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 for 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>
42 #include <linux/spinlock.h>
43 #include <linux/completion.h>
44 #include <linux/buffer_head.h>
45 #include <linux/sort.h>
47 #include <linux/bio.h>
48 #include <linux/gfs2_ondisk.h>
49 #include <linux/kthread.h>
50 #include <linux/freezer.h>
51 #include <linux/quota.h>
52 #include <linux/dqblk_xfs.h>
53 #include <linux/lockref.h>
54 #include <linux/list_lru.h>
70 struct gfs2_quota_change_host {
72 u32 qc_flags; /* GFS2_QCF_... */
76 /* Lock order: qd_lock -> qd->lockref.lock -> lru lock */
77 static DEFINE_SPINLOCK(qd_lock);
78 struct list_lru gfs2_qd_lru;
80 static void gfs2_qd_dispose(struct list_head *list)
82 struct gfs2_quota_data *qd;
85 while (!list_empty(list)) {
86 qd = list_entry(list->next, struct gfs2_quota_data, qd_lru);
87 sdp = qd->qd_gl->gl_sbd;
89 list_del(&qd->qd_lru);
91 /* Free from the filesystem-specific list */
93 list_del(&qd->qd_list);
94 spin_unlock(&qd_lock);
96 gfs2_assert_warn(sdp, !qd->qd_change);
97 gfs2_assert_warn(sdp, !qd->qd_slot_count);
98 gfs2_assert_warn(sdp, !qd->qd_bh_count);
100 gfs2_glock_put(qd->qd_gl);
101 atomic_dec(&sdp->sd_quota_count);
103 /* Delete it from the common reclaim list */
104 kmem_cache_free(gfs2_quotad_cachep, qd);
109 static enum lru_status gfs2_qd_isolate(struct list_head *item, spinlock_t *lock, void *arg)
111 struct list_head *dispose = arg;
112 struct gfs2_quota_data *qd = list_entry(item, struct gfs2_quota_data, qd_lru);
114 if (!spin_trylock(&qd->qd_lockref.lock))
117 if (qd->qd_lockref.count == 0) {
118 lockref_mark_dead(&qd->qd_lockref);
119 list_move(&qd->qd_lru, dispose);
122 spin_unlock(&qd->qd_lockref.lock);
126 static unsigned long gfs2_qd_shrink_scan(struct shrinker *shrink,
127 struct shrink_control *sc)
132 if (!(sc->gfp_mask & __GFP_FS))
135 freed = list_lru_walk_node(&gfs2_qd_lru, sc->nid, gfs2_qd_isolate,
136 &dispose, &sc->nr_to_scan);
138 gfs2_qd_dispose(&dispose);
143 static unsigned long gfs2_qd_shrink_count(struct shrinker *shrink,
144 struct shrink_control *sc)
146 return vfs_pressure_ratio(list_lru_count_node(&gfs2_qd_lru, sc->nid));
149 struct shrinker gfs2_qd_shrinker = {
150 .count_objects = gfs2_qd_shrink_count,
151 .scan_objects = gfs2_qd_shrink_scan,
152 .seeks = DEFAULT_SEEKS,
153 .flags = SHRINKER_NUMA_AWARE,
157 static u64 qd2index(struct gfs2_quota_data *qd)
159 struct kqid qid = qd->qd_id;
160 return (2 * (u64)from_kqid(&init_user_ns, qid)) +
161 ((qid.type == USRQUOTA) ? 0 : 1);
164 static u64 qd2offset(struct gfs2_quota_data *qd)
168 offset = qd2index(qd);
169 offset *= sizeof(struct gfs2_quota);
174 static int qd_alloc(struct gfs2_sbd *sdp, struct kqid qid,
175 struct gfs2_quota_data **qdp)
177 struct gfs2_quota_data *qd;
180 qd = kmem_cache_zalloc(gfs2_quotad_cachep, GFP_NOFS);
184 qd->qd_lockref.count = 1;
185 spin_lock_init(&qd->qd_lockref.lock);
188 INIT_LIST_HEAD(&qd->qd_lru);
190 error = gfs2_glock_get(sdp, qd2index(qd),
191 &gfs2_quota_glops, CREATE, &qd->qd_gl);
200 kmem_cache_free(gfs2_quotad_cachep, qd);
204 static int qd_get(struct gfs2_sbd *sdp, struct kqid qid,
205 struct gfs2_quota_data **qdp)
207 struct gfs2_quota_data *qd = NULL, *new_qd = NULL;
215 list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
216 if (qid_eq(qd->qd_id, qid) &&
217 lockref_get_not_dead(&qd->qd_lockref)) {
218 list_lru_del(&gfs2_qd_lru, &qd->qd_lru);
229 list_add(&qd->qd_list, &sdp->sd_quota_list);
230 atomic_inc(&sdp->sd_quota_count);
234 spin_unlock(&qd_lock);
238 gfs2_glock_put(new_qd->qd_gl);
239 kmem_cache_free(gfs2_quotad_cachep, new_qd);
245 error = qd_alloc(sdp, qid, &new_qd);
251 static void qd_hold(struct gfs2_quota_data *qd)
253 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
254 gfs2_assert(sdp, !__lockref_is_dead(&qd->qd_lockref));
255 lockref_get(&qd->qd_lockref);
258 static void qd_put(struct gfs2_quota_data *qd)
260 if (lockref_put_or_lock(&qd->qd_lockref))
263 qd->qd_lockref.count = 0;
264 list_lru_add(&gfs2_qd_lru, &qd->qd_lru);
265 spin_unlock(&qd->qd_lockref.lock);
269 static int slot_get(struct gfs2_quota_data *qd)
271 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
272 unsigned int c, o = 0, b;
273 unsigned char byte = 0;
277 if (qd->qd_slot_count++) {
278 spin_unlock(&qd_lock);
282 for (c = 0; c < sdp->sd_quota_chunks; c++)
283 for (o = 0; o < PAGE_SIZE; o++) {
284 byte = sdp->sd_quota_bitmap[c][o];
292 for (b = 0; b < 8; b++)
293 if (!(byte & (1 << b)))
295 qd->qd_slot = c * (8 * PAGE_SIZE) + o * 8 + b;
297 if (qd->qd_slot >= sdp->sd_quota_slots)
300 sdp->sd_quota_bitmap[c][o] |= 1 << b;
302 spin_unlock(&qd_lock);
308 spin_unlock(&qd_lock);
312 static void slot_hold(struct gfs2_quota_data *qd)
314 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
317 gfs2_assert(sdp, qd->qd_slot_count);
319 spin_unlock(&qd_lock);
322 static void gfs2_icbit_munge(struct gfs2_sbd *sdp, unsigned char **bitmap,
323 unsigned int bit, int new_value)
325 unsigned int c, o, b = bit;
328 c = b / (8 * PAGE_SIZE);
333 old_value = (bitmap[c][o] & (1 << b));
334 gfs2_assert_withdraw(sdp, !old_value != !new_value);
337 bitmap[c][o] |= 1 << b;
339 bitmap[c][o] &= ~(1 << b);
342 static void slot_put(struct gfs2_quota_data *qd)
344 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
347 gfs2_assert(sdp, qd->qd_slot_count);
348 if (!--qd->qd_slot_count) {
349 gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, qd->qd_slot, 0);
352 spin_unlock(&qd_lock);
355 static int bh_get(struct gfs2_quota_data *qd)
357 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
358 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
359 unsigned int block, offset;
360 struct buffer_head *bh;
362 struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
364 mutex_lock(&sdp->sd_quota_mutex);
366 if (qd->qd_bh_count++) {
367 mutex_unlock(&sdp->sd_quota_mutex);
371 block = qd->qd_slot / sdp->sd_qc_per_block;
372 offset = qd->qd_slot % sdp->sd_qc_per_block;
374 bh_map.b_size = 1 << ip->i_inode.i_blkbits;
375 error = gfs2_block_map(&ip->i_inode, block, &bh_map, 0);
378 error = gfs2_meta_read(ip->i_gl, bh_map.b_blocknr, DIO_WAIT, &bh);
382 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC))
386 qd->qd_bh_qc = (struct gfs2_quota_change *)
387 (bh->b_data + sizeof(struct gfs2_meta_header) +
388 offset * sizeof(struct gfs2_quota_change));
390 mutex_unlock(&sdp->sd_quota_mutex);
398 mutex_unlock(&sdp->sd_quota_mutex);
402 static void bh_put(struct gfs2_quota_data *qd)
404 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
406 mutex_lock(&sdp->sd_quota_mutex);
407 gfs2_assert(sdp, qd->qd_bh_count);
408 if (!--qd->qd_bh_count) {
413 mutex_unlock(&sdp->sd_quota_mutex);
416 static int qd_check_sync(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd,
419 if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
420 !test_bit(QDF_CHANGE, &qd->qd_flags) ||
421 (sync_gen && (qd->qd_sync_gen >= *sync_gen)))
424 if (!lockref_get_not_dead(&qd->qd_lockref))
427 list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
428 set_bit(QDF_LOCKED, &qd->qd_flags);
429 qd->qd_change_sync = qd->qd_change;
430 gfs2_assert_warn(sdp, qd->qd_slot_count);
435 static int qd_fish(struct gfs2_sbd *sdp, struct gfs2_quota_data **qdp)
437 struct gfs2_quota_data *qd = NULL;
443 if (sdp->sd_vfs->s_flags & MS_RDONLY)
448 list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
449 found = qd_check_sync(sdp, qd, &sdp->sd_quota_sync_gen);
457 spin_unlock(&qd_lock);
460 gfs2_assert_warn(sdp, qd->qd_change_sync);
463 clear_bit(QDF_LOCKED, &qd->qd_flags);
475 static void qd_unlock(struct gfs2_quota_data *qd)
477 gfs2_assert_warn(qd->qd_gl->gl_sbd,
478 test_bit(QDF_LOCKED, &qd->qd_flags));
479 clear_bit(QDF_LOCKED, &qd->qd_flags);
485 static int qdsb_get(struct gfs2_sbd *sdp, struct kqid qid,
486 struct gfs2_quota_data **qdp)
490 error = qd_get(sdp, qid, qdp);
494 error = slot_get(*qdp);
498 error = bh_get(*qdp);
511 static void qdsb_put(struct gfs2_quota_data *qd)
518 int gfs2_quota_hold(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
520 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
521 struct gfs2_quota_data **qd;
524 if (ip->i_res == NULL) {
525 error = gfs2_rs_alloc(ip);
530 qd = ip->i_res->rs_qa_qd;
532 if (gfs2_assert_warn(sdp, !ip->i_res->rs_qa_qd_num) ||
533 gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags)))
536 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
539 error = qdsb_get(sdp, make_kqid_uid(ip->i_inode.i_uid), qd);
542 ip->i_res->rs_qa_qd_num++;
545 error = qdsb_get(sdp, make_kqid_gid(ip->i_inode.i_gid), qd);
548 ip->i_res->rs_qa_qd_num++;
551 if (!uid_eq(uid, NO_UID_QUOTA_CHANGE) &&
552 !uid_eq(uid, ip->i_inode.i_uid)) {
553 error = qdsb_get(sdp, make_kqid_uid(uid), qd);
556 ip->i_res->rs_qa_qd_num++;
560 if (!gid_eq(gid, NO_GID_QUOTA_CHANGE) &&
561 !gid_eq(gid, ip->i_inode.i_gid)) {
562 error = qdsb_get(sdp, make_kqid_gid(gid), qd);
565 ip->i_res->rs_qa_qd_num++;
571 gfs2_quota_unhold(ip);
575 void gfs2_quota_unhold(struct gfs2_inode *ip)
577 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
580 if (ip->i_res == NULL)
582 gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags));
584 for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
585 qdsb_put(ip->i_res->rs_qa_qd[x]);
586 ip->i_res->rs_qa_qd[x] = NULL;
588 ip->i_res->rs_qa_qd_num = 0;
591 static int sort_qd(const void *a, const void *b)
593 const struct gfs2_quota_data *qd_a = *(const struct gfs2_quota_data **)a;
594 const struct gfs2_quota_data *qd_b = *(const struct gfs2_quota_data **)b;
596 if (qid_lt(qd_a->qd_id, qd_b->qd_id))
598 if (qid_lt(qd_b->qd_id, qd_a->qd_id))
603 static void do_qc(struct gfs2_quota_data *qd, s64 change)
605 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
606 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
607 struct gfs2_quota_change *qc = qd->qd_bh_qc;
610 mutex_lock(&sdp->sd_quota_mutex);
611 gfs2_trans_add_meta(ip->i_gl, qd->qd_bh);
613 if (!test_bit(QDF_CHANGE, &qd->qd_flags)) {
616 if (qd->qd_id.type == USRQUOTA)
617 qc->qc_flags = cpu_to_be32(GFS2_QCF_USER);
618 qc->qc_id = cpu_to_be32(from_kqid(&init_user_ns, qd->qd_id));
621 x = be64_to_cpu(qc->qc_change) + change;
622 qc->qc_change = cpu_to_be64(x);
626 spin_unlock(&qd_lock);
629 gfs2_assert_warn(sdp, test_bit(QDF_CHANGE, &qd->qd_flags));
630 clear_bit(QDF_CHANGE, &qd->qd_flags);
635 } else if (!test_and_set_bit(QDF_CHANGE, &qd->qd_flags)) {
640 mutex_unlock(&sdp->sd_quota_mutex);
644 * gfs2_adjust_quota - adjust record of current block usage
645 * @ip: The quota inode
646 * @loc: Offset of the entry in the quota file
647 * @change: The amount of usage change to record
648 * @qd: The quota data
649 * @fdq: The updated limits to record
651 * This function was mostly borrowed from gfs2_block_truncate_page which was
652 * in turn mostly borrowed from ext3
654 * Returns: 0 or -ve on error
657 static int gfs2_adjust_quota(struct gfs2_inode *ip, loff_t loc,
658 s64 change, struct gfs2_quota_data *qd,
659 struct fs_disk_quota *fdq)
661 struct inode *inode = &ip->i_inode;
662 struct gfs2_sbd *sdp = GFS2_SB(inode);
663 struct address_space *mapping = inode->i_mapping;
664 unsigned long index = loc >> PAGE_CACHE_SHIFT;
665 unsigned offset = loc & (PAGE_CACHE_SIZE - 1);
666 unsigned blocksize, iblock, pos;
667 struct buffer_head *bh;
674 if (gfs2_is_stuffed(ip)) {
675 err = gfs2_unstuff_dinode(ip, NULL);
680 memset(&q, 0, sizeof(struct gfs2_quota));
681 err = gfs2_internal_read(ip, (char *)&q, &loc, sizeof(q));
686 be64_add_cpu(&q.qu_value, change);
687 qd->qd_qb.qb_value = q.qu_value;
689 if (fdq->d_fieldmask & FS_DQ_BSOFT) {
690 q.qu_warn = cpu_to_be64(fdq->d_blk_softlimit >> sdp->sd_fsb2bb_shift);
691 qd->qd_qb.qb_warn = q.qu_warn;
693 if (fdq->d_fieldmask & FS_DQ_BHARD) {
694 q.qu_limit = cpu_to_be64(fdq->d_blk_hardlimit >> sdp->sd_fsb2bb_shift);
695 qd->qd_qb.qb_limit = q.qu_limit;
697 if (fdq->d_fieldmask & FS_DQ_BCOUNT) {
698 q.qu_value = cpu_to_be64(fdq->d_bcount >> sdp->sd_fsb2bb_shift);
699 qd->qd_qb.qb_value = q.qu_value;
703 /* Write the quota into the quota file on disk */
705 nbytes = sizeof(struct gfs2_quota);
707 page = find_or_create_page(mapping, index, GFP_NOFS);
711 blocksize = inode->i_sb->s_blocksize;
712 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
714 if (!page_has_buffers(page))
715 create_empty_buffers(page, blocksize, 0);
717 bh = page_buffers(page);
719 while (offset >= pos) {
720 bh = bh->b_this_page;
725 if (!buffer_mapped(bh)) {
726 gfs2_block_map(inode, iblock, bh, 1);
727 if (!buffer_mapped(bh))
729 /* If it's a newly allocated disk block for quota, zero it */
731 zero_user(page, pos - blocksize, bh->b_size);
734 if (PageUptodate(page))
735 set_buffer_uptodate(bh);
737 if (!buffer_uptodate(bh)) {
738 ll_rw_block(READ | REQ_META, 1, &bh);
740 if (!buffer_uptodate(bh))
744 gfs2_trans_add_data(ip->i_gl, bh);
746 kaddr = kmap_atomic(page);
747 if (offset + sizeof(struct gfs2_quota) > PAGE_CACHE_SIZE)
748 nbytes = PAGE_CACHE_SIZE - offset;
749 memcpy(kaddr + offset, ptr, nbytes);
750 flush_dcache_page(page);
751 kunmap_atomic(kaddr);
753 page_cache_release(page);
755 /* If quota straddles page boundary, we need to update the rest of the
756 * quota at the beginning of the next page */
757 if ((offset + sizeof(struct gfs2_quota)) > PAGE_CACHE_SIZE) {
759 nbytes = sizeof(struct gfs2_quota) - nbytes;
765 size = loc + sizeof(struct gfs2_quota);
766 if (size > inode->i_size)
767 i_size_write(inode, size);
768 inode->i_mtime = inode->i_atime = CURRENT_TIME;
769 mark_inode_dirty(inode);
774 page_cache_release(page);
778 static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
780 struct gfs2_sbd *sdp = (*qda)->qd_gl->gl_sbd;
781 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
782 struct gfs2_alloc_parms ap = { .aflags = 0, };
783 unsigned int data_blocks, ind_blocks;
784 struct gfs2_holder *ghs, i_gh;
786 struct gfs2_quota_data *qd;
789 unsigned int nalloc = 0, blocks;
792 error = gfs2_rs_alloc(ip);
796 gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
797 &data_blocks, &ind_blocks);
799 ghs = kcalloc(num_qd, sizeof(struct gfs2_holder), GFP_NOFS);
803 sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL);
804 mutex_lock(&ip->i_inode.i_mutex);
805 for (qx = 0; qx < num_qd; qx++) {
806 error = gfs2_glock_nq_init(qda[qx]->qd_gl, LM_ST_EXCLUSIVE,
807 GL_NOCACHE, &ghs[qx]);
812 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
816 for (x = 0; x < num_qd; x++) {
817 offset = qd2offset(qda[x]);
818 if (gfs2_write_alloc_required(ip, offset,
819 sizeof(struct gfs2_quota)))
824 * 1 blk for unstuffing inode if stuffed. We add this extra
825 * block to the reservation unconditionally. If the inode
826 * doesn't need unstuffing, the block will be released to the
827 * rgrp since it won't be allocated during the transaction
829 /* +3 in the end for unstuffing block, inode size update block
830 * and another block in case quota straddles page boundary and
831 * two blocks need to be updated instead of 1 */
832 blocks = num_qd * data_blocks + RES_DINODE + num_qd + 3;
834 reserved = 1 + (nalloc * (data_blocks + ind_blocks));
835 ap.target = reserved;
836 error = gfs2_inplace_reserve(ip, &ap);
841 blocks += gfs2_rg_blocks(ip, reserved) + nalloc * ind_blocks + RES_STATFS;
843 error = gfs2_trans_begin(sdp, blocks, 0);
847 for (x = 0; x < num_qd; x++) {
849 offset = qd2offset(qd);
850 error = gfs2_adjust_quota(ip, offset, qd->qd_change_sync, qd, NULL);
854 do_qc(qd, -qd->qd_change_sync);
855 set_bit(QDF_REFRESH, &qd->qd_flags);
863 gfs2_inplace_release(ip);
865 gfs2_glock_dq_uninit(&i_gh);
868 gfs2_glock_dq_uninit(&ghs[qx]);
869 mutex_unlock(&ip->i_inode.i_mutex);
871 gfs2_log_flush(ip->i_gl->gl_sbd, ip->i_gl);
875 static int update_qd(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd)
877 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
879 struct gfs2_quota_lvb *qlvb;
883 memset(&q, 0, sizeof(struct gfs2_quota));
885 error = gfs2_internal_read(ip, (char *)&q, &pos, sizeof(q));
889 qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
890 qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC);
892 qlvb->qb_limit = q.qu_limit;
893 qlvb->qb_warn = q.qu_warn;
894 qlvb->qb_value = q.qu_value;
900 static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
901 struct gfs2_holder *q_gh)
903 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
904 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
905 struct gfs2_holder i_gh;
909 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
913 qd->qd_qb = *(struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
915 if (force_refresh || qd->qd_qb.qb_magic != cpu_to_be32(GFS2_MAGIC)) {
916 gfs2_glock_dq_uninit(q_gh);
917 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE,
922 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
926 error = update_qd(sdp, qd);
930 gfs2_glock_dq_uninit(&i_gh);
931 gfs2_glock_dq_uninit(q_gh);
939 gfs2_glock_dq_uninit(&i_gh);
941 gfs2_glock_dq_uninit(q_gh);
945 int gfs2_quota_lock(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
947 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
948 struct gfs2_quota_data *qd;
952 error = gfs2_quota_hold(ip, uid, gid);
956 if (capable(CAP_SYS_RESOURCE) ||
957 sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
960 sort(ip->i_res->rs_qa_qd, ip->i_res->rs_qa_qd_num,
961 sizeof(struct gfs2_quota_data *), sort_qd, NULL);
963 for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
964 int force = NO_FORCE;
965 qd = ip->i_res->rs_qa_qd[x];
966 if (test_and_clear_bit(QDF_REFRESH, &qd->qd_flags))
968 error = do_glock(qd, force, &ip->i_res->rs_qa_qd_ghs[x]);
974 set_bit(GIF_QD_LOCKED, &ip->i_flags);
977 gfs2_glock_dq_uninit(&ip->i_res->rs_qa_qd_ghs[x]);
978 gfs2_quota_unhold(ip);
984 static int need_sync(struct gfs2_quota_data *qd)
986 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
987 struct gfs2_tune *gt = &sdp->sd_tune;
989 unsigned int num, den;
992 if (!qd->qd_qb.qb_limit)
996 value = qd->qd_change;
997 spin_unlock(&qd_lock);
999 spin_lock(>->gt_spin);
1000 num = gt->gt_quota_scale_num;
1001 den = gt->gt_quota_scale_den;
1002 spin_unlock(>->gt_spin);
1006 else if ((s64)be64_to_cpu(qd->qd_qb.qb_value) >=
1007 (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1010 value *= gfs2_jindex_size(sdp) * num;
1011 value = div_s64(value, den);
1012 value += (s64)be64_to_cpu(qd->qd_qb.qb_value);
1013 if (value < (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1020 void gfs2_quota_unlock(struct gfs2_inode *ip)
1022 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1023 struct gfs2_quota_data *qda[4];
1024 unsigned int count = 0;
1028 if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
1031 for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1032 struct gfs2_quota_data *qd;
1035 qd = ip->i_res->rs_qa_qd[x];
1036 sync = need_sync(qd);
1038 gfs2_glock_dq_uninit(&ip->i_res->rs_qa_qd_ghs[x]);
1042 spin_lock(&qd_lock);
1043 found = qd_check_sync(sdp, qd, NULL);
1044 spin_unlock(&qd_lock);
1049 gfs2_assert_warn(sdp, qd->qd_change_sync);
1051 clear_bit(QDF_LOCKED, &qd->qd_flags);
1061 do_sync(count, qda);
1062 for (x = 0; x < count; x++)
1067 gfs2_quota_unhold(ip);
1070 #define MAX_LINE 256
1072 static int print_message(struct gfs2_quota_data *qd, char *type)
1074 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
1076 printk(KERN_INFO "GFS2: fsid=%s: quota %s for %s %u\n",
1077 sdp->sd_fsname, type,
1078 (qd->qd_id.type == USRQUOTA) ? "user" : "group",
1079 from_kqid(&init_user_ns, qd->qd_id));
1084 int gfs2_quota_check(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
1086 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1087 struct gfs2_quota_data *qd;
1092 if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
1095 if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
1098 for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1099 qd = ip->i_res->rs_qa_qd[x];
1101 if (!(qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1102 qid_eq(qd->qd_id, make_kqid_gid(gid))))
1105 value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
1106 spin_lock(&qd_lock);
1107 value += qd->qd_change;
1108 spin_unlock(&qd_lock);
1110 if (be64_to_cpu(qd->qd_qb.qb_limit) && (s64)be64_to_cpu(qd->qd_qb.qb_limit) < value) {
1111 print_message(qd, "exceeded");
1112 quota_send_warning(qd->qd_id,
1113 sdp->sd_vfs->s_dev, QUOTA_NL_BHARDWARN);
1117 } else if (be64_to_cpu(qd->qd_qb.qb_warn) &&
1118 (s64)be64_to_cpu(qd->qd_qb.qb_warn) < value &&
1119 time_after_eq(jiffies, qd->qd_last_warn +
1121 gt_quota_warn_period) * HZ)) {
1122 quota_send_warning(qd->qd_id,
1123 sdp->sd_vfs->s_dev, QUOTA_NL_BSOFTWARN);
1124 error = print_message(qd, "warning");
1125 qd->qd_last_warn = jiffies;
1132 void gfs2_quota_change(struct gfs2_inode *ip, s64 change,
1133 kuid_t uid, kgid_t gid)
1135 struct gfs2_quota_data *qd;
1138 if (gfs2_assert_warn(GFS2_SB(&ip->i_inode), change))
1140 if (ip->i_diskflags & GFS2_DIF_SYSTEM)
1143 for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1144 qd = ip->i_res->rs_qa_qd[x];
1146 if (qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1147 qid_eq(qd->qd_id, make_kqid_gid(gid))) {
1153 int gfs2_quota_sync(struct super_block *sb, int type)
1155 struct gfs2_sbd *sdp = sb->s_fs_info;
1156 struct gfs2_quota_data **qda;
1157 unsigned int max_qd = PAGE_SIZE/sizeof(struct gfs2_holder);
1158 unsigned int num_qd;
1162 qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
1166 mutex_lock(&sdp->sd_quota_sync_mutex);
1167 sdp->sd_quota_sync_gen++;
1173 error = qd_fish(sdp, qda + num_qd);
1174 if (error || !qda[num_qd])
1176 if (++num_qd == max_qd)
1182 error = do_sync(num_qd, qda);
1184 for (x = 0; x < num_qd; x++)
1185 qda[x]->qd_sync_gen =
1186 sdp->sd_quota_sync_gen;
1188 for (x = 0; x < num_qd; x++)
1191 } while (!error && num_qd == max_qd);
1193 mutex_unlock(&sdp->sd_quota_sync_mutex);
1199 int gfs2_quota_refresh(struct gfs2_sbd *sdp, struct kqid qid)
1201 struct gfs2_quota_data *qd;
1202 struct gfs2_holder q_gh;
1205 error = qd_get(sdp, qid, &qd);
1209 error = do_glock(qd, FORCE, &q_gh);
1211 gfs2_glock_dq_uninit(&q_gh);
1217 static void gfs2_quota_change_in(struct gfs2_quota_change_host *qc, const void *buf)
1219 const struct gfs2_quota_change *str = buf;
1221 qc->qc_change = be64_to_cpu(str->qc_change);
1222 qc->qc_flags = be32_to_cpu(str->qc_flags);
1223 qc->qc_id = make_kqid(&init_user_ns,
1224 (qc->qc_flags & GFS2_QCF_USER)?USRQUOTA:GRPQUOTA,
1225 be32_to_cpu(str->qc_id));
1228 int gfs2_quota_init(struct gfs2_sbd *sdp)
1230 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
1231 u64 size = i_size_read(sdp->sd_qc_inode);
1232 unsigned int blocks = size >> sdp->sd_sb.sb_bsize_shift;
1233 unsigned int x, slot = 0;
1234 unsigned int found = 0;
1239 if (gfs2_check_internal_file_size(sdp->sd_qc_inode, 1, 64 << 20))
1242 sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
1243 sdp->sd_quota_chunks = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * PAGE_SIZE);
1247 sdp->sd_quota_bitmap = kcalloc(sdp->sd_quota_chunks,
1248 sizeof(unsigned char *), GFP_NOFS);
1249 if (!sdp->sd_quota_bitmap)
1252 for (x = 0; x < sdp->sd_quota_chunks; x++) {
1253 sdp->sd_quota_bitmap[x] = kzalloc(PAGE_SIZE, GFP_NOFS);
1254 if (!sdp->sd_quota_bitmap[x])
1258 for (x = 0; x < blocks; x++) {
1259 struct buffer_head *bh;
1264 error = gfs2_extent_map(&ip->i_inode, x, &new, &dblock, &extlen);
1269 bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
1272 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
1277 for (y = 0; y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1279 struct gfs2_quota_change_host qc;
1280 struct gfs2_quota_data *qd;
1282 gfs2_quota_change_in(&qc, bh->b_data +
1283 sizeof(struct gfs2_meta_header) +
1284 y * sizeof(struct gfs2_quota_change));
1288 error = qd_alloc(sdp, qc.qc_id, &qd);
1294 set_bit(QDF_CHANGE, &qd->qd_flags);
1295 qd->qd_change = qc.qc_change;
1297 qd->qd_slot_count = 1;
1299 spin_lock(&qd_lock);
1300 gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, slot, 1);
1301 list_add(&qd->qd_list, &sdp->sd_quota_list);
1302 atomic_inc(&sdp->sd_quota_count);
1303 spin_unlock(&qd_lock);
1314 fs_info(sdp, "found %u quota changes\n", found);
1319 gfs2_quota_cleanup(sdp);
1323 void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1325 struct list_head *head = &sdp->sd_quota_list;
1326 struct gfs2_quota_data *qd;
1329 spin_lock(&qd_lock);
1330 while (!list_empty(head)) {
1331 qd = list_entry(head->prev, struct gfs2_quota_data, qd_list);
1334 * To be removed in due course... we should be able to
1335 * ensure that all refs to the qd have done by this point
1336 * so that this rather odd test is not required
1338 spin_lock(&qd->qd_lockref.lock);
1339 if (qd->qd_lockref.count > 1 ||
1340 (qd->qd_lockref.count && !test_bit(QDF_CHANGE, &qd->qd_flags))) {
1341 spin_unlock(&qd->qd_lockref.lock);
1342 list_move(&qd->qd_list, head);
1343 spin_unlock(&qd_lock);
1345 spin_lock(&qd_lock);
1348 spin_unlock(&qd->qd_lockref.lock);
1350 list_del(&qd->qd_list);
1351 /* Also remove if this qd exists in the reclaim list */
1352 list_lru_del(&gfs2_qd_lru, &qd->qd_lru);
1353 atomic_dec(&sdp->sd_quota_count);
1354 spin_unlock(&qd_lock);
1356 if (!qd->qd_lockref.count) {
1357 gfs2_assert_warn(sdp, !qd->qd_change);
1358 gfs2_assert_warn(sdp, !qd->qd_slot_count);
1360 gfs2_assert_warn(sdp, qd->qd_slot_count == 1);
1361 gfs2_assert_warn(sdp, !qd->qd_bh_count);
1363 gfs2_glock_put(qd->qd_gl);
1364 kmem_cache_free(gfs2_quotad_cachep, qd);
1366 spin_lock(&qd_lock);
1368 spin_unlock(&qd_lock);
1370 gfs2_assert_warn(sdp, !atomic_read(&sdp->sd_quota_count));
1372 if (sdp->sd_quota_bitmap) {
1373 for (x = 0; x < sdp->sd_quota_chunks; x++)
1374 kfree(sdp->sd_quota_bitmap[x]);
1375 kfree(sdp->sd_quota_bitmap);
1379 static void quotad_error(struct gfs2_sbd *sdp, const char *msg, int error)
1381 if (error == 0 || error == -EROFS)
1383 if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
1384 fs_err(sdp, "gfs2_quotad: %s error %d\n", msg, error);
1387 static void quotad_check_timeo(struct gfs2_sbd *sdp, const char *msg,
1388 int (*fxn)(struct super_block *sb, int type),
1389 unsigned long t, unsigned long *timeo,
1390 unsigned int *new_timeo)
1393 int error = fxn(sdp->sd_vfs, 0);
1394 quotad_error(sdp, msg, error);
1395 *timeo = gfs2_tune_get_i(&sdp->sd_tune, new_timeo) * HZ;
1401 static void quotad_check_trunc_list(struct gfs2_sbd *sdp)
1403 struct gfs2_inode *ip;
1407 spin_lock(&sdp->sd_trunc_lock);
1408 if (!list_empty(&sdp->sd_trunc_list)) {
1409 ip = list_entry(sdp->sd_trunc_list.next,
1410 struct gfs2_inode, i_trunc_list);
1411 list_del_init(&ip->i_trunc_list);
1413 spin_unlock(&sdp->sd_trunc_lock);
1416 gfs2_glock_finish_truncate(ip);
1420 void gfs2_wake_up_statfs(struct gfs2_sbd *sdp) {
1421 if (!sdp->sd_statfs_force_sync) {
1422 sdp->sd_statfs_force_sync = 1;
1423 wake_up(&sdp->sd_quota_wait);
1429 * gfs2_quotad - Write cached quota changes into the quota file
1430 * @sdp: Pointer to GFS2 superblock
1434 int gfs2_quotad(void *data)
1436 struct gfs2_sbd *sdp = data;
1437 struct gfs2_tune *tune = &sdp->sd_tune;
1438 unsigned long statfs_timeo = 0;
1439 unsigned long quotad_timeo = 0;
1440 unsigned long t = 0;
1444 while (!kthread_should_stop()) {
1446 /* Update the master statfs file */
1447 if (sdp->sd_statfs_force_sync) {
1448 int error = gfs2_statfs_sync(sdp->sd_vfs, 0);
1449 quotad_error(sdp, "statfs", error);
1450 statfs_timeo = gfs2_tune_get(sdp, gt_statfs_quantum) * HZ;
1453 quotad_check_timeo(sdp, "statfs", gfs2_statfs_sync, t,
1455 &tune->gt_statfs_quantum);
1457 /* Update quota file */
1458 quotad_check_timeo(sdp, "sync", gfs2_quota_sync, t,
1459 "ad_timeo, &tune->gt_quota_quantum);
1461 /* Check for & recover partially truncated inodes */
1462 quotad_check_trunc_list(sdp);
1466 t = min(quotad_timeo, statfs_timeo);
1468 prepare_to_wait(&sdp->sd_quota_wait, &wait, TASK_INTERRUPTIBLE);
1469 spin_lock(&sdp->sd_trunc_lock);
1470 empty = list_empty(&sdp->sd_trunc_list);
1471 spin_unlock(&sdp->sd_trunc_lock);
1472 if (empty && !sdp->sd_statfs_force_sync)
1473 t -= schedule_timeout(t);
1476 finish_wait(&sdp->sd_quota_wait, &wait);
1482 static int gfs2_quota_get_xstate(struct super_block *sb,
1483 struct fs_quota_stat *fqs)
1485 struct gfs2_sbd *sdp = sb->s_fs_info;
1487 memset(fqs, 0, sizeof(struct fs_quota_stat));
1488 fqs->qs_version = FS_QSTAT_VERSION;
1490 switch (sdp->sd_args.ar_quota) {
1492 fqs->qs_flags |= (FS_QUOTA_UDQ_ENFD | FS_QUOTA_GDQ_ENFD);
1494 case GFS2_QUOTA_ACCOUNT:
1495 fqs->qs_flags |= (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT);
1497 case GFS2_QUOTA_OFF:
1501 if (sdp->sd_quota_inode) {
1502 fqs->qs_uquota.qfs_ino = GFS2_I(sdp->sd_quota_inode)->i_no_addr;
1503 fqs->qs_uquota.qfs_nblks = sdp->sd_quota_inode->i_blocks;
1505 fqs->qs_uquota.qfs_nextents = 1; /* unsupported */
1506 fqs->qs_gquota = fqs->qs_uquota; /* its the same inode in both cases */
1507 fqs->qs_incoredqs = list_lru_count(&gfs2_qd_lru);
1511 static int gfs2_get_dqblk(struct super_block *sb, struct kqid qid,
1512 struct fs_disk_quota *fdq)
1514 struct gfs2_sbd *sdp = sb->s_fs_info;
1515 struct gfs2_quota_lvb *qlvb;
1516 struct gfs2_quota_data *qd;
1517 struct gfs2_holder q_gh;
1520 memset(fdq, 0, sizeof(struct fs_disk_quota));
1522 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1523 return -ESRCH; /* Crazy XFS error code */
1525 if ((qid.type != USRQUOTA) &&
1526 (qid.type != GRPQUOTA))
1529 error = qd_get(sdp, qid, &qd);
1532 error = do_glock(qd, FORCE, &q_gh);
1536 qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1537 fdq->d_version = FS_DQUOT_VERSION;
1538 fdq->d_flags = (qid.type == USRQUOTA) ? FS_USER_QUOTA : FS_GROUP_QUOTA;
1539 fdq->d_id = from_kqid_munged(current_user_ns(), qid);
1540 fdq->d_blk_hardlimit = be64_to_cpu(qlvb->qb_limit) << sdp->sd_fsb2bb_shift;
1541 fdq->d_blk_softlimit = be64_to_cpu(qlvb->qb_warn) << sdp->sd_fsb2bb_shift;
1542 fdq->d_bcount = be64_to_cpu(qlvb->qb_value) << sdp->sd_fsb2bb_shift;
1544 gfs2_glock_dq_uninit(&q_gh);
1550 /* GFS2 only supports a subset of the XFS fields */
1551 #define GFS2_FIELDMASK (FS_DQ_BSOFT|FS_DQ_BHARD|FS_DQ_BCOUNT)
1553 static int gfs2_set_dqblk(struct super_block *sb, struct kqid qid,
1554 struct fs_disk_quota *fdq)
1556 struct gfs2_sbd *sdp = sb->s_fs_info;
1557 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1558 struct gfs2_quota_data *qd;
1559 struct gfs2_holder q_gh, i_gh;
1560 unsigned int data_blocks, ind_blocks;
1561 unsigned int blocks = 0;
1566 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1567 return -ESRCH; /* Crazy XFS error code */
1569 if ((qid.type != USRQUOTA) &&
1570 (qid.type != GRPQUOTA))
1573 if (fdq->d_fieldmask & ~GFS2_FIELDMASK)
1576 error = qd_get(sdp, qid, &qd);
1580 error = gfs2_rs_alloc(ip);
1584 mutex_lock(&ip->i_inode.i_mutex);
1585 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE, 0, &q_gh);
1588 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1592 /* Check for existing entry, if none then alloc new blocks */
1593 error = update_qd(sdp, qd);
1597 /* If nothing has changed, this is a no-op */
1598 if ((fdq->d_fieldmask & FS_DQ_BSOFT) &&
1599 ((fdq->d_blk_softlimit >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_warn)))
1600 fdq->d_fieldmask ^= FS_DQ_BSOFT;
1602 if ((fdq->d_fieldmask & FS_DQ_BHARD) &&
1603 ((fdq->d_blk_hardlimit >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_limit)))
1604 fdq->d_fieldmask ^= FS_DQ_BHARD;
1606 if ((fdq->d_fieldmask & FS_DQ_BCOUNT) &&
1607 ((fdq->d_bcount >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_value)))
1608 fdq->d_fieldmask ^= FS_DQ_BCOUNT;
1610 if (fdq->d_fieldmask == 0)
1613 offset = qd2offset(qd);
1614 alloc_required = gfs2_write_alloc_required(ip, offset, sizeof(struct gfs2_quota));
1615 if (gfs2_is_stuffed(ip))
1617 if (alloc_required) {
1618 struct gfs2_alloc_parms ap = { .aflags = 0, };
1619 gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
1620 &data_blocks, &ind_blocks);
1621 blocks = 1 + data_blocks + ind_blocks;
1623 error = gfs2_inplace_reserve(ip, &ap);
1626 blocks += gfs2_rg_blocks(ip, blocks);
1629 /* Some quotas span block boundaries and can update two blocks,
1630 adding an extra block to the transaction to handle such quotas */
1631 error = gfs2_trans_begin(sdp, blocks + RES_DINODE + 2, 0);
1636 error = gfs2_adjust_quota(ip, offset, 0, qd, fdq);
1638 gfs2_trans_end(sdp);
1641 gfs2_inplace_release(ip);
1643 gfs2_glock_dq_uninit(&i_gh);
1645 gfs2_glock_dq_uninit(&q_gh);
1647 mutex_unlock(&ip->i_inode.i_mutex);
1653 const struct quotactl_ops gfs2_quotactl_ops = {
1654 .quota_sync = gfs2_quota_sync,
1655 .get_xstate = gfs2_quota_get_xstate,
1656 .get_dqblk = gfs2_get_dqblk,
1657 .set_dqblk = gfs2_set_dqblk,