xref: /titanic_41/usr/src/uts/common/fs/zev/zev_checksums.c (revision 32a712da90cea6ff9a05f51e7844944ccfa28d5e)
1 #include <sys/zfs_events.h>
2 #include <sys/zev_checksums.h>
3 #include <sys/fs/zev.h>
4 #include <sys/zfs_znode.h>
5 #include <sys/sha1.h>
6 #include <sys/avl.h>
7 #include <sys/sysmacros.h>
8 #include <sys/fs/zev.h>
9 #include <sys/zfs_rlock.h>
10 #include <sys/list.h>
11 
12 typedef struct zev_sig_cache_chksums_t {
13 	/* begin of key */
14 	uint64_t			offset_l1;
15 	/* end of key */
16 	avl_node_t			avl_node;
17 	uint8_t		sigs[ZEV_L1_SIZE/ZEV_L0_SIZE][SHA1_DIGEST_LENGTH];
18 } zev_sig_cache_chksums_t;
19 
20 typedef struct zev_sig_cache_file_t {
21 	/* begin of key */
22 	uint64_t			guid;
23 	uint64_t			ino;
24 	uint64_t			gen;
25 	/* end of key */
26 	uint32_t			refcnt;
27 	list_node_t			lru_node;
28 	avl_node_t			avl_node;
29 	avl_tree_t			chksums;
30 } zev_sig_cache_file_t;
31 
32 typedef struct zev_sig_cache_t {
33 	kmutex_t			mutex;
34 	uint64_t			cache_size;
35 	uint64_t			max_cache_size;
36 	uint64_t			hits;
37 	uint64_t			misses;
38 	list_t				lru;
39 	avl_tree_t			files;
40 } zev_sig_cache_t;
41 
42 extern offset_t zfs_read_chunk_size;	/* tuneable from zfs_vnops.c */
43 
44 static uint8_t all_zero_sig[SHA1_DIGEST_LENGTH] = {
45 	0x1c, 0xea, 0xf7, 0x3d, 0xf4, 0x0e, 0x53, 0x1d, 0xf3, 0xbf,
46 	0xb2, 0x6b, 0x4f, 0xb7, 0xcd, 0x95, 0xfb, 0x7b, 0xff, 0x1d
47 };
48 
49 static uint8_t unknown_sig[SHA1_DIGEST_LENGTH] = {
50 	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
51 	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00
52 };
53 
54 static zev_sig_cache_t	zev_sig_cache;
55 
56 static int
57 zev_cache_file_cmp(const void *entry_a, const void *entry_b)
58 {
59 	const zev_sig_cache_file_t *a = entry_a;
60 	const zev_sig_cache_file_t *b = entry_b;
61 
62 	if (a->guid < b->guid)
63 		return -1;
64 	if (a->guid > b->guid)
65 		return 1;
66 	if (a->ino < b->ino)
67 		return -1;
68 	if (a->ino > b->ino)
69 		return 1;
70 	if (a->gen < b->gen)
71 		return -1;
72 	if (a->gen > b->gen)
73 		return 1;
74 	return 0;
75 }
76 
77 static int
78 zev_chksum_cache_cmp(const void *entry_a, const void *entry_b)
79 {
80 	const zev_sig_cache_chksums_t *a = entry_a;
81 	const zev_sig_cache_chksums_t *b = entry_b;
82 
83 	if (a->offset_l1 < b->offset_l1)
84 		return -1;
85 	if (a->offset_l1 > b->offset_l1)
86 		return 1;
87 	return 0;
88 }
89 
90 /* must be called with zev_sig_cache.mutex held */
91 static void
92 zev_chksum_cache_file_free(zev_sig_cache_file_t *file)
93 {
94 	zev_sig_cache_chksums_t *cs;
95 	void *c = NULL; /* cookie */
96 
97 	/* remove from lru list */
98 	list_remove(&zev_sig_cache.lru, file);
99 	/* free resources */
100 	avl_remove(&zev_sig_cache.files, file);
101 	while ((cs = avl_destroy_nodes(&file->chksums, &c)) != NULL) {
102 		zev_sig_cache.cache_size -= sizeof(*cs);
103 		zev_free(cs, sizeof(*cs));
104 	}
105 	avl_destroy(&file->chksums);
106 	zev_free(file, sizeof(*file));
107 	zev_sig_cache.cache_size -= sizeof(*file);
108 }
109 
110 void
111 zev_chksum_init(void)
112 {
113 	memset(&zev_sig_cache, 0, sizeof(zev_sig_cache));
114 	mutex_init(&zev_sig_cache.mutex, NULL, MUTEX_DRIVER, NULL);
115 	avl_create(&zev_sig_cache.files, zev_cache_file_cmp,
116 	           sizeof(zev_sig_cache_file_t),
117 	           offsetof(zev_sig_cache_file_t, avl_node));
118 	list_create(&zev_sig_cache.lru,
119 	            sizeof(zev_sig_cache_file_t),
120 	            offsetof(zev_sig_cache_file_t, lru_node));
121 	zev_sig_cache.max_cache_size = ZEV_CHKSUM_DEFAULT_CACHE_SIZE;
122 }
123 
124 void
125 zev_chksum_fini(void)
126 {
127 	zev_sig_cache_file_t *file;
128 
129 	mutex_destroy(&zev_sig_cache.mutex);
130 	while ((file = avl_first(&zev_sig_cache.files)) != NULL)
131 		zev_chksum_cache_file_free(file);
132 	list_destroy(&zev_sig_cache.lru);
133 	avl_destroy(&zev_sig_cache.files);
134 }
135 
136 static zev_sig_cache_file_t *
137 zev_chksum_cache_file_get_and_hold(znode_t *zp)
138 {
139 	zev_sig_cache_file_t find_file;
140 	zev_sig_cache_file_t *file;
141 	avl_index_t where;
142 
143 	find_file.guid = zp->z_zfsvfs->z_os->os_dsl_dataset->ds_phys->ds_guid;
144 	find_file.ino = zp->z_id;
145 	find_file.gen = zp->z_gen;
146 
147 	mutex_enter(&zev_sig_cache.mutex);
148 	file = avl_find(&zev_sig_cache.files, &find_file, &where);
149 	if (!file) {
150 		file = zev_alloc(sizeof(*file));
151 		file->guid =
152 		    zp->z_zfsvfs->z_os->os_dsl_dataset->ds_phys->ds_guid;
153 		file->ino = zp->z_id;
154 		file->gen = zp->z_gen;
155 		file->refcnt = 0;
156 		avl_create(&file->chksums, zev_chksum_cache_cmp,
157 		           sizeof(zev_sig_cache_chksums_t),
158 		           offsetof(zev_sig_cache_chksums_t, avl_node));
159 		list_insert_head(&zev_sig_cache.lru, file);
160 		avl_insert(&zev_sig_cache.files, file, where);
161 		zev_sig_cache.cache_size += sizeof(*file);
162 	}
163 	file->refcnt++;
164 	mutex_exit(&zev_sig_cache.mutex);
165 	return file;
166 }
167 
168 static void
169 zev_chksum_cache_file_release(zev_sig_cache_file_t *file)
170 {
171 	mutex_enter(&zev_sig_cache.mutex);
172 
173 	/* We don't invalidate/free/destroy *file. Cache expiry does that */
174 	file->refcnt--;
175 
176 	/* Move file to front of lru list */
177 	list_remove(&zev_sig_cache.lru, file);
178 	list_insert_head(&zev_sig_cache.lru, file);
179 
180 	mutex_exit(&zev_sig_cache.mutex);
181 }
182 
183 static  zev_sig_cache_chksums_t *
184 zev_chksum_cache_get_lv1_entry(zev_sig_cache_file_t *file, uint64_t off_l1)
185 {
186 	zev_sig_cache_chksums_t find_chksum;
187 	zev_sig_cache_chksums_t *cs;
188 	avl_index_t where;
189 
190 	mutex_enter(&zev_sig_cache.mutex);
191 
192 	find_chksum.offset_l1 = off_l1;
193 	cs = avl_find(&file->chksums, &find_chksum, &where);
194 	if (!cs) {
195 		cs = zev_zalloc(sizeof(*cs));
196 		cs->offset_l1 = off_l1;
197 		avl_insert(&file->chksums, cs, where);
198 		zev_sig_cache.cache_size += sizeof(*cs);
199 	}
200 
201 	mutex_exit(&zev_sig_cache.mutex);
202 
203 	return cs;
204 }
205 
206 void
207 zev_chksum_stats(uint64_t *c_size, uint64_t *c_hits, uint64_t *c_misses)
208 {
209 	mutex_enter(&zev_sig_cache.mutex);
210 	*c_size = zev_sig_cache.cache_size;
211 	*c_hits = zev_sig_cache.hits;
212 	*c_misses = zev_sig_cache.misses;
213 	mutex_exit(&zev_sig_cache.mutex);
214 }
215 
216 static void
217 zev_chksum_cache_invalidate(zev_sig_cache_file_t *file,
218                             znode_t *zp,
219                             zev_chksum_mode_t mode,
220                             uint64_t off,
221                             uint64_t len)
222 {
223 	zev_sig_cache_chksums_t find_chksum;
224 	zev_sig_cache_chksums_t *cs;
225 	int idx;
226 	uint64_t off_l1;
227 	uint64_t len_l1;
228 	uint64_t pos_l0;
229 	uint64_t pos_l1;
230 
231 	mutex_enter(&zev_sig_cache.mutex);
232 
233 	/* start of this megabyte */
234 	off_l1 = P2ALIGN(off, ZEV_L1_SIZE);
235 
236 	if (len == 0) {
237 		/* truncate() to EOF */
238 		len_l1 = ZEV_L1_SIZE;
239 	} else {
240 		/* full megabytes */
241 		len_l1 = len + (off - off_l1);
242 		len_l1 = P2ROUNDUP(len_l1, ZEV_L1_SIZE);
243 	}
244 
245 	for (pos_l1 = off_l1; pos_l1 < (off_l1+len_l1); pos_l1 += ZEV_L1_SIZE) {
246 
247 		find_chksum.offset_l1 = pos_l1;
248 		cs = avl_find(&file->chksums, &find_chksum, NULL);
249 		if (!cs)
250 			continue;
251 
252 		for (pos_l0 = MAX(pos_l1, P2ALIGN(off, ZEV_L0_SIZE));
253 		     pos_l0 < (pos_l1 + ZEV_L1_SIZE);
254 		     pos_l0 += ZEV_L0_SIZE){
255 
256 			if ((len > 0) && (pos_l0 > (off + len - 1)))
257 				break;
258 
259 			idx = (pos_l0 % ZEV_L1_SIZE) / ZEV_L0_SIZE;
260 			memcpy(cs->sigs[idx], unknown_sig, SHA1_DIGEST_LENGTH);
261 		}
262 	}
263 
264 	if (len == 0) {
265 		/* truncate() to EOF -> invalidate all l1 sigs beyond EOF */
266 		while ((cs = avl_last(&file->chksums)) != NULL) {
267 			if (cs->offset_l1 < zp->z_size)
268 				break;
269 			avl_remove(&file->chksums, cs);
270 			zev_sig_cache.cache_size -= sizeof(*cs);
271 			zev_free(cs, sizeof(*cs));
272 		}
273 	}
274 
275 	mutex_exit(&zev_sig_cache.mutex);
276 }
277 
278 static int
279 zev_chksum_cache_get(uint8_t *dst,
280                      zev_sig_cache_file_t *file,
281                      zev_sig_cache_chksums_t *cs,
282                      uint64_t off_l0)
283 {
284 	int idx;
285 
286 	mutex_enter(&zev_sig_cache.mutex);
287 
288 	idx = (off_l0 % ZEV_L1_SIZE) / ZEV_L0_SIZE;
289 	if (!memcmp(cs->sigs[idx], unknown_sig, SHA1_DIGEST_LENGTH)) {
290 		zev_sig_cache.misses++;
291 		mutex_exit(&zev_sig_cache.mutex);
292 		return ENOENT;
293 	}
294 	memcpy(dst, cs->sigs[idx], SHA1_DIGEST_LENGTH);
295 	zev_sig_cache.hits++;
296 
297 	mutex_exit(&zev_sig_cache.mutex);
298 	return 0;
299 }
300 
301 static void
302 zev_chksum_cache_put(uint8_t *sig,
303                      zev_sig_cache_file_t *file,
304                      zev_sig_cache_chksums_t *cs,
305                      uint64_t off_l0)
306 {
307 	zev_sig_cache_file_t *f;
308 	zev_sig_cache_file_t *tmp;
309 	int idx;
310 
311 	mutex_enter(&zev_sig_cache.mutex);
312 
313 	if (zev_sig_cache.max_cache_size == 0) {
314 		/* cache disabled */
315 		mutex_exit(&zev_sig_cache.mutex);
316 		return;
317 	}
318 
319 	/* expire entries until there's room in the cache */
320 	f = list_tail(&zev_sig_cache.lru);
321 	while (f && (zev_sig_cache.cache_size > zev_sig_cache.max_cache_size)){
322 		tmp = f;
323 		f = list_prev(&zev_sig_cache.lru, f);
324 		if (tmp->refcnt == 0)
325 			zev_chksum_cache_file_free(tmp);
326 	}
327 
328 	idx = (off_l0 % ZEV_L1_SIZE) / ZEV_L0_SIZE;
329 	memcpy(cs->sigs[idx], sig, SHA1_DIGEST_LENGTH);
330 
331 	mutex_exit(&zev_sig_cache.mutex);
332 	return;
333 }
334 
335 /* verbatim from zfs_vnops.c (unfortunatly it's declared static, there) */
336 static int
337 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
338 {
339 	znode_t *zp = VTOZ(vp);
340 	objset_t *os = zp->z_zfsvfs->z_os;
341 	int64_t	start, off;
342 	int len = nbytes;
343 	int error = 0;
344 
345 	start = uio->uio_loffset;
346 	off = start & PAGEOFFSET;
347 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
348 		page_t *pp;
349 		uint64_t bytes = MIN(PAGESIZE - off, len);
350 
351 		if (pp = page_lookup(vp, start, SE_SHARED)) {
352 			caddr_t va;
353 
354 			va = zfs_map_page(pp, S_READ);
355 			error = uiomove(va + off, bytes, UIO_READ, uio);
356 			zfs_unmap_page(pp, va);
357 			page_unlock(pp);
358 		} else {
359 			error = dmu_read_uio(os, zp->z_id, uio, bytes);
360 		}
361 		len -= bytes;
362 		off = 0;
363 		if (error)
364 			break;
365 	}
366 	return (error);
367 }
368 
369 static int
370 zev_safe_read(znode_t *zp, char *buf, uint64_t off, uint64_t len)
371 {
372 	uio_t		uio;
373 	struct iovec	iov;
374 	ssize_t		n;
375 	ssize_t		nbytes;
376 	int		error = 0;
377 	vnode_t		*vp = ZTOV(zp);
378 	objset_t	*os = zp->z_zfsvfs->z_os;
379 
380 	/* set up uio */
381 
382 	iov.iov_base = buf;
383 	iov.iov_len = ZEV_L0_SIZE;
384 
385 	uio.uio_iov = &iov;
386 	uio.uio_iovcnt = 1;
387 	uio.uio_segflg = (short)UIO_SYSSPACE;
388 	uio.uio_llimit = RLIM64_INFINITY;
389 	uio.uio_fmode = FREAD;
390 	uio.uio_extflg = UIO_COPY_DEFAULT;
391 
392 	uio.uio_loffset = off;
393 	uio.uio_resid = len;
394 
395 again:
396 	if (uio.uio_loffset >= zp->z_size)
397 		return EINVAL;
398 
399 	/* don't read past EOF */
400 	n = MIN(uio.uio_resid, zp->z_size - uio.uio_loffset);
401 
402 	/* this block was essentially copied from zfs_read() in zfs_vnops.c */
403 	while (n > 0) {
404 		nbytes = MIN(n, zfs_read_chunk_size -
405 		    P2PHASE(uio.uio_loffset, zfs_read_chunk_size));
406 
407 		if (vn_has_cached_data(vp)) {
408 			error = mappedread(vp, nbytes, &uio);
409 		} else {
410 			error = dmu_read_uio(os, zp->z_id, &uio, nbytes);
411 		}
412 		if (error) {
413 			if (error = EINTR)
414 				goto again;
415 			/* convert checksum errors into IO errors */
416 			if (error == ECKSUM)
417 				error = SET_ERROR(EIO);
418 			break;
419 		}
420 
421 		n -= nbytes;
422 	}
423 
424 	if (error)
425 		return error;
426 	return len - uio.uio_resid;
427 }
428 
429 static void
430 zev_l0_sig(uint8_t *sig, char *buf)
431 {
432 	SHA1_CTX	ctx;
433 
434 	SHA1Init(&ctx);
435 	SHA1Update(&ctx, buf, ZEV_L0_SIZE);
436 	SHA1Final(sig, &ctx);
437 	return;
438 }
439 
440 static void
441 zev_l0_blocksig(uint8_t *blk_sig, uint8_t *l0_sig, uint8_t block_no)
442 {
443 	SHA1_CTX	ctx;
444 
445 	SHA1Init(&ctx);
446 	SHA1Update(&ctx, l0_sig, SHA1_DIGEST_LENGTH);
447 	SHA1Update(&ctx, &block_no, sizeof(block_no));
448 	SHA1Final(blk_sig, &ctx);
449 	return;
450 }
451 
452 static void
453 zev_l1_add(uint8_t *sig_l1, uint8_t *sig_l0)
454 {
455 	int	i;
456 	int	s;
457 	int	carry = 0;
458 
459 	for (i = SHA1_DIGEST_LENGTH - 1; i >= 0; --i) {
460 		s = sig_l1[i] + sig_l0[i] + carry;
461 		carry = s > 255 ? 1 : 0;
462 		sig_l1[i] = s & 0xff;
463 	}
464 }
465 
466 static int
467 zev_get_result_buffer(zev_sig_t **buffer,
468                       uint64_t *buffer_len,
469                       uint64_t max_buffer_len,
470                       znode_t *zp,
471                       uint64_t off,
472                       uint64_t len,
473                       zev_chksum_mode_t mode)
474 {
475 	uint64_t	blk_start;
476 	uint64_t	blk_end;
477 	uint64_t	l0_blocks;
478 	uint64_t	l1_blocks;
479 	uint64_t	sigs;
480 	int buflen;
481 
482 	/* calculate result set size: how many checksums will we provide? */
483 
484 	ASSERT(len > 0 || (mode == zev_truncate && len == 0));
485 
486 	if (len == 0) {
487 		/* truncate */
488 		l0_blocks = ((off % ZEV_L0_SIZE) == 0) ? 0 : 1;
489 		l1_blocks = ((off % ZEV_L1_SIZE) == 0) ? 0 : 1;
490 	} else {
491 		/* how many lv1 checksums do we update? */
492 		blk_start = off / ZEV_L1_SIZE;
493 		blk_end = (off + len - 1) / ZEV_L1_SIZE;
494 		l1_blocks = blk_end - blk_start + 1;
495 		/* how many lv0 checksums do we update? */
496 		blk_start = off / ZEV_L0_SIZE;
497 		blk_end = (off + len - 1) / ZEV_L0_SIZE;
498 		l0_blocks = blk_end - blk_start + 1;
499 	}
500 
501 	sigs = l1_blocks + l0_blocks;
502 	if (sigs == 0) {
503 		*buffer = NULL;
504 		*buffer_len = 0;
505 		return 0;
506 	}
507 
508 	buflen = sigs * sizeof(zev_sig_t);
509 	if (max_buffer_len && (buflen > max_buffer_len)) {
510 		*buffer = NULL;
511 		*buffer_len = 0;
512 		return ENOSPC;
513 	}
514 	*buffer_len = buflen;
515 	*buffer = zev_alloc(buflen);
516 	return 0;
517 }
518 
519 static void
520 zev_append_sig(zev_sig_t *s, int level, uint64_t off, uint8_t *sig)
521 {
522 	s->level = level;
523 	s->block_offset = off;
524 	memcpy(s->value, sig, SHA1_DIGEST_LENGTH);
525 }
526 
527 /*
528  * Calculate all l0 and l1 checksums that are affected by the given range.
529  *
530  * This function assumes that the ranges it needs to read are already
531  * range-locked.
532  */
533 int
534 zev_get_checksums(zev_sig_t **result,
535                   uint64_t *result_buf_len,
536                   uint64_t *signature_cnt,
537                   uint64_t max_result_len,
538                   znode_t *zp,
539                   uint64_t off,
540                   uint64_t len,
541                   zev_chksum_mode_t mode)
542 {
543 	uint64_t	off_l1;
544 	uint64_t	len_l1;
545 	uint64_t	pos_l1;
546 	uint64_t	pos_l0;
547 	char		*buf;
548 	int64_t		ret;
549 	uint8_t		sig_l0[SHA1_DIGEST_LENGTH];
550 	uint8_t		blk_sig_l0[SHA1_DIGEST_LENGTH];
551 	uint8_t		sig_l1[SHA1_DIGEST_LENGTH];
552 	uint8_t		l0_block_no;
553 	zev_sig_t	*sig;
554 	int		non_empty_l0_blocks;
555 	zev_sig_cache_file_t *file;
556 	zev_sig_cache_chksums_t *cs;
557 
558 	/*
559 	 * Note: for write events, the callback is called via
560 	 *    zfs_write() -> zfs_log_write() -> zev_znode_write_cb()
561 	 *
562 	 * The transaction is not commited, yet.
563 	 *
564 	 * A write() syscall might be split into smaller chunks by zfs_write()
565 	 *
566 	 * zfs_write() has a range lock when this is called. (zfs_vnops.c:925)
567 	 * In zev mode, the range lock will encompass all data we need
568 	 * to calculate our checksums.
569 	 *
570 	 * The same is true for truncates with non-zero length. ("punch hole")
571 	 */
572 
573 	ASSERT(len > 0 || (mode == zev_truncate && len == 0));
574 	*signature_cnt = 0;
575 
576 	/*
577 	 * Under certain circumstances we need the first l0 block's
578 	 * checksum, because we didn't store it in the database and
579 	 * can't easily get it from userspace.  Not for this exact point
580 	 * in time, anyway.  So we cheat a little.
581 	 */
582 	if (mode == zev_truncate && len == 0 && off == 4096) {
583 		/*
584 		 * Normally, we'd report no checkums:
585 		 *  - no l0 sum, because no remaining l0 block is changed
586 		 *  - no l1 sum, because the file is now too short for l1 sums
587 		 * Let's pretend we changed the first l0 block, then.
588 		 * Luckily the entire file is range locked during truncate().
589 		 */
590 		off = 0;
591 		len = 4096;
592 	}
593 
594 	/* start of this megabyte */
595 	off_l1 = P2ALIGN(off, ZEV_L1_SIZE);
596 	/* full megabytes */
597 	if (len == 0) {
598 		/* truncate(): we'll look at the last lv1 block, only. */
599 		len_l1 = ZEV_L1_SIZE;
600 	} else {
601 		len_l1 = len + (off - off_l1);
602 		len_l1 = P2ROUNDUP(len_l1, ZEV_L1_SIZE);
603 	}
604 
605 	file = zev_chksum_cache_file_get_and_hold(zp);
606 	zev_chksum_cache_invalidate(file, zp, mode, off, len);
607 	buf = zev_alloc(ZEV_L0_SIZE);
608 
609 	ret = zev_get_result_buffer(result, result_buf_len, max_result_len,
610 	                            zp, off, len, mode);
611 	if (ret) {
612 		zev_free(buf, ZEV_L0_SIZE);
613 		zev_chksum_cache_file_release(file);
614 		return ret;
615 	}
616 	if (*result == NULL) {
617 		/* we're done */
618 		zev_free(buf, ZEV_L0_SIZE);
619 		zev_chksum_cache_file_release(file);
620 		return 0;
621 	}
622 	sig = *result;
623 
624 	for (pos_l1 = off_l1; pos_l1 < (off_l1+len_l1); pos_l1 += ZEV_L1_SIZE) {
625 
626 		if (pos_l1 > zp->z_size) {
627 			cmn_err(CE_WARN, "zev_get_checksums: off+len beyond "
628 			        "EOF.  Unexpected behaviour; please fix!");
629 			zev_free(*result, *result_buf_len);
630 			cmn_err(CE_WARN, "off=%" PRIu64 ", len=%" PRIu64 ", "
631 			        "dataset='%s', inode=%" PRIu64,
632 			        off, len, zp->z_zfsvfs->z_os->
633 			        os_dsl_dataset->ds_dir->dd_myname, zp->z_id);
634 			zev_free(*result, *result_buf_len);
635 			*result = NULL;
636 			break;
637 		}
638 
639 		/*
640 		 * Since we have a reference to 'file' 'cs' can't be expired.
641 		 * Since our ranges are range locked, other threads woun't
642 		 * touch our checksum entries. (not even read them)
643 		 * Hence, we don't need to hold() or release() 'cs'.
644 		 */
645 		cs = zev_chksum_cache_get_lv1_entry(file, pos_l1);
646 
647 		l0_block_no = 0;
648 		non_empty_l0_blocks = 0;
649 		bzero(sig_l1, sizeof(sig_l1));
650 		for (pos_l0 = pos_l1;
651 		     pos_l0 < (pos_l1 + ZEV_L1_SIZE);
652 		     pos_l0 += ZEV_L0_SIZE){
653 
654 			if (pos_l0 >= zp->z_size)
655 				break;	/* EOF */
656 
657 			if (zev_chksum_cache_get(sig_l0, file,cs,pos_l0) != 0) {
658 
659 				/* signature is not cached, yet. */
660 				ret = zev_safe_read(zp, buf,
661 				                    pos_l0, ZEV_L0_SIZE);
662 				if (ret < 0) {
663 					zev_free(*result, *result_buf_len);
664 					zev_free(buf, ZEV_L0_SIZE);
665 					zev_chksum_cache_file_release(file);
666 					return ret;
667 				}
668 				/* pad buffer with zeros if necessary */
669 				if (ret < ZEV_L0_SIZE)
670 					bzero(buf + ret, ZEV_L0_SIZE - ret);
671 
672 				/* calculate signature */
673 				zev_l0_sig(sig_l0, buf);
674 
675 				zev_chksum_cache_put(sig_l0, file, cs, pos_l0);
676 			}
677 
678 			if (!memcmp(sig_l0, all_zero_sig, SHA1_DIGEST_LENGTH)) {
679 				/* all-zero l0 block.  omit signature. */
680 				l0_block_no++;
681 				continue;
682 			}
683 			non_empty_l0_blocks++;
684 			zev_l0_blocksig(blk_sig_l0, sig_l0, l0_block_no);
685 			zev_l1_add(sig_l1, blk_sig_l0);
686 
687 			if (((pos_l0 + ZEV_L0_SIZE - 1) >= off) &&
688 			    (pos_l0 <= (off + len - 1))) {
689 				zev_append_sig(sig++, 0, pos_l0, sig_l0);
690 			}
691 
692 			l0_block_no++;
693 		}
694 
695 		if (non_empty_l0_blocks && (zp->z_size > ZEV_L0_SIZE))
696 			zev_append_sig(sig++, 1, pos_l1, sig_l1);
697 	}
698 
699 	*signature_cnt = ((char *)sig - (char *)*result) / sizeof(zev_sig_t);
700 
701 	zev_free(buf, ZEV_L0_SIZE);
702 	zev_chksum_cache_file_release(file);
703 	return 0;
704 }
705 
706 int
707 zev_ioc_get_signatures(intptr_t arg, int mode)
708 {
709 	zev_ioctl_get_signatures_t gs;
710 	file_t *fp;
711 	int ret = 0;
712 	znode_t *zp;
713 	zev_sig_t *sig_buf = NULL;
714 	uint64_t sig_buf_len;
715 	uint64_t sig_cnt = 0;
716 	uint64_t sig_len;
717 	char *dst;
718 	int range_locked = 0;
719 	rl_t *rl;
720 	ssize_t	lock_off;
721 	ssize_t lock_len;
722 
723 	if (ddi_copyin((void *)arg, &gs, sizeof(gs), mode) != 0)
724 		return EFAULT;
725 	fp = getf(gs.zev_fd);
726 	if (fp == NULL)
727 		return EBADF;
728 	if (fp->f_vnode->v_vfsp->vfs_fstype != zfsfstype) {
729 		ret = EINVAL;
730 		goto out;
731 	}
732 	if (fp->f_vnode->v_type != VREG) {
733 		ret = EINVAL;
734 		goto out;
735 	}
736 	zp = VTOZ(fp->f_vnode);
737 	if (gs.zev_offset >= zp->z_size) {
738 		ret = EINVAL;
739 		goto out;
740 	}
741 
742 	/* range lock data */
743 	lock_off = P2ALIGN(gs.zev_offset, ZEV_L1_SIZE);
744 	lock_len = gs.zev_len + (gs.zev_offset - lock_off);
745 	lock_len = P2ROUNDUP(lock_len, ZEV_L1_SIZE);
746 	rl = zfs_range_lock(zp, lock_off, lock_len, RL_READER);
747 	range_locked = 1;
748 
749 	/* get checksums */
750 	ret = zev_get_checksums(&sig_buf, &sig_buf_len, &sig_cnt,
751 	                        gs.zev_bufsize,
752 	                        zp, gs.zev_offset, gs.zev_len, zev_write);
753 	if (ret)
754 		goto out;
755 
756 	/* copy to userland */
757 	sig_len = sig_cnt * sizeof(zev_sig_t);
758 	gs.zev_signature_cnt = sig_cnt;
759 	if (ddi_copyout(&gs, (void *)arg, sizeof(gs), mode) != 0) {
760 		ret = EFAULT;
761 		goto out;
762 	}
763 	if (sig_cnt && sig_buf) {
764 		dst = (char *)arg + sizeof(gs);
765 		if (ddi_copyout(sig_buf, (void *)dst, sig_len, mode) != 0) {
766 			ret = EFAULT;
767 			goto out;
768 		}
769 	}
770 out:
771 	if (sig_buf)
772 		zev_free(sig_buf, sig_buf_len);
773 	if (range_locked)
774 		zfs_range_unlock(rl);
775 	releasef(gs.zev_fd);
776 	return ret;
777 }
778 
779 void
780 zev_symlink_checksum(zev_znode_symlink_t *rec, char *link)
781 {
782 	char buf[ZEV_L0_SIZE];
783 
784 	memset(buf, 0, sizeof(buf));
785 	strcpy(buf, link);
786 	zev_l0_sig(rec->signature.value, buf);
787 	rec->signature.level = 0;
788 	rec->signature.block_offset = 0;
789 }
790 
791 
792 void
793 zev_create_checksum(zev_znode_create_t *rec, znode_t *zp)
794 {
795 	char buf[ZEV_L0_SIZE];
796 	vnode_t *vp;
797 	uint64_t rdev;
798 
799 	vp = ZTOV(zp);
800 	if (vp->v_type == VBLK || vp->v_type == VCHR) {
801 		sa_lookup(zp->z_sa_hdl, SA_ZPL_RDEV(zp->z_zfsvfs),
802 		          &rdev, sizeof(rdev));
803 		memset(buf, 0, sizeof(buf));
804 		snprintf(buf, sizeof(buf), "%c%d,%d",
805 		         vp->v_type == VBLK ? 'b' : 'c',
806 		         getmajor(rdev),
807 		         getminor(rdev));
808 		zev_l0_sig(rec->signature.value, buf);
809 	} else {
810 		memset(rec->signature.value, 0, sizeof(rec->signature.value));
811 	}
812 	rec->signature.level = 0;
813 	rec->signature.block_offset = 0;
814 }
815 
816