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>
32 #include "trace_gfs2.h"
34 /* This doesn't need to be that large as max 64 bit pointers in a 4k
35 * block is 512, so __u16 is fine for that. It saves stack space to
39 struct buffer_head *mp_bh[GFS2_MAX_META_HEIGHT];
40 __u16 mp_list[GFS2_MAX_META_HEIGHT];
41 int mp_fheight; /* find_metapath height */
42 int mp_aheight; /* actual height (lookup height) */
45 static int punch_hole(struct gfs2_inode *ip, u64 offset, u64 length);
48 * gfs2_unstuffer_page - unstuff a stuffed inode into a block cached by a page
50 * @dibh: the dinode buffer
51 * @block: the block number that was allocated
52 * @page: The (optional) page. This is looked up if @page is NULL
57 static int gfs2_unstuffer_page(struct gfs2_inode *ip, struct buffer_head *dibh,
58 u64 block, struct page *page)
60 struct inode *inode = &ip->i_inode;
61 struct buffer_head *bh;
64 if (!page || page->index) {
65 page = find_or_create_page(inode->i_mapping, 0, GFP_NOFS);
71 if (!PageUptodate(page)) {
72 void *kaddr = kmap(page);
73 u64 dsize = i_size_read(inode);
75 if (dsize > gfs2_max_stuffed_size(ip))
76 dsize = gfs2_max_stuffed_size(ip);
78 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize);
79 memset(kaddr + dsize, 0, PAGE_SIZE - dsize);
82 SetPageUptodate(page);
85 if (!page_has_buffers(page))
86 create_empty_buffers(page, BIT(inode->i_blkbits),
89 bh = page_buffers(page);
91 if (!buffer_mapped(bh))
92 map_bh(bh, inode->i_sb, block);
94 set_buffer_uptodate(bh);
95 if (gfs2_is_jdata(ip))
96 gfs2_trans_add_data(ip->i_gl, bh);
98 mark_buffer_dirty(bh);
99 gfs2_ordered_add_inode(ip);
111 * gfs2_unstuff_dinode - Unstuff a dinode when the data has grown too big
112 * @ip: The GFS2 inode to unstuff
113 * @page: The (optional) page. This is looked up if the @page is NULL
115 * This routine unstuffs a dinode and returns it to a "normal" state such
116 * that the height can be grown in the traditional way.
121 int gfs2_unstuff_dinode(struct gfs2_inode *ip, struct page *page)
123 struct buffer_head *bh, *dibh;
124 struct gfs2_dinode *di;
126 int isdir = gfs2_is_dir(ip);
129 down_write(&ip->i_rw_mutex);
131 error = gfs2_meta_inode_buffer(ip, &dibh);
135 if (i_size_read(&ip->i_inode)) {
136 /* Get a free block, fill it with the stuffed data,
137 and write it out to disk */
140 error = gfs2_alloc_blocks(ip, &block, &n, 0, NULL);
144 gfs2_trans_add_unrevoke(GFS2_SB(&ip->i_inode), block, 1);
145 error = gfs2_dir_get_new_buffer(ip, block, &bh);
148 gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_meta_header),
149 dibh, sizeof(struct gfs2_dinode));
152 error = gfs2_unstuffer_page(ip, dibh, block, page);
158 /* Set up the pointer to the new block */
160 gfs2_trans_add_meta(ip->i_gl, dibh);
161 di = (struct gfs2_dinode *)dibh->b_data;
162 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
164 if (i_size_read(&ip->i_inode)) {
165 *(__be64 *)(di + 1) = cpu_to_be64(block);
166 gfs2_add_inode_blocks(&ip->i_inode, 1);
167 di->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
171 di->di_height = cpu_to_be16(1);
176 up_write(&ip->i_rw_mutex);
182 * find_metapath - Find path through the metadata tree
183 * @sdp: The superblock
184 * @block: The disk block to look up
185 * @mp: The metapath to return the result in
186 * @height: The pre-calculated height of the metadata tree
188 * This routine returns a struct metapath structure that defines a path
189 * through the metadata of inode "ip" to get to block "block".
192 * Given: "ip" is a height 3 file, "offset" is 101342453, and this is a
193 * filesystem with a blocksize of 4096.
195 * find_metapath() would return a struct metapath structure set to:
196 * mp_fheight = 3, mp_list[0] = 0, mp_list[1] = 48, and mp_list[2] = 165.
198 * That means that in order to get to the block containing the byte at
199 * offset 101342453, we would load the indirect block pointed to by pointer
200 * 0 in the dinode. We would then load the indirect block pointed to by
201 * pointer 48 in that indirect block. We would then load the data block
202 * pointed to by pointer 165 in that indirect block.
204 * ----------------------------------------
209 * ----------------------------------------
213 * ----------------------------------------
217 * |0 5 6 7 8 9 0 1 2|
218 * ----------------------------------------
222 * ----------------------------------------
227 * ----------------------------------------
231 * ----------------------------------------
232 * | Data block containing offset |
236 * ----------------------------------------
240 static void find_metapath(const struct gfs2_sbd *sdp, u64 block,
241 struct metapath *mp, unsigned int height)
245 mp->mp_fheight = height;
246 for (i = height; i--;)
247 mp->mp_list[i] = do_div(block, sdp->sd_inptrs);
250 static inline unsigned int metapath_branch_start(const struct metapath *mp)
252 if (mp->mp_list[0] == 0)
258 * metaptr1 - Return the first possible metadata pointer in a metapath buffer
259 * @height: The metadata height (0 = dinode)
262 static inline __be64 *metaptr1(unsigned int height, const struct metapath *mp)
264 struct buffer_head *bh = mp->mp_bh[height];
266 return ((__be64 *)(bh->b_data + sizeof(struct gfs2_dinode)));
267 return ((__be64 *)(bh->b_data + sizeof(struct gfs2_meta_header)));
271 * metapointer - Return pointer to start of metadata in a buffer
272 * @height: The metadata height (0 = dinode)
275 * Return a pointer to the block number of the next height of the metadata
276 * tree given a buffer containing the pointer to the current height of the
280 static inline __be64 *metapointer(unsigned int height, const struct metapath *mp)
282 __be64 *p = metaptr1(height, mp);
283 return p + mp->mp_list[height];
286 static inline const __be64 *metaend(unsigned int height, const struct metapath *mp)
288 const struct buffer_head *bh = mp->mp_bh[height];
289 return (const __be64 *)(bh->b_data + bh->b_size);
292 static void clone_metapath(struct metapath *clone, struct metapath *mp)
297 for (hgt = 0; hgt < mp->mp_aheight; hgt++)
298 get_bh(clone->mp_bh[hgt]);
301 static void gfs2_metapath_ra(struct gfs2_glock *gl, __be64 *start, __be64 *end)
305 for (t = start; t < end; t++) {
306 struct buffer_head *rabh;
311 rabh = gfs2_getbuf(gl, be64_to_cpu(*t), CREATE);
312 if (trylock_buffer(rabh)) {
313 if (!buffer_uptodate(rabh)) {
314 rabh->b_end_io = end_buffer_read_sync;
315 submit_bh(REQ_OP_READ,
316 REQ_RAHEAD | REQ_META | REQ_PRIO,
326 static int __fillup_metapath(struct gfs2_inode *ip, struct metapath *mp,
327 unsigned int x, unsigned int h)
330 __be64 *ptr = metapointer(x, mp);
331 u64 dblock = be64_to_cpu(*ptr);
336 ret = gfs2_meta_indirect_buffer(ip, x + 1, dblock, &mp->mp_bh[x + 1]);
340 mp->mp_aheight = x + 1;
345 * lookup_metapath - Walk the metadata tree to a specific point
349 * Assumes that the inode's buffer has already been looked up and
350 * hooked onto mp->mp_bh[0] and that the metapath has been initialised
351 * by find_metapath().
353 * If this function encounters part of the tree which has not been
354 * allocated, it returns the current height of the tree at the point
355 * at which it found the unallocated block. Blocks which are found are
356 * added to the mp->mp_bh[] list.
361 static int lookup_metapath(struct gfs2_inode *ip, struct metapath *mp)
363 return __fillup_metapath(ip, mp, 0, ip->i_height - 1);
367 * fillup_metapath - fill up buffers for the metadata path to a specific height
370 * @h: The height to which it should be mapped
372 * Similar to lookup_metapath, but does lookups for a range of heights
374 * Returns: error or the number of buffers filled
377 static int fillup_metapath(struct gfs2_inode *ip, struct metapath *mp, int h)
383 /* find the first buffer we need to look up. */
384 for (x = h - 1; x > 0; x--) {
389 ret = __fillup_metapath(ip, mp, x, h);
392 return mp->mp_aheight - x - 1;
395 static void release_metapath(struct metapath *mp)
399 for (i = 0; i < GFS2_MAX_META_HEIGHT; i++) {
400 if (mp->mp_bh[i] == NULL)
402 brelse(mp->mp_bh[i]);
408 * gfs2_extent_length - Returns length of an extent of blocks
409 * @bh: The metadata block
410 * @ptr: Current position in @bh
411 * @limit: Max extent length to return
412 * @eob: Set to 1 if we hit "end of block"
414 * Returns: The length of the extent (minimum of one block)
417 static inline unsigned int gfs2_extent_length(struct buffer_head *bh, __be64 *ptr, size_t limit, int *eob)
419 const __be64 *end = (__be64 *)(bh->b_data + bh->b_size);
420 const __be64 *first = ptr;
421 u64 d = be64_to_cpu(*ptr);
429 } while(be64_to_cpu(*ptr) == d);
435 typedef const __be64 *(*gfs2_metadata_walker)(
437 const __be64 *start, const __be64 *end,
438 u64 factor, void *data);
440 #define WALK_STOP ((__be64 *)0)
441 #define WALK_NEXT ((__be64 *)1)
443 static int gfs2_walk_metadata(struct inode *inode, sector_t lblock,
444 u64 len, struct metapath *mp, gfs2_metadata_walker walker,
447 struct metapath clone;
448 struct gfs2_inode *ip = GFS2_I(inode);
449 struct gfs2_sbd *sdp = GFS2_SB(inode);
450 const __be64 *start, *end, *ptr;
455 for (hgt = ip->i_height - 1; hgt >= mp->mp_aheight; hgt--)
456 factor *= sdp->sd_inptrs;
461 /* Walk indirect block. */
462 start = metapointer(hgt, mp);
463 end = metaend(hgt, mp);
465 step = (end - start) * factor;
467 end = start + DIV_ROUND_UP_ULL(len, factor);
469 ptr = walker(mp, start, end, factor, data);
470 if (ptr == WALK_STOP)
475 if (ptr != WALK_NEXT) {
477 mp->mp_list[hgt] += ptr - start;
478 goto fill_up_metapath;
482 /* Decrease height of metapath. */
484 clone_metapath(&clone, mp);
487 brelse(mp->mp_bh[hgt]);
488 mp->mp_bh[hgt] = NULL;
492 factor *= sdp->sd_inptrs;
494 /* Advance in metadata tree. */
495 (mp->mp_list[hgt])++;
496 start = metapointer(hgt, mp);
497 end = metaend(hgt, mp);
499 mp->mp_list[hgt] = 0;
506 /* Increase height of metapath. */
508 clone_metapath(&clone, mp);
511 ret = fillup_metapath(ip, mp, ip->i_height - 1);
516 do_div(factor, sdp->sd_inptrs);
517 mp->mp_aheight = hgt + 1;
520 release_metapath(mp);
524 struct gfs2_hole_walker_args {
528 static const __be64 *gfs2_hole_walker(struct metapath *mp,
529 const __be64 *start, const __be64 *end,
530 u64 factor, void *data)
532 struct gfs2_hole_walker_args *args = data;
535 for (ptr = start; ptr < end; ptr++) {
537 args->blocks += (ptr - start) * factor;
538 if (mp->mp_aheight == mp->mp_fheight)
540 return ptr; /* increase height */
543 args->blocks += (end - start) * factor;
548 * gfs2_hole_size - figure out the size of a hole
550 * @lblock: The logical starting block number
551 * @len: How far to look (in blocks)
552 * @mp: The metapath at lblock
553 * @iomap: The iomap to store the hole size in
555 * This function modifies @mp.
557 * Returns: errno on error
559 static int gfs2_hole_size(struct inode *inode, sector_t lblock, u64 len,
560 struct metapath *mp, struct iomap *iomap)
562 struct gfs2_hole_walker_args args = { };
565 ret = gfs2_walk_metadata(inode, lblock, len, mp, gfs2_hole_walker, &args);
567 iomap->length = args.blocks << inode->i_blkbits;
571 static inline __be64 *gfs2_indirect_init(struct metapath *mp,
572 struct gfs2_glock *gl, unsigned int i,
573 unsigned offset, u64 bn)
575 __be64 *ptr = (__be64 *)(mp->mp_bh[i - 1]->b_data +
576 ((i > 1) ? sizeof(struct gfs2_meta_header) :
577 sizeof(struct gfs2_dinode)));
579 BUG_ON(mp->mp_bh[i] != NULL);
580 mp->mp_bh[i] = gfs2_meta_new(gl, bn);
581 gfs2_trans_add_meta(gl, mp->mp_bh[i]);
582 gfs2_metatype_set(mp->mp_bh[i], GFS2_METATYPE_IN, GFS2_FORMAT_IN);
583 gfs2_buffer_clear_tail(mp->mp_bh[i], sizeof(struct gfs2_meta_header));
585 *ptr = cpu_to_be64(bn);
591 ALLOC_GROW_DEPTH = 1,
592 ALLOC_GROW_HEIGHT = 2,
593 /* ALLOC_UNSTUFF = 3, TBD and rather complicated */
597 * gfs2_iomap_alloc - Build a metadata tree of the requested height
598 * @inode: The GFS2 inode
599 * @iomap: The iomap structure
600 * @flags: iomap flags
601 * @mp: The metapath, with proper height information calculated
603 * In this routine we may have to alloc:
604 * i) Indirect blocks to grow the metadata tree height
605 * ii) Indirect blocks to fill in lower part of the metadata tree
608 * This function is called after gfs2_iomap_get, which works out the
609 * total number of blocks which we need via gfs2_alloc_size.
611 * We then do the actual allocation asking for an extent at a time (if
612 * enough contiguous free blocks are available, there will only be one
613 * allocation request per call) and uses the state machine to initialise
614 * the blocks in order.
616 * Right now, this function will allocate at most one indirect block
617 * worth of data -- with a default block size of 4K, that's slightly
618 * less than 2M. If this limitation is ever removed to allow huge
619 * allocations, we would probably still want to limit the iomap size we
620 * return to avoid stalling other tasks during huge writes; the next
621 * iomap iteration would then find the blocks already allocated.
623 * Returns: errno on error
626 static int gfs2_iomap_alloc(struct inode *inode, struct iomap *iomap,
627 unsigned flags, struct metapath *mp)
629 struct gfs2_inode *ip = GFS2_I(inode);
630 struct gfs2_sbd *sdp = GFS2_SB(inode);
631 struct buffer_head *dibh = mp->mp_bh[0];
633 unsigned n, i, blks, alloced = 0, iblks = 0, branch_start = 0;
634 size_t dblks = iomap->length >> inode->i_blkbits;
635 const unsigned end_of_metadata = mp->mp_fheight - 1;
637 enum alloc_state state;
641 BUG_ON(mp->mp_aheight < 1);
642 BUG_ON(dibh == NULL);
645 gfs2_trans_add_meta(ip->i_gl, dibh);
647 down_write(&ip->i_rw_mutex);
649 if (mp->mp_fheight == mp->mp_aheight) {
650 /* Bottom indirect block exists */
653 /* Need to allocate indirect blocks */
654 if (mp->mp_fheight == ip->i_height) {
655 /* Writing into existing tree, extend tree down */
656 iblks = mp->mp_fheight - mp->mp_aheight;
657 state = ALLOC_GROW_DEPTH;
659 /* Building up tree height */
660 state = ALLOC_GROW_HEIGHT;
661 iblks = mp->mp_fheight - ip->i_height;
662 branch_start = metapath_branch_start(mp);
663 iblks += (mp->mp_fheight - branch_start);
667 /* start of the second part of the function (state machine) */
669 blks = dblks + iblks;
673 ret = gfs2_alloc_blocks(ip, &bn, &n, 0, NULL);
677 if (state != ALLOC_DATA || gfs2_is_jdata(ip))
678 gfs2_trans_add_unrevoke(sdp, bn, n);
680 /* Growing height of tree */
681 case ALLOC_GROW_HEIGHT:
683 ptr = (__be64 *)(dibh->b_data +
684 sizeof(struct gfs2_dinode));
687 for (; i - 1 < mp->mp_fheight - ip->i_height && n > 0;
689 gfs2_indirect_init(mp, ip->i_gl, i, 0, bn++);
690 if (i - 1 == mp->mp_fheight - ip->i_height) {
692 gfs2_buffer_copy_tail(mp->mp_bh[i],
693 sizeof(struct gfs2_meta_header),
694 dibh, sizeof(struct gfs2_dinode));
695 gfs2_buffer_clear_tail(dibh,
696 sizeof(struct gfs2_dinode) +
698 ptr = (__be64 *)(mp->mp_bh[i]->b_data +
699 sizeof(struct gfs2_meta_header));
701 state = ALLOC_GROW_DEPTH;
702 for(i = branch_start; i < mp->mp_fheight; i++) {
703 if (mp->mp_bh[i] == NULL)
705 brelse(mp->mp_bh[i]);
712 /* Branching from existing tree */
713 case ALLOC_GROW_DEPTH:
714 if (i > 1 && i < mp->mp_fheight)
715 gfs2_trans_add_meta(ip->i_gl, mp->mp_bh[i-1]);
716 for (; i < mp->mp_fheight && n > 0; i++, n--)
717 gfs2_indirect_init(mp, ip->i_gl, i,
718 mp->mp_list[i-1], bn++);
719 if (i == mp->mp_fheight)
723 /* Tree complete, adding data blocks */
726 BUG_ON(mp->mp_bh[end_of_metadata] == NULL);
727 gfs2_trans_add_meta(ip->i_gl, mp->mp_bh[end_of_metadata]);
729 ptr = metapointer(end_of_metadata, mp);
730 iomap->addr = bn << inode->i_blkbits;
731 iomap->flags |= IOMAP_F_MERGED | IOMAP_F_NEW;
733 *ptr++ = cpu_to_be64(bn++);
736 } while (iomap->addr == IOMAP_NULL_ADDR);
738 iomap->type = IOMAP_MAPPED;
739 iomap->length = (u64)dblks << inode->i_blkbits;
740 ip->i_height = mp->mp_fheight;
741 gfs2_add_inode_blocks(&ip->i_inode, alloced);
742 gfs2_dinode_out(ip, dibh->b_data);
744 up_write(&ip->i_rw_mutex);
748 #define IOMAP_F_GFS2_BOUNDARY IOMAP_F_PRIVATE
751 * gfs2_alloc_size - Compute the maximum allocation size
754 * @size: Requested size in blocks
756 * Compute the maximum size of the next allocation at @mp.
758 * Returns: size in blocks
760 static u64 gfs2_alloc_size(struct inode *inode, struct metapath *mp, u64 size)
762 struct gfs2_inode *ip = GFS2_I(inode);
763 struct gfs2_sbd *sdp = GFS2_SB(inode);
764 const __be64 *first, *ptr, *end;
767 * For writes to stuffed files, this function is called twice via
768 * gfs2_iomap_get, before and after unstuffing. The size we return the
769 * first time needs to be large enough to get the reservation and
770 * allocation sizes right. The size we return the second time must
771 * be exact or else gfs2_iomap_alloc won't do the right thing.
774 if (gfs2_is_stuffed(ip) || mp->mp_fheight != mp->mp_aheight) {
775 unsigned int maxsize = mp->mp_fheight > 1 ?
776 sdp->sd_inptrs : sdp->sd_diptrs;
777 maxsize -= mp->mp_list[mp->mp_fheight - 1];
783 first = metapointer(ip->i_height - 1, mp);
784 end = metaend(ip->i_height - 1, mp);
785 if (end - first > size)
787 for (ptr = first; ptr < end; ptr++) {
795 * gfs2_iomap_get - Map blocks from an inode to disk blocks
797 * @pos: Starting position in bytes
798 * @length: Length to map, in bytes
799 * @flags: iomap flags
800 * @iomap: The iomap structure
805 static int gfs2_iomap_get(struct inode *inode, loff_t pos, loff_t length,
806 unsigned flags, struct iomap *iomap,
809 struct gfs2_inode *ip = GFS2_I(inode);
810 struct gfs2_sbd *sdp = GFS2_SB(inode);
811 loff_t size = i_size_read(inode);
814 sector_t lblock_stop;
818 struct buffer_head *dibh = NULL, *bh;
824 down_read(&ip->i_rw_mutex);
826 ret = gfs2_meta_inode_buffer(ip, &dibh);
829 iomap->private = dibh;
831 if (gfs2_is_stuffed(ip)) {
832 if (flags & IOMAP_WRITE) {
833 loff_t max_size = gfs2_max_stuffed_size(ip);
835 if (pos + length > max_size)
837 iomap->length = max_size;
840 if (flags & IOMAP_REPORT) {
846 iomap->length = length;
850 iomap->length = size;
852 iomap->addr = (ip->i_no_addr << inode->i_blkbits) +
853 sizeof(struct gfs2_dinode);
854 iomap->type = IOMAP_INLINE;
855 iomap->inline_data = dibh->b_data + sizeof(struct gfs2_dinode);
860 lblock = pos >> inode->i_blkbits;
861 iomap->offset = lblock << inode->i_blkbits;
862 lblock_stop = (pos + length - 1) >> inode->i_blkbits;
863 len = lblock_stop - lblock + 1;
864 iomap->length = len << inode->i_blkbits;
869 height = ip->i_height;
870 while ((lblock + 1) * sdp->sd_sb.sb_bsize > sdp->sd_heightsize[height])
872 find_metapath(sdp, lblock, mp, height);
873 if (height > ip->i_height || gfs2_is_stuffed(ip))
876 ret = lookup_metapath(ip, mp);
880 if (mp->mp_aheight != ip->i_height)
883 ptr = metapointer(ip->i_height - 1, mp);
887 bh = mp->mp_bh[ip->i_height - 1];
888 len = gfs2_extent_length(bh, ptr, len, &eob);
890 iomap->addr = be64_to_cpu(*ptr) << inode->i_blkbits;
891 iomap->length = len << inode->i_blkbits;
892 iomap->type = IOMAP_MAPPED;
893 iomap->flags = IOMAP_F_MERGED;
895 iomap->flags |= IOMAP_F_GFS2_BOUNDARY;
898 iomap->bdev = inode->i_sb->s_bdev;
900 up_read(&ip->i_rw_mutex);
906 iomap->addr = IOMAP_NULL_ADDR;
907 iomap->type = IOMAP_HOLE;
908 if (flags & IOMAP_REPORT) {
911 else if (height == ip->i_height)
912 ret = gfs2_hole_size(inode, lblock, len, mp, iomap);
914 iomap->length = size - pos;
915 } else if (flags & IOMAP_WRITE) {
918 if (flags & IOMAP_DIRECT)
919 goto out; /* (see gfs2_file_direct_write) */
921 len = gfs2_alloc_size(inode, mp, len);
922 alloc_size = len << inode->i_blkbits;
923 if (alloc_size < iomap->length)
924 iomap->length = alloc_size;
926 if (pos < size && height == ip->i_height)
927 ret = gfs2_hole_size(inode, lblock, len, mp, iomap);
932 static int gfs2_write_lock(struct inode *inode)
934 struct gfs2_inode *ip = GFS2_I(inode);
935 struct gfs2_sbd *sdp = GFS2_SB(inode);
938 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh);
939 error = gfs2_glock_nq(&ip->i_gh);
942 if (&ip->i_inode == sdp->sd_rindex) {
943 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
945 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE,
946 GL_NOCACHE, &m_ip->i_gh);
953 gfs2_glock_dq(&ip->i_gh);
955 gfs2_holder_uninit(&ip->i_gh);
959 static void gfs2_write_unlock(struct inode *inode)
961 struct gfs2_inode *ip = GFS2_I(inode);
962 struct gfs2_sbd *sdp = GFS2_SB(inode);
964 if (&ip->i_inode == sdp->sd_rindex) {
965 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
967 gfs2_glock_dq_uninit(&m_ip->i_gh);
969 gfs2_glock_dq_uninit(&ip->i_gh);
972 static void gfs2_iomap_journaled_page_done(struct inode *inode, loff_t pos,
973 unsigned copied, struct page *page,
976 struct gfs2_inode *ip = GFS2_I(inode);
978 gfs2_page_add_databufs(ip, page, offset_in_page(pos), copied);
981 static int gfs2_iomap_begin_write(struct inode *inode, loff_t pos,
982 loff_t length, unsigned flags,
985 struct metapath mp = { .mp_aheight = 1, };
986 struct gfs2_inode *ip = GFS2_I(inode);
987 struct gfs2_sbd *sdp = GFS2_SB(inode);
988 unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
989 bool unstuff, alloc_required;
992 ret = gfs2_write_lock(inode);
996 unstuff = gfs2_is_stuffed(ip) &&
997 pos + length > gfs2_max_stuffed_size(ip);
999 ret = gfs2_iomap_get(inode, pos, length, flags, iomap, &mp);
1003 alloc_required = unstuff || iomap->type == IOMAP_HOLE;
1005 if (alloc_required || gfs2_is_jdata(ip))
1006 gfs2_write_calc_reserv(ip, iomap->length, &data_blocks,
1009 if (alloc_required) {
1010 struct gfs2_alloc_parms ap = {
1011 .target = data_blocks + ind_blocks
1014 ret = gfs2_quota_lock_check(ip, &ap);
1018 ret = gfs2_inplace_reserve(ip, &ap);
1023 rblocks = RES_DINODE + ind_blocks;
1024 if (gfs2_is_jdata(ip))
1025 rblocks += data_blocks;
1026 if (ind_blocks || data_blocks)
1027 rblocks += RES_STATFS + RES_QUOTA;
1028 if (inode == sdp->sd_rindex)
1029 rblocks += 2 * RES_STATFS;
1031 rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks);
1033 ret = gfs2_trans_begin(sdp, rblocks, iomap->length >> inode->i_blkbits);
1035 goto out_trans_fail;
1038 ret = gfs2_unstuff_dinode(ip, NULL);
1041 release_metapath(&mp);
1042 brelse(iomap->private);
1043 iomap->private = NULL;
1044 ret = gfs2_iomap_get(inode, iomap->offset, iomap->length,
1050 if (iomap->type == IOMAP_HOLE) {
1051 ret = gfs2_iomap_alloc(inode, iomap, flags, &mp);
1053 gfs2_trans_end(sdp);
1054 gfs2_inplace_release(ip);
1055 punch_hole(ip, iomap->offset, iomap->length);
1059 release_metapath(&mp);
1060 if (gfs2_is_jdata(ip))
1061 iomap->page_done = gfs2_iomap_journaled_page_done;
1065 gfs2_trans_end(sdp);
1068 gfs2_inplace_release(ip);
1071 gfs2_quota_unlock(ip);
1074 brelse(iomap->private);
1075 release_metapath(&mp);
1076 gfs2_write_unlock(inode);
1080 static int gfs2_iomap_begin(struct inode *inode, loff_t pos, loff_t length,
1081 unsigned flags, struct iomap *iomap)
1083 struct gfs2_inode *ip = GFS2_I(inode);
1084 struct metapath mp = { .mp_aheight = 1, };
1087 trace_gfs2_iomap_start(ip, pos, length, flags);
1088 if ((flags & IOMAP_WRITE) && !(flags & IOMAP_DIRECT)) {
1089 ret = gfs2_iomap_begin_write(inode, pos, length, flags, iomap);
1091 ret = gfs2_iomap_get(inode, pos, length, flags, iomap, &mp);
1092 release_metapath(&mp);
1094 * Silently fall back to buffered I/O for stuffed files or if
1095 * we've hot a hole (see gfs2_file_direct_write).
1097 if ((flags & IOMAP_WRITE) && (flags & IOMAP_DIRECT) &&
1098 iomap->type != IOMAP_MAPPED)
1101 trace_gfs2_iomap_end(ip, iomap, ret);
1105 static int gfs2_iomap_end(struct inode *inode, loff_t pos, loff_t length,
1106 ssize_t written, unsigned flags, struct iomap *iomap)
1108 struct gfs2_inode *ip = GFS2_I(inode);
1109 struct gfs2_sbd *sdp = GFS2_SB(inode);
1110 struct gfs2_trans *tr = current->journal_info;
1111 struct buffer_head *dibh = iomap->private;
1113 if ((flags & (IOMAP_WRITE | IOMAP_DIRECT)) != IOMAP_WRITE)
1116 if (iomap->type != IOMAP_INLINE) {
1117 gfs2_ordered_add_inode(ip);
1119 if (tr->tr_num_buf_new)
1120 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
1122 gfs2_trans_add_meta(ip->i_gl, dibh);
1125 if (inode == sdp->sd_rindex) {
1126 adjust_fs_space(inode);
1127 sdp->sd_rindex_uptodate = 0;
1130 gfs2_trans_end(sdp);
1131 gfs2_inplace_release(ip);
1133 if (length != written && (iomap->flags & IOMAP_F_NEW)) {
1134 /* Deallocate blocks that were just allocated. */
1135 loff_t blockmask = i_blocksize(inode) - 1;
1136 loff_t end = (pos + length) & ~blockmask;
1138 pos = (pos + written + blockmask) & ~blockmask;
1140 truncate_pagecache_range(inode, pos, end - 1);
1141 punch_hole(ip, pos, end - pos);
1145 if (ip->i_qadata && ip->i_qadata->qa_qd_num)
1146 gfs2_quota_unlock(ip);
1147 gfs2_write_unlock(inode);
1155 const struct iomap_ops gfs2_iomap_ops = {
1156 .iomap_begin = gfs2_iomap_begin,
1157 .iomap_end = gfs2_iomap_end,
1161 * gfs2_block_map - Map one or more blocks of an inode to a disk block
1163 * @lblock: The logical block number
1164 * @bh_map: The bh to be mapped
1165 * @create: True if its ok to alloc blocks to satify the request
1167 * The size of the requested mapping is defined in bh_map->b_size.
1169 * Clears buffer_mapped(bh_map) and leaves bh_map->b_size unchanged
1170 * when @lblock is not mapped. Sets buffer_mapped(bh_map) and
1171 * bh_map->b_size to indicate the size of the mapping when @lblock and
1172 * successive blocks are mapped, up to the requested size.
1174 * Sets buffer_boundary() if a read of metadata will be required
1175 * before the next block can be mapped. Sets buffer_new() if new
1176 * blocks were allocated.
1181 int gfs2_block_map(struct inode *inode, sector_t lblock,
1182 struct buffer_head *bh_map, int create)
1184 struct gfs2_inode *ip = GFS2_I(inode);
1185 loff_t pos = (loff_t)lblock << inode->i_blkbits;
1186 loff_t length = bh_map->b_size;
1187 struct metapath mp = { .mp_aheight = 1, };
1188 struct iomap iomap = { };
1191 clear_buffer_mapped(bh_map);
1192 clear_buffer_new(bh_map);
1193 clear_buffer_boundary(bh_map);
1194 trace_gfs2_bmap(ip, bh_map, lblock, create, 1);
1197 ret = gfs2_iomap_get(inode, pos, length, IOMAP_WRITE, &iomap, &mp);
1198 if (!ret && iomap.type == IOMAP_HOLE)
1199 ret = gfs2_iomap_alloc(inode, &iomap, IOMAP_WRITE, &mp);
1200 release_metapath(&mp);
1202 ret = gfs2_iomap_get(inode, pos, length, 0, &iomap, &mp);
1203 release_metapath(&mp);
1208 if (iomap.length > bh_map->b_size) {
1209 iomap.length = bh_map->b_size;
1210 iomap.flags &= ~IOMAP_F_GFS2_BOUNDARY;
1212 if (iomap.addr != IOMAP_NULL_ADDR)
1213 map_bh(bh_map, inode->i_sb, iomap.addr >> inode->i_blkbits);
1214 bh_map->b_size = iomap.length;
1215 if (iomap.flags & IOMAP_F_GFS2_BOUNDARY)
1216 set_buffer_boundary(bh_map);
1217 if (iomap.flags & IOMAP_F_NEW)
1218 set_buffer_new(bh_map);
1221 trace_gfs2_bmap(ip, bh_map, lblock, create, ret);
1226 * Deprecated: do not use in new code
1228 int gfs2_extent_map(struct inode *inode, u64 lblock, int *new, u64 *dblock, unsigned *extlen)
1230 struct buffer_head bh = { .b_state = 0, .b_blocknr = 0 };
1238 bh.b_size = BIT(inode->i_blkbits + (create ? 0 : 5));
1239 ret = gfs2_block_map(inode, lblock, &bh, create);
1240 *extlen = bh.b_size >> inode->i_blkbits;
1241 *dblock = bh.b_blocknr;
1242 if (buffer_new(&bh))
1250 * gfs2_block_zero_range - Deal with zeroing out data
1252 * This is partly borrowed from ext3.
1254 static int gfs2_block_zero_range(struct inode *inode, loff_t from,
1255 unsigned int length)
1257 struct address_space *mapping = inode->i_mapping;
1258 struct gfs2_inode *ip = GFS2_I(inode);
1259 unsigned long index = from >> PAGE_SHIFT;
1260 unsigned offset = from & (PAGE_SIZE-1);
1261 unsigned blocksize, iblock, pos;
1262 struct buffer_head *bh;
1266 page = find_or_create_page(mapping, index, GFP_NOFS);
1270 blocksize = inode->i_sb->s_blocksize;
1271 iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
1273 if (!page_has_buffers(page))
1274 create_empty_buffers(page, blocksize, 0);
1276 /* Find the buffer that contains "offset" */
1277 bh = page_buffers(page);
1279 while (offset >= pos) {
1280 bh = bh->b_this_page;
1287 if (!buffer_mapped(bh)) {
1288 gfs2_block_map(inode, iblock, bh, 0);
1289 /* unmapped? It's a hole - nothing to do */
1290 if (!buffer_mapped(bh))
1294 /* Ok, it's mapped. Make sure it's up-to-date */
1295 if (PageUptodate(page))
1296 set_buffer_uptodate(bh);
1298 if (!buffer_uptodate(bh)) {
1300 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1302 /* Uhhuh. Read error. Complain and punt. */
1303 if (!buffer_uptodate(bh))
1308 if (gfs2_is_jdata(ip))
1309 gfs2_trans_add_data(ip->i_gl, bh);
1311 gfs2_ordered_add_inode(ip);
1313 zero_user(page, offset, length);
1314 mark_buffer_dirty(bh);
1321 #define GFS2_JTRUNC_REVOKES 8192
1324 * gfs2_journaled_truncate - Wrapper for truncate_pagecache for jdata files
1325 * @inode: The inode being truncated
1326 * @oldsize: The original (larger) size
1327 * @newsize: The new smaller size
1329 * With jdata files, we have to journal a revoke for each block which is
1330 * truncated. As a result, we need to split this into separate transactions
1331 * if the number of pages being truncated gets too large.
1334 static int gfs2_journaled_truncate(struct inode *inode, u64 oldsize, u64 newsize)
1336 struct gfs2_sbd *sdp = GFS2_SB(inode);
1337 u64 max_chunk = GFS2_JTRUNC_REVOKES * sdp->sd_vfs->s_blocksize;
1341 while (oldsize != newsize) {
1342 struct gfs2_trans *tr;
1345 chunk = oldsize - newsize;
1346 if (chunk > max_chunk)
1349 offs = oldsize & ~PAGE_MASK;
1350 if (offs && chunk > PAGE_SIZE)
1351 chunk = offs + ((chunk - offs) & PAGE_MASK);
1353 truncate_pagecache(inode, oldsize - chunk);
1356 tr = current->journal_info;
1357 if (!test_bit(TR_TOUCHED, &tr->tr_flags))
1360 gfs2_trans_end(sdp);
1361 error = gfs2_trans_begin(sdp, RES_DINODE, GFS2_JTRUNC_REVOKES);
1369 static int trunc_start(struct inode *inode, u64 newsize)
1371 struct gfs2_inode *ip = GFS2_I(inode);
1372 struct gfs2_sbd *sdp = GFS2_SB(inode);
1373 struct buffer_head *dibh = NULL;
1374 int journaled = gfs2_is_jdata(ip);
1375 u64 oldsize = inode->i_size;
1379 error = gfs2_trans_begin(sdp, RES_DINODE + RES_JDATA, GFS2_JTRUNC_REVOKES);
1381 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1385 error = gfs2_meta_inode_buffer(ip, &dibh);
1389 gfs2_trans_add_meta(ip->i_gl, dibh);
1391 if (gfs2_is_stuffed(ip)) {
1392 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode) + newsize);
1394 unsigned int blocksize = i_blocksize(inode);
1395 unsigned int offs = newsize & (blocksize - 1);
1397 error = gfs2_block_zero_range(inode, newsize,
1402 ip->i_diskflags |= GFS2_DIF_TRUNC_IN_PROG;
1405 i_size_write(inode, newsize);
1406 ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode);
1407 gfs2_dinode_out(ip, dibh->b_data);
1410 error = gfs2_journaled_truncate(inode, oldsize, newsize);
1412 truncate_pagecache(inode, newsize);
1416 if (current->journal_info)
1417 gfs2_trans_end(sdp);
1421 int gfs2_iomap_get_alloc(struct inode *inode, loff_t pos, loff_t length,
1422 struct iomap *iomap)
1424 struct metapath mp = { .mp_aheight = 1, };
1427 ret = gfs2_iomap_get(inode, pos, length, IOMAP_WRITE, iomap, &mp);
1428 if (!ret && iomap->type == IOMAP_HOLE)
1429 ret = gfs2_iomap_alloc(inode, iomap, IOMAP_WRITE, &mp);
1430 release_metapath(&mp);
1435 * sweep_bh_for_rgrps - find an rgrp in a meta buffer and free blocks therein
1437 * @rg_gh: holder of resource group glock
1438 * @bh: buffer head to sweep
1439 * @start: starting point in bh
1440 * @end: end point in bh
1441 * @meta: true if bh points to metadata (rather than data)
1442 * @btotal: place to keep count of total blocks freed
1444 * We sweep a metadata buffer (provided by the metapath) for blocks we need to
1445 * free, and free them all. However, we do it one rgrp at a time. If this
1446 * block has references to multiple rgrps, we break it into individual
1447 * transactions. This allows other processes to use the rgrps while we're
1448 * focused on a single one, for better concurrency / performance.
1449 * At every transaction boundary, we rewrite the inode into the journal.
1450 * That way the bitmaps are kept consistent with the inode and we can recover
1451 * if we're interrupted by power-outages.
1453 * Returns: 0, or return code if an error occurred.
1454 * *btotal has the total number of blocks freed
1456 static int sweep_bh_for_rgrps(struct gfs2_inode *ip, struct gfs2_holder *rd_gh,
1457 struct buffer_head *bh, __be64 *start, __be64 *end,
1458 bool meta, u32 *btotal)
1460 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1461 struct gfs2_rgrpd *rgd;
1462 struct gfs2_trans *tr;
1464 int blks_outside_rgrp;
1465 u64 bn, bstart, isize_blks;
1466 s64 blen; /* needs to be s64 or gfs2_add_inode_blocks breaks */
1468 bool buf_in_tr = false; /* buffer was added to transaction */
1472 if (gfs2_holder_initialized(rd_gh)) {
1473 rgd = gfs2_glock2rgrp(rd_gh->gh_gl);
1474 gfs2_assert_withdraw(sdp,
1475 gfs2_glock_is_locked_by_me(rd_gh->gh_gl));
1477 blks_outside_rgrp = 0;
1481 for (p = start; p < end; p++) {
1484 bn = be64_to_cpu(*p);
1487 if (!rgrp_contains_block(rgd, bn)) {
1488 blks_outside_rgrp++;
1492 rgd = gfs2_blk2rgrpd(sdp, bn, true);
1493 if (unlikely(!rgd)) {
1497 ret = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE,
1502 /* Must be done with the rgrp glock held: */
1503 if (gfs2_rs_active(&ip->i_res) &&
1504 rgd == ip->i_res.rs_rbm.rgd)
1505 gfs2_rs_deltree(&ip->i_res);
1508 /* The size of our transactions will be unknown until we
1509 actually process all the metadata blocks that relate to
1510 the rgrp. So we estimate. We know it can't be more than
1511 the dinode's i_blocks and we don't want to exceed the
1512 journal flush threshold, sd_log_thresh2. */
1513 if (current->journal_info == NULL) {
1514 unsigned int jblocks_rqsted, revokes;
1516 jblocks_rqsted = rgd->rd_length + RES_DINODE +
1518 isize_blks = gfs2_get_inode_blocks(&ip->i_inode);
1519 if (isize_blks > atomic_read(&sdp->sd_log_thresh2))
1521 atomic_read(&sdp->sd_log_thresh2);
1523 jblocks_rqsted += isize_blks;
1524 revokes = jblocks_rqsted;
1526 revokes += end - start;
1527 else if (ip->i_depth)
1528 revokes += sdp->sd_inptrs;
1529 ret = gfs2_trans_begin(sdp, jblocks_rqsted, revokes);
1532 down_write(&ip->i_rw_mutex);
1534 /* check if we will exceed the transaction blocks requested */
1535 tr = current->journal_info;
1536 if (tr->tr_num_buf_new + RES_STATFS +
1537 RES_QUOTA >= atomic_read(&sdp->sd_log_thresh2)) {
1538 /* We set blks_outside_rgrp to ensure the loop will
1539 be repeated for the same rgrp, but with a new
1541 blks_outside_rgrp++;
1542 /* This next part is tricky. If the buffer was added
1543 to the transaction, we've already set some block
1544 pointers to 0, so we better follow through and free
1545 them, or we will introduce corruption (so break).
1546 This may be impossible, or at least rare, but I
1547 decided to cover the case regardless.
1549 If the buffer was not added to the transaction
1550 (this call), doing so would exceed our transaction
1551 size, so we need to end the transaction and start a
1552 new one (so goto). */
1559 gfs2_trans_add_meta(ip->i_gl, bh);
1562 if (bstart + blen == bn) {
1567 __gfs2_free_blocks(ip, bstart, (u32)blen, meta);
1569 gfs2_add_inode_blocks(&ip->i_inode, -blen);
1575 __gfs2_free_blocks(ip, bstart, (u32)blen, meta);
1577 gfs2_add_inode_blocks(&ip->i_inode, -blen);
1580 if (!ret && blks_outside_rgrp) { /* If buffer still has non-zero blocks
1581 outside the rgrp we just processed,
1582 do it all over again. */
1583 if (current->journal_info) {
1584 struct buffer_head *dibh;
1586 ret = gfs2_meta_inode_buffer(ip, &dibh);
1590 /* Every transaction boundary, we rewrite the dinode
1591 to keep its di_blocks current in case of failure. */
1592 ip->i_inode.i_mtime = ip->i_inode.i_ctime =
1593 current_time(&ip->i_inode);
1594 gfs2_trans_add_meta(ip->i_gl, dibh);
1595 gfs2_dinode_out(ip, dibh->b_data);
1597 up_write(&ip->i_rw_mutex);
1598 gfs2_trans_end(sdp);
1600 gfs2_glock_dq_uninit(rd_gh);
1608 static bool mp_eq_to_hgt(struct metapath *mp, __u16 *list, unsigned int h)
1610 if (memcmp(mp->mp_list, list, h * sizeof(mp->mp_list[0])))
1616 * find_nonnull_ptr - find a non-null pointer given a metapath and height
1617 * @mp: starting metapath
1618 * @h: desired height to search
1620 * Assumes the metapath is valid (with buffers) out to height h.
1621 * Returns: true if a non-null pointer was found in the metapath buffer
1622 * false if all remaining pointers are NULL in the buffer
1624 static bool find_nonnull_ptr(struct gfs2_sbd *sdp, struct metapath *mp,
1626 __u16 *end_list, unsigned int end_aligned)
1628 struct buffer_head *bh = mp->mp_bh[h];
1629 __be64 *first, *ptr, *end;
1631 first = metaptr1(h, mp);
1632 ptr = first + mp->mp_list[h];
1633 end = (__be64 *)(bh->b_data + bh->b_size);
1634 if (end_list && mp_eq_to_hgt(mp, end_list, h)) {
1635 bool keep_end = h < end_aligned;
1636 end = first + end_list[h] + keep_end;
1640 if (*ptr) { /* if we have a non-null pointer */
1641 mp->mp_list[h] = ptr - first;
1643 if (h < GFS2_MAX_META_HEIGHT)
1652 enum dealloc_states {
1653 DEALLOC_MP_FULL = 0, /* Strip a metapath with all buffers read in */
1654 DEALLOC_MP_LOWER = 1, /* lower the metapath strip height */
1655 DEALLOC_FILL_MP = 2, /* Fill in the metapath to the given height. */
1656 DEALLOC_DONE = 3, /* process complete */
1660 metapointer_range(struct metapath *mp, int height,
1661 __u16 *start_list, unsigned int start_aligned,
1662 __u16 *end_list, unsigned int end_aligned,
1663 __be64 **start, __be64 **end)
1665 struct buffer_head *bh = mp->mp_bh[height];
1668 first = metaptr1(height, mp);
1670 if (mp_eq_to_hgt(mp, start_list, height)) {
1671 bool keep_start = height < start_aligned;
1672 *start = first + start_list[height] + keep_start;
1674 *end = (__be64 *)(bh->b_data + bh->b_size);
1675 if (end_list && mp_eq_to_hgt(mp, end_list, height)) {
1676 bool keep_end = height < end_aligned;
1677 *end = first + end_list[height] + keep_end;
1681 static inline bool walk_done(struct gfs2_sbd *sdp,
1682 struct metapath *mp, int height,
1683 __u16 *end_list, unsigned int end_aligned)
1688 bool keep_end = height < end_aligned;
1689 if (!mp_eq_to_hgt(mp, end_list, height))
1691 end = end_list[height] + keep_end;
1693 end = (height > 0) ? sdp->sd_inptrs : sdp->sd_diptrs;
1694 return mp->mp_list[height] >= end;
1698 * punch_hole - deallocate blocks in a file
1699 * @ip: inode to truncate
1700 * @offset: the start of the hole
1701 * @length: the size of the hole (or 0 for truncate)
1703 * Punch a hole into a file or truncate a file at a given position. This
1704 * function operates in whole blocks (@offset and @length are rounded
1705 * accordingly); partially filled blocks must be cleared otherwise.
1707 * This function works from the bottom up, and from the right to the left. In
1708 * other words, it strips off the highest layer (data) before stripping any of
1709 * the metadata. Doing it this way is best in case the operation is interrupted
1710 * by power failure, etc. The dinode is rewritten in every transaction to
1711 * guarantee integrity.
1713 static int punch_hole(struct gfs2_inode *ip, u64 offset, u64 length)
1715 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1716 u64 maxsize = sdp->sd_heightsize[ip->i_height];
1717 struct metapath mp = {};
1718 struct buffer_head *dibh, *bh;
1719 struct gfs2_holder rd_gh;
1720 unsigned int bsize_shift = sdp->sd_sb.sb_bsize_shift;
1721 u64 lblock = (offset + (1 << bsize_shift) - 1) >> bsize_shift;
1722 __u16 start_list[GFS2_MAX_META_HEIGHT];
1723 __u16 __end_list[GFS2_MAX_META_HEIGHT], *end_list = NULL;
1724 unsigned int start_aligned, uninitialized_var(end_aligned);
1725 unsigned int strip_h = ip->i_height - 1;
1728 int mp_h; /* metapath buffers are read in to this height */
1730 __be64 *start, *end;
1732 if (offset >= maxsize) {
1734 * The starting point lies beyond the allocated meta-data;
1735 * there are no blocks do deallocate.
1741 * The start position of the hole is defined by lblock, start_list, and
1742 * start_aligned. The end position of the hole is defined by lend,
1743 * end_list, and end_aligned.
1745 * start_aligned and end_aligned define down to which height the start
1746 * and end positions are aligned to the metadata tree (i.e., the
1747 * position is a multiple of the metadata granularity at the height
1748 * above). This determines at which heights additional meta pointers
1749 * needs to be preserved for the remaining data.
1753 u64 end_offset = offset + length;
1757 * Clip the end at the maximum file size for the given height:
1758 * that's how far the metadata goes; files bigger than that
1759 * will have additional layers of indirection.
1761 if (end_offset > maxsize)
1762 end_offset = maxsize;
1763 lend = end_offset >> bsize_shift;
1768 find_metapath(sdp, lend, &mp, ip->i_height);
1769 end_list = __end_list;
1770 memcpy(end_list, mp.mp_list, sizeof(mp.mp_list));
1772 for (mp_h = ip->i_height - 1; mp_h > 0; mp_h--) {
1779 find_metapath(sdp, lblock, &mp, ip->i_height);
1780 memcpy(start_list, mp.mp_list, sizeof(start_list));
1782 for (mp_h = ip->i_height - 1; mp_h > 0; mp_h--) {
1783 if (start_list[mp_h])
1786 start_aligned = mp_h;
1788 ret = gfs2_meta_inode_buffer(ip, &dibh);
1793 ret = lookup_metapath(ip, &mp);
1797 /* issue read-ahead on metadata */
1798 for (mp_h = 0; mp_h < mp.mp_aheight - 1; mp_h++) {
1799 metapointer_range(&mp, mp_h, start_list, start_aligned,
1800 end_list, end_aligned, &start, &end);
1801 gfs2_metapath_ra(ip->i_gl, start, end);
1804 if (mp.mp_aheight == ip->i_height)
1805 state = DEALLOC_MP_FULL; /* We have a complete metapath */
1807 state = DEALLOC_FILL_MP; /* deal with partial metapath */
1809 ret = gfs2_rindex_update(sdp);
1813 ret = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1816 gfs2_holder_mark_uninitialized(&rd_gh);
1820 while (state != DEALLOC_DONE) {
1822 /* Truncate a full metapath at the given strip height.
1823 * Note that strip_h == mp_h in order to be in this state. */
1824 case DEALLOC_MP_FULL:
1825 bh = mp.mp_bh[mp_h];
1826 gfs2_assert_withdraw(sdp, bh);
1827 if (gfs2_assert_withdraw(sdp,
1828 prev_bnr != bh->b_blocknr)) {
1829 printk(KERN_EMERG "GFS2: fsid=%s:inode %llu, "
1830 "block:%llu, i_h:%u, s_h:%u, mp_h:%u\n",
1832 (unsigned long long)ip->i_no_addr,
1833 prev_bnr, ip->i_height, strip_h, mp_h);
1835 prev_bnr = bh->b_blocknr;
1837 if (gfs2_metatype_check(sdp, bh,
1838 (mp_h ? GFS2_METATYPE_IN :
1839 GFS2_METATYPE_DI))) {
1845 * Below, passing end_aligned as 0 gives us the
1846 * metapointer range excluding the end point: the end
1847 * point is the first metapath we must not deallocate!
1850 metapointer_range(&mp, mp_h, start_list, start_aligned,
1851 end_list, 0 /* end_aligned */,
1853 ret = sweep_bh_for_rgrps(ip, &rd_gh, mp.mp_bh[mp_h],
1855 mp_h != ip->i_height - 1,
1858 /* If we hit an error or just swept dinode buffer,
1861 state = DEALLOC_DONE;
1864 state = DEALLOC_MP_LOWER;
1867 /* lower the metapath strip height */
1868 case DEALLOC_MP_LOWER:
1869 /* We're done with the current buffer, so release it,
1870 unless it's the dinode buffer. Then back up to the
1871 previous pointer. */
1873 brelse(mp.mp_bh[mp_h]);
1874 mp.mp_bh[mp_h] = NULL;
1876 /* If we can't get any lower in height, we've stripped
1877 off all we can. Next step is to back up and start
1878 stripping the previous level of metadata. */
1881 memcpy(mp.mp_list, start_list, sizeof(start_list));
1883 state = DEALLOC_FILL_MP;
1886 mp.mp_list[mp_h] = 0;
1887 mp_h--; /* search one metadata height down */
1889 if (walk_done(sdp, &mp, mp_h, end_list, end_aligned))
1891 /* Here we've found a part of the metapath that is not
1892 * allocated. We need to search at that height for the
1893 * next non-null pointer. */
1894 if (find_nonnull_ptr(sdp, &mp, mp_h, end_list, end_aligned)) {
1895 state = DEALLOC_FILL_MP;
1898 /* No more non-null pointers at this height. Back up
1899 to the previous height and try again. */
1900 break; /* loop around in the same state */
1902 /* Fill the metapath with buffers to the given height. */
1903 case DEALLOC_FILL_MP:
1904 /* Fill the buffers out to the current height. */
1905 ret = fillup_metapath(ip, &mp, mp_h);
1909 /* issue read-ahead on metadata */
1910 if (mp.mp_aheight > 1) {
1911 for (; ret > 1; ret--) {
1912 metapointer_range(&mp, mp.mp_aheight - ret,
1913 start_list, start_aligned,
1914 end_list, end_aligned,
1916 gfs2_metapath_ra(ip->i_gl, start, end);
1920 /* If buffers found for the entire strip height */
1921 if (mp.mp_aheight - 1 == strip_h) {
1922 state = DEALLOC_MP_FULL;
1925 if (mp.mp_aheight < ip->i_height) /* We have a partial height */
1926 mp_h = mp.mp_aheight - 1;
1928 /* If we find a non-null block pointer, crawl a bit
1929 higher up in the metapath and try again, otherwise
1930 we need to look lower for a new starting point. */
1931 if (find_nonnull_ptr(sdp, &mp, mp_h, end_list, end_aligned))
1934 state = DEALLOC_MP_LOWER;
1940 if (current->journal_info == NULL) {
1941 ret = gfs2_trans_begin(sdp, RES_DINODE + RES_STATFS +
1945 down_write(&ip->i_rw_mutex);
1947 gfs2_statfs_change(sdp, 0, +btotal, 0);
1948 gfs2_quota_change(ip, -(s64)btotal, ip->i_inode.i_uid,
1950 ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode);
1951 gfs2_trans_add_meta(ip->i_gl, dibh);
1952 gfs2_dinode_out(ip, dibh->b_data);
1953 up_write(&ip->i_rw_mutex);
1954 gfs2_trans_end(sdp);
1958 if (gfs2_holder_initialized(&rd_gh))
1959 gfs2_glock_dq_uninit(&rd_gh);
1960 if (current->journal_info) {
1961 up_write(&ip->i_rw_mutex);
1962 gfs2_trans_end(sdp);
1965 gfs2_quota_unhold(ip);
1967 release_metapath(&mp);
1971 static int trunc_end(struct gfs2_inode *ip)
1973 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1974 struct buffer_head *dibh;
1977 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1981 down_write(&ip->i_rw_mutex);
1983 error = gfs2_meta_inode_buffer(ip, &dibh);
1987 if (!i_size_read(&ip->i_inode)) {
1989 ip->i_goal = ip->i_no_addr;
1990 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
1991 gfs2_ordered_del_inode(ip);
1993 ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode);
1994 ip->i_diskflags &= ~GFS2_DIF_TRUNC_IN_PROG;
1996 gfs2_trans_add_meta(ip->i_gl, dibh);
1997 gfs2_dinode_out(ip, dibh->b_data);
2001 up_write(&ip->i_rw_mutex);
2002 gfs2_trans_end(sdp);
2007 * do_shrink - make a file smaller
2009 * @newsize: the size to make the file
2011 * Called with an exclusive lock on @inode. The @size must
2012 * be equal to or smaller than the current inode size.
2017 static int do_shrink(struct inode *inode, u64 newsize)
2019 struct gfs2_inode *ip = GFS2_I(inode);
2022 error = trunc_start(inode, newsize);
2025 if (gfs2_is_stuffed(ip))
2028 error = punch_hole(ip, newsize, 0);
2030 error = trunc_end(ip);
2035 void gfs2_trim_blocks(struct inode *inode)
2039 ret = do_shrink(inode, inode->i_size);
2044 * do_grow - Touch and update inode size
2046 * @size: The new size
2048 * This function updates the timestamps on the inode and
2049 * may also increase the size of the inode. This function
2050 * must not be called with @size any smaller than the current
2053 * Although it is not strictly required to unstuff files here,
2054 * earlier versions of GFS2 have a bug in the stuffed file reading
2055 * code which will result in a buffer overrun if the size is larger
2056 * than the max stuffed file size. In order to prevent this from
2057 * occurring, such files are unstuffed, but in other cases we can
2058 * just update the inode size directly.
2060 * Returns: 0 on success, or -ve on error
2063 static int do_grow(struct inode *inode, u64 size)
2065 struct gfs2_inode *ip = GFS2_I(inode);
2066 struct gfs2_sbd *sdp = GFS2_SB(inode);
2067 struct gfs2_alloc_parms ap = { .target = 1, };
2068 struct buffer_head *dibh;
2072 if (gfs2_is_stuffed(ip) && size > gfs2_max_stuffed_size(ip)) {
2073 error = gfs2_quota_lock_check(ip, &ap);
2077 error = gfs2_inplace_reserve(ip, &ap);
2079 goto do_grow_qunlock;
2083 error = gfs2_trans_begin(sdp, RES_DINODE + RES_STATFS + RES_RG_BIT +
2084 (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF ?
2087 goto do_grow_release;
2090 error = gfs2_unstuff_dinode(ip, NULL);
2095 error = gfs2_meta_inode_buffer(ip, &dibh);
2099 i_size_write(inode, size);
2100 ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode);
2101 gfs2_trans_add_meta(ip->i_gl, dibh);
2102 gfs2_dinode_out(ip, dibh->b_data);
2106 gfs2_trans_end(sdp);
2109 gfs2_inplace_release(ip);
2111 gfs2_quota_unlock(ip);
2117 * gfs2_setattr_size - make a file a given size
2119 * @newsize: the size to make the file
2121 * The file size can grow, shrink, or stay the same size. This
2122 * is called holding i_rwsem and an exclusive glock on the inode
2128 int gfs2_setattr_size(struct inode *inode, u64 newsize)
2130 struct gfs2_inode *ip = GFS2_I(inode);
2133 BUG_ON(!S_ISREG(inode->i_mode));
2135 ret = inode_newsize_ok(inode, newsize);
2139 inode_dio_wait(inode);
2141 ret = gfs2_rsqa_alloc(ip);
2145 if (newsize >= inode->i_size) {
2146 ret = do_grow(inode, newsize);
2150 ret = do_shrink(inode, newsize);
2152 gfs2_rsqa_delete(ip, NULL);
2156 int gfs2_truncatei_resume(struct gfs2_inode *ip)
2159 error = punch_hole(ip, i_size_read(&ip->i_inode), 0);
2161 error = trunc_end(ip);
2165 int gfs2_file_dealloc(struct gfs2_inode *ip)
2167 return punch_hole(ip, 0, 0);
2171 * gfs2_free_journal_extents - Free cached journal bmap info
2176 void gfs2_free_journal_extents(struct gfs2_jdesc *jd)
2178 struct gfs2_journal_extent *jext;
2180 while(!list_empty(&jd->extent_list)) {
2181 jext = list_entry(jd->extent_list.next, struct gfs2_journal_extent, list);
2182 list_del(&jext->list);
2188 * gfs2_add_jextent - Add or merge a new extent to extent cache
2189 * @jd: The journal descriptor
2190 * @lblock: The logical block at start of new extent
2191 * @dblock: The physical block at start of new extent
2192 * @blocks: Size of extent in fs blocks
2194 * Returns: 0 on success or -ENOMEM
2197 static int gfs2_add_jextent(struct gfs2_jdesc *jd, u64 lblock, u64 dblock, u64 blocks)
2199 struct gfs2_journal_extent *jext;
2201 if (!list_empty(&jd->extent_list)) {
2202 jext = list_entry(jd->extent_list.prev, struct gfs2_journal_extent, list);
2203 if ((jext->dblock + jext->blocks) == dblock) {
2204 jext->blocks += blocks;
2209 jext = kzalloc(sizeof(struct gfs2_journal_extent), GFP_NOFS);
2212 jext->dblock = dblock;
2213 jext->lblock = lblock;
2214 jext->blocks = blocks;
2215 list_add_tail(&jext->list, &jd->extent_list);
2221 * gfs2_map_journal_extents - Cache journal bmap info
2222 * @sdp: The super block
2223 * @jd: The journal to map
2225 * Create a reusable "extent" mapping from all logical
2226 * blocks to all physical blocks for the given journal. This will save
2227 * us time when writing journal blocks. Most journals will have only one
2228 * extent that maps all their logical blocks. That's because gfs2.mkfs
2229 * arranges the journal blocks sequentially to maximize performance.
2230 * So the extent would map the first block for the entire file length.
2231 * However, gfs2_jadd can happen while file activity is happening, so
2232 * those journals may not be sequential. Less likely is the case where
2233 * the users created their own journals by mounting the metafs and
2234 * laying it out. But it's still possible. These journals might have
2237 * Returns: 0 on success, or error on failure
2240 int gfs2_map_journal_extents(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd)
2244 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
2245 struct buffer_head bh;
2246 unsigned int shift = sdp->sd_sb.sb_bsize_shift;
2250 lblock_stop = i_size_read(jd->jd_inode) >> shift;
2251 size = (lblock_stop - lblock) << shift;
2253 WARN_ON(!list_empty(&jd->extent_list));
2259 rc = gfs2_block_map(jd->jd_inode, lblock, &bh, 0);
2260 if (rc || !buffer_mapped(&bh))
2262 rc = gfs2_add_jextent(jd, lblock, bh.b_blocknr, bh.b_size >> shift);
2266 lblock += (bh.b_size >> ip->i_inode.i_blkbits);
2269 fs_info(sdp, "journal %d mapped with %u extents\n", jd->jd_jid,
2274 fs_warn(sdp, "error %d mapping journal %u at offset %llu (extent %u)\n",
2276 (unsigned long long)(i_size_read(jd->jd_inode) - size),
2278 fs_warn(sdp, "bmap=%d lblock=%llu block=%llu, state=0x%08lx, size=%llu\n",
2279 rc, (unsigned long long)lblock, (unsigned long long)bh.b_blocknr,
2280 bh.b_state, (unsigned long long)bh.b_size);
2281 gfs2_free_journal_extents(jd);
2286 * gfs2_write_alloc_required - figure out if a write will require an allocation
2287 * @ip: the file being written to
2288 * @offset: the offset to write to
2289 * @len: the number of bytes being written
2291 * Returns: 1 if an alloc is required, 0 otherwise
2294 int gfs2_write_alloc_required(struct gfs2_inode *ip, u64 offset,
2297 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
2298 struct buffer_head bh;
2300 u64 lblock, lblock_stop, size;
2306 if (gfs2_is_stuffed(ip)) {
2307 if (offset + len > gfs2_max_stuffed_size(ip))
2312 shift = sdp->sd_sb.sb_bsize_shift;
2313 BUG_ON(gfs2_is_dir(ip));
2314 end_of_file = (i_size_read(&ip->i_inode) + sdp->sd_sb.sb_bsize - 1) >> shift;
2315 lblock = offset >> shift;
2316 lblock_stop = (offset + len + sdp->sd_sb.sb_bsize - 1) >> shift;
2317 if (lblock_stop > end_of_file)
2320 size = (lblock_stop - lblock) << shift;
2324 gfs2_block_map(&ip->i_inode, lblock, &bh, 0);
2325 if (!buffer_mapped(&bh))
2328 lblock += (bh.b_size >> ip->i_inode.i_blkbits);
2334 static int stuffed_zero_range(struct inode *inode, loff_t offset, loff_t length)
2336 struct gfs2_inode *ip = GFS2_I(inode);
2337 struct buffer_head *dibh;
2340 if (offset >= inode->i_size)
2342 if (offset + length > inode->i_size)
2343 length = inode->i_size - offset;
2345 error = gfs2_meta_inode_buffer(ip, &dibh);
2348 gfs2_trans_add_meta(ip->i_gl, dibh);
2349 memset(dibh->b_data + sizeof(struct gfs2_dinode) + offset, 0,
2355 static int gfs2_journaled_truncate_range(struct inode *inode, loff_t offset,
2358 struct gfs2_sbd *sdp = GFS2_SB(inode);
2359 loff_t max_chunk = GFS2_JTRUNC_REVOKES * sdp->sd_vfs->s_blocksize;
2363 struct gfs2_trans *tr;
2368 if (chunk > max_chunk)
2371 offs = offset & ~PAGE_MASK;
2372 if (offs && chunk > PAGE_SIZE)
2373 chunk = offs + ((chunk - offs) & PAGE_MASK);
2375 truncate_pagecache_range(inode, offset, chunk);
2379 tr = current->journal_info;
2380 if (!test_bit(TR_TOUCHED, &tr->tr_flags))
2383 gfs2_trans_end(sdp);
2384 error = gfs2_trans_begin(sdp, RES_DINODE, GFS2_JTRUNC_REVOKES);
2391 int __gfs2_punch_hole(struct file *file, loff_t offset, loff_t length)
2393 struct inode *inode = file_inode(file);
2394 struct gfs2_inode *ip = GFS2_I(inode);
2395 struct gfs2_sbd *sdp = GFS2_SB(inode);
2398 if (gfs2_is_jdata(ip))
2399 error = gfs2_trans_begin(sdp, RES_DINODE + 2 * RES_JDATA,
2400 GFS2_JTRUNC_REVOKES);
2402 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
2406 if (gfs2_is_stuffed(ip)) {
2407 error = stuffed_zero_range(inode, offset, length);
2411 unsigned int start_off, end_len, blocksize;
2413 blocksize = i_blocksize(inode);
2414 start_off = offset & (blocksize - 1);
2415 end_len = (offset + length) & (blocksize - 1);
2417 unsigned int len = length;
2418 if (length > blocksize - start_off)
2419 len = blocksize - start_off;
2420 error = gfs2_block_zero_range(inode, offset, len);
2423 if (start_off + length < blocksize)
2427 error = gfs2_block_zero_range(inode,
2428 offset + length - end_len, end_len);
2434 if (gfs2_is_jdata(ip)) {
2435 BUG_ON(!current->journal_info);
2436 gfs2_journaled_truncate_range(inode, offset, length);
2438 truncate_pagecache_range(inode, offset, offset + length - 1);
2440 file_update_time(file);
2441 mark_inode_dirty(inode);
2443 if (current->journal_info)
2444 gfs2_trans_end(sdp);
2446 if (!gfs2_is_stuffed(ip))
2447 error = punch_hole(ip, offset, length);
2450 if (current->journal_info)
2451 gfs2_trans_end(sdp);