xref: /openbmc/linux/fs/gfs2/quota.c (revision 609e478b)
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 	set_bit(QDF_REFRESH, &qd->qd_flags);
782 	return 0;
783 
784 unlock_out:
785 	unlock_page(page);
786 	page_cache_release(page);
787 	return err;
788 }
789 
790 static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
791 {
792 	struct gfs2_sbd *sdp = (*qda)->qd_gl->gl_sbd;
793 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
794 	struct gfs2_alloc_parms ap = { .aflags = 0, };
795 	unsigned int data_blocks, ind_blocks;
796 	struct gfs2_holder *ghs, i_gh;
797 	unsigned int qx, x;
798 	struct gfs2_quota_data *qd;
799 	unsigned reserved;
800 	loff_t offset;
801 	unsigned int nalloc = 0, blocks;
802 	int error;
803 
804 	error = gfs2_rs_alloc(ip);
805 	if (error)
806 		return error;
807 
808 	gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
809 			      &data_blocks, &ind_blocks);
810 
811 	ghs = kcalloc(num_qd, sizeof(struct gfs2_holder), GFP_NOFS);
812 	if (!ghs)
813 		return -ENOMEM;
814 
815 	sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL);
816 	mutex_lock(&ip->i_inode.i_mutex);
817 	for (qx = 0; qx < num_qd; qx++) {
818 		error = gfs2_glock_nq_init(qda[qx]->qd_gl, LM_ST_EXCLUSIVE,
819 					   GL_NOCACHE, &ghs[qx]);
820 		if (error)
821 			goto out;
822 	}
823 
824 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
825 	if (error)
826 		goto out;
827 
828 	for (x = 0; x < num_qd; x++) {
829 		offset = qd2offset(qda[x]);
830 		if (gfs2_write_alloc_required(ip, offset,
831 					      sizeof(struct gfs2_quota)))
832 			nalloc++;
833 	}
834 
835 	/*
836 	 * 1 blk for unstuffing inode if stuffed. We add this extra
837 	 * block to the reservation unconditionally. If the inode
838 	 * doesn't need unstuffing, the block will be released to the
839 	 * rgrp since it won't be allocated during the transaction
840 	 */
841 	/* +3 in the end for unstuffing block, inode size update block
842 	 * and another block in case quota straddles page boundary and
843 	 * two blocks need to be updated instead of 1 */
844 	blocks = num_qd * data_blocks + RES_DINODE + num_qd + 3;
845 
846 	reserved = 1 + (nalloc * (data_blocks + ind_blocks));
847 	ap.target = reserved;
848 	error = gfs2_inplace_reserve(ip, &ap);
849 	if (error)
850 		goto out_alloc;
851 
852 	if (nalloc)
853 		blocks += gfs2_rg_blocks(ip, reserved) + nalloc * ind_blocks + RES_STATFS;
854 
855 	error = gfs2_trans_begin(sdp, blocks, 0);
856 	if (error)
857 		goto out_ipres;
858 
859 	for (x = 0; x < num_qd; x++) {
860 		qd = qda[x];
861 		offset = qd2offset(qd);
862 		error = gfs2_adjust_quota(ip, offset, qd->qd_change_sync, qd, NULL);
863 		if (error)
864 			goto out_end_trans;
865 
866 		do_qc(qd, -qd->qd_change_sync);
867 		set_bit(QDF_REFRESH, &qd->qd_flags);
868 	}
869 
870 	error = 0;
871 
872 out_end_trans:
873 	gfs2_trans_end(sdp);
874 out_ipres:
875 	gfs2_inplace_release(ip);
876 out_alloc:
877 	gfs2_glock_dq_uninit(&i_gh);
878 out:
879 	while (qx--)
880 		gfs2_glock_dq_uninit(&ghs[qx]);
881 	mutex_unlock(&ip->i_inode.i_mutex);
882 	kfree(ghs);
883 	gfs2_log_flush(ip->i_gl->gl_sbd, ip->i_gl, NORMAL_FLUSH);
884 	return error;
885 }
886 
887 static int update_qd(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd)
888 {
889 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
890 	struct gfs2_quota q;
891 	struct gfs2_quota_lvb *qlvb;
892 	loff_t pos;
893 	int error;
894 
895 	memset(&q, 0, sizeof(struct gfs2_quota));
896 	pos = qd2offset(qd);
897 	error = gfs2_internal_read(ip, (char *)&q, &pos, sizeof(q));
898 	if (error < 0)
899 		return error;
900 
901 	qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
902 	qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC);
903 	qlvb->__pad = 0;
904 	qlvb->qb_limit = q.qu_limit;
905 	qlvb->qb_warn = q.qu_warn;
906 	qlvb->qb_value = q.qu_value;
907 	qd->qd_qb = *qlvb;
908 
909 	return 0;
910 }
911 
912 static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
913 		    struct gfs2_holder *q_gh)
914 {
915 	struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
916 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
917 	struct gfs2_holder i_gh;
918 	int error;
919 
920 restart:
921 	error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
922 	if (error)
923 		return error;
924 
925 	qd->qd_qb = *(struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
926 
927 	if (force_refresh || qd->qd_qb.qb_magic != cpu_to_be32(GFS2_MAGIC)) {
928 		gfs2_glock_dq_uninit(q_gh);
929 		error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE,
930 					   GL_NOCACHE, q_gh);
931 		if (error)
932 			return error;
933 
934 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
935 		if (error)
936 			goto fail;
937 
938 		error = update_qd(sdp, qd);
939 		if (error)
940 			goto fail_gunlock;
941 
942 		gfs2_glock_dq_uninit(&i_gh);
943 		gfs2_glock_dq_uninit(q_gh);
944 		force_refresh = 0;
945 		goto restart;
946 	}
947 
948 	return 0;
949 
950 fail_gunlock:
951 	gfs2_glock_dq_uninit(&i_gh);
952 fail:
953 	gfs2_glock_dq_uninit(q_gh);
954 	return error;
955 }
956 
957 int gfs2_quota_lock(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
958 {
959 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
960 	struct gfs2_quota_data *qd;
961 	unsigned int x;
962 	int error = 0;
963 
964 	error = gfs2_quota_hold(ip, uid, gid);
965 	if (error)
966 		return error;
967 
968 	if (capable(CAP_SYS_RESOURCE) ||
969 	    sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
970 		return 0;
971 
972 	sort(ip->i_res->rs_qa_qd, ip->i_res->rs_qa_qd_num,
973 	     sizeof(struct gfs2_quota_data *), sort_qd, NULL);
974 
975 	for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
976 		int force = NO_FORCE;
977 		qd = ip->i_res->rs_qa_qd[x];
978 		if (test_and_clear_bit(QDF_REFRESH, &qd->qd_flags))
979 			force = FORCE;
980 		error = do_glock(qd, force, &ip->i_res->rs_qa_qd_ghs[x]);
981 		if (error)
982 			break;
983 	}
984 
985 	if (!error)
986 		set_bit(GIF_QD_LOCKED, &ip->i_flags);
987 	else {
988 		while (x--)
989 			gfs2_glock_dq_uninit(&ip->i_res->rs_qa_qd_ghs[x]);
990 		gfs2_quota_unhold(ip);
991 	}
992 
993 	return error;
994 }
995 
996 static int need_sync(struct gfs2_quota_data *qd)
997 {
998 	struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
999 	struct gfs2_tune *gt = &sdp->sd_tune;
1000 	s64 value;
1001 	unsigned int num, den;
1002 	int do_sync = 1;
1003 
1004 	if (!qd->qd_qb.qb_limit)
1005 		return 0;
1006 
1007 	spin_lock(&qd_lock);
1008 	value = qd->qd_change;
1009 	spin_unlock(&qd_lock);
1010 
1011 	spin_lock(&gt->gt_spin);
1012 	num = gt->gt_quota_scale_num;
1013 	den = gt->gt_quota_scale_den;
1014 	spin_unlock(&gt->gt_spin);
1015 
1016 	if (value < 0)
1017 		do_sync = 0;
1018 	else if ((s64)be64_to_cpu(qd->qd_qb.qb_value) >=
1019 		 (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1020 		do_sync = 0;
1021 	else {
1022 		value *= gfs2_jindex_size(sdp) * num;
1023 		value = div_s64(value, den);
1024 		value += (s64)be64_to_cpu(qd->qd_qb.qb_value);
1025 		if (value < (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1026 			do_sync = 0;
1027 	}
1028 
1029 	return do_sync;
1030 }
1031 
1032 void gfs2_quota_unlock(struct gfs2_inode *ip)
1033 {
1034 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1035 	struct gfs2_quota_data *qda[4];
1036 	unsigned int count = 0;
1037 	unsigned int x;
1038 	int found;
1039 
1040 	if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
1041 		goto out;
1042 
1043 	for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1044 		struct gfs2_quota_data *qd;
1045 		int sync;
1046 
1047 		qd = ip->i_res->rs_qa_qd[x];
1048 		sync = need_sync(qd);
1049 
1050 		gfs2_glock_dq_uninit(&ip->i_res->rs_qa_qd_ghs[x]);
1051 		if (!sync)
1052 			continue;
1053 
1054 		spin_lock(&qd_lock);
1055 		found = qd_check_sync(sdp, qd, NULL);
1056 		spin_unlock(&qd_lock);
1057 
1058 		if (!found)
1059 			continue;
1060 
1061 		gfs2_assert_warn(sdp, qd->qd_change_sync);
1062 		if (bh_get(qd)) {
1063 			clear_bit(QDF_LOCKED, &qd->qd_flags);
1064 			slot_put(qd);
1065 			qd_put(qd);
1066 			continue;
1067 		}
1068 
1069 		qda[count++] = qd;
1070 	}
1071 
1072 	if (count) {
1073 		do_sync(count, qda);
1074 		for (x = 0; x < count; x++)
1075 			qd_unlock(qda[x]);
1076 	}
1077 
1078 out:
1079 	gfs2_quota_unhold(ip);
1080 }
1081 
1082 #define MAX_LINE 256
1083 
1084 static int print_message(struct gfs2_quota_data *qd, char *type)
1085 {
1086 	struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
1087 
1088 	fs_info(sdp, "quota %s for %s %u\n",
1089 		type,
1090 		(qd->qd_id.type == USRQUOTA) ? "user" : "group",
1091 		from_kqid(&init_user_ns, qd->qd_id));
1092 
1093 	return 0;
1094 }
1095 
1096 int gfs2_quota_check(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
1097 {
1098 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1099 	struct gfs2_quota_data *qd;
1100 	s64 value;
1101 	unsigned int x;
1102 	int error = 0;
1103 
1104 	if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
1105 		return 0;
1106 
1107         if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
1108                 return 0;
1109 
1110 	for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1111 		qd = ip->i_res->rs_qa_qd[x];
1112 
1113 		if (!(qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1114 		      qid_eq(qd->qd_id, make_kqid_gid(gid))))
1115 			continue;
1116 
1117 		value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
1118 		spin_lock(&qd_lock);
1119 		value += qd->qd_change;
1120 		spin_unlock(&qd_lock);
1121 
1122 		if (be64_to_cpu(qd->qd_qb.qb_limit) && (s64)be64_to_cpu(qd->qd_qb.qb_limit) < value) {
1123 			print_message(qd, "exceeded");
1124 			quota_send_warning(qd->qd_id,
1125 					   sdp->sd_vfs->s_dev, QUOTA_NL_BHARDWARN);
1126 
1127 			error = -EDQUOT;
1128 			break;
1129 		} else if (be64_to_cpu(qd->qd_qb.qb_warn) &&
1130 			   (s64)be64_to_cpu(qd->qd_qb.qb_warn) < value &&
1131 			   time_after_eq(jiffies, qd->qd_last_warn +
1132 					 gfs2_tune_get(sdp,
1133 						gt_quota_warn_period) * HZ)) {
1134 			quota_send_warning(qd->qd_id,
1135 					   sdp->sd_vfs->s_dev, QUOTA_NL_BSOFTWARN);
1136 			error = print_message(qd, "warning");
1137 			qd->qd_last_warn = jiffies;
1138 		}
1139 	}
1140 
1141 	return error;
1142 }
1143 
1144 void gfs2_quota_change(struct gfs2_inode *ip, s64 change,
1145 		       kuid_t uid, kgid_t gid)
1146 {
1147 	struct gfs2_quota_data *qd;
1148 	unsigned int x;
1149 
1150 	if (gfs2_assert_warn(GFS2_SB(&ip->i_inode), change))
1151 		return;
1152 	if (ip->i_diskflags & GFS2_DIF_SYSTEM)
1153 		return;
1154 
1155 	for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1156 		qd = ip->i_res->rs_qa_qd[x];
1157 
1158 		if (qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1159 		    qid_eq(qd->qd_id, make_kqid_gid(gid))) {
1160 			do_qc(qd, change);
1161 		}
1162 	}
1163 }
1164 
1165 int gfs2_quota_sync(struct super_block *sb, int type)
1166 {
1167 	struct gfs2_sbd *sdp = sb->s_fs_info;
1168 	struct gfs2_quota_data **qda;
1169 	unsigned int max_qd = PAGE_SIZE/sizeof(struct gfs2_holder);
1170 	unsigned int num_qd;
1171 	unsigned int x;
1172 	int error = 0;
1173 
1174 	qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
1175 	if (!qda)
1176 		return -ENOMEM;
1177 
1178 	mutex_lock(&sdp->sd_quota_sync_mutex);
1179 	sdp->sd_quota_sync_gen++;
1180 
1181 	do {
1182 		num_qd = 0;
1183 
1184 		for (;;) {
1185 			error = qd_fish(sdp, qda + num_qd);
1186 			if (error || !qda[num_qd])
1187 				break;
1188 			if (++num_qd == max_qd)
1189 				break;
1190 		}
1191 
1192 		if (num_qd) {
1193 			if (!error)
1194 				error = do_sync(num_qd, qda);
1195 			if (!error)
1196 				for (x = 0; x < num_qd; x++)
1197 					qda[x]->qd_sync_gen =
1198 						sdp->sd_quota_sync_gen;
1199 
1200 			for (x = 0; x < num_qd; x++)
1201 				qd_unlock(qda[x]);
1202 		}
1203 	} while (!error && num_qd == max_qd);
1204 
1205 	mutex_unlock(&sdp->sd_quota_sync_mutex);
1206 	kfree(qda);
1207 
1208 	return error;
1209 }
1210 
1211 int gfs2_quota_refresh(struct gfs2_sbd *sdp, struct kqid qid)
1212 {
1213 	struct gfs2_quota_data *qd;
1214 	struct gfs2_holder q_gh;
1215 	int error;
1216 
1217 	error = qd_get(sdp, qid, &qd);
1218 	if (error)
1219 		return error;
1220 
1221 	error = do_glock(qd, FORCE, &q_gh);
1222 	if (!error)
1223 		gfs2_glock_dq_uninit(&q_gh);
1224 
1225 	qd_put(qd);
1226 	return error;
1227 }
1228 
1229 int gfs2_quota_init(struct gfs2_sbd *sdp)
1230 {
1231 	struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
1232 	u64 size = i_size_read(sdp->sd_qc_inode);
1233 	unsigned int blocks = size >> sdp->sd_sb.sb_bsize_shift;
1234 	unsigned int x, slot = 0;
1235 	unsigned int found = 0;
1236 	unsigned int hash;
1237 	unsigned int bm_size;
1238 	u64 dblock;
1239 	u32 extlen = 0;
1240 	int error;
1241 
1242 	if (gfs2_check_internal_file_size(sdp->sd_qc_inode, 1, 64 << 20))
1243 		return -EIO;
1244 
1245 	sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
1246 	bm_size = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * sizeof(unsigned long));
1247 	bm_size *= sizeof(unsigned long);
1248 	error = -ENOMEM;
1249 	sdp->sd_quota_bitmap = kzalloc(bm_size, GFP_NOFS | __GFP_NOWARN);
1250 	if (sdp->sd_quota_bitmap == NULL)
1251 		sdp->sd_quota_bitmap = __vmalloc(bm_size, GFP_NOFS |
1252 						 __GFP_ZERO, PAGE_KERNEL);
1253 	if (!sdp->sd_quota_bitmap)
1254 		return error;
1255 
1256 	for (x = 0; x < blocks; x++) {
1257 		struct buffer_head *bh;
1258 		const struct gfs2_quota_change *qc;
1259 		unsigned int y;
1260 
1261 		if (!extlen) {
1262 			int new = 0;
1263 			error = gfs2_extent_map(&ip->i_inode, x, &new, &dblock, &extlen);
1264 			if (error)
1265 				goto fail;
1266 		}
1267 		error = -EIO;
1268 		bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
1269 		if (!bh)
1270 			goto fail;
1271 		if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
1272 			brelse(bh);
1273 			goto fail;
1274 		}
1275 
1276 		qc = (const struct gfs2_quota_change *)(bh->b_data + sizeof(struct gfs2_meta_header));
1277 		for (y = 0; y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1278 		     y++, slot++) {
1279 			struct gfs2_quota_data *qd;
1280 			s64 qc_change = be64_to_cpu(qc->qc_change);
1281 			u32 qc_flags = be32_to_cpu(qc->qc_flags);
1282 			enum quota_type qtype = (qc_flags & GFS2_QCF_USER) ?
1283 						USRQUOTA : GRPQUOTA;
1284 			struct kqid qc_id = make_kqid(&init_user_ns, qtype,
1285 						      be32_to_cpu(qc->qc_id));
1286 			qc++;
1287 			if (!qc_change)
1288 				continue;
1289 
1290 			hash = gfs2_qd_hash(sdp, qc_id);
1291 			qd = qd_alloc(hash, sdp, qc_id);
1292 			if (qd == NULL) {
1293 				brelse(bh);
1294 				goto fail;
1295 			}
1296 
1297 			set_bit(QDF_CHANGE, &qd->qd_flags);
1298 			qd->qd_change = qc_change;
1299 			qd->qd_slot = slot;
1300 			qd->qd_slot_count = 1;
1301 
1302 			spin_lock(&qd_lock);
1303 			BUG_ON(test_and_set_bit(slot, sdp->sd_quota_bitmap));
1304 			list_add(&qd->qd_list, &sdp->sd_quota_list);
1305 			atomic_inc(&sdp->sd_quota_count);
1306 			spin_unlock(&qd_lock);
1307 
1308 			spin_lock_bucket(hash);
1309 			hlist_bl_add_head_rcu(&qd->qd_hlist, &qd_hash_table[hash]);
1310 			spin_unlock_bucket(hash);
1311 
1312 			found++;
1313 		}
1314 
1315 		brelse(bh);
1316 		dblock++;
1317 		extlen--;
1318 	}
1319 
1320 	if (found)
1321 		fs_info(sdp, "found %u quota changes\n", found);
1322 
1323 	return 0;
1324 
1325 fail:
1326 	gfs2_quota_cleanup(sdp);
1327 	return error;
1328 }
1329 
1330 void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1331 {
1332 	struct list_head *head = &sdp->sd_quota_list;
1333 	struct gfs2_quota_data *qd;
1334 
1335 	spin_lock(&qd_lock);
1336 	while (!list_empty(head)) {
1337 		qd = list_entry(head->prev, struct gfs2_quota_data, qd_list);
1338 
1339 		list_del(&qd->qd_list);
1340 
1341 		/* Also remove if this qd exists in the reclaim list */
1342 		list_lru_del(&gfs2_qd_lru, &qd->qd_lru);
1343 		atomic_dec(&sdp->sd_quota_count);
1344 		spin_unlock(&qd_lock);
1345 
1346 		spin_lock_bucket(qd->qd_hash);
1347 		hlist_bl_del_rcu(&qd->qd_hlist);
1348 		spin_unlock_bucket(qd->qd_hash);
1349 
1350 		gfs2_assert_warn(sdp, !qd->qd_change);
1351 		gfs2_assert_warn(sdp, !qd->qd_slot_count);
1352 		gfs2_assert_warn(sdp, !qd->qd_bh_count);
1353 
1354 		gfs2_glock_put(qd->qd_gl);
1355 		call_rcu(&qd->qd_rcu, gfs2_qd_dealloc);
1356 
1357 		spin_lock(&qd_lock);
1358 	}
1359 	spin_unlock(&qd_lock);
1360 
1361 	gfs2_assert_warn(sdp, !atomic_read(&sdp->sd_quota_count));
1362 
1363 	if (sdp->sd_quota_bitmap) {
1364 		if (is_vmalloc_addr(sdp->sd_quota_bitmap))
1365 			vfree(sdp->sd_quota_bitmap);
1366 		else
1367 			kfree(sdp->sd_quota_bitmap);
1368 		sdp->sd_quota_bitmap = NULL;
1369 	}
1370 }
1371 
1372 static void quotad_error(struct gfs2_sbd *sdp, const char *msg, int error)
1373 {
1374 	if (error == 0 || error == -EROFS)
1375 		return;
1376 	if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
1377 		fs_err(sdp, "gfs2_quotad: %s error %d\n", msg, error);
1378 }
1379 
1380 static void quotad_check_timeo(struct gfs2_sbd *sdp, const char *msg,
1381 			       int (*fxn)(struct super_block *sb, int type),
1382 			       unsigned long t, unsigned long *timeo,
1383 			       unsigned int *new_timeo)
1384 {
1385 	if (t >= *timeo) {
1386 		int error = fxn(sdp->sd_vfs, 0);
1387 		quotad_error(sdp, msg, error);
1388 		*timeo = gfs2_tune_get_i(&sdp->sd_tune, new_timeo) * HZ;
1389 	} else {
1390 		*timeo -= t;
1391 	}
1392 }
1393 
1394 static void quotad_check_trunc_list(struct gfs2_sbd *sdp)
1395 {
1396 	struct gfs2_inode *ip;
1397 
1398 	while(1) {
1399 		ip = NULL;
1400 		spin_lock(&sdp->sd_trunc_lock);
1401 		if (!list_empty(&sdp->sd_trunc_list)) {
1402 			ip = list_entry(sdp->sd_trunc_list.next,
1403 					struct gfs2_inode, i_trunc_list);
1404 			list_del_init(&ip->i_trunc_list);
1405 		}
1406 		spin_unlock(&sdp->sd_trunc_lock);
1407 		if (ip == NULL)
1408 			return;
1409 		gfs2_glock_finish_truncate(ip);
1410 	}
1411 }
1412 
1413 void gfs2_wake_up_statfs(struct gfs2_sbd *sdp) {
1414 	if (!sdp->sd_statfs_force_sync) {
1415 		sdp->sd_statfs_force_sync = 1;
1416 		wake_up(&sdp->sd_quota_wait);
1417 	}
1418 }
1419 
1420 
1421 /**
1422  * gfs2_quotad - Write cached quota changes into the quota file
1423  * @sdp: Pointer to GFS2 superblock
1424  *
1425  */
1426 
1427 int gfs2_quotad(void *data)
1428 {
1429 	struct gfs2_sbd *sdp = data;
1430 	struct gfs2_tune *tune = &sdp->sd_tune;
1431 	unsigned long statfs_timeo = 0;
1432 	unsigned long quotad_timeo = 0;
1433 	unsigned long t = 0;
1434 	DEFINE_WAIT(wait);
1435 	int empty;
1436 
1437 	while (!kthread_should_stop()) {
1438 
1439 		/* Update the master statfs file */
1440 		if (sdp->sd_statfs_force_sync) {
1441 			int error = gfs2_statfs_sync(sdp->sd_vfs, 0);
1442 			quotad_error(sdp, "statfs", error);
1443 			statfs_timeo = gfs2_tune_get(sdp, gt_statfs_quantum) * HZ;
1444 		}
1445 		else
1446 			quotad_check_timeo(sdp, "statfs", gfs2_statfs_sync, t,
1447 				   	   &statfs_timeo,
1448 					   &tune->gt_statfs_quantum);
1449 
1450 		/* Update quota file */
1451 		quotad_check_timeo(sdp, "sync", gfs2_quota_sync, t,
1452 				   &quotad_timeo, &tune->gt_quota_quantum);
1453 
1454 		/* Check for & recover partially truncated inodes */
1455 		quotad_check_trunc_list(sdp);
1456 
1457 		try_to_freeze();
1458 
1459 		t = min(quotad_timeo, statfs_timeo);
1460 
1461 		prepare_to_wait(&sdp->sd_quota_wait, &wait, TASK_INTERRUPTIBLE);
1462 		spin_lock(&sdp->sd_trunc_lock);
1463 		empty = list_empty(&sdp->sd_trunc_list);
1464 		spin_unlock(&sdp->sd_trunc_lock);
1465 		if (empty && !sdp->sd_statfs_force_sync)
1466 			t -= schedule_timeout(t);
1467 		else
1468 			t = 0;
1469 		finish_wait(&sdp->sd_quota_wait, &wait);
1470 	}
1471 
1472 	return 0;
1473 }
1474 
1475 static int gfs2_quota_get_xstate(struct super_block *sb,
1476 				 struct fs_quota_stat *fqs)
1477 {
1478 	struct gfs2_sbd *sdp = sb->s_fs_info;
1479 
1480 	memset(fqs, 0, sizeof(struct fs_quota_stat));
1481 	fqs->qs_version = FS_QSTAT_VERSION;
1482 
1483 	switch (sdp->sd_args.ar_quota) {
1484 	case GFS2_QUOTA_ON:
1485 		fqs->qs_flags |= (FS_QUOTA_UDQ_ENFD | FS_QUOTA_GDQ_ENFD);
1486 		/*FALLTHRU*/
1487 	case GFS2_QUOTA_ACCOUNT:
1488 		fqs->qs_flags |= (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT);
1489 		break;
1490 	case GFS2_QUOTA_OFF:
1491 		break;
1492 	}
1493 
1494 	if (sdp->sd_quota_inode) {
1495 		fqs->qs_uquota.qfs_ino = GFS2_I(sdp->sd_quota_inode)->i_no_addr;
1496 		fqs->qs_uquota.qfs_nblks = sdp->sd_quota_inode->i_blocks;
1497 	}
1498 	fqs->qs_uquota.qfs_nextents = 1; /* unsupported */
1499 	fqs->qs_gquota = fqs->qs_uquota; /* its the same inode in both cases */
1500 	fqs->qs_incoredqs = list_lru_count(&gfs2_qd_lru);
1501 	return 0;
1502 }
1503 
1504 static int gfs2_get_dqblk(struct super_block *sb, struct kqid qid,
1505 			  struct fs_disk_quota *fdq)
1506 {
1507 	struct gfs2_sbd *sdp = sb->s_fs_info;
1508 	struct gfs2_quota_lvb *qlvb;
1509 	struct gfs2_quota_data *qd;
1510 	struct gfs2_holder q_gh;
1511 	int error;
1512 
1513 	memset(fdq, 0, sizeof(struct fs_disk_quota));
1514 
1515 	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1516 		return -ESRCH; /* Crazy XFS error code */
1517 
1518 	if ((qid.type != USRQUOTA) &&
1519 	    (qid.type != GRPQUOTA))
1520 		return -EINVAL;
1521 
1522 	error = qd_get(sdp, qid, &qd);
1523 	if (error)
1524 		return error;
1525 	error = do_glock(qd, FORCE, &q_gh);
1526 	if (error)
1527 		goto out;
1528 
1529 	qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1530 	fdq->d_version = FS_DQUOT_VERSION;
1531 	fdq->d_flags = (qid.type == USRQUOTA) ? FS_USER_QUOTA : FS_GROUP_QUOTA;
1532 	fdq->d_id = from_kqid_munged(current_user_ns(), qid);
1533 	fdq->d_blk_hardlimit = be64_to_cpu(qlvb->qb_limit) << sdp->sd_fsb2bb_shift;
1534 	fdq->d_blk_softlimit = be64_to_cpu(qlvb->qb_warn) << sdp->sd_fsb2bb_shift;
1535 	fdq->d_bcount = be64_to_cpu(qlvb->qb_value) << sdp->sd_fsb2bb_shift;
1536 
1537 	gfs2_glock_dq_uninit(&q_gh);
1538 out:
1539 	qd_put(qd);
1540 	return error;
1541 }
1542 
1543 /* GFS2 only supports a subset of the XFS fields */
1544 #define GFS2_FIELDMASK (FS_DQ_BSOFT|FS_DQ_BHARD|FS_DQ_BCOUNT)
1545 
1546 static int gfs2_set_dqblk(struct super_block *sb, struct kqid qid,
1547 			  struct fs_disk_quota *fdq)
1548 {
1549 	struct gfs2_sbd *sdp = sb->s_fs_info;
1550 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1551 	struct gfs2_quota_data *qd;
1552 	struct gfs2_holder q_gh, i_gh;
1553 	unsigned int data_blocks, ind_blocks;
1554 	unsigned int blocks = 0;
1555 	int alloc_required;
1556 	loff_t offset;
1557 	int error;
1558 
1559 	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1560 		return -ESRCH; /* Crazy XFS error code */
1561 
1562 	if ((qid.type != USRQUOTA) &&
1563 	    (qid.type != GRPQUOTA))
1564 		return -EINVAL;
1565 
1566 	if (fdq->d_fieldmask & ~GFS2_FIELDMASK)
1567 		return -EINVAL;
1568 
1569 	error = qd_get(sdp, qid, &qd);
1570 	if (error)
1571 		return error;
1572 
1573 	error = gfs2_rs_alloc(ip);
1574 	if (error)
1575 		goto out_put;
1576 
1577 	mutex_lock(&ip->i_inode.i_mutex);
1578 	error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE, 0, &q_gh);
1579 	if (error)
1580 		goto out_unlockput;
1581 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1582 	if (error)
1583 		goto out_q;
1584 
1585 	/* Check for existing entry, if none then alloc new blocks */
1586 	error = update_qd(sdp, qd);
1587 	if (error)
1588 		goto out_i;
1589 
1590 	/* If nothing has changed, this is a no-op */
1591 	if ((fdq->d_fieldmask & FS_DQ_BSOFT) &&
1592 	    ((fdq->d_blk_softlimit >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_warn)))
1593 		fdq->d_fieldmask ^= FS_DQ_BSOFT;
1594 
1595 	if ((fdq->d_fieldmask & FS_DQ_BHARD) &&
1596 	    ((fdq->d_blk_hardlimit >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_limit)))
1597 		fdq->d_fieldmask ^= FS_DQ_BHARD;
1598 
1599 	if ((fdq->d_fieldmask & FS_DQ_BCOUNT) &&
1600 	    ((fdq->d_bcount >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_value)))
1601 		fdq->d_fieldmask ^= FS_DQ_BCOUNT;
1602 
1603 	if (fdq->d_fieldmask == 0)
1604 		goto out_i;
1605 
1606 	offset = qd2offset(qd);
1607 	alloc_required = gfs2_write_alloc_required(ip, offset, sizeof(struct gfs2_quota));
1608 	if (gfs2_is_stuffed(ip))
1609 		alloc_required = 1;
1610 	if (alloc_required) {
1611 		struct gfs2_alloc_parms ap = { .aflags = 0, };
1612 		gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
1613 				       &data_blocks, &ind_blocks);
1614 		blocks = 1 + data_blocks + ind_blocks;
1615 		ap.target = blocks;
1616 		error = gfs2_inplace_reserve(ip, &ap);
1617 		if (error)
1618 			goto out_i;
1619 		blocks += gfs2_rg_blocks(ip, blocks);
1620 	}
1621 
1622 	/* Some quotas span block boundaries and can update two blocks,
1623 	   adding an extra block to the transaction to handle such quotas */
1624 	error = gfs2_trans_begin(sdp, blocks + RES_DINODE + 2, 0);
1625 	if (error)
1626 		goto out_release;
1627 
1628 	/* Apply changes */
1629 	error = gfs2_adjust_quota(ip, offset, 0, qd, fdq);
1630 
1631 	gfs2_trans_end(sdp);
1632 out_release:
1633 	if (alloc_required)
1634 		gfs2_inplace_release(ip);
1635 out_i:
1636 	gfs2_glock_dq_uninit(&i_gh);
1637 out_q:
1638 	gfs2_glock_dq_uninit(&q_gh);
1639 out_unlockput:
1640 	mutex_unlock(&ip->i_inode.i_mutex);
1641 out_put:
1642 	qd_put(qd);
1643 	return error;
1644 }
1645 
1646 const struct quotactl_ops gfs2_quotactl_ops = {
1647 	.quota_sync     = gfs2_quota_sync,
1648 	.get_xstate     = gfs2_quota_get_xstate,
1649 	.get_dqblk	= gfs2_get_dqblk,
1650 	.set_dqblk	= gfs2_set_dqblk,
1651 };
1652 
1653 void __init gfs2_quota_hash_init(void)
1654 {
1655 	unsigned i;
1656 
1657 	for(i = 0; i < GFS2_QD_HASH_SIZE; i++)
1658 		INIT_HLIST_BL_HEAD(&qd_hash_table[i]);
1659 }
1660