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>
71 struct gfs2_quota_change_host {
73 u32 qc_flags; /* GFS2_QCF_... */
77 static LIST_HEAD(qd_lru_list);
78 static atomic_t qd_lru_count = ATOMIC_INIT(0);
79 static DEFINE_SPINLOCK(qd_lru_lock);
81 int gfs2_shrink_qd_memory(struct shrinker *shrink, struct shrink_control *sc)
83 struct gfs2_quota_data *qd;
85 int nr_to_scan = sc->nr_to_scan;
90 if (!(sc->gfp_mask & __GFP_FS))
93 spin_lock(&qd_lru_lock);
94 while (nr_to_scan && !list_empty(&qd_lru_list)) {
95 qd = list_entry(qd_lru_list.next,
96 struct gfs2_quota_data, qd_reclaim);
97 sdp = qd->qd_gl->gl_sbd;
99 /* Free from the filesystem-specific list */
100 list_del(&qd->qd_list);
102 gfs2_assert_warn(sdp, !qd->qd_change);
103 gfs2_assert_warn(sdp, !qd->qd_slot_count);
104 gfs2_assert_warn(sdp, !qd->qd_bh_count);
106 gfs2_glock_put(qd->qd_gl);
107 atomic_dec(&sdp->sd_quota_count);
109 /* Delete it from the common reclaim list */
110 list_del_init(&qd->qd_reclaim);
111 atomic_dec(&qd_lru_count);
112 spin_unlock(&qd_lru_lock);
113 kmem_cache_free(gfs2_quotad_cachep, qd);
114 spin_lock(&qd_lru_lock);
117 spin_unlock(&qd_lru_lock);
120 return (atomic_read(&qd_lru_count) * sysctl_vfs_cache_pressure) / 100;
123 static u64 qd2offset(struct gfs2_quota_data *qd)
127 offset = 2 * (u64)qd->qd_id + !test_bit(QDF_USER, &qd->qd_flags);
128 offset *= sizeof(struct gfs2_quota);
133 static int qd_alloc(struct gfs2_sbd *sdp, int user, u32 id,
134 struct gfs2_quota_data **qdp)
136 struct gfs2_quota_data *qd;
139 qd = kmem_cache_zalloc(gfs2_quotad_cachep, GFP_NOFS);
143 atomic_set(&qd->qd_count, 1);
146 set_bit(QDF_USER, &qd->qd_flags);
148 INIT_LIST_HEAD(&qd->qd_reclaim);
150 error = gfs2_glock_get(sdp, 2 * (u64)id + !user,
151 &gfs2_quota_glops, CREATE, &qd->qd_gl);
160 kmem_cache_free(gfs2_quotad_cachep, qd);
164 static int qd_get(struct gfs2_sbd *sdp, int user, u32 id,
165 struct gfs2_quota_data **qdp)
167 struct gfs2_quota_data *qd = NULL, *new_qd = NULL;
174 spin_lock(&qd_lru_lock);
175 list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
176 if (qd->qd_id == id &&
177 !test_bit(QDF_USER, &qd->qd_flags) == !user) {
178 if (!atomic_read(&qd->qd_count) &&
179 !list_empty(&qd->qd_reclaim)) {
180 /* Remove it from reclaim list */
181 list_del_init(&qd->qd_reclaim);
182 atomic_dec(&qd_lru_count);
184 atomic_inc(&qd->qd_count);
195 list_add(&qd->qd_list, &sdp->sd_quota_list);
196 atomic_inc(&sdp->sd_quota_count);
200 spin_unlock(&qd_lru_lock);
204 gfs2_glock_put(new_qd->qd_gl);
205 kmem_cache_free(gfs2_quotad_cachep, new_qd);
211 error = qd_alloc(sdp, user, id, &new_qd);
217 static void qd_hold(struct gfs2_quota_data *qd)
219 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
220 gfs2_assert(sdp, atomic_read(&qd->qd_count));
221 atomic_inc(&qd->qd_count);
224 static void qd_put(struct gfs2_quota_data *qd)
226 if (atomic_dec_and_lock(&qd->qd_count, &qd_lru_lock)) {
227 /* Add to the reclaim list */
228 list_add_tail(&qd->qd_reclaim, &qd_lru_list);
229 atomic_inc(&qd_lru_count);
230 spin_unlock(&qd_lru_lock);
234 static int slot_get(struct gfs2_quota_data *qd)
236 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
237 unsigned int c, o = 0, b;
238 unsigned char byte = 0;
240 spin_lock(&qd_lru_lock);
242 if (qd->qd_slot_count++) {
243 spin_unlock(&qd_lru_lock);
247 for (c = 0; c < sdp->sd_quota_chunks; c++)
248 for (o = 0; o < PAGE_SIZE; o++) {
249 byte = sdp->sd_quota_bitmap[c][o];
257 for (b = 0; b < 8; b++)
258 if (!(byte & (1 << b)))
260 qd->qd_slot = c * (8 * PAGE_SIZE) + o * 8 + b;
262 if (qd->qd_slot >= sdp->sd_quota_slots)
265 sdp->sd_quota_bitmap[c][o] |= 1 << b;
267 spin_unlock(&qd_lru_lock);
273 spin_unlock(&qd_lru_lock);
277 static void slot_hold(struct gfs2_quota_data *qd)
279 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
281 spin_lock(&qd_lru_lock);
282 gfs2_assert(sdp, qd->qd_slot_count);
284 spin_unlock(&qd_lru_lock);
287 static void slot_put(struct gfs2_quota_data *qd)
289 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
291 spin_lock(&qd_lru_lock);
292 gfs2_assert(sdp, qd->qd_slot_count);
293 if (!--qd->qd_slot_count) {
294 gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, qd->qd_slot, 0);
297 spin_unlock(&qd_lru_lock);
300 static int bh_get(struct gfs2_quota_data *qd)
302 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
303 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
304 unsigned int block, offset;
305 struct buffer_head *bh;
307 struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
309 mutex_lock(&sdp->sd_quota_mutex);
311 if (qd->qd_bh_count++) {
312 mutex_unlock(&sdp->sd_quota_mutex);
316 block = qd->qd_slot / sdp->sd_qc_per_block;
317 offset = qd->qd_slot % sdp->sd_qc_per_block;
319 bh_map.b_size = 1 << ip->i_inode.i_blkbits;
320 error = gfs2_block_map(&ip->i_inode, block, &bh_map, 0);
323 error = gfs2_meta_read(ip->i_gl, bh_map.b_blocknr, DIO_WAIT, &bh);
327 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC))
331 qd->qd_bh_qc = (struct gfs2_quota_change *)
332 (bh->b_data + sizeof(struct gfs2_meta_header) +
333 offset * sizeof(struct gfs2_quota_change));
335 mutex_unlock(&sdp->sd_quota_mutex);
343 mutex_unlock(&sdp->sd_quota_mutex);
347 static void bh_put(struct gfs2_quota_data *qd)
349 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
351 mutex_lock(&sdp->sd_quota_mutex);
352 gfs2_assert(sdp, qd->qd_bh_count);
353 if (!--qd->qd_bh_count) {
358 mutex_unlock(&sdp->sd_quota_mutex);
361 static int qd_fish(struct gfs2_sbd *sdp, struct gfs2_quota_data **qdp)
363 struct gfs2_quota_data *qd = NULL;
369 if (sdp->sd_vfs->s_flags & MS_RDONLY)
372 spin_lock(&qd_lru_lock);
374 list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
375 if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
376 !test_bit(QDF_CHANGE, &qd->qd_flags) ||
377 qd->qd_sync_gen >= sdp->sd_quota_sync_gen)
380 list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
382 set_bit(QDF_LOCKED, &qd->qd_flags);
383 gfs2_assert_warn(sdp, atomic_read(&qd->qd_count));
384 atomic_inc(&qd->qd_count);
385 qd->qd_change_sync = qd->qd_change;
386 gfs2_assert_warn(sdp, qd->qd_slot_count);
396 spin_unlock(&qd_lru_lock);
399 gfs2_assert_warn(sdp, qd->qd_change_sync);
402 clear_bit(QDF_LOCKED, &qd->qd_flags);
414 static int qd_trylock(struct gfs2_quota_data *qd)
416 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
418 if (sdp->sd_vfs->s_flags & MS_RDONLY)
421 spin_lock(&qd_lru_lock);
423 if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
424 !test_bit(QDF_CHANGE, &qd->qd_flags)) {
425 spin_unlock(&qd_lru_lock);
429 list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
431 set_bit(QDF_LOCKED, &qd->qd_flags);
432 gfs2_assert_warn(sdp, atomic_read(&qd->qd_count));
433 atomic_inc(&qd->qd_count);
434 qd->qd_change_sync = qd->qd_change;
435 gfs2_assert_warn(sdp, qd->qd_slot_count);
438 spin_unlock(&qd_lru_lock);
440 gfs2_assert_warn(sdp, qd->qd_change_sync);
442 clear_bit(QDF_LOCKED, &qd->qd_flags);
451 static void qd_unlock(struct gfs2_quota_data *qd)
453 gfs2_assert_warn(qd->qd_gl->gl_sbd,
454 test_bit(QDF_LOCKED, &qd->qd_flags));
455 clear_bit(QDF_LOCKED, &qd->qd_flags);
461 static int qdsb_get(struct gfs2_sbd *sdp, int user, u32 id,
462 struct gfs2_quota_data **qdp)
466 error = qd_get(sdp, user, id, qdp);
470 error = slot_get(*qdp);
474 error = bh_get(*qdp);
487 static void qdsb_put(struct gfs2_quota_data *qd)
494 int gfs2_quota_hold(struct gfs2_inode *ip, u32 uid, u32 gid)
496 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
497 struct gfs2_quota_data **qd;
500 if (ip->i_res == NULL)
503 qd = ip->i_res->rs_qa_qd;
505 if (gfs2_assert_warn(sdp, !ip->i_res->rs_qa_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, qd);
515 ip->i_res->rs_qa_qd_num++;
518 error = qdsb_get(sdp, QUOTA_GROUP, ip->i_inode.i_gid, qd);
521 ip->i_res->rs_qa_qd_num++;
524 if (uid != NO_QUOTA_CHANGE && uid != ip->i_inode.i_uid) {
525 error = qdsb_get(sdp, QUOTA_USER, uid, qd);
528 ip->i_res->rs_qa_qd_num++;
532 if (gid != NO_QUOTA_CHANGE && gid != ip->i_inode.i_gid) {
533 error = qdsb_get(sdp, QUOTA_GROUP, gid, qd);
536 ip->i_res->rs_qa_qd_num++;
542 gfs2_quota_unhold(ip);
546 void gfs2_quota_unhold(struct gfs2_inode *ip)
548 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
551 if (ip->i_res == NULL)
553 gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags));
555 for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
556 qdsb_put(ip->i_res->rs_qa_qd[x]);
557 ip->i_res->rs_qa_qd[x] = NULL;
559 ip->i_res->rs_qa_qd_num = 0;
562 static int sort_qd(const void *a, const void *b)
564 const struct gfs2_quota_data *qd_a = *(const struct gfs2_quota_data **)a;
565 const struct gfs2_quota_data *qd_b = *(const struct gfs2_quota_data **)b;
567 if (!test_bit(QDF_USER, &qd_a->qd_flags) !=
568 !test_bit(QDF_USER, &qd_b->qd_flags)) {
569 if (test_bit(QDF_USER, &qd_a->qd_flags))
574 if (qd_a->qd_id < qd_b->qd_id)
576 if (qd_a->qd_id > qd_b->qd_id)
582 static void do_qc(struct gfs2_quota_data *qd, s64 change)
584 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
585 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
586 struct gfs2_quota_change *qc = qd->qd_bh_qc;
589 mutex_lock(&sdp->sd_quota_mutex);
590 gfs2_trans_add_bh(ip->i_gl, qd->qd_bh, 1);
592 if (!test_bit(QDF_CHANGE, &qd->qd_flags)) {
595 if (test_bit(QDF_USER, &qd->qd_flags))
596 qc->qc_flags = cpu_to_be32(GFS2_QCF_USER);
597 qc->qc_id = cpu_to_be32(qd->qd_id);
600 x = be64_to_cpu(qc->qc_change) + change;
601 qc->qc_change = cpu_to_be64(x);
603 spin_lock(&qd_lru_lock);
605 spin_unlock(&qd_lru_lock);
608 gfs2_assert_warn(sdp, test_bit(QDF_CHANGE, &qd->qd_flags));
609 clear_bit(QDF_CHANGE, &qd->qd_flags);
614 } else if (!test_and_set_bit(QDF_CHANGE, &qd->qd_flags)) {
619 mutex_unlock(&sdp->sd_quota_mutex);
623 * gfs2_adjust_quota - adjust record of current block usage
624 * @ip: The quota inode
625 * @loc: Offset of the entry in the quota file
626 * @change: The amount of usage change to record
627 * @qd: The quota data
628 * @fdq: The updated limits to record
630 * This function was mostly borrowed from gfs2_block_truncate_page which was
631 * in turn mostly borrowed from ext3
633 * Returns: 0 or -ve on error
636 static int gfs2_adjust_quota(struct gfs2_inode *ip, loff_t loc,
637 s64 change, struct gfs2_quota_data *qd,
638 struct fs_disk_quota *fdq)
640 struct inode *inode = &ip->i_inode;
641 struct gfs2_sbd *sdp = GFS2_SB(inode);
642 struct address_space *mapping = inode->i_mapping;
643 unsigned long index = loc >> PAGE_CACHE_SHIFT;
644 unsigned offset = loc & (PAGE_CACHE_SIZE - 1);
645 unsigned blocksize, iblock, pos;
646 struct buffer_head *bh;
649 struct gfs2_quota q, *qp;
653 if (gfs2_is_stuffed(ip)) {
654 err = gfs2_unstuff_dinode(ip, NULL);
659 memset(&q, 0, sizeof(struct gfs2_quota));
660 err = gfs2_internal_read(ip, (char *)&q, &loc, sizeof(q));
666 qp->qu_value = be64_to_cpu(qp->qu_value);
667 qp->qu_value += change;
668 qp->qu_value = cpu_to_be64(qp->qu_value);
669 qd->qd_qb.qb_value = qp->qu_value;
671 if (fdq->d_fieldmask & FS_DQ_BSOFT) {
672 qp->qu_warn = cpu_to_be64(fdq->d_blk_softlimit >> sdp->sd_fsb2bb_shift);
673 qd->qd_qb.qb_warn = qp->qu_warn;
675 if (fdq->d_fieldmask & FS_DQ_BHARD) {
676 qp->qu_limit = cpu_to_be64(fdq->d_blk_hardlimit >> sdp->sd_fsb2bb_shift);
677 qd->qd_qb.qb_limit = qp->qu_limit;
679 if (fdq->d_fieldmask & FS_DQ_BCOUNT) {
680 qp->qu_value = cpu_to_be64(fdq->d_bcount >> sdp->sd_fsb2bb_shift);
681 qd->qd_qb.qb_value = qp->qu_value;
685 /* Write the quota into the quota file on disk */
687 nbytes = sizeof(struct gfs2_quota);
689 page = find_or_create_page(mapping, index, GFP_NOFS);
693 blocksize = inode->i_sb->s_blocksize;
694 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
696 if (!page_has_buffers(page))
697 create_empty_buffers(page, blocksize, 0);
699 bh = page_buffers(page);
701 while (offset >= pos) {
702 bh = bh->b_this_page;
707 if (!buffer_mapped(bh)) {
708 gfs2_block_map(inode, iblock, bh, 1);
709 if (!buffer_mapped(bh))
711 /* If it's a newly allocated disk block for quota, zero it */
713 zero_user(page, pos - blocksize, bh->b_size);
716 if (PageUptodate(page))
717 set_buffer_uptodate(bh);
719 if (!buffer_uptodate(bh)) {
720 ll_rw_block(READ | REQ_META, 1, &bh);
722 if (!buffer_uptodate(bh))
726 gfs2_trans_add_bh(ip->i_gl, bh, 0);
728 kaddr = kmap_atomic(page);
729 if (offset + sizeof(struct gfs2_quota) > PAGE_CACHE_SIZE)
730 nbytes = PAGE_CACHE_SIZE - offset;
731 memcpy(kaddr + offset, ptr, nbytes);
732 flush_dcache_page(page);
733 kunmap_atomic(kaddr);
735 page_cache_release(page);
737 /* If quota straddles page boundary, we need to update the rest of the
738 * quota at the beginning of the next page */
739 if ((offset + sizeof(struct gfs2_quota)) > PAGE_CACHE_SIZE) {
741 nbytes = sizeof(struct gfs2_quota) - nbytes;
747 size = loc + sizeof(struct gfs2_quota);
748 if (size > inode->i_size)
749 i_size_write(inode, size);
750 inode->i_mtime = inode->i_atime = CURRENT_TIME;
751 mark_inode_dirty(inode);
756 page_cache_release(page);
760 static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
762 struct gfs2_sbd *sdp = (*qda)->qd_gl->gl_sbd;
763 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
764 unsigned int data_blocks, ind_blocks;
765 struct gfs2_holder *ghs, i_gh;
767 struct gfs2_quota_data *qd;
769 unsigned int nalloc = 0, blocks;
772 error = gfs2_rs_alloc(ip);
776 gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
777 &data_blocks, &ind_blocks);
779 ghs = kcalloc(num_qd, sizeof(struct gfs2_holder), GFP_NOFS);
783 sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL);
784 mutex_lock_nested(&ip->i_inode.i_mutex, I_MUTEX_QUOTA);
785 for (qx = 0; qx < num_qd; qx++) {
786 error = gfs2_glock_nq_init(qda[qx]->qd_gl, LM_ST_EXCLUSIVE,
787 GL_NOCACHE, &ghs[qx]);
792 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
796 for (x = 0; x < num_qd; x++) {
797 offset = qd2offset(qda[x]);
798 if (gfs2_write_alloc_required(ip, offset,
799 sizeof(struct gfs2_quota)))
804 * 1 blk for unstuffing inode if stuffed. We add this extra
805 * block to the reservation unconditionally. If the inode
806 * doesn't need unstuffing, the block will be released to the
807 * rgrp since it won't be allocated during the transaction
809 /* +3 in the end for unstuffing block, inode size update block
810 * and another block in case quota straddles page boundary and
811 * two blocks need to be updated instead of 1 */
812 blocks = num_qd * data_blocks + RES_DINODE + num_qd + 3;
814 error = gfs2_inplace_reserve(ip, 1 +
815 (nalloc * (data_blocks + ind_blocks)));
820 blocks += gfs2_rg_blocks(ip) + nalloc * ind_blocks + RES_STATFS;
822 error = gfs2_trans_begin(sdp, blocks, 0);
826 for (x = 0; x < num_qd; x++) {
828 offset = qd2offset(qd);
829 error = gfs2_adjust_quota(ip, offset, qd->qd_change_sync, qd, NULL);
833 do_qc(qd, -qd->qd_change_sync);
834 set_bit(QDF_REFRESH, &qd->qd_flags);
842 gfs2_inplace_release(ip);
844 gfs2_glock_dq_uninit(&i_gh);
847 gfs2_glock_dq_uninit(&ghs[qx]);
848 mutex_unlock(&ip->i_inode.i_mutex);
850 gfs2_log_flush(ip->i_gl->gl_sbd, ip->i_gl);
854 static int update_qd(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd)
856 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
858 struct gfs2_quota_lvb *qlvb;
862 memset(&q, 0, sizeof(struct gfs2_quota));
864 error = gfs2_internal_read(ip, (char *)&q, &pos, sizeof(q));
868 qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lvb;
869 qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC);
871 qlvb->qb_limit = q.qu_limit;
872 qlvb->qb_warn = q.qu_warn;
873 qlvb->qb_value = q.qu_value;
879 static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
880 struct gfs2_holder *q_gh)
882 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
883 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
884 struct gfs2_holder i_gh;
888 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
892 qd->qd_qb = *(struct gfs2_quota_lvb *)qd->qd_gl->gl_lvb;
894 if (force_refresh || qd->qd_qb.qb_magic != cpu_to_be32(GFS2_MAGIC)) {
895 gfs2_glock_dq_uninit(q_gh);
896 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE,
901 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
905 error = update_qd(sdp, qd);
909 gfs2_glock_dq_uninit(&i_gh);
910 gfs2_glock_dq_uninit(q_gh);
918 gfs2_glock_dq_uninit(&i_gh);
920 gfs2_glock_dq_uninit(q_gh);
924 int gfs2_quota_lock(struct gfs2_inode *ip, u32 uid, u32 gid)
926 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
927 struct gfs2_quota_data *qd;
931 error = gfs2_quota_hold(ip, uid, gid);
935 if (capable(CAP_SYS_RESOURCE) ||
936 sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
939 sort(ip->i_res->rs_qa_qd, ip->i_res->rs_qa_qd_num,
940 sizeof(struct gfs2_quota_data *), sort_qd, NULL);
942 for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
943 int force = NO_FORCE;
944 qd = ip->i_res->rs_qa_qd[x];
945 if (test_and_clear_bit(QDF_REFRESH, &qd->qd_flags))
947 error = do_glock(qd, force, &ip->i_res->rs_qa_qd_ghs[x]);
953 set_bit(GIF_QD_LOCKED, &ip->i_flags);
956 gfs2_glock_dq_uninit(&ip->i_res->rs_qa_qd_ghs[x]);
957 gfs2_quota_unhold(ip);
963 static int need_sync(struct gfs2_quota_data *qd)
965 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
966 struct gfs2_tune *gt = &sdp->sd_tune;
968 unsigned int num, den;
971 if (!qd->qd_qb.qb_limit)
974 spin_lock(&qd_lru_lock);
975 value = qd->qd_change;
976 spin_unlock(&qd_lru_lock);
978 spin_lock(>->gt_spin);
979 num = gt->gt_quota_scale_num;
980 den = gt->gt_quota_scale_den;
981 spin_unlock(>->gt_spin);
985 else if ((s64)be64_to_cpu(qd->qd_qb.qb_value) >=
986 (s64)be64_to_cpu(qd->qd_qb.qb_limit))
989 value *= gfs2_jindex_size(sdp) * num;
990 value = div_s64(value, den);
991 value += (s64)be64_to_cpu(qd->qd_qb.qb_value);
992 if (value < (s64)be64_to_cpu(qd->qd_qb.qb_limit))
999 void gfs2_quota_unlock(struct gfs2_inode *ip)
1001 struct gfs2_quota_data *qda[4];
1002 unsigned int count = 0;
1005 if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
1008 for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1009 struct gfs2_quota_data *qd;
1012 qd = ip->i_res->rs_qa_qd[x];
1013 sync = need_sync(qd);
1015 gfs2_glock_dq_uninit(&ip->i_res->rs_qa_qd_ghs[x]);
1017 if (sync && qd_trylock(qd))
1022 do_sync(count, qda);
1023 for (x = 0; x < count; x++)
1028 gfs2_quota_unhold(ip);
1031 #define MAX_LINE 256
1033 static int print_message(struct gfs2_quota_data *qd, char *type)
1035 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
1037 printk(KERN_INFO "GFS2: fsid=%s: quota %s for %s %u\n",
1038 sdp->sd_fsname, type,
1039 (test_bit(QDF_USER, &qd->qd_flags)) ? "user" : "group",
1045 int gfs2_quota_check(struct gfs2_inode *ip, u32 uid, u32 gid)
1047 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1048 struct gfs2_quota_data *qd;
1053 if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
1056 if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
1059 for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1060 qd = ip->i_res->rs_qa_qd[x];
1062 if (!((qd->qd_id == uid && test_bit(QDF_USER, &qd->qd_flags)) ||
1063 (qd->qd_id == gid && !test_bit(QDF_USER, &qd->qd_flags))))
1066 value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
1067 spin_lock(&qd_lru_lock);
1068 value += qd->qd_change;
1069 spin_unlock(&qd_lru_lock);
1071 if (be64_to_cpu(qd->qd_qb.qb_limit) && (s64)be64_to_cpu(qd->qd_qb.qb_limit) < value) {
1072 print_message(qd, "exceeded");
1073 quota_send_warning(test_bit(QDF_USER, &qd->qd_flags) ?
1074 USRQUOTA : GRPQUOTA, qd->qd_id,
1075 sdp->sd_vfs->s_dev, QUOTA_NL_BHARDWARN);
1079 } else if (be64_to_cpu(qd->qd_qb.qb_warn) &&
1080 (s64)be64_to_cpu(qd->qd_qb.qb_warn) < value &&
1081 time_after_eq(jiffies, qd->qd_last_warn +
1083 gt_quota_warn_period) * HZ)) {
1084 quota_send_warning(test_bit(QDF_USER, &qd->qd_flags) ?
1085 USRQUOTA : GRPQUOTA, qd->qd_id,
1086 sdp->sd_vfs->s_dev, QUOTA_NL_BSOFTWARN);
1087 error = print_message(qd, "warning");
1088 qd->qd_last_warn = jiffies;
1095 void gfs2_quota_change(struct gfs2_inode *ip, s64 change,
1098 struct gfs2_quota_data *qd;
1101 if (gfs2_assert_warn(GFS2_SB(&ip->i_inode), change))
1103 if (ip->i_diskflags & GFS2_DIF_SYSTEM)
1106 for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1107 qd = ip->i_res->rs_qa_qd[x];
1109 if ((qd->qd_id == uid && test_bit(QDF_USER, &qd->qd_flags)) ||
1110 (qd->qd_id == gid && !test_bit(QDF_USER, &qd->qd_flags))) {
1116 int gfs2_quota_sync(struct super_block *sb, int type)
1118 struct gfs2_sbd *sdp = sb->s_fs_info;
1119 struct gfs2_quota_data **qda;
1120 unsigned int max_qd = gfs2_tune_get(sdp, gt_quota_simul_sync);
1121 unsigned int num_qd;
1125 sdp->sd_quota_sync_gen++;
1127 qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
1135 error = qd_fish(sdp, qda + num_qd);
1136 if (error || !qda[num_qd])
1138 if (++num_qd == max_qd)
1144 error = do_sync(num_qd, qda);
1146 for (x = 0; x < num_qd; x++)
1147 qda[x]->qd_sync_gen =
1148 sdp->sd_quota_sync_gen;
1150 for (x = 0; x < num_qd; x++)
1153 } while (!error && num_qd == max_qd);
1160 static int gfs2_quota_sync_timeo(struct super_block *sb, int type)
1162 return gfs2_quota_sync(sb, type);
1165 int gfs2_quota_refresh(struct gfs2_sbd *sdp, int user, u32 id)
1167 struct gfs2_quota_data *qd;
1168 struct gfs2_holder q_gh;
1171 error = qd_get(sdp, user, id, &qd);
1175 error = do_glock(qd, FORCE, &q_gh);
1177 gfs2_glock_dq_uninit(&q_gh);
1183 static void gfs2_quota_change_in(struct gfs2_quota_change_host *qc, const void *buf)
1185 const struct gfs2_quota_change *str = buf;
1187 qc->qc_change = be64_to_cpu(str->qc_change);
1188 qc->qc_flags = be32_to_cpu(str->qc_flags);
1189 qc->qc_id = be32_to_cpu(str->qc_id);
1192 int gfs2_quota_init(struct gfs2_sbd *sdp)
1194 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
1195 u64 size = i_size_read(sdp->sd_qc_inode);
1196 unsigned int blocks = size >> sdp->sd_sb.sb_bsize_shift;
1197 unsigned int x, slot = 0;
1198 unsigned int found = 0;
1203 if (gfs2_check_internal_file_size(sdp->sd_qc_inode, 1, 64 << 20))
1206 sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
1207 sdp->sd_quota_chunks = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * PAGE_SIZE);
1211 sdp->sd_quota_bitmap = kcalloc(sdp->sd_quota_chunks,
1212 sizeof(unsigned char *), GFP_NOFS);
1213 if (!sdp->sd_quota_bitmap)
1216 for (x = 0; x < sdp->sd_quota_chunks; x++) {
1217 sdp->sd_quota_bitmap[x] = kzalloc(PAGE_SIZE, GFP_NOFS);
1218 if (!sdp->sd_quota_bitmap[x])
1222 for (x = 0; x < blocks; x++) {
1223 struct buffer_head *bh;
1228 error = gfs2_extent_map(&ip->i_inode, x, &new, &dblock, &extlen);
1233 bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
1236 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
1241 for (y = 0; y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1243 struct gfs2_quota_change_host qc;
1244 struct gfs2_quota_data *qd;
1246 gfs2_quota_change_in(&qc, bh->b_data +
1247 sizeof(struct gfs2_meta_header) +
1248 y * sizeof(struct gfs2_quota_change));
1252 error = qd_alloc(sdp, (qc.qc_flags & GFS2_QCF_USER),
1259 set_bit(QDF_CHANGE, &qd->qd_flags);
1260 qd->qd_change = qc.qc_change;
1262 qd->qd_slot_count = 1;
1264 spin_lock(&qd_lru_lock);
1265 gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, slot, 1);
1266 list_add(&qd->qd_list, &sdp->sd_quota_list);
1267 atomic_inc(&sdp->sd_quota_count);
1268 spin_unlock(&qd_lru_lock);
1279 fs_info(sdp, "found %u quota changes\n", found);
1284 gfs2_quota_cleanup(sdp);
1288 void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1290 struct list_head *head = &sdp->sd_quota_list;
1291 struct gfs2_quota_data *qd;
1294 spin_lock(&qd_lru_lock);
1295 while (!list_empty(head)) {
1296 qd = list_entry(head->prev, struct gfs2_quota_data, qd_list);
1298 if (atomic_read(&qd->qd_count) > 1 ||
1299 (atomic_read(&qd->qd_count) &&
1300 !test_bit(QDF_CHANGE, &qd->qd_flags))) {
1301 list_move(&qd->qd_list, head);
1302 spin_unlock(&qd_lru_lock);
1304 spin_lock(&qd_lru_lock);
1308 list_del(&qd->qd_list);
1309 /* Also remove if this qd exists in the reclaim list */
1310 if (!list_empty(&qd->qd_reclaim)) {
1311 list_del_init(&qd->qd_reclaim);
1312 atomic_dec(&qd_lru_count);
1314 atomic_dec(&sdp->sd_quota_count);
1315 spin_unlock(&qd_lru_lock);
1317 if (!atomic_read(&qd->qd_count)) {
1318 gfs2_assert_warn(sdp, !qd->qd_change);
1319 gfs2_assert_warn(sdp, !qd->qd_slot_count);
1321 gfs2_assert_warn(sdp, qd->qd_slot_count == 1);
1322 gfs2_assert_warn(sdp, !qd->qd_bh_count);
1324 gfs2_glock_put(qd->qd_gl);
1325 kmem_cache_free(gfs2_quotad_cachep, qd);
1327 spin_lock(&qd_lru_lock);
1329 spin_unlock(&qd_lru_lock);
1331 gfs2_assert_warn(sdp, !atomic_read(&sdp->sd_quota_count));
1333 if (sdp->sd_quota_bitmap) {
1334 for (x = 0; x < sdp->sd_quota_chunks; x++)
1335 kfree(sdp->sd_quota_bitmap[x]);
1336 kfree(sdp->sd_quota_bitmap);
1340 static void quotad_error(struct gfs2_sbd *sdp, const char *msg, int error)
1342 if (error == 0 || error == -EROFS)
1344 if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
1345 fs_err(sdp, "gfs2_quotad: %s error %d\n", msg, error);
1348 static void quotad_check_timeo(struct gfs2_sbd *sdp, const char *msg,
1349 int (*fxn)(struct super_block *sb, int type),
1350 unsigned long t, unsigned long *timeo,
1351 unsigned int *new_timeo)
1354 int error = fxn(sdp->sd_vfs, 0);
1355 quotad_error(sdp, msg, error);
1356 *timeo = gfs2_tune_get_i(&sdp->sd_tune, new_timeo) * HZ;
1362 static void quotad_check_trunc_list(struct gfs2_sbd *sdp)
1364 struct gfs2_inode *ip;
1368 spin_lock(&sdp->sd_trunc_lock);
1369 if (!list_empty(&sdp->sd_trunc_list)) {
1370 ip = list_entry(sdp->sd_trunc_list.next,
1371 struct gfs2_inode, i_trunc_list);
1372 list_del_init(&ip->i_trunc_list);
1374 spin_unlock(&sdp->sd_trunc_lock);
1377 gfs2_glock_finish_truncate(ip);
1381 void gfs2_wake_up_statfs(struct gfs2_sbd *sdp) {
1382 if (!sdp->sd_statfs_force_sync) {
1383 sdp->sd_statfs_force_sync = 1;
1384 wake_up(&sdp->sd_quota_wait);
1390 * gfs2_quotad - Write cached quota changes into the quota file
1391 * @sdp: Pointer to GFS2 superblock
1395 int gfs2_quotad(void *data)
1397 struct gfs2_sbd *sdp = data;
1398 struct gfs2_tune *tune = &sdp->sd_tune;
1399 unsigned long statfs_timeo = 0;
1400 unsigned long quotad_timeo = 0;
1401 unsigned long t = 0;
1405 while (!kthread_should_stop()) {
1407 /* Update the master statfs file */
1408 if (sdp->sd_statfs_force_sync) {
1409 int error = gfs2_statfs_sync(sdp->sd_vfs, 0);
1410 quotad_error(sdp, "statfs", error);
1411 statfs_timeo = gfs2_tune_get(sdp, gt_statfs_quantum) * HZ;
1414 quotad_check_timeo(sdp, "statfs", gfs2_statfs_sync, t,
1416 &tune->gt_statfs_quantum);
1418 /* Update quota file */
1419 quotad_check_timeo(sdp, "sync", gfs2_quota_sync_timeo, t,
1420 "ad_timeo, &tune->gt_quota_quantum);
1422 /* Check for & recover partially truncated inodes */
1423 quotad_check_trunc_list(sdp);
1427 t = min(quotad_timeo, statfs_timeo);
1429 prepare_to_wait(&sdp->sd_quota_wait, &wait, TASK_INTERRUPTIBLE);
1430 spin_lock(&sdp->sd_trunc_lock);
1431 empty = list_empty(&sdp->sd_trunc_list);
1432 spin_unlock(&sdp->sd_trunc_lock);
1433 if (empty && !sdp->sd_statfs_force_sync)
1434 t -= schedule_timeout(t);
1437 finish_wait(&sdp->sd_quota_wait, &wait);
1443 static int gfs2_quota_get_xstate(struct super_block *sb,
1444 struct fs_quota_stat *fqs)
1446 struct gfs2_sbd *sdp = sb->s_fs_info;
1448 memset(fqs, 0, sizeof(struct fs_quota_stat));
1449 fqs->qs_version = FS_QSTAT_VERSION;
1451 switch (sdp->sd_args.ar_quota) {
1453 fqs->qs_flags |= (FS_QUOTA_UDQ_ENFD | FS_QUOTA_GDQ_ENFD);
1455 case GFS2_QUOTA_ACCOUNT:
1456 fqs->qs_flags |= (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT);
1458 case GFS2_QUOTA_OFF:
1462 if (sdp->sd_quota_inode) {
1463 fqs->qs_uquota.qfs_ino = GFS2_I(sdp->sd_quota_inode)->i_no_addr;
1464 fqs->qs_uquota.qfs_nblks = sdp->sd_quota_inode->i_blocks;
1466 fqs->qs_uquota.qfs_nextents = 1; /* unsupported */
1467 fqs->qs_gquota = fqs->qs_uquota; /* its the same inode in both cases */
1468 fqs->qs_incoredqs = atomic_read(&qd_lru_count);
1472 static int gfs2_get_dqblk(struct super_block *sb, int type, qid_t id,
1473 struct fs_disk_quota *fdq)
1475 struct gfs2_sbd *sdp = sb->s_fs_info;
1476 struct gfs2_quota_lvb *qlvb;
1477 struct gfs2_quota_data *qd;
1478 struct gfs2_holder q_gh;
1481 memset(fdq, 0, sizeof(struct fs_disk_quota));
1483 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1484 return -ESRCH; /* Crazy XFS error code */
1486 if (type == USRQUOTA)
1488 else if (type == GRPQUOTA)
1493 error = qd_get(sdp, type, id, &qd);
1496 error = do_glock(qd, FORCE, &q_gh);
1500 qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lvb;
1501 fdq->d_version = FS_DQUOT_VERSION;
1502 fdq->d_flags = (type == QUOTA_USER) ? FS_USER_QUOTA : FS_GROUP_QUOTA;
1504 fdq->d_blk_hardlimit = be64_to_cpu(qlvb->qb_limit) << sdp->sd_fsb2bb_shift;
1505 fdq->d_blk_softlimit = be64_to_cpu(qlvb->qb_warn) << sdp->sd_fsb2bb_shift;
1506 fdq->d_bcount = be64_to_cpu(qlvb->qb_value) << sdp->sd_fsb2bb_shift;
1508 gfs2_glock_dq_uninit(&q_gh);
1514 /* GFS2 only supports a subset of the XFS fields */
1515 #define GFS2_FIELDMASK (FS_DQ_BSOFT|FS_DQ_BHARD|FS_DQ_BCOUNT)
1517 static int gfs2_set_dqblk(struct super_block *sb, int type, qid_t id,
1518 struct fs_disk_quota *fdq)
1520 struct gfs2_sbd *sdp = sb->s_fs_info;
1521 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1522 struct gfs2_quota_data *qd;
1523 struct gfs2_holder q_gh, i_gh;
1524 unsigned int data_blocks, ind_blocks;
1525 unsigned int blocks = 0;
1530 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1531 return -ESRCH; /* Crazy XFS error code */
1536 if (fdq->d_flags != FS_USER_QUOTA)
1541 if (fdq->d_flags != FS_GROUP_QUOTA)
1548 if (fdq->d_fieldmask & ~GFS2_FIELDMASK)
1550 if (fdq->d_id != id)
1553 error = qd_get(sdp, type, id, &qd);
1557 error = gfs2_rs_alloc(ip);
1561 mutex_lock(&ip->i_inode.i_mutex);
1562 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE, 0, &q_gh);
1565 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1569 /* Check for existing entry, if none then alloc new blocks */
1570 error = update_qd(sdp, qd);
1574 /* If nothing has changed, this is a no-op */
1575 if ((fdq->d_fieldmask & FS_DQ_BSOFT) &&
1576 ((fdq->d_blk_softlimit >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_warn)))
1577 fdq->d_fieldmask ^= FS_DQ_BSOFT;
1579 if ((fdq->d_fieldmask & FS_DQ_BHARD) &&
1580 ((fdq->d_blk_hardlimit >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_limit)))
1581 fdq->d_fieldmask ^= FS_DQ_BHARD;
1583 if ((fdq->d_fieldmask & FS_DQ_BCOUNT) &&
1584 ((fdq->d_bcount >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_value)))
1585 fdq->d_fieldmask ^= FS_DQ_BCOUNT;
1587 if (fdq->d_fieldmask == 0)
1590 offset = qd2offset(qd);
1591 alloc_required = gfs2_write_alloc_required(ip, offset, sizeof(struct gfs2_quota));
1592 if (gfs2_is_stuffed(ip))
1594 if (alloc_required) {
1595 gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
1596 &data_blocks, &ind_blocks);
1597 blocks = 1 + data_blocks + ind_blocks;
1598 error = gfs2_inplace_reserve(ip, blocks);
1601 blocks += gfs2_rg_blocks(ip);
1604 /* Some quotas span block boundaries and can update two blocks,
1605 adding an extra block to the transaction to handle such quotas */
1606 error = gfs2_trans_begin(sdp, blocks + RES_DINODE + 2, 0);
1611 error = gfs2_adjust_quota(ip, offset, 0, qd, fdq);
1613 gfs2_trans_end(sdp);
1616 gfs2_inplace_release(ip);
1618 gfs2_glock_dq_uninit(&i_gh);
1620 gfs2_glock_dq_uninit(&q_gh);
1622 mutex_unlock(&ip->i_inode.i_mutex);
1628 const struct quotactl_ops gfs2_quotactl_ops = {
1629 .quota_sync = gfs2_quota_sync,
1630 .get_xstate = gfs2_quota_get_xstate,
1631 .get_dqblk = gfs2_get_dqblk,
1632 .set_dqblk = gfs2_set_dqblk,