xref: /linux/fs/gfs2/quota.c (revision 2a2c74b2efcb1a0ca3fdcb5fbb96ad8de6a29177)
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2007 Red Hat, Inc.  All rights reserved.
4  *
5  * This copyrighted material is made available to anyone wishing to use,
6  * modify, copy, or redistribute it subject to the terms and conditions
7  * of the GNU General Public License version 2.
8  */
9 
10 /*
11  * Quota change tags are associated with each transaction that allocates or
12  * deallocates space.  Those changes are accumulated locally to each node (in a
13  * per-node file) and then are periodically synced to the quota file.  This
14  * avoids the bottleneck of constantly touching the quota file, but introduces
15  * fuzziness in the current usage value of IDs that are being used on different
16  * nodes in the cluster simultaneously.  So, it is possible for a user on
17  * multiple nodes to overrun their quota, but that overrun is controlable.
18  * Since quota tags are part of transactions, there is no need for a quota check
19  * program to be run on node crashes or anything like that.
20  *
21  * There are couple of knobs that let the administrator manage the quota
22  * fuzziness.  "quota_quantum" sets the maximum time a quota change can be
23  * sitting on one node before being synced to the quota file.  (The default is
24  * 60 seconds.)  Another knob, "quota_scale" controls how quickly the frequency
25  * of quota file syncs increases as the user moves closer to their limit.  The
26  * more frequent the syncs, the more accurate the quota enforcement, but that
27  * means that there is more contention between the nodes for the quota file.
28  * The default value is one.  This sets the maximum theoretical quota overrun
29  * (with infinite node with infinite bandwidth) to twice the user's limit.  (In
30  * practice, the maximum overrun you see should be much less.)  A "quota_scale"
31  * number greater than one makes quota syncs more frequent and reduces the
32  * maximum overrun.  Numbers less than one (but greater than zero) make quota
33  * syncs less frequent.
34  *
35  * GFS quotas also use per-ID Lock Value Blocks (LVBs) to cache the contents of
36  * the quota file, so it is not being constantly read.
37  */
38 
39 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
40 
41 #include <linux/sched.h>
42 #include <linux/slab.h>
43 #include <linux/mm.h>
44 #include <linux/spinlock.h>
45 #include <linux/completion.h>
46 #include <linux/buffer_head.h>
47 #include <linux/sort.h>
48 #include <linux/fs.h>
49 #include <linux/bio.h>
50 #include <linux/gfs2_ondisk.h>
51 #include <linux/kthread.h>
52 #include <linux/freezer.h>
53 #include <linux/quota.h>
54 #include <linux/dqblk_xfs.h>
55 #include <linux/lockref.h>
56 #include <linux/list_lru.h>
57 #include <linux/rcupdate.h>
58 #include <linux/rculist_bl.h>
59 #include <linux/bit_spinlock.h>
60 #include <linux/jhash.h>
61 #include <linux/vmalloc.h>
62 
63 #include "gfs2.h"
64 #include "incore.h"
65 #include "bmap.h"
66 #include "glock.h"
67 #include "glops.h"
68 #include "log.h"
69 #include "meta_io.h"
70 #include "quota.h"
71 #include "rgrp.h"
72 #include "super.h"
73 #include "trans.h"
74 #include "inode.h"
75 #include "util.h"
76 
77 #define GFS2_QD_HASH_SHIFT      12
78 #define GFS2_QD_HASH_SIZE       (1 << GFS2_QD_HASH_SHIFT)
79 #define GFS2_QD_HASH_MASK       (GFS2_QD_HASH_SIZE - 1)
80 
81 /* Lock order: qd_lock -> bucket lock -> qd->lockref.lock -> lru lock */
82 /*                     -> sd_bitmap_lock                              */
83 static DEFINE_SPINLOCK(qd_lock);
84 struct list_lru gfs2_qd_lru;
85 
86 static struct hlist_bl_head qd_hash_table[GFS2_QD_HASH_SIZE];
87 
88 static unsigned int gfs2_qd_hash(const struct gfs2_sbd *sdp,
89 				 const struct kqid qid)
90 {
91 	unsigned int h;
92 
93 	h = jhash(&sdp, sizeof(struct gfs2_sbd *), 0);
94 	h = jhash(&qid, sizeof(struct kqid), h);
95 
96 	return h & GFS2_QD_HASH_MASK;
97 }
98 
99 static inline void spin_lock_bucket(unsigned int hash)
100 {
101         hlist_bl_lock(&qd_hash_table[hash]);
102 }
103 
104 static inline void spin_unlock_bucket(unsigned int hash)
105 {
106         hlist_bl_unlock(&qd_hash_table[hash]);
107 }
108 
109 static void gfs2_qd_dealloc(struct rcu_head *rcu)
110 {
111 	struct gfs2_quota_data *qd = container_of(rcu, struct gfs2_quota_data, qd_rcu);
112 	kmem_cache_free(gfs2_quotad_cachep, qd);
113 }
114 
115 static void gfs2_qd_dispose(struct list_head *list)
116 {
117 	struct gfs2_quota_data *qd;
118 	struct gfs2_sbd *sdp;
119 
120 	while (!list_empty(list)) {
121 		qd = list_entry(list->next, struct gfs2_quota_data, qd_lru);
122 		sdp = qd->qd_gl->gl_sbd;
123 
124 		list_del(&qd->qd_lru);
125 
126 		/* Free from the filesystem-specific list */
127 		spin_lock(&qd_lock);
128 		list_del(&qd->qd_list);
129 		spin_unlock(&qd_lock);
130 
131 		spin_lock_bucket(qd->qd_hash);
132 		hlist_bl_del_rcu(&qd->qd_hlist);
133 		spin_unlock_bucket(qd->qd_hash);
134 
135 		gfs2_assert_warn(sdp, !qd->qd_change);
136 		gfs2_assert_warn(sdp, !qd->qd_slot_count);
137 		gfs2_assert_warn(sdp, !qd->qd_bh_count);
138 
139 		gfs2_glock_put(qd->qd_gl);
140 		atomic_dec(&sdp->sd_quota_count);
141 
142 		/* Delete it from the common reclaim list */
143 		call_rcu(&qd->qd_rcu, gfs2_qd_dealloc);
144 	}
145 }
146 
147 
148 static enum lru_status gfs2_qd_isolate(struct list_head *item, spinlock_t *lock, void *arg)
149 {
150 	struct list_head *dispose = arg;
151 	struct gfs2_quota_data *qd = list_entry(item, struct gfs2_quota_data, qd_lru);
152 
153 	if (!spin_trylock(&qd->qd_lockref.lock))
154 		return LRU_SKIP;
155 
156 	if (qd->qd_lockref.count == 0) {
157 		lockref_mark_dead(&qd->qd_lockref);
158 		list_move(&qd->qd_lru, dispose);
159 	}
160 
161 	spin_unlock(&qd->qd_lockref.lock);
162 	return LRU_REMOVED;
163 }
164 
165 static unsigned long gfs2_qd_shrink_scan(struct shrinker *shrink,
166 					 struct shrink_control *sc)
167 {
168 	LIST_HEAD(dispose);
169 	unsigned long freed;
170 
171 	if (!(sc->gfp_mask & __GFP_FS))
172 		return SHRINK_STOP;
173 
174 	freed = list_lru_walk_node(&gfs2_qd_lru, sc->nid, gfs2_qd_isolate,
175 				   &dispose, &sc->nr_to_scan);
176 
177 	gfs2_qd_dispose(&dispose);
178 
179 	return freed;
180 }
181 
182 static unsigned long gfs2_qd_shrink_count(struct shrinker *shrink,
183 					  struct shrink_control *sc)
184 {
185 	return vfs_pressure_ratio(list_lru_count_node(&gfs2_qd_lru, sc->nid));
186 }
187 
188 struct shrinker gfs2_qd_shrinker = {
189 	.count_objects = gfs2_qd_shrink_count,
190 	.scan_objects = gfs2_qd_shrink_scan,
191 	.seeks = DEFAULT_SEEKS,
192 	.flags = SHRINKER_NUMA_AWARE,
193 };
194 
195 
196 static u64 qd2index(struct gfs2_quota_data *qd)
197 {
198 	struct kqid qid = qd->qd_id;
199 	return (2 * (u64)from_kqid(&init_user_ns, qid)) +
200 		((qid.type == USRQUOTA) ? 0 : 1);
201 }
202 
203 static u64 qd2offset(struct gfs2_quota_data *qd)
204 {
205 	u64 offset;
206 
207 	offset = qd2index(qd);
208 	offset *= sizeof(struct gfs2_quota);
209 
210 	return offset;
211 }
212 
213 static struct gfs2_quota_data *qd_alloc(unsigned hash, struct gfs2_sbd *sdp, struct kqid qid)
214 {
215 	struct gfs2_quota_data *qd;
216 	int error;
217 
218 	qd = kmem_cache_zalloc(gfs2_quotad_cachep, GFP_NOFS);
219 	if (!qd)
220 		return NULL;
221 
222 	qd->qd_sbd = sdp;
223 	qd->qd_lockref.count = 1;
224 	spin_lock_init(&qd->qd_lockref.lock);
225 	qd->qd_id = qid;
226 	qd->qd_slot = -1;
227 	INIT_LIST_HEAD(&qd->qd_lru);
228 	qd->qd_hash = hash;
229 
230 	error = gfs2_glock_get(sdp, qd2index(qd),
231 			      &gfs2_quota_glops, CREATE, &qd->qd_gl);
232 	if (error)
233 		goto fail;
234 
235 	return qd;
236 
237 fail:
238 	kmem_cache_free(gfs2_quotad_cachep, qd);
239 	return NULL;
240 }
241 
242 static struct gfs2_quota_data *gfs2_qd_search_bucket(unsigned int hash,
243 						     const struct gfs2_sbd *sdp,
244 						     struct kqid qid)
245 {
246 	struct gfs2_quota_data *qd;
247 	struct hlist_bl_node *h;
248 
249 	hlist_bl_for_each_entry_rcu(qd, h, &qd_hash_table[hash], qd_hlist) {
250 		if (!qid_eq(qd->qd_id, qid))
251 			continue;
252 		if (qd->qd_sbd != sdp)
253 			continue;
254 		if (lockref_get_not_dead(&qd->qd_lockref)) {
255 			list_lru_del(&gfs2_qd_lru, &qd->qd_lru);
256 			return qd;
257 		}
258 	}
259 
260 	return NULL;
261 }
262 
263 
264 static int qd_get(struct gfs2_sbd *sdp, struct kqid qid,
265 		  struct gfs2_quota_data **qdp)
266 {
267 	struct gfs2_quota_data *qd, *new_qd;
268 	unsigned int hash = gfs2_qd_hash(sdp, qid);
269 
270 	rcu_read_lock();
271 	*qdp = qd = gfs2_qd_search_bucket(hash, sdp, qid);
272 	rcu_read_unlock();
273 
274 	if (qd)
275 		return 0;
276 
277 	new_qd = qd_alloc(hash, sdp, qid);
278 	if (!new_qd)
279 		return -ENOMEM;
280 
281 	spin_lock(&qd_lock);
282 	spin_lock_bucket(hash);
283 	*qdp = qd = gfs2_qd_search_bucket(hash, sdp, qid);
284 	if (qd == NULL) {
285 		*qdp = new_qd;
286 		list_add(&new_qd->qd_list, &sdp->sd_quota_list);
287 		hlist_bl_add_head_rcu(&new_qd->qd_hlist, &qd_hash_table[hash]);
288 		atomic_inc(&sdp->sd_quota_count);
289 	}
290 	spin_unlock_bucket(hash);
291 	spin_unlock(&qd_lock);
292 
293 	if (qd) {
294 		gfs2_glock_put(new_qd->qd_gl);
295 		kmem_cache_free(gfs2_quotad_cachep, new_qd);
296 	}
297 
298 	return 0;
299 }
300 
301 
302 static void qd_hold(struct gfs2_quota_data *qd)
303 {
304 	struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
305 	gfs2_assert(sdp, !__lockref_is_dead(&qd->qd_lockref));
306 	lockref_get(&qd->qd_lockref);
307 }
308 
309 static void qd_put(struct gfs2_quota_data *qd)
310 {
311 	if (lockref_put_or_lock(&qd->qd_lockref))
312 		return;
313 
314 	qd->qd_lockref.count = 0;
315 	list_lru_add(&gfs2_qd_lru, &qd->qd_lru);
316 	spin_unlock(&qd->qd_lockref.lock);
317 
318 }
319 
320 static int slot_get(struct gfs2_quota_data *qd)
321 {
322 	struct gfs2_sbd *sdp = qd->qd_sbd;
323 	unsigned int bit;
324 	int error = 0;
325 
326 	spin_lock(&sdp->sd_bitmap_lock);
327 	if (qd->qd_slot_count != 0)
328 		goto out;
329 
330 	error = -ENOSPC;
331 	bit = find_first_zero_bit(sdp->sd_quota_bitmap, sdp->sd_quota_slots);
332 	if (bit < sdp->sd_quota_slots) {
333 		set_bit(bit, sdp->sd_quota_bitmap);
334 		qd->qd_slot = bit;
335 		error = 0;
336 out:
337 		qd->qd_slot_count++;
338 	}
339 	spin_unlock(&sdp->sd_bitmap_lock);
340 
341 	return error;
342 }
343 
344 static void slot_hold(struct gfs2_quota_data *qd)
345 {
346 	struct gfs2_sbd *sdp = qd->qd_sbd;
347 
348 	spin_lock(&sdp->sd_bitmap_lock);
349 	gfs2_assert(sdp, qd->qd_slot_count);
350 	qd->qd_slot_count++;
351 	spin_unlock(&sdp->sd_bitmap_lock);
352 }
353 
354 static void slot_put(struct gfs2_quota_data *qd)
355 {
356 	struct gfs2_sbd *sdp = qd->qd_sbd;
357 
358 	spin_lock(&sdp->sd_bitmap_lock);
359 	gfs2_assert(sdp, qd->qd_slot_count);
360 	if (!--qd->qd_slot_count) {
361 		BUG_ON(!test_and_clear_bit(qd->qd_slot, sdp->sd_quota_bitmap));
362 		qd->qd_slot = -1;
363 	}
364 	spin_unlock(&sdp->sd_bitmap_lock);
365 }
366 
367 static int bh_get(struct gfs2_quota_data *qd)
368 {
369 	struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
370 	struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
371 	unsigned int block, offset;
372 	struct buffer_head *bh;
373 	int error;
374 	struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
375 
376 	mutex_lock(&sdp->sd_quota_mutex);
377 
378 	if (qd->qd_bh_count++) {
379 		mutex_unlock(&sdp->sd_quota_mutex);
380 		return 0;
381 	}
382 
383 	block = qd->qd_slot / sdp->sd_qc_per_block;
384 	offset = qd->qd_slot % sdp->sd_qc_per_block;
385 
386 	bh_map.b_size = 1 << ip->i_inode.i_blkbits;
387 	error = gfs2_block_map(&ip->i_inode, block, &bh_map, 0);
388 	if (error)
389 		goto fail;
390 	error = gfs2_meta_read(ip->i_gl, bh_map.b_blocknr, DIO_WAIT, &bh);
391 	if (error)
392 		goto fail;
393 	error = -EIO;
394 	if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC))
395 		goto fail_brelse;
396 
397 	qd->qd_bh = bh;
398 	qd->qd_bh_qc = (struct gfs2_quota_change *)
399 		(bh->b_data + sizeof(struct gfs2_meta_header) +
400 		 offset * sizeof(struct gfs2_quota_change));
401 
402 	mutex_unlock(&sdp->sd_quota_mutex);
403 
404 	return 0;
405 
406 fail_brelse:
407 	brelse(bh);
408 fail:
409 	qd->qd_bh_count--;
410 	mutex_unlock(&sdp->sd_quota_mutex);
411 	return error;
412 }
413 
414 static void bh_put(struct gfs2_quota_data *qd)
415 {
416 	struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
417 
418 	mutex_lock(&sdp->sd_quota_mutex);
419 	gfs2_assert(sdp, qd->qd_bh_count);
420 	if (!--qd->qd_bh_count) {
421 		brelse(qd->qd_bh);
422 		qd->qd_bh = NULL;
423 		qd->qd_bh_qc = NULL;
424 	}
425 	mutex_unlock(&sdp->sd_quota_mutex);
426 }
427 
428 static int qd_check_sync(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd,
429 			 u64 *sync_gen)
430 {
431 	if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
432 	    !test_bit(QDF_CHANGE, &qd->qd_flags) ||
433 	    (sync_gen && (qd->qd_sync_gen >= *sync_gen)))
434 		return 0;
435 
436 	if (!lockref_get_not_dead(&qd->qd_lockref))
437 		return 0;
438 
439 	list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
440 	set_bit(QDF_LOCKED, &qd->qd_flags);
441 	qd->qd_change_sync = qd->qd_change;
442 	slot_hold(qd);
443 	return 1;
444 }
445 
446 static int qd_fish(struct gfs2_sbd *sdp, struct gfs2_quota_data **qdp)
447 {
448 	struct gfs2_quota_data *qd = NULL;
449 	int error;
450 	int found = 0;
451 
452 	*qdp = NULL;
453 
454 	if (sdp->sd_vfs->s_flags & MS_RDONLY)
455 		return 0;
456 
457 	spin_lock(&qd_lock);
458 
459 	list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
460 		found = qd_check_sync(sdp, qd, &sdp->sd_quota_sync_gen);
461 		if (found)
462 			break;
463 	}
464 
465 	if (!found)
466 		qd = NULL;
467 
468 	spin_unlock(&qd_lock);
469 
470 	if (qd) {
471 		gfs2_assert_warn(sdp, qd->qd_change_sync);
472 		error = bh_get(qd);
473 		if (error) {
474 			clear_bit(QDF_LOCKED, &qd->qd_flags);
475 			slot_put(qd);
476 			qd_put(qd);
477 			return error;
478 		}
479 	}
480 
481 	*qdp = qd;
482 
483 	return 0;
484 }
485 
486 static void qd_unlock(struct gfs2_quota_data *qd)
487 {
488 	gfs2_assert_warn(qd->qd_gl->gl_sbd,
489 			 test_bit(QDF_LOCKED, &qd->qd_flags));
490 	clear_bit(QDF_LOCKED, &qd->qd_flags);
491 	bh_put(qd);
492 	slot_put(qd);
493 	qd_put(qd);
494 }
495 
496 static int qdsb_get(struct gfs2_sbd *sdp, struct kqid qid,
497 		    struct gfs2_quota_data **qdp)
498 {
499 	int error;
500 
501 	error = qd_get(sdp, qid, qdp);
502 	if (error)
503 		return error;
504 
505 	error = slot_get(*qdp);
506 	if (error)
507 		goto fail;
508 
509 	error = bh_get(*qdp);
510 	if (error)
511 		goto fail_slot;
512 
513 	return 0;
514 
515 fail_slot:
516 	slot_put(*qdp);
517 fail:
518 	qd_put(*qdp);
519 	return error;
520 }
521 
522 static void qdsb_put(struct gfs2_quota_data *qd)
523 {
524 	bh_put(qd);
525 	slot_put(qd);
526 	qd_put(qd);
527 }
528 
529 int gfs2_quota_hold(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
530 {
531 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
532 	struct gfs2_quota_data **qd;
533 	int error;
534 
535 	if (ip->i_res == NULL) {
536 		error = gfs2_rs_alloc(ip);
537 		if (error)
538 			return error;
539 	}
540 
541 	qd = ip->i_res->rs_qa_qd;
542 
543 	if (gfs2_assert_warn(sdp, !ip->i_res->rs_qa_qd_num) ||
544 	    gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags)))
545 		return -EIO;
546 
547 	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
548 		return 0;
549 
550 	error = qdsb_get(sdp, make_kqid_uid(ip->i_inode.i_uid), qd);
551 	if (error)
552 		goto out;
553 	ip->i_res->rs_qa_qd_num++;
554 	qd++;
555 
556 	error = qdsb_get(sdp, make_kqid_gid(ip->i_inode.i_gid), qd);
557 	if (error)
558 		goto out;
559 	ip->i_res->rs_qa_qd_num++;
560 	qd++;
561 
562 	if (!uid_eq(uid, NO_UID_QUOTA_CHANGE) &&
563 	    !uid_eq(uid, ip->i_inode.i_uid)) {
564 		error = qdsb_get(sdp, make_kqid_uid(uid), qd);
565 		if (error)
566 			goto out;
567 		ip->i_res->rs_qa_qd_num++;
568 		qd++;
569 	}
570 
571 	if (!gid_eq(gid, NO_GID_QUOTA_CHANGE) &&
572 	    !gid_eq(gid, ip->i_inode.i_gid)) {
573 		error = qdsb_get(sdp, make_kqid_gid(gid), qd);
574 		if (error)
575 			goto out;
576 		ip->i_res->rs_qa_qd_num++;
577 		qd++;
578 	}
579 
580 out:
581 	if (error)
582 		gfs2_quota_unhold(ip);
583 	return error;
584 }
585 
586 void gfs2_quota_unhold(struct gfs2_inode *ip)
587 {
588 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
589 	unsigned int x;
590 
591 	if (ip->i_res == NULL)
592 		return;
593 	gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags));
594 
595 	for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
596 		qdsb_put(ip->i_res->rs_qa_qd[x]);
597 		ip->i_res->rs_qa_qd[x] = NULL;
598 	}
599 	ip->i_res->rs_qa_qd_num = 0;
600 }
601 
602 static int sort_qd(const void *a, const void *b)
603 {
604 	const struct gfs2_quota_data *qd_a = *(const struct gfs2_quota_data **)a;
605 	const struct gfs2_quota_data *qd_b = *(const struct gfs2_quota_data **)b;
606 
607 	if (qid_lt(qd_a->qd_id, qd_b->qd_id))
608 		return -1;
609 	if (qid_lt(qd_b->qd_id, qd_a->qd_id))
610 		return 1;
611 	return 0;
612 }
613 
614 static void do_qc(struct gfs2_quota_data *qd, s64 change)
615 {
616 	struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
617 	struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
618 	struct gfs2_quota_change *qc = qd->qd_bh_qc;
619 	s64 x;
620 
621 	mutex_lock(&sdp->sd_quota_mutex);
622 	gfs2_trans_add_meta(ip->i_gl, qd->qd_bh);
623 
624 	if (!test_bit(QDF_CHANGE, &qd->qd_flags)) {
625 		qc->qc_change = 0;
626 		qc->qc_flags = 0;
627 		if (qd->qd_id.type == USRQUOTA)
628 			qc->qc_flags = cpu_to_be32(GFS2_QCF_USER);
629 		qc->qc_id = cpu_to_be32(from_kqid(&init_user_ns, qd->qd_id));
630 	}
631 
632 	x = be64_to_cpu(qc->qc_change) + change;
633 	qc->qc_change = cpu_to_be64(x);
634 
635 	spin_lock(&qd_lock);
636 	qd->qd_change = x;
637 	spin_unlock(&qd_lock);
638 
639 	if (!x) {
640 		gfs2_assert_warn(sdp, test_bit(QDF_CHANGE, &qd->qd_flags));
641 		clear_bit(QDF_CHANGE, &qd->qd_flags);
642 		qc->qc_flags = 0;
643 		qc->qc_id = 0;
644 		slot_put(qd);
645 		qd_put(qd);
646 	} else if (!test_and_set_bit(QDF_CHANGE, &qd->qd_flags)) {
647 		qd_hold(qd);
648 		slot_hold(qd);
649 	}
650 
651 	mutex_unlock(&sdp->sd_quota_mutex);
652 }
653 
654 /**
655  * gfs2_adjust_quota - adjust record of current block usage
656  * @ip: The quota inode
657  * @loc: Offset of the entry in the quota file
658  * @change: The amount of usage change to record
659  * @qd: The quota data
660  * @fdq: The updated limits to record
661  *
662  * This function was mostly borrowed from gfs2_block_truncate_page which was
663  * in turn mostly borrowed from ext3
664  *
665  * Returns: 0 or -ve on error
666  */
667 
668 static int gfs2_adjust_quota(struct gfs2_inode *ip, loff_t loc,
669 			     s64 change, struct gfs2_quota_data *qd,
670 			     struct fs_disk_quota *fdq)
671 {
672 	struct inode *inode = &ip->i_inode;
673 	struct gfs2_sbd *sdp = GFS2_SB(inode);
674 	struct address_space *mapping = inode->i_mapping;
675 	unsigned long index = loc >> PAGE_CACHE_SHIFT;
676 	unsigned offset = loc & (PAGE_CACHE_SIZE - 1);
677 	unsigned blocksize, iblock, pos;
678 	struct buffer_head *bh;
679 	struct page *page;
680 	void *kaddr, *ptr;
681 	struct gfs2_quota q;
682 	int err, nbytes;
683 	u64 size;
684 
685 	if (gfs2_is_stuffed(ip)) {
686 		err = gfs2_unstuff_dinode(ip, NULL);
687 		if (err)
688 			return err;
689 	}
690 
691 	memset(&q, 0, sizeof(struct gfs2_quota));
692 	err = gfs2_internal_read(ip, (char *)&q, &loc, sizeof(q));
693 	if (err < 0)
694 		return err;
695 
696 	err = -EIO;
697 	be64_add_cpu(&q.qu_value, change);
698 	qd->qd_qb.qb_value = q.qu_value;
699 	if (fdq) {
700 		if (fdq->d_fieldmask & FS_DQ_BSOFT) {
701 			q.qu_warn = cpu_to_be64(fdq->d_blk_softlimit >> sdp->sd_fsb2bb_shift);
702 			qd->qd_qb.qb_warn = q.qu_warn;
703 		}
704 		if (fdq->d_fieldmask & FS_DQ_BHARD) {
705 			q.qu_limit = cpu_to_be64(fdq->d_blk_hardlimit >> sdp->sd_fsb2bb_shift);
706 			qd->qd_qb.qb_limit = q.qu_limit;
707 		}
708 		if (fdq->d_fieldmask & FS_DQ_BCOUNT) {
709 			q.qu_value = cpu_to_be64(fdq->d_bcount >> sdp->sd_fsb2bb_shift);
710 			qd->qd_qb.qb_value = q.qu_value;
711 		}
712 	}
713 
714 	/* Write the quota into the quota file on disk */
715 	ptr = &q;
716 	nbytes = sizeof(struct gfs2_quota);
717 get_a_page:
718 	page = find_or_create_page(mapping, index, GFP_NOFS);
719 	if (!page)
720 		return -ENOMEM;
721 
722 	blocksize = inode->i_sb->s_blocksize;
723 	iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
724 
725 	if (!page_has_buffers(page))
726 		create_empty_buffers(page, blocksize, 0);
727 
728 	bh = page_buffers(page);
729 	pos = blocksize;
730 	while (offset >= pos) {
731 		bh = bh->b_this_page;
732 		iblock++;
733 		pos += blocksize;
734 	}
735 
736 	if (!buffer_mapped(bh)) {
737 		gfs2_block_map(inode, iblock, bh, 1);
738 		if (!buffer_mapped(bh))
739 			goto unlock_out;
740 		/* If it's a newly allocated disk block for quota, zero it */
741 		if (buffer_new(bh))
742 			zero_user(page, pos - blocksize, bh->b_size);
743 	}
744 
745 	if (PageUptodate(page))
746 		set_buffer_uptodate(bh);
747 
748 	if (!buffer_uptodate(bh)) {
749 		ll_rw_block(READ | REQ_META, 1, &bh);
750 		wait_on_buffer(bh);
751 		if (!buffer_uptodate(bh))
752 			goto unlock_out;
753 	}
754 
755 	gfs2_trans_add_data(ip->i_gl, bh);
756 
757 	kaddr = kmap_atomic(page);
758 	if (offset + sizeof(struct gfs2_quota) > PAGE_CACHE_SIZE)
759 		nbytes = PAGE_CACHE_SIZE - offset;
760 	memcpy(kaddr + offset, ptr, nbytes);
761 	flush_dcache_page(page);
762 	kunmap_atomic(kaddr);
763 	unlock_page(page);
764 	page_cache_release(page);
765 
766 	/* If quota straddles page boundary, we need to update the rest of the
767 	 * quota at the beginning of the next page */
768 	if ((offset + sizeof(struct gfs2_quota)) > PAGE_CACHE_SIZE) {
769 		ptr = ptr + nbytes;
770 		nbytes = sizeof(struct gfs2_quota) - nbytes;
771 		offset = 0;
772 		index++;
773 		goto get_a_page;
774 	}
775 
776 	size = loc + sizeof(struct gfs2_quota);
777 	if (size > inode->i_size)
778 		i_size_write(inode, size);
779 	inode->i_mtime = inode->i_atime = CURRENT_TIME;
780 	mark_inode_dirty(inode);
781 	return 0;
782 
783 unlock_out:
784 	unlock_page(page);
785 	page_cache_release(page);
786 	return err;
787 }
788 
789 static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
790 {
791 	struct gfs2_sbd *sdp = (*qda)->qd_gl->gl_sbd;
792 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
793 	struct gfs2_alloc_parms ap = { .aflags = 0, };
794 	unsigned int data_blocks, ind_blocks;
795 	struct gfs2_holder *ghs, i_gh;
796 	unsigned int qx, x;
797 	struct gfs2_quota_data *qd;
798 	unsigned reserved;
799 	loff_t offset;
800 	unsigned int nalloc = 0, blocks;
801 	int error;
802 
803 	error = gfs2_rs_alloc(ip);
804 	if (error)
805 		return error;
806 
807 	gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
808 			      &data_blocks, &ind_blocks);
809 
810 	ghs = kcalloc(num_qd, sizeof(struct gfs2_holder), GFP_NOFS);
811 	if (!ghs)
812 		return -ENOMEM;
813 
814 	sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL);
815 	mutex_lock(&ip->i_inode.i_mutex);
816 	for (qx = 0; qx < num_qd; qx++) {
817 		error = gfs2_glock_nq_init(qda[qx]->qd_gl, LM_ST_EXCLUSIVE,
818 					   GL_NOCACHE, &ghs[qx]);
819 		if (error)
820 			goto out;
821 	}
822 
823 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
824 	if (error)
825 		goto out;
826 
827 	for (x = 0; x < num_qd; x++) {
828 		offset = qd2offset(qda[x]);
829 		if (gfs2_write_alloc_required(ip, offset,
830 					      sizeof(struct gfs2_quota)))
831 			nalloc++;
832 	}
833 
834 	/*
835 	 * 1 blk for unstuffing inode if stuffed. We add this extra
836 	 * block to the reservation unconditionally. If the inode
837 	 * doesn't need unstuffing, the block will be released to the
838 	 * rgrp since it won't be allocated during the transaction
839 	 */
840 	/* +3 in the end for unstuffing block, inode size update block
841 	 * and another block in case quota straddles page boundary and
842 	 * two blocks need to be updated instead of 1 */
843 	blocks = num_qd * data_blocks + RES_DINODE + num_qd + 3;
844 
845 	reserved = 1 + (nalloc * (data_blocks + ind_blocks));
846 	ap.target = reserved;
847 	error = gfs2_inplace_reserve(ip, &ap);
848 	if (error)
849 		goto out_alloc;
850 
851 	if (nalloc)
852 		blocks += gfs2_rg_blocks(ip, reserved) + nalloc * ind_blocks + RES_STATFS;
853 
854 	error = gfs2_trans_begin(sdp, blocks, 0);
855 	if (error)
856 		goto out_ipres;
857 
858 	for (x = 0; x < num_qd; x++) {
859 		qd = qda[x];
860 		offset = qd2offset(qd);
861 		error = gfs2_adjust_quota(ip, offset, qd->qd_change_sync, qd, NULL);
862 		if (error)
863 			goto out_end_trans;
864 
865 		do_qc(qd, -qd->qd_change_sync);
866 		set_bit(QDF_REFRESH, &qd->qd_flags);
867 	}
868 
869 	error = 0;
870 
871 out_end_trans:
872 	gfs2_trans_end(sdp);
873 out_ipres:
874 	gfs2_inplace_release(ip);
875 out_alloc:
876 	gfs2_glock_dq_uninit(&i_gh);
877 out:
878 	while (qx--)
879 		gfs2_glock_dq_uninit(&ghs[qx]);
880 	mutex_unlock(&ip->i_inode.i_mutex);
881 	kfree(ghs);
882 	gfs2_log_flush(ip->i_gl->gl_sbd, ip->i_gl);
883 	return error;
884 }
885 
886 static int update_qd(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd)
887 {
888 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
889 	struct gfs2_quota q;
890 	struct gfs2_quota_lvb *qlvb;
891 	loff_t pos;
892 	int error;
893 
894 	memset(&q, 0, sizeof(struct gfs2_quota));
895 	pos = qd2offset(qd);
896 	error = gfs2_internal_read(ip, (char *)&q, &pos, sizeof(q));
897 	if (error < 0)
898 		return error;
899 
900 	qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
901 	qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC);
902 	qlvb->__pad = 0;
903 	qlvb->qb_limit = q.qu_limit;
904 	qlvb->qb_warn = q.qu_warn;
905 	qlvb->qb_value = q.qu_value;
906 	qd->qd_qb = *qlvb;
907 
908 	return 0;
909 }
910 
911 static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
912 		    struct gfs2_holder *q_gh)
913 {
914 	struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
915 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
916 	struct gfs2_holder i_gh;
917 	int error;
918 
919 restart:
920 	error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
921 	if (error)
922 		return error;
923 
924 	qd->qd_qb = *(struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
925 
926 	if (force_refresh || qd->qd_qb.qb_magic != cpu_to_be32(GFS2_MAGIC)) {
927 		gfs2_glock_dq_uninit(q_gh);
928 		error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE,
929 					   GL_NOCACHE, q_gh);
930 		if (error)
931 			return error;
932 
933 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
934 		if (error)
935 			goto fail;
936 
937 		error = update_qd(sdp, qd);
938 		if (error)
939 			goto fail_gunlock;
940 
941 		gfs2_glock_dq_uninit(&i_gh);
942 		gfs2_glock_dq_uninit(q_gh);
943 		force_refresh = 0;
944 		goto restart;
945 	}
946 
947 	return 0;
948 
949 fail_gunlock:
950 	gfs2_glock_dq_uninit(&i_gh);
951 fail:
952 	gfs2_glock_dq_uninit(q_gh);
953 	return error;
954 }
955 
956 int gfs2_quota_lock(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
957 {
958 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
959 	struct gfs2_quota_data *qd;
960 	unsigned int x;
961 	int error = 0;
962 
963 	error = gfs2_quota_hold(ip, uid, gid);
964 	if (error)
965 		return error;
966 
967 	if (capable(CAP_SYS_RESOURCE) ||
968 	    sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
969 		return 0;
970 
971 	sort(ip->i_res->rs_qa_qd, ip->i_res->rs_qa_qd_num,
972 	     sizeof(struct gfs2_quota_data *), sort_qd, NULL);
973 
974 	for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
975 		int force = NO_FORCE;
976 		qd = ip->i_res->rs_qa_qd[x];
977 		if (test_and_clear_bit(QDF_REFRESH, &qd->qd_flags))
978 			force = FORCE;
979 		error = do_glock(qd, force, &ip->i_res->rs_qa_qd_ghs[x]);
980 		if (error)
981 			break;
982 	}
983 
984 	if (!error)
985 		set_bit(GIF_QD_LOCKED, &ip->i_flags);
986 	else {
987 		while (x--)
988 			gfs2_glock_dq_uninit(&ip->i_res->rs_qa_qd_ghs[x]);
989 		gfs2_quota_unhold(ip);
990 	}
991 
992 	return error;
993 }
994 
995 static int need_sync(struct gfs2_quota_data *qd)
996 {
997 	struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
998 	struct gfs2_tune *gt = &sdp->sd_tune;
999 	s64 value;
1000 	unsigned int num, den;
1001 	int do_sync = 1;
1002 
1003 	if (!qd->qd_qb.qb_limit)
1004 		return 0;
1005 
1006 	spin_lock(&qd_lock);
1007 	value = qd->qd_change;
1008 	spin_unlock(&qd_lock);
1009 
1010 	spin_lock(&gt->gt_spin);
1011 	num = gt->gt_quota_scale_num;
1012 	den = gt->gt_quota_scale_den;
1013 	spin_unlock(&gt->gt_spin);
1014 
1015 	if (value < 0)
1016 		do_sync = 0;
1017 	else if ((s64)be64_to_cpu(qd->qd_qb.qb_value) >=
1018 		 (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1019 		do_sync = 0;
1020 	else {
1021 		value *= gfs2_jindex_size(sdp) * num;
1022 		value = div_s64(value, den);
1023 		value += (s64)be64_to_cpu(qd->qd_qb.qb_value);
1024 		if (value < (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1025 			do_sync = 0;
1026 	}
1027 
1028 	return do_sync;
1029 }
1030 
1031 void gfs2_quota_unlock(struct gfs2_inode *ip)
1032 {
1033 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1034 	struct gfs2_quota_data *qda[4];
1035 	unsigned int count = 0;
1036 	unsigned int x;
1037 	int found;
1038 
1039 	if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
1040 		goto out;
1041 
1042 	for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1043 		struct gfs2_quota_data *qd;
1044 		int sync;
1045 
1046 		qd = ip->i_res->rs_qa_qd[x];
1047 		sync = need_sync(qd);
1048 
1049 		gfs2_glock_dq_uninit(&ip->i_res->rs_qa_qd_ghs[x]);
1050 		if (!sync)
1051 			continue;
1052 
1053 		spin_lock(&qd_lock);
1054 		found = qd_check_sync(sdp, qd, NULL);
1055 		spin_unlock(&qd_lock);
1056 
1057 		if (!found)
1058 			continue;
1059 
1060 		gfs2_assert_warn(sdp, qd->qd_change_sync);
1061 		if (bh_get(qd)) {
1062 			clear_bit(QDF_LOCKED, &qd->qd_flags);
1063 			slot_put(qd);
1064 			qd_put(qd);
1065 			continue;
1066 		}
1067 
1068 		qda[count++] = qd;
1069 	}
1070 
1071 	if (count) {
1072 		do_sync(count, qda);
1073 		for (x = 0; x < count; x++)
1074 			qd_unlock(qda[x]);
1075 	}
1076 
1077 out:
1078 	gfs2_quota_unhold(ip);
1079 }
1080 
1081 #define MAX_LINE 256
1082 
1083 static int print_message(struct gfs2_quota_data *qd, char *type)
1084 {
1085 	struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
1086 
1087 	fs_info(sdp, "quota %s for %s %u\n",
1088 		type,
1089 		(qd->qd_id.type == USRQUOTA) ? "user" : "group",
1090 		from_kqid(&init_user_ns, qd->qd_id));
1091 
1092 	return 0;
1093 }
1094 
1095 int gfs2_quota_check(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
1096 {
1097 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1098 	struct gfs2_quota_data *qd;
1099 	s64 value;
1100 	unsigned int x;
1101 	int error = 0;
1102 
1103 	if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
1104 		return 0;
1105 
1106         if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
1107                 return 0;
1108 
1109 	for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1110 		qd = ip->i_res->rs_qa_qd[x];
1111 
1112 		if (!(qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1113 		      qid_eq(qd->qd_id, make_kqid_gid(gid))))
1114 			continue;
1115 
1116 		value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
1117 		spin_lock(&qd_lock);
1118 		value += qd->qd_change;
1119 		spin_unlock(&qd_lock);
1120 
1121 		if (be64_to_cpu(qd->qd_qb.qb_limit) && (s64)be64_to_cpu(qd->qd_qb.qb_limit) < value) {
1122 			print_message(qd, "exceeded");
1123 			quota_send_warning(qd->qd_id,
1124 					   sdp->sd_vfs->s_dev, QUOTA_NL_BHARDWARN);
1125 
1126 			error = -EDQUOT;
1127 			break;
1128 		} else if (be64_to_cpu(qd->qd_qb.qb_warn) &&
1129 			   (s64)be64_to_cpu(qd->qd_qb.qb_warn) < value &&
1130 			   time_after_eq(jiffies, qd->qd_last_warn +
1131 					 gfs2_tune_get(sdp,
1132 						gt_quota_warn_period) * HZ)) {
1133 			quota_send_warning(qd->qd_id,
1134 					   sdp->sd_vfs->s_dev, QUOTA_NL_BSOFTWARN);
1135 			error = print_message(qd, "warning");
1136 			qd->qd_last_warn = jiffies;
1137 		}
1138 	}
1139 
1140 	return error;
1141 }
1142 
1143 void gfs2_quota_change(struct gfs2_inode *ip, s64 change,
1144 		       kuid_t uid, kgid_t gid)
1145 {
1146 	struct gfs2_quota_data *qd;
1147 	unsigned int x;
1148 
1149 	if (gfs2_assert_warn(GFS2_SB(&ip->i_inode), change))
1150 		return;
1151 	if (ip->i_diskflags & GFS2_DIF_SYSTEM)
1152 		return;
1153 
1154 	for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1155 		qd = ip->i_res->rs_qa_qd[x];
1156 
1157 		if (qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1158 		    qid_eq(qd->qd_id, make_kqid_gid(gid))) {
1159 			do_qc(qd, change);
1160 		}
1161 	}
1162 }
1163 
1164 int gfs2_quota_sync(struct super_block *sb, int type)
1165 {
1166 	struct gfs2_sbd *sdp = sb->s_fs_info;
1167 	struct gfs2_quota_data **qda;
1168 	unsigned int max_qd = PAGE_SIZE/sizeof(struct gfs2_holder);
1169 	unsigned int num_qd;
1170 	unsigned int x;
1171 	int error = 0;
1172 
1173 	qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
1174 	if (!qda)
1175 		return -ENOMEM;
1176 
1177 	mutex_lock(&sdp->sd_quota_sync_mutex);
1178 	sdp->sd_quota_sync_gen++;
1179 
1180 	do {
1181 		num_qd = 0;
1182 
1183 		for (;;) {
1184 			error = qd_fish(sdp, qda + num_qd);
1185 			if (error || !qda[num_qd])
1186 				break;
1187 			if (++num_qd == max_qd)
1188 				break;
1189 		}
1190 
1191 		if (num_qd) {
1192 			if (!error)
1193 				error = do_sync(num_qd, qda);
1194 			if (!error)
1195 				for (x = 0; x < num_qd; x++)
1196 					qda[x]->qd_sync_gen =
1197 						sdp->sd_quota_sync_gen;
1198 
1199 			for (x = 0; x < num_qd; x++)
1200 				qd_unlock(qda[x]);
1201 		}
1202 	} while (!error && num_qd == max_qd);
1203 
1204 	mutex_unlock(&sdp->sd_quota_sync_mutex);
1205 	kfree(qda);
1206 
1207 	return error;
1208 }
1209 
1210 int gfs2_quota_refresh(struct gfs2_sbd *sdp, struct kqid qid)
1211 {
1212 	struct gfs2_quota_data *qd;
1213 	struct gfs2_holder q_gh;
1214 	int error;
1215 
1216 	error = qd_get(sdp, qid, &qd);
1217 	if (error)
1218 		return error;
1219 
1220 	error = do_glock(qd, FORCE, &q_gh);
1221 	if (!error)
1222 		gfs2_glock_dq_uninit(&q_gh);
1223 
1224 	qd_put(qd);
1225 	return error;
1226 }
1227 
1228 int gfs2_quota_init(struct gfs2_sbd *sdp)
1229 {
1230 	struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
1231 	u64 size = i_size_read(sdp->sd_qc_inode);
1232 	unsigned int blocks = size >> sdp->sd_sb.sb_bsize_shift;
1233 	unsigned int x, slot = 0;
1234 	unsigned int found = 0;
1235 	unsigned int hash;
1236 	unsigned int bm_size;
1237 	u64 dblock;
1238 	u32 extlen = 0;
1239 	int error;
1240 
1241 	if (gfs2_check_internal_file_size(sdp->sd_qc_inode, 1, 64 << 20))
1242 		return -EIO;
1243 
1244 	sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
1245 	bm_size = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * sizeof(unsigned long));
1246 	bm_size *= sizeof(unsigned long);
1247 	error = -ENOMEM;
1248 	sdp->sd_quota_bitmap = kzalloc(bm_size, GFP_NOFS | __GFP_NOWARN);
1249 	if (sdp->sd_quota_bitmap == NULL)
1250 		sdp->sd_quota_bitmap = __vmalloc(bm_size, GFP_NOFS |
1251 						 __GFP_ZERO, PAGE_KERNEL);
1252 	if (!sdp->sd_quota_bitmap)
1253 		return error;
1254 
1255 	for (x = 0; x < blocks; x++) {
1256 		struct buffer_head *bh;
1257 		const struct gfs2_quota_change *qc;
1258 		unsigned int y;
1259 
1260 		if (!extlen) {
1261 			int new = 0;
1262 			error = gfs2_extent_map(&ip->i_inode, x, &new, &dblock, &extlen);
1263 			if (error)
1264 				goto fail;
1265 		}
1266 		error = -EIO;
1267 		bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
1268 		if (!bh)
1269 			goto fail;
1270 		if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
1271 			brelse(bh);
1272 			goto fail;
1273 		}
1274 
1275 		qc = (const struct gfs2_quota_change *)(bh->b_data + sizeof(struct gfs2_meta_header));
1276 		for (y = 0; y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1277 		     y++, slot++) {
1278 			struct gfs2_quota_data *qd;
1279 			s64 qc_change = be64_to_cpu(qc->qc_change);
1280 			u32 qc_flags = be32_to_cpu(qc->qc_flags);
1281 			enum quota_type qtype = (qc_flags & GFS2_QCF_USER) ?
1282 						USRQUOTA : GRPQUOTA;
1283 			struct kqid qc_id = make_kqid(&init_user_ns, qtype,
1284 						      be32_to_cpu(qc->qc_id));
1285 			qc++;
1286 			if (!qc_change)
1287 				continue;
1288 
1289 			hash = gfs2_qd_hash(sdp, qc_id);
1290 			qd = qd_alloc(hash, sdp, qc_id);
1291 			if (qd == NULL) {
1292 				brelse(bh);
1293 				goto fail;
1294 			}
1295 
1296 			set_bit(QDF_CHANGE, &qd->qd_flags);
1297 			qd->qd_change = qc_change;
1298 			qd->qd_slot = slot;
1299 			qd->qd_slot_count = 1;
1300 
1301 			spin_lock(&qd_lock);
1302 			BUG_ON(test_and_set_bit(slot, sdp->sd_quota_bitmap));
1303 			list_add(&qd->qd_list, &sdp->sd_quota_list);
1304 			atomic_inc(&sdp->sd_quota_count);
1305 			spin_unlock(&qd_lock);
1306 
1307 			spin_lock_bucket(hash);
1308 			hlist_bl_add_head_rcu(&qd->qd_hlist, &qd_hash_table[hash]);
1309 			spin_unlock_bucket(hash);
1310 
1311 			found++;
1312 		}
1313 
1314 		brelse(bh);
1315 		dblock++;
1316 		extlen--;
1317 	}
1318 
1319 	if (found)
1320 		fs_info(sdp, "found %u quota changes\n", found);
1321 
1322 	return 0;
1323 
1324 fail:
1325 	gfs2_quota_cleanup(sdp);
1326 	return error;
1327 }
1328 
1329 void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1330 {
1331 	struct list_head *head = &sdp->sd_quota_list;
1332 	struct gfs2_quota_data *qd;
1333 
1334 	spin_lock(&qd_lock);
1335 	while (!list_empty(head)) {
1336 		qd = list_entry(head->prev, struct gfs2_quota_data, qd_list);
1337 
1338 		list_del(&qd->qd_list);
1339 
1340 		/* Also remove if this qd exists in the reclaim list */
1341 		list_lru_del(&gfs2_qd_lru, &qd->qd_lru);
1342 		atomic_dec(&sdp->sd_quota_count);
1343 		spin_unlock(&qd_lock);
1344 
1345 		spin_lock_bucket(qd->qd_hash);
1346 		hlist_bl_del_rcu(&qd->qd_hlist);
1347 		spin_unlock_bucket(qd->qd_hash);
1348 
1349 		gfs2_assert_warn(sdp, !qd->qd_change);
1350 		gfs2_assert_warn(sdp, !qd->qd_slot_count);
1351 		gfs2_assert_warn(sdp, !qd->qd_bh_count);
1352 
1353 		gfs2_glock_put(qd->qd_gl);
1354 		call_rcu(&qd->qd_rcu, gfs2_qd_dealloc);
1355 
1356 		spin_lock(&qd_lock);
1357 	}
1358 	spin_unlock(&qd_lock);
1359 
1360 	gfs2_assert_warn(sdp, !atomic_read(&sdp->sd_quota_count));
1361 
1362 	if (sdp->sd_quota_bitmap) {
1363 		if (is_vmalloc_addr(sdp->sd_quota_bitmap))
1364 			vfree(sdp->sd_quota_bitmap);
1365 		else
1366 			kfree(sdp->sd_quota_bitmap);
1367 		sdp->sd_quota_bitmap = NULL;
1368 	}
1369 }
1370 
1371 static void quotad_error(struct gfs2_sbd *sdp, const char *msg, int error)
1372 {
1373 	if (error == 0 || error == -EROFS)
1374 		return;
1375 	if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
1376 		fs_err(sdp, "gfs2_quotad: %s error %d\n", msg, error);
1377 }
1378 
1379 static void quotad_check_timeo(struct gfs2_sbd *sdp, const char *msg,
1380 			       int (*fxn)(struct super_block *sb, int type),
1381 			       unsigned long t, unsigned long *timeo,
1382 			       unsigned int *new_timeo)
1383 {
1384 	if (t >= *timeo) {
1385 		int error = fxn(sdp->sd_vfs, 0);
1386 		quotad_error(sdp, msg, error);
1387 		*timeo = gfs2_tune_get_i(&sdp->sd_tune, new_timeo) * HZ;
1388 	} else {
1389 		*timeo -= t;
1390 	}
1391 }
1392 
1393 static void quotad_check_trunc_list(struct gfs2_sbd *sdp)
1394 {
1395 	struct gfs2_inode *ip;
1396 
1397 	while(1) {
1398 		ip = NULL;
1399 		spin_lock(&sdp->sd_trunc_lock);
1400 		if (!list_empty(&sdp->sd_trunc_list)) {
1401 			ip = list_entry(sdp->sd_trunc_list.next,
1402 					struct gfs2_inode, i_trunc_list);
1403 			list_del_init(&ip->i_trunc_list);
1404 		}
1405 		spin_unlock(&sdp->sd_trunc_lock);
1406 		if (ip == NULL)
1407 			return;
1408 		gfs2_glock_finish_truncate(ip);
1409 	}
1410 }
1411 
1412 void gfs2_wake_up_statfs(struct gfs2_sbd *sdp) {
1413 	if (!sdp->sd_statfs_force_sync) {
1414 		sdp->sd_statfs_force_sync = 1;
1415 		wake_up(&sdp->sd_quota_wait);
1416 	}
1417 }
1418 
1419 
1420 /**
1421  * gfs2_quotad - Write cached quota changes into the quota file
1422  * @sdp: Pointer to GFS2 superblock
1423  *
1424  */
1425 
1426 int gfs2_quotad(void *data)
1427 {
1428 	struct gfs2_sbd *sdp = data;
1429 	struct gfs2_tune *tune = &sdp->sd_tune;
1430 	unsigned long statfs_timeo = 0;
1431 	unsigned long quotad_timeo = 0;
1432 	unsigned long t = 0;
1433 	DEFINE_WAIT(wait);
1434 	int empty;
1435 
1436 	while (!kthread_should_stop()) {
1437 
1438 		/* Update the master statfs file */
1439 		if (sdp->sd_statfs_force_sync) {
1440 			int error = gfs2_statfs_sync(sdp->sd_vfs, 0);
1441 			quotad_error(sdp, "statfs", error);
1442 			statfs_timeo = gfs2_tune_get(sdp, gt_statfs_quantum) * HZ;
1443 		}
1444 		else
1445 			quotad_check_timeo(sdp, "statfs", gfs2_statfs_sync, t,
1446 				   	   &statfs_timeo,
1447 					   &tune->gt_statfs_quantum);
1448 
1449 		/* Update quota file */
1450 		quotad_check_timeo(sdp, "sync", gfs2_quota_sync, t,
1451 				   &quotad_timeo, &tune->gt_quota_quantum);
1452 
1453 		/* Check for & recover partially truncated inodes */
1454 		quotad_check_trunc_list(sdp);
1455 
1456 		try_to_freeze();
1457 
1458 		t = min(quotad_timeo, statfs_timeo);
1459 
1460 		prepare_to_wait(&sdp->sd_quota_wait, &wait, TASK_INTERRUPTIBLE);
1461 		spin_lock(&sdp->sd_trunc_lock);
1462 		empty = list_empty(&sdp->sd_trunc_list);
1463 		spin_unlock(&sdp->sd_trunc_lock);
1464 		if (empty && !sdp->sd_statfs_force_sync)
1465 			t -= schedule_timeout(t);
1466 		else
1467 			t = 0;
1468 		finish_wait(&sdp->sd_quota_wait, &wait);
1469 	}
1470 
1471 	return 0;
1472 }
1473 
1474 static int gfs2_quota_get_xstate(struct super_block *sb,
1475 				 struct fs_quota_stat *fqs)
1476 {
1477 	struct gfs2_sbd *sdp = sb->s_fs_info;
1478 
1479 	memset(fqs, 0, sizeof(struct fs_quota_stat));
1480 	fqs->qs_version = FS_QSTAT_VERSION;
1481 
1482 	switch (sdp->sd_args.ar_quota) {
1483 	case GFS2_QUOTA_ON:
1484 		fqs->qs_flags |= (FS_QUOTA_UDQ_ENFD | FS_QUOTA_GDQ_ENFD);
1485 		/*FALLTHRU*/
1486 	case GFS2_QUOTA_ACCOUNT:
1487 		fqs->qs_flags |= (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT);
1488 		break;
1489 	case GFS2_QUOTA_OFF:
1490 		break;
1491 	}
1492 
1493 	if (sdp->sd_quota_inode) {
1494 		fqs->qs_uquota.qfs_ino = GFS2_I(sdp->sd_quota_inode)->i_no_addr;
1495 		fqs->qs_uquota.qfs_nblks = sdp->sd_quota_inode->i_blocks;
1496 	}
1497 	fqs->qs_uquota.qfs_nextents = 1; /* unsupported */
1498 	fqs->qs_gquota = fqs->qs_uquota; /* its the same inode in both cases */
1499 	fqs->qs_incoredqs = list_lru_count(&gfs2_qd_lru);
1500 	return 0;
1501 }
1502 
1503 static int gfs2_get_dqblk(struct super_block *sb, struct kqid qid,
1504 			  struct fs_disk_quota *fdq)
1505 {
1506 	struct gfs2_sbd *sdp = sb->s_fs_info;
1507 	struct gfs2_quota_lvb *qlvb;
1508 	struct gfs2_quota_data *qd;
1509 	struct gfs2_holder q_gh;
1510 	int error;
1511 
1512 	memset(fdq, 0, sizeof(struct fs_disk_quota));
1513 
1514 	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1515 		return -ESRCH; /* Crazy XFS error code */
1516 
1517 	if ((qid.type != USRQUOTA) &&
1518 	    (qid.type != GRPQUOTA))
1519 		return -EINVAL;
1520 
1521 	error = qd_get(sdp, qid, &qd);
1522 	if (error)
1523 		return error;
1524 	error = do_glock(qd, FORCE, &q_gh);
1525 	if (error)
1526 		goto out;
1527 
1528 	qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1529 	fdq->d_version = FS_DQUOT_VERSION;
1530 	fdq->d_flags = (qid.type == USRQUOTA) ? FS_USER_QUOTA : FS_GROUP_QUOTA;
1531 	fdq->d_id = from_kqid_munged(current_user_ns(), qid);
1532 	fdq->d_blk_hardlimit = be64_to_cpu(qlvb->qb_limit) << sdp->sd_fsb2bb_shift;
1533 	fdq->d_blk_softlimit = be64_to_cpu(qlvb->qb_warn) << sdp->sd_fsb2bb_shift;
1534 	fdq->d_bcount = be64_to_cpu(qlvb->qb_value) << sdp->sd_fsb2bb_shift;
1535 
1536 	gfs2_glock_dq_uninit(&q_gh);
1537 out:
1538 	qd_put(qd);
1539 	return error;
1540 }
1541 
1542 /* GFS2 only supports a subset of the XFS fields */
1543 #define GFS2_FIELDMASK (FS_DQ_BSOFT|FS_DQ_BHARD|FS_DQ_BCOUNT)
1544 
1545 static int gfs2_set_dqblk(struct super_block *sb, struct kqid qid,
1546 			  struct fs_disk_quota *fdq)
1547 {
1548 	struct gfs2_sbd *sdp = sb->s_fs_info;
1549 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1550 	struct gfs2_quota_data *qd;
1551 	struct gfs2_holder q_gh, i_gh;
1552 	unsigned int data_blocks, ind_blocks;
1553 	unsigned int blocks = 0;
1554 	int alloc_required;
1555 	loff_t offset;
1556 	int error;
1557 
1558 	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1559 		return -ESRCH; /* Crazy XFS error code */
1560 
1561 	if ((qid.type != USRQUOTA) &&
1562 	    (qid.type != GRPQUOTA))
1563 		return -EINVAL;
1564 
1565 	if (fdq->d_fieldmask & ~GFS2_FIELDMASK)
1566 		return -EINVAL;
1567 
1568 	error = qd_get(sdp, qid, &qd);
1569 	if (error)
1570 		return error;
1571 
1572 	error = gfs2_rs_alloc(ip);
1573 	if (error)
1574 		goto out_put;
1575 
1576 	mutex_lock(&ip->i_inode.i_mutex);
1577 	error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE, 0, &q_gh);
1578 	if (error)
1579 		goto out_unlockput;
1580 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1581 	if (error)
1582 		goto out_q;
1583 
1584 	/* Check for existing entry, if none then alloc new blocks */
1585 	error = update_qd(sdp, qd);
1586 	if (error)
1587 		goto out_i;
1588 
1589 	/* If nothing has changed, this is a no-op */
1590 	if ((fdq->d_fieldmask & FS_DQ_BSOFT) &&
1591 	    ((fdq->d_blk_softlimit >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_warn)))
1592 		fdq->d_fieldmask ^= FS_DQ_BSOFT;
1593 
1594 	if ((fdq->d_fieldmask & FS_DQ_BHARD) &&
1595 	    ((fdq->d_blk_hardlimit >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_limit)))
1596 		fdq->d_fieldmask ^= FS_DQ_BHARD;
1597 
1598 	if ((fdq->d_fieldmask & FS_DQ_BCOUNT) &&
1599 	    ((fdq->d_bcount >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_value)))
1600 		fdq->d_fieldmask ^= FS_DQ_BCOUNT;
1601 
1602 	if (fdq->d_fieldmask == 0)
1603 		goto out_i;
1604 
1605 	offset = qd2offset(qd);
1606 	alloc_required = gfs2_write_alloc_required(ip, offset, sizeof(struct gfs2_quota));
1607 	if (gfs2_is_stuffed(ip))
1608 		alloc_required = 1;
1609 	if (alloc_required) {
1610 		struct gfs2_alloc_parms ap = { .aflags = 0, };
1611 		gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
1612 				       &data_blocks, &ind_blocks);
1613 		blocks = 1 + data_blocks + ind_blocks;
1614 		ap.target = blocks;
1615 		error = gfs2_inplace_reserve(ip, &ap);
1616 		if (error)
1617 			goto out_i;
1618 		blocks += gfs2_rg_blocks(ip, blocks);
1619 	}
1620 
1621 	/* Some quotas span block boundaries and can update two blocks,
1622 	   adding an extra block to the transaction to handle such quotas */
1623 	error = gfs2_trans_begin(sdp, blocks + RES_DINODE + 2, 0);
1624 	if (error)
1625 		goto out_release;
1626 
1627 	/* Apply changes */
1628 	error = gfs2_adjust_quota(ip, offset, 0, qd, fdq);
1629 
1630 	gfs2_trans_end(sdp);
1631 out_release:
1632 	if (alloc_required)
1633 		gfs2_inplace_release(ip);
1634 out_i:
1635 	gfs2_glock_dq_uninit(&i_gh);
1636 out_q:
1637 	gfs2_glock_dq_uninit(&q_gh);
1638 out_unlockput:
1639 	mutex_unlock(&ip->i_inode.i_mutex);
1640 out_put:
1641 	qd_put(qd);
1642 	return error;
1643 }
1644 
1645 const struct quotactl_ops gfs2_quotactl_ops = {
1646 	.quota_sync     = gfs2_quota_sync,
1647 	.get_xstate     = gfs2_quota_get_xstate,
1648 	.get_dqblk	= gfs2_get_dqblk,
1649 	.set_dqblk	= gfs2_set_dqblk,
1650 };
1651 
1652 void __init gfs2_quota_hash_init(void)
1653 {
1654 	unsigned i;
1655 
1656 	for(i = 0; i < GFS2_QD_HASH_SIZE; i++)
1657 		INIT_HLIST_BL_HEAD(&qd_hash_table[i]);
1658 }
1659