xref: /linux/fs/sysv/itree.c (revision b6ebbac51bedf9e98e837688bc838f400196da5e)
1 /*
2  *  linux/fs/sysv/itree.c
3  *
4  *  Handling of indirect blocks' trees.
5  *  AV, Sep--Dec 2000
6  */
7 
8 #include <linux/buffer_head.h>
9 #include <linux/mount.h>
10 #include <linux/string.h>
11 #include "sysv.h"
12 
13 enum {DIRECT = 10, DEPTH = 4};	/* Have triple indirect */
14 
15 static inline void dirty_indirect(struct buffer_head *bh, struct inode *inode)
16 {
17 	mark_buffer_dirty_inode(bh, inode);
18 	if (IS_SYNC(inode))
19 		sync_dirty_buffer(bh);
20 }
21 
22 static int block_to_path(struct inode *inode, long block, int offsets[DEPTH])
23 {
24 	struct super_block *sb = inode->i_sb;
25 	struct sysv_sb_info *sbi = SYSV_SB(sb);
26 	int ptrs_bits = sbi->s_ind_per_block_bits;
27 	unsigned long	indirect_blocks = sbi->s_ind_per_block,
28 			double_blocks = sbi->s_ind_per_block_2;
29 	int n = 0;
30 
31 	if (block < 0) {
32 		printk("sysv_block_map: block < 0\n");
33 	} else if (block < DIRECT) {
34 		offsets[n++] = block;
35 	} else if ( (block -= DIRECT) < indirect_blocks) {
36 		offsets[n++] = DIRECT;
37 		offsets[n++] = block;
38 	} else if ((block -= indirect_blocks) < double_blocks) {
39 		offsets[n++] = DIRECT+1;
40 		offsets[n++] = block >> ptrs_bits;
41 		offsets[n++] = block & (indirect_blocks - 1);
42 	} else if (((block -= double_blocks) >> (ptrs_bits * 2)) < indirect_blocks) {
43 		offsets[n++] = DIRECT+2;
44 		offsets[n++] = block >> (ptrs_bits * 2);
45 		offsets[n++] = (block >> ptrs_bits) & (indirect_blocks - 1);
46 		offsets[n++] = block & (indirect_blocks - 1);
47 	} else {
48 		/* nothing */;
49 	}
50 	return n;
51 }
52 
53 static inline int block_to_cpu(struct sysv_sb_info *sbi, sysv_zone_t nr)
54 {
55 	return sbi->s_block_base + fs32_to_cpu(sbi, nr);
56 }
57 
58 typedef struct {
59 	sysv_zone_t     *p;
60 	sysv_zone_t     key;
61 	struct buffer_head *bh;
62 } Indirect;
63 
64 static DEFINE_RWLOCK(pointers_lock);
65 
66 static inline void add_chain(Indirect *p, struct buffer_head *bh, sysv_zone_t *v)
67 {
68 	p->key = *(p->p = v);
69 	p->bh = bh;
70 }
71 
72 static inline int verify_chain(Indirect *from, Indirect *to)
73 {
74 	while (from <= to && from->key == *from->p)
75 		from++;
76 	return (from > to);
77 }
78 
79 static inline sysv_zone_t *block_end(struct buffer_head *bh)
80 {
81 	return (sysv_zone_t*)((char*)bh->b_data + bh->b_size);
82 }
83 
84 /*
85  * Requires read_lock(&pointers_lock) or write_lock(&pointers_lock)
86  */
87 static Indirect *get_branch(struct inode *inode,
88 			    int depth,
89 			    int offsets[],
90 			    Indirect chain[],
91 			    int *err)
92 {
93 	struct super_block *sb = inode->i_sb;
94 	Indirect *p = chain;
95 	struct buffer_head *bh;
96 
97 	*err = 0;
98 	add_chain(chain, NULL, SYSV_I(inode)->i_data + *offsets);
99 	if (!p->key)
100 		goto no_block;
101 	while (--depth) {
102 		int block = block_to_cpu(SYSV_SB(sb), p->key);
103 		bh = sb_bread(sb, block);
104 		if (!bh)
105 			goto failure;
106 		if (!verify_chain(chain, p))
107 			goto changed;
108 		add_chain(++p, bh, (sysv_zone_t*)bh->b_data + *++offsets);
109 		if (!p->key)
110 			goto no_block;
111 	}
112 	return NULL;
113 
114 changed:
115 	brelse(bh);
116 	*err = -EAGAIN;
117 	goto no_block;
118 failure:
119 	*err = -EIO;
120 no_block:
121 	return p;
122 }
123 
124 static int alloc_branch(struct inode *inode,
125 			int num,
126 			int *offsets,
127 			Indirect *branch)
128 {
129 	int blocksize = inode->i_sb->s_blocksize;
130 	int n = 0;
131 	int i;
132 
133 	branch[0].key = sysv_new_block(inode->i_sb);
134 	if (branch[0].key) for (n = 1; n < num; n++) {
135 		struct buffer_head *bh;
136 		int parent;
137 		/* Allocate the next block */
138 		branch[n].key = sysv_new_block(inode->i_sb);
139 		if (!branch[n].key)
140 			break;
141 		/*
142 		 * Get buffer_head for parent block, zero it out and set
143 		 * the pointer to new one, then send parent to disk.
144 		 */
145 		parent = block_to_cpu(SYSV_SB(inode->i_sb), branch[n-1].key);
146 		bh = sb_getblk(inode->i_sb, parent);
147 		lock_buffer(bh);
148 		memset(bh->b_data, 0, blocksize);
149 		branch[n].bh = bh;
150 		branch[n].p = (sysv_zone_t*) bh->b_data + offsets[n];
151 		*branch[n].p = branch[n].key;
152 		set_buffer_uptodate(bh);
153 		unlock_buffer(bh);
154 		dirty_indirect(bh, inode);
155 	}
156 	if (n == num)
157 		return 0;
158 
159 	/* Allocation failed, free what we already allocated */
160 	for (i = 1; i < n; i++)
161 		bforget(branch[i].bh);
162 	for (i = 0; i < n; i++)
163 		sysv_free_block(inode->i_sb, branch[i].key);
164 	return -ENOSPC;
165 }
166 
167 static inline int splice_branch(struct inode *inode,
168 				Indirect chain[],
169 				Indirect *where,
170 				int num)
171 {
172 	int i;
173 
174 	/* Verify that place we are splicing to is still there and vacant */
175 	write_lock(&pointers_lock);
176 	if (!verify_chain(chain, where-1) || *where->p)
177 		goto changed;
178 	*where->p = where->key;
179 	write_unlock(&pointers_lock);
180 
181 	inode->i_ctime = CURRENT_TIME_SEC;
182 
183 	/* had we spliced it onto indirect block? */
184 	if (where->bh)
185 		dirty_indirect(where->bh, inode);
186 
187 	if (IS_SYNC(inode))
188 		sysv_sync_inode(inode);
189 	else
190 		mark_inode_dirty(inode);
191 	return 0;
192 
193 changed:
194 	write_unlock(&pointers_lock);
195 	for (i = 1; i < num; i++)
196 		bforget(where[i].bh);
197 	for (i = 0; i < num; i++)
198 		sysv_free_block(inode->i_sb, where[i].key);
199 	return -EAGAIN;
200 }
201 
202 static int get_block(struct inode *inode, sector_t iblock, struct buffer_head *bh_result, int create)
203 {
204 	int err = -EIO;
205 	int offsets[DEPTH];
206 	Indirect chain[DEPTH];
207 	struct super_block *sb = inode->i_sb;
208 	Indirect *partial;
209 	int left;
210 	int depth = block_to_path(inode, iblock, offsets);
211 
212 	if (depth == 0)
213 		goto out;
214 
215 reread:
216 	read_lock(&pointers_lock);
217 	partial = get_branch(inode, depth, offsets, chain, &err);
218 	read_unlock(&pointers_lock);
219 
220 	/* Simplest case - block found, no allocation needed */
221 	if (!partial) {
222 got_it:
223 		map_bh(bh_result, sb, block_to_cpu(SYSV_SB(sb),
224 					chain[depth-1].key));
225 		/* Clean up and exit */
226 		partial = chain+depth-1; /* the whole chain */
227 		goto cleanup;
228 	}
229 
230 	/* Next simple case - plain lookup or failed read of indirect block */
231 	if (!create || err == -EIO) {
232 cleanup:
233 		while (partial > chain) {
234 			brelse(partial->bh);
235 			partial--;
236 		}
237 out:
238 		return err;
239 	}
240 
241 	/*
242 	 * Indirect block might be removed by truncate while we were
243 	 * reading it. Handling of that case (forget what we've got and
244 	 * reread) is taken out of the main path.
245 	 */
246 	if (err == -EAGAIN)
247 		goto changed;
248 
249 	left = (chain + depth) - partial;
250 	err = alloc_branch(inode, left, offsets+(partial-chain), partial);
251 	if (err)
252 		goto cleanup;
253 
254 	if (splice_branch(inode, chain, partial, left) < 0)
255 		goto changed;
256 
257 	set_buffer_new(bh_result);
258 	goto got_it;
259 
260 changed:
261 	while (partial > chain) {
262 		brelse(partial->bh);
263 		partial--;
264 	}
265 	goto reread;
266 }
267 
268 static inline int all_zeroes(sysv_zone_t *p, sysv_zone_t *q)
269 {
270 	while (p < q)
271 		if (*p++)
272 			return 0;
273 	return 1;
274 }
275 
276 static Indirect *find_shared(struct inode *inode,
277 				int depth,
278 				int offsets[],
279 				Indirect chain[],
280 				sysv_zone_t *top)
281 {
282 	Indirect *partial, *p;
283 	int k, err;
284 
285 	*top = 0;
286 	for (k = depth; k > 1 && !offsets[k-1]; k--)
287 		;
288 
289 	write_lock(&pointers_lock);
290 	partial = get_branch(inode, k, offsets, chain, &err);
291 	if (!partial)
292 		partial = chain + k-1;
293 	/*
294 	 * If the branch acquired continuation since we've looked at it -
295 	 * fine, it should all survive and (new) top doesn't belong to us.
296 	 */
297 	if (!partial->key && *partial->p) {
298 		write_unlock(&pointers_lock);
299 		goto no_top;
300 	}
301 	for (p=partial; p>chain && all_zeroes((sysv_zone_t*)p->bh->b_data,p->p); p--)
302 		;
303 	/*
304 	 * OK, we've found the last block that must survive. The rest of our
305 	 * branch should be detached before unlocking. However, if that rest
306 	 * of branch is all ours and does not grow immediately from the inode
307 	 * it's easier to cheat and just decrement partial->p.
308 	 */
309 	if (p == chain + k - 1 && p > chain) {
310 		p->p--;
311 	} else {
312 		*top = *p->p;
313 		*p->p = 0;
314 	}
315 	write_unlock(&pointers_lock);
316 
317 	while (partial > p) {
318 		brelse(partial->bh);
319 		partial--;
320 	}
321 no_top:
322 	return partial;
323 }
324 
325 static inline void free_data(struct inode *inode, sysv_zone_t *p, sysv_zone_t *q)
326 {
327 	for ( ; p < q ; p++) {
328 		sysv_zone_t nr = *p;
329 		if (nr) {
330 			*p = 0;
331 			sysv_free_block(inode->i_sb, nr);
332 			mark_inode_dirty(inode);
333 		}
334 	}
335 }
336 
337 static void free_branches(struct inode *inode, sysv_zone_t *p, sysv_zone_t *q, int depth)
338 {
339 	struct buffer_head * bh;
340 	struct super_block *sb = inode->i_sb;
341 
342 	if (depth--) {
343 		for ( ; p < q ; p++) {
344 			int block;
345 			sysv_zone_t nr = *p;
346 			if (!nr)
347 				continue;
348 			*p = 0;
349 			block = block_to_cpu(SYSV_SB(sb), nr);
350 			bh = sb_bread(sb, block);
351 			if (!bh)
352 				continue;
353 			free_branches(inode, (sysv_zone_t*)bh->b_data,
354 					block_end(bh), depth);
355 			bforget(bh);
356 			sysv_free_block(sb, nr);
357 			mark_inode_dirty(inode);
358 		}
359 	} else
360 		free_data(inode, p, q);
361 }
362 
363 void sysv_truncate (struct inode * inode)
364 {
365 	sysv_zone_t *i_data = SYSV_I(inode)->i_data;
366 	int offsets[DEPTH];
367 	Indirect chain[DEPTH];
368 	Indirect *partial;
369 	sysv_zone_t nr = 0;
370 	int n;
371 	long iblock;
372 	unsigned blocksize;
373 
374 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
375 	    S_ISLNK(inode->i_mode)))
376 		return;
377 
378 	blocksize = inode->i_sb->s_blocksize;
379 	iblock = (inode->i_size + blocksize-1)
380 					>> inode->i_sb->s_blocksize_bits;
381 
382 	block_truncate_page(inode->i_mapping, inode->i_size, get_block);
383 
384 	n = block_to_path(inode, iblock, offsets);
385 	if (n == 0)
386 		return;
387 
388 	if (n == 1) {
389 		free_data(inode, i_data+offsets[0], i_data + DIRECT);
390 		goto do_indirects;
391 	}
392 
393 	partial = find_shared(inode, n, offsets, chain, &nr);
394 	/* Kill the top of shared branch (already detached) */
395 	if (nr) {
396 		if (partial == chain)
397 			mark_inode_dirty(inode);
398 		else
399 			dirty_indirect(partial->bh, inode);
400 		free_branches(inode, &nr, &nr+1, (chain+n-1) - partial);
401 	}
402 	/* Clear the ends of indirect blocks on the shared branch */
403 	while (partial > chain) {
404 		free_branches(inode, partial->p + 1, block_end(partial->bh),
405 				(chain+n-1) - partial);
406 		dirty_indirect(partial->bh, inode);
407 		brelse (partial->bh);
408 		partial--;
409 	}
410 do_indirects:
411 	/* Kill the remaining (whole) subtrees (== subtrees deeper than...) */
412 	while (n < DEPTH) {
413 		nr = i_data[DIRECT + n - 1];
414 		if (nr) {
415 			i_data[DIRECT + n - 1] = 0;
416 			mark_inode_dirty(inode);
417 			free_branches(inode, &nr, &nr+1, n);
418 		}
419 		n++;
420 	}
421 	inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
422 	if (IS_SYNC(inode))
423 		sysv_sync_inode (inode);
424 	else
425 		mark_inode_dirty(inode);
426 }
427 
428 static unsigned sysv_nblocks(struct super_block *s, loff_t size)
429 {
430 	struct sysv_sb_info *sbi = SYSV_SB(s);
431 	int ptrs_bits = sbi->s_ind_per_block_bits;
432 	unsigned blocks, res, direct = DIRECT, i = DEPTH;
433 	blocks = (size + s->s_blocksize - 1) >> s->s_blocksize_bits;
434 	res = blocks;
435 	while (--i && blocks > direct) {
436 		blocks = ((blocks - direct - 1) >> ptrs_bits) + 1;
437 		res += blocks;
438 		direct = 1;
439 	}
440 	return blocks;
441 }
442 
443 int sysv_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
444 {
445 	struct super_block *s = dentry->d_sb;
446 	generic_fillattr(d_inode(dentry), stat);
447 	stat->blocks = (s->s_blocksize / 512) * sysv_nblocks(s, stat->size);
448 	stat->blksize = s->s_blocksize;
449 	return 0;
450 }
451 
452 static int sysv_writepage(struct page *page, struct writeback_control *wbc)
453 {
454 	return block_write_full_page(page,get_block,wbc);
455 }
456 
457 static int sysv_readpage(struct file *file, struct page *page)
458 {
459 	return block_read_full_page(page,get_block);
460 }
461 
462 int sysv_prepare_chunk(struct page *page, loff_t pos, unsigned len)
463 {
464 	return __block_write_begin(page, pos, len, get_block);
465 }
466 
467 static void sysv_write_failed(struct address_space *mapping, loff_t to)
468 {
469 	struct inode *inode = mapping->host;
470 
471 	if (to > inode->i_size) {
472 		truncate_pagecache(inode, inode->i_size);
473 		sysv_truncate(inode);
474 	}
475 }
476 
477 static int sysv_write_begin(struct file *file, struct address_space *mapping,
478 			loff_t pos, unsigned len, unsigned flags,
479 			struct page **pagep, void **fsdata)
480 {
481 	int ret;
482 
483 	ret = block_write_begin(mapping, pos, len, flags, pagep, get_block);
484 	if (unlikely(ret))
485 		sysv_write_failed(mapping, pos + len);
486 
487 	return ret;
488 }
489 
490 static sector_t sysv_bmap(struct address_space *mapping, sector_t block)
491 {
492 	return generic_block_bmap(mapping,block,get_block);
493 }
494 
495 const struct address_space_operations sysv_aops = {
496 	.readpage = sysv_readpage,
497 	.writepage = sysv_writepage,
498 	.write_begin = sysv_write_begin,
499 	.write_end = generic_write_end,
500 	.bmap = sysv_bmap
501 };
502