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
gfs2_qd_hash(const struct gfs2_sbd * sdp,const struct kqid qid)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
spin_lock_bucket(unsigned int hash)96 static inline void spin_lock_bucket(unsigned int hash)
97 {
98 hlist_bl_lock(&qd_hash_table[hash]);
99 }
100
spin_unlock_bucket(unsigned int hash)101 static inline void spin_unlock_bucket(unsigned int hash)
102 {
103 hlist_bl_unlock(&qd_hash_table[hash]);
104 }
105
gfs2_qd_dealloc(struct rcu_head * rcu)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 struct gfs2_sbd *sdp = qd->qd_sbd;
110
111 kmem_cache_free(gfs2_quotad_cachep, qd);
112 if (atomic_dec_and_test(&sdp->sd_quota_count))
113 wake_up(&sdp->sd_kill_wait);
114 }
115
gfs2_qd_dispose(struct gfs2_quota_data * qd)116 static void gfs2_qd_dispose(struct gfs2_quota_data *qd)
117 {
118 struct gfs2_sbd *sdp = qd->qd_sbd;
119
120 spin_lock(&qd_lock);
121 list_del(&qd->qd_list);
122 spin_unlock(&qd_lock);
123
124 spin_lock_bucket(qd->qd_hash);
125 hlist_bl_del_rcu(&qd->qd_hlist);
126 spin_unlock_bucket(qd->qd_hash);
127
128 if (!gfs2_withdrawn(sdp)) {
129 gfs2_assert_warn(sdp, !qd->qd_change);
130 gfs2_assert_warn(sdp, !qd->qd_slot_ref);
131 gfs2_assert_warn(sdp, !qd->qd_bh_count);
132 }
133
134 gfs2_glock_put(qd->qd_gl);
135 call_rcu(&qd->qd_rcu, gfs2_qd_dealloc);
136 }
137
gfs2_qd_list_dispose(struct list_head * list)138 static void gfs2_qd_list_dispose(struct list_head *list)
139 {
140 struct gfs2_quota_data *qd;
141
142 while (!list_empty(list)) {
143 qd = list_first_entry(list, struct gfs2_quota_data, qd_lru);
144 list_del(&qd->qd_lru);
145
146 gfs2_qd_dispose(qd);
147 }
148 }
149
150
gfs2_qd_isolate(struct list_head * item,struct list_lru_one * lru,void * arg)151 static enum lru_status gfs2_qd_isolate(struct list_head *item,
152 struct list_lru_one *lru, void *arg)
153 {
154 struct list_head *dispose = arg;
155 struct gfs2_quota_data *qd =
156 list_entry(item, struct gfs2_quota_data, qd_lru);
157 enum lru_status status;
158
159 if (!spin_trylock(&qd->qd_lockref.lock))
160 return LRU_SKIP;
161
162 status = LRU_SKIP;
163 if (qd->qd_lockref.count == 0) {
164 lockref_mark_dead(&qd->qd_lockref);
165 list_lru_isolate_move(lru, &qd->qd_lru, dispose);
166 status = LRU_REMOVED;
167 }
168
169 spin_unlock(&qd->qd_lockref.lock);
170 return status;
171 }
172
gfs2_qd_shrink_scan(struct shrinker * shrink,struct shrink_control * sc)173 static unsigned long gfs2_qd_shrink_scan(struct shrinker *shrink,
174 struct shrink_control *sc)
175 {
176 LIST_HEAD(dispose);
177 unsigned long freed;
178
179 if (!(sc->gfp_mask & __GFP_FS))
180 return SHRINK_STOP;
181
182 freed = list_lru_shrink_walk(&gfs2_qd_lru, sc,
183 gfs2_qd_isolate, &dispose);
184
185 gfs2_qd_list_dispose(&dispose);
186
187 return freed;
188 }
189
gfs2_qd_shrink_count(struct shrinker * shrink,struct shrink_control * sc)190 static unsigned long gfs2_qd_shrink_count(struct shrinker *shrink,
191 struct shrink_control *sc)
192 {
193 return vfs_pressure_ratio(list_lru_shrink_count(&gfs2_qd_lru, sc));
194 }
195
196 static struct shrinker *gfs2_qd_shrinker;
197
gfs2_qd_shrinker_init(void)198 int __init gfs2_qd_shrinker_init(void)
199 {
200 gfs2_qd_shrinker = shrinker_alloc(SHRINKER_NUMA_AWARE, "gfs2-qd");
201 if (!gfs2_qd_shrinker)
202 return -ENOMEM;
203
204 gfs2_qd_shrinker->count_objects = gfs2_qd_shrink_count;
205 gfs2_qd_shrinker->scan_objects = gfs2_qd_shrink_scan;
206
207 shrinker_register(gfs2_qd_shrinker);
208
209 return 0;
210 }
211
gfs2_qd_shrinker_exit(void)212 void gfs2_qd_shrinker_exit(void)
213 {
214 shrinker_free(gfs2_qd_shrinker);
215 }
216
qd2index(struct gfs2_quota_data * qd)217 static u64 qd2index(struct gfs2_quota_data *qd)
218 {
219 struct kqid qid = qd->qd_id;
220 return (2 * (u64)from_kqid(&init_user_ns, qid)) +
221 ((qid.type == USRQUOTA) ? 0 : 1);
222 }
223
qd2offset(struct gfs2_quota_data * qd)224 static u64 qd2offset(struct gfs2_quota_data *qd)
225 {
226 return qd2index(qd) * sizeof(struct gfs2_quota);
227 }
228
qd_alloc(unsigned hash,struct gfs2_sbd * sdp,struct kqid qid)229 static struct gfs2_quota_data *qd_alloc(unsigned hash, struct gfs2_sbd *sdp, struct kqid qid)
230 {
231 struct gfs2_quota_data *qd;
232 int error;
233
234 qd = kmem_cache_zalloc(gfs2_quotad_cachep, GFP_NOFS);
235 if (!qd)
236 return NULL;
237
238 qd->qd_sbd = sdp;
239 lockref_init(&qd->qd_lockref);
240 qd->qd_id = qid;
241 qd->qd_slot = -1;
242 INIT_LIST_HEAD(&qd->qd_lru);
243 qd->qd_hash = hash;
244
245 error = gfs2_glock_get(sdp, qd2index(qd),
246 &gfs2_quota_glops, CREATE, &qd->qd_gl);
247 if (error)
248 goto fail;
249
250 return qd;
251
252 fail:
253 kmem_cache_free(gfs2_quotad_cachep, qd);
254 return NULL;
255 }
256
gfs2_qd_search_bucket(unsigned int hash,const struct gfs2_sbd * sdp,struct kqid qid)257 static struct gfs2_quota_data *gfs2_qd_search_bucket(unsigned int hash,
258 const struct gfs2_sbd *sdp,
259 struct kqid qid)
260 {
261 struct gfs2_quota_data *qd;
262 struct hlist_bl_node *h;
263
264 hlist_bl_for_each_entry_rcu(qd, h, &qd_hash_table[hash], qd_hlist) {
265 if (!qid_eq(qd->qd_id, qid))
266 continue;
267 if (qd->qd_sbd != sdp)
268 continue;
269 if (lockref_get_not_dead(&qd->qd_lockref)) {
270 list_lru_del_obj(&gfs2_qd_lru, &qd->qd_lru);
271 return qd;
272 }
273 }
274
275 return NULL;
276 }
277
278
qd_get(struct gfs2_sbd * sdp,struct kqid qid,struct gfs2_quota_data ** qdp)279 static int qd_get(struct gfs2_sbd *sdp, struct kqid qid,
280 struct gfs2_quota_data **qdp)
281 {
282 struct gfs2_quota_data *qd, *new_qd;
283 unsigned int hash = gfs2_qd_hash(sdp, qid);
284
285 rcu_read_lock();
286 *qdp = qd = gfs2_qd_search_bucket(hash, sdp, qid);
287 rcu_read_unlock();
288
289 if (qd)
290 return 0;
291
292 new_qd = qd_alloc(hash, sdp, qid);
293 if (!new_qd)
294 return -ENOMEM;
295
296 spin_lock(&qd_lock);
297 spin_lock_bucket(hash);
298 *qdp = qd = gfs2_qd_search_bucket(hash, sdp, qid);
299 if (qd == NULL) {
300 *qdp = new_qd;
301 list_add(&new_qd->qd_list, &sdp->sd_quota_list);
302 hlist_bl_add_head_rcu(&new_qd->qd_hlist, &qd_hash_table[hash]);
303 atomic_inc(&sdp->sd_quota_count);
304 }
305 spin_unlock_bucket(hash);
306 spin_unlock(&qd_lock);
307
308 if (qd) {
309 gfs2_glock_put(new_qd->qd_gl);
310 kmem_cache_free(gfs2_quotad_cachep, new_qd);
311 }
312
313 return 0;
314 }
315
316
__qd_hold(struct gfs2_quota_data * qd)317 static void __qd_hold(struct gfs2_quota_data *qd)
318 {
319 struct gfs2_sbd *sdp = qd->qd_sbd;
320 gfs2_assert(sdp, qd->qd_lockref.count > 0);
321 qd->qd_lockref.count++;
322 }
323
qd_put(struct gfs2_quota_data * qd)324 static void qd_put(struct gfs2_quota_data *qd)
325 {
326 struct gfs2_sbd *sdp;
327
328 if (lockref_put_or_lock(&qd->qd_lockref))
329 return;
330
331 BUG_ON(__lockref_is_dead(&qd->qd_lockref));
332 sdp = qd->qd_sbd;
333 if (unlikely(!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))) {
334 lockref_mark_dead(&qd->qd_lockref);
335 spin_unlock(&qd->qd_lockref.lock);
336
337 list_lru_del_obj(&gfs2_qd_lru, &qd->qd_lru);
338 gfs2_qd_dispose(qd);
339 return;
340 }
341
342 qd->qd_lockref.count = 0;
343 list_lru_add_obj(&gfs2_qd_lru, &qd->qd_lru);
344 spin_unlock(&qd->qd_lockref.lock);
345 }
346
slot_get(struct gfs2_quota_data * qd)347 static int slot_get(struct gfs2_quota_data *qd)
348 {
349 struct gfs2_sbd *sdp = qd->qd_sbd;
350 unsigned int bit;
351 int error = 0;
352
353 spin_lock(&sdp->sd_bitmap_lock);
354 if (qd->qd_slot_ref == 0) {
355 bit = find_first_zero_bit(sdp->sd_quota_bitmap,
356 sdp->sd_quota_slots);
357 if (bit >= sdp->sd_quota_slots) {
358 error = -ENOSPC;
359 goto out;
360 }
361 set_bit(bit, sdp->sd_quota_bitmap);
362 qd->qd_slot = bit;
363 }
364 qd->qd_slot_ref++;
365 out:
366 spin_unlock(&sdp->sd_bitmap_lock);
367 return error;
368 }
369
slot_hold(struct gfs2_quota_data * qd)370 static void slot_hold(struct gfs2_quota_data *qd)
371 {
372 struct gfs2_sbd *sdp = qd->qd_sbd;
373
374 spin_lock(&sdp->sd_bitmap_lock);
375 gfs2_assert(sdp, qd->qd_slot_ref);
376 qd->qd_slot_ref++;
377 spin_unlock(&sdp->sd_bitmap_lock);
378 }
379
slot_put(struct gfs2_quota_data * qd)380 static void slot_put(struct gfs2_quota_data *qd)
381 {
382 struct gfs2_sbd *sdp = qd->qd_sbd;
383
384 spin_lock(&sdp->sd_bitmap_lock);
385 gfs2_assert(sdp, qd->qd_slot_ref);
386 if (!--qd->qd_slot_ref) {
387 BUG_ON(!test_and_clear_bit(qd->qd_slot, sdp->sd_quota_bitmap));
388 qd->qd_slot = -1;
389 }
390 spin_unlock(&sdp->sd_bitmap_lock);
391 }
392
bh_get(struct gfs2_quota_data * qd)393 static int bh_get(struct gfs2_quota_data *qd)
394 {
395 struct gfs2_sbd *sdp = qd->qd_sbd;
396 struct inode *inode = sdp->sd_qc_inode;
397 struct gfs2_inode *ip = GFS2_I(inode);
398 unsigned int block, offset;
399 struct buffer_head *bh = NULL;
400 struct iomap iomap = { };
401 int error;
402
403 spin_lock(&qd->qd_lockref.lock);
404 if (qd->qd_bh_count) {
405 qd->qd_bh_count++;
406 spin_unlock(&qd->qd_lockref.lock);
407 return 0;
408 }
409 spin_unlock(&qd->qd_lockref.lock);
410
411 block = qd->qd_slot / sdp->sd_qc_per_block;
412 offset = qd->qd_slot % sdp->sd_qc_per_block;
413
414 error = gfs2_iomap_get(inode,
415 (loff_t)block << inode->i_blkbits,
416 i_blocksize(inode), &iomap);
417 if (error)
418 return error;
419 error = -ENOENT;
420 if (iomap.type != IOMAP_MAPPED)
421 return error;
422
423 error = gfs2_meta_read(ip->i_gl, iomap.addr >> inode->i_blkbits,
424 DIO_WAIT, 0, &bh);
425 if (error)
426 return error;
427 error = -EIO;
428 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC))
429 goto out;
430
431 spin_lock(&qd->qd_lockref.lock);
432 if (qd->qd_bh == NULL) {
433 qd->qd_bh = bh;
434 qd->qd_bh_qc = (struct gfs2_quota_change *)
435 (bh->b_data + sizeof(struct gfs2_meta_header) +
436 offset * sizeof(struct gfs2_quota_change));
437 bh = NULL;
438 }
439 qd->qd_bh_count++;
440 spin_unlock(&qd->qd_lockref.lock);
441 error = 0;
442
443 out:
444 brelse(bh);
445 return error;
446 }
447
bh_put(struct gfs2_quota_data * qd)448 static void bh_put(struct gfs2_quota_data *qd)
449 {
450 struct gfs2_sbd *sdp = qd->qd_sbd;
451 struct buffer_head *bh = NULL;
452
453 spin_lock(&qd->qd_lockref.lock);
454 gfs2_assert(sdp, qd->qd_bh_count);
455 if (!--qd->qd_bh_count) {
456 bh = qd->qd_bh;
457 qd->qd_bh = NULL;
458 qd->qd_bh_qc = NULL;
459 }
460 spin_unlock(&qd->qd_lockref.lock);
461 brelse(bh);
462 }
463
qd_grab_sync(struct gfs2_sbd * sdp,struct gfs2_quota_data * qd,u64 sync_gen)464 static bool qd_grab_sync(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd,
465 u64 sync_gen)
466 {
467 bool ret = false;
468
469 spin_lock(&qd->qd_lockref.lock);
470 if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
471 !test_bit(QDF_CHANGE, &qd->qd_flags) ||
472 qd->qd_sync_gen >= sync_gen)
473 goto out;
474
475 if (__lockref_is_dead(&qd->qd_lockref))
476 goto out;
477 qd->qd_lockref.count++;
478
479 list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
480 set_bit(QDF_LOCKED, &qd->qd_flags);
481 qd->qd_change_sync = qd->qd_change;
482 slot_hold(qd);
483 ret = true;
484
485 out:
486 spin_unlock(&qd->qd_lockref.lock);
487 return ret;
488 }
489
qd_ungrab_sync(struct gfs2_quota_data * qd)490 static void qd_ungrab_sync(struct gfs2_quota_data *qd)
491 {
492 clear_bit(QDF_LOCKED, &qd->qd_flags);
493 slot_put(qd);
494 qd_put(qd);
495 }
496
qdsb_put(struct gfs2_quota_data * qd)497 static void qdsb_put(struct gfs2_quota_data *qd)
498 {
499 bh_put(qd);
500 slot_put(qd);
501 qd_put(qd);
502 }
503
qd_unlock(struct gfs2_quota_data * qd)504 static void qd_unlock(struct gfs2_quota_data *qd)
505 {
506 spin_lock(&qd->qd_lockref.lock);
507 gfs2_assert_warn(qd->qd_sbd, test_bit(QDF_LOCKED, &qd->qd_flags));
508 clear_bit(QDF_LOCKED, &qd->qd_flags);
509 spin_unlock(&qd->qd_lockref.lock);
510 qdsb_put(qd);
511 }
512
qdsb_get(struct gfs2_sbd * sdp,struct kqid qid,struct gfs2_quota_data ** qdp)513 static int qdsb_get(struct gfs2_sbd *sdp, struct kqid qid,
514 struct gfs2_quota_data **qdp)
515 {
516 int error;
517
518 error = qd_get(sdp, qid, qdp);
519 if (error)
520 return error;
521
522 error = slot_get(*qdp);
523 if (error)
524 goto fail;
525
526 error = bh_get(*qdp);
527 if (error)
528 goto fail_slot;
529
530 return 0;
531
532 fail_slot:
533 slot_put(*qdp);
534 fail:
535 qd_put(*qdp);
536 return error;
537 }
538
539 /**
540 * gfs2_qa_get - make sure we have a quota allocations data structure,
541 * if necessary
542 * @ip: the inode for this reservation
543 */
gfs2_qa_get(struct gfs2_inode * ip)544 int gfs2_qa_get(struct gfs2_inode *ip)
545 {
546 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
547 struct inode *inode = &ip->i_inode;
548
549 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
550 return 0;
551
552 spin_lock(&inode->i_lock);
553 if (ip->i_qadata == NULL) {
554 struct gfs2_qadata *tmp;
555
556 spin_unlock(&inode->i_lock);
557 tmp = kmem_cache_zalloc(gfs2_qadata_cachep, GFP_NOFS);
558 if (!tmp)
559 return -ENOMEM;
560
561 spin_lock(&inode->i_lock);
562 if (ip->i_qadata == NULL)
563 ip->i_qadata = tmp;
564 else
565 kmem_cache_free(gfs2_qadata_cachep, tmp);
566 }
567 ip->i_qadata->qa_ref++;
568 spin_unlock(&inode->i_lock);
569 return 0;
570 }
571
gfs2_qa_put(struct gfs2_inode * ip)572 void gfs2_qa_put(struct gfs2_inode *ip)
573 {
574 struct inode *inode = &ip->i_inode;
575
576 spin_lock(&inode->i_lock);
577 if (ip->i_qadata && --ip->i_qadata->qa_ref == 0) {
578 kmem_cache_free(gfs2_qadata_cachep, ip->i_qadata);
579 ip->i_qadata = NULL;
580 }
581 spin_unlock(&inode->i_lock);
582 }
583
gfs2_quota_hold(struct gfs2_inode * ip,kuid_t uid,kgid_t gid)584 int gfs2_quota_hold(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
585 {
586 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
587 struct gfs2_quota_data **qd;
588 int error;
589
590 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
591 return 0;
592
593 error = gfs2_qa_get(ip);
594 if (error)
595 return error;
596
597 qd = ip->i_qadata->qa_qd;
598
599 if (gfs2_assert_warn(sdp, !ip->i_qadata->qa_qd_num) ||
600 gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags))) {
601 error = -EIO;
602 gfs2_qa_put(ip);
603 goto out;
604 }
605
606 error = qdsb_get(sdp, make_kqid_uid(ip->i_inode.i_uid), qd);
607 if (error)
608 goto out_unhold;
609 ip->i_qadata->qa_qd_num++;
610 qd++;
611
612 error = qdsb_get(sdp, make_kqid_gid(ip->i_inode.i_gid), qd);
613 if (error)
614 goto out_unhold;
615 ip->i_qadata->qa_qd_num++;
616 qd++;
617
618 if (!uid_eq(uid, NO_UID_QUOTA_CHANGE) &&
619 !uid_eq(uid, ip->i_inode.i_uid)) {
620 error = qdsb_get(sdp, make_kqid_uid(uid), qd);
621 if (error)
622 goto out_unhold;
623 ip->i_qadata->qa_qd_num++;
624 qd++;
625 }
626
627 if (!gid_eq(gid, NO_GID_QUOTA_CHANGE) &&
628 !gid_eq(gid, ip->i_inode.i_gid)) {
629 error = qdsb_get(sdp, make_kqid_gid(gid), qd);
630 if (error)
631 goto out_unhold;
632 ip->i_qadata->qa_qd_num++;
633 qd++;
634 }
635
636 out_unhold:
637 if (error)
638 gfs2_quota_unhold(ip);
639 out:
640 return error;
641 }
642
gfs2_quota_unhold(struct gfs2_inode * ip)643 void gfs2_quota_unhold(struct gfs2_inode *ip)
644 {
645 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
646 u32 x;
647
648 if (ip->i_qadata == NULL)
649 return;
650
651 gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags));
652
653 for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
654 qdsb_put(ip->i_qadata->qa_qd[x]);
655 ip->i_qadata->qa_qd[x] = NULL;
656 }
657 ip->i_qadata->qa_qd_num = 0;
658 gfs2_qa_put(ip);
659 }
660
sort_qd(const void * a,const void * b)661 static int sort_qd(const void *a, const void *b)
662 {
663 const struct gfs2_quota_data *qd_a = *(const struct gfs2_quota_data **)a;
664 const struct gfs2_quota_data *qd_b = *(const struct gfs2_quota_data **)b;
665
666 if (qid_lt(qd_a->qd_id, qd_b->qd_id))
667 return -1;
668 if (qid_lt(qd_b->qd_id, qd_a->qd_id))
669 return 1;
670 return 0;
671 }
672
do_qc(struct gfs2_quota_data * qd,s64 change)673 static void do_qc(struct gfs2_quota_data *qd, s64 change)
674 {
675 struct gfs2_sbd *sdp = qd->qd_sbd;
676 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
677 struct gfs2_quota_change *qc = qd->qd_bh_qc;
678 bool needs_put = false;
679 s64 x;
680
681 gfs2_trans_add_meta(ip->i_gl, qd->qd_bh);
682
683 /*
684 * The QDF_CHANGE flag indicates that the slot in the quota change file
685 * is used. Here, we use the value of qc->qc_change when the slot is
686 * used, and we assume a value of 0 otherwise.
687 */
688
689 spin_lock(&qd->qd_lockref.lock);
690
691 x = 0;
692 if (test_bit(QDF_CHANGE, &qd->qd_flags))
693 x = be64_to_cpu(qc->qc_change);
694 x += change;
695 qd->qd_change += change;
696
697 if (!x && test_bit(QDF_CHANGE, &qd->qd_flags)) {
698 /* The slot in the quota change file becomes unused. */
699 clear_bit(QDF_CHANGE, &qd->qd_flags);
700 qc->qc_flags = 0;
701 qc->qc_id = 0;
702 needs_put = true;
703 } else if (x && !test_bit(QDF_CHANGE, &qd->qd_flags)) {
704 /* The slot in the quota change file becomes used. */
705 set_bit(QDF_CHANGE, &qd->qd_flags);
706 __qd_hold(qd);
707 slot_hold(qd);
708
709 qc->qc_flags = 0;
710 if (qd->qd_id.type == USRQUOTA)
711 qc->qc_flags = cpu_to_be32(GFS2_QCF_USER);
712 qc->qc_id = cpu_to_be32(from_kqid(&init_user_ns, qd->qd_id));
713 }
714 qc->qc_change = cpu_to_be64(x);
715
716 spin_unlock(&qd->qd_lockref.lock);
717
718 if (needs_put) {
719 slot_put(qd);
720 qd_put(qd);
721 }
722 if (change < 0) /* Reset quiet flag if we freed some blocks */
723 clear_bit(QDF_QMSG_QUIET, &qd->qd_flags);
724 }
725
gfs2_write_buf_to_page(struct gfs2_sbd * sdp,unsigned long index,unsigned off,void * buf,unsigned bytes)726 static int gfs2_write_buf_to_page(struct gfs2_sbd *sdp, unsigned long index,
727 unsigned off, void *buf, unsigned bytes)
728 {
729 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
730 struct inode *inode = &ip->i_inode;
731 struct address_space *mapping = inode->i_mapping;
732 struct folio *folio;
733 struct buffer_head *bh;
734 u64 blk;
735 unsigned bsize = sdp->sd_sb.sb_bsize, bnum = 0, boff = 0;
736 unsigned to_write = bytes, pg_off = off;
737
738 blk = index << (PAGE_SHIFT - sdp->sd_sb.sb_bsize_shift);
739 boff = off % bsize;
740
741 folio = filemap_grab_folio(mapping, index);
742 if (IS_ERR(folio))
743 return PTR_ERR(folio);
744 bh = folio_buffers(folio);
745 if (!bh)
746 bh = create_empty_buffers(folio, bsize, 0);
747
748 for (;;) {
749 /* Find the beginning block within the folio */
750 if (pg_off >= ((bnum * bsize) + bsize)) {
751 bh = bh->b_this_page;
752 bnum++;
753 blk++;
754 continue;
755 }
756 if (!buffer_mapped(bh)) {
757 gfs2_block_map(inode, blk, bh, 1);
758 if (!buffer_mapped(bh))
759 goto unlock_out;
760 /* If it's a newly allocated disk block, zero it */
761 if (buffer_new(bh))
762 folio_zero_range(folio, bnum * bsize,
763 bh->b_size);
764 }
765 if (folio_test_uptodate(folio))
766 set_buffer_uptodate(bh);
767 if (bh_read(bh, REQ_META | REQ_PRIO) < 0)
768 goto unlock_out;
769 gfs2_trans_add_data(ip->i_gl, bh);
770
771 /* If we need to write to the next block as well */
772 if (to_write > (bsize - boff)) {
773 pg_off += (bsize - boff);
774 to_write -= (bsize - boff);
775 boff = pg_off % bsize;
776 continue;
777 }
778 break;
779 }
780
781 /* Write to the folio, now that we have setup the buffer(s) */
782 memcpy_to_folio(folio, off, buf, bytes);
783 flush_dcache_folio(folio);
784 folio_unlock(folio);
785 folio_put(folio);
786
787 return 0;
788
789 unlock_out:
790 folio_unlock(folio);
791 folio_put(folio);
792 return -EIO;
793 }
794
gfs2_write_disk_quota(struct gfs2_sbd * sdp,struct gfs2_quota * qp,loff_t loc)795 static int gfs2_write_disk_quota(struct gfs2_sbd *sdp, struct gfs2_quota *qp,
796 loff_t loc)
797 {
798 unsigned long pg_beg;
799 unsigned pg_off, nbytes, overflow = 0;
800 int error;
801 void *ptr;
802
803 nbytes = sizeof(struct gfs2_quota);
804
805 pg_beg = loc >> PAGE_SHIFT;
806 pg_off = offset_in_page(loc);
807
808 /* If the quota straddles a page boundary, split the write in two */
809 if ((pg_off + nbytes) > PAGE_SIZE)
810 overflow = (pg_off + nbytes) - PAGE_SIZE;
811
812 ptr = qp;
813 error = gfs2_write_buf_to_page(sdp, pg_beg, pg_off, ptr,
814 nbytes - overflow);
815 /* If there's an overflow, write the remaining bytes to the next page */
816 if (!error && overflow)
817 error = gfs2_write_buf_to_page(sdp, pg_beg + 1, 0,
818 ptr + nbytes - overflow,
819 overflow);
820 return error;
821 }
822
823 /**
824 * gfs2_adjust_quota - adjust record of current block usage
825 * @sdp: The superblock
826 * @loc: Offset of the entry in the quota file
827 * @change: The amount of usage change to record
828 * @qd: The quota data
829 * @fdq: The updated limits to record
830 *
831 * This function was mostly borrowed from gfs2_block_truncate_page which was
832 * in turn mostly borrowed from ext3
833 *
834 * Returns: 0 or -ve on error
835 */
836
gfs2_adjust_quota(struct gfs2_sbd * sdp,loff_t loc,s64 change,struct gfs2_quota_data * qd,struct qc_dqblk * fdq)837 static int gfs2_adjust_quota(struct gfs2_sbd *sdp, loff_t loc,
838 s64 change, struct gfs2_quota_data *qd,
839 struct qc_dqblk *fdq)
840 {
841 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
842 struct inode *inode = &ip->i_inode;
843 struct gfs2_quota q;
844 int err;
845 u64 size;
846
847 if (gfs2_is_stuffed(ip)) {
848 err = gfs2_unstuff_dinode(ip);
849 if (err)
850 return err;
851 }
852
853 memset(&q, 0, sizeof(struct gfs2_quota));
854 err = gfs2_internal_read(ip, (char *)&q, &loc, sizeof(q));
855 if (err < 0)
856 return err;
857
858 loc -= sizeof(q); /* gfs2_internal_read would've advanced the loc ptr */
859 be64_add_cpu(&q.qu_value, change);
860 if (((s64)be64_to_cpu(q.qu_value)) < 0)
861 q.qu_value = 0; /* Never go negative on quota usage */
862 spin_lock(&qd->qd_lockref.lock);
863 qd->qd_qb.qb_value = q.qu_value;
864 if (fdq) {
865 if (fdq->d_fieldmask & QC_SPC_SOFT) {
866 q.qu_warn = cpu_to_be64(fdq->d_spc_softlimit >> sdp->sd_sb.sb_bsize_shift);
867 qd->qd_qb.qb_warn = q.qu_warn;
868 }
869 if (fdq->d_fieldmask & QC_SPC_HARD) {
870 q.qu_limit = cpu_to_be64(fdq->d_spc_hardlimit >> sdp->sd_sb.sb_bsize_shift);
871 qd->qd_qb.qb_limit = q.qu_limit;
872 }
873 if (fdq->d_fieldmask & QC_SPACE) {
874 q.qu_value = cpu_to_be64(fdq->d_space >> sdp->sd_sb.sb_bsize_shift);
875 qd->qd_qb.qb_value = q.qu_value;
876 }
877 }
878 spin_unlock(&qd->qd_lockref.lock);
879
880 err = gfs2_write_disk_quota(sdp, &q, loc);
881 if (!err) {
882 size = loc + sizeof(struct gfs2_quota);
883 if (size > inode->i_size)
884 i_size_write(inode, size);
885 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
886 mark_inode_dirty(inode);
887 set_bit(QDF_REFRESH, &qd->qd_flags);
888 }
889
890 return err;
891 }
892
do_sync(unsigned int num_qd,struct gfs2_quota_data ** qda,u64 sync_gen)893 static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda,
894 u64 sync_gen)
895 {
896 struct gfs2_sbd *sdp = (*qda)->qd_sbd;
897 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
898 struct gfs2_alloc_parms ap = {};
899 unsigned int data_blocks, ind_blocks;
900 struct gfs2_holder *ghs, i_gh;
901 unsigned int qx, x;
902 struct gfs2_quota_data *qd;
903 unsigned reserved;
904 loff_t offset;
905 unsigned int nalloc = 0, blocks;
906 int error;
907
908 gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
909 &data_blocks, &ind_blocks);
910
911 ghs = kmalloc_objs(struct gfs2_holder, num_qd, GFP_NOFS);
912 if (!ghs)
913 return -ENOMEM;
914
915 sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL);
916 inode_lock(&ip->i_inode);
917 for (qx = 0; qx < num_qd; qx++) {
918 error = gfs2_glock_nq_init(qda[qx]->qd_gl, LM_ST_EXCLUSIVE,
919 GL_NOCACHE, &ghs[qx]);
920 if (error)
921 goto out_dq;
922 }
923
924 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
925 if (error)
926 goto out_dq;
927
928 for (x = 0; x < num_qd; x++) {
929 offset = qd2offset(qda[x]);
930 if (gfs2_write_alloc_required(ip, offset,
931 sizeof(struct gfs2_quota)))
932 nalloc++;
933 }
934
935 /*
936 * 1 blk for unstuffing inode if stuffed. We add this extra
937 * block to the reservation unconditionally. If the inode
938 * doesn't need unstuffing, the block will be released to the
939 * rgrp since it won't be allocated during the transaction
940 */
941 /* +3 in the end for unstuffing block, inode size update block
942 * and another block in case quota straddles page boundary and
943 * two blocks need to be updated instead of 1 */
944 blocks = num_qd * data_blocks + RES_DINODE + num_qd + 3;
945
946 reserved = 1 + (nalloc * (data_blocks + ind_blocks));
947 ap.target = reserved;
948 error = gfs2_inplace_reserve(ip, &ap);
949 if (error)
950 goto out_alloc;
951
952 if (nalloc)
953 blocks += gfs2_rg_blocks(ip, reserved) + nalloc * ind_blocks + RES_STATFS;
954
955 error = gfs2_trans_begin(sdp, blocks, 0);
956 if (error)
957 goto out_ipres;
958
959 for (x = 0; x < num_qd; x++) {
960 qd = qda[x];
961 offset = qd2offset(qd);
962 error = gfs2_adjust_quota(sdp, offset, qd->qd_change_sync, qd,
963 NULL);
964 if (error)
965 goto out_end_trans;
966
967 do_qc(qd, -qd->qd_change_sync);
968 set_bit(QDF_REFRESH, &qd->qd_flags);
969 }
970
971 out_end_trans:
972 gfs2_trans_end(sdp);
973 out_ipres:
974 gfs2_inplace_release(ip);
975 out_alloc:
976 gfs2_glock_dq_uninit(&i_gh);
977 out_dq:
978 while (qx--)
979 gfs2_glock_dq_uninit(&ghs[qx]);
980 inode_unlock(&ip->i_inode);
981 kfree(ghs);
982 gfs2_log_flush(glock_sbd(ip->i_gl), ip->i_gl,
983 GFS2_LOG_HEAD_FLUSH_NORMAL | GFS2_LFC_DO_SYNC);
984 if (!error) {
985 for (x = 0; x < num_qd; x++) {
986 qd = qda[x];
987 spin_lock(&qd->qd_lockref.lock);
988 if (qd->qd_sync_gen < sync_gen)
989 qd->qd_sync_gen = sync_gen;
990 spin_unlock(&qd->qd_lockref.lock);
991 }
992 }
993 return error;
994 }
995
update_qd(struct gfs2_sbd * sdp,struct gfs2_quota_data * qd)996 static int update_qd(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd)
997 {
998 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
999 struct gfs2_quota q;
1000 struct gfs2_quota_lvb *qlvb;
1001 loff_t pos;
1002 int error;
1003
1004 memset(&q, 0, sizeof(struct gfs2_quota));
1005 pos = qd2offset(qd);
1006 error = gfs2_internal_read(ip, (char *)&q, &pos, sizeof(q));
1007 if (error < 0)
1008 return error;
1009
1010 qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1011 qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC);
1012 qlvb->__pad = 0;
1013 qlvb->qb_limit = q.qu_limit;
1014 qlvb->qb_warn = q.qu_warn;
1015 qlvb->qb_value = q.qu_value;
1016 spin_lock(&qd->qd_lockref.lock);
1017 qd->qd_qb = *qlvb;
1018 spin_unlock(&qd->qd_lockref.lock);
1019
1020 return 0;
1021 }
1022
do_glock(struct gfs2_quota_data * qd,int force_refresh,struct gfs2_holder * q_gh)1023 static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
1024 struct gfs2_holder *q_gh)
1025 {
1026 struct gfs2_sbd *sdp = qd->qd_sbd;
1027 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1028 struct gfs2_holder i_gh;
1029 int error;
1030
1031 gfs2_assert_warn(sdp, sdp == glock_sbd(qd->qd_gl));
1032 restart:
1033 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
1034 if (error)
1035 return error;
1036
1037 if (test_and_clear_bit(QDF_REFRESH, &qd->qd_flags))
1038 force_refresh = FORCE;
1039
1040 spin_lock(&qd->qd_lockref.lock);
1041 qd->qd_qb = *(struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1042 spin_unlock(&qd->qd_lockref.lock);
1043
1044 if (force_refresh || qd->qd_qb.qb_magic != cpu_to_be32(GFS2_MAGIC)) {
1045 gfs2_glock_dq_uninit(q_gh);
1046 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE,
1047 GL_NOCACHE, q_gh);
1048 if (error)
1049 return error;
1050
1051 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
1052 if (error)
1053 goto fail;
1054
1055 error = update_qd(sdp, qd);
1056 if (error)
1057 goto fail_gunlock;
1058
1059 gfs2_glock_dq_uninit(&i_gh);
1060 gfs2_glock_dq_uninit(q_gh);
1061 force_refresh = 0;
1062 goto restart;
1063 }
1064
1065 return 0;
1066
1067 fail_gunlock:
1068 gfs2_glock_dq_uninit(&i_gh);
1069 fail:
1070 gfs2_glock_dq_uninit(q_gh);
1071 return error;
1072 }
1073
gfs2_quota_lock(struct gfs2_inode * ip,kuid_t uid,kgid_t gid)1074 int gfs2_quota_lock(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
1075 {
1076 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1077 struct gfs2_quota_data *qd;
1078 u32 x;
1079 int error;
1080
1081 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1082 return 0;
1083
1084 error = gfs2_quota_hold(ip, uid, gid);
1085 if (error)
1086 return error;
1087
1088 sort(ip->i_qadata->qa_qd, ip->i_qadata->qa_qd_num,
1089 sizeof(struct gfs2_quota_data *), sort_qd, NULL);
1090
1091 for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1092 qd = ip->i_qadata->qa_qd[x];
1093 error = do_glock(qd, NO_FORCE, &ip->i_qadata->qa_qd_ghs[x]);
1094 if (error)
1095 break;
1096 }
1097
1098 if (!error)
1099 set_bit(GIF_QD_LOCKED, &ip->i_flags);
1100 else {
1101 while (x--)
1102 gfs2_glock_dq_uninit(&ip->i_qadata->qa_qd_ghs[x]);
1103 gfs2_quota_unhold(ip);
1104 }
1105
1106 return error;
1107 }
1108
need_sync(struct gfs2_quota_data * qd)1109 static bool need_sync(struct gfs2_quota_data *qd)
1110 {
1111 struct gfs2_sbd *sdp = qd->qd_sbd;
1112 struct gfs2_tune *gt = &sdp->sd_tune;
1113 s64 value, change, limit;
1114 unsigned int num, den;
1115 int ret = false;
1116
1117 spin_lock(&qd->qd_lockref.lock);
1118 if (!qd->qd_qb.qb_limit)
1119 goto out;
1120
1121 change = qd->qd_change;
1122 if (change <= 0)
1123 goto out;
1124 value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
1125 limit = (s64)be64_to_cpu(qd->qd_qb.qb_limit);
1126 if (value >= limit)
1127 goto out;
1128
1129 spin_lock(>->gt_spin);
1130 num = gt->gt_quota_scale_num;
1131 den = gt->gt_quota_scale_den;
1132 spin_unlock(>->gt_spin);
1133
1134 change *= gfs2_jindex_size(sdp) * num;
1135 change = div_s64(change, den);
1136 if (value + change < limit)
1137 goto out;
1138
1139 ret = true;
1140 out:
1141 spin_unlock(&qd->qd_lockref.lock);
1142 return ret;
1143 }
1144
gfs2_quota_unlock(struct gfs2_inode * ip)1145 void gfs2_quota_unlock(struct gfs2_inode *ip)
1146 {
1147 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1148 struct gfs2_quota_data *qda[2 * GFS2_MAXQUOTAS];
1149 unsigned int count = 0;
1150 u32 x;
1151
1152 if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
1153 return;
1154
1155 for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1156 struct gfs2_quota_data *qd;
1157 bool sync;
1158 int error;
1159
1160 qd = ip->i_qadata->qa_qd[x];
1161 sync = need_sync(qd);
1162
1163 gfs2_glock_dq_uninit(&ip->i_qadata->qa_qd_ghs[x]);
1164 if (!sync)
1165 continue;
1166
1167 spin_lock(&qd_lock);
1168 sync = qd_grab_sync(sdp, qd, U64_MAX);
1169 spin_unlock(&qd_lock);
1170
1171 if (!sync)
1172 continue;
1173
1174 gfs2_assert_warn(sdp, qd->qd_change_sync);
1175 error = bh_get(qd);
1176 if (error) {
1177 qd_ungrab_sync(qd);
1178 continue;
1179 }
1180
1181 qda[count++] = qd;
1182 }
1183
1184 if (count) {
1185 u64 sync_gen = READ_ONCE(sdp->sd_quota_sync_gen);
1186
1187 do_sync(count, qda, sync_gen);
1188 for (x = 0; x < count; x++)
1189 qd_unlock(qda[x]);
1190 }
1191
1192 gfs2_quota_unhold(ip);
1193 }
1194
1195 #define MAX_LINE 256
1196
print_message(struct gfs2_quota_data * qd,char * type)1197 static void print_message(struct gfs2_quota_data *qd, char *type)
1198 {
1199 struct gfs2_sbd *sdp = qd->qd_sbd;
1200
1201 if (sdp->sd_args.ar_quota != GFS2_QUOTA_QUIET) {
1202 fs_info(sdp, "quota %s for %s %u\n",
1203 type,
1204 (qd->qd_id.type == USRQUOTA) ? "user" : "group",
1205 from_kqid(&init_user_ns, qd->qd_id));
1206 }
1207 }
1208
1209 /**
1210 * gfs2_quota_check - check if allocating new blocks will exceed quota
1211 * @ip: The inode for which this check is being performed
1212 * @uid: The uid to check against
1213 * @gid: The gid to check against
1214 * @ap: The allocation parameters. ap->target contains the requested
1215 * blocks. ap->min_target, if set, contains the minimum blks
1216 * requested.
1217 *
1218 * Returns: 0 on success.
1219 * min_req = ap->min_target ? ap->min_target : ap->target;
1220 * quota must allow at least min_req blks for success and
1221 * ap->allowed is set to the number of blocks allowed
1222 *
1223 * -EDQUOT otherwise, quota violation. ap->allowed is set to number
1224 * of blocks available.
1225 */
gfs2_quota_check(struct gfs2_inode * ip,kuid_t uid,kgid_t gid,struct gfs2_alloc_parms * ap)1226 int gfs2_quota_check(struct gfs2_inode *ip, kuid_t uid, kgid_t gid,
1227 struct gfs2_alloc_parms *ap)
1228 {
1229 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1230 struct gfs2_quota_data *qd;
1231 s64 value, warn, limit;
1232 u32 x;
1233 int error = 0;
1234
1235 ap->allowed = UINT_MAX; /* Assume we are permitted a whole lot */
1236 if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
1237 return 0;
1238
1239 for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1240 qd = ip->i_qadata->qa_qd[x];
1241
1242 if (!(qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1243 qid_eq(qd->qd_id, make_kqid_gid(gid))))
1244 continue;
1245
1246 spin_lock(&qd->qd_lockref.lock);
1247 warn = (s64)be64_to_cpu(qd->qd_qb.qb_warn);
1248 limit = (s64)be64_to_cpu(qd->qd_qb.qb_limit);
1249 value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
1250 value += qd->qd_change;
1251 spin_unlock(&qd->qd_lockref.lock);
1252
1253 if (limit > 0 && (limit - value) < ap->allowed)
1254 ap->allowed = limit - value;
1255 /* If we can't meet the target */
1256 if (limit && limit < (value + (s64)ap->target)) {
1257 /* If no min_target specified or we don't meet
1258 * min_target, return -EDQUOT */
1259 if (!ap->min_target || ap->min_target > ap->allowed) {
1260 if (!test_and_set_bit(QDF_QMSG_QUIET,
1261 &qd->qd_flags)) {
1262 print_message(qd, "exceeded");
1263 quota_send_warning(qd->qd_id,
1264 sdp->sd_vfs->s_dev,
1265 QUOTA_NL_BHARDWARN);
1266 }
1267 error = -EDQUOT;
1268 break;
1269 }
1270 } else if (warn && warn < value &&
1271 time_after_eq(jiffies, qd->qd_last_warn +
1272 gfs2_tune_get(sdp, gt_quota_warn_period)
1273 * HZ)) {
1274 quota_send_warning(qd->qd_id,
1275 sdp->sd_vfs->s_dev, QUOTA_NL_BSOFTWARN);
1276 print_message(qd, "warning");
1277 error = 0;
1278 qd->qd_last_warn = jiffies;
1279 }
1280 }
1281 return error;
1282 }
1283
gfs2_quota_change(struct gfs2_inode * ip,s64 change,kuid_t uid,kgid_t gid)1284 void gfs2_quota_change(struct gfs2_inode *ip, s64 change,
1285 kuid_t uid, kgid_t gid)
1286 {
1287 struct gfs2_quota_data *qd;
1288 u32 x;
1289 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1290
1291 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF ||
1292 gfs2_assert_warn(sdp, change))
1293 return;
1294 if (ip->i_diskflags & GFS2_DIF_SYSTEM)
1295 return;
1296
1297 if (gfs2_assert_withdraw(sdp, ip->i_qadata &&
1298 ip->i_qadata->qa_ref > 0))
1299 return;
1300 for (x = 0; x < ip->i_qadata->qa_qd_num; x++) {
1301 qd = ip->i_qadata->qa_qd[x];
1302
1303 if (qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1304 qid_eq(qd->qd_id, make_kqid_gid(gid))) {
1305 do_qc(qd, change);
1306 }
1307 }
1308 }
1309
gfs2_quota_sync(struct super_block * sb,int type)1310 int gfs2_quota_sync(struct super_block *sb, int type)
1311 {
1312 struct gfs2_sbd *sdp = sb->s_fs_info;
1313 struct gfs2_quota_data **qda;
1314 unsigned int max_qd = PAGE_SIZE / sizeof(struct gfs2_holder);
1315 u64 sync_gen;
1316 int error = 0;
1317
1318 if (sb_rdonly(sdp->sd_vfs))
1319 return 0;
1320
1321 qda = kzalloc_objs(struct gfs2_quota_data *, max_qd);
1322 if (!qda)
1323 return -ENOMEM;
1324
1325 mutex_lock(&sdp->sd_quota_sync_mutex);
1326 sync_gen = sdp->sd_quota_sync_gen + 1;
1327
1328 do {
1329 struct gfs2_quota_data *iter;
1330 unsigned int num_qd = 0;
1331 unsigned int x;
1332
1333 spin_lock(&qd_lock);
1334 list_for_each_entry(iter, &sdp->sd_quota_list, qd_list) {
1335 if (qd_grab_sync(sdp, iter, sync_gen)) {
1336 qda[num_qd++] = iter;
1337 if (num_qd == max_qd)
1338 break;
1339 }
1340 }
1341 spin_unlock(&qd_lock);
1342
1343 if (!num_qd)
1344 break;
1345
1346 for (x = 0; x < num_qd; x++) {
1347 error = bh_get(qda[x]);
1348 if (!error)
1349 continue;
1350
1351 while (x < num_qd)
1352 qd_ungrab_sync(qda[--num_qd]);
1353 break;
1354 }
1355
1356 if (!error) {
1357 WRITE_ONCE(sdp->sd_quota_sync_gen, sync_gen);
1358 error = do_sync(num_qd, qda, sync_gen);
1359 }
1360
1361 for (x = 0; x < num_qd; x++)
1362 qd_unlock(qda[x]);
1363 } while (!error);
1364
1365 mutex_unlock(&sdp->sd_quota_sync_mutex);
1366 kfree(qda);
1367
1368 return error;
1369 }
1370
gfs2_quota_refresh(struct gfs2_sbd * sdp,struct kqid qid)1371 int gfs2_quota_refresh(struct gfs2_sbd *sdp, struct kqid qid)
1372 {
1373 struct gfs2_quota_data *qd;
1374 struct gfs2_holder q_gh;
1375 int error;
1376
1377 error = qd_get(sdp, qid, &qd);
1378 if (error)
1379 return error;
1380
1381 error = do_glock(qd, FORCE, &q_gh);
1382 if (!error)
1383 gfs2_glock_dq_uninit(&q_gh);
1384
1385 qd_put(qd);
1386 return error;
1387 }
1388
gfs2_quota_init(struct gfs2_sbd * sdp)1389 int gfs2_quota_init(struct gfs2_sbd *sdp)
1390 {
1391 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
1392 u64 size = i_size_read(sdp->sd_qc_inode);
1393 unsigned int blocks = size >> sdp->sd_sb.sb_bsize_shift;
1394 unsigned int x, slot = 0;
1395 unsigned int found = 0;
1396 unsigned int hash;
1397 unsigned int bm_size;
1398 struct buffer_head *bh;
1399 u64 dblock;
1400 u32 extlen = 0;
1401 int error;
1402
1403 if (gfs2_check_internal_file_size(sdp->sd_qc_inode, 1, 64 << 20))
1404 return -EIO;
1405
1406 sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
1407 bm_size = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * sizeof(unsigned long));
1408 bm_size *= sizeof(unsigned long);
1409 error = -ENOMEM;
1410 sdp->sd_quota_bitmap = kzalloc(bm_size, GFP_NOFS | __GFP_NOWARN);
1411 if (sdp->sd_quota_bitmap == NULL)
1412 sdp->sd_quota_bitmap = __vmalloc(bm_size, GFP_NOFS |
1413 __GFP_ZERO);
1414 if (!sdp->sd_quota_bitmap)
1415 return error;
1416
1417 for (x = 0; x < blocks; x++) {
1418 struct gfs2_quota_change *qc;
1419 unsigned int y;
1420
1421 if (!extlen) {
1422 extlen = 32;
1423 error = gfs2_get_extent(&ip->i_inode, x, &dblock, &extlen);
1424 if (error)
1425 goto fail;
1426 }
1427 error = -EIO;
1428 bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
1429 if (!bh)
1430 goto fail;
1431 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC))
1432 goto fail_brelse;
1433
1434 qc = (struct gfs2_quota_change *)(bh->b_data + sizeof(struct gfs2_meta_header));
1435 for (y = 0; y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1436 y++, slot++) {
1437 struct gfs2_quota_data *old_qd, *qd;
1438 s64 qc_change = be64_to_cpu(qc->qc_change);
1439 u32 qc_flags = be32_to_cpu(qc->qc_flags);
1440 enum quota_type qtype = (qc_flags & GFS2_QCF_USER) ?
1441 USRQUOTA : GRPQUOTA;
1442 struct kqid qc_id = make_kqid(&init_user_ns, qtype,
1443 be32_to_cpu(qc->qc_id));
1444 qc++;
1445 if (!qc_change)
1446 continue;
1447
1448 hash = gfs2_qd_hash(sdp, qc_id);
1449 qd = qd_alloc(hash, sdp, qc_id);
1450 if (qd == NULL)
1451 goto fail_brelse;
1452
1453 qd->qd_lockref.count = 0;
1454 set_bit(QDF_CHANGE, &qd->qd_flags);
1455 qd->qd_change = qc_change;
1456 qd->qd_slot = slot;
1457 qd->qd_slot_ref = 1;
1458
1459 spin_lock(&qd_lock);
1460 spin_lock_bucket(hash);
1461 old_qd = gfs2_qd_search_bucket(hash, sdp, qc_id);
1462 if (old_qd) {
1463 fs_err(sdp, "Corruption found in quota_change%u"
1464 "file: duplicate identifier in "
1465 "slot %u\n",
1466 sdp->sd_jdesc->jd_jid, slot);
1467
1468 spin_unlock_bucket(hash);
1469 spin_unlock(&qd_lock);
1470 qd_put(old_qd);
1471
1472 gfs2_glock_put(qd->qd_gl);
1473 kmem_cache_free(gfs2_quotad_cachep, qd);
1474
1475 /* zero out the duplicate slot */
1476 lock_buffer(bh);
1477 memset(qc, 0, sizeof(*qc));
1478 mark_buffer_dirty(bh);
1479 unlock_buffer(bh);
1480
1481 continue;
1482 }
1483 BUG_ON(test_and_set_bit(slot, sdp->sd_quota_bitmap));
1484 list_add(&qd->qd_list, &sdp->sd_quota_list);
1485 atomic_inc(&sdp->sd_quota_count);
1486 hlist_bl_add_head_rcu(&qd->qd_hlist, &qd_hash_table[hash]);
1487 spin_unlock_bucket(hash);
1488 spin_unlock(&qd_lock);
1489
1490 found++;
1491 }
1492
1493 if (buffer_dirty(bh))
1494 sync_dirty_buffer(bh);
1495 brelse(bh);
1496 dblock++;
1497 extlen--;
1498 }
1499
1500 if (found)
1501 fs_info(sdp, "found %u quota changes\n", found);
1502
1503 return 0;
1504
1505 fail_brelse:
1506 if (buffer_dirty(bh))
1507 sync_dirty_buffer(bh);
1508 brelse(bh);
1509 fail:
1510 gfs2_quota_cleanup(sdp);
1511 return error;
1512 }
1513
gfs2_quota_cleanup(struct gfs2_sbd * sdp)1514 void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1515 {
1516 struct gfs2_quota_data *qd;
1517 LIST_HEAD(dispose);
1518 int count;
1519
1520 BUG_ON(!test_bit(SDF_NORECOVERY, &sdp->sd_flags) &&
1521 test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags));
1522
1523 spin_lock(&qd_lock);
1524 list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
1525 spin_lock(&qd->qd_lockref.lock);
1526 if (qd->qd_lockref.count != 0) {
1527 spin_unlock(&qd->qd_lockref.lock);
1528 continue;
1529 }
1530 lockref_mark_dead(&qd->qd_lockref);
1531 spin_unlock(&qd->qd_lockref.lock);
1532
1533 list_lru_del_obj(&gfs2_qd_lru, &qd->qd_lru);
1534 list_add(&qd->qd_lru, &dispose);
1535 }
1536 spin_unlock(&qd_lock);
1537
1538 gfs2_qd_list_dispose(&dispose);
1539
1540 wait_event_timeout(sdp->sd_kill_wait,
1541 (count = atomic_read(&sdp->sd_quota_count)) == 0,
1542 HZ * 60);
1543
1544 if (count != 0)
1545 fs_err(sdp, "%d left-over quota data objects\n", count);
1546
1547 kvfree(sdp->sd_quota_bitmap);
1548 sdp->sd_quota_bitmap = NULL;
1549 }
1550
quotad_error(struct gfs2_sbd * sdp,const char * msg,int error)1551 static void quotad_error(struct gfs2_sbd *sdp, const char *msg, int error)
1552 {
1553 if (error == 0 || error == -EROFS)
1554 return;
1555 if (!gfs2_withdrawn(sdp)) {
1556 if (!cmpxchg(&sdp->sd_log_error, 0, error))
1557 fs_err(sdp, "gfs2_quotad: %s error %d\n", msg, error);
1558 wake_up(&sdp->sd_logd_waitq);
1559 }
1560 }
1561
gfs2_wake_up_statfs(struct gfs2_sbd * sdp)1562 void gfs2_wake_up_statfs(struct gfs2_sbd *sdp) {
1563 if (!sdp->sd_statfs_force_sync) {
1564 sdp->sd_statfs_force_sync = 1;
1565 wake_up(&sdp->sd_quota_wait);
1566 }
1567 }
1568
1569
1570 /**
1571 * gfs2_quotad - Write cached quota changes into the quota file
1572 * @data: Pointer to GFS2 superblock
1573 *
1574 */
1575
gfs2_quotad(void * data)1576 int gfs2_quotad(void *data)
1577 {
1578 struct gfs2_sbd *sdp = data;
1579 unsigned long now = jiffies;
1580 unsigned long statfs_deadline = now;
1581 unsigned long quotad_deadline = now;
1582
1583 set_freezable();
1584 while (!kthread_should_stop()) {
1585 unsigned long t;
1586
1587 if (gfs2_withdrawn(sdp))
1588 break;
1589
1590 now = jiffies;
1591 if (sdp->sd_statfs_force_sync ||
1592 time_after(now, statfs_deadline)) {
1593 unsigned int quantum;
1594 int error;
1595
1596 /* Update the master statfs file */
1597 error = gfs2_statfs_sync(sdp->sd_vfs, 0);
1598 quotad_error(sdp, "statfs", error);
1599
1600 quantum = gfs2_tune_get(sdp, gt_statfs_quantum);
1601 statfs_deadline = now + quantum * HZ;
1602 }
1603 if (time_after(now, quotad_deadline)) {
1604 unsigned int quantum;
1605 int error;
1606
1607 /* Update the quota file */
1608 error = gfs2_quota_sync(sdp->sd_vfs, 0);
1609 quotad_error(sdp, "sync", error);
1610
1611 quantum = gfs2_tune_get(sdp, gt_quota_quantum);
1612 quotad_deadline = now + quantum * HZ;
1613 }
1614
1615 t = min(statfs_deadline - now, quotad_deadline - now);
1616 wait_event_freezable_timeout(sdp->sd_quota_wait,
1617 sdp->sd_statfs_force_sync ||
1618 gfs2_withdrawn(sdp) ||
1619 kthread_should_stop(),
1620 t);
1621
1622 if (sdp->sd_statfs_force_sync)
1623 t = 0;
1624 }
1625
1626 return 0;
1627 }
1628
gfs2_quota_get_state(struct super_block * sb,struct qc_state * state)1629 static int gfs2_quota_get_state(struct super_block *sb, struct qc_state *state)
1630 {
1631 struct gfs2_sbd *sdp = sb->s_fs_info;
1632
1633 memset(state, 0, sizeof(*state));
1634
1635 switch (sdp->sd_args.ar_quota) {
1636 case GFS2_QUOTA_QUIET:
1637 fallthrough;
1638 case GFS2_QUOTA_ON:
1639 state->s_state[USRQUOTA].flags |= QCI_LIMITS_ENFORCED;
1640 state->s_state[GRPQUOTA].flags |= QCI_LIMITS_ENFORCED;
1641 fallthrough;
1642 case GFS2_QUOTA_ACCOUNT:
1643 state->s_state[USRQUOTA].flags |= QCI_ACCT_ENABLED |
1644 QCI_SYSFILE;
1645 state->s_state[GRPQUOTA].flags |= QCI_ACCT_ENABLED |
1646 QCI_SYSFILE;
1647 break;
1648 case GFS2_QUOTA_OFF:
1649 break;
1650 }
1651 if (sdp->sd_quota_inode) {
1652 state->s_state[USRQUOTA].ino =
1653 GFS2_I(sdp->sd_quota_inode)->i_no_addr;
1654 state->s_state[USRQUOTA].blocks = sdp->sd_quota_inode->i_blocks;
1655 }
1656 state->s_state[USRQUOTA].nextents = 1; /* unsupported */
1657 state->s_state[GRPQUOTA] = state->s_state[USRQUOTA];
1658 state->s_incoredqs = list_lru_count(&gfs2_qd_lru);
1659 return 0;
1660 }
1661
gfs2_get_dqblk(struct super_block * sb,struct kqid qid,struct qc_dqblk * fdq)1662 static int gfs2_get_dqblk(struct super_block *sb, struct kqid qid,
1663 struct qc_dqblk *fdq)
1664 {
1665 struct gfs2_sbd *sdp = sb->s_fs_info;
1666 struct gfs2_quota_lvb *qlvb;
1667 struct gfs2_quota_data *qd;
1668 struct gfs2_holder q_gh;
1669 int error;
1670
1671 memset(fdq, 0, sizeof(*fdq));
1672
1673 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1674 return -ESRCH; /* Crazy XFS error code */
1675
1676 if ((qid.type != USRQUOTA) &&
1677 (qid.type != GRPQUOTA))
1678 return -EINVAL;
1679
1680 error = qd_get(sdp, qid, &qd);
1681 if (error)
1682 return error;
1683 error = do_glock(qd, FORCE, &q_gh);
1684 if (error)
1685 goto out;
1686
1687 qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1688 fdq->d_spc_hardlimit = be64_to_cpu(qlvb->qb_limit) << sdp->sd_sb.sb_bsize_shift;
1689 fdq->d_spc_softlimit = be64_to_cpu(qlvb->qb_warn) << sdp->sd_sb.sb_bsize_shift;
1690 fdq->d_space = be64_to_cpu(qlvb->qb_value) << sdp->sd_sb.sb_bsize_shift;
1691
1692 gfs2_glock_dq_uninit(&q_gh);
1693 out:
1694 qd_put(qd);
1695 return error;
1696 }
1697
1698 /* GFS2 only supports a subset of the XFS fields */
1699 #define GFS2_FIELDMASK (QC_SPC_SOFT|QC_SPC_HARD|QC_SPACE)
1700
gfs2_set_dqblk(struct super_block * sb,struct kqid qid,struct qc_dqblk * fdq)1701 static int gfs2_set_dqblk(struct super_block *sb, struct kqid qid,
1702 struct qc_dqblk *fdq)
1703 {
1704 struct gfs2_sbd *sdp = sb->s_fs_info;
1705 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1706 struct gfs2_quota_data *qd;
1707 struct gfs2_holder q_gh, i_gh;
1708 unsigned int data_blocks, ind_blocks;
1709 unsigned int blocks = 0;
1710 int alloc_required;
1711 loff_t offset;
1712 int error;
1713
1714 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1715 return -ESRCH; /* Crazy XFS error code */
1716
1717 if ((qid.type != USRQUOTA) &&
1718 (qid.type != GRPQUOTA))
1719 return -EINVAL;
1720
1721 if (fdq->d_fieldmask & ~GFS2_FIELDMASK)
1722 return -EINVAL;
1723
1724 error = qd_get(sdp, qid, &qd);
1725 if (error)
1726 return error;
1727
1728 error = gfs2_qa_get(ip);
1729 if (error)
1730 goto out_put;
1731
1732 inode_lock(&ip->i_inode);
1733 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE, 0, &q_gh);
1734 if (error)
1735 goto out_unlockput;
1736 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1737 if (error)
1738 goto out_q;
1739
1740 /* Check for existing entry, if none then alloc new blocks */
1741 error = update_qd(sdp, qd);
1742 if (error)
1743 goto out_i;
1744
1745 /* If nothing has changed, this is a no-op */
1746 if ((fdq->d_fieldmask & QC_SPC_SOFT) &&
1747 ((fdq->d_spc_softlimit >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_warn)))
1748 fdq->d_fieldmask ^= QC_SPC_SOFT;
1749
1750 if ((fdq->d_fieldmask & QC_SPC_HARD) &&
1751 ((fdq->d_spc_hardlimit >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_limit)))
1752 fdq->d_fieldmask ^= QC_SPC_HARD;
1753
1754 if ((fdq->d_fieldmask & QC_SPACE) &&
1755 ((fdq->d_space >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_value)))
1756 fdq->d_fieldmask ^= QC_SPACE;
1757
1758 if (fdq->d_fieldmask == 0)
1759 goto out_i;
1760
1761 offset = qd2offset(qd);
1762 alloc_required = gfs2_write_alloc_required(ip, offset, sizeof(struct gfs2_quota));
1763 if (gfs2_is_stuffed(ip))
1764 alloc_required = 1;
1765 if (alloc_required) {
1766 struct gfs2_alloc_parms ap = {};
1767 gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
1768 &data_blocks, &ind_blocks);
1769 blocks = 1 + data_blocks + ind_blocks;
1770 ap.target = blocks;
1771 error = gfs2_inplace_reserve(ip, &ap);
1772 if (error)
1773 goto out_i;
1774 blocks += gfs2_rg_blocks(ip, blocks);
1775 }
1776
1777 /* Some quotas span block boundaries and can update two blocks,
1778 adding an extra block to the transaction to handle such quotas */
1779 error = gfs2_trans_begin(sdp, blocks + RES_DINODE + 2, 0);
1780 if (error)
1781 goto out_release;
1782
1783 /* Apply changes */
1784 error = gfs2_adjust_quota(sdp, offset, 0, qd, fdq);
1785 if (!error)
1786 clear_bit(QDF_QMSG_QUIET, &qd->qd_flags);
1787
1788 gfs2_trans_end(sdp);
1789 out_release:
1790 if (alloc_required)
1791 gfs2_inplace_release(ip);
1792 out_i:
1793 gfs2_glock_dq_uninit(&i_gh);
1794 out_q:
1795 gfs2_glock_dq_uninit(&q_gh);
1796 out_unlockput:
1797 gfs2_qa_put(ip);
1798 inode_unlock(&ip->i_inode);
1799 out_put:
1800 qd_put(qd);
1801 return error;
1802 }
1803
1804 const struct quotactl_ops gfs2_quotactl_ops = {
1805 .quota_sync = gfs2_quota_sync,
1806 .get_state = gfs2_quota_get_state,
1807 .get_dqblk = gfs2_get_dqblk,
1808 .set_dqblk = gfs2_set_dqblk,
1809 };
1810
gfs2_quota_hash_init(void)1811 void __init gfs2_quota_hash_init(void)
1812 {
1813 unsigned i;
1814
1815 for(i = 0; i < GFS2_QD_HASH_SIZE; i++)
1816 INIT_HLIST_BL_HEAD(&qd_hash_table[i]);
1817 }
1818