2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2007 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.
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>
46 #include <linux/bio.h>
47 #include <linux/gfs2_ondisk.h>
48 #include <linux/kthread.h>
49 #include <linux/freezer.h>
63 #include "ops_address.h"
69 struct gfs2_quota_host {
76 struct gfs2_quota_change_host {
78 u32 qc_flags; /* GFS2_QCF_... */
82 static LIST_HEAD(qd_lru_list);
83 static atomic_t qd_lru_count = ATOMIC_INIT(0);
84 static DEFINE_SPINLOCK(qd_lru_lock);
86 int gfs2_shrink_qd_memory(int nr, gfp_t gfp_mask)
88 struct gfs2_quota_data *qd;
94 if (!(gfp_mask & __GFP_FS))
97 spin_lock(&qd_lru_lock);
98 while (nr && !list_empty(&qd_lru_list)) {
99 qd = list_entry(qd_lru_list.next,
100 struct gfs2_quota_data, qd_reclaim);
101 sdp = qd->qd_gl->gl_sbd;
103 /* Free from the filesystem-specific list */
104 list_del(&qd->qd_list);
106 gfs2_assert_warn(sdp, !qd->qd_change);
107 gfs2_assert_warn(sdp, !qd->qd_slot_count);
108 gfs2_assert_warn(sdp, !qd->qd_bh_count);
110 gfs2_glock_put(qd->qd_gl);
111 atomic_dec(&sdp->sd_quota_count);
113 /* Delete it from the common reclaim list */
114 list_del_init(&qd->qd_reclaim);
115 atomic_dec(&qd_lru_count);
116 spin_unlock(&qd_lru_lock);
117 kmem_cache_free(gfs2_quotad_cachep, qd);
118 spin_lock(&qd_lru_lock);
121 spin_unlock(&qd_lru_lock);
124 return (atomic_read(&qd_lru_count) * sysctl_vfs_cache_pressure) / 100;
127 static u64 qd2offset(struct gfs2_quota_data *qd)
131 offset = 2 * (u64)qd->qd_id + !test_bit(QDF_USER, &qd->qd_flags);
132 offset *= sizeof(struct gfs2_quota);
137 static int qd_alloc(struct gfs2_sbd *sdp, int user, u32 id,
138 struct gfs2_quota_data **qdp)
140 struct gfs2_quota_data *qd;
143 qd = kmem_cache_zalloc(gfs2_quotad_cachep, GFP_NOFS);
147 atomic_set(&qd->qd_count, 1);
150 set_bit(QDF_USER, &qd->qd_flags);
152 INIT_LIST_HEAD(&qd->qd_reclaim);
154 error = gfs2_glock_get(sdp, 2 * (u64)id + !user,
155 &gfs2_quota_glops, CREATE, &qd->qd_gl);
164 kmem_cache_free(gfs2_quotad_cachep, qd);
168 static int qd_get(struct gfs2_sbd *sdp, int user, u32 id, int create,
169 struct gfs2_quota_data **qdp)
171 struct gfs2_quota_data *qd = NULL, *new_qd = NULL;
178 spin_lock(&qd_lru_lock);
179 list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
180 if (qd->qd_id == id &&
181 !test_bit(QDF_USER, &qd->qd_flags) == !user) {
182 if (!atomic_read(&qd->qd_count) &&
183 !list_empty(&qd->qd_reclaim)) {
184 /* Remove it from reclaim list */
185 list_del_init(&qd->qd_reclaim);
186 atomic_dec(&qd_lru_count);
188 atomic_inc(&qd->qd_count);
199 list_add(&qd->qd_list, &sdp->sd_quota_list);
200 atomic_inc(&sdp->sd_quota_count);
204 spin_unlock(&qd_lru_lock);
208 gfs2_glock_put(new_qd->qd_gl);
209 kmem_cache_free(gfs2_quotad_cachep, new_qd);
215 error = qd_alloc(sdp, user, id, &new_qd);
221 static void qd_hold(struct gfs2_quota_data *qd)
223 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
224 gfs2_assert(sdp, atomic_read(&qd->qd_count));
225 atomic_inc(&qd->qd_count);
228 static void qd_put(struct gfs2_quota_data *qd)
230 if (atomic_dec_and_lock(&qd->qd_count, &qd_lru_lock)) {
231 /* Add to the reclaim list */
232 list_add_tail(&qd->qd_reclaim, &qd_lru_list);
233 atomic_inc(&qd_lru_count);
234 spin_unlock(&qd_lru_lock);
238 static int slot_get(struct gfs2_quota_data *qd)
240 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
241 unsigned int c, o = 0, b;
242 unsigned char byte = 0;
244 spin_lock(&qd_lru_lock);
246 if (qd->qd_slot_count++) {
247 spin_unlock(&qd_lru_lock);
251 for (c = 0; c < sdp->sd_quota_chunks; c++)
252 for (o = 0; o < PAGE_SIZE; o++) {
253 byte = sdp->sd_quota_bitmap[c][o];
261 for (b = 0; b < 8; b++)
262 if (!(byte & (1 << b)))
264 qd->qd_slot = c * (8 * PAGE_SIZE) + o * 8 + b;
266 if (qd->qd_slot >= sdp->sd_quota_slots)
269 sdp->sd_quota_bitmap[c][o] |= 1 << b;
271 spin_unlock(&qd_lru_lock);
277 spin_unlock(&qd_lru_lock);
281 static void slot_hold(struct gfs2_quota_data *qd)
283 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
285 spin_lock(&qd_lru_lock);
286 gfs2_assert(sdp, qd->qd_slot_count);
288 spin_unlock(&qd_lru_lock);
291 static void slot_put(struct gfs2_quota_data *qd)
293 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
295 spin_lock(&qd_lru_lock);
296 gfs2_assert(sdp, qd->qd_slot_count);
297 if (!--qd->qd_slot_count) {
298 gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, qd->qd_slot, 0);
301 spin_unlock(&qd_lru_lock);
304 static int bh_get(struct gfs2_quota_data *qd)
306 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
307 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
308 unsigned int block, offset;
309 struct buffer_head *bh;
311 struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
313 mutex_lock(&sdp->sd_quota_mutex);
315 if (qd->qd_bh_count++) {
316 mutex_unlock(&sdp->sd_quota_mutex);
320 block = qd->qd_slot / sdp->sd_qc_per_block;
321 offset = qd->qd_slot % sdp->sd_qc_per_block;
323 bh_map.b_size = 1 << ip->i_inode.i_blkbits;
324 error = gfs2_block_map(&ip->i_inode, block, &bh_map, 0);
327 error = gfs2_meta_read(ip->i_gl, bh_map.b_blocknr, DIO_WAIT, &bh);
331 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC))
335 qd->qd_bh_qc = (struct gfs2_quota_change *)
336 (bh->b_data + sizeof(struct gfs2_meta_header) +
337 offset * sizeof(struct gfs2_quota_change));
339 mutex_unlock(&sdp->sd_quota_mutex);
347 mutex_unlock(&sdp->sd_quota_mutex);
351 static void bh_put(struct gfs2_quota_data *qd)
353 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
355 mutex_lock(&sdp->sd_quota_mutex);
356 gfs2_assert(sdp, qd->qd_bh_count);
357 if (!--qd->qd_bh_count) {
362 mutex_unlock(&sdp->sd_quota_mutex);
365 static int qd_fish(struct gfs2_sbd *sdp, struct gfs2_quota_data **qdp)
367 struct gfs2_quota_data *qd = NULL;
373 if (sdp->sd_vfs->s_flags & MS_RDONLY)
376 spin_lock(&qd_lru_lock);
378 list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
379 if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
380 !test_bit(QDF_CHANGE, &qd->qd_flags) ||
381 qd->qd_sync_gen >= sdp->sd_quota_sync_gen)
384 list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
386 set_bit(QDF_LOCKED, &qd->qd_flags);
387 gfs2_assert_warn(sdp, atomic_read(&qd->qd_count));
388 atomic_inc(&qd->qd_count);
389 qd->qd_change_sync = qd->qd_change;
390 gfs2_assert_warn(sdp, qd->qd_slot_count);
400 spin_unlock(&qd_lru_lock);
403 gfs2_assert_warn(sdp, qd->qd_change_sync);
406 clear_bit(QDF_LOCKED, &qd->qd_flags);
418 static int qd_trylock(struct gfs2_quota_data *qd)
420 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
422 if (sdp->sd_vfs->s_flags & MS_RDONLY)
425 spin_lock(&qd_lru_lock);
427 if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
428 !test_bit(QDF_CHANGE, &qd->qd_flags)) {
429 spin_unlock(&qd_lru_lock);
433 list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
435 set_bit(QDF_LOCKED, &qd->qd_flags);
436 gfs2_assert_warn(sdp, atomic_read(&qd->qd_count));
437 atomic_inc(&qd->qd_count);
438 qd->qd_change_sync = qd->qd_change;
439 gfs2_assert_warn(sdp, qd->qd_slot_count);
442 spin_unlock(&qd_lru_lock);
444 gfs2_assert_warn(sdp, qd->qd_change_sync);
446 clear_bit(QDF_LOCKED, &qd->qd_flags);
455 static void qd_unlock(struct gfs2_quota_data *qd)
457 gfs2_assert_warn(qd->qd_gl->gl_sbd,
458 test_bit(QDF_LOCKED, &qd->qd_flags));
459 clear_bit(QDF_LOCKED, &qd->qd_flags);
465 static int qdsb_get(struct gfs2_sbd *sdp, int user, u32 id, int create,
466 struct gfs2_quota_data **qdp)
470 error = qd_get(sdp, user, id, create, qdp);
474 error = slot_get(*qdp);
478 error = bh_get(*qdp);
491 static void qdsb_put(struct gfs2_quota_data *qd)
498 int gfs2_quota_hold(struct gfs2_inode *ip, u32 uid, u32 gid)
500 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
501 struct gfs2_alloc *al = ip->i_alloc;
502 struct gfs2_quota_data **qd = al->al_qd;
505 if (gfs2_assert_warn(sdp, !al->al_qd_num) ||
506 gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags)))
509 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
512 error = qdsb_get(sdp, QUOTA_USER, ip->i_inode.i_uid, CREATE, qd);
518 error = qdsb_get(sdp, QUOTA_GROUP, ip->i_inode.i_gid, CREATE, qd);
524 if (uid != NO_QUOTA_CHANGE && uid != ip->i_inode.i_uid) {
525 error = qdsb_get(sdp, QUOTA_USER, uid, CREATE, qd);
532 if (gid != NO_QUOTA_CHANGE && gid != ip->i_inode.i_gid) {
533 error = qdsb_get(sdp, QUOTA_GROUP, gid, CREATE, qd);
542 gfs2_quota_unhold(ip);
546 void gfs2_quota_unhold(struct gfs2_inode *ip)
548 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
549 struct gfs2_alloc *al = ip->i_alloc;
552 gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags));
554 for (x = 0; x < al->al_qd_num; x++) {
555 qdsb_put(al->al_qd[x]);
561 static int sort_qd(const void *a, const void *b)
563 const struct gfs2_quota_data *qd_a = *(const struct gfs2_quota_data **)a;
564 const struct gfs2_quota_data *qd_b = *(const struct gfs2_quota_data **)b;
566 if (!test_bit(QDF_USER, &qd_a->qd_flags) !=
567 !test_bit(QDF_USER, &qd_b->qd_flags)) {
568 if (test_bit(QDF_USER, &qd_a->qd_flags))
573 if (qd_a->qd_id < qd_b->qd_id)
575 if (qd_a->qd_id > qd_b->qd_id)
581 static void do_qc(struct gfs2_quota_data *qd, s64 change)
583 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
584 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
585 struct gfs2_quota_change *qc = qd->qd_bh_qc;
588 mutex_lock(&sdp->sd_quota_mutex);
589 gfs2_trans_add_bh(ip->i_gl, qd->qd_bh, 1);
591 if (!test_bit(QDF_CHANGE, &qd->qd_flags)) {
594 if (test_bit(QDF_USER, &qd->qd_flags))
595 qc->qc_flags = cpu_to_be32(GFS2_QCF_USER);
596 qc->qc_id = cpu_to_be32(qd->qd_id);
599 x = be64_to_cpu(qc->qc_change) + change;
600 qc->qc_change = cpu_to_be64(x);
602 spin_lock(&qd_lru_lock);
604 spin_unlock(&qd_lru_lock);
607 gfs2_assert_warn(sdp, test_bit(QDF_CHANGE, &qd->qd_flags));
608 clear_bit(QDF_CHANGE, &qd->qd_flags);
613 } else if (!test_and_set_bit(QDF_CHANGE, &qd->qd_flags)) {
618 mutex_unlock(&sdp->sd_quota_mutex);
621 static void gfs2_quota_in(struct gfs2_quota_host *qu, const void *buf)
623 const struct gfs2_quota *str = buf;
625 qu->qu_limit = be64_to_cpu(str->qu_limit);
626 qu->qu_warn = be64_to_cpu(str->qu_warn);
627 qu->qu_value = be64_to_cpu(str->qu_value);
628 qu->qu_ll_next = be32_to_cpu(str->qu_ll_next);
631 static void gfs2_quota_out(const struct gfs2_quota_host *qu, void *buf)
633 struct gfs2_quota *str = buf;
635 str->qu_limit = cpu_to_be64(qu->qu_limit);
636 str->qu_warn = cpu_to_be64(qu->qu_warn);
637 str->qu_value = cpu_to_be64(qu->qu_value);
638 str->qu_ll_next = cpu_to_be32(qu->qu_ll_next);
639 memset(&str->qu_reserved, 0, sizeof(str->qu_reserved));
645 * This function was mostly borrowed from gfs2_block_truncate_page which was
646 * in turn mostly borrowed from ext3
648 static int gfs2_adjust_quota(struct gfs2_inode *ip, loff_t loc,
649 s64 change, struct gfs2_quota_data *qd)
651 struct inode *inode = &ip->i_inode;
652 struct address_space *mapping = inode->i_mapping;
653 unsigned long index = loc >> PAGE_CACHE_SHIFT;
654 unsigned offset = loc & (PAGE_CACHE_SIZE - 1);
655 unsigned blocksize, iblock, pos;
656 struct buffer_head *bh;
660 struct gfs2_quota_host qp;
664 if (gfs2_is_stuffed(ip))
665 gfs2_unstuff_dinode(ip, NULL);
667 page = grab_cache_page(mapping, index);
671 blocksize = inode->i_sb->s_blocksize;
672 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
674 if (!page_has_buffers(page))
675 create_empty_buffers(page, blocksize, 0);
677 bh = page_buffers(page);
679 while (offset >= pos) {
680 bh = bh->b_this_page;
685 if (!buffer_mapped(bh)) {
686 gfs2_block_map(inode, iblock, bh, 1);
687 if (!buffer_mapped(bh))
691 if (PageUptodate(page))
692 set_buffer_uptodate(bh);
694 if (!buffer_uptodate(bh)) {
695 ll_rw_block(READ_META, 1, &bh);
697 if (!buffer_uptodate(bh))
701 gfs2_trans_add_bh(ip->i_gl, bh, 0);
703 kaddr = kmap_atomic(page, KM_USER0);
704 ptr = kaddr + offset;
705 gfs2_quota_in(&qp, ptr);
706 qp.qu_value += change;
708 gfs2_quota_out(&qp, ptr);
709 flush_dcache_page(page);
710 kunmap_atomic(kaddr, KM_USER0);
712 qd->qd_qb.qb_magic = cpu_to_be32(GFS2_MAGIC);
713 qd->qd_qb.qb_value = cpu_to_be64(value);
714 ((struct gfs2_quota_lvb*)(qd->qd_gl->gl_lvb))->qb_magic = cpu_to_be32(GFS2_MAGIC);
715 ((struct gfs2_quota_lvb*)(qd->qd_gl->gl_lvb))->qb_value = cpu_to_be64(value);
718 page_cache_release(page);
722 static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
724 struct gfs2_sbd *sdp = (*qda)->qd_gl->gl_sbd;
725 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
726 unsigned int data_blocks, ind_blocks;
727 struct gfs2_holder *ghs, i_gh;
729 struct gfs2_quota_data *qd;
731 unsigned int nalloc = 0, blocks;
732 struct gfs2_alloc *al = NULL;
735 gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
736 &data_blocks, &ind_blocks);
738 ghs = kcalloc(num_qd, sizeof(struct gfs2_holder), GFP_NOFS);
742 sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL);
743 for (qx = 0; qx < num_qd; qx++) {
744 error = gfs2_glock_nq_init(qda[qx]->qd_gl,
746 GL_NOCACHE, &ghs[qx]);
751 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
755 for (x = 0; x < num_qd; x++) {
758 offset = qd2offset(qda[x]);
759 error = gfs2_write_alloc_required(ip, offset,
760 sizeof(struct gfs2_quota),
768 al = gfs2_alloc_get(ip);
774 * 1 blk for unstuffing inode if stuffed. We add this extra
775 * block to the reservation unconditionally. If the inode
776 * doesn't need unstuffing, the block will be released to the
777 * rgrp since it won't be allocated during the transaction
779 al->al_requested = 1;
780 /* +1 in the end for block requested above for unstuffing */
781 blocks = num_qd * data_blocks + RES_DINODE + num_qd + 1;
784 al->al_requested += nalloc * (data_blocks + ind_blocks);
785 error = gfs2_inplace_reserve(ip);
790 blocks += al->al_rgd->rd_length + nalloc * ind_blocks + RES_STATFS;
792 error = gfs2_trans_begin(sdp, blocks, 0);
796 for (x = 0; x < num_qd; x++) {
798 offset = qd2offset(qd);
799 error = gfs2_adjust_quota(ip, offset, qd->qd_change_sync,
800 (struct gfs2_quota_data *)
805 do_qc(qd, -qd->qd_change_sync);
813 gfs2_inplace_release(ip);
817 gfs2_glock_dq_uninit(&i_gh);
820 gfs2_glock_dq_uninit(&ghs[qx]);
822 gfs2_log_flush(ip->i_gl->gl_sbd, ip->i_gl);
826 static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
827 struct gfs2_holder *q_gh)
829 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
830 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
831 struct gfs2_holder i_gh;
832 struct gfs2_quota_host q;
833 char buf[sizeof(struct gfs2_quota)];
835 struct gfs2_quota_lvb *qlvb;
838 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
842 qd->qd_qb = *(struct gfs2_quota_lvb *)qd->qd_gl->gl_lvb;
844 if (force_refresh || qd->qd_qb.qb_magic != cpu_to_be32(GFS2_MAGIC)) {
846 gfs2_glock_dq_uninit(q_gh);
847 error = gfs2_glock_nq_init(qd->qd_gl,
848 LM_ST_EXCLUSIVE, GL_NOCACHE,
853 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
857 memset(buf, 0, sizeof(struct gfs2_quota));
859 error = gfs2_internal_read(ip, NULL, buf, &pos,
860 sizeof(struct gfs2_quota));
864 gfs2_glock_dq_uninit(&i_gh);
866 gfs2_quota_in(&q, buf);
867 qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lvb;
868 qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC);
870 qlvb->qb_limit = cpu_to_be64(q.qu_limit);
871 qlvb->qb_warn = cpu_to_be64(q.qu_warn);
872 qlvb->qb_value = cpu_to_be64(q.qu_value);
875 if (gfs2_glock_is_blocking(qd->qd_gl)) {
876 gfs2_glock_dq_uninit(q_gh);
885 gfs2_glock_dq_uninit(&i_gh);
887 gfs2_glock_dq_uninit(q_gh);
891 int gfs2_quota_lock(struct gfs2_inode *ip, u32 uid, u32 gid)
893 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
894 struct gfs2_alloc *al = ip->i_alloc;
898 gfs2_quota_hold(ip, uid, gid);
900 if (capable(CAP_SYS_RESOURCE) ||
901 sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
904 sort(al->al_qd, al->al_qd_num, sizeof(struct gfs2_quota_data *),
907 for (x = 0; x < al->al_qd_num; x++) {
908 error = do_glock(al->al_qd[x], NO_FORCE, &al->al_qd_ghs[x]);
914 set_bit(GIF_QD_LOCKED, &ip->i_flags);
917 gfs2_glock_dq_uninit(&al->al_qd_ghs[x]);
918 gfs2_quota_unhold(ip);
924 static int need_sync(struct gfs2_quota_data *qd)
926 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
927 struct gfs2_tune *gt = &sdp->sd_tune;
929 unsigned int num, den;
932 if (!qd->qd_qb.qb_limit)
935 spin_lock(&qd_lru_lock);
936 value = qd->qd_change;
937 spin_unlock(&qd_lru_lock);
939 spin_lock(>->gt_spin);
940 num = gt->gt_quota_scale_num;
941 den = gt->gt_quota_scale_den;
942 spin_unlock(>->gt_spin);
946 else if ((s64)be64_to_cpu(qd->qd_qb.qb_value) >=
947 (s64)be64_to_cpu(qd->qd_qb.qb_limit))
950 value *= gfs2_jindex_size(sdp) * num;
951 value = div_s64(value, den);
952 value += (s64)be64_to_cpu(qd->qd_qb.qb_value);
953 if (value < (s64)be64_to_cpu(qd->qd_qb.qb_limit))
960 void gfs2_quota_unlock(struct gfs2_inode *ip)
962 struct gfs2_alloc *al = ip->i_alloc;
963 struct gfs2_quota_data *qda[4];
964 unsigned int count = 0;
967 if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
970 for (x = 0; x < al->al_qd_num; x++) {
971 struct gfs2_quota_data *qd;
975 sync = need_sync(qd);
977 gfs2_glock_dq_uninit(&al->al_qd_ghs[x]);
979 if (sync && qd_trylock(qd))
985 for (x = 0; x < count; x++)
990 gfs2_quota_unhold(ip);
995 static int print_message(struct gfs2_quota_data *qd, char *type)
997 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
999 printk(KERN_INFO "GFS2: fsid=%s: quota %s for %s %u\r\n",
1000 sdp->sd_fsname, type,
1001 (test_bit(QDF_USER, &qd->qd_flags)) ? "user" : "group",
1007 int gfs2_quota_check(struct gfs2_inode *ip, u32 uid, u32 gid)
1009 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1010 struct gfs2_alloc *al = ip->i_alloc;
1011 struct gfs2_quota_data *qd;
1016 if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
1019 if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
1022 for (x = 0; x < al->al_qd_num; x++) {
1025 if (!((qd->qd_id == uid && test_bit(QDF_USER, &qd->qd_flags)) ||
1026 (qd->qd_id == gid && !test_bit(QDF_USER, &qd->qd_flags))))
1029 value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
1030 spin_lock(&qd_lru_lock);
1031 value += qd->qd_change;
1032 spin_unlock(&qd_lru_lock);
1034 if (be64_to_cpu(qd->qd_qb.qb_limit) && (s64)be64_to_cpu(qd->qd_qb.qb_limit) < value) {
1035 print_message(qd, "exceeded");
1038 } else if (be64_to_cpu(qd->qd_qb.qb_warn) &&
1039 (s64)be64_to_cpu(qd->qd_qb.qb_warn) < value &&
1040 time_after_eq(jiffies, qd->qd_last_warn +
1042 gt_quota_warn_period) * HZ)) {
1043 error = print_message(qd, "warning");
1044 qd->qd_last_warn = jiffies;
1051 void gfs2_quota_change(struct gfs2_inode *ip, s64 change,
1054 struct gfs2_alloc *al = ip->i_alloc;
1055 struct gfs2_quota_data *qd;
1058 if (gfs2_assert_warn(GFS2_SB(&ip->i_inode), change))
1060 if (ip->i_diskflags & GFS2_DIF_SYSTEM)
1063 for (x = 0; x < al->al_qd_num; x++) {
1066 if ((qd->qd_id == uid && test_bit(QDF_USER, &qd->qd_flags)) ||
1067 (qd->qd_id == gid && !test_bit(QDF_USER, &qd->qd_flags))) {
1073 int gfs2_quota_sync(struct gfs2_sbd *sdp)
1075 struct gfs2_quota_data **qda;
1076 unsigned int max_qd = gfs2_tune_get(sdp, gt_quota_simul_sync);
1077 unsigned int num_qd;
1081 sdp->sd_quota_sync_gen++;
1083 qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
1091 error = qd_fish(sdp, qda + num_qd);
1092 if (error || !qda[num_qd])
1094 if (++num_qd == max_qd)
1100 error = do_sync(num_qd, qda);
1102 for (x = 0; x < num_qd; x++)
1103 qda[x]->qd_sync_gen =
1104 sdp->sd_quota_sync_gen;
1106 for (x = 0; x < num_qd; x++)
1109 } while (!error && num_qd == max_qd);
1116 int gfs2_quota_refresh(struct gfs2_sbd *sdp, int user, u32 id)
1118 struct gfs2_quota_data *qd;
1119 struct gfs2_holder q_gh;
1122 error = qd_get(sdp, user, id, CREATE, &qd);
1126 error = do_glock(qd, FORCE, &q_gh);
1128 gfs2_glock_dq_uninit(&q_gh);
1135 static void gfs2_quota_change_in(struct gfs2_quota_change_host *qc, const void *buf)
1137 const struct gfs2_quota_change *str = buf;
1139 qc->qc_change = be64_to_cpu(str->qc_change);
1140 qc->qc_flags = be32_to_cpu(str->qc_flags);
1141 qc->qc_id = be32_to_cpu(str->qc_id);
1144 int gfs2_quota_init(struct gfs2_sbd *sdp)
1146 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
1147 unsigned int blocks = ip->i_disksize >> sdp->sd_sb.sb_bsize_shift;
1148 unsigned int x, slot = 0;
1149 unsigned int found = 0;
1154 if (!ip->i_disksize || ip->i_disksize > (64 << 20) ||
1155 ip->i_disksize & (sdp->sd_sb.sb_bsize - 1)) {
1156 gfs2_consist_inode(ip);
1159 sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
1160 sdp->sd_quota_chunks = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * PAGE_SIZE);
1164 sdp->sd_quota_bitmap = kcalloc(sdp->sd_quota_chunks,
1165 sizeof(unsigned char *), GFP_NOFS);
1166 if (!sdp->sd_quota_bitmap)
1169 for (x = 0; x < sdp->sd_quota_chunks; x++) {
1170 sdp->sd_quota_bitmap[x] = kzalloc(PAGE_SIZE, GFP_NOFS);
1171 if (!sdp->sd_quota_bitmap[x])
1175 for (x = 0; x < blocks; x++) {
1176 struct buffer_head *bh;
1181 error = gfs2_extent_map(&ip->i_inode, x, &new, &dblock, &extlen);
1186 bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
1189 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
1194 for (y = 0; y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1196 struct gfs2_quota_change_host qc;
1197 struct gfs2_quota_data *qd;
1199 gfs2_quota_change_in(&qc, bh->b_data +
1200 sizeof(struct gfs2_meta_header) +
1201 y * sizeof(struct gfs2_quota_change));
1205 error = qd_alloc(sdp, (qc.qc_flags & GFS2_QCF_USER),
1212 set_bit(QDF_CHANGE, &qd->qd_flags);
1213 qd->qd_change = qc.qc_change;
1215 qd->qd_slot_count = 1;
1217 spin_lock(&qd_lru_lock);
1218 gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, slot, 1);
1219 list_add(&qd->qd_list, &sdp->sd_quota_list);
1220 atomic_inc(&sdp->sd_quota_count);
1221 spin_unlock(&qd_lru_lock);
1232 fs_info(sdp, "found %u quota changes\n", found);
1237 gfs2_quota_cleanup(sdp);
1241 void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1243 struct list_head *head = &sdp->sd_quota_list;
1244 struct gfs2_quota_data *qd;
1247 spin_lock(&qd_lru_lock);
1248 while (!list_empty(head)) {
1249 qd = list_entry(head->prev, struct gfs2_quota_data, qd_list);
1251 if (atomic_read(&qd->qd_count) > 1 ||
1252 (atomic_read(&qd->qd_count) &&
1253 !test_bit(QDF_CHANGE, &qd->qd_flags))) {
1254 list_move(&qd->qd_list, head);
1255 spin_unlock(&qd_lru_lock);
1257 spin_lock(&qd_lru_lock);
1261 list_del(&qd->qd_list);
1262 /* Also remove if this qd exists in the reclaim list */
1263 if (!list_empty(&qd->qd_reclaim)) {
1264 list_del_init(&qd->qd_reclaim);
1265 atomic_dec(&qd_lru_count);
1267 atomic_dec(&sdp->sd_quota_count);
1268 spin_unlock(&qd_lru_lock);
1270 if (!atomic_read(&qd->qd_count)) {
1271 gfs2_assert_warn(sdp, !qd->qd_change);
1272 gfs2_assert_warn(sdp, !qd->qd_slot_count);
1274 gfs2_assert_warn(sdp, qd->qd_slot_count == 1);
1275 gfs2_assert_warn(sdp, !qd->qd_bh_count);
1277 gfs2_glock_put(qd->qd_gl);
1278 kmem_cache_free(gfs2_quotad_cachep, qd);
1280 spin_lock(&qd_lru_lock);
1282 spin_unlock(&qd_lru_lock);
1284 gfs2_assert_warn(sdp, !atomic_read(&sdp->sd_quota_count));
1286 if (sdp->sd_quota_bitmap) {
1287 for (x = 0; x < sdp->sd_quota_chunks; x++)
1288 kfree(sdp->sd_quota_bitmap[x]);
1289 kfree(sdp->sd_quota_bitmap);
1293 static void quotad_error(struct gfs2_sbd *sdp, const char *msg, int error)
1295 if (error == 0 || error == -EROFS)
1297 if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
1298 fs_err(sdp, "gfs2_quotad: %s error %d\n", msg, error);
1301 static void quotad_check_timeo(struct gfs2_sbd *sdp, const char *msg,
1302 int (*fxn)(struct gfs2_sbd *sdp),
1303 unsigned long t, unsigned long *timeo,
1304 unsigned int *new_timeo)
1307 int error = fxn(sdp);
1308 quotad_error(sdp, msg, error);
1309 *timeo = gfs2_tune_get_i(&sdp->sd_tune, new_timeo) * HZ;
1315 static void quotad_check_trunc_list(struct gfs2_sbd *sdp)
1317 struct gfs2_inode *ip;
1321 spin_lock(&sdp->sd_trunc_lock);
1322 if (!list_empty(&sdp->sd_trunc_list)) {
1323 ip = list_entry(sdp->sd_trunc_list.next,
1324 struct gfs2_inode, i_trunc_list);
1325 list_del_init(&ip->i_trunc_list);
1327 spin_unlock(&sdp->sd_trunc_lock);
1330 gfs2_glock_finish_truncate(ip);
1335 * gfs2_quotad - Write cached quota changes into the quota file
1336 * @sdp: Pointer to GFS2 superblock
1340 int gfs2_quotad(void *data)
1342 struct gfs2_sbd *sdp = data;
1343 struct gfs2_tune *tune = &sdp->sd_tune;
1344 unsigned long statfs_timeo = 0;
1345 unsigned long quotad_timeo = 0;
1346 unsigned long t = 0;
1350 while (!kthread_should_stop()) {
1352 /* Update the master statfs file */
1353 quotad_check_timeo(sdp, "statfs", gfs2_statfs_sync, t,
1354 &statfs_timeo, &tune->gt_statfs_quantum);
1356 /* Update quota file */
1357 quotad_check_timeo(sdp, "sync", gfs2_quota_sync, t,
1358 "ad_timeo, &tune->gt_quota_quantum);
1360 /* Check for & recover partially truncated inodes */
1361 quotad_check_trunc_list(sdp);
1363 if (freezing(current))
1365 t = min(quotad_timeo, statfs_timeo);
1367 prepare_to_wait(&sdp->sd_quota_wait, &wait, TASK_INTERRUPTIBLE);
1368 spin_lock(&sdp->sd_trunc_lock);
1369 empty = list_empty(&sdp->sd_trunc_list);
1370 spin_unlock(&sdp->sd_trunc_lock);
1372 t -= schedule_timeout(t);
1375 finish_wait(&sdp->sd_quota_wait, &wait);