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_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);
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, size_t 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,
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, (size_t)(ptrs_per_blk -
487 mp->mp_list[end_of_metadata]));
488 if (height == ip->i_height) {
489 /* Writing into existing tree, extend tree down */
490 iblks = height - sheight;
491 state = ALLOC_GROW_DEPTH;
493 /* Building up tree height */
494 state = ALLOC_GROW_HEIGHT;
495 iblks = height - ip->i_height;
496 branch_start = metapath_branch_start(mp);
497 iblks += (height - branch_start);
501 /* start of the second part of the function (state machine) */
503 blks = dblks + iblks;
508 error = gfs2_alloc_blocks(ip, &bn, &n, 0, NULL);
512 if (state != ALLOC_DATA || gfs2_is_jdata(ip))
513 gfs2_trans_add_unrevoke(sdp, bn, n);
515 /* Growing height of tree */
516 case ALLOC_GROW_HEIGHT:
518 ptr = (__be64 *)(dibh->b_data +
519 sizeof(struct gfs2_dinode));
522 for (; i - 1 < height - ip->i_height && n > 0; i++, n--)
523 gfs2_indirect_init(mp, ip->i_gl, i, 0, bn++);
524 if (i - 1 == height - ip->i_height) {
526 gfs2_buffer_copy_tail(mp->mp_bh[i],
527 sizeof(struct gfs2_meta_header),
528 dibh, sizeof(struct gfs2_dinode));
529 gfs2_buffer_clear_tail(dibh,
530 sizeof(struct gfs2_dinode) +
532 ptr = (__be64 *)(mp->mp_bh[i]->b_data +
533 sizeof(struct gfs2_meta_header));
535 state = ALLOC_GROW_DEPTH;
536 for(i = branch_start; i < height; i++) {
537 if (mp->mp_bh[i] == NULL)
539 brelse(mp->mp_bh[i]);
546 /* Branching from existing tree */
547 case ALLOC_GROW_DEPTH:
548 if (i > 1 && i < height)
549 gfs2_trans_add_meta(ip->i_gl, mp->mp_bh[i-1]);
550 for (; i < height && n > 0; i++, n--)
551 gfs2_indirect_init(mp, ip->i_gl, i,
552 mp->mp_list[i-1], bn++);
557 /* Tree complete, adding data blocks */
560 BUG_ON(mp->mp_bh[end_of_metadata] == NULL);
561 gfs2_trans_add_meta(ip->i_gl, mp->mp_bh[end_of_metadata]);
563 ptr = metapointer(end_of_metadata, mp);
566 *ptr++ = cpu_to_be64(bn++);
567 if (buffer_zeronew(bh_map)) {
568 ret = sb_issue_zeroout(sb, dblock, dblks,
572 "Failed to zero data buffers\n");
573 clear_buffer_zeronew(bh_map);
578 } while ((state != ALLOC_DATA) || !dblock);
580 ip->i_height = height;
581 gfs2_add_inode_blocks(&ip->i_inode, alloced);
582 gfs2_dinode_out(ip, mp->mp_bh[0]->b_data);
583 map_bh(bh_map, inode->i_sb, dblock);
584 bh_map->b_size = dblks << inode->i_blkbits;
585 set_buffer_new(bh_map);
590 * gfs2_block_map - Map a block from an inode to a disk block
592 * @lblock: The logical block number
593 * @bh_map: The bh to be mapped
594 * @create: True if its ok to alloc blocks to satify the request
596 * Sets buffer_mapped() if successful, sets buffer_boundary() if a
597 * read of metadata will be required before the next block can be
598 * mapped. Sets buffer_new() if new blocks were allocated.
603 int gfs2_block_map(struct inode *inode, sector_t lblock,
604 struct buffer_head *bh_map, int create)
606 struct gfs2_inode *ip = GFS2_I(inode);
607 struct gfs2_sbd *sdp = GFS2_SB(inode);
608 unsigned int bsize = sdp->sd_sb.sb_bsize;
609 const size_t maxlen = bh_map->b_size >> inode->i_blkbits;
610 const u64 *arr = sdp->sd_heightsize;
617 struct buffer_head *bh;
622 memset(mp.mp_bh, 0, sizeof(mp.mp_bh));
623 bmap_lock(ip, create);
624 clear_buffer_mapped(bh_map);
625 clear_buffer_new(bh_map);
626 clear_buffer_boundary(bh_map);
627 trace_gfs2_bmap(ip, bh_map, lblock, create, 1);
628 if (gfs2_is_dir(ip)) {
629 bsize = sdp->sd_jbsize;
630 arr = sdp->sd_jheightsize;
633 ret = gfs2_meta_inode_buffer(ip, &mp.mp_bh[0]);
637 height = ip->i_height;
638 size = (lblock + 1) * bsize;
639 while (size > arr[height])
641 find_metapath(sdp, lblock, &mp, height);
643 if (height > ip->i_height || gfs2_is_stuffed(ip))
645 ret = lookup_metapath(ip, &mp);
648 if (ret != ip->i_height)
650 ptr = metapointer(ip->i_height - 1, &mp);
653 map_bh(bh_map, inode->i_sb, be64_to_cpu(*ptr));
654 bh = mp.mp_bh[ip->i_height - 1];
655 len = gfs2_extent_length(bh->b_data, bh->b_size, ptr, maxlen, &eob);
656 bh_map->b_size = (len << inode->i_blkbits);
658 set_buffer_boundary(bh_map);
661 release_metapath(&mp);
662 trace_gfs2_bmap(ip, bh_map, lblock, create, ret);
663 bmap_unlock(ip, create);
667 /* All allocations are done here, firstly check create flag */
669 BUG_ON(gfs2_is_stuffed(ip));
674 /* At this point ret is the tree depth of already allocated blocks */
675 ret = gfs2_bmap_alloc(inode, lblock, bh_map, &mp, ret, height, maxlen);
680 * Deprecated: do not use in new code
682 int gfs2_extent_map(struct inode *inode, u64 lblock, int *new, u64 *dblock, unsigned *extlen)
684 struct buffer_head bh = { .b_state = 0, .b_blocknr = 0 };
692 bh.b_size = 1 << (inode->i_blkbits + (create ? 0 : 5));
693 ret = gfs2_block_map(inode, lblock, &bh, create);
694 *extlen = bh.b_size >> inode->i_blkbits;
695 *dblock = bh.b_blocknr;
704 * do_strip - Look for a layer a particular layer of the file and strip it off
706 * @dibh: the dinode buffer
707 * @bh: A buffer of pointers
708 * @top: The first pointer in the buffer
709 * @bottom: One more than the last pointer
710 * @height: the height this buffer is at
711 * @sm: a pointer to a struct strip_mine
716 static int do_strip(struct gfs2_inode *ip, struct buffer_head *dibh,
717 struct buffer_head *bh, __be64 *top, __be64 *bottom,
718 unsigned int height, struct strip_mine *sm)
720 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
721 struct gfs2_rgrp_list rlist;
725 unsigned int rg_blocks = 0;
727 unsigned int revokes = 0;
731 error = gfs2_rindex_update(sdp);
738 if (height != sm->sm_height)
746 metadata = (height != ip->i_height - 1);
748 revokes = (height) ? sdp->sd_inptrs : sdp->sd_diptrs;
749 else if (ip->i_depth)
750 revokes = sdp->sd_inptrs;
752 memset(&rlist, 0, sizeof(struct gfs2_rgrp_list));
756 for (p = top; p < bottom; p++) {
760 bn = be64_to_cpu(*p);
762 if (bstart + blen == bn)
766 gfs2_rlist_add(ip, &rlist, bstart);
774 gfs2_rlist_add(ip, &rlist, bstart);
776 goto out; /* Nothing to do */
778 gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE);
780 for (x = 0; x < rlist.rl_rgrps; x++) {
781 struct gfs2_rgrpd *rgd;
782 rgd = rlist.rl_ghs[x].gh_gl->gl_object;
783 rg_blocks += rgd->rd_length;
786 error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs);
790 if (gfs2_rs_active(&ip->i_res)) /* needs to be done with the rgrp glock held */
791 gfs2_rs_deltree(&ip->i_res);
793 error = gfs2_trans_begin(sdp, rg_blocks + RES_DINODE +
794 RES_INDIRECT + RES_STATFS + RES_QUOTA,
799 down_write(&ip->i_rw_mutex);
801 gfs2_trans_add_meta(ip->i_gl, dibh);
802 gfs2_trans_add_meta(ip->i_gl, bh);
808 for (p = top; p < bottom; p++) {
812 bn = be64_to_cpu(*p);
814 if (bstart + blen == bn)
818 __gfs2_free_blocks(ip, bstart, blen, metadata);
827 gfs2_add_inode_blocks(&ip->i_inode, -1);
830 __gfs2_free_blocks(ip, bstart, blen, metadata);
834 gfs2_statfs_change(sdp, 0, +btotal, 0);
835 gfs2_quota_change(ip, -(s64)btotal, ip->i_inode.i_uid,
838 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
840 gfs2_dinode_out(ip, dibh->b_data);
842 up_write(&ip->i_rw_mutex);
847 gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs);
849 gfs2_rlist_free(&rlist);
855 * recursive_scan - recursively scan through the end of a file
857 * @dibh: the dinode buffer
858 * @mp: the path through the metadata to the point to start
859 * @height: the height the recursion is at
860 * @block: the indirect block to look at
861 * @first: 1 if this is the first block
862 * @sm: data opaque to this function to pass to @bc
864 * When this is first called @height and @block should be zero and
865 * @first should be 1.
870 static int recursive_scan(struct gfs2_inode *ip, struct buffer_head *dibh,
871 struct metapath *mp, unsigned int height,
872 u64 block, int first, struct strip_mine *sm)
874 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
875 struct buffer_head *bh = NULL;
876 __be64 *top, *bottom;
879 int mh_size = sizeof(struct gfs2_meta_header);
882 error = gfs2_meta_inode_buffer(ip, &bh);
887 top = (__be64 *)(bh->b_data + sizeof(struct gfs2_dinode)) + mp->mp_list[0];
888 bottom = (__be64 *)(bh->b_data + sizeof(struct gfs2_dinode)) + sdp->sd_diptrs;
890 error = gfs2_meta_indirect_buffer(ip, height, block, &bh);
894 top = (__be64 *)(bh->b_data + mh_size) +
895 (first ? mp->mp_list[height] : 0);
897 bottom = (__be64 *)(bh->b_data + mh_size) + sdp->sd_inptrs;
900 error = do_strip(ip, dibh, bh, top, bottom, height, sm);
904 if (height < ip->i_height - 1) {
906 gfs2_metapath_ra(ip->i_gl, bh, top);
908 for (; top < bottom; top++, first = 0) {
912 bn = be64_to_cpu(*top);
914 error = recursive_scan(ip, dibh, mp, height + 1, bn,
927 * gfs2_block_truncate_page - Deal with zeroing out data for truncate
929 * This is partly borrowed from ext3.
931 static int gfs2_block_truncate_page(struct address_space *mapping, loff_t from)
933 struct inode *inode = mapping->host;
934 struct gfs2_inode *ip = GFS2_I(inode);
935 unsigned long index = from >> PAGE_SHIFT;
936 unsigned offset = from & (PAGE_SIZE-1);
937 unsigned blocksize, iblock, length, pos;
938 struct buffer_head *bh;
942 page = find_or_create_page(mapping, index, GFP_NOFS);
946 blocksize = inode->i_sb->s_blocksize;
947 length = blocksize - (offset & (blocksize - 1));
948 iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
950 if (!page_has_buffers(page))
951 create_empty_buffers(page, blocksize, 0);
953 /* Find the buffer that contains "offset" */
954 bh = page_buffers(page);
956 while (offset >= pos) {
957 bh = bh->b_this_page;
964 if (!buffer_mapped(bh)) {
965 gfs2_block_map(inode, iblock, bh, 0);
966 /* unmapped? It's a hole - nothing to do */
967 if (!buffer_mapped(bh))
971 /* Ok, it's mapped. Make sure it's up-to-date */
972 if (PageUptodate(page))
973 set_buffer_uptodate(bh);
975 if (!buffer_uptodate(bh)) {
977 ll_rw_block(READ, 1, &bh);
979 /* Uhhuh. Read error. Complain and punt. */
980 if (!buffer_uptodate(bh))
985 if (!gfs2_is_writeback(ip))
986 gfs2_trans_add_data(ip->i_gl, bh);
988 zero_user(page, offset, length);
989 mark_buffer_dirty(bh);
996 #define GFS2_JTRUNC_REVOKES 8192
999 * gfs2_journaled_truncate - Wrapper for truncate_pagecache for jdata files
1000 * @inode: The inode being truncated
1001 * @oldsize: The original (larger) size
1002 * @newsize: The new smaller size
1004 * With jdata files, we have to journal a revoke for each block which is
1005 * truncated. As a result, we need to split this into separate transactions
1006 * if the number of pages being truncated gets too large.
1009 static int gfs2_journaled_truncate(struct inode *inode, u64 oldsize, u64 newsize)
1011 struct gfs2_sbd *sdp = GFS2_SB(inode);
1012 u64 max_chunk = GFS2_JTRUNC_REVOKES * sdp->sd_vfs->s_blocksize;
1016 while (oldsize != newsize) {
1017 chunk = oldsize - newsize;
1018 if (chunk > max_chunk)
1020 truncate_pagecache(inode, oldsize - chunk);
1022 gfs2_trans_end(sdp);
1023 error = gfs2_trans_begin(sdp, RES_DINODE, GFS2_JTRUNC_REVOKES);
1031 static int trunc_start(struct inode *inode, u64 oldsize, u64 newsize)
1033 struct gfs2_inode *ip = GFS2_I(inode);
1034 struct gfs2_sbd *sdp = GFS2_SB(inode);
1035 struct address_space *mapping = inode->i_mapping;
1036 struct buffer_head *dibh;
1037 int journaled = gfs2_is_jdata(ip);
1041 error = gfs2_trans_begin(sdp, RES_DINODE + RES_JDATA, GFS2_JTRUNC_REVOKES);
1043 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1047 error = gfs2_meta_inode_buffer(ip, &dibh);
1051 gfs2_trans_add_meta(ip->i_gl, dibh);
1053 if (gfs2_is_stuffed(ip)) {
1054 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode) + newsize);
1056 if (newsize & (u64)(sdp->sd_sb.sb_bsize - 1)) {
1057 error = gfs2_block_truncate_page(mapping, newsize);
1061 ip->i_diskflags |= GFS2_DIF_TRUNC_IN_PROG;
1064 i_size_write(inode, newsize);
1065 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
1066 gfs2_dinode_out(ip, dibh->b_data);
1069 error = gfs2_journaled_truncate(inode, oldsize, newsize);
1071 truncate_pagecache(inode, newsize);
1081 gfs2_trans_end(sdp);
1085 static int trunc_dealloc(struct gfs2_inode *ip, u64 size)
1087 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1088 unsigned int height = ip->i_height;
1096 lblock = (size - 1) >> sdp->sd_sb.sb_bsize_shift;
1098 find_metapath(sdp, lblock, &mp, ip->i_height);
1099 error = gfs2_rindex_update(sdp);
1103 error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1108 struct strip_mine sm;
1109 sm.sm_first = !!size;
1110 sm.sm_height = height;
1112 error = recursive_scan(ip, NULL, &mp, 0, 0, 1, &sm);
1117 gfs2_quota_unhold(ip);
1122 static int trunc_end(struct gfs2_inode *ip)
1124 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1125 struct buffer_head *dibh;
1128 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1132 down_write(&ip->i_rw_mutex);
1134 error = gfs2_meta_inode_buffer(ip, &dibh);
1138 if (!i_size_read(&ip->i_inode)) {
1140 ip->i_goal = ip->i_no_addr;
1141 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
1142 gfs2_ordered_del_inode(ip);
1144 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
1145 ip->i_diskflags &= ~GFS2_DIF_TRUNC_IN_PROG;
1147 gfs2_trans_add_meta(ip->i_gl, dibh);
1148 gfs2_dinode_out(ip, dibh->b_data);
1152 up_write(&ip->i_rw_mutex);
1153 gfs2_trans_end(sdp);
1158 * do_shrink - make a file smaller
1160 * @oldsize: the current inode size
1161 * @newsize: the size to make the file
1163 * Called with an exclusive lock on @inode. The @size must
1164 * be equal to or smaller than the current inode size.
1169 static int do_shrink(struct inode *inode, u64 oldsize, u64 newsize)
1171 struct gfs2_inode *ip = GFS2_I(inode);
1174 error = trunc_start(inode, oldsize, newsize);
1177 if (gfs2_is_stuffed(ip))
1180 error = trunc_dealloc(ip, newsize);
1182 error = trunc_end(ip);
1187 void gfs2_trim_blocks(struct inode *inode)
1189 u64 size = inode->i_size;
1192 ret = do_shrink(inode, size, size);
1197 * do_grow - Touch and update inode size
1199 * @size: The new size
1201 * This function updates the timestamps on the inode and
1202 * may also increase the size of the inode. This function
1203 * must not be called with @size any smaller than the current
1206 * Although it is not strictly required to unstuff files here,
1207 * earlier versions of GFS2 have a bug in the stuffed file reading
1208 * code which will result in a buffer overrun if the size is larger
1209 * than the max stuffed file size. In order to prevent this from
1210 * occurring, such files are unstuffed, but in other cases we can
1211 * just update the inode size directly.
1213 * Returns: 0 on success, or -ve on error
1216 static int do_grow(struct inode *inode, u64 size)
1218 struct gfs2_inode *ip = GFS2_I(inode);
1219 struct gfs2_sbd *sdp = GFS2_SB(inode);
1220 struct gfs2_alloc_parms ap = { .target = 1, };
1221 struct buffer_head *dibh;
1225 if (gfs2_is_stuffed(ip) &&
1226 (size > (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)))) {
1227 error = gfs2_quota_lock_check(ip, &ap);
1231 error = gfs2_inplace_reserve(ip, &ap);
1233 goto do_grow_qunlock;
1237 error = gfs2_trans_begin(sdp, RES_DINODE + RES_STATFS + RES_RG_BIT +
1238 (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF ?
1241 goto do_grow_release;
1244 error = gfs2_unstuff_dinode(ip, NULL);
1249 error = gfs2_meta_inode_buffer(ip, &dibh);
1253 i_size_write(inode, size);
1254 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
1255 gfs2_trans_add_meta(ip->i_gl, dibh);
1256 gfs2_dinode_out(ip, dibh->b_data);
1260 gfs2_trans_end(sdp);
1263 gfs2_inplace_release(ip);
1265 gfs2_quota_unlock(ip);
1271 * gfs2_setattr_size - make a file a given size
1273 * @newsize: the size to make the file
1275 * The file size can grow, shrink, or stay the same size. This
1276 * is called holding i_mutex and an exclusive glock on the inode
1282 int gfs2_setattr_size(struct inode *inode, u64 newsize)
1284 struct gfs2_inode *ip = GFS2_I(inode);
1288 BUG_ON(!S_ISREG(inode->i_mode));
1290 ret = inode_newsize_ok(inode, newsize);
1294 inode_dio_wait(inode);
1296 ret = gfs2_rsqa_alloc(ip);
1300 oldsize = inode->i_size;
1301 if (newsize >= oldsize) {
1302 ret = do_grow(inode, newsize);
1306 ret = do_shrink(inode, oldsize, newsize);
1308 gfs2_rsqa_delete(ip, NULL);
1312 int gfs2_truncatei_resume(struct gfs2_inode *ip)
1315 error = trunc_dealloc(ip, i_size_read(&ip->i_inode));
1317 error = trunc_end(ip);
1321 int gfs2_file_dealloc(struct gfs2_inode *ip)
1323 return trunc_dealloc(ip, 0);
1327 * gfs2_free_journal_extents - Free cached journal bmap info
1332 void gfs2_free_journal_extents(struct gfs2_jdesc *jd)
1334 struct gfs2_journal_extent *jext;
1336 while(!list_empty(&jd->extent_list)) {
1337 jext = list_entry(jd->extent_list.next, struct gfs2_journal_extent, list);
1338 list_del(&jext->list);
1344 * gfs2_add_jextent - Add or merge a new extent to extent cache
1345 * @jd: The journal descriptor
1346 * @lblock: The logical block at start of new extent
1347 * @dblock: The physical block at start of new extent
1348 * @blocks: Size of extent in fs blocks
1350 * Returns: 0 on success or -ENOMEM
1353 static int gfs2_add_jextent(struct gfs2_jdesc *jd, u64 lblock, u64 dblock, u64 blocks)
1355 struct gfs2_journal_extent *jext;
1357 if (!list_empty(&jd->extent_list)) {
1358 jext = list_entry(jd->extent_list.prev, struct gfs2_journal_extent, list);
1359 if ((jext->dblock + jext->blocks) == dblock) {
1360 jext->blocks += blocks;
1365 jext = kzalloc(sizeof(struct gfs2_journal_extent), GFP_NOFS);
1368 jext->dblock = dblock;
1369 jext->lblock = lblock;
1370 jext->blocks = blocks;
1371 list_add_tail(&jext->list, &jd->extent_list);
1377 * gfs2_map_journal_extents - Cache journal bmap info
1378 * @sdp: The super block
1379 * @jd: The journal to map
1381 * Create a reusable "extent" mapping from all logical
1382 * blocks to all physical blocks for the given journal. This will save
1383 * us time when writing journal blocks. Most journals will have only one
1384 * extent that maps all their logical blocks. That's because gfs2.mkfs
1385 * arranges the journal blocks sequentially to maximize performance.
1386 * So the extent would map the first block for the entire file length.
1387 * However, gfs2_jadd can happen while file activity is happening, so
1388 * those journals may not be sequential. Less likely is the case where
1389 * the users created their own journals by mounting the metafs and
1390 * laying it out. But it's still possible. These journals might have
1393 * Returns: 0 on success, or error on failure
1396 int gfs2_map_journal_extents(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd)
1400 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
1401 struct buffer_head bh;
1402 unsigned int shift = sdp->sd_sb.sb_bsize_shift;
1406 lblock_stop = i_size_read(jd->jd_inode) >> shift;
1407 size = (lblock_stop - lblock) << shift;
1409 WARN_ON(!list_empty(&jd->extent_list));
1415 rc = gfs2_block_map(jd->jd_inode, lblock, &bh, 0);
1416 if (rc || !buffer_mapped(&bh))
1418 rc = gfs2_add_jextent(jd, lblock, bh.b_blocknr, bh.b_size >> shift);
1422 lblock += (bh.b_size >> ip->i_inode.i_blkbits);
1425 fs_info(sdp, "journal %d mapped with %u extents\n", jd->jd_jid,
1430 fs_warn(sdp, "error %d mapping journal %u at offset %llu (extent %u)\n",
1432 (unsigned long long)(i_size_read(jd->jd_inode) - size),
1434 fs_warn(sdp, "bmap=%d lblock=%llu block=%llu, state=0x%08lx, size=%llu\n",
1435 rc, (unsigned long long)lblock, (unsigned long long)bh.b_blocknr,
1436 bh.b_state, (unsigned long long)bh.b_size);
1437 gfs2_free_journal_extents(jd);
1442 * gfs2_write_alloc_required - figure out if a write will require an allocation
1443 * @ip: the file being written to
1444 * @offset: the offset to write to
1445 * @len: the number of bytes being written
1447 * Returns: 1 if an alloc is required, 0 otherwise
1450 int gfs2_write_alloc_required(struct gfs2_inode *ip, u64 offset,
1453 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1454 struct buffer_head bh;
1456 u64 lblock, lblock_stop, size;
1462 if (gfs2_is_stuffed(ip)) {
1464 sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode))
1469 shift = sdp->sd_sb.sb_bsize_shift;
1470 BUG_ON(gfs2_is_dir(ip));
1471 end_of_file = (i_size_read(&ip->i_inode) + sdp->sd_sb.sb_bsize - 1) >> shift;
1472 lblock = offset >> shift;
1473 lblock_stop = (offset + len + sdp->sd_sb.sb_bsize - 1) >> shift;
1474 if (lblock_stop > end_of_file)
1477 size = (lblock_stop - lblock) << shift;
1481 gfs2_block_map(&ip->i_inode, lblock, &bh, 0);
1482 if (!buffer_mapped(&bh))
1485 lblock += (bh.b_size >> ip->i_inode.i_blkbits);