xref: /openbmc/linux/fs/gfs2/quota.c (revision a72594ca)
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       BIT(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 = BIT(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, 0, &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 (sb_rdonly(sdp->sd_vfs))
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 /**
531  * gfs2_qa_alloc - make sure we have a quota allocations data structure,
532  *                 if necessary
533  * @ip: the inode for this reservation
534  */
535 int gfs2_qa_alloc(struct gfs2_inode *ip)
536 {
537 	int error = 0;
538 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
539 
540 	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
541 		return 0;
542 
543 	down_write(&ip->i_rw_mutex);
544 	if (ip->i_qadata == NULL) {
545 		ip->i_qadata = kmem_cache_zalloc(gfs2_qadata_cachep, GFP_NOFS);
546 		if (!ip->i_qadata)
547 			error = -ENOMEM;
548 	}
549 	up_write(&ip->i_rw_mutex);
550 	return error;
551 }
552 
553 void gfs2_qa_delete(struct gfs2_inode *ip, atomic_t *wcount)
554 {
555 	down_write(&ip->i_rw_mutex);
556 	if (ip->i_qadata && ((wcount == NULL) || (atomic_read(wcount) <= 1))) {
557 		kmem_cache_free(gfs2_qadata_cachep, ip->i_qadata);
558 		ip->i_qadata = NULL;
559 	}
560 	up_write(&ip->i_rw_mutex);
561 }
562 
563 int gfs2_quota_hold(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
564 {
565 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
566 	struct gfs2_quota_data **qd;
567 	int error;
568 
569 	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
570 		return 0;
571 
572 	if (ip->i_qadata == NULL) {
573 		error = gfs2_rsqa_alloc(ip);
574 		if (error)
575 			return error;
576 	}
577 
578 	qd = ip->i_qadata->qa_qd;
579 
580 	if (gfs2_assert_warn(sdp, !ip->i_qadata->qa_qd_num) ||
581 	    gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags)))
582 		return -EIO;
583 
584 	error = qdsb_get(sdp, make_kqid_uid(ip->i_inode.i_uid), qd);
585 	if (error)
586 		goto out;
587 	ip->i_qadata->qa_qd_num++;
588 	qd++;
589 
590 	error = qdsb_get(sdp, make_kqid_gid(ip->i_inode.i_gid), qd);
591 	if (error)
592 		goto out;
593 	ip->i_qadata->qa_qd_num++;
594 	qd++;
595 
596 	if (!uid_eq(uid, NO_UID_QUOTA_CHANGE) &&
597 	    !uid_eq(uid, ip->i_inode.i_uid)) {
598 		error = qdsb_get(sdp, make_kqid_uid(uid), qd);
599 		if (error)
600 			goto out;
601 		ip->i_qadata->qa_qd_num++;
602 		qd++;
603 	}
604 
605 	if (!gid_eq(gid, NO_GID_QUOTA_CHANGE) &&
606 	    !gid_eq(gid, ip->i_inode.i_gid)) {
607 		error = qdsb_get(sdp, make_kqid_gid(gid), qd);
608 		if (error)
609 			goto out;
610 		ip->i_qadata->qa_qd_num++;
611 		qd++;
612 	}
613 
614 out:
615 	if (error)
616 		gfs2_quota_unhold(ip);
617 	return error;
618 }
619 
620 void gfs2_quota_unhold(struct gfs2_inode *ip)
621 {
622 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
623 	u32 x;
624 
625 	if (ip->i_qadata == NULL)
626 		return;
627 	gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags));
628 
629 	for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
630 		qdsb_put(ip->i_qadata->qa_qd[x]);
631 		ip->i_qadata->qa_qd[x] = NULL;
632 	}
633 	ip->i_qadata->qa_qd_num = 0;
634 }
635 
636 static int sort_qd(const void *a, const void *b)
637 {
638 	const struct gfs2_quota_data *qd_a = *(const struct gfs2_quota_data **)a;
639 	const struct gfs2_quota_data *qd_b = *(const struct gfs2_quota_data **)b;
640 
641 	if (qid_lt(qd_a->qd_id, qd_b->qd_id))
642 		return -1;
643 	if (qid_lt(qd_b->qd_id, qd_a->qd_id))
644 		return 1;
645 	return 0;
646 }
647 
648 static void do_qc(struct gfs2_quota_data *qd, s64 change)
649 {
650 	struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
651 	struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
652 	struct gfs2_quota_change *qc = qd->qd_bh_qc;
653 	s64 x;
654 
655 	mutex_lock(&sdp->sd_quota_mutex);
656 	gfs2_trans_add_meta(ip->i_gl, qd->qd_bh);
657 
658 	if (!test_bit(QDF_CHANGE, &qd->qd_flags)) {
659 		qc->qc_change = 0;
660 		qc->qc_flags = 0;
661 		if (qd->qd_id.type == USRQUOTA)
662 			qc->qc_flags = cpu_to_be32(GFS2_QCF_USER);
663 		qc->qc_id = cpu_to_be32(from_kqid(&init_user_ns, qd->qd_id));
664 	}
665 
666 	x = be64_to_cpu(qc->qc_change) + change;
667 	qc->qc_change = cpu_to_be64(x);
668 
669 	spin_lock(&qd_lock);
670 	qd->qd_change = x;
671 	spin_unlock(&qd_lock);
672 
673 	if (!x) {
674 		gfs2_assert_warn(sdp, test_bit(QDF_CHANGE, &qd->qd_flags));
675 		clear_bit(QDF_CHANGE, &qd->qd_flags);
676 		qc->qc_flags = 0;
677 		qc->qc_id = 0;
678 		slot_put(qd);
679 		qd_put(qd);
680 	} else if (!test_and_set_bit(QDF_CHANGE, &qd->qd_flags)) {
681 		qd_hold(qd);
682 		slot_hold(qd);
683 	}
684 
685 	if (change < 0) /* Reset quiet flag if we freed some blocks */
686 		clear_bit(QDF_QMSG_QUIET, &qd->qd_flags);
687 	mutex_unlock(&sdp->sd_quota_mutex);
688 }
689 
690 static int gfs2_write_buf_to_page(struct gfs2_inode *ip, unsigned long index,
691 				  unsigned off, void *buf, unsigned bytes)
692 {
693 	struct inode *inode = &ip->i_inode;
694 	struct gfs2_sbd *sdp = GFS2_SB(inode);
695 	struct address_space *mapping = inode->i_mapping;
696 	struct page *page;
697 	struct buffer_head *bh;
698 	void *kaddr;
699 	u64 blk;
700 	unsigned bsize = sdp->sd_sb.sb_bsize, bnum = 0, boff = 0;
701 	unsigned to_write = bytes, pg_off = off;
702 	int done = 0;
703 
704 	blk = index << (PAGE_SHIFT - sdp->sd_sb.sb_bsize_shift);
705 	boff = off % bsize;
706 
707 	page = find_or_create_page(mapping, index, GFP_NOFS);
708 	if (!page)
709 		return -ENOMEM;
710 	if (!page_has_buffers(page))
711 		create_empty_buffers(page, bsize, 0);
712 
713 	bh = page_buffers(page);
714 	while (!done) {
715 		/* Find the beginning block within the page */
716 		if (pg_off >= ((bnum * bsize) + bsize)) {
717 			bh = bh->b_this_page;
718 			bnum++;
719 			blk++;
720 			continue;
721 		}
722 		if (!buffer_mapped(bh)) {
723 			gfs2_block_map(inode, blk, bh, 1);
724 			if (!buffer_mapped(bh))
725 				goto unlock_out;
726 			/* If it's a newly allocated disk block, zero it */
727 			if (buffer_new(bh))
728 				zero_user(page, bnum * bsize, bh->b_size);
729 		}
730 		if (PageUptodate(page))
731 			set_buffer_uptodate(bh);
732 		if (!buffer_uptodate(bh)) {
733 			ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &bh);
734 			wait_on_buffer(bh);
735 			if (!buffer_uptodate(bh))
736 				goto unlock_out;
737 		}
738 		gfs2_trans_add_data(ip->i_gl, bh);
739 
740 		/* If we need to write to the next block as well */
741 		if (to_write > (bsize - boff)) {
742 			pg_off += (bsize - boff);
743 			to_write -= (bsize - boff);
744 			boff = pg_off % bsize;
745 			continue;
746 		}
747 		done = 1;
748 	}
749 
750 	/* Write to the page, now that we have setup the buffer(s) */
751 	kaddr = kmap_atomic(page);
752 	memcpy(kaddr + off, buf, bytes);
753 	flush_dcache_page(page);
754 	kunmap_atomic(kaddr);
755 	unlock_page(page);
756 	put_page(page);
757 
758 	return 0;
759 
760 unlock_out:
761 	unlock_page(page);
762 	put_page(page);
763 	return -EIO;
764 }
765 
766 static int gfs2_write_disk_quota(struct gfs2_inode *ip, struct gfs2_quota *qp,
767 				 loff_t loc)
768 {
769 	unsigned long pg_beg;
770 	unsigned pg_off, nbytes, overflow = 0;
771 	int pg_oflow = 0, error;
772 	void *ptr;
773 
774 	nbytes = sizeof(struct gfs2_quota);
775 
776 	pg_beg = loc >> PAGE_SHIFT;
777 	pg_off = loc % PAGE_SIZE;
778 
779 	/* If the quota straddles a page boundary, split the write in two */
780 	if ((pg_off + nbytes) > PAGE_SIZE) {
781 		pg_oflow = 1;
782 		overflow = (pg_off + nbytes) - PAGE_SIZE;
783 	}
784 
785 	ptr = qp;
786 	error = gfs2_write_buf_to_page(ip, pg_beg, pg_off, ptr,
787 				       nbytes - overflow);
788 	/* If there's an overflow, write the remaining bytes to the next page */
789 	if (!error && pg_oflow)
790 		error = gfs2_write_buf_to_page(ip, pg_beg + 1, 0,
791 					       ptr + nbytes - overflow,
792 					       overflow);
793 	return error;
794 }
795 
796 /**
797  * gfs2_adjust_quota - adjust record of current block usage
798  * @ip: The quota inode
799  * @loc: Offset of the entry in the quota file
800  * @change: The amount of usage change to record
801  * @qd: The quota data
802  * @fdq: The updated limits to record
803  *
804  * This function was mostly borrowed from gfs2_block_truncate_page which was
805  * in turn mostly borrowed from ext3
806  *
807  * Returns: 0 or -ve on error
808  */
809 
810 static int gfs2_adjust_quota(struct gfs2_inode *ip, loff_t loc,
811 			     s64 change, struct gfs2_quota_data *qd,
812 			     struct qc_dqblk *fdq)
813 {
814 	struct inode *inode = &ip->i_inode;
815 	struct gfs2_sbd *sdp = GFS2_SB(inode);
816 	struct gfs2_quota q;
817 	int err;
818 	u64 size;
819 
820 	if (gfs2_is_stuffed(ip)) {
821 		err = gfs2_unstuff_dinode(ip, NULL);
822 		if (err)
823 			return err;
824 	}
825 
826 	memset(&q, 0, sizeof(struct gfs2_quota));
827 	err = gfs2_internal_read(ip, (char *)&q, &loc, sizeof(q));
828 	if (err < 0)
829 		return err;
830 
831 	loc -= sizeof(q); /* gfs2_internal_read would've advanced the loc ptr */
832 	err = -EIO;
833 	be64_add_cpu(&q.qu_value, change);
834 	if (((s64)be64_to_cpu(q.qu_value)) < 0)
835 		q.qu_value = 0; /* Never go negative on quota usage */
836 	qd->qd_qb.qb_value = q.qu_value;
837 	if (fdq) {
838 		if (fdq->d_fieldmask & QC_SPC_SOFT) {
839 			q.qu_warn = cpu_to_be64(fdq->d_spc_softlimit >> sdp->sd_sb.sb_bsize_shift);
840 			qd->qd_qb.qb_warn = q.qu_warn;
841 		}
842 		if (fdq->d_fieldmask & QC_SPC_HARD) {
843 			q.qu_limit = cpu_to_be64(fdq->d_spc_hardlimit >> sdp->sd_sb.sb_bsize_shift);
844 			qd->qd_qb.qb_limit = q.qu_limit;
845 		}
846 		if (fdq->d_fieldmask & QC_SPACE) {
847 			q.qu_value = cpu_to_be64(fdq->d_space >> sdp->sd_sb.sb_bsize_shift);
848 			qd->qd_qb.qb_value = q.qu_value;
849 		}
850 	}
851 
852 	err = gfs2_write_disk_quota(ip, &q, loc);
853 	if (!err) {
854 		size = loc + sizeof(struct gfs2_quota);
855 		if (size > inode->i_size)
856 			i_size_write(inode, size);
857 		inode->i_mtime = inode->i_atime = current_time(inode);
858 		mark_inode_dirty(inode);
859 		set_bit(QDF_REFRESH, &qd->qd_flags);
860 	}
861 
862 	return err;
863 }
864 
865 static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
866 {
867 	struct gfs2_sbd *sdp = (*qda)->qd_gl->gl_name.ln_sbd;
868 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
869 	struct gfs2_alloc_parms ap = { .aflags = 0, };
870 	unsigned int data_blocks, ind_blocks;
871 	struct gfs2_holder *ghs, i_gh;
872 	unsigned int qx, x;
873 	struct gfs2_quota_data *qd;
874 	unsigned reserved;
875 	loff_t offset;
876 	unsigned int nalloc = 0, blocks;
877 	int error;
878 
879 	error = gfs2_rsqa_alloc(ip);
880 	if (error)
881 		return error;
882 
883 	gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
884 			      &data_blocks, &ind_blocks);
885 
886 	ghs = kmalloc(num_qd * sizeof(struct gfs2_holder), GFP_NOFS);
887 	if (!ghs)
888 		return -ENOMEM;
889 
890 	sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL);
891 	inode_lock(&ip->i_inode);
892 	for (qx = 0; qx < num_qd; qx++) {
893 		error = gfs2_glock_nq_init(qda[qx]->qd_gl, LM_ST_EXCLUSIVE,
894 					   GL_NOCACHE, &ghs[qx]);
895 		if (error)
896 			goto out;
897 	}
898 
899 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
900 	if (error)
901 		goto out;
902 
903 	for (x = 0; x < num_qd; x++) {
904 		offset = qd2offset(qda[x]);
905 		if (gfs2_write_alloc_required(ip, offset,
906 					      sizeof(struct gfs2_quota)))
907 			nalloc++;
908 	}
909 
910 	/*
911 	 * 1 blk for unstuffing inode if stuffed. We add this extra
912 	 * block to the reservation unconditionally. If the inode
913 	 * doesn't need unstuffing, the block will be released to the
914 	 * rgrp since it won't be allocated during the transaction
915 	 */
916 	/* +3 in the end for unstuffing block, inode size update block
917 	 * and another block in case quota straddles page boundary and
918 	 * two blocks need to be updated instead of 1 */
919 	blocks = num_qd * data_blocks + RES_DINODE + num_qd + 3;
920 
921 	reserved = 1 + (nalloc * (data_blocks + ind_blocks));
922 	ap.target = reserved;
923 	error = gfs2_inplace_reserve(ip, &ap);
924 	if (error)
925 		goto out_alloc;
926 
927 	if (nalloc)
928 		blocks += gfs2_rg_blocks(ip, reserved) + nalloc * ind_blocks + RES_STATFS;
929 
930 	error = gfs2_trans_begin(sdp, blocks, 0);
931 	if (error)
932 		goto out_ipres;
933 
934 	for (x = 0; x < num_qd; x++) {
935 		qd = qda[x];
936 		offset = qd2offset(qd);
937 		error = gfs2_adjust_quota(ip, offset, qd->qd_change_sync, qd, NULL);
938 		if (error)
939 			goto out_end_trans;
940 
941 		do_qc(qd, -qd->qd_change_sync);
942 		set_bit(QDF_REFRESH, &qd->qd_flags);
943 	}
944 
945 	error = 0;
946 
947 out_end_trans:
948 	gfs2_trans_end(sdp);
949 out_ipres:
950 	gfs2_inplace_release(ip);
951 out_alloc:
952 	gfs2_glock_dq_uninit(&i_gh);
953 out:
954 	while (qx--)
955 		gfs2_glock_dq_uninit(&ghs[qx]);
956 	inode_unlock(&ip->i_inode);
957 	kfree(ghs);
958 	gfs2_log_flush(ip->i_gl->gl_name.ln_sbd, ip->i_gl, NORMAL_FLUSH);
959 	return error;
960 }
961 
962 static int update_qd(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd)
963 {
964 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
965 	struct gfs2_quota q;
966 	struct gfs2_quota_lvb *qlvb;
967 	loff_t pos;
968 	int error;
969 
970 	memset(&q, 0, sizeof(struct gfs2_quota));
971 	pos = qd2offset(qd);
972 	error = gfs2_internal_read(ip, (char *)&q, &pos, sizeof(q));
973 	if (error < 0)
974 		return error;
975 
976 	qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
977 	qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC);
978 	qlvb->__pad = 0;
979 	qlvb->qb_limit = q.qu_limit;
980 	qlvb->qb_warn = q.qu_warn;
981 	qlvb->qb_value = q.qu_value;
982 	qd->qd_qb = *qlvb;
983 
984 	return 0;
985 }
986 
987 static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
988 		    struct gfs2_holder *q_gh)
989 {
990 	struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
991 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
992 	struct gfs2_holder i_gh;
993 	int error;
994 
995 restart:
996 	error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
997 	if (error)
998 		return error;
999 
1000 	if (test_and_clear_bit(QDF_REFRESH, &qd->qd_flags))
1001 		force_refresh = FORCE;
1002 
1003 	qd->qd_qb = *(struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1004 
1005 	if (force_refresh || qd->qd_qb.qb_magic != cpu_to_be32(GFS2_MAGIC)) {
1006 		gfs2_glock_dq_uninit(q_gh);
1007 		error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE,
1008 					   GL_NOCACHE, q_gh);
1009 		if (error)
1010 			return error;
1011 
1012 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
1013 		if (error)
1014 			goto fail;
1015 
1016 		error = update_qd(sdp, qd);
1017 		if (error)
1018 			goto fail_gunlock;
1019 
1020 		gfs2_glock_dq_uninit(&i_gh);
1021 		gfs2_glock_dq_uninit(q_gh);
1022 		force_refresh = 0;
1023 		goto restart;
1024 	}
1025 
1026 	return 0;
1027 
1028 fail_gunlock:
1029 	gfs2_glock_dq_uninit(&i_gh);
1030 fail:
1031 	gfs2_glock_dq_uninit(q_gh);
1032 	return error;
1033 }
1034 
1035 int gfs2_quota_lock(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
1036 {
1037 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1038 	struct gfs2_quota_data *qd;
1039 	u32 x;
1040 	int error = 0;
1041 
1042 	if (capable(CAP_SYS_RESOURCE) ||
1043 	    sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
1044 		return 0;
1045 
1046 	error = gfs2_quota_hold(ip, uid, gid);
1047 	if (error)
1048 		return error;
1049 
1050 	sort(ip->i_qadata->qa_qd, ip->i_qadata->qa_qd_num,
1051 	     sizeof(struct gfs2_quota_data *), sort_qd, NULL);
1052 
1053 	for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1054 		qd = ip->i_qadata->qa_qd[x];
1055 		error = do_glock(qd, NO_FORCE, &ip->i_qadata->qa_qd_ghs[x]);
1056 		if (error)
1057 			break;
1058 	}
1059 
1060 	if (!error)
1061 		set_bit(GIF_QD_LOCKED, &ip->i_flags);
1062 	else {
1063 		while (x--)
1064 			gfs2_glock_dq_uninit(&ip->i_qadata->qa_qd_ghs[x]);
1065 		gfs2_quota_unhold(ip);
1066 	}
1067 
1068 	return error;
1069 }
1070 
1071 static int need_sync(struct gfs2_quota_data *qd)
1072 {
1073 	struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
1074 	struct gfs2_tune *gt = &sdp->sd_tune;
1075 	s64 value;
1076 	unsigned int num, den;
1077 	int do_sync = 1;
1078 
1079 	if (!qd->qd_qb.qb_limit)
1080 		return 0;
1081 
1082 	spin_lock(&qd_lock);
1083 	value = qd->qd_change;
1084 	spin_unlock(&qd_lock);
1085 
1086 	spin_lock(&gt->gt_spin);
1087 	num = gt->gt_quota_scale_num;
1088 	den = gt->gt_quota_scale_den;
1089 	spin_unlock(&gt->gt_spin);
1090 
1091 	if (value < 0)
1092 		do_sync = 0;
1093 	else if ((s64)be64_to_cpu(qd->qd_qb.qb_value) >=
1094 		 (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1095 		do_sync = 0;
1096 	else {
1097 		value *= gfs2_jindex_size(sdp) * num;
1098 		value = div_s64(value, den);
1099 		value += (s64)be64_to_cpu(qd->qd_qb.qb_value);
1100 		if (value < (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1101 			do_sync = 0;
1102 	}
1103 
1104 	return do_sync;
1105 }
1106 
1107 void gfs2_quota_unlock(struct gfs2_inode *ip)
1108 {
1109 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1110 	struct gfs2_quota_data *qda[4];
1111 	unsigned int count = 0;
1112 	u32 x;
1113 	int found;
1114 
1115 	if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
1116 		goto out;
1117 
1118 	for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1119 		struct gfs2_quota_data *qd;
1120 		int sync;
1121 
1122 		qd = ip->i_qadata->qa_qd[x];
1123 		sync = need_sync(qd);
1124 
1125 		gfs2_glock_dq_uninit(&ip->i_qadata->qa_qd_ghs[x]);
1126 		if (!sync)
1127 			continue;
1128 
1129 		spin_lock(&qd_lock);
1130 		found = qd_check_sync(sdp, qd, NULL);
1131 		spin_unlock(&qd_lock);
1132 
1133 		if (!found)
1134 			continue;
1135 
1136 		gfs2_assert_warn(sdp, qd->qd_change_sync);
1137 		if (bh_get(qd)) {
1138 			clear_bit(QDF_LOCKED, &qd->qd_flags);
1139 			slot_put(qd);
1140 			qd_put(qd);
1141 			continue;
1142 		}
1143 
1144 		qda[count++] = qd;
1145 	}
1146 
1147 	if (count) {
1148 		do_sync(count, qda);
1149 		for (x = 0; x < count; x++)
1150 			qd_unlock(qda[x]);
1151 	}
1152 
1153 out:
1154 	gfs2_quota_unhold(ip);
1155 }
1156 
1157 #define MAX_LINE 256
1158 
1159 static int print_message(struct gfs2_quota_data *qd, char *type)
1160 {
1161 	struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
1162 
1163 	fs_info(sdp, "quota %s for %s %u\n",
1164 		type,
1165 		(qd->qd_id.type == USRQUOTA) ? "user" : "group",
1166 		from_kqid(&init_user_ns, qd->qd_id));
1167 
1168 	return 0;
1169 }
1170 
1171 /**
1172  * gfs2_quota_check - check if allocating new blocks will exceed quota
1173  * @ip:  The inode for which this check is being performed
1174  * @uid: The uid to check against
1175  * @gid: The gid to check against
1176  * @ap:  The allocation parameters. ap->target contains the requested
1177  *       blocks. ap->min_target, if set, contains the minimum blks
1178  *       requested.
1179  *
1180  * Returns: 0 on success.
1181  *                  min_req = ap->min_target ? ap->min_target : ap->target;
1182  *                  quota must allow atleast min_req blks for success and
1183  *                  ap->allowed is set to the number of blocks allowed
1184  *
1185  *          -EDQUOT otherwise, quota violation. ap->allowed is set to number
1186  *                  of blocks available.
1187  */
1188 int gfs2_quota_check(struct gfs2_inode *ip, kuid_t uid, kgid_t gid,
1189 		     struct gfs2_alloc_parms *ap)
1190 {
1191 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1192 	struct gfs2_quota_data *qd;
1193 	s64 value, warn, limit;
1194 	u32 x;
1195 	int error = 0;
1196 
1197 	ap->allowed = UINT_MAX; /* Assume we are permitted a whole lot */
1198 	if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
1199 		return 0;
1200 
1201         if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
1202                 return 0;
1203 
1204 	for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1205 		qd = ip->i_qadata->qa_qd[x];
1206 
1207 		if (!(qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1208 		      qid_eq(qd->qd_id, make_kqid_gid(gid))))
1209 			continue;
1210 
1211 		warn = (s64)be64_to_cpu(qd->qd_qb.qb_warn);
1212 		limit = (s64)be64_to_cpu(qd->qd_qb.qb_limit);
1213 		value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
1214 		spin_lock(&qd_lock);
1215 		value += qd->qd_change;
1216 		spin_unlock(&qd_lock);
1217 
1218 		if (limit > 0 && (limit - value) < ap->allowed)
1219 			ap->allowed = limit - value;
1220 		/* If we can't meet the target */
1221 		if (limit && limit < (value + (s64)ap->target)) {
1222 			/* If no min_target specified or we don't meet
1223 			 * min_target, return -EDQUOT */
1224 			if (!ap->min_target || ap->min_target > ap->allowed) {
1225 				if (!test_and_set_bit(QDF_QMSG_QUIET,
1226 						      &qd->qd_flags)) {
1227 					print_message(qd, "exceeded");
1228 					quota_send_warning(qd->qd_id,
1229 							   sdp->sd_vfs->s_dev,
1230 							   QUOTA_NL_BHARDWARN);
1231 				}
1232 				error = -EDQUOT;
1233 				break;
1234 			}
1235 		} else if (warn && warn < value &&
1236 			   time_after_eq(jiffies, qd->qd_last_warn +
1237 					 gfs2_tune_get(sdp, gt_quota_warn_period)
1238 					 * HZ)) {
1239 			quota_send_warning(qd->qd_id,
1240 					   sdp->sd_vfs->s_dev, QUOTA_NL_BSOFTWARN);
1241 			error = print_message(qd, "warning");
1242 			qd->qd_last_warn = jiffies;
1243 		}
1244 	}
1245 	return error;
1246 }
1247 
1248 void gfs2_quota_change(struct gfs2_inode *ip, s64 change,
1249 		       kuid_t uid, kgid_t gid)
1250 {
1251 	struct gfs2_quota_data *qd;
1252 	u32 x;
1253 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1254 
1255 	if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON ||
1256 	    gfs2_assert_warn(sdp, change))
1257 		return;
1258 	if (ip->i_diskflags & GFS2_DIF_SYSTEM)
1259 		return;
1260 
1261 	for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1262 		qd = ip->i_qadata->qa_qd[x];
1263 
1264 		if (qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1265 		    qid_eq(qd->qd_id, make_kqid_gid(gid))) {
1266 			do_qc(qd, change);
1267 		}
1268 	}
1269 }
1270 
1271 int gfs2_quota_sync(struct super_block *sb, int type)
1272 {
1273 	struct gfs2_sbd *sdp = sb->s_fs_info;
1274 	struct gfs2_quota_data **qda;
1275 	unsigned int max_qd = PAGE_SIZE/sizeof(struct gfs2_holder);
1276 	unsigned int num_qd;
1277 	unsigned int x;
1278 	int error = 0;
1279 
1280 	qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
1281 	if (!qda)
1282 		return -ENOMEM;
1283 
1284 	mutex_lock(&sdp->sd_quota_sync_mutex);
1285 	sdp->sd_quota_sync_gen++;
1286 
1287 	do {
1288 		num_qd = 0;
1289 
1290 		for (;;) {
1291 			error = qd_fish(sdp, qda + num_qd);
1292 			if (error || !qda[num_qd])
1293 				break;
1294 			if (++num_qd == max_qd)
1295 				break;
1296 		}
1297 
1298 		if (num_qd) {
1299 			if (!error)
1300 				error = do_sync(num_qd, qda);
1301 			if (!error)
1302 				for (x = 0; x < num_qd; x++)
1303 					qda[x]->qd_sync_gen =
1304 						sdp->sd_quota_sync_gen;
1305 
1306 			for (x = 0; x < num_qd; x++)
1307 				qd_unlock(qda[x]);
1308 		}
1309 	} while (!error && num_qd == max_qd);
1310 
1311 	mutex_unlock(&sdp->sd_quota_sync_mutex);
1312 	kfree(qda);
1313 
1314 	return error;
1315 }
1316 
1317 int gfs2_quota_refresh(struct gfs2_sbd *sdp, struct kqid qid)
1318 {
1319 	struct gfs2_quota_data *qd;
1320 	struct gfs2_holder q_gh;
1321 	int error;
1322 
1323 	error = qd_get(sdp, qid, &qd);
1324 	if (error)
1325 		return error;
1326 
1327 	error = do_glock(qd, FORCE, &q_gh);
1328 	if (!error)
1329 		gfs2_glock_dq_uninit(&q_gh);
1330 
1331 	qd_put(qd);
1332 	return error;
1333 }
1334 
1335 int gfs2_quota_init(struct gfs2_sbd *sdp)
1336 {
1337 	struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
1338 	u64 size = i_size_read(sdp->sd_qc_inode);
1339 	unsigned int blocks = size >> sdp->sd_sb.sb_bsize_shift;
1340 	unsigned int x, slot = 0;
1341 	unsigned int found = 0;
1342 	unsigned int hash;
1343 	unsigned int bm_size;
1344 	u64 dblock;
1345 	u32 extlen = 0;
1346 	int error;
1347 
1348 	if (gfs2_check_internal_file_size(sdp->sd_qc_inode, 1, 64 << 20))
1349 		return -EIO;
1350 
1351 	sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
1352 	bm_size = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * sizeof(unsigned long));
1353 	bm_size *= sizeof(unsigned long);
1354 	error = -ENOMEM;
1355 	sdp->sd_quota_bitmap = kzalloc(bm_size, GFP_NOFS | __GFP_NOWARN);
1356 	if (sdp->sd_quota_bitmap == NULL)
1357 		sdp->sd_quota_bitmap = __vmalloc(bm_size, GFP_NOFS |
1358 						 __GFP_ZERO, PAGE_KERNEL);
1359 	if (!sdp->sd_quota_bitmap)
1360 		return error;
1361 
1362 	for (x = 0; x < blocks; x++) {
1363 		struct buffer_head *bh;
1364 		const struct gfs2_quota_change *qc;
1365 		unsigned int y;
1366 
1367 		if (!extlen) {
1368 			int new = 0;
1369 			error = gfs2_extent_map(&ip->i_inode, x, &new, &dblock, &extlen);
1370 			if (error)
1371 				goto fail;
1372 		}
1373 		error = -EIO;
1374 		bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
1375 		if (!bh)
1376 			goto fail;
1377 		if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
1378 			brelse(bh);
1379 			goto fail;
1380 		}
1381 
1382 		qc = (const struct gfs2_quota_change *)(bh->b_data + sizeof(struct gfs2_meta_header));
1383 		for (y = 0; y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1384 		     y++, slot++) {
1385 			struct gfs2_quota_data *qd;
1386 			s64 qc_change = be64_to_cpu(qc->qc_change);
1387 			u32 qc_flags = be32_to_cpu(qc->qc_flags);
1388 			enum quota_type qtype = (qc_flags & GFS2_QCF_USER) ?
1389 						USRQUOTA : GRPQUOTA;
1390 			struct kqid qc_id = make_kqid(&init_user_ns, qtype,
1391 						      be32_to_cpu(qc->qc_id));
1392 			qc++;
1393 			if (!qc_change)
1394 				continue;
1395 
1396 			hash = gfs2_qd_hash(sdp, qc_id);
1397 			qd = qd_alloc(hash, sdp, qc_id);
1398 			if (qd == NULL) {
1399 				brelse(bh);
1400 				goto fail;
1401 			}
1402 
1403 			set_bit(QDF_CHANGE, &qd->qd_flags);
1404 			qd->qd_change = qc_change;
1405 			qd->qd_slot = slot;
1406 			qd->qd_slot_count = 1;
1407 
1408 			spin_lock(&qd_lock);
1409 			BUG_ON(test_and_set_bit(slot, sdp->sd_quota_bitmap));
1410 			list_add(&qd->qd_list, &sdp->sd_quota_list);
1411 			atomic_inc(&sdp->sd_quota_count);
1412 			spin_unlock(&qd_lock);
1413 
1414 			spin_lock_bucket(hash);
1415 			hlist_bl_add_head_rcu(&qd->qd_hlist, &qd_hash_table[hash]);
1416 			spin_unlock_bucket(hash);
1417 
1418 			found++;
1419 		}
1420 
1421 		brelse(bh);
1422 		dblock++;
1423 		extlen--;
1424 	}
1425 
1426 	if (found)
1427 		fs_info(sdp, "found %u quota changes\n", found);
1428 
1429 	return 0;
1430 
1431 fail:
1432 	gfs2_quota_cleanup(sdp);
1433 	return error;
1434 }
1435 
1436 void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1437 {
1438 	struct list_head *head = &sdp->sd_quota_list;
1439 	struct gfs2_quota_data *qd;
1440 
1441 	spin_lock(&qd_lock);
1442 	while (!list_empty(head)) {
1443 		qd = list_entry(head->prev, struct gfs2_quota_data, qd_list);
1444 
1445 		list_del(&qd->qd_list);
1446 
1447 		/* Also remove if this qd exists in the reclaim list */
1448 		list_lru_del(&gfs2_qd_lru, &qd->qd_lru);
1449 		atomic_dec(&sdp->sd_quota_count);
1450 		spin_unlock(&qd_lock);
1451 
1452 		spin_lock_bucket(qd->qd_hash);
1453 		hlist_bl_del_rcu(&qd->qd_hlist);
1454 		spin_unlock_bucket(qd->qd_hash);
1455 
1456 		gfs2_assert_warn(sdp, !qd->qd_change);
1457 		gfs2_assert_warn(sdp, !qd->qd_slot_count);
1458 		gfs2_assert_warn(sdp, !qd->qd_bh_count);
1459 
1460 		gfs2_glock_put(qd->qd_gl);
1461 		call_rcu(&qd->qd_rcu, gfs2_qd_dealloc);
1462 
1463 		spin_lock(&qd_lock);
1464 	}
1465 	spin_unlock(&qd_lock);
1466 
1467 	gfs2_assert_warn(sdp, !atomic_read(&sdp->sd_quota_count));
1468 
1469 	kvfree(sdp->sd_quota_bitmap);
1470 	sdp->sd_quota_bitmap = NULL;
1471 }
1472 
1473 static void quotad_error(struct gfs2_sbd *sdp, const char *msg, int error)
1474 {
1475 	if (error == 0 || error == -EROFS)
1476 		return;
1477 	if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) {
1478 		fs_err(sdp, "gfs2_quotad: %s error %d\n", msg, error);
1479 		sdp->sd_log_error = error;
1480 		wake_up(&sdp->sd_logd_waitq);
1481 	}
1482 }
1483 
1484 static void quotad_check_timeo(struct gfs2_sbd *sdp, const char *msg,
1485 			       int (*fxn)(struct super_block *sb, int type),
1486 			       unsigned long t, unsigned long *timeo,
1487 			       unsigned int *new_timeo)
1488 {
1489 	if (t >= *timeo) {
1490 		int error = fxn(sdp->sd_vfs, 0);
1491 		quotad_error(sdp, msg, error);
1492 		*timeo = gfs2_tune_get_i(&sdp->sd_tune, new_timeo) * HZ;
1493 	} else {
1494 		*timeo -= t;
1495 	}
1496 }
1497 
1498 static void quotad_check_trunc_list(struct gfs2_sbd *sdp)
1499 {
1500 	struct gfs2_inode *ip;
1501 
1502 	while(1) {
1503 		ip = NULL;
1504 		spin_lock(&sdp->sd_trunc_lock);
1505 		if (!list_empty(&sdp->sd_trunc_list)) {
1506 			ip = list_entry(sdp->sd_trunc_list.next,
1507 					struct gfs2_inode, i_trunc_list);
1508 			list_del_init(&ip->i_trunc_list);
1509 		}
1510 		spin_unlock(&sdp->sd_trunc_lock);
1511 		if (ip == NULL)
1512 			return;
1513 		gfs2_glock_finish_truncate(ip);
1514 	}
1515 }
1516 
1517 void gfs2_wake_up_statfs(struct gfs2_sbd *sdp) {
1518 	if (!sdp->sd_statfs_force_sync) {
1519 		sdp->sd_statfs_force_sync = 1;
1520 		wake_up(&sdp->sd_quota_wait);
1521 	}
1522 }
1523 
1524 
1525 /**
1526  * gfs2_quotad - Write cached quota changes into the quota file
1527  * @sdp: Pointer to GFS2 superblock
1528  *
1529  */
1530 
1531 int gfs2_quotad(void *data)
1532 {
1533 	struct gfs2_sbd *sdp = data;
1534 	struct gfs2_tune *tune = &sdp->sd_tune;
1535 	unsigned long statfs_timeo = 0;
1536 	unsigned long quotad_timeo = 0;
1537 	unsigned long t = 0;
1538 	DEFINE_WAIT(wait);
1539 	int empty;
1540 
1541 	while (!kthread_should_stop()) {
1542 
1543 		/* Update the master statfs file */
1544 		if (sdp->sd_statfs_force_sync) {
1545 			int error = gfs2_statfs_sync(sdp->sd_vfs, 0);
1546 			quotad_error(sdp, "statfs", error);
1547 			statfs_timeo = gfs2_tune_get(sdp, gt_statfs_quantum) * HZ;
1548 		}
1549 		else
1550 			quotad_check_timeo(sdp, "statfs", gfs2_statfs_sync, t,
1551 				   	   &statfs_timeo,
1552 					   &tune->gt_statfs_quantum);
1553 
1554 		/* Update quota file */
1555 		quotad_check_timeo(sdp, "sync", gfs2_quota_sync, t,
1556 				   &quotad_timeo, &tune->gt_quota_quantum);
1557 
1558 		/* Check for & recover partially truncated inodes */
1559 		quotad_check_trunc_list(sdp);
1560 
1561 		try_to_freeze();
1562 
1563 		t = min(quotad_timeo, statfs_timeo);
1564 
1565 		prepare_to_wait(&sdp->sd_quota_wait, &wait, TASK_INTERRUPTIBLE);
1566 		spin_lock(&sdp->sd_trunc_lock);
1567 		empty = list_empty(&sdp->sd_trunc_list);
1568 		spin_unlock(&sdp->sd_trunc_lock);
1569 		if (empty && !sdp->sd_statfs_force_sync)
1570 			t -= schedule_timeout(t);
1571 		else
1572 			t = 0;
1573 		finish_wait(&sdp->sd_quota_wait, &wait);
1574 	}
1575 
1576 	return 0;
1577 }
1578 
1579 static int gfs2_quota_get_state(struct super_block *sb, struct qc_state *state)
1580 {
1581 	struct gfs2_sbd *sdp = sb->s_fs_info;
1582 
1583 	memset(state, 0, sizeof(*state));
1584 
1585 	switch (sdp->sd_args.ar_quota) {
1586 	case GFS2_QUOTA_ON:
1587 		state->s_state[USRQUOTA].flags |= QCI_LIMITS_ENFORCED;
1588 		state->s_state[GRPQUOTA].flags |= QCI_LIMITS_ENFORCED;
1589 		/*FALLTHRU*/
1590 	case GFS2_QUOTA_ACCOUNT:
1591 		state->s_state[USRQUOTA].flags |= QCI_ACCT_ENABLED |
1592 						  QCI_SYSFILE;
1593 		state->s_state[GRPQUOTA].flags |= QCI_ACCT_ENABLED |
1594 						  QCI_SYSFILE;
1595 		break;
1596 	case GFS2_QUOTA_OFF:
1597 		break;
1598 	}
1599 	if (sdp->sd_quota_inode) {
1600 		state->s_state[USRQUOTA].ino =
1601 					GFS2_I(sdp->sd_quota_inode)->i_no_addr;
1602 		state->s_state[USRQUOTA].blocks = sdp->sd_quota_inode->i_blocks;
1603 	}
1604 	state->s_state[USRQUOTA].nextents = 1;	/* unsupported */
1605 	state->s_state[GRPQUOTA] = state->s_state[USRQUOTA];
1606 	state->s_incoredqs = list_lru_count(&gfs2_qd_lru);
1607 	return 0;
1608 }
1609 
1610 static int gfs2_get_dqblk(struct super_block *sb, struct kqid qid,
1611 			  struct qc_dqblk *fdq)
1612 {
1613 	struct gfs2_sbd *sdp = sb->s_fs_info;
1614 	struct gfs2_quota_lvb *qlvb;
1615 	struct gfs2_quota_data *qd;
1616 	struct gfs2_holder q_gh;
1617 	int error;
1618 
1619 	memset(fdq, 0, sizeof(*fdq));
1620 
1621 	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1622 		return -ESRCH; /* Crazy XFS error code */
1623 
1624 	if ((qid.type != USRQUOTA) &&
1625 	    (qid.type != GRPQUOTA))
1626 		return -EINVAL;
1627 
1628 	error = qd_get(sdp, qid, &qd);
1629 	if (error)
1630 		return error;
1631 	error = do_glock(qd, FORCE, &q_gh);
1632 	if (error)
1633 		goto out;
1634 
1635 	qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1636 	fdq->d_spc_hardlimit = be64_to_cpu(qlvb->qb_limit) << sdp->sd_sb.sb_bsize_shift;
1637 	fdq->d_spc_softlimit = be64_to_cpu(qlvb->qb_warn) << sdp->sd_sb.sb_bsize_shift;
1638 	fdq->d_space = be64_to_cpu(qlvb->qb_value) << sdp->sd_sb.sb_bsize_shift;
1639 
1640 	gfs2_glock_dq_uninit(&q_gh);
1641 out:
1642 	qd_put(qd);
1643 	return error;
1644 }
1645 
1646 /* GFS2 only supports a subset of the XFS fields */
1647 #define GFS2_FIELDMASK (QC_SPC_SOFT|QC_SPC_HARD|QC_SPACE)
1648 
1649 static int gfs2_set_dqblk(struct super_block *sb, struct kqid qid,
1650 			  struct qc_dqblk *fdq)
1651 {
1652 	struct gfs2_sbd *sdp = sb->s_fs_info;
1653 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1654 	struct gfs2_quota_data *qd;
1655 	struct gfs2_holder q_gh, i_gh;
1656 	unsigned int data_blocks, ind_blocks;
1657 	unsigned int blocks = 0;
1658 	int alloc_required;
1659 	loff_t offset;
1660 	int error;
1661 
1662 	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1663 		return -ESRCH; /* Crazy XFS error code */
1664 
1665 	if ((qid.type != USRQUOTA) &&
1666 	    (qid.type != GRPQUOTA))
1667 		return -EINVAL;
1668 
1669 	if (fdq->d_fieldmask & ~GFS2_FIELDMASK)
1670 		return -EINVAL;
1671 
1672 	error = qd_get(sdp, qid, &qd);
1673 	if (error)
1674 		return error;
1675 
1676 	error = gfs2_rsqa_alloc(ip);
1677 	if (error)
1678 		goto out_put;
1679 
1680 	inode_lock(&ip->i_inode);
1681 	error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE, 0, &q_gh);
1682 	if (error)
1683 		goto out_unlockput;
1684 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1685 	if (error)
1686 		goto out_q;
1687 
1688 	/* Check for existing entry, if none then alloc new blocks */
1689 	error = update_qd(sdp, qd);
1690 	if (error)
1691 		goto out_i;
1692 
1693 	/* If nothing has changed, this is a no-op */
1694 	if ((fdq->d_fieldmask & QC_SPC_SOFT) &&
1695 	    ((fdq->d_spc_softlimit >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_warn)))
1696 		fdq->d_fieldmask ^= QC_SPC_SOFT;
1697 
1698 	if ((fdq->d_fieldmask & QC_SPC_HARD) &&
1699 	    ((fdq->d_spc_hardlimit >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_limit)))
1700 		fdq->d_fieldmask ^= QC_SPC_HARD;
1701 
1702 	if ((fdq->d_fieldmask & QC_SPACE) &&
1703 	    ((fdq->d_space >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_value)))
1704 		fdq->d_fieldmask ^= QC_SPACE;
1705 
1706 	if (fdq->d_fieldmask == 0)
1707 		goto out_i;
1708 
1709 	offset = qd2offset(qd);
1710 	alloc_required = gfs2_write_alloc_required(ip, offset, sizeof(struct gfs2_quota));
1711 	if (gfs2_is_stuffed(ip))
1712 		alloc_required = 1;
1713 	if (alloc_required) {
1714 		struct gfs2_alloc_parms ap = { .aflags = 0, };
1715 		gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
1716 				       &data_blocks, &ind_blocks);
1717 		blocks = 1 + data_blocks + ind_blocks;
1718 		ap.target = blocks;
1719 		error = gfs2_inplace_reserve(ip, &ap);
1720 		if (error)
1721 			goto out_i;
1722 		blocks += gfs2_rg_blocks(ip, blocks);
1723 	}
1724 
1725 	/* Some quotas span block boundaries and can update two blocks,
1726 	   adding an extra block to the transaction to handle such quotas */
1727 	error = gfs2_trans_begin(sdp, blocks + RES_DINODE + 2, 0);
1728 	if (error)
1729 		goto out_release;
1730 
1731 	/* Apply changes */
1732 	error = gfs2_adjust_quota(ip, offset, 0, qd, fdq);
1733 	if (!error)
1734 		clear_bit(QDF_QMSG_QUIET, &qd->qd_flags);
1735 
1736 	gfs2_trans_end(sdp);
1737 out_release:
1738 	if (alloc_required)
1739 		gfs2_inplace_release(ip);
1740 out_i:
1741 	gfs2_glock_dq_uninit(&i_gh);
1742 out_q:
1743 	gfs2_glock_dq_uninit(&q_gh);
1744 out_unlockput:
1745 	inode_unlock(&ip->i_inode);
1746 out_put:
1747 	qd_put(qd);
1748 	return error;
1749 }
1750 
1751 const struct quotactl_ops gfs2_quotactl_ops = {
1752 	.quota_sync     = gfs2_quota_sync,
1753 	.get_state	= gfs2_quota_get_state,
1754 	.get_dqblk	= gfs2_get_dqblk,
1755 	.set_dqblk	= gfs2_set_dqblk,
1756 };
1757 
1758 void __init gfs2_quota_hash_init(void)
1759 {
1760 	unsigned i;
1761 
1762 	for(i = 0; i < GFS2_QD_HASH_SIZE; i++)
1763 		INIT_HLIST_BL_HEAD(&qd_hash_table[i]);
1764 }
1765