xref: /openbmc/linux/fs/quota/dquot.c (revision f7d84fa7)
1 /*
2  * Implementation of the diskquota system for the LINUX operating system. QUOTA
3  * is implemented using the BSD system call interface as the means of
4  * communication with the user level. This file contains the generic routines
5  * called by the different filesystems on allocation of an inode or block.
6  * These routines take care of the administration needed to have a consistent
7  * diskquota tracking system. The ideas of both user and group quotas are based
8  * on the Melbourne quota system as used on BSD derived systems. The internal
9  * implementation is based on one of the several variants of the LINUX
10  * inode-subsystem with added complexity of the diskquota system.
11  *
12  * Author:	Marco van Wieringen <mvw@planets.elm.net>
13  *
14  * Fixes:   Dmitry Gorodchanin <pgmdsg@ibi.com>, 11 Feb 96
15  *
16  *		Revised list management to avoid races
17  *		-- Bill Hawes, <whawes@star.net>, 9/98
18  *
19  *		Fixed races in dquot_transfer(), dqget() and dquot_alloc_...().
20  *		As the consequence the locking was moved from dquot_decr_...(),
21  *		dquot_incr_...() to calling functions.
22  *		invalidate_dquots() now writes modified dquots.
23  *		Serialized quota_off() and quota_on() for mount point.
24  *		Fixed a few bugs in grow_dquots().
25  *		Fixed deadlock in write_dquot() - we no longer account quotas on
26  *		quota files
27  *		remove_dquot_ref() moved to inode.c - it now traverses through inodes
28  *		add_dquot_ref() restarts after blocking
29  *		Added check for bogus uid and fixed check for group in quotactl.
30  *		Jan Kara, <jack@suse.cz>, sponsored by SuSE CR, 10-11/99
31  *
32  *		Used struct list_head instead of own list struct
33  *		Invalidation of referenced dquots is no longer possible
34  *		Improved free_dquots list management
35  *		Quota and i_blocks are now updated in one place to avoid races
36  *		Warnings are now delayed so we won't block in critical section
37  *		Write updated not to require dquot lock
38  *		Jan Kara, <jack@suse.cz>, 9/2000
39  *
40  *		Added dynamic quota structure allocation
41  *		Jan Kara <jack@suse.cz> 12/2000
42  *
43  *		Rewritten quota interface. Implemented new quota format and
44  *		formats registering.
45  *		Jan Kara, <jack@suse.cz>, 2001,2002
46  *
47  *		New SMP locking.
48  *		Jan Kara, <jack@suse.cz>, 10/2002
49  *
50  *		Added journalled quota support, fix lock inversion problems
51  *		Jan Kara, <jack@suse.cz>, 2003,2004
52  *
53  * (C) Copyright 1994 - 1997 Marco van Wieringen
54  */
55 
56 #include <linux/errno.h>
57 #include <linux/kernel.h>
58 #include <linux/fs.h>
59 #include <linux/mount.h>
60 #include <linux/mm.h>
61 #include <linux/time.h>
62 #include <linux/types.h>
63 #include <linux/string.h>
64 #include <linux/fcntl.h>
65 #include <linux/stat.h>
66 #include <linux/tty.h>
67 #include <linux/file.h>
68 #include <linux/slab.h>
69 #include <linux/sysctl.h>
70 #include <linux/init.h>
71 #include <linux/module.h>
72 #include <linux/proc_fs.h>
73 #include <linux/security.h>
74 #include <linux/sched.h>
75 #include <linux/cred.h>
76 #include <linux/kmod.h>
77 #include <linux/namei.h>
78 #include <linux/capability.h>
79 #include <linux/quotaops.h>
80 #include "../internal.h" /* ugh */
81 
82 #include <linux/uaccess.h>
83 
84 /*
85  * There are three quota SMP locks. dq_list_lock protects all lists with quotas
86  * and quota formats.
87  * dq_data_lock protects data from dq_dqb and also mem_dqinfo structures and
88  * also guards consistency of dquot->dq_dqb with inode->i_blocks, i_bytes.
89  * i_blocks and i_bytes updates itself are guarded by i_lock acquired directly
90  * in inode_add_bytes() and inode_sub_bytes(). dq_state_lock protects
91  * modifications of quota state (on quotaon and quotaoff) and readers who care
92  * about latest values take it as well.
93  *
94  * The spinlock ordering is hence: dq_data_lock > dq_list_lock > i_lock,
95  *   dq_list_lock > dq_state_lock
96  *
97  * Note that some things (eg. sb pointer, type, id) doesn't change during
98  * the life of the dquot structure and so needn't to be protected by a lock
99  *
100  * Operation accessing dquots via inode pointers are protected by dquot_srcu.
101  * Operation of reading pointer needs srcu_read_lock(&dquot_srcu), and
102  * synchronize_srcu(&dquot_srcu) is called after clearing pointers from
103  * inode and before dropping dquot references to avoid use of dquots after
104  * they are freed. dq_data_lock is used to serialize the pointer setting and
105  * clearing operations.
106  * Special care needs to be taken about S_NOQUOTA inode flag (marking that
107  * inode is a quota file). Functions adding pointers from inode to dquots have
108  * to check this flag under dq_data_lock and then (if S_NOQUOTA is not set) they
109  * have to do all pointer modifications before dropping dq_data_lock. This makes
110  * sure they cannot race with quotaon which first sets S_NOQUOTA flag and
111  * then drops all pointers to dquots from an inode.
112  *
113  * Each dquot has its dq_lock mutex. Locked dquots might not be referenced
114  * from inodes (dquot_alloc_space() and such don't check the dq_lock).
115  * Currently dquot is locked only when it is being read to memory (or space for
116  * it is being allocated) on the first dqget() and when it is being released on
117  * the last dqput(). The allocation and release oparations are serialized by
118  * the dq_lock and by checking the use count in dquot_release().  Write
119  * operations on dquots don't hold dq_lock as they copy data under dq_data_lock
120  * spinlock to internal buffers before writing.
121  *
122  * Lock ordering (including related VFS locks) is the following:
123  *   s_umount > i_mutex > journal_lock > dquot->dq_lock > dqio_mutex
124  */
125 
126 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_list_lock);
127 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_state_lock);
128 __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_data_lock);
129 EXPORT_SYMBOL(dq_data_lock);
130 DEFINE_STATIC_SRCU(dquot_srcu);
131 
132 void __quota_error(struct super_block *sb, const char *func,
133 		   const char *fmt, ...)
134 {
135 	if (printk_ratelimit()) {
136 		va_list args;
137 		struct va_format vaf;
138 
139 		va_start(args, fmt);
140 
141 		vaf.fmt = fmt;
142 		vaf.va = &args;
143 
144 		printk(KERN_ERR "Quota error (device %s): %s: %pV\n",
145 		       sb->s_id, func, &vaf);
146 
147 		va_end(args);
148 	}
149 }
150 EXPORT_SYMBOL(__quota_error);
151 
152 #if defined(CONFIG_QUOTA_DEBUG) || defined(CONFIG_PRINT_QUOTA_WARNING)
153 static char *quotatypes[] = INITQFNAMES;
154 #endif
155 static struct quota_format_type *quota_formats;	/* List of registered formats */
156 static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
157 
158 /* SLAB cache for dquot structures */
159 static struct kmem_cache *dquot_cachep;
160 
161 int register_quota_format(struct quota_format_type *fmt)
162 {
163 	spin_lock(&dq_list_lock);
164 	fmt->qf_next = quota_formats;
165 	quota_formats = fmt;
166 	spin_unlock(&dq_list_lock);
167 	return 0;
168 }
169 EXPORT_SYMBOL(register_quota_format);
170 
171 void unregister_quota_format(struct quota_format_type *fmt)
172 {
173 	struct quota_format_type **actqf;
174 
175 	spin_lock(&dq_list_lock);
176 	for (actqf = &quota_formats; *actqf && *actqf != fmt;
177 	     actqf = &(*actqf)->qf_next)
178 		;
179 	if (*actqf)
180 		*actqf = (*actqf)->qf_next;
181 	spin_unlock(&dq_list_lock);
182 }
183 EXPORT_SYMBOL(unregister_quota_format);
184 
185 static struct quota_format_type *find_quota_format(int id)
186 {
187 	struct quota_format_type *actqf;
188 
189 	spin_lock(&dq_list_lock);
190 	for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
191 	     actqf = actqf->qf_next)
192 		;
193 	if (!actqf || !try_module_get(actqf->qf_owner)) {
194 		int qm;
195 
196 		spin_unlock(&dq_list_lock);
197 
198 		for (qm = 0; module_names[qm].qm_fmt_id &&
199 			     module_names[qm].qm_fmt_id != id; qm++)
200 			;
201 		if (!module_names[qm].qm_fmt_id ||
202 		    request_module(module_names[qm].qm_mod_name))
203 			return NULL;
204 
205 		spin_lock(&dq_list_lock);
206 		for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
207 		     actqf = actqf->qf_next)
208 			;
209 		if (actqf && !try_module_get(actqf->qf_owner))
210 			actqf = NULL;
211 	}
212 	spin_unlock(&dq_list_lock);
213 	return actqf;
214 }
215 
216 static void put_quota_format(struct quota_format_type *fmt)
217 {
218 	module_put(fmt->qf_owner);
219 }
220 
221 /*
222  * Dquot List Management:
223  * The quota code uses three lists for dquot management: the inuse_list,
224  * free_dquots, and dquot_hash[] array. A single dquot structure may be
225  * on all three lists, depending on its current state.
226  *
227  * All dquots are placed to the end of inuse_list when first created, and this
228  * list is used for invalidate operation, which must look at every dquot.
229  *
230  * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
231  * and this list is searched whenever we need an available dquot.  Dquots are
232  * removed from the list as soon as they are used again, and
233  * dqstats.free_dquots gives the number of dquots on the list. When
234  * dquot is invalidated it's completely released from memory.
235  *
236  * Dquots with a specific identity (device, type and id) are placed on
237  * one of the dquot_hash[] hash chains. The provides an efficient search
238  * mechanism to locate a specific dquot.
239  */
240 
241 static LIST_HEAD(inuse_list);
242 static LIST_HEAD(free_dquots);
243 static unsigned int dq_hash_bits, dq_hash_mask;
244 static struct hlist_head *dquot_hash;
245 
246 struct dqstats dqstats;
247 EXPORT_SYMBOL(dqstats);
248 
249 static qsize_t inode_get_rsv_space(struct inode *inode);
250 static int __dquot_initialize(struct inode *inode, int type);
251 
252 static inline unsigned int
253 hashfn(const struct super_block *sb, struct kqid qid)
254 {
255 	unsigned int id = from_kqid(&init_user_ns, qid);
256 	int type = qid.type;
257 	unsigned long tmp;
258 
259 	tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
260 	return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
261 }
262 
263 /*
264  * Following list functions expect dq_list_lock to be held
265  */
266 static inline void insert_dquot_hash(struct dquot *dquot)
267 {
268 	struct hlist_head *head;
269 	head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id);
270 	hlist_add_head(&dquot->dq_hash, head);
271 }
272 
273 static inline void remove_dquot_hash(struct dquot *dquot)
274 {
275 	hlist_del_init(&dquot->dq_hash);
276 }
277 
278 static struct dquot *find_dquot(unsigned int hashent, struct super_block *sb,
279 				struct kqid qid)
280 {
281 	struct hlist_node *node;
282 	struct dquot *dquot;
283 
284 	hlist_for_each (node, dquot_hash+hashent) {
285 		dquot = hlist_entry(node, struct dquot, dq_hash);
286 		if (dquot->dq_sb == sb && qid_eq(dquot->dq_id, qid))
287 			return dquot;
288 	}
289 	return NULL;
290 }
291 
292 /* Add a dquot to the tail of the free list */
293 static inline void put_dquot_last(struct dquot *dquot)
294 {
295 	list_add_tail(&dquot->dq_free, &free_dquots);
296 	dqstats_inc(DQST_FREE_DQUOTS);
297 }
298 
299 static inline void remove_free_dquot(struct dquot *dquot)
300 {
301 	if (list_empty(&dquot->dq_free))
302 		return;
303 	list_del_init(&dquot->dq_free);
304 	dqstats_dec(DQST_FREE_DQUOTS);
305 }
306 
307 static inline void put_inuse(struct dquot *dquot)
308 {
309 	/* We add to the back of inuse list so we don't have to restart
310 	 * when traversing this list and we block */
311 	list_add_tail(&dquot->dq_inuse, &inuse_list);
312 	dqstats_inc(DQST_ALLOC_DQUOTS);
313 }
314 
315 static inline void remove_inuse(struct dquot *dquot)
316 {
317 	dqstats_dec(DQST_ALLOC_DQUOTS);
318 	list_del(&dquot->dq_inuse);
319 }
320 /*
321  * End of list functions needing dq_list_lock
322  */
323 
324 static void wait_on_dquot(struct dquot *dquot)
325 {
326 	mutex_lock(&dquot->dq_lock);
327 	mutex_unlock(&dquot->dq_lock);
328 }
329 
330 static inline int dquot_dirty(struct dquot *dquot)
331 {
332 	return test_bit(DQ_MOD_B, &dquot->dq_flags);
333 }
334 
335 static inline int mark_dquot_dirty(struct dquot *dquot)
336 {
337 	return dquot->dq_sb->dq_op->mark_dirty(dquot);
338 }
339 
340 /* Mark dquot dirty in atomic manner, and return it's old dirty flag state */
341 int dquot_mark_dquot_dirty(struct dquot *dquot)
342 {
343 	int ret = 1;
344 
345 	/* If quota is dirty already, we don't have to acquire dq_list_lock */
346 	if (test_bit(DQ_MOD_B, &dquot->dq_flags))
347 		return 1;
348 
349 	spin_lock(&dq_list_lock);
350 	if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags)) {
351 		list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
352 				info[dquot->dq_id.type].dqi_dirty_list);
353 		ret = 0;
354 	}
355 	spin_unlock(&dq_list_lock);
356 	return ret;
357 }
358 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
359 
360 /* Dirtify all the dquots - this can block when journalling */
361 static inline int mark_all_dquot_dirty(struct dquot * const *dquot)
362 {
363 	int ret, err, cnt;
364 
365 	ret = err = 0;
366 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
367 		if (dquot[cnt])
368 			/* Even in case of error we have to continue */
369 			ret = mark_dquot_dirty(dquot[cnt]);
370 		if (!err)
371 			err = ret;
372 	}
373 	return err;
374 }
375 
376 static inline void dqput_all(struct dquot **dquot)
377 {
378 	unsigned int cnt;
379 
380 	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
381 		dqput(dquot[cnt]);
382 }
383 
384 /* This function needs dq_list_lock */
385 static inline int clear_dquot_dirty(struct dquot *dquot)
386 {
387 	if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags))
388 		return 0;
389 	list_del_init(&dquot->dq_dirty);
390 	return 1;
391 }
392 
393 void mark_info_dirty(struct super_block *sb, int type)
394 {
395 	set_bit(DQF_INFO_DIRTY_B, &sb_dqopt(sb)->info[type].dqi_flags);
396 }
397 EXPORT_SYMBOL(mark_info_dirty);
398 
399 /*
400  *	Read dquot from disk and alloc space for it
401  */
402 
403 int dquot_acquire(struct dquot *dquot)
404 {
405 	int ret = 0, ret2 = 0;
406 	struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
407 
408 	mutex_lock(&dquot->dq_lock);
409 	mutex_lock(&dqopt->dqio_mutex);
410 	if (!test_bit(DQ_READ_B, &dquot->dq_flags))
411 		ret = dqopt->ops[dquot->dq_id.type]->read_dqblk(dquot);
412 	if (ret < 0)
413 		goto out_iolock;
414 	/* Make sure flags update is visible after dquot has been filled */
415 	smp_mb__before_atomic();
416 	set_bit(DQ_READ_B, &dquot->dq_flags);
417 	/* Instantiate dquot if needed */
418 	if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
419 		ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
420 		/* Write the info if needed */
421 		if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
422 			ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
423 					dquot->dq_sb, dquot->dq_id.type);
424 		}
425 		if (ret < 0)
426 			goto out_iolock;
427 		if (ret2 < 0) {
428 			ret = ret2;
429 			goto out_iolock;
430 		}
431 	}
432 	/*
433 	 * Make sure flags update is visible after on-disk struct has been
434 	 * allocated. Paired with smp_rmb() in dqget().
435 	 */
436 	smp_mb__before_atomic();
437 	set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
438 out_iolock:
439 	mutex_unlock(&dqopt->dqio_mutex);
440 	mutex_unlock(&dquot->dq_lock);
441 	return ret;
442 }
443 EXPORT_SYMBOL(dquot_acquire);
444 
445 /*
446  *	Write dquot to disk
447  */
448 int dquot_commit(struct dquot *dquot)
449 {
450 	int ret = 0;
451 	struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
452 
453 	mutex_lock(&dqopt->dqio_mutex);
454 	spin_lock(&dq_list_lock);
455 	if (!clear_dquot_dirty(dquot)) {
456 		spin_unlock(&dq_list_lock);
457 		goto out_sem;
458 	}
459 	spin_unlock(&dq_list_lock);
460 	/* Inactive dquot can be only if there was error during read/init
461 	 * => we have better not writing it */
462 	if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
463 		ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
464 	else
465 		ret = -EIO;
466 out_sem:
467 	mutex_unlock(&dqopt->dqio_mutex);
468 	return ret;
469 }
470 EXPORT_SYMBOL(dquot_commit);
471 
472 /*
473  *	Release dquot
474  */
475 int dquot_release(struct dquot *dquot)
476 {
477 	int ret = 0, ret2 = 0;
478 	struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
479 
480 	mutex_lock(&dquot->dq_lock);
481 	/* Check whether we are not racing with some other dqget() */
482 	if (atomic_read(&dquot->dq_count) > 1)
483 		goto out_dqlock;
484 	mutex_lock(&dqopt->dqio_mutex);
485 	if (dqopt->ops[dquot->dq_id.type]->release_dqblk) {
486 		ret = dqopt->ops[dquot->dq_id.type]->release_dqblk(dquot);
487 		/* Write the info */
488 		if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
489 			ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
490 						dquot->dq_sb, dquot->dq_id.type);
491 		}
492 		if (ret >= 0)
493 			ret = ret2;
494 	}
495 	clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
496 	mutex_unlock(&dqopt->dqio_mutex);
497 out_dqlock:
498 	mutex_unlock(&dquot->dq_lock);
499 	return ret;
500 }
501 EXPORT_SYMBOL(dquot_release);
502 
503 void dquot_destroy(struct dquot *dquot)
504 {
505 	kmem_cache_free(dquot_cachep, dquot);
506 }
507 EXPORT_SYMBOL(dquot_destroy);
508 
509 static inline void do_destroy_dquot(struct dquot *dquot)
510 {
511 	dquot->dq_sb->dq_op->destroy_dquot(dquot);
512 }
513 
514 /* Invalidate all dquots on the list. Note that this function is called after
515  * quota is disabled and pointers from inodes removed so there cannot be new
516  * quota users. There can still be some users of quotas due to inodes being
517  * just deleted or pruned by prune_icache() (those are not attached to any
518  * list) or parallel quotactl call. We have to wait for such users.
519  */
520 static void invalidate_dquots(struct super_block *sb, int type)
521 {
522 	struct dquot *dquot, *tmp;
523 
524 restart:
525 	spin_lock(&dq_list_lock);
526 	list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
527 		if (dquot->dq_sb != sb)
528 			continue;
529 		if (dquot->dq_id.type != type)
530 			continue;
531 		/* Wait for dquot users */
532 		if (atomic_read(&dquot->dq_count)) {
533 			DEFINE_WAIT(wait);
534 
535 			dqgrab(dquot);
536 			prepare_to_wait(&dquot->dq_wait_unused, &wait,
537 					TASK_UNINTERRUPTIBLE);
538 			spin_unlock(&dq_list_lock);
539 			/* Once dqput() wakes us up, we know it's time to free
540 			 * the dquot.
541 			 * IMPORTANT: we rely on the fact that there is always
542 			 * at most one process waiting for dquot to free.
543 			 * Otherwise dq_count would be > 1 and we would never
544 			 * wake up.
545 			 */
546 			if (atomic_read(&dquot->dq_count) > 1)
547 				schedule();
548 			finish_wait(&dquot->dq_wait_unused, &wait);
549 			dqput(dquot);
550 			/* At this moment dquot() need not exist (it could be
551 			 * reclaimed by prune_dqcache(). Hence we must
552 			 * restart. */
553 			goto restart;
554 		}
555 		/*
556 		 * Quota now has no users and it has been written on last
557 		 * dqput()
558 		 */
559 		remove_dquot_hash(dquot);
560 		remove_free_dquot(dquot);
561 		remove_inuse(dquot);
562 		do_destroy_dquot(dquot);
563 	}
564 	spin_unlock(&dq_list_lock);
565 }
566 
567 /* Call callback for every active dquot on given filesystem */
568 int dquot_scan_active(struct super_block *sb,
569 		      int (*fn)(struct dquot *dquot, unsigned long priv),
570 		      unsigned long priv)
571 {
572 	struct dquot *dquot, *old_dquot = NULL;
573 	int ret = 0;
574 
575 	WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
576 
577 	spin_lock(&dq_list_lock);
578 	list_for_each_entry(dquot, &inuse_list, dq_inuse) {
579 		if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
580 			continue;
581 		if (dquot->dq_sb != sb)
582 			continue;
583 		/* Now we have active dquot so we can just increase use count */
584 		atomic_inc(&dquot->dq_count);
585 		spin_unlock(&dq_list_lock);
586 		dqstats_inc(DQST_LOOKUPS);
587 		dqput(old_dquot);
588 		old_dquot = dquot;
589 		/*
590 		 * ->release_dquot() can be racing with us. Our reference
591 		 * protects us from new calls to it so just wait for any
592 		 * outstanding call and recheck the DQ_ACTIVE_B after that.
593 		 */
594 		wait_on_dquot(dquot);
595 		if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
596 			ret = fn(dquot, priv);
597 			if (ret < 0)
598 				goto out;
599 		}
600 		spin_lock(&dq_list_lock);
601 		/* We are safe to continue now because our dquot could not
602 		 * be moved out of the inuse list while we hold the reference */
603 	}
604 	spin_unlock(&dq_list_lock);
605 out:
606 	dqput(old_dquot);
607 	return ret;
608 }
609 EXPORT_SYMBOL(dquot_scan_active);
610 
611 /* Write all dquot structures to quota files */
612 int dquot_writeback_dquots(struct super_block *sb, int type)
613 {
614 	struct list_head *dirty;
615 	struct dquot *dquot;
616 	struct quota_info *dqopt = sb_dqopt(sb);
617 	int cnt;
618 	int err, ret = 0;
619 
620 	WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
621 
622 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
623 		if (type != -1 && cnt != type)
624 			continue;
625 		if (!sb_has_quota_active(sb, cnt))
626 			continue;
627 		spin_lock(&dq_list_lock);
628 		dirty = &dqopt->info[cnt].dqi_dirty_list;
629 		while (!list_empty(dirty)) {
630 			dquot = list_first_entry(dirty, struct dquot,
631 						 dq_dirty);
632 			/* Dirty and inactive can be only bad dquot... */
633 			if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
634 				clear_dquot_dirty(dquot);
635 				continue;
636 			}
637 			/* Now we have active dquot from which someone is
638  			 * holding reference so we can safely just increase
639 			 * use count */
640 			dqgrab(dquot);
641 			spin_unlock(&dq_list_lock);
642 			dqstats_inc(DQST_LOOKUPS);
643 			err = sb->dq_op->write_dquot(dquot);
644 			if (!ret && err)
645 				ret = err;
646 			dqput(dquot);
647 			spin_lock(&dq_list_lock);
648 		}
649 		spin_unlock(&dq_list_lock);
650 	}
651 
652 	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
653 		if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
654 		    && info_dirty(&dqopt->info[cnt]))
655 			sb->dq_op->write_info(sb, cnt);
656 	dqstats_inc(DQST_SYNCS);
657 
658 	return ret;
659 }
660 EXPORT_SYMBOL(dquot_writeback_dquots);
661 
662 /* Write all dquot structures to disk and make them visible from userspace */
663 int dquot_quota_sync(struct super_block *sb, int type)
664 {
665 	struct quota_info *dqopt = sb_dqopt(sb);
666 	int cnt;
667 	int ret;
668 
669 	ret = dquot_writeback_dquots(sb, type);
670 	if (ret)
671 		return ret;
672 	if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
673 		return 0;
674 
675 	/* This is not very clever (and fast) but currently I don't know about
676 	 * any other simple way of getting quota data to disk and we must get
677 	 * them there for userspace to be visible... */
678 	if (sb->s_op->sync_fs)
679 		sb->s_op->sync_fs(sb, 1);
680 	sync_blockdev(sb->s_bdev);
681 
682 	/*
683 	 * Now when everything is written we can discard the pagecache so
684 	 * that userspace sees the changes.
685 	 */
686 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
687 		if (type != -1 && cnt != type)
688 			continue;
689 		if (!sb_has_quota_active(sb, cnt))
690 			continue;
691 		inode_lock(dqopt->files[cnt]);
692 		truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
693 		inode_unlock(dqopt->files[cnt]);
694 	}
695 
696 	return 0;
697 }
698 EXPORT_SYMBOL(dquot_quota_sync);
699 
700 static unsigned long
701 dqcache_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
702 {
703 	struct list_head *head;
704 	struct dquot *dquot;
705 	unsigned long freed = 0;
706 
707 	spin_lock(&dq_list_lock);
708 	head = free_dquots.prev;
709 	while (head != &free_dquots && sc->nr_to_scan) {
710 		dquot = list_entry(head, struct dquot, dq_free);
711 		remove_dquot_hash(dquot);
712 		remove_free_dquot(dquot);
713 		remove_inuse(dquot);
714 		do_destroy_dquot(dquot);
715 		sc->nr_to_scan--;
716 		freed++;
717 		head = free_dquots.prev;
718 	}
719 	spin_unlock(&dq_list_lock);
720 	return freed;
721 }
722 
723 static unsigned long
724 dqcache_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
725 {
726 	return vfs_pressure_ratio(
727 	percpu_counter_read_positive(&dqstats.counter[DQST_FREE_DQUOTS]));
728 }
729 
730 static struct shrinker dqcache_shrinker = {
731 	.count_objects = dqcache_shrink_count,
732 	.scan_objects = dqcache_shrink_scan,
733 	.seeks = DEFAULT_SEEKS,
734 };
735 
736 /*
737  * Put reference to dquot
738  */
739 void dqput(struct dquot *dquot)
740 {
741 	int ret;
742 
743 	if (!dquot)
744 		return;
745 #ifdef CONFIG_QUOTA_DEBUG
746 	if (!atomic_read(&dquot->dq_count)) {
747 		quota_error(dquot->dq_sb, "trying to free free dquot of %s %d",
748 			    quotatypes[dquot->dq_id.type],
749 			    from_kqid(&init_user_ns, dquot->dq_id));
750 		BUG();
751 	}
752 #endif
753 	dqstats_inc(DQST_DROPS);
754 we_slept:
755 	spin_lock(&dq_list_lock);
756 	if (atomic_read(&dquot->dq_count) > 1) {
757 		/* We have more than one user... nothing to do */
758 		atomic_dec(&dquot->dq_count);
759 		/* Releasing dquot during quotaoff phase? */
760 		if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_id.type) &&
761 		    atomic_read(&dquot->dq_count) == 1)
762 			wake_up(&dquot->dq_wait_unused);
763 		spin_unlock(&dq_list_lock);
764 		return;
765 	}
766 	/* Need to release dquot? */
767 	if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && dquot_dirty(dquot)) {
768 		spin_unlock(&dq_list_lock);
769 		/* Commit dquot before releasing */
770 		ret = dquot->dq_sb->dq_op->write_dquot(dquot);
771 		if (ret < 0) {
772 			quota_error(dquot->dq_sb, "Can't write quota structure"
773 				    " (error %d). Quota may get out of sync!",
774 				    ret);
775 			/*
776 			 * We clear dirty bit anyway, so that we avoid
777 			 * infinite loop here
778 			 */
779 			spin_lock(&dq_list_lock);
780 			clear_dquot_dirty(dquot);
781 			spin_unlock(&dq_list_lock);
782 		}
783 		goto we_slept;
784 	}
785 	/* Clear flag in case dquot was inactive (something bad happened) */
786 	clear_dquot_dirty(dquot);
787 	if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
788 		spin_unlock(&dq_list_lock);
789 		dquot->dq_sb->dq_op->release_dquot(dquot);
790 		goto we_slept;
791 	}
792 	atomic_dec(&dquot->dq_count);
793 #ifdef CONFIG_QUOTA_DEBUG
794 	/* sanity check */
795 	BUG_ON(!list_empty(&dquot->dq_free));
796 #endif
797 	put_dquot_last(dquot);
798 	spin_unlock(&dq_list_lock);
799 }
800 EXPORT_SYMBOL(dqput);
801 
802 struct dquot *dquot_alloc(struct super_block *sb, int type)
803 {
804 	return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
805 }
806 EXPORT_SYMBOL(dquot_alloc);
807 
808 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
809 {
810 	struct dquot *dquot;
811 
812 	dquot = sb->dq_op->alloc_dquot(sb, type);
813 	if(!dquot)
814 		return NULL;
815 
816 	mutex_init(&dquot->dq_lock);
817 	INIT_LIST_HEAD(&dquot->dq_free);
818 	INIT_LIST_HEAD(&dquot->dq_inuse);
819 	INIT_HLIST_NODE(&dquot->dq_hash);
820 	INIT_LIST_HEAD(&dquot->dq_dirty);
821 	init_waitqueue_head(&dquot->dq_wait_unused);
822 	dquot->dq_sb = sb;
823 	dquot->dq_id = make_kqid_invalid(type);
824 	atomic_set(&dquot->dq_count, 1);
825 
826 	return dquot;
827 }
828 
829 /*
830  * Get reference to dquot
831  *
832  * Locking is slightly tricky here. We are guarded from parallel quotaoff()
833  * destroying our dquot by:
834  *   a) checking for quota flags under dq_list_lock and
835  *   b) getting a reference to dquot before we release dq_list_lock
836  */
837 struct dquot *dqget(struct super_block *sb, struct kqid qid)
838 {
839 	unsigned int hashent = hashfn(sb, qid);
840 	struct dquot *dquot, *empty = NULL;
841 
842 	if (!qid_has_mapping(sb->s_user_ns, qid))
843 		return ERR_PTR(-EINVAL);
844 
845         if (!sb_has_quota_active(sb, qid.type))
846 		return ERR_PTR(-ESRCH);
847 we_slept:
848 	spin_lock(&dq_list_lock);
849 	spin_lock(&dq_state_lock);
850 	if (!sb_has_quota_active(sb, qid.type)) {
851 		spin_unlock(&dq_state_lock);
852 		spin_unlock(&dq_list_lock);
853 		dquot = ERR_PTR(-ESRCH);
854 		goto out;
855 	}
856 	spin_unlock(&dq_state_lock);
857 
858 	dquot = find_dquot(hashent, sb, qid);
859 	if (!dquot) {
860 		if (!empty) {
861 			spin_unlock(&dq_list_lock);
862 			empty = get_empty_dquot(sb, qid.type);
863 			if (!empty)
864 				schedule();	/* Try to wait for a moment... */
865 			goto we_slept;
866 		}
867 		dquot = empty;
868 		empty = NULL;
869 		dquot->dq_id = qid;
870 		/* all dquots go on the inuse_list */
871 		put_inuse(dquot);
872 		/* hash it first so it can be found */
873 		insert_dquot_hash(dquot);
874 		spin_unlock(&dq_list_lock);
875 		dqstats_inc(DQST_LOOKUPS);
876 	} else {
877 		if (!atomic_read(&dquot->dq_count))
878 			remove_free_dquot(dquot);
879 		atomic_inc(&dquot->dq_count);
880 		spin_unlock(&dq_list_lock);
881 		dqstats_inc(DQST_CACHE_HITS);
882 		dqstats_inc(DQST_LOOKUPS);
883 	}
884 	/* Wait for dq_lock - after this we know that either dquot_release() is
885 	 * already finished or it will be canceled due to dq_count > 1 test */
886 	wait_on_dquot(dquot);
887 	/* Read the dquot / allocate space in quota file */
888 	if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
889 		int err;
890 
891 		err = sb->dq_op->acquire_dquot(dquot);
892 		if (err < 0) {
893 			dqput(dquot);
894 			dquot = ERR_PTR(err);
895 			goto out;
896 		}
897 	}
898 	/*
899 	 * Make sure following reads see filled structure - paired with
900 	 * smp_mb__before_atomic() in dquot_acquire().
901 	 */
902 	smp_rmb();
903 #ifdef CONFIG_QUOTA_DEBUG
904 	BUG_ON(!dquot->dq_sb);	/* Has somebody invalidated entry under us? */
905 #endif
906 out:
907 	if (empty)
908 		do_destroy_dquot(empty);
909 
910 	return dquot;
911 }
912 EXPORT_SYMBOL(dqget);
913 
914 static inline struct dquot **i_dquot(struct inode *inode)
915 {
916 	return inode->i_sb->s_op->get_dquots(inode);
917 }
918 
919 static int dqinit_needed(struct inode *inode, int type)
920 {
921 	struct dquot * const *dquots;
922 	int cnt;
923 
924 	if (IS_NOQUOTA(inode))
925 		return 0;
926 
927 	dquots = i_dquot(inode);
928 	if (type != -1)
929 		return !dquots[type];
930 	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
931 		if (!dquots[cnt])
932 			return 1;
933 	return 0;
934 }
935 
936 /* This routine is guarded by s_umount semaphore */
937 static void add_dquot_ref(struct super_block *sb, int type)
938 {
939 	struct inode *inode, *old_inode = NULL;
940 #ifdef CONFIG_QUOTA_DEBUG
941 	int reserved = 0;
942 #endif
943 
944 	spin_lock(&sb->s_inode_list_lock);
945 	list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
946 		spin_lock(&inode->i_lock);
947 		if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
948 		    !atomic_read(&inode->i_writecount) ||
949 		    !dqinit_needed(inode, type)) {
950 			spin_unlock(&inode->i_lock);
951 			continue;
952 		}
953 		__iget(inode);
954 		spin_unlock(&inode->i_lock);
955 		spin_unlock(&sb->s_inode_list_lock);
956 
957 #ifdef CONFIG_QUOTA_DEBUG
958 		if (unlikely(inode_get_rsv_space(inode) > 0))
959 			reserved = 1;
960 #endif
961 		iput(old_inode);
962 		__dquot_initialize(inode, type);
963 
964 		/*
965 		 * We hold a reference to 'inode' so it couldn't have been
966 		 * removed from s_inodes list while we dropped the
967 		 * s_inode_list_lock. We cannot iput the inode now as we can be
968 		 * holding the last reference and we cannot iput it under
969 		 * s_inode_list_lock. So we keep the reference and iput it
970 		 * later.
971 		 */
972 		old_inode = inode;
973 		spin_lock(&sb->s_inode_list_lock);
974 	}
975 	spin_unlock(&sb->s_inode_list_lock);
976 	iput(old_inode);
977 
978 #ifdef CONFIG_QUOTA_DEBUG
979 	if (reserved) {
980 		quota_error(sb, "Writes happened before quota was turned on "
981 			"thus quota information is probably inconsistent. "
982 			"Please run quotacheck(8)");
983 	}
984 #endif
985 }
986 
987 /*
988  * Remove references to dquots from inode and add dquot to list for freeing
989  * if we have the last reference to dquot
990  */
991 static void remove_inode_dquot_ref(struct inode *inode, int type,
992 				   struct list_head *tofree_head)
993 {
994 	struct dquot **dquots = i_dquot(inode);
995 	struct dquot *dquot = dquots[type];
996 
997 	if (!dquot)
998 		return;
999 
1000 	dquots[type] = NULL;
1001 	if (list_empty(&dquot->dq_free)) {
1002 		/*
1003 		 * The inode still has reference to dquot so it can't be in the
1004 		 * free list
1005 		 */
1006 		spin_lock(&dq_list_lock);
1007 		list_add(&dquot->dq_free, tofree_head);
1008 		spin_unlock(&dq_list_lock);
1009 	} else {
1010 		/*
1011 		 * Dquot is already in a list to put so we won't drop the last
1012 		 * reference here.
1013 		 */
1014 		dqput(dquot);
1015 	}
1016 }
1017 
1018 /*
1019  * Free list of dquots
1020  * Dquots are removed from inodes and no new references can be got so we are
1021  * the only ones holding reference
1022  */
1023 static void put_dquot_list(struct list_head *tofree_head)
1024 {
1025 	struct list_head *act_head;
1026 	struct dquot *dquot;
1027 
1028 	act_head = tofree_head->next;
1029 	while (act_head != tofree_head) {
1030 		dquot = list_entry(act_head, struct dquot, dq_free);
1031 		act_head = act_head->next;
1032 		/* Remove dquot from the list so we won't have problems... */
1033 		list_del_init(&dquot->dq_free);
1034 		dqput(dquot);
1035 	}
1036 }
1037 
1038 static void remove_dquot_ref(struct super_block *sb, int type,
1039 		struct list_head *tofree_head)
1040 {
1041 	struct inode *inode;
1042 	int reserved = 0;
1043 
1044 	spin_lock(&sb->s_inode_list_lock);
1045 	list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1046 		/*
1047 		 *  We have to scan also I_NEW inodes because they can already
1048 		 *  have quota pointer initialized. Luckily, we need to touch
1049 		 *  only quota pointers and these have separate locking
1050 		 *  (dq_data_lock).
1051 		 */
1052 		spin_lock(&dq_data_lock);
1053 		if (!IS_NOQUOTA(inode)) {
1054 			if (unlikely(inode_get_rsv_space(inode) > 0))
1055 				reserved = 1;
1056 			remove_inode_dquot_ref(inode, type, tofree_head);
1057 		}
1058 		spin_unlock(&dq_data_lock);
1059 	}
1060 	spin_unlock(&sb->s_inode_list_lock);
1061 #ifdef CONFIG_QUOTA_DEBUG
1062 	if (reserved) {
1063 		printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1064 			" was disabled thus quota information is probably "
1065 			"inconsistent. Please run quotacheck(8).\n", sb->s_id);
1066 	}
1067 #endif
1068 }
1069 
1070 /* Gather all references from inodes and drop them */
1071 static void drop_dquot_ref(struct super_block *sb, int type)
1072 {
1073 	LIST_HEAD(tofree_head);
1074 
1075 	if (sb->dq_op) {
1076 		remove_dquot_ref(sb, type, &tofree_head);
1077 		synchronize_srcu(&dquot_srcu);
1078 		put_dquot_list(&tofree_head);
1079 	}
1080 }
1081 
1082 static inline void dquot_incr_inodes(struct dquot *dquot, qsize_t number)
1083 {
1084 	dquot->dq_dqb.dqb_curinodes += number;
1085 }
1086 
1087 static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
1088 {
1089 	dquot->dq_dqb.dqb_curspace += number;
1090 }
1091 
1092 static inline void dquot_resv_space(struct dquot *dquot, qsize_t number)
1093 {
1094 	dquot->dq_dqb.dqb_rsvspace += number;
1095 }
1096 
1097 /*
1098  * Claim reserved quota space
1099  */
1100 static void dquot_claim_reserved_space(struct dquot *dquot, qsize_t number)
1101 {
1102 	if (dquot->dq_dqb.dqb_rsvspace < number) {
1103 		WARN_ON_ONCE(1);
1104 		number = dquot->dq_dqb.dqb_rsvspace;
1105 	}
1106 	dquot->dq_dqb.dqb_curspace += number;
1107 	dquot->dq_dqb.dqb_rsvspace -= number;
1108 }
1109 
1110 static void dquot_reclaim_reserved_space(struct dquot *dquot, qsize_t number)
1111 {
1112 	if (WARN_ON_ONCE(dquot->dq_dqb.dqb_curspace < number))
1113 		number = dquot->dq_dqb.dqb_curspace;
1114 	dquot->dq_dqb.dqb_rsvspace += number;
1115 	dquot->dq_dqb.dqb_curspace -= number;
1116 }
1117 
1118 static inline
1119 void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1120 {
1121 	if (dquot->dq_dqb.dqb_rsvspace >= number)
1122 		dquot->dq_dqb.dqb_rsvspace -= number;
1123 	else {
1124 		WARN_ON_ONCE(1);
1125 		dquot->dq_dqb.dqb_rsvspace = 0;
1126 	}
1127 }
1128 
1129 static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1130 {
1131 	if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1132 	    dquot->dq_dqb.dqb_curinodes >= number)
1133 		dquot->dq_dqb.dqb_curinodes -= number;
1134 	else
1135 		dquot->dq_dqb.dqb_curinodes = 0;
1136 	if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1137 		dquot->dq_dqb.dqb_itime = (time64_t) 0;
1138 	clear_bit(DQ_INODES_B, &dquot->dq_flags);
1139 }
1140 
1141 static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1142 {
1143 	if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1144 	    dquot->dq_dqb.dqb_curspace >= number)
1145 		dquot->dq_dqb.dqb_curspace -= number;
1146 	else
1147 		dquot->dq_dqb.dqb_curspace = 0;
1148 	if (dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1149 		dquot->dq_dqb.dqb_btime = (time64_t) 0;
1150 	clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1151 }
1152 
1153 struct dquot_warn {
1154 	struct super_block *w_sb;
1155 	struct kqid w_dq_id;
1156 	short w_type;
1157 };
1158 
1159 static int warning_issued(struct dquot *dquot, const int warntype)
1160 {
1161 	int flag = (warntype == QUOTA_NL_BHARDWARN ||
1162 		warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1163 		((warntype == QUOTA_NL_IHARDWARN ||
1164 		warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1165 
1166 	if (!flag)
1167 		return 0;
1168 	return test_and_set_bit(flag, &dquot->dq_flags);
1169 }
1170 
1171 #ifdef CONFIG_PRINT_QUOTA_WARNING
1172 static int flag_print_warnings = 1;
1173 
1174 static int need_print_warning(struct dquot_warn *warn)
1175 {
1176 	if (!flag_print_warnings)
1177 		return 0;
1178 
1179 	switch (warn->w_dq_id.type) {
1180 		case USRQUOTA:
1181 			return uid_eq(current_fsuid(), warn->w_dq_id.uid);
1182 		case GRPQUOTA:
1183 			return in_group_p(warn->w_dq_id.gid);
1184 		case PRJQUOTA:
1185 			return 1;
1186 	}
1187 	return 0;
1188 }
1189 
1190 /* Print warning to user which exceeded quota */
1191 static void print_warning(struct dquot_warn *warn)
1192 {
1193 	char *msg = NULL;
1194 	struct tty_struct *tty;
1195 	int warntype = warn->w_type;
1196 
1197 	if (warntype == QUOTA_NL_IHARDBELOW ||
1198 	    warntype == QUOTA_NL_ISOFTBELOW ||
1199 	    warntype == QUOTA_NL_BHARDBELOW ||
1200 	    warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(warn))
1201 		return;
1202 
1203 	tty = get_current_tty();
1204 	if (!tty)
1205 		return;
1206 	tty_write_message(tty, warn->w_sb->s_id);
1207 	if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1208 		tty_write_message(tty, ": warning, ");
1209 	else
1210 		tty_write_message(tty, ": write failed, ");
1211 	tty_write_message(tty, quotatypes[warn->w_dq_id.type]);
1212 	switch (warntype) {
1213 		case QUOTA_NL_IHARDWARN:
1214 			msg = " file limit reached.\r\n";
1215 			break;
1216 		case QUOTA_NL_ISOFTLONGWARN:
1217 			msg = " file quota exceeded too long.\r\n";
1218 			break;
1219 		case QUOTA_NL_ISOFTWARN:
1220 			msg = " file quota exceeded.\r\n";
1221 			break;
1222 		case QUOTA_NL_BHARDWARN:
1223 			msg = " block limit reached.\r\n";
1224 			break;
1225 		case QUOTA_NL_BSOFTLONGWARN:
1226 			msg = " block quota exceeded too long.\r\n";
1227 			break;
1228 		case QUOTA_NL_BSOFTWARN:
1229 			msg = " block quota exceeded.\r\n";
1230 			break;
1231 	}
1232 	tty_write_message(tty, msg);
1233 	tty_kref_put(tty);
1234 }
1235 #endif
1236 
1237 static void prepare_warning(struct dquot_warn *warn, struct dquot *dquot,
1238 			    int warntype)
1239 {
1240 	if (warning_issued(dquot, warntype))
1241 		return;
1242 	warn->w_type = warntype;
1243 	warn->w_sb = dquot->dq_sb;
1244 	warn->w_dq_id = dquot->dq_id;
1245 }
1246 
1247 /*
1248  * Write warnings to the console and send warning messages over netlink.
1249  *
1250  * Note that this function can call into tty and networking code.
1251  */
1252 static void flush_warnings(struct dquot_warn *warn)
1253 {
1254 	int i;
1255 
1256 	for (i = 0; i < MAXQUOTAS; i++) {
1257 		if (warn[i].w_type == QUOTA_NL_NOWARN)
1258 			continue;
1259 #ifdef CONFIG_PRINT_QUOTA_WARNING
1260 		print_warning(&warn[i]);
1261 #endif
1262 		quota_send_warning(warn[i].w_dq_id,
1263 				   warn[i].w_sb->s_dev, warn[i].w_type);
1264 	}
1265 }
1266 
1267 static int ignore_hardlimit(struct dquot *dquot)
1268 {
1269 	struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
1270 
1271 	return capable(CAP_SYS_RESOURCE) &&
1272 	       (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1273 		!(info->dqi_flags & DQF_ROOT_SQUASH));
1274 }
1275 
1276 /* needs dq_data_lock */
1277 static int check_idq(struct dquot *dquot, qsize_t inodes,
1278 		     struct dquot_warn *warn)
1279 {
1280 	qsize_t newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1281 
1282 	if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type) ||
1283 	    test_bit(DQ_FAKE_B, &dquot->dq_flags))
1284 		return 0;
1285 
1286 	if (dquot->dq_dqb.dqb_ihardlimit &&
1287 	    newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1288             !ignore_hardlimit(dquot)) {
1289 		prepare_warning(warn, dquot, QUOTA_NL_IHARDWARN);
1290 		return -EDQUOT;
1291 	}
1292 
1293 	if (dquot->dq_dqb.dqb_isoftlimit &&
1294 	    newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1295 	    dquot->dq_dqb.dqb_itime &&
1296 	    ktime_get_real_seconds() >= dquot->dq_dqb.dqb_itime &&
1297             !ignore_hardlimit(dquot)) {
1298 		prepare_warning(warn, dquot, QUOTA_NL_ISOFTLONGWARN);
1299 		return -EDQUOT;
1300 	}
1301 
1302 	if (dquot->dq_dqb.dqb_isoftlimit &&
1303 	    newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1304 	    dquot->dq_dqb.dqb_itime == 0) {
1305 		prepare_warning(warn, dquot, QUOTA_NL_ISOFTWARN);
1306 		dquot->dq_dqb.dqb_itime = ktime_get_real_seconds() +
1307 		    sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type].dqi_igrace;
1308 	}
1309 
1310 	return 0;
1311 }
1312 
1313 /* needs dq_data_lock */
1314 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc,
1315 		     struct dquot_warn *warn)
1316 {
1317 	qsize_t tspace;
1318 	struct super_block *sb = dquot->dq_sb;
1319 
1320 	if (!sb_has_quota_limits_enabled(sb, dquot->dq_id.type) ||
1321 	    test_bit(DQ_FAKE_B, &dquot->dq_flags))
1322 		return 0;
1323 
1324 	tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1325 		+ space;
1326 
1327 	if (dquot->dq_dqb.dqb_bhardlimit &&
1328 	    tspace > dquot->dq_dqb.dqb_bhardlimit &&
1329             !ignore_hardlimit(dquot)) {
1330 		if (!prealloc)
1331 			prepare_warning(warn, dquot, QUOTA_NL_BHARDWARN);
1332 		return -EDQUOT;
1333 	}
1334 
1335 	if (dquot->dq_dqb.dqb_bsoftlimit &&
1336 	    tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1337 	    dquot->dq_dqb.dqb_btime &&
1338 	    ktime_get_real_seconds() >= dquot->dq_dqb.dqb_btime &&
1339             !ignore_hardlimit(dquot)) {
1340 		if (!prealloc)
1341 			prepare_warning(warn, dquot, QUOTA_NL_BSOFTLONGWARN);
1342 		return -EDQUOT;
1343 	}
1344 
1345 	if (dquot->dq_dqb.dqb_bsoftlimit &&
1346 	    tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1347 	    dquot->dq_dqb.dqb_btime == 0) {
1348 		if (!prealloc) {
1349 			prepare_warning(warn, dquot, QUOTA_NL_BSOFTWARN);
1350 			dquot->dq_dqb.dqb_btime = ktime_get_real_seconds() +
1351 			    sb_dqopt(sb)->info[dquot->dq_id.type].dqi_bgrace;
1352 		}
1353 		else
1354 			/*
1355 			 * We don't allow preallocation to exceed softlimit so exceeding will
1356 			 * be always printed
1357 			 */
1358 			return -EDQUOT;
1359 	}
1360 
1361 	return 0;
1362 }
1363 
1364 static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1365 {
1366 	qsize_t newinodes;
1367 
1368 	if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1369 	    dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1370 	    !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type))
1371 		return QUOTA_NL_NOWARN;
1372 
1373 	newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1374 	if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1375 		return QUOTA_NL_ISOFTBELOW;
1376 	if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1377 	    newinodes < dquot->dq_dqb.dqb_ihardlimit)
1378 		return QUOTA_NL_IHARDBELOW;
1379 	return QUOTA_NL_NOWARN;
1380 }
1381 
1382 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1383 {
1384 	if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1385 	    dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1386 		return QUOTA_NL_NOWARN;
1387 
1388 	if (dquot->dq_dqb.dqb_curspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1389 		return QUOTA_NL_BSOFTBELOW;
1390 	if (dquot->dq_dqb.dqb_curspace >= dquot->dq_dqb.dqb_bhardlimit &&
1391 	    dquot->dq_dqb.dqb_curspace - space < dquot->dq_dqb.dqb_bhardlimit)
1392 		return QUOTA_NL_BHARDBELOW;
1393 	return QUOTA_NL_NOWARN;
1394 }
1395 
1396 static int dquot_active(const struct inode *inode)
1397 {
1398 	struct super_block *sb = inode->i_sb;
1399 
1400 	if (IS_NOQUOTA(inode))
1401 		return 0;
1402 	return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1403 }
1404 
1405 /*
1406  * Initialize quota pointers in inode
1407  *
1408  * It is better to call this function outside of any transaction as it
1409  * might need a lot of space in journal for dquot structure allocation.
1410  */
1411 static int __dquot_initialize(struct inode *inode, int type)
1412 {
1413 	int cnt, init_needed = 0;
1414 	struct dquot **dquots, *got[MAXQUOTAS] = {};
1415 	struct super_block *sb = inode->i_sb;
1416 	qsize_t rsv;
1417 	int ret = 0;
1418 
1419 	if (!dquot_active(inode))
1420 		return 0;
1421 
1422 	dquots = i_dquot(inode);
1423 
1424 	/* First get references to structures we might need. */
1425 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1426 		struct kqid qid;
1427 		kprojid_t projid;
1428 		int rc;
1429 		struct dquot *dquot;
1430 
1431 		if (type != -1 && cnt != type)
1432 			continue;
1433 		/*
1434 		 * The i_dquot should have been initialized in most cases,
1435 		 * we check it without locking here to avoid unnecessary
1436 		 * dqget()/dqput() calls.
1437 		 */
1438 		if (dquots[cnt])
1439 			continue;
1440 
1441 		if (!sb_has_quota_active(sb, cnt))
1442 			continue;
1443 
1444 		init_needed = 1;
1445 
1446 		switch (cnt) {
1447 		case USRQUOTA:
1448 			qid = make_kqid_uid(inode->i_uid);
1449 			break;
1450 		case GRPQUOTA:
1451 			qid = make_kqid_gid(inode->i_gid);
1452 			break;
1453 		case PRJQUOTA:
1454 			rc = inode->i_sb->dq_op->get_projid(inode, &projid);
1455 			if (rc)
1456 				continue;
1457 			qid = make_kqid_projid(projid);
1458 			break;
1459 		}
1460 		dquot = dqget(sb, qid);
1461 		if (IS_ERR(dquot)) {
1462 			/* We raced with somebody turning quotas off... */
1463 			if (PTR_ERR(dquot) != -ESRCH) {
1464 				ret = PTR_ERR(dquot);
1465 				goto out_put;
1466 			}
1467 			dquot = NULL;
1468 		}
1469 		got[cnt] = dquot;
1470 	}
1471 
1472 	/* All required i_dquot has been initialized */
1473 	if (!init_needed)
1474 		return 0;
1475 
1476 	spin_lock(&dq_data_lock);
1477 	if (IS_NOQUOTA(inode))
1478 		goto out_lock;
1479 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1480 		if (type != -1 && cnt != type)
1481 			continue;
1482 		/* Avoid races with quotaoff() */
1483 		if (!sb_has_quota_active(sb, cnt))
1484 			continue;
1485 		/* We could race with quotaon or dqget() could have failed */
1486 		if (!got[cnt])
1487 			continue;
1488 		if (!dquots[cnt]) {
1489 			dquots[cnt] = got[cnt];
1490 			got[cnt] = NULL;
1491 			/*
1492 			 * Make quota reservation system happy if someone
1493 			 * did a write before quota was turned on
1494 			 */
1495 			rsv = inode_get_rsv_space(inode);
1496 			if (unlikely(rsv))
1497 				dquot_resv_space(dquots[cnt], rsv);
1498 		}
1499 	}
1500 out_lock:
1501 	spin_unlock(&dq_data_lock);
1502 out_put:
1503 	/* Drop unused references */
1504 	dqput_all(got);
1505 
1506 	return ret;
1507 }
1508 
1509 int dquot_initialize(struct inode *inode)
1510 {
1511 	return __dquot_initialize(inode, -1);
1512 }
1513 EXPORT_SYMBOL(dquot_initialize);
1514 
1515 bool dquot_initialize_needed(struct inode *inode)
1516 {
1517 	struct dquot **dquots;
1518 	int i;
1519 
1520 	if (!dquot_active(inode))
1521 		return false;
1522 
1523 	dquots = i_dquot(inode);
1524 	for (i = 0; i < MAXQUOTAS; i++)
1525 		if (!dquots[i] && sb_has_quota_active(inode->i_sb, i))
1526 			return true;
1527 	return false;
1528 }
1529 EXPORT_SYMBOL(dquot_initialize_needed);
1530 
1531 /*
1532  * Release all quotas referenced by inode.
1533  *
1534  * This function only be called on inode free or converting
1535  * a file to quota file, no other users for the i_dquot in
1536  * both cases, so we needn't call synchronize_srcu() after
1537  * clearing i_dquot.
1538  */
1539 static void __dquot_drop(struct inode *inode)
1540 {
1541 	int cnt;
1542 	struct dquot **dquots = i_dquot(inode);
1543 	struct dquot *put[MAXQUOTAS];
1544 
1545 	spin_lock(&dq_data_lock);
1546 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1547 		put[cnt] = dquots[cnt];
1548 		dquots[cnt] = NULL;
1549 	}
1550 	spin_unlock(&dq_data_lock);
1551 	dqput_all(put);
1552 }
1553 
1554 void dquot_drop(struct inode *inode)
1555 {
1556 	struct dquot * const *dquots;
1557 	int cnt;
1558 
1559 	if (IS_NOQUOTA(inode))
1560 		return;
1561 
1562 	/*
1563 	 * Test before calling to rule out calls from proc and such
1564 	 * where we are not allowed to block. Note that this is
1565 	 * actually reliable test even without the lock - the caller
1566 	 * must assure that nobody can come after the DQUOT_DROP and
1567 	 * add quota pointers back anyway.
1568 	 */
1569 	dquots = i_dquot(inode);
1570 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1571 		if (dquots[cnt])
1572 			break;
1573 	}
1574 
1575 	if (cnt < MAXQUOTAS)
1576 		__dquot_drop(inode);
1577 }
1578 EXPORT_SYMBOL(dquot_drop);
1579 
1580 /*
1581  * inode_reserved_space is managed internally by quota, and protected by
1582  * i_lock similar to i_blocks+i_bytes.
1583  */
1584 static qsize_t *inode_reserved_space(struct inode * inode)
1585 {
1586 	/* Filesystem must explicitly define it's own method in order to use
1587 	 * quota reservation interface */
1588 	BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1589 	return inode->i_sb->dq_op->get_reserved_space(inode);
1590 }
1591 
1592 void inode_add_rsv_space(struct inode *inode, qsize_t number)
1593 {
1594 	spin_lock(&inode->i_lock);
1595 	*inode_reserved_space(inode) += number;
1596 	spin_unlock(&inode->i_lock);
1597 }
1598 EXPORT_SYMBOL(inode_add_rsv_space);
1599 
1600 void inode_claim_rsv_space(struct inode *inode, qsize_t number)
1601 {
1602 	spin_lock(&inode->i_lock);
1603 	*inode_reserved_space(inode) -= number;
1604 	__inode_add_bytes(inode, number);
1605 	spin_unlock(&inode->i_lock);
1606 }
1607 EXPORT_SYMBOL(inode_claim_rsv_space);
1608 
1609 void inode_reclaim_rsv_space(struct inode *inode, qsize_t number)
1610 {
1611 	spin_lock(&inode->i_lock);
1612 	*inode_reserved_space(inode) += number;
1613 	__inode_sub_bytes(inode, number);
1614 	spin_unlock(&inode->i_lock);
1615 }
1616 EXPORT_SYMBOL(inode_reclaim_rsv_space);
1617 
1618 void inode_sub_rsv_space(struct inode *inode, qsize_t number)
1619 {
1620 	spin_lock(&inode->i_lock);
1621 	*inode_reserved_space(inode) -= number;
1622 	spin_unlock(&inode->i_lock);
1623 }
1624 EXPORT_SYMBOL(inode_sub_rsv_space);
1625 
1626 static qsize_t inode_get_rsv_space(struct inode *inode)
1627 {
1628 	qsize_t ret;
1629 
1630 	if (!inode->i_sb->dq_op->get_reserved_space)
1631 		return 0;
1632 	spin_lock(&inode->i_lock);
1633 	ret = *inode_reserved_space(inode);
1634 	spin_unlock(&inode->i_lock);
1635 	return ret;
1636 }
1637 
1638 static void inode_incr_space(struct inode *inode, qsize_t number,
1639 				int reserve)
1640 {
1641 	if (reserve)
1642 		inode_add_rsv_space(inode, number);
1643 	else
1644 		inode_add_bytes(inode, number);
1645 }
1646 
1647 static void inode_decr_space(struct inode *inode, qsize_t number, int reserve)
1648 {
1649 	if (reserve)
1650 		inode_sub_rsv_space(inode, number);
1651 	else
1652 		inode_sub_bytes(inode, number);
1653 }
1654 
1655 /*
1656  * This functions updates i_blocks+i_bytes fields and quota information
1657  * (together with appropriate checks).
1658  *
1659  * NOTE: We absolutely rely on the fact that caller dirties the inode
1660  * (usually helpers in quotaops.h care about this) and holds a handle for
1661  * the current transaction so that dquot write and inode write go into the
1662  * same transaction.
1663  */
1664 
1665 /*
1666  * This operation can block, but only after everything is updated
1667  */
1668 int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
1669 {
1670 	int cnt, ret = 0, index;
1671 	struct dquot_warn warn[MAXQUOTAS];
1672 	int reserve = flags & DQUOT_SPACE_RESERVE;
1673 	struct dquot **dquots;
1674 
1675 	if (!dquot_active(inode)) {
1676 		inode_incr_space(inode, number, reserve);
1677 		goto out;
1678 	}
1679 
1680 	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1681 		warn[cnt].w_type = QUOTA_NL_NOWARN;
1682 
1683 	dquots = i_dquot(inode);
1684 	index = srcu_read_lock(&dquot_srcu);
1685 	spin_lock(&dq_data_lock);
1686 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1687 		if (!dquots[cnt])
1688 			continue;
1689 		ret = check_bdq(dquots[cnt], number,
1690 				!(flags & DQUOT_SPACE_WARN), &warn[cnt]);
1691 		if (ret && !(flags & DQUOT_SPACE_NOFAIL)) {
1692 			spin_unlock(&dq_data_lock);
1693 			goto out_flush_warn;
1694 		}
1695 	}
1696 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1697 		if (!dquots[cnt])
1698 			continue;
1699 		if (reserve)
1700 			dquot_resv_space(dquots[cnt], number);
1701 		else
1702 			dquot_incr_space(dquots[cnt], number);
1703 	}
1704 	inode_incr_space(inode, number, reserve);
1705 	spin_unlock(&dq_data_lock);
1706 
1707 	if (reserve)
1708 		goto out_flush_warn;
1709 	mark_all_dquot_dirty(dquots);
1710 out_flush_warn:
1711 	srcu_read_unlock(&dquot_srcu, index);
1712 	flush_warnings(warn);
1713 out:
1714 	return ret;
1715 }
1716 EXPORT_SYMBOL(__dquot_alloc_space);
1717 
1718 /*
1719  * This operation can block, but only after everything is updated
1720  */
1721 int dquot_alloc_inode(struct inode *inode)
1722 {
1723 	int cnt, ret = 0, index;
1724 	struct dquot_warn warn[MAXQUOTAS];
1725 	struct dquot * const *dquots;
1726 
1727 	if (!dquot_active(inode))
1728 		return 0;
1729 	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1730 		warn[cnt].w_type = QUOTA_NL_NOWARN;
1731 
1732 	dquots = i_dquot(inode);
1733 	index = srcu_read_lock(&dquot_srcu);
1734 	spin_lock(&dq_data_lock);
1735 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1736 		if (!dquots[cnt])
1737 			continue;
1738 		ret = check_idq(dquots[cnt], 1, &warn[cnt]);
1739 		if (ret)
1740 			goto warn_put_all;
1741 	}
1742 
1743 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1744 		if (!dquots[cnt])
1745 			continue;
1746 		dquot_incr_inodes(dquots[cnt], 1);
1747 	}
1748 
1749 warn_put_all:
1750 	spin_unlock(&dq_data_lock);
1751 	if (ret == 0)
1752 		mark_all_dquot_dirty(dquots);
1753 	srcu_read_unlock(&dquot_srcu, index);
1754 	flush_warnings(warn);
1755 	return ret;
1756 }
1757 EXPORT_SYMBOL(dquot_alloc_inode);
1758 
1759 /*
1760  * Convert in-memory reserved quotas to real consumed quotas
1761  */
1762 int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1763 {
1764 	struct dquot **dquots;
1765 	int cnt, index;
1766 
1767 	if (!dquot_active(inode)) {
1768 		inode_claim_rsv_space(inode, number);
1769 		return 0;
1770 	}
1771 
1772 	dquots = i_dquot(inode);
1773 	index = srcu_read_lock(&dquot_srcu);
1774 	spin_lock(&dq_data_lock);
1775 	/* Claim reserved quotas to allocated quotas */
1776 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1777 		if (dquots[cnt])
1778 			dquot_claim_reserved_space(dquots[cnt], number);
1779 	}
1780 	/* Update inode bytes */
1781 	inode_claim_rsv_space(inode, number);
1782 	spin_unlock(&dq_data_lock);
1783 	mark_all_dquot_dirty(dquots);
1784 	srcu_read_unlock(&dquot_srcu, index);
1785 	return 0;
1786 }
1787 EXPORT_SYMBOL(dquot_claim_space_nodirty);
1788 
1789 /*
1790  * Convert allocated space back to in-memory reserved quotas
1791  */
1792 void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
1793 {
1794 	struct dquot **dquots;
1795 	int cnt, index;
1796 
1797 	if (!dquot_active(inode)) {
1798 		inode_reclaim_rsv_space(inode, number);
1799 		return;
1800 	}
1801 
1802 	dquots = i_dquot(inode);
1803 	index = srcu_read_lock(&dquot_srcu);
1804 	spin_lock(&dq_data_lock);
1805 	/* Claim reserved quotas to allocated quotas */
1806 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1807 		if (dquots[cnt])
1808 			dquot_reclaim_reserved_space(dquots[cnt], number);
1809 	}
1810 	/* Update inode bytes */
1811 	inode_reclaim_rsv_space(inode, number);
1812 	spin_unlock(&dq_data_lock);
1813 	mark_all_dquot_dirty(dquots);
1814 	srcu_read_unlock(&dquot_srcu, index);
1815 	return;
1816 }
1817 EXPORT_SYMBOL(dquot_reclaim_space_nodirty);
1818 
1819 /*
1820  * This operation can block, but only after everything is updated
1821  */
1822 void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
1823 {
1824 	unsigned int cnt;
1825 	struct dquot_warn warn[MAXQUOTAS];
1826 	struct dquot **dquots;
1827 	int reserve = flags & DQUOT_SPACE_RESERVE, index;
1828 
1829 	if (!dquot_active(inode)) {
1830 		inode_decr_space(inode, number, reserve);
1831 		return;
1832 	}
1833 
1834 	dquots = i_dquot(inode);
1835 	index = srcu_read_lock(&dquot_srcu);
1836 	spin_lock(&dq_data_lock);
1837 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1838 		int wtype;
1839 
1840 		warn[cnt].w_type = QUOTA_NL_NOWARN;
1841 		if (!dquots[cnt])
1842 			continue;
1843 		wtype = info_bdq_free(dquots[cnt], number);
1844 		if (wtype != QUOTA_NL_NOWARN)
1845 			prepare_warning(&warn[cnt], dquots[cnt], wtype);
1846 		if (reserve)
1847 			dquot_free_reserved_space(dquots[cnt], number);
1848 		else
1849 			dquot_decr_space(dquots[cnt], number);
1850 	}
1851 	inode_decr_space(inode, number, reserve);
1852 	spin_unlock(&dq_data_lock);
1853 
1854 	if (reserve)
1855 		goto out_unlock;
1856 	mark_all_dquot_dirty(dquots);
1857 out_unlock:
1858 	srcu_read_unlock(&dquot_srcu, index);
1859 	flush_warnings(warn);
1860 }
1861 EXPORT_SYMBOL(__dquot_free_space);
1862 
1863 /*
1864  * This operation can block, but only after everything is updated
1865  */
1866 void dquot_free_inode(struct inode *inode)
1867 {
1868 	unsigned int cnt;
1869 	struct dquot_warn warn[MAXQUOTAS];
1870 	struct dquot * const *dquots;
1871 	int index;
1872 
1873 	if (!dquot_active(inode))
1874 		return;
1875 
1876 	dquots = i_dquot(inode);
1877 	index = srcu_read_lock(&dquot_srcu);
1878 	spin_lock(&dq_data_lock);
1879 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1880 		int wtype;
1881 
1882 		warn[cnt].w_type = QUOTA_NL_NOWARN;
1883 		if (!dquots[cnt])
1884 			continue;
1885 		wtype = info_idq_free(dquots[cnt], 1);
1886 		if (wtype != QUOTA_NL_NOWARN)
1887 			prepare_warning(&warn[cnt], dquots[cnt], wtype);
1888 		dquot_decr_inodes(dquots[cnt], 1);
1889 	}
1890 	spin_unlock(&dq_data_lock);
1891 	mark_all_dquot_dirty(dquots);
1892 	srcu_read_unlock(&dquot_srcu, index);
1893 	flush_warnings(warn);
1894 }
1895 EXPORT_SYMBOL(dquot_free_inode);
1896 
1897 /*
1898  * Transfer the number of inode and blocks from one diskquota to an other.
1899  * On success, dquot references in transfer_to are consumed and references
1900  * to original dquots that need to be released are placed there. On failure,
1901  * references are kept untouched.
1902  *
1903  * This operation can block, but only after everything is updated
1904  * A transaction must be started when entering this function.
1905  *
1906  * We are holding reference on transfer_from & transfer_to, no need to
1907  * protect them by srcu_read_lock().
1908  */
1909 int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
1910 {
1911 	qsize_t space, cur_space;
1912 	qsize_t rsv_space = 0;
1913 	struct dquot *transfer_from[MAXQUOTAS] = {};
1914 	int cnt, ret = 0;
1915 	char is_valid[MAXQUOTAS] = {};
1916 	struct dquot_warn warn_to[MAXQUOTAS];
1917 	struct dquot_warn warn_from_inodes[MAXQUOTAS];
1918 	struct dquot_warn warn_from_space[MAXQUOTAS];
1919 
1920 	if (IS_NOQUOTA(inode))
1921 		return 0;
1922 	/* Initialize the arrays */
1923 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1924 		warn_to[cnt].w_type = QUOTA_NL_NOWARN;
1925 		warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
1926 		warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
1927 	}
1928 
1929 	spin_lock(&dq_data_lock);
1930 	if (IS_NOQUOTA(inode)) {	/* File without quota accounting? */
1931 		spin_unlock(&dq_data_lock);
1932 		return 0;
1933 	}
1934 	cur_space = inode_get_bytes(inode);
1935 	rsv_space = inode_get_rsv_space(inode);
1936 	space = cur_space + rsv_space;
1937 	/* Build the transfer_from list and check the limits */
1938 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1939 		/*
1940 		 * Skip changes for same uid or gid or for turned off quota-type.
1941 		 */
1942 		if (!transfer_to[cnt])
1943 			continue;
1944 		/* Avoid races with quotaoff() */
1945 		if (!sb_has_quota_active(inode->i_sb, cnt))
1946 			continue;
1947 		is_valid[cnt] = 1;
1948 		transfer_from[cnt] = i_dquot(inode)[cnt];
1949 		ret = check_idq(transfer_to[cnt], 1, &warn_to[cnt]);
1950 		if (ret)
1951 			goto over_quota;
1952 		ret = check_bdq(transfer_to[cnt], space, 0, &warn_to[cnt]);
1953 		if (ret)
1954 			goto over_quota;
1955 	}
1956 
1957 	/*
1958 	 * Finally perform the needed transfer from transfer_from to transfer_to
1959 	 */
1960 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1961 		if (!is_valid[cnt])
1962 			continue;
1963 		/* Due to IO error we might not have transfer_from[] structure */
1964 		if (transfer_from[cnt]) {
1965 			int wtype;
1966 			wtype = info_idq_free(transfer_from[cnt], 1);
1967 			if (wtype != QUOTA_NL_NOWARN)
1968 				prepare_warning(&warn_from_inodes[cnt],
1969 						transfer_from[cnt], wtype);
1970 			wtype = info_bdq_free(transfer_from[cnt], space);
1971 			if (wtype != QUOTA_NL_NOWARN)
1972 				prepare_warning(&warn_from_space[cnt],
1973 						transfer_from[cnt], wtype);
1974 			dquot_decr_inodes(transfer_from[cnt], 1);
1975 			dquot_decr_space(transfer_from[cnt], cur_space);
1976 			dquot_free_reserved_space(transfer_from[cnt],
1977 						  rsv_space);
1978 		}
1979 
1980 		dquot_incr_inodes(transfer_to[cnt], 1);
1981 		dquot_incr_space(transfer_to[cnt], cur_space);
1982 		dquot_resv_space(transfer_to[cnt], rsv_space);
1983 
1984 		i_dquot(inode)[cnt] = transfer_to[cnt];
1985 	}
1986 	spin_unlock(&dq_data_lock);
1987 
1988 	mark_all_dquot_dirty(transfer_from);
1989 	mark_all_dquot_dirty(transfer_to);
1990 	flush_warnings(warn_to);
1991 	flush_warnings(warn_from_inodes);
1992 	flush_warnings(warn_from_space);
1993 	/* Pass back references to put */
1994 	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1995 		if (is_valid[cnt])
1996 			transfer_to[cnt] = transfer_from[cnt];
1997 	return 0;
1998 over_quota:
1999 	spin_unlock(&dq_data_lock);
2000 	flush_warnings(warn_to);
2001 	return ret;
2002 }
2003 EXPORT_SYMBOL(__dquot_transfer);
2004 
2005 /* Wrapper for transferring ownership of an inode for uid/gid only
2006  * Called from FSXXX_setattr()
2007  */
2008 int dquot_transfer(struct inode *inode, struct iattr *iattr)
2009 {
2010 	struct dquot *transfer_to[MAXQUOTAS] = {};
2011 	struct dquot *dquot;
2012 	struct super_block *sb = inode->i_sb;
2013 	int ret;
2014 
2015 	if (!dquot_active(inode))
2016 		return 0;
2017 
2018 	if (iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)){
2019 		dquot = dqget(sb, make_kqid_uid(iattr->ia_uid));
2020 		if (IS_ERR(dquot)) {
2021 			if (PTR_ERR(dquot) != -ESRCH) {
2022 				ret = PTR_ERR(dquot);
2023 				goto out_put;
2024 			}
2025 			dquot = NULL;
2026 		}
2027 		transfer_to[USRQUOTA] = dquot;
2028 	}
2029 	if (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid)){
2030 		dquot = dqget(sb, make_kqid_gid(iattr->ia_gid));
2031 		if (IS_ERR(dquot)) {
2032 			if (PTR_ERR(dquot) != -ESRCH) {
2033 				ret = PTR_ERR(dquot);
2034 				goto out_put;
2035 			}
2036 			dquot = NULL;
2037 		}
2038 		transfer_to[GRPQUOTA] = dquot;
2039 	}
2040 	ret = __dquot_transfer(inode, transfer_to);
2041 out_put:
2042 	dqput_all(transfer_to);
2043 	return ret;
2044 }
2045 EXPORT_SYMBOL(dquot_transfer);
2046 
2047 /*
2048  * Write info of quota file to disk
2049  */
2050 int dquot_commit_info(struct super_block *sb, int type)
2051 {
2052 	int ret;
2053 	struct quota_info *dqopt = sb_dqopt(sb);
2054 
2055 	mutex_lock(&dqopt->dqio_mutex);
2056 	ret = dqopt->ops[type]->write_file_info(sb, type);
2057 	mutex_unlock(&dqopt->dqio_mutex);
2058 	return ret;
2059 }
2060 EXPORT_SYMBOL(dquot_commit_info);
2061 
2062 int dquot_get_next_id(struct super_block *sb, struct kqid *qid)
2063 {
2064 	struct quota_info *dqopt = sb_dqopt(sb);
2065 	int err;
2066 
2067 	if (!sb_has_quota_active(sb, qid->type))
2068 		return -ESRCH;
2069 	if (!dqopt->ops[qid->type]->get_next_id)
2070 		return -ENOSYS;
2071 	mutex_lock(&dqopt->dqio_mutex);
2072 	err = dqopt->ops[qid->type]->get_next_id(sb, qid);
2073 	mutex_unlock(&dqopt->dqio_mutex);
2074 	return err;
2075 }
2076 EXPORT_SYMBOL(dquot_get_next_id);
2077 
2078 /*
2079  * Definitions of diskquota operations.
2080  */
2081 const struct dquot_operations dquot_operations = {
2082 	.write_dquot	= dquot_commit,
2083 	.acquire_dquot	= dquot_acquire,
2084 	.release_dquot	= dquot_release,
2085 	.mark_dirty	= dquot_mark_dquot_dirty,
2086 	.write_info	= dquot_commit_info,
2087 	.alloc_dquot	= dquot_alloc,
2088 	.destroy_dquot	= dquot_destroy,
2089 	.get_next_id	= dquot_get_next_id,
2090 };
2091 EXPORT_SYMBOL(dquot_operations);
2092 
2093 /*
2094  * Generic helper for ->open on filesystems supporting disk quotas.
2095  */
2096 int dquot_file_open(struct inode *inode, struct file *file)
2097 {
2098 	int error;
2099 
2100 	error = generic_file_open(inode, file);
2101 	if (!error && (file->f_mode & FMODE_WRITE))
2102 		dquot_initialize(inode);
2103 	return error;
2104 }
2105 EXPORT_SYMBOL(dquot_file_open);
2106 
2107 /*
2108  * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
2109  */
2110 int dquot_disable(struct super_block *sb, int type, unsigned int flags)
2111 {
2112 	int cnt, ret = 0;
2113 	struct quota_info *dqopt = sb_dqopt(sb);
2114 	struct inode *toputinode[MAXQUOTAS];
2115 
2116 	/* s_umount should be held in exclusive mode */
2117 	if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2118 		up_read(&sb->s_umount);
2119 
2120 	/* Cannot turn off usage accounting without turning off limits, or
2121 	 * suspend quotas and simultaneously turn quotas off. */
2122 	if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
2123 	    || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
2124 	    DQUOT_USAGE_ENABLED)))
2125 		return -EINVAL;
2126 
2127 	/*
2128 	 * Skip everything if there's nothing to do. We have to do this because
2129 	 * sometimes we are called when fill_super() failed and calling
2130 	 * sync_fs() in such cases does no good.
2131 	 */
2132 	if (!sb_any_quota_loaded(sb))
2133 		return 0;
2134 
2135 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2136 		toputinode[cnt] = NULL;
2137 		if (type != -1 && cnt != type)
2138 			continue;
2139 		if (!sb_has_quota_loaded(sb, cnt))
2140 			continue;
2141 
2142 		if (flags & DQUOT_SUSPENDED) {
2143 			spin_lock(&dq_state_lock);
2144 			dqopt->flags |=
2145 				dquot_state_flag(DQUOT_SUSPENDED, cnt);
2146 			spin_unlock(&dq_state_lock);
2147 		} else {
2148 			spin_lock(&dq_state_lock);
2149 			dqopt->flags &= ~dquot_state_flag(flags, cnt);
2150 			/* Turning off suspended quotas? */
2151 			if (!sb_has_quota_loaded(sb, cnt) &&
2152 			    sb_has_quota_suspended(sb, cnt)) {
2153 				dqopt->flags &=	~dquot_state_flag(
2154 							DQUOT_SUSPENDED, cnt);
2155 				spin_unlock(&dq_state_lock);
2156 				iput(dqopt->files[cnt]);
2157 				dqopt->files[cnt] = NULL;
2158 				continue;
2159 			}
2160 			spin_unlock(&dq_state_lock);
2161 		}
2162 
2163 		/* We still have to keep quota loaded? */
2164 		if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
2165 			continue;
2166 
2167 		/* Note: these are blocking operations */
2168 		drop_dquot_ref(sb, cnt);
2169 		invalidate_dquots(sb, cnt);
2170 		/*
2171 		 * Now all dquots should be invalidated, all writes done so we
2172 		 * should be only users of the info. No locks needed.
2173 		 */
2174 		if (info_dirty(&dqopt->info[cnt]))
2175 			sb->dq_op->write_info(sb, cnt);
2176 		if (dqopt->ops[cnt]->free_file_info)
2177 			dqopt->ops[cnt]->free_file_info(sb, cnt);
2178 		put_quota_format(dqopt->info[cnt].dqi_format);
2179 
2180 		toputinode[cnt] = dqopt->files[cnt];
2181 		if (!sb_has_quota_loaded(sb, cnt))
2182 			dqopt->files[cnt] = NULL;
2183 		dqopt->info[cnt].dqi_flags = 0;
2184 		dqopt->info[cnt].dqi_igrace = 0;
2185 		dqopt->info[cnt].dqi_bgrace = 0;
2186 		dqopt->ops[cnt] = NULL;
2187 	}
2188 
2189 	/* Skip syncing and setting flags if quota files are hidden */
2190 	if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2191 		goto put_inodes;
2192 
2193 	/* Sync the superblock so that buffers with quota data are written to
2194 	 * disk (and so userspace sees correct data afterwards). */
2195 	if (sb->s_op->sync_fs)
2196 		sb->s_op->sync_fs(sb, 1);
2197 	sync_blockdev(sb->s_bdev);
2198 	/* Now the quota files are just ordinary files and we can set the
2199 	 * inode flags back. Moreover we discard the pagecache so that
2200 	 * userspace sees the writes we did bypassing the pagecache. We
2201 	 * must also discard the blockdev buffers so that we see the
2202 	 * changes done by userspace on the next quotaon() */
2203 	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2204 		/* This can happen when suspending quotas on remount-ro... */
2205 		if (toputinode[cnt] && !sb_has_quota_loaded(sb, cnt)) {
2206 			inode_lock(toputinode[cnt]);
2207 			toputinode[cnt]->i_flags &= ~S_NOQUOTA;
2208 			truncate_inode_pages(&toputinode[cnt]->i_data, 0);
2209 			inode_unlock(toputinode[cnt]);
2210 			mark_inode_dirty_sync(toputinode[cnt]);
2211 		}
2212 	if (sb->s_bdev)
2213 		invalidate_bdev(sb->s_bdev);
2214 put_inodes:
2215 	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2216 		if (toputinode[cnt]) {
2217 			/* On remount RO, we keep the inode pointer so that we
2218 			 * can reenable quota on the subsequent remount RW. We
2219 			 * have to check 'flags' variable and not use sb_has_
2220 			 * function because another quotaon / quotaoff could
2221 			 * change global state before we got here. We refuse
2222 			 * to suspend quotas when there is pending delete on
2223 			 * the quota file... */
2224 			if (!(flags & DQUOT_SUSPENDED))
2225 				iput(toputinode[cnt]);
2226 			else if (!toputinode[cnt]->i_nlink)
2227 				ret = -EBUSY;
2228 		}
2229 	return ret;
2230 }
2231 EXPORT_SYMBOL(dquot_disable);
2232 
2233 int dquot_quota_off(struct super_block *sb, int type)
2234 {
2235 	return dquot_disable(sb, type,
2236 			     DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2237 }
2238 EXPORT_SYMBOL(dquot_quota_off);
2239 
2240 /*
2241  *	Turn quotas on on a device
2242  */
2243 
2244 /*
2245  * Helper function to turn quotas on when we already have the inode of
2246  * quota file and no quota information is loaded.
2247  */
2248 static int vfs_load_quota_inode(struct inode *inode, int type, int format_id,
2249 	unsigned int flags)
2250 {
2251 	struct quota_format_type *fmt = find_quota_format(format_id);
2252 	struct super_block *sb = inode->i_sb;
2253 	struct quota_info *dqopt = sb_dqopt(sb);
2254 	int error;
2255 
2256 	if (!fmt)
2257 		return -ESRCH;
2258 	if (!S_ISREG(inode->i_mode)) {
2259 		error = -EACCES;
2260 		goto out_fmt;
2261 	}
2262 	if (IS_RDONLY(inode)) {
2263 		error = -EROFS;
2264 		goto out_fmt;
2265 	}
2266 	if (!sb->s_op->quota_write || !sb->s_op->quota_read ||
2267 	    (type == PRJQUOTA && sb->dq_op->get_projid == NULL)) {
2268 		error = -EINVAL;
2269 		goto out_fmt;
2270 	}
2271 	/* Filesystems outside of init_user_ns not yet supported */
2272 	if (sb->s_user_ns != &init_user_ns) {
2273 		error = -EINVAL;
2274 		goto out_fmt;
2275 	}
2276 	/* Usage always has to be set... */
2277 	if (!(flags & DQUOT_USAGE_ENABLED)) {
2278 		error = -EINVAL;
2279 		goto out_fmt;
2280 	}
2281 	if (sb_has_quota_loaded(sb, type)) {
2282 		error = -EBUSY;
2283 		goto out_fmt;
2284 	}
2285 
2286 	if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2287 		/* As we bypass the pagecache we must now flush all the
2288 		 * dirty data and invalidate caches so that kernel sees
2289 		 * changes from userspace. It is not enough to just flush
2290 		 * the quota file since if blocksize < pagesize, invalidation
2291 		 * of the cache could fail because of other unrelated dirty
2292 		 * data */
2293 		sync_filesystem(sb);
2294 		invalidate_bdev(sb->s_bdev);
2295 	}
2296 
2297 	if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2298 		/* We don't want quota and atime on quota files (deadlocks
2299 		 * possible) Also nobody should write to the file - we use
2300 		 * special IO operations which ignore the immutable bit. */
2301 		inode_lock(inode);
2302 		inode->i_flags |= S_NOQUOTA;
2303 		inode_unlock(inode);
2304 		/*
2305 		 * When S_NOQUOTA is set, remove dquot references as no more
2306 		 * references can be added
2307 		 */
2308 		__dquot_drop(inode);
2309 	}
2310 
2311 	error = -EIO;
2312 	dqopt->files[type] = igrab(inode);
2313 	if (!dqopt->files[type])
2314 		goto out_file_flags;
2315 	error = -EINVAL;
2316 	if (!fmt->qf_ops->check_quota_file(sb, type))
2317 		goto out_file_init;
2318 
2319 	dqopt->ops[type] = fmt->qf_ops;
2320 	dqopt->info[type].dqi_format = fmt;
2321 	dqopt->info[type].dqi_fmt_id = format_id;
2322 	INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2323 	mutex_lock(&dqopt->dqio_mutex);
2324 	error = dqopt->ops[type]->read_file_info(sb, type);
2325 	if (error < 0) {
2326 		mutex_unlock(&dqopt->dqio_mutex);
2327 		goto out_file_init;
2328 	}
2329 	if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2330 		dqopt->info[type].dqi_flags |= DQF_SYS_FILE;
2331 	mutex_unlock(&dqopt->dqio_mutex);
2332 	spin_lock(&dq_state_lock);
2333 	dqopt->flags |= dquot_state_flag(flags, type);
2334 	spin_unlock(&dq_state_lock);
2335 
2336 	add_dquot_ref(sb, type);
2337 
2338 	return 0;
2339 
2340 out_file_init:
2341 	dqopt->files[type] = NULL;
2342 	iput(inode);
2343 out_file_flags:
2344 	inode_lock(inode);
2345 	inode->i_flags &= ~S_NOQUOTA;
2346 	inode_unlock(inode);
2347 out_fmt:
2348 	put_quota_format(fmt);
2349 
2350 	return error;
2351 }
2352 
2353 /* Reenable quotas on remount RW */
2354 int dquot_resume(struct super_block *sb, int type)
2355 {
2356 	struct quota_info *dqopt = sb_dqopt(sb);
2357 	struct inode *inode;
2358 	int ret = 0, cnt;
2359 	unsigned int flags;
2360 
2361 	/* s_umount should be held in exclusive mode */
2362 	if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2363 		up_read(&sb->s_umount);
2364 
2365 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2366 		if (type != -1 && cnt != type)
2367 			continue;
2368 		if (!sb_has_quota_suspended(sb, cnt))
2369 			continue;
2370 
2371 		inode = dqopt->files[cnt];
2372 		dqopt->files[cnt] = NULL;
2373 		spin_lock(&dq_state_lock);
2374 		flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2375 							DQUOT_LIMITS_ENABLED,
2376 							cnt);
2377 		dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2378 		spin_unlock(&dq_state_lock);
2379 
2380 		flags = dquot_generic_flag(flags, cnt);
2381 		ret = vfs_load_quota_inode(inode, cnt,
2382 				dqopt->info[cnt].dqi_fmt_id, flags);
2383 		iput(inode);
2384 	}
2385 
2386 	return ret;
2387 }
2388 EXPORT_SYMBOL(dquot_resume);
2389 
2390 int dquot_quota_on(struct super_block *sb, int type, int format_id,
2391 		   const struct path *path)
2392 {
2393 	int error = security_quota_on(path->dentry);
2394 	if (error)
2395 		return error;
2396 	/* Quota file not on the same filesystem? */
2397 	if (path->dentry->d_sb != sb)
2398 		error = -EXDEV;
2399 	else
2400 		error = vfs_load_quota_inode(d_inode(path->dentry), type,
2401 					     format_id, DQUOT_USAGE_ENABLED |
2402 					     DQUOT_LIMITS_ENABLED);
2403 	return error;
2404 }
2405 EXPORT_SYMBOL(dquot_quota_on);
2406 
2407 /*
2408  * More powerful function for turning on quotas allowing setting
2409  * of individual quota flags
2410  */
2411 int dquot_enable(struct inode *inode, int type, int format_id,
2412 		 unsigned int flags)
2413 {
2414 	struct super_block *sb = inode->i_sb;
2415 
2416 	/* Just unsuspend quotas? */
2417 	BUG_ON(flags & DQUOT_SUSPENDED);
2418 	/* s_umount should be held in exclusive mode */
2419 	if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2420 		up_read(&sb->s_umount);
2421 
2422 	if (!flags)
2423 		return 0;
2424 	/* Just updating flags needed? */
2425 	if (sb_has_quota_loaded(sb, type)) {
2426 		if (flags & DQUOT_USAGE_ENABLED &&
2427 		    sb_has_quota_usage_enabled(sb, type))
2428 			return -EBUSY;
2429 		if (flags & DQUOT_LIMITS_ENABLED &&
2430 		    sb_has_quota_limits_enabled(sb, type))
2431 			return -EBUSY;
2432 		spin_lock(&dq_state_lock);
2433 		sb_dqopt(sb)->flags |= dquot_state_flag(flags, type);
2434 		spin_unlock(&dq_state_lock);
2435 		return 0;
2436 	}
2437 
2438 	return vfs_load_quota_inode(inode, type, format_id, flags);
2439 }
2440 EXPORT_SYMBOL(dquot_enable);
2441 
2442 /*
2443  * This function is used when filesystem needs to initialize quotas
2444  * during mount time.
2445  */
2446 int dquot_quota_on_mount(struct super_block *sb, char *qf_name,
2447 		int format_id, int type)
2448 {
2449 	struct dentry *dentry;
2450 	int error;
2451 
2452 	dentry = lookup_one_len_unlocked(qf_name, sb->s_root, strlen(qf_name));
2453 	if (IS_ERR(dentry))
2454 		return PTR_ERR(dentry);
2455 
2456 	if (d_really_is_negative(dentry)) {
2457 		error = -ENOENT;
2458 		goto out;
2459 	}
2460 
2461 	error = security_quota_on(dentry);
2462 	if (!error)
2463 		error = vfs_load_quota_inode(d_inode(dentry), type, format_id,
2464 				DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2465 
2466 out:
2467 	dput(dentry);
2468 	return error;
2469 }
2470 EXPORT_SYMBOL(dquot_quota_on_mount);
2471 
2472 static int dquot_quota_enable(struct super_block *sb, unsigned int flags)
2473 {
2474 	int ret;
2475 	int type;
2476 	struct quota_info *dqopt = sb_dqopt(sb);
2477 
2478 	if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2479 		return -ENOSYS;
2480 	/* Accounting cannot be turned on while fs is mounted */
2481 	flags &= ~(FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT);
2482 	if (!flags)
2483 		return -EINVAL;
2484 	for (type = 0; type < MAXQUOTAS; type++) {
2485 		if (!(flags & qtype_enforce_flag(type)))
2486 			continue;
2487 		/* Can't enforce without accounting */
2488 		if (!sb_has_quota_usage_enabled(sb, type))
2489 			return -EINVAL;
2490 		ret = dquot_enable(dqopt->files[type], type,
2491 				   dqopt->info[type].dqi_fmt_id,
2492 				   DQUOT_LIMITS_ENABLED);
2493 		if (ret < 0)
2494 			goto out_err;
2495 	}
2496 	return 0;
2497 out_err:
2498 	/* Backout enforcement enablement we already did */
2499 	for (type--; type >= 0; type--)  {
2500 		if (flags & qtype_enforce_flag(type))
2501 			dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2502 	}
2503 	/* Error code translation for better compatibility with XFS */
2504 	if (ret == -EBUSY)
2505 		ret = -EEXIST;
2506 	return ret;
2507 }
2508 
2509 static int dquot_quota_disable(struct super_block *sb, unsigned int flags)
2510 {
2511 	int ret;
2512 	int type;
2513 	struct quota_info *dqopt = sb_dqopt(sb);
2514 
2515 	if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2516 		return -ENOSYS;
2517 	/*
2518 	 * We don't support turning off accounting via quotactl. In principle
2519 	 * quota infrastructure can do this but filesystems don't expect
2520 	 * userspace to be able to do it.
2521 	 */
2522 	if (flags &
2523 		  (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT))
2524 		return -EOPNOTSUPP;
2525 
2526 	/* Filter out limits not enabled */
2527 	for (type = 0; type < MAXQUOTAS; type++)
2528 		if (!sb_has_quota_limits_enabled(sb, type))
2529 			flags &= ~qtype_enforce_flag(type);
2530 	/* Nothing left? */
2531 	if (!flags)
2532 		return -EEXIST;
2533 	for (type = 0; type < MAXQUOTAS; type++) {
2534 		if (flags & qtype_enforce_flag(type)) {
2535 			ret = dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2536 			if (ret < 0)
2537 				goto out_err;
2538 		}
2539 	}
2540 	return 0;
2541 out_err:
2542 	/* Backout enforcement disabling we already did */
2543 	for (type--; type >= 0; type--)  {
2544 		if (flags & qtype_enforce_flag(type))
2545 			dquot_enable(dqopt->files[type], type,
2546 				     dqopt->info[type].dqi_fmt_id,
2547 				     DQUOT_LIMITS_ENABLED);
2548 	}
2549 	return ret;
2550 }
2551 
2552 /* Generic routine for getting common part of quota structure */
2553 static void do_get_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2554 {
2555 	struct mem_dqblk *dm = &dquot->dq_dqb;
2556 
2557 	memset(di, 0, sizeof(*di));
2558 	spin_lock(&dq_data_lock);
2559 	di->d_spc_hardlimit = dm->dqb_bhardlimit;
2560 	di->d_spc_softlimit = dm->dqb_bsoftlimit;
2561 	di->d_ino_hardlimit = dm->dqb_ihardlimit;
2562 	di->d_ino_softlimit = dm->dqb_isoftlimit;
2563 	di->d_space = dm->dqb_curspace + dm->dqb_rsvspace;
2564 	di->d_ino_count = dm->dqb_curinodes;
2565 	di->d_spc_timer = dm->dqb_btime;
2566 	di->d_ino_timer = dm->dqb_itime;
2567 	spin_unlock(&dq_data_lock);
2568 }
2569 
2570 int dquot_get_dqblk(struct super_block *sb, struct kqid qid,
2571 		    struct qc_dqblk *di)
2572 {
2573 	struct dquot *dquot;
2574 
2575 	dquot = dqget(sb, qid);
2576 	if (IS_ERR(dquot))
2577 		return PTR_ERR(dquot);
2578 	do_get_dqblk(dquot, di);
2579 	dqput(dquot);
2580 
2581 	return 0;
2582 }
2583 EXPORT_SYMBOL(dquot_get_dqblk);
2584 
2585 int dquot_get_next_dqblk(struct super_block *sb, struct kqid *qid,
2586 			 struct qc_dqblk *di)
2587 {
2588 	struct dquot *dquot;
2589 	int err;
2590 
2591 	if (!sb->dq_op->get_next_id)
2592 		return -ENOSYS;
2593 	err = sb->dq_op->get_next_id(sb, qid);
2594 	if (err < 0)
2595 		return err;
2596 	dquot = dqget(sb, *qid);
2597 	if (IS_ERR(dquot))
2598 		return PTR_ERR(dquot);
2599 	do_get_dqblk(dquot, di);
2600 	dqput(dquot);
2601 
2602 	return 0;
2603 }
2604 EXPORT_SYMBOL(dquot_get_next_dqblk);
2605 
2606 #define VFS_QC_MASK \
2607 	(QC_SPACE | QC_SPC_SOFT | QC_SPC_HARD | \
2608 	 QC_INO_COUNT | QC_INO_SOFT | QC_INO_HARD | \
2609 	 QC_SPC_TIMER | QC_INO_TIMER)
2610 
2611 /* Generic routine for setting common part of quota structure */
2612 static int do_set_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2613 {
2614 	struct mem_dqblk *dm = &dquot->dq_dqb;
2615 	int check_blim = 0, check_ilim = 0;
2616 	struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
2617 
2618 	if (di->d_fieldmask & ~VFS_QC_MASK)
2619 		return -EINVAL;
2620 
2621 	if (((di->d_fieldmask & QC_SPC_SOFT) &&
2622 	     di->d_spc_softlimit > dqi->dqi_max_spc_limit) ||
2623 	    ((di->d_fieldmask & QC_SPC_HARD) &&
2624 	     di->d_spc_hardlimit > dqi->dqi_max_spc_limit) ||
2625 	    ((di->d_fieldmask & QC_INO_SOFT) &&
2626 	     (di->d_ino_softlimit > dqi->dqi_max_ino_limit)) ||
2627 	    ((di->d_fieldmask & QC_INO_HARD) &&
2628 	     (di->d_ino_hardlimit > dqi->dqi_max_ino_limit)))
2629 		return -ERANGE;
2630 
2631 	spin_lock(&dq_data_lock);
2632 	if (di->d_fieldmask & QC_SPACE) {
2633 		dm->dqb_curspace = di->d_space - dm->dqb_rsvspace;
2634 		check_blim = 1;
2635 		set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2636 	}
2637 
2638 	if (di->d_fieldmask & QC_SPC_SOFT)
2639 		dm->dqb_bsoftlimit = di->d_spc_softlimit;
2640 	if (di->d_fieldmask & QC_SPC_HARD)
2641 		dm->dqb_bhardlimit = di->d_spc_hardlimit;
2642 	if (di->d_fieldmask & (QC_SPC_SOFT | QC_SPC_HARD)) {
2643 		check_blim = 1;
2644 		set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2645 	}
2646 
2647 	if (di->d_fieldmask & QC_INO_COUNT) {
2648 		dm->dqb_curinodes = di->d_ino_count;
2649 		check_ilim = 1;
2650 		set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2651 	}
2652 
2653 	if (di->d_fieldmask & QC_INO_SOFT)
2654 		dm->dqb_isoftlimit = di->d_ino_softlimit;
2655 	if (di->d_fieldmask & QC_INO_HARD)
2656 		dm->dqb_ihardlimit = di->d_ino_hardlimit;
2657 	if (di->d_fieldmask & (QC_INO_SOFT | QC_INO_HARD)) {
2658 		check_ilim = 1;
2659 		set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2660 	}
2661 
2662 	if (di->d_fieldmask & QC_SPC_TIMER) {
2663 		dm->dqb_btime = di->d_spc_timer;
2664 		check_blim = 1;
2665 		set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2666 	}
2667 
2668 	if (di->d_fieldmask & QC_INO_TIMER) {
2669 		dm->dqb_itime = di->d_ino_timer;
2670 		check_ilim = 1;
2671 		set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2672 	}
2673 
2674 	if (check_blim) {
2675 		if (!dm->dqb_bsoftlimit ||
2676 		    dm->dqb_curspace < dm->dqb_bsoftlimit) {
2677 			dm->dqb_btime = 0;
2678 			clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2679 		} else if (!(di->d_fieldmask & QC_SPC_TIMER))
2680 			/* Set grace only if user hasn't provided his own... */
2681 			dm->dqb_btime = ktime_get_real_seconds() + dqi->dqi_bgrace;
2682 	}
2683 	if (check_ilim) {
2684 		if (!dm->dqb_isoftlimit ||
2685 		    dm->dqb_curinodes < dm->dqb_isoftlimit) {
2686 			dm->dqb_itime = 0;
2687 			clear_bit(DQ_INODES_B, &dquot->dq_flags);
2688 		} else if (!(di->d_fieldmask & QC_INO_TIMER))
2689 			/* Set grace only if user hasn't provided his own... */
2690 			dm->dqb_itime = ktime_get_real_seconds() + dqi->dqi_igrace;
2691 	}
2692 	if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2693 	    dm->dqb_isoftlimit)
2694 		clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2695 	else
2696 		set_bit(DQ_FAKE_B, &dquot->dq_flags);
2697 	spin_unlock(&dq_data_lock);
2698 	mark_dquot_dirty(dquot);
2699 
2700 	return 0;
2701 }
2702 
2703 int dquot_set_dqblk(struct super_block *sb, struct kqid qid,
2704 		  struct qc_dqblk *di)
2705 {
2706 	struct dquot *dquot;
2707 	int rc;
2708 
2709 	dquot = dqget(sb, qid);
2710 	if (IS_ERR(dquot)) {
2711 		rc = PTR_ERR(dquot);
2712 		goto out;
2713 	}
2714 	rc = do_set_dqblk(dquot, di);
2715 	dqput(dquot);
2716 out:
2717 	return rc;
2718 }
2719 EXPORT_SYMBOL(dquot_set_dqblk);
2720 
2721 /* Generic routine for getting common part of quota file information */
2722 int dquot_get_state(struct super_block *sb, struct qc_state *state)
2723 {
2724 	struct mem_dqinfo *mi;
2725 	struct qc_type_state *tstate;
2726 	struct quota_info *dqopt = sb_dqopt(sb);
2727 	int type;
2728 
2729 	memset(state, 0, sizeof(*state));
2730 	for (type = 0; type < MAXQUOTAS; type++) {
2731 		if (!sb_has_quota_active(sb, type))
2732 			continue;
2733 		tstate = state->s_state + type;
2734 		mi = sb_dqopt(sb)->info + type;
2735 		tstate->flags = QCI_ACCT_ENABLED;
2736 		spin_lock(&dq_data_lock);
2737 		if (mi->dqi_flags & DQF_SYS_FILE)
2738 			tstate->flags |= QCI_SYSFILE;
2739 		if (mi->dqi_flags & DQF_ROOT_SQUASH)
2740 			tstate->flags |= QCI_ROOT_SQUASH;
2741 		if (sb_has_quota_limits_enabled(sb, type))
2742 			tstate->flags |= QCI_LIMITS_ENFORCED;
2743 		tstate->spc_timelimit = mi->dqi_bgrace;
2744 		tstate->ino_timelimit = mi->dqi_igrace;
2745 		tstate->ino = dqopt->files[type]->i_ino;
2746 		tstate->blocks = dqopt->files[type]->i_blocks;
2747 		tstate->nextents = 1;	/* We don't know... */
2748 		spin_unlock(&dq_data_lock);
2749 	}
2750 	return 0;
2751 }
2752 EXPORT_SYMBOL(dquot_get_state);
2753 
2754 /* Generic routine for setting common part of quota file information */
2755 int dquot_set_dqinfo(struct super_block *sb, int type, struct qc_info *ii)
2756 {
2757 	struct mem_dqinfo *mi;
2758 	int err = 0;
2759 
2760 	if ((ii->i_fieldmask & QC_WARNS_MASK) ||
2761 	    (ii->i_fieldmask & QC_RT_SPC_TIMER))
2762 		return -EINVAL;
2763 	if (!sb_has_quota_active(sb, type))
2764 		return -ESRCH;
2765 	mi = sb_dqopt(sb)->info + type;
2766 	if (ii->i_fieldmask & QC_FLAGS) {
2767 		if ((ii->i_flags & QCI_ROOT_SQUASH &&
2768 		     mi->dqi_format->qf_fmt_id != QFMT_VFS_OLD))
2769 			return -EINVAL;
2770 	}
2771 	spin_lock(&dq_data_lock);
2772 	if (ii->i_fieldmask & QC_SPC_TIMER)
2773 		mi->dqi_bgrace = ii->i_spc_timelimit;
2774 	if (ii->i_fieldmask & QC_INO_TIMER)
2775 		mi->dqi_igrace = ii->i_ino_timelimit;
2776 	if (ii->i_fieldmask & QC_FLAGS) {
2777 		if (ii->i_flags & QCI_ROOT_SQUASH)
2778 			mi->dqi_flags |= DQF_ROOT_SQUASH;
2779 		else
2780 			mi->dqi_flags &= ~DQF_ROOT_SQUASH;
2781 	}
2782 	spin_unlock(&dq_data_lock);
2783 	mark_info_dirty(sb, type);
2784 	/* Force write to disk */
2785 	sb->dq_op->write_info(sb, type);
2786 	return err;
2787 }
2788 EXPORT_SYMBOL(dquot_set_dqinfo);
2789 
2790 const struct quotactl_ops dquot_quotactl_sysfile_ops = {
2791 	.quota_enable	= dquot_quota_enable,
2792 	.quota_disable	= dquot_quota_disable,
2793 	.quota_sync	= dquot_quota_sync,
2794 	.get_state	= dquot_get_state,
2795 	.set_info	= dquot_set_dqinfo,
2796 	.get_dqblk	= dquot_get_dqblk,
2797 	.get_nextdqblk	= dquot_get_next_dqblk,
2798 	.set_dqblk	= dquot_set_dqblk
2799 };
2800 EXPORT_SYMBOL(dquot_quotactl_sysfile_ops);
2801 
2802 static int do_proc_dqstats(struct ctl_table *table, int write,
2803 		     void __user *buffer, size_t *lenp, loff_t *ppos)
2804 {
2805 	unsigned int type = (int *)table->data - dqstats.stat;
2806 
2807 	/* Update global table */
2808 	dqstats.stat[type] =
2809 			percpu_counter_sum_positive(&dqstats.counter[type]);
2810 	return proc_dointvec(table, write, buffer, lenp, ppos);
2811 }
2812 
2813 static struct ctl_table fs_dqstats_table[] = {
2814 	{
2815 		.procname	= "lookups",
2816 		.data		= &dqstats.stat[DQST_LOOKUPS],
2817 		.maxlen		= sizeof(int),
2818 		.mode		= 0444,
2819 		.proc_handler	= do_proc_dqstats,
2820 	},
2821 	{
2822 		.procname	= "drops",
2823 		.data		= &dqstats.stat[DQST_DROPS],
2824 		.maxlen		= sizeof(int),
2825 		.mode		= 0444,
2826 		.proc_handler	= do_proc_dqstats,
2827 	},
2828 	{
2829 		.procname	= "reads",
2830 		.data		= &dqstats.stat[DQST_READS],
2831 		.maxlen		= sizeof(int),
2832 		.mode		= 0444,
2833 		.proc_handler	= do_proc_dqstats,
2834 	},
2835 	{
2836 		.procname	= "writes",
2837 		.data		= &dqstats.stat[DQST_WRITES],
2838 		.maxlen		= sizeof(int),
2839 		.mode		= 0444,
2840 		.proc_handler	= do_proc_dqstats,
2841 	},
2842 	{
2843 		.procname	= "cache_hits",
2844 		.data		= &dqstats.stat[DQST_CACHE_HITS],
2845 		.maxlen		= sizeof(int),
2846 		.mode		= 0444,
2847 		.proc_handler	= do_proc_dqstats,
2848 	},
2849 	{
2850 		.procname	= "allocated_dquots",
2851 		.data		= &dqstats.stat[DQST_ALLOC_DQUOTS],
2852 		.maxlen		= sizeof(int),
2853 		.mode		= 0444,
2854 		.proc_handler	= do_proc_dqstats,
2855 	},
2856 	{
2857 		.procname	= "free_dquots",
2858 		.data		= &dqstats.stat[DQST_FREE_DQUOTS],
2859 		.maxlen		= sizeof(int),
2860 		.mode		= 0444,
2861 		.proc_handler	= do_proc_dqstats,
2862 	},
2863 	{
2864 		.procname	= "syncs",
2865 		.data		= &dqstats.stat[DQST_SYNCS],
2866 		.maxlen		= sizeof(int),
2867 		.mode		= 0444,
2868 		.proc_handler	= do_proc_dqstats,
2869 	},
2870 #ifdef CONFIG_PRINT_QUOTA_WARNING
2871 	{
2872 		.procname	= "warnings",
2873 		.data		= &flag_print_warnings,
2874 		.maxlen		= sizeof(int),
2875 		.mode		= 0644,
2876 		.proc_handler	= proc_dointvec,
2877 	},
2878 #endif
2879 	{ },
2880 };
2881 
2882 static struct ctl_table fs_table[] = {
2883 	{
2884 		.procname	= "quota",
2885 		.mode		= 0555,
2886 		.child		= fs_dqstats_table,
2887 	},
2888 	{ },
2889 };
2890 
2891 static struct ctl_table sys_table[] = {
2892 	{
2893 		.procname	= "fs",
2894 		.mode		= 0555,
2895 		.child		= fs_table,
2896 	},
2897 	{ },
2898 };
2899 
2900 static int __init dquot_init(void)
2901 {
2902 	int i, ret;
2903 	unsigned long nr_hash, order;
2904 
2905 	printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2906 
2907 	register_sysctl_table(sys_table);
2908 
2909 	dquot_cachep = kmem_cache_create("dquot",
2910 			sizeof(struct dquot), sizeof(unsigned long) * 4,
2911 			(SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2912 				SLAB_MEM_SPREAD|SLAB_PANIC),
2913 			NULL);
2914 
2915 	order = 0;
2916 	dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
2917 	if (!dquot_hash)
2918 		panic("Cannot create dquot hash table");
2919 
2920 	for (i = 0; i < _DQST_DQSTAT_LAST; i++) {
2921 		ret = percpu_counter_init(&dqstats.counter[i], 0, GFP_KERNEL);
2922 		if (ret)
2923 			panic("Cannot create dquot stat counters");
2924 	}
2925 
2926 	/* Find power-of-two hlist_heads which can fit into allocation */
2927 	nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2928 	dq_hash_bits = 0;
2929 	do {
2930 		dq_hash_bits++;
2931 	} while (nr_hash >> dq_hash_bits);
2932 	dq_hash_bits--;
2933 
2934 	nr_hash = 1UL << dq_hash_bits;
2935 	dq_hash_mask = nr_hash - 1;
2936 	for (i = 0; i < nr_hash; i++)
2937 		INIT_HLIST_HEAD(dquot_hash + i);
2938 
2939 	pr_info("VFS: Dquot-cache hash table entries: %ld (order %ld,"
2940 		" %ld bytes)\n", nr_hash, order, (PAGE_SIZE << order));
2941 
2942 	register_shrinker(&dqcache_shrinker);
2943 
2944 	return 0;
2945 }
2946 fs_initcall(dquot_init);
2947