xref: /linux/fs/ocfs2/blockcheck.c (revision 8be98d2f2a0a262f8bf8a0bc1fdf522b3c7aab17)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * blockcheck.c
4  *
5  * Checksum and ECC codes for the OCFS2 userspace library.
6  *
7  * Copyright (C) 2006, 2008 Oracle.  All rights reserved.
8  */
9 
10 #include <linux/kernel.h>
11 #include <linux/types.h>
12 #include <linux/crc32.h>
13 #include <linux/buffer_head.h>
14 #include <linux/bitops.h>
15 #include <linux/debugfs.h>
16 #include <linux/module.h>
17 #include <linux/fs.h>
18 #include <asm/byteorder.h>
19 
20 #include <cluster/masklog.h>
21 
22 #include "ocfs2.h"
23 
24 #include "blockcheck.h"
25 
26 
27 /*
28  * We use the following conventions:
29  *
30  * d = # data bits
31  * p = # parity bits
32  * c = # total code bits (d + p)
33  */
34 
35 
36 /*
37  * Calculate the bit offset in the hamming code buffer based on the bit's
38  * offset in the data buffer.  Since the hamming code reserves all
39  * power-of-two bits for parity, the data bit number and the code bit
40  * number are offset by all the parity bits beforehand.
41  *
42  * Recall that bit numbers in hamming code are 1-based.  This function
43  * takes the 0-based data bit from the caller.
44  *
45  * An example.  Take bit 1 of the data buffer.  1 is a power of two (2^0),
46  * so it's a parity bit.  2 is a power of two (2^1), so it's a parity bit.
47  * 3 is not a power of two.  So bit 1 of the data buffer ends up as bit 3
48  * in the code buffer.
49  *
50  * The caller can pass in *p if it wants to keep track of the most recent
51  * number of parity bits added.  This allows the function to start the
52  * calculation at the last place.
53  */
calc_code_bit(unsigned int i,unsigned int * p_cache)54 static unsigned int calc_code_bit(unsigned int i, unsigned int *p_cache)
55 {
56 	unsigned int b, p = 0;
57 
58 	/*
59 	 * Data bits are 0-based, but we're talking code bits, which
60 	 * are 1-based.
61 	 */
62 	b = i + 1;
63 
64 	/* Use the cache if it is there */
65 	if (p_cache)
66 		p = *p_cache;
67         b += p;
68 
69 	/*
70 	 * For every power of two below our bit number, bump our bit.
71 	 *
72 	 * We compare with (b + 1) because we have to compare with what b
73 	 * would be _if_ it were bumped up by the parity bit.  Capice?
74 	 *
75 	 * p is set above.
76 	 */
77 	for (; (1 << p) < (b + 1); p++)
78 		b++;
79 
80 	if (p_cache)
81 		*p_cache = p;
82 
83 	return b;
84 }
85 
86 /*
87  * This is the low level encoder function.  It can be called across
88  * multiple hunks just like the crc32 code.  'd' is the number of bits
89  * _in_this_hunk_.  nr is the bit offset of this hunk.  So, if you had
90  * two 512B buffers, you would do it like so:
91  *
92  * parity = ocfs2_hamming_encode(0, buf1, 512 * 8, 0);
93  * parity = ocfs2_hamming_encode(parity, buf2, 512 * 8, 512 * 8);
94  *
95  * If you just have one buffer, use ocfs2_hamming_encode_block().
96  */
ocfs2_hamming_encode(u32 parity,void * data,unsigned int d,unsigned int nr)97 u32 ocfs2_hamming_encode(u32 parity, void *data, unsigned int d, unsigned int nr)
98 {
99 	unsigned int i, b, p = 0;
100 
101 	BUG_ON(!d);
102 
103 	/*
104 	 * b is the hamming code bit number.  Hamming code specifies a
105 	 * 1-based array, but C uses 0-based.  So 'i' is for C, and 'b' is
106 	 * for the algorithm.
107 	 *
108 	 * The i++ in the for loop is so that the start offset passed
109 	 * to ocfs2_find_next_bit_set() is one greater than the previously
110 	 * found bit.
111 	 */
112 	for (i = 0; (i = ocfs2_find_next_bit(data, d, i)) < d; i++)
113 	{
114 		/*
115 		 * i is the offset in this hunk, nr + i is the total bit
116 		 * offset.
117 		 */
118 		b = calc_code_bit(nr + i, &p);
119 
120 		/*
121 		 * Data bits in the resultant code are checked by
122 		 * parity bits that are part of the bit number
123 		 * representation.  Huh?
124 		 *
125 		 * <wikipedia href="https://en.wikipedia.org/wiki/Hamming_code">
126 		 * In other words, the parity bit at position 2^k
127 		 * checks bits in positions having bit k set in
128 		 * their binary representation.  Conversely, for
129 		 * instance, bit 13, i.e. 1101(2), is checked by
130 		 * bits 1000(2) = 8, 0100(2)=4 and 0001(2) = 1.
131 		 * </wikipedia>
132 		 *
133 		 * Note that 'k' is the _code_ bit number.  'b' in
134 		 * our loop.
135 		 */
136 		parity ^= b;
137 	}
138 
139 	/* While the data buffer was treated as little endian, the
140 	 * return value is in host endian. */
141 	return parity;
142 }
143 
ocfs2_hamming_encode_block(void * data,unsigned int blocksize)144 u32 ocfs2_hamming_encode_block(void *data, unsigned int blocksize)
145 {
146 	return ocfs2_hamming_encode(0, data, blocksize * 8, 0);
147 }
148 
149 /*
150  * Like ocfs2_hamming_encode(), this can handle hunks.  nr is the bit
151  * offset of the current hunk.  If bit to be fixed is not part of the
152  * current hunk, this does nothing.
153  *
154  * If you only have one hunk, use ocfs2_hamming_fix_block().
155  */
ocfs2_hamming_fix(void * data,unsigned int d,unsigned int nr,unsigned int fix)156 void ocfs2_hamming_fix(void *data, unsigned int d, unsigned int nr,
157 		       unsigned int fix)
158 {
159 	unsigned int i, b;
160 
161 	BUG_ON(!d);
162 
163 	/*
164 	 * If the bit to fix has an hweight of 1, it's a parity bit.  One
165 	 * busted parity bit is its own error.  Nothing to do here.
166 	 */
167 	if (hweight32(fix) == 1)
168 		return;
169 
170 	/*
171 	 * nr + d is the bit right past the data hunk we're looking at.
172 	 * If fix after that, nothing to do
173 	 */
174 	if (fix >= calc_code_bit(nr + d, NULL))
175 		return;
176 
177 	/*
178 	 * nr is the offset in the data hunk we're starting at.  Let's
179 	 * start b at the offset in the code buffer.  See hamming_encode()
180 	 * for a more detailed description of 'b'.
181 	 */
182 	b = calc_code_bit(nr, NULL);
183 	/* If the fix is before this hunk, nothing to do */
184 	if (fix < b)
185 		return;
186 
187 	for (i = 0; i < d; i++, b++)
188 	{
189 		/* Skip past parity bits */
190 		while (hweight32(b) == 1)
191 			b++;
192 
193 		/*
194 		 * i is the offset in this data hunk.
195 		 * nr + i is the offset in the total data buffer.
196 		 * b is the offset in the total code buffer.
197 		 *
198 		 * Thus, when b == fix, bit i in the current hunk needs
199 		 * fixing.
200 		 */
201 		if (b == fix)
202 		{
203 			if (ocfs2_test_bit(i, data))
204 				ocfs2_clear_bit(i, data);
205 			else
206 				ocfs2_set_bit(i, data);
207 			break;
208 		}
209 	}
210 }
211 
ocfs2_hamming_fix_block(void * data,unsigned int blocksize,unsigned int fix)212 void ocfs2_hamming_fix_block(void *data, unsigned int blocksize,
213 			     unsigned int fix)
214 {
215 	ocfs2_hamming_fix(data, blocksize * 8, 0, fix);
216 }
217 
218 
219 /*
220  * Debugfs handling.
221  */
222 
223 #ifdef CONFIG_DEBUG_FS
224 
blockcheck_u64_get(void * data,u64 * val)225 static int blockcheck_u64_get(void *data, u64 *val)
226 {
227 	*val = *(u64 *)data;
228 	return 0;
229 }
230 DEFINE_DEBUGFS_ATTRIBUTE(blockcheck_fops, blockcheck_u64_get, NULL, "%llu\n");
231 
ocfs2_blockcheck_debug_remove(struct ocfs2_blockcheck_stats * stats)232 static void ocfs2_blockcheck_debug_remove(struct ocfs2_blockcheck_stats *stats)
233 {
234 	if (stats) {
235 		debugfs_remove_recursive(stats->b_debug_dir);
236 		stats->b_debug_dir = NULL;
237 	}
238 }
239 
ocfs2_blockcheck_debug_install(struct ocfs2_blockcheck_stats * stats,struct dentry * parent)240 static void ocfs2_blockcheck_debug_install(struct ocfs2_blockcheck_stats *stats,
241 					   struct dentry *parent)
242 {
243 	struct dentry *dir;
244 
245 	dir = debugfs_create_dir("blockcheck", parent);
246 	stats->b_debug_dir = dir;
247 
248 	debugfs_create_file("blocks_checked", S_IFREG | S_IRUSR, dir,
249 			    &stats->b_check_count, &blockcheck_fops);
250 
251 	debugfs_create_file("checksums_failed", S_IFREG | S_IRUSR, dir,
252 			    &stats->b_failure_count, &blockcheck_fops);
253 
254 	debugfs_create_file("ecc_recoveries", S_IFREG | S_IRUSR, dir,
255 			    &stats->b_recover_count, &blockcheck_fops);
256 
257 }
258 #else
ocfs2_blockcheck_debug_install(struct ocfs2_blockcheck_stats * stats,struct dentry * parent)259 static inline void ocfs2_blockcheck_debug_install(struct ocfs2_blockcheck_stats *stats,
260 						  struct dentry *parent)
261 {
262 }
263 
ocfs2_blockcheck_debug_remove(struct ocfs2_blockcheck_stats * stats)264 static inline void ocfs2_blockcheck_debug_remove(struct ocfs2_blockcheck_stats *stats)
265 {
266 }
267 #endif  /* CONFIG_DEBUG_FS */
268 
269 /* Always-called wrappers for starting and stopping the debugfs files */
ocfs2_blockcheck_stats_debugfs_install(struct ocfs2_blockcheck_stats * stats,struct dentry * parent)270 void ocfs2_blockcheck_stats_debugfs_install(struct ocfs2_blockcheck_stats *stats,
271 					    struct dentry *parent)
272 {
273 	ocfs2_blockcheck_debug_install(stats, parent);
274 }
275 
ocfs2_blockcheck_stats_debugfs_remove(struct ocfs2_blockcheck_stats * stats)276 void ocfs2_blockcheck_stats_debugfs_remove(struct ocfs2_blockcheck_stats *stats)
277 {
278 	ocfs2_blockcheck_debug_remove(stats);
279 }
280 
ocfs2_blockcheck_inc_check(struct ocfs2_blockcheck_stats * stats)281 static void ocfs2_blockcheck_inc_check(struct ocfs2_blockcheck_stats *stats)
282 {
283 	u64 new_count;
284 
285 	if (!stats)
286 		return;
287 
288 	spin_lock(&stats->b_lock);
289 	stats->b_check_count++;
290 	new_count = stats->b_check_count;
291 	spin_unlock(&stats->b_lock);
292 
293 	if (!new_count)
294 		mlog(ML_NOTICE, "Block check count has wrapped\n");
295 }
296 
ocfs2_blockcheck_inc_failure(struct ocfs2_blockcheck_stats * stats)297 static void ocfs2_blockcheck_inc_failure(struct ocfs2_blockcheck_stats *stats)
298 {
299 	u64 new_count;
300 
301 	if (!stats)
302 		return;
303 
304 	spin_lock(&stats->b_lock);
305 	stats->b_failure_count++;
306 	new_count = stats->b_failure_count;
307 	spin_unlock(&stats->b_lock);
308 
309 	if (!new_count)
310 		mlog(ML_NOTICE, "Checksum failure count has wrapped\n");
311 }
312 
ocfs2_blockcheck_inc_recover(struct ocfs2_blockcheck_stats * stats)313 static void ocfs2_blockcheck_inc_recover(struct ocfs2_blockcheck_stats *stats)
314 {
315 	u64 new_count;
316 
317 	if (!stats)
318 		return;
319 
320 	spin_lock(&stats->b_lock);
321 	stats->b_recover_count++;
322 	new_count = stats->b_recover_count;
323 	spin_unlock(&stats->b_lock);
324 
325 	if (!new_count)
326 		mlog(ML_NOTICE, "ECC recovery count has wrapped\n");
327 }
328 
329 
330 
331 /*
332  * These are the low-level APIs for using the ocfs2_block_check structure.
333  */
334 
335 /*
336  * This function generates check information for a block.
337  * data is the block to be checked.  bc is a pointer to the
338  * ocfs2_block_check structure describing the crc32 and the ecc.
339  *
340  * bc should be a pointer inside data, as the function will
341  * take care of zeroing it before calculating the check information.  If
342  * bc does not point inside data, the caller must make sure any inline
343  * ocfs2_block_check structures are zeroed.
344  *
345  * The data buffer must be in on-disk endian (little endian for ocfs2).
346  * bc will be filled with little-endian values and will be ready to go to
347  * disk.
348  */
ocfs2_block_check_compute(void * data,size_t blocksize,struct ocfs2_block_check * bc)349 void ocfs2_block_check_compute(void *data, size_t blocksize,
350 			       struct ocfs2_block_check *bc)
351 {
352 	u32 crc;
353 	u32 ecc;
354 
355 	memset(bc, 0, sizeof(struct ocfs2_block_check));
356 
357 	crc = crc32_le(~0, data, blocksize);
358 	ecc = ocfs2_hamming_encode_block(data, blocksize);
359 
360 	/*
361 	 * No ecc'd ocfs2 structure is larger than 4K, so ecc will be no
362 	 * larger than 16 bits.
363 	 */
364 	BUG_ON(ecc > USHRT_MAX);
365 
366 	bc->bc_crc32e = cpu_to_le32(crc);
367 	bc->bc_ecc = cpu_to_le16((u16)ecc);
368 }
369 
370 /*
371  * This function validates existing check information.  Like _compute,
372  * the function will take care of zeroing bc before calculating check codes.
373  * If bc is not a pointer inside data, the caller must have zeroed any
374  * inline ocfs2_block_check structures.
375  *
376  * Again, the data passed in should be the on-disk endian.
377  */
ocfs2_block_check_validate(void * data,size_t blocksize,struct ocfs2_block_check * bc,struct ocfs2_blockcheck_stats * stats)378 int ocfs2_block_check_validate(void *data, size_t blocksize,
379 			       struct ocfs2_block_check *bc,
380 			       struct ocfs2_blockcheck_stats *stats)
381 {
382 	int rc = 0;
383 	u32 bc_crc32e;
384 	u16 bc_ecc;
385 	u32 crc, ecc;
386 
387 	ocfs2_blockcheck_inc_check(stats);
388 
389 	bc_crc32e = le32_to_cpu(bc->bc_crc32e);
390 	bc_ecc = le16_to_cpu(bc->bc_ecc);
391 
392 	memset(bc, 0, sizeof(struct ocfs2_block_check));
393 
394 	/* Fast path - if the crc32 validates, we're good to go */
395 	crc = crc32_le(~0, data, blocksize);
396 	if (crc == bc_crc32e)
397 		goto out;
398 
399 	ocfs2_blockcheck_inc_failure(stats);
400 	mlog(ML_ERROR,
401 	     "CRC32 failed: stored: 0x%x, computed 0x%x. Applying ECC.\n",
402 	     (unsigned int)bc_crc32e, (unsigned int)crc);
403 
404 	/* Ok, try ECC fixups */
405 	ecc = ocfs2_hamming_encode_block(data, blocksize);
406 	ocfs2_hamming_fix_block(data, blocksize, ecc ^ bc_ecc);
407 
408 	/* And check the crc32 again */
409 	crc = crc32_le(~0, data, blocksize);
410 	if (crc == bc_crc32e) {
411 		ocfs2_blockcheck_inc_recover(stats);
412 		goto out;
413 	}
414 
415 	mlog(ML_ERROR, "Fixed CRC32 failed: stored: 0x%x, computed 0x%x\n",
416 	     (unsigned int)bc_crc32e, (unsigned int)crc);
417 
418 	rc = -EIO;
419 
420 out:
421 	bc->bc_crc32e = cpu_to_le32(bc_crc32e);
422 	bc->bc_ecc = cpu_to_le16(bc_ecc);
423 
424 	return rc;
425 }
426 
427 /*
428  * This function generates check information for a list of buffer_heads.
429  * bhs is the blocks to be checked.  bc is a pointer to the
430  * ocfs2_block_check structure describing the crc32 and the ecc.
431  *
432  * bc should be a pointer inside data, as the function will
433  * take care of zeroing it before calculating the check information.  If
434  * bc does not point inside data, the caller must make sure any inline
435  * ocfs2_block_check structures are zeroed.
436  *
437  * The data buffer must be in on-disk endian (little endian for ocfs2).
438  * bc will be filled with little-endian values and will be ready to go to
439  * disk.
440  */
ocfs2_block_check_compute_bhs(struct buffer_head ** bhs,int nr,struct ocfs2_block_check * bc)441 void ocfs2_block_check_compute_bhs(struct buffer_head **bhs, int nr,
442 				   struct ocfs2_block_check *bc)
443 {
444 	int i;
445 	u32 crc, ecc;
446 
447 	BUG_ON(nr < 0);
448 
449 	if (!nr)
450 		return;
451 
452 	memset(bc, 0, sizeof(struct ocfs2_block_check));
453 
454 	for (i = 0, crc = ~0, ecc = 0; i < nr; i++) {
455 		crc = crc32_le(crc, bhs[i]->b_data, bhs[i]->b_size);
456 		/*
457 		 * The number of bits in a buffer is obviously b_size*8.
458 		 * The offset of this buffer is b_size*i, so the bit offset
459 		 * of this buffer is b_size*8*i.
460 		 */
461 		ecc = (u16)ocfs2_hamming_encode(ecc, bhs[i]->b_data,
462 						bhs[i]->b_size * 8,
463 						bhs[i]->b_size * 8 * i);
464 	}
465 
466 	/*
467 	 * No ecc'd ocfs2 structure is larger than 4K, so ecc will be no
468 	 * larger than 16 bits.
469 	 */
470 	BUG_ON(ecc > USHRT_MAX);
471 
472 	bc->bc_crc32e = cpu_to_le32(crc);
473 	bc->bc_ecc = cpu_to_le16((u16)ecc);
474 }
475 
476 /*
477  * This function validates existing check information on a list of
478  * buffer_heads.  Like _compute_bhs, the function will take care of
479  * zeroing bc before calculating check codes.  If bc is not a pointer
480  * inside data, the caller must have zeroed any inline
481  * ocfs2_block_check structures.
482  *
483  * Again, the data passed in should be the on-disk endian.
484  */
ocfs2_block_check_validate_bhs(struct buffer_head ** bhs,int nr,struct ocfs2_block_check * bc,struct ocfs2_blockcheck_stats * stats)485 int ocfs2_block_check_validate_bhs(struct buffer_head **bhs, int nr,
486 				   struct ocfs2_block_check *bc,
487 				   struct ocfs2_blockcheck_stats *stats)
488 {
489 	int i, rc = 0;
490 	u32 bc_crc32e;
491 	u16 bc_ecc;
492 	u32 crc, ecc, fix;
493 
494 	BUG_ON(nr < 0);
495 
496 	if (!nr)
497 		return 0;
498 
499 	ocfs2_blockcheck_inc_check(stats);
500 
501 	bc_crc32e = le32_to_cpu(bc->bc_crc32e);
502 	bc_ecc = le16_to_cpu(bc->bc_ecc);
503 
504 	memset(bc, 0, sizeof(struct ocfs2_block_check));
505 
506 	/* Fast path - if the crc32 validates, we're good to go */
507 	for (i = 0, crc = ~0; i < nr; i++)
508 		crc = crc32_le(crc, bhs[i]->b_data, bhs[i]->b_size);
509 	if (crc == bc_crc32e)
510 		goto out;
511 
512 	ocfs2_blockcheck_inc_failure(stats);
513 	mlog(ML_ERROR,
514 	     "CRC32 failed: stored: %u, computed %u.  Applying ECC.\n",
515 	     (unsigned int)bc_crc32e, (unsigned int)crc);
516 
517 	/* Ok, try ECC fixups */
518 	for (i = 0, ecc = 0; i < nr; i++) {
519 		/*
520 		 * The number of bits in a buffer is obviously b_size*8.
521 		 * The offset of this buffer is b_size*i, so the bit offset
522 		 * of this buffer is b_size*8*i.
523 		 */
524 		ecc = (u16)ocfs2_hamming_encode(ecc, bhs[i]->b_data,
525 						bhs[i]->b_size * 8,
526 						bhs[i]->b_size * 8 * i);
527 	}
528 	fix = ecc ^ bc_ecc;
529 	for (i = 0; i < nr; i++) {
530 		/*
531 		 * Try the fix against each buffer.  It will only affect
532 		 * one of them.
533 		 */
534 		ocfs2_hamming_fix(bhs[i]->b_data, bhs[i]->b_size * 8,
535 				  bhs[i]->b_size * 8 * i, fix);
536 	}
537 
538 	/* And check the crc32 again */
539 	for (i = 0, crc = ~0; i < nr; i++)
540 		crc = crc32_le(crc, bhs[i]->b_data, bhs[i]->b_size);
541 	if (crc == bc_crc32e) {
542 		ocfs2_blockcheck_inc_recover(stats);
543 		goto out;
544 	}
545 
546 	mlog(ML_ERROR, "Fixed CRC32 failed: stored: %u, computed %u\n",
547 	     (unsigned int)bc_crc32e, (unsigned int)crc);
548 
549 	rc = -EIO;
550 
551 out:
552 	bc->bc_crc32e = cpu_to_le32(bc_crc32e);
553 	bc->bc_ecc = cpu_to_le16(bc_ecc);
554 
555 	return rc;
556 }
557 
558 /*
559  * These are the main API.  They check the superblock flag before
560  * calling the underlying operations.
561  *
562  * They expect the buffer(s) to be in disk format.
563  */
ocfs2_compute_meta_ecc(struct super_block * sb,void * data,struct ocfs2_block_check * bc)564 void ocfs2_compute_meta_ecc(struct super_block *sb, void *data,
565 			    struct ocfs2_block_check *bc)
566 {
567 	if (ocfs2_meta_ecc(OCFS2_SB(sb)))
568 		ocfs2_block_check_compute(data, sb->s_blocksize, bc);
569 }
570 
ocfs2_validate_meta_ecc(struct super_block * sb,void * data,struct ocfs2_block_check * bc)571 int ocfs2_validate_meta_ecc(struct super_block *sb, void *data,
572 			    struct ocfs2_block_check *bc)
573 {
574 	int rc = 0;
575 	struct ocfs2_super *osb = OCFS2_SB(sb);
576 
577 	if (ocfs2_meta_ecc(osb))
578 		rc = ocfs2_block_check_validate(data, sb->s_blocksize, bc,
579 						&osb->osb_ecc_stats);
580 
581 	return rc;
582 }
583 
ocfs2_compute_meta_ecc_bhs(struct super_block * sb,struct buffer_head ** bhs,int nr,struct ocfs2_block_check * bc)584 void ocfs2_compute_meta_ecc_bhs(struct super_block *sb,
585 				struct buffer_head **bhs, int nr,
586 				struct ocfs2_block_check *bc)
587 {
588 	if (ocfs2_meta_ecc(OCFS2_SB(sb)))
589 		ocfs2_block_check_compute_bhs(bhs, nr, bc);
590 }
591 
ocfs2_validate_meta_ecc_bhs(struct super_block * sb,struct buffer_head ** bhs,int nr,struct ocfs2_block_check * bc)592 int ocfs2_validate_meta_ecc_bhs(struct super_block *sb,
593 				struct buffer_head **bhs, int nr,
594 				struct ocfs2_block_check *bc)
595 {
596 	int rc = 0;
597 	struct ocfs2_super *osb = OCFS2_SB(sb);
598 
599 	if (ocfs2_meta_ecc(osb))
600 		rc = ocfs2_block_check_validate_bhs(bhs, nr, bc,
601 						    &osb->osb_ecc_stats);
602 
603 	return rc;
604 }
605 
606