xref: /linux/fs/gfs2/quota.c (revision c5d3cdad688ed75fb311a3a671eb30ba7106d7d3)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
4  * Copyright (C) 2004-2007 Red Hat, Inc.  All rights reserved.
5  */
6 
7 /*
8  * Quota change tags are associated with each transaction that allocates or
9  * deallocates space.  Those changes are accumulated locally to each node (in a
10  * per-node file) and then are periodically synced to the quota file.  This
11  * avoids the bottleneck of constantly touching the quota file, but introduces
12  * fuzziness in the current usage value of IDs that are being used on different
13  * nodes in the cluster simultaneously.  So, it is possible for a user on
14  * multiple nodes to overrun their quota, but that overrun is controlable.
15  * Since quota tags are part of transactions, there is no need for a quota check
16  * program to be run on node crashes or anything like that.
17  *
18  * There are couple of knobs that let the administrator manage the quota
19  * fuzziness.  "quota_quantum" sets the maximum time a quota change can be
20  * sitting on one node before being synced to the quota file.  (The default is
21  * 60 seconds.)  Another knob, "quota_scale" controls how quickly the frequency
22  * of quota file syncs increases as the user moves closer to their limit.  The
23  * more frequent the syncs, the more accurate the quota enforcement, but that
24  * means that there is more contention between the nodes for the quota file.
25  * The default value is one.  This sets the maximum theoretical quota overrun
26  * (with infinite node with infinite bandwidth) to twice the user's limit.  (In
27  * practice, the maximum overrun you see should be much less.)  A "quota_scale"
28  * number greater than one makes quota syncs more frequent and reduces the
29  * maximum overrun.  Numbers less than one (but greater than zero) make quota
30  * syncs less frequent.
31  *
32  * GFS quotas also use per-ID Lock Value Blocks (LVBs) to cache the contents of
33  * the quota file, so it is not being constantly read.
34  */
35 
36 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
37 
38 #include <linux/sched.h>
39 #include <linux/slab.h>
40 #include <linux/mm.h>
41 #include <linux/spinlock.h>
42 #include <linux/completion.h>
43 #include <linux/buffer_head.h>
44 #include <linux/sort.h>
45 #include <linux/fs.h>
46 #include <linux/bio.h>
47 #include <linux/gfs2_ondisk.h>
48 #include <linux/kthread.h>
49 #include <linux/freezer.h>
50 #include <linux/quota.h>
51 #include <linux/dqblk_xfs.h>
52 #include <linux/lockref.h>
53 #include <linux/list_lru.h>
54 #include <linux/rcupdate.h>
55 #include <linux/rculist_bl.h>
56 #include <linux/bit_spinlock.h>
57 #include <linux/jhash.h>
58 #include <linux/vmalloc.h>
59 
60 #include "gfs2.h"
61 #include "incore.h"
62 #include "bmap.h"
63 #include "glock.h"
64 #include "glops.h"
65 #include "log.h"
66 #include "meta_io.h"
67 #include "quota.h"
68 #include "rgrp.h"
69 #include "super.h"
70 #include "trans.h"
71 #include "inode.h"
72 #include "util.h"
73 
74 #define GFS2_QD_HASH_SHIFT      12
75 #define GFS2_QD_HASH_SIZE       BIT(GFS2_QD_HASH_SHIFT)
76 #define GFS2_QD_HASH_MASK       (GFS2_QD_HASH_SIZE - 1)
77 
78 /* Lock order: qd_lock -> bucket lock -> qd->lockref.lock -> lru lock */
79 /*                     -> sd_bitmap_lock                              */
80 static DEFINE_SPINLOCK(qd_lock);
81 struct list_lru gfs2_qd_lru;
82 
83 static struct hlist_bl_head qd_hash_table[GFS2_QD_HASH_SIZE];
84 
85 static unsigned int gfs2_qd_hash(const struct gfs2_sbd *sdp,
86 				 const struct kqid qid)
87 {
88 	unsigned int h;
89 
90 	h = jhash(&sdp, sizeof(struct gfs2_sbd *), 0);
91 	h = jhash(&qid, sizeof(struct kqid), h);
92 
93 	return h & GFS2_QD_HASH_MASK;
94 }
95 
96 static inline void spin_lock_bucket(unsigned int hash)
97 {
98         hlist_bl_lock(&qd_hash_table[hash]);
99 }
100 
101 static inline void spin_unlock_bucket(unsigned int hash)
102 {
103         hlist_bl_unlock(&qd_hash_table[hash]);
104 }
105 
106 static void gfs2_qd_dealloc(struct rcu_head *rcu)
107 {
108 	struct gfs2_quota_data *qd = container_of(rcu, struct gfs2_quota_data, qd_rcu);
109 	kmem_cache_free(gfs2_quotad_cachep, qd);
110 }
111 
112 static void gfs2_qd_dispose(struct list_head *list)
113 {
114 	struct gfs2_quota_data *qd;
115 	struct gfs2_sbd *sdp;
116 
117 	while (!list_empty(list)) {
118 		qd = list_first_entry(list, struct gfs2_quota_data, qd_lru);
119 		sdp = qd->qd_gl->gl_name.ln_sbd;
120 
121 		list_del(&qd->qd_lru);
122 
123 		/* Free from the filesystem-specific list */
124 		spin_lock(&qd_lock);
125 		list_del(&qd->qd_list);
126 		spin_unlock(&qd_lock);
127 
128 		spin_lock_bucket(qd->qd_hash);
129 		hlist_bl_del_rcu(&qd->qd_hlist);
130 		spin_unlock_bucket(qd->qd_hash);
131 
132 		gfs2_assert_warn(sdp, !qd->qd_change);
133 		gfs2_assert_warn(sdp, !qd->qd_slot_count);
134 		gfs2_assert_warn(sdp, !qd->qd_bh_count);
135 
136 		gfs2_glock_put(qd->qd_gl);
137 		atomic_dec(&sdp->sd_quota_count);
138 
139 		/* Delete it from the common reclaim list */
140 		call_rcu(&qd->qd_rcu, gfs2_qd_dealloc);
141 	}
142 }
143 
144 
145 static enum lru_status gfs2_qd_isolate(struct list_head *item,
146 		struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
147 {
148 	struct list_head *dispose = arg;
149 	struct gfs2_quota_data *qd = list_entry(item, struct gfs2_quota_data, qd_lru);
150 
151 	if (!spin_trylock(&qd->qd_lockref.lock))
152 		return LRU_SKIP;
153 
154 	if (qd->qd_lockref.count == 0) {
155 		lockref_mark_dead(&qd->qd_lockref);
156 		list_lru_isolate_move(lru, &qd->qd_lru, dispose);
157 	}
158 
159 	spin_unlock(&qd->qd_lockref.lock);
160 	return LRU_REMOVED;
161 }
162 
163 static unsigned long gfs2_qd_shrink_scan(struct shrinker *shrink,
164 					 struct shrink_control *sc)
165 {
166 	LIST_HEAD(dispose);
167 	unsigned long freed;
168 
169 	if (!(sc->gfp_mask & __GFP_FS))
170 		return SHRINK_STOP;
171 
172 	freed = list_lru_shrink_walk(&gfs2_qd_lru, sc,
173 				     gfs2_qd_isolate, &dispose);
174 
175 	gfs2_qd_dispose(&dispose);
176 
177 	return freed;
178 }
179 
180 static unsigned long gfs2_qd_shrink_count(struct shrinker *shrink,
181 					  struct shrink_control *sc)
182 {
183 	return vfs_pressure_ratio(list_lru_shrink_count(&gfs2_qd_lru, sc));
184 }
185 
186 struct shrinker gfs2_qd_shrinker = {
187 	.count_objects = gfs2_qd_shrink_count,
188 	.scan_objects = gfs2_qd_shrink_scan,
189 	.seeks = DEFAULT_SEEKS,
190 	.flags = SHRINKER_NUMA_AWARE,
191 };
192 
193 
194 static u64 qd2index(struct gfs2_quota_data *qd)
195 {
196 	struct kqid qid = qd->qd_id;
197 	return (2 * (u64)from_kqid(&init_user_ns, qid)) +
198 		((qid.type == USRQUOTA) ? 0 : 1);
199 }
200 
201 static u64 qd2offset(struct gfs2_quota_data *qd)
202 {
203 	u64 offset;
204 
205 	offset = qd2index(qd);
206 	offset *= sizeof(struct gfs2_quota);
207 
208 	return offset;
209 }
210 
211 static struct gfs2_quota_data *qd_alloc(unsigned hash, struct gfs2_sbd *sdp, struct kqid qid)
212 {
213 	struct gfs2_quota_data *qd;
214 	int error;
215 
216 	qd = kmem_cache_zalloc(gfs2_quotad_cachep, GFP_NOFS);
217 	if (!qd)
218 		return NULL;
219 
220 	qd->qd_sbd = sdp;
221 	qd->qd_lockref.count = 1;
222 	spin_lock_init(&qd->qd_lockref.lock);
223 	qd->qd_id = qid;
224 	qd->qd_slot = -1;
225 	INIT_LIST_HEAD(&qd->qd_lru);
226 	qd->qd_hash = hash;
227 
228 	error = gfs2_glock_get(sdp, qd2index(qd),
229 			      &gfs2_quota_glops, CREATE, &qd->qd_gl);
230 	if (error)
231 		goto fail;
232 
233 	return qd;
234 
235 fail:
236 	kmem_cache_free(gfs2_quotad_cachep, qd);
237 	return NULL;
238 }
239 
240 static struct gfs2_quota_data *gfs2_qd_search_bucket(unsigned int hash,
241 						     const struct gfs2_sbd *sdp,
242 						     struct kqid qid)
243 {
244 	struct gfs2_quota_data *qd;
245 	struct hlist_bl_node *h;
246 
247 	hlist_bl_for_each_entry_rcu(qd, h, &qd_hash_table[hash], qd_hlist) {
248 		if (!qid_eq(qd->qd_id, qid))
249 			continue;
250 		if (qd->qd_sbd != sdp)
251 			continue;
252 		if (lockref_get_not_dead(&qd->qd_lockref)) {
253 			list_lru_del(&gfs2_qd_lru, &qd->qd_lru);
254 			return qd;
255 		}
256 	}
257 
258 	return NULL;
259 }
260 
261 
262 static int qd_get(struct gfs2_sbd *sdp, struct kqid qid,
263 		  struct gfs2_quota_data **qdp)
264 {
265 	struct gfs2_quota_data *qd, *new_qd;
266 	unsigned int hash = gfs2_qd_hash(sdp, qid);
267 
268 	rcu_read_lock();
269 	*qdp = qd = gfs2_qd_search_bucket(hash, sdp, qid);
270 	rcu_read_unlock();
271 
272 	if (qd)
273 		return 0;
274 
275 	new_qd = qd_alloc(hash, sdp, qid);
276 	if (!new_qd)
277 		return -ENOMEM;
278 
279 	spin_lock(&qd_lock);
280 	spin_lock_bucket(hash);
281 	*qdp = qd = gfs2_qd_search_bucket(hash, sdp, qid);
282 	if (qd == NULL) {
283 		*qdp = new_qd;
284 		list_add(&new_qd->qd_list, &sdp->sd_quota_list);
285 		hlist_bl_add_head_rcu(&new_qd->qd_hlist, &qd_hash_table[hash]);
286 		atomic_inc(&sdp->sd_quota_count);
287 	}
288 	spin_unlock_bucket(hash);
289 	spin_unlock(&qd_lock);
290 
291 	if (qd) {
292 		gfs2_glock_put(new_qd->qd_gl);
293 		kmem_cache_free(gfs2_quotad_cachep, new_qd);
294 	}
295 
296 	return 0;
297 }
298 
299 
300 static void qd_hold(struct gfs2_quota_data *qd)
301 {
302 	struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
303 	gfs2_assert(sdp, !__lockref_is_dead(&qd->qd_lockref));
304 	lockref_get(&qd->qd_lockref);
305 }
306 
307 static void qd_put(struct gfs2_quota_data *qd)
308 {
309 	if (lockref_put_or_lock(&qd->qd_lockref))
310 		return;
311 
312 	qd->qd_lockref.count = 0;
313 	list_lru_add(&gfs2_qd_lru, &qd->qd_lru);
314 	spin_unlock(&qd->qd_lockref.lock);
315 
316 }
317 
318 static int slot_get(struct gfs2_quota_data *qd)
319 {
320 	struct gfs2_sbd *sdp = qd->qd_sbd;
321 	unsigned int bit;
322 	int error = 0;
323 
324 	spin_lock(&sdp->sd_bitmap_lock);
325 	if (qd->qd_slot_count != 0)
326 		goto out;
327 
328 	error = -ENOSPC;
329 	bit = find_first_zero_bit(sdp->sd_quota_bitmap, sdp->sd_quota_slots);
330 	if (bit < sdp->sd_quota_slots) {
331 		set_bit(bit, sdp->sd_quota_bitmap);
332 		qd->qd_slot = bit;
333 		error = 0;
334 out:
335 		qd->qd_slot_count++;
336 	}
337 	spin_unlock(&sdp->sd_bitmap_lock);
338 
339 	return error;
340 }
341 
342 static void slot_hold(struct gfs2_quota_data *qd)
343 {
344 	struct gfs2_sbd *sdp = qd->qd_sbd;
345 
346 	spin_lock(&sdp->sd_bitmap_lock);
347 	gfs2_assert(sdp, qd->qd_slot_count);
348 	qd->qd_slot_count++;
349 	spin_unlock(&sdp->sd_bitmap_lock);
350 }
351 
352 static void slot_put(struct gfs2_quota_data *qd)
353 {
354 	struct gfs2_sbd *sdp = qd->qd_sbd;
355 
356 	spin_lock(&sdp->sd_bitmap_lock);
357 	gfs2_assert(sdp, qd->qd_slot_count);
358 	if (!--qd->qd_slot_count) {
359 		BUG_ON(!test_and_clear_bit(qd->qd_slot, sdp->sd_quota_bitmap));
360 		qd->qd_slot = -1;
361 	}
362 	spin_unlock(&sdp->sd_bitmap_lock);
363 }
364 
365 static int bh_get(struct gfs2_quota_data *qd)
366 {
367 	struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
368 	struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
369 	unsigned int block, offset;
370 	struct buffer_head *bh;
371 	int error;
372 	struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
373 
374 	mutex_lock(&sdp->sd_quota_mutex);
375 
376 	if (qd->qd_bh_count++) {
377 		mutex_unlock(&sdp->sd_quota_mutex);
378 		return 0;
379 	}
380 
381 	block = qd->qd_slot / sdp->sd_qc_per_block;
382 	offset = qd->qd_slot % sdp->sd_qc_per_block;
383 
384 	bh_map.b_size = BIT(ip->i_inode.i_blkbits);
385 	error = gfs2_block_map(&ip->i_inode, block, &bh_map, 0);
386 	if (error)
387 		goto fail;
388 	error = gfs2_meta_read(ip->i_gl, bh_map.b_blocknr, DIO_WAIT, 0, &bh);
389 	if (error)
390 		goto fail;
391 	error = -EIO;
392 	if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC))
393 		goto fail_brelse;
394 
395 	qd->qd_bh = bh;
396 	qd->qd_bh_qc = (struct gfs2_quota_change *)
397 		(bh->b_data + sizeof(struct gfs2_meta_header) +
398 		 offset * sizeof(struct gfs2_quota_change));
399 
400 	mutex_unlock(&sdp->sd_quota_mutex);
401 
402 	return 0;
403 
404 fail_brelse:
405 	brelse(bh);
406 fail:
407 	qd->qd_bh_count--;
408 	mutex_unlock(&sdp->sd_quota_mutex);
409 	return error;
410 }
411 
412 static void bh_put(struct gfs2_quota_data *qd)
413 {
414 	struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
415 
416 	mutex_lock(&sdp->sd_quota_mutex);
417 	gfs2_assert(sdp, qd->qd_bh_count);
418 	if (!--qd->qd_bh_count) {
419 		brelse(qd->qd_bh);
420 		qd->qd_bh = NULL;
421 		qd->qd_bh_qc = NULL;
422 	}
423 	mutex_unlock(&sdp->sd_quota_mutex);
424 }
425 
426 static int qd_check_sync(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd,
427 			 u64 *sync_gen)
428 {
429 	if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
430 	    !test_bit(QDF_CHANGE, &qd->qd_flags) ||
431 	    (sync_gen && (qd->qd_sync_gen >= *sync_gen)))
432 		return 0;
433 
434 	if (!lockref_get_not_dead(&qd->qd_lockref))
435 		return 0;
436 
437 	list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
438 	set_bit(QDF_LOCKED, &qd->qd_flags);
439 	qd->qd_change_sync = qd->qd_change;
440 	slot_hold(qd);
441 	return 1;
442 }
443 
444 static int qd_fish(struct gfs2_sbd *sdp, struct gfs2_quota_data **qdp)
445 {
446 	struct gfs2_quota_data *qd = NULL;
447 	int error;
448 	int found = 0;
449 
450 	*qdp = NULL;
451 
452 	if (sb_rdonly(sdp->sd_vfs))
453 		return 0;
454 
455 	spin_lock(&qd_lock);
456 
457 	list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
458 		found = qd_check_sync(sdp, qd, &sdp->sd_quota_sync_gen);
459 		if (found)
460 			break;
461 	}
462 
463 	if (!found)
464 		qd = NULL;
465 
466 	spin_unlock(&qd_lock);
467 
468 	if (qd) {
469 		gfs2_assert_warn(sdp, qd->qd_change_sync);
470 		error = bh_get(qd);
471 		if (error) {
472 			clear_bit(QDF_LOCKED, &qd->qd_flags);
473 			slot_put(qd);
474 			qd_put(qd);
475 			return error;
476 		}
477 	}
478 
479 	*qdp = qd;
480 
481 	return 0;
482 }
483 
484 static void qd_unlock(struct gfs2_quota_data *qd)
485 {
486 	gfs2_assert_warn(qd->qd_gl->gl_name.ln_sbd,
487 			 test_bit(QDF_LOCKED, &qd->qd_flags));
488 	clear_bit(QDF_LOCKED, &qd->qd_flags);
489 	bh_put(qd);
490 	slot_put(qd);
491 	qd_put(qd);
492 }
493 
494 static int qdsb_get(struct gfs2_sbd *sdp, struct kqid qid,
495 		    struct gfs2_quota_data **qdp)
496 {
497 	int error;
498 
499 	error = qd_get(sdp, qid, qdp);
500 	if (error)
501 		return error;
502 
503 	error = slot_get(*qdp);
504 	if (error)
505 		goto fail;
506 
507 	error = bh_get(*qdp);
508 	if (error)
509 		goto fail_slot;
510 
511 	return 0;
512 
513 fail_slot:
514 	slot_put(*qdp);
515 fail:
516 	qd_put(*qdp);
517 	return error;
518 }
519 
520 static void qdsb_put(struct gfs2_quota_data *qd)
521 {
522 	bh_put(qd);
523 	slot_put(qd);
524 	qd_put(qd);
525 }
526 
527 /**
528  * gfs2_qa_get - make sure we have a quota allocations data structure,
529  *               if necessary
530  * @ip: the inode for this reservation
531  */
532 int gfs2_qa_get(struct gfs2_inode *ip)
533 {
534 	int error = 0;
535 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
536 
537 	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
538 		return 0;
539 
540 	down_write(&ip->i_rw_mutex);
541 	if (ip->i_qadata == NULL) {
542 		ip->i_qadata = kmem_cache_zalloc(gfs2_qadata_cachep, GFP_NOFS);
543 		if (!ip->i_qadata) {
544 			error = -ENOMEM;
545 			goto out;
546 		}
547 	}
548 	ip->i_qadata->qa_ref++;
549 out:
550 	up_write(&ip->i_rw_mutex);
551 	return error;
552 }
553 
554 void gfs2_qa_put(struct gfs2_inode *ip)
555 {
556 	down_write(&ip->i_rw_mutex);
557 	if (ip->i_qadata && --ip->i_qadata->qa_ref == 0) {
558 		kmem_cache_free(gfs2_qadata_cachep, ip->i_qadata);
559 		ip->i_qadata = NULL;
560 	}
561 	up_write(&ip->i_rw_mutex);
562 }
563 
564 int gfs2_quota_hold(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
565 {
566 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
567 	struct gfs2_quota_data **qd;
568 	int error;
569 
570 	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
571 		return 0;
572 
573 	error = gfs2_qa_get(ip);
574 	if (error)
575 		return error;
576 
577 	qd = ip->i_qadata->qa_qd;
578 
579 	if (gfs2_assert_warn(sdp, !ip->i_qadata->qa_qd_num) ||
580 	    gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags))) {
581 		error = -EIO;
582 		goto out;
583 	}
584 
585 	error = qdsb_get(sdp, make_kqid_uid(ip->i_inode.i_uid), qd);
586 	if (error)
587 		goto out_unhold;
588 	ip->i_qadata->qa_qd_num++;
589 	qd++;
590 
591 	error = qdsb_get(sdp, make_kqid_gid(ip->i_inode.i_gid), qd);
592 	if (error)
593 		goto out_unhold;
594 	ip->i_qadata->qa_qd_num++;
595 	qd++;
596 
597 	if (!uid_eq(uid, NO_UID_QUOTA_CHANGE) &&
598 	    !uid_eq(uid, ip->i_inode.i_uid)) {
599 		error = qdsb_get(sdp, make_kqid_uid(uid), qd);
600 		if (error)
601 			goto out_unhold;
602 		ip->i_qadata->qa_qd_num++;
603 		qd++;
604 	}
605 
606 	if (!gid_eq(gid, NO_GID_QUOTA_CHANGE) &&
607 	    !gid_eq(gid, ip->i_inode.i_gid)) {
608 		error = qdsb_get(sdp, make_kqid_gid(gid), qd);
609 		if (error)
610 			goto out_unhold;
611 		ip->i_qadata->qa_qd_num++;
612 		qd++;
613 	}
614 
615 out_unhold:
616 	if (error)
617 		gfs2_quota_unhold(ip);
618 out:
619 	return error;
620 }
621 
622 void gfs2_quota_unhold(struct gfs2_inode *ip)
623 {
624 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
625 	u32 x;
626 
627 	if (ip->i_qadata == NULL)
628 		return;
629 
630 	gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags));
631 
632 	for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
633 		qdsb_put(ip->i_qadata->qa_qd[x]);
634 		ip->i_qadata->qa_qd[x] = NULL;
635 	}
636 	ip->i_qadata->qa_qd_num = 0;
637 	gfs2_qa_put(ip);
638 }
639 
640 static int sort_qd(const void *a, const void *b)
641 {
642 	const struct gfs2_quota_data *qd_a = *(const struct gfs2_quota_data **)a;
643 	const struct gfs2_quota_data *qd_b = *(const struct gfs2_quota_data **)b;
644 
645 	if (qid_lt(qd_a->qd_id, qd_b->qd_id))
646 		return -1;
647 	if (qid_lt(qd_b->qd_id, qd_a->qd_id))
648 		return 1;
649 	return 0;
650 }
651 
652 static void do_qc(struct gfs2_quota_data *qd, s64 change)
653 {
654 	struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
655 	struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
656 	struct gfs2_quota_change *qc = qd->qd_bh_qc;
657 	s64 x;
658 
659 	mutex_lock(&sdp->sd_quota_mutex);
660 	gfs2_trans_add_meta(ip->i_gl, qd->qd_bh);
661 
662 	if (!test_bit(QDF_CHANGE, &qd->qd_flags)) {
663 		qc->qc_change = 0;
664 		qc->qc_flags = 0;
665 		if (qd->qd_id.type == USRQUOTA)
666 			qc->qc_flags = cpu_to_be32(GFS2_QCF_USER);
667 		qc->qc_id = cpu_to_be32(from_kqid(&init_user_ns, qd->qd_id));
668 	}
669 
670 	x = be64_to_cpu(qc->qc_change) + change;
671 	qc->qc_change = cpu_to_be64(x);
672 
673 	spin_lock(&qd_lock);
674 	qd->qd_change = x;
675 	spin_unlock(&qd_lock);
676 
677 	if (!x) {
678 		gfs2_assert_warn(sdp, test_bit(QDF_CHANGE, &qd->qd_flags));
679 		clear_bit(QDF_CHANGE, &qd->qd_flags);
680 		qc->qc_flags = 0;
681 		qc->qc_id = 0;
682 		slot_put(qd);
683 		qd_put(qd);
684 	} else if (!test_and_set_bit(QDF_CHANGE, &qd->qd_flags)) {
685 		qd_hold(qd);
686 		slot_hold(qd);
687 	}
688 
689 	if (change < 0) /* Reset quiet flag if we freed some blocks */
690 		clear_bit(QDF_QMSG_QUIET, &qd->qd_flags);
691 	mutex_unlock(&sdp->sd_quota_mutex);
692 }
693 
694 static int gfs2_write_buf_to_page(struct gfs2_inode *ip, unsigned long index,
695 				  unsigned off, void *buf, unsigned bytes)
696 {
697 	struct inode *inode = &ip->i_inode;
698 	struct gfs2_sbd *sdp = GFS2_SB(inode);
699 	struct address_space *mapping = inode->i_mapping;
700 	struct page *page;
701 	struct buffer_head *bh;
702 	void *kaddr;
703 	u64 blk;
704 	unsigned bsize = sdp->sd_sb.sb_bsize, bnum = 0, boff = 0;
705 	unsigned to_write = bytes, pg_off = off;
706 	int done = 0;
707 
708 	blk = index << (PAGE_SHIFT - sdp->sd_sb.sb_bsize_shift);
709 	boff = off % bsize;
710 
711 	page = find_or_create_page(mapping, index, GFP_NOFS);
712 	if (!page)
713 		return -ENOMEM;
714 	if (!page_has_buffers(page))
715 		create_empty_buffers(page, bsize, 0);
716 
717 	bh = page_buffers(page);
718 	while (!done) {
719 		/* Find the beginning block within the page */
720 		if (pg_off >= ((bnum * bsize) + bsize)) {
721 			bh = bh->b_this_page;
722 			bnum++;
723 			blk++;
724 			continue;
725 		}
726 		if (!buffer_mapped(bh)) {
727 			gfs2_block_map(inode, blk, bh, 1);
728 			if (!buffer_mapped(bh))
729 				goto unlock_out;
730 			/* If it's a newly allocated disk block, zero it */
731 			if (buffer_new(bh))
732 				zero_user(page, bnum * bsize, bh->b_size);
733 		}
734 		if (PageUptodate(page))
735 			set_buffer_uptodate(bh);
736 		if (!buffer_uptodate(bh)) {
737 			ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &bh);
738 			wait_on_buffer(bh);
739 			if (!buffer_uptodate(bh))
740 				goto unlock_out;
741 		}
742 		if (gfs2_is_jdata(ip))
743 			gfs2_trans_add_data(ip->i_gl, bh);
744 		else
745 			gfs2_ordered_add_inode(ip);
746 
747 		/* If we need to write to the next block as well */
748 		if (to_write > (bsize - boff)) {
749 			pg_off += (bsize - boff);
750 			to_write -= (bsize - boff);
751 			boff = pg_off % bsize;
752 			continue;
753 		}
754 		done = 1;
755 	}
756 
757 	/* Write to the page, now that we have setup the buffer(s) */
758 	kaddr = kmap_atomic(page);
759 	memcpy(kaddr + off, buf, bytes);
760 	flush_dcache_page(page);
761 	kunmap_atomic(kaddr);
762 	unlock_page(page);
763 	put_page(page);
764 
765 	return 0;
766 
767 unlock_out:
768 	unlock_page(page);
769 	put_page(page);
770 	return -EIO;
771 }
772 
773 static int gfs2_write_disk_quota(struct gfs2_inode *ip, struct gfs2_quota *qp,
774 				 loff_t loc)
775 {
776 	unsigned long pg_beg;
777 	unsigned pg_off, nbytes, overflow = 0;
778 	int pg_oflow = 0, error;
779 	void *ptr;
780 
781 	nbytes = sizeof(struct gfs2_quota);
782 
783 	pg_beg = loc >> PAGE_SHIFT;
784 	pg_off = offset_in_page(loc);
785 
786 	/* If the quota straddles a page boundary, split the write in two */
787 	if ((pg_off + nbytes) > PAGE_SIZE) {
788 		pg_oflow = 1;
789 		overflow = (pg_off + nbytes) - PAGE_SIZE;
790 	}
791 
792 	ptr = qp;
793 	error = gfs2_write_buf_to_page(ip, pg_beg, pg_off, ptr,
794 				       nbytes - overflow);
795 	/* If there's an overflow, write the remaining bytes to the next page */
796 	if (!error && pg_oflow)
797 		error = gfs2_write_buf_to_page(ip, pg_beg + 1, 0,
798 					       ptr + nbytes - overflow,
799 					       overflow);
800 	return error;
801 }
802 
803 /**
804  * gfs2_adjust_quota - adjust record of current block usage
805  * @ip: The quota inode
806  * @loc: Offset of the entry in the quota file
807  * @change: The amount of usage change to record
808  * @qd: The quota data
809  * @fdq: The updated limits to record
810  *
811  * This function was mostly borrowed from gfs2_block_truncate_page which was
812  * in turn mostly borrowed from ext3
813  *
814  * Returns: 0 or -ve on error
815  */
816 
817 static int gfs2_adjust_quota(struct gfs2_inode *ip, loff_t loc,
818 			     s64 change, struct gfs2_quota_data *qd,
819 			     struct qc_dqblk *fdq)
820 {
821 	struct inode *inode = &ip->i_inode;
822 	struct gfs2_sbd *sdp = GFS2_SB(inode);
823 	struct gfs2_quota q;
824 	int err;
825 	u64 size;
826 
827 	if (gfs2_is_stuffed(ip)) {
828 		err = gfs2_unstuff_dinode(ip, NULL);
829 		if (err)
830 			return err;
831 	}
832 
833 	memset(&q, 0, sizeof(struct gfs2_quota));
834 	err = gfs2_internal_read(ip, (char *)&q, &loc, sizeof(q));
835 	if (err < 0)
836 		return err;
837 
838 	loc -= sizeof(q); /* gfs2_internal_read would've advanced the loc ptr */
839 	err = -EIO;
840 	be64_add_cpu(&q.qu_value, change);
841 	if (((s64)be64_to_cpu(q.qu_value)) < 0)
842 		q.qu_value = 0; /* Never go negative on quota usage */
843 	qd->qd_qb.qb_value = q.qu_value;
844 	if (fdq) {
845 		if (fdq->d_fieldmask & QC_SPC_SOFT) {
846 			q.qu_warn = cpu_to_be64(fdq->d_spc_softlimit >> sdp->sd_sb.sb_bsize_shift);
847 			qd->qd_qb.qb_warn = q.qu_warn;
848 		}
849 		if (fdq->d_fieldmask & QC_SPC_HARD) {
850 			q.qu_limit = cpu_to_be64(fdq->d_spc_hardlimit >> sdp->sd_sb.sb_bsize_shift);
851 			qd->qd_qb.qb_limit = q.qu_limit;
852 		}
853 		if (fdq->d_fieldmask & QC_SPACE) {
854 			q.qu_value = cpu_to_be64(fdq->d_space >> sdp->sd_sb.sb_bsize_shift);
855 			qd->qd_qb.qb_value = q.qu_value;
856 		}
857 	}
858 
859 	err = gfs2_write_disk_quota(ip, &q, loc);
860 	if (!err) {
861 		size = loc + sizeof(struct gfs2_quota);
862 		if (size > inode->i_size)
863 			i_size_write(inode, size);
864 		inode->i_mtime = inode->i_atime = current_time(inode);
865 		mark_inode_dirty(inode);
866 		set_bit(QDF_REFRESH, &qd->qd_flags);
867 	}
868 
869 	return err;
870 }
871 
872 static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
873 {
874 	struct gfs2_sbd *sdp = (*qda)->qd_gl->gl_name.ln_sbd;
875 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
876 	struct gfs2_alloc_parms ap = { .aflags = 0, };
877 	unsigned int data_blocks, ind_blocks;
878 	struct gfs2_holder *ghs, i_gh;
879 	unsigned int qx, x;
880 	struct gfs2_quota_data *qd;
881 	unsigned reserved;
882 	loff_t offset;
883 	unsigned int nalloc = 0, blocks;
884 	int error;
885 
886 	error = gfs2_qa_get(ip);
887 	if (error)
888 		return error;
889 
890 	gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
891 			      &data_blocks, &ind_blocks);
892 
893 	ghs = kmalloc_array(num_qd, sizeof(struct gfs2_holder), GFP_NOFS);
894 	if (!ghs) {
895 		error = -ENOMEM;
896 		goto out;
897 	}
898 
899 	sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL);
900 	inode_lock(&ip->i_inode);
901 	for (qx = 0; qx < num_qd; qx++) {
902 		error = gfs2_glock_nq_init(qda[qx]->qd_gl, LM_ST_EXCLUSIVE,
903 					   GL_NOCACHE, &ghs[qx]);
904 		if (error)
905 			goto out_dq;
906 	}
907 
908 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
909 	if (error)
910 		goto out_dq;
911 
912 	for (x = 0; x < num_qd; x++) {
913 		offset = qd2offset(qda[x]);
914 		if (gfs2_write_alloc_required(ip, offset,
915 					      sizeof(struct gfs2_quota)))
916 			nalloc++;
917 	}
918 
919 	/*
920 	 * 1 blk for unstuffing inode if stuffed. We add this extra
921 	 * block to the reservation unconditionally. If the inode
922 	 * doesn't need unstuffing, the block will be released to the
923 	 * rgrp since it won't be allocated during the transaction
924 	 */
925 	/* +3 in the end for unstuffing block, inode size update block
926 	 * and another block in case quota straddles page boundary and
927 	 * two blocks need to be updated instead of 1 */
928 	blocks = num_qd * data_blocks + RES_DINODE + num_qd + 3;
929 
930 	reserved = 1 + (nalloc * (data_blocks + ind_blocks));
931 	ap.target = reserved;
932 	error = gfs2_inplace_reserve(ip, &ap);
933 	if (error)
934 		goto out_alloc;
935 
936 	if (nalloc)
937 		blocks += gfs2_rg_blocks(ip, reserved) + nalloc * ind_blocks + RES_STATFS;
938 
939 	error = gfs2_trans_begin(sdp, blocks, 0);
940 	if (error)
941 		goto out_ipres;
942 
943 	for (x = 0; x < num_qd; x++) {
944 		qd = qda[x];
945 		offset = qd2offset(qd);
946 		error = gfs2_adjust_quota(ip, offset, qd->qd_change_sync, qd, NULL);
947 		if (error)
948 			goto out_end_trans;
949 
950 		do_qc(qd, -qd->qd_change_sync);
951 		set_bit(QDF_REFRESH, &qd->qd_flags);
952 	}
953 
954 	error = 0;
955 
956 out_end_trans:
957 	gfs2_trans_end(sdp);
958 out_ipres:
959 	gfs2_inplace_release(ip);
960 out_alloc:
961 	gfs2_glock_dq_uninit(&i_gh);
962 out_dq:
963 	while (qx--)
964 		gfs2_glock_dq_uninit(&ghs[qx]);
965 	inode_unlock(&ip->i_inode);
966 	kfree(ghs);
967 	gfs2_log_flush(ip->i_gl->gl_name.ln_sbd, ip->i_gl,
968 		       GFS2_LOG_HEAD_FLUSH_NORMAL | GFS2_LFC_DO_SYNC);
969 out:
970 	gfs2_qa_put(ip);
971 	return error;
972 }
973 
974 static int update_qd(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd)
975 {
976 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
977 	struct gfs2_quota q;
978 	struct gfs2_quota_lvb *qlvb;
979 	loff_t pos;
980 	int error;
981 
982 	memset(&q, 0, sizeof(struct gfs2_quota));
983 	pos = qd2offset(qd);
984 	error = gfs2_internal_read(ip, (char *)&q, &pos, sizeof(q));
985 	if (error < 0)
986 		return error;
987 
988 	qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
989 	qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC);
990 	qlvb->__pad = 0;
991 	qlvb->qb_limit = q.qu_limit;
992 	qlvb->qb_warn = q.qu_warn;
993 	qlvb->qb_value = q.qu_value;
994 	qd->qd_qb = *qlvb;
995 
996 	return 0;
997 }
998 
999 static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
1000 		    struct gfs2_holder *q_gh)
1001 {
1002 	struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
1003 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1004 	struct gfs2_holder i_gh;
1005 	int error;
1006 
1007 restart:
1008 	error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
1009 	if (error)
1010 		return error;
1011 
1012 	if (test_and_clear_bit(QDF_REFRESH, &qd->qd_flags))
1013 		force_refresh = FORCE;
1014 
1015 	qd->qd_qb = *(struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1016 
1017 	if (force_refresh || qd->qd_qb.qb_magic != cpu_to_be32(GFS2_MAGIC)) {
1018 		gfs2_glock_dq_uninit(q_gh);
1019 		error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE,
1020 					   GL_NOCACHE, q_gh);
1021 		if (error)
1022 			return error;
1023 
1024 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
1025 		if (error)
1026 			goto fail;
1027 
1028 		error = update_qd(sdp, qd);
1029 		if (error)
1030 			goto fail_gunlock;
1031 
1032 		gfs2_glock_dq_uninit(&i_gh);
1033 		gfs2_glock_dq_uninit(q_gh);
1034 		force_refresh = 0;
1035 		goto restart;
1036 	}
1037 
1038 	return 0;
1039 
1040 fail_gunlock:
1041 	gfs2_glock_dq_uninit(&i_gh);
1042 fail:
1043 	gfs2_glock_dq_uninit(q_gh);
1044 	return error;
1045 }
1046 
1047 int gfs2_quota_lock(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
1048 {
1049 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1050 	struct gfs2_quota_data *qd;
1051 	u32 x;
1052 	int error = 0;
1053 
1054 	if (capable(CAP_SYS_RESOURCE) ||
1055 	    sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
1056 		return 0;
1057 
1058 	error = gfs2_quota_hold(ip, uid, gid);
1059 	if (error)
1060 		return error;
1061 
1062 	sort(ip->i_qadata->qa_qd, ip->i_qadata->qa_qd_num,
1063 	     sizeof(struct gfs2_quota_data *), sort_qd, NULL);
1064 
1065 	for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1066 		qd = ip->i_qadata->qa_qd[x];
1067 		error = do_glock(qd, NO_FORCE, &ip->i_qadata->qa_qd_ghs[x]);
1068 		if (error)
1069 			break;
1070 	}
1071 
1072 	if (!error)
1073 		set_bit(GIF_QD_LOCKED, &ip->i_flags);
1074 	else {
1075 		while (x--)
1076 			gfs2_glock_dq_uninit(&ip->i_qadata->qa_qd_ghs[x]);
1077 		gfs2_quota_unhold(ip);
1078 	}
1079 
1080 	return error;
1081 }
1082 
1083 static int need_sync(struct gfs2_quota_data *qd)
1084 {
1085 	struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
1086 	struct gfs2_tune *gt = &sdp->sd_tune;
1087 	s64 value;
1088 	unsigned int num, den;
1089 	int do_sync = 1;
1090 
1091 	if (!qd->qd_qb.qb_limit)
1092 		return 0;
1093 
1094 	spin_lock(&qd_lock);
1095 	value = qd->qd_change;
1096 	spin_unlock(&qd_lock);
1097 
1098 	spin_lock(&gt->gt_spin);
1099 	num = gt->gt_quota_scale_num;
1100 	den = gt->gt_quota_scale_den;
1101 	spin_unlock(&gt->gt_spin);
1102 
1103 	if (value < 0)
1104 		do_sync = 0;
1105 	else if ((s64)be64_to_cpu(qd->qd_qb.qb_value) >=
1106 		 (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1107 		do_sync = 0;
1108 	else {
1109 		value *= gfs2_jindex_size(sdp) * num;
1110 		value = div_s64(value, den);
1111 		value += (s64)be64_to_cpu(qd->qd_qb.qb_value);
1112 		if (value < (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1113 			do_sync = 0;
1114 	}
1115 
1116 	return do_sync;
1117 }
1118 
1119 void gfs2_quota_unlock(struct gfs2_inode *ip)
1120 {
1121 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1122 	struct gfs2_quota_data *qda[4];
1123 	unsigned int count = 0;
1124 	u32 x;
1125 	int found;
1126 
1127 	if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
1128 		goto out;
1129 
1130 	for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1131 		struct gfs2_quota_data *qd;
1132 		int sync;
1133 
1134 		qd = ip->i_qadata->qa_qd[x];
1135 		sync = need_sync(qd);
1136 
1137 		gfs2_glock_dq_uninit(&ip->i_qadata->qa_qd_ghs[x]);
1138 		if (!sync)
1139 			continue;
1140 
1141 		spin_lock(&qd_lock);
1142 		found = qd_check_sync(sdp, qd, NULL);
1143 		spin_unlock(&qd_lock);
1144 
1145 		if (!found)
1146 			continue;
1147 
1148 		gfs2_assert_warn(sdp, qd->qd_change_sync);
1149 		if (bh_get(qd)) {
1150 			clear_bit(QDF_LOCKED, &qd->qd_flags);
1151 			slot_put(qd);
1152 			qd_put(qd);
1153 			continue;
1154 		}
1155 
1156 		qda[count++] = qd;
1157 	}
1158 
1159 	if (count) {
1160 		do_sync(count, qda);
1161 		for (x = 0; x < count; x++)
1162 			qd_unlock(qda[x]);
1163 	}
1164 
1165 out:
1166 	gfs2_quota_unhold(ip);
1167 }
1168 
1169 #define MAX_LINE 256
1170 
1171 static int print_message(struct gfs2_quota_data *qd, char *type)
1172 {
1173 	struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
1174 
1175 	fs_info(sdp, "quota %s for %s %u\n",
1176 		type,
1177 		(qd->qd_id.type == USRQUOTA) ? "user" : "group",
1178 		from_kqid(&init_user_ns, qd->qd_id));
1179 
1180 	return 0;
1181 }
1182 
1183 /**
1184  * gfs2_quota_check - check if allocating new blocks will exceed quota
1185  * @ip:  The inode for which this check is being performed
1186  * @uid: The uid to check against
1187  * @gid: The gid to check against
1188  * @ap:  The allocation parameters. ap->target contains the requested
1189  *       blocks. ap->min_target, if set, contains the minimum blks
1190  *       requested.
1191  *
1192  * Returns: 0 on success.
1193  *                  min_req = ap->min_target ? ap->min_target : ap->target;
1194  *                  quota must allow at least min_req blks for success and
1195  *                  ap->allowed is set to the number of blocks allowed
1196  *
1197  *          -EDQUOT otherwise, quota violation. ap->allowed is set to number
1198  *                  of blocks available.
1199  */
1200 int gfs2_quota_check(struct gfs2_inode *ip, kuid_t uid, kgid_t gid,
1201 		     struct gfs2_alloc_parms *ap)
1202 {
1203 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1204 	struct gfs2_quota_data *qd;
1205 	s64 value, warn, limit;
1206 	u32 x;
1207 	int error = 0;
1208 
1209 	ap->allowed = UINT_MAX; /* Assume we are permitted a whole lot */
1210 	if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
1211 		return 0;
1212 
1213         if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
1214                 return 0;
1215 
1216 	for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1217 		qd = ip->i_qadata->qa_qd[x];
1218 
1219 		if (!(qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1220 		      qid_eq(qd->qd_id, make_kqid_gid(gid))))
1221 			continue;
1222 
1223 		warn = (s64)be64_to_cpu(qd->qd_qb.qb_warn);
1224 		limit = (s64)be64_to_cpu(qd->qd_qb.qb_limit);
1225 		value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
1226 		spin_lock(&qd_lock);
1227 		value += qd->qd_change;
1228 		spin_unlock(&qd_lock);
1229 
1230 		if (limit > 0 && (limit - value) < ap->allowed)
1231 			ap->allowed = limit - value;
1232 		/* If we can't meet the target */
1233 		if (limit && limit < (value + (s64)ap->target)) {
1234 			/* If no min_target specified or we don't meet
1235 			 * min_target, return -EDQUOT */
1236 			if (!ap->min_target || ap->min_target > ap->allowed) {
1237 				if (!test_and_set_bit(QDF_QMSG_QUIET,
1238 						      &qd->qd_flags)) {
1239 					print_message(qd, "exceeded");
1240 					quota_send_warning(qd->qd_id,
1241 							   sdp->sd_vfs->s_dev,
1242 							   QUOTA_NL_BHARDWARN);
1243 				}
1244 				error = -EDQUOT;
1245 				break;
1246 			}
1247 		} else if (warn && warn < value &&
1248 			   time_after_eq(jiffies, qd->qd_last_warn +
1249 					 gfs2_tune_get(sdp, gt_quota_warn_period)
1250 					 * HZ)) {
1251 			quota_send_warning(qd->qd_id,
1252 					   sdp->sd_vfs->s_dev, QUOTA_NL_BSOFTWARN);
1253 			error = print_message(qd, "warning");
1254 			qd->qd_last_warn = jiffies;
1255 		}
1256 	}
1257 	return error;
1258 }
1259 
1260 void gfs2_quota_change(struct gfs2_inode *ip, s64 change,
1261 		       kuid_t uid, kgid_t gid)
1262 {
1263 	struct gfs2_quota_data *qd;
1264 	u32 x;
1265 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1266 
1267 	if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON ||
1268 	    gfs2_assert_warn(sdp, change))
1269 		return;
1270 	if (ip->i_diskflags & GFS2_DIF_SYSTEM)
1271 		return;
1272 
1273 	BUG_ON(ip->i_qadata->qa_ref <= 0);
1274 	for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1275 		qd = ip->i_qadata->qa_qd[x];
1276 
1277 		if (qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1278 		    qid_eq(qd->qd_id, make_kqid_gid(gid))) {
1279 			do_qc(qd, change);
1280 		}
1281 	}
1282 }
1283 
1284 int gfs2_quota_sync(struct super_block *sb, int type)
1285 {
1286 	struct gfs2_sbd *sdp = sb->s_fs_info;
1287 	struct gfs2_quota_data **qda;
1288 	unsigned int max_qd = PAGE_SIZE / sizeof(struct gfs2_holder);
1289 	unsigned int num_qd;
1290 	unsigned int x;
1291 	int error = 0;
1292 
1293 	qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
1294 	if (!qda)
1295 		return -ENOMEM;
1296 
1297 	mutex_lock(&sdp->sd_quota_sync_mutex);
1298 	sdp->sd_quota_sync_gen++;
1299 
1300 	do {
1301 		num_qd = 0;
1302 
1303 		for (;;) {
1304 			error = qd_fish(sdp, qda + num_qd);
1305 			if (error || !qda[num_qd])
1306 				break;
1307 			if (++num_qd == max_qd)
1308 				break;
1309 		}
1310 
1311 		if (num_qd) {
1312 			if (!error)
1313 				error = do_sync(num_qd, qda);
1314 			if (!error)
1315 				for (x = 0; x < num_qd; x++)
1316 					qda[x]->qd_sync_gen =
1317 						sdp->sd_quota_sync_gen;
1318 
1319 			for (x = 0; x < num_qd; x++)
1320 				qd_unlock(qda[x]);
1321 		}
1322 	} while (!error && num_qd == max_qd);
1323 
1324 	mutex_unlock(&sdp->sd_quota_sync_mutex);
1325 	kfree(qda);
1326 
1327 	return error;
1328 }
1329 
1330 int gfs2_quota_refresh(struct gfs2_sbd *sdp, struct kqid qid)
1331 {
1332 	struct gfs2_quota_data *qd;
1333 	struct gfs2_holder q_gh;
1334 	int error;
1335 
1336 	error = qd_get(sdp, qid, &qd);
1337 	if (error)
1338 		return error;
1339 
1340 	error = do_glock(qd, FORCE, &q_gh);
1341 	if (!error)
1342 		gfs2_glock_dq_uninit(&q_gh);
1343 
1344 	qd_put(qd);
1345 	return error;
1346 }
1347 
1348 int gfs2_quota_init(struct gfs2_sbd *sdp)
1349 {
1350 	struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
1351 	u64 size = i_size_read(sdp->sd_qc_inode);
1352 	unsigned int blocks = size >> sdp->sd_sb.sb_bsize_shift;
1353 	unsigned int x, slot = 0;
1354 	unsigned int found = 0;
1355 	unsigned int hash;
1356 	unsigned int bm_size;
1357 	u64 dblock;
1358 	u32 extlen = 0;
1359 	int error;
1360 
1361 	if (gfs2_check_internal_file_size(sdp->sd_qc_inode, 1, 64 << 20))
1362 		return -EIO;
1363 
1364 	sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
1365 	bm_size = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * sizeof(unsigned long));
1366 	bm_size *= sizeof(unsigned long);
1367 	error = -ENOMEM;
1368 	sdp->sd_quota_bitmap = kzalloc(bm_size, GFP_NOFS | __GFP_NOWARN);
1369 	if (sdp->sd_quota_bitmap == NULL)
1370 		sdp->sd_quota_bitmap = __vmalloc(bm_size, GFP_NOFS |
1371 						 __GFP_ZERO, PAGE_KERNEL);
1372 	if (!sdp->sd_quota_bitmap)
1373 		return error;
1374 
1375 	for (x = 0; x < blocks; x++) {
1376 		struct buffer_head *bh;
1377 		const struct gfs2_quota_change *qc;
1378 		unsigned int y;
1379 
1380 		if (!extlen) {
1381 			int new = 0;
1382 			error = gfs2_extent_map(&ip->i_inode, x, &new, &dblock, &extlen);
1383 			if (error)
1384 				goto fail;
1385 		}
1386 		error = -EIO;
1387 		bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
1388 		if (!bh)
1389 			goto fail;
1390 		if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
1391 			brelse(bh);
1392 			goto fail;
1393 		}
1394 
1395 		qc = (const struct gfs2_quota_change *)(bh->b_data + sizeof(struct gfs2_meta_header));
1396 		for (y = 0; y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1397 		     y++, slot++) {
1398 			struct gfs2_quota_data *qd;
1399 			s64 qc_change = be64_to_cpu(qc->qc_change);
1400 			u32 qc_flags = be32_to_cpu(qc->qc_flags);
1401 			enum quota_type qtype = (qc_flags & GFS2_QCF_USER) ?
1402 						USRQUOTA : GRPQUOTA;
1403 			struct kqid qc_id = make_kqid(&init_user_ns, qtype,
1404 						      be32_to_cpu(qc->qc_id));
1405 			qc++;
1406 			if (!qc_change)
1407 				continue;
1408 
1409 			hash = gfs2_qd_hash(sdp, qc_id);
1410 			qd = qd_alloc(hash, sdp, qc_id);
1411 			if (qd == NULL) {
1412 				brelse(bh);
1413 				goto fail;
1414 			}
1415 
1416 			set_bit(QDF_CHANGE, &qd->qd_flags);
1417 			qd->qd_change = qc_change;
1418 			qd->qd_slot = slot;
1419 			qd->qd_slot_count = 1;
1420 
1421 			spin_lock(&qd_lock);
1422 			BUG_ON(test_and_set_bit(slot, sdp->sd_quota_bitmap));
1423 			list_add(&qd->qd_list, &sdp->sd_quota_list);
1424 			atomic_inc(&sdp->sd_quota_count);
1425 			spin_unlock(&qd_lock);
1426 
1427 			spin_lock_bucket(hash);
1428 			hlist_bl_add_head_rcu(&qd->qd_hlist, &qd_hash_table[hash]);
1429 			spin_unlock_bucket(hash);
1430 
1431 			found++;
1432 		}
1433 
1434 		brelse(bh);
1435 		dblock++;
1436 		extlen--;
1437 	}
1438 
1439 	if (found)
1440 		fs_info(sdp, "found %u quota changes\n", found);
1441 
1442 	return 0;
1443 
1444 fail:
1445 	gfs2_quota_cleanup(sdp);
1446 	return error;
1447 }
1448 
1449 void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1450 {
1451 	struct list_head *head = &sdp->sd_quota_list;
1452 	struct gfs2_quota_data *qd;
1453 
1454 	spin_lock(&qd_lock);
1455 	while (!list_empty(head)) {
1456 		qd = list_last_entry(head, struct gfs2_quota_data, qd_list);
1457 
1458 		list_del(&qd->qd_list);
1459 
1460 		/* Also remove if this qd exists in the reclaim list */
1461 		list_lru_del(&gfs2_qd_lru, &qd->qd_lru);
1462 		atomic_dec(&sdp->sd_quota_count);
1463 		spin_unlock(&qd_lock);
1464 
1465 		spin_lock_bucket(qd->qd_hash);
1466 		hlist_bl_del_rcu(&qd->qd_hlist);
1467 		spin_unlock_bucket(qd->qd_hash);
1468 
1469 		gfs2_assert_warn(sdp, !qd->qd_change);
1470 		gfs2_assert_warn(sdp, !qd->qd_slot_count);
1471 		gfs2_assert_warn(sdp, !qd->qd_bh_count);
1472 
1473 		gfs2_glock_put(qd->qd_gl);
1474 		call_rcu(&qd->qd_rcu, gfs2_qd_dealloc);
1475 
1476 		spin_lock(&qd_lock);
1477 	}
1478 	spin_unlock(&qd_lock);
1479 
1480 	gfs2_assert_warn(sdp, !atomic_read(&sdp->sd_quota_count));
1481 
1482 	kvfree(sdp->sd_quota_bitmap);
1483 	sdp->sd_quota_bitmap = NULL;
1484 }
1485 
1486 static void quotad_error(struct gfs2_sbd *sdp, const char *msg, int error)
1487 {
1488 	if (error == 0 || error == -EROFS)
1489 		return;
1490 	if (!gfs2_withdrawn(sdp)) {
1491 		if (!cmpxchg(&sdp->sd_log_error, 0, error))
1492 			fs_err(sdp, "gfs2_quotad: %s error %d\n", msg, error);
1493 		wake_up(&sdp->sd_logd_waitq);
1494 	}
1495 }
1496 
1497 static void quotad_check_timeo(struct gfs2_sbd *sdp, const char *msg,
1498 			       int (*fxn)(struct super_block *sb, int type),
1499 			       unsigned long t, unsigned long *timeo,
1500 			       unsigned int *new_timeo)
1501 {
1502 	if (t >= *timeo) {
1503 		int error = fxn(sdp->sd_vfs, 0);
1504 		quotad_error(sdp, msg, error);
1505 		*timeo = gfs2_tune_get_i(&sdp->sd_tune, new_timeo) * HZ;
1506 	} else {
1507 		*timeo -= t;
1508 	}
1509 }
1510 
1511 static void quotad_check_trunc_list(struct gfs2_sbd *sdp)
1512 {
1513 	struct gfs2_inode *ip;
1514 
1515 	while(1) {
1516 		ip = NULL;
1517 		spin_lock(&sdp->sd_trunc_lock);
1518 		if (!list_empty(&sdp->sd_trunc_list)) {
1519 			ip = list_first_entry(&sdp->sd_trunc_list,
1520 					struct gfs2_inode, i_trunc_list);
1521 			list_del_init(&ip->i_trunc_list);
1522 		}
1523 		spin_unlock(&sdp->sd_trunc_lock);
1524 		if (ip == NULL)
1525 			return;
1526 		gfs2_glock_finish_truncate(ip);
1527 	}
1528 }
1529 
1530 void gfs2_wake_up_statfs(struct gfs2_sbd *sdp) {
1531 	if (!sdp->sd_statfs_force_sync) {
1532 		sdp->sd_statfs_force_sync = 1;
1533 		wake_up(&sdp->sd_quota_wait);
1534 	}
1535 }
1536 
1537 
1538 /**
1539  * gfs2_quotad - Write cached quota changes into the quota file
1540  * @sdp: Pointer to GFS2 superblock
1541  *
1542  */
1543 
1544 int gfs2_quotad(void *data)
1545 {
1546 	struct gfs2_sbd *sdp = data;
1547 	struct gfs2_tune *tune = &sdp->sd_tune;
1548 	unsigned long statfs_timeo = 0;
1549 	unsigned long quotad_timeo = 0;
1550 	unsigned long t = 0;
1551 	DEFINE_WAIT(wait);
1552 	int empty;
1553 
1554 	while (!kthread_should_stop()) {
1555 
1556 		if (gfs2_withdrawn(sdp))
1557 			goto bypass;
1558 		/* Update the master statfs file */
1559 		if (sdp->sd_statfs_force_sync) {
1560 			int error = gfs2_statfs_sync(sdp->sd_vfs, 0);
1561 			quotad_error(sdp, "statfs", error);
1562 			statfs_timeo = gfs2_tune_get(sdp, gt_statfs_quantum) * HZ;
1563 		}
1564 		else
1565 			quotad_check_timeo(sdp, "statfs", gfs2_statfs_sync, t,
1566 				   	   &statfs_timeo,
1567 					   &tune->gt_statfs_quantum);
1568 
1569 		/* Update quota file */
1570 		quotad_check_timeo(sdp, "sync", gfs2_quota_sync, t,
1571 				   &quotad_timeo, &tune->gt_quota_quantum);
1572 
1573 		/* Check for & recover partially truncated inodes */
1574 		quotad_check_trunc_list(sdp);
1575 
1576 		try_to_freeze();
1577 
1578 bypass:
1579 		t = min(quotad_timeo, statfs_timeo);
1580 
1581 		prepare_to_wait(&sdp->sd_quota_wait, &wait, TASK_INTERRUPTIBLE);
1582 		spin_lock(&sdp->sd_trunc_lock);
1583 		empty = list_empty(&sdp->sd_trunc_list);
1584 		spin_unlock(&sdp->sd_trunc_lock);
1585 		if (empty && !sdp->sd_statfs_force_sync)
1586 			t -= schedule_timeout(t);
1587 		else
1588 			t = 0;
1589 		finish_wait(&sdp->sd_quota_wait, &wait);
1590 	}
1591 
1592 	return 0;
1593 }
1594 
1595 static int gfs2_quota_get_state(struct super_block *sb, struct qc_state *state)
1596 {
1597 	struct gfs2_sbd *sdp = sb->s_fs_info;
1598 
1599 	memset(state, 0, sizeof(*state));
1600 
1601 	switch (sdp->sd_args.ar_quota) {
1602 	case GFS2_QUOTA_ON:
1603 		state->s_state[USRQUOTA].flags |= QCI_LIMITS_ENFORCED;
1604 		state->s_state[GRPQUOTA].flags |= QCI_LIMITS_ENFORCED;
1605 		/*FALLTHRU*/
1606 	case GFS2_QUOTA_ACCOUNT:
1607 		state->s_state[USRQUOTA].flags |= QCI_ACCT_ENABLED |
1608 						  QCI_SYSFILE;
1609 		state->s_state[GRPQUOTA].flags |= QCI_ACCT_ENABLED |
1610 						  QCI_SYSFILE;
1611 		break;
1612 	case GFS2_QUOTA_OFF:
1613 		break;
1614 	}
1615 	if (sdp->sd_quota_inode) {
1616 		state->s_state[USRQUOTA].ino =
1617 					GFS2_I(sdp->sd_quota_inode)->i_no_addr;
1618 		state->s_state[USRQUOTA].blocks = sdp->sd_quota_inode->i_blocks;
1619 	}
1620 	state->s_state[USRQUOTA].nextents = 1;	/* unsupported */
1621 	state->s_state[GRPQUOTA] = state->s_state[USRQUOTA];
1622 	state->s_incoredqs = list_lru_count(&gfs2_qd_lru);
1623 	return 0;
1624 }
1625 
1626 static int gfs2_get_dqblk(struct super_block *sb, struct kqid qid,
1627 			  struct qc_dqblk *fdq)
1628 {
1629 	struct gfs2_sbd *sdp = sb->s_fs_info;
1630 	struct gfs2_quota_lvb *qlvb;
1631 	struct gfs2_quota_data *qd;
1632 	struct gfs2_holder q_gh;
1633 	int error;
1634 
1635 	memset(fdq, 0, sizeof(*fdq));
1636 
1637 	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1638 		return -ESRCH; /* Crazy XFS error code */
1639 
1640 	if ((qid.type != USRQUOTA) &&
1641 	    (qid.type != GRPQUOTA))
1642 		return -EINVAL;
1643 
1644 	error = qd_get(sdp, qid, &qd);
1645 	if (error)
1646 		return error;
1647 	error = do_glock(qd, FORCE, &q_gh);
1648 	if (error)
1649 		goto out;
1650 
1651 	qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1652 	fdq->d_spc_hardlimit = be64_to_cpu(qlvb->qb_limit) << sdp->sd_sb.sb_bsize_shift;
1653 	fdq->d_spc_softlimit = be64_to_cpu(qlvb->qb_warn) << sdp->sd_sb.sb_bsize_shift;
1654 	fdq->d_space = be64_to_cpu(qlvb->qb_value) << sdp->sd_sb.sb_bsize_shift;
1655 
1656 	gfs2_glock_dq_uninit(&q_gh);
1657 out:
1658 	qd_put(qd);
1659 	return error;
1660 }
1661 
1662 /* GFS2 only supports a subset of the XFS fields */
1663 #define GFS2_FIELDMASK (QC_SPC_SOFT|QC_SPC_HARD|QC_SPACE)
1664 
1665 static int gfs2_set_dqblk(struct super_block *sb, struct kqid qid,
1666 			  struct qc_dqblk *fdq)
1667 {
1668 	struct gfs2_sbd *sdp = sb->s_fs_info;
1669 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1670 	struct gfs2_quota_data *qd;
1671 	struct gfs2_holder q_gh, i_gh;
1672 	unsigned int data_blocks, ind_blocks;
1673 	unsigned int blocks = 0;
1674 	int alloc_required;
1675 	loff_t offset;
1676 	int error;
1677 
1678 	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1679 		return -ESRCH; /* Crazy XFS error code */
1680 
1681 	if ((qid.type != USRQUOTA) &&
1682 	    (qid.type != GRPQUOTA))
1683 		return -EINVAL;
1684 
1685 	if (fdq->d_fieldmask & ~GFS2_FIELDMASK)
1686 		return -EINVAL;
1687 
1688 	error = qd_get(sdp, qid, &qd);
1689 	if (error)
1690 		return error;
1691 
1692 	error = gfs2_qa_get(ip);
1693 	if (error)
1694 		goto out_put;
1695 
1696 	inode_lock(&ip->i_inode);
1697 	error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE, 0, &q_gh);
1698 	if (error)
1699 		goto out_unlockput;
1700 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1701 	if (error)
1702 		goto out_q;
1703 
1704 	/* Check for existing entry, if none then alloc new blocks */
1705 	error = update_qd(sdp, qd);
1706 	if (error)
1707 		goto out_i;
1708 
1709 	/* If nothing has changed, this is a no-op */
1710 	if ((fdq->d_fieldmask & QC_SPC_SOFT) &&
1711 	    ((fdq->d_spc_softlimit >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_warn)))
1712 		fdq->d_fieldmask ^= QC_SPC_SOFT;
1713 
1714 	if ((fdq->d_fieldmask & QC_SPC_HARD) &&
1715 	    ((fdq->d_spc_hardlimit >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_limit)))
1716 		fdq->d_fieldmask ^= QC_SPC_HARD;
1717 
1718 	if ((fdq->d_fieldmask & QC_SPACE) &&
1719 	    ((fdq->d_space >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_value)))
1720 		fdq->d_fieldmask ^= QC_SPACE;
1721 
1722 	if (fdq->d_fieldmask == 0)
1723 		goto out_i;
1724 
1725 	offset = qd2offset(qd);
1726 	alloc_required = gfs2_write_alloc_required(ip, offset, sizeof(struct gfs2_quota));
1727 	if (gfs2_is_stuffed(ip))
1728 		alloc_required = 1;
1729 	if (alloc_required) {
1730 		struct gfs2_alloc_parms ap = { .aflags = 0, };
1731 		gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
1732 				       &data_blocks, &ind_blocks);
1733 		blocks = 1 + data_blocks + ind_blocks;
1734 		ap.target = blocks;
1735 		error = gfs2_inplace_reserve(ip, &ap);
1736 		if (error)
1737 			goto out_i;
1738 		blocks += gfs2_rg_blocks(ip, blocks);
1739 	}
1740 
1741 	/* Some quotas span block boundaries and can update two blocks,
1742 	   adding an extra block to the transaction to handle such quotas */
1743 	error = gfs2_trans_begin(sdp, blocks + RES_DINODE + 2, 0);
1744 	if (error)
1745 		goto out_release;
1746 
1747 	/* Apply changes */
1748 	error = gfs2_adjust_quota(ip, offset, 0, qd, fdq);
1749 	if (!error)
1750 		clear_bit(QDF_QMSG_QUIET, &qd->qd_flags);
1751 
1752 	gfs2_trans_end(sdp);
1753 out_release:
1754 	if (alloc_required)
1755 		gfs2_inplace_release(ip);
1756 out_i:
1757 	gfs2_glock_dq_uninit(&i_gh);
1758 out_q:
1759 	gfs2_glock_dq_uninit(&q_gh);
1760 out_unlockput:
1761 	gfs2_qa_put(ip);
1762 	inode_unlock(&ip->i_inode);
1763 out_put:
1764 	qd_put(qd);
1765 	return error;
1766 }
1767 
1768 const struct quotactl_ops gfs2_quotactl_ops = {
1769 	.quota_sync     = gfs2_quota_sync,
1770 	.get_state	= gfs2_quota_get_state,
1771 	.get_dqblk	= gfs2_get_dqblk,
1772 	.set_dqblk	= gfs2_set_dqblk,
1773 };
1774 
1775 void __init gfs2_quota_hash_init(void)
1776 {
1777 	unsigned i;
1778 
1779 	for(i = 0; i < GFS2_QD_HASH_SIZE; i++)
1780 		INIT_HLIST_BL_HEAD(&qd_hash_table[i]);
1781 }
1782