xref: /linux/fs/gfs2/quota.c (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
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_name.ln_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,
149 		struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
150 {
151 	struct list_head *dispose = arg;
152 	struct gfs2_quota_data *qd = list_entry(item, struct gfs2_quota_data, qd_lru);
153 
154 	if (!spin_trylock(&qd->qd_lockref.lock))
155 		return LRU_SKIP;
156 
157 	if (qd->qd_lockref.count == 0) {
158 		lockref_mark_dead(&qd->qd_lockref);
159 		list_lru_isolate_move(lru, &qd->qd_lru, dispose);
160 	}
161 
162 	spin_unlock(&qd->qd_lockref.lock);
163 	return LRU_REMOVED;
164 }
165 
166 static unsigned long gfs2_qd_shrink_scan(struct shrinker *shrink,
167 					 struct shrink_control *sc)
168 {
169 	LIST_HEAD(dispose);
170 	unsigned long freed;
171 
172 	if (!(sc->gfp_mask & __GFP_FS))
173 		return SHRINK_STOP;
174 
175 	freed = list_lru_shrink_walk(&gfs2_qd_lru, sc,
176 				     gfs2_qd_isolate, &dispose);
177 
178 	gfs2_qd_dispose(&dispose);
179 
180 	return freed;
181 }
182 
183 static unsigned long gfs2_qd_shrink_count(struct shrinker *shrink,
184 					  struct shrink_control *sc)
185 {
186 	return vfs_pressure_ratio(list_lru_shrink_count(&gfs2_qd_lru, sc));
187 }
188 
189 struct shrinker gfs2_qd_shrinker = {
190 	.count_objects = gfs2_qd_shrink_count,
191 	.scan_objects = gfs2_qd_shrink_scan,
192 	.seeks = DEFAULT_SEEKS,
193 	.flags = SHRINKER_NUMA_AWARE,
194 };
195 
196 
197 static u64 qd2index(struct gfs2_quota_data *qd)
198 {
199 	struct kqid qid = qd->qd_id;
200 	return (2 * (u64)from_kqid(&init_user_ns, qid)) +
201 		((qid.type == USRQUOTA) ? 0 : 1);
202 }
203 
204 static u64 qd2offset(struct gfs2_quota_data *qd)
205 {
206 	u64 offset;
207 
208 	offset = qd2index(qd);
209 	offset *= sizeof(struct gfs2_quota);
210 
211 	return offset;
212 }
213 
214 static struct gfs2_quota_data *qd_alloc(unsigned hash, struct gfs2_sbd *sdp, struct kqid qid)
215 {
216 	struct gfs2_quota_data *qd;
217 	int error;
218 
219 	qd = kmem_cache_zalloc(gfs2_quotad_cachep, GFP_NOFS);
220 	if (!qd)
221 		return NULL;
222 
223 	qd->qd_sbd = sdp;
224 	qd->qd_lockref.count = 1;
225 	spin_lock_init(&qd->qd_lockref.lock);
226 	qd->qd_id = qid;
227 	qd->qd_slot = -1;
228 	INIT_LIST_HEAD(&qd->qd_lru);
229 	qd->qd_hash = hash;
230 
231 	error = gfs2_glock_get(sdp, qd2index(qd),
232 			      &gfs2_quota_glops, CREATE, &qd->qd_gl);
233 	if (error)
234 		goto fail;
235 
236 	return qd;
237 
238 fail:
239 	kmem_cache_free(gfs2_quotad_cachep, qd);
240 	return NULL;
241 }
242 
243 static struct gfs2_quota_data *gfs2_qd_search_bucket(unsigned int hash,
244 						     const struct gfs2_sbd *sdp,
245 						     struct kqid qid)
246 {
247 	struct gfs2_quota_data *qd;
248 	struct hlist_bl_node *h;
249 
250 	hlist_bl_for_each_entry_rcu(qd, h, &qd_hash_table[hash], qd_hlist) {
251 		if (!qid_eq(qd->qd_id, qid))
252 			continue;
253 		if (qd->qd_sbd != sdp)
254 			continue;
255 		if (lockref_get_not_dead(&qd->qd_lockref)) {
256 			list_lru_del(&gfs2_qd_lru, &qd->qd_lru);
257 			return qd;
258 		}
259 	}
260 
261 	return NULL;
262 }
263 
264 
265 static int qd_get(struct gfs2_sbd *sdp, struct kqid qid,
266 		  struct gfs2_quota_data **qdp)
267 {
268 	struct gfs2_quota_data *qd, *new_qd;
269 	unsigned int hash = gfs2_qd_hash(sdp, qid);
270 
271 	rcu_read_lock();
272 	*qdp = qd = gfs2_qd_search_bucket(hash, sdp, qid);
273 	rcu_read_unlock();
274 
275 	if (qd)
276 		return 0;
277 
278 	new_qd = qd_alloc(hash, sdp, qid);
279 	if (!new_qd)
280 		return -ENOMEM;
281 
282 	spin_lock(&qd_lock);
283 	spin_lock_bucket(hash);
284 	*qdp = qd = gfs2_qd_search_bucket(hash, sdp, qid);
285 	if (qd == NULL) {
286 		*qdp = new_qd;
287 		list_add(&new_qd->qd_list, &sdp->sd_quota_list);
288 		hlist_bl_add_head_rcu(&new_qd->qd_hlist, &qd_hash_table[hash]);
289 		atomic_inc(&sdp->sd_quota_count);
290 	}
291 	spin_unlock_bucket(hash);
292 	spin_unlock(&qd_lock);
293 
294 	if (qd) {
295 		gfs2_glock_put(new_qd->qd_gl);
296 		kmem_cache_free(gfs2_quotad_cachep, new_qd);
297 	}
298 
299 	return 0;
300 }
301 
302 
303 static void qd_hold(struct gfs2_quota_data *qd)
304 {
305 	struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
306 	gfs2_assert(sdp, !__lockref_is_dead(&qd->qd_lockref));
307 	lockref_get(&qd->qd_lockref);
308 }
309 
310 static void qd_put(struct gfs2_quota_data *qd)
311 {
312 	if (lockref_put_or_lock(&qd->qd_lockref))
313 		return;
314 
315 	qd->qd_lockref.count = 0;
316 	list_lru_add(&gfs2_qd_lru, &qd->qd_lru);
317 	spin_unlock(&qd->qd_lockref.lock);
318 
319 }
320 
321 static int slot_get(struct gfs2_quota_data *qd)
322 {
323 	struct gfs2_sbd *sdp = qd->qd_sbd;
324 	unsigned int bit;
325 	int error = 0;
326 
327 	spin_lock(&sdp->sd_bitmap_lock);
328 	if (qd->qd_slot_count != 0)
329 		goto out;
330 
331 	error = -ENOSPC;
332 	bit = find_first_zero_bit(sdp->sd_quota_bitmap, sdp->sd_quota_slots);
333 	if (bit < sdp->sd_quota_slots) {
334 		set_bit(bit, sdp->sd_quota_bitmap);
335 		qd->qd_slot = bit;
336 		error = 0;
337 out:
338 		qd->qd_slot_count++;
339 	}
340 	spin_unlock(&sdp->sd_bitmap_lock);
341 
342 	return error;
343 }
344 
345 static void slot_hold(struct gfs2_quota_data *qd)
346 {
347 	struct gfs2_sbd *sdp = qd->qd_sbd;
348 
349 	spin_lock(&sdp->sd_bitmap_lock);
350 	gfs2_assert(sdp, qd->qd_slot_count);
351 	qd->qd_slot_count++;
352 	spin_unlock(&sdp->sd_bitmap_lock);
353 }
354 
355 static void slot_put(struct gfs2_quota_data *qd)
356 {
357 	struct gfs2_sbd *sdp = qd->qd_sbd;
358 
359 	spin_lock(&sdp->sd_bitmap_lock);
360 	gfs2_assert(sdp, qd->qd_slot_count);
361 	if (!--qd->qd_slot_count) {
362 		BUG_ON(!test_and_clear_bit(qd->qd_slot, sdp->sd_quota_bitmap));
363 		qd->qd_slot = -1;
364 	}
365 	spin_unlock(&sdp->sd_bitmap_lock);
366 }
367 
368 static int bh_get(struct gfs2_quota_data *qd)
369 {
370 	struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
371 	struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
372 	unsigned int block, offset;
373 	struct buffer_head *bh;
374 	int error;
375 	struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
376 
377 	mutex_lock(&sdp->sd_quota_mutex);
378 
379 	if (qd->qd_bh_count++) {
380 		mutex_unlock(&sdp->sd_quota_mutex);
381 		return 0;
382 	}
383 
384 	block = qd->qd_slot / sdp->sd_qc_per_block;
385 	offset = qd->qd_slot % sdp->sd_qc_per_block;
386 
387 	bh_map.b_size = 1 << ip->i_inode.i_blkbits;
388 	error = gfs2_block_map(&ip->i_inode, block, &bh_map, 0);
389 	if (error)
390 		goto fail;
391 	error = gfs2_meta_read(ip->i_gl, bh_map.b_blocknr, DIO_WAIT, &bh);
392 	if (error)
393 		goto fail;
394 	error = -EIO;
395 	if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC))
396 		goto fail_brelse;
397 
398 	qd->qd_bh = bh;
399 	qd->qd_bh_qc = (struct gfs2_quota_change *)
400 		(bh->b_data + sizeof(struct gfs2_meta_header) +
401 		 offset * sizeof(struct gfs2_quota_change));
402 
403 	mutex_unlock(&sdp->sd_quota_mutex);
404 
405 	return 0;
406 
407 fail_brelse:
408 	brelse(bh);
409 fail:
410 	qd->qd_bh_count--;
411 	mutex_unlock(&sdp->sd_quota_mutex);
412 	return error;
413 }
414 
415 static void bh_put(struct gfs2_quota_data *qd)
416 {
417 	struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
418 
419 	mutex_lock(&sdp->sd_quota_mutex);
420 	gfs2_assert(sdp, qd->qd_bh_count);
421 	if (!--qd->qd_bh_count) {
422 		brelse(qd->qd_bh);
423 		qd->qd_bh = NULL;
424 		qd->qd_bh_qc = NULL;
425 	}
426 	mutex_unlock(&sdp->sd_quota_mutex);
427 }
428 
429 static int qd_check_sync(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd,
430 			 u64 *sync_gen)
431 {
432 	if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
433 	    !test_bit(QDF_CHANGE, &qd->qd_flags) ||
434 	    (sync_gen && (qd->qd_sync_gen >= *sync_gen)))
435 		return 0;
436 
437 	if (!lockref_get_not_dead(&qd->qd_lockref))
438 		return 0;
439 
440 	list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
441 	set_bit(QDF_LOCKED, &qd->qd_flags);
442 	qd->qd_change_sync = qd->qd_change;
443 	slot_hold(qd);
444 	return 1;
445 }
446 
447 static int qd_fish(struct gfs2_sbd *sdp, struct gfs2_quota_data **qdp)
448 {
449 	struct gfs2_quota_data *qd = NULL;
450 	int error;
451 	int found = 0;
452 
453 	*qdp = NULL;
454 
455 	if (sdp->sd_vfs->s_flags & MS_RDONLY)
456 		return 0;
457 
458 	spin_lock(&qd_lock);
459 
460 	list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
461 		found = qd_check_sync(sdp, qd, &sdp->sd_quota_sync_gen);
462 		if (found)
463 			break;
464 	}
465 
466 	if (!found)
467 		qd = NULL;
468 
469 	spin_unlock(&qd_lock);
470 
471 	if (qd) {
472 		gfs2_assert_warn(sdp, qd->qd_change_sync);
473 		error = bh_get(qd);
474 		if (error) {
475 			clear_bit(QDF_LOCKED, &qd->qd_flags);
476 			slot_put(qd);
477 			qd_put(qd);
478 			return error;
479 		}
480 	}
481 
482 	*qdp = qd;
483 
484 	return 0;
485 }
486 
487 static void qd_unlock(struct gfs2_quota_data *qd)
488 {
489 	gfs2_assert_warn(qd->qd_gl->gl_name.ln_sbd,
490 			 test_bit(QDF_LOCKED, &qd->qd_flags));
491 	clear_bit(QDF_LOCKED, &qd->qd_flags);
492 	bh_put(qd);
493 	slot_put(qd);
494 	qd_put(qd);
495 }
496 
497 static int qdsb_get(struct gfs2_sbd *sdp, struct kqid qid,
498 		    struct gfs2_quota_data **qdp)
499 {
500 	int error;
501 
502 	error = qd_get(sdp, qid, qdp);
503 	if (error)
504 		return error;
505 
506 	error = slot_get(*qdp);
507 	if (error)
508 		goto fail;
509 
510 	error = bh_get(*qdp);
511 	if (error)
512 		goto fail_slot;
513 
514 	return 0;
515 
516 fail_slot:
517 	slot_put(*qdp);
518 fail:
519 	qd_put(*qdp);
520 	return error;
521 }
522 
523 static void qdsb_put(struct gfs2_quota_data *qd)
524 {
525 	bh_put(qd);
526 	slot_put(qd);
527 	qd_put(qd);
528 }
529 
530 int gfs2_quota_hold(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
531 {
532 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
533 	struct gfs2_quota_data **qd;
534 	int error;
535 
536 	if (ip->i_res == NULL) {
537 		error = gfs2_rs_alloc(ip);
538 		if (error)
539 			return error;
540 	}
541 
542 	qd = ip->i_res->rs_qa_qd;
543 
544 	if (gfs2_assert_warn(sdp, !ip->i_res->rs_qa_qd_num) ||
545 	    gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags)))
546 		return -EIO;
547 
548 	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
549 		return 0;
550 
551 	error = qdsb_get(sdp, make_kqid_uid(ip->i_inode.i_uid), qd);
552 	if (error)
553 		goto out;
554 	ip->i_res->rs_qa_qd_num++;
555 	qd++;
556 
557 	error = qdsb_get(sdp, make_kqid_gid(ip->i_inode.i_gid), qd);
558 	if (error)
559 		goto out;
560 	ip->i_res->rs_qa_qd_num++;
561 	qd++;
562 
563 	if (!uid_eq(uid, NO_UID_QUOTA_CHANGE) &&
564 	    !uid_eq(uid, ip->i_inode.i_uid)) {
565 		error = qdsb_get(sdp, make_kqid_uid(uid), qd);
566 		if (error)
567 			goto out;
568 		ip->i_res->rs_qa_qd_num++;
569 		qd++;
570 	}
571 
572 	if (!gid_eq(gid, NO_GID_QUOTA_CHANGE) &&
573 	    !gid_eq(gid, ip->i_inode.i_gid)) {
574 		error = qdsb_get(sdp, make_kqid_gid(gid), qd);
575 		if (error)
576 			goto out;
577 		ip->i_res->rs_qa_qd_num++;
578 		qd++;
579 	}
580 
581 out:
582 	if (error)
583 		gfs2_quota_unhold(ip);
584 	return error;
585 }
586 
587 void gfs2_quota_unhold(struct gfs2_inode *ip)
588 {
589 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
590 	unsigned int x;
591 
592 	if (ip->i_res == NULL)
593 		return;
594 	gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags));
595 
596 	for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
597 		qdsb_put(ip->i_res->rs_qa_qd[x]);
598 		ip->i_res->rs_qa_qd[x] = NULL;
599 	}
600 	ip->i_res->rs_qa_qd_num = 0;
601 }
602 
603 static int sort_qd(const void *a, const void *b)
604 {
605 	const struct gfs2_quota_data *qd_a = *(const struct gfs2_quota_data **)a;
606 	const struct gfs2_quota_data *qd_b = *(const struct gfs2_quota_data **)b;
607 
608 	if (qid_lt(qd_a->qd_id, qd_b->qd_id))
609 		return -1;
610 	if (qid_lt(qd_b->qd_id, qd_a->qd_id))
611 		return 1;
612 	return 0;
613 }
614 
615 static void do_qc(struct gfs2_quota_data *qd, s64 change)
616 {
617 	struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
618 	struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
619 	struct gfs2_quota_change *qc = qd->qd_bh_qc;
620 	s64 x;
621 
622 	mutex_lock(&sdp->sd_quota_mutex);
623 	gfs2_trans_add_meta(ip->i_gl, qd->qd_bh);
624 
625 	if (!test_bit(QDF_CHANGE, &qd->qd_flags)) {
626 		qc->qc_change = 0;
627 		qc->qc_flags = 0;
628 		if (qd->qd_id.type == USRQUOTA)
629 			qc->qc_flags = cpu_to_be32(GFS2_QCF_USER);
630 		qc->qc_id = cpu_to_be32(from_kqid(&init_user_ns, qd->qd_id));
631 	}
632 
633 	x = be64_to_cpu(qc->qc_change) + change;
634 	qc->qc_change = cpu_to_be64(x);
635 
636 	spin_lock(&qd_lock);
637 	qd->qd_change = x;
638 	spin_unlock(&qd_lock);
639 
640 	if (!x) {
641 		gfs2_assert_warn(sdp, test_bit(QDF_CHANGE, &qd->qd_flags));
642 		clear_bit(QDF_CHANGE, &qd->qd_flags);
643 		qc->qc_flags = 0;
644 		qc->qc_id = 0;
645 		slot_put(qd);
646 		qd_put(qd);
647 	} else if (!test_and_set_bit(QDF_CHANGE, &qd->qd_flags)) {
648 		qd_hold(qd);
649 		slot_hold(qd);
650 	}
651 
652 	if (change < 0) /* Reset quiet flag if we freed some blocks */
653 		clear_bit(QDF_QMSG_QUIET, &qd->qd_flags);
654 	mutex_unlock(&sdp->sd_quota_mutex);
655 }
656 
657 static int gfs2_write_buf_to_page(struct gfs2_inode *ip, unsigned long index,
658 				  unsigned off, void *buf, unsigned bytes)
659 {
660 	struct inode *inode = &ip->i_inode;
661 	struct gfs2_sbd *sdp = GFS2_SB(inode);
662 	struct address_space *mapping = inode->i_mapping;
663 	struct page *page;
664 	struct buffer_head *bh;
665 	void *kaddr;
666 	u64 blk;
667 	unsigned bsize = sdp->sd_sb.sb_bsize, bnum = 0, boff = 0;
668 	unsigned to_write = bytes, pg_off = off;
669 	int done = 0;
670 
671 	blk = index << (PAGE_CACHE_SHIFT - sdp->sd_sb.sb_bsize_shift);
672 	boff = off % bsize;
673 
674 	page = find_or_create_page(mapping, index, GFP_NOFS);
675 	if (!page)
676 		return -ENOMEM;
677 	if (!page_has_buffers(page))
678 		create_empty_buffers(page, bsize, 0);
679 
680 	bh = page_buffers(page);
681 	while (!done) {
682 		/* Find the beginning block within the page */
683 		if (pg_off >= ((bnum * bsize) + bsize)) {
684 			bh = bh->b_this_page;
685 			bnum++;
686 			blk++;
687 			continue;
688 		}
689 		if (!buffer_mapped(bh)) {
690 			gfs2_block_map(inode, blk, bh, 1);
691 			if (!buffer_mapped(bh))
692 				goto unlock_out;
693 			/* If it's a newly allocated disk block, zero it */
694 			if (buffer_new(bh))
695 				zero_user(page, bnum * bsize, bh->b_size);
696 		}
697 		if (PageUptodate(page))
698 			set_buffer_uptodate(bh);
699 		if (!buffer_uptodate(bh)) {
700 			ll_rw_block(READ | REQ_META, 1, &bh);
701 			wait_on_buffer(bh);
702 			if (!buffer_uptodate(bh))
703 				goto unlock_out;
704 		}
705 		gfs2_trans_add_data(ip->i_gl, bh);
706 
707 		/* If we need to write to the next block as well */
708 		if (to_write > (bsize - boff)) {
709 			pg_off += (bsize - boff);
710 			to_write -= (bsize - boff);
711 			boff = pg_off % bsize;
712 			continue;
713 		}
714 		done = 1;
715 	}
716 
717 	/* Write to the page, now that we have setup the buffer(s) */
718 	kaddr = kmap_atomic(page);
719 	memcpy(kaddr + off, buf, bytes);
720 	flush_dcache_page(page);
721 	kunmap_atomic(kaddr);
722 	unlock_page(page);
723 	page_cache_release(page);
724 
725 	return 0;
726 
727 unlock_out:
728 	unlock_page(page);
729 	page_cache_release(page);
730 	return -EIO;
731 }
732 
733 static int gfs2_write_disk_quota(struct gfs2_inode *ip, struct gfs2_quota *qp,
734 				 loff_t loc)
735 {
736 	unsigned long pg_beg;
737 	unsigned pg_off, nbytes, overflow = 0;
738 	int pg_oflow = 0, error;
739 	void *ptr;
740 
741 	nbytes = sizeof(struct gfs2_quota);
742 
743 	pg_beg = loc >> PAGE_CACHE_SHIFT;
744 	pg_off = loc % PAGE_CACHE_SIZE;
745 
746 	/* If the quota straddles a page boundary, split the write in two */
747 	if ((pg_off + nbytes) > PAGE_CACHE_SIZE) {
748 		pg_oflow = 1;
749 		overflow = (pg_off + nbytes) - PAGE_CACHE_SIZE;
750 	}
751 
752 	ptr = qp;
753 	error = gfs2_write_buf_to_page(ip, pg_beg, pg_off, ptr,
754 				       nbytes - overflow);
755 	/* If there's an overflow, write the remaining bytes to the next page */
756 	if (!error && pg_oflow)
757 		error = gfs2_write_buf_to_page(ip, pg_beg + 1, 0,
758 					       ptr + nbytes - overflow,
759 					       overflow);
760 	return error;
761 }
762 
763 /**
764  * gfs2_adjust_quota - adjust record of current block usage
765  * @ip: The quota inode
766  * @loc: Offset of the entry in the quota file
767  * @change: The amount of usage change to record
768  * @qd: The quota data
769  * @fdq: The updated limits to record
770  *
771  * This function was mostly borrowed from gfs2_block_truncate_page which was
772  * in turn mostly borrowed from ext3
773  *
774  * Returns: 0 or -ve on error
775  */
776 
777 static int gfs2_adjust_quota(struct gfs2_inode *ip, loff_t loc,
778 			     s64 change, struct gfs2_quota_data *qd,
779 			     struct qc_dqblk *fdq)
780 {
781 	struct inode *inode = &ip->i_inode;
782 	struct gfs2_sbd *sdp = GFS2_SB(inode);
783 	struct gfs2_quota q;
784 	int err;
785 	u64 size;
786 
787 	if (gfs2_is_stuffed(ip)) {
788 		err = gfs2_unstuff_dinode(ip, NULL);
789 		if (err)
790 			return err;
791 	}
792 
793 	memset(&q, 0, sizeof(struct gfs2_quota));
794 	err = gfs2_internal_read(ip, (char *)&q, &loc, sizeof(q));
795 	if (err < 0)
796 		return err;
797 
798 	loc -= sizeof(q); /* gfs2_internal_read would've advanced the loc ptr */
799 	err = -EIO;
800 	be64_add_cpu(&q.qu_value, change);
801 	if (((s64)be64_to_cpu(q.qu_value)) < 0)
802 		q.qu_value = 0; /* Never go negative on quota usage */
803 	qd->qd_qb.qb_value = q.qu_value;
804 	if (fdq) {
805 		if (fdq->d_fieldmask & QC_SPC_SOFT) {
806 			q.qu_warn = cpu_to_be64(fdq->d_spc_softlimit >> sdp->sd_sb.sb_bsize_shift);
807 			qd->qd_qb.qb_warn = q.qu_warn;
808 		}
809 		if (fdq->d_fieldmask & QC_SPC_HARD) {
810 			q.qu_limit = cpu_to_be64(fdq->d_spc_hardlimit >> sdp->sd_sb.sb_bsize_shift);
811 			qd->qd_qb.qb_limit = q.qu_limit;
812 		}
813 		if (fdq->d_fieldmask & QC_SPACE) {
814 			q.qu_value = cpu_to_be64(fdq->d_space >> sdp->sd_sb.sb_bsize_shift);
815 			qd->qd_qb.qb_value = q.qu_value;
816 		}
817 	}
818 
819 	err = gfs2_write_disk_quota(ip, &q, loc);
820 	if (!err) {
821 		size = loc + sizeof(struct gfs2_quota);
822 		if (size > inode->i_size)
823 			i_size_write(inode, size);
824 		inode->i_mtime = inode->i_atime = CURRENT_TIME;
825 		mark_inode_dirty(inode);
826 		set_bit(QDF_REFRESH, &qd->qd_flags);
827 	}
828 
829 	return err;
830 }
831 
832 static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
833 {
834 	struct gfs2_sbd *sdp = (*qda)->qd_gl->gl_name.ln_sbd;
835 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
836 	struct gfs2_alloc_parms ap = { .aflags = 0, };
837 	unsigned int data_blocks, ind_blocks;
838 	struct gfs2_holder *ghs, i_gh;
839 	unsigned int qx, x;
840 	struct gfs2_quota_data *qd;
841 	unsigned reserved;
842 	loff_t offset;
843 	unsigned int nalloc = 0, blocks;
844 	int error;
845 
846 	error = gfs2_rs_alloc(ip);
847 	if (error)
848 		return error;
849 
850 	gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
851 			      &data_blocks, &ind_blocks);
852 
853 	ghs = kcalloc(num_qd, sizeof(struct gfs2_holder), GFP_NOFS);
854 	if (!ghs)
855 		return -ENOMEM;
856 
857 	sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL);
858 	mutex_lock(&ip->i_inode.i_mutex);
859 	for (qx = 0; qx < num_qd; qx++) {
860 		error = gfs2_glock_nq_init(qda[qx]->qd_gl, LM_ST_EXCLUSIVE,
861 					   GL_NOCACHE, &ghs[qx]);
862 		if (error)
863 			goto out;
864 	}
865 
866 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
867 	if (error)
868 		goto out;
869 
870 	for (x = 0; x < num_qd; x++) {
871 		offset = qd2offset(qda[x]);
872 		if (gfs2_write_alloc_required(ip, offset,
873 					      sizeof(struct gfs2_quota)))
874 			nalloc++;
875 	}
876 
877 	/*
878 	 * 1 blk for unstuffing inode if stuffed. We add this extra
879 	 * block to the reservation unconditionally. If the inode
880 	 * doesn't need unstuffing, the block will be released to the
881 	 * rgrp since it won't be allocated during the transaction
882 	 */
883 	/* +3 in the end for unstuffing block, inode size update block
884 	 * and another block in case quota straddles page boundary and
885 	 * two blocks need to be updated instead of 1 */
886 	blocks = num_qd * data_blocks + RES_DINODE + num_qd + 3;
887 
888 	reserved = 1 + (nalloc * (data_blocks + ind_blocks));
889 	ap.target = reserved;
890 	error = gfs2_inplace_reserve(ip, &ap);
891 	if (error)
892 		goto out_alloc;
893 
894 	if (nalloc)
895 		blocks += gfs2_rg_blocks(ip, reserved) + nalloc * ind_blocks + RES_STATFS;
896 
897 	error = gfs2_trans_begin(sdp, blocks, 0);
898 	if (error)
899 		goto out_ipres;
900 
901 	for (x = 0; x < num_qd; x++) {
902 		qd = qda[x];
903 		offset = qd2offset(qd);
904 		error = gfs2_adjust_quota(ip, offset, qd->qd_change_sync, qd, NULL);
905 		if (error)
906 			goto out_end_trans;
907 
908 		do_qc(qd, -qd->qd_change_sync);
909 		set_bit(QDF_REFRESH, &qd->qd_flags);
910 	}
911 
912 	error = 0;
913 
914 out_end_trans:
915 	gfs2_trans_end(sdp);
916 out_ipres:
917 	gfs2_inplace_release(ip);
918 out_alloc:
919 	gfs2_glock_dq_uninit(&i_gh);
920 out:
921 	while (qx--)
922 		gfs2_glock_dq_uninit(&ghs[qx]);
923 	mutex_unlock(&ip->i_inode.i_mutex);
924 	kfree(ghs);
925 	gfs2_log_flush(ip->i_gl->gl_name.ln_sbd, ip->i_gl, NORMAL_FLUSH);
926 	return error;
927 }
928 
929 static int update_qd(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd)
930 {
931 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
932 	struct gfs2_quota q;
933 	struct gfs2_quota_lvb *qlvb;
934 	loff_t pos;
935 	int error;
936 
937 	memset(&q, 0, sizeof(struct gfs2_quota));
938 	pos = qd2offset(qd);
939 	error = gfs2_internal_read(ip, (char *)&q, &pos, sizeof(q));
940 	if (error < 0)
941 		return error;
942 
943 	qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
944 	qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC);
945 	qlvb->__pad = 0;
946 	qlvb->qb_limit = q.qu_limit;
947 	qlvb->qb_warn = q.qu_warn;
948 	qlvb->qb_value = q.qu_value;
949 	qd->qd_qb = *qlvb;
950 
951 	return 0;
952 }
953 
954 static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
955 		    struct gfs2_holder *q_gh)
956 {
957 	struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
958 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
959 	struct gfs2_holder i_gh;
960 	int error;
961 
962 restart:
963 	error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
964 	if (error)
965 		return error;
966 
967 	if (test_and_clear_bit(QDF_REFRESH, &qd->qd_flags))
968 		force_refresh = FORCE;
969 
970 	qd->qd_qb = *(struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
971 
972 	if (force_refresh || qd->qd_qb.qb_magic != cpu_to_be32(GFS2_MAGIC)) {
973 		gfs2_glock_dq_uninit(q_gh);
974 		error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE,
975 					   GL_NOCACHE, q_gh);
976 		if (error)
977 			return error;
978 
979 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
980 		if (error)
981 			goto fail;
982 
983 		error = update_qd(sdp, qd);
984 		if (error)
985 			goto fail_gunlock;
986 
987 		gfs2_glock_dq_uninit(&i_gh);
988 		gfs2_glock_dq_uninit(q_gh);
989 		force_refresh = 0;
990 		goto restart;
991 	}
992 
993 	return 0;
994 
995 fail_gunlock:
996 	gfs2_glock_dq_uninit(&i_gh);
997 fail:
998 	gfs2_glock_dq_uninit(q_gh);
999 	return error;
1000 }
1001 
1002 int gfs2_quota_lock(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
1003 {
1004 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1005 	struct gfs2_quota_data *qd;
1006 	unsigned int x;
1007 	int error = 0;
1008 
1009 	error = gfs2_quota_hold(ip, uid, gid);
1010 	if (error)
1011 		return error;
1012 
1013 	if (capable(CAP_SYS_RESOURCE) ||
1014 	    sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
1015 		return 0;
1016 
1017 	sort(ip->i_res->rs_qa_qd, ip->i_res->rs_qa_qd_num,
1018 	     sizeof(struct gfs2_quota_data *), sort_qd, NULL);
1019 
1020 	for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1021 		qd = ip->i_res->rs_qa_qd[x];
1022 		error = do_glock(qd, NO_FORCE, &ip->i_res->rs_qa_qd_ghs[x]);
1023 		if (error)
1024 			break;
1025 	}
1026 
1027 	if (!error)
1028 		set_bit(GIF_QD_LOCKED, &ip->i_flags);
1029 	else {
1030 		while (x--)
1031 			gfs2_glock_dq_uninit(&ip->i_res->rs_qa_qd_ghs[x]);
1032 		gfs2_quota_unhold(ip);
1033 	}
1034 
1035 	return error;
1036 }
1037 
1038 static int need_sync(struct gfs2_quota_data *qd)
1039 {
1040 	struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
1041 	struct gfs2_tune *gt = &sdp->sd_tune;
1042 	s64 value;
1043 	unsigned int num, den;
1044 	int do_sync = 1;
1045 
1046 	if (!qd->qd_qb.qb_limit)
1047 		return 0;
1048 
1049 	spin_lock(&qd_lock);
1050 	value = qd->qd_change;
1051 	spin_unlock(&qd_lock);
1052 
1053 	spin_lock(&gt->gt_spin);
1054 	num = gt->gt_quota_scale_num;
1055 	den = gt->gt_quota_scale_den;
1056 	spin_unlock(&gt->gt_spin);
1057 
1058 	if (value < 0)
1059 		do_sync = 0;
1060 	else if ((s64)be64_to_cpu(qd->qd_qb.qb_value) >=
1061 		 (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1062 		do_sync = 0;
1063 	else {
1064 		value *= gfs2_jindex_size(sdp) * num;
1065 		value = div_s64(value, den);
1066 		value += (s64)be64_to_cpu(qd->qd_qb.qb_value);
1067 		if (value < (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1068 			do_sync = 0;
1069 	}
1070 
1071 	return do_sync;
1072 }
1073 
1074 void gfs2_quota_unlock(struct gfs2_inode *ip)
1075 {
1076 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1077 	struct gfs2_quota_data *qda[4];
1078 	unsigned int count = 0;
1079 	unsigned int x;
1080 	int found;
1081 
1082 	if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
1083 		goto out;
1084 
1085 	for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1086 		struct gfs2_quota_data *qd;
1087 		int sync;
1088 
1089 		qd = ip->i_res->rs_qa_qd[x];
1090 		sync = need_sync(qd);
1091 
1092 		gfs2_glock_dq_uninit(&ip->i_res->rs_qa_qd_ghs[x]);
1093 		if (!sync)
1094 			continue;
1095 
1096 		spin_lock(&qd_lock);
1097 		found = qd_check_sync(sdp, qd, NULL);
1098 		spin_unlock(&qd_lock);
1099 
1100 		if (!found)
1101 			continue;
1102 
1103 		gfs2_assert_warn(sdp, qd->qd_change_sync);
1104 		if (bh_get(qd)) {
1105 			clear_bit(QDF_LOCKED, &qd->qd_flags);
1106 			slot_put(qd);
1107 			qd_put(qd);
1108 			continue;
1109 		}
1110 
1111 		qda[count++] = qd;
1112 	}
1113 
1114 	if (count) {
1115 		do_sync(count, qda);
1116 		for (x = 0; x < count; x++)
1117 			qd_unlock(qda[x]);
1118 	}
1119 
1120 out:
1121 	gfs2_quota_unhold(ip);
1122 }
1123 
1124 #define MAX_LINE 256
1125 
1126 static int print_message(struct gfs2_quota_data *qd, char *type)
1127 {
1128 	struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
1129 
1130 	fs_info(sdp, "quota %s for %s %u\n",
1131 		type,
1132 		(qd->qd_id.type == USRQUOTA) ? "user" : "group",
1133 		from_kqid(&init_user_ns, qd->qd_id));
1134 
1135 	return 0;
1136 }
1137 
1138 /**
1139  * gfs2_quota_check - check if allocating new blocks will exceed quota
1140  * @ip:  The inode for which this check is being performed
1141  * @uid: The uid to check against
1142  * @gid: The gid to check against
1143  * @ap:  The allocation parameters. ap->target contains the requested
1144  *       blocks. ap->min_target, if set, contains the minimum blks
1145  *       requested.
1146  *
1147  * Returns: 0 on success.
1148  *                  min_req = ap->min_target ? ap->min_target : ap->target;
1149  *                  quota must allow atleast min_req blks for success and
1150  *                  ap->allowed is set to the number of blocks allowed
1151  *
1152  *          -EDQUOT otherwise, quota violation. ap->allowed is set to number
1153  *                  of blocks available.
1154  */
1155 int gfs2_quota_check(struct gfs2_inode *ip, kuid_t uid, kgid_t gid,
1156 		     struct gfs2_alloc_parms *ap)
1157 {
1158 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1159 	struct gfs2_quota_data *qd;
1160 	s64 value, warn, limit;
1161 	unsigned int x;
1162 	int error = 0;
1163 
1164 	ap->allowed = UINT_MAX; /* Assume we are permitted a whole lot */
1165 	if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
1166 		return 0;
1167 
1168         if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
1169                 return 0;
1170 
1171 	for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1172 		qd = ip->i_res->rs_qa_qd[x];
1173 
1174 		if (!(qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1175 		      qid_eq(qd->qd_id, make_kqid_gid(gid))))
1176 			continue;
1177 
1178 		warn = (s64)be64_to_cpu(qd->qd_qb.qb_warn);
1179 		limit = (s64)be64_to_cpu(qd->qd_qb.qb_limit);
1180 		value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
1181 		spin_lock(&qd_lock);
1182 		value += qd->qd_change;
1183 		spin_unlock(&qd_lock);
1184 
1185 		if (limit > 0 && (limit - value) < ap->allowed)
1186 			ap->allowed = limit - value;
1187 		/* If we can't meet the target */
1188 		if (limit && limit < (value + (s64)ap->target)) {
1189 			/* If no min_target specified or we don't meet
1190 			 * min_target, return -EDQUOT */
1191 			if (!ap->min_target || ap->min_target > ap->allowed) {
1192 				if (!test_and_set_bit(QDF_QMSG_QUIET,
1193 						      &qd->qd_flags)) {
1194 					print_message(qd, "exceeded");
1195 					quota_send_warning(qd->qd_id,
1196 							   sdp->sd_vfs->s_dev,
1197 							   QUOTA_NL_BHARDWARN);
1198 				}
1199 				error = -EDQUOT;
1200 				break;
1201 			}
1202 		} else if (warn && warn < value &&
1203 			   time_after_eq(jiffies, qd->qd_last_warn +
1204 					 gfs2_tune_get(sdp, gt_quota_warn_period)
1205 					 * HZ)) {
1206 			quota_send_warning(qd->qd_id,
1207 					   sdp->sd_vfs->s_dev, QUOTA_NL_BSOFTWARN);
1208 			error = print_message(qd, "warning");
1209 			qd->qd_last_warn = jiffies;
1210 		}
1211 	}
1212 	return error;
1213 }
1214 
1215 void gfs2_quota_change(struct gfs2_inode *ip, s64 change,
1216 		       kuid_t uid, kgid_t gid)
1217 {
1218 	struct gfs2_quota_data *qd;
1219 	unsigned int x;
1220 
1221 	if (gfs2_assert_warn(GFS2_SB(&ip->i_inode), change))
1222 		return;
1223 	if (ip->i_diskflags & GFS2_DIF_SYSTEM)
1224 		return;
1225 
1226 	for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1227 		qd = ip->i_res->rs_qa_qd[x];
1228 
1229 		if (qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1230 		    qid_eq(qd->qd_id, make_kqid_gid(gid))) {
1231 			do_qc(qd, change);
1232 		}
1233 	}
1234 }
1235 
1236 int gfs2_quota_sync(struct super_block *sb, int type)
1237 {
1238 	struct gfs2_sbd *sdp = sb->s_fs_info;
1239 	struct gfs2_quota_data **qda;
1240 	unsigned int max_qd = PAGE_SIZE/sizeof(struct gfs2_holder);
1241 	unsigned int num_qd;
1242 	unsigned int x;
1243 	int error = 0;
1244 
1245 	qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
1246 	if (!qda)
1247 		return -ENOMEM;
1248 
1249 	mutex_lock(&sdp->sd_quota_sync_mutex);
1250 	sdp->sd_quota_sync_gen++;
1251 
1252 	do {
1253 		num_qd = 0;
1254 
1255 		for (;;) {
1256 			error = qd_fish(sdp, qda + num_qd);
1257 			if (error || !qda[num_qd])
1258 				break;
1259 			if (++num_qd == max_qd)
1260 				break;
1261 		}
1262 
1263 		if (num_qd) {
1264 			if (!error)
1265 				error = do_sync(num_qd, qda);
1266 			if (!error)
1267 				for (x = 0; x < num_qd; x++)
1268 					qda[x]->qd_sync_gen =
1269 						sdp->sd_quota_sync_gen;
1270 
1271 			for (x = 0; x < num_qd; x++)
1272 				qd_unlock(qda[x]);
1273 		}
1274 	} while (!error && num_qd == max_qd);
1275 
1276 	mutex_unlock(&sdp->sd_quota_sync_mutex);
1277 	kfree(qda);
1278 
1279 	return error;
1280 }
1281 
1282 int gfs2_quota_refresh(struct gfs2_sbd *sdp, struct kqid qid)
1283 {
1284 	struct gfs2_quota_data *qd;
1285 	struct gfs2_holder q_gh;
1286 	int error;
1287 
1288 	error = qd_get(sdp, qid, &qd);
1289 	if (error)
1290 		return error;
1291 
1292 	error = do_glock(qd, FORCE, &q_gh);
1293 	if (!error)
1294 		gfs2_glock_dq_uninit(&q_gh);
1295 
1296 	qd_put(qd);
1297 	return error;
1298 }
1299 
1300 int gfs2_quota_init(struct gfs2_sbd *sdp)
1301 {
1302 	struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
1303 	u64 size = i_size_read(sdp->sd_qc_inode);
1304 	unsigned int blocks = size >> sdp->sd_sb.sb_bsize_shift;
1305 	unsigned int x, slot = 0;
1306 	unsigned int found = 0;
1307 	unsigned int hash;
1308 	unsigned int bm_size;
1309 	u64 dblock;
1310 	u32 extlen = 0;
1311 	int error;
1312 
1313 	if (gfs2_check_internal_file_size(sdp->sd_qc_inode, 1, 64 << 20))
1314 		return -EIO;
1315 
1316 	sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
1317 	bm_size = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * sizeof(unsigned long));
1318 	bm_size *= sizeof(unsigned long);
1319 	error = -ENOMEM;
1320 	sdp->sd_quota_bitmap = kzalloc(bm_size, GFP_NOFS | __GFP_NOWARN);
1321 	if (sdp->sd_quota_bitmap == NULL)
1322 		sdp->sd_quota_bitmap = __vmalloc(bm_size, GFP_NOFS |
1323 						 __GFP_ZERO, PAGE_KERNEL);
1324 	if (!sdp->sd_quota_bitmap)
1325 		return error;
1326 
1327 	for (x = 0; x < blocks; x++) {
1328 		struct buffer_head *bh;
1329 		const struct gfs2_quota_change *qc;
1330 		unsigned int y;
1331 
1332 		if (!extlen) {
1333 			int new = 0;
1334 			error = gfs2_extent_map(&ip->i_inode, x, &new, &dblock, &extlen);
1335 			if (error)
1336 				goto fail;
1337 		}
1338 		error = -EIO;
1339 		bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
1340 		if (!bh)
1341 			goto fail;
1342 		if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
1343 			brelse(bh);
1344 			goto fail;
1345 		}
1346 
1347 		qc = (const struct gfs2_quota_change *)(bh->b_data + sizeof(struct gfs2_meta_header));
1348 		for (y = 0; y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1349 		     y++, slot++) {
1350 			struct gfs2_quota_data *qd;
1351 			s64 qc_change = be64_to_cpu(qc->qc_change);
1352 			u32 qc_flags = be32_to_cpu(qc->qc_flags);
1353 			enum quota_type qtype = (qc_flags & GFS2_QCF_USER) ?
1354 						USRQUOTA : GRPQUOTA;
1355 			struct kqid qc_id = make_kqid(&init_user_ns, qtype,
1356 						      be32_to_cpu(qc->qc_id));
1357 			qc++;
1358 			if (!qc_change)
1359 				continue;
1360 
1361 			hash = gfs2_qd_hash(sdp, qc_id);
1362 			qd = qd_alloc(hash, sdp, qc_id);
1363 			if (qd == NULL) {
1364 				brelse(bh);
1365 				goto fail;
1366 			}
1367 
1368 			set_bit(QDF_CHANGE, &qd->qd_flags);
1369 			qd->qd_change = qc_change;
1370 			qd->qd_slot = slot;
1371 			qd->qd_slot_count = 1;
1372 
1373 			spin_lock(&qd_lock);
1374 			BUG_ON(test_and_set_bit(slot, sdp->sd_quota_bitmap));
1375 			list_add(&qd->qd_list, &sdp->sd_quota_list);
1376 			atomic_inc(&sdp->sd_quota_count);
1377 			spin_unlock(&qd_lock);
1378 
1379 			spin_lock_bucket(hash);
1380 			hlist_bl_add_head_rcu(&qd->qd_hlist, &qd_hash_table[hash]);
1381 			spin_unlock_bucket(hash);
1382 
1383 			found++;
1384 		}
1385 
1386 		brelse(bh);
1387 		dblock++;
1388 		extlen--;
1389 	}
1390 
1391 	if (found)
1392 		fs_info(sdp, "found %u quota changes\n", found);
1393 
1394 	return 0;
1395 
1396 fail:
1397 	gfs2_quota_cleanup(sdp);
1398 	return error;
1399 }
1400 
1401 void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1402 {
1403 	struct list_head *head = &sdp->sd_quota_list;
1404 	struct gfs2_quota_data *qd;
1405 
1406 	spin_lock(&qd_lock);
1407 	while (!list_empty(head)) {
1408 		qd = list_entry(head->prev, struct gfs2_quota_data, qd_list);
1409 
1410 		list_del(&qd->qd_list);
1411 
1412 		/* Also remove if this qd exists in the reclaim list */
1413 		list_lru_del(&gfs2_qd_lru, &qd->qd_lru);
1414 		atomic_dec(&sdp->sd_quota_count);
1415 		spin_unlock(&qd_lock);
1416 
1417 		spin_lock_bucket(qd->qd_hash);
1418 		hlist_bl_del_rcu(&qd->qd_hlist);
1419 		spin_unlock_bucket(qd->qd_hash);
1420 
1421 		gfs2_assert_warn(sdp, !qd->qd_change);
1422 		gfs2_assert_warn(sdp, !qd->qd_slot_count);
1423 		gfs2_assert_warn(sdp, !qd->qd_bh_count);
1424 
1425 		gfs2_glock_put(qd->qd_gl);
1426 		call_rcu(&qd->qd_rcu, gfs2_qd_dealloc);
1427 
1428 		spin_lock(&qd_lock);
1429 	}
1430 	spin_unlock(&qd_lock);
1431 
1432 	gfs2_assert_warn(sdp, !atomic_read(&sdp->sd_quota_count));
1433 
1434 	kvfree(sdp->sd_quota_bitmap);
1435 	sdp->sd_quota_bitmap = NULL;
1436 }
1437 
1438 static void quotad_error(struct gfs2_sbd *sdp, const char *msg, int error)
1439 {
1440 	if (error == 0 || error == -EROFS)
1441 		return;
1442 	if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
1443 		fs_err(sdp, "gfs2_quotad: %s error %d\n", msg, error);
1444 }
1445 
1446 static void quotad_check_timeo(struct gfs2_sbd *sdp, const char *msg,
1447 			       int (*fxn)(struct super_block *sb, int type),
1448 			       unsigned long t, unsigned long *timeo,
1449 			       unsigned int *new_timeo)
1450 {
1451 	if (t >= *timeo) {
1452 		int error = fxn(sdp->sd_vfs, 0);
1453 		quotad_error(sdp, msg, error);
1454 		*timeo = gfs2_tune_get_i(&sdp->sd_tune, new_timeo) * HZ;
1455 	} else {
1456 		*timeo -= t;
1457 	}
1458 }
1459 
1460 static void quotad_check_trunc_list(struct gfs2_sbd *sdp)
1461 {
1462 	struct gfs2_inode *ip;
1463 
1464 	while(1) {
1465 		ip = NULL;
1466 		spin_lock(&sdp->sd_trunc_lock);
1467 		if (!list_empty(&sdp->sd_trunc_list)) {
1468 			ip = list_entry(sdp->sd_trunc_list.next,
1469 					struct gfs2_inode, i_trunc_list);
1470 			list_del_init(&ip->i_trunc_list);
1471 		}
1472 		spin_unlock(&sdp->sd_trunc_lock);
1473 		if (ip == NULL)
1474 			return;
1475 		gfs2_glock_finish_truncate(ip);
1476 	}
1477 }
1478 
1479 void gfs2_wake_up_statfs(struct gfs2_sbd *sdp) {
1480 	if (!sdp->sd_statfs_force_sync) {
1481 		sdp->sd_statfs_force_sync = 1;
1482 		wake_up(&sdp->sd_quota_wait);
1483 	}
1484 }
1485 
1486 
1487 /**
1488  * gfs2_quotad - Write cached quota changes into the quota file
1489  * @sdp: Pointer to GFS2 superblock
1490  *
1491  */
1492 
1493 int gfs2_quotad(void *data)
1494 {
1495 	struct gfs2_sbd *sdp = data;
1496 	struct gfs2_tune *tune = &sdp->sd_tune;
1497 	unsigned long statfs_timeo = 0;
1498 	unsigned long quotad_timeo = 0;
1499 	unsigned long t = 0;
1500 	DEFINE_WAIT(wait);
1501 	int empty;
1502 
1503 	while (!kthread_should_stop()) {
1504 
1505 		/* Update the master statfs file */
1506 		if (sdp->sd_statfs_force_sync) {
1507 			int error = gfs2_statfs_sync(sdp->sd_vfs, 0);
1508 			quotad_error(sdp, "statfs", error);
1509 			statfs_timeo = gfs2_tune_get(sdp, gt_statfs_quantum) * HZ;
1510 		}
1511 		else
1512 			quotad_check_timeo(sdp, "statfs", gfs2_statfs_sync, t,
1513 				   	   &statfs_timeo,
1514 					   &tune->gt_statfs_quantum);
1515 
1516 		/* Update quota file */
1517 		quotad_check_timeo(sdp, "sync", gfs2_quota_sync, t,
1518 				   &quotad_timeo, &tune->gt_quota_quantum);
1519 
1520 		/* Check for & recover partially truncated inodes */
1521 		quotad_check_trunc_list(sdp);
1522 
1523 		try_to_freeze();
1524 
1525 		t = min(quotad_timeo, statfs_timeo);
1526 
1527 		prepare_to_wait(&sdp->sd_quota_wait, &wait, TASK_INTERRUPTIBLE);
1528 		spin_lock(&sdp->sd_trunc_lock);
1529 		empty = list_empty(&sdp->sd_trunc_list);
1530 		spin_unlock(&sdp->sd_trunc_lock);
1531 		if (empty && !sdp->sd_statfs_force_sync)
1532 			t -= schedule_timeout(t);
1533 		else
1534 			t = 0;
1535 		finish_wait(&sdp->sd_quota_wait, &wait);
1536 	}
1537 
1538 	return 0;
1539 }
1540 
1541 static int gfs2_quota_get_state(struct super_block *sb, struct qc_state *state)
1542 {
1543 	struct gfs2_sbd *sdp = sb->s_fs_info;
1544 
1545 	memset(state, 0, sizeof(*state));
1546 
1547 	switch (sdp->sd_args.ar_quota) {
1548 	case GFS2_QUOTA_ON:
1549 		state->s_state[USRQUOTA].flags |= QCI_LIMITS_ENFORCED;
1550 		state->s_state[GRPQUOTA].flags |= QCI_LIMITS_ENFORCED;
1551 		/*FALLTHRU*/
1552 	case GFS2_QUOTA_ACCOUNT:
1553 		state->s_state[USRQUOTA].flags |= QCI_ACCT_ENABLED |
1554 						  QCI_SYSFILE;
1555 		state->s_state[GRPQUOTA].flags |= QCI_ACCT_ENABLED |
1556 						  QCI_SYSFILE;
1557 		break;
1558 	case GFS2_QUOTA_OFF:
1559 		break;
1560 	}
1561 	if (sdp->sd_quota_inode) {
1562 		state->s_state[USRQUOTA].ino =
1563 					GFS2_I(sdp->sd_quota_inode)->i_no_addr;
1564 		state->s_state[USRQUOTA].blocks = sdp->sd_quota_inode->i_blocks;
1565 	}
1566 	state->s_state[USRQUOTA].nextents = 1;	/* unsupported */
1567 	state->s_state[GRPQUOTA] = state->s_state[USRQUOTA];
1568 	state->s_incoredqs = list_lru_count(&gfs2_qd_lru);
1569 	return 0;
1570 }
1571 
1572 static int gfs2_get_dqblk(struct super_block *sb, struct kqid qid,
1573 			  struct qc_dqblk *fdq)
1574 {
1575 	struct gfs2_sbd *sdp = sb->s_fs_info;
1576 	struct gfs2_quota_lvb *qlvb;
1577 	struct gfs2_quota_data *qd;
1578 	struct gfs2_holder q_gh;
1579 	int error;
1580 
1581 	memset(fdq, 0, sizeof(*fdq));
1582 
1583 	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1584 		return -ESRCH; /* Crazy XFS error code */
1585 
1586 	if ((qid.type != USRQUOTA) &&
1587 	    (qid.type != GRPQUOTA))
1588 		return -EINVAL;
1589 
1590 	error = qd_get(sdp, qid, &qd);
1591 	if (error)
1592 		return error;
1593 	error = do_glock(qd, FORCE, &q_gh);
1594 	if (error)
1595 		goto out;
1596 
1597 	qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1598 	fdq->d_spc_hardlimit = be64_to_cpu(qlvb->qb_limit) << sdp->sd_sb.sb_bsize_shift;
1599 	fdq->d_spc_softlimit = be64_to_cpu(qlvb->qb_warn) << sdp->sd_sb.sb_bsize_shift;
1600 	fdq->d_space = be64_to_cpu(qlvb->qb_value) << sdp->sd_sb.sb_bsize_shift;
1601 
1602 	gfs2_glock_dq_uninit(&q_gh);
1603 out:
1604 	qd_put(qd);
1605 	return error;
1606 }
1607 
1608 /* GFS2 only supports a subset of the XFS fields */
1609 #define GFS2_FIELDMASK (QC_SPC_SOFT|QC_SPC_HARD|QC_SPACE)
1610 
1611 static int gfs2_set_dqblk(struct super_block *sb, struct kqid qid,
1612 			  struct qc_dqblk *fdq)
1613 {
1614 	struct gfs2_sbd *sdp = sb->s_fs_info;
1615 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1616 	struct gfs2_quota_data *qd;
1617 	struct gfs2_holder q_gh, i_gh;
1618 	unsigned int data_blocks, ind_blocks;
1619 	unsigned int blocks = 0;
1620 	int alloc_required;
1621 	loff_t offset;
1622 	int error;
1623 
1624 	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1625 		return -ESRCH; /* Crazy XFS error code */
1626 
1627 	if ((qid.type != USRQUOTA) &&
1628 	    (qid.type != GRPQUOTA))
1629 		return -EINVAL;
1630 
1631 	if (fdq->d_fieldmask & ~GFS2_FIELDMASK)
1632 		return -EINVAL;
1633 
1634 	error = qd_get(sdp, qid, &qd);
1635 	if (error)
1636 		return error;
1637 
1638 	error = gfs2_rs_alloc(ip);
1639 	if (error)
1640 		goto out_put;
1641 
1642 	mutex_lock(&ip->i_inode.i_mutex);
1643 	error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE, 0, &q_gh);
1644 	if (error)
1645 		goto out_unlockput;
1646 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1647 	if (error)
1648 		goto out_q;
1649 
1650 	/* Check for existing entry, if none then alloc new blocks */
1651 	error = update_qd(sdp, qd);
1652 	if (error)
1653 		goto out_i;
1654 
1655 	/* If nothing has changed, this is a no-op */
1656 	if ((fdq->d_fieldmask & QC_SPC_SOFT) &&
1657 	    ((fdq->d_spc_softlimit >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_warn)))
1658 		fdq->d_fieldmask ^= QC_SPC_SOFT;
1659 
1660 	if ((fdq->d_fieldmask & QC_SPC_HARD) &&
1661 	    ((fdq->d_spc_hardlimit >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_limit)))
1662 		fdq->d_fieldmask ^= QC_SPC_HARD;
1663 
1664 	if ((fdq->d_fieldmask & QC_SPACE) &&
1665 	    ((fdq->d_space >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_value)))
1666 		fdq->d_fieldmask ^= QC_SPACE;
1667 
1668 	if (fdq->d_fieldmask == 0)
1669 		goto out_i;
1670 
1671 	offset = qd2offset(qd);
1672 	alloc_required = gfs2_write_alloc_required(ip, offset, sizeof(struct gfs2_quota));
1673 	if (gfs2_is_stuffed(ip))
1674 		alloc_required = 1;
1675 	if (alloc_required) {
1676 		struct gfs2_alloc_parms ap = { .aflags = 0, };
1677 		gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
1678 				       &data_blocks, &ind_blocks);
1679 		blocks = 1 + data_blocks + ind_blocks;
1680 		ap.target = blocks;
1681 		error = gfs2_inplace_reserve(ip, &ap);
1682 		if (error)
1683 			goto out_i;
1684 		blocks += gfs2_rg_blocks(ip, blocks);
1685 	}
1686 
1687 	/* Some quotas span block boundaries and can update two blocks,
1688 	   adding an extra block to the transaction to handle such quotas */
1689 	error = gfs2_trans_begin(sdp, blocks + RES_DINODE + 2, 0);
1690 	if (error)
1691 		goto out_release;
1692 
1693 	/* Apply changes */
1694 	error = gfs2_adjust_quota(ip, offset, 0, qd, fdq);
1695 	if (!error)
1696 		clear_bit(QDF_QMSG_QUIET, &qd->qd_flags);
1697 
1698 	gfs2_trans_end(sdp);
1699 out_release:
1700 	if (alloc_required)
1701 		gfs2_inplace_release(ip);
1702 out_i:
1703 	gfs2_glock_dq_uninit(&i_gh);
1704 out_q:
1705 	gfs2_glock_dq_uninit(&q_gh);
1706 out_unlockput:
1707 	mutex_unlock(&ip->i_inode.i_mutex);
1708 out_put:
1709 	qd_put(qd);
1710 	return error;
1711 }
1712 
1713 const struct quotactl_ops gfs2_quotactl_ops = {
1714 	.quota_sync     = gfs2_quota_sync,
1715 	.get_state	= gfs2_quota_get_state,
1716 	.get_dqblk	= gfs2_get_dqblk,
1717 	.set_dqblk	= gfs2_set_dqblk,
1718 };
1719 
1720 void __init gfs2_quota_hash_init(void)
1721 {
1722 	unsigned i;
1723 
1724 	for(i = 0; i < GFS2_QD_HASH_SIZE; i++)
1725 		INIT_HLIST_BL_HEAD(&qd_hash_table[i]);
1726 }
1727