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>
68 struct gfs2_quota_change_host {
70 u32 qc_flags; /* GFS2_QCF_... */
74 static LIST_HEAD(qd_lru_list);
75 static atomic_t qd_lru_count = ATOMIC_INIT(0);
76 static DEFINE_SPINLOCK(qd_lru_lock);
78 unsigned long gfs2_qd_shrink_scan(struct shrinker *shrink,
79 struct shrink_control *sc)
81 struct gfs2_quota_data *qd;
83 int nr_to_scan = sc->nr_to_scan;
86 if (!(sc->gfp_mask & __GFP_FS))
89 spin_lock(&qd_lru_lock);
90 while (nr_to_scan && !list_empty(&qd_lru_list)) {
91 qd = list_entry(qd_lru_list.next,
92 struct gfs2_quota_data, qd_reclaim);
93 sdp = qd->qd_gl->gl_sbd;
95 /* Free from the filesystem-specific list */
96 list_del(&qd->qd_list);
98 gfs2_assert_warn(sdp, !qd->qd_change);
99 gfs2_assert_warn(sdp, !qd->qd_slot_count);
100 gfs2_assert_warn(sdp, !qd->qd_bh_count);
102 gfs2_glock_put(qd->qd_gl);
103 atomic_dec(&sdp->sd_quota_count);
105 /* Delete it from the common reclaim list */
106 list_del_init(&qd->qd_reclaim);
107 atomic_dec(&qd_lru_count);
108 spin_unlock(&qd_lru_lock);
109 kmem_cache_free(gfs2_quotad_cachep, qd);
110 spin_lock(&qd_lru_lock);
114 spin_unlock(&qd_lru_lock);
118 unsigned long gfs2_qd_shrink_count(struct shrinker *shrink,
119 struct shrink_control *sc)
121 return vfs_pressure_ratio(atomic_read(&qd_lru_count));
124 static u64 qd2index(struct gfs2_quota_data *qd)
126 struct kqid qid = qd->qd_id;
127 return (2 * (u64)from_kqid(&init_user_ns, qid)) +
128 ((qid.type == USRQUOTA) ? 0 : 1);
131 static u64 qd2offset(struct gfs2_quota_data *qd)
135 offset = qd2index(qd);
136 offset *= sizeof(struct gfs2_quota);
141 static int qd_alloc(struct gfs2_sbd *sdp, struct kqid qid,
142 struct gfs2_quota_data **qdp)
144 struct gfs2_quota_data *qd;
147 qd = kmem_cache_zalloc(gfs2_quotad_cachep, GFP_NOFS);
151 atomic_set(&qd->qd_count, 1);
154 INIT_LIST_HEAD(&qd->qd_reclaim);
156 error = gfs2_glock_get(sdp, qd2index(qd),
157 &gfs2_quota_glops, CREATE, &qd->qd_gl);
166 kmem_cache_free(gfs2_quotad_cachep, qd);
170 static int qd_get(struct gfs2_sbd *sdp, struct kqid qid,
171 struct gfs2_quota_data **qdp)
173 struct gfs2_quota_data *qd = NULL, *new_qd = NULL;
180 spin_lock(&qd_lru_lock);
181 list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
182 if (qid_eq(qd->qd_id, qid)) {
183 if (!atomic_read(&qd->qd_count) &&
184 !list_empty(&qd->qd_reclaim)) {
185 /* Remove it from reclaim list */
186 list_del_init(&qd->qd_reclaim);
187 atomic_dec(&qd_lru_count);
189 atomic_inc(&qd->qd_count);
200 list_add(&qd->qd_list, &sdp->sd_quota_list);
201 atomic_inc(&sdp->sd_quota_count);
205 spin_unlock(&qd_lru_lock);
209 gfs2_glock_put(new_qd->qd_gl);
210 kmem_cache_free(gfs2_quotad_cachep, new_qd);
216 error = qd_alloc(sdp, qid, &new_qd);
222 static void qd_hold(struct gfs2_quota_data *qd)
224 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
225 gfs2_assert(sdp, atomic_read(&qd->qd_count));
226 atomic_inc(&qd->qd_count);
229 static void qd_put(struct gfs2_quota_data *qd)
231 if (atomic_dec_and_lock(&qd->qd_count, &qd_lru_lock)) {
232 /* Add to the reclaim list */
233 list_add_tail(&qd->qd_reclaim, &qd_lru_list);
234 atomic_inc(&qd_lru_count);
235 spin_unlock(&qd_lru_lock);
239 static int slot_get(struct gfs2_quota_data *qd)
241 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
242 unsigned int c, o = 0, b;
243 unsigned char byte = 0;
245 spin_lock(&qd_lru_lock);
247 if (qd->qd_slot_count++) {
248 spin_unlock(&qd_lru_lock);
252 for (c = 0; c < sdp->sd_quota_chunks; c++)
253 for (o = 0; o < PAGE_SIZE; o++) {
254 byte = sdp->sd_quota_bitmap[c][o];
262 for (b = 0; b < 8; b++)
263 if (!(byte & (1 << b)))
265 qd->qd_slot = c * (8 * PAGE_SIZE) + o * 8 + b;
267 if (qd->qd_slot >= sdp->sd_quota_slots)
270 sdp->sd_quota_bitmap[c][o] |= 1 << b;
272 spin_unlock(&qd_lru_lock);
278 spin_unlock(&qd_lru_lock);
282 static void slot_hold(struct gfs2_quota_data *qd)
284 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
286 spin_lock(&qd_lru_lock);
287 gfs2_assert(sdp, qd->qd_slot_count);
289 spin_unlock(&qd_lru_lock);
292 static void gfs2_icbit_munge(struct gfs2_sbd *sdp, unsigned char **bitmap,
293 unsigned int bit, int new_value)
295 unsigned int c, o, b = bit;
298 c = b / (8 * PAGE_SIZE);
303 old_value = (bitmap[c][o] & (1 << b));
304 gfs2_assert_withdraw(sdp, !old_value != !new_value);
307 bitmap[c][o] |= 1 << b;
309 bitmap[c][o] &= ~(1 << b);
312 static void slot_put(struct gfs2_quota_data *qd)
314 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
316 spin_lock(&qd_lru_lock);
317 gfs2_assert(sdp, qd->qd_slot_count);
318 if (!--qd->qd_slot_count) {
319 gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, qd->qd_slot, 0);
322 spin_unlock(&qd_lru_lock);
325 static int bh_get(struct gfs2_quota_data *qd)
327 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
328 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
329 unsigned int block, offset;
330 struct buffer_head *bh;
332 struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
334 mutex_lock(&sdp->sd_quota_mutex);
336 if (qd->qd_bh_count++) {
337 mutex_unlock(&sdp->sd_quota_mutex);
341 block = qd->qd_slot / sdp->sd_qc_per_block;
342 offset = qd->qd_slot % sdp->sd_qc_per_block;
344 bh_map.b_size = 1 << ip->i_inode.i_blkbits;
345 error = gfs2_block_map(&ip->i_inode, block, &bh_map, 0);
348 error = gfs2_meta_read(ip->i_gl, bh_map.b_blocknr, DIO_WAIT, &bh);
352 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC))
356 qd->qd_bh_qc = (struct gfs2_quota_change *)
357 (bh->b_data + sizeof(struct gfs2_meta_header) +
358 offset * sizeof(struct gfs2_quota_change));
360 mutex_unlock(&sdp->sd_quota_mutex);
368 mutex_unlock(&sdp->sd_quota_mutex);
372 static void bh_put(struct gfs2_quota_data *qd)
374 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
376 mutex_lock(&sdp->sd_quota_mutex);
377 gfs2_assert(sdp, qd->qd_bh_count);
378 if (!--qd->qd_bh_count) {
383 mutex_unlock(&sdp->sd_quota_mutex);
386 static int qd_check_sync(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd,
389 if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
390 !test_bit(QDF_CHANGE, &qd->qd_flags) ||
391 (sync_gen && (qd->qd_sync_gen >= *sync_gen)))
394 list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
396 set_bit(QDF_LOCKED, &qd->qd_flags);
397 gfs2_assert_warn(sdp, atomic_read(&qd->qd_count));
398 atomic_inc(&qd->qd_count);
399 qd->qd_change_sync = qd->qd_change;
400 gfs2_assert_warn(sdp, qd->qd_slot_count);
405 static int qd_fish(struct gfs2_sbd *sdp, struct gfs2_quota_data **qdp)
407 struct gfs2_quota_data *qd = NULL;
413 if (sdp->sd_vfs->s_flags & MS_RDONLY)
416 spin_lock(&qd_lru_lock);
418 list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
419 found = qd_check_sync(sdp, qd, &sdp->sd_quota_sync_gen);
427 spin_unlock(&qd_lru_lock);
430 gfs2_assert_warn(sdp, qd->qd_change_sync);
433 clear_bit(QDF_LOCKED, &qd->qd_flags);
445 static void qd_unlock(struct gfs2_quota_data *qd)
447 gfs2_assert_warn(qd->qd_gl->gl_sbd,
448 test_bit(QDF_LOCKED, &qd->qd_flags));
449 clear_bit(QDF_LOCKED, &qd->qd_flags);
455 static int qdsb_get(struct gfs2_sbd *sdp, struct kqid qid,
456 struct gfs2_quota_data **qdp)
460 error = qd_get(sdp, qid, qdp);
464 error = slot_get(*qdp);
468 error = bh_get(*qdp);
481 static void qdsb_put(struct gfs2_quota_data *qd)
488 int gfs2_quota_hold(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
490 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
491 struct gfs2_quota_data **qd;
494 if (ip->i_res == NULL) {
495 error = gfs2_rs_alloc(ip);
500 qd = ip->i_res->rs_qa_qd;
502 if (gfs2_assert_warn(sdp, !ip->i_res->rs_qa_qd_num) ||
503 gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags)))
506 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
509 error = qdsb_get(sdp, make_kqid_uid(ip->i_inode.i_uid), qd);
512 ip->i_res->rs_qa_qd_num++;
515 error = qdsb_get(sdp, make_kqid_gid(ip->i_inode.i_gid), qd);
518 ip->i_res->rs_qa_qd_num++;
521 if (!uid_eq(uid, NO_UID_QUOTA_CHANGE) &&
522 !uid_eq(uid, ip->i_inode.i_uid)) {
523 error = qdsb_get(sdp, make_kqid_uid(uid), qd);
526 ip->i_res->rs_qa_qd_num++;
530 if (!gid_eq(gid, NO_GID_QUOTA_CHANGE) &&
531 !gid_eq(gid, ip->i_inode.i_gid)) {
532 error = qdsb_get(sdp, make_kqid_gid(gid), qd);
535 ip->i_res->rs_qa_qd_num++;
541 gfs2_quota_unhold(ip);
545 void gfs2_quota_unhold(struct gfs2_inode *ip)
547 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
550 if (ip->i_res == NULL)
552 gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags));
554 for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
555 qdsb_put(ip->i_res->rs_qa_qd[x]);
556 ip->i_res->rs_qa_qd[x] = NULL;
558 ip->i_res->rs_qa_qd_num = 0;
561 static int sort_qd(const void *a, const void *b)
563 const struct gfs2_quota_data *qd_a = *(const struct gfs2_quota_data **)a;
564 const struct gfs2_quota_data *qd_b = *(const struct gfs2_quota_data **)b;
566 if (qid_lt(qd_a->qd_id, qd_b->qd_id))
568 if (qid_lt(qd_b->qd_id, qd_a->qd_id))
573 static void do_qc(struct gfs2_quota_data *qd, s64 change)
575 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
576 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
577 struct gfs2_quota_change *qc = qd->qd_bh_qc;
580 mutex_lock(&sdp->sd_quota_mutex);
581 gfs2_trans_add_meta(ip->i_gl, qd->qd_bh);
583 if (!test_bit(QDF_CHANGE, &qd->qd_flags)) {
586 if (qd->qd_id.type == USRQUOTA)
587 qc->qc_flags = cpu_to_be32(GFS2_QCF_USER);
588 qc->qc_id = cpu_to_be32(from_kqid(&init_user_ns, qd->qd_id));
591 x = be64_to_cpu(qc->qc_change) + change;
592 qc->qc_change = cpu_to_be64(x);
594 spin_lock(&qd_lru_lock);
596 spin_unlock(&qd_lru_lock);
599 gfs2_assert_warn(sdp, test_bit(QDF_CHANGE, &qd->qd_flags));
600 clear_bit(QDF_CHANGE, &qd->qd_flags);
605 } else if (!test_and_set_bit(QDF_CHANGE, &qd->qd_flags)) {
610 mutex_unlock(&sdp->sd_quota_mutex);
614 * gfs2_adjust_quota - adjust record of current block usage
615 * @ip: The quota inode
616 * @loc: Offset of the entry in the quota file
617 * @change: The amount of usage change to record
618 * @qd: The quota data
619 * @fdq: The updated limits to record
621 * This function was mostly borrowed from gfs2_block_truncate_page which was
622 * in turn mostly borrowed from ext3
624 * Returns: 0 or -ve on error
627 static int gfs2_adjust_quota(struct gfs2_inode *ip, loff_t loc,
628 s64 change, struct gfs2_quota_data *qd,
629 struct fs_disk_quota *fdq)
631 struct inode *inode = &ip->i_inode;
632 struct gfs2_sbd *sdp = GFS2_SB(inode);
633 struct address_space *mapping = inode->i_mapping;
634 unsigned long index = loc >> PAGE_CACHE_SHIFT;
635 unsigned offset = loc & (PAGE_CACHE_SIZE - 1);
636 unsigned blocksize, iblock, pos;
637 struct buffer_head *bh;
640 struct gfs2_quota q, *qp;
644 if (gfs2_is_stuffed(ip)) {
645 err = gfs2_unstuff_dinode(ip, NULL);
650 memset(&q, 0, sizeof(struct gfs2_quota));
651 err = gfs2_internal_read(ip, (char *)&q, &loc, sizeof(q));
657 qp->qu_value = be64_to_cpu(qp->qu_value);
658 qp->qu_value += change;
659 qp->qu_value = cpu_to_be64(qp->qu_value);
660 qd->qd_qb.qb_value = qp->qu_value;
662 if (fdq->d_fieldmask & FS_DQ_BSOFT) {
663 qp->qu_warn = cpu_to_be64(fdq->d_blk_softlimit >> sdp->sd_fsb2bb_shift);
664 qd->qd_qb.qb_warn = qp->qu_warn;
666 if (fdq->d_fieldmask & FS_DQ_BHARD) {
667 qp->qu_limit = cpu_to_be64(fdq->d_blk_hardlimit >> sdp->sd_fsb2bb_shift);
668 qd->qd_qb.qb_limit = qp->qu_limit;
670 if (fdq->d_fieldmask & FS_DQ_BCOUNT) {
671 qp->qu_value = cpu_to_be64(fdq->d_bcount >> sdp->sd_fsb2bb_shift);
672 qd->qd_qb.qb_value = qp->qu_value;
676 /* Write the quota into the quota file on disk */
678 nbytes = sizeof(struct gfs2_quota);
680 page = find_or_create_page(mapping, index, GFP_NOFS);
684 blocksize = inode->i_sb->s_blocksize;
685 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
687 if (!page_has_buffers(page))
688 create_empty_buffers(page, blocksize, 0);
690 bh = page_buffers(page);
692 while (offset >= pos) {
693 bh = bh->b_this_page;
698 if (!buffer_mapped(bh)) {
699 gfs2_block_map(inode, iblock, bh, 1);
700 if (!buffer_mapped(bh))
702 /* If it's a newly allocated disk block for quota, zero it */
704 zero_user(page, pos - blocksize, bh->b_size);
707 if (PageUptodate(page))
708 set_buffer_uptodate(bh);
710 if (!buffer_uptodate(bh)) {
711 ll_rw_block(READ | REQ_META, 1, &bh);
713 if (!buffer_uptodate(bh))
717 gfs2_trans_add_data(ip->i_gl, bh);
719 kaddr = kmap_atomic(page);
720 if (offset + sizeof(struct gfs2_quota) > PAGE_CACHE_SIZE)
721 nbytes = PAGE_CACHE_SIZE - offset;
722 memcpy(kaddr + offset, ptr, nbytes);
723 flush_dcache_page(page);
724 kunmap_atomic(kaddr);
726 page_cache_release(page);
728 /* If quota straddles page boundary, we need to update the rest of the
729 * quota at the beginning of the next page */
730 if ((offset + sizeof(struct gfs2_quota)) > PAGE_CACHE_SIZE) {
732 nbytes = sizeof(struct gfs2_quota) - nbytes;
738 size = loc + sizeof(struct gfs2_quota);
739 if (size > inode->i_size)
740 i_size_write(inode, size);
741 inode->i_mtime = inode->i_atime = CURRENT_TIME;
742 mark_inode_dirty(inode);
747 page_cache_release(page);
751 static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
753 struct gfs2_sbd *sdp = (*qda)->qd_gl->gl_sbd;
754 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
755 struct gfs2_alloc_parms ap = { .aflags = 0, };
756 unsigned int data_blocks, ind_blocks;
757 struct gfs2_holder *ghs, i_gh;
759 struct gfs2_quota_data *qd;
762 unsigned int nalloc = 0, blocks;
765 error = gfs2_rs_alloc(ip);
769 gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
770 &data_blocks, &ind_blocks);
772 ghs = kcalloc(num_qd, sizeof(struct gfs2_holder), GFP_NOFS);
776 sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL);
777 mutex_lock(&ip->i_inode.i_mutex);
778 for (qx = 0; qx < num_qd; qx++) {
779 error = gfs2_glock_nq_init(qda[qx]->qd_gl, LM_ST_EXCLUSIVE,
780 GL_NOCACHE, &ghs[qx]);
785 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
789 for (x = 0; x < num_qd; x++) {
790 offset = qd2offset(qda[x]);
791 if (gfs2_write_alloc_required(ip, offset,
792 sizeof(struct gfs2_quota)))
797 * 1 blk for unstuffing inode if stuffed. We add this extra
798 * block to the reservation unconditionally. If the inode
799 * doesn't need unstuffing, the block will be released to the
800 * rgrp since it won't be allocated during the transaction
802 /* +3 in the end for unstuffing block, inode size update block
803 * and another block in case quota straddles page boundary and
804 * two blocks need to be updated instead of 1 */
805 blocks = num_qd * data_blocks + RES_DINODE + num_qd + 3;
807 reserved = 1 + (nalloc * (data_blocks + ind_blocks));
808 ap.target = reserved;
809 error = gfs2_inplace_reserve(ip, &ap);
814 blocks += gfs2_rg_blocks(ip, reserved) + nalloc * ind_blocks + RES_STATFS;
816 error = gfs2_trans_begin(sdp, blocks, 0);
820 for (x = 0; x < num_qd; x++) {
822 offset = qd2offset(qd);
823 error = gfs2_adjust_quota(ip, offset, qd->qd_change_sync, qd, NULL);
827 do_qc(qd, -qd->qd_change_sync);
828 set_bit(QDF_REFRESH, &qd->qd_flags);
836 gfs2_inplace_release(ip);
838 gfs2_glock_dq_uninit(&i_gh);
841 gfs2_glock_dq_uninit(&ghs[qx]);
842 mutex_unlock(&ip->i_inode.i_mutex);
844 gfs2_log_flush(ip->i_gl->gl_sbd, ip->i_gl);
848 static int update_qd(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd)
850 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
852 struct gfs2_quota_lvb *qlvb;
856 memset(&q, 0, sizeof(struct gfs2_quota));
858 error = gfs2_internal_read(ip, (char *)&q, &pos, sizeof(q));
862 qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
863 qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC);
865 qlvb->qb_limit = q.qu_limit;
866 qlvb->qb_warn = q.qu_warn;
867 qlvb->qb_value = q.qu_value;
873 static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
874 struct gfs2_holder *q_gh)
876 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
877 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
878 struct gfs2_holder i_gh;
882 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
886 qd->qd_qb = *(struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
888 if (force_refresh || qd->qd_qb.qb_magic != cpu_to_be32(GFS2_MAGIC)) {
889 gfs2_glock_dq_uninit(q_gh);
890 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE,
895 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
899 error = update_qd(sdp, qd);
903 gfs2_glock_dq_uninit(&i_gh);
904 gfs2_glock_dq_uninit(q_gh);
912 gfs2_glock_dq_uninit(&i_gh);
914 gfs2_glock_dq_uninit(q_gh);
918 int gfs2_quota_lock(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
920 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
921 struct gfs2_quota_data *qd;
925 error = gfs2_quota_hold(ip, uid, gid);
929 if (capable(CAP_SYS_RESOURCE) ||
930 sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
933 sort(ip->i_res->rs_qa_qd, ip->i_res->rs_qa_qd_num,
934 sizeof(struct gfs2_quota_data *), sort_qd, NULL);
936 for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
937 int force = NO_FORCE;
938 qd = ip->i_res->rs_qa_qd[x];
939 if (test_and_clear_bit(QDF_REFRESH, &qd->qd_flags))
941 error = do_glock(qd, force, &ip->i_res->rs_qa_qd_ghs[x]);
947 set_bit(GIF_QD_LOCKED, &ip->i_flags);
950 gfs2_glock_dq_uninit(&ip->i_res->rs_qa_qd_ghs[x]);
951 gfs2_quota_unhold(ip);
957 static int need_sync(struct gfs2_quota_data *qd)
959 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
960 struct gfs2_tune *gt = &sdp->sd_tune;
962 unsigned int num, den;
965 if (!qd->qd_qb.qb_limit)
968 spin_lock(&qd_lru_lock);
969 value = qd->qd_change;
970 spin_unlock(&qd_lru_lock);
972 spin_lock(>->gt_spin);
973 num = gt->gt_quota_scale_num;
974 den = gt->gt_quota_scale_den;
975 spin_unlock(>->gt_spin);
979 else if ((s64)be64_to_cpu(qd->qd_qb.qb_value) >=
980 (s64)be64_to_cpu(qd->qd_qb.qb_limit))
983 value *= gfs2_jindex_size(sdp) * num;
984 value = div_s64(value, den);
985 value += (s64)be64_to_cpu(qd->qd_qb.qb_value);
986 if (value < (s64)be64_to_cpu(qd->qd_qb.qb_limit))
993 void gfs2_quota_unlock(struct gfs2_inode *ip)
995 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
996 struct gfs2_quota_data *qda[4];
997 unsigned int count = 0;
1001 if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
1004 for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1005 struct gfs2_quota_data *qd;
1008 qd = ip->i_res->rs_qa_qd[x];
1009 sync = need_sync(qd);
1011 gfs2_glock_dq_uninit(&ip->i_res->rs_qa_qd_ghs[x]);
1015 spin_lock(&qd_lru_lock);
1016 found = qd_check_sync(sdp, qd, NULL);
1017 spin_unlock(&qd_lru_lock);
1022 gfs2_assert_warn(sdp, qd->qd_change_sync);
1024 clear_bit(QDF_LOCKED, &qd->qd_flags);
1034 do_sync(count, qda);
1035 for (x = 0; x < count; x++)
1040 gfs2_quota_unhold(ip);
1043 #define MAX_LINE 256
1045 static int print_message(struct gfs2_quota_data *qd, char *type)
1047 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
1049 printk(KERN_INFO "GFS2: fsid=%s: quota %s for %s %u\n",
1050 sdp->sd_fsname, type,
1051 (qd->qd_id.type == USRQUOTA) ? "user" : "group",
1052 from_kqid(&init_user_ns, qd->qd_id));
1057 int gfs2_quota_check(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
1059 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1060 struct gfs2_quota_data *qd;
1065 if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
1068 if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
1071 for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1072 qd = ip->i_res->rs_qa_qd[x];
1074 if (!(qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1075 qid_eq(qd->qd_id, make_kqid_gid(gid))))
1078 value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
1079 spin_lock(&qd_lru_lock);
1080 value += qd->qd_change;
1081 spin_unlock(&qd_lru_lock);
1083 if (be64_to_cpu(qd->qd_qb.qb_limit) && (s64)be64_to_cpu(qd->qd_qb.qb_limit) < value) {
1084 print_message(qd, "exceeded");
1085 quota_send_warning(qd->qd_id,
1086 sdp->sd_vfs->s_dev, QUOTA_NL_BHARDWARN);
1090 } else if (be64_to_cpu(qd->qd_qb.qb_warn) &&
1091 (s64)be64_to_cpu(qd->qd_qb.qb_warn) < value &&
1092 time_after_eq(jiffies, qd->qd_last_warn +
1094 gt_quota_warn_period) * HZ)) {
1095 quota_send_warning(qd->qd_id,
1096 sdp->sd_vfs->s_dev, QUOTA_NL_BSOFTWARN);
1097 error = print_message(qd, "warning");
1098 qd->qd_last_warn = jiffies;
1105 void gfs2_quota_change(struct gfs2_inode *ip, s64 change,
1106 kuid_t uid, kgid_t gid)
1108 struct gfs2_quota_data *qd;
1111 if (gfs2_assert_warn(GFS2_SB(&ip->i_inode), change))
1113 if (ip->i_diskflags & GFS2_DIF_SYSTEM)
1116 for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1117 qd = ip->i_res->rs_qa_qd[x];
1119 if (qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1120 qid_eq(qd->qd_id, make_kqid_gid(gid))) {
1126 int gfs2_quota_sync(struct super_block *sb, int type)
1128 struct gfs2_sbd *sdp = sb->s_fs_info;
1129 struct gfs2_quota_data **qda;
1130 unsigned int max_qd = PAGE_SIZE/sizeof(struct gfs2_holder);
1131 unsigned int num_qd;
1135 qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
1139 mutex_lock(&sdp->sd_quota_sync_mutex);
1140 sdp->sd_quota_sync_gen++;
1146 error = qd_fish(sdp, qda + num_qd);
1147 if (error || !qda[num_qd])
1149 if (++num_qd == max_qd)
1155 error = do_sync(num_qd, qda);
1157 for (x = 0; x < num_qd; x++)
1158 qda[x]->qd_sync_gen =
1159 sdp->sd_quota_sync_gen;
1161 for (x = 0; x < num_qd; x++)
1164 } while (!error && num_qd == max_qd);
1166 mutex_unlock(&sdp->sd_quota_sync_mutex);
1172 int gfs2_quota_refresh(struct gfs2_sbd *sdp, struct kqid qid)
1174 struct gfs2_quota_data *qd;
1175 struct gfs2_holder q_gh;
1178 error = qd_get(sdp, qid, &qd);
1182 error = do_glock(qd, FORCE, &q_gh);
1184 gfs2_glock_dq_uninit(&q_gh);
1190 static void gfs2_quota_change_in(struct gfs2_quota_change_host *qc, const void *buf)
1192 const struct gfs2_quota_change *str = buf;
1194 qc->qc_change = be64_to_cpu(str->qc_change);
1195 qc->qc_flags = be32_to_cpu(str->qc_flags);
1196 qc->qc_id = make_kqid(&init_user_ns,
1197 (qc->qc_flags & GFS2_QCF_USER)?USRQUOTA:GRPQUOTA,
1198 be32_to_cpu(str->qc_id));
1201 int gfs2_quota_init(struct gfs2_sbd *sdp)
1203 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
1204 u64 size = i_size_read(sdp->sd_qc_inode);
1205 unsigned int blocks = size >> sdp->sd_sb.sb_bsize_shift;
1206 unsigned int x, slot = 0;
1207 unsigned int found = 0;
1212 if (gfs2_check_internal_file_size(sdp->sd_qc_inode, 1, 64 << 20))
1215 sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
1216 sdp->sd_quota_chunks = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * PAGE_SIZE);
1220 sdp->sd_quota_bitmap = kcalloc(sdp->sd_quota_chunks,
1221 sizeof(unsigned char *), GFP_NOFS);
1222 if (!sdp->sd_quota_bitmap)
1225 for (x = 0; x < sdp->sd_quota_chunks; x++) {
1226 sdp->sd_quota_bitmap[x] = kzalloc(PAGE_SIZE, GFP_NOFS);
1227 if (!sdp->sd_quota_bitmap[x])
1231 for (x = 0; x < blocks; x++) {
1232 struct buffer_head *bh;
1237 error = gfs2_extent_map(&ip->i_inode, x, &new, &dblock, &extlen);
1242 bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
1245 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
1250 for (y = 0; y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1252 struct gfs2_quota_change_host qc;
1253 struct gfs2_quota_data *qd;
1255 gfs2_quota_change_in(&qc, bh->b_data +
1256 sizeof(struct gfs2_meta_header) +
1257 y * sizeof(struct gfs2_quota_change));
1261 error = qd_alloc(sdp, qc.qc_id, &qd);
1267 set_bit(QDF_CHANGE, &qd->qd_flags);
1268 qd->qd_change = qc.qc_change;
1270 qd->qd_slot_count = 1;
1272 spin_lock(&qd_lru_lock);
1273 gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, slot, 1);
1274 list_add(&qd->qd_list, &sdp->sd_quota_list);
1275 atomic_inc(&sdp->sd_quota_count);
1276 spin_unlock(&qd_lru_lock);
1287 fs_info(sdp, "found %u quota changes\n", found);
1292 gfs2_quota_cleanup(sdp);
1296 void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1298 struct list_head *head = &sdp->sd_quota_list;
1299 struct gfs2_quota_data *qd;
1302 spin_lock(&qd_lru_lock);
1303 while (!list_empty(head)) {
1304 qd = list_entry(head->prev, struct gfs2_quota_data, qd_list);
1306 if (atomic_read(&qd->qd_count) > 1 ||
1307 (atomic_read(&qd->qd_count) &&
1308 !test_bit(QDF_CHANGE, &qd->qd_flags))) {
1309 list_move(&qd->qd_list, head);
1310 spin_unlock(&qd_lru_lock);
1312 spin_lock(&qd_lru_lock);
1316 list_del(&qd->qd_list);
1317 /* Also remove if this qd exists in the reclaim list */
1318 if (!list_empty(&qd->qd_reclaim)) {
1319 list_del_init(&qd->qd_reclaim);
1320 atomic_dec(&qd_lru_count);
1322 atomic_dec(&sdp->sd_quota_count);
1323 spin_unlock(&qd_lru_lock);
1325 if (!atomic_read(&qd->qd_count)) {
1326 gfs2_assert_warn(sdp, !qd->qd_change);
1327 gfs2_assert_warn(sdp, !qd->qd_slot_count);
1329 gfs2_assert_warn(sdp, qd->qd_slot_count == 1);
1330 gfs2_assert_warn(sdp, !qd->qd_bh_count);
1332 gfs2_glock_put(qd->qd_gl);
1333 kmem_cache_free(gfs2_quotad_cachep, qd);
1335 spin_lock(&qd_lru_lock);
1337 spin_unlock(&qd_lru_lock);
1339 gfs2_assert_warn(sdp, !atomic_read(&sdp->sd_quota_count));
1341 if (sdp->sd_quota_bitmap) {
1342 for (x = 0; x < sdp->sd_quota_chunks; x++)
1343 kfree(sdp->sd_quota_bitmap[x]);
1344 kfree(sdp->sd_quota_bitmap);
1348 static void quotad_error(struct gfs2_sbd *sdp, const char *msg, int error)
1350 if (error == 0 || error == -EROFS)
1352 if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
1353 fs_err(sdp, "gfs2_quotad: %s error %d\n", msg, error);
1356 static void quotad_check_timeo(struct gfs2_sbd *sdp, const char *msg,
1357 int (*fxn)(struct super_block *sb, int type),
1358 unsigned long t, unsigned long *timeo,
1359 unsigned int *new_timeo)
1362 int error = fxn(sdp->sd_vfs, 0);
1363 quotad_error(sdp, msg, error);
1364 *timeo = gfs2_tune_get_i(&sdp->sd_tune, new_timeo) * HZ;
1370 static void quotad_check_trunc_list(struct gfs2_sbd *sdp)
1372 struct gfs2_inode *ip;
1376 spin_lock(&sdp->sd_trunc_lock);
1377 if (!list_empty(&sdp->sd_trunc_list)) {
1378 ip = list_entry(sdp->sd_trunc_list.next,
1379 struct gfs2_inode, i_trunc_list);
1380 list_del_init(&ip->i_trunc_list);
1382 spin_unlock(&sdp->sd_trunc_lock);
1385 gfs2_glock_finish_truncate(ip);
1389 void gfs2_wake_up_statfs(struct gfs2_sbd *sdp) {
1390 if (!sdp->sd_statfs_force_sync) {
1391 sdp->sd_statfs_force_sync = 1;
1392 wake_up(&sdp->sd_quota_wait);
1398 * gfs2_quotad - Write cached quota changes into the quota file
1399 * @sdp: Pointer to GFS2 superblock
1403 int gfs2_quotad(void *data)
1405 struct gfs2_sbd *sdp = data;
1406 struct gfs2_tune *tune = &sdp->sd_tune;
1407 unsigned long statfs_timeo = 0;
1408 unsigned long quotad_timeo = 0;
1409 unsigned long t = 0;
1413 while (!kthread_should_stop()) {
1415 /* Update the master statfs file */
1416 if (sdp->sd_statfs_force_sync) {
1417 int error = gfs2_statfs_sync(sdp->sd_vfs, 0);
1418 quotad_error(sdp, "statfs", error);
1419 statfs_timeo = gfs2_tune_get(sdp, gt_statfs_quantum) * HZ;
1422 quotad_check_timeo(sdp, "statfs", gfs2_statfs_sync, t,
1424 &tune->gt_statfs_quantum);
1426 /* Update quota file */
1427 quotad_check_timeo(sdp, "sync", gfs2_quota_sync, t,
1428 "ad_timeo, &tune->gt_quota_quantum);
1430 /* Check for & recover partially truncated inodes */
1431 quotad_check_trunc_list(sdp);
1435 t = min(quotad_timeo, statfs_timeo);
1437 prepare_to_wait(&sdp->sd_quota_wait, &wait, TASK_INTERRUPTIBLE);
1438 spin_lock(&sdp->sd_trunc_lock);
1439 empty = list_empty(&sdp->sd_trunc_list);
1440 spin_unlock(&sdp->sd_trunc_lock);
1441 if (empty && !sdp->sd_statfs_force_sync)
1442 t -= schedule_timeout(t);
1445 finish_wait(&sdp->sd_quota_wait, &wait);
1451 static int gfs2_quota_get_xstate(struct super_block *sb,
1452 struct fs_quota_stat *fqs)
1454 struct gfs2_sbd *sdp = sb->s_fs_info;
1456 memset(fqs, 0, sizeof(struct fs_quota_stat));
1457 fqs->qs_version = FS_QSTAT_VERSION;
1459 switch (sdp->sd_args.ar_quota) {
1461 fqs->qs_flags |= (FS_QUOTA_UDQ_ENFD | FS_QUOTA_GDQ_ENFD);
1463 case GFS2_QUOTA_ACCOUNT:
1464 fqs->qs_flags |= (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT);
1466 case GFS2_QUOTA_OFF:
1470 if (sdp->sd_quota_inode) {
1471 fqs->qs_uquota.qfs_ino = GFS2_I(sdp->sd_quota_inode)->i_no_addr;
1472 fqs->qs_uquota.qfs_nblks = sdp->sd_quota_inode->i_blocks;
1474 fqs->qs_uquota.qfs_nextents = 1; /* unsupported */
1475 fqs->qs_gquota = fqs->qs_uquota; /* its the same inode in both cases */
1476 fqs->qs_incoredqs = atomic_read(&qd_lru_count);
1480 static int gfs2_get_dqblk(struct super_block *sb, struct kqid qid,
1481 struct fs_disk_quota *fdq)
1483 struct gfs2_sbd *sdp = sb->s_fs_info;
1484 struct gfs2_quota_lvb *qlvb;
1485 struct gfs2_quota_data *qd;
1486 struct gfs2_holder q_gh;
1489 memset(fdq, 0, sizeof(struct fs_disk_quota));
1491 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1492 return -ESRCH; /* Crazy XFS error code */
1494 if ((qid.type != USRQUOTA) &&
1495 (qid.type != GRPQUOTA))
1498 error = qd_get(sdp, qid, &qd);
1501 error = do_glock(qd, FORCE, &q_gh);
1505 qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1506 fdq->d_version = FS_DQUOT_VERSION;
1507 fdq->d_flags = (qid.type == USRQUOTA) ? FS_USER_QUOTA : FS_GROUP_QUOTA;
1508 fdq->d_id = from_kqid_munged(current_user_ns(), qid);
1509 fdq->d_blk_hardlimit = be64_to_cpu(qlvb->qb_limit) << sdp->sd_fsb2bb_shift;
1510 fdq->d_blk_softlimit = be64_to_cpu(qlvb->qb_warn) << sdp->sd_fsb2bb_shift;
1511 fdq->d_bcount = be64_to_cpu(qlvb->qb_value) << sdp->sd_fsb2bb_shift;
1513 gfs2_glock_dq_uninit(&q_gh);
1519 /* GFS2 only supports a subset of the XFS fields */
1520 #define GFS2_FIELDMASK (FS_DQ_BSOFT|FS_DQ_BHARD|FS_DQ_BCOUNT)
1522 static int gfs2_set_dqblk(struct super_block *sb, struct kqid qid,
1523 struct fs_disk_quota *fdq)
1525 struct gfs2_sbd *sdp = sb->s_fs_info;
1526 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1527 struct gfs2_quota_data *qd;
1528 struct gfs2_holder q_gh, i_gh;
1529 unsigned int data_blocks, ind_blocks;
1530 unsigned int blocks = 0;
1535 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1536 return -ESRCH; /* Crazy XFS error code */
1538 if ((qid.type != USRQUOTA) &&
1539 (qid.type != GRPQUOTA))
1542 if (fdq->d_fieldmask & ~GFS2_FIELDMASK)
1545 error = qd_get(sdp, qid, &qd);
1549 error = gfs2_rs_alloc(ip);
1553 mutex_lock(&ip->i_inode.i_mutex);
1554 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE, 0, &q_gh);
1557 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1561 /* Check for existing entry, if none then alloc new blocks */
1562 error = update_qd(sdp, qd);
1566 /* If nothing has changed, this is a no-op */
1567 if ((fdq->d_fieldmask & FS_DQ_BSOFT) &&
1568 ((fdq->d_blk_softlimit >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_warn)))
1569 fdq->d_fieldmask ^= FS_DQ_BSOFT;
1571 if ((fdq->d_fieldmask & FS_DQ_BHARD) &&
1572 ((fdq->d_blk_hardlimit >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_limit)))
1573 fdq->d_fieldmask ^= FS_DQ_BHARD;
1575 if ((fdq->d_fieldmask & FS_DQ_BCOUNT) &&
1576 ((fdq->d_bcount >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_value)))
1577 fdq->d_fieldmask ^= FS_DQ_BCOUNT;
1579 if (fdq->d_fieldmask == 0)
1582 offset = qd2offset(qd);
1583 alloc_required = gfs2_write_alloc_required(ip, offset, sizeof(struct gfs2_quota));
1584 if (gfs2_is_stuffed(ip))
1586 if (alloc_required) {
1587 struct gfs2_alloc_parms ap = { .aflags = 0, };
1588 gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
1589 &data_blocks, &ind_blocks);
1590 blocks = 1 + data_blocks + ind_blocks;
1592 error = gfs2_inplace_reserve(ip, &ap);
1595 blocks += gfs2_rg_blocks(ip, blocks);
1598 /* Some quotas span block boundaries and can update two blocks,
1599 adding an extra block to the transaction to handle such quotas */
1600 error = gfs2_trans_begin(sdp, blocks + RES_DINODE + 2, 0);
1605 error = gfs2_adjust_quota(ip, offset, 0, qd, fdq);
1607 gfs2_trans_end(sdp);
1610 gfs2_inplace_release(ip);
1612 gfs2_glock_dq_uninit(&i_gh);
1614 gfs2_glock_dq_uninit(&q_gh);
1616 mutex_unlock(&ip->i_inode.i_mutex);
1622 const struct quotactl_ops gfs2_quotactl_ops = {
1623 .quota_sync = gfs2_quota_sync,
1624 .get_xstate = gfs2_quota_get_xstate,
1625 .get_dqblk = gfs2_get_dqblk,
1626 .set_dqblk = gfs2_set_dqblk,