2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2008 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/prefetch.h>
17 #include <linux/blkdev.h>
18 #include <linux/rbtree.h>
19 #include <linux/random.h>
34 #include "trace_gfs2.h"
36 #define BFITNOENT ((u32)~0)
37 #define NO_BLOCK ((u64)~0)
39 #if BITS_PER_LONG == 32
40 #define LBITMASK (0x55555555UL)
41 #define LBITSKIP55 (0x55555555UL)
42 #define LBITSKIP00 (0x00000000UL)
44 #define LBITMASK (0x5555555555555555UL)
45 #define LBITSKIP55 (0x5555555555555555UL)
46 #define LBITSKIP00 (0x0000000000000000UL)
50 * These routines are used by the resource group routines (rgrp.c)
51 * to keep track of block allocation. Each block is represented by two
52 * bits. So, each byte represents GFS2_NBBY (i.e. 4) blocks.
55 * 1 = Used (not metadata)
56 * 2 = Unlinked (still in use) inode
60 static const char valid_change
[16] = {
68 static int gfs2_rbm_find(struct gfs2_rbm
*rbm
, u8 state
, u32 minext
,
69 const struct gfs2_inode
*ip
, bool nowrap
);
73 * gfs2_setbit - Set a bit in the bitmaps
74 * @rbm: The position of the bit to set
75 * @do_clone: Also set the clone bitmap, if it exists
76 * @new_state: the new state of the block
80 static inline void gfs2_setbit(const struct gfs2_rbm
*rbm
, bool do_clone
,
81 unsigned char new_state
)
83 unsigned char *byte1
, *byte2
, *end
, cur_state
;
84 unsigned int buflen
= rbm
->bi
->bi_len
;
85 const unsigned int bit
= (rbm
->offset
% GFS2_NBBY
) * GFS2_BIT_SIZE
;
87 byte1
= rbm
->bi
->bi_bh
->b_data
+ rbm
->bi
->bi_offset
+ (rbm
->offset
/ GFS2_NBBY
);
88 end
= rbm
->bi
->bi_bh
->b_data
+ rbm
->bi
->bi_offset
+ buflen
;
92 cur_state
= (*byte1
>> bit
) & GFS2_BIT_MASK
;
94 if (unlikely(!valid_change
[new_state
* 4 + cur_state
])) {
95 printk(KERN_WARNING
"GFS2: buf_blk = 0x%x old_state=%d, "
96 "new_state=%d\n", rbm
->offset
, cur_state
, new_state
);
97 printk(KERN_WARNING
"GFS2: rgrp=0x%llx bi_start=0x%x\n",
98 (unsigned long long)rbm
->rgd
->rd_addr
,
100 printk(KERN_WARNING
"GFS2: bi_offset=0x%x bi_len=0x%x\n",
101 rbm
->bi
->bi_offset
, rbm
->bi
->bi_len
);
103 gfs2_consist_rgrpd(rbm
->rgd
);
106 *byte1
^= (cur_state
^ new_state
) << bit
;
108 if (do_clone
&& rbm
->bi
->bi_clone
) {
109 byte2
= rbm
->bi
->bi_clone
+ rbm
->bi
->bi_offset
+ (rbm
->offset
/ GFS2_NBBY
);
110 cur_state
= (*byte2
>> bit
) & GFS2_BIT_MASK
;
111 *byte2
^= (cur_state
^ new_state
) << bit
;
116 * gfs2_testbit - test a bit in the bitmaps
117 * @rbm: The bit to test
119 * Returns: The two bit block state of the requested bit
122 static inline u8
gfs2_testbit(const struct gfs2_rbm
*rbm
)
124 const u8
*buffer
= rbm
->bi
->bi_bh
->b_data
+ rbm
->bi
->bi_offset
;
128 byte
= buffer
+ (rbm
->offset
/ GFS2_NBBY
);
129 bit
= (rbm
->offset
% GFS2_NBBY
) * GFS2_BIT_SIZE
;
131 return (*byte
>> bit
) & GFS2_BIT_MASK
;
136 * @ptr: Pointer to bitmap data
137 * @mask: Mask to use (normally 0x55555.... but adjusted for search start)
138 * @state: The state we are searching for
140 * We xor the bitmap data with a patter which is the bitwise opposite
141 * of what we are looking for, this gives rise to a pattern of ones
142 * wherever there is a match. Since we have two bits per entry, we
143 * take this pattern, shift it down by one place and then and it with
144 * the original. All the even bit positions (0,2,4, etc) then represent
145 * successful matches, so we mask with 0x55555..... to remove the unwanted
148 * This allows searching of a whole u64 at once (32 blocks) with a
149 * single test (on 64 bit arches).
152 static inline u64
gfs2_bit_search(const __le64
*ptr
, u64 mask
, u8 state
)
155 static const u64 search
[] = {
156 [0] = 0xffffffffffffffffULL
,
157 [1] = 0xaaaaaaaaaaaaaaaaULL
,
158 [2] = 0x5555555555555555ULL
,
159 [3] = 0x0000000000000000ULL
,
161 tmp
= le64_to_cpu(*ptr
) ^ search
[state
];
168 * rs_cmp - multi-block reservation range compare
169 * @blk: absolute file system block number of the new reservation
170 * @len: number of blocks in the new reservation
171 * @rs: existing reservation to compare against
173 * returns: 1 if the block range is beyond the reach of the reservation
174 * -1 if the block range is before the start of the reservation
175 * 0 if the block range overlaps with the reservation
177 static inline int rs_cmp(u64 blk
, u32 len
, struct gfs2_blkreserv
*rs
)
179 u64 startblk
= gfs2_rbm_to_block(&rs
->rs_rbm
);
181 if (blk
>= startblk
+ rs
->rs_free
)
183 if (blk
+ len
- 1 < startblk
)
189 * gfs2_bitfit - Search an rgrp's bitmap buffer to find a bit-pair representing
190 * a block in a given allocation state.
191 * @buf: the buffer that holds the bitmaps
192 * @len: the length (in bytes) of the buffer
193 * @goal: start search at this block's bit-pair (within @buffer)
194 * @state: GFS2_BLKST_XXX the state of the block we're looking for.
196 * Scope of @goal and returned block number is only within this bitmap buffer,
197 * not entire rgrp or filesystem. @buffer will be offset from the actual
198 * beginning of a bitmap block buffer, skipping any header structures, but
199 * headers are always a multiple of 64 bits long so that the buffer is
200 * always aligned to a 64 bit boundary.
202 * The size of the buffer is in bytes, but is it assumed that it is
203 * always ok to read a complete multiple of 64 bits at the end
204 * of the block in case the end is no aligned to a natural boundary.
206 * Return: the block number (bitmap buffer scope) that was found
209 static u32
gfs2_bitfit(const u8
*buf
, const unsigned int len
,
212 u32 spoint
= (goal
<< 1) & ((8*sizeof(u64
)) - 1);
213 const __le64
*ptr
= ((__le64
*)buf
) + (goal
>> 5);
214 const __le64
*end
= (__le64
*)(buf
+ ALIGN(len
, sizeof(u64
)));
216 u64 mask
= 0x5555555555555555ULL
;
219 /* Mask off bits we don't care about at the start of the search */
221 tmp
= gfs2_bit_search(ptr
, mask
, state
);
223 while(tmp
== 0 && ptr
< end
) {
224 tmp
= gfs2_bit_search(ptr
, 0x5555555555555555ULL
, state
);
227 /* Mask off any bits which are more than len bytes from the start */
228 if (ptr
== end
&& (len
& (sizeof(u64
) - 1)))
229 tmp
&= (((u64
)~0) >> (64 - 8*(len
& (sizeof(u64
) - 1))));
230 /* Didn't find anything, so return */
235 bit
/= 2; /* two bits per entry in the bitmap */
236 return (((const unsigned char *)ptr
- buf
) * GFS2_NBBY
) + bit
;
240 * gfs2_rbm_from_block - Set the rbm based upon rgd and block number
241 * @rbm: The rbm with rgd already set correctly
242 * @block: The block number (filesystem relative)
244 * This sets the bi and offset members of an rbm based on a
245 * resource group and a filesystem relative block number. The
246 * resource group must be set in the rbm on entry, the bi and
247 * offset members will be set by this function.
249 * Returns: 0 on success, or an error code
252 static int gfs2_rbm_from_block(struct gfs2_rbm
*rbm
, u64 block
)
254 u64 rblock
= block
- rbm
->rgd
->rd_data0
;
257 if (WARN_ON_ONCE(rblock
> UINT_MAX
))
259 if (block
>= rbm
->rgd
->rd_data0
+ rbm
->rgd
->rd_data
)
262 rbm
->bi
= rbm
->rgd
->rd_bits
;
263 rbm
->offset
= (u32
)(rblock
);
264 /* Check if the block is within the first block */
265 if (rbm
->offset
< rbm
->bi
->bi_blocks
)
268 /* Adjust for the size diff between gfs2_meta_header and gfs2_rgrp */
269 rbm
->offset
+= (sizeof(struct gfs2_rgrp
) -
270 sizeof(struct gfs2_meta_header
)) * GFS2_NBBY
;
271 x
= rbm
->offset
/ rbm
->rgd
->rd_sbd
->sd_blocks_per_bitmap
;
272 rbm
->offset
-= x
* rbm
->rgd
->rd_sbd
->sd_blocks_per_bitmap
;
278 * gfs2_unaligned_extlen - Look for free blocks which are not byte aligned
279 * @rbm: Position to search (value/result)
280 * @n_unaligned: Number of unaligned blocks to check
281 * @len: Decremented for each block found (terminate on zero)
283 * Returns: true if a non-free block is encountered
286 static bool gfs2_unaligned_extlen(struct gfs2_rbm
*rbm
, u32 n_unaligned
, u32
*len
)
292 for (n
= 0; n
< n_unaligned
; n
++) {
293 res
= gfs2_testbit(rbm
);
294 if (res
!= GFS2_BLKST_FREE
)
299 block
= gfs2_rbm_to_block(rbm
);
300 if (gfs2_rbm_from_block(rbm
, block
+ 1))
308 * gfs2_free_extlen - Return extent length of free blocks
309 * @rbm: Starting position
310 * @len: Max length to check
312 * Starting at the block specified by the rbm, see how many free blocks
313 * there are, not reading more than len blocks ahead. This can be done
314 * using memchr_inv when the blocks are byte aligned, but has to be done
315 * on a block by block basis in case of unaligned blocks. Also this
316 * function can cope with bitmap boundaries (although it must stop on
317 * a resource group boundary)
319 * Returns: Number of free blocks in the extent
322 static u32
gfs2_free_extlen(const struct gfs2_rbm
*rrbm
, u32 len
)
324 struct gfs2_rbm rbm
= *rrbm
;
325 u32 n_unaligned
= rbm
.offset
& 3;
329 u8
*ptr
, *start
, *end
;
333 gfs2_unaligned_extlen(&rbm
, 4 - n_unaligned
, &len
))
336 n_unaligned
= len
& 3;
337 /* Start is now byte aligned */
339 start
= rbm
.bi
->bi_bh
->b_data
;
340 if (rbm
.bi
->bi_clone
)
341 start
= rbm
.bi
->bi_clone
;
342 end
= start
+ rbm
.bi
->bi_bh
->b_size
;
343 start
+= rbm
.bi
->bi_offset
;
344 BUG_ON(rbm
.offset
& 3);
345 start
+= (rbm
.offset
/ GFS2_NBBY
);
346 bytes
= min_t(u32
, len
/ GFS2_NBBY
, (end
- start
));
347 ptr
= memchr_inv(start
, 0, bytes
);
348 chunk_size
= ((ptr
== NULL
) ? bytes
: (ptr
- start
));
349 chunk_size
*= GFS2_NBBY
;
350 BUG_ON(len
< chunk_size
);
352 block
= gfs2_rbm_to_block(&rbm
);
353 if (gfs2_rbm_from_block(&rbm
, block
+ chunk_size
)) {
361 n_unaligned
= len
& 3;
364 /* Deal with any bits left over at the end */
366 gfs2_unaligned_extlen(&rbm
, n_unaligned
, &len
);
372 * gfs2_bitcount - count the number of bits in a certain state
373 * @rgd: the resource group descriptor
374 * @buffer: the buffer that holds the bitmaps
375 * @buflen: the length (in bytes) of the buffer
376 * @state: the state of the block we're looking for
378 * Returns: The number of bits
381 static u32
gfs2_bitcount(struct gfs2_rgrpd
*rgd
, const u8
*buffer
,
382 unsigned int buflen
, u8 state
)
384 const u8
*byte
= buffer
;
385 const u8
*end
= buffer
+ buflen
;
386 const u8 state1
= state
<< 2;
387 const u8 state2
= state
<< 4;
388 const u8 state3
= state
<< 6;
391 for (; byte
< end
; byte
++) {
392 if (((*byte
) & 0x03) == state
)
394 if (((*byte
) & 0x0C) == state1
)
396 if (((*byte
) & 0x30) == state2
)
398 if (((*byte
) & 0xC0) == state3
)
406 * gfs2_rgrp_verify - Verify that a resource group is consistent
411 void gfs2_rgrp_verify(struct gfs2_rgrpd
*rgd
)
413 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
414 struct gfs2_bitmap
*bi
= NULL
;
415 u32 length
= rgd
->rd_length
;
419 memset(count
, 0, 4 * sizeof(u32
));
421 /* Count # blocks in each of 4 possible allocation states */
422 for (buf
= 0; buf
< length
; buf
++) {
423 bi
= rgd
->rd_bits
+ buf
;
424 for (x
= 0; x
< 4; x
++)
425 count
[x
] += gfs2_bitcount(rgd
,
431 if (count
[0] != rgd
->rd_free
) {
432 if (gfs2_consist_rgrpd(rgd
))
433 fs_err(sdp
, "free data mismatch: %u != %u\n",
434 count
[0], rgd
->rd_free
);
438 tmp
= rgd
->rd_data
- rgd
->rd_free
- rgd
->rd_dinodes
;
439 if (count
[1] != tmp
) {
440 if (gfs2_consist_rgrpd(rgd
))
441 fs_err(sdp
, "used data mismatch: %u != %u\n",
446 if (count
[2] + count
[3] != rgd
->rd_dinodes
) {
447 if (gfs2_consist_rgrpd(rgd
))
448 fs_err(sdp
, "used metadata mismatch: %u != %u\n",
449 count
[2] + count
[3], rgd
->rd_dinodes
);
454 static inline int rgrp_contains_block(struct gfs2_rgrpd
*rgd
, u64 block
)
456 u64 first
= rgd
->rd_data0
;
457 u64 last
= first
+ rgd
->rd_data
;
458 return first
<= block
&& block
< last
;
462 * gfs2_blk2rgrpd - Find resource group for a given data/meta block number
463 * @sdp: The GFS2 superblock
464 * @blk: The data block number
465 * @exact: True if this needs to be an exact match
467 * Returns: The resource group, or NULL if not found
470 struct gfs2_rgrpd
*gfs2_blk2rgrpd(struct gfs2_sbd
*sdp
, u64 blk
, bool exact
)
472 struct rb_node
*n
, *next
;
473 struct gfs2_rgrpd
*cur
;
475 spin_lock(&sdp
->sd_rindex_spin
);
476 n
= sdp
->sd_rindex_tree
.rb_node
;
478 cur
= rb_entry(n
, struct gfs2_rgrpd
, rd_node
);
480 if (blk
< cur
->rd_addr
)
482 else if (blk
>= cur
->rd_data0
+ cur
->rd_data
)
485 spin_unlock(&sdp
->sd_rindex_spin
);
487 if (blk
< cur
->rd_addr
)
489 if (blk
>= cur
->rd_data0
+ cur
->rd_data
)
496 spin_unlock(&sdp
->sd_rindex_spin
);
502 * gfs2_rgrpd_get_first - get the first Resource Group in the filesystem
503 * @sdp: The GFS2 superblock
505 * Returns: The first rgrp in the filesystem
508 struct gfs2_rgrpd
*gfs2_rgrpd_get_first(struct gfs2_sbd
*sdp
)
510 const struct rb_node
*n
;
511 struct gfs2_rgrpd
*rgd
;
513 spin_lock(&sdp
->sd_rindex_spin
);
514 n
= rb_first(&sdp
->sd_rindex_tree
);
515 rgd
= rb_entry(n
, struct gfs2_rgrpd
, rd_node
);
516 spin_unlock(&sdp
->sd_rindex_spin
);
522 * gfs2_rgrpd_get_next - get the next RG
523 * @rgd: the resource group descriptor
525 * Returns: The next rgrp
528 struct gfs2_rgrpd
*gfs2_rgrpd_get_next(struct gfs2_rgrpd
*rgd
)
530 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
531 const struct rb_node
*n
;
533 spin_lock(&sdp
->sd_rindex_spin
);
534 n
= rb_next(&rgd
->rd_node
);
536 n
= rb_first(&sdp
->sd_rindex_tree
);
538 if (unlikely(&rgd
->rd_node
== n
)) {
539 spin_unlock(&sdp
->sd_rindex_spin
);
542 rgd
= rb_entry(n
, struct gfs2_rgrpd
, rd_node
);
543 spin_unlock(&sdp
->sd_rindex_spin
);
547 void gfs2_free_clones(struct gfs2_rgrpd
*rgd
)
551 for (x
= 0; x
< rgd
->rd_length
; x
++) {
552 struct gfs2_bitmap
*bi
= rgd
->rd_bits
+ x
;
559 * gfs2_rs_alloc - make sure we have a reservation assigned to the inode
560 * @ip: the inode for this reservation
562 int gfs2_rs_alloc(struct gfs2_inode
*ip
)
566 down_write(&ip
->i_rw_mutex
);
570 ip
->i_res
= kmem_cache_zalloc(gfs2_rsrv_cachep
, GFP_NOFS
);
576 RB_CLEAR_NODE(&ip
->i_res
->rs_node
);
578 up_write(&ip
->i_rw_mutex
);
582 static void dump_rs(struct seq_file
*seq
, const struct gfs2_blkreserv
*rs
)
584 gfs2_print_dbg(seq
, " B: n:%llu s:%llu b:%u f:%u\n",
585 (unsigned long long)rs
->rs_inum
,
586 (unsigned long long)gfs2_rbm_to_block(&rs
->rs_rbm
),
587 rs
->rs_rbm
.offset
, rs
->rs_free
);
591 * __rs_deltree - remove a multi-block reservation from the rgd tree
592 * @rs: The reservation to remove
595 static void __rs_deltree(struct gfs2_blkreserv
*rs
)
597 struct gfs2_rgrpd
*rgd
;
599 if (!gfs2_rs_active(rs
))
602 rgd
= rs
->rs_rbm
.rgd
;
603 trace_gfs2_rs(rs
, TRACE_RS_TREEDEL
);
604 rb_erase(&rs
->rs_node
, &rgd
->rd_rstree
);
605 RB_CLEAR_NODE(&rs
->rs_node
);
608 /* return reserved blocks to the rgrp */
609 BUG_ON(rs
->rs_rbm
.rgd
->rd_reserved
< rs
->rs_free
);
610 rs
->rs_rbm
.rgd
->rd_reserved
-= rs
->rs_free
;
612 clear_bit(GBF_FULL
, &rs
->rs_rbm
.bi
->bi_flags
);
613 smp_mb__after_clear_bit();
618 * gfs2_rs_deltree - remove a multi-block reservation from the rgd tree
619 * @rs: The reservation to remove
622 void gfs2_rs_deltree(struct gfs2_blkreserv
*rs
)
624 struct gfs2_rgrpd
*rgd
;
626 rgd
= rs
->rs_rbm
.rgd
;
628 spin_lock(&rgd
->rd_rsspin
);
630 spin_unlock(&rgd
->rd_rsspin
);
635 * gfs2_rs_delete - delete a multi-block reservation
636 * @ip: The inode for this reservation
639 void gfs2_rs_delete(struct gfs2_inode
*ip
)
641 struct inode
*inode
= &ip
->i_inode
;
643 down_write(&ip
->i_rw_mutex
);
644 if (ip
->i_res
&& atomic_read(&inode
->i_writecount
) <= 1) {
645 gfs2_rs_deltree(ip
->i_res
);
646 BUG_ON(ip
->i_res
->rs_free
);
647 kmem_cache_free(gfs2_rsrv_cachep
, ip
->i_res
);
650 up_write(&ip
->i_rw_mutex
);
654 * return_all_reservations - return all reserved blocks back to the rgrp.
655 * @rgd: the rgrp that needs its space back
657 * We previously reserved a bunch of blocks for allocation. Now we need to
658 * give them back. This leave the reservation structures in tact, but removes
659 * all of their corresponding "no-fly zones".
661 static void return_all_reservations(struct gfs2_rgrpd
*rgd
)
664 struct gfs2_blkreserv
*rs
;
666 spin_lock(&rgd
->rd_rsspin
);
667 while ((n
= rb_first(&rgd
->rd_rstree
))) {
668 rs
= rb_entry(n
, struct gfs2_blkreserv
, rs_node
);
671 spin_unlock(&rgd
->rd_rsspin
);
674 void gfs2_clear_rgrpd(struct gfs2_sbd
*sdp
)
677 struct gfs2_rgrpd
*rgd
;
678 struct gfs2_glock
*gl
;
680 while ((n
= rb_first(&sdp
->sd_rindex_tree
))) {
681 rgd
= rb_entry(n
, struct gfs2_rgrpd
, rd_node
);
684 rb_erase(n
, &sdp
->sd_rindex_tree
);
687 spin_lock(&gl
->gl_spin
);
688 gl
->gl_object
= NULL
;
689 spin_unlock(&gl
->gl_spin
);
690 gfs2_glock_add_to_lru(gl
);
694 gfs2_free_clones(rgd
);
696 return_all_reservations(rgd
);
697 kmem_cache_free(gfs2_rgrpd_cachep
, rgd
);
701 static void gfs2_rindex_print(const struct gfs2_rgrpd
*rgd
)
703 printk(KERN_INFO
" ri_addr = %llu\n", (unsigned long long)rgd
->rd_addr
);
704 printk(KERN_INFO
" ri_length = %u\n", rgd
->rd_length
);
705 printk(KERN_INFO
" ri_data0 = %llu\n", (unsigned long long)rgd
->rd_data0
);
706 printk(KERN_INFO
" ri_data = %u\n", rgd
->rd_data
);
707 printk(KERN_INFO
" ri_bitbytes = %u\n", rgd
->rd_bitbytes
);
711 * gfs2_compute_bitstructs - Compute the bitmap sizes
712 * @rgd: The resource group descriptor
714 * Calculates bitmap descriptors, one for each block that contains bitmap data
719 static int compute_bitstructs(struct gfs2_rgrpd
*rgd
)
721 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
722 struct gfs2_bitmap
*bi
;
723 u32 length
= rgd
->rd_length
; /* # blocks in hdr & bitmap */
724 u32 bytes_left
, bytes
;
730 rgd
->rd_bits
= kcalloc(length
, sizeof(struct gfs2_bitmap
), GFP_NOFS
);
734 bytes_left
= rgd
->rd_bitbytes
;
736 for (x
= 0; x
< length
; x
++) {
737 bi
= rgd
->rd_bits
+ x
;
740 /* small rgrp; bitmap stored completely in header block */
743 bi
->bi_offset
= sizeof(struct gfs2_rgrp
);
746 bi
->bi_blocks
= bytes
* GFS2_NBBY
;
749 bytes
= sdp
->sd_sb
.sb_bsize
- sizeof(struct gfs2_rgrp
);
750 bi
->bi_offset
= sizeof(struct gfs2_rgrp
);
753 bi
->bi_blocks
= bytes
* GFS2_NBBY
;
755 } else if (x
+ 1 == length
) {
757 bi
->bi_offset
= sizeof(struct gfs2_meta_header
);
758 bi
->bi_start
= rgd
->rd_bitbytes
- bytes_left
;
760 bi
->bi_blocks
= bytes
* GFS2_NBBY
;
763 bytes
= sdp
->sd_sb
.sb_bsize
-
764 sizeof(struct gfs2_meta_header
);
765 bi
->bi_offset
= sizeof(struct gfs2_meta_header
);
766 bi
->bi_start
= rgd
->rd_bitbytes
- bytes_left
;
768 bi
->bi_blocks
= bytes
* GFS2_NBBY
;
775 gfs2_consist_rgrpd(rgd
);
778 bi
= rgd
->rd_bits
+ (length
- 1);
779 if ((bi
->bi_start
+ bi
->bi_len
) * GFS2_NBBY
!= rgd
->rd_data
) {
780 if (gfs2_consist_rgrpd(rgd
)) {
781 gfs2_rindex_print(rgd
);
782 fs_err(sdp
, "start=%u len=%u offset=%u\n",
783 bi
->bi_start
, bi
->bi_len
, bi
->bi_offset
);
792 * gfs2_ri_total - Total up the file system space, according to the rindex.
793 * @sdp: the filesystem
796 u64
gfs2_ri_total(struct gfs2_sbd
*sdp
)
799 struct inode
*inode
= sdp
->sd_rindex
;
800 struct gfs2_inode
*ip
= GFS2_I(inode
);
801 char buf
[sizeof(struct gfs2_rindex
)];
804 for (rgrps
= 0;; rgrps
++) {
805 loff_t pos
= rgrps
* sizeof(struct gfs2_rindex
);
807 if (pos
+ sizeof(struct gfs2_rindex
) > i_size_read(inode
))
809 error
= gfs2_internal_read(ip
, buf
, &pos
,
810 sizeof(struct gfs2_rindex
));
811 if (error
!= sizeof(struct gfs2_rindex
))
813 total_data
+= be32_to_cpu(((struct gfs2_rindex
*)buf
)->ri_data
);
818 static int rgd_insert(struct gfs2_rgrpd
*rgd
)
820 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
821 struct rb_node
**newn
= &sdp
->sd_rindex_tree
.rb_node
, *parent
= NULL
;
823 /* Figure out where to put new node */
825 struct gfs2_rgrpd
*cur
= rb_entry(*newn
, struct gfs2_rgrpd
,
829 if (rgd
->rd_addr
< cur
->rd_addr
)
830 newn
= &((*newn
)->rb_left
);
831 else if (rgd
->rd_addr
> cur
->rd_addr
)
832 newn
= &((*newn
)->rb_right
);
837 rb_link_node(&rgd
->rd_node
, parent
, newn
);
838 rb_insert_color(&rgd
->rd_node
, &sdp
->sd_rindex_tree
);
844 * read_rindex_entry - Pull in a new resource index entry from the disk
845 * @ip: Pointer to the rindex inode
847 * Returns: 0 on success, > 0 on EOF, error code otherwise
850 static int read_rindex_entry(struct gfs2_inode
*ip
)
852 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
853 loff_t pos
= sdp
->sd_rgrps
* sizeof(struct gfs2_rindex
);
854 struct gfs2_rindex buf
;
856 struct gfs2_rgrpd
*rgd
;
858 if (pos
>= i_size_read(&ip
->i_inode
))
861 error
= gfs2_internal_read(ip
, (char *)&buf
, &pos
,
862 sizeof(struct gfs2_rindex
));
864 if (error
!= sizeof(struct gfs2_rindex
))
865 return (error
== 0) ? 1 : error
;
867 rgd
= kmem_cache_zalloc(gfs2_rgrpd_cachep
, GFP_NOFS
);
873 rgd
->rd_addr
= be64_to_cpu(buf
.ri_addr
);
874 rgd
->rd_length
= be32_to_cpu(buf
.ri_length
);
875 rgd
->rd_data0
= be64_to_cpu(buf
.ri_data0
);
876 rgd
->rd_data
= be32_to_cpu(buf
.ri_data
);
877 rgd
->rd_bitbytes
= be32_to_cpu(buf
.ri_bitbytes
);
878 spin_lock_init(&rgd
->rd_rsspin
);
880 error
= compute_bitstructs(rgd
);
884 error
= gfs2_glock_get(sdp
, rgd
->rd_addr
,
885 &gfs2_rgrp_glops
, CREATE
, &rgd
->rd_gl
);
889 rgd
->rd_gl
->gl_object
= rgd
;
890 rgd
->rd_rgl
= (struct gfs2_rgrp_lvb
*)rgd
->rd_gl
->gl_lksb
.sb_lvbptr
;
891 rgd
->rd_flags
&= ~GFS2_RDF_UPTODATE
;
892 if (rgd
->rd_data
> sdp
->sd_max_rg_data
)
893 sdp
->sd_max_rg_data
= rgd
->rd_data
;
894 spin_lock(&sdp
->sd_rindex_spin
);
895 error
= rgd_insert(rgd
);
896 spin_unlock(&sdp
->sd_rindex_spin
);
900 error
= 0; /* someone else read in the rgrp; free it and ignore it */
901 gfs2_glock_put(rgd
->rd_gl
);
905 kmem_cache_free(gfs2_rgrpd_cachep
, rgd
);
910 * gfs2_ri_update - Pull in a new resource index from the disk
911 * @ip: pointer to the rindex inode
913 * Returns: 0 on successful update, error code otherwise
916 static int gfs2_ri_update(struct gfs2_inode
*ip
)
918 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
922 error
= read_rindex_entry(ip
);
923 } while (error
== 0);
928 sdp
->sd_rindex_uptodate
= 1;
933 * gfs2_rindex_update - Update the rindex if required
934 * @sdp: The GFS2 superblock
936 * We grab a lock on the rindex inode to make sure that it doesn't
937 * change whilst we are performing an operation. We keep this lock
938 * for quite long periods of time compared to other locks. This
939 * doesn't matter, since it is shared and it is very, very rarely
940 * accessed in the exclusive mode (i.e. only when expanding the filesystem).
942 * This makes sure that we're using the latest copy of the resource index
943 * special file, which might have been updated if someone expanded the
944 * filesystem (via gfs2_grow utility), which adds new resource groups.
946 * Returns: 0 on succeess, error code otherwise
949 int gfs2_rindex_update(struct gfs2_sbd
*sdp
)
951 struct gfs2_inode
*ip
= GFS2_I(sdp
->sd_rindex
);
952 struct gfs2_glock
*gl
= ip
->i_gl
;
953 struct gfs2_holder ri_gh
;
955 int unlock_required
= 0;
957 /* Read new copy from disk if we don't have the latest */
958 if (!sdp
->sd_rindex_uptodate
) {
959 if (!gfs2_glock_is_locked_by_me(gl
)) {
960 error
= gfs2_glock_nq_init(gl
, LM_ST_SHARED
, 0, &ri_gh
);
965 if (!sdp
->sd_rindex_uptodate
)
966 error
= gfs2_ri_update(ip
);
968 gfs2_glock_dq_uninit(&ri_gh
);
974 static void gfs2_rgrp_in(struct gfs2_rgrpd
*rgd
, const void *buf
)
976 const struct gfs2_rgrp
*str
= buf
;
979 rg_flags
= be32_to_cpu(str
->rg_flags
);
980 rg_flags
&= ~GFS2_RDF_MASK
;
981 rgd
->rd_flags
&= GFS2_RDF_MASK
;
982 rgd
->rd_flags
|= rg_flags
;
983 rgd
->rd_free
= be32_to_cpu(str
->rg_free
);
984 rgd
->rd_dinodes
= be32_to_cpu(str
->rg_dinodes
);
985 rgd
->rd_igeneration
= be64_to_cpu(str
->rg_igeneration
);
988 static void gfs2_rgrp_out(struct gfs2_rgrpd
*rgd
, void *buf
)
990 struct gfs2_rgrp
*str
= buf
;
992 str
->rg_flags
= cpu_to_be32(rgd
->rd_flags
& ~GFS2_RDF_MASK
);
993 str
->rg_free
= cpu_to_be32(rgd
->rd_free
);
994 str
->rg_dinodes
= cpu_to_be32(rgd
->rd_dinodes
);
995 str
->__pad
= cpu_to_be32(0);
996 str
->rg_igeneration
= cpu_to_be64(rgd
->rd_igeneration
);
997 memset(&str
->rg_reserved
, 0, sizeof(str
->rg_reserved
));
1000 static int gfs2_rgrp_lvb_valid(struct gfs2_rgrpd
*rgd
)
1002 struct gfs2_rgrp_lvb
*rgl
= rgd
->rd_rgl
;
1003 struct gfs2_rgrp
*str
= (struct gfs2_rgrp
*)rgd
->rd_bits
[0].bi_bh
->b_data
;
1005 if (rgl
->rl_flags
!= str
->rg_flags
|| rgl
->rl_free
!= str
->rg_free
||
1006 rgl
->rl_dinodes
!= str
->rg_dinodes
||
1007 rgl
->rl_igeneration
!= str
->rg_igeneration
)
1012 static void gfs2_rgrp_ondisk2lvb(struct gfs2_rgrp_lvb
*rgl
, const void *buf
)
1014 const struct gfs2_rgrp
*str
= buf
;
1016 rgl
->rl_magic
= cpu_to_be32(GFS2_MAGIC
);
1017 rgl
->rl_flags
= str
->rg_flags
;
1018 rgl
->rl_free
= str
->rg_free
;
1019 rgl
->rl_dinodes
= str
->rg_dinodes
;
1020 rgl
->rl_igeneration
= str
->rg_igeneration
;
1024 static void update_rgrp_lvb_unlinked(struct gfs2_rgrpd
*rgd
, u32 change
)
1026 struct gfs2_rgrp_lvb
*rgl
= rgd
->rd_rgl
;
1027 u32 unlinked
= be32_to_cpu(rgl
->rl_unlinked
) + change
;
1028 rgl
->rl_unlinked
= cpu_to_be32(unlinked
);
1031 static u32
count_unlinked(struct gfs2_rgrpd
*rgd
)
1033 struct gfs2_bitmap
*bi
;
1034 const u32 length
= rgd
->rd_length
;
1035 const u8
*buffer
= NULL
;
1036 u32 i
, goal
, count
= 0;
1038 for (i
= 0, bi
= rgd
->rd_bits
; i
< length
; i
++, bi
++) {
1040 buffer
= bi
->bi_bh
->b_data
+ bi
->bi_offset
;
1041 WARN_ON(!buffer_uptodate(bi
->bi_bh
));
1042 while (goal
< bi
->bi_len
* GFS2_NBBY
) {
1043 goal
= gfs2_bitfit(buffer
, bi
->bi_len
, goal
,
1044 GFS2_BLKST_UNLINKED
);
1045 if (goal
== BFITNOENT
)
1057 * gfs2_rgrp_bh_get - Read in a RG's header and bitmaps
1058 * @rgd: the struct gfs2_rgrpd describing the RG to read in
1060 * Read in all of a Resource Group's header and bitmap blocks.
1061 * Caller must eventually call gfs2_rgrp_relse() to free the bitmaps.
1066 int gfs2_rgrp_bh_get(struct gfs2_rgrpd
*rgd
)
1068 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
1069 struct gfs2_glock
*gl
= rgd
->rd_gl
;
1070 unsigned int length
= rgd
->rd_length
;
1071 struct gfs2_bitmap
*bi
;
1075 if (rgd
->rd_bits
[0].bi_bh
!= NULL
)
1078 for (x
= 0; x
< length
; x
++) {
1079 bi
= rgd
->rd_bits
+ x
;
1080 error
= gfs2_meta_read(gl
, rgd
->rd_addr
+ x
, 0, &bi
->bi_bh
);
1085 for (y
= length
; y
--;) {
1086 bi
= rgd
->rd_bits
+ y
;
1087 error
= gfs2_meta_wait(sdp
, bi
->bi_bh
);
1090 if (gfs2_metatype_check(sdp
, bi
->bi_bh
, y
? GFS2_METATYPE_RB
:
1091 GFS2_METATYPE_RG
)) {
1097 if (!(rgd
->rd_flags
& GFS2_RDF_UPTODATE
)) {
1098 for (x
= 0; x
< length
; x
++)
1099 clear_bit(GBF_FULL
, &rgd
->rd_bits
[x
].bi_flags
);
1100 gfs2_rgrp_in(rgd
, (rgd
->rd_bits
[0].bi_bh
)->b_data
);
1101 rgd
->rd_flags
|= (GFS2_RDF_UPTODATE
| GFS2_RDF_CHECK
);
1102 rgd
->rd_free_clone
= rgd
->rd_free
;
1104 if (be32_to_cpu(GFS2_MAGIC
) != rgd
->rd_rgl
->rl_magic
) {
1105 rgd
->rd_rgl
->rl_unlinked
= cpu_to_be32(count_unlinked(rgd
));
1106 gfs2_rgrp_ondisk2lvb(rgd
->rd_rgl
,
1107 rgd
->rd_bits
[0].bi_bh
->b_data
);
1109 else if (sdp
->sd_args
.ar_rgrplvb
) {
1110 if (!gfs2_rgrp_lvb_valid(rgd
)){
1111 gfs2_consist_rgrpd(rgd
);
1115 if (rgd
->rd_rgl
->rl_unlinked
== 0)
1116 rgd
->rd_flags
&= ~GFS2_RDF_CHECK
;
1122 bi
= rgd
->rd_bits
+ x
;
1125 gfs2_assert_warn(sdp
, !bi
->bi_clone
);
1131 int update_rgrp_lvb(struct gfs2_rgrpd
*rgd
)
1135 if (rgd
->rd_flags
& GFS2_RDF_UPTODATE
)
1138 if (be32_to_cpu(GFS2_MAGIC
) != rgd
->rd_rgl
->rl_magic
)
1139 return gfs2_rgrp_bh_get(rgd
);
1141 rl_flags
= be32_to_cpu(rgd
->rd_rgl
->rl_flags
);
1142 rl_flags
&= ~GFS2_RDF_MASK
;
1143 rgd
->rd_flags
&= GFS2_RDF_MASK
;
1144 rgd
->rd_flags
|= (rl_flags
| GFS2_RDF_UPTODATE
| GFS2_RDF_CHECK
);
1145 if (rgd
->rd_rgl
->rl_unlinked
== 0)
1146 rgd
->rd_flags
&= ~GFS2_RDF_CHECK
;
1147 rgd
->rd_free
= be32_to_cpu(rgd
->rd_rgl
->rl_free
);
1148 rgd
->rd_free_clone
= rgd
->rd_free
;
1149 rgd
->rd_dinodes
= be32_to_cpu(rgd
->rd_rgl
->rl_dinodes
);
1150 rgd
->rd_igeneration
= be64_to_cpu(rgd
->rd_rgl
->rl_igeneration
);
1154 int gfs2_rgrp_go_lock(struct gfs2_holder
*gh
)
1156 struct gfs2_rgrpd
*rgd
= gh
->gh_gl
->gl_object
;
1157 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
1159 if (gh
->gh_flags
& GL_SKIP
&& sdp
->sd_args
.ar_rgrplvb
)
1161 return gfs2_rgrp_bh_get((struct gfs2_rgrpd
*)gh
->gh_gl
->gl_object
);
1165 * gfs2_rgrp_go_unlock - Release RG bitmaps read in with gfs2_rgrp_bh_get()
1166 * @gh: The glock holder for the resource group
1170 void gfs2_rgrp_go_unlock(struct gfs2_holder
*gh
)
1172 struct gfs2_rgrpd
*rgd
= gh
->gh_gl
->gl_object
;
1173 int x
, length
= rgd
->rd_length
;
1175 for (x
= 0; x
< length
; x
++) {
1176 struct gfs2_bitmap
*bi
= rgd
->rd_bits
+ x
;
1185 int gfs2_rgrp_send_discards(struct gfs2_sbd
*sdp
, u64 offset
,
1186 struct buffer_head
*bh
,
1187 const struct gfs2_bitmap
*bi
, unsigned minlen
, u64
*ptrimmed
)
1189 struct super_block
*sb
= sdp
->sd_vfs
;
1192 sector_t nr_blks
= 0;
1198 for (x
= 0; x
< bi
->bi_len
; x
++) {
1199 const u8
*clone
= bi
->bi_clone
? bi
->bi_clone
: bi
->bi_bh
->b_data
;
1200 clone
+= bi
->bi_offset
;
1203 const u8
*orig
= bh
->b_data
+ bi
->bi_offset
+ x
;
1204 diff
= ~(*orig
| (*orig
>> 1)) & (*clone
| (*clone
>> 1));
1206 diff
= ~(*clone
| (*clone
>> 1));
1211 blk
= offset
+ ((bi
->bi_start
+ x
) * GFS2_NBBY
);
1215 goto start_new_extent
;
1216 if ((start
+ nr_blks
) != blk
) {
1217 if (nr_blks
>= minlen
) {
1218 rv
= sb_issue_discard(sb
,
1235 if (nr_blks
>= minlen
) {
1236 rv
= sb_issue_discard(sb
, start
, nr_blks
, GFP_NOFS
, 0);
1242 *ptrimmed
= trimmed
;
1246 if (sdp
->sd_args
.ar_discard
)
1247 fs_warn(sdp
, "error %d on discard request, turning discards off for this filesystem", rv
);
1248 sdp
->sd_args
.ar_discard
= 0;
1253 * gfs2_fitrim - Generate discard requests for unused bits of the filesystem
1254 * @filp: Any file on the filesystem
1255 * @argp: Pointer to the arguments (also used to pass result)
1257 * Returns: 0 on success, otherwise error code
1260 int gfs2_fitrim(struct file
*filp
, void __user
*argp
)
1262 struct inode
*inode
= file_inode(filp
);
1263 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
1264 struct request_queue
*q
= bdev_get_queue(sdp
->sd_vfs
->s_bdev
);
1265 struct buffer_head
*bh
;
1266 struct gfs2_rgrpd
*rgd
;
1267 struct gfs2_rgrpd
*rgd_end
;
1268 struct gfs2_holder gh
;
1269 struct fstrim_range r
;
1273 u64 start
, end
, minlen
;
1275 unsigned bs_shift
= sdp
->sd_sb
.sb_bsize_shift
;
1277 if (!capable(CAP_SYS_ADMIN
))
1280 if (!blk_queue_discard(q
))
1283 if (copy_from_user(&r
, argp
, sizeof(r
)))
1286 ret
= gfs2_rindex_update(sdp
);
1290 start
= r
.start
>> bs_shift
;
1291 end
= start
+ (r
.len
>> bs_shift
);
1292 minlen
= max_t(u64
, r
.minlen
,
1293 q
->limits
.discard_granularity
) >> bs_shift
;
1295 if (end
<= start
|| minlen
> sdp
->sd_max_rg_data
)
1298 rgd
= gfs2_blk2rgrpd(sdp
, start
, 0);
1299 rgd_end
= gfs2_blk2rgrpd(sdp
, end
, 0);
1301 if ((gfs2_rgrpd_get_first(sdp
) == gfs2_rgrpd_get_next(rgd_end
))
1302 && (start
> rgd_end
->rd_data0
+ rgd_end
->rd_data
))
1303 return -EINVAL
; /* start is beyond the end of the fs */
1307 ret
= gfs2_glock_nq_init(rgd
->rd_gl
, LM_ST_EXCLUSIVE
, 0, &gh
);
1311 if (!(rgd
->rd_flags
& GFS2_RGF_TRIMMED
)) {
1312 /* Trim each bitmap in the rgrp */
1313 for (x
= 0; x
< rgd
->rd_length
; x
++) {
1314 struct gfs2_bitmap
*bi
= rgd
->rd_bits
+ x
;
1315 ret
= gfs2_rgrp_send_discards(sdp
,
1316 rgd
->rd_data0
, NULL
, bi
, minlen
,
1319 gfs2_glock_dq_uninit(&gh
);
1325 /* Mark rgrp as having been trimmed */
1326 ret
= gfs2_trans_begin(sdp
, RES_RG_HDR
, 0);
1328 bh
= rgd
->rd_bits
[0].bi_bh
;
1329 rgd
->rd_flags
|= GFS2_RGF_TRIMMED
;
1330 gfs2_trans_add_meta(rgd
->rd_gl
, bh
);
1331 gfs2_rgrp_out(rgd
, bh
->b_data
);
1332 gfs2_rgrp_ondisk2lvb(rgd
->rd_rgl
, bh
->b_data
);
1333 gfs2_trans_end(sdp
);
1336 gfs2_glock_dq_uninit(&gh
);
1341 rgd
= gfs2_rgrpd_get_next(rgd
);
1345 r
.len
= trimmed
<< bs_shift
;
1346 if (copy_to_user(argp
, &r
, sizeof(r
)))
1353 * rs_insert - insert a new multi-block reservation into the rgrp's rb_tree
1354 * @ip: the inode structure
1357 static void rs_insert(struct gfs2_inode
*ip
)
1359 struct rb_node
**newn
, *parent
= NULL
;
1361 struct gfs2_blkreserv
*rs
= ip
->i_res
;
1362 struct gfs2_rgrpd
*rgd
= rs
->rs_rbm
.rgd
;
1363 u64 fsblock
= gfs2_rbm_to_block(&rs
->rs_rbm
);
1365 BUG_ON(gfs2_rs_active(rs
));
1367 spin_lock(&rgd
->rd_rsspin
);
1368 newn
= &rgd
->rd_rstree
.rb_node
;
1370 struct gfs2_blkreserv
*cur
=
1371 rb_entry(*newn
, struct gfs2_blkreserv
, rs_node
);
1374 rc
= rs_cmp(fsblock
, rs
->rs_free
, cur
);
1376 newn
= &((*newn
)->rb_right
);
1378 newn
= &((*newn
)->rb_left
);
1380 spin_unlock(&rgd
->rd_rsspin
);
1386 rb_link_node(&rs
->rs_node
, parent
, newn
);
1387 rb_insert_color(&rs
->rs_node
, &rgd
->rd_rstree
);
1389 /* Do our rgrp accounting for the reservation */
1390 rgd
->rd_reserved
+= rs
->rs_free
; /* blocks reserved */
1391 spin_unlock(&rgd
->rd_rsspin
);
1392 trace_gfs2_rs(rs
, TRACE_RS_INSERT
);
1396 * rg_mblk_search - find a group of multiple free blocks to form a reservation
1397 * @rgd: the resource group descriptor
1398 * @ip: pointer to the inode for which we're reserving blocks
1399 * @requested: number of blocks required for this allocation
1403 static void rg_mblk_search(struct gfs2_rgrpd
*rgd
, struct gfs2_inode
*ip
,
1406 struct gfs2_rbm rbm
= { .rgd
= rgd
, };
1408 struct gfs2_blkreserv
*rs
= ip
->i_res
;
1410 u32 free_blocks
= rgd
->rd_free_clone
- rgd
->rd_reserved
;
1412 struct inode
*inode
= &ip
->i_inode
;
1414 if (S_ISDIR(inode
->i_mode
))
1417 extlen
= max_t(u32
, atomic_read(&rs
->rs_sizehint
), requested
);
1418 extlen
= clamp(extlen
, RGRP_RSRV_MINBLKS
, free_blocks
);
1420 if ((rgd
->rd_free_clone
< rgd
->rd_reserved
) || (free_blocks
< extlen
))
1423 /* Find bitmap block that contains bits for goal block */
1424 if (rgrp_contains_block(rgd
, ip
->i_goal
))
1427 goal
= rgd
->rd_last_alloc
+ rgd
->rd_data0
;
1429 if (WARN_ON(gfs2_rbm_from_block(&rbm
, goal
)))
1432 ret
= gfs2_rbm_find(&rbm
, GFS2_BLKST_FREE
, extlen
, ip
, true);
1435 rs
->rs_free
= extlen
;
1436 rs
->rs_inum
= ip
->i_no_addr
;
1439 if (goal
== rgd
->rd_last_alloc
+ rgd
->rd_data0
)
1440 rgd
->rd_last_alloc
= 0;
1445 * gfs2_next_unreserved_block - Return next block that is not reserved
1446 * @rgd: The resource group
1447 * @block: The starting block
1448 * @length: The required length
1449 * @ip: Ignore any reservations for this inode
1451 * If the block does not appear in any reservation, then return the
1452 * block number unchanged. If it does appear in the reservation, then
1453 * keep looking through the tree of reservations in order to find the
1454 * first block number which is not reserved.
1457 static u64
gfs2_next_unreserved_block(struct gfs2_rgrpd
*rgd
, u64 block
,
1459 const struct gfs2_inode
*ip
)
1461 struct gfs2_blkreserv
*rs
;
1465 spin_lock(&rgd
->rd_rsspin
);
1466 n
= rgd
->rd_rstree
.rb_node
;
1468 rs
= rb_entry(n
, struct gfs2_blkreserv
, rs_node
);
1469 rc
= rs_cmp(block
, length
, rs
);
1479 while ((rs_cmp(block
, length
, rs
) == 0) && (ip
->i_res
!= rs
)) {
1480 block
= gfs2_rbm_to_block(&rs
->rs_rbm
) + rs
->rs_free
;
1484 rs
= rb_entry(n
, struct gfs2_blkreserv
, rs_node
);
1488 spin_unlock(&rgd
->rd_rsspin
);
1493 * gfs2_reservation_check_and_update - Check for reservations during block alloc
1494 * @rbm: The current position in the resource group
1495 * @ip: The inode for which we are searching for blocks
1496 * @minext: The minimum extent length
1498 * This checks the current position in the rgrp to see whether there is
1499 * a reservation covering this block. If not then this function is a
1500 * no-op. If there is, then the position is moved to the end of the
1501 * contiguous reservation(s) so that we are pointing at the first
1502 * non-reserved block.
1504 * Returns: 0 if no reservation, 1 if @rbm has changed, otherwise an error
1507 static int gfs2_reservation_check_and_update(struct gfs2_rbm
*rbm
,
1508 const struct gfs2_inode
*ip
,
1511 u64 block
= gfs2_rbm_to_block(rbm
);
1517 * If we have a minimum extent length, then skip over any extent
1518 * which is less than the min extent length in size.
1521 extlen
= gfs2_free_extlen(rbm
, minext
);
1522 nblock
= block
+ extlen
;
1523 if (extlen
< minext
)
1528 * Check the extent which has been found against the reservations
1529 * and skip if parts of it are already reserved
1531 nblock
= gfs2_next_unreserved_block(rbm
->rgd
, block
, extlen
, ip
);
1532 if (nblock
== block
)
1535 ret
= gfs2_rbm_from_block(rbm
, nblock
);
1542 * gfs2_rbm_find - Look for blocks of a particular state
1543 * @rbm: Value/result starting position and final position
1544 * @state: The state which we want to find
1545 * @minext: The requested extent length (0 for a single block)
1546 * @ip: If set, check for reservations
1547 * @nowrap: Stop looking at the end of the rgrp, rather than wrapping
1548 * around until we've reached the starting point.
1551 * - If looking for free blocks, we set GBF_FULL on each bitmap which
1552 * has no free blocks in it.
1554 * Returns: 0 on success, -ENOSPC if there is no block of the requested state
1557 static int gfs2_rbm_find(struct gfs2_rbm
*rbm
, u8 state
, u32 minext
,
1558 const struct gfs2_inode
*ip
, bool nowrap
)
1560 struct buffer_head
*bh
;
1561 struct gfs2_bitmap
*initial_bi
;
1567 int iters
= rbm
->rgd
->rd_length
;
1570 /* If we are not starting at the beginning of a bitmap, then we
1571 * need to add one to the bitmap count to ensure that we search
1572 * the starting bitmap twice.
1574 if (rbm
->offset
!= 0)
1578 if (test_bit(GBF_FULL
, &rbm
->bi
->bi_flags
) &&
1579 (state
== GFS2_BLKST_FREE
))
1582 bh
= rbm
->bi
->bi_bh
;
1583 buffer
= bh
->b_data
+ rbm
->bi
->bi_offset
;
1584 WARN_ON(!buffer_uptodate(bh
));
1585 if (state
!= GFS2_BLKST_UNLINKED
&& rbm
->bi
->bi_clone
)
1586 buffer
= rbm
->bi
->bi_clone
+ rbm
->bi
->bi_offset
;
1587 initial_offset
= rbm
->offset
;
1588 offset
= gfs2_bitfit(buffer
, rbm
->bi
->bi_len
, rbm
->offset
, state
);
1589 if (offset
== BFITNOENT
)
1591 rbm
->offset
= offset
;
1595 initial_bi
= rbm
->bi
;
1596 ret
= gfs2_reservation_check_and_update(rbm
, ip
, minext
);
1600 n
+= (rbm
->bi
- initial_bi
);
1603 if (ret
== -E2BIG
) {
1606 n
+= (rbm
->bi
- initial_bi
);
1607 goto res_covered_end_of_rgrp
;
1611 bitmap_full
: /* Mark bitmap as full and fall through */
1612 if ((state
== GFS2_BLKST_FREE
) && initial_offset
== 0)
1613 set_bit(GBF_FULL
, &rbm
->bi
->bi_flags
);
1615 next_bitmap
: /* Find next bitmap in the rgrp */
1617 index
= rbm
->bi
- rbm
->rgd
->rd_bits
;
1619 if (index
== rbm
->rgd
->rd_length
)
1621 res_covered_end_of_rgrp
:
1622 rbm
->bi
= &rbm
->rgd
->rd_bits
[index
];
1623 if ((index
== 0) && nowrap
)
1635 * try_rgrp_unlink - Look for any unlinked, allocated, but unused inodes
1637 * @last_unlinked: block address of the last dinode we unlinked
1638 * @skip: block address we should explicitly not unlink
1640 * Returns: 0 if no error
1641 * The inode, if one has been found, in inode.
1644 static void try_rgrp_unlink(struct gfs2_rgrpd
*rgd
, u64
*last_unlinked
, u64 skip
)
1647 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
1648 struct gfs2_glock
*gl
;
1649 struct gfs2_inode
*ip
;
1652 struct gfs2_rbm rbm
= { .rgd
= rgd
, .bi
= rgd
->rd_bits
, .offset
= 0 };
1655 down_write(&sdp
->sd_log_flush_lock
);
1656 error
= gfs2_rbm_find(&rbm
, GFS2_BLKST_UNLINKED
, 0, NULL
, true);
1657 up_write(&sdp
->sd_log_flush_lock
);
1658 if (error
== -ENOSPC
)
1660 if (WARN_ON_ONCE(error
))
1663 block
= gfs2_rbm_to_block(&rbm
);
1664 if (gfs2_rbm_from_block(&rbm
, block
+ 1))
1666 if (*last_unlinked
!= NO_BLOCK
&& block
<= *last_unlinked
)
1670 *last_unlinked
= block
;
1672 error
= gfs2_glock_get(sdp
, block
, &gfs2_inode_glops
, CREATE
, &gl
);
1676 /* If the inode is already in cache, we can ignore it here
1677 * because the existing inode disposal code will deal with
1678 * it when all refs have gone away. Accessing gl_object like
1679 * this is not safe in general. Here it is ok because we do
1680 * not dereference the pointer, and we only need an approx
1681 * answer to whether it is NULL or not.
1685 if (ip
|| queue_work(gfs2_delete_workqueue
, &gl
->gl_delete
) == 0)
1690 /* Limit reclaim to sensible number of tasks */
1691 if (found
> NR_CPUS
)
1695 rgd
->rd_flags
&= ~GFS2_RDF_CHECK
;
1700 * gfs2_rgrp_congested - Use stats to figure out whether an rgrp is congested
1701 * @rgd: The rgrp in question
1702 * @loops: An indication of how picky we can be (0=very, 1=less so)
1704 * This function uses the recently added glock statistics in order to
1705 * figure out whether a parciular resource group is suffering from
1706 * contention from multiple nodes. This is done purely on the basis
1707 * of timings, since this is the only data we have to work with and
1708 * our aim here is to reject a resource group which is highly contended
1709 * but (very important) not to do this too often in order to ensure that
1710 * we do not land up introducing fragmentation by changing resource
1711 * groups when not actually required.
1713 * The calculation is fairly simple, we want to know whether the SRTTB
1714 * (i.e. smoothed round trip time for blocking operations) to acquire
1715 * the lock for this rgrp's glock is significantly greater than the
1716 * time taken for resource groups on average. We introduce a margin in
1717 * the form of the variable @var which is computed as the sum of the two
1718 * respective variences, and multiplied by a factor depending on @loops
1719 * and whether we have a lot of data to base the decision on. This is
1720 * then tested against the square difference of the means in order to
1721 * decide whether the result is statistically significant or not.
1723 * Returns: A boolean verdict on the congestion status
1726 static bool gfs2_rgrp_congested(const struct gfs2_rgrpd
*rgd
, int loops
)
1728 const struct gfs2_glock
*gl
= rgd
->rd_gl
;
1729 const struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
1730 struct gfs2_lkstats
*st
;
1731 s64 r_dcount
, l_dcount
;
1732 s64 r_srttb
, l_srttb
;
1738 st
= &this_cpu_ptr(sdp
->sd_lkstats
)->lkstats
[LM_TYPE_RGRP
];
1739 r_srttb
= st
->stats
[GFS2_LKS_SRTTB
];
1740 r_dcount
= st
->stats
[GFS2_LKS_DCOUNT
];
1741 var
= st
->stats
[GFS2_LKS_SRTTVARB
] +
1742 gl
->gl_stats
.stats
[GFS2_LKS_SRTTVARB
];
1745 l_srttb
= gl
->gl_stats
.stats
[GFS2_LKS_SRTTB
];
1746 l_dcount
= gl
->gl_stats
.stats
[GFS2_LKS_DCOUNT
];
1748 if ((l_dcount
< 1) || (r_dcount
< 1) || (r_srttb
== 0))
1751 srttb_diff
= r_srttb
- l_srttb
;
1752 sqr_diff
= srttb_diff
* srttb_diff
;
1755 if (l_dcount
< 8 || r_dcount
< 8)
1760 return ((srttb_diff
< 0) && (sqr_diff
> var
));
1764 * gfs2_rgrp_used_recently
1765 * @rs: The block reservation with the rgrp to test
1766 * @msecs: The time limit in milliseconds
1768 * Returns: True if the rgrp glock has been used within the time limit
1770 static bool gfs2_rgrp_used_recently(const struct gfs2_blkreserv
*rs
,
1775 tdiff
= ktime_to_ns(ktime_sub(ktime_get_real(),
1776 rs
->rs_rbm
.rgd
->rd_gl
->gl_dstamp
));
1778 return tdiff
> (msecs
* 1000 * 1000);
1781 static u32
gfs2_orlov_skip(const struct gfs2_inode
*ip
)
1783 const struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
1786 get_random_bytes(&skip
, sizeof(skip
));
1787 return skip
% sdp
->sd_rgrps
;
1790 static bool gfs2_select_rgrp(struct gfs2_rgrpd
**pos
, const struct gfs2_rgrpd
*begin
)
1792 struct gfs2_rgrpd
*rgd
= *pos
;
1793 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
1795 rgd
= gfs2_rgrpd_get_next(rgd
);
1797 rgd
= gfs2_rgrpd_get_first(sdp
);
1799 if (rgd
!= begin
) /* If we didn't wrap */
1805 * gfs2_inplace_reserve - Reserve space in the filesystem
1806 * @ip: the inode to reserve space for
1807 * @requested: the number of blocks to be reserved
1812 int gfs2_inplace_reserve(struct gfs2_inode
*ip
, u32 requested
, u32 aflags
)
1814 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
1815 struct gfs2_rgrpd
*begin
= NULL
;
1816 struct gfs2_blkreserv
*rs
= ip
->i_res
;
1817 int error
= 0, rg_locked
, flags
= 0;
1818 u64 last_unlinked
= NO_BLOCK
;
1822 if (sdp
->sd_args
.ar_rgrplvb
)
1824 if (gfs2_assert_warn(sdp
, requested
))
1826 if (gfs2_rs_active(rs
)) {
1827 begin
= rs
->rs_rbm
.rgd
;
1828 } else if (ip
->i_rgd
&& rgrp_contains_block(ip
->i_rgd
, ip
->i_goal
)) {
1829 rs
->rs_rbm
.rgd
= begin
= ip
->i_rgd
;
1831 rs
->rs_rbm
.rgd
= begin
= gfs2_blk2rgrpd(sdp
, ip
->i_goal
, 1);
1833 if (S_ISDIR(ip
->i_inode
.i_mode
) && (aflags
& GFS2_AF_ORLOV
))
1834 skip
= gfs2_orlov_skip(ip
);
1835 if (rs
->rs_rbm
.rgd
== NULL
)
1841 if (!gfs2_glock_is_locked_by_me(rs
->rs_rbm
.rgd
->rd_gl
)) {
1845 if (!gfs2_rs_active(rs
) && (loops
< 2) &&
1846 gfs2_rgrp_used_recently(rs
, 1000) &&
1847 gfs2_rgrp_congested(rs
->rs_rbm
.rgd
, loops
))
1849 error
= gfs2_glock_nq_init(rs
->rs_rbm
.rgd
->rd_gl
,
1850 LM_ST_EXCLUSIVE
, flags
,
1852 if (unlikely(error
))
1854 if (!gfs2_rs_active(rs
) && (loops
< 2) &&
1855 gfs2_rgrp_congested(rs
->rs_rbm
.rgd
, loops
))
1857 if (sdp
->sd_args
.ar_rgrplvb
) {
1858 error
= update_rgrp_lvb(rs
->rs_rbm
.rgd
);
1859 if (unlikely(error
)) {
1860 gfs2_glock_dq_uninit(&rs
->rs_rgd_gh
);
1866 /* Skip unuseable resource groups */
1867 if (rs
->rs_rbm
.rgd
->rd_flags
& (GFS2_RGF_NOALLOC
| GFS2_RDF_ERROR
))
1870 if (sdp
->sd_args
.ar_rgrplvb
)
1871 gfs2_rgrp_bh_get(rs
->rs_rbm
.rgd
);
1873 /* Get a reservation if we don't already have one */
1874 if (!gfs2_rs_active(rs
))
1875 rg_mblk_search(rs
->rs_rbm
.rgd
, ip
, requested
);
1877 /* Skip rgrps when we can't get a reservation on first pass */
1878 if (!gfs2_rs_active(rs
) && (loops
< 1))
1881 /* If rgrp has enough free space, use it */
1882 if (rs
->rs_rbm
.rgd
->rd_free_clone
>= requested
) {
1883 ip
->i_rgd
= rs
->rs_rbm
.rgd
;
1887 /* Drop reservation, if we couldn't use reserved rgrp */
1888 if (gfs2_rs_active(rs
))
1889 gfs2_rs_deltree(rs
);
1891 /* Check for unlinked inodes which can be reclaimed */
1892 if (rs
->rs_rbm
.rgd
->rd_flags
& GFS2_RDF_CHECK
)
1893 try_rgrp_unlink(rs
->rs_rbm
.rgd
, &last_unlinked
,
1896 /* Unlock rgrp if required */
1898 gfs2_glock_dq_uninit(&rs
->rs_rgd_gh
);
1900 /* Find the next rgrp, and continue looking */
1901 if (gfs2_select_rgrp(&rs
->rs_rbm
.rgd
, begin
))
1906 /* If we've scanned all the rgrps, but found no free blocks
1907 * then this checks for some less likely conditions before
1911 /* Check that fs hasn't grown if writing to rindex */
1912 if (ip
== GFS2_I(sdp
->sd_rindex
) && !sdp
->sd_rindex_uptodate
) {
1913 error
= gfs2_ri_update(ip
);
1917 /* Flushing the log may release space */
1919 gfs2_log_flush(sdp
, NULL
);
1926 * gfs2_inplace_release - release an inplace reservation
1927 * @ip: the inode the reservation was taken out on
1929 * Release a reservation made by gfs2_inplace_reserve().
1932 void gfs2_inplace_release(struct gfs2_inode
*ip
)
1934 struct gfs2_blkreserv
*rs
= ip
->i_res
;
1936 if (rs
->rs_rgd_gh
.gh_gl
)
1937 gfs2_glock_dq_uninit(&rs
->rs_rgd_gh
);
1941 * gfs2_get_block_type - Check a block in a RG is of given type
1942 * @rgd: the resource group holding the block
1943 * @block: the block number
1945 * Returns: The block type (GFS2_BLKST_*)
1948 static unsigned char gfs2_get_block_type(struct gfs2_rgrpd
*rgd
, u64 block
)
1950 struct gfs2_rbm rbm
= { .rgd
= rgd
, };
1953 ret
= gfs2_rbm_from_block(&rbm
, block
);
1954 WARN_ON_ONCE(ret
!= 0);
1956 return gfs2_testbit(&rbm
);
1961 * gfs2_alloc_extent - allocate an extent from a given bitmap
1962 * @rbm: the resource group information
1963 * @dinode: TRUE if the first block we allocate is for a dinode
1964 * @n: The extent length (value/result)
1966 * Add the bitmap buffer to the transaction.
1967 * Set the found bits to @new_state to change block's allocation state.
1969 static void gfs2_alloc_extent(const struct gfs2_rbm
*rbm
, bool dinode
,
1972 struct gfs2_rbm pos
= { .rgd
= rbm
->rgd
, };
1973 const unsigned int elen
= *n
;
1978 block
= gfs2_rbm_to_block(rbm
);
1979 gfs2_trans_add_meta(rbm
->rgd
->rd_gl
, rbm
->bi
->bi_bh
);
1980 gfs2_setbit(rbm
, true, dinode
? GFS2_BLKST_DINODE
: GFS2_BLKST_USED
);
1983 ret
= gfs2_rbm_from_block(&pos
, block
);
1984 if (ret
|| gfs2_testbit(&pos
) != GFS2_BLKST_FREE
)
1986 gfs2_trans_add_meta(pos
.rgd
->rd_gl
, pos
.bi
->bi_bh
);
1987 gfs2_setbit(&pos
, true, GFS2_BLKST_USED
);
1994 * rgblk_free - Change alloc state of given block(s)
1995 * @sdp: the filesystem
1996 * @bstart: the start of a run of blocks to free
1997 * @blen: the length of the block run (all must lie within ONE RG!)
1998 * @new_state: GFS2_BLKST_XXX the after-allocation block state
2000 * Returns: Resource group containing the block(s)
2003 static struct gfs2_rgrpd
*rgblk_free(struct gfs2_sbd
*sdp
, u64 bstart
,
2004 u32 blen
, unsigned char new_state
)
2006 struct gfs2_rbm rbm
;
2008 rbm
.rgd
= gfs2_blk2rgrpd(sdp
, bstart
, 1);
2010 if (gfs2_consist(sdp
))
2011 fs_err(sdp
, "block = %llu\n", (unsigned long long)bstart
);
2016 gfs2_rbm_from_block(&rbm
, bstart
);
2018 if (!rbm
.bi
->bi_clone
) {
2019 rbm
.bi
->bi_clone
= kmalloc(rbm
.bi
->bi_bh
->b_size
,
2020 GFP_NOFS
| __GFP_NOFAIL
);
2021 memcpy(rbm
.bi
->bi_clone
+ rbm
.bi
->bi_offset
,
2022 rbm
.bi
->bi_bh
->b_data
+ rbm
.bi
->bi_offset
,
2025 gfs2_trans_add_meta(rbm
.rgd
->rd_gl
, rbm
.bi
->bi_bh
);
2026 gfs2_setbit(&rbm
, false, new_state
);
2033 * gfs2_rgrp_dump - print out an rgrp
2034 * @seq: The iterator
2035 * @gl: The glock in question
2039 int gfs2_rgrp_dump(struct seq_file
*seq
, const struct gfs2_glock
*gl
)
2041 struct gfs2_rgrpd
*rgd
= gl
->gl_object
;
2042 struct gfs2_blkreserv
*trs
;
2043 const struct rb_node
*n
;
2047 gfs2_print_dbg(seq
, " R: n:%llu f:%02x b:%u/%u i:%u r:%u\n",
2048 (unsigned long long)rgd
->rd_addr
, rgd
->rd_flags
,
2049 rgd
->rd_free
, rgd
->rd_free_clone
, rgd
->rd_dinodes
,
2051 spin_lock(&rgd
->rd_rsspin
);
2052 for (n
= rb_first(&rgd
->rd_rstree
); n
; n
= rb_next(&trs
->rs_node
)) {
2053 trs
= rb_entry(n
, struct gfs2_blkreserv
, rs_node
);
2056 spin_unlock(&rgd
->rd_rsspin
);
2060 static void gfs2_rgrp_error(struct gfs2_rgrpd
*rgd
)
2062 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
2063 fs_warn(sdp
, "rgrp %llu has an error, marking it readonly until umount\n",
2064 (unsigned long long)rgd
->rd_addr
);
2065 fs_warn(sdp
, "umount on all nodes and run fsck.gfs2 to fix the error\n");
2066 gfs2_rgrp_dump(NULL
, rgd
->rd_gl
);
2067 rgd
->rd_flags
|= GFS2_RDF_ERROR
;
2071 * gfs2_adjust_reservation - Adjust (or remove) a reservation after allocation
2072 * @ip: The inode we have just allocated blocks for
2073 * @rbm: The start of the allocated blocks
2074 * @len: The extent length
2076 * Adjusts a reservation after an allocation has taken place. If the
2077 * reservation does not match the allocation, or if it is now empty
2078 * then it is removed.
2081 static void gfs2_adjust_reservation(struct gfs2_inode
*ip
,
2082 const struct gfs2_rbm
*rbm
, unsigned len
)
2084 struct gfs2_blkreserv
*rs
= ip
->i_res
;
2085 struct gfs2_rgrpd
*rgd
= rbm
->rgd
;
2090 spin_lock(&rgd
->rd_rsspin
);
2091 if (gfs2_rs_active(rs
)) {
2092 if (gfs2_rbm_eq(&rs
->rs_rbm
, rbm
)) {
2093 block
= gfs2_rbm_to_block(rbm
);
2094 ret
= gfs2_rbm_from_block(&rs
->rs_rbm
, block
+ len
);
2095 rlen
= min(rs
->rs_free
, len
);
2096 rs
->rs_free
-= rlen
;
2097 rgd
->rd_reserved
-= rlen
;
2098 trace_gfs2_rs(rs
, TRACE_RS_CLAIM
);
2099 if (rs
->rs_free
&& !ret
)
2105 spin_unlock(&rgd
->rd_rsspin
);
2109 * gfs2_alloc_blocks - Allocate one or more blocks of data and/or a dinode
2110 * @ip: the inode to allocate the block for
2111 * @bn: Used to return the starting block number
2112 * @nblocks: requested number of blocks/extent length (value/result)
2113 * @dinode: 1 if we're allocating a dinode block, else 0
2114 * @generation: the generation number of the inode
2116 * Returns: 0 or error
2119 int gfs2_alloc_blocks(struct gfs2_inode
*ip
, u64
*bn
, unsigned int *nblocks
,
2120 bool dinode
, u64
*generation
)
2122 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
2123 struct buffer_head
*dibh
;
2124 struct gfs2_rbm rbm
= { .rgd
= ip
->i_rgd
, };
2127 u64 block
; /* block, within the file system scope */
2130 if (gfs2_rs_active(ip
->i_res
))
2131 goal
= gfs2_rbm_to_block(&ip
->i_res
->rs_rbm
);
2132 else if (!dinode
&& rgrp_contains_block(rbm
.rgd
, ip
->i_goal
))
2135 goal
= rbm
.rgd
->rd_last_alloc
+ rbm
.rgd
->rd_data0
;
2137 gfs2_rbm_from_block(&rbm
, goal
);
2138 error
= gfs2_rbm_find(&rbm
, GFS2_BLKST_FREE
, 0, ip
, false);
2140 if (error
== -ENOSPC
) {
2141 gfs2_rbm_from_block(&rbm
, goal
);
2142 error
= gfs2_rbm_find(&rbm
, GFS2_BLKST_FREE
, 0, NULL
, false);
2145 /* Since all blocks are reserved in advance, this shouldn't happen */
2147 fs_warn(sdp
, "inum=%llu error=%d, nblocks=%u, full=%d\n",
2148 (unsigned long long)ip
->i_no_addr
, error
, *nblocks
,
2149 test_bit(GBF_FULL
, &rbm
.rgd
->rd_bits
->bi_flags
));
2153 gfs2_alloc_extent(&rbm
, dinode
, nblocks
);
2154 block
= gfs2_rbm_to_block(&rbm
);
2155 rbm
.rgd
->rd_last_alloc
= block
- rbm
.rgd
->rd_data0
;
2156 if (gfs2_rs_active(ip
->i_res
))
2157 gfs2_adjust_reservation(ip
, &rbm
, *nblocks
);
2163 ip
->i_goal
= block
+ ndata
- 1;
2164 error
= gfs2_meta_inode_buffer(ip
, &dibh
);
2166 struct gfs2_dinode
*di
=
2167 (struct gfs2_dinode
*)dibh
->b_data
;
2168 gfs2_trans_add_meta(ip
->i_gl
, dibh
);
2169 di
->di_goal_meta
= di
->di_goal_data
=
2170 cpu_to_be64(ip
->i_goal
);
2174 if (rbm
.rgd
->rd_free
< *nblocks
) {
2175 printk(KERN_WARNING
"nblocks=%u\n", *nblocks
);
2179 rbm
.rgd
->rd_free
-= *nblocks
;
2181 rbm
.rgd
->rd_dinodes
++;
2182 *generation
= rbm
.rgd
->rd_igeneration
++;
2183 if (*generation
== 0)
2184 *generation
= rbm
.rgd
->rd_igeneration
++;
2187 gfs2_trans_add_meta(rbm
.rgd
->rd_gl
, rbm
.rgd
->rd_bits
[0].bi_bh
);
2188 gfs2_rgrp_out(rbm
.rgd
, rbm
.rgd
->rd_bits
[0].bi_bh
->b_data
);
2189 gfs2_rgrp_ondisk2lvb(rbm
.rgd
->rd_rgl
, rbm
.rgd
->rd_bits
[0].bi_bh
->b_data
);
2191 gfs2_statfs_change(sdp
, 0, -(s64
)*nblocks
, dinode
? 1 : 0);
2193 gfs2_trans_add_unrevoke(sdp
, block
, 1);
2195 gfs2_quota_change(ip
, *nblocks
, ip
->i_inode
.i_uid
, ip
->i_inode
.i_gid
);
2197 rbm
.rgd
->rd_free_clone
-= *nblocks
;
2198 trace_gfs2_block_alloc(ip
, rbm
.rgd
, block
, *nblocks
,
2199 dinode
? GFS2_BLKST_DINODE
: GFS2_BLKST_USED
);
2204 gfs2_rgrp_error(rbm
.rgd
);
2209 * __gfs2_free_blocks - free a contiguous run of block(s)
2210 * @ip: the inode these blocks are being freed from
2211 * @bstart: first block of a run of contiguous blocks
2212 * @blen: the length of the block run
2213 * @meta: 1 if the blocks represent metadata
2217 void __gfs2_free_blocks(struct gfs2_inode
*ip
, u64 bstart
, u32 blen
, int meta
)
2219 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
2220 struct gfs2_rgrpd
*rgd
;
2222 rgd
= rgblk_free(sdp
, bstart
, blen
, GFS2_BLKST_FREE
);
2225 trace_gfs2_block_alloc(ip
, rgd
, bstart
, blen
, GFS2_BLKST_FREE
);
2226 rgd
->rd_free
+= blen
;
2227 rgd
->rd_flags
&= ~GFS2_RGF_TRIMMED
;
2228 gfs2_trans_add_meta(rgd
->rd_gl
, rgd
->rd_bits
[0].bi_bh
);
2229 gfs2_rgrp_out(rgd
, rgd
->rd_bits
[0].bi_bh
->b_data
);
2230 gfs2_rgrp_ondisk2lvb(rgd
->rd_rgl
, rgd
->rd_bits
[0].bi_bh
->b_data
);
2232 /* Directories keep their data in the metadata address space */
2233 if (meta
|| ip
->i_depth
)
2234 gfs2_meta_wipe(ip
, bstart
, blen
);
2238 * gfs2_free_meta - free a contiguous run of data block(s)
2239 * @ip: the inode these blocks are being freed from
2240 * @bstart: first block of a run of contiguous blocks
2241 * @blen: the length of the block run
2245 void gfs2_free_meta(struct gfs2_inode
*ip
, u64 bstart
, u32 blen
)
2247 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
2249 __gfs2_free_blocks(ip
, bstart
, blen
, 1);
2250 gfs2_statfs_change(sdp
, 0, +blen
, 0);
2251 gfs2_quota_change(ip
, -(s64
)blen
, ip
->i_inode
.i_uid
, ip
->i_inode
.i_gid
);
2254 void gfs2_unlink_di(struct inode
*inode
)
2256 struct gfs2_inode
*ip
= GFS2_I(inode
);
2257 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
2258 struct gfs2_rgrpd
*rgd
;
2259 u64 blkno
= ip
->i_no_addr
;
2261 rgd
= rgblk_free(sdp
, blkno
, 1, GFS2_BLKST_UNLINKED
);
2264 trace_gfs2_block_alloc(ip
, rgd
, blkno
, 1, GFS2_BLKST_UNLINKED
);
2265 gfs2_trans_add_meta(rgd
->rd_gl
, rgd
->rd_bits
[0].bi_bh
);
2266 gfs2_rgrp_out(rgd
, rgd
->rd_bits
[0].bi_bh
->b_data
);
2267 gfs2_rgrp_ondisk2lvb(rgd
->rd_rgl
, rgd
->rd_bits
[0].bi_bh
->b_data
);
2268 update_rgrp_lvb_unlinked(rgd
, 1);
2271 static void gfs2_free_uninit_di(struct gfs2_rgrpd
*rgd
, u64 blkno
)
2273 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
2274 struct gfs2_rgrpd
*tmp_rgd
;
2276 tmp_rgd
= rgblk_free(sdp
, blkno
, 1, GFS2_BLKST_FREE
);
2279 gfs2_assert_withdraw(sdp
, rgd
== tmp_rgd
);
2281 if (!rgd
->rd_dinodes
)
2282 gfs2_consist_rgrpd(rgd
);
2286 gfs2_trans_add_meta(rgd
->rd_gl
, rgd
->rd_bits
[0].bi_bh
);
2287 gfs2_rgrp_out(rgd
, rgd
->rd_bits
[0].bi_bh
->b_data
);
2288 gfs2_rgrp_ondisk2lvb(rgd
->rd_rgl
, rgd
->rd_bits
[0].bi_bh
->b_data
);
2289 update_rgrp_lvb_unlinked(rgd
, -1);
2291 gfs2_statfs_change(sdp
, 0, +1, -1);
2295 void gfs2_free_di(struct gfs2_rgrpd
*rgd
, struct gfs2_inode
*ip
)
2297 gfs2_free_uninit_di(rgd
, ip
->i_no_addr
);
2298 trace_gfs2_block_alloc(ip
, rgd
, ip
->i_no_addr
, 1, GFS2_BLKST_FREE
);
2299 gfs2_quota_change(ip
, -1, ip
->i_inode
.i_uid
, ip
->i_inode
.i_gid
);
2300 gfs2_meta_wipe(ip
, ip
->i_no_addr
, 1);
2304 * gfs2_check_blk_type - Check the type of a block
2305 * @sdp: The superblock
2306 * @no_addr: The block number to check
2307 * @type: The block type we are looking for
2309 * Returns: 0 if the block type matches the expected type
2310 * -ESTALE if it doesn't match
2311 * or -ve errno if something went wrong while checking
2314 int gfs2_check_blk_type(struct gfs2_sbd
*sdp
, u64 no_addr
, unsigned int type
)
2316 struct gfs2_rgrpd
*rgd
;
2317 struct gfs2_holder rgd_gh
;
2318 int error
= -EINVAL
;
2320 rgd
= gfs2_blk2rgrpd(sdp
, no_addr
, 1);
2324 error
= gfs2_glock_nq_init(rgd
->rd_gl
, LM_ST_SHARED
, 0, &rgd_gh
);
2328 if (gfs2_get_block_type(rgd
, no_addr
) != type
)
2331 gfs2_glock_dq_uninit(&rgd_gh
);
2337 * gfs2_rlist_add - add a RG to a list of RGs
2339 * @rlist: the list of resource groups
2342 * Figure out what RG a block belongs to and add that RG to the list
2344 * FIXME: Don't use NOFAIL
2348 void gfs2_rlist_add(struct gfs2_inode
*ip
, struct gfs2_rgrp_list
*rlist
,
2351 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
2352 struct gfs2_rgrpd
*rgd
;
2353 struct gfs2_rgrpd
**tmp
;
2354 unsigned int new_space
;
2357 if (gfs2_assert_warn(sdp
, !rlist
->rl_ghs
))
2360 if (ip
->i_rgd
&& rgrp_contains_block(ip
->i_rgd
, block
))
2363 rgd
= gfs2_blk2rgrpd(sdp
, block
, 1);
2365 fs_err(sdp
, "rlist_add: no rgrp for block %llu\n", (unsigned long long)block
);
2370 for (x
= 0; x
< rlist
->rl_rgrps
; x
++)
2371 if (rlist
->rl_rgd
[x
] == rgd
)
2374 if (rlist
->rl_rgrps
== rlist
->rl_space
) {
2375 new_space
= rlist
->rl_space
+ 10;
2377 tmp
= kcalloc(new_space
, sizeof(struct gfs2_rgrpd
*),
2378 GFP_NOFS
| __GFP_NOFAIL
);
2380 if (rlist
->rl_rgd
) {
2381 memcpy(tmp
, rlist
->rl_rgd
,
2382 rlist
->rl_space
* sizeof(struct gfs2_rgrpd
*));
2383 kfree(rlist
->rl_rgd
);
2386 rlist
->rl_space
= new_space
;
2387 rlist
->rl_rgd
= tmp
;
2390 rlist
->rl_rgd
[rlist
->rl_rgrps
++] = rgd
;
2394 * gfs2_rlist_alloc - all RGs have been added to the rlist, now allocate
2395 * and initialize an array of glock holders for them
2396 * @rlist: the list of resource groups
2397 * @state: the lock state to acquire the RG lock in
2399 * FIXME: Don't use NOFAIL
2403 void gfs2_rlist_alloc(struct gfs2_rgrp_list
*rlist
, unsigned int state
)
2407 rlist
->rl_ghs
= kcalloc(rlist
->rl_rgrps
, sizeof(struct gfs2_holder
),
2408 GFP_NOFS
| __GFP_NOFAIL
);
2409 for (x
= 0; x
< rlist
->rl_rgrps
; x
++)
2410 gfs2_holder_init(rlist
->rl_rgd
[x
]->rd_gl
,
2416 * gfs2_rlist_free - free a resource group list
2417 * @list: the list of resource groups
2421 void gfs2_rlist_free(struct gfs2_rgrp_list
*rlist
)
2425 kfree(rlist
->rl_rgd
);
2427 if (rlist
->rl_ghs
) {
2428 for (x
= 0; x
< rlist
->rl_rgrps
; x
++)
2429 gfs2_holder_uninit(&rlist
->rl_ghs
[x
]);
2430 kfree(rlist
->rl_ghs
);
2431 rlist
->rl_ghs
= NULL
;
This page took 0.194015 seconds and 6 git commands to generate.