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;
670 struct gfs2_quota q, *qp;
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));
687 qp->qu_value = be64_to_cpu(qp->qu_value);
688 qp->qu_value += change;
689 qp->qu_value = cpu_to_be64(qp->qu_value);
690 qd->qd_qb.qb_value = qp->qu_value;
692 if (fdq->d_fieldmask & FS_DQ_BSOFT) {
693 qp->qu_warn = cpu_to_be64(fdq->d_blk_softlimit >> sdp->sd_fsb2bb_shift);
694 qd->qd_qb.qb_warn = qp->qu_warn;
696 if (fdq->d_fieldmask & FS_DQ_BHARD) {
697 qp->qu_limit = cpu_to_be64(fdq->d_blk_hardlimit >> sdp->sd_fsb2bb_shift);
698 qd->qd_qb.qb_limit = qp->qu_limit;
700 if (fdq->d_fieldmask & FS_DQ_BCOUNT) {
701 qp->qu_value = cpu_to_be64(fdq->d_bcount >> sdp->sd_fsb2bb_shift);
702 qd->qd_qb.qb_value = qp->qu_value;
706 /* Write the quota into the quota file on disk */
708 nbytes = sizeof(struct gfs2_quota);
710 page = find_or_create_page(mapping, index, GFP_NOFS);
714 blocksize = inode->i_sb->s_blocksize;
715 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
717 if (!page_has_buffers(page))
718 create_empty_buffers(page, blocksize, 0);
720 bh = page_buffers(page);
722 while (offset >= pos) {
723 bh = bh->b_this_page;
728 if (!buffer_mapped(bh)) {
729 gfs2_block_map(inode, iblock, bh, 1);
730 if (!buffer_mapped(bh))
732 /* If it's a newly allocated disk block for quota, zero it */
734 zero_user(page, pos - blocksize, bh->b_size);
737 if (PageUptodate(page))
738 set_buffer_uptodate(bh);
740 if (!buffer_uptodate(bh)) {
741 ll_rw_block(READ | REQ_META, 1, &bh);
743 if (!buffer_uptodate(bh))
747 gfs2_trans_add_data(ip->i_gl, bh);
749 kaddr = kmap_atomic(page);
750 if (offset + sizeof(struct gfs2_quota) > PAGE_CACHE_SIZE)
751 nbytes = PAGE_CACHE_SIZE - offset;
752 memcpy(kaddr + offset, ptr, nbytes);
753 flush_dcache_page(page);
754 kunmap_atomic(kaddr);
756 page_cache_release(page);
758 /* If quota straddles page boundary, we need to update the rest of the
759 * quota at the beginning of the next page */
760 if ((offset + sizeof(struct gfs2_quota)) > PAGE_CACHE_SIZE) {
762 nbytes = sizeof(struct gfs2_quota) - nbytes;
768 size = loc + sizeof(struct gfs2_quota);
769 if (size > inode->i_size)
770 i_size_write(inode, size);
771 inode->i_mtime = inode->i_atime = CURRENT_TIME;
772 mark_inode_dirty(inode);
777 page_cache_release(page);
781 static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
783 struct gfs2_sbd *sdp = (*qda)->qd_gl->gl_sbd;
784 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
785 struct gfs2_alloc_parms ap = { .aflags = 0, };
786 unsigned int data_blocks, ind_blocks;
787 struct gfs2_holder *ghs, i_gh;
789 struct gfs2_quota_data *qd;
792 unsigned int nalloc = 0, blocks;
795 error = gfs2_rs_alloc(ip);
799 gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
800 &data_blocks, &ind_blocks);
802 ghs = kcalloc(num_qd, sizeof(struct gfs2_holder), GFP_NOFS);
806 sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL);
807 mutex_lock(&ip->i_inode.i_mutex);
808 for (qx = 0; qx < num_qd; qx++) {
809 error = gfs2_glock_nq_init(qda[qx]->qd_gl, LM_ST_EXCLUSIVE,
810 GL_NOCACHE, &ghs[qx]);
815 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
819 for (x = 0; x < num_qd; x++) {
820 offset = qd2offset(qda[x]);
821 if (gfs2_write_alloc_required(ip, offset,
822 sizeof(struct gfs2_quota)))
827 * 1 blk for unstuffing inode if stuffed. We add this extra
828 * block to the reservation unconditionally. If the inode
829 * doesn't need unstuffing, the block will be released to the
830 * rgrp since it won't be allocated during the transaction
832 /* +3 in the end for unstuffing block, inode size update block
833 * and another block in case quota straddles page boundary and
834 * two blocks need to be updated instead of 1 */
835 blocks = num_qd * data_blocks + RES_DINODE + num_qd + 3;
837 reserved = 1 + (nalloc * (data_blocks + ind_blocks));
838 ap.target = reserved;
839 error = gfs2_inplace_reserve(ip, &ap);
844 blocks += gfs2_rg_blocks(ip, reserved) + nalloc * ind_blocks + RES_STATFS;
846 error = gfs2_trans_begin(sdp, blocks, 0);
850 for (x = 0; x < num_qd; x++) {
852 offset = qd2offset(qd);
853 error = gfs2_adjust_quota(ip, offset, qd->qd_change_sync, qd, NULL);
857 do_qc(qd, -qd->qd_change_sync);
858 set_bit(QDF_REFRESH, &qd->qd_flags);
866 gfs2_inplace_release(ip);
868 gfs2_glock_dq_uninit(&i_gh);
871 gfs2_glock_dq_uninit(&ghs[qx]);
872 mutex_unlock(&ip->i_inode.i_mutex);
874 gfs2_log_flush(ip->i_gl->gl_sbd, ip->i_gl);
878 static int update_qd(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd)
880 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
882 struct gfs2_quota_lvb *qlvb;
886 memset(&q, 0, sizeof(struct gfs2_quota));
888 error = gfs2_internal_read(ip, (char *)&q, &pos, sizeof(q));
892 qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
893 qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC);
895 qlvb->qb_limit = q.qu_limit;
896 qlvb->qb_warn = q.qu_warn;
897 qlvb->qb_value = q.qu_value;
903 static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
904 struct gfs2_holder *q_gh)
906 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
907 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
908 struct gfs2_holder i_gh;
912 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
916 qd->qd_qb = *(struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
918 if (force_refresh || qd->qd_qb.qb_magic != cpu_to_be32(GFS2_MAGIC)) {
919 gfs2_glock_dq_uninit(q_gh);
920 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE,
925 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
929 error = update_qd(sdp, qd);
933 gfs2_glock_dq_uninit(&i_gh);
934 gfs2_glock_dq_uninit(q_gh);
942 gfs2_glock_dq_uninit(&i_gh);
944 gfs2_glock_dq_uninit(q_gh);
948 int gfs2_quota_lock(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
950 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
951 struct gfs2_quota_data *qd;
955 error = gfs2_quota_hold(ip, uid, gid);
959 if (capable(CAP_SYS_RESOURCE) ||
960 sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
963 sort(ip->i_res->rs_qa_qd, ip->i_res->rs_qa_qd_num,
964 sizeof(struct gfs2_quota_data *), sort_qd, NULL);
966 for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
967 int force = NO_FORCE;
968 qd = ip->i_res->rs_qa_qd[x];
969 if (test_and_clear_bit(QDF_REFRESH, &qd->qd_flags))
971 error = do_glock(qd, force, &ip->i_res->rs_qa_qd_ghs[x]);
977 set_bit(GIF_QD_LOCKED, &ip->i_flags);
980 gfs2_glock_dq_uninit(&ip->i_res->rs_qa_qd_ghs[x]);
981 gfs2_quota_unhold(ip);
987 static int need_sync(struct gfs2_quota_data *qd)
989 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
990 struct gfs2_tune *gt = &sdp->sd_tune;
992 unsigned int num, den;
995 if (!qd->qd_qb.qb_limit)
999 value = qd->qd_change;
1000 spin_unlock(&qd_lock);
1002 spin_lock(>->gt_spin);
1003 num = gt->gt_quota_scale_num;
1004 den = gt->gt_quota_scale_den;
1005 spin_unlock(>->gt_spin);
1009 else if ((s64)be64_to_cpu(qd->qd_qb.qb_value) >=
1010 (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1013 value *= gfs2_jindex_size(sdp) * num;
1014 value = div_s64(value, den);
1015 value += (s64)be64_to_cpu(qd->qd_qb.qb_value);
1016 if (value < (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1023 void gfs2_quota_unlock(struct gfs2_inode *ip)
1025 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1026 struct gfs2_quota_data *qda[4];
1027 unsigned int count = 0;
1031 if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
1034 for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1035 struct gfs2_quota_data *qd;
1038 qd = ip->i_res->rs_qa_qd[x];
1039 sync = need_sync(qd);
1041 gfs2_glock_dq_uninit(&ip->i_res->rs_qa_qd_ghs[x]);
1045 spin_lock(&qd_lock);
1046 found = qd_check_sync(sdp, qd, NULL);
1047 spin_unlock(&qd_lock);
1052 gfs2_assert_warn(sdp, qd->qd_change_sync);
1054 clear_bit(QDF_LOCKED, &qd->qd_flags);
1064 do_sync(count, qda);
1065 for (x = 0; x < count; x++)
1070 gfs2_quota_unhold(ip);
1073 #define MAX_LINE 256
1075 static int print_message(struct gfs2_quota_data *qd, char *type)
1077 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
1079 printk(KERN_INFO "GFS2: fsid=%s: quota %s for %s %u\n",
1080 sdp->sd_fsname, type,
1081 (qd->qd_id.type == USRQUOTA) ? "user" : "group",
1082 from_kqid(&init_user_ns, qd->qd_id));
1087 int gfs2_quota_check(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
1089 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1090 struct gfs2_quota_data *qd;
1095 if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
1098 if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
1101 for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1102 qd = ip->i_res->rs_qa_qd[x];
1104 if (!(qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1105 qid_eq(qd->qd_id, make_kqid_gid(gid))))
1108 value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
1109 spin_lock(&qd_lock);
1110 value += qd->qd_change;
1111 spin_unlock(&qd_lock);
1113 if (be64_to_cpu(qd->qd_qb.qb_limit) && (s64)be64_to_cpu(qd->qd_qb.qb_limit) < value) {
1114 print_message(qd, "exceeded");
1115 quota_send_warning(qd->qd_id,
1116 sdp->sd_vfs->s_dev, QUOTA_NL_BHARDWARN);
1120 } else if (be64_to_cpu(qd->qd_qb.qb_warn) &&
1121 (s64)be64_to_cpu(qd->qd_qb.qb_warn) < value &&
1122 time_after_eq(jiffies, qd->qd_last_warn +
1124 gt_quota_warn_period) * HZ)) {
1125 quota_send_warning(qd->qd_id,
1126 sdp->sd_vfs->s_dev, QUOTA_NL_BSOFTWARN);
1127 error = print_message(qd, "warning");
1128 qd->qd_last_warn = jiffies;
1135 void gfs2_quota_change(struct gfs2_inode *ip, s64 change,
1136 kuid_t uid, kgid_t gid)
1138 struct gfs2_quota_data *qd;
1141 if (gfs2_assert_warn(GFS2_SB(&ip->i_inode), change))
1143 if (ip->i_diskflags & GFS2_DIF_SYSTEM)
1146 for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1147 qd = ip->i_res->rs_qa_qd[x];
1149 if (qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1150 qid_eq(qd->qd_id, make_kqid_gid(gid))) {
1156 int gfs2_quota_sync(struct super_block *sb, int type)
1158 struct gfs2_sbd *sdp = sb->s_fs_info;
1159 struct gfs2_quota_data **qda;
1160 unsigned int max_qd = PAGE_SIZE/sizeof(struct gfs2_holder);
1161 unsigned int num_qd;
1165 qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
1169 mutex_lock(&sdp->sd_quota_sync_mutex);
1170 sdp->sd_quota_sync_gen++;
1176 error = qd_fish(sdp, qda + num_qd);
1177 if (error || !qda[num_qd])
1179 if (++num_qd == max_qd)
1185 error = do_sync(num_qd, qda);
1187 for (x = 0; x < num_qd; x++)
1188 qda[x]->qd_sync_gen =
1189 sdp->sd_quota_sync_gen;
1191 for (x = 0; x < num_qd; x++)
1194 } while (!error && num_qd == max_qd);
1196 mutex_unlock(&sdp->sd_quota_sync_mutex);
1202 int gfs2_quota_refresh(struct gfs2_sbd *sdp, struct kqid qid)
1204 struct gfs2_quota_data *qd;
1205 struct gfs2_holder q_gh;
1208 error = qd_get(sdp, qid, &qd);
1212 error = do_glock(qd, FORCE, &q_gh);
1214 gfs2_glock_dq_uninit(&q_gh);
1220 static void gfs2_quota_change_in(struct gfs2_quota_change_host *qc, const void *buf)
1222 const struct gfs2_quota_change *str = buf;
1224 qc->qc_change = be64_to_cpu(str->qc_change);
1225 qc->qc_flags = be32_to_cpu(str->qc_flags);
1226 qc->qc_id = make_kqid(&init_user_ns,
1227 (qc->qc_flags & GFS2_QCF_USER)?USRQUOTA:GRPQUOTA,
1228 be32_to_cpu(str->qc_id));
1231 int gfs2_quota_init(struct gfs2_sbd *sdp)
1233 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
1234 u64 size = i_size_read(sdp->sd_qc_inode);
1235 unsigned int blocks = size >> sdp->sd_sb.sb_bsize_shift;
1236 unsigned int x, slot = 0;
1237 unsigned int found = 0;
1242 if (gfs2_check_internal_file_size(sdp->sd_qc_inode, 1, 64 << 20))
1245 sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
1246 sdp->sd_quota_chunks = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * PAGE_SIZE);
1250 sdp->sd_quota_bitmap = kcalloc(sdp->sd_quota_chunks,
1251 sizeof(unsigned char *), GFP_NOFS);
1252 if (!sdp->sd_quota_bitmap)
1255 for (x = 0; x < sdp->sd_quota_chunks; x++) {
1256 sdp->sd_quota_bitmap[x] = kzalloc(PAGE_SIZE, GFP_NOFS);
1257 if (!sdp->sd_quota_bitmap[x])
1261 for (x = 0; x < blocks; x++) {
1262 struct buffer_head *bh;
1267 error = gfs2_extent_map(&ip->i_inode, x, &new, &dblock, &extlen);
1272 bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
1275 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
1280 for (y = 0; y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1282 struct gfs2_quota_change_host qc;
1283 struct gfs2_quota_data *qd;
1285 gfs2_quota_change_in(&qc, bh->b_data +
1286 sizeof(struct gfs2_meta_header) +
1287 y * sizeof(struct gfs2_quota_change));
1291 error = qd_alloc(sdp, qc.qc_id, &qd);
1297 set_bit(QDF_CHANGE, &qd->qd_flags);
1298 qd->qd_change = qc.qc_change;
1300 qd->qd_slot_count = 1;
1302 spin_lock(&qd_lock);
1303 gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, slot, 1);
1304 list_add(&qd->qd_list, &sdp->sd_quota_list);
1305 atomic_inc(&sdp->sd_quota_count);
1306 spin_unlock(&qd_lock);
1317 fs_info(sdp, "found %u quota changes\n", found);
1322 gfs2_quota_cleanup(sdp);
1326 void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1328 struct list_head *head = &sdp->sd_quota_list;
1329 struct gfs2_quota_data *qd;
1332 spin_lock(&qd_lock);
1333 while (!list_empty(head)) {
1334 qd = list_entry(head->prev, struct gfs2_quota_data, qd_list);
1337 * To be removed in due course... we should be able to
1338 * ensure that all refs to the qd have done by this point
1339 * so that this rather odd test is not required
1341 spin_lock(&qd->qd_lockref.lock);
1342 if (qd->qd_lockref.count > 1 ||
1343 (qd->qd_lockref.count && !test_bit(QDF_CHANGE, &qd->qd_flags))) {
1344 spin_unlock(&qd->qd_lockref.lock);
1345 list_move(&qd->qd_list, head);
1346 spin_unlock(&qd_lock);
1348 spin_lock(&qd_lock);
1351 spin_unlock(&qd->qd_lockref.lock);
1353 list_del(&qd->qd_list);
1354 /* Also remove if this qd exists in the reclaim list */
1355 list_lru_del(&gfs2_qd_lru, &qd->qd_lru);
1356 atomic_dec(&sdp->sd_quota_count);
1357 spin_unlock(&qd_lock);
1359 if (!qd->qd_lockref.count) {
1360 gfs2_assert_warn(sdp, !qd->qd_change);
1361 gfs2_assert_warn(sdp, !qd->qd_slot_count);
1363 gfs2_assert_warn(sdp, qd->qd_slot_count == 1);
1364 gfs2_assert_warn(sdp, !qd->qd_bh_count);
1366 gfs2_glock_put(qd->qd_gl);
1367 kmem_cache_free(gfs2_quotad_cachep, qd);
1369 spin_lock(&qd_lock);
1371 spin_unlock(&qd_lock);
1373 gfs2_assert_warn(sdp, !atomic_read(&sdp->sd_quota_count));
1375 if (sdp->sd_quota_bitmap) {
1376 for (x = 0; x < sdp->sd_quota_chunks; x++)
1377 kfree(sdp->sd_quota_bitmap[x]);
1378 kfree(sdp->sd_quota_bitmap);
1382 static void quotad_error(struct gfs2_sbd *sdp, const char *msg, int error)
1384 if (error == 0 || error == -EROFS)
1386 if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
1387 fs_err(sdp, "gfs2_quotad: %s error %d\n", msg, error);
1390 static void quotad_check_timeo(struct gfs2_sbd *sdp, const char *msg,
1391 int (*fxn)(struct super_block *sb, int type),
1392 unsigned long t, unsigned long *timeo,
1393 unsigned int *new_timeo)
1396 int error = fxn(sdp->sd_vfs, 0);
1397 quotad_error(sdp, msg, error);
1398 *timeo = gfs2_tune_get_i(&sdp->sd_tune, new_timeo) * HZ;
1404 static void quotad_check_trunc_list(struct gfs2_sbd *sdp)
1406 struct gfs2_inode *ip;
1410 spin_lock(&sdp->sd_trunc_lock);
1411 if (!list_empty(&sdp->sd_trunc_list)) {
1412 ip = list_entry(sdp->sd_trunc_list.next,
1413 struct gfs2_inode, i_trunc_list);
1414 list_del_init(&ip->i_trunc_list);
1416 spin_unlock(&sdp->sd_trunc_lock);
1419 gfs2_glock_finish_truncate(ip);
1423 void gfs2_wake_up_statfs(struct gfs2_sbd *sdp) {
1424 if (!sdp->sd_statfs_force_sync) {
1425 sdp->sd_statfs_force_sync = 1;
1426 wake_up(&sdp->sd_quota_wait);
1432 * gfs2_quotad - Write cached quota changes into the quota file
1433 * @sdp: Pointer to GFS2 superblock
1437 int gfs2_quotad(void *data)
1439 struct gfs2_sbd *sdp = data;
1440 struct gfs2_tune *tune = &sdp->sd_tune;
1441 unsigned long statfs_timeo = 0;
1442 unsigned long quotad_timeo = 0;
1443 unsigned long t = 0;
1447 while (!kthread_should_stop()) {
1449 /* Update the master statfs file */
1450 if (sdp->sd_statfs_force_sync) {
1451 int error = gfs2_statfs_sync(sdp->sd_vfs, 0);
1452 quotad_error(sdp, "statfs", error);
1453 statfs_timeo = gfs2_tune_get(sdp, gt_statfs_quantum) * HZ;
1456 quotad_check_timeo(sdp, "statfs", gfs2_statfs_sync, t,
1458 &tune->gt_statfs_quantum);
1460 /* Update quota file */
1461 quotad_check_timeo(sdp, "sync", gfs2_quota_sync, t,
1462 "ad_timeo, &tune->gt_quota_quantum);
1464 /* Check for & recover partially truncated inodes */
1465 quotad_check_trunc_list(sdp);
1469 t = min(quotad_timeo, statfs_timeo);
1471 prepare_to_wait(&sdp->sd_quota_wait, &wait, TASK_INTERRUPTIBLE);
1472 spin_lock(&sdp->sd_trunc_lock);
1473 empty = list_empty(&sdp->sd_trunc_list);
1474 spin_unlock(&sdp->sd_trunc_lock);
1475 if (empty && !sdp->sd_statfs_force_sync)
1476 t -= schedule_timeout(t);
1479 finish_wait(&sdp->sd_quota_wait, &wait);
1485 static int gfs2_quota_get_xstate(struct super_block *sb,
1486 struct fs_quota_stat *fqs)
1488 struct gfs2_sbd *sdp = sb->s_fs_info;
1490 memset(fqs, 0, sizeof(struct fs_quota_stat));
1491 fqs->qs_version = FS_QSTAT_VERSION;
1493 switch (sdp->sd_args.ar_quota) {
1495 fqs->qs_flags |= (FS_QUOTA_UDQ_ENFD | FS_QUOTA_GDQ_ENFD);
1497 case GFS2_QUOTA_ACCOUNT:
1498 fqs->qs_flags |= (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT);
1500 case GFS2_QUOTA_OFF:
1504 if (sdp->sd_quota_inode) {
1505 fqs->qs_uquota.qfs_ino = GFS2_I(sdp->sd_quota_inode)->i_no_addr;
1506 fqs->qs_uquota.qfs_nblks = sdp->sd_quota_inode->i_blocks;
1508 fqs->qs_uquota.qfs_nextents = 1; /* unsupported */
1509 fqs->qs_gquota = fqs->qs_uquota; /* its the same inode in both cases */
1510 fqs->qs_incoredqs = list_lru_count(&gfs2_qd_lru);
1514 static int gfs2_get_dqblk(struct super_block *sb, struct kqid qid,
1515 struct fs_disk_quota *fdq)
1517 struct gfs2_sbd *sdp = sb->s_fs_info;
1518 struct gfs2_quota_lvb *qlvb;
1519 struct gfs2_quota_data *qd;
1520 struct gfs2_holder q_gh;
1523 memset(fdq, 0, sizeof(struct fs_disk_quota));
1525 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1526 return -ESRCH; /* Crazy XFS error code */
1528 if ((qid.type != USRQUOTA) &&
1529 (qid.type != GRPQUOTA))
1532 error = qd_get(sdp, qid, &qd);
1535 error = do_glock(qd, FORCE, &q_gh);
1539 qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1540 fdq->d_version = FS_DQUOT_VERSION;
1541 fdq->d_flags = (qid.type == USRQUOTA) ? FS_USER_QUOTA : FS_GROUP_QUOTA;
1542 fdq->d_id = from_kqid_munged(current_user_ns(), qid);
1543 fdq->d_blk_hardlimit = be64_to_cpu(qlvb->qb_limit) << sdp->sd_fsb2bb_shift;
1544 fdq->d_blk_softlimit = be64_to_cpu(qlvb->qb_warn) << sdp->sd_fsb2bb_shift;
1545 fdq->d_bcount = be64_to_cpu(qlvb->qb_value) << sdp->sd_fsb2bb_shift;
1547 gfs2_glock_dq_uninit(&q_gh);
1553 /* GFS2 only supports a subset of the XFS fields */
1554 #define GFS2_FIELDMASK (FS_DQ_BSOFT|FS_DQ_BHARD|FS_DQ_BCOUNT)
1556 static int gfs2_set_dqblk(struct super_block *sb, struct kqid qid,
1557 struct fs_disk_quota *fdq)
1559 struct gfs2_sbd *sdp = sb->s_fs_info;
1560 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1561 struct gfs2_quota_data *qd;
1562 struct gfs2_holder q_gh, i_gh;
1563 unsigned int data_blocks, ind_blocks;
1564 unsigned int blocks = 0;
1569 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1570 return -ESRCH; /* Crazy XFS error code */
1572 if ((qid.type != USRQUOTA) &&
1573 (qid.type != GRPQUOTA))
1576 if (fdq->d_fieldmask & ~GFS2_FIELDMASK)
1579 error = qd_get(sdp, qid, &qd);
1583 error = gfs2_rs_alloc(ip);
1587 mutex_lock(&ip->i_inode.i_mutex);
1588 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE, 0, &q_gh);
1591 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1595 /* Check for existing entry, if none then alloc new blocks */
1596 error = update_qd(sdp, qd);
1600 /* If nothing has changed, this is a no-op */
1601 if ((fdq->d_fieldmask & FS_DQ_BSOFT) &&
1602 ((fdq->d_blk_softlimit >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_warn)))
1603 fdq->d_fieldmask ^= FS_DQ_BSOFT;
1605 if ((fdq->d_fieldmask & FS_DQ_BHARD) &&
1606 ((fdq->d_blk_hardlimit >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_limit)))
1607 fdq->d_fieldmask ^= FS_DQ_BHARD;
1609 if ((fdq->d_fieldmask & FS_DQ_BCOUNT) &&
1610 ((fdq->d_bcount >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_value)))
1611 fdq->d_fieldmask ^= FS_DQ_BCOUNT;
1613 if (fdq->d_fieldmask == 0)
1616 offset = qd2offset(qd);
1617 alloc_required = gfs2_write_alloc_required(ip, offset, sizeof(struct gfs2_quota));
1618 if (gfs2_is_stuffed(ip))
1620 if (alloc_required) {
1621 struct gfs2_alloc_parms ap = { .aflags = 0, };
1622 gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
1623 &data_blocks, &ind_blocks);
1624 blocks = 1 + data_blocks + ind_blocks;
1626 error = gfs2_inplace_reserve(ip, &ap);
1629 blocks += gfs2_rg_blocks(ip, blocks);
1632 /* Some quotas span block boundaries and can update two blocks,
1633 adding an extra block to the transaction to handle such quotas */
1634 error = gfs2_trans_begin(sdp, blocks + RES_DINODE + 2, 0);
1639 error = gfs2_adjust_quota(ip, offset, 0, qd, fdq);
1641 gfs2_trans_end(sdp);
1644 gfs2_inplace_release(ip);
1646 gfs2_glock_dq_uninit(&i_gh);
1648 gfs2_glock_dq_uninit(&q_gh);
1650 mutex_unlock(&ip->i_inode.i_mutex);
1656 const struct quotactl_ops gfs2_quotactl_ops = {
1657 .quota_sync = gfs2_quota_sync,
1658 .get_xstate = gfs2_quota_get_xstate,
1659 .get_dqblk = gfs2_get_dqblk,
1660 .set_dqblk = gfs2_set_dqblk,