xref: /openbmc/linux/fs/sysv/itree.c (revision 297ce026) 
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  linux/fs/sysv/itree.c
4  *
5  *  Handling of indirect blocks' trees.
6  *  AV, Sep--Dec 2000
7  */
8 
9 #include <linux/buffer_head.h>
10 #include <linux/mount.h>
11 #include <linux/string.h>
12 #include "sysv.h"
13 
14 enum {DIRECT = 10, DEPTH = 4};	/* Have triple indirect */
15 
16 static inline void dirty_indirect(struct buffer_head *bh, struct inode *inode)
17 {
18 	mark_buffer_dirty_inode(bh, inode);
19 	if (IS_SYNC(inode))
20 		sync_dirty_buffer(bh);
21 }
22 
23 static int block_to_path(struct inode *inode, long block, int offsets[DEPTH])
24 {
25 	struct super_block *sb = inode->i_sb;
26 	struct sysv_sb_info *sbi = SYSV_SB(sb);
27 	int ptrs_bits = sbi->s_ind_per_block_bits;
28 	unsigned long	indirect_blocks = sbi->s_ind_per_block,
29 			double_blocks = sbi->s_ind_per_block_2;
30 	int n = 0;
31 
32 	if (block < 0) {
33 		printk("sysv_block_map: block < 0\n");
34 	} else if (block < DIRECT) {
35 		offsets[n++] = block;
36 	} else if ( (block -= DIRECT) < indirect_blocks) {
37 		offsets[n++] = DIRECT;
38 		offsets[n++] = block;
39 	} else if ((block -= indirect_blocks) < double_blocks) {
40 		offsets[n++] = DIRECT+1;
41 		offsets[n++] = block >> ptrs_bits;
42 		offsets[n++] = block & (indirect_blocks - 1);
43 	} else if (((block -= double_blocks) >> (ptrs_bits * 2)) < indirect_blocks) {
44 		offsets[n++] = DIRECT+2;
45 		offsets[n++] = block >> (ptrs_bits * 2);
46 		offsets[n++] = (block >> ptrs_bits) & (indirect_blocks - 1);
47 		offsets[n++] = block & (indirect_blocks - 1);
48 	} else {
49 		/* nothing */;
50 	}
51 	return n;
52 }
53 
54 static inline int block_to_cpu(struct sysv_sb_info *sbi, sysv_zone_t nr)
55 {
56 	return sbi->s_block_base + fs32_to_cpu(sbi, nr);
57 }
58 
59 typedef struct {
60 	sysv_zone_t     *p;
61 	sysv_zone_t     key;
62 	struct buffer_head *bh;
63 } Indirect;
64 
65 static DEFINE_RWLOCK(pointers_lock);
66 
67 static inline void add_chain(Indirect *p, struct buffer_head *bh, sysv_zone_t *v)
68 {
69 	p->key = *(p->p = v);
70 	p->bh = bh;
71 }
72 
73 static inline int verify_chain(Indirect *from, Indirect *to)
74 {
75 	while (from <= to && from->key == *from->p)
76 		from++;
77 	return (from > to);
78 }
79 
80 static inline sysv_zone_t *block_end(struct buffer_head *bh)
81 {
82 	return (sysv_zone_t*)((char*)bh->b_data + bh->b_size);
83 }
84 
85 /*
86  * Requires read_lock(&pointers_lock) or write_lock(&pointers_lock)
87  */
88 static Indirect *get_branch(struct inode *inode,
89 			    int depth,
90 			    int offsets[],
91 			    Indirect chain[],
92 			    int *err)
93 {
94 	struct super_block *sb = inode->i_sb;
95 	Indirect *p = chain;
96 	struct buffer_head *bh;
97 
98 	*err = 0;
99 	add_chain(chain, NULL, SYSV_I(inode)->i_data + *offsets);
100 	if (!p->key)
101 		goto no_block;
102 	while (--depth) {
103 		int block = block_to_cpu(SYSV_SB(sb), p->key);
104 		bh = sb_bread(sb, block);
105 		if (!bh)
106 			goto failure;
107 		if (!verify_chain(chain, p))
108 			goto changed;
109 		add_chain(++p, bh, (sysv_zone_t*)bh->b_data + *++offsets);
110 		if (!p->key)
111 			goto no_block;
112 	}
113 	return NULL;
114 
115 changed:
116 	brelse(bh);
117 	*err = -EAGAIN;
118 	goto no_block;
119 failure:
120 	*err = -EIO;
121 no_block:
122 	return p;
123 }
124 
125 static int alloc_branch(struct inode *inode,
126 			int num,
127 			int *offsets,
128 			Indirect *branch)
129 {
130 	int blocksize = inode->i_sb->s_blocksize;
131 	int n = 0;
132 	int i;
133 
134 	branch[0].key = sysv_new_block(inode->i_sb);
135 	if (branch[0].key) for (n = 1; n < num; n++) {
136 		struct buffer_head *bh;
137 		int parent;
138 		/* Allocate the next block */
139 		branch[n].key = sysv_new_block(inode->i_sb);
140 		if (!branch[n].key)
141 			break;
142 		/*
143 		 * Get buffer_head for parent block, zero it out and set
144 		 * the pointer to new one, then send parent to disk.
145 		 */
146 		parent = block_to_cpu(SYSV_SB(inode->i_sb), branch[n-1].key);
147 		bh = sb_getblk(inode->i_sb, parent);
148 		lock_buffer(bh);
149 		memset(bh->b_data, 0, blocksize);
150 		branch[n].bh = bh;
151 		branch[n].p = (sysv_zone_t*) bh->b_data + offsets[n];
152 		*branch[n].p = branch[n].key;
153 		set_buffer_uptodate(bh);
154 		unlock_buffer(bh);
155 		dirty_indirect(bh, inode);
156 	}
157 	if (n == num)
158 		return 0;
159 
160 	/* Allocation failed, free what we already allocated */
161 	for (i = 1; i < n; i++)
162 		bforget(branch[i].bh);
163 	for (i = 0; i < n; i++)
164 		sysv_free_block(inode->i_sb, branch[i].key);
165 	return -ENOSPC;
166 }
167 
168 static inline int splice_branch(struct inode *inode,
169 				Indirect chain[],
170 				Indirect *where,
171 				int num)
172 {
173 	int i;
174 
175 	/* Verify that place we are splicing to is still there and vacant */
176 	write_lock(&pointers_lock);
177 	if (!verify_chain(chain, where-1) || *where->p)
178 		goto changed;
179 	*where->p = where->key;
180 	write_unlock(&pointers_lock);
181 
182 	inode->i_ctime = current_time(inode);
183 
184 	/* had we spliced it onto indirect block? */
185 	if (where->bh)
186 		dirty_indirect(where->bh, inode);
187 
188 	if (IS_SYNC(inode))
189 		sysv_sync_inode(inode);
190 	else
191 		mark_inode_dirty(inode);
192 	return 0;
193 
194 changed:
195 	write_unlock(&pointers_lock);
196 	for (i = 1; i < num; i++)
197 		bforget(where[i].bh);
198 	for (i = 0; i < num; i++)
199 		sysv_free_block(inode->i_sb, where[i].key);
200 	return -EAGAIN;
201 }
202 
203 static int get_block(struct inode *inode, sector_t iblock, struct buffer_head *bh_result, int create)
204 {
205 	int err = -EIO;
206 	int offsets[DEPTH];
207 	Indirect chain[DEPTH];
208 	struct super_block *sb = inode->i_sb;
209 	Indirect *partial;
210 	int left;
211 	int depth = block_to_path(inode, iblock, offsets);
212 
213 	if (depth == 0)
214 		goto out;
215 
216 reread:
217 	read_lock(&pointers_lock);
218 	partial = get_branch(inode, depth, offsets, chain, &err);
219 	read_unlock(&pointers_lock);
220 
221 	/* Simplest case - block found, no allocation needed */
222 	if (!partial) {
223 got_it:
224 		map_bh(bh_result, sb, block_to_cpu(SYSV_SB(sb),
225 					chain[depth-1].key));
226 		/* Clean up and exit */
227 		partial = chain+depth-1; /* the whole chain */
228 		goto cleanup;
229 	}
230 
231 	/* Next simple case - plain lookup or failed read of indirect block */
232 	if (!create || err == -EIO) {
233 cleanup:
234 		while (partial > chain) {
235 			brelse(partial->bh);
236 			partial--;
237 		}
238 out:
239 		return err;
240 	}
241 
242 	/*
243 	 * Indirect block might be removed by truncate while we were
244 	 * reading it. Handling of that case (forget what we've got and
245 	 * reread) is taken out of the main path.
246 	 */
247 	if (err == -EAGAIN)
248 		goto changed;
249 
250 	left = (chain + depth) - partial;
251 	err = alloc_branch(inode, left, offsets+(partial-chain), partial);
252 	if (err)
253 		goto cleanup;
254 
255 	if (splice_branch(inode, chain, partial, left) < 0)
256 		goto changed;
257 
258 	set_buffer_new(bh_result);
259 	goto got_it;
260 
261 changed:
262 	while (partial > chain) {
263 		brelse(partial->bh);
264 		partial--;
265 	}
266 	goto reread;
267 }
268 
269 static inline int all_zeroes(sysv_zone_t *p, sysv_zone_t *q)
270 {
271 	while (p < q)
272 		if (*p++)
273 			return 0;
274 	return 1;
275 }
276 
277 static Indirect *find_shared(struct inode *inode,
278 				int depth,
279 				int offsets[],
280 				Indirect chain[],
281 				sysv_zone_t *top)
282 {
283 	Indirect *partial, *p;
284 	int k, err;
285 
286 	*top = 0;
287 	for (k = depth; k > 1 && !offsets[k-1]; k--)
288 		;
289 
290 	write_lock(&pointers_lock);
291 	partial = get_branch(inode, k, offsets, chain, &err);
292 	if (!partial)
293 		partial = chain + k-1;
294 	/*
295 	 * If the branch acquired continuation since we've looked at it -
296 	 * fine, it should all survive and (new) top doesn't belong to us.
297 	 */
298 	if (!partial->key && *partial->p) {
299 		write_unlock(&pointers_lock);
300 		goto no_top;
301 	}
302 	for (p=partial; p>chain && all_zeroes((sysv_zone_t*)p->bh->b_data,p->p); p--)
303 		;
304 	/*
305 	 * OK, we've found the last block that must survive. The rest of our
306 	 * branch should be detached before unlocking. However, if that rest
307 	 * of branch is all ours and does not grow immediately from the inode
308 	 * it's easier to cheat and just decrement partial->p.
309 	 */
310 	if (p == chain + k - 1 && p > chain) {
311 		p->p--;
312 	} else {
313 		*top = *p->p;
314 		*p->p = 0;
315 	}
316 	write_unlock(&pointers_lock);
317 
318 	while (partial > p) {
319 		brelse(partial->bh);
320 		partial--;
321 	}
322 no_top:
323 	return partial;
324 }
325 
326 static inline void free_data(struct inode *inode, sysv_zone_t *p, sysv_zone_t *q)
327 {
328 	for ( ; p < q ; p++) {
329 		sysv_zone_t nr = *p;
330 		if (nr) {
331 			*p = 0;
332 			sysv_free_block(inode->i_sb, nr);
333 			mark_inode_dirty(inode);
334 		}
335 	}
336 }
337 
338 static void free_branches(struct inode *inode, sysv_zone_t *p, sysv_zone_t *q, int depth)
339 {
340 	struct buffer_head * bh;
341 	struct super_block *sb = inode->i_sb;
342 
343 	if (depth--) {
344 		for ( ; p < q ; p++) {
345 			int block;
346 			sysv_zone_t nr = *p;
347 			if (!nr)
348 				continue;
349 			*p = 0;
350 			block = block_to_cpu(SYSV_SB(sb), nr);
351 			bh = sb_bread(sb, block);
352 			if (!bh)
353 				continue;
354 			free_branches(inode, (sysv_zone_t*)bh->b_data,
355 					block_end(bh), depth);
356 			bforget(bh);
357 			sysv_free_block(sb, nr);
358 			mark_inode_dirty(inode);
359 		}
360 	} else
361 		free_data(inode, p, q);
362 }
363 
364 void sysv_truncate (struct inode * inode)
365 {
366 	sysv_zone_t *i_data = SYSV_I(inode)->i_data;
367 	int offsets[DEPTH];
368 	Indirect chain[DEPTH];
369 	Indirect *partial;
370 	sysv_zone_t nr = 0;
371 	int n;
372 	long iblock;
373 	unsigned blocksize;
374 
375 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
376 	    S_ISLNK(inode->i_mode)))
377 		return;
378 
379 	blocksize = inode->i_sb->s_blocksize;
380 	iblock = (inode->i_size + blocksize-1)
381 					>> inode->i_sb->s_blocksize_bits;
382 
383 	block_truncate_page(inode->i_mapping, inode->i_size, get_block);
384 
385 	n = block_to_path(inode, iblock, offsets);
386 	if (n == 0)
387 		return;
388 
389 	if (n == 1) {
390 		free_data(inode, i_data+offsets[0], i_data + DIRECT);
391 		goto do_indirects;
392 	}
393 
394 	partial = find_shared(inode, n, offsets, chain, &nr);
395 	/* Kill the top of shared branch (already detached) */
396 	if (nr) {
397 		if (partial == chain)
398 			mark_inode_dirty(inode);
399 		else
400 			dirty_indirect(partial->bh, inode);
401 		free_branches(inode, &nr, &nr+1, (chain+n-1) - partial);
402 	}
403 	/* Clear the ends of indirect blocks on the shared branch */
404 	while (partial > chain) {
405 		free_branches(inode, partial->p + 1, block_end(partial->bh),
406 				(chain+n-1) - partial);
407 		dirty_indirect(partial->bh, inode);
408 		brelse (partial->bh);
409 		partial--;
410 	}
411 do_indirects:
412 	/* Kill the remaining (whole) subtrees (== subtrees deeper than...) */
413 	while (n < DEPTH) {
414 		nr = i_data[DIRECT + n - 1];
415 		if (nr) {
416 			i_data[DIRECT + n - 1] = 0;
417 			mark_inode_dirty(inode);
418 			free_branches(inode, &nr, &nr+1, n);
419 		}
420 		n++;
421 	}
422 	inode->i_mtime = inode->i_ctime = current_time(inode);
423 	if (IS_SYNC(inode))
424 		sysv_sync_inode (inode);
425 	else
426 		mark_inode_dirty(inode);
427 }
428 
429 static unsigned sysv_nblocks(struct super_block *s, loff_t size)
430 {
431 	struct sysv_sb_info *sbi = SYSV_SB(s);
432 	int ptrs_bits = sbi->s_ind_per_block_bits;
433 	unsigned blocks, res, direct = DIRECT, i = DEPTH;
434 	blocks = (size + s->s_blocksize - 1) >> s->s_blocksize_bits;
435 	res = blocks;
436 	while (--i && blocks > direct) {
437 		blocks = ((blocks - direct - 1) >> ptrs_bits) + 1;
438 		res += blocks;
439 		direct = 1;
440 	}
441 	return blocks;
442 }
443 
444 int sysv_getattr(struct user_namespace *mnt_userns, const struct path *path,
445 		 struct kstat *stat, u32 request_mask, unsigned int flags)
446 {
447 	struct super_block *s = path->dentry->d_sb;
448 	generic_fillattr(&init_user_ns, d_inode(path->dentry), stat);
449 	stat->blocks = (s->s_blocksize / 512) * sysv_nblocks(s, stat->size);
450 	stat->blksize = s->s_blocksize;
451 	return 0;
452 }
453 
454 static int sysv_writepage(struct page *page, struct writeback_control *wbc)
455 {
456 	return block_write_full_page(page,get_block,wbc);
457 }
458 
459 static int sysv_readpage(struct file *file, struct page *page)
460 {
461 	return block_read_full_page(page,get_block);
462 }
463 
464 int sysv_prepare_chunk(struct page *page, loff_t pos, unsigned len)
465 {
466 	return __block_write_begin(page, pos, len, get_block);
467 }
468 
469 static void sysv_write_failed(struct address_space *mapping, loff_t to)
470 {
471 	struct inode *inode = mapping->host;
472 
473 	if (to > inode->i_size) {
474 		truncate_pagecache(inode, inode->i_size);
475 		sysv_truncate(inode);
476 	}
477 }
478 
479 static int sysv_write_begin(struct file *file, struct address_space *mapping,
480 			loff_t pos, unsigned len, unsigned flags,
481 			struct page **pagep, void **fsdata)
482 {
483 	int ret;
484 
485 	ret = block_write_begin(mapping, pos, len, flags, pagep, get_block);
486 	if (unlikely(ret))
487 		sysv_write_failed(mapping, pos + len);
488 
489 	return ret;
490 }
491 
492 static sector_t sysv_bmap(struct address_space *mapping, sector_t block)
493 {
494 	return generic_block_bmap(mapping,block,get_block);
495 }
496 
497 const struct address_space_operations sysv_aops = {
498 	.dirty_folio = block_dirty_folio,
499 	.invalidate_folio = block_invalidate_folio,
500 	.readpage = sysv_readpage,
501 	.writepage = sysv_writepage,
502 	.write_begin = sysv_write_begin,
503 	.write_end = generic_write_end,
504 	.bmap = sysv_bmap
505 };
506