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 to 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/lm_interface.h>
49 #include <linux/kthread.h>
50 #include <linux/freezer.h>
64 #include "ops_address.h"
70 struct gfs2_quota_host
{
77 struct gfs2_quota_change_host
{
79 u32 qc_flags
; /* GFS2_QCF_... */
83 static u64
qd2offset(struct gfs2_quota_data
*qd
)
87 offset
= 2 * (u64
)qd
->qd_id
+ !test_bit(QDF_USER
, &qd
->qd_flags
);
88 offset
*= sizeof(struct gfs2_quota
);
93 static int qd_alloc(struct gfs2_sbd
*sdp
, int user
, u32 id
,
94 struct gfs2_quota_data
**qdp
)
96 struct gfs2_quota_data
*qd
;
99 qd
= kmem_cache_zalloc(gfs2_quotad_cachep
, GFP_NOFS
);
106 set_bit(QDF_USER
, &qd
->qd_flags
);
109 error
= gfs2_glock_get(sdp
, 2 * (u64
)id
+ !user
,
110 &gfs2_quota_glops
, CREATE
, &qd
->qd_gl
);
114 error
= gfs2_lvb_hold(qd
->qd_gl
);
115 gfs2_glock_put(qd
->qd_gl
);
124 kmem_cache_free(gfs2_quotad_cachep
, qd
);
128 static int qd_get(struct gfs2_sbd
*sdp
, int user
, u32 id
, int create
,
129 struct gfs2_quota_data
**qdp
)
131 struct gfs2_quota_data
*qd
= NULL
, *new_qd
= NULL
;
138 spin_lock(&sdp
->sd_quota_spin
);
139 list_for_each_entry(qd
, &sdp
->sd_quota_list
, qd_list
) {
140 if (qd
->qd_id
== id
&&
141 !test_bit(QDF_USER
, &qd
->qd_flags
) == !user
) {
153 list_add(&qd
->qd_list
, &sdp
->sd_quota_list
);
154 atomic_inc(&sdp
->sd_quota_count
);
158 spin_unlock(&sdp
->sd_quota_spin
);
162 gfs2_lvb_unhold(new_qd
->qd_gl
);
163 kmem_cache_free(gfs2_quotad_cachep
, new_qd
);
169 error
= qd_alloc(sdp
, user
, id
, &new_qd
);
175 static void qd_hold(struct gfs2_quota_data
*qd
)
177 struct gfs2_sbd
*sdp
= qd
->qd_gl
->gl_sbd
;
179 spin_lock(&sdp
->sd_quota_spin
);
180 gfs2_assert(sdp
, qd
->qd_count
);
182 spin_unlock(&sdp
->sd_quota_spin
);
185 static void qd_put(struct gfs2_quota_data
*qd
)
187 struct gfs2_sbd
*sdp
= qd
->qd_gl
->gl_sbd
;
188 spin_lock(&sdp
->sd_quota_spin
);
189 gfs2_assert(sdp
, qd
->qd_count
);
191 qd
->qd_last_touched
= jiffies
;
192 spin_unlock(&sdp
->sd_quota_spin
);
195 static int slot_get(struct gfs2_quota_data
*qd
)
197 struct gfs2_sbd
*sdp
= qd
->qd_gl
->gl_sbd
;
198 unsigned int c
, o
= 0, b
;
199 unsigned char byte
= 0;
201 spin_lock(&sdp
->sd_quota_spin
);
203 if (qd
->qd_slot_count
++) {
204 spin_unlock(&sdp
->sd_quota_spin
);
208 for (c
= 0; c
< sdp
->sd_quota_chunks
; c
++)
209 for (o
= 0; o
< PAGE_SIZE
; o
++) {
210 byte
= sdp
->sd_quota_bitmap
[c
][o
];
218 for (b
= 0; b
< 8; b
++)
219 if (!(byte
& (1 << b
)))
221 qd
->qd_slot
= c
* (8 * PAGE_SIZE
) + o
* 8 + b
;
223 if (qd
->qd_slot
>= sdp
->sd_quota_slots
)
226 sdp
->sd_quota_bitmap
[c
][o
] |= 1 << b
;
228 spin_unlock(&sdp
->sd_quota_spin
);
234 spin_unlock(&sdp
->sd_quota_spin
);
238 static void slot_hold(struct gfs2_quota_data
*qd
)
240 struct gfs2_sbd
*sdp
= qd
->qd_gl
->gl_sbd
;
242 spin_lock(&sdp
->sd_quota_spin
);
243 gfs2_assert(sdp
, qd
->qd_slot_count
);
245 spin_unlock(&sdp
->sd_quota_spin
);
248 static void slot_put(struct gfs2_quota_data
*qd
)
250 struct gfs2_sbd
*sdp
= qd
->qd_gl
->gl_sbd
;
252 spin_lock(&sdp
->sd_quota_spin
);
253 gfs2_assert(sdp
, qd
->qd_slot_count
);
254 if (!--qd
->qd_slot_count
) {
255 gfs2_icbit_munge(sdp
, sdp
->sd_quota_bitmap
, qd
->qd_slot
, 0);
258 spin_unlock(&sdp
->sd_quota_spin
);
261 static int bh_get(struct gfs2_quota_data
*qd
)
263 struct gfs2_sbd
*sdp
= qd
->qd_gl
->gl_sbd
;
264 struct gfs2_inode
*ip
= GFS2_I(sdp
->sd_qc_inode
);
265 unsigned int block
, offset
;
266 struct buffer_head
*bh
;
268 struct buffer_head bh_map
= { .b_state
= 0, .b_blocknr
= 0 };
270 mutex_lock(&sdp
->sd_quota_mutex
);
272 if (qd
->qd_bh_count
++) {
273 mutex_unlock(&sdp
->sd_quota_mutex
);
277 block
= qd
->qd_slot
/ sdp
->sd_qc_per_block
;
278 offset
= qd
->qd_slot
% sdp
->sd_qc_per_block
;
280 bh_map
.b_size
= 1 << ip
->i_inode
.i_blkbits
;
281 error
= gfs2_block_map(&ip
->i_inode
, block
, &bh_map
, 0);
284 error
= gfs2_meta_read(ip
->i_gl
, bh_map
.b_blocknr
, DIO_WAIT
, &bh
);
288 if (gfs2_metatype_check(sdp
, bh
, GFS2_METATYPE_QC
))
292 qd
->qd_bh_qc
= (struct gfs2_quota_change
*)
293 (bh
->b_data
+ sizeof(struct gfs2_meta_header
) +
294 offset
* sizeof(struct gfs2_quota_change
));
296 mutex_unlock(&sdp
->sd_quota_mutex
);
304 mutex_unlock(&sdp
->sd_quota_mutex
);
308 static void bh_put(struct gfs2_quota_data
*qd
)
310 struct gfs2_sbd
*sdp
= qd
->qd_gl
->gl_sbd
;
312 mutex_lock(&sdp
->sd_quota_mutex
);
313 gfs2_assert(sdp
, qd
->qd_bh_count
);
314 if (!--qd
->qd_bh_count
) {
319 mutex_unlock(&sdp
->sd_quota_mutex
);
322 static int qd_fish(struct gfs2_sbd
*sdp
, struct gfs2_quota_data
**qdp
)
324 struct gfs2_quota_data
*qd
= NULL
;
330 if (sdp
->sd_vfs
->s_flags
& MS_RDONLY
)
333 spin_lock(&sdp
->sd_quota_spin
);
335 list_for_each_entry(qd
, &sdp
->sd_quota_list
, qd_list
) {
336 if (test_bit(QDF_LOCKED
, &qd
->qd_flags
) ||
337 !test_bit(QDF_CHANGE
, &qd
->qd_flags
) ||
338 qd
->qd_sync_gen
>= sdp
->sd_quota_sync_gen
)
341 list_move_tail(&qd
->qd_list
, &sdp
->sd_quota_list
);
343 set_bit(QDF_LOCKED
, &qd
->qd_flags
);
344 gfs2_assert_warn(sdp
, qd
->qd_count
);
346 qd
->qd_change_sync
= qd
->qd_change
;
347 gfs2_assert_warn(sdp
, qd
->qd_slot_count
);
357 spin_unlock(&sdp
->sd_quota_spin
);
360 gfs2_assert_warn(sdp
, qd
->qd_change_sync
);
363 clear_bit(QDF_LOCKED
, &qd
->qd_flags
);
375 static int qd_trylock(struct gfs2_quota_data
*qd
)
377 struct gfs2_sbd
*sdp
= qd
->qd_gl
->gl_sbd
;
379 if (sdp
->sd_vfs
->s_flags
& MS_RDONLY
)
382 spin_lock(&sdp
->sd_quota_spin
);
384 if (test_bit(QDF_LOCKED
, &qd
->qd_flags
) ||
385 !test_bit(QDF_CHANGE
, &qd
->qd_flags
)) {
386 spin_unlock(&sdp
->sd_quota_spin
);
390 list_move_tail(&qd
->qd_list
, &sdp
->sd_quota_list
);
392 set_bit(QDF_LOCKED
, &qd
->qd_flags
);
393 gfs2_assert_warn(sdp
, qd
->qd_count
);
395 qd
->qd_change_sync
= qd
->qd_change
;
396 gfs2_assert_warn(sdp
, qd
->qd_slot_count
);
399 spin_unlock(&sdp
->sd_quota_spin
);
401 gfs2_assert_warn(sdp
, qd
->qd_change_sync
);
403 clear_bit(QDF_LOCKED
, &qd
->qd_flags
);
412 static void qd_unlock(struct gfs2_quota_data
*qd
)
414 gfs2_assert_warn(qd
->qd_gl
->gl_sbd
,
415 test_bit(QDF_LOCKED
, &qd
->qd_flags
));
416 clear_bit(QDF_LOCKED
, &qd
->qd_flags
);
422 static int qdsb_get(struct gfs2_sbd
*sdp
, int user
, u32 id
, int create
,
423 struct gfs2_quota_data
**qdp
)
427 error
= qd_get(sdp
, user
, id
, create
, qdp
);
431 error
= slot_get(*qdp
);
435 error
= bh_get(*qdp
);
448 static void qdsb_put(struct gfs2_quota_data
*qd
)
455 int gfs2_quota_hold(struct gfs2_inode
*ip
, u32 uid
, u32 gid
)
457 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
458 struct gfs2_alloc
*al
= ip
->i_alloc
;
459 struct gfs2_quota_data
**qd
= al
->al_qd
;
462 if (gfs2_assert_warn(sdp
, !al
->al_qd_num
) ||
463 gfs2_assert_warn(sdp
, !test_bit(GIF_QD_LOCKED
, &ip
->i_flags
)))
466 if (sdp
->sd_args
.ar_quota
== GFS2_QUOTA_OFF
)
469 error
= qdsb_get(sdp
, QUOTA_USER
, ip
->i_inode
.i_uid
, CREATE
, qd
);
475 error
= qdsb_get(sdp
, QUOTA_GROUP
, ip
->i_inode
.i_gid
, CREATE
, qd
);
481 if (uid
!= NO_QUOTA_CHANGE
&& uid
!= ip
->i_inode
.i_uid
) {
482 error
= qdsb_get(sdp
, QUOTA_USER
, uid
, CREATE
, qd
);
489 if (gid
!= NO_QUOTA_CHANGE
&& gid
!= ip
->i_inode
.i_gid
) {
490 error
= qdsb_get(sdp
, QUOTA_GROUP
, gid
, CREATE
, qd
);
499 gfs2_quota_unhold(ip
);
503 void gfs2_quota_unhold(struct gfs2_inode
*ip
)
505 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
506 struct gfs2_alloc
*al
= ip
->i_alloc
;
509 gfs2_assert_warn(sdp
, !test_bit(GIF_QD_LOCKED
, &ip
->i_flags
));
511 for (x
= 0; x
< al
->al_qd_num
; x
++) {
512 qdsb_put(al
->al_qd
[x
]);
518 static int sort_qd(const void *a
, const void *b
)
520 const struct gfs2_quota_data
*qd_a
= *(const struct gfs2_quota_data
**)a
;
521 const struct gfs2_quota_data
*qd_b
= *(const struct gfs2_quota_data
**)b
;
523 if (!test_bit(QDF_USER
, &qd_a
->qd_flags
) !=
524 !test_bit(QDF_USER
, &qd_b
->qd_flags
)) {
525 if (test_bit(QDF_USER
, &qd_a
->qd_flags
))
530 if (qd_a
->qd_id
< qd_b
->qd_id
)
532 if (qd_a
->qd_id
> qd_b
->qd_id
)
538 static void do_qc(struct gfs2_quota_data
*qd
, s64 change
)
540 struct gfs2_sbd
*sdp
= qd
->qd_gl
->gl_sbd
;
541 struct gfs2_inode
*ip
= GFS2_I(sdp
->sd_qc_inode
);
542 struct gfs2_quota_change
*qc
= qd
->qd_bh_qc
;
545 mutex_lock(&sdp
->sd_quota_mutex
);
546 gfs2_trans_add_bh(ip
->i_gl
, qd
->qd_bh
, 1);
548 if (!test_bit(QDF_CHANGE
, &qd
->qd_flags
)) {
551 if (test_bit(QDF_USER
, &qd
->qd_flags
))
552 qc
->qc_flags
= cpu_to_be32(GFS2_QCF_USER
);
553 qc
->qc_id
= cpu_to_be32(qd
->qd_id
);
556 x
= be64_to_cpu(qc
->qc_change
) + change
;
557 qc
->qc_change
= cpu_to_be64(x
);
559 spin_lock(&sdp
->sd_quota_spin
);
561 spin_unlock(&sdp
->sd_quota_spin
);
564 gfs2_assert_warn(sdp
, test_bit(QDF_CHANGE
, &qd
->qd_flags
));
565 clear_bit(QDF_CHANGE
, &qd
->qd_flags
);
570 } else if (!test_and_set_bit(QDF_CHANGE
, &qd
->qd_flags
)) {
575 mutex_unlock(&sdp
->sd_quota_mutex
);
578 static void gfs2_quota_in(struct gfs2_quota_host
*qu
, const void *buf
)
580 const struct gfs2_quota
*str
= buf
;
582 qu
->qu_limit
= be64_to_cpu(str
->qu_limit
);
583 qu
->qu_warn
= be64_to_cpu(str
->qu_warn
);
584 qu
->qu_value
= be64_to_cpu(str
->qu_value
);
585 qu
->qu_ll_next
= be32_to_cpu(str
->qu_ll_next
);
588 static void gfs2_quota_out(const struct gfs2_quota_host
*qu
, void *buf
)
590 struct gfs2_quota
*str
= buf
;
592 str
->qu_limit
= cpu_to_be64(qu
->qu_limit
);
593 str
->qu_warn
= cpu_to_be64(qu
->qu_warn
);
594 str
->qu_value
= cpu_to_be64(qu
->qu_value
);
595 str
->qu_ll_next
= cpu_to_be32(qu
->qu_ll_next
);
596 memset(&str
->qu_reserved
, 0, sizeof(str
->qu_reserved
));
602 * This function was mostly borrowed from gfs2_block_truncate_page which was
603 * in turn mostly borrowed from ext3
605 static int gfs2_adjust_quota(struct gfs2_inode
*ip
, loff_t loc
,
606 s64 change
, struct gfs2_quota_data
*qd
)
608 struct inode
*inode
= &ip
->i_inode
;
609 struct address_space
*mapping
= inode
->i_mapping
;
610 unsigned long index
= loc
>> PAGE_CACHE_SHIFT
;
611 unsigned offset
= loc
& (PAGE_CACHE_SIZE
- 1);
612 unsigned blocksize
, iblock
, pos
;
613 struct buffer_head
*bh
;
617 struct gfs2_quota_host qp
;
621 if (gfs2_is_stuffed(ip
))
622 gfs2_unstuff_dinode(ip
, NULL
);
624 page
= grab_cache_page(mapping
, index
);
628 blocksize
= inode
->i_sb
->s_blocksize
;
629 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
631 if (!page_has_buffers(page
))
632 create_empty_buffers(page
, blocksize
, 0);
634 bh
= page_buffers(page
);
636 while (offset
>= pos
) {
637 bh
= bh
->b_this_page
;
642 if (!buffer_mapped(bh
)) {
643 gfs2_block_map(inode
, iblock
, bh
, 1);
644 if (!buffer_mapped(bh
))
648 if (PageUptodate(page
))
649 set_buffer_uptodate(bh
);
651 if (!buffer_uptodate(bh
)) {
652 ll_rw_block(READ_META
, 1, &bh
);
654 if (!buffer_uptodate(bh
))
658 gfs2_trans_add_bh(ip
->i_gl
, bh
, 0);
660 kaddr
= kmap_atomic(page
, KM_USER0
);
661 ptr
= kaddr
+ offset
;
662 gfs2_quota_in(&qp
, ptr
);
663 qp
.qu_value
+= change
;
665 gfs2_quota_out(&qp
, ptr
);
666 flush_dcache_page(page
);
667 kunmap_atomic(kaddr
, KM_USER0
);
669 qd
->qd_qb
.qb_magic
= cpu_to_be32(GFS2_MAGIC
);
670 qd
->qd_qb
.qb_value
= cpu_to_be64(value
);
671 ((struct gfs2_quota_lvb
*)(qd
->qd_gl
->gl_lvb
))->qb_magic
= cpu_to_be32(GFS2_MAGIC
);
672 ((struct gfs2_quota_lvb
*)(qd
->qd_gl
->gl_lvb
))->qb_value
= cpu_to_be64(value
);
675 page_cache_release(page
);
679 static int do_sync(unsigned int num_qd
, struct gfs2_quota_data
**qda
)
681 struct gfs2_sbd
*sdp
= (*qda
)->qd_gl
->gl_sbd
;
682 struct gfs2_inode
*ip
= GFS2_I(sdp
->sd_quota_inode
);
683 unsigned int data_blocks
, ind_blocks
;
684 struct gfs2_holder
*ghs
, i_gh
;
686 struct gfs2_quota_data
*qd
;
688 unsigned int nalloc
= 0, blocks
;
689 struct gfs2_alloc
*al
= NULL
;
692 gfs2_write_calc_reserv(ip
, sizeof(struct gfs2_quota
),
693 &data_blocks
, &ind_blocks
);
695 ghs
= kcalloc(num_qd
, sizeof(struct gfs2_holder
), GFP_NOFS
);
699 sort(qda
, num_qd
, sizeof(struct gfs2_quota_data
*), sort_qd
, NULL
);
700 for (qx
= 0; qx
< num_qd
; qx
++) {
701 error
= gfs2_glock_nq_init(qda
[qx
]->qd_gl
,
703 GL_NOCACHE
, &ghs
[qx
]);
708 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, 0, &i_gh
);
712 for (x
= 0; x
< num_qd
; x
++) {
715 offset
= qd2offset(qda
[x
]);
716 error
= gfs2_write_alloc_required(ip
, offset
,
717 sizeof(struct gfs2_quota
),
725 al
= gfs2_alloc_get(ip
);
731 * 1 blk for unstuffing inode if stuffed. We add this extra
732 * block to the reservation unconditionally. If the inode
733 * doesn't need unstuffing, the block will be released to the
734 * rgrp since it won't be allocated during the transaction
736 al
->al_requested
= 1;
737 /* +1 in the end for block requested above for unstuffing */
738 blocks
= num_qd
* data_blocks
+ RES_DINODE
+ num_qd
+ 1;
741 al
->al_requested
+= nalloc
* (data_blocks
+ ind_blocks
);
742 error
= gfs2_inplace_reserve(ip
);
747 blocks
+= al
->al_rgd
->rd_length
+ nalloc
* ind_blocks
+ RES_STATFS
;
749 error
= gfs2_trans_begin(sdp
, blocks
, 0);
753 for (x
= 0; x
< num_qd
; x
++) {
755 offset
= qd2offset(qd
);
756 error
= gfs2_adjust_quota(ip
, offset
, qd
->qd_change_sync
,
757 (struct gfs2_quota_data
*)
762 do_qc(qd
, -qd
->qd_change_sync
);
770 gfs2_inplace_release(ip
);
774 gfs2_glock_dq_uninit(&i_gh
);
777 gfs2_glock_dq_uninit(&ghs
[qx
]);
779 gfs2_log_flush(ip
->i_gl
->gl_sbd
, ip
->i_gl
);
783 static int do_glock(struct gfs2_quota_data
*qd
, int force_refresh
,
784 struct gfs2_holder
*q_gh
)
786 struct gfs2_sbd
*sdp
= qd
->qd_gl
->gl_sbd
;
787 struct gfs2_inode
*ip
= GFS2_I(sdp
->sd_quota_inode
);
788 struct gfs2_holder i_gh
;
789 struct gfs2_quota_host q
;
790 char buf
[sizeof(struct gfs2_quota
)];
792 struct gfs2_quota_lvb
*qlvb
;
795 error
= gfs2_glock_nq_init(qd
->qd_gl
, LM_ST_SHARED
, 0, q_gh
);
799 qd
->qd_qb
= *(struct gfs2_quota_lvb
*)qd
->qd_gl
->gl_lvb
;
801 if (force_refresh
|| qd
->qd_qb
.qb_magic
!= cpu_to_be32(GFS2_MAGIC
)) {
803 gfs2_glock_dq_uninit(q_gh
);
804 error
= gfs2_glock_nq_init(qd
->qd_gl
,
805 LM_ST_EXCLUSIVE
, GL_NOCACHE
,
810 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
, 0, &i_gh
);
814 memset(buf
, 0, sizeof(struct gfs2_quota
));
816 error
= gfs2_internal_read(ip
, NULL
, buf
, &pos
,
817 sizeof(struct gfs2_quota
));
821 gfs2_glock_dq_uninit(&i_gh
);
824 gfs2_quota_in(&q
, buf
);
825 qlvb
= (struct gfs2_quota_lvb
*)qd
->qd_gl
->gl_lvb
;
826 qlvb
->qb_magic
= cpu_to_be32(GFS2_MAGIC
);
828 qlvb
->qb_limit
= cpu_to_be64(q
.qu_limit
);
829 qlvb
->qb_warn
= cpu_to_be64(q
.qu_warn
);
830 qlvb
->qb_value
= cpu_to_be64(q
.qu_value
);
833 if (gfs2_glock_is_blocking(qd
->qd_gl
)) {
834 gfs2_glock_dq_uninit(q_gh
);
843 gfs2_glock_dq_uninit(&i_gh
);
845 gfs2_glock_dq_uninit(q_gh
);
849 int gfs2_quota_lock(struct gfs2_inode
*ip
, u32 uid
, u32 gid
)
851 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
852 struct gfs2_alloc
*al
= ip
->i_alloc
;
856 gfs2_quota_hold(ip
, uid
, gid
);
858 if (capable(CAP_SYS_RESOURCE
) ||
859 sdp
->sd_args
.ar_quota
!= GFS2_QUOTA_ON
)
862 sort(al
->al_qd
, al
->al_qd_num
, sizeof(struct gfs2_quota_data
*),
865 for (x
= 0; x
< al
->al_qd_num
; x
++) {
866 error
= do_glock(al
->al_qd
[x
], NO_FORCE
, &al
->al_qd_ghs
[x
]);
872 set_bit(GIF_QD_LOCKED
, &ip
->i_flags
);
875 gfs2_glock_dq_uninit(&al
->al_qd_ghs
[x
]);
876 gfs2_quota_unhold(ip
);
882 static int need_sync(struct gfs2_quota_data
*qd
)
884 struct gfs2_sbd
*sdp
= qd
->qd_gl
->gl_sbd
;
885 struct gfs2_tune
*gt
= &sdp
->sd_tune
;
887 unsigned int num
, den
;
890 if (!qd
->qd_qb
.qb_limit
)
893 spin_lock(&sdp
->sd_quota_spin
);
894 value
= qd
->qd_change
;
895 spin_unlock(&sdp
->sd_quota_spin
);
897 spin_lock(>
->gt_spin
);
898 num
= gt
->gt_quota_scale_num
;
899 den
= gt
->gt_quota_scale_den
;
900 spin_unlock(>
->gt_spin
);
904 else if ((s64
)be64_to_cpu(qd
->qd_qb
.qb_value
) >=
905 (s64
)be64_to_cpu(qd
->qd_qb
.qb_limit
))
908 value
*= gfs2_jindex_size(sdp
) * num
;
909 value
= div_s64(value
, den
);
910 value
+= (s64
)be64_to_cpu(qd
->qd_qb
.qb_value
);
911 if (value
< (s64
)be64_to_cpu(qd
->qd_qb
.qb_limit
))
918 void gfs2_quota_unlock(struct gfs2_inode
*ip
)
920 struct gfs2_alloc
*al
= ip
->i_alloc
;
921 struct gfs2_quota_data
*qda
[4];
922 unsigned int count
= 0;
925 if (!test_and_clear_bit(GIF_QD_LOCKED
, &ip
->i_flags
))
928 for (x
= 0; x
< al
->al_qd_num
; x
++) {
929 struct gfs2_quota_data
*qd
;
933 sync
= need_sync(qd
);
935 gfs2_glock_dq_uninit(&al
->al_qd_ghs
[x
]);
937 if (sync
&& qd_trylock(qd
))
943 for (x
= 0; x
< count
; x
++)
948 gfs2_quota_unhold(ip
);
953 static int print_message(struct gfs2_quota_data
*qd
, char *type
)
955 struct gfs2_sbd
*sdp
= qd
->qd_gl
->gl_sbd
;
957 printk(KERN_INFO
"GFS2: fsid=%s: quota %s for %s %u\r\n",
958 sdp
->sd_fsname
, type
,
959 (test_bit(QDF_USER
, &qd
->qd_flags
)) ? "user" : "group",
965 int gfs2_quota_check(struct gfs2_inode
*ip
, u32 uid
, u32 gid
)
967 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
968 struct gfs2_alloc
*al
= ip
->i_alloc
;
969 struct gfs2_quota_data
*qd
;
974 if (!test_bit(GIF_QD_LOCKED
, &ip
->i_flags
))
977 if (sdp
->sd_args
.ar_quota
!= GFS2_QUOTA_ON
)
980 for (x
= 0; x
< al
->al_qd_num
; x
++) {
983 if (!((qd
->qd_id
== uid
&& test_bit(QDF_USER
, &qd
->qd_flags
)) ||
984 (qd
->qd_id
== gid
&& !test_bit(QDF_USER
, &qd
->qd_flags
))))
987 value
= (s64
)be64_to_cpu(qd
->qd_qb
.qb_value
);
988 spin_lock(&sdp
->sd_quota_spin
);
989 value
+= qd
->qd_change
;
990 spin_unlock(&sdp
->sd_quota_spin
);
992 if (be64_to_cpu(qd
->qd_qb
.qb_limit
) && (s64
)be64_to_cpu(qd
->qd_qb
.qb_limit
) < value
) {
993 print_message(qd
, "exceeded");
996 } else if (be64_to_cpu(qd
->qd_qb
.qb_warn
) &&
997 (s64
)be64_to_cpu(qd
->qd_qb
.qb_warn
) < value
&&
998 time_after_eq(jiffies
, qd
->qd_last_warn
+
1000 gt_quota_warn_period
) * HZ
)) {
1001 error
= print_message(qd
, "warning");
1002 qd
->qd_last_warn
= jiffies
;
1009 void gfs2_quota_change(struct gfs2_inode
*ip
, s64 change
,
1012 struct gfs2_alloc
*al
= ip
->i_alloc
;
1013 struct gfs2_quota_data
*qd
;
1016 if (gfs2_assert_warn(GFS2_SB(&ip
->i_inode
), change
))
1018 if (ip
->i_diskflags
& GFS2_DIF_SYSTEM
)
1021 for (x
= 0; x
< al
->al_qd_num
; x
++) {
1024 if ((qd
->qd_id
== uid
&& test_bit(QDF_USER
, &qd
->qd_flags
)) ||
1025 (qd
->qd_id
== gid
&& !test_bit(QDF_USER
, &qd
->qd_flags
))) {
1031 int gfs2_quota_sync(struct gfs2_sbd
*sdp
)
1033 struct gfs2_quota_data
**qda
;
1034 unsigned int max_qd
= gfs2_tune_get(sdp
, gt_quota_simul_sync
);
1035 unsigned int num_qd
;
1039 sdp
->sd_quota_sync_gen
++;
1041 qda
= kcalloc(max_qd
, sizeof(struct gfs2_quota_data
*), GFP_KERNEL
);
1049 error
= qd_fish(sdp
, qda
+ num_qd
);
1050 if (error
|| !qda
[num_qd
])
1052 if (++num_qd
== max_qd
)
1058 error
= do_sync(num_qd
, qda
);
1060 for (x
= 0; x
< num_qd
; x
++)
1061 qda
[x
]->qd_sync_gen
=
1062 sdp
->sd_quota_sync_gen
;
1064 for (x
= 0; x
< num_qd
; x
++)
1067 } while (!error
&& num_qd
== max_qd
);
1074 int gfs2_quota_refresh(struct gfs2_sbd
*sdp
, int user
, u32 id
)
1076 struct gfs2_quota_data
*qd
;
1077 struct gfs2_holder q_gh
;
1080 error
= qd_get(sdp
, user
, id
, CREATE
, &qd
);
1084 error
= do_glock(qd
, FORCE
, &q_gh
);
1086 gfs2_glock_dq_uninit(&q_gh
);
1093 static void gfs2_quota_change_in(struct gfs2_quota_change_host
*qc
, const void *buf
)
1095 const struct gfs2_quota_change
*str
= buf
;
1097 qc
->qc_change
= be64_to_cpu(str
->qc_change
);
1098 qc
->qc_flags
= be32_to_cpu(str
->qc_flags
);
1099 qc
->qc_id
= be32_to_cpu(str
->qc_id
);
1102 int gfs2_quota_init(struct gfs2_sbd
*sdp
)
1104 struct gfs2_inode
*ip
= GFS2_I(sdp
->sd_qc_inode
);
1105 unsigned int blocks
= ip
->i_disksize
>> sdp
->sd_sb
.sb_bsize_shift
;
1106 unsigned int x
, slot
= 0;
1107 unsigned int found
= 0;
1112 if (!ip
->i_disksize
|| ip
->i_disksize
> (64 << 20) ||
1113 ip
->i_disksize
& (sdp
->sd_sb
.sb_bsize
- 1)) {
1114 gfs2_consist_inode(ip
);
1117 sdp
->sd_quota_slots
= blocks
* sdp
->sd_qc_per_block
;
1118 sdp
->sd_quota_chunks
= DIV_ROUND_UP(sdp
->sd_quota_slots
, 8 * PAGE_SIZE
);
1122 sdp
->sd_quota_bitmap
= kcalloc(sdp
->sd_quota_chunks
,
1123 sizeof(unsigned char *), GFP_NOFS
);
1124 if (!sdp
->sd_quota_bitmap
)
1127 for (x
= 0; x
< sdp
->sd_quota_chunks
; x
++) {
1128 sdp
->sd_quota_bitmap
[x
] = kzalloc(PAGE_SIZE
, GFP_NOFS
);
1129 if (!sdp
->sd_quota_bitmap
[x
])
1133 for (x
= 0; x
< blocks
; x
++) {
1134 struct buffer_head
*bh
;
1139 error
= gfs2_extent_map(&ip
->i_inode
, x
, &new, &dblock
, &extlen
);
1144 bh
= gfs2_meta_ra(ip
->i_gl
, dblock
, extlen
);
1147 if (gfs2_metatype_check(sdp
, bh
, GFS2_METATYPE_QC
)) {
1152 for (y
= 0; y
< sdp
->sd_qc_per_block
&& slot
< sdp
->sd_quota_slots
;
1154 struct gfs2_quota_change_host qc
;
1155 struct gfs2_quota_data
*qd
;
1157 gfs2_quota_change_in(&qc
, bh
->b_data
+
1158 sizeof(struct gfs2_meta_header
) +
1159 y
* sizeof(struct gfs2_quota_change
));
1163 error
= qd_alloc(sdp
, (qc
.qc_flags
& GFS2_QCF_USER
),
1170 set_bit(QDF_CHANGE
, &qd
->qd_flags
);
1171 qd
->qd_change
= qc
.qc_change
;
1173 qd
->qd_slot_count
= 1;
1174 qd
->qd_last_touched
= jiffies
;
1176 spin_lock(&sdp
->sd_quota_spin
);
1177 gfs2_icbit_munge(sdp
, sdp
->sd_quota_bitmap
, slot
, 1);
1178 list_add(&qd
->qd_list
, &sdp
->sd_quota_list
);
1179 atomic_inc(&sdp
->sd_quota_count
);
1180 spin_unlock(&sdp
->sd_quota_spin
);
1191 fs_info(sdp
, "found %u quota changes\n", found
);
1196 gfs2_quota_cleanup(sdp
);
1200 static void gfs2_quota_scan(struct gfs2_sbd
*sdp
)
1202 struct gfs2_quota_data
*qd
, *safe
;
1205 spin_lock(&sdp
->sd_quota_spin
);
1206 list_for_each_entry_safe(qd
, safe
, &sdp
->sd_quota_list
, qd_list
) {
1207 if (!qd
->qd_count
&&
1208 time_after_eq(jiffies
, qd
->qd_last_touched
+
1209 gfs2_tune_get(sdp
, gt_quota_cache_secs
) * HZ
)) {
1210 list_move(&qd
->qd_list
, &dead
);
1211 gfs2_assert_warn(sdp
,
1212 atomic_read(&sdp
->sd_quota_count
) > 0);
1213 atomic_dec(&sdp
->sd_quota_count
);
1216 spin_unlock(&sdp
->sd_quota_spin
);
1218 while (!list_empty(&dead
)) {
1219 qd
= list_entry(dead
.next
, struct gfs2_quota_data
, qd_list
);
1220 list_del(&qd
->qd_list
);
1222 gfs2_assert_warn(sdp
, !qd
->qd_change
);
1223 gfs2_assert_warn(sdp
, !qd
->qd_slot_count
);
1224 gfs2_assert_warn(sdp
, !qd
->qd_bh_count
);
1226 gfs2_lvb_unhold(qd
->qd_gl
);
1227 kmem_cache_free(gfs2_quotad_cachep
, qd
);
1231 void gfs2_quota_cleanup(struct gfs2_sbd
*sdp
)
1233 struct list_head
*head
= &sdp
->sd_quota_list
;
1234 struct gfs2_quota_data
*qd
;
1237 spin_lock(&sdp
->sd_quota_spin
);
1238 while (!list_empty(head
)) {
1239 qd
= list_entry(head
->prev
, struct gfs2_quota_data
, qd_list
);
1241 if (qd
->qd_count
> 1 ||
1242 (qd
->qd_count
&& !test_bit(QDF_CHANGE
, &qd
->qd_flags
))) {
1243 list_move(&qd
->qd_list
, head
);
1244 spin_unlock(&sdp
->sd_quota_spin
);
1246 spin_lock(&sdp
->sd_quota_spin
);
1250 list_del(&qd
->qd_list
);
1251 atomic_dec(&sdp
->sd_quota_count
);
1252 spin_unlock(&sdp
->sd_quota_spin
);
1254 if (!qd
->qd_count
) {
1255 gfs2_assert_warn(sdp
, !qd
->qd_change
);
1256 gfs2_assert_warn(sdp
, !qd
->qd_slot_count
);
1258 gfs2_assert_warn(sdp
, qd
->qd_slot_count
== 1);
1259 gfs2_assert_warn(sdp
, !qd
->qd_bh_count
);
1261 gfs2_lvb_unhold(qd
->qd_gl
);
1262 kmem_cache_free(gfs2_quotad_cachep
, qd
);
1264 spin_lock(&sdp
->sd_quota_spin
);
1266 spin_unlock(&sdp
->sd_quota_spin
);
1268 gfs2_assert_warn(sdp
, !atomic_read(&sdp
->sd_quota_count
));
1270 if (sdp
->sd_quota_bitmap
) {
1271 for (x
= 0; x
< sdp
->sd_quota_chunks
; x
++)
1272 kfree(sdp
->sd_quota_bitmap
[x
]);
1273 kfree(sdp
->sd_quota_bitmap
);
1277 static void quotad_error(struct gfs2_sbd
*sdp
, const char *msg
, int error
)
1279 if (error
== 0 || error
== -EROFS
)
1281 if (!test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
))
1282 fs_err(sdp
, "gfs2_quotad: %s error %d\n", msg
, error
);
1285 static void quotad_check_timeo(struct gfs2_sbd
*sdp
, const char *msg
,
1286 int (*fxn
)(struct gfs2_sbd
*sdp
),
1287 unsigned long t
, unsigned long *timeo
,
1288 unsigned int *new_timeo
)
1291 int error
= fxn(sdp
);
1292 quotad_error(sdp
, msg
, error
);
1293 *timeo
= gfs2_tune_get_i(&sdp
->sd_tune
, new_timeo
) * HZ
;
1299 static void quotad_check_trunc_list(struct gfs2_sbd
*sdp
)
1301 struct gfs2_inode
*ip
;
1305 spin_lock(&sdp
->sd_trunc_lock
);
1306 if (!list_empty(&sdp
->sd_trunc_list
)) {
1307 ip
= list_entry(sdp
->sd_trunc_list
.next
,
1308 struct gfs2_inode
, i_trunc_list
);
1309 list_del_init(&ip
->i_trunc_list
);
1311 spin_unlock(&sdp
->sd_trunc_lock
);
1314 gfs2_glock_finish_truncate(ip
);
1319 * gfs2_quotad - Write cached quota changes into the quota file
1320 * @sdp: Pointer to GFS2 superblock
1324 int gfs2_quotad(void *data
)
1326 struct gfs2_sbd
*sdp
= data
;
1327 struct gfs2_tune
*tune
= &sdp
->sd_tune
;
1328 unsigned long statfs_timeo
= 0;
1329 unsigned long quotad_timeo
= 0;
1330 unsigned long t
= 0;
1334 while (!kthread_should_stop()) {
1336 /* Update the master statfs file */
1337 quotad_check_timeo(sdp
, "statfs", gfs2_statfs_sync
, t
,
1338 &statfs_timeo
, &tune
->gt_statfs_quantum
);
1340 /* Update quota file */
1341 quotad_check_timeo(sdp
, "sync", gfs2_quota_sync
, t
,
1342 "ad_timeo
, &tune
->gt_quota_quantum
);
1344 /* FIXME: This should be turned into a shrinker */
1345 gfs2_quota_scan(sdp
);
1347 /* Check for & recover partially truncated inodes */
1348 quotad_check_trunc_list(sdp
);
1350 if (freezing(current
))
1352 t
= min(quotad_timeo
, statfs_timeo
);
1354 prepare_to_wait(&sdp
->sd_quota_wait
, &wait
, TASK_UNINTERRUPTIBLE
);
1355 spin_lock(&sdp
->sd_trunc_lock
);
1356 empty
= list_empty(&sdp
->sd_trunc_list
);
1357 spin_unlock(&sdp
->sd_trunc_lock
);
1359 t
-= schedule_timeout(t
);
1362 finish_wait(&sdp
->sd_quota_wait
, &wait
);