Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
[linux/fpc-iii.git] / fs / gfs2 / bmap.c
blobfe0500c0af7aab88ca4276234df6894cca7290b3
1 /*
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.
8 */
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>
17 #include "gfs2.h"
18 #include "incore.h"
19 #include "bmap.h"
20 #include "glock.h"
21 #include "inode.h"
22 #include "meta_io.h"
23 #include "quota.h"
24 #include "rgrp.h"
25 #include "log.h"
26 #include "super.h"
27 #include "trans.h"
28 #include "dir.h"
29 #include "util.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
34 * keep it small.
36 struct metapath {
37 struct buffer_head *mp_bh[GFS2_MAX_META_HEIGHT];
38 __u16 mp_list[GFS2_MAX_META_HEIGHT];
41 struct strip_mine {
42 int sm_first;
43 unsigned int sm_height;
46 /**
47 * gfs2_unstuffer_page - unstuff a stuffed inode into a block cached by a page
48 * @ip: the inode
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
53 * Returns: errno
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;
61 int release = 0;
63 if (!page || page->index) {
64 page = find_or_create_page(inode->i_mapping, 0, GFP_NOFS);
65 if (!page)
66 return -ENOMEM;
67 release = 1;
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);
79 kunmap(page);
81 SetPageUptodate(page);
84 if (!page_has_buffers(page))
85 create_empty_buffers(page, 1 << inode->i_blkbits,
86 (1 << BH_Uptodate));
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);
99 if (release) {
100 unlock_page(page);
101 page_cache_release(page);
104 return 0;
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.
115 * Returns: errno
118 int gfs2_unstuff_dinode(struct gfs2_inode *ip, struct page *page)
120 struct buffer_head *bh, *dibh;
121 struct gfs2_dinode *di;
122 u64 block = 0;
123 int isdir = gfs2_is_dir(ip);
124 int error;
126 down_write(&ip->i_rw_mutex);
128 error = gfs2_meta_inode_buffer(ip, &dibh);
129 if (error)
130 goto out;
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 */
136 unsigned int n = 1;
137 error = gfs2_alloc_blocks(ip, &block, &n, 0, NULL);
138 if (error)
139 goto out_brelse;
140 if (isdir) {
141 gfs2_trans_add_unrevoke(GFS2_SB(&ip->i_inode), block, 1);
142 error = gfs2_dir_get_new_buffer(ip, block, &bh);
143 if (error)
144 goto out_brelse;
145 gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_meta_header),
146 dibh, sizeof(struct gfs2_dinode));
147 brelse(bh);
148 } else {
149 error = gfs2_unstuffer_page(ip, dibh, block, page);
150 if (error)
151 goto out_brelse;
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));
167 ip->i_height = 1;
168 di->di_height = cpu_to_be16(1);
170 out_brelse:
171 brelse(dibh);
172 out:
173 up_write(&ip->i_rw_mutex);
174 return error;
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".
188 * Example:
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 * ----------------------------------------
203 * | Dinode | |
204 * | | 4|
205 * | |0 1 2 3 4 5 9|
206 * | | 6|
207 * ----------------------------------------
211 * ----------------------------------------
212 * | Indirect Block |
213 * | 5|
214 * | 4 4 4 4 4 5 5 1|
215 * |0 5 6 7 8 9 0 1 2|
216 * ----------------------------------------
220 * ----------------------------------------
221 * | Indirect Block |
222 * | 1 1 1 1 1 5|
223 * | 6 6 6 6 6 1|
224 * |0 3 4 5 6 7 2|
225 * ----------------------------------------
229 * ----------------------------------------
230 * | Data block containing offset |
231 * | 101342453 |
232 * | |
233 * | |
234 * ----------------------------------------
238 static void find_metapath(const struct gfs2_sbd *sdp, u64 block,
239 struct metapath *mp, unsigned int height)
241 unsigned int i;
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)
251 return 2;
252 return 1;
256 * metapointer - Return pointer to start of metadata in a buffer
257 * @height: The metadata height (0 = dinode)
258 * @mp: The metapath
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
262 * metadata tree.
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);
278 const __be64 *t;
280 for (t = pos; t < endp; t++) {
281 if (!*t)
282 continue;
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);
289 continue;
291 unlock_buffer(rabh);
293 brelse(rabh);
298 * lookup_metapath - Walk the metadata tree to a specific point
299 * @ip: The inode
300 * @mp: The metapath
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;
317 unsigned int x;
318 __be64 *ptr;
319 u64 dblock;
320 int ret;
322 for (x = 0; x < end_of_metadata; x++) {
323 ptr = metapointer(x, mp);
324 dblock = be64_to_cpu(*ptr);
325 if (!dblock)
326 return x + 1;
328 ret = gfs2_meta_indirect_buffer(ip, x+1, dblock, &mp->mp_bh[x+1]);
329 if (ret)
330 return ret;
333 return ip->i_height;
336 static inline void release_metapath(struct metapath *mp)
338 int i;
340 for (i = 0; i < GFS2_MAX_META_HEIGHT; i++) {
341 if (mp->mp_bh[i] == NULL)
342 break;
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);
368 *eob = 0;
369 do {
370 ptr++;
371 if (ptr >= end)
372 break;
373 if (limit && --limit == 0)
374 break;
375 if (d)
376 d++;
377 } while(be64_to_cpu(*ptr) == d);
378 if (ptr >= end)
379 *eob = 1;
380 return (ptr - first);
383 static inline void bmap_lock(struct gfs2_inode *ip, int create)
385 if (create)
386 down_write(&ip->i_rw_mutex);
387 else
388 down_read(&ip->i_rw_mutex);
391 static inline void bmap_unlock(struct gfs2_inode *ip, int create)
393 if (create)
394 up_write(&ip->i_rw_mutex);
395 else
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)));
406 BUG_ON(i < 1);
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));
412 ptr += offset;
413 *ptr = cpu_to_be64(bn);
414 return ptr;
417 enum alloc_state {
418 ALLOC_DATA = 0,
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
429 * @mp: The metapath
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
437 * iii) Data blocks
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];
458 u64 bn, dblock = 0;
459 unsigned n, i, blks, alloced = 0, iblks = 0, branch_start = 0;
460 unsigned dblks = 0;
461 unsigned ptrs_per_blk;
462 const unsigned end_of_metadata = height - 1;
463 int ret;
464 int eob = 0;
465 enum alloc_state state;
466 __be64 *ptr;
467 __be64 zero_bn = 0;
469 BUG_ON(sheight < 1);
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,
480 &eob);
481 BUG_ON(dblks < 1);
482 state = ALLOC_DATA;
483 } else {
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;
491 } else {
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;
503 i = sheight;
504 do {
505 int error;
506 n = blks - alloced;
507 error = gfs2_alloc_blocks(ip, &bn, &n, 0, NULL);
508 if (error)
509 return error;
510 alloced += n;
511 if (state != ALLOC_DATA || gfs2_is_jdata(ip))
512 gfs2_trans_add_unrevoke(sdp, bn, n);
513 switch (state) {
514 /* Growing height of tree */
515 case ALLOC_GROW_HEIGHT:
516 if (i == 1) {
517 ptr = (__be64 *)(dibh->b_data +
518 sizeof(struct gfs2_dinode));
519 zero_bn = *ptr;
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) {
524 i--;
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) +
530 sizeof(__be64));
531 ptr = (__be64 *)(mp->mp_bh[i]->b_data +
532 sizeof(struct gfs2_meta_header));
533 *ptr = zero_bn;
534 state = ALLOC_GROW_DEPTH;
535 for(i = branch_start; i < height; i++) {
536 if (mp->mp_bh[i] == NULL)
537 break;
538 brelse(mp->mp_bh[i]);
539 mp->mp_bh[i] = NULL;
541 i = branch_start;
543 if (n == 0)
544 break;
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++);
552 if (i == height)
553 state = ALLOC_DATA;
554 if (n == 0)
555 break;
556 /* Tree complete, adding data blocks */
557 case ALLOC_DATA:
558 BUG_ON(n > dblks);
559 BUG_ON(mp->mp_bh[end_of_metadata] == NULL);
560 gfs2_trans_add_meta(ip->i_gl, mp->mp_bh[end_of_metadata]);
561 dblks = n;
562 ptr = metapointer(end_of_metadata, mp);
563 dblock = bn;
564 while (n-- > 0)
565 *ptr++ = cpu_to_be64(bn++);
566 if (buffer_zeronew(bh_map)) {
567 ret = sb_issue_zeroout(sb, dblock, dblks,
568 GFP_NOFS);
569 if (ret) {
570 fs_err(sdp,
571 "Failed to zero data buffers\n");
572 clear_buffer_zeronew(bh_map);
575 break;
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);
585 return 0;
589 * gfs2_block_map - Map a block from an inode to a disk block
590 * @inode: The inode
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.
599 * Returns: errno
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;
610 __be64 *ptr;
611 u64 size;
612 struct metapath mp;
613 int ret;
614 int eob;
615 unsigned int len;
616 struct buffer_head *bh;
617 u8 height;
619 BUG_ON(maxlen == 0);
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]);
633 if (ret)
634 goto out;
636 height = ip->i_height;
637 size = (lblock + 1) * bsize;
638 while (size > arr[height])
639 height++;
640 find_metapath(sdp, lblock, &mp, height);
641 ret = 1;
642 if (height > ip->i_height || gfs2_is_stuffed(ip))
643 goto do_alloc;
644 ret = lookup_metapath(ip, &mp);
645 if (ret < 0)
646 goto out;
647 if (ret != ip->i_height)
648 goto do_alloc;
649 ptr = metapointer(ip->i_height - 1, &mp);
650 if (*ptr == 0)
651 goto do_alloc;
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);
656 if (eob)
657 set_buffer_boundary(bh_map);
658 ret = 0;
659 out:
660 release_metapath(&mp);
661 trace_gfs2_bmap(ip, bh_map, lblock, create, ret);
662 bmap_unlock(ip, create);
663 return ret;
665 do_alloc:
666 /* All allocations are done here, firstly check create flag */
667 if (!create) {
668 BUG_ON(gfs2_is_stuffed(ip));
669 ret = 0;
670 goto out;
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);
675 goto out;
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 };
684 int ret;
685 int create = *new;
687 BUG_ON(!extlen);
688 BUG_ON(!dblock);
689 BUG_ON(!new);
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;
695 if (buffer_new(&bh))
696 *new = 1;
697 else
698 *new = 0;
699 return ret;
703 * do_strip - Look for a layer a particular layer of the file and strip it off
704 * @ip: the inode
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
712 * Returns: errno
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;
721 u64 bn, bstart;
722 u32 blen, btotal;
723 __be64 *p;
724 unsigned int rg_blocks = 0;
725 int metadata;
726 unsigned int revokes = 0;
727 int x;
728 int error;
730 error = gfs2_rindex_update(sdp);
731 if (error)
732 return error;
734 if (!*top)
735 sm->sm_first = 0;
737 if (height != sm->sm_height)
738 return 0;
740 if (sm->sm_first) {
741 top++;
742 sm->sm_first = 0;
745 metadata = (height != ip->i_height - 1);
746 if (metadata)
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));
752 bstart = 0;
753 blen = 0;
755 for (p = top; p < bottom; p++) {
756 if (!*p)
757 continue;
759 bn = be64_to_cpu(*p);
761 if (bstart + blen == bn)
762 blen++;
763 else {
764 if (bstart)
765 gfs2_rlist_add(ip, &rlist, bstart);
767 bstart = bn;
768 blen = 1;
772 if (bstart)
773 gfs2_rlist_add(ip, &rlist, bstart);
774 else
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);
786 if (error)
787 goto out_rlist;
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,
794 revokes);
795 if (error)
796 goto out_rg_gunlock;
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);
803 bstart = 0;
804 blen = 0;
805 btotal = 0;
807 for (p = top; p < bottom; p++) {
808 if (!*p)
809 continue;
811 bn = be64_to_cpu(*p);
813 if (bstart + blen == bn)
814 blen++;
815 else {
816 if (bstart) {
817 __gfs2_free_blocks(ip, bstart, blen, metadata);
818 btotal += blen;
821 bstart = bn;
822 blen = 1;
825 *p = 0;
826 gfs2_add_inode_blocks(&ip->i_inode, -1);
828 if (bstart) {
829 __gfs2_free_blocks(ip, bstart, blen, metadata);
830 btotal += blen;
833 gfs2_statfs_change(sdp, 0, +btotal, 0);
834 gfs2_quota_change(ip, -(s64)btotal, ip->i_inode.i_uid,
835 ip->i_inode.i_gid);
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);
843 gfs2_trans_end(sdp);
845 out_rg_gunlock:
846 gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs);
847 out_rlist:
848 gfs2_rlist_free(&rlist);
849 out:
850 return error;
854 * recursive_scan - recursively scan through the end of a file
855 * @ip: the inode
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.
866 * Returns: errno
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;
876 u64 bn;
877 int error;
878 int mh_size = sizeof(struct gfs2_meta_header);
880 if (!height) {
881 error = gfs2_meta_inode_buffer(ip, &bh);
882 if (error)
883 return error;
884 dibh = 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;
888 } else {
889 error = gfs2_meta_indirect_buffer(ip, height, block, &bh);
890 if (error)
891 return error;
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);
900 if (error)
901 goto out;
903 if (height < ip->i_height - 1) {
905 gfs2_metapath_ra(ip->i_gl, bh, top);
907 for (; top < bottom; top++, first = 0) {
908 if (!*top)
909 continue;
911 bn = be64_to_cpu(*top);
913 error = recursive_scan(ip, dibh, mp, height + 1, bn,
914 first, sm);
915 if (error)
916 break;
919 out:
920 brelse(bh);
921 return error;
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;
938 struct page *page;
939 int err;
941 page = find_or_create_page(mapping, index, GFP_NOFS);
942 if (!page)
943 return 0;
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);
954 pos = blocksize;
955 while (offset >= pos) {
956 bh = bh->b_this_page;
957 iblock++;
958 pos += blocksize;
961 err = 0;
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))
967 goto unlock;
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)) {
975 err = -EIO;
976 ll_rw_block(READ, 1, &bh);
977 wait_on_buffer(bh);
978 /* Uhhuh. Read error. Complain and punt. */
979 if (!buffer_uptodate(bh))
980 goto unlock;
981 err = 0;
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);
989 unlock:
990 unlock_page(page);
991 page_cache_release(page);
992 return err;
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;
1012 u64 chunk;
1013 int error;
1015 while (oldsize != newsize) {
1016 chunk = oldsize - newsize;
1017 if (chunk > max_chunk)
1018 chunk = max_chunk;
1019 truncate_pagecache(inode, oldsize - chunk);
1020 oldsize -= chunk;
1021 gfs2_trans_end(sdp);
1022 error = gfs2_trans_begin(sdp, RES_DINODE, GFS2_JTRUNC_REVOKES);
1023 if (error)
1024 return error;
1027 return 0;
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);
1037 int error;
1039 if (journaled)
1040 error = gfs2_trans_begin(sdp, RES_DINODE + RES_JDATA, GFS2_JTRUNC_REVOKES);
1041 else
1042 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1043 if (error)
1044 return error;
1046 error = gfs2_meta_inode_buffer(ip, &dibh);
1047 if (error)
1048 goto out;
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);
1054 } else {
1055 if (newsize & (u64)(sdp->sd_sb.sb_bsize - 1)) {
1056 error = gfs2_block_truncate_page(mapping, newsize);
1057 if (error)
1058 goto out_brelse;
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);
1067 if (journaled)
1068 error = gfs2_journaled_truncate(inode, oldsize, newsize);
1069 else
1070 truncate_pagecache(inode, newsize);
1072 if (error) {
1073 brelse(dibh);
1074 return error;
1077 out_brelse:
1078 brelse(dibh);
1079 out:
1080 gfs2_trans_end(sdp);
1081 return error;
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;
1088 u64 lblock;
1089 struct metapath mp;
1090 int error;
1092 if (!size)
1093 lblock = 0;
1094 else
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);
1099 if (error)
1100 return error;
1102 error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1103 if (error)
1104 return error;
1106 while (height--) {
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);
1112 if (error)
1113 break;
1116 gfs2_quota_unhold(ip);
1118 return error;
1121 static int trunc_end(struct gfs2_inode *ip)
1123 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1124 struct buffer_head *dibh;
1125 int error;
1127 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1128 if (error)
1129 return error;
1131 down_write(&ip->i_rw_mutex);
1133 error = gfs2_meta_inode_buffer(ip, &dibh);
1134 if (error)
1135 goto out;
1137 if (!i_size_read(&ip->i_inode)) {
1138 ip->i_height = 0;
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);
1148 brelse(dibh);
1150 out:
1151 up_write(&ip->i_rw_mutex);
1152 gfs2_trans_end(sdp);
1153 return error;
1157 * do_shrink - make a file smaller
1158 * @inode: the inode
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.
1165 * Returns: errno
1168 static int do_shrink(struct inode *inode, u64 oldsize, u64 newsize)
1170 struct gfs2_inode *ip = GFS2_I(inode);
1171 int error;
1173 error = trunc_start(inode, oldsize, newsize);
1174 if (error < 0)
1175 return error;
1176 if (gfs2_is_stuffed(ip))
1177 return 0;
1179 error = trunc_dealloc(ip, newsize);
1180 if (error == 0)
1181 error = trunc_end(ip);
1183 return error;
1186 void gfs2_trim_blocks(struct inode *inode)
1188 u64 size = inode->i_size;
1189 int ret;
1191 ret = do_shrink(inode, size, size);
1192 WARN_ON(ret != 0);
1196 * do_grow - Touch and update inode size
1197 * @inode: The inode
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
1203 * inode size.
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;
1221 int error;
1222 int unstuff = 0;
1224 if (gfs2_is_stuffed(ip) &&
1225 (size > (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)))) {
1226 error = gfs2_quota_lock_check(ip);
1227 if (error)
1228 return error;
1230 error = gfs2_inplace_reserve(ip, &ap);
1231 if (error)
1232 goto do_grow_qunlock;
1233 unstuff = 1;
1236 error = gfs2_trans_begin(sdp, RES_DINODE + RES_STATFS + RES_RG_BIT +
1237 (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF ?
1238 0 : RES_QUOTA), 0);
1239 if (error)
1240 goto do_grow_release;
1242 if (unstuff) {
1243 error = gfs2_unstuff_dinode(ip, NULL);
1244 if (error)
1245 goto do_end_trans;
1248 error = gfs2_meta_inode_buffer(ip, &dibh);
1249 if (error)
1250 goto do_end_trans;
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);
1256 brelse(dibh);
1258 do_end_trans:
1259 gfs2_trans_end(sdp);
1260 do_grow_release:
1261 if (unstuff) {
1262 gfs2_inplace_release(ip);
1263 do_grow_qunlock:
1264 gfs2_quota_unlock(ip);
1266 return error;
1270 * gfs2_setattr_size - make a file a given size
1271 * @inode: the inode
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
1276 * in question.
1278 * Returns: errno
1281 int gfs2_setattr_size(struct inode *inode, u64 newsize)
1283 struct gfs2_inode *ip = GFS2_I(inode);
1284 int ret;
1285 u64 oldsize;
1287 BUG_ON(!S_ISREG(inode->i_mode));
1289 ret = inode_newsize_ok(inode, newsize);
1290 if (ret)
1291 return ret;
1293 ret = get_write_access(inode);
1294 if (ret)
1295 return ret;
1297 inode_dio_wait(inode);
1299 ret = gfs2_rs_alloc(ip);
1300 if (ret)
1301 goto out;
1303 oldsize = inode->i_size;
1304 if (newsize >= oldsize) {
1305 ret = do_grow(inode, newsize);
1306 goto out;
1309 gfs2_rs_deltree(ip->i_res);
1310 ret = do_shrink(inode, oldsize, newsize);
1311 out:
1312 put_write_access(inode);
1313 return ret;
1316 int gfs2_truncatei_resume(struct gfs2_inode *ip)
1318 int error;
1319 error = trunc_dealloc(ip, i_size_read(&ip->i_inode));
1320 if (!error)
1321 error = trunc_end(ip);
1322 return error;
1325 int gfs2_file_dealloc(struct gfs2_inode *ip)
1327 return trunc_dealloc(ip, 0);
1331 * gfs2_write_alloc_required - figure out if a write will require an allocation
1332 * @ip: the file being written to
1333 * @offset: the offset to write to
1334 * @len: the number of bytes being written
1336 * Returns: 1 if an alloc is required, 0 otherwise
1339 int gfs2_write_alloc_required(struct gfs2_inode *ip, u64 offset,
1340 unsigned int len)
1342 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1343 struct buffer_head bh;
1344 unsigned int shift;
1345 u64 lblock, lblock_stop, size;
1346 u64 end_of_file;
1348 if (!len)
1349 return 0;
1351 if (gfs2_is_stuffed(ip)) {
1352 if (offset + len >
1353 sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode))
1354 return 1;
1355 return 0;
1358 shift = sdp->sd_sb.sb_bsize_shift;
1359 BUG_ON(gfs2_is_dir(ip));
1360 end_of_file = (i_size_read(&ip->i_inode) + sdp->sd_sb.sb_bsize - 1) >> shift;
1361 lblock = offset >> shift;
1362 lblock_stop = (offset + len + sdp->sd_sb.sb_bsize - 1) >> shift;
1363 if (lblock_stop > end_of_file)
1364 return 1;
1366 size = (lblock_stop - lblock) << shift;
1367 do {
1368 bh.b_state = 0;
1369 bh.b_size = size;
1370 gfs2_block_map(&ip->i_inode, lblock, &bh, 0);
1371 if (!buffer_mapped(&bh))
1372 return 1;
1373 size -= bh.b_size;
1374 lblock += (bh.b_size >> ip->i_inode.i_blkbits);
1375 } while(size > 0);
1377 return 0;