xref: /linux/fs/gfs2/quota.c (revision 827634added7f38b7d724cab1dccdb2b004c13c3)
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,
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_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_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_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_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_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 	mutex_unlock(&sdp->sd_quota_mutex);
653 }
654 
655 /**
656  * gfs2_adjust_quota - adjust record of current block usage
657  * @ip: The quota inode
658  * @loc: Offset of the entry in the quota file
659  * @change: The amount of usage change to record
660  * @qd: The quota data
661  * @fdq: The updated limits to record
662  *
663  * This function was mostly borrowed from gfs2_block_truncate_page which was
664  * in turn mostly borrowed from ext3
665  *
666  * Returns: 0 or -ve on error
667  */
668 
669 static int gfs2_adjust_quota(struct gfs2_inode *ip, loff_t loc,
670 			     s64 change, struct gfs2_quota_data *qd,
671 			     struct qc_dqblk *fdq)
672 {
673 	struct inode *inode = &ip->i_inode;
674 	struct gfs2_sbd *sdp = GFS2_SB(inode);
675 	struct address_space *mapping = inode->i_mapping;
676 	unsigned long index = loc >> PAGE_CACHE_SHIFT;
677 	unsigned offset = loc & (PAGE_CACHE_SIZE - 1);
678 	unsigned blocksize, iblock, pos;
679 	struct buffer_head *bh;
680 	struct page *page;
681 	void *kaddr, *ptr;
682 	struct gfs2_quota q;
683 	int err, nbytes;
684 	u64 size;
685 
686 	if (gfs2_is_stuffed(ip)) {
687 		err = gfs2_unstuff_dinode(ip, NULL);
688 		if (err)
689 			return err;
690 	}
691 
692 	memset(&q, 0, sizeof(struct gfs2_quota));
693 	err = gfs2_internal_read(ip, (char *)&q, &loc, sizeof(q));
694 	if (err < 0)
695 		return err;
696 
697 	err = -EIO;
698 	be64_add_cpu(&q.qu_value, change);
699 	qd->qd_qb.qb_value = q.qu_value;
700 	if (fdq) {
701 		if (fdq->d_fieldmask & QC_SPC_SOFT) {
702 			q.qu_warn = cpu_to_be64(fdq->d_spc_softlimit >> sdp->sd_sb.sb_bsize_shift);
703 			qd->qd_qb.qb_warn = q.qu_warn;
704 		}
705 		if (fdq->d_fieldmask & QC_SPC_HARD) {
706 			q.qu_limit = cpu_to_be64(fdq->d_spc_hardlimit >> sdp->sd_sb.sb_bsize_shift);
707 			qd->qd_qb.qb_limit = q.qu_limit;
708 		}
709 		if (fdq->d_fieldmask & QC_SPACE) {
710 			q.qu_value = cpu_to_be64(fdq->d_space >> sdp->sd_sb.sb_bsize_shift);
711 			qd->qd_qb.qb_value = q.qu_value;
712 		}
713 	}
714 
715 	/* Write the quota into the quota file on disk */
716 	ptr = &q;
717 	nbytes = sizeof(struct gfs2_quota);
718 get_a_page:
719 	page = find_or_create_page(mapping, index, GFP_NOFS);
720 	if (!page)
721 		return -ENOMEM;
722 
723 	blocksize = inode->i_sb->s_blocksize;
724 	iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
725 
726 	if (!page_has_buffers(page))
727 		create_empty_buffers(page, blocksize, 0);
728 
729 	bh = page_buffers(page);
730 	pos = blocksize;
731 	while (offset >= pos) {
732 		bh = bh->b_this_page;
733 		iblock++;
734 		pos += blocksize;
735 	}
736 
737 	if (!buffer_mapped(bh)) {
738 		gfs2_block_map(inode, iblock, bh, 1);
739 		if (!buffer_mapped(bh))
740 			goto unlock_out;
741 		/* If it's a newly allocated disk block for quota, zero it */
742 		if (buffer_new(bh))
743 			zero_user(page, pos - blocksize, bh->b_size);
744 	}
745 
746 	if (PageUptodate(page))
747 		set_buffer_uptodate(bh);
748 
749 	if (!buffer_uptodate(bh)) {
750 		ll_rw_block(READ | REQ_META, 1, &bh);
751 		wait_on_buffer(bh);
752 		if (!buffer_uptodate(bh))
753 			goto unlock_out;
754 	}
755 
756 	gfs2_trans_add_data(ip->i_gl, bh);
757 
758 	kaddr = kmap_atomic(page);
759 	if (offset + sizeof(struct gfs2_quota) > PAGE_CACHE_SIZE)
760 		nbytes = PAGE_CACHE_SIZE - offset;
761 	memcpy(kaddr + offset, ptr, nbytes);
762 	flush_dcache_page(page);
763 	kunmap_atomic(kaddr);
764 	unlock_page(page);
765 	page_cache_release(page);
766 
767 	/* If quota straddles page boundary, we need to update the rest of the
768 	 * quota at the beginning of the next page */
769 	if ((offset + sizeof(struct gfs2_quota)) > PAGE_CACHE_SIZE) {
770 		ptr = ptr + nbytes;
771 		nbytes = sizeof(struct gfs2_quota) - nbytes;
772 		offset = 0;
773 		index++;
774 		goto get_a_page;
775 	}
776 
777 	size = loc + sizeof(struct gfs2_quota);
778 	if (size > inode->i_size)
779 		i_size_write(inode, size);
780 	inode->i_mtime = inode->i_atime = CURRENT_TIME;
781 	mark_inode_dirty(inode);
782 	set_bit(QDF_REFRESH, &qd->qd_flags);
783 	return 0;
784 
785 unlock_out:
786 	unlock_page(page);
787 	page_cache_release(page);
788 	return err;
789 }
790 
791 static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
792 {
793 	struct gfs2_sbd *sdp = (*qda)->qd_gl->gl_sbd;
794 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
795 	struct gfs2_alloc_parms ap = { .aflags = 0, };
796 	unsigned int data_blocks, ind_blocks;
797 	struct gfs2_holder *ghs, i_gh;
798 	unsigned int qx, x;
799 	struct gfs2_quota_data *qd;
800 	unsigned reserved;
801 	loff_t offset;
802 	unsigned int nalloc = 0, blocks;
803 	int error;
804 
805 	error = gfs2_rs_alloc(ip);
806 	if (error)
807 		return error;
808 
809 	gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
810 			      &data_blocks, &ind_blocks);
811 
812 	ghs = kcalloc(num_qd, sizeof(struct gfs2_holder), GFP_NOFS);
813 	if (!ghs)
814 		return -ENOMEM;
815 
816 	sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL);
817 	mutex_lock(&ip->i_inode.i_mutex);
818 	for (qx = 0; qx < num_qd; qx++) {
819 		error = gfs2_glock_nq_init(qda[qx]->qd_gl, LM_ST_EXCLUSIVE,
820 					   GL_NOCACHE, &ghs[qx]);
821 		if (error)
822 			goto out;
823 	}
824 
825 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
826 	if (error)
827 		goto out;
828 
829 	for (x = 0; x < num_qd; x++) {
830 		offset = qd2offset(qda[x]);
831 		if (gfs2_write_alloc_required(ip, offset,
832 					      sizeof(struct gfs2_quota)))
833 			nalloc++;
834 	}
835 
836 	/*
837 	 * 1 blk for unstuffing inode if stuffed. We add this extra
838 	 * block to the reservation unconditionally. If the inode
839 	 * doesn't need unstuffing, the block will be released to the
840 	 * rgrp since it won't be allocated during the transaction
841 	 */
842 	/* +3 in the end for unstuffing block, inode size update block
843 	 * and another block in case quota straddles page boundary and
844 	 * two blocks need to be updated instead of 1 */
845 	blocks = num_qd * data_blocks + RES_DINODE + num_qd + 3;
846 
847 	reserved = 1 + (nalloc * (data_blocks + ind_blocks));
848 	ap.target = reserved;
849 	error = gfs2_inplace_reserve(ip, &ap);
850 	if (error)
851 		goto out_alloc;
852 
853 	if (nalloc)
854 		blocks += gfs2_rg_blocks(ip, reserved) + nalloc * ind_blocks + RES_STATFS;
855 
856 	error = gfs2_trans_begin(sdp, blocks, 0);
857 	if (error)
858 		goto out_ipres;
859 
860 	for (x = 0; x < num_qd; x++) {
861 		qd = qda[x];
862 		offset = qd2offset(qd);
863 		error = gfs2_adjust_quota(ip, offset, qd->qd_change_sync, qd, NULL);
864 		if (error)
865 			goto out_end_trans;
866 
867 		do_qc(qd, -qd->qd_change_sync);
868 		set_bit(QDF_REFRESH, &qd->qd_flags);
869 	}
870 
871 	error = 0;
872 
873 out_end_trans:
874 	gfs2_trans_end(sdp);
875 out_ipres:
876 	gfs2_inplace_release(ip);
877 out_alloc:
878 	gfs2_glock_dq_uninit(&i_gh);
879 out:
880 	while (qx--)
881 		gfs2_glock_dq_uninit(&ghs[qx]);
882 	mutex_unlock(&ip->i_inode.i_mutex);
883 	kfree(ghs);
884 	gfs2_log_flush(ip->i_gl->gl_sbd, ip->i_gl, NORMAL_FLUSH);
885 	return error;
886 }
887 
888 static int update_qd(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd)
889 {
890 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
891 	struct gfs2_quota q;
892 	struct gfs2_quota_lvb *qlvb;
893 	loff_t pos;
894 	int error;
895 
896 	memset(&q, 0, sizeof(struct gfs2_quota));
897 	pos = qd2offset(qd);
898 	error = gfs2_internal_read(ip, (char *)&q, &pos, sizeof(q));
899 	if (error < 0)
900 		return error;
901 
902 	qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
903 	qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC);
904 	qlvb->__pad = 0;
905 	qlvb->qb_limit = q.qu_limit;
906 	qlvb->qb_warn = q.qu_warn;
907 	qlvb->qb_value = q.qu_value;
908 	qd->qd_qb = *qlvb;
909 
910 	return 0;
911 }
912 
913 static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
914 		    struct gfs2_holder *q_gh)
915 {
916 	struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
917 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
918 	struct gfs2_holder i_gh;
919 	int error;
920 
921 restart:
922 	error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
923 	if (error)
924 		return error;
925 
926 	if (test_and_clear_bit(QDF_REFRESH, &qd->qd_flags))
927 		force_refresh = FORCE;
928 
929 	qd->qd_qb = *(struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
930 
931 	if (force_refresh || qd->qd_qb.qb_magic != cpu_to_be32(GFS2_MAGIC)) {
932 		gfs2_glock_dq_uninit(q_gh);
933 		error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE,
934 					   GL_NOCACHE, q_gh);
935 		if (error)
936 			return error;
937 
938 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
939 		if (error)
940 			goto fail;
941 
942 		error = update_qd(sdp, qd);
943 		if (error)
944 			goto fail_gunlock;
945 
946 		gfs2_glock_dq_uninit(&i_gh);
947 		gfs2_glock_dq_uninit(q_gh);
948 		force_refresh = 0;
949 		goto restart;
950 	}
951 
952 	return 0;
953 
954 fail_gunlock:
955 	gfs2_glock_dq_uninit(&i_gh);
956 fail:
957 	gfs2_glock_dq_uninit(q_gh);
958 	return error;
959 }
960 
961 int gfs2_quota_lock(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
962 {
963 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
964 	struct gfs2_quota_data *qd;
965 	unsigned int x;
966 	int error = 0;
967 
968 	error = gfs2_quota_hold(ip, uid, gid);
969 	if (error)
970 		return error;
971 
972 	if (capable(CAP_SYS_RESOURCE) ||
973 	    sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
974 		return 0;
975 
976 	sort(ip->i_res->rs_qa_qd, ip->i_res->rs_qa_qd_num,
977 	     sizeof(struct gfs2_quota_data *), sort_qd, NULL);
978 
979 	for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
980 		qd = ip->i_res->rs_qa_qd[x];
981 		error = do_glock(qd, NO_FORCE, &ip->i_res->rs_qa_qd_ghs[x]);
982 		if (error)
983 			break;
984 	}
985 
986 	if (!error)
987 		set_bit(GIF_QD_LOCKED, &ip->i_flags);
988 	else {
989 		while (x--)
990 			gfs2_glock_dq_uninit(&ip->i_res->rs_qa_qd_ghs[x]);
991 		gfs2_quota_unhold(ip);
992 	}
993 
994 	return error;
995 }
996 
997 static int need_sync(struct gfs2_quota_data *qd)
998 {
999 	struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
1000 	struct gfs2_tune *gt = &sdp->sd_tune;
1001 	s64 value;
1002 	unsigned int num, den;
1003 	int do_sync = 1;
1004 
1005 	if (!qd->qd_qb.qb_limit)
1006 		return 0;
1007 
1008 	spin_lock(&qd_lock);
1009 	value = qd->qd_change;
1010 	spin_unlock(&qd_lock);
1011 
1012 	spin_lock(&gt->gt_spin);
1013 	num = gt->gt_quota_scale_num;
1014 	den = gt->gt_quota_scale_den;
1015 	spin_unlock(&gt->gt_spin);
1016 
1017 	if (value < 0)
1018 		do_sync = 0;
1019 	else if ((s64)be64_to_cpu(qd->qd_qb.qb_value) >=
1020 		 (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1021 		do_sync = 0;
1022 	else {
1023 		value *= gfs2_jindex_size(sdp) * num;
1024 		value = div_s64(value, den);
1025 		value += (s64)be64_to_cpu(qd->qd_qb.qb_value);
1026 		if (value < (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1027 			do_sync = 0;
1028 	}
1029 
1030 	return do_sync;
1031 }
1032 
1033 void gfs2_quota_unlock(struct gfs2_inode *ip)
1034 {
1035 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1036 	struct gfs2_quota_data *qda[4];
1037 	unsigned int count = 0;
1038 	unsigned int x;
1039 	int found;
1040 
1041 	if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
1042 		goto out;
1043 
1044 	for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1045 		struct gfs2_quota_data *qd;
1046 		int sync;
1047 
1048 		qd = ip->i_res->rs_qa_qd[x];
1049 		sync = need_sync(qd);
1050 
1051 		gfs2_glock_dq_uninit(&ip->i_res->rs_qa_qd_ghs[x]);
1052 		if (!sync)
1053 			continue;
1054 
1055 		spin_lock(&qd_lock);
1056 		found = qd_check_sync(sdp, qd, NULL);
1057 		spin_unlock(&qd_lock);
1058 
1059 		if (!found)
1060 			continue;
1061 
1062 		gfs2_assert_warn(sdp, qd->qd_change_sync);
1063 		if (bh_get(qd)) {
1064 			clear_bit(QDF_LOCKED, &qd->qd_flags);
1065 			slot_put(qd);
1066 			qd_put(qd);
1067 			continue;
1068 		}
1069 
1070 		qda[count++] = qd;
1071 	}
1072 
1073 	if (count) {
1074 		do_sync(count, qda);
1075 		for (x = 0; x < count; x++)
1076 			qd_unlock(qda[x]);
1077 	}
1078 
1079 out:
1080 	gfs2_quota_unhold(ip);
1081 }
1082 
1083 #define MAX_LINE 256
1084 
1085 static int print_message(struct gfs2_quota_data *qd, char *type)
1086 {
1087 	struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
1088 
1089 	fs_info(sdp, "quota %s for %s %u\n",
1090 		type,
1091 		(qd->qd_id.type == USRQUOTA) ? "user" : "group",
1092 		from_kqid(&init_user_ns, qd->qd_id));
1093 
1094 	return 0;
1095 }
1096 
1097 /**
1098  * gfs2_quota_check - check if allocating new blocks will exceed quota
1099  * @ip:  The inode for which this check is being performed
1100  * @uid: The uid to check against
1101  * @gid: The gid to check against
1102  * @ap:  The allocation parameters. ap->target contains the requested
1103  *       blocks. ap->min_target, if set, contains the minimum blks
1104  *       requested.
1105  *
1106  * Returns: 0 on success.
1107  *                  min_req = ap->min_target ? ap->min_target : ap->target;
1108  *                  quota must allow atleast min_req blks for success and
1109  *                  ap->allowed is set to the number of blocks allowed
1110  *
1111  *          -EDQUOT otherwise, quota violation. ap->allowed is set to number
1112  *                  of blocks available.
1113  */
1114 int gfs2_quota_check(struct gfs2_inode *ip, kuid_t uid, kgid_t gid,
1115 		     struct gfs2_alloc_parms *ap)
1116 {
1117 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1118 	struct gfs2_quota_data *qd;
1119 	s64 value, warn, limit;
1120 	unsigned int x;
1121 	int error = 0;
1122 
1123 	ap->allowed = UINT_MAX; /* Assume we are permitted a whole lot */
1124 	if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
1125 		return 0;
1126 
1127         if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
1128                 return 0;
1129 
1130 	for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1131 		qd = ip->i_res->rs_qa_qd[x];
1132 
1133 		if (!(qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1134 		      qid_eq(qd->qd_id, make_kqid_gid(gid))))
1135 			continue;
1136 
1137 		warn = (s64)be64_to_cpu(qd->qd_qb.qb_warn);
1138 		limit = (s64)be64_to_cpu(qd->qd_qb.qb_limit);
1139 		value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
1140 		spin_lock(&qd_lock);
1141 		value += qd->qd_change;
1142 		spin_unlock(&qd_lock);
1143 
1144 		if (limit > 0 && (limit - value) < ap->allowed)
1145 			ap->allowed = limit - value;
1146 		/* If we can't meet the target */
1147 		if (limit && limit < (value + (s64)ap->target)) {
1148 			/* If no min_target specified or we don't meet
1149 			 * min_target, return -EDQUOT */
1150 			if (!ap->min_target || ap->min_target > ap->allowed) {
1151 				print_message(qd, "exceeded");
1152 				quota_send_warning(qd->qd_id,
1153 						   sdp->sd_vfs->s_dev,
1154 						   QUOTA_NL_BHARDWARN);
1155 				error = -EDQUOT;
1156 				break;
1157 			}
1158 		} else if (warn && warn < value &&
1159 			   time_after_eq(jiffies, qd->qd_last_warn +
1160 					 gfs2_tune_get(sdp, gt_quota_warn_period)
1161 					 * HZ)) {
1162 			quota_send_warning(qd->qd_id,
1163 					   sdp->sd_vfs->s_dev, QUOTA_NL_BSOFTWARN);
1164 			error = print_message(qd, "warning");
1165 			qd->qd_last_warn = jiffies;
1166 		}
1167 	}
1168 	return error;
1169 }
1170 
1171 void gfs2_quota_change(struct gfs2_inode *ip, s64 change,
1172 		       kuid_t uid, kgid_t gid)
1173 {
1174 	struct gfs2_quota_data *qd;
1175 	unsigned int x;
1176 
1177 	if (gfs2_assert_warn(GFS2_SB(&ip->i_inode), change))
1178 		return;
1179 	if (ip->i_diskflags & GFS2_DIF_SYSTEM)
1180 		return;
1181 
1182 	for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1183 		qd = ip->i_res->rs_qa_qd[x];
1184 
1185 		if (qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1186 		    qid_eq(qd->qd_id, make_kqid_gid(gid))) {
1187 			do_qc(qd, change);
1188 		}
1189 	}
1190 }
1191 
1192 int gfs2_quota_sync(struct super_block *sb, int type)
1193 {
1194 	struct gfs2_sbd *sdp = sb->s_fs_info;
1195 	struct gfs2_quota_data **qda;
1196 	unsigned int max_qd = PAGE_SIZE/sizeof(struct gfs2_holder);
1197 	unsigned int num_qd;
1198 	unsigned int x;
1199 	int error = 0;
1200 
1201 	qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
1202 	if (!qda)
1203 		return -ENOMEM;
1204 
1205 	mutex_lock(&sdp->sd_quota_sync_mutex);
1206 	sdp->sd_quota_sync_gen++;
1207 
1208 	do {
1209 		num_qd = 0;
1210 
1211 		for (;;) {
1212 			error = qd_fish(sdp, qda + num_qd);
1213 			if (error || !qda[num_qd])
1214 				break;
1215 			if (++num_qd == max_qd)
1216 				break;
1217 		}
1218 
1219 		if (num_qd) {
1220 			if (!error)
1221 				error = do_sync(num_qd, qda);
1222 			if (!error)
1223 				for (x = 0; x < num_qd; x++)
1224 					qda[x]->qd_sync_gen =
1225 						sdp->sd_quota_sync_gen;
1226 
1227 			for (x = 0; x < num_qd; x++)
1228 				qd_unlock(qda[x]);
1229 		}
1230 	} while (!error && num_qd == max_qd);
1231 
1232 	mutex_unlock(&sdp->sd_quota_sync_mutex);
1233 	kfree(qda);
1234 
1235 	return error;
1236 }
1237 
1238 int gfs2_quota_refresh(struct gfs2_sbd *sdp, struct kqid qid)
1239 {
1240 	struct gfs2_quota_data *qd;
1241 	struct gfs2_holder q_gh;
1242 	int error;
1243 
1244 	error = qd_get(sdp, qid, &qd);
1245 	if (error)
1246 		return error;
1247 
1248 	error = do_glock(qd, FORCE, &q_gh);
1249 	if (!error)
1250 		gfs2_glock_dq_uninit(&q_gh);
1251 
1252 	qd_put(qd);
1253 	return error;
1254 }
1255 
1256 int gfs2_quota_init(struct gfs2_sbd *sdp)
1257 {
1258 	struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
1259 	u64 size = i_size_read(sdp->sd_qc_inode);
1260 	unsigned int blocks = size >> sdp->sd_sb.sb_bsize_shift;
1261 	unsigned int x, slot = 0;
1262 	unsigned int found = 0;
1263 	unsigned int hash;
1264 	unsigned int bm_size;
1265 	u64 dblock;
1266 	u32 extlen = 0;
1267 	int error;
1268 
1269 	if (gfs2_check_internal_file_size(sdp->sd_qc_inode, 1, 64 << 20))
1270 		return -EIO;
1271 
1272 	sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
1273 	bm_size = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * sizeof(unsigned long));
1274 	bm_size *= sizeof(unsigned long);
1275 	error = -ENOMEM;
1276 	sdp->sd_quota_bitmap = kzalloc(bm_size, GFP_NOFS | __GFP_NOWARN);
1277 	if (sdp->sd_quota_bitmap == NULL)
1278 		sdp->sd_quota_bitmap = __vmalloc(bm_size, GFP_NOFS |
1279 						 __GFP_ZERO, PAGE_KERNEL);
1280 	if (!sdp->sd_quota_bitmap)
1281 		return error;
1282 
1283 	for (x = 0; x < blocks; x++) {
1284 		struct buffer_head *bh;
1285 		const struct gfs2_quota_change *qc;
1286 		unsigned int y;
1287 
1288 		if (!extlen) {
1289 			int new = 0;
1290 			error = gfs2_extent_map(&ip->i_inode, x, &new, &dblock, &extlen);
1291 			if (error)
1292 				goto fail;
1293 		}
1294 		error = -EIO;
1295 		bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
1296 		if (!bh)
1297 			goto fail;
1298 		if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
1299 			brelse(bh);
1300 			goto fail;
1301 		}
1302 
1303 		qc = (const struct gfs2_quota_change *)(bh->b_data + sizeof(struct gfs2_meta_header));
1304 		for (y = 0; y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1305 		     y++, slot++) {
1306 			struct gfs2_quota_data *qd;
1307 			s64 qc_change = be64_to_cpu(qc->qc_change);
1308 			u32 qc_flags = be32_to_cpu(qc->qc_flags);
1309 			enum quota_type qtype = (qc_flags & GFS2_QCF_USER) ?
1310 						USRQUOTA : GRPQUOTA;
1311 			struct kqid qc_id = make_kqid(&init_user_ns, qtype,
1312 						      be32_to_cpu(qc->qc_id));
1313 			qc++;
1314 			if (!qc_change)
1315 				continue;
1316 
1317 			hash = gfs2_qd_hash(sdp, qc_id);
1318 			qd = qd_alloc(hash, sdp, qc_id);
1319 			if (qd == NULL) {
1320 				brelse(bh);
1321 				goto fail;
1322 			}
1323 
1324 			set_bit(QDF_CHANGE, &qd->qd_flags);
1325 			qd->qd_change = qc_change;
1326 			qd->qd_slot = slot;
1327 			qd->qd_slot_count = 1;
1328 
1329 			spin_lock(&qd_lock);
1330 			BUG_ON(test_and_set_bit(slot, sdp->sd_quota_bitmap));
1331 			list_add(&qd->qd_list, &sdp->sd_quota_list);
1332 			atomic_inc(&sdp->sd_quota_count);
1333 			spin_unlock(&qd_lock);
1334 
1335 			spin_lock_bucket(hash);
1336 			hlist_bl_add_head_rcu(&qd->qd_hlist, &qd_hash_table[hash]);
1337 			spin_unlock_bucket(hash);
1338 
1339 			found++;
1340 		}
1341 
1342 		brelse(bh);
1343 		dblock++;
1344 		extlen--;
1345 	}
1346 
1347 	if (found)
1348 		fs_info(sdp, "found %u quota changes\n", found);
1349 
1350 	return 0;
1351 
1352 fail:
1353 	gfs2_quota_cleanup(sdp);
1354 	return error;
1355 }
1356 
1357 void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1358 {
1359 	struct list_head *head = &sdp->sd_quota_list;
1360 	struct gfs2_quota_data *qd;
1361 
1362 	spin_lock(&qd_lock);
1363 	while (!list_empty(head)) {
1364 		qd = list_entry(head->prev, struct gfs2_quota_data, qd_list);
1365 
1366 		list_del(&qd->qd_list);
1367 
1368 		/* Also remove if this qd exists in the reclaim list */
1369 		list_lru_del(&gfs2_qd_lru, &qd->qd_lru);
1370 		atomic_dec(&sdp->sd_quota_count);
1371 		spin_unlock(&qd_lock);
1372 
1373 		spin_lock_bucket(qd->qd_hash);
1374 		hlist_bl_del_rcu(&qd->qd_hlist);
1375 		spin_unlock_bucket(qd->qd_hash);
1376 
1377 		gfs2_assert_warn(sdp, !qd->qd_change);
1378 		gfs2_assert_warn(sdp, !qd->qd_slot_count);
1379 		gfs2_assert_warn(sdp, !qd->qd_bh_count);
1380 
1381 		gfs2_glock_put(qd->qd_gl);
1382 		call_rcu(&qd->qd_rcu, gfs2_qd_dealloc);
1383 
1384 		spin_lock(&qd_lock);
1385 	}
1386 	spin_unlock(&qd_lock);
1387 
1388 	gfs2_assert_warn(sdp, !atomic_read(&sdp->sd_quota_count));
1389 
1390 	kvfree(sdp->sd_quota_bitmap);
1391 	sdp->sd_quota_bitmap = NULL;
1392 }
1393 
1394 static void quotad_error(struct gfs2_sbd *sdp, const char *msg, int error)
1395 {
1396 	if (error == 0 || error == -EROFS)
1397 		return;
1398 	if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
1399 		fs_err(sdp, "gfs2_quotad: %s error %d\n", msg, error);
1400 }
1401 
1402 static void quotad_check_timeo(struct gfs2_sbd *sdp, const char *msg,
1403 			       int (*fxn)(struct super_block *sb, int type),
1404 			       unsigned long t, unsigned long *timeo,
1405 			       unsigned int *new_timeo)
1406 {
1407 	if (t >= *timeo) {
1408 		int error = fxn(sdp->sd_vfs, 0);
1409 		quotad_error(sdp, msg, error);
1410 		*timeo = gfs2_tune_get_i(&sdp->sd_tune, new_timeo) * HZ;
1411 	} else {
1412 		*timeo -= t;
1413 	}
1414 }
1415 
1416 static void quotad_check_trunc_list(struct gfs2_sbd *sdp)
1417 {
1418 	struct gfs2_inode *ip;
1419 
1420 	while(1) {
1421 		ip = NULL;
1422 		spin_lock(&sdp->sd_trunc_lock);
1423 		if (!list_empty(&sdp->sd_trunc_list)) {
1424 			ip = list_entry(sdp->sd_trunc_list.next,
1425 					struct gfs2_inode, i_trunc_list);
1426 			list_del_init(&ip->i_trunc_list);
1427 		}
1428 		spin_unlock(&sdp->sd_trunc_lock);
1429 		if (ip == NULL)
1430 			return;
1431 		gfs2_glock_finish_truncate(ip);
1432 	}
1433 }
1434 
1435 void gfs2_wake_up_statfs(struct gfs2_sbd *sdp) {
1436 	if (!sdp->sd_statfs_force_sync) {
1437 		sdp->sd_statfs_force_sync = 1;
1438 		wake_up(&sdp->sd_quota_wait);
1439 	}
1440 }
1441 
1442 
1443 /**
1444  * gfs2_quotad - Write cached quota changes into the quota file
1445  * @sdp: Pointer to GFS2 superblock
1446  *
1447  */
1448 
1449 int gfs2_quotad(void *data)
1450 {
1451 	struct gfs2_sbd *sdp = data;
1452 	struct gfs2_tune *tune = &sdp->sd_tune;
1453 	unsigned long statfs_timeo = 0;
1454 	unsigned long quotad_timeo = 0;
1455 	unsigned long t = 0;
1456 	DEFINE_WAIT(wait);
1457 	int empty;
1458 
1459 	while (!kthread_should_stop()) {
1460 
1461 		/* Update the master statfs file */
1462 		if (sdp->sd_statfs_force_sync) {
1463 			int error = gfs2_statfs_sync(sdp->sd_vfs, 0);
1464 			quotad_error(sdp, "statfs", error);
1465 			statfs_timeo = gfs2_tune_get(sdp, gt_statfs_quantum) * HZ;
1466 		}
1467 		else
1468 			quotad_check_timeo(sdp, "statfs", gfs2_statfs_sync, t,
1469 				   	   &statfs_timeo,
1470 					   &tune->gt_statfs_quantum);
1471 
1472 		/* Update quota file */
1473 		quotad_check_timeo(sdp, "sync", gfs2_quota_sync, t,
1474 				   &quotad_timeo, &tune->gt_quota_quantum);
1475 
1476 		/* Check for & recover partially truncated inodes */
1477 		quotad_check_trunc_list(sdp);
1478 
1479 		try_to_freeze();
1480 
1481 		t = min(quotad_timeo, statfs_timeo);
1482 
1483 		prepare_to_wait(&sdp->sd_quota_wait, &wait, TASK_INTERRUPTIBLE);
1484 		spin_lock(&sdp->sd_trunc_lock);
1485 		empty = list_empty(&sdp->sd_trunc_list);
1486 		spin_unlock(&sdp->sd_trunc_lock);
1487 		if (empty && !sdp->sd_statfs_force_sync)
1488 			t -= schedule_timeout(t);
1489 		else
1490 			t = 0;
1491 		finish_wait(&sdp->sd_quota_wait, &wait);
1492 	}
1493 
1494 	return 0;
1495 }
1496 
1497 static int gfs2_quota_get_state(struct super_block *sb, struct qc_state *state)
1498 {
1499 	struct gfs2_sbd *sdp = sb->s_fs_info;
1500 
1501 	memset(state, 0, sizeof(*state));
1502 
1503 	switch (sdp->sd_args.ar_quota) {
1504 	case GFS2_QUOTA_ON:
1505 		state->s_state[USRQUOTA].flags |= QCI_LIMITS_ENFORCED;
1506 		state->s_state[GRPQUOTA].flags |= QCI_LIMITS_ENFORCED;
1507 		/*FALLTHRU*/
1508 	case GFS2_QUOTA_ACCOUNT:
1509 		state->s_state[USRQUOTA].flags |= QCI_ACCT_ENABLED |
1510 						  QCI_SYSFILE;
1511 		state->s_state[GRPQUOTA].flags |= QCI_ACCT_ENABLED |
1512 						  QCI_SYSFILE;
1513 		break;
1514 	case GFS2_QUOTA_OFF:
1515 		break;
1516 	}
1517 	if (sdp->sd_quota_inode) {
1518 		state->s_state[USRQUOTA].ino =
1519 					GFS2_I(sdp->sd_quota_inode)->i_no_addr;
1520 		state->s_state[USRQUOTA].blocks = sdp->sd_quota_inode->i_blocks;
1521 	}
1522 	state->s_state[USRQUOTA].nextents = 1;	/* unsupported */
1523 	state->s_state[GRPQUOTA] = state->s_state[USRQUOTA];
1524 	state->s_incoredqs = list_lru_count(&gfs2_qd_lru);
1525 	return 0;
1526 }
1527 
1528 static int gfs2_get_dqblk(struct super_block *sb, struct kqid qid,
1529 			  struct qc_dqblk *fdq)
1530 {
1531 	struct gfs2_sbd *sdp = sb->s_fs_info;
1532 	struct gfs2_quota_lvb *qlvb;
1533 	struct gfs2_quota_data *qd;
1534 	struct gfs2_holder q_gh;
1535 	int error;
1536 
1537 	memset(fdq, 0, sizeof(*fdq));
1538 
1539 	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1540 		return -ESRCH; /* Crazy XFS error code */
1541 
1542 	if ((qid.type != USRQUOTA) &&
1543 	    (qid.type != GRPQUOTA))
1544 		return -EINVAL;
1545 
1546 	error = qd_get(sdp, qid, &qd);
1547 	if (error)
1548 		return error;
1549 	error = do_glock(qd, FORCE, &q_gh);
1550 	if (error)
1551 		goto out;
1552 
1553 	qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1554 	fdq->d_spc_hardlimit = be64_to_cpu(qlvb->qb_limit) << sdp->sd_sb.sb_bsize_shift;
1555 	fdq->d_spc_softlimit = be64_to_cpu(qlvb->qb_warn) << sdp->sd_sb.sb_bsize_shift;
1556 	fdq->d_space = be64_to_cpu(qlvb->qb_value) << sdp->sd_sb.sb_bsize_shift;
1557 
1558 	gfs2_glock_dq_uninit(&q_gh);
1559 out:
1560 	qd_put(qd);
1561 	return error;
1562 }
1563 
1564 /* GFS2 only supports a subset of the XFS fields */
1565 #define GFS2_FIELDMASK (QC_SPC_SOFT|QC_SPC_HARD|QC_SPACE)
1566 
1567 static int gfs2_set_dqblk(struct super_block *sb, struct kqid qid,
1568 			  struct qc_dqblk *fdq)
1569 {
1570 	struct gfs2_sbd *sdp = sb->s_fs_info;
1571 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1572 	struct gfs2_quota_data *qd;
1573 	struct gfs2_holder q_gh, i_gh;
1574 	unsigned int data_blocks, ind_blocks;
1575 	unsigned int blocks = 0;
1576 	int alloc_required;
1577 	loff_t offset;
1578 	int error;
1579 
1580 	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1581 		return -ESRCH; /* Crazy XFS error code */
1582 
1583 	if ((qid.type != USRQUOTA) &&
1584 	    (qid.type != GRPQUOTA))
1585 		return -EINVAL;
1586 
1587 	if (fdq->d_fieldmask & ~GFS2_FIELDMASK)
1588 		return -EINVAL;
1589 
1590 	error = qd_get(sdp, qid, &qd);
1591 	if (error)
1592 		return error;
1593 
1594 	error = gfs2_rs_alloc(ip);
1595 	if (error)
1596 		goto out_put;
1597 
1598 	mutex_lock(&ip->i_inode.i_mutex);
1599 	error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE, 0, &q_gh);
1600 	if (error)
1601 		goto out_unlockput;
1602 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1603 	if (error)
1604 		goto out_q;
1605 
1606 	/* Check for existing entry, if none then alloc new blocks */
1607 	error = update_qd(sdp, qd);
1608 	if (error)
1609 		goto out_i;
1610 
1611 	/* If nothing has changed, this is a no-op */
1612 	if ((fdq->d_fieldmask & QC_SPC_SOFT) &&
1613 	    ((fdq->d_spc_softlimit >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_warn)))
1614 		fdq->d_fieldmask ^= QC_SPC_SOFT;
1615 
1616 	if ((fdq->d_fieldmask & QC_SPC_HARD) &&
1617 	    ((fdq->d_spc_hardlimit >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_limit)))
1618 		fdq->d_fieldmask ^= QC_SPC_HARD;
1619 
1620 	if ((fdq->d_fieldmask & QC_SPACE) &&
1621 	    ((fdq->d_space >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_value)))
1622 		fdq->d_fieldmask ^= QC_SPACE;
1623 
1624 	if (fdq->d_fieldmask == 0)
1625 		goto out_i;
1626 
1627 	offset = qd2offset(qd);
1628 	alloc_required = gfs2_write_alloc_required(ip, offset, sizeof(struct gfs2_quota));
1629 	if (gfs2_is_stuffed(ip))
1630 		alloc_required = 1;
1631 	if (alloc_required) {
1632 		struct gfs2_alloc_parms ap = { .aflags = 0, };
1633 		gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
1634 				       &data_blocks, &ind_blocks);
1635 		blocks = 1 + data_blocks + ind_blocks;
1636 		ap.target = blocks;
1637 		error = gfs2_inplace_reserve(ip, &ap);
1638 		if (error)
1639 			goto out_i;
1640 		blocks += gfs2_rg_blocks(ip, blocks);
1641 	}
1642 
1643 	/* Some quotas span block boundaries and can update two blocks,
1644 	   adding an extra block to the transaction to handle such quotas */
1645 	error = gfs2_trans_begin(sdp, blocks + RES_DINODE + 2, 0);
1646 	if (error)
1647 		goto out_release;
1648 
1649 	/* Apply changes */
1650 	error = gfs2_adjust_quota(ip, offset, 0, qd, fdq);
1651 
1652 	gfs2_trans_end(sdp);
1653 out_release:
1654 	if (alloc_required)
1655 		gfs2_inplace_release(ip);
1656 out_i:
1657 	gfs2_glock_dq_uninit(&i_gh);
1658 out_q:
1659 	gfs2_glock_dq_uninit(&q_gh);
1660 out_unlockput:
1661 	mutex_unlock(&ip->i_inode.i_mutex);
1662 out_put:
1663 	qd_put(qd);
1664 	return error;
1665 }
1666 
1667 const struct quotactl_ops gfs2_quotactl_ops = {
1668 	.quota_sync     = gfs2_quota_sync,
1669 	.get_state	= gfs2_quota_get_state,
1670 	.get_dqblk	= gfs2_get_dqblk,
1671 	.set_dqblk	= gfs2_set_dqblk,
1672 };
1673 
1674 void __init gfs2_quota_hash_init(void)
1675 {
1676 	unsigned i;
1677 
1678 	for(i = 0; i < GFS2_QD_HASH_SIZE; i++)
1679 		INIT_HLIST_BL_HEAD(&qd_hash_table[i]);
1680 }
1681