2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2006 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.
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 <linux/lm_interface.h>
32 #define BFITNOENT ((u32)~0)
35 * These routines are used by the resource group routines (rgrp.c)
36 * to keep track of block allocation. Each block is represented by two
37 * bits. So, each byte represents GFS2_NBBY (i.e. 4) blocks.
40 * 1 = Used (not metadata)
41 * 2 = Unlinked (still in use) inode
45 static const char valid_change
[16] = {
54 * gfs2_setbit - Set a bit in the bitmaps
55 * @buffer: the buffer that holds the bitmaps
56 * @buflen: the length (in bytes) of the buffer
57 * @block: the block to set
58 * @new_state: the new state of the block
62 static void gfs2_setbit(struct gfs2_rgrpd
*rgd
, unsigned char *buffer
,
63 unsigned int buflen
, u32 block
,
64 unsigned char new_state
)
66 unsigned char *byte
, *end
, cur_state
;
69 byte
= buffer
+ (block
/ GFS2_NBBY
);
70 bit
= (block
% GFS2_NBBY
) * GFS2_BIT_SIZE
;
71 end
= buffer
+ buflen
;
73 gfs2_assert(rgd
->rd_sbd
, byte
< end
);
75 cur_state
= (*byte
>> bit
) & GFS2_BIT_MASK
;
77 if (valid_change
[new_state
* 4 + cur_state
]) {
78 *byte
^= cur_state
<< bit
;
79 *byte
|= new_state
<< bit
;
81 gfs2_consist_rgrpd(rgd
);
85 * gfs2_testbit - test a bit in the bitmaps
86 * @buffer: the buffer that holds the bitmaps
87 * @buflen: the length (in bytes) of the buffer
88 * @block: the block to read
92 static unsigned char gfs2_testbit(struct gfs2_rgrpd
*rgd
, unsigned char *buffer
,
93 unsigned int buflen
, u32 block
)
95 unsigned char *byte
, *end
, cur_state
;
98 byte
= buffer
+ (block
/ GFS2_NBBY
);
99 bit
= (block
% GFS2_NBBY
) * GFS2_BIT_SIZE
;
100 end
= buffer
+ buflen
;
102 gfs2_assert(rgd
->rd_sbd
, byte
< end
);
104 cur_state
= (*byte
>> bit
) & GFS2_BIT_MASK
;
110 * gfs2_bitfit - Search an rgrp's bitmap buffer to find a bit-pair representing
111 * a block in a given allocation state.
112 * @buffer: the buffer that holds the bitmaps
113 * @buflen: the length (in bytes) of the buffer
114 * @goal: start search at this block's bit-pair (within @buffer)
115 * @old_state: GFS2_BLKST_XXX the state of the block we're looking for;
116 * bit 0 = alloc(1)/free(0), bit 1 = meta(1)/data(0)
118 * Scope of @goal and returned block number is only within this bitmap buffer,
119 * not entire rgrp or filesystem. @buffer will be offset from the actual
120 * beginning of a bitmap block buffer, skipping any header structures.
122 * Return: the block number (bitmap buffer scope) that was found
125 static u32
gfs2_bitfit(struct gfs2_rgrpd
*rgd
, unsigned char *buffer
,
126 unsigned int buflen
, u32 goal
,
127 unsigned char old_state
)
129 unsigned char *byte
, *end
, alloc
;
133 byte
= buffer
+ (goal
/ GFS2_NBBY
);
134 bit
= (goal
% GFS2_NBBY
) * GFS2_BIT_SIZE
;
135 end
= buffer
+ buflen
;
136 alloc
= (old_state
& 1) ? 0 : 0x55;
139 if ((*byte
& 0x55) == alloc
) {
140 blk
+= (8 - bit
) >> 1;
148 if (((*byte
>> bit
) & GFS2_BIT_MASK
) == old_state
)
151 bit
+= GFS2_BIT_SIZE
;
164 * gfs2_bitcount - count the number of bits in a certain state
165 * @buffer: the buffer that holds the bitmaps
166 * @buflen: the length (in bytes) of the buffer
167 * @state: the state of the block we're looking for
169 * Returns: The number of bits
172 static u32
gfs2_bitcount(struct gfs2_rgrpd
*rgd
, unsigned char *buffer
,
173 unsigned int buflen
, unsigned char state
)
175 unsigned char *byte
= buffer
;
176 unsigned char *end
= buffer
+ buflen
;
177 unsigned char state1
= state
<< 2;
178 unsigned char state2
= state
<< 4;
179 unsigned char state3
= state
<< 6;
182 for (; byte
< end
; byte
++) {
183 if (((*byte
) & 0x03) == state
)
185 if (((*byte
) & 0x0C) == state1
)
187 if (((*byte
) & 0x30) == state2
)
189 if (((*byte
) & 0xC0) == state3
)
197 * gfs2_rgrp_verify - Verify that a resource group is consistent
198 * @sdp: the filesystem
203 void gfs2_rgrp_verify(struct gfs2_rgrpd
*rgd
)
205 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
206 struct gfs2_bitmap
*bi
= NULL
;
207 u32 length
= rgd
->rd_ri
.ri_length
;
211 memset(count
, 0, 4 * sizeof(u32
));
213 /* Count # blocks in each of 4 possible allocation states */
214 for (buf
= 0; buf
< length
; buf
++) {
215 bi
= rgd
->rd_bits
+ buf
;
216 for (x
= 0; x
< 4; x
++)
217 count
[x
] += gfs2_bitcount(rgd
,
223 if (count
[0] != rgd
->rd_rg
.rg_free
) {
224 if (gfs2_consist_rgrpd(rgd
))
225 fs_err(sdp
, "free data mismatch: %u != %u\n",
226 count
[0], rgd
->rd_rg
.rg_free
);
230 tmp
= rgd
->rd_ri
.ri_data
-
232 rgd
->rd_rg
.rg_dinodes
;
233 if (count
[1] + count
[2] != tmp
) {
234 if (gfs2_consist_rgrpd(rgd
))
235 fs_err(sdp
, "used data mismatch: %u != %u\n",
240 if (count
[3] != rgd
->rd_rg
.rg_dinodes
) {
241 if (gfs2_consist_rgrpd(rgd
))
242 fs_err(sdp
, "used metadata mismatch: %u != %u\n",
243 count
[3], rgd
->rd_rg
.rg_dinodes
);
247 if (count
[2] > count
[3]) {
248 if (gfs2_consist_rgrpd(rgd
))
249 fs_err(sdp
, "unlinked inodes > inodes: %u\n",
256 static inline int rgrp_contains_block(struct gfs2_rindex
*ri
, u64 block
)
258 u64 first
= ri
->ri_data0
;
259 u64 last
= first
+ ri
->ri_data
;
260 return first
<= block
&& block
< last
;
264 * gfs2_blk2rgrpd - Find resource group for a given data/meta block number
265 * @sdp: The GFS2 superblock
266 * @n: The data block number
268 * Returns: The resource group, or NULL if not found
271 struct gfs2_rgrpd
*gfs2_blk2rgrpd(struct gfs2_sbd
*sdp
, u64 blk
)
273 struct gfs2_rgrpd
*rgd
;
275 spin_lock(&sdp
->sd_rindex_spin
);
277 list_for_each_entry(rgd
, &sdp
->sd_rindex_mru_list
, rd_list_mru
) {
278 if (rgrp_contains_block(&rgd
->rd_ri
, blk
)) {
279 list_move(&rgd
->rd_list_mru
, &sdp
->sd_rindex_mru_list
);
280 spin_unlock(&sdp
->sd_rindex_spin
);
285 spin_unlock(&sdp
->sd_rindex_spin
);
291 * gfs2_rgrpd_get_first - get the first Resource Group in the filesystem
292 * @sdp: The GFS2 superblock
294 * Returns: The first rgrp in the filesystem
297 struct gfs2_rgrpd
*gfs2_rgrpd_get_first(struct gfs2_sbd
*sdp
)
299 gfs2_assert(sdp
, !list_empty(&sdp
->sd_rindex_list
));
300 return list_entry(sdp
->sd_rindex_list
.next
, struct gfs2_rgrpd
, rd_list
);
304 * gfs2_rgrpd_get_next - get the next RG
307 * Returns: The next rgrp
310 struct gfs2_rgrpd
*gfs2_rgrpd_get_next(struct gfs2_rgrpd
*rgd
)
312 if (rgd
->rd_list
.next
== &rgd
->rd_sbd
->sd_rindex_list
)
314 return list_entry(rgd
->rd_list
.next
, struct gfs2_rgrpd
, rd_list
);
317 static void clear_rgrpdi(struct gfs2_sbd
*sdp
)
319 struct list_head
*head
;
320 struct gfs2_rgrpd
*rgd
;
321 struct gfs2_glock
*gl
;
323 spin_lock(&sdp
->sd_rindex_spin
);
324 sdp
->sd_rindex_forward
= NULL
;
325 head
= &sdp
->sd_rindex_recent_list
;
326 while (!list_empty(head
)) {
327 rgd
= list_entry(head
->next
, struct gfs2_rgrpd
, rd_recent
);
328 list_del(&rgd
->rd_recent
);
330 spin_unlock(&sdp
->sd_rindex_spin
);
332 head
= &sdp
->sd_rindex_list
;
333 while (!list_empty(head
)) {
334 rgd
= list_entry(head
->next
, struct gfs2_rgrpd
, rd_list
);
337 list_del(&rgd
->rd_list
);
338 list_del(&rgd
->rd_list_mru
);
341 gl
->gl_object
= NULL
;
350 void gfs2_clear_rgrpd(struct gfs2_sbd
*sdp
)
352 mutex_lock(&sdp
->sd_rindex_mutex
);
354 mutex_unlock(&sdp
->sd_rindex_mutex
);
358 * gfs2_compute_bitstructs - Compute the bitmap sizes
359 * @rgd: The resource group descriptor
361 * Calculates bitmap descriptors, one for each block that contains bitmap data
366 static int compute_bitstructs(struct gfs2_rgrpd
*rgd
)
368 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
369 struct gfs2_bitmap
*bi
;
370 u32 length
= rgd
->rd_ri
.ri_length
; /* # blocks in hdr & bitmap */
371 u32 bytes_left
, bytes
;
377 rgd
->rd_bits
= kcalloc(length
, sizeof(struct gfs2_bitmap
), GFP_NOFS
);
381 bytes_left
= rgd
->rd_ri
.ri_bitbytes
;
383 for (x
= 0; x
< length
; x
++) {
384 bi
= rgd
->rd_bits
+ x
;
386 /* small rgrp; bitmap stored completely in header block */
389 bi
->bi_offset
= sizeof(struct gfs2_rgrp
);
394 bytes
= sdp
->sd_sb
.sb_bsize
- sizeof(struct gfs2_rgrp
);
395 bi
->bi_offset
= sizeof(struct gfs2_rgrp
);
399 } else if (x
+ 1 == length
) {
401 bi
->bi_offset
= sizeof(struct gfs2_meta_header
);
402 bi
->bi_start
= rgd
->rd_ri
.ri_bitbytes
- bytes_left
;
406 bytes
= sdp
->sd_sb
.sb_bsize
-
407 sizeof(struct gfs2_meta_header
);
408 bi
->bi_offset
= sizeof(struct gfs2_meta_header
);
409 bi
->bi_start
= rgd
->rd_ri
.ri_bitbytes
- bytes_left
;
417 gfs2_consist_rgrpd(rgd
);
420 bi
= rgd
->rd_bits
+ (length
- 1);
421 if ((bi
->bi_start
+ bi
->bi_len
) * GFS2_NBBY
!= rgd
->rd_ri
.ri_data
) {
422 if (gfs2_consist_rgrpd(rgd
)) {
423 gfs2_rindex_print(&rgd
->rd_ri
);
424 fs_err(sdp
, "start=%u len=%u offset=%u\n",
425 bi
->bi_start
, bi
->bi_len
, bi
->bi_offset
);
434 * gfs2_ri_update - Pull in a new resource index from the disk
435 * @gl: The glock covering the rindex inode
437 * Returns: 0 on successful update, error code otherwise
440 static int gfs2_ri_update(struct gfs2_inode
*ip
)
442 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
443 struct inode
*inode
= &ip
->i_inode
;
444 struct gfs2_rgrpd
*rgd
;
445 char buf
[sizeof(struct gfs2_rindex
)];
446 struct file_ra_state ra_state
;
447 u64 junk
= ip
->i_di
.di_size
;
450 if (do_div(junk
, sizeof(struct gfs2_rindex
))) {
451 gfs2_consist_inode(ip
);
457 file_ra_state_init(&ra_state
, inode
->i_mapping
);
458 for (sdp
->sd_rgrps
= 0;; sdp
->sd_rgrps
++) {
459 loff_t pos
= sdp
->sd_rgrps
* sizeof(struct gfs2_rindex
);
460 error
= gfs2_internal_read(ip
, &ra_state
, buf
, &pos
,
461 sizeof(struct gfs2_rindex
));
464 if (error
!= sizeof(struct gfs2_rindex
)) {
470 rgd
= kzalloc(sizeof(struct gfs2_rgrpd
), GFP_NOFS
);
475 mutex_init(&rgd
->rd_mutex
);
476 lops_init_le(&rgd
->rd_le
, &gfs2_rg_lops
);
479 list_add_tail(&rgd
->rd_list
, &sdp
->sd_rindex_list
);
480 list_add_tail(&rgd
->rd_list_mru
, &sdp
->sd_rindex_mru_list
);
482 gfs2_rindex_in(&rgd
->rd_ri
, buf
);
483 error
= compute_bitstructs(rgd
);
487 error
= gfs2_glock_get(sdp
, rgd
->rd_ri
.ri_addr
,
488 &gfs2_rgrp_glops
, CREATE
, &rgd
->rd_gl
);
492 rgd
->rd_gl
->gl_object
= rgd
;
493 rgd
->rd_rg_vn
= rgd
->rd_gl
->gl_vn
- 1;
496 sdp
->sd_rindex_vn
= ip
->i_gl
->gl_vn
;
505 * gfs2_rindex_hold - Grab a lock on the rindex
506 * @sdp: The GFS2 superblock
507 * @ri_gh: the glock holder
509 * We grab a lock on the rindex inode to make sure that it doesn't
510 * change whilst we are performing an operation. We keep this lock
511 * for quite long periods of time compared to other locks. This
512 * doesn't matter, since it is shared and it is very, very rarely
513 * accessed in the exclusive mode (i.e. only when expanding the filesystem).
515 * This makes sure that we're using the latest copy of the resource index
516 * special file, which might have been updated if someone expanded the
517 * filesystem (via gfs2_grow utility), which adds new resource groups.
519 * Returns: 0 on success, error code otherwise
522 int gfs2_rindex_hold(struct gfs2_sbd
*sdp
, struct gfs2_holder
*ri_gh
)
524 struct gfs2_inode
*ip
= GFS2_I(sdp
->sd_rindex
);
525 struct gfs2_glock
*gl
= ip
->i_gl
;
528 error
= gfs2_glock_nq_init(gl
, LM_ST_SHARED
, 0, ri_gh
);
532 /* Read new copy from disk if we don't have the latest */
533 if (sdp
->sd_rindex_vn
!= gl
->gl_vn
) {
534 mutex_lock(&sdp
->sd_rindex_mutex
);
535 if (sdp
->sd_rindex_vn
!= gl
->gl_vn
) {
536 error
= gfs2_ri_update(ip
);
538 gfs2_glock_dq_uninit(ri_gh
);
540 mutex_unlock(&sdp
->sd_rindex_mutex
);
547 * gfs2_rgrp_bh_get - Read in a RG's header and bitmaps
548 * @rgd: the struct gfs2_rgrpd describing the RG to read in
550 * Read in all of a Resource Group's header and bitmap blocks.
551 * Caller must eventually call gfs2_rgrp_relse() to free the bitmaps.
556 int gfs2_rgrp_bh_get(struct gfs2_rgrpd
*rgd
)
558 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
559 struct gfs2_glock
*gl
= rgd
->rd_gl
;
560 unsigned int length
= rgd
->rd_ri
.ri_length
;
561 struct gfs2_bitmap
*bi
;
565 mutex_lock(&rgd
->rd_mutex
);
567 spin_lock(&sdp
->sd_rindex_spin
);
568 if (rgd
->rd_bh_count
) {
570 spin_unlock(&sdp
->sd_rindex_spin
);
571 mutex_unlock(&rgd
->rd_mutex
);
574 spin_unlock(&sdp
->sd_rindex_spin
);
576 for (x
= 0; x
< length
; x
++) {
577 bi
= rgd
->rd_bits
+ x
;
578 error
= gfs2_meta_read(gl
, rgd
->rd_ri
.ri_addr
+ x
, 0, &bi
->bi_bh
);
583 for (y
= length
; y
--;) {
584 bi
= rgd
->rd_bits
+ y
;
585 error
= gfs2_meta_wait(sdp
, bi
->bi_bh
);
588 if (gfs2_metatype_check(sdp
, bi
->bi_bh
, 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
, bi
->bi_len
);
674 spin_lock(&sdp
->sd_rindex_spin
);
675 rgd
->rd_free_clone
= rgd
->rd_rg
.rg_free
;
676 spin_unlock(&sdp
->sd_rindex_spin
);
680 * gfs2_alloc_get - get the struct gfs2_alloc structure for an inode
681 * @ip: the incore GFS2 inode structure
683 * Returns: the struct gfs2_alloc
686 struct gfs2_alloc
*gfs2_alloc_get(struct gfs2_inode
*ip
)
688 struct gfs2_alloc
*al
= &ip
->i_alloc
;
690 /* FIXME: Should assert that the correct locks are held here... */
691 memset(al
, 0, sizeof(*al
));
696 * try_rgrp_fit - See if a given reservation will fit in a given RG
698 * @al: the struct gfs2_alloc structure describing the reservation
700 * If there's room for the requested blocks to be allocated from the RG:
701 * Sets the $al_reserved_data field in @al.
702 * Sets the $al_reserved_meta field in @al.
703 * Sets the $al_rgd field in @al.
705 * Returns: 1 on success (it fits), 0 on failure (it doesn't fit)
708 static int try_rgrp_fit(struct gfs2_rgrpd
*rgd
, struct gfs2_alloc
*al
)
710 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
713 spin_lock(&sdp
->sd_rindex_spin
);
714 if (rgd
->rd_free_clone
>= al
->al_requested
) {
718 spin_unlock(&sdp
->sd_rindex_spin
);
724 * recent_rgrp_first - get first RG from "recent" list
725 * @sdp: The GFS2 superblock
726 * @rglast: address of the rgrp used last
728 * Returns: The first rgrp in the recent list
731 static struct gfs2_rgrpd
*recent_rgrp_first(struct gfs2_sbd
*sdp
,
734 struct gfs2_rgrpd
*rgd
= NULL
;
736 spin_lock(&sdp
->sd_rindex_spin
);
738 if (list_empty(&sdp
->sd_rindex_recent_list
))
744 list_for_each_entry(rgd
, &sdp
->sd_rindex_recent_list
, rd_recent
) {
745 if (rgd
->rd_ri
.ri_addr
== rglast
)
750 rgd
= list_entry(sdp
->sd_rindex_recent_list
.next
, struct gfs2_rgrpd
,
753 spin_unlock(&sdp
->sd_rindex_spin
);
758 * recent_rgrp_next - get next RG from "recent" list
759 * @cur_rgd: current rgrp
762 * Returns: The next rgrp in the recent list
765 static struct gfs2_rgrpd
*recent_rgrp_next(struct gfs2_rgrpd
*cur_rgd
,
768 struct gfs2_sbd
*sdp
= cur_rgd
->rd_sbd
;
769 struct list_head
*head
;
770 struct gfs2_rgrpd
*rgd
;
772 spin_lock(&sdp
->sd_rindex_spin
);
774 head
= &sdp
->sd_rindex_recent_list
;
776 list_for_each_entry(rgd
, head
, rd_recent
) {
777 if (rgd
== cur_rgd
) {
778 if (cur_rgd
->rd_recent
.next
!= head
)
779 rgd
= list_entry(cur_rgd
->rd_recent
.next
,
780 struct gfs2_rgrpd
, rd_recent
);
785 list_del(&cur_rgd
->rd_recent
);
792 if (!list_empty(head
))
793 rgd
= list_entry(head
->next
, struct gfs2_rgrpd
, rd_recent
);
796 spin_unlock(&sdp
->sd_rindex_spin
);
801 * recent_rgrp_add - add an RG to tail of "recent" list
802 * @new_rgd: The rgrp to add
806 static void recent_rgrp_add(struct gfs2_rgrpd
*new_rgd
)
808 struct gfs2_sbd
*sdp
= new_rgd
->rd_sbd
;
809 struct gfs2_rgrpd
*rgd
;
810 unsigned int count
= 0;
811 unsigned int max
= sdp
->sd_rgrps
/ gfs2_jindex_size(sdp
);
813 spin_lock(&sdp
->sd_rindex_spin
);
815 list_for_each_entry(rgd
, &sdp
->sd_rindex_recent_list
, rd_recent
) {
822 list_add_tail(&new_rgd
->rd_recent
, &sdp
->sd_rindex_recent_list
);
825 spin_unlock(&sdp
->sd_rindex_spin
);
829 * forward_rgrp_get - get an rgrp to try next from full list
830 * @sdp: The GFS2 superblock
832 * Returns: The rgrp to try next
835 static struct gfs2_rgrpd
*forward_rgrp_get(struct gfs2_sbd
*sdp
)
837 struct gfs2_rgrpd
*rgd
;
838 unsigned int journals
= gfs2_jindex_size(sdp
);
839 unsigned int rg
= 0, x
;
841 spin_lock(&sdp
->sd_rindex_spin
);
843 rgd
= sdp
->sd_rindex_forward
;
845 if (sdp
->sd_rgrps
>= journals
)
846 rg
= sdp
->sd_rgrps
* sdp
->sd_jdesc
->jd_jid
/ journals
;
848 for (x
= 0, rgd
= gfs2_rgrpd_get_first(sdp
); x
< rg
;
849 x
++, rgd
= gfs2_rgrpd_get_next(rgd
))
852 sdp
->sd_rindex_forward
= rgd
;
855 spin_unlock(&sdp
->sd_rindex_spin
);
861 * forward_rgrp_set - set the forward rgrp pointer
862 * @sdp: the filesystem
863 * @rgd: The new forward rgrp
867 static void forward_rgrp_set(struct gfs2_sbd
*sdp
, struct gfs2_rgrpd
*rgd
)
869 spin_lock(&sdp
->sd_rindex_spin
);
870 sdp
->sd_rindex_forward
= rgd
;
871 spin_unlock(&sdp
->sd_rindex_spin
);
875 * get_local_rgrp - Choose and lock a rgrp for allocation
876 * @ip: the inode to reserve space for
877 * @rgp: the chosen and locked rgrp
879 * Try to acquire rgrp in way which avoids contending with others.
884 static int get_local_rgrp(struct gfs2_inode
*ip
)
886 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
887 struct gfs2_rgrpd
*rgd
, *begin
= NULL
;
888 struct gfs2_alloc
*al
= &ip
->i_alloc
;
889 int flags
= LM_FLAG_TRY
;
894 /* Try recently successful rgrps */
896 rgd
= recent_rgrp_first(sdp
, ip
->i_last_rg_alloc
);
899 error
= gfs2_glock_nq_init(rgd
->rd_gl
, LM_ST_EXCLUSIVE
,
900 LM_FLAG_TRY
, &al
->al_rgd_gh
);
903 if (try_rgrp_fit(rgd
, al
))
905 gfs2_glock_dq_uninit(&al
->al_rgd_gh
);
906 rgd
= recent_rgrp_next(rgd
, 1);
910 rgd
= recent_rgrp_next(rgd
, 0);
918 /* Go through full list of rgrps */
920 begin
= rgd
= forward_rgrp_get(sdp
);
923 error
= gfs2_glock_nq_init(rgd
->rd_gl
, LM_ST_EXCLUSIVE
, flags
,
927 if (try_rgrp_fit(rgd
, al
))
929 gfs2_glock_dq_uninit(&al
->al_rgd_gh
);
940 rgd
= gfs2_rgrpd_get_next(rgd
);
942 rgd
= gfs2_rgrpd_get_first(sdp
);
945 if (++loops
>= 2 || !skipped
)
952 ip
->i_last_rg_alloc
= rgd
->rd_ri
.ri_addr
;
955 recent_rgrp_add(rgd
);
956 rgd
= gfs2_rgrpd_get_next(rgd
);
958 rgd
= gfs2_rgrpd_get_first(sdp
);
959 forward_rgrp_set(sdp
, rgd
);
966 * gfs2_inplace_reserve_i - Reserve space in the filesystem
967 * @ip: the inode to reserve space for
972 int gfs2_inplace_reserve_i(struct gfs2_inode
*ip
, char *file
, unsigned int line
)
974 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
975 struct gfs2_alloc
*al
= &ip
->i_alloc
;
978 if (gfs2_assert_warn(sdp
, al
->al_requested
))
981 error
= gfs2_rindex_hold(sdp
, &al
->al_ri_gh
);
985 error
= get_local_rgrp(ip
);
987 gfs2_glock_dq_uninit(&al
->al_ri_gh
);
998 * gfs2_inplace_release - release an inplace reservation
999 * @ip: the inode the reservation was taken out on
1001 * Release a reservation made by gfs2_inplace_reserve().
1004 void gfs2_inplace_release(struct gfs2_inode
*ip
)
1006 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
1007 struct gfs2_alloc
*al
= &ip
->i_alloc
;
1009 if (gfs2_assert_warn(sdp
, al
->al_alloced
<= al
->al_requested
) == -1)
1010 fs_warn(sdp
, "al_alloced = %u, al_requested = %u "
1011 "al_file = %s, al_line = %u\n",
1012 al
->al_alloced
, al
->al_requested
, al
->al_file
,
1016 gfs2_glock_dq_uninit(&al
->al_rgd_gh
);
1017 gfs2_glock_dq_uninit(&al
->al_ri_gh
);
1021 * gfs2_get_block_type - Check a block in a RG is of given type
1022 * @rgd: the resource group holding the block
1023 * @block: the block number
1025 * Returns: The block type (GFS2_BLKST_*)
1028 unsigned char gfs2_get_block_type(struct gfs2_rgrpd
*rgd
, u64 block
)
1030 struct gfs2_bitmap
*bi
= NULL
;
1031 u32 length
, rgrp_block
, buf_block
;
1035 length
= rgd
->rd_ri
.ri_length
;
1036 rgrp_block
= block
- rgd
->rd_ri
.ri_data0
;
1038 for (buf
= 0; buf
< length
; buf
++) {
1039 bi
= rgd
->rd_bits
+ buf
;
1040 if (rgrp_block
< (bi
->bi_start
+ bi
->bi_len
) * GFS2_NBBY
)
1044 gfs2_assert(rgd
->rd_sbd
, buf
< length
);
1045 buf_block
= rgrp_block
- bi
->bi_start
* GFS2_NBBY
;
1047 type
= gfs2_testbit(rgd
, bi
->bi_bh
->b_data
+ bi
->bi_offset
,
1048 bi
->bi_len
, buf_block
);
1054 * rgblk_search - find a block in @old_state, change allocation
1055 * state to @new_state
1056 * @rgd: the resource group descriptor
1057 * @goal: the goal block within the RG (start here to search for avail block)
1058 * @old_state: GFS2_BLKST_XXX the before-allocation state to find
1059 * @new_state: GFS2_BLKST_XXX the after-allocation block state
1061 * Walk rgrp's bitmap to find bits that represent a block in @old_state.
1062 * Add the found bitmap buffer to the transaction.
1063 * Set the found bits to @new_state to change block's allocation state.
1065 * This function never fails, because we wouldn't call it unless we
1066 * know (from reservation results, etc.) that a block is available.
1068 * Scope of @goal and returned block is just within rgrp, not the whole
1071 * Returns: the block number allocated
1074 static u32
rgblk_search(struct gfs2_rgrpd
*rgd
, u32 goal
,
1075 unsigned char old_state
, unsigned char new_state
)
1077 struct gfs2_bitmap
*bi
= NULL
;
1078 u32 length
= rgd
->rd_ri
.ri_length
;
1080 unsigned int buf
, x
;
1082 /* Find bitmap block that contains bits for goal block */
1083 for (buf
= 0; buf
< length
; buf
++) {
1084 bi
= rgd
->rd_bits
+ buf
;
1085 if (goal
< (bi
->bi_start
+ bi
->bi_len
) * GFS2_NBBY
)
1089 gfs2_assert(rgd
->rd_sbd
, buf
< length
);
1091 /* Convert scope of "goal" from rgrp-wide to within found bit block */
1092 goal
-= bi
->bi_start
* GFS2_NBBY
;
1094 /* Search (up to entire) bitmap in this rgrp for allocatable block.
1095 "x <= length", instead of "x < length", because we typically start
1096 the search in the middle of a bit block, but if we can't find an
1097 allocatable block anywhere else, we want to be able wrap around and
1098 search in the first part of our first-searched bit block. */
1099 for (x
= 0; x
<= length
; x
++) {
1101 blk
= gfs2_bitfit(rgd
, bi
->bi_clone
+ bi
->bi_offset
,
1102 bi
->bi_len
, goal
, old_state
);
1104 blk
= gfs2_bitfit(rgd
,
1105 bi
->bi_bh
->b_data
+ bi
->bi_offset
,
1106 bi
->bi_len
, goal
, old_state
);
1107 if (blk
!= BFITNOENT
)
1110 /* Try next bitmap block (wrap back to rgrp header if at end) */
1111 buf
= (buf
+ 1) % length
;
1112 bi
= rgd
->rd_bits
+ buf
;
1116 if (gfs2_assert_withdraw(rgd
->rd_sbd
, x
<= length
))
1119 gfs2_trans_add_bh(rgd
->rd_gl
, bi
->bi_bh
, 1);
1120 gfs2_setbit(rgd
, bi
->bi_bh
->b_data
+ bi
->bi_offset
,
1121 bi
->bi_len
, blk
, new_state
);
1123 gfs2_setbit(rgd
, bi
->bi_clone
+ bi
->bi_offset
,
1124 bi
->bi_len
, blk
, new_state
);
1126 return bi
->bi_start
* GFS2_NBBY
+ blk
;
1130 * rgblk_free - Change alloc state of given block(s)
1131 * @sdp: the filesystem
1132 * @bstart: the start of a run of blocks to free
1133 * @blen: the length of the block run (all must lie within ONE RG!)
1134 * @new_state: GFS2_BLKST_XXX the after-allocation block state
1136 * Returns: Resource group containing the block(s)
1139 static struct gfs2_rgrpd
*rgblk_free(struct gfs2_sbd
*sdp
, u64 bstart
,
1140 u32 blen
, unsigned char new_state
)
1142 struct gfs2_rgrpd
*rgd
;
1143 struct gfs2_bitmap
*bi
= NULL
;
1144 u32 length
, rgrp_blk
, buf_blk
;
1147 rgd
= gfs2_blk2rgrpd(sdp
, bstart
);
1149 if (gfs2_consist(sdp
))
1150 fs_err(sdp
, "block = %llu\n", (unsigned long long)bstart
);
1154 length
= rgd
->rd_ri
.ri_length
;
1156 rgrp_blk
= bstart
- rgd
->rd_ri
.ri_data0
;
1159 for (buf
= 0; buf
< length
; buf
++) {
1160 bi
= rgd
->rd_bits
+ buf
;
1161 if (rgrp_blk
< (bi
->bi_start
+ bi
->bi_len
) * GFS2_NBBY
)
1165 gfs2_assert(rgd
->rd_sbd
, buf
< length
);
1167 buf_blk
= rgrp_blk
- bi
->bi_start
* GFS2_NBBY
;
1170 if (!bi
->bi_clone
) {
1171 bi
->bi_clone
= kmalloc(bi
->bi_bh
->b_size
,
1172 GFP_NOFS
| __GFP_NOFAIL
);
1173 memcpy(bi
->bi_clone
+ bi
->bi_offset
,
1174 bi
->bi_bh
->b_data
+ bi
->bi_offset
,
1177 gfs2_trans_add_bh(rgd
->rd_gl
, bi
->bi_bh
, 1);
1178 gfs2_setbit(rgd
, bi
->bi_bh
->b_data
+ bi
->bi_offset
,
1179 bi
->bi_len
, buf_blk
, new_state
);
1186 * gfs2_alloc_data - Allocate a data block
1187 * @ip: the inode to allocate the data block for
1189 * Returns: the allocated block
1192 u64
gfs2_alloc_data(struct gfs2_inode
*ip
)
1194 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
1195 struct gfs2_alloc
*al
= &ip
->i_alloc
;
1196 struct gfs2_rgrpd
*rgd
= al
->al_rgd
;
1200 if (rgrp_contains_block(&rgd
->rd_ri
, ip
->i_di
.di_goal_data
))
1201 goal
= ip
->i_di
.di_goal_data
- rgd
->rd_ri
.ri_data0
;
1203 goal
= rgd
->rd_last_alloc_data
;
1205 blk
= rgblk_search(rgd
, goal
, GFS2_BLKST_FREE
, GFS2_BLKST_USED
);
1206 rgd
->rd_last_alloc_data
= blk
;
1208 block
= rgd
->rd_ri
.ri_data0
+ blk
;
1209 ip
->i_di
.di_goal_data
= block
;
1211 gfs2_assert_withdraw(sdp
, rgd
->rd_rg
.rg_free
);
1212 rgd
->rd_rg
.rg_free
--;
1214 gfs2_trans_add_bh(rgd
->rd_gl
, rgd
->rd_bits
[0].bi_bh
, 1);
1215 gfs2_rgrp_out(&rgd
->rd_rg
, rgd
->rd_bits
[0].bi_bh
->b_data
);
1219 gfs2_statfs_change(sdp
, 0, -1, 0);
1220 gfs2_quota_change(ip
, +1, ip
->i_di
.di_uid
, ip
->i_di
.di_gid
);
1222 spin_lock(&sdp
->sd_rindex_spin
);
1223 rgd
->rd_free_clone
--;
1224 spin_unlock(&sdp
->sd_rindex_spin
);
1230 * gfs2_alloc_meta - Allocate a metadata block
1231 * @ip: the inode to allocate the metadata block for
1233 * Returns: the allocated block
1236 u64
gfs2_alloc_meta(struct gfs2_inode
*ip
)
1238 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
1239 struct gfs2_alloc
*al
= &ip
->i_alloc
;
1240 struct gfs2_rgrpd
*rgd
= al
->al_rgd
;
1244 if (rgrp_contains_block(&rgd
->rd_ri
, ip
->i_di
.di_goal_meta
))
1245 goal
= ip
->i_di
.di_goal_meta
- rgd
->rd_ri
.ri_data0
;
1247 goal
= rgd
->rd_last_alloc_meta
;
1249 blk
= rgblk_search(rgd
, goal
, GFS2_BLKST_FREE
, GFS2_BLKST_USED
);
1250 rgd
->rd_last_alloc_meta
= blk
;
1252 block
= rgd
->rd_ri
.ri_data0
+ blk
;
1253 ip
->i_di
.di_goal_meta
= block
;
1255 gfs2_assert_withdraw(sdp
, rgd
->rd_rg
.rg_free
);
1256 rgd
->rd_rg
.rg_free
--;
1258 gfs2_trans_add_bh(rgd
->rd_gl
, rgd
->rd_bits
[0].bi_bh
, 1);
1259 gfs2_rgrp_out(&rgd
->rd_rg
, rgd
->rd_bits
[0].bi_bh
->b_data
);
1263 gfs2_statfs_change(sdp
, 0, -1, 0);
1264 gfs2_quota_change(ip
, +1, ip
->i_di
.di_uid
, ip
->i_di
.di_gid
);
1265 gfs2_trans_add_unrevoke(sdp
, block
);
1267 spin_lock(&sdp
->sd_rindex_spin
);
1268 rgd
->rd_free_clone
--;
1269 spin_unlock(&sdp
->sd_rindex_spin
);
1275 * gfs2_alloc_di - Allocate a dinode
1276 * @dip: the directory that the inode is going in
1278 * Returns: the block allocated
1281 u64
gfs2_alloc_di(struct gfs2_inode
*dip
, u64
*generation
)
1283 struct gfs2_sbd
*sdp
= GFS2_SB(&dip
->i_inode
);
1284 struct gfs2_alloc
*al
= &dip
->i_alloc
;
1285 struct gfs2_rgrpd
*rgd
= al
->al_rgd
;
1289 blk
= rgblk_search(rgd
, rgd
->rd_last_alloc_meta
,
1290 GFS2_BLKST_FREE
, GFS2_BLKST_DINODE
);
1292 rgd
->rd_last_alloc_meta
= blk
;
1294 block
= rgd
->rd_ri
.ri_data0
+ blk
;
1296 gfs2_assert_withdraw(sdp
, rgd
->rd_rg
.rg_free
);
1297 rgd
->rd_rg
.rg_free
--;
1298 rgd
->rd_rg
.rg_dinodes
++;
1299 *generation
= rgd
->rd_rg
.rg_igeneration
++;
1300 gfs2_trans_add_bh(rgd
->rd_gl
, rgd
->rd_bits
[0].bi_bh
, 1);
1301 gfs2_rgrp_out(&rgd
->rd_rg
, rgd
->rd_bits
[0].bi_bh
->b_data
);
1305 gfs2_statfs_change(sdp
, 0, -1, +1);
1306 gfs2_trans_add_unrevoke(sdp
, block
);
1308 spin_lock(&sdp
->sd_rindex_spin
);
1309 rgd
->rd_free_clone
--;
1310 spin_unlock(&sdp
->sd_rindex_spin
);
1316 * gfs2_free_data - free a contiguous run of data block(s)
1317 * @ip: the inode these blocks are being freed from
1318 * @bstart: first block of a run of contiguous blocks
1319 * @blen: the length of the block run
1323 void gfs2_free_data(struct gfs2_inode
*ip
, u64 bstart
, u32 blen
)
1325 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
1326 struct gfs2_rgrpd
*rgd
;
1328 rgd
= rgblk_free(sdp
, bstart
, blen
, GFS2_BLKST_FREE
);
1332 rgd
->rd_rg
.rg_free
+= blen
;
1334 gfs2_trans_add_bh(rgd
->rd_gl
, rgd
->rd_bits
[0].bi_bh
, 1);
1335 gfs2_rgrp_out(&rgd
->rd_rg
, rgd
->rd_bits
[0].bi_bh
->b_data
);
1337 gfs2_trans_add_rg(rgd
);
1339 gfs2_statfs_change(sdp
, 0, +blen
, 0);
1340 gfs2_quota_change(ip
, -(s64
)blen
,
1341 ip
->i_di
.di_uid
, ip
->i_di
.di_gid
);
1345 * gfs2_free_meta - free a contiguous run of data block(s)
1346 * @ip: the inode these blocks are being freed from
1347 * @bstart: first block of a run of contiguous blocks
1348 * @blen: the length of the block run
1352 void gfs2_free_meta(struct gfs2_inode
*ip
, u64 bstart
, u32 blen
)
1354 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
1355 struct gfs2_rgrpd
*rgd
;
1357 rgd
= rgblk_free(sdp
, bstart
, blen
, GFS2_BLKST_FREE
);
1361 rgd
->rd_rg
.rg_free
+= blen
;
1363 gfs2_trans_add_bh(rgd
->rd_gl
, rgd
->rd_bits
[0].bi_bh
, 1);
1364 gfs2_rgrp_out(&rgd
->rd_rg
, rgd
->rd_bits
[0].bi_bh
->b_data
);
1366 gfs2_trans_add_rg(rgd
);
1368 gfs2_statfs_change(sdp
, 0, +blen
, 0);
1369 gfs2_quota_change(ip
, -(s64
)blen
, ip
->i_di
.di_uid
, ip
->i_di
.di_gid
);
1370 gfs2_meta_wipe(ip
, bstart
, blen
);
1373 void gfs2_unlink_di(struct inode
*inode
)
1375 struct gfs2_inode
*ip
= GFS2_I(inode
);
1376 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
1377 struct gfs2_rgrpd
*rgd
;
1378 u64 blkno
= ip
->i_num
.no_addr
;
1380 rgd
= rgblk_free(sdp
, blkno
, 1, GFS2_BLKST_UNLINKED
);
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
);
1385 gfs2_trans_add_rg(rgd
);
1388 static void gfs2_free_uninit_di(struct gfs2_rgrpd
*rgd
, u64 blkno
)
1390 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
1391 struct gfs2_rgrpd
*tmp_rgd
;
1393 tmp_rgd
= rgblk_free(sdp
, blkno
, 1, GFS2_BLKST_FREE
);
1396 gfs2_assert_withdraw(sdp
, rgd
== tmp_rgd
);
1398 if (!rgd
->rd_rg
.rg_dinodes
)
1399 gfs2_consist_rgrpd(rgd
);
1400 rgd
->rd_rg
.rg_dinodes
--;
1401 rgd
->rd_rg
.rg_free
++;
1403 gfs2_trans_add_bh(rgd
->rd_gl
, rgd
->rd_bits
[0].bi_bh
, 1);
1404 gfs2_rgrp_out(&rgd
->rd_rg
, rgd
->rd_bits
[0].bi_bh
->b_data
);
1406 gfs2_statfs_change(sdp
, 0, +1, -1);
1407 gfs2_trans_add_rg(rgd
);
1411 void gfs2_free_di(struct gfs2_rgrpd
*rgd
, struct gfs2_inode
*ip
)
1413 gfs2_free_uninit_di(rgd
, ip
->i_num
.no_addr
);
1414 gfs2_quota_change(ip
, -1, ip
->i_di
.di_uid
, ip
->i_di
.di_gid
);
1415 gfs2_meta_wipe(ip
, ip
->i_num
.no_addr
, 1);
1419 * gfs2_rlist_add - add a RG to a list of RGs
1420 * @sdp: the filesystem
1421 * @rlist: the list of resource groups
1424 * Figure out what RG a block belongs to and add that RG to the list
1426 * FIXME: Don't use NOFAIL
1430 void gfs2_rlist_add(struct gfs2_sbd
*sdp
, struct gfs2_rgrp_list
*rlist
,
1433 struct gfs2_rgrpd
*rgd
;
1434 struct gfs2_rgrpd
**tmp
;
1435 unsigned int new_space
;
1438 if (gfs2_assert_warn(sdp
, !rlist
->rl_ghs
))
1441 rgd
= gfs2_blk2rgrpd(sdp
, block
);
1443 if (gfs2_consist(sdp
))
1444 fs_err(sdp
, "block = %llu\n", (unsigned long long)block
);
1448 for (x
= 0; x
< rlist
->rl_rgrps
; x
++)
1449 if (rlist
->rl_rgd
[x
] == rgd
)
1452 if (rlist
->rl_rgrps
== rlist
->rl_space
) {
1453 new_space
= rlist
->rl_space
+ 10;
1455 tmp
= kcalloc(new_space
, sizeof(struct gfs2_rgrpd
*),
1456 GFP_NOFS
| __GFP_NOFAIL
);
1458 if (rlist
->rl_rgd
) {
1459 memcpy(tmp
, rlist
->rl_rgd
,
1460 rlist
->rl_space
* sizeof(struct gfs2_rgrpd
*));
1461 kfree(rlist
->rl_rgd
);
1464 rlist
->rl_space
= new_space
;
1465 rlist
->rl_rgd
= tmp
;
1468 rlist
->rl_rgd
[rlist
->rl_rgrps
++] = rgd
;
1472 * gfs2_rlist_alloc - all RGs have been added to the rlist, now allocate
1473 * and initialize an array of glock holders for them
1474 * @rlist: the list of resource groups
1475 * @state: the lock state to acquire the RG lock in
1476 * @flags: the modifier flags for the holder structures
1478 * FIXME: Don't use NOFAIL
1482 void gfs2_rlist_alloc(struct gfs2_rgrp_list
*rlist
, unsigned int state
,
1487 rlist
->rl_ghs
= kcalloc(rlist
->rl_rgrps
, sizeof(struct gfs2_holder
),
1488 GFP_NOFS
| __GFP_NOFAIL
);
1489 for (x
= 0; x
< rlist
->rl_rgrps
; x
++)
1490 gfs2_holder_init(rlist
->rl_rgd
[x
]->rd_gl
,
1496 * gfs2_rlist_free - free a resource group list
1497 * @list: the list of resource groups
1501 void gfs2_rlist_free(struct gfs2_rgrp_list
*rlist
)
1505 kfree(rlist
->rl_rgd
);
1507 if (rlist
->rl_ghs
) {
1508 for (x
= 0; x
< rlist
->rl_rgrps
; x
++)
1509 gfs2_holder_uninit(&rlist
->rl_ghs
[x
]);
1510 kfree(rlist
->rl_ghs
);