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>
41 #include <linux/spinlock.h>
42 #include <linux/completion.h>
43 #include <linux/buffer_head.h>
44 #include <linux/sort.h>
46 #include <linux/bio.h>
47 #include <linux/gfs2_ondisk.h>
48 #include <linux/kthread.h>
49 #include <linux/freezer.h>
50 #include <linux/quota.h>
51 #include <linux/dqblk_xfs.h>
70 struct gfs2_quota_change_host {
72 u32 qc_flags; /* GFS2_QCF_... */
76 static LIST_HEAD(qd_lru_list);
77 static atomic_t qd_lru_count = ATOMIC_INIT(0);
78 static DEFINE_SPINLOCK(qd_lru_lock);
80 int gfs2_shrink_qd_memory(int nr, gfp_t gfp_mask)
82 struct gfs2_quota_data *qd;
88 if (!(gfp_mask & __GFP_FS))
91 spin_lock(&qd_lru_lock);
92 while (nr && !list_empty(&qd_lru_list)) {
93 qd = list_entry(qd_lru_list.next,
94 struct gfs2_quota_data, qd_reclaim);
95 sdp = qd->qd_gl->gl_sbd;
97 /* Free from the filesystem-specific list */
98 list_del(&qd->qd_list);
100 gfs2_assert_warn(sdp, !qd->qd_change);
101 gfs2_assert_warn(sdp, !qd->qd_slot_count);
102 gfs2_assert_warn(sdp, !qd->qd_bh_count);
104 gfs2_glock_put(qd->qd_gl);
105 atomic_dec(&sdp->sd_quota_count);
107 /* Delete it from the common reclaim list */
108 list_del_init(&qd->qd_reclaim);
109 atomic_dec(&qd_lru_count);
110 spin_unlock(&qd_lru_lock);
111 kmem_cache_free(gfs2_quotad_cachep, qd);
112 spin_lock(&qd_lru_lock);
115 spin_unlock(&qd_lru_lock);
118 return (atomic_read(&qd_lru_count) * sysctl_vfs_cache_pressure) / 100;
121 static u64 qd2offset(struct gfs2_quota_data *qd)
125 offset = 2 * (u64)qd->qd_id + !test_bit(QDF_USER, &qd->qd_flags);
126 offset *= sizeof(struct gfs2_quota);
131 static int qd_alloc(struct gfs2_sbd *sdp, int user, u32 id,
132 struct gfs2_quota_data **qdp)
134 struct gfs2_quota_data *qd;
137 qd = kmem_cache_zalloc(gfs2_quotad_cachep, GFP_NOFS);
141 atomic_set(&qd->qd_count, 1);
144 set_bit(QDF_USER, &qd->qd_flags);
146 INIT_LIST_HEAD(&qd->qd_reclaim);
148 error = gfs2_glock_get(sdp, 2 * (u64)id + !user,
149 &gfs2_quota_glops, CREATE, &qd->qd_gl);
158 kmem_cache_free(gfs2_quotad_cachep, qd);
162 static int qd_get(struct gfs2_sbd *sdp, int user, u32 id,
163 struct gfs2_quota_data **qdp)
165 struct gfs2_quota_data *qd = NULL, *new_qd = NULL;
172 spin_lock(&qd_lru_lock);
173 list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
174 if (qd->qd_id == id &&
175 !test_bit(QDF_USER, &qd->qd_flags) == !user) {
176 if (!atomic_read(&qd->qd_count) &&
177 !list_empty(&qd->qd_reclaim)) {
178 /* Remove it from reclaim list */
179 list_del_init(&qd->qd_reclaim);
180 atomic_dec(&qd_lru_count);
182 atomic_inc(&qd->qd_count);
193 list_add(&qd->qd_list, &sdp->sd_quota_list);
194 atomic_inc(&sdp->sd_quota_count);
198 spin_unlock(&qd_lru_lock);
202 gfs2_glock_put(new_qd->qd_gl);
203 kmem_cache_free(gfs2_quotad_cachep, new_qd);
209 error = qd_alloc(sdp, user, id, &new_qd);
215 static void qd_hold(struct gfs2_quota_data *qd)
217 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
218 gfs2_assert(sdp, atomic_read(&qd->qd_count));
219 atomic_inc(&qd->qd_count);
222 static void qd_put(struct gfs2_quota_data *qd)
224 if (atomic_dec_and_lock(&qd->qd_count, &qd_lru_lock)) {
225 /* Add to the reclaim list */
226 list_add_tail(&qd->qd_reclaim, &qd_lru_list);
227 atomic_inc(&qd_lru_count);
228 spin_unlock(&qd_lru_lock);
232 static int slot_get(struct gfs2_quota_data *qd)
234 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
235 unsigned int c, o = 0, b;
236 unsigned char byte = 0;
238 spin_lock(&qd_lru_lock);
240 if (qd->qd_slot_count++) {
241 spin_unlock(&qd_lru_lock);
245 for (c = 0; c < sdp->sd_quota_chunks; c++)
246 for (o = 0; o < PAGE_SIZE; o++) {
247 byte = sdp->sd_quota_bitmap[c][o];
255 for (b = 0; b < 8; b++)
256 if (!(byte & (1 << b)))
258 qd->qd_slot = c * (8 * PAGE_SIZE) + o * 8 + b;
260 if (qd->qd_slot >= sdp->sd_quota_slots)
263 sdp->sd_quota_bitmap[c][o] |= 1 << b;
265 spin_unlock(&qd_lru_lock);
271 spin_unlock(&qd_lru_lock);
275 static void slot_hold(struct gfs2_quota_data *qd)
277 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
279 spin_lock(&qd_lru_lock);
280 gfs2_assert(sdp, qd->qd_slot_count);
282 spin_unlock(&qd_lru_lock);
285 static void slot_put(struct gfs2_quota_data *qd)
287 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
289 spin_lock(&qd_lru_lock);
290 gfs2_assert(sdp, qd->qd_slot_count);
291 if (!--qd->qd_slot_count) {
292 gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, qd->qd_slot, 0);
295 spin_unlock(&qd_lru_lock);
298 static int bh_get(struct gfs2_quota_data *qd)
300 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
301 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
302 unsigned int block, offset;
303 struct buffer_head *bh;
305 struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
307 mutex_lock(&sdp->sd_quota_mutex);
309 if (qd->qd_bh_count++) {
310 mutex_unlock(&sdp->sd_quota_mutex);
314 block = qd->qd_slot / sdp->sd_qc_per_block;
315 offset = qd->qd_slot % sdp->sd_qc_per_block;
317 bh_map.b_size = 1 << ip->i_inode.i_blkbits;
318 error = gfs2_block_map(&ip->i_inode, block, &bh_map, 0);
321 error = gfs2_meta_read(ip->i_gl, bh_map.b_blocknr, DIO_WAIT, &bh);
325 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC))
329 qd->qd_bh_qc = (struct gfs2_quota_change *)
330 (bh->b_data + sizeof(struct gfs2_meta_header) +
331 offset * sizeof(struct gfs2_quota_change));
333 mutex_unlock(&sdp->sd_quota_mutex);
341 mutex_unlock(&sdp->sd_quota_mutex);
345 static void bh_put(struct gfs2_quota_data *qd)
347 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
349 mutex_lock(&sdp->sd_quota_mutex);
350 gfs2_assert(sdp, qd->qd_bh_count);
351 if (!--qd->qd_bh_count) {
356 mutex_unlock(&sdp->sd_quota_mutex);
359 static int qd_fish(struct gfs2_sbd *sdp, struct gfs2_quota_data **qdp)
361 struct gfs2_quota_data *qd = NULL;
367 if (sdp->sd_vfs->s_flags & MS_RDONLY)
370 spin_lock(&qd_lru_lock);
372 list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
373 if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
374 !test_bit(QDF_CHANGE, &qd->qd_flags) ||
375 qd->qd_sync_gen >= sdp->sd_quota_sync_gen)
378 list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
380 set_bit(QDF_LOCKED, &qd->qd_flags);
381 gfs2_assert_warn(sdp, atomic_read(&qd->qd_count));
382 atomic_inc(&qd->qd_count);
383 qd->qd_change_sync = qd->qd_change;
384 gfs2_assert_warn(sdp, qd->qd_slot_count);
394 spin_unlock(&qd_lru_lock);
397 gfs2_assert_warn(sdp, qd->qd_change_sync);
400 clear_bit(QDF_LOCKED, &qd->qd_flags);
412 static int qd_trylock(struct gfs2_quota_data *qd)
414 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
416 if (sdp->sd_vfs->s_flags & MS_RDONLY)
419 spin_lock(&qd_lru_lock);
421 if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
422 !test_bit(QDF_CHANGE, &qd->qd_flags)) {
423 spin_unlock(&qd_lru_lock);
427 list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
429 set_bit(QDF_LOCKED, &qd->qd_flags);
430 gfs2_assert_warn(sdp, atomic_read(&qd->qd_count));
431 atomic_inc(&qd->qd_count);
432 qd->qd_change_sync = qd->qd_change;
433 gfs2_assert_warn(sdp, qd->qd_slot_count);
436 spin_unlock(&qd_lru_lock);
438 gfs2_assert_warn(sdp, qd->qd_change_sync);
440 clear_bit(QDF_LOCKED, &qd->qd_flags);
449 static void qd_unlock(struct gfs2_quota_data *qd)
451 gfs2_assert_warn(qd->qd_gl->gl_sbd,
452 test_bit(QDF_LOCKED, &qd->qd_flags));
453 clear_bit(QDF_LOCKED, &qd->qd_flags);
459 static int qdsb_get(struct gfs2_sbd *sdp, int user, u32 id,
460 struct gfs2_quota_data **qdp)
464 error = qd_get(sdp, user, id, qdp);
468 error = slot_get(*qdp);
472 error = bh_get(*qdp);
485 static void qdsb_put(struct gfs2_quota_data *qd)
492 int gfs2_quota_hold(struct gfs2_inode *ip, u32 uid, u32 gid)
494 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
495 struct gfs2_alloc *al = ip->i_alloc;
496 struct gfs2_quota_data **qd = al->al_qd;
499 if (gfs2_assert_warn(sdp, !al->al_qd_num) ||
500 gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags)))
503 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
506 error = qdsb_get(sdp, QUOTA_USER, ip->i_inode.i_uid, qd);
512 error = qdsb_get(sdp, QUOTA_GROUP, ip->i_inode.i_gid, qd);
518 if (uid != NO_QUOTA_CHANGE && uid != ip->i_inode.i_uid) {
519 error = qdsb_get(sdp, QUOTA_USER, uid, qd);
526 if (gid != NO_QUOTA_CHANGE && gid != ip->i_inode.i_gid) {
527 error = qdsb_get(sdp, QUOTA_GROUP, gid, qd);
536 gfs2_quota_unhold(ip);
540 void gfs2_quota_unhold(struct gfs2_inode *ip)
542 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
543 struct gfs2_alloc *al = ip->i_alloc;
546 gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags));
548 for (x = 0; x < al->al_qd_num; x++) {
549 qdsb_put(al->al_qd[x]);
555 static int sort_qd(const void *a, const void *b)
557 const struct gfs2_quota_data *qd_a = *(const struct gfs2_quota_data **)a;
558 const struct gfs2_quota_data *qd_b = *(const struct gfs2_quota_data **)b;
560 if (!test_bit(QDF_USER, &qd_a->qd_flags) !=
561 !test_bit(QDF_USER, &qd_b->qd_flags)) {
562 if (test_bit(QDF_USER, &qd_a->qd_flags))
567 if (qd_a->qd_id < qd_b->qd_id)
569 if (qd_a->qd_id > qd_b->qd_id)
575 static void do_qc(struct gfs2_quota_data *qd, s64 change)
577 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
578 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
579 struct gfs2_quota_change *qc = qd->qd_bh_qc;
582 mutex_lock(&sdp->sd_quota_mutex);
583 gfs2_trans_add_bh(ip->i_gl, qd->qd_bh, 1);
585 if (!test_bit(QDF_CHANGE, &qd->qd_flags)) {
588 if (test_bit(QDF_USER, &qd->qd_flags))
589 qc->qc_flags = cpu_to_be32(GFS2_QCF_USER);
590 qc->qc_id = cpu_to_be32(qd->qd_id);
593 x = be64_to_cpu(qc->qc_change) + change;
594 qc->qc_change = cpu_to_be64(x);
596 spin_lock(&qd_lru_lock);
598 spin_unlock(&qd_lru_lock);
601 gfs2_assert_warn(sdp, test_bit(QDF_CHANGE, &qd->qd_flags));
602 clear_bit(QDF_CHANGE, &qd->qd_flags);
607 } else if (!test_and_set_bit(QDF_CHANGE, &qd->qd_flags)) {
612 mutex_unlock(&sdp->sd_quota_mutex);
616 * gfs2_adjust_quota - adjust record of current block usage
617 * @ip: The quota inode
618 * @loc: Offset of the entry in the quota file
619 * @change: The amount of usage change to record
620 * @qd: The quota data
621 * @fdq: The updated limits to record
623 * This function was mostly borrowed from gfs2_block_truncate_page which was
624 * in turn mostly borrowed from ext3
626 * Returns: 0 or -ve on error
629 static int gfs2_adjust_quota(struct gfs2_inode *ip, loff_t loc,
630 s64 change, struct gfs2_quota_data *qd,
631 struct fs_disk_quota *fdq)
633 struct inode *inode = &ip->i_inode;
634 struct address_space *mapping = inode->i_mapping;
635 unsigned long index = loc >> PAGE_CACHE_SHIFT;
636 unsigned offset = loc & (PAGE_CACHE_SIZE - 1);
637 unsigned blocksize, iblock, pos;
638 struct buffer_head *bh, *dibh;
641 struct gfs2_quota q, *qp;
645 if (gfs2_is_stuffed(ip))
646 gfs2_unstuff_dinode(ip, NULL);
648 memset(&q, 0, sizeof(struct gfs2_quota));
649 err = gfs2_internal_read(ip, NULL, (char *)&q, &loc, sizeof(q));
655 qp->qu_value = be64_to_cpu(qp->qu_value);
656 qp->qu_value += change;
657 qp->qu_value = cpu_to_be64(qp->qu_value);
658 qd->qd_qb.qb_value = qp->qu_value;
660 if (fdq->d_fieldmask & FS_DQ_BSOFT) {
661 qp->qu_warn = cpu_to_be64(fdq->d_blk_softlimit);
662 qd->qd_qb.qb_warn = qp->qu_warn;
664 if (fdq->d_fieldmask & FS_DQ_BHARD) {
665 qp->qu_limit = cpu_to_be64(fdq->d_blk_hardlimit);
666 qd->qd_qb.qb_limit = qp->qu_limit;
670 /* Write the quota into the quota file on disk */
672 nbytes = sizeof(struct gfs2_quota);
674 page = grab_cache_page(mapping, index);
678 blocksize = inode->i_sb->s_blocksize;
679 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
681 if (!page_has_buffers(page))
682 create_empty_buffers(page, blocksize, 0);
684 bh = page_buffers(page);
686 while (offset >= pos) {
687 bh = bh->b_this_page;
692 if (!buffer_mapped(bh)) {
693 gfs2_block_map(inode, iblock, bh, 1);
694 if (!buffer_mapped(bh))
696 /* If it's a newly allocated disk block for quota, zero it */
697 if (buffer_new(bh)) {
698 memset(bh->b_data, 0, bh->b_size);
699 set_buffer_uptodate(bh);
703 if (PageUptodate(page))
704 set_buffer_uptodate(bh);
706 if (!buffer_uptodate(bh)) {
707 ll_rw_block(READ_META, 1, &bh);
709 if (!buffer_uptodate(bh))
713 gfs2_trans_add_bh(ip->i_gl, bh, 0);
715 kaddr = kmap_atomic(page, KM_USER0);
716 if (offset + sizeof(struct gfs2_quota) > PAGE_CACHE_SIZE)
717 nbytes = PAGE_CACHE_SIZE - offset;
718 memcpy(kaddr + offset, ptr, nbytes);
719 flush_dcache_page(page);
720 kunmap_atomic(kaddr, KM_USER0);
722 page_cache_release(page);
724 /* If quota straddles page boundary, we need to update the rest of the
725 * quota at the beginning of the next page */
726 if (offset != 0) { /* first page, offset is closer to PAGE_CACHE_SIZE */
728 nbytes = sizeof(struct gfs2_quota) - nbytes;
734 /* Update the disk inode timestamp and size (if extended) */
735 err = gfs2_meta_inode_buffer(ip, &dibh);
739 size = loc + sizeof(struct gfs2_quota);
740 if (size > inode->i_size) {
741 ip->i_disksize = size;
742 i_size_write(inode, size);
744 inode->i_mtime = inode->i_atime = CURRENT_TIME;
745 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
746 gfs2_dinode_out(ip, dibh->b_data);
748 mark_inode_dirty(inode);
754 page_cache_release(page);
758 static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
760 struct gfs2_sbd *sdp = (*qda)->qd_gl->gl_sbd;
761 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
762 unsigned int data_blocks, ind_blocks;
763 struct gfs2_holder *ghs, i_gh;
765 struct gfs2_quota_data *qd;
767 unsigned int nalloc = 0, blocks;
768 struct gfs2_alloc *al = NULL;
771 gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
772 &data_blocks, &ind_blocks);
774 ghs = kcalloc(num_qd, sizeof(struct gfs2_holder), GFP_NOFS);
778 sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL);
779 mutex_lock_nested(&ip->i_inode.i_mutex, I_MUTEX_QUOTA);
780 for (qx = 0; qx < num_qd; qx++) {
781 error = gfs2_glock_nq_init(qda[qx]->qd_gl, LM_ST_EXCLUSIVE,
782 GL_NOCACHE, &ghs[qx]);
787 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
791 for (x = 0; x < num_qd; x++) {
794 offset = qd2offset(qda[x]);
795 error = gfs2_write_alloc_required(ip, offset,
796 sizeof(struct gfs2_quota),
804 al = gfs2_alloc_get(ip);
810 * 1 blk for unstuffing inode if stuffed. We add this extra
811 * block to the reservation unconditionally. If the inode
812 * doesn't need unstuffing, the block will be released to the
813 * rgrp since it won't be allocated during the transaction
815 al->al_requested = 1;
816 /* +3 in the end for unstuffing block, inode size update block
817 * and another block in case quota straddles page boundary and
818 * two blocks need to be updated instead of 1 */
819 blocks = num_qd * data_blocks + RES_DINODE + num_qd + 3;
822 al->al_requested += nalloc * (data_blocks + ind_blocks);
823 error = gfs2_inplace_reserve(ip);
828 blocks += al->al_rgd->rd_length + nalloc * ind_blocks + RES_STATFS;
830 error = gfs2_trans_begin(sdp, blocks, 0);
834 for (x = 0; x < num_qd; x++) {
836 offset = qd2offset(qd);
837 error = gfs2_adjust_quota(ip, offset, qd->qd_change_sync, qd, NULL);
841 do_qc(qd, -qd->qd_change_sync);
849 gfs2_inplace_release(ip);
853 gfs2_glock_dq_uninit(&i_gh);
856 gfs2_glock_dq_uninit(&ghs[qx]);
857 mutex_unlock(&ip->i_inode.i_mutex);
859 gfs2_log_flush(ip->i_gl->gl_sbd, ip->i_gl);
863 static int update_qd(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd)
865 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
867 struct gfs2_quota_lvb *qlvb;
871 memset(&q, 0, sizeof(struct gfs2_quota));
873 error = gfs2_internal_read(ip, NULL, (char *)&q, &pos, sizeof(q));
877 qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lvb;
878 qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC);
880 qlvb->qb_limit = q.qu_limit;
881 qlvb->qb_warn = q.qu_warn;
882 qlvb->qb_value = q.qu_value;
888 static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
889 struct gfs2_holder *q_gh)
891 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
892 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
893 struct gfs2_holder i_gh;
897 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
901 qd->qd_qb = *(struct gfs2_quota_lvb *)qd->qd_gl->gl_lvb;
903 if (force_refresh || qd->qd_qb.qb_magic != cpu_to_be32(GFS2_MAGIC)) {
904 gfs2_glock_dq_uninit(q_gh);
905 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE,
910 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
914 error = update_qd(sdp, qd);
918 gfs2_glock_dq_uninit(&i_gh);
919 gfs2_glock_dq_uninit(q_gh);
927 gfs2_glock_dq_uninit(&i_gh);
929 gfs2_glock_dq_uninit(q_gh);
933 int gfs2_quota_lock(struct gfs2_inode *ip, u32 uid, u32 gid)
935 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
936 struct gfs2_alloc *al = ip->i_alloc;
940 gfs2_quota_hold(ip, uid, gid);
942 if (capable(CAP_SYS_RESOURCE) ||
943 sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
946 sort(al->al_qd, al->al_qd_num, sizeof(struct gfs2_quota_data *),
949 for (x = 0; x < al->al_qd_num; x++) {
950 error = do_glock(al->al_qd[x], NO_FORCE, &al->al_qd_ghs[x]);
956 set_bit(GIF_QD_LOCKED, &ip->i_flags);
959 gfs2_glock_dq_uninit(&al->al_qd_ghs[x]);
960 gfs2_quota_unhold(ip);
966 static int need_sync(struct gfs2_quota_data *qd)
968 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
969 struct gfs2_tune *gt = &sdp->sd_tune;
971 unsigned int num, den;
974 if (!qd->qd_qb.qb_limit)
977 spin_lock(&qd_lru_lock);
978 value = qd->qd_change;
979 spin_unlock(&qd_lru_lock);
981 spin_lock(>->gt_spin);
982 num = gt->gt_quota_scale_num;
983 den = gt->gt_quota_scale_den;
984 spin_unlock(>->gt_spin);
988 else if ((s64)be64_to_cpu(qd->qd_qb.qb_value) >=
989 (s64)be64_to_cpu(qd->qd_qb.qb_limit))
992 value *= gfs2_jindex_size(sdp) * num;
993 value = div_s64(value, den);
994 value += (s64)be64_to_cpu(qd->qd_qb.qb_value);
995 if (value < (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1002 void gfs2_quota_unlock(struct gfs2_inode *ip)
1004 struct gfs2_alloc *al = ip->i_alloc;
1005 struct gfs2_quota_data *qda[4];
1006 unsigned int count = 0;
1009 if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
1012 for (x = 0; x < al->al_qd_num; x++) {
1013 struct gfs2_quota_data *qd;
1017 sync = need_sync(qd);
1019 gfs2_glock_dq_uninit(&al->al_qd_ghs[x]);
1021 if (sync && qd_trylock(qd))
1026 do_sync(count, qda);
1027 for (x = 0; x < count; x++)
1032 gfs2_quota_unhold(ip);
1035 #define MAX_LINE 256
1037 static int print_message(struct gfs2_quota_data *qd, char *type)
1039 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
1041 printk(KERN_INFO "GFS2: fsid=%s: quota %s for %s %u\n",
1042 sdp->sd_fsname, type,
1043 (test_bit(QDF_USER, &qd->qd_flags)) ? "user" : "group",
1049 int gfs2_quota_check(struct gfs2_inode *ip, u32 uid, u32 gid)
1051 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1052 struct gfs2_alloc *al = ip->i_alloc;
1053 struct gfs2_quota_data *qd;
1058 if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
1061 if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
1064 for (x = 0; x < al->al_qd_num; x++) {
1067 if (!((qd->qd_id == uid && test_bit(QDF_USER, &qd->qd_flags)) ||
1068 (qd->qd_id == gid && !test_bit(QDF_USER, &qd->qd_flags))))
1071 value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
1072 spin_lock(&qd_lru_lock);
1073 value += qd->qd_change;
1074 spin_unlock(&qd_lru_lock);
1076 if (be64_to_cpu(qd->qd_qb.qb_limit) && (s64)be64_to_cpu(qd->qd_qb.qb_limit) < value) {
1077 print_message(qd, "exceeded");
1078 quota_send_warning(test_bit(QDF_USER, &qd->qd_flags) ?
1079 USRQUOTA : GRPQUOTA, qd->qd_id,
1080 sdp->sd_vfs->s_dev, QUOTA_NL_BHARDWARN);
1084 } else if (be64_to_cpu(qd->qd_qb.qb_warn) &&
1085 (s64)be64_to_cpu(qd->qd_qb.qb_warn) < value &&
1086 time_after_eq(jiffies, qd->qd_last_warn +
1088 gt_quota_warn_period) * HZ)) {
1089 quota_send_warning(test_bit(QDF_USER, &qd->qd_flags) ?
1090 USRQUOTA : GRPQUOTA, qd->qd_id,
1091 sdp->sd_vfs->s_dev, QUOTA_NL_BSOFTWARN);
1092 error = print_message(qd, "warning");
1093 qd->qd_last_warn = jiffies;
1100 void gfs2_quota_change(struct gfs2_inode *ip, s64 change,
1103 struct gfs2_alloc *al = ip->i_alloc;
1104 struct gfs2_quota_data *qd;
1107 if (gfs2_assert_warn(GFS2_SB(&ip->i_inode), change))
1109 if (ip->i_diskflags & GFS2_DIF_SYSTEM)
1112 for (x = 0; x < al->al_qd_num; x++) {
1115 if ((qd->qd_id == uid && test_bit(QDF_USER, &qd->qd_flags)) ||
1116 (qd->qd_id == gid && !test_bit(QDF_USER, &qd->qd_flags))) {
1122 int gfs2_quota_sync(struct super_block *sb, int type, int wait)
1124 struct gfs2_sbd *sdp = sb->s_fs_info;
1125 struct gfs2_quota_data **qda;
1126 unsigned int max_qd = gfs2_tune_get(sdp, gt_quota_simul_sync);
1127 unsigned int num_qd;
1131 sdp->sd_quota_sync_gen++;
1133 qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
1141 error = qd_fish(sdp, qda + num_qd);
1142 if (error || !qda[num_qd])
1144 if (++num_qd == max_qd)
1150 error = do_sync(num_qd, qda);
1152 for (x = 0; x < num_qd; x++)
1153 qda[x]->qd_sync_gen =
1154 sdp->sd_quota_sync_gen;
1156 for (x = 0; x < num_qd; x++)
1159 } while (!error && num_qd == max_qd);
1166 static int gfs2_quota_sync_timeo(struct super_block *sb, int type)
1168 return gfs2_quota_sync(sb, type, 0);
1171 int gfs2_quota_refresh(struct gfs2_sbd *sdp, int user, u32 id)
1173 struct gfs2_quota_data *qd;
1174 struct gfs2_holder q_gh;
1177 error = qd_get(sdp, user, id, &qd);
1181 error = do_glock(qd, FORCE, &q_gh);
1183 gfs2_glock_dq_uninit(&q_gh);
1189 static void gfs2_quota_change_in(struct gfs2_quota_change_host *qc, const void *buf)
1191 const struct gfs2_quota_change *str = buf;
1193 qc->qc_change = be64_to_cpu(str->qc_change);
1194 qc->qc_flags = be32_to_cpu(str->qc_flags);
1195 qc->qc_id = be32_to_cpu(str->qc_id);
1198 int gfs2_quota_init(struct gfs2_sbd *sdp)
1200 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
1201 unsigned int blocks = ip->i_disksize >> sdp->sd_sb.sb_bsize_shift;
1202 unsigned int x, slot = 0;
1203 unsigned int found = 0;
1208 if (!ip->i_disksize || ip->i_disksize > (64 << 20) ||
1209 ip->i_disksize & (sdp->sd_sb.sb_bsize - 1)) {
1210 gfs2_consist_inode(ip);
1213 sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
1214 sdp->sd_quota_chunks = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * PAGE_SIZE);
1218 sdp->sd_quota_bitmap = kcalloc(sdp->sd_quota_chunks,
1219 sizeof(unsigned char *), GFP_NOFS);
1220 if (!sdp->sd_quota_bitmap)
1223 for (x = 0; x < sdp->sd_quota_chunks; x++) {
1224 sdp->sd_quota_bitmap[x] = kzalloc(PAGE_SIZE, GFP_NOFS);
1225 if (!sdp->sd_quota_bitmap[x])
1229 for (x = 0; x < blocks; x++) {
1230 struct buffer_head *bh;
1235 error = gfs2_extent_map(&ip->i_inode, x, &new, &dblock, &extlen);
1240 bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
1243 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
1248 for (y = 0; y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1250 struct gfs2_quota_change_host qc;
1251 struct gfs2_quota_data *qd;
1253 gfs2_quota_change_in(&qc, bh->b_data +
1254 sizeof(struct gfs2_meta_header) +
1255 y * sizeof(struct gfs2_quota_change));
1259 error = qd_alloc(sdp, (qc.qc_flags & GFS2_QCF_USER),
1266 set_bit(QDF_CHANGE, &qd->qd_flags);
1267 qd->qd_change = qc.qc_change;
1269 qd->qd_slot_count = 1;
1271 spin_lock(&qd_lru_lock);
1272 gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, slot, 1);
1273 list_add(&qd->qd_list, &sdp->sd_quota_list);
1274 atomic_inc(&sdp->sd_quota_count);
1275 spin_unlock(&qd_lru_lock);
1286 fs_info(sdp, "found %u quota changes\n", found);
1291 gfs2_quota_cleanup(sdp);
1295 void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1297 struct list_head *head = &sdp->sd_quota_list;
1298 struct gfs2_quota_data *qd;
1301 spin_lock(&qd_lru_lock);
1302 while (!list_empty(head)) {
1303 qd = list_entry(head->prev, struct gfs2_quota_data, qd_list);
1305 if (atomic_read(&qd->qd_count) > 1 ||
1306 (atomic_read(&qd->qd_count) &&
1307 !test_bit(QDF_CHANGE, &qd->qd_flags))) {
1308 list_move(&qd->qd_list, head);
1309 spin_unlock(&qd_lru_lock);
1311 spin_lock(&qd_lru_lock);
1315 list_del(&qd->qd_list);
1316 /* Also remove if this qd exists in the reclaim list */
1317 if (!list_empty(&qd->qd_reclaim)) {
1318 list_del_init(&qd->qd_reclaim);
1319 atomic_dec(&qd_lru_count);
1321 atomic_dec(&sdp->sd_quota_count);
1322 spin_unlock(&qd_lru_lock);
1324 if (!atomic_read(&qd->qd_count)) {
1325 gfs2_assert_warn(sdp, !qd->qd_change);
1326 gfs2_assert_warn(sdp, !qd->qd_slot_count);
1328 gfs2_assert_warn(sdp, qd->qd_slot_count == 1);
1329 gfs2_assert_warn(sdp, !qd->qd_bh_count);
1331 gfs2_glock_put(qd->qd_gl);
1332 kmem_cache_free(gfs2_quotad_cachep, qd);
1334 spin_lock(&qd_lru_lock);
1336 spin_unlock(&qd_lru_lock);
1338 gfs2_assert_warn(sdp, !atomic_read(&sdp->sd_quota_count));
1340 if (sdp->sd_quota_bitmap) {
1341 for (x = 0; x < sdp->sd_quota_chunks; x++)
1342 kfree(sdp->sd_quota_bitmap[x]);
1343 kfree(sdp->sd_quota_bitmap);
1347 static void quotad_error(struct gfs2_sbd *sdp, const char *msg, int error)
1349 if (error == 0 || error == -EROFS)
1351 if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
1352 fs_err(sdp, "gfs2_quotad: %s error %d\n", msg, error);
1355 static void quotad_check_timeo(struct gfs2_sbd *sdp, const char *msg,
1356 int (*fxn)(struct super_block *sb, int type),
1357 unsigned long t, unsigned long *timeo,
1358 unsigned int *new_timeo)
1361 int error = fxn(sdp->sd_vfs, 0);
1362 quotad_error(sdp, msg, error);
1363 *timeo = gfs2_tune_get_i(&sdp->sd_tune, new_timeo) * HZ;
1369 static void quotad_check_trunc_list(struct gfs2_sbd *sdp)
1371 struct gfs2_inode *ip;
1375 spin_lock(&sdp->sd_trunc_lock);
1376 if (!list_empty(&sdp->sd_trunc_list)) {
1377 ip = list_entry(sdp->sd_trunc_list.next,
1378 struct gfs2_inode, i_trunc_list);
1379 list_del_init(&ip->i_trunc_list);
1381 spin_unlock(&sdp->sd_trunc_lock);
1384 gfs2_glock_finish_truncate(ip);
1388 void gfs2_wake_up_statfs(struct gfs2_sbd *sdp) {
1389 if (!sdp->sd_statfs_force_sync) {
1390 sdp->sd_statfs_force_sync = 1;
1391 wake_up(&sdp->sd_quota_wait);
1397 * gfs2_quotad - Write cached quota changes into the quota file
1398 * @sdp: Pointer to GFS2 superblock
1402 int gfs2_quotad(void *data)
1404 struct gfs2_sbd *sdp = data;
1405 struct gfs2_tune *tune = &sdp->sd_tune;
1406 unsigned long statfs_timeo = 0;
1407 unsigned long quotad_timeo = 0;
1408 unsigned long t = 0;
1412 while (!kthread_should_stop()) {
1414 /* Update the master statfs file */
1415 if (sdp->sd_statfs_force_sync) {
1416 int error = gfs2_statfs_sync(sdp->sd_vfs, 0);
1417 quotad_error(sdp, "statfs", error);
1418 statfs_timeo = gfs2_tune_get(sdp, gt_statfs_quantum) * HZ;
1421 quotad_check_timeo(sdp, "statfs", gfs2_statfs_sync, t,
1423 &tune->gt_statfs_quantum);
1425 /* Update quota file */
1426 quotad_check_timeo(sdp, "sync", gfs2_quota_sync_timeo, t,
1427 "ad_timeo, &tune->gt_quota_quantum);
1429 /* Check for & recover partially truncated inodes */
1430 quotad_check_trunc_list(sdp);
1432 if (freezing(current))
1434 t = min(quotad_timeo, statfs_timeo);
1436 prepare_to_wait(&sdp->sd_quota_wait, &wait, TASK_INTERRUPTIBLE);
1437 spin_lock(&sdp->sd_trunc_lock);
1438 empty = list_empty(&sdp->sd_trunc_list);
1439 spin_unlock(&sdp->sd_trunc_lock);
1440 if (empty && !sdp->sd_statfs_force_sync)
1441 t -= schedule_timeout(t);
1444 finish_wait(&sdp->sd_quota_wait, &wait);
1450 static int gfs2_quota_get_xstate(struct super_block *sb,
1451 struct fs_quota_stat *fqs)
1453 struct gfs2_sbd *sdp = sb->s_fs_info;
1455 memset(fqs, 0, sizeof(struct fs_quota_stat));
1456 fqs->qs_version = FS_QSTAT_VERSION;
1458 switch (sdp->sd_args.ar_quota) {
1460 fqs->qs_flags |= (XFS_QUOTA_UDQ_ENFD | XFS_QUOTA_GDQ_ENFD);
1462 case GFS2_QUOTA_ACCOUNT:
1463 fqs->qs_flags |= (XFS_QUOTA_UDQ_ACCT | XFS_QUOTA_GDQ_ACCT);
1465 case GFS2_QUOTA_OFF:
1469 if (sdp->sd_quota_inode) {
1470 fqs->qs_uquota.qfs_ino = GFS2_I(sdp->sd_quota_inode)->i_no_addr;
1471 fqs->qs_uquota.qfs_nblks = sdp->sd_quota_inode->i_blocks;
1473 fqs->qs_uquota.qfs_nextents = 1; /* unsupported */
1474 fqs->qs_gquota = fqs->qs_uquota; /* its the same inode in both cases */
1475 fqs->qs_incoredqs = atomic_read(&qd_lru_count);
1479 static int gfs2_get_dqblk(struct super_block *sb, int type, qid_t id,
1480 struct fs_disk_quota *fdq)
1482 struct gfs2_sbd *sdp = sb->s_fs_info;
1483 struct gfs2_quota_lvb *qlvb;
1484 struct gfs2_quota_data *qd;
1485 struct gfs2_holder q_gh;
1488 memset(fdq, 0, sizeof(struct fs_disk_quota));
1490 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1491 return -ESRCH; /* Crazy XFS error code */
1493 if (type == USRQUOTA)
1495 else if (type == GRPQUOTA)
1500 error = qd_get(sdp, type, id, &qd);
1503 error = do_glock(qd, FORCE, &q_gh);
1507 qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lvb;
1508 fdq->d_version = FS_DQUOT_VERSION;
1509 fdq->d_flags = (type == QUOTA_USER) ? XFS_USER_QUOTA : XFS_GROUP_QUOTA;
1511 fdq->d_blk_hardlimit = be64_to_cpu(qlvb->qb_limit);
1512 fdq->d_blk_softlimit = be64_to_cpu(qlvb->qb_warn);
1513 fdq->d_bcount = be64_to_cpu(qlvb->qb_value);
1515 gfs2_glock_dq_uninit(&q_gh);
1521 /* GFS2 only supports a subset of the XFS fields */
1522 #define GFS2_FIELDMASK (FS_DQ_BSOFT|FS_DQ_BHARD)
1524 static int gfs2_set_dqblk(struct super_block *sb, int type, qid_t id,
1525 struct fs_disk_quota *fdq)
1527 struct gfs2_sbd *sdp = sb->s_fs_info;
1528 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1529 struct gfs2_quota_data *qd;
1530 struct gfs2_holder q_gh, i_gh;
1531 unsigned int data_blocks, ind_blocks;
1532 unsigned int blocks = 0;
1534 struct gfs2_alloc *al;
1538 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1539 return -ESRCH; /* Crazy XFS error code */
1544 if (fdq->d_flags != XFS_USER_QUOTA)
1549 if (fdq->d_flags != XFS_GROUP_QUOTA)
1556 if (fdq->d_fieldmask & ~GFS2_FIELDMASK)
1558 if (fdq->d_id != id)
1561 error = qd_get(sdp, type, id, &qd);
1565 mutex_lock(&ip->i_inode.i_mutex);
1566 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE, 0, &q_gh);
1569 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1573 /* Check for existing entry, if none then alloc new blocks */
1574 error = update_qd(sdp, qd);
1578 /* If nothing has changed, this is a no-op */
1579 if ((fdq->d_fieldmask & FS_DQ_BSOFT) &&
1580 (fdq->d_blk_softlimit == be64_to_cpu(qd->qd_qb.qb_warn)))
1581 fdq->d_fieldmask ^= FS_DQ_BSOFT;
1582 if ((fdq->d_fieldmask & FS_DQ_BHARD) &&
1583 (fdq->d_blk_hardlimit == be64_to_cpu(qd->qd_qb.qb_limit)))
1584 fdq->d_fieldmask ^= FS_DQ_BHARD;
1585 if (fdq->d_fieldmask == 0)
1588 offset = qd2offset(qd);
1589 error = gfs2_write_alloc_required(ip, offset, sizeof(struct gfs2_quota),
1593 if (alloc_required) {
1594 al = gfs2_alloc_get(ip);
1597 gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
1598 &data_blocks, &ind_blocks);
1599 blocks = al->al_requested = 1 + data_blocks + ind_blocks;
1600 error = gfs2_inplace_reserve(ip);
1605 error = gfs2_trans_begin(sdp, blocks + RES_DINODE + 1, 0);
1610 error = gfs2_adjust_quota(ip, offset, 0, qd, fdq);
1612 gfs2_trans_end(sdp);
1614 if (alloc_required) {
1615 gfs2_inplace_release(ip);
1620 gfs2_glock_dq_uninit(&i_gh);
1622 gfs2_glock_dq_uninit(&q_gh);
1624 mutex_unlock(&ip->i_inode.i_mutex);
1629 const struct quotactl_ops gfs2_quotactl_ops = {
1630 .quota_sync = gfs2_quota_sync,
1631 .get_xstate = gfs2_quota_get_xstate,
1632 .get_dqblk = gfs2_get_dqblk,
1633 .set_dqblk = gfs2_set_dqblk,