]>
git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - fs/gfs2/bmap.c
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
10 #include <linux/spinlock.h>
11 #include <linux/completion.h>
12 #include <linux/buffer_head.h>
13 #include <linux/blkdev.h>
14 #include <linux/gfs2_ondisk.h>
15 #include <linux/crc32.h>
30 #include "trace_gfs2.h"
32 /* This doesn't need to be that large as max 64 bit pointers in a 4k
33 * block is 512, so __u16 is fine for that. It saves stack space to
37 struct buffer_head
*mp_bh
[GFS2_MAX_META_HEIGHT
];
38 __u16 mp_list
[GFS2_MAX_META_HEIGHT
];
43 unsigned int sm_height
;
47 * gfs2_unstuffer_page - unstuff a stuffed inode into a block cached by a page
49 * @dibh: the dinode buffer
50 * @block: the block number that was allocated
51 * @page: The (optional) page. This is looked up if @page is NULL
56 static int gfs2_unstuffer_page(struct gfs2_inode
*ip
, struct buffer_head
*dibh
,
57 u64 block
, struct page
*page
)
59 struct inode
*inode
= &ip
->i_inode
;
60 struct buffer_head
*bh
;
63 if (!page
|| page
->index
) {
64 page
= find_or_create_page(inode
->i_mapping
, 0, GFP_NOFS
);
70 if (!PageUptodate(page
)) {
71 void *kaddr
= kmap(page
);
72 u64 dsize
= i_size_read(inode
);
74 if (dsize
> (dibh
->b_size
- sizeof(struct gfs2_dinode
)))
75 dsize
= dibh
->b_size
- sizeof(struct gfs2_dinode
);
77 memcpy(kaddr
, dibh
->b_data
+ sizeof(struct gfs2_dinode
), dsize
);
78 memset(kaddr
+ dsize
, 0, PAGE_CACHE_SIZE
- dsize
);
81 SetPageUptodate(page
);
84 if (!page_has_buffers(page
))
85 create_empty_buffers(page
, 1 << inode
->i_blkbits
,
88 bh
= page_buffers(page
);
90 if (!buffer_mapped(bh
))
91 map_bh(bh
, inode
->i_sb
, block
);
93 set_buffer_uptodate(bh
);
94 if (!gfs2_is_jdata(ip
))
95 mark_buffer_dirty(bh
);
96 if (!gfs2_is_writeback(ip
))
97 gfs2_trans_add_data(ip
->i_gl
, bh
);
101 page_cache_release(page
);
108 * gfs2_unstuff_dinode - Unstuff a dinode when the data has grown too big
109 * @ip: The GFS2 inode to unstuff
110 * @page: The (optional) page. This is looked up if the @page is NULL
112 * This routine unstuffs a dinode and returns it to a "normal" state such
113 * that the height can be grown in the traditional way.
118 int gfs2_unstuff_dinode(struct gfs2_inode
*ip
, struct page
*page
)
120 struct buffer_head
*bh
, *dibh
;
121 struct gfs2_dinode
*di
;
123 int isdir
= gfs2_is_dir(ip
);
126 down_write(&ip
->i_rw_mutex
);
128 error
= gfs2_meta_inode_buffer(ip
, &dibh
);
132 if (i_size_read(&ip
->i_inode
)) {
133 /* Get a free block, fill it with the stuffed data,
134 and write it out to disk */
137 error
= gfs2_alloc_blocks(ip
, &block
, &n
, 0, NULL
);
141 gfs2_trans_add_unrevoke(GFS2_SB(&ip
->i_inode
), block
, 1);
142 error
= gfs2_dir_get_new_buffer(ip
, block
, &bh
);
145 gfs2_buffer_copy_tail(bh
, sizeof(struct gfs2_meta_header
),
146 dibh
, sizeof(struct gfs2_dinode
));
149 error
= gfs2_unstuffer_page(ip
, dibh
, block
, page
);
155 /* Set up the pointer to the new block */
157 gfs2_trans_add_meta(ip
->i_gl
, dibh
);
158 di
= (struct gfs2_dinode
*)dibh
->b_data
;
159 gfs2_buffer_clear_tail(dibh
, sizeof(struct gfs2_dinode
));
161 if (i_size_read(&ip
->i_inode
)) {
162 *(__be64
*)(di
+ 1) = cpu_to_be64(block
);
163 gfs2_add_inode_blocks(&ip
->i_inode
, 1);
164 di
->di_blocks
= cpu_to_be64(gfs2_get_inode_blocks(&ip
->i_inode
));
168 di
->di_height
= cpu_to_be16(1);
173 up_write(&ip
->i_rw_mutex
);
179 * find_metapath - Find path through the metadata tree
180 * @sdp: The superblock
181 * @mp: The metapath to return the result in
182 * @block: The disk block to look up
183 * @height: The pre-calculated height of the metadata tree
185 * This routine returns a struct metapath structure that defines a path
186 * through the metadata of inode "ip" to get to block "block".
189 * Given: "ip" is a height 3 file, "offset" is 101342453, and this is a
190 * filesystem with a blocksize of 4096.
192 * find_metapath() would return a struct metapath structure set to:
193 * mp_offset = 101342453, mp_height = 3, mp_list[0] = 0, mp_list[1] = 48,
194 * and mp_list[2] = 165.
196 * That means that in order to get to the block containing the byte at
197 * offset 101342453, we would load the indirect block pointed to by pointer
198 * 0 in the dinode. We would then load the indirect block pointed to by
199 * pointer 48 in that indirect block. We would then load the data block
200 * pointed to by pointer 165 in that indirect block.
202 * ----------------------------------------
207 * ----------------------------------------
211 * ----------------------------------------
215 * |0 5 6 7 8 9 0 1 2|
216 * ----------------------------------------
220 * ----------------------------------------
225 * ----------------------------------------
229 * ----------------------------------------
230 * | Data block containing offset |
234 * ----------------------------------------
238 static void find_metapath(const struct gfs2_sbd
*sdp
, u64 block
,
239 struct metapath
*mp
, unsigned int height
)
243 for (i
= height
; i
--;)
244 mp
->mp_list
[i
] = do_div(block
, sdp
->sd_inptrs
);
248 static inline unsigned int metapath_branch_start(const struct metapath
*mp
)
250 if (mp
->mp_list
[0] == 0)
256 * metapointer - Return pointer to start of metadata in a buffer
257 * @height: The metadata height (0 = dinode)
260 * Return a pointer to the block number of the next height of the metadata
261 * tree given a buffer containing the pointer to the current height of the
265 static inline __be64
*metapointer(unsigned int height
, const struct metapath
*mp
)
267 struct buffer_head
*bh
= mp
->mp_bh
[height
];
268 unsigned int head_size
= (height
> 0) ?
269 sizeof(struct gfs2_meta_header
) : sizeof(struct gfs2_dinode
);
270 return ((__be64
*)(bh
->b_data
+ head_size
)) + mp
->mp_list
[height
];
273 static void gfs2_metapath_ra(struct gfs2_glock
*gl
,
274 const struct buffer_head
*bh
, const __be64
*pos
)
276 struct buffer_head
*rabh
;
277 const __be64
*endp
= (const __be64
*)(bh
->b_data
+ bh
->b_size
);
280 for (t
= pos
; t
< endp
; t
++) {
284 rabh
= gfs2_getbuf(gl
, be64_to_cpu(*t
), CREATE
);
285 if (trylock_buffer(rabh
)) {
286 if (!buffer_uptodate(rabh
)) {
287 rabh
->b_end_io
= end_buffer_read_sync
;
288 submit_bh(READA
| REQ_META
, rabh
);
298 * lookup_metapath - Walk the metadata tree to a specific point
302 * Assumes that the inode's buffer has already been looked up and
303 * hooked onto mp->mp_bh[0] and that the metapath has been initialised
304 * by find_metapath().
306 * If this function encounters part of the tree which has not been
307 * allocated, it returns the current height of the tree at the point
308 * at which it found the unallocated block. Blocks which are found are
309 * added to the mp->mp_bh[] list.
311 * Returns: error or height of metadata tree
314 static int lookup_metapath(struct gfs2_inode
*ip
, struct metapath
*mp
)
316 unsigned int end_of_metadata
= ip
->i_height
- 1;
322 for (x
= 0; x
< end_of_metadata
; x
++) {
323 ptr
= metapointer(x
, mp
);
324 dblock
= be64_to_cpu(*ptr
);
328 ret
= gfs2_meta_indirect_buffer(ip
, x
+1, dblock
, &mp
->mp_bh
[x
+1]);
336 static inline void release_metapath(struct metapath
*mp
)
340 for (i
= 0; i
< GFS2_MAX_META_HEIGHT
; i
++) {
341 if (mp
->mp_bh
[i
] == NULL
)
343 brelse(mp
->mp_bh
[i
]);
348 * gfs2_extent_length - Returns length of an extent of blocks
349 * @start: Start of the buffer
350 * @len: Length of the buffer in bytes
351 * @ptr: Current position in the buffer
352 * @limit: Max extent length to return (0 = unlimited)
353 * @eob: Set to 1 if we hit "end of block"
355 * If the first block is zero (unallocated) it will return the number of
356 * unallocated blocks in the extent, otherwise it will return the number
357 * of contiguous blocks in the extent.
359 * Returns: The length of the extent (minimum of one block)
362 static inline unsigned int gfs2_extent_length(void *start
, unsigned int len
, __be64
*ptr
, unsigned limit
, int *eob
)
364 const __be64
*end
= (start
+ len
);
365 const __be64
*first
= ptr
;
366 u64 d
= be64_to_cpu(*ptr
);
373 if (limit
&& --limit
== 0)
377 } while(be64_to_cpu(*ptr
) == d
);
380 return (ptr
- first
);
383 static inline void bmap_lock(struct gfs2_inode
*ip
, int create
)
386 down_write(&ip
->i_rw_mutex
);
388 down_read(&ip
->i_rw_mutex
);
391 static inline void bmap_unlock(struct gfs2_inode
*ip
, int create
)
394 up_write(&ip
->i_rw_mutex
);
396 up_read(&ip
->i_rw_mutex
);
399 static inline __be64
*gfs2_indirect_init(struct metapath
*mp
,
400 struct gfs2_glock
*gl
, unsigned int i
,
401 unsigned offset
, u64 bn
)
403 __be64
*ptr
= (__be64
*)(mp
->mp_bh
[i
- 1]->b_data
+
404 ((i
> 1) ? sizeof(struct gfs2_meta_header
) :
405 sizeof(struct gfs2_dinode
)));
407 BUG_ON(mp
->mp_bh
[i
] != NULL
);
408 mp
->mp_bh
[i
] = gfs2_meta_new(gl
, bn
);
409 gfs2_trans_add_meta(gl
, mp
->mp_bh
[i
]);
410 gfs2_metatype_set(mp
->mp_bh
[i
], GFS2_METATYPE_IN
, GFS2_FORMAT_IN
);
411 gfs2_buffer_clear_tail(mp
->mp_bh
[i
], sizeof(struct gfs2_meta_header
));
413 *ptr
= cpu_to_be64(bn
);
419 ALLOC_GROW_DEPTH
= 1,
420 ALLOC_GROW_HEIGHT
= 2,
421 /* ALLOC_UNSTUFF = 3, TBD and rather complicated */
425 * gfs2_bmap_alloc - Build a metadata tree of the requested height
426 * @inode: The GFS2 inode
427 * @lblock: The logical starting block of the extent
428 * @bh_map: This is used to return the mapping details
430 * @sheight: The starting height (i.e. whats already mapped)
431 * @height: The height to build to
432 * @maxlen: The max number of data blocks to alloc
434 * In this routine we may have to alloc:
435 * i) Indirect blocks to grow the metadata tree height
436 * ii) Indirect blocks to fill in lower part of the metadata tree
439 * The function is in two parts. The first part works out the total
440 * number of blocks which we need. The second part does the actual
441 * allocation asking for an extent at a time (if enough contiguous free
442 * blocks are available, there will only be one request per bmap call)
443 * and uses the state machine to initialise the blocks in order.
445 * Returns: errno on error
448 static int gfs2_bmap_alloc(struct inode
*inode
, const sector_t lblock
,
449 struct buffer_head
*bh_map
, struct metapath
*mp
,
450 const unsigned int sheight
,
451 const unsigned int height
,
452 const unsigned int maxlen
)
454 struct gfs2_inode
*ip
= GFS2_I(inode
);
455 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
456 struct super_block
*sb
= sdp
->sd_vfs
;
457 struct buffer_head
*dibh
= mp
->mp_bh
[0];
459 unsigned n
, i
, blks
, alloced
= 0, iblks
= 0, branch_start
= 0;
461 unsigned ptrs_per_blk
;
462 const unsigned end_of_metadata
= height
- 1;
465 enum alloc_state state
;
470 BUG_ON(dibh
== NULL
);
472 gfs2_trans_add_meta(ip
->i_gl
, dibh
);
474 if (height
== sheight
) {
475 struct buffer_head
*bh
;
476 /* Bottom indirect block exists, find unalloced extent size */
477 ptr
= metapointer(end_of_metadata
, mp
);
478 bh
= mp
->mp_bh
[end_of_metadata
];
479 dblks
= gfs2_extent_length(bh
->b_data
, bh
->b_size
, ptr
, maxlen
,
484 /* Need to allocate indirect blocks */
485 ptrs_per_blk
= height
> 1 ? sdp
->sd_inptrs
: sdp
->sd_diptrs
;
486 dblks
= min(maxlen
, ptrs_per_blk
- mp
->mp_list
[end_of_metadata
]);
487 if (height
== ip
->i_height
) {
488 /* Writing into existing tree, extend tree down */
489 iblks
= height
- sheight
;
490 state
= ALLOC_GROW_DEPTH
;
492 /* Building up tree height */
493 state
= ALLOC_GROW_HEIGHT
;
494 iblks
= height
- ip
->i_height
;
495 branch_start
= metapath_branch_start(mp
);
496 iblks
+= (height
- branch_start
);
500 /* start of the second part of the function (state machine) */
502 blks
= dblks
+ iblks
;
507 error
= gfs2_alloc_blocks(ip
, &bn
, &n
, 0, NULL
);
511 if (state
!= ALLOC_DATA
|| gfs2_is_jdata(ip
))
512 gfs2_trans_add_unrevoke(sdp
, bn
, n
);
514 /* Growing height of tree */
515 case ALLOC_GROW_HEIGHT
:
517 ptr
= (__be64
*)(dibh
->b_data
+
518 sizeof(struct gfs2_dinode
));
521 for (; i
- 1 < height
- ip
->i_height
&& n
> 0; i
++, n
--)
522 gfs2_indirect_init(mp
, ip
->i_gl
, i
, 0, bn
++);
523 if (i
- 1 == height
- ip
->i_height
) {
525 gfs2_buffer_copy_tail(mp
->mp_bh
[i
],
526 sizeof(struct gfs2_meta_header
),
527 dibh
, sizeof(struct gfs2_dinode
));
528 gfs2_buffer_clear_tail(dibh
,
529 sizeof(struct gfs2_dinode
) +
531 ptr
= (__be64
*)(mp
->mp_bh
[i
]->b_data
+
532 sizeof(struct gfs2_meta_header
));
534 state
= ALLOC_GROW_DEPTH
;
535 for(i
= branch_start
; i
< height
; i
++) {
536 if (mp
->mp_bh
[i
] == NULL
)
538 brelse(mp
->mp_bh
[i
]);
545 /* Branching from existing tree */
546 case ALLOC_GROW_DEPTH
:
547 if (i
> 1 && i
< height
)
548 gfs2_trans_add_meta(ip
->i_gl
, mp
->mp_bh
[i
-1]);
549 for (; i
< height
&& n
> 0; i
++, n
--)
550 gfs2_indirect_init(mp
, ip
->i_gl
, i
,
551 mp
->mp_list
[i
-1], bn
++);
556 /* Tree complete, adding data blocks */
559 BUG_ON(mp
->mp_bh
[end_of_metadata
] == NULL
);
560 gfs2_trans_add_meta(ip
->i_gl
, mp
->mp_bh
[end_of_metadata
]);
562 ptr
= metapointer(end_of_metadata
, mp
);
565 *ptr
++ = cpu_to_be64(bn
++);
566 if (buffer_zeronew(bh_map
)) {
567 ret
= sb_issue_zeroout(sb
, dblock
, dblks
,
571 "Failed to zero data buffers\n");
572 clear_buffer_zeronew(bh_map
);
577 } while ((state
!= ALLOC_DATA
) || !dblock
);
579 ip
->i_height
= height
;
580 gfs2_add_inode_blocks(&ip
->i_inode
, alloced
);
581 gfs2_dinode_out(ip
, mp
->mp_bh
[0]->b_data
);
582 map_bh(bh_map
, inode
->i_sb
, dblock
);
583 bh_map
->b_size
= dblks
<< inode
->i_blkbits
;
584 set_buffer_new(bh_map
);
589 * gfs2_block_map - Map a block from an inode to a disk block
591 * @lblock: The logical block number
592 * @bh_map: The bh to be mapped
593 * @create: True if its ok to alloc blocks to satify the request
595 * Sets buffer_mapped() if successful, sets buffer_boundary() if a
596 * read of metadata will be required before the next block can be
597 * mapped. Sets buffer_new() if new blocks were allocated.
602 int gfs2_block_map(struct inode
*inode
, sector_t lblock
,
603 struct buffer_head
*bh_map
, int create
)
605 struct gfs2_inode
*ip
= GFS2_I(inode
);
606 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
607 unsigned int bsize
= sdp
->sd_sb
.sb_bsize
;
608 const unsigned int maxlen
= bh_map
->b_size
>> inode
->i_blkbits
;
609 const u64
*arr
= sdp
->sd_heightsize
;
616 struct buffer_head
*bh
;
621 memset(mp
.mp_bh
, 0, sizeof(mp
.mp_bh
));
622 bmap_lock(ip
, create
);
623 clear_buffer_mapped(bh_map
);
624 clear_buffer_new(bh_map
);
625 clear_buffer_boundary(bh_map
);
626 trace_gfs2_bmap(ip
, bh_map
, lblock
, create
, 1);
627 if (gfs2_is_dir(ip
)) {
628 bsize
= sdp
->sd_jbsize
;
629 arr
= sdp
->sd_jheightsize
;
632 ret
= gfs2_meta_inode_buffer(ip
, &mp
.mp_bh
[0]);
636 height
= ip
->i_height
;
637 size
= (lblock
+ 1) * bsize
;
638 while (size
> arr
[height
])
640 find_metapath(sdp
, lblock
, &mp
, height
);
642 if (height
> ip
->i_height
|| gfs2_is_stuffed(ip
))
644 ret
= lookup_metapath(ip
, &mp
);
647 if (ret
!= ip
->i_height
)
649 ptr
= metapointer(ip
->i_height
- 1, &mp
);
652 map_bh(bh_map
, inode
->i_sb
, be64_to_cpu(*ptr
));
653 bh
= mp
.mp_bh
[ip
->i_height
- 1];
654 len
= gfs2_extent_length(bh
->b_data
, bh
->b_size
, ptr
, maxlen
, &eob
);
655 bh_map
->b_size
= (len
<< inode
->i_blkbits
);
657 set_buffer_boundary(bh_map
);
660 release_metapath(&mp
);
661 trace_gfs2_bmap(ip
, bh_map
, lblock
, create
, ret
);
662 bmap_unlock(ip
, create
);
666 /* All allocations are done here, firstly check create flag */
668 BUG_ON(gfs2_is_stuffed(ip
));
673 /* At this point ret is the tree depth of already allocated blocks */
674 ret
= gfs2_bmap_alloc(inode
, lblock
, bh_map
, &mp
, ret
, height
, maxlen
);
679 * Deprecated: do not use in new code
681 int gfs2_extent_map(struct inode
*inode
, u64 lblock
, int *new, u64
*dblock
, unsigned *extlen
)
683 struct buffer_head bh
= { .b_state
= 0, .b_blocknr
= 0 };
691 bh
.b_size
= 1 << (inode
->i_blkbits
+ (create
? 0 : 5));
692 ret
= gfs2_block_map(inode
, lblock
, &bh
, create
);
693 *extlen
= bh
.b_size
>> inode
->i_blkbits
;
694 *dblock
= bh
.b_blocknr
;
703 * do_strip - Look for a layer a particular layer of the file and strip it off
705 * @dibh: the dinode buffer
706 * @bh: A buffer of pointers
707 * @top: The first pointer in the buffer
708 * @bottom: One more than the last pointer
709 * @height: the height this buffer is at
710 * @data: a pointer to a struct strip_mine
715 static int do_strip(struct gfs2_inode
*ip
, struct buffer_head
*dibh
,
716 struct buffer_head
*bh
, __be64
*top
, __be64
*bottom
,
717 unsigned int height
, struct strip_mine
*sm
)
719 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
720 struct gfs2_rgrp_list rlist
;
724 unsigned int rg_blocks
= 0;
726 unsigned int revokes
= 0;
730 error
= gfs2_rindex_update(sdp
);
737 if (height
!= sm
->sm_height
)
745 metadata
= (height
!= ip
->i_height
- 1);
747 revokes
= (height
) ? sdp
->sd_inptrs
: sdp
->sd_diptrs
;
748 else if (ip
->i_depth
)
749 revokes
= sdp
->sd_inptrs
;
751 memset(&rlist
, 0, sizeof(struct gfs2_rgrp_list
));
755 for (p
= top
; p
< bottom
; p
++) {
759 bn
= be64_to_cpu(*p
);
761 if (bstart
+ blen
== bn
)
765 gfs2_rlist_add(ip
, &rlist
, bstart
);
773 gfs2_rlist_add(ip
, &rlist
, bstart
);
775 goto out
; /* Nothing to do */
777 gfs2_rlist_alloc(&rlist
, LM_ST_EXCLUSIVE
);
779 for (x
= 0; x
< rlist
.rl_rgrps
; x
++) {
780 struct gfs2_rgrpd
*rgd
;
781 rgd
= rlist
.rl_ghs
[x
].gh_gl
->gl_object
;
782 rg_blocks
+= rgd
->rd_length
;
785 error
= gfs2_glock_nq_m(rlist
.rl_rgrps
, rlist
.rl_ghs
);
789 if (gfs2_rs_active(ip
->i_res
)) /* needs to be done with the rgrp glock held */
790 gfs2_rs_deltree(ip
->i_res
);
792 error
= gfs2_trans_begin(sdp
, rg_blocks
+ RES_DINODE
+
793 RES_INDIRECT
+ RES_STATFS
+ RES_QUOTA
,
798 down_write(&ip
->i_rw_mutex
);
800 gfs2_trans_add_meta(ip
->i_gl
, dibh
);
801 gfs2_trans_add_meta(ip
->i_gl
, bh
);
807 for (p
= top
; p
< bottom
; p
++) {
811 bn
= be64_to_cpu(*p
);
813 if (bstart
+ blen
== bn
)
817 __gfs2_free_blocks(ip
, bstart
, blen
, metadata
);
826 gfs2_add_inode_blocks(&ip
->i_inode
, -1);
829 __gfs2_free_blocks(ip
, bstart
, blen
, metadata
);
833 gfs2_statfs_change(sdp
, 0, +btotal
, 0);
834 gfs2_quota_change(ip
, -(s64
)btotal
, ip
->i_inode
.i_uid
,
837 ip
->i_inode
.i_mtime
= ip
->i_inode
.i_ctime
= CURRENT_TIME
;
839 gfs2_dinode_out(ip
, dibh
->b_data
);
841 up_write(&ip
->i_rw_mutex
);
846 gfs2_glock_dq_m(rlist
.rl_rgrps
, rlist
.rl_ghs
);
848 gfs2_rlist_free(&rlist
);
854 * recursive_scan - recursively scan through the end of a file
856 * @dibh: the dinode buffer
857 * @mp: the path through the metadata to the point to start
858 * @height: the height the recursion is at
859 * @block: the indirect block to look at
860 * @first: 1 if this is the first block
861 * @sm: data opaque to this function to pass to @bc
863 * When this is first called @height and @block should be zero and
864 * @first should be 1.
869 static int recursive_scan(struct gfs2_inode
*ip
, struct buffer_head
*dibh
,
870 struct metapath
*mp
, unsigned int height
,
871 u64 block
, int first
, struct strip_mine
*sm
)
873 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
874 struct buffer_head
*bh
= NULL
;
875 __be64
*top
, *bottom
;
878 int mh_size
= sizeof(struct gfs2_meta_header
);
881 error
= gfs2_meta_inode_buffer(ip
, &bh
);
886 top
= (__be64
*)(bh
->b_data
+ sizeof(struct gfs2_dinode
)) + mp
->mp_list
[0];
887 bottom
= (__be64
*)(bh
->b_data
+ sizeof(struct gfs2_dinode
)) + sdp
->sd_diptrs
;
889 error
= gfs2_meta_indirect_buffer(ip
, height
, block
, &bh
);
893 top
= (__be64
*)(bh
->b_data
+ mh_size
) +
894 (first
? mp
->mp_list
[height
] : 0);
896 bottom
= (__be64
*)(bh
->b_data
+ mh_size
) + sdp
->sd_inptrs
;
899 error
= do_strip(ip
, dibh
, bh
, top
, bottom
, height
, sm
);
903 if (height
< ip
->i_height
- 1) {
905 gfs2_metapath_ra(ip
->i_gl
, bh
, top
);
907 for (; top
< bottom
; top
++, first
= 0) {
911 bn
= be64_to_cpu(*top
);
913 error
= recursive_scan(ip
, dibh
, mp
, height
+ 1, bn
,
926 * gfs2_block_truncate_page - Deal with zeroing out data for truncate
928 * This is partly borrowed from ext3.
930 static int gfs2_block_truncate_page(struct address_space
*mapping
, loff_t from
)
932 struct inode
*inode
= mapping
->host
;
933 struct gfs2_inode
*ip
= GFS2_I(inode
);
934 unsigned long index
= from
>> PAGE_CACHE_SHIFT
;
935 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
936 unsigned blocksize
, iblock
, length
, pos
;
937 struct buffer_head
*bh
;
941 page
= find_or_create_page(mapping
, index
, GFP_NOFS
);
945 blocksize
= inode
->i_sb
->s_blocksize
;
946 length
= blocksize
- (offset
& (blocksize
- 1));
947 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
949 if (!page_has_buffers(page
))
950 create_empty_buffers(page
, blocksize
, 0);
952 /* Find the buffer that contains "offset" */
953 bh
= page_buffers(page
);
955 while (offset
>= pos
) {
956 bh
= bh
->b_this_page
;
963 if (!buffer_mapped(bh
)) {
964 gfs2_block_map(inode
, iblock
, bh
, 0);
965 /* unmapped? It's a hole - nothing to do */
966 if (!buffer_mapped(bh
))
970 /* Ok, it's mapped. Make sure it's up-to-date */
971 if (PageUptodate(page
))
972 set_buffer_uptodate(bh
);
974 if (!buffer_uptodate(bh
)) {
976 ll_rw_block(READ
, 1, &bh
);
978 /* Uhhuh. Read error. Complain and punt. */
979 if (!buffer_uptodate(bh
))
984 if (!gfs2_is_writeback(ip
))
985 gfs2_trans_add_data(ip
->i_gl
, bh
);
987 zero_user(page
, offset
, length
);
988 mark_buffer_dirty(bh
);
991 page_cache_release(page
);
996 * gfs2_journaled_truncate - Wrapper for truncate_pagecache for jdata files
997 * @inode: The inode being truncated
998 * @oldsize: The original (larger) size
999 * @newsize: The new smaller size
1001 * With jdata files, we have to journal a revoke for each block which is
1002 * truncated. As a result, we need to split this into separate transactions
1003 * if the number of pages being truncated gets too large.
1006 #define GFS2_JTRUNC_REVOKES 8192
1008 static int gfs2_journaled_truncate(struct inode
*inode
, u64 oldsize
, u64 newsize
)
1010 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
1011 u64 max_chunk
= GFS2_JTRUNC_REVOKES
* sdp
->sd_vfs
->s_blocksize
;
1015 while (oldsize
!= newsize
) {
1016 chunk
= oldsize
- newsize
;
1017 if (chunk
> max_chunk
)
1019 truncate_pagecache(inode
, oldsize
- chunk
);
1021 gfs2_trans_end(sdp
);
1022 error
= gfs2_trans_begin(sdp
, RES_DINODE
, GFS2_JTRUNC_REVOKES
);
1030 static int trunc_start(struct inode
*inode
, u64 oldsize
, u64 newsize
)
1032 struct gfs2_inode
*ip
= GFS2_I(inode
);
1033 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
1034 struct address_space
*mapping
= inode
->i_mapping
;
1035 struct buffer_head
*dibh
;
1036 int journaled
= gfs2_is_jdata(ip
);
1040 error
= gfs2_trans_begin(sdp
, RES_DINODE
+ RES_JDATA
, GFS2_JTRUNC_REVOKES
);
1042 error
= gfs2_trans_begin(sdp
, RES_DINODE
, 0);
1046 error
= gfs2_meta_inode_buffer(ip
, &dibh
);
1050 gfs2_trans_add_meta(ip
->i_gl
, dibh
);
1052 if (gfs2_is_stuffed(ip
)) {
1053 gfs2_buffer_clear_tail(dibh
, sizeof(struct gfs2_dinode
) + newsize
);
1055 if (newsize
& (u64
)(sdp
->sd_sb
.sb_bsize
- 1)) {
1056 error
= gfs2_block_truncate_page(mapping
, newsize
);
1060 ip
->i_diskflags
|= GFS2_DIF_TRUNC_IN_PROG
;
1063 i_size_write(inode
, newsize
);
1064 ip
->i_inode
.i_mtime
= ip
->i_inode
.i_ctime
= CURRENT_TIME
;
1065 gfs2_dinode_out(ip
, dibh
->b_data
);
1068 error
= gfs2_journaled_truncate(inode
, oldsize
, newsize
);
1070 truncate_pagecache(inode
, newsize
);
1080 gfs2_trans_end(sdp
);
1084 static int trunc_dealloc(struct gfs2_inode
*ip
, u64 size
)
1086 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
1087 unsigned int height
= ip
->i_height
;
1095 lblock
= (size
- 1) >> sdp
->sd_sb
.sb_bsize_shift
;
1097 find_metapath(sdp
, lblock
, &mp
, ip
->i_height
);
1098 error
= gfs2_rindex_update(sdp
);
1102 error
= gfs2_quota_hold(ip
, NO_UID_QUOTA_CHANGE
, NO_GID_QUOTA_CHANGE
);
1107 struct strip_mine sm
;
1108 sm
.sm_first
= !!size
;
1109 sm
.sm_height
= height
;
1111 error
= recursive_scan(ip
, NULL
, &mp
, 0, 0, 1, &sm
);
1116 gfs2_quota_unhold(ip
);
1121 static int trunc_end(struct gfs2_inode
*ip
)
1123 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
1124 struct buffer_head
*dibh
;
1127 error
= gfs2_trans_begin(sdp
, RES_DINODE
, 0);
1131 down_write(&ip
->i_rw_mutex
);
1133 error
= gfs2_meta_inode_buffer(ip
, &dibh
);
1137 if (!i_size_read(&ip
->i_inode
)) {
1139 ip
->i_goal
= ip
->i_no_addr
;
1140 gfs2_buffer_clear_tail(dibh
, sizeof(struct gfs2_dinode
));
1141 gfs2_ordered_del_inode(ip
);
1143 ip
->i_inode
.i_mtime
= ip
->i_inode
.i_ctime
= CURRENT_TIME
;
1144 ip
->i_diskflags
&= ~GFS2_DIF_TRUNC_IN_PROG
;
1146 gfs2_trans_add_meta(ip
->i_gl
, dibh
);
1147 gfs2_dinode_out(ip
, dibh
->b_data
);
1151 up_write(&ip
->i_rw_mutex
);
1152 gfs2_trans_end(sdp
);
1157 * do_shrink - make a file smaller
1159 * @oldsize: the current inode size
1160 * @newsize: the size to make the file
1162 * Called with an exclusive lock on @inode. The @size must
1163 * be equal to or smaller than the current inode size.
1168 static int do_shrink(struct inode
*inode
, u64 oldsize
, u64 newsize
)
1170 struct gfs2_inode
*ip
= GFS2_I(inode
);
1173 error
= trunc_start(inode
, oldsize
, newsize
);
1176 if (gfs2_is_stuffed(ip
))
1179 error
= trunc_dealloc(ip
, newsize
);
1181 error
= trunc_end(ip
);
1186 void gfs2_trim_blocks(struct inode
*inode
)
1188 u64 size
= inode
->i_size
;
1191 ret
= do_shrink(inode
, size
, size
);
1196 * do_grow - Touch and update inode size
1198 * @size: The new size
1200 * This function updates the timestamps on the inode and
1201 * may also increase the size of the inode. This function
1202 * must not be called with @size any smaller than the current
1205 * Although it is not strictly required to unstuff files here,
1206 * earlier versions of GFS2 have a bug in the stuffed file reading
1207 * code which will result in a buffer overrun if the size is larger
1208 * than the max stuffed file size. In order to prevent this from
1209 * occurring, such files are unstuffed, but in other cases we can
1210 * just update the inode size directly.
1212 * Returns: 0 on success, or -ve on error
1215 static int do_grow(struct inode
*inode
, u64 size
)
1217 struct gfs2_inode
*ip
= GFS2_I(inode
);
1218 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
1219 struct gfs2_alloc_parms ap
= { .target
= 1, };
1220 struct buffer_head
*dibh
;
1224 if (gfs2_is_stuffed(ip
) &&
1225 (size
> (sdp
->sd_sb
.sb_bsize
- sizeof(struct gfs2_dinode
)))) {
1226 error
= gfs2_quota_lock_check(ip
);
1230 error
= gfs2_inplace_reserve(ip
, &ap
);
1232 goto do_grow_qunlock
;
1236 error
= gfs2_trans_begin(sdp
, RES_DINODE
+ RES_STATFS
+ RES_RG_BIT
+
1237 (sdp
->sd_args
.ar_quota
== GFS2_QUOTA_OFF
?
1240 goto do_grow_release
;
1243 error
= gfs2_unstuff_dinode(ip
, NULL
);
1248 error
= gfs2_meta_inode_buffer(ip
, &dibh
);
1252 i_size_write(inode
, size
);
1253 ip
->i_inode
.i_mtime
= ip
->i_inode
.i_ctime
= CURRENT_TIME
;
1254 gfs2_trans_add_meta(ip
->i_gl
, dibh
);
1255 gfs2_dinode_out(ip
, dibh
->b_data
);
1259 gfs2_trans_end(sdp
);
1262 gfs2_inplace_release(ip
);
1264 gfs2_quota_unlock(ip
);
1270 * gfs2_setattr_size - make a file a given size
1272 * @newsize: the size to make the file
1274 * The file size can grow, shrink, or stay the same size. This
1275 * is called holding i_mutex and an exclusive glock on the inode
1281 int gfs2_setattr_size(struct inode
*inode
, u64 newsize
)
1283 struct gfs2_inode
*ip
= GFS2_I(inode
);
1287 BUG_ON(!S_ISREG(inode
->i_mode
));
1289 ret
= inode_newsize_ok(inode
, newsize
);
1293 ret
= get_write_access(inode
);
1297 inode_dio_wait(inode
);
1299 ret
= gfs2_rs_alloc(ip
);
1303 oldsize
= inode
->i_size
;
1304 if (newsize
>= oldsize
) {
1305 ret
= do_grow(inode
, newsize
);
1309 gfs2_rs_deltree(ip
->i_res
);
1310 ret
= do_shrink(inode
, oldsize
, newsize
);
1312 put_write_access(inode
);
1316 int gfs2_truncatei_resume(struct gfs2_inode
*ip
)
1319 error
= trunc_dealloc(ip
, i_size_read(&ip
->i_inode
));
1321 error
= trunc_end(ip
);
1325 int gfs2_file_dealloc(struct gfs2_inode
*ip
)
1327 return trunc_dealloc(ip
, 0);
1331 * gfs2_free_journal_extents - Free cached journal bmap info
1336 void gfs2_free_journal_extents(struct gfs2_jdesc
*jd
)
1338 struct gfs2_journal_extent
*jext
;
1340 while(!list_empty(&jd
->extent_list
)) {
1341 jext
= list_entry(jd
->extent_list
.next
, struct gfs2_journal_extent
, list
);
1342 list_del(&jext
->list
);
1348 * gfs2_add_jextent - Add or merge a new extent to extent cache
1349 * @jd: The journal descriptor
1350 * @lblock: The logical block at start of new extent
1351 * @pblock: The physical block at start of new extent
1352 * @blocks: Size of extent in fs blocks
1354 * Returns: 0 on success or -ENOMEM
1357 static int gfs2_add_jextent(struct gfs2_jdesc
*jd
, u64 lblock
, u64 dblock
, u64 blocks
)
1359 struct gfs2_journal_extent
*jext
;
1361 if (!list_empty(&jd
->extent_list
)) {
1362 jext
= list_entry(jd
->extent_list
.prev
, struct gfs2_journal_extent
, list
);
1363 if ((jext
->dblock
+ jext
->blocks
) == dblock
) {
1364 jext
->blocks
+= blocks
;
1369 jext
= kzalloc(sizeof(struct gfs2_journal_extent
), GFP_NOFS
);
1372 jext
->dblock
= dblock
;
1373 jext
->lblock
= lblock
;
1374 jext
->blocks
= blocks
;
1375 list_add_tail(&jext
->list
, &jd
->extent_list
);
1381 * gfs2_map_journal_extents - Cache journal bmap info
1382 * @sdp: The super block
1383 * @jd: The journal to map
1385 * Create a reusable "extent" mapping from all logical
1386 * blocks to all physical blocks for the given journal. This will save
1387 * us time when writing journal blocks. Most journals will have only one
1388 * extent that maps all their logical blocks. That's because gfs2.mkfs
1389 * arranges the journal blocks sequentially to maximize performance.
1390 * So the extent would map the first block for the entire file length.
1391 * However, gfs2_jadd can happen while file activity is happening, so
1392 * those journals may not be sequential. Less likely is the case where
1393 * the users created their own journals by mounting the metafs and
1394 * laying it out. But it's still possible. These journals might have
1397 * Returns: 0 on success, or error on failure
1400 int gfs2_map_journal_extents(struct gfs2_sbd
*sdp
, struct gfs2_jdesc
*jd
)
1404 struct gfs2_inode
*ip
= GFS2_I(jd
->jd_inode
);
1405 struct buffer_head bh
;
1406 unsigned int shift
= sdp
->sd_sb
.sb_bsize_shift
;
1410 lblock_stop
= i_size_read(jd
->jd_inode
) >> shift
;
1411 size
= (lblock_stop
- lblock
) << shift
;
1413 WARN_ON(!list_empty(&jd
->extent_list
));
1419 rc
= gfs2_block_map(jd
->jd_inode
, lblock
, &bh
, 0);
1420 if (rc
|| !buffer_mapped(&bh
))
1422 rc
= gfs2_add_jextent(jd
, lblock
, bh
.b_blocknr
, bh
.b_size
>> shift
);
1426 lblock
+= (bh
.b_size
>> ip
->i_inode
.i_blkbits
);
1429 fs_info(sdp
, "journal %d mapped with %u extents\n", jd
->jd_jid
,
1434 fs_warn(sdp
, "error %d mapping journal %u at offset %llu (extent %u)\n",
1436 (unsigned long long)(i_size_read(jd
->jd_inode
) - size
),
1438 fs_warn(sdp
, "bmap=%d lblock=%llu block=%llu, state=0x%08lx, size=%llu\n",
1439 rc
, (unsigned long long)lblock
, (unsigned long long)bh
.b_blocknr
,
1440 bh
.b_state
, (unsigned long long)bh
.b_size
);
1441 gfs2_free_journal_extents(jd
);
1446 * gfs2_write_alloc_required - figure out if a write will require an allocation
1447 * @ip: the file being written to
1448 * @offset: the offset to write to
1449 * @len: the number of bytes being written
1451 * Returns: 1 if an alloc is required, 0 otherwise
1454 int gfs2_write_alloc_required(struct gfs2_inode
*ip
, u64 offset
,
1457 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
1458 struct buffer_head bh
;
1460 u64 lblock
, lblock_stop
, size
;
1466 if (gfs2_is_stuffed(ip
)) {
1468 sdp
->sd_sb
.sb_bsize
- sizeof(struct gfs2_dinode
))
1473 shift
= sdp
->sd_sb
.sb_bsize_shift
;
1474 BUG_ON(gfs2_is_dir(ip
));
1475 end_of_file
= (i_size_read(&ip
->i_inode
) + sdp
->sd_sb
.sb_bsize
- 1) >> shift
;
1476 lblock
= offset
>> shift
;
1477 lblock_stop
= (offset
+ len
+ sdp
->sd_sb
.sb_bsize
- 1) >> shift
;
1478 if (lblock_stop
> end_of_file
)
1481 size
= (lblock_stop
- lblock
) << shift
;
1485 gfs2_block_map(&ip
->i_inode
, lblock
, &bh
, 0);
1486 if (!buffer_mapped(&bh
))
1489 lblock
+= (bh
.b_size
>> ip
->i_inode
.i_blkbits
);