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/slab.h>
11 #include <linux/spinlock.h>
12 #include <linux/completion.h>
13 #include <linux/buffer_head.h>
15 #include <linux/gfs2_ondisk.h>
16 #include <linux/lm_interface.h>
31 #define BFITNOENT ((u32)~0)
34 * These routines are used by the resource group routines (rgrp.c)
35 * to keep track of block allocation. Each block is represented by two
36 * bits. So, each byte represents GFS2_NBBY (i.e. 4) blocks.
39 * 1 = Used (not metadata)
40 * 2 = Unlinked (still in use) inode
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
, u32 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
, u32 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 u32
gfs2_bitfit(struct gfs2_rgrpd
*rgd
, unsigned char *buffer
,
125 unsigned int buflen
, u32 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 u32
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 u32 length
= rgd
->rd_ri
.ri_length
;
210 memset(count
, 0, 4 * sizeof(u32
));
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] + count
[2] != tmp
) {
233 if (gfs2_consist_rgrpd(rgd
))
234 fs_err(sdp
, "used data mismatch: %u != %u\n",
239 if (count
[3] != rgd
->rd_rg
.rg_dinodes
) {
240 if (gfs2_consist_rgrpd(rgd
))
241 fs_err(sdp
, "used metadata mismatch: %u != %u\n",
242 count
[3], rgd
->rd_rg
.rg_dinodes
);
246 if (count
[2] > count
[3]) {
247 if (gfs2_consist_rgrpd(rgd
))
248 fs_err(sdp
, "unlinked inodes > inodes: %u\n",
255 static inline int rgrp_contains_block(struct gfs2_rindex_host
*ri
, u64 block
)
257 u64 first
= ri
->ri_data0
;
258 u64 last
= first
+ ri
->ri_data
;
259 return first
<= block
&& block
< last
;
263 * gfs2_blk2rgrpd - Find resource group for a given data/meta block number
264 * @sdp: The GFS2 superblock
265 * @n: The data block number
267 * Returns: The resource group, or NULL if not found
270 struct gfs2_rgrpd
*gfs2_blk2rgrpd(struct gfs2_sbd
*sdp
, u64 blk
)
272 struct gfs2_rgrpd
*rgd
;
274 spin_lock(&sdp
->sd_rindex_spin
);
276 list_for_each_entry(rgd
, &sdp
->sd_rindex_mru_list
, rd_list_mru
) {
277 if (rgrp_contains_block(&rgd
->rd_ri
, blk
)) {
278 list_move(&rgd
->rd_list_mru
, &sdp
->sd_rindex_mru_list
);
279 spin_unlock(&sdp
->sd_rindex_spin
);
284 spin_unlock(&sdp
->sd_rindex_spin
);
290 * gfs2_rgrpd_get_first - get the first Resource Group in the filesystem
291 * @sdp: The GFS2 superblock
293 * Returns: The first rgrp in the filesystem
296 struct gfs2_rgrpd
*gfs2_rgrpd_get_first(struct gfs2_sbd
*sdp
)
298 gfs2_assert(sdp
, !list_empty(&sdp
->sd_rindex_list
));
299 return list_entry(sdp
->sd_rindex_list
.next
, struct gfs2_rgrpd
, rd_list
);
303 * gfs2_rgrpd_get_next - get the next RG
306 * Returns: The next rgrp
309 struct gfs2_rgrpd
*gfs2_rgrpd_get_next(struct gfs2_rgrpd
*rgd
)
311 if (rgd
->rd_list
.next
== &rgd
->rd_sbd
->sd_rindex_list
)
313 return list_entry(rgd
->rd_list
.next
, struct gfs2_rgrpd
, rd_list
);
316 static void clear_rgrpdi(struct gfs2_sbd
*sdp
)
318 struct list_head
*head
;
319 struct gfs2_rgrpd
*rgd
;
320 struct gfs2_glock
*gl
;
322 spin_lock(&sdp
->sd_rindex_spin
);
323 sdp
->sd_rindex_forward
= NULL
;
324 head
= &sdp
->sd_rindex_recent_list
;
325 while (!list_empty(head
)) {
326 rgd
= list_entry(head
->next
, struct gfs2_rgrpd
, rd_recent
);
327 list_del(&rgd
->rd_recent
);
329 spin_unlock(&sdp
->sd_rindex_spin
);
331 head
= &sdp
->sd_rindex_list
;
332 while (!list_empty(head
)) {
333 rgd
= list_entry(head
->next
, struct gfs2_rgrpd
, rd_list
);
336 list_del(&rgd
->rd_list
);
337 list_del(&rgd
->rd_list_mru
);
340 gl
->gl_object
= NULL
;
349 void gfs2_clear_rgrpd(struct gfs2_sbd
*sdp
)
351 mutex_lock(&sdp
->sd_rindex_mutex
);
353 mutex_unlock(&sdp
->sd_rindex_mutex
);
357 * gfs2_compute_bitstructs - Compute the bitmap sizes
358 * @rgd: The resource group descriptor
360 * Calculates bitmap descriptors, one for each block that contains bitmap data
365 static int compute_bitstructs(struct gfs2_rgrpd
*rgd
)
367 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
368 struct gfs2_bitmap
*bi
;
369 u32 length
= rgd
->rd_ri
.ri_length
; /* # blocks in hdr & bitmap */
370 u32 bytes_left
, bytes
;
376 rgd
->rd_bits
= kcalloc(length
, sizeof(struct gfs2_bitmap
), GFP_NOFS
);
380 bytes_left
= rgd
->rd_ri
.ri_bitbytes
;
382 for (x
= 0; x
< length
; x
++) {
383 bi
= rgd
->rd_bits
+ x
;
385 /* small rgrp; bitmap stored completely in header block */
388 bi
->bi_offset
= sizeof(struct gfs2_rgrp
);
393 bytes
= sdp
->sd_sb
.sb_bsize
- sizeof(struct gfs2_rgrp
);
394 bi
->bi_offset
= sizeof(struct gfs2_rgrp
);
398 } else if (x
+ 1 == length
) {
400 bi
->bi_offset
= sizeof(struct gfs2_meta_header
);
401 bi
->bi_start
= rgd
->rd_ri
.ri_bitbytes
- bytes_left
;
405 bytes
= sdp
->sd_sb
.sb_bsize
-
406 sizeof(struct gfs2_meta_header
);
407 bi
->bi_offset
= sizeof(struct gfs2_meta_header
);
408 bi
->bi_start
= rgd
->rd_ri
.ri_bitbytes
- bytes_left
;
416 gfs2_consist_rgrpd(rgd
);
419 bi
= rgd
->rd_bits
+ (length
- 1);
420 if ((bi
->bi_start
+ bi
->bi_len
) * GFS2_NBBY
!= rgd
->rd_ri
.ri_data
) {
421 if (gfs2_consist_rgrpd(rgd
)) {
422 gfs2_rindex_print(&rgd
->rd_ri
);
423 fs_err(sdp
, "start=%u len=%u offset=%u\n",
424 bi
->bi_start
, bi
->bi_len
, bi
->bi_offset
);
433 * gfs2_ri_update - Pull in a new resource index from the disk
434 * @gl: The glock covering the rindex inode
436 * Returns: 0 on successful update, error code otherwise
439 static int gfs2_ri_update(struct gfs2_inode
*ip
)
441 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
442 struct inode
*inode
= &ip
->i_inode
;
443 struct gfs2_rgrpd
*rgd
;
444 char buf
[sizeof(struct gfs2_rindex
)];
445 struct file_ra_state ra_state
;
446 u64 junk
= ip
->i_di
.di_size
;
449 if (do_div(junk
, sizeof(struct gfs2_rindex
))) {
450 gfs2_consist_inode(ip
);
456 file_ra_state_init(&ra_state
, inode
->i_mapping
);
457 for (sdp
->sd_rgrps
= 0;; sdp
->sd_rgrps
++) {
458 loff_t pos
= sdp
->sd_rgrps
* sizeof(struct gfs2_rindex
);
459 error
= gfs2_internal_read(ip
, &ra_state
, buf
, &pos
,
460 sizeof(struct gfs2_rindex
));
463 if (error
!= sizeof(struct gfs2_rindex
)) {
469 rgd
= kzalloc(sizeof(struct gfs2_rgrpd
), GFP_NOFS
);
474 mutex_init(&rgd
->rd_mutex
);
475 lops_init_le(&rgd
->rd_le
, &gfs2_rg_lops
);
478 list_add_tail(&rgd
->rd_list
, &sdp
->sd_rindex_list
);
479 list_add_tail(&rgd
->rd_list_mru
, &sdp
->sd_rindex_mru_list
);
481 gfs2_rindex_in(&rgd
->rd_ri
, buf
);
482 error
= compute_bitstructs(rgd
);
486 error
= gfs2_glock_get(sdp
, rgd
->rd_ri
.ri_addr
,
487 &gfs2_rgrp_glops
, CREATE
, &rgd
->rd_gl
);
491 rgd
->rd_gl
->gl_object
= rgd
;
492 rgd
->rd_rg_vn
= rgd
->rd_gl
->gl_vn
- 1;
495 sdp
->sd_rindex_vn
= ip
->i_gl
->gl_vn
;
504 * gfs2_rindex_hold - Grab a lock on the rindex
505 * @sdp: The GFS2 superblock
506 * @ri_gh: the glock holder
508 * We grab a lock on the rindex inode to make sure that it doesn't
509 * change whilst we are performing an operation. We keep this lock
510 * for quite long periods of time compared to other locks. This
511 * doesn't matter, since it is shared and it is very, very rarely
512 * accessed in the exclusive mode (i.e. only when expanding the filesystem).
514 * This makes sure that we're using the latest copy of the resource index
515 * special file, which might have been updated if someone expanded the
516 * filesystem (via gfs2_grow utility), which adds new resource groups.
518 * Returns: 0 on success, error code otherwise
521 int gfs2_rindex_hold(struct gfs2_sbd
*sdp
, struct gfs2_holder
*ri_gh
)
523 struct gfs2_inode
*ip
= GFS2_I(sdp
->sd_rindex
);
524 struct gfs2_glock
*gl
= ip
->i_gl
;
527 error
= gfs2_glock_nq_init(gl
, LM_ST_SHARED
, 0, ri_gh
);
531 /* Read new copy from disk if we don't have the latest */
532 if (sdp
->sd_rindex_vn
!= gl
->gl_vn
) {
533 mutex_lock(&sdp
->sd_rindex_mutex
);
534 if (sdp
->sd_rindex_vn
!= gl
->gl_vn
) {
535 error
= gfs2_ri_update(ip
);
537 gfs2_glock_dq_uninit(ri_gh
);
539 mutex_unlock(&sdp
->sd_rindex_mutex
);
546 * gfs2_rgrp_bh_get - Read in a RG's header and bitmaps
547 * @rgd: the struct gfs2_rgrpd describing the RG to read in
549 * Read in all of a Resource Group's header and bitmap blocks.
550 * Caller must eventually call gfs2_rgrp_relse() to free the bitmaps.
555 int gfs2_rgrp_bh_get(struct gfs2_rgrpd
*rgd
)
557 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
558 struct gfs2_glock
*gl
= rgd
->rd_gl
;
559 unsigned int length
= rgd
->rd_ri
.ri_length
;
560 struct gfs2_bitmap
*bi
;
564 mutex_lock(&rgd
->rd_mutex
);
566 spin_lock(&sdp
->sd_rindex_spin
);
567 if (rgd
->rd_bh_count
) {
569 spin_unlock(&sdp
->sd_rindex_spin
);
570 mutex_unlock(&rgd
->rd_mutex
);
573 spin_unlock(&sdp
->sd_rindex_spin
);
575 for (x
= 0; x
< length
; x
++) {
576 bi
= rgd
->rd_bits
+ x
;
577 error
= gfs2_meta_read(gl
, rgd
->rd_ri
.ri_addr
+ x
, 0, &bi
->bi_bh
);
582 for (y
= length
; y
--;) {
583 bi
= rgd
->rd_bits
+ y
;
584 error
= gfs2_meta_wait(sdp
, bi
->bi_bh
);
587 if (gfs2_metatype_check(sdp
, bi
->bi_bh
, y
? GFS2_METATYPE_RB
:
594 if (rgd
->rd_rg_vn
!= gl
->gl_vn
) {
595 gfs2_rgrp_in(&rgd
->rd_rg
, (rgd
->rd_bits
[0].bi_bh
)->b_data
);
596 rgd
->rd_rg_vn
= gl
->gl_vn
;
599 spin_lock(&sdp
->sd_rindex_spin
);
600 rgd
->rd_free_clone
= rgd
->rd_rg
.rg_free
;
602 spin_unlock(&sdp
->sd_rindex_spin
);
604 mutex_unlock(&rgd
->rd_mutex
);
610 bi
= rgd
->rd_bits
+ x
;
613 gfs2_assert_warn(sdp
, !bi
->bi_clone
);
615 mutex_unlock(&rgd
->rd_mutex
);
620 void gfs2_rgrp_bh_hold(struct gfs2_rgrpd
*rgd
)
622 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
624 spin_lock(&sdp
->sd_rindex_spin
);
625 gfs2_assert_warn(rgd
->rd_sbd
, rgd
->rd_bh_count
);
627 spin_unlock(&sdp
->sd_rindex_spin
);
631 * gfs2_rgrp_bh_put - Release RG bitmaps read in with gfs2_rgrp_bh_get()
632 * @rgd: the struct gfs2_rgrpd describing the RG to read in
636 void gfs2_rgrp_bh_put(struct gfs2_rgrpd
*rgd
)
638 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
639 int x
, length
= rgd
->rd_ri
.ri_length
;
641 spin_lock(&sdp
->sd_rindex_spin
);
642 gfs2_assert_warn(rgd
->rd_sbd
, rgd
->rd_bh_count
);
643 if (--rgd
->rd_bh_count
) {
644 spin_unlock(&sdp
->sd_rindex_spin
);
648 for (x
= 0; x
< length
; x
++) {
649 struct gfs2_bitmap
*bi
= rgd
->rd_bits
+ x
;
656 spin_unlock(&sdp
->sd_rindex_spin
);
659 void gfs2_rgrp_repolish_clones(struct gfs2_rgrpd
*rgd
)
661 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
662 unsigned int length
= rgd
->rd_ri
.ri_length
;
665 for (x
= 0; x
< length
; x
++) {
666 struct gfs2_bitmap
*bi
= rgd
->rd_bits
+ x
;
669 memcpy(bi
->bi_clone
+ bi
->bi_offset
,
670 bi
->bi_bh
->b_data
+ bi
->bi_offset
, bi
->bi_len
);
673 spin_lock(&sdp
->sd_rindex_spin
);
674 rgd
->rd_free_clone
= rgd
->rd_rg
.rg_free
;
675 spin_unlock(&sdp
->sd_rindex_spin
);
679 * gfs2_alloc_get - get the struct gfs2_alloc structure for an inode
680 * @ip: the incore GFS2 inode structure
682 * Returns: the struct gfs2_alloc
685 struct gfs2_alloc
*gfs2_alloc_get(struct gfs2_inode
*ip
)
687 struct gfs2_alloc
*al
= &ip
->i_alloc
;
689 /* FIXME: Should assert that the correct locks are held here... */
690 memset(al
, 0, sizeof(*al
));
695 * try_rgrp_fit - See if a given reservation will fit in a given RG
697 * @al: the struct gfs2_alloc structure describing the reservation
699 * If there's room for the requested blocks to be allocated from the RG:
700 * Sets the $al_reserved_data field in @al.
701 * Sets the $al_reserved_meta field in @al.
702 * Sets the $al_rgd field in @al.
704 * Returns: 1 on success (it fits), 0 on failure (it doesn't fit)
707 static int try_rgrp_fit(struct gfs2_rgrpd
*rgd
, struct gfs2_alloc
*al
)
709 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
712 spin_lock(&sdp
->sd_rindex_spin
);
713 if (rgd
->rd_free_clone
>= al
->al_requested
) {
717 spin_unlock(&sdp
->sd_rindex_spin
);
723 * recent_rgrp_first - get first RG from "recent" list
724 * @sdp: The GFS2 superblock
725 * @rglast: address of the rgrp used last
727 * Returns: The first rgrp in the recent list
730 static struct gfs2_rgrpd
*recent_rgrp_first(struct gfs2_sbd
*sdp
,
733 struct gfs2_rgrpd
*rgd
= NULL
;
735 spin_lock(&sdp
->sd_rindex_spin
);
737 if (list_empty(&sdp
->sd_rindex_recent_list
))
743 list_for_each_entry(rgd
, &sdp
->sd_rindex_recent_list
, rd_recent
) {
744 if (rgd
->rd_ri
.ri_addr
== rglast
)
749 rgd
= list_entry(sdp
->sd_rindex_recent_list
.next
, struct gfs2_rgrpd
,
752 spin_unlock(&sdp
->sd_rindex_spin
);
757 * recent_rgrp_next - get next RG from "recent" list
758 * @cur_rgd: current rgrp
761 * Returns: The next rgrp in the recent list
764 static struct gfs2_rgrpd
*recent_rgrp_next(struct gfs2_rgrpd
*cur_rgd
,
767 struct gfs2_sbd
*sdp
= cur_rgd
->rd_sbd
;
768 struct list_head
*head
;
769 struct gfs2_rgrpd
*rgd
;
771 spin_lock(&sdp
->sd_rindex_spin
);
773 head
= &sdp
->sd_rindex_recent_list
;
775 list_for_each_entry(rgd
, head
, rd_recent
) {
776 if (rgd
== cur_rgd
) {
777 if (cur_rgd
->rd_recent
.next
!= head
)
778 rgd
= list_entry(cur_rgd
->rd_recent
.next
,
779 struct gfs2_rgrpd
, rd_recent
);
784 list_del(&cur_rgd
->rd_recent
);
791 if (!list_empty(head
))
792 rgd
= list_entry(head
->next
, struct gfs2_rgrpd
, rd_recent
);
795 spin_unlock(&sdp
->sd_rindex_spin
);
800 * recent_rgrp_add - add an RG to tail of "recent" list
801 * @new_rgd: The rgrp to add
805 static void recent_rgrp_add(struct gfs2_rgrpd
*new_rgd
)
807 struct gfs2_sbd
*sdp
= new_rgd
->rd_sbd
;
808 struct gfs2_rgrpd
*rgd
;
809 unsigned int count
= 0;
810 unsigned int max
= sdp
->sd_rgrps
/ gfs2_jindex_size(sdp
);
812 spin_lock(&sdp
->sd_rindex_spin
);
814 list_for_each_entry(rgd
, &sdp
->sd_rindex_recent_list
, rd_recent
) {
821 list_add_tail(&new_rgd
->rd_recent
, &sdp
->sd_rindex_recent_list
);
824 spin_unlock(&sdp
->sd_rindex_spin
);
828 * forward_rgrp_get - get an rgrp to try next from full list
829 * @sdp: The GFS2 superblock
831 * Returns: The rgrp to try next
834 static struct gfs2_rgrpd
*forward_rgrp_get(struct gfs2_sbd
*sdp
)
836 struct gfs2_rgrpd
*rgd
;
837 unsigned int journals
= gfs2_jindex_size(sdp
);
838 unsigned int rg
= 0, x
;
840 spin_lock(&sdp
->sd_rindex_spin
);
842 rgd
= sdp
->sd_rindex_forward
;
844 if (sdp
->sd_rgrps
>= journals
)
845 rg
= sdp
->sd_rgrps
* sdp
->sd_jdesc
->jd_jid
/ journals
;
847 for (x
= 0, rgd
= gfs2_rgrpd_get_first(sdp
); x
< rg
;
848 x
++, rgd
= gfs2_rgrpd_get_next(rgd
))
851 sdp
->sd_rindex_forward
= rgd
;
854 spin_unlock(&sdp
->sd_rindex_spin
);
860 * forward_rgrp_set - set the forward rgrp pointer
861 * @sdp: the filesystem
862 * @rgd: The new forward rgrp
866 static void forward_rgrp_set(struct gfs2_sbd
*sdp
, struct gfs2_rgrpd
*rgd
)
868 spin_lock(&sdp
->sd_rindex_spin
);
869 sdp
->sd_rindex_forward
= rgd
;
870 spin_unlock(&sdp
->sd_rindex_spin
);
874 * get_local_rgrp - Choose and lock a rgrp for allocation
875 * @ip: the inode to reserve space for
876 * @rgp: the chosen and locked rgrp
878 * Try to acquire rgrp in way which avoids contending with others.
883 static int get_local_rgrp(struct gfs2_inode
*ip
)
885 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
886 struct gfs2_rgrpd
*rgd
, *begin
= NULL
;
887 struct gfs2_alloc
*al
= &ip
->i_alloc
;
888 int flags
= LM_FLAG_TRY
;
893 /* Try recently successful rgrps */
895 rgd
= recent_rgrp_first(sdp
, ip
->i_last_rg_alloc
);
898 error
= gfs2_glock_nq_init(rgd
->rd_gl
, LM_ST_EXCLUSIVE
,
899 LM_FLAG_TRY
, &al
->al_rgd_gh
);
902 if (try_rgrp_fit(rgd
, al
))
904 gfs2_glock_dq_uninit(&al
->al_rgd_gh
);
905 rgd
= recent_rgrp_next(rgd
, 1);
909 rgd
= recent_rgrp_next(rgd
, 0);
917 /* Go through full list of rgrps */
919 begin
= rgd
= forward_rgrp_get(sdp
);
922 error
= gfs2_glock_nq_init(rgd
->rd_gl
, LM_ST_EXCLUSIVE
, flags
,
926 if (try_rgrp_fit(rgd
, al
))
928 gfs2_glock_dq_uninit(&al
->al_rgd_gh
);
939 rgd
= gfs2_rgrpd_get_next(rgd
);
941 rgd
= gfs2_rgrpd_get_first(sdp
);
944 if (++loops
>= 2 || !skipped
)
951 ip
->i_last_rg_alloc
= rgd
->rd_ri
.ri_addr
;
954 recent_rgrp_add(rgd
);
955 rgd
= gfs2_rgrpd_get_next(rgd
);
957 rgd
= gfs2_rgrpd_get_first(sdp
);
958 forward_rgrp_set(sdp
, rgd
);
965 * gfs2_inplace_reserve_i - Reserve space in the filesystem
966 * @ip: the inode to reserve space for
971 int gfs2_inplace_reserve_i(struct gfs2_inode
*ip
, char *file
, unsigned int line
)
973 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
974 struct gfs2_alloc
*al
= &ip
->i_alloc
;
977 if (gfs2_assert_warn(sdp
, al
->al_requested
))
980 error
= gfs2_rindex_hold(sdp
, &al
->al_ri_gh
);
984 error
= get_local_rgrp(ip
);
986 gfs2_glock_dq_uninit(&al
->al_ri_gh
);
997 * gfs2_inplace_release - release an inplace reservation
998 * @ip: the inode the reservation was taken out on
1000 * Release a reservation made by gfs2_inplace_reserve().
1003 void gfs2_inplace_release(struct gfs2_inode
*ip
)
1005 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
1006 struct gfs2_alloc
*al
= &ip
->i_alloc
;
1008 if (gfs2_assert_warn(sdp
, al
->al_alloced
<= al
->al_requested
) == -1)
1009 fs_warn(sdp
, "al_alloced = %u, al_requested = %u "
1010 "al_file = %s, al_line = %u\n",
1011 al
->al_alloced
, al
->al_requested
, al
->al_file
,
1015 gfs2_glock_dq_uninit(&al
->al_rgd_gh
);
1016 gfs2_glock_dq_uninit(&al
->al_ri_gh
);
1020 * gfs2_get_block_type - Check a block in a RG is of given type
1021 * @rgd: the resource group holding the block
1022 * @block: the block number
1024 * Returns: The block type (GFS2_BLKST_*)
1027 unsigned char gfs2_get_block_type(struct gfs2_rgrpd
*rgd
, u64 block
)
1029 struct gfs2_bitmap
*bi
= NULL
;
1030 u32 length
, rgrp_block
, buf_block
;
1034 length
= rgd
->rd_ri
.ri_length
;
1035 rgrp_block
= block
- rgd
->rd_ri
.ri_data0
;
1037 for (buf
= 0; buf
< length
; buf
++) {
1038 bi
= rgd
->rd_bits
+ buf
;
1039 if (rgrp_block
< (bi
->bi_start
+ bi
->bi_len
) * GFS2_NBBY
)
1043 gfs2_assert(rgd
->rd_sbd
, buf
< length
);
1044 buf_block
= rgrp_block
- bi
->bi_start
* GFS2_NBBY
;
1046 type
= gfs2_testbit(rgd
, bi
->bi_bh
->b_data
+ bi
->bi_offset
,
1047 bi
->bi_len
, buf_block
);
1053 * rgblk_search - find a block in @old_state, change allocation
1054 * state to @new_state
1055 * @rgd: the resource group descriptor
1056 * @goal: the goal block within the RG (start here to search for avail block)
1057 * @old_state: GFS2_BLKST_XXX the before-allocation state to find
1058 * @new_state: GFS2_BLKST_XXX the after-allocation block state
1060 * Walk rgrp's bitmap to find bits that represent a block in @old_state.
1061 * Add the found bitmap buffer to the transaction.
1062 * Set the found bits to @new_state to change block's allocation state.
1064 * This function never fails, because we wouldn't call it unless we
1065 * know (from reservation results, etc.) that a block is available.
1067 * Scope of @goal and returned block is just within rgrp, not the whole
1070 * Returns: the block number allocated
1073 static u32
rgblk_search(struct gfs2_rgrpd
*rgd
, u32 goal
,
1074 unsigned char old_state
, unsigned char new_state
)
1076 struct gfs2_bitmap
*bi
= NULL
;
1077 u32 length
= rgd
->rd_ri
.ri_length
;
1079 unsigned int buf
, x
;
1081 /* Find bitmap block that contains bits for goal block */
1082 for (buf
= 0; buf
< length
; buf
++) {
1083 bi
= rgd
->rd_bits
+ buf
;
1084 if (goal
< (bi
->bi_start
+ bi
->bi_len
) * GFS2_NBBY
)
1088 gfs2_assert(rgd
->rd_sbd
, buf
< length
);
1090 /* Convert scope of "goal" from rgrp-wide to within found bit block */
1091 goal
-= bi
->bi_start
* GFS2_NBBY
;
1093 /* Search (up to entire) bitmap in this rgrp for allocatable block.
1094 "x <= length", instead of "x < length", because we typically start
1095 the search in the middle of a bit block, but if we can't find an
1096 allocatable block anywhere else, we want to be able wrap around and
1097 search in the first part of our first-searched bit block. */
1098 for (x
= 0; x
<= length
; x
++) {
1100 blk
= gfs2_bitfit(rgd
, bi
->bi_clone
+ bi
->bi_offset
,
1101 bi
->bi_len
, goal
, old_state
);
1103 blk
= gfs2_bitfit(rgd
,
1104 bi
->bi_bh
->b_data
+ bi
->bi_offset
,
1105 bi
->bi_len
, goal
, old_state
);
1106 if (blk
!= BFITNOENT
)
1109 /* Try next bitmap block (wrap back to rgrp header if at end) */
1110 buf
= (buf
+ 1) % length
;
1111 bi
= rgd
->rd_bits
+ buf
;
1115 if (gfs2_assert_withdraw(rgd
->rd_sbd
, x
<= length
))
1118 gfs2_trans_add_bh(rgd
->rd_gl
, bi
->bi_bh
, 1);
1119 gfs2_setbit(rgd
, bi
->bi_bh
->b_data
+ bi
->bi_offset
,
1120 bi
->bi_len
, blk
, new_state
);
1122 gfs2_setbit(rgd
, bi
->bi_clone
+ bi
->bi_offset
,
1123 bi
->bi_len
, blk
, new_state
);
1125 return bi
->bi_start
* GFS2_NBBY
+ blk
;
1129 * rgblk_free - Change alloc state of given block(s)
1130 * @sdp: the filesystem
1131 * @bstart: the start of a run of blocks to free
1132 * @blen: the length of the block run (all must lie within ONE RG!)
1133 * @new_state: GFS2_BLKST_XXX the after-allocation block state
1135 * Returns: Resource group containing the block(s)
1138 static struct gfs2_rgrpd
*rgblk_free(struct gfs2_sbd
*sdp
, u64 bstart
,
1139 u32 blen
, unsigned char new_state
)
1141 struct gfs2_rgrpd
*rgd
;
1142 struct gfs2_bitmap
*bi
= NULL
;
1143 u32 length
, rgrp_blk
, buf_blk
;
1146 rgd
= gfs2_blk2rgrpd(sdp
, bstart
);
1148 if (gfs2_consist(sdp
))
1149 fs_err(sdp
, "block = %llu\n", (unsigned long long)bstart
);
1153 length
= rgd
->rd_ri
.ri_length
;
1155 rgrp_blk
= bstart
- rgd
->rd_ri
.ri_data0
;
1158 for (buf
= 0; buf
< length
; buf
++) {
1159 bi
= rgd
->rd_bits
+ buf
;
1160 if (rgrp_blk
< (bi
->bi_start
+ bi
->bi_len
) * GFS2_NBBY
)
1164 gfs2_assert(rgd
->rd_sbd
, buf
< length
);
1166 buf_blk
= rgrp_blk
- bi
->bi_start
* GFS2_NBBY
;
1169 if (!bi
->bi_clone
) {
1170 bi
->bi_clone
= kmalloc(bi
->bi_bh
->b_size
,
1171 GFP_NOFS
| __GFP_NOFAIL
);
1172 memcpy(bi
->bi_clone
+ bi
->bi_offset
,
1173 bi
->bi_bh
->b_data
+ bi
->bi_offset
,
1176 gfs2_trans_add_bh(rgd
->rd_gl
, bi
->bi_bh
, 1);
1177 gfs2_setbit(rgd
, bi
->bi_bh
->b_data
+ bi
->bi_offset
,
1178 bi
->bi_len
, buf_blk
, new_state
);
1185 * gfs2_alloc_data - Allocate a data block
1186 * @ip: the inode to allocate the data block for
1188 * Returns: the allocated block
1191 u64
gfs2_alloc_data(struct gfs2_inode
*ip
)
1193 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
1194 struct gfs2_alloc
*al
= &ip
->i_alloc
;
1195 struct gfs2_rgrpd
*rgd
= al
->al_rgd
;
1199 if (rgrp_contains_block(&rgd
->rd_ri
, ip
->i_di
.di_goal_data
))
1200 goal
= ip
->i_di
.di_goal_data
- rgd
->rd_ri
.ri_data0
;
1202 goal
= rgd
->rd_last_alloc_data
;
1204 blk
= rgblk_search(rgd
, goal
, GFS2_BLKST_FREE
, GFS2_BLKST_USED
);
1205 rgd
->rd_last_alloc_data
= blk
;
1207 block
= rgd
->rd_ri
.ri_data0
+ blk
;
1208 ip
->i_di
.di_goal_data
= block
;
1210 gfs2_assert_withdraw(sdp
, rgd
->rd_rg
.rg_free
);
1211 rgd
->rd_rg
.rg_free
--;
1213 gfs2_trans_add_bh(rgd
->rd_gl
, rgd
->rd_bits
[0].bi_bh
, 1);
1214 gfs2_rgrp_out(&rgd
->rd_rg
, rgd
->rd_bits
[0].bi_bh
->b_data
);
1218 gfs2_statfs_change(sdp
, 0, -1, 0);
1219 gfs2_quota_change(ip
, +1, ip
->i_inode
.i_uid
, ip
->i_inode
.i_gid
);
1221 spin_lock(&sdp
->sd_rindex_spin
);
1222 rgd
->rd_free_clone
--;
1223 spin_unlock(&sdp
->sd_rindex_spin
);
1229 * gfs2_alloc_meta - Allocate a metadata block
1230 * @ip: the inode to allocate the metadata block for
1232 * Returns: the allocated block
1235 u64
gfs2_alloc_meta(struct gfs2_inode
*ip
)
1237 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
1238 struct gfs2_alloc
*al
= &ip
->i_alloc
;
1239 struct gfs2_rgrpd
*rgd
= al
->al_rgd
;
1243 if (rgrp_contains_block(&rgd
->rd_ri
, ip
->i_di
.di_goal_meta
))
1244 goal
= ip
->i_di
.di_goal_meta
- rgd
->rd_ri
.ri_data0
;
1246 goal
= rgd
->rd_last_alloc_meta
;
1248 blk
= rgblk_search(rgd
, goal
, GFS2_BLKST_FREE
, GFS2_BLKST_USED
);
1249 rgd
->rd_last_alloc_meta
= blk
;
1251 block
= rgd
->rd_ri
.ri_data0
+ blk
;
1252 ip
->i_di
.di_goal_meta
= block
;
1254 gfs2_assert_withdraw(sdp
, rgd
->rd_rg
.rg_free
);
1255 rgd
->rd_rg
.rg_free
--;
1257 gfs2_trans_add_bh(rgd
->rd_gl
, rgd
->rd_bits
[0].bi_bh
, 1);
1258 gfs2_rgrp_out(&rgd
->rd_rg
, rgd
->rd_bits
[0].bi_bh
->b_data
);
1262 gfs2_statfs_change(sdp
, 0, -1, 0);
1263 gfs2_quota_change(ip
, +1, ip
->i_inode
.i_uid
, ip
->i_inode
.i_gid
);
1264 gfs2_trans_add_unrevoke(sdp
, block
);
1266 spin_lock(&sdp
->sd_rindex_spin
);
1267 rgd
->rd_free_clone
--;
1268 spin_unlock(&sdp
->sd_rindex_spin
);
1274 * gfs2_alloc_di - Allocate a dinode
1275 * @dip: the directory that the inode is going in
1277 * Returns: the block allocated
1280 u64
gfs2_alloc_di(struct gfs2_inode
*dip
, u64
*generation
)
1282 struct gfs2_sbd
*sdp
= GFS2_SB(&dip
->i_inode
);
1283 struct gfs2_alloc
*al
= &dip
->i_alloc
;
1284 struct gfs2_rgrpd
*rgd
= al
->al_rgd
;
1288 blk
= rgblk_search(rgd
, rgd
->rd_last_alloc_meta
,
1289 GFS2_BLKST_FREE
, GFS2_BLKST_DINODE
);
1291 rgd
->rd_last_alloc_meta
= blk
;
1293 block
= rgd
->rd_ri
.ri_data0
+ blk
;
1295 gfs2_assert_withdraw(sdp
, rgd
->rd_rg
.rg_free
);
1296 rgd
->rd_rg
.rg_free
--;
1297 rgd
->rd_rg
.rg_dinodes
++;
1298 *generation
= rgd
->rd_rg
.rg_igeneration
++;
1299 gfs2_trans_add_bh(rgd
->rd_gl
, rgd
->rd_bits
[0].bi_bh
, 1);
1300 gfs2_rgrp_out(&rgd
->rd_rg
, rgd
->rd_bits
[0].bi_bh
->b_data
);
1304 gfs2_statfs_change(sdp
, 0, -1, +1);
1305 gfs2_trans_add_unrevoke(sdp
, block
);
1307 spin_lock(&sdp
->sd_rindex_spin
);
1308 rgd
->rd_free_clone
--;
1309 spin_unlock(&sdp
->sd_rindex_spin
);
1315 * gfs2_free_data - free a contiguous run of data block(s)
1316 * @ip: the inode these blocks are being freed from
1317 * @bstart: first block of a run of contiguous blocks
1318 * @blen: the length of the block run
1322 void gfs2_free_data(struct gfs2_inode
*ip
, u64 bstart
, u32 blen
)
1324 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
1325 struct gfs2_rgrpd
*rgd
;
1327 rgd
= rgblk_free(sdp
, bstart
, blen
, GFS2_BLKST_FREE
);
1331 rgd
->rd_rg
.rg_free
+= blen
;
1333 gfs2_trans_add_bh(rgd
->rd_gl
, rgd
->rd_bits
[0].bi_bh
, 1);
1334 gfs2_rgrp_out(&rgd
->rd_rg
, rgd
->rd_bits
[0].bi_bh
->b_data
);
1336 gfs2_trans_add_rg(rgd
);
1338 gfs2_statfs_change(sdp
, 0, +blen
, 0);
1339 gfs2_quota_change(ip
, -(s64
)blen
, ip
->i_inode
.i_uid
, ip
->i_inode
.i_gid
);
1343 * gfs2_free_meta - free a contiguous run of data block(s)
1344 * @ip: the inode these blocks are being freed from
1345 * @bstart: first block of a run of contiguous blocks
1346 * @blen: the length of the block run
1350 void gfs2_free_meta(struct gfs2_inode
*ip
, u64 bstart
, u32 blen
)
1352 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
1353 struct gfs2_rgrpd
*rgd
;
1355 rgd
= rgblk_free(sdp
, bstart
, blen
, GFS2_BLKST_FREE
);
1359 rgd
->rd_rg
.rg_free
+= blen
;
1361 gfs2_trans_add_bh(rgd
->rd_gl
, rgd
->rd_bits
[0].bi_bh
, 1);
1362 gfs2_rgrp_out(&rgd
->rd_rg
, rgd
->rd_bits
[0].bi_bh
->b_data
);
1364 gfs2_trans_add_rg(rgd
);
1366 gfs2_statfs_change(sdp
, 0, +blen
, 0);
1367 gfs2_quota_change(ip
, -(s64
)blen
, ip
->i_inode
.i_uid
, ip
->i_inode
.i_gid
);
1368 gfs2_meta_wipe(ip
, bstart
, blen
);
1371 void gfs2_unlink_di(struct inode
*inode
)
1373 struct gfs2_inode
*ip
= GFS2_I(inode
);
1374 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
1375 struct gfs2_rgrpd
*rgd
;
1376 u64 blkno
= ip
->i_num
.no_addr
;
1378 rgd
= rgblk_free(sdp
, blkno
, 1, GFS2_BLKST_UNLINKED
);
1381 gfs2_trans_add_bh(rgd
->rd_gl
, rgd
->rd_bits
[0].bi_bh
, 1);
1382 gfs2_rgrp_out(&rgd
->rd_rg
, rgd
->rd_bits
[0].bi_bh
->b_data
);
1383 gfs2_trans_add_rg(rgd
);
1386 static void gfs2_free_uninit_di(struct gfs2_rgrpd
*rgd
, u64 blkno
)
1388 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
1389 struct gfs2_rgrpd
*tmp_rgd
;
1391 tmp_rgd
= rgblk_free(sdp
, blkno
, 1, GFS2_BLKST_FREE
);
1394 gfs2_assert_withdraw(sdp
, rgd
== tmp_rgd
);
1396 if (!rgd
->rd_rg
.rg_dinodes
)
1397 gfs2_consist_rgrpd(rgd
);
1398 rgd
->rd_rg
.rg_dinodes
--;
1399 rgd
->rd_rg
.rg_free
++;
1401 gfs2_trans_add_bh(rgd
->rd_gl
, rgd
->rd_bits
[0].bi_bh
, 1);
1402 gfs2_rgrp_out(&rgd
->rd_rg
, rgd
->rd_bits
[0].bi_bh
->b_data
);
1404 gfs2_statfs_change(sdp
, 0, +1, -1);
1405 gfs2_trans_add_rg(rgd
);
1409 void gfs2_free_di(struct gfs2_rgrpd
*rgd
, struct gfs2_inode
*ip
)
1411 gfs2_free_uninit_di(rgd
, ip
->i_num
.no_addr
);
1412 gfs2_quota_change(ip
, -1, ip
->i_inode
.i_uid
, ip
->i_inode
.i_gid
);
1413 gfs2_meta_wipe(ip
, ip
->i_num
.no_addr
, 1);
1417 * gfs2_rlist_add - add a RG to a list of RGs
1418 * @sdp: the filesystem
1419 * @rlist: the list of resource groups
1422 * Figure out what RG a block belongs to and add that RG to the list
1424 * FIXME: Don't use NOFAIL
1428 void gfs2_rlist_add(struct gfs2_sbd
*sdp
, struct gfs2_rgrp_list
*rlist
,
1431 struct gfs2_rgrpd
*rgd
;
1432 struct gfs2_rgrpd
**tmp
;
1433 unsigned int new_space
;
1436 if (gfs2_assert_warn(sdp
, !rlist
->rl_ghs
))
1439 rgd
= gfs2_blk2rgrpd(sdp
, block
);
1441 if (gfs2_consist(sdp
))
1442 fs_err(sdp
, "block = %llu\n", (unsigned long long)block
);
1446 for (x
= 0; x
< rlist
->rl_rgrps
; x
++)
1447 if (rlist
->rl_rgd
[x
] == rgd
)
1450 if (rlist
->rl_rgrps
== rlist
->rl_space
) {
1451 new_space
= rlist
->rl_space
+ 10;
1453 tmp
= kcalloc(new_space
, sizeof(struct gfs2_rgrpd
*),
1454 GFP_NOFS
| __GFP_NOFAIL
);
1456 if (rlist
->rl_rgd
) {
1457 memcpy(tmp
, rlist
->rl_rgd
,
1458 rlist
->rl_space
* sizeof(struct gfs2_rgrpd
*));
1459 kfree(rlist
->rl_rgd
);
1462 rlist
->rl_space
= new_space
;
1463 rlist
->rl_rgd
= tmp
;
1466 rlist
->rl_rgd
[rlist
->rl_rgrps
++] = rgd
;
1470 * gfs2_rlist_alloc - all RGs have been added to the rlist, now allocate
1471 * and initialize an array of glock holders for them
1472 * @rlist: the list of resource groups
1473 * @state: the lock state to acquire the RG lock in
1474 * @flags: the modifier flags for the holder structures
1476 * FIXME: Don't use NOFAIL
1480 void gfs2_rlist_alloc(struct gfs2_rgrp_list
*rlist
, unsigned int state
,
1485 rlist
->rl_ghs
= kcalloc(rlist
->rl_rgrps
, sizeof(struct gfs2_holder
),
1486 GFP_NOFS
| __GFP_NOFAIL
);
1487 for (x
= 0; x
< rlist
->rl_rgrps
; x
++)
1488 gfs2_holder_init(rlist
->rl_rgd
[x
]->rd_gl
,
1494 * gfs2_rlist_free - free a resource group list
1495 * @list: the list of resource groups
1499 void gfs2_rlist_free(struct gfs2_rgrp_list
*rlist
)
1503 kfree(rlist
->rl_rgd
);
1505 if (rlist
->rl_ghs
) {
1506 for (x
= 0; x
< rlist
->rl_rgrps
; x
++)
1507 gfs2_holder_uninit(&rlist
->rl_ghs
[x
]);
1508 kfree(rlist
->rl_ghs
);