xref: /openbmc/linux/fs/gfs2/quota.c (revision dff03381)
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 inode *inode = sdp->sd_qc_inode;
369 	struct gfs2_inode *ip = GFS2_I(inode);
370 	unsigned int block, offset;
371 	struct buffer_head *bh;
372 	struct iomap iomap = { };
373 	int error;
374 
375 	mutex_lock(&sdp->sd_quota_mutex);
376 
377 	if (qd->qd_bh_count++) {
378 		mutex_unlock(&sdp->sd_quota_mutex);
379 		return 0;
380 	}
381 
382 	block = qd->qd_slot / sdp->sd_qc_per_block;
383 	offset = qd->qd_slot % sdp->sd_qc_per_block;
384 
385 	error = gfs2_iomap_get(inode,
386 			       (loff_t)block << inode->i_blkbits,
387 			       i_blocksize(inode), &iomap);
388 	if (error)
389 		goto fail;
390 	error = -ENOENT;
391 	if (iomap.type != IOMAP_MAPPED)
392 		goto fail;
393 
394 	error = gfs2_meta_read(ip->i_gl, iomap.addr >> inode->i_blkbits,
395 			       DIO_WAIT, 0, &bh);
396 	if (error)
397 		goto fail;
398 	error = -EIO;
399 	if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC))
400 		goto fail_brelse;
401 
402 	qd->qd_bh = bh;
403 	qd->qd_bh_qc = (struct gfs2_quota_change *)
404 		(bh->b_data + sizeof(struct gfs2_meta_header) +
405 		 offset * sizeof(struct gfs2_quota_change));
406 
407 	mutex_unlock(&sdp->sd_quota_mutex);
408 
409 	return 0;
410 
411 fail_brelse:
412 	brelse(bh);
413 fail:
414 	qd->qd_bh_count--;
415 	mutex_unlock(&sdp->sd_quota_mutex);
416 	return error;
417 }
418 
419 static void bh_put(struct gfs2_quota_data *qd)
420 {
421 	struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
422 
423 	mutex_lock(&sdp->sd_quota_mutex);
424 	gfs2_assert(sdp, qd->qd_bh_count);
425 	if (!--qd->qd_bh_count) {
426 		brelse(qd->qd_bh);
427 		qd->qd_bh = NULL;
428 		qd->qd_bh_qc = NULL;
429 	}
430 	mutex_unlock(&sdp->sd_quota_mutex);
431 }
432 
433 static int qd_check_sync(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd,
434 			 u64 *sync_gen)
435 {
436 	if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
437 	    !test_bit(QDF_CHANGE, &qd->qd_flags) ||
438 	    (sync_gen && (qd->qd_sync_gen >= *sync_gen)))
439 		return 0;
440 
441 	if (!lockref_get_not_dead(&qd->qd_lockref))
442 		return 0;
443 
444 	list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
445 	set_bit(QDF_LOCKED, &qd->qd_flags);
446 	qd->qd_change_sync = qd->qd_change;
447 	slot_hold(qd);
448 	return 1;
449 }
450 
451 static int qd_fish(struct gfs2_sbd *sdp, struct gfs2_quota_data **qdp)
452 {
453 	struct gfs2_quota_data *qd = NULL, *iter;
454 	int error;
455 
456 	*qdp = NULL;
457 
458 	if (sb_rdonly(sdp->sd_vfs))
459 		return 0;
460 
461 	spin_lock(&qd_lock);
462 
463 	list_for_each_entry(iter, &sdp->sd_quota_list, qd_list) {
464 		if (qd_check_sync(sdp, iter, &sdp->sd_quota_sync_gen)) {
465 			qd = iter;
466 			break;
467 		}
468 	}
469 
470 	spin_unlock(&qd_lock);
471 
472 	if (qd) {
473 		gfs2_assert_warn(sdp, qd->qd_change_sync);
474 		error = bh_get(qd);
475 		if (error) {
476 			clear_bit(QDF_LOCKED, &qd->qd_flags);
477 			slot_put(qd);
478 			qd_put(qd);
479 			return error;
480 		}
481 	}
482 
483 	*qdp = qd;
484 
485 	return 0;
486 }
487 
488 static void qd_unlock(struct gfs2_quota_data *qd)
489 {
490 	gfs2_assert_warn(qd->qd_gl->gl_name.ln_sbd,
491 			 test_bit(QDF_LOCKED, &qd->qd_flags));
492 	clear_bit(QDF_LOCKED, &qd->qd_flags);
493 	bh_put(qd);
494 	slot_put(qd);
495 	qd_put(qd);
496 }
497 
498 static int qdsb_get(struct gfs2_sbd *sdp, struct kqid qid,
499 		    struct gfs2_quota_data **qdp)
500 {
501 	int error;
502 
503 	error = qd_get(sdp, qid, qdp);
504 	if (error)
505 		return error;
506 
507 	error = slot_get(*qdp);
508 	if (error)
509 		goto fail;
510 
511 	error = bh_get(*qdp);
512 	if (error)
513 		goto fail_slot;
514 
515 	return 0;
516 
517 fail_slot:
518 	slot_put(*qdp);
519 fail:
520 	qd_put(*qdp);
521 	return error;
522 }
523 
524 static void qdsb_put(struct gfs2_quota_data *qd)
525 {
526 	bh_put(qd);
527 	slot_put(qd);
528 	qd_put(qd);
529 }
530 
531 /**
532  * gfs2_qa_get - make sure we have a quota allocations data structure,
533  *               if necessary
534  * @ip: the inode for this reservation
535  */
536 int gfs2_qa_get(struct gfs2_inode *ip)
537 {
538 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
539 	struct inode *inode = &ip->i_inode;
540 
541 	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
542 		return 0;
543 
544 	spin_lock(&inode->i_lock);
545 	if (ip->i_qadata == NULL) {
546 		struct gfs2_qadata *tmp;
547 
548 		spin_unlock(&inode->i_lock);
549 		tmp = kmem_cache_zalloc(gfs2_qadata_cachep, GFP_NOFS);
550 		if (!tmp)
551 			return -ENOMEM;
552 
553 		spin_lock(&inode->i_lock);
554 		if (ip->i_qadata == NULL)
555 			ip->i_qadata = tmp;
556 		else
557 			kmem_cache_free(gfs2_qadata_cachep, tmp);
558 	}
559 	ip->i_qadata->qa_ref++;
560 	spin_unlock(&inode->i_lock);
561 	return 0;
562 }
563 
564 void gfs2_qa_put(struct gfs2_inode *ip)
565 {
566 	struct inode *inode = &ip->i_inode;
567 
568 	spin_lock(&inode->i_lock);
569 	if (ip->i_qadata && --ip->i_qadata->qa_ref == 0) {
570 		kmem_cache_free(gfs2_qadata_cachep, ip->i_qadata);
571 		ip->i_qadata = NULL;
572 	}
573 	spin_unlock(&inode->i_lock);
574 }
575 
576 int gfs2_quota_hold(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
577 {
578 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
579 	struct gfs2_quota_data **qd;
580 	int error;
581 
582 	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
583 		return 0;
584 
585 	error = gfs2_qa_get(ip);
586 	if (error)
587 		return error;
588 
589 	qd = ip->i_qadata->qa_qd;
590 
591 	if (gfs2_assert_warn(sdp, !ip->i_qadata->qa_qd_num) ||
592 	    gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags))) {
593 		error = -EIO;
594 		goto out;
595 	}
596 
597 	error = qdsb_get(sdp, make_kqid_uid(ip->i_inode.i_uid), qd);
598 	if (error)
599 		goto out_unhold;
600 	ip->i_qadata->qa_qd_num++;
601 	qd++;
602 
603 	error = qdsb_get(sdp, make_kqid_gid(ip->i_inode.i_gid), qd);
604 	if (error)
605 		goto out_unhold;
606 	ip->i_qadata->qa_qd_num++;
607 	qd++;
608 
609 	if (!uid_eq(uid, NO_UID_QUOTA_CHANGE) &&
610 	    !uid_eq(uid, ip->i_inode.i_uid)) {
611 		error = qdsb_get(sdp, make_kqid_uid(uid), qd);
612 		if (error)
613 			goto out_unhold;
614 		ip->i_qadata->qa_qd_num++;
615 		qd++;
616 	}
617 
618 	if (!gid_eq(gid, NO_GID_QUOTA_CHANGE) &&
619 	    !gid_eq(gid, ip->i_inode.i_gid)) {
620 		error = qdsb_get(sdp, make_kqid_gid(gid), qd);
621 		if (error)
622 			goto out_unhold;
623 		ip->i_qadata->qa_qd_num++;
624 		qd++;
625 	}
626 
627 out_unhold:
628 	if (error)
629 		gfs2_quota_unhold(ip);
630 out:
631 	return error;
632 }
633 
634 void gfs2_quota_unhold(struct gfs2_inode *ip)
635 {
636 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
637 	u32 x;
638 
639 	if (ip->i_qadata == NULL)
640 		return;
641 
642 	gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags));
643 
644 	for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
645 		qdsb_put(ip->i_qadata->qa_qd[x]);
646 		ip->i_qadata->qa_qd[x] = NULL;
647 	}
648 	ip->i_qadata->qa_qd_num = 0;
649 	gfs2_qa_put(ip);
650 }
651 
652 static int sort_qd(const void *a, const void *b)
653 {
654 	const struct gfs2_quota_data *qd_a = *(const struct gfs2_quota_data **)a;
655 	const struct gfs2_quota_data *qd_b = *(const struct gfs2_quota_data **)b;
656 
657 	if (qid_lt(qd_a->qd_id, qd_b->qd_id))
658 		return -1;
659 	if (qid_lt(qd_b->qd_id, qd_a->qd_id))
660 		return 1;
661 	return 0;
662 }
663 
664 static void do_qc(struct gfs2_quota_data *qd, s64 change)
665 {
666 	struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
667 	struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
668 	struct gfs2_quota_change *qc = qd->qd_bh_qc;
669 	s64 x;
670 
671 	mutex_lock(&sdp->sd_quota_mutex);
672 	gfs2_trans_add_meta(ip->i_gl, qd->qd_bh);
673 
674 	if (!test_bit(QDF_CHANGE, &qd->qd_flags)) {
675 		qc->qc_change = 0;
676 		qc->qc_flags = 0;
677 		if (qd->qd_id.type == USRQUOTA)
678 			qc->qc_flags = cpu_to_be32(GFS2_QCF_USER);
679 		qc->qc_id = cpu_to_be32(from_kqid(&init_user_ns, qd->qd_id));
680 	}
681 
682 	x = be64_to_cpu(qc->qc_change) + change;
683 	qc->qc_change = cpu_to_be64(x);
684 
685 	spin_lock(&qd_lock);
686 	qd->qd_change = x;
687 	spin_unlock(&qd_lock);
688 
689 	if (!x) {
690 		gfs2_assert_warn(sdp, test_bit(QDF_CHANGE, &qd->qd_flags));
691 		clear_bit(QDF_CHANGE, &qd->qd_flags);
692 		qc->qc_flags = 0;
693 		qc->qc_id = 0;
694 		slot_put(qd);
695 		qd_put(qd);
696 	} else if (!test_and_set_bit(QDF_CHANGE, &qd->qd_flags)) {
697 		qd_hold(qd);
698 		slot_hold(qd);
699 	}
700 
701 	if (change < 0) /* Reset quiet flag if we freed some blocks */
702 		clear_bit(QDF_QMSG_QUIET, &qd->qd_flags);
703 	mutex_unlock(&sdp->sd_quota_mutex);
704 }
705 
706 static int gfs2_write_buf_to_page(struct gfs2_inode *ip, unsigned long index,
707 				  unsigned off, void *buf, unsigned bytes)
708 {
709 	struct inode *inode = &ip->i_inode;
710 	struct gfs2_sbd *sdp = GFS2_SB(inode);
711 	struct address_space *mapping = inode->i_mapping;
712 	struct page *page;
713 	struct buffer_head *bh;
714 	void *kaddr;
715 	u64 blk;
716 	unsigned bsize = sdp->sd_sb.sb_bsize, bnum = 0, boff = 0;
717 	unsigned to_write = bytes, pg_off = off;
718 	int done = 0;
719 
720 	blk = index << (PAGE_SHIFT - sdp->sd_sb.sb_bsize_shift);
721 	boff = off % bsize;
722 
723 	page = find_or_create_page(mapping, index, GFP_NOFS);
724 	if (!page)
725 		return -ENOMEM;
726 	if (!page_has_buffers(page))
727 		create_empty_buffers(page, bsize, 0);
728 
729 	bh = page_buffers(page);
730 	while (!done) {
731 		/* Find the beginning block within the page */
732 		if (pg_off >= ((bnum * bsize) + bsize)) {
733 			bh = bh->b_this_page;
734 			bnum++;
735 			blk++;
736 			continue;
737 		}
738 		if (!buffer_mapped(bh)) {
739 			gfs2_block_map(inode, blk, bh, 1);
740 			if (!buffer_mapped(bh))
741 				goto unlock_out;
742 			/* If it's a newly allocated disk block, zero it */
743 			if (buffer_new(bh))
744 				zero_user(page, bnum * bsize, bh->b_size);
745 		}
746 		if (PageUptodate(page))
747 			set_buffer_uptodate(bh);
748 		if (!buffer_uptodate(bh)) {
749 			ll_rw_block(REQ_OP_READ | REQ_META | REQ_PRIO, 1, &bh);
750 			wait_on_buffer(bh);
751 			if (!buffer_uptodate(bh))
752 				goto unlock_out;
753 		}
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 	kaddr = kmap_atomic(page);
771 	memcpy(kaddr + off, buf, bytes);
772 	flush_dcache_page(page);
773 	kunmap_atomic(kaddr);
774 	unlock_page(page);
775 	put_page(page);
776 
777 	return 0;
778 
779 unlock_out:
780 	unlock_page(page);
781 	put_page(page);
782 	return -EIO;
783 }
784 
785 static int gfs2_write_disk_quota(struct gfs2_inode *ip, struct gfs2_quota *qp,
786 				 loff_t loc)
787 {
788 	unsigned long pg_beg;
789 	unsigned pg_off, nbytes, overflow = 0;
790 	int pg_oflow = 0, error;
791 	void *ptr;
792 
793 	nbytes = sizeof(struct gfs2_quota);
794 
795 	pg_beg = loc >> PAGE_SHIFT;
796 	pg_off = offset_in_page(loc);
797 
798 	/* If the quota straddles a page boundary, split the write in two */
799 	if ((pg_off + nbytes) > PAGE_SIZE) {
800 		pg_oflow = 1;
801 		overflow = (pg_off + nbytes) - PAGE_SIZE;
802 	}
803 
804 	ptr = qp;
805 	error = gfs2_write_buf_to_page(ip, pg_beg, pg_off, ptr,
806 				       nbytes - overflow);
807 	/* If there's an overflow, write the remaining bytes to the next page */
808 	if (!error && pg_oflow)
809 		error = gfs2_write_buf_to_page(ip, pg_beg + 1, 0,
810 					       ptr + nbytes - overflow,
811 					       overflow);
812 	return error;
813 }
814 
815 /**
816  * gfs2_adjust_quota - adjust record of current block usage
817  * @ip: The quota inode
818  * @loc: Offset of the entry in the quota file
819  * @change: The amount of usage change to record
820  * @qd: The quota data
821  * @fdq: The updated limits to record
822  *
823  * This function was mostly borrowed from gfs2_block_truncate_page which was
824  * in turn mostly borrowed from ext3
825  *
826  * Returns: 0 or -ve on error
827  */
828 
829 static int gfs2_adjust_quota(struct gfs2_inode *ip, loff_t loc,
830 			     s64 change, struct gfs2_quota_data *qd,
831 			     struct qc_dqblk *fdq)
832 {
833 	struct inode *inode = &ip->i_inode;
834 	struct gfs2_sbd *sdp = GFS2_SB(inode);
835 	struct gfs2_quota q;
836 	int err;
837 	u64 size;
838 
839 	if (gfs2_is_stuffed(ip)) {
840 		err = gfs2_unstuff_dinode(ip);
841 		if (err)
842 			return err;
843 	}
844 
845 	memset(&q, 0, sizeof(struct gfs2_quota));
846 	err = gfs2_internal_read(ip, (char *)&q, &loc, sizeof(q));
847 	if (err < 0)
848 		return err;
849 
850 	loc -= sizeof(q); /* gfs2_internal_read would've advanced the loc ptr */
851 	err = -EIO;
852 	be64_add_cpu(&q.qu_value, change);
853 	if (((s64)be64_to_cpu(q.qu_value)) < 0)
854 		q.qu_value = 0; /* Never go negative on quota usage */
855 	qd->qd_qb.qb_value = q.qu_value;
856 	if (fdq) {
857 		if (fdq->d_fieldmask & QC_SPC_SOFT) {
858 			q.qu_warn = cpu_to_be64(fdq->d_spc_softlimit >> sdp->sd_sb.sb_bsize_shift);
859 			qd->qd_qb.qb_warn = q.qu_warn;
860 		}
861 		if (fdq->d_fieldmask & QC_SPC_HARD) {
862 			q.qu_limit = cpu_to_be64(fdq->d_spc_hardlimit >> sdp->sd_sb.sb_bsize_shift);
863 			qd->qd_qb.qb_limit = q.qu_limit;
864 		}
865 		if (fdq->d_fieldmask & QC_SPACE) {
866 			q.qu_value = cpu_to_be64(fdq->d_space >> sdp->sd_sb.sb_bsize_shift);
867 			qd->qd_qb.qb_value = q.qu_value;
868 		}
869 	}
870 
871 	err = gfs2_write_disk_quota(ip, &q, loc);
872 	if (!err) {
873 		size = loc + sizeof(struct gfs2_quota);
874 		if (size > inode->i_size)
875 			i_size_write(inode, size);
876 		inode->i_mtime = inode->i_atime = current_time(inode);
877 		mark_inode_dirty(inode);
878 		set_bit(QDF_REFRESH, &qd->qd_flags);
879 	}
880 
881 	return err;
882 }
883 
884 static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
885 {
886 	struct gfs2_sbd *sdp = (*qda)->qd_gl->gl_name.ln_sbd;
887 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
888 	struct gfs2_alloc_parms ap = { .aflags = 0, };
889 	unsigned int data_blocks, ind_blocks;
890 	struct gfs2_holder *ghs, i_gh;
891 	unsigned int qx, x;
892 	struct gfs2_quota_data *qd;
893 	unsigned reserved;
894 	loff_t offset;
895 	unsigned int nalloc = 0, blocks;
896 	int error;
897 
898 	error = gfs2_qa_get(ip);
899 	if (error)
900 		return error;
901 
902 	gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
903 			      &data_blocks, &ind_blocks);
904 
905 	ghs = kmalloc_array(num_qd, sizeof(struct gfs2_holder), GFP_NOFS);
906 	if (!ghs) {
907 		error = -ENOMEM;
908 		goto out;
909 	}
910 
911 	sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL);
912 	inode_lock(&ip->i_inode);
913 	for (qx = 0; qx < num_qd; qx++) {
914 		error = gfs2_glock_nq_init(qda[qx]->qd_gl, LM_ST_EXCLUSIVE,
915 					   GL_NOCACHE, &ghs[qx]);
916 		if (error)
917 			goto out_dq;
918 	}
919 
920 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
921 	if (error)
922 		goto out_dq;
923 
924 	for (x = 0; x < num_qd; x++) {
925 		offset = qd2offset(qda[x]);
926 		if (gfs2_write_alloc_required(ip, offset,
927 					      sizeof(struct gfs2_quota)))
928 			nalloc++;
929 	}
930 
931 	/*
932 	 * 1 blk for unstuffing inode if stuffed. We add this extra
933 	 * block to the reservation unconditionally. If the inode
934 	 * doesn't need unstuffing, the block will be released to the
935 	 * rgrp since it won't be allocated during the transaction
936 	 */
937 	/* +3 in the end for unstuffing block, inode size update block
938 	 * and another block in case quota straddles page boundary and
939 	 * two blocks need to be updated instead of 1 */
940 	blocks = num_qd * data_blocks + RES_DINODE + num_qd + 3;
941 
942 	reserved = 1 + (nalloc * (data_blocks + ind_blocks));
943 	ap.target = reserved;
944 	error = gfs2_inplace_reserve(ip, &ap);
945 	if (error)
946 		goto out_alloc;
947 
948 	if (nalloc)
949 		blocks += gfs2_rg_blocks(ip, reserved) + nalloc * ind_blocks + RES_STATFS;
950 
951 	error = gfs2_trans_begin(sdp, blocks, 0);
952 	if (error)
953 		goto out_ipres;
954 
955 	for (x = 0; x < num_qd; x++) {
956 		qd = qda[x];
957 		offset = qd2offset(qd);
958 		error = gfs2_adjust_quota(ip, offset, qd->qd_change_sync, qd, NULL);
959 		if (error)
960 			goto out_end_trans;
961 
962 		do_qc(qd, -qd->qd_change_sync);
963 		set_bit(QDF_REFRESH, &qd->qd_flags);
964 	}
965 
966 	error = 0;
967 
968 out_end_trans:
969 	gfs2_trans_end(sdp);
970 out_ipres:
971 	gfs2_inplace_release(ip);
972 out_alloc:
973 	gfs2_glock_dq_uninit(&i_gh);
974 out_dq:
975 	while (qx--)
976 		gfs2_glock_dq_uninit(&ghs[qx]);
977 	inode_unlock(&ip->i_inode);
978 	kfree(ghs);
979 	gfs2_log_flush(ip->i_gl->gl_name.ln_sbd, ip->i_gl,
980 		       GFS2_LOG_HEAD_FLUSH_NORMAL | GFS2_LFC_DO_SYNC);
981 out:
982 	gfs2_qa_put(ip);
983 	return error;
984 }
985 
986 static int update_qd(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd)
987 {
988 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
989 	struct gfs2_quota q;
990 	struct gfs2_quota_lvb *qlvb;
991 	loff_t pos;
992 	int error;
993 
994 	memset(&q, 0, sizeof(struct gfs2_quota));
995 	pos = qd2offset(qd);
996 	error = gfs2_internal_read(ip, (char *)&q, &pos, sizeof(q));
997 	if (error < 0)
998 		return error;
999 
1000 	qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1001 	qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC);
1002 	qlvb->__pad = 0;
1003 	qlvb->qb_limit = q.qu_limit;
1004 	qlvb->qb_warn = q.qu_warn;
1005 	qlvb->qb_value = q.qu_value;
1006 	qd->qd_qb = *qlvb;
1007 
1008 	return 0;
1009 }
1010 
1011 static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
1012 		    struct gfs2_holder *q_gh)
1013 {
1014 	struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
1015 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1016 	struct gfs2_holder i_gh;
1017 	int error;
1018 
1019 restart:
1020 	error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
1021 	if (error)
1022 		return error;
1023 
1024 	if (test_and_clear_bit(QDF_REFRESH, &qd->qd_flags))
1025 		force_refresh = FORCE;
1026 
1027 	qd->qd_qb = *(struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1028 
1029 	if (force_refresh || qd->qd_qb.qb_magic != cpu_to_be32(GFS2_MAGIC)) {
1030 		gfs2_glock_dq_uninit(q_gh);
1031 		error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE,
1032 					   GL_NOCACHE, q_gh);
1033 		if (error)
1034 			return error;
1035 
1036 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
1037 		if (error)
1038 			goto fail;
1039 
1040 		error = update_qd(sdp, qd);
1041 		if (error)
1042 			goto fail_gunlock;
1043 
1044 		gfs2_glock_dq_uninit(&i_gh);
1045 		gfs2_glock_dq_uninit(q_gh);
1046 		force_refresh = 0;
1047 		goto restart;
1048 	}
1049 
1050 	return 0;
1051 
1052 fail_gunlock:
1053 	gfs2_glock_dq_uninit(&i_gh);
1054 fail:
1055 	gfs2_glock_dq_uninit(q_gh);
1056 	return error;
1057 }
1058 
1059 int gfs2_quota_lock(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
1060 {
1061 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1062 	struct gfs2_quota_data *qd;
1063 	u32 x;
1064 	int error = 0;
1065 
1066 	if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
1067 		return 0;
1068 
1069 	error = gfs2_quota_hold(ip, uid, gid);
1070 	if (error)
1071 		return error;
1072 
1073 	sort(ip->i_qadata->qa_qd, ip->i_qadata->qa_qd_num,
1074 	     sizeof(struct gfs2_quota_data *), sort_qd, NULL);
1075 
1076 	for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1077 		qd = ip->i_qadata->qa_qd[x];
1078 		error = do_glock(qd, NO_FORCE, &ip->i_qadata->qa_qd_ghs[x]);
1079 		if (error)
1080 			break;
1081 	}
1082 
1083 	if (!error)
1084 		set_bit(GIF_QD_LOCKED, &ip->i_flags);
1085 	else {
1086 		while (x--)
1087 			gfs2_glock_dq_uninit(&ip->i_qadata->qa_qd_ghs[x]);
1088 		gfs2_quota_unhold(ip);
1089 	}
1090 
1091 	return error;
1092 }
1093 
1094 static int need_sync(struct gfs2_quota_data *qd)
1095 {
1096 	struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
1097 	struct gfs2_tune *gt = &sdp->sd_tune;
1098 	s64 value;
1099 	unsigned int num, den;
1100 	int do_sync = 1;
1101 
1102 	if (!qd->qd_qb.qb_limit)
1103 		return 0;
1104 
1105 	spin_lock(&qd_lock);
1106 	value = qd->qd_change;
1107 	spin_unlock(&qd_lock);
1108 
1109 	spin_lock(&gt->gt_spin);
1110 	num = gt->gt_quota_scale_num;
1111 	den = gt->gt_quota_scale_den;
1112 	spin_unlock(&gt->gt_spin);
1113 
1114 	if (value < 0)
1115 		do_sync = 0;
1116 	else if ((s64)be64_to_cpu(qd->qd_qb.qb_value) >=
1117 		 (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1118 		do_sync = 0;
1119 	else {
1120 		value *= gfs2_jindex_size(sdp) * num;
1121 		value = div_s64(value, den);
1122 		value += (s64)be64_to_cpu(qd->qd_qb.qb_value);
1123 		if (value < (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1124 			do_sync = 0;
1125 	}
1126 
1127 	return do_sync;
1128 }
1129 
1130 void gfs2_quota_unlock(struct gfs2_inode *ip)
1131 {
1132 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1133 	struct gfs2_quota_data *qda[4];
1134 	unsigned int count = 0;
1135 	u32 x;
1136 	int found;
1137 
1138 	if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
1139 		return;
1140 
1141 	for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1142 		struct gfs2_quota_data *qd;
1143 		int sync;
1144 
1145 		qd = ip->i_qadata->qa_qd[x];
1146 		sync = need_sync(qd);
1147 
1148 		gfs2_glock_dq_uninit(&ip->i_qadata->qa_qd_ghs[x]);
1149 		if (!sync)
1150 			continue;
1151 
1152 		spin_lock(&qd_lock);
1153 		found = qd_check_sync(sdp, qd, NULL);
1154 		spin_unlock(&qd_lock);
1155 
1156 		if (!found)
1157 			continue;
1158 
1159 		gfs2_assert_warn(sdp, qd->qd_change_sync);
1160 		if (bh_get(qd)) {
1161 			clear_bit(QDF_LOCKED, &qd->qd_flags);
1162 			slot_put(qd);
1163 			qd_put(qd);
1164 			continue;
1165 		}
1166 
1167 		qda[count++] = qd;
1168 	}
1169 
1170 	if (count) {
1171 		do_sync(count, qda);
1172 		for (x = 0; x < count; x++)
1173 			qd_unlock(qda[x]);
1174 	}
1175 
1176 	gfs2_quota_unhold(ip);
1177 }
1178 
1179 #define MAX_LINE 256
1180 
1181 static int print_message(struct gfs2_quota_data *qd, char *type)
1182 {
1183 	struct gfs2_sbd *sdp = qd->qd_gl->gl_name.ln_sbd;
1184 
1185 	fs_info(sdp, "quota %s for %s %u\n",
1186 		type,
1187 		(qd->qd_id.type == USRQUOTA) ? "user" : "group",
1188 		from_kqid(&init_user_ns, qd->qd_id));
1189 
1190 	return 0;
1191 }
1192 
1193 /**
1194  * gfs2_quota_check - check if allocating new blocks will exceed quota
1195  * @ip:  The inode for which this check is being performed
1196  * @uid: The uid to check against
1197  * @gid: The gid to check against
1198  * @ap:  The allocation parameters. ap->target contains the requested
1199  *       blocks. ap->min_target, if set, contains the minimum blks
1200  *       requested.
1201  *
1202  * Returns: 0 on success.
1203  *                  min_req = ap->min_target ? ap->min_target : ap->target;
1204  *                  quota must allow at least min_req blks for success and
1205  *                  ap->allowed is set to the number of blocks allowed
1206  *
1207  *          -EDQUOT otherwise, quota violation. ap->allowed is set to number
1208  *                  of blocks available.
1209  */
1210 int gfs2_quota_check(struct gfs2_inode *ip, kuid_t uid, kgid_t gid,
1211 		     struct gfs2_alloc_parms *ap)
1212 {
1213 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1214 	struct gfs2_quota_data *qd;
1215 	s64 value, warn, limit;
1216 	u32 x;
1217 	int error = 0;
1218 
1219 	ap->allowed = UINT_MAX; /* Assume we are permitted a whole lot */
1220 	if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
1221 		return 0;
1222 
1223 	for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1224 		qd = ip->i_qadata->qa_qd[x];
1225 
1226 		if (!(qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1227 		      qid_eq(qd->qd_id, make_kqid_gid(gid))))
1228 			continue;
1229 
1230 		warn = (s64)be64_to_cpu(qd->qd_qb.qb_warn);
1231 		limit = (s64)be64_to_cpu(qd->qd_qb.qb_limit);
1232 		value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
1233 		spin_lock(&qd_lock);
1234 		value += qd->qd_change;
1235 		spin_unlock(&qd_lock);
1236 
1237 		if (limit > 0 && (limit - value) < ap->allowed)
1238 			ap->allowed = limit - value;
1239 		/* If we can't meet the target */
1240 		if (limit && limit < (value + (s64)ap->target)) {
1241 			/* If no min_target specified or we don't meet
1242 			 * min_target, return -EDQUOT */
1243 			if (!ap->min_target || ap->min_target > ap->allowed) {
1244 				if (!test_and_set_bit(QDF_QMSG_QUIET,
1245 						      &qd->qd_flags)) {
1246 					print_message(qd, "exceeded");
1247 					quota_send_warning(qd->qd_id,
1248 							   sdp->sd_vfs->s_dev,
1249 							   QUOTA_NL_BHARDWARN);
1250 				}
1251 				error = -EDQUOT;
1252 				break;
1253 			}
1254 		} else if (warn && warn < value &&
1255 			   time_after_eq(jiffies, qd->qd_last_warn +
1256 					 gfs2_tune_get(sdp, gt_quota_warn_period)
1257 					 * HZ)) {
1258 			quota_send_warning(qd->qd_id,
1259 					   sdp->sd_vfs->s_dev, QUOTA_NL_BSOFTWARN);
1260 			error = print_message(qd, "warning");
1261 			qd->qd_last_warn = jiffies;
1262 		}
1263 	}
1264 	return error;
1265 }
1266 
1267 void gfs2_quota_change(struct gfs2_inode *ip, s64 change,
1268 		       kuid_t uid, kgid_t gid)
1269 {
1270 	struct gfs2_quota_data *qd;
1271 	u32 x;
1272 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1273 
1274 	if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON ||
1275 	    gfs2_assert_warn(sdp, change))
1276 		return;
1277 	if (ip->i_diskflags & GFS2_DIF_SYSTEM)
1278 		return;
1279 
1280 	if (gfs2_assert_withdraw(sdp, ip->i_qadata &&
1281 				 ip->i_qadata->qa_ref > 0))
1282 		return;
1283 	for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1284 		qd = ip->i_qadata->qa_qd[x];
1285 
1286 		if (qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1287 		    qid_eq(qd->qd_id, make_kqid_gid(gid))) {
1288 			do_qc(qd, change);
1289 		}
1290 	}
1291 }
1292 
1293 int gfs2_quota_sync(struct super_block *sb, int type)
1294 {
1295 	struct gfs2_sbd *sdp = sb->s_fs_info;
1296 	struct gfs2_quota_data **qda;
1297 	unsigned int max_qd = PAGE_SIZE / sizeof(struct gfs2_holder);
1298 	unsigned int num_qd;
1299 	unsigned int x;
1300 	int error = 0;
1301 
1302 	qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
1303 	if (!qda)
1304 		return -ENOMEM;
1305 
1306 	mutex_lock(&sdp->sd_quota_sync_mutex);
1307 	sdp->sd_quota_sync_gen++;
1308 
1309 	do {
1310 		num_qd = 0;
1311 
1312 		for (;;) {
1313 			error = qd_fish(sdp, qda + num_qd);
1314 			if (error || !qda[num_qd])
1315 				break;
1316 			if (++num_qd == max_qd)
1317 				break;
1318 		}
1319 
1320 		if (num_qd) {
1321 			if (!error)
1322 				error = do_sync(num_qd, qda);
1323 			if (!error)
1324 				for (x = 0; x < num_qd; x++)
1325 					qda[x]->qd_sync_gen =
1326 						sdp->sd_quota_sync_gen;
1327 
1328 			for (x = 0; x < num_qd; x++)
1329 				qd_unlock(qda[x]);
1330 		}
1331 	} while (!error && num_qd == max_qd);
1332 
1333 	mutex_unlock(&sdp->sd_quota_sync_mutex);
1334 	kfree(qda);
1335 
1336 	return error;
1337 }
1338 
1339 int gfs2_quota_refresh(struct gfs2_sbd *sdp, struct kqid qid)
1340 {
1341 	struct gfs2_quota_data *qd;
1342 	struct gfs2_holder q_gh;
1343 	int error;
1344 
1345 	error = qd_get(sdp, qid, &qd);
1346 	if (error)
1347 		return error;
1348 
1349 	error = do_glock(qd, FORCE, &q_gh);
1350 	if (!error)
1351 		gfs2_glock_dq_uninit(&q_gh);
1352 
1353 	qd_put(qd);
1354 	return error;
1355 }
1356 
1357 int gfs2_quota_init(struct gfs2_sbd *sdp)
1358 {
1359 	struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
1360 	u64 size = i_size_read(sdp->sd_qc_inode);
1361 	unsigned int blocks = size >> sdp->sd_sb.sb_bsize_shift;
1362 	unsigned int x, slot = 0;
1363 	unsigned int found = 0;
1364 	unsigned int hash;
1365 	unsigned int bm_size;
1366 	u64 dblock;
1367 	u32 extlen = 0;
1368 	int error;
1369 
1370 	if (gfs2_check_internal_file_size(sdp->sd_qc_inode, 1, 64 << 20))
1371 		return -EIO;
1372 
1373 	sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
1374 	bm_size = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * sizeof(unsigned long));
1375 	bm_size *= sizeof(unsigned long);
1376 	error = -ENOMEM;
1377 	sdp->sd_quota_bitmap = kzalloc(bm_size, GFP_NOFS | __GFP_NOWARN);
1378 	if (sdp->sd_quota_bitmap == NULL)
1379 		sdp->sd_quota_bitmap = __vmalloc(bm_size, GFP_NOFS |
1380 						 __GFP_ZERO);
1381 	if (!sdp->sd_quota_bitmap)
1382 		return error;
1383 
1384 	for (x = 0; x < blocks; x++) {
1385 		struct buffer_head *bh;
1386 		const struct gfs2_quota_change *qc;
1387 		unsigned int y;
1388 
1389 		if (!extlen) {
1390 			extlen = 32;
1391 			error = gfs2_get_extent(&ip->i_inode, x, &dblock, &extlen);
1392 			if (error)
1393 				goto fail;
1394 		}
1395 		error = -EIO;
1396 		bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
1397 		if (!bh)
1398 			goto fail;
1399 		if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
1400 			brelse(bh);
1401 			goto fail;
1402 		}
1403 
1404 		qc = (const struct gfs2_quota_change *)(bh->b_data + sizeof(struct gfs2_meta_header));
1405 		for (y = 0; y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1406 		     y++, slot++) {
1407 			struct gfs2_quota_data *qd;
1408 			s64 qc_change = be64_to_cpu(qc->qc_change);
1409 			u32 qc_flags = be32_to_cpu(qc->qc_flags);
1410 			enum quota_type qtype = (qc_flags & GFS2_QCF_USER) ?
1411 						USRQUOTA : GRPQUOTA;
1412 			struct kqid qc_id = make_kqid(&init_user_ns, qtype,
1413 						      be32_to_cpu(qc->qc_id));
1414 			qc++;
1415 			if (!qc_change)
1416 				continue;
1417 
1418 			hash = gfs2_qd_hash(sdp, qc_id);
1419 			qd = qd_alloc(hash, sdp, qc_id);
1420 			if (qd == NULL) {
1421 				brelse(bh);
1422 				goto fail;
1423 			}
1424 
1425 			set_bit(QDF_CHANGE, &qd->qd_flags);
1426 			qd->qd_change = qc_change;
1427 			qd->qd_slot = slot;
1428 			qd->qd_slot_count = 1;
1429 
1430 			spin_lock(&qd_lock);
1431 			BUG_ON(test_and_set_bit(slot, sdp->sd_quota_bitmap));
1432 			list_add(&qd->qd_list, &sdp->sd_quota_list);
1433 			atomic_inc(&sdp->sd_quota_count);
1434 			spin_unlock(&qd_lock);
1435 
1436 			spin_lock_bucket(hash);
1437 			hlist_bl_add_head_rcu(&qd->qd_hlist, &qd_hash_table[hash]);
1438 			spin_unlock_bucket(hash);
1439 
1440 			found++;
1441 		}
1442 
1443 		brelse(bh);
1444 		dblock++;
1445 		extlen--;
1446 	}
1447 
1448 	if (found)
1449 		fs_info(sdp, "found %u quota changes\n", found);
1450 
1451 	return 0;
1452 
1453 fail:
1454 	gfs2_quota_cleanup(sdp);
1455 	return error;
1456 }
1457 
1458 void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1459 {
1460 	struct list_head *head = &sdp->sd_quota_list;
1461 	struct gfs2_quota_data *qd;
1462 
1463 	spin_lock(&qd_lock);
1464 	while (!list_empty(head)) {
1465 		qd = list_last_entry(head, struct gfs2_quota_data, qd_list);
1466 
1467 		list_del(&qd->qd_list);
1468 
1469 		/* Also remove if this qd exists in the reclaim list */
1470 		list_lru_del(&gfs2_qd_lru, &qd->qd_lru);
1471 		atomic_dec(&sdp->sd_quota_count);
1472 		spin_unlock(&qd_lock);
1473 
1474 		spin_lock_bucket(qd->qd_hash);
1475 		hlist_bl_del_rcu(&qd->qd_hlist);
1476 		spin_unlock_bucket(qd->qd_hash);
1477 
1478 		gfs2_assert_warn(sdp, !qd->qd_change);
1479 		gfs2_assert_warn(sdp, !qd->qd_slot_count);
1480 		gfs2_assert_warn(sdp, !qd->qd_bh_count);
1481 
1482 		gfs2_glock_put(qd->qd_gl);
1483 		call_rcu(&qd->qd_rcu, gfs2_qd_dealloc);
1484 
1485 		spin_lock(&qd_lock);
1486 	}
1487 	spin_unlock(&qd_lock);
1488 
1489 	gfs2_assert_warn(sdp, !atomic_read(&sdp->sd_quota_count));
1490 
1491 	kvfree(sdp->sd_quota_bitmap);
1492 	sdp->sd_quota_bitmap = NULL;
1493 }
1494 
1495 static void quotad_error(struct gfs2_sbd *sdp, const char *msg, int error)
1496 {
1497 	if (error == 0 || error == -EROFS)
1498 		return;
1499 	if (!gfs2_withdrawn(sdp)) {
1500 		if (!cmpxchg(&sdp->sd_log_error, 0, error))
1501 			fs_err(sdp, "gfs2_quotad: %s error %d\n", msg, error);
1502 		wake_up(&sdp->sd_logd_waitq);
1503 	}
1504 }
1505 
1506 static void quotad_check_timeo(struct gfs2_sbd *sdp, const char *msg,
1507 			       int (*fxn)(struct super_block *sb, int type),
1508 			       unsigned long t, unsigned long *timeo,
1509 			       unsigned int *new_timeo)
1510 {
1511 	if (t >= *timeo) {
1512 		int error = fxn(sdp->sd_vfs, 0);
1513 		quotad_error(sdp, msg, error);
1514 		*timeo = gfs2_tune_get_i(&sdp->sd_tune, new_timeo) * HZ;
1515 	} else {
1516 		*timeo -= t;
1517 	}
1518 }
1519 
1520 void gfs2_wake_up_statfs(struct gfs2_sbd *sdp) {
1521 	if (!sdp->sd_statfs_force_sync) {
1522 		sdp->sd_statfs_force_sync = 1;
1523 		wake_up(&sdp->sd_quota_wait);
1524 	}
1525 }
1526 
1527 
1528 /**
1529  * gfs2_quotad - Write cached quota changes into the quota file
1530  * @data: Pointer to GFS2 superblock
1531  *
1532  */
1533 
1534 int gfs2_quotad(void *data)
1535 {
1536 	struct gfs2_sbd *sdp = data;
1537 	struct gfs2_tune *tune = &sdp->sd_tune;
1538 	unsigned long statfs_timeo = 0;
1539 	unsigned long quotad_timeo = 0;
1540 	unsigned long t = 0;
1541 	DEFINE_WAIT(wait);
1542 
1543 	while (!kthread_should_stop()) {
1544 
1545 		if (gfs2_withdrawn(sdp))
1546 			goto bypass;
1547 		/* Update the master statfs file */
1548 		if (sdp->sd_statfs_force_sync) {
1549 			int error = gfs2_statfs_sync(sdp->sd_vfs, 0);
1550 			quotad_error(sdp, "statfs", error);
1551 			statfs_timeo = gfs2_tune_get(sdp, gt_statfs_quantum) * HZ;
1552 		}
1553 		else
1554 			quotad_check_timeo(sdp, "statfs", gfs2_statfs_sync, t,
1555 				   	   &statfs_timeo,
1556 					   &tune->gt_statfs_quantum);
1557 
1558 		/* Update quota file */
1559 		quotad_check_timeo(sdp, "sync", gfs2_quota_sync, t,
1560 				   &quotad_timeo, &tune->gt_quota_quantum);
1561 
1562 		try_to_freeze();
1563 
1564 bypass:
1565 		t = min(quotad_timeo, statfs_timeo);
1566 
1567 		prepare_to_wait(&sdp->sd_quota_wait, &wait, TASK_INTERRUPTIBLE);
1568 		if (!sdp->sd_statfs_force_sync)
1569 			t -= schedule_timeout(t);
1570 		else
1571 			t = 0;
1572 		finish_wait(&sdp->sd_quota_wait, &wait);
1573 	}
1574 
1575 	return 0;
1576 }
1577 
1578 static int gfs2_quota_get_state(struct super_block *sb, struct qc_state *state)
1579 {
1580 	struct gfs2_sbd *sdp = sb->s_fs_info;
1581 
1582 	memset(state, 0, sizeof(*state));
1583 
1584 	switch (sdp->sd_args.ar_quota) {
1585 	case GFS2_QUOTA_ON:
1586 		state->s_state[USRQUOTA].flags |= QCI_LIMITS_ENFORCED;
1587 		state->s_state[GRPQUOTA].flags |= QCI_LIMITS_ENFORCED;
1588 		fallthrough;
1589 	case GFS2_QUOTA_ACCOUNT:
1590 		state->s_state[USRQUOTA].flags |= QCI_ACCT_ENABLED |
1591 						  QCI_SYSFILE;
1592 		state->s_state[GRPQUOTA].flags |= QCI_ACCT_ENABLED |
1593 						  QCI_SYSFILE;
1594 		break;
1595 	case GFS2_QUOTA_OFF:
1596 		break;
1597 	}
1598 	if (sdp->sd_quota_inode) {
1599 		state->s_state[USRQUOTA].ino =
1600 					GFS2_I(sdp->sd_quota_inode)->i_no_addr;
1601 		state->s_state[USRQUOTA].blocks = sdp->sd_quota_inode->i_blocks;
1602 	}
1603 	state->s_state[USRQUOTA].nextents = 1;	/* unsupported */
1604 	state->s_state[GRPQUOTA] = state->s_state[USRQUOTA];
1605 	state->s_incoredqs = list_lru_count(&gfs2_qd_lru);
1606 	return 0;
1607 }
1608 
1609 static int gfs2_get_dqblk(struct super_block *sb, struct kqid qid,
1610 			  struct qc_dqblk *fdq)
1611 {
1612 	struct gfs2_sbd *sdp = sb->s_fs_info;
1613 	struct gfs2_quota_lvb *qlvb;
1614 	struct gfs2_quota_data *qd;
1615 	struct gfs2_holder q_gh;
1616 	int error;
1617 
1618 	memset(fdq, 0, sizeof(*fdq));
1619 
1620 	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1621 		return -ESRCH; /* Crazy XFS error code */
1622 
1623 	if ((qid.type != USRQUOTA) &&
1624 	    (qid.type != GRPQUOTA))
1625 		return -EINVAL;
1626 
1627 	error = qd_get(sdp, qid, &qd);
1628 	if (error)
1629 		return error;
1630 	error = do_glock(qd, FORCE, &q_gh);
1631 	if (error)
1632 		goto out;
1633 
1634 	qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1635 	fdq->d_spc_hardlimit = be64_to_cpu(qlvb->qb_limit) << sdp->sd_sb.sb_bsize_shift;
1636 	fdq->d_spc_softlimit = be64_to_cpu(qlvb->qb_warn) << sdp->sd_sb.sb_bsize_shift;
1637 	fdq->d_space = be64_to_cpu(qlvb->qb_value) << sdp->sd_sb.sb_bsize_shift;
1638 
1639 	gfs2_glock_dq_uninit(&q_gh);
1640 out:
1641 	qd_put(qd);
1642 	return error;
1643 }
1644 
1645 /* GFS2 only supports a subset of the XFS fields */
1646 #define GFS2_FIELDMASK (QC_SPC_SOFT|QC_SPC_HARD|QC_SPACE)
1647 
1648 static int gfs2_set_dqblk(struct super_block *sb, struct kqid qid,
1649 			  struct qc_dqblk *fdq)
1650 {
1651 	struct gfs2_sbd *sdp = sb->s_fs_info;
1652 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1653 	struct gfs2_quota_data *qd;
1654 	struct gfs2_holder q_gh, i_gh;
1655 	unsigned int data_blocks, ind_blocks;
1656 	unsigned int blocks = 0;
1657 	int alloc_required;
1658 	loff_t offset;
1659 	int error;
1660 
1661 	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1662 		return -ESRCH; /* Crazy XFS error code */
1663 
1664 	if ((qid.type != USRQUOTA) &&
1665 	    (qid.type != GRPQUOTA))
1666 		return -EINVAL;
1667 
1668 	if (fdq->d_fieldmask & ~GFS2_FIELDMASK)
1669 		return -EINVAL;
1670 
1671 	error = qd_get(sdp, qid, &qd);
1672 	if (error)
1673 		return error;
1674 
1675 	error = gfs2_qa_get(ip);
1676 	if (error)
1677 		goto out_put;
1678 
1679 	inode_lock(&ip->i_inode);
1680 	error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE, 0, &q_gh);
1681 	if (error)
1682 		goto out_unlockput;
1683 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1684 	if (error)
1685 		goto out_q;
1686 
1687 	/* Check for existing entry, if none then alloc new blocks */
1688 	error = update_qd(sdp, qd);
1689 	if (error)
1690 		goto out_i;
1691 
1692 	/* If nothing has changed, this is a no-op */
1693 	if ((fdq->d_fieldmask & QC_SPC_SOFT) &&
1694 	    ((fdq->d_spc_softlimit >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_warn)))
1695 		fdq->d_fieldmask ^= QC_SPC_SOFT;
1696 
1697 	if ((fdq->d_fieldmask & QC_SPC_HARD) &&
1698 	    ((fdq->d_spc_hardlimit >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_limit)))
1699 		fdq->d_fieldmask ^= QC_SPC_HARD;
1700 
1701 	if ((fdq->d_fieldmask & QC_SPACE) &&
1702 	    ((fdq->d_space >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_value)))
1703 		fdq->d_fieldmask ^= QC_SPACE;
1704 
1705 	if (fdq->d_fieldmask == 0)
1706 		goto out_i;
1707 
1708 	offset = qd2offset(qd);
1709 	alloc_required = gfs2_write_alloc_required(ip, offset, sizeof(struct gfs2_quota));
1710 	if (gfs2_is_stuffed(ip))
1711 		alloc_required = 1;
1712 	if (alloc_required) {
1713 		struct gfs2_alloc_parms ap = { .aflags = 0, };
1714 		gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
1715 				       &data_blocks, &ind_blocks);
1716 		blocks = 1 + data_blocks + ind_blocks;
1717 		ap.target = blocks;
1718 		error = gfs2_inplace_reserve(ip, &ap);
1719 		if (error)
1720 			goto out_i;
1721 		blocks += gfs2_rg_blocks(ip, blocks);
1722 	}
1723 
1724 	/* Some quotas span block boundaries and can update two blocks,
1725 	   adding an extra block to the transaction to handle such quotas */
1726 	error = gfs2_trans_begin(sdp, blocks + RES_DINODE + 2, 0);
1727 	if (error)
1728 		goto out_release;
1729 
1730 	/* Apply changes */
1731 	error = gfs2_adjust_quota(ip, offset, 0, qd, fdq);
1732 	if (!error)
1733 		clear_bit(QDF_QMSG_QUIET, &qd->qd_flags);
1734 
1735 	gfs2_trans_end(sdp);
1736 out_release:
1737 	if (alloc_required)
1738 		gfs2_inplace_release(ip);
1739 out_i:
1740 	gfs2_glock_dq_uninit(&i_gh);
1741 out_q:
1742 	gfs2_glock_dq_uninit(&q_gh);
1743 out_unlockput:
1744 	gfs2_qa_put(ip);
1745 	inode_unlock(&ip->i_inode);
1746 out_put:
1747 	qd_put(qd);
1748 	return error;
1749 }
1750 
1751 const struct quotactl_ops gfs2_quotactl_ops = {
1752 	.quota_sync     = gfs2_quota_sync,
1753 	.get_state	= gfs2_quota_get_state,
1754 	.get_dqblk	= gfs2_get_dqblk,
1755 	.set_dqblk	= gfs2_set_dqblk,
1756 };
1757 
1758 void __init gfs2_quota_hash_init(void)
1759 {
1760 	unsigned i;
1761 
1762 	for(i = 0; i < GFS2_QD_HASH_SIZE; i++)
1763 		INIT_HLIST_BL_HEAD(&qd_hash_table[i]);
1764 }
1765