1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Implementation of the diskquota system for the LINUX operating system. QUOTA
4 * is implemented using the BSD system call interface as the means of
5 * communication with the user level. This file contains the generic routines
6 * called by the different filesystems on allocation of an inode or block.
7 * These routines take care of the administration needed to have a consistent
8 * diskquota tracking system. The ideas of both user and group quotas are based
9 * on the Melbourne quota system as used on BSD derived systems. The internal
10 * implementation is based on one of the several variants of the LINUX
11 * inode-subsystem with added complexity of the diskquota system.
12 *
13 * Author: Marco van Wieringen <mvw@planets.elm.net>
14 *
15 * Fixes: Dmitry Gorodchanin <pgmdsg@ibi.com>, 11 Feb 96
16 *
17 * Revised list management to avoid races
18 * -- Bill Hawes, <whawes@star.net>, 9/98
19 *
20 * Fixed races in dquot_transfer(), dqget() and dquot_alloc_...().
21 * As the consequence the locking was moved from dquot_decr_...(),
22 * dquot_incr_...() to calling functions.
23 * invalidate_dquots() now writes modified dquots.
24 * Serialized quota_off() and quota_on() for mount point.
25 * Fixed a few bugs in grow_dquots().
26 * Fixed deadlock in write_dquot() - we no longer account quotas on
27 * quota files
28 * remove_dquot_ref() moved to inode.c - it now traverses through inodes
29 * add_dquot_ref() restarts after blocking
30 * Added check for bogus uid and fixed check for group in quotactl.
31 * Jan Kara, <jack@suse.cz>, sponsored by SuSE CR, 10-11/99
32 *
33 * Used struct list_head instead of own list struct
34 * Invalidation of referenced dquots is no longer possible
35 * Improved free_dquots list management
36 * Quota and i_blocks are now updated in one place to avoid races
37 * Warnings are now delayed so we won't block in critical section
38 * Write updated not to require dquot lock
39 * Jan Kara, <jack@suse.cz>, 9/2000
40 *
41 * Added dynamic quota structure allocation
42 * Jan Kara <jack@suse.cz> 12/2000
43 *
44 * Rewritten quota interface. Implemented new quota format and
45 * formats registering.
46 * Jan Kara, <jack@suse.cz>, 2001,2002
47 *
48 * New SMP locking.
49 * Jan Kara, <jack@suse.cz>, 10/2002
50 *
51 * Added journalled quota support, fix lock inversion problems
52 * Jan Kara, <jack@suse.cz>, 2003,2004
53 *
54 * (C) Copyright 1994 - 1997 Marco van Wieringen
55 */
56
57 #include <linux/errno.h>
58 #include <linux/kernel.h>
59 #include <linux/fs.h>
60 #include <linux/mount.h>
61 #include <linux/mm.h>
62 #include <linux/time.h>
63 #include <linux/types.h>
64 #include <linux/string.h>
65 #include <linux/fcntl.h>
66 #include <linux/stat.h>
67 #include <linux/tty.h>
68 #include <linux/file.h>
69 #include <linux/slab.h>
70 #include <linux/sysctl.h>
71 #include <linux/init.h>
72 #include <linux/module.h>
73 #include <linux/proc_fs.h>
74 #include <linux/security.h>
75 #include <linux/sched.h>
76 #include <linux/cred.h>
77 #include <linux/kmod.h>
78 #include <linux/namei.h>
79 #include <linux/capability.h>
80 #include <linux/quotaops.h>
81 #include <linux/blkdev.h>
82 #include <linux/sched/mm.h>
83 #include "../internal.h" /* ugh */
84
85 #include <linux/uaccess.h>
86
87 /*
88 * There are five quota SMP locks:
89 * * dq_list_lock protects all lists with quotas and quota formats.
90 * * dquot->dq_dqb_lock protects data from dq_dqb
91 * * inode->i_lock protects inode->i_blocks, i_bytes and also guards
92 * consistency of dquot->dq_dqb with inode->i_blocks, i_bytes so that
93 * dquot_transfer() can stabilize amount it transfers
94 * * dq_data_lock protects mem_dqinfo structures and modifications of dquot
95 * pointers in the inode
96 * * dq_state_lock protects modifications of quota state (on quotaon and
97 * quotaoff) and readers who care about latest values take it as well.
98 *
99 * The spinlock ordering is hence:
100 * dq_data_lock > dq_list_lock > i_lock > dquot->dq_dqb_lock,
101 * dq_list_lock > dq_state_lock
102 *
103 * Note that some things (eg. sb pointer, type, id) doesn't change during
104 * the life of the dquot structure and so needn't to be protected by a lock
105 *
106 * Operation accessing dquots via inode pointers are protected by dquot_srcu.
107 * Operation of reading pointer needs srcu_read_lock(&dquot_srcu), and
108 * synchronize_srcu(&dquot_srcu) is called after clearing pointers from
109 * inode and before dropping dquot references to avoid use of dquots after
110 * they are freed. dq_data_lock is used to serialize the pointer setting and
111 * clearing operations.
112 * Special care needs to be taken about S_NOQUOTA inode flag (marking that
113 * inode is a quota file). Functions adding pointers from inode to dquots have
114 * to check this flag under dq_data_lock and then (if S_NOQUOTA is not set) they
115 * have to do all pointer modifications before dropping dq_data_lock. This makes
116 * sure they cannot race with quotaon which first sets S_NOQUOTA flag and
117 * then drops all pointers to dquots from an inode.
118 *
119 * Each dquot has its dq_lock mutex. Dquot is locked when it is being read to
120 * memory (or space for it is being allocated) on the first dqget(), when it is
121 * being written out, and when it is being released on the last dqput(). The
122 * allocation and release operations are serialized by the dq_lock and by
123 * checking the use count in dquot_release().
124 *
125 * Lock ordering (including related VFS locks) is the following:
126 * s_umount > i_mutex > journal_lock > dquot->dq_lock > dqio_sem
127 */
128
129 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_list_lock);
130 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_state_lock);
131 __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_data_lock);
132 EXPORT_SYMBOL(dq_data_lock);
133 DEFINE_STATIC_SRCU(dquot_srcu);
134
135 static DECLARE_WAIT_QUEUE_HEAD(dquot_ref_wq);
136
__quota_error(struct super_block * sb,const char * func,const char * fmt,...)137 void __quota_error(struct super_block *sb, const char *func,
138 const char *fmt, ...)
139 {
140 if (printk_ratelimit()) {
141 va_list args;
142 struct va_format vaf;
143
144 va_start(args, fmt);
145
146 vaf.fmt = fmt;
147 vaf.va = &args;
148
149 printk(KERN_ERR "Quota error (device %s): %s: %pV\n",
150 sb->s_id, func, &vaf);
151
152 va_end(args);
153 }
154 }
155 EXPORT_SYMBOL(__quota_error);
156
157 #if defined(CONFIG_QUOTA_DEBUG) || defined(CONFIG_PRINT_QUOTA_WARNING)
158 static char *quotatypes[] = INITQFNAMES;
159 #endif
160 static struct quota_format_type *quota_formats; /* List of registered formats */
161 static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
162
163 /* SLAB cache for dquot structures */
164 static struct kmem_cache *dquot_cachep;
165
register_quota_format(struct quota_format_type * fmt)166 int register_quota_format(struct quota_format_type *fmt)
167 {
168 spin_lock(&dq_list_lock);
169 fmt->qf_next = quota_formats;
170 quota_formats = fmt;
171 spin_unlock(&dq_list_lock);
172 return 0;
173 }
174 EXPORT_SYMBOL(register_quota_format);
175
unregister_quota_format(struct quota_format_type * fmt)176 void unregister_quota_format(struct quota_format_type *fmt)
177 {
178 struct quota_format_type **actqf;
179
180 spin_lock(&dq_list_lock);
181 for (actqf = "a_formats; *actqf && *actqf != fmt;
182 actqf = &(*actqf)->qf_next)
183 ;
184 if (*actqf)
185 *actqf = (*actqf)->qf_next;
186 spin_unlock(&dq_list_lock);
187 }
188 EXPORT_SYMBOL(unregister_quota_format);
189
find_quota_format(int id)190 static struct quota_format_type *find_quota_format(int id)
191 {
192 struct quota_format_type *actqf;
193
194 spin_lock(&dq_list_lock);
195 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
196 actqf = actqf->qf_next)
197 ;
198 if (!actqf || !try_module_get(actqf->qf_owner)) {
199 int qm;
200
201 spin_unlock(&dq_list_lock);
202
203 for (qm = 0; module_names[qm].qm_fmt_id &&
204 module_names[qm].qm_fmt_id != id; qm++)
205 ;
206 if (!module_names[qm].qm_fmt_id ||
207 request_module(module_names[qm].qm_mod_name))
208 return NULL;
209
210 spin_lock(&dq_list_lock);
211 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
212 actqf = actqf->qf_next)
213 ;
214 if (actqf && !try_module_get(actqf->qf_owner))
215 actqf = NULL;
216 }
217 spin_unlock(&dq_list_lock);
218 return actqf;
219 }
220
put_quota_format(struct quota_format_type * fmt)221 static void put_quota_format(struct quota_format_type *fmt)
222 {
223 module_put(fmt->qf_owner);
224 }
225
226 /*
227 * Dquot List Management:
228 * The quota code uses five lists for dquot management: the inuse_list,
229 * releasing_dquots, free_dquots, dqi_dirty_list, and dquot_hash[] array.
230 * A single dquot structure may be on some of those lists, depending on
231 * its current state.
232 *
233 * All dquots are placed to the end of inuse_list when first created, and this
234 * list is used for invalidate operation, which must look at every dquot.
235 *
236 * When the last reference of a dquot will be dropped, the dquot will be
237 * added to releasing_dquots. We'd then queue work item which would call
238 * synchronize_srcu() and after that perform the final cleanup of all the
239 * dquots on the list. Both releasing_dquots and free_dquots use the
240 * dq_free list_head in the dquot struct. When a dquot is removed from
241 * releasing_dquots, a reference count is always subtracted, and if
242 * dq_count == 0 at that point, the dquot will be added to the free_dquots.
243 *
244 * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
245 * and this list is searched whenever we need an available dquot. Dquots are
246 * removed from the list as soon as they are used again, and
247 * dqstats.free_dquots gives the number of dquots on the list. When
248 * dquot is invalidated it's completely released from memory.
249 *
250 * Dirty dquots are added to the dqi_dirty_list of quota_info when mark
251 * dirtied, and this list is searched when writing dirty dquots back to
252 * quota file. Note that some filesystems do dirty dquot tracking on their
253 * own (e.g. in a journal) and thus don't use dqi_dirty_list.
254 *
255 * Dquots with a specific identity (device, type and id) are placed on
256 * one of the dquot_hash[] hash chains. The provides an efficient search
257 * mechanism to locate a specific dquot.
258 */
259
260 static LIST_HEAD(inuse_list);
261 static LIST_HEAD(free_dquots);
262 static LIST_HEAD(releasing_dquots);
263 static unsigned int dq_hash_bits, dq_hash_mask;
264 static struct hlist_head *dquot_hash;
265
266 struct dqstats dqstats;
267 EXPORT_SYMBOL(dqstats);
268
269 static qsize_t inode_get_rsv_space(struct inode *inode);
270 static qsize_t __inode_get_rsv_space(struct inode *inode);
271 static int __dquot_initialize(struct inode *inode, int type);
272
273 static void quota_release_workfn(struct work_struct *work);
274 static DECLARE_DELAYED_WORK(quota_release_work, quota_release_workfn);
275
276 static inline unsigned int
hashfn(const struct super_block * sb,struct kqid qid)277 hashfn(const struct super_block *sb, struct kqid qid)
278 {
279 unsigned int id = from_kqid(&init_user_ns, qid);
280 int type = qid.type;
281 unsigned long tmp;
282
283 tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
284 return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
285 }
286
287 /*
288 * Following list functions expect dq_list_lock to be held
289 */
insert_dquot_hash(struct dquot * dquot)290 static inline void insert_dquot_hash(struct dquot *dquot)
291 {
292 struct hlist_head *head;
293 head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id);
294 hlist_add_head(&dquot->dq_hash, head);
295 }
296
remove_dquot_hash(struct dquot * dquot)297 static inline void remove_dquot_hash(struct dquot *dquot)
298 {
299 hlist_del_init(&dquot->dq_hash);
300 }
301
find_dquot(unsigned int hashent,struct super_block * sb,struct kqid qid)302 static struct dquot *find_dquot(unsigned int hashent, struct super_block *sb,
303 struct kqid qid)
304 {
305 struct hlist_node *node;
306 struct dquot *dquot;
307
308 hlist_for_each (node, dquot_hash+hashent) {
309 dquot = hlist_entry(node, struct dquot, dq_hash);
310 if (dquot->dq_sb == sb && qid_eq(dquot->dq_id, qid))
311 return dquot;
312 }
313 return NULL;
314 }
315
316 /* Add a dquot to the tail of the free list */
put_dquot_last(struct dquot * dquot)317 static inline void put_dquot_last(struct dquot *dquot)
318 {
319 list_add_tail(&dquot->dq_free, &free_dquots);
320 dqstats_inc(DQST_FREE_DQUOTS);
321 }
322
put_releasing_dquots(struct dquot * dquot)323 static inline void put_releasing_dquots(struct dquot *dquot)
324 {
325 list_add_tail(&dquot->dq_free, &releasing_dquots);
326 }
327
remove_free_dquot(struct dquot * dquot)328 static inline void remove_free_dquot(struct dquot *dquot)
329 {
330 if (list_empty(&dquot->dq_free))
331 return;
332 list_del_init(&dquot->dq_free);
333 if (!atomic_read(&dquot->dq_count))
334 dqstats_dec(DQST_FREE_DQUOTS);
335 }
336
put_inuse(struct dquot * dquot)337 static inline void put_inuse(struct dquot *dquot)
338 {
339 /* We add to the back of inuse list so we don't have to restart
340 * when traversing this list and we block */
341 list_add_tail(&dquot->dq_inuse, &inuse_list);
342 dqstats_inc(DQST_ALLOC_DQUOTS);
343 }
344
remove_inuse(struct dquot * dquot)345 static inline void remove_inuse(struct dquot *dquot)
346 {
347 dqstats_dec(DQST_ALLOC_DQUOTS);
348 list_del(&dquot->dq_inuse);
349 }
350 /*
351 * End of list functions needing dq_list_lock
352 */
353
wait_on_dquot(struct dquot * dquot)354 static void wait_on_dquot(struct dquot *dquot)
355 {
356 mutex_lock(&dquot->dq_lock);
357 mutex_unlock(&dquot->dq_lock);
358 }
359
dquot_active(struct dquot * dquot)360 static inline int dquot_active(struct dquot *dquot)
361 {
362 return test_bit(DQ_ACTIVE_B, &dquot->dq_flags);
363 }
364
dquot_dirty(struct dquot * dquot)365 static inline int dquot_dirty(struct dquot *dquot)
366 {
367 return test_bit(DQ_MOD_B, &dquot->dq_flags);
368 }
369
mark_dquot_dirty(struct dquot * dquot)370 static inline int mark_dquot_dirty(struct dquot *dquot)
371 {
372 return dquot->dq_sb->dq_op->mark_dirty(dquot);
373 }
374
375 /* Mark dquot dirty in atomic manner, and return it's old dirty flag state */
dquot_mark_dquot_dirty(struct dquot * dquot)376 int dquot_mark_dquot_dirty(struct dquot *dquot)
377 {
378 int ret = 1;
379
380 if (!dquot_active(dquot))
381 return 0;
382
383 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NOLIST_DIRTY)
384 return test_and_set_bit(DQ_MOD_B, &dquot->dq_flags);
385
386 /* If quota is dirty already, we don't have to acquire dq_list_lock */
387 if (dquot_dirty(dquot))
388 return 1;
389
390 spin_lock(&dq_list_lock);
391 if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags)) {
392 list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
393 info[dquot->dq_id.type].dqi_dirty_list);
394 ret = 0;
395 }
396 spin_unlock(&dq_list_lock);
397 return ret;
398 }
399 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
400
401 /* Dirtify all the dquots - this can block when journalling */
mark_all_dquot_dirty(struct dquot * const * dquot)402 static inline int mark_all_dquot_dirty(struct dquot * const *dquot)
403 {
404 int ret, err, cnt;
405
406 ret = err = 0;
407 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
408 if (dquot[cnt])
409 /* Even in case of error we have to continue */
410 ret = mark_dquot_dirty(dquot[cnt]);
411 if (!err)
412 err = ret;
413 }
414 return err;
415 }
416
dqput_all(struct dquot ** dquot)417 static inline void dqput_all(struct dquot **dquot)
418 {
419 unsigned int cnt;
420
421 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
422 dqput(dquot[cnt]);
423 }
424
clear_dquot_dirty(struct dquot * dquot)425 static inline int clear_dquot_dirty(struct dquot *dquot)
426 {
427 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NOLIST_DIRTY)
428 return test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags);
429
430 spin_lock(&dq_list_lock);
431 if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags)) {
432 spin_unlock(&dq_list_lock);
433 return 0;
434 }
435 list_del_init(&dquot->dq_dirty);
436 spin_unlock(&dq_list_lock);
437 return 1;
438 }
439
mark_info_dirty(struct super_block * sb,int type)440 void mark_info_dirty(struct super_block *sb, int type)
441 {
442 spin_lock(&dq_data_lock);
443 sb_dqopt(sb)->info[type].dqi_flags |= DQF_INFO_DIRTY;
444 spin_unlock(&dq_data_lock);
445 }
446 EXPORT_SYMBOL(mark_info_dirty);
447
448 /*
449 * Read dquot from disk and alloc space for it
450 */
451
dquot_acquire(struct dquot * dquot)452 int dquot_acquire(struct dquot *dquot)
453 {
454 int ret = 0, ret2 = 0;
455 unsigned int memalloc;
456 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
457
458 mutex_lock(&dquot->dq_lock);
459 memalloc = memalloc_nofs_save();
460 if (!test_bit(DQ_READ_B, &dquot->dq_flags)) {
461 ret = dqopt->ops[dquot->dq_id.type]->read_dqblk(dquot);
462 if (ret < 0)
463 goto out_iolock;
464 }
465 /* Make sure flags update is visible after dquot has been filled */
466 smp_mb__before_atomic();
467 set_bit(DQ_READ_B, &dquot->dq_flags);
468 /* Instantiate dquot if needed */
469 if (!dquot_active(dquot) && !dquot->dq_off) {
470 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
471 /* Write the info if needed */
472 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
473 ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
474 dquot->dq_sb, dquot->dq_id.type);
475 }
476 if (ret < 0)
477 goto out_iolock;
478 if (ret2 < 0) {
479 ret = ret2;
480 goto out_iolock;
481 }
482 }
483 /*
484 * Make sure flags update is visible after on-disk struct has been
485 * allocated. Paired with smp_rmb() in dqget().
486 */
487 smp_mb__before_atomic();
488 set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
489 out_iolock:
490 memalloc_nofs_restore(memalloc);
491 mutex_unlock(&dquot->dq_lock);
492 return ret;
493 }
494 EXPORT_SYMBOL(dquot_acquire);
495
496 /*
497 * Write dquot to disk
498 */
dquot_commit(struct dquot * dquot)499 int dquot_commit(struct dquot *dquot)
500 {
501 int ret = 0;
502 unsigned int memalloc;
503 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
504
505 mutex_lock(&dquot->dq_lock);
506 memalloc = memalloc_nofs_save();
507 if (!clear_dquot_dirty(dquot))
508 goto out_lock;
509 /* Inactive dquot can be only if there was error during read/init
510 * => we have better not writing it */
511 if (dquot_active(dquot))
512 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
513 else
514 ret = -EIO;
515 out_lock:
516 memalloc_nofs_restore(memalloc);
517 mutex_unlock(&dquot->dq_lock);
518 return ret;
519 }
520 EXPORT_SYMBOL(dquot_commit);
521
522 /*
523 * Release dquot
524 */
dquot_release(struct dquot * dquot)525 int dquot_release(struct dquot *dquot)
526 {
527 int ret = 0, ret2 = 0;
528 unsigned int memalloc;
529 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
530
531 mutex_lock(&dquot->dq_lock);
532 memalloc = memalloc_nofs_save();
533 /* Check whether we are not racing with some other dqget() */
534 if (dquot_is_busy(dquot))
535 goto out_dqlock;
536 if (dqopt->ops[dquot->dq_id.type]->release_dqblk) {
537 ret = dqopt->ops[dquot->dq_id.type]->release_dqblk(dquot);
538 /* Write the info */
539 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
540 ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
541 dquot->dq_sb, dquot->dq_id.type);
542 }
543 if (ret >= 0)
544 ret = ret2;
545 }
546 clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
547 out_dqlock:
548 memalloc_nofs_restore(memalloc);
549 mutex_unlock(&dquot->dq_lock);
550 return ret;
551 }
552 EXPORT_SYMBOL(dquot_release);
553
dquot_destroy(struct dquot * dquot)554 void dquot_destroy(struct dquot *dquot)
555 {
556 kmem_cache_free(dquot_cachep, dquot);
557 }
558 EXPORT_SYMBOL(dquot_destroy);
559
do_destroy_dquot(struct dquot * dquot)560 static inline void do_destroy_dquot(struct dquot *dquot)
561 {
562 dquot->dq_sb->dq_op->destroy_dquot(dquot);
563 }
564
565 /* Invalidate all dquots on the list. Note that this function is called after
566 * quota is disabled and pointers from inodes removed so there cannot be new
567 * quota users. There can still be some users of quotas due to inodes being
568 * just deleted or pruned by prune_icache() (those are not attached to any
569 * list) or parallel quotactl call. We have to wait for such users.
570 */
invalidate_dquots(struct super_block * sb,int type)571 static void invalidate_dquots(struct super_block *sb, int type)
572 {
573 struct dquot *dquot, *tmp;
574
575 restart:
576 flush_delayed_work("a_release_work);
577
578 spin_lock(&dq_list_lock);
579 list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
580 if (dquot->dq_sb != sb)
581 continue;
582 if (dquot->dq_id.type != type)
583 continue;
584 /* Wait for dquot users */
585 if (atomic_read(&dquot->dq_count)) {
586 /* dquot in releasing_dquots, flush and retry */
587 if (!list_empty(&dquot->dq_free)) {
588 spin_unlock(&dq_list_lock);
589 goto restart;
590 }
591
592 atomic_inc(&dquot->dq_count);
593 spin_unlock(&dq_list_lock);
594 /*
595 * Once dqput() wakes us up, we know it's time to free
596 * the dquot.
597 * IMPORTANT: we rely on the fact that there is always
598 * at most one process waiting for dquot to free.
599 * Otherwise dq_count would be > 1 and we would never
600 * wake up.
601 */
602 wait_event(dquot_ref_wq,
603 atomic_read(&dquot->dq_count) == 1);
604 dqput(dquot);
605 /* At this moment dquot() need not exist (it could be
606 * reclaimed by prune_dqcache(). Hence we must
607 * restart. */
608 goto restart;
609 }
610 /*
611 * Quota now has no users and it has been written on last
612 * dqput()
613 */
614 remove_dquot_hash(dquot);
615 remove_free_dquot(dquot);
616 remove_inuse(dquot);
617 do_destroy_dquot(dquot);
618 }
619 spin_unlock(&dq_list_lock);
620 }
621
622 /* Call callback for every active dquot on given filesystem */
dquot_scan_active(struct super_block * sb,int (* fn)(struct dquot * dquot,unsigned long priv),unsigned long priv)623 int dquot_scan_active(struct super_block *sb,
624 int (*fn)(struct dquot *dquot, unsigned long priv),
625 unsigned long priv)
626 {
627 struct dquot *dquot, *old_dquot = NULL;
628 int ret = 0;
629
630 WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
631
632 spin_lock(&dq_list_lock);
633 list_for_each_entry(dquot, &inuse_list, dq_inuse) {
634 if (!dquot_active(dquot))
635 continue;
636 if (dquot->dq_sb != sb)
637 continue;
638 /* Now we have active dquot so we can just increase use count */
639 atomic_inc(&dquot->dq_count);
640 spin_unlock(&dq_list_lock);
641 dqput(old_dquot);
642 old_dquot = dquot;
643 /*
644 * ->release_dquot() can be racing with us. Our reference
645 * protects us from new calls to it so just wait for any
646 * outstanding call and recheck the DQ_ACTIVE_B after that.
647 */
648 wait_on_dquot(dquot);
649 if (dquot_active(dquot)) {
650 ret = fn(dquot, priv);
651 if (ret < 0)
652 goto out;
653 }
654 spin_lock(&dq_list_lock);
655 /* We are safe to continue now because our dquot could not
656 * be moved out of the inuse list while we hold the reference */
657 }
658 spin_unlock(&dq_list_lock);
659 out:
660 dqput(old_dquot);
661 return ret;
662 }
663 EXPORT_SYMBOL(dquot_scan_active);
664
dquot_write_dquot(struct dquot * dquot)665 static inline int dquot_write_dquot(struct dquot *dquot)
666 {
667 int ret = dquot->dq_sb->dq_op->write_dquot(dquot);
668 if (ret < 0) {
669 quota_error(dquot->dq_sb, "Can't write quota structure "
670 "(error %d). Quota may get out of sync!", ret);
671 /* Clear dirty bit anyway to avoid infinite loop. */
672 clear_dquot_dirty(dquot);
673 }
674 return ret;
675 }
676
677 /* Write all dquot structures to quota files */
dquot_writeback_dquots(struct super_block * sb,int type)678 int dquot_writeback_dquots(struct super_block *sb, int type)
679 {
680 struct list_head dirty;
681 struct dquot *dquot;
682 struct quota_info *dqopt = sb_dqopt(sb);
683 int cnt;
684 int err, ret = 0;
685
686 WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
687
688 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
689 if (type != -1 && cnt != type)
690 continue;
691 if (!sb_has_quota_active(sb, cnt))
692 continue;
693 spin_lock(&dq_list_lock);
694 /* Move list away to avoid livelock. */
695 list_replace_init(&dqopt->info[cnt].dqi_dirty_list, &dirty);
696 while (!list_empty(&dirty)) {
697 dquot = list_first_entry(&dirty, struct dquot,
698 dq_dirty);
699
700 WARN_ON(!dquot_active(dquot));
701
702 /* Now we have active dquot from which someone is
703 * holding reference so we can safely just increase
704 * use count */
705 dqgrab(dquot);
706 spin_unlock(&dq_list_lock);
707 err = dquot_write_dquot(dquot);
708 if (err && !ret)
709 ret = err;
710 dqput(dquot);
711 spin_lock(&dq_list_lock);
712 }
713 spin_unlock(&dq_list_lock);
714 }
715
716 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
717 if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
718 && info_dirty(&dqopt->info[cnt]))
719 sb->dq_op->write_info(sb, cnt);
720 dqstats_inc(DQST_SYNCS);
721
722 return ret;
723 }
724 EXPORT_SYMBOL(dquot_writeback_dquots);
725
726 /* Write all dquot structures to disk and make them visible from userspace */
dquot_quota_sync(struct super_block * sb,int type)727 int dquot_quota_sync(struct super_block *sb, int type)
728 {
729 struct quota_info *dqopt = sb_dqopt(sb);
730 int cnt;
731 int ret;
732
733 ret = dquot_writeback_dquots(sb, type);
734 if (ret)
735 return ret;
736 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
737 return 0;
738
739 /* This is not very clever (and fast) but currently I don't know about
740 * any other simple way of getting quota data to disk and we must get
741 * them there for userspace to be visible... */
742 if (sb->s_op->sync_fs) {
743 ret = sb->s_op->sync_fs(sb, 1);
744 if (ret)
745 return ret;
746 }
747 ret = sync_blockdev(sb->s_bdev);
748 if (ret)
749 return ret;
750
751 /*
752 * Now when everything is written we can discard the pagecache so
753 * that userspace sees the changes.
754 */
755 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
756 if (type != -1 && cnt != type)
757 continue;
758 if (!sb_has_quota_active(sb, cnt))
759 continue;
760 inode_lock(dqopt->files[cnt]);
761 truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
762 inode_unlock(dqopt->files[cnt]);
763 }
764
765 return 0;
766 }
767 EXPORT_SYMBOL(dquot_quota_sync);
768
769 static unsigned long
dqcache_shrink_scan(struct shrinker * shrink,struct shrink_control * sc)770 dqcache_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
771 {
772 struct dquot *dquot;
773 unsigned long freed = 0;
774
775 spin_lock(&dq_list_lock);
776 while (!list_empty(&free_dquots) && sc->nr_to_scan) {
777 dquot = list_first_entry(&free_dquots, struct dquot, dq_free);
778 remove_dquot_hash(dquot);
779 remove_free_dquot(dquot);
780 remove_inuse(dquot);
781 do_destroy_dquot(dquot);
782 sc->nr_to_scan--;
783 freed++;
784 }
785 spin_unlock(&dq_list_lock);
786 return freed;
787 }
788
789 static unsigned long
dqcache_shrink_count(struct shrinker * shrink,struct shrink_control * sc)790 dqcache_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
791 {
792 return vfs_pressure_ratio(
793 percpu_counter_read_positive(&dqstats.counter[DQST_FREE_DQUOTS]));
794 }
795
796 static struct shrinker dqcache_shrinker = {
797 .count_objects = dqcache_shrink_count,
798 .scan_objects = dqcache_shrink_scan,
799 .seeks = DEFAULT_SEEKS,
800 };
801
802 /*
803 * Safely release dquot and put reference to dquot.
804 */
quota_release_workfn(struct work_struct * work)805 static void quota_release_workfn(struct work_struct *work)
806 {
807 struct dquot *dquot;
808 struct list_head rls_head;
809
810 spin_lock(&dq_list_lock);
811 /* Exchange the list head to avoid livelock. */
812 list_replace_init(&releasing_dquots, &rls_head);
813 spin_unlock(&dq_list_lock);
814
815 restart:
816 synchronize_srcu(&dquot_srcu);
817 spin_lock(&dq_list_lock);
818 while (!list_empty(&rls_head)) {
819 dquot = list_first_entry(&rls_head, struct dquot, dq_free);
820 /* Dquot got used again? */
821 if (atomic_read(&dquot->dq_count) > 1) {
822 remove_free_dquot(dquot);
823 atomic_dec(&dquot->dq_count);
824 continue;
825 }
826 if (dquot_dirty(dquot)) {
827 spin_unlock(&dq_list_lock);
828 /* Commit dquot before releasing */
829 dquot_write_dquot(dquot);
830 goto restart;
831 }
832 if (dquot_active(dquot)) {
833 spin_unlock(&dq_list_lock);
834 dquot->dq_sb->dq_op->release_dquot(dquot);
835 goto restart;
836 }
837 /* Dquot is inactive and clean, now move it to free list */
838 remove_free_dquot(dquot);
839 atomic_dec(&dquot->dq_count);
840 put_dquot_last(dquot);
841 }
842 spin_unlock(&dq_list_lock);
843 }
844
845 /*
846 * Put reference to dquot
847 */
dqput(struct dquot * dquot)848 void dqput(struct dquot *dquot)
849 {
850 if (!dquot)
851 return;
852 #ifdef CONFIG_QUOTA_DEBUG
853 if (!atomic_read(&dquot->dq_count)) {
854 quota_error(dquot->dq_sb, "trying to free free dquot of %s %d",
855 quotatypes[dquot->dq_id.type],
856 from_kqid(&init_user_ns, dquot->dq_id));
857 BUG();
858 }
859 #endif
860 dqstats_inc(DQST_DROPS);
861
862 spin_lock(&dq_list_lock);
863 if (atomic_read(&dquot->dq_count) > 1) {
864 /* We have more than one user... nothing to do */
865 atomic_dec(&dquot->dq_count);
866 /* Releasing dquot during quotaoff phase? */
867 if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_id.type) &&
868 atomic_read(&dquot->dq_count) == 1)
869 wake_up(&dquot_ref_wq);
870 spin_unlock(&dq_list_lock);
871 return;
872 }
873
874 /* Need to release dquot? */
875 #ifdef CONFIG_QUOTA_DEBUG
876 /* sanity check */
877 BUG_ON(!list_empty(&dquot->dq_free));
878 #endif
879 put_releasing_dquots(dquot);
880 spin_unlock(&dq_list_lock);
881 queue_delayed_work(system_unbound_wq, "a_release_work, 1);
882 }
883 EXPORT_SYMBOL(dqput);
884
dquot_alloc(struct super_block * sb,int type)885 struct dquot *dquot_alloc(struct super_block *sb, int type)
886 {
887 return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
888 }
889 EXPORT_SYMBOL(dquot_alloc);
890
get_empty_dquot(struct super_block * sb,int type)891 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
892 {
893 struct dquot *dquot;
894
895 dquot = sb->dq_op->alloc_dquot(sb, type);
896 if(!dquot)
897 return NULL;
898
899 mutex_init(&dquot->dq_lock);
900 INIT_LIST_HEAD(&dquot->dq_free);
901 INIT_LIST_HEAD(&dquot->dq_inuse);
902 INIT_HLIST_NODE(&dquot->dq_hash);
903 INIT_LIST_HEAD(&dquot->dq_dirty);
904 dquot->dq_sb = sb;
905 dquot->dq_id = make_kqid_invalid(type);
906 atomic_set(&dquot->dq_count, 1);
907 spin_lock_init(&dquot->dq_dqb_lock);
908
909 return dquot;
910 }
911
912 /*
913 * Get reference to dquot
914 *
915 * Locking is slightly tricky here. We are guarded from parallel quotaoff()
916 * destroying our dquot by:
917 * a) checking for quota flags under dq_list_lock and
918 * b) getting a reference to dquot before we release dq_list_lock
919 */
dqget(struct super_block * sb,struct kqid qid)920 struct dquot *dqget(struct super_block *sb, struct kqid qid)
921 {
922 unsigned int hashent = hashfn(sb, qid);
923 struct dquot *dquot, *empty = NULL;
924
925 if (!qid_has_mapping(sb->s_user_ns, qid))
926 return ERR_PTR(-EINVAL);
927
928 if (!sb_has_quota_active(sb, qid.type))
929 return ERR_PTR(-ESRCH);
930 we_slept:
931 spin_lock(&dq_list_lock);
932 spin_lock(&dq_state_lock);
933 if (!sb_has_quota_active(sb, qid.type)) {
934 spin_unlock(&dq_state_lock);
935 spin_unlock(&dq_list_lock);
936 dquot = ERR_PTR(-ESRCH);
937 goto out;
938 }
939 spin_unlock(&dq_state_lock);
940
941 dquot = find_dquot(hashent, sb, qid);
942 if (!dquot) {
943 if (!empty) {
944 spin_unlock(&dq_list_lock);
945 empty = get_empty_dquot(sb, qid.type);
946 if (!empty)
947 schedule(); /* Try to wait for a moment... */
948 goto we_slept;
949 }
950 dquot = empty;
951 empty = NULL;
952 dquot->dq_id = qid;
953 /* all dquots go on the inuse_list */
954 put_inuse(dquot);
955 /* hash it first so it can be found */
956 insert_dquot_hash(dquot);
957 spin_unlock(&dq_list_lock);
958 dqstats_inc(DQST_LOOKUPS);
959 } else {
960 if (!atomic_read(&dquot->dq_count))
961 remove_free_dquot(dquot);
962 atomic_inc(&dquot->dq_count);
963 spin_unlock(&dq_list_lock);
964 dqstats_inc(DQST_CACHE_HITS);
965 dqstats_inc(DQST_LOOKUPS);
966 }
967 /* Wait for dq_lock - after this we know that either dquot_release() is
968 * already finished or it will be canceled due to dq_count > 1 test */
969 wait_on_dquot(dquot);
970 /* Read the dquot / allocate space in quota file */
971 if (!dquot_active(dquot)) {
972 int err;
973
974 err = sb->dq_op->acquire_dquot(dquot);
975 if (err < 0) {
976 dqput(dquot);
977 dquot = ERR_PTR(err);
978 goto out;
979 }
980 }
981 /*
982 * Make sure following reads see filled structure - paired with
983 * smp_mb__before_atomic() in dquot_acquire().
984 */
985 smp_rmb();
986 #ifdef CONFIG_QUOTA_DEBUG
987 BUG_ON(!dquot->dq_sb); /* Has somebody invalidated entry under us? */
988 #endif
989 out:
990 if (empty)
991 do_destroy_dquot(empty);
992
993 return dquot;
994 }
995 EXPORT_SYMBOL(dqget);
996
i_dquot(struct inode * inode)997 static inline struct dquot **i_dquot(struct inode *inode)
998 {
999 return inode->i_sb->s_op->get_dquots(inode);
1000 }
1001
dqinit_needed(struct inode * inode,int type)1002 static int dqinit_needed(struct inode *inode, int type)
1003 {
1004 struct dquot * const *dquots;
1005 int cnt;
1006
1007 if (IS_NOQUOTA(inode))
1008 return 0;
1009
1010 dquots = i_dquot(inode);
1011 if (type != -1)
1012 return !dquots[type];
1013 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1014 if (!dquots[cnt])
1015 return 1;
1016 return 0;
1017 }
1018
1019 /* This routine is guarded by s_umount semaphore */
add_dquot_ref(struct super_block * sb,int type)1020 static int add_dquot_ref(struct super_block *sb, int type)
1021 {
1022 struct inode *inode, *old_inode = NULL;
1023 #ifdef CONFIG_QUOTA_DEBUG
1024 int reserved = 0;
1025 #endif
1026 int err = 0;
1027
1028 spin_lock(&sb->s_inode_list_lock);
1029 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1030 spin_lock(&inode->i_lock);
1031 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
1032 !atomic_read(&inode->i_writecount) ||
1033 !dqinit_needed(inode, type)) {
1034 spin_unlock(&inode->i_lock);
1035 continue;
1036 }
1037 __iget(inode);
1038 spin_unlock(&inode->i_lock);
1039 spin_unlock(&sb->s_inode_list_lock);
1040
1041 #ifdef CONFIG_QUOTA_DEBUG
1042 if (unlikely(inode_get_rsv_space(inode) > 0))
1043 reserved = 1;
1044 #endif
1045 iput(old_inode);
1046 err = __dquot_initialize(inode, type);
1047 if (err) {
1048 iput(inode);
1049 goto out;
1050 }
1051
1052 /*
1053 * We hold a reference to 'inode' so it couldn't have been
1054 * removed from s_inodes list while we dropped the
1055 * s_inode_list_lock. We cannot iput the inode now as we can be
1056 * holding the last reference and we cannot iput it under
1057 * s_inode_list_lock. So we keep the reference and iput it
1058 * later.
1059 */
1060 old_inode = inode;
1061 cond_resched();
1062 spin_lock(&sb->s_inode_list_lock);
1063 }
1064 spin_unlock(&sb->s_inode_list_lock);
1065 iput(old_inode);
1066 out:
1067 #ifdef CONFIG_QUOTA_DEBUG
1068 if (reserved) {
1069 quota_error(sb, "Writes happened before quota was turned on "
1070 "thus quota information is probably inconsistent. "
1071 "Please run quotacheck(8)");
1072 }
1073 #endif
1074 return err;
1075 }
1076
1077 /*
1078 * Remove references to dquots from inode and add dquot to list for freeing
1079 * if we have the last reference to dquot
1080 */
remove_inode_dquot_ref(struct inode * inode,int type,struct list_head * tofree_head)1081 static void remove_inode_dquot_ref(struct inode *inode, int type,
1082 struct list_head *tofree_head)
1083 {
1084 struct dquot **dquots = i_dquot(inode);
1085 struct dquot *dquot = dquots[type];
1086
1087 if (!dquot)
1088 return;
1089
1090 dquots[type] = NULL;
1091 if (list_empty(&dquot->dq_free)) {
1092 /*
1093 * The inode still has reference to dquot so it can't be in the
1094 * free list
1095 */
1096 spin_lock(&dq_list_lock);
1097 list_add(&dquot->dq_free, tofree_head);
1098 spin_unlock(&dq_list_lock);
1099 } else {
1100 /*
1101 * Dquot is already in a list to put so we won't drop the last
1102 * reference here.
1103 */
1104 dqput(dquot);
1105 }
1106 }
1107
1108 /*
1109 * Free list of dquots
1110 * Dquots are removed from inodes and no new references can be got so we are
1111 * the only ones holding reference
1112 */
put_dquot_list(struct list_head * tofree_head)1113 static void put_dquot_list(struct list_head *tofree_head)
1114 {
1115 struct list_head *act_head;
1116 struct dquot *dquot;
1117
1118 act_head = tofree_head->next;
1119 while (act_head != tofree_head) {
1120 dquot = list_entry(act_head, struct dquot, dq_free);
1121 act_head = act_head->next;
1122 /* Remove dquot from the list so we won't have problems... */
1123 list_del_init(&dquot->dq_free);
1124 dqput(dquot);
1125 }
1126 }
1127
remove_dquot_ref(struct super_block * sb,int type,struct list_head * tofree_head)1128 static void remove_dquot_ref(struct super_block *sb, int type,
1129 struct list_head *tofree_head)
1130 {
1131 struct inode *inode;
1132 #ifdef CONFIG_QUOTA_DEBUG
1133 int reserved = 0;
1134 #endif
1135
1136 spin_lock(&sb->s_inode_list_lock);
1137 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1138 /*
1139 * We have to scan also I_NEW inodes because they can already
1140 * have quota pointer initialized. Luckily, we need to touch
1141 * only quota pointers and these have separate locking
1142 * (dq_data_lock).
1143 */
1144 spin_lock(&dq_data_lock);
1145 if (!IS_NOQUOTA(inode)) {
1146 #ifdef CONFIG_QUOTA_DEBUG
1147 if (unlikely(inode_get_rsv_space(inode) > 0))
1148 reserved = 1;
1149 #endif
1150 remove_inode_dquot_ref(inode, type, tofree_head);
1151 }
1152 spin_unlock(&dq_data_lock);
1153 }
1154 spin_unlock(&sb->s_inode_list_lock);
1155 #ifdef CONFIG_QUOTA_DEBUG
1156 if (reserved) {
1157 printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1158 " was disabled thus quota information is probably "
1159 "inconsistent. Please run quotacheck(8).\n", sb->s_id);
1160 }
1161 #endif
1162 }
1163
1164 /* Gather all references from inodes and drop them */
drop_dquot_ref(struct super_block * sb,int type)1165 static void drop_dquot_ref(struct super_block *sb, int type)
1166 {
1167 LIST_HEAD(tofree_head);
1168
1169 if (sb->dq_op) {
1170 remove_dquot_ref(sb, type, &tofree_head);
1171 synchronize_srcu(&dquot_srcu);
1172 put_dquot_list(&tofree_head);
1173 }
1174 }
1175
1176 static inline
dquot_free_reserved_space(struct dquot * dquot,qsize_t number)1177 void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1178 {
1179 if (dquot->dq_dqb.dqb_rsvspace >= number)
1180 dquot->dq_dqb.dqb_rsvspace -= number;
1181 else {
1182 WARN_ON_ONCE(1);
1183 dquot->dq_dqb.dqb_rsvspace = 0;
1184 }
1185 if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1186 dquot->dq_dqb.dqb_bsoftlimit)
1187 dquot->dq_dqb.dqb_btime = (time64_t) 0;
1188 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1189 }
1190
dquot_decr_inodes(struct dquot * dquot,qsize_t number)1191 static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1192 {
1193 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1194 dquot->dq_dqb.dqb_curinodes >= number)
1195 dquot->dq_dqb.dqb_curinodes -= number;
1196 else
1197 dquot->dq_dqb.dqb_curinodes = 0;
1198 if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1199 dquot->dq_dqb.dqb_itime = (time64_t) 0;
1200 clear_bit(DQ_INODES_B, &dquot->dq_flags);
1201 }
1202
dquot_decr_space(struct dquot * dquot,qsize_t number)1203 static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1204 {
1205 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1206 dquot->dq_dqb.dqb_curspace >= number)
1207 dquot->dq_dqb.dqb_curspace -= number;
1208 else
1209 dquot->dq_dqb.dqb_curspace = 0;
1210 if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1211 dquot->dq_dqb.dqb_bsoftlimit)
1212 dquot->dq_dqb.dqb_btime = (time64_t) 0;
1213 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1214 }
1215
1216 struct dquot_warn {
1217 struct super_block *w_sb;
1218 struct kqid w_dq_id;
1219 short w_type;
1220 };
1221
warning_issued(struct dquot * dquot,const int warntype)1222 static int warning_issued(struct dquot *dquot, const int warntype)
1223 {
1224 int flag = (warntype == QUOTA_NL_BHARDWARN ||
1225 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1226 ((warntype == QUOTA_NL_IHARDWARN ||
1227 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1228
1229 if (!flag)
1230 return 0;
1231 return test_and_set_bit(flag, &dquot->dq_flags);
1232 }
1233
1234 #ifdef CONFIG_PRINT_QUOTA_WARNING
1235 static int flag_print_warnings = 1;
1236
need_print_warning(struct dquot_warn * warn)1237 static int need_print_warning(struct dquot_warn *warn)
1238 {
1239 if (!flag_print_warnings)
1240 return 0;
1241
1242 switch (warn->w_dq_id.type) {
1243 case USRQUOTA:
1244 return uid_eq(current_fsuid(), warn->w_dq_id.uid);
1245 case GRPQUOTA:
1246 return in_group_p(warn->w_dq_id.gid);
1247 case PRJQUOTA:
1248 return 1;
1249 }
1250 return 0;
1251 }
1252
1253 /* Print warning to user which exceeded quota */
print_warning(struct dquot_warn * warn)1254 static void print_warning(struct dquot_warn *warn)
1255 {
1256 char *msg = NULL;
1257 struct tty_struct *tty;
1258 int warntype = warn->w_type;
1259
1260 if (warntype == QUOTA_NL_IHARDBELOW ||
1261 warntype == QUOTA_NL_ISOFTBELOW ||
1262 warntype == QUOTA_NL_BHARDBELOW ||
1263 warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(warn))
1264 return;
1265
1266 tty = get_current_tty();
1267 if (!tty)
1268 return;
1269 tty_write_message(tty, warn->w_sb->s_id);
1270 if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1271 tty_write_message(tty, ": warning, ");
1272 else
1273 tty_write_message(tty, ": write failed, ");
1274 tty_write_message(tty, quotatypes[warn->w_dq_id.type]);
1275 switch (warntype) {
1276 case QUOTA_NL_IHARDWARN:
1277 msg = " file limit reached.\r\n";
1278 break;
1279 case QUOTA_NL_ISOFTLONGWARN:
1280 msg = " file quota exceeded too long.\r\n";
1281 break;
1282 case QUOTA_NL_ISOFTWARN:
1283 msg = " file quota exceeded.\r\n";
1284 break;
1285 case QUOTA_NL_BHARDWARN:
1286 msg = " block limit reached.\r\n";
1287 break;
1288 case QUOTA_NL_BSOFTLONGWARN:
1289 msg = " block quota exceeded too long.\r\n";
1290 break;
1291 case QUOTA_NL_BSOFTWARN:
1292 msg = " block quota exceeded.\r\n";
1293 break;
1294 }
1295 tty_write_message(tty, msg);
1296 tty_kref_put(tty);
1297 }
1298 #endif
1299
prepare_warning(struct dquot_warn * warn,struct dquot * dquot,int warntype)1300 static void prepare_warning(struct dquot_warn *warn, struct dquot *dquot,
1301 int warntype)
1302 {
1303 if (warning_issued(dquot, warntype))
1304 return;
1305 warn->w_type = warntype;
1306 warn->w_sb = dquot->dq_sb;
1307 warn->w_dq_id = dquot->dq_id;
1308 }
1309
1310 /*
1311 * Write warnings to the console and send warning messages over netlink.
1312 *
1313 * Note that this function can call into tty and networking code.
1314 */
flush_warnings(struct dquot_warn * warn)1315 static void flush_warnings(struct dquot_warn *warn)
1316 {
1317 int i;
1318
1319 for (i = 0; i < MAXQUOTAS; i++) {
1320 if (warn[i].w_type == QUOTA_NL_NOWARN)
1321 continue;
1322 #ifdef CONFIG_PRINT_QUOTA_WARNING
1323 print_warning(&warn[i]);
1324 #endif
1325 quota_send_warning(warn[i].w_dq_id,
1326 warn[i].w_sb->s_dev, warn[i].w_type);
1327 }
1328 }
1329
ignore_hardlimit(struct dquot * dquot)1330 static int ignore_hardlimit(struct dquot *dquot)
1331 {
1332 struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
1333
1334 return capable(CAP_SYS_RESOURCE) &&
1335 (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1336 !(info->dqi_flags & DQF_ROOT_SQUASH));
1337 }
1338
dquot_add_inodes(struct dquot * dquot,qsize_t inodes,struct dquot_warn * warn)1339 static int dquot_add_inodes(struct dquot *dquot, qsize_t inodes,
1340 struct dquot_warn *warn)
1341 {
1342 qsize_t newinodes;
1343 int ret = 0;
1344
1345 spin_lock(&dquot->dq_dqb_lock);
1346 newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1347 if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type) ||
1348 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1349 goto add;
1350
1351 if (dquot->dq_dqb.dqb_ihardlimit &&
1352 newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1353 !ignore_hardlimit(dquot)) {
1354 prepare_warning(warn, dquot, QUOTA_NL_IHARDWARN);
1355 ret = -EDQUOT;
1356 goto out;
1357 }
1358
1359 if (dquot->dq_dqb.dqb_isoftlimit &&
1360 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1361 dquot->dq_dqb.dqb_itime &&
1362 ktime_get_real_seconds() >= dquot->dq_dqb.dqb_itime &&
1363 !ignore_hardlimit(dquot)) {
1364 prepare_warning(warn, dquot, QUOTA_NL_ISOFTLONGWARN);
1365 ret = -EDQUOT;
1366 goto out;
1367 }
1368
1369 if (dquot->dq_dqb.dqb_isoftlimit &&
1370 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1371 dquot->dq_dqb.dqb_itime == 0) {
1372 prepare_warning(warn, dquot, QUOTA_NL_ISOFTWARN);
1373 dquot->dq_dqb.dqb_itime = ktime_get_real_seconds() +
1374 sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type].dqi_igrace;
1375 }
1376 add:
1377 dquot->dq_dqb.dqb_curinodes = newinodes;
1378
1379 out:
1380 spin_unlock(&dquot->dq_dqb_lock);
1381 return ret;
1382 }
1383
dquot_add_space(struct dquot * dquot,qsize_t space,qsize_t rsv_space,unsigned int flags,struct dquot_warn * warn)1384 static int dquot_add_space(struct dquot *dquot, qsize_t space,
1385 qsize_t rsv_space, unsigned int flags,
1386 struct dquot_warn *warn)
1387 {
1388 qsize_t tspace;
1389 struct super_block *sb = dquot->dq_sb;
1390 int ret = 0;
1391
1392 spin_lock(&dquot->dq_dqb_lock);
1393 if (!sb_has_quota_limits_enabled(sb, dquot->dq_id.type) ||
1394 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1395 goto finish;
1396
1397 tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1398 + space + rsv_space;
1399
1400 if (dquot->dq_dqb.dqb_bhardlimit &&
1401 tspace > dquot->dq_dqb.dqb_bhardlimit &&
1402 !ignore_hardlimit(dquot)) {
1403 if (flags & DQUOT_SPACE_WARN)
1404 prepare_warning(warn, dquot, QUOTA_NL_BHARDWARN);
1405 ret = -EDQUOT;
1406 goto finish;
1407 }
1408
1409 if (dquot->dq_dqb.dqb_bsoftlimit &&
1410 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1411 dquot->dq_dqb.dqb_btime &&
1412 ktime_get_real_seconds() >= dquot->dq_dqb.dqb_btime &&
1413 !ignore_hardlimit(dquot)) {
1414 if (flags & DQUOT_SPACE_WARN)
1415 prepare_warning(warn, dquot, QUOTA_NL_BSOFTLONGWARN);
1416 ret = -EDQUOT;
1417 goto finish;
1418 }
1419
1420 if (dquot->dq_dqb.dqb_bsoftlimit &&
1421 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1422 dquot->dq_dqb.dqb_btime == 0) {
1423 if (flags & DQUOT_SPACE_WARN) {
1424 prepare_warning(warn, dquot, QUOTA_NL_BSOFTWARN);
1425 dquot->dq_dqb.dqb_btime = ktime_get_real_seconds() +
1426 sb_dqopt(sb)->info[dquot->dq_id.type].dqi_bgrace;
1427 } else {
1428 /*
1429 * We don't allow preallocation to exceed softlimit so exceeding will
1430 * be always printed
1431 */
1432 ret = -EDQUOT;
1433 goto finish;
1434 }
1435 }
1436 finish:
1437 /*
1438 * We have to be careful and go through warning generation & grace time
1439 * setting even if DQUOT_SPACE_NOFAIL is set. That's why we check it
1440 * only here...
1441 */
1442 if (flags & DQUOT_SPACE_NOFAIL)
1443 ret = 0;
1444 if (!ret) {
1445 dquot->dq_dqb.dqb_rsvspace += rsv_space;
1446 dquot->dq_dqb.dqb_curspace += space;
1447 }
1448 spin_unlock(&dquot->dq_dqb_lock);
1449 return ret;
1450 }
1451
info_idq_free(struct dquot * dquot,qsize_t inodes)1452 static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1453 {
1454 qsize_t newinodes;
1455
1456 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1457 dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1458 !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type))
1459 return QUOTA_NL_NOWARN;
1460
1461 newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1462 if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1463 return QUOTA_NL_ISOFTBELOW;
1464 if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1465 newinodes < dquot->dq_dqb.dqb_ihardlimit)
1466 return QUOTA_NL_IHARDBELOW;
1467 return QUOTA_NL_NOWARN;
1468 }
1469
info_bdq_free(struct dquot * dquot,qsize_t space)1470 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1471 {
1472 qsize_t tspace;
1473
1474 tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace;
1475
1476 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1477 tspace <= dquot->dq_dqb.dqb_bsoftlimit)
1478 return QUOTA_NL_NOWARN;
1479
1480 if (tspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1481 return QUOTA_NL_BSOFTBELOW;
1482 if (tspace >= dquot->dq_dqb.dqb_bhardlimit &&
1483 tspace - space < dquot->dq_dqb.dqb_bhardlimit)
1484 return QUOTA_NL_BHARDBELOW;
1485 return QUOTA_NL_NOWARN;
1486 }
1487
inode_quota_active(const struct inode * inode)1488 static int inode_quota_active(const struct inode *inode)
1489 {
1490 struct super_block *sb = inode->i_sb;
1491
1492 if (IS_NOQUOTA(inode))
1493 return 0;
1494 return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1495 }
1496
1497 /*
1498 * Initialize quota pointers in inode
1499 *
1500 * It is better to call this function outside of any transaction as it
1501 * might need a lot of space in journal for dquot structure allocation.
1502 */
__dquot_initialize(struct inode * inode,int type)1503 static int __dquot_initialize(struct inode *inode, int type)
1504 {
1505 int cnt, init_needed = 0;
1506 struct dquot **dquots, *got[MAXQUOTAS] = {};
1507 struct super_block *sb = inode->i_sb;
1508 qsize_t rsv;
1509 int ret = 0;
1510
1511 if (!inode_quota_active(inode))
1512 return 0;
1513
1514 dquots = i_dquot(inode);
1515
1516 /* First get references to structures we might need. */
1517 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1518 struct kqid qid;
1519 kprojid_t projid;
1520 int rc;
1521 struct dquot *dquot;
1522
1523 if (type != -1 && cnt != type)
1524 continue;
1525 /*
1526 * The i_dquot should have been initialized in most cases,
1527 * we check it without locking here to avoid unnecessary
1528 * dqget()/dqput() calls.
1529 */
1530 if (dquots[cnt])
1531 continue;
1532
1533 if (!sb_has_quota_active(sb, cnt))
1534 continue;
1535
1536 init_needed = 1;
1537
1538 switch (cnt) {
1539 case USRQUOTA:
1540 qid = make_kqid_uid(inode->i_uid);
1541 break;
1542 case GRPQUOTA:
1543 qid = make_kqid_gid(inode->i_gid);
1544 break;
1545 case PRJQUOTA:
1546 rc = inode->i_sb->dq_op->get_projid(inode, &projid);
1547 if (rc)
1548 continue;
1549 qid = make_kqid_projid(projid);
1550 break;
1551 }
1552 dquot = dqget(sb, qid);
1553 if (IS_ERR(dquot)) {
1554 /* We raced with somebody turning quotas off... */
1555 if (PTR_ERR(dquot) != -ESRCH) {
1556 ret = PTR_ERR(dquot);
1557 goto out_put;
1558 }
1559 dquot = NULL;
1560 }
1561 got[cnt] = dquot;
1562 }
1563
1564 /* All required i_dquot has been initialized */
1565 if (!init_needed)
1566 return 0;
1567
1568 spin_lock(&dq_data_lock);
1569 if (IS_NOQUOTA(inode))
1570 goto out_lock;
1571 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1572 if (type != -1 && cnt != type)
1573 continue;
1574 /* Avoid races with quotaoff() */
1575 if (!sb_has_quota_active(sb, cnt))
1576 continue;
1577 /* We could race with quotaon or dqget() could have failed */
1578 if (!got[cnt])
1579 continue;
1580 if (!dquots[cnt]) {
1581 dquots[cnt] = got[cnt];
1582 got[cnt] = NULL;
1583 /*
1584 * Make quota reservation system happy if someone
1585 * did a write before quota was turned on
1586 */
1587 rsv = inode_get_rsv_space(inode);
1588 if (unlikely(rsv)) {
1589 spin_lock(&inode->i_lock);
1590 /* Get reservation again under proper lock */
1591 rsv = __inode_get_rsv_space(inode);
1592 spin_lock(&dquots[cnt]->dq_dqb_lock);
1593 dquots[cnt]->dq_dqb.dqb_rsvspace += rsv;
1594 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1595 spin_unlock(&inode->i_lock);
1596 }
1597 }
1598 }
1599 out_lock:
1600 spin_unlock(&dq_data_lock);
1601 out_put:
1602 /* Drop unused references */
1603 dqput_all(got);
1604
1605 return ret;
1606 }
1607
dquot_initialize(struct inode * inode)1608 int dquot_initialize(struct inode *inode)
1609 {
1610 return __dquot_initialize(inode, -1);
1611 }
1612 EXPORT_SYMBOL(dquot_initialize);
1613
dquot_initialize_needed(struct inode * inode)1614 bool dquot_initialize_needed(struct inode *inode)
1615 {
1616 struct dquot **dquots;
1617 int i;
1618
1619 if (!inode_quota_active(inode))
1620 return false;
1621
1622 dquots = i_dquot(inode);
1623 for (i = 0; i < MAXQUOTAS; i++)
1624 if (!dquots[i] && sb_has_quota_active(inode->i_sb, i))
1625 return true;
1626 return false;
1627 }
1628 EXPORT_SYMBOL(dquot_initialize_needed);
1629
1630 /*
1631 * Release all quotas referenced by inode.
1632 *
1633 * This function only be called on inode free or converting
1634 * a file to quota file, no other users for the i_dquot in
1635 * both cases, so we needn't call synchronize_srcu() after
1636 * clearing i_dquot.
1637 */
__dquot_drop(struct inode * inode)1638 static void __dquot_drop(struct inode *inode)
1639 {
1640 int cnt;
1641 struct dquot **dquots = i_dquot(inode);
1642 struct dquot *put[MAXQUOTAS];
1643
1644 spin_lock(&dq_data_lock);
1645 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1646 put[cnt] = dquots[cnt];
1647 dquots[cnt] = NULL;
1648 }
1649 spin_unlock(&dq_data_lock);
1650 dqput_all(put);
1651 }
1652
dquot_drop(struct inode * inode)1653 void dquot_drop(struct inode *inode)
1654 {
1655 struct dquot * const *dquots;
1656 int cnt;
1657
1658 if (IS_NOQUOTA(inode))
1659 return;
1660
1661 /*
1662 * Test before calling to rule out calls from proc and such
1663 * where we are not allowed to block. Note that this is
1664 * actually reliable test even without the lock - the caller
1665 * must assure that nobody can come after the DQUOT_DROP and
1666 * add quota pointers back anyway.
1667 */
1668 dquots = i_dquot(inode);
1669 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1670 if (dquots[cnt])
1671 break;
1672 }
1673
1674 if (cnt < MAXQUOTAS)
1675 __dquot_drop(inode);
1676 }
1677 EXPORT_SYMBOL(dquot_drop);
1678
1679 /*
1680 * inode_reserved_space is managed internally by quota, and protected by
1681 * i_lock similar to i_blocks+i_bytes.
1682 */
inode_reserved_space(struct inode * inode)1683 static qsize_t *inode_reserved_space(struct inode * inode)
1684 {
1685 /* Filesystem must explicitly define it's own method in order to use
1686 * quota reservation interface */
1687 BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1688 return inode->i_sb->dq_op->get_reserved_space(inode);
1689 }
1690
__inode_get_rsv_space(struct inode * inode)1691 static qsize_t __inode_get_rsv_space(struct inode *inode)
1692 {
1693 if (!inode->i_sb->dq_op->get_reserved_space)
1694 return 0;
1695 return *inode_reserved_space(inode);
1696 }
1697
inode_get_rsv_space(struct inode * inode)1698 static qsize_t inode_get_rsv_space(struct inode *inode)
1699 {
1700 qsize_t ret;
1701
1702 if (!inode->i_sb->dq_op->get_reserved_space)
1703 return 0;
1704 spin_lock(&inode->i_lock);
1705 ret = __inode_get_rsv_space(inode);
1706 spin_unlock(&inode->i_lock);
1707 return ret;
1708 }
1709
1710 /*
1711 * This functions updates i_blocks+i_bytes fields and quota information
1712 * (together with appropriate checks).
1713 *
1714 * NOTE: We absolutely rely on the fact that caller dirties the inode
1715 * (usually helpers in quotaops.h care about this) and holds a handle for
1716 * the current transaction so that dquot write and inode write go into the
1717 * same transaction.
1718 */
1719
1720 /*
1721 * This operation can block, but only after everything is updated
1722 */
__dquot_alloc_space(struct inode * inode,qsize_t number,int flags)1723 int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
1724 {
1725 int cnt, ret = 0, index;
1726 struct dquot_warn warn[MAXQUOTAS];
1727 int reserve = flags & DQUOT_SPACE_RESERVE;
1728 struct dquot **dquots;
1729
1730 if (!inode_quota_active(inode)) {
1731 if (reserve) {
1732 spin_lock(&inode->i_lock);
1733 *inode_reserved_space(inode) += number;
1734 spin_unlock(&inode->i_lock);
1735 } else {
1736 inode_add_bytes(inode, number);
1737 }
1738 goto out;
1739 }
1740
1741 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1742 warn[cnt].w_type = QUOTA_NL_NOWARN;
1743
1744 dquots = i_dquot(inode);
1745 index = srcu_read_lock(&dquot_srcu);
1746 spin_lock(&inode->i_lock);
1747 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1748 if (!dquots[cnt])
1749 continue;
1750 if (reserve) {
1751 ret = dquot_add_space(dquots[cnt], 0, number, flags,
1752 &warn[cnt]);
1753 } else {
1754 ret = dquot_add_space(dquots[cnt], number, 0, flags,
1755 &warn[cnt]);
1756 }
1757 if (ret) {
1758 /* Back out changes we already did */
1759 for (cnt--; cnt >= 0; cnt--) {
1760 if (!dquots[cnt])
1761 continue;
1762 spin_lock(&dquots[cnt]->dq_dqb_lock);
1763 if (reserve)
1764 dquot_free_reserved_space(dquots[cnt],
1765 number);
1766 else
1767 dquot_decr_space(dquots[cnt], number);
1768 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1769 }
1770 spin_unlock(&inode->i_lock);
1771 goto out_flush_warn;
1772 }
1773 }
1774 if (reserve)
1775 *inode_reserved_space(inode) += number;
1776 else
1777 __inode_add_bytes(inode, number);
1778 spin_unlock(&inode->i_lock);
1779
1780 if (reserve)
1781 goto out_flush_warn;
1782 mark_all_dquot_dirty(dquots);
1783 out_flush_warn:
1784 srcu_read_unlock(&dquot_srcu, index);
1785 flush_warnings(warn);
1786 out:
1787 return ret;
1788 }
1789 EXPORT_SYMBOL(__dquot_alloc_space);
1790
1791 /*
1792 * This operation can block, but only after everything is updated
1793 */
dquot_alloc_inode(struct inode * inode)1794 int dquot_alloc_inode(struct inode *inode)
1795 {
1796 int cnt, ret = 0, index;
1797 struct dquot_warn warn[MAXQUOTAS];
1798 struct dquot * const *dquots;
1799
1800 if (!inode_quota_active(inode))
1801 return 0;
1802 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1803 warn[cnt].w_type = QUOTA_NL_NOWARN;
1804
1805 dquots = i_dquot(inode);
1806 index = srcu_read_lock(&dquot_srcu);
1807 spin_lock(&inode->i_lock);
1808 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1809 if (!dquots[cnt])
1810 continue;
1811 ret = dquot_add_inodes(dquots[cnt], 1, &warn[cnt]);
1812 if (ret) {
1813 for (cnt--; cnt >= 0; cnt--) {
1814 if (!dquots[cnt])
1815 continue;
1816 /* Back out changes we already did */
1817 spin_lock(&dquots[cnt]->dq_dqb_lock);
1818 dquot_decr_inodes(dquots[cnt], 1);
1819 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1820 }
1821 goto warn_put_all;
1822 }
1823 }
1824
1825 warn_put_all:
1826 spin_unlock(&inode->i_lock);
1827 if (ret == 0)
1828 mark_all_dquot_dirty(dquots);
1829 srcu_read_unlock(&dquot_srcu, index);
1830 flush_warnings(warn);
1831 return ret;
1832 }
1833 EXPORT_SYMBOL(dquot_alloc_inode);
1834
1835 /*
1836 * Convert in-memory reserved quotas to real consumed quotas
1837 */
dquot_claim_space_nodirty(struct inode * inode,qsize_t number)1838 int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1839 {
1840 struct dquot **dquots;
1841 int cnt, index;
1842
1843 if (!inode_quota_active(inode)) {
1844 spin_lock(&inode->i_lock);
1845 *inode_reserved_space(inode) -= number;
1846 __inode_add_bytes(inode, number);
1847 spin_unlock(&inode->i_lock);
1848 return 0;
1849 }
1850
1851 dquots = i_dquot(inode);
1852 index = srcu_read_lock(&dquot_srcu);
1853 spin_lock(&inode->i_lock);
1854 /* Claim reserved quotas to allocated quotas */
1855 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1856 if (dquots[cnt]) {
1857 struct dquot *dquot = dquots[cnt];
1858
1859 spin_lock(&dquot->dq_dqb_lock);
1860 if (WARN_ON_ONCE(dquot->dq_dqb.dqb_rsvspace < number))
1861 number = dquot->dq_dqb.dqb_rsvspace;
1862 dquot->dq_dqb.dqb_curspace += number;
1863 dquot->dq_dqb.dqb_rsvspace -= number;
1864 spin_unlock(&dquot->dq_dqb_lock);
1865 }
1866 }
1867 /* Update inode bytes */
1868 *inode_reserved_space(inode) -= number;
1869 __inode_add_bytes(inode, number);
1870 spin_unlock(&inode->i_lock);
1871 mark_all_dquot_dirty(dquots);
1872 srcu_read_unlock(&dquot_srcu, index);
1873 return 0;
1874 }
1875 EXPORT_SYMBOL(dquot_claim_space_nodirty);
1876
1877 /*
1878 * Convert allocated space back to in-memory reserved quotas
1879 */
dquot_reclaim_space_nodirty(struct inode * inode,qsize_t number)1880 void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
1881 {
1882 struct dquot **dquots;
1883 int cnt, index;
1884
1885 if (!inode_quota_active(inode)) {
1886 spin_lock(&inode->i_lock);
1887 *inode_reserved_space(inode) += number;
1888 __inode_sub_bytes(inode, number);
1889 spin_unlock(&inode->i_lock);
1890 return;
1891 }
1892
1893 dquots = i_dquot(inode);
1894 index = srcu_read_lock(&dquot_srcu);
1895 spin_lock(&inode->i_lock);
1896 /* Claim reserved quotas to allocated quotas */
1897 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1898 if (dquots[cnt]) {
1899 struct dquot *dquot = dquots[cnt];
1900
1901 spin_lock(&dquot->dq_dqb_lock);
1902 if (WARN_ON_ONCE(dquot->dq_dqb.dqb_curspace < number))
1903 number = dquot->dq_dqb.dqb_curspace;
1904 dquot->dq_dqb.dqb_rsvspace += number;
1905 dquot->dq_dqb.dqb_curspace -= number;
1906 spin_unlock(&dquot->dq_dqb_lock);
1907 }
1908 }
1909 /* Update inode bytes */
1910 *inode_reserved_space(inode) += number;
1911 __inode_sub_bytes(inode, number);
1912 spin_unlock(&inode->i_lock);
1913 mark_all_dquot_dirty(dquots);
1914 srcu_read_unlock(&dquot_srcu, index);
1915 return;
1916 }
1917 EXPORT_SYMBOL(dquot_reclaim_space_nodirty);
1918
1919 /*
1920 * This operation can block, but only after everything is updated
1921 */
__dquot_free_space(struct inode * inode,qsize_t number,int flags)1922 void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
1923 {
1924 unsigned int cnt;
1925 struct dquot_warn warn[MAXQUOTAS];
1926 struct dquot **dquots;
1927 int reserve = flags & DQUOT_SPACE_RESERVE, index;
1928
1929 if (!inode_quota_active(inode)) {
1930 if (reserve) {
1931 spin_lock(&inode->i_lock);
1932 *inode_reserved_space(inode) -= number;
1933 spin_unlock(&inode->i_lock);
1934 } else {
1935 inode_sub_bytes(inode, number);
1936 }
1937 return;
1938 }
1939
1940 dquots = i_dquot(inode);
1941 index = srcu_read_lock(&dquot_srcu);
1942 spin_lock(&inode->i_lock);
1943 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1944 int wtype;
1945
1946 warn[cnt].w_type = QUOTA_NL_NOWARN;
1947 if (!dquots[cnt])
1948 continue;
1949 spin_lock(&dquots[cnt]->dq_dqb_lock);
1950 wtype = info_bdq_free(dquots[cnt], number);
1951 if (wtype != QUOTA_NL_NOWARN)
1952 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1953 if (reserve)
1954 dquot_free_reserved_space(dquots[cnt], number);
1955 else
1956 dquot_decr_space(dquots[cnt], number);
1957 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1958 }
1959 if (reserve)
1960 *inode_reserved_space(inode) -= number;
1961 else
1962 __inode_sub_bytes(inode, number);
1963 spin_unlock(&inode->i_lock);
1964
1965 if (reserve)
1966 goto out_unlock;
1967 mark_all_dquot_dirty(dquots);
1968 out_unlock:
1969 srcu_read_unlock(&dquot_srcu, index);
1970 flush_warnings(warn);
1971 }
1972 EXPORT_SYMBOL(__dquot_free_space);
1973
1974 /*
1975 * This operation can block, but only after everything is updated
1976 */
dquot_free_inode(struct inode * inode)1977 void dquot_free_inode(struct inode *inode)
1978 {
1979 unsigned int cnt;
1980 struct dquot_warn warn[MAXQUOTAS];
1981 struct dquot * const *dquots;
1982 int index;
1983
1984 if (!inode_quota_active(inode))
1985 return;
1986
1987 dquots = i_dquot(inode);
1988 index = srcu_read_lock(&dquot_srcu);
1989 spin_lock(&inode->i_lock);
1990 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1991 int wtype;
1992
1993 warn[cnt].w_type = QUOTA_NL_NOWARN;
1994 if (!dquots[cnt])
1995 continue;
1996 spin_lock(&dquots[cnt]->dq_dqb_lock);
1997 wtype = info_idq_free(dquots[cnt], 1);
1998 if (wtype != QUOTA_NL_NOWARN)
1999 prepare_warning(&warn[cnt], dquots[cnt], wtype);
2000 dquot_decr_inodes(dquots[cnt], 1);
2001 spin_unlock(&dquots[cnt]->dq_dqb_lock);
2002 }
2003 spin_unlock(&inode->i_lock);
2004 mark_all_dquot_dirty(dquots);
2005 srcu_read_unlock(&dquot_srcu, index);
2006 flush_warnings(warn);
2007 }
2008 EXPORT_SYMBOL(dquot_free_inode);
2009
2010 /*
2011 * Transfer the number of inode and blocks from one diskquota to an other.
2012 * On success, dquot references in transfer_to are consumed and references
2013 * to original dquots that need to be released are placed there. On failure,
2014 * references are kept untouched.
2015 *
2016 * This operation can block, but only after everything is updated
2017 * A transaction must be started when entering this function.
2018 *
2019 * We are holding reference on transfer_from & transfer_to, no need to
2020 * protect them by srcu_read_lock().
2021 */
__dquot_transfer(struct inode * inode,struct dquot ** transfer_to)2022 int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
2023 {
2024 qsize_t cur_space;
2025 qsize_t rsv_space = 0;
2026 qsize_t inode_usage = 1;
2027 struct dquot *transfer_from[MAXQUOTAS] = {};
2028 int cnt, ret = 0;
2029 char is_valid[MAXQUOTAS] = {};
2030 struct dquot_warn warn_to[MAXQUOTAS];
2031 struct dquot_warn warn_from_inodes[MAXQUOTAS];
2032 struct dquot_warn warn_from_space[MAXQUOTAS];
2033
2034 if (IS_NOQUOTA(inode))
2035 return 0;
2036
2037 if (inode->i_sb->dq_op->get_inode_usage) {
2038 ret = inode->i_sb->dq_op->get_inode_usage(inode, &inode_usage);
2039 if (ret)
2040 return ret;
2041 }
2042
2043 /* Initialize the arrays */
2044 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2045 warn_to[cnt].w_type = QUOTA_NL_NOWARN;
2046 warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
2047 warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
2048 }
2049
2050 spin_lock(&dq_data_lock);
2051 spin_lock(&inode->i_lock);
2052 if (IS_NOQUOTA(inode)) { /* File without quota accounting? */
2053 spin_unlock(&inode->i_lock);
2054 spin_unlock(&dq_data_lock);
2055 return 0;
2056 }
2057 cur_space = __inode_get_bytes(inode);
2058 rsv_space = __inode_get_rsv_space(inode);
2059 /*
2060 * Build the transfer_from list, check limits, and update usage in
2061 * the target structures.
2062 */
2063 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2064 /*
2065 * Skip changes for same uid or gid or for turned off quota-type.
2066 */
2067 if (!transfer_to[cnt])
2068 continue;
2069 /* Avoid races with quotaoff() */
2070 if (!sb_has_quota_active(inode->i_sb, cnt))
2071 continue;
2072 is_valid[cnt] = 1;
2073 transfer_from[cnt] = i_dquot(inode)[cnt];
2074 ret = dquot_add_inodes(transfer_to[cnt], inode_usage,
2075 &warn_to[cnt]);
2076 if (ret)
2077 goto over_quota;
2078 ret = dquot_add_space(transfer_to[cnt], cur_space, rsv_space,
2079 DQUOT_SPACE_WARN, &warn_to[cnt]);
2080 if (ret) {
2081 spin_lock(&transfer_to[cnt]->dq_dqb_lock);
2082 dquot_decr_inodes(transfer_to[cnt], inode_usage);
2083 spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2084 goto over_quota;
2085 }
2086 }
2087
2088 /* Decrease usage for source structures and update quota pointers */
2089 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2090 if (!is_valid[cnt])
2091 continue;
2092 /* Due to IO error we might not have transfer_from[] structure */
2093 if (transfer_from[cnt]) {
2094 int wtype;
2095
2096 spin_lock(&transfer_from[cnt]->dq_dqb_lock);
2097 wtype = info_idq_free(transfer_from[cnt], inode_usage);
2098 if (wtype != QUOTA_NL_NOWARN)
2099 prepare_warning(&warn_from_inodes[cnt],
2100 transfer_from[cnt], wtype);
2101 wtype = info_bdq_free(transfer_from[cnt],
2102 cur_space + rsv_space);
2103 if (wtype != QUOTA_NL_NOWARN)
2104 prepare_warning(&warn_from_space[cnt],
2105 transfer_from[cnt], wtype);
2106 dquot_decr_inodes(transfer_from[cnt], inode_usage);
2107 dquot_decr_space(transfer_from[cnt], cur_space);
2108 dquot_free_reserved_space(transfer_from[cnt],
2109 rsv_space);
2110 spin_unlock(&transfer_from[cnt]->dq_dqb_lock);
2111 }
2112 i_dquot(inode)[cnt] = transfer_to[cnt];
2113 }
2114 spin_unlock(&inode->i_lock);
2115 spin_unlock(&dq_data_lock);
2116
2117 mark_all_dquot_dirty(transfer_from);
2118 mark_all_dquot_dirty(transfer_to);
2119 flush_warnings(warn_to);
2120 flush_warnings(warn_from_inodes);
2121 flush_warnings(warn_from_space);
2122 /* Pass back references to put */
2123 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2124 if (is_valid[cnt])
2125 transfer_to[cnt] = transfer_from[cnt];
2126 return 0;
2127 over_quota:
2128 /* Back out changes we already did */
2129 for (cnt--; cnt >= 0; cnt--) {
2130 if (!is_valid[cnt])
2131 continue;
2132 spin_lock(&transfer_to[cnt]->dq_dqb_lock);
2133 dquot_decr_inodes(transfer_to[cnt], inode_usage);
2134 dquot_decr_space(transfer_to[cnt], cur_space);
2135 dquot_free_reserved_space(transfer_to[cnt], rsv_space);
2136 spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2137 }
2138 spin_unlock(&inode->i_lock);
2139 spin_unlock(&dq_data_lock);
2140 flush_warnings(warn_to);
2141 return ret;
2142 }
2143 EXPORT_SYMBOL(__dquot_transfer);
2144
2145 /* Wrapper for transferring ownership of an inode for uid/gid only
2146 * Called from FSXXX_setattr()
2147 */
dquot_transfer(struct inode * inode,struct iattr * iattr)2148 int dquot_transfer(struct inode *inode, struct iattr *iattr)
2149 {
2150 struct dquot *transfer_to[MAXQUOTAS] = {};
2151 struct dquot *dquot;
2152 struct super_block *sb = inode->i_sb;
2153 int ret;
2154
2155 if (!inode_quota_active(inode))
2156 return 0;
2157
2158 if (iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)){
2159 dquot = dqget(sb, make_kqid_uid(iattr->ia_uid));
2160 if (IS_ERR(dquot)) {
2161 if (PTR_ERR(dquot) != -ESRCH) {
2162 ret = PTR_ERR(dquot);
2163 goto out_put;
2164 }
2165 dquot = NULL;
2166 }
2167 transfer_to[USRQUOTA] = dquot;
2168 }
2169 if (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid)){
2170 dquot = dqget(sb, make_kqid_gid(iattr->ia_gid));
2171 if (IS_ERR(dquot)) {
2172 if (PTR_ERR(dquot) != -ESRCH) {
2173 ret = PTR_ERR(dquot);
2174 goto out_put;
2175 }
2176 dquot = NULL;
2177 }
2178 transfer_to[GRPQUOTA] = dquot;
2179 }
2180 ret = __dquot_transfer(inode, transfer_to);
2181 out_put:
2182 dqput_all(transfer_to);
2183 return ret;
2184 }
2185 EXPORT_SYMBOL(dquot_transfer);
2186
2187 /*
2188 * Write info of quota file to disk
2189 */
dquot_commit_info(struct super_block * sb,int type)2190 int dquot_commit_info(struct super_block *sb, int type)
2191 {
2192 struct quota_info *dqopt = sb_dqopt(sb);
2193
2194 return dqopt->ops[type]->write_file_info(sb, type);
2195 }
2196 EXPORT_SYMBOL(dquot_commit_info);
2197
dquot_get_next_id(struct super_block * sb,struct kqid * qid)2198 int dquot_get_next_id(struct super_block *sb, struct kqid *qid)
2199 {
2200 struct quota_info *dqopt = sb_dqopt(sb);
2201
2202 if (!sb_has_quota_active(sb, qid->type))
2203 return -ESRCH;
2204 if (!dqopt->ops[qid->type]->get_next_id)
2205 return -ENOSYS;
2206 return dqopt->ops[qid->type]->get_next_id(sb, qid);
2207 }
2208 EXPORT_SYMBOL(dquot_get_next_id);
2209
2210 /*
2211 * Definitions of diskquota operations.
2212 */
2213 const struct dquot_operations dquot_operations = {
2214 .write_dquot = dquot_commit,
2215 .acquire_dquot = dquot_acquire,
2216 .release_dquot = dquot_release,
2217 .mark_dirty = dquot_mark_dquot_dirty,
2218 .write_info = dquot_commit_info,
2219 .alloc_dquot = dquot_alloc,
2220 .destroy_dquot = dquot_destroy,
2221 .get_next_id = dquot_get_next_id,
2222 };
2223 EXPORT_SYMBOL(dquot_operations);
2224
2225 /*
2226 * Generic helper for ->open on filesystems supporting disk quotas.
2227 */
dquot_file_open(struct inode * inode,struct file * file)2228 int dquot_file_open(struct inode *inode, struct file *file)
2229 {
2230 int error;
2231
2232 error = generic_file_open(inode, file);
2233 if (!error && (file->f_mode & FMODE_WRITE))
2234 error = dquot_initialize(inode);
2235 return error;
2236 }
2237 EXPORT_SYMBOL(dquot_file_open);
2238
vfs_cleanup_quota_inode(struct super_block * sb,int type)2239 static void vfs_cleanup_quota_inode(struct super_block *sb, int type)
2240 {
2241 struct quota_info *dqopt = sb_dqopt(sb);
2242 struct inode *inode = dqopt->files[type];
2243
2244 if (!inode)
2245 return;
2246 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2247 inode_lock(inode);
2248 inode->i_flags &= ~S_NOQUOTA;
2249 inode_unlock(inode);
2250 }
2251 dqopt->files[type] = NULL;
2252 iput(inode);
2253 }
2254
2255 /*
2256 * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
2257 */
dquot_disable(struct super_block * sb,int type,unsigned int flags)2258 int dquot_disable(struct super_block *sb, int type, unsigned int flags)
2259 {
2260 int cnt;
2261 struct quota_info *dqopt = sb_dqopt(sb);
2262
2263 /* s_umount should be held in exclusive mode */
2264 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2265 up_read(&sb->s_umount);
2266
2267 /* Cannot turn off usage accounting without turning off limits, or
2268 * suspend quotas and simultaneously turn quotas off. */
2269 if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
2270 || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
2271 DQUOT_USAGE_ENABLED)))
2272 return -EINVAL;
2273
2274 /*
2275 * Skip everything if there's nothing to do. We have to do this because
2276 * sometimes we are called when fill_super() failed and calling
2277 * sync_fs() in such cases does no good.
2278 */
2279 if (!sb_any_quota_loaded(sb))
2280 return 0;
2281
2282 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2283 if (type != -1 && cnt != type)
2284 continue;
2285 if (!sb_has_quota_loaded(sb, cnt))
2286 continue;
2287
2288 if (flags & DQUOT_SUSPENDED) {
2289 spin_lock(&dq_state_lock);
2290 dqopt->flags |=
2291 dquot_state_flag(DQUOT_SUSPENDED, cnt);
2292 spin_unlock(&dq_state_lock);
2293 } else {
2294 spin_lock(&dq_state_lock);
2295 dqopt->flags &= ~dquot_state_flag(flags, cnt);
2296 /* Turning off suspended quotas? */
2297 if (!sb_has_quota_loaded(sb, cnt) &&
2298 sb_has_quota_suspended(sb, cnt)) {
2299 dqopt->flags &= ~dquot_state_flag(
2300 DQUOT_SUSPENDED, cnt);
2301 spin_unlock(&dq_state_lock);
2302 vfs_cleanup_quota_inode(sb, cnt);
2303 continue;
2304 }
2305 spin_unlock(&dq_state_lock);
2306 }
2307
2308 /* We still have to keep quota loaded? */
2309 if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
2310 continue;
2311
2312 /* Note: these are blocking operations */
2313 drop_dquot_ref(sb, cnt);
2314 invalidate_dquots(sb, cnt);
2315 /*
2316 * Now all dquots should be invalidated, all writes done so we
2317 * should be only users of the info. No locks needed.
2318 */
2319 if (info_dirty(&dqopt->info[cnt]))
2320 sb->dq_op->write_info(sb, cnt);
2321 if (dqopt->ops[cnt]->free_file_info)
2322 dqopt->ops[cnt]->free_file_info(sb, cnt);
2323 put_quota_format(dqopt->info[cnt].dqi_format);
2324 dqopt->info[cnt].dqi_flags = 0;
2325 dqopt->info[cnt].dqi_igrace = 0;
2326 dqopt->info[cnt].dqi_bgrace = 0;
2327 dqopt->ops[cnt] = NULL;
2328 }
2329
2330 /* Skip syncing and setting flags if quota files are hidden */
2331 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2332 goto put_inodes;
2333
2334 /* Sync the superblock so that buffers with quota data are written to
2335 * disk (and so userspace sees correct data afterwards). */
2336 if (sb->s_op->sync_fs)
2337 sb->s_op->sync_fs(sb, 1);
2338 sync_blockdev(sb->s_bdev);
2339 /* Now the quota files are just ordinary files and we can set the
2340 * inode flags back. Moreover we discard the pagecache so that
2341 * userspace sees the writes we did bypassing the pagecache. We
2342 * must also discard the blockdev buffers so that we see the
2343 * changes done by userspace on the next quotaon() */
2344 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2345 if (!sb_has_quota_loaded(sb, cnt) && dqopt->files[cnt]) {
2346 inode_lock(dqopt->files[cnt]);
2347 truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
2348 inode_unlock(dqopt->files[cnt]);
2349 }
2350 if (sb->s_bdev)
2351 invalidate_bdev(sb->s_bdev);
2352 put_inodes:
2353 /* We are done when suspending quotas */
2354 if (flags & DQUOT_SUSPENDED)
2355 return 0;
2356
2357 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2358 if (!sb_has_quota_loaded(sb, cnt))
2359 vfs_cleanup_quota_inode(sb, cnt);
2360 return 0;
2361 }
2362 EXPORT_SYMBOL(dquot_disable);
2363
dquot_quota_off(struct super_block * sb,int type)2364 int dquot_quota_off(struct super_block *sb, int type)
2365 {
2366 return dquot_disable(sb, type,
2367 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2368 }
2369 EXPORT_SYMBOL(dquot_quota_off);
2370
2371 /*
2372 * Turn quotas on on a device
2373 */
2374
vfs_setup_quota_inode(struct inode * inode,int type)2375 static int vfs_setup_quota_inode(struct inode *inode, int type)
2376 {
2377 struct super_block *sb = inode->i_sb;
2378 struct quota_info *dqopt = sb_dqopt(sb);
2379
2380 if (is_bad_inode(inode))
2381 return -EUCLEAN;
2382 if (!S_ISREG(inode->i_mode))
2383 return -EACCES;
2384 if (IS_RDONLY(inode))
2385 return -EROFS;
2386 if (sb_has_quota_loaded(sb, type))
2387 return -EBUSY;
2388
2389 dqopt->files[type] = igrab(inode);
2390 if (!dqopt->files[type])
2391 return -EIO;
2392 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2393 /* We don't want quota and atime on quota files (deadlocks
2394 * possible) Also nobody should write to the file - we use
2395 * special IO operations which ignore the immutable bit. */
2396 inode_lock(inode);
2397 inode->i_flags |= S_NOQUOTA;
2398 inode_unlock(inode);
2399 /*
2400 * When S_NOQUOTA is set, remove dquot references as no more
2401 * references can be added
2402 */
2403 __dquot_drop(inode);
2404 }
2405 return 0;
2406 }
2407
dquot_load_quota_sb(struct super_block * sb,int type,int format_id,unsigned int flags)2408 int dquot_load_quota_sb(struct super_block *sb, int type, int format_id,
2409 unsigned int flags)
2410 {
2411 struct quota_format_type *fmt = find_quota_format(format_id);
2412 struct quota_info *dqopt = sb_dqopt(sb);
2413 int error;
2414
2415 /* Just unsuspend quotas? */
2416 BUG_ON(flags & DQUOT_SUSPENDED);
2417 /* s_umount should be held in exclusive mode */
2418 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2419 up_read(&sb->s_umount);
2420
2421 if (!fmt)
2422 return -ESRCH;
2423 if (!sb->s_op->quota_write || !sb->s_op->quota_read ||
2424 (type == PRJQUOTA && sb->dq_op->get_projid == NULL)) {
2425 error = -EINVAL;
2426 goto out_fmt;
2427 }
2428 /* Filesystems outside of init_user_ns not yet supported */
2429 if (sb->s_user_ns != &init_user_ns) {
2430 error = -EINVAL;
2431 goto out_fmt;
2432 }
2433 /* Usage always has to be set... */
2434 if (!(flags & DQUOT_USAGE_ENABLED)) {
2435 error = -EINVAL;
2436 goto out_fmt;
2437 }
2438 if (sb_has_quota_loaded(sb, type)) {
2439 error = -EBUSY;
2440 goto out_fmt;
2441 }
2442
2443 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2444 /* As we bypass the pagecache we must now flush all the
2445 * dirty data and invalidate caches so that kernel sees
2446 * changes from userspace. It is not enough to just flush
2447 * the quota file since if blocksize < pagesize, invalidation
2448 * of the cache could fail because of other unrelated dirty
2449 * data */
2450 sync_filesystem(sb);
2451 invalidate_bdev(sb->s_bdev);
2452 }
2453
2454 error = -EINVAL;
2455 if (!fmt->qf_ops->check_quota_file(sb, type))
2456 goto out_fmt;
2457
2458 dqopt->ops[type] = fmt->qf_ops;
2459 dqopt->info[type].dqi_format = fmt;
2460 dqopt->info[type].dqi_fmt_id = format_id;
2461 INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2462 error = dqopt->ops[type]->read_file_info(sb, type);
2463 if (error < 0)
2464 goto out_fmt;
2465 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE) {
2466 spin_lock(&dq_data_lock);
2467 dqopt->info[type].dqi_flags |= DQF_SYS_FILE;
2468 spin_unlock(&dq_data_lock);
2469 }
2470 spin_lock(&dq_state_lock);
2471 dqopt->flags |= dquot_state_flag(flags, type);
2472 spin_unlock(&dq_state_lock);
2473
2474 error = add_dquot_ref(sb, type);
2475 if (error)
2476 dquot_disable(sb, type,
2477 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2478
2479 return error;
2480 out_fmt:
2481 put_quota_format(fmt);
2482
2483 return error;
2484 }
2485 EXPORT_SYMBOL(dquot_load_quota_sb);
2486
2487 /*
2488 * More powerful function for turning on quotas on given quota inode allowing
2489 * setting of individual quota flags
2490 */
dquot_load_quota_inode(struct inode * inode,int type,int format_id,unsigned int flags)2491 int dquot_load_quota_inode(struct inode *inode, int type, int format_id,
2492 unsigned int flags)
2493 {
2494 int err;
2495
2496 err = vfs_setup_quota_inode(inode, type);
2497 if (err < 0)
2498 return err;
2499 err = dquot_load_quota_sb(inode->i_sb, type, format_id, flags);
2500 if (err < 0)
2501 vfs_cleanup_quota_inode(inode->i_sb, type);
2502 return err;
2503 }
2504 EXPORT_SYMBOL(dquot_load_quota_inode);
2505
2506 /* Reenable quotas on remount RW */
dquot_resume(struct super_block * sb,int type)2507 int dquot_resume(struct super_block *sb, int type)
2508 {
2509 struct quota_info *dqopt = sb_dqopt(sb);
2510 int ret = 0, cnt;
2511 unsigned int flags;
2512
2513 /* s_umount should be held in exclusive mode */
2514 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2515 up_read(&sb->s_umount);
2516
2517 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2518 if (type != -1 && cnt != type)
2519 continue;
2520 if (!sb_has_quota_suspended(sb, cnt))
2521 continue;
2522
2523 spin_lock(&dq_state_lock);
2524 flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2525 DQUOT_LIMITS_ENABLED,
2526 cnt);
2527 dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2528 spin_unlock(&dq_state_lock);
2529
2530 flags = dquot_generic_flag(flags, cnt);
2531 ret = dquot_load_quota_sb(sb, cnt, dqopt->info[cnt].dqi_fmt_id,
2532 flags);
2533 if (ret < 0)
2534 vfs_cleanup_quota_inode(sb, cnt);
2535 }
2536
2537 return ret;
2538 }
2539 EXPORT_SYMBOL(dquot_resume);
2540
dquot_quota_on(struct super_block * sb,int type,int format_id,const struct path * path)2541 int dquot_quota_on(struct super_block *sb, int type, int format_id,
2542 const struct path *path)
2543 {
2544 int error = security_quota_on(path->dentry);
2545 if (error)
2546 return error;
2547 /* Quota file not on the same filesystem? */
2548 if (path->dentry->d_sb != sb)
2549 error = -EXDEV;
2550 else
2551 error = dquot_load_quota_inode(d_inode(path->dentry), type,
2552 format_id, DQUOT_USAGE_ENABLED |
2553 DQUOT_LIMITS_ENABLED);
2554 return error;
2555 }
2556 EXPORT_SYMBOL(dquot_quota_on);
2557
2558 /*
2559 * This function is used when filesystem needs to initialize quotas
2560 * during mount time.
2561 */
dquot_quota_on_mount(struct super_block * sb,char * qf_name,int format_id,int type)2562 int dquot_quota_on_mount(struct super_block *sb, char *qf_name,
2563 int format_id, int type)
2564 {
2565 struct dentry *dentry;
2566 int error;
2567
2568 dentry = lookup_positive_unlocked(qf_name, sb->s_root, strlen(qf_name));
2569 if (IS_ERR(dentry))
2570 return PTR_ERR(dentry);
2571
2572 error = security_quota_on(dentry);
2573 if (!error)
2574 error = dquot_load_quota_inode(d_inode(dentry), type, format_id,
2575 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2576
2577 dput(dentry);
2578 return error;
2579 }
2580 EXPORT_SYMBOL(dquot_quota_on_mount);
2581
dquot_quota_enable(struct super_block * sb,unsigned int flags)2582 static int dquot_quota_enable(struct super_block *sb, unsigned int flags)
2583 {
2584 int ret;
2585 int type;
2586 struct quota_info *dqopt = sb_dqopt(sb);
2587
2588 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2589 return -ENOSYS;
2590 /* Accounting cannot be turned on while fs is mounted */
2591 flags &= ~(FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT);
2592 if (!flags)
2593 return -EINVAL;
2594 for (type = 0; type < MAXQUOTAS; type++) {
2595 if (!(flags & qtype_enforce_flag(type)))
2596 continue;
2597 /* Can't enforce without accounting */
2598 if (!sb_has_quota_usage_enabled(sb, type)) {
2599 ret = -EINVAL;
2600 goto out_err;
2601 }
2602 if (sb_has_quota_limits_enabled(sb, type)) {
2603 ret = -EBUSY;
2604 goto out_err;
2605 }
2606 spin_lock(&dq_state_lock);
2607 dqopt->flags |= dquot_state_flag(DQUOT_LIMITS_ENABLED, type);
2608 spin_unlock(&dq_state_lock);
2609 }
2610 return 0;
2611 out_err:
2612 /* Backout enforcement enablement we already did */
2613 for (type--; type >= 0; type--) {
2614 if (flags & qtype_enforce_flag(type))
2615 dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2616 }
2617 /* Error code translation for better compatibility with XFS */
2618 if (ret == -EBUSY)
2619 ret = -EEXIST;
2620 return ret;
2621 }
2622
dquot_quota_disable(struct super_block * sb,unsigned int flags)2623 static int dquot_quota_disable(struct super_block *sb, unsigned int flags)
2624 {
2625 int ret;
2626 int type;
2627 struct quota_info *dqopt = sb_dqopt(sb);
2628
2629 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2630 return -ENOSYS;
2631 /*
2632 * We don't support turning off accounting via quotactl. In principle
2633 * quota infrastructure can do this but filesystems don't expect
2634 * userspace to be able to do it.
2635 */
2636 if (flags &
2637 (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT))
2638 return -EOPNOTSUPP;
2639
2640 /* Filter out limits not enabled */
2641 for (type = 0; type < MAXQUOTAS; type++)
2642 if (!sb_has_quota_limits_enabled(sb, type))
2643 flags &= ~qtype_enforce_flag(type);
2644 /* Nothing left? */
2645 if (!flags)
2646 return -EEXIST;
2647 for (type = 0; type < MAXQUOTAS; type++) {
2648 if (flags & qtype_enforce_flag(type)) {
2649 ret = dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2650 if (ret < 0)
2651 goto out_err;
2652 }
2653 }
2654 return 0;
2655 out_err:
2656 /* Backout enforcement disabling we already did */
2657 for (type--; type >= 0; type--) {
2658 if (flags & qtype_enforce_flag(type)) {
2659 spin_lock(&dq_state_lock);
2660 dqopt->flags |=
2661 dquot_state_flag(DQUOT_LIMITS_ENABLED, type);
2662 spin_unlock(&dq_state_lock);
2663 }
2664 }
2665 return ret;
2666 }
2667
2668 /* Generic routine for getting common part of quota structure */
do_get_dqblk(struct dquot * dquot,struct qc_dqblk * di)2669 static void do_get_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2670 {
2671 struct mem_dqblk *dm = &dquot->dq_dqb;
2672
2673 memset(di, 0, sizeof(*di));
2674 spin_lock(&dquot->dq_dqb_lock);
2675 di->d_spc_hardlimit = dm->dqb_bhardlimit;
2676 di->d_spc_softlimit = dm->dqb_bsoftlimit;
2677 di->d_ino_hardlimit = dm->dqb_ihardlimit;
2678 di->d_ino_softlimit = dm->dqb_isoftlimit;
2679 di->d_space = dm->dqb_curspace + dm->dqb_rsvspace;
2680 di->d_ino_count = dm->dqb_curinodes;
2681 di->d_spc_timer = dm->dqb_btime;
2682 di->d_ino_timer = dm->dqb_itime;
2683 spin_unlock(&dquot->dq_dqb_lock);
2684 }
2685
dquot_get_dqblk(struct super_block * sb,struct kqid qid,struct qc_dqblk * di)2686 int dquot_get_dqblk(struct super_block *sb, struct kqid qid,
2687 struct qc_dqblk *di)
2688 {
2689 struct dquot *dquot;
2690
2691 dquot = dqget(sb, qid);
2692 if (IS_ERR(dquot))
2693 return PTR_ERR(dquot);
2694 do_get_dqblk(dquot, di);
2695 dqput(dquot);
2696
2697 return 0;
2698 }
2699 EXPORT_SYMBOL(dquot_get_dqblk);
2700
dquot_get_next_dqblk(struct super_block * sb,struct kqid * qid,struct qc_dqblk * di)2701 int dquot_get_next_dqblk(struct super_block *sb, struct kqid *qid,
2702 struct qc_dqblk *di)
2703 {
2704 struct dquot *dquot;
2705 int err;
2706
2707 if (!sb->dq_op->get_next_id)
2708 return -ENOSYS;
2709 err = sb->dq_op->get_next_id(sb, qid);
2710 if (err < 0)
2711 return err;
2712 dquot = dqget(sb, *qid);
2713 if (IS_ERR(dquot))
2714 return PTR_ERR(dquot);
2715 do_get_dqblk(dquot, di);
2716 dqput(dquot);
2717
2718 return 0;
2719 }
2720 EXPORT_SYMBOL(dquot_get_next_dqblk);
2721
2722 #define VFS_QC_MASK \
2723 (QC_SPACE | QC_SPC_SOFT | QC_SPC_HARD | \
2724 QC_INO_COUNT | QC_INO_SOFT | QC_INO_HARD | \
2725 QC_SPC_TIMER | QC_INO_TIMER)
2726
2727 /* Generic routine for setting common part of quota structure */
do_set_dqblk(struct dquot * dquot,struct qc_dqblk * di)2728 static int do_set_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2729 {
2730 struct mem_dqblk *dm = &dquot->dq_dqb;
2731 int check_blim = 0, check_ilim = 0;
2732 struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
2733
2734 if (di->d_fieldmask & ~VFS_QC_MASK)
2735 return -EINVAL;
2736
2737 if (((di->d_fieldmask & QC_SPC_SOFT) &&
2738 di->d_spc_softlimit > dqi->dqi_max_spc_limit) ||
2739 ((di->d_fieldmask & QC_SPC_HARD) &&
2740 di->d_spc_hardlimit > dqi->dqi_max_spc_limit) ||
2741 ((di->d_fieldmask & QC_INO_SOFT) &&
2742 (di->d_ino_softlimit > dqi->dqi_max_ino_limit)) ||
2743 ((di->d_fieldmask & QC_INO_HARD) &&
2744 (di->d_ino_hardlimit > dqi->dqi_max_ino_limit)))
2745 return -ERANGE;
2746
2747 spin_lock(&dquot->dq_dqb_lock);
2748 if (di->d_fieldmask & QC_SPACE) {
2749 dm->dqb_curspace = di->d_space - dm->dqb_rsvspace;
2750 check_blim = 1;
2751 set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2752 }
2753
2754 if (di->d_fieldmask & QC_SPC_SOFT)
2755 dm->dqb_bsoftlimit = di->d_spc_softlimit;
2756 if (di->d_fieldmask & QC_SPC_HARD)
2757 dm->dqb_bhardlimit = di->d_spc_hardlimit;
2758 if (di->d_fieldmask & (QC_SPC_SOFT | QC_SPC_HARD)) {
2759 check_blim = 1;
2760 set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2761 }
2762
2763 if (di->d_fieldmask & QC_INO_COUNT) {
2764 dm->dqb_curinodes = di->d_ino_count;
2765 check_ilim = 1;
2766 set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2767 }
2768
2769 if (di->d_fieldmask & QC_INO_SOFT)
2770 dm->dqb_isoftlimit = di->d_ino_softlimit;
2771 if (di->d_fieldmask & QC_INO_HARD)
2772 dm->dqb_ihardlimit = di->d_ino_hardlimit;
2773 if (di->d_fieldmask & (QC_INO_SOFT | QC_INO_HARD)) {
2774 check_ilim = 1;
2775 set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2776 }
2777
2778 if (di->d_fieldmask & QC_SPC_TIMER) {
2779 dm->dqb_btime = di->d_spc_timer;
2780 check_blim = 1;
2781 set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2782 }
2783
2784 if (di->d_fieldmask & QC_INO_TIMER) {
2785 dm->dqb_itime = di->d_ino_timer;
2786 check_ilim = 1;
2787 set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2788 }
2789
2790 if (check_blim) {
2791 if (!dm->dqb_bsoftlimit ||
2792 dm->dqb_curspace + dm->dqb_rsvspace <= dm->dqb_bsoftlimit) {
2793 dm->dqb_btime = 0;
2794 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2795 } else if (!(di->d_fieldmask & QC_SPC_TIMER))
2796 /* Set grace only if user hasn't provided his own... */
2797 dm->dqb_btime = ktime_get_real_seconds() + dqi->dqi_bgrace;
2798 }
2799 if (check_ilim) {
2800 if (!dm->dqb_isoftlimit ||
2801 dm->dqb_curinodes <= dm->dqb_isoftlimit) {
2802 dm->dqb_itime = 0;
2803 clear_bit(DQ_INODES_B, &dquot->dq_flags);
2804 } else if (!(di->d_fieldmask & QC_INO_TIMER))
2805 /* Set grace only if user hasn't provided his own... */
2806 dm->dqb_itime = ktime_get_real_seconds() + dqi->dqi_igrace;
2807 }
2808 if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2809 dm->dqb_isoftlimit)
2810 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2811 else
2812 set_bit(DQ_FAKE_B, &dquot->dq_flags);
2813 spin_unlock(&dquot->dq_dqb_lock);
2814 mark_dquot_dirty(dquot);
2815
2816 return 0;
2817 }
2818
dquot_set_dqblk(struct super_block * sb,struct kqid qid,struct qc_dqblk * di)2819 int dquot_set_dqblk(struct super_block *sb, struct kqid qid,
2820 struct qc_dqblk *di)
2821 {
2822 struct dquot *dquot;
2823 int rc;
2824
2825 dquot = dqget(sb, qid);
2826 if (IS_ERR(dquot)) {
2827 rc = PTR_ERR(dquot);
2828 goto out;
2829 }
2830 rc = do_set_dqblk(dquot, di);
2831 dqput(dquot);
2832 out:
2833 return rc;
2834 }
2835 EXPORT_SYMBOL(dquot_set_dqblk);
2836
2837 /* Generic routine for getting common part of quota file information */
dquot_get_state(struct super_block * sb,struct qc_state * state)2838 int dquot_get_state(struct super_block *sb, struct qc_state *state)
2839 {
2840 struct mem_dqinfo *mi;
2841 struct qc_type_state *tstate;
2842 struct quota_info *dqopt = sb_dqopt(sb);
2843 int type;
2844
2845 memset(state, 0, sizeof(*state));
2846 for (type = 0; type < MAXQUOTAS; type++) {
2847 if (!sb_has_quota_active(sb, type))
2848 continue;
2849 tstate = state->s_state + type;
2850 mi = sb_dqopt(sb)->info + type;
2851 tstate->flags = QCI_ACCT_ENABLED;
2852 spin_lock(&dq_data_lock);
2853 if (mi->dqi_flags & DQF_SYS_FILE)
2854 tstate->flags |= QCI_SYSFILE;
2855 if (mi->dqi_flags & DQF_ROOT_SQUASH)
2856 tstate->flags |= QCI_ROOT_SQUASH;
2857 if (sb_has_quota_limits_enabled(sb, type))
2858 tstate->flags |= QCI_LIMITS_ENFORCED;
2859 tstate->spc_timelimit = mi->dqi_bgrace;
2860 tstate->ino_timelimit = mi->dqi_igrace;
2861 if (dqopt->files[type]) {
2862 tstate->ino = dqopt->files[type]->i_ino;
2863 tstate->blocks = dqopt->files[type]->i_blocks;
2864 }
2865 tstate->nextents = 1; /* We don't know... */
2866 spin_unlock(&dq_data_lock);
2867 }
2868 return 0;
2869 }
2870 EXPORT_SYMBOL(dquot_get_state);
2871
2872 /* Generic routine for setting common part of quota file information */
dquot_set_dqinfo(struct super_block * sb,int type,struct qc_info * ii)2873 int dquot_set_dqinfo(struct super_block *sb, int type, struct qc_info *ii)
2874 {
2875 struct mem_dqinfo *mi;
2876 int err = 0;
2877
2878 if ((ii->i_fieldmask & QC_WARNS_MASK) ||
2879 (ii->i_fieldmask & QC_RT_SPC_TIMER))
2880 return -EINVAL;
2881 if (!sb_has_quota_active(sb, type))
2882 return -ESRCH;
2883 mi = sb_dqopt(sb)->info + type;
2884 if (ii->i_fieldmask & QC_FLAGS) {
2885 if ((ii->i_flags & QCI_ROOT_SQUASH &&
2886 mi->dqi_format->qf_fmt_id != QFMT_VFS_OLD))
2887 return -EINVAL;
2888 }
2889 spin_lock(&dq_data_lock);
2890 if (ii->i_fieldmask & QC_SPC_TIMER)
2891 mi->dqi_bgrace = ii->i_spc_timelimit;
2892 if (ii->i_fieldmask & QC_INO_TIMER)
2893 mi->dqi_igrace = ii->i_ino_timelimit;
2894 if (ii->i_fieldmask & QC_FLAGS) {
2895 if (ii->i_flags & QCI_ROOT_SQUASH)
2896 mi->dqi_flags |= DQF_ROOT_SQUASH;
2897 else
2898 mi->dqi_flags &= ~DQF_ROOT_SQUASH;
2899 }
2900 spin_unlock(&dq_data_lock);
2901 mark_info_dirty(sb, type);
2902 /* Force write to disk */
2903 sb->dq_op->write_info(sb, type);
2904 return err;
2905 }
2906 EXPORT_SYMBOL(dquot_set_dqinfo);
2907
2908 const struct quotactl_ops dquot_quotactl_sysfile_ops = {
2909 .quota_enable = dquot_quota_enable,
2910 .quota_disable = dquot_quota_disable,
2911 .quota_sync = dquot_quota_sync,
2912 .get_state = dquot_get_state,
2913 .set_info = dquot_set_dqinfo,
2914 .get_dqblk = dquot_get_dqblk,
2915 .get_nextdqblk = dquot_get_next_dqblk,
2916 .set_dqblk = dquot_set_dqblk
2917 };
2918 EXPORT_SYMBOL(dquot_quotactl_sysfile_ops);
2919
do_proc_dqstats(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)2920 static int do_proc_dqstats(struct ctl_table *table, int write,
2921 void *buffer, size_t *lenp, loff_t *ppos)
2922 {
2923 unsigned int type = (unsigned long *)table->data - dqstats.stat;
2924 s64 value = percpu_counter_sum(&dqstats.counter[type]);
2925
2926 /* Filter negative values for non-monotonic counters */
2927 if (value < 0 && (type == DQST_ALLOC_DQUOTS ||
2928 type == DQST_FREE_DQUOTS))
2929 value = 0;
2930
2931 /* Update global table */
2932 dqstats.stat[type] = value;
2933 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
2934 }
2935
2936 static struct ctl_table fs_dqstats_table[] = {
2937 {
2938 .procname = "lookups",
2939 .data = &dqstats.stat[DQST_LOOKUPS],
2940 .maxlen = sizeof(unsigned long),
2941 .mode = 0444,
2942 .proc_handler = do_proc_dqstats,
2943 },
2944 {
2945 .procname = "drops",
2946 .data = &dqstats.stat[DQST_DROPS],
2947 .maxlen = sizeof(unsigned long),
2948 .mode = 0444,
2949 .proc_handler = do_proc_dqstats,
2950 },
2951 {
2952 .procname = "reads",
2953 .data = &dqstats.stat[DQST_READS],
2954 .maxlen = sizeof(unsigned long),
2955 .mode = 0444,
2956 .proc_handler = do_proc_dqstats,
2957 },
2958 {
2959 .procname = "writes",
2960 .data = &dqstats.stat[DQST_WRITES],
2961 .maxlen = sizeof(unsigned long),
2962 .mode = 0444,
2963 .proc_handler = do_proc_dqstats,
2964 },
2965 {
2966 .procname = "cache_hits",
2967 .data = &dqstats.stat[DQST_CACHE_HITS],
2968 .maxlen = sizeof(unsigned long),
2969 .mode = 0444,
2970 .proc_handler = do_proc_dqstats,
2971 },
2972 {
2973 .procname = "allocated_dquots",
2974 .data = &dqstats.stat[DQST_ALLOC_DQUOTS],
2975 .maxlen = sizeof(unsigned long),
2976 .mode = 0444,
2977 .proc_handler = do_proc_dqstats,
2978 },
2979 {
2980 .procname = "free_dquots",
2981 .data = &dqstats.stat[DQST_FREE_DQUOTS],
2982 .maxlen = sizeof(unsigned long),
2983 .mode = 0444,
2984 .proc_handler = do_proc_dqstats,
2985 },
2986 {
2987 .procname = "syncs",
2988 .data = &dqstats.stat[DQST_SYNCS],
2989 .maxlen = sizeof(unsigned long),
2990 .mode = 0444,
2991 .proc_handler = do_proc_dqstats,
2992 },
2993 #ifdef CONFIG_PRINT_QUOTA_WARNING
2994 {
2995 .procname = "warnings",
2996 .data = &flag_print_warnings,
2997 .maxlen = sizeof(int),
2998 .mode = 0644,
2999 .proc_handler = proc_dointvec,
3000 },
3001 #endif
3002 { },
3003 };
3004
3005 static struct ctl_table fs_table[] = {
3006 {
3007 .procname = "quota",
3008 .mode = 0555,
3009 .child = fs_dqstats_table,
3010 },
3011 { },
3012 };
3013
3014 static struct ctl_table sys_table[] = {
3015 {
3016 .procname = "fs",
3017 .mode = 0555,
3018 .child = fs_table,
3019 },
3020 { },
3021 };
3022
dquot_init(void)3023 static int __init dquot_init(void)
3024 {
3025 int i, ret;
3026 unsigned long nr_hash, order;
3027
3028 printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
3029
3030 register_sysctl_table(sys_table);
3031
3032 dquot_cachep = kmem_cache_create("dquot",
3033 sizeof(struct dquot), sizeof(unsigned long) * 4,
3034 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
3035 SLAB_MEM_SPREAD|SLAB_PANIC),
3036 NULL);
3037
3038 order = 0;
3039 dquot_hash = (struct hlist_head *)__get_free_pages(GFP_KERNEL, order);
3040 if (!dquot_hash)
3041 panic("Cannot create dquot hash table");
3042
3043 for (i = 0; i < _DQST_DQSTAT_LAST; i++) {
3044 ret = percpu_counter_init(&dqstats.counter[i], 0, GFP_KERNEL);
3045 if (ret)
3046 panic("Cannot create dquot stat counters");
3047 }
3048
3049 /* Find power-of-two hlist_heads which can fit into allocation */
3050 nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
3051 dq_hash_bits = ilog2(nr_hash);
3052
3053 nr_hash = 1UL << dq_hash_bits;
3054 dq_hash_mask = nr_hash - 1;
3055 for (i = 0; i < nr_hash; i++)
3056 INIT_HLIST_HEAD(dquot_hash + i);
3057
3058 pr_info("VFS: Dquot-cache hash table entries: %ld (order %ld,"
3059 " %ld bytes)\n", nr_hash, order, (PAGE_SIZE << order));
3060
3061 if (register_shrinker(&dqcache_shrinker))
3062 panic("Cannot register dquot shrinker");
3063
3064 return 0;
3065 }
3066 fs_initcall(dquot_init);
3067