2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2005 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 v.2.
10 #include <linux/sched.h>
11 #include <linux/slab.h>
12 #include <linux/spinlock.h>
13 #include <linux/completion.h>
14 #include <linux/buffer_head.h>
16 #include <linux/gfs2_ondisk.h>
17 #include <asm/semaphore.h>
20 #include "lm_interface.h"
33 #define BFITNOENT 0xFFFFFFFF
36 * These routines are used by the resource group routines (rgrp.c)
37 * to keep track of block allocation. Each block is represented by two
38 * bits. One bit indicates whether or not the block is used. (1=used,
39 * 0=free) The other bit indicates whether or not the block contains a
40 * dinode or not. (1=dinode, 0=not-dinode) So, each byte represents
41 * GFS2_NBBY (i.e. 4) blocks.
44 static const char valid_change
[16] = {
53 * gfs2_setbit - Set a bit in the bitmaps
54 * @buffer: the buffer that holds the bitmaps
55 * @buflen: the length (in bytes) of the buffer
56 * @block: the block to set
57 * @new_state: the new state of the block
61 static void gfs2_setbit(struct gfs2_rgrpd
*rgd
, unsigned char *buffer
,
62 unsigned int buflen
, uint32_t block
,
63 unsigned char new_state
)
65 unsigned char *byte
, *end
, cur_state
;
68 byte
= buffer
+ (block
/ GFS2_NBBY
);
69 bit
= (block
% GFS2_NBBY
) * GFS2_BIT_SIZE
;
70 end
= buffer
+ buflen
;
72 gfs2_assert(rgd
->rd_sbd
, byte
< end
);
74 cur_state
= (*byte
>> bit
) & GFS2_BIT_MASK
;
76 if (valid_change
[new_state
* 4 + cur_state
]) {
77 *byte
^= cur_state
<< bit
;
78 *byte
|= new_state
<< bit
;
80 gfs2_consist_rgrpd(rgd
);
84 * gfs2_testbit - test a bit in the bitmaps
85 * @buffer: the buffer that holds the bitmaps
86 * @buflen: the length (in bytes) of the buffer
87 * @block: the block to read
91 static unsigned char gfs2_testbit(struct gfs2_rgrpd
*rgd
, unsigned char *buffer
,
92 unsigned int buflen
, uint32_t block
)
94 unsigned char *byte
, *end
, cur_state
;
97 byte
= buffer
+ (block
/ GFS2_NBBY
);
98 bit
= (block
% GFS2_NBBY
) * GFS2_BIT_SIZE
;
99 end
= buffer
+ buflen
;
101 gfs2_assert(rgd
->rd_sbd
, byte
< end
);
103 cur_state
= (*byte
>> bit
) & GFS2_BIT_MASK
;
109 * gfs2_bitfit - Search an rgrp's bitmap buffer to find a bit-pair representing
110 * a block in a given allocation state.
111 * @buffer: the buffer that holds the bitmaps
112 * @buflen: the length (in bytes) of the buffer
113 * @goal: start search at this block's bit-pair (within @buffer)
114 * @old_state: GFS2_BLKST_XXX the state of the block we're looking for;
115 * bit 0 = alloc(1)/free(0), bit 1 = meta(1)/data(0)
117 * Scope of @goal and returned block number is only within this bitmap buffer,
118 * not entire rgrp or filesystem. @buffer will be offset from the actual
119 * beginning of a bitmap block buffer, skipping any header structures.
121 * Return: the block number (bitmap buffer scope) that was found
124 static uint32_t gfs2_bitfit(struct gfs2_rgrpd
*rgd
, unsigned char *buffer
,
125 unsigned int buflen
, uint32_t goal
,
126 unsigned char old_state
)
128 unsigned char *byte
, *end
, alloc
;
132 byte
= buffer
+ (goal
/ GFS2_NBBY
);
133 bit
= (goal
% GFS2_NBBY
) * GFS2_BIT_SIZE
;
134 end
= buffer
+ buflen
;
135 alloc
= (old_state
& 1) ? 0 : 0x55;
138 if ((*byte
& 0x55) == alloc
) {
139 blk
+= (8 - bit
) >> 1;
147 if (((*byte
>> bit
) & GFS2_BIT_MASK
) == old_state
)
150 bit
+= GFS2_BIT_SIZE
;
163 * gfs2_bitcount - count the number of bits in a certain state
164 * @buffer: the buffer that holds the bitmaps
165 * @buflen: the length (in bytes) of the buffer
166 * @state: the state of the block we're looking for
168 * Returns: The number of bits
171 static uint32_t gfs2_bitcount(struct gfs2_rgrpd
*rgd
, unsigned char *buffer
,
172 unsigned int buflen
, unsigned char state
)
174 unsigned char *byte
= buffer
;
175 unsigned char *end
= buffer
+ buflen
;
176 unsigned char state1
= state
<< 2;
177 unsigned char state2
= state
<< 4;
178 unsigned char state3
= state
<< 6;
181 for (; byte
< end
; byte
++) {
182 if (((*byte
) & 0x03) == state
)
184 if (((*byte
) & 0x0C) == state1
)
186 if (((*byte
) & 0x30) == state2
)
188 if (((*byte
) & 0xC0) == state3
)
196 * gfs2_rgrp_verify - Verify that a resource group is consistent
197 * @sdp: the filesystem
202 void gfs2_rgrp_verify(struct gfs2_rgrpd
*rgd
)
204 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
205 struct gfs2_bitmap
*bi
= NULL
;
206 uint32_t length
= rgd
->rd_ri
.ri_length
;
207 uint32_t count
[4], tmp
;
210 memset(count
, 0, 4 * sizeof(uint32_t));
212 /* Count # blocks in each of 4 possible allocation states */
213 for (buf
= 0; buf
< length
; buf
++) {
214 bi
= rgd
->rd_bits
+ buf
;
215 for (x
= 0; x
< 4; x
++)
216 count
[x
] += gfs2_bitcount(rgd
,
222 if (count
[0] != rgd
->rd_rg
.rg_free
) {
223 if (gfs2_consist_rgrpd(rgd
))
224 fs_err(sdp
, "free data mismatch: %u != %u\n",
225 count
[0], rgd
->rd_rg
.rg_free
);
229 tmp
= rgd
->rd_ri
.ri_data
-
231 rgd
->rd_rg
.rg_dinodes
;
232 if (count
[1] != tmp
) {
233 if (gfs2_consist_rgrpd(rgd
))
234 fs_err(sdp
, "used data mismatch: %u != %u\n",
240 if (gfs2_consist_rgrpd(rgd
))
241 fs_err(sdp
, "free metadata mismatch: %u != 0\n",
246 if (count
[3] != rgd
->rd_rg
.rg_dinodes
) {
247 if (gfs2_consist_rgrpd(rgd
))
248 fs_err(sdp
, "used metadata mismatch: %u != %u\n",
249 count
[3], rgd
->rd_rg
.rg_dinodes
);
254 static inline int rgrp_contains_block(struct gfs2_rindex
*ri
, uint64_t block
)
256 uint64_t first
= ri
->ri_data0
;
257 uint64_t last
= first
+ ri
->ri_data
;
258 return !!(first
<= block
&& block
< last
);
262 * gfs2_blk2rgrpd - Find resource group for a given data/meta block number
263 * @sdp: The GFS2 superblock
264 * @n: The data block number
266 * Returns: The resource group, or NULL if not found
269 struct gfs2_rgrpd
*gfs2_blk2rgrpd(struct gfs2_sbd
*sdp
, uint64_t blk
)
271 struct gfs2_rgrpd
*rgd
;
273 spin_lock(&sdp
->sd_rindex_spin
);
275 list_for_each_entry(rgd
, &sdp
->sd_rindex_mru_list
, rd_list_mru
) {
276 if (rgrp_contains_block(&rgd
->rd_ri
, blk
)) {
277 list_move(&rgd
->rd_list_mru
, &sdp
->sd_rindex_mru_list
);
278 spin_unlock(&sdp
->sd_rindex_spin
);
283 spin_unlock(&sdp
->sd_rindex_spin
);
289 * gfs2_rgrpd_get_first - get the first Resource Group in the filesystem
290 * @sdp: The GFS2 superblock
292 * Returns: The first rgrp in the filesystem
295 struct gfs2_rgrpd
*gfs2_rgrpd_get_first(struct gfs2_sbd
*sdp
)
297 gfs2_assert(sdp
, !list_empty(&sdp
->sd_rindex_list
));
298 return list_entry(sdp
->sd_rindex_list
.next
, struct gfs2_rgrpd
, rd_list
);
302 * gfs2_rgrpd_get_next - get the next RG
305 * Returns: The next rgrp
308 struct gfs2_rgrpd
*gfs2_rgrpd_get_next(struct gfs2_rgrpd
*rgd
)
310 if (rgd
->rd_list
.next
== &rgd
->rd_sbd
->sd_rindex_list
)
312 return list_entry(rgd
->rd_list
.next
, struct gfs2_rgrpd
, rd_list
);
315 static void clear_rgrpdi(struct gfs2_sbd
*sdp
)
317 struct list_head
*head
;
318 struct gfs2_rgrpd
*rgd
;
319 struct gfs2_glock
*gl
;
321 spin_lock(&sdp
->sd_rindex_spin
);
322 sdp
->sd_rindex_forward
= NULL
;
323 head
= &sdp
->sd_rindex_recent_list
;
324 while (!list_empty(head
)) {
325 rgd
= list_entry(head
->next
, struct gfs2_rgrpd
, rd_recent
);
326 list_del(&rgd
->rd_recent
);
328 spin_unlock(&sdp
->sd_rindex_spin
);
330 head
= &sdp
->sd_rindex_list
;
331 while (!list_empty(head
)) {
332 rgd
= list_entry(head
->next
, struct gfs2_rgrpd
, rd_list
);
335 list_del(&rgd
->rd_list
);
336 list_del(&rgd
->rd_list_mru
);
339 gl
->gl_object
= NULL
;
348 void gfs2_clear_rgrpd(struct gfs2_sbd
*sdp
)
350 mutex_lock(&sdp
->sd_rindex_mutex
);
352 mutex_unlock(&sdp
->sd_rindex_mutex
);
356 * gfs2_compute_bitstructs - Compute the bitmap sizes
357 * @rgd: The resource group descriptor
359 * Calculates bitmap descriptors, one for each block that contains bitmap data
364 static int compute_bitstructs(struct gfs2_rgrpd
*rgd
)
366 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
367 struct gfs2_bitmap
*bi
;
368 uint32_t length
= rgd
->rd_ri
.ri_length
; /* # blocks in hdr & bitmap */
369 uint32_t bytes_left
, bytes
;
372 rgd
->rd_bits
= kcalloc(length
, sizeof(struct gfs2_bitmap
), GFP_KERNEL
);
376 bytes_left
= rgd
->rd_ri
.ri_bitbytes
;
378 for (x
= 0; x
< length
; x
++) {
379 bi
= rgd
->rd_bits
+ x
;
381 /* small rgrp; bitmap stored completely in header block */
384 bi
->bi_offset
= sizeof(struct gfs2_rgrp
);
389 bytes
= sdp
->sd_sb
.sb_bsize
- sizeof(struct gfs2_rgrp
);
390 bi
->bi_offset
= sizeof(struct gfs2_rgrp
);
394 } else if (x
+ 1 == length
) {
396 bi
->bi_offset
= sizeof(struct gfs2_meta_header
);
397 bi
->bi_start
= rgd
->rd_ri
.ri_bitbytes
- bytes_left
;
401 bytes
= sdp
->sd_sb
.sb_bsize
-
402 sizeof(struct gfs2_meta_header
);
403 bi
->bi_offset
= sizeof(struct gfs2_meta_header
);
404 bi
->bi_start
= rgd
->rd_ri
.ri_bitbytes
- bytes_left
;
412 gfs2_consist_rgrpd(rgd
);
415 bi
= rgd
->rd_bits
+ (length
- 1);
416 if ((bi
->bi_start
+ bi
->bi_len
) * GFS2_NBBY
!= rgd
->rd_ri
.ri_data
) {
417 if (gfs2_consist_rgrpd(rgd
)) {
418 gfs2_rindex_print(&rgd
->rd_ri
);
419 fs_err(sdp
, "start=%u len=%u offset=%u\n",
420 bi
->bi_start
, bi
->bi_len
, bi
->bi_offset
);
429 * gfs2_ri_update - Pull in a new resource index from the disk
430 * @gl: The glock covering the rindex inode
432 * Returns: 0 on successful update, error code otherwise
435 static int gfs2_ri_update(struct gfs2_inode
*ip
)
437 struct gfs2_sbd
*sdp
= ip
->i_sbd
;
438 struct inode
*inode
= ip
->i_vnode
;
439 struct gfs2_rgrpd
*rgd
;
440 char buf
[sizeof(struct gfs2_rindex
)];
441 struct file_ra_state ra_state
;
442 uint64_t junk
= ip
->i_di
.di_size
;
445 if (do_div(junk
, sizeof(struct gfs2_rindex
))) {
446 gfs2_consist_inode(ip
);
452 file_ra_state_init(&ra_state
, inode
->i_mapping
);
453 for (sdp
->sd_rgrps
= 0;; sdp
->sd_rgrps
++) {
454 loff_t pos
= sdp
->sd_rgrps
* sizeof(struct gfs2_rindex
);
455 error
= gfs2_internal_read(ip
, &ra_state
, buf
, &pos
,
456 sizeof(struct gfs2_rindex
));
459 if (error
!= sizeof(struct gfs2_rindex
)) {
465 rgd
= kzalloc(sizeof(struct gfs2_rgrpd
), GFP_KERNEL
);
470 mutex_init(&rgd
->rd_mutex
);
471 lops_init_le(&rgd
->rd_le
, &gfs2_rg_lops
);
474 list_add_tail(&rgd
->rd_list
, &sdp
->sd_rindex_list
);
475 list_add_tail(&rgd
->rd_list_mru
, &sdp
->sd_rindex_mru_list
);
477 gfs2_rindex_in(&rgd
->rd_ri
, buf
);
479 error
= compute_bitstructs(rgd
);
483 error
= gfs2_glock_get(sdp
, rgd
->rd_ri
.ri_addr
,
484 &gfs2_rgrp_glops
, CREATE
, &rgd
->rd_gl
);
488 rgd
->rd_gl
->gl_object
= rgd
;
489 rgd
->rd_rg_vn
= rgd
->rd_gl
->gl_vn
- 1;
492 sdp
->sd_rindex_vn
= ip
->i_gl
->gl_vn
;
503 * gfs2_rindex_hold - Grab a lock on the rindex
504 * @sdp: The GFS2 superblock
505 * @ri_gh: the glock holder
507 * We grab a lock on the rindex inode to make sure that it doesn't
508 * change whilst we are performing an operation. We keep this lock
509 * for quite long periods of time compared to other locks. This
510 * doesn't matter, since it is shared and it is very, very rarely
511 * accessed in the exclusive mode (i.e. only when expanding the filesystem).
513 * This makes sure that we're using the latest copy of the resource index
514 * special file, which might have been updated if someone expanded the
515 * filesystem (via gfs2_grow utility), which adds new resource groups.
517 * Returns: 0 on success, error code otherwise
520 int gfs2_rindex_hold(struct gfs2_sbd
*sdp
, struct gfs2_holder
*ri_gh
)
522 struct gfs2_inode
*ip
= sdp
->sd_rindex
->u
.generic_ip
;
523 struct gfs2_glock
*gl
= ip
->i_gl
;
526 error
= gfs2_glock_nq_init(gl
, LM_ST_SHARED
, 0, ri_gh
);
530 /* Read new copy from disk if we don't have the latest */
531 if (sdp
->sd_rindex_vn
!= gl
->gl_vn
) {
532 mutex_lock(&sdp
->sd_rindex_mutex
);
533 if (sdp
->sd_rindex_vn
!= gl
->gl_vn
) {
534 error
= gfs2_ri_update(ip
);
536 gfs2_glock_dq_uninit(ri_gh
);
538 mutex_unlock(&sdp
->sd_rindex_mutex
);
545 * gfs2_rgrp_bh_get - Read in a RG's header and bitmaps
546 * @rgd: the struct gfs2_rgrpd describing the RG to read in
548 * Read in all of a Resource Group's header and bitmap blocks.
549 * Caller must eventually call gfs2_rgrp_relse() to free the bitmaps.
554 int gfs2_rgrp_bh_get(struct gfs2_rgrpd
*rgd
)
556 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
557 struct gfs2_glock
*gl
= rgd
->rd_gl
;
558 unsigned int length
= rgd
->rd_ri
.ri_length
;
559 struct gfs2_bitmap
*bi
;
563 mutex_lock(&rgd
->rd_mutex
);
565 spin_lock(&sdp
->sd_rindex_spin
);
566 if (rgd
->rd_bh_count
) {
568 spin_unlock(&sdp
->sd_rindex_spin
);
569 mutex_unlock(&rgd
->rd_mutex
);
572 spin_unlock(&sdp
->sd_rindex_spin
);
574 for (x
= 0; x
< length
; x
++) {
575 bi
= rgd
->rd_bits
+ x
;
576 error
= gfs2_meta_read(gl
, rgd
->rd_ri
.ri_addr
+ x
, DIO_START
,
582 for (y
= length
; y
--;) {
583 bi
= rgd
->rd_bits
+ y
;
584 error
= gfs2_meta_reread(sdp
, bi
->bi_bh
, DIO_WAIT
);
587 if (gfs2_metatype_check(sdp
, bi
->bi_bh
,
588 (y
) ? GFS2_METATYPE_RB
:
595 if (rgd
->rd_rg_vn
!= gl
->gl_vn
) {
596 gfs2_rgrp_in(&rgd
->rd_rg
, (rgd
->rd_bits
[0].bi_bh
)->b_data
);
597 rgd
->rd_rg_vn
= gl
->gl_vn
;
600 spin_lock(&sdp
->sd_rindex_spin
);
601 rgd
->rd_free_clone
= rgd
->rd_rg
.rg_free
;
603 spin_unlock(&sdp
->sd_rindex_spin
);
605 mutex_unlock(&rgd
->rd_mutex
);
611 bi
= rgd
->rd_bits
+ x
;
614 gfs2_assert_warn(sdp
, !bi
->bi_clone
);
616 mutex_unlock(&rgd
->rd_mutex
);
621 void gfs2_rgrp_bh_hold(struct gfs2_rgrpd
*rgd
)
623 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
625 spin_lock(&sdp
->sd_rindex_spin
);
626 gfs2_assert_warn(rgd
->rd_sbd
, rgd
->rd_bh_count
);
628 spin_unlock(&sdp
->sd_rindex_spin
);
632 * gfs2_rgrp_bh_put - Release RG bitmaps read in with gfs2_rgrp_bh_get()
633 * @rgd: the struct gfs2_rgrpd describing the RG to read in
637 void gfs2_rgrp_bh_put(struct gfs2_rgrpd
*rgd
)
639 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
640 int x
, length
= rgd
->rd_ri
.ri_length
;
642 spin_lock(&sdp
->sd_rindex_spin
);
643 gfs2_assert_warn(rgd
->rd_sbd
, rgd
->rd_bh_count
);
644 if (--rgd
->rd_bh_count
) {
645 spin_unlock(&sdp
->sd_rindex_spin
);
649 for (x
= 0; x
< length
; x
++) {
650 struct gfs2_bitmap
*bi
= rgd
->rd_bits
+ x
;
657 spin_unlock(&sdp
->sd_rindex_spin
);
660 void gfs2_rgrp_repolish_clones(struct gfs2_rgrpd
*rgd
)
662 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
663 unsigned int length
= rgd
->rd_ri
.ri_length
;
666 for (x
= 0; x
< length
; x
++) {
667 struct gfs2_bitmap
*bi
= rgd
->rd_bits
+ x
;
670 memcpy(bi
->bi_clone
+ bi
->bi_offset
,
671 bi
->bi_bh
->b_data
+ bi
->bi_offset
,
675 spin_lock(&sdp
->sd_rindex_spin
);
676 rgd
->rd_free_clone
= rgd
->rd_rg
.rg_free
;
677 spin_unlock(&sdp
->sd_rindex_spin
);
681 * gfs2_alloc_get - get the struct gfs2_alloc structure for an inode
682 * @ip: the incore GFS2 inode structure
684 * Returns: the struct gfs2_alloc
687 struct gfs2_alloc
*gfs2_alloc_get(struct gfs2_inode
*ip
)
689 struct gfs2_alloc
*al
= &ip
->i_alloc
;
691 /* FIXME: Should assert that the correct locks are held here... */
692 memset(al
, 0, sizeof(*al
));
697 * gfs2_alloc_put - throw away the struct gfs2_alloc for an inode
702 void gfs2_alloc_put(struct gfs2_inode
*ip
)
708 * try_rgrp_fit - See if a given reservation will fit in a given RG
710 * @al: the struct gfs2_alloc structure describing the reservation
712 * If there's room for the requested blocks to be allocated from the RG:
713 * Sets the $al_reserved_data field in @al.
714 * Sets the $al_reserved_meta field in @al.
715 * Sets the $al_rgd field in @al.
717 * Returns: 1 on success (it fits), 0 on failure (it doesn't fit)
720 static int try_rgrp_fit(struct gfs2_rgrpd
*rgd
, struct gfs2_alloc
*al
)
722 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
725 spin_lock(&sdp
->sd_rindex_spin
);
726 if (rgd
->rd_free_clone
>= al
->al_requested
) {
730 spin_unlock(&sdp
->sd_rindex_spin
);
736 * recent_rgrp_first - get first RG from "recent" list
737 * @sdp: The GFS2 superblock
738 * @rglast: address of the rgrp used last
740 * Returns: The first rgrp in the recent list
743 static struct gfs2_rgrpd
*recent_rgrp_first(struct gfs2_sbd
*sdp
,
746 struct gfs2_rgrpd
*rgd
= NULL
;
748 spin_lock(&sdp
->sd_rindex_spin
);
750 if (list_empty(&sdp
->sd_rindex_recent_list
))
756 list_for_each_entry(rgd
, &sdp
->sd_rindex_recent_list
, rd_recent
) {
757 if (rgd
->rd_ri
.ri_addr
== rglast
)
762 rgd
= list_entry(sdp
->sd_rindex_recent_list
.next
, struct gfs2_rgrpd
,
766 spin_unlock(&sdp
->sd_rindex_spin
);
772 * recent_rgrp_next - get next RG from "recent" list
773 * @cur_rgd: current rgrp
776 * Returns: The next rgrp in the recent list
779 static struct gfs2_rgrpd
*recent_rgrp_next(struct gfs2_rgrpd
*cur_rgd
,
782 struct gfs2_sbd
*sdp
= cur_rgd
->rd_sbd
;
783 struct list_head
*head
;
784 struct gfs2_rgrpd
*rgd
;
786 spin_lock(&sdp
->sd_rindex_spin
);
788 head
= &sdp
->sd_rindex_recent_list
;
790 list_for_each_entry(rgd
, head
, rd_recent
) {
791 if (rgd
== cur_rgd
) {
792 if (cur_rgd
->rd_recent
.next
!= head
)
793 rgd
= list_entry(cur_rgd
->rd_recent
.next
,
794 struct gfs2_rgrpd
, rd_recent
);
799 list_del(&cur_rgd
->rd_recent
);
806 if (!list_empty(head
))
807 rgd
= list_entry(head
->next
, struct gfs2_rgrpd
, rd_recent
);
810 spin_unlock(&sdp
->sd_rindex_spin
);
816 * recent_rgrp_add - add an RG to tail of "recent" list
817 * @new_rgd: The rgrp to add
821 static void recent_rgrp_add(struct gfs2_rgrpd
*new_rgd
)
823 struct gfs2_sbd
*sdp
= new_rgd
->rd_sbd
;
824 struct gfs2_rgrpd
*rgd
;
825 unsigned int count
= 0;
826 unsigned int max
= sdp
->sd_rgrps
/ gfs2_jindex_size(sdp
);
828 spin_lock(&sdp
->sd_rindex_spin
);
830 list_for_each_entry(rgd
, &sdp
->sd_rindex_recent_list
, rd_recent
) {
837 list_add_tail(&new_rgd
->rd_recent
, &sdp
->sd_rindex_recent_list
);
840 spin_unlock(&sdp
->sd_rindex_spin
);
844 * forward_rgrp_get - get an rgrp to try next from full list
845 * @sdp: The GFS2 superblock
847 * Returns: The rgrp to try next
850 static struct gfs2_rgrpd
*forward_rgrp_get(struct gfs2_sbd
*sdp
)
852 struct gfs2_rgrpd
*rgd
;
853 unsigned int journals
= gfs2_jindex_size(sdp
);
854 unsigned int rg
= 0, x
;
856 spin_lock(&sdp
->sd_rindex_spin
);
858 rgd
= sdp
->sd_rindex_forward
;
860 if (sdp
->sd_rgrps
>= journals
)
861 rg
= sdp
->sd_rgrps
* sdp
->sd_jdesc
->jd_jid
/ journals
;
863 for (x
= 0, rgd
= gfs2_rgrpd_get_first(sdp
);
865 x
++, rgd
= gfs2_rgrpd_get_next(rgd
))
868 sdp
->sd_rindex_forward
= rgd
;
871 spin_unlock(&sdp
->sd_rindex_spin
);
877 * forward_rgrp_set - set the forward rgrp pointer
878 * @sdp: the filesystem
879 * @rgd: The new forward rgrp
883 static void forward_rgrp_set(struct gfs2_sbd
*sdp
, struct gfs2_rgrpd
*rgd
)
885 spin_lock(&sdp
->sd_rindex_spin
);
886 sdp
->sd_rindex_forward
= rgd
;
887 spin_unlock(&sdp
->sd_rindex_spin
);
891 * get_local_rgrp - Choose and lock a rgrp for allocation
892 * @ip: the inode to reserve space for
893 * @rgp: the chosen and locked rgrp
895 * Try to acquire rgrp in way which avoids contending with others.
900 static int get_local_rgrp(struct gfs2_inode
*ip
)
902 struct gfs2_sbd
*sdp
= ip
->i_sbd
;
903 struct gfs2_rgrpd
*rgd
, *begin
= NULL
;
904 struct gfs2_alloc
*al
= &ip
->i_alloc
;
905 int flags
= LM_FLAG_TRY
;
910 /* Try recently successful rgrps */
912 rgd
= recent_rgrp_first(sdp
, ip
->i_last_rg_alloc
);
915 error
= gfs2_glock_nq_init(rgd
->rd_gl
,
916 LM_ST_EXCLUSIVE
, LM_FLAG_TRY
,
920 if (try_rgrp_fit(rgd
, al
))
922 gfs2_glock_dq_uninit(&al
->al_rgd_gh
);
923 rgd
= recent_rgrp_next(rgd
, 1);
927 rgd
= recent_rgrp_next(rgd
, 0);
935 /* Go through full list of rgrps */
937 begin
= rgd
= forward_rgrp_get(sdp
);
940 error
= gfs2_glock_nq_init(rgd
->rd_gl
,
941 LM_ST_EXCLUSIVE
, flags
,
945 if (try_rgrp_fit(rgd
, al
))
947 gfs2_glock_dq_uninit(&al
->al_rgd_gh
);
958 rgd
= gfs2_rgrpd_get_next(rgd
);
960 rgd
= gfs2_rgrpd_get_first(sdp
);
963 if (++loops
>= 2 || !skipped
)
970 ip
->i_last_rg_alloc
= rgd
->rd_ri
.ri_addr
;
973 recent_rgrp_add(rgd
);
974 rgd
= gfs2_rgrpd_get_next(rgd
);
976 rgd
= gfs2_rgrpd_get_first(sdp
);
977 forward_rgrp_set(sdp
, rgd
);
984 * gfs2_inplace_reserve_i - Reserve space in the filesystem
985 * @ip: the inode to reserve space for
990 int gfs2_inplace_reserve_i(struct gfs2_inode
*ip
, char *file
, unsigned int line
)
992 struct gfs2_sbd
*sdp
= ip
->i_sbd
;
993 struct gfs2_alloc
*al
= &ip
->i_alloc
;
996 if (gfs2_assert_warn(sdp
, al
->al_requested
))
999 error
= gfs2_rindex_hold(sdp
, &al
->al_ri_gh
);
1003 error
= get_local_rgrp(ip
);
1005 gfs2_glock_dq_uninit(&al
->al_ri_gh
);
1016 * gfs2_inplace_release - release an inplace reservation
1017 * @ip: the inode the reservation was taken out on
1019 * Release a reservation made by gfs2_inplace_reserve().
1022 void gfs2_inplace_release(struct gfs2_inode
*ip
)
1024 struct gfs2_sbd
*sdp
= ip
->i_sbd
;
1025 struct gfs2_alloc
*al
= &ip
->i_alloc
;
1027 if (gfs2_assert_warn(sdp
, al
->al_alloced
<= al
->al_requested
) == -1)
1028 fs_warn(sdp
, "al_alloced = %u, al_requested = %u "
1029 "al_file = %s, al_line = %u\n",
1030 al
->al_alloced
, al
->al_requested
, al
->al_file
,
1034 gfs2_glock_dq_uninit(&al
->al_rgd_gh
);
1035 gfs2_glock_dq_uninit(&al
->al_ri_gh
);
1039 * gfs2_get_block_type - Check a block in a RG is of given type
1040 * @rgd: the resource group holding the block
1041 * @block: the block number
1043 * Returns: The block type (GFS2_BLKST_*)
1046 unsigned char gfs2_get_block_type(struct gfs2_rgrpd
*rgd
, uint64_t block
)
1048 struct gfs2_bitmap
*bi
= NULL
;
1049 uint32_t length
, rgrp_block
, buf_block
;
1053 length
= rgd
->rd_ri
.ri_length
;
1054 rgrp_block
= block
- rgd
->rd_ri
.ri_data0
;
1056 for (buf
= 0; buf
< length
; buf
++) {
1057 bi
= rgd
->rd_bits
+ buf
;
1058 if (rgrp_block
< (bi
->bi_start
+ bi
->bi_len
) * GFS2_NBBY
)
1062 gfs2_assert(rgd
->rd_sbd
, buf
< length
);
1063 buf_block
= rgrp_block
- bi
->bi_start
* GFS2_NBBY
;
1065 type
= gfs2_testbit(rgd
,
1066 bi
->bi_bh
->b_data
+ bi
->bi_offset
,
1067 bi
->bi_len
, buf_block
);
1073 * rgblk_search - find a block in @old_state, change allocation
1074 * state to @new_state
1075 * @rgd: the resource group descriptor
1076 * @goal: the goal block within the RG (start here to search for avail block)
1077 * @old_state: GFS2_BLKST_XXX the before-allocation state to find
1078 * @new_state: GFS2_BLKST_XXX the after-allocation block state
1080 * Walk rgrp's bitmap to find bits that represent a block in @old_state.
1081 * Add the found bitmap buffer to the transaction.
1082 * Set the found bits to @new_state to change block's allocation state.
1084 * This function never fails, because we wouldn't call it unless we
1085 * know (from reservation results, etc.) that a block is available.
1087 * Scope of @goal and returned block is just within rgrp, not the whole
1090 * Returns: the block number allocated
1093 static uint32_t rgblk_search(struct gfs2_rgrpd
*rgd
, uint32_t goal
,
1094 unsigned char old_state
, unsigned char new_state
)
1096 struct gfs2_bitmap
*bi
= NULL
;
1097 uint32_t length
= rgd
->rd_ri
.ri_length
;
1099 unsigned int buf
, x
;
1101 /* Find bitmap block that contains bits for goal block */
1102 for (buf
= 0; buf
< length
; buf
++) {
1103 bi
= rgd
->rd_bits
+ buf
;
1104 if (goal
< (bi
->bi_start
+ bi
->bi_len
) * GFS2_NBBY
)
1108 gfs2_assert(rgd
->rd_sbd
, buf
< length
);
1110 /* Convert scope of "goal" from rgrp-wide to within found bit block */
1111 goal
-= bi
->bi_start
* GFS2_NBBY
;
1113 /* Search (up to entire) bitmap in this rgrp for allocatable block.
1114 "x <= length", instead of "x < length", because we typically start
1115 the search in the middle of a bit block, but if we can't find an
1116 allocatable block anywhere else, we want to be able wrap around and
1117 search in the first part of our first-searched bit block. */
1118 for (x
= 0; x
<= length
; x
++) {
1120 blk
= gfs2_bitfit(rgd
, bi
->bi_clone
+ bi
->bi_offset
,
1121 bi
->bi_len
, goal
, old_state
);
1123 blk
= gfs2_bitfit(rgd
,
1124 bi
->bi_bh
->b_data
+ bi
->bi_offset
,
1125 bi
->bi_len
, goal
, old_state
);
1126 if (blk
!= BFITNOENT
)
1129 /* Try next bitmap block (wrap back to rgrp header if at end) */
1130 buf
= (buf
+ 1) % length
;
1131 bi
= rgd
->rd_bits
+ buf
;
1135 if (gfs2_assert_withdraw(rgd
->rd_sbd
, x
<= length
))
1138 gfs2_trans_add_bh(rgd
->rd_gl
, bi
->bi_bh
, 1);
1139 gfs2_setbit(rgd
, bi
->bi_bh
->b_data
+ bi
->bi_offset
,
1140 bi
->bi_len
, blk
, new_state
);
1142 gfs2_setbit(rgd
, bi
->bi_clone
+ bi
->bi_offset
,
1143 bi
->bi_len
, blk
, new_state
);
1145 return bi
->bi_start
* GFS2_NBBY
+ blk
;
1149 * rgblk_free - Change alloc state of given block(s)
1150 * @sdp: the filesystem
1151 * @bstart: the start of a run of blocks to free
1152 * @blen: the length of the block run (all must lie within ONE RG!)
1153 * @new_state: GFS2_BLKST_XXX the after-allocation block state
1155 * Returns: Resource group containing the block(s)
1158 static struct gfs2_rgrpd
*rgblk_free(struct gfs2_sbd
*sdp
, uint64_t bstart
,
1159 uint32_t blen
, unsigned char new_state
)
1161 struct gfs2_rgrpd
*rgd
;
1162 struct gfs2_bitmap
*bi
= NULL
;
1163 uint32_t length
, rgrp_blk
, buf_blk
;
1166 rgd
= gfs2_blk2rgrpd(sdp
, bstart
);
1168 if (gfs2_consist(sdp
))
1169 fs_err(sdp
, "block = %llu\n", bstart
);
1173 length
= rgd
->rd_ri
.ri_length
;
1175 rgrp_blk
= bstart
- rgd
->rd_ri
.ri_data0
;
1178 for (buf
= 0; buf
< length
; buf
++) {
1179 bi
= rgd
->rd_bits
+ buf
;
1180 if (rgrp_blk
< (bi
->bi_start
+ bi
->bi_len
) * GFS2_NBBY
)
1184 gfs2_assert(rgd
->rd_sbd
, buf
< length
);
1186 buf_blk
= rgrp_blk
- bi
->bi_start
* GFS2_NBBY
;
1189 if (!bi
->bi_clone
) {
1190 bi
->bi_clone
= kmalloc(bi
->bi_bh
->b_size
,
1191 GFP_KERNEL
| __GFP_NOFAIL
);
1192 memcpy(bi
->bi_clone
+ bi
->bi_offset
,
1193 bi
->bi_bh
->b_data
+ bi
->bi_offset
,
1196 gfs2_trans_add_bh(rgd
->rd_gl
, bi
->bi_bh
, 1);
1198 bi
->bi_bh
->b_data
+ bi
->bi_offset
,
1199 bi
->bi_len
, buf_blk
, new_state
);
1206 * gfs2_alloc_data - Allocate a data block
1207 * @ip: the inode to allocate the data block for
1209 * Returns: the allocated block
1212 uint64_t gfs2_alloc_data(struct gfs2_inode
*ip
)
1214 struct gfs2_sbd
*sdp
= ip
->i_sbd
;
1215 struct gfs2_alloc
*al
= &ip
->i_alloc
;
1216 struct gfs2_rgrpd
*rgd
= al
->al_rgd
;
1220 if (rgrp_contains_block(&rgd
->rd_ri
, ip
->i_di
.di_goal_data
))
1221 goal
= ip
->i_di
.di_goal_data
- rgd
->rd_ri
.ri_data0
;
1223 goal
= rgd
->rd_last_alloc_data
;
1225 blk
= rgblk_search(rgd
, goal
, GFS2_BLKST_FREE
, GFS2_BLKST_USED
);
1226 rgd
->rd_last_alloc_data
= blk
;
1228 block
= rgd
->rd_ri
.ri_data0
+ blk
;
1229 ip
->i_di
.di_goal_data
= block
;
1231 gfs2_assert_withdraw(sdp
, rgd
->rd_rg
.rg_free
);
1232 rgd
->rd_rg
.rg_free
--;
1234 gfs2_trans_add_bh(rgd
->rd_gl
, rgd
->rd_bits
[0].bi_bh
, 1);
1235 gfs2_rgrp_out(&rgd
->rd_rg
, rgd
->rd_bits
[0].bi_bh
->b_data
);
1239 gfs2_statfs_change(sdp
, 0, -1, 0);
1240 gfs2_quota_change(ip
, +1, ip
->i_di
.di_uid
, ip
->i_di
.di_gid
);
1242 spin_lock(&sdp
->sd_rindex_spin
);
1243 rgd
->rd_free_clone
--;
1244 spin_unlock(&sdp
->sd_rindex_spin
);
1250 * gfs2_alloc_meta - Allocate a metadata block
1251 * @ip: the inode to allocate the metadata block for
1253 * Returns: the allocated block
1256 uint64_t gfs2_alloc_meta(struct gfs2_inode
*ip
)
1258 struct gfs2_sbd
*sdp
= ip
->i_sbd
;
1259 struct gfs2_alloc
*al
= &ip
->i_alloc
;
1260 struct gfs2_rgrpd
*rgd
= al
->al_rgd
;
1264 if (rgrp_contains_block(&rgd
->rd_ri
, ip
->i_di
.di_goal_meta
))
1265 goal
= ip
->i_di
.di_goal_meta
- rgd
->rd_ri
.ri_data0
;
1267 goal
= rgd
->rd_last_alloc_meta
;
1269 blk
= rgblk_search(rgd
, goal
, GFS2_BLKST_FREE
, GFS2_BLKST_USED
);
1270 rgd
->rd_last_alloc_meta
= blk
;
1272 block
= rgd
->rd_ri
.ri_data0
+ blk
;
1273 ip
->i_di
.di_goal_meta
= block
;
1275 gfs2_assert_withdraw(sdp
, rgd
->rd_rg
.rg_free
);
1276 rgd
->rd_rg
.rg_free
--;
1278 gfs2_trans_add_bh(rgd
->rd_gl
, rgd
->rd_bits
[0].bi_bh
, 1);
1279 gfs2_rgrp_out(&rgd
->rd_rg
, rgd
->rd_bits
[0].bi_bh
->b_data
);
1283 gfs2_statfs_change(sdp
, 0, -1, 0);
1284 gfs2_quota_change(ip
, +1, ip
->i_di
.di_uid
, ip
->i_di
.di_gid
);
1285 gfs2_trans_add_unrevoke(sdp
, block
);
1287 spin_lock(&sdp
->sd_rindex_spin
);
1288 rgd
->rd_free_clone
--;
1289 spin_unlock(&sdp
->sd_rindex_spin
);
1295 * gfs2_alloc_di - Allocate a dinode
1296 * @dip: the directory that the inode is going in
1298 * Returns: the block allocated
1301 uint64_t gfs2_alloc_di(struct gfs2_inode
*dip
)
1303 struct gfs2_sbd
*sdp
= dip
->i_sbd
;
1304 struct gfs2_alloc
*al
= &dip
->i_alloc
;
1305 struct gfs2_rgrpd
*rgd
= al
->al_rgd
;
1309 blk
= rgblk_search(rgd
, rgd
->rd_last_alloc_meta
,
1310 GFS2_BLKST_FREE
, GFS2_BLKST_DINODE
);
1312 rgd
->rd_last_alloc_meta
= blk
;
1314 block
= rgd
->rd_ri
.ri_data0
+ blk
;
1316 gfs2_assert_withdraw(sdp
, rgd
->rd_rg
.rg_free
);
1317 rgd
->rd_rg
.rg_free
--;
1318 rgd
->rd_rg
.rg_dinodes
++;
1320 gfs2_trans_add_bh(rgd
->rd_gl
, rgd
->rd_bits
[0].bi_bh
, 1);
1321 gfs2_rgrp_out(&rgd
->rd_rg
, rgd
->rd_bits
[0].bi_bh
->b_data
);
1325 gfs2_statfs_change(sdp
, 0, -1, +1);
1326 gfs2_trans_add_unrevoke(sdp
, block
);
1328 spin_lock(&sdp
->sd_rindex_spin
);
1329 rgd
->rd_free_clone
--;
1330 spin_unlock(&sdp
->sd_rindex_spin
);
1336 * gfs2_free_data - free a contiguous run of data block(s)
1337 * @ip: the inode these blocks are being freed from
1338 * @bstart: first block of a run of contiguous blocks
1339 * @blen: the length of the block run
1343 void gfs2_free_data(struct gfs2_inode
*ip
, uint64_t bstart
, uint32_t blen
)
1345 struct gfs2_sbd
*sdp
= ip
->i_sbd
;
1346 struct gfs2_rgrpd
*rgd
;
1348 rgd
= rgblk_free(sdp
, bstart
, blen
, GFS2_BLKST_FREE
);
1352 rgd
->rd_rg
.rg_free
+= blen
;
1354 gfs2_trans_add_bh(rgd
->rd_gl
, rgd
->rd_bits
[0].bi_bh
, 1);
1355 gfs2_rgrp_out(&rgd
->rd_rg
, rgd
->rd_bits
[0].bi_bh
->b_data
);
1357 gfs2_trans_add_rg(rgd
);
1359 gfs2_statfs_change(sdp
, 0, +blen
, 0);
1360 gfs2_quota_change(ip
, -(int64_t)blen
,
1361 ip
->i_di
.di_uid
, ip
->i_di
.di_gid
);
1365 * gfs2_free_meta - free a contiguous run of data block(s)
1366 * @ip: the inode these blocks are being freed from
1367 * @bstart: first block of a run of contiguous blocks
1368 * @blen: the length of the block run
1372 void gfs2_free_meta(struct gfs2_inode
*ip
, uint64_t bstart
, uint32_t blen
)
1374 struct gfs2_sbd
*sdp
= ip
->i_sbd
;
1375 struct gfs2_rgrpd
*rgd
;
1377 rgd
= rgblk_free(sdp
, bstart
, blen
, GFS2_BLKST_FREE
);
1381 rgd
->rd_rg
.rg_free
+= blen
;
1383 gfs2_trans_add_bh(rgd
->rd_gl
, rgd
->rd_bits
[0].bi_bh
, 1);
1384 gfs2_rgrp_out(&rgd
->rd_rg
, rgd
->rd_bits
[0].bi_bh
->b_data
);
1386 gfs2_trans_add_rg(rgd
);
1388 gfs2_statfs_change(sdp
, 0, +blen
, 0);
1389 gfs2_quota_change(ip
, -(int64_t)blen
,
1390 ip
->i_di
.di_uid
, ip
->i_di
.di_gid
);
1391 gfs2_meta_wipe(ip
, bstart
, blen
);
1394 void gfs2_free_uninit_di(struct gfs2_rgrpd
*rgd
, uint64_t blkno
)
1396 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
1397 struct gfs2_rgrpd
*tmp_rgd
;
1399 tmp_rgd
= rgblk_free(sdp
, blkno
, 1, GFS2_BLKST_FREE
);
1402 gfs2_assert_withdraw(sdp
, rgd
== tmp_rgd
);
1404 if (!rgd
->rd_rg
.rg_dinodes
)
1405 gfs2_consist_rgrpd(rgd
);
1406 rgd
->rd_rg
.rg_dinodes
--;
1407 rgd
->rd_rg
.rg_free
++;
1409 gfs2_trans_add_bh(rgd
->rd_gl
, rgd
->rd_bits
[0].bi_bh
, 1);
1410 gfs2_rgrp_out(&rgd
->rd_rg
, rgd
->rd_bits
[0].bi_bh
->b_data
);
1412 gfs2_statfs_change(sdp
, 0, +1, -1);
1413 gfs2_trans_add_rg(rgd
);
1417 * gfs2_free_uninit_di - free a dinode block
1418 * @rgd: the resource group that contains the dinode
1423 void gfs2_free_di(struct gfs2_rgrpd
*rgd
, struct gfs2_inode
*ip
)
1425 gfs2_free_uninit_di(rgd
, ip
->i_num
.no_addr
);
1426 gfs2_quota_change(ip
, -1, ip
->i_di
.di_uid
, ip
->i_di
.di_gid
);
1427 gfs2_meta_wipe(ip
, ip
->i_num
.no_addr
, 1);
1431 * gfs2_rlist_add - add a RG to a list of RGs
1432 * @sdp: the filesystem
1433 * @rlist: the list of resource groups
1436 * Figure out what RG a block belongs to and add that RG to the list
1438 * FIXME: Don't use NOFAIL
1442 void gfs2_rlist_add(struct gfs2_sbd
*sdp
, struct gfs2_rgrp_list
*rlist
,
1445 struct gfs2_rgrpd
*rgd
;
1446 struct gfs2_rgrpd
**tmp
;
1447 unsigned int new_space
;
1450 if (gfs2_assert_warn(sdp
, !rlist
->rl_ghs
))
1453 rgd
= gfs2_blk2rgrpd(sdp
, block
);
1455 if (gfs2_consist(sdp
))
1456 fs_err(sdp
, "block = %llu\n", block
);
1460 for (x
= 0; x
< rlist
->rl_rgrps
; x
++)
1461 if (rlist
->rl_rgd
[x
] == rgd
)
1464 if (rlist
->rl_rgrps
== rlist
->rl_space
) {
1465 new_space
= rlist
->rl_space
+ 10;
1467 tmp
= kcalloc(new_space
, sizeof(struct gfs2_rgrpd
*),
1468 GFP_KERNEL
| __GFP_NOFAIL
);
1470 if (rlist
->rl_rgd
) {
1471 memcpy(tmp
, rlist
->rl_rgd
,
1472 rlist
->rl_space
* sizeof(struct gfs2_rgrpd
*));
1473 kfree(rlist
->rl_rgd
);
1476 rlist
->rl_space
= new_space
;
1477 rlist
->rl_rgd
= tmp
;
1480 rlist
->rl_rgd
[rlist
->rl_rgrps
++] = rgd
;
1484 * gfs2_rlist_alloc - all RGs have been added to the rlist, now allocate
1485 * and initialize an array of glock holders for them
1486 * @rlist: the list of resource groups
1487 * @state: the lock state to acquire the RG lock in
1488 * @flags: the modifier flags for the holder structures
1490 * FIXME: Don't use NOFAIL
1494 void gfs2_rlist_alloc(struct gfs2_rgrp_list
*rlist
, unsigned int state
,
1499 rlist
->rl_ghs
= kcalloc(rlist
->rl_rgrps
, sizeof(struct gfs2_holder
),
1500 GFP_KERNEL
| __GFP_NOFAIL
);
1501 for (x
= 0; x
< rlist
->rl_rgrps
; x
++)
1502 gfs2_holder_init(rlist
->rl_rgd
[x
]->rd_gl
,
1508 * gfs2_rlist_free - free a resource group list
1509 * @list: the list of resource groups
1513 void gfs2_rlist_free(struct gfs2_rgrp_list
*rlist
)
1517 kfree(rlist
->rl_rgd
);
1519 if (rlist
->rl_ghs
) {
1520 for (x
= 0; x
< rlist
->rl_rgrps
; x
++)
1521 gfs2_holder_uninit(&rlist
->rl_ghs
[x
]);
1522 kfree(rlist
->rl_ghs
);