS_DQUOT",
954         .version = 1,
955         .maxattr = QUOTA_NL_A_MAX,
956 };
957
958 /* Send warning to userspace about user which exceeded quota */
959 static void send_warning(const struct dquot *dquot, const char warntype)
960 {
961         static atomic_t seq;
962         struct sk_buff *skb;
963         void *msg_head;
964         int ret;
965         int msg_size = 4 * nla_total_size(sizeof(u32)) +
966                        2 * nla_total_size(sizeof(u64));
967
968         /* We have to allocate using GFP_NOFS as we are called from a
969          * filesystem performing write and thus further recursion into
970          * the fs to free some data could cause deadlocks. */
971         skb = genlmsg_new(msg_size, GFP_NOFS);
972         if (!skb) {
973                 printk(KERN_ERR
974                   "VFS: Not enough memory to send quota warning.\n");
975                 return;
976         }
977         msg_head = genlmsg_put(skb, 0, atomic_add_return(1, &seq),
978                         &quota_genl_family, 0, QUOTA_NL_C_WARNING);
979         if (!msg_head) {
980                 printk(KERN_ERR
981                   "VFS: Cannot store netlink header in quota warning.\n");
982                 goto err_out;
983         }
984         ret = nla_put_u32(skb, QUOTA_NL_A_QTYPE, dquot->dq_type);
985         if (ret)
986                 goto attr_err_out;
987         ret = nla_put_u64(skb, QUOTA_NL_A_EXCESS_ID, dquot->dq_id);
988         if (ret)
989                 goto attr_err_out;
990         ret = nla_put_u32(skb, QUOTA_NL_A_WARNING, warntype);
991         if (ret)
992                 goto attr_err_out;
993         ret = nla_put_u32(skb, QUOTA_NL_A_DEV_MAJOR,
994                 MAJOR(dquot->dq_sb->s_dev));
995         if (ret)
996                 goto attr_err_out;
997         ret = nla_put_u32(skb, QUOTA_NL_A_DEV_MINOR,
998                 MINOR(dquot->dq_sb->s_dev));
999         if (ret)
1000                 goto attr_err_out;
1001         ret = nla_put_u64(skb, QUOTA_NL_A_CAUSED_ID, current_uid());
1002         if (ret)
1003                 goto attr_err_out;
1004         genlmsg_end(skb, msg_head);
1005
1006         ret = genlmsg_multicast(skb, 0, quota_genl_family.id, GFP_NOFS);
1007         if (ret < 0 && ret != -ESRCH)
1008                 printk(KERN_ERR
1009                         "VFS: Failed to send notification message: %d\n", ret);
1010         return;
1011 attr_err_out:
1012         printk(KERN_ERR "VFS: Not enough space to compose quota message!\n");
1013 err_out:
1014         kfree_skb(skb);
1015 }
1016 #endif
1017
1018 static inline void flush_warnings(struct dquot * const *dquots, char *warntype)
1019 {
1020         int i;
1021
1022         for (i = 0; i < MAXQUOTAS; i++)
1023                 if (dquots[i] != NODQUOT && warntype[i] != QUOTA_NL_NOWARN &&
1024                     !warning_issued(dquots[i], warntype[i])) {
1025 #ifdef CONFIG_PRINT_QUOTA_WARNING
1026                         print_warning(dquots[i], warntype[i]);
1027 #endif
1028 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
1029                         send_warning(dquots[i], warntype[i]);
1030 #endif
1031                 }
1032 }
1033
1034 static inline char ignore_hardlimit(struct dquot *dquot)
1035 {
1036         struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_type];
1037
1038         return capable(CAP_SYS_RESOURCE) &&
1039             (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD || !(info->dqi_flags & V1_DQF_RSQUASH));
1040 }
1041
1042 /* needs dq_data_lock */
1043 static int check_idq(struct dquot *dquot, qsize_t inodes, char *warntype)
1044 {
1045         *warntype = QUOTA_NL_NOWARN;
1046         if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_type) ||
1047             test_bit(DQ_FAKE_B, &dquot->dq_flags))
1048                 return QUOTA_OK;
1049
1050         if (dquot->dq_dqb.dqb_ihardlimit &&
1051            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_ihardlimit &&
1052             !ignore_hardlimit(dquot)) {
1053                 *warntype = QUOTA_NL_IHARDWARN;
1054                 return NO_QUOTA;
1055         }
1056
1057         if (dquot->dq_dqb.dqb_isoftlimit &&
1058            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
1059             dquot->dq_dqb.dqb_itime && get_seconds() >= dquot->dq_dqb.dqb_itime &&
1060             !ignore_hardlimit(dquot)) {
1061                 *warntype = QUOTA_NL_ISOFTLONGWARN;
1062                 return NO_QUOTA;
1063         }
1064
1065         if (dquot->dq_dqb.dqb_isoftlimit &&
1066            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
1067             dquot->dq_dqb.dqb_itime == 0) {
1068                 *warntype = QUOTA_NL_ISOFTWARN;
1069                 dquot->dq_dqb.dqb_itime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
1070         }
1071
1072         return QUOTA_OK;
1073 }
1074
1075 /* needs dq_data_lock */
1076 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc, char *warntype)
1077 {
1078         *warntype = QUOTA_NL_NOWARN;
1079         if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_type) ||
1080             test_bit(DQ_FAKE_B, &dquot->dq_flags))
1081                 return QUOTA_OK;
1082
1083         if (dquot->dq_dqb.dqb_bhardlimit &&
1084             dquot->dq_dqb.dqb_curspace + space > dquot->dq_dqb.dqb_bhardlimit &&
1085             !ignore_hardlimit(dquot)) {
1086                 if (!prealloc)
1087                         *warntype = QUOTA_NL_BHARDWARN;
1088                 return NO_QUOTA;
1089         }
1090
1091         if (dquot->dq_dqb.dqb_bsoftlimit &&
1092             dquot->dq_dqb.dqb_curspace + space > dquot->dq_dqb.dqb_bsoftlimit &&
1093             dquot->dq_dqb.dqb_btime && get_seconds() >= dquot->dq_dqb.dqb_btime &&
1094             !ignore_hardlimit(dquot)) {
1095                 if (!prealloc)
1096                         *warntype = QUOTA_NL_BSOFTLONGWARN;
1097                 return NO_QUOTA;
1098         }
1099
1100         if (dquot->dq_dqb.dqb_bsoftlimit &&
1101             dquot->dq_dqb.dqb_curspace + space > dquot->dq_dqb.dqb_bsoftlimit &&
1102             dquot->dq_dqb.dqb_btime == 0) {
1103                 if (!prealloc) {
1104                         *warntype = QUOTA_NL_BSOFTWARN;
1105                         dquot->dq_dqb.dqb_btime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_bgrace;
1106                 }
1107                 else
1108                         /*
1109                          * We don't allow preallocation to exceed softlimit so exceeding will
1110                          * be always printed
1111                          */
1112                         return NO_QUOTA;
1113         }
1114
1115         return QUOTA_OK;
1116 }
1117
1118 static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1119 {
1120         if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1121             dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1122             !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_type))
1123                 return QUOTA_NL_NOWARN;
1124
1125         if (dquot->dq_dqb.dqb_curinodes - inodes <= dquot->dq_dqb.dqb_isoftlimit)
1126                 return QUOTA_NL_ISOFTBELOW;
1127         if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1128             dquot->dq_dqb.dqb_curinodes - inodes < dquot->dq_dqb.dqb_ihardlimit)
1129                 return QUOTA_NL_IHARDBELOW;
1130         return QUOTA_NL_NOWARN;
1131 }
1132
1133 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1134 {
1135         if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1136             dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1137                 return QUOTA_NL_NOWARN;
1138
1139         if (dquot->dq_dqb.dqb_curspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1140                 return QUOTA_NL_BSOFTBELOW;
1141         if (dquot->dq_dqb.dqb_curspace >= dquot->dq_dqb.dqb_bhardlimit &&
1142             dquot->dq_dqb.dqb_curspace - space < dquot->dq_dqb.dqb_bhardlimit)
1143                 return QUOTA_NL_BHARDBELOW;
1144         return QUOTA_NL_NOWARN;
1145 }
1146 /*
1147  *      Initialize quota pointers in inode
1148  *      Transaction must be started at entry
1149  */
1150 int dquot_initialize(struct inode *inode, int type)
1151 {
1152         unsigned int id = 0;
1153         int cnt, ret = 0;
1154
1155         /* First test before acquiring mutex - solves deadlocks when we
1156          * re-enter the quota code and are already holding the mutex */
1157         if (IS_NOQUOTA(inode))
1158                 return 0;
1159         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1160         /* Having dqptr_sem we know NOQUOTA flags can't be altered... */
1161         if (IS_NOQUOTA(inode))
1162                 goto out_err;
1163         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1164                 if (type != -1 && cnt != type)
1165                         continue;
1166                 if (inode->i_dquot[cnt] == NODQUOT) {
1167                         switch (cnt) {
1168                                 case USRQUOTA:
1169                                         id = inode->i_uid;
1170                                         break;
1171                                 case GRPQUOTA:
1172                                         id = inode->i_gid;
1173                                         break;
1174                         }
1175                         inode->i_dquot[cnt] = dqget(inode->i_sb, id, cnt);
1176                 }
1177         }
1178 out_err:
1179         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1180         return ret;
1181 }
1182
1183 /*
1184  *      Release all quotas referenced by inode
1185  *      Transaction must be started at an entry
1186  */
1187 int dquot_drop(struct inode *inode)
1188 {
1189         int cnt;
1190
1191         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1192         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1193                 if (inode->i_dquot[cnt] != NODQUOT) {
1194                         dqput(inode->i_dquot[cnt]);
1195                         inode->i_dquot[cnt] = NODQUOT;
1196                 }
1197         }
1198         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1199         return 0;
1200 }
1201
1202 /* Wrapper to remove references to quota structures from inode */
1203 void vfs_dq_drop(struct inode *inode)
1204 {
1205         /* Here we can get arbitrary inode from clear_inode() so we have
1206          * to be careful. OTOH we don't need locking as quota operations
1207          * are allowed to change only at mount time */
1208         if (!IS_NOQUOTA(inode) && inode->i_sb && inode->i_sb->dq_op
1209             && inode->i_sb->dq_op->drop) {
1210                 int cnt;
1211                 /* Test before calling to rule out calls from proc and such
1212                  * where we are not allowed to block. Note that this is
1213                  * actually reliable test even without the lock - the caller
1214                  * must assure that nobody can come after the DQUOT_DROP and
1215                  * add quota pointers back anyway */
1216                 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1217                         if (inode->i_dquot[cnt] != NODQUOT)
1218                                 break;
1219                 if (cnt < MAXQUOTAS)
1220                         inode->i_sb->dq_op->drop(inode);
1221         }
1222 }
1223
1224 /*
1225  * Following four functions update i_blocks+i_bytes fields and
1226  * quota information (together with appropriate checks)
1227  * NOTE: We absolutely rely on the fact that caller dirties
1228  * the inode (usually macros in quotaops.h care about this) and
1229  * holds a handle for the current transaction so that dquot write and
1230  * inode write go into the same transaction.
1231  */
1232
1233 /*
1234  * This operation can block, but only after everything is updated
1235  */
1236 int dquot_alloc_space(struct inode *inode, qsize_t number, int warn)
1237 {
1238         int cnt, ret = NO_QUOTA;
1239         char warntype[MAXQUOTAS];
1240
1241         /* First test before acquiring mutex - solves deadlocks when we
1242          * re-enter the quota code and are already holding the mutex */
1243         if (IS_NOQUOTA(inode)) {
1244 out_add:
1245                 inode_add_bytes(inode, number);
1246                 return QUOTA_OK;
1247         }
1248         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1249                 warntype[cnt] = QUOTA_NL_NOWARN;
1250
1251         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1252         if (IS_NOQUOTA(inode)) {        /* Now we can do reliable test... */
1253                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1254                 goto out_add;
1255         }
1256         spin_lock(&dq_data_lock);
1257         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1258                 if (inode->i_dquot[cnt] == NODQUOT)
1259                         continue;
1260                 if (check_bdq(inode->i_dquot[cnt], number, warn, warntype+cnt) == NO_QUOTA)
1261                         goto warn_put_all;
1262         }
1263         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1264                 if (inode->i_dquot[cnt] == NODQUOT)
1265                         continue;
1266                 dquot_incr_space(inode->i_dquot[cnt], number);
1267         }
1268         inode_add_bytes(inode, number);
1269         ret = QUOTA_OK;
1270 warn_put_all:
1271         spin_unlock(&dq_data_lock);
1272         if (ret == QUOTA_OK)
1273                 /* Dirtify all the dquots - this can block when journalling */
1274                 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1275                         if (inode->i_dquot[cnt])
1276                                 mark_dquot_dirty(inode->i_dquot[cnt]);
1277         flush_warnings(inode->i_dquot, warntype);
1278         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1279         return ret;
1280 }
1281
1282 /*
1283  * This operation can block, but only after everything is updated
1284  */
1285 int dquot_alloc_inode(const struct inode *inode, qsize_t number)
1286 {
1287         int cnt, ret = NO_QUOTA;
1288         char warntype[MAXQUOTAS];
1289
1290         /* First test before acquiring mutex - solves deadlocks when we
1291          * re-enter the quota code and are already holding the mutex */
1292         if (IS_NOQUOTA(inode))
1293                 return QUOTA_OK;
1294         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1295                 warntype[cnt] = QUOTA_NL_NOWARN;
1296         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1297         if (IS_NOQUOTA(inode)) {
1298                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1299                 return QUOTA_OK;
1300         }
1301         spin_lock(&dq_data_lock);
1302         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1303                 if (inode->i_dquot[cnt] == NODQUOT)
1304                         continue;
1305                 if (check_idq(inode->i_dquot[cnt], number, warntype+cnt) == NO_QUOTA)
1306                         goto warn_put_all;
1307         }
1308
1309         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1310                 if (inode->i_dquot[cnt] == NODQUOT)
1311                         continue;
1312                 dquot_incr_inodes(inode->i_dquot[cnt], number);
1313         }
1314         ret = QUOTA_OK;
1315 warn_put_all:
1316         spin_unlock(&dq_data_lock);
1317         if (ret == QUOTA_OK)
1318                 /* Dirtify all the dquots - this can block when journalling */
1319                 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1320                         if (inode->i_dquot[cnt])
1321                                 mark_dquot_dirty(inode->i_dquot[cnt]);
1322         flush_warnings(inode->i_dquot, warntype);
1323         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1324         return ret;
1325 }
1326
1327 /*
1328  * This operation can block, but only after everything is updated
1329  */
1330 int dquot_free_space(struct inode *inode, qsize_t number)
1331 {
1332         unsigned int cnt;
1333         char warntype[MAXQUOTAS];
1334
1335         /* First test before acquiring mutex - solves deadlocks when we
1336          * re-enter the quota code and are already holding the mutex */
1337         if (IS_NOQUOTA(inode)) {
1338 out_sub:
1339                 inode_sub_bytes(inode, number);
1340                 return QUOTA_OK;
1341         }
1342
1343         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1344         /* Now recheck reliably when holding dqptr_sem */
1345         if (IS_NOQUOTA(inode)) {
1346                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1347                 goto out_sub;
1348         }
1349         spin_lock(&dq_data_lock);
1350         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1351                 if (inode->i_dquot[cnt] == NODQUOT)
1352                         continue;
1353                 warntype[cnt] = info_bdq_free(inode->i_dquot[cnt], number);
1354                 dquot_decr_space(inode->i_dquot[cnt], number);
1355         }
1356         inode_sub_bytes(inode, number);
1357         spin_unlock(&dq_data_lock);
1358         /* Dirtify all the dquots - this can block when journalling */
1359         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1360                 if (inode->i_dquot[cnt])
1361                         mark_dquot_dirty(inode->i_dquot[cnt]);
1362         flush_warnings(inode->i_dquot, warntype);
1363         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1364         return QUOTA_OK;
1365 }
1366
1367 /*
1368  * This operation can block, but only after everything is updated
1369  */
1370 int dquot_free_inode(const struct inode *inode, qsize_t number)
1371 {
1372         unsigned int cnt;
1373         char warntype[MAXQUOTAS];
1374
1375         /* First test before acquiring mutex - solves deadlocks when we
1376          * re-enter the quota code and are already holding the mutex */
1377         if (IS_NOQUOTA(inode))
1378                 return QUOTA_OK;
1379
1380         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1381         /* Now recheck reliably when holding dqptr_sem */
1382         if (IS_NOQUOTA(inode)) {
1383                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1384                 return QUOTA_OK;
1385         }
1386         spin_lock(&dq_data_lock);
1387         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1388                 if (inode->i_dquot[cnt] == NODQUOT)
1389                         continue;
1390                 warntype[cnt] = info_idq_free(inode->i_dquot[cnt], number);
1391                 dquot_decr_inodes(inode->i_dquot[cnt], number);
1392         }
1393         spin_unlock(&dq_data_lock);
1394         /* Dirtify all the dquots - this can block when journalling */
1395         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1396                 if (inode->i_dquot[cnt])
1397                         mark_dquot_dirty(inode->i_dquot[cnt]);
1398         flush_warnings(inode->i_dquot, warntype);
1399         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1400         return QUOTA_OK;
1401 }
1402
1403 /*
1404  * Transfer the number of inode and blocks from one diskquota to an other.
1405  *
1406  * This operation can block, but only after everything is updated
1407  * A transaction must be started when entering this function.
1408  */
1409 int dquot_transfer(struct inode *inode, struct iattr *iattr)
1410 {
1411         qsize_t space;
1412         struct dquot *transfer_from[MAXQUOTAS];
1413         struct dquot *transfer_to[MAXQUOTAS];
1414         int cnt, ret = NO_QUOTA, chuid = (iattr->ia_valid & ATTR_UID) && inode->i_uid != iattr->ia_uid,
1415             chgid = (iattr->ia_valid & ATTR_GID) && inode->i_gid != iattr->ia_gid;
1416         char warntype_to[MAXQUOTAS];
1417         char warntype_from_inodes[MAXQUOTAS], warntype_from_space[MAXQUOTAS];
1418
1419         /* First test before acquiring mutex - solves deadlocks when we
1420          * re-enter the quota code and are already holding the mutex */
1421         if (IS_NOQUOTA(inode))
1422                 return QUOTA_OK;
1423         /* Clear the arrays */
1424         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1425                 transfer_to[cnt] = transfer_from[cnt] = NODQUOT;
1426                 warntype_to[cnt] = QUOTA_NL_NOWARN;
1427         }
1428         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1429         /* Now recheck reliably when holding dqptr_sem */
1430         if (IS_NOQUOTA(inode)) {        /* File without quota accounting? */
1431                 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1432                 return QUOTA_OK;
1433         }
1434         /* First build the transfer_to list - here we can block on
1435          * reading/instantiating of dquots.  We know that the transaction for
1436          * us was already started so we don't violate lock ranking here */
1437         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1438                 switch (cnt) {
1439                         case USRQUOTA:
1440                                 if (!chuid)
1441                                         continue;
1442                                 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_uid, cnt);
1443                                 break;
1444                         case GRPQUOTA:
1445                                 if (!chgid)
1446                                         continue;
1447                                 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_gid, cnt);
1448                                 break;
1449                 }
1450         }
1451         spin_lock(&dq_data_lock);
1452         space = inode_get_bytes(inode);
1453         /* Build the transfer_from list and check the limits */
1454         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1455                 if (transfer_to[cnt] == NODQUOT)
1456                         continue;
1457                 transfer_from[cnt] = inode->i_dquot[cnt];
1458                 if (check_idq(transfer_to[cnt], 1, warntype_to + cnt) ==
1459                     NO_QUOTA || check_bdq(transfer_to[cnt], space, 0,
1460                     warntype_to + cnt) == NO_QUOTA)
1461                         goto warn_put_all;
1462         }
1463
1464         /*
1465          * Finally perform the needed transfer from transfer_from to transfer_to
1466          */
1467         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1468                 /*
1469                  * Skip changes for same uid or gid or for turned off quota-type.
1470                  */
1471                 if (transfer_to[cnt] == NODQUOT)
1472                         continue;
1473
1474                 /* Due to IO error we might not have transfer_from[] structure */
1475                 if (transfer_from[cnt]) {
1476                         warntype_from_inodes[cnt] =
1477                                 info_idq_free(transfer_from[cnt], 1);
1478                         warntype_from_space[cnt] =
1479                                 info_bdq_free(transfer_from[cnt], space);
1480                         dquot_decr_inodes(transfer_from[cnt], 1);
1481                         dquot_decr_space(transfer_from[cnt], space);
1482                 }
1483
1484                 dquot_incr_inodes(transfer_to[cnt], 1);
1485                 dquot_incr_space(transfer_to[cnt], space);
1486
1487                 inode->i_dquot[cnt] = transfer_to[cnt];
1488         }
1489         ret = QUOTA_OK;
1490 warn_put_all:
1491         spin_unlock(&dq_data_lock);
1492         /* Dirtify all the dquots - this can block when journalling */
1493         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1494                 if (transfer_from[cnt])
1495                         mark_dquot_dirty(transfer_from[cnt]);
1496                 if (transfer_to[cnt])
1497                         mark_dquot_dirty(transfer_to[cnt]);
1498         }
1499         flush_warnings(transfer_to, warntype_to);
1500         flush_warnings(transfer_from, warntype_from_inodes);
1501         flush_warnings(transfer_from, warntype_from_space);
1502         
1503         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1504                 if (ret == QUOTA_OK && transfer_from[cnt] != NODQUOT)
1505                         dqput(transfer_from[cnt]);
1506                 if (ret == NO_QUOTA && transfer_to[cnt] != NODQUOT)
1507                         dqput(transfer_to[cnt]);
1508         }
1509         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1510         return ret;
1511 }
1512
1513 /* Wrapper for transferring ownership of an inode */
1514 int vfs_dq_transfer(struct inode *inode, struct iattr *iattr)
1515 {
1516         if (sb_any_quota_active(inode->i_sb) && !IS_NOQUOTA(inode)) {
1517                 vfs_dq_init(inode);
1518                 if (inode->i_sb->dq_op->transfer(inode, iattr) == NO_QUOTA)
1519                         return 1;
1520         }
1521         return 0;
1522 }
1523
1524
1525 /*
1526  * Write info of quota file to disk
1527  */
1528 int dquot_commit_info(struct super_block *sb, int type)
1529 {
1530         int ret;
1531         struct quota_info *dqopt = sb_dqopt(sb);
1532
1533         mutex_lock(&dqopt->dqio_mutex);
1534         ret = dqopt->ops[type]->write_file_info(sb, type);
1535         mutex_unlock(&dqopt->dqio_mutex);
1536         return ret;
1537 }
1538
1539 /*
1540  * Definitions of diskquota operations.
1541  */
1542 struct dquot_operations dquot_operations = {
1543         .initialize     = dquot_initialize,
1544         .drop           = dquot_drop,
1545         .alloc_space    = dquot_alloc_space,
1546         .alloc_inode    = dquot_alloc_inode,
1547         .free_space     = dquot_free_space,
1548         .free_inode     = dquot_free_inode,
1549         .transfer       = dquot_transfer,
1550         .write_dquot    = dquot_commit,
1551         .acquire_dquot  = dquot_acquire,
1552         .release_dquot  = dquot_release,
1553         .mark_dirty     = dquot_mark_dquot_dirty,
1554         .write_info     = dquot_commit_info,
1555         .alloc_dquot    = dquot_alloc,
1556         .destroy_dquot  = dquot_destroy,
1557 };
1558
1559 /*
1560  * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
1561  */
1562 int vfs_quota_disable(struct super_block *sb, int type, unsigned int flags)
1563 {
1564         int cnt, ret = 0;
1565         struct quota_info *dqopt = sb_dqopt(sb);
1566         struct inode *toputinode[MAXQUOTAS];
1567
1568         /* Cannot turn off usage accounting without turning off limits, or
1569          * suspend quotas and simultaneously turn quotas off. */
1570         if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
1571             || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
1572             DQUOT_USAGE_ENABLED)))
1573                 return -EINVAL;
1574
1575         /* We need to serialize quota_off() for device */
1576         mutex_lock(&dqopt->dqonoff_mutex);
1577
1578         /*
1579          * Skip everything if there's nothing to do. We have to do this because
1580          * sometimes we are called when fill_super() failed and calling
1581          * sync_fs() in such cases does no good.
1582          */
1583         if (!sb_any_quota_loaded(sb)) {
1584                 mutex_unlock(&dqopt->dqonoff_mutex);
1585                 return 0;
1586         }
1587         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1588                 toputinode[cnt] = NULL;
1589                 if (type != -1 && cnt != type)
1590                         continue;
1591                 if (!sb_has_quota_loaded(sb, cnt))
1592                         continue;
1593
1594                 if (flags & DQUOT_SUSPENDED) {
1595                         dqopt->flags |=
1596                                 dquot_state_flag(DQUOT_SUSPENDED, cnt);
1597                 } else {
1598                         dqopt->flags &= ~dquot_state_flag(flags, cnt);
1599                         /* Turning off suspended quotas? */
1600                         if (!sb_has_quota_loaded(sb, cnt) &&
1601                             sb_has_quota_suspended(sb, cnt)) {
1602                                 dqopt->flags &= ~dquot_state_flag(
1603                                                         DQUOT_SUSPENDED, cnt);
1604                                 iput(dqopt->files[cnt]);
1605                                 dqopt->files[cnt] = NULL;
1606                                 continue;
1607                         }
1608                 }
1609
1610                 /* We still have to keep quota loaded? */
1611                 if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
1612                         continue;
1613
1614                 /* Note: these are blocking operations */
1615                 drop_dquot_ref(sb, cnt);
1616                 invalidate_dquots(sb, cnt);
1617                 /*
1618                  * Now all dquots should be invalidated, all writes done so we should be only
1619                  * users of the info. No locks needed.
1620                  */
1621                 if (info_dirty(&dqopt->info[cnt]))
1622                         sb->dq_op->write_info(sb, cnt);
1623                 if (dqopt->ops[cnt]->free_file_info)
1624                         dqopt->ops[cnt]->free_file_info(sb, cnt);
1625                 put_quota_format(dqopt->info[cnt].dqi_format);
1626
1627                 toputinode[cnt] = dqopt->files[cnt];
1628                 if (!sb_has_quota_loaded(sb, cnt))
1629                         dqopt->files[cnt] = NULL;
1630                 dqopt->info[cnt].dqi_flags = 0;
1631                 dqopt->info[cnt].dqi_igrace = 0;
1632                 dqopt->info[cnt].dqi_bgrace = 0;
1633                 dqopt->ops[cnt] = NULL;
1634         }
1635         mutex_unlock(&dqopt->dqonoff_mutex);
1636
1637         /* Skip syncing and setting flags if quota files are hidden */
1638         if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
1639                 goto put_inodes;
1640
1641         /* Sync the superblock so that buffers with quota data are written to
1642          * disk (and so userspace sees correct data afterwards). */
1643         if (sb->s_op->sync_fs)
1644                 sb->s_op->sync_fs(sb, 1);
1645         sync_blockdev(sb->s_bdev);
1646         /* Now the quota files are just ordinary files and we can set the
1647          * inode flags back. Moreover we discard the pagecache so that
1648          * userspace sees the writes we did bypassing the pagecache. We
1649          * must also discard the blockdev buffers so that we see the
1650          * changes done by userspace on the next quotaon() */
1651         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1652                 if (toputinode[cnt]) {
1653                         mutex_lock(&dqopt->dqonoff_mutex);
1654                         /* If quota was reenabled in the meantime, we have
1655                          * nothing to do */
1656                         if (!sb_has_quota_loaded(sb, cnt)) {
1657                                 mutex_lock_nested(&toputinode[cnt]->i_mutex, I_MUTEX_QUOTA);
1658                                 toputinode[cnt]->i_flags &= ~(S_IMMUTABLE |
1659                                   S_NOATIME | S_NOQUOTA);
1660                                 truncate_inode_pages(&toputinode[cnt]->i_data, 0);
1661                                 mutex_unlock(&toputinode[cnt]->i_mutex);
1662                                 mark_inode_dirty(toputinode[cnt]);
1663                         }
1664                         mutex_unlock(&dqopt->dqonoff_mutex);
1665                 }
1666         if (sb->s_bdev)
1667                 invalidate_bdev(sb->s_bdev);
1668 put_inodes:
1669         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1670                 if (toputinode[cnt]) {
1671                         /* On remount RO, we keep the inode pointer so that we
1672                          * can reenable quota on the subsequent remount RW. We
1673                          * have to check 'flags' variable and not use sb_has_
1674                          * function because another quotaon / quotaoff could
1675                          * change global state before we got here. We refuse
1676                          * to suspend quotas when there is pending delete on
1677                          * the quota file... */
1678                         if (!(flags & DQUOT_SUSPENDED))
1679                                 iput(toputinode[cnt]);
1680                         else if (!toputinode[cnt]->i_nlink)
1681                                 ret = -EBUSY;
1682                 }
1683         return ret;
1684 }
1685
1686 int vfs_quota_off(struct super_block *sb, int type, int remount)
1687 {
1688         return vfs_quota_disable(sb, type, remount ? DQUOT_SUSPENDED :
1689                                  (DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED));
1690 }
1691
1692 /*
1693  *      Turn quotas on on a device
1694  */
1695
1696 /*
1697  * Helper function to turn quotas on when we already have the inode of
1698  * quota file and no quota information is loaded.
1699  */
1700 static int vfs_load_quota_inode(struct inode *inode, int type, int format_id,
1701         unsigned int flags)
1702 {
1703         struct quota_format_type *fmt = find_quota_format(format_id);
1704         struct super_block *sb = inode->i_sb;
1705         struct quota_info *dqopt = sb_dqopt(sb);
1706         int error;
1707         int oldflags = -1;
1708
1709         if (!fmt)
1710                 return -ESRCH;
1711         if (!S_ISREG(inode->i_mode)) {
1712                 error = -EACCES;
1713                 goto out_fmt;
1714         }
1715         if (IS_RDONLY(inode)) {
1716                 error = -EROFS;
1717                 goto out_fmt;
1718         }
1719         if (!sb->s_op->quota_write || !sb->s_op->quota_read) {
1720                 error = -EINVAL;
1721                 goto out_fmt;
1722         }
1723         /* Usage always has to be set... */
1724         if (!(flags & DQUOT_USAGE_ENABLED)) {
1725                 error = -EINVAL;
1726                 goto out_fmt;
1727         }
1728
1729         if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
1730                 /* As we bypass the pagecache we must now flush the inode so
1731                  * that we see all the changes from userspace... */
1732                 write_inode_now(inode, 1);
1733                 /* And now flush the block cache so that kernel sees the
1734                  * changes */
1735                 invalidate_bdev(sb->s_bdev);
1736         }
1737         mutex_lock(&inode->i_mutex);
1738         mutex_lock(&dqopt->dqonoff_mutex);
1739         if (sb_has_quota_loaded(sb, type)) {
1740                 error = -EBUSY;
1741                 goto out_lock;
1742         }
1743
1744         if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
1745                 /* We don't want quota and atime on quota files (deadlocks
1746                  * possible) Also nobody should write to the file - we use
1747                  * special IO operations which ignore the immutable bit. */
1748                 down_write(&dqopt->dqptr_sem);
1749                 oldflags = inode->i_flags & (S_NOATIME | S_IMMUTABLE | S_NOQUOTA);
1750                 inode->i_flags |= S_NOQUOTA | S_NOATIME | S_IMMUTABLE;
1751                 up_write(&dqopt->dqptr_sem);
1752                 sb->dq_op->drop(inode);
1753         }
1754
1755         error = -EIO;
1756         dqopt->files[type] = igrab(inode);
1757         if (!dqopt->files[type])
1758                 goto out_lock;
1759         error = -EINVAL;
1760         if (!fmt->qf_ops->check_quota_file(sb, type))
1761                 goto out_file_init;
1762
1763         dqopt->ops[type] = fmt->qf_ops;
1764         dqopt->info[type].dqi_format = fmt;
1765         dqopt->info[type].dqi_fmt_id = format_id;
1766         INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
1767         mutex_lock(&dqopt->dqio_mutex);
1768         if ((error = dqopt->ops[type]->read_file_info(sb, type)) < 0) {
1769                 mutex_unlock(&dqopt->dqio_mutex);
1770                 goto out_file_init;
1771         }
1772         mutex_unlock(&dqopt->dqio_mutex);
1773         mutex_unlock(&inode->i_mutex);
1774         dqopt->flags |= dquot_state_flag(flags, type);
1775
1776         add_dquot_ref(sb, type);
1777         mutex_unlock(&dqopt->dqonoff_mutex);
1778
1779         return 0;
1780
1781 out_file_init:
1782         dqopt->files[type] = NULL;
1783         iput(inode);
1784 out_lock:
1785         mutex_unlock(&dqopt->dqonoff_mutex);
1786         if (oldflags != -1) {
1787                 down_write(&dqopt->dqptr_sem);
1788                 /* Set the flags back (in the case of accidental quotaon()
1789                  * on a wrong file we don't want to mess up the flags) */
1790                 inode->i_flags &= ~(S_NOATIME | S_NOQUOTA | S_IMMUTABLE);
1791                 inode->i_flags |= oldflags;
1792                 up_write(&dqopt->dqptr_sem);
1793         }
1794         mutex_unlock(&inode->i_mutex);
1795 out_fmt:
1796         put_quota_format(fmt);
1797
1798         return error; 
1799 }
1800
1801 /* Reenable quotas on remount RW */
1802 static int vfs_quota_on_remount(struct super_block *sb, int type)
1803 {
1804         struct quota_info *dqopt = sb_dqopt(sb);
1805         struct inode *inode;
1806         int ret;
1807         unsigned int flags;
1808
1809         mutex_lock(&dqopt->dqonoff_mutex);
1810         if (!sb_has_quota_suspended(sb, type)) {
1811                 mutex_unlock(&dqopt->dqonoff_mutex);
1812                 return 0;
1813         }
1814         inode = dqopt->files[type];
1815         dqopt->files[type] = NULL;
1816         flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
1817                                                 DQUOT_LIMITS_ENABLED, type);
1818         dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, type);
1819         mutex_unlock(&dqopt->dqonoff_mutex);
1820
1821         flags = dquot_generic_flag(flags, type);
1822         ret = vfs_load_quota_inode(inode, type, dqopt->info[type].dqi_fmt_id,
1823                                    flags);
1824         iput(inode);
1825
1826         return ret;
1827 }
1828
1829 int vfs_quota_on_path(struct super_block *sb, int type, int format_id,
1830                       struct path *path)
1831 {
1832         int error = security_quota_on(path->dentry);
1833         if (error)
1834                 return error;
1835         /* Quota file not on the same filesystem? */
1836         if (path->mnt->mnt_sb != sb)
1837                 error = -EXDEV;
1838         else
1839                 error = vfs_load_quota_inode(path->dentry->d_inode, type,
1840                                              format_id, DQUOT_USAGE_ENABLED |
1841                                              DQUOT_LIMITS_ENABLED);
1842         return error;
1843 }
1844
1845 int vfs_quota_on(struct super_block *sb, int type, int format_id, char *name,
1846                  int remount)
1847 {
1848         struct path path;
1849         int error;
1850
1851         if (remount)
1852                 return vfs_quota_on_remount(sb, type);
1853
1854         error = kern_path(name, LOOKUP_FOLLOW, &path);
1855         if (!error) {
1856                 error = vfs_quota_on_path(sb, type, format_id, &path);
1857                 path_put(&path);
1858         }
1859         return error;
1860 }
1861
1862 /*
1863  * More powerful function for turning on quotas allowing setting
1864  * of individual quota flags
1865  */
1866 int vfs_quota_enable(struct inode *inode, int type, int format_id,
1867                 unsigned int flags)
1868 {
1869         int ret = 0;
1870         struct super_block *sb = inode->i_sb;
1871         struct quota_info *dqopt = sb_dqopt(sb);
1872
1873         /* Just unsuspend quotas? */
1874         if (flags & DQUOT_SUSPENDED)
1875                 return vfs_quota_on_remount(sb, type);
1876         if (!flags)
1877                 return 0;
1878         /* Just updating flags needed? */
1879         if (sb_has_quota_loaded(sb, type)) {
1880                 mutex_lock(&dqopt->dqonoff_mutex);
1881                 /* Now do a reliable test... */
1882                 if (!sb_has_quota_loaded(sb, type)) {
1883                         mutex_unlock(&dqopt->dqonoff_mutex);
1884                         goto load_quota;
1885                 }
1886                 if (flags & DQUOT_USAGE_ENABLED &&
1887                     sb_has_quota_usage_enabled(sb, type)) {
1888                         ret = -EBUSY;
1889                         goto out_lock;
1890                 }
1891                 if (flags & DQUOT_LIMITS_ENABLED &&
1892                     sb_has_quota_limits_enabled(sb, type)) {
1893                         ret = -EBUSY;
1894                         goto out_lock;
1895                 }
1896                 sb_dqopt(sb)->flags |= dquot_state_flag(flags, type);
1897 out_lock:
1898                 mutex_unlock(&dqopt->dqonoff_mutex);
1899                 return ret;
1900         }
1901
1902 load_quota:
1903         return vfs_load_quota_inode(inode, type, format_id, flags);
1904 }
1905
1906 /*
1907  * This function is used when filesystem needs to initialize quotas
1908  * during mount time.
1909  */
1910 int vfs_quota_on_mount(struct super_block *sb, char *qf_name,
1911                 int format_id, int type)
1912 {
1913         struct dentry *dentry;
1914         int error;
1915
1916         dentry = lookup_one_len(qf_name, sb->s_root, strlen(qf_name));
1917         if (IS_ERR(dentry))
1918                 return PTR_ERR(dentry);
1919
1920         if (!dentry->d_inode) {
1921                 error = -ENOENT;
1922                 goto out;
1923         }
1924
1925         error = security_quota_on(dentry);
1926         if (!error)
1927                 error = vfs_load_quota_inode(dentry->d_inode, type, format_id,
1928                                 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
1929
1930 out:
1931         dput(dentry);
1932         return error;
1933 }
1934
1935 /* Wrapper to turn on quotas when remounting rw */
1936 int vfs_dq_quota_on_remount(struct super_block *sb)
1937 {
1938         int cnt;
1939         int ret = 0, err;
1940
1941         if (!sb->s_qcop || !sb->s_qcop->quota_on)
1942                 return -ENOSYS;
1943         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1944                 err = sb->s_qcop->quota_on(sb, cnt, 0, NULL, 1);
1945                 if (err < 0 && !ret)
1946                         ret = err;
1947         }
1948         return ret;
1949 }
1950
1951 static inline qsize_t qbtos(qsize_t blocks)
1952 {
1953         return blocks << QIF_DQBLKSIZE_BITS;
1954 }
1955
1956 static inline qsize_t stoqb(qsize_t space)
1957 {
1958         return (space + QIF_DQBLKSIZE - 1) >> QIF_DQBLKSIZE_BITS;
1959 }
1960
1961 /* Generic routine for getting common part of quota structure */
1962 static void do_get_dqblk(struct dquot *dquot, struct if_dqblk *di)
1963 {
1964         struct mem_dqblk *dm = &dquot->dq_dqb;
1965
1966         spin_lock(&dq_data_lock);
1967         di->dqb_bhardlimit = stoqb(dm->dqb_bhardlimit);
1968         di->dqb_bsoftlimit = stoqb(dm->dqb_bsoftlimit);
1969         di->dqb_curspace = dm->dqb_curspace;
1970         di->dqb_ihardlimit = dm->dqb_ihardlimit;
1971         di->dqb_isoftlimit = dm->dqb_isoftlimit;
1972         di->dqb_curinodes = dm->dqb_curinodes;
1973         di->dqb_btime = dm->dqb_btime;
1974         di->dqb_itime = dm->dqb_itime;
1975         di->dqb_valid = QIF_ALL;
1976         spin_unlock(&dq_data_lock);
1977 }
1978
1979 int vfs_get_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
1980 {
1981         struct dquot *dquot;
1982
1983         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1984         if (!(dquot = dqget(sb, id, type))) {
1985                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1986                 return -ESRCH;
1987         }
1988         do_get_dqblk(dquot, di);
1989         dqput(dquot);
1990         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1991         return 0;
1992 }
1993
1994 /* Generic routine for setting common part of quota structure */
1995 static int do_set_dqblk(struct dquot *dquot, struct if_dqblk *di)
1996 {
1997         struct mem_dqblk *dm = &dquot->dq_dqb;
1998         int check_blim = 0, check_ilim = 0;
1999         struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_type];
2000
2001         if ((di->dqb_valid & QIF_BLIMITS &&
2002              (di->dqb_bhardlimit > dqi->dqi_maxblimit ||
2003               di->dqb_bsoftlimit > dqi->dqi_maxblimit)) ||
2004             (di->dqb_valid & QIF_ILIMITS &&
2005              (di->dqb_ihardlimit > dqi->dqi_maxilimit ||
2006               di->dqb_isoftlimit > dqi->dqi_maxilimit)))
2007                 return -ERANGE;
2008
2009         spin_lock(&dq_data_lock);
2010         if (di->dqb_valid & QIF_SPACE) {
2011                 dm->dqb_curspace = di->dqb_curspace;
2012                 check_blim = 1;
2013         }
2014         if (di->dqb_valid & QIF_BLIMITS) {
2015                 dm->dqb_bsoftlimit = qbtos(di->dqb_bsoftlimit);
2016                 dm->dqb_bhardlimit = qbtos(di->dqb_bhardlimit);
2017                 check_blim = 1;
2018         }
2019         if (di->dqb_valid & QIF_INODES) {
2020                 dm->dqb_curinodes = di->dqb_curinodes;
2021                 check_ilim = 1;
2022         }
2023         if (di->dqb_valid & QIF_ILIMITS) {
2024                 dm->dqb_isoftlimit = di->dqb_isoftlimit;
2025                 dm->dqb_ihardlimit = di->dqb_ihardlimit;
2026                 check_ilim = 1;
2027         }
2028         if (di->dqb_valid & QIF_BTIME)
2029                 dm->dqb_btime = di->dqb_btime;
2030         if (di->dqb_valid & QIF_ITIME)
2031                 dm->dqb_itime = di->dqb_itime;
2032
2033         if (check_blim) {
2034                 if (!dm->dqb_bsoftlimit || dm->dqb_curspace < dm->dqb_bsoftlimit) {
2035                         dm->dqb_btime = 0;
2036                         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2037                 }
2038                 else if (!(di->dqb_valid & QIF_BTIME))  /* Set grace only if user hasn't provided his own... */
2039                         dm->dqb_btime = get_seconds() + dqi->dqi_bgrace;
2040         }
2041         if (check_ilim) {
2042                 if (!dm->dqb_isoftlimit || dm->dqb_curinodes < dm->dqb_isoftlimit) {
2043                         dm->dqb_itime = 0;
2044                         clear_bit(DQ_INODES_B, &dquot->dq_flags);
2045                 }
2046                 else if (!(di->dqb_valid & QIF_ITIME))  /* Set grace only if user hasn't provided his own... */
2047                         dm->dqb_itime = get_seconds() + dqi->dqi_igrace;
2048         }
2049         if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit || dm->dqb_isoftlimit)
2050                 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2051         else
2052                 set_bit(DQ_FAKE_B, &dquot->dq_flags);
2053         spin_unlock(&dq_data_lock);
2054         mark_dquot_dirty(dquot);
2055
2056         return 0;
2057 }
2058
2059 int vfs_set_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
2060 {
2061         struct dquot *dquot;
2062         int rc;
2063
2064         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
2065         dquot = dqget(sb, id, type);
2066         if (!dquot) {
2067                 rc = -ESRCH;
2068                 goto out;
2069         }
2070         rc = do_set_dqblk(dquot, di);
2071         dqput(dquot);
2072 out:
2073         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2074         return rc;
2075 }
2076
2077 /* Generic routine for getting common part of quota file information */
2078 int vfs_get_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
2079 {
2080         struct mem_dqinfo *mi;
2081   
2082         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
2083         if (!sb_has_quota_active(sb, type)) {
2084                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2085                 return -ESRCH;
2086         }
2087         mi = sb_dqopt(sb)->info + type;
2088         spin_lock(&dq_data_lock);
2089         ii->dqi_bgrace = mi->dqi_bgrace;
2090         ii->dqi_igrace = mi->dqi_igrace;
2091         ii->dqi_flags = mi->dqi_flags & DQF_MASK;
2092         ii->dqi_valid = IIF_ALL;
2093         spin_unlock(&dq_data_lock);
2094         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2095         return 0;
2096 }
2097
2098 /* Generic routine for setting common part of quota file information */
2099 int vfs_set_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
2100 {
2101         struct mem_dqinfo *mi;
2102         int err = 0;
2103
2104         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
2105         if (!sb_has_quota_active(sb, type)) {
2106                 err = -ESRCH;
2107                 goto out;
2108         }
2109         mi = sb_dqopt(sb)->info + type;
2110         spin_lock(&dq_data_lock);
2111         if (ii->dqi_valid & IIF_BGRACE)
2112                 mi->dqi_bgrace = ii->dqi_bgrace;
2113         if (ii->dqi_valid & IIF_IGRACE)
2114                 mi->dqi_igrace = ii->dqi_igrace;
2115         if (ii->dqi_valid & IIF_FLAGS)
2116                 mi->dqi_flags = (mi->dqi_flags & ~DQF_MASK) | (ii->dqi_flags & DQF_MASK);
2117         spin_unlock(&dq_data_lock);
2118         mark_info_dirty(sb, type);
2119         /* Force write to disk */
2120         sb->dq_op->write_info(sb, type);
2121 out:
2122         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2123         return err;
2124 }
2125
2126 struct quotactl_ops vfs_quotactl_ops = {
2127         .quota_on       = vfs_quota_on,
2128         .quota_off      = vfs_quota_off,
2129         .quota_sync     = vfs_quota_sync,
2130         .get_info       = vfs_get_dqinfo,
2131         .set_info       = vfs_set_dqinfo,
2132         .get_dqblk      = vfs_get_dqblk,
2133         .set_dqblk      = vfs_set_dqblk
2134 };
2135
2136 static ctl_table fs_dqstats_table[] = {
2137         {
2138                 .ctl_name       = FS_DQ_LOOKUPS,
2139                 .procname       = "lookups",
2140                 .data           = &dqstats.lookups,
2141                 .maxlen         = sizeof(int),
2142                 .mode           = 0444,
2143                 .proc_handler   = &proc_dointvec,
2144         },
2145         {
2146                 .ctl_name       = FS_DQ_DROPS,
2147                 .procname       = "drops",
2148                 .data           = &dqstats.drops,
2149                 .maxlen         = sizeof(int),
2150                 .mode           = 0444,
2151                 .proc_handler   = &proc_dointvec,
2152         },
2153         {
2154                 .ctl_name       = FS_DQ_READS,
2155                 .procname       = "reads",
2156                 .data           = &dqstats.reads,
2157                 .maxlen         = sizeof(int),
2158                 .mode           = 0444,
2159                 .proc_handler   = &proc_dointvec,
2160         },
2161         {
2162                 .ctl_name       = FS_DQ_WRITES,
2163                 .procname       = "writes",
2164                 .data           = &dqstats.writes,
2165                 .maxlen         = sizeof(int),
2166                 .mode           = 0444,
2167                 .proc_handler   = &proc_dointvec,
2168         },
2169         {
2170                 .ctl_name       = FS_DQ_CACHE_HITS,
2171                 .procname       = "cache_hits",
2172                 .data           = &dqstats.cache_hits,
2173                 .maxlen         = sizeof(int),
2174                 .mode           = 0444,
2175                 .proc_handler   = &proc_dointvec,
2176         },
2177         {
2178                 .ctl_name       = FS_DQ_ALLOCATED,
2179                 .procname       = "allocated_dquots",
2180                 .data           = &dqstats.allocated_dquots,
2181                 .maxlen         = sizeof(int),
2182                 .mode           = 0444,
2183                 .proc_handler   = &proc_dointvec,
2184         },
2185         {
2186                 .ctl_name       = FS_DQ_FREE,
2187                 .procname       = "free_dquots",
2188                 .data           = &dqstats.free_dquots,
2189                 .maxlen         = sizeof(int),
2190                 .mode           = 0444,
2191                 .proc_handler   = &proc_dointvec,
2192         },
2193         {
2194                 .ctl_name       = FS_DQ_SYNCS,
2195                 .procname       = "syncs",
2196                 .data           = &dqstats.syncs,
2197                 .maxlen         = sizeof(int),
2198                 .mode           = 0444,
2199                 .proc_handler   = &proc_dointvec,
2200         },
2201 #ifdef CONFIG_PRINT_QUOTA_WARNING
2202         {
2203                 .ctl_name       = FS_DQ_WARNINGS,
2204                 .procname       = "warnings",
2205                 .data           = &flag_print_warnings,
2206                 .maxlen         = sizeof(int),
2207                 .mode           = 0644,
2208                 .proc_handler   = &proc_dointvec,
2209         },
2210 #endif
2211         { .ctl_name = 0 },
2212 };
2213
2214 static ctl_table fs_table[] = {
2215         {
2216                 .ctl_name       = FS_DQSTATS,
2217                 .procname       = "quota",
2218                 .mode           = 0555,
2219                 .child          = fs_dqstats_table,
2220         },
2221         { .ctl_name = 0 },
2222 };
2223
2224 static ctl_table sys_table[] = {
2225         {
2226                 .ctl_name       = CTL_FS,
2227                 .procname       = "fs",
2228                 .mode           = 0555,
2229                 .child          = fs_table,
2230         },
2231         { .ctl_name = 0 },
2232 };
2233
2234 static int __init dquot_init(void)
2235 {
2236         int i;
2237         unsigned long nr_hash, order;
2238
2239         printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2240
2241         register_sysctl_table(sys_table);
2242
2243         dquot_cachep = kmem_cache_create("dquot",
2244                         sizeof(struct dquot), sizeof(unsigned long) * 4,
2245                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2246                                 SLAB_MEM_SPREAD|SLAB_PANIC),
2247                         NULL);
2248
2249         order = 0;
2250         dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
2251         if (!dquot_hash)
2252                 panic("Cannot create dquot hash table");
2253
2254         /* Find power-of-two hlist_heads which can fit into allocation */
2255         nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2256         dq_hash_bits = 0;
2257         do {
2258                 dq_hash_bits++;
2259         } while (nr_hash >> dq_hash_bits);
2260         dq_hash_bits--;
2261
2262         nr_hash = 1UL << dq_hash_bits;
2263         dq_hash_mask = nr_hash - 1;
2264         for (i = 0; i < nr_hash; i++)
2265                 INIT_HLIST_HEAD(dquot_hash + i);
2266
2267         printk("Dquot-cache hash table entries: %ld (order %ld, %ld bytes)\n",
2268                         nr_hash, order, (PAGE_SIZE << order));
2269
2270         register_shrinker(&dqcache_shrinker);
2271
2272 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
2273         if (genl_register_family(&quota_genl_family) != 0)
2274                 printk(KERN_ERR "VFS: Failed to create quota netlink interface.\n");
2275 #endif
2276
2277         return 0;
2278 }
2279 module_init(dquot_init);
2280
2281 EXPORT_SYMBOL(register_quota_format);
2282 EXPORT_SYMBOL(unregister_quota_format);
2283 EXPORT_SYMBOL(dqstats);
2284 EXPORT_SYMBOL(dq_data_lock);
2285 EXPORT_SYMBOL(vfs_quota_enable);
2286 EXPORT_SYMBOL(vfs_quota_on);
2287 EXPORT_SYMBOL(vfs_quota_on_path);
2288 EXPORT_SYMBOL(vfs_quota_on_mount);
2289 EXPORT_SYMBOL(vfs_quota_disable);
2290 EXPORT_SYMBOL(vfs_quota_off);
2291 EXPORT_SYMBOL(vfs_quota_sync);
2292 EXPORT_SYMBOL(vfs_get_dqinfo);
2293 EXPORT_SYMBOL(vfs_set_dqinfo);
2294 EXPORT_SYMBOL(vfs_get_dqblk);
2295 EXPORT_SYMBOL(vfs_set_dqblk);
2296 EXPORT_SYMBOL(dquot_commit);
2297 EXPORT_SYMBOL(dquot_commit_info);
2298 EXPORT_SYMBOL(dquot_acquire);
2299 EXPORT_SYMBOL(dquot_release);
2300 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
2301 EXPORT_SYMBOL(dquot_initialize);
2302 EXPORT_SYMBOL(dquot_drop);
2303 EXPORT_SYMBOL(vfs_dq_drop);
2304 EXPORT_SYMBOL(dquot_alloc_space);
2305 EXPORT_SYMBOL(dquot_alloc_inode);
2306 EXPORT_SYMBOL(dquot_free_space);
2307 EXPORT_SYMBOL(dquot_free_inode);
2308 EXPORT_SYMBOL(dquot_transfer);
2309 EXPORT_SYMBOL(vfs_dq_transfer);
2310 EXPORT_SYMBOL(vfs_dq_quota_on_remount);