Loading...
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 */
36struct metapath {
37 struct buffer_head *mp_bh[GFS2_MAX_META_HEIGHT];
38 __u16 mp_list[GFS2_MAX_META_HEIGHT];
39};
40
41struct strip_mine {
42 int sm_first;
43 unsigned int sm_height;
44};
45
46/**
47 * gfs2_unstuffer_page - unstuff a stuffed inode into a block cached by a page
48 * @ip: the inode
49 * @dibh: the dinode buffer
50 * @block: the block number that was allocated
51 * @page: The (optional) page. This is looked up if @page is NULL
52 *
53 * Returns: errno
54 */
55
56static int gfs2_unstuffer_page(struct gfs2_inode *ip, struct buffer_head *dibh,
57 u64 block, struct page *page)
58{
59 struct inode *inode = &ip->i_inode;
60 struct buffer_head *bh;
61 int release = 0;
62
63 if (!page || page->index) {
64 page = find_or_create_page(inode->i_mapping, 0, GFP_NOFS);
65 if (!page)
66 return -ENOMEM;
67 release = 1;
68 }
69
70 if (!PageUptodate(page)) {
71 void *kaddr = kmap(page);
72 u64 dsize = i_size_read(inode);
73
74 if (dsize > (dibh->b_size - sizeof(struct gfs2_dinode)))
75 dsize = dibh->b_size - sizeof(struct gfs2_dinode);
76
77 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize);
78 memset(kaddr + dsize, 0, PAGE_CACHE_SIZE - dsize);
79 kunmap(page);
80
81 SetPageUptodate(page);
82 }
83
84 if (!page_has_buffers(page))
85 create_empty_buffers(page, 1 << inode->i_blkbits,
86 (1 << BH_Uptodate));
87
88 bh = page_buffers(page);
89
90 if (!buffer_mapped(bh))
91 map_bh(bh, inode->i_sb, block);
92
93 set_buffer_uptodate(bh);
94 if (!gfs2_is_jdata(ip))
95 mark_buffer_dirty(bh);
96 if (!gfs2_is_writeback(ip))
97 gfs2_trans_add_data(ip->i_gl, bh);
98
99 if (release) {
100 unlock_page(page);
101 page_cache_release(page);
102 }
103
104 return 0;
105}
106
107/**
108 * gfs2_unstuff_dinode - Unstuff a dinode when the data has grown too big
109 * @ip: The GFS2 inode to unstuff
110 * @page: The (optional) page. This is looked up if the @page is NULL
111 *
112 * This routine unstuffs a dinode and returns it to a "normal" state such
113 * that the height can be grown in the traditional way.
114 *
115 * Returns: errno
116 */
117
118int gfs2_unstuff_dinode(struct gfs2_inode *ip, struct page *page)
119{
120 struct buffer_head *bh, *dibh;
121 struct gfs2_dinode *di;
122 u64 block = 0;
123 int isdir = gfs2_is_dir(ip);
124 int error;
125
126 down_write(&ip->i_rw_mutex);
127
128 error = gfs2_meta_inode_buffer(ip, &dibh);
129 if (error)
130 goto out;
131
132 if (i_size_read(&ip->i_inode)) {
133 /* Get a free block, fill it with the stuffed data,
134 and write it out to disk */
135
136 unsigned int n = 1;
137 error = gfs2_alloc_blocks(ip, &block, &n, 0, NULL);
138 if (error)
139 goto out_brelse;
140 if (isdir) {
141 gfs2_trans_add_unrevoke(GFS2_SB(&ip->i_inode), block, 1);
142 error = gfs2_dir_get_new_buffer(ip, block, &bh);
143 if (error)
144 goto out_brelse;
145 gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_meta_header),
146 dibh, sizeof(struct gfs2_dinode));
147 brelse(bh);
148 } else {
149 error = gfs2_unstuffer_page(ip, dibh, block, page);
150 if (error)
151 goto out_brelse;
152 }
153 }
154
155 /* Set up the pointer to the new block */
156
157 gfs2_trans_add_meta(ip->i_gl, dibh);
158 di = (struct gfs2_dinode *)dibh->b_data;
159 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
160
161 if (i_size_read(&ip->i_inode)) {
162 *(__be64 *)(di + 1) = cpu_to_be64(block);
163 gfs2_add_inode_blocks(&ip->i_inode, 1);
164 di->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
165 }
166
167 ip->i_height = 1;
168 di->di_height = cpu_to_be16(1);
169
170out_brelse:
171 brelse(dibh);
172out:
173 up_write(&ip->i_rw_mutex);
174 return error;
175}
176
177
178/**
179 * find_metapath - Find path through the metadata tree
180 * @sdp: The superblock
181 * @mp: The metapath to return the result in
182 * @block: The disk block to look up
183 * @height: The pre-calculated height of the metadata tree
184 *
185 * This routine returns a struct metapath structure that defines a path
186 * through the metadata of inode "ip" to get to block "block".
187 *
188 * Example:
189 * Given: "ip" is a height 3 file, "offset" is 101342453, and this is a
190 * filesystem with a blocksize of 4096.
191 *
192 * find_metapath() would return a struct metapath structure set to:
193 * mp_offset = 101342453, mp_height = 3, mp_list[0] = 0, mp_list[1] = 48,
194 * and mp_list[2] = 165.
195 *
196 * That means that in order to get to the block containing the byte at
197 * offset 101342453, we would load the indirect block pointed to by pointer
198 * 0 in the dinode. We would then load the indirect block pointed to by
199 * pointer 48 in that indirect block. We would then load the data block
200 * pointed to by pointer 165 in that indirect block.
201 *
202 * ----------------------------------------
203 * | Dinode | |
204 * | | 4|
205 * | |0 1 2 3 4 5 9|
206 * | | 6|
207 * ----------------------------------------
208 * |
209 * |
210 * V
211 * ----------------------------------------
212 * | Indirect Block |
213 * | 5|
214 * | 4 4 4 4 4 5 5 1|
215 * |0 5 6 7 8 9 0 1 2|
216 * ----------------------------------------
217 * |
218 * |
219 * V
220 * ----------------------------------------
221 * | Indirect Block |
222 * | 1 1 1 1 1 5|
223 * | 6 6 6 6 6 1|
224 * |0 3 4 5 6 7 2|
225 * ----------------------------------------
226 * |
227 * |
228 * V
229 * ----------------------------------------
230 * | Data block containing offset |
231 * | 101342453 |
232 * | |
233 * | |
234 * ----------------------------------------
235 *
236 */
237
238static void find_metapath(const struct gfs2_sbd *sdp, u64 block,
239 struct metapath *mp, unsigned int height)
240{
241 unsigned int i;
242
243 for (i = height; i--;)
244 mp->mp_list[i] = do_div(block, sdp->sd_inptrs);
245
246}
247
248static inline unsigned int metapath_branch_start(const struct metapath *mp)
249{
250 if (mp->mp_list[0] == 0)
251 return 2;
252 return 1;
253}
254
255/**
256 * metapointer - Return pointer to start of metadata in a buffer
257 * @height: The metadata height (0 = dinode)
258 * @mp: The metapath
259 *
260 * Return a pointer to the block number of the next height of the metadata
261 * tree given a buffer containing the pointer to the current height of the
262 * metadata tree.
263 */
264
265static inline __be64 *metapointer(unsigned int height, const struct metapath *mp)
266{
267 struct buffer_head *bh = mp->mp_bh[height];
268 unsigned int head_size = (height > 0) ?
269 sizeof(struct gfs2_meta_header) : sizeof(struct gfs2_dinode);
270 return ((__be64 *)(bh->b_data + head_size)) + mp->mp_list[height];
271}
272
273static void gfs2_metapath_ra(struct gfs2_glock *gl,
274 const struct buffer_head *bh, const __be64 *pos)
275{
276 struct buffer_head *rabh;
277 const __be64 *endp = (const __be64 *)(bh->b_data + bh->b_size);
278 const __be64 *t;
279
280 for (t = pos; t < endp; t++) {
281 if (!*t)
282 continue;
283
284 rabh = gfs2_getbuf(gl, be64_to_cpu(*t), CREATE);
285 if (trylock_buffer(rabh)) {
286 if (!buffer_uptodate(rabh)) {
287 rabh->b_end_io = end_buffer_read_sync;
288 submit_bh(READA | REQ_META, rabh);
289 continue;
290 }
291 unlock_buffer(rabh);
292 }
293 brelse(rabh);
294 }
295}
296
297/**
298 * lookup_metapath - Walk the metadata tree to a specific point
299 * @ip: The inode
300 * @mp: The metapath
301 *
302 * Assumes that the inode's buffer has already been looked up and
303 * hooked onto mp->mp_bh[0] and that the metapath has been initialised
304 * by find_metapath().
305 *
306 * If this function encounters part of the tree which has not been
307 * allocated, it returns the current height of the tree at the point
308 * at which it found the unallocated block. Blocks which are found are
309 * added to the mp->mp_bh[] list.
310 *
311 * Returns: error or height of metadata tree
312 */
313
314static int lookup_metapath(struct gfs2_inode *ip, struct metapath *mp)
315{
316 unsigned int end_of_metadata = ip->i_height - 1;
317 unsigned int x;
318 __be64 *ptr;
319 u64 dblock;
320 int ret;
321
322 for (x = 0; x < end_of_metadata; x++) {
323 ptr = metapointer(x, mp);
324 dblock = be64_to_cpu(*ptr);
325 if (!dblock)
326 return x + 1;
327
328 ret = gfs2_meta_indirect_buffer(ip, x+1, dblock, &mp->mp_bh[x+1]);
329 if (ret)
330 return ret;
331 }
332
333 return ip->i_height;
334}
335
336static inline void release_metapath(struct metapath *mp)
337{
338 int i;
339
340 for (i = 0; i < GFS2_MAX_META_HEIGHT; i++) {
341 if (mp->mp_bh[i] == NULL)
342 break;
343 brelse(mp->mp_bh[i]);
344 }
345}
346
347/**
348 * gfs2_extent_length - Returns length of an extent of blocks
349 * @start: Start of the buffer
350 * @len: Length of the buffer in bytes
351 * @ptr: Current position in the buffer
352 * @limit: Max extent length to return (0 = unlimited)
353 * @eob: Set to 1 if we hit "end of block"
354 *
355 * If the first block is zero (unallocated) it will return the number of
356 * unallocated blocks in the extent, otherwise it will return the number
357 * of contiguous blocks in the extent.
358 *
359 * Returns: The length of the extent (minimum of one block)
360 */
361
362static inline unsigned int gfs2_extent_length(void *start, unsigned int len, __be64 *ptr, unsigned limit, int *eob)
363{
364 const __be64 *end = (start + len);
365 const __be64 *first = ptr;
366 u64 d = be64_to_cpu(*ptr);
367
368 *eob = 0;
369 do {
370 ptr++;
371 if (ptr >= end)
372 break;
373 if (limit && --limit == 0)
374 break;
375 if (d)
376 d++;
377 } while(be64_to_cpu(*ptr) == d);
378 if (ptr >= end)
379 *eob = 1;
380 return (ptr - first);
381}
382
383static inline void bmap_lock(struct gfs2_inode *ip, int create)
384{
385 if (create)
386 down_write(&ip->i_rw_mutex);
387 else
388 down_read(&ip->i_rw_mutex);
389}
390
391static inline void bmap_unlock(struct gfs2_inode *ip, int create)
392{
393 if (create)
394 up_write(&ip->i_rw_mutex);
395 else
396 up_read(&ip->i_rw_mutex);
397}
398
399static inline __be64 *gfs2_indirect_init(struct metapath *mp,
400 struct gfs2_glock *gl, unsigned int i,
401 unsigned offset, u64 bn)
402{
403 __be64 *ptr = (__be64 *)(mp->mp_bh[i - 1]->b_data +
404 ((i > 1) ? sizeof(struct gfs2_meta_header) :
405 sizeof(struct gfs2_dinode)));
406 BUG_ON(i < 1);
407 BUG_ON(mp->mp_bh[i] != NULL);
408 mp->mp_bh[i] = gfs2_meta_new(gl, bn);
409 gfs2_trans_add_meta(gl, mp->mp_bh[i]);
410 gfs2_metatype_set(mp->mp_bh[i], GFS2_METATYPE_IN, GFS2_FORMAT_IN);
411 gfs2_buffer_clear_tail(mp->mp_bh[i], sizeof(struct gfs2_meta_header));
412 ptr += offset;
413 *ptr = cpu_to_be64(bn);
414 return ptr;
415}
416
417enum alloc_state {
418 ALLOC_DATA = 0,
419 ALLOC_GROW_DEPTH = 1,
420 ALLOC_GROW_HEIGHT = 2,
421 /* ALLOC_UNSTUFF = 3, TBD and rather complicated */
422};
423
424/**
425 * gfs2_bmap_alloc - Build a metadata tree of the requested height
426 * @inode: The GFS2 inode
427 * @lblock: The logical starting block of the extent
428 * @bh_map: This is used to return the mapping details
429 * @mp: The metapath
430 * @sheight: The starting height (i.e. whats already mapped)
431 * @height: The height to build to
432 * @maxlen: The max number of data blocks to alloc
433 *
434 * In this routine we may have to alloc:
435 * i) Indirect blocks to grow the metadata tree height
436 * ii) Indirect blocks to fill in lower part of the metadata tree
437 * iii) Data blocks
438 *
439 * The function is in two parts. The first part works out the total
440 * number of blocks which we need. The second part does the actual
441 * allocation asking for an extent at a time (if enough contiguous free
442 * blocks are available, there will only be one request per bmap call)
443 * and uses the state machine to initialise the blocks in order.
444 *
445 * Returns: errno on error
446 */
447
448static int gfs2_bmap_alloc(struct inode *inode, const sector_t lblock,
449 struct buffer_head *bh_map, struct metapath *mp,
450 const unsigned int sheight,
451 const unsigned int height,
452 const unsigned int maxlen)
453{
454 struct gfs2_inode *ip = GFS2_I(inode);
455 struct gfs2_sbd *sdp = GFS2_SB(inode);
456 struct super_block *sb = sdp->sd_vfs;
457 struct buffer_head *dibh = mp->mp_bh[0];
458 u64 bn, dblock = 0;
459 unsigned n, i, blks, alloced = 0, iblks = 0, branch_start = 0;
460 unsigned dblks = 0;
461 unsigned ptrs_per_blk;
462 const unsigned end_of_metadata = height - 1;
463 int ret;
464 int eob = 0;
465 enum alloc_state state;
466 __be64 *ptr;
467 __be64 zero_bn = 0;
468
469 BUG_ON(sheight < 1);
470 BUG_ON(dibh == NULL);
471
472 gfs2_trans_add_meta(ip->i_gl, dibh);
473
474 if (height == sheight) {
475 struct buffer_head *bh;
476 /* Bottom indirect block exists, find unalloced extent size */
477 ptr = metapointer(end_of_metadata, mp);
478 bh = mp->mp_bh[end_of_metadata];
479 dblks = gfs2_extent_length(bh->b_data, bh->b_size, ptr, maxlen,
480 &eob);
481 BUG_ON(dblks < 1);
482 state = ALLOC_DATA;
483 } else {
484 /* Need to allocate indirect blocks */
485 ptrs_per_blk = height > 1 ? sdp->sd_inptrs : sdp->sd_diptrs;
486 dblks = min(maxlen, ptrs_per_blk - mp->mp_list[end_of_metadata]);
487 if (height == ip->i_height) {
488 /* Writing into existing tree, extend tree down */
489 iblks = height - sheight;
490 state = ALLOC_GROW_DEPTH;
491 } else {
492 /* Building up tree height */
493 state = ALLOC_GROW_HEIGHT;
494 iblks = height - ip->i_height;
495 branch_start = metapath_branch_start(mp);
496 iblks += (height - branch_start);
497 }
498 }
499
500 /* start of the second part of the function (state machine) */
501
502 blks = dblks + iblks;
503 i = sheight;
504 do {
505 int error;
506 n = blks - alloced;
507 error = gfs2_alloc_blocks(ip, &bn, &n, 0, NULL);
508 if (error)
509 return error;
510 alloced += n;
511 if (state != ALLOC_DATA || gfs2_is_jdata(ip))
512 gfs2_trans_add_unrevoke(sdp, bn, n);
513 switch (state) {
514 /* Growing height of tree */
515 case ALLOC_GROW_HEIGHT:
516 if (i == 1) {
517 ptr = (__be64 *)(dibh->b_data +
518 sizeof(struct gfs2_dinode));
519 zero_bn = *ptr;
520 }
521 for (; i - 1 < height - ip->i_height && n > 0; i++, n--)
522 gfs2_indirect_init(mp, ip->i_gl, i, 0, bn++);
523 if (i - 1 == height - ip->i_height) {
524 i--;
525 gfs2_buffer_copy_tail(mp->mp_bh[i],
526 sizeof(struct gfs2_meta_header),
527 dibh, sizeof(struct gfs2_dinode));
528 gfs2_buffer_clear_tail(dibh,
529 sizeof(struct gfs2_dinode) +
530 sizeof(__be64));
531 ptr = (__be64 *)(mp->mp_bh[i]->b_data +
532 sizeof(struct gfs2_meta_header));
533 *ptr = zero_bn;
534 state = ALLOC_GROW_DEPTH;
535 for(i = branch_start; i < height; i++) {
536 if (mp->mp_bh[i] == NULL)
537 break;
538 brelse(mp->mp_bh[i]);
539 mp->mp_bh[i] = NULL;
540 }
541 i = branch_start;
542 }
543 if (n == 0)
544 break;
545 /* Branching from existing tree */
546 case ALLOC_GROW_DEPTH:
547 if (i > 1 && i < height)
548 gfs2_trans_add_meta(ip->i_gl, mp->mp_bh[i-1]);
549 for (; i < height && n > 0; i++, n--)
550 gfs2_indirect_init(mp, ip->i_gl, i,
551 mp->mp_list[i-1], bn++);
552 if (i == height)
553 state = ALLOC_DATA;
554 if (n == 0)
555 break;
556 /* Tree complete, adding data blocks */
557 case ALLOC_DATA:
558 BUG_ON(n > dblks);
559 BUG_ON(mp->mp_bh[end_of_metadata] == NULL);
560 gfs2_trans_add_meta(ip->i_gl, mp->mp_bh[end_of_metadata]);
561 dblks = n;
562 ptr = metapointer(end_of_metadata, mp);
563 dblock = bn;
564 while (n-- > 0)
565 *ptr++ = cpu_to_be64(bn++);
566 if (buffer_zeronew(bh_map)) {
567 ret = sb_issue_zeroout(sb, dblock, dblks,
568 GFP_NOFS);
569 if (ret) {
570 fs_err(sdp,
571 "Failed to zero data buffers\n");
572 clear_buffer_zeronew(bh_map);
573 }
574 }
575 break;
576 }
577 } while ((state != ALLOC_DATA) || !dblock);
578
579 ip->i_height = height;
580 gfs2_add_inode_blocks(&ip->i_inode, alloced);
581 gfs2_dinode_out(ip, mp->mp_bh[0]->b_data);
582 map_bh(bh_map, inode->i_sb, dblock);
583 bh_map->b_size = dblks << inode->i_blkbits;
584 set_buffer_new(bh_map);
585 return 0;
586}
587
588/**
589 * gfs2_block_map - Map a block from an inode to a disk block
590 * @inode: The inode
591 * @lblock: The logical block number
592 * @bh_map: The bh to be mapped
593 * @create: True if its ok to alloc blocks to satify the request
594 *
595 * Sets buffer_mapped() if successful, sets buffer_boundary() if a
596 * read of metadata will be required before the next block can be
597 * mapped. Sets buffer_new() if new blocks were allocated.
598 *
599 * Returns: errno
600 */
601
602int gfs2_block_map(struct inode *inode, sector_t lblock,
603 struct buffer_head *bh_map, int create)
604{
605 struct gfs2_inode *ip = GFS2_I(inode);
606 struct gfs2_sbd *sdp = GFS2_SB(inode);
607 unsigned int bsize = sdp->sd_sb.sb_bsize;
608 const unsigned int maxlen = bh_map->b_size >> inode->i_blkbits;
609 const u64 *arr = sdp->sd_heightsize;
610 __be64 *ptr;
611 u64 size;
612 struct metapath mp;
613 int ret;
614 int eob;
615 unsigned int len;
616 struct buffer_head *bh;
617 u8 height;
618
619 BUG_ON(maxlen == 0);
620
621 memset(mp.mp_bh, 0, sizeof(mp.mp_bh));
622 bmap_lock(ip, create);
623 clear_buffer_mapped(bh_map);
624 clear_buffer_new(bh_map);
625 clear_buffer_boundary(bh_map);
626 trace_gfs2_bmap(ip, bh_map, lblock, create, 1);
627 if (gfs2_is_dir(ip)) {
628 bsize = sdp->sd_jbsize;
629 arr = sdp->sd_jheightsize;
630 }
631
632 ret = gfs2_meta_inode_buffer(ip, &mp.mp_bh[0]);
633 if (ret)
634 goto out;
635
636 height = ip->i_height;
637 size = (lblock + 1) * bsize;
638 while (size > arr[height])
639 height++;
640 find_metapath(sdp, lblock, &mp, height);
641 ret = 1;
642 if (height > ip->i_height || gfs2_is_stuffed(ip))
643 goto do_alloc;
644 ret = lookup_metapath(ip, &mp);
645 if (ret < 0)
646 goto out;
647 if (ret != ip->i_height)
648 goto do_alloc;
649 ptr = metapointer(ip->i_height - 1, &mp);
650 if (*ptr == 0)
651 goto do_alloc;
652 map_bh(bh_map, inode->i_sb, be64_to_cpu(*ptr));
653 bh = mp.mp_bh[ip->i_height - 1];
654 len = gfs2_extent_length(bh->b_data, bh->b_size, ptr, maxlen, &eob);
655 bh_map->b_size = (len << inode->i_blkbits);
656 if (eob)
657 set_buffer_boundary(bh_map);
658 ret = 0;
659out:
660 release_metapath(&mp);
661 trace_gfs2_bmap(ip, bh_map, lblock, create, ret);
662 bmap_unlock(ip, create);
663 return ret;
664
665do_alloc:
666 /* All allocations are done here, firstly check create flag */
667 if (!create) {
668 BUG_ON(gfs2_is_stuffed(ip));
669 ret = 0;
670 goto out;
671 }
672
673 /* At this point ret is the tree depth of already allocated blocks */
674 ret = gfs2_bmap_alloc(inode, lblock, bh_map, &mp, ret, height, maxlen);
675 goto out;
676}
677
678/*
679 * Deprecated: do not use in new code
680 */
681int gfs2_extent_map(struct inode *inode, u64 lblock, int *new, u64 *dblock, unsigned *extlen)
682{
683 struct buffer_head bh = { .b_state = 0, .b_blocknr = 0 };
684 int ret;
685 int create = *new;
686
687 BUG_ON(!extlen);
688 BUG_ON(!dblock);
689 BUG_ON(!new);
690
691 bh.b_size = 1 << (inode->i_blkbits + (create ? 0 : 5));
692 ret = gfs2_block_map(inode, lblock, &bh, create);
693 *extlen = bh.b_size >> inode->i_blkbits;
694 *dblock = bh.b_blocknr;
695 if (buffer_new(&bh))
696 *new = 1;
697 else
698 *new = 0;
699 return ret;
700}
701
702/**
703 * do_strip - Look for a layer a particular layer of the file and strip it off
704 * @ip: the inode
705 * @dibh: the dinode buffer
706 * @bh: A buffer of pointers
707 * @top: The first pointer in the buffer
708 * @bottom: One more than the last pointer
709 * @height: the height this buffer is at
710 * @data: a pointer to a struct strip_mine
711 *
712 * Returns: errno
713 */
714
715static int do_strip(struct gfs2_inode *ip, struct buffer_head *dibh,
716 struct buffer_head *bh, __be64 *top, __be64 *bottom,
717 unsigned int height, struct strip_mine *sm)
718{
719 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
720 struct gfs2_rgrp_list rlist;
721 u64 bn, bstart;
722 u32 blen, btotal;
723 __be64 *p;
724 unsigned int rg_blocks = 0;
725 int metadata;
726 unsigned int revokes = 0;
727 int x;
728 int error;
729
730 error = gfs2_rindex_update(sdp);
731 if (error)
732 return error;
733
734 if (!*top)
735 sm->sm_first = 0;
736
737 if (height != sm->sm_height)
738 return 0;
739
740 if (sm->sm_first) {
741 top++;
742 sm->sm_first = 0;
743 }
744
745 metadata = (height != ip->i_height - 1);
746 if (metadata)
747 revokes = (height) ? sdp->sd_inptrs : sdp->sd_diptrs;
748 else if (ip->i_depth)
749 revokes = sdp->sd_inptrs;
750
751 memset(&rlist, 0, sizeof(struct gfs2_rgrp_list));
752 bstart = 0;
753 blen = 0;
754
755 for (p = top; p < bottom; p++) {
756 if (!*p)
757 continue;
758
759 bn = be64_to_cpu(*p);
760
761 if (bstart + blen == bn)
762 blen++;
763 else {
764 if (bstart)
765 gfs2_rlist_add(ip, &rlist, bstart);
766
767 bstart = bn;
768 blen = 1;
769 }
770 }
771
772 if (bstart)
773 gfs2_rlist_add(ip, &rlist, bstart);
774 else
775 goto out; /* Nothing to do */
776
777 gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE);
778
779 for (x = 0; x < rlist.rl_rgrps; x++) {
780 struct gfs2_rgrpd *rgd;
781 rgd = rlist.rl_ghs[x].gh_gl->gl_object;
782 rg_blocks += rgd->rd_length;
783 }
784
785 error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs);
786 if (error)
787 goto out_rlist;
788
789 if (gfs2_rs_active(ip->i_res)) /* needs to be done with the rgrp glock held */
790 gfs2_rs_deltree(ip->i_res);
791
792 error = gfs2_trans_begin(sdp, rg_blocks + RES_DINODE +
793 RES_INDIRECT + RES_STATFS + RES_QUOTA,
794 revokes);
795 if (error)
796 goto out_rg_gunlock;
797
798 down_write(&ip->i_rw_mutex);
799
800 gfs2_trans_add_meta(ip->i_gl, dibh);
801 gfs2_trans_add_meta(ip->i_gl, bh);
802
803 bstart = 0;
804 blen = 0;
805 btotal = 0;
806
807 for (p = top; p < bottom; p++) {
808 if (!*p)
809 continue;
810
811 bn = be64_to_cpu(*p);
812
813 if (bstart + blen == bn)
814 blen++;
815 else {
816 if (bstart) {
817 __gfs2_free_blocks(ip, bstart, blen, metadata);
818 btotal += blen;
819 }
820
821 bstart = bn;
822 blen = 1;
823 }
824
825 *p = 0;
826 gfs2_add_inode_blocks(&ip->i_inode, -1);
827 }
828 if (bstart) {
829 __gfs2_free_blocks(ip, bstart, blen, metadata);
830 btotal += blen;
831 }
832
833 gfs2_statfs_change(sdp, 0, +btotal, 0);
834 gfs2_quota_change(ip, -(s64)btotal, ip->i_inode.i_uid,
835 ip->i_inode.i_gid);
836
837 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
838
839 gfs2_dinode_out(ip, dibh->b_data);
840
841 up_write(&ip->i_rw_mutex);
842
843 gfs2_trans_end(sdp);
844
845out_rg_gunlock:
846 gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs);
847out_rlist:
848 gfs2_rlist_free(&rlist);
849out:
850 return error;
851}
852
853/**
854 * recursive_scan - recursively scan through the end of a file
855 * @ip: the inode
856 * @dibh: the dinode buffer
857 * @mp: the path through the metadata to the point to start
858 * @height: the height the recursion is at
859 * @block: the indirect block to look at
860 * @first: 1 if this is the first block
861 * @sm: data opaque to this function to pass to @bc
862 *
863 * When this is first called @height and @block should be zero and
864 * @first should be 1.
865 *
866 * Returns: errno
867 */
868
869static int recursive_scan(struct gfs2_inode *ip, struct buffer_head *dibh,
870 struct metapath *mp, unsigned int height,
871 u64 block, int first, struct strip_mine *sm)
872{
873 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
874 struct buffer_head *bh = NULL;
875 __be64 *top, *bottom;
876 u64 bn;
877 int error;
878 int mh_size = sizeof(struct gfs2_meta_header);
879
880 if (!height) {
881 error = gfs2_meta_inode_buffer(ip, &bh);
882 if (error)
883 return error;
884 dibh = bh;
885
886 top = (__be64 *)(bh->b_data + sizeof(struct gfs2_dinode)) + mp->mp_list[0];
887 bottom = (__be64 *)(bh->b_data + sizeof(struct gfs2_dinode)) + sdp->sd_diptrs;
888 } else {
889 error = gfs2_meta_indirect_buffer(ip, height, block, &bh);
890 if (error)
891 return error;
892
893 top = (__be64 *)(bh->b_data + mh_size) +
894 (first ? mp->mp_list[height] : 0);
895
896 bottom = (__be64 *)(bh->b_data + mh_size) + sdp->sd_inptrs;
897 }
898
899 error = do_strip(ip, dibh, bh, top, bottom, height, sm);
900 if (error)
901 goto out;
902
903 if (height < ip->i_height - 1) {
904
905 gfs2_metapath_ra(ip->i_gl, bh, top);
906
907 for (; top < bottom; top++, first = 0) {
908 if (!*top)
909 continue;
910
911 bn = be64_to_cpu(*top);
912
913 error = recursive_scan(ip, dibh, mp, height + 1, bn,
914 first, sm);
915 if (error)
916 break;
917 }
918 }
919out:
920 brelse(bh);
921 return error;
922}
923
924
925/**
926 * gfs2_block_truncate_page - Deal with zeroing out data for truncate
927 *
928 * This is partly borrowed from ext3.
929 */
930static int gfs2_block_truncate_page(struct address_space *mapping, loff_t from)
931{
932 struct inode *inode = mapping->host;
933 struct gfs2_inode *ip = GFS2_I(inode);
934 unsigned long index = from >> PAGE_CACHE_SHIFT;
935 unsigned offset = from & (PAGE_CACHE_SIZE-1);
936 unsigned blocksize, iblock, length, pos;
937 struct buffer_head *bh;
938 struct page *page;
939 int err;
940
941 page = find_or_create_page(mapping, index, GFP_NOFS);
942 if (!page)
943 return 0;
944
945 blocksize = inode->i_sb->s_blocksize;
946 length = blocksize - (offset & (blocksize - 1));
947 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
948
949 if (!page_has_buffers(page))
950 create_empty_buffers(page, blocksize, 0);
951
952 /* Find the buffer that contains "offset" */
953 bh = page_buffers(page);
954 pos = blocksize;
955 while (offset >= pos) {
956 bh = bh->b_this_page;
957 iblock++;
958 pos += blocksize;
959 }
960
961 err = 0;
962
963 if (!buffer_mapped(bh)) {
964 gfs2_block_map(inode, iblock, bh, 0);
965 /* unmapped? It's a hole - nothing to do */
966 if (!buffer_mapped(bh))
967 goto unlock;
968 }
969
970 /* Ok, it's mapped. Make sure it's up-to-date */
971 if (PageUptodate(page))
972 set_buffer_uptodate(bh);
973
974 if (!buffer_uptodate(bh)) {
975 err = -EIO;
976 ll_rw_block(READ, 1, &bh);
977 wait_on_buffer(bh);
978 /* Uhhuh. Read error. Complain and punt. */
979 if (!buffer_uptodate(bh))
980 goto unlock;
981 err = 0;
982 }
983
984 if (!gfs2_is_writeback(ip))
985 gfs2_trans_add_data(ip->i_gl, bh);
986
987 zero_user(page, offset, length);
988 mark_buffer_dirty(bh);
989unlock:
990 unlock_page(page);
991 page_cache_release(page);
992 return err;
993}
994
995/**
996 * gfs2_journaled_truncate - Wrapper for truncate_pagecache for jdata files
997 * @inode: The inode being truncated
998 * @oldsize: The original (larger) size
999 * @newsize: The new smaller size
1000 *
1001 * With jdata files, we have to journal a revoke for each block which is
1002 * truncated. As a result, we need to split this into separate transactions
1003 * if the number of pages being truncated gets too large.
1004 */
1005
1006#define GFS2_JTRUNC_REVOKES 8192
1007
1008static int gfs2_journaled_truncate(struct inode *inode, u64 oldsize, u64 newsize)
1009{
1010 struct gfs2_sbd *sdp = GFS2_SB(inode);
1011 u64 max_chunk = GFS2_JTRUNC_REVOKES * sdp->sd_vfs->s_blocksize;
1012 u64 chunk;
1013 int error;
1014
1015 while (oldsize != newsize) {
1016 chunk = oldsize - newsize;
1017 if (chunk > max_chunk)
1018 chunk = max_chunk;
1019 truncate_pagecache(inode, oldsize - chunk);
1020 oldsize -= chunk;
1021 gfs2_trans_end(sdp);
1022 error = gfs2_trans_begin(sdp, RES_DINODE, GFS2_JTRUNC_REVOKES);
1023 if (error)
1024 return error;
1025 }
1026
1027 return 0;
1028}
1029
1030static int trunc_start(struct inode *inode, u64 oldsize, u64 newsize)
1031{
1032 struct gfs2_inode *ip = GFS2_I(inode);
1033 struct gfs2_sbd *sdp = GFS2_SB(inode);
1034 struct address_space *mapping = inode->i_mapping;
1035 struct buffer_head *dibh;
1036 int journaled = gfs2_is_jdata(ip);
1037 int error;
1038
1039 if (journaled)
1040 error = gfs2_trans_begin(sdp, RES_DINODE + RES_JDATA, GFS2_JTRUNC_REVOKES);
1041 else
1042 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1043 if (error)
1044 return error;
1045
1046 error = gfs2_meta_inode_buffer(ip, &dibh);
1047 if (error)
1048 goto out;
1049
1050 gfs2_trans_add_meta(ip->i_gl, dibh);
1051
1052 if (gfs2_is_stuffed(ip)) {
1053 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode) + newsize);
1054 } else {
1055 if (newsize & (u64)(sdp->sd_sb.sb_bsize - 1)) {
1056 error = gfs2_block_truncate_page(mapping, newsize);
1057 if (error)
1058 goto out_brelse;
1059 }
1060 ip->i_diskflags |= GFS2_DIF_TRUNC_IN_PROG;
1061 }
1062
1063 i_size_write(inode, newsize);
1064 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
1065 gfs2_dinode_out(ip, dibh->b_data);
1066
1067 if (journaled)
1068 error = gfs2_journaled_truncate(inode, oldsize, newsize);
1069 else
1070 truncate_pagecache(inode, newsize);
1071
1072 if (error) {
1073 brelse(dibh);
1074 return error;
1075 }
1076
1077out_brelse:
1078 brelse(dibh);
1079out:
1080 gfs2_trans_end(sdp);
1081 return error;
1082}
1083
1084static int trunc_dealloc(struct gfs2_inode *ip, u64 size)
1085{
1086 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1087 unsigned int height = ip->i_height;
1088 u64 lblock;
1089 struct metapath mp;
1090 int error;
1091
1092 if (!size)
1093 lblock = 0;
1094 else
1095 lblock = (size - 1) >> sdp->sd_sb.sb_bsize_shift;
1096
1097 find_metapath(sdp, lblock, &mp, ip->i_height);
1098 error = gfs2_rindex_update(sdp);
1099 if (error)
1100 return error;
1101
1102 error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1103 if (error)
1104 return error;
1105
1106 while (height--) {
1107 struct strip_mine sm;
1108 sm.sm_first = !!size;
1109 sm.sm_height = height;
1110
1111 error = recursive_scan(ip, NULL, &mp, 0, 0, 1, &sm);
1112 if (error)
1113 break;
1114 }
1115
1116 gfs2_quota_unhold(ip);
1117
1118 return error;
1119}
1120
1121static int trunc_end(struct gfs2_inode *ip)
1122{
1123 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1124 struct buffer_head *dibh;
1125 int error;
1126
1127 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1128 if (error)
1129 return error;
1130
1131 down_write(&ip->i_rw_mutex);
1132
1133 error = gfs2_meta_inode_buffer(ip, &dibh);
1134 if (error)
1135 goto out;
1136
1137 if (!i_size_read(&ip->i_inode)) {
1138 ip->i_height = 0;
1139 ip->i_goal = ip->i_no_addr;
1140 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
1141 gfs2_ordered_del_inode(ip);
1142 }
1143 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
1144 ip->i_diskflags &= ~GFS2_DIF_TRUNC_IN_PROG;
1145
1146 gfs2_trans_add_meta(ip->i_gl, dibh);
1147 gfs2_dinode_out(ip, dibh->b_data);
1148 brelse(dibh);
1149
1150out:
1151 up_write(&ip->i_rw_mutex);
1152 gfs2_trans_end(sdp);
1153 return error;
1154}
1155
1156/**
1157 * do_shrink - make a file smaller
1158 * @inode: the inode
1159 * @oldsize: the current inode size
1160 * @newsize: the size to make the file
1161 *
1162 * Called with an exclusive lock on @inode. The @size must
1163 * be equal to or smaller than the current inode size.
1164 *
1165 * Returns: errno
1166 */
1167
1168static int do_shrink(struct inode *inode, u64 oldsize, u64 newsize)
1169{
1170 struct gfs2_inode *ip = GFS2_I(inode);
1171 int error;
1172
1173 error = trunc_start(inode, oldsize, newsize);
1174 if (error < 0)
1175 return error;
1176 if (gfs2_is_stuffed(ip))
1177 return 0;
1178
1179 error = trunc_dealloc(ip, newsize);
1180 if (error == 0)
1181 error = trunc_end(ip);
1182
1183 return error;
1184}
1185
1186void gfs2_trim_blocks(struct inode *inode)
1187{
1188 u64 size = inode->i_size;
1189 int ret;
1190
1191 ret = do_shrink(inode, size, size);
1192 WARN_ON(ret != 0);
1193}
1194
1195/**
1196 * do_grow - Touch and update inode size
1197 * @inode: The inode
1198 * @size: The new size
1199 *
1200 * This function updates the timestamps on the inode and
1201 * may also increase the size of the inode. This function
1202 * must not be called with @size any smaller than the current
1203 * inode size.
1204 *
1205 * Although it is not strictly required to unstuff files here,
1206 * earlier versions of GFS2 have a bug in the stuffed file reading
1207 * code which will result in a buffer overrun if the size is larger
1208 * than the max stuffed file size. In order to prevent this from
1209 * occurring, such files are unstuffed, but in other cases we can
1210 * just update the inode size directly.
1211 *
1212 * Returns: 0 on success, or -ve on error
1213 */
1214
1215static int do_grow(struct inode *inode, u64 size)
1216{
1217 struct gfs2_inode *ip = GFS2_I(inode);
1218 struct gfs2_sbd *sdp = GFS2_SB(inode);
1219 struct gfs2_alloc_parms ap = { .target = 1, };
1220 struct buffer_head *dibh;
1221 int error;
1222 int unstuff = 0;
1223
1224 if (gfs2_is_stuffed(ip) &&
1225 (size > (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)))) {
1226 error = gfs2_quota_lock_check(ip);
1227 if (error)
1228 return error;
1229
1230 error = gfs2_inplace_reserve(ip, &ap);
1231 if (error)
1232 goto do_grow_qunlock;
1233 unstuff = 1;
1234 }
1235
1236 error = gfs2_trans_begin(sdp, RES_DINODE + RES_STATFS + RES_RG_BIT +
1237 (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF ?
1238 0 : RES_QUOTA), 0);
1239 if (error)
1240 goto do_grow_release;
1241
1242 if (unstuff) {
1243 error = gfs2_unstuff_dinode(ip, NULL);
1244 if (error)
1245 goto do_end_trans;
1246 }
1247
1248 error = gfs2_meta_inode_buffer(ip, &dibh);
1249 if (error)
1250 goto do_end_trans;
1251
1252 i_size_write(inode, size);
1253 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
1254 gfs2_trans_add_meta(ip->i_gl, dibh);
1255 gfs2_dinode_out(ip, dibh->b_data);
1256 brelse(dibh);
1257
1258do_end_trans:
1259 gfs2_trans_end(sdp);
1260do_grow_release:
1261 if (unstuff) {
1262 gfs2_inplace_release(ip);
1263do_grow_qunlock:
1264 gfs2_quota_unlock(ip);
1265 }
1266 return error;
1267}
1268
1269/**
1270 * gfs2_setattr_size - make a file a given size
1271 * @inode: the inode
1272 * @newsize: the size to make the file
1273 *
1274 * The file size can grow, shrink, or stay the same size. This
1275 * is called holding i_mutex and an exclusive glock on the inode
1276 * in question.
1277 *
1278 * Returns: errno
1279 */
1280
1281int gfs2_setattr_size(struct inode *inode, u64 newsize)
1282{
1283 struct gfs2_inode *ip = GFS2_I(inode);
1284 int ret;
1285 u64 oldsize;
1286
1287 BUG_ON(!S_ISREG(inode->i_mode));
1288
1289 ret = inode_newsize_ok(inode, newsize);
1290 if (ret)
1291 return ret;
1292
1293 ret = get_write_access(inode);
1294 if (ret)
1295 return ret;
1296
1297 inode_dio_wait(inode);
1298
1299 ret = gfs2_rs_alloc(ip);
1300 if (ret)
1301 goto out;
1302
1303 oldsize = inode->i_size;
1304 if (newsize >= oldsize) {
1305 ret = do_grow(inode, newsize);
1306 goto out;
1307 }
1308
1309 gfs2_rs_deltree(ip->i_res);
1310 ret = do_shrink(inode, oldsize, newsize);
1311out:
1312 put_write_access(inode);
1313 return ret;
1314}
1315
1316int gfs2_truncatei_resume(struct gfs2_inode *ip)
1317{
1318 int error;
1319 error = trunc_dealloc(ip, i_size_read(&ip->i_inode));
1320 if (!error)
1321 error = trunc_end(ip);
1322 return error;
1323}
1324
1325int gfs2_file_dealloc(struct gfs2_inode *ip)
1326{
1327 return trunc_dealloc(ip, 0);
1328}
1329
1330/**
1331 * gfs2_free_journal_extents - Free cached journal bmap info
1332 * @jd: The journal
1333 *
1334 */
1335
1336void gfs2_free_journal_extents(struct gfs2_jdesc *jd)
1337{
1338 struct gfs2_journal_extent *jext;
1339
1340 while(!list_empty(&jd->extent_list)) {
1341 jext = list_entry(jd->extent_list.next, struct gfs2_journal_extent, list);
1342 list_del(&jext->list);
1343 kfree(jext);
1344 }
1345}
1346
1347/**
1348 * gfs2_add_jextent - Add or merge a new extent to extent cache
1349 * @jd: The journal descriptor
1350 * @lblock: The logical block at start of new extent
1351 * @pblock: The physical block at start of new extent
1352 * @blocks: Size of extent in fs blocks
1353 *
1354 * Returns: 0 on success or -ENOMEM
1355 */
1356
1357static int gfs2_add_jextent(struct gfs2_jdesc *jd, u64 lblock, u64 dblock, u64 blocks)
1358{
1359 struct gfs2_journal_extent *jext;
1360
1361 if (!list_empty(&jd->extent_list)) {
1362 jext = list_entry(jd->extent_list.prev, struct gfs2_journal_extent, list);
1363 if ((jext->dblock + jext->blocks) == dblock) {
1364 jext->blocks += blocks;
1365 return 0;
1366 }
1367 }
1368
1369 jext = kzalloc(sizeof(struct gfs2_journal_extent), GFP_NOFS);
1370 if (jext == NULL)
1371 return -ENOMEM;
1372 jext->dblock = dblock;
1373 jext->lblock = lblock;
1374 jext->blocks = blocks;
1375 list_add_tail(&jext->list, &jd->extent_list);
1376 jd->nr_extents++;
1377 return 0;
1378}
1379
1380/**
1381 * gfs2_map_journal_extents - Cache journal bmap info
1382 * @sdp: The super block
1383 * @jd: The journal to map
1384 *
1385 * Create a reusable "extent" mapping from all logical
1386 * blocks to all physical blocks for the given journal. This will save
1387 * us time when writing journal blocks. Most journals will have only one
1388 * extent that maps all their logical blocks. That's because gfs2.mkfs
1389 * arranges the journal blocks sequentially to maximize performance.
1390 * So the extent would map the first block for the entire file length.
1391 * However, gfs2_jadd can happen while file activity is happening, so
1392 * those journals may not be sequential. Less likely is the case where
1393 * the users created their own journals by mounting the metafs and
1394 * laying it out. But it's still possible. These journals might have
1395 * several extents.
1396 *
1397 * Returns: 0 on success, or error on failure
1398 */
1399
1400int gfs2_map_journal_extents(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd)
1401{
1402 u64 lblock = 0;
1403 u64 lblock_stop;
1404 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
1405 struct buffer_head bh;
1406 unsigned int shift = sdp->sd_sb.sb_bsize_shift;
1407 u64 size;
1408 int rc;
1409
1410 lblock_stop = i_size_read(jd->jd_inode) >> shift;
1411 size = (lblock_stop - lblock) << shift;
1412 jd->nr_extents = 0;
1413 WARN_ON(!list_empty(&jd->extent_list));
1414
1415 do {
1416 bh.b_state = 0;
1417 bh.b_blocknr = 0;
1418 bh.b_size = size;
1419 rc = gfs2_block_map(jd->jd_inode, lblock, &bh, 0);
1420 if (rc || !buffer_mapped(&bh))
1421 goto fail;
1422 rc = gfs2_add_jextent(jd, lblock, bh.b_blocknr, bh.b_size >> shift);
1423 if (rc)
1424 goto fail;
1425 size -= bh.b_size;
1426 lblock += (bh.b_size >> ip->i_inode.i_blkbits);
1427 } while(size > 0);
1428
1429 fs_info(sdp, "journal %d mapped with %u extents\n", jd->jd_jid,
1430 jd->nr_extents);
1431 return 0;
1432
1433fail:
1434 fs_warn(sdp, "error %d mapping journal %u at offset %llu (extent %u)\n",
1435 rc, jd->jd_jid,
1436 (unsigned long long)(i_size_read(jd->jd_inode) - size),
1437 jd->nr_extents);
1438 fs_warn(sdp, "bmap=%d lblock=%llu block=%llu, state=0x%08lx, size=%llu\n",
1439 rc, (unsigned long long)lblock, (unsigned long long)bh.b_blocknr,
1440 bh.b_state, (unsigned long long)bh.b_size);
1441 gfs2_free_journal_extents(jd);
1442 return rc;
1443}
1444
1445/**
1446 * gfs2_write_alloc_required - figure out if a write will require an allocation
1447 * @ip: the file being written to
1448 * @offset: the offset to write to
1449 * @len: the number of bytes being written
1450 *
1451 * Returns: 1 if an alloc is required, 0 otherwise
1452 */
1453
1454int gfs2_write_alloc_required(struct gfs2_inode *ip, u64 offset,
1455 unsigned int len)
1456{
1457 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1458 struct buffer_head bh;
1459 unsigned int shift;
1460 u64 lblock, lblock_stop, size;
1461 u64 end_of_file;
1462
1463 if (!len)
1464 return 0;
1465
1466 if (gfs2_is_stuffed(ip)) {
1467 if (offset + len >
1468 sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode))
1469 return 1;
1470 return 0;
1471 }
1472
1473 shift = sdp->sd_sb.sb_bsize_shift;
1474 BUG_ON(gfs2_is_dir(ip));
1475 end_of_file = (i_size_read(&ip->i_inode) + sdp->sd_sb.sb_bsize - 1) >> shift;
1476 lblock = offset >> shift;
1477 lblock_stop = (offset + len + sdp->sd_sb.sb_bsize - 1) >> shift;
1478 if (lblock_stop > end_of_file)
1479 return 1;
1480
1481 size = (lblock_stop - lblock) << shift;
1482 do {
1483 bh.b_state = 0;
1484 bh.b_size = size;
1485 gfs2_block_map(&ip->i_inode, lblock, &bh, 0);
1486 if (!buffer_mapped(&bh))
1487 return 1;
1488 size -= bh.b_size;
1489 lblock += (bh.b_size >> ip->i_inode.i_blkbits);
1490 } while(size > 0);
1491
1492 return 0;
1493}
1494
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 "super.h"
26#include "trans.h"
27#include "dir.h"
28#include "util.h"
29#include "trace_gfs2.h"
30
31/* This doesn't need to be that large as max 64 bit pointers in a 4k
32 * block is 512, so __u16 is fine for that. It saves stack space to
33 * keep it small.
34 */
35struct metapath {
36 struct buffer_head *mp_bh[GFS2_MAX_META_HEIGHT];
37 __u16 mp_list[GFS2_MAX_META_HEIGHT];
38};
39
40struct strip_mine {
41 int sm_first;
42 unsigned int sm_height;
43};
44
45/**
46 * gfs2_unstuffer_page - unstuff a stuffed inode into a block cached by a page
47 * @ip: the inode
48 * @dibh: the dinode buffer
49 * @block: the block number that was allocated
50 * @page: The (optional) page. This is looked up if @page is NULL
51 *
52 * Returns: errno
53 */
54
55static int gfs2_unstuffer_page(struct gfs2_inode *ip, struct buffer_head *dibh,
56 u64 block, struct page *page)
57{
58 struct inode *inode = &ip->i_inode;
59 struct buffer_head *bh;
60 int release = 0;
61
62 if (!page || page->index) {
63 page = find_or_create_page(inode->i_mapping, 0, GFP_NOFS);
64 if (!page)
65 return -ENOMEM;
66 release = 1;
67 }
68
69 if (!PageUptodate(page)) {
70 void *kaddr = kmap(page);
71 u64 dsize = i_size_read(inode);
72
73 if (dsize > (dibh->b_size - sizeof(struct gfs2_dinode)))
74 dsize = dibh->b_size - sizeof(struct gfs2_dinode);
75
76 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize);
77 memset(kaddr + dsize, 0, PAGE_CACHE_SIZE - dsize);
78 kunmap(page);
79
80 SetPageUptodate(page);
81 }
82
83 if (!page_has_buffers(page))
84 create_empty_buffers(page, 1 << inode->i_blkbits,
85 (1 << BH_Uptodate));
86
87 bh = page_buffers(page);
88
89 if (!buffer_mapped(bh))
90 map_bh(bh, inode->i_sb, block);
91
92 set_buffer_uptodate(bh);
93 if (!gfs2_is_jdata(ip))
94 mark_buffer_dirty(bh);
95 if (!gfs2_is_writeback(ip))
96 gfs2_trans_add_bh(ip->i_gl, bh, 0);
97
98 if (release) {
99 unlock_page(page);
100 page_cache_release(page);
101 }
102
103 return 0;
104}
105
106/**
107 * gfs2_unstuff_dinode - Unstuff a dinode when the data has grown too big
108 * @ip: The GFS2 inode to unstuff
109 * @page: The (optional) page. This is looked up if the @page is NULL
110 *
111 * This routine unstuffs a dinode and returns it to a "normal" state such
112 * that the height can be grown in the traditional way.
113 *
114 * Returns: errno
115 */
116
117int gfs2_unstuff_dinode(struct gfs2_inode *ip, struct page *page)
118{
119 struct buffer_head *bh, *dibh;
120 struct gfs2_dinode *di;
121 u64 block = 0;
122 int isdir = gfs2_is_dir(ip);
123 int error;
124
125 down_write(&ip->i_rw_mutex);
126
127 error = gfs2_meta_inode_buffer(ip, &dibh);
128 if (error)
129 goto out;
130
131 if (i_size_read(&ip->i_inode)) {
132 /* Get a free block, fill it with the stuffed data,
133 and write it out to disk */
134
135 unsigned int n = 1;
136 error = gfs2_alloc_blocks(ip, &block, &n, 0, NULL);
137 if (error)
138 goto out_brelse;
139 if (isdir) {
140 gfs2_trans_add_unrevoke(GFS2_SB(&ip->i_inode), block, 1);
141 error = gfs2_dir_get_new_buffer(ip, block, &bh);
142 if (error)
143 goto out_brelse;
144 gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_meta_header),
145 dibh, sizeof(struct gfs2_dinode));
146 brelse(bh);
147 } else {
148 error = gfs2_unstuffer_page(ip, dibh, block, page);
149 if (error)
150 goto out_brelse;
151 }
152 }
153
154 /* Set up the pointer to the new block */
155
156 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
157 di = (struct gfs2_dinode *)dibh->b_data;
158 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
159
160 if (i_size_read(&ip->i_inode)) {
161 *(__be64 *)(di + 1) = cpu_to_be64(block);
162 gfs2_add_inode_blocks(&ip->i_inode, 1);
163 di->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
164 }
165
166 ip->i_height = 1;
167 di->di_height = cpu_to_be16(1);
168
169out_brelse:
170 brelse(dibh);
171out:
172 up_write(&ip->i_rw_mutex);
173 return error;
174}
175
176
177/**
178 * find_metapath - Find path through the metadata tree
179 * @sdp: The superblock
180 * @mp: The metapath to return the result in
181 * @block: The disk block to look up
182 * @height: The pre-calculated height of the metadata tree
183 *
184 * This routine returns a struct metapath structure that defines a path
185 * through the metadata of inode "ip" to get to block "block".
186 *
187 * Example:
188 * Given: "ip" is a height 3 file, "offset" is 101342453, and this is a
189 * filesystem with a blocksize of 4096.
190 *
191 * find_metapath() would return a struct metapath structure set to:
192 * mp_offset = 101342453, mp_height = 3, mp_list[0] = 0, mp_list[1] = 48,
193 * and mp_list[2] = 165.
194 *
195 * That means that in order to get to the block containing the byte at
196 * offset 101342453, we would load the indirect block pointed to by pointer
197 * 0 in the dinode. We would then load the indirect block pointed to by
198 * pointer 48 in that indirect block. We would then load the data block
199 * pointed to by pointer 165 in that indirect block.
200 *
201 * ----------------------------------------
202 * | Dinode | |
203 * | | 4|
204 * | |0 1 2 3 4 5 9|
205 * | | 6|
206 * ----------------------------------------
207 * |
208 * |
209 * V
210 * ----------------------------------------
211 * | Indirect Block |
212 * | 5|
213 * | 4 4 4 4 4 5 5 1|
214 * |0 5 6 7 8 9 0 1 2|
215 * ----------------------------------------
216 * |
217 * |
218 * V
219 * ----------------------------------------
220 * | Indirect Block |
221 * | 1 1 1 1 1 5|
222 * | 6 6 6 6 6 1|
223 * |0 3 4 5 6 7 2|
224 * ----------------------------------------
225 * |
226 * |
227 * V
228 * ----------------------------------------
229 * | Data block containing offset |
230 * | 101342453 |
231 * | |
232 * | |
233 * ----------------------------------------
234 *
235 */
236
237static void find_metapath(const struct gfs2_sbd *sdp, u64 block,
238 struct metapath *mp, unsigned int height)
239{
240 unsigned int i;
241
242 for (i = height; i--;)
243 mp->mp_list[i] = do_div(block, sdp->sd_inptrs);
244
245}
246
247static inline unsigned int metapath_branch_start(const struct metapath *mp)
248{
249 if (mp->mp_list[0] == 0)
250 return 2;
251 return 1;
252}
253
254/**
255 * metapointer - Return pointer to start of metadata in a buffer
256 * @height: The metadata height (0 = dinode)
257 * @mp: The metapath
258 *
259 * Return a pointer to the block number of the next height of the metadata
260 * tree given a buffer containing the pointer to the current height of the
261 * metadata tree.
262 */
263
264static inline __be64 *metapointer(unsigned int height, const struct metapath *mp)
265{
266 struct buffer_head *bh = mp->mp_bh[height];
267 unsigned int head_size = (height > 0) ?
268 sizeof(struct gfs2_meta_header) : sizeof(struct gfs2_dinode);
269 return ((__be64 *)(bh->b_data + head_size)) + mp->mp_list[height];
270}
271
272static void gfs2_metapath_ra(struct gfs2_glock *gl,
273 const struct buffer_head *bh, const __be64 *pos)
274{
275 struct buffer_head *rabh;
276 const __be64 *endp = (const __be64 *)(bh->b_data + bh->b_size);
277 const __be64 *t;
278
279 for (t = pos; t < endp; t++) {
280 if (!*t)
281 continue;
282
283 rabh = gfs2_getbuf(gl, be64_to_cpu(*t), CREATE);
284 if (trylock_buffer(rabh)) {
285 if (!buffer_uptodate(rabh)) {
286 rabh->b_end_io = end_buffer_read_sync;
287 submit_bh(READA | REQ_META, rabh);
288 continue;
289 }
290 unlock_buffer(rabh);
291 }
292 brelse(rabh);
293 }
294}
295
296/**
297 * lookup_metapath - Walk the metadata tree to a specific point
298 * @ip: The inode
299 * @mp: The metapath
300 *
301 * Assumes that the inode's buffer has already been looked up and
302 * hooked onto mp->mp_bh[0] and that the metapath has been initialised
303 * by find_metapath().
304 *
305 * If this function encounters part of the tree which has not been
306 * allocated, it returns the current height of the tree at the point
307 * at which it found the unallocated block. Blocks which are found are
308 * added to the mp->mp_bh[] list.
309 *
310 * Returns: error or height of metadata tree
311 */
312
313static int lookup_metapath(struct gfs2_inode *ip, struct metapath *mp)
314{
315 unsigned int end_of_metadata = ip->i_height - 1;
316 unsigned int x;
317 __be64 *ptr;
318 u64 dblock;
319 int ret;
320
321 for (x = 0; x < end_of_metadata; x++) {
322 ptr = metapointer(x, mp);
323 dblock = be64_to_cpu(*ptr);
324 if (!dblock)
325 return x + 1;
326
327 ret = gfs2_meta_indirect_buffer(ip, x+1, dblock, &mp->mp_bh[x+1]);
328 if (ret)
329 return ret;
330 }
331
332 return ip->i_height;
333}
334
335static inline void release_metapath(struct metapath *mp)
336{
337 int i;
338
339 for (i = 0; i < GFS2_MAX_META_HEIGHT; i++) {
340 if (mp->mp_bh[i] == NULL)
341 break;
342 brelse(mp->mp_bh[i]);
343 }
344}
345
346/**
347 * gfs2_extent_length - Returns length of an extent of blocks
348 * @start: Start of the buffer
349 * @len: Length of the buffer in bytes
350 * @ptr: Current position in the buffer
351 * @limit: Max extent length to return (0 = unlimited)
352 * @eob: Set to 1 if we hit "end of block"
353 *
354 * If the first block is zero (unallocated) it will return the number of
355 * unallocated blocks in the extent, otherwise it will return the number
356 * of contiguous blocks in the extent.
357 *
358 * Returns: The length of the extent (minimum of one block)
359 */
360
361static inline unsigned int gfs2_extent_length(void *start, unsigned int len, __be64 *ptr, unsigned limit, int *eob)
362{
363 const __be64 *end = (start + len);
364 const __be64 *first = ptr;
365 u64 d = be64_to_cpu(*ptr);
366
367 *eob = 0;
368 do {
369 ptr++;
370 if (ptr >= end)
371 break;
372 if (limit && --limit == 0)
373 break;
374 if (d)
375 d++;
376 } while(be64_to_cpu(*ptr) == d);
377 if (ptr >= end)
378 *eob = 1;
379 return (ptr - first);
380}
381
382static inline void bmap_lock(struct gfs2_inode *ip, int create)
383{
384 if (create)
385 down_write(&ip->i_rw_mutex);
386 else
387 down_read(&ip->i_rw_mutex);
388}
389
390static inline void bmap_unlock(struct gfs2_inode *ip, int create)
391{
392 if (create)
393 up_write(&ip->i_rw_mutex);
394 else
395 up_read(&ip->i_rw_mutex);
396}
397
398static inline __be64 *gfs2_indirect_init(struct metapath *mp,
399 struct gfs2_glock *gl, unsigned int i,
400 unsigned offset, u64 bn)
401{
402 __be64 *ptr = (__be64 *)(mp->mp_bh[i - 1]->b_data +
403 ((i > 1) ? sizeof(struct gfs2_meta_header) :
404 sizeof(struct gfs2_dinode)));
405 BUG_ON(i < 1);
406 BUG_ON(mp->mp_bh[i] != NULL);
407 mp->mp_bh[i] = gfs2_meta_new(gl, bn);
408 gfs2_trans_add_bh(gl, mp->mp_bh[i], 1);
409 gfs2_metatype_set(mp->mp_bh[i], GFS2_METATYPE_IN, GFS2_FORMAT_IN);
410 gfs2_buffer_clear_tail(mp->mp_bh[i], sizeof(struct gfs2_meta_header));
411 ptr += offset;
412 *ptr = cpu_to_be64(bn);
413 return ptr;
414}
415
416enum alloc_state {
417 ALLOC_DATA = 0,
418 ALLOC_GROW_DEPTH = 1,
419 ALLOC_GROW_HEIGHT = 2,
420 /* ALLOC_UNSTUFF = 3, TBD and rather complicated */
421};
422
423/**
424 * gfs2_bmap_alloc - Build a metadata tree of the requested height
425 * @inode: The GFS2 inode
426 * @lblock: The logical starting block of the extent
427 * @bh_map: This is used to return the mapping details
428 * @mp: The metapath
429 * @sheight: The starting height (i.e. whats already mapped)
430 * @height: The height to build to
431 * @maxlen: The max number of data blocks to alloc
432 *
433 * In this routine we may have to alloc:
434 * i) Indirect blocks to grow the metadata tree height
435 * ii) Indirect blocks to fill in lower part of the metadata tree
436 * iii) Data blocks
437 *
438 * The function is in two parts. The first part works out the total
439 * number of blocks which we need. The second part does the actual
440 * allocation asking for an extent at a time (if enough contiguous free
441 * blocks are available, there will only be one request per bmap call)
442 * and uses the state machine to initialise the blocks in order.
443 *
444 * Returns: errno on error
445 */
446
447static int gfs2_bmap_alloc(struct inode *inode, const sector_t lblock,
448 struct buffer_head *bh_map, struct metapath *mp,
449 const unsigned int sheight,
450 const unsigned int height,
451 const unsigned int maxlen)
452{
453 struct gfs2_inode *ip = GFS2_I(inode);
454 struct gfs2_sbd *sdp = GFS2_SB(inode);
455 struct super_block *sb = sdp->sd_vfs;
456 struct buffer_head *dibh = mp->mp_bh[0];
457 u64 bn, dblock = 0;
458 unsigned n, i, blks, alloced = 0, iblks = 0, branch_start = 0;
459 unsigned dblks = 0;
460 unsigned ptrs_per_blk;
461 const unsigned end_of_metadata = height - 1;
462 int ret;
463 int eob = 0;
464 enum alloc_state state;
465 __be64 *ptr;
466 __be64 zero_bn = 0;
467
468 BUG_ON(sheight < 1);
469 BUG_ON(dibh == NULL);
470
471 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
472
473 if (height == sheight) {
474 struct buffer_head *bh;
475 /* Bottom indirect block exists, find unalloced extent size */
476 ptr = metapointer(end_of_metadata, mp);
477 bh = mp->mp_bh[end_of_metadata];
478 dblks = gfs2_extent_length(bh->b_data, bh->b_size, ptr, maxlen,
479 &eob);
480 BUG_ON(dblks < 1);
481 state = ALLOC_DATA;
482 } else {
483 /* Need to allocate indirect blocks */
484 ptrs_per_blk = height > 1 ? sdp->sd_inptrs : sdp->sd_diptrs;
485 dblks = min(maxlen, ptrs_per_blk - mp->mp_list[end_of_metadata]);
486 if (height == ip->i_height) {
487 /* Writing into existing tree, extend tree down */
488 iblks = height - sheight;
489 state = ALLOC_GROW_DEPTH;
490 } else {
491 /* Building up tree height */
492 state = ALLOC_GROW_HEIGHT;
493 iblks = height - ip->i_height;
494 branch_start = metapath_branch_start(mp);
495 iblks += (height - branch_start);
496 }
497 }
498
499 /* start of the second part of the function (state machine) */
500
501 blks = dblks + iblks;
502 i = sheight;
503 do {
504 int error;
505 n = blks - alloced;
506 error = gfs2_alloc_blocks(ip, &bn, &n, 0, NULL);
507 if (error)
508 return error;
509 alloced += n;
510 if (state != ALLOC_DATA || gfs2_is_jdata(ip))
511 gfs2_trans_add_unrevoke(sdp, bn, n);
512 switch (state) {
513 /* Growing height of tree */
514 case ALLOC_GROW_HEIGHT:
515 if (i == 1) {
516 ptr = (__be64 *)(dibh->b_data +
517 sizeof(struct gfs2_dinode));
518 zero_bn = *ptr;
519 }
520 for (; i - 1 < height - ip->i_height && n > 0; i++, n--)
521 gfs2_indirect_init(mp, ip->i_gl, i, 0, bn++);
522 if (i - 1 == height - ip->i_height) {
523 i--;
524 gfs2_buffer_copy_tail(mp->mp_bh[i],
525 sizeof(struct gfs2_meta_header),
526 dibh, sizeof(struct gfs2_dinode));
527 gfs2_buffer_clear_tail(dibh,
528 sizeof(struct gfs2_dinode) +
529 sizeof(__be64));
530 ptr = (__be64 *)(mp->mp_bh[i]->b_data +
531 sizeof(struct gfs2_meta_header));
532 *ptr = zero_bn;
533 state = ALLOC_GROW_DEPTH;
534 for(i = branch_start; i < height; i++) {
535 if (mp->mp_bh[i] == NULL)
536 break;
537 brelse(mp->mp_bh[i]);
538 mp->mp_bh[i] = NULL;
539 }
540 i = branch_start;
541 }
542 if (n == 0)
543 break;
544 /* Branching from existing tree */
545 case ALLOC_GROW_DEPTH:
546 if (i > 1 && i < height)
547 gfs2_trans_add_bh(ip->i_gl, mp->mp_bh[i-1], 1);
548 for (; i < height && n > 0; i++, n--)
549 gfs2_indirect_init(mp, ip->i_gl, i,
550 mp->mp_list[i-1], bn++);
551 if (i == height)
552 state = ALLOC_DATA;
553 if (n == 0)
554 break;
555 /* Tree complete, adding data blocks */
556 case ALLOC_DATA:
557 BUG_ON(n > dblks);
558 BUG_ON(mp->mp_bh[end_of_metadata] == NULL);
559 gfs2_trans_add_bh(ip->i_gl, mp->mp_bh[end_of_metadata], 1);
560 dblks = n;
561 ptr = metapointer(end_of_metadata, mp);
562 dblock = bn;
563 while (n-- > 0)
564 *ptr++ = cpu_to_be64(bn++);
565 if (buffer_zeronew(bh_map)) {
566 ret = sb_issue_zeroout(sb, dblock, dblks,
567 GFP_NOFS);
568 if (ret) {
569 fs_err(sdp,
570 "Failed to zero data buffers\n");
571 clear_buffer_zeronew(bh_map);
572 }
573 }
574 break;
575 }
576 } while ((state != ALLOC_DATA) || !dblock);
577
578 ip->i_height = height;
579 gfs2_add_inode_blocks(&ip->i_inode, alloced);
580 gfs2_dinode_out(ip, mp->mp_bh[0]->b_data);
581 map_bh(bh_map, inode->i_sb, dblock);
582 bh_map->b_size = dblks << inode->i_blkbits;
583 set_buffer_new(bh_map);
584 return 0;
585}
586
587/**
588 * gfs2_block_map - Map a block from an inode to a disk block
589 * @inode: The inode
590 * @lblock: The logical block number
591 * @bh_map: The bh to be mapped
592 * @create: True if its ok to alloc blocks to satify the request
593 *
594 * Sets buffer_mapped() if successful, sets buffer_boundary() if a
595 * read of metadata will be required before the next block can be
596 * mapped. Sets buffer_new() if new blocks were allocated.
597 *
598 * Returns: errno
599 */
600
601int gfs2_block_map(struct inode *inode, sector_t lblock,
602 struct buffer_head *bh_map, int create)
603{
604 struct gfs2_inode *ip = GFS2_I(inode);
605 struct gfs2_sbd *sdp = GFS2_SB(inode);
606 unsigned int bsize = sdp->sd_sb.sb_bsize;
607 const unsigned int maxlen = bh_map->b_size >> inode->i_blkbits;
608 const u64 *arr = sdp->sd_heightsize;
609 __be64 *ptr;
610 u64 size;
611 struct metapath mp;
612 int ret;
613 int eob;
614 unsigned int len;
615 struct buffer_head *bh;
616 u8 height;
617
618 BUG_ON(maxlen == 0);
619
620 memset(mp.mp_bh, 0, sizeof(mp.mp_bh));
621 bmap_lock(ip, create);
622 clear_buffer_mapped(bh_map);
623 clear_buffer_new(bh_map);
624 clear_buffer_boundary(bh_map);
625 trace_gfs2_bmap(ip, bh_map, lblock, create, 1);
626 if (gfs2_is_dir(ip)) {
627 bsize = sdp->sd_jbsize;
628 arr = sdp->sd_jheightsize;
629 }
630
631 ret = gfs2_meta_inode_buffer(ip, &mp.mp_bh[0]);
632 if (ret)
633 goto out;
634
635 height = ip->i_height;
636 size = (lblock + 1) * bsize;
637 while (size > arr[height])
638 height++;
639 find_metapath(sdp, lblock, &mp, height);
640 ret = 1;
641 if (height > ip->i_height || gfs2_is_stuffed(ip))
642 goto do_alloc;
643 ret = lookup_metapath(ip, &mp);
644 if (ret < 0)
645 goto out;
646 if (ret != ip->i_height)
647 goto do_alloc;
648 ptr = metapointer(ip->i_height - 1, &mp);
649 if (*ptr == 0)
650 goto do_alloc;
651 map_bh(bh_map, inode->i_sb, be64_to_cpu(*ptr));
652 bh = mp.mp_bh[ip->i_height - 1];
653 len = gfs2_extent_length(bh->b_data, bh->b_size, ptr, maxlen, &eob);
654 bh_map->b_size = (len << inode->i_blkbits);
655 if (eob)
656 set_buffer_boundary(bh_map);
657 ret = 0;
658out:
659 release_metapath(&mp);
660 trace_gfs2_bmap(ip, bh_map, lblock, create, ret);
661 bmap_unlock(ip, create);
662 return ret;
663
664do_alloc:
665 /* All allocations are done here, firstly check create flag */
666 if (!create) {
667 BUG_ON(gfs2_is_stuffed(ip));
668 ret = 0;
669 goto out;
670 }
671
672 /* At this point ret is the tree depth of already allocated blocks */
673 ret = gfs2_bmap_alloc(inode, lblock, bh_map, &mp, ret, height, maxlen);
674 goto out;
675}
676
677/*
678 * Deprecated: do not use in new code
679 */
680int gfs2_extent_map(struct inode *inode, u64 lblock, int *new, u64 *dblock, unsigned *extlen)
681{
682 struct buffer_head bh = { .b_state = 0, .b_blocknr = 0 };
683 int ret;
684 int create = *new;
685
686 BUG_ON(!extlen);
687 BUG_ON(!dblock);
688 BUG_ON(!new);
689
690 bh.b_size = 1 << (inode->i_blkbits + (create ? 0 : 5));
691 ret = gfs2_block_map(inode, lblock, &bh, create);
692 *extlen = bh.b_size >> inode->i_blkbits;
693 *dblock = bh.b_blocknr;
694 if (buffer_new(&bh))
695 *new = 1;
696 else
697 *new = 0;
698 return ret;
699}
700
701/**
702 * do_strip - Look for a layer a particular layer of the file and strip it off
703 * @ip: the inode
704 * @dibh: the dinode buffer
705 * @bh: A buffer of pointers
706 * @top: The first pointer in the buffer
707 * @bottom: One more than the last pointer
708 * @height: the height this buffer is at
709 * @data: a pointer to a struct strip_mine
710 *
711 * Returns: errno
712 */
713
714static int do_strip(struct gfs2_inode *ip, struct buffer_head *dibh,
715 struct buffer_head *bh, __be64 *top, __be64 *bottom,
716 unsigned int height, struct strip_mine *sm)
717{
718 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
719 struct gfs2_rgrp_list rlist;
720 u64 bn, bstart;
721 u32 blen, btotal;
722 __be64 *p;
723 unsigned int rg_blocks = 0;
724 int metadata;
725 unsigned int revokes = 0;
726 int x;
727 int error;
728
729 error = gfs2_rindex_update(sdp);
730 if (error)
731 return error;
732
733 if (!*top)
734 sm->sm_first = 0;
735
736 if (height != sm->sm_height)
737 return 0;
738
739 if (sm->sm_first) {
740 top++;
741 sm->sm_first = 0;
742 }
743
744 metadata = (height != ip->i_height - 1);
745 if (metadata)
746 revokes = (height) ? sdp->sd_inptrs : sdp->sd_diptrs;
747 else if (ip->i_depth)
748 revokes = sdp->sd_inptrs;
749
750 memset(&rlist, 0, sizeof(struct gfs2_rgrp_list));
751 bstart = 0;
752 blen = 0;
753
754 for (p = top; p < bottom; p++) {
755 if (!*p)
756 continue;
757
758 bn = be64_to_cpu(*p);
759
760 if (bstart + blen == bn)
761 blen++;
762 else {
763 if (bstart)
764 gfs2_rlist_add(ip, &rlist, bstart);
765
766 bstart = bn;
767 blen = 1;
768 }
769 }
770
771 if (bstart)
772 gfs2_rlist_add(ip, &rlist, bstart);
773 else
774 goto out; /* Nothing to do */
775
776 gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE);
777
778 for (x = 0; x < rlist.rl_rgrps; x++) {
779 struct gfs2_rgrpd *rgd;
780 rgd = rlist.rl_ghs[x].gh_gl->gl_object;
781 rg_blocks += rgd->rd_length;
782 }
783
784 error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs);
785 if (error)
786 goto out_rlist;
787
788 error = gfs2_trans_begin(sdp, rg_blocks + RES_DINODE +
789 RES_INDIRECT + RES_STATFS + RES_QUOTA,
790 revokes);
791 if (error)
792 goto out_rg_gunlock;
793
794 down_write(&ip->i_rw_mutex);
795
796 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
797 gfs2_trans_add_bh(ip->i_gl, bh, 1);
798
799 bstart = 0;
800 blen = 0;
801 btotal = 0;
802
803 for (p = top; p < bottom; p++) {
804 if (!*p)
805 continue;
806
807 bn = be64_to_cpu(*p);
808
809 if (bstart + blen == bn)
810 blen++;
811 else {
812 if (bstart) {
813 __gfs2_free_blocks(ip, bstart, blen, metadata);
814 btotal += blen;
815 }
816
817 bstart = bn;
818 blen = 1;
819 }
820
821 *p = 0;
822 gfs2_add_inode_blocks(&ip->i_inode, -1);
823 }
824 if (bstart) {
825 __gfs2_free_blocks(ip, bstart, blen, metadata);
826 btotal += blen;
827 }
828
829 gfs2_statfs_change(sdp, 0, +btotal, 0);
830 gfs2_quota_change(ip, -(s64)btotal, ip->i_inode.i_uid,
831 ip->i_inode.i_gid);
832
833 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
834
835 gfs2_dinode_out(ip, dibh->b_data);
836
837 up_write(&ip->i_rw_mutex);
838
839 gfs2_trans_end(sdp);
840
841out_rg_gunlock:
842 gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs);
843out_rlist:
844 gfs2_rlist_free(&rlist);
845out:
846 return error;
847}
848
849/**
850 * recursive_scan - recursively scan through the end of a file
851 * @ip: the inode
852 * @dibh: the dinode buffer
853 * @mp: the path through the metadata to the point to start
854 * @height: the height the recursion is at
855 * @block: the indirect block to look at
856 * @first: 1 if this is the first block
857 * @sm: data opaque to this function to pass to @bc
858 *
859 * When this is first called @height and @block should be zero and
860 * @first should be 1.
861 *
862 * Returns: errno
863 */
864
865static int recursive_scan(struct gfs2_inode *ip, struct buffer_head *dibh,
866 struct metapath *mp, unsigned int height,
867 u64 block, int first, struct strip_mine *sm)
868{
869 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
870 struct buffer_head *bh = NULL;
871 __be64 *top, *bottom;
872 u64 bn;
873 int error;
874 int mh_size = sizeof(struct gfs2_meta_header);
875
876 if (!height) {
877 error = gfs2_meta_inode_buffer(ip, &bh);
878 if (error)
879 return error;
880 dibh = bh;
881
882 top = (__be64 *)(bh->b_data + sizeof(struct gfs2_dinode)) + mp->mp_list[0];
883 bottom = (__be64 *)(bh->b_data + sizeof(struct gfs2_dinode)) + sdp->sd_diptrs;
884 } else {
885 error = gfs2_meta_indirect_buffer(ip, height, block, &bh);
886 if (error)
887 return error;
888
889 top = (__be64 *)(bh->b_data + mh_size) +
890 (first ? mp->mp_list[height] : 0);
891
892 bottom = (__be64 *)(bh->b_data + mh_size) + sdp->sd_inptrs;
893 }
894
895 error = do_strip(ip, dibh, bh, top, bottom, height, sm);
896 if (error)
897 goto out;
898
899 if (height < ip->i_height - 1) {
900
901 gfs2_metapath_ra(ip->i_gl, bh, top);
902
903 for (; top < bottom; top++, first = 0) {
904 if (!*top)
905 continue;
906
907 bn = be64_to_cpu(*top);
908
909 error = recursive_scan(ip, dibh, mp, height + 1, bn,
910 first, sm);
911 if (error)
912 break;
913 }
914 }
915out:
916 brelse(bh);
917 return error;
918}
919
920
921/**
922 * gfs2_block_truncate_page - Deal with zeroing out data for truncate
923 *
924 * This is partly borrowed from ext3.
925 */
926static int gfs2_block_truncate_page(struct address_space *mapping, loff_t from)
927{
928 struct inode *inode = mapping->host;
929 struct gfs2_inode *ip = GFS2_I(inode);
930 unsigned long index = from >> PAGE_CACHE_SHIFT;
931 unsigned offset = from & (PAGE_CACHE_SIZE-1);
932 unsigned blocksize, iblock, length, pos;
933 struct buffer_head *bh;
934 struct page *page;
935 int err;
936
937 page = find_or_create_page(mapping, index, GFP_NOFS);
938 if (!page)
939 return 0;
940
941 blocksize = inode->i_sb->s_blocksize;
942 length = blocksize - (offset & (blocksize - 1));
943 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
944
945 if (!page_has_buffers(page))
946 create_empty_buffers(page, blocksize, 0);
947
948 /* Find the buffer that contains "offset" */
949 bh = page_buffers(page);
950 pos = blocksize;
951 while (offset >= pos) {
952 bh = bh->b_this_page;
953 iblock++;
954 pos += blocksize;
955 }
956
957 err = 0;
958
959 if (!buffer_mapped(bh)) {
960 gfs2_block_map(inode, iblock, bh, 0);
961 /* unmapped? It's a hole - nothing to do */
962 if (!buffer_mapped(bh))
963 goto unlock;
964 }
965
966 /* Ok, it's mapped. Make sure it's up-to-date */
967 if (PageUptodate(page))
968 set_buffer_uptodate(bh);
969
970 if (!buffer_uptodate(bh)) {
971 err = -EIO;
972 ll_rw_block(READ, 1, &bh);
973 wait_on_buffer(bh);
974 /* Uhhuh. Read error. Complain and punt. */
975 if (!buffer_uptodate(bh))
976 goto unlock;
977 err = 0;
978 }
979
980 if (!gfs2_is_writeback(ip))
981 gfs2_trans_add_bh(ip->i_gl, bh, 0);
982
983 zero_user(page, offset, length);
984 mark_buffer_dirty(bh);
985unlock:
986 unlock_page(page);
987 page_cache_release(page);
988 return err;
989}
990
991static int trunc_start(struct inode *inode, u64 oldsize, u64 newsize)
992{
993 struct gfs2_inode *ip = GFS2_I(inode);
994 struct gfs2_sbd *sdp = GFS2_SB(inode);
995 struct address_space *mapping = inode->i_mapping;
996 struct buffer_head *dibh;
997 int journaled = gfs2_is_jdata(ip);
998 int error;
999
1000 error = gfs2_trans_begin(sdp,
1001 RES_DINODE + (journaled ? RES_JDATA : 0), 0);
1002 if (error)
1003 return error;
1004
1005 error = gfs2_meta_inode_buffer(ip, &dibh);
1006 if (error)
1007 goto out;
1008
1009 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1010
1011 if (gfs2_is_stuffed(ip)) {
1012 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode) + newsize);
1013 } else {
1014 if (newsize & (u64)(sdp->sd_sb.sb_bsize - 1)) {
1015 error = gfs2_block_truncate_page(mapping, newsize);
1016 if (error)
1017 goto out_brelse;
1018 }
1019 ip->i_diskflags |= GFS2_DIF_TRUNC_IN_PROG;
1020 }
1021
1022 i_size_write(inode, newsize);
1023 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
1024 gfs2_dinode_out(ip, dibh->b_data);
1025
1026 truncate_pagecache(inode, oldsize, newsize);
1027out_brelse:
1028 brelse(dibh);
1029out:
1030 gfs2_trans_end(sdp);
1031 return error;
1032}
1033
1034static int trunc_dealloc(struct gfs2_inode *ip, u64 size)
1035{
1036 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1037 unsigned int height = ip->i_height;
1038 u64 lblock;
1039 struct metapath mp;
1040 int error;
1041
1042 if (!size)
1043 lblock = 0;
1044 else
1045 lblock = (size - 1) >> sdp->sd_sb.sb_bsize_shift;
1046
1047 find_metapath(sdp, lblock, &mp, ip->i_height);
1048 if (!gfs2_qadata_get(ip))
1049 return -ENOMEM;
1050
1051 error = gfs2_quota_hold(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
1052 if (error)
1053 goto out;
1054
1055 while (height--) {
1056 struct strip_mine sm;
1057 sm.sm_first = !!size;
1058 sm.sm_height = height;
1059
1060 error = recursive_scan(ip, NULL, &mp, 0, 0, 1, &sm);
1061 if (error)
1062 break;
1063 }
1064
1065 gfs2_quota_unhold(ip);
1066
1067out:
1068 gfs2_qadata_put(ip);
1069 return error;
1070}
1071
1072static int trunc_end(struct gfs2_inode *ip)
1073{
1074 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1075 struct buffer_head *dibh;
1076 int error;
1077
1078 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1079 if (error)
1080 return error;
1081
1082 down_write(&ip->i_rw_mutex);
1083
1084 error = gfs2_meta_inode_buffer(ip, &dibh);
1085 if (error)
1086 goto out;
1087
1088 if (!i_size_read(&ip->i_inode)) {
1089 ip->i_height = 0;
1090 ip->i_goal = ip->i_no_addr;
1091 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
1092 }
1093 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
1094 ip->i_diskflags &= ~GFS2_DIF_TRUNC_IN_PROG;
1095
1096 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1097 gfs2_dinode_out(ip, dibh->b_data);
1098 brelse(dibh);
1099
1100out:
1101 up_write(&ip->i_rw_mutex);
1102 gfs2_trans_end(sdp);
1103 return error;
1104}
1105
1106/**
1107 * do_shrink - make a file smaller
1108 * @inode: the inode
1109 * @oldsize: the current inode size
1110 * @newsize: the size to make the file
1111 *
1112 * Called with an exclusive lock on @inode. The @size must
1113 * be equal to or smaller than the current inode size.
1114 *
1115 * Returns: errno
1116 */
1117
1118static int do_shrink(struct inode *inode, u64 oldsize, u64 newsize)
1119{
1120 struct gfs2_inode *ip = GFS2_I(inode);
1121 int error;
1122
1123 error = trunc_start(inode, oldsize, newsize);
1124 if (error < 0)
1125 return error;
1126 if (gfs2_is_stuffed(ip))
1127 return 0;
1128
1129 error = trunc_dealloc(ip, newsize);
1130 if (error == 0)
1131 error = trunc_end(ip);
1132
1133 return error;
1134}
1135
1136void gfs2_trim_blocks(struct inode *inode)
1137{
1138 u64 size = inode->i_size;
1139 int ret;
1140
1141 ret = do_shrink(inode, size, size);
1142 WARN_ON(ret != 0);
1143}
1144
1145/**
1146 * do_grow - Touch and update inode size
1147 * @inode: The inode
1148 * @size: The new size
1149 *
1150 * This function updates the timestamps on the inode and
1151 * may also increase the size of the inode. This function
1152 * must not be called with @size any smaller than the current
1153 * inode size.
1154 *
1155 * Although it is not strictly required to unstuff files here,
1156 * earlier versions of GFS2 have a bug in the stuffed file reading
1157 * code which will result in a buffer overrun if the size is larger
1158 * than the max stuffed file size. In order to prevent this from
1159 * occurring, such files are unstuffed, but in other cases we can
1160 * just update the inode size directly.
1161 *
1162 * Returns: 0 on success, or -ve on error
1163 */
1164
1165static int do_grow(struct inode *inode, u64 size)
1166{
1167 struct gfs2_inode *ip = GFS2_I(inode);
1168 struct gfs2_sbd *sdp = GFS2_SB(inode);
1169 struct buffer_head *dibh;
1170 struct gfs2_qadata *qa = NULL;
1171 int error;
1172 int unstuff = 0;
1173
1174 if (gfs2_is_stuffed(ip) &&
1175 (size > (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)))) {
1176 qa = gfs2_qadata_get(ip);
1177 if (qa == NULL)
1178 return -ENOMEM;
1179
1180 error = gfs2_quota_lock_check(ip);
1181 if (error)
1182 goto do_grow_alloc_put;
1183
1184 error = gfs2_inplace_reserve(ip, 1);
1185 if (error)
1186 goto do_grow_qunlock;
1187 unstuff = 1;
1188 }
1189
1190 error = gfs2_trans_begin(sdp, RES_DINODE + RES_STATFS + RES_RG_BIT, 0);
1191 if (error)
1192 goto do_grow_release;
1193
1194 if (unstuff) {
1195 error = gfs2_unstuff_dinode(ip, NULL);
1196 if (error)
1197 goto do_end_trans;
1198 }
1199
1200 error = gfs2_meta_inode_buffer(ip, &dibh);
1201 if (error)
1202 goto do_end_trans;
1203
1204 i_size_write(inode, size);
1205 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
1206 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1207 gfs2_dinode_out(ip, dibh->b_data);
1208 brelse(dibh);
1209
1210do_end_trans:
1211 gfs2_trans_end(sdp);
1212do_grow_release:
1213 if (unstuff) {
1214 gfs2_inplace_release(ip);
1215do_grow_qunlock:
1216 gfs2_quota_unlock(ip);
1217do_grow_alloc_put:
1218 gfs2_qadata_put(ip);
1219 }
1220 return error;
1221}
1222
1223/**
1224 * gfs2_setattr_size - make a file a given size
1225 * @inode: the inode
1226 * @newsize: the size to make the file
1227 *
1228 * The file size can grow, shrink, or stay the same size. This
1229 * is called holding i_mutex and an exclusive glock on the inode
1230 * in question.
1231 *
1232 * Returns: errno
1233 */
1234
1235int gfs2_setattr_size(struct inode *inode, u64 newsize)
1236{
1237 int ret;
1238 u64 oldsize;
1239
1240 BUG_ON(!S_ISREG(inode->i_mode));
1241
1242 ret = inode_newsize_ok(inode, newsize);
1243 if (ret)
1244 return ret;
1245
1246 inode_dio_wait(inode);
1247
1248 oldsize = inode->i_size;
1249 if (newsize >= oldsize)
1250 return do_grow(inode, newsize);
1251
1252 return do_shrink(inode, oldsize, newsize);
1253}
1254
1255int gfs2_truncatei_resume(struct gfs2_inode *ip)
1256{
1257 int error;
1258 error = trunc_dealloc(ip, i_size_read(&ip->i_inode));
1259 if (!error)
1260 error = trunc_end(ip);
1261 return error;
1262}
1263
1264int gfs2_file_dealloc(struct gfs2_inode *ip)
1265{
1266 return trunc_dealloc(ip, 0);
1267}
1268
1269/**
1270 * gfs2_write_alloc_required - figure out if a write will require an allocation
1271 * @ip: the file being written to
1272 * @offset: the offset to write to
1273 * @len: the number of bytes being written
1274 *
1275 * Returns: 1 if an alloc is required, 0 otherwise
1276 */
1277
1278int gfs2_write_alloc_required(struct gfs2_inode *ip, u64 offset,
1279 unsigned int len)
1280{
1281 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1282 struct buffer_head bh;
1283 unsigned int shift;
1284 u64 lblock, lblock_stop, size;
1285 u64 end_of_file;
1286
1287 if (!len)
1288 return 0;
1289
1290 if (gfs2_is_stuffed(ip)) {
1291 if (offset + len >
1292 sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode))
1293 return 1;
1294 return 0;
1295 }
1296
1297 shift = sdp->sd_sb.sb_bsize_shift;
1298 BUG_ON(gfs2_is_dir(ip));
1299 end_of_file = (i_size_read(&ip->i_inode) + sdp->sd_sb.sb_bsize - 1) >> shift;
1300 lblock = offset >> shift;
1301 lblock_stop = (offset + len + sdp->sd_sb.sb_bsize - 1) >> shift;
1302 if (lblock_stop > end_of_file)
1303 return 1;
1304
1305 size = (lblock_stop - lblock) << shift;
1306 do {
1307 bh.b_state = 0;
1308 bh.b_size = size;
1309 gfs2_block_map(&ip->i_inode, lblock, &bh, 0);
1310 if (!buffer_mapped(&bh))
1311 return 1;
1312 size -= bh.b_size;
1313 lblock += (bh.b_size >> ip->i_inode.i_blkbits);
1314 } while(size > 0);
1315
1316 return 0;
1317}
1318