Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / fs / quota / 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/sched.h>
75 #include <linux/kmod.h>
76 #include <linux/namei.h>
77 #include <linux/capability.h>
78 #include <linux/quotaops.h>
79 #include "../internal.h" /* ugh */
80
81 #include <linux/uaccess.h>
82
83 /*
84  * There are three quota SMP locks. dq_list_lock protects all lists with quotas
85  * and quota formats.
86  * dq_data_lock protects data from dq_dqb and also mem_dqinfo structures and
87  * also guards consistency of dquot->dq_dqb with inode->i_blocks, i_bytes.
88  * i_blocks and i_bytes updates itself are guarded by i_lock acquired directly
89  * in inode_add_bytes() and inode_sub_bytes(). dq_state_lock protects
90  * modifications of quota state (on quotaon and quotaoff) and readers who care
91  * about latest values take it as well.
92  *
93  * The spinlock ordering is hence: dq_data_lock > dq_list_lock > i_lock,
94  *   dq_list_lock > dq_state_lock
95  *
96  * Note that some things (eg. sb pointer, type, id) doesn't change during
97  * the life of the dquot structure and so needn't to be protected by a lock
98  *
99  * Operation accessing dquots via inode pointers are protected by dquot_srcu.
100  * Operation of reading pointer needs srcu_read_lock(&dquot_srcu), and
101  * synchronize_srcu(&dquot_srcu) is called after clearing pointers from
102  * inode and before dropping dquot references to avoid use of dquots after
103  * they are freed. dq_data_lock is used to serialize the pointer setting and
104  * clearing operations.
105  * Special care needs to be taken about S_NOQUOTA inode flag (marking that
106  * inode is a quota file). Functions adding pointers from inode to dquots have
107  * to check this flag under dq_data_lock and then (if S_NOQUOTA is not set) they
108  * have to do all pointer modifications before dropping dq_data_lock. This makes
109  * sure they cannot race with quotaon which first sets S_NOQUOTA flag and
110  * then drops all pointers to dquots from an inode.
111  *
112  * Each dquot has its dq_lock mutex. Locked dquots might not be referenced
113  * from inodes (dquot_alloc_space() and such don't check the dq_lock).
114  * Currently dquot is locked only when it is being read to memory (or space for
115  * it is being allocated) on the first dqget() and when it is being released on
116  * the last dqput(). The allocation and release oparations are serialized by
117  * the dq_lock and by checking the use count in dquot_release().  Write
118  * operations on dquots don't hold dq_lock as they copy data under dq_data_lock
119  * spinlock to internal buffers before writing.
120  *
121  * Lock ordering (including related VFS locks) is the following:
122  *   dqonoff_mutex > i_mutex > journal_lock > dquot->dq_lock > dqio_mutex
123  * dqonoff_mutex > i_mutex comes from dquot_quota_sync, dquot_enable, etc.
124  */
125
126 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_list_lock);
127 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_state_lock);
128 __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_data_lock);
129 EXPORT_SYMBOL(dq_data_lock);
130 DEFINE_STATIC_SRCU(dquot_srcu);
131
132 void __quota_error(struct super_block *sb, const char *func,
133                    const char *fmt, ...)
134 {
135         if (printk_ratelimit()) {
136                 va_list args;
137                 struct va_format vaf;
138
139                 va_start(args, fmt);
140
141                 vaf.fmt = fmt;
142                 vaf.va = &args;
143
144                 printk(KERN_ERR "Quota error (device %s): %s: %pV\n",
145                        sb->s_id, func, &vaf);
146
147                 va_end(args);
148         }
149 }
150 EXPORT_SYMBOL(__quota_error);
151
152 #if defined(CONFIG_QUOTA_DEBUG) || defined(CONFIG_PRINT_QUOTA_WARNING)
153 static char *quotatypes[] = INITQFNAMES;
154 #endif
155 static struct quota_format_type *quota_formats; /* List of registered formats */
156 static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
157
158 /* SLAB cache for dquot structures */
159 static struct kmem_cache *dquot_cachep;
160
161 int register_quota_format(struct quota_format_type *fmt)
162 {
163         spin_lock(&dq_list_lock);
164         fmt->qf_next = quota_formats;
165         quota_formats = fmt;
166         spin_unlock(&dq_list_lock);
167         return 0;
168 }
169 EXPORT_SYMBOL(register_quota_format);
170
171 void unregister_quota_format(struct quota_format_type *fmt)
172 {
173         struct quota_format_type **actqf;
174
175         spin_lock(&dq_list_lock);
176         for (actqf = &quota_formats; *actqf && *actqf != fmt;
177              actqf = &(*actqf)->qf_next)
178                 ;
179         if (*actqf)
180                 *actqf = (*actqf)->qf_next;
181         spin_unlock(&dq_list_lock);
182 }
183 EXPORT_SYMBOL(unregister_quota_format);
184
185 static struct quota_format_type *find_quota_format(int id)
186 {
187         struct quota_format_type *actqf;
188
189         spin_lock(&dq_list_lock);
190         for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
191              actqf = actqf->qf_next)
192                 ;
193         if (!actqf || !try_module_get(actqf->qf_owner)) {
194                 int qm;
195
196                 spin_unlock(&dq_list_lock);
197                 
198                 for (qm = 0; module_names[qm].qm_fmt_id &&
199                              module_names[qm].qm_fmt_id != id; qm++)
200                         ;
201                 if (!module_names[qm].qm_fmt_id ||
202                     request_module(module_names[qm].qm_mod_name))
203                         return NULL;
204
205                 spin_lock(&dq_list_lock);
206                 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
207                      actqf = actqf->qf_next)
208                         ;
209                 if (actqf && !try_module_get(actqf->qf_owner))
210                         actqf = NULL;
211         }
212         spin_unlock(&dq_list_lock);
213         return actqf;
214 }
215
216 static void put_quota_format(struct quota_format_type *fmt)
217 {
218         module_put(fmt->qf_owner);
219 }
220
221 /*
222  * Dquot List Management:
223  * The quota code uses three lists for dquot management: the inuse_list,
224  * free_dquots, and dquot_hash[] array. A single dquot structure may be
225  * on all three lists, depending on its current state.
226  *
227  * All dquots are placed to the end of inuse_list when first created, and this
228  * list is used for invalidate operation, which must look at every dquot.
229  *
230  * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
231  * and this list is searched whenever we need an available dquot.  Dquots are
232  * removed from the list as soon as they are used again, and
233  * dqstats.free_dquots gives the number of dquots on the list. When
234  * dquot is invalidated it's completely released from memory.
235  *
236  * Dquots with a specific identity (device, type and id) are placed on
237  * one of the dquot_hash[] hash chains. The provides an efficient search
238  * mechanism to locate a specific dquot.
239  */
240
241 static LIST_HEAD(inuse_list);
242 static LIST_HEAD(free_dquots);
243 static unsigned int dq_hash_bits, dq_hash_mask;
244 static struct hlist_head *dquot_hash;
245
246 struct dqstats dqstats;
247 EXPORT_SYMBOL(dqstats);
248
249 static qsize_t inode_get_rsv_space(struct inode *inode);
250 static void __dquot_initialize(struct inode *inode, int type);
251
252 static inline unsigned int
253 hashfn(const struct super_block *sb, struct kqid qid)
254 {
255         unsigned int id = from_kqid(&init_user_ns, qid);
256         int type = qid.type;
257         unsigned long tmp;
258
259         tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
260         return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
261 }
262
263 /*
264  * Following list functions expect dq_list_lock to be held
265  */
266 static inline void insert_dquot_hash(struct dquot *dquot)
267 {
268         struct hlist_head *head;
269         head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id);
270         hlist_add_head(&dquot->dq_hash, head);
271 }
272
273 static inline void remove_dquot_hash(struct dquot *dquot)
274 {
275         hlist_del_init(&dquot->dq_hash);
276 }
277
278 static struct dquot *find_dquot(unsigned int hashent, struct super_block *sb,
279                                 struct kqid qid)
280 {
281         struct hlist_node *node;
282         struct dquot *dquot;
283
284         hlist_for_each (node, dquot_hash+hashent) {
285                 dquot = hlist_entry(node, struct dquot, dq_hash);
286                 if (dquot->dq_sb == sb && qid_eq(dquot->dq_id, qid))
287                         return dquot;
288         }
289         return NULL;
290 }
291
292 /* Add a dquot to the tail of the free list */
293 static inline void put_dquot_last(struct dquot *dquot)
294 {
295         list_add_tail(&dquot->dq_free, &free_dquots);
296         dqstats_inc(DQST_FREE_DQUOTS);
297 }
298
299 static inline void remove_free_dquot(struct dquot *dquot)
300 {
301         if (list_empty(&dquot->dq_free))
302                 return;
303         list_del_init(&dquot->dq_free);
304         dqstats_dec(DQST_FREE_DQUOTS);
305 }
306
307 static inline void put_inuse(struct dquot *dquot)
308 {
309         /* We add to the back of inuse list so we don't have to restart
310          * when traversing this list and we block */
311         list_add_tail(&dquot->dq_inuse, &inuse_list);
312         dqstats_inc(DQST_ALLOC_DQUOTS);
313 }
314
315 static inline void remove_inuse(struct dquot *dquot)
316 {
317         dqstats_dec(DQST_ALLOC_DQUOTS);
318         list_del(&dquot->dq_inuse);
319 }
320 /*
321  * End of list functions needing dq_list_lock
322  */
323
324 static void wait_on_dquot(struct dquot *dquot)
325 {
326         mutex_lock(&dquot->dq_lock);
327         mutex_unlock(&dquot->dq_lock);
328 }
329
330 static inline int dquot_dirty(struct dquot *dquot)
331 {
332         return test_bit(DQ_MOD_B, &dquot->dq_flags);
333 }
334
335 static inline int mark_dquot_dirty(struct dquot *dquot)
336 {
337         return dquot->dq_sb->dq_op->mark_dirty(dquot);
338 }
339
340 /* Mark dquot dirty in atomic manner, and return it's old dirty flag state */
341 int dquot_mark_dquot_dirty(struct dquot *dquot)
342 {
343         int ret = 1;
344
345         /* If quota is dirty already, we don't have to acquire dq_list_lock */
346         if (test_bit(DQ_MOD_B, &dquot->dq_flags))
347                 return 1;
348
349         spin_lock(&dq_list_lock);
350         if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags)) {
351                 list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
352                                 info[dquot->dq_id.type].dqi_dirty_list);
353                 ret = 0;
354         }
355         spin_unlock(&dq_list_lock);
356         return ret;
357 }
358 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
359
360 /* Dirtify all the dquots - this can block when journalling */
361 static inline int mark_all_dquot_dirty(struct dquot * const *dquot)
362 {
363         int ret, err, cnt;
364
365         ret = err = 0;
366         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
367                 if (dquot[cnt])
368                         /* Even in case of error we have to continue */
369                         ret = mark_dquot_dirty(dquot[cnt]);
370                 if (!err)
371                         err = ret;
372         }
373         return err;
374 }
375
376 static inline void dqput_all(struct dquot **dquot)
377 {
378         unsigned int cnt;
379
380         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
381                 dqput(dquot[cnt]);
382 }
383
384 /* This function needs dq_list_lock */
385 static inline int clear_dquot_dirty(struct dquot *dquot)
386 {
387         if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags))
388                 return 0;
389         list_del_init(&dquot->dq_dirty);
390         return 1;
391 }
392
393 void mark_info_dirty(struct super_block *sb, int type)
394 {
395         set_bit(DQF_INFO_DIRTY_B, &sb_dqopt(sb)->info[type].dqi_flags);
396 }
397 EXPORT_SYMBOL(mark_info_dirty);
398
399 /*
400  *      Read dquot from disk and alloc space for it
401  */
402
403 int dquot_acquire(struct dquot *dquot)
404 {
405         int ret = 0, ret2 = 0;
406         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
407
408         mutex_lock(&dquot->dq_lock);
409         mutex_lock(&dqopt->dqio_mutex);
410         if (!test_bit(DQ_READ_B, &dquot->dq_flags))
411                 ret = dqopt->ops[dquot->dq_id.type]->read_dqblk(dquot);
412         if (ret < 0)
413                 goto out_iolock;
414         set_bit(DQ_READ_B, &dquot->dq_flags);
415         /* Instantiate dquot if needed */
416         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
417                 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
418                 /* Write the info if needed */
419                 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
420                         ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
421                                         dquot->dq_sb, dquot->dq_id.type);
422                 }
423                 if (ret < 0)
424                         goto out_iolock;
425                 if (ret2 < 0) {
426                         ret = ret2;
427                         goto out_iolock;
428                 }
429         }
430         set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
431 out_iolock:
432         mutex_unlock(&dqopt->dqio_mutex);
433         mutex_unlock(&dquot->dq_lock);
434         return ret;
435 }
436 EXPORT_SYMBOL(dquot_acquire);
437
438 /*
439  *      Write dquot to disk
440  */
441 int dquot_commit(struct dquot *dquot)
442 {
443         int ret = 0;
444         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
445
446         mutex_lock(&dqopt->dqio_mutex);
447         spin_lock(&dq_list_lock);
448         if (!clear_dquot_dirty(dquot)) {
449                 spin_unlock(&dq_list_lock);
450                 goto out_sem;
451         }
452         spin_unlock(&dq_list_lock);
453         /* Inactive dquot can be only if there was error during read/init
454          * => we have better not writing it */
455         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
456                 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
457         else
458                 ret = -EIO;
459 out_sem:
460         mutex_unlock(&dqopt->dqio_mutex);
461         return ret;
462 }
463 EXPORT_SYMBOL(dquot_commit);
464
465 /*
466  *      Release dquot
467  */
468 int dquot_release(struct dquot *dquot)
469 {
470         int ret = 0, ret2 = 0;
471         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
472
473         mutex_lock(&dquot->dq_lock);
474         /* Check whether we are not racing with some other dqget() */
475         if (atomic_read(&dquot->dq_count) > 1)
476                 goto out_dqlock;
477         mutex_lock(&dqopt->dqio_mutex);
478         if (dqopt->ops[dquot->dq_id.type]->release_dqblk) {
479                 ret = dqopt->ops[dquot->dq_id.type]->release_dqblk(dquot);
480                 /* Write the info */
481                 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
482                         ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
483                                                 dquot->dq_sb, dquot->dq_id.type);
484                 }
485                 if (ret >= 0)
486                         ret = ret2;
487         }
488         clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
489         mutex_unlock(&dqopt->dqio_mutex);
490 out_dqlock:
491         mutex_unlock(&dquot->dq_lock);
492         return ret;
493 }
494 EXPORT_SYMBOL(dquot_release);
495
496 void dquot_destroy(struct dquot *dquot)
497 {
498         kmem_cache_free(dquot_cachep, dquot);
499 }
500 EXPORT_SYMBOL(dquot_destroy);
501
502 static inline void do_destroy_dquot(struct dquot *dquot)
503 {
504         dquot->dq_sb->dq_op->destroy_dquot(dquot);
505 }
506
507 /* Invalidate all dquots on the list. Note that this function is called after
508  * quota is disabled and pointers from inodes removed so there cannot be new
509  * quota users. There can still be some users of quotas due to inodes being
510  * just deleted or pruned by prune_icache() (those are not attached to any
511  * list) or parallel quotactl call. We have to wait for such users.
512  */
513 static void invalidate_dquots(struct super_block *sb, int type)
514 {
515         struct dquot *dquot, *tmp;
516
517 restart:
518         spin_lock(&dq_list_lock);
519         list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
520                 if (dquot->dq_sb != sb)
521                         continue;
522                 if (dquot->dq_id.type != type)
523                         continue;
524                 /* Wait for dquot users */
525                 if (atomic_read(&dquot->dq_count)) {
526                         DEFINE_WAIT(wait);
527
528                         dqgrab(dquot);
529                         prepare_to_wait(&dquot->dq_wait_unused, &wait,
530                                         TASK_UNINTERRUPTIBLE);
531                         spin_unlock(&dq_list_lock);
532                         /* Once dqput() wakes us up, we know it's time to free
533                          * the dquot.
534                          * IMPORTANT: we rely on the fact that there is always
535                          * at most one process waiting for dquot to free.
536                          * Otherwise dq_count would be > 1 and we would never
537                          * wake up.
538                          */
539                         if (atomic_read(&dquot->dq_count) > 1)
540                                 schedule();
541                         finish_wait(&dquot->dq_wait_unused, &wait);
542                         dqput(dquot);
543                         /* At this moment dquot() need not exist (it could be
544                          * reclaimed by prune_dqcache(). Hence we must
545                          * restart. */
546                         goto restart;
547                 }
548                 /*
549                  * Quota now has no users and it has been written on last
550                  * dqput()
551                  */
552                 remove_dquot_hash(dquot);
553                 remove_free_dquot(dquot);
554                 remove_inuse(dquot);
555                 do_destroy_dquot(dquot);
556         }
557         spin_unlock(&dq_list_lock);
558 }
559
560 /* Call callback for every active dquot on given filesystem */
561 int dquot_scan_active(struct super_block *sb,
562                       int (*fn)(struct dquot *dquot, unsigned long priv),
563                       unsigned long priv)
564 {
565         struct dquot *dquot, *old_dquot = NULL;
566         int ret = 0;
567
568         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
569         spin_lock(&dq_list_lock);
570         list_for_each_entry(dquot, &inuse_list, dq_inuse) {
571                 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
572                         continue;
573                 if (dquot->dq_sb != sb)
574                         continue;
575                 /* Now we have active dquot so we can just increase use count */
576                 atomic_inc(&dquot->dq_count);
577                 spin_unlock(&dq_list_lock);
578                 dqstats_inc(DQST_LOOKUPS);
579                 dqput(old_dquot);
580                 old_dquot = dquot;
581                 /*
582                  * ->release_dquot() can be racing with us. Our reference
583                  * protects us from new calls to it so just wait for any
584                  * outstanding call and recheck the DQ_ACTIVE_B after that.
585                  */
586                 wait_on_dquot(dquot);
587                 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
588                         ret = fn(dquot, priv);
589                         if (ret < 0)
590                                 goto out;
591                 }
592                 spin_lock(&dq_list_lock);
593                 /* We are safe to continue now because our dquot could not
594                  * be moved out of the inuse list while we hold the reference */
595         }
596         spin_unlock(&dq_list_lock);
597 out:
598         dqput(old_dquot);
599         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
600         return ret;
601 }
602 EXPORT_SYMBOL(dquot_scan_active);
603
604 /* Write all dquot structures to quota files */
605 int dquot_writeback_dquots(struct super_block *sb, int type)
606 {
607         struct list_head *dirty;
608         struct dquot *dquot;
609         struct quota_info *dqopt = sb_dqopt(sb);
610         int cnt;
611         int err, ret = 0;
612
613         mutex_lock(&dqopt->dqonoff_mutex);
614         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
615                 if (type != -1 && cnt != type)
616                         continue;
617                 if (!sb_has_quota_active(sb, cnt))
618                         continue;
619                 spin_lock(&dq_list_lock);
620                 dirty = &dqopt->info[cnt].dqi_dirty_list;
621                 while (!list_empty(dirty)) {
622                         dquot = list_first_entry(dirty, struct dquot,
623                                                  dq_dirty);
624                         /* Dirty and inactive can be only bad dquot... */
625                         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
626                                 clear_dquot_dirty(dquot);
627                                 continue;
628                         }
629                         /* Now we have active dquot from which someone is
630                          * holding reference so we can safely just increase
631                          * use count */
632                         dqgrab(dquot);
633                         spin_unlock(&dq_list_lock);
634                         dqstats_inc(DQST_LOOKUPS);
635                         err = sb->dq_op->write_dquot(dquot);
636                         if (!ret && err)
637                                 ret = err;
638                         dqput(dquot);
639                         spin_lock(&dq_list_lock);
640                 }
641                 spin_unlock(&dq_list_lock);
642         }
643
644         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
645                 if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
646                     && info_dirty(&dqopt->info[cnt]))
647                         sb->dq_op->write_info(sb, cnt);
648         dqstats_inc(DQST_SYNCS);
649         mutex_unlock(&dqopt->dqonoff_mutex);
650
651         return ret;
652 }
653 EXPORT_SYMBOL(dquot_writeback_dquots);
654
655 /* Write all dquot structures to disk and make them visible from userspace */
656 int dquot_quota_sync(struct super_block *sb, int type)
657 {
658         struct quota_info *dqopt = sb_dqopt(sb);
659         int cnt;
660         int ret;
661
662         ret = dquot_writeback_dquots(sb, type);
663         if (ret)
664                 return ret;
665         if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
666                 return 0;
667
668         /* This is not very clever (and fast) but currently I don't know about
669          * any other simple way of getting quota data to disk and we must get
670          * them there for userspace to be visible... */
671         if (sb->s_op->sync_fs)
672                 sb->s_op->sync_fs(sb, 1);
673         sync_blockdev(sb->s_bdev);
674
675         /*
676          * Now when everything is written we can discard the pagecache so
677          * that userspace sees the changes.
678          */
679         mutex_lock(&dqopt->dqonoff_mutex);
680         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
681                 if (type != -1 && cnt != type)
682                         continue;
683                 if (!sb_has_quota_active(sb, cnt))
684                         continue;
685                 mutex_lock(&dqopt->files[cnt]->i_mutex);
686                 truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
687                 mutex_unlock(&dqopt->files[cnt]->i_mutex);
688         }
689         mutex_unlock(&dqopt->dqonoff_mutex);
690
691         return 0;
692 }
693 EXPORT_SYMBOL(dquot_quota_sync);
694
695 static unsigned long
696 dqcache_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
697 {
698         struct list_head *head;
699         struct dquot *dquot;
700         unsigned long freed = 0;
701
702         spin_lock(&dq_list_lock);
703         head = free_dquots.prev;
704         while (head != &free_dquots && sc->nr_to_scan) {
705                 dquot = list_entry(head, struct dquot, dq_free);
706                 remove_dquot_hash(dquot);
707                 remove_free_dquot(dquot);
708                 remove_inuse(dquot);
709                 do_destroy_dquot(dquot);
710                 sc->nr_to_scan--;
711                 freed++;
712                 head = free_dquots.prev;
713         }
714         spin_unlock(&dq_list_lock);
715         return freed;
716 }
717
718 static unsigned long
719 dqcache_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
720 {
721         return vfs_pressure_ratio(
722         percpu_counter_read_positive(&dqstats.counter[DQST_FREE_DQUOTS]));
723 }
724
725 static struct shrinker dqcache_shrinker = {
726         .count_objects = dqcache_shrink_count,
727         .scan_objects = dqcache_shrink_scan,
728         .seeks = DEFAULT_SEEKS,
729 };
730
731 /*
732  * Put reference to dquot
733  */
734 void dqput(struct dquot *dquot)
735 {
736         int ret;
737
738         if (!dquot)
739                 return;
740 #ifdef CONFIG_QUOTA_DEBUG
741         if (!atomic_read(&dquot->dq_count)) {
742                 quota_error(dquot->dq_sb, "trying to free free dquot of %s %d",
743                             quotatypes[dquot->dq_id.type],
744                             from_kqid(&init_user_ns, dquot->dq_id));
745                 BUG();
746         }
747 #endif
748         dqstats_inc(DQST_DROPS);
749 we_slept:
750         spin_lock(&dq_list_lock);
751         if (atomic_read(&dquot->dq_count) > 1) {
752                 /* We have more than one user... nothing to do */
753                 atomic_dec(&dquot->dq_count);
754                 /* Releasing dquot during quotaoff phase? */
755                 if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_id.type) &&
756                     atomic_read(&dquot->dq_count) == 1)
757                         wake_up(&dquot->dq_wait_unused);
758                 spin_unlock(&dq_list_lock);
759                 return;
760         }
761         /* Need to release dquot? */
762         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && dquot_dirty(dquot)) {
763                 spin_unlock(&dq_list_lock);
764                 /* Commit dquot before releasing */
765                 ret = dquot->dq_sb->dq_op->write_dquot(dquot);
766                 if (ret < 0) {
767                         quota_error(dquot->dq_sb, "Can't write quota structure"
768                                     " (error %d). Quota may get out of sync!",
769                                     ret);
770                         /*
771                          * We clear dirty bit anyway, so that we avoid
772                          * infinite loop here
773                          */
774                         spin_lock(&dq_list_lock);
775                         clear_dquot_dirty(dquot);
776                         spin_unlock(&dq_list_lock);
777                 }
778                 goto we_slept;
779         }
780         /* Clear flag in case dquot was inactive (something bad happened) */
781         clear_dquot_dirty(dquot);
782         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
783                 spin_unlock(&dq_list_lock);
784                 dquot->dq_sb->dq_op->release_dquot(dquot);
785                 goto we_slept;
786         }
787         atomic_dec(&dquot->dq_count);
788 #ifdef CONFIG_QUOTA_DEBUG
789         /* sanity check */
790         BUG_ON(!list_empty(&dquot->dq_free));
791 #endif
792         put_dquot_last(dquot);
793         spin_unlock(&dq_list_lock);
794 }
795 EXPORT_SYMBOL(dqput);
796
797 struct dquot *dquot_alloc(struct super_block *sb, int type)
798 {
799         return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
800 }
801 EXPORT_SYMBOL(dquot_alloc);
802
803 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
804 {
805         struct dquot *dquot;
806
807         dquot = sb->dq_op->alloc_dquot(sb, type);
808         if(!dquot)
809                 return NULL;
810
811         mutex_init(&dquot->dq_lock);
812         INIT_LIST_HEAD(&dquot->dq_free);
813         INIT_LIST_HEAD(&dquot->dq_inuse);
814         INIT_HLIST_NODE(&dquot->dq_hash);
815         INIT_LIST_HEAD(&dquot->dq_dirty);
816         init_waitqueue_head(&dquot->dq_wait_unused);
817         dquot->dq_sb = sb;
818         dquot->dq_id = make_kqid_invalid(type);
819         atomic_set(&dquot->dq_count, 1);
820
821         return dquot;
822 }
823
824 /*
825  * Get reference to dquot
826  *
827  * Locking is slightly tricky here. We are guarded from parallel quotaoff()
828  * destroying our dquot by:
829  *   a) checking for quota flags under dq_list_lock and
830  *   b) getting a reference to dquot before we release dq_list_lock
831  */
832 struct dquot *dqget(struct super_block *sb, struct kqid qid)
833 {
834         unsigned int hashent = hashfn(sb, qid);
835         struct dquot *dquot = NULL, *empty = NULL;
836
837         if (!sb_has_quota_active(sb, qid.type))
838                 return NULL;
839 we_slept:
840         spin_lock(&dq_list_lock);
841         spin_lock(&dq_state_lock);
842         if (!sb_has_quota_active(sb, qid.type)) {
843                 spin_unlock(&dq_state_lock);
844                 spin_unlock(&dq_list_lock);
845                 goto out;
846         }
847         spin_unlock(&dq_state_lock);
848
849         dquot = find_dquot(hashent, sb, qid);
850         if (!dquot) {
851                 if (!empty) {
852                         spin_unlock(&dq_list_lock);
853                         empty = get_empty_dquot(sb, qid.type);
854                         if (!empty)
855                                 schedule();     /* Try to wait for a moment... */
856                         goto we_slept;
857                 }
858                 dquot = empty;
859                 empty = NULL;
860                 dquot->dq_id = qid;
861                 /* all dquots go on the inuse_list */
862                 put_inuse(dquot);
863                 /* hash it first so it can be found */
864                 insert_dquot_hash(dquot);
865                 spin_unlock(&dq_list_lock);
866                 dqstats_inc(DQST_LOOKUPS);
867         } else {
868                 if (!atomic_read(&dquot->dq_count))
869                         remove_free_dquot(dquot);
870                 atomic_inc(&dquot->dq_count);
871                 spin_unlock(&dq_list_lock);
872                 dqstats_inc(DQST_CACHE_HITS);
873                 dqstats_inc(DQST_LOOKUPS);
874         }
875         /* Wait for dq_lock - after this we know that either dquot_release() is
876          * already finished or it will be canceled due to dq_count > 1 test */
877         wait_on_dquot(dquot);
878         /* Read the dquot / allocate space in quota file */
879         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) &&
880             sb->dq_op->acquire_dquot(dquot) < 0) {
881                 dqput(dquot);
882                 dquot = NULL;
883                 goto out;
884         }
885 #ifdef CONFIG_QUOTA_DEBUG
886         BUG_ON(!dquot->dq_sb);  /* Has somebody invalidated entry under us? */
887 #endif
888 out:
889         if (empty)
890                 do_destroy_dquot(empty);
891
892         return dquot;
893 }
894 EXPORT_SYMBOL(dqget);
895
896 static inline struct dquot **i_dquot(struct inode *inode)
897 {
898         return inode->i_sb->s_op->get_dquots(inode);
899 }
900
901 static int dqinit_needed(struct inode *inode, int type)
902 {
903         struct dquot * const *dquots;
904         int cnt;
905
906         if (IS_NOQUOTA(inode))
907                 return 0;
908
909         dquots = i_dquot(inode);
910         if (type != -1)
911                 return !dquots[type];
912         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
913                 if (!dquots[cnt])
914                         return 1;
915         return 0;
916 }
917
918 /* This routine is guarded by dqonoff_mutex mutex */
919 static void add_dquot_ref(struct super_block *sb, int type)
920 {
921         struct inode *inode, *old_inode = NULL;
922 #ifdef CONFIG_QUOTA_DEBUG
923         int reserved = 0;
924 #endif
925
926         spin_lock(&inode_sb_list_lock);
927         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
928                 spin_lock(&inode->i_lock);
929                 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
930                     !atomic_read(&inode->i_writecount) ||
931                     !dqinit_needed(inode, type)) {
932                         spin_unlock(&inode->i_lock);
933                         continue;
934                 }
935                 __iget(inode);
936                 spin_unlock(&inode->i_lock);
937                 spin_unlock(&inode_sb_list_lock);
938
939 #ifdef CONFIG_QUOTA_DEBUG
940                 if (unlikely(inode_get_rsv_space(inode) > 0))
941                         reserved = 1;
942 #endif
943                 iput(old_inode);
944                 __dquot_initialize(inode, type);
945
946                 /*
947                  * We hold a reference to 'inode' so it couldn't have been
948                  * removed from s_inodes list while we dropped the
949                  * inode_sb_list_lock We cannot iput the inode now as we can be
950                  * holding the last reference and we cannot iput it under
951                  * inode_sb_list_lock. So we keep the reference and iput it
952                  * later.
953                  */
954                 old_inode = inode;
955                 spin_lock(&inode_sb_list_lock);
956         }
957         spin_unlock(&inode_sb_list_lock);
958         iput(old_inode);
959
960 #ifdef CONFIG_QUOTA_DEBUG
961         if (reserved) {
962                 quota_error(sb, "Writes happened before quota was turned on "
963                         "thus quota information is probably inconsistent. "
964                         "Please run quotacheck(8)");
965         }
966 #endif
967 }
968
969 /*
970  * Remove references to dquots from inode and add dquot to list for freeing
971  * if we have the last reference to dquot
972  */
973 static void remove_inode_dquot_ref(struct inode *inode, int type,
974                                    struct list_head *tofree_head)
975 {
976         struct dquot **dquots = i_dquot(inode);
977         struct dquot *dquot = dquots[type];
978
979         if (!dquot)
980                 return;
981
982         dquots[type] = NULL;
983         if (list_empty(&dquot->dq_free)) {
984                 /*
985                  * The inode still has reference to dquot so it can't be in the
986                  * free list
987                  */
988                 spin_lock(&dq_list_lock);
989                 list_add(&dquot->dq_free, tofree_head);
990                 spin_unlock(&dq_list_lock);
991         } else {
992                 /*
993                  * Dquot is already in a list to put so we won't drop the last
994                  * reference here.
995                  */
996                 dqput(dquot);
997         }
998 }
999
1000 /*
1001  * Free list of dquots
1002  * Dquots are removed from inodes and no new references can be got so we are
1003  * the only ones holding reference
1004  */
1005 static void put_dquot_list(struct list_head *tofree_head)
1006 {
1007         struct list_head *act_head;
1008         struct dquot *dquot;
1009
1010         act_head = tofree_head->next;
1011         while (act_head != tofree_head) {
1012                 dquot = list_entry(act_head, struct dquot, dq_free);
1013                 act_head = act_head->next;
1014                 /* Remove dquot from the list so we won't have problems... */
1015                 list_del_init(&dquot->dq_free);
1016                 dqput(dquot);
1017         }
1018 }
1019
1020 static void remove_dquot_ref(struct super_block *sb, int type,
1021                 struct list_head *tofree_head)
1022 {
1023         struct inode *inode;
1024         int reserved = 0;
1025
1026         spin_lock(&inode_sb_list_lock);
1027         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1028                 /*
1029                  *  We have to scan also I_NEW inodes because they can already
1030                  *  have quota pointer initialized. Luckily, we need to touch
1031                  *  only quota pointers and these have separate locking
1032                  *  (dq_data_lock).
1033                  */
1034                 spin_lock(&dq_data_lock);
1035                 if (!IS_NOQUOTA(inode)) {
1036                         if (unlikely(inode_get_rsv_space(inode) > 0))
1037                                 reserved = 1;
1038                         remove_inode_dquot_ref(inode, type, tofree_head);
1039                 }
1040                 spin_unlock(&dq_data_lock);
1041         }
1042         spin_unlock(&inode_sb_list_lock);
1043 #ifdef CONFIG_QUOTA_DEBUG
1044         if (reserved) {
1045                 printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1046                         " was disabled thus quota information is probably "
1047                         "inconsistent. Please run quotacheck(8).\n", sb->s_id);
1048         }
1049 #endif
1050 }
1051
1052 /* Gather all references from inodes and drop them */
1053 static void drop_dquot_ref(struct super_block *sb, int type)
1054 {
1055         LIST_HEAD(tofree_head);
1056
1057         if (sb->dq_op) {
1058                 remove_dquot_ref(sb, type, &tofree_head);
1059                 synchronize_srcu(&dquot_srcu);
1060                 put_dquot_list(&tofree_head);
1061         }
1062 }
1063
1064 static inline void dquot_incr_inodes(struct dquot *dquot, qsize_t number)
1065 {
1066         dquot->dq_dqb.dqb_curinodes += number;
1067 }
1068
1069 static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
1070 {
1071         dquot->dq_dqb.dqb_curspace += number;
1072 }
1073
1074 static inline void dquot_resv_space(struct dquot *dquot, qsize_t number)
1075 {
1076         dquot->dq_dqb.dqb_rsvspace += number;
1077 }
1078
1079 /*
1080  * Claim reserved quota space
1081  */
1082 static void dquot_claim_reserved_space(struct dquot *dquot, qsize_t number)
1083 {
1084         if (dquot->dq_dqb.dqb_rsvspace < number) {
1085                 WARN_ON_ONCE(1);
1086                 number = dquot->dq_dqb.dqb_rsvspace;
1087         }
1088         dquot->dq_dqb.dqb_curspace += number;
1089         dquot->dq_dqb.dqb_rsvspace -= number;
1090 }
1091
1092 static void dquot_reclaim_reserved_space(struct dquot *dquot, qsize_t number)
1093 {
1094         if (WARN_ON_ONCE(dquot->dq_dqb.dqb_curspace < number))
1095                 number = dquot->dq_dqb.dqb_curspace;
1096         dquot->dq_dqb.dqb_rsvspace += number;
1097         dquot->dq_dqb.dqb_curspace -= number;
1098 }
1099
1100 static inline
1101 void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1102 {
1103         if (dquot->dq_dqb.dqb_rsvspace >= number)
1104                 dquot->dq_dqb.dqb_rsvspace -= number;
1105         else {
1106                 WARN_ON_ONCE(1);
1107                 dquot->dq_dqb.dqb_rsvspace = 0;
1108         }
1109 }
1110
1111 static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1112 {
1113         if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1114             dquot->dq_dqb.dqb_curinodes >= number)
1115                 dquot->dq_dqb.dqb_curinodes -= number;
1116         else
1117                 dquot->dq_dqb.dqb_curinodes = 0;
1118         if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1119                 dquot->dq_dqb.dqb_itime = (time_t) 0;
1120         clear_bit(DQ_INODES_B, &dquot->dq_flags);
1121 }
1122
1123 static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1124 {
1125         if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1126             dquot->dq_dqb.dqb_curspace >= number)
1127                 dquot->dq_dqb.dqb_curspace -= number;
1128         else
1129                 dquot->dq_dqb.dqb_curspace = 0;
1130         if (dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1131                 dquot->dq_dqb.dqb_btime = (time_t) 0;
1132         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1133 }
1134
1135 struct dquot_warn {
1136         struct super_block *w_sb;
1137         struct kqid w_dq_id;
1138         short w_type;
1139 };
1140
1141 static int warning_issued(struct dquot *dquot, const int warntype)
1142 {
1143         int flag = (warntype == QUOTA_NL_BHARDWARN ||
1144                 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1145                 ((warntype == QUOTA_NL_IHARDWARN ||
1146                 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1147
1148         if (!flag)
1149                 return 0;
1150         return test_and_set_bit(flag, &dquot->dq_flags);
1151 }
1152
1153 #ifdef CONFIG_PRINT_QUOTA_WARNING
1154 static int flag_print_warnings = 1;
1155
1156 static int need_print_warning(struct dquot_warn *warn)
1157 {
1158         if (!flag_print_warnings)
1159                 return 0;
1160
1161         switch (warn->w_dq_id.type) {
1162                 case USRQUOTA:
1163                         return uid_eq(current_fsuid(), warn->w_dq_id.uid);
1164                 case GRPQUOTA:
1165                         return in_group_p(warn->w_dq_id.gid);
1166                 case PRJQUOTA:
1167                         return 1;
1168         }
1169         return 0;
1170 }
1171
1172 /* Print warning to user which exceeded quota */
1173 static void print_warning(struct dquot_warn *warn)
1174 {
1175         char *msg = NULL;
1176         struct tty_struct *tty;
1177         int warntype = warn->w_type;
1178
1179         if (warntype == QUOTA_NL_IHARDBELOW ||
1180             warntype == QUOTA_NL_ISOFTBELOW ||
1181             warntype == QUOTA_NL_BHARDBELOW ||
1182             warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(warn))
1183                 return;
1184
1185         tty = get_current_tty();
1186         if (!tty)
1187                 return;
1188         tty_write_message(tty, warn->w_sb->s_id);
1189         if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1190                 tty_write_message(tty, ": warning, ");
1191         else
1192                 tty_write_message(tty, ": write failed, ");
1193         tty_write_message(tty, quotatypes[warn->w_dq_id.type]);
1194         switch (warntype) {
1195                 case QUOTA_NL_IHARDWARN:
1196                         msg = " file limit reached.\r\n";
1197                         break;
1198                 case QUOTA_NL_ISOFTLONGWARN:
1199                         msg = " file quota exceeded too long.\r\n";
1200                         break;
1201                 case QUOTA_NL_ISOFTWARN:
1202                         msg = " file quota exceeded.\r\n";
1203                         break;
1204                 case QUOTA_NL_BHARDWARN:
1205                         msg = " block limit reached.\r\n";
1206                         break;
1207                 case QUOTA_NL_BSOFTLONGWARN:
1208                         msg = " block quota exceeded too long.\r\n";
1209                         break;
1210                 case QUOTA_NL_BSOFTWARN:
1211                         msg = " block quota exceeded.\r\n";
1212                         break;
1213         }
1214         tty_write_message(tty, msg);
1215         tty_kref_put(tty);
1216 }
1217 #endif
1218
1219 static void prepare_warning(struct dquot_warn *warn, struct dquot *dquot,
1220                             int warntype)
1221 {
1222         if (warning_issued(dquot, warntype))
1223                 return;
1224         warn->w_type = warntype;
1225         warn->w_sb = dquot->dq_sb;
1226         warn->w_dq_id = dquot->dq_id;
1227 }
1228
1229 /*
1230  * Write warnings to the console and send warning messages over netlink.
1231  *
1232  * Note that this function can call into tty and networking code.
1233  */
1234 static void flush_warnings(struct dquot_warn *warn)
1235 {
1236         int i;
1237
1238         for (i = 0; i < MAXQUOTAS; i++) {
1239                 if (warn[i].w_type == QUOTA_NL_NOWARN)
1240                         continue;
1241 #ifdef CONFIG_PRINT_QUOTA_WARNING
1242                 print_warning(&warn[i]);
1243 #endif
1244                 quota_send_warning(warn[i].w_dq_id,
1245                                    warn[i].w_sb->s_dev, warn[i].w_type);
1246         }
1247 }
1248
1249 static int ignore_hardlimit(struct dquot *dquot)
1250 {
1251         struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
1252
1253         return capable(CAP_SYS_RESOURCE) &&
1254                (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1255                 !(info->dqi_flags & DQF_ROOT_SQUASH));
1256 }
1257
1258 /* needs dq_data_lock */
1259 static int check_idq(struct dquot *dquot, qsize_t inodes,
1260                      struct dquot_warn *warn)
1261 {
1262         qsize_t newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1263
1264         if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type) ||
1265             test_bit(DQ_FAKE_B, &dquot->dq_flags))
1266                 return 0;
1267
1268         if (dquot->dq_dqb.dqb_ihardlimit &&
1269             newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1270             !ignore_hardlimit(dquot)) {
1271                 prepare_warning(warn, dquot, QUOTA_NL_IHARDWARN);
1272                 return -EDQUOT;
1273         }
1274
1275         if (dquot->dq_dqb.dqb_isoftlimit &&
1276             newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1277             dquot->dq_dqb.dqb_itime &&
1278             get_seconds() >= dquot->dq_dqb.dqb_itime &&
1279             !ignore_hardlimit(dquot)) {
1280                 prepare_warning(warn, dquot, QUOTA_NL_ISOFTLONGWARN);
1281                 return -EDQUOT;
1282         }
1283
1284         if (dquot->dq_dqb.dqb_isoftlimit &&
1285             newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1286             dquot->dq_dqb.dqb_itime == 0) {
1287                 prepare_warning(warn, dquot, QUOTA_NL_ISOFTWARN);
1288                 dquot->dq_dqb.dqb_itime = get_seconds() +
1289                     sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type].dqi_igrace;
1290         }
1291
1292         return 0;
1293 }
1294
1295 /* needs dq_data_lock */
1296 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc,
1297                      struct dquot_warn *warn)
1298 {
1299         qsize_t tspace;
1300         struct super_block *sb = dquot->dq_sb;
1301
1302         if (!sb_has_quota_limits_enabled(sb, dquot->dq_id.type) ||
1303             test_bit(DQ_FAKE_B, &dquot->dq_flags))
1304                 return 0;
1305
1306         tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1307                 + space;
1308
1309         if (dquot->dq_dqb.dqb_bhardlimit &&
1310             tspace > dquot->dq_dqb.dqb_bhardlimit &&
1311             !ignore_hardlimit(dquot)) {
1312                 if (!prealloc)
1313                         prepare_warning(warn, dquot, QUOTA_NL_BHARDWARN);
1314                 return -EDQUOT;
1315         }
1316
1317         if (dquot->dq_dqb.dqb_bsoftlimit &&
1318             tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1319             dquot->dq_dqb.dqb_btime &&
1320             get_seconds() >= dquot->dq_dqb.dqb_btime &&
1321             !ignore_hardlimit(dquot)) {
1322                 if (!prealloc)
1323                         prepare_warning(warn, dquot, QUOTA_NL_BSOFTLONGWARN);
1324                 return -EDQUOT;
1325         }
1326
1327         if (dquot->dq_dqb.dqb_bsoftlimit &&
1328             tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1329             dquot->dq_dqb.dqb_btime == 0) {
1330                 if (!prealloc) {
1331                         prepare_warning(warn, dquot, QUOTA_NL_BSOFTWARN);
1332                         dquot->dq_dqb.dqb_btime = get_seconds() +
1333                             sb_dqopt(sb)->info[dquot->dq_id.type].dqi_bgrace;
1334                 }
1335                 else
1336                         /*
1337                          * We don't allow preallocation to exceed softlimit so exceeding will
1338                          * be always printed
1339                          */
1340                         return -EDQUOT;
1341         }
1342
1343         return 0;
1344 }
1345
1346 static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1347 {
1348         qsize_t newinodes;
1349
1350         if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1351             dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1352             !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type))
1353                 return QUOTA_NL_NOWARN;
1354
1355         newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1356         if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1357                 return QUOTA_NL_ISOFTBELOW;
1358         if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1359             newinodes < dquot->dq_dqb.dqb_ihardlimit)
1360                 return QUOTA_NL_IHARDBELOW;
1361         return QUOTA_NL_NOWARN;
1362 }
1363
1364 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1365 {
1366         if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1367             dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1368                 return QUOTA_NL_NOWARN;
1369
1370         if (dquot->dq_dqb.dqb_curspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1371                 return QUOTA_NL_BSOFTBELOW;
1372         if (dquot->dq_dqb.dqb_curspace >= dquot->dq_dqb.dqb_bhardlimit &&
1373             dquot->dq_dqb.dqb_curspace - space < dquot->dq_dqb.dqb_bhardlimit)
1374                 return QUOTA_NL_BHARDBELOW;
1375         return QUOTA_NL_NOWARN;
1376 }
1377
1378 static int dquot_active(const struct inode *inode)
1379 {
1380         struct super_block *sb = inode->i_sb;
1381
1382         if (IS_NOQUOTA(inode))
1383                 return 0;
1384         return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1385 }
1386
1387 /*
1388  * Initialize quota pointers in inode
1389  *
1390  * It is better to call this function outside of any transaction as it
1391  * might need a lot of space in journal for dquot structure allocation.
1392  */
1393 static void __dquot_initialize(struct inode *inode, int type)
1394 {
1395         int cnt, init_needed = 0;
1396         struct dquot **dquots, *got[MAXQUOTAS];
1397         struct super_block *sb = inode->i_sb;
1398         qsize_t rsv;
1399
1400         if (!dquot_active(inode))
1401                 return;
1402
1403         dquots = i_dquot(inode);
1404
1405         /* First get references to structures we might need. */
1406         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1407                 struct kqid qid;
1408                 kprojid_t projid;
1409                 int rc;
1410
1411                 got[cnt] = NULL;
1412                 if (type != -1 && cnt != type)
1413                         continue;
1414                 /*
1415                  * The i_dquot should have been initialized in most cases,
1416                  * we check it without locking here to avoid unnecessary
1417                  * dqget()/dqput() calls.
1418                  */
1419                 if (dquots[cnt])
1420                         continue;
1421
1422                 if (!sb_has_quota_active(sb, cnt))
1423                         continue;
1424
1425                 init_needed = 1;
1426
1427                 switch (cnt) {
1428                 case USRQUOTA:
1429                         qid = make_kqid_uid(inode->i_uid);
1430                         break;
1431                 case GRPQUOTA:
1432                         qid = make_kqid_gid(inode->i_gid);
1433                         break;
1434                 case PRJQUOTA:
1435                         rc = inode->i_sb->dq_op->get_projid(inode, &projid);
1436                         if (rc)
1437                                 continue;
1438                         qid = make_kqid_projid(projid);
1439                         break;
1440                 }
1441                 got[cnt] = dqget(sb, qid);
1442         }
1443
1444         /* All required i_dquot has been initialized */
1445         if (!init_needed)
1446                 return;
1447
1448         spin_lock(&dq_data_lock);
1449         if (IS_NOQUOTA(inode))
1450                 goto out_err;
1451         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1452                 if (type != -1 && cnt != type)
1453                         continue;
1454                 /* Avoid races with quotaoff() */
1455                 if (!sb_has_quota_active(sb, cnt))
1456                         continue;
1457                 /* We could race with quotaon or dqget() could have failed */
1458                 if (!got[cnt])
1459                         continue;
1460                 if (!dquots[cnt]) {
1461                         dquots[cnt] = got[cnt];
1462                         got[cnt] = NULL;
1463                         /*
1464                          * Make quota reservation system happy if someone
1465                          * did a write before quota was turned on
1466                          */
1467                         rsv = inode_get_rsv_space(inode);
1468                         if (unlikely(rsv))
1469                                 dquot_resv_space(dquots[cnt], rsv);
1470                 }
1471         }
1472 out_err:
1473         spin_unlock(&dq_data_lock);
1474         /* Drop unused references */
1475         dqput_all(got);
1476 }
1477
1478 void dquot_initialize(struct inode *inode)
1479 {
1480         __dquot_initialize(inode, -1);
1481 }
1482 EXPORT_SYMBOL(dquot_initialize);
1483
1484 /*
1485  * Release all quotas referenced by inode.
1486  *
1487  * This function only be called on inode free or converting
1488  * a file to quota file, no other users for the i_dquot in
1489  * both cases, so we needn't call synchronize_srcu() after
1490  * clearing i_dquot.
1491  */
1492 static void __dquot_drop(struct inode *inode)
1493 {
1494         int cnt;
1495         struct dquot **dquots = i_dquot(inode);
1496         struct dquot *put[MAXQUOTAS];
1497
1498         spin_lock(&dq_data_lock);
1499         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1500                 put[cnt] = dquots[cnt];
1501                 dquots[cnt] = NULL;
1502         }
1503         spin_unlock(&dq_data_lock);
1504         dqput_all(put);
1505 }
1506
1507 void dquot_drop(struct inode *inode)
1508 {
1509         struct dquot * const *dquots;
1510         int cnt;
1511
1512         if (IS_NOQUOTA(inode))
1513                 return;
1514
1515         /*
1516          * Test before calling to rule out calls from proc and such
1517          * where we are not allowed to block. Note that this is
1518          * actually reliable test even without the lock - the caller
1519          * must assure that nobody can come after the DQUOT_DROP and
1520          * add quota pointers back anyway.
1521          */
1522         dquots = i_dquot(inode);
1523         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1524                 if (dquots[cnt])
1525                         break;
1526         }
1527
1528         if (cnt < MAXQUOTAS)
1529                 __dquot_drop(inode);
1530 }
1531 EXPORT_SYMBOL(dquot_drop);
1532
1533 /*
1534  * inode_reserved_space is managed internally by quota, and protected by
1535  * i_lock similar to i_blocks+i_bytes.
1536  */
1537 static qsize_t *inode_reserved_space(struct inode * inode)
1538 {
1539         /* Filesystem must explicitly define it's own method in order to use
1540          * quota reservation interface */
1541         BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1542         return inode->i_sb->dq_op->get_reserved_space(inode);
1543 }
1544
1545 void inode_add_rsv_space(struct inode *inode, qsize_t number)
1546 {
1547         spin_lock(&inode->i_lock);
1548         *inode_reserved_space(inode) += number;
1549         spin_unlock(&inode->i_lock);
1550 }
1551 EXPORT_SYMBOL(inode_add_rsv_space);
1552
1553 void inode_claim_rsv_space(struct inode *inode, qsize_t number)
1554 {
1555         spin_lock(&inode->i_lock);
1556         *inode_reserved_space(inode) -= number;
1557         __inode_add_bytes(inode, number);
1558         spin_unlock(&inode->i_lock);
1559 }
1560 EXPORT_SYMBOL(inode_claim_rsv_space);
1561
1562 void inode_reclaim_rsv_space(struct inode *inode, qsize_t number)
1563 {
1564         spin_lock(&inode->i_lock);
1565         *inode_reserved_space(inode) += number;
1566         __inode_sub_bytes(inode, number);
1567         spin_unlock(&inode->i_lock);
1568 }
1569 EXPORT_SYMBOL(inode_reclaim_rsv_space);
1570
1571 void inode_sub_rsv_space(struct inode *inode, qsize_t number)
1572 {
1573         spin_lock(&inode->i_lock);
1574         *inode_reserved_space(inode) -= number;
1575         spin_unlock(&inode->i_lock);
1576 }
1577 EXPORT_SYMBOL(inode_sub_rsv_space);
1578
1579 static qsize_t inode_get_rsv_space(struct inode *inode)
1580 {
1581         qsize_t ret;
1582
1583         if (!inode->i_sb->dq_op->get_reserved_space)
1584                 return 0;
1585         spin_lock(&inode->i_lock);
1586         ret = *inode_reserved_space(inode);
1587         spin_unlock(&inode->i_lock);
1588         return ret;
1589 }
1590
1591 static void inode_incr_space(struct inode *inode, qsize_t number,
1592                                 int reserve)
1593 {
1594         if (reserve)
1595                 inode_add_rsv_space(inode, number);
1596         else
1597                 inode_add_bytes(inode, number);
1598 }
1599
1600 static void inode_decr_space(struct inode *inode, qsize_t number, int reserve)
1601 {
1602         if (reserve)
1603                 inode_sub_rsv_space(inode, number);
1604         else
1605                 inode_sub_bytes(inode, number);
1606 }
1607
1608 /*
1609  * This functions updates i_blocks+i_bytes fields and quota information
1610  * (together with appropriate checks).
1611  *
1612  * NOTE: We absolutely rely on the fact that caller dirties the inode
1613  * (usually helpers in quotaops.h care about this) and holds a handle for
1614  * the current transaction so that dquot write and inode write go into the
1615  * same transaction.
1616  */
1617
1618 /*
1619  * This operation can block, but only after everything is updated
1620  */
1621 int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
1622 {
1623         int cnt, ret = 0, index;
1624         struct dquot_warn warn[MAXQUOTAS];
1625         int reserve = flags & DQUOT_SPACE_RESERVE;
1626         struct dquot **dquots;
1627
1628         if (!dquot_active(inode)) {
1629                 inode_incr_space(inode, number, reserve);
1630                 goto out;
1631         }
1632
1633         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1634                 warn[cnt].w_type = QUOTA_NL_NOWARN;
1635
1636         dquots = i_dquot(inode);
1637         index = srcu_read_lock(&dquot_srcu);
1638         spin_lock(&dq_data_lock);
1639         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1640                 if (!dquots[cnt])
1641                         continue;
1642                 ret = check_bdq(dquots[cnt], number,
1643                                 !(flags & DQUOT_SPACE_WARN), &warn[cnt]);
1644                 if (ret && !(flags & DQUOT_SPACE_NOFAIL)) {
1645                         spin_unlock(&dq_data_lock);
1646                         goto out_flush_warn;
1647                 }
1648         }
1649         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1650                 if (!dquots[cnt])
1651                         continue;
1652                 if (reserve)
1653                         dquot_resv_space(dquots[cnt], number);
1654                 else
1655                         dquot_incr_space(dquots[cnt], number);
1656         }
1657         inode_incr_space(inode, number, reserve);
1658         spin_unlock(&dq_data_lock);
1659
1660         if (reserve)
1661                 goto out_flush_warn;
1662         mark_all_dquot_dirty(dquots);
1663 out_flush_warn:
1664         srcu_read_unlock(&dquot_srcu, index);
1665         flush_warnings(warn);
1666 out:
1667         return ret;
1668 }
1669 EXPORT_SYMBOL(__dquot_alloc_space);
1670
1671 /*
1672  * This operation can block, but only after everything is updated
1673  */
1674 int dquot_alloc_inode(struct inode *inode)
1675 {
1676         int cnt, ret = 0, index;
1677         struct dquot_warn warn[MAXQUOTAS];
1678         struct dquot * const *dquots;
1679
1680         if (!dquot_active(inode))
1681                 return 0;
1682         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1683                 warn[cnt].w_type = QUOTA_NL_NOWARN;
1684
1685         dquots = i_dquot(inode);
1686         index = srcu_read_lock(&dquot_srcu);
1687         spin_lock(&dq_data_lock);
1688         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1689                 if (!dquots[cnt])
1690                         continue;
1691                 ret = check_idq(dquots[cnt], 1, &warn[cnt]);
1692                 if (ret)
1693                         goto warn_put_all;
1694         }
1695
1696         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1697                 if (!dquots[cnt])
1698                         continue;
1699                 dquot_incr_inodes(dquots[cnt], 1);
1700         }
1701
1702 warn_put_all:
1703         spin_unlock(&dq_data_lock);
1704         if (ret == 0)
1705                 mark_all_dquot_dirty(dquots);
1706         srcu_read_unlock(&dquot_srcu, index);
1707         flush_warnings(warn);
1708         return ret;
1709 }
1710 EXPORT_SYMBOL(dquot_alloc_inode);
1711
1712 /*
1713  * Convert in-memory reserved quotas to real consumed quotas
1714  */
1715 int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1716 {
1717         struct dquot **dquots;
1718         int cnt, index;
1719
1720         if (!dquot_active(inode)) {
1721                 inode_claim_rsv_space(inode, number);
1722                 return 0;
1723         }
1724
1725         dquots = i_dquot(inode);
1726         index = srcu_read_lock(&dquot_srcu);
1727         spin_lock(&dq_data_lock);
1728         /* Claim reserved quotas to allocated quotas */
1729         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1730                 if (dquots[cnt])
1731                         dquot_claim_reserved_space(dquots[cnt], number);
1732         }
1733         /* Update inode bytes */
1734         inode_claim_rsv_space(inode, number);
1735         spin_unlock(&dq_data_lock);
1736         mark_all_dquot_dirty(dquots);
1737         srcu_read_unlock(&dquot_srcu, index);
1738         return 0;
1739 }
1740 EXPORT_SYMBOL(dquot_claim_space_nodirty);
1741
1742 /*
1743  * Convert allocated space back to in-memory reserved quotas
1744  */
1745 void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
1746 {
1747         struct dquot **dquots;
1748         int cnt, index;
1749
1750         if (!dquot_active(inode)) {
1751                 inode_reclaim_rsv_space(inode, number);
1752                 return;
1753         }
1754
1755         dquots = i_dquot(inode);
1756         index = srcu_read_lock(&dquot_srcu);
1757         spin_lock(&dq_data_lock);
1758         /* Claim reserved quotas to allocated quotas */
1759         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1760                 if (dquots[cnt])
1761                         dquot_reclaim_reserved_space(dquots[cnt], number);
1762         }
1763         /* Update inode bytes */
1764         inode_reclaim_rsv_space(inode, number);
1765         spin_unlock(&dq_data_lock);
1766         mark_all_dquot_dirty(dquots);
1767         srcu_read_unlock(&dquot_srcu, index);
1768         return;
1769 }
1770 EXPORT_SYMBOL(dquot_reclaim_space_nodirty);
1771
1772 /*
1773  * This operation can block, but only after everything is updated
1774  */
1775 void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
1776 {
1777         unsigned int cnt;
1778         struct dquot_warn warn[MAXQUOTAS];
1779         struct dquot **dquots;
1780         int reserve = flags & DQUOT_SPACE_RESERVE, index;
1781
1782         if (!dquot_active(inode)) {
1783                 inode_decr_space(inode, number, reserve);
1784                 return;
1785         }
1786
1787         dquots = i_dquot(inode);
1788         index = srcu_read_lock(&dquot_srcu);
1789         spin_lock(&dq_data_lock);
1790         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1791                 int wtype;
1792
1793                 warn[cnt].w_type = QUOTA_NL_NOWARN;
1794                 if (!dquots[cnt])
1795                         continue;
1796                 wtype = info_bdq_free(dquots[cnt], number);
1797                 if (wtype != QUOTA_NL_NOWARN)
1798                         prepare_warning(&warn[cnt], dquots[cnt], wtype);
1799                 if (reserve)
1800                         dquot_free_reserved_space(dquots[cnt], number);
1801                 else
1802                         dquot_decr_space(dquots[cnt], number);
1803         }
1804         inode_decr_space(inode, number, reserve);
1805         spin_unlock(&dq_data_lock);
1806
1807         if (reserve)
1808                 goto out_unlock;
1809         mark_all_dquot_dirty(dquots);
1810 out_unlock:
1811         srcu_read_unlock(&dquot_srcu, index);
1812         flush_warnings(warn);
1813 }
1814 EXPORT_SYMBOL(__dquot_free_space);
1815
1816 /*
1817  * This operation can block, but only after everything is updated
1818  */
1819 void dquot_free_inode(struct inode *inode)
1820 {
1821         unsigned int cnt;
1822         struct dquot_warn warn[MAXQUOTAS];
1823         struct dquot * const *dquots;
1824         int index;
1825
1826         if (!dquot_active(inode))
1827                 return;
1828
1829         dquots = i_dquot(inode);
1830         index = srcu_read_lock(&dquot_srcu);
1831         spin_lock(&dq_data_lock);
1832         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1833                 int wtype;
1834
1835                 warn[cnt].w_type = QUOTA_NL_NOWARN;
1836                 if (!dquots[cnt])
1837                         continue;
1838                 wtype = info_idq_free(dquots[cnt], 1);
1839                 if (wtype != QUOTA_NL_NOWARN)
1840                         prepare_warning(&warn[cnt], dquots[cnt], wtype);
1841                 dquot_decr_inodes(dquots[cnt], 1);
1842         }
1843         spin_unlock(&dq_data_lock);
1844         mark_all_dquot_dirty(dquots);
1845         srcu_read_unlock(&dquot_srcu, index);
1846         flush_warnings(warn);
1847 }
1848 EXPORT_SYMBOL(dquot_free_inode);
1849
1850 /*
1851  * Transfer the number of inode and blocks from one diskquota to an other.
1852  * On success, dquot references in transfer_to are consumed and references
1853  * to original dquots that need to be released are placed there. On failure,
1854  * references are kept untouched.
1855  *
1856  * This operation can block, but only after everything is updated
1857  * A transaction must be started when entering this function.
1858  *
1859  * We are holding reference on transfer_from & transfer_to, no need to
1860  * protect them by srcu_read_lock().
1861  */
1862 int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
1863 {
1864         qsize_t space, cur_space;
1865         qsize_t rsv_space = 0;
1866         struct dquot *transfer_from[MAXQUOTAS] = {};
1867         int cnt, ret = 0;
1868         char is_valid[MAXQUOTAS] = {};
1869         struct dquot_warn warn_to[MAXQUOTAS];
1870         struct dquot_warn warn_from_inodes[MAXQUOTAS];
1871         struct dquot_warn warn_from_space[MAXQUOTAS];
1872
1873         if (IS_NOQUOTA(inode))
1874                 return 0;
1875         /* Initialize the arrays */
1876         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1877                 warn_to[cnt].w_type = QUOTA_NL_NOWARN;
1878                 warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
1879                 warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
1880         }
1881
1882         spin_lock(&dq_data_lock);
1883         if (IS_NOQUOTA(inode)) {        /* File without quota accounting? */
1884                 spin_unlock(&dq_data_lock);
1885                 return 0;
1886         }
1887         cur_space = inode_get_bytes(inode);
1888         rsv_space = inode_get_rsv_space(inode);
1889         space = cur_space + rsv_space;
1890         /* Build the transfer_from list and check the limits */
1891         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1892                 /*
1893                  * Skip changes for same uid or gid or for turned off quota-type.
1894                  */
1895                 if (!transfer_to[cnt])
1896                         continue;
1897                 /* Avoid races with quotaoff() */
1898                 if (!sb_has_quota_active(inode->i_sb, cnt))
1899                         continue;
1900                 is_valid[cnt] = 1;
1901                 transfer_from[cnt] = i_dquot(inode)[cnt];
1902                 ret = check_idq(transfer_to[cnt], 1, &warn_to[cnt]);
1903                 if (ret)
1904                         goto over_quota;
1905                 ret = check_bdq(transfer_to[cnt], space, 0, &warn_to[cnt]);
1906                 if (ret)
1907                         goto over_quota;
1908         }
1909
1910         /*
1911          * Finally perform the needed transfer from transfer_from to transfer_to
1912          */
1913         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1914                 if (!is_valid[cnt])
1915                         continue;
1916                 /* Due to IO error we might not have transfer_from[] structure */
1917                 if (transfer_from[cnt]) {
1918                         int wtype;
1919                         wtype = info_idq_free(transfer_from[cnt], 1);
1920                         if (wtype != QUOTA_NL_NOWARN)
1921                                 prepare_warning(&warn_from_inodes[cnt],
1922                                                 transfer_from[cnt], wtype);
1923                         wtype = info_bdq_free(transfer_from[cnt], space);
1924                         if (wtype != QUOTA_NL_NOWARN)
1925                                 prepare_warning(&warn_from_space[cnt],
1926                                                 transfer_from[cnt], wtype);
1927                         dquot_decr_inodes(transfer_from[cnt], 1);
1928                         dquot_decr_space(transfer_from[cnt], cur_space);
1929                         dquot_free_reserved_space(transfer_from[cnt],
1930                                                   rsv_space);
1931                 }
1932
1933                 dquot_incr_inodes(transfer_to[cnt], 1);
1934                 dquot_incr_space(transfer_to[cnt], cur_space);
1935                 dquot_resv_space(transfer_to[cnt], rsv_space);
1936
1937                 i_dquot(inode)[cnt] = transfer_to[cnt];
1938         }
1939         spin_unlock(&dq_data_lock);
1940
1941         mark_all_dquot_dirty(transfer_from);
1942         mark_all_dquot_dirty(transfer_to);
1943         flush_warnings(warn_to);
1944         flush_warnings(warn_from_inodes);
1945         flush_warnings(warn_from_space);
1946         /* Pass back references to put */
1947         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1948                 if (is_valid[cnt])
1949                         transfer_to[cnt] = transfer_from[cnt];
1950         return 0;
1951 over_quota:
1952         spin_unlock(&dq_data_lock);
1953         flush_warnings(warn_to);
1954         return ret;
1955 }
1956 EXPORT_SYMBOL(__dquot_transfer);
1957
1958 /* Wrapper for transferring ownership of an inode for uid/gid only
1959  * Called from FSXXX_setattr()
1960  */
1961 int dquot_transfer(struct inode *inode, struct iattr *iattr)
1962 {
1963         struct dquot *transfer_to[MAXQUOTAS] = {};
1964         struct super_block *sb = inode->i_sb;
1965         int ret;
1966
1967         if (!dquot_active(inode))
1968                 return 0;
1969
1970         if (iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid))
1971                 transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(iattr->ia_uid));
1972         if (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid))
1973                 transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(iattr->ia_gid));
1974
1975         ret = __dquot_transfer(inode, transfer_to);
1976         dqput_all(transfer_to);
1977         return ret;
1978 }
1979 EXPORT_SYMBOL(dquot_transfer);
1980
1981 /*
1982  * Write info of quota file to disk
1983  */
1984 int dquot_commit_info(struct super_block *sb, int type)
1985 {
1986         int ret;
1987         struct quota_info *dqopt = sb_dqopt(sb);
1988
1989         mutex_lock(&dqopt->dqio_mutex);
1990         ret = dqopt->ops[type]->write_file_info(sb, type);
1991         mutex_unlock(&dqopt->dqio_mutex);
1992         return ret;
1993 }
1994 EXPORT_SYMBOL(dquot_commit_info);
1995
1996 /*
1997  * Definitions of diskquota operations.
1998  */
1999 const struct dquot_operations dquot_operations = {
2000         .write_dquot    = dquot_commit,
2001         .acquire_dquot  = dquot_acquire,
2002         .release_dquot  = dquot_release,
2003         .mark_dirty     = dquot_mark_dquot_dirty,
2004         .write_info     = dquot_commit_info,
2005         .alloc_dquot    = dquot_alloc,
2006         .destroy_dquot  = dquot_destroy,
2007 };
2008 EXPORT_SYMBOL(dquot_operations);
2009
2010 /*
2011  * Generic helper for ->open on filesystems supporting disk quotas.
2012  */
2013 int dquot_file_open(struct inode *inode, struct file *file)
2014 {
2015         int error;
2016
2017         error = generic_file_open(inode, file);
2018         if (!error && (file->f_mode & FMODE_WRITE))
2019                 dquot_initialize(inode);
2020         return error;
2021 }
2022 EXPORT_SYMBOL(dquot_file_open);
2023
2024 /*
2025  * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
2026  */
2027 int dquot_disable(struct super_block *sb, int type, unsigned int flags)
2028 {
2029         int cnt, ret = 0;
2030         struct quota_info *dqopt = sb_dqopt(sb);
2031         struct inode *toputinode[MAXQUOTAS];
2032
2033         /* Cannot turn off usage accounting without turning off limits, or
2034          * suspend quotas and simultaneously turn quotas off. */
2035         if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
2036             || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
2037             DQUOT_USAGE_ENABLED)))
2038                 return -EINVAL;
2039
2040         /* We need to serialize quota_off() for device */
2041         mutex_lock(&dqopt->dqonoff_mutex);
2042
2043         /*
2044          * Skip everything if there's nothing to do. We have to do this because
2045          * sometimes we are called when fill_super() failed and calling
2046          * sync_fs() in such cases does no good.
2047          */
2048         if (!sb_any_quota_loaded(sb)) {
2049                 mutex_unlock(&dqopt->dqonoff_mutex);
2050                 return 0;
2051         }
2052         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2053                 toputinode[cnt] = NULL;
2054                 if (type != -1 && cnt != type)
2055                         continue;
2056                 if (!sb_has_quota_loaded(sb, cnt))
2057                         continue;
2058
2059                 if (flags & DQUOT_SUSPENDED) {
2060                         spin_lock(&dq_state_lock);
2061                         dqopt->flags |=
2062                                 dquot_state_flag(DQUOT_SUSPENDED, cnt);
2063                         spin_unlock(&dq_state_lock);
2064                 } else {
2065                         spin_lock(&dq_state_lock);
2066                         dqopt->flags &= ~dquot_state_flag(flags, cnt);
2067                         /* Turning off suspended quotas? */
2068                         if (!sb_has_quota_loaded(sb, cnt) &&
2069                             sb_has_quota_suspended(sb, cnt)) {
2070                                 dqopt->flags &= ~dquot_state_flag(
2071                                                         DQUOT_SUSPENDED, cnt);
2072                                 spin_unlock(&dq_state_lock);
2073                                 iput(dqopt->files[cnt]);
2074                                 dqopt->files[cnt] = NULL;
2075                                 continue;
2076                         }
2077                         spin_unlock(&dq_state_lock);
2078                 }
2079
2080                 /* We still have to keep quota loaded? */
2081                 if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
2082                         continue;
2083
2084                 /* Note: these are blocking operations */
2085                 drop_dquot_ref(sb, cnt);
2086                 invalidate_dquots(sb, cnt);
2087                 /*
2088                  * Now all dquots should be invalidated, all writes done so we
2089                  * should be only users of the info. No locks needed.
2090                  */
2091                 if (info_dirty(&dqopt->info[cnt]))
2092                         sb->dq_op->write_info(sb, cnt);
2093                 if (dqopt->ops[cnt]->free_file_info)
2094                         dqopt->ops[cnt]->free_file_info(sb, cnt);
2095                 put_quota_format(dqopt->info[cnt].dqi_format);
2096
2097                 toputinode[cnt] = dqopt->files[cnt];
2098                 if (!sb_has_quota_loaded(sb, cnt))
2099                         dqopt->files[cnt] = NULL;
2100                 dqopt->info[cnt].dqi_flags = 0;
2101                 dqopt->info[cnt].dqi_igrace = 0;
2102                 dqopt->info[cnt].dqi_bgrace = 0;
2103                 dqopt->ops[cnt] = NULL;
2104         }
2105         mutex_unlock(&dqopt->dqonoff_mutex);
2106
2107         /* Skip syncing and setting flags if quota files are hidden */
2108         if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2109                 goto put_inodes;
2110
2111         /* Sync the superblock so that buffers with quota data are written to
2112          * disk (and so userspace sees correct data afterwards). */
2113         if (sb->s_op->sync_fs)
2114                 sb->s_op->sync_fs(sb, 1);
2115         sync_blockdev(sb->s_bdev);
2116         /* Now the quota files are just ordinary files and we can set the
2117          * inode flags back. Moreover we discard the pagecache so that
2118          * userspace sees the writes we did bypassing the pagecache. We
2119          * must also discard the blockdev buffers so that we see the
2120          * changes done by userspace on the next quotaon() */
2121         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2122                 if (toputinode[cnt]) {
2123                         mutex_lock(&dqopt->dqonoff_mutex);
2124                         /* If quota was reenabled in the meantime, we have
2125                          * nothing to do */
2126                         if (!sb_has_quota_loaded(sb, cnt)) {
2127                                 mutex_lock(&toputinode[cnt]->i_mutex);
2128                                 toputinode[cnt]->i_flags &= ~(S_IMMUTABLE |
2129                                   S_NOATIME | S_NOQUOTA);
2130                                 truncate_inode_pages(&toputinode[cnt]->i_data,
2131                                                      0);
2132                                 mutex_unlock(&toputinode[cnt]->i_mutex);
2133                                 mark_inode_dirty_sync(toputinode[cnt]);
2134                         }
2135                         mutex_unlock(&dqopt->dqonoff_mutex);
2136                 }
2137         if (sb->s_bdev)
2138                 invalidate_bdev(sb->s_bdev);
2139 put_inodes:
2140         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2141                 if (toputinode[cnt]) {
2142                         /* On remount RO, we keep the inode pointer so that we
2143                          * can reenable quota on the subsequent remount RW. We
2144                          * have to check 'flags' variable and not use sb_has_
2145                          * function because another quotaon / quotaoff could
2146                          * change global state before we got here. We refuse
2147                          * to suspend quotas when there is pending delete on
2148                          * the quota file... */
2149                         if (!(flags & DQUOT_SUSPENDED))
2150                                 iput(toputinode[cnt]);
2151                         else if (!toputinode[cnt]->i_nlink)
2152                                 ret = -EBUSY;
2153                 }
2154         return ret;
2155 }
2156 EXPORT_SYMBOL(dquot_disable);
2157
2158 int dquot_quota_off(struct super_block *sb, int type)
2159 {
2160         return dquot_disable(sb, type,
2161                              DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2162 }
2163 EXPORT_SYMBOL(dquot_quota_off);
2164
2165 /*
2166  *      Turn quotas on on a device
2167  */
2168
2169 /*
2170  * Helper function to turn quotas on when we already have the inode of
2171  * quota file and no quota information is loaded.
2172  */
2173 static int vfs_load_quota_inode(struct inode *inode, int type, int format_id,
2174         unsigned int flags)
2175 {
2176         struct quota_format_type *fmt = find_quota_format(format_id);
2177         struct super_block *sb = inode->i_sb;
2178         struct quota_info *dqopt = sb_dqopt(sb);
2179         int error;
2180         int oldflags = -1;
2181
2182         if (!fmt)
2183                 return -ESRCH;
2184         if (!S_ISREG(inode->i_mode)) {
2185                 error = -EACCES;
2186                 goto out_fmt;
2187         }
2188         if (IS_RDONLY(inode)) {
2189                 error = -EROFS;
2190                 goto out_fmt;
2191         }
2192         if (!sb->s_op->quota_write || !sb->s_op->quota_read ||
2193             (type == PRJQUOTA && sb->dq_op->get_projid == NULL)) {
2194                 error = -EINVAL;
2195                 goto out_fmt;
2196         }
2197         /* Usage always has to be set... */
2198         if (!(flags & DQUOT_USAGE_ENABLED)) {
2199                 error = -EINVAL;
2200                 goto out_fmt;
2201         }
2202
2203         if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2204                 /* As we bypass the pagecache we must now flush all the
2205                  * dirty data and invalidate caches so that kernel sees
2206                  * changes from userspace. It is not enough to just flush
2207                  * the quota file since if blocksize < pagesize, invalidation
2208                  * of the cache could fail because of other unrelated dirty
2209                  * data */
2210                 sync_filesystem(sb);
2211                 invalidate_bdev(sb->s_bdev);
2212         }
2213         mutex_lock(&dqopt->dqonoff_mutex);
2214         if (sb_has_quota_loaded(sb, type)) {
2215                 error = -EBUSY;
2216                 goto out_lock;
2217         }
2218
2219         if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2220                 /* We don't want quota and atime on quota files (deadlocks
2221                  * possible) Also nobody should write to the file - we use
2222                  * special IO operations which ignore the immutable bit. */
2223                 mutex_lock(&inode->i_mutex);
2224                 oldflags = inode->i_flags & (S_NOATIME | S_IMMUTABLE |
2225                                              S_NOQUOTA);
2226                 inode->i_flags |= S_NOQUOTA | S_NOATIME | S_IMMUTABLE;
2227                 mutex_unlock(&inode->i_mutex);
2228                 /*
2229                  * When S_NOQUOTA is set, remove dquot references as no more
2230                  * references can be added
2231                  */
2232                 __dquot_drop(inode);
2233         }
2234
2235         error = -EIO;
2236         dqopt->files[type] = igrab(inode);
2237         if (!dqopt->files[type])
2238                 goto out_lock;
2239         error = -EINVAL;
2240         if (!fmt->qf_ops->check_quota_file(sb, type))
2241                 goto out_file_init;
2242
2243         dqopt->ops[type] = fmt->qf_ops;
2244         dqopt->info[type].dqi_format = fmt;
2245         dqopt->info[type].dqi_fmt_id = format_id;
2246         INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2247         mutex_lock(&dqopt->dqio_mutex);
2248         error = dqopt->ops[type]->read_file_info(sb, type);
2249         if (error < 0) {
2250                 mutex_unlock(&dqopt->dqio_mutex);
2251                 goto out_file_init;
2252         }
2253         if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2254                 dqopt->info[type].dqi_flags |= DQF_SYS_FILE;
2255         mutex_unlock(&dqopt->dqio_mutex);
2256         spin_lock(&dq_state_lock);
2257         dqopt->flags |= dquot_state_flag(flags, type);
2258         spin_unlock(&dq_state_lock);
2259
2260         add_dquot_ref(sb, type);
2261         mutex_unlock(&dqopt->dqonoff_mutex);
2262
2263         return 0;
2264
2265 out_file_init:
2266         dqopt->files[type] = NULL;
2267         iput(inode);
2268 out_lock:
2269         if (oldflags != -1) {
2270                 mutex_lock(&inode->i_mutex);
2271                 /* Set the flags back (in the case of accidental quotaon()
2272                  * on a wrong file we don't want to mess up the flags) */
2273                 inode->i_flags &= ~(S_NOATIME | S_NOQUOTA | S_IMMUTABLE);
2274                 inode->i_flags |= oldflags;
2275                 mutex_unlock(&inode->i_mutex);
2276         }
2277         mutex_unlock(&dqopt->dqonoff_mutex);
2278 out_fmt:
2279         put_quota_format(fmt);
2280
2281         return error; 
2282 }
2283
2284 /* Reenable quotas on remount RW */
2285 int dquot_resume(struct super_block *sb, int type)
2286 {
2287         struct quota_info *dqopt = sb_dqopt(sb);
2288         struct inode *inode;
2289         int ret = 0, cnt;
2290         unsigned int flags;
2291
2292         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2293                 if (type != -1 && cnt != type)
2294                         continue;
2295
2296                 mutex_lock(&dqopt->dqonoff_mutex);
2297                 if (!sb_has_quota_suspended(sb, cnt)) {
2298                         mutex_unlock(&dqopt->dqonoff_mutex);
2299                         continue;
2300                 }
2301                 inode = dqopt->files[cnt];
2302                 dqopt->files[cnt] = NULL;
2303                 spin_lock(&dq_state_lock);
2304                 flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2305                                                         DQUOT_LIMITS_ENABLED,
2306                                                         cnt);
2307                 dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2308                 spin_unlock(&dq_state_lock);
2309                 mutex_unlock(&dqopt->dqonoff_mutex);
2310
2311                 flags = dquot_generic_flag(flags, cnt);
2312                 ret = vfs_load_quota_inode(inode, cnt,
2313                                 dqopt->info[cnt].dqi_fmt_id, flags);
2314                 iput(inode);
2315         }
2316
2317         return ret;
2318 }
2319 EXPORT_SYMBOL(dquot_resume);
2320
2321 int dquot_quota_on(struct super_block *sb, int type, int format_id,
2322                    struct path *path)
2323 {
2324         int error = security_quota_on(path->dentry);
2325         if (error)
2326                 return error;
2327         /* Quota file not on the same filesystem? */
2328         if (path->dentry->d_sb != sb)
2329                 error = -EXDEV;
2330         else
2331                 error = vfs_load_quota_inode(d_inode(path->dentry), type,
2332                                              format_id, DQUOT_USAGE_ENABLED |
2333                                              DQUOT_LIMITS_ENABLED);
2334         return error;
2335 }
2336 EXPORT_SYMBOL(dquot_quota_on);
2337
2338 /*
2339  * More powerful function for turning on quotas allowing setting
2340  * of individual quota flags
2341  */
2342 int dquot_enable(struct inode *inode, int type, int format_id,
2343                  unsigned int flags)
2344 {
2345         int ret = 0;
2346         struct super_block *sb = inode->i_sb;
2347         struct quota_info *dqopt = sb_dqopt(sb);
2348
2349         /* Just unsuspend quotas? */
2350         BUG_ON(flags & DQUOT_SUSPENDED);
2351
2352         if (!flags)
2353                 return 0;
2354         /* Just updating flags needed? */
2355         if (sb_has_quota_loaded(sb, type)) {
2356                 mutex_lock(&dqopt->dqonoff_mutex);
2357                 /* Now do a reliable test... */
2358                 if (!sb_has_quota_loaded(sb, type)) {
2359                         mutex_unlock(&dqopt->dqonoff_mutex);
2360                         goto load_quota;
2361                 }
2362                 if (flags & DQUOT_USAGE_ENABLED &&
2363                     sb_has_quota_usage_enabled(sb, type)) {
2364                         ret = -EBUSY;
2365                         goto out_lock;
2366                 }
2367                 if (flags & DQUOT_LIMITS_ENABLED &&
2368                     sb_has_quota_limits_enabled(sb, type)) {
2369                         ret = -EBUSY;
2370                         goto out_lock;
2371                 }
2372                 spin_lock(&dq_state_lock);
2373                 sb_dqopt(sb)->flags |= dquot_state_flag(flags, type);
2374                 spin_unlock(&dq_state_lock);
2375 out_lock:
2376                 mutex_unlock(&dqopt->dqonoff_mutex);
2377                 return ret;
2378         }
2379
2380 load_quota:
2381         return vfs_load_quota_inode(inode, type, format_id, flags);
2382 }
2383 EXPORT_SYMBOL(dquot_enable);
2384
2385 /*
2386  * This function is used when filesystem needs to initialize quotas
2387  * during mount time.
2388  */
2389 int dquot_quota_on_mount(struct super_block *sb, char *qf_name,
2390                 int format_id, int type)
2391 {
2392         struct dentry *dentry;
2393         int error;
2394
2395         mutex_lock(&d_inode(sb->s_root)->i_mutex);
2396         dentry = lookup_one_len(qf_name, sb->s_root, strlen(qf_name));
2397         mutex_unlock(&d_inode(sb->s_root)->i_mutex);
2398         if (IS_ERR(dentry))
2399                 return PTR_ERR(dentry);
2400
2401         if (d_really_is_negative(dentry)) {
2402                 error = -ENOENT;
2403                 goto out;
2404         }
2405
2406         error = security_quota_on(dentry);
2407         if (!error)
2408                 error = vfs_load_quota_inode(d_inode(dentry), type, format_id,
2409                                 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2410
2411 out:
2412         dput(dentry);
2413         return error;
2414 }
2415 EXPORT_SYMBOL(dquot_quota_on_mount);
2416
2417 static int dquot_quota_enable(struct super_block *sb, unsigned int flags)
2418 {
2419         int ret;
2420         int type;
2421         struct quota_info *dqopt = sb_dqopt(sb);
2422
2423         if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2424                 return -ENOSYS;
2425         /* Accounting cannot be turned on while fs is mounted */
2426         flags &= ~(FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT);
2427         if (!flags)
2428                 return -EINVAL;
2429         for (type = 0; type < MAXQUOTAS; type++) {
2430                 if (!(flags & qtype_enforce_flag(type)))
2431                         continue;
2432                 /* Can't enforce without accounting */
2433                 if (!sb_has_quota_usage_enabled(sb, type))
2434                         return -EINVAL;
2435                 ret = dquot_enable(dqopt->files[type], type,
2436                                    dqopt->info[type].dqi_fmt_id,
2437                                    DQUOT_LIMITS_ENABLED);
2438                 if (ret < 0)
2439                         goto out_err;
2440         }
2441         return 0;
2442 out_err:
2443         /* Backout enforcement enablement we already did */
2444         for (type--; type >= 0; type--)  {
2445                 if (flags & qtype_enforce_flag(type))
2446                         dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2447         }
2448         /* Error code translation for better compatibility with XFS */
2449         if (ret == -EBUSY)
2450                 ret = -EEXIST;
2451         return ret;
2452 }
2453
2454 static int dquot_quota_disable(struct super_block *sb, unsigned int flags)
2455 {
2456         int ret;
2457         int type;
2458         struct quota_info *dqopt = sb_dqopt(sb);
2459
2460         if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2461                 return -ENOSYS;
2462         /*
2463          * We don't support turning off accounting via quotactl. In principle
2464          * quota infrastructure can do this but filesystems don't expect
2465          * userspace to be able to do it.
2466          */
2467         if (flags &
2468                   (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT))
2469                 return -EOPNOTSUPP;
2470
2471         /* Filter out limits not enabled */
2472         for (type = 0; type < MAXQUOTAS; type++)
2473                 if (!sb_has_quota_limits_enabled(sb, type))
2474                         flags &= ~qtype_enforce_flag(type);
2475         /* Nothing left? */
2476         if (!flags)
2477                 return -EEXIST;
2478         for (type = 0; type < MAXQUOTAS; type++) {
2479                 if (flags & qtype_enforce_flag(type)) {
2480                         ret = dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2481                         if (ret < 0)
2482                                 goto out_err;
2483                 }
2484         }
2485         return 0;
2486 out_err:
2487         /* Backout enforcement disabling we already did */
2488         for (type--; type >= 0; type--)  {
2489                 if (flags & qtype_enforce_flag(type))
2490                         dquot_enable(dqopt->files[type], type,
2491                                      dqopt->info[type].dqi_fmt_id,
2492                                      DQUOT_LIMITS_ENABLED);
2493         }
2494         return ret;
2495 }
2496
2497 /* Generic routine for getting common part of quota structure */
2498 static void do_get_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2499 {
2500         struct mem_dqblk *dm = &dquot->dq_dqb;
2501
2502         memset(di, 0, sizeof(*di));
2503         spin_lock(&dq_data_lock);
2504         di->d_spc_hardlimit = dm->dqb_bhardlimit;
2505         di->d_spc_softlimit = dm->dqb_bsoftlimit;
2506         di->d_ino_hardlimit = dm->dqb_ihardlimit;
2507         di->d_ino_softlimit = dm->dqb_isoftlimit;
2508         di->d_space = dm->dqb_curspace + dm->dqb_rsvspace;
2509         di->d_ino_count = dm->dqb_curinodes;
2510         di->d_spc_timer = dm->dqb_btime;
2511         di->d_ino_timer = dm->dqb_itime;
2512         spin_unlock(&dq_data_lock);
2513 }
2514
2515 int dquot_get_dqblk(struct super_block *sb, struct kqid qid,
2516                     struct qc_dqblk *di)
2517 {
2518         struct dquot *dquot;
2519
2520         dquot = dqget(sb, qid);
2521         if (!dquot)
2522                 return -ESRCH;
2523         do_get_dqblk(dquot, di);
2524         dqput(dquot);
2525
2526         return 0;
2527 }
2528 EXPORT_SYMBOL(dquot_get_dqblk);
2529
2530 #define VFS_QC_MASK \
2531         (QC_SPACE | QC_SPC_SOFT | QC_SPC_HARD | \
2532          QC_INO_COUNT | QC_INO_SOFT | QC_INO_HARD | \
2533          QC_SPC_TIMER | QC_INO_TIMER)
2534
2535 /* Generic routine for setting common part of quota structure */
2536 static int do_set_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2537 {
2538         struct mem_dqblk *dm = &dquot->dq_dqb;
2539         int check_blim = 0, check_ilim = 0;
2540         struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
2541
2542         if (di->d_fieldmask & ~VFS_QC_MASK)
2543                 return -EINVAL;
2544
2545         if (((di->d_fieldmask & QC_SPC_SOFT) &&
2546              di->d_spc_softlimit > dqi->dqi_max_spc_limit) ||
2547             ((di->d_fieldmask & QC_SPC_HARD) &&
2548              di->d_spc_hardlimit > dqi->dqi_max_spc_limit) ||
2549             ((di->d_fieldmask & QC_INO_SOFT) &&
2550              (di->d_ino_softlimit > dqi->dqi_max_ino_limit)) ||
2551             ((di->d_fieldmask & QC_INO_HARD) &&
2552              (di->d_ino_hardlimit > dqi->dqi_max_ino_limit)))
2553                 return -ERANGE;
2554
2555         spin_lock(&dq_data_lock);
2556         if (di->d_fieldmask & QC_SPACE) {
2557                 dm->dqb_curspace = di->d_space - dm->dqb_rsvspace;
2558                 check_blim = 1;
2559                 set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2560         }
2561
2562         if (di->d_fieldmask & QC_SPC_SOFT)
2563                 dm->dqb_bsoftlimit = di->d_spc_softlimit;
2564         if (di->d_fieldmask & QC_SPC_HARD)
2565                 dm->dqb_bhardlimit = di->d_spc_hardlimit;
2566         if (di->d_fieldmask & (QC_SPC_SOFT | QC_SPC_HARD)) {
2567                 check_blim = 1;
2568                 set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2569         }
2570
2571         if (di->d_fieldmask & QC_INO_COUNT) {
2572                 dm->dqb_curinodes = di->d_ino_count;
2573                 check_ilim = 1;
2574                 set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2575         }
2576
2577         if (di->d_fieldmask & QC_INO_SOFT)
2578                 dm->dqb_isoftlimit = di->d_ino_softlimit;
2579         if (di->d_fieldmask & QC_INO_HARD)
2580                 dm->dqb_ihardlimit = di->d_ino_hardlimit;
2581         if (di->d_fieldmask & (QC_INO_SOFT | QC_INO_HARD)) {
2582                 check_ilim = 1;
2583                 set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2584         }
2585
2586         if (di->d_fieldmask & QC_SPC_TIMER) {
2587                 dm->dqb_btime = di->d_spc_timer;
2588                 check_blim = 1;
2589                 set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2590         }
2591
2592         if (di->d_fieldmask & QC_INO_TIMER) {
2593                 dm->dqb_itime = di->d_ino_timer;
2594                 check_ilim = 1;
2595                 set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2596         }
2597
2598         if (check_blim) {
2599                 if (!dm->dqb_bsoftlimit ||
2600                     dm->dqb_curspace < dm->dqb_bsoftlimit) {
2601                         dm->dqb_btime = 0;
2602                         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2603                 } else if (!(di->d_fieldmask & QC_SPC_TIMER))
2604                         /* Set grace only if user hasn't provided his own... */
2605                         dm->dqb_btime = get_seconds() + dqi->dqi_bgrace;
2606         }
2607         if (check_ilim) {
2608                 if (!dm->dqb_isoftlimit ||
2609                     dm->dqb_curinodes < dm->dqb_isoftlimit) {
2610                         dm->dqb_itime = 0;
2611                         clear_bit(DQ_INODES_B, &dquot->dq_flags);
2612                 } else if (!(di->d_fieldmask & QC_INO_TIMER))
2613                         /* Set grace only if user hasn't provided his own... */
2614                         dm->dqb_itime = get_seconds() + dqi->dqi_igrace;
2615         }
2616         if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2617             dm->dqb_isoftlimit)
2618                 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2619         else
2620                 set_bit(DQ_FAKE_B, &dquot->dq_flags);
2621         spin_unlock(&dq_data_lock);
2622         mark_dquot_dirty(dquot);
2623
2624         return 0;
2625 }
2626
2627 int dquot_set_dqblk(struct super_block *sb, struct kqid qid,
2628                   struct qc_dqblk *di)
2629 {
2630         struct dquot *dquot;
2631         int rc;
2632
2633         dquot = dqget(sb, qid);
2634         if (!dquot) {
2635                 rc = -ESRCH;
2636                 goto out;
2637         }
2638         rc = do_set_dqblk(dquot, di);
2639         dqput(dquot);
2640 out:
2641         return rc;
2642 }
2643 EXPORT_SYMBOL(dquot_set_dqblk);
2644
2645 /* Generic routine for getting common part of quota file information */
2646 int dquot_get_state(struct super_block *sb, struct qc_state *state)
2647 {
2648         struct mem_dqinfo *mi;
2649         struct qc_type_state *tstate;
2650         struct quota_info *dqopt = sb_dqopt(sb);
2651         int type;
2652   
2653         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
2654         memset(state, 0, sizeof(*state));
2655         for (type = 0; type < MAXQUOTAS; type++) {
2656                 if (!sb_has_quota_active(sb, type))
2657                         continue;
2658                 tstate = state->s_state + type;
2659                 mi = sb_dqopt(sb)->info + type;
2660                 tstate->flags = QCI_ACCT_ENABLED;
2661                 spin_lock(&dq_data_lock);
2662                 if (mi->dqi_flags & DQF_SYS_FILE)
2663                         tstate->flags |= QCI_SYSFILE;
2664                 if (mi->dqi_flags & DQF_ROOT_SQUASH)
2665                         tstate->flags |= QCI_ROOT_SQUASH;
2666                 if (sb_has_quota_limits_enabled(sb, type))
2667                         tstate->flags |= QCI_LIMITS_ENFORCED;
2668                 tstate->spc_timelimit = mi->dqi_bgrace;
2669                 tstate->ino_timelimit = mi->dqi_igrace;
2670                 tstate->ino = dqopt->files[type]->i_ino;
2671                 tstate->blocks = dqopt->files[type]->i_blocks;
2672                 tstate->nextents = 1;   /* We don't know... */
2673                 spin_unlock(&dq_data_lock);
2674         }
2675         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2676         return 0;
2677 }
2678 EXPORT_SYMBOL(dquot_get_state);
2679
2680 /* Generic routine for setting common part of quota file information */
2681 int dquot_set_dqinfo(struct super_block *sb, int type, struct qc_info *ii)
2682 {
2683         struct mem_dqinfo *mi;
2684         int err = 0;
2685
2686         if ((ii->i_fieldmask & QC_WARNS_MASK) ||
2687             (ii->i_fieldmask & QC_RT_SPC_TIMER))
2688                 return -EINVAL;
2689         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
2690         if (!sb_has_quota_active(sb, type)) {
2691                 err = -ESRCH;
2692                 goto out;
2693         }
2694         mi = sb_dqopt(sb)->info + type;
2695         if (ii->i_fieldmask & QC_FLAGS) {
2696                 if ((ii->i_flags & QCI_ROOT_SQUASH &&
2697                      mi->dqi_format->qf_fmt_id != QFMT_VFS_OLD)) {
2698                         err = -EINVAL;
2699                         goto out;
2700                 }
2701         }
2702         spin_lock(&dq_data_lock);
2703         if (ii->i_fieldmask & QC_SPC_TIMER)
2704                 mi->dqi_bgrace = ii->i_spc_timelimit;
2705         if (ii->i_fieldmask & QC_INO_TIMER)
2706                 mi->dqi_igrace = ii->i_ino_timelimit;
2707         if (ii->i_fieldmask & QC_FLAGS) {
2708                 if (ii->i_flags & QCI_ROOT_SQUASH)
2709                         mi->dqi_flags |= DQF_ROOT_SQUASH;
2710                 else
2711                         mi->dqi_flags &= ~DQF_ROOT_SQUASH;
2712         }
2713         spin_unlock(&dq_data_lock);
2714         mark_info_dirty(sb, type);
2715         /* Force write to disk */
2716         sb->dq_op->write_info(sb, type);
2717 out:
2718         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2719         return err;
2720 }
2721 EXPORT_SYMBOL(dquot_set_dqinfo);
2722
2723 const struct quotactl_ops dquot_quotactl_ops = {
2724         .quota_on       = dquot_quota_on,
2725         .quota_off      = dquot_quota_off,
2726         .quota_sync     = dquot_quota_sync,
2727         .get_state      = dquot_get_state,
2728         .set_info       = dquot_set_dqinfo,
2729         .get_dqblk      = dquot_get_dqblk,
2730         .set_dqblk      = dquot_set_dqblk
2731 };
2732 EXPORT_SYMBOL(dquot_quotactl_ops);
2733
2734 const struct quotactl_ops dquot_quotactl_sysfile_ops = {
2735         .quota_enable   = dquot_quota_enable,
2736         .quota_disable  = dquot_quota_disable,
2737         .quota_sync     = dquot_quota_sync,
2738         .get_state      = dquot_get_state,
2739         .set_info       = dquot_set_dqinfo,
2740         .get_dqblk      = dquot_get_dqblk,
2741         .set_dqblk      = dquot_set_dqblk
2742 };
2743 EXPORT_SYMBOL(dquot_quotactl_sysfile_ops);
2744
2745 static int do_proc_dqstats(struct ctl_table *table, int write,
2746                      void __user *buffer, size_t *lenp, loff_t *ppos)
2747 {
2748         unsigned int type = (int *)table->data - dqstats.stat;
2749
2750         /* Update global table */
2751         dqstats.stat[type] =
2752                         percpu_counter_sum_positive(&dqstats.counter[type]);
2753         return proc_dointvec(table, write, buffer, lenp, ppos);
2754 }
2755
2756 static struct ctl_table fs_dqstats_table[] = {
2757         {
2758                 .procname       = "lookups",
2759                 .data           = &dqstats.stat[DQST_LOOKUPS],
2760                 .maxlen         = sizeof(int),
2761                 .mode           = 0444,
2762                 .proc_handler   = do_proc_dqstats,
2763         },
2764         {
2765                 .procname       = "drops",
2766                 .data           = &dqstats.stat[DQST_DROPS],
2767                 .maxlen         = sizeof(int),
2768                 .mode           = 0444,
2769                 .proc_handler   = do_proc_dqstats,
2770         },
2771         {
2772                 .procname       = "reads",
2773                 .data           = &dqstats.stat[DQST_READS],
2774                 .maxlen         = sizeof(int),
2775                 .mode           = 0444,
2776                 .proc_handler   = do_proc_dqstats,
2777         },
2778         {
2779                 .procname       = "writes",
2780                 .data           = &dqstats.stat[DQST_WRITES],
2781                 .maxlen         = sizeof(int),
2782                 .mode           = 0444,
2783                 .proc_handler   = do_proc_dqstats,
2784         },
2785         {
2786                 .procname       = "cache_hits",
2787                 .data           = &dqstats.stat[DQST_CACHE_HITS],
2788                 .maxlen         = sizeof(int),
2789                 .mode           = 0444,
2790                 .proc_handler   = do_proc_dqstats,
2791         },
2792         {
2793                 .procname       = "allocated_dquots",
2794                 .data           = &dqstats.stat[DQST_ALLOC_DQUOTS],
2795                 .maxlen         = sizeof(int),
2796                 .mode           = 0444,
2797                 .proc_handler   = do_proc_dqstats,
2798         },
2799         {
2800                 .procname       = "free_dquots",
2801                 .data           = &dqstats.stat[DQST_FREE_DQUOTS],
2802                 .maxlen         = sizeof(int),
2803                 .mode           = 0444,
2804                 .proc_handler   = do_proc_dqstats,
2805         },
2806         {
2807                 .procname       = "syncs",
2808                 .data           = &dqstats.stat[DQST_SYNCS],
2809                 .maxlen         = sizeof(int),
2810                 .mode           = 0444,
2811                 .proc_handler   = do_proc_dqstats,
2812         },
2813 #ifdef CONFIG_PRINT_QUOTA_WARNING
2814         {
2815                 .procname       = "warnings",
2816                 .data           = &flag_print_warnings,
2817                 .maxlen         = sizeof(int),
2818                 .mode           = 0644,
2819                 .proc_handler   = proc_dointvec,
2820         },
2821 #endif
2822         { },
2823 };
2824
2825 static struct ctl_table fs_table[] = {
2826         {
2827                 .procname       = "quota",
2828                 .mode           = 0555,
2829                 .child          = fs_dqstats_table,
2830         },
2831         { },
2832 };
2833
2834 static struct ctl_table sys_table[] = {
2835         {
2836                 .procname       = "fs",
2837                 .mode           = 0555,
2838                 .child          = fs_table,
2839         },
2840         { },
2841 };
2842
2843 static int __init dquot_init(void)
2844 {
2845         int i, ret;
2846         unsigned long nr_hash, order;
2847
2848         printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2849
2850         register_sysctl_table(sys_table);
2851
2852         dquot_cachep = kmem_cache_create("dquot",
2853                         sizeof(struct dquot), sizeof(unsigned long) * 4,
2854                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2855                                 SLAB_MEM_SPREAD|SLAB_PANIC),
2856                         NULL);
2857
2858         order = 0;
2859         dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
2860         if (!dquot_hash)
2861                 panic("Cannot create dquot hash table");
2862
2863         for (i = 0; i < _DQST_DQSTAT_LAST; i++) {
2864                 ret = percpu_counter_init(&dqstats.counter[i], 0, GFP_KERNEL);
2865                 if (ret)
2866                         panic("Cannot create dquot stat counters");
2867         }
2868
2869         /* Find power-of-two hlist_heads which can fit into allocation */
2870         nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2871         dq_hash_bits = 0;
2872         do {
2873                 dq_hash_bits++;
2874         } while (nr_hash >> dq_hash_bits);
2875         dq_hash_bits--;
2876
2877         nr_hash = 1UL << dq_hash_bits;
2878         dq_hash_mask = nr_hash - 1;
2879         for (i = 0; i < nr_hash; i++)
2880                 INIT_HLIST_HEAD(dquot_hash + i);
2881
2882         pr_info("VFS: Dquot-cache hash table entries: %ld (order %ld,"
2883                 " %ld bytes)\n", nr_hash, order, (PAGE_SIZE << order));
2884
2885         register_shrinker(&dqcache_shrinker);
2886
2887         return 0;
2888 }
2889 module_init(dquot_init);