xref: /openbmc/linux/fs/gfs2/quota.c (revision 7559e757)
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 #define QC_CHANGE 0
79 #define QC_SYNC 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_first_entry(list, 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 inode *inode = sdp->sd_qc_inode;
372 	struct gfs2_inode *ip = GFS2_I(inode);
373 	unsigned int block, offset;
374 	struct buffer_head *bh;
375 	struct iomap iomap = { };
376 	int error;
377 
378 	mutex_lock(&sdp->sd_quota_mutex);
379 
380 	if (qd->qd_bh_count++) {
381 		mutex_unlock(&sdp->sd_quota_mutex);
382 		return 0;
383 	}
384 
385 	block = qd->qd_slot / sdp->sd_qc_per_block;
386 	offset = qd->qd_slot % sdp->sd_qc_per_block;
387 
388 	error = gfs2_iomap_get(inode,
389 			       (loff_t)block << inode->i_blkbits,
390 			       i_blocksize(inode), &iomap);
391 	if (error)
392 		goto fail;
393 	error = -ENOENT;
394 	if (iomap.type != IOMAP_MAPPED)
395 		goto fail;
396 
397 	error = gfs2_meta_read(ip->i_gl, iomap.addr >> inode->i_blkbits,
398 			       DIO_WAIT, 0, &bh);
399 	if (error)
400 		goto fail;
401 	error = -EIO;
402 	if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC))
403 		goto fail_brelse;
404 
405 	qd->qd_bh = bh;
406 	qd->qd_bh_qc = (struct gfs2_quota_change *)
407 		(bh->b_data + sizeof(struct gfs2_meta_header) +
408 		 offset * sizeof(struct gfs2_quota_change));
409 
410 	mutex_unlock(&sdp->sd_quota_mutex);
411 
412 	return 0;
413 
414 fail_brelse:
415 	brelse(bh);
416 fail:
417 	qd->qd_bh_count--;
418 	mutex_unlock(&sdp->sd_quota_mutex);
419 	return error;
420 }
421 
422 static void bh_put(struct gfs2_quota_data *qd)
423 {
424 	struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
425 
426 	mutex_lock(&sdp->sd_quota_mutex);
427 	gfs2_assert(sdp, qd->qd_bh_count);
428 	if (!--qd->qd_bh_count) {
429 		brelse(qd->qd_bh);
430 		qd->qd_bh = NULL;
431 		qd->qd_bh_qc = NULL;
432 	}
433 	mutex_unlock(&sdp->sd_quota_mutex);
434 }
435 
436 static int qd_check_sync(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd,
437 			 u64 *sync_gen)
438 {
439 	if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
440 	    !test_bit(QDF_CHANGE, &qd->qd_flags) ||
441 	    (sync_gen && (qd->qd_sync_gen >= *sync_gen)))
442 		return 0;
443 
444 	if (!lockref_get_not_dead(&qd->qd_lockref))
445 		return 0;
446 
447 	list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
448 	set_bit(QDF_LOCKED, &qd->qd_flags);
449 	qd->qd_change_sync = qd->qd_change;
450 	slot_hold(qd);
451 	return 1;
452 }
453 
454 static int qd_fish(struct gfs2_sbd *sdp, struct gfs2_quota_data **qdp)
455 {
456 	struct gfs2_quota_data *qd = NULL, *iter;
457 	int error;
458 
459 	*qdp = NULL;
460 
461 	if (sb_rdonly(sdp->sd_vfs))
462 		return 0;
463 
464 	spin_lock(&qd_lock);
465 
466 	list_for_each_entry(iter, &sdp->sd_quota_list, qd_list) {
467 		if (qd_check_sync(sdp, iter, &sdp->sd_quota_sync_gen)) {
468 			qd = iter;
469 			break;
470 		}
471 	}
472 
473 	spin_unlock(&qd_lock);
474 
475 	if (qd) {
476 		error = bh_get(qd);
477 		if (error) {
478 			clear_bit(QDF_LOCKED, &qd->qd_flags);
479 			slot_put(qd);
480 			qd_put(qd);
481 			return error;
482 		}
483 	}
484 
485 	*qdp = qd;
486 
487 	return 0;
488 }
489 
490 static void qd_unlock(struct gfs2_quota_data *qd)
491 {
492 	gfs2_assert_warn(qd->qd_gl->gl_name.ln_sbd,
493 			 test_bit(QDF_LOCKED, &qd->qd_flags));
494 	clear_bit(QDF_LOCKED, &qd->qd_flags);
495 	bh_put(qd);
496 	slot_put(qd);
497 	qd_put(qd);
498 }
499 
500 static int qdsb_get(struct gfs2_sbd *sdp, struct kqid qid,
501 		    struct gfs2_quota_data **qdp)
502 {
503 	int error;
504 
505 	error = qd_get(sdp, qid, qdp);
506 	if (error)
507 		return error;
508 
509 	error = slot_get(*qdp);
510 	if (error)
511 		goto fail;
512 
513 	error = bh_get(*qdp);
514 	if (error)
515 		goto fail_slot;
516 
517 	return 0;
518 
519 fail_slot:
520 	slot_put(*qdp);
521 fail:
522 	qd_put(*qdp);
523 	return error;
524 }
525 
526 static void qdsb_put(struct gfs2_quota_data *qd)
527 {
528 	bh_put(qd);
529 	slot_put(qd);
530 	qd_put(qd);
531 }
532 
533 /**
534  * gfs2_qa_get - make sure we have a quota allocations data structure,
535  *               if necessary
536  * @ip: the inode for this reservation
537  */
538 int gfs2_qa_get(struct gfs2_inode *ip)
539 {
540 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
541 	struct inode *inode = &ip->i_inode;
542 
543 	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
544 		return 0;
545 
546 	spin_lock(&inode->i_lock);
547 	if (ip->i_qadata == NULL) {
548 		struct gfs2_qadata *tmp;
549 
550 		spin_unlock(&inode->i_lock);
551 		tmp = kmem_cache_zalloc(gfs2_qadata_cachep, GFP_NOFS);
552 		if (!tmp)
553 			return -ENOMEM;
554 
555 		spin_lock(&inode->i_lock);
556 		if (ip->i_qadata == NULL)
557 			ip->i_qadata = tmp;
558 		else
559 			kmem_cache_free(gfs2_qadata_cachep, tmp);
560 	}
561 	ip->i_qadata->qa_ref++;
562 	spin_unlock(&inode->i_lock);
563 	return 0;
564 }
565 
566 void gfs2_qa_put(struct gfs2_inode *ip)
567 {
568 	struct inode *inode = &ip->i_inode;
569 
570 	spin_lock(&inode->i_lock);
571 	if (ip->i_qadata && --ip->i_qadata->qa_ref == 0) {
572 		kmem_cache_free(gfs2_qadata_cachep, ip->i_qadata);
573 		ip->i_qadata = NULL;
574 	}
575 	spin_unlock(&inode->i_lock);
576 }
577 
578 int gfs2_quota_hold(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
579 {
580 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
581 	struct gfs2_quota_data **qd;
582 	int error;
583 
584 	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
585 		return 0;
586 
587 	error = gfs2_qa_get(ip);
588 	if (error)
589 		return error;
590 
591 	qd = ip->i_qadata->qa_qd;
592 
593 	if (gfs2_assert_warn(sdp, !ip->i_qadata->qa_qd_num) ||
594 	    gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags))) {
595 		error = -EIO;
596 		gfs2_qa_put(ip);
597 		goto out;
598 	}
599 
600 	error = qdsb_get(sdp, make_kqid_uid(ip->i_inode.i_uid), qd);
601 	if (error)
602 		goto out_unhold;
603 	ip->i_qadata->qa_qd_num++;
604 	qd++;
605 
606 	error = qdsb_get(sdp, make_kqid_gid(ip->i_inode.i_gid), qd);
607 	if (error)
608 		goto out_unhold;
609 	ip->i_qadata->qa_qd_num++;
610 	qd++;
611 
612 	if (!uid_eq(uid, NO_UID_QUOTA_CHANGE) &&
613 	    !uid_eq(uid, ip->i_inode.i_uid)) {
614 		error = qdsb_get(sdp, make_kqid_uid(uid), qd);
615 		if (error)
616 			goto out_unhold;
617 		ip->i_qadata->qa_qd_num++;
618 		qd++;
619 	}
620 
621 	if (!gid_eq(gid, NO_GID_QUOTA_CHANGE) &&
622 	    !gid_eq(gid, ip->i_inode.i_gid)) {
623 		error = qdsb_get(sdp, make_kqid_gid(gid), qd);
624 		if (error)
625 			goto out_unhold;
626 		ip->i_qadata->qa_qd_num++;
627 		qd++;
628 	}
629 
630 out_unhold:
631 	if (error)
632 		gfs2_quota_unhold(ip);
633 out:
634 	return error;
635 }
636 
637 void gfs2_quota_unhold(struct gfs2_inode *ip)
638 {
639 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
640 	u32 x;
641 
642 	if (ip->i_qadata == NULL)
643 		return;
644 
645 	gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags));
646 
647 	for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
648 		qdsb_put(ip->i_qadata->qa_qd[x]);
649 		ip->i_qadata->qa_qd[x] = NULL;
650 	}
651 	ip->i_qadata->qa_qd_num = 0;
652 	gfs2_qa_put(ip);
653 }
654 
655 static int sort_qd(const void *a, const void *b)
656 {
657 	const struct gfs2_quota_data *qd_a = *(const struct gfs2_quota_data **)a;
658 	const struct gfs2_quota_data *qd_b = *(const struct gfs2_quota_data **)b;
659 
660 	if (qid_lt(qd_a->qd_id, qd_b->qd_id))
661 		return -1;
662 	if (qid_lt(qd_b->qd_id, qd_a->qd_id))
663 		return 1;
664 	return 0;
665 }
666 
667 static void do_qc(struct gfs2_quota_data *qd, s64 change, int qc_type)
668 {
669 	struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
670 	struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
671 	struct gfs2_quota_change *qc = qd->qd_bh_qc;
672 	s64 x;
673 
674 	mutex_lock(&sdp->sd_quota_mutex);
675 	gfs2_trans_add_meta(ip->i_gl, qd->qd_bh);
676 
677 	if (!test_bit(QDF_CHANGE, &qd->qd_flags)) {
678 		qc->qc_change = 0;
679 		qc->qc_flags = 0;
680 		if (qd->qd_id.type == USRQUOTA)
681 			qc->qc_flags = cpu_to_be32(GFS2_QCF_USER);
682 		qc->qc_id = cpu_to_be32(from_kqid(&init_user_ns, qd->qd_id));
683 	}
684 
685 	x = be64_to_cpu(qc->qc_change) + change;
686 	qc->qc_change = cpu_to_be64(x);
687 
688 	spin_lock(&qd_lock);
689 	qd->qd_change = x;
690 	spin_unlock(&qd_lock);
691 
692 	if (qc_type == QC_CHANGE) {
693 		if (!test_and_set_bit(QDF_CHANGE, &qd->qd_flags)) {
694 			qd_hold(qd);
695 			slot_hold(qd);
696 		}
697 	} else {
698 		gfs2_assert_warn(sdp, test_bit(QDF_CHANGE, &qd->qd_flags));
699 		clear_bit(QDF_CHANGE, &qd->qd_flags);
700 		qc->qc_flags = 0;
701 		qc->qc_id = 0;
702 		slot_put(qd);
703 		qd_put(qd);
704 	}
705 
706 	if (change < 0) /* Reset quiet flag if we freed some blocks */
707 		clear_bit(QDF_QMSG_QUIET, &qd->qd_flags);
708 	mutex_unlock(&sdp->sd_quota_mutex);
709 }
710 
711 static int gfs2_write_buf_to_page(struct gfs2_inode *ip, unsigned long index,
712 				  unsigned off, void *buf, unsigned bytes)
713 {
714 	struct inode *inode = &ip->i_inode;
715 	struct gfs2_sbd *sdp = GFS2_SB(inode);
716 	struct address_space *mapping = inode->i_mapping;
717 	struct page *page;
718 	struct buffer_head *bh;
719 	u64 blk;
720 	unsigned bsize = sdp->sd_sb.sb_bsize, bnum = 0, boff = 0;
721 	unsigned to_write = bytes, pg_off = off;
722 	int done = 0;
723 
724 	blk = index << (PAGE_SHIFT - sdp->sd_sb.sb_bsize_shift);
725 	boff = off % bsize;
726 
727 	page = find_or_create_page(mapping, index, GFP_NOFS);
728 	if (!page)
729 		return -ENOMEM;
730 	if (!page_has_buffers(page))
731 		create_empty_buffers(page, bsize, 0);
732 
733 	bh = page_buffers(page);
734 	while (!done) {
735 		/* Find the beginning block within the page */
736 		if (pg_off >= ((bnum * bsize) + bsize)) {
737 			bh = bh->b_this_page;
738 			bnum++;
739 			blk++;
740 			continue;
741 		}
742 		if (!buffer_mapped(bh)) {
743 			gfs2_block_map(inode, blk, bh, 1);
744 			if (!buffer_mapped(bh))
745 				goto unlock_out;
746 			/* If it's a newly allocated disk block, zero it */
747 			if (buffer_new(bh))
748 				zero_user(page, bnum * bsize, bh->b_size);
749 		}
750 		if (PageUptodate(page))
751 			set_buffer_uptodate(bh);
752 		if (bh_read(bh, REQ_META | REQ_PRIO) < 0)
753 			goto unlock_out;
754 		if (gfs2_is_jdata(ip))
755 			gfs2_trans_add_data(ip->i_gl, bh);
756 		else
757 			gfs2_ordered_add_inode(ip);
758 
759 		/* If we need to write to the next block as well */
760 		if (to_write > (bsize - boff)) {
761 			pg_off += (bsize - boff);
762 			to_write -= (bsize - boff);
763 			boff = pg_off % bsize;
764 			continue;
765 		}
766 		done = 1;
767 	}
768 
769 	/* Write to the page, now that we have setup the buffer(s) */
770 	memcpy_to_page(page, off, buf, bytes);
771 	flush_dcache_page(page);
772 	unlock_page(page);
773 	put_page(page);
774 
775 	return 0;
776 
777 unlock_out:
778 	unlock_page(page);
779 	put_page(page);
780 	return -EIO;
781 }
782 
783 static int gfs2_write_disk_quota(struct gfs2_inode *ip, struct gfs2_quota *qp,
784 				 loff_t loc)
785 {
786 	unsigned long pg_beg;
787 	unsigned pg_off, nbytes, overflow = 0;
788 	int pg_oflow = 0, error;
789 	void *ptr;
790 
791 	nbytes = sizeof(struct gfs2_quota);
792 
793 	pg_beg = loc >> PAGE_SHIFT;
794 	pg_off = offset_in_page(loc);
795 
796 	/* If the quota straddles a page boundary, split the write in two */
797 	if ((pg_off + nbytes) > PAGE_SIZE) {
798 		pg_oflow = 1;
799 		overflow = (pg_off + nbytes) - PAGE_SIZE;
800 	}
801 
802 	ptr = qp;
803 	error = gfs2_write_buf_to_page(ip, pg_beg, pg_off, ptr,
804 				       nbytes - overflow);
805 	/* If there's an overflow, write the remaining bytes to the next page */
806 	if (!error && pg_oflow)
807 		error = gfs2_write_buf_to_page(ip, pg_beg + 1, 0,
808 					       ptr + nbytes - overflow,
809 					       overflow);
810 	return error;
811 }
812 
813 /**
814  * gfs2_adjust_quota - adjust record of current block usage
815  * @ip: The quota inode
816  * @loc: Offset of the entry in the quota file
817  * @change: The amount of usage change to record
818  * @qd: The quota data
819  * @fdq: The updated limits to record
820  *
821  * This function was mostly borrowed from gfs2_block_truncate_page which was
822  * in turn mostly borrowed from ext3
823  *
824  * Returns: 0 or -ve on error
825  */
826 
827 static int gfs2_adjust_quota(struct gfs2_inode *ip, loff_t loc,
828 			     s64 change, struct gfs2_quota_data *qd,
829 			     struct qc_dqblk *fdq)
830 {
831 	struct inode *inode = &ip->i_inode;
832 	struct gfs2_sbd *sdp = GFS2_SB(inode);
833 	struct gfs2_quota q;
834 	int err;
835 	u64 size;
836 
837 	if (gfs2_is_stuffed(ip)) {
838 		err = gfs2_unstuff_dinode(ip);
839 		if (err)
840 			return err;
841 	}
842 
843 	memset(&q, 0, sizeof(struct gfs2_quota));
844 	err = gfs2_internal_read(ip, (char *)&q, &loc, sizeof(q));
845 	if (err < 0)
846 		return err;
847 
848 	loc -= sizeof(q); /* gfs2_internal_read would've advanced the loc ptr */
849 	err = -EIO;
850 	be64_add_cpu(&q.qu_value, change);
851 	if (((s64)be64_to_cpu(q.qu_value)) < 0)
852 		q.qu_value = 0; /* Never go negative on quota usage */
853 	qd->qd_qb.qb_value = q.qu_value;
854 	if (fdq) {
855 		if (fdq->d_fieldmask & QC_SPC_SOFT) {
856 			q.qu_warn = cpu_to_be64(fdq->d_spc_softlimit >> sdp->sd_sb.sb_bsize_shift);
857 			qd->qd_qb.qb_warn = q.qu_warn;
858 		}
859 		if (fdq->d_fieldmask & QC_SPC_HARD) {
860 			q.qu_limit = cpu_to_be64(fdq->d_spc_hardlimit >> sdp->sd_sb.sb_bsize_shift);
861 			qd->qd_qb.qb_limit = q.qu_limit;
862 		}
863 		if (fdq->d_fieldmask & QC_SPACE) {
864 			q.qu_value = cpu_to_be64(fdq->d_space >> sdp->sd_sb.sb_bsize_shift);
865 			qd->qd_qb.qb_value = q.qu_value;
866 		}
867 	}
868 
869 	err = gfs2_write_disk_quota(ip, &q, loc);
870 	if (!err) {
871 		size = loc + sizeof(struct gfs2_quota);
872 		if (size > inode->i_size)
873 			i_size_write(inode, size);
874 		inode->i_mtime = inode->i_atime = current_time(inode);
875 		mark_inode_dirty(inode);
876 		set_bit(QDF_REFRESH, &qd->qd_flags);
877 	}
878 
879 	return err;
880 }
881 
882 static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
883 {
884 	struct gfs2_sbd *sdp = (*qda)->qd_gl->gl_name.ln_sbd;
885 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
886 	struct gfs2_alloc_parms ap = { .aflags = 0, };
887 	unsigned int data_blocks, ind_blocks;
888 	struct gfs2_holder *ghs, i_gh;
889 	unsigned int qx, x;
890 	struct gfs2_quota_data *qd;
891 	unsigned reserved;
892 	loff_t offset;
893 	unsigned int nalloc = 0, blocks;
894 	int error;
895 
896 	error = gfs2_qa_get(ip);
897 	if (error)
898 		return error;
899 
900 	gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
901 			      &data_blocks, &ind_blocks);
902 
903 	ghs = kmalloc_array(num_qd, sizeof(struct gfs2_holder), GFP_NOFS);
904 	if (!ghs) {
905 		error = -ENOMEM;
906 		goto out;
907 	}
908 
909 	sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL);
910 	inode_lock(&ip->i_inode);
911 	for (qx = 0; qx < num_qd; qx++) {
912 		error = gfs2_glock_nq_init(qda[qx]->qd_gl, LM_ST_EXCLUSIVE,
913 					   GL_NOCACHE, &ghs[qx]);
914 		if (error)
915 			goto out_dq;
916 	}
917 
918 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
919 	if (error)
920 		goto out_dq;
921 
922 	for (x = 0; x < num_qd; x++) {
923 		offset = qd2offset(qda[x]);
924 		if (gfs2_write_alloc_required(ip, offset,
925 					      sizeof(struct gfs2_quota)))
926 			nalloc++;
927 	}
928 
929 	/*
930 	 * 1 blk for unstuffing inode if stuffed. We add this extra
931 	 * block to the reservation unconditionally. If the inode
932 	 * doesn't need unstuffing, the block will be released to the
933 	 * rgrp since it won't be allocated during the transaction
934 	 */
935 	/* +3 in the end for unstuffing block, inode size update block
936 	 * and another block in case quota straddles page boundary and
937 	 * two blocks need to be updated instead of 1 */
938 	blocks = num_qd * data_blocks + RES_DINODE + num_qd + 3;
939 
940 	reserved = 1 + (nalloc * (data_blocks + ind_blocks));
941 	ap.target = reserved;
942 	error = gfs2_inplace_reserve(ip, &ap);
943 	if (error)
944 		goto out_alloc;
945 
946 	if (nalloc)
947 		blocks += gfs2_rg_blocks(ip, reserved) + nalloc * ind_blocks + RES_STATFS;
948 
949 	error = gfs2_trans_begin(sdp, blocks, 0);
950 	if (error)
951 		goto out_ipres;
952 
953 	for (x = 0; x < num_qd; x++) {
954 		qd = qda[x];
955 		offset = qd2offset(qd);
956 		error = gfs2_adjust_quota(ip, offset, qd->qd_change_sync, qd, NULL);
957 		if (error)
958 			goto out_end_trans;
959 
960 		do_qc(qd, -qd->qd_change_sync, QC_SYNC);
961 		set_bit(QDF_REFRESH, &qd->qd_flags);
962 	}
963 
964 	error = 0;
965 
966 out_end_trans:
967 	gfs2_trans_end(sdp);
968 out_ipres:
969 	gfs2_inplace_release(ip);
970 out_alloc:
971 	gfs2_glock_dq_uninit(&i_gh);
972 out_dq:
973 	while (qx--)
974 		gfs2_glock_dq_uninit(&ghs[qx]);
975 	inode_unlock(&ip->i_inode);
976 	kfree(ghs);
977 	gfs2_log_flush(ip->i_gl->gl_name.ln_sbd, ip->i_gl,
978 		       GFS2_LOG_HEAD_FLUSH_NORMAL | GFS2_LFC_DO_SYNC);
979 out:
980 	gfs2_qa_put(ip);
981 	return error;
982 }
983 
984 static int update_qd(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd)
985 {
986 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
987 	struct gfs2_quota q;
988 	struct gfs2_quota_lvb *qlvb;
989 	loff_t pos;
990 	int error;
991 
992 	memset(&q, 0, sizeof(struct gfs2_quota));
993 	pos = qd2offset(qd);
994 	error = gfs2_internal_read(ip, (char *)&q, &pos, sizeof(q));
995 	if (error < 0)
996 		return error;
997 
998 	qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
999 	qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC);
1000 	qlvb->__pad = 0;
1001 	qlvb->qb_limit = q.qu_limit;
1002 	qlvb->qb_warn = q.qu_warn;
1003 	qlvb->qb_value = q.qu_value;
1004 	qd->qd_qb = *qlvb;
1005 
1006 	return 0;
1007 }
1008 
1009 static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
1010 		    struct gfs2_holder *q_gh)
1011 {
1012 	struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
1013 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1014 	struct gfs2_holder i_gh;
1015 	int error;
1016 
1017 restart:
1018 	error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
1019 	if (error)
1020 		return error;
1021 
1022 	if (test_and_clear_bit(QDF_REFRESH, &qd->qd_flags))
1023 		force_refresh = FORCE;
1024 
1025 	qd->qd_qb = *(struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1026 
1027 	if (force_refresh || qd->qd_qb.qb_magic != cpu_to_be32(GFS2_MAGIC)) {
1028 		gfs2_glock_dq_uninit(q_gh);
1029 		error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE,
1030 					   GL_NOCACHE, q_gh);
1031 		if (error)
1032 			return error;
1033 
1034 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
1035 		if (error)
1036 			goto fail;
1037 
1038 		error = update_qd(sdp, qd);
1039 		if (error)
1040 			goto fail_gunlock;
1041 
1042 		gfs2_glock_dq_uninit(&i_gh);
1043 		gfs2_glock_dq_uninit(q_gh);
1044 		force_refresh = 0;
1045 		goto restart;
1046 	}
1047 
1048 	return 0;
1049 
1050 fail_gunlock:
1051 	gfs2_glock_dq_uninit(&i_gh);
1052 fail:
1053 	gfs2_glock_dq_uninit(q_gh);
1054 	return error;
1055 }
1056 
1057 int gfs2_quota_lock(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
1058 {
1059 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1060 	struct gfs2_quota_data *qd;
1061 	u32 x;
1062 	int error = 0;
1063 
1064 	if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
1065 		return 0;
1066 
1067 	error = gfs2_quota_hold(ip, uid, gid);
1068 	if (error)
1069 		return error;
1070 
1071 	sort(ip->i_qadata->qa_qd, ip->i_qadata->qa_qd_num,
1072 	     sizeof(struct gfs2_quota_data *), sort_qd, NULL);
1073 
1074 	for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1075 		qd = ip->i_qadata->qa_qd[x];
1076 		error = do_glock(qd, NO_FORCE, &ip->i_qadata->qa_qd_ghs[x]);
1077 		if (error)
1078 			break;
1079 	}
1080 
1081 	if (!error)
1082 		set_bit(GIF_QD_LOCKED, &ip->i_flags);
1083 	else {
1084 		while (x--)
1085 			gfs2_glock_dq_uninit(&ip->i_qadata->qa_qd_ghs[x]);
1086 		gfs2_quota_unhold(ip);
1087 	}
1088 
1089 	return error;
1090 }
1091 
1092 static int need_sync(struct gfs2_quota_data *qd)
1093 {
1094 	struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
1095 	struct gfs2_tune *gt = &sdp->sd_tune;
1096 	s64 value;
1097 	unsigned int num, den;
1098 	int do_sync = 1;
1099 
1100 	if (!qd->qd_qb.qb_limit)
1101 		return 0;
1102 
1103 	spin_lock(&qd_lock);
1104 	value = qd->qd_change;
1105 	spin_unlock(&qd_lock);
1106 
1107 	spin_lock(&gt->gt_spin);
1108 	num = gt->gt_quota_scale_num;
1109 	den = gt->gt_quota_scale_den;
1110 	spin_unlock(&gt->gt_spin);
1111 
1112 	if (value < 0)
1113 		do_sync = 0;
1114 	else if ((s64)be64_to_cpu(qd->qd_qb.qb_value) >=
1115 		 (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1116 		do_sync = 0;
1117 	else {
1118 		value *= gfs2_jindex_size(sdp) * num;
1119 		value = div_s64(value, den);
1120 		value += (s64)be64_to_cpu(qd->qd_qb.qb_value);
1121 		if (value < (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1122 			do_sync = 0;
1123 	}
1124 
1125 	return do_sync;
1126 }
1127 
1128 void gfs2_quota_unlock(struct gfs2_inode *ip)
1129 {
1130 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1131 	struct gfs2_quota_data *qda[4];
1132 	unsigned int count = 0;
1133 	u32 x;
1134 	int found;
1135 
1136 	if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
1137 		return;
1138 
1139 	for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1140 		struct gfs2_quota_data *qd;
1141 		int sync;
1142 
1143 		qd = ip->i_qadata->qa_qd[x];
1144 		sync = need_sync(qd);
1145 
1146 		gfs2_glock_dq_uninit(&ip->i_qadata->qa_qd_ghs[x]);
1147 		if (!sync)
1148 			continue;
1149 
1150 		spin_lock(&qd_lock);
1151 		found = qd_check_sync(sdp, qd, NULL);
1152 		spin_unlock(&qd_lock);
1153 
1154 		if (!found)
1155 			continue;
1156 
1157 		gfs2_assert_warn(sdp, qd->qd_change_sync);
1158 		if (bh_get(qd)) {
1159 			clear_bit(QDF_LOCKED, &qd->qd_flags);
1160 			slot_put(qd);
1161 			qd_put(qd);
1162 			continue;
1163 		}
1164 
1165 		qda[count++] = qd;
1166 	}
1167 
1168 	if (count) {
1169 		do_sync(count, qda);
1170 		for (x = 0; x < count; x++)
1171 			qd_unlock(qda[x]);
1172 	}
1173 
1174 	gfs2_quota_unhold(ip);
1175 }
1176 
1177 #define MAX_LINE 256
1178 
1179 static int print_message(struct gfs2_quota_data *qd, char *type)
1180 {
1181 	struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
1182 
1183 	fs_info(sdp, "quota %s for %s %u\n",
1184 		type,
1185 		(qd->qd_id.type == USRQUOTA) ? "user" : "group",
1186 		from_kqid(&init_user_ns, qd->qd_id));
1187 
1188 	return 0;
1189 }
1190 
1191 /**
1192  * gfs2_quota_check - check if allocating new blocks will exceed quota
1193  * @ip:  The inode for which this check is being performed
1194  * @uid: The uid to check against
1195  * @gid: The gid to check against
1196  * @ap:  The allocation parameters. ap->target contains the requested
1197  *       blocks. ap->min_target, if set, contains the minimum blks
1198  *       requested.
1199  *
1200  * Returns: 0 on success.
1201  *                  min_req = ap->min_target ? ap->min_target : ap->target;
1202  *                  quota must allow at least min_req blks for success and
1203  *                  ap->allowed is set to the number of blocks allowed
1204  *
1205  *          -EDQUOT otherwise, quota violation. ap->allowed is set to number
1206  *                  of blocks available.
1207  */
1208 int gfs2_quota_check(struct gfs2_inode *ip, kuid_t uid, kgid_t gid,
1209 		     struct gfs2_alloc_parms *ap)
1210 {
1211 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1212 	struct gfs2_quota_data *qd;
1213 	s64 value, warn, limit;
1214 	u32 x;
1215 	int error = 0;
1216 
1217 	ap->allowed = UINT_MAX; /* Assume we are permitted a whole lot */
1218 	if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
1219 		return 0;
1220 
1221 	for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1222 		qd = ip->i_qadata->qa_qd[x];
1223 
1224 		if (!(qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1225 		      qid_eq(qd->qd_id, make_kqid_gid(gid))))
1226 			continue;
1227 
1228 		warn = (s64)be64_to_cpu(qd->qd_qb.qb_warn);
1229 		limit = (s64)be64_to_cpu(qd->qd_qb.qb_limit);
1230 		value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
1231 		spin_lock(&qd_lock);
1232 		value += qd->qd_change;
1233 		spin_unlock(&qd_lock);
1234 
1235 		if (limit > 0 && (limit - value) < ap->allowed)
1236 			ap->allowed = limit - value;
1237 		/* If we can't meet the target */
1238 		if (limit && limit < (value + (s64)ap->target)) {
1239 			/* If no min_target specified or we don't meet
1240 			 * min_target, return -EDQUOT */
1241 			if (!ap->min_target || ap->min_target > ap->allowed) {
1242 				if (!test_and_set_bit(QDF_QMSG_QUIET,
1243 						      &qd->qd_flags)) {
1244 					print_message(qd, "exceeded");
1245 					quota_send_warning(qd->qd_id,
1246 							   sdp->sd_vfs->s_dev,
1247 							   QUOTA_NL_BHARDWARN);
1248 				}
1249 				error = -EDQUOT;
1250 				break;
1251 			}
1252 		} else if (warn && warn < value &&
1253 			   time_after_eq(jiffies, qd->qd_last_warn +
1254 					 gfs2_tune_get(sdp, gt_quota_warn_period)
1255 					 * HZ)) {
1256 			quota_send_warning(qd->qd_id,
1257 					   sdp->sd_vfs->s_dev, QUOTA_NL_BSOFTWARN);
1258 			error = print_message(qd, "warning");
1259 			qd->qd_last_warn = jiffies;
1260 		}
1261 	}
1262 	return error;
1263 }
1264 
1265 void gfs2_quota_change(struct gfs2_inode *ip, s64 change,
1266 		       kuid_t uid, kgid_t gid)
1267 {
1268 	struct gfs2_quota_data *qd;
1269 	u32 x;
1270 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1271 
1272 	if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON ||
1273 	    gfs2_assert_warn(sdp, change))
1274 		return;
1275 	if (ip->i_diskflags & GFS2_DIF_SYSTEM)
1276 		return;
1277 
1278 	if (gfs2_assert_withdraw(sdp, ip->i_qadata &&
1279 				 ip->i_qadata->qa_ref > 0))
1280 		return;
1281 	for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1282 		qd = ip->i_qadata->qa_qd[x];
1283 
1284 		if (qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1285 		    qid_eq(qd->qd_id, make_kqid_gid(gid))) {
1286 			do_qc(qd, change, QC_CHANGE);
1287 		}
1288 	}
1289 }
1290 
1291 int gfs2_quota_sync(struct super_block *sb, int type)
1292 {
1293 	struct gfs2_sbd *sdp = sb->s_fs_info;
1294 	struct gfs2_quota_data **qda;
1295 	unsigned int max_qd = PAGE_SIZE / sizeof(struct gfs2_holder);
1296 	unsigned int num_qd;
1297 	unsigned int x;
1298 	int error = 0;
1299 
1300 	qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
1301 	if (!qda)
1302 		return -ENOMEM;
1303 
1304 	mutex_lock(&sdp->sd_quota_sync_mutex);
1305 	sdp->sd_quota_sync_gen++;
1306 
1307 	do {
1308 		num_qd = 0;
1309 
1310 		for (;;) {
1311 			error = qd_fish(sdp, qda + num_qd);
1312 			if (error || !qda[num_qd])
1313 				break;
1314 			if (++num_qd == max_qd)
1315 				break;
1316 		}
1317 
1318 		if (num_qd) {
1319 			if (!error)
1320 				error = do_sync(num_qd, qda);
1321 			if (!error)
1322 				for (x = 0; x < num_qd; x++)
1323 					qda[x]->qd_sync_gen =
1324 						sdp->sd_quota_sync_gen;
1325 
1326 			for (x = 0; x < num_qd; x++)
1327 				qd_unlock(qda[x]);
1328 		}
1329 	} while (!error && num_qd == max_qd);
1330 
1331 	mutex_unlock(&sdp->sd_quota_sync_mutex);
1332 	kfree(qda);
1333 
1334 	return error;
1335 }
1336 
1337 int gfs2_quota_refresh(struct gfs2_sbd *sdp, struct kqid qid)
1338 {
1339 	struct gfs2_quota_data *qd;
1340 	struct gfs2_holder q_gh;
1341 	int error;
1342 
1343 	error = qd_get(sdp, qid, &qd);
1344 	if (error)
1345 		return error;
1346 
1347 	error = do_glock(qd, FORCE, &q_gh);
1348 	if (!error)
1349 		gfs2_glock_dq_uninit(&q_gh);
1350 
1351 	qd_put(qd);
1352 	return error;
1353 }
1354 
1355 int gfs2_quota_init(struct gfs2_sbd *sdp)
1356 {
1357 	struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
1358 	u64 size = i_size_read(sdp->sd_qc_inode);
1359 	unsigned int blocks = size >> sdp->sd_sb.sb_bsize_shift;
1360 	unsigned int x, slot = 0;
1361 	unsigned int found = 0;
1362 	unsigned int hash;
1363 	unsigned int bm_size;
1364 	u64 dblock;
1365 	u32 extlen = 0;
1366 	int error;
1367 
1368 	if (gfs2_check_internal_file_size(sdp->sd_qc_inode, 1, 64 << 20))
1369 		return -EIO;
1370 
1371 	sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
1372 	bm_size = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * sizeof(unsigned long));
1373 	bm_size *= sizeof(unsigned long);
1374 	error = -ENOMEM;
1375 	sdp->sd_quota_bitmap = kzalloc(bm_size, GFP_NOFS | __GFP_NOWARN);
1376 	if (sdp->sd_quota_bitmap == NULL)
1377 		sdp->sd_quota_bitmap = __vmalloc(bm_size, GFP_NOFS |
1378 						 __GFP_ZERO);
1379 	if (!sdp->sd_quota_bitmap)
1380 		return error;
1381 
1382 	for (x = 0; x < blocks; x++) {
1383 		struct buffer_head *bh;
1384 		const struct gfs2_quota_change *qc;
1385 		unsigned int y;
1386 
1387 		if (!extlen) {
1388 			extlen = 32;
1389 			error = gfs2_get_extent(&ip->i_inode, x, &dblock, &extlen);
1390 			if (error)
1391 				goto fail;
1392 		}
1393 		error = -EIO;
1394 		bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
1395 		if (!bh)
1396 			goto fail;
1397 		if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
1398 			brelse(bh);
1399 			goto fail;
1400 		}
1401 
1402 		qc = (const struct gfs2_quota_change *)(bh->b_data + sizeof(struct gfs2_meta_header));
1403 		for (y = 0; y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1404 		     y++, slot++) {
1405 			struct gfs2_quota_data *qd;
1406 			s64 qc_change = be64_to_cpu(qc->qc_change);
1407 			u32 qc_flags = be32_to_cpu(qc->qc_flags);
1408 			enum quota_type qtype = (qc_flags & GFS2_QCF_USER) ?
1409 						USRQUOTA : GRPQUOTA;
1410 			struct kqid qc_id = make_kqid(&init_user_ns, qtype,
1411 						      be32_to_cpu(qc->qc_id));
1412 			qc++;
1413 			if (!qc_change)
1414 				continue;
1415 
1416 			hash = gfs2_qd_hash(sdp, qc_id);
1417 			qd = qd_alloc(hash, sdp, qc_id);
1418 			if (qd == NULL) {
1419 				brelse(bh);
1420 				goto fail;
1421 			}
1422 
1423 			set_bit(QDF_CHANGE, &qd->qd_flags);
1424 			qd->qd_change = qc_change;
1425 			qd->qd_slot = slot;
1426 			qd->qd_slot_count = 1;
1427 
1428 			spin_lock(&qd_lock);
1429 			BUG_ON(test_and_set_bit(slot, sdp->sd_quota_bitmap));
1430 			list_add(&qd->qd_list, &sdp->sd_quota_list);
1431 			atomic_inc(&sdp->sd_quota_count);
1432 			spin_unlock(&qd_lock);
1433 
1434 			spin_lock_bucket(hash);
1435 			hlist_bl_add_head_rcu(&qd->qd_hlist, &qd_hash_table[hash]);
1436 			spin_unlock_bucket(hash);
1437 
1438 			found++;
1439 		}
1440 
1441 		brelse(bh);
1442 		dblock++;
1443 		extlen--;
1444 	}
1445 
1446 	if (found)
1447 		fs_info(sdp, "found %u quota changes\n", found);
1448 
1449 	return 0;
1450 
1451 fail:
1452 	gfs2_quota_cleanup(sdp);
1453 	return error;
1454 }
1455 
1456 void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1457 {
1458 	struct list_head *head = &sdp->sd_quota_list;
1459 	struct gfs2_quota_data *qd;
1460 
1461 	spin_lock(&qd_lock);
1462 	while (!list_empty(head)) {
1463 		qd = list_last_entry(head, struct gfs2_quota_data, qd_list);
1464 
1465 		list_del(&qd->qd_list);
1466 
1467 		/* Also remove if this qd exists in the reclaim list */
1468 		list_lru_del(&gfs2_qd_lru, &qd->qd_lru);
1469 		atomic_dec(&sdp->sd_quota_count);
1470 		spin_unlock(&qd_lock);
1471 
1472 		spin_lock_bucket(qd->qd_hash);
1473 		hlist_bl_del_rcu(&qd->qd_hlist);
1474 		spin_unlock_bucket(qd->qd_hash);
1475 
1476 		gfs2_assert_warn(sdp, !qd->qd_change);
1477 		gfs2_assert_warn(sdp, !qd->qd_slot_count);
1478 		gfs2_assert_warn(sdp, !qd->qd_bh_count);
1479 
1480 		gfs2_glock_put(qd->qd_gl);
1481 		call_rcu(&qd->qd_rcu, gfs2_qd_dealloc);
1482 
1483 		spin_lock(&qd_lock);
1484 	}
1485 	spin_unlock(&qd_lock);
1486 
1487 	gfs2_assert_warn(sdp, !atomic_read(&sdp->sd_quota_count));
1488 
1489 	kvfree(sdp->sd_quota_bitmap);
1490 	sdp->sd_quota_bitmap = NULL;
1491 }
1492 
1493 static void quotad_error(struct gfs2_sbd *sdp, const char *msg, int error)
1494 {
1495 	if (error == 0 || error == -EROFS)
1496 		return;
1497 	if (!gfs2_withdrawn(sdp)) {
1498 		if (!cmpxchg(&sdp->sd_log_error, 0, error))
1499 			fs_err(sdp, "gfs2_quotad: %s error %d\n", msg, error);
1500 		wake_up(&sdp->sd_logd_waitq);
1501 	}
1502 }
1503 
1504 static void quotad_check_timeo(struct gfs2_sbd *sdp, const char *msg,
1505 			       int (*fxn)(struct super_block *sb, int type),
1506 			       unsigned long t, unsigned long *timeo,
1507 			       unsigned int *new_timeo)
1508 {
1509 	if (t >= *timeo) {
1510 		int error = fxn(sdp->sd_vfs, 0);
1511 		quotad_error(sdp, msg, error);
1512 		*timeo = gfs2_tune_get_i(&sdp->sd_tune, new_timeo) * HZ;
1513 	} else {
1514 		*timeo -= t;
1515 	}
1516 }
1517 
1518 void gfs2_wake_up_statfs(struct gfs2_sbd *sdp) {
1519 	if (!sdp->sd_statfs_force_sync) {
1520 		sdp->sd_statfs_force_sync = 1;
1521 		wake_up(&sdp->sd_quota_wait);
1522 	}
1523 }
1524 
1525 
1526 /**
1527  * gfs2_quotad - Write cached quota changes into the quota file
1528  * @data: Pointer to GFS2 superblock
1529  *
1530  */
1531 
1532 int gfs2_quotad(void *data)
1533 {
1534 	struct gfs2_sbd *sdp = data;
1535 	struct gfs2_tune *tune = &sdp->sd_tune;
1536 	unsigned long statfs_timeo = 0;
1537 	unsigned long quotad_timeo = 0;
1538 	unsigned long t = 0;
1539 	DEFINE_WAIT(wait);
1540 
1541 	while (!kthread_should_stop()) {
1542 
1543 		if (gfs2_withdrawn(sdp))
1544 			goto bypass;
1545 		/* Update the master statfs file */
1546 		if (sdp->sd_statfs_force_sync) {
1547 			int error = gfs2_statfs_sync(sdp->sd_vfs, 0);
1548 			quotad_error(sdp, "statfs", error);
1549 			statfs_timeo = gfs2_tune_get(sdp, gt_statfs_quantum) * HZ;
1550 		}
1551 		else
1552 			quotad_check_timeo(sdp, "statfs", gfs2_statfs_sync, t,
1553 				   	   &statfs_timeo,
1554 					   &tune->gt_statfs_quantum);
1555 
1556 		/* Update quota file */
1557 		quotad_check_timeo(sdp, "sync", gfs2_quota_sync, t,
1558 				   &quotad_timeo, &tune->gt_quota_quantum);
1559 
1560 		try_to_freeze();
1561 
1562 bypass:
1563 		t = min(quotad_timeo, statfs_timeo);
1564 
1565 		prepare_to_wait(&sdp->sd_quota_wait, &wait, TASK_INTERRUPTIBLE);
1566 		if (!sdp->sd_statfs_force_sync)
1567 			t -= schedule_timeout(t);
1568 		else
1569 			t = 0;
1570 		finish_wait(&sdp->sd_quota_wait, &wait);
1571 	}
1572 
1573 	return 0;
1574 }
1575 
1576 static int gfs2_quota_get_state(struct super_block *sb, struct qc_state *state)
1577 {
1578 	struct gfs2_sbd *sdp = sb->s_fs_info;
1579 
1580 	memset(state, 0, sizeof(*state));
1581 
1582 	switch (sdp->sd_args.ar_quota) {
1583 	case GFS2_QUOTA_ON:
1584 		state->s_state[USRQUOTA].flags |= QCI_LIMITS_ENFORCED;
1585 		state->s_state[GRPQUOTA].flags |= QCI_LIMITS_ENFORCED;
1586 		fallthrough;
1587 	case GFS2_QUOTA_ACCOUNT:
1588 		state->s_state[USRQUOTA].flags |= QCI_ACCT_ENABLED |
1589 						  QCI_SYSFILE;
1590 		state->s_state[GRPQUOTA].flags |= QCI_ACCT_ENABLED |
1591 						  QCI_SYSFILE;
1592 		break;
1593 	case GFS2_QUOTA_OFF:
1594 		break;
1595 	}
1596 	if (sdp->sd_quota_inode) {
1597 		state->s_state[USRQUOTA].ino =
1598 					GFS2_I(sdp->sd_quota_inode)->i_no_addr;
1599 		state->s_state[USRQUOTA].blocks = sdp->sd_quota_inode->i_blocks;
1600 	}
1601 	state->s_state[USRQUOTA].nextents = 1;	/* unsupported */
1602 	state->s_state[GRPQUOTA] = state->s_state[USRQUOTA];
1603 	state->s_incoredqs = list_lru_count(&gfs2_qd_lru);
1604 	return 0;
1605 }
1606 
1607 static int gfs2_get_dqblk(struct super_block *sb, struct kqid qid,
1608 			  struct qc_dqblk *fdq)
1609 {
1610 	struct gfs2_sbd *sdp = sb->s_fs_info;
1611 	struct gfs2_quota_lvb *qlvb;
1612 	struct gfs2_quota_data *qd;
1613 	struct gfs2_holder q_gh;
1614 	int error;
1615 
1616 	memset(fdq, 0, sizeof(*fdq));
1617 
1618 	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1619 		return -ESRCH; /* Crazy XFS error code */
1620 
1621 	if ((qid.type != USRQUOTA) &&
1622 	    (qid.type != GRPQUOTA))
1623 		return -EINVAL;
1624 
1625 	error = qd_get(sdp, qid, &qd);
1626 	if (error)
1627 		return error;
1628 	error = do_glock(qd, FORCE, &q_gh);
1629 	if (error)
1630 		goto out;
1631 
1632 	qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1633 	fdq->d_spc_hardlimit = be64_to_cpu(qlvb->qb_limit) << sdp->sd_sb.sb_bsize_shift;
1634 	fdq->d_spc_softlimit = be64_to_cpu(qlvb->qb_warn) << sdp->sd_sb.sb_bsize_shift;
1635 	fdq->d_space = be64_to_cpu(qlvb->qb_value) << sdp->sd_sb.sb_bsize_shift;
1636 
1637 	gfs2_glock_dq_uninit(&q_gh);
1638 out:
1639 	qd_put(qd);
1640 	return error;
1641 }
1642 
1643 /* GFS2 only supports a subset of the XFS fields */
1644 #define GFS2_FIELDMASK (QC_SPC_SOFT|QC_SPC_HARD|QC_SPACE)
1645 
1646 static int gfs2_set_dqblk(struct super_block *sb, struct kqid qid,
1647 			  struct qc_dqblk *fdq)
1648 {
1649 	struct gfs2_sbd *sdp = sb->s_fs_info;
1650 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1651 	struct gfs2_quota_data *qd;
1652 	struct gfs2_holder q_gh, i_gh;
1653 	unsigned int data_blocks, ind_blocks;
1654 	unsigned int blocks = 0;
1655 	int alloc_required;
1656 	loff_t offset;
1657 	int error;
1658 
1659 	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1660 		return -ESRCH; /* Crazy XFS error code */
1661 
1662 	if ((qid.type != USRQUOTA) &&
1663 	    (qid.type != GRPQUOTA))
1664 		return -EINVAL;
1665 
1666 	if (fdq->d_fieldmask & ~GFS2_FIELDMASK)
1667 		return -EINVAL;
1668 
1669 	error = qd_get(sdp, qid, &qd);
1670 	if (error)
1671 		return error;
1672 
1673 	error = gfs2_qa_get(ip);
1674 	if (error)
1675 		goto out_put;
1676 
1677 	inode_lock(&ip->i_inode);
1678 	error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE, 0, &q_gh);
1679 	if (error)
1680 		goto out_unlockput;
1681 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1682 	if (error)
1683 		goto out_q;
1684 
1685 	/* Check for existing entry, if none then alloc new blocks */
1686 	error = update_qd(sdp, qd);
1687 	if (error)
1688 		goto out_i;
1689 
1690 	/* If nothing has changed, this is a no-op */
1691 	if ((fdq->d_fieldmask & QC_SPC_SOFT) &&
1692 	    ((fdq->d_spc_softlimit >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_warn)))
1693 		fdq->d_fieldmask ^= QC_SPC_SOFT;
1694 
1695 	if ((fdq->d_fieldmask & QC_SPC_HARD) &&
1696 	    ((fdq->d_spc_hardlimit >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_limit)))
1697 		fdq->d_fieldmask ^= QC_SPC_HARD;
1698 
1699 	if ((fdq->d_fieldmask & QC_SPACE) &&
1700 	    ((fdq->d_space >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_value)))
1701 		fdq->d_fieldmask ^= QC_SPACE;
1702 
1703 	if (fdq->d_fieldmask == 0)
1704 		goto out_i;
1705 
1706 	offset = qd2offset(qd);
1707 	alloc_required = gfs2_write_alloc_required(ip, offset, sizeof(struct gfs2_quota));
1708 	if (gfs2_is_stuffed(ip))
1709 		alloc_required = 1;
1710 	if (alloc_required) {
1711 		struct gfs2_alloc_parms ap = { .aflags = 0, };
1712 		gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
1713 				       &data_blocks, &ind_blocks);
1714 		blocks = 1 + data_blocks + ind_blocks;
1715 		ap.target = blocks;
1716 		error = gfs2_inplace_reserve(ip, &ap);
1717 		if (error)
1718 			goto out_i;
1719 		blocks += gfs2_rg_blocks(ip, blocks);
1720 	}
1721 
1722 	/* Some quotas span block boundaries and can update two blocks,
1723 	   adding an extra block to the transaction to handle such quotas */
1724 	error = gfs2_trans_begin(sdp, blocks + RES_DINODE + 2, 0);
1725 	if (error)
1726 		goto out_release;
1727 
1728 	/* Apply changes */
1729 	error = gfs2_adjust_quota(ip, offset, 0, qd, fdq);
1730 	if (!error)
1731 		clear_bit(QDF_QMSG_QUIET, &qd->qd_flags);
1732 
1733 	gfs2_trans_end(sdp);
1734 out_release:
1735 	if (alloc_required)
1736 		gfs2_inplace_release(ip);
1737 out_i:
1738 	gfs2_glock_dq_uninit(&i_gh);
1739 out_q:
1740 	gfs2_glock_dq_uninit(&q_gh);
1741 out_unlockput:
1742 	gfs2_qa_put(ip);
1743 	inode_unlock(&ip->i_inode);
1744 out_put:
1745 	qd_put(qd);
1746 	return error;
1747 }
1748 
1749 const struct quotactl_ops gfs2_quotactl_ops = {
1750 	.quota_sync     = gfs2_quota_sync,
1751 	.get_state	= gfs2_quota_get_state,
1752 	.get_dqblk	= gfs2_get_dqblk,
1753 	.set_dqblk	= gfs2_set_dqblk,
1754 };
1755 
1756 void __init gfs2_quota_hash_init(void)
1757 {
1758 	unsigned i;
1759 
1760 	for(i = 0; i < GFS2_QD_HASH_SIZE; i++)
1761 		INIT_HLIST_BL_HEAD(&qd_hash_table[i]);
1762 }
1763