1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
|
/*-
* Copyright (c) 2014-2020 MongoDB, Inc.
* Copyright (c) 2008-2014 WiredTiger, Inc.
* All rights reserved.
*
* See the file LICENSE for redistribution information.
*/
#include "wt_internal.h"
static int __desc_read(WT_SESSION_IMPL *, uint32_t allocsize, WT_BLOCK *);
/*
* __wt_block_manager_drop --
* Drop a file.
*/
int
__wt_block_manager_drop(WT_SESSION_IMPL *session, const char *filename, bool durable)
{
return (__wt_remove_if_exists(session, filename, durable));
}
/*
* __wt_block_manager_create --
* Create a file.
*/
int
__wt_block_manager_create(WT_SESSION_IMPL *session, const char *filename, uint32_t allocsize)
{
WT_DECL_ITEM(tmp);
WT_DECL_RET;
WT_FH *fh;
int suffix;
bool exists;
/*
* Create the underlying file and open a handle.
*
* Since WiredTiger schema operations are (currently) non-transactional, it's possible to see a
* partially-created file left from a previous create. Further, there's nothing to prevent users
* from creating files in our space. Move any existing files out of the way and complain.
*/
for (;;) {
if ((ret = __wt_open(session, filename, WT_FS_OPEN_FILE_TYPE_DATA,
WT_FS_OPEN_CREATE | WT_FS_OPEN_DURABLE | WT_FS_OPEN_EXCLUSIVE, &fh)) == 0)
break;
WT_ERR_TEST(ret != EEXIST, ret, false);
if (tmp == NULL)
WT_ERR(__wt_scr_alloc(session, 0, &tmp));
for (suffix = 1;; ++suffix) {
WT_ERR(__wt_buf_fmt(session, tmp, "%s.%d", filename, suffix));
WT_ERR(__wt_fs_exist(session, tmp->data, &exists));
if (!exists) {
WT_ERR(__wt_fs_rename(session, filename, tmp->data, false));
WT_ERR(__wt_msg(session, "unexpected file %s found, renamed to %s", filename,
(const char *)tmp->data));
break;
}
}
}
/* Write out the file's meta-data. */
ret = __wt_desc_write(session, fh, allocsize);
/*
* Ensure the truncated file has made it to disk, then the upper-level is never surprised.
*/
WT_TRET(__wt_fsync(session, fh, true));
/* Close the file handle. */
WT_TRET(__wt_close(session, &fh));
/* Undo any create on error. */
if (ret != 0)
WT_TRET(__wt_fs_remove(session, filename, false));
err:
__wt_scr_free(session, &tmp);
return (ret);
}
/*
* __block_destroy --
* Destroy a block handle.
*/
static int
__block_destroy(WT_SESSION_IMPL *session, WT_BLOCK *block)
{
WT_CONNECTION_IMPL *conn;
WT_DECL_RET;
uint64_t bucket;
conn = S2C(session);
bucket = block->name_hash & (conn->hash_size - 1);
WT_CONN_BLOCK_REMOVE(conn, block, bucket);
__wt_free(session, block->name);
if (block->fh != NULL)
WT_TRET(__wt_close(session, &block->fh));
__wt_spin_destroy(session, &block->live_lock);
__wt_overwrite_and_free(session, block);
return (ret);
}
/*
* __wt_block_configure_first_fit --
* Configure first-fit allocation.
*/
void
__wt_block_configure_first_fit(WT_BLOCK *block, bool on)
{
/*
* Switch to first-fit allocation so we rewrite blocks at the start of the file; use atomic
* instructions because checkpoints also configure first-fit allocation, and this way we stay on
* first-fit allocation as long as any operation wants it.
*/
if (on)
(void)__wt_atomic_add32(&block->allocfirst, 1);
else
(void)__wt_atomic_sub32(&block->allocfirst, 1);
}
/*
* __wt_block_open --
* Open a block handle.
*/
int
__wt_block_open(WT_SESSION_IMPL *session, const char *filename, const char *cfg[],
bool forced_salvage, bool readonly, uint32_t allocsize, WT_BLOCK **blockp)
{
WT_BLOCK *block;
WT_CONFIG_ITEM cval;
WT_CONNECTION_IMPL *conn;
WT_DECL_RET;
uint64_t bucket, hash;
uint32_t flags;
*blockp = block = NULL;
__wt_verbose(session, WT_VERB_BLOCK, "open: %s", filename);
conn = S2C(session);
hash = __wt_hash_city64(filename, strlen(filename));
bucket = hash & (conn->hash_size - 1);
__wt_spin_lock(session, &conn->block_lock);
TAILQ_FOREACH (block, &conn->blockhash[bucket], hashq) {
if (strcmp(filename, block->name) == 0) {
++block->ref;
*blockp = block;
__wt_spin_unlock(session, &conn->block_lock);
return (0);
}
}
/*
* Basic structure allocation, initialization.
*
* Note: set the block's name-hash value before any work that can fail because cleanup calls the
* block destroy code which uses that hash value to remove the block from the underlying linked
* lists.
*/
WT_ERR(__wt_calloc_one(session, &block));
block->ref = 1;
block->name_hash = hash;
block->allocsize = allocsize;
WT_CONN_BLOCK_INSERT(conn, block, bucket);
WT_ERR(__wt_strdup(session, filename, &block->name));
WT_ERR(__wt_config_gets(session, cfg, "block_allocation", &cval));
block->allocfirst = WT_STRING_MATCH("first", cval.str, cval.len);
/* Configuration: optional OS buffer cache maximum size. */
WT_ERR(__wt_config_gets(session, cfg, "os_cache_max", &cval));
block->os_cache_max = (size_t)cval.val;
/* Configuration: optional immediate write scheduling flag. */
WT_ERR(__wt_config_gets(session, cfg, "os_cache_dirty_max", &cval));
block->os_cache_dirty_max = (size_t)cval.val;
/* Set the file extension information. */
block->extend_len = conn->data_extend_len;
/*
* Open the underlying file handle.
*
* "direct_io=checkpoint" configures direct I/O for readonly data files.
*/
flags = 0;
WT_ERR(__wt_config_gets(session, cfg, "access_pattern_hint", &cval));
if (WT_STRING_MATCH("random", cval.str, cval.len))
LF_SET(WT_FS_OPEN_ACCESS_RAND);
else if (WT_STRING_MATCH("sequential", cval.str, cval.len))
LF_SET(WT_FS_OPEN_ACCESS_SEQ);
if (readonly && FLD_ISSET(conn->direct_io, WT_DIRECT_IO_CHECKPOINT))
LF_SET(WT_FS_OPEN_DIRECTIO);
if (!readonly && FLD_ISSET(conn->direct_io, WT_DIRECT_IO_DATA))
LF_SET(WT_FS_OPEN_DIRECTIO);
WT_ERR(__wt_open(session, filename, WT_FS_OPEN_FILE_TYPE_DATA, flags, &block->fh));
/* Set the file's size. */
WT_ERR(__wt_filesize(session, block->fh, &block->size));
/*
* If we're opening a file and it only contains a header and we're doing incremental backup
* indicate this so that the first checkpoint is sure to set all the bits as dirty to cover the
* header so that the header gets copied.
*/
if (block->size == allocsize && F_ISSET(conn, WT_CONN_INCR_BACKUP))
block->created_during_backup = true;
/* Initialize the live checkpoint's lock. */
WT_ERR(__wt_spin_init(session, &block->live_lock, "block manager"));
/*
* Read the description information from the first block.
*
* Salvage is a special case: if we're forcing the salvage, we don't look at anything, including
* the description information.
*/
if (!forced_salvage)
WT_ERR(__desc_read(session, allocsize, block));
*blockp = block;
__wt_spin_unlock(session, &conn->block_lock);
return (0);
err:
if (block != NULL)
WT_TRET(__block_destroy(session, block));
__wt_spin_unlock(session, &conn->block_lock);
return (ret);
}
/*
* __wt_block_close --
* Close a block handle.
*/
int
__wt_block_close(WT_SESSION_IMPL *session, WT_BLOCK *block)
{
WT_CONNECTION_IMPL *conn;
WT_DECL_RET;
if (block == NULL) /* Safety check */
return (0);
conn = S2C(session);
__wt_verbose(session, WT_VERB_BLOCK, "close: %s", block->name == NULL ? "" : block->name);
__wt_spin_lock(session, &conn->block_lock);
/* Reference count is initialized to 1. */
if (block->ref == 0 || --block->ref == 0)
ret = __block_destroy(session, block);
__wt_spin_unlock(session, &conn->block_lock);
return (ret);
}
/*
* __wt_desc_write --
* Write a file's initial descriptor structure.
*/
int
__wt_desc_write(WT_SESSION_IMPL *session, WT_FH *fh, uint32_t allocsize)
{
WT_BLOCK_DESC *desc;
WT_DECL_ITEM(buf);
WT_DECL_RET;
/* If in-memory, we don't read or write the descriptor structure. */
if (F_ISSET(S2C(session), WT_CONN_IN_MEMORY))
return (0);
/* Use a scratch buffer to get correct alignment for direct I/O. */
WT_RET(__wt_scr_alloc(session, allocsize, &buf));
memset(buf->mem, 0, allocsize);
/*
* Checksum a little-endian version of the header, and write everything in little-endian format.
* The checksum is (potentially) returned in a big-endian format, swap it into place in a
* separate step.
*/
desc = buf->mem;
desc->magic = WT_BLOCK_MAGIC;
desc->majorv = WT_BLOCK_MAJOR_VERSION;
desc->minorv = WT_BLOCK_MINOR_VERSION;
desc->checksum = 0;
__wt_block_desc_byteswap(desc);
desc->checksum = __wt_checksum(desc, allocsize);
#ifdef WORDS_BIGENDIAN
desc->checksum = __wt_bswap32(desc->checksum);
#endif
ret = __wt_write(session, fh, (wt_off_t)0, (size_t)allocsize, desc);
__wt_scr_free(session, &buf);
return (ret);
}
/*
* __desc_read --
* Read and verify the file's metadata.
*/
static int
__desc_read(WT_SESSION_IMPL *session, uint32_t allocsize, WT_BLOCK *block)
{
WT_BLOCK_DESC *desc;
WT_DECL_ITEM(buf);
WT_DECL_RET;
uint32_t checksum_saved, checksum_tmp;
bool checksum_matched;
/* If in-memory, we don't read or write the descriptor structure. */
if (F_ISSET(S2C(session), WT_CONN_IN_MEMORY))
return (0);
/*
* If a data file is smaller than the allocation size, we're not going to be able to read the
* descriptor block.
*
* If we're performing rollback to stable as part of recovery, we should treat this as if the
* file has been deleted; that is, to log an error but continue on.
*
* In the general case, we should return a generic error and signal that we've detected data
* corruption.
*
* FIXME-WT-5832: MongoDB relies heavily on the error codes reported when opening cursors (which
* hits this logic if the relevant data handle isn't already open). However this code gets run
* in rollback to stable as part of recovery where we want to skip any corrupted data files
* temporarily to allow MongoDB to initiate salvage. This is why we've been forced into this
* situation. We should address this as part of WT-5832 and clarify what error codes we expect
* to be returning across the API boundary.
*/
if (block->size < allocsize) {
if (F_ISSET(session, WT_SESSION_ROLLBACK_TO_STABLE))
ret = ENOENT;
else {
ret = WT_ERROR;
F_SET(S2C(session), WT_CONN_DATA_CORRUPTION);
}
WT_RET_MSG(session, ret,
"File %s is smaller than allocation size; file size=%" PRId64 ", alloc size=%" PRIu32,
block->name, block->size, allocsize);
}
/* Use a scratch buffer to get correct alignment for direct I/O. */
WT_RET(__wt_scr_alloc(session, allocsize, &buf));
/* Read the first allocation-sized block and verify the file format. */
WT_ERR(__wt_read(session, block->fh, (wt_off_t)0, (size_t)allocsize, buf->mem));
/*
* Handle little- and big-endian objects. Objects are written in little- endian format: save the
* header checksum, and calculate the checksum for the header in its little-endian form. Then,
* restore the header's checksum, and byte-swap the whole thing as necessary, leaving us with a
* calculated checksum that should match the checksum in the header.
*/
desc = buf->mem;
checksum_saved = checksum_tmp = desc->checksum;
#ifdef WORDS_BIGENDIAN
checksum_tmp = __wt_bswap32(checksum_tmp);
#endif
desc->checksum = 0;
checksum_matched = __wt_checksum_match(desc, allocsize, checksum_tmp);
desc->checksum = checksum_saved;
__wt_block_desc_byteswap(desc);
/*
* We fail the open if the checksum fails, or the magic number is wrong or the major/minor
* numbers are unsupported for this version. This test is done even if the caller is verifying
* or salvaging the file: it makes sense for verify, and for salvage we don't overwrite files
* without some reason to believe they are WiredTiger files. The user may have entered the wrong
* file name, and is now frantically pounding their interrupt key.
*/
if (desc->magic != WT_BLOCK_MAGIC || !checksum_matched) {
if (strcmp(block->name, WT_METAFILE) == 0 || strcmp(block->name, WT_HS_FILE) == 0)
WT_ERR_MSG(session, WT_TRY_SALVAGE, "%s is corrupted", block->name);
/*
* If we're doing an import, we can't expect to be able to verify checksums since we don't
* know the allocation size being used. This isn't an error so we should just return success
* and let import get whatever information it needs.
*/
if (F_ISSET(session, WT_SESSION_IMPORT_REPAIR))
goto err;
if (F_ISSET(session, WT_SESSION_ROLLBACK_TO_STABLE))
ret = ENOENT;
else
WT_ERR_MSG(
session, WT_ERROR, "%s does not appear to be a WiredTiger file", block->name);
}
if (desc->majorv > WT_BLOCK_MAJOR_VERSION ||
(desc->majorv == WT_BLOCK_MAJOR_VERSION && desc->minorv > WT_BLOCK_MINOR_VERSION))
WT_ERR_MSG(session, WT_ERROR,
"unsupported WiredTiger file version: this build only supports major/minor versions up "
"to %d/%d, and the file is version %" PRIu16 "/%" PRIu16,
WT_BLOCK_MAJOR_VERSION, WT_BLOCK_MINOR_VERSION, desc->majorv, desc->minorv);
__wt_verbose(session, WT_VERB_BLOCK, "%s: magic %" PRIu32 ", major/minor: %" PRIu32 "/%" PRIu32,
block->name, desc->magic, desc->majorv, desc->minorv);
err:
__wt_scr_free(session, &buf);
return (ret);
}
/*
* __wt_block_stat --
* Set the statistics for a live block handle.
*/
void
__wt_block_stat(WT_SESSION_IMPL *session, WT_BLOCK *block, WT_DSRC_STATS *stats)
{
/*
* Reading from the live system's structure normally requires locking, but it's an 8B statistics
* read, there's no need.
*/
WT_STAT_WRITE(session, stats, allocation_size, block->allocsize);
WT_STAT_WRITE(session, stats, block_checkpoint_size, (int64_t)block->live.ckpt_size);
WT_STAT_WRITE(session, stats, block_magic, WT_BLOCK_MAGIC);
WT_STAT_WRITE(session, stats, block_major, WT_BLOCK_MAJOR_VERSION);
WT_STAT_WRITE(session, stats, block_minor, WT_BLOCK_MINOR_VERSION);
WT_STAT_WRITE(session, stats, block_reuse_bytes, (int64_t)block->live.avail.bytes);
WT_STAT_WRITE(session, stats, block_size, block->size);
}
/*
* __wt_block_manager_size --
* Return the size of a live block handle.
*/
int
__wt_block_manager_size(WT_BM *bm, WT_SESSION_IMPL *session, wt_off_t *sizep)
{
WT_UNUSED(session);
*sizep = bm->block->size;
return (0);
}
/*
* __wt_block_manager_named_size --
* Return the size of a named file.
*/
int
__wt_block_manager_named_size(WT_SESSION_IMPL *session, const char *name, wt_off_t *sizep)
{
return (__wt_fs_size(session, name, sizep));
}
|