i386-resolve-dependency-of-asm-i386-pgtableh-on-highmemh
[linux-2.6/linux-trees-mm.git] / fs / gfs2 / quota.c
blob4996f0ef30076aa0ba30425c4d5b42fe96141226
1 /*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
8 */
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.
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.
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.
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>
44 #include <linux/sort.h>
45 #include <linux/fs.h>
46 #include <linux/bio.h>
47 #include <linux/gfs2_ondisk.h>
48 #include <linux/lm_interface.h>
50 #include "gfs2.h"
51 #include "incore.h"
52 #include "bmap.h"
53 #include "glock.h"
54 #include "glops.h"
55 #include "log.h"
56 #include "meta_io.h"
57 #include "quota.h"
58 #include "rgrp.h"
59 #include "super.h"
60 #include "trans.h"
61 #include "inode.h"
62 #include "ops_address.h"
63 #include "util.h"
65 #define QUOTA_USER 1
66 #define QUOTA_GROUP 0
68 struct gfs2_quota_host {
69 u64 qu_limit;
70 u64 qu_warn;
71 s64 qu_value;
72 u32 qu_ll_next;
75 struct gfs2_quota_change_host {
76 u64 qc_change;
77 u32 qc_flags; /* GFS2_QCF_... */
78 u32 qc_id;
81 static u64 qd2offset(struct gfs2_quota_data *qd)
83 u64 offset;
85 offset = 2 * (u64)qd->qd_id + !test_bit(QDF_USER, &qd->qd_flags);
86 offset *= sizeof(struct gfs2_quota);
88 return offset;
91 static int qd_alloc(struct gfs2_sbd *sdp, int user, u32 id,
92 struct gfs2_quota_data **qdp)
94 struct gfs2_quota_data *qd;
95 int error;
97 qd = kzalloc(sizeof(struct gfs2_quota_data), GFP_KERNEL);
98 if (!qd)
99 return -ENOMEM;
101 qd->qd_count = 1;
102 qd->qd_id = id;
103 if (user)
104 set_bit(QDF_USER, &qd->qd_flags);
105 qd->qd_slot = -1;
107 error = gfs2_glock_get(sdp, 2 * (u64)id + !user,
108 &gfs2_quota_glops, CREATE, &qd->qd_gl);
109 if (error)
110 goto fail;
112 error = gfs2_lvb_hold(qd->qd_gl);
113 gfs2_glock_put(qd->qd_gl);
114 if (error)
115 goto fail;
117 *qdp = qd;
119 return 0;
121 fail:
122 kfree(qd);
123 return error;
126 static int qd_get(struct gfs2_sbd *sdp, int user, u32 id, int create,
127 struct gfs2_quota_data **qdp)
129 struct gfs2_quota_data *qd = NULL, *new_qd = NULL;
130 int error, found;
132 *qdp = NULL;
134 for (;;) {
135 found = 0;
136 spin_lock(&sdp->sd_quota_spin);
137 list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
138 if (qd->qd_id == id &&
139 !test_bit(QDF_USER, &qd->qd_flags) == !user) {
140 qd->qd_count++;
141 found = 1;
142 break;
146 if (!found)
147 qd = NULL;
149 if (!qd && new_qd) {
150 qd = new_qd;
151 list_add(&qd->qd_list, &sdp->sd_quota_list);
152 atomic_inc(&sdp->sd_quota_count);
153 new_qd = NULL;
156 spin_unlock(&sdp->sd_quota_spin);
158 if (qd || !create) {
159 if (new_qd) {
160 gfs2_lvb_unhold(new_qd->qd_gl);
161 kfree(new_qd);
163 *qdp = qd;
164 return 0;
167 error = qd_alloc(sdp, user, id, &new_qd);
168 if (error)
169 return error;
173 static void qd_hold(struct gfs2_quota_data *qd)
175 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
177 spin_lock(&sdp->sd_quota_spin);
178 gfs2_assert(sdp, qd->qd_count);
179 qd->qd_count++;
180 spin_unlock(&sdp->sd_quota_spin);
183 static void qd_put(struct gfs2_quota_data *qd)
185 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
186 spin_lock(&sdp->sd_quota_spin);
187 gfs2_assert(sdp, qd->qd_count);
188 if (!--qd->qd_count)
189 qd->qd_last_touched = jiffies;
190 spin_unlock(&sdp->sd_quota_spin);
193 static int slot_get(struct gfs2_quota_data *qd)
195 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
196 unsigned int c, o = 0, b;
197 unsigned char byte = 0;
199 spin_lock(&sdp->sd_quota_spin);
201 if (qd->qd_slot_count++) {
202 spin_unlock(&sdp->sd_quota_spin);
203 return 0;
206 for (c = 0; c < sdp->sd_quota_chunks; c++)
207 for (o = 0; o < PAGE_SIZE; o++) {
208 byte = sdp->sd_quota_bitmap[c][o];
209 if (byte != 0xFF)
210 goto found;
213 goto fail;
215 found:
216 for (b = 0; b < 8; b++)
217 if (!(byte & (1 << b)))
218 break;
219 qd->qd_slot = c * (8 * PAGE_SIZE) + o * 8 + b;
221 if (qd->qd_slot >= sdp->sd_quota_slots)
222 goto fail;
224 sdp->sd_quota_bitmap[c][o] |= 1 << b;
226 spin_unlock(&sdp->sd_quota_spin);
228 return 0;
230 fail:
231 qd->qd_slot_count--;
232 spin_unlock(&sdp->sd_quota_spin);
233 return -ENOSPC;
236 static void slot_hold(struct gfs2_quota_data *qd)
238 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
240 spin_lock(&sdp->sd_quota_spin);
241 gfs2_assert(sdp, qd->qd_slot_count);
242 qd->qd_slot_count++;
243 spin_unlock(&sdp->sd_quota_spin);
246 static void slot_put(struct gfs2_quota_data *qd)
248 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
250 spin_lock(&sdp->sd_quota_spin);
251 gfs2_assert(sdp, qd->qd_slot_count);
252 if (!--qd->qd_slot_count) {
253 gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, qd->qd_slot, 0);
254 qd->qd_slot = -1;
256 spin_unlock(&sdp->sd_quota_spin);
259 static int bh_get(struct gfs2_quota_data *qd)
261 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
262 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
263 unsigned int block, offset;
264 struct buffer_head *bh;
265 int error;
266 struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
268 mutex_lock(&sdp->sd_quota_mutex);
270 if (qd->qd_bh_count++) {
271 mutex_unlock(&sdp->sd_quota_mutex);
272 return 0;
275 block = qd->qd_slot / sdp->sd_qc_per_block;
276 offset = qd->qd_slot % sdp->sd_qc_per_block;;
278 bh_map.b_size = 1 << ip->i_inode.i_blkbits;
279 error = gfs2_block_map(&ip->i_inode, block, 0, &bh_map);
280 if (error)
281 goto fail;
282 error = gfs2_meta_read(ip->i_gl, bh_map.b_blocknr, DIO_WAIT, &bh);
283 if (error)
284 goto fail;
285 error = -EIO;
286 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC))
287 goto fail_brelse;
289 qd->qd_bh = bh;
290 qd->qd_bh_qc = (struct gfs2_quota_change *)
291 (bh->b_data + sizeof(struct gfs2_meta_header) +
292 offset * sizeof(struct gfs2_quota_change));
294 mutex_unlock(&sdp->sd_quota_mutex);
296 return 0;
298 fail_brelse:
299 brelse(bh);
300 fail:
301 qd->qd_bh_count--;
302 mutex_unlock(&sdp->sd_quota_mutex);
303 return error;
306 static void bh_put(struct gfs2_quota_data *qd)
308 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
310 mutex_lock(&sdp->sd_quota_mutex);
311 gfs2_assert(sdp, qd->qd_bh_count);
312 if (!--qd->qd_bh_count) {
313 brelse(qd->qd_bh);
314 qd->qd_bh = NULL;
315 qd->qd_bh_qc = NULL;
317 mutex_unlock(&sdp->sd_quota_mutex);
320 static int qd_fish(struct gfs2_sbd *sdp, struct gfs2_quota_data **qdp)
322 struct gfs2_quota_data *qd = NULL;
323 int error;
324 int found = 0;
326 *qdp = NULL;
328 if (sdp->sd_vfs->s_flags & MS_RDONLY)
329 return 0;
331 spin_lock(&sdp->sd_quota_spin);
333 list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
334 if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
335 !test_bit(QDF_CHANGE, &qd->qd_flags) ||
336 qd->qd_sync_gen >= sdp->sd_quota_sync_gen)
337 continue;
339 list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
341 set_bit(QDF_LOCKED, &qd->qd_flags);
342 gfs2_assert_warn(sdp, qd->qd_count);
343 qd->qd_count++;
344 qd->qd_change_sync = qd->qd_change;
345 gfs2_assert_warn(sdp, qd->qd_slot_count);
346 qd->qd_slot_count++;
347 found = 1;
349 break;
352 if (!found)
353 qd = NULL;
355 spin_unlock(&sdp->sd_quota_spin);
357 if (qd) {
358 gfs2_assert_warn(sdp, qd->qd_change_sync);
359 error = bh_get(qd);
360 if (error) {
361 clear_bit(QDF_LOCKED, &qd->qd_flags);
362 slot_put(qd);
363 qd_put(qd);
364 return error;
368 *qdp = qd;
370 return 0;
373 static int qd_trylock(struct gfs2_quota_data *qd)
375 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
377 if (sdp->sd_vfs->s_flags & MS_RDONLY)
378 return 0;
380 spin_lock(&sdp->sd_quota_spin);
382 if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
383 !test_bit(QDF_CHANGE, &qd->qd_flags)) {
384 spin_unlock(&sdp->sd_quota_spin);
385 return 0;
388 list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
390 set_bit(QDF_LOCKED, &qd->qd_flags);
391 gfs2_assert_warn(sdp, qd->qd_count);
392 qd->qd_count++;
393 qd->qd_change_sync = qd->qd_change;
394 gfs2_assert_warn(sdp, qd->qd_slot_count);
395 qd->qd_slot_count++;
397 spin_unlock(&sdp->sd_quota_spin);
399 gfs2_assert_warn(sdp, qd->qd_change_sync);
400 if (bh_get(qd)) {
401 clear_bit(QDF_LOCKED, &qd->qd_flags);
402 slot_put(qd);
403 qd_put(qd);
404 return 0;
407 return 1;
410 static void qd_unlock(struct gfs2_quota_data *qd)
412 gfs2_assert_warn(qd->qd_gl->gl_sbd,
413 test_bit(QDF_LOCKED, &qd->qd_flags));
414 clear_bit(QDF_LOCKED, &qd->qd_flags);
415 bh_put(qd);
416 slot_put(qd);
417 qd_put(qd);
420 static int qdsb_get(struct gfs2_sbd *sdp, int user, u32 id, int create,
421 struct gfs2_quota_data **qdp)
423 int error;
425 error = qd_get(sdp, user, id, create, qdp);
426 if (error)
427 return error;
429 error = slot_get(*qdp);
430 if (error)
431 goto fail;
433 error = bh_get(*qdp);
434 if (error)
435 goto fail_slot;
437 return 0;
439 fail_slot:
440 slot_put(*qdp);
441 fail:
442 qd_put(*qdp);
443 return error;
446 static void qdsb_put(struct gfs2_quota_data *qd)
448 bh_put(qd);
449 slot_put(qd);
450 qd_put(qd);
453 int gfs2_quota_hold(struct gfs2_inode *ip, u32 uid, u32 gid)
455 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
456 struct gfs2_alloc *al = &ip->i_alloc;
457 struct gfs2_quota_data **qd = al->al_qd;
458 int error;
460 if (gfs2_assert_warn(sdp, !al->al_qd_num) ||
461 gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags)))
462 return -EIO;
464 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
465 return 0;
467 error = qdsb_get(sdp, QUOTA_USER, ip->i_inode.i_uid, CREATE, qd);
468 if (error)
469 goto out;
470 al->al_qd_num++;
471 qd++;
473 error = qdsb_get(sdp, QUOTA_GROUP, ip->i_inode.i_gid, CREATE, qd);
474 if (error)
475 goto out;
476 al->al_qd_num++;
477 qd++;
479 if (uid != NO_QUOTA_CHANGE && uid != ip->i_inode.i_uid) {
480 error = qdsb_get(sdp, QUOTA_USER, uid, CREATE, qd);
481 if (error)
482 goto out;
483 al->al_qd_num++;
484 qd++;
487 if (gid != NO_QUOTA_CHANGE && gid != ip->i_inode.i_gid) {
488 error = qdsb_get(sdp, QUOTA_GROUP, gid, CREATE, qd);
489 if (error)
490 goto out;
491 al->al_qd_num++;
492 qd++;
495 out:
496 if (error)
497 gfs2_quota_unhold(ip);
498 return error;
501 void gfs2_quota_unhold(struct gfs2_inode *ip)
503 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
504 struct gfs2_alloc *al = &ip->i_alloc;
505 unsigned int x;
507 gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags));
509 for (x = 0; x < al->al_qd_num; x++) {
510 qdsb_put(al->al_qd[x]);
511 al->al_qd[x] = NULL;
513 al->al_qd_num = 0;
516 static int sort_qd(const void *a, const void *b)
518 const struct gfs2_quota_data *qd_a = *(const struct gfs2_quota_data **)a;
519 const struct gfs2_quota_data *qd_b = *(const struct gfs2_quota_data **)b;
521 if (!test_bit(QDF_USER, &qd_a->qd_flags) !=
522 !test_bit(QDF_USER, &qd_b->qd_flags)) {
523 if (test_bit(QDF_USER, &qd_a->qd_flags))
524 return -1;
525 else
526 return 1;
528 if (qd_a->qd_id < qd_b->qd_id)
529 return -1;
530 if (qd_a->qd_id > qd_b->qd_id)
531 return 1;
533 return 0;
536 static void do_qc(struct gfs2_quota_data *qd, s64 change)
538 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
539 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
540 struct gfs2_quota_change *qc = qd->qd_bh_qc;
541 s64 x;
543 mutex_lock(&sdp->sd_quota_mutex);
544 gfs2_trans_add_bh(ip->i_gl, qd->qd_bh, 1);
546 if (!test_bit(QDF_CHANGE, &qd->qd_flags)) {
547 qc->qc_change = 0;
548 qc->qc_flags = 0;
549 if (test_bit(QDF_USER, &qd->qd_flags))
550 qc->qc_flags = cpu_to_be32(GFS2_QCF_USER);
551 qc->qc_id = cpu_to_be32(qd->qd_id);
554 x = be64_to_cpu(qc->qc_change) + change;
555 qc->qc_change = cpu_to_be64(x);
557 spin_lock(&sdp->sd_quota_spin);
558 qd->qd_change = x;
559 spin_unlock(&sdp->sd_quota_spin);
561 if (!x) {
562 gfs2_assert_warn(sdp, test_bit(QDF_CHANGE, &qd->qd_flags));
563 clear_bit(QDF_CHANGE, &qd->qd_flags);
564 qc->qc_flags = 0;
565 qc->qc_id = 0;
566 slot_put(qd);
567 qd_put(qd);
568 } else if (!test_and_set_bit(QDF_CHANGE, &qd->qd_flags)) {
569 qd_hold(qd);
570 slot_hold(qd);
573 mutex_unlock(&sdp->sd_quota_mutex);
576 static void gfs2_quota_in(struct gfs2_quota_host *qu, const void *buf)
578 const struct gfs2_quota *str = buf;
580 qu->qu_limit = be64_to_cpu(str->qu_limit);
581 qu->qu_warn = be64_to_cpu(str->qu_warn);
582 qu->qu_value = be64_to_cpu(str->qu_value);
583 qu->qu_ll_next = be32_to_cpu(str->qu_ll_next);
586 static void gfs2_quota_out(const struct gfs2_quota_host *qu, void *buf)
588 struct gfs2_quota *str = buf;
590 str->qu_limit = cpu_to_be64(qu->qu_limit);
591 str->qu_warn = cpu_to_be64(qu->qu_warn);
592 str->qu_value = cpu_to_be64(qu->qu_value);
593 str->qu_ll_next = cpu_to_be32(qu->qu_ll_next);
594 memset(&str->qu_reserved, 0, sizeof(str->qu_reserved));
598 * gfs2_adjust_quota
600 * This function was mostly borrowed from gfs2_block_truncate_page which was
601 * in turn mostly borrowed from ext3
603 static int gfs2_adjust_quota(struct gfs2_inode *ip, loff_t loc,
604 s64 change, struct gfs2_quota_data *qd)
606 struct inode *inode = &ip->i_inode;
607 struct address_space *mapping = inode->i_mapping;
608 unsigned long index = loc >> PAGE_CACHE_SHIFT;
609 unsigned offset = loc & (PAGE_CACHE_SIZE - 1);
610 unsigned blocksize, iblock, pos;
611 struct buffer_head *bh;
612 struct page *page;
613 void *kaddr;
614 char *ptr;
615 struct gfs2_quota_host qp;
616 s64 value;
617 int err = -EIO;
619 if (gfs2_is_stuffed(ip)) {
620 struct gfs2_alloc *al = NULL;
621 al = gfs2_alloc_get(ip);
622 /* just request 1 blk */
623 al->al_requested = 1;
624 gfs2_inplace_reserve(ip);
625 gfs2_unstuff_dinode(ip, NULL);
626 gfs2_inplace_release(ip);
627 gfs2_alloc_put(ip);
629 page = grab_cache_page(mapping, index);
630 if (!page)
631 return -ENOMEM;
633 blocksize = inode->i_sb->s_blocksize;
634 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
636 if (!page_has_buffers(page))
637 create_empty_buffers(page, blocksize, 0);
639 bh = page_buffers(page);
640 pos = blocksize;
641 while (offset >= pos) {
642 bh = bh->b_this_page;
643 iblock++;
644 pos += blocksize;
647 if (!buffer_mapped(bh)) {
648 gfs2_get_block(inode, iblock, bh, 1);
649 if (!buffer_mapped(bh))
650 goto unlock;
653 if (PageUptodate(page))
654 set_buffer_uptodate(bh);
656 if (!buffer_uptodate(bh)) {
657 ll_rw_block(READ_META, 1, &bh);
658 wait_on_buffer(bh);
659 if (!buffer_uptodate(bh))
660 goto unlock;
663 gfs2_trans_add_bh(ip->i_gl, bh, 0);
665 kaddr = kmap_atomic(page, KM_USER0);
666 ptr = kaddr + offset;
667 gfs2_quota_in(&qp, ptr);
668 qp.qu_value += change;
669 value = qp.qu_value;
670 gfs2_quota_out(&qp, ptr);
671 flush_dcache_page(page);
672 kunmap_atomic(kaddr, KM_USER0);
673 err = 0;
674 qd->qd_qb.qb_magic = cpu_to_be32(GFS2_MAGIC);
675 qd->qd_qb.qb_value = cpu_to_be64(value);
676 ((struct gfs2_quota_lvb*)(qd->qd_gl->gl_lvb))->qb_magic = cpu_to_be32(GFS2_MAGIC);
677 ((struct gfs2_quota_lvb*)(qd->qd_gl->gl_lvb))->qb_value = cpu_to_be64(value);
678 unlock:
679 unlock_page(page);
680 page_cache_release(page);
681 return err;
684 static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
686 struct gfs2_sbd *sdp = (*qda)->qd_gl->gl_sbd;
687 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
688 unsigned int data_blocks, ind_blocks;
689 struct gfs2_holder *ghs, i_gh;
690 unsigned int qx, x;
691 struct gfs2_quota_data *qd;
692 loff_t offset;
693 unsigned int nalloc = 0;
694 struct gfs2_alloc *al = NULL;
695 int error;
697 gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
698 &data_blocks, &ind_blocks);
700 ghs = kcalloc(num_qd, sizeof(struct gfs2_holder), GFP_KERNEL);
701 if (!ghs)
702 return -ENOMEM;
704 sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL);
705 for (qx = 0; qx < num_qd; qx++) {
706 error = gfs2_glock_nq_init(qda[qx]->qd_gl,
707 LM_ST_EXCLUSIVE,
708 GL_NOCACHE, &ghs[qx]);
709 if (error)
710 goto out;
713 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
714 if (error)
715 goto out;
717 for (x = 0; x < num_qd; x++) {
718 int alloc_required;
720 offset = qd2offset(qda[x]);
721 error = gfs2_write_alloc_required(ip, offset,
722 sizeof(struct gfs2_quota),
723 &alloc_required);
724 if (error)
725 goto out_gunlock;
726 if (alloc_required)
727 nalloc++;
730 if (nalloc) {
731 al = gfs2_alloc_get(ip);
733 al->al_requested = nalloc * (data_blocks + ind_blocks);
735 error = gfs2_inplace_reserve(ip);
736 if (error)
737 goto out_alloc;
739 error = gfs2_trans_begin(sdp,
740 al->al_rgd->rd_length +
741 num_qd * data_blocks +
742 nalloc * ind_blocks +
743 RES_DINODE + num_qd +
744 RES_STATFS, 0);
745 if (error)
746 goto out_ipres;
747 } else {
748 error = gfs2_trans_begin(sdp,
749 num_qd * data_blocks +
750 RES_DINODE + num_qd, 0);
751 if (error)
752 goto out_gunlock;
755 for (x = 0; x < num_qd; x++) {
756 qd = qda[x];
757 offset = qd2offset(qd);
758 error = gfs2_adjust_quota(ip, offset, qd->qd_change_sync,
759 (struct gfs2_quota_data *)
760 qd);
761 if (error)
762 goto out_end_trans;
764 do_qc(qd, -qd->qd_change_sync);
767 error = 0;
769 out_end_trans:
770 gfs2_trans_end(sdp);
771 out_ipres:
772 if (nalloc)
773 gfs2_inplace_release(ip);
774 out_alloc:
775 if (nalloc)
776 gfs2_alloc_put(ip);
777 out_gunlock:
778 gfs2_glock_dq_uninit(&i_gh);
779 out:
780 while (qx--)
781 gfs2_glock_dq_uninit(&ghs[qx]);
782 kfree(ghs);
783 gfs2_log_flush(ip->i_gl->gl_sbd, ip->i_gl);
784 return error;
787 static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
788 struct gfs2_holder *q_gh)
790 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
791 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
792 struct gfs2_holder i_gh;
793 struct gfs2_quota_host q;
794 char buf[sizeof(struct gfs2_quota)];
795 int error;
796 struct gfs2_quota_lvb *qlvb;
798 restart:
799 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
800 if (error)
801 return error;
803 qd->qd_qb = *(struct gfs2_quota_lvb *)qd->qd_gl->gl_lvb;
805 if (force_refresh || qd->qd_qb.qb_magic != cpu_to_be32(GFS2_MAGIC)) {
806 loff_t pos;
807 gfs2_glock_dq_uninit(q_gh);
808 error = gfs2_glock_nq_init(qd->qd_gl,
809 LM_ST_EXCLUSIVE, GL_NOCACHE,
810 q_gh);
811 if (error)
812 return error;
814 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
815 if (error)
816 goto fail;
818 memset(buf, 0, sizeof(struct gfs2_quota));
819 pos = qd2offset(qd);
820 error = gfs2_internal_read(ip, NULL, buf, &pos,
821 sizeof(struct gfs2_quota));
822 if (error < 0)
823 goto fail_gunlock;
825 gfs2_glock_dq_uninit(&i_gh);
828 gfs2_quota_in(&q, buf);
829 qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lvb;
830 qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC);
831 qlvb->__pad = 0;
832 qlvb->qb_limit = cpu_to_be64(q.qu_limit);
833 qlvb->qb_warn = cpu_to_be64(q.qu_warn);
834 qlvb->qb_value = cpu_to_be64(q.qu_value);
835 qd->qd_qb = *qlvb;
837 if (gfs2_glock_is_blocking(qd->qd_gl)) {
838 gfs2_glock_dq_uninit(q_gh);
839 force_refresh = 0;
840 goto restart;
844 return 0;
846 fail_gunlock:
847 gfs2_glock_dq_uninit(&i_gh);
848 fail:
849 gfs2_glock_dq_uninit(q_gh);
850 return error;
853 int gfs2_quota_lock(struct gfs2_inode *ip, u32 uid, u32 gid)
855 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
856 struct gfs2_alloc *al = &ip->i_alloc;
857 unsigned int x;
858 int error = 0;
860 gfs2_quota_hold(ip, uid, gid);
862 if (capable(CAP_SYS_RESOURCE) ||
863 sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
864 return 0;
866 sort(al->al_qd, al->al_qd_num, sizeof(struct gfs2_quota_data *),
867 sort_qd, NULL);
869 for (x = 0; x < al->al_qd_num; x++) {
870 error = do_glock(al->al_qd[x], NO_FORCE, &al->al_qd_ghs[x]);
871 if (error)
872 break;
875 if (!error)
876 set_bit(GIF_QD_LOCKED, &ip->i_flags);
877 else {
878 while (x--)
879 gfs2_glock_dq_uninit(&al->al_qd_ghs[x]);
880 gfs2_quota_unhold(ip);
883 return error;
886 static int need_sync(struct gfs2_quota_data *qd)
888 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
889 struct gfs2_tune *gt = &sdp->sd_tune;
890 s64 value;
891 unsigned int num, den;
892 int do_sync = 1;
894 if (!qd->qd_qb.qb_limit)
895 return 0;
897 spin_lock(&sdp->sd_quota_spin);
898 value = qd->qd_change;
899 spin_unlock(&sdp->sd_quota_spin);
901 spin_lock(&gt->gt_spin);
902 num = gt->gt_quota_scale_num;
903 den = gt->gt_quota_scale_den;
904 spin_unlock(&gt->gt_spin);
906 if (value < 0)
907 do_sync = 0;
908 else if ((s64)be64_to_cpu(qd->qd_qb.qb_value) >=
909 (s64)be64_to_cpu(qd->qd_qb.qb_limit))
910 do_sync = 0;
911 else {
912 value *= gfs2_jindex_size(sdp) * num;
913 do_div(value, den);
914 value += (s64)be64_to_cpu(qd->qd_qb.qb_value);
915 if (value < (s64)be64_to_cpu(qd->qd_qb.qb_limit))
916 do_sync = 0;
919 return do_sync;
922 void gfs2_quota_unlock(struct gfs2_inode *ip)
924 struct gfs2_alloc *al = &ip->i_alloc;
925 struct gfs2_quota_data *qda[4];
926 unsigned int count = 0;
927 unsigned int x;
929 if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
930 goto out;
932 for (x = 0; x < al->al_qd_num; x++) {
933 struct gfs2_quota_data *qd;
934 int sync;
936 qd = al->al_qd[x];
937 sync = need_sync(qd);
939 gfs2_glock_dq_uninit(&al->al_qd_ghs[x]);
941 if (sync && qd_trylock(qd))
942 qda[count++] = qd;
945 if (count) {
946 do_sync(count, qda);
947 for (x = 0; x < count; x++)
948 qd_unlock(qda[x]);
951 out:
952 gfs2_quota_unhold(ip);
955 #define MAX_LINE 256
957 static int print_message(struct gfs2_quota_data *qd, char *type)
959 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
961 printk(KERN_INFO "GFS2: fsid=%s: quota %s for %s %u\r\n",
962 sdp->sd_fsname, type,
963 (test_bit(QDF_USER, &qd->qd_flags)) ? "user" : "group",
964 qd->qd_id);
966 return 0;
969 int gfs2_quota_check(struct gfs2_inode *ip, u32 uid, u32 gid)
971 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
972 struct gfs2_alloc *al = &ip->i_alloc;
973 struct gfs2_quota_data *qd;
974 s64 value;
975 unsigned int x;
976 int error = 0;
978 if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
979 return 0;
981 if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
982 return 0;
984 for (x = 0; x < al->al_qd_num; x++) {
985 qd = al->al_qd[x];
987 if (!((qd->qd_id == uid && test_bit(QDF_USER, &qd->qd_flags)) ||
988 (qd->qd_id == gid && !test_bit(QDF_USER, &qd->qd_flags))))
989 continue;
991 value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
992 spin_lock(&sdp->sd_quota_spin);
993 value += qd->qd_change;
994 spin_unlock(&sdp->sd_quota_spin);
996 if (be64_to_cpu(qd->qd_qb.qb_limit) && (s64)be64_to_cpu(qd->qd_qb.qb_limit) < value) {
997 print_message(qd, "exceeded");
998 error = -EDQUOT;
999 break;
1000 } else if (be64_to_cpu(qd->qd_qb.qb_warn) &&
1001 (s64)be64_to_cpu(qd->qd_qb.qb_warn) < value &&
1002 time_after_eq(jiffies, qd->qd_last_warn +
1003 gfs2_tune_get(sdp,
1004 gt_quota_warn_period) * HZ)) {
1005 error = print_message(qd, "warning");
1006 qd->qd_last_warn = jiffies;
1010 return error;
1013 void gfs2_quota_change(struct gfs2_inode *ip, s64 change,
1014 u32 uid, u32 gid)
1016 struct gfs2_alloc *al = &ip->i_alloc;
1017 struct gfs2_quota_data *qd;
1018 unsigned int x;
1019 unsigned int found = 0;
1021 if (gfs2_assert_warn(GFS2_SB(&ip->i_inode), change))
1022 return;
1023 if (ip->i_di.di_flags & GFS2_DIF_SYSTEM)
1024 return;
1026 for (x = 0; x < al->al_qd_num; x++) {
1027 qd = al->al_qd[x];
1029 if ((qd->qd_id == uid && test_bit(QDF_USER, &qd->qd_flags)) ||
1030 (qd->qd_id == gid && !test_bit(QDF_USER, &qd->qd_flags))) {
1031 do_qc(qd, change);
1032 found++;
1037 int gfs2_quota_sync(struct gfs2_sbd *sdp)
1039 struct gfs2_quota_data **qda;
1040 unsigned int max_qd = gfs2_tune_get(sdp, gt_quota_simul_sync);
1041 unsigned int num_qd;
1042 unsigned int x;
1043 int error = 0;
1045 sdp->sd_quota_sync_gen++;
1047 qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
1048 if (!qda)
1049 return -ENOMEM;
1051 do {
1052 num_qd = 0;
1054 for (;;) {
1055 error = qd_fish(sdp, qda + num_qd);
1056 if (error || !qda[num_qd])
1057 break;
1058 if (++num_qd == max_qd)
1059 break;
1062 if (num_qd) {
1063 if (!error)
1064 error = do_sync(num_qd, qda);
1065 if (!error)
1066 for (x = 0; x < num_qd; x++)
1067 qda[x]->qd_sync_gen =
1068 sdp->sd_quota_sync_gen;
1070 for (x = 0; x < num_qd; x++)
1071 qd_unlock(qda[x]);
1073 } while (!error && num_qd == max_qd);
1075 kfree(qda);
1077 return error;
1080 int gfs2_quota_refresh(struct gfs2_sbd *sdp, int user, u32 id)
1082 struct gfs2_quota_data *qd;
1083 struct gfs2_holder q_gh;
1084 int error;
1086 error = qd_get(sdp, user, id, CREATE, &qd);
1087 if (error)
1088 return error;
1090 error = do_glock(qd, FORCE, &q_gh);
1091 if (!error)
1092 gfs2_glock_dq_uninit(&q_gh);
1094 qd_put(qd);
1096 return error;
1099 static void gfs2_quota_change_in(struct gfs2_quota_change_host *qc, const void *buf)
1101 const struct gfs2_quota_change *str = buf;
1103 qc->qc_change = be64_to_cpu(str->qc_change);
1104 qc->qc_flags = be32_to_cpu(str->qc_flags);
1105 qc->qc_id = be32_to_cpu(str->qc_id);
1108 int gfs2_quota_init(struct gfs2_sbd *sdp)
1110 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
1111 unsigned int blocks = ip->i_di.di_size >> sdp->sd_sb.sb_bsize_shift;
1112 unsigned int x, slot = 0;
1113 unsigned int found = 0;
1114 u64 dblock;
1115 u32 extlen = 0;
1116 int error;
1118 if (!ip->i_di.di_size || ip->i_di.di_size > (64 << 20) ||
1119 ip->i_di.di_size & (sdp->sd_sb.sb_bsize - 1)) {
1120 gfs2_consist_inode(ip);
1121 return -EIO;
1123 sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
1124 sdp->sd_quota_chunks = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * PAGE_SIZE);
1126 error = -ENOMEM;
1128 sdp->sd_quota_bitmap = kcalloc(sdp->sd_quota_chunks,
1129 sizeof(unsigned char *), GFP_KERNEL);
1130 if (!sdp->sd_quota_bitmap)
1131 return error;
1133 for (x = 0; x < sdp->sd_quota_chunks; x++) {
1134 sdp->sd_quota_bitmap[x] = kzalloc(PAGE_SIZE, GFP_KERNEL);
1135 if (!sdp->sd_quota_bitmap[x])
1136 goto fail;
1139 for (x = 0; x < blocks; x++) {
1140 struct buffer_head *bh;
1141 unsigned int y;
1143 if (!extlen) {
1144 int new = 0;
1145 error = gfs2_extent_map(&ip->i_inode, x, &new, &dblock, &extlen);
1146 if (error)
1147 goto fail;
1149 error = -EIO;
1150 bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
1151 if (!bh)
1152 goto fail;
1153 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
1154 brelse(bh);
1155 goto fail;
1158 for (y = 0; y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1159 y++, slot++) {
1160 struct gfs2_quota_change_host qc;
1161 struct gfs2_quota_data *qd;
1163 gfs2_quota_change_in(&qc, bh->b_data +
1164 sizeof(struct gfs2_meta_header) +
1165 y * sizeof(struct gfs2_quota_change));
1166 if (!qc.qc_change)
1167 continue;
1169 error = qd_alloc(sdp, (qc.qc_flags & GFS2_QCF_USER),
1170 qc.qc_id, &qd);
1171 if (error) {
1172 brelse(bh);
1173 goto fail;
1176 set_bit(QDF_CHANGE, &qd->qd_flags);
1177 qd->qd_change = qc.qc_change;
1178 qd->qd_slot = slot;
1179 qd->qd_slot_count = 1;
1180 qd->qd_last_touched = jiffies;
1182 spin_lock(&sdp->sd_quota_spin);
1183 gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, slot, 1);
1184 list_add(&qd->qd_list, &sdp->sd_quota_list);
1185 atomic_inc(&sdp->sd_quota_count);
1186 spin_unlock(&sdp->sd_quota_spin);
1188 found++;
1191 brelse(bh);
1192 dblock++;
1193 extlen--;
1196 if (found)
1197 fs_info(sdp, "found %u quota changes\n", found);
1199 return 0;
1201 fail:
1202 gfs2_quota_cleanup(sdp);
1203 return error;
1206 void gfs2_quota_scan(struct gfs2_sbd *sdp)
1208 struct gfs2_quota_data *qd, *safe;
1209 LIST_HEAD(dead);
1211 spin_lock(&sdp->sd_quota_spin);
1212 list_for_each_entry_safe(qd, safe, &sdp->sd_quota_list, qd_list) {
1213 if (!qd->qd_count &&
1214 time_after_eq(jiffies, qd->qd_last_touched +
1215 gfs2_tune_get(sdp, gt_quota_cache_secs) * HZ)) {
1216 list_move(&qd->qd_list, &dead);
1217 gfs2_assert_warn(sdp,
1218 atomic_read(&sdp->sd_quota_count) > 0);
1219 atomic_dec(&sdp->sd_quota_count);
1222 spin_unlock(&sdp->sd_quota_spin);
1224 while (!list_empty(&dead)) {
1225 qd = list_entry(dead.next, struct gfs2_quota_data, qd_list);
1226 list_del(&qd->qd_list);
1228 gfs2_assert_warn(sdp, !qd->qd_change);
1229 gfs2_assert_warn(sdp, !qd->qd_slot_count);
1230 gfs2_assert_warn(sdp, !qd->qd_bh_count);
1232 gfs2_lvb_unhold(qd->qd_gl);
1233 kfree(qd);
1237 void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1239 struct list_head *head = &sdp->sd_quota_list;
1240 struct gfs2_quota_data *qd;
1241 unsigned int x;
1243 spin_lock(&sdp->sd_quota_spin);
1244 while (!list_empty(head)) {
1245 qd = list_entry(head->prev, struct gfs2_quota_data, qd_list);
1247 if (qd->qd_count > 1 ||
1248 (qd->qd_count && !test_bit(QDF_CHANGE, &qd->qd_flags))) {
1249 list_move(&qd->qd_list, head);
1250 spin_unlock(&sdp->sd_quota_spin);
1251 schedule();
1252 spin_lock(&sdp->sd_quota_spin);
1253 continue;
1256 list_del(&qd->qd_list);
1257 atomic_dec(&sdp->sd_quota_count);
1258 spin_unlock(&sdp->sd_quota_spin);
1260 if (!qd->qd_count) {
1261 gfs2_assert_warn(sdp, !qd->qd_change);
1262 gfs2_assert_warn(sdp, !qd->qd_slot_count);
1263 } else
1264 gfs2_assert_warn(sdp, qd->qd_slot_count == 1);
1265 gfs2_assert_warn(sdp, !qd->qd_bh_count);
1267 gfs2_lvb_unhold(qd->qd_gl);
1268 kfree(qd);
1270 spin_lock(&sdp->sd_quota_spin);
1272 spin_unlock(&sdp->sd_quota_spin);
1274 gfs2_assert_warn(sdp, !atomic_read(&sdp->sd_quota_count));
1276 if (sdp->sd_quota_bitmap) {
1277 for (x = 0; x < sdp->sd_quota_chunks; x++)
1278 kfree(sdp->sd_quota_bitmap[x]);
1279 kfree(sdp->sd_quota_bitmap);