xref: /linux/fs/gfs2/bmap.c (revision b24413180f5600bcb3bb70fbed5cf186b60864bd)
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2006 Red Hat, Inc.  All rights reserved.
4  *
5  * This copyrighted material is made available to anyone wishing to use,
6  * modify, copy, or redistribute it subject to the terms and conditions
7  * of the GNU General Public License version 2.
8  */
9 
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 
17 #include "gfs2.h"
18 #include "incore.h"
19 #include "bmap.h"
20 #include "glock.h"
21 #include "inode.h"
22 #include "meta_io.h"
23 #include "quota.h"
24 #include "rgrp.h"
25 #include "log.h"
26 #include "super.h"
27 #include "trans.h"
28 #include "dir.h"
29 #include "util.h"
30 #include "trace_gfs2.h"
31 
32 /* This doesn't need to be that large as max 64 bit pointers in a 4k
33  * block is 512, so __u16 is fine for that. It saves stack space to
34  * keep it small.
35  */
36 struct metapath {
37 	struct buffer_head *mp_bh[GFS2_MAX_META_HEIGHT];
38 	__u16 mp_list[GFS2_MAX_META_HEIGHT];
39 };
40 
41 /**
42  * gfs2_unstuffer_page - unstuff a stuffed inode into a block cached by a page
43  * @ip: the inode
44  * @dibh: the dinode buffer
45  * @block: the block number that was allocated
46  * @page: The (optional) page. This is looked up if @page is NULL
47  *
48  * Returns: errno
49  */
50 
51 static int gfs2_unstuffer_page(struct gfs2_inode *ip, struct buffer_head *dibh,
52 			       u64 block, struct page *page)
53 {
54 	struct inode *inode = &ip->i_inode;
55 	struct buffer_head *bh;
56 	int release = 0;
57 
58 	if (!page || page->index) {
59 		page = find_or_create_page(inode->i_mapping, 0, GFP_NOFS);
60 		if (!page)
61 			return -ENOMEM;
62 		release = 1;
63 	}
64 
65 	if (!PageUptodate(page)) {
66 		void *kaddr = kmap(page);
67 		u64 dsize = i_size_read(inode);
68 
69 		if (dsize > (dibh->b_size - sizeof(struct gfs2_dinode)))
70 			dsize = dibh->b_size - sizeof(struct gfs2_dinode);
71 
72 		memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize);
73 		memset(kaddr + dsize, 0, PAGE_SIZE - dsize);
74 		kunmap(page);
75 
76 		SetPageUptodate(page);
77 	}
78 
79 	if (!page_has_buffers(page))
80 		create_empty_buffers(page, BIT(inode->i_blkbits),
81 				     BIT(BH_Uptodate));
82 
83 	bh = page_buffers(page);
84 
85 	if (!buffer_mapped(bh))
86 		map_bh(bh, inode->i_sb, block);
87 
88 	set_buffer_uptodate(bh);
89 	if (!gfs2_is_jdata(ip))
90 		mark_buffer_dirty(bh);
91 	if (!gfs2_is_writeback(ip))
92 		gfs2_trans_add_data(ip->i_gl, bh);
93 
94 	if (release) {
95 		unlock_page(page);
96 		put_page(page);
97 	}
98 
99 	return 0;
100 }
101 
102 /**
103  * gfs2_unstuff_dinode - Unstuff a dinode when the data has grown too big
104  * @ip: The GFS2 inode to unstuff
105  * @page: The (optional) page. This is looked up if the @page is NULL
106  *
107  * This routine unstuffs a dinode and returns it to a "normal" state such
108  * that the height can be grown in the traditional way.
109  *
110  * Returns: errno
111  */
112 
113 int gfs2_unstuff_dinode(struct gfs2_inode *ip, struct page *page)
114 {
115 	struct buffer_head *bh, *dibh;
116 	struct gfs2_dinode *di;
117 	u64 block = 0;
118 	int isdir = gfs2_is_dir(ip);
119 	int error;
120 
121 	down_write(&ip->i_rw_mutex);
122 
123 	error = gfs2_meta_inode_buffer(ip, &dibh);
124 	if (error)
125 		goto out;
126 
127 	if (i_size_read(&ip->i_inode)) {
128 		/* Get a free block, fill it with the stuffed data,
129 		   and write it out to disk */
130 
131 		unsigned int n = 1;
132 		error = gfs2_alloc_blocks(ip, &block, &n, 0, NULL);
133 		if (error)
134 			goto out_brelse;
135 		if (isdir) {
136 			gfs2_trans_add_unrevoke(GFS2_SB(&ip->i_inode), block, 1);
137 			error = gfs2_dir_get_new_buffer(ip, block, &bh);
138 			if (error)
139 				goto out_brelse;
140 			gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_meta_header),
141 					      dibh, sizeof(struct gfs2_dinode));
142 			brelse(bh);
143 		} else {
144 			error = gfs2_unstuffer_page(ip, dibh, block, page);
145 			if (error)
146 				goto out_brelse;
147 		}
148 	}
149 
150 	/*  Set up the pointer to the new block  */
151 
152 	gfs2_trans_add_meta(ip->i_gl, dibh);
153 	di = (struct gfs2_dinode *)dibh->b_data;
154 	gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
155 
156 	if (i_size_read(&ip->i_inode)) {
157 		*(__be64 *)(di + 1) = cpu_to_be64(block);
158 		gfs2_add_inode_blocks(&ip->i_inode, 1);
159 		di->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
160 	}
161 
162 	ip->i_height = 1;
163 	di->di_height = cpu_to_be16(1);
164 
165 out_brelse:
166 	brelse(dibh);
167 out:
168 	up_write(&ip->i_rw_mutex);
169 	return error;
170 }
171 
172 
173 /**
174  * find_metapath - Find path through the metadata tree
175  * @sdp: The superblock
176  * @mp: The metapath to return the result in
177  * @block: The disk block to look up
178  * @height: The pre-calculated height of the metadata tree
179  *
180  *   This routine returns a struct metapath structure that defines a path
181  *   through the metadata of inode "ip" to get to block "block".
182  *
183  *   Example:
184  *   Given:  "ip" is a height 3 file, "offset" is 101342453, and this is a
185  *   filesystem with a blocksize of 4096.
186  *
187  *   find_metapath() would return a struct metapath structure set to:
188  *   mp_offset = 101342453, mp_height = 3, mp_list[0] = 0, mp_list[1] = 48,
189  *   and mp_list[2] = 165.
190  *
191  *   That means that in order to get to the block containing the byte at
192  *   offset 101342453, we would load the indirect block pointed to by pointer
193  *   0 in the dinode.  We would then load the indirect block pointed to by
194  *   pointer 48 in that indirect block.  We would then load the data block
195  *   pointed to by pointer 165 in that indirect block.
196  *
197  *             ----------------------------------------
198  *             | Dinode |                             |
199  *             |        |                            4|
200  *             |        |0 1 2 3 4 5                 9|
201  *             |        |                            6|
202  *             ----------------------------------------
203  *                       |
204  *                       |
205  *                       V
206  *             ----------------------------------------
207  *             | Indirect Block                       |
208  *             |                                     5|
209  *             |            4 4 4 4 4 5 5            1|
210  *             |0           5 6 7 8 9 0 1            2|
211  *             ----------------------------------------
212  *                                |
213  *                                |
214  *                                V
215  *             ----------------------------------------
216  *             | Indirect Block                       |
217  *             |                         1 1 1 1 1   5|
218  *             |                         6 6 6 6 6   1|
219  *             |0                        3 4 5 6 7   2|
220  *             ----------------------------------------
221  *                                           |
222  *                                           |
223  *                                           V
224  *             ----------------------------------------
225  *             | Data block containing offset         |
226  *             |            101342453                 |
227  *             |                                      |
228  *             |                                      |
229  *             ----------------------------------------
230  *
231  */
232 
233 static void find_metapath(const struct gfs2_sbd *sdp, u64 block,
234 			  struct metapath *mp, unsigned int height)
235 {
236 	unsigned int i;
237 
238 	for (i = height; i--;)
239 		mp->mp_list[i] = do_div(block, sdp->sd_inptrs);
240 
241 }
242 
243 static inline unsigned int metapath_branch_start(const struct metapath *mp)
244 {
245 	if (mp->mp_list[0] == 0)
246 		return 2;
247 	return 1;
248 }
249 
250 /**
251  * metaptr1 - Return the first possible metadata pointer in a metaath buffer
252  * @height: The metadata height (0 = dinode)
253  * @mp: The metapath
254  */
255 static inline __be64 *metaptr1(unsigned int height, const struct metapath *mp)
256 {
257 	struct buffer_head *bh = mp->mp_bh[height];
258 	if (height == 0)
259 		return ((__be64 *)(bh->b_data + sizeof(struct gfs2_dinode)));
260 	return ((__be64 *)(bh->b_data + sizeof(struct gfs2_meta_header)));
261 }
262 
263 /**
264  * metapointer - Return pointer to start of metadata in a buffer
265  * @height: The metadata height (0 = dinode)
266  * @mp: The metapath
267  *
268  * Return a pointer to the block number of the next height of the metadata
269  * tree given a buffer containing the pointer to the current height of the
270  * metadata tree.
271  */
272 
273 static inline __be64 *metapointer(unsigned int height, const struct metapath *mp)
274 {
275 	__be64 *p = metaptr1(height, mp);
276 	return p + mp->mp_list[height];
277 }
278 
279 static void gfs2_metapath_ra(struct gfs2_glock *gl,
280 			     const struct buffer_head *bh, const __be64 *pos)
281 {
282 	struct buffer_head *rabh;
283 	const __be64 *endp = (const __be64 *)(bh->b_data + bh->b_size);
284 	const __be64 *t;
285 
286 	for (t = pos; t < endp; t++) {
287 		if (!*t)
288 			continue;
289 
290 		rabh = gfs2_getbuf(gl, be64_to_cpu(*t), CREATE);
291 		if (trylock_buffer(rabh)) {
292 			if (!buffer_uptodate(rabh)) {
293 				rabh->b_end_io = end_buffer_read_sync;
294 				submit_bh(REQ_OP_READ,
295 					  REQ_RAHEAD | REQ_META | REQ_PRIO,
296 					  rabh);
297 				continue;
298 			}
299 			unlock_buffer(rabh);
300 		}
301 		brelse(rabh);
302 	}
303 }
304 
305 /**
306  * lookup_mp_height - helper function for lookup_metapath
307  * @ip: the inode
308  * @mp: the metapath
309  * @h: the height which needs looking up
310  */
311 static int lookup_mp_height(struct gfs2_inode *ip, struct metapath *mp, int h)
312 {
313 	__be64 *ptr = metapointer(h, mp);
314 	u64 dblock = be64_to_cpu(*ptr);
315 
316 	if (!dblock)
317 		return h + 1;
318 
319 	return gfs2_meta_indirect_buffer(ip, h + 1, dblock, &mp->mp_bh[h + 1]);
320 }
321 
322 /**
323  * lookup_metapath - Walk the metadata tree to a specific point
324  * @ip: The inode
325  * @mp: The metapath
326  *
327  * Assumes that the inode's buffer has already been looked up and
328  * hooked onto mp->mp_bh[0] and that the metapath has been initialised
329  * by find_metapath().
330  *
331  * If this function encounters part of the tree which has not been
332  * allocated, it returns the current height of the tree at the point
333  * at which it found the unallocated block. Blocks which are found are
334  * added to the mp->mp_bh[] list.
335  *
336  * Returns: error or height of metadata tree
337  */
338 
339 static int lookup_metapath(struct gfs2_inode *ip, struct metapath *mp)
340 {
341 	unsigned int end_of_metadata = ip->i_height - 1;
342 	unsigned int x;
343 	int ret;
344 
345 	for (x = 0; x < end_of_metadata; x++) {
346 		ret = lookup_mp_height(ip, mp, x);
347 		if (ret)
348 			return ret;
349 	}
350 
351 	return ip->i_height;
352 }
353 
354 /**
355  * fillup_metapath - fill up buffers for the metadata path to a specific height
356  * @ip: The inode
357  * @mp: The metapath
358  * @h: The height to which it should be mapped
359  *
360  * Similar to lookup_metapath, but does lookups for a range of heights
361  *
362  * Returns: error or height of metadata tree
363  */
364 
365 static int fillup_metapath(struct gfs2_inode *ip, struct metapath *mp, int h)
366 {
367 	unsigned int start_h = h - 1;
368 	int ret;
369 
370 	if (h) {
371 		/* find the first buffer we need to look up. */
372 		while (start_h > 0 && mp->mp_bh[start_h] == NULL)
373 			start_h--;
374 		for (; start_h < h; start_h++) {
375 			ret = lookup_mp_height(ip, mp, start_h);
376 			if (ret)
377 				return ret;
378 		}
379 	}
380 	return ip->i_height;
381 }
382 
383 static inline void release_metapath(struct metapath *mp)
384 {
385 	int i;
386 
387 	for (i = 0; i < GFS2_MAX_META_HEIGHT; i++) {
388 		if (mp->mp_bh[i] == NULL)
389 			break;
390 		brelse(mp->mp_bh[i]);
391 	}
392 }
393 
394 /**
395  * gfs2_extent_length - Returns length of an extent of blocks
396  * @start: Start of the buffer
397  * @len: Length of the buffer in bytes
398  * @ptr: Current position in the buffer
399  * @limit: Max extent length to return (0 = unlimited)
400  * @eob: Set to 1 if we hit "end of block"
401  *
402  * If the first block is zero (unallocated) it will return the number of
403  * unallocated blocks in the extent, otherwise it will return the number
404  * of contiguous blocks in the extent.
405  *
406  * Returns: The length of the extent (minimum of one block)
407  */
408 
409 static inline unsigned int gfs2_extent_length(void *start, unsigned int len, __be64 *ptr, size_t limit, int *eob)
410 {
411 	const __be64 *end = (start + len);
412 	const __be64 *first = ptr;
413 	u64 d = be64_to_cpu(*ptr);
414 
415 	*eob = 0;
416 	do {
417 		ptr++;
418 		if (ptr >= end)
419 			break;
420 		if (limit && --limit == 0)
421 			break;
422 		if (d)
423 			d++;
424 	} while(be64_to_cpu(*ptr) == d);
425 	if (ptr >= end)
426 		*eob = 1;
427 	return (ptr - first);
428 }
429 
430 static inline void bmap_lock(struct gfs2_inode *ip, int create)
431 {
432 	if (create)
433 		down_write(&ip->i_rw_mutex);
434 	else
435 		down_read(&ip->i_rw_mutex);
436 }
437 
438 static inline void bmap_unlock(struct gfs2_inode *ip, int create)
439 {
440 	if (create)
441 		up_write(&ip->i_rw_mutex);
442 	else
443 		up_read(&ip->i_rw_mutex);
444 }
445 
446 static inline __be64 *gfs2_indirect_init(struct metapath *mp,
447 					 struct gfs2_glock *gl, unsigned int i,
448 					 unsigned offset, u64 bn)
449 {
450 	__be64 *ptr = (__be64 *)(mp->mp_bh[i - 1]->b_data +
451 		       ((i > 1) ? sizeof(struct gfs2_meta_header) :
452 				 sizeof(struct gfs2_dinode)));
453 	BUG_ON(i < 1);
454 	BUG_ON(mp->mp_bh[i] != NULL);
455 	mp->mp_bh[i] = gfs2_meta_new(gl, bn);
456 	gfs2_trans_add_meta(gl, mp->mp_bh[i]);
457 	gfs2_metatype_set(mp->mp_bh[i], GFS2_METATYPE_IN, GFS2_FORMAT_IN);
458 	gfs2_buffer_clear_tail(mp->mp_bh[i], sizeof(struct gfs2_meta_header));
459 	ptr += offset;
460 	*ptr = cpu_to_be64(bn);
461 	return ptr;
462 }
463 
464 enum alloc_state {
465 	ALLOC_DATA = 0,
466 	ALLOC_GROW_DEPTH = 1,
467 	ALLOC_GROW_HEIGHT = 2,
468 	/* ALLOC_UNSTUFF = 3,   TBD and rather complicated */
469 };
470 
471 static inline unsigned int hptrs(struct gfs2_sbd *sdp, const unsigned int hgt)
472 {
473 	if (hgt)
474 		return sdp->sd_inptrs;
475 	return sdp->sd_diptrs;
476 }
477 
478 /**
479  * gfs2_bmap_alloc - Build a metadata tree of the requested height
480  * @inode: The GFS2 inode
481  * @lblock: The logical starting block of the extent
482  * @bh_map: This is used to return the mapping details
483  * @mp: The metapath
484  * @sheight: The starting height (i.e. whats already mapped)
485  * @height: The height to build to
486  * @maxlen: The max number of data blocks to alloc
487  *
488  * In this routine we may have to alloc:
489  *   i) Indirect blocks to grow the metadata tree height
490  *  ii) Indirect blocks to fill in lower part of the metadata tree
491  * iii) Data blocks
492  *
493  * The function is in two parts. The first part works out the total
494  * number of blocks which we need. The second part does the actual
495  * allocation asking for an extent at a time (if enough contiguous free
496  * blocks are available, there will only be one request per bmap call)
497  * and uses the state machine to initialise the blocks in order.
498  *
499  * Returns: errno on error
500  */
501 
502 static int gfs2_bmap_alloc(struct inode *inode, const sector_t lblock,
503 			   struct buffer_head *bh_map, struct metapath *mp,
504 			   const unsigned int sheight,
505 			   const unsigned int height,
506 			   const size_t maxlen)
507 {
508 	struct gfs2_inode *ip = GFS2_I(inode);
509 	struct gfs2_sbd *sdp = GFS2_SB(inode);
510 	struct super_block *sb = sdp->sd_vfs;
511 	struct buffer_head *dibh = mp->mp_bh[0];
512 	u64 bn, dblock = 0;
513 	unsigned n, i, blks, alloced = 0, iblks = 0, branch_start = 0;
514 	unsigned dblks = 0;
515 	unsigned ptrs_per_blk;
516 	const unsigned end_of_metadata = height - 1;
517 	int ret;
518 	int eob = 0;
519 	enum alloc_state state;
520 	__be64 *ptr;
521 	__be64 zero_bn = 0;
522 
523 	BUG_ON(sheight < 1);
524 	BUG_ON(dibh == NULL);
525 
526 	gfs2_trans_add_meta(ip->i_gl, dibh);
527 
528 	if (height == sheight) {
529 		struct buffer_head *bh;
530 		/* Bottom indirect block exists, find unalloced extent size */
531 		ptr = metapointer(end_of_metadata, mp);
532 		bh = mp->mp_bh[end_of_metadata];
533 		dblks = gfs2_extent_length(bh->b_data, bh->b_size, ptr, maxlen,
534 					   &eob);
535 		BUG_ON(dblks < 1);
536 		state = ALLOC_DATA;
537 	} else {
538 		/* Need to allocate indirect blocks */
539 		ptrs_per_blk = height > 1 ? sdp->sd_inptrs : sdp->sd_diptrs;
540 		dblks = min(maxlen, (size_t)(ptrs_per_blk -
541 					     mp->mp_list[end_of_metadata]));
542 		if (height == ip->i_height) {
543 			/* Writing into existing tree, extend tree down */
544 			iblks = height - sheight;
545 			state = ALLOC_GROW_DEPTH;
546 		} else {
547 			/* Building up tree height */
548 			state = ALLOC_GROW_HEIGHT;
549 			iblks = height - ip->i_height;
550 			branch_start = metapath_branch_start(mp);
551 			iblks += (height - branch_start);
552 		}
553 	}
554 
555 	/* start of the second part of the function (state machine) */
556 
557 	blks = dblks + iblks;
558 	i = sheight;
559 	do {
560 		int error;
561 		n = blks - alloced;
562 		error = gfs2_alloc_blocks(ip, &bn, &n, 0, NULL);
563 		if (error)
564 			return error;
565 		alloced += n;
566 		if (state != ALLOC_DATA || gfs2_is_jdata(ip))
567 			gfs2_trans_add_unrevoke(sdp, bn, n);
568 		switch (state) {
569 		/* Growing height of tree */
570 		case ALLOC_GROW_HEIGHT:
571 			if (i == 1) {
572 				ptr = (__be64 *)(dibh->b_data +
573 						 sizeof(struct gfs2_dinode));
574 				zero_bn = *ptr;
575 			}
576 			for (; i - 1 < height - ip->i_height && n > 0; i++, n--)
577 				gfs2_indirect_init(mp, ip->i_gl, i, 0, bn++);
578 			if (i - 1 == height - ip->i_height) {
579 				i--;
580 				gfs2_buffer_copy_tail(mp->mp_bh[i],
581 						sizeof(struct gfs2_meta_header),
582 						dibh, sizeof(struct gfs2_dinode));
583 				gfs2_buffer_clear_tail(dibh,
584 						sizeof(struct gfs2_dinode) +
585 						sizeof(__be64));
586 				ptr = (__be64 *)(mp->mp_bh[i]->b_data +
587 					sizeof(struct gfs2_meta_header));
588 				*ptr = zero_bn;
589 				state = ALLOC_GROW_DEPTH;
590 				for(i = branch_start; i < height; i++) {
591 					if (mp->mp_bh[i] == NULL)
592 						break;
593 					brelse(mp->mp_bh[i]);
594 					mp->mp_bh[i] = NULL;
595 				}
596 				i = branch_start;
597 			}
598 			if (n == 0)
599 				break;
600 		/* Branching from existing tree */
601 		case ALLOC_GROW_DEPTH:
602 			if (i > 1 && i < height)
603 				gfs2_trans_add_meta(ip->i_gl, mp->mp_bh[i-1]);
604 			for (; i < height && n > 0; i++, n--)
605 				gfs2_indirect_init(mp, ip->i_gl, i,
606 						   mp->mp_list[i-1], bn++);
607 			if (i == height)
608 				state = ALLOC_DATA;
609 			if (n == 0)
610 				break;
611 		/* Tree complete, adding data blocks */
612 		case ALLOC_DATA:
613 			BUG_ON(n > dblks);
614 			BUG_ON(mp->mp_bh[end_of_metadata] == NULL);
615 			gfs2_trans_add_meta(ip->i_gl, mp->mp_bh[end_of_metadata]);
616 			dblks = n;
617 			ptr = metapointer(end_of_metadata, mp);
618 			dblock = bn;
619 			while (n-- > 0)
620 				*ptr++ = cpu_to_be64(bn++);
621 			if (buffer_zeronew(bh_map)) {
622 				ret = sb_issue_zeroout(sb, dblock, dblks,
623 						       GFP_NOFS);
624 				if (ret) {
625 					fs_err(sdp,
626 					       "Failed to zero data buffers\n");
627 					clear_buffer_zeronew(bh_map);
628 				}
629 			}
630 			break;
631 		}
632 	} while ((state != ALLOC_DATA) || !dblock);
633 
634 	ip->i_height = height;
635 	gfs2_add_inode_blocks(&ip->i_inode, alloced);
636 	gfs2_dinode_out(ip, mp->mp_bh[0]->b_data);
637 	map_bh(bh_map, inode->i_sb, dblock);
638 	bh_map->b_size = dblks << inode->i_blkbits;
639 	set_buffer_new(bh_map);
640 	return 0;
641 }
642 
643 /**
644  * gfs2_block_map - Map a block from an inode to a disk block
645  * @inode: The inode
646  * @lblock: The logical block number
647  * @bh_map: The bh to be mapped
648  * @create: True if its ok to alloc blocks to satify the request
649  *
650  * Sets buffer_mapped() if successful, sets buffer_boundary() if a
651  * read of metadata will be required before the next block can be
652  * mapped. Sets buffer_new() if new blocks were allocated.
653  *
654  * Returns: errno
655  */
656 
657 int gfs2_block_map(struct inode *inode, sector_t lblock,
658 		   struct buffer_head *bh_map, int create)
659 {
660 	struct gfs2_inode *ip = GFS2_I(inode);
661 	struct gfs2_sbd *sdp = GFS2_SB(inode);
662 	unsigned int bsize = sdp->sd_sb.sb_bsize;
663 	const size_t maxlen = bh_map->b_size >> inode->i_blkbits;
664 	const u64 *arr = sdp->sd_heightsize;
665 	__be64 *ptr;
666 	u64 size;
667 	struct metapath mp;
668 	int ret;
669 	int eob;
670 	unsigned int len;
671 	struct buffer_head *bh;
672 	u8 height;
673 
674 	BUG_ON(maxlen == 0);
675 
676 	memset(&mp, 0, sizeof(mp));
677 	bmap_lock(ip, create);
678 	clear_buffer_mapped(bh_map);
679 	clear_buffer_new(bh_map);
680 	clear_buffer_boundary(bh_map);
681 	trace_gfs2_bmap(ip, bh_map, lblock, create, 1);
682 	if (gfs2_is_dir(ip)) {
683 		bsize = sdp->sd_jbsize;
684 		arr = sdp->sd_jheightsize;
685 	}
686 
687 	ret = gfs2_meta_inode_buffer(ip, &mp.mp_bh[0]);
688 	if (ret)
689 		goto out;
690 
691 	height = ip->i_height;
692 	size = (lblock + 1) * bsize;
693 	while (size > arr[height])
694 		height++;
695 	find_metapath(sdp, lblock, &mp, height);
696 	ret = 1;
697 	if (height > ip->i_height || gfs2_is_stuffed(ip))
698 		goto do_alloc;
699 	ret = lookup_metapath(ip, &mp);
700 	if (ret < 0)
701 		goto out;
702 	if (ret != ip->i_height)
703 		goto do_alloc;
704 	ptr = metapointer(ip->i_height - 1, &mp);
705 	if (*ptr == 0)
706 		goto do_alloc;
707 	map_bh(bh_map, inode->i_sb, be64_to_cpu(*ptr));
708 	bh = mp.mp_bh[ip->i_height - 1];
709 	len = gfs2_extent_length(bh->b_data, bh->b_size, ptr, maxlen, &eob);
710 	bh_map->b_size = (len << inode->i_blkbits);
711 	if (eob)
712 		set_buffer_boundary(bh_map);
713 	ret = 0;
714 out:
715 	release_metapath(&mp);
716 	trace_gfs2_bmap(ip, bh_map, lblock, create, ret);
717 	bmap_unlock(ip, create);
718 	return ret;
719 
720 do_alloc:
721 	/* All allocations are done here, firstly check create flag */
722 	if (!create) {
723 		BUG_ON(gfs2_is_stuffed(ip));
724 		ret = 0;
725 		goto out;
726 	}
727 
728 	/* At this point ret is the tree depth of already allocated blocks */
729 	ret = gfs2_bmap_alloc(inode, lblock, bh_map, &mp, ret, height, maxlen);
730 	goto out;
731 }
732 
733 /*
734  * Deprecated: do not use in new code
735  */
736 int gfs2_extent_map(struct inode *inode, u64 lblock, int *new, u64 *dblock, unsigned *extlen)
737 {
738 	struct buffer_head bh = { .b_state = 0, .b_blocknr = 0 };
739 	int ret;
740 	int create = *new;
741 
742 	BUG_ON(!extlen);
743 	BUG_ON(!dblock);
744 	BUG_ON(!new);
745 
746 	bh.b_size = BIT(inode->i_blkbits + (create ? 0 : 5));
747 	ret = gfs2_block_map(inode, lblock, &bh, create);
748 	*extlen = bh.b_size >> inode->i_blkbits;
749 	*dblock = bh.b_blocknr;
750 	if (buffer_new(&bh))
751 		*new = 1;
752 	else
753 		*new = 0;
754 	return ret;
755 }
756 
757 /**
758  * gfs2_block_truncate_page - Deal with zeroing out data for truncate
759  *
760  * This is partly borrowed from ext3.
761  */
762 static int gfs2_block_truncate_page(struct address_space *mapping, loff_t from)
763 {
764 	struct inode *inode = mapping->host;
765 	struct gfs2_inode *ip = GFS2_I(inode);
766 	unsigned long index = from >> PAGE_SHIFT;
767 	unsigned offset = from & (PAGE_SIZE-1);
768 	unsigned blocksize, iblock, length, pos;
769 	struct buffer_head *bh;
770 	struct page *page;
771 	int err;
772 
773 	page = find_or_create_page(mapping, index, GFP_NOFS);
774 	if (!page)
775 		return 0;
776 
777 	blocksize = inode->i_sb->s_blocksize;
778 	length = blocksize - (offset & (blocksize - 1));
779 	iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
780 
781 	if (!page_has_buffers(page))
782 		create_empty_buffers(page, blocksize, 0);
783 
784 	/* Find the buffer that contains "offset" */
785 	bh = page_buffers(page);
786 	pos = blocksize;
787 	while (offset >= pos) {
788 		bh = bh->b_this_page;
789 		iblock++;
790 		pos += blocksize;
791 	}
792 
793 	err = 0;
794 
795 	if (!buffer_mapped(bh)) {
796 		gfs2_block_map(inode, iblock, bh, 0);
797 		/* unmapped? It's a hole - nothing to do */
798 		if (!buffer_mapped(bh))
799 			goto unlock;
800 	}
801 
802 	/* Ok, it's mapped. Make sure it's up-to-date */
803 	if (PageUptodate(page))
804 		set_buffer_uptodate(bh);
805 
806 	if (!buffer_uptodate(bh)) {
807 		err = -EIO;
808 		ll_rw_block(REQ_OP_READ, 0, 1, &bh);
809 		wait_on_buffer(bh);
810 		/* Uhhuh. Read error. Complain and punt. */
811 		if (!buffer_uptodate(bh))
812 			goto unlock;
813 		err = 0;
814 	}
815 
816 	if (!gfs2_is_writeback(ip))
817 		gfs2_trans_add_data(ip->i_gl, bh);
818 
819 	zero_user(page, offset, length);
820 	mark_buffer_dirty(bh);
821 unlock:
822 	unlock_page(page);
823 	put_page(page);
824 	return err;
825 }
826 
827 #define GFS2_JTRUNC_REVOKES 8192
828 
829 /**
830  * gfs2_journaled_truncate - Wrapper for truncate_pagecache for jdata files
831  * @inode: The inode being truncated
832  * @oldsize: The original (larger) size
833  * @newsize: The new smaller size
834  *
835  * With jdata files, we have to journal a revoke for each block which is
836  * truncated. As a result, we need to split this into separate transactions
837  * if the number of pages being truncated gets too large.
838  */
839 
840 static int gfs2_journaled_truncate(struct inode *inode, u64 oldsize, u64 newsize)
841 {
842 	struct gfs2_sbd *sdp = GFS2_SB(inode);
843 	u64 max_chunk = GFS2_JTRUNC_REVOKES * sdp->sd_vfs->s_blocksize;
844 	u64 chunk;
845 	int error;
846 
847 	while (oldsize != newsize) {
848 		chunk = oldsize - newsize;
849 		if (chunk > max_chunk)
850 			chunk = max_chunk;
851 		truncate_pagecache(inode, oldsize - chunk);
852 		oldsize -= chunk;
853 		gfs2_trans_end(sdp);
854 		error = gfs2_trans_begin(sdp, RES_DINODE, GFS2_JTRUNC_REVOKES);
855 		if (error)
856 			return error;
857 	}
858 
859 	return 0;
860 }
861 
862 static int trunc_start(struct inode *inode, u64 oldsize, u64 newsize)
863 {
864 	struct gfs2_inode *ip = GFS2_I(inode);
865 	struct gfs2_sbd *sdp = GFS2_SB(inode);
866 	struct address_space *mapping = inode->i_mapping;
867 	struct buffer_head *dibh;
868 	int journaled = gfs2_is_jdata(ip);
869 	int error;
870 
871 	if (journaled)
872 		error = gfs2_trans_begin(sdp, RES_DINODE + RES_JDATA, GFS2_JTRUNC_REVOKES);
873 	else
874 		error = gfs2_trans_begin(sdp, RES_DINODE, 0);
875 	if (error)
876 		return error;
877 
878 	error = gfs2_meta_inode_buffer(ip, &dibh);
879 	if (error)
880 		goto out;
881 
882 	gfs2_trans_add_meta(ip->i_gl, dibh);
883 
884 	if (gfs2_is_stuffed(ip)) {
885 		gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode) + newsize);
886 	} else {
887 		if (newsize & (u64)(sdp->sd_sb.sb_bsize - 1)) {
888 			error = gfs2_block_truncate_page(mapping, newsize);
889 			if (error)
890 				goto out_brelse;
891 		}
892 		ip->i_diskflags |= GFS2_DIF_TRUNC_IN_PROG;
893 	}
894 
895 	i_size_write(inode, newsize);
896 	ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode);
897 	gfs2_dinode_out(ip, dibh->b_data);
898 
899 	if (journaled)
900 		error = gfs2_journaled_truncate(inode, oldsize, newsize);
901 	else
902 		truncate_pagecache(inode, newsize);
903 
904 	if (error) {
905 		brelse(dibh);
906 		return error;
907 	}
908 
909 out_brelse:
910 	brelse(dibh);
911 out:
912 	gfs2_trans_end(sdp);
913 	return error;
914 }
915 
916 /**
917  * sweep_bh_for_rgrps - find an rgrp in a meta buffer and free blocks therein
918  * @ip: inode
919  * @rg_gh: holder of resource group glock
920  * @mp: current metapath fully populated with buffers
921  * @btotal: place to keep count of total blocks freed
922  * @hgt: height we're processing
923  * @first: true if this is the first call to this function for this height
924  *
925  * We sweep a metadata buffer (provided by the metapath) for blocks we need to
926  * free, and free them all. However, we do it one rgrp at a time. If this
927  * block has references to multiple rgrps, we break it into individual
928  * transactions. This allows other processes to use the rgrps while we're
929  * focused on a single one, for better concurrency / performance.
930  * At every transaction boundary, we rewrite the inode into the journal.
931  * That way the bitmaps are kept consistent with the inode and we can recover
932  * if we're interrupted by power-outages.
933  *
934  * Returns: 0, or return code if an error occurred.
935  *          *btotal has the total number of blocks freed
936  */
937 static int sweep_bh_for_rgrps(struct gfs2_inode *ip, struct gfs2_holder *rd_gh,
938 			      const struct metapath *mp, u32 *btotal, int hgt,
939 			      bool preserve1)
940 {
941 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
942 	struct gfs2_rgrpd *rgd;
943 	struct gfs2_trans *tr;
944 	struct buffer_head *bh = mp->mp_bh[hgt];
945 	__be64 *top, *bottom, *p;
946 	int blks_outside_rgrp;
947 	u64 bn, bstart, isize_blks;
948 	s64 blen; /* needs to be s64 or gfs2_add_inode_blocks breaks */
949 	int meta = ((hgt != ip->i_height - 1) ? 1 : 0);
950 	int ret = 0;
951 	bool buf_in_tr = false; /* buffer was added to transaction */
952 
953 	if (gfs2_metatype_check(sdp, bh,
954 				(hgt ? GFS2_METATYPE_IN : GFS2_METATYPE_DI)))
955 		return -EIO;
956 
957 more_rgrps:
958 	blks_outside_rgrp = 0;
959 	bstart = 0;
960 	blen = 0;
961 	top = metapointer(hgt, mp); /* first ptr from metapath */
962 	/* If we're keeping some data at the truncation point, we've got to
963 	   preserve the metadata tree by adding 1 to the starting metapath. */
964 	if (preserve1)
965 		top++;
966 
967 	bottom = (__be64 *)(bh->b_data + bh->b_size);
968 
969 	for (p = top; p < bottom; p++) {
970 		if (!*p)
971 			continue;
972 		bn = be64_to_cpu(*p);
973 		if (gfs2_holder_initialized(rd_gh)) {
974 			rgd = gfs2_glock2rgrp(rd_gh->gh_gl);
975 			gfs2_assert_withdraw(sdp,
976 				     gfs2_glock_is_locked_by_me(rd_gh->gh_gl));
977 		} else {
978 			rgd = gfs2_blk2rgrpd(sdp, bn, false);
979 			ret = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE,
980 						 0, rd_gh);
981 			if (ret)
982 				goto out;
983 
984 			/* Must be done with the rgrp glock held: */
985 			if (gfs2_rs_active(&ip->i_res) &&
986 			    rgd == ip->i_res.rs_rbm.rgd)
987 				gfs2_rs_deltree(&ip->i_res);
988 		}
989 
990 		if (!rgrp_contains_block(rgd, bn)) {
991 			blks_outside_rgrp++;
992 			continue;
993 		}
994 
995 		/* The size of our transactions will be unknown until we
996 		   actually process all the metadata blocks that relate to
997 		   the rgrp. So we estimate. We know it can't be more than
998 		   the dinode's i_blocks and we don't want to exceed the
999 		   journal flush threshold, sd_log_thresh2. */
1000 		if (current->journal_info == NULL) {
1001 			unsigned int jblocks_rqsted, revokes;
1002 
1003 			jblocks_rqsted = rgd->rd_length + RES_DINODE +
1004 				RES_INDIRECT;
1005 			isize_blks = gfs2_get_inode_blocks(&ip->i_inode);
1006 			if (isize_blks > atomic_read(&sdp->sd_log_thresh2))
1007 				jblocks_rqsted +=
1008 					atomic_read(&sdp->sd_log_thresh2);
1009 			else
1010 				jblocks_rqsted += isize_blks;
1011 			revokes = jblocks_rqsted;
1012 			if (meta)
1013 				revokes += hptrs(sdp, hgt);
1014 			else if (ip->i_depth)
1015 				revokes += sdp->sd_inptrs;
1016 			ret = gfs2_trans_begin(sdp, jblocks_rqsted, revokes);
1017 			if (ret)
1018 				goto out_unlock;
1019 			down_write(&ip->i_rw_mutex);
1020 		}
1021 		/* check if we will exceed the transaction blocks requested */
1022 		tr = current->journal_info;
1023 		if (tr->tr_num_buf_new + RES_STATFS +
1024 		    RES_QUOTA >= atomic_read(&sdp->sd_log_thresh2)) {
1025 			/* We set blks_outside_rgrp to ensure the loop will
1026 			   be repeated for the same rgrp, but with a new
1027 			   transaction. */
1028 			blks_outside_rgrp++;
1029 			/* This next part is tricky. If the buffer was added
1030 			   to the transaction, we've already set some block
1031 			   pointers to 0, so we better follow through and free
1032 			   them, or we will introduce corruption (so break).
1033 			   This may be impossible, or at least rare, but I
1034 			   decided to cover the case regardless.
1035 
1036 			   If the buffer was not added to the transaction
1037 			   (this call), doing so would exceed our transaction
1038 			   size, so we need to end the transaction and start a
1039 			   new one (so goto). */
1040 
1041 			if (buf_in_tr)
1042 				break;
1043 			goto out_unlock;
1044 		}
1045 
1046 		gfs2_trans_add_meta(ip->i_gl, bh);
1047 		buf_in_tr = true;
1048 		*p = 0;
1049 		if (bstart + blen == bn) {
1050 			blen++;
1051 			continue;
1052 		}
1053 		if (bstart) {
1054 			__gfs2_free_blocks(ip, bstart, (u32)blen, meta);
1055 			(*btotal) += blen;
1056 			gfs2_add_inode_blocks(&ip->i_inode, -blen);
1057 		}
1058 		bstart = bn;
1059 		blen = 1;
1060 	}
1061 	if (bstart) {
1062 		__gfs2_free_blocks(ip, bstart, (u32)blen, meta);
1063 		(*btotal) += blen;
1064 		gfs2_add_inode_blocks(&ip->i_inode, -blen);
1065 	}
1066 out_unlock:
1067 	if (!ret && blks_outside_rgrp) { /* If buffer still has non-zero blocks
1068 					    outside the rgrp we just processed,
1069 					    do it all over again. */
1070 		if (current->journal_info) {
1071 			struct buffer_head *dibh = mp->mp_bh[0];
1072 
1073 			/* Every transaction boundary, we rewrite the dinode
1074 			   to keep its di_blocks current in case of failure. */
1075 			ip->i_inode.i_mtime = ip->i_inode.i_ctime =
1076 				current_time(&ip->i_inode);
1077 			gfs2_trans_add_meta(ip->i_gl, dibh);
1078 			gfs2_dinode_out(ip, dibh->b_data);
1079 			up_write(&ip->i_rw_mutex);
1080 			gfs2_trans_end(sdp);
1081 		}
1082 		gfs2_glock_dq_uninit(rd_gh);
1083 		cond_resched();
1084 		goto more_rgrps;
1085 	}
1086 out:
1087 	return ret;
1088 }
1089 
1090 /**
1091  * find_nonnull_ptr - find a non-null pointer given a metapath and height
1092  * assumes the metapath is valid (with buffers) out to height h
1093  * @mp: starting metapath
1094  * @h: desired height to search
1095  *
1096  * Returns: true if a non-null pointer was found in the metapath buffer
1097  *          false if all remaining pointers are NULL in the buffer
1098  */
1099 static bool find_nonnull_ptr(struct gfs2_sbd *sdp, struct metapath *mp,
1100 			     unsigned int h)
1101 {
1102 	__be64 *ptr;
1103 	unsigned int ptrs = hptrs(sdp, h) - 1;
1104 
1105 	while (true) {
1106 		ptr = metapointer(h, mp);
1107 		if (*ptr) { /* if we have a non-null pointer */
1108 			/* Now zero the metapath after the current height. */
1109 			h++;
1110 			if (h < GFS2_MAX_META_HEIGHT)
1111 				memset(&mp->mp_list[h], 0,
1112 				       (GFS2_MAX_META_HEIGHT - h) *
1113 				       sizeof(mp->mp_list[0]));
1114 			return true;
1115 		}
1116 
1117 		if (mp->mp_list[h] < ptrs)
1118 			mp->mp_list[h]++;
1119 		else
1120 			return false; /* no more pointers in this buffer */
1121 	}
1122 }
1123 
1124 enum dealloc_states {
1125 	DEALLOC_MP_FULL = 0,    /* Strip a metapath with all buffers read in */
1126 	DEALLOC_MP_LOWER = 1,   /* lower the metapath strip height */
1127 	DEALLOC_FILL_MP = 2,  /* Fill in the metapath to the given height. */
1128 	DEALLOC_DONE = 3,       /* process complete */
1129 };
1130 
1131 static bool mp_eq_to_hgt(struct metapath *mp, __u16 *nbof, unsigned int h)
1132 {
1133 	if (memcmp(mp->mp_list, nbof, h * sizeof(mp->mp_list[0])))
1134 		return false;
1135 	return true;
1136 }
1137 
1138 /**
1139  * trunc_dealloc - truncate a file down to a desired size
1140  * @ip: inode to truncate
1141  * @newsize: The desired size of the file
1142  *
1143  * This function truncates a file to newsize. It works from the
1144  * bottom up, and from the right to the left. In other words, it strips off
1145  * the highest layer (data) before stripping any of the metadata. Doing it
1146  * this way is best in case the operation is interrupted by power failure, etc.
1147  * The dinode is rewritten in every transaction to guarantee integrity.
1148  */
1149 static int trunc_dealloc(struct gfs2_inode *ip, u64 newsize)
1150 {
1151 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1152 	struct metapath mp;
1153 	struct buffer_head *dibh, *bh;
1154 	struct gfs2_holder rd_gh;
1155 	u64 lblock;
1156 	__u16 nbof[GFS2_MAX_META_HEIGHT]; /* new beginning of truncation */
1157 	unsigned int strip_h = ip->i_height - 1;
1158 	u32 btotal = 0;
1159 	int ret, state;
1160 	int mp_h; /* metapath buffers are read in to this height */
1161 	sector_t last_ra = 0;
1162 	u64 prev_bnr = 0;
1163 	bool preserve1; /* need to preserve the first meta pointer? */
1164 
1165 	if (!newsize)
1166 		lblock = 0;
1167 	else
1168 		lblock = (newsize - 1) >> sdp->sd_sb.sb_bsize_shift;
1169 
1170 	memset(&mp, 0, sizeof(mp));
1171 	find_metapath(sdp, lblock, &mp, ip->i_height);
1172 
1173 	memcpy(&nbof, &mp.mp_list, sizeof(nbof));
1174 
1175 	ret = gfs2_meta_inode_buffer(ip, &dibh);
1176 	if (ret)
1177 		return ret;
1178 
1179 	mp.mp_bh[0] = dibh;
1180 	ret = lookup_metapath(ip, &mp);
1181 	if (ret == ip->i_height)
1182 		state = DEALLOC_MP_FULL; /* We have a complete metapath */
1183 	else
1184 		state = DEALLOC_FILL_MP; /* deal with partial metapath */
1185 
1186 	ret = gfs2_rindex_update(sdp);
1187 	if (ret)
1188 		goto out_metapath;
1189 
1190 	ret = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1191 	if (ret)
1192 		goto out_metapath;
1193 	gfs2_holder_mark_uninitialized(&rd_gh);
1194 
1195 	mp_h = strip_h;
1196 
1197 	while (state != DEALLOC_DONE) {
1198 		switch (state) {
1199 		/* Truncate a full metapath at the given strip height.
1200 		 * Note that strip_h == mp_h in order to be in this state. */
1201 		case DEALLOC_MP_FULL:
1202 			if (mp_h > 0) { /* issue read-ahead on metadata */
1203 				__be64 *top;
1204 
1205 				bh = mp.mp_bh[mp_h - 1];
1206 				if (bh->b_blocknr != last_ra) {
1207 					last_ra = bh->b_blocknr;
1208 					top = metaptr1(mp_h - 1, &mp);
1209 					gfs2_metapath_ra(ip->i_gl, bh, top);
1210 				}
1211 			}
1212 			/* If we're truncating to a non-zero size and the mp is
1213 			   at the beginning of file for the strip height, we
1214 			   need to preserve the first metadata pointer. */
1215 			preserve1 = (newsize && mp_eq_to_hgt(&mp, nbof, mp_h));
1216 			bh = mp.mp_bh[mp_h];
1217 			gfs2_assert_withdraw(sdp, bh);
1218 			if (gfs2_assert_withdraw(sdp,
1219 						 prev_bnr != bh->b_blocknr)) {
1220 				printk(KERN_EMERG "GFS2: fsid=%s:inode %llu, "
1221 				       "block:%llu, i_h:%u, s_h:%u, mp_h:%u\n",
1222 				       sdp->sd_fsname,
1223 				       (unsigned long long)ip->i_no_addr,
1224 				       prev_bnr, ip->i_height, strip_h, mp_h);
1225 			}
1226 			prev_bnr = bh->b_blocknr;
1227 			ret = sweep_bh_for_rgrps(ip, &rd_gh, &mp, &btotal,
1228 						 mp_h, preserve1);
1229 			/* If we hit an error or just swept dinode buffer,
1230 			   just exit. */
1231 			if (ret || !mp_h) {
1232 				state = DEALLOC_DONE;
1233 				break;
1234 			}
1235 			state = DEALLOC_MP_LOWER;
1236 			break;
1237 
1238 		/* lower the metapath strip height */
1239 		case DEALLOC_MP_LOWER:
1240 			/* We're done with the current buffer, so release it,
1241 			   unless it's the dinode buffer. Then back up to the
1242 			   previous pointer. */
1243 			if (mp_h) {
1244 				brelse(mp.mp_bh[mp_h]);
1245 				mp.mp_bh[mp_h] = NULL;
1246 			}
1247 			/* If we can't get any lower in height, we've stripped
1248 			   off all we can. Next step is to back up and start
1249 			   stripping the previous level of metadata. */
1250 			if (mp_h == 0) {
1251 				strip_h--;
1252 				memcpy(&mp.mp_list, &nbof, sizeof(nbof));
1253 				mp_h = strip_h;
1254 				state = DEALLOC_FILL_MP;
1255 				break;
1256 			}
1257 			mp.mp_list[mp_h] = 0;
1258 			mp_h--; /* search one metadata height down */
1259 			if (mp.mp_list[mp_h] >= hptrs(sdp, mp_h) - 1)
1260 				break; /* loop around in the same state */
1261 			mp.mp_list[mp_h]++;
1262 			/* Here we've found a part of the metapath that is not
1263 			 * allocated. We need to search at that height for the
1264 			 * next non-null pointer. */
1265 			if (find_nonnull_ptr(sdp, &mp, mp_h)) {
1266 				state = DEALLOC_FILL_MP;
1267 				mp_h++;
1268 			}
1269 			/* No more non-null pointers at this height. Back up
1270 			   to the previous height and try again. */
1271 			break; /* loop around in the same state */
1272 
1273 		/* Fill the metapath with buffers to the given height. */
1274 		case DEALLOC_FILL_MP:
1275 			/* Fill the buffers out to the current height. */
1276 			ret = fillup_metapath(ip, &mp, mp_h);
1277 			if (ret < 0)
1278 				goto out;
1279 
1280 			/* If buffers found for the entire strip height */
1281 			if ((ret == ip->i_height) && (mp_h == strip_h)) {
1282 				state = DEALLOC_MP_FULL;
1283 				break;
1284 			}
1285 			if (ret < ip->i_height) /* We have a partial height */
1286 				mp_h = ret - 1;
1287 
1288 			/* If we find a non-null block pointer, crawl a bit
1289 			   higher up in the metapath and try again, otherwise
1290 			   we need to look lower for a new starting point. */
1291 			if (find_nonnull_ptr(sdp, &mp, mp_h))
1292 				mp_h++;
1293 			else
1294 				state = DEALLOC_MP_LOWER;
1295 			break;
1296 		}
1297 	}
1298 
1299 	if (btotal) {
1300 		if (current->journal_info == NULL) {
1301 			ret = gfs2_trans_begin(sdp, RES_DINODE + RES_STATFS +
1302 					       RES_QUOTA, 0);
1303 			if (ret)
1304 				goto out;
1305 			down_write(&ip->i_rw_mutex);
1306 		}
1307 		gfs2_statfs_change(sdp, 0, +btotal, 0);
1308 		gfs2_quota_change(ip, -(s64)btotal, ip->i_inode.i_uid,
1309 				  ip->i_inode.i_gid);
1310 		ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode);
1311 		gfs2_trans_add_meta(ip->i_gl, dibh);
1312 		gfs2_dinode_out(ip, dibh->b_data);
1313 		up_write(&ip->i_rw_mutex);
1314 		gfs2_trans_end(sdp);
1315 	}
1316 
1317 out:
1318 	if (gfs2_holder_initialized(&rd_gh))
1319 		gfs2_glock_dq_uninit(&rd_gh);
1320 	if (current->journal_info) {
1321 		up_write(&ip->i_rw_mutex);
1322 		gfs2_trans_end(sdp);
1323 		cond_resched();
1324 	}
1325 	gfs2_quota_unhold(ip);
1326 out_metapath:
1327 	release_metapath(&mp);
1328 	return ret;
1329 }
1330 
1331 static int trunc_end(struct gfs2_inode *ip)
1332 {
1333 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1334 	struct buffer_head *dibh;
1335 	int error;
1336 
1337 	error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1338 	if (error)
1339 		return error;
1340 
1341 	down_write(&ip->i_rw_mutex);
1342 
1343 	error = gfs2_meta_inode_buffer(ip, &dibh);
1344 	if (error)
1345 		goto out;
1346 
1347 	if (!i_size_read(&ip->i_inode)) {
1348 		ip->i_height = 0;
1349 		ip->i_goal = ip->i_no_addr;
1350 		gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
1351 		gfs2_ordered_del_inode(ip);
1352 	}
1353 	ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode);
1354 	ip->i_diskflags &= ~GFS2_DIF_TRUNC_IN_PROG;
1355 
1356 	gfs2_trans_add_meta(ip->i_gl, dibh);
1357 	gfs2_dinode_out(ip, dibh->b_data);
1358 	brelse(dibh);
1359 
1360 out:
1361 	up_write(&ip->i_rw_mutex);
1362 	gfs2_trans_end(sdp);
1363 	return error;
1364 }
1365 
1366 /**
1367  * do_shrink - make a file smaller
1368  * @inode: the inode
1369  * @oldsize: the current inode size
1370  * @newsize: the size to make the file
1371  *
1372  * Called with an exclusive lock on @inode. The @size must
1373  * be equal to or smaller than the current inode size.
1374  *
1375  * Returns: errno
1376  */
1377 
1378 static int do_shrink(struct inode *inode, u64 oldsize, u64 newsize)
1379 {
1380 	struct gfs2_inode *ip = GFS2_I(inode);
1381 	int error;
1382 
1383 	error = trunc_start(inode, oldsize, newsize);
1384 	if (error < 0)
1385 		return error;
1386 	if (gfs2_is_stuffed(ip))
1387 		return 0;
1388 
1389 	error = trunc_dealloc(ip, newsize);
1390 	if (error == 0)
1391 		error = trunc_end(ip);
1392 
1393 	return error;
1394 }
1395 
1396 void gfs2_trim_blocks(struct inode *inode)
1397 {
1398 	u64 size = inode->i_size;
1399 	int ret;
1400 
1401 	ret = do_shrink(inode, size, size);
1402 	WARN_ON(ret != 0);
1403 }
1404 
1405 /**
1406  * do_grow - Touch and update inode size
1407  * @inode: The inode
1408  * @size: The new size
1409  *
1410  * This function updates the timestamps on the inode and
1411  * may also increase the size of the inode. This function
1412  * must not be called with @size any smaller than the current
1413  * inode size.
1414  *
1415  * Although it is not strictly required to unstuff files here,
1416  * earlier versions of GFS2 have a bug in the stuffed file reading
1417  * code which will result in a buffer overrun if the size is larger
1418  * than the max stuffed file size. In order to prevent this from
1419  * occurring, such files are unstuffed, but in other cases we can
1420  * just update the inode size directly.
1421  *
1422  * Returns: 0 on success, or -ve on error
1423  */
1424 
1425 static int do_grow(struct inode *inode, u64 size)
1426 {
1427 	struct gfs2_inode *ip = GFS2_I(inode);
1428 	struct gfs2_sbd *sdp = GFS2_SB(inode);
1429 	struct gfs2_alloc_parms ap = { .target = 1, };
1430 	struct buffer_head *dibh;
1431 	int error;
1432 	int unstuff = 0;
1433 
1434 	if (gfs2_is_stuffed(ip) &&
1435 	    (size > (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)))) {
1436 		error = gfs2_quota_lock_check(ip, &ap);
1437 		if (error)
1438 			return error;
1439 
1440 		error = gfs2_inplace_reserve(ip, &ap);
1441 		if (error)
1442 			goto do_grow_qunlock;
1443 		unstuff = 1;
1444 	}
1445 
1446 	error = gfs2_trans_begin(sdp, RES_DINODE + RES_STATFS + RES_RG_BIT +
1447 				 (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF ?
1448 				  0 : RES_QUOTA), 0);
1449 	if (error)
1450 		goto do_grow_release;
1451 
1452 	if (unstuff) {
1453 		error = gfs2_unstuff_dinode(ip, NULL);
1454 		if (error)
1455 			goto do_end_trans;
1456 	}
1457 
1458 	error = gfs2_meta_inode_buffer(ip, &dibh);
1459 	if (error)
1460 		goto do_end_trans;
1461 
1462 	i_size_write(inode, size);
1463 	ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode);
1464 	gfs2_trans_add_meta(ip->i_gl, dibh);
1465 	gfs2_dinode_out(ip, dibh->b_data);
1466 	brelse(dibh);
1467 
1468 do_end_trans:
1469 	gfs2_trans_end(sdp);
1470 do_grow_release:
1471 	if (unstuff) {
1472 		gfs2_inplace_release(ip);
1473 do_grow_qunlock:
1474 		gfs2_quota_unlock(ip);
1475 	}
1476 	return error;
1477 }
1478 
1479 /**
1480  * gfs2_setattr_size - make a file a given size
1481  * @inode: the inode
1482  * @newsize: the size to make the file
1483  *
1484  * The file size can grow, shrink, or stay the same size. This
1485  * is called holding i_mutex and an exclusive glock on the inode
1486  * in question.
1487  *
1488  * Returns: errno
1489  */
1490 
1491 int gfs2_setattr_size(struct inode *inode, u64 newsize)
1492 {
1493 	struct gfs2_inode *ip = GFS2_I(inode);
1494 	int ret;
1495 	u64 oldsize;
1496 
1497 	BUG_ON(!S_ISREG(inode->i_mode));
1498 
1499 	ret = inode_newsize_ok(inode, newsize);
1500 	if (ret)
1501 		return ret;
1502 
1503 	inode_dio_wait(inode);
1504 
1505 	ret = gfs2_rsqa_alloc(ip);
1506 	if (ret)
1507 		goto out;
1508 
1509 	oldsize = inode->i_size;
1510 	if (newsize >= oldsize) {
1511 		ret = do_grow(inode, newsize);
1512 		goto out;
1513 	}
1514 
1515 	ret = do_shrink(inode, oldsize, newsize);
1516 out:
1517 	gfs2_rsqa_delete(ip, NULL);
1518 	return ret;
1519 }
1520 
1521 int gfs2_truncatei_resume(struct gfs2_inode *ip)
1522 {
1523 	int error;
1524 	error = trunc_dealloc(ip, i_size_read(&ip->i_inode));
1525 	if (!error)
1526 		error = trunc_end(ip);
1527 	return error;
1528 }
1529 
1530 int gfs2_file_dealloc(struct gfs2_inode *ip)
1531 {
1532 	return trunc_dealloc(ip, 0);
1533 }
1534 
1535 /**
1536  * gfs2_free_journal_extents - Free cached journal bmap info
1537  * @jd: The journal
1538  *
1539  */
1540 
1541 void gfs2_free_journal_extents(struct gfs2_jdesc *jd)
1542 {
1543 	struct gfs2_journal_extent *jext;
1544 
1545 	while(!list_empty(&jd->extent_list)) {
1546 		jext = list_entry(jd->extent_list.next, struct gfs2_journal_extent, list);
1547 		list_del(&jext->list);
1548 		kfree(jext);
1549 	}
1550 }
1551 
1552 /**
1553  * gfs2_add_jextent - Add or merge a new extent to extent cache
1554  * @jd: The journal descriptor
1555  * @lblock: The logical block at start of new extent
1556  * @dblock: The physical block at start of new extent
1557  * @blocks: Size of extent in fs blocks
1558  *
1559  * Returns: 0 on success or -ENOMEM
1560  */
1561 
1562 static int gfs2_add_jextent(struct gfs2_jdesc *jd, u64 lblock, u64 dblock, u64 blocks)
1563 {
1564 	struct gfs2_journal_extent *jext;
1565 
1566 	if (!list_empty(&jd->extent_list)) {
1567 		jext = list_entry(jd->extent_list.prev, struct gfs2_journal_extent, list);
1568 		if ((jext->dblock + jext->blocks) == dblock) {
1569 			jext->blocks += blocks;
1570 			return 0;
1571 		}
1572 	}
1573 
1574 	jext = kzalloc(sizeof(struct gfs2_journal_extent), GFP_NOFS);
1575 	if (jext == NULL)
1576 		return -ENOMEM;
1577 	jext->dblock = dblock;
1578 	jext->lblock = lblock;
1579 	jext->blocks = blocks;
1580 	list_add_tail(&jext->list, &jd->extent_list);
1581 	jd->nr_extents++;
1582 	return 0;
1583 }
1584 
1585 /**
1586  * gfs2_map_journal_extents - Cache journal bmap info
1587  * @sdp: The super block
1588  * @jd: The journal to map
1589  *
1590  * Create a reusable "extent" mapping from all logical
1591  * blocks to all physical blocks for the given journal.  This will save
1592  * us time when writing journal blocks.  Most journals will have only one
1593  * extent that maps all their logical blocks.  That's because gfs2.mkfs
1594  * arranges the journal blocks sequentially to maximize performance.
1595  * So the extent would map the first block for the entire file length.
1596  * However, gfs2_jadd can happen while file activity is happening, so
1597  * those journals may not be sequential.  Less likely is the case where
1598  * the users created their own journals by mounting the metafs and
1599  * laying it out.  But it's still possible.  These journals might have
1600  * several extents.
1601  *
1602  * Returns: 0 on success, or error on failure
1603  */
1604 
1605 int gfs2_map_journal_extents(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd)
1606 {
1607 	u64 lblock = 0;
1608 	u64 lblock_stop;
1609 	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
1610 	struct buffer_head bh;
1611 	unsigned int shift = sdp->sd_sb.sb_bsize_shift;
1612 	u64 size;
1613 	int rc;
1614 
1615 	lblock_stop = i_size_read(jd->jd_inode) >> shift;
1616 	size = (lblock_stop - lblock) << shift;
1617 	jd->nr_extents = 0;
1618 	WARN_ON(!list_empty(&jd->extent_list));
1619 
1620 	do {
1621 		bh.b_state = 0;
1622 		bh.b_blocknr = 0;
1623 		bh.b_size = size;
1624 		rc = gfs2_block_map(jd->jd_inode, lblock, &bh, 0);
1625 		if (rc || !buffer_mapped(&bh))
1626 			goto fail;
1627 		rc = gfs2_add_jextent(jd, lblock, bh.b_blocknr, bh.b_size >> shift);
1628 		if (rc)
1629 			goto fail;
1630 		size -= bh.b_size;
1631 		lblock += (bh.b_size >> ip->i_inode.i_blkbits);
1632 	} while(size > 0);
1633 
1634 	fs_info(sdp, "journal %d mapped with %u extents\n", jd->jd_jid,
1635 		jd->nr_extents);
1636 	return 0;
1637 
1638 fail:
1639 	fs_warn(sdp, "error %d mapping journal %u at offset %llu (extent %u)\n",
1640 		rc, jd->jd_jid,
1641 		(unsigned long long)(i_size_read(jd->jd_inode) - size),
1642 		jd->nr_extents);
1643 	fs_warn(sdp, "bmap=%d lblock=%llu block=%llu, state=0x%08lx, size=%llu\n",
1644 		rc, (unsigned long long)lblock, (unsigned long long)bh.b_blocknr,
1645 		bh.b_state, (unsigned long long)bh.b_size);
1646 	gfs2_free_journal_extents(jd);
1647 	return rc;
1648 }
1649 
1650 /**
1651  * gfs2_write_alloc_required - figure out if a write will require an allocation
1652  * @ip: the file being written to
1653  * @offset: the offset to write to
1654  * @len: the number of bytes being written
1655  *
1656  * Returns: 1 if an alloc is required, 0 otherwise
1657  */
1658 
1659 int gfs2_write_alloc_required(struct gfs2_inode *ip, u64 offset,
1660 			      unsigned int len)
1661 {
1662 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1663 	struct buffer_head bh;
1664 	unsigned int shift;
1665 	u64 lblock, lblock_stop, size;
1666 	u64 end_of_file;
1667 
1668 	if (!len)
1669 		return 0;
1670 
1671 	if (gfs2_is_stuffed(ip)) {
1672 		if (offset + len >
1673 		    sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode))
1674 			return 1;
1675 		return 0;
1676 	}
1677 
1678 	shift = sdp->sd_sb.sb_bsize_shift;
1679 	BUG_ON(gfs2_is_dir(ip));
1680 	end_of_file = (i_size_read(&ip->i_inode) + sdp->sd_sb.sb_bsize - 1) >> shift;
1681 	lblock = offset >> shift;
1682 	lblock_stop = (offset + len + sdp->sd_sb.sb_bsize - 1) >> shift;
1683 	if (lblock_stop > end_of_file)
1684 		return 1;
1685 
1686 	size = (lblock_stop - lblock) << shift;
1687 	do {
1688 		bh.b_state = 0;
1689 		bh.b_size = size;
1690 		gfs2_block_map(&ip->i_inode, lblock, &bh, 0);
1691 		if (!buffer_mapped(&bh))
1692 			return 1;
1693 		size -= bh.b_size;
1694 		lblock += (bh.b_size >> ip->i_inode.i_blkbits);
1695 	} while(size > 0);
1696 
1697 	return 0;
1698 }
1699 
1700