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>
16 #include <linux/iomap.h>
31 #include "trace_gfs2.h"
33 /* This doesn't need to be that large as max 64 bit pointers in a 4k
34 * block is 512, so __u16 is fine for that. It saves stack space to
38 struct buffer_head *mp_bh[GFS2_MAX_META_HEIGHT];
39 __u16 mp_list[GFS2_MAX_META_HEIGHT];
40 int mp_fheight; /* find_metapath height */
41 int mp_aheight; /* actual height (lookup height) */
45 * gfs2_unstuffer_page - unstuff a stuffed inode into a block cached by a page
47 * @dibh: the dinode buffer
48 * @block: the block number that was allocated
49 * @page: The (optional) page. This is looked up if @page is NULL
54 static int gfs2_unstuffer_page(struct gfs2_inode *ip, struct buffer_head *dibh,
55 u64 block, struct page *page)
57 struct inode *inode = &ip->i_inode;
58 struct buffer_head *bh;
61 if (!page || page->index) {
62 page = find_or_create_page(inode->i_mapping, 0, GFP_NOFS);
68 if (!PageUptodate(page)) {
69 void *kaddr = kmap(page);
70 u64 dsize = i_size_read(inode);
72 if (dsize > (dibh->b_size - sizeof(struct gfs2_dinode)))
73 dsize = dibh->b_size - sizeof(struct gfs2_dinode);
75 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize);
76 memset(kaddr + dsize, 0, PAGE_SIZE - dsize);
79 SetPageUptodate(page);
82 if (!page_has_buffers(page))
83 create_empty_buffers(page, BIT(inode->i_blkbits),
86 bh = page_buffers(page);
88 if (!buffer_mapped(bh))
89 map_bh(bh, inode->i_sb, block);
91 set_buffer_uptodate(bh);
92 if (!gfs2_is_jdata(ip))
93 mark_buffer_dirty(bh);
94 if (!gfs2_is_writeback(ip))
95 gfs2_trans_add_data(ip->i_gl, bh);
106 * gfs2_unstuff_dinode - Unstuff a dinode when the data has grown too big
107 * @ip: The GFS2 inode to unstuff
108 * @page: The (optional) page. This is looked up if the @page is NULL
110 * This routine unstuffs a dinode and returns it to a "normal" state such
111 * that the height can be grown in the traditional way.
116 int gfs2_unstuff_dinode(struct gfs2_inode *ip, struct page *page)
118 struct buffer_head *bh, *dibh;
119 struct gfs2_dinode *di;
121 int isdir = gfs2_is_dir(ip);
124 down_write(&ip->i_rw_mutex);
126 error = gfs2_meta_inode_buffer(ip, &dibh);
130 if (i_size_read(&ip->i_inode)) {
131 /* Get a free block, fill it with the stuffed data,
132 and write it out to disk */
135 error = gfs2_alloc_blocks(ip, &block, &n, 0, NULL);
139 gfs2_trans_add_unrevoke(GFS2_SB(&ip->i_inode), block, 1);
140 error = gfs2_dir_get_new_buffer(ip, block, &bh);
143 gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_meta_header),
144 dibh, sizeof(struct gfs2_dinode));
147 error = gfs2_unstuffer_page(ip, dibh, block, page);
153 /* Set up the pointer to the new block */
155 gfs2_trans_add_meta(ip->i_gl, dibh);
156 di = (struct gfs2_dinode *)dibh->b_data;
157 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
159 if (i_size_read(&ip->i_inode)) {
160 *(__be64 *)(di + 1) = cpu_to_be64(block);
161 gfs2_add_inode_blocks(&ip->i_inode, 1);
162 di->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
166 di->di_height = cpu_to_be16(1);
171 up_write(&ip->i_rw_mutex);
177 * find_metapath - Find path through the metadata tree
178 * @sdp: The superblock
179 * @mp: The metapath to return the result in
180 * @block: The disk block to look up
181 * @height: The pre-calculated height of the metadata tree
183 * This routine returns a struct metapath structure that defines a path
184 * through the metadata of inode "ip" to get to block "block".
187 * Given: "ip" is a height 3 file, "offset" is 101342453, and this is a
188 * filesystem with a blocksize of 4096.
190 * find_metapath() would return a struct metapath structure set to:
191 * mp_offset = 101342453, mp_height = 3, mp_list[0] = 0, mp_list[1] = 48,
192 * and mp_list[2] = 165.
194 * That means that in order to get to the block containing the byte at
195 * offset 101342453, we would load the indirect block pointed to by pointer
196 * 0 in the dinode. We would then load the indirect block pointed to by
197 * pointer 48 in that indirect block. We would then load the data block
198 * pointed to by pointer 165 in that indirect block.
200 * ----------------------------------------
205 * ----------------------------------------
209 * ----------------------------------------
213 * |0 5 6 7 8 9 0 1 2|
214 * ----------------------------------------
218 * ----------------------------------------
223 * ----------------------------------------
227 * ----------------------------------------
228 * | Data block containing offset |
232 * ----------------------------------------
236 static void find_metapath(const struct gfs2_sbd *sdp, u64 block,
237 struct metapath *mp, unsigned int height)
241 mp->mp_fheight = height;
242 for (i = height; i--;)
243 mp->mp_list[i] = do_div(block, sdp->sd_inptrs);
246 static inline unsigned int metapath_branch_start(const struct metapath *mp)
248 if (mp->mp_list[0] == 0)
254 * metaptr1 - Return the first possible metadata pointer in a metapath buffer
255 * @height: The metadata height (0 = dinode)
258 static inline __be64 *metaptr1(unsigned int height, const struct metapath *mp)
260 struct buffer_head *bh = mp->mp_bh[height];
262 return ((__be64 *)(bh->b_data + sizeof(struct gfs2_dinode)));
263 return ((__be64 *)(bh->b_data + sizeof(struct gfs2_meta_header)));
267 * metapointer - Return pointer to start of metadata in a buffer
268 * @height: The metadata height (0 = dinode)
271 * Return a pointer to the block number of the next height of the metadata
272 * tree given a buffer containing the pointer to the current height of the
276 static inline __be64 *metapointer(unsigned int height, const struct metapath *mp)
278 __be64 *p = metaptr1(height, mp);
279 return p + mp->mp_list[height];
282 static void gfs2_metapath_ra(struct gfs2_glock *gl, struct metapath *mp,
285 struct buffer_head *bh = mp->mp_bh[height];
286 const __be64 *pos = metapointer(height, mp);
287 const __be64 *endp = (const __be64 *)(bh->b_data + bh->b_size);
290 for (t = pos; t < endp; t++) {
291 struct buffer_head *rabh;
296 rabh = gfs2_getbuf(gl, be64_to_cpu(*t), CREATE);
297 if (trylock_buffer(rabh)) {
298 if (!buffer_uptodate(rabh)) {
299 rabh->b_end_io = end_buffer_read_sync;
300 submit_bh(REQ_OP_READ,
301 REQ_RAHEAD | REQ_META | REQ_PRIO,
311 static int __fillup_metapath(struct gfs2_inode *ip, struct metapath *mp,
312 unsigned int x, unsigned int h)
315 __be64 *ptr = metapointer(x, mp);
316 u64 dblock = be64_to_cpu(*ptr);
321 ret = gfs2_meta_indirect_buffer(ip, x + 1, dblock, &mp->mp_bh[x + 1]);
325 mp->mp_aheight = x + 1;
330 * lookup_metapath - Walk the metadata tree to a specific point
334 * Assumes that the inode's buffer has already been looked up and
335 * hooked onto mp->mp_bh[0] and that the metapath has been initialised
336 * by find_metapath().
338 * If this function encounters part of the tree which has not been
339 * allocated, it returns the current height of the tree at the point
340 * at which it found the unallocated block. Blocks which are found are
341 * added to the mp->mp_bh[] list.
346 static int lookup_metapath(struct gfs2_inode *ip, struct metapath *mp)
348 return __fillup_metapath(ip, mp, 0, ip->i_height - 1);
352 * fillup_metapath - fill up buffers for the metadata path to a specific height
355 * @h: The height to which it should be mapped
357 * Similar to lookup_metapath, but does lookups for a range of heights
359 * Returns: error or the number of buffers filled
362 static int fillup_metapath(struct gfs2_inode *ip, struct metapath *mp, int h)
368 /* find the first buffer we need to look up. */
369 for (x = h - 1; x > 0; x--) {
374 ret = __fillup_metapath(ip, mp, x, h);
377 return mp->mp_aheight - x - 1;
380 static inline void release_metapath(struct metapath *mp)
384 for (i = 0; i < GFS2_MAX_META_HEIGHT; i++) {
385 if (mp->mp_bh[i] == NULL)
387 brelse(mp->mp_bh[i]);
392 * gfs2_extent_length - Returns length of an extent of blocks
393 * @start: Start of the buffer
394 * @len: Length of the buffer in bytes
395 * @ptr: Current position in the buffer
396 * @limit: Max extent length to return (0 = unlimited)
397 * @eob: Set to 1 if we hit "end of block"
399 * If the first block is zero (unallocated) it will return the number of
400 * unallocated blocks in the extent, otherwise it will return the number
401 * of contiguous blocks in the extent.
403 * Returns: The length of the extent (minimum of one block)
406 static inline unsigned int gfs2_extent_length(void *start, unsigned int len, __be64 *ptr, size_t limit, int *eob)
408 const __be64 *end = (start + len);
409 const __be64 *first = ptr;
410 u64 d = be64_to_cpu(*ptr);
417 if (limit && --limit == 0)
421 } while(be64_to_cpu(*ptr) == d);
424 return (ptr - first);
427 static inline void bmap_lock(struct gfs2_inode *ip, int create)
430 down_write(&ip->i_rw_mutex);
432 down_read(&ip->i_rw_mutex);
435 static inline void bmap_unlock(struct gfs2_inode *ip, int create)
438 up_write(&ip->i_rw_mutex);
440 up_read(&ip->i_rw_mutex);
443 static inline __be64 *gfs2_indirect_init(struct metapath *mp,
444 struct gfs2_glock *gl, unsigned int i,
445 unsigned offset, u64 bn)
447 __be64 *ptr = (__be64 *)(mp->mp_bh[i - 1]->b_data +
448 ((i > 1) ? sizeof(struct gfs2_meta_header) :
449 sizeof(struct gfs2_dinode)));
451 BUG_ON(mp->mp_bh[i] != NULL);
452 mp->mp_bh[i] = gfs2_meta_new(gl, bn);
453 gfs2_trans_add_meta(gl, mp->mp_bh[i]);
454 gfs2_metatype_set(mp->mp_bh[i], GFS2_METATYPE_IN, GFS2_FORMAT_IN);
455 gfs2_buffer_clear_tail(mp->mp_bh[i], sizeof(struct gfs2_meta_header));
457 *ptr = cpu_to_be64(bn);
463 ALLOC_GROW_DEPTH = 1,
464 ALLOC_GROW_HEIGHT = 2,
465 /* ALLOC_UNSTUFF = 3, TBD and rather complicated */
468 static inline unsigned int hptrs(struct gfs2_sbd *sdp, const unsigned int hgt)
471 return sdp->sd_inptrs;
472 return sdp->sd_diptrs;
476 * gfs2_bmap_alloc - Build a metadata tree of the requested height
477 * @inode: The GFS2 inode
478 * @lblock: The logical starting block of the extent
479 * @bh_map: This is used to return the mapping details
480 * @zero_new: True if newly allocated blocks should be zeroed
481 * @mp: The metapath, with proper height information calculated
482 * @maxlen: The max number of data blocks to alloc
483 * @dblock: Pointer to return the resulting new block
484 * @dblks: Pointer to return the number of blocks allocated
486 * In this routine we may have to alloc:
487 * i) Indirect blocks to grow the metadata tree height
488 * ii) Indirect blocks to fill in lower part of the metadata tree
491 * The function is in two parts. The first part works out the total
492 * number of blocks which we need. The second part does the actual
493 * allocation asking for an extent at a time (if enough contiguous free
494 * blocks are available, there will only be one request per bmap call)
495 * and uses the state machine to initialise the blocks in order.
497 * Returns: errno on error
500 static int gfs2_iomap_alloc(struct inode *inode, struct iomap *iomap,
501 unsigned flags, struct metapath *mp)
503 struct gfs2_inode *ip = GFS2_I(inode);
504 struct gfs2_sbd *sdp = GFS2_SB(inode);
505 struct super_block *sb = sdp->sd_vfs;
506 struct buffer_head *dibh = mp->mp_bh[0];
508 unsigned n, i, blks, alloced = 0, iblks = 0, branch_start = 0;
510 unsigned ptrs_per_blk;
511 const unsigned end_of_metadata = mp->mp_fheight - 1;
513 enum alloc_state state;
516 size_t maxlen = iomap->length >> inode->i_blkbits;
518 BUG_ON(mp->mp_aheight < 1);
519 BUG_ON(dibh == NULL);
521 gfs2_trans_add_meta(ip->i_gl, dibh);
523 if (mp->mp_fheight == mp->mp_aheight) {
524 struct buffer_head *bh;
527 /* Bottom indirect block exists, find unalloced extent size */
528 ptr = metapointer(end_of_metadata, mp);
529 bh = mp->mp_bh[end_of_metadata];
530 dblks = gfs2_extent_length(bh->b_data, bh->b_size, ptr,
535 /* Need to allocate indirect blocks */
536 ptrs_per_blk = mp->mp_fheight > 1 ? sdp->sd_inptrs :
538 dblks = min(maxlen, (size_t)(ptrs_per_blk -
539 mp->mp_list[end_of_metadata]));
540 if (mp->mp_fheight == ip->i_height) {
541 /* Writing into existing tree, extend tree down */
542 iblks = mp->mp_fheight - mp->mp_aheight;
543 state = ALLOC_GROW_DEPTH;
545 /* Building up tree height */
546 state = ALLOC_GROW_HEIGHT;
547 iblks = mp->mp_fheight - ip->i_height;
548 branch_start = metapath_branch_start(mp);
549 iblks += (mp->mp_fheight - branch_start);
553 /* start of the second part of the function (state machine) */
555 blks = dblks + iblks;
560 error = gfs2_alloc_blocks(ip, &bn, &n, 0, NULL);
564 if (state != ALLOC_DATA || gfs2_is_jdata(ip))
565 gfs2_trans_add_unrevoke(sdp, bn, n);
567 /* Growing height of tree */
568 case ALLOC_GROW_HEIGHT:
570 ptr = (__be64 *)(dibh->b_data +
571 sizeof(struct gfs2_dinode));
574 for (; i - 1 < mp->mp_fheight - ip->i_height && n > 0;
576 gfs2_indirect_init(mp, ip->i_gl, i, 0, bn++);
577 if (i - 1 == mp->mp_fheight - ip->i_height) {
579 gfs2_buffer_copy_tail(mp->mp_bh[i],
580 sizeof(struct gfs2_meta_header),
581 dibh, sizeof(struct gfs2_dinode));
582 gfs2_buffer_clear_tail(dibh,
583 sizeof(struct gfs2_dinode) +
585 ptr = (__be64 *)(mp->mp_bh[i]->b_data +
586 sizeof(struct gfs2_meta_header));
588 state = ALLOC_GROW_DEPTH;
589 for(i = branch_start; i < mp->mp_fheight; i++) {
590 if (mp->mp_bh[i] == NULL)
592 brelse(mp->mp_bh[i]);
599 /* Branching from existing tree */
600 case ALLOC_GROW_DEPTH:
601 if (i > 1 && i < mp->mp_fheight)
602 gfs2_trans_add_meta(ip->i_gl, mp->mp_bh[i-1]);
603 for (; i < mp->mp_fheight && n > 0; i++, n--)
604 gfs2_indirect_init(mp, ip->i_gl, i,
605 mp->mp_list[i-1], bn++);
606 if (i == mp->mp_fheight)
610 /* Tree complete, adding data blocks */
613 BUG_ON(mp->mp_bh[end_of_metadata] == NULL);
614 gfs2_trans_add_meta(ip->i_gl, mp->mp_bh[end_of_metadata]);
616 ptr = metapointer(end_of_metadata, mp);
617 iomap->addr = bn << inode->i_blkbits;
618 iomap->flags |= IOMAP_F_NEW;
620 *ptr++ = cpu_to_be64(bn++);
621 if (flags & IOMAP_ZERO) {
622 ret = sb_issue_zeroout(sb, iomap->addr >> inode->i_blkbits,
626 "Failed to zero data buffers\n");
627 flags &= ~IOMAP_ZERO;
632 } while (iomap->addr == IOMAP_NULL_ADDR);
634 iomap->length = (u64)dblks << inode->i_blkbits;
635 ip->i_height = mp->mp_fheight;
636 gfs2_add_inode_blocks(&ip->i_inode, alloced);
637 gfs2_dinode_out(ip, mp->mp_bh[0]->b_data);
642 * hole_size - figure out the size of a hole
644 * @lblock: The logical starting block number
647 * Returns: The hole size in bytes
650 static u64 hole_size(struct inode *inode, sector_t lblock, struct metapath *mp)
652 struct gfs2_inode *ip = GFS2_I(inode);
653 struct gfs2_sbd *sdp = GFS2_SB(inode);
654 struct metapath mp_eof;
658 const __be64 *first, *end, *ptr;
659 const struct buffer_head *bh;
660 u64 lblock_stop = (i_size_read(inode) - 1) >> inode->i_blkbits;
664 /* Get another metapath, to the very last byte */
665 find_metapath(sdp, lblock_stop, &mp_eof, ip->i_height);
666 for (hgt = ip->i_height - 1; hgt >= 0 && !done; hgt--) {
670 first = metapointer(hgt, mp);
671 end = (const __be64 *)(bh->b_data + bh->b_size);
673 for (ptr = first; ptr < end; ptr++) {
682 zeroptrs = sdp->sd_inptrs;
684 if (factor * zeroptrs >= lblock_stop - lblock + 1) {
685 holesz = lblock_stop - lblock + 1;
688 holesz += factor * zeroptrs;
690 factor *= sdp->sd_inptrs;
691 if (hgt && (mp->mp_list[hgt - 1] < mp_eof.mp_list[hgt - 1]))
692 (mp->mp_list[hgt - 1])++;
694 return holesz << inode->i_blkbits;
697 static void gfs2_stuffed_iomap(struct inode *inode, struct iomap *iomap)
699 struct gfs2_inode *ip = GFS2_I(inode);
701 iomap->addr = (ip->i_no_addr << inode->i_blkbits) +
702 sizeof(struct gfs2_dinode);
704 iomap->length = i_size_read(inode);
705 iomap->type = IOMAP_MAPPED;
706 iomap->flags = IOMAP_F_DATA_INLINE;
710 * gfs2_iomap_begin - Map blocks from an inode to disk blocks
712 * @pos: Starting position in bytes
713 * @length: Length to map, in bytes
714 * @flags: iomap flags
715 * @iomap: The iomap structure
719 int gfs2_iomap_begin(struct inode *inode, loff_t pos, loff_t length,
720 unsigned flags, struct iomap *iomap)
722 struct gfs2_inode *ip = GFS2_I(inode);
723 struct gfs2_sbd *sdp = GFS2_SB(inode);
724 struct metapath mp = { .mp_aheight = 1, };
725 unsigned int factor = sdp->sd_sb.sb_bsize;
726 const u64 *arr = sdp->sd_heightsize;
733 struct buffer_head *bh;
736 trace_gfs2_iomap_start(ip, pos, length, flags);
742 if ((flags & IOMAP_REPORT) && gfs2_is_stuffed(ip)) {
743 gfs2_stuffed_iomap(inode, iomap);
744 if (pos >= iomap->length)
750 lblock = pos >> inode->i_blkbits;
751 lend = (pos + length + sdp->sd_sb.sb_bsize - 1) >> inode->i_blkbits;
753 iomap->offset = lblock << inode->i_blkbits;
754 iomap->addr = IOMAP_NULL_ADDR;
755 iomap->type = IOMAP_HOLE;
756 iomap->length = (u64)(lend - lblock) << inode->i_blkbits;
757 iomap->flags = IOMAP_F_MERGED;
761 * Directory data blocks have a struct gfs2_meta_header header, so the
762 * remaining size is smaller than the filesystem block size. Logical
763 * block numbers for directories are in units of this remaining size!
765 if (gfs2_is_dir(ip)) {
766 factor = sdp->sd_jbsize;
767 arr = sdp->sd_jheightsize;
770 ret = gfs2_meta_inode_buffer(ip, &mp.mp_bh[0]);
774 height = ip->i_height;
775 while ((lblock + 1) * factor > arr[height])
777 find_metapath(sdp, lblock, &mp, height);
778 if (height > ip->i_height || gfs2_is_stuffed(ip))
781 ret = lookup_metapath(ip, &mp);
785 if (mp.mp_aheight != ip->i_height)
788 ptr = metapointer(ip->i_height - 1, &mp);
792 iomap->type = IOMAP_MAPPED;
793 iomap->addr = be64_to_cpu(*ptr) << inode->i_blkbits;
795 bh = mp.mp_bh[ip->i_height - 1];
796 len = gfs2_extent_length(bh->b_data, bh->b_size, ptr, lend - lblock, &eob);
798 iomap->flags |= IOMAP_F_BOUNDARY;
799 iomap->length = (u64)len << inode->i_blkbits;
804 release_metapath(&mp);
807 trace_gfs2_iomap_end(ip, iomap, ret);
811 if (!(flags & IOMAP_WRITE)) {
812 if (pos >= i_size_read(inode)) {
817 iomap->length = hole_size(inode, lblock, &mp);
821 ret = gfs2_iomap_alloc(inode, iomap, flags, &mp);
826 * gfs2_block_map - Map a block from an inode to a disk block
828 * @lblock: The logical block number
829 * @bh_map: The bh to be mapped
830 * @create: True if its ok to alloc blocks to satify the request
832 * Sets buffer_mapped() if successful, sets buffer_boundary() if a
833 * read of metadata will be required before the next block can be
834 * mapped. Sets buffer_new() if new blocks were allocated.
839 int gfs2_block_map(struct inode *inode, sector_t lblock,
840 struct buffer_head *bh_map, int create)
842 struct gfs2_inode *ip = GFS2_I(inode);
846 clear_buffer_mapped(bh_map);
847 clear_buffer_new(bh_map);
848 clear_buffer_boundary(bh_map);
849 trace_gfs2_bmap(ip, bh_map, lblock, create, 1);
852 flags |= IOMAP_WRITE;
853 if (buffer_zeronew(bh_map))
855 ret = gfs2_iomap_begin(inode, (loff_t)lblock << inode->i_blkbits,
856 bh_map->b_size, flags, &iomap);
858 if (!create && ret == -ENOENT) {
859 /* Return unmapped buffer beyond the end of file. */
865 if (iomap.length > bh_map->b_size) {
866 iomap.length = bh_map->b_size;
867 iomap.flags &= ~IOMAP_F_BOUNDARY;
869 if (iomap.addr != IOMAP_NULL_ADDR)
870 map_bh(bh_map, inode->i_sb, iomap.addr >> inode->i_blkbits);
871 bh_map->b_size = iomap.length;
872 if (iomap.flags & IOMAP_F_BOUNDARY)
873 set_buffer_boundary(bh_map);
874 if (iomap.flags & IOMAP_F_NEW)
875 set_buffer_new(bh_map);
878 trace_gfs2_bmap(ip, bh_map, lblock, create, ret);
883 * Deprecated: do not use in new code
885 int gfs2_extent_map(struct inode *inode, u64 lblock, int *new, u64 *dblock, unsigned *extlen)
887 struct buffer_head bh = { .b_state = 0, .b_blocknr = 0 };
895 bh.b_size = BIT(inode->i_blkbits + (create ? 0 : 5));
896 ret = gfs2_block_map(inode, lblock, &bh, create);
897 *extlen = bh.b_size >> inode->i_blkbits;
898 *dblock = bh.b_blocknr;
907 * gfs2_block_truncate_page - Deal with zeroing out data for truncate
909 * This is partly borrowed from ext3.
911 static int gfs2_block_truncate_page(struct address_space *mapping, loff_t from)
913 struct inode *inode = mapping->host;
914 struct gfs2_inode *ip = GFS2_I(inode);
915 unsigned long index = from >> PAGE_SHIFT;
916 unsigned offset = from & (PAGE_SIZE-1);
917 unsigned blocksize, iblock, length, pos;
918 struct buffer_head *bh;
922 page = find_or_create_page(mapping, index, GFP_NOFS);
926 blocksize = inode->i_sb->s_blocksize;
927 length = blocksize - (offset & (blocksize - 1));
928 iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
930 if (!page_has_buffers(page))
931 create_empty_buffers(page, blocksize, 0);
933 /* Find the buffer that contains "offset" */
934 bh = page_buffers(page);
936 while (offset >= pos) {
937 bh = bh->b_this_page;
944 if (!buffer_mapped(bh)) {
945 gfs2_block_map(inode, iblock, bh, 0);
946 /* unmapped? It's a hole - nothing to do */
947 if (!buffer_mapped(bh))
951 /* Ok, it's mapped. Make sure it's up-to-date */
952 if (PageUptodate(page))
953 set_buffer_uptodate(bh);
955 if (!buffer_uptodate(bh)) {
957 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
959 /* Uhhuh. Read error. Complain and punt. */
960 if (!buffer_uptodate(bh))
965 if (!gfs2_is_writeback(ip))
966 gfs2_trans_add_data(ip->i_gl, bh);
968 zero_user(page, offset, length);
969 mark_buffer_dirty(bh);
976 #define GFS2_JTRUNC_REVOKES 8192
979 * gfs2_journaled_truncate - Wrapper for truncate_pagecache for jdata files
980 * @inode: The inode being truncated
981 * @oldsize: The original (larger) size
982 * @newsize: The new smaller size
984 * With jdata files, we have to journal a revoke for each block which is
985 * truncated. As a result, we need to split this into separate transactions
986 * if the number of pages being truncated gets too large.
989 static int gfs2_journaled_truncate(struct inode *inode, u64 oldsize, u64 newsize)
991 struct gfs2_sbd *sdp = GFS2_SB(inode);
992 u64 max_chunk = GFS2_JTRUNC_REVOKES * sdp->sd_vfs->s_blocksize;
996 while (oldsize != newsize) {
997 struct gfs2_trans *tr;
1000 chunk = oldsize - newsize;
1001 if (chunk > max_chunk)
1004 offs = oldsize & ~PAGE_MASK;
1005 if (offs && chunk > PAGE_SIZE)
1006 chunk = offs + ((chunk - offs) & PAGE_MASK);
1008 truncate_pagecache(inode, oldsize - chunk);
1011 tr = current->journal_info;
1012 if (!test_bit(TR_TOUCHED, &tr->tr_flags))
1015 gfs2_trans_end(sdp);
1016 error = gfs2_trans_begin(sdp, RES_DINODE, GFS2_JTRUNC_REVOKES);
1024 static int trunc_start(struct inode *inode, u64 newsize)
1026 struct gfs2_inode *ip = GFS2_I(inode);
1027 struct gfs2_sbd *sdp = GFS2_SB(inode);
1028 struct address_space *mapping = inode->i_mapping;
1029 struct buffer_head *dibh = NULL;
1030 int journaled = gfs2_is_jdata(ip);
1031 u64 oldsize = inode->i_size;
1035 error = gfs2_trans_begin(sdp, RES_DINODE + RES_JDATA, GFS2_JTRUNC_REVOKES);
1037 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1041 error = gfs2_meta_inode_buffer(ip, &dibh);
1045 gfs2_trans_add_meta(ip->i_gl, dibh);
1047 if (gfs2_is_stuffed(ip)) {
1048 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode) + newsize);
1050 if (newsize & (u64)(sdp->sd_sb.sb_bsize - 1)) {
1051 error = gfs2_block_truncate_page(mapping, newsize);
1055 ip->i_diskflags |= GFS2_DIF_TRUNC_IN_PROG;
1058 i_size_write(inode, newsize);
1059 ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode);
1060 gfs2_dinode_out(ip, dibh->b_data);
1063 error = gfs2_journaled_truncate(inode, oldsize, newsize);
1065 truncate_pagecache(inode, newsize);
1069 if (current->journal_info)
1070 gfs2_trans_end(sdp);
1075 * sweep_bh_for_rgrps - find an rgrp in a meta buffer and free blocks therein
1077 * @rg_gh: holder of resource group glock
1078 * @mp: current metapath fully populated with buffers
1079 * @btotal: place to keep count of total blocks freed
1080 * @hgt: height we're processing
1081 * @first: true if this is the first call to this function for this height
1083 * We sweep a metadata buffer (provided by the metapath) for blocks we need to
1084 * free, and free them all. However, we do it one rgrp at a time. If this
1085 * block has references to multiple rgrps, we break it into individual
1086 * transactions. This allows other processes to use the rgrps while we're
1087 * focused on a single one, for better concurrency / performance.
1088 * At every transaction boundary, we rewrite the inode into the journal.
1089 * That way the bitmaps are kept consistent with the inode and we can recover
1090 * if we're interrupted by power-outages.
1092 * Returns: 0, or return code if an error occurred.
1093 * *btotal has the total number of blocks freed
1095 static int sweep_bh_for_rgrps(struct gfs2_inode *ip, struct gfs2_holder *rd_gh,
1096 const struct metapath *mp, u32 *btotal, int hgt,
1099 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1100 struct gfs2_rgrpd *rgd;
1101 struct gfs2_trans *tr;
1102 struct buffer_head *bh = mp->mp_bh[hgt];
1103 __be64 *top, *bottom, *p;
1104 int blks_outside_rgrp;
1105 u64 bn, bstart, isize_blks;
1106 s64 blen; /* needs to be s64 or gfs2_add_inode_blocks breaks */
1107 int meta = ((hgt != ip->i_height - 1) ? 1 : 0);
1109 bool buf_in_tr = false; /* buffer was added to transaction */
1111 if (gfs2_metatype_check(sdp, bh,
1112 (hgt ? GFS2_METATYPE_IN : GFS2_METATYPE_DI)))
1116 blks_outside_rgrp = 0;
1119 top = metapointer(hgt, mp); /* first ptr from metapath */
1120 /* If we're keeping some data at the truncation point, we've got to
1121 preserve the metadata tree by adding 1 to the starting metapath. */
1125 bottom = (__be64 *)(bh->b_data + bh->b_size);
1127 for (p = top; p < bottom; p++) {
1130 bn = be64_to_cpu(*p);
1131 if (gfs2_holder_initialized(rd_gh)) {
1132 rgd = gfs2_glock2rgrp(rd_gh->gh_gl);
1133 gfs2_assert_withdraw(sdp,
1134 gfs2_glock_is_locked_by_me(rd_gh->gh_gl));
1136 rgd = gfs2_blk2rgrpd(sdp, bn, true);
1137 ret = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE,
1142 /* Must be done with the rgrp glock held: */
1143 if (gfs2_rs_active(&ip->i_res) &&
1144 rgd == ip->i_res.rs_rbm.rgd)
1145 gfs2_rs_deltree(&ip->i_res);
1148 if (!rgrp_contains_block(rgd, bn)) {
1149 blks_outside_rgrp++;
1153 /* The size of our transactions will be unknown until we
1154 actually process all the metadata blocks that relate to
1155 the rgrp. So we estimate. We know it can't be more than
1156 the dinode's i_blocks and we don't want to exceed the
1157 journal flush threshold, sd_log_thresh2. */
1158 if (current->journal_info == NULL) {
1159 unsigned int jblocks_rqsted, revokes;
1161 jblocks_rqsted = rgd->rd_length + RES_DINODE +
1163 isize_blks = gfs2_get_inode_blocks(&ip->i_inode);
1164 if (isize_blks > atomic_read(&sdp->sd_log_thresh2))
1166 atomic_read(&sdp->sd_log_thresh2);
1168 jblocks_rqsted += isize_blks;
1169 revokes = jblocks_rqsted;
1171 revokes += hptrs(sdp, hgt);
1172 else if (ip->i_depth)
1173 revokes += sdp->sd_inptrs;
1174 ret = gfs2_trans_begin(sdp, jblocks_rqsted, revokes);
1177 down_write(&ip->i_rw_mutex);
1179 /* check if we will exceed the transaction blocks requested */
1180 tr = current->journal_info;
1181 if (tr->tr_num_buf_new + RES_STATFS +
1182 RES_QUOTA >= atomic_read(&sdp->sd_log_thresh2)) {
1183 /* We set blks_outside_rgrp to ensure the loop will
1184 be repeated for the same rgrp, but with a new
1186 blks_outside_rgrp++;
1187 /* This next part is tricky. If the buffer was added
1188 to the transaction, we've already set some block
1189 pointers to 0, so we better follow through and free
1190 them, or we will introduce corruption (so break).
1191 This may be impossible, or at least rare, but I
1192 decided to cover the case regardless.
1194 If the buffer was not added to the transaction
1195 (this call), doing so would exceed our transaction
1196 size, so we need to end the transaction and start a
1197 new one (so goto). */
1204 gfs2_trans_add_meta(ip->i_gl, bh);
1207 if (bstart + blen == bn) {
1212 __gfs2_free_blocks(ip, bstart, (u32)blen, meta);
1214 gfs2_add_inode_blocks(&ip->i_inode, -blen);
1220 __gfs2_free_blocks(ip, bstart, (u32)blen, meta);
1222 gfs2_add_inode_blocks(&ip->i_inode, -blen);
1225 if (!ret && blks_outside_rgrp) { /* If buffer still has non-zero blocks
1226 outside the rgrp we just processed,
1227 do it all over again. */
1228 if (current->journal_info) {
1229 struct buffer_head *dibh = mp->mp_bh[0];
1231 /* Every transaction boundary, we rewrite the dinode
1232 to keep its di_blocks current in case of failure. */
1233 ip->i_inode.i_mtime = ip->i_inode.i_ctime =
1234 current_time(&ip->i_inode);
1235 gfs2_trans_add_meta(ip->i_gl, dibh);
1236 gfs2_dinode_out(ip, dibh->b_data);
1237 up_write(&ip->i_rw_mutex);
1238 gfs2_trans_end(sdp);
1240 gfs2_glock_dq_uninit(rd_gh);
1249 * find_nonnull_ptr - find a non-null pointer given a metapath and height
1250 * assumes the metapath is valid (with buffers) out to height h
1251 * @mp: starting metapath
1252 * @h: desired height to search
1254 * Returns: true if a non-null pointer was found in the metapath buffer
1255 * false if all remaining pointers are NULL in the buffer
1257 static bool find_nonnull_ptr(struct gfs2_sbd *sdp, struct metapath *mp,
1261 unsigned int ptrs = hptrs(sdp, h) - 1;
1264 ptr = metapointer(h, mp);
1265 if (*ptr) { /* if we have a non-null pointer */
1266 /* Now zero the metapath after the current height. */
1268 if (h < GFS2_MAX_META_HEIGHT)
1269 memset(&mp->mp_list[h], 0,
1270 (GFS2_MAX_META_HEIGHT - h) *
1271 sizeof(mp->mp_list[0]));
1275 if (mp->mp_list[h] < ptrs)
1278 return false; /* no more pointers in this buffer */
1282 enum dealloc_states {
1283 DEALLOC_MP_FULL = 0, /* Strip a metapath with all buffers read in */
1284 DEALLOC_MP_LOWER = 1, /* lower the metapath strip height */
1285 DEALLOC_FILL_MP = 2, /* Fill in the metapath to the given height. */
1286 DEALLOC_DONE = 3, /* process complete */
1289 static bool mp_eq_to_hgt(struct metapath *mp, __u16 *nbof, unsigned int h)
1291 if (memcmp(mp->mp_list, nbof, h * sizeof(mp->mp_list[0])))
1297 * trunc_dealloc - truncate a file down to a desired size
1298 * @ip: inode to truncate
1299 * @newsize: The desired size of the file
1301 * This function truncates a file to newsize. It works from the
1302 * bottom up, and from the right to the left. In other words, it strips off
1303 * the highest layer (data) before stripping any of the metadata. Doing it
1304 * this way is best in case the operation is interrupted by power failure, etc.
1305 * The dinode is rewritten in every transaction to guarantee integrity.
1307 static int trunc_dealloc(struct gfs2_inode *ip, u64 newsize)
1309 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1311 struct buffer_head *dibh, *bh;
1312 struct gfs2_holder rd_gh;
1314 __u16 nbof[GFS2_MAX_META_HEIGHT]; /* new beginning of truncation */
1315 unsigned int strip_h = ip->i_height - 1;
1318 int mp_h; /* metapath buffers are read in to this height */
1320 bool preserve1; /* need to preserve the first meta pointer? */
1325 lblock = (newsize - 1) >> sdp->sd_sb.sb_bsize_shift;
1327 memset(&mp, 0, sizeof(mp));
1328 find_metapath(sdp, lblock, &mp, ip->i_height);
1330 memcpy(&nbof, &mp.mp_list, sizeof(nbof));
1332 ret = gfs2_meta_inode_buffer(ip, &dibh);
1337 ret = lookup_metapath(ip, &mp);
1341 /* issue read-ahead on metadata */
1342 for (mp_h = 0; mp_h < mp.mp_aheight - 1; mp_h++)
1343 gfs2_metapath_ra(ip->i_gl, &mp, mp_h);
1345 if (mp.mp_aheight == ip->i_height)
1346 state = DEALLOC_MP_FULL; /* We have a complete metapath */
1348 state = DEALLOC_FILL_MP; /* deal with partial metapath */
1350 ret = gfs2_rindex_update(sdp);
1354 ret = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1357 gfs2_holder_mark_uninitialized(&rd_gh);
1361 while (state != DEALLOC_DONE) {
1363 /* Truncate a full metapath at the given strip height.
1364 * Note that strip_h == mp_h in order to be in this state. */
1365 case DEALLOC_MP_FULL:
1366 /* If we're truncating to a non-zero size and the mp is
1367 at the beginning of file for the strip height, we
1368 need to preserve the first metadata pointer. */
1369 preserve1 = (newsize && mp_eq_to_hgt(&mp, nbof, mp_h));
1370 bh = mp.mp_bh[mp_h];
1371 gfs2_assert_withdraw(sdp, bh);
1372 if (gfs2_assert_withdraw(sdp,
1373 prev_bnr != bh->b_blocknr)) {
1374 printk(KERN_EMERG "GFS2: fsid=%s:inode %llu, "
1375 "block:%llu, i_h:%u, s_h:%u, mp_h:%u\n",
1377 (unsigned long long)ip->i_no_addr,
1378 prev_bnr, ip->i_height, strip_h, mp_h);
1380 prev_bnr = bh->b_blocknr;
1381 ret = sweep_bh_for_rgrps(ip, &rd_gh, &mp, &btotal,
1383 /* If we hit an error or just swept dinode buffer,
1386 state = DEALLOC_DONE;
1389 state = DEALLOC_MP_LOWER;
1392 /* lower the metapath strip height */
1393 case DEALLOC_MP_LOWER:
1394 /* We're done with the current buffer, so release it,
1395 unless it's the dinode buffer. Then back up to the
1396 previous pointer. */
1398 brelse(mp.mp_bh[mp_h]);
1399 mp.mp_bh[mp_h] = NULL;
1401 /* If we can't get any lower in height, we've stripped
1402 off all we can. Next step is to back up and start
1403 stripping the previous level of metadata. */
1406 memcpy(&mp.mp_list, &nbof, sizeof(nbof));
1408 state = DEALLOC_FILL_MP;
1411 mp.mp_list[mp_h] = 0;
1412 mp_h--; /* search one metadata height down */
1413 if (mp.mp_list[mp_h] >= hptrs(sdp, mp_h) - 1)
1414 break; /* loop around in the same state */
1416 /* Here we've found a part of the metapath that is not
1417 * allocated. We need to search at that height for the
1418 * next non-null pointer. */
1419 if (find_nonnull_ptr(sdp, &mp, mp_h)) {
1420 state = DEALLOC_FILL_MP;
1423 /* No more non-null pointers at this height. Back up
1424 to the previous height and try again. */
1425 break; /* loop around in the same state */
1427 /* Fill the metapath with buffers to the given height. */
1428 case DEALLOC_FILL_MP:
1429 /* Fill the buffers out to the current height. */
1430 ret = fillup_metapath(ip, &mp, mp_h);
1434 /* issue read-ahead on metadata */
1435 if (mp.mp_aheight > 1) {
1436 for (; ret > 1; ret--)
1437 gfs2_metapath_ra(ip->i_gl, &mp,
1438 mp.mp_aheight - ret);
1441 /* If buffers found for the entire strip height */
1442 if (mp.mp_aheight - 1 == strip_h) {
1443 state = DEALLOC_MP_FULL;
1446 if (mp.mp_aheight < ip->i_height) /* We have a partial height */
1447 mp_h = mp.mp_aheight - 1;
1449 /* If we find a non-null block pointer, crawl a bit
1450 higher up in the metapath and try again, otherwise
1451 we need to look lower for a new starting point. */
1452 if (find_nonnull_ptr(sdp, &mp, mp_h))
1455 state = DEALLOC_MP_LOWER;
1461 if (current->journal_info == NULL) {
1462 ret = gfs2_trans_begin(sdp, RES_DINODE + RES_STATFS +
1466 down_write(&ip->i_rw_mutex);
1468 gfs2_statfs_change(sdp, 0, +btotal, 0);
1469 gfs2_quota_change(ip, -(s64)btotal, ip->i_inode.i_uid,
1471 ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode);
1472 gfs2_trans_add_meta(ip->i_gl, dibh);
1473 gfs2_dinode_out(ip, dibh->b_data);
1474 up_write(&ip->i_rw_mutex);
1475 gfs2_trans_end(sdp);
1479 if (gfs2_holder_initialized(&rd_gh))
1480 gfs2_glock_dq_uninit(&rd_gh);
1481 if (current->journal_info) {
1482 up_write(&ip->i_rw_mutex);
1483 gfs2_trans_end(sdp);
1486 gfs2_quota_unhold(ip);
1488 release_metapath(&mp);
1492 static int trunc_end(struct gfs2_inode *ip)
1494 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1495 struct buffer_head *dibh;
1498 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1502 down_write(&ip->i_rw_mutex);
1504 error = gfs2_meta_inode_buffer(ip, &dibh);
1508 if (!i_size_read(&ip->i_inode)) {
1510 ip->i_goal = ip->i_no_addr;
1511 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
1512 gfs2_ordered_del_inode(ip);
1514 ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode);
1515 ip->i_diskflags &= ~GFS2_DIF_TRUNC_IN_PROG;
1517 gfs2_trans_add_meta(ip->i_gl, dibh);
1518 gfs2_dinode_out(ip, dibh->b_data);
1522 up_write(&ip->i_rw_mutex);
1523 gfs2_trans_end(sdp);
1528 * do_shrink - make a file smaller
1530 * @newsize: the size to make the file
1532 * Called with an exclusive lock on @inode. The @size must
1533 * be equal to or smaller than the current inode size.
1538 static int do_shrink(struct inode *inode, u64 newsize)
1540 struct gfs2_inode *ip = GFS2_I(inode);
1543 error = trunc_start(inode, newsize);
1546 if (gfs2_is_stuffed(ip))
1549 error = trunc_dealloc(ip, newsize);
1551 error = trunc_end(ip);
1556 void gfs2_trim_blocks(struct inode *inode)
1560 ret = do_shrink(inode, inode->i_size);
1565 * do_grow - Touch and update inode size
1567 * @size: The new size
1569 * This function updates the timestamps on the inode and
1570 * may also increase the size of the inode. This function
1571 * must not be called with @size any smaller than the current
1574 * Although it is not strictly required to unstuff files here,
1575 * earlier versions of GFS2 have a bug in the stuffed file reading
1576 * code which will result in a buffer overrun if the size is larger
1577 * than the max stuffed file size. In order to prevent this from
1578 * occurring, such files are unstuffed, but in other cases we can
1579 * just update the inode size directly.
1581 * Returns: 0 on success, or -ve on error
1584 static int do_grow(struct inode *inode, u64 size)
1586 struct gfs2_inode *ip = GFS2_I(inode);
1587 struct gfs2_sbd *sdp = GFS2_SB(inode);
1588 struct gfs2_alloc_parms ap = { .target = 1, };
1589 struct buffer_head *dibh;
1593 if (gfs2_is_stuffed(ip) &&
1594 (size > (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)))) {
1595 error = gfs2_quota_lock_check(ip, &ap);
1599 error = gfs2_inplace_reserve(ip, &ap);
1601 goto do_grow_qunlock;
1605 error = gfs2_trans_begin(sdp, RES_DINODE + RES_STATFS + RES_RG_BIT +
1606 (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF ?
1609 goto do_grow_release;
1612 error = gfs2_unstuff_dinode(ip, NULL);
1617 error = gfs2_meta_inode_buffer(ip, &dibh);
1621 i_size_write(inode, size);
1622 ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode);
1623 gfs2_trans_add_meta(ip->i_gl, dibh);
1624 gfs2_dinode_out(ip, dibh->b_data);
1628 gfs2_trans_end(sdp);
1631 gfs2_inplace_release(ip);
1633 gfs2_quota_unlock(ip);
1639 * gfs2_setattr_size - make a file a given size
1641 * @newsize: the size to make the file
1643 * The file size can grow, shrink, or stay the same size. This
1644 * is called holding i_mutex and an exclusive glock on the inode
1650 int gfs2_setattr_size(struct inode *inode, u64 newsize)
1652 struct gfs2_inode *ip = GFS2_I(inode);
1655 BUG_ON(!S_ISREG(inode->i_mode));
1657 ret = inode_newsize_ok(inode, newsize);
1661 inode_dio_wait(inode);
1663 ret = gfs2_rsqa_alloc(ip);
1667 if (newsize >= inode->i_size) {
1668 ret = do_grow(inode, newsize);
1672 ret = do_shrink(inode, newsize);
1674 gfs2_rsqa_delete(ip, NULL);
1678 int gfs2_truncatei_resume(struct gfs2_inode *ip)
1681 error = trunc_dealloc(ip, i_size_read(&ip->i_inode));
1683 error = trunc_end(ip);
1687 int gfs2_file_dealloc(struct gfs2_inode *ip)
1689 return trunc_dealloc(ip, 0);
1693 * gfs2_free_journal_extents - Free cached journal bmap info
1698 void gfs2_free_journal_extents(struct gfs2_jdesc *jd)
1700 struct gfs2_journal_extent *jext;
1702 while(!list_empty(&jd->extent_list)) {
1703 jext = list_entry(jd->extent_list.next, struct gfs2_journal_extent, list);
1704 list_del(&jext->list);
1710 * gfs2_add_jextent - Add or merge a new extent to extent cache
1711 * @jd: The journal descriptor
1712 * @lblock: The logical block at start of new extent
1713 * @dblock: The physical block at start of new extent
1714 * @blocks: Size of extent in fs blocks
1716 * Returns: 0 on success or -ENOMEM
1719 static int gfs2_add_jextent(struct gfs2_jdesc *jd, u64 lblock, u64 dblock, u64 blocks)
1721 struct gfs2_journal_extent *jext;
1723 if (!list_empty(&jd->extent_list)) {
1724 jext = list_entry(jd->extent_list.prev, struct gfs2_journal_extent, list);
1725 if ((jext->dblock + jext->blocks) == dblock) {
1726 jext->blocks += blocks;
1731 jext = kzalloc(sizeof(struct gfs2_journal_extent), GFP_NOFS);
1734 jext->dblock = dblock;
1735 jext->lblock = lblock;
1736 jext->blocks = blocks;
1737 list_add_tail(&jext->list, &jd->extent_list);
1743 * gfs2_map_journal_extents - Cache journal bmap info
1744 * @sdp: The super block
1745 * @jd: The journal to map
1747 * Create a reusable "extent" mapping from all logical
1748 * blocks to all physical blocks for the given journal. This will save
1749 * us time when writing journal blocks. Most journals will have only one
1750 * extent that maps all their logical blocks. That's because gfs2.mkfs
1751 * arranges the journal blocks sequentially to maximize performance.
1752 * So the extent would map the first block for the entire file length.
1753 * However, gfs2_jadd can happen while file activity is happening, so
1754 * those journals may not be sequential. Less likely is the case where
1755 * the users created their own journals by mounting the metafs and
1756 * laying it out. But it's still possible. These journals might have
1759 * Returns: 0 on success, or error on failure
1762 int gfs2_map_journal_extents(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd)
1766 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
1767 struct buffer_head bh;
1768 unsigned int shift = sdp->sd_sb.sb_bsize_shift;
1772 lblock_stop = i_size_read(jd->jd_inode) >> shift;
1773 size = (lblock_stop - lblock) << shift;
1775 WARN_ON(!list_empty(&jd->extent_list));
1781 rc = gfs2_block_map(jd->jd_inode, lblock, &bh, 0);
1782 if (rc || !buffer_mapped(&bh))
1784 rc = gfs2_add_jextent(jd, lblock, bh.b_blocknr, bh.b_size >> shift);
1788 lblock += (bh.b_size >> ip->i_inode.i_blkbits);
1791 fs_info(sdp, "journal %d mapped with %u extents\n", jd->jd_jid,
1796 fs_warn(sdp, "error %d mapping journal %u at offset %llu (extent %u)\n",
1798 (unsigned long long)(i_size_read(jd->jd_inode) - size),
1800 fs_warn(sdp, "bmap=%d lblock=%llu block=%llu, state=0x%08lx, size=%llu\n",
1801 rc, (unsigned long long)lblock, (unsigned long long)bh.b_blocknr,
1802 bh.b_state, (unsigned long long)bh.b_size);
1803 gfs2_free_journal_extents(jd);
1808 * gfs2_write_alloc_required - figure out if a write will require an allocation
1809 * @ip: the file being written to
1810 * @offset: the offset to write to
1811 * @len: the number of bytes being written
1813 * Returns: 1 if an alloc is required, 0 otherwise
1816 int gfs2_write_alloc_required(struct gfs2_inode *ip, u64 offset,
1819 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1820 struct buffer_head bh;
1822 u64 lblock, lblock_stop, size;
1828 if (gfs2_is_stuffed(ip)) {
1830 sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode))
1835 shift = sdp->sd_sb.sb_bsize_shift;
1836 BUG_ON(gfs2_is_dir(ip));
1837 end_of_file = (i_size_read(&ip->i_inode) + sdp->sd_sb.sb_bsize - 1) >> shift;
1838 lblock = offset >> shift;
1839 lblock_stop = (offset + len + sdp->sd_sb.sb_bsize - 1) >> shift;
1840 if (lblock_stop > end_of_file)
1843 size = (lblock_stop - lblock) << shift;
1847 gfs2_block_map(&ip->i_inode, lblock, &bh, 0);
1848 if (!buffer_mapped(&bh))
1851 lblock += (bh.b_size >> ip->i_inode.i_blkbits);