xref: /kernel/linux/linux-5.10/fs/quota/dquot.c (revision 8c2ecf20)
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 hlist_node *node;
305	struct dquot *dquot;
306
307	hlist_for_each (node, dquot_hash+hashent) {
308		dquot = hlist_entry(node, struct dquot, dq_hash);
309		if (dquot->dq_sb == sb && qid_eq(dquot->dq_id, qid))
310			return dquot;
311	}
312	return NULL;
313}
314
315/* Add a dquot to the tail of the free list */
316static inline void put_dquot_last(struct dquot *dquot)
317{
318	list_add_tail(&dquot->dq_free, &free_dquots);
319	dqstats_inc(DQST_FREE_DQUOTS);
320}
321
322static inline void put_releasing_dquots(struct dquot *dquot)
323{
324	list_add_tail(&dquot->dq_free, &releasing_dquots);
325	set_bit(DQ_RELEASING_B, &dquot->dq_flags);
326}
327
328static inline void remove_free_dquot(struct dquot *dquot)
329{
330	if (list_empty(&dquot->dq_free))
331		return;
332	list_del_init(&dquot->dq_free);
333	if (!test_bit(DQ_RELEASING_B, &dquot->dq_flags))
334		dqstats_dec(DQST_FREE_DQUOTS);
335	else
336		clear_bit(DQ_RELEASING_B, &dquot->dq_flags);
337}
338
339static inline void put_inuse(struct dquot *dquot)
340{
341	/* We add to the back of inuse list so we don't have to restart
342	 * when traversing this list and we block */
343	list_add_tail(&dquot->dq_inuse, &inuse_list);
344	dqstats_inc(DQST_ALLOC_DQUOTS);
345}
346
347static inline void remove_inuse(struct dquot *dquot)
348{
349	dqstats_dec(DQST_ALLOC_DQUOTS);
350	list_del(&dquot->dq_inuse);
351}
352/*
353 * End of list functions needing dq_list_lock
354 */
355
356static void wait_on_dquot(struct dquot *dquot)
357{
358	mutex_lock(&dquot->dq_lock);
359	mutex_unlock(&dquot->dq_lock);
360}
361
362static inline int dquot_active(struct dquot *dquot)
363{
364	return test_bit(DQ_ACTIVE_B, &dquot->dq_flags);
365}
366
367static inline int dquot_dirty(struct dquot *dquot)
368{
369	return test_bit(DQ_MOD_B, &dquot->dq_flags);
370}
371
372static inline int mark_dquot_dirty(struct dquot *dquot)
373{
374	return dquot->dq_sb->dq_op->mark_dirty(dquot);
375}
376
377/* Mark dquot dirty in atomic manner, and return it's old dirty flag state */
378int dquot_mark_dquot_dirty(struct dquot *dquot)
379{
380	int ret = 1;
381
382	if (!dquot_active(dquot))
383		return 0;
384
385	if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NOLIST_DIRTY)
386		return test_and_set_bit(DQ_MOD_B, &dquot->dq_flags);
387
388	/* If quota is dirty already, we don't have to acquire dq_list_lock */
389	if (dquot_dirty(dquot))
390		return 1;
391
392	spin_lock(&dq_list_lock);
393	if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags)) {
394		list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
395				info[dquot->dq_id.type].dqi_dirty_list);
396		ret = 0;
397	}
398	spin_unlock(&dq_list_lock);
399	return ret;
400}
401EXPORT_SYMBOL(dquot_mark_dquot_dirty);
402
403/* Dirtify all the dquots - this can block when journalling */
404static inline int mark_all_dquot_dirty(struct dquot * const *dquots)
405{
406	int ret, err, cnt;
407	struct dquot *dquot;
408
409	ret = err = 0;
410	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
411		dquot = srcu_dereference(dquots[cnt], &dquot_srcu);
412		if (dquot)
413			/* Even in case of error we have to continue */
414			ret = mark_dquot_dirty(dquot);
415		if (!err)
416			err = ret;
417	}
418	return err;
419}
420
421static inline void dqput_all(struct dquot **dquot)
422{
423	unsigned int cnt;
424
425	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
426		dqput(dquot[cnt]);
427}
428
429static inline int clear_dquot_dirty(struct dquot *dquot)
430{
431	if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NOLIST_DIRTY)
432		return test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags);
433
434	spin_lock(&dq_list_lock);
435	if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags)) {
436		spin_unlock(&dq_list_lock);
437		return 0;
438	}
439	list_del_init(&dquot->dq_dirty);
440	spin_unlock(&dq_list_lock);
441	return 1;
442}
443
444void mark_info_dirty(struct super_block *sb, int type)
445{
446	spin_lock(&dq_data_lock);
447	sb_dqopt(sb)->info[type].dqi_flags |= DQF_INFO_DIRTY;
448	spin_unlock(&dq_data_lock);
449}
450EXPORT_SYMBOL(mark_info_dirty);
451
452/*
453 *	Read dquot from disk and alloc space for it
454 */
455
456int dquot_acquire(struct dquot *dquot)
457{
458	int ret = 0, ret2 = 0;
459	unsigned int memalloc;
460	struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
461
462	mutex_lock(&dquot->dq_lock);
463	memalloc = memalloc_nofs_save();
464	if (!test_bit(DQ_READ_B, &dquot->dq_flags)) {
465		ret = dqopt->ops[dquot->dq_id.type]->read_dqblk(dquot);
466		if (ret < 0)
467			goto out_iolock;
468	}
469	/* Make sure flags update is visible after dquot has been filled */
470	smp_mb__before_atomic();
471	set_bit(DQ_READ_B, &dquot->dq_flags);
472	/* Instantiate dquot if needed */
473	if (!dquot_active(dquot) && !dquot->dq_off) {
474		ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
475		/* Write the info if needed */
476		if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
477			ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
478					dquot->dq_sb, dquot->dq_id.type);
479		}
480		if (ret < 0)
481			goto out_iolock;
482		if (ret2 < 0) {
483			ret = ret2;
484			goto out_iolock;
485		}
486	}
487	/*
488	 * Make sure flags update is visible after on-disk struct has been
489	 * allocated. Paired with smp_rmb() in dqget().
490	 */
491	smp_mb__before_atomic();
492	set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
493out_iolock:
494	memalloc_nofs_restore(memalloc);
495	mutex_unlock(&dquot->dq_lock);
496	return ret;
497}
498EXPORT_SYMBOL(dquot_acquire);
499
500/*
501 *	Write dquot to disk
502 */
503int dquot_commit(struct dquot *dquot)
504{
505	int ret = 0;
506	unsigned int memalloc;
507	struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
508
509	mutex_lock(&dquot->dq_lock);
510	memalloc = memalloc_nofs_save();
511	if (!clear_dquot_dirty(dquot))
512		goto out_lock;
513	/* Inactive dquot can be only if there was error during read/init
514	 * => we have better not writing it */
515	if (dquot_active(dquot))
516		ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
517	else
518		ret = -EIO;
519out_lock:
520	memalloc_nofs_restore(memalloc);
521	mutex_unlock(&dquot->dq_lock);
522	return ret;
523}
524EXPORT_SYMBOL(dquot_commit);
525
526/*
527 *	Release dquot
528 */
529int dquot_release(struct dquot *dquot)
530{
531	int ret = 0, ret2 = 0;
532	unsigned int memalloc;
533	struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
534
535	mutex_lock(&dquot->dq_lock);
536	memalloc = memalloc_nofs_save();
537	/* Check whether we are not racing with some other dqget() */
538	if (dquot_is_busy(dquot))
539		goto out_dqlock;
540	if (dqopt->ops[dquot->dq_id.type]->release_dqblk) {
541		ret = dqopt->ops[dquot->dq_id.type]->release_dqblk(dquot);
542		/* Write the info */
543		if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
544			ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
545						dquot->dq_sb, dquot->dq_id.type);
546		}
547		if (ret >= 0)
548			ret = ret2;
549	}
550	clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
551out_dqlock:
552	memalloc_nofs_restore(memalloc);
553	mutex_unlock(&dquot->dq_lock);
554	return ret;
555}
556EXPORT_SYMBOL(dquot_release);
557
558void dquot_destroy(struct dquot *dquot)
559{
560	kmem_cache_free(dquot_cachep, dquot);
561}
562EXPORT_SYMBOL(dquot_destroy);
563
564static inline void do_destroy_dquot(struct dquot *dquot)
565{
566	dquot->dq_sb->dq_op->destroy_dquot(dquot);
567}
568
569/* Invalidate all dquots on the list. Note that this function is called after
570 * quota is disabled and pointers from inodes removed so there cannot be new
571 * quota users. There can still be some users of quotas due to inodes being
572 * just deleted or pruned by prune_icache() (those are not attached to any
573 * list) or parallel quotactl call. We have to wait for such users.
574 */
575static void invalidate_dquots(struct super_block *sb, int type)
576{
577	struct dquot *dquot, *tmp;
578
579restart:
580	flush_delayed_work(&quota_release_work);
581
582	spin_lock(&dq_list_lock);
583	list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
584		if (dquot->dq_sb != sb)
585			continue;
586		if (dquot->dq_id.type != type)
587			continue;
588		/* Wait for dquot users */
589		if (atomic_read(&dquot->dq_count)) {
590			atomic_inc(&dquot->dq_count);
591			spin_unlock(&dq_list_lock);
592			/*
593			 * Once dqput() wakes us up, we know it's time to free
594			 * the dquot.
595			 * IMPORTANT: we rely on the fact that there is always
596			 * at most one process waiting for dquot to free.
597			 * Otherwise dq_count would be > 1 and we would never
598			 * wake up.
599			 */
600			wait_event(dquot_ref_wq,
601				   atomic_read(&dquot->dq_count) == 1);
602			dqput(dquot);
603			/* At this moment dquot() need not exist (it could be
604			 * reclaimed by prune_dqcache(). Hence we must
605			 * restart. */
606			goto restart;
607		}
608		/*
609		 * The last user already dropped its reference but dquot didn't
610		 * get fully cleaned up yet. Restart the scan which flushes the
611		 * work cleaning up released dquots.
612		 */
613		if (test_bit(DQ_RELEASING_B, &dquot->dq_flags)) {
614			spin_unlock(&dq_list_lock);
615			goto restart;
616		}
617		/*
618		 * Quota now has no users and it has been written on last
619		 * dqput()
620		 */
621		remove_dquot_hash(dquot);
622		remove_free_dquot(dquot);
623		remove_inuse(dquot);
624		do_destroy_dquot(dquot);
625	}
626	spin_unlock(&dq_list_lock);
627}
628
629/* Call callback for every active dquot on given filesystem */
630int dquot_scan_active(struct super_block *sb,
631		      int (*fn)(struct dquot *dquot, unsigned long priv),
632		      unsigned long priv)
633{
634	struct dquot *dquot, *old_dquot = NULL;
635	int ret = 0;
636
637	WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
638
639	spin_lock(&dq_list_lock);
640	list_for_each_entry(dquot, &inuse_list, dq_inuse) {
641		if (!dquot_active(dquot))
642			continue;
643		if (dquot->dq_sb != sb)
644			continue;
645		/* Now we have active dquot so we can just increase use count */
646		atomic_inc(&dquot->dq_count);
647		spin_unlock(&dq_list_lock);
648		dqput(old_dquot);
649		old_dquot = dquot;
650		/*
651		 * ->release_dquot() can be racing with us. Our reference
652		 * protects us from new calls to it so just wait for any
653		 * outstanding call and recheck the DQ_ACTIVE_B after that.
654		 */
655		wait_on_dquot(dquot);
656		if (dquot_active(dquot)) {
657			ret = fn(dquot, priv);
658			if (ret < 0)
659				goto out;
660		}
661		spin_lock(&dq_list_lock);
662		/* We are safe to continue now because our dquot could not
663		 * be moved out of the inuse list while we hold the reference */
664	}
665	spin_unlock(&dq_list_lock);
666out:
667	dqput(old_dquot);
668	return ret;
669}
670EXPORT_SYMBOL(dquot_scan_active);
671
672static inline int dquot_write_dquot(struct dquot *dquot)
673{
674	int ret = dquot->dq_sb->dq_op->write_dquot(dquot);
675	if (ret < 0) {
676		quota_error(dquot->dq_sb, "Can't write quota structure "
677			    "(error %d). Quota may get out of sync!", ret);
678		/* Clear dirty bit anyway to avoid infinite loop. */
679		clear_dquot_dirty(dquot);
680	}
681	return ret;
682}
683
684/* Write all dquot structures to quota files */
685int dquot_writeback_dquots(struct super_block *sb, int type)
686{
687	struct list_head dirty;
688	struct dquot *dquot;
689	struct quota_info *dqopt = sb_dqopt(sb);
690	int cnt;
691	int err, ret = 0;
692
693	WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
694
695	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
696		if (type != -1 && cnt != type)
697			continue;
698		if (!sb_has_quota_active(sb, cnt))
699			continue;
700		spin_lock(&dq_list_lock);
701		/* Move list away to avoid livelock. */
702		list_replace_init(&dqopt->info[cnt].dqi_dirty_list, &dirty);
703		while (!list_empty(&dirty)) {
704			dquot = list_first_entry(&dirty, struct dquot,
705						 dq_dirty);
706
707			WARN_ON(!dquot_active(dquot));
708			/* If the dquot is releasing we should not touch it */
709			if (test_bit(DQ_RELEASING_B, &dquot->dq_flags)) {
710				spin_unlock(&dq_list_lock);
711				flush_delayed_work(&quota_release_work);
712				spin_lock(&dq_list_lock);
713				continue;
714			}
715
716			/* Now we have active dquot from which someone is
717 			 * holding reference so we can safely just increase
718			 * use count */
719			dqgrab(dquot);
720			spin_unlock(&dq_list_lock);
721			err = dquot_write_dquot(dquot);
722			if (err && !ret)
723				ret = err;
724			dqput(dquot);
725			spin_lock(&dq_list_lock);
726		}
727		spin_unlock(&dq_list_lock);
728	}
729
730	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
731		if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
732		    && info_dirty(&dqopt->info[cnt]))
733			sb->dq_op->write_info(sb, cnt);
734	dqstats_inc(DQST_SYNCS);
735
736	return ret;
737}
738EXPORT_SYMBOL(dquot_writeback_dquots);
739
740/* Write all dquot structures to disk and make them visible from userspace */
741int dquot_quota_sync(struct super_block *sb, int type)
742{
743	struct quota_info *dqopt = sb_dqopt(sb);
744	int cnt;
745	int ret;
746
747	ret = dquot_writeback_dquots(sb, type);
748	if (ret)
749		return ret;
750	if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
751		return 0;
752
753	/* This is not very clever (and fast) but currently I don't know about
754	 * any other simple way of getting quota data to disk and we must get
755	 * them there for userspace to be visible... */
756	if (sb->s_op->sync_fs) {
757		ret = sb->s_op->sync_fs(sb, 1);
758		if (ret)
759			return ret;
760	}
761	ret = sync_blockdev(sb->s_bdev);
762	if (ret)
763		return ret;
764
765	/*
766	 * Now when everything is written we can discard the pagecache so
767	 * that userspace sees the changes.
768	 */
769	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
770		if (type != -1 && cnt != type)
771			continue;
772		if (!sb_has_quota_active(sb, cnt))
773			continue;
774		inode_lock(dqopt->files[cnt]);
775		truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
776		inode_unlock(dqopt->files[cnt]);
777	}
778
779	return 0;
780}
781EXPORT_SYMBOL(dquot_quota_sync);
782
783static unsigned long
784dqcache_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
785{
786	struct dquot *dquot;
787	unsigned long freed = 0;
788
789	spin_lock(&dq_list_lock);
790	while (!list_empty(&free_dquots) && sc->nr_to_scan) {
791		dquot = list_first_entry(&free_dquots, struct dquot, dq_free);
792		remove_dquot_hash(dquot);
793		remove_free_dquot(dquot);
794		remove_inuse(dquot);
795		do_destroy_dquot(dquot);
796		sc->nr_to_scan--;
797		freed++;
798	}
799	spin_unlock(&dq_list_lock);
800	return freed;
801}
802
803static unsigned long
804dqcache_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
805{
806	return vfs_pressure_ratio(
807	percpu_counter_read_positive(&dqstats.counter[DQST_FREE_DQUOTS]));
808}
809
810static struct shrinker dqcache_shrinker = {
811	.count_objects = dqcache_shrink_count,
812	.scan_objects = dqcache_shrink_scan,
813	.seeks = DEFAULT_SEEKS,
814};
815
816/*
817 * Safely release dquot and put reference to dquot.
818 */
819static void quota_release_workfn(struct work_struct *work)
820{
821	struct dquot *dquot;
822	struct list_head rls_head;
823
824	spin_lock(&dq_list_lock);
825	/* Exchange the list head to avoid livelock. */
826	list_replace_init(&releasing_dquots, &rls_head);
827	spin_unlock(&dq_list_lock);
828	synchronize_srcu(&dquot_srcu);
829
830restart:
831	spin_lock(&dq_list_lock);
832	while (!list_empty(&rls_head)) {
833		dquot = list_first_entry(&rls_head, struct dquot, dq_free);
834		WARN_ON_ONCE(atomic_read(&dquot->dq_count));
835		/*
836		 * Note that DQ_RELEASING_B protects us from racing with
837		 * invalidate_dquots() calls so we are safe to work with the
838		 * dquot even after we drop dq_list_lock.
839		 */
840		if (dquot_dirty(dquot)) {
841			spin_unlock(&dq_list_lock);
842			/* Commit dquot before releasing */
843			dquot_write_dquot(dquot);
844			goto restart;
845		}
846		if (dquot_active(dquot)) {
847			spin_unlock(&dq_list_lock);
848			dquot->dq_sb->dq_op->release_dquot(dquot);
849			goto restart;
850		}
851		/* Dquot is inactive and clean, now move it to free list */
852		remove_free_dquot(dquot);
853		put_dquot_last(dquot);
854	}
855	spin_unlock(&dq_list_lock);
856}
857
858/*
859 * Put reference to dquot
860 */
861void dqput(struct dquot *dquot)
862{
863	if (!dquot)
864		return;
865#ifdef CONFIG_QUOTA_DEBUG
866	if (!atomic_read(&dquot->dq_count)) {
867		quota_error(dquot->dq_sb, "trying to free free dquot of %s %d",
868			    quotatypes[dquot->dq_id.type],
869			    from_kqid(&init_user_ns, dquot->dq_id));
870		BUG();
871	}
872#endif
873	dqstats_inc(DQST_DROPS);
874
875	spin_lock(&dq_list_lock);
876	if (atomic_read(&dquot->dq_count) > 1) {
877		/* We have more than one user... nothing to do */
878		atomic_dec(&dquot->dq_count);
879		/* Releasing dquot during quotaoff phase? */
880		if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_id.type) &&
881		    atomic_read(&dquot->dq_count) == 1)
882			wake_up(&dquot_ref_wq);
883		spin_unlock(&dq_list_lock);
884		return;
885	}
886
887	/* Need to release dquot? */
888#ifdef CONFIG_QUOTA_DEBUG
889	/* sanity check */
890	BUG_ON(!list_empty(&dquot->dq_free));
891#endif
892	put_releasing_dquots(dquot);
893	atomic_dec(&dquot->dq_count);
894	spin_unlock(&dq_list_lock);
895	queue_delayed_work(system_unbound_wq, &quota_release_work, 1);
896}
897EXPORT_SYMBOL(dqput);
898
899struct dquot *dquot_alloc(struct super_block *sb, int type)
900{
901	return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
902}
903EXPORT_SYMBOL(dquot_alloc);
904
905static struct dquot *get_empty_dquot(struct super_block *sb, int type)
906{
907	struct dquot *dquot;
908
909	dquot = sb->dq_op->alloc_dquot(sb, type);
910	if(!dquot)
911		return NULL;
912
913	mutex_init(&dquot->dq_lock);
914	INIT_LIST_HEAD(&dquot->dq_free);
915	INIT_LIST_HEAD(&dquot->dq_inuse);
916	INIT_HLIST_NODE(&dquot->dq_hash);
917	INIT_LIST_HEAD(&dquot->dq_dirty);
918	dquot->dq_sb = sb;
919	dquot->dq_id = make_kqid_invalid(type);
920	atomic_set(&dquot->dq_count, 1);
921	spin_lock_init(&dquot->dq_dqb_lock);
922
923	return dquot;
924}
925
926/*
927 * Get reference to dquot
928 *
929 * Locking is slightly tricky here. We are guarded from parallel quotaoff()
930 * destroying our dquot by:
931 *   a) checking for quota flags under dq_list_lock and
932 *   b) getting a reference to dquot before we release dq_list_lock
933 */
934struct dquot *dqget(struct super_block *sb, struct kqid qid)
935{
936	unsigned int hashent = hashfn(sb, qid);
937	struct dquot *dquot, *empty = NULL;
938
939	if (!qid_has_mapping(sb->s_user_ns, qid))
940		return ERR_PTR(-EINVAL);
941
942        if (!sb_has_quota_active(sb, qid.type))
943		return ERR_PTR(-ESRCH);
944we_slept:
945	spin_lock(&dq_list_lock);
946	spin_lock(&dq_state_lock);
947	if (!sb_has_quota_active(sb, qid.type)) {
948		spin_unlock(&dq_state_lock);
949		spin_unlock(&dq_list_lock);
950		dquot = ERR_PTR(-ESRCH);
951		goto out;
952	}
953	spin_unlock(&dq_state_lock);
954
955	dquot = find_dquot(hashent, sb, qid);
956	if (!dquot) {
957		if (!empty) {
958			spin_unlock(&dq_list_lock);
959			empty = get_empty_dquot(sb, qid.type);
960			if (!empty)
961				schedule();	/* Try to wait for a moment... */
962			goto we_slept;
963		}
964		dquot = empty;
965		empty = NULL;
966		dquot->dq_id = qid;
967		/* all dquots go on the inuse_list */
968		put_inuse(dquot);
969		/* hash it first so it can be found */
970		insert_dquot_hash(dquot);
971		spin_unlock(&dq_list_lock);
972		dqstats_inc(DQST_LOOKUPS);
973	} else {
974		if (!atomic_read(&dquot->dq_count))
975			remove_free_dquot(dquot);
976		atomic_inc(&dquot->dq_count);
977		spin_unlock(&dq_list_lock);
978		dqstats_inc(DQST_CACHE_HITS);
979		dqstats_inc(DQST_LOOKUPS);
980	}
981	/* Wait for dq_lock - after this we know that either dquot_release() is
982	 * already finished or it will be canceled due to dq_count > 0 test */
983	wait_on_dquot(dquot);
984	/* Read the dquot / allocate space in quota file */
985	if (!dquot_active(dquot)) {
986		int err;
987
988		err = sb->dq_op->acquire_dquot(dquot);
989		if (err < 0) {
990			dqput(dquot);
991			dquot = ERR_PTR(err);
992			goto out;
993		}
994	}
995	/*
996	 * Make sure following reads see filled structure - paired with
997	 * smp_mb__before_atomic() in dquot_acquire().
998	 */
999	smp_rmb();
1000#ifdef CONFIG_QUOTA_DEBUG
1001	BUG_ON(!dquot->dq_sb);	/* Has somebody invalidated entry under us? */
1002#endif
1003out:
1004	if (empty)
1005		do_destroy_dquot(empty);
1006
1007	return dquot;
1008}
1009EXPORT_SYMBOL(dqget);
1010
1011static inline struct dquot **i_dquot(struct inode *inode)
1012{
1013	return inode->i_sb->s_op->get_dquots(inode);
1014}
1015
1016static int dqinit_needed(struct inode *inode, int type)
1017{
1018	struct dquot * const *dquots;
1019	int cnt;
1020
1021	if (IS_NOQUOTA(inode))
1022		return 0;
1023
1024	dquots = i_dquot(inode);
1025	if (type != -1)
1026		return !dquots[type];
1027	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1028		if (!dquots[cnt])
1029			return 1;
1030	return 0;
1031}
1032
1033/* This routine is guarded by s_umount semaphore */
1034static int add_dquot_ref(struct super_block *sb, int type)
1035{
1036	struct inode *inode, *old_inode = NULL;
1037#ifdef CONFIG_QUOTA_DEBUG
1038	int reserved = 0;
1039#endif
1040	int err = 0;
1041
1042	spin_lock(&sb->s_inode_list_lock);
1043	list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1044		spin_lock(&inode->i_lock);
1045		if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
1046		    !atomic_read(&inode->i_writecount) ||
1047		    !dqinit_needed(inode, type)) {
1048			spin_unlock(&inode->i_lock);
1049			continue;
1050		}
1051		__iget(inode);
1052		spin_unlock(&inode->i_lock);
1053		spin_unlock(&sb->s_inode_list_lock);
1054
1055#ifdef CONFIG_QUOTA_DEBUG
1056		if (unlikely(inode_get_rsv_space(inode) > 0))
1057			reserved = 1;
1058#endif
1059		iput(old_inode);
1060		err = __dquot_initialize(inode, type);
1061		if (err) {
1062			iput(inode);
1063			goto out;
1064		}
1065
1066		/*
1067		 * We hold a reference to 'inode' so it couldn't have been
1068		 * removed from s_inodes list while we dropped the
1069		 * s_inode_list_lock. We cannot iput the inode now as we can be
1070		 * holding the last reference and we cannot iput it under
1071		 * s_inode_list_lock. So we keep the reference and iput it
1072		 * later.
1073		 */
1074		old_inode = inode;
1075		cond_resched();
1076		spin_lock(&sb->s_inode_list_lock);
1077	}
1078	spin_unlock(&sb->s_inode_list_lock);
1079	iput(old_inode);
1080out:
1081#ifdef CONFIG_QUOTA_DEBUG
1082	if (reserved) {
1083		quota_error(sb, "Writes happened before quota was turned on "
1084			"thus quota information is probably inconsistent. "
1085			"Please run quotacheck(8)");
1086	}
1087#endif
1088	return err;
1089}
1090
1091/*
1092 * Remove references to dquots from inode and add dquot to list for freeing
1093 * if we have the last reference to dquot
1094 */
1095static void remove_inode_dquot_ref(struct inode *inode, int type,
1096				   struct list_head *tofree_head)
1097{
1098	struct dquot **dquots = i_dquot(inode);
1099	struct dquot *dquot = dquots[type];
1100
1101	if (!dquot)
1102		return;
1103
1104	dquots[type] = NULL;
1105	if (list_empty(&dquot->dq_free)) {
1106		/*
1107		 * The inode still has reference to dquot so it can't be in the
1108		 * free list
1109		 */
1110		spin_lock(&dq_list_lock);
1111		list_add(&dquot->dq_free, tofree_head);
1112		spin_unlock(&dq_list_lock);
1113	} else {
1114		/*
1115		 * Dquot is already in a list to put so we won't drop the last
1116		 * reference here.
1117		 */
1118		dqput(dquot);
1119	}
1120}
1121
1122/*
1123 * Free list of dquots
1124 * Dquots are removed from inodes and no new references can be got so we are
1125 * the only ones holding reference
1126 */
1127static void put_dquot_list(struct list_head *tofree_head)
1128{
1129	struct list_head *act_head;
1130	struct dquot *dquot;
1131
1132	act_head = tofree_head->next;
1133	while (act_head != tofree_head) {
1134		dquot = list_entry(act_head, struct dquot, dq_free);
1135		act_head = act_head->next;
1136		/* Remove dquot from the list so we won't have problems... */
1137		list_del_init(&dquot->dq_free);
1138		dqput(dquot);
1139	}
1140}
1141
1142static void remove_dquot_ref(struct super_block *sb, int type,
1143		struct list_head *tofree_head)
1144{
1145	struct inode *inode;
1146#ifdef CONFIG_QUOTA_DEBUG
1147	int reserved = 0;
1148#endif
1149
1150	spin_lock(&sb->s_inode_list_lock);
1151	list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1152		/*
1153		 *  We have to scan also I_NEW inodes because they can already
1154		 *  have quota pointer initialized. Luckily, we need to touch
1155		 *  only quota pointers and these have separate locking
1156		 *  (dq_data_lock).
1157		 */
1158		spin_lock(&dq_data_lock);
1159		if (!IS_NOQUOTA(inode)) {
1160#ifdef CONFIG_QUOTA_DEBUG
1161			if (unlikely(inode_get_rsv_space(inode) > 0))
1162				reserved = 1;
1163#endif
1164			remove_inode_dquot_ref(inode, type, tofree_head);
1165		}
1166		spin_unlock(&dq_data_lock);
1167	}
1168	spin_unlock(&sb->s_inode_list_lock);
1169#ifdef CONFIG_QUOTA_DEBUG
1170	if (reserved) {
1171		printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1172			" was disabled thus quota information is probably "
1173			"inconsistent. Please run quotacheck(8).\n", sb->s_id);
1174	}
1175#endif
1176}
1177
1178/* Gather all references from inodes and drop them */
1179static void drop_dquot_ref(struct super_block *sb, int type)
1180{
1181	LIST_HEAD(tofree_head);
1182
1183	if (sb->dq_op) {
1184		remove_dquot_ref(sb, type, &tofree_head);
1185		synchronize_srcu(&dquot_srcu);
1186		put_dquot_list(&tofree_head);
1187	}
1188}
1189
1190static inline
1191void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1192{
1193	if (dquot->dq_dqb.dqb_rsvspace >= number)
1194		dquot->dq_dqb.dqb_rsvspace -= number;
1195	else {
1196		WARN_ON_ONCE(1);
1197		dquot->dq_dqb.dqb_rsvspace = 0;
1198	}
1199	if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1200	    dquot->dq_dqb.dqb_bsoftlimit)
1201		dquot->dq_dqb.dqb_btime = (time64_t) 0;
1202	clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1203}
1204
1205static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1206{
1207	if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1208	    dquot->dq_dqb.dqb_curinodes >= number)
1209		dquot->dq_dqb.dqb_curinodes -= number;
1210	else
1211		dquot->dq_dqb.dqb_curinodes = 0;
1212	if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1213		dquot->dq_dqb.dqb_itime = (time64_t) 0;
1214	clear_bit(DQ_INODES_B, &dquot->dq_flags);
1215}
1216
1217static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1218{
1219	if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1220	    dquot->dq_dqb.dqb_curspace >= number)
1221		dquot->dq_dqb.dqb_curspace -= number;
1222	else
1223		dquot->dq_dqb.dqb_curspace = 0;
1224	if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1225	    dquot->dq_dqb.dqb_bsoftlimit)
1226		dquot->dq_dqb.dqb_btime = (time64_t) 0;
1227	clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1228}
1229
1230struct dquot_warn {
1231	struct super_block *w_sb;
1232	struct kqid w_dq_id;
1233	short w_type;
1234};
1235
1236static int warning_issued(struct dquot *dquot, const int warntype)
1237{
1238	int flag = (warntype == QUOTA_NL_BHARDWARN ||
1239		warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1240		((warntype == QUOTA_NL_IHARDWARN ||
1241		warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1242
1243	if (!flag)
1244		return 0;
1245	return test_and_set_bit(flag, &dquot->dq_flags);
1246}
1247
1248#ifdef CONFIG_PRINT_QUOTA_WARNING
1249static int flag_print_warnings = 1;
1250
1251static int need_print_warning(struct dquot_warn *warn)
1252{
1253	if (!flag_print_warnings)
1254		return 0;
1255
1256	switch (warn->w_dq_id.type) {
1257		case USRQUOTA:
1258			return uid_eq(current_fsuid(), warn->w_dq_id.uid);
1259		case GRPQUOTA:
1260			return in_group_p(warn->w_dq_id.gid);
1261		case PRJQUOTA:
1262			return 1;
1263	}
1264	return 0;
1265}
1266
1267/* Print warning to user which exceeded quota */
1268static void print_warning(struct dquot_warn *warn)
1269{
1270	char *msg = NULL;
1271	struct tty_struct *tty;
1272	int warntype = warn->w_type;
1273
1274	if (warntype == QUOTA_NL_IHARDBELOW ||
1275	    warntype == QUOTA_NL_ISOFTBELOW ||
1276	    warntype == QUOTA_NL_BHARDBELOW ||
1277	    warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(warn))
1278		return;
1279
1280	tty = get_current_tty();
1281	if (!tty)
1282		return;
1283	tty_write_message(tty, warn->w_sb->s_id);
1284	if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1285		tty_write_message(tty, ": warning, ");
1286	else
1287		tty_write_message(tty, ": write failed, ");
1288	tty_write_message(tty, quotatypes[warn->w_dq_id.type]);
1289	switch (warntype) {
1290		case QUOTA_NL_IHARDWARN:
1291			msg = " file limit reached.\r\n";
1292			break;
1293		case QUOTA_NL_ISOFTLONGWARN:
1294			msg = " file quota exceeded too long.\r\n";
1295			break;
1296		case QUOTA_NL_ISOFTWARN:
1297			msg = " file quota exceeded.\r\n";
1298			break;
1299		case QUOTA_NL_BHARDWARN:
1300			msg = " block limit reached.\r\n";
1301			break;
1302		case QUOTA_NL_BSOFTLONGWARN:
1303			msg = " block quota exceeded too long.\r\n";
1304			break;
1305		case QUOTA_NL_BSOFTWARN:
1306			msg = " block quota exceeded.\r\n";
1307			break;
1308	}
1309	tty_write_message(tty, msg);
1310	tty_kref_put(tty);
1311}
1312#endif
1313
1314static void prepare_warning(struct dquot_warn *warn, struct dquot *dquot,
1315			    int warntype)
1316{
1317	if (warning_issued(dquot, warntype))
1318		return;
1319	warn->w_type = warntype;
1320	warn->w_sb = dquot->dq_sb;
1321	warn->w_dq_id = dquot->dq_id;
1322}
1323
1324/*
1325 * Write warnings to the console and send warning messages over netlink.
1326 *
1327 * Note that this function can call into tty and networking code.
1328 */
1329static void flush_warnings(struct dquot_warn *warn)
1330{
1331	int i;
1332
1333	for (i = 0; i < MAXQUOTAS; i++) {
1334		if (warn[i].w_type == QUOTA_NL_NOWARN)
1335			continue;
1336#ifdef CONFIG_PRINT_QUOTA_WARNING
1337		print_warning(&warn[i]);
1338#endif
1339		quota_send_warning(warn[i].w_dq_id,
1340				   warn[i].w_sb->s_dev, warn[i].w_type);
1341	}
1342}
1343
1344static int ignore_hardlimit(struct dquot *dquot)
1345{
1346	struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
1347
1348	return capable(CAP_SYS_RESOURCE) &&
1349	       (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1350		!(info->dqi_flags & DQF_ROOT_SQUASH));
1351}
1352
1353static int dquot_add_inodes(struct dquot *dquot, qsize_t inodes,
1354			    struct dquot_warn *warn)
1355{
1356	qsize_t newinodes;
1357	int ret = 0;
1358
1359	spin_lock(&dquot->dq_dqb_lock);
1360	newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1361	if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type) ||
1362	    test_bit(DQ_FAKE_B, &dquot->dq_flags))
1363		goto add;
1364
1365	if (dquot->dq_dqb.dqb_ihardlimit &&
1366	    newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1367            !ignore_hardlimit(dquot)) {
1368		prepare_warning(warn, dquot, QUOTA_NL_IHARDWARN);
1369		ret = -EDQUOT;
1370		goto out;
1371	}
1372
1373	if (dquot->dq_dqb.dqb_isoftlimit &&
1374	    newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1375	    dquot->dq_dqb.dqb_itime &&
1376	    ktime_get_real_seconds() >= dquot->dq_dqb.dqb_itime &&
1377            !ignore_hardlimit(dquot)) {
1378		prepare_warning(warn, dquot, QUOTA_NL_ISOFTLONGWARN);
1379		ret = -EDQUOT;
1380		goto out;
1381	}
1382
1383	if (dquot->dq_dqb.dqb_isoftlimit &&
1384	    newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1385	    dquot->dq_dqb.dqb_itime == 0) {
1386		prepare_warning(warn, dquot, QUOTA_NL_ISOFTWARN);
1387		dquot->dq_dqb.dqb_itime = ktime_get_real_seconds() +
1388		    sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type].dqi_igrace;
1389	}
1390add:
1391	dquot->dq_dqb.dqb_curinodes = newinodes;
1392
1393out:
1394	spin_unlock(&dquot->dq_dqb_lock);
1395	return ret;
1396}
1397
1398static int dquot_add_space(struct dquot *dquot, qsize_t space,
1399			   qsize_t rsv_space, unsigned int flags,
1400			   struct dquot_warn *warn)
1401{
1402	qsize_t tspace;
1403	struct super_block *sb = dquot->dq_sb;
1404	int ret = 0;
1405
1406	spin_lock(&dquot->dq_dqb_lock);
1407	if (!sb_has_quota_limits_enabled(sb, dquot->dq_id.type) ||
1408	    test_bit(DQ_FAKE_B, &dquot->dq_flags))
1409		goto finish;
1410
1411	tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1412		+ space + rsv_space;
1413
1414	if (dquot->dq_dqb.dqb_bhardlimit &&
1415	    tspace > dquot->dq_dqb.dqb_bhardlimit &&
1416            !ignore_hardlimit(dquot)) {
1417		if (flags & DQUOT_SPACE_WARN)
1418			prepare_warning(warn, dquot, QUOTA_NL_BHARDWARN);
1419		ret = -EDQUOT;
1420		goto finish;
1421	}
1422
1423	if (dquot->dq_dqb.dqb_bsoftlimit &&
1424	    tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1425	    dquot->dq_dqb.dqb_btime &&
1426	    ktime_get_real_seconds() >= dquot->dq_dqb.dqb_btime &&
1427            !ignore_hardlimit(dquot)) {
1428		if (flags & DQUOT_SPACE_WARN)
1429			prepare_warning(warn, dquot, QUOTA_NL_BSOFTLONGWARN);
1430		ret = -EDQUOT;
1431		goto finish;
1432	}
1433
1434	if (dquot->dq_dqb.dqb_bsoftlimit &&
1435	    tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1436	    dquot->dq_dqb.dqb_btime == 0) {
1437		if (flags & DQUOT_SPACE_WARN) {
1438			prepare_warning(warn, dquot, QUOTA_NL_BSOFTWARN);
1439			dquot->dq_dqb.dqb_btime = ktime_get_real_seconds() +
1440			    sb_dqopt(sb)->info[dquot->dq_id.type].dqi_bgrace;
1441		} else {
1442			/*
1443			 * We don't allow preallocation to exceed softlimit so exceeding will
1444			 * be always printed
1445			 */
1446			ret = -EDQUOT;
1447			goto finish;
1448		}
1449	}
1450finish:
1451	/*
1452	 * We have to be careful and go through warning generation & grace time
1453	 * setting even if DQUOT_SPACE_NOFAIL is set. That's why we check it
1454	 * only here...
1455	 */
1456	if (flags & DQUOT_SPACE_NOFAIL)
1457		ret = 0;
1458	if (!ret) {
1459		dquot->dq_dqb.dqb_rsvspace += rsv_space;
1460		dquot->dq_dqb.dqb_curspace += space;
1461	}
1462	spin_unlock(&dquot->dq_dqb_lock);
1463	return ret;
1464}
1465
1466static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1467{
1468	qsize_t newinodes;
1469
1470	if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1471	    dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1472	    !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type))
1473		return QUOTA_NL_NOWARN;
1474
1475	newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1476	if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1477		return QUOTA_NL_ISOFTBELOW;
1478	if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1479	    newinodes < dquot->dq_dqb.dqb_ihardlimit)
1480		return QUOTA_NL_IHARDBELOW;
1481	return QUOTA_NL_NOWARN;
1482}
1483
1484static int info_bdq_free(struct dquot *dquot, qsize_t space)
1485{
1486	qsize_t tspace;
1487
1488	tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace;
1489
1490	if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1491	    tspace <= dquot->dq_dqb.dqb_bsoftlimit)
1492		return QUOTA_NL_NOWARN;
1493
1494	if (tspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1495		return QUOTA_NL_BSOFTBELOW;
1496	if (tspace >= dquot->dq_dqb.dqb_bhardlimit &&
1497	    tspace - space < dquot->dq_dqb.dqb_bhardlimit)
1498		return QUOTA_NL_BHARDBELOW;
1499	return QUOTA_NL_NOWARN;
1500}
1501
1502static int inode_quota_active(const struct inode *inode)
1503{
1504	struct super_block *sb = inode->i_sb;
1505
1506	if (IS_NOQUOTA(inode))
1507		return 0;
1508	return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1509}
1510
1511/*
1512 * Initialize quota pointers in inode
1513 *
1514 * It is better to call this function outside of any transaction as it
1515 * might need a lot of space in journal for dquot structure allocation.
1516 */
1517static int __dquot_initialize(struct inode *inode, int type)
1518{
1519	int cnt, init_needed = 0;
1520	struct dquot **dquots, *got[MAXQUOTAS] = {};
1521	struct super_block *sb = inode->i_sb;
1522	qsize_t rsv;
1523	int ret = 0;
1524
1525	if (!inode_quota_active(inode))
1526		return 0;
1527
1528	dquots = i_dquot(inode);
1529
1530	/* First get references to structures we might need. */
1531	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1532		struct kqid qid;
1533		kprojid_t projid;
1534		int rc;
1535		struct dquot *dquot;
1536
1537		if (type != -1 && cnt != type)
1538			continue;
1539		/*
1540		 * The i_dquot should have been initialized in most cases,
1541		 * we check it without locking here to avoid unnecessary
1542		 * dqget()/dqput() calls.
1543		 */
1544		if (dquots[cnt])
1545			continue;
1546
1547		if (!sb_has_quota_active(sb, cnt))
1548			continue;
1549
1550		init_needed = 1;
1551
1552		switch (cnt) {
1553		case USRQUOTA:
1554			qid = make_kqid_uid(inode->i_uid);
1555			break;
1556		case GRPQUOTA:
1557			qid = make_kqid_gid(inode->i_gid);
1558			break;
1559		case PRJQUOTA:
1560			rc = inode->i_sb->dq_op->get_projid(inode, &projid);
1561			if (rc)
1562				continue;
1563			qid = make_kqid_projid(projid);
1564			break;
1565		}
1566		dquot = dqget(sb, qid);
1567		if (IS_ERR(dquot)) {
1568			/* We raced with somebody turning quotas off... */
1569			if (PTR_ERR(dquot) != -ESRCH) {
1570				ret = PTR_ERR(dquot);
1571				goto out_put;
1572			}
1573			dquot = NULL;
1574		}
1575		got[cnt] = dquot;
1576	}
1577
1578	/* All required i_dquot has been initialized */
1579	if (!init_needed)
1580		return 0;
1581
1582	spin_lock(&dq_data_lock);
1583	if (IS_NOQUOTA(inode))
1584		goto out_lock;
1585	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1586		if (type != -1 && cnt != type)
1587			continue;
1588		/* Avoid races with quotaoff() */
1589		if (!sb_has_quota_active(sb, cnt))
1590			continue;
1591		/* We could race with quotaon or dqget() could have failed */
1592		if (!got[cnt])
1593			continue;
1594		if (!dquots[cnt]) {
1595			dquots[cnt] = got[cnt];
1596			got[cnt] = NULL;
1597			/*
1598			 * Make quota reservation system happy if someone
1599			 * did a write before quota was turned on
1600			 */
1601			rsv = inode_get_rsv_space(inode);
1602			if (unlikely(rsv)) {
1603				spin_lock(&inode->i_lock);
1604				/* Get reservation again under proper lock */
1605				rsv = __inode_get_rsv_space(inode);
1606				spin_lock(&dquots[cnt]->dq_dqb_lock);
1607				dquots[cnt]->dq_dqb.dqb_rsvspace += rsv;
1608				spin_unlock(&dquots[cnt]->dq_dqb_lock);
1609				spin_unlock(&inode->i_lock);
1610			}
1611		}
1612	}
1613out_lock:
1614	spin_unlock(&dq_data_lock);
1615out_put:
1616	/* Drop unused references */
1617	dqput_all(got);
1618
1619	return ret;
1620}
1621
1622int dquot_initialize(struct inode *inode)
1623{
1624	return __dquot_initialize(inode, -1);
1625}
1626EXPORT_SYMBOL(dquot_initialize);
1627
1628bool dquot_initialize_needed(struct inode *inode)
1629{
1630	struct dquot **dquots;
1631	int i;
1632
1633	if (!inode_quota_active(inode))
1634		return false;
1635
1636	dquots = i_dquot(inode);
1637	for (i = 0; i < MAXQUOTAS; i++)
1638		if (!dquots[i] && sb_has_quota_active(inode->i_sb, i))
1639			return true;
1640	return false;
1641}
1642EXPORT_SYMBOL(dquot_initialize_needed);
1643
1644/*
1645 * Release all quotas referenced by inode.
1646 *
1647 * This function only be called on inode free or converting
1648 * a file to quota file, no other users for the i_dquot in
1649 * both cases, so we needn't call synchronize_srcu() after
1650 * clearing i_dquot.
1651 */
1652static void __dquot_drop(struct inode *inode)
1653{
1654	int cnt;
1655	struct dquot **dquots = i_dquot(inode);
1656	struct dquot *put[MAXQUOTAS];
1657
1658	spin_lock(&dq_data_lock);
1659	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1660		put[cnt] = dquots[cnt];
1661		dquots[cnt] = NULL;
1662	}
1663	spin_unlock(&dq_data_lock);
1664	dqput_all(put);
1665}
1666
1667void dquot_drop(struct inode *inode)
1668{
1669	struct dquot * const *dquots;
1670	int cnt;
1671
1672	if (IS_NOQUOTA(inode))
1673		return;
1674
1675	/*
1676	 * Test before calling to rule out calls from proc and such
1677	 * where we are not allowed to block. Note that this is
1678	 * actually reliable test even without the lock - the caller
1679	 * must assure that nobody can come after the DQUOT_DROP and
1680	 * add quota pointers back anyway.
1681	 */
1682	dquots = i_dquot(inode);
1683	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1684		if (dquots[cnt])
1685			break;
1686	}
1687
1688	if (cnt < MAXQUOTAS)
1689		__dquot_drop(inode);
1690}
1691EXPORT_SYMBOL(dquot_drop);
1692
1693/*
1694 * inode_reserved_space is managed internally by quota, and protected by
1695 * i_lock similar to i_blocks+i_bytes.
1696 */
1697static qsize_t *inode_reserved_space(struct inode * inode)
1698{
1699	/* Filesystem must explicitly define it's own method in order to use
1700	 * quota reservation interface */
1701	BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1702	return inode->i_sb->dq_op->get_reserved_space(inode);
1703}
1704
1705static qsize_t __inode_get_rsv_space(struct inode *inode)
1706{
1707	if (!inode->i_sb->dq_op->get_reserved_space)
1708		return 0;
1709	return *inode_reserved_space(inode);
1710}
1711
1712static qsize_t inode_get_rsv_space(struct inode *inode)
1713{
1714	qsize_t ret;
1715
1716	if (!inode->i_sb->dq_op->get_reserved_space)
1717		return 0;
1718	spin_lock(&inode->i_lock);
1719	ret = __inode_get_rsv_space(inode);
1720	spin_unlock(&inode->i_lock);
1721	return ret;
1722}
1723
1724/*
1725 * This functions updates i_blocks+i_bytes fields and quota information
1726 * (together with appropriate checks).
1727 *
1728 * NOTE: We absolutely rely on the fact that caller dirties the inode
1729 * (usually helpers in quotaops.h care about this) and holds a handle for
1730 * the current transaction so that dquot write and inode write go into the
1731 * same transaction.
1732 */
1733
1734/*
1735 * This operation can block, but only after everything is updated
1736 */
1737int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
1738{
1739	int cnt, ret = 0, index;
1740	struct dquot_warn warn[MAXQUOTAS];
1741	int reserve = flags & DQUOT_SPACE_RESERVE;
1742	struct dquot **dquots;
1743	struct dquot *dquot;
1744
1745	if (!inode_quota_active(inode)) {
1746		if (reserve) {
1747			spin_lock(&inode->i_lock);
1748			*inode_reserved_space(inode) += number;
1749			spin_unlock(&inode->i_lock);
1750		} else {
1751			inode_add_bytes(inode, number);
1752		}
1753		goto out;
1754	}
1755
1756	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1757		warn[cnt].w_type = QUOTA_NL_NOWARN;
1758
1759	dquots = i_dquot(inode);
1760	index = srcu_read_lock(&dquot_srcu);
1761	spin_lock(&inode->i_lock);
1762	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1763		dquot = srcu_dereference(dquots[cnt], &dquot_srcu);
1764		if (!dquot)
1765			continue;
1766		if (reserve) {
1767			ret = dquot_add_space(dquot, 0, number, flags, &warn[cnt]);
1768		} else {
1769			ret = dquot_add_space(dquot, number, 0, flags, &warn[cnt]);
1770		}
1771		if (ret) {
1772			/* Back out changes we already did */
1773			for (cnt--; cnt >= 0; cnt--) {
1774				dquot = srcu_dereference(dquots[cnt], &dquot_srcu);
1775				if (!dquot)
1776					continue;
1777				spin_lock(&dquot->dq_dqb_lock);
1778				if (reserve)
1779					dquot_free_reserved_space(dquot, number);
1780				else
1781					dquot_decr_space(dquot, number);
1782				spin_unlock(&dquot->dq_dqb_lock);
1783			}
1784			spin_unlock(&inode->i_lock);
1785			goto out_flush_warn;
1786		}
1787	}
1788	if (reserve)
1789		*inode_reserved_space(inode) += number;
1790	else
1791		__inode_add_bytes(inode, number);
1792	spin_unlock(&inode->i_lock);
1793
1794	if (reserve)
1795		goto out_flush_warn;
1796	mark_all_dquot_dirty(dquots);
1797out_flush_warn:
1798	srcu_read_unlock(&dquot_srcu, index);
1799	flush_warnings(warn);
1800out:
1801	return ret;
1802}
1803EXPORT_SYMBOL(__dquot_alloc_space);
1804
1805/*
1806 * This operation can block, but only after everything is updated
1807 */
1808int dquot_alloc_inode(struct inode *inode)
1809{
1810	int cnt, ret = 0, index;
1811	struct dquot_warn warn[MAXQUOTAS];
1812	struct dquot * const *dquots;
1813	struct dquot *dquot;
1814
1815	if (!inode_quota_active(inode))
1816		return 0;
1817	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1818		warn[cnt].w_type = QUOTA_NL_NOWARN;
1819
1820	dquots = i_dquot(inode);
1821	index = srcu_read_lock(&dquot_srcu);
1822	spin_lock(&inode->i_lock);
1823	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1824		dquot = srcu_dereference(dquots[cnt], &dquot_srcu);
1825		if (!dquot)
1826			continue;
1827		ret = dquot_add_inodes(dquot, 1, &warn[cnt]);
1828		if (ret) {
1829			for (cnt--; cnt >= 0; cnt--) {
1830				dquot = srcu_dereference(dquots[cnt], &dquot_srcu);
1831				if (!dquot)
1832					continue;
1833				/* Back out changes we already did */
1834				spin_lock(&dquot->dq_dqb_lock);
1835				dquot_decr_inodes(dquot, 1);
1836				spin_unlock(&dquot->dq_dqb_lock);
1837			}
1838			goto warn_put_all;
1839		}
1840	}
1841
1842warn_put_all:
1843	spin_unlock(&inode->i_lock);
1844	if (ret == 0)
1845		mark_all_dquot_dirty(dquots);
1846	srcu_read_unlock(&dquot_srcu, index);
1847	flush_warnings(warn);
1848	return ret;
1849}
1850EXPORT_SYMBOL(dquot_alloc_inode);
1851
1852/*
1853 * Convert in-memory reserved quotas to real consumed quotas
1854 */
1855int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1856{
1857	struct dquot **dquots;
1858	struct dquot *dquot;
1859	int cnt, index;
1860
1861	if (!inode_quota_active(inode)) {
1862		spin_lock(&inode->i_lock);
1863		*inode_reserved_space(inode) -= number;
1864		__inode_add_bytes(inode, number);
1865		spin_unlock(&inode->i_lock);
1866		return 0;
1867	}
1868
1869	dquots = i_dquot(inode);
1870	index = srcu_read_lock(&dquot_srcu);
1871	spin_lock(&inode->i_lock);
1872	/* Claim reserved quotas to allocated quotas */
1873	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1874		dquot = srcu_dereference(dquots[cnt], &dquot_srcu);
1875		if (dquot) {
1876			spin_lock(&dquot->dq_dqb_lock);
1877			if (WARN_ON_ONCE(dquot->dq_dqb.dqb_rsvspace < number))
1878				number = dquot->dq_dqb.dqb_rsvspace;
1879			dquot->dq_dqb.dqb_curspace += number;
1880			dquot->dq_dqb.dqb_rsvspace -= number;
1881			spin_unlock(&dquot->dq_dqb_lock);
1882		}
1883	}
1884	/* Update inode bytes */
1885	*inode_reserved_space(inode) -= number;
1886	__inode_add_bytes(inode, number);
1887	spin_unlock(&inode->i_lock);
1888	mark_all_dquot_dirty(dquots);
1889	srcu_read_unlock(&dquot_srcu, index);
1890	return 0;
1891}
1892EXPORT_SYMBOL(dquot_claim_space_nodirty);
1893
1894/*
1895 * Convert allocated space back to in-memory reserved quotas
1896 */
1897void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
1898{
1899	struct dquot **dquots;
1900	struct dquot *dquot;
1901	int cnt, index;
1902
1903	if (!inode_quota_active(inode)) {
1904		spin_lock(&inode->i_lock);
1905		*inode_reserved_space(inode) += number;
1906		__inode_sub_bytes(inode, number);
1907		spin_unlock(&inode->i_lock);
1908		return;
1909	}
1910
1911	dquots = i_dquot(inode);
1912	index = srcu_read_lock(&dquot_srcu);
1913	spin_lock(&inode->i_lock);
1914	/* Claim reserved quotas to allocated quotas */
1915	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1916		dquot = srcu_dereference(dquots[cnt], &dquot_srcu);
1917		if (dquot) {
1918			spin_lock(&dquot->dq_dqb_lock);
1919			if (WARN_ON_ONCE(dquot->dq_dqb.dqb_curspace < number))
1920				number = dquot->dq_dqb.dqb_curspace;
1921			dquot->dq_dqb.dqb_rsvspace += number;
1922			dquot->dq_dqb.dqb_curspace -= number;
1923			spin_unlock(&dquot->dq_dqb_lock);
1924		}
1925	}
1926	/* Update inode bytes */
1927	*inode_reserved_space(inode) += number;
1928	__inode_sub_bytes(inode, number);
1929	spin_unlock(&inode->i_lock);
1930	mark_all_dquot_dirty(dquots);
1931	srcu_read_unlock(&dquot_srcu, index);
1932	return;
1933}
1934EXPORT_SYMBOL(dquot_reclaim_space_nodirty);
1935
1936/*
1937 * This operation can block, but only after everything is updated
1938 */
1939void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
1940{
1941	unsigned int cnt;
1942	struct dquot_warn warn[MAXQUOTAS];
1943	struct dquot **dquots;
1944	struct dquot *dquot;
1945	int reserve = flags & DQUOT_SPACE_RESERVE, index;
1946
1947	if (!inode_quota_active(inode)) {
1948		if (reserve) {
1949			spin_lock(&inode->i_lock);
1950			*inode_reserved_space(inode) -= number;
1951			spin_unlock(&inode->i_lock);
1952		} else {
1953			inode_sub_bytes(inode, number);
1954		}
1955		return;
1956	}
1957
1958	dquots = i_dquot(inode);
1959	index = srcu_read_lock(&dquot_srcu);
1960	spin_lock(&inode->i_lock);
1961	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1962		int wtype;
1963
1964		warn[cnt].w_type = QUOTA_NL_NOWARN;
1965		dquot = srcu_dereference(dquots[cnt], &dquot_srcu);
1966		if (!dquot)
1967			continue;
1968		spin_lock(&dquot->dq_dqb_lock);
1969		wtype = info_bdq_free(dquot, number);
1970		if (wtype != QUOTA_NL_NOWARN)
1971			prepare_warning(&warn[cnt], dquot, wtype);
1972		if (reserve)
1973			dquot_free_reserved_space(dquot, number);
1974		else
1975			dquot_decr_space(dquot, number);
1976		spin_unlock(&dquot->dq_dqb_lock);
1977	}
1978	if (reserve)
1979		*inode_reserved_space(inode) -= number;
1980	else
1981		__inode_sub_bytes(inode, number);
1982	spin_unlock(&inode->i_lock);
1983
1984	if (reserve)
1985		goto out_unlock;
1986	mark_all_dquot_dirty(dquots);
1987out_unlock:
1988	srcu_read_unlock(&dquot_srcu, index);
1989	flush_warnings(warn);
1990}
1991EXPORT_SYMBOL(__dquot_free_space);
1992
1993/*
1994 * This operation can block, but only after everything is updated
1995 */
1996void dquot_free_inode(struct inode *inode)
1997{
1998	unsigned int cnt;
1999	struct dquot_warn warn[MAXQUOTAS];
2000	struct dquot * const *dquots;
2001	struct dquot *dquot;
2002	int index;
2003
2004	if (!inode_quota_active(inode))
2005		return;
2006
2007	dquots = i_dquot(inode);
2008	index = srcu_read_lock(&dquot_srcu);
2009	spin_lock(&inode->i_lock);
2010	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2011		int wtype;
2012		warn[cnt].w_type = QUOTA_NL_NOWARN;
2013		dquot = srcu_dereference(dquots[cnt], &dquot_srcu);
2014		if (!dquot)
2015			continue;
2016		spin_lock(&dquot->dq_dqb_lock);
2017		wtype = info_idq_free(dquot, 1);
2018		if (wtype != QUOTA_NL_NOWARN)
2019			prepare_warning(&warn[cnt], dquot, wtype);
2020		dquot_decr_inodes(dquot, 1);
2021		spin_unlock(&dquot->dq_dqb_lock);
2022	}
2023	spin_unlock(&inode->i_lock);
2024	mark_all_dquot_dirty(dquots);
2025	srcu_read_unlock(&dquot_srcu, index);
2026	flush_warnings(warn);
2027}
2028EXPORT_SYMBOL(dquot_free_inode);
2029
2030/*
2031 * Transfer the number of inode and blocks from one diskquota to an other.
2032 * On success, dquot references in transfer_to are consumed and references
2033 * to original dquots that need to be released are placed there. On failure,
2034 * references are kept untouched.
2035 *
2036 * This operation can block, but only after everything is updated
2037 * A transaction must be started when entering this function.
2038 *
2039 * We are holding reference on transfer_from & transfer_to, no need to
2040 * protect them by srcu_read_lock().
2041 */
2042int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
2043{
2044	qsize_t cur_space;
2045	qsize_t rsv_space = 0;
2046	qsize_t inode_usage = 1;
2047	struct dquot *transfer_from[MAXQUOTAS] = {};
2048	int cnt, index, ret = 0;
2049	char is_valid[MAXQUOTAS] = {};
2050	struct dquot_warn warn_to[MAXQUOTAS];
2051	struct dquot_warn warn_from_inodes[MAXQUOTAS];
2052	struct dquot_warn warn_from_space[MAXQUOTAS];
2053
2054	if (IS_NOQUOTA(inode))
2055		return 0;
2056
2057	if (inode->i_sb->dq_op->get_inode_usage) {
2058		ret = inode->i_sb->dq_op->get_inode_usage(inode, &inode_usage);
2059		if (ret)
2060			return ret;
2061	}
2062
2063	/* Initialize the arrays */
2064	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2065		warn_to[cnt].w_type = QUOTA_NL_NOWARN;
2066		warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
2067		warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
2068	}
2069
2070	spin_lock(&dq_data_lock);
2071	spin_lock(&inode->i_lock);
2072	if (IS_NOQUOTA(inode)) {	/* File without quota accounting? */
2073		spin_unlock(&inode->i_lock);
2074		spin_unlock(&dq_data_lock);
2075		return 0;
2076	}
2077	cur_space = __inode_get_bytes(inode);
2078	rsv_space = __inode_get_rsv_space(inode);
2079	/*
2080	 * Build the transfer_from list, check limits, and update usage in
2081	 * the target structures.
2082	 */
2083	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2084		/*
2085		 * Skip changes for same uid or gid or for turned off quota-type.
2086		 */
2087		if (!transfer_to[cnt])
2088			continue;
2089		/* Avoid races with quotaoff() */
2090		if (!sb_has_quota_active(inode->i_sb, cnt))
2091			continue;
2092		is_valid[cnt] = 1;
2093		transfer_from[cnt] = i_dquot(inode)[cnt];
2094		ret = dquot_add_inodes(transfer_to[cnt], inode_usage,
2095				       &warn_to[cnt]);
2096		if (ret)
2097			goto over_quota;
2098		ret = dquot_add_space(transfer_to[cnt], cur_space, rsv_space,
2099				      DQUOT_SPACE_WARN, &warn_to[cnt]);
2100		if (ret) {
2101			spin_lock(&transfer_to[cnt]->dq_dqb_lock);
2102			dquot_decr_inodes(transfer_to[cnt], inode_usage);
2103			spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2104			goto over_quota;
2105		}
2106	}
2107
2108	/* Decrease usage for source structures and update quota pointers */
2109	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2110		if (!is_valid[cnt])
2111			continue;
2112		/* Due to IO error we might not have transfer_from[] structure */
2113		if (transfer_from[cnt]) {
2114			int wtype;
2115
2116			spin_lock(&transfer_from[cnt]->dq_dqb_lock);
2117			wtype = info_idq_free(transfer_from[cnt], inode_usage);
2118			if (wtype != QUOTA_NL_NOWARN)
2119				prepare_warning(&warn_from_inodes[cnt],
2120						transfer_from[cnt], wtype);
2121			wtype = info_bdq_free(transfer_from[cnt],
2122					      cur_space + rsv_space);
2123			if (wtype != QUOTA_NL_NOWARN)
2124				prepare_warning(&warn_from_space[cnt],
2125						transfer_from[cnt], wtype);
2126			dquot_decr_inodes(transfer_from[cnt], inode_usage);
2127			dquot_decr_space(transfer_from[cnt], cur_space);
2128			dquot_free_reserved_space(transfer_from[cnt],
2129						  rsv_space);
2130			spin_unlock(&transfer_from[cnt]->dq_dqb_lock);
2131		}
2132		i_dquot(inode)[cnt] = transfer_to[cnt];
2133	}
2134	spin_unlock(&inode->i_lock);
2135	spin_unlock(&dq_data_lock);
2136
2137	/*
2138	 * These arrays are local and we hold dquot references so we don't need
2139	 * the srcu protection but still take dquot_srcu to avoid warning in
2140	 * mark_all_dquot_dirty().
2141	 */
2142	index = srcu_read_lock(&dquot_srcu);
2143	mark_all_dquot_dirty(transfer_from);
2144	mark_all_dquot_dirty(transfer_to);
2145	srcu_read_unlock(&dquot_srcu, index);
2146
2147	flush_warnings(warn_to);
2148	flush_warnings(warn_from_inodes);
2149	flush_warnings(warn_from_space);
2150	/* Pass back references to put */
2151	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2152		if (is_valid[cnt])
2153			transfer_to[cnt] = transfer_from[cnt];
2154	return 0;
2155over_quota:
2156	/* Back out changes we already did */
2157	for (cnt--; cnt >= 0; cnt--) {
2158		if (!is_valid[cnt])
2159			continue;
2160		spin_lock(&transfer_to[cnt]->dq_dqb_lock);
2161		dquot_decr_inodes(transfer_to[cnt], inode_usage);
2162		dquot_decr_space(transfer_to[cnt], cur_space);
2163		dquot_free_reserved_space(transfer_to[cnt], rsv_space);
2164		spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2165	}
2166	spin_unlock(&inode->i_lock);
2167	spin_unlock(&dq_data_lock);
2168	flush_warnings(warn_to);
2169	return ret;
2170}
2171EXPORT_SYMBOL(__dquot_transfer);
2172
2173/* Wrapper for transferring ownership of an inode for uid/gid only
2174 * Called from FSXXX_setattr()
2175 */
2176int dquot_transfer(struct inode *inode, struct iattr *iattr)
2177{
2178	struct dquot *transfer_to[MAXQUOTAS] = {};
2179	struct dquot *dquot;
2180	struct super_block *sb = inode->i_sb;
2181	int ret;
2182
2183	if (!inode_quota_active(inode))
2184		return 0;
2185
2186	if (iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)){
2187		dquot = dqget(sb, make_kqid_uid(iattr->ia_uid));
2188		if (IS_ERR(dquot)) {
2189			if (PTR_ERR(dquot) != -ESRCH) {
2190				ret = PTR_ERR(dquot);
2191				goto out_put;
2192			}
2193			dquot = NULL;
2194		}
2195		transfer_to[USRQUOTA] = dquot;
2196	}
2197	if (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid)){
2198		dquot = dqget(sb, make_kqid_gid(iattr->ia_gid));
2199		if (IS_ERR(dquot)) {
2200			if (PTR_ERR(dquot) != -ESRCH) {
2201				ret = PTR_ERR(dquot);
2202				goto out_put;
2203			}
2204			dquot = NULL;
2205		}
2206		transfer_to[GRPQUOTA] = dquot;
2207	}
2208	ret = __dquot_transfer(inode, transfer_to);
2209out_put:
2210	dqput_all(transfer_to);
2211	return ret;
2212}
2213EXPORT_SYMBOL(dquot_transfer);
2214
2215/*
2216 * Write info of quota file to disk
2217 */
2218int dquot_commit_info(struct super_block *sb, int type)
2219{
2220	struct quota_info *dqopt = sb_dqopt(sb);
2221
2222	return dqopt->ops[type]->write_file_info(sb, type);
2223}
2224EXPORT_SYMBOL(dquot_commit_info);
2225
2226int dquot_get_next_id(struct super_block *sb, struct kqid *qid)
2227{
2228	struct quota_info *dqopt = sb_dqopt(sb);
2229
2230	if (!sb_has_quota_active(sb, qid->type))
2231		return -ESRCH;
2232	if (!dqopt->ops[qid->type]->get_next_id)
2233		return -ENOSYS;
2234	return dqopt->ops[qid->type]->get_next_id(sb, qid);
2235}
2236EXPORT_SYMBOL(dquot_get_next_id);
2237
2238/*
2239 * Definitions of diskquota operations.
2240 */
2241const struct dquot_operations dquot_operations = {
2242	.write_dquot	= dquot_commit,
2243	.acquire_dquot	= dquot_acquire,
2244	.release_dquot	= dquot_release,
2245	.mark_dirty	= dquot_mark_dquot_dirty,
2246	.write_info	= dquot_commit_info,
2247	.alloc_dquot	= dquot_alloc,
2248	.destroy_dquot	= dquot_destroy,
2249	.get_next_id	= dquot_get_next_id,
2250};
2251EXPORT_SYMBOL(dquot_operations);
2252
2253/*
2254 * Generic helper for ->open on filesystems supporting disk quotas.
2255 */
2256int dquot_file_open(struct inode *inode, struct file *file)
2257{
2258	int error;
2259
2260	error = generic_file_open(inode, file);
2261	if (!error && (file->f_mode & FMODE_WRITE))
2262		error = dquot_initialize(inode);
2263	return error;
2264}
2265EXPORT_SYMBOL(dquot_file_open);
2266
2267static void vfs_cleanup_quota_inode(struct super_block *sb, int type)
2268{
2269	struct quota_info *dqopt = sb_dqopt(sb);
2270	struct inode *inode = dqopt->files[type];
2271
2272	if (!inode)
2273		return;
2274	if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2275		inode_lock(inode);
2276		inode->i_flags &= ~S_NOQUOTA;
2277		inode_unlock(inode);
2278	}
2279	dqopt->files[type] = NULL;
2280	iput(inode);
2281}
2282
2283/*
2284 * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
2285 */
2286int dquot_disable(struct super_block *sb, int type, unsigned int flags)
2287{
2288	int cnt;
2289	struct quota_info *dqopt = sb_dqopt(sb);
2290
2291	/* s_umount should be held in exclusive mode */
2292	if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2293		up_read(&sb->s_umount);
2294
2295	/* Cannot turn off usage accounting without turning off limits, or
2296	 * suspend quotas and simultaneously turn quotas off. */
2297	if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
2298	    || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
2299	    DQUOT_USAGE_ENABLED)))
2300		return -EINVAL;
2301
2302	/*
2303	 * Skip everything if there's nothing to do. We have to do this because
2304	 * sometimes we are called when fill_super() failed and calling
2305	 * sync_fs() in such cases does no good.
2306	 */
2307	if (!sb_any_quota_loaded(sb))
2308		return 0;
2309
2310	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2311		if (type != -1 && cnt != type)
2312			continue;
2313		if (!sb_has_quota_loaded(sb, cnt))
2314			continue;
2315
2316		if (flags & DQUOT_SUSPENDED) {
2317			spin_lock(&dq_state_lock);
2318			dqopt->flags |=
2319				dquot_state_flag(DQUOT_SUSPENDED, cnt);
2320			spin_unlock(&dq_state_lock);
2321		} else {
2322			spin_lock(&dq_state_lock);
2323			dqopt->flags &= ~dquot_state_flag(flags, cnt);
2324			/* Turning off suspended quotas? */
2325			if (!sb_has_quota_loaded(sb, cnt) &&
2326			    sb_has_quota_suspended(sb, cnt)) {
2327				dqopt->flags &=	~dquot_state_flag(
2328							DQUOT_SUSPENDED, cnt);
2329				spin_unlock(&dq_state_lock);
2330				vfs_cleanup_quota_inode(sb, cnt);
2331				continue;
2332			}
2333			spin_unlock(&dq_state_lock);
2334		}
2335
2336		/* We still have to keep quota loaded? */
2337		if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
2338			continue;
2339
2340		/* Note: these are blocking operations */
2341		drop_dquot_ref(sb, cnt);
2342		invalidate_dquots(sb, cnt);
2343		/*
2344		 * Now all dquots should be invalidated, all writes done so we
2345		 * should be only users of the info. No locks needed.
2346		 */
2347		if (info_dirty(&dqopt->info[cnt]))
2348			sb->dq_op->write_info(sb, cnt);
2349		if (dqopt->ops[cnt]->free_file_info)
2350			dqopt->ops[cnt]->free_file_info(sb, cnt);
2351		put_quota_format(dqopt->info[cnt].dqi_format);
2352		dqopt->info[cnt].dqi_flags = 0;
2353		dqopt->info[cnt].dqi_igrace = 0;
2354		dqopt->info[cnt].dqi_bgrace = 0;
2355		dqopt->ops[cnt] = NULL;
2356	}
2357
2358	/* Skip syncing and setting flags if quota files are hidden */
2359	if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2360		goto put_inodes;
2361
2362	/* Sync the superblock so that buffers with quota data are written to
2363	 * disk (and so userspace sees correct data afterwards). */
2364	if (sb->s_op->sync_fs)
2365		sb->s_op->sync_fs(sb, 1);
2366	sync_blockdev(sb->s_bdev);
2367	/* Now the quota files are just ordinary files and we can set the
2368	 * inode flags back. Moreover we discard the pagecache so that
2369	 * userspace sees the writes we did bypassing the pagecache. We
2370	 * must also discard the blockdev buffers so that we see the
2371	 * changes done by userspace on the next quotaon() */
2372	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2373		if (!sb_has_quota_loaded(sb, cnt) && dqopt->files[cnt]) {
2374			inode_lock(dqopt->files[cnt]);
2375			truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
2376			inode_unlock(dqopt->files[cnt]);
2377		}
2378	if (sb->s_bdev)
2379		invalidate_bdev(sb->s_bdev);
2380put_inodes:
2381	/* We are done when suspending quotas */
2382	if (flags & DQUOT_SUSPENDED)
2383		return 0;
2384
2385	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2386		if (!sb_has_quota_loaded(sb, cnt))
2387			vfs_cleanup_quota_inode(sb, cnt);
2388	return 0;
2389}
2390EXPORT_SYMBOL(dquot_disable);
2391
2392int dquot_quota_off(struct super_block *sb, int type)
2393{
2394	return dquot_disable(sb, type,
2395			     DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2396}
2397EXPORT_SYMBOL(dquot_quota_off);
2398
2399/*
2400 *	Turn quotas on on a device
2401 */
2402
2403static int vfs_setup_quota_inode(struct inode *inode, int type)
2404{
2405	struct super_block *sb = inode->i_sb;
2406	struct quota_info *dqopt = sb_dqopt(sb);
2407
2408	if (is_bad_inode(inode))
2409		return -EUCLEAN;
2410	if (!S_ISREG(inode->i_mode))
2411		return -EACCES;
2412	if (IS_RDONLY(inode))
2413		return -EROFS;
2414	if (sb_has_quota_loaded(sb, type))
2415		return -EBUSY;
2416
2417	/*
2418	 * Quota files should never be encrypted.  They should be thought of as
2419	 * filesystem metadata, not user data.  New-style internal quota files
2420	 * cannot be encrypted by users anyway, but old-style external quota
2421	 * files could potentially be incorrectly created in an encrypted
2422	 * directory, hence this explicit check.  Some reasons why encrypted
2423	 * quota files don't work include: (1) some filesystems that support
2424	 * encryption don't handle it in their quota_read and quota_write, and
2425	 * (2) cleaning up encrypted quota files at unmount would need special
2426	 * consideration, as quota files are cleaned up later than user files.
2427	 */
2428	if (IS_ENCRYPTED(inode))
2429		return -EINVAL;
2430
2431	dqopt->files[type] = igrab(inode);
2432	if (!dqopt->files[type])
2433		return -EIO;
2434	if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2435		/* We don't want quota and atime on quota files (deadlocks
2436		 * possible) Also nobody should write to the file - we use
2437		 * special IO operations which ignore the immutable bit. */
2438		inode_lock(inode);
2439		inode->i_flags |= S_NOQUOTA;
2440		inode_unlock(inode);
2441		/*
2442		 * When S_NOQUOTA is set, remove dquot references as no more
2443		 * references can be added
2444		 */
2445		__dquot_drop(inode);
2446	}
2447	return 0;
2448}
2449
2450int dquot_load_quota_sb(struct super_block *sb, int type, int format_id,
2451	unsigned int flags)
2452{
2453	struct quota_format_type *fmt = find_quota_format(format_id);
2454	struct quota_info *dqopt = sb_dqopt(sb);
2455	int error;
2456
2457	/* Just unsuspend quotas? */
2458	BUG_ON(flags & DQUOT_SUSPENDED);
2459	/* s_umount should be held in exclusive mode */
2460	if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2461		up_read(&sb->s_umount);
2462
2463	if (!fmt)
2464		return -ESRCH;
2465	if (!sb->s_op->quota_write || !sb->s_op->quota_read ||
2466	    (type == PRJQUOTA && sb->dq_op->get_projid == NULL)) {
2467		error = -EINVAL;
2468		goto out_fmt;
2469	}
2470	/* Filesystems outside of init_user_ns not yet supported */
2471	if (sb->s_user_ns != &init_user_ns) {
2472		error = -EINVAL;
2473		goto out_fmt;
2474	}
2475	/* Usage always has to be set... */
2476	if (!(flags & DQUOT_USAGE_ENABLED)) {
2477		error = -EINVAL;
2478		goto out_fmt;
2479	}
2480	if (sb_has_quota_loaded(sb, type)) {
2481		error = -EBUSY;
2482		goto out_fmt;
2483	}
2484
2485	if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2486		/* As we bypass the pagecache we must now flush all the
2487		 * dirty data and invalidate caches so that kernel sees
2488		 * changes from userspace. It is not enough to just flush
2489		 * the quota file since if blocksize < pagesize, invalidation
2490		 * of the cache could fail because of other unrelated dirty
2491		 * data */
2492		sync_filesystem(sb);
2493		invalidate_bdev(sb->s_bdev);
2494	}
2495
2496	error = -EINVAL;
2497	if (!fmt->qf_ops->check_quota_file(sb, type))
2498		goto out_fmt;
2499
2500	dqopt->ops[type] = fmt->qf_ops;
2501	dqopt->info[type].dqi_format = fmt;
2502	dqopt->info[type].dqi_fmt_id = format_id;
2503	INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2504	error = dqopt->ops[type]->read_file_info(sb, type);
2505	if (error < 0)
2506		goto out_fmt;
2507	if (dqopt->flags & DQUOT_QUOTA_SYS_FILE) {
2508		spin_lock(&dq_data_lock);
2509		dqopt->info[type].dqi_flags |= DQF_SYS_FILE;
2510		spin_unlock(&dq_data_lock);
2511	}
2512	spin_lock(&dq_state_lock);
2513	dqopt->flags |= dquot_state_flag(flags, type);
2514	spin_unlock(&dq_state_lock);
2515
2516	error = add_dquot_ref(sb, type);
2517	if (error)
2518		dquot_disable(sb, type,
2519			      DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2520
2521	return error;
2522out_fmt:
2523	put_quota_format(fmt);
2524
2525	return error;
2526}
2527EXPORT_SYMBOL(dquot_load_quota_sb);
2528
2529/*
2530 * More powerful function for turning on quotas on given quota inode allowing
2531 * setting of individual quota flags
2532 */
2533int dquot_load_quota_inode(struct inode *inode, int type, int format_id,
2534	unsigned int flags)
2535{
2536	int err;
2537
2538	err = vfs_setup_quota_inode(inode, type);
2539	if (err < 0)
2540		return err;
2541	err = dquot_load_quota_sb(inode->i_sb, type, format_id, flags);
2542	if (err < 0)
2543		vfs_cleanup_quota_inode(inode->i_sb, type);
2544	return err;
2545}
2546EXPORT_SYMBOL(dquot_load_quota_inode);
2547
2548/* Reenable quotas on remount RW */
2549int dquot_resume(struct super_block *sb, int type)
2550{
2551	struct quota_info *dqopt = sb_dqopt(sb);
2552	int ret = 0, cnt;
2553	unsigned int flags;
2554
2555	/* s_umount should be held in exclusive mode */
2556	if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2557		up_read(&sb->s_umount);
2558
2559	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2560		if (type != -1 && cnt != type)
2561			continue;
2562		if (!sb_has_quota_suspended(sb, cnt))
2563			continue;
2564
2565		spin_lock(&dq_state_lock);
2566		flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2567							DQUOT_LIMITS_ENABLED,
2568							cnt);
2569		dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2570		spin_unlock(&dq_state_lock);
2571
2572		flags = dquot_generic_flag(flags, cnt);
2573		ret = dquot_load_quota_sb(sb, cnt, dqopt->info[cnt].dqi_fmt_id,
2574					  flags);
2575		if (ret < 0)
2576			vfs_cleanup_quota_inode(sb, cnt);
2577	}
2578
2579	return ret;
2580}
2581EXPORT_SYMBOL(dquot_resume);
2582
2583int dquot_quota_on(struct super_block *sb, int type, int format_id,
2584		   const struct path *path)
2585{
2586	int error = security_quota_on(path->dentry);
2587	if (error)
2588		return error;
2589	/* Quota file not on the same filesystem? */
2590	if (path->dentry->d_sb != sb)
2591		error = -EXDEV;
2592	else
2593		error = dquot_load_quota_inode(d_inode(path->dentry), type,
2594					     format_id, DQUOT_USAGE_ENABLED |
2595					     DQUOT_LIMITS_ENABLED);
2596	return error;
2597}
2598EXPORT_SYMBOL(dquot_quota_on);
2599
2600/*
2601 * This function is used when filesystem needs to initialize quotas
2602 * during mount time.
2603 */
2604int dquot_quota_on_mount(struct super_block *sb, char *qf_name,
2605		int format_id, int type)
2606{
2607	struct dentry *dentry;
2608	int error;
2609
2610	dentry = lookup_positive_unlocked(qf_name, sb->s_root, strlen(qf_name));
2611	if (IS_ERR(dentry))
2612		return PTR_ERR(dentry);
2613
2614	error = security_quota_on(dentry);
2615	if (!error)
2616		error = dquot_load_quota_inode(d_inode(dentry), type, format_id,
2617				DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2618
2619	dput(dentry);
2620	return error;
2621}
2622EXPORT_SYMBOL(dquot_quota_on_mount);
2623
2624static int dquot_quota_enable(struct super_block *sb, unsigned int flags)
2625{
2626	int ret;
2627	int type;
2628	struct quota_info *dqopt = sb_dqopt(sb);
2629
2630	if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2631		return -ENOSYS;
2632	/* Accounting cannot be turned on while fs is mounted */
2633	flags &= ~(FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT);
2634	if (!flags)
2635		return -EINVAL;
2636	for (type = 0; type < MAXQUOTAS; type++) {
2637		if (!(flags & qtype_enforce_flag(type)))
2638			continue;
2639		/* Can't enforce without accounting */
2640		if (!sb_has_quota_usage_enabled(sb, type)) {
2641			ret = -EINVAL;
2642			goto out_err;
2643		}
2644		if (sb_has_quota_limits_enabled(sb, type)) {
2645			ret = -EBUSY;
2646			goto out_err;
2647		}
2648		spin_lock(&dq_state_lock);
2649		dqopt->flags |= dquot_state_flag(DQUOT_LIMITS_ENABLED, type);
2650		spin_unlock(&dq_state_lock);
2651	}
2652	return 0;
2653out_err:
2654	/* Backout enforcement enablement we already did */
2655	for (type--; type >= 0; type--)  {
2656		if (flags & qtype_enforce_flag(type))
2657			dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2658	}
2659	/* Error code translation for better compatibility with XFS */
2660	if (ret == -EBUSY)
2661		ret = -EEXIST;
2662	return ret;
2663}
2664
2665static int dquot_quota_disable(struct super_block *sb, unsigned int flags)
2666{
2667	int ret;
2668	int type;
2669	struct quota_info *dqopt = sb_dqopt(sb);
2670
2671	if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2672		return -ENOSYS;
2673	/*
2674	 * We don't support turning off accounting via quotactl. In principle
2675	 * quota infrastructure can do this but filesystems don't expect
2676	 * userspace to be able to do it.
2677	 */
2678	if (flags &
2679		  (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT))
2680		return -EOPNOTSUPP;
2681
2682	/* Filter out limits not enabled */
2683	for (type = 0; type < MAXQUOTAS; type++)
2684		if (!sb_has_quota_limits_enabled(sb, type))
2685			flags &= ~qtype_enforce_flag(type);
2686	/* Nothing left? */
2687	if (!flags)
2688		return -EEXIST;
2689	for (type = 0; type < MAXQUOTAS; type++) {
2690		if (flags & qtype_enforce_flag(type)) {
2691			ret = dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2692			if (ret < 0)
2693				goto out_err;
2694		}
2695	}
2696	return 0;
2697out_err:
2698	/* Backout enforcement disabling we already did */
2699	for (type--; type >= 0; type--)  {
2700		if (flags & qtype_enforce_flag(type)) {
2701			spin_lock(&dq_state_lock);
2702			dqopt->flags |=
2703				dquot_state_flag(DQUOT_LIMITS_ENABLED, type);
2704			spin_unlock(&dq_state_lock);
2705		}
2706	}
2707	return ret;
2708}
2709
2710/* Generic routine for getting common part of quota structure */
2711static void do_get_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2712{
2713	struct mem_dqblk *dm = &dquot->dq_dqb;
2714
2715	memset(di, 0, sizeof(*di));
2716	spin_lock(&dquot->dq_dqb_lock);
2717	di->d_spc_hardlimit = dm->dqb_bhardlimit;
2718	di->d_spc_softlimit = dm->dqb_bsoftlimit;
2719	di->d_ino_hardlimit = dm->dqb_ihardlimit;
2720	di->d_ino_softlimit = dm->dqb_isoftlimit;
2721	di->d_space = dm->dqb_curspace + dm->dqb_rsvspace;
2722	di->d_ino_count = dm->dqb_curinodes;
2723	di->d_spc_timer = dm->dqb_btime;
2724	di->d_ino_timer = dm->dqb_itime;
2725	spin_unlock(&dquot->dq_dqb_lock);
2726}
2727
2728int dquot_get_dqblk(struct super_block *sb, struct kqid qid,
2729		    struct qc_dqblk *di)
2730{
2731	struct dquot *dquot;
2732
2733	dquot = dqget(sb, qid);
2734	if (IS_ERR(dquot))
2735		return PTR_ERR(dquot);
2736	do_get_dqblk(dquot, di);
2737	dqput(dquot);
2738
2739	return 0;
2740}
2741EXPORT_SYMBOL(dquot_get_dqblk);
2742
2743int dquot_get_next_dqblk(struct super_block *sb, struct kqid *qid,
2744			 struct qc_dqblk *di)
2745{
2746	struct dquot *dquot;
2747	int err;
2748
2749	if (!sb->dq_op->get_next_id)
2750		return -ENOSYS;
2751	err = sb->dq_op->get_next_id(sb, qid);
2752	if (err < 0)
2753		return err;
2754	dquot = dqget(sb, *qid);
2755	if (IS_ERR(dquot))
2756		return PTR_ERR(dquot);
2757	do_get_dqblk(dquot, di);
2758	dqput(dquot);
2759
2760	return 0;
2761}
2762EXPORT_SYMBOL(dquot_get_next_dqblk);
2763
2764#define VFS_QC_MASK \
2765	(QC_SPACE | QC_SPC_SOFT | QC_SPC_HARD | \
2766	 QC_INO_COUNT | QC_INO_SOFT | QC_INO_HARD | \
2767	 QC_SPC_TIMER | QC_INO_TIMER)
2768
2769/* Generic routine for setting common part of quota structure */
2770static int do_set_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2771{
2772	struct mem_dqblk *dm = &dquot->dq_dqb;
2773	int check_blim = 0, check_ilim = 0;
2774	struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
2775
2776	if (di->d_fieldmask & ~VFS_QC_MASK)
2777		return -EINVAL;
2778
2779	if (((di->d_fieldmask & QC_SPC_SOFT) &&
2780	     di->d_spc_softlimit > dqi->dqi_max_spc_limit) ||
2781	    ((di->d_fieldmask & QC_SPC_HARD) &&
2782	     di->d_spc_hardlimit > dqi->dqi_max_spc_limit) ||
2783	    ((di->d_fieldmask & QC_INO_SOFT) &&
2784	     (di->d_ino_softlimit > dqi->dqi_max_ino_limit)) ||
2785	    ((di->d_fieldmask & QC_INO_HARD) &&
2786	     (di->d_ino_hardlimit > dqi->dqi_max_ino_limit)))
2787		return -ERANGE;
2788
2789	spin_lock(&dquot->dq_dqb_lock);
2790	if (di->d_fieldmask & QC_SPACE) {
2791		dm->dqb_curspace = di->d_space - dm->dqb_rsvspace;
2792		check_blim = 1;
2793		set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2794	}
2795
2796	if (di->d_fieldmask & QC_SPC_SOFT)
2797		dm->dqb_bsoftlimit = di->d_spc_softlimit;
2798	if (di->d_fieldmask & QC_SPC_HARD)
2799		dm->dqb_bhardlimit = di->d_spc_hardlimit;
2800	if (di->d_fieldmask & (QC_SPC_SOFT | QC_SPC_HARD)) {
2801		check_blim = 1;
2802		set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2803	}
2804
2805	if (di->d_fieldmask & QC_INO_COUNT) {
2806		dm->dqb_curinodes = di->d_ino_count;
2807		check_ilim = 1;
2808		set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2809	}
2810
2811	if (di->d_fieldmask & QC_INO_SOFT)
2812		dm->dqb_isoftlimit = di->d_ino_softlimit;
2813	if (di->d_fieldmask & QC_INO_HARD)
2814		dm->dqb_ihardlimit = di->d_ino_hardlimit;
2815	if (di->d_fieldmask & (QC_INO_SOFT | QC_INO_HARD)) {
2816		check_ilim = 1;
2817		set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2818	}
2819
2820	if (di->d_fieldmask & QC_SPC_TIMER) {
2821		dm->dqb_btime = di->d_spc_timer;
2822		check_blim = 1;
2823		set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2824	}
2825
2826	if (di->d_fieldmask & QC_INO_TIMER) {
2827		dm->dqb_itime = di->d_ino_timer;
2828		check_ilim = 1;
2829		set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2830	}
2831
2832	if (check_blim) {
2833		if (!dm->dqb_bsoftlimit ||
2834		    dm->dqb_curspace + dm->dqb_rsvspace <= dm->dqb_bsoftlimit) {
2835			dm->dqb_btime = 0;
2836			clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2837		} else if (!(di->d_fieldmask & QC_SPC_TIMER))
2838			/* Set grace only if user hasn't provided his own... */
2839			dm->dqb_btime = ktime_get_real_seconds() + dqi->dqi_bgrace;
2840	}
2841	if (check_ilim) {
2842		if (!dm->dqb_isoftlimit ||
2843		    dm->dqb_curinodes <= dm->dqb_isoftlimit) {
2844			dm->dqb_itime = 0;
2845			clear_bit(DQ_INODES_B, &dquot->dq_flags);
2846		} else if (!(di->d_fieldmask & QC_INO_TIMER))
2847			/* Set grace only if user hasn't provided his own... */
2848			dm->dqb_itime = ktime_get_real_seconds() + dqi->dqi_igrace;
2849	}
2850	if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2851	    dm->dqb_isoftlimit)
2852		clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2853	else
2854		set_bit(DQ_FAKE_B, &dquot->dq_flags);
2855	spin_unlock(&dquot->dq_dqb_lock);
2856	mark_dquot_dirty(dquot);
2857
2858	return 0;
2859}
2860
2861int dquot_set_dqblk(struct super_block *sb, struct kqid qid,
2862		  struct qc_dqblk *di)
2863{
2864	struct dquot *dquot;
2865	int rc;
2866
2867	dquot = dqget(sb, qid);
2868	if (IS_ERR(dquot)) {
2869		rc = PTR_ERR(dquot);
2870		goto out;
2871	}
2872	rc = do_set_dqblk(dquot, di);
2873	dqput(dquot);
2874out:
2875	return rc;
2876}
2877EXPORT_SYMBOL(dquot_set_dqblk);
2878
2879/* Generic routine for getting common part of quota file information */
2880int dquot_get_state(struct super_block *sb, struct qc_state *state)
2881{
2882	struct mem_dqinfo *mi;
2883	struct qc_type_state *tstate;
2884	struct quota_info *dqopt = sb_dqopt(sb);
2885	int type;
2886
2887	memset(state, 0, sizeof(*state));
2888	for (type = 0; type < MAXQUOTAS; type++) {
2889		if (!sb_has_quota_active(sb, type))
2890			continue;
2891		tstate = state->s_state + type;
2892		mi = sb_dqopt(sb)->info + type;
2893		tstate->flags = QCI_ACCT_ENABLED;
2894		spin_lock(&dq_data_lock);
2895		if (mi->dqi_flags & DQF_SYS_FILE)
2896			tstate->flags |= QCI_SYSFILE;
2897		if (mi->dqi_flags & DQF_ROOT_SQUASH)
2898			tstate->flags |= QCI_ROOT_SQUASH;
2899		if (sb_has_quota_limits_enabled(sb, type))
2900			tstate->flags |= QCI_LIMITS_ENFORCED;
2901		tstate->spc_timelimit = mi->dqi_bgrace;
2902		tstate->ino_timelimit = mi->dqi_igrace;
2903		if (dqopt->files[type]) {
2904			tstate->ino = dqopt->files[type]->i_ino;
2905			tstate->blocks = dqopt->files[type]->i_blocks;
2906		}
2907		tstate->nextents = 1;	/* We don't know... */
2908		spin_unlock(&dq_data_lock);
2909	}
2910	return 0;
2911}
2912EXPORT_SYMBOL(dquot_get_state);
2913
2914/* Generic routine for setting common part of quota file information */
2915int dquot_set_dqinfo(struct super_block *sb, int type, struct qc_info *ii)
2916{
2917	struct mem_dqinfo *mi;
2918	int err = 0;
2919
2920	if ((ii->i_fieldmask & QC_WARNS_MASK) ||
2921	    (ii->i_fieldmask & QC_RT_SPC_TIMER))
2922		return -EINVAL;
2923	if (!sb_has_quota_active(sb, type))
2924		return -ESRCH;
2925	mi = sb_dqopt(sb)->info + type;
2926	if (ii->i_fieldmask & QC_FLAGS) {
2927		if ((ii->i_flags & QCI_ROOT_SQUASH &&
2928		     mi->dqi_format->qf_fmt_id != QFMT_VFS_OLD))
2929			return -EINVAL;
2930	}
2931	spin_lock(&dq_data_lock);
2932	if (ii->i_fieldmask & QC_SPC_TIMER)
2933		mi->dqi_bgrace = ii->i_spc_timelimit;
2934	if (ii->i_fieldmask & QC_INO_TIMER)
2935		mi->dqi_igrace = ii->i_ino_timelimit;
2936	if (ii->i_fieldmask & QC_FLAGS) {
2937		if (ii->i_flags & QCI_ROOT_SQUASH)
2938			mi->dqi_flags |= DQF_ROOT_SQUASH;
2939		else
2940			mi->dqi_flags &= ~DQF_ROOT_SQUASH;
2941	}
2942	spin_unlock(&dq_data_lock);
2943	mark_info_dirty(sb, type);
2944	/* Force write to disk */
2945	sb->dq_op->write_info(sb, type);
2946	return err;
2947}
2948EXPORT_SYMBOL(dquot_set_dqinfo);
2949
2950const struct quotactl_ops dquot_quotactl_sysfile_ops = {
2951	.quota_enable	= dquot_quota_enable,
2952	.quota_disable	= dquot_quota_disable,
2953	.quota_sync	= dquot_quota_sync,
2954	.get_state	= dquot_get_state,
2955	.set_info	= dquot_set_dqinfo,
2956	.get_dqblk	= dquot_get_dqblk,
2957	.get_nextdqblk	= dquot_get_next_dqblk,
2958	.set_dqblk	= dquot_set_dqblk
2959};
2960EXPORT_SYMBOL(dquot_quotactl_sysfile_ops);
2961
2962static int do_proc_dqstats(struct ctl_table *table, int write,
2963		     void *buffer, size_t *lenp, loff_t *ppos)
2964{
2965	unsigned int type = (unsigned long *)table->data - dqstats.stat;
2966	s64 value = percpu_counter_sum(&dqstats.counter[type]);
2967
2968	/* Filter negative values for non-monotonic counters */
2969	if (value < 0 && (type == DQST_ALLOC_DQUOTS ||
2970			  type == DQST_FREE_DQUOTS))
2971		value = 0;
2972
2973	/* Update global table */
2974	dqstats.stat[type] = value;
2975	return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
2976}
2977
2978static struct ctl_table fs_dqstats_table[] = {
2979	{
2980		.procname	= "lookups",
2981		.data		= &dqstats.stat[DQST_LOOKUPS],
2982		.maxlen		= sizeof(unsigned long),
2983		.mode		= 0444,
2984		.proc_handler	= do_proc_dqstats,
2985	},
2986	{
2987		.procname	= "drops",
2988		.data		= &dqstats.stat[DQST_DROPS],
2989		.maxlen		= sizeof(unsigned long),
2990		.mode		= 0444,
2991		.proc_handler	= do_proc_dqstats,
2992	},
2993	{
2994		.procname	= "reads",
2995		.data		= &dqstats.stat[DQST_READS],
2996		.maxlen		= sizeof(unsigned long),
2997		.mode		= 0444,
2998		.proc_handler	= do_proc_dqstats,
2999	},
3000	{
3001		.procname	= "writes",
3002		.data		= &dqstats.stat[DQST_WRITES],
3003		.maxlen		= sizeof(unsigned long),
3004		.mode		= 0444,
3005		.proc_handler	= do_proc_dqstats,
3006	},
3007	{
3008		.procname	= "cache_hits",
3009		.data		= &dqstats.stat[DQST_CACHE_HITS],
3010		.maxlen		= sizeof(unsigned long),
3011		.mode		= 0444,
3012		.proc_handler	= do_proc_dqstats,
3013	},
3014	{
3015		.procname	= "allocated_dquots",
3016		.data		= &dqstats.stat[DQST_ALLOC_DQUOTS],
3017		.maxlen		= sizeof(unsigned long),
3018		.mode		= 0444,
3019		.proc_handler	= do_proc_dqstats,
3020	},
3021	{
3022		.procname	= "free_dquots",
3023		.data		= &dqstats.stat[DQST_FREE_DQUOTS],
3024		.maxlen		= sizeof(unsigned long),
3025		.mode		= 0444,
3026		.proc_handler	= do_proc_dqstats,
3027	},
3028	{
3029		.procname	= "syncs",
3030		.data		= &dqstats.stat[DQST_SYNCS],
3031		.maxlen		= sizeof(unsigned long),
3032		.mode		= 0444,
3033		.proc_handler	= do_proc_dqstats,
3034	},
3035#ifdef CONFIG_PRINT_QUOTA_WARNING
3036	{
3037		.procname	= "warnings",
3038		.data		= &flag_print_warnings,
3039		.maxlen		= sizeof(int),
3040		.mode		= 0644,
3041		.proc_handler	= proc_dointvec,
3042	},
3043#endif
3044	{ },
3045};
3046
3047static struct ctl_table fs_table[] = {
3048	{
3049		.procname	= "quota",
3050		.mode		= 0555,
3051		.child		= fs_dqstats_table,
3052	},
3053	{ },
3054};
3055
3056static struct ctl_table sys_table[] = {
3057	{
3058		.procname	= "fs",
3059		.mode		= 0555,
3060		.child		= fs_table,
3061	},
3062	{ },
3063};
3064
3065static int __init dquot_init(void)
3066{
3067	int i, ret;
3068	unsigned long nr_hash, order;
3069
3070	printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
3071
3072	register_sysctl_table(sys_table);
3073
3074	dquot_cachep = kmem_cache_create("dquot",
3075			sizeof(struct dquot), sizeof(unsigned long) * 4,
3076			(SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
3077				SLAB_MEM_SPREAD|SLAB_PANIC),
3078			NULL);
3079
3080	order = 0;
3081	dquot_hash = (struct hlist_head *)__get_free_pages(GFP_KERNEL, order);
3082	if (!dquot_hash)
3083		panic("Cannot create dquot hash table");
3084
3085	for (i = 0; i < _DQST_DQSTAT_LAST; i++) {
3086		ret = percpu_counter_init(&dqstats.counter[i], 0, GFP_KERNEL);
3087		if (ret)
3088			panic("Cannot create dquot stat counters");
3089	}
3090
3091	/* Find power-of-two hlist_heads which can fit into allocation */
3092	nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
3093	dq_hash_bits = ilog2(nr_hash);
3094
3095	nr_hash = 1UL << dq_hash_bits;
3096	dq_hash_mask = nr_hash - 1;
3097	for (i = 0; i < nr_hash; i++)
3098		INIT_HLIST_HEAD(dquot_hash + i);
3099
3100	pr_info("VFS: Dquot-cache hash table entries: %ld (order %ld,"
3101		" %ld bytes)\n", nr_hash, order, (PAGE_SIZE << order));
3102
3103	if (register_shrinker(&dqcache_shrinker))
3104		panic("Cannot register dquot shrinker");
3105
3106	return 0;
3107}
3108fs_initcall(dquot_init);
3109