]> bbs.cooldavid.org Git - net-next-2.6.git/blame - fs/gfs2/quota.c
git://bbs.cooldavid.org / net-next-2.6.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
[GFS2] inode-diet-eliminate-i_blksize-and-use-a-per-superblock-default-vs-gfs2
[net-next-2.6.git] / fs / gfs2 / quota.c
CommitLineData
b3b94faa
DT		 1/*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3a8a9a10 3 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
b3b94faa
DT		 4 *
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
e9fc2aa0 7 * of the GNU General Public License version 2.
b3b94faa
DT		 8 */
9
10/*
11 * Quota change tags are associated with each transaction that allocates or
12 * deallocates space. Those changes are accumulated locally to each node (in a
13 * per-node file) and then are periodically synced to the quota file. This
14 * avoids the bottleneck of constantly touching the quota file, but introduces
15 * fuzziness in the current usage value of IDs that are being used on different
16 * nodes in the cluster simultaneously. So, it is possible for a user on
17 * multiple nodes to overrun their quota, but that overrun is controlable.
18 * Since quota tags are part of transactions, there is no need to a quota check
19 * program to be run on node crashes or anything like that.
20 *
21 * There are couple of knobs that let the administrator manage the quota
22 * fuzziness. "quota_quantum" sets the maximum time a quota change can be
23 * sitting on one node before being synced to the quota file. (The default is
24 * 60 seconds.) Another knob, "quota_scale" controls how quickly the frequency
25 * of quota file syncs increases as the user moves closer to their limit. The
26 * more frequent the syncs, the more accurate the quota enforcement, but that
27 * means that there is more contention between the nodes for the quota file.
28 * The default value is one. This sets the maximum theoretical quota overrun
29 * (with infinite node with infinite bandwidth) to twice the user's limit. (In
30 * practice, the maximum overrun you see should be much less.) A "quota_scale"
31 * number greater than one makes quota syncs more frequent and reduces the
32 * maximum overrun. Numbers less than one (but greater than zero) make quota
33 * syncs less frequent.
34 *
35 * GFS quotas also use per-ID Lock Value Blocks (LVBs) to cache the contents of
36 * the quota file, so it is not being constantly read.
37 */
38
39#include <linux/sched.h>
40#include <linux/slab.h>
41#include <linux/spinlock.h>
42#include <linux/completion.h>
43#include <linux/buffer_head.h>
b3b94faa 44#include <linux/sort.h>
18ec7d5c 45#include <linux/fs.h>
5c676f6d 46#include <linux/gfs2_ondisk.h>
b3b94faa
DT		 47
48#include "gfs2.h"
5c676f6d
SW		 49#include "lm_interface.h"
50#include "incore.h"
b3b94faa
DT		 51#include "bmap.h"
52#include "glock.h"
53#include "glops.h"
b3b94faa
DT		 54#include "log.h"
55#include "meta_io.h"
56#include "quota.h"
57#include "rgrp.h"
58#include "super.h"
59#include "trans.h"
18ec7d5c 60#include "inode.h"
f42faf4f 61#include "ops_file.h"
18ec7d5c 62#include "ops_address.h"
5c676f6d 63#include "util.h"
b3b94faa
DT		 64
65#define QUOTA_USER 1
66#define QUOTA_GROUP 0
67
cd915493 68static u64 qd2offset(struct gfs2_quota_data *qd)
b3b94faa 69{
cd915493 70 u64 offset;
b3b94faa 71
cd915493 72 offset = 2 * (u64)qd->qd_id + !test_bit(QDF_USER, &qd->qd_flags);
b3b94faa
DT		 73 offset *= sizeof(struct gfs2_quota);
74
75 return offset;
76}
77
cd915493 78static int qd_alloc(struct gfs2_sbd *sdp, int user, u32 id,
b3b94faa
DT		 79 struct gfs2_quota_data **qdp)
80{
81 struct gfs2_quota_data *qd;
82 int error;
83
84 qd = kzalloc(sizeof(struct gfs2_quota_data), GFP_KERNEL);
85 if (!qd)
86 return -ENOMEM;
87
88 qd->qd_count = 1;
89 qd->qd_id = id;
90 if (user)
91 set_bit(QDF_USER, &qd->qd_flags);
92 qd->qd_slot = -1;
93
cd915493 94 error = gfs2_glock_get(sdp, 2 * (u64)id + !user,
b3b94faa
DT		 95 &gfs2_quota_glops, CREATE, &qd->qd_gl);
96 if (error)
97 goto fail;
98
99 error = gfs2_lvb_hold(qd->qd_gl);
100 gfs2_glock_put(qd->qd_gl);
101 if (error)
102 goto fail;
103
104 *qdp = qd;
105
106 return 0;
107
a91ea69f 108fail:
b3b94faa
DT		 109 kfree(qd);
110 return error;
111}
112
cd915493 113static int qd_get(struct gfs2_sbd *sdp, int user, u32 id, int create,
b3b94faa
DT		 114 struct gfs2_quota_data **qdp)
115{
116 struct gfs2_quota_data *qd = NULL, *new_qd = NULL;
117 int error, found;
118
119 *qdp = NULL;
120
121 for (;;) {
122 found = 0;
123 spin_lock(&sdp->sd_quota_spin);
124 list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
125 if (qd->qd_id == id &&
126 !test_bit(QDF_USER, &qd->qd_flags) == !user) {
127 qd->qd_count++;
128 found = 1;
129 break;
130 }
131 }
132
133 if (!found)
134 qd = NULL;
135
136 if (!qd && new_qd) {
137 qd = new_qd;
138 list_add(&qd->qd_list, &sdp->sd_quota_list);
139 atomic_inc(&sdp->sd_quota_count);
140 new_qd = NULL;
141 }
142
143 spin_unlock(&sdp->sd_quota_spin);
144
145 if (qd || !create) {
146 if (new_qd) {
147 gfs2_lvb_unhold(new_qd->qd_gl);
148 kfree(new_qd);
149 }
150 *qdp = qd;
151 return 0;
152 }
153
154 error = qd_alloc(sdp, user, id, &new_qd);
155 if (error)
156 return error;
157 }
158}
159
160static void qd_hold(struct gfs2_quota_data *qd)
161{
162 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
163
164 spin_lock(&sdp->sd_quota_spin);
165 gfs2_assert(sdp, qd->qd_count);
166 qd->qd_count++;
167 spin_unlock(&sdp->sd_quota_spin);
168}
169
170static void qd_put(struct gfs2_quota_data *qd)
171{
172 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
173 spin_lock(&sdp->sd_quota_spin);
174 gfs2_assert(sdp, qd->qd_count);
175 if (!--qd->qd_count)
176 qd->qd_last_touched = jiffies;
177 spin_unlock(&sdp->sd_quota_spin);
178}
179
180static int slot_get(struct gfs2_quota_data *qd)
181{
182 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
183 unsigned int c, o = 0, b;
184 unsigned char byte = 0;
185
186 spin_lock(&sdp->sd_quota_spin);
187
188 if (qd->qd_slot_count++) {
189 spin_unlock(&sdp->sd_quota_spin);
190 return 0;
191 }
192
193 for (c = 0; c < sdp->sd_quota_chunks; c++)
194 for (o = 0; o < PAGE_SIZE; o++) {
195 byte = sdp->sd_quota_bitmap[c][o];
196 if (byte != 0xFF)
197 goto found;
198 }
199
200 goto fail;
201
a91ea69f 202found:
b3b94faa
DT		 203 for (b = 0; b < 8; b++)
204 if (!(byte & (1 << b)))
205 break;
206 qd->qd_slot = c * (8 * PAGE_SIZE) + o * 8 + b;
207
208 if (qd->qd_slot >= sdp->sd_quota_slots)
209 goto fail;
210
211 sdp->sd_quota_bitmap[c][o] |= 1 << b;
212
213 spin_unlock(&sdp->sd_quota_spin);
214
215 return 0;
216
a91ea69f 217fail:
b3b94faa
DT		 218 qd->qd_slot_count--;
219 spin_unlock(&sdp->sd_quota_spin);
220 return -ENOSPC;
221}
222
223static void slot_hold(struct gfs2_quota_data *qd)
224{
225 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
226
227 spin_lock(&sdp->sd_quota_spin);
228 gfs2_assert(sdp, qd->qd_slot_count);
229 qd->qd_slot_count++;
230 spin_unlock(&sdp->sd_quota_spin);
231}
232
233static void slot_put(struct gfs2_quota_data *qd)
234{
235 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
236
237 spin_lock(&sdp->sd_quota_spin);
238 gfs2_assert(sdp, qd->qd_slot_count);
239 if (!--qd->qd_slot_count) {
240 gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, qd->qd_slot, 0);
241 qd->qd_slot = -1;
242 }
243 spin_unlock(&sdp->sd_quota_spin);
244}
245
246static int bh_get(struct gfs2_quota_data *qd)
247{
248 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
feaa7bba 249 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
b3b94faa 250 unsigned int block, offset;
b3b94faa
DT		 251 struct buffer_head *bh;
252 int error;
7a6bbacb 253 struct buffer_head bh_map;
b3b94faa 254
f55ab26a 255 mutex_lock(&sdp->sd_quota_mutex);
b3b94faa
DT		 256
257 if (qd->qd_bh_count++) {
f55ab26a 258 mutex_unlock(&sdp->sd_quota_mutex);
b3b94faa
DT		 259 return 0;
260 }
261
262 block = qd->qd_slot / sdp->sd_qc_per_block;
263 offset = qd->qd_slot % sdp->sd_qc_per_block;;
264
7a6bbacb 265 error = gfs2_block_map(&ip->i_inode, block, 0, &bh_map, 1);
b3b94faa
DT		 266 if (error)
267 goto fail;
7a6bbacb 268 error = gfs2_meta_read(ip->i_gl, bh_map.b_blocknr, DIO_START | DIO_WAIT, &bh);
b3b94faa
DT		 269 if (error)
270 goto fail;
271 error = -EIO;
272 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC))
273 goto fail_brelse;
274
275 qd->qd_bh = bh;
276 qd->qd_bh_qc = (struct gfs2_quota_change *)
277 (bh->b_data + sizeof(struct gfs2_meta_header) +
278 offset * sizeof(struct gfs2_quota_change));
279
f55ab26a 280 mutex_lock(&sdp->sd_quota_mutex);
b3b94faa
DT		 281
282 return 0;
283
a91ea69f 284fail_brelse:
b3b94faa 285 brelse(bh);
a91ea69f 286fail:
b3b94faa 287 qd->qd_bh_count--;
f55ab26a 288 mutex_unlock(&sdp->sd_quota_mutex);
b3b94faa
DT		 289 return error;
290}
291
292static void bh_put(struct gfs2_quota_data *qd)
293{
294 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
295
f55ab26a 296 mutex_lock(&sdp->sd_quota_mutex);
b3b94faa
DT		 297 gfs2_assert(sdp, qd->qd_bh_count);
298 if (!--qd->qd_bh_count) {
299 brelse(qd->qd_bh);
300 qd->qd_bh = NULL;
301 qd->qd_bh_qc = NULL;
302 }
f55ab26a 303 mutex_unlock(&sdp->sd_quota_mutex);
b3b94faa
DT		 304}
305
306static int qd_fish(struct gfs2_sbd *sdp, struct gfs2_quota_data **qdp)
307{
308 struct gfs2_quota_data *qd = NULL;
309 int error;
310 int found = 0;
311
312 *qdp = NULL;
313
314 if (sdp->sd_vfs->s_flags & MS_RDONLY)
315 return 0;
316
317 spin_lock(&sdp->sd_quota_spin);
318
319 list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
320 if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
321 !test_bit(QDF_CHANGE, &qd->qd_flags) ||
322 qd->qd_sync_gen >= sdp->sd_quota_sync_gen)
323 continue;
324
325 list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
326
327 set_bit(QDF_LOCKED, &qd->qd_flags);
328 gfs2_assert_warn(sdp, qd->qd_count);
329 qd->qd_count++;
330 qd->qd_change_sync = qd->qd_change;
331 gfs2_assert_warn(sdp, qd->qd_slot_count);
332 qd->qd_slot_count++;
333 found = 1;
334
335 break;
336 }
337
338 if (!found)
339 qd = NULL;
340
341 spin_unlock(&sdp->sd_quota_spin);
342
343 if (qd) {
344 gfs2_assert_warn(sdp, qd->qd_change_sync);
345 error = bh_get(qd);
346 if (error) {
347 clear_bit(QDF_LOCKED, &qd->qd_flags);
348 slot_put(qd);
349 qd_put(qd);
350 return error;
351 }
352 }
353
354 *qdp = qd;
355
356 return 0;
357}
358
359static int qd_trylock(struct gfs2_quota_data *qd)
360{
361 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
362
363 if (sdp->sd_vfs->s_flags & MS_RDONLY)
364 return 0;
365
366 spin_lock(&sdp->sd_quota_spin);
367
368 if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
369 !test_bit(QDF_CHANGE, &qd->qd_flags)) {
370 spin_unlock(&sdp->sd_quota_spin);
371 return 0;
372 }
373
374 list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
375
376 set_bit(QDF_LOCKED, &qd->qd_flags);
377 gfs2_assert_warn(sdp, qd->qd_count);
378 qd->qd_count++;
379 qd->qd_change_sync = qd->qd_change;
380 gfs2_assert_warn(sdp, qd->qd_slot_count);
381 qd->qd_slot_count++;
382
383 spin_unlock(&sdp->sd_quota_spin);
384
385 gfs2_assert_warn(sdp, qd->qd_change_sync);
386 if (bh_get(qd)) {
387 clear_bit(QDF_LOCKED, &qd->qd_flags);
388 slot_put(qd);
389 qd_put(qd);
390 return 0;
391 }
392
393 return 1;
394}
395
396static void qd_unlock(struct gfs2_quota_data *qd)
397{
568f4c96
SW		 398 gfs2_assert_warn(qd->qd_gl->gl_sbd,
399 test_bit(QDF_LOCKED, &qd->qd_flags));
b3b94faa
DT		 400 clear_bit(QDF_LOCKED, &qd->qd_flags);
401 bh_put(qd);
402 slot_put(qd);
403 qd_put(qd);
404}
405
cd915493 406static int qdsb_get(struct gfs2_sbd *sdp, int user, u32 id, int create,
b3b94faa
DT		 407 struct gfs2_quota_data **qdp)
408{
409 int error;
410
411 error = qd_get(sdp, user, id, create, qdp);
412 if (error)
413 return error;
414
415 error = slot_get(*qdp);
416 if (error)
417 goto fail;
418
419 error = bh_get(*qdp);
420 if (error)
421 goto fail_slot;
422
423 return 0;
424
a91ea69f 425fail_slot:
b3b94faa 426 slot_put(*qdp);
a91ea69f 427fail:
b3b94faa
DT		 428 qd_put(*qdp);
429 return error;
430}
431
432static void qdsb_put(struct gfs2_quota_data *qd)
433{
434 bh_put(qd);
435 slot_put(qd);
436 qd_put(qd);
437}
438
cd915493 439int gfs2_quota_hold(struct gfs2_inode *ip, u32 uid, u32 gid)
b3b94faa 440{
feaa7bba 441 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
b3b94faa
DT		 442 struct gfs2_alloc *al = &ip->i_alloc;
443 struct gfs2_quota_data **qd = al->al_qd;
444 int error;
445
446 if (gfs2_assert_warn(sdp, !al->al_qd_num) ||
447 gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags)))
448 return -EIO;
449
450 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
451 return 0;
452
453 error = qdsb_get(sdp, QUOTA_USER, ip->i_di.di_uid, CREATE, qd);
454 if (error)
455 goto out;
456 al->al_qd_num++;
457 qd++;
458
459 error = qdsb_get(sdp, QUOTA_GROUP, ip->i_di.di_gid, CREATE, qd);
460 if (error)
461 goto out;
462 al->al_qd_num++;
463 qd++;
464
465 if (uid != NO_QUOTA_CHANGE && uid != ip->i_di.di_uid) {
466 error = qdsb_get(sdp, QUOTA_USER, uid, CREATE, qd);
467 if (error)
468 goto out;
469 al->al_qd_num++;
470 qd++;
471 }
472
473 if (gid != NO_QUOTA_CHANGE && gid != ip->i_di.di_gid) {
474 error = qdsb_get(sdp, QUOTA_GROUP, gid, CREATE, qd);
475 if (error)
476 goto out;
477 al->al_qd_num++;
478 qd++;
479 }
480
a91ea69f 481out:
b3b94faa
DT		 482 if (error)
483 gfs2_quota_unhold(ip);
b3b94faa
DT		 484 return error;
485}
486
487void gfs2_quota_unhold(struct gfs2_inode *ip)
488{
feaa7bba 489 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
b3b94faa
DT		 490 struct gfs2_alloc *al = &ip->i_alloc;
491 unsigned int x;
492
493 gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags));
494
495 for (x = 0; x < al->al_qd_num; x++) {
496 qdsb_put(al->al_qd[x]);
497 al->al_qd[x] = NULL;
498 }
499 al->al_qd_num = 0;
500}
501
502static int sort_qd(const void *a, const void *b)
503{
48fac179
SW		 504 const struct gfs2_quota_data *qd_a = *(const struct gfs2_quota_data **)a;
505 const struct gfs2_quota_data *qd_b = *(const struct gfs2_quota_data **)b;
b3b94faa
DT		 506
507 if (!test_bit(QDF_USER, &qd_a->qd_flags) !=
508 !test_bit(QDF_USER, &qd_b->qd_flags)) {
509 if (test_bit(QDF_USER, &qd_a->qd_flags))
48fac179 510 return -1;
b3b94faa 511 else
48fac179 512 return 1;
b3b94faa 513 }
48fac179
SW		 514 if (qd_a->qd_id < qd_b->qd_id)
515 return -1;
516 if (qd_a->qd_id > qd_b->qd_id)
517 return 1;
b3b94faa 518
48fac179 519 return 0;
b3b94faa
DT		 520}
521
cd915493 522static void do_qc(struct gfs2_quota_data *qd, s64 change)
b3b94faa
DT		 523{
524 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
feaa7bba 525 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
b3b94faa 526 struct gfs2_quota_change *qc = qd->qd_bh_qc;
cd915493 527 s64 x;
b3b94faa 528
f55ab26a 529 mutex_lock(&sdp->sd_quota_mutex);
d4e9c4c3 530 gfs2_trans_add_bh(ip->i_gl, qd->qd_bh, 1);
b3b94faa
DT		 531
532 if (!test_bit(QDF_CHANGE, &qd->qd_flags)) {
533 qc->qc_change = 0;
534 qc->qc_flags = 0;
535 if (test_bit(QDF_USER, &qd->qd_flags))
536 qc->qc_flags = cpu_to_be32(GFS2_QCF_USER);
537 qc->qc_id = cpu_to_be32(qd->qd_id);
538 }
539
540 x = qc->qc_change;
541 x = be64_to_cpu(x) + change;
542 qc->qc_change = cpu_to_be64(x);
543
544 spin_lock(&sdp->sd_quota_spin);
545 qd->qd_change = x;
546 spin_unlock(&sdp->sd_quota_spin);
547
548 if (!x) {
549 gfs2_assert_warn(sdp, test_bit(QDF_CHANGE, &qd->qd_flags));
550 clear_bit(QDF_CHANGE, &qd->qd_flags);
551 qc->qc_flags = 0;
552 qc->qc_id = 0;
553 slot_put(qd);
554 qd_put(qd);
555 } else if (!test_and_set_bit(QDF_CHANGE, &qd->qd_flags)) {
556 qd_hold(qd);
557 slot_hold(qd);
558 }
559
f55ab26a 560 mutex_unlock(&sdp->sd_quota_mutex);
b3b94faa
DT		 561}
562
18ec7d5c
SW		 563/**
564 * gfs2_adjust_quota
565 *
566 * This function was mostly borrowed from gfs2_block_truncate_page which was
567 * in turn mostly borrowed from ext3
568 */
569static int gfs2_adjust_quota(struct gfs2_inode *ip, loff_t loc,
cd915493 570 s64 change, struct gfs2_quota_data *qd)
18ec7d5c 571{
feaa7bba 572 struct inode *inode = &ip->i_inode;
18ec7d5c
SW		 573 struct address_space *mapping = inode->i_mapping;
574 unsigned long index = loc >> PAGE_CACHE_SHIFT;
575 unsigned offset = loc & (PAGE_CACHE_SHIFT - 1);
576 unsigned blocksize, iblock, pos;
577 struct buffer_head *bh;
578 struct page *page;
579 void *kaddr;
580 __be64 *ptr;
e9fc2aa0 581 s64 value;
18ec7d5c
SW		 582 int err = -EIO;
583
584 page = grab_cache_page(mapping, index);
585 if (!page)
586 return -ENOMEM;
587
588 blocksize = inode->i_sb->s_blocksize;
589 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
590
591 if (!page_has_buffers(page))
592 create_empty_buffers(page, blocksize, 0);
593
594 bh = page_buffers(page);
595 pos = blocksize;
596 while (offset >= pos) {
597 bh = bh->b_this_page;
598 iblock++;
599 pos += blocksize;
600 }
601
602 if (!buffer_mapped(bh)) {
603 gfs2_get_block(inode, iblock, bh, 1);
604 if (!buffer_mapped(bh))
605 goto unlock;
606 }
607
608 if (PageUptodate(page))
609 set_buffer_uptodate(bh);
610
611 if (!buffer_uptodate(bh)) {
612 ll_rw_block(READ, 1, &bh);
613 wait_on_buffer(bh);
614 if (!buffer_uptodate(bh))
615 goto unlock;
616 }
617
618 gfs2_trans_add_bh(ip->i_gl, bh, 0);
619
620 kaddr = kmap_atomic(page, KM_USER0);
48fac179 621 ptr = kaddr + offset;
e9fc2aa0
SW		 622 value = (s64)be64_to_cpu(*ptr) + change;
623 *ptr = cpu_to_be64(value);
18ec7d5c
SW		 624 flush_dcache_page(page);
625 kunmap_atomic(kaddr, KM_USER0);
626 err = 0;
627 qd->qd_qb.qb_magic = cpu_to_be32(GFS2_MAGIC);
18ec7d5c
SW		 628 qd->qd_qb.qb_value = cpu_to_be64(value);
629unlock:
630 unlock_page(page);
631 page_cache_release(page);
632 return err;
633}
634
b3b94faa
DT		 635static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
636{
637 struct gfs2_sbd *sdp = (*qda)->qd_gl->gl_sbd;
feaa7bba 638 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
b3b94faa
DT		 639 unsigned int data_blocks, ind_blocks;
640 struct gfs2_holder *ghs, i_gh;
641 unsigned int qx, x;
642 struct gfs2_quota_data *qd;
f42faf4f 643 loff_t offset;
b3b94faa
DT		 644 unsigned int nalloc = 0;
645 struct gfs2_alloc *al = NULL;
646 int error;
647
648 gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
649 &data_blocks, &ind_blocks);
650
651 ghs = kcalloc(num_qd, sizeof(struct gfs2_holder), GFP_KERNEL);
652 if (!ghs)
653 return -ENOMEM;
654
655 sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL);
656 for (qx = 0; qx < num_qd; qx++) {
657 error = gfs2_glock_nq_init(qda[qx]->qd_gl,
658 LM_ST_EXCLUSIVE,
659 GL_NOCACHE, &ghs[qx]);
660 if (error)
661 goto out;
662 }
663
664 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
665 if (error)
666 goto out;
667
668 for (x = 0; x < num_qd; x++) {
669 int alloc_required;
670
671 offset = qd2offset(qda[x]);
672 error = gfs2_write_alloc_required(ip, offset,
673 sizeof(struct gfs2_quota),
674 &alloc_required);
675 if (error)
676 goto out_gunlock;
677 if (alloc_required)
678 nalloc++;
679 }
680
681 if (nalloc) {
682 al = gfs2_alloc_get(ip);
683
684 al->al_requested = nalloc * (data_blocks + ind_blocks);
685
686 error = gfs2_inplace_reserve(ip);
687 if (error)
688 goto out_alloc;
689
690 error = gfs2_trans_begin(sdp,
691 al->al_rgd->rd_ri.ri_length +
692 num_qd * data_blocks +
693 nalloc * ind_blocks +
694 RES_DINODE + num_qd +
695 RES_STATFS, 0);
696 if (error)
697 goto out_ipres;
698 } else {
699 error = gfs2_trans_begin(sdp,
700 num_qd * data_blocks +
701 RES_DINODE + num_qd, 0);
702 if (error)
703 goto out_gunlock;
704 }
705
706 for (x = 0; x < num_qd; x++) {
b3b94faa
DT		 707 qd = qda[x];
708 offset = qd2offset(qd);
18ec7d5c 709 error = gfs2_adjust_quota(ip, offset, qd->qd_change_sync,
568f4c96
SW		 710 (struct gfs2_quota_data *)
711 qd->qd_gl->gl_lvb);
18ec7d5c 712 if (error)
b3b94faa 713 goto out_end_trans;
b3b94faa
DT		 714
715 do_qc(qd, -qd->qd_change_sync);
b3b94faa
DT		 716 }
717
718 error = 0;
719
a91ea69f 720out_end_trans:
b3b94faa 721 gfs2_trans_end(sdp);
a91ea69f 722out_ipres:
b3b94faa
DT		 723 if (nalloc)
724 gfs2_inplace_release(ip);
a91ea69f 725out_alloc:
b3b94faa
DT		 726 if (nalloc)
727 gfs2_alloc_put(ip);
a91ea69f 728out_gunlock:
b3b94faa 729 gfs2_glock_dq_uninit(&i_gh);
a91ea69f 730out:
b3b94faa
DT		 731 while (qx--)
732 gfs2_glock_dq_uninit(&ghs[qx]);
733 kfree(ghs);
b09e593d 734 gfs2_log_flush(ip->i_gl->gl_sbd, ip->i_gl);
b3b94faa
DT		 735 return error;
736}
737
738static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
739 struct gfs2_holder *q_gh)
740{
741 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
feaa7bba 742 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
b3b94faa
DT		 743 struct gfs2_holder i_gh;
744 struct gfs2_quota q;
745 char buf[sizeof(struct gfs2_quota)];
f42faf4f 746 struct file_ra_state ra_state;
b3b94faa 747 int error;
e9fc2aa0 748 struct gfs2_quota_lvb *qlvb;
b3b94faa 749
f42faf4f 750 file_ra_state_init(&ra_state, sdp->sd_quota_inode->i_mapping);
a91ea69f 751restart:
b3b94faa
DT		 752 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
753 if (error)
754 return error;
755
e9fc2aa0 756 qd->qd_qb = *(struct gfs2_quota_lvb *)qd->qd_gl->gl_lvb;
b3b94faa 757
e9fc2aa0 758 if (force_refresh || qd->qd_qb.qb_magic != cpu_to_be32(GFS2_MAGIC)) {
f42faf4f 759 loff_t pos;
b3b94faa
DT		 760 gfs2_glock_dq_uninit(q_gh);
761 error = gfs2_glock_nq_init(qd->qd_gl,
762 LM_ST_EXCLUSIVE, GL_NOCACHE,
763 q_gh);
764 if (error)
765 return error;
766
e9fc2aa0 767 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
b3b94faa
DT		 768 if (error)
769 goto fail;
770
771 memset(buf, 0, sizeof(struct gfs2_quota));
f42faf4f 772 pos = qd2offset(qd);
0d42e542
SW		 773 error = gfs2_internal_read(ip, &ra_state, buf,
774 &pos, sizeof(struct gfs2_quota));
b3b94faa
DT		 775 if (error < 0)
776 goto fail_gunlock;
777
778 gfs2_glock_dq_uninit(&i_gh);
779
e9fc2aa0 780
b3b94faa 781 gfs2_quota_in(&q, buf);
e9fc2aa0
SW		 782 qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lvb;
783 qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC);
784 qlvb->__pad = 0;
785 qlvb->qb_limit = cpu_to_be64(q.qu_limit);
786 qlvb->qb_warn = cpu_to_be64(q.qu_warn);
787 qlvb->qb_value = cpu_to_be64(q.qu_value);
788 qd->qd_qb = *qlvb;
b3b94faa
DT		 789
790 if (gfs2_glock_is_blocking(qd->qd_gl)) {
791 gfs2_glock_dq_uninit(q_gh);
792 force_refresh = 0;
793 goto restart;
794 }
795 }
796
797 return 0;
798
a91ea69f 799fail_gunlock:
b3b94faa 800 gfs2_glock_dq_uninit(&i_gh);
a91ea69f 801fail:
b3b94faa 802 gfs2_glock_dq_uninit(q_gh);
b3b94faa
DT		 803 return error;
804}
805
cd915493 806int gfs2_quota_lock(struct gfs2_inode *ip, u32 uid, u32 gid)
b3b94faa 807{
feaa7bba 808 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
b3b94faa
DT		 809 struct gfs2_alloc *al = &ip->i_alloc;
810 unsigned int x;
811 int error = 0;
812
813 gfs2_quota_hold(ip, uid, gid);
814
815 if (capable(CAP_SYS_RESOURCE) ||
816 sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
817 return 0;
818
819 sort(al->al_qd, al->al_qd_num, sizeof(struct gfs2_quota_data *),
820 sort_qd, NULL);
821
822 for (x = 0; x < al->al_qd_num; x++) {
823 error = do_glock(al->al_qd[x], NO_FORCE, &al->al_qd_ghs[x]);
824 if (error)
825 break;
826 }
827
828 if (!error)
829 set_bit(GIF_QD_LOCKED, &ip->i_flags);
830 else {
831 while (x--)
832 gfs2_glock_dq_uninit(&al->al_qd_ghs[x]);
833 gfs2_quota_unhold(ip);
834 }
835
836 return error;
837}
838
839static int need_sync(struct gfs2_quota_data *qd)
840{
841 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
842 struct gfs2_tune *gt = &sdp->sd_tune;
cd915493 843 s64 value;
b3b94faa
DT		 844 unsigned int num, den;
845 int do_sync = 1;
846
847 if (!qd->qd_qb.qb_limit)
848 return 0;
849
850 spin_lock(&sdp->sd_quota_spin);
851 value = qd->qd_change;
852 spin_unlock(&sdp->sd_quota_spin);
853
854 spin_lock(&gt->gt_spin);
855 num = gt->gt_quota_scale_num;
856 den = gt->gt_quota_scale_den;
857 spin_unlock(&gt->gt_spin);
858
859 if (value < 0)
860 do_sync = 0;
e9fc2aa0
SW		 861 else if ((s64)be64_to_cpu(qd->qd_qb.qb_value) >=
862 (s64)be64_to_cpu(qd->qd_qb.qb_limit))
b3b94faa
DT		 863 do_sync = 0;
864 else {
865 value *= gfs2_jindex_size(sdp) * num;
866 do_div(value, den);
e9fc2aa0 867 value += (s64)be64_to_cpu(qd->qd_qb.qb_value);
cd915493 868 if (value < (s64)be64_to_cpu(qd->qd_qb.qb_limit))
b3b94faa
DT		 869 do_sync = 0;
870 }
871
872 return do_sync;
873}
874
875void gfs2_quota_unlock(struct gfs2_inode *ip)
876{
877 struct gfs2_alloc *al = &ip->i_alloc;
878 struct gfs2_quota_data *qda[4];
879 unsigned int count = 0;
880 unsigned int x;
881
882 if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
883 goto out;
884
885 for (x = 0; x < al->al_qd_num; x++) {
886 struct gfs2_quota_data *qd;
887 int sync;
888
889 qd = al->al_qd[x];
890 sync = need_sync(qd);
891
892 gfs2_glock_dq_uninit(&al->al_qd_ghs[x]);
893
894 if (sync && qd_trylock(qd))
895 qda[count++] = qd;
896 }
897
898 if (count) {
899 do_sync(count, qda);
900 for (x = 0; x < count; x++)
901 qd_unlock(qda[x]);
902 }
903
a91ea69f 904out:
b3b94faa
DT		 905 gfs2_quota_unhold(ip);
906}
907
908#define MAX_LINE 256
909
910static int print_message(struct gfs2_quota_data *qd, char *type)
911{
912 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
b3b94faa 913
02630a12
SW		 914 printk(KERN_INFO "GFS2: fsid=%s: quota %s for %s %u\r\n",
915 sdp->sd_fsname, type,
916 (test_bit(QDF_USER, &qd->qd_flags)) ? "user" : "group",
917 qd->qd_id);
b3b94faa
DT		 918
919 return 0;
920}
921
cd915493 922int gfs2_quota_check(struct gfs2_inode *ip, u32 uid, u32 gid)
b3b94faa 923{
feaa7bba 924 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
b3b94faa
DT		 925 struct gfs2_alloc *al = &ip->i_alloc;
926 struct gfs2_quota_data *qd;
cd915493 927 s64 value;
b3b94faa
DT		 928 unsigned int x;
929 int error = 0;
930
931 if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
932 return 0;
933
934 if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
935 return 0;
936
937 for (x = 0; x < al->al_qd_num; x++) {
938 qd = al->al_qd[x];
939
940 if (!((qd->qd_id == uid && test_bit(QDF_USER, &qd->qd_flags)) ||
941 (qd->qd_id == gid && !test_bit(QDF_USER, &qd->qd_flags))))
942 continue;
943
e9fc2aa0 944 value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
b3b94faa
DT		 945 spin_lock(&sdp->sd_quota_spin);
946 value += qd->qd_change;
947 spin_unlock(&sdp->sd_quota_spin);
948
cd915493 949 if (be64_to_cpu(qd->qd_qb.qb_limit) && (s64)be64_to_cpu(qd->qd_qb.qb_limit) < value) {
b3b94faa
DT		 950 print_message(qd, "exceeded");
951 error = -EDQUOT;
952 break;
e9fc2aa0 953 } else if (be64_to_cpu(qd->qd_qb.qb_warn) &&
cd915493 954 (s64)be64_to_cpu(qd->qd_qb.qb_warn) < value &&
b3b94faa 955 time_after_eq(jiffies, qd->qd_last_warn +
568f4c96
SW		 956 gfs2_tune_get(sdp,
957 gt_quota_warn_period) * HZ)) {
b3b94faa
DT		 958 error = print_message(qd, "warning");
959 qd->qd_last_warn = jiffies;
960 }
961 }
962
963 return error;
964}
965
cd915493
SW		 966void gfs2_quota_change(struct gfs2_inode *ip, s64 change,
967 u32 uid, u32 gid)
b3b94faa
DT		 968{
969 struct gfs2_alloc *al = &ip->i_alloc;
970 struct gfs2_quota_data *qd;
971 unsigned int x;
972 unsigned int found = 0;
973
feaa7bba 974 if (gfs2_assert_warn(GFS2_SB(&ip->i_inode), change))
b3b94faa
DT		 975 return;
976 if (ip->i_di.di_flags & GFS2_DIF_SYSTEM)
977 return;
978
979 for (x = 0; x < al->al_qd_num; x++) {
980 qd = al->al_qd[x];
981
982 if ((qd->qd_id == uid && test_bit(QDF_USER, &qd->qd_flags)) ||
983 (qd->qd_id == gid && !test_bit(QDF_USER, &qd->qd_flags))) {
984 do_qc(qd, change);
985 found++;
986 }
987 }
988}
989
990int gfs2_quota_sync(struct gfs2_sbd *sdp)
991{
992 struct gfs2_quota_data **qda;
993 unsigned int max_qd = gfs2_tune_get(sdp, gt_quota_simul_sync);
994 unsigned int num_qd;
995 unsigned int x;
996 int error = 0;
997
998 sdp->sd_quota_sync_gen++;
999
1000 qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
1001 if (!qda)
1002 return -ENOMEM;
1003
1004 do {
1005 num_qd = 0;
1006
1007 for (;;) {
1008 error = qd_fish(sdp, qda + num_qd);
1009 if (error || !qda[num_qd])
1010 break;
1011 if (++num_qd == max_qd)
1012 break;
1013 }
1014
1015 if (num_qd) {
1016 if (!error)
1017 error = do_sync(num_qd, qda);
1018 if (!error)
1019 for (x = 0; x < num_qd; x++)
1020 qda[x]->qd_sync_gen =
1021 sdp->sd_quota_sync_gen;
1022
1023 for (x = 0; x < num_qd; x++)
1024 qd_unlock(qda[x]);
1025 }
1026 } while (!error && num_qd == max_qd);
1027
1028 kfree(qda);
1029
1030 return error;
1031}
1032
cd915493 1033int gfs2_quota_refresh(struct gfs2_sbd *sdp, int user, u32 id)
b3b94faa
DT		 1034{
1035 struct gfs2_quota_data *qd;
1036 struct gfs2_holder q_gh;
1037 int error;
1038
1039 error = qd_get(sdp, user, id, CREATE, &qd);
1040 if (error)
1041 return error;
1042
1043 error = do_glock(qd, FORCE, &q_gh);
1044 if (!error)
1045 gfs2_glock_dq_uninit(&q_gh);
1046
1047 qd_put(qd);
1048
1049 return error;
1050}
1051
b3b94faa
DT		 1052int gfs2_quota_init(struct gfs2_sbd *sdp)
1053{
feaa7bba 1054 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
b3b94faa
DT		 1055 unsigned int blocks = ip->i_di.di_size >> sdp->sd_sb.sb_bsize_shift;
1056 unsigned int x, slot = 0;
1057 unsigned int found = 0;
cd915493
SW		 1058 u64 dblock;
1059 u32 extlen = 0;
b3b94faa
DT		 1060 int error;
1061
1062 if (!ip->i_di.di_size ||
1063 ip->i_di.di_size > (64 << 20) ||
1064 ip->i_di.di_size & (sdp->sd_sb.sb_bsize - 1)) {
1065 gfs2_consist_inode(ip);
1066 return -EIO;
1067 }
1068 sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
5c676f6d 1069 sdp->sd_quota_chunks = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * PAGE_SIZE);
b3b94faa
DT		 1070
1071 error = -ENOMEM;
1072
1073 sdp->sd_quota_bitmap = kcalloc(sdp->sd_quota_chunks,
1074 sizeof(unsigned char *), GFP_KERNEL);
1075 if (!sdp->sd_quota_bitmap)
1076 return error;
1077
1078 for (x = 0; x < sdp->sd_quota_chunks; x++) {
1079 sdp->sd_quota_bitmap[x] = kzalloc(PAGE_SIZE, GFP_KERNEL);
1080 if (!sdp->sd_quota_bitmap[x])
1081 goto fail;
1082 }
1083
1084 for (x = 0; x < blocks; x++) {
1085 struct buffer_head *bh;
1086 unsigned int y;
1087
1088 if (!extlen) {
1089 int new = 0;
feaa7bba 1090 error = gfs2_extent_map(&ip->i_inode, x, &new, &dblock, &extlen);
b3b94faa
DT		 1091 if (error)
1092 goto fail;
1093 }
1094 gfs2_meta_ra(ip->i_gl, dblock, extlen);
1095 error = gfs2_meta_read(ip->i_gl, dblock, DIO_START | DIO_WAIT,
1096 &bh);
1097 if (error)
1098 goto fail;
1099 error = -EIO;
1100 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
1101 brelse(bh);
1102 goto fail;
1103 }
1104
1105 for (y = 0;
1106 y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1107 y++, slot++) {
1108 struct gfs2_quota_change qc;
1109 struct gfs2_quota_data *qd;
1110
1111 gfs2_quota_change_in(&qc, bh->b_data +
1112 sizeof(struct gfs2_meta_header) +
1113 y * sizeof(struct gfs2_quota_change));
1114 if (!qc.qc_change)
1115 continue;
1116
1117 error = qd_alloc(sdp, (qc.qc_flags & GFS2_QCF_USER),
1118 qc.qc_id, &qd);
1119 if (error) {
1120 brelse(bh);
1121 goto fail;
1122 }
1123
1124 set_bit(QDF_CHANGE, &qd->qd_flags);
1125 qd->qd_change = qc.qc_change;
1126 qd->qd_slot = slot;
1127 qd->qd_slot_count = 1;
1128 qd->qd_last_touched = jiffies;
1129
1130 spin_lock(&sdp->sd_quota_spin);
1131 gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, slot, 1);
1132 list_add(&qd->qd_list, &sdp->sd_quota_list);
1133 atomic_inc(&sdp->sd_quota_count);
1134 spin_unlock(&sdp->sd_quota_spin);
1135
1136 found++;
1137 }
1138
1139 brelse(bh);
1140 dblock++;
1141 extlen--;
1142 }
1143
1144 if (found)
1145 fs_info(sdp, "found %u quota changes\n", found);
1146
1147 return 0;
1148
a91ea69f 1149fail:
b3b94faa
DT		 1150 gfs2_quota_cleanup(sdp);
1151 return error;
1152}
1153
1154void gfs2_quota_scan(struct gfs2_sbd *sdp)
1155{
1156 struct gfs2_quota_data *qd, *safe;
1157 LIST_HEAD(dead);
1158
1159 spin_lock(&sdp->sd_quota_spin);
1160 list_for_each_entry_safe(qd, safe, &sdp->sd_quota_list, qd_list) {
1161 if (!qd->qd_count &&
1162 time_after_eq(jiffies, qd->qd_last_touched +
1163 gfs2_tune_get(sdp, gt_quota_cache_secs) * HZ)) {
1164 list_move(&qd->qd_list, &dead);
1165 gfs2_assert_warn(sdp,
1166 atomic_read(&sdp->sd_quota_count) > 0);
1167 atomic_dec(&sdp->sd_quota_count);
1168 }
1169 }
1170 spin_unlock(&sdp->sd_quota_spin);
1171
1172 while (!list_empty(&dead)) {
1173 qd = list_entry(dead.next, struct gfs2_quota_data, qd_list);
1174 list_del(&qd->qd_list);
1175
1176 gfs2_assert_warn(sdp, !qd->qd_change);
1177 gfs2_assert_warn(sdp, !qd->qd_slot_count);
1178 gfs2_assert_warn(sdp, !qd->qd_bh_count);
1179
1180 gfs2_lvb_unhold(qd->qd_gl);
1181 kfree(qd);
1182 }
1183}
1184
1185void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1186{
1187 struct list_head *head = &sdp->sd_quota_list;
1188 struct gfs2_quota_data *qd;
1189 unsigned int x;
1190
1191 spin_lock(&sdp->sd_quota_spin);
1192 while (!list_empty(head)) {
1193 qd = list_entry(head->prev, struct gfs2_quota_data, qd_list);
1194
1195 if (qd->qd_count > 1 ||
1196 (qd->qd_count && !test_bit(QDF_CHANGE, &qd->qd_flags))) {
1197 list_move(&qd->qd_list, head);
1198 spin_unlock(&sdp->sd_quota_spin);
1199 schedule();
1200 spin_lock(&sdp->sd_quota_spin);
1201 continue;
1202 }
1203
1204 list_del(&qd->qd_list);
1205 atomic_dec(&sdp->sd_quota_count);
1206 spin_unlock(&sdp->sd_quota_spin);
1207
1208 if (!qd->qd_count) {
1209 gfs2_assert_warn(sdp, !qd->qd_change);
1210 gfs2_assert_warn(sdp, !qd->qd_slot_count);
1211 } else
1212 gfs2_assert_warn(sdp, qd->qd_slot_count == 1);
1213 gfs2_assert_warn(sdp, !qd->qd_bh_count);
1214
1215 gfs2_lvb_unhold(qd->qd_gl);
1216 kfree(qd);
1217
1218 spin_lock(&sdp->sd_quota_spin);
1219 }
1220 spin_unlock(&sdp->sd_quota_spin);
1221
1222 gfs2_assert_warn(sdp, !atomic_read(&sdp->sd_quota_count));
1223
1224 if (sdp->sd_quota_bitmap) {
1225 for (x = 0; x < sdp->sd_quota_chunks; x++)
1226 kfree(sdp->sd_quota_bitmap[x]);
1227 kfree(sdp->sd_quota_bitmap);
1228 }
1229}
1230
Clone of davem's net-next-2.6 tree