]>
Commit | Line | Data |
---|---|---|
34dc7c2f BB |
1 | /* |
2 | * CDDL HEADER START | |
3 | * | |
4 | * The contents of this file are subject to the terms of the | |
5 | * Common Development and Distribution License (the "License"). | |
6 | * You may not use this file except in compliance with the License. | |
7 | * | |
8 | * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE | |
9 | * or http://www.opensolaris.org/os/licensing. | |
10 | * See the License for the specific language governing permissions | |
11 | * and limitations under the License. | |
12 | * | |
13 | * When distributing Covered Code, include this CDDL HEADER in each | |
14 | * file and include the License file at usr/src/OPENSOLARIS.LICENSE. | |
15 | * If applicable, add the following below this CDDL HEADER, with the | |
16 | * fields enclosed by brackets "[]" replaced with your own identifying | |
17 | * information: Portions Copyright [yyyy] [name of copyright owner] | |
18 | * | |
19 | * CDDL HEADER END | |
20 | */ | |
21 | /* | |
428870ff | 22 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. |
34dc7c2f BB |
23 | */ |
24 | ||
34dc7c2f BB |
25 | /* |
26 | * The objective of this program is to provide a DMU/ZAP/SPA stress test | |
27 | * that runs entirely in userland, is easy to use, and easy to extend. | |
28 | * | |
29 | * The overall design of the ztest program is as follows: | |
30 | * | |
31 | * (1) For each major functional area (e.g. adding vdevs to a pool, | |
32 | * creating and destroying datasets, reading and writing objects, etc) | |
33 | * we have a simple routine to test that functionality. These | |
34 | * individual routines do not have to do anything "stressful". | |
35 | * | |
36 | * (2) We turn these simple functionality tests into a stress test by | |
37 | * running them all in parallel, with as many threads as desired, | |
38 | * and spread across as many datasets, objects, and vdevs as desired. | |
39 | * | |
40 | * (3) While all this is happening, we inject faults into the pool to | |
41 | * verify that self-healing data really works. | |
42 | * | |
43 | * (4) Every time we open a dataset, we change its checksum and compression | |
44 | * functions. Thus even individual objects vary from block to block | |
45 | * in which checksum they use and whether they're compressed. | |
46 | * | |
47 | * (5) To verify that we never lose on-disk consistency after a crash, | |
48 | * we run the entire test in a child of the main process. | |
49 | * At random times, the child self-immolates with a SIGKILL. | |
50 | * This is the software equivalent of pulling the power cord. | |
51 | * The parent then runs the test again, using the existing | |
52 | * storage pool, as many times as desired. | |
53 | * | |
54 | * (6) To verify that we don't have future leaks or temporal incursions, | |
55 | * many of the functional tests record the transaction group number | |
56 | * as part of their data. When reading old data, they verify that | |
57 | * the transaction group number is less than the current, open txg. | |
58 | * If you add a new test, please do this if applicable. | |
59 | * | |
1e33ac1e BB |
60 | * (7) Threads are created with a reduced stack size, for sanity checking. |
61 | * Therefore, it's important not to allocate huge buffers on the stack. | |
62 | * | |
34dc7c2f BB |
63 | * When run with no arguments, ztest runs for about five minutes and |
64 | * produces no output if successful. To get a little bit of information, | |
65 | * specify -V. To get more information, specify -VV, and so on. | |
66 | * | |
67 | * To turn this into an overnight stress test, use -T to specify run time. | |
68 | * | |
69 | * You can ask more more vdevs [-v], datasets [-d], or threads [-t] | |
70 | * to increase the pool capacity, fanout, and overall stress level. | |
71 | * | |
72 | * The -N(okill) option will suppress kills, so each child runs to completion. | |
73 | * This can be useful when you're trying to distinguish temporal incursions | |
74 | * from plain old race conditions. | |
75 | */ | |
76 | ||
77 | #include <sys/zfs_context.h> | |
78 | #include <sys/spa.h> | |
79 | #include <sys/dmu.h> | |
80 | #include <sys/txg.h> | |
9babb374 | 81 | #include <sys/dbuf.h> |
34dc7c2f | 82 | #include <sys/zap.h> |
34dc7c2f BB |
83 | #include <sys/dmu_objset.h> |
84 | #include <sys/poll.h> | |
85 | #include <sys/stat.h> | |
86 | #include <sys/time.h> | |
87 | #include <sys/wait.h> | |
88 | #include <sys/mman.h> | |
89 | #include <sys/resource.h> | |
90 | #include <sys/zio.h> | |
34dc7c2f | 91 | #include <sys/zil.h> |
428870ff | 92 | #include <sys/zil_impl.h> |
34dc7c2f | 93 | #include <sys/vdev_impl.h> |
b128c09f | 94 | #include <sys/vdev_file.h> |
34dc7c2f | 95 | #include <sys/spa_impl.h> |
428870ff | 96 | #include <sys/metaslab_impl.h> |
34dc7c2f | 97 | #include <sys/dsl_prop.h> |
9babb374 | 98 | #include <sys/dsl_dataset.h> |
428870ff BB |
99 | #include <sys/dsl_scan.h> |
100 | #include <sys/zio_checksum.h> | |
34dc7c2f BB |
101 | #include <sys/refcount.h> |
102 | #include <stdio.h> | |
103 | #include <stdio_ext.h> | |
104 | #include <stdlib.h> | |
105 | #include <unistd.h> | |
106 | #include <signal.h> | |
107 | #include <umem.h> | |
108 | #include <dlfcn.h> | |
109 | #include <ctype.h> | |
110 | #include <math.h> | |
111 | #include <sys/fs/zfs.h> | |
428870ff | 112 | #include <libnvpair.h> |
34dc7c2f BB |
113 | |
114 | static char cmdname[] = "ztest"; | |
115 | static char *zopt_pool = cmdname; | |
116 | ||
117 | static uint64_t zopt_vdevs = 5; | |
118 | static uint64_t zopt_vdevtime; | |
119 | static int zopt_ashift = SPA_MINBLOCKSHIFT; | |
120 | static int zopt_mirrors = 2; | |
121 | static int zopt_raidz = 4; | |
122 | static int zopt_raidz_parity = 1; | |
123 | static size_t zopt_vdev_size = SPA_MINDEVSIZE; | |
124 | static int zopt_datasets = 7; | |
125 | static int zopt_threads = 23; | |
126 | static uint64_t zopt_passtime = 60; /* 60 seconds */ | |
127 | static uint64_t zopt_killrate = 70; /* 70% kill rate */ | |
128 | static int zopt_verbose = 0; | |
129 | static int zopt_init = 1; | |
130 | static char *zopt_dir = "/tmp"; | |
131 | static uint64_t zopt_time = 300; /* 5 minutes */ | |
428870ff BB |
132 | static uint64_t zopt_maxloops = 50; /* max loops during spa_freeze() */ |
133 | ||
134 | #define BT_MAGIC 0x123456789abcdefULL | |
135 | #define MAXFAULTS() (MAX(zs->zs_mirrors, 1) * (zopt_raidz_parity + 1) - 1) | |
136 | ||
137 | enum ztest_io_type { | |
138 | ZTEST_IO_WRITE_TAG, | |
139 | ZTEST_IO_WRITE_PATTERN, | |
140 | ZTEST_IO_WRITE_ZEROES, | |
141 | ZTEST_IO_TRUNCATE, | |
142 | ZTEST_IO_SETATTR, | |
143 | ZTEST_IO_TYPES | |
144 | }; | |
34dc7c2f BB |
145 | |
146 | typedef struct ztest_block_tag { | |
428870ff | 147 | uint64_t bt_magic; |
34dc7c2f BB |
148 | uint64_t bt_objset; |
149 | uint64_t bt_object; | |
150 | uint64_t bt_offset; | |
428870ff | 151 | uint64_t bt_gen; |
34dc7c2f | 152 | uint64_t bt_txg; |
428870ff | 153 | uint64_t bt_crtxg; |
34dc7c2f BB |
154 | } ztest_block_tag_t; |
155 | ||
428870ff BB |
156 | typedef struct bufwad { |
157 | uint64_t bw_index; | |
158 | uint64_t bw_txg; | |
159 | uint64_t bw_data; | |
160 | } bufwad_t; | |
161 | ||
162 | /* | |
163 | * XXX -- fix zfs range locks to be generic so we can use them here. | |
164 | */ | |
165 | typedef enum { | |
166 | RL_READER, | |
167 | RL_WRITER, | |
168 | RL_APPEND | |
169 | } rl_type_t; | |
170 | ||
171 | typedef struct rll { | |
172 | void *rll_writer; | |
173 | int rll_readers; | |
1e33ac1e BB |
174 | kmutex_t rll_lock; |
175 | kcondvar_t rll_cv; | |
428870ff BB |
176 | } rll_t; |
177 | ||
178 | typedef struct rl { | |
179 | uint64_t rl_object; | |
180 | uint64_t rl_offset; | |
181 | uint64_t rl_size; | |
182 | rll_t *rl_lock; | |
183 | } rl_t; | |
184 | ||
185 | #define ZTEST_RANGE_LOCKS 64 | |
186 | #define ZTEST_OBJECT_LOCKS 64 | |
187 | ||
188 | /* | |
189 | * Object descriptor. Used as a template for object lookup/create/remove. | |
190 | */ | |
191 | typedef struct ztest_od { | |
192 | uint64_t od_dir; | |
193 | uint64_t od_object; | |
194 | dmu_object_type_t od_type; | |
195 | dmu_object_type_t od_crtype; | |
196 | uint64_t od_blocksize; | |
197 | uint64_t od_crblocksize; | |
198 | uint64_t od_gen; | |
199 | uint64_t od_crgen; | |
200 | char od_name[MAXNAMELEN]; | |
201 | } ztest_od_t; | |
34dc7c2f | 202 | |
428870ff BB |
203 | /* |
204 | * Per-dataset state. | |
205 | */ | |
206 | typedef struct ztest_ds { | |
207 | objset_t *zd_os; | |
208 | zilog_t *zd_zilog; | |
209 | uint64_t zd_seq; | |
210 | ztest_od_t *zd_od; /* debugging aid */ | |
211 | char zd_name[MAXNAMELEN]; | |
1e33ac1e | 212 | kmutex_t zd_dirobj_lock; |
428870ff BB |
213 | rll_t zd_object_lock[ZTEST_OBJECT_LOCKS]; |
214 | rll_t zd_range_lock[ZTEST_RANGE_LOCKS]; | |
215 | } ztest_ds_t; | |
216 | ||
217 | /* | |
218 | * Per-iteration state. | |
219 | */ | |
220 | typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id); | |
221 | ||
222 | typedef struct ztest_info { | |
223 | ztest_func_t *zi_func; /* test function */ | |
224 | uint64_t zi_iters; /* iterations per execution */ | |
225 | uint64_t *zi_interval; /* execute every <interval> seconds */ | |
226 | uint64_t zi_call_count; /* per-pass count */ | |
227 | uint64_t zi_call_time; /* per-pass time */ | |
228 | uint64_t zi_call_next; /* next time to call this function */ | |
229 | } ztest_info_t; | |
34dc7c2f BB |
230 | |
231 | /* | |
232 | * Note: these aren't static because we want dladdr() to work. | |
233 | */ | |
234 | ztest_func_t ztest_dmu_read_write; | |
235 | ztest_func_t ztest_dmu_write_parallel; | |
236 | ztest_func_t ztest_dmu_object_alloc_free; | |
428870ff | 237 | ztest_func_t ztest_dmu_commit_callbacks; |
34dc7c2f BB |
238 | ztest_func_t ztest_zap; |
239 | ztest_func_t ztest_zap_parallel; | |
428870ff BB |
240 | ztest_func_t ztest_zil_commit; |
241 | ztest_func_t ztest_dmu_read_write_zcopy; | |
34dc7c2f | 242 | ztest_func_t ztest_dmu_objset_create_destroy; |
428870ff BB |
243 | ztest_func_t ztest_dmu_prealloc; |
244 | ztest_func_t ztest_fzap; | |
34dc7c2f | 245 | ztest_func_t ztest_dmu_snapshot_create_destroy; |
428870ff BB |
246 | ztest_func_t ztest_dsl_prop_get_set; |
247 | ztest_func_t ztest_spa_prop_get_set; | |
34dc7c2f BB |
248 | ztest_func_t ztest_spa_create_destroy; |
249 | ztest_func_t ztest_fault_inject; | |
428870ff BB |
250 | ztest_func_t ztest_ddt_repair; |
251 | ztest_func_t ztest_dmu_snapshot_hold; | |
b128c09f | 252 | ztest_func_t ztest_spa_rename; |
428870ff BB |
253 | ztest_func_t ztest_scrub; |
254 | ztest_func_t ztest_dsl_dataset_promote_busy; | |
34dc7c2f BB |
255 | ztest_func_t ztest_vdev_attach_detach; |
256 | ztest_func_t ztest_vdev_LUN_growth; | |
257 | ztest_func_t ztest_vdev_add_remove; | |
b128c09f | 258 | ztest_func_t ztest_vdev_aux_add_remove; |
428870ff | 259 | ztest_func_t ztest_split_pool; |
34dc7c2f | 260 | |
428870ff BB |
261 | uint64_t zopt_always = 0ULL * NANOSEC; /* all the time */ |
262 | uint64_t zopt_incessant = 1ULL * NANOSEC / 10; /* every 1/10 second */ | |
263 | uint64_t zopt_often = 1ULL * NANOSEC; /* every second */ | |
264 | uint64_t zopt_sometimes = 10ULL * NANOSEC; /* every 10 seconds */ | |
265 | uint64_t zopt_rarely = 60ULL * NANOSEC; /* every 60 seconds */ | |
34dc7c2f BB |
266 | |
267 | ztest_info_t ztest_info[] = { | |
268 | { ztest_dmu_read_write, 1, &zopt_always }, | |
428870ff | 269 | { ztest_dmu_write_parallel, 10, &zopt_always }, |
34dc7c2f | 270 | { ztest_dmu_object_alloc_free, 1, &zopt_always }, |
428870ff | 271 | { ztest_dmu_commit_callbacks, 1, &zopt_always }, |
34dc7c2f BB |
272 | { ztest_zap, 30, &zopt_always }, |
273 | { ztest_zap_parallel, 100, &zopt_always }, | |
428870ff BB |
274 | { ztest_split_pool, 1, &zopt_always }, |
275 | { ztest_zil_commit, 1, &zopt_incessant }, | |
276 | { ztest_dmu_read_write_zcopy, 1, &zopt_often }, | |
277 | { ztest_dmu_objset_create_destroy, 1, &zopt_often }, | |
278 | { ztest_dsl_prop_get_set, 1, &zopt_often }, | |
279 | { ztest_spa_prop_get_set, 1, &zopt_sometimes }, | |
280 | #if 0 | |
281 | { ztest_dmu_prealloc, 1, &zopt_sometimes }, | |
282 | #endif | |
283 | { ztest_fzap, 1, &zopt_sometimes }, | |
284 | { ztest_dmu_snapshot_create_destroy, 1, &zopt_sometimes }, | |
285 | { ztest_spa_create_destroy, 1, &zopt_sometimes }, | |
34dc7c2f | 286 | { ztest_fault_inject, 1, &zopt_sometimes }, |
428870ff BB |
287 | { ztest_ddt_repair, 1, &zopt_sometimes }, |
288 | { ztest_dmu_snapshot_hold, 1, &zopt_sometimes }, | |
34dc7c2f | 289 | { ztest_spa_rename, 1, &zopt_rarely }, |
428870ff | 290 | { ztest_scrub, 1, &zopt_rarely }, |
9babb374 | 291 | { ztest_dsl_dataset_promote_busy, 1, &zopt_rarely }, |
428870ff BB |
292 | { ztest_vdev_attach_detach, 1, &zopt_rarely }, |
293 | { ztest_vdev_LUN_growth, 1, &zopt_rarely }, | |
294 | { ztest_vdev_add_remove, 1, &zopt_vdevtime }, | |
b128c09f | 295 | { ztest_vdev_aux_add_remove, 1, &zopt_vdevtime }, |
34dc7c2f BB |
296 | }; |
297 | ||
298 | #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t)) | |
299 | ||
428870ff BB |
300 | /* |
301 | * The following struct is used to hold a list of uncalled commit callbacks. | |
302 | * The callbacks are ordered by txg number. | |
303 | */ | |
304 | typedef struct ztest_cb_list { | |
1e33ac1e BB |
305 | kmutex_t zcl_callbacks_lock; |
306 | list_t zcl_callbacks; | |
428870ff | 307 | } ztest_cb_list_t; |
34dc7c2f BB |
308 | |
309 | /* | |
310 | * Stuff we need to share writably between parent and child. | |
311 | */ | |
312 | typedef struct ztest_shared { | |
428870ff BB |
313 | char *zs_pool; |
314 | spa_t *zs_spa; | |
315 | hrtime_t zs_proc_start; | |
316 | hrtime_t zs_proc_stop; | |
317 | hrtime_t zs_thread_start; | |
318 | hrtime_t zs_thread_stop; | |
319 | hrtime_t zs_thread_kill; | |
34dc7c2f | 320 | uint64_t zs_enospc_count; |
428870ff BB |
321 | uint64_t zs_vdev_next_leaf; |
322 | uint64_t zs_vdev_aux; | |
34dc7c2f BB |
323 | uint64_t zs_alloc; |
324 | uint64_t zs_space; | |
1e33ac1e BB |
325 | kmutex_t zs_vdev_lock; |
326 | krwlock_t zs_name_lock; | |
34dc7c2f | 327 | ztest_info_t zs_info[ZTEST_FUNCS]; |
428870ff BB |
328 | uint64_t zs_splits; |
329 | uint64_t zs_mirrors; | |
330 | ztest_ds_t zs_zd[]; | |
34dc7c2f BB |
331 | } ztest_shared_t; |
332 | ||
428870ff BB |
333 | #define ID_PARALLEL -1ULL |
334 | ||
34dc7c2f | 335 | static char ztest_dev_template[] = "%s/%s.%llua"; |
b128c09f | 336 | static char ztest_aux_template[] = "%s/%s.%s.%llu"; |
428870ff BB |
337 | ztest_shared_t *ztest_shared; |
338 | uint64_t *ztest_seq; | |
34dc7c2f BB |
339 | |
340 | static int ztest_random_fd; | |
341 | static int ztest_dump_core = 1; | |
342 | ||
b128c09f | 343 | static boolean_t ztest_exiting; |
34dc7c2f | 344 | |
428870ff BB |
345 | /* Global commit callback list */ |
346 | static ztest_cb_list_t zcl; | |
090ff092 RC |
347 | /* Commit cb delay */ |
348 | static uint64_t zc_min_txg_delay = UINT64_MAX; | |
349 | static int zc_cb_counter = 0; | |
350 | ||
351 | /* | |
352 | * Minimum number of commit callbacks that need to be registered for us to check | |
353 | * whether the minimum txg delay is acceptable. | |
354 | */ | |
355 | #define ZTEST_COMMIT_CB_MIN_REG 100 | |
356 | ||
357 | /* | |
358 | * If a number of txgs equal to this threshold have been created after a commit | |
359 | * callback has been registered but not called, then we assume there is an | |
360 | * implementation bug. | |
361 | */ | |
362 | #define ZTEST_COMMIT_CB_THRESH (TXG_CONCURRENT_STATES + 1000) | |
428870ff | 363 | |
34dc7c2f | 364 | extern uint64_t metaslab_gang_bang; |
9babb374 | 365 | extern uint64_t metaslab_df_alloc_threshold; |
428870ff | 366 | static uint64_t metaslab_sz; |
34dc7c2f | 367 | |
428870ff BB |
368 | enum ztest_object { |
369 | ZTEST_META_DNODE = 0, | |
370 | ZTEST_DIROBJ, | |
371 | ZTEST_OBJECTS | |
372 | }; | |
34dc7c2f BB |
373 | |
374 | static void usage(boolean_t) __NORETURN; | |
375 | ||
376 | /* | |
377 | * These libumem hooks provide a reasonable set of defaults for the allocator's | |
378 | * debugging facilities. | |
379 | */ | |
380 | const char * | |
0bc8fd78 | 381 | _umem_debug_init(void) |
34dc7c2f BB |
382 | { |
383 | return ("default,verbose"); /* $UMEM_DEBUG setting */ | |
384 | } | |
385 | ||
386 | const char * | |
387 | _umem_logging_init(void) | |
388 | { | |
389 | return ("fail,contents"); /* $UMEM_LOGGING setting */ | |
390 | } | |
391 | ||
392 | #define FATAL_MSG_SZ 1024 | |
393 | ||
394 | char *fatal_msg; | |
395 | ||
396 | static void | |
397 | fatal(int do_perror, char *message, ...) | |
398 | { | |
399 | va_list args; | |
400 | int save_errno = errno; | |
401 | char buf[FATAL_MSG_SZ]; | |
402 | ||
403 | (void) fflush(stdout); | |
404 | ||
405 | va_start(args, message); | |
406 | (void) sprintf(buf, "ztest: "); | |
407 | /* LINTED */ | |
408 | (void) vsprintf(buf + strlen(buf), message, args); | |
409 | va_end(args); | |
410 | if (do_perror) { | |
411 | (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf), | |
412 | ": %s", strerror(save_errno)); | |
413 | } | |
414 | (void) fprintf(stderr, "%s\n", buf); | |
415 | fatal_msg = buf; /* to ease debugging */ | |
416 | if (ztest_dump_core) | |
417 | abort(); | |
418 | exit(3); | |
419 | } | |
420 | ||
421 | static int | |
422 | str2shift(const char *buf) | |
423 | { | |
424 | const char *ends = "BKMGTPEZ"; | |
425 | int i; | |
426 | ||
427 | if (buf[0] == '\0') | |
428 | return (0); | |
429 | for (i = 0; i < strlen(ends); i++) { | |
430 | if (toupper(buf[0]) == ends[i]) | |
431 | break; | |
432 | } | |
433 | if (i == strlen(ends)) { | |
434 | (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", | |
435 | buf); | |
436 | usage(B_FALSE); | |
437 | } | |
438 | if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) { | |
439 | return (10*i); | |
440 | } | |
441 | (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf); | |
442 | usage(B_FALSE); | |
443 | /* NOTREACHED */ | |
444 | } | |
445 | ||
446 | static uint64_t | |
447 | nicenumtoull(const char *buf) | |
448 | { | |
449 | char *end; | |
450 | uint64_t val; | |
451 | ||
452 | val = strtoull(buf, &end, 0); | |
453 | if (end == buf) { | |
454 | (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf); | |
455 | usage(B_FALSE); | |
456 | } else if (end[0] == '.') { | |
457 | double fval = strtod(buf, &end); | |
458 | fval *= pow(2, str2shift(end)); | |
459 | if (fval > UINT64_MAX) { | |
460 | (void) fprintf(stderr, "ztest: value too large: %s\n", | |
461 | buf); | |
462 | usage(B_FALSE); | |
463 | } | |
464 | val = (uint64_t)fval; | |
465 | } else { | |
466 | int shift = str2shift(end); | |
467 | if (shift >= 64 || (val << shift) >> shift != val) { | |
468 | (void) fprintf(stderr, "ztest: value too large: %s\n", | |
469 | buf); | |
470 | usage(B_FALSE); | |
471 | } | |
472 | val <<= shift; | |
473 | } | |
474 | return (val); | |
475 | } | |
476 | ||
477 | static void | |
478 | usage(boolean_t requested) | |
479 | { | |
480 | char nice_vdev_size[10]; | |
481 | char nice_gang_bang[10]; | |
482 | FILE *fp = requested ? stdout : stderr; | |
483 | ||
484 | nicenum(zopt_vdev_size, nice_vdev_size); | |
485 | nicenum(metaslab_gang_bang, nice_gang_bang); | |
486 | ||
487 | (void) fprintf(fp, "Usage: %s\n" | |
488 | "\t[-v vdevs (default: %llu)]\n" | |
489 | "\t[-s size_of_each_vdev (default: %s)]\n" | |
428870ff | 490 | "\t[-a alignment_shift (default: %d)] use 0 for random\n" |
34dc7c2f BB |
491 | "\t[-m mirror_copies (default: %d)]\n" |
492 | "\t[-r raidz_disks (default: %d)]\n" | |
493 | "\t[-R raidz_parity (default: %d)]\n" | |
494 | "\t[-d datasets (default: %d)]\n" | |
495 | "\t[-t threads (default: %d)]\n" | |
496 | "\t[-g gang_block_threshold (default: %s)]\n" | |
428870ff BB |
497 | "\t[-i init_count (default: %d)] initialize pool i times\n" |
498 | "\t[-k kill_percentage (default: %llu%%)]\n" | |
34dc7c2f | 499 | "\t[-p pool_name (default: %s)]\n" |
428870ff BB |
500 | "\t[-f dir (default: %s)] file directory for vdev files\n" |
501 | "\t[-V] verbose (use multiple times for ever more blather)\n" | |
502 | "\t[-E] use existing pool instead of creating new one\n" | |
503 | "\t[-T time (default: %llu sec)] total run time\n" | |
504 | "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n" | |
505 | "\t[-P passtime (default: %llu sec)] time per pass\n" | |
34dc7c2f BB |
506 | "\t[-h] (print help)\n" |
507 | "", | |
508 | cmdname, | |
509 | (u_longlong_t)zopt_vdevs, /* -v */ | |
510 | nice_vdev_size, /* -s */ | |
511 | zopt_ashift, /* -a */ | |
512 | zopt_mirrors, /* -m */ | |
513 | zopt_raidz, /* -r */ | |
514 | zopt_raidz_parity, /* -R */ | |
515 | zopt_datasets, /* -d */ | |
516 | zopt_threads, /* -t */ | |
517 | nice_gang_bang, /* -g */ | |
518 | zopt_init, /* -i */ | |
519 | (u_longlong_t)zopt_killrate, /* -k */ | |
520 | zopt_pool, /* -p */ | |
521 | zopt_dir, /* -f */ | |
522 | (u_longlong_t)zopt_time, /* -T */ | |
428870ff | 523 | (u_longlong_t)zopt_maxloops, /* -F */ |
b128c09f | 524 | (u_longlong_t)zopt_passtime); /* -P */ |
34dc7c2f BB |
525 | exit(requested ? 0 : 1); |
526 | } | |
527 | ||
34dc7c2f BB |
528 | static void |
529 | process_options(int argc, char **argv) | |
530 | { | |
531 | int opt; | |
532 | uint64_t value; | |
533 | ||
534 | /* By default, test gang blocks for blocks 32K and greater */ | |
535 | metaslab_gang_bang = 32 << 10; | |
536 | ||
34dc7c2f | 537 | while ((opt = getopt(argc, argv, |
428870ff | 538 | "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:")) != EOF) { |
34dc7c2f BB |
539 | value = 0; |
540 | switch (opt) { | |
541 | case 'v': | |
542 | case 's': | |
543 | case 'a': | |
544 | case 'm': | |
545 | case 'r': | |
546 | case 'R': | |
547 | case 'd': | |
548 | case 't': | |
549 | case 'g': | |
550 | case 'i': | |
551 | case 'k': | |
552 | case 'T': | |
553 | case 'P': | |
428870ff | 554 | case 'F': |
34dc7c2f BB |
555 | value = nicenumtoull(optarg); |
556 | } | |
557 | switch (opt) { | |
558 | case 'v': | |
559 | zopt_vdevs = value; | |
560 | break; | |
561 | case 's': | |
562 | zopt_vdev_size = MAX(SPA_MINDEVSIZE, value); | |
563 | break; | |
564 | case 'a': | |
565 | zopt_ashift = value; | |
566 | break; | |
567 | case 'm': | |
568 | zopt_mirrors = value; | |
569 | break; | |
570 | case 'r': | |
571 | zopt_raidz = MAX(1, value); | |
572 | break; | |
573 | case 'R': | |
45d1cae3 | 574 | zopt_raidz_parity = MIN(MAX(value, 1), 3); |
34dc7c2f BB |
575 | break; |
576 | case 'd': | |
577 | zopt_datasets = MAX(1, value); | |
578 | break; | |
579 | case 't': | |
580 | zopt_threads = MAX(1, value); | |
581 | break; | |
582 | case 'g': | |
583 | metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1, value); | |
584 | break; | |
585 | case 'i': | |
586 | zopt_init = value; | |
587 | break; | |
588 | case 'k': | |
589 | zopt_killrate = value; | |
590 | break; | |
591 | case 'p': | |
592 | zopt_pool = strdup(optarg); | |
593 | break; | |
594 | case 'f': | |
595 | zopt_dir = strdup(optarg); | |
596 | break; | |
597 | case 'V': | |
598 | zopt_verbose++; | |
599 | break; | |
600 | case 'E': | |
601 | zopt_init = 0; | |
602 | break; | |
603 | case 'T': | |
604 | zopt_time = value; | |
605 | break; | |
606 | case 'P': | |
607 | zopt_passtime = MAX(1, value); | |
608 | break; | |
428870ff BB |
609 | case 'F': |
610 | zopt_maxloops = MAX(1, value); | |
611 | break; | |
34dc7c2f BB |
612 | case 'h': |
613 | usage(B_TRUE); | |
614 | break; | |
615 | case '?': | |
616 | default: | |
617 | usage(B_FALSE); | |
618 | break; | |
619 | } | |
620 | } | |
621 | ||
622 | zopt_raidz_parity = MIN(zopt_raidz_parity, zopt_raidz - 1); | |
623 | ||
428870ff BB |
624 | zopt_vdevtime = (zopt_vdevs > 0 ? zopt_time * NANOSEC / zopt_vdevs : |
625 | UINT64_MAX >> 2); | |
626 | } | |
627 | ||
628 | static void | |
629 | ztest_kill(ztest_shared_t *zs) | |
630 | { | |
631 | zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(zs->zs_spa)); | |
632 | zs->zs_space = metaslab_class_get_space(spa_normal_class(zs->zs_spa)); | |
633 | (void) kill(getpid(), SIGKILL); | |
634 | } | |
635 | ||
636 | static uint64_t | |
637 | ztest_random(uint64_t range) | |
638 | { | |
639 | uint64_t r; | |
640 | ||
641 | if (range == 0) | |
642 | return (0); | |
643 | ||
644 | if (read(ztest_random_fd, &r, sizeof (r)) != sizeof (r)) | |
645 | fatal(1, "short read from /dev/urandom"); | |
646 | ||
647 | return (r % range); | |
648 | } | |
649 | ||
650 | /* ARGSUSED */ | |
651 | static void | |
652 | ztest_record_enospc(const char *s) | |
653 | { | |
654 | ztest_shared->zs_enospc_count++; | |
34dc7c2f BB |
655 | } |
656 | ||
657 | static uint64_t | |
658 | ztest_get_ashift(void) | |
659 | { | |
660 | if (zopt_ashift == 0) | |
661 | return (SPA_MINBLOCKSHIFT + ztest_random(3)); | |
662 | return (zopt_ashift); | |
663 | } | |
664 | ||
665 | static nvlist_t * | |
b128c09f | 666 | make_vdev_file(char *path, char *aux, size_t size, uint64_t ashift) |
34dc7c2f | 667 | { |
b128c09f | 668 | char pathbuf[MAXPATHLEN]; |
34dc7c2f | 669 | uint64_t vdev; |
34dc7c2f BB |
670 | nvlist_t *file; |
671 | ||
b128c09f BB |
672 | if (ashift == 0) |
673 | ashift = ztest_get_ashift(); | |
674 | ||
675 | if (path == NULL) { | |
676 | path = pathbuf; | |
677 | ||
678 | if (aux != NULL) { | |
679 | vdev = ztest_shared->zs_vdev_aux; | |
680 | (void) sprintf(path, ztest_aux_template, | |
681 | zopt_dir, zopt_pool, aux, vdev); | |
682 | } else { | |
428870ff | 683 | vdev = ztest_shared->zs_vdev_next_leaf++; |
b128c09f BB |
684 | (void) sprintf(path, ztest_dev_template, |
685 | zopt_dir, zopt_pool, vdev); | |
686 | } | |
687 | } | |
34dc7c2f | 688 | |
b128c09f BB |
689 | if (size != 0) { |
690 | int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666); | |
34dc7c2f | 691 | if (fd == -1) |
b128c09f | 692 | fatal(1, "can't open %s", path); |
34dc7c2f | 693 | if (ftruncate(fd, size) != 0) |
b128c09f | 694 | fatal(1, "can't ftruncate %s", path); |
34dc7c2f BB |
695 | (void) close(fd); |
696 | } | |
697 | ||
698 | VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0); | |
699 | VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0); | |
b128c09f | 700 | VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0); |
34dc7c2f BB |
701 | VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0); |
702 | ||
703 | return (file); | |
704 | } | |
705 | ||
706 | static nvlist_t * | |
b128c09f | 707 | make_vdev_raidz(char *path, char *aux, size_t size, uint64_t ashift, int r) |
34dc7c2f BB |
708 | { |
709 | nvlist_t *raidz, **child; | |
710 | int c; | |
711 | ||
712 | if (r < 2) | |
b128c09f | 713 | return (make_vdev_file(path, aux, size, ashift)); |
34dc7c2f BB |
714 | child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL); |
715 | ||
716 | for (c = 0; c < r; c++) | |
b128c09f | 717 | child[c] = make_vdev_file(path, aux, size, ashift); |
34dc7c2f BB |
718 | |
719 | VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0); | |
720 | VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE, | |
721 | VDEV_TYPE_RAIDZ) == 0); | |
722 | VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY, | |
723 | zopt_raidz_parity) == 0); | |
724 | VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN, | |
725 | child, r) == 0); | |
726 | ||
727 | for (c = 0; c < r; c++) | |
728 | nvlist_free(child[c]); | |
729 | ||
730 | umem_free(child, r * sizeof (nvlist_t *)); | |
731 | ||
732 | return (raidz); | |
733 | } | |
734 | ||
735 | static nvlist_t * | |
b128c09f BB |
736 | make_vdev_mirror(char *path, char *aux, size_t size, uint64_t ashift, |
737 | int r, int m) | |
34dc7c2f BB |
738 | { |
739 | nvlist_t *mirror, **child; | |
740 | int c; | |
741 | ||
742 | if (m < 1) | |
b128c09f | 743 | return (make_vdev_raidz(path, aux, size, ashift, r)); |
34dc7c2f BB |
744 | |
745 | child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL); | |
746 | ||
747 | for (c = 0; c < m; c++) | |
b128c09f | 748 | child[c] = make_vdev_raidz(path, aux, size, ashift, r); |
34dc7c2f BB |
749 | |
750 | VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0); | |
751 | VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE, | |
752 | VDEV_TYPE_MIRROR) == 0); | |
753 | VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN, | |
754 | child, m) == 0); | |
34dc7c2f BB |
755 | |
756 | for (c = 0; c < m; c++) | |
757 | nvlist_free(child[c]); | |
758 | ||
759 | umem_free(child, m * sizeof (nvlist_t *)); | |
760 | ||
761 | return (mirror); | |
762 | } | |
763 | ||
764 | static nvlist_t * | |
b128c09f BB |
765 | make_vdev_root(char *path, char *aux, size_t size, uint64_t ashift, |
766 | int log, int r, int m, int t) | |
34dc7c2f BB |
767 | { |
768 | nvlist_t *root, **child; | |
769 | int c; | |
770 | ||
771 | ASSERT(t > 0); | |
772 | ||
773 | child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL); | |
774 | ||
b128c09f BB |
775 | for (c = 0; c < t; c++) { |
776 | child[c] = make_vdev_mirror(path, aux, size, ashift, r, m); | |
777 | VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG, | |
778 | log) == 0); | |
779 | } | |
34dc7c2f BB |
780 | |
781 | VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0); | |
782 | VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0); | |
b128c09f | 783 | VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN, |
34dc7c2f BB |
784 | child, t) == 0); |
785 | ||
786 | for (c = 0; c < t; c++) | |
787 | nvlist_free(child[c]); | |
788 | ||
789 | umem_free(child, t * sizeof (nvlist_t *)); | |
790 | ||
791 | return (root); | |
792 | } | |
793 | ||
428870ff BB |
794 | static int |
795 | ztest_random_blocksize(void) | |
34dc7c2f | 796 | { |
428870ff BB |
797 | return (1 << (SPA_MINBLOCKSHIFT + |
798 | ztest_random(SPA_MAXBLOCKSHIFT - SPA_MINBLOCKSHIFT + 1))); | |
799 | } | |
34dc7c2f | 800 | |
428870ff BB |
801 | static int |
802 | ztest_random_ibshift(void) | |
803 | { | |
804 | return (DN_MIN_INDBLKSHIFT + | |
805 | ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1)); | |
34dc7c2f BB |
806 | } |
807 | ||
428870ff BB |
808 | static uint64_t |
809 | ztest_random_vdev_top(spa_t *spa, boolean_t log_ok) | |
34dc7c2f | 810 | { |
428870ff BB |
811 | uint64_t top; |
812 | vdev_t *rvd = spa->spa_root_vdev; | |
813 | vdev_t *tvd; | |
34dc7c2f | 814 | |
428870ff | 815 | ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); |
34dc7c2f | 816 | |
428870ff BB |
817 | do { |
818 | top = ztest_random(rvd->vdev_children); | |
819 | tvd = rvd->vdev_child[top]; | |
820 | } while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) || | |
821 | tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL); | |
34dc7c2f | 822 | |
428870ff | 823 | return (top); |
34dc7c2f BB |
824 | } |
825 | ||
428870ff BB |
826 | static uint64_t |
827 | ztest_random_dsl_prop(zfs_prop_t prop) | |
34dc7c2f | 828 | { |
428870ff BB |
829 | uint64_t value; |
830 | ||
831 | do { | |
832 | value = zfs_prop_random_value(prop, ztest_random(-1ULL)); | |
833 | } while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF); | |
834 | ||
835 | return (value); | |
34dc7c2f BB |
836 | } |
837 | ||
34dc7c2f | 838 | static int |
428870ff BB |
839 | ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value, |
840 | boolean_t inherit) | |
34dc7c2f | 841 | { |
428870ff BB |
842 | const char *propname = zfs_prop_to_name(prop); |
843 | const char *valname; | |
844 | char setpoint[MAXPATHLEN]; | |
845 | uint64_t curval; | |
34dc7c2f BB |
846 | int error; |
847 | ||
428870ff BB |
848 | error = dsl_prop_set(osname, propname, |
849 | (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), | |
850 | sizeof (value), 1, &value); | |
34dc7c2f | 851 | |
428870ff BB |
852 | if (error == ENOSPC) { |
853 | ztest_record_enospc(FTAG); | |
34dc7c2f BB |
854 | return (error); |
855 | } | |
34dc7c2f | 856 | ASSERT3U(error, ==, 0); |
34dc7c2f | 857 | |
428870ff BB |
858 | VERIFY3U(dsl_prop_get(osname, propname, sizeof (curval), |
859 | 1, &curval, setpoint), ==, 0); | |
860 | ||
861 | if (zopt_verbose >= 6) { | |
862 | VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0); | |
863 | (void) printf("%s %s = %s at '%s'\n", | |
864 | osname, propname, valname, setpoint); | |
34dc7c2f BB |
865 | } |
866 | ||
867 | return (error); | |
868 | } | |
869 | ||
870 | static int | |
428870ff | 871 | ztest_spa_prop_set_uint64(ztest_shared_t *zs, zpool_prop_t prop, uint64_t value) |
34dc7c2f | 872 | { |
428870ff BB |
873 | spa_t *spa = zs->zs_spa; |
874 | nvlist_t *props = NULL; | |
34dc7c2f BB |
875 | int error; |
876 | ||
428870ff BB |
877 | VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0); |
878 | VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0); | |
34dc7c2f | 879 | |
428870ff BB |
880 | error = spa_prop_set(spa, props); |
881 | ||
882 | nvlist_free(props); | |
883 | ||
884 | if (error == ENOSPC) { | |
885 | ztest_record_enospc(FTAG); | |
34dc7c2f BB |
886 | return (error); |
887 | } | |
428870ff | 888 | ASSERT3U(error, ==, 0); |
34dc7c2f BB |
889 | |
890 | return (error); | |
891 | } | |
892 | ||
428870ff BB |
893 | static void |
894 | ztest_rll_init(rll_t *rll) | |
895 | { | |
896 | rll->rll_writer = NULL; | |
897 | rll->rll_readers = 0; | |
1e33ac1e BB |
898 | mutex_init(&rll->rll_lock, NULL, MUTEX_DEFAULT, NULL); |
899 | cv_init(&rll->rll_cv, NULL, CV_DEFAULT, NULL); | |
428870ff | 900 | } |
34dc7c2f | 901 | |
428870ff BB |
902 | static void |
903 | ztest_rll_destroy(rll_t *rll) | |
34dc7c2f | 904 | { |
428870ff BB |
905 | ASSERT(rll->rll_writer == NULL); |
906 | ASSERT(rll->rll_readers == 0); | |
1e33ac1e BB |
907 | mutex_destroy(&rll->rll_lock); |
908 | cv_destroy(&rll->rll_cv); | |
428870ff | 909 | } |
34dc7c2f | 910 | |
428870ff BB |
911 | static void |
912 | ztest_rll_lock(rll_t *rll, rl_type_t type) | |
913 | { | |
1e33ac1e | 914 | mutex_enter(&rll->rll_lock); |
34dc7c2f | 915 | |
428870ff BB |
916 | if (type == RL_READER) { |
917 | while (rll->rll_writer != NULL) | |
1e33ac1e | 918 | (void) cv_wait(&rll->rll_cv, &rll->rll_lock); |
428870ff BB |
919 | rll->rll_readers++; |
920 | } else { | |
921 | while (rll->rll_writer != NULL || rll->rll_readers) | |
1e33ac1e | 922 | (void) cv_wait(&rll->rll_cv, &rll->rll_lock); |
428870ff BB |
923 | rll->rll_writer = curthread; |
924 | } | |
34dc7c2f | 925 | |
1e33ac1e | 926 | mutex_exit(&rll->rll_lock); |
428870ff | 927 | } |
34dc7c2f | 928 | |
428870ff BB |
929 | static void |
930 | ztest_rll_unlock(rll_t *rll) | |
931 | { | |
1e33ac1e | 932 | mutex_enter(&rll->rll_lock); |
34dc7c2f | 933 | |
428870ff BB |
934 | if (rll->rll_writer) { |
935 | ASSERT(rll->rll_readers == 0); | |
936 | rll->rll_writer = NULL; | |
937 | } else { | |
938 | ASSERT(rll->rll_readers != 0); | |
939 | ASSERT(rll->rll_writer == NULL); | |
940 | rll->rll_readers--; | |
941 | } | |
34dc7c2f | 942 | |
428870ff | 943 | if (rll->rll_writer == NULL && rll->rll_readers == 0) |
1e33ac1e | 944 | cv_broadcast(&rll->rll_cv); |
428870ff | 945 | |
1e33ac1e | 946 | mutex_exit(&rll->rll_lock); |
34dc7c2f BB |
947 | } |
948 | ||
428870ff BB |
949 | static void |
950 | ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type) | |
b128c09f | 951 | { |
428870ff | 952 | rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)]; |
b128c09f | 953 | |
428870ff BB |
954 | ztest_rll_lock(rll, type); |
955 | } | |
b128c09f | 956 | |
428870ff BB |
957 | static void |
958 | ztest_object_unlock(ztest_ds_t *zd, uint64_t object) | |
959 | { | |
960 | rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)]; | |
b128c09f | 961 | |
428870ff | 962 | ztest_rll_unlock(rll); |
b128c09f BB |
963 | } |
964 | ||
428870ff BB |
965 | static rl_t * |
966 | ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset, | |
967 | uint64_t size, rl_type_t type) | |
34dc7c2f | 968 | { |
428870ff BB |
969 | uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1)); |
970 | rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)]; | |
971 | rl_t *rl; | |
34dc7c2f | 972 | |
428870ff BB |
973 | rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL); |
974 | rl->rl_object = object; | |
975 | rl->rl_offset = offset; | |
976 | rl->rl_size = size; | |
977 | rl->rl_lock = rll; | |
34dc7c2f | 978 | |
428870ff BB |
979 | ztest_rll_lock(rll, type); |
980 | ||
981 | return (rl); | |
982 | } | |
34dc7c2f | 983 | |
428870ff BB |
984 | static void |
985 | ztest_range_unlock(rl_t *rl) | |
986 | { | |
987 | rll_t *rll = rl->rl_lock; | |
34dc7c2f | 988 | |
428870ff | 989 | ztest_rll_unlock(rll); |
34dc7c2f | 990 | |
428870ff BB |
991 | umem_free(rl, sizeof (*rl)); |
992 | } | |
34dc7c2f | 993 | |
428870ff BB |
994 | static void |
995 | ztest_zd_init(ztest_ds_t *zd, objset_t *os) | |
996 | { | |
997 | zd->zd_os = os; | |
998 | zd->zd_zilog = dmu_objset_zil(os); | |
999 | zd->zd_seq = 0; | |
1000 | dmu_objset_name(os, zd->zd_name); | |
d6320ddb | 1001 | int l; |
428870ff | 1002 | |
1e33ac1e | 1003 | mutex_init(&zd->zd_dirobj_lock, NULL, MUTEX_DEFAULT, NULL); |
34dc7c2f | 1004 | |
d6320ddb | 1005 | for (l = 0; l < ZTEST_OBJECT_LOCKS; l++) |
428870ff | 1006 | ztest_rll_init(&zd->zd_object_lock[l]); |
34dc7c2f | 1007 | |
d6320ddb | 1008 | for (l = 0; l < ZTEST_RANGE_LOCKS; l++) |
428870ff | 1009 | ztest_rll_init(&zd->zd_range_lock[l]); |
34dc7c2f BB |
1010 | } |
1011 | ||
428870ff BB |
1012 | static void |
1013 | ztest_zd_fini(ztest_ds_t *zd) | |
34dc7c2f | 1014 | { |
d6320ddb BB |
1015 | int l; |
1016 | ||
1e33ac1e | 1017 | mutex_destroy(&zd->zd_dirobj_lock); |
34dc7c2f | 1018 | |
d6320ddb | 1019 | for (l = 0; l < ZTEST_OBJECT_LOCKS; l++) |
428870ff | 1020 | ztest_rll_destroy(&zd->zd_object_lock[l]); |
b128c09f | 1021 | |
d6320ddb | 1022 | for (l = 0; l < ZTEST_RANGE_LOCKS; l++) |
428870ff BB |
1023 | ztest_rll_destroy(&zd->zd_range_lock[l]); |
1024 | } | |
b128c09f | 1025 | |
428870ff | 1026 | #define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT) |
b128c09f | 1027 | |
428870ff BB |
1028 | static uint64_t |
1029 | ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag) | |
1030 | { | |
1031 | uint64_t txg; | |
1032 | int error; | |
1033 | ||
1034 | /* | |
1035 | * Attempt to assign tx to some transaction group. | |
1036 | */ | |
1037 | error = dmu_tx_assign(tx, txg_how); | |
1038 | if (error) { | |
1039 | if (error == ERESTART) { | |
1040 | ASSERT(txg_how == TXG_NOWAIT); | |
1041 | dmu_tx_wait(tx); | |
1042 | } else { | |
1043 | ASSERT3U(error, ==, ENOSPC); | |
1044 | ztest_record_enospc(tag); | |
1045 | } | |
1046 | dmu_tx_abort(tx); | |
1047 | return (0); | |
1048 | } | |
1049 | txg = dmu_tx_get_txg(tx); | |
1050 | ASSERT(txg != 0); | |
1051 | return (txg); | |
1052 | } | |
1053 | ||
1054 | static void | |
1055 | ztest_pattern_set(void *buf, uint64_t size, uint64_t value) | |
1056 | { | |
1057 | uint64_t *ip = buf; | |
1058 | uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size); | |
1059 | ||
1060 | while (ip < ip_end) | |
1061 | *ip++ = value; | |
1062 | } | |
1063 | ||
1fde1e37 | 1064 | #ifndef NDEBUG |
428870ff BB |
1065 | static boolean_t |
1066 | ztest_pattern_match(void *buf, uint64_t size, uint64_t value) | |
1067 | { | |
1068 | uint64_t *ip = buf; | |
1069 | uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size); | |
1070 | uint64_t diff = 0; | |
1071 | ||
1072 | while (ip < ip_end) | |
1073 | diff |= (value - *ip++); | |
1074 | ||
1075 | return (diff == 0); | |
1076 | } | |
1fde1e37 | 1077 | #endif |
428870ff BB |
1078 | |
1079 | static void | |
1080 | ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object, | |
1081 | uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg) | |
1082 | { | |
1083 | bt->bt_magic = BT_MAGIC; | |
1084 | bt->bt_objset = dmu_objset_id(os); | |
1085 | bt->bt_object = object; | |
1086 | bt->bt_offset = offset; | |
1087 | bt->bt_gen = gen; | |
1088 | bt->bt_txg = txg; | |
1089 | bt->bt_crtxg = crtxg; | |
1090 | } | |
1091 | ||
1092 | static void | |
1093 | ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object, | |
1094 | uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg) | |
1095 | { | |
1096 | ASSERT(bt->bt_magic == BT_MAGIC); | |
1097 | ASSERT(bt->bt_objset == dmu_objset_id(os)); | |
1098 | ASSERT(bt->bt_object == object); | |
1099 | ASSERT(bt->bt_offset == offset); | |
1100 | ASSERT(bt->bt_gen <= gen); | |
1101 | ASSERT(bt->bt_txg <= txg); | |
1102 | ASSERT(bt->bt_crtxg == crtxg); | |
1103 | } | |
1104 | ||
1105 | static ztest_block_tag_t * | |
1106 | ztest_bt_bonus(dmu_buf_t *db) | |
1107 | { | |
1108 | dmu_object_info_t doi; | |
1109 | ztest_block_tag_t *bt; | |
1110 | ||
1111 | dmu_object_info_from_db(db, &doi); | |
1112 | ASSERT3U(doi.doi_bonus_size, <=, db->db_size); | |
1113 | ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt)); | |
1114 | bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt)); | |
1115 | ||
1116 | return (bt); | |
1117 | } | |
1118 | ||
1119 | /* | |
1120 | * ZIL logging ops | |
1121 | */ | |
1122 | ||
1123 | #define lrz_type lr_mode | |
1124 | #define lrz_blocksize lr_uid | |
1125 | #define lrz_ibshift lr_gid | |
1126 | #define lrz_bonustype lr_rdev | |
1127 | #define lrz_bonuslen lr_crtime[1] | |
1128 | ||
572e2857 | 1129 | static void |
428870ff BB |
1130 | ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr) |
1131 | { | |
1132 | char *name = (void *)(lr + 1); /* name follows lr */ | |
1133 | size_t namesize = strlen(name) + 1; | |
1134 | itx_t *itx; | |
1135 | ||
1136 | if (zil_replaying(zd->zd_zilog, tx)) | |
572e2857 | 1137 | return; |
428870ff BB |
1138 | |
1139 | itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize); | |
1140 | bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, | |
1141 | sizeof (*lr) + namesize - sizeof (lr_t)); | |
1142 | ||
572e2857 | 1143 | zil_itx_assign(zd->zd_zilog, itx, tx); |
428870ff BB |
1144 | } |
1145 | ||
572e2857 BB |
1146 | static void |
1147 | ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object) | |
428870ff BB |
1148 | { |
1149 | char *name = (void *)(lr + 1); /* name follows lr */ | |
1150 | size_t namesize = strlen(name) + 1; | |
1151 | itx_t *itx; | |
1152 | ||
1153 | if (zil_replaying(zd->zd_zilog, tx)) | |
572e2857 | 1154 | return; |
428870ff BB |
1155 | |
1156 | itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize); | |
1157 | bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, | |
1158 | sizeof (*lr) + namesize - sizeof (lr_t)); | |
1159 | ||
572e2857 BB |
1160 | itx->itx_oid = object; |
1161 | zil_itx_assign(zd->zd_zilog, itx, tx); | |
428870ff BB |
1162 | } |
1163 | ||
572e2857 | 1164 | static void |
428870ff BB |
1165 | ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr) |
1166 | { | |
1167 | itx_t *itx; | |
1168 | itx_wr_state_t write_state = ztest_random(WR_NUM_STATES); | |
1169 | ||
1170 | if (zil_replaying(zd->zd_zilog, tx)) | |
572e2857 | 1171 | return; |
428870ff BB |
1172 | |
1173 | if (lr->lr_length > ZIL_MAX_LOG_DATA) | |
1174 | write_state = WR_INDIRECT; | |
1175 | ||
1176 | itx = zil_itx_create(TX_WRITE, | |
1177 | sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0)); | |
1178 | ||
1179 | if (write_state == WR_COPIED && | |
1180 | dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length, | |
1181 | ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) { | |
1182 | zil_itx_destroy(itx); | |
1183 | itx = zil_itx_create(TX_WRITE, sizeof (*lr)); | |
1184 | write_state = WR_NEED_COPY; | |
1185 | } | |
1186 | itx->itx_private = zd; | |
1187 | itx->itx_wr_state = write_state; | |
1188 | itx->itx_sync = (ztest_random(8) == 0); | |
1189 | itx->itx_sod += (write_state == WR_NEED_COPY ? lr->lr_length : 0); | |
1190 | ||
1191 | bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, | |
1192 | sizeof (*lr) - sizeof (lr_t)); | |
1193 | ||
572e2857 | 1194 | zil_itx_assign(zd->zd_zilog, itx, tx); |
428870ff BB |
1195 | } |
1196 | ||
572e2857 | 1197 | static void |
428870ff BB |
1198 | ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr) |
1199 | { | |
1200 | itx_t *itx; | |
1201 | ||
1202 | if (zil_replaying(zd->zd_zilog, tx)) | |
572e2857 | 1203 | return; |
428870ff BB |
1204 | |
1205 | itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr)); | |
1206 | bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, | |
1207 | sizeof (*lr) - sizeof (lr_t)); | |
1208 | ||
572e2857 BB |
1209 | itx->itx_sync = B_FALSE; |
1210 | zil_itx_assign(zd->zd_zilog, itx, tx); | |
428870ff BB |
1211 | } |
1212 | ||
572e2857 | 1213 | static void |
428870ff BB |
1214 | ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr) |
1215 | { | |
1216 | itx_t *itx; | |
1217 | ||
1218 | if (zil_replaying(zd->zd_zilog, tx)) | |
572e2857 | 1219 | return; |
428870ff BB |
1220 | |
1221 | itx = zil_itx_create(TX_SETATTR, sizeof (*lr)); | |
1222 | bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, | |
1223 | sizeof (*lr) - sizeof (lr_t)); | |
1224 | ||
572e2857 BB |
1225 | itx->itx_sync = B_FALSE; |
1226 | zil_itx_assign(zd->zd_zilog, itx, tx); | |
428870ff BB |
1227 | } |
1228 | ||
1229 | /* | |
1230 | * ZIL replay ops | |
1231 | */ | |
1232 | static int | |
1233 | ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap) | |
1234 | { | |
1235 | char *name = (void *)(lr + 1); /* name follows lr */ | |
1236 | objset_t *os = zd->zd_os; | |
1237 | ztest_block_tag_t *bbt; | |
1238 | dmu_buf_t *db; | |
1239 | dmu_tx_t *tx; | |
1240 | uint64_t txg; | |
1241 | int error = 0; | |
1242 | ||
1243 | if (byteswap) | |
1244 | byteswap_uint64_array(lr, sizeof (*lr)); | |
1245 | ||
1246 | ASSERT(lr->lr_doid == ZTEST_DIROBJ); | |
1247 | ASSERT(name[0] != '\0'); | |
1248 | ||
1249 | tx = dmu_tx_create(os); | |
1250 | ||
1251 | dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name); | |
1252 | ||
1253 | if (lr->lrz_type == DMU_OT_ZAP_OTHER) { | |
1254 | dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL); | |
1255 | } else { | |
1256 | dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); | |
1257 | } | |
1258 | ||
1259 | txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); | |
1260 | if (txg == 0) | |
1261 | return (ENOSPC); | |
1262 | ||
1263 | ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid); | |
1264 | ||
1265 | if (lr->lrz_type == DMU_OT_ZAP_OTHER) { | |
1266 | if (lr->lr_foid == 0) { | |
1267 | lr->lr_foid = zap_create(os, | |
1268 | lr->lrz_type, lr->lrz_bonustype, | |
1269 | lr->lrz_bonuslen, tx); | |
1270 | } else { | |
1271 | error = zap_create_claim(os, lr->lr_foid, | |
1272 | lr->lrz_type, lr->lrz_bonustype, | |
1273 | lr->lrz_bonuslen, tx); | |
1274 | } | |
1275 | } else { | |
1276 | if (lr->lr_foid == 0) { | |
1277 | lr->lr_foid = dmu_object_alloc(os, | |
1278 | lr->lrz_type, 0, lr->lrz_bonustype, | |
1279 | lr->lrz_bonuslen, tx); | |
1280 | } else { | |
1281 | error = dmu_object_claim(os, lr->lr_foid, | |
1282 | lr->lrz_type, 0, lr->lrz_bonustype, | |
1283 | lr->lrz_bonuslen, tx); | |
1284 | } | |
1285 | } | |
1286 | ||
1287 | if (error) { | |
1288 | ASSERT3U(error, ==, EEXIST); | |
1289 | ASSERT(zd->zd_zilog->zl_replay); | |
1290 | dmu_tx_commit(tx); | |
1291 | return (error); | |
1292 | } | |
1293 | ||
1294 | ASSERT(lr->lr_foid != 0); | |
1295 | ||
1296 | if (lr->lrz_type != DMU_OT_ZAP_OTHER) | |
1297 | VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid, | |
1298 | lr->lrz_blocksize, lr->lrz_ibshift, tx)); | |
1299 | ||
1300 | VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db)); | |
1301 | bbt = ztest_bt_bonus(db); | |
1302 | dmu_buf_will_dirty(db, tx); | |
1303 | ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg); | |
1304 | dmu_buf_rele(db, FTAG); | |
1305 | ||
1306 | VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1, | |
1307 | &lr->lr_foid, tx)); | |
1308 | ||
1309 | (void) ztest_log_create(zd, tx, lr); | |
1310 | ||
1311 | dmu_tx_commit(tx); | |
1312 | ||
1313 | return (0); | |
1314 | } | |
1315 | ||
1316 | static int | |
1317 | ztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap) | |
1318 | { | |
1319 | char *name = (void *)(lr + 1); /* name follows lr */ | |
1320 | objset_t *os = zd->zd_os; | |
1321 | dmu_object_info_t doi; | |
1322 | dmu_tx_t *tx; | |
1323 | uint64_t object, txg; | |
1324 | ||
1325 | if (byteswap) | |
1326 | byteswap_uint64_array(lr, sizeof (*lr)); | |
1327 | ||
1328 | ASSERT(lr->lr_doid == ZTEST_DIROBJ); | |
1329 | ASSERT(name[0] != '\0'); | |
1330 | ||
1331 | VERIFY3U(0, ==, | |
1332 | zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object)); | |
1333 | ASSERT(object != 0); | |
1334 | ||
1335 | ztest_object_lock(zd, object, RL_WRITER); | |
1336 | ||
1337 | VERIFY3U(0, ==, dmu_object_info(os, object, &doi)); | |
1338 | ||
1339 | tx = dmu_tx_create(os); | |
1340 | ||
1341 | dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name); | |
1342 | dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END); | |
1343 | ||
1344 | txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); | |
1345 | if (txg == 0) { | |
1346 | ztest_object_unlock(zd, object); | |
1347 | return (ENOSPC); | |
1348 | } | |
1349 | ||
1350 | if (doi.doi_type == DMU_OT_ZAP_OTHER) { | |
1351 | VERIFY3U(0, ==, zap_destroy(os, object, tx)); | |
1352 | } else { | |
1353 | VERIFY3U(0, ==, dmu_object_free(os, object, tx)); | |
1354 | } | |
1355 | ||
1356 | VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx)); | |
1357 | ||
572e2857 | 1358 | (void) ztest_log_remove(zd, tx, lr, object); |
428870ff BB |
1359 | |
1360 | dmu_tx_commit(tx); | |
1361 | ||
1362 | ztest_object_unlock(zd, object); | |
1363 | ||
1364 | return (0); | |
1365 | } | |
1366 | ||
1367 | static int | |
1368 | ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap) | |
1369 | { | |
1370 | objset_t *os = zd->zd_os; | |
1371 | void *data = lr + 1; /* data follows lr */ | |
1372 | uint64_t offset, length; | |
1373 | ztest_block_tag_t *bt = data; | |
1374 | ztest_block_tag_t *bbt; | |
1375 | uint64_t gen, txg, lrtxg, crtxg; | |
1376 | dmu_object_info_t doi; | |
1377 | dmu_tx_t *tx; | |
1378 | dmu_buf_t *db; | |
1379 | arc_buf_t *abuf = NULL; | |
1380 | rl_t *rl; | |
1381 | ||
1382 | if (byteswap) | |
1383 | byteswap_uint64_array(lr, sizeof (*lr)); | |
1384 | ||
1385 | offset = lr->lr_offset; | |
1386 | length = lr->lr_length; | |
1387 | ||
1388 | /* If it's a dmu_sync() block, write the whole block */ | |
1389 | if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) { | |
1390 | uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr); | |
1391 | if (length < blocksize) { | |
1392 | offset -= offset % blocksize; | |
1393 | length = blocksize; | |
1394 | } | |
1395 | } | |
1396 | ||
1397 | if (bt->bt_magic == BSWAP_64(BT_MAGIC)) | |
1398 | byteswap_uint64_array(bt, sizeof (*bt)); | |
1399 | ||
1400 | if (bt->bt_magic != BT_MAGIC) | |
1401 | bt = NULL; | |
1402 | ||
1403 | ztest_object_lock(zd, lr->lr_foid, RL_READER); | |
1404 | rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER); | |
1405 | ||
1406 | VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db)); | |
1407 | ||
1408 | dmu_object_info_from_db(db, &doi); | |
1409 | ||
1410 | bbt = ztest_bt_bonus(db); | |
1411 | ASSERT3U(bbt->bt_magic, ==, BT_MAGIC); | |
1412 | gen = bbt->bt_gen; | |
1413 | crtxg = bbt->bt_crtxg; | |
1414 | lrtxg = lr->lr_common.lrc_txg; | |
1415 | ||
1416 | tx = dmu_tx_create(os); | |
1417 | ||
1418 | dmu_tx_hold_write(tx, lr->lr_foid, offset, length); | |
1419 | ||
1420 | if (ztest_random(8) == 0 && length == doi.doi_data_block_size && | |
1421 | P2PHASE(offset, length) == 0) | |
1422 | abuf = dmu_request_arcbuf(db, length); | |
1423 | ||
1424 | txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); | |
1425 | if (txg == 0) { | |
1426 | if (abuf != NULL) | |
1427 | dmu_return_arcbuf(abuf); | |
1428 | dmu_buf_rele(db, FTAG); | |
1429 | ztest_range_unlock(rl); | |
1430 | ztest_object_unlock(zd, lr->lr_foid); | |
1431 | return (ENOSPC); | |
1432 | } | |
1433 | ||
1434 | if (bt != NULL) { | |
1435 | /* | |
1436 | * Usually, verify the old data before writing new data -- | |
1437 | * but not always, because we also want to verify correct | |
1438 | * behavior when the data was not recently read into cache. | |
1439 | */ | |
1440 | ASSERT(offset % doi.doi_data_block_size == 0); | |
1441 | if (ztest_random(4) != 0) { | |
1442 | int prefetch = ztest_random(2) ? | |
1443 | DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH; | |
1444 | ztest_block_tag_t rbt; | |
1445 | ||
1446 | VERIFY(dmu_read(os, lr->lr_foid, offset, | |
1447 | sizeof (rbt), &rbt, prefetch) == 0); | |
1448 | if (rbt.bt_magic == BT_MAGIC) { | |
1449 | ztest_bt_verify(&rbt, os, lr->lr_foid, | |
1450 | offset, gen, txg, crtxg); | |
1451 | } | |
1452 | } | |
1453 | ||
1454 | /* | |
1455 | * Writes can appear to be newer than the bonus buffer because | |
1456 | * the ztest_get_data() callback does a dmu_read() of the | |
1457 | * open-context data, which may be different than the data | |
1458 | * as it was when the write was generated. | |
1459 | */ | |
1460 | if (zd->zd_zilog->zl_replay) { | |
1461 | ztest_bt_verify(bt, os, lr->lr_foid, offset, | |
1462 | MAX(gen, bt->bt_gen), MAX(txg, lrtxg), | |
1463 | bt->bt_crtxg); | |
1464 | } | |
1465 | ||
1466 | /* | |
1467 | * Set the bt's gen/txg to the bonus buffer's gen/txg | |
1468 | * so that all of the usual ASSERTs will work. | |
1469 | */ | |
1470 | ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg); | |
1471 | } | |
1472 | ||
1473 | if (abuf == NULL) { | |
1474 | dmu_write(os, lr->lr_foid, offset, length, data, tx); | |
1475 | } else { | |
1476 | bcopy(data, abuf->b_data, length); | |
1477 | dmu_assign_arcbuf(db, offset, abuf, tx); | |
1478 | } | |
1479 | ||
1480 | (void) ztest_log_write(zd, tx, lr); | |
1481 | ||
1482 | dmu_buf_rele(db, FTAG); | |
1483 | ||
1484 | dmu_tx_commit(tx); | |
1485 | ||
1486 | ztest_range_unlock(rl); | |
1487 | ztest_object_unlock(zd, lr->lr_foid); | |
1488 | ||
1489 | return (0); | |
1490 | } | |
1491 | ||
1492 | static int | |
1493 | ztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap) | |
1494 | { | |
1495 | objset_t *os = zd->zd_os; | |
1496 | dmu_tx_t *tx; | |
1497 | uint64_t txg; | |
1498 | rl_t *rl; | |
1499 | ||
1500 | if (byteswap) | |
1501 | byteswap_uint64_array(lr, sizeof (*lr)); | |
1502 | ||
1503 | ztest_object_lock(zd, lr->lr_foid, RL_READER); | |
1504 | rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length, | |
1505 | RL_WRITER); | |
1506 | ||
1507 | tx = dmu_tx_create(os); | |
1508 | ||
1509 | dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length); | |
1510 | ||
1511 | txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); | |
1512 | if (txg == 0) { | |
1513 | ztest_range_unlock(rl); | |
1514 | ztest_object_unlock(zd, lr->lr_foid); | |
1515 | return (ENOSPC); | |
1516 | } | |
1517 | ||
1518 | VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset, | |
1519 | lr->lr_length, tx) == 0); | |
1520 | ||
1521 | (void) ztest_log_truncate(zd, tx, lr); | |
1522 | ||
1523 | dmu_tx_commit(tx); | |
1524 | ||
1525 | ztest_range_unlock(rl); | |
1526 | ztest_object_unlock(zd, lr->lr_foid); | |
1527 | ||
1528 | return (0); | |
1529 | } | |
1530 | ||
1531 | static int | |
1532 | ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap) | |
1533 | { | |
1534 | objset_t *os = zd->zd_os; | |
1535 | dmu_tx_t *tx; | |
1536 | dmu_buf_t *db; | |
1537 | ztest_block_tag_t *bbt; | |
1538 | uint64_t txg, lrtxg, crtxg; | |
1539 | ||
1540 | if (byteswap) | |
1541 | byteswap_uint64_array(lr, sizeof (*lr)); | |
1542 | ||
1543 | ztest_object_lock(zd, lr->lr_foid, RL_WRITER); | |
1544 | ||
1545 | VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db)); | |
1546 | ||
1547 | tx = dmu_tx_create(os); | |
1548 | dmu_tx_hold_bonus(tx, lr->lr_foid); | |
1549 | ||
1550 | txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); | |
1551 | if (txg == 0) { | |
1552 | dmu_buf_rele(db, FTAG); | |
1553 | ztest_object_unlock(zd, lr->lr_foid); | |
1554 | return (ENOSPC); | |
1555 | } | |
1556 | ||
1557 | bbt = ztest_bt_bonus(db); | |
1558 | ASSERT3U(bbt->bt_magic, ==, BT_MAGIC); | |
1559 | crtxg = bbt->bt_crtxg; | |
1560 | lrtxg = lr->lr_common.lrc_txg; | |
1561 | ||
1562 | if (zd->zd_zilog->zl_replay) { | |
1563 | ASSERT(lr->lr_size != 0); | |
1564 | ASSERT(lr->lr_mode != 0); | |
1565 | ASSERT(lrtxg != 0); | |
1566 | } else { | |
1567 | /* | |
1568 | * Randomly change the size and increment the generation. | |
1569 | */ | |
1570 | lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) * | |
1571 | sizeof (*bbt); | |
1572 | lr->lr_mode = bbt->bt_gen + 1; | |
1573 | ASSERT(lrtxg == 0); | |
1574 | } | |
1575 | ||
1576 | /* | |
1577 | * Verify that the current bonus buffer is not newer than our txg. | |
1578 | */ | |
1579 | ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, | |
1580 | MAX(txg, lrtxg), crtxg); | |
1581 | ||
1582 | dmu_buf_will_dirty(db, tx); | |
1583 | ||
1584 | ASSERT3U(lr->lr_size, >=, sizeof (*bbt)); | |
1585 | ASSERT3U(lr->lr_size, <=, db->db_size); | |
1586 | VERIFY3U(dmu_set_bonus(db, lr->lr_size, tx), ==, 0); | |
1587 | bbt = ztest_bt_bonus(db); | |
1588 | ||
1589 | ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg); | |
1590 | ||
1591 | dmu_buf_rele(db, FTAG); | |
1592 | ||
1593 | (void) ztest_log_setattr(zd, tx, lr); | |
1594 | ||
1595 | dmu_tx_commit(tx); | |
1596 | ||
1597 | ztest_object_unlock(zd, lr->lr_foid); | |
1598 | ||
1599 | return (0); | |
1600 | } | |
1601 | ||
1602 | zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = { | |
0bc8fd78 BB |
1603 | NULL, /* 0 no such transaction type */ |
1604 | (zil_replay_func_t *)ztest_replay_create, /* TX_CREATE */ | |
1605 | NULL, /* TX_MKDIR */ | |
1606 | NULL, /* TX_MKXATTR */ | |
1607 | NULL, /* TX_SYMLINK */ | |
1608 | (zil_replay_func_t *)ztest_replay_remove, /* TX_REMOVE */ | |
1609 | NULL, /* TX_RMDIR */ | |
1610 | NULL, /* TX_LINK */ | |
1611 | NULL, /* TX_RENAME */ | |
1612 | (zil_replay_func_t *)ztest_replay_write, /* TX_WRITE */ | |
1613 | (zil_replay_func_t *)ztest_replay_truncate, /* TX_TRUNCATE */ | |
1614 | (zil_replay_func_t *)ztest_replay_setattr, /* TX_SETATTR */ | |
1615 | NULL, /* TX_ACL */ | |
1616 | NULL, /* TX_CREATE_ACL */ | |
1617 | NULL, /* TX_CREATE_ATTR */ | |
1618 | NULL, /* TX_CREATE_ACL_ATTR */ | |
1619 | NULL, /* TX_MKDIR_ACL */ | |
1620 | NULL, /* TX_MKDIR_ATTR */ | |
1621 | NULL, /* TX_MKDIR_ACL_ATTR */ | |
1622 | NULL, /* TX_WRITE2 */ | |
428870ff BB |
1623 | }; |
1624 | ||
1625 | /* | |
1626 | * ZIL get_data callbacks | |
1627 | */ | |
1628 | ||
1629 | static void | |
1630 | ztest_get_done(zgd_t *zgd, int error) | |
1631 | { | |
1632 | ztest_ds_t *zd = zgd->zgd_private; | |
1633 | uint64_t object = zgd->zgd_rl->rl_object; | |
1634 | ||
1635 | if (zgd->zgd_db) | |
1636 | dmu_buf_rele(zgd->zgd_db, zgd); | |
1637 | ||
1638 | ztest_range_unlock(zgd->zgd_rl); | |
1639 | ztest_object_unlock(zd, object); | |
1640 | ||
1641 | if (error == 0 && zgd->zgd_bp) | |
1642 | zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); | |
1643 | ||
1644 | umem_free(zgd, sizeof (*zgd)); | |
1645 | } | |
1646 | ||
1647 | static int | |
1648 | ztest_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) | |
1649 | { | |
1650 | ztest_ds_t *zd = arg; | |
1651 | objset_t *os = zd->zd_os; | |
1652 | uint64_t object = lr->lr_foid; | |
1653 | uint64_t offset = lr->lr_offset; | |
1654 | uint64_t size = lr->lr_length; | |
1655 | blkptr_t *bp = &lr->lr_blkptr; | |
1656 | uint64_t txg = lr->lr_common.lrc_txg; | |
1657 | uint64_t crtxg; | |
1658 | dmu_object_info_t doi; | |
1659 | dmu_buf_t *db; | |
1660 | zgd_t *zgd; | |
1661 | int error; | |
1662 | ||
1663 | ztest_object_lock(zd, object, RL_READER); | |
1664 | error = dmu_bonus_hold(os, object, FTAG, &db); | |
1665 | if (error) { | |
1666 | ztest_object_unlock(zd, object); | |
1667 | return (error); | |
1668 | } | |
1669 | ||
1670 | crtxg = ztest_bt_bonus(db)->bt_crtxg; | |
1671 | ||
1672 | if (crtxg == 0 || crtxg > txg) { | |
1673 | dmu_buf_rele(db, FTAG); | |
1674 | ztest_object_unlock(zd, object); | |
1675 | return (ENOENT); | |
1676 | } | |
1677 | ||
1678 | dmu_object_info_from_db(db, &doi); | |
1679 | dmu_buf_rele(db, FTAG); | |
1680 | db = NULL; | |
1681 | ||
1682 | zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL); | |
1683 | zgd->zgd_zilog = zd->zd_zilog; | |
1684 | zgd->zgd_private = zd; | |
1685 | ||
1686 | if (buf != NULL) { /* immediate write */ | |
1687 | zgd->zgd_rl = ztest_range_lock(zd, object, offset, size, | |
1688 | RL_READER); | |
1689 | ||
1690 | error = dmu_read(os, object, offset, size, buf, | |
1691 | DMU_READ_NO_PREFETCH); | |
1692 | ASSERT(error == 0); | |
1693 | } else { | |
1694 | size = doi.doi_data_block_size; | |
1695 | if (ISP2(size)) { | |
1696 | offset = P2ALIGN(offset, size); | |
1697 | } else { | |
1698 | ASSERT(offset < size); | |
1699 | offset = 0; | |
1700 | } | |
1701 | ||
1702 | zgd->zgd_rl = ztest_range_lock(zd, object, offset, size, | |
1703 | RL_READER); | |
1704 | ||
1705 | error = dmu_buf_hold(os, object, offset, zgd, &db, | |
1706 | DMU_READ_NO_PREFETCH); | |
1707 | ||
1708 | if (error == 0) { | |
1709 | zgd->zgd_db = db; | |
1710 | zgd->zgd_bp = bp; | |
1711 | ||
1712 | ASSERT(db->db_offset == offset); | |
1713 | ASSERT(db->db_size == size); | |
1714 | ||
1715 | error = dmu_sync(zio, lr->lr_common.lrc_txg, | |
1716 | ztest_get_done, zgd); | |
1717 | ||
1718 | if (error == 0) | |
1719 | return (0); | |
1720 | } | |
1721 | } | |
1722 | ||
1723 | ztest_get_done(zgd, error); | |
1724 | ||
1725 | return (error); | |
1726 | } | |
1727 | ||
1728 | static void * | |
1729 | ztest_lr_alloc(size_t lrsize, char *name) | |
1730 | { | |
1731 | char *lr; | |
1732 | size_t namesize = name ? strlen(name) + 1 : 0; | |
1733 | ||
1734 | lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL); | |
1735 | ||
1736 | if (name) | |
1737 | bcopy(name, lr + lrsize, namesize); | |
1738 | ||
1739 | return (lr); | |
1740 | } | |
1741 | ||
1742 | void | |
1743 | ztest_lr_free(void *lr, size_t lrsize, char *name) | |
1744 | { | |
1745 | size_t namesize = name ? strlen(name) + 1 : 0; | |
1746 | ||
1747 | umem_free(lr, lrsize + namesize); | |
1748 | } | |
1749 | ||
1750 | /* | |
1751 | * Lookup a bunch of objects. Returns the number of objects not found. | |
1752 | */ | |
1753 | static int | |
1754 | ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count) | |
1755 | { | |
1756 | int missing = 0; | |
1757 | int error; | |
d6320ddb | 1758 | int i; |
428870ff | 1759 | |
1e33ac1e | 1760 | ASSERT(mutex_held(&zd->zd_dirobj_lock)); |
428870ff | 1761 | |
d6320ddb | 1762 | for (i = 0; i < count; i++, od++) { |
428870ff BB |
1763 | od->od_object = 0; |
1764 | error = zap_lookup(zd->zd_os, od->od_dir, od->od_name, | |
1765 | sizeof (uint64_t), 1, &od->od_object); | |
1766 | if (error) { | |
1767 | ASSERT(error == ENOENT); | |
1768 | ASSERT(od->od_object == 0); | |
1769 | missing++; | |
1770 | } else { | |
1771 | dmu_buf_t *db; | |
1772 | ztest_block_tag_t *bbt; | |
1773 | dmu_object_info_t doi; | |
1774 | ||
1775 | ASSERT(od->od_object != 0); | |
1776 | ASSERT(missing == 0); /* there should be no gaps */ | |
1777 | ||
1778 | ztest_object_lock(zd, od->od_object, RL_READER); | |
1779 | VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os, | |
1780 | od->od_object, FTAG, &db)); | |
1781 | dmu_object_info_from_db(db, &doi); | |
1782 | bbt = ztest_bt_bonus(db); | |
1783 | ASSERT3U(bbt->bt_magic, ==, BT_MAGIC); | |
1784 | od->od_type = doi.doi_type; | |
1785 | od->od_blocksize = doi.doi_data_block_size; | |
1786 | od->od_gen = bbt->bt_gen; | |
1787 | dmu_buf_rele(db, FTAG); | |
1788 | ztest_object_unlock(zd, od->od_object); | |
1789 | } | |
1790 | } | |
1791 | ||
1792 | return (missing); | |
1793 | } | |
1794 | ||
1795 | static int | |
1796 | ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count) | |
1797 | { | |
1798 | int missing = 0; | |
d6320ddb | 1799 | int i; |
428870ff | 1800 | |
1e33ac1e | 1801 | ASSERT(mutex_held(&zd->zd_dirobj_lock)); |
428870ff | 1802 | |
d6320ddb | 1803 | for (i = 0; i < count; i++, od++) { |
428870ff BB |
1804 | if (missing) { |
1805 | od->od_object = 0; | |
1806 | missing++; | |
1807 | continue; | |
1808 | } | |
1809 | ||
1810 | lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name); | |
1811 | ||
1812 | lr->lr_doid = od->od_dir; | |
1813 | lr->lr_foid = 0; /* 0 to allocate, > 0 to claim */ | |
1814 | lr->lrz_type = od->od_crtype; | |
1815 | lr->lrz_blocksize = od->od_crblocksize; | |
1816 | lr->lrz_ibshift = ztest_random_ibshift(); | |
1817 | lr->lrz_bonustype = DMU_OT_UINT64_OTHER; | |
1818 | lr->lrz_bonuslen = dmu_bonus_max(); | |
1819 | lr->lr_gen = od->od_crgen; | |
1820 | lr->lr_crtime[0] = time(NULL); | |
1821 | ||
1822 | if (ztest_replay_create(zd, lr, B_FALSE) != 0) { | |
1823 | ASSERT(missing == 0); | |
1824 | od->od_object = 0; | |
1825 | missing++; | |
1826 | } else { | |
1827 | od->od_object = lr->lr_foid; | |
1828 | od->od_type = od->od_crtype; | |
1829 | od->od_blocksize = od->od_crblocksize; | |
1830 | od->od_gen = od->od_crgen; | |
1831 | ASSERT(od->od_object != 0); | |
1832 | } | |
1833 | ||
1834 | ztest_lr_free(lr, sizeof (*lr), od->od_name); | |
1835 | } | |
1836 | ||
1837 | return (missing); | |
1838 | } | |
1839 | ||
1840 | static int | |
1841 | ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count) | |
1842 | { | |
1843 | int missing = 0; | |
1844 | int error; | |
d6320ddb | 1845 | int i; |
428870ff | 1846 | |
1e33ac1e | 1847 | ASSERT(mutex_held(&zd->zd_dirobj_lock)); |
428870ff BB |
1848 | |
1849 | od += count - 1; | |
1850 | ||
d6320ddb | 1851 | for (i = count - 1; i >= 0; i--, od--) { |
428870ff BB |
1852 | if (missing) { |
1853 | missing++; | |
1854 | continue; | |
1855 | } | |
1856 | ||
1857 | if (od->od_object == 0) | |
1858 | continue; | |
1859 | ||
1860 | lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name); | |
1861 | ||
1862 | lr->lr_doid = od->od_dir; | |
1863 | ||
1864 | if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) { | |
1865 | ASSERT3U(error, ==, ENOSPC); | |
1866 | missing++; | |
1867 | } else { | |
1868 | od->od_object = 0; | |
1869 | } | |
1870 | ztest_lr_free(lr, sizeof (*lr), od->od_name); | |
1871 | } | |
1872 | ||
1873 | return (missing); | |
1874 | } | |
1875 | ||
1876 | static int | |
1877 | ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size, | |
1878 | void *data) | |
1879 | { | |
1880 | lr_write_t *lr; | |
1881 | int error; | |
1882 | ||
1883 | lr = ztest_lr_alloc(sizeof (*lr) + size, NULL); | |
1884 | ||
1885 | lr->lr_foid = object; | |
1886 | lr->lr_offset = offset; | |
1887 | lr->lr_length = size; | |
1888 | lr->lr_blkoff = 0; | |
1889 | BP_ZERO(&lr->lr_blkptr); | |
1890 | ||
1891 | bcopy(data, lr + 1, size); | |
1892 | ||
1893 | error = ztest_replay_write(zd, lr, B_FALSE); | |
1894 | ||
1895 | ztest_lr_free(lr, sizeof (*lr) + size, NULL); | |
1896 | ||
1897 | return (error); | |
1898 | } | |
1899 | ||
1900 | static int | |
1901 | ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size) | |
1902 | { | |
1903 | lr_truncate_t *lr; | |
1904 | int error; | |
1905 | ||
1906 | lr = ztest_lr_alloc(sizeof (*lr), NULL); | |
1907 | ||
1908 | lr->lr_foid = object; | |
1909 | lr->lr_offset = offset; | |
1910 | lr->lr_length = size; | |
1911 | ||
1912 | error = ztest_replay_truncate(zd, lr, B_FALSE); | |
1913 | ||
1914 | ztest_lr_free(lr, sizeof (*lr), NULL); | |
1915 | ||
1916 | return (error); | |
1917 | } | |
1918 | ||
1919 | static int | |
1920 | ztest_setattr(ztest_ds_t *zd, uint64_t object) | |
1921 | { | |
1922 | lr_setattr_t *lr; | |
1923 | int error; | |
1924 | ||
1925 | lr = ztest_lr_alloc(sizeof (*lr), NULL); | |
1926 | ||
1927 | lr->lr_foid = object; | |
1928 | lr->lr_size = 0; | |
1929 | lr->lr_mode = 0; | |
1930 | ||
1931 | error = ztest_replay_setattr(zd, lr, B_FALSE); | |
1932 | ||
1933 | ztest_lr_free(lr, sizeof (*lr), NULL); | |
1934 | ||
1935 | return (error); | |
1936 | } | |
1937 | ||
1938 | static void | |
1939 | ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size) | |
1940 | { | |
1941 | objset_t *os = zd->zd_os; | |
1942 | dmu_tx_t *tx; | |
1943 | uint64_t txg; | |
1944 | rl_t *rl; | |
1945 | ||
1946 | txg_wait_synced(dmu_objset_pool(os), 0); | |
1947 | ||
1948 | ztest_object_lock(zd, object, RL_READER); | |
1949 | rl = ztest_range_lock(zd, object, offset, size, RL_WRITER); | |
1950 | ||
1951 | tx = dmu_tx_create(os); | |
1952 | ||
1953 | dmu_tx_hold_write(tx, object, offset, size); | |
1954 | ||
1955 | txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); | |
1956 | ||
1957 | if (txg != 0) { | |
1958 | dmu_prealloc(os, object, offset, size, tx); | |
1959 | dmu_tx_commit(tx); | |
1960 | txg_wait_synced(dmu_objset_pool(os), txg); | |
1961 | } else { | |
1962 | (void) dmu_free_long_range(os, object, offset, size); | |
1963 | } | |
1964 | ||
1965 | ztest_range_unlock(rl); | |
1966 | ztest_object_unlock(zd, object); | |
1967 | } | |
1968 | ||
1969 | static void | |
1970 | ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset) | |
1971 | { | |
1972 | ztest_block_tag_t wbt; | |
1973 | dmu_object_info_t doi; | |
1974 | enum ztest_io_type io_type; | |
1975 | uint64_t blocksize; | |
1976 | void *data; | |
1977 | ||
1978 | VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0); | |
1979 | blocksize = doi.doi_data_block_size; | |
1980 | data = umem_alloc(blocksize, UMEM_NOFAIL); | |
1981 | ||
1982 | /* | |
1983 | * Pick an i/o type at random, biased toward writing block tags. | |
1984 | */ | |
1985 | io_type = ztest_random(ZTEST_IO_TYPES); | |
1986 | if (ztest_random(2) == 0) | |
1987 | io_type = ZTEST_IO_WRITE_TAG; | |
1988 | ||
1989 | switch (io_type) { | |
1990 | ||
1991 | case ZTEST_IO_WRITE_TAG: | |
1992 | ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0); | |
1993 | (void) ztest_write(zd, object, offset, sizeof (wbt), &wbt); | |
1994 | break; | |
1995 | ||
1996 | case ZTEST_IO_WRITE_PATTERN: | |
1997 | (void) memset(data, 'a' + (object + offset) % 5, blocksize); | |
1998 | if (ztest_random(2) == 0) { | |
1999 | /* | |
2000 | * Induce fletcher2 collisions to ensure that | |
2001 | * zio_ddt_collision() detects and resolves them | |
2002 | * when using fletcher2-verify for deduplication. | |
2003 | */ | |
2004 | ((uint64_t *)data)[0] ^= 1ULL << 63; | |
2005 | ((uint64_t *)data)[4] ^= 1ULL << 63; | |
2006 | } | |
2007 | (void) ztest_write(zd, object, offset, blocksize, data); | |
2008 | break; | |
2009 | ||
2010 | case ZTEST_IO_WRITE_ZEROES: | |
2011 | bzero(data, blocksize); | |
2012 | (void) ztest_write(zd, object, offset, blocksize, data); | |
2013 | break; | |
2014 | ||
2015 | case ZTEST_IO_TRUNCATE: | |
2016 | (void) ztest_truncate(zd, object, offset, blocksize); | |
2017 | break; | |
2018 | ||
2019 | case ZTEST_IO_SETATTR: | |
2020 | (void) ztest_setattr(zd, object); | |
2021 | break; | |
e75c13c3 BB |
2022 | default: |
2023 | break; | |
428870ff BB |
2024 | } |
2025 | ||
2026 | umem_free(data, blocksize); | |
2027 | } | |
2028 | ||
2029 | /* | |
2030 | * Initialize an object description template. | |
2031 | */ | |
2032 | static void | |
2033 | ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index, | |
2034 | dmu_object_type_t type, uint64_t blocksize, uint64_t gen) | |
2035 | { | |
2036 | od->od_dir = ZTEST_DIROBJ; | |
2037 | od->od_object = 0; | |
2038 | ||
2039 | od->od_crtype = type; | |
2040 | od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize(); | |
2041 | od->od_crgen = gen; | |
2042 | ||
2043 | od->od_type = DMU_OT_NONE; | |
2044 | od->od_blocksize = 0; | |
2045 | od->od_gen = 0; | |
2046 | ||
2047 | (void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]", | |
b8864a23 | 2048 | tag, (longlong_t)id, (u_longlong_t)index); |
428870ff BB |
2049 | } |
2050 | ||
2051 | /* | |
2052 | * Lookup or create the objects for a test using the od template. | |
2053 | * If the objects do not all exist, or if 'remove' is specified, | |
2054 | * remove any existing objects and create new ones. Otherwise, | |
2055 | * use the existing objects. | |
2056 | */ | |
2057 | static int | |
2058 | ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove) | |
2059 | { | |
2060 | int count = size / sizeof (*od); | |
2061 | int rv = 0; | |
2062 | ||
1e33ac1e | 2063 | mutex_enter(&zd->zd_dirobj_lock); |
428870ff BB |
2064 | if ((ztest_lookup(zd, od, count) != 0 || remove) && |
2065 | (ztest_remove(zd, od, count) != 0 || | |
2066 | ztest_create(zd, od, count) != 0)) | |
2067 | rv = -1; | |
2068 | zd->zd_od = od; | |
1e33ac1e | 2069 | mutex_exit(&zd->zd_dirobj_lock); |
428870ff BB |
2070 | |
2071 | return (rv); | |
2072 | } | |
2073 | ||
2074 | /* ARGSUSED */ | |
2075 | void | |
2076 | ztest_zil_commit(ztest_ds_t *zd, uint64_t id) | |
2077 | { | |
2078 | zilog_t *zilog = zd->zd_zilog; | |
2079 | ||
572e2857 | 2080 | zil_commit(zilog, ztest_random(ZTEST_OBJECTS)); |
428870ff BB |
2081 | |
2082 | /* | |
2083 | * Remember the committed values in zd, which is in parent/child | |
2084 | * shared memory. If we die, the next iteration of ztest_run() | |
2085 | * will verify that the log really does contain this record. | |
2086 | */ | |
2087 | mutex_enter(&zilog->zl_lock); | |
2088 | ASSERT(zd->zd_seq <= zilog->zl_commit_lr_seq); | |
2089 | zd->zd_seq = zilog->zl_commit_lr_seq; | |
2090 | mutex_exit(&zilog->zl_lock); | |
2091 | } | |
2092 | ||
2093 | /* | |
2094 | * Verify that we can't destroy an active pool, create an existing pool, | |
2095 | * or create a pool with a bad vdev spec. | |
2096 | */ | |
2097 | /* ARGSUSED */ | |
2098 | void | |
2099 | ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id) | |
2100 | { | |
2101 | ztest_shared_t *zs = ztest_shared; | |
2102 | spa_t *spa; | |
2103 | nvlist_t *nvroot; | |
2104 | ||
2105 | /* | |
2106 | * Attempt to create using a bad file. | |
2107 | */ | |
2108 | nvroot = make_vdev_root("/dev/bogus", NULL, 0, 0, 0, 0, 0, 1); | |
2109 | VERIFY3U(ENOENT, ==, | |
2110 | spa_create("ztest_bad_file", nvroot, NULL, NULL, NULL)); | |
2111 | nvlist_free(nvroot); | |
2112 | ||
2113 | /* | |
2114 | * Attempt to create using a bad mirror. | |
2115 | */ | |
2116 | nvroot = make_vdev_root("/dev/bogus", NULL, 0, 0, 0, 0, 2, 1); | |
2117 | VERIFY3U(ENOENT, ==, | |
2118 | spa_create("ztest_bad_mirror", nvroot, NULL, NULL, NULL)); | |
2119 | nvlist_free(nvroot); | |
2120 | ||
2121 | /* | |
2122 | * Attempt to create an existing pool. It shouldn't matter | |
2123 | * what's in the nvroot; we should fail with EEXIST. | |
2124 | */ | |
1e33ac1e | 2125 | (void) rw_enter(&zs->zs_name_lock, RW_READER); |
428870ff BB |
2126 | nvroot = make_vdev_root("/dev/bogus", NULL, 0, 0, 0, 0, 0, 1); |
2127 | VERIFY3U(EEXIST, ==, spa_create(zs->zs_pool, nvroot, NULL, NULL, NULL)); | |
2128 | nvlist_free(nvroot); | |
2129 | VERIFY3U(0, ==, spa_open(zs->zs_pool, &spa, FTAG)); | |
2130 | VERIFY3U(EBUSY, ==, spa_destroy(zs->zs_pool)); | |
2131 | spa_close(spa, FTAG); | |
2132 | ||
1e33ac1e | 2133 | (void) rw_exit(&zs->zs_name_lock); |
428870ff BB |
2134 | } |
2135 | ||
2136 | static vdev_t * | |
2137 | vdev_lookup_by_path(vdev_t *vd, const char *path) | |
2138 | { | |
2139 | vdev_t *mvd; | |
d6320ddb | 2140 | int c; |
428870ff BB |
2141 | |
2142 | if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0) | |
2143 | return (vd); | |
2144 | ||
d6320ddb | 2145 | for (c = 0; c < vd->vdev_children; c++) |
428870ff BB |
2146 | if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) != |
2147 | NULL) | |
2148 | return (mvd); | |
2149 | ||
2150 | return (NULL); | |
2151 | } | |
2152 | ||
2153 | /* | |
2154 | * Find the first available hole which can be used as a top-level. | |
2155 | */ | |
2156 | int | |
2157 | find_vdev_hole(spa_t *spa) | |
2158 | { | |
2159 | vdev_t *rvd = spa->spa_root_vdev; | |
2160 | int c; | |
2161 | ||
2162 | ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV); | |
2163 | ||
2164 | for (c = 0; c < rvd->vdev_children; c++) { | |
2165 | vdev_t *cvd = rvd->vdev_child[c]; | |
2166 | ||
2167 | if (cvd->vdev_ishole) | |
2168 | break; | |
2169 | } | |
2170 | return (c); | |
2171 | } | |
2172 | ||
2173 | /* | |
2174 | * Verify that vdev_add() works as expected. | |
2175 | */ | |
2176 | /* ARGSUSED */ | |
2177 | void | |
2178 | ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id) | |
2179 | { | |
2180 | ztest_shared_t *zs = ztest_shared; | |
2181 | spa_t *spa = zs->zs_spa; | |
2182 | uint64_t leaves; | |
2183 | uint64_t guid; | |
2184 | nvlist_t *nvroot; | |
2185 | int error; | |
2186 | ||
1e33ac1e | 2187 | mutex_enter(&zs->zs_vdev_lock); |
428870ff BB |
2188 | leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * zopt_raidz; |
2189 | ||
2190 | spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); | |
2191 | ||
2192 | ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves; | |
2193 | ||
2194 | /* | |
2195 | * If we have slogs then remove them 1/4 of the time. | |
2196 | */ | |
2197 | if (spa_has_slogs(spa) && ztest_random(4) == 0) { | |
2198 | /* | |
2199 | * Grab the guid from the head of the log class rotor. | |
2200 | */ | |
2201 | guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid; | |
2202 | ||
2203 | spa_config_exit(spa, SCL_VDEV, FTAG); | |
2204 | ||
2205 | /* | |
2206 | * We have to grab the zs_name_lock as writer to | |
2207 | * prevent a race between removing a slog (dmu_objset_find) | |
2208 | * and destroying a dataset. Removing the slog will | |
2209 | * grab a reference on the dataset which may cause | |
2210 | * dmu_objset_destroy() to fail with EBUSY thus | |
2211 | * leaving the dataset in an inconsistent state. | |
2212 | */ | |
1e33ac1e | 2213 | rw_enter(&ztest_shared->zs_name_lock, RW_WRITER); |
428870ff | 2214 | error = spa_vdev_remove(spa, guid, B_FALSE); |
1e33ac1e | 2215 | rw_exit(&ztest_shared->zs_name_lock); |
428870ff BB |
2216 | |
2217 | if (error && error != EEXIST) | |
2218 | fatal(0, "spa_vdev_remove() = %d", error); | |
2219 | } else { | |
2220 | spa_config_exit(spa, SCL_VDEV, FTAG); | |
2221 | ||
2222 | /* | |
2223 | * Make 1/4 of the devices be log devices. | |
2224 | */ | |
2225 | nvroot = make_vdev_root(NULL, NULL, zopt_vdev_size, 0, | |
2226 | ztest_random(4) == 0, zopt_raidz, zs->zs_mirrors, 1); | |
2227 | ||
2228 | error = spa_vdev_add(spa, nvroot); | |
2229 | nvlist_free(nvroot); | |
2230 | ||
2231 | if (error == ENOSPC) | |
2232 | ztest_record_enospc("spa_vdev_add"); | |
2233 | else if (error != 0) | |
2234 | fatal(0, "spa_vdev_add() = %d", error); | |
2235 | } | |
2236 | ||
1e33ac1e | 2237 | mutex_exit(&ztest_shared->zs_vdev_lock); |
428870ff BB |
2238 | } |
2239 | ||
2240 | /* | |
2241 | * Verify that adding/removing aux devices (l2arc, hot spare) works as expected. | |
2242 | */ | |
2243 | /* ARGSUSED */ | |
2244 | void | |
2245 | ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id) | |
2246 | { | |
2247 | ztest_shared_t *zs = ztest_shared; | |
2248 | spa_t *spa = zs->zs_spa; | |
2249 | vdev_t *rvd = spa->spa_root_vdev; | |
2250 | spa_aux_vdev_t *sav; | |
2251 | char *aux; | |
2252 | uint64_t guid = 0; | |
2253 | int error; | |
2254 | ||
2255 | if (ztest_random(2) == 0) { | |
2256 | sav = &spa->spa_spares; | |
2257 | aux = ZPOOL_CONFIG_SPARES; | |
2258 | } else { | |
2259 | sav = &spa->spa_l2cache; | |
2260 | aux = ZPOOL_CONFIG_L2CACHE; | |
2261 | } | |
2262 | ||
1e33ac1e | 2263 | mutex_enter(&zs->zs_vdev_lock); |
428870ff BB |
2264 | |
2265 | spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); | |
2266 | ||
2267 | if (sav->sav_count != 0 && ztest_random(4) == 0) { | |
b128c09f BB |
2268 | /* |
2269 | * Pick a random device to remove. | |
2270 | */ | |
2271 | guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid; | |
2272 | } else { | |
2273 | /* | |
2274 | * Find an unused device we can add. | |
2275 | */ | |
428870ff | 2276 | zs->zs_vdev_aux = 0; |
b128c09f BB |
2277 | for (;;) { |
2278 | char path[MAXPATHLEN]; | |
2279 | int c; | |
2280 | (void) sprintf(path, ztest_aux_template, zopt_dir, | |
428870ff | 2281 | zopt_pool, aux, zs->zs_vdev_aux); |
b128c09f BB |
2282 | for (c = 0; c < sav->sav_count; c++) |
2283 | if (strcmp(sav->sav_vdevs[c]->vdev_path, | |
2284 | path) == 0) | |
2285 | break; | |
2286 | if (c == sav->sav_count && | |
2287 | vdev_lookup_by_path(rvd, path) == NULL) | |
2288 | break; | |
428870ff | 2289 | zs->zs_vdev_aux++; |
34dc7c2f BB |
2290 | } |
2291 | } | |
2292 | ||
b128c09f | 2293 | spa_config_exit(spa, SCL_VDEV, FTAG); |
34dc7c2f | 2294 | |
b128c09f BB |
2295 | if (guid == 0) { |
2296 | /* | |
2297 | * Add a new device. | |
2298 | */ | |
2299 | nvlist_t *nvroot = make_vdev_root(NULL, aux, | |
2300 | (zopt_vdev_size * 5) / 4, 0, 0, 0, 0, 1); | |
2301 | error = spa_vdev_add(spa, nvroot); | |
2302 | if (error != 0) | |
2303 | fatal(0, "spa_vdev_add(%p) = %d", nvroot, error); | |
2304 | nvlist_free(nvroot); | |
2305 | } else { | |
2306 | /* | |
2307 | * Remove an existing device. Sometimes, dirty its | |
2308 | * vdev state first to make sure we handle removal | |
2309 | * of devices that have pending state changes. | |
2310 | */ | |
2311 | if (ztest_random(2) == 0) | |
9babb374 | 2312 | (void) vdev_online(spa, guid, 0, NULL); |
b128c09f BB |
2313 | |
2314 | error = spa_vdev_remove(spa, guid, B_FALSE); | |
2315 | if (error != 0 && error != EBUSY) | |
2316 | fatal(0, "spa_vdev_remove(%llu) = %d", guid, error); | |
2317 | } | |
2318 | ||
1e33ac1e | 2319 | mutex_exit(&zs->zs_vdev_lock); |
428870ff BB |
2320 | } |
2321 | ||
2322 | /* | |
2323 | * split a pool if it has mirror tlvdevs | |
2324 | */ | |
2325 | /* ARGSUSED */ | |
2326 | void | |
2327 | ztest_split_pool(ztest_ds_t *zd, uint64_t id) | |
2328 | { | |
2329 | ztest_shared_t *zs = ztest_shared; | |
2330 | spa_t *spa = zs->zs_spa; | |
2331 | vdev_t *rvd = spa->spa_root_vdev; | |
2332 | nvlist_t *tree, **child, *config, *split, **schild; | |
2333 | uint_t c, children, schildren = 0, lastlogid = 0; | |
2334 | int error = 0; | |
2335 | ||
1e33ac1e | 2336 | mutex_enter(&zs->zs_vdev_lock); |
428870ff BB |
2337 | |
2338 | /* ensure we have a useable config; mirrors of raidz aren't supported */ | |
2339 | if (zs->zs_mirrors < 3 || zopt_raidz > 1) { | |
1e33ac1e | 2340 | mutex_exit(&zs->zs_vdev_lock); |
428870ff BB |
2341 | return; |
2342 | } | |
2343 | ||
2344 | /* clean up the old pool, if any */ | |
2345 | (void) spa_destroy("splitp"); | |
2346 | ||
2347 | spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); | |
2348 | ||
2349 | /* generate a config from the existing config */ | |
2350 | mutex_enter(&spa->spa_props_lock); | |
2351 | VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE, | |
2352 | &tree) == 0); | |
2353 | mutex_exit(&spa->spa_props_lock); | |
2354 | ||
2355 | VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child, | |
2356 | &children) == 0); | |
2357 | ||
2358 | schild = malloc(rvd->vdev_children * sizeof (nvlist_t *)); | |
2359 | for (c = 0; c < children; c++) { | |
2360 | vdev_t *tvd = rvd->vdev_child[c]; | |
2361 | nvlist_t **mchild; | |
2362 | uint_t mchildren; | |
2363 | ||
2364 | if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) { | |
2365 | VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME, | |
2366 | 0) == 0); | |
2367 | VERIFY(nvlist_add_string(schild[schildren], | |
2368 | ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0); | |
2369 | VERIFY(nvlist_add_uint64(schild[schildren], | |
2370 | ZPOOL_CONFIG_IS_HOLE, 1) == 0); | |
2371 | if (lastlogid == 0) | |
2372 | lastlogid = schildren; | |
2373 | ++schildren; | |
2374 | continue; | |
2375 | } | |
2376 | lastlogid = 0; | |
2377 | VERIFY(nvlist_lookup_nvlist_array(child[c], | |
2378 | ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0); | |
2379 | VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0); | |
2380 | } | |
2381 | ||
2382 | /* OK, create a config that can be used to split */ | |
2383 | VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0); | |
2384 | VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE, | |
2385 | VDEV_TYPE_ROOT) == 0); | |
2386 | VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild, | |
2387 | lastlogid != 0 ? lastlogid : schildren) == 0); | |
2388 | ||
2389 | VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0); | |
2390 | VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0); | |
2391 | ||
2392 | for (c = 0; c < schildren; c++) | |
2393 | nvlist_free(schild[c]); | |
2394 | free(schild); | |
2395 | nvlist_free(split); | |
2396 | ||
2397 | spa_config_exit(spa, SCL_VDEV, FTAG); | |
2398 | ||
1e33ac1e | 2399 | (void) rw_enter(&zs->zs_name_lock, RW_WRITER); |
428870ff | 2400 | error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE); |
1e33ac1e | 2401 | (void) rw_exit(&zs->zs_name_lock); |
428870ff BB |
2402 | |
2403 | nvlist_free(config); | |
2404 | ||
2405 | if (error == 0) { | |
2406 | (void) printf("successful split - results:\n"); | |
2407 | mutex_enter(&spa_namespace_lock); | |
2408 | show_pool_stats(spa); | |
2409 | show_pool_stats(spa_lookup("splitp")); | |
2410 | mutex_exit(&spa_namespace_lock); | |
2411 | ++zs->zs_splits; | |
2412 | --zs->zs_mirrors; | |
2413 | } | |
1e33ac1e | 2414 | mutex_exit(&zs->zs_vdev_lock); |
428870ff | 2415 | |
34dc7c2f BB |
2416 | } |
2417 | ||
2418 | /* | |
2419 | * Verify that we can attach and detach devices. | |
2420 | */ | |
428870ff | 2421 | /* ARGSUSED */ |
34dc7c2f | 2422 | void |
428870ff | 2423 | ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id) |
34dc7c2f | 2424 | { |
428870ff BB |
2425 | ztest_shared_t *zs = ztest_shared; |
2426 | spa_t *spa = zs->zs_spa; | |
b128c09f | 2427 | spa_aux_vdev_t *sav = &spa->spa_spares; |
34dc7c2f BB |
2428 | vdev_t *rvd = spa->spa_root_vdev; |
2429 | vdev_t *oldvd, *newvd, *pvd; | |
b128c09f | 2430 | nvlist_t *root; |
428870ff | 2431 | uint64_t leaves; |
34dc7c2f BB |
2432 | uint64_t leaf, top; |
2433 | uint64_t ashift = ztest_get_ashift(); | |
fb5f0bc8 | 2434 | uint64_t oldguid, pguid; |
34dc7c2f BB |
2435 | size_t oldsize, newsize; |
2436 | char oldpath[MAXPATHLEN], newpath[MAXPATHLEN]; | |
2437 | int replacing; | |
b128c09f BB |
2438 | int oldvd_has_siblings = B_FALSE; |
2439 | int newvd_is_spare = B_FALSE; | |
2440 | int oldvd_is_log; | |
34dc7c2f | 2441 | int error, expected_error; |
34dc7c2f | 2442 | |
1e33ac1e | 2443 | mutex_enter(&zs->zs_vdev_lock); |
428870ff | 2444 | leaves = MAX(zs->zs_mirrors, 1) * zopt_raidz; |
34dc7c2f | 2445 | |
b128c09f | 2446 | spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); |
34dc7c2f BB |
2447 | |
2448 | /* | |
2449 | * Decide whether to do an attach or a replace. | |
2450 | */ | |
2451 | replacing = ztest_random(2); | |
2452 | ||
2453 | /* | |
2454 | * Pick a random top-level vdev. | |
2455 | */ | |
428870ff | 2456 | top = ztest_random_vdev_top(spa, B_TRUE); |
34dc7c2f BB |
2457 | |
2458 | /* | |
2459 | * Pick a random leaf within it. | |
2460 | */ | |
2461 | leaf = ztest_random(leaves); | |
2462 | ||
2463 | /* | |
b128c09f | 2464 | * Locate this vdev. |
34dc7c2f | 2465 | */ |
b128c09f | 2466 | oldvd = rvd->vdev_child[top]; |
428870ff | 2467 | if (zs->zs_mirrors >= 1) { |
fb5f0bc8 | 2468 | ASSERT(oldvd->vdev_ops == &vdev_mirror_ops); |
428870ff | 2469 | ASSERT(oldvd->vdev_children >= zs->zs_mirrors); |
b128c09f | 2470 | oldvd = oldvd->vdev_child[leaf / zopt_raidz]; |
fb5f0bc8 BB |
2471 | } |
2472 | if (zopt_raidz > 1) { | |
2473 | ASSERT(oldvd->vdev_ops == &vdev_raidz_ops); | |
2474 | ASSERT(oldvd->vdev_children == zopt_raidz); | |
b128c09f | 2475 | oldvd = oldvd->vdev_child[leaf % zopt_raidz]; |
fb5f0bc8 | 2476 | } |
34dc7c2f BB |
2477 | |
2478 | /* | |
b128c09f BB |
2479 | * If we're already doing an attach or replace, oldvd may be a |
2480 | * mirror vdev -- in which case, pick a random child. | |
34dc7c2f | 2481 | */ |
b128c09f BB |
2482 | while (oldvd->vdev_children != 0) { |
2483 | oldvd_has_siblings = B_TRUE; | |
fb5f0bc8 BB |
2484 | ASSERT(oldvd->vdev_children >= 2); |
2485 | oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)]; | |
b128c09f BB |
2486 | } |
2487 | ||
2488 | oldguid = oldvd->vdev_guid; | |
9babb374 | 2489 | oldsize = vdev_get_min_asize(oldvd); |
b128c09f BB |
2490 | oldvd_is_log = oldvd->vdev_top->vdev_islog; |
2491 | (void) strcpy(oldpath, oldvd->vdev_path); | |
2492 | pvd = oldvd->vdev_parent; | |
fb5f0bc8 | 2493 | pguid = pvd->vdev_guid; |
34dc7c2f BB |
2494 | |
2495 | /* | |
b128c09f | 2496 | * If oldvd has siblings, then half of the time, detach it. |
34dc7c2f | 2497 | */ |
b128c09f BB |
2498 | if (oldvd_has_siblings && ztest_random(2) == 0) { |
2499 | spa_config_exit(spa, SCL_VDEV, FTAG); | |
fb5f0bc8 BB |
2500 | error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE); |
2501 | if (error != 0 && error != ENODEV && error != EBUSY && | |
2502 | error != ENOTSUP) | |
2503 | fatal(0, "detach (%s) returned %d", oldpath, error); | |
1e33ac1e | 2504 | mutex_exit(&zs->zs_vdev_lock); |
b128c09f BB |
2505 | return; |
2506 | } | |
34dc7c2f BB |
2507 | |
2508 | /* | |
b128c09f BB |
2509 | * For the new vdev, choose with equal probability between the two |
2510 | * standard paths (ending in either 'a' or 'b') or a random hot spare. | |
34dc7c2f | 2511 | */ |
b128c09f BB |
2512 | if (sav->sav_count != 0 && ztest_random(3) == 0) { |
2513 | newvd = sav->sav_vdevs[ztest_random(sav->sav_count)]; | |
2514 | newvd_is_spare = B_TRUE; | |
2515 | (void) strcpy(newpath, newvd->vdev_path); | |
2516 | } else { | |
2517 | (void) snprintf(newpath, sizeof (newpath), ztest_dev_template, | |
2518 | zopt_dir, zopt_pool, top * leaves + leaf); | |
2519 | if (ztest_random(2) == 0) | |
2520 | newpath[strlen(newpath) - 1] = 'b'; | |
2521 | newvd = vdev_lookup_by_path(rvd, newpath); | |
2522 | } | |
2523 | ||
2524 | if (newvd) { | |
9babb374 | 2525 | newsize = vdev_get_min_asize(newvd); |
b128c09f BB |
2526 | } else { |
2527 | /* | |
2528 | * Make newsize a little bigger or smaller than oldsize. | |
2529 | * If it's smaller, the attach should fail. | |
2530 | * If it's larger, and we're doing a replace, | |
2531 | * we should get dynamic LUN growth when we're done. | |
2532 | */ | |
2533 | newsize = 10 * oldsize / (9 + ztest_random(3)); | |
2534 | } | |
34dc7c2f BB |
2535 | |
2536 | /* | |
2537 | * If pvd is not a mirror or root, the attach should fail with ENOTSUP, | |
2538 | * unless it's a replace; in that case any non-replacing parent is OK. | |
2539 | * | |
2540 | * If newvd is already part of the pool, it should fail with EBUSY. | |
2541 | * | |
2542 | * If newvd is too small, it should fail with EOVERFLOW. | |
2543 | */ | |
b128c09f BB |
2544 | if (pvd->vdev_ops != &vdev_mirror_ops && |
2545 | pvd->vdev_ops != &vdev_root_ops && (!replacing || | |
2546 | pvd->vdev_ops == &vdev_replacing_ops || | |
2547 | pvd->vdev_ops == &vdev_spare_ops)) | |
34dc7c2f | 2548 | expected_error = ENOTSUP; |
b128c09f BB |
2549 | else if (newvd_is_spare && (!replacing || oldvd_is_log)) |
2550 | expected_error = ENOTSUP; | |
2551 | else if (newvd == oldvd) | |
2552 | expected_error = replacing ? 0 : EBUSY; | |
2553 | else if (vdev_lookup_by_path(rvd, newpath) != NULL) | |
2554 | expected_error = EBUSY; | |
34dc7c2f BB |
2555 | else if (newsize < oldsize) |
2556 | expected_error = EOVERFLOW; | |
2557 | else if (ashift > oldvd->vdev_top->vdev_ashift) | |
2558 | expected_error = EDOM; | |
2559 | else | |
2560 | expected_error = 0; | |
2561 | ||
b128c09f | 2562 | spa_config_exit(spa, SCL_VDEV, FTAG); |
34dc7c2f BB |
2563 | |
2564 | /* | |
2565 | * Build the nvlist describing newpath. | |
2566 | */ | |
b128c09f BB |
2567 | root = make_vdev_root(newpath, NULL, newvd == NULL ? newsize : 0, |
2568 | ashift, 0, 0, 0, 1); | |
34dc7c2f | 2569 | |
b128c09f | 2570 | error = spa_vdev_attach(spa, oldguid, root, replacing); |
34dc7c2f | 2571 | |
34dc7c2f BB |
2572 | nvlist_free(root); |
2573 | ||
2574 | /* | |
2575 | * If our parent was the replacing vdev, but the replace completed, | |
2576 | * then instead of failing with ENOTSUP we may either succeed, | |
2577 | * fail with ENODEV, or fail with EOVERFLOW. | |
2578 | */ | |
2579 | if (expected_error == ENOTSUP && | |
2580 | (error == 0 || error == ENODEV || error == EOVERFLOW)) | |
2581 | expected_error = error; | |
2582 | ||
2583 | /* | |
2584 | * If someone grew the LUN, the replacement may be too small. | |
2585 | */ | |
b128c09f | 2586 | if (error == EOVERFLOW || error == EBUSY) |
34dc7c2f BB |
2587 | expected_error = error; |
2588 | ||
b128c09f BB |
2589 | /* XXX workaround 6690467 */ |
2590 | if (error != expected_error && expected_error != EBUSY) { | |
2591 | fatal(0, "attach (%s %llu, %s %llu, %d) " | |
2592 | "returned %d, expected %d", | |
2593 | oldpath, (longlong_t)oldsize, newpath, | |
2594 | (longlong_t)newsize, replacing, error, expected_error); | |
34dc7c2f BB |
2595 | } |
2596 | ||
1e33ac1e | 2597 | mutex_exit(&zs->zs_vdev_lock); |
34dc7c2f BB |
2598 | } |
2599 | ||
9babb374 BB |
2600 | /* |
2601 | * Callback function which expands the physical size of the vdev. | |
2602 | */ | |
2603 | vdev_t * | |
2604 | grow_vdev(vdev_t *vd, void *arg) | |
2605 | { | |
1fde1e37 | 2606 | ASSERTV(spa_t *spa = vd->vdev_spa); |
9babb374 BB |
2607 | size_t *newsize = arg; |
2608 | size_t fsize; | |
2609 | int fd; | |
2610 | ||
2611 | ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE); | |
2612 | ASSERT(vd->vdev_ops->vdev_op_leaf); | |
2613 | ||
2614 | if ((fd = open(vd->vdev_path, O_RDWR)) == -1) | |
2615 | return (vd); | |
2616 | ||
2617 | fsize = lseek(fd, 0, SEEK_END); | |
0e5b68e0 | 2618 | VERIFY(ftruncate(fd, *newsize) == 0); |
9babb374 BB |
2619 | |
2620 | if (zopt_verbose >= 6) { | |
2621 | (void) printf("%s grew from %lu to %lu bytes\n", | |
2622 | vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize); | |
2623 | } | |
2624 | (void) close(fd); | |
2625 | return (NULL); | |
2626 | } | |
2627 | ||
2628 | /* | |
2629 | * Callback function which expands a given vdev by calling vdev_online(). | |
2630 | */ | |
2631 | /* ARGSUSED */ | |
2632 | vdev_t * | |
2633 | online_vdev(vdev_t *vd, void *arg) | |
2634 | { | |
2635 | spa_t *spa = vd->vdev_spa; | |
2636 | vdev_t *tvd = vd->vdev_top; | |
9babb374 | 2637 | uint64_t guid = vd->vdev_guid; |
428870ff BB |
2638 | uint64_t generation = spa->spa_config_generation + 1; |
2639 | vdev_state_t newstate = VDEV_STATE_UNKNOWN; | |
2640 | int error; | |
9babb374 BB |
2641 | |
2642 | ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE); | |
2643 | ASSERT(vd->vdev_ops->vdev_op_leaf); | |
2644 | ||
2645 | /* Calling vdev_online will initialize the new metaslabs */ | |
2646 | spa_config_exit(spa, SCL_STATE, spa); | |
428870ff | 2647 | error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate); |
9babb374 BB |
2648 | spa_config_enter(spa, SCL_STATE, spa, RW_READER); |
2649 | ||
428870ff BB |
2650 | /* |
2651 | * If vdev_online returned an error or the underlying vdev_open | |
2652 | * failed then we abort the expand. The only way to know that | |
2653 | * vdev_open fails is by checking the returned newstate. | |
2654 | */ | |
2655 | if (error || newstate != VDEV_STATE_HEALTHY) { | |
2656 | if (zopt_verbose >= 5) { | |
2657 | (void) printf("Unable to expand vdev, state %llu, " | |
2658 | "error %d\n", (u_longlong_t)newstate, error); | |
2659 | } | |
2660 | return (vd); | |
2661 | } | |
2662 | ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY); | |
2663 | ||
9babb374 BB |
2664 | /* |
2665 | * Since we dropped the lock we need to ensure that we're | |
2666 | * still talking to the original vdev. It's possible this | |
2667 | * vdev may have been detached/replaced while we were | |
2668 | * trying to online it. | |
2669 | */ | |
428870ff BB |
2670 | if (generation != spa->spa_config_generation) { |
2671 | if (zopt_verbose >= 5) { | |
2672 | (void) printf("vdev configuration has changed, " | |
2673 | "guid %llu, state %llu, expected gen %llu, " | |
2674 | "got gen %llu\n", | |
2675 | (u_longlong_t)guid, | |
2676 | (u_longlong_t)tvd->vdev_state, | |
2677 | (u_longlong_t)generation, | |
2678 | (u_longlong_t)spa->spa_config_generation); | |
9babb374 BB |
2679 | } |
2680 | return (vd); | |
2681 | } | |
2682 | return (NULL); | |
2683 | } | |
2684 | ||
2685 | /* | |
2686 | * Traverse the vdev tree calling the supplied function. | |
2687 | * We continue to walk the tree until we either have walked all | |
2688 | * children or we receive a non-NULL return from the callback. | |
2689 | * If a NULL callback is passed, then we just return back the first | |
2690 | * leaf vdev we encounter. | |
2691 | */ | |
2692 | vdev_t * | |
2693 | vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg) | |
2694 | { | |
d6320ddb BB |
2695 | uint_t c; |
2696 | ||
9babb374 BB |
2697 | if (vd->vdev_ops->vdev_op_leaf) { |
2698 | if (func == NULL) | |
2699 | return (vd); | |
2700 | else | |
2701 | return (func(vd, arg)); | |
2702 | } | |
2703 | ||
d6320ddb | 2704 | for (c = 0; c < vd->vdev_children; c++) { |
9babb374 BB |
2705 | vdev_t *cvd = vd->vdev_child[c]; |
2706 | if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL) | |
2707 | return (cvd); | |
2708 | } | |
2709 | return (NULL); | |
2710 | } | |
2711 | ||
34dc7c2f BB |
2712 | /* |
2713 | * Verify that dynamic LUN growth works as expected. | |
2714 | */ | |
428870ff | 2715 | /* ARGSUSED */ |
34dc7c2f | 2716 | void |
428870ff | 2717 | ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id) |
34dc7c2f | 2718 | { |
428870ff BB |
2719 | ztest_shared_t *zs = ztest_shared; |
2720 | spa_t *spa = zs->zs_spa; | |
2721 | vdev_t *vd, *tvd; | |
2722 | metaslab_class_t *mc; | |
2723 | metaslab_group_t *mg; | |
9babb374 | 2724 | size_t psize, newsize; |
428870ff BB |
2725 | uint64_t top; |
2726 | uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count; | |
34dc7c2f | 2727 | |
1e33ac1e | 2728 | mutex_enter(&zs->zs_vdev_lock); |
9babb374 BB |
2729 | spa_config_enter(spa, SCL_STATE, spa, RW_READER); |
2730 | ||
428870ff | 2731 | top = ztest_random_vdev_top(spa, B_TRUE); |
9babb374 | 2732 | |
428870ff BB |
2733 | tvd = spa->spa_root_vdev->vdev_child[top]; |
2734 | mg = tvd->vdev_mg; | |
2735 | mc = mg->mg_class; | |
2736 | old_ms_count = tvd->vdev_ms_count; | |
2737 | old_class_space = metaslab_class_get_space(mc); | |
34dc7c2f BB |
2738 | |
2739 | /* | |
9babb374 BB |
2740 | * Determine the size of the first leaf vdev associated with |
2741 | * our top-level device. | |
34dc7c2f | 2742 | */ |
9babb374 BB |
2743 | vd = vdev_walk_tree(tvd, NULL, NULL); |
2744 | ASSERT3P(vd, !=, NULL); | |
2745 | ASSERT(vd->vdev_ops->vdev_op_leaf); | |
34dc7c2f | 2746 | |
9babb374 | 2747 | psize = vd->vdev_psize; |
34dc7c2f | 2748 | |
9babb374 | 2749 | /* |
428870ff BB |
2750 | * We only try to expand the vdev if it's healthy, less than 4x its |
2751 | * original size, and it has a valid psize. | |
9babb374 | 2752 | */ |
428870ff BB |
2753 | if (tvd->vdev_state != VDEV_STATE_HEALTHY || |
2754 | psize == 0 || psize >= 4 * zopt_vdev_size) { | |
9babb374 | 2755 | spa_config_exit(spa, SCL_STATE, spa); |
1e33ac1e | 2756 | mutex_exit(&zs->zs_vdev_lock); |
9babb374 BB |
2757 | return; |
2758 | } | |
2759 | ASSERT(psize > 0); | |
2760 | newsize = psize + psize / 8; | |
2761 | ASSERT3U(newsize, >, psize); | |
34dc7c2f | 2762 | |
9babb374 | 2763 | if (zopt_verbose >= 6) { |
428870ff | 2764 | (void) printf("Expanding LUN %s from %lu to %lu\n", |
9babb374 BB |
2765 | vd->vdev_path, (ulong_t)psize, (ulong_t)newsize); |
2766 | } | |
2767 | ||
9babb374 BB |
2768 | /* |
2769 | * Growing the vdev is a two step process: | |
2770 | * 1). expand the physical size (i.e. relabel) | |
2771 | * 2). online the vdev to create the new metaslabs | |
2772 | */ | |
2773 | if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL || | |
2774 | vdev_walk_tree(tvd, online_vdev, NULL) != NULL || | |
2775 | tvd->vdev_state != VDEV_STATE_HEALTHY) { | |
2776 | if (zopt_verbose >= 5) { | |
2777 | (void) printf("Could not expand LUN because " | |
428870ff | 2778 | "the vdev configuration changed.\n"); |
34dc7c2f | 2779 | } |
428870ff | 2780 | spa_config_exit(spa, SCL_STATE, spa); |
1e33ac1e | 2781 | mutex_exit(&zs->zs_vdev_lock); |
9babb374 | 2782 | return; |
34dc7c2f BB |
2783 | } |
2784 | ||
428870ff | 2785 | spa_config_exit(spa, SCL_STATE, spa); |
9babb374 BB |
2786 | |
2787 | /* | |
2788 | * Expanding the LUN will update the config asynchronously, | |
2789 | * thus we must wait for the async thread to complete any | |
2790 | * pending tasks before proceeding. | |
2791 | */ | |
428870ff BB |
2792 | for (;;) { |
2793 | boolean_t done; | |
2794 | mutex_enter(&spa->spa_async_lock); | |
2795 | done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks); | |
2796 | mutex_exit(&spa->spa_async_lock); | |
2797 | if (done) | |
2798 | break; | |
2799 | txg_wait_synced(spa_get_dsl(spa), 0); | |
2800 | (void) poll(NULL, 0, 100); | |
2801 | } | |
9babb374 BB |
2802 | |
2803 | spa_config_enter(spa, SCL_STATE, spa, RW_READER); | |
428870ff BB |
2804 | |
2805 | tvd = spa->spa_root_vdev->vdev_child[top]; | |
2806 | new_ms_count = tvd->vdev_ms_count; | |
2807 | new_class_space = metaslab_class_get_space(mc); | |
2808 | ||
2809 | if (tvd->vdev_mg != mg || mg->mg_class != mc) { | |
2810 | if (zopt_verbose >= 5) { | |
2811 | (void) printf("Could not verify LUN expansion due to " | |
2812 | "intervening vdev offline or remove.\n"); | |
2813 | } | |
2814 | spa_config_exit(spa, SCL_STATE, spa); | |
1e33ac1e | 2815 | mutex_exit(&zs->zs_vdev_lock); |
428870ff BB |
2816 | return; |
2817 | } | |
2818 | ||
2819 | /* | |
2820 | * Make sure we were able to grow the vdev. | |
2821 | */ | |
2822 | if (new_ms_count <= old_ms_count) | |
2823 | fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n", | |
2824 | old_ms_count, new_ms_count); | |
9babb374 BB |
2825 | |
2826 | /* | |
2827 | * Make sure we were able to grow the pool. | |
2828 | */ | |
428870ff BB |
2829 | if (new_class_space <= old_class_space) |
2830 | fatal(0, "LUN expansion failed: class_space %llu <= %llu\n", | |
2831 | old_class_space, new_class_space); | |
2832 | ||
2833 | if (zopt_verbose >= 5) { | |
9babb374 BB |
2834 | char oldnumbuf[6], newnumbuf[6]; |
2835 | ||
428870ff BB |
2836 | nicenum(old_class_space, oldnumbuf); |
2837 | nicenum(new_class_space, newnumbuf); | |
9babb374 BB |
2838 | (void) printf("%s grew from %s to %s\n", |
2839 | spa->spa_name, oldnumbuf, newnumbuf); | |
2840 | } | |
428870ff | 2841 | |
9babb374 | 2842 | spa_config_exit(spa, SCL_STATE, spa); |
1e33ac1e | 2843 | mutex_exit(&zs->zs_vdev_lock); |
34dc7c2f BB |
2844 | } |
2845 | ||
428870ff BB |
2846 | /* |
2847 | * Verify that dmu_objset_{create,destroy,open,close} work as expected. | |
2848 | */ | |
34dc7c2f BB |
2849 | /* ARGSUSED */ |
2850 | static void | |
428870ff | 2851 | ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx) |
34dc7c2f BB |
2852 | { |
2853 | /* | |
428870ff | 2854 | * Create the objects common to all ztest datasets. |
34dc7c2f | 2855 | */ |
428870ff | 2856 | VERIFY(zap_create_claim(os, ZTEST_DIROBJ, |
34dc7c2f | 2857 | DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0); |
428870ff | 2858 | } |
34dc7c2f | 2859 | |
428870ff BB |
2860 | static int |
2861 | ztest_dataset_create(char *dsname) | |
2862 | { | |
2863 | uint64_t zilset = ztest_random(100); | |
2864 | int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0, | |
2865 | ztest_objset_create_cb, NULL); | |
2866 | ||
2867 | if (err || zilset < 80) | |
2868 | return (err); | |
2869 | ||
2870 | (void) printf("Setting dataset %s to sync always\n", dsname); | |
2871 | return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC, | |
2872 | ZFS_SYNC_ALWAYS, B_FALSE)); | |
34dc7c2f BB |
2873 | } |
2874 | ||
428870ff | 2875 | /* ARGSUSED */ |
34dc7c2f | 2876 | static int |
428870ff | 2877 | ztest_objset_destroy_cb(const char *name, void *arg) |
34dc7c2f | 2878 | { |
34dc7c2f | 2879 | objset_t *os; |
428870ff | 2880 | dmu_object_info_t doi; |
34dc7c2f BB |
2881 | int error; |
2882 | ||
2883 | /* | |
2884 | * Verify that the dataset contains a directory object. | |
2885 | */ | |
428870ff BB |
2886 | VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os)); |
2887 | error = dmu_object_info(os, ZTEST_DIROBJ, &doi); | |
34dc7c2f BB |
2888 | if (error != ENOENT) { |
2889 | /* We could have crashed in the middle of destroying it */ | |
2890 | ASSERT3U(error, ==, 0); | |
428870ff BB |
2891 | ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER); |
2892 | ASSERT3S(doi.doi_physical_blocks_512, >=, 0); | |
34dc7c2f | 2893 | } |
428870ff | 2894 | dmu_objset_rele(os, FTAG); |
34dc7c2f BB |
2895 | |
2896 | /* | |
2897 | * Destroy the dataset. | |
2898 | */ | |
428870ff | 2899 | VERIFY3U(0, ==, dmu_objset_destroy(name, B_FALSE)); |
34dc7c2f BB |
2900 | return (0); |
2901 | } | |
2902 | ||
428870ff BB |
2903 | static boolean_t |
2904 | ztest_snapshot_create(char *osname, uint64_t id) | |
34dc7c2f | 2905 | { |
428870ff BB |
2906 | char snapname[MAXNAMELEN]; |
2907 | int error; | |
2908 | ||
2909 | (void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname, | |
2910 | (u_longlong_t)id); | |
2911 | ||
2912 | error = dmu_objset_snapshot(osname, strchr(snapname, '@') + 1, | |
572e2857 | 2913 | NULL, NULL, B_FALSE, B_FALSE, -1); |
428870ff BB |
2914 | if (error == ENOSPC) { |
2915 | ztest_record_enospc(FTAG); | |
2916 | return (B_FALSE); | |
2917 | } | |
2918 | if (error != 0 && error != EEXIST) | |
2919 | fatal(0, "ztest_snapshot_create(%s) = %d", snapname, error); | |
2920 | return (B_TRUE); | |
2921 | } | |
2922 | ||
2923 | static boolean_t | |
2924 | ztest_snapshot_destroy(char *osname, uint64_t id) | |
2925 | { | |
2926 | char snapname[MAXNAMELEN]; | |
2927 | int error; | |
2928 | ||
2929 | (void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname, | |
2930 | (u_longlong_t)id); | |
2931 | ||
2932 | error = dmu_objset_destroy(snapname, B_FALSE); | |
2933 | if (error != 0 && error != ENOENT) | |
2934 | fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error); | |
2935 | return (B_TRUE); | |
34dc7c2f BB |
2936 | } |
2937 | ||
428870ff | 2938 | /* ARGSUSED */ |
34dc7c2f | 2939 | void |
428870ff | 2940 | ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id) |
34dc7c2f | 2941 | { |
428870ff BB |
2942 | ztest_shared_t *zs = ztest_shared; |
2943 | ztest_ds_t zdtmp; | |
2944 | int iters; | |
34dc7c2f | 2945 | int error; |
b128c09f | 2946 | objset_t *os, *os2; |
428870ff | 2947 | char name[MAXNAMELEN]; |
34dc7c2f | 2948 | zilog_t *zilog; |
d6320ddb | 2949 | int i; |
34dc7c2f | 2950 | |
1e33ac1e | 2951 | (void) rw_enter(&zs->zs_name_lock, RW_READER); |
34dc7c2f | 2952 | |
428870ff BB |
2953 | (void) snprintf(name, MAXNAMELEN, "%s/temp_%llu", |
2954 | zs->zs_pool, (u_longlong_t)id); | |
34dc7c2f BB |
2955 | |
2956 | /* | |
2957 | * If this dataset exists from a previous run, process its replay log | |
2958 | * half of the time. If we don't replay it, then dmu_objset_destroy() | |
428870ff | 2959 | * (invoked from ztest_objset_destroy_cb()) should just throw it away. |
34dc7c2f BB |
2960 | */ |
2961 | if (ztest_random(2) == 0 && | |
428870ff BB |
2962 | dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) { |
2963 | ztest_zd_init(&zdtmp, os); | |
2964 | zil_replay(os, &zdtmp, ztest_replay_vector); | |
2965 | ztest_zd_fini(&zdtmp); | |
2966 | dmu_objset_disown(os, FTAG); | |
34dc7c2f BB |
2967 | } |
2968 | ||
2969 | /* | |
2970 | * There may be an old instance of the dataset we're about to | |
2971 | * create lying around from a previous run. If so, destroy it | |
2972 | * and all of its snapshots. | |
2973 | */ | |
428870ff | 2974 | (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL, |
34dc7c2f BB |
2975 | DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS); |
2976 | ||
2977 | /* | |
2978 | * Verify that the destroyed dataset is no longer in the namespace. | |
2979 | */ | |
428870ff | 2980 | VERIFY3U(ENOENT, ==, dmu_objset_hold(name, FTAG, &os)); |
34dc7c2f BB |
2981 | |
2982 | /* | |
2983 | * Verify that we can create a new dataset. | |
2984 | */ | |
428870ff | 2985 | error = ztest_dataset_create(name); |
34dc7c2f BB |
2986 | if (error) { |
2987 | if (error == ENOSPC) { | |
428870ff | 2988 | ztest_record_enospc(FTAG); |
1e33ac1e | 2989 | (void) rw_exit(&zs->zs_name_lock); |
34dc7c2f BB |
2990 | return; |
2991 | } | |
2992 | fatal(0, "dmu_objset_create(%s) = %d", name, error); | |
2993 | } | |
2994 | ||
428870ff BB |
2995 | VERIFY3U(0, ==, |
2996 | dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os)); | |
2997 | ||
2998 | ztest_zd_init(&zdtmp, os); | |
34dc7c2f BB |
2999 | |
3000 | /* | |
3001 | * Open the intent log for it. | |
3002 | */ | |
428870ff | 3003 | zilog = zil_open(os, ztest_get_data); |
34dc7c2f BB |
3004 | |
3005 | /* | |
428870ff BB |
3006 | * Put some objects in there, do a little I/O to them, |
3007 | * and randomly take a couple of snapshots along the way. | |
34dc7c2f | 3008 | */ |
428870ff | 3009 | iters = ztest_random(5); |
d6320ddb | 3010 | for (i = 0; i < iters; i++) { |
428870ff BB |
3011 | ztest_dmu_object_alloc_free(&zdtmp, id); |
3012 | if (ztest_random(iters) == 0) | |
3013 | (void) ztest_snapshot_create(name, i); | |
34dc7c2f BB |
3014 | } |
3015 | ||
3016 | /* | |
3017 | * Verify that we cannot create an existing dataset. | |
3018 | */ | |
428870ff BB |
3019 | VERIFY3U(EEXIST, ==, |
3020 | dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL)); | |
34dc7c2f BB |
3021 | |
3022 | /* | |
428870ff | 3023 | * Verify that we can hold an objset that is also owned. |
b128c09f | 3024 | */ |
428870ff BB |
3025 | VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2)); |
3026 | dmu_objset_rele(os2, FTAG); | |
34dc7c2f | 3027 | |
428870ff BB |
3028 | /* |
3029 | * Verify that we cannot own an objset that is already owned. | |
3030 | */ | |
3031 | VERIFY3U(EBUSY, ==, | |
3032 | dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2)); | |
34dc7c2f | 3033 | |
428870ff BB |
3034 | zil_close(zilog); |
3035 | dmu_objset_disown(os, FTAG); | |
3036 | ztest_zd_fini(&zdtmp); | |
34dc7c2f | 3037 | |
1e33ac1e | 3038 | (void) rw_exit(&zs->zs_name_lock); |
34dc7c2f BB |
3039 | } |
3040 | ||
3041 | /* | |
3042 | * Verify that dmu_snapshot_{create,destroy,open,close} work as expected. | |
3043 | */ | |
3044 | void | |
428870ff | 3045 | ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id) |
34dc7c2f | 3046 | { |
428870ff | 3047 | ztest_shared_t *zs = ztest_shared; |
34dc7c2f | 3048 | |
1e33ac1e | 3049 | (void) rw_enter(&zs->zs_name_lock, RW_READER); |
428870ff BB |
3050 | (void) ztest_snapshot_destroy(zd->zd_name, id); |
3051 | (void) ztest_snapshot_create(zd->zd_name, id); | |
1e33ac1e | 3052 | (void) rw_exit(&zs->zs_name_lock); |
34dc7c2f BB |
3053 | } |
3054 | ||
9babb374 BB |
3055 | /* |
3056 | * Cleanup non-standard snapshots and clones. | |
3057 | */ | |
3058 | void | |
428870ff | 3059 | ztest_dsl_dataset_cleanup(char *osname, uint64_t id) |
9babb374 | 3060 | { |
428870ff BB |
3061 | char snap1name[MAXNAMELEN]; |
3062 | char clone1name[MAXNAMELEN]; | |
3063 | char snap2name[MAXNAMELEN]; | |
3064 | char clone2name[MAXNAMELEN]; | |
3065 | char snap3name[MAXNAMELEN]; | |
9babb374 BB |
3066 | int error; |
3067 | ||
b8864a23 BB |
3068 | (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", |
3069 | osname, (u_longlong_t)id); | |
3070 | (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", | |
3071 | osname, (u_longlong_t)id); | |
3072 | (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", | |
3073 | clone1name, (u_longlong_t)id); | |
3074 | (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", | |
3075 | osname, (u_longlong_t)id); | |
3076 | (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", | |
3077 | clone1name, (u_longlong_t)id); | |
9babb374 | 3078 | |
45d1cae3 | 3079 | error = dmu_objset_destroy(clone2name, B_FALSE); |
9babb374 BB |
3080 | if (error && error != ENOENT) |
3081 | fatal(0, "dmu_objset_destroy(%s) = %d", clone2name, error); | |
45d1cae3 | 3082 | error = dmu_objset_destroy(snap3name, B_FALSE); |
9babb374 BB |
3083 | if (error && error != ENOENT) |
3084 | fatal(0, "dmu_objset_destroy(%s) = %d", snap3name, error); | |
45d1cae3 | 3085 | error = dmu_objset_destroy(snap2name, B_FALSE); |
9babb374 BB |
3086 | if (error && error != ENOENT) |
3087 | fatal(0, "dmu_objset_destroy(%s) = %d", snap2name, error); | |
45d1cae3 | 3088 | error = dmu_objset_destroy(clone1name, B_FALSE); |
9babb374 BB |
3089 | if (error && error != ENOENT) |
3090 | fatal(0, "dmu_objset_destroy(%s) = %d", clone1name, error); | |
45d1cae3 | 3091 | error = dmu_objset_destroy(snap1name, B_FALSE); |
9babb374 BB |
3092 | if (error && error != ENOENT) |
3093 | fatal(0, "dmu_objset_destroy(%s) = %d", snap1name, error); | |
3094 | } | |
3095 | ||
3096 | /* | |
3097 | * Verify dsl_dataset_promote handles EBUSY | |
3098 | */ | |
3099 | void | |
428870ff | 3100 | ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id) |
9babb374 | 3101 | { |
428870ff | 3102 | ztest_shared_t *zs = ztest_shared; |
9babb374 BB |
3103 | objset_t *clone; |
3104 | dsl_dataset_t *ds; | |
428870ff BB |
3105 | char snap1name[MAXNAMELEN]; |
3106 | char clone1name[MAXNAMELEN]; | |
3107 | char snap2name[MAXNAMELEN]; | |
3108 | char clone2name[MAXNAMELEN]; | |
3109 | char snap3name[MAXNAMELEN]; | |
3110 | char *osname = zd->zd_name; | |
3111 | int error; | |
9babb374 | 3112 | |
1e33ac1e | 3113 | (void) rw_enter(&zs->zs_name_lock, RW_READER); |
9babb374 | 3114 | |
428870ff | 3115 | ztest_dsl_dataset_cleanup(osname, id); |
9babb374 | 3116 | |
b8864a23 BB |
3117 | (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", |
3118 | osname, (u_longlong_t)id); | |
3119 | (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", | |
3120 | osname, (u_longlong_t)id); | |
3121 | (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", | |
3122 | clone1name, (u_longlong_t)id); | |
3123 | (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", | |
3124 | osname, (u_longlong_t)id); | |
3125 | (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", | |
3126 | clone1name, (u_longlong_t)id); | |
9babb374 BB |
3127 | |
3128 | error = dmu_objset_snapshot(osname, strchr(snap1name, '@')+1, | |
572e2857 | 3129 | NULL, NULL, B_FALSE, B_FALSE, -1); |
9babb374 BB |
3130 | if (error && error != EEXIST) { |
3131 | if (error == ENOSPC) { | |
428870ff | 3132 | ztest_record_enospc(FTAG); |
9babb374 BB |
3133 | goto out; |
3134 | } | |
3135 | fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error); | |
3136 | } | |
3137 | ||
428870ff | 3138 | error = dmu_objset_hold(snap1name, FTAG, &clone); |
9babb374 BB |
3139 | if (error) |
3140 | fatal(0, "dmu_open_snapshot(%s) = %d", snap1name, error); | |
3141 | ||
428870ff BB |
3142 | error = dmu_objset_clone(clone1name, dmu_objset_ds(clone), 0); |
3143 | dmu_objset_rele(clone, FTAG); | |
9babb374 BB |
3144 | if (error) { |
3145 | if (error == ENOSPC) { | |
428870ff | 3146 | ztest_record_enospc(FTAG); |
9babb374 BB |
3147 | goto out; |
3148 | } | |
3149 | fatal(0, "dmu_objset_create(%s) = %d", clone1name, error); | |
3150 | } | |
3151 | ||
428870ff | 3152 | error = dmu_objset_snapshot(clone1name, strchr(snap2name, '@')+1, |
572e2857 | 3153 | NULL, NULL, B_FALSE, B_FALSE, -1); |
428870ff BB |
3154 | if (error && error != EEXIST) { |
3155 | if (error == ENOSPC) { | |
3156 | ztest_record_enospc(FTAG); | |
3157 | goto out; | |
34dc7c2f | 3158 | } |
428870ff BB |
3159 | fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error); |
3160 | } | |
34dc7c2f | 3161 | |
428870ff | 3162 | error = dmu_objset_snapshot(clone1name, strchr(snap3name, '@')+1, |
572e2857 | 3163 | NULL, NULL, B_FALSE, B_FALSE, -1); |
428870ff BB |
3164 | if (error && error != EEXIST) { |
3165 | if (error == ENOSPC) { | |
3166 | ztest_record_enospc(FTAG); | |
3167 | goto out; | |
3168 | } | |
3169 | fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error); | |
3170 | } | |
34dc7c2f | 3171 | |
428870ff BB |
3172 | error = dmu_objset_hold(snap3name, FTAG, &clone); |
3173 | if (error) | |
3174 | fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error); | |
34dc7c2f | 3175 | |
428870ff BB |
3176 | error = dmu_objset_clone(clone2name, dmu_objset_ds(clone), 0); |
3177 | dmu_objset_rele(clone, FTAG); | |
3178 | if (error) { | |
3179 | if (error == ENOSPC) { | |
3180 | ztest_record_enospc(FTAG); | |
3181 | goto out; | |
3182 | } | |
3183 | fatal(0, "dmu_objset_create(%s) = %d", clone2name, error); | |
3184 | } | |
34dc7c2f | 3185 | |
428870ff BB |
3186 | error = dsl_dataset_own(snap2name, B_FALSE, FTAG, &ds); |
3187 | if (error) | |
3188 | fatal(0, "dsl_dataset_own(%s) = %d", snap2name, error); | |
3189 | error = dsl_dataset_promote(clone2name, NULL); | |
3190 | if (error != EBUSY) | |
3191 | fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name, | |
3192 | error); | |
3193 | dsl_dataset_disown(ds, FTAG); | |
34dc7c2f | 3194 | |
428870ff BB |
3195 | out: |
3196 | ztest_dsl_dataset_cleanup(osname, id); | |
34dc7c2f | 3197 | |
1e33ac1e | 3198 | (void) rw_exit(&zs->zs_name_lock); |
428870ff | 3199 | } |
34dc7c2f | 3200 | |
428870ff BB |
3201 | /* |
3202 | * Verify that dmu_object_{alloc,free} work as expected. | |
3203 | */ | |
3204 | void | |
3205 | ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id) | |
3206 | { | |
3207 | ztest_od_t od[4]; | |
3208 | int batchsize = sizeof (od) / sizeof (od[0]); | |
d6320ddb | 3209 | int b; |
34dc7c2f | 3210 | |
d6320ddb | 3211 | for (b = 0; b < batchsize; b++) |
428870ff | 3212 | ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0); |
34dc7c2f | 3213 | |
428870ff BB |
3214 | /* |
3215 | * Destroy the previous batch of objects, create a new batch, | |
3216 | * and do some I/O on the new objects. | |
3217 | */ | |
3218 | if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0) | |
3219 | return; | |
34dc7c2f | 3220 | |
428870ff BB |
3221 | while (ztest_random(4 * batchsize) != 0) |
3222 | ztest_io(zd, od[ztest_random(batchsize)].od_object, | |
3223 | ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT); | |
34dc7c2f BB |
3224 | } |
3225 | ||
3226 | /* | |
3227 | * Verify that dmu_{read,write} work as expected. | |
3228 | */ | |
34dc7c2f | 3229 | void |
428870ff | 3230 | ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id) |
34dc7c2f | 3231 | { |
428870ff BB |
3232 | objset_t *os = zd->zd_os; |
3233 | ztest_od_t od[2]; | |
34dc7c2f BB |
3234 | dmu_tx_t *tx; |
3235 | int i, freeit, error; | |
3236 | uint64_t n, s, txg; | |
3237 | bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT; | |
428870ff BB |
3238 | uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize; |
3239 | uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t); | |
34dc7c2f BB |
3240 | uint64_t regions = 997; |
3241 | uint64_t stride = 123456789ULL; | |
3242 | uint64_t width = 40; | |
3243 | int free_percent = 5; | |
3244 | ||
3245 | /* | |
3246 | * This test uses two objects, packobj and bigobj, that are always | |
3247 | * updated together (i.e. in the same tx) so that their contents are | |
3248 | * in sync and can be compared. Their contents relate to each other | |
3249 | * in a simple way: packobj is a dense array of 'bufwad' structures, | |
3250 | * while bigobj is a sparse array of the same bufwads. Specifically, | |
3251 | * for any index n, there are three bufwads that should be identical: | |
3252 | * | |
3253 | * packobj, at offset n * sizeof (bufwad_t) | |
3254 | * bigobj, at the head of the nth chunk | |
3255 | * bigobj, at the tail of the nth chunk | |
3256 | * | |
3257 | * The chunk size is arbitrary. It doesn't have to be a power of two, | |
3258 | * and it doesn't have any relation to the object blocksize. | |
3259 | * The only requirement is that it can hold at least two bufwads. | |
3260 | * | |
3261 | * Normally, we write the bufwad to each of these locations. | |
3262 | * However, free_percent of the time we instead write zeroes to | |
3263 | * packobj and perform a dmu_free_range() on bigobj. By comparing | |
3264 | * bigobj to packobj, we can verify that the DMU is correctly | |
3265 | * tracking which parts of an object are allocated and free, | |
3266 | * and that the contents of the allocated blocks are correct. | |
3267 | */ | |
3268 | ||
3269 | /* | |
3270 | * Read the directory info. If it's the first time, set things up. | |
3271 | */ | |
428870ff BB |
3272 | ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize); |
3273 | ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize); | |
34dc7c2f | 3274 | |
428870ff BB |
3275 | if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0) |
3276 | return; | |
34dc7c2f | 3277 | |
428870ff BB |
3278 | bigobj = od[0].od_object; |
3279 | packobj = od[1].od_object; | |
3280 | chunksize = od[0].od_gen; | |
3281 | ASSERT(chunksize == od[1].od_gen); | |
34dc7c2f BB |
3282 | |
3283 | /* | |
3284 | * Prefetch a random chunk of the big object. | |
3285 | * Our aim here is to get some async reads in flight | |
3286 | * for blocks that we may free below; the DMU should | |
3287 | * handle this race correctly. | |
3288 | */ | |
3289 | n = ztest_random(regions) * stride + ztest_random(width); | |
3290 | s = 1 + ztest_random(2 * width - 1); | |
428870ff | 3291 | dmu_prefetch(os, bigobj, n * chunksize, s * chunksize); |
34dc7c2f BB |
3292 | |
3293 | /* | |
3294 | * Pick a random index and compute the offsets into packobj and bigobj. | |
3295 | */ | |
3296 | n = ztest_random(regions) * stride + ztest_random(width); | |
3297 | s = 1 + ztest_random(width - 1); | |
3298 | ||
3299 | packoff = n * sizeof (bufwad_t); | |
3300 | packsize = s * sizeof (bufwad_t); | |
3301 | ||
428870ff BB |
3302 | bigoff = n * chunksize; |
3303 | bigsize = s * chunksize; | |
34dc7c2f BB |
3304 | |
3305 | packbuf = umem_alloc(packsize, UMEM_NOFAIL); | |
3306 | bigbuf = umem_alloc(bigsize, UMEM_NOFAIL); | |
3307 | ||
3308 | /* | |
3309 | * free_percent of the time, free a range of bigobj rather than | |
3310 | * overwriting it. | |
3311 | */ | |
3312 | freeit = (ztest_random(100) < free_percent); | |
3313 | ||
3314 | /* | |
3315 | * Read the current contents of our objects. | |
3316 | */ | |
428870ff | 3317 | error = dmu_read(os, packobj, packoff, packsize, packbuf, |
9babb374 | 3318 | DMU_READ_PREFETCH); |
34dc7c2f | 3319 | ASSERT3U(error, ==, 0); |
428870ff | 3320 | error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf, |
9babb374 | 3321 | DMU_READ_PREFETCH); |
34dc7c2f BB |
3322 | ASSERT3U(error, ==, 0); |
3323 | ||
3324 | /* | |
3325 | * Get a tx for the mods to both packobj and bigobj. | |
3326 | */ | |
3327 | tx = dmu_tx_create(os); | |
3328 | ||
428870ff | 3329 | dmu_tx_hold_write(tx, packobj, packoff, packsize); |
34dc7c2f BB |
3330 | |
3331 | if (freeit) | |
428870ff | 3332 | dmu_tx_hold_free(tx, bigobj, bigoff, bigsize); |
34dc7c2f | 3333 | else |
428870ff | 3334 | dmu_tx_hold_write(tx, bigobj, bigoff, bigsize); |
34dc7c2f | 3335 | |
428870ff BB |
3336 | txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); |
3337 | if (txg == 0) { | |
34dc7c2f BB |
3338 | umem_free(packbuf, packsize); |
3339 | umem_free(bigbuf, bigsize); | |
3340 | return; | |
3341 | } | |
3342 | ||
428870ff BB |
3343 | dmu_object_set_checksum(os, bigobj, |
3344 | (enum zio_checksum)ztest_random_dsl_prop(ZFS_PROP_CHECKSUM), tx); | |
3345 | ||
3346 | dmu_object_set_compress(os, bigobj, | |
3347 | (enum zio_compress)ztest_random_dsl_prop(ZFS_PROP_COMPRESSION), tx); | |
34dc7c2f BB |
3348 | |
3349 | /* | |
3350 | * For each index from n to n + s, verify that the existing bufwad | |
3351 | * in packobj matches the bufwads at the head and tail of the | |
3352 | * corresponding chunk in bigobj. Then update all three bufwads | |
3353 | * with the new values we want to write out. | |
3354 | */ | |
3355 | for (i = 0; i < s; i++) { | |
3356 | /* LINTED */ | |
3357 | pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t)); | |
3358 | /* LINTED */ | |
428870ff | 3359 | bigH = (bufwad_t *)((char *)bigbuf + i * chunksize); |
34dc7c2f | 3360 | /* LINTED */ |
428870ff | 3361 | bigT = (bufwad_t *)((char *)bigH + chunksize) - 1; |
34dc7c2f BB |
3362 | |
3363 | ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize); | |
3364 | ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize); | |
3365 | ||
3366 | if (pack->bw_txg > txg) | |
3367 | fatal(0, "future leak: got %llx, open txg is %llx", | |
3368 | pack->bw_txg, txg); | |
3369 | ||
3370 | if (pack->bw_data != 0 && pack->bw_index != n + i) | |
3371 | fatal(0, "wrong index: got %llx, wanted %llx+%llx", | |
3372 | pack->bw_index, n, i); | |
3373 | ||
3374 | if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0) | |
3375 | fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH); | |
3376 | ||
3377 | if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0) | |
3378 | fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT); | |
3379 | ||
3380 | if (freeit) { | |
3381 | bzero(pack, sizeof (bufwad_t)); | |
3382 | } else { | |
3383 | pack->bw_index = n + i; | |
3384 | pack->bw_txg = txg; | |
3385 | pack->bw_data = 1 + ztest_random(-2ULL); | |
3386 | } | |
3387 | *bigH = *pack; | |
3388 | *bigT = *pack; | |
3389 | } | |
3390 | ||
3391 | /* | |
3392 | * We've verified all the old bufwads, and made new ones. | |
3393 | * Now write them out. | |
3394 | */ | |
428870ff | 3395 | dmu_write(os, packobj, packoff, packsize, packbuf, tx); |
34dc7c2f BB |
3396 | |
3397 | if (freeit) { | |
428870ff | 3398 | if (zopt_verbose >= 7) { |
34dc7c2f BB |
3399 | (void) printf("freeing offset %llx size %llx" |
3400 | " txg %llx\n", | |
3401 | (u_longlong_t)bigoff, | |
3402 | (u_longlong_t)bigsize, | |
3403 | (u_longlong_t)txg); | |
3404 | } | |
428870ff | 3405 | VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx)); |
34dc7c2f | 3406 | } else { |
428870ff | 3407 | if (zopt_verbose >= 7) { |
34dc7c2f BB |
3408 | (void) printf("writing offset %llx size %llx" |
3409 | " txg %llx\n", | |
3410 | (u_longlong_t)bigoff, | |
3411 | (u_longlong_t)bigsize, | |
3412 | (u_longlong_t)txg); | |
3413 | } | |
428870ff | 3414 | dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx); |
34dc7c2f BB |
3415 | } |
3416 | ||
3417 | dmu_tx_commit(tx); | |
3418 | ||
3419 | /* | |
3420 | * Sanity check the stuff we just wrote. | |
3421 | */ | |
3422 | { | |
3423 | void *packcheck = umem_alloc(packsize, UMEM_NOFAIL); | |
3424 | void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL); | |
3425 | ||
428870ff | 3426 | VERIFY(0 == dmu_read(os, packobj, packoff, |
9babb374 | 3427 | packsize, packcheck, DMU_READ_PREFETCH)); |
428870ff | 3428 | VERIFY(0 == dmu_read(os, bigobj, bigoff, |
9babb374 | 3429 | bigsize, bigcheck, DMU_READ_PREFETCH)); |
34dc7c2f BB |
3430 | |
3431 | ASSERT(bcmp(packbuf, packcheck, packsize) == 0); | |
3432 | ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0); | |
3433 | ||
3434 | umem_free(packcheck, packsize); | |
3435 | umem_free(bigcheck, bigsize); | |
3436 | } | |
3437 | ||
3438 | umem_free(packbuf, packsize); | |
3439 | umem_free(bigbuf, bigsize); | |
3440 | } | |
3441 | ||
9babb374 BB |
3442 | void |
3443 | compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf, | |
428870ff | 3444 | uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg) |
9babb374 BB |
3445 | { |
3446 | uint64_t i; | |
3447 | bufwad_t *pack; | |
3448 | bufwad_t *bigH; | |
3449 | bufwad_t *bigT; | |
3450 | ||
3451 | /* | |
3452 | * For each index from n to n + s, verify that the existing bufwad | |
3453 | * in packobj matches the bufwads at the head and tail of the | |
3454 | * corresponding chunk in bigobj. Then update all three bufwads | |
3455 | * with the new values we want to write out. | |
3456 | */ | |
3457 | for (i = 0; i < s; i++) { | |
3458 | /* LINTED */ | |
3459 | pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t)); | |
3460 | /* LINTED */ | |
428870ff | 3461 | bigH = (bufwad_t *)((char *)bigbuf + i * chunksize); |
9babb374 | 3462 | /* LINTED */ |
428870ff | 3463 | bigT = (bufwad_t *)((char *)bigH + chunksize) - 1; |
9babb374 BB |
3464 | |
3465 | ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize); | |
3466 | ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize); | |
3467 | ||
3468 | if (pack->bw_txg > txg) | |
3469 | fatal(0, "future leak: got %llx, open txg is %llx", | |
3470 | pack->bw_txg, txg); | |
3471 | ||
3472 | if (pack->bw_data != 0 && pack->bw_index != n + i) | |
3473 | fatal(0, "wrong index: got %llx, wanted %llx+%llx", | |
3474 | pack->bw_index, n, i); | |
3475 | ||
3476 | if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0) | |
3477 | fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH); | |
3478 | ||
3479 | if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0) | |
3480 | fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT); | |
3481 | ||
3482 | pack->bw_index = n + i; | |
3483 | pack->bw_txg = txg; | |
3484 | pack->bw_data = 1 + ztest_random(-2ULL); | |
3485 | ||
3486 | *bigH = *pack; | |
3487 | *bigT = *pack; | |
3488 | } | |
3489 | } | |
3490 | ||
3491 | void | |
428870ff | 3492 | ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id) |
9babb374 | 3493 | { |
428870ff BB |
3494 | objset_t *os = zd->zd_os; |
3495 | ztest_od_t od[2]; | |
9babb374 BB |
3496 | dmu_tx_t *tx; |
3497 | uint64_t i; | |
3498 | int error; | |
3499 | uint64_t n, s, txg; | |
3500 | bufwad_t *packbuf, *bigbuf; | |
428870ff BB |
3501 | uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize; |
3502 | uint64_t blocksize = ztest_random_blocksize(); | |
3503 | uint64_t chunksize = blocksize; | |
9babb374 BB |
3504 | uint64_t regions = 997; |
3505 | uint64_t stride = 123456789ULL; | |
3506 | uint64_t width = 9; | |
3507 | dmu_buf_t *bonus_db; | |
3508 | arc_buf_t **bigbuf_arcbufs; | |
428870ff | 3509 | dmu_object_info_t doi; |
9babb374 BB |
3510 | |
3511 | /* | |
3512 | * This test uses two objects, packobj and bigobj, that are always | |
3513 | * updated together (i.e. in the same tx) so that their contents are | |
3514 | * in sync and can be compared. Their contents relate to each other | |
3515 | * in a simple way: packobj is a dense array of 'bufwad' structures, | |
3516 | * while bigobj is a sparse array of the same bufwads. Specifically, | |
3517 | * for any index n, there are three bufwads that should be identical: | |
3518 | * | |
3519 | * packobj, at offset n * sizeof (bufwad_t) | |
3520 | * bigobj, at the head of the nth chunk | |
3521 | * bigobj, at the tail of the nth chunk | |
3522 | * | |
3523 | * The chunk size is set equal to bigobj block size so that | |
3524 | * dmu_assign_arcbuf() can be tested for object updates. | |
3525 | */ | |
3526 | ||
3527 | /* | |
3528 | * Read the directory info. If it's the first time, set things up. | |
3529 | */ | |
428870ff BB |
3530 | ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0); |
3531 | ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize); | |
9babb374 | 3532 | |
428870ff BB |
3533 | if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0) |
3534 | return; | |
9babb374 | 3535 | |
428870ff BB |
3536 | bigobj = od[0].od_object; |
3537 | packobj = od[1].od_object; | |
3538 | blocksize = od[0].od_blocksize; | |
3539 | chunksize = blocksize; | |
3540 | ASSERT(chunksize == od[1].od_gen); | |
9babb374 | 3541 | |
428870ff BB |
3542 | VERIFY(dmu_object_info(os, bigobj, &doi) == 0); |
3543 | VERIFY(ISP2(doi.doi_data_block_size)); | |
3544 | VERIFY(chunksize == doi.doi_data_block_size); | |
3545 | VERIFY(chunksize >= 2 * sizeof (bufwad_t)); | |
9babb374 BB |
3546 | |
3547 | /* | |
3548 | * Pick a random index and compute the offsets into packobj and bigobj. | |
3549 | */ | |
3550 | n = ztest_random(regions) * stride + ztest_random(width); | |
3551 | s = 1 + ztest_random(width - 1); | |
3552 | ||
3553 | packoff = n * sizeof (bufwad_t); | |
3554 | packsize = s * sizeof (bufwad_t); | |
3555 | ||
428870ff BB |
3556 | bigoff = n * chunksize; |
3557 | bigsize = s * chunksize; | |
9babb374 BB |
3558 | |
3559 | packbuf = umem_zalloc(packsize, UMEM_NOFAIL); | |
3560 | bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL); | |
3561 | ||
428870ff | 3562 | VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db)); |
9babb374 BB |
3563 | |
3564 | bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL); | |
3565 | ||
3566 | /* | |
3567 | * Iteration 0 test zcopy for DB_UNCACHED dbufs. | |
3568 | * Iteration 1 test zcopy to already referenced dbufs. | |
3569 | * Iteration 2 test zcopy to dirty dbuf in the same txg. | |
3570 | * Iteration 3 test zcopy to dbuf dirty in previous txg. | |
3571 | * Iteration 4 test zcopy when dbuf is no longer dirty. | |
3572 | * Iteration 5 test zcopy when it can't be done. | |
3573 | * Iteration 6 one more zcopy write. | |
3574 | */ | |
3575 | for (i = 0; i < 7; i++) { | |
3576 | uint64_t j; | |
3577 | uint64_t off; | |
3578 | ||
3579 | /* | |
3580 | * In iteration 5 (i == 5) use arcbufs | |
3581 | * that don't match bigobj blksz to test | |
3582 | * dmu_assign_arcbuf() when it can't directly | |
3583 | * assign an arcbuf to a dbuf. | |
3584 | */ | |
3585 | for (j = 0; j < s; j++) { | |
3586 | if (i != 5) { | |
3587 | bigbuf_arcbufs[j] = | |
428870ff | 3588 | dmu_request_arcbuf(bonus_db, chunksize); |
9babb374 BB |
3589 | } else { |
3590 | bigbuf_arcbufs[2 * j] = | |
428870ff | 3591 | dmu_request_arcbuf(bonus_db, chunksize / 2); |
9babb374 | 3592 | bigbuf_arcbufs[2 * j + 1] = |
428870ff | 3593 | dmu_request_arcbuf(bonus_db, chunksize / 2); |
9babb374 BB |
3594 | } |
3595 | } | |
3596 | ||
3597 | /* | |
3598 | * Get a tx for the mods to both packobj and bigobj. | |
3599 | */ | |
3600 | tx = dmu_tx_create(os); | |
3601 | ||
428870ff BB |
3602 | dmu_tx_hold_write(tx, packobj, packoff, packsize); |
3603 | dmu_tx_hold_write(tx, bigobj, bigoff, bigsize); | |
9babb374 | 3604 | |
428870ff BB |
3605 | txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); |
3606 | if (txg == 0) { | |
9babb374 BB |
3607 | umem_free(packbuf, packsize); |
3608 | umem_free(bigbuf, bigsize); | |
3609 | for (j = 0; j < s; j++) { | |
3610 | if (i != 5) { | |
3611 | dmu_return_arcbuf(bigbuf_arcbufs[j]); | |
3612 | } else { | |
3613 | dmu_return_arcbuf( | |
3614 | bigbuf_arcbufs[2 * j]); | |
3615 | dmu_return_arcbuf( | |
3616 | bigbuf_arcbufs[2 * j + 1]); | |
3617 | } | |
3618 | } | |
3619 | umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *)); | |
3620 | dmu_buf_rele(bonus_db, FTAG); | |
3621 | return; | |
3622 | } | |
3623 | ||
9babb374 BB |
3624 | /* |
3625 | * 50% of the time don't read objects in the 1st iteration to | |
3626 | * test dmu_assign_arcbuf() for the case when there're no | |
3627 | * existing dbufs for the specified offsets. | |
3628 | */ | |
3629 | if (i != 0 || ztest_random(2) != 0) { | |
428870ff | 3630 | error = dmu_read(os, packobj, packoff, |
9babb374 BB |
3631 | packsize, packbuf, DMU_READ_PREFETCH); |
3632 | ASSERT3U(error, ==, 0); | |
428870ff | 3633 | error = dmu_read(os, bigobj, bigoff, bigsize, |
9babb374 BB |
3634 | bigbuf, DMU_READ_PREFETCH); |
3635 | ASSERT3U(error, ==, 0); | |
3636 | } | |
3637 | compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize, | |
428870ff | 3638 | n, chunksize, txg); |
9babb374 BB |
3639 | |
3640 | /* | |
3641 | * We've verified all the old bufwads, and made new ones. | |
3642 | * Now write them out. | |
3643 | */ | |
428870ff BB |
3644 | dmu_write(os, packobj, packoff, packsize, packbuf, tx); |
3645 | if (zopt_verbose >= 7) { | |
9babb374 BB |
3646 | (void) printf("writing offset %llx size %llx" |
3647 | " txg %llx\n", | |
3648 | (u_longlong_t)bigoff, | |
3649 | (u_longlong_t)bigsize, | |
3650 | (u_longlong_t)txg); | |
3651 | } | |
428870ff | 3652 | for (off = bigoff, j = 0; j < s; j++, off += chunksize) { |
9babb374 BB |
3653 | dmu_buf_t *dbt; |
3654 | if (i != 5) { | |
3655 | bcopy((caddr_t)bigbuf + (off - bigoff), | |
428870ff | 3656 | bigbuf_arcbufs[j]->b_data, chunksize); |
9babb374 BB |
3657 | } else { |
3658 | bcopy((caddr_t)bigbuf + (off - bigoff), | |
3659 | bigbuf_arcbufs[2 * j]->b_data, | |
428870ff | 3660 | chunksize / 2); |
9babb374 | 3661 | bcopy((caddr_t)bigbuf + (off - bigoff) + |
428870ff | 3662 | chunksize / 2, |
9babb374 | 3663 | bigbuf_arcbufs[2 * j + 1]->b_data, |
428870ff | 3664 | chunksize / 2); |
9babb374 BB |
3665 | } |
3666 | ||
3667 | if (i == 1) { | |
428870ff BB |
3668 | VERIFY(dmu_buf_hold(os, bigobj, off, |
3669 | FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0); | |
9babb374 BB |
3670 | } |
3671 | if (i != 5) { | |
3672 | dmu_assign_arcbuf(bonus_db, off, | |
3673 | bigbuf_arcbufs[j], tx); | |
3674 | } else { | |
3675 | dmu_assign_arcbuf(bonus_db, off, | |
3676 | bigbuf_arcbufs[2 * j], tx); | |
3677 | dmu_assign_arcbuf(bonus_db, | |
428870ff | 3678 | off + chunksize / 2, |
9babb374 BB |
3679 | bigbuf_arcbufs[2 * j + 1], tx); |
3680 | } | |
3681 | if (i == 1) { | |
3682 | dmu_buf_rele(dbt, FTAG); | |
3683 | } | |
3684 | } | |
3685 | dmu_tx_commit(tx); | |
3686 | ||
3687 | /* | |
3688 | * Sanity check the stuff we just wrote. | |
3689 | */ | |
3690 | { | |
3691 | void *packcheck = umem_alloc(packsize, UMEM_NOFAIL); | |
3692 | void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL); | |
3693 | ||
428870ff | 3694 | VERIFY(0 == dmu_read(os, packobj, packoff, |
9babb374 | 3695 | packsize, packcheck, DMU_READ_PREFETCH)); |
428870ff | 3696 | VERIFY(0 == dmu_read(os, bigobj, bigoff, |
9babb374 BB |
3697 | bigsize, bigcheck, DMU_READ_PREFETCH)); |
3698 | ||
3699 | ASSERT(bcmp(packbuf, packcheck, packsize) == 0); | |
3700 | ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0); | |
3701 | ||
3702 | umem_free(packcheck, packsize); | |
3703 | umem_free(bigcheck, bigsize); | |
3704 | } | |
3705 | if (i == 2) { | |
3706 | txg_wait_open(dmu_objset_pool(os), 0); | |
3707 | } else if (i == 3) { | |
3708 | txg_wait_synced(dmu_objset_pool(os), 0); | |
3709 | } | |
3710 | } | |
3711 | ||
3712 | dmu_buf_rele(bonus_db, FTAG); | |
3713 | umem_free(packbuf, packsize); | |
3714 | umem_free(bigbuf, bigsize); | |
3715 | umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *)); | |
3716 | } | |
3717 | ||
428870ff | 3718 | /* ARGSUSED */ |
34dc7c2f | 3719 | void |
428870ff | 3720 | ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id) |
34dc7c2f | 3721 | { |
428870ff BB |
3722 | ztest_od_t od[1]; |
3723 | uint64_t offset = (1ULL << (ztest_random(20) + 43)) + | |
3724 | (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT); | |
34dc7c2f BB |
3725 | |
3726 | /* | |
428870ff BB |
3727 | * Have multiple threads write to large offsets in an object |
3728 | * to verify that parallel writes to an object -- even to the | |
3729 | * same blocks within the object -- doesn't cause any trouble. | |
34dc7c2f | 3730 | */ |
428870ff | 3731 | ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0); |
9babb374 | 3732 | |
428870ff | 3733 | if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0) |
34dc7c2f | 3734 | return; |
34dc7c2f | 3735 | |
428870ff BB |
3736 | while (ztest_random(10) != 0) |
3737 | ztest_io(zd, od[0].od_object, offset); | |
3738 | } | |
34dc7c2f | 3739 | |
428870ff BB |
3740 | void |
3741 | ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id) | |
3742 | { | |
3743 | ztest_od_t od[1]; | |
3744 | uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) + | |
3745 | (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT); | |
3746 | uint64_t count = ztest_random(20) + 1; | |
3747 | uint64_t blocksize = ztest_random_blocksize(); | |
3748 | void *data; | |
34dc7c2f | 3749 | |
428870ff | 3750 | ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0); |
34dc7c2f | 3751 | |
428870ff | 3752 | if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0) |
34dc7c2f BB |
3753 | return; |
3754 | ||
428870ff | 3755 | if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0) |
34dc7c2f BB |
3756 | return; |
3757 | ||
428870ff | 3758 | ztest_prealloc(zd, od[0].od_object, offset, count * blocksize); |
34dc7c2f | 3759 | |
428870ff | 3760 | data = umem_zalloc(blocksize, UMEM_NOFAIL); |
34dc7c2f | 3761 | |
428870ff BB |
3762 | while (ztest_random(count) != 0) { |
3763 | uint64_t randoff = offset + (ztest_random(count) * blocksize); | |
3764 | if (ztest_write(zd, od[0].od_object, randoff, blocksize, | |
3765 | data) != 0) | |
3766 | break; | |
3767 | while (ztest_random(4) != 0) | |
3768 | ztest_io(zd, od[0].od_object, randoff); | |
9babb374 | 3769 | } |
34dc7c2f | 3770 | |
428870ff | 3771 | umem_free(data, blocksize); |
34dc7c2f BB |
3772 | } |
3773 | ||
3774 | /* | |
3775 | * Verify that zap_{create,destroy,add,remove,update} work as expected. | |
3776 | */ | |
3777 | #define ZTEST_ZAP_MIN_INTS 1 | |
3778 | #define ZTEST_ZAP_MAX_INTS 4 | |
3779 | #define ZTEST_ZAP_MAX_PROPS 1000 | |
3780 | ||
3781 | void | |
428870ff | 3782 | ztest_zap(ztest_ds_t *zd, uint64_t id) |
34dc7c2f | 3783 | { |
428870ff BB |
3784 | objset_t *os = zd->zd_os; |
3785 | ztest_od_t od[1]; | |
34dc7c2f BB |
3786 | uint64_t object; |
3787 | uint64_t txg, last_txg; | |
3788 | uint64_t value[ZTEST_ZAP_MAX_INTS]; | |
3789 | uint64_t zl_ints, zl_intsize, prop; | |
3790 | int i, ints; | |
3791 | dmu_tx_t *tx; | |
3792 | char propname[100], txgname[100]; | |
3793 | int error; | |
34dc7c2f BB |
3794 | char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" }; |
3795 | ||
428870ff | 3796 | ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0); |
34dc7c2f | 3797 | |
428870ff BB |
3798 | if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0) |
3799 | return; | |
34dc7c2f | 3800 | |
428870ff | 3801 | object = od[0].od_object; |
34dc7c2f | 3802 | |
428870ff BB |
3803 | /* |
3804 | * Generate a known hash collision, and verify that | |
3805 | * we can lookup and remove both entries. | |
3806 | */ | |
3807 | tx = dmu_tx_create(os); | |
3808 | dmu_tx_hold_zap(tx, object, B_TRUE, NULL); | |
3809 | txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); | |
3810 | if (txg == 0) | |
3811 | return; | |
3812 | for (i = 0; i < 2; i++) { | |
3813 | value[i] = i; | |
3814 | VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t), | |
3815 | 1, &value[i], tx)); | |
3816 | } | |
3817 | for (i = 0; i < 2; i++) { | |
3818 | VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i], | |
3819 | sizeof (uint64_t), 1, &value[i], tx)); | |
3820 | VERIFY3U(0, ==, | |
3821 | zap_length(os, object, hc[i], &zl_intsize, &zl_ints)); | |
3822 | ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); | |
3823 | ASSERT3U(zl_ints, ==, 1); | |
3824 | } | |
3825 | for (i = 0; i < 2; i++) { | |
3826 | VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx)); | |
34dc7c2f | 3827 | } |
428870ff | 3828 | dmu_tx_commit(tx); |
34dc7c2f | 3829 | |
428870ff BB |
3830 | /* |
3831 | * Generate a buch of random entries. | |
3832 | */ | |
34dc7c2f BB |
3833 | ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS); |
3834 | ||
3835 | prop = ztest_random(ZTEST_ZAP_MAX_PROPS); | |
3836 | (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop); | |
3837 | (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop); | |
3838 | bzero(value, sizeof (value)); | |
3839 | last_txg = 0; | |
3840 | ||
3841 | /* | |
3842 | * If these zap entries already exist, validate their contents. | |
3843 | */ | |
3844 | error = zap_length(os, object, txgname, &zl_intsize, &zl_ints); | |
3845 | if (error == 0) { | |
3846 | ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); | |
3847 | ASSERT3U(zl_ints, ==, 1); | |
3848 | ||
3849 | VERIFY(zap_lookup(os, object, txgname, zl_intsize, | |
3850 | zl_ints, &last_txg) == 0); | |
3851 | ||
3852 | VERIFY(zap_length(os, object, propname, &zl_intsize, | |
3853 | &zl_ints) == 0); | |
3854 | ||
3855 | ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); | |
3856 | ASSERT3U(zl_ints, ==, ints); | |
3857 | ||
3858 | VERIFY(zap_lookup(os, object, propname, zl_intsize, | |
3859 | zl_ints, value) == 0); | |
3860 | ||
3861 | for (i = 0; i < ints; i++) { | |
3862 | ASSERT3U(value[i], ==, last_txg + object + i); | |
3863 | } | |
3864 | } else { | |
3865 | ASSERT3U(error, ==, ENOENT); | |
3866 | } | |
3867 | ||
3868 | /* | |
3869 | * Atomically update two entries in our zap object. | |
3870 | * The first is named txg_%llu, and contains the txg | |
3871 | * in which the property was last updated. The second | |
3872 | * is named prop_%llu, and the nth element of its value | |
3873 | * should be txg + object + n. | |
3874 | */ | |
3875 | tx = dmu_tx_create(os); | |
428870ff BB |
3876 | dmu_tx_hold_zap(tx, object, B_TRUE, NULL); |
3877 | txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); | |
3878 | if (txg == 0) | |
34dc7c2f | 3879 | return; |
34dc7c2f BB |
3880 | |
3881 | if (last_txg > txg) | |
3882 | fatal(0, "zap future leak: old %llu new %llu", last_txg, txg); | |
3883 | ||
3884 | for (i = 0; i < ints; i++) | |
3885 | value[i] = txg + object + i; | |
3886 | ||
428870ff BB |
3887 | VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t), |
3888 | 1, &txg, tx)); | |
3889 | VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t), | |
3890 | ints, value, tx)); | |
34dc7c2f BB |
3891 | |
3892 | dmu_tx_commit(tx); | |
3893 | ||
3894 | /* | |
3895 | * Remove a random pair of entries. | |
3896 | */ | |
3897 | prop = ztest_random(ZTEST_ZAP_MAX_PROPS); | |
3898 | (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop); | |
3899 | (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop); | |
3900 | ||
3901 | error = zap_length(os, object, txgname, &zl_intsize, &zl_ints); | |
3902 | ||
3903 | if (error == ENOENT) | |
3904 | return; | |
3905 | ||
3906 | ASSERT3U(error, ==, 0); | |
3907 | ||
3908 | tx = dmu_tx_create(os); | |
428870ff BB |
3909 | dmu_tx_hold_zap(tx, object, B_TRUE, NULL); |
3910 | txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); | |
3911 | if (txg == 0) | |
34dc7c2f | 3912 | return; |
428870ff BB |
3913 | VERIFY3U(0, ==, zap_remove(os, object, txgname, tx)); |
3914 | VERIFY3U(0, ==, zap_remove(os, object, propname, tx)); | |
3915 | dmu_tx_commit(tx); | |
3916 | } | |
34dc7c2f | 3917 | |
428870ff BB |
3918 | /* |
3919 | * Testcase to test the upgrading of a microzap to fatzap. | |
3920 | */ | |
3921 | void | |
3922 | ztest_fzap(ztest_ds_t *zd, uint64_t id) | |
3923 | { | |
3924 | objset_t *os = zd->zd_os; | |
3925 | ztest_od_t od[1]; | |
3926 | uint64_t object, txg; | |
d6320ddb | 3927 | int i; |
34dc7c2f | 3928 | |
428870ff BB |
3929 | ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0); |
3930 | ||
3931 | if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0) | |
3932 | return; | |
3933 | ||
3934 | object = od[0].od_object; | |
34dc7c2f BB |
3935 | |
3936 | /* | |
428870ff BB |
3937 | * Add entries to this ZAP and make sure it spills over |
3938 | * and gets upgraded to a fatzap. Also, since we are adding | |
3939 | * 2050 entries we should see ptrtbl growth and leaf-block split. | |
34dc7c2f | 3940 | */ |
d6320ddb | 3941 | for (i = 0; i < 2050; i++) { |
428870ff BB |
3942 | char name[MAXNAMELEN]; |
3943 | uint64_t value = i; | |
3944 | dmu_tx_t *tx; | |
3945 | int error; | |
34dc7c2f | 3946 | |
428870ff | 3947 | (void) snprintf(name, sizeof (name), "fzap-%llu-%llu", |
b8864a23 | 3948 | (u_longlong_t)id, (u_longlong_t)value); |
428870ff BB |
3949 | |
3950 | tx = dmu_tx_create(os); | |
3951 | dmu_tx_hold_zap(tx, object, B_TRUE, name); | |
3952 | txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); | |
3953 | if (txg == 0) | |
3954 | return; | |
3955 | error = zap_add(os, object, name, sizeof (uint64_t), 1, | |
3956 | &value, tx); | |
3957 | ASSERT(error == 0 || error == EEXIST); | |
3958 | dmu_tx_commit(tx); | |
34dc7c2f | 3959 | } |
34dc7c2f BB |
3960 | } |
3961 | ||
428870ff | 3962 | /* ARGSUSED */ |
34dc7c2f | 3963 | void |
428870ff | 3964 | ztest_zap_parallel(ztest_ds_t *zd, uint64_t id) |
34dc7c2f | 3965 | { |
428870ff BB |
3966 | objset_t *os = zd->zd_os; |
3967 | ztest_od_t od[1]; | |
34dc7c2f BB |
3968 | uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc; |
3969 | dmu_tx_t *tx; | |
3970 | int i, namelen, error; | |
428870ff | 3971 | int micro = ztest_random(2); |
34dc7c2f BB |
3972 | char name[20], string_value[20]; |
3973 | void *data; | |
3974 | ||
428870ff BB |
3975 | ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0); |
3976 | ||
3977 | if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0) | |
3978 | return; | |
3979 | ||
3980 | object = od[0].od_object; | |
3981 | ||
34dc7c2f BB |
3982 | /* |
3983 | * Generate a random name of the form 'xxx.....' where each | |
3984 | * x is a random printable character and the dots are dots. | |
3985 | * There are 94 such characters, and the name length goes from | |
3986 | * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names. | |
3987 | */ | |
3988 | namelen = ztest_random(sizeof (name) - 5) + 5 + 1; | |
3989 | ||
3990 | for (i = 0; i < 3; i++) | |
3991 | name[i] = '!' + ztest_random('~' - '!' + 1); | |
3992 | for (; i < namelen - 1; i++) | |
3993 | name[i] = '.'; | |
3994 | name[i] = '\0'; | |
3995 | ||
428870ff | 3996 | if ((namelen & 1) || micro) { |
34dc7c2f BB |
3997 | wsize = sizeof (txg); |
3998 | wc = 1; | |
3999 | data = &txg; | |
4000 | } else { | |
4001 | wsize = 1; | |
4002 | wc = namelen; | |
4003 | data = string_value; | |
4004 | } | |
4005 | ||
4006 | count = -1ULL; | |
4007 | VERIFY(zap_count(os, object, &count) == 0); | |
4008 | ASSERT(count != -1ULL); | |
4009 | ||
4010 | /* | |
4011 | * Select an operation: length, lookup, add, update, remove. | |
4012 | */ | |
4013 | i = ztest_random(5); | |
4014 | ||
4015 | if (i >= 2) { | |
4016 | tx = dmu_tx_create(os); | |
428870ff BB |
4017 | dmu_tx_hold_zap(tx, object, B_TRUE, NULL); |
4018 | txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); | |
4019 | if (txg == 0) | |
34dc7c2f | 4020 | return; |
34dc7c2f BB |
4021 | bcopy(name, string_value, namelen); |
4022 | } else { | |
4023 | tx = NULL; | |
4024 | txg = 0; | |
4025 | bzero(string_value, namelen); | |
4026 | } | |
4027 | ||
4028 | switch (i) { | |
4029 | ||
4030 | case 0: | |
4031 | error = zap_length(os, object, name, &zl_wsize, &zl_wc); | |
4032 | if (error == 0) { | |
4033 | ASSERT3U(wsize, ==, zl_wsize); | |
4034 | ASSERT3U(wc, ==, zl_wc); | |
4035 | } else { | |
4036 | ASSERT3U(error, ==, ENOENT); | |
4037 | } | |
4038 | break; | |
4039 | ||
4040 | case 1: | |
4041 | error = zap_lookup(os, object, name, wsize, wc, data); | |
4042 | if (error == 0) { | |
4043 | if (data == string_value && | |
4044 | bcmp(name, data, namelen) != 0) | |
4045 | fatal(0, "name '%s' != val '%s' len %d", | |
4046 | name, data, namelen); | |
4047 | } else { | |
4048 | ASSERT3U(error, ==, ENOENT); | |
4049 | } | |
4050 | break; | |
4051 | ||
4052 | case 2: | |
4053 | error = zap_add(os, object, name, wsize, wc, data, tx); | |
4054 | ASSERT(error == 0 || error == EEXIST); | |
4055 | break; | |
4056 | ||
4057 | case 3: | |
4058 | VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0); | |
4059 | break; | |
4060 | ||
4061 | case 4: | |
4062 | error = zap_remove(os, object, name, tx); | |
4063 | ASSERT(error == 0 || error == ENOENT); | |
4064 | break; | |
4065 | } | |
4066 | ||
4067 | if (tx != NULL) | |
4068 | dmu_tx_commit(tx); | |
4069 | } | |
4070 | ||
428870ff BB |
4071 | /* |
4072 | * Commit callback data. | |
4073 | */ | |
4074 | typedef struct ztest_cb_data { | |
4075 | list_node_t zcd_node; | |
4076 | uint64_t zcd_txg; | |
4077 | int zcd_expected_err; | |
4078 | boolean_t zcd_added; | |
4079 | boolean_t zcd_called; | |
4080 | spa_t *zcd_spa; | |
4081 | } ztest_cb_data_t; | |
4082 | ||
4083 | /* This is the actual commit callback function */ | |
4084 | static void | |
4085 | ztest_commit_callback(void *arg, int error) | |
4086 | { | |
4087 | ztest_cb_data_t *data = arg; | |
4088 | uint64_t synced_txg; | |
4089 | ||
4090 | VERIFY(data != NULL); | |
4091 | VERIFY3S(data->zcd_expected_err, ==, error); | |
4092 | VERIFY(!data->zcd_called); | |
4093 | ||
4094 | synced_txg = spa_last_synced_txg(data->zcd_spa); | |
4095 | if (data->zcd_txg > synced_txg) | |
4096 | fatal(0, "commit callback of txg %" PRIu64 " called prematurely" | |
4097 | ", last synced txg = %" PRIu64 "\n", data->zcd_txg, | |
4098 | synced_txg); | |
4099 | ||
4100 | data->zcd_called = B_TRUE; | |
4101 | ||
4102 | if (error == ECANCELED) { | |
4103 | ASSERT3U(data->zcd_txg, ==, 0); | |
4104 | ASSERT(!data->zcd_added); | |
4105 | ||
4106 | /* | |
4107 | * The private callback data should be destroyed here, but | |
4108 | * since we are going to check the zcd_called field after | |
4109 | * dmu_tx_abort(), we will destroy it there. | |
4110 | */ | |
4111 | return; | |
4112 | } | |
4113 | ||
090ff092 | 4114 | ASSERT(data->zcd_added); |
428870ff BB |
4115 | ASSERT3U(data->zcd_txg, !=, 0); |
4116 | ||
1e33ac1e | 4117 | (void) mutex_enter(&zcl.zcl_callbacks_lock); |
090ff092 RC |
4118 | |
4119 | /* See if this cb was called more quickly */ | |
4120 | if ((synced_txg - data->zcd_txg) < zc_min_txg_delay) | |
4121 | zc_min_txg_delay = synced_txg - data->zcd_txg; | |
4122 | ||
4123 | /* Remove our callback from the list */ | |
428870ff | 4124 | list_remove(&zcl.zcl_callbacks, data); |
090ff092 | 4125 | |
1e33ac1e | 4126 | (void) mutex_exit(&zcl.zcl_callbacks_lock); |
428870ff | 4127 | |
428870ff BB |
4128 | umem_free(data, sizeof (ztest_cb_data_t)); |
4129 | } | |
4130 | ||
4131 | /* Allocate and initialize callback data structure */ | |
4132 | static ztest_cb_data_t * | |
4133 | ztest_create_cb_data(objset_t *os, uint64_t txg) | |
4134 | { | |
4135 | ztest_cb_data_t *cb_data; | |
4136 | ||
4137 | cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL); | |
4138 | ||
4139 | cb_data->zcd_txg = txg; | |
4140 | cb_data->zcd_spa = dmu_objset_spa(os); | |
1e33ac1e | 4141 | list_link_init(&cb_data->zcd_node); |
428870ff BB |
4142 | |
4143 | return (cb_data); | |
4144 | } | |
4145 | ||
428870ff BB |
4146 | /* |
4147 | * Commit callback test. | |
4148 | */ | |
34dc7c2f | 4149 | void |
428870ff BB |
4150 | ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id) |
4151 | { | |
4152 | objset_t *os = zd->zd_os; | |
4153 | ztest_od_t od[1]; | |
4154 | dmu_tx_t *tx; | |
4155 | ztest_cb_data_t *cb_data[3], *tmp_cb; | |
4156 | uint64_t old_txg, txg; | |
090ff092 | 4157 | int i, error = 0; |
428870ff BB |
4158 | |
4159 | ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0); | |
4160 | ||
4161 | if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0) | |
4162 | return; | |
4163 | ||
4164 | tx = dmu_tx_create(os); | |
4165 | ||
4166 | cb_data[0] = ztest_create_cb_data(os, 0); | |
4167 | dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]); | |
4168 | ||
4169 | dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t)); | |
4170 | ||
4171 | /* Every once in a while, abort the transaction on purpose */ | |
4172 | if (ztest_random(100) == 0) | |
4173 | error = -1; | |
4174 | ||
4175 | if (!error) | |
4176 | error = dmu_tx_assign(tx, TXG_NOWAIT); | |
4177 | ||
4178 | txg = error ? 0 : dmu_tx_get_txg(tx); | |
4179 | ||
4180 | cb_data[0]->zcd_txg = txg; | |
4181 | cb_data[1] = ztest_create_cb_data(os, txg); | |
4182 | dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]); | |
4183 | ||
4184 | if (error) { | |
4185 | /* | |
4186 | * It's not a strict requirement to call the registered | |
4187 | * callbacks from inside dmu_tx_abort(), but that's what | |
4188 | * it's supposed to happen in the current implementation | |
4189 | * so we will check for that. | |
4190 | */ | |
4191 | for (i = 0; i < 2; i++) { | |
4192 | cb_data[i]->zcd_expected_err = ECANCELED; | |
4193 | VERIFY(!cb_data[i]->zcd_called); | |
4194 | } | |
4195 | ||
4196 | dmu_tx_abort(tx); | |
4197 | ||
4198 | for (i = 0; i < 2; i++) { | |
4199 | VERIFY(cb_data[i]->zcd_called); | |
4200 | umem_free(cb_data[i], sizeof (ztest_cb_data_t)); | |
4201 | } | |
4202 | ||
4203 | return; | |
4204 | } | |
4205 | ||
4206 | cb_data[2] = ztest_create_cb_data(os, txg); | |
4207 | dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]); | |
4208 | ||
4209 | /* | |
4210 | * Read existing data to make sure there isn't a future leak. | |
4211 | */ | |
4212 | VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t), | |
4213 | &old_txg, DMU_READ_PREFETCH)); | |
4214 | ||
4215 | if (old_txg > txg) | |
4216 | fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64, | |
4217 | old_txg, txg); | |
4218 | ||
4219 | dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx); | |
4220 | ||
1e33ac1e | 4221 | (void) mutex_enter(&zcl.zcl_callbacks_lock); |
428870ff BB |
4222 | |
4223 | /* | |
4224 | * Since commit callbacks don't have any ordering requirement and since | |
4225 | * it is theoretically possible for a commit callback to be called | |
4226 | * after an arbitrary amount of time has elapsed since its txg has been | |
4227 | * synced, it is difficult to reliably determine whether a commit | |
4228 | * callback hasn't been called due to high load or due to a flawed | |
4229 | * implementation. | |
4230 | * | |
4231 | * In practice, we will assume that if after a certain number of txgs a | |
4232 | * commit callback hasn't been called, then most likely there's an | |
4233 | * implementation bug.. | |
4234 | */ | |
4235 | tmp_cb = list_head(&zcl.zcl_callbacks); | |
4236 | if (tmp_cb != NULL && | |
090ff092 | 4237 | tmp_cb->zcd_txg + ZTEST_COMMIT_CB_THRESH < txg) { |
428870ff BB |
4238 | fatal(0, "Commit callback threshold exceeded, oldest txg: %" |
4239 | PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg); | |
4240 | } | |
4241 | ||
4242 | /* | |
4243 | * Let's find the place to insert our callbacks. | |
4244 | * | |
4245 | * Even though the list is ordered by txg, it is possible for the | |
4246 | * insertion point to not be the end because our txg may already be | |
4247 | * quiescing at this point and other callbacks in the open txg | |
4248 | * (from other objsets) may have sneaked in. | |
4249 | */ | |
4250 | tmp_cb = list_tail(&zcl.zcl_callbacks); | |
4251 | while (tmp_cb != NULL && tmp_cb->zcd_txg > txg) | |
4252 | tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb); | |
4253 | ||
4254 | /* Add the 3 callbacks to the list */ | |
4255 | for (i = 0; i < 3; i++) { | |
4256 | if (tmp_cb == NULL) | |
4257 | list_insert_head(&zcl.zcl_callbacks, cb_data[i]); | |
4258 | else | |
4259 | list_insert_after(&zcl.zcl_callbacks, tmp_cb, | |
4260 | cb_data[i]); | |
4261 | ||
4262 | cb_data[i]->zcd_added = B_TRUE; | |
4263 | VERIFY(!cb_data[i]->zcd_called); | |
4264 | ||
4265 | tmp_cb = cb_data[i]; | |
4266 | } | |
4267 | ||
090ff092 RC |
4268 | zc_cb_counter += 3; |
4269 | ||
1e33ac1e | 4270 | (void) mutex_exit(&zcl.zcl_callbacks_lock); |
428870ff BB |
4271 | |
4272 | dmu_tx_commit(tx); | |
4273 | } | |
4274 | ||
4275 | /* ARGSUSED */ | |
4276 | void | |
4277 | ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id) | |
4278 | { | |
4279 | zfs_prop_t proplist[] = { | |
4280 | ZFS_PROP_CHECKSUM, | |
4281 | ZFS_PROP_COMPRESSION, | |
4282 | ZFS_PROP_COPIES, | |
4283 | ZFS_PROP_DEDUP | |
4284 | }; | |
4285 | ztest_shared_t *zs = ztest_shared; | |
d6320ddb | 4286 | int p; |
428870ff | 4287 | |
1e33ac1e | 4288 | (void) rw_enter(&zs->zs_name_lock, RW_READER); |
428870ff | 4289 | |
d6320ddb | 4290 | for (p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++) |
428870ff BB |
4291 | (void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p], |
4292 | ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2)); | |
4293 | ||
1e33ac1e | 4294 | (void) rw_exit(&zs->zs_name_lock); |
428870ff BB |
4295 | } |
4296 | ||
4297 | /* ARGSUSED */ | |
4298 | void | |
4299 | ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id) | |
4300 | { | |
4301 | ztest_shared_t *zs = ztest_shared; | |
4302 | nvlist_t *props = NULL; | |
4303 | ||
1e33ac1e | 4304 | (void) rw_enter(&zs->zs_name_lock, RW_READER); |
428870ff BB |
4305 | |
4306 | (void) ztest_spa_prop_set_uint64(zs, ZPOOL_PROP_DEDUPDITTO, | |
4307 | ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN)); | |
4308 | ||
4309 | VERIFY3U(spa_prop_get(zs->zs_spa, &props), ==, 0); | |
4310 | ||
4311 | if (zopt_verbose >= 6) | |
4312 | dump_nvlist(props, 4); | |
4313 | ||
4314 | nvlist_free(props); | |
4315 | ||
1e33ac1e | 4316 | (void) rw_exit(&zs->zs_name_lock); |
428870ff BB |
4317 | } |
4318 | ||
4319 | /* | |
4320 | * Test snapshot hold/release and deferred destroy. | |
4321 | */ | |
4322 | void | |
4323 | ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id) | |
34dc7c2f | 4324 | { |
34dc7c2f | 4325 | int error; |
428870ff BB |
4326 | objset_t *os = zd->zd_os; |
4327 | objset_t *origin; | |
4328 | char snapname[100]; | |
4329 | char fullname[100]; | |
4330 | char clonename[100]; | |
4331 | char tag[100]; | |
4332 | char osname[MAXNAMELEN]; | |
34dc7c2f | 4333 | |
1e33ac1e | 4334 | (void) rw_enter(&ztest_shared->zs_name_lock, RW_READER); |
34dc7c2f BB |
4335 | |
4336 | dmu_objset_name(os, osname); | |
4337 | ||
1e33ac1e | 4338 | (void) snprintf(snapname, 100, "sh1_%llu", (u_longlong_t)id); |
428870ff | 4339 | (void) snprintf(fullname, 100, "%s@%s", osname, snapname); |
1e33ac1e BB |
4340 | (void) snprintf(clonename, 100, "%s/ch1_%llu",osname,(u_longlong_t)id); |
4341 | (void) snprintf(tag, 100, "tag_%llu", (u_longlong_t)id); | |
428870ff BB |
4342 | |
4343 | /* | |
4344 | * Clean up from any previous run. | |
4345 | */ | |
4346 | (void) dmu_objset_destroy(clonename, B_FALSE); | |
4347 | (void) dsl_dataset_user_release(osname, snapname, tag, B_FALSE); | |
4348 | (void) dmu_objset_destroy(fullname, B_FALSE); | |
4349 | ||
4350 | /* | |
4351 | * Create snapshot, clone it, mark snap for deferred destroy, | |
4352 | * destroy clone, verify snap was also destroyed. | |
4353 | */ | |
572e2857 BB |
4354 | error = dmu_objset_snapshot(osname, snapname, NULL, NULL, FALSE, |
4355 | FALSE, -1); | |
428870ff BB |
4356 | if (error) { |
4357 | if (error == ENOSPC) { | |
4358 | ztest_record_enospc("dmu_objset_snapshot"); | |
4359 | goto out; | |
34dc7c2f | 4360 | } |
428870ff BB |
4361 | fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error); |
4362 | } | |
34dc7c2f | 4363 | |
428870ff BB |
4364 | error = dmu_objset_hold(fullname, FTAG, &origin); |
4365 | if (error) | |
4366 | fatal(0, "dmu_objset_hold(%s) = %d", fullname, error); | |
34dc7c2f | 4367 | |
428870ff BB |
4368 | error = dmu_objset_clone(clonename, dmu_objset_ds(origin), 0); |
4369 | dmu_objset_rele(origin, FTAG); | |
4370 | if (error) { | |
34dc7c2f | 4371 | if (error == ENOSPC) { |
428870ff BB |
4372 | ztest_record_enospc("dmu_objset_clone"); |
4373 | goto out; | |
34dc7c2f | 4374 | } |
428870ff BB |
4375 | fatal(0, "dmu_objset_clone(%s) = %d", clonename, error); |
4376 | } | |
34dc7c2f | 4377 | |
428870ff BB |
4378 | error = dmu_objset_destroy(fullname, B_TRUE); |
4379 | if (error) { | |
4380 | fatal(0, "dmu_objset_destroy(%s, B_TRUE) = %d", | |
4381 | fullname, error); | |
4382 | } | |
34dc7c2f | 4383 | |
428870ff BB |
4384 | error = dmu_objset_destroy(clonename, B_FALSE); |
4385 | if (error) | |
4386 | fatal(0, "dmu_objset_destroy(%s) = %d", clonename, error); | |
34dc7c2f | 4387 | |
428870ff BB |
4388 | error = dmu_objset_hold(fullname, FTAG, &origin); |
4389 | if (error != ENOENT) | |
4390 | fatal(0, "dmu_objset_hold(%s) = %d", fullname, error); | |
34dc7c2f | 4391 | |
428870ff BB |
4392 | /* |
4393 | * Create snapshot, add temporary hold, verify that we can't | |
4394 | * destroy a held snapshot, mark for deferred destroy, | |
4395 | * release hold, verify snapshot was destroyed. | |
4396 | */ | |
572e2857 BB |
4397 | error = dmu_objset_snapshot(osname, snapname, NULL, NULL, FALSE, |
4398 | FALSE, -1); | |
428870ff BB |
4399 | if (error) { |
4400 | if (error == ENOSPC) { | |
4401 | ztest_record_enospc("dmu_objset_snapshot"); | |
4402 | goto out; | |
34dc7c2f | 4403 | } |
428870ff BB |
4404 | fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error); |
4405 | } | |
4406 | ||
572e2857 BB |
4407 | error = dsl_dataset_user_hold(osname, snapname, tag, B_FALSE, |
4408 | B_TRUE, -1); | |
428870ff BB |
4409 | if (error) |
4410 | fatal(0, "dsl_dataset_user_hold(%s)", fullname, tag); | |
4411 | ||
4412 | error = dmu_objset_destroy(fullname, B_FALSE); | |
4413 | if (error != EBUSY) { | |
4414 | fatal(0, "dmu_objset_destroy(%s, B_FALSE) = %d", | |
4415 | fullname, error); | |
4416 | } | |
4417 | ||
4418 | error = dmu_objset_destroy(fullname, B_TRUE); | |
4419 | if (error) { | |
4420 | fatal(0, "dmu_objset_destroy(%s, B_TRUE) = %d", | |
4421 | fullname, error); | |
34dc7c2f BB |
4422 | } |
4423 | ||
428870ff BB |
4424 | error = dsl_dataset_user_release(osname, snapname, tag, B_FALSE); |
4425 | if (error) | |
4426 | fatal(0, "dsl_dataset_user_release(%s)", fullname, tag); | |
4427 | ||
4428 | VERIFY(dmu_objset_hold(fullname, FTAG, &origin) == ENOENT); | |
4429 | ||
4430 | out: | |
1e33ac1e | 4431 | (void) rw_exit(&ztest_shared->zs_name_lock); |
34dc7c2f BB |
4432 | } |
4433 | ||
34dc7c2f BB |
4434 | /* |
4435 | * Inject random faults into the on-disk data. | |
4436 | */ | |
428870ff | 4437 | /* ARGSUSED */ |
34dc7c2f | 4438 | void |
428870ff | 4439 | ztest_fault_inject(ztest_ds_t *zd, uint64_t id) |
34dc7c2f | 4440 | { |
428870ff BB |
4441 | ztest_shared_t *zs = ztest_shared; |
4442 | spa_t *spa = zs->zs_spa; | |
34dc7c2f BB |
4443 | int fd; |
4444 | uint64_t offset; | |
428870ff | 4445 | uint64_t leaves; |
c5b3a7bb | 4446 | uint64_t bad = 0x1990c0ffeedecadeull; |
34dc7c2f BB |
4447 | uint64_t top, leaf; |
4448 | char path0[MAXPATHLEN]; | |
4449 | char pathrand[MAXPATHLEN]; | |
4450 | size_t fsize; | |
34dc7c2f BB |
4451 | int bshift = SPA_MAXBLOCKSHIFT + 2; /* don't scrog all labels */ |
4452 | int iters = 1000; | |
428870ff BB |
4453 | int maxfaults; |
4454 | int mirror_save; | |
b128c09f | 4455 | vdev_t *vd0 = NULL; |
34dc7c2f | 4456 | uint64_t guid0 = 0; |
428870ff BB |
4457 | boolean_t islog = B_FALSE; |
4458 | ||
1e33ac1e | 4459 | mutex_enter(&zs->zs_vdev_lock); |
428870ff BB |
4460 | maxfaults = MAXFAULTS(); |
4461 | leaves = MAX(zs->zs_mirrors, 1) * zopt_raidz; | |
4462 | mirror_save = zs->zs_mirrors; | |
1e33ac1e | 4463 | mutex_exit(&zs->zs_vdev_lock); |
34dc7c2f | 4464 | |
b128c09f | 4465 | ASSERT(leaves >= 1); |
34dc7c2f BB |
4466 | |
4467 | /* | |
b128c09f | 4468 | * We need SCL_STATE here because we're going to look at vd0->vdev_tsd. |
34dc7c2f | 4469 | */ |
b128c09f | 4470 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); |
34dc7c2f | 4471 | |
b128c09f BB |
4472 | if (ztest_random(2) == 0) { |
4473 | /* | |
428870ff | 4474 | * Inject errors on a normal data device or slog device. |
b128c09f | 4475 | */ |
428870ff BB |
4476 | top = ztest_random_vdev_top(spa, B_TRUE); |
4477 | leaf = ztest_random(leaves) + zs->zs_splits; | |
34dc7c2f | 4478 | |
b128c09f BB |
4479 | /* |
4480 | * Generate paths to the first leaf in this top-level vdev, | |
4481 | * and to the random leaf we selected. We'll induce transient | |
4482 | * write failures and random online/offline activity on leaf 0, | |
4483 | * and we'll write random garbage to the randomly chosen leaf. | |
4484 | */ | |
4485 | (void) snprintf(path0, sizeof (path0), ztest_dev_template, | |
428870ff | 4486 | zopt_dir, zopt_pool, top * leaves + zs->zs_splits); |
b128c09f BB |
4487 | (void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template, |
4488 | zopt_dir, zopt_pool, top * leaves + leaf); | |
34dc7c2f | 4489 | |
b128c09f | 4490 | vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0); |
428870ff BB |
4491 | if (vd0 != NULL && vd0->vdev_top->vdev_islog) |
4492 | islog = B_TRUE; | |
4493 | ||
b128c09f BB |
4494 | if (vd0 != NULL && maxfaults != 1) { |
4495 | /* | |
4496 | * Make vd0 explicitly claim to be unreadable, | |
4497 | * or unwriteable, or reach behind its back | |
4498 | * and close the underlying fd. We can do this if | |
4499 | * maxfaults == 0 because we'll fail and reexecute, | |
4500 | * and we can do it if maxfaults >= 2 because we'll | |
4501 | * have enough redundancy. If maxfaults == 1, the | |
4502 | * combination of this with injection of random data | |
4503 | * corruption below exceeds the pool's fault tolerance. | |
4504 | */ | |
4505 | vdev_file_t *vf = vd0->vdev_tsd; | |
4506 | ||
4507 | if (vf != NULL && ztest_random(3) == 0) { | |
4508 | (void) close(vf->vf_vnode->v_fd); | |
4509 | vf->vf_vnode->v_fd = -1; | |
4510 | } else if (ztest_random(2) == 0) { | |
4511 | vd0->vdev_cant_read = B_TRUE; | |
4512 | } else { | |
4513 | vd0->vdev_cant_write = B_TRUE; | |
4514 | } | |
4515 | guid0 = vd0->vdev_guid; | |
4516 | } | |
4517 | } else { | |
4518 | /* | |
4519 | * Inject errors on an l2cache device. | |
4520 | */ | |
4521 | spa_aux_vdev_t *sav = &spa->spa_l2cache; | |
34dc7c2f | 4522 | |
b128c09f BB |
4523 | if (sav->sav_count == 0) { |
4524 | spa_config_exit(spa, SCL_STATE, FTAG); | |
4525 | return; | |
4526 | } | |
4527 | vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)]; | |
34dc7c2f | 4528 | guid0 = vd0->vdev_guid; |
b128c09f BB |
4529 | (void) strcpy(path0, vd0->vdev_path); |
4530 | (void) strcpy(pathrand, vd0->vdev_path); | |
4531 | ||
4532 | leaf = 0; | |
4533 | leaves = 1; | |
4534 | maxfaults = INT_MAX; /* no limit on cache devices */ | |
34dc7c2f BB |
4535 | } |
4536 | ||
b128c09f BB |
4537 | spa_config_exit(spa, SCL_STATE, FTAG); |
4538 | ||
34dc7c2f | 4539 | /* |
428870ff BB |
4540 | * If we can tolerate two or more faults, or we're dealing |
4541 | * with a slog, randomly online/offline vd0. | |
34dc7c2f | 4542 | */ |
428870ff | 4543 | if ((maxfaults >= 2 || islog) && guid0 != 0) { |
fb5f0bc8 BB |
4544 | if (ztest_random(10) < 6) { |
4545 | int flags = (ztest_random(2) == 0 ? | |
4546 | ZFS_OFFLINE_TEMPORARY : 0); | |
428870ff BB |
4547 | |
4548 | /* | |
4549 | * We have to grab the zs_name_lock as writer to | |
4550 | * prevent a race between offlining a slog and | |
4551 | * destroying a dataset. Offlining the slog will | |
4552 | * grab a reference on the dataset which may cause | |
4553 | * dmu_objset_destroy() to fail with EBUSY thus | |
4554 | * leaving the dataset in an inconsistent state. | |
4555 | */ | |
4556 | if (islog) | |
1e33ac1e BB |
4557 | (void) rw_enter(&ztest_shared->zs_name_lock, |
4558 | RW_WRITER); | |
428870ff | 4559 | |
fb5f0bc8 | 4560 | VERIFY(vdev_offline(spa, guid0, flags) != EBUSY); |
428870ff BB |
4561 | |
4562 | if (islog) | |
1e33ac1e | 4563 | (void) rw_exit(&ztest_shared->zs_name_lock); |
fb5f0bc8 BB |
4564 | } else { |
4565 | (void) vdev_online(spa, guid0, 0, NULL); | |
4566 | } | |
34dc7c2f BB |
4567 | } |
4568 | ||
428870ff BB |
4569 | if (maxfaults == 0) |
4570 | return; | |
4571 | ||
34dc7c2f BB |
4572 | /* |
4573 | * We have at least single-fault tolerance, so inject data corruption. | |
4574 | */ | |
4575 | fd = open(pathrand, O_RDWR); | |
4576 | ||
4577 | if (fd == -1) /* we hit a gap in the device namespace */ | |
4578 | return; | |
4579 | ||
4580 | fsize = lseek(fd, 0, SEEK_END); | |
4581 | ||
4582 | while (--iters != 0) { | |
4583 | offset = ztest_random(fsize / (leaves << bshift)) * | |
4584 | (leaves << bshift) + (leaf << bshift) + | |
4585 | (ztest_random(1ULL << (bshift - 1)) & -8ULL); | |
4586 | ||
4587 | if (offset >= fsize) | |
4588 | continue; | |
4589 | ||
1e33ac1e | 4590 | mutex_enter(&zs->zs_vdev_lock); |
428870ff | 4591 | if (mirror_save != zs->zs_mirrors) { |
1e33ac1e | 4592 | mutex_exit(&zs->zs_vdev_lock); |
428870ff BB |
4593 | (void) close(fd); |
4594 | return; | |
4595 | } | |
34dc7c2f BB |
4596 | |
4597 | if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad)) | |
4598 | fatal(1, "can't inject bad word at 0x%llx in %s", | |
4599 | offset, pathrand); | |
428870ff | 4600 | |
1e33ac1e | 4601 | mutex_exit(&zs->zs_vdev_lock); |
428870ff BB |
4602 | |
4603 | if (zopt_verbose >= 7) | |
4604 | (void) printf("injected bad word into %s," | |
4605 | " offset 0x%llx\n", pathrand, (u_longlong_t)offset); | |
34dc7c2f BB |
4606 | } |
4607 | ||
4608 | (void) close(fd); | |
4609 | } | |
4610 | ||
4611 | /* | |
428870ff | 4612 | * Verify that DDT repair works as expected. |
34dc7c2f BB |
4613 | */ |
4614 | void | |
428870ff | 4615 | ztest_ddt_repair(ztest_ds_t *zd, uint64_t id) |
34dc7c2f | 4616 | { |
428870ff BB |
4617 | ztest_shared_t *zs = ztest_shared; |
4618 | spa_t *spa = zs->zs_spa; | |
4619 | objset_t *os = zd->zd_os; | |
4620 | ztest_od_t od[1]; | |
4621 | uint64_t object, blocksize, txg, pattern, psize; | |
4622 | enum zio_checksum checksum = spa_dedup_checksum(spa); | |
4623 | dmu_buf_t *db; | |
4624 | dmu_tx_t *tx; | |
4625 | void *buf; | |
4626 | blkptr_t blk; | |
4627 | int copies = 2 * ZIO_DEDUPDITTO_MIN; | |
d6320ddb | 4628 | int i; |
34dc7c2f | 4629 | |
428870ff BB |
4630 | blocksize = ztest_random_blocksize(); |
4631 | blocksize = MIN(blocksize, 2048); /* because we write so many */ | |
34dc7c2f | 4632 | |
428870ff | 4633 | ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0); |
34dc7c2f | 4634 | |
428870ff BB |
4635 | if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0) |
4636 | return; | |
34dc7c2f BB |
4637 | |
4638 | /* | |
428870ff BB |
4639 | * Take the name lock as writer to prevent anyone else from changing |
4640 | * the pool and dataset properies we need to maintain during this test. | |
34dc7c2f | 4641 | */ |
1e33ac1e | 4642 | (void) rw_enter(&zs->zs_name_lock, RW_WRITER); |
34dc7c2f | 4643 | |
428870ff BB |
4644 | if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum, |
4645 | B_FALSE) != 0 || | |
4646 | ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1, | |
4647 | B_FALSE) != 0) { | |
1e33ac1e | 4648 | (void) rw_exit(&zs->zs_name_lock); |
428870ff BB |
4649 | return; |
4650 | } | |
4651 | ||
4652 | object = od[0].od_object; | |
4653 | blocksize = od[0].od_blocksize; | |
4654 | pattern = spa_guid(spa) ^ dmu_objset_fsid_guid(os); | |
4655 | ||
4656 | ASSERT(object != 0); | |
4657 | ||
4658 | tx = dmu_tx_create(os); | |
4659 | dmu_tx_hold_write(tx, object, 0, copies * blocksize); | |
4660 | txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); | |
4661 | if (txg == 0) { | |
1e33ac1e | 4662 | (void) rw_exit(&zs->zs_name_lock); |
428870ff BB |
4663 | return; |
4664 | } | |
34dc7c2f BB |
4665 | |
4666 | /* | |
428870ff | 4667 | * Write all the copies of our block. |
34dc7c2f | 4668 | */ |
d6320ddb | 4669 | for (i = 0; i < copies; i++) { |
428870ff BB |
4670 | uint64_t offset = i * blocksize; |
4671 | VERIFY(dmu_buf_hold(os, object, offset, FTAG, &db, | |
4672 | DMU_READ_NO_PREFETCH) == 0); | |
4673 | ASSERT(db->db_offset == offset); | |
4674 | ASSERT(db->db_size == blocksize); | |
4675 | ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) || | |
4676 | ztest_pattern_match(db->db_data, db->db_size, 0ULL)); | |
4677 | dmu_buf_will_fill(db, tx); | |
4678 | ztest_pattern_set(db->db_data, db->db_size, pattern); | |
4679 | dmu_buf_rele(db, FTAG); | |
4680 | } | |
34dc7c2f | 4681 | |
428870ff BB |
4682 | dmu_tx_commit(tx); |
4683 | txg_wait_synced(spa_get_dsl(spa), txg); | |
34dc7c2f BB |
4684 | |
4685 | /* | |
428870ff | 4686 | * Find out what block we got. |
34dc7c2f | 4687 | */ |
428870ff BB |
4688 | VERIFY(dmu_buf_hold(os, object, 0, FTAG, &db, |
4689 | DMU_READ_NO_PREFETCH) == 0); | |
4690 | blk = *((dmu_buf_impl_t *)db)->db_blkptr; | |
4691 | dmu_buf_rele(db, FTAG); | |
34dc7c2f BB |
4692 | |
4693 | /* | |
428870ff | 4694 | * Damage the block. Dedup-ditto will save us when we read it later. |
34dc7c2f | 4695 | */ |
428870ff BB |
4696 | psize = BP_GET_PSIZE(&blk); |
4697 | buf = zio_buf_alloc(psize); | |
4698 | ztest_pattern_set(buf, psize, ~pattern); | |
34dc7c2f | 4699 | |
428870ff BB |
4700 | (void) zio_wait(zio_rewrite(NULL, spa, 0, &blk, |
4701 | buf, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE, | |
4702 | ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL)); | |
34dc7c2f | 4703 | |
428870ff | 4704 | zio_buf_free(buf, psize); |
34dc7c2f | 4705 | |
1e33ac1e | 4706 | (void) rw_exit(&zs->zs_name_lock); |
34dc7c2f BB |
4707 | } |
4708 | ||
34dc7c2f | 4709 | /* |
428870ff | 4710 | * Scrub the pool. |
34dc7c2f | 4711 | */ |
428870ff BB |
4712 | /* ARGSUSED */ |
4713 | void | |
4714 | ztest_scrub(ztest_ds_t *zd, uint64_t id) | |
34dc7c2f | 4715 | { |
428870ff BB |
4716 | ztest_shared_t *zs = ztest_shared; |
4717 | spa_t *spa = zs->zs_spa; | |
34dc7c2f | 4718 | |
428870ff BB |
4719 | (void) spa_scan(spa, POOL_SCAN_SCRUB); |
4720 | (void) poll(NULL, 0, 100); /* wait a moment, then force a restart */ | |
4721 | (void) spa_scan(spa, POOL_SCAN_SCRUB); | |
4722 | } | |
34dc7c2f | 4723 | |
428870ff BB |
4724 | /* |
4725 | * Rename the pool to a different name and then rename it back. | |
4726 | */ | |
4727 | /* ARGSUSED */ | |
4728 | void | |
4729 | ztest_spa_rename(ztest_ds_t *zd, uint64_t id) | |
4730 | { | |
4731 | ztest_shared_t *zs = ztest_shared; | |
4732 | char *oldname, *newname; | |
4733 | spa_t *spa; | |
34dc7c2f | 4734 | |
1e33ac1e | 4735 | (void) rw_enter(&zs->zs_name_lock, RW_WRITER); |
34dc7c2f | 4736 | |
428870ff BB |
4737 | oldname = zs->zs_pool; |
4738 | newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL); | |
4739 | (void) strcpy(newname, oldname); | |
4740 | (void) strcat(newname, "_tmp"); | |
34dc7c2f BB |
4741 | |
4742 | /* | |
428870ff | 4743 | * Do the rename |
34dc7c2f | 4744 | */ |
428870ff | 4745 | VERIFY3U(0, ==, spa_rename(oldname, newname)); |
34dc7c2f BB |
4746 | |
4747 | /* | |
428870ff | 4748 | * Try to open it under the old name, which shouldn't exist |
34dc7c2f | 4749 | */ |
428870ff | 4750 | VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG)); |
34dc7c2f BB |
4751 | |
4752 | /* | |
428870ff BB |
4753 | * Open it under the new name and make sure it's still the same spa_t. |
4754 | */ | |
4755 | VERIFY3U(0, ==, spa_open(newname, &spa, FTAG)); | |
34dc7c2f | 4756 | |
428870ff BB |
4757 | ASSERT(spa == zs->zs_spa); |
4758 | spa_close(spa, FTAG); | |
34dc7c2f BB |
4759 | |
4760 | /* | |
428870ff | 4761 | * Rename it back to the original |
34dc7c2f | 4762 | */ |
428870ff | 4763 | VERIFY3U(0, ==, spa_rename(newname, oldname)); |
34dc7c2f | 4764 | |
428870ff BB |
4765 | /* |
4766 | * Make sure it can still be opened | |
4767 | */ | |
4768 | VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG)); | |
34dc7c2f | 4769 | |
428870ff BB |
4770 | ASSERT(spa == zs->zs_spa); |
4771 | spa_close(spa, FTAG); | |
4772 | ||
4773 | umem_free(newname, strlen(newname) + 1); | |
4774 | ||
1e33ac1e | 4775 | (void) rw_exit(&zs->zs_name_lock); |
34dc7c2f BB |
4776 | } |
4777 | ||
428870ff BB |
4778 | /* |
4779 | * Verify pool integrity by running zdb. | |
4780 | */ | |
34dc7c2f | 4781 | static void |
428870ff | 4782 | ztest_run_zdb(char *pool) |
34dc7c2f BB |
4783 | { |
4784 | int status; | |
4785 | char zdb[MAXPATHLEN + MAXNAMELEN + 20]; | |
4786 | char zbuf[1024]; | |
4787 | char *bin; | |
4788 | char *ztest; | |
4789 | char *isa; | |
4790 | int isalen; | |
4791 | FILE *fp; | |
4792 | ||
4793 | (void) realpath(getexecname(), zdb); | |
4794 | ||
4795 | /* zdb lives in /usr/sbin, while ztest lives in /usr/bin */ | |
4796 | bin = strstr(zdb, "/usr/bin/"); | |
4797 | ztest = strstr(bin, "/ztest"); | |
4798 | isa = bin + 8; | |
4799 | isalen = ztest - isa; | |
4800 | isa = strdup(isa); | |
4801 | /* LINTED */ | |
4802 | (void) sprintf(bin, | |
428870ff | 4803 | "/usr/sbin%.*s/zdb -bcc%s%s -U %s %s", |
34dc7c2f BB |
4804 | isalen, |
4805 | isa, | |
4806 | zopt_verbose >= 3 ? "s" : "", | |
4807 | zopt_verbose >= 4 ? "v" : "", | |
428870ff | 4808 | spa_config_path, |
b128c09f | 4809 | pool); |
34dc7c2f BB |
4810 | free(isa); |
4811 | ||
4812 | if (zopt_verbose >= 5) | |
4813 | (void) printf("Executing %s\n", strstr(zdb, "zdb ")); | |
4814 | ||
4815 | fp = popen(zdb, "r"); | |
4816 | ||
4817 | while (fgets(zbuf, sizeof (zbuf), fp) != NULL) | |
4818 | if (zopt_verbose >= 3) | |
4819 | (void) printf("%s", zbuf); | |
4820 | ||
4821 | status = pclose(fp); | |
4822 | ||
4823 | if (status == 0) | |
4824 | return; | |
4825 | ||
4826 | ztest_dump_core = 0; | |
4827 | if (WIFEXITED(status)) | |
4828 | fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status)); | |
4829 | else | |
4830 | fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status)); | |
4831 | } | |
4832 | ||
4833 | static void | |
4834 | ztest_walk_pool_directory(char *header) | |
4835 | { | |
4836 | spa_t *spa = NULL; | |
4837 | ||
4838 | if (zopt_verbose >= 6) | |
4839 | (void) printf("%s\n", header); | |
4840 | ||
4841 | mutex_enter(&spa_namespace_lock); | |
4842 | while ((spa = spa_next(spa)) != NULL) | |
4843 | if (zopt_verbose >= 6) | |
4844 | (void) printf("\t%s\n", spa_name(spa)); | |
4845 | mutex_exit(&spa_namespace_lock); | |
4846 | } | |
4847 | ||
4848 | static void | |
4849 | ztest_spa_import_export(char *oldname, char *newname) | |
4850 | { | |
fb5f0bc8 | 4851 | nvlist_t *config, *newconfig; |
34dc7c2f BB |
4852 | uint64_t pool_guid; |
4853 | spa_t *spa; | |
34dc7c2f BB |
4854 | |
4855 | if (zopt_verbose >= 4) { | |
4856 | (void) printf("import/export: old = %s, new = %s\n", | |
4857 | oldname, newname); | |
4858 | } | |
4859 | ||
4860 | /* | |
4861 | * Clean up from previous runs. | |
4862 | */ | |
4863 | (void) spa_destroy(newname); | |
4864 | ||
4865 | /* | |
4866 | * Get the pool's configuration and guid. | |
4867 | */ | |
428870ff | 4868 | VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG)); |
34dc7c2f | 4869 | |
fb5f0bc8 BB |
4870 | /* |
4871 | * Kick off a scrub to tickle scrub/export races. | |
4872 | */ | |
4873 | if (ztest_random(2) == 0) | |
428870ff | 4874 | (void) spa_scan(spa, POOL_SCAN_SCRUB); |
fb5f0bc8 | 4875 | |
34dc7c2f BB |
4876 | pool_guid = spa_guid(spa); |
4877 | spa_close(spa, FTAG); | |
4878 | ||
4879 | ztest_walk_pool_directory("pools before export"); | |
4880 | ||
4881 | /* | |
4882 | * Export it. | |
4883 | */ | |
428870ff | 4884 | VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE)); |
34dc7c2f BB |
4885 | |
4886 | ztest_walk_pool_directory("pools after export"); | |
4887 | ||
fb5f0bc8 BB |
4888 | /* |
4889 | * Try to import it. | |
4890 | */ | |
4891 | newconfig = spa_tryimport(config); | |
4892 | ASSERT(newconfig != NULL); | |
4893 | nvlist_free(newconfig); | |
4894 | ||
34dc7c2f BB |
4895 | /* |
4896 | * Import it under the new name. | |
4897 | */ | |
572e2857 | 4898 | VERIFY3U(0, ==, spa_import(newname, config, NULL, 0)); |
34dc7c2f BB |
4899 | |
4900 | ztest_walk_pool_directory("pools after import"); | |
4901 | ||
4902 | /* | |
4903 | * Try to import it again -- should fail with EEXIST. | |
4904 | */ | |
572e2857 | 4905 | VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0)); |
34dc7c2f BB |
4906 | |
4907 | /* | |
4908 | * Try to import it under a different name -- should fail with EEXIST. | |
4909 | */ | |
572e2857 | 4910 | VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0)); |
34dc7c2f BB |
4911 | |
4912 | /* | |
4913 | * Verify that the pool is no longer visible under the old name. | |
4914 | */ | |
428870ff | 4915 | VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG)); |
34dc7c2f BB |
4916 | |
4917 | /* | |
4918 | * Verify that we can open and close the pool using the new name. | |
4919 | */ | |
428870ff | 4920 | VERIFY3U(0, ==, spa_open(newname, &spa, FTAG)); |
34dc7c2f BB |
4921 | ASSERT(pool_guid == spa_guid(spa)); |
4922 | spa_close(spa, FTAG); | |
4923 | ||
4924 | nvlist_free(config); | |
4925 | } | |
4926 | ||
fb5f0bc8 BB |
4927 | static void |
4928 | ztest_resume(spa_t *spa) | |
4929 | { | |
428870ff BB |
4930 | if (spa_suspended(spa) && zopt_verbose >= 6) |
4931 | (void) printf("resuming from suspended state\n"); | |
4932 | spa_vdev_state_enter(spa, SCL_NONE); | |
4933 | vdev_clear(spa, NULL); | |
4934 | (void) spa_vdev_state_exit(spa, NULL, 0); | |
4935 | (void) zio_resume(spa); | |
fb5f0bc8 BB |
4936 | } |
4937 | ||
34dc7c2f | 4938 | static void * |
fb5f0bc8 | 4939 | ztest_resume_thread(void *arg) |
34dc7c2f | 4940 | { |
b128c09f | 4941 | spa_t *spa = arg; |
34dc7c2f BB |
4942 | |
4943 | while (!ztest_exiting) { | |
428870ff BB |
4944 | if (spa_suspended(spa)) |
4945 | ztest_resume(spa); | |
4946 | (void) poll(NULL, 0, 100); | |
34dc7c2f | 4947 | } |
34dc7c2f | 4948 | |
1e33ac1e | 4949 | thread_exit(); |
428870ff | 4950 | |
1e33ac1e BB |
4951 | return (NULL); |
4952 | } | |
428870ff | 4953 | |
1e33ac1e | 4954 | #define GRACE 300 |
428870ff | 4955 | |
1e33ac1e BB |
4956 | static void |
4957 | ztest_deadman_alarm(int sig) | |
4958 | { | |
4959 | fatal(0, "failed to complete within %d seconds of deadline", GRACE); | |
428870ff BB |
4960 | } |
4961 | ||
4962 | static void | |
4963 | ztest_execute(ztest_info_t *zi, uint64_t id) | |
4964 | { | |
4965 | ztest_shared_t *zs = ztest_shared; | |
4966 | ztest_ds_t *zd = &zs->zs_zd[id % zopt_datasets]; | |
4967 | hrtime_t functime = gethrtime(); | |
d6320ddb | 4968 | int i; |
428870ff | 4969 | |
d6320ddb | 4970 | for (i = 0; i < zi->zi_iters; i++) |
428870ff BB |
4971 | zi->zi_func(zd, id); |
4972 | ||
4973 | functime = gethrtime() - functime; | |
4974 | ||
4975 | atomic_add_64(&zi->zi_call_count, 1); | |
4976 | atomic_add_64(&zi->zi_call_time, functime); | |
4977 | ||
4978 | if (zopt_verbose >= 4) { | |
4979 | Dl_info dli; | |
4980 | (void) dladdr((void *)zi->zi_func, &dli); | |
4981 | (void) printf("%6.2f sec in %s\n", | |
4982 | (double)functime / NANOSEC, dli.dli_sname); | |
4983 | } | |
4984 | } | |
4985 | ||
34dc7c2f BB |
4986 | static void * |
4987 | ztest_thread(void *arg) | |
4988 | { | |
428870ff | 4989 | uint64_t id = (uintptr_t)arg; |
34dc7c2f | 4990 | ztest_shared_t *zs = ztest_shared; |
428870ff BB |
4991 | uint64_t call_next; |
4992 | hrtime_t now; | |
34dc7c2f | 4993 | ztest_info_t *zi; |
34dc7c2f | 4994 | |
428870ff | 4995 | while ((now = gethrtime()) < zs->zs_thread_stop) { |
34dc7c2f BB |
4996 | /* |
4997 | * See if it's time to force a crash. | |
4998 | */ | |
428870ff BB |
4999 | if (now > zs->zs_thread_kill) |
5000 | ztest_kill(zs); | |
34dc7c2f BB |
5001 | |
5002 | /* | |
428870ff | 5003 | * If we're getting ENOSPC with some regularity, stop. |
34dc7c2f | 5004 | */ |
428870ff BB |
5005 | if (zs->zs_enospc_count > 10) |
5006 | break; | |
34dc7c2f BB |
5007 | |
5008 | /* | |
428870ff | 5009 | * Pick a random function to execute. |
34dc7c2f | 5010 | */ |
428870ff BB |
5011 | zi = &zs->zs_info[ztest_random(ZTEST_FUNCS)]; |
5012 | call_next = zi->zi_call_next; | |
34dc7c2f | 5013 | |
428870ff BB |
5014 | if (now >= call_next && |
5015 | atomic_cas_64(&zi->zi_call_next, call_next, call_next + | |
5016 | ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) | |
5017 | ztest_execute(zi, id); | |
5018 | } | |
34dc7c2f | 5019 | |
1e33ac1e BB |
5020 | thread_exit(); |
5021 | ||
428870ff BB |
5022 | return (NULL); |
5023 | } | |
34dc7c2f | 5024 | |
428870ff BB |
5025 | static void |
5026 | ztest_dataset_name(char *dsname, char *pool, int d) | |
5027 | { | |
5028 | (void) snprintf(dsname, MAXNAMELEN, "%s/ds_%d", pool, d); | |
5029 | } | |
34dc7c2f | 5030 | |
428870ff BB |
5031 | static void |
5032 | ztest_dataset_destroy(ztest_shared_t *zs, int d) | |
5033 | { | |
5034 | char name[MAXNAMELEN]; | |
d6320ddb | 5035 | int t; |
34dc7c2f | 5036 | |
428870ff | 5037 | ztest_dataset_name(name, zs->zs_pool, d); |
34dc7c2f | 5038 | |
428870ff BB |
5039 | if (zopt_verbose >= 3) |
5040 | (void) printf("Destroying %s to free up space\n", name); | |
34dc7c2f | 5041 | |
428870ff BB |
5042 | /* |
5043 | * Cleanup any non-standard clones and snapshots. In general, | |
5044 | * ztest thread t operates on dataset (t % zopt_datasets), | |
5045 | * so there may be more than one thing to clean up. | |
5046 | */ | |
d6320ddb | 5047 | for (t = d; t < zopt_threads; t += zopt_datasets) |
428870ff BB |
5048 | ztest_dsl_dataset_cleanup(name, t); |
5049 | ||
5050 | (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL, | |
5051 | DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN); | |
5052 | } | |
5053 | ||
5054 | static void | |
5055 | ztest_dataset_dirobj_verify(ztest_ds_t *zd) | |
5056 | { | |
5057 | uint64_t usedobjs, dirobjs, scratch; | |
5058 | ||
5059 | /* | |
5060 | * ZTEST_DIROBJ is the object directory for the entire dataset. | |
5061 | * Therefore, the number of objects in use should equal the | |
5062 | * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself. | |
5063 | * If not, we have an object leak. | |
5064 | * | |
5065 | * Note that we can only check this in ztest_dataset_open(), | |
5066 | * when the open-context and syncing-context values agree. | |
5067 | * That's because zap_count() returns the open-context value, | |
5068 | * while dmu_objset_space() returns the rootbp fill count. | |
5069 | */ | |
5070 | VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs)); | |
5071 | dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch); | |
5072 | ASSERT3U(dirobjs + 1, ==, usedobjs); | |
5073 | } | |
5074 | ||
5075 | static int | |
5076 | ztest_dataset_open(ztest_shared_t *zs, int d) | |
5077 | { | |
5078 | ztest_ds_t *zd = &zs->zs_zd[d]; | |
5079 | uint64_t committed_seq = zd->zd_seq; | |
5080 | objset_t *os; | |
5081 | zilog_t *zilog; | |
5082 | char name[MAXNAMELEN]; | |
5083 | int error; | |
5084 | ||
5085 | ztest_dataset_name(name, zs->zs_pool, d); | |
5086 | ||
1e33ac1e | 5087 | (void) rw_enter(&zs->zs_name_lock, RW_READER); |
428870ff BB |
5088 | |
5089 | error = ztest_dataset_create(name); | |
5090 | if (error == ENOSPC) { | |
1e33ac1e | 5091 | (void) rw_exit(&zs->zs_name_lock); |
428870ff BB |
5092 | ztest_record_enospc(FTAG); |
5093 | return (error); | |
34dc7c2f | 5094 | } |
428870ff | 5095 | ASSERT(error == 0 || error == EEXIST); |
34dc7c2f | 5096 | |
428870ff | 5097 | VERIFY3U(dmu_objset_hold(name, zd, &os), ==, 0); |
1e33ac1e | 5098 | (void) rw_exit(&zs->zs_name_lock); |
428870ff BB |
5099 | |
5100 | ztest_zd_init(zd, os); | |
5101 | ||
5102 | zilog = zd->zd_zilog; | |
5103 | ||
5104 | if (zilog->zl_header->zh_claim_lr_seq != 0 && | |
5105 | zilog->zl_header->zh_claim_lr_seq < committed_seq) | |
5106 | fatal(0, "missing log records: claimed %llu < committed %llu", | |
5107 | zilog->zl_header->zh_claim_lr_seq, committed_seq); | |
5108 | ||
5109 | ztest_dataset_dirobj_verify(zd); | |
5110 | ||
5111 | zil_replay(os, zd, ztest_replay_vector); | |
5112 | ||
5113 | ztest_dataset_dirobj_verify(zd); | |
5114 | ||
5115 | if (zopt_verbose >= 6) | |
5116 | (void) printf("%s replay %llu blocks, %llu records, seq %llu\n", | |
5117 | zd->zd_name, | |
5118 | (u_longlong_t)zilog->zl_parse_blk_count, | |
5119 | (u_longlong_t)zilog->zl_parse_lr_count, | |
5120 | (u_longlong_t)zilog->zl_replaying_seq); | |
5121 | ||
5122 | zilog = zil_open(os, ztest_get_data); | |
5123 | ||
5124 | if (zilog->zl_replaying_seq != 0 && | |
5125 | zilog->zl_replaying_seq < committed_seq) | |
5126 | fatal(0, "missing log records: replayed %llu < committed %llu", | |
5127 | zilog->zl_replaying_seq, committed_seq); | |
5128 | ||
5129 | return (0); | |
5130 | } | |
5131 | ||
5132 | static void | |
5133 | ztest_dataset_close(ztest_shared_t *zs, int d) | |
5134 | { | |
5135 | ztest_ds_t *zd = &zs->zs_zd[d]; | |
5136 | ||
5137 | zil_close(zd->zd_zilog); | |
5138 | dmu_objset_rele(zd->zd_os, zd); | |
5139 | ||
5140 | ztest_zd_fini(zd); | |
34dc7c2f BB |
5141 | } |
5142 | ||
5143 | /* | |
5144 | * Kick off threads to run tests on all datasets in parallel. | |
5145 | */ | |
5146 | static void | |
428870ff | 5147 | ztest_run(ztest_shared_t *zs) |
34dc7c2f | 5148 | { |
1e33ac1e | 5149 | kt_did_t *tid; |
34dc7c2f | 5150 | spa_t *spa; |
1e33ac1e BB |
5151 | kthread_t *resume_thread; |
5152 | uint64_t object; | |
428870ff | 5153 | int error; |
d6320ddb | 5154 | int t, d; |
b128c09f BB |
5155 | |
5156 | ztest_exiting = B_FALSE; | |
34dc7c2f | 5157 | |
34dc7c2f | 5158 | /* |
428870ff | 5159 | * Initialize parent/child shared state. |
34dc7c2f | 5160 | */ |
1e33ac1e BB |
5161 | mutex_init(&zs->zs_vdev_lock, NULL, MUTEX_DEFAULT, NULL); |
5162 | rw_init(&zs->zs_name_lock, NULL, RW_DEFAULT, NULL); | |
34dc7c2f | 5163 | |
428870ff BB |
5164 | zs->zs_thread_start = gethrtime(); |
5165 | zs->zs_thread_stop = zs->zs_thread_start + zopt_passtime * NANOSEC; | |
5166 | zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop); | |
5167 | zs->zs_thread_kill = zs->zs_thread_stop; | |
5168 | if (ztest_random(100) < zopt_killrate) | |
5169 | zs->zs_thread_kill -= ztest_random(zopt_passtime * NANOSEC); | |
34dc7c2f | 5170 | |
1e33ac1e | 5171 | mutex_init(&zcl.zcl_callbacks_lock, NULL, MUTEX_DEFAULT, NULL); |
34dc7c2f | 5172 | |
428870ff BB |
5173 | list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t), |
5174 | offsetof(ztest_cb_data_t, zcd_node)); | |
34dc7c2f BB |
5175 | |
5176 | /* | |
b128c09f | 5177 | * Open our pool. |
34dc7c2f | 5178 | */ |
428870ff BB |
5179 | kernel_init(FREAD | FWRITE); |
5180 | VERIFY(spa_open(zs->zs_pool, &spa, FTAG) == 0); | |
5181 | zs->zs_spa = spa; | |
5182 | ||
5183 | spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN; | |
34dc7c2f | 5184 | |
fb5f0bc8 BB |
5185 | /* |
5186 | * We don't expect the pool to suspend unless maxfaults == 0, | |
5187 | * in which case ztest_fault_inject() temporarily takes away | |
5188 | * the only valid replica. | |
5189 | */ | |
428870ff | 5190 | if (MAXFAULTS() == 0) |
fb5f0bc8 BB |
5191 | spa->spa_failmode = ZIO_FAILURE_MODE_WAIT; |
5192 | else | |
5193 | spa->spa_failmode = ZIO_FAILURE_MODE_PANIC; | |
5194 | ||
34dc7c2f | 5195 | /* |
b128c09f | 5196 | * Create a thread to periodically resume suspended I/O. |
34dc7c2f | 5197 | */ |
1e33ac1e BB |
5198 | VERIFY3P((resume_thread = thread_create(NULL, 0, ztest_resume_thread, |
5199 | spa, TS_RUN, NULL, 0, 0)), !=, NULL); | |
34dc7c2f | 5200 | |
428870ff | 5201 | /* |
1e33ac1e | 5202 | * Set a deadman alarm to abort() if we hang. |
428870ff | 5203 | */ |
1e33ac1e BB |
5204 | signal(SIGALRM, ztest_deadman_alarm); |
5205 | alarm((zs->zs_thread_stop - zs->zs_thread_start) / NANOSEC + GRACE); | |
428870ff | 5206 | |
34dc7c2f BB |
5207 | /* |
5208 | * Verify that we can safely inquire about about any object, | |
5209 | * whether it's allocated or not. To make it interesting, | |
5210 | * we probe a 5-wide window around each power of two. | |
5211 | * This hits all edge cases, including zero and the max. | |
5212 | */ | |
d6320ddb BB |
5213 | for (t = 0; t < 64; t++) { |
5214 | for (d = -5; d <= 5; d++) { | |
34dc7c2f BB |
5215 | error = dmu_object_info(spa->spa_meta_objset, |
5216 | (1ULL << t) + d, NULL); | |
5217 | ASSERT(error == 0 || error == ENOENT || | |
5218 | error == EINVAL); | |
5219 | } | |
5220 | } | |
5221 | ||
5222 | /* | |
428870ff | 5223 | * If we got any ENOSPC errors on the previous run, destroy something. |
34dc7c2f | 5224 | */ |
428870ff BB |
5225 | if (zs->zs_enospc_count != 0) { |
5226 | int d = ztest_random(zopt_datasets); | |
5227 | ztest_dataset_destroy(zs, d); | |
5228 | } | |
34dc7c2f BB |
5229 | zs->zs_enospc_count = 0; |
5230 | ||
1e33ac1e | 5231 | tid = umem_zalloc(zopt_threads * sizeof (kt_did_t), UMEM_NOFAIL); |
34dc7c2f BB |
5232 | |
5233 | if (zopt_verbose >= 4) | |
5234 | (void) printf("starting main threads...\n"); | |
5235 | ||
428870ff BB |
5236 | /* |
5237 | * Kick off all the tests that run in parallel. | |
5238 | */ | |
d6320ddb | 5239 | for (t = 0; t < zopt_threads; t++) { |
1e33ac1e BB |
5240 | kthread_t *thread; |
5241 | ||
428870ff BB |
5242 | if (t < zopt_datasets && ztest_dataset_open(zs, t) != 0) |
5243 | return; | |
1e33ac1e BB |
5244 | |
5245 | VERIFY3P(thread = thread_create(NULL, 0, ztest_thread, | |
5246 | (void *)(uintptr_t)t, TS_RUN, NULL, 0, 0), !=, NULL); | |
5247 | tid[t] = thread->t_tid; | |
34dc7c2f BB |
5248 | } |
5249 | ||
428870ff BB |
5250 | /* |
5251 | * Wait for all of the tests to complete. We go in reverse order | |
5252 | * so we don't close datasets while threads are still using them. | |
5253 | */ | |
d6320ddb | 5254 | for (t = zopt_threads - 1; t >= 0; t--) { |
1e33ac1e | 5255 | thread_join(tid[t]); |
428870ff BB |
5256 | if (t < zopt_datasets) |
5257 | ztest_dataset_close(zs, t); | |
34dc7c2f BB |
5258 | } |
5259 | ||
34dc7c2f BB |
5260 | txg_wait_synced(spa_get_dsl(spa), 0); |
5261 | ||
428870ff BB |
5262 | zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa)); |
5263 | zs->zs_space = metaslab_class_get_space(spa_normal_class(spa)); | |
5264 | ||
1e33ac1e | 5265 | umem_free(tid, zopt_threads * sizeof (kt_did_t)); |
428870ff BB |
5266 | |
5267 | /* Kill the resume thread */ | |
5268 | ztest_exiting = B_TRUE; | |
1e33ac1e | 5269 | thread_join(resume_thread->t_tid); |
428870ff | 5270 | ztest_resume(spa); |
34dc7c2f BB |
5271 | |
5272 | /* | |
428870ff BB |
5273 | * Right before closing the pool, kick off a bunch of async I/O; |
5274 | * spa_close() should wait for it to complete. | |
34dc7c2f | 5275 | */ |
1e33ac1e | 5276 | for (object = 1; object < 50; object++) |
428870ff BB |
5277 | dmu_prefetch(spa->spa_meta_objset, object, 0, 1ULL << 20); |
5278 | ||
090ff092 RC |
5279 | /* Verify that at least one commit cb was called in a timely fashion */ |
5280 | if (zc_cb_counter >= ZTEST_COMMIT_CB_MIN_REG) | |
5281 | VERIFY3U(zc_min_txg_delay, ==, 0); | |
5282 | ||
428870ff BB |
5283 | spa_close(spa, FTAG); |
5284 | ||
5285 | /* | |
5286 | * Verify that we can loop over all pools. | |
5287 | */ | |
5288 | mutex_enter(&spa_namespace_lock); | |
5289 | for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa)) | |
5290 | if (zopt_verbose > 3) | |
5291 | (void) printf("spa_next: found %s\n", spa_name(spa)); | |
5292 | mutex_exit(&spa_namespace_lock); | |
5293 | ||
5294 | /* | |
5295 | * Verify that we can export the pool and reimport it under a | |
5296 | * different name. | |
5297 | */ | |
5298 | if (ztest_random(2) == 0) { | |
5299 | char name[MAXNAMELEN]; | |
5300 | (void) snprintf(name, MAXNAMELEN, "%s_import", zs->zs_pool); | |
5301 | ztest_spa_import_export(zs->zs_pool, name); | |
5302 | ztest_spa_import_export(name, zs->zs_pool); | |
5303 | } | |
5304 | ||
5305 | kernel_fini(); | |
572e2857 BB |
5306 | |
5307 | list_destroy(&zcl.zcl_callbacks); | |
5308 | ||
5309 | (void) _mutex_destroy(&zcl.zcl_callbacks_lock); | |
5310 | ||
5311 | (void) rwlock_destroy(&zs->zs_name_lock); | |
5312 | (void) _mutex_destroy(&zs->zs_vdev_lock); | |
428870ff BB |
5313 | } |
5314 | ||
5315 | static void | |
5316 | ztest_freeze(ztest_shared_t *zs) | |
5317 | { | |
5318 | ztest_ds_t *zd = &zs->zs_zd[0]; | |
5319 | spa_t *spa; | |
5320 | int numloops = 0; | |
5321 | ||
5322 | if (zopt_verbose >= 3) | |
5323 | (void) printf("testing spa_freeze()...\n"); | |
9babb374 | 5324 | |
428870ff BB |
5325 | kernel_init(FREAD | FWRITE); |
5326 | VERIFY3U(0, ==, spa_open(zs->zs_pool, &spa, FTAG)); | |
5327 | VERIFY3U(0, ==, ztest_dataset_open(zs, 0)); | |
9babb374 | 5328 | |
428870ff BB |
5329 | /* |
5330 | * Force the first log block to be transactionally allocated. | |
5331 | * We have to do this before we freeze the pool -- otherwise | |
5332 | * the log chain won't be anchored. | |
5333 | */ | |
5334 | while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) { | |
5335 | ztest_dmu_object_alloc_free(zd, 0); | |
572e2857 | 5336 | zil_commit(zd->zd_zilog, 0); |
34dc7c2f BB |
5337 | } |
5338 | ||
5339 | txg_wait_synced(spa_get_dsl(spa), 0); | |
5340 | ||
428870ff BB |
5341 | /* |
5342 | * Freeze the pool. This stops spa_sync() from doing anything, | |
5343 | * so that the only way to record changes from now on is the ZIL. | |
5344 | */ | |
5345 | spa_freeze(spa); | |
b128c09f | 5346 | |
428870ff BB |
5347 | /* |
5348 | * Run tests that generate log records but don't alter the pool config | |
5349 | * or depend on DSL sync tasks (snapshots, objset create/destroy, etc). | |
5350 | * We do a txg_wait_synced() after each iteration to force the txg | |
5351 | * to increase well beyond the last synced value in the uberblock. | |
5352 | * The ZIL should be OK with that. | |
5353 | */ | |
5354 | while (ztest_random(10) != 0 && numloops++ < zopt_maxloops) { | |
5355 | ztest_dmu_write_parallel(zd, 0); | |
5356 | ztest_dmu_object_alloc_free(zd, 0); | |
5357 | txg_wait_synced(spa_get_dsl(spa), 0); | |
5358 | } | |
b128c09f | 5359 | |
34dc7c2f | 5360 | /* |
428870ff | 5361 | * Commit all of the changes we just generated. |
34dc7c2f | 5362 | */ |
572e2857 | 5363 | zil_commit(zd->zd_zilog, 0); |
428870ff | 5364 | txg_wait_synced(spa_get_dsl(spa), 0); |
34dc7c2f | 5365 | |
428870ff BB |
5366 | /* |
5367 | * Close our dataset and close the pool. | |
5368 | */ | |
5369 | ztest_dataset_close(zs, 0); | |
34dc7c2f | 5370 | spa_close(spa, FTAG); |
428870ff | 5371 | kernel_fini(); |
34dc7c2f | 5372 | |
428870ff BB |
5373 | /* |
5374 | * Open and close the pool and dataset to induce log replay. | |
5375 | */ | |
5376 | kernel_init(FREAD | FWRITE); | |
5377 | VERIFY3U(0, ==, spa_open(zs->zs_pool, &spa, FTAG)); | |
5378 | VERIFY3U(0, ==, ztest_dataset_open(zs, 0)); | |
5379 | ztest_dataset_close(zs, 0); | |
5380 | spa_close(spa, FTAG); | |
34dc7c2f BB |
5381 | kernel_fini(); |
5382 | } | |
5383 | ||
5384 | void | |
5385 | print_time(hrtime_t t, char *timebuf) | |
5386 | { | |
5387 | hrtime_t s = t / NANOSEC; | |
5388 | hrtime_t m = s / 60; | |
5389 | hrtime_t h = m / 60; | |
5390 | hrtime_t d = h / 24; | |
5391 | ||
5392 | s -= m * 60; | |
5393 | m -= h * 60; | |
5394 | h -= d * 24; | |
5395 | ||
5396 | timebuf[0] = '\0'; | |
5397 | ||
5398 | if (d) | |
5399 | (void) sprintf(timebuf, | |
5400 | "%llud%02lluh%02llum%02llus", d, h, m, s); | |
5401 | else if (h) | |
5402 | (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s); | |
5403 | else if (m) | |
5404 | (void) sprintf(timebuf, "%llum%02llus", m, s); | |
5405 | else | |
5406 | (void) sprintf(timebuf, "%llus", s); | |
5407 | } | |
5408 | ||
428870ff | 5409 | static nvlist_t * |
0bc8fd78 | 5410 | make_random_props(void) |
428870ff BB |
5411 | { |
5412 | nvlist_t *props; | |
5413 | ||
5414 | if (ztest_random(2) == 0) | |
5415 | return (NULL); | |
5416 | ||
5417 | VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0); | |
5418 | VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0); | |
5419 | ||
5420 | (void) printf("props:\n"); | |
5421 | dump_nvlist(props, 4); | |
5422 | ||
5423 | return (props); | |
5424 | } | |
5425 | ||
34dc7c2f BB |
5426 | /* |
5427 | * Create a storage pool with the given name and initial vdev size. | |
428870ff | 5428 | * Then test spa_freeze() functionality. |
34dc7c2f BB |
5429 | */ |
5430 | static void | |
428870ff | 5431 | ztest_init(ztest_shared_t *zs) |
34dc7c2f BB |
5432 | { |
5433 | spa_t *spa; | |
428870ff BB |
5434 | nvlist_t *nvroot, *props; |
5435 | ||
1e33ac1e BB |
5436 | mutex_init(&zs->zs_vdev_lock, NULL, MUTEX_DEFAULT, NULL); |
5437 | rw_init(&zs->zs_name_lock, NULL, RW_DEFAULT, NULL); | |
34dc7c2f BB |
5438 | |
5439 | kernel_init(FREAD | FWRITE); | |
5440 | ||
5441 | /* | |
5442 | * Create the storage pool. | |
5443 | */ | |
428870ff BB |
5444 | (void) spa_destroy(zs->zs_pool); |
5445 | ztest_shared->zs_vdev_next_leaf = 0; | |
5446 | zs->zs_splits = 0; | |
5447 | zs->zs_mirrors = zopt_mirrors; | |
b128c09f | 5448 | nvroot = make_vdev_root(NULL, NULL, zopt_vdev_size, 0, |
428870ff BB |
5449 | 0, zopt_raidz, zs->zs_mirrors, 1); |
5450 | props = make_random_props(); | |
5451 | VERIFY3U(0, ==, spa_create(zs->zs_pool, nvroot, props, NULL, NULL)); | |
34dc7c2f BB |
5452 | nvlist_free(nvroot); |
5453 | ||
428870ff | 5454 | VERIFY3U(0, ==, spa_open(zs->zs_pool, &spa, FTAG)); |
9babb374 | 5455 | metaslab_sz = 1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift; |
34dc7c2f BB |
5456 | spa_close(spa, FTAG); |
5457 | ||
5458 | kernel_fini(); | |
428870ff BB |
5459 | |
5460 | ztest_run_zdb(zs->zs_pool); | |
5461 | ||
5462 | ztest_freeze(zs); | |
5463 | ||
5464 | ztest_run_zdb(zs->zs_pool); | |
572e2857 | 5465 | |
1e33ac1e BB |
5466 | (void) rw_destroy(&zs->zs_name_lock); |
5467 | (void) mutex_destroy(&zs->zs_vdev_lock); | |
34dc7c2f BB |
5468 | } |
5469 | ||
5470 | int | |
5471 | main(int argc, char **argv) | |
5472 | { | |
5473 | int kills = 0; | |
5474 | int iters = 0; | |
34dc7c2f | 5475 | ztest_shared_t *zs; |
428870ff | 5476 | size_t shared_size; |
34dc7c2f BB |
5477 | ztest_info_t *zi; |
5478 | char timebuf[100]; | |
5479 | char numbuf[6]; | |
428870ff | 5480 | spa_t *spa; |
d6320ddb | 5481 | int i, f; |
34dc7c2f BB |
5482 | |
5483 | (void) setvbuf(stdout, NULL, _IOLBF, 0); | |
5484 | ||
34dc7c2f BB |
5485 | ztest_random_fd = open("/dev/urandom", O_RDONLY); |
5486 | ||
5487 | process_options(argc, argv); | |
5488 | ||
428870ff | 5489 | /* Override location of zpool.cache */ |
0e5b68e0 BB |
5490 | VERIFY(asprintf((char **)&spa_config_path, "%s/zpool.cache", |
5491 | zopt_dir) != -1); | |
428870ff | 5492 | |
34dc7c2f BB |
5493 | /* |
5494 | * Blow away any existing copy of zpool.cache | |
5495 | */ | |
5496 | if (zopt_init != 0) | |
428870ff BB |
5497 | (void) remove(spa_config_path); |
5498 | ||
5499 | shared_size = sizeof (*zs) + zopt_datasets * sizeof (ztest_ds_t); | |
34dc7c2f BB |
5500 | |
5501 | zs = ztest_shared = (void *)mmap(0, | |
428870ff | 5502 | P2ROUNDUP(shared_size, getpagesize()), |
34dc7c2f BB |
5503 | PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANON, -1, 0); |
5504 | ||
5505 | if (zopt_verbose >= 1) { | |
5506 | (void) printf("%llu vdevs, %d datasets, %d threads," | |
5507 | " %llu seconds...\n", | |
5508 | (u_longlong_t)zopt_vdevs, zopt_datasets, zopt_threads, | |
5509 | (u_longlong_t)zopt_time); | |
5510 | } | |
5511 | ||
5512 | /* | |
5513 | * Create and initialize our storage pool. | |
5514 | */ | |
d6320ddb | 5515 | for (i = 1; i <= zopt_init; i++) { |
34dc7c2f BB |
5516 | bzero(zs, sizeof (ztest_shared_t)); |
5517 | if (zopt_verbose >= 3 && zopt_init != 1) | |
5518 | (void) printf("ztest_init(), pass %d\n", i); | |
428870ff BB |
5519 | zs->zs_pool = zopt_pool; |
5520 | ztest_init(zs); | |
34dc7c2f BB |
5521 | } |
5522 | ||
428870ff BB |
5523 | zs->zs_pool = zopt_pool; |
5524 | zs->zs_proc_start = gethrtime(); | |
5525 | zs->zs_proc_stop = zs->zs_proc_start + zopt_time * NANOSEC; | |
34dc7c2f | 5526 | |
d6320ddb | 5527 | for (f = 0; f < ZTEST_FUNCS; f++) { |
428870ff | 5528 | zi = &zs->zs_info[f]; |
34dc7c2f | 5529 | *zi = ztest_info[f]; |
428870ff BB |
5530 | if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop) |
5531 | zi->zi_call_next = UINT64_MAX; | |
34dc7c2f | 5532 | else |
428870ff BB |
5533 | zi->zi_call_next = zs->zs_proc_start + |
5534 | ztest_random(2 * zi->zi_interval[0] + 1); | |
34dc7c2f BB |
5535 | } |
5536 | ||
34dc7c2f BB |
5537 | /* |
5538 | * Run the tests in a loop. These tests include fault injection | |
5539 | * to verify that self-healing data works, and forced crashes | |
5540 | * to verify that we never lose on-disk consistency. | |
5541 | */ | |
428870ff | 5542 | while (gethrtime() < zs->zs_proc_stop) { |
34dc7c2f BB |
5543 | int status; |
5544 | pid_t pid; | |
34dc7c2f BB |
5545 | |
5546 | /* | |
5547 | * Initialize the workload counters for each function. | |
5548 | */ | |
d6320ddb | 5549 | for (f = 0; f < ZTEST_FUNCS; f++) { |
34dc7c2f | 5550 | zi = &zs->zs_info[f]; |
428870ff | 5551 | zi->zi_call_count = 0; |
34dc7c2f BB |
5552 | zi->zi_call_time = 0; |
5553 | } | |
5554 | ||
9babb374 BB |
5555 | /* Set the allocation switch size */ |
5556 | metaslab_df_alloc_threshold = ztest_random(metaslab_sz / 4) + 1; | |
5557 | ||
34dc7c2f BB |
5558 | pid = fork(); |
5559 | ||
5560 | if (pid == -1) | |
5561 | fatal(1, "fork failed"); | |
5562 | ||
5563 | if (pid == 0) { /* child */ | |
5564 | struct rlimit rl = { 1024, 1024 }; | |
5565 | (void) setrlimit(RLIMIT_NOFILE, &rl); | |
5566 | (void) enable_extended_FILE_stdio(-1, -1); | |
428870ff | 5567 | ztest_run(zs); |
34dc7c2f BB |
5568 | exit(0); |
5569 | } | |
5570 | ||
5571 | while (waitpid(pid, &status, 0) != pid) | |
5572 | continue; | |
5573 | ||
5574 | if (WIFEXITED(status)) { | |
5575 | if (WEXITSTATUS(status) != 0) { | |
5576 | (void) fprintf(stderr, | |
5577 | "child exited with code %d\n", | |
5578 | WEXITSTATUS(status)); | |
5579 | exit(2); | |
5580 | } | |
5581 | } else if (WIFSIGNALED(status)) { | |
5582 | if (WTERMSIG(status) != SIGKILL) { | |
5583 | (void) fprintf(stderr, | |
5584 | "child died with signal %d\n", | |
5585 | WTERMSIG(status)); | |
5586 | exit(3); | |
5587 | } | |
5588 | kills++; | |
5589 | } else { | |
5590 | (void) fprintf(stderr, "something strange happened " | |
5591 | "to child\n"); | |
5592 | exit(4); | |
5593 | } | |
5594 | ||
5595 | iters++; | |
5596 | ||
5597 | if (zopt_verbose >= 1) { | |
5598 | hrtime_t now = gethrtime(); | |
5599 | ||
428870ff BB |
5600 | now = MIN(now, zs->zs_proc_stop); |
5601 | print_time(zs->zs_proc_stop - now, timebuf); | |
34dc7c2f BB |
5602 | nicenum(zs->zs_space, numbuf); |
5603 | ||
5604 | (void) printf("Pass %3d, %8s, %3llu ENOSPC, " | |
5605 | "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n", | |
5606 | iters, | |
5607 | WIFEXITED(status) ? "Complete" : "SIGKILL", | |
5608 | (u_longlong_t)zs->zs_enospc_count, | |
5609 | 100.0 * zs->zs_alloc / zs->zs_space, | |
5610 | numbuf, | |
428870ff | 5611 | 100.0 * (now - zs->zs_proc_start) / |
34dc7c2f BB |
5612 | (zopt_time * NANOSEC), timebuf); |
5613 | } | |
5614 | ||
5615 | if (zopt_verbose >= 2) { | |
5616 | (void) printf("\nWorkload summary:\n\n"); | |
5617 | (void) printf("%7s %9s %s\n", | |
5618 | "Calls", "Time", "Function"); | |
5619 | (void) printf("%7s %9s %s\n", | |
5620 | "-----", "----", "--------"); | |
d6320ddb | 5621 | for (f = 0; f < ZTEST_FUNCS; f++) { |
34dc7c2f BB |
5622 | Dl_info dli; |
5623 | ||
5624 | zi = &zs->zs_info[f]; | |
5625 | print_time(zi->zi_call_time, timebuf); | |
5626 | (void) dladdr((void *)zi->zi_func, &dli); | |
5627 | (void) printf("%7llu %9s %s\n", | |
428870ff | 5628 | (u_longlong_t)zi->zi_call_count, timebuf, |
34dc7c2f BB |
5629 | dli.dli_sname); |
5630 | } | |
5631 | (void) printf("\n"); | |
5632 | } | |
5633 | ||
5634 | /* | |
428870ff BB |
5635 | * It's possible that we killed a child during a rename test, |
5636 | * in which case we'll have a 'ztest_tmp' pool lying around | |
5637 | * instead of 'ztest'. Do a blind rename in case this happened. | |
34dc7c2f | 5638 | */ |
428870ff BB |
5639 | kernel_init(FREAD); |
5640 | if (spa_open(zopt_pool, &spa, FTAG) == 0) { | |
5641 | spa_close(spa, FTAG); | |
5642 | } else { | |
5643 | char tmpname[MAXNAMELEN]; | |
5644 | kernel_fini(); | |
5645 | kernel_init(FREAD | FWRITE); | |
5646 | (void) snprintf(tmpname, sizeof (tmpname), "%s_tmp", | |
5647 | zopt_pool); | |
5648 | (void) spa_rename(tmpname, zopt_pool); | |
5649 | } | |
34dc7c2f | 5650 | kernel_fini(); |
34dc7c2f | 5651 | |
428870ff BB |
5652 | ztest_run_zdb(zopt_pool); |
5653 | } | |
34dc7c2f BB |
5654 | |
5655 | if (zopt_verbose >= 1) { | |
5656 | (void) printf("%d killed, %d completed, %.0f%% kill rate\n", | |
5657 | kills, iters - kills, (100.0 * kills) / MAX(1, iters)); | |
5658 | } | |
5659 | ||
5660 | return (0); | |
5661 | } |