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 struct gfs2_bitmap
*bi
= rbm_bi(rbm
);
85 unsigned int buflen
= bi
->bi_len
;
86 const unsigned int bit
= (rbm
->offset
% GFS2_NBBY
) * GFS2_BIT_SIZE
;
88 byte1
= bi
->bi_bh
->b_data
+ bi
->bi_offset
+ (rbm
->offset
/ GFS2_NBBY
);
89 end
= bi
->bi_bh
->b_data
+ bi
->bi_offset
+ buflen
;
93 cur_state
= (*byte1
>> bit
) & GFS2_BIT_MASK
;
95 if (unlikely(!valid_change
[new_state
* 4 + cur_state
])) {
96 printk(KERN_WARNING
"GFS2: buf_blk = 0x%x old_state=%d, "
97 "new_state=%d\n", rbm
->offset
, cur_state
, new_state
);
98 printk(KERN_WARNING
"GFS2: rgrp=0x%llx bi_start=0x%x\n",
99 (unsigned long long)rbm
->rgd
->rd_addr
, bi
->bi_start
);
100 printk(KERN_WARNING
"GFS2: bi_offset=0x%x bi_len=0x%x\n",
101 bi
->bi_offset
, bi
->bi_len
);
103 gfs2_consist_rgrpd(rbm
->rgd
);
106 *byte1
^= (cur_state
^ new_state
) << bit
;
108 if (do_clone
&& bi
->bi_clone
) {
109 byte2
= bi
->bi_clone
+ 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 struct gfs2_bitmap
*bi
= rbm_bi(rbm
);
125 const u8
*buffer
= bi
->bi_bh
->b_data
+ bi
->bi_offset
;
129 byte
= buffer
+ (rbm
->offset
/ GFS2_NBBY
);
130 bit
= (rbm
->offset
% GFS2_NBBY
) * GFS2_BIT_SIZE
;
132 return (*byte
>> bit
) & GFS2_BIT_MASK
;
137 * @ptr: Pointer to bitmap data
138 * @mask: Mask to use (normally 0x55555.... but adjusted for search start)
139 * @state: The state we are searching for
141 * We xor the bitmap data with a patter which is the bitwise opposite
142 * of what we are looking for, this gives rise to a pattern of ones
143 * wherever there is a match. Since we have two bits per entry, we
144 * take this pattern, shift it down by one place and then and it with
145 * the original. All the even bit positions (0,2,4, etc) then represent
146 * successful matches, so we mask with 0x55555..... to remove the unwanted
149 * This allows searching of a whole u64 at once (32 blocks) with a
150 * single test (on 64 bit arches).
153 static inline u64
gfs2_bit_search(const __le64
*ptr
, u64 mask
, u8 state
)
156 static const u64 search
[] = {
157 [0] = 0xffffffffffffffffULL
,
158 [1] = 0xaaaaaaaaaaaaaaaaULL
,
159 [2] = 0x5555555555555555ULL
,
160 [3] = 0x0000000000000000ULL
,
162 tmp
= le64_to_cpu(*ptr
) ^ search
[state
];
169 * rs_cmp - multi-block reservation range compare
170 * @blk: absolute file system block number of the new reservation
171 * @len: number of blocks in the new reservation
172 * @rs: existing reservation to compare against
174 * returns: 1 if the block range is beyond the reach of the reservation
175 * -1 if the block range is before the start of the reservation
176 * 0 if the block range overlaps with the reservation
178 static inline int rs_cmp(u64 blk
, u32 len
, struct gfs2_blkreserv
*rs
)
180 u64 startblk
= gfs2_rbm_to_block(&rs
->rs_rbm
);
182 if (blk
>= startblk
+ rs
->rs_free
)
184 if (blk
+ len
- 1 < startblk
)
190 * gfs2_bitfit - Search an rgrp's bitmap buffer to find a bit-pair representing
191 * a block in a given allocation state.
192 * @buf: the buffer that holds the bitmaps
193 * @len: the length (in bytes) of the buffer
194 * @goal: start search at this block's bit-pair (within @buffer)
195 * @state: GFS2_BLKST_XXX the state of the block we're looking for.
197 * Scope of @goal and returned block number is only within this bitmap buffer,
198 * not entire rgrp or filesystem. @buffer will be offset from the actual
199 * beginning of a bitmap block buffer, skipping any header structures, but
200 * headers are always a multiple of 64 bits long so that the buffer is
201 * always aligned to a 64 bit boundary.
203 * The size of the buffer is in bytes, but is it assumed that it is
204 * always ok to read a complete multiple of 64 bits at the end
205 * of the block in case the end is no aligned to a natural boundary.
207 * Return: the block number (bitmap buffer scope) that was found
210 static u32
gfs2_bitfit(const u8
*buf
, const unsigned int len
,
213 u32 spoint
= (goal
<< 1) & ((8*sizeof(u64
)) - 1);
214 const __le64
*ptr
= ((__le64
*)buf
) + (goal
>> 5);
215 const __le64
*end
= (__le64
*)(buf
+ ALIGN(len
, sizeof(u64
)));
217 u64 mask
= 0x5555555555555555ULL
;
220 /* Mask off bits we don't care about at the start of the search */
222 tmp
= gfs2_bit_search(ptr
, mask
, state
);
224 while(tmp
== 0 && ptr
< end
) {
225 tmp
= gfs2_bit_search(ptr
, 0x5555555555555555ULL
, state
);
228 /* Mask off any bits which are more than len bytes from the start */
229 if (ptr
== end
&& (len
& (sizeof(u64
) - 1)))
230 tmp
&= (((u64
)~0) >> (64 - 8*(len
& (sizeof(u64
) - 1))));
231 /* Didn't find anything, so return */
236 bit
/= 2; /* two bits per entry in the bitmap */
237 return (((const unsigned char *)ptr
- buf
) * GFS2_NBBY
) + bit
;
241 * gfs2_rbm_from_block - Set the rbm based upon rgd and block number
242 * @rbm: The rbm with rgd already set correctly
243 * @block: The block number (filesystem relative)
245 * This sets the bi and offset members of an rbm based on a
246 * resource group and a filesystem relative block number. The
247 * resource group must be set in the rbm on entry, the bi and
248 * offset members will be set by this function.
250 * Returns: 0 on success, or an error code
253 static int gfs2_rbm_from_block(struct gfs2_rbm
*rbm
, u64 block
)
255 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
)
263 rbm
->offset
= (u32
)(rblock
);
264 /* Check if the block is within the first block */
265 if (rbm
->offset
< rbm_bi(rbm
)->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 rbm
->bii
= rbm
->offset
/ rbm
->rgd
->rd_sbd
->sd_blocks_per_bitmap
;
272 rbm
->offset
-= rbm
->bii
* rbm
->rgd
->rd_sbd
->sd_blocks_per_bitmap
;
277 * gfs2_rbm_incr - increment an rbm structure
278 * @rbm: The rbm with rgd already set correctly
280 * This function takes an existing rbm structure and increments it to the next
281 * viable block offset.
283 * Returns: If incrementing the offset would cause the rbm to go past the
284 * end of the rgrp, true is returned, otherwise false.
288 static bool gfs2_rbm_incr(struct gfs2_rbm
*rbm
)
290 if (rbm
->offset
+ 1 < rbm_bi(rbm
)->bi_blocks
) { /* in the same bitmap */
294 if (rbm
->bii
== rbm
->rgd
->rd_length
- 1) /* at the last bitmap */
303 * gfs2_unaligned_extlen - Look for free blocks which are not byte aligned
304 * @rbm: Position to search (value/result)
305 * @n_unaligned: Number of unaligned blocks to check
306 * @len: Decremented for each block found (terminate on zero)
308 * Returns: true if a non-free block is encountered
311 static bool gfs2_unaligned_extlen(struct gfs2_rbm
*rbm
, u32 n_unaligned
, u32
*len
)
316 for (n
= 0; n
< n_unaligned
; n
++) {
317 res
= gfs2_testbit(rbm
);
318 if (res
!= GFS2_BLKST_FREE
)
323 if (gfs2_rbm_incr(rbm
))
331 * gfs2_free_extlen - Return extent length of free blocks
332 * @rbm: Starting position
333 * @len: Max length to check
335 * Starting at the block specified by the rbm, see how many free blocks
336 * there are, not reading more than len blocks ahead. This can be done
337 * using memchr_inv when the blocks are byte aligned, but has to be done
338 * on a block by block basis in case of unaligned blocks. Also this
339 * function can cope with bitmap boundaries (although it must stop on
340 * a resource group boundary)
342 * Returns: Number of free blocks in the extent
345 static u32
gfs2_free_extlen(const struct gfs2_rbm
*rrbm
, u32 len
)
347 struct gfs2_rbm rbm
= *rrbm
;
348 u32 n_unaligned
= rbm
.offset
& 3;
352 u8
*ptr
, *start
, *end
;
354 struct gfs2_bitmap
*bi
;
357 gfs2_unaligned_extlen(&rbm
, 4 - n_unaligned
, &len
))
360 n_unaligned
= len
& 3;
361 /* Start is now byte aligned */
364 start
= bi
->bi_bh
->b_data
;
366 start
= bi
->bi_clone
;
367 end
= start
+ bi
->bi_bh
->b_size
;
368 start
+= bi
->bi_offset
;
369 BUG_ON(rbm
.offset
& 3);
370 start
+= (rbm
.offset
/ GFS2_NBBY
);
371 bytes
= min_t(u32
, len
/ GFS2_NBBY
, (end
- start
));
372 ptr
= memchr_inv(start
, 0, bytes
);
373 chunk_size
= ((ptr
== NULL
) ? bytes
: (ptr
- start
));
374 chunk_size
*= GFS2_NBBY
;
375 BUG_ON(len
< chunk_size
);
377 block
= gfs2_rbm_to_block(&rbm
);
378 if (gfs2_rbm_from_block(&rbm
, block
+ chunk_size
)) {
386 n_unaligned
= len
& 3;
389 /* Deal with any bits left over at the end */
391 gfs2_unaligned_extlen(&rbm
, n_unaligned
, &len
);
397 * gfs2_bitcount - count the number of bits in a certain state
398 * @rgd: the resource group descriptor
399 * @buffer: the buffer that holds the bitmaps
400 * @buflen: the length (in bytes) of the buffer
401 * @state: the state of the block we're looking for
403 * Returns: The number of bits
406 static u32
gfs2_bitcount(struct gfs2_rgrpd
*rgd
, const u8
*buffer
,
407 unsigned int buflen
, u8 state
)
409 const u8
*byte
= buffer
;
410 const u8
*end
= buffer
+ buflen
;
411 const u8 state1
= state
<< 2;
412 const u8 state2
= state
<< 4;
413 const u8 state3
= state
<< 6;
416 for (; byte
< end
; byte
++) {
417 if (((*byte
) & 0x03) == state
)
419 if (((*byte
) & 0x0C) == state1
)
421 if (((*byte
) & 0x30) == state2
)
423 if (((*byte
) & 0xC0) == state3
)
431 * gfs2_rgrp_verify - Verify that a resource group is consistent
436 void gfs2_rgrp_verify(struct gfs2_rgrpd
*rgd
)
438 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
439 struct gfs2_bitmap
*bi
= NULL
;
440 u32 length
= rgd
->rd_length
;
444 memset(count
, 0, 4 * sizeof(u32
));
446 /* Count # blocks in each of 4 possible allocation states */
447 for (buf
= 0; buf
< length
; buf
++) {
448 bi
= rgd
->rd_bits
+ buf
;
449 for (x
= 0; x
< 4; x
++)
450 count
[x
] += gfs2_bitcount(rgd
,
456 if (count
[0] != rgd
->rd_free
) {
457 if (gfs2_consist_rgrpd(rgd
))
458 fs_err(sdp
, "free data mismatch: %u != %u\n",
459 count
[0], rgd
->rd_free
);
463 tmp
= rgd
->rd_data
- rgd
->rd_free
- rgd
->rd_dinodes
;
464 if (count
[1] != tmp
) {
465 if (gfs2_consist_rgrpd(rgd
))
466 fs_err(sdp
, "used data mismatch: %u != %u\n",
471 if (count
[2] + count
[3] != rgd
->rd_dinodes
) {
472 if (gfs2_consist_rgrpd(rgd
))
473 fs_err(sdp
, "used metadata mismatch: %u != %u\n",
474 count
[2] + count
[3], rgd
->rd_dinodes
);
479 static inline int rgrp_contains_block(struct gfs2_rgrpd
*rgd
, u64 block
)
481 u64 first
= rgd
->rd_data0
;
482 u64 last
= first
+ rgd
->rd_data
;
483 return first
<= block
&& block
< last
;
487 * gfs2_blk2rgrpd - Find resource group for a given data/meta block number
488 * @sdp: The GFS2 superblock
489 * @blk: The data block number
490 * @exact: True if this needs to be an exact match
492 * Returns: The resource group, or NULL if not found
495 struct gfs2_rgrpd
*gfs2_blk2rgrpd(struct gfs2_sbd
*sdp
, u64 blk
, bool exact
)
497 struct rb_node
*n
, *next
;
498 struct gfs2_rgrpd
*cur
;
500 spin_lock(&sdp
->sd_rindex_spin
);
501 n
= sdp
->sd_rindex_tree
.rb_node
;
503 cur
= rb_entry(n
, struct gfs2_rgrpd
, rd_node
);
505 if (blk
< cur
->rd_addr
)
507 else if (blk
>= cur
->rd_data0
+ cur
->rd_data
)
510 spin_unlock(&sdp
->sd_rindex_spin
);
512 if (blk
< cur
->rd_addr
)
514 if (blk
>= cur
->rd_data0
+ cur
->rd_data
)
521 spin_unlock(&sdp
->sd_rindex_spin
);
527 * gfs2_rgrpd_get_first - get the first Resource Group in the filesystem
528 * @sdp: The GFS2 superblock
530 * Returns: The first rgrp in the filesystem
533 struct gfs2_rgrpd
*gfs2_rgrpd_get_first(struct gfs2_sbd
*sdp
)
535 const struct rb_node
*n
;
536 struct gfs2_rgrpd
*rgd
;
538 spin_lock(&sdp
->sd_rindex_spin
);
539 n
= rb_first(&sdp
->sd_rindex_tree
);
540 rgd
= rb_entry(n
, struct gfs2_rgrpd
, rd_node
);
541 spin_unlock(&sdp
->sd_rindex_spin
);
547 * gfs2_rgrpd_get_next - get the next RG
548 * @rgd: the resource group descriptor
550 * Returns: The next rgrp
553 struct gfs2_rgrpd
*gfs2_rgrpd_get_next(struct gfs2_rgrpd
*rgd
)
555 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
556 const struct rb_node
*n
;
558 spin_lock(&sdp
->sd_rindex_spin
);
559 n
= rb_next(&rgd
->rd_node
);
561 n
= rb_first(&sdp
->sd_rindex_tree
);
563 if (unlikely(&rgd
->rd_node
== n
)) {
564 spin_unlock(&sdp
->sd_rindex_spin
);
567 rgd
= rb_entry(n
, struct gfs2_rgrpd
, rd_node
);
568 spin_unlock(&sdp
->sd_rindex_spin
);
572 void gfs2_free_clones(struct gfs2_rgrpd
*rgd
)
576 for (x
= 0; x
< rgd
->rd_length
; x
++) {
577 struct gfs2_bitmap
*bi
= rgd
->rd_bits
+ x
;
584 * gfs2_rs_alloc - make sure we have a reservation assigned to the inode
585 * @ip: the inode for this reservation
587 int gfs2_rs_alloc(struct gfs2_inode
*ip
)
591 down_write(&ip
->i_rw_mutex
);
595 ip
->i_res
= kmem_cache_zalloc(gfs2_rsrv_cachep
, GFP_NOFS
);
601 RB_CLEAR_NODE(&ip
->i_res
->rs_node
);
603 up_write(&ip
->i_rw_mutex
);
607 static void dump_rs(struct seq_file
*seq
, const struct gfs2_blkreserv
*rs
)
609 gfs2_print_dbg(seq
, " B: n:%llu s:%llu b:%u f:%u\n",
610 (unsigned long long)rs
->rs_inum
,
611 (unsigned long long)gfs2_rbm_to_block(&rs
->rs_rbm
),
612 rs
->rs_rbm
.offset
, rs
->rs_free
);
616 * __rs_deltree - remove a multi-block reservation from the rgd tree
617 * @rs: The reservation to remove
620 static void __rs_deltree(struct gfs2_blkreserv
*rs
)
622 struct gfs2_rgrpd
*rgd
;
624 if (!gfs2_rs_active(rs
))
627 rgd
= rs
->rs_rbm
.rgd
;
628 trace_gfs2_rs(rs
, TRACE_RS_TREEDEL
);
629 rb_erase(&rs
->rs_node
, &rgd
->rd_rstree
);
630 RB_CLEAR_NODE(&rs
->rs_node
);
633 struct gfs2_bitmap
*bi
= rbm_bi(&rs
->rs_rbm
);
635 /* return reserved blocks to the rgrp */
636 BUG_ON(rs
->rs_rbm
.rgd
->rd_reserved
< rs
->rs_free
);
637 rs
->rs_rbm
.rgd
->rd_reserved
-= rs
->rs_free
;
639 clear_bit(GBF_FULL
, &bi
->bi_flags
);
640 smp_mb__after_clear_bit();
645 * gfs2_rs_deltree - remove a multi-block reservation from the rgd tree
646 * @rs: The reservation to remove
649 void gfs2_rs_deltree(struct gfs2_blkreserv
*rs
)
651 struct gfs2_rgrpd
*rgd
;
653 rgd
= rs
->rs_rbm
.rgd
;
655 spin_lock(&rgd
->rd_rsspin
);
657 spin_unlock(&rgd
->rd_rsspin
);
662 * gfs2_rs_delete - delete a multi-block reservation
663 * @ip: The inode for this reservation
666 void gfs2_rs_delete(struct gfs2_inode
*ip
)
668 struct inode
*inode
= &ip
->i_inode
;
670 down_write(&ip
->i_rw_mutex
);
671 if (ip
->i_res
&& atomic_read(&inode
->i_writecount
) <= 1) {
672 gfs2_rs_deltree(ip
->i_res
);
673 BUG_ON(ip
->i_res
->rs_free
);
674 kmem_cache_free(gfs2_rsrv_cachep
, ip
->i_res
);
677 up_write(&ip
->i_rw_mutex
);
681 * return_all_reservations - return all reserved blocks back to the rgrp.
682 * @rgd: the rgrp that needs its space back
684 * We previously reserved a bunch of blocks for allocation. Now we need to
685 * give them back. This leave the reservation structures in tact, but removes
686 * all of their corresponding "no-fly zones".
688 static void return_all_reservations(struct gfs2_rgrpd
*rgd
)
691 struct gfs2_blkreserv
*rs
;
693 spin_lock(&rgd
->rd_rsspin
);
694 while ((n
= rb_first(&rgd
->rd_rstree
))) {
695 rs
= rb_entry(n
, struct gfs2_blkreserv
, rs_node
);
698 spin_unlock(&rgd
->rd_rsspin
);
701 void gfs2_clear_rgrpd(struct gfs2_sbd
*sdp
)
704 struct gfs2_rgrpd
*rgd
;
705 struct gfs2_glock
*gl
;
707 while ((n
= rb_first(&sdp
->sd_rindex_tree
))) {
708 rgd
= rb_entry(n
, struct gfs2_rgrpd
, rd_node
);
711 rb_erase(n
, &sdp
->sd_rindex_tree
);
714 spin_lock(&gl
->gl_spin
);
715 gl
->gl_object
= NULL
;
716 spin_unlock(&gl
->gl_spin
);
717 gfs2_glock_add_to_lru(gl
);
721 gfs2_free_clones(rgd
);
723 return_all_reservations(rgd
);
724 kmem_cache_free(gfs2_rgrpd_cachep
, rgd
);
728 static void gfs2_rindex_print(const struct gfs2_rgrpd
*rgd
)
730 printk(KERN_INFO
" ri_addr = %llu\n", (unsigned long long)rgd
->rd_addr
);
731 printk(KERN_INFO
" ri_length = %u\n", rgd
->rd_length
);
732 printk(KERN_INFO
" ri_data0 = %llu\n", (unsigned long long)rgd
->rd_data0
);
733 printk(KERN_INFO
" ri_data = %u\n", rgd
->rd_data
);
734 printk(KERN_INFO
" ri_bitbytes = %u\n", rgd
->rd_bitbytes
);
738 * gfs2_compute_bitstructs - Compute the bitmap sizes
739 * @rgd: The resource group descriptor
741 * Calculates bitmap descriptors, one for each block that contains bitmap data
746 static int compute_bitstructs(struct gfs2_rgrpd
*rgd
)
748 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
749 struct gfs2_bitmap
*bi
;
750 u32 length
= rgd
->rd_length
; /* # blocks in hdr & bitmap */
751 u32 bytes_left
, bytes
;
757 rgd
->rd_bits
= kcalloc(length
, sizeof(struct gfs2_bitmap
), GFP_NOFS
);
761 bytes_left
= rgd
->rd_bitbytes
;
763 for (x
= 0; x
< length
; x
++) {
764 bi
= rgd
->rd_bits
+ x
;
767 /* small rgrp; bitmap stored completely in header block */
770 bi
->bi_offset
= sizeof(struct gfs2_rgrp
);
773 bi
->bi_blocks
= bytes
* GFS2_NBBY
;
776 bytes
= sdp
->sd_sb
.sb_bsize
- sizeof(struct gfs2_rgrp
);
777 bi
->bi_offset
= sizeof(struct gfs2_rgrp
);
780 bi
->bi_blocks
= bytes
* GFS2_NBBY
;
782 } else if (x
+ 1 == length
) {
784 bi
->bi_offset
= sizeof(struct gfs2_meta_header
);
785 bi
->bi_start
= rgd
->rd_bitbytes
- bytes_left
;
787 bi
->bi_blocks
= bytes
* GFS2_NBBY
;
790 bytes
= sdp
->sd_sb
.sb_bsize
-
791 sizeof(struct gfs2_meta_header
);
792 bi
->bi_offset
= sizeof(struct gfs2_meta_header
);
793 bi
->bi_start
= rgd
->rd_bitbytes
- bytes_left
;
795 bi
->bi_blocks
= bytes
* GFS2_NBBY
;
802 gfs2_consist_rgrpd(rgd
);
805 bi
= rgd
->rd_bits
+ (length
- 1);
806 if ((bi
->bi_start
+ bi
->bi_len
) * GFS2_NBBY
!= rgd
->rd_data
) {
807 if (gfs2_consist_rgrpd(rgd
)) {
808 gfs2_rindex_print(rgd
);
809 fs_err(sdp
, "start=%u len=%u offset=%u\n",
810 bi
->bi_start
, bi
->bi_len
, bi
->bi_offset
);
819 * gfs2_ri_total - Total up the file system space, according to the rindex.
820 * @sdp: the filesystem
823 u64
gfs2_ri_total(struct gfs2_sbd
*sdp
)
826 struct inode
*inode
= sdp
->sd_rindex
;
827 struct gfs2_inode
*ip
= GFS2_I(inode
);
828 char buf
[sizeof(struct gfs2_rindex
)];
831 for (rgrps
= 0;; rgrps
++) {
832 loff_t pos
= rgrps
* sizeof(struct gfs2_rindex
);
834 if (pos
+ sizeof(struct gfs2_rindex
) > i_size_read(inode
))
836 error
= gfs2_internal_read(ip
, buf
, &pos
,
837 sizeof(struct gfs2_rindex
));
838 if (error
!= sizeof(struct gfs2_rindex
))
840 total_data
+= be32_to_cpu(((struct gfs2_rindex
*)buf
)->ri_data
);
845 static int rgd_insert(struct gfs2_rgrpd
*rgd
)
847 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
848 struct rb_node
**newn
= &sdp
->sd_rindex_tree
.rb_node
, *parent
= NULL
;
850 /* Figure out where to put new node */
852 struct gfs2_rgrpd
*cur
= rb_entry(*newn
, struct gfs2_rgrpd
,
856 if (rgd
->rd_addr
< cur
->rd_addr
)
857 newn
= &((*newn
)->rb_left
);
858 else if (rgd
->rd_addr
> cur
->rd_addr
)
859 newn
= &((*newn
)->rb_right
);
864 rb_link_node(&rgd
->rd_node
, parent
, newn
);
865 rb_insert_color(&rgd
->rd_node
, &sdp
->sd_rindex_tree
);
871 * read_rindex_entry - Pull in a new resource index entry from the disk
872 * @ip: Pointer to the rindex inode
874 * Returns: 0 on success, > 0 on EOF, error code otherwise
877 static int read_rindex_entry(struct gfs2_inode
*ip
)
879 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
880 loff_t pos
= sdp
->sd_rgrps
* sizeof(struct gfs2_rindex
);
881 struct gfs2_rindex buf
;
883 struct gfs2_rgrpd
*rgd
;
885 if (pos
>= i_size_read(&ip
->i_inode
))
888 error
= gfs2_internal_read(ip
, (char *)&buf
, &pos
,
889 sizeof(struct gfs2_rindex
));
891 if (error
!= sizeof(struct gfs2_rindex
))
892 return (error
== 0) ? 1 : error
;
894 rgd
= kmem_cache_zalloc(gfs2_rgrpd_cachep
, GFP_NOFS
);
900 rgd
->rd_addr
= be64_to_cpu(buf
.ri_addr
);
901 rgd
->rd_length
= be32_to_cpu(buf
.ri_length
);
902 rgd
->rd_data0
= be64_to_cpu(buf
.ri_data0
);
903 rgd
->rd_data
= be32_to_cpu(buf
.ri_data
);
904 rgd
->rd_bitbytes
= be32_to_cpu(buf
.ri_bitbytes
);
905 spin_lock_init(&rgd
->rd_rsspin
);
907 error
= compute_bitstructs(rgd
);
911 error
= gfs2_glock_get(sdp
, rgd
->rd_addr
,
912 &gfs2_rgrp_glops
, CREATE
, &rgd
->rd_gl
);
916 rgd
->rd_gl
->gl_object
= rgd
;
917 rgd
->rd_rgl
= (struct gfs2_rgrp_lvb
*)rgd
->rd_gl
->gl_lksb
.sb_lvbptr
;
918 rgd
->rd_flags
&= ~GFS2_RDF_UPTODATE
;
919 if (rgd
->rd_data
> sdp
->sd_max_rg_data
)
920 sdp
->sd_max_rg_data
= rgd
->rd_data
;
921 spin_lock(&sdp
->sd_rindex_spin
);
922 error
= rgd_insert(rgd
);
923 spin_unlock(&sdp
->sd_rindex_spin
);
927 error
= 0; /* someone else read in the rgrp; free it and ignore it */
928 gfs2_glock_put(rgd
->rd_gl
);
932 kmem_cache_free(gfs2_rgrpd_cachep
, rgd
);
937 * gfs2_ri_update - Pull in a new resource index from the disk
938 * @ip: pointer to the rindex inode
940 * Returns: 0 on successful update, error code otherwise
943 static int gfs2_ri_update(struct gfs2_inode
*ip
)
945 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
949 error
= read_rindex_entry(ip
);
950 } while (error
== 0);
955 sdp
->sd_rindex_uptodate
= 1;
960 * gfs2_rindex_update - Update the rindex if required
961 * @sdp: The GFS2 superblock
963 * We grab a lock on the rindex inode to make sure that it doesn't
964 * change whilst we are performing an operation. We keep this lock
965 * for quite long periods of time compared to other locks. This
966 * doesn't matter, since it is shared and it is very, very rarely
967 * accessed in the exclusive mode (i.e. only when expanding the filesystem).
969 * This makes sure that we're using the latest copy of the resource index
970 * special file, which might have been updated if someone expanded the
971 * filesystem (via gfs2_grow utility), which adds new resource groups.
973 * Returns: 0 on succeess, error code otherwise
976 int gfs2_rindex_update(struct gfs2_sbd
*sdp
)
978 struct gfs2_inode
*ip
= GFS2_I(sdp
->sd_rindex
);
979 struct gfs2_glock
*gl
= ip
->i_gl
;
980 struct gfs2_holder ri_gh
;
982 int unlock_required
= 0;
984 /* Read new copy from disk if we don't have the latest */
985 if (!sdp
->sd_rindex_uptodate
) {
986 if (!gfs2_glock_is_locked_by_me(gl
)) {
987 error
= gfs2_glock_nq_init(gl
, LM_ST_SHARED
, 0, &ri_gh
);
992 if (!sdp
->sd_rindex_uptodate
)
993 error
= gfs2_ri_update(ip
);
995 gfs2_glock_dq_uninit(&ri_gh
);
1001 static void gfs2_rgrp_in(struct gfs2_rgrpd
*rgd
, const void *buf
)
1003 const struct gfs2_rgrp
*str
= buf
;
1006 rg_flags
= be32_to_cpu(str
->rg_flags
);
1007 rg_flags
&= ~GFS2_RDF_MASK
;
1008 rgd
->rd_flags
&= GFS2_RDF_MASK
;
1009 rgd
->rd_flags
|= rg_flags
;
1010 rgd
->rd_free
= be32_to_cpu(str
->rg_free
);
1011 rgd
->rd_dinodes
= be32_to_cpu(str
->rg_dinodes
);
1012 rgd
->rd_igeneration
= be64_to_cpu(str
->rg_igeneration
);
1015 static void gfs2_rgrp_out(struct gfs2_rgrpd
*rgd
, void *buf
)
1017 struct gfs2_rgrp
*str
= buf
;
1019 str
->rg_flags
= cpu_to_be32(rgd
->rd_flags
& ~GFS2_RDF_MASK
);
1020 str
->rg_free
= cpu_to_be32(rgd
->rd_free
);
1021 str
->rg_dinodes
= cpu_to_be32(rgd
->rd_dinodes
);
1022 str
->__pad
= cpu_to_be32(0);
1023 str
->rg_igeneration
= cpu_to_be64(rgd
->rd_igeneration
);
1024 memset(&str
->rg_reserved
, 0, sizeof(str
->rg_reserved
));
1027 static int gfs2_rgrp_lvb_valid(struct gfs2_rgrpd
*rgd
)
1029 struct gfs2_rgrp_lvb
*rgl
= rgd
->rd_rgl
;
1030 struct gfs2_rgrp
*str
= (struct gfs2_rgrp
*)rgd
->rd_bits
[0].bi_bh
->b_data
;
1032 if (rgl
->rl_flags
!= str
->rg_flags
|| rgl
->rl_free
!= str
->rg_free
||
1033 rgl
->rl_dinodes
!= str
->rg_dinodes
||
1034 rgl
->rl_igeneration
!= str
->rg_igeneration
)
1039 static void gfs2_rgrp_ondisk2lvb(struct gfs2_rgrp_lvb
*rgl
, const void *buf
)
1041 const struct gfs2_rgrp
*str
= buf
;
1043 rgl
->rl_magic
= cpu_to_be32(GFS2_MAGIC
);
1044 rgl
->rl_flags
= str
->rg_flags
;
1045 rgl
->rl_free
= str
->rg_free
;
1046 rgl
->rl_dinodes
= str
->rg_dinodes
;
1047 rgl
->rl_igeneration
= str
->rg_igeneration
;
1051 static void update_rgrp_lvb_unlinked(struct gfs2_rgrpd
*rgd
, u32 change
)
1053 struct gfs2_rgrp_lvb
*rgl
= rgd
->rd_rgl
;
1054 u32 unlinked
= be32_to_cpu(rgl
->rl_unlinked
) + change
;
1055 rgl
->rl_unlinked
= cpu_to_be32(unlinked
);
1058 static u32
count_unlinked(struct gfs2_rgrpd
*rgd
)
1060 struct gfs2_bitmap
*bi
;
1061 const u32 length
= rgd
->rd_length
;
1062 const u8
*buffer
= NULL
;
1063 u32 i
, goal
, count
= 0;
1065 for (i
= 0, bi
= rgd
->rd_bits
; i
< length
; i
++, bi
++) {
1067 buffer
= bi
->bi_bh
->b_data
+ bi
->bi_offset
;
1068 WARN_ON(!buffer_uptodate(bi
->bi_bh
));
1069 while (goal
< bi
->bi_len
* GFS2_NBBY
) {
1070 goal
= gfs2_bitfit(buffer
, bi
->bi_len
, goal
,
1071 GFS2_BLKST_UNLINKED
);
1072 if (goal
== BFITNOENT
)
1084 * gfs2_rgrp_bh_get - Read in a RG's header and bitmaps
1085 * @rgd: the struct gfs2_rgrpd describing the RG to read in
1087 * Read in all of a Resource Group's header and bitmap blocks.
1088 * Caller must eventually call gfs2_rgrp_relse() to free the bitmaps.
1093 int gfs2_rgrp_bh_get(struct gfs2_rgrpd
*rgd
)
1095 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
1096 struct gfs2_glock
*gl
= rgd
->rd_gl
;
1097 unsigned int length
= rgd
->rd_length
;
1098 struct gfs2_bitmap
*bi
;
1102 if (rgd
->rd_bits
[0].bi_bh
!= NULL
)
1105 for (x
= 0; x
< length
; x
++) {
1106 bi
= rgd
->rd_bits
+ x
;
1107 error
= gfs2_meta_read(gl
, rgd
->rd_addr
+ x
, 0, &bi
->bi_bh
);
1112 for (y
= length
; y
--;) {
1113 bi
= rgd
->rd_bits
+ y
;
1114 error
= gfs2_meta_wait(sdp
, bi
->bi_bh
);
1117 if (gfs2_metatype_check(sdp
, bi
->bi_bh
, y
? GFS2_METATYPE_RB
:
1118 GFS2_METATYPE_RG
)) {
1124 if (!(rgd
->rd_flags
& GFS2_RDF_UPTODATE
)) {
1125 for (x
= 0; x
< length
; x
++)
1126 clear_bit(GBF_FULL
, &rgd
->rd_bits
[x
].bi_flags
);
1127 gfs2_rgrp_in(rgd
, (rgd
->rd_bits
[0].bi_bh
)->b_data
);
1128 rgd
->rd_flags
|= (GFS2_RDF_UPTODATE
| GFS2_RDF_CHECK
);
1129 rgd
->rd_free_clone
= rgd
->rd_free
;
1131 if (be32_to_cpu(GFS2_MAGIC
) != rgd
->rd_rgl
->rl_magic
) {
1132 rgd
->rd_rgl
->rl_unlinked
= cpu_to_be32(count_unlinked(rgd
));
1133 gfs2_rgrp_ondisk2lvb(rgd
->rd_rgl
,
1134 rgd
->rd_bits
[0].bi_bh
->b_data
);
1136 else if (sdp
->sd_args
.ar_rgrplvb
) {
1137 if (!gfs2_rgrp_lvb_valid(rgd
)){
1138 gfs2_consist_rgrpd(rgd
);
1142 if (rgd
->rd_rgl
->rl_unlinked
== 0)
1143 rgd
->rd_flags
&= ~GFS2_RDF_CHECK
;
1149 bi
= rgd
->rd_bits
+ x
;
1152 gfs2_assert_warn(sdp
, !bi
->bi_clone
);
1158 int update_rgrp_lvb(struct gfs2_rgrpd
*rgd
)
1162 if (rgd
->rd_flags
& GFS2_RDF_UPTODATE
)
1165 if (be32_to_cpu(GFS2_MAGIC
) != rgd
->rd_rgl
->rl_magic
)
1166 return gfs2_rgrp_bh_get(rgd
);
1168 rl_flags
= be32_to_cpu(rgd
->rd_rgl
->rl_flags
);
1169 rl_flags
&= ~GFS2_RDF_MASK
;
1170 rgd
->rd_flags
&= GFS2_RDF_MASK
;
1171 rgd
->rd_flags
|= (rl_flags
| GFS2_RDF_UPTODATE
| GFS2_RDF_CHECK
);
1172 if (rgd
->rd_rgl
->rl_unlinked
== 0)
1173 rgd
->rd_flags
&= ~GFS2_RDF_CHECK
;
1174 rgd
->rd_free
= be32_to_cpu(rgd
->rd_rgl
->rl_free
);
1175 rgd
->rd_free_clone
= rgd
->rd_free
;
1176 rgd
->rd_dinodes
= be32_to_cpu(rgd
->rd_rgl
->rl_dinodes
);
1177 rgd
->rd_igeneration
= be64_to_cpu(rgd
->rd_rgl
->rl_igeneration
);
1181 int gfs2_rgrp_go_lock(struct gfs2_holder
*gh
)
1183 struct gfs2_rgrpd
*rgd
= gh
->gh_gl
->gl_object
;
1184 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
1186 if (gh
->gh_flags
& GL_SKIP
&& sdp
->sd_args
.ar_rgrplvb
)
1188 return gfs2_rgrp_bh_get((struct gfs2_rgrpd
*)gh
->gh_gl
->gl_object
);
1192 * gfs2_rgrp_go_unlock - Release RG bitmaps read in with gfs2_rgrp_bh_get()
1193 * @gh: The glock holder for the resource group
1197 void gfs2_rgrp_go_unlock(struct gfs2_holder
*gh
)
1199 struct gfs2_rgrpd
*rgd
= gh
->gh_gl
->gl_object
;
1200 int x
, length
= rgd
->rd_length
;
1202 for (x
= 0; x
< length
; x
++) {
1203 struct gfs2_bitmap
*bi
= rgd
->rd_bits
+ x
;
1212 int gfs2_rgrp_send_discards(struct gfs2_sbd
*sdp
, u64 offset
,
1213 struct buffer_head
*bh
,
1214 const struct gfs2_bitmap
*bi
, unsigned minlen
, u64
*ptrimmed
)
1216 struct super_block
*sb
= sdp
->sd_vfs
;
1219 sector_t nr_blks
= 0;
1225 for (x
= 0; x
< bi
->bi_len
; x
++) {
1226 const u8
*clone
= bi
->bi_clone
? bi
->bi_clone
: bi
->bi_bh
->b_data
;
1227 clone
+= bi
->bi_offset
;
1230 const u8
*orig
= bh
->b_data
+ bi
->bi_offset
+ x
;
1231 diff
= ~(*orig
| (*orig
>> 1)) & (*clone
| (*clone
>> 1));
1233 diff
= ~(*clone
| (*clone
>> 1));
1238 blk
= offset
+ ((bi
->bi_start
+ x
) * GFS2_NBBY
);
1242 goto start_new_extent
;
1243 if ((start
+ nr_blks
) != blk
) {
1244 if (nr_blks
>= minlen
) {
1245 rv
= sb_issue_discard(sb
,
1262 if (nr_blks
>= minlen
) {
1263 rv
= sb_issue_discard(sb
, start
, nr_blks
, GFP_NOFS
, 0);
1269 *ptrimmed
= trimmed
;
1273 if (sdp
->sd_args
.ar_discard
)
1274 fs_warn(sdp
, "error %d on discard request, turning discards off for this filesystem", rv
);
1275 sdp
->sd_args
.ar_discard
= 0;
1280 * gfs2_fitrim - Generate discard requests for unused bits of the filesystem
1281 * @filp: Any file on the filesystem
1282 * @argp: Pointer to the arguments (also used to pass result)
1284 * Returns: 0 on success, otherwise error code
1287 int gfs2_fitrim(struct file
*filp
, void __user
*argp
)
1289 struct inode
*inode
= file_inode(filp
);
1290 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
1291 struct request_queue
*q
= bdev_get_queue(sdp
->sd_vfs
->s_bdev
);
1292 struct buffer_head
*bh
;
1293 struct gfs2_rgrpd
*rgd
;
1294 struct gfs2_rgrpd
*rgd_end
;
1295 struct gfs2_holder gh
;
1296 struct fstrim_range r
;
1300 u64 start
, end
, minlen
;
1302 unsigned bs_shift
= sdp
->sd_sb
.sb_bsize_shift
;
1304 if (!capable(CAP_SYS_ADMIN
))
1307 if (!blk_queue_discard(q
))
1310 if (copy_from_user(&r
, argp
, sizeof(r
)))
1313 ret
= gfs2_rindex_update(sdp
);
1317 start
= r
.start
>> bs_shift
;
1318 end
= start
+ (r
.len
>> bs_shift
);
1319 minlen
= max_t(u64
, r
.minlen
,
1320 q
->limits
.discard_granularity
) >> bs_shift
;
1322 if (end
<= start
|| minlen
> sdp
->sd_max_rg_data
)
1325 rgd
= gfs2_blk2rgrpd(sdp
, start
, 0);
1326 rgd_end
= gfs2_blk2rgrpd(sdp
, end
, 0);
1328 if ((gfs2_rgrpd_get_first(sdp
) == gfs2_rgrpd_get_next(rgd_end
))
1329 && (start
> rgd_end
->rd_data0
+ rgd_end
->rd_data
))
1330 return -EINVAL
; /* start is beyond the end of the fs */
1334 ret
= gfs2_glock_nq_init(rgd
->rd_gl
, LM_ST_EXCLUSIVE
, 0, &gh
);
1338 if (!(rgd
->rd_flags
& GFS2_RGF_TRIMMED
)) {
1339 /* Trim each bitmap in the rgrp */
1340 for (x
= 0; x
< rgd
->rd_length
; x
++) {
1341 struct gfs2_bitmap
*bi
= rgd
->rd_bits
+ x
;
1342 ret
= gfs2_rgrp_send_discards(sdp
,
1343 rgd
->rd_data0
, NULL
, bi
, minlen
,
1346 gfs2_glock_dq_uninit(&gh
);
1352 /* Mark rgrp as having been trimmed */
1353 ret
= gfs2_trans_begin(sdp
, RES_RG_HDR
, 0);
1355 bh
= rgd
->rd_bits
[0].bi_bh
;
1356 rgd
->rd_flags
|= GFS2_RGF_TRIMMED
;
1357 gfs2_trans_add_meta(rgd
->rd_gl
, bh
);
1358 gfs2_rgrp_out(rgd
, bh
->b_data
);
1359 gfs2_rgrp_ondisk2lvb(rgd
->rd_rgl
, bh
->b_data
);
1360 gfs2_trans_end(sdp
);
1363 gfs2_glock_dq_uninit(&gh
);
1368 rgd
= gfs2_rgrpd_get_next(rgd
);
1372 r
.len
= trimmed
<< bs_shift
;
1373 if (copy_to_user(argp
, &r
, sizeof(r
)))
1380 * rs_insert - insert a new multi-block reservation into the rgrp's rb_tree
1381 * @ip: the inode structure
1384 static void rs_insert(struct gfs2_inode
*ip
)
1386 struct rb_node
**newn
, *parent
= NULL
;
1388 struct gfs2_blkreserv
*rs
= ip
->i_res
;
1389 struct gfs2_rgrpd
*rgd
= rs
->rs_rbm
.rgd
;
1390 u64 fsblock
= gfs2_rbm_to_block(&rs
->rs_rbm
);
1392 BUG_ON(gfs2_rs_active(rs
));
1394 spin_lock(&rgd
->rd_rsspin
);
1395 newn
= &rgd
->rd_rstree
.rb_node
;
1397 struct gfs2_blkreserv
*cur
=
1398 rb_entry(*newn
, struct gfs2_blkreserv
, rs_node
);
1401 rc
= rs_cmp(fsblock
, rs
->rs_free
, cur
);
1403 newn
= &((*newn
)->rb_right
);
1405 newn
= &((*newn
)->rb_left
);
1407 spin_unlock(&rgd
->rd_rsspin
);
1413 rb_link_node(&rs
->rs_node
, parent
, newn
);
1414 rb_insert_color(&rs
->rs_node
, &rgd
->rd_rstree
);
1416 /* Do our rgrp accounting for the reservation */
1417 rgd
->rd_reserved
+= rs
->rs_free
; /* blocks reserved */
1418 spin_unlock(&rgd
->rd_rsspin
);
1419 trace_gfs2_rs(rs
, TRACE_RS_INSERT
);
1423 * rg_mblk_search - find a group of multiple free blocks to form a reservation
1424 * @rgd: the resource group descriptor
1425 * @ip: pointer to the inode for which we're reserving blocks
1426 * @requested: number of blocks required for this allocation
1430 static void rg_mblk_search(struct gfs2_rgrpd
*rgd
, struct gfs2_inode
*ip
,
1433 struct gfs2_rbm rbm
= { .rgd
= rgd
, };
1435 struct gfs2_blkreserv
*rs
= ip
->i_res
;
1437 u32 free_blocks
= rgd
->rd_free_clone
- rgd
->rd_reserved
;
1439 struct inode
*inode
= &ip
->i_inode
;
1441 if (S_ISDIR(inode
->i_mode
))
1444 extlen
= max_t(u32
, atomic_read(&rs
->rs_sizehint
), requested
);
1445 extlen
= clamp(extlen
, RGRP_RSRV_MINBLKS
, free_blocks
);
1447 if ((rgd
->rd_free_clone
< rgd
->rd_reserved
) || (free_blocks
< extlen
))
1450 /* Find bitmap block that contains bits for goal block */
1451 if (rgrp_contains_block(rgd
, ip
->i_goal
))
1454 goal
= rgd
->rd_last_alloc
+ rgd
->rd_data0
;
1456 if (WARN_ON(gfs2_rbm_from_block(&rbm
, goal
)))
1459 ret
= gfs2_rbm_find(&rbm
, GFS2_BLKST_FREE
, extlen
, ip
, true);
1462 rs
->rs_free
= extlen
;
1463 rs
->rs_inum
= ip
->i_no_addr
;
1466 if (goal
== rgd
->rd_last_alloc
+ rgd
->rd_data0
)
1467 rgd
->rd_last_alloc
= 0;
1472 * gfs2_next_unreserved_block - Return next block that is not reserved
1473 * @rgd: The resource group
1474 * @block: The starting block
1475 * @length: The required length
1476 * @ip: Ignore any reservations for this inode
1478 * If the block does not appear in any reservation, then return the
1479 * block number unchanged. If it does appear in the reservation, then
1480 * keep looking through the tree of reservations in order to find the
1481 * first block number which is not reserved.
1484 static u64
gfs2_next_unreserved_block(struct gfs2_rgrpd
*rgd
, u64 block
,
1486 const struct gfs2_inode
*ip
)
1488 struct gfs2_blkreserv
*rs
;
1492 spin_lock(&rgd
->rd_rsspin
);
1493 n
= rgd
->rd_rstree
.rb_node
;
1495 rs
= rb_entry(n
, struct gfs2_blkreserv
, rs_node
);
1496 rc
= rs_cmp(block
, length
, rs
);
1506 while ((rs_cmp(block
, length
, rs
) == 0) && (ip
->i_res
!= rs
)) {
1507 block
= gfs2_rbm_to_block(&rs
->rs_rbm
) + rs
->rs_free
;
1511 rs
= rb_entry(n
, struct gfs2_blkreserv
, rs_node
);
1515 spin_unlock(&rgd
->rd_rsspin
);
1520 * gfs2_reservation_check_and_update - Check for reservations during block alloc
1521 * @rbm: The current position in the resource group
1522 * @ip: The inode for which we are searching for blocks
1523 * @minext: The minimum extent length
1525 * This checks the current position in the rgrp to see whether there is
1526 * a reservation covering this block. If not then this function is a
1527 * no-op. If there is, then the position is moved to the end of the
1528 * contiguous reservation(s) so that we are pointing at the first
1529 * non-reserved block.
1531 * Returns: 0 if no reservation, 1 if @rbm has changed, otherwise an error
1534 static int gfs2_reservation_check_and_update(struct gfs2_rbm
*rbm
,
1535 const struct gfs2_inode
*ip
,
1538 u64 block
= gfs2_rbm_to_block(rbm
);
1544 * If we have a minimum extent length, then skip over any extent
1545 * which is less than the min extent length in size.
1548 extlen
= gfs2_free_extlen(rbm
, minext
);
1549 nblock
= block
+ extlen
;
1550 if (extlen
< minext
)
1555 * Check the extent which has been found against the reservations
1556 * and skip if parts of it are already reserved
1558 nblock
= gfs2_next_unreserved_block(rbm
->rgd
, block
, extlen
, ip
);
1559 if (nblock
== block
)
1562 ret
= gfs2_rbm_from_block(rbm
, nblock
);
1569 * gfs2_rbm_find - Look for blocks of a particular state
1570 * @rbm: Value/result starting position and final position
1571 * @state: The state which we want to find
1572 * @minext: The requested extent length (0 for a single block)
1573 * @ip: If set, check for reservations
1574 * @nowrap: Stop looking at the end of the rgrp, rather than wrapping
1575 * around until we've reached the starting point.
1578 * - If looking for free blocks, we set GBF_FULL on each bitmap which
1579 * has no free blocks in it.
1581 * Returns: 0 on success, -ENOSPC if there is no block of the requested state
1584 static int gfs2_rbm_find(struct gfs2_rbm
*rbm
, u8 state
, u32 minext
,
1585 const struct gfs2_inode
*ip
, bool nowrap
)
1587 struct buffer_head
*bh
;
1593 int iters
= rbm
->rgd
->rd_length
;
1595 struct gfs2_bitmap
*bi
;
1597 /* If we are not starting at the beginning of a bitmap, then we
1598 * need to add one to the bitmap count to ensure that we search
1599 * the starting bitmap twice.
1601 if (rbm
->offset
!= 0)
1606 if (test_bit(GBF_FULL
, &bi
->bi_flags
) &&
1607 (state
== GFS2_BLKST_FREE
))
1611 buffer
= bh
->b_data
+ bi
->bi_offset
;
1612 WARN_ON(!buffer_uptodate(bh
));
1613 if (state
!= GFS2_BLKST_UNLINKED
&& bi
->bi_clone
)
1614 buffer
= bi
->bi_clone
+ bi
->bi_offset
;
1615 initial_offset
= rbm
->offset
;
1616 offset
= gfs2_bitfit(buffer
, bi
->bi_len
, rbm
->offset
, state
);
1617 if (offset
== BFITNOENT
)
1619 rbm
->offset
= offset
;
1623 initial_bii
= rbm
->bii
;
1624 ret
= gfs2_reservation_check_and_update(rbm
, ip
, minext
);
1628 n
+= (rbm
->bii
- initial_bii
);
1631 if (ret
== -E2BIG
) {
1634 n
+= (rbm
->bii
- initial_bii
);
1635 goto res_covered_end_of_rgrp
;
1639 bitmap_full
: /* Mark bitmap as full and fall through */
1640 if ((state
== GFS2_BLKST_FREE
) && initial_offset
== 0) {
1641 struct gfs2_bitmap
*bi
= rbm_bi(rbm
);
1642 set_bit(GBF_FULL
, &bi
->bi_flags
);
1645 next_bitmap
: /* Find next bitmap in the rgrp */
1648 if (rbm
->bii
== rbm
->rgd
->rd_length
)
1650 res_covered_end_of_rgrp
:
1651 if ((rbm
->bii
== 0) && nowrap
)
1663 * try_rgrp_unlink - Look for any unlinked, allocated, but unused inodes
1665 * @last_unlinked: block address of the last dinode we unlinked
1666 * @skip: block address we should explicitly not unlink
1668 * Returns: 0 if no error
1669 * The inode, if one has been found, in inode.
1672 static void try_rgrp_unlink(struct gfs2_rgrpd
*rgd
, u64
*last_unlinked
, u64 skip
)
1675 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
1676 struct gfs2_glock
*gl
;
1677 struct gfs2_inode
*ip
;
1680 struct gfs2_rbm rbm
= { .rgd
= rgd
, .bii
= 0, .offset
= 0 };
1683 down_write(&sdp
->sd_log_flush_lock
);
1684 error
= gfs2_rbm_find(&rbm
, GFS2_BLKST_UNLINKED
, 0, NULL
, true);
1685 up_write(&sdp
->sd_log_flush_lock
);
1686 if (error
== -ENOSPC
)
1688 if (WARN_ON_ONCE(error
))
1691 block
= gfs2_rbm_to_block(&rbm
);
1692 if (gfs2_rbm_from_block(&rbm
, block
+ 1))
1694 if (*last_unlinked
!= NO_BLOCK
&& block
<= *last_unlinked
)
1698 *last_unlinked
= block
;
1700 error
= gfs2_glock_get(sdp
, block
, &gfs2_inode_glops
, CREATE
, &gl
);
1704 /* If the inode is already in cache, we can ignore it here
1705 * because the existing inode disposal code will deal with
1706 * it when all refs have gone away. Accessing gl_object like
1707 * this is not safe in general. Here it is ok because we do
1708 * not dereference the pointer, and we only need an approx
1709 * answer to whether it is NULL or not.
1713 if (ip
|| queue_work(gfs2_delete_workqueue
, &gl
->gl_delete
) == 0)
1718 /* Limit reclaim to sensible number of tasks */
1719 if (found
> NR_CPUS
)
1723 rgd
->rd_flags
&= ~GFS2_RDF_CHECK
;
1728 * gfs2_rgrp_congested - Use stats to figure out whether an rgrp is congested
1729 * @rgd: The rgrp in question
1730 * @loops: An indication of how picky we can be (0=very, 1=less so)
1732 * This function uses the recently added glock statistics in order to
1733 * figure out whether a parciular resource group is suffering from
1734 * contention from multiple nodes. This is done purely on the basis
1735 * of timings, since this is the only data we have to work with and
1736 * our aim here is to reject a resource group which is highly contended
1737 * but (very important) not to do this too often in order to ensure that
1738 * we do not land up introducing fragmentation by changing resource
1739 * groups when not actually required.
1741 * The calculation is fairly simple, we want to know whether the SRTTB
1742 * (i.e. smoothed round trip time for blocking operations) to acquire
1743 * the lock for this rgrp's glock is significantly greater than the
1744 * time taken for resource groups on average. We introduce a margin in
1745 * the form of the variable @var which is computed as the sum of the two
1746 * respective variences, and multiplied by a factor depending on @loops
1747 * and whether we have a lot of data to base the decision on. This is
1748 * then tested against the square difference of the means in order to
1749 * decide whether the result is statistically significant or not.
1751 * Returns: A boolean verdict on the congestion status
1754 static bool gfs2_rgrp_congested(const struct gfs2_rgrpd
*rgd
, int loops
)
1756 const struct gfs2_glock
*gl
= rgd
->rd_gl
;
1757 const struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
1758 struct gfs2_lkstats
*st
;
1759 s64 r_dcount
, l_dcount
;
1760 s64 r_srttb
, l_srttb
;
1766 st
= &this_cpu_ptr(sdp
->sd_lkstats
)->lkstats
[LM_TYPE_RGRP
];
1767 r_srttb
= st
->stats
[GFS2_LKS_SRTTB
];
1768 r_dcount
= st
->stats
[GFS2_LKS_DCOUNT
];
1769 var
= st
->stats
[GFS2_LKS_SRTTVARB
] +
1770 gl
->gl_stats
.stats
[GFS2_LKS_SRTTVARB
];
1773 l_srttb
= gl
->gl_stats
.stats
[GFS2_LKS_SRTTB
];
1774 l_dcount
= gl
->gl_stats
.stats
[GFS2_LKS_DCOUNT
];
1776 if ((l_dcount
< 1) || (r_dcount
< 1) || (r_srttb
== 0))
1779 srttb_diff
= r_srttb
- l_srttb
;
1780 sqr_diff
= srttb_diff
* srttb_diff
;
1783 if (l_dcount
< 8 || r_dcount
< 8)
1788 return ((srttb_diff
< 0) && (sqr_diff
> var
));
1792 * gfs2_rgrp_used_recently
1793 * @rs: The block reservation with the rgrp to test
1794 * @msecs: The time limit in milliseconds
1796 * Returns: True if the rgrp glock has been used within the time limit
1798 static bool gfs2_rgrp_used_recently(const struct gfs2_blkreserv
*rs
,
1803 tdiff
= ktime_to_ns(ktime_sub(ktime_get_real(),
1804 rs
->rs_rbm
.rgd
->rd_gl
->gl_dstamp
));
1806 return tdiff
> (msecs
* 1000 * 1000);
1809 static u32
gfs2_orlov_skip(const struct gfs2_inode
*ip
)
1811 const struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
1814 get_random_bytes(&skip
, sizeof(skip
));
1815 return skip
% sdp
->sd_rgrps
;
1818 static bool gfs2_select_rgrp(struct gfs2_rgrpd
**pos
, const struct gfs2_rgrpd
*begin
)
1820 struct gfs2_rgrpd
*rgd
= *pos
;
1821 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
1823 rgd
= gfs2_rgrpd_get_next(rgd
);
1825 rgd
= gfs2_rgrpd_get_first(sdp
);
1827 if (rgd
!= begin
) /* If we didn't wrap */
1833 * gfs2_inplace_reserve - Reserve space in the filesystem
1834 * @ip: the inode to reserve space for
1835 * @requested: the number of blocks to be reserved
1840 int gfs2_inplace_reserve(struct gfs2_inode
*ip
, u32 requested
, u32 aflags
)
1842 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
1843 struct gfs2_rgrpd
*begin
= NULL
;
1844 struct gfs2_blkreserv
*rs
= ip
->i_res
;
1845 int error
= 0, rg_locked
, flags
= 0;
1846 u64 last_unlinked
= NO_BLOCK
;
1850 if (sdp
->sd_args
.ar_rgrplvb
)
1852 if (gfs2_assert_warn(sdp
, requested
))
1854 if (gfs2_rs_active(rs
)) {
1855 begin
= rs
->rs_rbm
.rgd
;
1856 } else if (ip
->i_rgd
&& rgrp_contains_block(ip
->i_rgd
, ip
->i_goal
)) {
1857 rs
->rs_rbm
.rgd
= begin
= ip
->i_rgd
;
1859 rs
->rs_rbm
.rgd
= begin
= gfs2_blk2rgrpd(sdp
, ip
->i_goal
, 1);
1861 if (S_ISDIR(ip
->i_inode
.i_mode
) && (aflags
& GFS2_AF_ORLOV
))
1862 skip
= gfs2_orlov_skip(ip
);
1863 if (rs
->rs_rbm
.rgd
== NULL
)
1869 if (!gfs2_glock_is_locked_by_me(rs
->rs_rbm
.rgd
->rd_gl
)) {
1873 if (!gfs2_rs_active(rs
) && (loops
< 2) &&
1874 gfs2_rgrp_used_recently(rs
, 1000) &&
1875 gfs2_rgrp_congested(rs
->rs_rbm
.rgd
, loops
))
1877 error
= gfs2_glock_nq_init(rs
->rs_rbm
.rgd
->rd_gl
,
1878 LM_ST_EXCLUSIVE
, flags
,
1880 if (unlikely(error
))
1882 if (!gfs2_rs_active(rs
) && (loops
< 2) &&
1883 gfs2_rgrp_congested(rs
->rs_rbm
.rgd
, loops
))
1885 if (sdp
->sd_args
.ar_rgrplvb
) {
1886 error
= update_rgrp_lvb(rs
->rs_rbm
.rgd
);
1887 if (unlikely(error
)) {
1888 gfs2_glock_dq_uninit(&rs
->rs_rgd_gh
);
1894 /* Skip unuseable resource groups */
1895 if (rs
->rs_rbm
.rgd
->rd_flags
& (GFS2_RGF_NOALLOC
| GFS2_RDF_ERROR
))
1898 if (sdp
->sd_args
.ar_rgrplvb
)
1899 gfs2_rgrp_bh_get(rs
->rs_rbm
.rgd
);
1901 /* Get a reservation if we don't already have one */
1902 if (!gfs2_rs_active(rs
))
1903 rg_mblk_search(rs
->rs_rbm
.rgd
, ip
, requested
);
1905 /* Skip rgrps when we can't get a reservation on first pass */
1906 if (!gfs2_rs_active(rs
) && (loops
< 1))
1909 /* If rgrp has enough free space, use it */
1910 if (rs
->rs_rbm
.rgd
->rd_free_clone
>= requested
) {
1911 ip
->i_rgd
= rs
->rs_rbm
.rgd
;
1915 /* Drop reservation, if we couldn't use reserved rgrp */
1916 if (gfs2_rs_active(rs
))
1917 gfs2_rs_deltree(rs
);
1919 /* Check for unlinked inodes which can be reclaimed */
1920 if (rs
->rs_rbm
.rgd
->rd_flags
& GFS2_RDF_CHECK
)
1921 try_rgrp_unlink(rs
->rs_rbm
.rgd
, &last_unlinked
,
1924 /* Unlock rgrp if required */
1926 gfs2_glock_dq_uninit(&rs
->rs_rgd_gh
);
1928 /* Find the next rgrp, and continue looking */
1929 if (gfs2_select_rgrp(&rs
->rs_rbm
.rgd
, begin
))
1934 /* If we've scanned all the rgrps, but found no free blocks
1935 * then this checks for some less likely conditions before
1939 /* Check that fs hasn't grown if writing to rindex */
1940 if (ip
== GFS2_I(sdp
->sd_rindex
) && !sdp
->sd_rindex_uptodate
) {
1941 error
= gfs2_ri_update(ip
);
1945 /* Flushing the log may release space */
1947 gfs2_log_flush(sdp
, NULL
);
1954 * gfs2_inplace_release - release an inplace reservation
1955 * @ip: the inode the reservation was taken out on
1957 * Release a reservation made by gfs2_inplace_reserve().
1960 void gfs2_inplace_release(struct gfs2_inode
*ip
)
1962 struct gfs2_blkreserv
*rs
= ip
->i_res
;
1964 if (rs
->rs_rgd_gh
.gh_gl
)
1965 gfs2_glock_dq_uninit(&rs
->rs_rgd_gh
);
1969 * gfs2_get_block_type - Check a block in a RG is of given type
1970 * @rgd: the resource group holding the block
1971 * @block: the block number
1973 * Returns: The block type (GFS2_BLKST_*)
1976 static unsigned char gfs2_get_block_type(struct gfs2_rgrpd
*rgd
, u64 block
)
1978 struct gfs2_rbm rbm
= { .rgd
= rgd
, };
1981 ret
= gfs2_rbm_from_block(&rbm
, block
);
1982 WARN_ON_ONCE(ret
!= 0);
1984 return gfs2_testbit(&rbm
);
1989 * gfs2_alloc_extent - allocate an extent from a given bitmap
1990 * @rbm: the resource group information
1991 * @dinode: TRUE if the first block we allocate is for a dinode
1992 * @n: The extent length (value/result)
1994 * Add the bitmap buffer to the transaction.
1995 * Set the found bits to @new_state to change block's allocation state.
1997 static void gfs2_alloc_extent(const struct gfs2_rbm
*rbm
, bool dinode
,
2000 struct gfs2_rbm pos
= { .rgd
= rbm
->rgd
, };
2001 const unsigned int elen
= *n
;
2006 block
= gfs2_rbm_to_block(rbm
);
2007 gfs2_trans_add_meta(rbm
->rgd
->rd_gl
, rbm_bi(rbm
)->bi_bh
);
2008 gfs2_setbit(rbm
, true, dinode
? GFS2_BLKST_DINODE
: GFS2_BLKST_USED
);
2011 ret
= gfs2_rbm_from_block(&pos
, block
);
2012 if (ret
|| gfs2_testbit(&pos
) != GFS2_BLKST_FREE
)
2014 gfs2_trans_add_meta(pos
.rgd
->rd_gl
, rbm_bi(&pos
)->bi_bh
);
2015 gfs2_setbit(&pos
, true, GFS2_BLKST_USED
);
2022 * rgblk_free - Change alloc state of given block(s)
2023 * @sdp: the filesystem
2024 * @bstart: the start of a run of blocks to free
2025 * @blen: the length of the block run (all must lie within ONE RG!)
2026 * @new_state: GFS2_BLKST_XXX the after-allocation block state
2028 * Returns: Resource group containing the block(s)
2031 static struct gfs2_rgrpd
*rgblk_free(struct gfs2_sbd
*sdp
, u64 bstart
,
2032 u32 blen
, unsigned char new_state
)
2034 struct gfs2_rbm rbm
;
2035 struct gfs2_bitmap
*bi
;
2037 rbm
.rgd
= gfs2_blk2rgrpd(sdp
, bstart
, 1);
2039 if (gfs2_consist(sdp
))
2040 fs_err(sdp
, "block = %llu\n", (unsigned long long)bstart
);
2045 gfs2_rbm_from_block(&rbm
, bstart
);
2048 if (!bi
->bi_clone
) {
2049 bi
->bi_clone
= kmalloc(bi
->bi_bh
->b_size
,
2050 GFP_NOFS
| __GFP_NOFAIL
);
2051 memcpy(bi
->bi_clone
+ bi
->bi_offset
,
2052 bi
->bi_bh
->b_data
+ bi
->bi_offset
, bi
->bi_len
);
2054 gfs2_trans_add_meta(rbm
.rgd
->rd_gl
, bi
->bi_bh
);
2055 gfs2_setbit(&rbm
, false, new_state
);
2062 * gfs2_rgrp_dump - print out an rgrp
2063 * @seq: The iterator
2064 * @gl: The glock in question
2068 int gfs2_rgrp_dump(struct seq_file
*seq
, const struct gfs2_glock
*gl
)
2070 struct gfs2_rgrpd
*rgd
= gl
->gl_object
;
2071 struct gfs2_blkreserv
*trs
;
2072 const struct rb_node
*n
;
2076 gfs2_print_dbg(seq
, " R: n:%llu f:%02x b:%u/%u i:%u r:%u\n",
2077 (unsigned long long)rgd
->rd_addr
, rgd
->rd_flags
,
2078 rgd
->rd_free
, rgd
->rd_free_clone
, rgd
->rd_dinodes
,
2080 spin_lock(&rgd
->rd_rsspin
);
2081 for (n
= rb_first(&rgd
->rd_rstree
); n
; n
= rb_next(&trs
->rs_node
)) {
2082 trs
= rb_entry(n
, struct gfs2_blkreserv
, rs_node
);
2085 spin_unlock(&rgd
->rd_rsspin
);
2089 static void gfs2_rgrp_error(struct gfs2_rgrpd
*rgd
)
2091 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
2092 fs_warn(sdp
, "rgrp %llu has an error, marking it readonly until umount\n",
2093 (unsigned long long)rgd
->rd_addr
);
2094 fs_warn(sdp
, "umount on all nodes and run fsck.gfs2 to fix the error\n");
2095 gfs2_rgrp_dump(NULL
, rgd
->rd_gl
);
2096 rgd
->rd_flags
|= GFS2_RDF_ERROR
;
2100 * gfs2_adjust_reservation - Adjust (or remove) a reservation after allocation
2101 * @ip: The inode we have just allocated blocks for
2102 * @rbm: The start of the allocated blocks
2103 * @len: The extent length
2105 * Adjusts a reservation after an allocation has taken place. If the
2106 * reservation does not match the allocation, or if it is now empty
2107 * then it is removed.
2110 static void gfs2_adjust_reservation(struct gfs2_inode
*ip
,
2111 const struct gfs2_rbm
*rbm
, unsigned len
)
2113 struct gfs2_blkreserv
*rs
= ip
->i_res
;
2114 struct gfs2_rgrpd
*rgd
= rbm
->rgd
;
2119 spin_lock(&rgd
->rd_rsspin
);
2120 if (gfs2_rs_active(rs
)) {
2121 if (gfs2_rbm_eq(&rs
->rs_rbm
, rbm
)) {
2122 block
= gfs2_rbm_to_block(rbm
);
2123 ret
= gfs2_rbm_from_block(&rs
->rs_rbm
, block
+ len
);
2124 rlen
= min(rs
->rs_free
, len
);
2125 rs
->rs_free
-= rlen
;
2126 rgd
->rd_reserved
-= rlen
;
2127 trace_gfs2_rs(rs
, TRACE_RS_CLAIM
);
2128 if (rs
->rs_free
&& !ret
)
2134 spin_unlock(&rgd
->rd_rsspin
);
2138 * gfs2_alloc_blocks - Allocate one or more blocks of data and/or a dinode
2139 * @ip: the inode to allocate the block for
2140 * @bn: Used to return the starting block number
2141 * @nblocks: requested number of blocks/extent length (value/result)
2142 * @dinode: 1 if we're allocating a dinode block, else 0
2143 * @generation: the generation number of the inode
2145 * Returns: 0 or error
2148 int gfs2_alloc_blocks(struct gfs2_inode
*ip
, u64
*bn
, unsigned int *nblocks
,
2149 bool dinode
, u64
*generation
)
2151 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
2152 struct buffer_head
*dibh
;
2153 struct gfs2_rbm rbm
= { .rgd
= ip
->i_rgd
, };
2156 u64 block
; /* block, within the file system scope */
2159 if (gfs2_rs_active(ip
->i_res
))
2160 goal
= gfs2_rbm_to_block(&ip
->i_res
->rs_rbm
);
2161 else if (!dinode
&& rgrp_contains_block(rbm
.rgd
, ip
->i_goal
))
2164 goal
= rbm
.rgd
->rd_last_alloc
+ rbm
.rgd
->rd_data0
;
2166 gfs2_rbm_from_block(&rbm
, goal
);
2167 error
= gfs2_rbm_find(&rbm
, GFS2_BLKST_FREE
, 0, ip
, false);
2169 if (error
== -ENOSPC
) {
2170 gfs2_rbm_from_block(&rbm
, goal
);
2171 error
= gfs2_rbm_find(&rbm
, GFS2_BLKST_FREE
, 0, NULL
, false);
2174 /* Since all blocks are reserved in advance, this shouldn't happen */
2176 fs_warn(sdp
, "inum=%llu error=%d, nblocks=%u, full=%d\n",
2177 (unsigned long long)ip
->i_no_addr
, error
, *nblocks
,
2178 test_bit(GBF_FULL
, &rbm
.rgd
->rd_bits
->bi_flags
));
2182 gfs2_alloc_extent(&rbm
, dinode
, nblocks
);
2183 block
= gfs2_rbm_to_block(&rbm
);
2184 rbm
.rgd
->rd_last_alloc
= block
- rbm
.rgd
->rd_data0
;
2185 if (gfs2_rs_active(ip
->i_res
))
2186 gfs2_adjust_reservation(ip
, &rbm
, *nblocks
);
2192 ip
->i_goal
= block
+ ndata
- 1;
2193 error
= gfs2_meta_inode_buffer(ip
, &dibh
);
2195 struct gfs2_dinode
*di
=
2196 (struct gfs2_dinode
*)dibh
->b_data
;
2197 gfs2_trans_add_meta(ip
->i_gl
, dibh
);
2198 di
->di_goal_meta
= di
->di_goal_data
=
2199 cpu_to_be64(ip
->i_goal
);
2203 if (rbm
.rgd
->rd_free
< *nblocks
) {
2204 printk(KERN_WARNING
"nblocks=%u\n", *nblocks
);
2208 rbm
.rgd
->rd_free
-= *nblocks
;
2210 rbm
.rgd
->rd_dinodes
++;
2211 *generation
= rbm
.rgd
->rd_igeneration
++;
2212 if (*generation
== 0)
2213 *generation
= rbm
.rgd
->rd_igeneration
++;
2216 gfs2_trans_add_meta(rbm
.rgd
->rd_gl
, rbm
.rgd
->rd_bits
[0].bi_bh
);
2217 gfs2_rgrp_out(rbm
.rgd
, rbm
.rgd
->rd_bits
[0].bi_bh
->b_data
);
2218 gfs2_rgrp_ondisk2lvb(rbm
.rgd
->rd_rgl
, rbm
.rgd
->rd_bits
[0].bi_bh
->b_data
);
2220 gfs2_statfs_change(sdp
, 0, -(s64
)*nblocks
, dinode
? 1 : 0);
2222 gfs2_trans_add_unrevoke(sdp
, block
, 1);
2224 gfs2_quota_change(ip
, *nblocks
, ip
->i_inode
.i_uid
, ip
->i_inode
.i_gid
);
2226 rbm
.rgd
->rd_free_clone
-= *nblocks
;
2227 trace_gfs2_block_alloc(ip
, rbm
.rgd
, block
, *nblocks
,
2228 dinode
? GFS2_BLKST_DINODE
: GFS2_BLKST_USED
);
2233 gfs2_rgrp_error(rbm
.rgd
);
2238 * __gfs2_free_blocks - free a contiguous run of 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
2242 * @meta: 1 if the blocks represent metadata
2246 void __gfs2_free_blocks(struct gfs2_inode
*ip
, u64 bstart
, u32 blen
, int meta
)
2248 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
2249 struct gfs2_rgrpd
*rgd
;
2251 rgd
= rgblk_free(sdp
, bstart
, blen
, GFS2_BLKST_FREE
);
2254 trace_gfs2_block_alloc(ip
, rgd
, bstart
, blen
, GFS2_BLKST_FREE
);
2255 rgd
->rd_free
+= blen
;
2256 rgd
->rd_flags
&= ~GFS2_RGF_TRIMMED
;
2257 gfs2_trans_add_meta(rgd
->rd_gl
, rgd
->rd_bits
[0].bi_bh
);
2258 gfs2_rgrp_out(rgd
, rgd
->rd_bits
[0].bi_bh
->b_data
);
2259 gfs2_rgrp_ondisk2lvb(rgd
->rd_rgl
, rgd
->rd_bits
[0].bi_bh
->b_data
);
2261 /* Directories keep their data in the metadata address space */
2262 if (meta
|| ip
->i_depth
)
2263 gfs2_meta_wipe(ip
, bstart
, blen
);
2267 * gfs2_free_meta - free a contiguous run of data block(s)
2268 * @ip: the inode these blocks are being freed from
2269 * @bstart: first block of a run of contiguous blocks
2270 * @blen: the length of the block run
2274 void gfs2_free_meta(struct gfs2_inode
*ip
, u64 bstart
, u32 blen
)
2276 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
2278 __gfs2_free_blocks(ip
, bstart
, blen
, 1);
2279 gfs2_statfs_change(sdp
, 0, +blen
, 0);
2280 gfs2_quota_change(ip
, -(s64
)blen
, ip
->i_inode
.i_uid
, ip
->i_inode
.i_gid
);
2283 void gfs2_unlink_di(struct inode
*inode
)
2285 struct gfs2_inode
*ip
= GFS2_I(inode
);
2286 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
2287 struct gfs2_rgrpd
*rgd
;
2288 u64 blkno
= ip
->i_no_addr
;
2290 rgd
= rgblk_free(sdp
, blkno
, 1, GFS2_BLKST_UNLINKED
);
2293 trace_gfs2_block_alloc(ip
, rgd
, blkno
, 1, GFS2_BLKST_UNLINKED
);
2294 gfs2_trans_add_meta(rgd
->rd_gl
, rgd
->rd_bits
[0].bi_bh
);
2295 gfs2_rgrp_out(rgd
, rgd
->rd_bits
[0].bi_bh
->b_data
);
2296 gfs2_rgrp_ondisk2lvb(rgd
->rd_rgl
, rgd
->rd_bits
[0].bi_bh
->b_data
);
2297 update_rgrp_lvb_unlinked(rgd
, 1);
2300 static void gfs2_free_uninit_di(struct gfs2_rgrpd
*rgd
, u64 blkno
)
2302 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
2303 struct gfs2_rgrpd
*tmp_rgd
;
2305 tmp_rgd
= rgblk_free(sdp
, blkno
, 1, GFS2_BLKST_FREE
);
2308 gfs2_assert_withdraw(sdp
, rgd
== tmp_rgd
);
2310 if (!rgd
->rd_dinodes
)
2311 gfs2_consist_rgrpd(rgd
);
2315 gfs2_trans_add_meta(rgd
->rd_gl
, rgd
->rd_bits
[0].bi_bh
);
2316 gfs2_rgrp_out(rgd
, rgd
->rd_bits
[0].bi_bh
->b_data
);
2317 gfs2_rgrp_ondisk2lvb(rgd
->rd_rgl
, rgd
->rd_bits
[0].bi_bh
->b_data
);
2318 update_rgrp_lvb_unlinked(rgd
, -1);
2320 gfs2_statfs_change(sdp
, 0, +1, -1);
2324 void gfs2_free_di(struct gfs2_rgrpd
*rgd
, struct gfs2_inode
*ip
)
2326 gfs2_free_uninit_di(rgd
, ip
->i_no_addr
);
2327 trace_gfs2_block_alloc(ip
, rgd
, ip
->i_no_addr
, 1, GFS2_BLKST_FREE
);
2328 gfs2_quota_change(ip
, -1, ip
->i_inode
.i_uid
, ip
->i_inode
.i_gid
);
2329 gfs2_meta_wipe(ip
, ip
->i_no_addr
, 1);
2333 * gfs2_check_blk_type - Check the type of a block
2334 * @sdp: The superblock
2335 * @no_addr: The block number to check
2336 * @type: The block type we are looking for
2338 * Returns: 0 if the block type matches the expected type
2339 * -ESTALE if it doesn't match
2340 * or -ve errno if something went wrong while checking
2343 int gfs2_check_blk_type(struct gfs2_sbd
*sdp
, u64 no_addr
, unsigned int type
)
2345 struct gfs2_rgrpd
*rgd
;
2346 struct gfs2_holder rgd_gh
;
2347 int error
= -EINVAL
;
2349 rgd
= gfs2_blk2rgrpd(sdp
, no_addr
, 1);
2353 error
= gfs2_glock_nq_init(rgd
->rd_gl
, LM_ST_SHARED
, 0, &rgd_gh
);
2357 if (gfs2_get_block_type(rgd
, no_addr
) != type
)
2360 gfs2_glock_dq_uninit(&rgd_gh
);
2366 * gfs2_rlist_add - add a RG to a list of RGs
2368 * @rlist: the list of resource groups
2371 * Figure out what RG a block belongs to and add that RG to the list
2373 * FIXME: Don't use NOFAIL
2377 void gfs2_rlist_add(struct gfs2_inode
*ip
, struct gfs2_rgrp_list
*rlist
,
2380 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
2381 struct gfs2_rgrpd
*rgd
;
2382 struct gfs2_rgrpd
**tmp
;
2383 unsigned int new_space
;
2386 if (gfs2_assert_warn(sdp
, !rlist
->rl_ghs
))
2389 if (ip
->i_rgd
&& rgrp_contains_block(ip
->i_rgd
, block
))
2392 rgd
= gfs2_blk2rgrpd(sdp
, block
, 1);
2394 fs_err(sdp
, "rlist_add: no rgrp for block %llu\n", (unsigned long long)block
);
2399 for (x
= 0; x
< rlist
->rl_rgrps
; x
++)
2400 if (rlist
->rl_rgd
[x
] == rgd
)
2403 if (rlist
->rl_rgrps
== rlist
->rl_space
) {
2404 new_space
= rlist
->rl_space
+ 10;
2406 tmp
= kcalloc(new_space
, sizeof(struct gfs2_rgrpd
*),
2407 GFP_NOFS
| __GFP_NOFAIL
);
2409 if (rlist
->rl_rgd
) {
2410 memcpy(tmp
, rlist
->rl_rgd
,
2411 rlist
->rl_space
* sizeof(struct gfs2_rgrpd
*));
2412 kfree(rlist
->rl_rgd
);
2415 rlist
->rl_space
= new_space
;
2416 rlist
->rl_rgd
= tmp
;
2419 rlist
->rl_rgd
[rlist
->rl_rgrps
++] = rgd
;
2423 * gfs2_rlist_alloc - all RGs have been added to the rlist, now allocate
2424 * and initialize an array of glock holders for them
2425 * @rlist: the list of resource groups
2426 * @state: the lock state to acquire the RG lock in
2428 * FIXME: Don't use NOFAIL
2432 void gfs2_rlist_alloc(struct gfs2_rgrp_list
*rlist
, unsigned int state
)
2436 rlist
->rl_ghs
= kcalloc(rlist
->rl_rgrps
, sizeof(struct gfs2_holder
),
2437 GFP_NOFS
| __GFP_NOFAIL
);
2438 for (x
= 0; x
< rlist
->rl_rgrps
; x
++)
2439 gfs2_holder_init(rlist
->rl_rgd
[x
]->rd_gl
,
2445 * gfs2_rlist_free - free a resource group list
2446 * @list: the list of resource groups
2450 void gfs2_rlist_free(struct gfs2_rgrp_list
*rlist
)
2454 kfree(rlist
->rl_rgd
);
2456 if (rlist
->rl_ghs
) {
2457 for (x
= 0; x
< rlist
->rl_rgrps
; x
++)
2458 gfs2_holder_uninit(&rlist
->rl_ghs
[x
]);
2459 kfree(rlist
->rl_ghs
);
2460 rlist
->rl_ghs
= NULL
;
This page took 0.137379 seconds and 6 git commands to generate.