xref: /kernel/linux/linux-6.6/fs/quota/dquot.c (revision 62306a36)
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Implementation of the diskquota system for the LINUX operating system. QUOTA
4 * is implemented using the BSD system call interface as the means of
5 * communication with the user level. This file contains the generic routines
6 * called by the different filesystems on allocation of an inode or block.
7 * These routines take care of the administration needed to have a consistent
8 * diskquota tracking system. The ideas of both user and group quotas are based
9 * on the Melbourne quota system as used on BSD derived systems. The internal
10 * implementation is based on one of the several variants of the LINUX
11 * inode-subsystem with added complexity of the diskquota system.
12 *
13 * Author:	Marco van Wieringen <mvw@planets.elm.net>
14 *
15 * Fixes:   Dmitry Gorodchanin <pgmdsg@ibi.com>, 11 Feb 96
16 *
17 *		Revised list management to avoid races
18 *		-- Bill Hawes, <whawes@star.net>, 9/98
19 *
20 *		Fixed races in dquot_transfer(), dqget() and dquot_alloc_...().
21 *		As the consequence the locking was moved from dquot_decr_...(),
22 *		dquot_incr_...() to calling functions.
23 *		invalidate_dquots() now writes modified dquots.
24 *		Serialized quota_off() and quota_on() for mount point.
25 *		Fixed a few bugs in grow_dquots().
26 *		Fixed deadlock in write_dquot() - we no longer account quotas on
27 *		quota files
28 *		remove_dquot_ref() moved to inode.c - it now traverses through inodes
29 *		add_dquot_ref() restarts after blocking
30 *		Added check for bogus uid and fixed check for group in quotactl.
31 *		Jan Kara, <jack@suse.cz>, sponsored by SuSE CR, 10-11/99
32 *
33 *		Used struct list_head instead of own list struct
34 *		Invalidation of referenced dquots is no longer possible
35 *		Improved free_dquots list management
36 *		Quota and i_blocks are now updated in one place to avoid races
37 *		Warnings are now delayed so we won't block in critical section
38 *		Write updated not to require dquot lock
39 *		Jan Kara, <jack@suse.cz>, 9/2000
40 *
41 *		Added dynamic quota structure allocation
42 *		Jan Kara <jack@suse.cz> 12/2000
43 *
44 *		Rewritten quota interface. Implemented new quota format and
45 *		formats registering.
46 *		Jan Kara, <jack@suse.cz>, 2001,2002
47 *
48 *		New SMP locking.
49 *		Jan Kara, <jack@suse.cz>, 10/2002
50 *
51 *		Added journalled quota support, fix lock inversion problems
52 *		Jan Kara, <jack@suse.cz>, 2003,2004
53 *
54 * (C) Copyright 1994 - 1997 Marco van Wieringen
55 */
56
57#include <linux/errno.h>
58#include <linux/kernel.h>
59#include <linux/fs.h>
60#include <linux/mount.h>
61#include <linux/mm.h>
62#include <linux/time.h>
63#include <linux/types.h>
64#include <linux/string.h>
65#include <linux/fcntl.h>
66#include <linux/stat.h>
67#include <linux/tty.h>
68#include <linux/file.h>
69#include <linux/slab.h>
70#include <linux/sysctl.h>
71#include <linux/init.h>
72#include <linux/module.h>
73#include <linux/proc_fs.h>
74#include <linux/security.h>
75#include <linux/sched.h>
76#include <linux/cred.h>
77#include <linux/kmod.h>
78#include <linux/namei.h>
79#include <linux/capability.h>
80#include <linux/quotaops.h>
81#include <linux/blkdev.h>
82#include <linux/sched/mm.h>
83#include "../internal.h" /* ugh */
84
85#include <linux/uaccess.h>
86
87/*
88 * There are five quota SMP locks:
89 * * dq_list_lock protects all lists with quotas and quota formats.
90 * * dquot->dq_dqb_lock protects data from dq_dqb
91 * * inode->i_lock protects inode->i_blocks, i_bytes and also guards
92 *   consistency of dquot->dq_dqb with inode->i_blocks, i_bytes so that
93 *   dquot_transfer() can stabilize amount it transfers
94 * * dq_data_lock protects mem_dqinfo structures and modifications of dquot
95 *   pointers in the inode
96 * * dq_state_lock protects modifications of quota state (on quotaon and
97 *   quotaoff) and readers who care about latest values take it as well.
98 *
99 * The spinlock ordering is hence:
100 *   dq_data_lock > dq_list_lock > i_lock > dquot->dq_dqb_lock,
101 *   dq_list_lock > dq_state_lock
102 *
103 * Note that some things (eg. sb pointer, type, id) doesn't change during
104 * the life of the dquot structure and so needn't to be protected by a lock
105 *
106 * Operation accessing dquots via inode pointers are protected by dquot_srcu.
107 * Operation of reading pointer needs srcu_read_lock(&dquot_srcu), and
108 * synchronize_srcu(&dquot_srcu) is called after clearing pointers from
109 * inode and before dropping dquot references to avoid use of dquots after
110 * they are freed. dq_data_lock is used to serialize the pointer setting and
111 * clearing operations.
112 * Special care needs to be taken about S_NOQUOTA inode flag (marking that
113 * inode is a quota file). Functions adding pointers from inode to dquots have
114 * to check this flag under dq_data_lock and then (if S_NOQUOTA is not set) they
115 * have to do all pointer modifications before dropping dq_data_lock. This makes
116 * sure they cannot race with quotaon which first sets S_NOQUOTA flag and
117 * then drops all pointers to dquots from an inode.
118 *
119 * Each dquot has its dq_lock mutex.  Dquot is locked when it is being read to
120 * memory (or space for it is being allocated) on the first dqget(), when it is
121 * being written out, and when it is being released on the last dqput(). The
122 * allocation and release operations are serialized by the dq_lock and by
123 * checking the use count in dquot_release().
124 *
125 * Lock ordering (including related VFS locks) is the following:
126 *   s_umount > i_mutex > journal_lock > dquot->dq_lock > dqio_sem
127 */
128
129static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_list_lock);
130static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_state_lock);
131__cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_data_lock);
132EXPORT_SYMBOL(dq_data_lock);
133DEFINE_STATIC_SRCU(dquot_srcu);
134
135static DECLARE_WAIT_QUEUE_HEAD(dquot_ref_wq);
136
137void __quota_error(struct super_block *sb, const char *func,
138		   const char *fmt, ...)
139{
140	if (printk_ratelimit()) {
141		va_list args;
142		struct va_format vaf;
143
144		va_start(args, fmt);
145
146		vaf.fmt = fmt;
147		vaf.va = &args;
148
149		printk(KERN_ERR "Quota error (device %s): %s: %pV\n",
150		       sb->s_id, func, &vaf);
151
152		va_end(args);
153	}
154}
155EXPORT_SYMBOL(__quota_error);
156
157#if defined(CONFIG_QUOTA_DEBUG) || defined(CONFIG_PRINT_QUOTA_WARNING)
158static char *quotatypes[] = INITQFNAMES;
159#endif
160static struct quota_format_type *quota_formats;	/* List of registered formats */
161static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
162
163/* SLAB cache for dquot structures */
164static struct kmem_cache *dquot_cachep;
165
166int register_quota_format(struct quota_format_type *fmt)
167{
168	spin_lock(&dq_list_lock);
169	fmt->qf_next = quota_formats;
170	quota_formats = fmt;
171	spin_unlock(&dq_list_lock);
172	return 0;
173}
174EXPORT_SYMBOL(register_quota_format);
175
176void unregister_quota_format(struct quota_format_type *fmt)
177{
178	struct quota_format_type **actqf;
179
180	spin_lock(&dq_list_lock);
181	for (actqf = &quota_formats; *actqf && *actqf != fmt;
182	     actqf = &(*actqf)->qf_next)
183		;
184	if (*actqf)
185		*actqf = (*actqf)->qf_next;
186	spin_unlock(&dq_list_lock);
187}
188EXPORT_SYMBOL(unregister_quota_format);
189
190static struct quota_format_type *find_quota_format(int id)
191{
192	struct quota_format_type *actqf;
193
194	spin_lock(&dq_list_lock);
195	for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
196	     actqf = actqf->qf_next)
197		;
198	if (!actqf || !try_module_get(actqf->qf_owner)) {
199		int qm;
200
201		spin_unlock(&dq_list_lock);
202
203		for (qm = 0; module_names[qm].qm_fmt_id &&
204			     module_names[qm].qm_fmt_id != id; qm++)
205			;
206		if (!module_names[qm].qm_fmt_id ||
207		    request_module(module_names[qm].qm_mod_name))
208			return NULL;
209
210		spin_lock(&dq_list_lock);
211		for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
212		     actqf = actqf->qf_next)
213			;
214		if (actqf && !try_module_get(actqf->qf_owner))
215			actqf = NULL;
216	}
217	spin_unlock(&dq_list_lock);
218	return actqf;
219}
220
221static void put_quota_format(struct quota_format_type *fmt)
222{
223	module_put(fmt->qf_owner);
224}
225
226/*
227 * Dquot List Management:
228 * The quota code uses five lists for dquot management: the inuse_list,
229 * releasing_dquots, free_dquots, dqi_dirty_list, and dquot_hash[] array.
230 * A single dquot structure may be on some of those lists, depending on
231 * its current state.
232 *
233 * All dquots are placed to the end of inuse_list when first created, and this
234 * list is used for invalidate operation, which must look at every dquot.
235 *
236 * When the last reference of a dquot is dropped, the dquot is added to
237 * releasing_dquots. We'll then queue work item which will call
238 * synchronize_srcu() and after that perform the final cleanup of all the
239 * dquots on the list. Each cleaned up dquot is moved to free_dquots list.
240 * Both releasing_dquots and free_dquots use the dq_free list_head in the dquot
241 * struct.
242 *
243 * Unused and cleaned up dquots are in the free_dquots list and this list is
244 * searched whenever we need an available dquot. Dquots are removed from the
245 * list as soon as they are used again and dqstats.free_dquots gives the number
246 * of dquots on the list. When dquot is invalidated it's completely released
247 * from memory.
248 *
249 * Dirty dquots are added to the dqi_dirty_list of quota_info when mark
250 * dirtied, and this list is searched when writing dirty dquots back to
251 * quota file. Note that some filesystems do dirty dquot tracking on their
252 * own (e.g. in a journal) and thus don't use dqi_dirty_list.
253 *
254 * Dquots with a specific identity (device, type and id) are placed on
255 * one of the dquot_hash[] hash chains. The provides an efficient search
256 * mechanism to locate a specific dquot.
257 */
258
259static LIST_HEAD(inuse_list);
260static LIST_HEAD(free_dquots);
261static LIST_HEAD(releasing_dquots);
262static unsigned int dq_hash_bits, dq_hash_mask;
263static struct hlist_head *dquot_hash;
264
265struct dqstats dqstats;
266EXPORT_SYMBOL(dqstats);
267
268static qsize_t inode_get_rsv_space(struct inode *inode);
269static qsize_t __inode_get_rsv_space(struct inode *inode);
270static int __dquot_initialize(struct inode *inode, int type);
271
272static void quota_release_workfn(struct work_struct *work);
273static DECLARE_DELAYED_WORK(quota_release_work, quota_release_workfn);
274
275static inline unsigned int
276hashfn(const struct super_block *sb, struct kqid qid)
277{
278	unsigned int id = from_kqid(&init_user_ns, qid);
279	int type = qid.type;
280	unsigned long tmp;
281
282	tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
283	return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
284}
285
286/*
287 * Following list functions expect dq_list_lock to be held
288 */
289static inline void insert_dquot_hash(struct dquot *dquot)
290{
291	struct hlist_head *head;
292	head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id);
293	hlist_add_head(&dquot->dq_hash, head);
294}
295
296static inline void remove_dquot_hash(struct dquot *dquot)
297{
298	hlist_del_init(&dquot->dq_hash);
299}
300
301static struct dquot *find_dquot(unsigned int hashent, struct super_block *sb,
302				struct kqid qid)
303{
304	struct dquot *dquot;
305
306	hlist_for_each_entry(dquot, dquot_hash+hashent, dq_hash)
307		if (dquot->dq_sb == sb && qid_eq(dquot->dq_id, qid))
308			return dquot;
309
310	return NULL;
311}
312
313/* Add a dquot to the tail of the free list */
314static inline void put_dquot_last(struct dquot *dquot)
315{
316	list_add_tail(&dquot->dq_free, &free_dquots);
317	dqstats_inc(DQST_FREE_DQUOTS);
318}
319
320static inline void put_releasing_dquots(struct dquot *dquot)
321{
322	list_add_tail(&dquot->dq_free, &releasing_dquots);
323	set_bit(DQ_RELEASING_B, &dquot->dq_flags);
324}
325
326static inline void remove_free_dquot(struct dquot *dquot)
327{
328	if (list_empty(&dquot->dq_free))
329		return;
330	list_del_init(&dquot->dq_free);
331	if (!test_bit(DQ_RELEASING_B, &dquot->dq_flags))
332		dqstats_dec(DQST_FREE_DQUOTS);
333	else
334		clear_bit(DQ_RELEASING_B, &dquot->dq_flags);
335}
336
337static inline void put_inuse(struct dquot *dquot)
338{
339	/* We add to the back of inuse list so we don't have to restart
340	 * when traversing this list and we block */
341	list_add_tail(&dquot->dq_inuse, &inuse_list);
342	dqstats_inc(DQST_ALLOC_DQUOTS);
343}
344
345static inline void remove_inuse(struct dquot *dquot)
346{
347	dqstats_dec(DQST_ALLOC_DQUOTS);
348	list_del(&dquot->dq_inuse);
349}
350/*
351 * End of list functions needing dq_list_lock
352 */
353
354static void wait_on_dquot(struct dquot *dquot)
355{
356	mutex_lock(&dquot->dq_lock);
357	mutex_unlock(&dquot->dq_lock);
358}
359
360static inline int dquot_active(struct dquot *dquot)
361{
362	return test_bit(DQ_ACTIVE_B, &dquot->dq_flags);
363}
364
365static inline int dquot_dirty(struct dquot *dquot)
366{
367	return test_bit(DQ_MOD_B, &dquot->dq_flags);
368}
369
370static inline int mark_dquot_dirty(struct dquot *dquot)
371{
372	return dquot->dq_sb->dq_op->mark_dirty(dquot);
373}
374
375/* Mark dquot dirty in atomic manner, and return it's old dirty flag state */
376int dquot_mark_dquot_dirty(struct dquot *dquot)
377{
378	int ret = 1;
379
380	if (!dquot_active(dquot))
381		return 0;
382
383	if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NOLIST_DIRTY)
384		return test_and_set_bit(DQ_MOD_B, &dquot->dq_flags);
385
386	/* If quota is dirty already, we don't have to acquire dq_list_lock */
387	if (dquot_dirty(dquot))
388		return 1;
389
390	spin_lock(&dq_list_lock);
391	if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags)) {
392		list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
393				info[dquot->dq_id.type].dqi_dirty_list);
394		ret = 0;
395	}
396	spin_unlock(&dq_list_lock);
397	return ret;
398}
399EXPORT_SYMBOL(dquot_mark_dquot_dirty);
400
401/* Dirtify all the dquots - this can block when journalling */
402static inline int mark_all_dquot_dirty(struct dquot __rcu * const *dquots)
403{
404	int ret, err, cnt;
405	struct dquot *dquot;
406
407	ret = err = 0;
408	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
409		dquot = srcu_dereference(dquots[cnt], &dquot_srcu);
410		if (dquot)
411			/* Even in case of error we have to continue */
412			ret = mark_dquot_dirty(dquot);
413		if (!err)
414			err = ret;
415	}
416	return err;
417}
418
419static inline void dqput_all(struct dquot **dquot)
420{
421	unsigned int cnt;
422
423	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
424		dqput(dquot[cnt]);
425}
426
427static inline int clear_dquot_dirty(struct dquot *dquot)
428{
429	if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NOLIST_DIRTY)
430		return test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags);
431
432	spin_lock(&dq_list_lock);
433	if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags)) {
434		spin_unlock(&dq_list_lock);
435		return 0;
436	}
437	list_del_init(&dquot->dq_dirty);
438	spin_unlock(&dq_list_lock);
439	return 1;
440}
441
442void mark_info_dirty(struct super_block *sb, int type)
443{
444	spin_lock(&dq_data_lock);
445	sb_dqopt(sb)->info[type].dqi_flags |= DQF_INFO_DIRTY;
446	spin_unlock(&dq_data_lock);
447}
448EXPORT_SYMBOL(mark_info_dirty);
449
450/*
451 *	Read dquot from disk and alloc space for it
452 */
453
454int dquot_acquire(struct dquot *dquot)
455{
456	int ret = 0, ret2 = 0;
457	unsigned int memalloc;
458	struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
459
460	mutex_lock(&dquot->dq_lock);
461	memalloc = memalloc_nofs_save();
462	if (!test_bit(DQ_READ_B, &dquot->dq_flags)) {
463		ret = dqopt->ops[dquot->dq_id.type]->read_dqblk(dquot);
464		if (ret < 0)
465			goto out_iolock;
466	}
467	/* Make sure flags update is visible after dquot has been filled */
468	smp_mb__before_atomic();
469	set_bit(DQ_READ_B, &dquot->dq_flags);
470	/* Instantiate dquot if needed */
471	if (!dquot_active(dquot) && !dquot->dq_off) {
472		ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
473		/* Write the info if needed */
474		if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
475			ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
476					dquot->dq_sb, dquot->dq_id.type);
477		}
478		if (ret < 0)
479			goto out_iolock;
480		if (ret2 < 0) {
481			ret = ret2;
482			goto out_iolock;
483		}
484	}
485	/*
486	 * Make sure flags update is visible after on-disk struct has been
487	 * allocated. Paired with smp_rmb() in dqget().
488	 */
489	smp_mb__before_atomic();
490	set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
491out_iolock:
492	memalloc_nofs_restore(memalloc);
493	mutex_unlock(&dquot->dq_lock);
494	return ret;
495}
496EXPORT_SYMBOL(dquot_acquire);
497
498/*
499 *	Write dquot to disk
500 */
501int dquot_commit(struct dquot *dquot)
502{
503	int ret = 0;
504	unsigned int memalloc;
505	struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
506
507	mutex_lock(&dquot->dq_lock);
508	memalloc = memalloc_nofs_save();
509	if (!clear_dquot_dirty(dquot))
510		goto out_lock;
511	/* Inactive dquot can be only if there was error during read/init
512	 * => we have better not writing it */
513	if (dquot_active(dquot))
514		ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
515	else
516		ret = -EIO;
517out_lock:
518	memalloc_nofs_restore(memalloc);
519	mutex_unlock(&dquot->dq_lock);
520	return ret;
521}
522EXPORT_SYMBOL(dquot_commit);
523
524/*
525 *	Release dquot
526 */
527int dquot_release(struct dquot *dquot)
528{
529	int ret = 0, ret2 = 0;
530	unsigned int memalloc;
531	struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
532
533	mutex_lock(&dquot->dq_lock);
534	memalloc = memalloc_nofs_save();
535	/* Check whether we are not racing with some other dqget() */
536	if (dquot_is_busy(dquot))
537		goto out_dqlock;
538	if (dqopt->ops[dquot->dq_id.type]->release_dqblk) {
539		ret = dqopt->ops[dquot->dq_id.type]->release_dqblk(dquot);
540		/* Write the info */
541		if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
542			ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
543						dquot->dq_sb, dquot->dq_id.type);
544		}
545		if (ret >= 0)
546			ret = ret2;
547	}
548	clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
549out_dqlock:
550	memalloc_nofs_restore(memalloc);
551	mutex_unlock(&dquot->dq_lock);
552	return ret;
553}
554EXPORT_SYMBOL(dquot_release);
555
556void dquot_destroy(struct dquot *dquot)
557{
558	kmem_cache_free(dquot_cachep, dquot);
559}
560EXPORT_SYMBOL(dquot_destroy);
561
562static inline void do_destroy_dquot(struct dquot *dquot)
563{
564	dquot->dq_sb->dq_op->destroy_dquot(dquot);
565}
566
567/* Invalidate all dquots on the list. Note that this function is called after
568 * quota is disabled and pointers from inodes removed so there cannot be new
569 * quota users. There can still be some users of quotas due to inodes being
570 * just deleted or pruned by prune_icache() (those are not attached to any
571 * list) or parallel quotactl call. We have to wait for such users.
572 */
573static void invalidate_dquots(struct super_block *sb, int type)
574{
575	struct dquot *dquot, *tmp;
576
577restart:
578	flush_delayed_work(&quota_release_work);
579
580	spin_lock(&dq_list_lock);
581	list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
582		if (dquot->dq_sb != sb)
583			continue;
584		if (dquot->dq_id.type != type)
585			continue;
586		/* Wait for dquot users */
587		if (atomic_read(&dquot->dq_count)) {
588			atomic_inc(&dquot->dq_count);
589			spin_unlock(&dq_list_lock);
590			/*
591			 * Once dqput() wakes us up, we know it's time to free
592			 * the dquot.
593			 * IMPORTANT: we rely on the fact that there is always
594			 * at most one process waiting for dquot to free.
595			 * Otherwise dq_count would be > 1 and we would never
596			 * wake up.
597			 */
598			wait_event(dquot_ref_wq,
599				   atomic_read(&dquot->dq_count) == 1);
600			dqput(dquot);
601			/* At this moment dquot() need not exist (it could be
602			 * reclaimed by prune_dqcache(). Hence we must
603			 * restart. */
604			goto restart;
605		}
606		/*
607		 * The last user already dropped its reference but dquot didn't
608		 * get fully cleaned up yet. Restart the scan which flushes the
609		 * work cleaning up released dquots.
610		 */
611		if (test_bit(DQ_RELEASING_B, &dquot->dq_flags)) {
612			spin_unlock(&dq_list_lock);
613			goto restart;
614		}
615		/*
616		 * Quota now has no users and it has been written on last
617		 * dqput()
618		 */
619		remove_dquot_hash(dquot);
620		remove_free_dquot(dquot);
621		remove_inuse(dquot);
622		do_destroy_dquot(dquot);
623	}
624	spin_unlock(&dq_list_lock);
625}
626
627/* Call callback for every active dquot on given filesystem */
628int dquot_scan_active(struct super_block *sb,
629		      int (*fn)(struct dquot *dquot, unsigned long priv),
630		      unsigned long priv)
631{
632	struct dquot *dquot, *old_dquot = NULL;
633	int ret = 0;
634
635	WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
636
637	spin_lock(&dq_list_lock);
638	list_for_each_entry(dquot, &inuse_list, dq_inuse) {
639		if (!dquot_active(dquot))
640			continue;
641		if (dquot->dq_sb != sb)
642			continue;
643		/* Now we have active dquot so we can just increase use count */
644		atomic_inc(&dquot->dq_count);
645		spin_unlock(&dq_list_lock);
646		dqput(old_dquot);
647		old_dquot = dquot;
648		/*
649		 * ->release_dquot() can be racing with us. Our reference
650		 * protects us from new calls to it so just wait for any
651		 * outstanding call and recheck the DQ_ACTIVE_B after that.
652		 */
653		wait_on_dquot(dquot);
654		if (dquot_active(dquot)) {
655			ret = fn(dquot, priv);
656			if (ret < 0)
657				goto out;
658		}
659		spin_lock(&dq_list_lock);
660		/* We are safe to continue now because our dquot could not
661		 * be moved out of the inuse list while we hold the reference */
662	}
663	spin_unlock(&dq_list_lock);
664out:
665	dqput(old_dquot);
666	return ret;
667}
668EXPORT_SYMBOL(dquot_scan_active);
669
670static inline int dquot_write_dquot(struct dquot *dquot)
671{
672	int ret = dquot->dq_sb->dq_op->write_dquot(dquot);
673	if (ret < 0) {
674		quota_error(dquot->dq_sb, "Can't write quota structure "
675			    "(error %d). Quota may get out of sync!", ret);
676		/* Clear dirty bit anyway to avoid infinite loop. */
677		clear_dquot_dirty(dquot);
678	}
679	return ret;
680}
681
682/* Write all dquot structures to quota files */
683int dquot_writeback_dquots(struct super_block *sb, int type)
684{
685	struct list_head dirty;
686	struct dquot *dquot;
687	struct quota_info *dqopt = sb_dqopt(sb);
688	int cnt;
689	int err, ret = 0;
690
691	WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
692
693	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
694		if (type != -1 && cnt != type)
695			continue;
696		if (!sb_has_quota_active(sb, cnt))
697			continue;
698		spin_lock(&dq_list_lock);
699		/* Move list away to avoid livelock. */
700		list_replace_init(&dqopt->info[cnt].dqi_dirty_list, &dirty);
701		while (!list_empty(&dirty)) {
702			dquot = list_first_entry(&dirty, struct dquot,
703						 dq_dirty);
704
705			WARN_ON(!dquot_active(dquot));
706			/* If the dquot is releasing we should not touch it */
707			if (test_bit(DQ_RELEASING_B, &dquot->dq_flags)) {
708				spin_unlock(&dq_list_lock);
709				flush_delayed_work(&quota_release_work);
710				spin_lock(&dq_list_lock);
711				continue;
712			}
713
714			/* Now we have active dquot from which someone is
715 			 * holding reference so we can safely just increase
716			 * use count */
717			dqgrab(dquot);
718			spin_unlock(&dq_list_lock);
719			err = dquot_write_dquot(dquot);
720			if (err && !ret)
721				ret = err;
722			dqput(dquot);
723			spin_lock(&dq_list_lock);
724		}
725		spin_unlock(&dq_list_lock);
726	}
727
728	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
729		if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
730		    && info_dirty(&dqopt->info[cnt]))
731			sb->dq_op->write_info(sb, cnt);
732	dqstats_inc(DQST_SYNCS);
733
734	return ret;
735}
736EXPORT_SYMBOL(dquot_writeback_dquots);
737
738/* Write all dquot structures to disk and make them visible from userspace */
739int dquot_quota_sync(struct super_block *sb, int type)
740{
741	struct quota_info *dqopt = sb_dqopt(sb);
742	int cnt;
743	int ret;
744
745	ret = dquot_writeback_dquots(sb, type);
746	if (ret)
747		return ret;
748	if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
749		return 0;
750
751	/* This is not very clever (and fast) but currently I don't know about
752	 * any other simple way of getting quota data to disk and we must get
753	 * them there for userspace to be visible... */
754	if (sb->s_op->sync_fs) {
755		ret = sb->s_op->sync_fs(sb, 1);
756		if (ret)
757			return ret;
758	}
759	ret = sync_blockdev(sb->s_bdev);
760	if (ret)
761		return ret;
762
763	/*
764	 * Now when everything is written we can discard the pagecache so
765	 * that userspace sees the changes.
766	 */
767	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
768		if (type != -1 && cnt != type)
769			continue;
770		if (!sb_has_quota_active(sb, cnt))
771			continue;
772		inode_lock(dqopt->files[cnt]);
773		truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
774		inode_unlock(dqopt->files[cnt]);
775	}
776
777	return 0;
778}
779EXPORT_SYMBOL(dquot_quota_sync);
780
781static unsigned long
782dqcache_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
783{
784	struct dquot *dquot;
785	unsigned long freed = 0;
786
787	spin_lock(&dq_list_lock);
788	while (!list_empty(&free_dquots) && sc->nr_to_scan) {
789		dquot = list_first_entry(&free_dquots, struct dquot, dq_free);
790		remove_dquot_hash(dquot);
791		remove_free_dquot(dquot);
792		remove_inuse(dquot);
793		do_destroy_dquot(dquot);
794		sc->nr_to_scan--;
795		freed++;
796	}
797	spin_unlock(&dq_list_lock);
798	return freed;
799}
800
801static unsigned long
802dqcache_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
803{
804	return vfs_pressure_ratio(
805	percpu_counter_read_positive(&dqstats.counter[DQST_FREE_DQUOTS]));
806}
807
808static struct shrinker dqcache_shrinker = {
809	.count_objects = dqcache_shrink_count,
810	.scan_objects = dqcache_shrink_scan,
811	.seeks = DEFAULT_SEEKS,
812};
813
814/*
815 * Safely release dquot and put reference to dquot.
816 */
817static void quota_release_workfn(struct work_struct *work)
818{
819	struct dquot *dquot;
820	struct list_head rls_head;
821
822	spin_lock(&dq_list_lock);
823	/* Exchange the list head to avoid livelock. */
824	list_replace_init(&releasing_dquots, &rls_head);
825	spin_unlock(&dq_list_lock);
826	synchronize_srcu(&dquot_srcu);
827
828restart:
829	spin_lock(&dq_list_lock);
830	while (!list_empty(&rls_head)) {
831		dquot = list_first_entry(&rls_head, struct dquot, dq_free);
832		WARN_ON_ONCE(atomic_read(&dquot->dq_count));
833		/*
834		 * Note that DQ_RELEASING_B protects us from racing with
835		 * invalidate_dquots() calls so we are safe to work with the
836		 * dquot even after we drop dq_list_lock.
837		 */
838		if (dquot_dirty(dquot)) {
839			spin_unlock(&dq_list_lock);
840			/* Commit dquot before releasing */
841			dquot_write_dquot(dquot);
842			goto restart;
843		}
844		if (dquot_active(dquot)) {
845			spin_unlock(&dq_list_lock);
846			dquot->dq_sb->dq_op->release_dquot(dquot);
847			goto restart;
848		}
849		/* Dquot is inactive and clean, now move it to free list */
850		remove_free_dquot(dquot);
851		put_dquot_last(dquot);
852	}
853	spin_unlock(&dq_list_lock);
854}
855
856/*
857 * Put reference to dquot
858 */
859void dqput(struct dquot *dquot)
860{
861	if (!dquot)
862		return;
863#ifdef CONFIG_QUOTA_DEBUG
864	if (!atomic_read(&dquot->dq_count)) {
865		quota_error(dquot->dq_sb, "trying to free free dquot of %s %d",
866			    quotatypes[dquot->dq_id.type],
867			    from_kqid(&init_user_ns, dquot->dq_id));
868		BUG();
869	}
870#endif
871	dqstats_inc(DQST_DROPS);
872
873	spin_lock(&dq_list_lock);
874	if (atomic_read(&dquot->dq_count) > 1) {
875		/* We have more than one user... nothing to do */
876		atomic_dec(&dquot->dq_count);
877		/* Releasing dquot during quotaoff phase? */
878		if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_id.type) &&
879		    atomic_read(&dquot->dq_count) == 1)
880			wake_up(&dquot_ref_wq);
881		spin_unlock(&dq_list_lock);
882		return;
883	}
884
885	/* Need to release dquot? */
886#ifdef CONFIG_QUOTA_DEBUG
887	/* sanity check */
888	BUG_ON(!list_empty(&dquot->dq_free));
889#endif
890	put_releasing_dquots(dquot);
891	atomic_dec(&dquot->dq_count);
892	spin_unlock(&dq_list_lock);
893	queue_delayed_work(system_unbound_wq, &quota_release_work, 1);
894}
895EXPORT_SYMBOL(dqput);
896
897struct dquot *dquot_alloc(struct super_block *sb, int type)
898{
899	return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
900}
901EXPORT_SYMBOL(dquot_alloc);
902
903static struct dquot *get_empty_dquot(struct super_block *sb, int type)
904{
905	struct dquot *dquot;
906
907	dquot = sb->dq_op->alloc_dquot(sb, type);
908	if(!dquot)
909		return NULL;
910
911	mutex_init(&dquot->dq_lock);
912	INIT_LIST_HEAD(&dquot->dq_free);
913	INIT_LIST_HEAD(&dquot->dq_inuse);
914	INIT_HLIST_NODE(&dquot->dq_hash);
915	INIT_LIST_HEAD(&dquot->dq_dirty);
916	dquot->dq_sb = sb;
917	dquot->dq_id = make_kqid_invalid(type);
918	atomic_set(&dquot->dq_count, 1);
919	spin_lock_init(&dquot->dq_dqb_lock);
920
921	return dquot;
922}
923
924/*
925 * Get reference to dquot
926 *
927 * Locking is slightly tricky here. We are guarded from parallel quotaoff()
928 * destroying our dquot by:
929 *   a) checking for quota flags under dq_list_lock and
930 *   b) getting a reference to dquot before we release dq_list_lock
931 */
932struct dquot *dqget(struct super_block *sb, struct kqid qid)
933{
934	unsigned int hashent = hashfn(sb, qid);
935	struct dquot *dquot, *empty = NULL;
936
937	if (!qid_has_mapping(sb->s_user_ns, qid))
938		return ERR_PTR(-EINVAL);
939
940        if (!sb_has_quota_active(sb, qid.type))
941		return ERR_PTR(-ESRCH);
942we_slept:
943	spin_lock(&dq_list_lock);
944	spin_lock(&dq_state_lock);
945	if (!sb_has_quota_active(sb, qid.type)) {
946		spin_unlock(&dq_state_lock);
947		spin_unlock(&dq_list_lock);
948		dquot = ERR_PTR(-ESRCH);
949		goto out;
950	}
951	spin_unlock(&dq_state_lock);
952
953	dquot = find_dquot(hashent, sb, qid);
954	if (!dquot) {
955		if (!empty) {
956			spin_unlock(&dq_list_lock);
957			empty = get_empty_dquot(sb, qid.type);
958			if (!empty)
959				schedule();	/* Try to wait for a moment... */
960			goto we_slept;
961		}
962		dquot = empty;
963		empty = NULL;
964		dquot->dq_id = qid;
965		/* all dquots go on the inuse_list */
966		put_inuse(dquot);
967		/* hash it first so it can be found */
968		insert_dquot_hash(dquot);
969		spin_unlock(&dq_list_lock);
970		dqstats_inc(DQST_LOOKUPS);
971	} else {
972		if (!atomic_read(&dquot->dq_count))
973			remove_free_dquot(dquot);
974		atomic_inc(&dquot->dq_count);
975		spin_unlock(&dq_list_lock);
976		dqstats_inc(DQST_CACHE_HITS);
977		dqstats_inc(DQST_LOOKUPS);
978	}
979	/* Wait for dq_lock - after this we know that either dquot_release() is
980	 * already finished or it will be canceled due to dq_count > 0 test */
981	wait_on_dquot(dquot);
982	/* Read the dquot / allocate space in quota file */
983	if (!dquot_active(dquot)) {
984		int err;
985
986		err = sb->dq_op->acquire_dquot(dquot);
987		if (err < 0) {
988			dqput(dquot);
989			dquot = ERR_PTR(err);
990			goto out;
991		}
992	}
993	/*
994	 * Make sure following reads see filled structure - paired with
995	 * smp_mb__before_atomic() in dquot_acquire().
996	 */
997	smp_rmb();
998#ifdef CONFIG_QUOTA_DEBUG
999	BUG_ON(!dquot->dq_sb);	/* Has somebody invalidated entry under us? */
1000#endif
1001out:
1002	if (empty)
1003		do_destroy_dquot(empty);
1004
1005	return dquot;
1006}
1007EXPORT_SYMBOL(dqget);
1008
1009static inline struct dquot __rcu **i_dquot(struct inode *inode)
1010{
1011	return inode->i_sb->s_op->get_dquots(inode);
1012}
1013
1014static int dqinit_needed(struct inode *inode, int type)
1015{
1016	struct dquot __rcu * const *dquots;
1017	int cnt;
1018
1019	if (IS_NOQUOTA(inode))
1020		return 0;
1021
1022	dquots = i_dquot(inode);
1023	if (type != -1)
1024		return !dquots[type];
1025	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1026		if (!dquots[cnt])
1027			return 1;
1028	return 0;
1029}
1030
1031/* This routine is guarded by s_umount semaphore */
1032static int add_dquot_ref(struct super_block *sb, int type)
1033{
1034	struct inode *inode, *old_inode = NULL;
1035#ifdef CONFIG_QUOTA_DEBUG
1036	int reserved = 0;
1037#endif
1038	int err = 0;
1039
1040	spin_lock(&sb->s_inode_list_lock);
1041	list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1042		spin_lock(&inode->i_lock);
1043		if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
1044		    !atomic_read(&inode->i_writecount) ||
1045		    !dqinit_needed(inode, type)) {
1046			spin_unlock(&inode->i_lock);
1047			continue;
1048		}
1049		__iget(inode);
1050		spin_unlock(&inode->i_lock);
1051		spin_unlock(&sb->s_inode_list_lock);
1052
1053#ifdef CONFIG_QUOTA_DEBUG
1054		if (unlikely(inode_get_rsv_space(inode) > 0))
1055			reserved = 1;
1056#endif
1057		iput(old_inode);
1058		err = __dquot_initialize(inode, type);
1059		if (err) {
1060			iput(inode);
1061			goto out;
1062		}
1063
1064		/*
1065		 * We hold a reference to 'inode' so it couldn't have been
1066		 * removed from s_inodes list while we dropped the
1067		 * s_inode_list_lock. We cannot iput the inode now as we can be
1068		 * holding the last reference and we cannot iput it under
1069		 * s_inode_list_lock. So we keep the reference and iput it
1070		 * later.
1071		 */
1072		old_inode = inode;
1073		cond_resched();
1074		spin_lock(&sb->s_inode_list_lock);
1075	}
1076	spin_unlock(&sb->s_inode_list_lock);
1077	iput(old_inode);
1078out:
1079#ifdef CONFIG_QUOTA_DEBUG
1080	if (reserved) {
1081		quota_error(sb, "Writes happened before quota was turned on "
1082			"thus quota information is probably inconsistent. "
1083			"Please run quotacheck(8)");
1084	}
1085#endif
1086	return err;
1087}
1088
1089static void remove_dquot_ref(struct super_block *sb, int type)
1090{
1091	struct inode *inode;
1092#ifdef CONFIG_QUOTA_DEBUG
1093	int reserved = 0;
1094#endif
1095
1096	spin_lock(&sb->s_inode_list_lock);
1097	list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1098		/*
1099		 *  We have to scan also I_NEW inodes because they can already
1100		 *  have quota pointer initialized. Luckily, we need to touch
1101		 *  only quota pointers and these have separate locking
1102		 *  (dq_data_lock).
1103		 */
1104		spin_lock(&dq_data_lock);
1105		if (!IS_NOQUOTA(inode)) {
1106			struct dquot __rcu **dquots = i_dquot(inode);
1107			struct dquot *dquot = srcu_dereference_check(
1108				dquots[type], &dquot_srcu,
1109				lockdep_is_held(&dq_data_lock));
1110
1111#ifdef CONFIG_QUOTA_DEBUG
1112			if (unlikely(inode_get_rsv_space(inode) > 0))
1113				reserved = 1;
1114#endif
1115			rcu_assign_pointer(dquots[type], NULL);
1116			if (dquot)
1117				dqput(dquot);
1118		}
1119		spin_unlock(&dq_data_lock);
1120	}
1121	spin_unlock(&sb->s_inode_list_lock);
1122#ifdef CONFIG_QUOTA_DEBUG
1123	if (reserved) {
1124		printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1125			" was disabled thus quota information is probably "
1126			"inconsistent. Please run quotacheck(8).\n", sb->s_id);
1127	}
1128#endif
1129}
1130
1131/* Gather all references from inodes and drop them */
1132static void drop_dquot_ref(struct super_block *sb, int type)
1133{
1134	if (sb->dq_op)
1135		remove_dquot_ref(sb, type);
1136}
1137
1138static inline
1139void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1140{
1141	if (dquot->dq_dqb.dqb_rsvspace >= number)
1142		dquot->dq_dqb.dqb_rsvspace -= number;
1143	else {
1144		WARN_ON_ONCE(1);
1145		dquot->dq_dqb.dqb_rsvspace = 0;
1146	}
1147	if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1148	    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
1153static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1154{
1155	if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1156	    dquot->dq_dqb.dqb_curinodes >= number)
1157		dquot->dq_dqb.dqb_curinodes -= number;
1158	else
1159		dquot->dq_dqb.dqb_curinodes = 0;
1160	if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1161		dquot->dq_dqb.dqb_itime = (time64_t) 0;
1162	clear_bit(DQ_INODES_B, &dquot->dq_flags);
1163}
1164
1165static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1166{
1167	if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1168	    dquot->dq_dqb.dqb_curspace >= number)
1169		dquot->dq_dqb.dqb_curspace -= number;
1170	else
1171		dquot->dq_dqb.dqb_curspace = 0;
1172	if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1173	    dquot->dq_dqb.dqb_bsoftlimit)
1174		dquot->dq_dqb.dqb_btime = (time64_t) 0;
1175	clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1176}
1177
1178struct dquot_warn {
1179	struct super_block *w_sb;
1180	struct kqid w_dq_id;
1181	short w_type;
1182};
1183
1184static int warning_issued(struct dquot *dquot, const int warntype)
1185{
1186	int flag = (warntype == QUOTA_NL_BHARDWARN ||
1187		warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1188		((warntype == QUOTA_NL_IHARDWARN ||
1189		warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1190
1191	if (!flag)
1192		return 0;
1193	return test_and_set_bit(flag, &dquot->dq_flags);
1194}
1195
1196#ifdef CONFIG_PRINT_QUOTA_WARNING
1197static int flag_print_warnings = 1;
1198
1199static int need_print_warning(struct dquot_warn *warn)
1200{
1201	if (!flag_print_warnings)
1202		return 0;
1203
1204	switch (warn->w_dq_id.type) {
1205		case USRQUOTA:
1206			return uid_eq(current_fsuid(), warn->w_dq_id.uid);
1207		case GRPQUOTA:
1208			return in_group_p(warn->w_dq_id.gid);
1209		case PRJQUOTA:
1210			return 1;
1211	}
1212	return 0;
1213}
1214
1215/* Print warning to user which exceeded quota */
1216static void print_warning(struct dquot_warn *warn)
1217{
1218	char *msg = NULL;
1219	struct tty_struct *tty;
1220	int warntype = warn->w_type;
1221
1222	if (warntype == QUOTA_NL_IHARDBELOW ||
1223	    warntype == QUOTA_NL_ISOFTBELOW ||
1224	    warntype == QUOTA_NL_BHARDBELOW ||
1225	    warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(warn))
1226		return;
1227
1228	tty = get_current_tty();
1229	if (!tty)
1230		return;
1231	tty_write_message(tty, warn->w_sb->s_id);
1232	if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1233		tty_write_message(tty, ": warning, ");
1234	else
1235		tty_write_message(tty, ": write failed, ");
1236	tty_write_message(tty, quotatypes[warn->w_dq_id.type]);
1237	switch (warntype) {
1238		case QUOTA_NL_IHARDWARN:
1239			msg = " file limit reached.\r\n";
1240			break;
1241		case QUOTA_NL_ISOFTLONGWARN:
1242			msg = " file quota exceeded too long.\r\n";
1243			break;
1244		case QUOTA_NL_ISOFTWARN:
1245			msg = " file quota exceeded.\r\n";
1246			break;
1247		case QUOTA_NL_BHARDWARN:
1248			msg = " block limit reached.\r\n";
1249			break;
1250		case QUOTA_NL_BSOFTLONGWARN:
1251			msg = " block quota exceeded too long.\r\n";
1252			break;
1253		case QUOTA_NL_BSOFTWARN:
1254			msg = " block quota exceeded.\r\n";
1255			break;
1256	}
1257	tty_write_message(tty, msg);
1258	tty_kref_put(tty);
1259}
1260#endif
1261
1262static void prepare_warning(struct dquot_warn *warn, struct dquot *dquot,
1263			    int warntype)
1264{
1265	if (warning_issued(dquot, warntype))
1266		return;
1267	warn->w_type = warntype;
1268	warn->w_sb = dquot->dq_sb;
1269	warn->w_dq_id = dquot->dq_id;
1270}
1271
1272/*
1273 * Write warnings to the console and send warning messages over netlink.
1274 *
1275 * Note that this function can call into tty and networking code.
1276 */
1277static void flush_warnings(struct dquot_warn *warn)
1278{
1279	int i;
1280
1281	for (i = 0; i < MAXQUOTAS; i++) {
1282		if (warn[i].w_type == QUOTA_NL_NOWARN)
1283			continue;
1284#ifdef CONFIG_PRINT_QUOTA_WARNING
1285		print_warning(&warn[i]);
1286#endif
1287		quota_send_warning(warn[i].w_dq_id,
1288				   warn[i].w_sb->s_dev, warn[i].w_type);
1289	}
1290}
1291
1292static int ignore_hardlimit(struct dquot *dquot)
1293{
1294	struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
1295
1296	return capable(CAP_SYS_RESOURCE) &&
1297	       (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1298		!(info->dqi_flags & DQF_ROOT_SQUASH));
1299}
1300
1301static int dquot_add_inodes(struct dquot *dquot, qsize_t inodes,
1302			    struct dquot_warn *warn)
1303{
1304	qsize_t newinodes;
1305	int ret = 0;
1306
1307	spin_lock(&dquot->dq_dqb_lock);
1308	newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1309	if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type) ||
1310	    test_bit(DQ_FAKE_B, &dquot->dq_flags))
1311		goto add;
1312
1313	if (dquot->dq_dqb.dqb_ihardlimit &&
1314	    newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1315            !ignore_hardlimit(dquot)) {
1316		prepare_warning(warn, dquot, QUOTA_NL_IHARDWARN);
1317		ret = -EDQUOT;
1318		goto out;
1319	}
1320
1321	if (dquot->dq_dqb.dqb_isoftlimit &&
1322	    newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1323	    dquot->dq_dqb.dqb_itime &&
1324	    ktime_get_real_seconds() >= dquot->dq_dqb.dqb_itime &&
1325            !ignore_hardlimit(dquot)) {
1326		prepare_warning(warn, dquot, QUOTA_NL_ISOFTLONGWARN);
1327		ret = -EDQUOT;
1328		goto out;
1329	}
1330
1331	if (dquot->dq_dqb.dqb_isoftlimit &&
1332	    newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1333	    dquot->dq_dqb.dqb_itime == 0) {
1334		prepare_warning(warn, dquot, QUOTA_NL_ISOFTWARN);
1335		dquot->dq_dqb.dqb_itime = ktime_get_real_seconds() +
1336		    sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type].dqi_igrace;
1337	}
1338add:
1339	dquot->dq_dqb.dqb_curinodes = newinodes;
1340
1341out:
1342	spin_unlock(&dquot->dq_dqb_lock);
1343	return ret;
1344}
1345
1346static int dquot_add_space(struct dquot *dquot, qsize_t space,
1347			   qsize_t rsv_space, unsigned int flags,
1348			   struct dquot_warn *warn)
1349{
1350	qsize_t tspace;
1351	struct super_block *sb = dquot->dq_sb;
1352	int ret = 0;
1353
1354	spin_lock(&dquot->dq_dqb_lock);
1355	if (!sb_has_quota_limits_enabled(sb, dquot->dq_id.type) ||
1356	    test_bit(DQ_FAKE_B, &dquot->dq_flags))
1357		goto finish;
1358
1359	tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1360		+ space + rsv_space;
1361
1362	if (dquot->dq_dqb.dqb_bhardlimit &&
1363	    tspace > dquot->dq_dqb.dqb_bhardlimit &&
1364            !ignore_hardlimit(dquot)) {
1365		if (flags & DQUOT_SPACE_WARN)
1366			prepare_warning(warn, dquot, QUOTA_NL_BHARDWARN);
1367		ret = -EDQUOT;
1368		goto finish;
1369	}
1370
1371	if (dquot->dq_dqb.dqb_bsoftlimit &&
1372	    tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1373	    dquot->dq_dqb.dqb_btime &&
1374	    ktime_get_real_seconds() >= dquot->dq_dqb.dqb_btime &&
1375            !ignore_hardlimit(dquot)) {
1376		if (flags & DQUOT_SPACE_WARN)
1377			prepare_warning(warn, dquot, QUOTA_NL_BSOFTLONGWARN);
1378		ret = -EDQUOT;
1379		goto finish;
1380	}
1381
1382	if (dquot->dq_dqb.dqb_bsoftlimit &&
1383	    tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1384	    dquot->dq_dqb.dqb_btime == 0) {
1385		if (flags & DQUOT_SPACE_WARN) {
1386			prepare_warning(warn, dquot, QUOTA_NL_BSOFTWARN);
1387			dquot->dq_dqb.dqb_btime = ktime_get_real_seconds() +
1388			    sb_dqopt(sb)->info[dquot->dq_id.type].dqi_bgrace;
1389		} else {
1390			/*
1391			 * We don't allow preallocation to exceed softlimit so exceeding will
1392			 * be always printed
1393			 */
1394			ret = -EDQUOT;
1395			goto finish;
1396		}
1397	}
1398finish:
1399	/*
1400	 * We have to be careful and go through warning generation & grace time
1401	 * setting even if DQUOT_SPACE_NOFAIL is set. That's why we check it
1402	 * only here...
1403	 */
1404	if (flags & DQUOT_SPACE_NOFAIL)
1405		ret = 0;
1406	if (!ret) {
1407		dquot->dq_dqb.dqb_rsvspace += rsv_space;
1408		dquot->dq_dqb.dqb_curspace += space;
1409	}
1410	spin_unlock(&dquot->dq_dqb_lock);
1411	return ret;
1412}
1413
1414static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1415{
1416	qsize_t newinodes;
1417
1418	if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1419	    dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1420	    !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type))
1421		return QUOTA_NL_NOWARN;
1422
1423	newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1424	if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1425		return QUOTA_NL_ISOFTBELOW;
1426	if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1427	    newinodes < dquot->dq_dqb.dqb_ihardlimit)
1428		return QUOTA_NL_IHARDBELOW;
1429	return QUOTA_NL_NOWARN;
1430}
1431
1432static int info_bdq_free(struct dquot *dquot, qsize_t space)
1433{
1434	qsize_t tspace;
1435
1436	tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace;
1437
1438	if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1439	    tspace <= dquot->dq_dqb.dqb_bsoftlimit)
1440		return QUOTA_NL_NOWARN;
1441
1442	if (tspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1443		return QUOTA_NL_BSOFTBELOW;
1444	if (tspace >= dquot->dq_dqb.dqb_bhardlimit &&
1445	    tspace - space < dquot->dq_dqb.dqb_bhardlimit)
1446		return QUOTA_NL_BHARDBELOW;
1447	return QUOTA_NL_NOWARN;
1448}
1449
1450static int inode_quota_active(const struct inode *inode)
1451{
1452	struct super_block *sb = inode->i_sb;
1453
1454	if (IS_NOQUOTA(inode))
1455		return 0;
1456	return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1457}
1458
1459/*
1460 * Initialize quota pointers in inode
1461 *
1462 * It is better to call this function outside of any transaction as it
1463 * might need a lot of space in journal for dquot structure allocation.
1464 */
1465static int __dquot_initialize(struct inode *inode, int type)
1466{
1467	int cnt, init_needed = 0;
1468	struct dquot __rcu **dquots;
1469	struct dquot *got[MAXQUOTAS] = {};
1470	struct super_block *sb = inode->i_sb;
1471	qsize_t rsv;
1472	int ret = 0;
1473
1474	if (!inode_quota_active(inode))
1475		return 0;
1476
1477	dquots = i_dquot(inode);
1478
1479	/* First get references to structures we might need. */
1480	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1481		struct kqid qid;
1482		kprojid_t projid;
1483		int rc;
1484		struct dquot *dquot;
1485
1486		if (type != -1 && cnt != type)
1487			continue;
1488		/*
1489		 * The i_dquot should have been initialized in most cases,
1490		 * we check it without locking here to avoid unnecessary
1491		 * dqget()/dqput() calls.
1492		 */
1493		if (dquots[cnt])
1494			continue;
1495
1496		if (!sb_has_quota_active(sb, cnt))
1497			continue;
1498
1499		init_needed = 1;
1500
1501		switch (cnt) {
1502		case USRQUOTA:
1503			qid = make_kqid_uid(inode->i_uid);
1504			break;
1505		case GRPQUOTA:
1506			qid = make_kqid_gid(inode->i_gid);
1507			break;
1508		case PRJQUOTA:
1509			rc = inode->i_sb->dq_op->get_projid(inode, &projid);
1510			if (rc)
1511				continue;
1512			qid = make_kqid_projid(projid);
1513			break;
1514		}
1515		dquot = dqget(sb, qid);
1516		if (IS_ERR(dquot)) {
1517			/* We raced with somebody turning quotas off... */
1518			if (PTR_ERR(dquot) != -ESRCH) {
1519				ret = PTR_ERR(dquot);
1520				goto out_put;
1521			}
1522			dquot = NULL;
1523		}
1524		got[cnt] = dquot;
1525	}
1526
1527	/* All required i_dquot has been initialized */
1528	if (!init_needed)
1529		return 0;
1530
1531	spin_lock(&dq_data_lock);
1532	if (IS_NOQUOTA(inode))
1533		goto out_lock;
1534	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1535		if (type != -1 && cnt != type)
1536			continue;
1537		/* Avoid races with quotaoff() */
1538		if (!sb_has_quota_active(sb, cnt))
1539			continue;
1540		/* We could race with quotaon or dqget() could have failed */
1541		if (!got[cnt])
1542			continue;
1543		if (!dquots[cnt]) {
1544			rcu_assign_pointer(dquots[cnt], got[cnt]);
1545			got[cnt] = NULL;
1546			/*
1547			 * Make quota reservation system happy if someone
1548			 * did a write before quota was turned on
1549			 */
1550			rsv = inode_get_rsv_space(inode);
1551			if (unlikely(rsv)) {
1552				struct dquot *dquot = srcu_dereference_check(
1553					dquots[cnt], &dquot_srcu,
1554					lockdep_is_held(&dq_data_lock));
1555
1556				spin_lock(&inode->i_lock);
1557				/* Get reservation again under proper lock */
1558				rsv = __inode_get_rsv_space(inode);
1559				spin_lock(&dquot->dq_dqb_lock);
1560				dquot->dq_dqb.dqb_rsvspace += rsv;
1561				spin_unlock(&dquot->dq_dqb_lock);
1562				spin_unlock(&inode->i_lock);
1563			}
1564		}
1565	}
1566out_lock:
1567	spin_unlock(&dq_data_lock);
1568out_put:
1569	/* Drop unused references */
1570	dqput_all(got);
1571
1572	return ret;
1573}
1574
1575int dquot_initialize(struct inode *inode)
1576{
1577	return __dquot_initialize(inode, -1);
1578}
1579EXPORT_SYMBOL(dquot_initialize);
1580
1581bool dquot_initialize_needed(struct inode *inode)
1582{
1583	struct dquot __rcu **dquots;
1584	int i;
1585
1586	if (!inode_quota_active(inode))
1587		return false;
1588
1589	dquots = i_dquot(inode);
1590	for (i = 0; i < MAXQUOTAS; i++)
1591		if (!dquots[i] && sb_has_quota_active(inode->i_sb, i))
1592			return true;
1593	return false;
1594}
1595EXPORT_SYMBOL(dquot_initialize_needed);
1596
1597/*
1598 * Release all quotas referenced by inode.
1599 *
1600 * This function only be called on inode free or converting
1601 * a file to quota file, no other users for the i_dquot in
1602 * both cases, so we needn't call synchronize_srcu() after
1603 * clearing i_dquot.
1604 */
1605static void __dquot_drop(struct inode *inode)
1606{
1607	int cnt;
1608	struct dquot __rcu **dquots = i_dquot(inode);
1609	struct dquot *put[MAXQUOTAS];
1610
1611	spin_lock(&dq_data_lock);
1612	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1613		put[cnt] = srcu_dereference_check(dquots[cnt], &dquot_srcu,
1614					lockdep_is_held(&dq_data_lock));
1615		rcu_assign_pointer(dquots[cnt], NULL);
1616	}
1617	spin_unlock(&dq_data_lock);
1618	dqput_all(put);
1619}
1620
1621void dquot_drop(struct inode *inode)
1622{
1623	struct dquot __rcu * const *dquots;
1624	int cnt;
1625
1626	if (IS_NOQUOTA(inode))
1627		return;
1628
1629	/*
1630	 * Test before calling to rule out calls from proc and such
1631	 * where we are not allowed to block. Note that this is
1632	 * actually reliable test even without the lock - the caller
1633	 * must assure that nobody can come after the DQUOT_DROP and
1634	 * add quota pointers back anyway.
1635	 */
1636	dquots = i_dquot(inode);
1637	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1638		if (dquots[cnt])
1639			break;
1640	}
1641
1642	if (cnt < MAXQUOTAS)
1643		__dquot_drop(inode);
1644}
1645EXPORT_SYMBOL(dquot_drop);
1646
1647/*
1648 * inode_reserved_space is managed internally by quota, and protected by
1649 * i_lock similar to i_blocks+i_bytes.
1650 */
1651static qsize_t *inode_reserved_space(struct inode * inode)
1652{
1653	/* Filesystem must explicitly define it's own method in order to use
1654	 * quota reservation interface */
1655	BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1656	return inode->i_sb->dq_op->get_reserved_space(inode);
1657}
1658
1659static qsize_t __inode_get_rsv_space(struct inode *inode)
1660{
1661	if (!inode->i_sb->dq_op->get_reserved_space)
1662		return 0;
1663	return *inode_reserved_space(inode);
1664}
1665
1666static qsize_t inode_get_rsv_space(struct inode *inode)
1667{
1668	qsize_t ret;
1669
1670	if (!inode->i_sb->dq_op->get_reserved_space)
1671		return 0;
1672	spin_lock(&inode->i_lock);
1673	ret = __inode_get_rsv_space(inode);
1674	spin_unlock(&inode->i_lock);
1675	return ret;
1676}
1677
1678/*
1679 * This functions updates i_blocks+i_bytes fields and quota information
1680 * (together with appropriate checks).
1681 *
1682 * NOTE: We absolutely rely on the fact that caller dirties the inode
1683 * (usually helpers in quotaops.h care about this) and holds a handle for
1684 * the current transaction so that dquot write and inode write go into the
1685 * same transaction.
1686 */
1687
1688/*
1689 * This operation can block, but only after everything is updated
1690 */
1691int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
1692{
1693	int cnt, ret = 0, index;
1694	struct dquot_warn warn[MAXQUOTAS];
1695	int reserve = flags & DQUOT_SPACE_RESERVE;
1696	struct dquot __rcu **dquots;
1697	struct dquot *dquot;
1698
1699	if (!inode_quota_active(inode)) {
1700		if (reserve) {
1701			spin_lock(&inode->i_lock);
1702			*inode_reserved_space(inode) += number;
1703			spin_unlock(&inode->i_lock);
1704		} else {
1705			inode_add_bytes(inode, number);
1706		}
1707		goto out;
1708	}
1709
1710	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1711		warn[cnt].w_type = QUOTA_NL_NOWARN;
1712
1713	dquots = i_dquot(inode);
1714	index = srcu_read_lock(&dquot_srcu);
1715	spin_lock(&inode->i_lock);
1716	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1717		dquot = srcu_dereference(dquots[cnt], &dquot_srcu);
1718		if (!dquot)
1719			continue;
1720		if (reserve) {
1721			ret = dquot_add_space(dquot, 0, number, flags, &warn[cnt]);
1722		} else {
1723			ret = dquot_add_space(dquot, number, 0, flags, &warn[cnt]);
1724		}
1725		if (ret) {
1726			/* Back out changes we already did */
1727			for (cnt--; cnt >= 0; cnt--) {
1728				dquot = srcu_dereference(dquots[cnt], &dquot_srcu);
1729				if (!dquot)
1730					continue;
1731				spin_lock(&dquot->dq_dqb_lock);
1732				if (reserve)
1733					dquot_free_reserved_space(dquot, number);
1734				else
1735					dquot_decr_space(dquot, number);
1736				spin_unlock(&dquot->dq_dqb_lock);
1737			}
1738			spin_unlock(&inode->i_lock);
1739			goto out_flush_warn;
1740		}
1741	}
1742	if (reserve)
1743		*inode_reserved_space(inode) += number;
1744	else
1745		__inode_add_bytes(inode, number);
1746	spin_unlock(&inode->i_lock);
1747
1748	if (reserve)
1749		goto out_flush_warn;
1750	mark_all_dquot_dirty(dquots);
1751out_flush_warn:
1752	srcu_read_unlock(&dquot_srcu, index);
1753	flush_warnings(warn);
1754out:
1755	return ret;
1756}
1757EXPORT_SYMBOL(__dquot_alloc_space);
1758
1759/*
1760 * This operation can block, but only after everything is updated
1761 */
1762int dquot_alloc_inode(struct inode *inode)
1763{
1764	int cnt, ret = 0, index;
1765	struct dquot_warn warn[MAXQUOTAS];
1766	struct dquot __rcu * const *dquots;
1767	struct dquot *dquot;
1768
1769	if (!inode_quota_active(inode))
1770		return 0;
1771	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1772		warn[cnt].w_type = QUOTA_NL_NOWARN;
1773
1774	dquots = i_dquot(inode);
1775	index = srcu_read_lock(&dquot_srcu);
1776	spin_lock(&inode->i_lock);
1777	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1778		dquot = srcu_dereference(dquots[cnt], &dquot_srcu);
1779		if (!dquot)
1780			continue;
1781		ret = dquot_add_inodes(dquot, 1, &warn[cnt]);
1782		if (ret) {
1783			for (cnt--; cnt >= 0; cnt--) {
1784				dquot = srcu_dereference(dquots[cnt], &dquot_srcu);
1785				if (!dquot)
1786					continue;
1787				/* Back out changes we already did */
1788				spin_lock(&dquot->dq_dqb_lock);
1789				dquot_decr_inodes(dquot, 1);
1790				spin_unlock(&dquot->dq_dqb_lock);
1791			}
1792			goto warn_put_all;
1793		}
1794	}
1795
1796warn_put_all:
1797	spin_unlock(&inode->i_lock);
1798	if (ret == 0)
1799		mark_all_dquot_dirty(dquots);
1800	srcu_read_unlock(&dquot_srcu, index);
1801	flush_warnings(warn);
1802	return ret;
1803}
1804EXPORT_SYMBOL(dquot_alloc_inode);
1805
1806/*
1807 * Convert in-memory reserved quotas to real consumed quotas
1808 */
1809int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1810{
1811	struct dquot __rcu **dquots;
1812	struct dquot *dquot;
1813	int cnt, index;
1814
1815	if (!inode_quota_active(inode)) {
1816		spin_lock(&inode->i_lock);
1817		*inode_reserved_space(inode) -= number;
1818		__inode_add_bytes(inode, number);
1819		spin_unlock(&inode->i_lock);
1820		return 0;
1821	}
1822
1823	dquots = i_dquot(inode);
1824	index = srcu_read_lock(&dquot_srcu);
1825	spin_lock(&inode->i_lock);
1826	/* Claim reserved quotas to allocated quotas */
1827	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1828		dquot = srcu_dereference(dquots[cnt], &dquot_srcu);
1829		if (dquot) {
1830			spin_lock(&dquot->dq_dqb_lock);
1831			if (WARN_ON_ONCE(dquot->dq_dqb.dqb_rsvspace < number))
1832				number = dquot->dq_dqb.dqb_rsvspace;
1833			dquot->dq_dqb.dqb_curspace += number;
1834			dquot->dq_dqb.dqb_rsvspace -= number;
1835			spin_unlock(&dquot->dq_dqb_lock);
1836		}
1837	}
1838	/* Update inode bytes */
1839	*inode_reserved_space(inode) -= number;
1840	__inode_add_bytes(inode, number);
1841	spin_unlock(&inode->i_lock);
1842	mark_all_dquot_dirty(dquots);
1843	srcu_read_unlock(&dquot_srcu, index);
1844	return 0;
1845}
1846EXPORT_SYMBOL(dquot_claim_space_nodirty);
1847
1848/*
1849 * Convert allocated space back to in-memory reserved quotas
1850 */
1851void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
1852{
1853	struct dquot __rcu **dquots;
1854	struct dquot *dquot;
1855	int cnt, index;
1856
1857	if (!inode_quota_active(inode)) {
1858		spin_lock(&inode->i_lock);
1859		*inode_reserved_space(inode) += number;
1860		__inode_sub_bytes(inode, number);
1861		spin_unlock(&inode->i_lock);
1862		return;
1863	}
1864
1865	dquots = i_dquot(inode);
1866	index = srcu_read_lock(&dquot_srcu);
1867	spin_lock(&inode->i_lock);
1868	/* Claim reserved quotas to allocated quotas */
1869	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1870		dquot = srcu_dereference(dquots[cnt], &dquot_srcu);
1871		if (dquot) {
1872			spin_lock(&dquot->dq_dqb_lock);
1873			if (WARN_ON_ONCE(dquot->dq_dqb.dqb_curspace < number))
1874				number = dquot->dq_dqb.dqb_curspace;
1875			dquot->dq_dqb.dqb_rsvspace += number;
1876			dquot->dq_dqb.dqb_curspace -= number;
1877			spin_unlock(&dquot->dq_dqb_lock);
1878		}
1879	}
1880	/* Update inode bytes */
1881	*inode_reserved_space(inode) += number;
1882	__inode_sub_bytes(inode, number);
1883	spin_unlock(&inode->i_lock);
1884	mark_all_dquot_dirty(dquots);
1885	srcu_read_unlock(&dquot_srcu, index);
1886	return;
1887}
1888EXPORT_SYMBOL(dquot_reclaim_space_nodirty);
1889
1890/*
1891 * This operation can block, but only after everything is updated
1892 */
1893void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
1894{
1895	unsigned int cnt;
1896	struct dquot_warn warn[MAXQUOTAS];
1897	struct dquot __rcu **dquots;
1898	struct dquot *dquot;
1899	int reserve = flags & DQUOT_SPACE_RESERVE, index;
1900
1901	if (!inode_quota_active(inode)) {
1902		if (reserve) {
1903			spin_lock(&inode->i_lock);
1904			*inode_reserved_space(inode) -= number;
1905			spin_unlock(&inode->i_lock);
1906		} else {
1907			inode_sub_bytes(inode, number);
1908		}
1909		return;
1910	}
1911
1912	dquots = i_dquot(inode);
1913	index = srcu_read_lock(&dquot_srcu);
1914	spin_lock(&inode->i_lock);
1915	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1916		int wtype;
1917
1918		warn[cnt].w_type = QUOTA_NL_NOWARN;
1919		dquot = srcu_dereference(dquots[cnt], &dquot_srcu);
1920		if (!dquot)
1921			continue;
1922		spin_lock(&dquot->dq_dqb_lock);
1923		wtype = info_bdq_free(dquot, number);
1924		if (wtype != QUOTA_NL_NOWARN)
1925			prepare_warning(&warn[cnt], dquot, wtype);
1926		if (reserve)
1927			dquot_free_reserved_space(dquot, number);
1928		else
1929			dquot_decr_space(dquot, number);
1930		spin_unlock(&dquot->dq_dqb_lock);
1931	}
1932	if (reserve)
1933		*inode_reserved_space(inode) -= number;
1934	else
1935		__inode_sub_bytes(inode, number);
1936	spin_unlock(&inode->i_lock);
1937
1938	if (reserve)
1939		goto out_unlock;
1940	mark_all_dquot_dirty(dquots);
1941out_unlock:
1942	srcu_read_unlock(&dquot_srcu, index);
1943	flush_warnings(warn);
1944}
1945EXPORT_SYMBOL(__dquot_free_space);
1946
1947/*
1948 * This operation can block, but only after everything is updated
1949 */
1950void dquot_free_inode(struct inode *inode)
1951{
1952	unsigned int cnt;
1953	struct dquot_warn warn[MAXQUOTAS];
1954	struct dquot __rcu * const *dquots;
1955	struct dquot *dquot;
1956	int index;
1957
1958	if (!inode_quota_active(inode))
1959		return;
1960
1961	dquots = i_dquot(inode);
1962	index = srcu_read_lock(&dquot_srcu);
1963	spin_lock(&inode->i_lock);
1964	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1965		int wtype;
1966		warn[cnt].w_type = QUOTA_NL_NOWARN;
1967		dquot = srcu_dereference(dquots[cnt], &dquot_srcu);
1968		if (!dquot)
1969			continue;
1970		spin_lock(&dquot->dq_dqb_lock);
1971		wtype = info_idq_free(dquot, 1);
1972		if (wtype != QUOTA_NL_NOWARN)
1973			prepare_warning(&warn[cnt], dquot, wtype);
1974		dquot_decr_inodes(dquot, 1);
1975		spin_unlock(&dquot->dq_dqb_lock);
1976	}
1977	spin_unlock(&inode->i_lock);
1978	mark_all_dquot_dirty(dquots);
1979	srcu_read_unlock(&dquot_srcu, index);
1980	flush_warnings(warn);
1981}
1982EXPORT_SYMBOL(dquot_free_inode);
1983
1984/*
1985 * Transfer the number of inode and blocks from one diskquota to an other.
1986 * On success, dquot references in transfer_to are consumed and references
1987 * to original dquots that need to be released are placed there. On failure,
1988 * references are kept untouched.
1989 *
1990 * This operation can block, but only after everything is updated
1991 * A transaction must be started when entering this function.
1992 *
1993 * We are holding reference on transfer_from & transfer_to, no need to
1994 * protect them by srcu_read_lock().
1995 */
1996int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
1997{
1998	qsize_t cur_space;
1999	qsize_t rsv_space = 0;
2000	qsize_t inode_usage = 1;
2001	struct dquot __rcu **dquots;
2002	struct dquot *transfer_from[MAXQUOTAS] = {};
2003	int cnt, index, ret = 0;
2004	char is_valid[MAXQUOTAS] = {};
2005	struct dquot_warn warn_to[MAXQUOTAS];
2006	struct dquot_warn warn_from_inodes[MAXQUOTAS];
2007	struct dquot_warn warn_from_space[MAXQUOTAS];
2008
2009	if (IS_NOQUOTA(inode))
2010		return 0;
2011
2012	if (inode->i_sb->dq_op->get_inode_usage) {
2013		ret = inode->i_sb->dq_op->get_inode_usage(inode, &inode_usage);
2014		if (ret)
2015			return ret;
2016	}
2017
2018	/* Initialize the arrays */
2019	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2020		warn_to[cnt].w_type = QUOTA_NL_NOWARN;
2021		warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
2022		warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
2023	}
2024
2025	spin_lock(&dq_data_lock);
2026	spin_lock(&inode->i_lock);
2027	if (IS_NOQUOTA(inode)) {	/* File without quota accounting? */
2028		spin_unlock(&inode->i_lock);
2029		spin_unlock(&dq_data_lock);
2030		return 0;
2031	}
2032	cur_space = __inode_get_bytes(inode);
2033	rsv_space = __inode_get_rsv_space(inode);
2034	dquots = i_dquot(inode);
2035	/*
2036	 * Build the transfer_from list, check limits, and update usage in
2037	 * the target structures.
2038	 */
2039	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2040		/*
2041		 * Skip changes for same uid or gid or for turned off quota-type.
2042		 */
2043		if (!transfer_to[cnt])
2044			continue;
2045		/* Avoid races with quotaoff() */
2046		if (!sb_has_quota_active(inode->i_sb, cnt))
2047			continue;
2048		is_valid[cnt] = 1;
2049		transfer_from[cnt] = srcu_dereference_check(dquots[cnt],
2050				&dquot_srcu, lockdep_is_held(&dq_data_lock));
2051		ret = dquot_add_inodes(transfer_to[cnt], inode_usage,
2052				       &warn_to[cnt]);
2053		if (ret)
2054			goto over_quota;
2055		ret = dquot_add_space(transfer_to[cnt], cur_space, rsv_space,
2056				      DQUOT_SPACE_WARN, &warn_to[cnt]);
2057		if (ret) {
2058			spin_lock(&transfer_to[cnt]->dq_dqb_lock);
2059			dquot_decr_inodes(transfer_to[cnt], inode_usage);
2060			spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2061			goto over_quota;
2062		}
2063	}
2064
2065	/* Decrease usage for source structures and update quota pointers */
2066	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2067		if (!is_valid[cnt])
2068			continue;
2069		/* Due to IO error we might not have transfer_from[] structure */
2070		if (transfer_from[cnt]) {
2071			int wtype;
2072
2073			spin_lock(&transfer_from[cnt]->dq_dqb_lock);
2074			wtype = info_idq_free(transfer_from[cnt], inode_usage);
2075			if (wtype != QUOTA_NL_NOWARN)
2076				prepare_warning(&warn_from_inodes[cnt],
2077						transfer_from[cnt], wtype);
2078			wtype = info_bdq_free(transfer_from[cnt],
2079					      cur_space + rsv_space);
2080			if (wtype != QUOTA_NL_NOWARN)
2081				prepare_warning(&warn_from_space[cnt],
2082						transfer_from[cnt], wtype);
2083			dquot_decr_inodes(transfer_from[cnt], inode_usage);
2084			dquot_decr_space(transfer_from[cnt], cur_space);
2085			dquot_free_reserved_space(transfer_from[cnt],
2086						  rsv_space);
2087			spin_unlock(&transfer_from[cnt]->dq_dqb_lock);
2088		}
2089		rcu_assign_pointer(dquots[cnt], transfer_to[cnt]);
2090	}
2091	spin_unlock(&inode->i_lock);
2092	spin_unlock(&dq_data_lock);
2093
2094	/*
2095	 * These arrays are local and we hold dquot references so we don't need
2096	 * the srcu protection but still take dquot_srcu to avoid warning in
2097	 * mark_all_dquot_dirty().
2098	 */
2099	index = srcu_read_lock(&dquot_srcu);
2100	mark_all_dquot_dirty((struct dquot __rcu **)transfer_from);
2101	mark_all_dquot_dirty((struct dquot __rcu **)transfer_to);
2102	srcu_read_unlock(&dquot_srcu, index);
2103
2104	flush_warnings(warn_to);
2105	flush_warnings(warn_from_inodes);
2106	flush_warnings(warn_from_space);
2107	/* Pass back references to put */
2108	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2109		if (is_valid[cnt])
2110			transfer_to[cnt] = transfer_from[cnt];
2111	return 0;
2112over_quota:
2113	/* Back out changes we already did */
2114	for (cnt--; cnt >= 0; cnt--) {
2115		if (!is_valid[cnt])
2116			continue;
2117		spin_lock(&transfer_to[cnt]->dq_dqb_lock);
2118		dquot_decr_inodes(transfer_to[cnt], inode_usage);
2119		dquot_decr_space(transfer_to[cnt], cur_space);
2120		dquot_free_reserved_space(transfer_to[cnt], rsv_space);
2121		spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2122	}
2123	spin_unlock(&inode->i_lock);
2124	spin_unlock(&dq_data_lock);
2125	flush_warnings(warn_to);
2126	return ret;
2127}
2128EXPORT_SYMBOL(__dquot_transfer);
2129
2130/* Wrapper for transferring ownership of an inode for uid/gid only
2131 * Called from FSXXX_setattr()
2132 */
2133int dquot_transfer(struct mnt_idmap *idmap, struct inode *inode,
2134		   struct iattr *iattr)
2135{
2136	struct dquot *transfer_to[MAXQUOTAS] = {};
2137	struct dquot *dquot;
2138	struct super_block *sb = inode->i_sb;
2139	int ret;
2140
2141	if (!inode_quota_active(inode))
2142		return 0;
2143
2144	if (i_uid_needs_update(idmap, iattr, inode)) {
2145		kuid_t kuid = from_vfsuid(idmap, i_user_ns(inode),
2146					  iattr->ia_vfsuid);
2147
2148		dquot = dqget(sb, make_kqid_uid(kuid));
2149		if (IS_ERR(dquot)) {
2150			if (PTR_ERR(dquot) != -ESRCH) {
2151				ret = PTR_ERR(dquot);
2152				goto out_put;
2153			}
2154			dquot = NULL;
2155		}
2156		transfer_to[USRQUOTA] = dquot;
2157	}
2158	if (i_gid_needs_update(idmap, iattr, inode)) {
2159		kgid_t kgid = from_vfsgid(idmap, i_user_ns(inode),
2160					  iattr->ia_vfsgid);
2161
2162		dquot = dqget(sb, make_kqid_gid(kgid));
2163		if (IS_ERR(dquot)) {
2164			if (PTR_ERR(dquot) != -ESRCH) {
2165				ret = PTR_ERR(dquot);
2166				goto out_put;
2167			}
2168			dquot = NULL;
2169		}
2170		transfer_to[GRPQUOTA] = dquot;
2171	}
2172	ret = __dquot_transfer(inode, transfer_to);
2173out_put:
2174	dqput_all(transfer_to);
2175	return ret;
2176}
2177EXPORT_SYMBOL(dquot_transfer);
2178
2179/*
2180 * Write info of quota file to disk
2181 */
2182int dquot_commit_info(struct super_block *sb, int type)
2183{
2184	struct quota_info *dqopt = sb_dqopt(sb);
2185
2186	return dqopt->ops[type]->write_file_info(sb, type);
2187}
2188EXPORT_SYMBOL(dquot_commit_info);
2189
2190int dquot_get_next_id(struct super_block *sb, struct kqid *qid)
2191{
2192	struct quota_info *dqopt = sb_dqopt(sb);
2193
2194	if (!sb_has_quota_active(sb, qid->type))
2195		return -ESRCH;
2196	if (!dqopt->ops[qid->type]->get_next_id)
2197		return -ENOSYS;
2198	return dqopt->ops[qid->type]->get_next_id(sb, qid);
2199}
2200EXPORT_SYMBOL(dquot_get_next_id);
2201
2202/*
2203 * Definitions of diskquota operations.
2204 */
2205const struct dquot_operations dquot_operations = {
2206	.write_dquot	= dquot_commit,
2207	.acquire_dquot	= dquot_acquire,
2208	.release_dquot	= dquot_release,
2209	.mark_dirty	= dquot_mark_dquot_dirty,
2210	.write_info	= dquot_commit_info,
2211	.alloc_dquot	= dquot_alloc,
2212	.destroy_dquot	= dquot_destroy,
2213	.get_next_id	= dquot_get_next_id,
2214};
2215EXPORT_SYMBOL(dquot_operations);
2216
2217/*
2218 * Generic helper for ->open on filesystems supporting disk quotas.
2219 */
2220int dquot_file_open(struct inode *inode, struct file *file)
2221{
2222	int error;
2223
2224	error = generic_file_open(inode, file);
2225	if (!error && (file->f_mode & FMODE_WRITE))
2226		error = dquot_initialize(inode);
2227	return error;
2228}
2229EXPORT_SYMBOL(dquot_file_open);
2230
2231static void vfs_cleanup_quota_inode(struct super_block *sb, int type)
2232{
2233	struct quota_info *dqopt = sb_dqopt(sb);
2234	struct inode *inode = dqopt->files[type];
2235
2236	if (!inode)
2237		return;
2238	if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2239		inode_lock(inode);
2240		inode->i_flags &= ~S_NOQUOTA;
2241		inode_unlock(inode);
2242	}
2243	dqopt->files[type] = NULL;
2244	iput(inode);
2245}
2246
2247/*
2248 * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
2249 */
2250int dquot_disable(struct super_block *sb, int type, unsigned int flags)
2251{
2252	int cnt;
2253	struct quota_info *dqopt = sb_dqopt(sb);
2254
2255	/* s_umount should be held in exclusive mode */
2256	if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2257		up_read(&sb->s_umount);
2258
2259	/* Cannot turn off usage accounting without turning off limits, or
2260	 * suspend quotas and simultaneously turn quotas off. */
2261	if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
2262	    || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
2263	    DQUOT_USAGE_ENABLED)))
2264		return -EINVAL;
2265
2266	/*
2267	 * Skip everything if there's nothing to do. We have to do this because
2268	 * sometimes we are called when fill_super() failed and calling
2269	 * sync_fs() in such cases does no good.
2270	 */
2271	if (!sb_any_quota_loaded(sb))
2272		return 0;
2273
2274	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2275		if (type != -1 && cnt != type)
2276			continue;
2277		if (!sb_has_quota_loaded(sb, cnt))
2278			continue;
2279
2280		if (flags & DQUOT_SUSPENDED) {
2281			spin_lock(&dq_state_lock);
2282			dqopt->flags |=
2283				dquot_state_flag(DQUOT_SUSPENDED, cnt);
2284			spin_unlock(&dq_state_lock);
2285		} else {
2286			spin_lock(&dq_state_lock);
2287			dqopt->flags &= ~dquot_state_flag(flags, cnt);
2288			/* Turning off suspended quotas? */
2289			if (!sb_has_quota_loaded(sb, cnt) &&
2290			    sb_has_quota_suspended(sb, cnt)) {
2291				dqopt->flags &=	~dquot_state_flag(
2292							DQUOT_SUSPENDED, cnt);
2293				spin_unlock(&dq_state_lock);
2294				vfs_cleanup_quota_inode(sb, cnt);
2295				continue;
2296			}
2297			spin_unlock(&dq_state_lock);
2298		}
2299
2300		/* We still have to keep quota loaded? */
2301		if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
2302			continue;
2303
2304		/* Note: these are blocking operations */
2305		drop_dquot_ref(sb, cnt);
2306		invalidate_dquots(sb, cnt);
2307		/*
2308		 * Now all dquots should be invalidated, all writes done so we
2309		 * should be only users of the info. No locks needed.
2310		 */
2311		if (info_dirty(&dqopt->info[cnt]))
2312			sb->dq_op->write_info(sb, cnt);
2313		if (dqopt->ops[cnt]->free_file_info)
2314			dqopt->ops[cnt]->free_file_info(sb, cnt);
2315		put_quota_format(dqopt->info[cnt].dqi_format);
2316		dqopt->info[cnt].dqi_flags = 0;
2317		dqopt->info[cnt].dqi_igrace = 0;
2318		dqopt->info[cnt].dqi_bgrace = 0;
2319		dqopt->ops[cnt] = NULL;
2320	}
2321
2322	/* Skip syncing and setting flags if quota files are hidden */
2323	if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2324		goto put_inodes;
2325
2326	/* Sync the superblock so that buffers with quota data are written to
2327	 * disk (and so userspace sees correct data afterwards). */
2328	if (sb->s_op->sync_fs)
2329		sb->s_op->sync_fs(sb, 1);
2330	sync_blockdev(sb->s_bdev);
2331	/* Now the quota files are just ordinary files and we can set the
2332	 * inode flags back. Moreover we discard the pagecache so that
2333	 * userspace sees the writes we did bypassing the pagecache. We
2334	 * must also discard the blockdev buffers so that we see the
2335	 * changes done by userspace on the next quotaon() */
2336	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2337		if (!sb_has_quota_loaded(sb, cnt) && dqopt->files[cnt]) {
2338			inode_lock(dqopt->files[cnt]);
2339			truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
2340			inode_unlock(dqopt->files[cnt]);
2341		}
2342	if (sb->s_bdev)
2343		invalidate_bdev(sb->s_bdev);
2344put_inodes:
2345	/* We are done when suspending quotas */
2346	if (flags & DQUOT_SUSPENDED)
2347		return 0;
2348
2349	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2350		if (!sb_has_quota_loaded(sb, cnt))
2351			vfs_cleanup_quota_inode(sb, cnt);
2352	return 0;
2353}
2354EXPORT_SYMBOL(dquot_disable);
2355
2356int dquot_quota_off(struct super_block *sb, int type)
2357{
2358	return dquot_disable(sb, type,
2359			     DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2360}
2361EXPORT_SYMBOL(dquot_quota_off);
2362
2363/*
2364 *	Turn quotas on on a device
2365 */
2366
2367static int vfs_setup_quota_inode(struct inode *inode, int type)
2368{
2369	struct super_block *sb = inode->i_sb;
2370	struct quota_info *dqopt = sb_dqopt(sb);
2371
2372	if (is_bad_inode(inode))
2373		return -EUCLEAN;
2374	if (!S_ISREG(inode->i_mode))
2375		return -EACCES;
2376	if (IS_RDONLY(inode))
2377		return -EROFS;
2378	if (sb_has_quota_loaded(sb, type))
2379		return -EBUSY;
2380
2381	/*
2382	 * Quota files should never be encrypted.  They should be thought of as
2383	 * filesystem metadata, not user data.  New-style internal quota files
2384	 * cannot be encrypted by users anyway, but old-style external quota
2385	 * files could potentially be incorrectly created in an encrypted
2386	 * directory, hence this explicit check.  Some reasons why encrypted
2387	 * quota files don't work include: (1) some filesystems that support
2388	 * encryption don't handle it in their quota_read and quota_write, and
2389	 * (2) cleaning up encrypted quota files at unmount would need special
2390	 * consideration, as quota files are cleaned up later than user files.
2391	 */
2392	if (IS_ENCRYPTED(inode))
2393		return -EINVAL;
2394
2395	dqopt->files[type] = igrab(inode);
2396	if (!dqopt->files[type])
2397		return -EIO;
2398	if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2399		/* We don't want quota and atime on quota files (deadlocks
2400		 * possible) Also nobody should write to the file - we use
2401		 * special IO operations which ignore the immutable bit. */
2402		inode_lock(inode);
2403		inode->i_flags |= S_NOQUOTA;
2404		inode_unlock(inode);
2405		/*
2406		 * When S_NOQUOTA is set, remove dquot references as no more
2407		 * references can be added
2408		 */
2409		__dquot_drop(inode);
2410	}
2411	return 0;
2412}
2413
2414int dquot_load_quota_sb(struct super_block *sb, int type, int format_id,
2415	unsigned int flags)
2416{
2417	struct quota_format_type *fmt = find_quota_format(format_id);
2418	struct quota_info *dqopt = sb_dqopt(sb);
2419	int error;
2420
2421	lockdep_assert_held_write(&sb->s_umount);
2422
2423	/* Just unsuspend quotas? */
2424	BUG_ON(flags & DQUOT_SUSPENDED);
2425
2426	if (!fmt)
2427		return -ESRCH;
2428	if (!sb->dq_op || !sb->s_qcop ||
2429	    (type == PRJQUOTA && sb->dq_op->get_projid == NULL)) {
2430		error = -EINVAL;
2431		goto out_fmt;
2432	}
2433	/* Filesystems outside of init_user_ns not yet supported */
2434	if (sb->s_user_ns != &init_user_ns) {
2435		error = -EINVAL;
2436		goto out_fmt;
2437	}
2438	/* Usage always has to be set... */
2439	if (!(flags & DQUOT_USAGE_ENABLED)) {
2440		error = -EINVAL;
2441		goto out_fmt;
2442	}
2443	if (sb_has_quota_loaded(sb, type)) {
2444		error = -EBUSY;
2445		goto out_fmt;
2446	}
2447
2448	if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2449		/* As we bypass the pagecache we must now flush all the
2450		 * dirty data and invalidate caches so that kernel sees
2451		 * changes from userspace. It is not enough to just flush
2452		 * the quota file since if blocksize < pagesize, invalidation
2453		 * of the cache could fail because of other unrelated dirty
2454		 * data */
2455		sync_filesystem(sb);
2456		invalidate_bdev(sb->s_bdev);
2457	}
2458
2459	error = -EINVAL;
2460	if (!fmt->qf_ops->check_quota_file(sb, type))
2461		goto out_fmt;
2462
2463	dqopt->ops[type] = fmt->qf_ops;
2464	dqopt->info[type].dqi_format = fmt;
2465	dqopt->info[type].dqi_fmt_id = format_id;
2466	INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2467	error = dqopt->ops[type]->read_file_info(sb, type);
2468	if (error < 0)
2469		goto out_fmt;
2470	if (dqopt->flags & DQUOT_QUOTA_SYS_FILE) {
2471		spin_lock(&dq_data_lock);
2472		dqopt->info[type].dqi_flags |= DQF_SYS_FILE;
2473		spin_unlock(&dq_data_lock);
2474	}
2475	spin_lock(&dq_state_lock);
2476	dqopt->flags |= dquot_state_flag(flags, type);
2477	spin_unlock(&dq_state_lock);
2478
2479	error = add_dquot_ref(sb, type);
2480	if (error)
2481		dquot_disable(sb, type,
2482			      DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2483
2484	return error;
2485out_fmt:
2486	put_quota_format(fmt);
2487
2488	return error;
2489}
2490EXPORT_SYMBOL(dquot_load_quota_sb);
2491
2492/*
2493 * More powerful function for turning on quotas on given quota inode allowing
2494 * setting of individual quota flags
2495 */
2496int dquot_load_quota_inode(struct inode *inode, int type, int format_id,
2497	unsigned int flags)
2498{
2499	int err;
2500
2501	err = vfs_setup_quota_inode(inode, type);
2502	if (err < 0)
2503		return err;
2504	err = dquot_load_quota_sb(inode->i_sb, type, format_id, flags);
2505	if (err < 0)
2506		vfs_cleanup_quota_inode(inode->i_sb, type);
2507	return err;
2508}
2509EXPORT_SYMBOL(dquot_load_quota_inode);
2510
2511/* Reenable quotas on remount RW */
2512int dquot_resume(struct super_block *sb, int type)
2513{
2514	struct quota_info *dqopt = sb_dqopt(sb);
2515	int ret = 0, cnt;
2516	unsigned int flags;
2517
2518	/* s_umount should be held in exclusive mode */
2519	if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2520		up_read(&sb->s_umount);
2521
2522	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2523		if (type != -1 && cnt != type)
2524			continue;
2525		if (!sb_has_quota_suspended(sb, cnt))
2526			continue;
2527
2528		spin_lock(&dq_state_lock);
2529		flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2530							DQUOT_LIMITS_ENABLED,
2531							cnt);
2532		dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2533		spin_unlock(&dq_state_lock);
2534
2535		flags = dquot_generic_flag(flags, cnt);
2536		ret = dquot_load_quota_sb(sb, cnt, dqopt->info[cnt].dqi_fmt_id,
2537					  flags);
2538		if (ret < 0)
2539			vfs_cleanup_quota_inode(sb, cnt);
2540	}
2541
2542	return ret;
2543}
2544EXPORT_SYMBOL(dquot_resume);
2545
2546int dquot_quota_on(struct super_block *sb, int type, int format_id,
2547		   const struct path *path)
2548{
2549	int error = security_quota_on(path->dentry);
2550	if (error)
2551		return error;
2552	/* Quota file not on the same filesystem? */
2553	if (path->dentry->d_sb != sb)
2554		error = -EXDEV;
2555	else
2556		error = dquot_load_quota_inode(d_inode(path->dentry), type,
2557					     format_id, DQUOT_USAGE_ENABLED |
2558					     DQUOT_LIMITS_ENABLED);
2559	return error;
2560}
2561EXPORT_SYMBOL(dquot_quota_on);
2562
2563/*
2564 * This function is used when filesystem needs to initialize quotas
2565 * during mount time.
2566 */
2567int dquot_quota_on_mount(struct super_block *sb, char *qf_name,
2568		int format_id, int type)
2569{
2570	struct dentry *dentry;
2571	int error;
2572
2573	dentry = lookup_positive_unlocked(qf_name, sb->s_root, strlen(qf_name));
2574	if (IS_ERR(dentry))
2575		return PTR_ERR(dentry);
2576
2577	error = security_quota_on(dentry);
2578	if (!error)
2579		error = dquot_load_quota_inode(d_inode(dentry), type, format_id,
2580				DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2581
2582	dput(dentry);
2583	return error;
2584}
2585EXPORT_SYMBOL(dquot_quota_on_mount);
2586
2587static int dquot_quota_enable(struct super_block *sb, unsigned int flags)
2588{
2589	int ret;
2590	int type;
2591	struct quota_info *dqopt = sb_dqopt(sb);
2592
2593	if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2594		return -ENOSYS;
2595	/* Accounting cannot be turned on while fs is mounted */
2596	flags &= ~(FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT);
2597	if (!flags)
2598		return -EINVAL;
2599	for (type = 0; type < MAXQUOTAS; type++) {
2600		if (!(flags & qtype_enforce_flag(type)))
2601			continue;
2602		/* Can't enforce without accounting */
2603		if (!sb_has_quota_usage_enabled(sb, type)) {
2604			ret = -EINVAL;
2605			goto out_err;
2606		}
2607		if (sb_has_quota_limits_enabled(sb, type)) {
2608			ret = -EBUSY;
2609			goto out_err;
2610		}
2611		spin_lock(&dq_state_lock);
2612		dqopt->flags |= dquot_state_flag(DQUOT_LIMITS_ENABLED, type);
2613		spin_unlock(&dq_state_lock);
2614	}
2615	return 0;
2616out_err:
2617	/* Backout enforcement enablement we already did */
2618	for (type--; type >= 0; type--)  {
2619		if (flags & qtype_enforce_flag(type))
2620			dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2621	}
2622	/* Error code translation for better compatibility with XFS */
2623	if (ret == -EBUSY)
2624		ret = -EEXIST;
2625	return ret;
2626}
2627
2628static int dquot_quota_disable(struct super_block *sb, unsigned int flags)
2629{
2630	int ret;
2631	int type;
2632	struct quota_info *dqopt = sb_dqopt(sb);
2633
2634	if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2635		return -ENOSYS;
2636	/*
2637	 * We don't support turning off accounting via quotactl. In principle
2638	 * quota infrastructure can do this but filesystems don't expect
2639	 * userspace to be able to do it.
2640	 */
2641	if (flags &
2642		  (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT))
2643		return -EOPNOTSUPP;
2644
2645	/* Filter out limits not enabled */
2646	for (type = 0; type < MAXQUOTAS; type++)
2647		if (!sb_has_quota_limits_enabled(sb, type))
2648			flags &= ~qtype_enforce_flag(type);
2649	/* Nothing left? */
2650	if (!flags)
2651		return -EEXIST;
2652	for (type = 0; type < MAXQUOTAS; type++) {
2653		if (flags & qtype_enforce_flag(type)) {
2654			ret = dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2655			if (ret < 0)
2656				goto out_err;
2657		}
2658	}
2659	return 0;
2660out_err:
2661	/* Backout enforcement disabling we already did */
2662	for (type--; type >= 0; type--)  {
2663		if (flags & qtype_enforce_flag(type)) {
2664			spin_lock(&dq_state_lock);
2665			dqopt->flags |=
2666				dquot_state_flag(DQUOT_LIMITS_ENABLED, type);
2667			spin_unlock(&dq_state_lock);
2668		}
2669	}
2670	return ret;
2671}
2672
2673/* Generic routine for getting common part of quota structure */
2674static void do_get_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2675{
2676	struct mem_dqblk *dm = &dquot->dq_dqb;
2677
2678	memset(di, 0, sizeof(*di));
2679	spin_lock(&dquot->dq_dqb_lock);
2680	di->d_spc_hardlimit = dm->dqb_bhardlimit;
2681	di->d_spc_softlimit = dm->dqb_bsoftlimit;
2682	di->d_ino_hardlimit = dm->dqb_ihardlimit;
2683	di->d_ino_softlimit = dm->dqb_isoftlimit;
2684	di->d_space = dm->dqb_curspace + dm->dqb_rsvspace;
2685	di->d_ino_count = dm->dqb_curinodes;
2686	di->d_spc_timer = dm->dqb_btime;
2687	di->d_ino_timer = dm->dqb_itime;
2688	spin_unlock(&dquot->dq_dqb_lock);
2689}
2690
2691int dquot_get_dqblk(struct super_block *sb, struct kqid qid,
2692		    struct qc_dqblk *di)
2693{
2694	struct dquot *dquot;
2695
2696	dquot = dqget(sb, qid);
2697	if (IS_ERR(dquot))
2698		return PTR_ERR(dquot);
2699	do_get_dqblk(dquot, di);
2700	dqput(dquot);
2701
2702	return 0;
2703}
2704EXPORT_SYMBOL(dquot_get_dqblk);
2705
2706int dquot_get_next_dqblk(struct super_block *sb, struct kqid *qid,
2707			 struct qc_dqblk *di)
2708{
2709	struct dquot *dquot;
2710	int err;
2711
2712	if (!sb->dq_op->get_next_id)
2713		return -ENOSYS;
2714	err = sb->dq_op->get_next_id(sb, qid);
2715	if (err < 0)
2716		return err;
2717	dquot = dqget(sb, *qid);
2718	if (IS_ERR(dquot))
2719		return PTR_ERR(dquot);
2720	do_get_dqblk(dquot, di);
2721	dqput(dquot);
2722
2723	return 0;
2724}
2725EXPORT_SYMBOL(dquot_get_next_dqblk);
2726
2727#define VFS_QC_MASK \
2728	(QC_SPACE | QC_SPC_SOFT | QC_SPC_HARD | \
2729	 QC_INO_COUNT | QC_INO_SOFT | QC_INO_HARD | \
2730	 QC_SPC_TIMER | QC_INO_TIMER)
2731
2732/* Generic routine for setting common part of quota structure */
2733static int do_set_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2734{
2735	struct mem_dqblk *dm = &dquot->dq_dqb;
2736	int check_blim = 0, check_ilim = 0;
2737	struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
2738
2739	if (di->d_fieldmask & ~VFS_QC_MASK)
2740		return -EINVAL;
2741
2742	if (((di->d_fieldmask & QC_SPC_SOFT) &&
2743	     di->d_spc_softlimit > dqi->dqi_max_spc_limit) ||
2744	    ((di->d_fieldmask & QC_SPC_HARD) &&
2745	     di->d_spc_hardlimit > dqi->dqi_max_spc_limit) ||
2746	    ((di->d_fieldmask & QC_INO_SOFT) &&
2747	     (di->d_ino_softlimit > dqi->dqi_max_ino_limit)) ||
2748	    ((di->d_fieldmask & QC_INO_HARD) &&
2749	     (di->d_ino_hardlimit > dqi->dqi_max_ino_limit)))
2750		return -ERANGE;
2751
2752	spin_lock(&dquot->dq_dqb_lock);
2753	if (di->d_fieldmask & QC_SPACE) {
2754		dm->dqb_curspace = di->d_space - dm->dqb_rsvspace;
2755		check_blim = 1;
2756		set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2757	}
2758
2759	if (di->d_fieldmask & QC_SPC_SOFT)
2760		dm->dqb_bsoftlimit = di->d_spc_softlimit;
2761	if (di->d_fieldmask & QC_SPC_HARD)
2762		dm->dqb_bhardlimit = di->d_spc_hardlimit;
2763	if (di->d_fieldmask & (QC_SPC_SOFT | QC_SPC_HARD)) {
2764		check_blim = 1;
2765		set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2766	}
2767
2768	if (di->d_fieldmask & QC_INO_COUNT) {
2769		dm->dqb_curinodes = di->d_ino_count;
2770		check_ilim = 1;
2771		set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2772	}
2773
2774	if (di->d_fieldmask & QC_INO_SOFT)
2775		dm->dqb_isoftlimit = di->d_ino_softlimit;
2776	if (di->d_fieldmask & QC_INO_HARD)
2777		dm->dqb_ihardlimit = di->d_ino_hardlimit;
2778	if (di->d_fieldmask & (QC_INO_SOFT | QC_INO_HARD)) {
2779		check_ilim = 1;
2780		set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2781	}
2782
2783	if (di->d_fieldmask & QC_SPC_TIMER) {
2784		dm->dqb_btime = di->d_spc_timer;
2785		check_blim = 1;
2786		set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2787	}
2788
2789	if (di->d_fieldmask & QC_INO_TIMER) {
2790		dm->dqb_itime = di->d_ino_timer;
2791		check_ilim = 1;
2792		set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2793	}
2794
2795	if (check_blim) {
2796		if (!dm->dqb_bsoftlimit ||
2797		    dm->dqb_curspace + dm->dqb_rsvspace <= dm->dqb_bsoftlimit) {
2798			dm->dqb_btime = 0;
2799			clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2800		} else if (!(di->d_fieldmask & QC_SPC_TIMER))
2801			/* Set grace only if user hasn't provided his own... */
2802			dm->dqb_btime = ktime_get_real_seconds() + dqi->dqi_bgrace;
2803	}
2804	if (check_ilim) {
2805		if (!dm->dqb_isoftlimit ||
2806		    dm->dqb_curinodes <= dm->dqb_isoftlimit) {
2807			dm->dqb_itime = 0;
2808			clear_bit(DQ_INODES_B, &dquot->dq_flags);
2809		} else if (!(di->d_fieldmask & QC_INO_TIMER))
2810			/* Set grace only if user hasn't provided his own... */
2811			dm->dqb_itime = ktime_get_real_seconds() + dqi->dqi_igrace;
2812	}
2813	if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2814	    dm->dqb_isoftlimit)
2815		clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2816	else
2817		set_bit(DQ_FAKE_B, &dquot->dq_flags);
2818	spin_unlock(&dquot->dq_dqb_lock);
2819	mark_dquot_dirty(dquot);
2820
2821	return 0;
2822}
2823
2824int dquot_set_dqblk(struct super_block *sb, struct kqid qid,
2825		  struct qc_dqblk *di)
2826{
2827	struct dquot *dquot;
2828	int rc;
2829
2830	dquot = dqget(sb, qid);
2831	if (IS_ERR(dquot)) {
2832		rc = PTR_ERR(dquot);
2833		goto out;
2834	}
2835	rc = do_set_dqblk(dquot, di);
2836	dqput(dquot);
2837out:
2838	return rc;
2839}
2840EXPORT_SYMBOL(dquot_set_dqblk);
2841
2842/* Generic routine for getting common part of quota file information */
2843int dquot_get_state(struct super_block *sb, struct qc_state *state)
2844{
2845	struct mem_dqinfo *mi;
2846	struct qc_type_state *tstate;
2847	struct quota_info *dqopt = sb_dqopt(sb);
2848	int type;
2849
2850	memset(state, 0, sizeof(*state));
2851	for (type = 0; type < MAXQUOTAS; type++) {
2852		if (!sb_has_quota_active(sb, type))
2853			continue;
2854		tstate = state->s_state + type;
2855		mi = sb_dqopt(sb)->info + type;
2856		tstate->flags = QCI_ACCT_ENABLED;
2857		spin_lock(&dq_data_lock);
2858		if (mi->dqi_flags & DQF_SYS_FILE)
2859			tstate->flags |= QCI_SYSFILE;
2860		if (mi->dqi_flags & DQF_ROOT_SQUASH)
2861			tstate->flags |= QCI_ROOT_SQUASH;
2862		if (sb_has_quota_limits_enabled(sb, type))
2863			tstate->flags |= QCI_LIMITS_ENFORCED;
2864		tstate->spc_timelimit = mi->dqi_bgrace;
2865		tstate->ino_timelimit = mi->dqi_igrace;
2866		if (dqopt->files[type]) {
2867			tstate->ino = dqopt->files[type]->i_ino;
2868			tstate->blocks = dqopt->files[type]->i_blocks;
2869		}
2870		tstate->nextents = 1;	/* We don't know... */
2871		spin_unlock(&dq_data_lock);
2872	}
2873	return 0;
2874}
2875EXPORT_SYMBOL(dquot_get_state);
2876
2877/* Generic routine for setting common part of quota file information */
2878int dquot_set_dqinfo(struct super_block *sb, int type, struct qc_info *ii)
2879{
2880	struct mem_dqinfo *mi;
2881
2882	if ((ii->i_fieldmask & QC_WARNS_MASK) ||
2883	    (ii->i_fieldmask & QC_RT_SPC_TIMER))
2884		return -EINVAL;
2885	if (!sb_has_quota_active(sb, type))
2886		return -ESRCH;
2887	mi = sb_dqopt(sb)->info + type;
2888	if (ii->i_fieldmask & QC_FLAGS) {
2889		if ((ii->i_flags & QCI_ROOT_SQUASH &&
2890		     mi->dqi_format->qf_fmt_id != QFMT_VFS_OLD))
2891			return -EINVAL;
2892	}
2893	spin_lock(&dq_data_lock);
2894	if (ii->i_fieldmask & QC_SPC_TIMER)
2895		mi->dqi_bgrace = ii->i_spc_timelimit;
2896	if (ii->i_fieldmask & QC_INO_TIMER)
2897		mi->dqi_igrace = ii->i_ino_timelimit;
2898	if (ii->i_fieldmask & QC_FLAGS) {
2899		if (ii->i_flags & QCI_ROOT_SQUASH)
2900			mi->dqi_flags |= DQF_ROOT_SQUASH;
2901		else
2902			mi->dqi_flags &= ~DQF_ROOT_SQUASH;
2903	}
2904	spin_unlock(&dq_data_lock);
2905	mark_info_dirty(sb, type);
2906	/* Force write to disk */
2907	return sb->dq_op->write_info(sb, type);
2908}
2909EXPORT_SYMBOL(dquot_set_dqinfo);
2910
2911const struct quotactl_ops dquot_quotactl_sysfile_ops = {
2912	.quota_enable	= dquot_quota_enable,
2913	.quota_disable	= dquot_quota_disable,
2914	.quota_sync	= dquot_quota_sync,
2915	.get_state	= dquot_get_state,
2916	.set_info	= dquot_set_dqinfo,
2917	.get_dqblk	= dquot_get_dqblk,
2918	.get_nextdqblk	= dquot_get_next_dqblk,
2919	.set_dqblk	= dquot_set_dqblk
2920};
2921EXPORT_SYMBOL(dquot_quotactl_sysfile_ops);
2922
2923static int do_proc_dqstats(struct ctl_table *table, int write,
2924		     void *buffer, size_t *lenp, loff_t *ppos)
2925{
2926	unsigned int type = (unsigned long *)table->data - dqstats.stat;
2927	s64 value = percpu_counter_sum(&dqstats.counter[type]);
2928
2929	/* Filter negative values for non-monotonic counters */
2930	if (value < 0 && (type == DQST_ALLOC_DQUOTS ||
2931			  type == DQST_FREE_DQUOTS))
2932		value = 0;
2933
2934	/* Update global table */
2935	dqstats.stat[type] = value;
2936	return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
2937}
2938
2939static struct ctl_table fs_dqstats_table[] = {
2940	{
2941		.procname	= "lookups",
2942		.data		= &dqstats.stat[DQST_LOOKUPS],
2943		.maxlen		= sizeof(unsigned long),
2944		.mode		= 0444,
2945		.proc_handler	= do_proc_dqstats,
2946	},
2947	{
2948		.procname	= "drops",
2949		.data		= &dqstats.stat[DQST_DROPS],
2950		.maxlen		= sizeof(unsigned long),
2951		.mode		= 0444,
2952		.proc_handler	= do_proc_dqstats,
2953	},
2954	{
2955		.procname	= "reads",
2956		.data		= &dqstats.stat[DQST_READS],
2957		.maxlen		= sizeof(unsigned long),
2958		.mode		= 0444,
2959		.proc_handler	= do_proc_dqstats,
2960	},
2961	{
2962		.procname	= "writes",
2963		.data		= &dqstats.stat[DQST_WRITES],
2964		.maxlen		= sizeof(unsigned long),
2965		.mode		= 0444,
2966		.proc_handler	= do_proc_dqstats,
2967	},
2968	{
2969		.procname	= "cache_hits",
2970		.data		= &dqstats.stat[DQST_CACHE_HITS],
2971		.maxlen		= sizeof(unsigned long),
2972		.mode		= 0444,
2973		.proc_handler	= do_proc_dqstats,
2974	},
2975	{
2976		.procname	= "allocated_dquots",
2977		.data		= &dqstats.stat[DQST_ALLOC_DQUOTS],
2978		.maxlen		= sizeof(unsigned long),
2979		.mode		= 0444,
2980		.proc_handler	= do_proc_dqstats,
2981	},
2982	{
2983		.procname	= "free_dquots",
2984		.data		= &dqstats.stat[DQST_FREE_DQUOTS],
2985		.maxlen		= sizeof(unsigned long),
2986		.mode		= 0444,
2987		.proc_handler	= do_proc_dqstats,
2988	},
2989	{
2990		.procname	= "syncs",
2991		.data		= &dqstats.stat[DQST_SYNCS],
2992		.maxlen		= sizeof(unsigned long),
2993		.mode		= 0444,
2994		.proc_handler	= do_proc_dqstats,
2995	},
2996#ifdef CONFIG_PRINT_QUOTA_WARNING
2997	{
2998		.procname	= "warnings",
2999		.data		= &flag_print_warnings,
3000		.maxlen		= sizeof(int),
3001		.mode		= 0644,
3002		.proc_handler	= proc_dointvec,
3003	},
3004#endif
3005	{ },
3006};
3007
3008static int __init dquot_init(void)
3009{
3010	int i, ret;
3011	unsigned long nr_hash, order;
3012
3013	printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
3014
3015	register_sysctl_init("fs/quota", fs_dqstats_table);
3016
3017	dquot_cachep = kmem_cache_create("dquot",
3018			sizeof(struct dquot), sizeof(unsigned long) * 4,
3019			(SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
3020				SLAB_MEM_SPREAD|SLAB_PANIC),
3021			NULL);
3022
3023	order = 0;
3024	dquot_hash = (struct hlist_head *)__get_free_pages(GFP_KERNEL, order);
3025	if (!dquot_hash)
3026		panic("Cannot create dquot hash table");
3027
3028	for (i = 0; i < _DQST_DQSTAT_LAST; i++) {
3029		ret = percpu_counter_init(&dqstats.counter[i], 0, GFP_KERNEL);
3030		if (ret)
3031			panic("Cannot create dquot stat counters");
3032	}
3033
3034	/* Find power-of-two hlist_heads which can fit into allocation */
3035	nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
3036	dq_hash_bits = ilog2(nr_hash);
3037
3038	nr_hash = 1UL << dq_hash_bits;
3039	dq_hash_mask = nr_hash - 1;
3040	for (i = 0; i < nr_hash; i++)
3041		INIT_HLIST_HEAD(dquot_hash + i);
3042
3043	pr_info("VFS: Dquot-cache hash table entries: %ld (order %ld,"
3044		" %ld bytes)\n", nr_hash, order, (PAGE_SIZE << order));
3045
3046	if (register_shrinker(&dqcache_shrinker, "dquota-cache"))
3047		panic("Cannot register dquot shrinker");
3048
3049	return 0;
3050}
3051fs_initcall(dquot_init);
3052