Merge tag 'binfmt-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb...
[sfrench/cifs-2.6.git] / 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 int __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         /* Make sure flags update is visible after dquot has been filled */
415         smp_mb__before_atomic();
416         set_bit(DQ_READ_B, &dquot->dq_flags);
417         /* Instantiate dquot if needed */
418         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
419                 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
420                 /* Write the info if needed */
421                 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
422                         ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
423                                         dquot->dq_sb, dquot->dq_id.type);
424                 }
425                 if (ret < 0)
426                         goto out_iolock;
427                 if (ret2 < 0) {
428                         ret = ret2;
429                         goto out_iolock;
430                 }
431         }
432         /*
433          * Make sure flags update is visible after on-disk struct has been
434          * allocated. Paired with smp_rmb() in dqget().
435          */
436         smp_mb__before_atomic();
437         set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
438 out_iolock:
439         mutex_unlock(&dqopt->dqio_mutex);
440         mutex_unlock(&dquot->dq_lock);
441         return ret;
442 }
443 EXPORT_SYMBOL(dquot_acquire);
444
445 /*
446  *      Write dquot to disk
447  */
448 int dquot_commit(struct dquot *dquot)
449 {
450         int ret = 0;
451         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
452
453         mutex_lock(&dqopt->dqio_mutex);
454         spin_lock(&dq_list_lock);
455         if (!clear_dquot_dirty(dquot)) {
456                 spin_unlock(&dq_list_lock);
457                 goto out_sem;
458         }
459         spin_unlock(&dq_list_lock);
460         /* Inactive dquot can be only if there was error during read/init
461          * => we have better not writing it */
462         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
463                 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
464         else
465                 ret = -EIO;
466 out_sem:
467         mutex_unlock(&dqopt->dqio_mutex);
468         return ret;
469 }
470 EXPORT_SYMBOL(dquot_commit);
471
472 /*
473  *      Release dquot
474  */
475 int dquot_release(struct dquot *dquot)
476 {
477         int ret = 0, ret2 = 0;
478         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
479
480         mutex_lock(&dquot->dq_lock);
481         /* Check whether we are not racing with some other dqget() */
482         if (atomic_read(&dquot->dq_count) > 1)
483                 goto out_dqlock;
484         mutex_lock(&dqopt->dqio_mutex);
485         if (dqopt->ops[dquot->dq_id.type]->release_dqblk) {
486                 ret = dqopt->ops[dquot->dq_id.type]->release_dqblk(dquot);
487                 /* Write the info */
488                 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
489                         ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
490                                                 dquot->dq_sb, dquot->dq_id.type);
491                 }
492                 if (ret >= 0)
493                         ret = ret2;
494         }
495         clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
496         mutex_unlock(&dqopt->dqio_mutex);
497 out_dqlock:
498         mutex_unlock(&dquot->dq_lock);
499         return ret;
500 }
501 EXPORT_SYMBOL(dquot_release);
502
503 void dquot_destroy(struct dquot *dquot)
504 {
505         kmem_cache_free(dquot_cachep, dquot);
506 }
507 EXPORT_SYMBOL(dquot_destroy);
508
509 static inline void do_destroy_dquot(struct dquot *dquot)
510 {
511         dquot->dq_sb->dq_op->destroy_dquot(dquot);
512 }
513
514 /* Invalidate all dquots on the list. Note that this function is called after
515  * quota is disabled and pointers from inodes removed so there cannot be new
516  * quota users. There can still be some users of quotas due to inodes being
517  * just deleted or pruned by prune_icache() (those are not attached to any
518  * list) or parallel quotactl call. We have to wait for such users.
519  */
520 static void invalidate_dquots(struct super_block *sb, int type)
521 {
522         struct dquot *dquot, *tmp;
523
524 restart:
525         spin_lock(&dq_list_lock);
526         list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
527                 if (dquot->dq_sb != sb)
528                         continue;
529                 if (dquot->dq_id.type != type)
530                         continue;
531                 /* Wait for dquot users */
532                 if (atomic_read(&dquot->dq_count)) {
533                         DEFINE_WAIT(wait);
534
535                         dqgrab(dquot);
536                         prepare_to_wait(&dquot->dq_wait_unused, &wait,
537                                         TASK_UNINTERRUPTIBLE);
538                         spin_unlock(&dq_list_lock);
539                         /* Once dqput() wakes us up, we know it's time to free
540                          * the dquot.
541                          * IMPORTANT: we rely on the fact that there is always
542                          * at most one process waiting for dquot to free.
543                          * Otherwise dq_count would be > 1 and we would never
544                          * wake up.
545                          */
546                         if (atomic_read(&dquot->dq_count) > 1)
547                                 schedule();
548                         finish_wait(&dquot->dq_wait_unused, &wait);
549                         dqput(dquot);
550                         /* At this moment dquot() need not exist (it could be
551                          * reclaimed by prune_dqcache(). Hence we must
552                          * restart. */
553                         goto restart;
554                 }
555                 /*
556                  * Quota now has no users and it has been written on last
557                  * dqput()
558                  */
559                 remove_dquot_hash(dquot);
560                 remove_free_dquot(dquot);
561                 remove_inuse(dquot);
562                 do_destroy_dquot(dquot);
563         }
564         spin_unlock(&dq_list_lock);
565 }
566
567 /* Call callback for every active dquot on given filesystem */
568 int dquot_scan_active(struct super_block *sb,
569                       int (*fn)(struct dquot *dquot, unsigned long priv),
570                       unsigned long priv)
571 {
572         struct dquot *dquot, *old_dquot = NULL;
573         int ret = 0;
574
575         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
576         spin_lock(&dq_list_lock);
577         list_for_each_entry(dquot, &inuse_list, dq_inuse) {
578                 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
579                         continue;
580                 if (dquot->dq_sb != sb)
581                         continue;
582                 /* Now we have active dquot so we can just increase use count */
583                 atomic_inc(&dquot->dq_count);
584                 spin_unlock(&dq_list_lock);
585                 dqstats_inc(DQST_LOOKUPS);
586                 dqput(old_dquot);
587                 old_dquot = dquot;
588                 /*
589                  * ->release_dquot() can be racing with us. Our reference
590                  * protects us from new calls to it so just wait for any
591                  * outstanding call and recheck the DQ_ACTIVE_B after that.
592                  */
593                 wait_on_dquot(dquot);
594                 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
595                         ret = fn(dquot, priv);
596                         if (ret < 0)
597                                 goto out;
598                 }
599                 spin_lock(&dq_list_lock);
600                 /* We are safe to continue now because our dquot could not
601                  * be moved out of the inuse list while we hold the reference */
602         }
603         spin_unlock(&dq_list_lock);
604 out:
605         dqput(old_dquot);
606         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
607         return ret;
608 }
609 EXPORT_SYMBOL(dquot_scan_active);
610
611 /* Write all dquot structures to quota files */
612 int dquot_writeback_dquots(struct super_block *sb, int type)
613 {
614         struct list_head *dirty;
615         struct dquot *dquot;
616         struct quota_info *dqopt = sb_dqopt(sb);
617         int cnt;
618         int err, ret = 0;
619
620         mutex_lock(&dqopt->dqonoff_mutex);
621         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
622                 if (type != -1 && cnt != type)
623                         continue;
624                 if (!sb_has_quota_active(sb, cnt))
625                         continue;
626                 spin_lock(&dq_list_lock);
627                 dirty = &dqopt->info[cnt].dqi_dirty_list;
628                 while (!list_empty(dirty)) {
629                         dquot = list_first_entry(dirty, struct dquot,
630                                                  dq_dirty);
631                         /* Dirty and inactive can be only bad dquot... */
632                         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
633                                 clear_dquot_dirty(dquot);
634                                 continue;
635                         }
636                         /* Now we have active dquot from which someone is
637                          * holding reference so we can safely just increase
638                          * use count */
639                         dqgrab(dquot);
640                         spin_unlock(&dq_list_lock);
641                         dqstats_inc(DQST_LOOKUPS);
642                         err = sb->dq_op->write_dquot(dquot);
643                         if (!ret && err)
644                                 ret = err;
645                         dqput(dquot);
646                         spin_lock(&dq_list_lock);
647                 }
648                 spin_unlock(&dq_list_lock);
649         }
650
651         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
652                 if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
653                     && info_dirty(&dqopt->info[cnt]))
654                         sb->dq_op->write_info(sb, cnt);
655         dqstats_inc(DQST_SYNCS);
656         mutex_unlock(&dqopt->dqonoff_mutex);
657
658         return ret;
659 }
660 EXPORT_SYMBOL(dquot_writeback_dquots);
661
662 /* Write all dquot structures to disk and make them visible from userspace */
663 int dquot_quota_sync(struct super_block *sb, int type)
664 {
665         struct quota_info *dqopt = sb_dqopt(sb);
666         int cnt;
667         int ret;
668
669         ret = dquot_writeback_dquots(sb, type);
670         if (ret)
671                 return ret;
672         if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
673                 return 0;
674
675         /* This is not very clever (and fast) but currently I don't know about
676          * any other simple way of getting quota data to disk and we must get
677          * them there for userspace to be visible... */
678         if (sb->s_op->sync_fs)
679                 sb->s_op->sync_fs(sb, 1);
680         sync_blockdev(sb->s_bdev);
681
682         /*
683          * Now when everything is written we can discard the pagecache so
684          * that userspace sees the changes.
685          */
686         mutex_lock(&dqopt->dqonoff_mutex);
687         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
688                 if (type != -1 && cnt != type)
689                         continue;
690                 if (!sb_has_quota_active(sb, cnt))
691                         continue;
692                 inode_lock(dqopt->files[cnt]);
693                 truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
694                 inode_unlock(dqopt->files[cnt]);
695         }
696         mutex_unlock(&dqopt->dqonoff_mutex);
697
698         return 0;
699 }
700 EXPORT_SYMBOL(dquot_quota_sync);
701
702 static unsigned long
703 dqcache_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
704 {
705         struct list_head *head;
706         struct dquot *dquot;
707         unsigned long freed = 0;
708
709         spin_lock(&dq_list_lock);
710         head = free_dquots.prev;
711         while (head != &free_dquots && sc->nr_to_scan) {
712                 dquot = list_entry(head, struct dquot, dq_free);
713                 remove_dquot_hash(dquot);
714                 remove_free_dquot(dquot);
715                 remove_inuse(dquot);
716                 do_destroy_dquot(dquot);
717                 sc->nr_to_scan--;
718                 freed++;
719                 head = free_dquots.prev;
720         }
721         spin_unlock(&dq_list_lock);
722         return freed;
723 }
724
725 static unsigned long
726 dqcache_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
727 {
728         return vfs_pressure_ratio(
729         percpu_counter_read_positive(&dqstats.counter[DQST_FREE_DQUOTS]));
730 }
731
732 static struct shrinker dqcache_shrinker = {
733         .count_objects = dqcache_shrink_count,
734         .scan_objects = dqcache_shrink_scan,
735         .seeks = DEFAULT_SEEKS,
736 };
737
738 /*
739  * Put reference to dquot
740  */
741 void dqput(struct dquot *dquot)
742 {
743         int ret;
744
745         if (!dquot)
746                 return;
747 #ifdef CONFIG_QUOTA_DEBUG
748         if (!atomic_read(&dquot->dq_count)) {
749                 quota_error(dquot->dq_sb, "trying to free free dquot of %s %d",
750                             quotatypes[dquot->dq_id.type],
751                             from_kqid(&init_user_ns, dquot->dq_id));
752                 BUG();
753         }
754 #endif
755         dqstats_inc(DQST_DROPS);
756 we_slept:
757         spin_lock(&dq_list_lock);
758         if (atomic_read(&dquot->dq_count) > 1) {
759                 /* We have more than one user... nothing to do */
760                 atomic_dec(&dquot->dq_count);
761                 /* Releasing dquot during quotaoff phase? */
762                 if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_id.type) &&
763                     atomic_read(&dquot->dq_count) == 1)
764                         wake_up(&dquot->dq_wait_unused);
765                 spin_unlock(&dq_list_lock);
766                 return;
767         }
768         /* Need to release dquot? */
769         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && dquot_dirty(dquot)) {
770                 spin_unlock(&dq_list_lock);
771                 /* Commit dquot before releasing */
772                 ret = dquot->dq_sb->dq_op->write_dquot(dquot);
773                 if (ret < 0) {
774                         quota_error(dquot->dq_sb, "Can't write quota structure"
775                                     " (error %d). Quota may get out of sync!",
776                                     ret);
777                         /*
778                          * We clear dirty bit anyway, so that we avoid
779                          * infinite loop here
780                          */
781                         spin_lock(&dq_list_lock);
782                         clear_dquot_dirty(dquot);
783                         spin_unlock(&dq_list_lock);
784                 }
785                 goto we_slept;
786         }
787         /* Clear flag in case dquot was inactive (something bad happened) */
788         clear_dquot_dirty(dquot);
789         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
790                 spin_unlock(&dq_list_lock);
791                 dquot->dq_sb->dq_op->release_dquot(dquot);
792                 goto we_slept;
793         }
794         atomic_dec(&dquot->dq_count);
795 #ifdef CONFIG_QUOTA_DEBUG
796         /* sanity check */
797         BUG_ON(!list_empty(&dquot->dq_free));
798 #endif
799         put_dquot_last(dquot);
800         spin_unlock(&dq_list_lock);
801 }
802 EXPORT_SYMBOL(dqput);
803
804 struct dquot *dquot_alloc(struct super_block *sb, int type)
805 {
806         return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
807 }
808 EXPORT_SYMBOL(dquot_alloc);
809
810 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
811 {
812         struct dquot *dquot;
813
814         dquot = sb->dq_op->alloc_dquot(sb, type);
815         if(!dquot)
816                 return NULL;
817
818         mutex_init(&dquot->dq_lock);
819         INIT_LIST_HEAD(&dquot->dq_free);
820         INIT_LIST_HEAD(&dquot->dq_inuse);
821         INIT_HLIST_NODE(&dquot->dq_hash);
822         INIT_LIST_HEAD(&dquot->dq_dirty);
823         init_waitqueue_head(&dquot->dq_wait_unused);
824         dquot->dq_sb = sb;
825         dquot->dq_id = make_kqid_invalid(type);
826         atomic_set(&dquot->dq_count, 1);
827
828         return dquot;
829 }
830
831 /*
832  * Get reference to dquot
833  *
834  * Locking is slightly tricky here. We are guarded from parallel quotaoff()
835  * destroying our dquot by:
836  *   a) checking for quota flags under dq_list_lock and
837  *   b) getting a reference to dquot before we release dq_list_lock
838  */
839 struct dquot *dqget(struct super_block *sb, struct kqid qid)
840 {
841         unsigned int hashent = hashfn(sb, qid);
842         struct dquot *dquot, *empty = NULL;
843
844         if (!qid_has_mapping(sb->s_user_ns, qid))
845                 return ERR_PTR(-EINVAL);
846
847         if (!sb_has_quota_active(sb, qid.type))
848                 return ERR_PTR(-ESRCH);
849 we_slept:
850         spin_lock(&dq_list_lock);
851         spin_lock(&dq_state_lock);
852         if (!sb_has_quota_active(sb, qid.type)) {
853                 spin_unlock(&dq_state_lock);
854                 spin_unlock(&dq_list_lock);
855                 dquot = ERR_PTR(-ESRCH);
856                 goto out;
857         }
858         spin_unlock(&dq_state_lock);
859
860         dquot = find_dquot(hashent, sb, qid);
861         if (!dquot) {
862                 if (!empty) {
863                         spin_unlock(&dq_list_lock);
864                         empty = get_empty_dquot(sb, qid.type);
865                         if (!empty)
866                                 schedule();     /* Try to wait for a moment... */
867                         goto we_slept;
868                 }
869                 dquot = empty;
870                 empty = NULL;
871                 dquot->dq_id = qid;
872                 /* all dquots go on the inuse_list */
873                 put_inuse(dquot);
874                 /* hash it first so it can be found */
875                 insert_dquot_hash(dquot);
876                 spin_unlock(&dq_list_lock);
877                 dqstats_inc(DQST_LOOKUPS);
878         } else {
879                 if (!atomic_read(&dquot->dq_count))
880                         remove_free_dquot(dquot);
881                 atomic_inc(&dquot->dq_count);
882                 spin_unlock(&dq_list_lock);
883                 dqstats_inc(DQST_CACHE_HITS);
884                 dqstats_inc(DQST_LOOKUPS);
885         }
886         /* Wait for dq_lock - after this we know that either dquot_release() is
887          * already finished or it will be canceled due to dq_count > 1 test */
888         wait_on_dquot(dquot);
889         /* Read the dquot / allocate space in quota file */
890         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
891                 int err;
892
893                 err = sb->dq_op->acquire_dquot(dquot);
894                 if (err < 0) {
895                         dqput(dquot);
896                         dquot = ERR_PTR(err);
897                         goto out;
898                 }
899         }
900         /*
901          * Make sure following reads see filled structure - paired with
902          * smp_mb__before_atomic() in dquot_acquire().
903          */
904         smp_rmb();
905 #ifdef CONFIG_QUOTA_DEBUG
906         BUG_ON(!dquot->dq_sb);  /* Has somebody invalidated entry under us? */
907 #endif
908 out:
909         if (empty)
910                 do_destroy_dquot(empty);
911
912         return dquot;
913 }
914 EXPORT_SYMBOL(dqget);
915
916 static inline struct dquot **i_dquot(struct inode *inode)
917 {
918         return inode->i_sb->s_op->get_dquots(inode);
919 }
920
921 static int dqinit_needed(struct inode *inode, int type)
922 {
923         struct dquot * const *dquots;
924         int cnt;
925
926         if (IS_NOQUOTA(inode))
927                 return 0;
928
929         dquots = i_dquot(inode);
930         if (type != -1)
931                 return !dquots[type];
932         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
933                 if (!dquots[cnt])
934                         return 1;
935         return 0;
936 }
937
938 /* This routine is guarded by dqonoff_mutex mutex */
939 static void add_dquot_ref(struct super_block *sb, int type)
940 {
941         struct inode *inode, *old_inode = NULL;
942 #ifdef CONFIG_QUOTA_DEBUG
943         int reserved = 0;
944 #endif
945
946         spin_lock(&sb->s_inode_list_lock);
947         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
948                 spin_lock(&inode->i_lock);
949                 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
950                     !atomic_read(&inode->i_writecount) ||
951                     !dqinit_needed(inode, type)) {
952                         spin_unlock(&inode->i_lock);
953                         continue;
954                 }
955                 __iget(inode);
956                 spin_unlock(&inode->i_lock);
957                 spin_unlock(&sb->s_inode_list_lock);
958
959 #ifdef CONFIG_QUOTA_DEBUG
960                 if (unlikely(inode_get_rsv_space(inode) > 0))
961                         reserved = 1;
962 #endif
963                 iput(old_inode);
964                 __dquot_initialize(inode, type);
965
966                 /*
967                  * We hold a reference to 'inode' so it couldn't have been
968                  * removed from s_inodes list while we dropped the
969                  * s_inode_list_lock. We cannot iput the inode now as we can be
970                  * holding the last reference and we cannot iput it under
971                  * s_inode_list_lock. So we keep the reference and iput it
972                  * later.
973                  */
974                 old_inode = inode;
975                 spin_lock(&sb->s_inode_list_lock);
976         }
977         spin_unlock(&sb->s_inode_list_lock);
978         iput(old_inode);
979
980 #ifdef CONFIG_QUOTA_DEBUG
981         if (reserved) {
982                 quota_error(sb, "Writes happened before quota was turned on "
983                         "thus quota information is probably inconsistent. "
984                         "Please run quotacheck(8)");
985         }
986 #endif
987 }
988
989 /*
990  * Remove references to dquots from inode and add dquot to list for freeing
991  * if we have the last reference to dquot
992  */
993 static void remove_inode_dquot_ref(struct inode *inode, int type,
994                                    struct list_head *tofree_head)
995 {
996         struct dquot **dquots = i_dquot(inode);
997         struct dquot *dquot = dquots[type];
998
999         if (!dquot)
1000                 return;
1001
1002         dquots[type] = NULL;
1003         if (list_empty(&dquot->dq_free)) {
1004                 /*
1005                  * The inode still has reference to dquot so it can't be in the
1006                  * free list
1007                  */
1008                 spin_lock(&dq_list_lock);
1009                 list_add(&dquot->dq_free, tofree_head);
1010                 spin_unlock(&dq_list_lock);
1011         } else {
1012                 /*
1013                  * Dquot is already in a list to put so we won't drop the last
1014                  * reference here.
1015                  */
1016                 dqput(dquot);
1017         }
1018 }
1019
1020 /*
1021  * Free list of dquots
1022  * Dquots are removed from inodes and no new references can be got so we are
1023  * the only ones holding reference
1024  */
1025 static void put_dquot_list(struct list_head *tofree_head)
1026 {
1027         struct list_head *act_head;
1028         struct dquot *dquot;
1029
1030         act_head = tofree_head->next;
1031         while (act_head != tofree_head) {
1032                 dquot = list_entry(act_head, struct dquot, dq_free);
1033                 act_head = act_head->next;
1034                 /* Remove dquot from the list so we won't have problems... */
1035                 list_del_init(&dquot->dq_free);
1036                 dqput(dquot);
1037         }
1038 }
1039
1040 static void remove_dquot_ref(struct super_block *sb, int type,
1041                 struct list_head *tofree_head)
1042 {
1043         struct inode *inode;
1044         int reserved = 0;
1045
1046         spin_lock(&sb->s_inode_list_lock);
1047         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1048                 /*
1049                  *  We have to scan also I_NEW inodes because they can already
1050                  *  have quota pointer initialized. Luckily, we need to touch
1051                  *  only quota pointers and these have separate locking
1052                  *  (dq_data_lock).
1053                  */
1054                 spin_lock(&dq_data_lock);
1055                 if (!IS_NOQUOTA(inode)) {
1056                         if (unlikely(inode_get_rsv_space(inode) > 0))
1057                                 reserved = 1;
1058                         remove_inode_dquot_ref(inode, type, tofree_head);
1059                 }
1060                 spin_unlock(&dq_data_lock);
1061         }
1062         spin_unlock(&sb->s_inode_list_lock);
1063 #ifdef CONFIG_QUOTA_DEBUG
1064         if (reserved) {
1065                 printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1066                         " was disabled thus quota information is probably "
1067                         "inconsistent. Please run quotacheck(8).\n", sb->s_id);
1068         }
1069 #endif
1070 }
1071
1072 /* Gather all references from inodes and drop them */
1073 static void drop_dquot_ref(struct super_block *sb, int type)
1074 {
1075         LIST_HEAD(tofree_head);
1076
1077         if (sb->dq_op) {
1078                 remove_dquot_ref(sb, type, &tofree_head);
1079                 synchronize_srcu(&dquot_srcu);
1080                 put_dquot_list(&tofree_head);
1081         }
1082 }
1083
1084 static inline void dquot_incr_inodes(struct dquot *dquot, qsize_t number)
1085 {
1086         dquot->dq_dqb.dqb_curinodes += number;
1087 }
1088
1089 static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
1090 {
1091         dquot->dq_dqb.dqb_curspace += number;
1092 }
1093
1094 static inline void dquot_resv_space(struct dquot *dquot, qsize_t number)
1095 {
1096         dquot->dq_dqb.dqb_rsvspace += number;
1097 }
1098
1099 /*
1100  * Claim reserved quota space
1101  */
1102 static void dquot_claim_reserved_space(struct dquot *dquot, qsize_t number)
1103 {
1104         if (dquot->dq_dqb.dqb_rsvspace < number) {
1105                 WARN_ON_ONCE(1);
1106                 number = dquot->dq_dqb.dqb_rsvspace;
1107         }
1108         dquot->dq_dqb.dqb_curspace += number;
1109         dquot->dq_dqb.dqb_rsvspace -= number;
1110 }
1111
1112 static void dquot_reclaim_reserved_space(struct dquot *dquot, qsize_t number)
1113 {
1114         if (WARN_ON_ONCE(dquot->dq_dqb.dqb_curspace < number))
1115                 number = dquot->dq_dqb.dqb_curspace;
1116         dquot->dq_dqb.dqb_rsvspace += number;
1117         dquot->dq_dqb.dqb_curspace -= number;
1118 }
1119
1120 static inline
1121 void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1122 {
1123         if (dquot->dq_dqb.dqb_rsvspace >= number)
1124                 dquot->dq_dqb.dqb_rsvspace -= number;
1125         else {
1126                 WARN_ON_ONCE(1);
1127                 dquot->dq_dqb.dqb_rsvspace = 0;
1128         }
1129 }
1130
1131 static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1132 {
1133         if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1134             dquot->dq_dqb.dqb_curinodes >= number)
1135                 dquot->dq_dqb.dqb_curinodes -= number;
1136         else
1137                 dquot->dq_dqb.dqb_curinodes = 0;
1138         if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1139                 dquot->dq_dqb.dqb_itime = (time64_t) 0;
1140         clear_bit(DQ_INODES_B, &dquot->dq_flags);
1141 }
1142
1143 static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1144 {
1145         if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1146             dquot->dq_dqb.dqb_curspace >= number)
1147                 dquot->dq_dqb.dqb_curspace -= number;
1148         else
1149                 dquot->dq_dqb.dqb_curspace = 0;
1150         if (dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1151                 dquot->dq_dqb.dqb_btime = (time64_t) 0;
1152         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1153 }
1154
1155 struct dquot_warn {
1156         struct super_block *w_sb;
1157         struct kqid w_dq_id;
1158         short w_type;
1159 };
1160
1161 static int warning_issued(struct dquot *dquot, const int warntype)
1162 {
1163         int flag = (warntype == QUOTA_NL_BHARDWARN ||
1164                 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1165                 ((warntype == QUOTA_NL_IHARDWARN ||
1166                 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1167
1168         if (!flag)
1169                 return 0;
1170         return test_and_set_bit(flag, &dquot->dq_flags);
1171 }
1172
1173 #ifdef CONFIG_PRINT_QUOTA_WARNING
1174 static int flag_print_warnings = 1;
1175
1176 static int need_print_warning(struct dquot_warn *warn)
1177 {
1178         if (!flag_print_warnings)
1179                 return 0;
1180
1181         switch (warn->w_dq_id.type) {
1182                 case USRQUOTA:
1183                         return uid_eq(current_fsuid(), warn->w_dq_id.uid);
1184                 case GRPQUOTA:
1185                         return in_group_p(warn->w_dq_id.gid);
1186                 case PRJQUOTA:
1187                         return 1;
1188         }
1189         return 0;
1190 }
1191
1192 /* Print warning to user which exceeded quota */
1193 static void print_warning(struct dquot_warn *warn)
1194 {
1195         char *msg = NULL;
1196         struct tty_struct *tty;
1197         int warntype = warn->w_type;
1198
1199         if (warntype == QUOTA_NL_IHARDBELOW ||
1200             warntype == QUOTA_NL_ISOFTBELOW ||
1201             warntype == QUOTA_NL_BHARDBELOW ||
1202             warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(warn))
1203                 return;
1204
1205         tty = get_current_tty();
1206         if (!tty)
1207                 return;
1208         tty_write_message(tty, warn->w_sb->s_id);
1209         if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1210                 tty_write_message(tty, ": warning, ");
1211         else
1212                 tty_write_message(tty, ": write failed, ");
1213         tty_write_message(tty, quotatypes[warn->w_dq_id.type]);
1214         switch (warntype) {
1215                 case QUOTA_NL_IHARDWARN:
1216                         msg = " file limit reached.\r\n";
1217                         break;
1218                 case QUOTA_NL_ISOFTLONGWARN:
1219                         msg = " file quota exceeded too long.\r\n";
1220                         break;
1221                 case QUOTA_NL_ISOFTWARN:
1222                         msg = " file quota exceeded.\r\n";
1223                         break;
1224                 case QUOTA_NL_BHARDWARN:
1225                         msg = " block limit reached.\r\n";
1226                         break;
1227                 case QUOTA_NL_BSOFTLONGWARN:
1228                         msg = " block quota exceeded too long.\r\n";
1229                         break;
1230                 case QUOTA_NL_BSOFTWARN:
1231                         msg = " block quota exceeded.\r\n";
1232                         break;
1233         }
1234         tty_write_message(tty, msg);
1235         tty_kref_put(tty);
1236 }
1237 #endif
1238
1239 static void prepare_warning(struct dquot_warn *warn, struct dquot *dquot,
1240                             int warntype)
1241 {
1242         if (warning_issued(dquot, warntype))
1243                 return;
1244         warn->w_type = warntype;
1245         warn->w_sb = dquot->dq_sb;
1246         warn->w_dq_id = dquot->dq_id;
1247 }
1248
1249 /*
1250  * Write warnings to the console and send warning messages over netlink.
1251  *
1252  * Note that this function can call into tty and networking code.
1253  */
1254 static void flush_warnings(struct dquot_warn *warn)
1255 {
1256         int i;
1257
1258         for (i = 0; i < MAXQUOTAS; i++) {
1259                 if (warn[i].w_type == QUOTA_NL_NOWARN)
1260                         continue;
1261 #ifdef CONFIG_PRINT_QUOTA_WARNING
1262                 print_warning(&warn[i]);
1263 #endif
1264                 quota_send_warning(warn[i].w_dq_id,
1265                                    warn[i].w_sb->s_dev, warn[i].w_type);
1266         }
1267 }
1268
1269 static int ignore_hardlimit(struct dquot *dquot)
1270 {
1271         struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
1272
1273         return capable(CAP_SYS_RESOURCE) &&
1274                (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1275                 !(info->dqi_flags & DQF_ROOT_SQUASH));
1276 }
1277
1278 /* needs dq_data_lock */
1279 static int check_idq(struct dquot *dquot, qsize_t inodes,
1280                      struct dquot_warn *warn)
1281 {
1282         qsize_t newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1283
1284         if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type) ||
1285             test_bit(DQ_FAKE_B, &dquot->dq_flags))
1286                 return 0;
1287
1288         if (dquot->dq_dqb.dqb_ihardlimit &&
1289             newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1290             !ignore_hardlimit(dquot)) {
1291                 prepare_warning(warn, dquot, QUOTA_NL_IHARDWARN);
1292                 return -EDQUOT;
1293         }
1294
1295         if (dquot->dq_dqb.dqb_isoftlimit &&
1296             newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1297             dquot->dq_dqb.dqb_itime &&
1298             ktime_get_real_seconds() >= dquot->dq_dqb.dqb_itime &&
1299             !ignore_hardlimit(dquot)) {
1300                 prepare_warning(warn, dquot, QUOTA_NL_ISOFTLONGWARN);
1301                 return -EDQUOT;
1302         }
1303
1304         if (dquot->dq_dqb.dqb_isoftlimit &&
1305             newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1306             dquot->dq_dqb.dqb_itime == 0) {
1307                 prepare_warning(warn, dquot, QUOTA_NL_ISOFTWARN);
1308                 dquot->dq_dqb.dqb_itime = ktime_get_real_seconds() +
1309                     sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type].dqi_igrace;
1310         }
1311
1312         return 0;
1313 }
1314
1315 /* needs dq_data_lock */
1316 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc,
1317                      struct dquot_warn *warn)
1318 {
1319         qsize_t tspace;
1320         struct super_block *sb = dquot->dq_sb;
1321
1322         if (!sb_has_quota_limits_enabled(sb, dquot->dq_id.type) ||
1323             test_bit(DQ_FAKE_B, &dquot->dq_flags))
1324                 return 0;
1325
1326         tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1327                 + space;
1328
1329         if (dquot->dq_dqb.dqb_bhardlimit &&
1330             tspace > dquot->dq_dqb.dqb_bhardlimit &&
1331             !ignore_hardlimit(dquot)) {
1332                 if (!prealloc)
1333                         prepare_warning(warn, dquot, QUOTA_NL_BHARDWARN);
1334                 return -EDQUOT;
1335         }
1336
1337         if (dquot->dq_dqb.dqb_bsoftlimit &&
1338             tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1339             dquot->dq_dqb.dqb_btime &&
1340             ktime_get_real_seconds() >= dquot->dq_dqb.dqb_btime &&
1341             !ignore_hardlimit(dquot)) {
1342                 if (!prealloc)
1343                         prepare_warning(warn, dquot, QUOTA_NL_BSOFTLONGWARN);
1344                 return -EDQUOT;
1345         }
1346
1347         if (dquot->dq_dqb.dqb_bsoftlimit &&
1348             tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1349             dquot->dq_dqb.dqb_btime == 0) {
1350                 if (!prealloc) {
1351                         prepare_warning(warn, dquot, QUOTA_NL_BSOFTWARN);
1352                         dquot->dq_dqb.dqb_btime = ktime_get_real_seconds() +
1353                             sb_dqopt(sb)->info[dquot->dq_id.type].dqi_bgrace;
1354                 }
1355                 else
1356                         /*
1357                          * We don't allow preallocation to exceed softlimit so exceeding will
1358                          * be always printed
1359                          */
1360                         return -EDQUOT;
1361         }
1362
1363         return 0;
1364 }
1365
1366 static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1367 {
1368         qsize_t newinodes;
1369
1370         if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1371             dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1372             !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type))
1373                 return QUOTA_NL_NOWARN;
1374
1375         newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1376         if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1377                 return QUOTA_NL_ISOFTBELOW;
1378         if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1379             newinodes < dquot->dq_dqb.dqb_ihardlimit)
1380                 return QUOTA_NL_IHARDBELOW;
1381         return QUOTA_NL_NOWARN;
1382 }
1383
1384 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1385 {
1386         if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1387             dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1388                 return QUOTA_NL_NOWARN;
1389
1390         if (dquot->dq_dqb.dqb_curspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1391                 return QUOTA_NL_BSOFTBELOW;
1392         if (dquot->dq_dqb.dqb_curspace >= dquot->dq_dqb.dqb_bhardlimit &&
1393             dquot->dq_dqb.dqb_curspace - space < dquot->dq_dqb.dqb_bhardlimit)
1394                 return QUOTA_NL_BHARDBELOW;
1395         return QUOTA_NL_NOWARN;
1396 }
1397
1398 static int dquot_active(const struct inode *inode)
1399 {
1400         struct super_block *sb = inode->i_sb;
1401
1402         if (IS_NOQUOTA(inode))
1403                 return 0;
1404         return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1405 }
1406
1407 /*
1408  * Initialize quota pointers in inode
1409  *
1410  * It is better to call this function outside of any transaction as it
1411  * might need a lot of space in journal for dquot structure allocation.
1412  */
1413 static int __dquot_initialize(struct inode *inode, int type)
1414 {
1415         int cnt, init_needed = 0;
1416         struct dquot **dquots, *got[MAXQUOTAS] = {};
1417         struct super_block *sb = inode->i_sb;
1418         qsize_t rsv;
1419         int ret = 0;
1420
1421         if (!dquot_active(inode))
1422                 return 0;
1423
1424         dquots = i_dquot(inode);
1425
1426         /* First get references to structures we might need. */
1427         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1428                 struct kqid qid;
1429                 kprojid_t projid;
1430                 int rc;
1431                 struct dquot *dquot;
1432
1433                 if (type != -1 && cnt != type)
1434                         continue;
1435                 /*
1436                  * The i_dquot should have been initialized in most cases,
1437                  * we check it without locking here to avoid unnecessary
1438                  * dqget()/dqput() calls.
1439                  */
1440                 if (dquots[cnt])
1441                         continue;
1442
1443                 if (!sb_has_quota_active(sb, cnt))
1444                         continue;
1445
1446                 init_needed = 1;
1447
1448                 switch (cnt) {
1449                 case USRQUOTA:
1450                         qid = make_kqid_uid(inode->i_uid);
1451                         break;
1452                 case GRPQUOTA:
1453                         qid = make_kqid_gid(inode->i_gid);
1454                         break;
1455                 case PRJQUOTA:
1456                         rc = inode->i_sb->dq_op->get_projid(inode, &projid);
1457                         if (rc)
1458                                 continue;
1459                         qid = make_kqid_projid(projid);
1460                         break;
1461                 }
1462                 dquot = dqget(sb, qid);
1463                 if (IS_ERR(dquot)) {
1464                         /* We raced with somebody turning quotas off... */
1465                         if (PTR_ERR(dquot) != -ESRCH) {
1466                                 ret = PTR_ERR(dquot);
1467                                 goto out_put;
1468                         }
1469                         dquot = NULL;
1470                 }
1471                 got[cnt] = dquot;
1472         }
1473
1474         /* All required i_dquot has been initialized */
1475         if (!init_needed)
1476                 return 0;
1477
1478         spin_lock(&dq_data_lock);
1479         if (IS_NOQUOTA(inode))
1480                 goto out_lock;
1481         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1482                 if (type != -1 && cnt != type)
1483                         continue;
1484                 /* Avoid races with quotaoff() */
1485                 if (!sb_has_quota_active(sb, cnt))
1486                         continue;
1487                 /* We could race with quotaon or dqget() could have failed */
1488                 if (!got[cnt])
1489                         continue;
1490                 if (!dquots[cnt]) {
1491                         dquots[cnt] = got[cnt];
1492                         got[cnt] = NULL;
1493                         /*
1494                          * Make quota reservation system happy if someone
1495                          * did a write before quota was turned on
1496                          */
1497                         rsv = inode_get_rsv_space(inode);
1498                         if (unlikely(rsv))
1499                                 dquot_resv_space(dquots[cnt], rsv);
1500                 }
1501         }
1502 out_lock:
1503         spin_unlock(&dq_data_lock);
1504 out_put:
1505         /* Drop unused references */
1506         dqput_all(got);
1507
1508         return ret;
1509 }
1510
1511 int dquot_initialize(struct inode *inode)
1512 {
1513         return __dquot_initialize(inode, -1);
1514 }
1515 EXPORT_SYMBOL(dquot_initialize);
1516
1517 /*
1518  * Release all quotas referenced by inode.
1519  *
1520  * This function only be called on inode free or converting
1521  * a file to quota file, no other users for the i_dquot in
1522  * both cases, so we needn't call synchronize_srcu() after
1523  * clearing i_dquot.
1524  */
1525 static void __dquot_drop(struct inode *inode)
1526 {
1527         int cnt;
1528         struct dquot **dquots = i_dquot(inode);
1529         struct dquot *put[MAXQUOTAS];
1530
1531         spin_lock(&dq_data_lock);
1532         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1533                 put[cnt] = dquots[cnt];
1534                 dquots[cnt] = NULL;
1535         }
1536         spin_unlock(&dq_data_lock);
1537         dqput_all(put);
1538 }
1539
1540 void dquot_drop(struct inode *inode)
1541 {
1542         struct dquot * const *dquots;
1543         int cnt;
1544
1545         if (IS_NOQUOTA(inode))
1546                 return;
1547
1548         /*
1549          * Test before calling to rule out calls from proc and such
1550          * where we are not allowed to block. Note that this is
1551          * actually reliable test even without the lock - the caller
1552          * must assure that nobody can come after the DQUOT_DROP and
1553          * add quota pointers back anyway.
1554          */
1555         dquots = i_dquot(inode);
1556         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1557                 if (dquots[cnt])
1558                         break;
1559         }
1560
1561         if (cnt < MAXQUOTAS)
1562                 __dquot_drop(inode);
1563 }
1564 EXPORT_SYMBOL(dquot_drop);
1565
1566 /*
1567  * inode_reserved_space is managed internally by quota, and protected by
1568  * i_lock similar to i_blocks+i_bytes.
1569  */
1570 static qsize_t *inode_reserved_space(struct inode * inode)
1571 {
1572         /* Filesystem must explicitly define it's own method in order to use
1573          * quota reservation interface */
1574         BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1575         return inode->i_sb->dq_op->get_reserved_space(inode);
1576 }
1577
1578 void inode_add_rsv_space(struct inode *inode, qsize_t number)
1579 {
1580         spin_lock(&inode->i_lock);
1581         *inode_reserved_space(inode) += number;
1582         spin_unlock(&inode->i_lock);
1583 }
1584 EXPORT_SYMBOL(inode_add_rsv_space);
1585
1586 void inode_claim_rsv_space(struct inode *inode, qsize_t number)
1587 {
1588         spin_lock(&inode->i_lock);
1589         *inode_reserved_space(inode) -= number;
1590         __inode_add_bytes(inode, number);
1591         spin_unlock(&inode->i_lock);
1592 }
1593 EXPORT_SYMBOL(inode_claim_rsv_space);
1594
1595 void inode_reclaim_rsv_space(struct inode *inode, qsize_t number)
1596 {
1597         spin_lock(&inode->i_lock);
1598         *inode_reserved_space(inode) += number;
1599         __inode_sub_bytes(inode, number);
1600         spin_unlock(&inode->i_lock);
1601 }
1602 EXPORT_SYMBOL(inode_reclaim_rsv_space);
1603
1604 void inode_sub_rsv_space(struct inode *inode, qsize_t number)
1605 {
1606         spin_lock(&inode->i_lock);
1607         *inode_reserved_space(inode) -= number;
1608         spin_unlock(&inode->i_lock);
1609 }
1610 EXPORT_SYMBOL(inode_sub_rsv_space);
1611
1612 static qsize_t inode_get_rsv_space(struct inode *inode)
1613 {
1614         qsize_t ret;
1615
1616         if (!inode->i_sb->dq_op->get_reserved_space)
1617                 return 0;
1618         spin_lock(&inode->i_lock);
1619         ret = *inode_reserved_space(inode);
1620         spin_unlock(&inode->i_lock);
1621         return ret;
1622 }
1623
1624 static void inode_incr_space(struct inode *inode, qsize_t number,
1625                                 int reserve)
1626 {
1627         if (reserve)
1628                 inode_add_rsv_space(inode, number);
1629         else
1630                 inode_add_bytes(inode, number);
1631 }
1632
1633 static void inode_decr_space(struct inode *inode, qsize_t number, int reserve)
1634 {
1635         if (reserve)
1636                 inode_sub_rsv_space(inode, number);
1637         else
1638                 inode_sub_bytes(inode, number);
1639 }
1640
1641 /*
1642  * This functions updates i_blocks+i_bytes fields and quota information
1643  * (together with appropriate checks).
1644  *
1645  * NOTE: We absolutely rely on the fact that caller dirties the inode
1646  * (usually helpers in quotaops.h care about this) and holds a handle for
1647  * the current transaction so that dquot write and inode write go into the
1648  * same transaction.
1649  */
1650
1651 /*
1652  * This operation can block, but only after everything is updated
1653  */
1654 int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
1655 {
1656         int cnt, ret = 0, index;
1657         struct dquot_warn warn[MAXQUOTAS];
1658         int reserve = flags & DQUOT_SPACE_RESERVE;
1659         struct dquot **dquots;
1660
1661         if (!dquot_active(inode)) {
1662                 inode_incr_space(inode, number, reserve);
1663                 goto out;
1664         }
1665
1666         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1667                 warn[cnt].w_type = QUOTA_NL_NOWARN;
1668
1669         dquots = i_dquot(inode);
1670         index = srcu_read_lock(&dquot_srcu);
1671         spin_lock(&dq_data_lock);
1672         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1673                 if (!dquots[cnt])
1674                         continue;
1675                 ret = check_bdq(dquots[cnt], number,
1676                                 !(flags & DQUOT_SPACE_WARN), &warn[cnt]);
1677                 if (ret && !(flags & DQUOT_SPACE_NOFAIL)) {
1678                         spin_unlock(&dq_data_lock);
1679                         goto out_flush_warn;
1680                 }
1681         }
1682         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1683                 if (!dquots[cnt])
1684                         continue;
1685                 if (reserve)
1686                         dquot_resv_space(dquots[cnt], number);
1687                 else
1688                         dquot_incr_space(dquots[cnt], number);
1689         }
1690         inode_incr_space(inode, number, reserve);
1691         spin_unlock(&dq_data_lock);
1692
1693         if (reserve)
1694                 goto out_flush_warn;
1695         mark_all_dquot_dirty(dquots);
1696 out_flush_warn:
1697         srcu_read_unlock(&dquot_srcu, index);
1698         flush_warnings(warn);
1699 out:
1700         return ret;
1701 }
1702 EXPORT_SYMBOL(__dquot_alloc_space);
1703
1704 /*
1705  * This operation can block, but only after everything is updated
1706  */
1707 int dquot_alloc_inode(struct inode *inode)
1708 {
1709         int cnt, ret = 0, index;
1710         struct dquot_warn warn[MAXQUOTAS];
1711         struct dquot * const *dquots;
1712
1713         if (!dquot_active(inode))
1714                 return 0;
1715         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1716                 warn[cnt].w_type = QUOTA_NL_NOWARN;
1717
1718         dquots = i_dquot(inode);
1719         index = srcu_read_lock(&dquot_srcu);
1720         spin_lock(&dq_data_lock);
1721         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1722                 if (!dquots[cnt])
1723                         continue;
1724                 ret = check_idq(dquots[cnt], 1, &warn[cnt]);
1725                 if (ret)
1726                         goto warn_put_all;
1727         }
1728
1729         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1730                 if (!dquots[cnt])
1731                         continue;
1732                 dquot_incr_inodes(dquots[cnt], 1);
1733         }
1734
1735 warn_put_all:
1736         spin_unlock(&dq_data_lock);
1737         if (ret == 0)
1738                 mark_all_dquot_dirty(dquots);
1739         srcu_read_unlock(&dquot_srcu, index);
1740         flush_warnings(warn);
1741         return ret;
1742 }
1743 EXPORT_SYMBOL(dquot_alloc_inode);
1744
1745 /*
1746  * Convert in-memory reserved quotas to real consumed quotas
1747  */
1748 int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1749 {
1750         struct dquot **dquots;
1751         int cnt, index;
1752
1753         if (!dquot_active(inode)) {
1754                 inode_claim_rsv_space(inode, number);
1755                 return 0;
1756         }
1757
1758         dquots = i_dquot(inode);
1759         index = srcu_read_lock(&dquot_srcu);
1760         spin_lock(&dq_data_lock);
1761         /* Claim reserved quotas to allocated quotas */
1762         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1763                 if (dquots[cnt])
1764                         dquot_claim_reserved_space(dquots[cnt], number);
1765         }
1766         /* Update inode bytes */
1767         inode_claim_rsv_space(inode, number);
1768         spin_unlock(&dq_data_lock);
1769         mark_all_dquot_dirty(dquots);
1770         srcu_read_unlock(&dquot_srcu, index);
1771         return 0;
1772 }
1773 EXPORT_SYMBOL(dquot_claim_space_nodirty);
1774
1775 /*
1776  * Convert allocated space back to in-memory reserved quotas
1777  */
1778 void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
1779 {
1780         struct dquot **dquots;
1781         int cnt, index;
1782
1783         if (!dquot_active(inode)) {
1784                 inode_reclaim_rsv_space(inode, number);
1785                 return;
1786         }
1787
1788         dquots = i_dquot(inode);
1789         index = srcu_read_lock(&dquot_srcu);
1790         spin_lock(&dq_data_lock);
1791         /* Claim reserved quotas to allocated quotas */
1792         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1793                 if (dquots[cnt])
1794                         dquot_reclaim_reserved_space(dquots[cnt], number);
1795         }
1796         /* Update inode bytes */
1797         inode_reclaim_rsv_space(inode, number);
1798         spin_unlock(&dq_data_lock);
1799         mark_all_dquot_dirty(dquots);
1800         srcu_read_unlock(&dquot_srcu, index);
1801         return;
1802 }
1803 EXPORT_SYMBOL(dquot_reclaim_space_nodirty);
1804
1805 /*
1806  * This operation can block, but only after everything is updated
1807  */
1808 void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
1809 {
1810         unsigned int cnt;
1811         struct dquot_warn warn[MAXQUOTAS];
1812         struct dquot **dquots;
1813         int reserve = flags & DQUOT_SPACE_RESERVE, index;
1814
1815         if (!dquot_active(inode)) {
1816                 inode_decr_space(inode, number, reserve);
1817                 return;
1818         }
1819
1820         dquots = i_dquot(inode);
1821         index = srcu_read_lock(&dquot_srcu);
1822         spin_lock(&dq_data_lock);
1823         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1824                 int wtype;
1825
1826                 warn[cnt].w_type = QUOTA_NL_NOWARN;
1827                 if (!dquots[cnt])
1828                         continue;
1829                 wtype = info_bdq_free(dquots[cnt], number);
1830                 if (wtype != QUOTA_NL_NOWARN)
1831                         prepare_warning(&warn[cnt], dquots[cnt], wtype);
1832                 if (reserve)
1833                         dquot_free_reserved_space(dquots[cnt], number);
1834                 else
1835                         dquot_decr_space(dquots[cnt], number);
1836         }
1837         inode_decr_space(inode, number, reserve);
1838         spin_unlock(&dq_data_lock);
1839
1840         if (reserve)
1841                 goto out_unlock;
1842         mark_all_dquot_dirty(dquots);
1843 out_unlock:
1844         srcu_read_unlock(&dquot_srcu, index);
1845         flush_warnings(warn);
1846 }
1847 EXPORT_SYMBOL(__dquot_free_space);
1848
1849 /*
1850  * This operation can block, but only after everything is updated
1851  */
1852 void dquot_free_inode(struct inode *inode)
1853 {
1854         unsigned int cnt;
1855         struct dquot_warn warn[MAXQUOTAS];
1856         struct dquot * const *dquots;
1857         int index;
1858
1859         if (!dquot_active(inode))
1860                 return;
1861
1862         dquots = i_dquot(inode);
1863         index = srcu_read_lock(&dquot_srcu);
1864         spin_lock(&dq_data_lock);
1865         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1866                 int wtype;
1867
1868                 warn[cnt].w_type = QUOTA_NL_NOWARN;
1869                 if (!dquots[cnt])
1870                         continue;
1871                 wtype = info_idq_free(dquots[cnt], 1);
1872                 if (wtype != QUOTA_NL_NOWARN)
1873                         prepare_warning(&warn[cnt], dquots[cnt], wtype);
1874                 dquot_decr_inodes(dquots[cnt], 1);
1875         }
1876         spin_unlock(&dq_data_lock);
1877         mark_all_dquot_dirty(dquots);
1878         srcu_read_unlock(&dquot_srcu, index);
1879         flush_warnings(warn);
1880 }
1881 EXPORT_SYMBOL(dquot_free_inode);
1882
1883 /*
1884  * Transfer the number of inode and blocks from one diskquota to an other.
1885  * On success, dquot references in transfer_to are consumed and references
1886  * to original dquots that need to be released are placed there. On failure,
1887  * references are kept untouched.
1888  *
1889  * This operation can block, but only after everything is updated
1890  * A transaction must be started when entering this function.
1891  *
1892  * We are holding reference on transfer_from & transfer_to, no need to
1893  * protect them by srcu_read_lock().
1894  */
1895 int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
1896 {
1897         qsize_t space, cur_space;
1898         qsize_t rsv_space = 0;
1899         struct dquot *transfer_from[MAXQUOTAS] = {};
1900         int cnt, ret = 0;
1901         char is_valid[MAXQUOTAS] = {};
1902         struct dquot_warn warn_to[MAXQUOTAS];
1903         struct dquot_warn warn_from_inodes[MAXQUOTAS];
1904         struct dquot_warn warn_from_space[MAXQUOTAS];
1905
1906         if (IS_NOQUOTA(inode))
1907                 return 0;
1908         /* Initialize the arrays */
1909         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1910                 warn_to[cnt].w_type = QUOTA_NL_NOWARN;
1911                 warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
1912                 warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
1913         }
1914
1915         spin_lock(&dq_data_lock);
1916         if (IS_NOQUOTA(inode)) {        /* File without quota accounting? */
1917                 spin_unlock(&dq_data_lock);
1918                 return 0;
1919         }
1920         cur_space = inode_get_bytes(inode);
1921         rsv_space = inode_get_rsv_space(inode);
1922         space = cur_space + rsv_space;
1923         /* Build the transfer_from list and check the limits */
1924         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1925                 /*
1926                  * Skip changes for same uid or gid or for turned off quota-type.
1927                  */
1928                 if (!transfer_to[cnt])
1929                         continue;
1930                 /* Avoid races with quotaoff() */
1931                 if (!sb_has_quota_active(inode->i_sb, cnt))
1932                         continue;
1933                 is_valid[cnt] = 1;
1934                 transfer_from[cnt] = i_dquot(inode)[cnt];
1935                 ret = check_idq(transfer_to[cnt], 1, &warn_to[cnt]);
1936                 if (ret)
1937                         goto over_quota;
1938                 ret = check_bdq(transfer_to[cnt], space, 0, &warn_to[cnt]);
1939                 if (ret)
1940                         goto over_quota;
1941         }
1942
1943         /*
1944          * Finally perform the needed transfer from transfer_from to transfer_to
1945          */
1946         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1947                 if (!is_valid[cnt])
1948                         continue;
1949                 /* Due to IO error we might not have transfer_from[] structure */
1950                 if (transfer_from[cnt]) {
1951                         int wtype;
1952                         wtype = info_idq_free(transfer_from[cnt], 1);
1953                         if (wtype != QUOTA_NL_NOWARN)
1954                                 prepare_warning(&warn_from_inodes[cnt],
1955                                                 transfer_from[cnt], wtype);
1956                         wtype = info_bdq_free(transfer_from[cnt], space);
1957                         if (wtype != QUOTA_NL_NOWARN)
1958                                 prepare_warning(&warn_from_space[cnt],
1959                                                 transfer_from[cnt], wtype);
1960                         dquot_decr_inodes(transfer_from[cnt], 1);
1961                         dquot_decr_space(transfer_from[cnt], cur_space);
1962                         dquot_free_reserved_space(transfer_from[cnt],
1963                                                   rsv_space);
1964                 }
1965
1966                 dquot_incr_inodes(transfer_to[cnt], 1);
1967                 dquot_incr_space(transfer_to[cnt], cur_space);
1968                 dquot_resv_space(transfer_to[cnt], rsv_space);
1969
1970                 i_dquot(inode)[cnt] = transfer_to[cnt];
1971         }
1972         spin_unlock(&dq_data_lock);
1973
1974         mark_all_dquot_dirty(transfer_from);
1975         mark_all_dquot_dirty(transfer_to);
1976         flush_warnings(warn_to);
1977         flush_warnings(warn_from_inodes);
1978         flush_warnings(warn_from_space);
1979         /* Pass back references to put */
1980         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1981                 if (is_valid[cnt])
1982                         transfer_to[cnt] = transfer_from[cnt];
1983         return 0;
1984 over_quota:
1985         spin_unlock(&dq_data_lock);
1986         flush_warnings(warn_to);
1987         return ret;
1988 }
1989 EXPORT_SYMBOL(__dquot_transfer);
1990
1991 /* Wrapper for transferring ownership of an inode for uid/gid only
1992  * Called from FSXXX_setattr()
1993  */
1994 int dquot_transfer(struct inode *inode, struct iattr *iattr)
1995 {
1996         struct dquot *transfer_to[MAXQUOTAS] = {};
1997         struct dquot *dquot;
1998         struct super_block *sb = inode->i_sb;
1999         int ret;
2000
2001         if (!dquot_active(inode))
2002                 return 0;
2003
2004         if (iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)){
2005                 dquot = dqget(sb, make_kqid_uid(iattr->ia_uid));
2006                 if (IS_ERR(dquot)) {
2007                         if (PTR_ERR(dquot) != -ESRCH) {
2008                                 ret = PTR_ERR(dquot);
2009                                 goto out_put;
2010                         }
2011                         dquot = NULL;
2012                 }
2013                 transfer_to[USRQUOTA] = dquot;
2014         }
2015         if (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid)){
2016                 dquot = dqget(sb, make_kqid_gid(iattr->ia_gid));
2017                 if (IS_ERR(dquot)) {
2018                         if (PTR_ERR(dquot) != -ESRCH) {
2019                                 ret = PTR_ERR(dquot);
2020                                 goto out_put;
2021                         }
2022                         dquot = NULL;
2023                 }
2024                 transfer_to[GRPQUOTA] = dquot;
2025         }
2026         ret = __dquot_transfer(inode, transfer_to);
2027 out_put:
2028         dqput_all(transfer_to);
2029         return ret;
2030 }
2031 EXPORT_SYMBOL(dquot_transfer);
2032
2033 /*
2034  * Write info of quota file to disk
2035  */
2036 int dquot_commit_info(struct super_block *sb, int type)
2037 {
2038         int ret;
2039         struct quota_info *dqopt = sb_dqopt(sb);
2040
2041         mutex_lock(&dqopt->dqio_mutex);
2042         ret = dqopt->ops[type]->write_file_info(sb, type);
2043         mutex_unlock(&dqopt->dqio_mutex);
2044         return ret;
2045 }
2046 EXPORT_SYMBOL(dquot_commit_info);
2047
2048 int dquot_get_next_id(struct super_block *sb, struct kqid *qid)
2049 {
2050         struct quota_info *dqopt = sb_dqopt(sb);
2051         int err;
2052
2053         mutex_lock(&dqopt->dqonoff_mutex);
2054         if (!sb_has_quota_active(sb, qid->type)) {
2055                 err = -ESRCH;
2056                 goto out;
2057         }
2058         if (!dqopt->ops[qid->type]->get_next_id) {
2059                 err = -ENOSYS;
2060                 goto out;
2061         }
2062         mutex_lock(&dqopt->dqio_mutex);
2063         err = dqopt->ops[qid->type]->get_next_id(sb, qid);
2064         mutex_unlock(&dqopt->dqio_mutex);
2065 out:
2066         mutex_unlock(&dqopt->dqonoff_mutex);
2067
2068         return err;
2069 }
2070 EXPORT_SYMBOL(dquot_get_next_id);
2071
2072 /*
2073  * Definitions of diskquota operations.
2074  */
2075 const struct dquot_operations dquot_operations = {
2076         .write_dquot    = dquot_commit,
2077         .acquire_dquot  = dquot_acquire,
2078         .release_dquot  = dquot_release,
2079         .mark_dirty     = dquot_mark_dquot_dirty,
2080         .write_info     = dquot_commit_info,
2081         .alloc_dquot    = dquot_alloc,
2082         .destroy_dquot  = dquot_destroy,
2083         .get_next_id    = dquot_get_next_id,
2084 };
2085 EXPORT_SYMBOL(dquot_operations);
2086
2087 /*
2088  * Generic helper for ->open on filesystems supporting disk quotas.
2089  */
2090 int dquot_file_open(struct inode *inode, struct file *file)
2091 {
2092         int error;
2093
2094         error = generic_file_open(inode, file);
2095         if (!error && (file->f_mode & FMODE_WRITE))
2096                 dquot_initialize(inode);
2097         return error;
2098 }
2099 EXPORT_SYMBOL(dquot_file_open);
2100
2101 /*
2102  * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
2103  */
2104 int dquot_disable(struct super_block *sb, int type, unsigned int flags)
2105 {
2106         int cnt, ret = 0;
2107         struct quota_info *dqopt = sb_dqopt(sb);
2108         struct inode *toputinode[MAXQUOTAS];
2109
2110         /* Cannot turn off usage accounting without turning off limits, or
2111          * suspend quotas and simultaneously turn quotas off. */
2112         if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
2113             || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
2114             DQUOT_USAGE_ENABLED)))
2115                 return -EINVAL;
2116
2117         /* We need to serialize quota_off() for device */
2118         mutex_lock(&dqopt->dqonoff_mutex);
2119
2120         /*
2121          * Skip everything if there's nothing to do. We have to do this because
2122          * sometimes we are called when fill_super() failed and calling
2123          * sync_fs() in such cases does no good.
2124          */
2125         if (!sb_any_quota_loaded(sb)) {
2126                 mutex_unlock(&dqopt->dqonoff_mutex);
2127                 return 0;
2128         }
2129         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2130                 toputinode[cnt] = NULL;
2131                 if (type != -1 && cnt != type)
2132                         continue;
2133                 if (!sb_has_quota_loaded(sb, cnt))
2134                         continue;
2135
2136                 if (flags & DQUOT_SUSPENDED) {
2137                         spin_lock(&dq_state_lock);
2138                         dqopt->flags |=
2139                                 dquot_state_flag(DQUOT_SUSPENDED, cnt);
2140                         spin_unlock(&dq_state_lock);
2141                 } else {
2142                         spin_lock(&dq_state_lock);
2143                         dqopt->flags &= ~dquot_state_flag(flags, cnt);
2144                         /* Turning off suspended quotas? */
2145                         if (!sb_has_quota_loaded(sb, cnt) &&
2146                             sb_has_quota_suspended(sb, cnt)) {
2147                                 dqopt->flags &= ~dquot_state_flag(
2148                                                         DQUOT_SUSPENDED, cnt);
2149                                 spin_unlock(&dq_state_lock);
2150                                 iput(dqopt->files[cnt]);
2151                                 dqopt->files[cnt] = NULL;
2152                                 continue;
2153                         }
2154                         spin_unlock(&dq_state_lock);
2155                 }
2156
2157                 /* We still have to keep quota loaded? */
2158                 if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
2159                         continue;
2160
2161                 /* Note: these are blocking operations */
2162                 drop_dquot_ref(sb, cnt);
2163                 invalidate_dquots(sb, cnt);
2164                 /*
2165                  * Now all dquots should be invalidated, all writes done so we
2166                  * should be only users of the info. No locks needed.
2167                  */
2168                 if (info_dirty(&dqopt->info[cnt]))
2169                         sb->dq_op->write_info(sb, cnt);
2170                 if (dqopt->ops[cnt]->free_file_info)
2171                         dqopt->ops[cnt]->free_file_info(sb, cnt);
2172                 put_quota_format(dqopt->info[cnt].dqi_format);
2173
2174                 toputinode[cnt] = dqopt->files[cnt];
2175                 if (!sb_has_quota_loaded(sb, cnt))
2176                         dqopt->files[cnt] = NULL;
2177                 dqopt->info[cnt].dqi_flags = 0;
2178                 dqopt->info[cnt].dqi_igrace = 0;
2179                 dqopt->info[cnt].dqi_bgrace = 0;
2180                 dqopt->ops[cnt] = NULL;
2181         }
2182         mutex_unlock(&dqopt->dqonoff_mutex);
2183
2184         /* Skip syncing and setting flags if quota files are hidden */
2185         if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2186                 goto put_inodes;
2187
2188         /* Sync the superblock so that buffers with quota data are written to
2189          * disk (and so userspace sees correct data afterwards). */
2190         if (sb->s_op->sync_fs)
2191                 sb->s_op->sync_fs(sb, 1);
2192         sync_blockdev(sb->s_bdev);
2193         /* Now the quota files are just ordinary files and we can set the
2194          * inode flags back. Moreover we discard the pagecache so that
2195          * userspace sees the writes we did bypassing the pagecache. We
2196          * must also discard the blockdev buffers so that we see the
2197          * changes done by userspace on the next quotaon() */
2198         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2199                 if (toputinode[cnt]) {
2200                         mutex_lock(&dqopt->dqonoff_mutex);
2201                         /* If quota was reenabled in the meantime, we have
2202                          * nothing to do */
2203                         if (!sb_has_quota_loaded(sb, cnt)) {
2204                                 inode_lock(toputinode[cnt]);
2205                                 toputinode[cnt]->i_flags &= ~(S_IMMUTABLE |
2206                                   S_NOATIME | S_NOQUOTA);
2207                                 truncate_inode_pages(&toputinode[cnt]->i_data,
2208                                                      0);
2209                                 inode_unlock(toputinode[cnt]);
2210                                 mark_inode_dirty_sync(toputinode[cnt]);
2211                         }
2212                         mutex_unlock(&dqopt->dqonoff_mutex);
2213                 }
2214         if (sb->s_bdev)
2215                 invalidate_bdev(sb->s_bdev);
2216 put_inodes:
2217         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2218                 if (toputinode[cnt]) {
2219                         /* On remount RO, we keep the inode pointer so that we
2220                          * can reenable quota on the subsequent remount RW. We
2221                          * have to check 'flags' variable and not use sb_has_
2222                          * function because another quotaon / quotaoff could
2223                          * change global state before we got here. We refuse
2224                          * to suspend quotas when there is pending delete on
2225                          * the quota file... */
2226                         if (!(flags & DQUOT_SUSPENDED))
2227                                 iput(toputinode[cnt]);
2228                         else if (!toputinode[cnt]->i_nlink)
2229                                 ret = -EBUSY;
2230                 }
2231         return ret;
2232 }
2233 EXPORT_SYMBOL(dquot_disable);
2234
2235 int dquot_quota_off(struct super_block *sb, int type)
2236 {
2237         return dquot_disable(sb, type,
2238                              DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2239 }
2240 EXPORT_SYMBOL(dquot_quota_off);
2241
2242 /*
2243  *      Turn quotas on on a device
2244  */
2245
2246 /*
2247  * Helper function to turn quotas on when we already have the inode of
2248  * quota file and no quota information is loaded.
2249  */
2250 static int vfs_load_quota_inode(struct inode *inode, int type, int format_id,
2251         unsigned int flags)
2252 {
2253         struct quota_format_type *fmt = find_quota_format(format_id);
2254         struct super_block *sb = inode->i_sb;
2255         struct quota_info *dqopt = sb_dqopt(sb);
2256         int error;
2257         int oldflags = -1;
2258
2259         if (!fmt)
2260                 return -ESRCH;
2261         if (!S_ISREG(inode->i_mode)) {
2262                 error = -EACCES;
2263                 goto out_fmt;
2264         }
2265         if (IS_RDONLY(inode)) {
2266                 error = -EROFS;
2267                 goto out_fmt;
2268         }
2269         if (!sb->s_op->quota_write || !sb->s_op->quota_read ||
2270             (type == PRJQUOTA && sb->dq_op->get_projid == NULL)) {
2271                 error = -EINVAL;
2272                 goto out_fmt;
2273         }
2274         /* Filesystems outside of init_user_ns not yet supported */
2275         if (sb->s_user_ns != &init_user_ns) {
2276                 error = -EINVAL;
2277                 goto out_fmt;
2278         }
2279         /* Usage always has to be set... */
2280         if (!(flags & DQUOT_USAGE_ENABLED)) {
2281                 error = -EINVAL;
2282                 goto out_fmt;
2283         }
2284
2285         if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2286                 /* As we bypass the pagecache we must now flush all the
2287                  * dirty data and invalidate caches so that kernel sees
2288                  * changes from userspace. It is not enough to just flush
2289                  * the quota file since if blocksize < pagesize, invalidation
2290                  * of the cache could fail because of other unrelated dirty
2291                  * data */
2292                 sync_filesystem(sb);
2293                 invalidate_bdev(sb->s_bdev);
2294         }
2295         mutex_lock(&dqopt->dqonoff_mutex);
2296         if (sb_has_quota_loaded(sb, type)) {
2297                 error = -EBUSY;
2298                 goto out_lock;
2299         }
2300
2301         if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2302                 /* We don't want quota and atime on quota files (deadlocks
2303                  * possible) Also nobody should write to the file - we use
2304                  * special IO operations which ignore the immutable bit. */
2305                 inode_lock(inode);
2306                 oldflags = inode->i_flags & (S_NOATIME | S_IMMUTABLE |
2307                                              S_NOQUOTA);
2308                 inode->i_flags |= S_NOQUOTA | S_NOATIME | S_IMMUTABLE;
2309                 inode_unlock(inode);
2310                 /*
2311                  * When S_NOQUOTA is set, remove dquot references as no more
2312                  * references can be added
2313                  */
2314                 __dquot_drop(inode);
2315         }
2316
2317         error = -EIO;
2318         dqopt->files[type] = igrab(inode);
2319         if (!dqopt->files[type])
2320                 goto out_lock;
2321         error = -EINVAL;
2322         if (!fmt->qf_ops->check_quota_file(sb, type))
2323                 goto out_file_init;
2324
2325         dqopt->ops[type] = fmt->qf_ops;
2326         dqopt->info[type].dqi_format = fmt;
2327         dqopt->info[type].dqi_fmt_id = format_id;
2328         INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2329         mutex_lock(&dqopt->dqio_mutex);
2330         error = dqopt->ops[type]->read_file_info(sb, type);
2331         if (error < 0) {
2332                 mutex_unlock(&dqopt->dqio_mutex);
2333                 goto out_file_init;
2334         }
2335         if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2336                 dqopt->info[type].dqi_flags |= DQF_SYS_FILE;
2337         mutex_unlock(&dqopt->dqio_mutex);
2338         spin_lock(&dq_state_lock);
2339         dqopt->flags |= dquot_state_flag(flags, type);
2340         spin_unlock(&dq_state_lock);
2341
2342         add_dquot_ref(sb, type);
2343         mutex_unlock(&dqopt->dqonoff_mutex);
2344
2345         return 0;
2346
2347 out_file_init:
2348         dqopt->files[type] = NULL;
2349         iput(inode);
2350 out_lock:
2351         if (oldflags != -1) {
2352                 inode_lock(inode);
2353                 /* Set the flags back (in the case of accidental quotaon()
2354                  * on a wrong file we don't want to mess up the flags) */
2355                 inode->i_flags &= ~(S_NOATIME | S_NOQUOTA | S_IMMUTABLE);
2356                 inode->i_flags |= oldflags;
2357                 inode_unlock(inode);
2358         }
2359         mutex_unlock(&dqopt->dqonoff_mutex);
2360 out_fmt:
2361         put_quota_format(fmt);
2362
2363         return error; 
2364 }
2365
2366 /* Reenable quotas on remount RW */
2367 int dquot_resume(struct super_block *sb, int type)
2368 {
2369         struct quota_info *dqopt = sb_dqopt(sb);
2370         struct inode *inode;
2371         int ret = 0, cnt;
2372         unsigned int flags;
2373
2374         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2375                 if (type != -1 && cnt != type)
2376                         continue;
2377
2378                 mutex_lock(&dqopt->dqonoff_mutex);
2379                 if (!sb_has_quota_suspended(sb, cnt)) {
2380                         mutex_unlock(&dqopt->dqonoff_mutex);
2381                         continue;
2382                 }
2383                 inode = dqopt->files[cnt];
2384                 dqopt->files[cnt] = NULL;
2385                 spin_lock(&dq_state_lock);
2386                 flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2387                                                         DQUOT_LIMITS_ENABLED,
2388                                                         cnt);
2389                 dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2390                 spin_unlock(&dq_state_lock);
2391                 mutex_unlock(&dqopt->dqonoff_mutex);
2392
2393                 flags = dquot_generic_flag(flags, cnt);
2394                 ret = vfs_load_quota_inode(inode, cnt,
2395                                 dqopt->info[cnt].dqi_fmt_id, flags);
2396                 iput(inode);
2397         }
2398
2399         return ret;
2400 }
2401 EXPORT_SYMBOL(dquot_resume);
2402
2403 int dquot_quota_on(struct super_block *sb, int type, int format_id,
2404                    struct path *path)
2405 {
2406         int error = security_quota_on(path->dentry);
2407         if (error)
2408                 return error;
2409         /* Quota file not on the same filesystem? */
2410         if (path->dentry->d_sb != sb)
2411                 error = -EXDEV;
2412         else
2413                 error = vfs_load_quota_inode(d_inode(path->dentry), type,
2414                                              format_id, DQUOT_USAGE_ENABLED |
2415                                              DQUOT_LIMITS_ENABLED);
2416         return error;
2417 }
2418 EXPORT_SYMBOL(dquot_quota_on);
2419
2420 /*
2421  * More powerful function for turning on quotas allowing setting
2422  * of individual quota flags
2423  */
2424 int dquot_enable(struct inode *inode, int type, int format_id,
2425                  unsigned int flags)
2426 {
2427         int ret = 0;
2428         struct super_block *sb = inode->i_sb;
2429         struct quota_info *dqopt = sb_dqopt(sb);
2430
2431         /* Just unsuspend quotas? */
2432         BUG_ON(flags & DQUOT_SUSPENDED);
2433
2434         if (!flags)
2435                 return 0;
2436         /* Just updating flags needed? */
2437         if (sb_has_quota_loaded(sb, type)) {
2438                 mutex_lock(&dqopt->dqonoff_mutex);
2439                 /* Now do a reliable test... */
2440                 if (!sb_has_quota_loaded(sb, type)) {
2441                         mutex_unlock(&dqopt->dqonoff_mutex);
2442                         goto load_quota;
2443                 }
2444                 if (flags & DQUOT_USAGE_ENABLED &&
2445                     sb_has_quota_usage_enabled(sb, type)) {
2446                         ret = -EBUSY;
2447                         goto out_lock;
2448                 }
2449                 if (flags & DQUOT_LIMITS_ENABLED &&
2450                     sb_has_quota_limits_enabled(sb, type)) {
2451                         ret = -EBUSY;
2452                         goto out_lock;
2453                 }
2454                 spin_lock(&dq_state_lock);
2455                 sb_dqopt(sb)->flags |= dquot_state_flag(flags, type);
2456                 spin_unlock(&dq_state_lock);
2457 out_lock:
2458                 mutex_unlock(&dqopt->dqonoff_mutex);
2459                 return ret;
2460         }
2461
2462 load_quota:
2463         return vfs_load_quota_inode(inode, type, format_id, flags);
2464 }
2465 EXPORT_SYMBOL(dquot_enable);
2466
2467 /*
2468  * This function is used when filesystem needs to initialize quotas
2469  * during mount time.
2470  */
2471 int dquot_quota_on_mount(struct super_block *sb, char *qf_name,
2472                 int format_id, int type)
2473 {
2474         struct dentry *dentry;
2475         int error;
2476
2477         dentry = lookup_one_len_unlocked(qf_name, sb->s_root, strlen(qf_name));
2478         if (IS_ERR(dentry))
2479                 return PTR_ERR(dentry);
2480
2481         if (d_really_is_negative(dentry)) {
2482                 error = -ENOENT;
2483                 goto out;
2484         }
2485
2486         error = security_quota_on(dentry);
2487         if (!error)
2488                 error = vfs_load_quota_inode(d_inode(dentry), type, format_id,
2489                                 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2490
2491 out:
2492         dput(dentry);
2493         return error;
2494 }
2495 EXPORT_SYMBOL(dquot_quota_on_mount);
2496
2497 static int dquot_quota_enable(struct super_block *sb, unsigned int flags)
2498 {
2499         int ret;
2500         int type;
2501         struct quota_info *dqopt = sb_dqopt(sb);
2502
2503         if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2504                 return -ENOSYS;
2505         /* Accounting cannot be turned on while fs is mounted */
2506         flags &= ~(FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT);
2507         if (!flags)
2508                 return -EINVAL;
2509         for (type = 0; type < MAXQUOTAS; type++) {
2510                 if (!(flags & qtype_enforce_flag(type)))
2511                         continue;
2512                 /* Can't enforce without accounting */
2513                 if (!sb_has_quota_usage_enabled(sb, type))
2514                         return -EINVAL;
2515                 ret = dquot_enable(dqopt->files[type], type,
2516                                    dqopt->info[type].dqi_fmt_id,
2517                                    DQUOT_LIMITS_ENABLED);
2518                 if (ret < 0)
2519                         goto out_err;
2520         }
2521         return 0;
2522 out_err:
2523         /* Backout enforcement enablement we already did */
2524         for (type--; type >= 0; type--)  {
2525                 if (flags & qtype_enforce_flag(type))
2526                         dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2527         }
2528         /* Error code translation for better compatibility with XFS */
2529         if (ret == -EBUSY)
2530                 ret = -EEXIST;
2531         return ret;
2532 }
2533
2534 static int dquot_quota_disable(struct super_block *sb, unsigned int flags)
2535 {
2536         int ret;
2537         int type;
2538         struct quota_info *dqopt = sb_dqopt(sb);
2539
2540         if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2541                 return -ENOSYS;
2542         /*
2543          * We don't support turning off accounting via quotactl. In principle
2544          * quota infrastructure can do this but filesystems don't expect
2545          * userspace to be able to do it.
2546          */
2547         if (flags &
2548                   (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT))
2549                 return -EOPNOTSUPP;
2550
2551         /* Filter out limits not enabled */
2552         for (type = 0; type < MAXQUOTAS; type++)
2553                 if (!sb_has_quota_limits_enabled(sb, type))
2554                         flags &= ~qtype_enforce_flag(type);
2555         /* Nothing left? */
2556         if (!flags)
2557                 return -EEXIST;
2558         for (type = 0; type < MAXQUOTAS; type++) {
2559                 if (flags & qtype_enforce_flag(type)) {
2560                         ret = dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2561                         if (ret < 0)
2562                                 goto out_err;
2563                 }
2564         }
2565         return 0;
2566 out_err:
2567         /* Backout enforcement disabling we already did */
2568         for (type--; type >= 0; type--)  {
2569                 if (flags & qtype_enforce_flag(type))
2570                         dquot_enable(dqopt->files[type], type,
2571                                      dqopt->info[type].dqi_fmt_id,
2572                                      DQUOT_LIMITS_ENABLED);
2573         }
2574         return ret;
2575 }
2576
2577 /* Generic routine for getting common part of quota structure */
2578 static void do_get_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2579 {
2580         struct mem_dqblk *dm = &dquot->dq_dqb;
2581
2582         memset(di, 0, sizeof(*di));
2583         spin_lock(&dq_data_lock);
2584         di->d_spc_hardlimit = dm->dqb_bhardlimit;
2585         di->d_spc_softlimit = dm->dqb_bsoftlimit;
2586         di->d_ino_hardlimit = dm->dqb_ihardlimit;
2587         di->d_ino_softlimit = dm->dqb_isoftlimit;
2588         di->d_space = dm->dqb_curspace + dm->dqb_rsvspace;
2589         di->d_ino_count = dm->dqb_curinodes;
2590         di->d_spc_timer = dm->dqb_btime;
2591         di->d_ino_timer = dm->dqb_itime;
2592         spin_unlock(&dq_data_lock);
2593 }
2594
2595 int dquot_get_dqblk(struct super_block *sb, struct kqid qid,
2596                     struct qc_dqblk *di)
2597 {
2598         struct dquot *dquot;
2599
2600         dquot = dqget(sb, qid);
2601         if (IS_ERR(dquot))
2602                 return PTR_ERR(dquot);
2603         do_get_dqblk(dquot, di);
2604         dqput(dquot);
2605
2606         return 0;
2607 }
2608 EXPORT_SYMBOL(dquot_get_dqblk);
2609
2610 int dquot_get_next_dqblk(struct super_block *sb, struct kqid *qid,
2611                          struct qc_dqblk *di)
2612 {
2613         struct dquot *dquot;
2614         int err;
2615
2616         if (!sb->dq_op->get_next_id)
2617                 return -ENOSYS;
2618         err = sb->dq_op->get_next_id(sb, qid);
2619         if (err < 0)
2620                 return err;
2621         dquot = dqget(sb, *qid);
2622         if (IS_ERR(dquot))
2623                 return PTR_ERR(dquot);
2624         do_get_dqblk(dquot, di);
2625         dqput(dquot);
2626
2627         return 0;
2628 }
2629 EXPORT_SYMBOL(dquot_get_next_dqblk);
2630
2631 #define VFS_QC_MASK \
2632         (QC_SPACE | QC_SPC_SOFT | QC_SPC_HARD | \
2633          QC_INO_COUNT | QC_INO_SOFT | QC_INO_HARD | \
2634          QC_SPC_TIMER | QC_INO_TIMER)
2635
2636 /* Generic routine for setting common part of quota structure */
2637 static int do_set_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2638 {
2639         struct mem_dqblk *dm = &dquot->dq_dqb;
2640         int check_blim = 0, check_ilim = 0;
2641         struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
2642
2643         if (di->d_fieldmask & ~VFS_QC_MASK)
2644                 return -EINVAL;
2645
2646         if (((di->d_fieldmask & QC_SPC_SOFT) &&
2647              di->d_spc_softlimit > dqi->dqi_max_spc_limit) ||
2648             ((di->d_fieldmask & QC_SPC_HARD) &&
2649              di->d_spc_hardlimit > dqi->dqi_max_spc_limit) ||
2650             ((di->d_fieldmask & QC_INO_SOFT) &&
2651              (di->d_ino_softlimit > dqi->dqi_max_ino_limit)) ||
2652             ((di->d_fieldmask & QC_INO_HARD) &&
2653              (di->d_ino_hardlimit > dqi->dqi_max_ino_limit)))
2654                 return -ERANGE;
2655
2656         spin_lock(&dq_data_lock);
2657         if (di->d_fieldmask & QC_SPACE) {
2658                 dm->dqb_curspace = di->d_space - dm->dqb_rsvspace;
2659                 check_blim = 1;
2660                 set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2661         }
2662
2663         if (di->d_fieldmask & QC_SPC_SOFT)
2664                 dm->dqb_bsoftlimit = di->d_spc_softlimit;
2665         if (di->d_fieldmask & QC_SPC_HARD)
2666                 dm->dqb_bhardlimit = di->d_spc_hardlimit;
2667         if (di->d_fieldmask & (QC_SPC_SOFT | QC_SPC_HARD)) {
2668                 check_blim = 1;
2669                 set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2670         }
2671
2672         if (di->d_fieldmask & QC_INO_COUNT) {
2673                 dm->dqb_curinodes = di->d_ino_count;
2674                 check_ilim = 1;
2675                 set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2676         }
2677
2678         if (di->d_fieldmask & QC_INO_SOFT)
2679                 dm->dqb_isoftlimit = di->d_ino_softlimit;
2680         if (di->d_fieldmask & QC_INO_HARD)
2681                 dm->dqb_ihardlimit = di->d_ino_hardlimit;
2682         if (di->d_fieldmask & (QC_INO_SOFT | QC_INO_HARD)) {
2683                 check_ilim = 1;
2684                 set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2685         }
2686
2687         if (di->d_fieldmask & QC_SPC_TIMER) {
2688                 dm->dqb_btime = di->d_spc_timer;
2689                 check_blim = 1;
2690                 set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2691         }
2692
2693         if (di->d_fieldmask & QC_INO_TIMER) {
2694                 dm->dqb_itime = di->d_ino_timer;
2695                 check_ilim = 1;
2696                 set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2697         }
2698
2699         if (check_blim) {
2700                 if (!dm->dqb_bsoftlimit ||
2701                     dm->dqb_curspace < dm->dqb_bsoftlimit) {
2702                         dm->dqb_btime = 0;
2703                         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2704                 } else if (!(di->d_fieldmask & QC_SPC_TIMER))
2705                         /* Set grace only if user hasn't provided his own... */
2706                         dm->dqb_btime = ktime_get_real_seconds() + dqi->dqi_bgrace;
2707         }
2708         if (check_ilim) {
2709                 if (!dm->dqb_isoftlimit ||
2710                     dm->dqb_curinodes < dm->dqb_isoftlimit) {
2711                         dm->dqb_itime = 0;
2712                         clear_bit(DQ_INODES_B, &dquot->dq_flags);
2713                 } else if (!(di->d_fieldmask & QC_INO_TIMER))
2714                         /* Set grace only if user hasn't provided his own... */
2715                         dm->dqb_itime = ktime_get_real_seconds() + dqi->dqi_igrace;
2716         }
2717         if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2718             dm->dqb_isoftlimit)
2719                 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2720         else
2721                 set_bit(DQ_FAKE_B, &dquot->dq_flags);
2722         spin_unlock(&dq_data_lock);
2723         mark_dquot_dirty(dquot);
2724
2725         return 0;
2726 }
2727
2728 int dquot_set_dqblk(struct super_block *sb, struct kqid qid,
2729                   struct qc_dqblk *di)
2730 {
2731         struct dquot *dquot;
2732         int rc;
2733
2734         dquot = dqget(sb, qid);
2735         if (IS_ERR(dquot)) {
2736                 rc = PTR_ERR(dquot);
2737                 goto out;
2738         }
2739         rc = do_set_dqblk(dquot, di);
2740         dqput(dquot);
2741 out:
2742         return rc;
2743 }
2744 EXPORT_SYMBOL(dquot_set_dqblk);
2745
2746 /* Generic routine for getting common part of quota file information */
2747 int dquot_get_state(struct super_block *sb, struct qc_state *state)
2748 {
2749         struct mem_dqinfo *mi;
2750         struct qc_type_state *tstate;
2751         struct quota_info *dqopt = sb_dqopt(sb);
2752         int type;
2753   
2754         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
2755         memset(state, 0, sizeof(*state));
2756         for (type = 0; type < MAXQUOTAS; type++) {
2757                 if (!sb_has_quota_active(sb, type))
2758                         continue;
2759                 tstate = state->s_state + type;
2760                 mi = sb_dqopt(sb)->info + type;
2761                 tstate->flags = QCI_ACCT_ENABLED;
2762                 spin_lock(&dq_data_lock);
2763                 if (mi->dqi_flags & DQF_SYS_FILE)
2764                         tstate->flags |= QCI_SYSFILE;
2765                 if (mi->dqi_flags & DQF_ROOT_SQUASH)
2766                         tstate->flags |= QCI_ROOT_SQUASH;
2767                 if (sb_has_quota_limits_enabled(sb, type))
2768                         tstate->flags |= QCI_LIMITS_ENFORCED;
2769                 tstate->spc_timelimit = mi->dqi_bgrace;
2770                 tstate->ino_timelimit = mi->dqi_igrace;
2771                 tstate->ino = dqopt->files[type]->i_ino;
2772                 tstate->blocks = dqopt->files[type]->i_blocks;
2773                 tstate->nextents = 1;   /* We don't know... */
2774                 spin_unlock(&dq_data_lock);
2775         }
2776         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2777         return 0;
2778 }
2779 EXPORT_SYMBOL(dquot_get_state);
2780
2781 /* Generic routine for setting common part of quota file information */
2782 int dquot_set_dqinfo(struct super_block *sb, int type, struct qc_info *ii)
2783 {
2784         struct mem_dqinfo *mi;
2785         int err = 0;
2786
2787         if ((ii->i_fieldmask & QC_WARNS_MASK) ||
2788             (ii->i_fieldmask & QC_RT_SPC_TIMER))
2789                 return -EINVAL;
2790         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
2791         if (!sb_has_quota_active(sb, type)) {
2792                 err = -ESRCH;
2793                 goto out;
2794         }
2795         mi = sb_dqopt(sb)->info + type;
2796         if (ii->i_fieldmask & QC_FLAGS) {
2797                 if ((ii->i_flags & QCI_ROOT_SQUASH &&
2798                      mi->dqi_format->qf_fmt_id != QFMT_VFS_OLD)) {
2799                         err = -EINVAL;
2800                         goto out;
2801                 }
2802         }
2803         spin_lock(&dq_data_lock);
2804         if (ii->i_fieldmask & QC_SPC_TIMER)
2805                 mi->dqi_bgrace = ii->i_spc_timelimit;
2806         if (ii->i_fieldmask & QC_INO_TIMER)
2807                 mi->dqi_igrace = ii->i_ino_timelimit;
2808         if (ii->i_fieldmask & QC_FLAGS) {
2809                 if (ii->i_flags & QCI_ROOT_SQUASH)
2810                         mi->dqi_flags |= DQF_ROOT_SQUASH;
2811                 else
2812                         mi->dqi_flags &= ~DQF_ROOT_SQUASH;
2813         }
2814         spin_unlock(&dq_data_lock);
2815         mark_info_dirty(sb, type);
2816         /* Force write to disk */
2817         sb->dq_op->write_info(sb, type);
2818 out:
2819         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2820         return err;
2821 }
2822 EXPORT_SYMBOL(dquot_set_dqinfo);
2823
2824 const struct quotactl_ops dquot_quotactl_ops = {
2825         .quota_on       = dquot_quota_on,
2826         .quota_off      = dquot_quota_off,
2827         .quota_sync     = dquot_quota_sync,
2828         .get_state      = dquot_get_state,
2829         .set_info       = dquot_set_dqinfo,
2830         .get_dqblk      = dquot_get_dqblk,
2831         .get_nextdqblk  = dquot_get_next_dqblk,
2832         .set_dqblk      = dquot_set_dqblk
2833 };
2834 EXPORT_SYMBOL(dquot_quotactl_ops);
2835
2836 const struct quotactl_ops dquot_quotactl_sysfile_ops = {
2837         .quota_enable   = dquot_quota_enable,
2838         .quota_disable  = dquot_quota_disable,
2839         .quota_sync     = dquot_quota_sync,
2840         .get_state      = dquot_get_state,
2841         .set_info       = dquot_set_dqinfo,
2842         .get_dqblk      = dquot_get_dqblk,
2843         .get_nextdqblk  = dquot_get_next_dqblk,
2844         .set_dqblk      = dquot_set_dqblk
2845 };
2846 EXPORT_SYMBOL(dquot_quotactl_sysfile_ops);
2847
2848 static int do_proc_dqstats(struct ctl_table *table, int write,
2849                      void __user *buffer, size_t *lenp, loff_t *ppos)
2850 {
2851         unsigned int type = (int *)table->data - dqstats.stat;
2852
2853         /* Update global table */
2854         dqstats.stat[type] =
2855                         percpu_counter_sum_positive(&dqstats.counter[type]);
2856         return proc_dointvec(table, write, buffer, lenp, ppos);
2857 }
2858
2859 static struct ctl_table fs_dqstats_table[] = {
2860         {
2861                 .procname       = "lookups",
2862                 .data           = &dqstats.stat[DQST_LOOKUPS],
2863                 .maxlen         = sizeof(int),
2864                 .mode           = 0444,
2865                 .proc_handler   = do_proc_dqstats,
2866         },
2867         {
2868                 .procname       = "drops",
2869                 .data           = &dqstats.stat[DQST_DROPS],
2870                 .maxlen         = sizeof(int),
2871                 .mode           = 0444,
2872                 .proc_handler   = do_proc_dqstats,
2873         },
2874         {
2875                 .procname       = "reads",
2876                 .data           = &dqstats.stat[DQST_READS],
2877                 .maxlen         = sizeof(int),
2878                 .mode           = 0444,
2879                 .proc_handler   = do_proc_dqstats,
2880         },
2881         {
2882                 .procname       = "writes",
2883                 .data           = &dqstats.stat[DQST_WRITES],
2884                 .maxlen         = sizeof(int),
2885                 .mode           = 0444,
2886                 .proc_handler   = do_proc_dqstats,
2887         },
2888         {
2889                 .procname       = "cache_hits",
2890                 .data           = &dqstats.stat[DQST_CACHE_HITS],
2891                 .maxlen         = sizeof(int),
2892                 .mode           = 0444,
2893                 .proc_handler   = do_proc_dqstats,
2894         },
2895         {
2896                 .procname       = "allocated_dquots",
2897                 .data           = &dqstats.stat[DQST_ALLOC_DQUOTS],
2898                 .maxlen         = sizeof(int),
2899                 .mode           = 0444,
2900                 .proc_handler   = do_proc_dqstats,
2901         },
2902         {
2903                 .procname       = "free_dquots",
2904                 .data           = &dqstats.stat[DQST_FREE_DQUOTS],
2905                 .maxlen         = sizeof(int),
2906                 .mode           = 0444,
2907                 .proc_handler   = do_proc_dqstats,
2908         },
2909         {
2910                 .procname       = "syncs",
2911                 .data           = &dqstats.stat[DQST_SYNCS],
2912                 .maxlen         = sizeof(int),
2913                 .mode           = 0444,
2914                 .proc_handler   = do_proc_dqstats,
2915         },
2916 #ifdef CONFIG_PRINT_QUOTA_WARNING
2917         {
2918                 .procname       = "warnings",
2919                 .data           = &flag_print_warnings,
2920                 .maxlen         = sizeof(int),
2921                 .mode           = 0644,
2922                 .proc_handler   = proc_dointvec,
2923         },
2924 #endif
2925         { },
2926 };
2927
2928 static struct ctl_table fs_table[] = {
2929         {
2930                 .procname       = "quota",
2931                 .mode           = 0555,
2932                 .child          = fs_dqstats_table,
2933         },
2934         { },
2935 };
2936
2937 static struct ctl_table sys_table[] = {
2938         {
2939                 .procname       = "fs",
2940                 .mode           = 0555,
2941                 .child          = fs_table,
2942         },
2943         { },
2944 };
2945
2946 static int __init dquot_init(void)
2947 {
2948         int i, ret;
2949         unsigned long nr_hash, order;
2950
2951         printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2952
2953         register_sysctl_table(sys_table);
2954
2955         dquot_cachep = kmem_cache_create("dquot",
2956                         sizeof(struct dquot), sizeof(unsigned long) * 4,
2957                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2958                                 SLAB_MEM_SPREAD|SLAB_PANIC),
2959                         NULL);
2960
2961         order = 0;
2962         dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
2963         if (!dquot_hash)
2964                 panic("Cannot create dquot hash table");
2965
2966         for (i = 0; i < _DQST_DQSTAT_LAST; i++) {
2967                 ret = percpu_counter_init(&dqstats.counter[i], 0, GFP_KERNEL);
2968                 if (ret)
2969                         panic("Cannot create dquot stat counters");
2970         }
2971
2972         /* Find power-of-two hlist_heads which can fit into allocation */
2973         nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2974         dq_hash_bits = 0;
2975         do {
2976                 dq_hash_bits++;
2977         } while (nr_hash >> dq_hash_bits);
2978         dq_hash_bits--;
2979
2980         nr_hash = 1UL << dq_hash_bits;
2981         dq_hash_mask = nr_hash - 1;
2982         for (i = 0; i < nr_hash; i++)
2983                 INIT_HLIST_HEAD(dquot_hash + i);
2984
2985         pr_info("VFS: Dquot-cache hash table entries: %ld (order %ld,"
2986                 " %ld bytes)\n", nr_hash, order, (PAGE_SIZE << order));
2987
2988         register_shrinker(&dqcache_shrinker);
2989
2990         return 0;
2991 }
2992 fs_initcall(dquot_init);