]>
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. |
6d974228 | 23 | * Copyright (c) 2011 by Delphix. All rights reserved. |
3541dc6d | 24 | * Copyright 2011 Nexenta Systems, Inc. All rights reserved. |
34dc7c2f BB |
25 | */ |
26 | ||
34dc7c2f BB |
27 | /* |
28 | * The objective of this program is to provide a DMU/ZAP/SPA stress test | |
29 | * that runs entirely in userland, is easy to use, and easy to extend. | |
30 | * | |
31 | * The overall design of the ztest program is as follows: | |
32 | * | |
33 | * (1) For each major functional area (e.g. adding vdevs to a pool, | |
34 | * creating and destroying datasets, reading and writing objects, etc) | |
35 | * we have a simple routine to test that functionality. These | |
36 | * individual routines do not have to do anything "stressful". | |
37 | * | |
38 | * (2) We turn these simple functionality tests into a stress test by | |
39 | * running them all in parallel, with as many threads as desired, | |
40 | * and spread across as many datasets, objects, and vdevs as desired. | |
41 | * | |
42 | * (3) While all this is happening, we inject faults into the pool to | |
43 | * verify that self-healing data really works. | |
44 | * | |
45 | * (4) Every time we open a dataset, we change its checksum and compression | |
46 | * functions. Thus even individual objects vary from block to block | |
47 | * in which checksum they use and whether they're compressed. | |
48 | * | |
49 | * (5) To verify that we never lose on-disk consistency after a crash, | |
50 | * we run the entire test in a child of the main process. | |
51 | * At random times, the child self-immolates with a SIGKILL. | |
52 | * This is the software equivalent of pulling the power cord. | |
53 | * The parent then runs the test again, using the existing | |
54 | * storage pool, as many times as desired. | |
55 | * | |
56 | * (6) To verify that we don't have future leaks or temporal incursions, | |
57 | * many of the functional tests record the transaction group number | |
58 | * as part of their data. When reading old data, they verify that | |
59 | * the transaction group number is less than the current, open txg. | |
60 | * If you add a new test, please do this if applicable. | |
61 | * | |
1e33ac1e BB |
62 | * (7) Threads are created with a reduced stack size, for sanity checking. |
63 | * Therefore, it's important not to allocate huge buffers on the stack. | |
64 | * | |
34dc7c2f BB |
65 | * When run with no arguments, ztest runs for about five minutes and |
66 | * produces no output if successful. To get a little bit of information, | |
67 | * specify -V. To get more information, specify -VV, and so on. | |
68 | * | |
69 | * To turn this into an overnight stress test, use -T to specify run time. | |
70 | * | |
71 | * You can ask more more vdevs [-v], datasets [-d], or threads [-t] | |
72 | * to increase the pool capacity, fanout, and overall stress level. | |
73 | * | |
74 | * The -N(okill) option will suppress kills, so each child runs to completion. | |
75 | * This can be useful when you're trying to distinguish temporal incursions | |
76 | * from plain old race conditions. | |
77 | */ | |
78 | ||
79 | #include <sys/zfs_context.h> | |
80 | #include <sys/spa.h> | |
81 | #include <sys/dmu.h> | |
82 | #include <sys/txg.h> | |
9babb374 | 83 | #include <sys/dbuf.h> |
34dc7c2f | 84 | #include <sys/zap.h> |
34dc7c2f BB |
85 | #include <sys/dmu_objset.h> |
86 | #include <sys/poll.h> | |
87 | #include <sys/stat.h> | |
88 | #include <sys/time.h> | |
89 | #include <sys/wait.h> | |
90 | #include <sys/mman.h> | |
91 | #include <sys/resource.h> | |
92 | #include <sys/zio.h> | |
34dc7c2f | 93 | #include <sys/zil.h> |
428870ff | 94 | #include <sys/zil_impl.h> |
34dc7c2f | 95 | #include <sys/vdev_impl.h> |
b128c09f | 96 | #include <sys/vdev_file.h> |
34dc7c2f | 97 | #include <sys/spa_impl.h> |
428870ff | 98 | #include <sys/metaslab_impl.h> |
34dc7c2f | 99 | #include <sys/dsl_prop.h> |
9babb374 | 100 | #include <sys/dsl_dataset.h> |
428870ff BB |
101 | #include <sys/dsl_scan.h> |
102 | #include <sys/zio_checksum.h> | |
34dc7c2f BB |
103 | #include <sys/refcount.h> |
104 | #include <stdio.h> | |
105 | #include <stdio_ext.h> | |
106 | #include <stdlib.h> | |
107 | #include <unistd.h> | |
108 | #include <signal.h> | |
109 | #include <umem.h> | |
110 | #include <dlfcn.h> | |
111 | #include <ctype.h> | |
112 | #include <math.h> | |
113 | #include <sys/fs/zfs.h> | |
428870ff | 114 | #include <libnvpair.h> |
34dc7c2f BB |
115 | |
116 | static char cmdname[] = "ztest"; | |
117 | static char *zopt_pool = cmdname; | |
118 | ||
119 | static uint64_t zopt_vdevs = 5; | |
120 | static uint64_t zopt_vdevtime; | |
121 | static int zopt_ashift = SPA_MINBLOCKSHIFT; | |
122 | static int zopt_mirrors = 2; | |
123 | static int zopt_raidz = 4; | |
124 | static int zopt_raidz_parity = 1; | |
125 | static size_t zopt_vdev_size = SPA_MINDEVSIZE; | |
126 | static int zopt_datasets = 7; | |
127 | static int zopt_threads = 23; | |
128 | static uint64_t zopt_passtime = 60; /* 60 seconds */ | |
129 | static uint64_t zopt_killrate = 70; /* 70% kill rate */ | |
130 | static int zopt_verbose = 0; | |
131 | static int zopt_init = 1; | |
132 | static char *zopt_dir = "/tmp"; | |
133 | static uint64_t zopt_time = 300; /* 5 minutes */ | |
428870ff BB |
134 | static uint64_t zopt_maxloops = 50; /* max loops during spa_freeze() */ |
135 | ||
136 | #define BT_MAGIC 0x123456789abcdefULL | |
137 | #define MAXFAULTS() (MAX(zs->zs_mirrors, 1) * (zopt_raidz_parity + 1) - 1) | |
138 | ||
139 | enum ztest_io_type { | |
140 | ZTEST_IO_WRITE_TAG, | |
141 | ZTEST_IO_WRITE_PATTERN, | |
142 | ZTEST_IO_WRITE_ZEROES, | |
143 | ZTEST_IO_TRUNCATE, | |
144 | ZTEST_IO_SETATTR, | |
145 | ZTEST_IO_TYPES | |
146 | }; | |
34dc7c2f BB |
147 | |
148 | typedef struct ztest_block_tag { | |
428870ff | 149 | uint64_t bt_magic; |
34dc7c2f BB |
150 | uint64_t bt_objset; |
151 | uint64_t bt_object; | |
152 | uint64_t bt_offset; | |
428870ff | 153 | uint64_t bt_gen; |
34dc7c2f | 154 | uint64_t bt_txg; |
428870ff | 155 | uint64_t bt_crtxg; |
34dc7c2f BB |
156 | } ztest_block_tag_t; |
157 | ||
428870ff BB |
158 | typedef struct bufwad { |
159 | uint64_t bw_index; | |
160 | uint64_t bw_txg; | |
161 | uint64_t bw_data; | |
162 | } bufwad_t; | |
163 | ||
164 | /* | |
165 | * XXX -- fix zfs range locks to be generic so we can use them here. | |
166 | */ | |
167 | typedef enum { | |
168 | RL_READER, | |
169 | RL_WRITER, | |
170 | RL_APPEND | |
171 | } rl_type_t; | |
172 | ||
173 | typedef struct rll { | |
174 | void *rll_writer; | |
175 | int rll_readers; | |
1e33ac1e BB |
176 | kmutex_t rll_lock; |
177 | kcondvar_t rll_cv; | |
428870ff BB |
178 | } rll_t; |
179 | ||
180 | typedef struct rl { | |
181 | uint64_t rl_object; | |
182 | uint64_t rl_offset; | |
183 | uint64_t rl_size; | |
184 | rll_t *rl_lock; | |
185 | } rl_t; | |
186 | ||
187 | #define ZTEST_RANGE_LOCKS 64 | |
188 | #define ZTEST_OBJECT_LOCKS 64 | |
189 | ||
190 | /* | |
191 | * Object descriptor. Used as a template for object lookup/create/remove. | |
192 | */ | |
193 | typedef struct ztest_od { | |
194 | uint64_t od_dir; | |
195 | uint64_t od_object; | |
196 | dmu_object_type_t od_type; | |
197 | dmu_object_type_t od_crtype; | |
198 | uint64_t od_blocksize; | |
199 | uint64_t od_crblocksize; | |
200 | uint64_t od_gen; | |
201 | uint64_t od_crgen; | |
202 | char od_name[MAXNAMELEN]; | |
203 | } ztest_od_t; | |
34dc7c2f | 204 | |
428870ff BB |
205 | /* |
206 | * Per-dataset state. | |
207 | */ | |
208 | typedef struct ztest_ds { | |
209 | objset_t *zd_os; | |
3e31d2b0 | 210 | krwlock_t zd_zilog_lock; |
428870ff BB |
211 | zilog_t *zd_zilog; |
212 | uint64_t zd_seq; | |
213 | ztest_od_t *zd_od; /* debugging aid */ | |
214 | char zd_name[MAXNAMELEN]; | |
1e33ac1e | 215 | kmutex_t zd_dirobj_lock; |
428870ff BB |
216 | rll_t zd_object_lock[ZTEST_OBJECT_LOCKS]; |
217 | rll_t zd_range_lock[ZTEST_RANGE_LOCKS]; | |
218 | } ztest_ds_t; | |
219 | ||
220 | /* | |
221 | * Per-iteration state. | |
222 | */ | |
223 | typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id); | |
224 | ||
225 | typedef struct ztest_info { | |
226 | ztest_func_t *zi_func; /* test function */ | |
227 | uint64_t zi_iters; /* iterations per execution */ | |
228 | uint64_t *zi_interval; /* execute every <interval> seconds */ | |
229 | uint64_t zi_call_count; /* per-pass count */ | |
230 | uint64_t zi_call_time; /* per-pass time */ | |
231 | uint64_t zi_call_next; /* next time to call this function */ | |
232 | } ztest_info_t; | |
34dc7c2f BB |
233 | |
234 | /* | |
235 | * Note: these aren't static because we want dladdr() to work. | |
236 | */ | |
237 | ztest_func_t ztest_dmu_read_write; | |
238 | ztest_func_t ztest_dmu_write_parallel; | |
239 | ztest_func_t ztest_dmu_object_alloc_free; | |
428870ff | 240 | ztest_func_t ztest_dmu_commit_callbacks; |
34dc7c2f BB |
241 | ztest_func_t ztest_zap; |
242 | ztest_func_t ztest_zap_parallel; | |
428870ff | 243 | ztest_func_t ztest_zil_commit; |
3e31d2b0 | 244 | ztest_func_t ztest_zil_remount; |
428870ff | 245 | ztest_func_t ztest_dmu_read_write_zcopy; |
34dc7c2f | 246 | ztest_func_t ztest_dmu_objset_create_destroy; |
428870ff BB |
247 | ztest_func_t ztest_dmu_prealloc; |
248 | ztest_func_t ztest_fzap; | |
34dc7c2f | 249 | ztest_func_t ztest_dmu_snapshot_create_destroy; |
428870ff BB |
250 | ztest_func_t ztest_dsl_prop_get_set; |
251 | ztest_func_t ztest_spa_prop_get_set; | |
34dc7c2f BB |
252 | ztest_func_t ztest_spa_create_destroy; |
253 | ztest_func_t ztest_fault_inject; | |
428870ff BB |
254 | ztest_func_t ztest_ddt_repair; |
255 | ztest_func_t ztest_dmu_snapshot_hold; | |
b128c09f | 256 | ztest_func_t ztest_spa_rename; |
428870ff BB |
257 | ztest_func_t ztest_scrub; |
258 | ztest_func_t ztest_dsl_dataset_promote_busy; | |
34dc7c2f BB |
259 | ztest_func_t ztest_vdev_attach_detach; |
260 | ztest_func_t ztest_vdev_LUN_growth; | |
261 | ztest_func_t ztest_vdev_add_remove; | |
b128c09f | 262 | ztest_func_t ztest_vdev_aux_add_remove; |
428870ff | 263 | ztest_func_t ztest_split_pool; |
3541dc6d | 264 | ztest_func_t ztest_reguid; |
34dc7c2f | 265 | |
428870ff BB |
266 | uint64_t zopt_always = 0ULL * NANOSEC; /* all the time */ |
267 | uint64_t zopt_incessant = 1ULL * NANOSEC / 10; /* every 1/10 second */ | |
268 | uint64_t zopt_often = 1ULL * NANOSEC; /* every second */ | |
269 | uint64_t zopt_sometimes = 10ULL * NANOSEC; /* every 10 seconds */ | |
270 | uint64_t zopt_rarely = 60ULL * NANOSEC; /* every 60 seconds */ | |
34dc7c2f BB |
271 | |
272 | ztest_info_t ztest_info[] = { | |
273 | { ztest_dmu_read_write, 1, &zopt_always }, | |
428870ff | 274 | { ztest_dmu_write_parallel, 10, &zopt_always }, |
34dc7c2f | 275 | { ztest_dmu_object_alloc_free, 1, &zopt_always }, |
428870ff | 276 | { ztest_dmu_commit_callbacks, 1, &zopt_always }, |
34dc7c2f BB |
277 | { ztest_zap, 30, &zopt_always }, |
278 | { ztest_zap_parallel, 100, &zopt_always }, | |
428870ff BB |
279 | { ztest_split_pool, 1, &zopt_always }, |
280 | { ztest_zil_commit, 1, &zopt_incessant }, | |
3e31d2b0 | 281 | { ztest_zil_remount, 1, &zopt_sometimes }, |
428870ff BB |
282 | { ztest_dmu_read_write_zcopy, 1, &zopt_often }, |
283 | { ztest_dmu_objset_create_destroy, 1, &zopt_often }, | |
284 | { ztest_dsl_prop_get_set, 1, &zopt_often }, | |
285 | { ztest_spa_prop_get_set, 1, &zopt_sometimes }, | |
286 | #if 0 | |
287 | { ztest_dmu_prealloc, 1, &zopt_sometimes }, | |
288 | #endif | |
289 | { ztest_fzap, 1, &zopt_sometimes }, | |
290 | { ztest_dmu_snapshot_create_destroy, 1, &zopt_sometimes }, | |
291 | { ztest_spa_create_destroy, 1, &zopt_sometimes }, | |
34dc7c2f | 292 | { ztest_fault_inject, 1, &zopt_sometimes }, |
428870ff BB |
293 | { ztest_ddt_repair, 1, &zopt_sometimes }, |
294 | { ztest_dmu_snapshot_hold, 1, &zopt_sometimes }, | |
d1352457 ED |
295 | /* |
296 | * The reguid test is currently broken. Disable it until | |
297 | * we get around to fixing it. | |
298 | */ | |
299 | #if 0 | |
3541dc6d | 300 | { ztest_reguid, 1, &zopt_sometimes }, |
d1352457 | 301 | #endif |
34dc7c2f | 302 | { ztest_spa_rename, 1, &zopt_rarely }, |
428870ff | 303 | { ztest_scrub, 1, &zopt_rarely }, |
9babb374 | 304 | { ztest_dsl_dataset_promote_busy, 1, &zopt_rarely }, |
428870ff BB |
305 | { ztest_vdev_attach_detach, 1, &zopt_rarely }, |
306 | { ztest_vdev_LUN_growth, 1, &zopt_rarely }, | |
307 | { ztest_vdev_add_remove, 1, &zopt_vdevtime }, | |
b128c09f | 308 | { ztest_vdev_aux_add_remove, 1, &zopt_vdevtime }, |
34dc7c2f BB |
309 | }; |
310 | ||
311 | #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t)) | |
312 | ||
428870ff BB |
313 | /* |
314 | * The following struct is used to hold a list of uncalled commit callbacks. | |
315 | * The callbacks are ordered by txg number. | |
316 | */ | |
317 | typedef struct ztest_cb_list { | |
1e33ac1e BB |
318 | kmutex_t zcl_callbacks_lock; |
319 | list_t zcl_callbacks; | |
428870ff | 320 | } ztest_cb_list_t; |
34dc7c2f BB |
321 | |
322 | /* | |
323 | * Stuff we need to share writably between parent and child. | |
324 | */ | |
325 | typedef struct ztest_shared { | |
428870ff BB |
326 | char *zs_pool; |
327 | spa_t *zs_spa; | |
328 | hrtime_t zs_proc_start; | |
329 | hrtime_t zs_proc_stop; | |
330 | hrtime_t zs_thread_start; | |
331 | hrtime_t zs_thread_stop; | |
332 | hrtime_t zs_thread_kill; | |
34dc7c2f | 333 | uint64_t zs_enospc_count; |
428870ff BB |
334 | uint64_t zs_vdev_next_leaf; |
335 | uint64_t zs_vdev_aux; | |
34dc7c2f BB |
336 | uint64_t zs_alloc; |
337 | uint64_t zs_space; | |
3541dc6d | 338 | uint64_t zs_guid; |
1e33ac1e BB |
339 | kmutex_t zs_vdev_lock; |
340 | krwlock_t zs_name_lock; | |
34dc7c2f | 341 | ztest_info_t zs_info[ZTEST_FUNCS]; |
428870ff BB |
342 | uint64_t zs_splits; |
343 | uint64_t zs_mirrors; | |
344 | ztest_ds_t zs_zd[]; | |
34dc7c2f BB |
345 | } ztest_shared_t; |
346 | ||
428870ff BB |
347 | #define ID_PARALLEL -1ULL |
348 | ||
34dc7c2f | 349 | static char ztest_dev_template[] = "%s/%s.%llua"; |
b128c09f | 350 | static char ztest_aux_template[] = "%s/%s.%s.%llu"; |
428870ff BB |
351 | ztest_shared_t *ztest_shared; |
352 | uint64_t *ztest_seq; | |
34dc7c2f BB |
353 | |
354 | static int ztest_random_fd; | |
355 | static int ztest_dump_core = 1; | |
356 | ||
b128c09f | 357 | static boolean_t ztest_exiting; |
34dc7c2f | 358 | |
428870ff BB |
359 | /* Global commit callback list */ |
360 | static ztest_cb_list_t zcl; | |
090ff092 RC |
361 | /* Commit cb delay */ |
362 | static uint64_t zc_min_txg_delay = UINT64_MAX; | |
363 | static int zc_cb_counter = 0; | |
364 | ||
365 | /* | |
366 | * Minimum number of commit callbacks that need to be registered for us to check | |
367 | * whether the minimum txg delay is acceptable. | |
368 | */ | |
369 | #define ZTEST_COMMIT_CB_MIN_REG 100 | |
370 | ||
371 | /* | |
372 | * If a number of txgs equal to this threshold have been created after a commit | |
373 | * callback has been registered but not called, then we assume there is an | |
374 | * implementation bug. | |
375 | */ | |
376 | #define ZTEST_COMMIT_CB_THRESH (TXG_CONCURRENT_STATES + 1000) | |
428870ff | 377 | |
34dc7c2f | 378 | extern uint64_t metaslab_gang_bang; |
9babb374 | 379 | extern uint64_t metaslab_df_alloc_threshold; |
428870ff | 380 | static uint64_t metaslab_sz; |
34dc7c2f | 381 | |
428870ff BB |
382 | enum ztest_object { |
383 | ZTEST_META_DNODE = 0, | |
384 | ZTEST_DIROBJ, | |
385 | ZTEST_OBJECTS | |
386 | }; | |
34dc7c2f BB |
387 | |
388 | static void usage(boolean_t) __NORETURN; | |
389 | ||
390 | /* | |
391 | * These libumem hooks provide a reasonable set of defaults for the allocator's | |
392 | * debugging facilities. | |
393 | */ | |
394 | const char * | |
0bc8fd78 | 395 | _umem_debug_init(void) |
34dc7c2f BB |
396 | { |
397 | return ("default,verbose"); /* $UMEM_DEBUG setting */ | |
398 | } | |
399 | ||
400 | const char * | |
401 | _umem_logging_init(void) | |
402 | { | |
403 | return ("fail,contents"); /* $UMEM_LOGGING setting */ | |
404 | } | |
405 | ||
406 | #define FATAL_MSG_SZ 1024 | |
407 | ||
408 | char *fatal_msg; | |
409 | ||
410 | static void | |
411 | fatal(int do_perror, char *message, ...) | |
412 | { | |
413 | va_list args; | |
414 | int save_errno = errno; | |
40b84e7a | 415 | char *buf; |
34dc7c2f BB |
416 | |
417 | (void) fflush(stdout); | |
40b84e7a | 418 | buf = umem_alloc(FATAL_MSG_SZ, UMEM_NOFAIL); |
34dc7c2f BB |
419 | |
420 | va_start(args, message); | |
421 | (void) sprintf(buf, "ztest: "); | |
422 | /* LINTED */ | |
423 | (void) vsprintf(buf + strlen(buf), message, args); | |
424 | va_end(args); | |
425 | if (do_perror) { | |
426 | (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf), | |
427 | ": %s", strerror(save_errno)); | |
428 | } | |
429 | (void) fprintf(stderr, "%s\n", buf); | |
430 | fatal_msg = buf; /* to ease debugging */ | |
431 | if (ztest_dump_core) | |
432 | abort(); | |
433 | exit(3); | |
434 | } | |
435 | ||
436 | static int | |
437 | str2shift(const char *buf) | |
438 | { | |
439 | const char *ends = "BKMGTPEZ"; | |
440 | int i; | |
441 | ||
442 | if (buf[0] == '\0') | |
443 | return (0); | |
444 | for (i = 0; i < strlen(ends); i++) { | |
445 | if (toupper(buf[0]) == ends[i]) | |
446 | break; | |
447 | } | |
448 | if (i == strlen(ends)) { | |
449 | (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", | |
450 | buf); | |
451 | usage(B_FALSE); | |
452 | } | |
453 | if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) { | |
454 | return (10*i); | |
455 | } | |
456 | (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf); | |
457 | usage(B_FALSE); | |
458 | /* NOTREACHED */ | |
459 | } | |
460 | ||
461 | static uint64_t | |
462 | nicenumtoull(const char *buf) | |
463 | { | |
464 | char *end; | |
465 | uint64_t val; | |
466 | ||
467 | val = strtoull(buf, &end, 0); | |
468 | if (end == buf) { | |
469 | (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf); | |
470 | usage(B_FALSE); | |
471 | } else if (end[0] == '.') { | |
472 | double fval = strtod(buf, &end); | |
473 | fval *= pow(2, str2shift(end)); | |
474 | if (fval > UINT64_MAX) { | |
475 | (void) fprintf(stderr, "ztest: value too large: %s\n", | |
476 | buf); | |
477 | usage(B_FALSE); | |
478 | } | |
479 | val = (uint64_t)fval; | |
480 | } else { | |
481 | int shift = str2shift(end); | |
482 | if (shift >= 64 || (val << shift) >> shift != val) { | |
483 | (void) fprintf(stderr, "ztest: value too large: %s\n", | |
484 | buf); | |
485 | usage(B_FALSE); | |
486 | } | |
487 | val <<= shift; | |
488 | } | |
489 | return (val); | |
490 | } | |
491 | ||
492 | static void | |
493 | usage(boolean_t requested) | |
494 | { | |
495 | char nice_vdev_size[10]; | |
496 | char nice_gang_bang[10]; | |
497 | FILE *fp = requested ? stdout : stderr; | |
498 | ||
499 | nicenum(zopt_vdev_size, nice_vdev_size); | |
500 | nicenum(metaslab_gang_bang, nice_gang_bang); | |
501 | ||
502 | (void) fprintf(fp, "Usage: %s\n" | |
503 | "\t[-v vdevs (default: %llu)]\n" | |
504 | "\t[-s size_of_each_vdev (default: %s)]\n" | |
428870ff | 505 | "\t[-a alignment_shift (default: %d)] use 0 for random\n" |
34dc7c2f BB |
506 | "\t[-m mirror_copies (default: %d)]\n" |
507 | "\t[-r raidz_disks (default: %d)]\n" | |
508 | "\t[-R raidz_parity (default: %d)]\n" | |
509 | "\t[-d datasets (default: %d)]\n" | |
510 | "\t[-t threads (default: %d)]\n" | |
511 | "\t[-g gang_block_threshold (default: %s)]\n" | |
428870ff BB |
512 | "\t[-i init_count (default: %d)] initialize pool i times\n" |
513 | "\t[-k kill_percentage (default: %llu%%)]\n" | |
34dc7c2f | 514 | "\t[-p pool_name (default: %s)]\n" |
428870ff BB |
515 | "\t[-f dir (default: %s)] file directory for vdev files\n" |
516 | "\t[-V] verbose (use multiple times for ever more blather)\n" | |
517 | "\t[-E] use existing pool instead of creating new one\n" | |
518 | "\t[-T time (default: %llu sec)] total run time\n" | |
519 | "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n" | |
520 | "\t[-P passtime (default: %llu sec)] time per pass\n" | |
34dc7c2f BB |
521 | "\t[-h] (print help)\n" |
522 | "", | |
523 | cmdname, | |
524 | (u_longlong_t)zopt_vdevs, /* -v */ | |
525 | nice_vdev_size, /* -s */ | |
526 | zopt_ashift, /* -a */ | |
527 | zopt_mirrors, /* -m */ | |
528 | zopt_raidz, /* -r */ | |
529 | zopt_raidz_parity, /* -R */ | |
530 | zopt_datasets, /* -d */ | |
531 | zopt_threads, /* -t */ | |
532 | nice_gang_bang, /* -g */ | |
533 | zopt_init, /* -i */ | |
534 | (u_longlong_t)zopt_killrate, /* -k */ | |
535 | zopt_pool, /* -p */ | |
536 | zopt_dir, /* -f */ | |
537 | (u_longlong_t)zopt_time, /* -T */ | |
428870ff | 538 | (u_longlong_t)zopt_maxloops, /* -F */ |
b128c09f | 539 | (u_longlong_t)zopt_passtime); /* -P */ |
34dc7c2f BB |
540 | exit(requested ? 0 : 1); |
541 | } | |
542 | ||
34dc7c2f BB |
543 | static void |
544 | process_options(int argc, char **argv) | |
545 | { | |
546 | int opt; | |
547 | uint64_t value; | |
548 | ||
549 | /* By default, test gang blocks for blocks 32K and greater */ | |
550 | metaslab_gang_bang = 32 << 10; | |
551 | ||
34dc7c2f | 552 | while ((opt = getopt(argc, argv, |
428870ff | 553 | "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:")) != EOF) { |
34dc7c2f BB |
554 | value = 0; |
555 | switch (opt) { | |
556 | case 'v': | |
557 | case 's': | |
558 | case 'a': | |
559 | case 'm': | |
560 | case 'r': | |
561 | case 'R': | |
562 | case 'd': | |
563 | case 't': | |
564 | case 'g': | |
565 | case 'i': | |
566 | case 'k': | |
567 | case 'T': | |
568 | case 'P': | |
428870ff | 569 | case 'F': |
34dc7c2f BB |
570 | value = nicenumtoull(optarg); |
571 | } | |
572 | switch (opt) { | |
573 | case 'v': | |
574 | zopt_vdevs = value; | |
575 | break; | |
576 | case 's': | |
577 | zopt_vdev_size = MAX(SPA_MINDEVSIZE, value); | |
578 | break; | |
579 | case 'a': | |
580 | zopt_ashift = value; | |
581 | break; | |
582 | case 'm': | |
583 | zopt_mirrors = value; | |
584 | break; | |
585 | case 'r': | |
586 | zopt_raidz = MAX(1, value); | |
587 | break; | |
588 | case 'R': | |
45d1cae3 | 589 | zopt_raidz_parity = MIN(MAX(value, 1), 3); |
34dc7c2f BB |
590 | break; |
591 | case 'd': | |
592 | zopt_datasets = MAX(1, value); | |
593 | break; | |
594 | case 't': | |
595 | zopt_threads = MAX(1, value); | |
596 | break; | |
597 | case 'g': | |
598 | metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1, value); | |
599 | break; | |
600 | case 'i': | |
601 | zopt_init = value; | |
602 | break; | |
603 | case 'k': | |
604 | zopt_killrate = value; | |
605 | break; | |
606 | case 'p': | |
607 | zopt_pool = strdup(optarg); | |
608 | break; | |
609 | case 'f': | |
610 | zopt_dir = strdup(optarg); | |
611 | break; | |
612 | case 'V': | |
613 | zopt_verbose++; | |
614 | break; | |
615 | case 'E': | |
616 | zopt_init = 0; | |
617 | break; | |
618 | case 'T': | |
619 | zopt_time = value; | |
620 | break; | |
621 | case 'P': | |
622 | zopt_passtime = MAX(1, value); | |
623 | break; | |
428870ff BB |
624 | case 'F': |
625 | zopt_maxloops = MAX(1, value); | |
626 | break; | |
34dc7c2f BB |
627 | case 'h': |
628 | usage(B_TRUE); | |
629 | break; | |
630 | case '?': | |
631 | default: | |
632 | usage(B_FALSE); | |
633 | break; | |
634 | } | |
635 | } | |
636 | ||
637 | zopt_raidz_parity = MIN(zopt_raidz_parity, zopt_raidz - 1); | |
638 | ||
428870ff BB |
639 | zopt_vdevtime = (zopt_vdevs > 0 ? zopt_time * NANOSEC / zopt_vdevs : |
640 | UINT64_MAX >> 2); | |
641 | } | |
642 | ||
643 | static void | |
644 | ztest_kill(ztest_shared_t *zs) | |
645 | { | |
646 | zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(zs->zs_spa)); | |
647 | zs->zs_space = metaslab_class_get_space(spa_normal_class(zs->zs_spa)); | |
648 | (void) kill(getpid(), SIGKILL); | |
649 | } | |
650 | ||
651 | static uint64_t | |
652 | ztest_random(uint64_t range) | |
653 | { | |
654 | uint64_t r; | |
655 | ||
656 | if (range == 0) | |
657 | return (0); | |
658 | ||
659 | if (read(ztest_random_fd, &r, sizeof (r)) != sizeof (r)) | |
660 | fatal(1, "short read from /dev/urandom"); | |
661 | ||
662 | return (r % range); | |
663 | } | |
664 | ||
665 | /* ARGSUSED */ | |
666 | static void | |
667 | ztest_record_enospc(const char *s) | |
668 | { | |
669 | ztest_shared->zs_enospc_count++; | |
34dc7c2f BB |
670 | } |
671 | ||
672 | static uint64_t | |
673 | ztest_get_ashift(void) | |
674 | { | |
675 | if (zopt_ashift == 0) | |
676 | return (SPA_MINBLOCKSHIFT + ztest_random(3)); | |
677 | return (zopt_ashift); | |
678 | } | |
679 | ||
680 | static nvlist_t * | |
b128c09f | 681 | make_vdev_file(char *path, char *aux, size_t size, uint64_t ashift) |
34dc7c2f | 682 | { |
40b84e7a | 683 | char *pathbuf; |
34dc7c2f | 684 | uint64_t vdev; |
34dc7c2f BB |
685 | nvlist_t *file; |
686 | ||
40b84e7a BB |
687 | pathbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); |
688 | ||
b128c09f BB |
689 | if (ashift == 0) |
690 | ashift = ztest_get_ashift(); | |
691 | ||
692 | if (path == NULL) { | |
693 | path = pathbuf; | |
694 | ||
695 | if (aux != NULL) { | |
696 | vdev = ztest_shared->zs_vdev_aux; | |
697 | (void) sprintf(path, ztest_aux_template, | |
698 | zopt_dir, zopt_pool, aux, vdev); | |
699 | } else { | |
428870ff | 700 | vdev = ztest_shared->zs_vdev_next_leaf++; |
b128c09f BB |
701 | (void) sprintf(path, ztest_dev_template, |
702 | zopt_dir, zopt_pool, vdev); | |
703 | } | |
704 | } | |
34dc7c2f | 705 | |
b128c09f BB |
706 | if (size != 0) { |
707 | int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666); | |
34dc7c2f | 708 | if (fd == -1) |
b128c09f | 709 | fatal(1, "can't open %s", path); |
34dc7c2f | 710 | if (ftruncate(fd, size) != 0) |
b128c09f | 711 | fatal(1, "can't ftruncate %s", path); |
34dc7c2f BB |
712 | (void) close(fd); |
713 | } | |
714 | ||
715 | VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0); | |
716 | VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0); | |
b128c09f | 717 | VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0); |
34dc7c2f | 718 | VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0); |
40b84e7a | 719 | umem_free(pathbuf, MAXPATHLEN); |
34dc7c2f BB |
720 | |
721 | return (file); | |
722 | } | |
723 | ||
724 | static nvlist_t * | |
b128c09f | 725 | make_vdev_raidz(char *path, char *aux, size_t size, uint64_t ashift, int r) |
34dc7c2f BB |
726 | { |
727 | nvlist_t *raidz, **child; | |
728 | int c; | |
729 | ||
730 | if (r < 2) | |
b128c09f | 731 | return (make_vdev_file(path, aux, size, ashift)); |
34dc7c2f BB |
732 | child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL); |
733 | ||
734 | for (c = 0; c < r; c++) | |
b128c09f | 735 | child[c] = make_vdev_file(path, aux, size, ashift); |
34dc7c2f BB |
736 | |
737 | VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0); | |
738 | VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE, | |
739 | VDEV_TYPE_RAIDZ) == 0); | |
740 | VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY, | |
741 | zopt_raidz_parity) == 0); | |
742 | VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN, | |
743 | child, r) == 0); | |
744 | ||
745 | for (c = 0; c < r; c++) | |
746 | nvlist_free(child[c]); | |
747 | ||
748 | umem_free(child, r * sizeof (nvlist_t *)); | |
749 | ||
750 | return (raidz); | |
751 | } | |
752 | ||
753 | static nvlist_t * | |
b128c09f BB |
754 | make_vdev_mirror(char *path, char *aux, size_t size, uint64_t ashift, |
755 | int r, int m) | |
34dc7c2f BB |
756 | { |
757 | nvlist_t *mirror, **child; | |
758 | int c; | |
759 | ||
760 | if (m < 1) | |
b128c09f | 761 | return (make_vdev_raidz(path, aux, size, ashift, r)); |
34dc7c2f BB |
762 | |
763 | child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL); | |
764 | ||
765 | for (c = 0; c < m; c++) | |
b128c09f | 766 | child[c] = make_vdev_raidz(path, aux, size, ashift, r); |
34dc7c2f BB |
767 | |
768 | VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0); | |
769 | VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE, | |
770 | VDEV_TYPE_MIRROR) == 0); | |
771 | VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN, | |
772 | child, m) == 0); | |
34dc7c2f BB |
773 | |
774 | for (c = 0; c < m; c++) | |
775 | nvlist_free(child[c]); | |
776 | ||
777 | umem_free(child, m * sizeof (nvlist_t *)); | |
778 | ||
779 | return (mirror); | |
780 | } | |
781 | ||
782 | static nvlist_t * | |
b128c09f BB |
783 | make_vdev_root(char *path, char *aux, size_t size, uint64_t ashift, |
784 | int log, int r, int m, int t) | |
34dc7c2f BB |
785 | { |
786 | nvlist_t *root, **child; | |
787 | int c; | |
788 | ||
789 | ASSERT(t > 0); | |
790 | ||
791 | child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL); | |
792 | ||
b128c09f BB |
793 | for (c = 0; c < t; c++) { |
794 | child[c] = make_vdev_mirror(path, aux, size, ashift, r, m); | |
795 | VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG, | |
796 | log) == 0); | |
797 | } | |
34dc7c2f BB |
798 | |
799 | VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0); | |
800 | VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0); | |
b128c09f | 801 | VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN, |
34dc7c2f BB |
802 | child, t) == 0); |
803 | ||
804 | for (c = 0; c < t; c++) | |
805 | nvlist_free(child[c]); | |
806 | ||
807 | umem_free(child, t * sizeof (nvlist_t *)); | |
808 | ||
809 | return (root); | |
810 | } | |
811 | ||
428870ff BB |
812 | static int |
813 | ztest_random_blocksize(void) | |
34dc7c2f | 814 | { |
428870ff BB |
815 | return (1 << (SPA_MINBLOCKSHIFT + |
816 | ztest_random(SPA_MAXBLOCKSHIFT - SPA_MINBLOCKSHIFT + 1))); | |
817 | } | |
34dc7c2f | 818 | |
428870ff BB |
819 | static int |
820 | ztest_random_ibshift(void) | |
821 | { | |
822 | return (DN_MIN_INDBLKSHIFT + | |
823 | ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1)); | |
34dc7c2f BB |
824 | } |
825 | ||
428870ff BB |
826 | static uint64_t |
827 | ztest_random_vdev_top(spa_t *spa, boolean_t log_ok) | |
34dc7c2f | 828 | { |
428870ff BB |
829 | uint64_t top; |
830 | vdev_t *rvd = spa->spa_root_vdev; | |
831 | vdev_t *tvd; | |
34dc7c2f | 832 | |
428870ff | 833 | ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); |
34dc7c2f | 834 | |
428870ff BB |
835 | do { |
836 | top = ztest_random(rvd->vdev_children); | |
837 | tvd = rvd->vdev_child[top]; | |
838 | } while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) || | |
839 | tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL); | |
34dc7c2f | 840 | |
428870ff | 841 | return (top); |
34dc7c2f BB |
842 | } |
843 | ||
428870ff BB |
844 | static uint64_t |
845 | ztest_random_dsl_prop(zfs_prop_t prop) | |
34dc7c2f | 846 | { |
428870ff BB |
847 | uint64_t value; |
848 | ||
849 | do { | |
850 | value = zfs_prop_random_value(prop, ztest_random(-1ULL)); | |
851 | } while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF); | |
852 | ||
853 | return (value); | |
34dc7c2f BB |
854 | } |
855 | ||
34dc7c2f | 856 | static int |
428870ff BB |
857 | ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value, |
858 | boolean_t inherit) | |
34dc7c2f | 859 | { |
428870ff BB |
860 | const char *propname = zfs_prop_to_name(prop); |
861 | const char *valname; | |
40b84e7a | 862 | char *setpoint; |
428870ff | 863 | uint64_t curval; |
34dc7c2f BB |
864 | int error; |
865 | ||
428870ff BB |
866 | error = dsl_prop_set(osname, propname, |
867 | (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), | |
868 | sizeof (value), 1, &value); | |
34dc7c2f | 869 | |
428870ff BB |
870 | if (error == ENOSPC) { |
871 | ztest_record_enospc(FTAG); | |
34dc7c2f BB |
872 | return (error); |
873 | } | |
34dc7c2f | 874 | ASSERT3U(error, ==, 0); |
34dc7c2f | 875 | |
40b84e7a | 876 | setpoint = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); |
428870ff BB |
877 | VERIFY3U(dsl_prop_get(osname, propname, sizeof (curval), |
878 | 1, &curval, setpoint), ==, 0); | |
879 | ||
880 | if (zopt_verbose >= 6) { | |
881 | VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0); | |
882 | (void) printf("%s %s = %s at '%s'\n", | |
883 | osname, propname, valname, setpoint); | |
34dc7c2f | 884 | } |
40b84e7a | 885 | umem_free(setpoint, MAXPATHLEN); |
34dc7c2f BB |
886 | |
887 | return (error); | |
888 | } | |
889 | ||
890 | static int | |
428870ff | 891 | ztest_spa_prop_set_uint64(ztest_shared_t *zs, zpool_prop_t prop, uint64_t value) |
34dc7c2f | 892 | { |
428870ff BB |
893 | spa_t *spa = zs->zs_spa; |
894 | nvlist_t *props = NULL; | |
34dc7c2f BB |
895 | int error; |
896 | ||
428870ff BB |
897 | VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0); |
898 | VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0); | |
34dc7c2f | 899 | |
428870ff BB |
900 | error = spa_prop_set(spa, props); |
901 | ||
902 | nvlist_free(props); | |
903 | ||
904 | if (error == ENOSPC) { | |
905 | ztest_record_enospc(FTAG); | |
34dc7c2f BB |
906 | return (error); |
907 | } | |
428870ff | 908 | ASSERT3U(error, ==, 0); |
34dc7c2f BB |
909 | |
910 | return (error); | |
911 | } | |
912 | ||
428870ff BB |
913 | static void |
914 | ztest_rll_init(rll_t *rll) | |
915 | { | |
916 | rll->rll_writer = NULL; | |
917 | rll->rll_readers = 0; | |
1e33ac1e BB |
918 | mutex_init(&rll->rll_lock, NULL, MUTEX_DEFAULT, NULL); |
919 | cv_init(&rll->rll_cv, NULL, CV_DEFAULT, NULL); | |
428870ff | 920 | } |
34dc7c2f | 921 | |
428870ff BB |
922 | static void |
923 | ztest_rll_destroy(rll_t *rll) | |
34dc7c2f | 924 | { |
428870ff BB |
925 | ASSERT(rll->rll_writer == NULL); |
926 | ASSERT(rll->rll_readers == 0); | |
1e33ac1e BB |
927 | mutex_destroy(&rll->rll_lock); |
928 | cv_destroy(&rll->rll_cv); | |
428870ff | 929 | } |
34dc7c2f | 930 | |
428870ff BB |
931 | static void |
932 | ztest_rll_lock(rll_t *rll, rl_type_t type) | |
933 | { | |
1e33ac1e | 934 | mutex_enter(&rll->rll_lock); |
34dc7c2f | 935 | |
428870ff BB |
936 | if (type == RL_READER) { |
937 | while (rll->rll_writer != NULL) | |
1e33ac1e | 938 | (void) cv_wait(&rll->rll_cv, &rll->rll_lock); |
428870ff BB |
939 | rll->rll_readers++; |
940 | } else { | |
941 | while (rll->rll_writer != NULL || rll->rll_readers) | |
1e33ac1e | 942 | (void) cv_wait(&rll->rll_cv, &rll->rll_lock); |
428870ff BB |
943 | rll->rll_writer = curthread; |
944 | } | |
34dc7c2f | 945 | |
1e33ac1e | 946 | mutex_exit(&rll->rll_lock); |
428870ff | 947 | } |
34dc7c2f | 948 | |
428870ff BB |
949 | static void |
950 | ztest_rll_unlock(rll_t *rll) | |
951 | { | |
1e33ac1e | 952 | mutex_enter(&rll->rll_lock); |
34dc7c2f | 953 | |
428870ff BB |
954 | if (rll->rll_writer) { |
955 | ASSERT(rll->rll_readers == 0); | |
956 | rll->rll_writer = NULL; | |
957 | } else { | |
958 | ASSERT(rll->rll_readers != 0); | |
959 | ASSERT(rll->rll_writer == NULL); | |
960 | rll->rll_readers--; | |
961 | } | |
34dc7c2f | 962 | |
428870ff | 963 | if (rll->rll_writer == NULL && rll->rll_readers == 0) |
1e33ac1e | 964 | cv_broadcast(&rll->rll_cv); |
428870ff | 965 | |
1e33ac1e | 966 | mutex_exit(&rll->rll_lock); |
34dc7c2f BB |
967 | } |
968 | ||
428870ff BB |
969 | static void |
970 | ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type) | |
b128c09f | 971 | { |
428870ff | 972 | rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)]; |
b128c09f | 973 | |
428870ff BB |
974 | ztest_rll_lock(rll, type); |
975 | } | |
b128c09f | 976 | |
428870ff BB |
977 | static void |
978 | ztest_object_unlock(ztest_ds_t *zd, uint64_t object) | |
979 | { | |
980 | rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)]; | |
b128c09f | 981 | |
428870ff | 982 | ztest_rll_unlock(rll); |
b128c09f BB |
983 | } |
984 | ||
428870ff BB |
985 | static rl_t * |
986 | ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset, | |
987 | uint64_t size, rl_type_t type) | |
34dc7c2f | 988 | { |
428870ff BB |
989 | uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1)); |
990 | rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)]; | |
991 | rl_t *rl; | |
34dc7c2f | 992 | |
428870ff BB |
993 | rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL); |
994 | rl->rl_object = object; | |
995 | rl->rl_offset = offset; | |
996 | rl->rl_size = size; | |
997 | rl->rl_lock = rll; | |
34dc7c2f | 998 | |
428870ff BB |
999 | ztest_rll_lock(rll, type); |
1000 | ||
1001 | return (rl); | |
1002 | } | |
34dc7c2f | 1003 | |
428870ff BB |
1004 | static void |
1005 | ztest_range_unlock(rl_t *rl) | |
1006 | { | |
1007 | rll_t *rll = rl->rl_lock; | |
34dc7c2f | 1008 | |
428870ff | 1009 | ztest_rll_unlock(rll); |
34dc7c2f | 1010 | |
428870ff BB |
1011 | umem_free(rl, sizeof (*rl)); |
1012 | } | |
34dc7c2f | 1013 | |
428870ff BB |
1014 | static void |
1015 | ztest_zd_init(ztest_ds_t *zd, objset_t *os) | |
1016 | { | |
1017 | zd->zd_os = os; | |
1018 | zd->zd_zilog = dmu_objset_zil(os); | |
1019 | zd->zd_seq = 0; | |
1020 | dmu_objset_name(os, zd->zd_name); | |
d6320ddb | 1021 | int l; |
428870ff | 1022 | |
3e31d2b0 | 1023 | rw_init(&zd->zd_zilog_lock, NULL, RW_DEFAULT, NULL); |
1e33ac1e | 1024 | mutex_init(&zd->zd_dirobj_lock, NULL, MUTEX_DEFAULT, NULL); |
34dc7c2f | 1025 | |
d6320ddb | 1026 | for (l = 0; l < ZTEST_OBJECT_LOCKS; l++) |
428870ff | 1027 | ztest_rll_init(&zd->zd_object_lock[l]); |
34dc7c2f | 1028 | |
d6320ddb | 1029 | for (l = 0; l < ZTEST_RANGE_LOCKS; l++) |
428870ff | 1030 | ztest_rll_init(&zd->zd_range_lock[l]); |
34dc7c2f BB |
1031 | } |
1032 | ||
428870ff BB |
1033 | static void |
1034 | ztest_zd_fini(ztest_ds_t *zd) | |
34dc7c2f | 1035 | { |
d6320ddb BB |
1036 | int l; |
1037 | ||
1e33ac1e | 1038 | mutex_destroy(&zd->zd_dirobj_lock); |
3e31d2b0 | 1039 | rw_destroy(&zd->zd_zilog_lock); |
34dc7c2f | 1040 | |
d6320ddb | 1041 | for (l = 0; l < ZTEST_OBJECT_LOCKS; l++) |
428870ff | 1042 | ztest_rll_destroy(&zd->zd_object_lock[l]); |
b128c09f | 1043 | |
d6320ddb | 1044 | for (l = 0; l < ZTEST_RANGE_LOCKS; l++) |
428870ff BB |
1045 | ztest_rll_destroy(&zd->zd_range_lock[l]); |
1046 | } | |
b128c09f | 1047 | |
428870ff | 1048 | #define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT) |
b128c09f | 1049 | |
428870ff BB |
1050 | static uint64_t |
1051 | ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag) | |
1052 | { | |
1053 | uint64_t txg; | |
1054 | int error; | |
1055 | ||
1056 | /* | |
1057 | * Attempt to assign tx to some transaction group. | |
1058 | */ | |
1059 | error = dmu_tx_assign(tx, txg_how); | |
1060 | if (error) { | |
1061 | if (error == ERESTART) { | |
1062 | ASSERT(txg_how == TXG_NOWAIT); | |
1063 | dmu_tx_wait(tx); | |
1064 | } else { | |
1065 | ASSERT3U(error, ==, ENOSPC); | |
1066 | ztest_record_enospc(tag); | |
1067 | } | |
1068 | dmu_tx_abort(tx); | |
1069 | return (0); | |
1070 | } | |
1071 | txg = dmu_tx_get_txg(tx); | |
1072 | ASSERT(txg != 0); | |
1073 | return (txg); | |
1074 | } | |
1075 | ||
1076 | static void | |
1077 | ztest_pattern_set(void *buf, uint64_t size, uint64_t value) | |
1078 | { | |
1079 | uint64_t *ip = buf; | |
1080 | uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size); | |
1081 | ||
1082 | while (ip < ip_end) | |
1083 | *ip++ = value; | |
1084 | } | |
1085 | ||
1fde1e37 | 1086 | #ifndef NDEBUG |
428870ff BB |
1087 | static boolean_t |
1088 | ztest_pattern_match(void *buf, uint64_t size, uint64_t value) | |
1089 | { | |
1090 | uint64_t *ip = buf; | |
1091 | uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size); | |
1092 | uint64_t diff = 0; | |
1093 | ||
1094 | while (ip < ip_end) | |
1095 | diff |= (value - *ip++); | |
1096 | ||
1097 | return (diff == 0); | |
1098 | } | |
1fde1e37 | 1099 | #endif |
428870ff BB |
1100 | |
1101 | static void | |
1102 | ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object, | |
1103 | uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg) | |
1104 | { | |
1105 | bt->bt_magic = BT_MAGIC; | |
1106 | bt->bt_objset = dmu_objset_id(os); | |
1107 | bt->bt_object = object; | |
1108 | bt->bt_offset = offset; | |
1109 | bt->bt_gen = gen; | |
1110 | bt->bt_txg = txg; | |
1111 | bt->bt_crtxg = crtxg; | |
1112 | } | |
1113 | ||
1114 | static void | |
1115 | ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object, | |
1116 | uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg) | |
1117 | { | |
1118 | ASSERT(bt->bt_magic == BT_MAGIC); | |
1119 | ASSERT(bt->bt_objset == dmu_objset_id(os)); | |
1120 | ASSERT(bt->bt_object == object); | |
1121 | ASSERT(bt->bt_offset == offset); | |
1122 | ASSERT(bt->bt_gen <= gen); | |
1123 | ASSERT(bt->bt_txg <= txg); | |
1124 | ASSERT(bt->bt_crtxg == crtxg); | |
1125 | } | |
1126 | ||
1127 | static ztest_block_tag_t * | |
1128 | ztest_bt_bonus(dmu_buf_t *db) | |
1129 | { | |
1130 | dmu_object_info_t doi; | |
1131 | ztest_block_tag_t *bt; | |
1132 | ||
1133 | dmu_object_info_from_db(db, &doi); | |
1134 | ASSERT3U(doi.doi_bonus_size, <=, db->db_size); | |
1135 | ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt)); | |
1136 | bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt)); | |
1137 | ||
1138 | return (bt); | |
1139 | } | |
1140 | ||
1141 | /* | |
1142 | * ZIL logging ops | |
1143 | */ | |
1144 | ||
1145 | #define lrz_type lr_mode | |
1146 | #define lrz_blocksize lr_uid | |
1147 | #define lrz_ibshift lr_gid | |
1148 | #define lrz_bonustype lr_rdev | |
1149 | #define lrz_bonuslen lr_crtime[1] | |
1150 | ||
572e2857 | 1151 | static void |
428870ff BB |
1152 | ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr) |
1153 | { | |
1154 | char *name = (void *)(lr + 1); /* name follows lr */ | |
1155 | size_t namesize = strlen(name) + 1; | |
1156 | itx_t *itx; | |
1157 | ||
1158 | if (zil_replaying(zd->zd_zilog, tx)) | |
572e2857 | 1159 | return; |
428870ff BB |
1160 | |
1161 | itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize); | |
1162 | bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, | |
1163 | sizeof (*lr) + namesize - sizeof (lr_t)); | |
1164 | ||
572e2857 | 1165 | zil_itx_assign(zd->zd_zilog, itx, tx); |
428870ff BB |
1166 | } |
1167 | ||
572e2857 BB |
1168 | static void |
1169 | ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object) | |
428870ff BB |
1170 | { |
1171 | char *name = (void *)(lr + 1); /* name follows lr */ | |
1172 | size_t namesize = strlen(name) + 1; | |
1173 | itx_t *itx; | |
1174 | ||
1175 | if (zil_replaying(zd->zd_zilog, tx)) | |
572e2857 | 1176 | return; |
428870ff BB |
1177 | |
1178 | itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize); | |
1179 | bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, | |
1180 | sizeof (*lr) + namesize - sizeof (lr_t)); | |
1181 | ||
572e2857 BB |
1182 | itx->itx_oid = object; |
1183 | zil_itx_assign(zd->zd_zilog, itx, tx); | |
428870ff BB |
1184 | } |
1185 | ||
572e2857 | 1186 | static void |
428870ff BB |
1187 | ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr) |
1188 | { | |
1189 | itx_t *itx; | |
1190 | itx_wr_state_t write_state = ztest_random(WR_NUM_STATES); | |
1191 | ||
1192 | if (zil_replaying(zd->zd_zilog, tx)) | |
572e2857 | 1193 | return; |
428870ff BB |
1194 | |
1195 | if (lr->lr_length > ZIL_MAX_LOG_DATA) | |
1196 | write_state = WR_INDIRECT; | |
1197 | ||
1198 | itx = zil_itx_create(TX_WRITE, | |
1199 | sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0)); | |
1200 | ||
1201 | if (write_state == WR_COPIED && | |
1202 | dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length, | |
1203 | ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) { | |
1204 | zil_itx_destroy(itx); | |
1205 | itx = zil_itx_create(TX_WRITE, sizeof (*lr)); | |
1206 | write_state = WR_NEED_COPY; | |
1207 | } | |
1208 | itx->itx_private = zd; | |
1209 | itx->itx_wr_state = write_state; | |
1210 | itx->itx_sync = (ztest_random(8) == 0); | |
1211 | itx->itx_sod += (write_state == WR_NEED_COPY ? lr->lr_length : 0); | |
1212 | ||
1213 | bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, | |
1214 | sizeof (*lr) - sizeof (lr_t)); | |
1215 | ||
572e2857 | 1216 | zil_itx_assign(zd->zd_zilog, itx, tx); |
428870ff BB |
1217 | } |
1218 | ||
572e2857 | 1219 | static void |
428870ff BB |
1220 | ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr) |
1221 | { | |
1222 | itx_t *itx; | |
1223 | ||
1224 | if (zil_replaying(zd->zd_zilog, tx)) | |
572e2857 | 1225 | return; |
428870ff BB |
1226 | |
1227 | itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr)); | |
1228 | bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, | |
1229 | sizeof (*lr) - sizeof (lr_t)); | |
1230 | ||
572e2857 BB |
1231 | itx->itx_sync = B_FALSE; |
1232 | zil_itx_assign(zd->zd_zilog, itx, tx); | |
428870ff BB |
1233 | } |
1234 | ||
572e2857 | 1235 | static void |
428870ff BB |
1236 | ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr) |
1237 | { | |
1238 | itx_t *itx; | |
1239 | ||
1240 | if (zil_replaying(zd->zd_zilog, tx)) | |
572e2857 | 1241 | return; |
428870ff BB |
1242 | |
1243 | itx = zil_itx_create(TX_SETATTR, sizeof (*lr)); | |
1244 | bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, | |
1245 | sizeof (*lr) - sizeof (lr_t)); | |
1246 | ||
572e2857 BB |
1247 | itx->itx_sync = B_FALSE; |
1248 | zil_itx_assign(zd->zd_zilog, itx, tx); | |
428870ff BB |
1249 | } |
1250 | ||
1251 | /* | |
1252 | * ZIL replay ops | |
1253 | */ | |
1254 | static int | |
1255 | ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap) | |
1256 | { | |
1257 | char *name = (void *)(lr + 1); /* name follows lr */ | |
1258 | objset_t *os = zd->zd_os; | |
1259 | ztest_block_tag_t *bbt; | |
1260 | dmu_buf_t *db; | |
1261 | dmu_tx_t *tx; | |
1262 | uint64_t txg; | |
1263 | int error = 0; | |
1264 | ||
1265 | if (byteswap) | |
1266 | byteswap_uint64_array(lr, sizeof (*lr)); | |
1267 | ||
1268 | ASSERT(lr->lr_doid == ZTEST_DIROBJ); | |
1269 | ASSERT(name[0] != '\0'); | |
1270 | ||
1271 | tx = dmu_tx_create(os); | |
1272 | ||
1273 | dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name); | |
1274 | ||
1275 | if (lr->lrz_type == DMU_OT_ZAP_OTHER) { | |
1276 | dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL); | |
1277 | } else { | |
1278 | dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); | |
1279 | } | |
1280 | ||
1281 | txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); | |
1282 | if (txg == 0) | |
1283 | return (ENOSPC); | |
1284 | ||
1285 | ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid); | |
1286 | ||
1287 | if (lr->lrz_type == DMU_OT_ZAP_OTHER) { | |
1288 | if (lr->lr_foid == 0) { | |
1289 | lr->lr_foid = zap_create(os, | |
1290 | lr->lrz_type, lr->lrz_bonustype, | |
1291 | lr->lrz_bonuslen, tx); | |
1292 | } else { | |
1293 | error = zap_create_claim(os, lr->lr_foid, | |
1294 | lr->lrz_type, lr->lrz_bonustype, | |
1295 | lr->lrz_bonuslen, tx); | |
1296 | } | |
1297 | } else { | |
1298 | if (lr->lr_foid == 0) { | |
1299 | lr->lr_foid = dmu_object_alloc(os, | |
1300 | lr->lrz_type, 0, lr->lrz_bonustype, | |
1301 | lr->lrz_bonuslen, tx); | |
1302 | } else { | |
1303 | error = dmu_object_claim(os, lr->lr_foid, | |
1304 | lr->lrz_type, 0, lr->lrz_bonustype, | |
1305 | lr->lrz_bonuslen, tx); | |
1306 | } | |
1307 | } | |
1308 | ||
1309 | if (error) { | |
1310 | ASSERT3U(error, ==, EEXIST); | |
1311 | ASSERT(zd->zd_zilog->zl_replay); | |
1312 | dmu_tx_commit(tx); | |
1313 | return (error); | |
1314 | } | |
1315 | ||
1316 | ASSERT(lr->lr_foid != 0); | |
1317 | ||
1318 | if (lr->lrz_type != DMU_OT_ZAP_OTHER) | |
1319 | VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid, | |
1320 | lr->lrz_blocksize, lr->lrz_ibshift, tx)); | |
1321 | ||
1322 | VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db)); | |
1323 | bbt = ztest_bt_bonus(db); | |
1324 | dmu_buf_will_dirty(db, tx); | |
1325 | ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg); | |
1326 | dmu_buf_rele(db, FTAG); | |
1327 | ||
1328 | VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1, | |
1329 | &lr->lr_foid, tx)); | |
1330 | ||
1331 | (void) ztest_log_create(zd, tx, lr); | |
1332 | ||
1333 | dmu_tx_commit(tx); | |
1334 | ||
1335 | return (0); | |
1336 | } | |
1337 | ||
1338 | static int | |
1339 | ztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap) | |
1340 | { | |
1341 | char *name = (void *)(lr + 1); /* name follows lr */ | |
1342 | objset_t *os = zd->zd_os; | |
1343 | dmu_object_info_t doi; | |
1344 | dmu_tx_t *tx; | |
1345 | uint64_t object, txg; | |
1346 | ||
1347 | if (byteswap) | |
1348 | byteswap_uint64_array(lr, sizeof (*lr)); | |
1349 | ||
1350 | ASSERT(lr->lr_doid == ZTEST_DIROBJ); | |
1351 | ASSERT(name[0] != '\0'); | |
1352 | ||
1353 | VERIFY3U(0, ==, | |
1354 | zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object)); | |
1355 | ASSERT(object != 0); | |
1356 | ||
1357 | ztest_object_lock(zd, object, RL_WRITER); | |
1358 | ||
1359 | VERIFY3U(0, ==, dmu_object_info(os, object, &doi)); | |
1360 | ||
1361 | tx = dmu_tx_create(os); | |
1362 | ||
1363 | dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name); | |
1364 | dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END); | |
1365 | ||
1366 | txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); | |
1367 | if (txg == 0) { | |
1368 | ztest_object_unlock(zd, object); | |
1369 | return (ENOSPC); | |
1370 | } | |
1371 | ||
1372 | if (doi.doi_type == DMU_OT_ZAP_OTHER) { | |
1373 | VERIFY3U(0, ==, zap_destroy(os, object, tx)); | |
1374 | } else { | |
1375 | VERIFY3U(0, ==, dmu_object_free(os, object, tx)); | |
1376 | } | |
1377 | ||
1378 | VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx)); | |
1379 | ||
572e2857 | 1380 | (void) ztest_log_remove(zd, tx, lr, object); |
428870ff BB |
1381 | |
1382 | dmu_tx_commit(tx); | |
1383 | ||
1384 | ztest_object_unlock(zd, object); | |
1385 | ||
1386 | return (0); | |
1387 | } | |
1388 | ||
1389 | static int | |
1390 | ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap) | |
1391 | { | |
1392 | objset_t *os = zd->zd_os; | |
1393 | void *data = lr + 1; /* data follows lr */ | |
1394 | uint64_t offset, length; | |
1395 | ztest_block_tag_t *bt = data; | |
1396 | ztest_block_tag_t *bbt; | |
1397 | uint64_t gen, txg, lrtxg, crtxg; | |
1398 | dmu_object_info_t doi; | |
1399 | dmu_tx_t *tx; | |
1400 | dmu_buf_t *db; | |
1401 | arc_buf_t *abuf = NULL; | |
1402 | rl_t *rl; | |
1403 | ||
1404 | if (byteswap) | |
1405 | byteswap_uint64_array(lr, sizeof (*lr)); | |
1406 | ||
1407 | offset = lr->lr_offset; | |
1408 | length = lr->lr_length; | |
1409 | ||
1410 | /* If it's a dmu_sync() block, write the whole block */ | |
1411 | if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) { | |
1412 | uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr); | |
1413 | if (length < blocksize) { | |
1414 | offset -= offset % blocksize; | |
1415 | length = blocksize; | |
1416 | } | |
1417 | } | |
1418 | ||
1419 | if (bt->bt_magic == BSWAP_64(BT_MAGIC)) | |
1420 | byteswap_uint64_array(bt, sizeof (*bt)); | |
1421 | ||
1422 | if (bt->bt_magic != BT_MAGIC) | |
1423 | bt = NULL; | |
1424 | ||
1425 | ztest_object_lock(zd, lr->lr_foid, RL_READER); | |
1426 | rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER); | |
1427 | ||
1428 | VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db)); | |
1429 | ||
1430 | dmu_object_info_from_db(db, &doi); | |
1431 | ||
1432 | bbt = ztest_bt_bonus(db); | |
1433 | ASSERT3U(bbt->bt_magic, ==, BT_MAGIC); | |
1434 | gen = bbt->bt_gen; | |
1435 | crtxg = bbt->bt_crtxg; | |
1436 | lrtxg = lr->lr_common.lrc_txg; | |
1437 | ||
1438 | tx = dmu_tx_create(os); | |
1439 | ||
1440 | dmu_tx_hold_write(tx, lr->lr_foid, offset, length); | |
1441 | ||
1442 | if (ztest_random(8) == 0 && length == doi.doi_data_block_size && | |
1443 | P2PHASE(offset, length) == 0) | |
1444 | abuf = dmu_request_arcbuf(db, length); | |
1445 | ||
1446 | txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); | |
1447 | if (txg == 0) { | |
1448 | if (abuf != NULL) | |
1449 | dmu_return_arcbuf(abuf); | |
1450 | dmu_buf_rele(db, FTAG); | |
1451 | ztest_range_unlock(rl); | |
1452 | ztest_object_unlock(zd, lr->lr_foid); | |
1453 | return (ENOSPC); | |
1454 | } | |
1455 | ||
1456 | if (bt != NULL) { | |
1457 | /* | |
1458 | * Usually, verify the old data before writing new data -- | |
1459 | * but not always, because we also want to verify correct | |
1460 | * behavior when the data was not recently read into cache. | |
1461 | */ | |
1462 | ASSERT(offset % doi.doi_data_block_size == 0); | |
1463 | if (ztest_random(4) != 0) { | |
1464 | int prefetch = ztest_random(2) ? | |
1465 | DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH; | |
1466 | ztest_block_tag_t rbt; | |
1467 | ||
1468 | VERIFY(dmu_read(os, lr->lr_foid, offset, | |
1469 | sizeof (rbt), &rbt, prefetch) == 0); | |
1470 | if (rbt.bt_magic == BT_MAGIC) { | |
1471 | ztest_bt_verify(&rbt, os, lr->lr_foid, | |
1472 | offset, gen, txg, crtxg); | |
1473 | } | |
1474 | } | |
1475 | ||
1476 | /* | |
1477 | * Writes can appear to be newer than the bonus buffer because | |
1478 | * the ztest_get_data() callback does a dmu_read() of the | |
1479 | * open-context data, which may be different than the data | |
1480 | * as it was when the write was generated. | |
1481 | */ | |
1482 | if (zd->zd_zilog->zl_replay) { | |
1483 | ztest_bt_verify(bt, os, lr->lr_foid, offset, | |
1484 | MAX(gen, bt->bt_gen), MAX(txg, lrtxg), | |
1485 | bt->bt_crtxg); | |
1486 | } | |
1487 | ||
1488 | /* | |
1489 | * Set the bt's gen/txg to the bonus buffer's gen/txg | |
1490 | * so that all of the usual ASSERTs will work. | |
1491 | */ | |
1492 | ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg); | |
1493 | } | |
1494 | ||
1495 | if (abuf == NULL) { | |
1496 | dmu_write(os, lr->lr_foid, offset, length, data, tx); | |
1497 | } else { | |
1498 | bcopy(data, abuf->b_data, length); | |
1499 | dmu_assign_arcbuf(db, offset, abuf, tx); | |
1500 | } | |
1501 | ||
1502 | (void) ztest_log_write(zd, tx, lr); | |
1503 | ||
1504 | dmu_buf_rele(db, FTAG); | |
1505 | ||
1506 | dmu_tx_commit(tx); | |
1507 | ||
1508 | ztest_range_unlock(rl); | |
1509 | ztest_object_unlock(zd, lr->lr_foid); | |
1510 | ||
1511 | return (0); | |
1512 | } | |
1513 | ||
1514 | static int | |
1515 | ztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap) | |
1516 | { | |
1517 | objset_t *os = zd->zd_os; | |
1518 | dmu_tx_t *tx; | |
1519 | uint64_t txg; | |
1520 | rl_t *rl; | |
1521 | ||
1522 | if (byteswap) | |
1523 | byteswap_uint64_array(lr, sizeof (*lr)); | |
1524 | ||
1525 | ztest_object_lock(zd, lr->lr_foid, RL_READER); | |
1526 | rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length, | |
1527 | RL_WRITER); | |
1528 | ||
1529 | tx = dmu_tx_create(os); | |
1530 | ||
1531 | dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length); | |
1532 | ||
1533 | txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); | |
1534 | if (txg == 0) { | |
1535 | ztest_range_unlock(rl); | |
1536 | ztest_object_unlock(zd, lr->lr_foid); | |
1537 | return (ENOSPC); | |
1538 | } | |
1539 | ||
1540 | VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset, | |
1541 | lr->lr_length, tx) == 0); | |
1542 | ||
1543 | (void) ztest_log_truncate(zd, tx, lr); | |
1544 | ||
1545 | dmu_tx_commit(tx); | |
1546 | ||
1547 | ztest_range_unlock(rl); | |
1548 | ztest_object_unlock(zd, lr->lr_foid); | |
1549 | ||
1550 | return (0); | |
1551 | } | |
1552 | ||
1553 | static int | |
1554 | ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap) | |
1555 | { | |
1556 | objset_t *os = zd->zd_os; | |
1557 | dmu_tx_t *tx; | |
1558 | dmu_buf_t *db; | |
1559 | ztest_block_tag_t *bbt; | |
1560 | uint64_t txg, lrtxg, crtxg; | |
1561 | ||
1562 | if (byteswap) | |
1563 | byteswap_uint64_array(lr, sizeof (*lr)); | |
1564 | ||
1565 | ztest_object_lock(zd, lr->lr_foid, RL_WRITER); | |
1566 | ||
1567 | VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db)); | |
1568 | ||
1569 | tx = dmu_tx_create(os); | |
1570 | dmu_tx_hold_bonus(tx, lr->lr_foid); | |
1571 | ||
1572 | txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); | |
1573 | if (txg == 0) { | |
1574 | dmu_buf_rele(db, FTAG); | |
1575 | ztest_object_unlock(zd, lr->lr_foid); | |
1576 | return (ENOSPC); | |
1577 | } | |
1578 | ||
1579 | bbt = ztest_bt_bonus(db); | |
1580 | ASSERT3U(bbt->bt_magic, ==, BT_MAGIC); | |
1581 | crtxg = bbt->bt_crtxg; | |
1582 | lrtxg = lr->lr_common.lrc_txg; | |
1583 | ||
1584 | if (zd->zd_zilog->zl_replay) { | |
1585 | ASSERT(lr->lr_size != 0); | |
1586 | ASSERT(lr->lr_mode != 0); | |
1587 | ASSERT(lrtxg != 0); | |
1588 | } else { | |
1589 | /* | |
1590 | * Randomly change the size and increment the generation. | |
1591 | */ | |
1592 | lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) * | |
1593 | sizeof (*bbt); | |
1594 | lr->lr_mode = bbt->bt_gen + 1; | |
1595 | ASSERT(lrtxg == 0); | |
1596 | } | |
1597 | ||
1598 | /* | |
1599 | * Verify that the current bonus buffer is not newer than our txg. | |
1600 | */ | |
1601 | ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, | |
1602 | MAX(txg, lrtxg), crtxg); | |
1603 | ||
1604 | dmu_buf_will_dirty(db, tx); | |
1605 | ||
1606 | ASSERT3U(lr->lr_size, >=, sizeof (*bbt)); | |
1607 | ASSERT3U(lr->lr_size, <=, db->db_size); | |
1608 | VERIFY3U(dmu_set_bonus(db, lr->lr_size, tx), ==, 0); | |
1609 | bbt = ztest_bt_bonus(db); | |
1610 | ||
1611 | ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg); | |
1612 | ||
1613 | dmu_buf_rele(db, FTAG); | |
1614 | ||
1615 | (void) ztest_log_setattr(zd, tx, lr); | |
1616 | ||
1617 | dmu_tx_commit(tx); | |
1618 | ||
1619 | ztest_object_unlock(zd, lr->lr_foid); | |
1620 | ||
1621 | return (0); | |
1622 | } | |
1623 | ||
1624 | zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = { | |
0bc8fd78 BB |
1625 | NULL, /* 0 no such transaction type */ |
1626 | (zil_replay_func_t *)ztest_replay_create, /* TX_CREATE */ | |
1627 | NULL, /* TX_MKDIR */ | |
1628 | NULL, /* TX_MKXATTR */ | |
1629 | NULL, /* TX_SYMLINK */ | |
1630 | (zil_replay_func_t *)ztest_replay_remove, /* TX_REMOVE */ | |
1631 | NULL, /* TX_RMDIR */ | |
1632 | NULL, /* TX_LINK */ | |
1633 | NULL, /* TX_RENAME */ | |
1634 | (zil_replay_func_t *)ztest_replay_write, /* TX_WRITE */ | |
1635 | (zil_replay_func_t *)ztest_replay_truncate, /* TX_TRUNCATE */ | |
1636 | (zil_replay_func_t *)ztest_replay_setattr, /* TX_SETATTR */ | |
1637 | NULL, /* TX_ACL */ | |
1638 | NULL, /* TX_CREATE_ACL */ | |
1639 | NULL, /* TX_CREATE_ATTR */ | |
1640 | NULL, /* TX_CREATE_ACL_ATTR */ | |
1641 | NULL, /* TX_MKDIR_ACL */ | |
1642 | NULL, /* TX_MKDIR_ATTR */ | |
1643 | NULL, /* TX_MKDIR_ACL_ATTR */ | |
1644 | NULL, /* TX_WRITE2 */ | |
428870ff BB |
1645 | }; |
1646 | ||
1647 | /* | |
1648 | * ZIL get_data callbacks | |
1649 | */ | |
1650 | ||
1651 | static void | |
1652 | ztest_get_done(zgd_t *zgd, int error) | |
1653 | { | |
1654 | ztest_ds_t *zd = zgd->zgd_private; | |
1655 | uint64_t object = zgd->zgd_rl->rl_object; | |
1656 | ||
1657 | if (zgd->zgd_db) | |
1658 | dmu_buf_rele(zgd->zgd_db, zgd); | |
1659 | ||
1660 | ztest_range_unlock(zgd->zgd_rl); | |
1661 | ztest_object_unlock(zd, object); | |
1662 | ||
1663 | if (error == 0 && zgd->zgd_bp) | |
1664 | zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); | |
1665 | ||
1666 | umem_free(zgd, sizeof (*zgd)); | |
1667 | } | |
1668 | ||
1669 | static int | |
1670 | ztest_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) | |
1671 | { | |
1672 | ztest_ds_t *zd = arg; | |
1673 | objset_t *os = zd->zd_os; | |
1674 | uint64_t object = lr->lr_foid; | |
1675 | uint64_t offset = lr->lr_offset; | |
1676 | uint64_t size = lr->lr_length; | |
1677 | blkptr_t *bp = &lr->lr_blkptr; | |
1678 | uint64_t txg = lr->lr_common.lrc_txg; | |
1679 | uint64_t crtxg; | |
1680 | dmu_object_info_t doi; | |
1681 | dmu_buf_t *db; | |
1682 | zgd_t *zgd; | |
1683 | int error; | |
1684 | ||
1685 | ztest_object_lock(zd, object, RL_READER); | |
1686 | error = dmu_bonus_hold(os, object, FTAG, &db); | |
1687 | if (error) { | |
1688 | ztest_object_unlock(zd, object); | |
1689 | return (error); | |
1690 | } | |
1691 | ||
1692 | crtxg = ztest_bt_bonus(db)->bt_crtxg; | |
1693 | ||
1694 | if (crtxg == 0 || crtxg > txg) { | |
1695 | dmu_buf_rele(db, FTAG); | |
1696 | ztest_object_unlock(zd, object); | |
1697 | return (ENOENT); | |
1698 | } | |
1699 | ||
1700 | dmu_object_info_from_db(db, &doi); | |
1701 | dmu_buf_rele(db, FTAG); | |
1702 | db = NULL; | |
1703 | ||
1704 | zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL); | |
1705 | zgd->zgd_zilog = zd->zd_zilog; | |
1706 | zgd->zgd_private = zd; | |
1707 | ||
1708 | if (buf != NULL) { /* immediate write */ | |
1709 | zgd->zgd_rl = ztest_range_lock(zd, object, offset, size, | |
1710 | RL_READER); | |
1711 | ||
1712 | error = dmu_read(os, object, offset, size, buf, | |
1713 | DMU_READ_NO_PREFETCH); | |
1714 | ASSERT(error == 0); | |
1715 | } else { | |
1716 | size = doi.doi_data_block_size; | |
1717 | if (ISP2(size)) { | |
1718 | offset = P2ALIGN(offset, size); | |
1719 | } else { | |
1720 | ASSERT(offset < size); | |
1721 | offset = 0; | |
1722 | } | |
1723 | ||
1724 | zgd->zgd_rl = ztest_range_lock(zd, object, offset, size, | |
1725 | RL_READER); | |
1726 | ||
1727 | error = dmu_buf_hold(os, object, offset, zgd, &db, | |
1728 | DMU_READ_NO_PREFETCH); | |
1729 | ||
1730 | if (error == 0) { | |
1731 | zgd->zgd_db = db; | |
1732 | zgd->zgd_bp = bp; | |
1733 | ||
1734 | ASSERT(db->db_offset == offset); | |
1735 | ASSERT(db->db_size == size); | |
1736 | ||
1737 | error = dmu_sync(zio, lr->lr_common.lrc_txg, | |
1738 | ztest_get_done, zgd); | |
1739 | ||
1740 | if (error == 0) | |
1741 | return (0); | |
1742 | } | |
1743 | } | |
1744 | ||
1745 | ztest_get_done(zgd, error); | |
1746 | ||
1747 | return (error); | |
1748 | } | |
1749 | ||
1750 | static void * | |
1751 | ztest_lr_alloc(size_t lrsize, char *name) | |
1752 | { | |
1753 | char *lr; | |
1754 | size_t namesize = name ? strlen(name) + 1 : 0; | |
1755 | ||
1756 | lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL); | |
1757 | ||
1758 | if (name) | |
1759 | bcopy(name, lr + lrsize, namesize); | |
1760 | ||
1761 | return (lr); | |
1762 | } | |
1763 | ||
1764 | void | |
1765 | ztest_lr_free(void *lr, size_t lrsize, char *name) | |
1766 | { | |
1767 | size_t namesize = name ? strlen(name) + 1 : 0; | |
1768 | ||
1769 | umem_free(lr, lrsize + namesize); | |
1770 | } | |
1771 | ||
1772 | /* | |
1773 | * Lookup a bunch of objects. Returns the number of objects not found. | |
1774 | */ | |
1775 | static int | |
1776 | ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count) | |
1777 | { | |
1778 | int missing = 0; | |
1779 | int error; | |
d6320ddb | 1780 | int i; |
428870ff | 1781 | |
1e33ac1e | 1782 | ASSERT(mutex_held(&zd->zd_dirobj_lock)); |
428870ff | 1783 | |
d6320ddb | 1784 | for (i = 0; i < count; i++, od++) { |
428870ff BB |
1785 | od->od_object = 0; |
1786 | error = zap_lookup(zd->zd_os, od->od_dir, od->od_name, | |
1787 | sizeof (uint64_t), 1, &od->od_object); | |
1788 | if (error) { | |
1789 | ASSERT(error == ENOENT); | |
1790 | ASSERT(od->od_object == 0); | |
1791 | missing++; | |
1792 | } else { | |
1793 | dmu_buf_t *db; | |
1794 | ztest_block_tag_t *bbt; | |
1795 | dmu_object_info_t doi; | |
1796 | ||
1797 | ASSERT(od->od_object != 0); | |
1798 | ASSERT(missing == 0); /* there should be no gaps */ | |
1799 | ||
1800 | ztest_object_lock(zd, od->od_object, RL_READER); | |
1801 | VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os, | |
1802 | od->od_object, FTAG, &db)); | |
1803 | dmu_object_info_from_db(db, &doi); | |
1804 | bbt = ztest_bt_bonus(db); | |
1805 | ASSERT3U(bbt->bt_magic, ==, BT_MAGIC); | |
1806 | od->od_type = doi.doi_type; | |
1807 | od->od_blocksize = doi.doi_data_block_size; | |
1808 | od->od_gen = bbt->bt_gen; | |
1809 | dmu_buf_rele(db, FTAG); | |
1810 | ztest_object_unlock(zd, od->od_object); | |
1811 | } | |
1812 | } | |
1813 | ||
1814 | return (missing); | |
1815 | } | |
1816 | ||
1817 | static int | |
1818 | ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count) | |
1819 | { | |
1820 | int missing = 0; | |
d6320ddb | 1821 | int i; |
428870ff | 1822 | |
1e33ac1e | 1823 | ASSERT(mutex_held(&zd->zd_dirobj_lock)); |
428870ff | 1824 | |
d6320ddb | 1825 | for (i = 0; i < count; i++, od++) { |
428870ff BB |
1826 | if (missing) { |
1827 | od->od_object = 0; | |
1828 | missing++; | |
1829 | continue; | |
1830 | } | |
1831 | ||
1832 | lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name); | |
1833 | ||
1834 | lr->lr_doid = od->od_dir; | |
1835 | lr->lr_foid = 0; /* 0 to allocate, > 0 to claim */ | |
1836 | lr->lrz_type = od->od_crtype; | |
1837 | lr->lrz_blocksize = od->od_crblocksize; | |
1838 | lr->lrz_ibshift = ztest_random_ibshift(); | |
1839 | lr->lrz_bonustype = DMU_OT_UINT64_OTHER; | |
1840 | lr->lrz_bonuslen = dmu_bonus_max(); | |
1841 | lr->lr_gen = od->od_crgen; | |
1842 | lr->lr_crtime[0] = time(NULL); | |
1843 | ||
1844 | if (ztest_replay_create(zd, lr, B_FALSE) != 0) { | |
1845 | ASSERT(missing == 0); | |
1846 | od->od_object = 0; | |
1847 | missing++; | |
1848 | } else { | |
1849 | od->od_object = lr->lr_foid; | |
1850 | od->od_type = od->od_crtype; | |
1851 | od->od_blocksize = od->od_crblocksize; | |
1852 | od->od_gen = od->od_crgen; | |
1853 | ASSERT(od->od_object != 0); | |
1854 | } | |
1855 | ||
1856 | ztest_lr_free(lr, sizeof (*lr), od->od_name); | |
1857 | } | |
1858 | ||
1859 | return (missing); | |
1860 | } | |
1861 | ||
1862 | static int | |
1863 | ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count) | |
1864 | { | |
1865 | int missing = 0; | |
1866 | int error; | |
d6320ddb | 1867 | int i; |
428870ff | 1868 | |
1e33ac1e | 1869 | ASSERT(mutex_held(&zd->zd_dirobj_lock)); |
428870ff BB |
1870 | |
1871 | od += count - 1; | |
1872 | ||
d6320ddb | 1873 | for (i = count - 1; i >= 0; i--, od--) { |
428870ff BB |
1874 | if (missing) { |
1875 | missing++; | |
1876 | continue; | |
1877 | } | |
1878 | ||
1879 | if (od->od_object == 0) | |
1880 | continue; | |
1881 | ||
1882 | lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name); | |
1883 | ||
1884 | lr->lr_doid = od->od_dir; | |
1885 | ||
1886 | if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) { | |
1887 | ASSERT3U(error, ==, ENOSPC); | |
1888 | missing++; | |
1889 | } else { | |
1890 | od->od_object = 0; | |
1891 | } | |
1892 | ztest_lr_free(lr, sizeof (*lr), od->od_name); | |
1893 | } | |
1894 | ||
1895 | return (missing); | |
1896 | } | |
1897 | ||
1898 | static int | |
1899 | ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size, | |
1900 | void *data) | |
1901 | { | |
1902 | lr_write_t *lr; | |
1903 | int error; | |
1904 | ||
1905 | lr = ztest_lr_alloc(sizeof (*lr) + size, NULL); | |
1906 | ||
1907 | lr->lr_foid = object; | |
1908 | lr->lr_offset = offset; | |
1909 | lr->lr_length = size; | |
1910 | lr->lr_blkoff = 0; | |
1911 | BP_ZERO(&lr->lr_blkptr); | |
1912 | ||
1913 | bcopy(data, lr + 1, size); | |
1914 | ||
1915 | error = ztest_replay_write(zd, lr, B_FALSE); | |
1916 | ||
1917 | ztest_lr_free(lr, sizeof (*lr) + size, NULL); | |
1918 | ||
1919 | return (error); | |
1920 | } | |
1921 | ||
1922 | static int | |
1923 | ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size) | |
1924 | { | |
1925 | lr_truncate_t *lr; | |
1926 | int error; | |
1927 | ||
1928 | lr = ztest_lr_alloc(sizeof (*lr), NULL); | |
1929 | ||
1930 | lr->lr_foid = object; | |
1931 | lr->lr_offset = offset; | |
1932 | lr->lr_length = size; | |
1933 | ||
1934 | error = ztest_replay_truncate(zd, lr, B_FALSE); | |
1935 | ||
1936 | ztest_lr_free(lr, sizeof (*lr), NULL); | |
1937 | ||
1938 | return (error); | |
1939 | } | |
1940 | ||
1941 | static int | |
1942 | ztest_setattr(ztest_ds_t *zd, uint64_t object) | |
1943 | { | |
1944 | lr_setattr_t *lr; | |
1945 | int error; | |
1946 | ||
1947 | lr = ztest_lr_alloc(sizeof (*lr), NULL); | |
1948 | ||
1949 | lr->lr_foid = object; | |
1950 | lr->lr_size = 0; | |
1951 | lr->lr_mode = 0; | |
1952 | ||
1953 | error = ztest_replay_setattr(zd, lr, B_FALSE); | |
1954 | ||
1955 | ztest_lr_free(lr, sizeof (*lr), NULL); | |
1956 | ||
1957 | return (error); | |
1958 | } | |
1959 | ||
1960 | static void | |
1961 | ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size) | |
1962 | { | |
1963 | objset_t *os = zd->zd_os; | |
1964 | dmu_tx_t *tx; | |
1965 | uint64_t txg; | |
1966 | rl_t *rl; | |
1967 | ||
1968 | txg_wait_synced(dmu_objset_pool(os), 0); | |
1969 | ||
1970 | ztest_object_lock(zd, object, RL_READER); | |
1971 | rl = ztest_range_lock(zd, object, offset, size, RL_WRITER); | |
1972 | ||
1973 | tx = dmu_tx_create(os); | |
1974 | ||
1975 | dmu_tx_hold_write(tx, object, offset, size); | |
1976 | ||
1977 | txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); | |
1978 | ||
1979 | if (txg != 0) { | |
1980 | dmu_prealloc(os, object, offset, size, tx); | |
1981 | dmu_tx_commit(tx); | |
1982 | txg_wait_synced(dmu_objset_pool(os), txg); | |
1983 | } else { | |
1984 | (void) dmu_free_long_range(os, object, offset, size); | |
1985 | } | |
1986 | ||
1987 | ztest_range_unlock(rl); | |
1988 | ztest_object_unlock(zd, object); | |
1989 | } | |
1990 | ||
1991 | static void | |
1992 | ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset) | |
1993 | { | |
1994 | ztest_block_tag_t wbt; | |
1995 | dmu_object_info_t doi; | |
1996 | enum ztest_io_type io_type; | |
1997 | uint64_t blocksize; | |
1998 | void *data; | |
1999 | ||
2000 | VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0); | |
2001 | blocksize = doi.doi_data_block_size; | |
2002 | data = umem_alloc(blocksize, UMEM_NOFAIL); | |
2003 | ||
2004 | /* | |
2005 | * Pick an i/o type at random, biased toward writing block tags. | |
2006 | */ | |
2007 | io_type = ztest_random(ZTEST_IO_TYPES); | |
2008 | if (ztest_random(2) == 0) | |
2009 | io_type = ZTEST_IO_WRITE_TAG; | |
2010 | ||
3e31d2b0 ES |
2011 | (void) rw_enter(&zd->zd_zilog_lock, RW_READER); |
2012 | ||
428870ff BB |
2013 | switch (io_type) { |
2014 | ||
2015 | case ZTEST_IO_WRITE_TAG: | |
2016 | ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0); | |
2017 | (void) ztest_write(zd, object, offset, sizeof (wbt), &wbt); | |
2018 | break; | |
2019 | ||
2020 | case ZTEST_IO_WRITE_PATTERN: | |
2021 | (void) memset(data, 'a' + (object + offset) % 5, blocksize); | |
2022 | if (ztest_random(2) == 0) { | |
2023 | /* | |
2024 | * Induce fletcher2 collisions to ensure that | |
2025 | * zio_ddt_collision() detects and resolves them | |
2026 | * when using fletcher2-verify for deduplication. | |
2027 | */ | |
2028 | ((uint64_t *)data)[0] ^= 1ULL << 63; | |
2029 | ((uint64_t *)data)[4] ^= 1ULL << 63; | |
2030 | } | |
2031 | (void) ztest_write(zd, object, offset, blocksize, data); | |
2032 | break; | |
2033 | ||
2034 | case ZTEST_IO_WRITE_ZEROES: | |
2035 | bzero(data, blocksize); | |
2036 | (void) ztest_write(zd, object, offset, blocksize, data); | |
2037 | break; | |
2038 | ||
2039 | case ZTEST_IO_TRUNCATE: | |
2040 | (void) ztest_truncate(zd, object, offset, blocksize); | |
2041 | break; | |
2042 | ||
2043 | case ZTEST_IO_SETATTR: | |
2044 | (void) ztest_setattr(zd, object); | |
2045 | break; | |
e75c13c3 BB |
2046 | default: |
2047 | break; | |
428870ff BB |
2048 | } |
2049 | ||
3e31d2b0 ES |
2050 | (void) rw_exit(&zd->zd_zilog_lock); |
2051 | ||
428870ff BB |
2052 | umem_free(data, blocksize); |
2053 | } | |
2054 | ||
2055 | /* | |
2056 | * Initialize an object description template. | |
2057 | */ | |
2058 | static void | |
2059 | ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index, | |
2060 | dmu_object_type_t type, uint64_t blocksize, uint64_t gen) | |
2061 | { | |
2062 | od->od_dir = ZTEST_DIROBJ; | |
2063 | od->od_object = 0; | |
2064 | ||
2065 | od->od_crtype = type; | |
2066 | od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize(); | |
2067 | od->od_crgen = gen; | |
2068 | ||
2069 | od->od_type = DMU_OT_NONE; | |
2070 | od->od_blocksize = 0; | |
2071 | od->od_gen = 0; | |
2072 | ||
2073 | (void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]", | |
b8864a23 | 2074 | tag, (longlong_t)id, (u_longlong_t)index); |
428870ff BB |
2075 | } |
2076 | ||
2077 | /* | |
2078 | * Lookup or create the objects for a test using the od template. | |
2079 | * If the objects do not all exist, or if 'remove' is specified, | |
2080 | * remove any existing objects and create new ones. Otherwise, | |
2081 | * use the existing objects. | |
2082 | */ | |
2083 | static int | |
2084 | ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove) | |
2085 | { | |
2086 | int count = size / sizeof (*od); | |
2087 | int rv = 0; | |
2088 | ||
1e33ac1e | 2089 | mutex_enter(&zd->zd_dirobj_lock); |
428870ff BB |
2090 | if ((ztest_lookup(zd, od, count) != 0 || remove) && |
2091 | (ztest_remove(zd, od, count) != 0 || | |
2092 | ztest_create(zd, od, count) != 0)) | |
2093 | rv = -1; | |
2094 | zd->zd_od = od; | |
1e33ac1e | 2095 | mutex_exit(&zd->zd_dirobj_lock); |
428870ff BB |
2096 | |
2097 | return (rv); | |
2098 | } | |
2099 | ||
2100 | /* ARGSUSED */ | |
2101 | void | |
2102 | ztest_zil_commit(ztest_ds_t *zd, uint64_t id) | |
2103 | { | |
2104 | zilog_t *zilog = zd->zd_zilog; | |
2105 | ||
3e31d2b0 ES |
2106 | (void) rw_enter(&zd->zd_zilog_lock, RW_READER); |
2107 | ||
572e2857 | 2108 | zil_commit(zilog, ztest_random(ZTEST_OBJECTS)); |
428870ff BB |
2109 | |
2110 | /* | |
2111 | * Remember the committed values in zd, which is in parent/child | |
2112 | * shared memory. If we die, the next iteration of ztest_run() | |
2113 | * will verify that the log really does contain this record. | |
2114 | */ | |
2115 | mutex_enter(&zilog->zl_lock); | |
2116 | ASSERT(zd->zd_seq <= zilog->zl_commit_lr_seq); | |
2117 | zd->zd_seq = zilog->zl_commit_lr_seq; | |
2118 | mutex_exit(&zilog->zl_lock); | |
3e31d2b0 ES |
2119 | |
2120 | (void) rw_exit(&zd->zd_zilog_lock); | |
2121 | } | |
2122 | ||
2123 | /* | |
2124 | * This function is designed to simulate the operations that occur during a | |
2125 | * mount/unmount operation. We hold the dataset across these operations in an | |
2126 | * attempt to expose any implicit assumptions about ZIL management. | |
2127 | */ | |
2128 | /* ARGSUSED */ | |
2129 | void | |
2130 | ztest_zil_remount(ztest_ds_t *zd, uint64_t id) | |
2131 | { | |
2132 | objset_t *os = zd->zd_os; | |
2133 | ||
2134 | (void) rw_enter(&zd->zd_zilog_lock, RW_WRITER); | |
2135 | ||
86f35f34 | 2136 | /* zfs_sb_teardown() */ |
3e31d2b0 ES |
2137 | zil_close(zd->zd_zilog); |
2138 | ||
2139 | /* zfsvfs_setup() */ | |
2140 | VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog); | |
2141 | zil_replay(os, zd, ztest_replay_vector); | |
2142 | ||
2143 | (void) rw_exit(&zd->zd_zilog_lock); | |
428870ff BB |
2144 | } |
2145 | ||
2146 | /* | |
2147 | * Verify that we can't destroy an active pool, create an existing pool, | |
2148 | * or create a pool with a bad vdev spec. | |
2149 | */ | |
2150 | /* ARGSUSED */ | |
2151 | void | |
2152 | ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id) | |
2153 | { | |
2154 | ztest_shared_t *zs = ztest_shared; | |
2155 | spa_t *spa; | |
2156 | nvlist_t *nvroot; | |
2157 | ||
2158 | /* | |
2159 | * Attempt to create using a bad file. | |
2160 | */ | |
2161 | nvroot = make_vdev_root("/dev/bogus", NULL, 0, 0, 0, 0, 0, 1); | |
2162 | VERIFY3U(ENOENT, ==, | |
2163 | spa_create("ztest_bad_file", nvroot, NULL, NULL, NULL)); | |
2164 | nvlist_free(nvroot); | |
2165 | ||
2166 | /* | |
2167 | * Attempt to create using a bad mirror. | |
2168 | */ | |
2169 | nvroot = make_vdev_root("/dev/bogus", NULL, 0, 0, 0, 0, 2, 1); | |
2170 | VERIFY3U(ENOENT, ==, | |
2171 | spa_create("ztest_bad_mirror", nvroot, NULL, NULL, NULL)); | |
2172 | nvlist_free(nvroot); | |
2173 | ||
2174 | /* | |
2175 | * Attempt to create an existing pool. It shouldn't matter | |
2176 | * what's in the nvroot; we should fail with EEXIST. | |
2177 | */ | |
1e33ac1e | 2178 | (void) rw_enter(&zs->zs_name_lock, RW_READER); |
428870ff BB |
2179 | nvroot = make_vdev_root("/dev/bogus", NULL, 0, 0, 0, 0, 0, 1); |
2180 | VERIFY3U(EEXIST, ==, spa_create(zs->zs_pool, nvroot, NULL, NULL, NULL)); | |
2181 | nvlist_free(nvroot); | |
2182 | VERIFY3U(0, ==, spa_open(zs->zs_pool, &spa, FTAG)); | |
2183 | VERIFY3U(EBUSY, ==, spa_destroy(zs->zs_pool)); | |
2184 | spa_close(spa, FTAG); | |
2185 | ||
1e33ac1e | 2186 | (void) rw_exit(&zs->zs_name_lock); |
428870ff BB |
2187 | } |
2188 | ||
2189 | static vdev_t * | |
2190 | vdev_lookup_by_path(vdev_t *vd, const char *path) | |
2191 | { | |
2192 | vdev_t *mvd; | |
d6320ddb | 2193 | int c; |
428870ff BB |
2194 | |
2195 | if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0) | |
2196 | return (vd); | |
2197 | ||
d6320ddb | 2198 | for (c = 0; c < vd->vdev_children; c++) |
428870ff BB |
2199 | if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) != |
2200 | NULL) | |
2201 | return (mvd); | |
2202 | ||
2203 | return (NULL); | |
2204 | } | |
2205 | ||
2206 | /* | |
2207 | * Find the first available hole which can be used as a top-level. | |
2208 | */ | |
2209 | int | |
2210 | find_vdev_hole(spa_t *spa) | |
2211 | { | |
2212 | vdev_t *rvd = spa->spa_root_vdev; | |
2213 | int c; | |
2214 | ||
2215 | ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV); | |
2216 | ||
2217 | for (c = 0; c < rvd->vdev_children; c++) { | |
2218 | vdev_t *cvd = rvd->vdev_child[c]; | |
2219 | ||
2220 | if (cvd->vdev_ishole) | |
2221 | break; | |
2222 | } | |
2223 | return (c); | |
2224 | } | |
2225 | ||
2226 | /* | |
2227 | * Verify that vdev_add() works as expected. | |
2228 | */ | |
2229 | /* ARGSUSED */ | |
2230 | void | |
2231 | ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id) | |
2232 | { | |
2233 | ztest_shared_t *zs = ztest_shared; | |
2234 | spa_t *spa = zs->zs_spa; | |
2235 | uint64_t leaves; | |
2236 | uint64_t guid; | |
2237 | nvlist_t *nvroot; | |
2238 | int error; | |
2239 | ||
1e33ac1e | 2240 | mutex_enter(&zs->zs_vdev_lock); |
428870ff BB |
2241 | leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * zopt_raidz; |
2242 | ||
2243 | spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); | |
2244 | ||
2245 | ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves; | |
2246 | ||
2247 | /* | |
2248 | * If we have slogs then remove them 1/4 of the time. | |
2249 | */ | |
2250 | if (spa_has_slogs(spa) && ztest_random(4) == 0) { | |
2251 | /* | |
2252 | * Grab the guid from the head of the log class rotor. | |
2253 | */ | |
2254 | guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid; | |
2255 | ||
2256 | spa_config_exit(spa, SCL_VDEV, FTAG); | |
2257 | ||
2258 | /* | |
2259 | * We have to grab the zs_name_lock as writer to | |
2260 | * prevent a race between removing a slog (dmu_objset_find) | |
2261 | * and destroying a dataset. Removing the slog will | |
2262 | * grab a reference on the dataset which may cause | |
2263 | * dmu_objset_destroy() to fail with EBUSY thus | |
2264 | * leaving the dataset in an inconsistent state. | |
2265 | */ | |
1e33ac1e | 2266 | rw_enter(&ztest_shared->zs_name_lock, RW_WRITER); |
428870ff | 2267 | error = spa_vdev_remove(spa, guid, B_FALSE); |
1e33ac1e | 2268 | rw_exit(&ztest_shared->zs_name_lock); |
428870ff BB |
2269 | |
2270 | if (error && error != EEXIST) | |
2271 | fatal(0, "spa_vdev_remove() = %d", error); | |
2272 | } else { | |
2273 | spa_config_exit(spa, SCL_VDEV, FTAG); | |
2274 | ||
2275 | /* | |
2276 | * Make 1/4 of the devices be log devices. | |
2277 | */ | |
2278 | nvroot = make_vdev_root(NULL, NULL, zopt_vdev_size, 0, | |
2279 | ztest_random(4) == 0, zopt_raidz, zs->zs_mirrors, 1); | |
2280 | ||
2281 | error = spa_vdev_add(spa, nvroot); | |
2282 | nvlist_free(nvroot); | |
2283 | ||
2284 | if (error == ENOSPC) | |
2285 | ztest_record_enospc("spa_vdev_add"); | |
2286 | else if (error != 0) | |
2287 | fatal(0, "spa_vdev_add() = %d", error); | |
2288 | } | |
2289 | ||
1e33ac1e | 2290 | mutex_exit(&ztest_shared->zs_vdev_lock); |
428870ff BB |
2291 | } |
2292 | ||
2293 | /* | |
2294 | * Verify that adding/removing aux devices (l2arc, hot spare) works as expected. | |
2295 | */ | |
2296 | /* ARGSUSED */ | |
2297 | void | |
2298 | ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id) | |
2299 | { | |
2300 | ztest_shared_t *zs = ztest_shared; | |
2301 | spa_t *spa = zs->zs_spa; | |
2302 | vdev_t *rvd = spa->spa_root_vdev; | |
2303 | spa_aux_vdev_t *sav; | |
2304 | char *aux; | |
40b84e7a | 2305 | char *path; |
428870ff BB |
2306 | uint64_t guid = 0; |
2307 | int error; | |
2308 | ||
40b84e7a BB |
2309 | path = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); |
2310 | ||
428870ff BB |
2311 | if (ztest_random(2) == 0) { |
2312 | sav = &spa->spa_spares; | |
2313 | aux = ZPOOL_CONFIG_SPARES; | |
2314 | } else { | |
2315 | sav = &spa->spa_l2cache; | |
2316 | aux = ZPOOL_CONFIG_L2CACHE; | |
2317 | } | |
2318 | ||
1e33ac1e | 2319 | mutex_enter(&zs->zs_vdev_lock); |
428870ff BB |
2320 | |
2321 | spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); | |
2322 | ||
2323 | if (sav->sav_count != 0 && ztest_random(4) == 0) { | |
b128c09f BB |
2324 | /* |
2325 | * Pick a random device to remove. | |
2326 | */ | |
2327 | guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid; | |
2328 | } else { | |
2329 | /* | |
2330 | * Find an unused device we can add. | |
2331 | */ | |
428870ff | 2332 | zs->zs_vdev_aux = 0; |
b128c09f | 2333 | for (;;) { |
b128c09f BB |
2334 | int c; |
2335 | (void) sprintf(path, ztest_aux_template, zopt_dir, | |
428870ff | 2336 | zopt_pool, aux, zs->zs_vdev_aux); |
b128c09f BB |
2337 | for (c = 0; c < sav->sav_count; c++) |
2338 | if (strcmp(sav->sav_vdevs[c]->vdev_path, | |
2339 | path) == 0) | |
2340 | break; | |
2341 | if (c == sav->sav_count && | |
2342 | vdev_lookup_by_path(rvd, path) == NULL) | |
2343 | break; | |
428870ff | 2344 | zs->zs_vdev_aux++; |
34dc7c2f BB |
2345 | } |
2346 | } | |
2347 | ||
b128c09f | 2348 | spa_config_exit(spa, SCL_VDEV, FTAG); |
34dc7c2f | 2349 | |
b128c09f BB |
2350 | if (guid == 0) { |
2351 | /* | |
2352 | * Add a new device. | |
2353 | */ | |
2354 | nvlist_t *nvroot = make_vdev_root(NULL, aux, | |
2355 | (zopt_vdev_size * 5) / 4, 0, 0, 0, 0, 1); | |
2356 | error = spa_vdev_add(spa, nvroot); | |
2357 | if (error != 0) | |
2358 | fatal(0, "spa_vdev_add(%p) = %d", nvroot, error); | |
2359 | nvlist_free(nvroot); | |
2360 | } else { | |
2361 | /* | |
2362 | * Remove an existing device. Sometimes, dirty its | |
2363 | * vdev state first to make sure we handle removal | |
2364 | * of devices that have pending state changes. | |
2365 | */ | |
2366 | if (ztest_random(2) == 0) | |
9babb374 | 2367 | (void) vdev_online(spa, guid, 0, NULL); |
b128c09f BB |
2368 | |
2369 | error = spa_vdev_remove(spa, guid, B_FALSE); | |
2370 | if (error != 0 && error != EBUSY) | |
2371 | fatal(0, "spa_vdev_remove(%llu) = %d", guid, error); | |
2372 | } | |
2373 | ||
1e33ac1e | 2374 | mutex_exit(&zs->zs_vdev_lock); |
40b84e7a BB |
2375 | |
2376 | umem_free(path, MAXPATHLEN); | |
428870ff BB |
2377 | } |
2378 | ||
2379 | /* | |
2380 | * split a pool if it has mirror tlvdevs | |
2381 | */ | |
2382 | /* ARGSUSED */ | |
2383 | void | |
2384 | ztest_split_pool(ztest_ds_t *zd, uint64_t id) | |
2385 | { | |
2386 | ztest_shared_t *zs = ztest_shared; | |
2387 | spa_t *spa = zs->zs_spa; | |
2388 | vdev_t *rvd = spa->spa_root_vdev; | |
2389 | nvlist_t *tree, **child, *config, *split, **schild; | |
2390 | uint_t c, children, schildren = 0, lastlogid = 0; | |
2391 | int error = 0; | |
2392 | ||
1e33ac1e | 2393 | mutex_enter(&zs->zs_vdev_lock); |
428870ff BB |
2394 | |
2395 | /* ensure we have a useable config; mirrors of raidz aren't supported */ | |
2396 | if (zs->zs_mirrors < 3 || zopt_raidz > 1) { | |
1e33ac1e | 2397 | mutex_exit(&zs->zs_vdev_lock); |
428870ff BB |
2398 | return; |
2399 | } | |
2400 | ||
2401 | /* clean up the old pool, if any */ | |
2402 | (void) spa_destroy("splitp"); | |
2403 | ||
2404 | spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); | |
2405 | ||
2406 | /* generate a config from the existing config */ | |
2407 | mutex_enter(&spa->spa_props_lock); | |
2408 | VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE, | |
2409 | &tree) == 0); | |
2410 | mutex_exit(&spa->spa_props_lock); | |
2411 | ||
2412 | VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child, | |
2413 | &children) == 0); | |
2414 | ||
2415 | schild = malloc(rvd->vdev_children * sizeof (nvlist_t *)); | |
2416 | for (c = 0; c < children; c++) { | |
2417 | vdev_t *tvd = rvd->vdev_child[c]; | |
2418 | nvlist_t **mchild; | |
2419 | uint_t mchildren; | |
2420 | ||
2421 | if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) { | |
2422 | VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME, | |
2423 | 0) == 0); | |
2424 | VERIFY(nvlist_add_string(schild[schildren], | |
2425 | ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0); | |
2426 | VERIFY(nvlist_add_uint64(schild[schildren], | |
2427 | ZPOOL_CONFIG_IS_HOLE, 1) == 0); | |
2428 | if (lastlogid == 0) | |
2429 | lastlogid = schildren; | |
2430 | ++schildren; | |
2431 | continue; | |
2432 | } | |
2433 | lastlogid = 0; | |
2434 | VERIFY(nvlist_lookup_nvlist_array(child[c], | |
2435 | ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0); | |
2436 | VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0); | |
2437 | } | |
2438 | ||
2439 | /* OK, create a config that can be used to split */ | |
2440 | VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0); | |
2441 | VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE, | |
2442 | VDEV_TYPE_ROOT) == 0); | |
2443 | VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild, | |
2444 | lastlogid != 0 ? lastlogid : schildren) == 0); | |
2445 | ||
2446 | VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0); | |
2447 | VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0); | |
2448 | ||
2449 | for (c = 0; c < schildren; c++) | |
2450 | nvlist_free(schild[c]); | |
2451 | free(schild); | |
2452 | nvlist_free(split); | |
2453 | ||
2454 | spa_config_exit(spa, SCL_VDEV, FTAG); | |
2455 | ||
1e33ac1e | 2456 | (void) rw_enter(&zs->zs_name_lock, RW_WRITER); |
428870ff | 2457 | error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE); |
1e33ac1e | 2458 | (void) rw_exit(&zs->zs_name_lock); |
428870ff BB |
2459 | |
2460 | nvlist_free(config); | |
2461 | ||
2462 | if (error == 0) { | |
2463 | (void) printf("successful split - results:\n"); | |
2464 | mutex_enter(&spa_namespace_lock); | |
2465 | show_pool_stats(spa); | |
2466 | show_pool_stats(spa_lookup("splitp")); | |
2467 | mutex_exit(&spa_namespace_lock); | |
2468 | ++zs->zs_splits; | |
2469 | --zs->zs_mirrors; | |
2470 | } | |
1e33ac1e | 2471 | mutex_exit(&zs->zs_vdev_lock); |
428870ff | 2472 | |
34dc7c2f BB |
2473 | } |
2474 | ||
2475 | /* | |
2476 | * Verify that we can attach and detach devices. | |
2477 | */ | |
428870ff | 2478 | /* ARGSUSED */ |
34dc7c2f | 2479 | void |
428870ff | 2480 | ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id) |
34dc7c2f | 2481 | { |
428870ff BB |
2482 | ztest_shared_t *zs = ztest_shared; |
2483 | spa_t *spa = zs->zs_spa; | |
b128c09f | 2484 | spa_aux_vdev_t *sav = &spa->spa_spares; |
34dc7c2f BB |
2485 | vdev_t *rvd = spa->spa_root_vdev; |
2486 | vdev_t *oldvd, *newvd, *pvd; | |
b128c09f | 2487 | nvlist_t *root; |
428870ff | 2488 | uint64_t leaves; |
34dc7c2f BB |
2489 | uint64_t leaf, top; |
2490 | uint64_t ashift = ztest_get_ashift(); | |
fb5f0bc8 | 2491 | uint64_t oldguid, pguid; |
34dc7c2f | 2492 | size_t oldsize, newsize; |
40b84e7a | 2493 | char *oldpath, *newpath; |
34dc7c2f | 2494 | int replacing; |
b128c09f BB |
2495 | int oldvd_has_siblings = B_FALSE; |
2496 | int newvd_is_spare = B_FALSE; | |
2497 | int oldvd_is_log; | |
34dc7c2f | 2498 | int error, expected_error; |
34dc7c2f | 2499 | |
40b84e7a BB |
2500 | oldpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); |
2501 | newpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); | |
2502 | ||
1e33ac1e | 2503 | mutex_enter(&zs->zs_vdev_lock); |
428870ff | 2504 | leaves = MAX(zs->zs_mirrors, 1) * zopt_raidz; |
34dc7c2f | 2505 | |
b128c09f | 2506 | spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); |
34dc7c2f BB |
2507 | |
2508 | /* | |
2509 | * Decide whether to do an attach or a replace. | |
2510 | */ | |
2511 | replacing = ztest_random(2); | |
2512 | ||
2513 | /* | |
2514 | * Pick a random top-level vdev. | |
2515 | */ | |
428870ff | 2516 | top = ztest_random_vdev_top(spa, B_TRUE); |
34dc7c2f BB |
2517 | |
2518 | /* | |
2519 | * Pick a random leaf within it. | |
2520 | */ | |
2521 | leaf = ztest_random(leaves); | |
2522 | ||
2523 | /* | |
b128c09f | 2524 | * Locate this vdev. |
34dc7c2f | 2525 | */ |
b128c09f | 2526 | oldvd = rvd->vdev_child[top]; |
428870ff | 2527 | if (zs->zs_mirrors >= 1) { |
fb5f0bc8 | 2528 | ASSERT(oldvd->vdev_ops == &vdev_mirror_ops); |
428870ff | 2529 | ASSERT(oldvd->vdev_children >= zs->zs_mirrors); |
b128c09f | 2530 | oldvd = oldvd->vdev_child[leaf / zopt_raidz]; |
fb5f0bc8 BB |
2531 | } |
2532 | if (zopt_raidz > 1) { | |
2533 | ASSERT(oldvd->vdev_ops == &vdev_raidz_ops); | |
2534 | ASSERT(oldvd->vdev_children == zopt_raidz); | |
b128c09f | 2535 | oldvd = oldvd->vdev_child[leaf % zopt_raidz]; |
fb5f0bc8 | 2536 | } |
34dc7c2f BB |
2537 | |
2538 | /* | |
b128c09f BB |
2539 | * If we're already doing an attach or replace, oldvd may be a |
2540 | * mirror vdev -- in which case, pick a random child. | |
34dc7c2f | 2541 | */ |
b128c09f BB |
2542 | while (oldvd->vdev_children != 0) { |
2543 | oldvd_has_siblings = B_TRUE; | |
fb5f0bc8 BB |
2544 | ASSERT(oldvd->vdev_children >= 2); |
2545 | oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)]; | |
b128c09f BB |
2546 | } |
2547 | ||
2548 | oldguid = oldvd->vdev_guid; | |
9babb374 | 2549 | oldsize = vdev_get_min_asize(oldvd); |
b128c09f BB |
2550 | oldvd_is_log = oldvd->vdev_top->vdev_islog; |
2551 | (void) strcpy(oldpath, oldvd->vdev_path); | |
2552 | pvd = oldvd->vdev_parent; | |
fb5f0bc8 | 2553 | pguid = pvd->vdev_guid; |
34dc7c2f BB |
2554 | |
2555 | /* | |
b128c09f | 2556 | * If oldvd has siblings, then half of the time, detach it. |
34dc7c2f | 2557 | */ |
b128c09f BB |
2558 | if (oldvd_has_siblings && ztest_random(2) == 0) { |
2559 | spa_config_exit(spa, SCL_VDEV, FTAG); | |
fb5f0bc8 BB |
2560 | error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE); |
2561 | if (error != 0 && error != ENODEV && error != EBUSY && | |
2562 | error != ENOTSUP) | |
2563 | fatal(0, "detach (%s) returned %d", oldpath, error); | |
40b84e7a | 2564 | goto out; |
b128c09f | 2565 | } |
34dc7c2f BB |
2566 | |
2567 | /* | |
b128c09f BB |
2568 | * For the new vdev, choose with equal probability between the two |
2569 | * standard paths (ending in either 'a' or 'b') or a random hot spare. | |
34dc7c2f | 2570 | */ |
b128c09f BB |
2571 | if (sav->sav_count != 0 && ztest_random(3) == 0) { |
2572 | newvd = sav->sav_vdevs[ztest_random(sav->sav_count)]; | |
2573 | newvd_is_spare = B_TRUE; | |
2574 | (void) strcpy(newpath, newvd->vdev_path); | |
2575 | } else { | |
6aec1cd5 | 2576 | (void) snprintf(newpath, MAXPATHLEN, ztest_dev_template, |
b128c09f BB |
2577 | zopt_dir, zopt_pool, top * leaves + leaf); |
2578 | if (ztest_random(2) == 0) | |
2579 | newpath[strlen(newpath) - 1] = 'b'; | |
2580 | newvd = vdev_lookup_by_path(rvd, newpath); | |
2581 | } | |
2582 | ||
2583 | if (newvd) { | |
9babb374 | 2584 | newsize = vdev_get_min_asize(newvd); |
b128c09f BB |
2585 | } else { |
2586 | /* | |
2587 | * Make newsize a little bigger or smaller than oldsize. | |
2588 | * If it's smaller, the attach should fail. | |
2589 | * If it's larger, and we're doing a replace, | |
2590 | * we should get dynamic LUN growth when we're done. | |
2591 | */ | |
2592 | newsize = 10 * oldsize / (9 + ztest_random(3)); | |
2593 | } | |
34dc7c2f BB |
2594 | |
2595 | /* | |
2596 | * If pvd is not a mirror or root, the attach should fail with ENOTSUP, | |
2597 | * unless it's a replace; in that case any non-replacing parent is OK. | |
2598 | * | |
2599 | * If newvd is already part of the pool, it should fail with EBUSY. | |
2600 | * | |
2601 | * If newvd is too small, it should fail with EOVERFLOW. | |
2602 | */ | |
b128c09f BB |
2603 | if (pvd->vdev_ops != &vdev_mirror_ops && |
2604 | pvd->vdev_ops != &vdev_root_ops && (!replacing || | |
2605 | pvd->vdev_ops == &vdev_replacing_ops || | |
2606 | pvd->vdev_ops == &vdev_spare_ops)) | |
34dc7c2f | 2607 | expected_error = ENOTSUP; |
b128c09f BB |
2608 | else if (newvd_is_spare && (!replacing || oldvd_is_log)) |
2609 | expected_error = ENOTSUP; | |
2610 | else if (newvd == oldvd) | |
2611 | expected_error = replacing ? 0 : EBUSY; | |
2612 | else if (vdev_lookup_by_path(rvd, newpath) != NULL) | |
2613 | expected_error = EBUSY; | |
34dc7c2f BB |
2614 | else if (newsize < oldsize) |
2615 | expected_error = EOVERFLOW; | |
2616 | else if (ashift > oldvd->vdev_top->vdev_ashift) | |
2617 | expected_error = EDOM; | |
2618 | else | |
2619 | expected_error = 0; | |
2620 | ||
b128c09f | 2621 | spa_config_exit(spa, SCL_VDEV, FTAG); |
34dc7c2f BB |
2622 | |
2623 | /* | |
2624 | * Build the nvlist describing newpath. | |
2625 | */ | |
b128c09f BB |
2626 | root = make_vdev_root(newpath, NULL, newvd == NULL ? newsize : 0, |
2627 | ashift, 0, 0, 0, 1); | |
34dc7c2f | 2628 | |
b128c09f | 2629 | error = spa_vdev_attach(spa, oldguid, root, replacing); |
34dc7c2f | 2630 | |
34dc7c2f BB |
2631 | nvlist_free(root); |
2632 | ||
2633 | /* | |
2634 | * If our parent was the replacing vdev, but the replace completed, | |
2635 | * then instead of failing with ENOTSUP we may either succeed, | |
2636 | * fail with ENODEV, or fail with EOVERFLOW. | |
2637 | */ | |
2638 | if (expected_error == ENOTSUP && | |
2639 | (error == 0 || error == ENODEV || error == EOVERFLOW)) | |
2640 | expected_error = error; | |
2641 | ||
2642 | /* | |
2643 | * If someone grew the LUN, the replacement may be too small. | |
2644 | */ | |
b128c09f | 2645 | if (error == EOVERFLOW || error == EBUSY) |
34dc7c2f BB |
2646 | expected_error = error; |
2647 | ||
b128c09f BB |
2648 | /* XXX workaround 6690467 */ |
2649 | if (error != expected_error && expected_error != EBUSY) { | |
2650 | fatal(0, "attach (%s %llu, %s %llu, %d) " | |
2651 | "returned %d, expected %d", | |
2652 | oldpath, (longlong_t)oldsize, newpath, | |
2653 | (longlong_t)newsize, replacing, error, expected_error); | |
34dc7c2f | 2654 | } |
40b84e7a | 2655 | out: |
1e33ac1e | 2656 | mutex_exit(&zs->zs_vdev_lock); |
40b84e7a BB |
2657 | |
2658 | umem_free(oldpath, MAXPATHLEN); | |
2659 | umem_free(newpath, MAXPATHLEN); | |
34dc7c2f BB |
2660 | } |
2661 | ||
9babb374 BB |
2662 | /* |
2663 | * Callback function which expands the physical size of the vdev. | |
2664 | */ | |
2665 | vdev_t * | |
2666 | grow_vdev(vdev_t *vd, void *arg) | |
2667 | { | |
1fde1e37 | 2668 | ASSERTV(spa_t *spa = vd->vdev_spa); |
9babb374 BB |
2669 | size_t *newsize = arg; |
2670 | size_t fsize; | |
2671 | int fd; | |
2672 | ||
2673 | ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE); | |
2674 | ASSERT(vd->vdev_ops->vdev_op_leaf); | |
2675 | ||
2676 | if ((fd = open(vd->vdev_path, O_RDWR)) == -1) | |
2677 | return (vd); | |
2678 | ||
2679 | fsize = lseek(fd, 0, SEEK_END); | |
0e5b68e0 | 2680 | VERIFY(ftruncate(fd, *newsize) == 0); |
9babb374 BB |
2681 | |
2682 | if (zopt_verbose >= 6) { | |
2683 | (void) printf("%s grew from %lu to %lu bytes\n", | |
2684 | vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize); | |
2685 | } | |
2686 | (void) close(fd); | |
2687 | return (NULL); | |
2688 | } | |
2689 | ||
2690 | /* | |
2691 | * Callback function which expands a given vdev by calling vdev_online(). | |
2692 | */ | |
2693 | /* ARGSUSED */ | |
2694 | vdev_t * | |
2695 | online_vdev(vdev_t *vd, void *arg) | |
2696 | { | |
2697 | spa_t *spa = vd->vdev_spa; | |
2698 | vdev_t *tvd = vd->vdev_top; | |
9babb374 | 2699 | uint64_t guid = vd->vdev_guid; |
428870ff BB |
2700 | uint64_t generation = spa->spa_config_generation + 1; |
2701 | vdev_state_t newstate = VDEV_STATE_UNKNOWN; | |
2702 | int error; | |
9babb374 BB |
2703 | |
2704 | ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE); | |
2705 | ASSERT(vd->vdev_ops->vdev_op_leaf); | |
2706 | ||
2707 | /* Calling vdev_online will initialize the new metaslabs */ | |
2708 | spa_config_exit(spa, SCL_STATE, spa); | |
428870ff | 2709 | error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate); |
9babb374 BB |
2710 | spa_config_enter(spa, SCL_STATE, spa, RW_READER); |
2711 | ||
428870ff BB |
2712 | /* |
2713 | * If vdev_online returned an error or the underlying vdev_open | |
2714 | * failed then we abort the expand. The only way to know that | |
2715 | * vdev_open fails is by checking the returned newstate. | |
2716 | */ | |
2717 | if (error || newstate != VDEV_STATE_HEALTHY) { | |
2718 | if (zopt_verbose >= 5) { | |
2719 | (void) printf("Unable to expand vdev, state %llu, " | |
2720 | "error %d\n", (u_longlong_t)newstate, error); | |
2721 | } | |
2722 | return (vd); | |
2723 | } | |
2724 | ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY); | |
2725 | ||
9babb374 BB |
2726 | /* |
2727 | * Since we dropped the lock we need to ensure that we're | |
2728 | * still talking to the original vdev. It's possible this | |
2729 | * vdev may have been detached/replaced while we were | |
2730 | * trying to online it. | |
2731 | */ | |
428870ff BB |
2732 | if (generation != spa->spa_config_generation) { |
2733 | if (zopt_verbose >= 5) { | |
2734 | (void) printf("vdev configuration has changed, " | |
2735 | "guid %llu, state %llu, expected gen %llu, " | |
2736 | "got gen %llu\n", | |
2737 | (u_longlong_t)guid, | |
2738 | (u_longlong_t)tvd->vdev_state, | |
2739 | (u_longlong_t)generation, | |
2740 | (u_longlong_t)spa->spa_config_generation); | |
9babb374 BB |
2741 | } |
2742 | return (vd); | |
2743 | } | |
2744 | return (NULL); | |
2745 | } | |
2746 | ||
2747 | /* | |
2748 | * Traverse the vdev tree calling the supplied function. | |
2749 | * We continue to walk the tree until we either have walked all | |
2750 | * children or we receive a non-NULL return from the callback. | |
2751 | * If a NULL callback is passed, then we just return back the first | |
2752 | * leaf vdev we encounter. | |
2753 | */ | |
2754 | vdev_t * | |
2755 | vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg) | |
2756 | { | |
d6320ddb BB |
2757 | uint_t c; |
2758 | ||
9babb374 BB |
2759 | if (vd->vdev_ops->vdev_op_leaf) { |
2760 | if (func == NULL) | |
2761 | return (vd); | |
2762 | else | |
2763 | return (func(vd, arg)); | |
2764 | } | |
2765 | ||
d6320ddb | 2766 | for (c = 0; c < vd->vdev_children; c++) { |
9babb374 BB |
2767 | vdev_t *cvd = vd->vdev_child[c]; |
2768 | if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL) | |
2769 | return (cvd); | |
2770 | } | |
2771 | return (NULL); | |
2772 | } | |
2773 | ||
34dc7c2f BB |
2774 | /* |
2775 | * Verify that dynamic LUN growth works as expected. | |
2776 | */ | |
428870ff | 2777 | /* ARGSUSED */ |
34dc7c2f | 2778 | void |
428870ff | 2779 | ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id) |
34dc7c2f | 2780 | { |
428870ff BB |
2781 | ztest_shared_t *zs = ztest_shared; |
2782 | spa_t *spa = zs->zs_spa; | |
2783 | vdev_t *vd, *tvd; | |
2784 | metaslab_class_t *mc; | |
2785 | metaslab_group_t *mg; | |
9babb374 | 2786 | size_t psize, newsize; |
428870ff BB |
2787 | uint64_t top; |
2788 | uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count; | |
34dc7c2f | 2789 | |
1e33ac1e | 2790 | mutex_enter(&zs->zs_vdev_lock); |
9babb374 BB |
2791 | spa_config_enter(spa, SCL_STATE, spa, RW_READER); |
2792 | ||
428870ff | 2793 | top = ztest_random_vdev_top(spa, B_TRUE); |
9babb374 | 2794 | |
428870ff BB |
2795 | tvd = spa->spa_root_vdev->vdev_child[top]; |
2796 | mg = tvd->vdev_mg; | |
2797 | mc = mg->mg_class; | |
2798 | old_ms_count = tvd->vdev_ms_count; | |
2799 | old_class_space = metaslab_class_get_space(mc); | |
34dc7c2f BB |
2800 | |
2801 | /* | |
9babb374 BB |
2802 | * Determine the size of the first leaf vdev associated with |
2803 | * our top-level device. | |
34dc7c2f | 2804 | */ |
9babb374 BB |
2805 | vd = vdev_walk_tree(tvd, NULL, NULL); |
2806 | ASSERT3P(vd, !=, NULL); | |
2807 | ASSERT(vd->vdev_ops->vdev_op_leaf); | |
34dc7c2f | 2808 | |
9babb374 | 2809 | psize = vd->vdev_psize; |
34dc7c2f | 2810 | |
9babb374 | 2811 | /* |
428870ff BB |
2812 | * We only try to expand the vdev if it's healthy, less than 4x its |
2813 | * original size, and it has a valid psize. | |
9babb374 | 2814 | */ |
428870ff BB |
2815 | if (tvd->vdev_state != VDEV_STATE_HEALTHY || |
2816 | psize == 0 || psize >= 4 * zopt_vdev_size) { | |
9babb374 | 2817 | spa_config_exit(spa, SCL_STATE, spa); |
1e33ac1e | 2818 | mutex_exit(&zs->zs_vdev_lock); |
9babb374 BB |
2819 | return; |
2820 | } | |
2821 | ASSERT(psize > 0); | |
2822 | newsize = psize + psize / 8; | |
2823 | ASSERT3U(newsize, >, psize); | |
34dc7c2f | 2824 | |
9babb374 | 2825 | if (zopt_verbose >= 6) { |
428870ff | 2826 | (void) printf("Expanding LUN %s from %lu to %lu\n", |
9babb374 BB |
2827 | vd->vdev_path, (ulong_t)psize, (ulong_t)newsize); |
2828 | } | |
2829 | ||
9babb374 BB |
2830 | /* |
2831 | * Growing the vdev is a two step process: | |
2832 | * 1). expand the physical size (i.e. relabel) | |
2833 | * 2). online the vdev to create the new metaslabs | |
2834 | */ | |
2835 | if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL || | |
2836 | vdev_walk_tree(tvd, online_vdev, NULL) != NULL || | |
2837 | tvd->vdev_state != VDEV_STATE_HEALTHY) { | |
2838 | if (zopt_verbose >= 5) { | |
2839 | (void) printf("Could not expand LUN because " | |
428870ff | 2840 | "the vdev configuration changed.\n"); |
34dc7c2f | 2841 | } |
428870ff | 2842 | spa_config_exit(spa, SCL_STATE, spa); |
1e33ac1e | 2843 | mutex_exit(&zs->zs_vdev_lock); |
9babb374 | 2844 | return; |
34dc7c2f BB |
2845 | } |
2846 | ||
428870ff | 2847 | spa_config_exit(spa, SCL_STATE, spa); |
9babb374 BB |
2848 | |
2849 | /* | |
2850 | * Expanding the LUN will update the config asynchronously, | |
2851 | * thus we must wait for the async thread to complete any | |
2852 | * pending tasks before proceeding. | |
2853 | */ | |
428870ff BB |
2854 | for (;;) { |
2855 | boolean_t done; | |
2856 | mutex_enter(&spa->spa_async_lock); | |
2857 | done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks); | |
2858 | mutex_exit(&spa->spa_async_lock); | |
2859 | if (done) | |
2860 | break; | |
2861 | txg_wait_synced(spa_get_dsl(spa), 0); | |
2862 | (void) poll(NULL, 0, 100); | |
2863 | } | |
9babb374 BB |
2864 | |
2865 | spa_config_enter(spa, SCL_STATE, spa, RW_READER); | |
428870ff BB |
2866 | |
2867 | tvd = spa->spa_root_vdev->vdev_child[top]; | |
2868 | new_ms_count = tvd->vdev_ms_count; | |
2869 | new_class_space = metaslab_class_get_space(mc); | |
2870 | ||
2871 | if (tvd->vdev_mg != mg || mg->mg_class != mc) { | |
2872 | if (zopt_verbose >= 5) { | |
2873 | (void) printf("Could not verify LUN expansion due to " | |
2874 | "intervening vdev offline or remove.\n"); | |
2875 | } | |
2876 | spa_config_exit(spa, SCL_STATE, spa); | |
1e33ac1e | 2877 | mutex_exit(&zs->zs_vdev_lock); |
428870ff BB |
2878 | return; |
2879 | } | |
2880 | ||
2881 | /* | |
2882 | * Make sure we were able to grow the vdev. | |
2883 | */ | |
2884 | if (new_ms_count <= old_ms_count) | |
2885 | fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n", | |
2886 | old_ms_count, new_ms_count); | |
9babb374 BB |
2887 | |
2888 | /* | |
2889 | * Make sure we were able to grow the pool. | |
2890 | */ | |
428870ff BB |
2891 | if (new_class_space <= old_class_space) |
2892 | fatal(0, "LUN expansion failed: class_space %llu <= %llu\n", | |
2893 | old_class_space, new_class_space); | |
2894 | ||
2895 | if (zopt_verbose >= 5) { | |
9babb374 BB |
2896 | char oldnumbuf[6], newnumbuf[6]; |
2897 | ||
428870ff BB |
2898 | nicenum(old_class_space, oldnumbuf); |
2899 | nicenum(new_class_space, newnumbuf); | |
9babb374 BB |
2900 | (void) printf("%s grew from %s to %s\n", |
2901 | spa->spa_name, oldnumbuf, newnumbuf); | |
2902 | } | |
428870ff | 2903 | |
9babb374 | 2904 | spa_config_exit(spa, SCL_STATE, spa); |
1e33ac1e | 2905 | mutex_exit(&zs->zs_vdev_lock); |
34dc7c2f BB |
2906 | } |
2907 | ||
428870ff BB |
2908 | /* |
2909 | * Verify that dmu_objset_{create,destroy,open,close} work as expected. | |
2910 | */ | |
34dc7c2f BB |
2911 | /* ARGSUSED */ |
2912 | static void | |
428870ff | 2913 | ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx) |
34dc7c2f BB |
2914 | { |
2915 | /* | |
428870ff | 2916 | * Create the objects common to all ztest datasets. |
34dc7c2f | 2917 | */ |
428870ff | 2918 | VERIFY(zap_create_claim(os, ZTEST_DIROBJ, |
34dc7c2f | 2919 | DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0); |
428870ff | 2920 | } |
34dc7c2f | 2921 | |
428870ff BB |
2922 | static int |
2923 | ztest_dataset_create(char *dsname) | |
2924 | { | |
2925 | uint64_t zilset = ztest_random(100); | |
2926 | int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0, | |
2927 | ztest_objset_create_cb, NULL); | |
2928 | ||
2929 | if (err || zilset < 80) | |
2930 | return (err); | |
2931 | ||
b815ff9a ED |
2932 | if (zopt_verbose >= 5) |
2933 | (void) printf("Setting dataset %s to sync always\n", dsname); | |
428870ff BB |
2934 | return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC, |
2935 | ZFS_SYNC_ALWAYS, B_FALSE)); | |
34dc7c2f BB |
2936 | } |
2937 | ||
428870ff | 2938 | /* ARGSUSED */ |
34dc7c2f | 2939 | static int |
428870ff | 2940 | ztest_objset_destroy_cb(const char *name, void *arg) |
34dc7c2f | 2941 | { |
34dc7c2f | 2942 | objset_t *os; |
428870ff | 2943 | dmu_object_info_t doi; |
34dc7c2f BB |
2944 | int error; |
2945 | ||
2946 | /* | |
2947 | * Verify that the dataset contains a directory object. | |
2948 | */ | |
428870ff BB |
2949 | VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os)); |
2950 | error = dmu_object_info(os, ZTEST_DIROBJ, &doi); | |
34dc7c2f BB |
2951 | if (error != ENOENT) { |
2952 | /* We could have crashed in the middle of destroying it */ | |
2953 | ASSERT3U(error, ==, 0); | |
428870ff BB |
2954 | ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER); |
2955 | ASSERT3S(doi.doi_physical_blocks_512, >=, 0); | |
34dc7c2f | 2956 | } |
428870ff | 2957 | dmu_objset_rele(os, FTAG); |
34dc7c2f BB |
2958 | |
2959 | /* | |
2960 | * Destroy the dataset. | |
2961 | */ | |
428870ff | 2962 | VERIFY3U(0, ==, dmu_objset_destroy(name, B_FALSE)); |
34dc7c2f BB |
2963 | return (0); |
2964 | } | |
2965 | ||
428870ff BB |
2966 | static boolean_t |
2967 | ztest_snapshot_create(char *osname, uint64_t id) | |
34dc7c2f | 2968 | { |
428870ff BB |
2969 | char snapname[MAXNAMELEN]; |
2970 | int error; | |
2971 | ||
2972 | (void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname, | |
2973 | (u_longlong_t)id); | |
2974 | ||
2975 | error = dmu_objset_snapshot(osname, strchr(snapname, '@') + 1, | |
572e2857 | 2976 | NULL, NULL, B_FALSE, B_FALSE, -1); |
428870ff BB |
2977 | if (error == ENOSPC) { |
2978 | ztest_record_enospc(FTAG); | |
2979 | return (B_FALSE); | |
2980 | } | |
2981 | if (error != 0 && error != EEXIST) | |
2982 | fatal(0, "ztest_snapshot_create(%s) = %d", snapname, error); | |
2983 | return (B_TRUE); | |
2984 | } | |
2985 | ||
2986 | static boolean_t | |
2987 | ztest_snapshot_destroy(char *osname, uint64_t id) | |
2988 | { | |
2989 | char snapname[MAXNAMELEN]; | |
2990 | int error; | |
2991 | ||
2992 | (void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname, | |
2993 | (u_longlong_t)id); | |
2994 | ||
2995 | error = dmu_objset_destroy(snapname, B_FALSE); | |
2996 | if (error != 0 && error != ENOENT) | |
2997 | fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error); | |
2998 | return (B_TRUE); | |
34dc7c2f BB |
2999 | } |
3000 | ||
428870ff | 3001 | /* ARGSUSED */ |
34dc7c2f | 3002 | void |
428870ff | 3003 | ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id) |
34dc7c2f | 3004 | { |
428870ff | 3005 | ztest_shared_t *zs = ztest_shared; |
40b84e7a | 3006 | ztest_ds_t *zdtmp; |
428870ff | 3007 | int iters; |
34dc7c2f | 3008 | int error; |
b128c09f | 3009 | objset_t *os, *os2; |
40b84e7a | 3010 | char *name; |
34dc7c2f | 3011 | zilog_t *zilog; |
d6320ddb | 3012 | int i; |
34dc7c2f | 3013 | |
40b84e7a BB |
3014 | zdtmp = umem_alloc(sizeof (ztest_ds_t), UMEM_NOFAIL); |
3015 | name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); | |
3016 | ||
1e33ac1e | 3017 | (void) rw_enter(&zs->zs_name_lock, RW_READER); |
34dc7c2f | 3018 | |
428870ff BB |
3019 | (void) snprintf(name, MAXNAMELEN, "%s/temp_%llu", |
3020 | zs->zs_pool, (u_longlong_t)id); | |
34dc7c2f BB |
3021 | |
3022 | /* | |
3023 | * If this dataset exists from a previous run, process its replay log | |
3024 | * half of the time. If we don't replay it, then dmu_objset_destroy() | |
428870ff | 3025 | * (invoked from ztest_objset_destroy_cb()) should just throw it away. |
34dc7c2f BB |
3026 | */ |
3027 | if (ztest_random(2) == 0 && | |
428870ff | 3028 | dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) { |
40b84e7a BB |
3029 | ztest_zd_init(zdtmp, os); |
3030 | zil_replay(os, zdtmp, ztest_replay_vector); | |
3031 | ztest_zd_fini(zdtmp); | |
428870ff | 3032 | dmu_objset_disown(os, FTAG); |
34dc7c2f BB |
3033 | } |
3034 | ||
3035 | /* | |
3036 | * There may be an old instance of the dataset we're about to | |
3037 | * create lying around from a previous run. If so, destroy it | |
3038 | * and all of its snapshots. | |
3039 | */ | |
428870ff | 3040 | (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL, |
34dc7c2f BB |
3041 | DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS); |
3042 | ||
3043 | /* | |
3044 | * Verify that the destroyed dataset is no longer in the namespace. | |
3045 | */ | |
428870ff | 3046 | VERIFY3U(ENOENT, ==, dmu_objset_hold(name, FTAG, &os)); |
34dc7c2f BB |
3047 | |
3048 | /* | |
3049 | * Verify that we can create a new dataset. | |
3050 | */ | |
428870ff | 3051 | error = ztest_dataset_create(name); |
34dc7c2f BB |
3052 | if (error) { |
3053 | if (error == ENOSPC) { | |
428870ff | 3054 | ztest_record_enospc(FTAG); |
40b84e7a | 3055 | goto out; |
34dc7c2f BB |
3056 | } |
3057 | fatal(0, "dmu_objset_create(%s) = %d", name, error); | |
3058 | } | |
3059 | ||
428870ff BB |
3060 | VERIFY3U(0, ==, |
3061 | dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os)); | |
3062 | ||
40b84e7a | 3063 | ztest_zd_init(zdtmp, os); |
34dc7c2f BB |
3064 | |
3065 | /* | |
3066 | * Open the intent log for it. | |
3067 | */ | |
428870ff | 3068 | zilog = zil_open(os, ztest_get_data); |
34dc7c2f BB |
3069 | |
3070 | /* | |
428870ff BB |
3071 | * Put some objects in there, do a little I/O to them, |
3072 | * and randomly take a couple of snapshots along the way. | |
34dc7c2f | 3073 | */ |
428870ff | 3074 | iters = ztest_random(5); |
d6320ddb | 3075 | for (i = 0; i < iters; i++) { |
40b84e7a | 3076 | ztest_dmu_object_alloc_free(zdtmp, id); |
428870ff BB |
3077 | if (ztest_random(iters) == 0) |
3078 | (void) ztest_snapshot_create(name, i); | |
34dc7c2f BB |
3079 | } |
3080 | ||
3081 | /* | |
3082 | * Verify that we cannot create an existing dataset. | |
3083 | */ | |
428870ff BB |
3084 | VERIFY3U(EEXIST, ==, |
3085 | dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL)); | |
34dc7c2f BB |
3086 | |
3087 | /* | |
428870ff | 3088 | * Verify that we can hold an objset that is also owned. |
b128c09f | 3089 | */ |
428870ff BB |
3090 | VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2)); |
3091 | dmu_objset_rele(os2, FTAG); | |
34dc7c2f | 3092 | |
428870ff BB |
3093 | /* |
3094 | * Verify that we cannot own an objset that is already owned. | |
3095 | */ | |
3096 | VERIFY3U(EBUSY, ==, | |
3097 | dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2)); | |
34dc7c2f | 3098 | |
428870ff BB |
3099 | zil_close(zilog); |
3100 | dmu_objset_disown(os, FTAG); | |
40b84e7a BB |
3101 | ztest_zd_fini(zdtmp); |
3102 | out: | |
1e33ac1e | 3103 | (void) rw_exit(&zs->zs_name_lock); |
40b84e7a BB |
3104 | |
3105 | umem_free(name, MAXNAMELEN); | |
3106 | umem_free(zdtmp, sizeof (ztest_ds_t)); | |
34dc7c2f BB |
3107 | } |
3108 | ||
3109 | /* | |
3110 | * Verify that dmu_snapshot_{create,destroy,open,close} work as expected. | |
3111 | */ | |
3112 | void | |
428870ff | 3113 | ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id) |
34dc7c2f | 3114 | { |
428870ff | 3115 | ztest_shared_t *zs = ztest_shared; |
34dc7c2f | 3116 | |
1e33ac1e | 3117 | (void) rw_enter(&zs->zs_name_lock, RW_READER); |
428870ff BB |
3118 | (void) ztest_snapshot_destroy(zd->zd_name, id); |
3119 | (void) ztest_snapshot_create(zd->zd_name, id); | |
1e33ac1e | 3120 | (void) rw_exit(&zs->zs_name_lock); |
34dc7c2f BB |
3121 | } |
3122 | ||
9babb374 BB |
3123 | /* |
3124 | * Cleanup non-standard snapshots and clones. | |
3125 | */ | |
3126 | void | |
428870ff | 3127 | ztest_dsl_dataset_cleanup(char *osname, uint64_t id) |
9babb374 | 3128 | { |
40b84e7a BB |
3129 | char *snap1name; |
3130 | char *clone1name; | |
3131 | char *snap2name; | |
3132 | char *clone2name; | |
3133 | char *snap3name; | |
9babb374 BB |
3134 | int error; |
3135 | ||
40b84e7a BB |
3136 | snap1name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); |
3137 | clone1name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); | |
3138 | snap2name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); | |
3139 | clone2name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); | |
3140 | snap3name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); | |
3141 | ||
b8864a23 BB |
3142 | (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", |
3143 | osname, (u_longlong_t)id); | |
3144 | (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", | |
3145 | osname, (u_longlong_t)id); | |
3146 | (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", | |
3147 | clone1name, (u_longlong_t)id); | |
3148 | (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", | |
3149 | osname, (u_longlong_t)id); | |
3150 | (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", | |
3151 | clone1name, (u_longlong_t)id); | |
9babb374 | 3152 | |
45d1cae3 | 3153 | error = dmu_objset_destroy(clone2name, B_FALSE); |
9babb374 BB |
3154 | if (error && error != ENOENT) |
3155 | fatal(0, "dmu_objset_destroy(%s) = %d", clone2name, error); | |
45d1cae3 | 3156 | error = dmu_objset_destroy(snap3name, B_FALSE); |
9babb374 BB |
3157 | if (error && error != ENOENT) |
3158 | fatal(0, "dmu_objset_destroy(%s) = %d", snap3name, error); | |
45d1cae3 | 3159 | error = dmu_objset_destroy(snap2name, B_FALSE); |
9babb374 BB |
3160 | if (error && error != ENOENT) |
3161 | fatal(0, "dmu_objset_destroy(%s) = %d", snap2name, error); | |
45d1cae3 | 3162 | error = dmu_objset_destroy(clone1name, B_FALSE); |
9babb374 BB |
3163 | if (error && error != ENOENT) |
3164 | fatal(0, "dmu_objset_destroy(%s) = %d", clone1name, error); | |
45d1cae3 | 3165 | error = dmu_objset_destroy(snap1name, B_FALSE); |
9babb374 BB |
3166 | if (error && error != ENOENT) |
3167 | fatal(0, "dmu_objset_destroy(%s) = %d", snap1name, error); | |
40b84e7a BB |
3168 | |
3169 | umem_free(snap1name, MAXNAMELEN); | |
3170 | umem_free(clone1name, MAXNAMELEN); | |
3171 | umem_free(snap2name, MAXNAMELEN); | |
3172 | umem_free(clone2name, MAXNAMELEN); | |
3173 | umem_free(snap3name, MAXNAMELEN); | |
9babb374 BB |
3174 | } |
3175 | ||
3176 | /* | |
3177 | * Verify dsl_dataset_promote handles EBUSY | |
3178 | */ | |
3179 | void | |
428870ff | 3180 | ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id) |
9babb374 | 3181 | { |
428870ff | 3182 | ztest_shared_t *zs = ztest_shared; |
9babb374 BB |
3183 | objset_t *clone; |
3184 | dsl_dataset_t *ds; | |
40b84e7a BB |
3185 | char *snap1name; |
3186 | char *clone1name; | |
3187 | char *snap2name; | |
3188 | char *clone2name; | |
3189 | char *snap3name; | |
428870ff BB |
3190 | char *osname = zd->zd_name; |
3191 | int error; | |
9babb374 | 3192 | |
40b84e7a BB |
3193 | snap1name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); |
3194 | clone1name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); | |
3195 | snap2name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); | |
3196 | clone2name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); | |
3197 | snap3name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); | |
3198 | ||
1e33ac1e | 3199 | (void) rw_enter(&zs->zs_name_lock, RW_READER); |
9babb374 | 3200 | |
428870ff | 3201 | ztest_dsl_dataset_cleanup(osname, id); |
9babb374 | 3202 | |
b8864a23 BB |
3203 | (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", |
3204 | osname, (u_longlong_t)id); | |
3205 | (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", | |
3206 | osname, (u_longlong_t)id); | |
3207 | (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", | |
3208 | clone1name, (u_longlong_t)id); | |
3209 | (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", | |
3210 | osname, (u_longlong_t)id); | |
3211 | (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", | |
3212 | clone1name, (u_longlong_t)id); | |
9babb374 BB |
3213 | |
3214 | error = dmu_objset_snapshot(osname, strchr(snap1name, '@')+1, | |
572e2857 | 3215 | NULL, NULL, B_FALSE, B_FALSE, -1); |
9babb374 BB |
3216 | if (error && error != EEXIST) { |
3217 | if (error == ENOSPC) { | |
428870ff | 3218 | ztest_record_enospc(FTAG); |
9babb374 BB |
3219 | goto out; |
3220 | } | |
3221 | fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error); | |
3222 | } | |
3223 | ||
428870ff | 3224 | error = dmu_objset_hold(snap1name, FTAG, &clone); |
9babb374 BB |
3225 | if (error) |
3226 | fatal(0, "dmu_open_snapshot(%s) = %d", snap1name, error); | |
3227 | ||
428870ff BB |
3228 | error = dmu_objset_clone(clone1name, dmu_objset_ds(clone), 0); |
3229 | dmu_objset_rele(clone, FTAG); | |
9babb374 BB |
3230 | if (error) { |
3231 | if (error == ENOSPC) { | |
428870ff | 3232 | ztest_record_enospc(FTAG); |
9babb374 BB |
3233 | goto out; |
3234 | } | |
3235 | fatal(0, "dmu_objset_create(%s) = %d", clone1name, error); | |
3236 | } | |
3237 | ||
428870ff | 3238 | error = dmu_objset_snapshot(clone1name, strchr(snap2name, '@')+1, |
572e2857 | 3239 | NULL, NULL, B_FALSE, B_FALSE, -1); |
428870ff BB |
3240 | if (error && error != EEXIST) { |
3241 | if (error == ENOSPC) { | |
3242 | ztest_record_enospc(FTAG); | |
3243 | goto out; | |
34dc7c2f | 3244 | } |
428870ff BB |
3245 | fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error); |
3246 | } | |
34dc7c2f | 3247 | |
428870ff | 3248 | error = dmu_objset_snapshot(clone1name, strchr(snap3name, '@')+1, |
572e2857 | 3249 | NULL, NULL, B_FALSE, B_FALSE, -1); |
428870ff BB |
3250 | if (error && error != EEXIST) { |
3251 | if (error == ENOSPC) { | |
3252 | ztest_record_enospc(FTAG); | |
3253 | goto out; | |
3254 | } | |
3255 | fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error); | |
3256 | } | |
34dc7c2f | 3257 | |
428870ff BB |
3258 | error = dmu_objset_hold(snap3name, FTAG, &clone); |
3259 | if (error) | |
3260 | fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error); | |
34dc7c2f | 3261 | |
428870ff BB |
3262 | error = dmu_objset_clone(clone2name, dmu_objset_ds(clone), 0); |
3263 | dmu_objset_rele(clone, FTAG); | |
3264 | if (error) { | |
3265 | if (error == ENOSPC) { | |
3266 | ztest_record_enospc(FTAG); | |
3267 | goto out; | |
3268 | } | |
3269 | fatal(0, "dmu_objset_create(%s) = %d", clone2name, error); | |
3270 | } | |
34dc7c2f | 3271 | |
428870ff BB |
3272 | error = dsl_dataset_own(snap2name, B_FALSE, FTAG, &ds); |
3273 | if (error) | |
3274 | fatal(0, "dsl_dataset_own(%s) = %d", snap2name, error); | |
3275 | error = dsl_dataset_promote(clone2name, NULL); | |
3276 | if (error != EBUSY) | |
3277 | fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name, | |
3278 | error); | |
3279 | dsl_dataset_disown(ds, FTAG); | |
34dc7c2f | 3280 | |
428870ff BB |
3281 | out: |
3282 | ztest_dsl_dataset_cleanup(osname, id); | |
34dc7c2f | 3283 | |
1e33ac1e | 3284 | (void) rw_exit(&zs->zs_name_lock); |
40b84e7a BB |
3285 | |
3286 | umem_free(snap1name, MAXNAMELEN); | |
3287 | umem_free(clone1name, MAXNAMELEN); | |
3288 | umem_free(snap2name, MAXNAMELEN); | |
3289 | umem_free(clone2name, MAXNAMELEN); | |
3290 | umem_free(snap3name, MAXNAMELEN); | |
428870ff | 3291 | } |
34dc7c2f | 3292 | |
40b84e7a BB |
3293 | #undef OD_ARRAY_SIZE |
3294 | #define OD_ARRAY_SIZE 4 | |
3295 | ||
428870ff BB |
3296 | /* |
3297 | * Verify that dmu_object_{alloc,free} work as expected. | |
3298 | */ | |
3299 | void | |
3300 | ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id) | |
3301 | { | |
40b84e7a BB |
3302 | ztest_od_t *od; |
3303 | int batchsize; | |
3304 | int size; | |
d6320ddb | 3305 | int b; |
34dc7c2f | 3306 | |
40b84e7a BB |
3307 | size = sizeof(ztest_od_t) * OD_ARRAY_SIZE; |
3308 | od = umem_alloc(size, UMEM_NOFAIL); | |
3309 | batchsize = OD_ARRAY_SIZE; | |
3310 | ||
d6320ddb | 3311 | for (b = 0; b < batchsize; b++) |
40b84e7a | 3312 | ztest_od_init(od + b, id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0); |
34dc7c2f | 3313 | |
428870ff BB |
3314 | /* |
3315 | * Destroy the previous batch of objects, create a new batch, | |
3316 | * and do some I/O on the new objects. | |
3317 | */ | |
40b84e7a | 3318 | if (ztest_object_init(zd, od, size, B_TRUE) != 0) |
428870ff | 3319 | return; |
34dc7c2f | 3320 | |
428870ff BB |
3321 | while (ztest_random(4 * batchsize) != 0) |
3322 | ztest_io(zd, od[ztest_random(batchsize)].od_object, | |
3323 | ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT); | |
40b84e7a BB |
3324 | |
3325 | umem_free(od, size); | |
34dc7c2f BB |
3326 | } |
3327 | ||
40b84e7a BB |
3328 | #undef OD_ARRAY_SIZE |
3329 | #define OD_ARRAY_SIZE 2 | |
3330 | ||
34dc7c2f BB |
3331 | /* |
3332 | * Verify that dmu_{read,write} work as expected. | |
3333 | */ | |
34dc7c2f | 3334 | void |
428870ff | 3335 | ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id) |
34dc7c2f | 3336 | { |
40b84e7a BB |
3337 | int size; |
3338 | ztest_od_t *od; | |
3339 | ||
428870ff | 3340 | objset_t *os = zd->zd_os; |
40b84e7a BB |
3341 | size = sizeof(ztest_od_t) * OD_ARRAY_SIZE; |
3342 | od = umem_alloc(size, UMEM_NOFAIL); | |
34dc7c2f BB |
3343 | dmu_tx_t *tx; |
3344 | int i, freeit, error; | |
3345 | uint64_t n, s, txg; | |
3346 | bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT; | |
428870ff BB |
3347 | uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize; |
3348 | uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t); | |
34dc7c2f BB |
3349 | uint64_t regions = 997; |
3350 | uint64_t stride = 123456789ULL; | |
3351 | uint64_t width = 40; | |
3352 | int free_percent = 5; | |
3353 | ||
3354 | /* | |
3355 | * This test uses two objects, packobj and bigobj, that are always | |
3356 | * updated together (i.e. in the same tx) so that their contents are | |
3357 | * in sync and can be compared. Their contents relate to each other | |
3358 | * in a simple way: packobj is a dense array of 'bufwad' structures, | |
3359 | * while bigobj is a sparse array of the same bufwads. Specifically, | |
3360 | * for any index n, there are three bufwads that should be identical: | |
3361 | * | |
3362 | * packobj, at offset n * sizeof (bufwad_t) | |
3363 | * bigobj, at the head of the nth chunk | |
3364 | * bigobj, at the tail of the nth chunk | |
3365 | * | |
3366 | * The chunk size is arbitrary. It doesn't have to be a power of two, | |
3367 | * and it doesn't have any relation to the object blocksize. | |
3368 | * The only requirement is that it can hold at least two bufwads. | |
3369 | * | |
3370 | * Normally, we write the bufwad to each of these locations. | |
3371 | * However, free_percent of the time we instead write zeroes to | |
3372 | * packobj and perform a dmu_free_range() on bigobj. By comparing | |
3373 | * bigobj to packobj, we can verify that the DMU is correctly | |
3374 | * tracking which parts of an object are allocated and free, | |
3375 | * and that the contents of the allocated blocks are correct. | |
3376 | */ | |
3377 | ||
3378 | /* | |
3379 | * Read the directory info. If it's the first time, set things up. | |
3380 | */ | |
40b84e7a BB |
3381 | ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize); |
3382 | ztest_od_init(od + 1, id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize); | |
34dc7c2f | 3383 | |
40b84e7a BB |
3384 | if (ztest_object_init(zd, od, size, B_FALSE) != 0) { |
3385 | umem_free(od, size); | |
428870ff | 3386 | return; |
40b84e7a | 3387 | } |
34dc7c2f | 3388 | |
428870ff BB |
3389 | bigobj = od[0].od_object; |
3390 | packobj = od[1].od_object; | |
3391 | chunksize = od[0].od_gen; | |
3392 | ASSERT(chunksize == od[1].od_gen); | |
34dc7c2f BB |
3393 | |
3394 | /* | |
3395 | * Prefetch a random chunk of the big object. | |
3396 | * Our aim here is to get some async reads in flight | |
3397 | * for blocks that we may free below; the DMU should | |
3398 | * handle this race correctly. | |
3399 | */ | |
3400 | n = ztest_random(regions) * stride + ztest_random(width); | |
3401 | s = 1 + ztest_random(2 * width - 1); | |
428870ff | 3402 | dmu_prefetch(os, bigobj, n * chunksize, s * chunksize); |
34dc7c2f BB |
3403 | |
3404 | /* | |
3405 | * Pick a random index and compute the offsets into packobj and bigobj. | |
3406 | */ | |
3407 | n = ztest_random(regions) * stride + ztest_random(width); | |
3408 | s = 1 + ztest_random(width - 1); | |
3409 | ||
3410 | packoff = n * sizeof (bufwad_t); | |
3411 | packsize = s * sizeof (bufwad_t); | |
3412 | ||
428870ff BB |
3413 | bigoff = n * chunksize; |
3414 | bigsize = s * chunksize; | |
34dc7c2f BB |
3415 | |
3416 | packbuf = umem_alloc(packsize, UMEM_NOFAIL); | |
3417 | bigbuf = umem_alloc(bigsize, UMEM_NOFAIL); | |
3418 | ||
3419 | /* | |
3420 | * free_percent of the time, free a range of bigobj rather than | |
3421 | * overwriting it. | |
3422 | */ | |
3423 | freeit = (ztest_random(100) < free_percent); | |
3424 | ||
3425 | /* | |
3426 | * Read the current contents of our objects. | |
3427 | */ | |
428870ff | 3428 | error = dmu_read(os, packobj, packoff, packsize, packbuf, |
9babb374 | 3429 | DMU_READ_PREFETCH); |
34dc7c2f | 3430 | ASSERT3U(error, ==, 0); |
428870ff | 3431 | error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf, |
9babb374 | 3432 | DMU_READ_PREFETCH); |
34dc7c2f BB |
3433 | ASSERT3U(error, ==, 0); |
3434 | ||
3435 | /* | |
3436 | * Get a tx for the mods to both packobj and bigobj. | |
3437 | */ | |
3438 | tx = dmu_tx_create(os); | |
3439 | ||
428870ff | 3440 | dmu_tx_hold_write(tx, packobj, packoff, packsize); |
34dc7c2f BB |
3441 | |
3442 | if (freeit) | |
428870ff | 3443 | dmu_tx_hold_free(tx, bigobj, bigoff, bigsize); |
34dc7c2f | 3444 | else |
428870ff | 3445 | dmu_tx_hold_write(tx, bigobj, bigoff, bigsize); |
34dc7c2f | 3446 | |
428870ff BB |
3447 | txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); |
3448 | if (txg == 0) { | |
34dc7c2f BB |
3449 | umem_free(packbuf, packsize); |
3450 | umem_free(bigbuf, bigsize); | |
40b84e7a | 3451 | umem_free(od, size); |
34dc7c2f BB |
3452 | return; |
3453 | } | |
3454 | ||
428870ff BB |
3455 | dmu_object_set_checksum(os, bigobj, |
3456 | (enum zio_checksum)ztest_random_dsl_prop(ZFS_PROP_CHECKSUM), tx); | |
3457 | ||
3458 | dmu_object_set_compress(os, bigobj, | |
3459 | (enum zio_compress)ztest_random_dsl_prop(ZFS_PROP_COMPRESSION), tx); | |
34dc7c2f BB |
3460 | |
3461 | /* | |
3462 | * For each index from n to n + s, verify that the existing bufwad | |
3463 | * in packobj matches the bufwads at the head and tail of the | |
3464 | * corresponding chunk in bigobj. Then update all three bufwads | |
3465 | * with the new values we want to write out. | |
3466 | */ | |
3467 | for (i = 0; i < s; i++) { | |
3468 | /* LINTED */ | |
3469 | pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t)); | |
3470 | /* LINTED */ | |
428870ff | 3471 | bigH = (bufwad_t *)((char *)bigbuf + i * chunksize); |
34dc7c2f | 3472 | /* LINTED */ |
428870ff | 3473 | bigT = (bufwad_t *)((char *)bigH + chunksize) - 1; |
34dc7c2f BB |
3474 | |
3475 | ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize); | |
3476 | ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize); | |
3477 | ||
3478 | if (pack->bw_txg > txg) | |
3479 | fatal(0, "future leak: got %llx, open txg is %llx", | |
3480 | pack->bw_txg, txg); | |
3481 | ||
3482 | if (pack->bw_data != 0 && pack->bw_index != n + i) | |
3483 | fatal(0, "wrong index: got %llx, wanted %llx+%llx", | |
3484 | pack->bw_index, n, i); | |
3485 | ||
3486 | if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0) | |
3487 | fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH); | |
3488 | ||
3489 | if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0) | |
3490 | fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT); | |
3491 | ||
3492 | if (freeit) { | |
3493 | bzero(pack, sizeof (bufwad_t)); | |
3494 | } else { | |
3495 | pack->bw_index = n + i; | |
3496 | pack->bw_txg = txg; | |
3497 | pack->bw_data = 1 + ztest_random(-2ULL); | |
3498 | } | |
3499 | *bigH = *pack; | |
3500 | *bigT = *pack; | |
3501 | } | |
3502 | ||
3503 | /* | |
3504 | * We've verified all the old bufwads, and made new ones. | |
3505 | * Now write them out. | |
3506 | */ | |
428870ff | 3507 | dmu_write(os, packobj, packoff, packsize, packbuf, tx); |
34dc7c2f BB |
3508 | |
3509 | if (freeit) { | |
428870ff | 3510 | if (zopt_verbose >= 7) { |
34dc7c2f BB |
3511 | (void) printf("freeing offset %llx size %llx" |
3512 | " txg %llx\n", | |
3513 | (u_longlong_t)bigoff, | |
3514 | (u_longlong_t)bigsize, | |
3515 | (u_longlong_t)txg); | |
3516 | } | |
428870ff | 3517 | VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx)); |
34dc7c2f | 3518 | } else { |
428870ff | 3519 | if (zopt_verbose >= 7) { |
34dc7c2f BB |
3520 | (void) printf("writing offset %llx size %llx" |
3521 | " txg %llx\n", | |
3522 | (u_longlong_t)bigoff, | |
3523 | (u_longlong_t)bigsize, | |
3524 | (u_longlong_t)txg); | |
3525 | } | |
428870ff | 3526 | dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx); |
34dc7c2f BB |
3527 | } |
3528 | ||
3529 | dmu_tx_commit(tx); | |
3530 | ||
3531 | /* | |
3532 | * Sanity check the stuff we just wrote. | |
3533 | */ | |
3534 | { | |
3535 | void *packcheck = umem_alloc(packsize, UMEM_NOFAIL); | |
3536 | void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL); | |
3537 | ||
428870ff | 3538 | VERIFY(0 == dmu_read(os, packobj, packoff, |
9babb374 | 3539 | packsize, packcheck, DMU_READ_PREFETCH)); |
428870ff | 3540 | VERIFY(0 == dmu_read(os, bigobj, bigoff, |
9babb374 | 3541 | bigsize, bigcheck, DMU_READ_PREFETCH)); |
34dc7c2f BB |
3542 | |
3543 | ASSERT(bcmp(packbuf, packcheck, packsize) == 0); | |
3544 | ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0); | |
3545 | ||
3546 | umem_free(packcheck, packsize); | |
3547 | umem_free(bigcheck, bigsize); | |
3548 | } | |
3549 | ||
3550 | umem_free(packbuf, packsize); | |
3551 | umem_free(bigbuf, bigsize); | |
40b84e7a | 3552 | umem_free(od, size); |
34dc7c2f BB |
3553 | } |
3554 | ||
9babb374 BB |
3555 | void |
3556 | compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf, | |
428870ff | 3557 | uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg) |
9babb374 BB |
3558 | { |
3559 | uint64_t i; | |
3560 | bufwad_t *pack; | |
3561 | bufwad_t *bigH; | |
3562 | bufwad_t *bigT; | |
3563 | ||
3564 | /* | |
3565 | * For each index from n to n + s, verify that the existing bufwad | |
3566 | * in packobj matches the bufwads at the head and tail of the | |
3567 | * corresponding chunk in bigobj. Then update all three bufwads | |
3568 | * with the new values we want to write out. | |
3569 | */ | |
3570 | for (i = 0; i < s; i++) { | |
3571 | /* LINTED */ | |
3572 | pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t)); | |
3573 | /* LINTED */ | |
428870ff | 3574 | bigH = (bufwad_t *)((char *)bigbuf + i * chunksize); |
9babb374 | 3575 | /* LINTED */ |
428870ff | 3576 | bigT = (bufwad_t *)((char *)bigH + chunksize) - 1; |
9babb374 BB |
3577 | |
3578 | ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize); | |
3579 | ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize); | |
3580 | ||
3581 | if (pack->bw_txg > txg) | |
3582 | fatal(0, "future leak: got %llx, open txg is %llx", | |
3583 | pack->bw_txg, txg); | |
3584 | ||
3585 | if (pack->bw_data != 0 && pack->bw_index != n + i) | |
3586 | fatal(0, "wrong index: got %llx, wanted %llx+%llx", | |
3587 | pack->bw_index, n, i); | |
3588 | ||
3589 | if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0) | |
3590 | fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH); | |
3591 | ||
3592 | if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0) | |
3593 | fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT); | |
3594 | ||
3595 | pack->bw_index = n + i; | |
3596 | pack->bw_txg = txg; | |
3597 | pack->bw_data = 1 + ztest_random(-2ULL); | |
3598 | ||
3599 | *bigH = *pack; | |
3600 | *bigT = *pack; | |
3601 | } | |
3602 | } | |
3603 | ||
40b84e7a BB |
3604 | #undef OD_ARRAY_SIZE |
3605 | #define OD_ARRAY_SIZE 2 | |
3606 | ||
9babb374 | 3607 | void |
428870ff | 3608 | ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id) |
9babb374 | 3609 | { |
428870ff | 3610 | objset_t *os = zd->zd_os; |
40b84e7a | 3611 | ztest_od_t *od; |
9babb374 BB |
3612 | dmu_tx_t *tx; |
3613 | uint64_t i; | |
3614 | int error; | |
40b84e7a | 3615 | int size; |
9babb374 BB |
3616 | uint64_t n, s, txg; |
3617 | bufwad_t *packbuf, *bigbuf; | |
428870ff BB |
3618 | uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize; |
3619 | uint64_t blocksize = ztest_random_blocksize(); | |
3620 | uint64_t chunksize = blocksize; | |
9babb374 BB |
3621 | uint64_t regions = 997; |
3622 | uint64_t stride = 123456789ULL; | |
3623 | uint64_t width = 9; | |
3624 | dmu_buf_t *bonus_db; | |
3625 | arc_buf_t **bigbuf_arcbufs; | |
428870ff | 3626 | dmu_object_info_t doi; |
9babb374 | 3627 | |
40b84e7a BB |
3628 | size = sizeof(ztest_od_t) * OD_ARRAY_SIZE; |
3629 | od = umem_alloc(size, UMEM_NOFAIL); | |
3630 | ||
9babb374 BB |
3631 | /* |
3632 | * This test uses two objects, packobj and bigobj, that are always | |
3633 | * updated together (i.e. in the same tx) so that their contents are | |
3634 | * in sync and can be compared. Their contents relate to each other | |
3635 | * in a simple way: packobj is a dense array of 'bufwad' structures, | |
3636 | * while bigobj is a sparse array of the same bufwads. Specifically, | |
3637 | * for any index n, there are three bufwads that should be identical: | |
3638 | * | |
3639 | * packobj, at offset n * sizeof (bufwad_t) | |
3640 | * bigobj, at the head of the nth chunk | |
3641 | * bigobj, at the tail of the nth chunk | |
3642 | * | |
3643 | * The chunk size is set equal to bigobj block size so that | |
3644 | * dmu_assign_arcbuf() can be tested for object updates. | |
3645 | */ | |
3646 | ||
3647 | /* | |
3648 | * Read the directory info. If it's the first time, set things up. | |
3649 | */ | |
40b84e7a BB |
3650 | ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0); |
3651 | ztest_od_init(od + 1, id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize); | |
3652 | ||
9babb374 | 3653 | |
40b84e7a BB |
3654 | if (ztest_object_init(zd, od, size, B_FALSE) != 0) { |
3655 | umem_free(od, size); | |
428870ff | 3656 | return; |
40b84e7a | 3657 | } |
9babb374 | 3658 | |
428870ff BB |
3659 | bigobj = od[0].od_object; |
3660 | packobj = od[1].od_object; | |
3661 | blocksize = od[0].od_blocksize; | |
3662 | chunksize = blocksize; | |
3663 | ASSERT(chunksize == od[1].od_gen); | |
9babb374 | 3664 | |
428870ff BB |
3665 | VERIFY(dmu_object_info(os, bigobj, &doi) == 0); |
3666 | VERIFY(ISP2(doi.doi_data_block_size)); | |
3667 | VERIFY(chunksize == doi.doi_data_block_size); | |
3668 | VERIFY(chunksize >= 2 * sizeof (bufwad_t)); | |
9babb374 BB |
3669 | |
3670 | /* | |
3671 | * Pick a random index and compute the offsets into packobj and bigobj. | |
3672 | */ | |
3673 | n = ztest_random(regions) * stride + ztest_random(width); | |
3674 | s = 1 + ztest_random(width - 1); | |
3675 | ||
3676 | packoff = n * sizeof (bufwad_t); | |
3677 | packsize = s * sizeof (bufwad_t); | |
3678 | ||
428870ff BB |
3679 | bigoff = n * chunksize; |
3680 | bigsize = s * chunksize; | |
9babb374 BB |
3681 | |
3682 | packbuf = umem_zalloc(packsize, UMEM_NOFAIL); | |
3683 | bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL); | |
3684 | ||
428870ff | 3685 | VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db)); |
9babb374 BB |
3686 | |
3687 | bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL); | |
3688 | ||
3689 | /* | |
3690 | * Iteration 0 test zcopy for DB_UNCACHED dbufs. | |
3691 | * Iteration 1 test zcopy to already referenced dbufs. | |
3692 | * Iteration 2 test zcopy to dirty dbuf in the same txg. | |
3693 | * Iteration 3 test zcopy to dbuf dirty in previous txg. | |
3694 | * Iteration 4 test zcopy when dbuf is no longer dirty. | |
3695 | * Iteration 5 test zcopy when it can't be done. | |
3696 | * Iteration 6 one more zcopy write. | |
3697 | */ | |
3698 | for (i = 0; i < 7; i++) { | |
3699 | uint64_t j; | |
3700 | uint64_t off; | |
3701 | ||
3702 | /* | |
3703 | * In iteration 5 (i == 5) use arcbufs | |
3704 | * that don't match bigobj blksz to test | |
3705 | * dmu_assign_arcbuf() when it can't directly | |
3706 | * assign an arcbuf to a dbuf. | |
3707 | */ | |
3708 | for (j = 0; j < s; j++) { | |
3709 | if (i != 5) { | |
3710 | bigbuf_arcbufs[j] = | |
428870ff | 3711 | dmu_request_arcbuf(bonus_db, chunksize); |
9babb374 BB |
3712 | } else { |
3713 | bigbuf_arcbufs[2 * j] = | |
428870ff | 3714 | dmu_request_arcbuf(bonus_db, chunksize / 2); |
9babb374 | 3715 | bigbuf_arcbufs[2 * j + 1] = |
428870ff | 3716 | dmu_request_arcbuf(bonus_db, chunksize / 2); |
9babb374 BB |
3717 | } |
3718 | } | |
3719 | ||
3720 | /* | |
3721 | * Get a tx for the mods to both packobj and bigobj. | |
3722 | */ | |
3723 | tx = dmu_tx_create(os); | |
3724 | ||
428870ff BB |
3725 | dmu_tx_hold_write(tx, packobj, packoff, packsize); |
3726 | dmu_tx_hold_write(tx, bigobj, bigoff, bigsize); | |
9babb374 | 3727 | |
428870ff BB |
3728 | txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); |
3729 | if (txg == 0) { | |
9babb374 BB |
3730 | umem_free(packbuf, packsize); |
3731 | umem_free(bigbuf, bigsize); | |
3732 | for (j = 0; j < s; j++) { | |
3733 | if (i != 5) { | |
3734 | dmu_return_arcbuf(bigbuf_arcbufs[j]); | |
3735 | } else { | |
3736 | dmu_return_arcbuf( | |
3737 | bigbuf_arcbufs[2 * j]); | |
3738 | dmu_return_arcbuf( | |
3739 | bigbuf_arcbufs[2 * j + 1]); | |
3740 | } | |
3741 | } | |
3742 | umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *)); | |
40b84e7a | 3743 | umem_free(od, size); |
9babb374 BB |
3744 | dmu_buf_rele(bonus_db, FTAG); |
3745 | return; | |
3746 | } | |
3747 | ||
9babb374 BB |
3748 | /* |
3749 | * 50% of the time don't read objects in the 1st iteration to | |
3750 | * test dmu_assign_arcbuf() for the case when there're no | |
3751 | * existing dbufs for the specified offsets. | |
3752 | */ | |
3753 | if (i != 0 || ztest_random(2) != 0) { | |
428870ff | 3754 | error = dmu_read(os, packobj, packoff, |
9babb374 BB |
3755 | packsize, packbuf, DMU_READ_PREFETCH); |
3756 | ASSERT3U(error, ==, 0); | |
428870ff | 3757 | error = dmu_read(os, bigobj, bigoff, bigsize, |
9babb374 BB |
3758 | bigbuf, DMU_READ_PREFETCH); |
3759 | ASSERT3U(error, ==, 0); | |
3760 | } | |
3761 | compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize, | |
428870ff | 3762 | n, chunksize, txg); |
9babb374 BB |
3763 | |
3764 | /* | |
3765 | * We've verified all the old bufwads, and made new ones. | |
3766 | * Now write them out. | |
3767 | */ | |
428870ff BB |
3768 | dmu_write(os, packobj, packoff, packsize, packbuf, tx); |
3769 | if (zopt_verbose >= 7) { | |
9babb374 BB |
3770 | (void) printf("writing offset %llx size %llx" |
3771 | " txg %llx\n", | |
3772 | (u_longlong_t)bigoff, | |
3773 | (u_longlong_t)bigsize, | |
3774 | (u_longlong_t)txg); | |
3775 | } | |
428870ff | 3776 | for (off = bigoff, j = 0; j < s; j++, off += chunksize) { |
9babb374 BB |
3777 | dmu_buf_t *dbt; |
3778 | if (i != 5) { | |
3779 | bcopy((caddr_t)bigbuf + (off - bigoff), | |
428870ff | 3780 | bigbuf_arcbufs[j]->b_data, chunksize); |
9babb374 BB |
3781 | } else { |
3782 | bcopy((caddr_t)bigbuf + (off - bigoff), | |
3783 | bigbuf_arcbufs[2 * j]->b_data, | |
428870ff | 3784 | chunksize / 2); |
9babb374 | 3785 | bcopy((caddr_t)bigbuf + (off - bigoff) + |
428870ff | 3786 | chunksize / 2, |
9babb374 | 3787 | bigbuf_arcbufs[2 * j + 1]->b_data, |
428870ff | 3788 | chunksize / 2); |
9babb374 BB |
3789 | } |
3790 | ||
3791 | if (i == 1) { | |
428870ff BB |
3792 | VERIFY(dmu_buf_hold(os, bigobj, off, |
3793 | FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0); | |
9babb374 BB |
3794 | } |
3795 | if (i != 5) { | |
3796 | dmu_assign_arcbuf(bonus_db, off, | |
3797 | bigbuf_arcbufs[j], tx); | |
3798 | } else { | |
3799 | dmu_assign_arcbuf(bonus_db, off, | |
3800 | bigbuf_arcbufs[2 * j], tx); | |
3801 | dmu_assign_arcbuf(bonus_db, | |
428870ff | 3802 | off + chunksize / 2, |
9babb374 BB |
3803 | bigbuf_arcbufs[2 * j + 1], tx); |
3804 | } | |
3805 | if (i == 1) { | |
3806 | dmu_buf_rele(dbt, FTAG); | |
3807 | } | |
3808 | } | |
3809 | dmu_tx_commit(tx); | |
3810 | ||
3811 | /* | |
3812 | * Sanity check the stuff we just wrote. | |
3813 | */ | |
3814 | { | |
3815 | void *packcheck = umem_alloc(packsize, UMEM_NOFAIL); | |
3816 | void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL); | |
3817 | ||
428870ff | 3818 | VERIFY(0 == dmu_read(os, packobj, packoff, |
9babb374 | 3819 | packsize, packcheck, DMU_READ_PREFETCH)); |
428870ff | 3820 | VERIFY(0 == dmu_read(os, bigobj, bigoff, |
9babb374 BB |
3821 | bigsize, bigcheck, DMU_READ_PREFETCH)); |
3822 | ||
3823 | ASSERT(bcmp(packbuf, packcheck, packsize) == 0); | |
3824 | ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0); | |
3825 | ||
3826 | umem_free(packcheck, packsize); | |
3827 | umem_free(bigcheck, bigsize); | |
3828 | } | |
3829 | if (i == 2) { | |
3830 | txg_wait_open(dmu_objset_pool(os), 0); | |
3831 | } else if (i == 3) { | |
3832 | txg_wait_synced(dmu_objset_pool(os), 0); | |
3833 | } | |
3834 | } | |
3835 | ||
3836 | dmu_buf_rele(bonus_db, FTAG); | |
3837 | umem_free(packbuf, packsize); | |
3838 | umem_free(bigbuf, bigsize); | |
3839 | umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *)); | |
40b84e7a | 3840 | umem_free(od, size); |
9babb374 BB |
3841 | } |
3842 | ||
428870ff | 3843 | /* ARGSUSED */ |
34dc7c2f | 3844 | void |
428870ff | 3845 | ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id) |
34dc7c2f | 3846 | { |
40b84e7a BB |
3847 | ztest_od_t *od; |
3848 | ||
3849 | od = umem_alloc(sizeof(ztest_od_t), UMEM_NOFAIL); | |
428870ff BB |
3850 | uint64_t offset = (1ULL << (ztest_random(20) + 43)) + |
3851 | (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT); | |
34dc7c2f BB |
3852 | |
3853 | /* | |
428870ff BB |
3854 | * Have multiple threads write to large offsets in an object |
3855 | * to verify that parallel writes to an object -- even to the | |
3856 | * same blocks within the object -- doesn't cause any trouble. | |
34dc7c2f | 3857 | */ |
40b84e7a | 3858 | ztest_od_init(od, ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0); |
9babb374 | 3859 | |
40b84e7a | 3860 | if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) |
34dc7c2f | 3861 | return; |
34dc7c2f | 3862 | |
428870ff | 3863 | while (ztest_random(10) != 0) |
40b84e7a BB |
3864 | ztest_io(zd, od->od_object, offset); |
3865 | ||
3866 | umem_free(od, sizeof(ztest_od_t)); | |
428870ff | 3867 | } |
34dc7c2f | 3868 | |
428870ff BB |
3869 | void |
3870 | ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id) | |
3871 | { | |
40b84e7a | 3872 | ztest_od_t *od; |
428870ff BB |
3873 | uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) + |
3874 | (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT); | |
3875 | uint64_t count = ztest_random(20) + 1; | |
3876 | uint64_t blocksize = ztest_random_blocksize(); | |
3877 | void *data; | |
34dc7c2f | 3878 | |
40b84e7a | 3879 | od = umem_alloc(sizeof(ztest_od_t), UMEM_NOFAIL); |
34dc7c2f | 3880 | |
40b84e7a BB |
3881 | ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0); |
3882 | ||
3883 | if (ztest_object_init(zd, od, sizeof (ztest_od_t), !ztest_random(2)) != 0) { | |
3884 | umem_free(od, sizeof(ztest_od_t)); | |
34dc7c2f | 3885 | return; |
40b84e7a | 3886 | } |
34dc7c2f | 3887 | |
40b84e7a BB |
3888 | if (ztest_truncate(zd, od->od_object, offset, count * blocksize) != 0) { |
3889 | umem_free(od, sizeof(ztest_od_t)); | |
34dc7c2f | 3890 | return; |
40b84e7a | 3891 | } |
34dc7c2f | 3892 | |
40b84e7a | 3893 | ztest_prealloc(zd, od->od_object, offset, count * blocksize); |
34dc7c2f | 3894 | |
428870ff | 3895 | data = umem_zalloc(blocksize, UMEM_NOFAIL); |
34dc7c2f | 3896 | |
428870ff BB |
3897 | while (ztest_random(count) != 0) { |
3898 | uint64_t randoff = offset + (ztest_random(count) * blocksize); | |
40b84e7a | 3899 | if (ztest_write(zd, od->od_object, randoff, blocksize, |
428870ff BB |
3900 | data) != 0) |
3901 | break; | |
3902 | while (ztest_random(4) != 0) | |
40b84e7a | 3903 | ztest_io(zd, od->od_object, randoff); |
9babb374 | 3904 | } |
34dc7c2f | 3905 | |
428870ff | 3906 | umem_free(data, blocksize); |
40b84e7a | 3907 | umem_free(od, sizeof(ztest_od_t)); |
34dc7c2f BB |
3908 | } |
3909 | ||
3910 | /* | |
3911 | * Verify that zap_{create,destroy,add,remove,update} work as expected. | |
3912 | */ | |
3913 | #define ZTEST_ZAP_MIN_INTS 1 | |
3914 | #define ZTEST_ZAP_MAX_INTS 4 | |
3915 | #define ZTEST_ZAP_MAX_PROPS 1000 | |
3916 | ||
3917 | void | |
428870ff | 3918 | ztest_zap(ztest_ds_t *zd, uint64_t id) |
34dc7c2f | 3919 | { |
428870ff | 3920 | objset_t *os = zd->zd_os; |
40b84e7a | 3921 | ztest_od_t *od; |
34dc7c2f BB |
3922 | uint64_t object; |
3923 | uint64_t txg, last_txg; | |
3924 | uint64_t value[ZTEST_ZAP_MAX_INTS]; | |
3925 | uint64_t zl_ints, zl_intsize, prop; | |
3926 | int i, ints; | |
3927 | dmu_tx_t *tx; | |
3928 | char propname[100], txgname[100]; | |
3929 | int error; | |
34dc7c2f BB |
3930 | char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" }; |
3931 | ||
40b84e7a BB |
3932 | od = umem_alloc(sizeof(ztest_od_t), UMEM_NOFAIL); |
3933 | ztest_od_init(od, id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0); | |
34dc7c2f | 3934 | |
40b84e7a BB |
3935 | if (ztest_object_init(zd, od, sizeof (ztest_od_t), |
3936 | !ztest_random(2)) != 0) | |
3937 | goto out; | |
34dc7c2f | 3938 | |
40b84e7a | 3939 | object = od->od_object; |
34dc7c2f | 3940 | |
428870ff BB |
3941 | /* |
3942 | * Generate a known hash collision, and verify that | |
3943 | * we can lookup and remove both entries. | |
3944 | */ | |
3945 | tx = dmu_tx_create(os); | |
3946 | dmu_tx_hold_zap(tx, object, B_TRUE, NULL); | |
3947 | txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); | |
3948 | if (txg == 0) | |
40b84e7a | 3949 | goto out; |
428870ff BB |
3950 | for (i = 0; i < 2; i++) { |
3951 | value[i] = i; | |
3952 | VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t), | |
3953 | 1, &value[i], tx)); | |
3954 | } | |
3955 | for (i = 0; i < 2; i++) { | |
3956 | VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i], | |
3957 | sizeof (uint64_t), 1, &value[i], tx)); | |
3958 | VERIFY3U(0, ==, | |
3959 | zap_length(os, object, hc[i], &zl_intsize, &zl_ints)); | |
3960 | ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); | |
3961 | ASSERT3U(zl_ints, ==, 1); | |
3962 | } | |
3963 | for (i = 0; i < 2; i++) { | |
3964 | VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx)); | |
34dc7c2f | 3965 | } |
428870ff | 3966 | dmu_tx_commit(tx); |
34dc7c2f | 3967 | |
428870ff BB |
3968 | /* |
3969 | * Generate a buch of random entries. | |
3970 | */ | |
34dc7c2f BB |
3971 | ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS); |
3972 | ||
3973 | prop = ztest_random(ZTEST_ZAP_MAX_PROPS); | |
3974 | (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop); | |
3975 | (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop); | |
3976 | bzero(value, sizeof (value)); | |
3977 | last_txg = 0; | |
3978 | ||
3979 | /* | |
3980 | * If these zap entries already exist, validate their contents. | |
3981 | */ | |
3982 | error = zap_length(os, object, txgname, &zl_intsize, &zl_ints); | |
3983 | if (error == 0) { | |
3984 | ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); | |
3985 | ASSERT3U(zl_ints, ==, 1); | |
3986 | ||
3987 | VERIFY(zap_lookup(os, object, txgname, zl_intsize, | |
3988 | zl_ints, &last_txg) == 0); | |
3989 | ||
3990 | VERIFY(zap_length(os, object, propname, &zl_intsize, | |
3991 | &zl_ints) == 0); | |
3992 | ||
3993 | ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); | |
3994 | ASSERT3U(zl_ints, ==, ints); | |
3995 | ||
3996 | VERIFY(zap_lookup(os, object, propname, zl_intsize, | |
3997 | zl_ints, value) == 0); | |
3998 | ||
3999 | for (i = 0; i < ints; i++) { | |
4000 | ASSERT3U(value[i], ==, last_txg + object + i); | |
4001 | } | |
4002 | } else { | |
4003 | ASSERT3U(error, ==, ENOENT); | |
4004 | } | |
4005 | ||
4006 | /* | |
4007 | * Atomically update two entries in our zap object. | |
4008 | * The first is named txg_%llu, and contains the txg | |
4009 | * in which the property was last updated. The second | |
4010 | * is named prop_%llu, and the nth element of its value | |
4011 | * should be txg + object + n. | |
4012 | */ | |
4013 | tx = dmu_tx_create(os); | |
428870ff BB |
4014 | dmu_tx_hold_zap(tx, object, B_TRUE, NULL); |
4015 | txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); | |
4016 | if (txg == 0) | |
40b84e7a | 4017 | goto out; |
34dc7c2f BB |
4018 | |
4019 | if (last_txg > txg) | |
4020 | fatal(0, "zap future leak: old %llu new %llu", last_txg, txg); | |
4021 | ||
4022 | for (i = 0; i < ints; i++) | |
4023 | value[i] = txg + object + i; | |
4024 | ||
428870ff BB |
4025 | VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t), |
4026 | 1, &txg, tx)); | |
4027 | VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t), | |
4028 | ints, value, tx)); | |
34dc7c2f BB |
4029 | |
4030 | dmu_tx_commit(tx); | |
4031 | ||
4032 | /* | |
4033 | * Remove a random pair of entries. | |
4034 | */ | |
4035 | prop = ztest_random(ZTEST_ZAP_MAX_PROPS); | |
4036 | (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop); | |
4037 | (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop); | |
4038 | ||
4039 | error = zap_length(os, object, txgname, &zl_intsize, &zl_ints); | |
4040 | ||
4041 | if (error == ENOENT) | |
40b84e7a | 4042 | goto out; |
34dc7c2f BB |
4043 | |
4044 | ASSERT3U(error, ==, 0); | |
4045 | ||
4046 | tx = dmu_tx_create(os); | |
428870ff BB |
4047 | dmu_tx_hold_zap(tx, object, B_TRUE, NULL); |
4048 | txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); | |
4049 | if (txg == 0) | |
40b84e7a | 4050 | goto out; |
428870ff BB |
4051 | VERIFY3U(0, ==, zap_remove(os, object, txgname, tx)); |
4052 | VERIFY3U(0, ==, zap_remove(os, object, propname, tx)); | |
4053 | dmu_tx_commit(tx); | |
40b84e7a BB |
4054 | out: |
4055 | umem_free(od, sizeof(ztest_od_t)); | |
428870ff | 4056 | } |
34dc7c2f | 4057 | |
428870ff BB |
4058 | /* |
4059 | * Testcase to test the upgrading of a microzap to fatzap. | |
4060 | */ | |
4061 | void | |
4062 | ztest_fzap(ztest_ds_t *zd, uint64_t id) | |
4063 | { | |
4064 | objset_t *os = zd->zd_os; | |
40b84e7a | 4065 | ztest_od_t *od; |
428870ff | 4066 | uint64_t object, txg; |
d6320ddb | 4067 | int i; |
34dc7c2f | 4068 | |
40b84e7a BB |
4069 | od = umem_alloc(sizeof(ztest_od_t), UMEM_NOFAIL); |
4070 | ztest_od_init(od, id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0); | |
428870ff | 4071 | |
40b84e7a BB |
4072 | if (ztest_object_init(zd, od, sizeof (ztest_od_t), |
4073 | !ztest_random(2)) != 0) | |
4074 | goto out; | |
4075 | object = od->od_object; | |
34dc7c2f BB |
4076 | |
4077 | /* | |
428870ff BB |
4078 | * Add entries to this ZAP and make sure it spills over |
4079 | * and gets upgraded to a fatzap. Also, since we are adding | |
4080 | * 2050 entries we should see ptrtbl growth and leaf-block split. | |
34dc7c2f | 4081 | */ |
d6320ddb | 4082 | for (i = 0; i < 2050; i++) { |
428870ff BB |
4083 | char name[MAXNAMELEN]; |
4084 | uint64_t value = i; | |
4085 | dmu_tx_t *tx; | |
4086 | int error; | |
34dc7c2f | 4087 | |
428870ff | 4088 | (void) snprintf(name, sizeof (name), "fzap-%llu-%llu", |
b8864a23 | 4089 | (u_longlong_t)id, (u_longlong_t)value); |
428870ff BB |
4090 | |
4091 | tx = dmu_tx_create(os); | |
4092 | dmu_tx_hold_zap(tx, object, B_TRUE, name); | |
4093 | txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); | |
4094 | if (txg == 0) | |
40b84e7a | 4095 | goto out; |
428870ff BB |
4096 | error = zap_add(os, object, name, sizeof (uint64_t), 1, |
4097 | &value, tx); | |
4098 | ASSERT(error == 0 || error == EEXIST); | |
4099 | dmu_tx_commit(tx); | |
34dc7c2f | 4100 | } |
40b84e7a BB |
4101 | out: |
4102 | umem_free(od, sizeof(ztest_od_t)); | |
34dc7c2f BB |
4103 | } |
4104 | ||
428870ff | 4105 | /* ARGSUSED */ |
34dc7c2f | 4106 | void |
428870ff | 4107 | ztest_zap_parallel(ztest_ds_t *zd, uint64_t id) |
34dc7c2f | 4108 | { |
428870ff | 4109 | objset_t *os = zd->zd_os; |
40b84e7a | 4110 | ztest_od_t *od; |
34dc7c2f BB |
4111 | uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc; |
4112 | dmu_tx_t *tx; | |
4113 | int i, namelen, error; | |
428870ff | 4114 | int micro = ztest_random(2); |
34dc7c2f BB |
4115 | char name[20], string_value[20]; |
4116 | void *data; | |
4117 | ||
40b84e7a BB |
4118 | od = umem_alloc(sizeof(ztest_od_t), UMEM_NOFAIL); |
4119 | ztest_od_init(od, ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0); | |
428870ff | 4120 | |
40b84e7a BB |
4121 | if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) { |
4122 | umem_free(od, sizeof(ztest_od_t)); | |
428870ff | 4123 | return; |
40b84e7a | 4124 | } |
428870ff | 4125 | |
40b84e7a | 4126 | object = od->od_object; |
428870ff | 4127 | |
34dc7c2f BB |
4128 | /* |
4129 | * Generate a random name of the form 'xxx.....' where each | |
4130 | * x is a random printable character and the dots are dots. | |
4131 | * There are 94 such characters, and the name length goes from | |
4132 | * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names. | |
4133 | */ | |
4134 | namelen = ztest_random(sizeof (name) - 5) + 5 + 1; | |
4135 | ||
4136 | for (i = 0; i < 3; i++) | |
4137 | name[i] = '!' + ztest_random('~' - '!' + 1); | |
4138 | for (; i < namelen - 1; i++) | |
4139 | name[i] = '.'; | |
4140 | name[i] = '\0'; | |
4141 | ||
428870ff | 4142 | if ((namelen & 1) || micro) { |
34dc7c2f BB |
4143 | wsize = sizeof (txg); |
4144 | wc = 1; | |
4145 | data = &txg; | |
4146 | } else { | |
4147 | wsize = 1; | |
4148 | wc = namelen; | |
4149 | data = string_value; | |
4150 | } | |
4151 | ||
4152 | count = -1ULL; | |
4153 | VERIFY(zap_count(os, object, &count) == 0); | |
4154 | ASSERT(count != -1ULL); | |
4155 | ||
4156 | /* | |
4157 | * Select an operation: length, lookup, add, update, remove. | |
4158 | */ | |
4159 | i = ztest_random(5); | |
4160 | ||
4161 | if (i >= 2) { | |
4162 | tx = dmu_tx_create(os); | |
428870ff BB |
4163 | dmu_tx_hold_zap(tx, object, B_TRUE, NULL); |
4164 | txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); | |
4165 | if (txg == 0) | |
34dc7c2f | 4166 | return; |
34dc7c2f BB |
4167 | bcopy(name, string_value, namelen); |
4168 | } else { | |
4169 | tx = NULL; | |
4170 | txg = 0; | |
4171 | bzero(string_value, namelen); | |
4172 | } | |
4173 | ||
4174 | switch (i) { | |
4175 | ||
4176 | case 0: | |
4177 | error = zap_length(os, object, name, &zl_wsize, &zl_wc); | |
4178 | if (error == 0) { | |
4179 | ASSERT3U(wsize, ==, zl_wsize); | |
4180 | ASSERT3U(wc, ==, zl_wc); | |
4181 | } else { | |
4182 | ASSERT3U(error, ==, ENOENT); | |
4183 | } | |
4184 | break; | |
4185 | ||
4186 | case 1: | |
4187 | error = zap_lookup(os, object, name, wsize, wc, data); | |
4188 | if (error == 0) { | |
4189 | if (data == string_value && | |
4190 | bcmp(name, data, namelen) != 0) | |
4191 | fatal(0, "name '%s' != val '%s' len %d", | |
4192 | name, data, namelen); | |
4193 | } else { | |
4194 | ASSERT3U(error, ==, ENOENT); | |
4195 | } | |
4196 | break; | |
4197 | ||
4198 | case 2: | |
4199 | error = zap_add(os, object, name, wsize, wc, data, tx); | |
4200 | ASSERT(error == 0 || error == EEXIST); | |
4201 | break; | |
4202 | ||
4203 | case 3: | |
4204 | VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0); | |
4205 | break; | |
4206 | ||
4207 | case 4: | |
4208 | error = zap_remove(os, object, name, tx); | |
4209 | ASSERT(error == 0 || error == ENOENT); | |
4210 | break; | |
4211 | } | |
4212 | ||
4213 | if (tx != NULL) | |
4214 | dmu_tx_commit(tx); | |
40b84e7a BB |
4215 | |
4216 | umem_free(od, sizeof(ztest_od_t)); | |
34dc7c2f BB |
4217 | } |
4218 | ||
428870ff BB |
4219 | /* |
4220 | * Commit callback data. | |
4221 | */ | |
4222 | typedef struct ztest_cb_data { | |
4223 | list_node_t zcd_node; | |
4224 | uint64_t zcd_txg; | |
4225 | int zcd_expected_err; | |
4226 | boolean_t zcd_added; | |
4227 | boolean_t zcd_called; | |
4228 | spa_t *zcd_spa; | |
4229 | } ztest_cb_data_t; | |
4230 | ||
4231 | /* This is the actual commit callback function */ | |
4232 | static void | |
4233 | ztest_commit_callback(void *arg, int error) | |
4234 | { | |
4235 | ztest_cb_data_t *data = arg; | |
4236 | uint64_t synced_txg; | |
4237 | ||
4238 | VERIFY(data != NULL); | |
4239 | VERIFY3S(data->zcd_expected_err, ==, error); | |
4240 | VERIFY(!data->zcd_called); | |
4241 | ||
4242 | synced_txg = spa_last_synced_txg(data->zcd_spa); | |
4243 | if (data->zcd_txg > synced_txg) | |
4244 | fatal(0, "commit callback of txg %" PRIu64 " called prematurely" | |
4245 | ", last synced txg = %" PRIu64 "\n", data->zcd_txg, | |
4246 | synced_txg); | |
4247 | ||
4248 | data->zcd_called = B_TRUE; | |
4249 | ||
4250 | if (error == ECANCELED) { | |
4251 | ASSERT3U(data->zcd_txg, ==, 0); | |
4252 | ASSERT(!data->zcd_added); | |
4253 | ||
4254 | /* | |
4255 | * The private callback data should be destroyed here, but | |
4256 | * since we are going to check the zcd_called field after | |
4257 | * dmu_tx_abort(), we will destroy it there. | |
4258 | */ | |
4259 | return; | |
4260 | } | |
4261 | ||
090ff092 | 4262 | ASSERT(data->zcd_added); |
428870ff BB |
4263 | ASSERT3U(data->zcd_txg, !=, 0); |
4264 | ||
1e33ac1e | 4265 | (void) mutex_enter(&zcl.zcl_callbacks_lock); |
090ff092 RC |
4266 | |
4267 | /* See if this cb was called more quickly */ | |
4268 | if ((synced_txg - data->zcd_txg) < zc_min_txg_delay) | |
4269 | zc_min_txg_delay = synced_txg - data->zcd_txg; | |
4270 | ||
4271 | /* Remove our callback from the list */ | |
428870ff | 4272 | list_remove(&zcl.zcl_callbacks, data); |
090ff092 | 4273 | |
1e33ac1e | 4274 | (void) mutex_exit(&zcl.zcl_callbacks_lock); |
428870ff | 4275 | |
428870ff BB |
4276 | umem_free(data, sizeof (ztest_cb_data_t)); |
4277 | } | |
4278 | ||
4279 | /* Allocate and initialize callback data structure */ | |
4280 | static ztest_cb_data_t * | |
4281 | ztest_create_cb_data(objset_t *os, uint64_t txg) | |
4282 | { | |
4283 | ztest_cb_data_t *cb_data; | |
4284 | ||
4285 | cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL); | |
4286 | ||
4287 | cb_data->zcd_txg = txg; | |
4288 | cb_data->zcd_spa = dmu_objset_spa(os); | |
1e33ac1e | 4289 | list_link_init(&cb_data->zcd_node); |
428870ff BB |
4290 | |
4291 | return (cb_data); | |
4292 | } | |
4293 | ||
428870ff BB |
4294 | /* |
4295 | * Commit callback test. | |
4296 | */ | |
34dc7c2f | 4297 | void |
428870ff BB |
4298 | ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id) |
4299 | { | |
4300 | objset_t *os = zd->zd_os; | |
40b84e7a | 4301 | ztest_od_t *od; |
428870ff BB |
4302 | dmu_tx_t *tx; |
4303 | ztest_cb_data_t *cb_data[3], *tmp_cb; | |
4304 | uint64_t old_txg, txg; | |
090ff092 | 4305 | int i, error = 0; |
428870ff | 4306 | |
40b84e7a BB |
4307 | od = umem_alloc(sizeof(ztest_od_t), UMEM_NOFAIL); |
4308 | ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0); | |
428870ff | 4309 | |
40b84e7a BB |
4310 | if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) { |
4311 | umem_free(od, sizeof(ztest_od_t)); | |
428870ff | 4312 | return; |
40b84e7a | 4313 | } |
428870ff BB |
4314 | |
4315 | tx = dmu_tx_create(os); | |
4316 | ||
4317 | cb_data[0] = ztest_create_cb_data(os, 0); | |
4318 | dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]); | |
4319 | ||
40b84e7a | 4320 | dmu_tx_hold_write(tx, od->od_object, 0, sizeof (uint64_t)); |
428870ff BB |
4321 | |
4322 | /* Every once in a while, abort the transaction on purpose */ | |
4323 | if (ztest_random(100) == 0) | |
4324 | error = -1; | |
4325 | ||
4326 | if (!error) | |
4327 | error = dmu_tx_assign(tx, TXG_NOWAIT); | |
4328 | ||
4329 | txg = error ? 0 : dmu_tx_get_txg(tx); | |
4330 | ||
4331 | cb_data[0]->zcd_txg = txg; | |
4332 | cb_data[1] = ztest_create_cb_data(os, txg); | |
4333 | dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]); | |
4334 | ||
4335 | if (error) { | |
4336 | /* | |
4337 | * It's not a strict requirement to call the registered | |
4338 | * callbacks from inside dmu_tx_abort(), but that's what | |
4339 | * it's supposed to happen in the current implementation | |
4340 | * so we will check for that. | |
4341 | */ | |
4342 | for (i = 0; i < 2; i++) { | |
4343 | cb_data[i]->zcd_expected_err = ECANCELED; | |
4344 | VERIFY(!cb_data[i]->zcd_called); | |
4345 | } | |
4346 | ||
4347 | dmu_tx_abort(tx); | |
4348 | ||
4349 | for (i = 0; i < 2; i++) { | |
4350 | VERIFY(cb_data[i]->zcd_called); | |
4351 | umem_free(cb_data[i], sizeof (ztest_cb_data_t)); | |
4352 | } | |
4353 | ||
40b84e7a | 4354 | umem_free(od, sizeof(ztest_od_t)); |
428870ff BB |
4355 | return; |
4356 | } | |
4357 | ||
4358 | cb_data[2] = ztest_create_cb_data(os, txg); | |
4359 | dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]); | |
4360 | ||
4361 | /* | |
4362 | * Read existing data to make sure there isn't a future leak. | |
4363 | */ | |
40b84e7a | 4364 | VERIFY(0 == dmu_read(os, od->od_object, 0, sizeof (uint64_t), |
428870ff BB |
4365 | &old_txg, DMU_READ_PREFETCH)); |
4366 | ||
4367 | if (old_txg > txg) | |
4368 | fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64, | |
4369 | old_txg, txg); | |
4370 | ||
40b84e7a | 4371 | dmu_write(os, od->od_object, 0, sizeof (uint64_t), &txg, tx); |
428870ff | 4372 | |
1e33ac1e | 4373 | (void) mutex_enter(&zcl.zcl_callbacks_lock); |
428870ff BB |
4374 | |
4375 | /* | |
4376 | * Since commit callbacks don't have any ordering requirement and since | |
4377 | * it is theoretically possible for a commit callback to be called | |
4378 | * after an arbitrary amount of time has elapsed since its txg has been | |
4379 | * synced, it is difficult to reliably determine whether a commit | |
4380 | * callback hasn't been called due to high load or due to a flawed | |
4381 | * implementation. | |
4382 | * | |
4383 | * In practice, we will assume that if after a certain number of txgs a | |
4384 | * commit callback hasn't been called, then most likely there's an | |
4385 | * implementation bug.. | |
4386 | */ | |
4387 | tmp_cb = list_head(&zcl.zcl_callbacks); | |
4388 | if (tmp_cb != NULL && | |
090ff092 | 4389 | tmp_cb->zcd_txg + ZTEST_COMMIT_CB_THRESH < txg) { |
428870ff BB |
4390 | fatal(0, "Commit callback threshold exceeded, oldest txg: %" |
4391 | PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg); | |
4392 | } | |
4393 | ||
4394 | /* | |
4395 | * Let's find the place to insert our callbacks. | |
4396 | * | |
4397 | * Even though the list is ordered by txg, it is possible for the | |
4398 | * insertion point to not be the end because our txg may already be | |
4399 | * quiescing at this point and other callbacks in the open txg | |
4400 | * (from other objsets) may have sneaked in. | |
4401 | */ | |
4402 | tmp_cb = list_tail(&zcl.zcl_callbacks); | |
4403 | while (tmp_cb != NULL && tmp_cb->zcd_txg > txg) | |
4404 | tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb); | |
4405 | ||
4406 | /* Add the 3 callbacks to the list */ | |
4407 | for (i = 0; i < 3; i++) { | |
4408 | if (tmp_cb == NULL) | |
4409 | list_insert_head(&zcl.zcl_callbacks, cb_data[i]); | |
4410 | else | |
4411 | list_insert_after(&zcl.zcl_callbacks, tmp_cb, | |
4412 | cb_data[i]); | |
4413 | ||
4414 | cb_data[i]->zcd_added = B_TRUE; | |
4415 | VERIFY(!cb_data[i]->zcd_called); | |
4416 | ||
4417 | tmp_cb = cb_data[i]; | |
4418 | } | |
4419 | ||
090ff092 RC |
4420 | zc_cb_counter += 3; |
4421 | ||
1e33ac1e | 4422 | (void) mutex_exit(&zcl.zcl_callbacks_lock); |
428870ff BB |
4423 | |
4424 | dmu_tx_commit(tx); | |
40b84e7a BB |
4425 | |
4426 | umem_free(od, sizeof(ztest_od_t)); | |
428870ff BB |
4427 | } |
4428 | ||
4429 | /* ARGSUSED */ | |
4430 | void | |
4431 | ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id) | |
4432 | { | |
4433 | zfs_prop_t proplist[] = { | |
4434 | ZFS_PROP_CHECKSUM, | |
4435 | ZFS_PROP_COMPRESSION, | |
4436 | ZFS_PROP_COPIES, | |
4437 | ZFS_PROP_DEDUP | |
4438 | }; | |
4439 | ztest_shared_t *zs = ztest_shared; | |
d6320ddb | 4440 | int p; |
428870ff | 4441 | |
1e33ac1e | 4442 | (void) rw_enter(&zs->zs_name_lock, RW_READER); |
428870ff | 4443 | |
d6320ddb | 4444 | for (p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++) |
428870ff BB |
4445 | (void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p], |
4446 | ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2)); | |
4447 | ||
1e33ac1e | 4448 | (void) rw_exit(&zs->zs_name_lock); |
428870ff BB |
4449 | } |
4450 | ||
4451 | /* ARGSUSED */ | |
4452 | void | |
4453 | ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id) | |
4454 | { | |
4455 | ztest_shared_t *zs = ztest_shared; | |
4456 | nvlist_t *props = NULL; | |
4457 | ||
1e33ac1e | 4458 | (void) rw_enter(&zs->zs_name_lock, RW_READER); |
428870ff BB |
4459 | |
4460 | (void) ztest_spa_prop_set_uint64(zs, ZPOOL_PROP_DEDUPDITTO, | |
4461 | ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN)); | |
4462 | ||
4463 | VERIFY3U(spa_prop_get(zs->zs_spa, &props), ==, 0); | |
4464 | ||
4465 | if (zopt_verbose >= 6) | |
4466 | dump_nvlist(props, 4); | |
4467 | ||
4468 | nvlist_free(props); | |
4469 | ||
1e33ac1e | 4470 | (void) rw_exit(&zs->zs_name_lock); |
428870ff BB |
4471 | } |
4472 | ||
4473 | /* | |
4474 | * Test snapshot hold/release and deferred destroy. | |
4475 | */ | |
4476 | void | |
4477 | ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id) | |
34dc7c2f | 4478 | { |
34dc7c2f | 4479 | int error; |
428870ff BB |
4480 | objset_t *os = zd->zd_os; |
4481 | objset_t *origin; | |
4482 | char snapname[100]; | |
4483 | char fullname[100]; | |
4484 | char clonename[100]; | |
4485 | char tag[100]; | |
4486 | char osname[MAXNAMELEN]; | |
34dc7c2f | 4487 | |
1e33ac1e | 4488 | (void) rw_enter(&ztest_shared->zs_name_lock, RW_READER); |
34dc7c2f BB |
4489 | |
4490 | dmu_objset_name(os, osname); | |
4491 | ||
1e33ac1e | 4492 | (void) snprintf(snapname, 100, "sh1_%llu", (u_longlong_t)id); |
428870ff | 4493 | (void) snprintf(fullname, 100, "%s@%s", osname, snapname); |
1e33ac1e BB |
4494 | (void) snprintf(clonename, 100, "%s/ch1_%llu",osname,(u_longlong_t)id); |
4495 | (void) snprintf(tag, 100, "tag_%llu", (u_longlong_t)id); | |
428870ff BB |
4496 | |
4497 | /* | |
4498 | * Clean up from any previous run. | |
4499 | */ | |
4500 | (void) dmu_objset_destroy(clonename, B_FALSE); | |
4501 | (void) dsl_dataset_user_release(osname, snapname, tag, B_FALSE); | |
4502 | (void) dmu_objset_destroy(fullname, B_FALSE); | |
4503 | ||
4504 | /* | |
4505 | * Create snapshot, clone it, mark snap for deferred destroy, | |
4506 | * destroy clone, verify snap was also destroyed. | |
4507 | */ | |
572e2857 BB |
4508 | error = dmu_objset_snapshot(osname, snapname, NULL, NULL, FALSE, |
4509 | FALSE, -1); | |
428870ff BB |
4510 | if (error) { |
4511 | if (error == ENOSPC) { | |
4512 | ztest_record_enospc("dmu_objset_snapshot"); | |
4513 | goto out; | |
34dc7c2f | 4514 | } |
428870ff BB |
4515 | fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error); |
4516 | } | |
34dc7c2f | 4517 | |
428870ff BB |
4518 | error = dmu_objset_hold(fullname, FTAG, &origin); |
4519 | if (error) | |
4520 | fatal(0, "dmu_objset_hold(%s) = %d", fullname, error); | |
34dc7c2f | 4521 | |
428870ff BB |
4522 | error = dmu_objset_clone(clonename, dmu_objset_ds(origin), 0); |
4523 | dmu_objset_rele(origin, FTAG); | |
4524 | if (error) { | |
34dc7c2f | 4525 | if (error == ENOSPC) { |
428870ff BB |
4526 | ztest_record_enospc("dmu_objset_clone"); |
4527 | goto out; | |
34dc7c2f | 4528 | } |
428870ff BB |
4529 | fatal(0, "dmu_objset_clone(%s) = %d", clonename, error); |
4530 | } | |
34dc7c2f | 4531 | |
428870ff BB |
4532 | error = dmu_objset_destroy(fullname, B_TRUE); |
4533 | if (error) { | |
4534 | fatal(0, "dmu_objset_destroy(%s, B_TRUE) = %d", | |
4535 | fullname, error); | |
4536 | } | |
34dc7c2f | 4537 | |
428870ff BB |
4538 | error = dmu_objset_destroy(clonename, B_FALSE); |
4539 | if (error) | |
4540 | fatal(0, "dmu_objset_destroy(%s) = %d", clonename, error); | |
34dc7c2f | 4541 | |
428870ff BB |
4542 | error = dmu_objset_hold(fullname, FTAG, &origin); |
4543 | if (error != ENOENT) | |
4544 | fatal(0, "dmu_objset_hold(%s) = %d", fullname, error); | |
34dc7c2f | 4545 | |
428870ff BB |
4546 | /* |
4547 | * Create snapshot, add temporary hold, verify that we can't | |
4548 | * destroy a held snapshot, mark for deferred destroy, | |
4549 | * release hold, verify snapshot was destroyed. | |
4550 | */ | |
572e2857 BB |
4551 | error = dmu_objset_snapshot(osname, snapname, NULL, NULL, FALSE, |
4552 | FALSE, -1); | |
428870ff BB |
4553 | if (error) { |
4554 | if (error == ENOSPC) { | |
4555 | ztest_record_enospc("dmu_objset_snapshot"); | |
4556 | goto out; | |
34dc7c2f | 4557 | } |
428870ff BB |
4558 | fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error); |
4559 | } | |
4560 | ||
572e2857 BB |
4561 | error = dsl_dataset_user_hold(osname, snapname, tag, B_FALSE, |
4562 | B_TRUE, -1); | |
428870ff BB |
4563 | if (error) |
4564 | fatal(0, "dsl_dataset_user_hold(%s)", fullname, tag); | |
4565 | ||
4566 | error = dmu_objset_destroy(fullname, B_FALSE); | |
4567 | if (error != EBUSY) { | |
4568 | fatal(0, "dmu_objset_destroy(%s, B_FALSE) = %d", | |
4569 | fullname, error); | |
4570 | } | |
4571 | ||
4572 | error = dmu_objset_destroy(fullname, B_TRUE); | |
4573 | if (error) { | |
4574 | fatal(0, "dmu_objset_destroy(%s, B_TRUE) = %d", | |
4575 | fullname, error); | |
34dc7c2f BB |
4576 | } |
4577 | ||
428870ff BB |
4578 | error = dsl_dataset_user_release(osname, snapname, tag, B_FALSE); |
4579 | if (error) | |
4580 | fatal(0, "dsl_dataset_user_release(%s)", fullname, tag); | |
4581 | ||
4582 | VERIFY(dmu_objset_hold(fullname, FTAG, &origin) == ENOENT); | |
4583 | ||
4584 | out: | |
1e33ac1e | 4585 | (void) rw_exit(&ztest_shared->zs_name_lock); |
34dc7c2f BB |
4586 | } |
4587 | ||
34dc7c2f BB |
4588 | /* |
4589 | * Inject random faults into the on-disk data. | |
4590 | */ | |
428870ff | 4591 | /* ARGSUSED */ |
34dc7c2f | 4592 | void |
428870ff | 4593 | ztest_fault_inject(ztest_ds_t *zd, uint64_t id) |
34dc7c2f | 4594 | { |
428870ff BB |
4595 | ztest_shared_t *zs = ztest_shared; |
4596 | spa_t *spa = zs->zs_spa; | |
34dc7c2f BB |
4597 | int fd; |
4598 | uint64_t offset; | |
428870ff | 4599 | uint64_t leaves; |
c5b3a7bb | 4600 | uint64_t bad = 0x1990c0ffeedecadeull; |
34dc7c2f | 4601 | uint64_t top, leaf; |
40b84e7a BB |
4602 | char *path0; |
4603 | char *pathrand; | |
34dc7c2f | 4604 | size_t fsize; |
34dc7c2f BB |
4605 | int bshift = SPA_MAXBLOCKSHIFT + 2; /* don't scrog all labels */ |
4606 | int iters = 1000; | |
428870ff BB |
4607 | int maxfaults; |
4608 | int mirror_save; | |
b128c09f | 4609 | vdev_t *vd0 = NULL; |
34dc7c2f | 4610 | uint64_t guid0 = 0; |
428870ff BB |
4611 | boolean_t islog = B_FALSE; |
4612 | ||
40b84e7a BB |
4613 | path0 = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); |
4614 | pathrand = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); | |
4615 | ||
1e33ac1e | 4616 | mutex_enter(&zs->zs_vdev_lock); |
428870ff BB |
4617 | maxfaults = MAXFAULTS(); |
4618 | leaves = MAX(zs->zs_mirrors, 1) * zopt_raidz; | |
4619 | mirror_save = zs->zs_mirrors; | |
1e33ac1e | 4620 | mutex_exit(&zs->zs_vdev_lock); |
34dc7c2f | 4621 | |
b128c09f | 4622 | ASSERT(leaves >= 1); |
34dc7c2f BB |
4623 | |
4624 | /* | |
b128c09f | 4625 | * We need SCL_STATE here because we're going to look at vd0->vdev_tsd. |
34dc7c2f | 4626 | */ |
b128c09f | 4627 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); |
34dc7c2f | 4628 | |
b128c09f BB |
4629 | if (ztest_random(2) == 0) { |
4630 | /* | |
428870ff | 4631 | * Inject errors on a normal data device or slog device. |
b128c09f | 4632 | */ |
428870ff BB |
4633 | top = ztest_random_vdev_top(spa, B_TRUE); |
4634 | leaf = ztest_random(leaves) + zs->zs_splits; | |
34dc7c2f | 4635 | |
b128c09f BB |
4636 | /* |
4637 | * Generate paths to the first leaf in this top-level vdev, | |
4638 | * and to the random leaf we selected. We'll induce transient | |
4639 | * write failures and random online/offline activity on leaf 0, | |
4640 | * and we'll write random garbage to the randomly chosen leaf. | |
4641 | */ | |
6aec1cd5 | 4642 | (void) snprintf(path0, MAXPATHLEN, ztest_dev_template, |
428870ff | 4643 | zopt_dir, zopt_pool, top * leaves + zs->zs_splits); |
6aec1cd5 | 4644 | (void) snprintf(pathrand, MAXPATHLEN, ztest_dev_template, |
b128c09f | 4645 | zopt_dir, zopt_pool, top * leaves + leaf); |
34dc7c2f | 4646 | |
b128c09f | 4647 | vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0); |
428870ff BB |
4648 | if (vd0 != NULL && vd0->vdev_top->vdev_islog) |
4649 | islog = B_TRUE; | |
4650 | ||
b128c09f BB |
4651 | if (vd0 != NULL && maxfaults != 1) { |
4652 | /* | |
4653 | * Make vd0 explicitly claim to be unreadable, | |
4654 | * or unwriteable, or reach behind its back | |
4655 | * and close the underlying fd. We can do this if | |
4656 | * maxfaults == 0 because we'll fail and reexecute, | |
4657 | * and we can do it if maxfaults >= 2 because we'll | |
4658 | * have enough redundancy. If maxfaults == 1, the | |
4659 | * combination of this with injection of random data | |
4660 | * corruption below exceeds the pool's fault tolerance. | |
4661 | */ | |
4662 | vdev_file_t *vf = vd0->vdev_tsd; | |
4663 | ||
4664 | if (vf != NULL && ztest_random(3) == 0) { | |
4665 | (void) close(vf->vf_vnode->v_fd); | |
4666 | vf->vf_vnode->v_fd = -1; | |
4667 | } else if (ztest_random(2) == 0) { | |
4668 | vd0->vdev_cant_read = B_TRUE; | |
4669 | } else { | |
4670 | vd0->vdev_cant_write = B_TRUE; | |
4671 | } | |
4672 | guid0 = vd0->vdev_guid; | |
4673 | } | |
4674 | } else { | |
4675 | /* | |
4676 | * Inject errors on an l2cache device. | |
4677 | */ | |
4678 | spa_aux_vdev_t *sav = &spa->spa_l2cache; | |
34dc7c2f | 4679 | |
b128c09f BB |
4680 | if (sav->sav_count == 0) { |
4681 | spa_config_exit(spa, SCL_STATE, FTAG); | |
40b84e7a | 4682 | goto out; |
b128c09f BB |
4683 | } |
4684 | vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)]; | |
34dc7c2f | 4685 | guid0 = vd0->vdev_guid; |
b128c09f BB |
4686 | (void) strcpy(path0, vd0->vdev_path); |
4687 | (void) strcpy(pathrand, vd0->vdev_path); | |
4688 | ||
4689 | leaf = 0; | |
4690 | leaves = 1; | |
4691 | maxfaults = INT_MAX; /* no limit on cache devices */ | |
34dc7c2f BB |
4692 | } |
4693 | ||
b128c09f BB |
4694 | spa_config_exit(spa, SCL_STATE, FTAG); |
4695 | ||
34dc7c2f | 4696 | /* |
428870ff BB |
4697 | * If we can tolerate two or more faults, or we're dealing |
4698 | * with a slog, randomly online/offline vd0. | |
34dc7c2f | 4699 | */ |
428870ff | 4700 | if ((maxfaults >= 2 || islog) && guid0 != 0) { |
fb5f0bc8 BB |
4701 | if (ztest_random(10) < 6) { |
4702 | int flags = (ztest_random(2) == 0 ? | |
4703 | ZFS_OFFLINE_TEMPORARY : 0); | |
428870ff BB |
4704 | |
4705 | /* | |
4706 | * We have to grab the zs_name_lock as writer to | |
4707 | * prevent a race between offlining a slog and | |
4708 | * destroying a dataset. Offlining the slog will | |
4709 | * grab a reference on the dataset which may cause | |
4710 | * dmu_objset_destroy() to fail with EBUSY thus | |
4711 | * leaving the dataset in an inconsistent state. | |
4712 | */ | |
4713 | if (islog) | |
1e33ac1e BB |
4714 | (void) rw_enter(&ztest_shared->zs_name_lock, |
4715 | RW_WRITER); | |
428870ff | 4716 | |
fb5f0bc8 | 4717 | VERIFY(vdev_offline(spa, guid0, flags) != EBUSY); |
428870ff BB |
4718 | |
4719 | if (islog) | |
1e33ac1e | 4720 | (void) rw_exit(&ztest_shared->zs_name_lock); |
fb5f0bc8 BB |
4721 | } else { |
4722 | (void) vdev_online(spa, guid0, 0, NULL); | |
4723 | } | |
34dc7c2f BB |
4724 | } |
4725 | ||
428870ff | 4726 | if (maxfaults == 0) |
40b84e7a | 4727 | goto out; |
428870ff | 4728 | |
34dc7c2f BB |
4729 | /* |
4730 | * We have at least single-fault tolerance, so inject data corruption. | |
4731 | */ | |
4732 | fd = open(pathrand, O_RDWR); | |
4733 | ||
4734 | if (fd == -1) /* we hit a gap in the device namespace */ | |
40b84e7a | 4735 | goto out; |
34dc7c2f BB |
4736 | |
4737 | fsize = lseek(fd, 0, SEEK_END); | |
4738 | ||
4739 | while (--iters != 0) { | |
4740 | offset = ztest_random(fsize / (leaves << bshift)) * | |
4741 | (leaves << bshift) + (leaf << bshift) + | |
4742 | (ztest_random(1ULL << (bshift - 1)) & -8ULL); | |
4743 | ||
4744 | if (offset >= fsize) | |
4745 | continue; | |
4746 | ||
1e33ac1e | 4747 | mutex_enter(&zs->zs_vdev_lock); |
428870ff | 4748 | if (mirror_save != zs->zs_mirrors) { |
1e33ac1e | 4749 | mutex_exit(&zs->zs_vdev_lock); |
428870ff | 4750 | (void) close(fd); |
40b84e7a | 4751 | goto out; |
428870ff | 4752 | } |
34dc7c2f BB |
4753 | |
4754 | if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad)) | |
4755 | fatal(1, "can't inject bad word at 0x%llx in %s", | |
4756 | offset, pathrand); | |
428870ff | 4757 | |
1e33ac1e | 4758 | mutex_exit(&zs->zs_vdev_lock); |
428870ff BB |
4759 | |
4760 | if (zopt_verbose >= 7) | |
4761 | (void) printf("injected bad word into %s," | |
4762 | " offset 0x%llx\n", pathrand, (u_longlong_t)offset); | |
34dc7c2f BB |
4763 | } |
4764 | ||
4765 | (void) close(fd); | |
40b84e7a BB |
4766 | out: |
4767 | umem_free(path0, MAXPATHLEN); | |
4768 | umem_free(pathrand, MAXPATHLEN); | |
34dc7c2f BB |
4769 | } |
4770 | ||
4771 | /* | |
428870ff | 4772 | * Verify that DDT repair works as expected. |
34dc7c2f BB |
4773 | */ |
4774 | void | |
428870ff | 4775 | ztest_ddt_repair(ztest_ds_t *zd, uint64_t id) |
34dc7c2f | 4776 | { |
428870ff BB |
4777 | ztest_shared_t *zs = ztest_shared; |
4778 | spa_t *spa = zs->zs_spa; | |
4779 | objset_t *os = zd->zd_os; | |
40b84e7a | 4780 | ztest_od_t *od; |
428870ff BB |
4781 | uint64_t object, blocksize, txg, pattern, psize; |
4782 | enum zio_checksum checksum = spa_dedup_checksum(spa); | |
4783 | dmu_buf_t *db; | |
4784 | dmu_tx_t *tx; | |
4785 | void *buf; | |
4786 | blkptr_t blk; | |
4787 | int copies = 2 * ZIO_DEDUPDITTO_MIN; | |
d6320ddb | 4788 | int i; |
34dc7c2f | 4789 | |
428870ff BB |
4790 | blocksize = ztest_random_blocksize(); |
4791 | blocksize = MIN(blocksize, 2048); /* because we write so many */ | |
34dc7c2f | 4792 | |
40b84e7a BB |
4793 | od = umem_alloc(sizeof(ztest_od_t), UMEM_NOFAIL); |
4794 | ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0); | |
34dc7c2f | 4795 | |
40b84e7a BB |
4796 | if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) { |
4797 | umem_free(od, sizeof(ztest_od_t)); | |
428870ff | 4798 | return; |
40b84e7a | 4799 | } |
34dc7c2f BB |
4800 | |
4801 | /* | |
428870ff BB |
4802 | * Take the name lock as writer to prevent anyone else from changing |
4803 | * the pool and dataset properies we need to maintain during this test. | |
34dc7c2f | 4804 | */ |
1e33ac1e | 4805 | (void) rw_enter(&zs->zs_name_lock, RW_WRITER); |
34dc7c2f | 4806 | |
428870ff BB |
4807 | if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum, |
4808 | B_FALSE) != 0 || | |
4809 | ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1, | |
4810 | B_FALSE) != 0) { | |
1e33ac1e | 4811 | (void) rw_exit(&zs->zs_name_lock); |
40b84e7a | 4812 | umem_free(od, sizeof(ztest_od_t)); |
428870ff BB |
4813 | return; |
4814 | } | |
4815 | ||
4816 | object = od[0].od_object; | |
4817 | blocksize = od[0].od_blocksize; | |
3541dc6d | 4818 | pattern = zs->zs_guid ^ dmu_objset_fsid_guid(os); |
428870ff BB |
4819 | |
4820 | ASSERT(object != 0); | |
4821 | ||
4822 | tx = dmu_tx_create(os); | |
4823 | dmu_tx_hold_write(tx, object, 0, copies * blocksize); | |
4824 | txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); | |
4825 | if (txg == 0) { | |
1e33ac1e | 4826 | (void) rw_exit(&zs->zs_name_lock); |
40b84e7a | 4827 | umem_free(od, sizeof(ztest_od_t)); |
428870ff BB |
4828 | return; |
4829 | } | |
34dc7c2f BB |
4830 | |
4831 | /* | |
428870ff | 4832 | * Write all the copies of our block. |
34dc7c2f | 4833 | */ |
d6320ddb | 4834 | for (i = 0; i < copies; i++) { |
428870ff BB |
4835 | uint64_t offset = i * blocksize; |
4836 | VERIFY(dmu_buf_hold(os, object, offset, FTAG, &db, | |
4837 | DMU_READ_NO_PREFETCH) == 0); | |
4838 | ASSERT(db->db_offset == offset); | |
4839 | ASSERT(db->db_size == blocksize); | |
4840 | ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) || | |
4841 | ztest_pattern_match(db->db_data, db->db_size, 0ULL)); | |
4842 | dmu_buf_will_fill(db, tx); | |
4843 | ztest_pattern_set(db->db_data, db->db_size, pattern); | |
4844 | dmu_buf_rele(db, FTAG); | |
4845 | } | |
34dc7c2f | 4846 | |
428870ff BB |
4847 | dmu_tx_commit(tx); |
4848 | txg_wait_synced(spa_get_dsl(spa), txg); | |
34dc7c2f BB |
4849 | |
4850 | /* | |
428870ff | 4851 | * Find out what block we got. |
34dc7c2f | 4852 | */ |
428870ff BB |
4853 | VERIFY(dmu_buf_hold(os, object, 0, FTAG, &db, |
4854 | DMU_READ_NO_PREFETCH) == 0); | |
4855 | blk = *((dmu_buf_impl_t *)db)->db_blkptr; | |
4856 | dmu_buf_rele(db, FTAG); | |
34dc7c2f BB |
4857 | |
4858 | /* | |
428870ff | 4859 | * Damage the block. Dedup-ditto will save us when we read it later. |
34dc7c2f | 4860 | */ |
428870ff BB |
4861 | psize = BP_GET_PSIZE(&blk); |
4862 | buf = zio_buf_alloc(psize); | |
4863 | ztest_pattern_set(buf, psize, ~pattern); | |
34dc7c2f | 4864 | |
428870ff BB |
4865 | (void) zio_wait(zio_rewrite(NULL, spa, 0, &blk, |
4866 | buf, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE, | |
4867 | ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL)); | |
34dc7c2f | 4868 | |
428870ff | 4869 | zio_buf_free(buf, psize); |
34dc7c2f | 4870 | |
1e33ac1e | 4871 | (void) rw_exit(&zs->zs_name_lock); |
40b84e7a | 4872 | umem_free(od, sizeof(ztest_od_t)); |
34dc7c2f BB |
4873 | } |
4874 | ||
34dc7c2f | 4875 | /* |
428870ff | 4876 | * Scrub the pool. |
34dc7c2f | 4877 | */ |
428870ff BB |
4878 | /* ARGSUSED */ |
4879 | void | |
4880 | ztest_scrub(ztest_ds_t *zd, uint64_t id) | |
34dc7c2f | 4881 | { |
428870ff BB |
4882 | ztest_shared_t *zs = ztest_shared; |
4883 | spa_t *spa = zs->zs_spa; | |
34dc7c2f | 4884 | |
428870ff BB |
4885 | (void) spa_scan(spa, POOL_SCAN_SCRUB); |
4886 | (void) poll(NULL, 0, 100); /* wait a moment, then force a restart */ | |
4887 | (void) spa_scan(spa, POOL_SCAN_SCRUB); | |
4888 | } | |
34dc7c2f | 4889 | |
3541dc6d GA |
4890 | /* |
4891 | * Change the guid for the pool. | |
4892 | */ | |
4893 | /* ARGSUSED */ | |
4894 | void | |
4895 | ztest_reguid(ztest_ds_t *zd, uint64_t id) | |
4896 | { | |
4897 | ztest_shared_t *zs = ztest_shared; | |
4898 | spa_t *spa = zs->zs_spa; | |
4899 | uint64_t orig, load; | |
4900 | ||
4901 | orig = spa_guid(spa); | |
4902 | load = spa_load_guid(spa); | |
4903 | if (spa_change_guid(spa) != 0) | |
4904 | return; | |
4905 | ||
4906 | if (zopt_verbose >= 3) { | |
4907 | (void) printf("Changed guid old %llu -> %llu\n", | |
4908 | (u_longlong_t)orig, (u_longlong_t)spa_guid(spa)); | |
4909 | } | |
4910 | ||
4911 | VERIFY3U(orig, !=, spa_guid(spa)); | |
4912 | VERIFY3U(load, ==, spa_load_guid(spa)); | |
4913 | } | |
4914 | ||
428870ff BB |
4915 | /* |
4916 | * Rename the pool to a different name and then rename it back. | |
4917 | */ | |
4918 | /* ARGSUSED */ | |
4919 | void | |
4920 | ztest_spa_rename(ztest_ds_t *zd, uint64_t id) | |
4921 | { | |
4922 | ztest_shared_t *zs = ztest_shared; | |
4923 | char *oldname, *newname; | |
4924 | spa_t *spa; | |
34dc7c2f | 4925 | |
1e33ac1e | 4926 | (void) rw_enter(&zs->zs_name_lock, RW_WRITER); |
34dc7c2f | 4927 | |
428870ff BB |
4928 | oldname = zs->zs_pool; |
4929 | newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL); | |
4930 | (void) strcpy(newname, oldname); | |
4931 | (void) strcat(newname, "_tmp"); | |
34dc7c2f BB |
4932 | |
4933 | /* | |
428870ff | 4934 | * Do the rename |
34dc7c2f | 4935 | */ |
428870ff | 4936 | VERIFY3U(0, ==, spa_rename(oldname, newname)); |
34dc7c2f BB |
4937 | |
4938 | /* | |
428870ff | 4939 | * Try to open it under the old name, which shouldn't exist |
34dc7c2f | 4940 | */ |
428870ff | 4941 | VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG)); |
34dc7c2f BB |
4942 | |
4943 | /* | |
428870ff BB |
4944 | * Open it under the new name and make sure it's still the same spa_t. |
4945 | */ | |
4946 | VERIFY3U(0, ==, spa_open(newname, &spa, FTAG)); | |
34dc7c2f | 4947 | |
428870ff BB |
4948 | ASSERT(spa == zs->zs_spa); |
4949 | spa_close(spa, FTAG); | |
34dc7c2f BB |
4950 | |
4951 | /* | |
428870ff | 4952 | * Rename it back to the original |
34dc7c2f | 4953 | */ |
428870ff | 4954 | VERIFY3U(0, ==, spa_rename(newname, oldname)); |
34dc7c2f | 4955 | |
428870ff BB |
4956 | /* |
4957 | * Make sure it can still be opened | |
4958 | */ | |
4959 | VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG)); | |
34dc7c2f | 4960 | |
428870ff BB |
4961 | ASSERT(spa == zs->zs_spa); |
4962 | spa_close(spa, FTAG); | |
4963 | ||
4964 | umem_free(newname, strlen(newname) + 1); | |
4965 | ||
1e33ac1e | 4966 | (void) rw_exit(&zs->zs_name_lock); |
34dc7c2f BB |
4967 | } |
4968 | ||
428870ff BB |
4969 | /* |
4970 | * Verify pool integrity by running zdb. | |
4971 | */ | |
34dc7c2f | 4972 | static void |
428870ff | 4973 | ztest_run_zdb(char *pool) |
34dc7c2f BB |
4974 | { |
4975 | int status; | |
34dc7c2f | 4976 | char *bin; |
0e8d1b2d BB |
4977 | char *zdb; |
4978 | char *zbuf; | |
34dc7c2f BB |
4979 | FILE *fp; |
4980 | ||
0e8d1b2d BB |
4981 | bin = umem_alloc(MAXPATHLEN + MAXNAMELEN + 20, UMEM_NOFAIL); |
4982 | zdb = umem_alloc(MAXPATHLEN + MAXNAMELEN + 20, UMEM_NOFAIL); | |
4983 | zbuf = umem_alloc(1024, UMEM_NOFAIL); | |
34dc7c2f | 4984 | |
0e8d1b2d | 4985 | VERIFY(realpath(getexecname(), bin) != NULL); |
341b5f1d | 4986 | if (strncmp(bin, "/usr/sbin/ztest", 15) == 0) { |
0e8d1b2d | 4987 | strcpy(bin, "/usr/sbin/zdb"); /* Installed */ |
341b5f1d BB |
4988 | } else if (strncmp(bin, "/sbin/ztest", 11) == 0) { |
4989 | strcpy(bin, "/sbin/zdb"); /* Installed */ | |
0e8d1b2d BB |
4990 | } else { |
4991 | strstr(bin, "/ztest/")[0] = '\0'; /* In-tree */ | |
4992 | strcat(bin, "/zdb/zdb"); | |
4993 | } | |
4994 | ||
4995 | (void) sprintf(zdb, | |
4996 | "%s -bcc%s%s -U %s %s", | |
4997 | bin, | |
34dc7c2f BB |
4998 | zopt_verbose >= 3 ? "s" : "", |
4999 | zopt_verbose >= 4 ? "v" : "", | |
428870ff | 5000 | spa_config_path, |
b128c09f | 5001 | pool); |
34dc7c2f BB |
5002 | |
5003 | if (zopt_verbose >= 5) | |
5004 | (void) printf("Executing %s\n", strstr(zdb, "zdb ")); | |
5005 | ||
5006 | fp = popen(zdb, "r"); | |
5007 | ||
6aec1cd5 | 5008 | while (fgets(zbuf, 1024, fp) != NULL) |
34dc7c2f BB |
5009 | if (zopt_verbose >= 3) |
5010 | (void) printf("%s", zbuf); | |
5011 | ||
5012 | status = pclose(fp); | |
5013 | ||
5014 | if (status == 0) | |
0e8d1b2d | 5015 | goto out; |
34dc7c2f BB |
5016 | |
5017 | ztest_dump_core = 0; | |
5018 | if (WIFEXITED(status)) | |
5019 | fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status)); | |
5020 | else | |
5021 | fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status)); | |
0e8d1b2d BB |
5022 | out: |
5023 | umem_free(bin, MAXPATHLEN + MAXNAMELEN + 20); | |
5024 | umem_free(zdb, MAXPATHLEN + MAXNAMELEN + 20); | |
5025 | umem_free(zbuf, 1024); | |
34dc7c2f BB |
5026 | } |
5027 | ||
5028 | static void | |
5029 | ztest_walk_pool_directory(char *header) | |
5030 | { | |
5031 | spa_t *spa = NULL; | |
5032 | ||
5033 | if (zopt_verbose >= 6) | |
5034 | (void) printf("%s\n", header); | |
5035 | ||
5036 | mutex_enter(&spa_namespace_lock); | |
5037 | while ((spa = spa_next(spa)) != NULL) | |
5038 | if (zopt_verbose >= 6) | |
5039 | (void) printf("\t%s\n", spa_name(spa)); | |
5040 | mutex_exit(&spa_namespace_lock); | |
5041 | } | |
5042 | ||
5043 | static void | |
5044 | ztest_spa_import_export(char *oldname, char *newname) | |
5045 | { | |
fb5f0bc8 | 5046 | nvlist_t *config, *newconfig; |
34dc7c2f BB |
5047 | uint64_t pool_guid; |
5048 | spa_t *spa; | |
34dc7c2f BB |
5049 | |
5050 | if (zopt_verbose >= 4) { | |
5051 | (void) printf("import/export: old = %s, new = %s\n", | |
5052 | oldname, newname); | |
5053 | } | |
5054 | ||
5055 | /* | |
5056 | * Clean up from previous runs. | |
5057 | */ | |
5058 | (void) spa_destroy(newname); | |
5059 | ||
5060 | /* | |
5061 | * Get the pool's configuration and guid. | |
5062 | */ | |
428870ff | 5063 | VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG)); |
34dc7c2f | 5064 | |
fb5f0bc8 BB |
5065 | /* |
5066 | * Kick off a scrub to tickle scrub/export races. | |
5067 | */ | |
5068 | if (ztest_random(2) == 0) | |
428870ff | 5069 | (void) spa_scan(spa, POOL_SCAN_SCRUB); |
fb5f0bc8 | 5070 | |
34dc7c2f BB |
5071 | pool_guid = spa_guid(spa); |
5072 | spa_close(spa, FTAG); | |
5073 | ||
5074 | ztest_walk_pool_directory("pools before export"); | |
5075 | ||
5076 | /* | |
5077 | * Export it. | |
5078 | */ | |
428870ff | 5079 | VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE)); |
34dc7c2f BB |
5080 | |
5081 | ztest_walk_pool_directory("pools after export"); | |
5082 | ||
fb5f0bc8 BB |
5083 | /* |
5084 | * Try to import it. | |
5085 | */ | |
5086 | newconfig = spa_tryimport(config); | |
5087 | ASSERT(newconfig != NULL); | |
5088 | nvlist_free(newconfig); | |
5089 | ||
34dc7c2f BB |
5090 | /* |
5091 | * Import it under the new name. | |
5092 | */ | |
572e2857 | 5093 | VERIFY3U(0, ==, spa_import(newname, config, NULL, 0)); |
34dc7c2f BB |
5094 | |
5095 | ztest_walk_pool_directory("pools after import"); | |
5096 | ||
5097 | /* | |
5098 | * Try to import it again -- should fail with EEXIST. | |
5099 | */ | |
572e2857 | 5100 | VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0)); |
34dc7c2f BB |
5101 | |
5102 | /* | |
5103 | * Try to import it under a different name -- should fail with EEXIST. | |
5104 | */ | |
572e2857 | 5105 | VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0)); |
34dc7c2f BB |
5106 | |
5107 | /* | |
5108 | * Verify that the pool is no longer visible under the old name. | |
5109 | */ | |
428870ff | 5110 | VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG)); |
34dc7c2f BB |
5111 | |
5112 | /* | |
5113 | * Verify that we can open and close the pool using the new name. | |
5114 | */ | |
428870ff | 5115 | VERIFY3U(0, ==, spa_open(newname, &spa, FTAG)); |
34dc7c2f BB |
5116 | ASSERT(pool_guid == spa_guid(spa)); |
5117 | spa_close(spa, FTAG); | |
5118 | ||
5119 | nvlist_free(config); | |
5120 | } | |
5121 | ||
fb5f0bc8 BB |
5122 | static void |
5123 | ztest_resume(spa_t *spa) | |
5124 | { | |
428870ff BB |
5125 | if (spa_suspended(spa) && zopt_verbose >= 6) |
5126 | (void) printf("resuming from suspended state\n"); | |
5127 | spa_vdev_state_enter(spa, SCL_NONE); | |
5128 | vdev_clear(spa, NULL); | |
5129 | (void) spa_vdev_state_exit(spa, NULL, 0); | |
5130 | (void) zio_resume(spa); | |
fb5f0bc8 BB |
5131 | } |
5132 | ||
34dc7c2f | 5133 | static void * |
fb5f0bc8 | 5134 | ztest_resume_thread(void *arg) |
34dc7c2f | 5135 | { |
b128c09f | 5136 | spa_t *spa = arg; |
34dc7c2f BB |
5137 | |
5138 | while (!ztest_exiting) { | |
428870ff BB |
5139 | if (spa_suspended(spa)) |
5140 | ztest_resume(spa); | |
5141 | (void) poll(NULL, 0, 100); | |
34dc7c2f | 5142 | } |
34dc7c2f | 5143 | |
1e33ac1e | 5144 | thread_exit(); |
428870ff | 5145 | |
1e33ac1e BB |
5146 | return (NULL); |
5147 | } | |
428870ff | 5148 | |
1e33ac1e | 5149 | #define GRACE 300 |
428870ff | 5150 | |
1e33ac1e BB |
5151 | static void |
5152 | ztest_deadman_alarm(int sig) | |
5153 | { | |
5154 | fatal(0, "failed to complete within %d seconds of deadline", GRACE); | |
428870ff BB |
5155 | } |
5156 | ||
5157 | static void | |
5158 | ztest_execute(ztest_info_t *zi, uint64_t id) | |
5159 | { | |
5160 | ztest_shared_t *zs = ztest_shared; | |
5161 | ztest_ds_t *zd = &zs->zs_zd[id % zopt_datasets]; | |
5162 | hrtime_t functime = gethrtime(); | |
d6320ddb | 5163 | int i; |
428870ff | 5164 | |
d6320ddb | 5165 | for (i = 0; i < zi->zi_iters; i++) |
428870ff BB |
5166 | zi->zi_func(zd, id); |
5167 | ||
5168 | functime = gethrtime() - functime; | |
5169 | ||
5170 | atomic_add_64(&zi->zi_call_count, 1); | |
5171 | atomic_add_64(&zi->zi_call_time, functime); | |
5172 | ||
5173 | if (zopt_verbose >= 4) { | |
5174 | Dl_info dli; | |
5175 | (void) dladdr((void *)zi->zi_func, &dli); | |
5176 | (void) printf("%6.2f sec in %s\n", | |
5177 | (double)functime / NANOSEC, dli.dli_sname); | |
5178 | } | |
5179 | } | |
5180 | ||
34dc7c2f BB |
5181 | static void * |
5182 | ztest_thread(void *arg) | |
5183 | { | |
428870ff | 5184 | uint64_t id = (uintptr_t)arg; |
34dc7c2f | 5185 | ztest_shared_t *zs = ztest_shared; |
428870ff BB |
5186 | uint64_t call_next; |
5187 | hrtime_t now; | |
34dc7c2f | 5188 | ztest_info_t *zi; |
34dc7c2f | 5189 | |
428870ff | 5190 | while ((now = gethrtime()) < zs->zs_thread_stop) { |
34dc7c2f BB |
5191 | /* |
5192 | * See if it's time to force a crash. | |
5193 | */ | |
428870ff BB |
5194 | if (now > zs->zs_thread_kill) |
5195 | ztest_kill(zs); | |
34dc7c2f BB |
5196 | |
5197 | /* | |
428870ff | 5198 | * If we're getting ENOSPC with some regularity, stop. |
34dc7c2f | 5199 | */ |
428870ff BB |
5200 | if (zs->zs_enospc_count > 10) |
5201 | break; | |
34dc7c2f BB |
5202 | |
5203 | /* | |
428870ff | 5204 | * Pick a random function to execute. |
34dc7c2f | 5205 | */ |
428870ff BB |
5206 | zi = &zs->zs_info[ztest_random(ZTEST_FUNCS)]; |
5207 | call_next = zi->zi_call_next; | |
34dc7c2f | 5208 | |
428870ff BB |
5209 | if (now >= call_next && |
5210 | atomic_cas_64(&zi->zi_call_next, call_next, call_next + | |
5211 | ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) | |
5212 | ztest_execute(zi, id); | |
5213 | } | |
34dc7c2f | 5214 | |
1e33ac1e BB |
5215 | thread_exit(); |
5216 | ||
428870ff BB |
5217 | return (NULL); |
5218 | } | |
34dc7c2f | 5219 | |
428870ff BB |
5220 | static void |
5221 | ztest_dataset_name(char *dsname, char *pool, int d) | |
5222 | { | |
5223 | (void) snprintf(dsname, MAXNAMELEN, "%s/ds_%d", pool, d); | |
5224 | } | |
34dc7c2f | 5225 | |
428870ff BB |
5226 | static void |
5227 | ztest_dataset_destroy(ztest_shared_t *zs, int d) | |
5228 | { | |
5229 | char name[MAXNAMELEN]; | |
d6320ddb | 5230 | int t; |
34dc7c2f | 5231 | |
428870ff | 5232 | ztest_dataset_name(name, zs->zs_pool, d); |
34dc7c2f | 5233 | |
428870ff BB |
5234 | if (zopt_verbose >= 3) |
5235 | (void) printf("Destroying %s to free up space\n", name); | |
34dc7c2f | 5236 | |
428870ff BB |
5237 | /* |
5238 | * Cleanup any non-standard clones and snapshots. In general, | |
5239 | * ztest thread t operates on dataset (t % zopt_datasets), | |
5240 | * so there may be more than one thing to clean up. | |
5241 | */ | |
d6320ddb | 5242 | for (t = d; t < zopt_threads; t += zopt_datasets) |
428870ff BB |
5243 | ztest_dsl_dataset_cleanup(name, t); |
5244 | ||
5245 | (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL, | |
5246 | DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN); | |
5247 | } | |
5248 | ||
5249 | static void | |
5250 | ztest_dataset_dirobj_verify(ztest_ds_t *zd) | |
5251 | { | |
5252 | uint64_t usedobjs, dirobjs, scratch; | |
5253 | ||
5254 | /* | |
5255 | * ZTEST_DIROBJ is the object directory for the entire dataset. | |
5256 | * Therefore, the number of objects in use should equal the | |
5257 | * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself. | |
5258 | * If not, we have an object leak. | |
5259 | * | |
5260 | * Note that we can only check this in ztest_dataset_open(), | |
5261 | * when the open-context and syncing-context values agree. | |
5262 | * That's because zap_count() returns the open-context value, | |
5263 | * while dmu_objset_space() returns the rootbp fill count. | |
5264 | */ | |
5265 | VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs)); | |
5266 | dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch); | |
5267 | ASSERT3U(dirobjs + 1, ==, usedobjs); | |
5268 | } | |
5269 | ||
5270 | static int | |
5271 | ztest_dataset_open(ztest_shared_t *zs, int d) | |
5272 | { | |
5273 | ztest_ds_t *zd = &zs->zs_zd[d]; | |
5274 | uint64_t committed_seq = zd->zd_seq; | |
5275 | objset_t *os; | |
5276 | zilog_t *zilog; | |
5277 | char name[MAXNAMELEN]; | |
5278 | int error; | |
5279 | ||
5280 | ztest_dataset_name(name, zs->zs_pool, d); | |
5281 | ||
1e33ac1e | 5282 | (void) rw_enter(&zs->zs_name_lock, RW_READER); |
428870ff BB |
5283 | |
5284 | error = ztest_dataset_create(name); | |
5285 | if (error == ENOSPC) { | |
1e33ac1e | 5286 | (void) rw_exit(&zs->zs_name_lock); |
428870ff BB |
5287 | ztest_record_enospc(FTAG); |
5288 | return (error); | |
34dc7c2f | 5289 | } |
428870ff | 5290 | ASSERT(error == 0 || error == EEXIST); |
34dc7c2f | 5291 | |
428870ff | 5292 | VERIFY3U(dmu_objset_hold(name, zd, &os), ==, 0); |
1e33ac1e | 5293 | (void) rw_exit(&zs->zs_name_lock); |
428870ff BB |
5294 | |
5295 | ztest_zd_init(zd, os); | |
5296 | ||
5297 | zilog = zd->zd_zilog; | |
5298 | ||
5299 | if (zilog->zl_header->zh_claim_lr_seq != 0 && | |
5300 | zilog->zl_header->zh_claim_lr_seq < committed_seq) | |
5301 | fatal(0, "missing log records: claimed %llu < committed %llu", | |
5302 | zilog->zl_header->zh_claim_lr_seq, committed_seq); | |
5303 | ||
5304 | ztest_dataset_dirobj_verify(zd); | |
5305 | ||
5306 | zil_replay(os, zd, ztest_replay_vector); | |
5307 | ||
5308 | ztest_dataset_dirobj_verify(zd); | |
5309 | ||
5310 | if (zopt_verbose >= 6) | |
5311 | (void) printf("%s replay %llu blocks, %llu records, seq %llu\n", | |
5312 | zd->zd_name, | |
5313 | (u_longlong_t)zilog->zl_parse_blk_count, | |
5314 | (u_longlong_t)zilog->zl_parse_lr_count, | |
5315 | (u_longlong_t)zilog->zl_replaying_seq); | |
5316 | ||
5317 | zilog = zil_open(os, ztest_get_data); | |
5318 | ||
5319 | if (zilog->zl_replaying_seq != 0 && | |
5320 | zilog->zl_replaying_seq < committed_seq) | |
5321 | fatal(0, "missing log records: replayed %llu < committed %llu", | |
5322 | zilog->zl_replaying_seq, committed_seq); | |
5323 | ||
5324 | return (0); | |
5325 | } | |
5326 | ||
5327 | static void | |
5328 | ztest_dataset_close(ztest_shared_t *zs, int d) | |
5329 | { | |
5330 | ztest_ds_t *zd = &zs->zs_zd[d]; | |
5331 | ||
5332 | zil_close(zd->zd_zilog); | |
5333 | dmu_objset_rele(zd->zd_os, zd); | |
5334 | ||
5335 | ztest_zd_fini(zd); | |
34dc7c2f BB |
5336 | } |
5337 | ||
5338 | /* | |
5339 | * Kick off threads to run tests on all datasets in parallel. | |
5340 | */ | |
5341 | static void | |
428870ff | 5342 | ztest_run(ztest_shared_t *zs) |
34dc7c2f | 5343 | { |
1e33ac1e | 5344 | kt_did_t *tid; |
34dc7c2f | 5345 | spa_t *spa; |
3541dc6d | 5346 | objset_t *os; |
1e33ac1e BB |
5347 | kthread_t *resume_thread; |
5348 | uint64_t object; | |
428870ff | 5349 | int error; |
d6320ddb | 5350 | int t, d; |
b128c09f BB |
5351 | |
5352 | ztest_exiting = B_FALSE; | |
34dc7c2f | 5353 | |
34dc7c2f | 5354 | /* |
428870ff | 5355 | * Initialize parent/child shared state. |
34dc7c2f | 5356 | */ |
1e33ac1e BB |
5357 | mutex_init(&zs->zs_vdev_lock, NULL, MUTEX_DEFAULT, NULL); |
5358 | rw_init(&zs->zs_name_lock, NULL, RW_DEFAULT, NULL); | |
34dc7c2f | 5359 | |
428870ff BB |
5360 | zs->zs_thread_start = gethrtime(); |
5361 | zs->zs_thread_stop = zs->zs_thread_start + zopt_passtime * NANOSEC; | |
5362 | zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop); | |
5363 | zs->zs_thread_kill = zs->zs_thread_stop; | |
5364 | if (ztest_random(100) < zopt_killrate) | |
5365 | zs->zs_thread_kill -= ztest_random(zopt_passtime * NANOSEC); | |
34dc7c2f | 5366 | |
1e33ac1e | 5367 | mutex_init(&zcl.zcl_callbacks_lock, NULL, MUTEX_DEFAULT, NULL); |
34dc7c2f | 5368 | |
428870ff BB |
5369 | list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t), |
5370 | offsetof(ztest_cb_data_t, zcd_node)); | |
34dc7c2f BB |
5371 | |
5372 | /* | |
b128c09f | 5373 | * Open our pool. |
34dc7c2f | 5374 | */ |
428870ff BB |
5375 | kernel_init(FREAD | FWRITE); |
5376 | VERIFY(spa_open(zs->zs_pool, &spa, FTAG) == 0); | |
6d974228 | 5377 | spa->spa_debug = B_TRUE; |
428870ff BB |
5378 | zs->zs_spa = spa; |
5379 | ||
3541dc6d GA |
5380 | VERIFY3U(0, ==, dmu_objset_hold(zs->zs_pool, FTAG, &os)); |
5381 | zs->zs_guid = dmu_objset_fsid_guid(os); | |
5382 | dmu_objset_rele(os, FTAG); | |
5383 | ||
428870ff | 5384 | spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN; |
34dc7c2f | 5385 | |
fb5f0bc8 BB |
5386 | /* |
5387 | * We don't expect the pool to suspend unless maxfaults == 0, | |
5388 | * in which case ztest_fault_inject() temporarily takes away | |
5389 | * the only valid replica. | |
5390 | */ | |
428870ff | 5391 | if (MAXFAULTS() == 0) |
fb5f0bc8 BB |
5392 | spa->spa_failmode = ZIO_FAILURE_MODE_WAIT; |
5393 | else | |
5394 | spa->spa_failmode = ZIO_FAILURE_MODE_PANIC; | |
5395 | ||
34dc7c2f | 5396 | /* |
b128c09f | 5397 | * Create a thread to periodically resume suspended I/O. |
34dc7c2f | 5398 | */ |
0aebd4f9 ED |
5399 | VERIFY3P((resume_thread = zk_thread_create(NULL, 0, |
5400 | (thread_func_t)ztest_resume_thread, spa, TS_RUN, NULL, 0, 0, | |
5401 | PTHREAD_CREATE_JOINABLE)), !=, NULL); | |
34dc7c2f | 5402 | |
428870ff | 5403 | /* |
1e33ac1e | 5404 | * Set a deadman alarm to abort() if we hang. |
428870ff | 5405 | */ |
1e33ac1e BB |
5406 | signal(SIGALRM, ztest_deadman_alarm); |
5407 | alarm((zs->zs_thread_stop - zs->zs_thread_start) / NANOSEC + GRACE); | |
428870ff | 5408 | |
34dc7c2f BB |
5409 | /* |
5410 | * Verify that we can safely inquire about about any object, | |
5411 | * whether it's allocated or not. To make it interesting, | |
5412 | * we probe a 5-wide window around each power of two. | |
5413 | * This hits all edge cases, including zero and the max. | |
5414 | */ | |
d6320ddb BB |
5415 | for (t = 0; t < 64; t++) { |
5416 | for (d = -5; d <= 5; d++) { | |
34dc7c2f BB |
5417 | error = dmu_object_info(spa->spa_meta_objset, |
5418 | (1ULL << t) + d, NULL); | |
5419 | ASSERT(error == 0 || error == ENOENT || | |
5420 | error == EINVAL); | |
5421 | } | |
5422 | } | |
5423 | ||
5424 | /* | |
428870ff | 5425 | * If we got any ENOSPC errors on the previous run, destroy something. |
34dc7c2f | 5426 | */ |
428870ff BB |
5427 | if (zs->zs_enospc_count != 0) { |
5428 | int d = ztest_random(zopt_datasets); | |
5429 | ztest_dataset_destroy(zs, d); | |
5430 | } | |
34dc7c2f BB |
5431 | zs->zs_enospc_count = 0; |
5432 | ||
1e33ac1e | 5433 | tid = umem_zalloc(zopt_threads * sizeof (kt_did_t), UMEM_NOFAIL); |
34dc7c2f BB |
5434 | |
5435 | if (zopt_verbose >= 4) | |
5436 | (void) printf("starting main threads...\n"); | |
5437 | ||
428870ff BB |
5438 | /* |
5439 | * Kick off all the tests that run in parallel. | |
5440 | */ | |
d6320ddb | 5441 | for (t = 0; t < zopt_threads; t++) { |
1e33ac1e BB |
5442 | kthread_t *thread; |
5443 | ||
428870ff BB |
5444 | if (t < zopt_datasets && ztest_dataset_open(zs, t) != 0) |
5445 | return; | |
1e33ac1e | 5446 | |
0aebd4f9 ED |
5447 | VERIFY3P(thread = zk_thread_create(NULL, 0, |
5448 | (thread_func_t)ztest_thread, | |
5449 | (void *)(uintptr_t)t, TS_RUN, NULL, 0, 0, | |
5450 | PTHREAD_CREATE_JOINABLE), !=, NULL); | |
1e33ac1e | 5451 | tid[t] = thread->t_tid; |
34dc7c2f BB |
5452 | } |
5453 | ||
428870ff BB |
5454 | /* |
5455 | * Wait for all of the tests to complete. We go in reverse order | |
5456 | * so we don't close datasets while threads are still using them. | |
5457 | */ | |
d6320ddb | 5458 | for (t = zopt_threads - 1; t >= 0; t--) { |
1e33ac1e | 5459 | thread_join(tid[t]); |
428870ff BB |
5460 | if (t < zopt_datasets) |
5461 | ztest_dataset_close(zs, t); | |
34dc7c2f BB |
5462 | } |
5463 | ||
34dc7c2f BB |
5464 | txg_wait_synced(spa_get_dsl(spa), 0); |
5465 | ||
428870ff BB |
5466 | zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa)); |
5467 | zs->zs_space = metaslab_class_get_space(spa_normal_class(spa)); | |
5468 | ||
1e33ac1e | 5469 | umem_free(tid, zopt_threads * sizeof (kt_did_t)); |
428870ff BB |
5470 | |
5471 | /* Kill the resume thread */ | |
5472 | ztest_exiting = B_TRUE; | |
1e33ac1e | 5473 | thread_join(resume_thread->t_tid); |
428870ff | 5474 | ztest_resume(spa); |
34dc7c2f BB |
5475 | |
5476 | /* | |
428870ff BB |
5477 | * Right before closing the pool, kick off a bunch of async I/O; |
5478 | * spa_close() should wait for it to complete. | |
34dc7c2f | 5479 | */ |
1e33ac1e | 5480 | for (object = 1; object < 50; object++) |
428870ff BB |
5481 | dmu_prefetch(spa->spa_meta_objset, object, 0, 1ULL << 20); |
5482 | ||
090ff092 RC |
5483 | /* Verify that at least one commit cb was called in a timely fashion */ |
5484 | if (zc_cb_counter >= ZTEST_COMMIT_CB_MIN_REG) | |
5485 | VERIFY3U(zc_min_txg_delay, ==, 0); | |
5486 | ||
428870ff BB |
5487 | spa_close(spa, FTAG); |
5488 | ||
5489 | /* | |
5490 | * Verify that we can loop over all pools. | |
5491 | */ | |
5492 | mutex_enter(&spa_namespace_lock); | |
5493 | for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa)) | |
5494 | if (zopt_verbose > 3) | |
5495 | (void) printf("spa_next: found %s\n", spa_name(spa)); | |
5496 | mutex_exit(&spa_namespace_lock); | |
5497 | ||
5498 | /* | |
5499 | * Verify that we can export the pool and reimport it under a | |
5500 | * different name. | |
5501 | */ | |
5502 | if (ztest_random(2) == 0) { | |
5503 | char name[MAXNAMELEN]; | |
5504 | (void) snprintf(name, MAXNAMELEN, "%s_import", zs->zs_pool); | |
5505 | ztest_spa_import_export(zs->zs_pool, name); | |
5506 | ztest_spa_import_export(name, zs->zs_pool); | |
5507 | } | |
5508 | ||
5509 | kernel_fini(); | |
572e2857 BB |
5510 | |
5511 | list_destroy(&zcl.zcl_callbacks); | |
0e8d1b2d BB |
5512 | mutex_destroy(&zcl.zcl_callbacks_lock); |
5513 | rw_destroy(&zs->zs_name_lock); | |
5514 | mutex_destroy(&zs->zs_vdev_lock); | |
428870ff BB |
5515 | } |
5516 | ||
5517 | static void | |
5518 | ztest_freeze(ztest_shared_t *zs) | |
5519 | { | |
5520 | ztest_ds_t *zd = &zs->zs_zd[0]; | |
5521 | spa_t *spa; | |
5522 | int numloops = 0; | |
5523 | ||
5524 | if (zopt_verbose >= 3) | |
5525 | (void) printf("testing spa_freeze()...\n"); | |
9babb374 | 5526 | |
428870ff BB |
5527 | kernel_init(FREAD | FWRITE); |
5528 | VERIFY3U(0, ==, spa_open(zs->zs_pool, &spa, FTAG)); | |
5529 | VERIFY3U(0, ==, ztest_dataset_open(zs, 0)); | |
9babb374 | 5530 | |
428870ff BB |
5531 | /* |
5532 | * Force the first log block to be transactionally allocated. | |
5533 | * We have to do this before we freeze the pool -- otherwise | |
5534 | * the log chain won't be anchored. | |
5535 | */ | |
5536 | while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) { | |
5537 | ztest_dmu_object_alloc_free(zd, 0); | |
572e2857 | 5538 | zil_commit(zd->zd_zilog, 0); |
34dc7c2f BB |
5539 | } |
5540 | ||
5541 | txg_wait_synced(spa_get_dsl(spa), 0); | |
5542 | ||
428870ff BB |
5543 | /* |
5544 | * Freeze the pool. This stops spa_sync() from doing anything, | |
5545 | * so that the only way to record changes from now on is the ZIL. | |
5546 | */ | |
5547 | spa_freeze(spa); | |
b128c09f | 5548 | |
428870ff BB |
5549 | /* |
5550 | * Run tests that generate log records but don't alter the pool config | |
5551 | * or depend on DSL sync tasks (snapshots, objset create/destroy, etc). | |
5552 | * We do a txg_wait_synced() after each iteration to force the txg | |
5553 | * to increase well beyond the last synced value in the uberblock. | |
5554 | * The ZIL should be OK with that. | |
5555 | */ | |
5556 | while (ztest_random(10) != 0 && numloops++ < zopt_maxloops) { | |
5557 | ztest_dmu_write_parallel(zd, 0); | |
5558 | ztest_dmu_object_alloc_free(zd, 0); | |
5559 | txg_wait_synced(spa_get_dsl(spa), 0); | |
5560 | } | |
b128c09f | 5561 | |
34dc7c2f | 5562 | /* |
428870ff | 5563 | * Commit all of the changes we just generated. |
34dc7c2f | 5564 | */ |
572e2857 | 5565 | zil_commit(zd->zd_zilog, 0); |
428870ff | 5566 | txg_wait_synced(spa_get_dsl(spa), 0); |
34dc7c2f | 5567 | |
428870ff BB |
5568 | /* |
5569 | * Close our dataset and close the pool. | |
5570 | */ | |
5571 | ztest_dataset_close(zs, 0); | |
34dc7c2f | 5572 | spa_close(spa, FTAG); |
428870ff | 5573 | kernel_fini(); |
34dc7c2f | 5574 | |
428870ff BB |
5575 | /* |
5576 | * Open and close the pool and dataset to induce log replay. | |
5577 | */ | |
5578 | kernel_init(FREAD | FWRITE); | |
5579 | VERIFY3U(0, ==, spa_open(zs->zs_pool, &spa, FTAG)); | |
5580 | VERIFY3U(0, ==, ztest_dataset_open(zs, 0)); | |
5581 | ztest_dataset_close(zs, 0); | |
5582 | spa_close(spa, FTAG); | |
34dc7c2f BB |
5583 | kernel_fini(); |
5584 | } | |
5585 | ||
5586 | void | |
5587 | print_time(hrtime_t t, char *timebuf) | |
5588 | { | |
5589 | hrtime_t s = t / NANOSEC; | |
5590 | hrtime_t m = s / 60; | |
5591 | hrtime_t h = m / 60; | |
5592 | hrtime_t d = h / 24; | |
5593 | ||
5594 | s -= m * 60; | |
5595 | m -= h * 60; | |
5596 | h -= d * 24; | |
5597 | ||
5598 | timebuf[0] = '\0'; | |
5599 | ||
5600 | if (d) | |
5601 | (void) sprintf(timebuf, | |
5602 | "%llud%02lluh%02llum%02llus", d, h, m, s); | |
5603 | else if (h) | |
5604 | (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s); | |
5605 | else if (m) | |
5606 | (void) sprintf(timebuf, "%llum%02llus", m, s); | |
5607 | else | |
5608 | (void) sprintf(timebuf, "%llus", s); | |
5609 | } | |
5610 | ||
428870ff | 5611 | static nvlist_t * |
0bc8fd78 | 5612 | make_random_props(void) |
428870ff BB |
5613 | { |
5614 | nvlist_t *props; | |
5615 | ||
5616 | if (ztest_random(2) == 0) | |
5617 | return (NULL); | |
5618 | ||
5619 | VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0); | |
5620 | VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0); | |
5621 | ||
5622 | (void) printf("props:\n"); | |
5623 | dump_nvlist(props, 4); | |
5624 | ||
5625 | return (props); | |
5626 | } | |
5627 | ||
34dc7c2f BB |
5628 | /* |
5629 | * Create a storage pool with the given name and initial vdev size. | |
428870ff | 5630 | * Then test spa_freeze() functionality. |
34dc7c2f BB |
5631 | */ |
5632 | static void | |
428870ff | 5633 | ztest_init(ztest_shared_t *zs) |
34dc7c2f BB |
5634 | { |
5635 | spa_t *spa; | |
428870ff BB |
5636 | nvlist_t *nvroot, *props; |
5637 | ||
1e33ac1e BB |
5638 | mutex_init(&zs->zs_vdev_lock, NULL, MUTEX_DEFAULT, NULL); |
5639 | rw_init(&zs->zs_name_lock, NULL, RW_DEFAULT, NULL); | |
34dc7c2f BB |
5640 | |
5641 | kernel_init(FREAD | FWRITE); | |
5642 | ||
5643 | /* | |
5644 | * Create the storage pool. | |
5645 | */ | |
428870ff BB |
5646 | (void) spa_destroy(zs->zs_pool); |
5647 | ztest_shared->zs_vdev_next_leaf = 0; | |
5648 | zs->zs_splits = 0; | |
5649 | zs->zs_mirrors = zopt_mirrors; | |
b128c09f | 5650 | nvroot = make_vdev_root(NULL, NULL, zopt_vdev_size, 0, |
428870ff BB |
5651 | 0, zopt_raidz, zs->zs_mirrors, 1); |
5652 | props = make_random_props(); | |
5653 | VERIFY3U(0, ==, spa_create(zs->zs_pool, nvroot, props, NULL, NULL)); | |
34dc7c2f BB |
5654 | nvlist_free(nvroot); |
5655 | ||
428870ff | 5656 | VERIFY3U(0, ==, spa_open(zs->zs_pool, &spa, FTAG)); |
9babb374 | 5657 | metaslab_sz = 1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift; |
34dc7c2f BB |
5658 | spa_close(spa, FTAG); |
5659 | ||
5660 | kernel_fini(); | |
428870ff BB |
5661 | |
5662 | ztest_run_zdb(zs->zs_pool); | |
5663 | ||
5664 | ztest_freeze(zs); | |
5665 | ||
5666 | ztest_run_zdb(zs->zs_pool); | |
572e2857 | 5667 | |
1e33ac1e BB |
5668 | (void) rw_destroy(&zs->zs_name_lock); |
5669 | (void) mutex_destroy(&zs->zs_vdev_lock); | |
34dc7c2f BB |
5670 | } |
5671 | ||
5672 | int | |
5673 | main(int argc, char **argv) | |
5674 | { | |
5675 | int kills = 0; | |
5676 | int iters = 0; | |
34dc7c2f | 5677 | ztest_shared_t *zs; |
428870ff | 5678 | size_t shared_size; |
34dc7c2f BB |
5679 | ztest_info_t *zi; |
5680 | char timebuf[100]; | |
5681 | char numbuf[6]; | |
428870ff | 5682 | spa_t *spa; |
d6320ddb | 5683 | int i, f; |
34dc7c2f BB |
5684 | |
5685 | (void) setvbuf(stdout, NULL, _IOLBF, 0); | |
5686 | ||
34dc7c2f BB |
5687 | ztest_random_fd = open("/dev/urandom", O_RDONLY); |
5688 | ||
0e8d1b2d | 5689 | dprintf_setup(&argc, argv); |
34dc7c2f BB |
5690 | process_options(argc, argv); |
5691 | ||
428870ff | 5692 | /* Override location of zpool.cache */ |
0e5b68e0 BB |
5693 | VERIFY(asprintf((char **)&spa_config_path, "%s/zpool.cache", |
5694 | zopt_dir) != -1); | |
428870ff | 5695 | |
34dc7c2f BB |
5696 | /* |
5697 | * Blow away any existing copy of zpool.cache | |
5698 | */ | |
5699 | if (zopt_init != 0) | |
428870ff BB |
5700 | (void) remove(spa_config_path); |
5701 | ||
5702 | shared_size = sizeof (*zs) + zopt_datasets * sizeof (ztest_ds_t); | |
34dc7c2f BB |
5703 | |
5704 | zs = ztest_shared = (void *)mmap(0, | |
428870ff | 5705 | P2ROUNDUP(shared_size, getpagesize()), |
34dc7c2f BB |
5706 | PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANON, -1, 0); |
5707 | ||
5708 | if (zopt_verbose >= 1) { | |
5709 | (void) printf("%llu vdevs, %d datasets, %d threads," | |
5710 | " %llu seconds...\n", | |
5711 | (u_longlong_t)zopt_vdevs, zopt_datasets, zopt_threads, | |
5712 | (u_longlong_t)zopt_time); | |
5713 | } | |
5714 | ||
5715 | /* | |
5716 | * Create and initialize our storage pool. | |
5717 | */ | |
d6320ddb | 5718 | for (i = 1; i <= zopt_init; i++) { |
34dc7c2f BB |
5719 | bzero(zs, sizeof (ztest_shared_t)); |
5720 | if (zopt_verbose >= 3 && zopt_init != 1) | |
5721 | (void) printf("ztest_init(), pass %d\n", i); | |
428870ff BB |
5722 | zs->zs_pool = zopt_pool; |
5723 | ztest_init(zs); | |
34dc7c2f BB |
5724 | } |
5725 | ||
428870ff BB |
5726 | zs->zs_pool = zopt_pool; |
5727 | zs->zs_proc_start = gethrtime(); | |
5728 | zs->zs_proc_stop = zs->zs_proc_start + zopt_time * NANOSEC; | |
34dc7c2f | 5729 | |
d6320ddb | 5730 | for (f = 0; f < ZTEST_FUNCS; f++) { |
428870ff | 5731 | zi = &zs->zs_info[f]; |
34dc7c2f | 5732 | *zi = ztest_info[f]; |
428870ff BB |
5733 | if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop) |
5734 | zi->zi_call_next = UINT64_MAX; | |
34dc7c2f | 5735 | else |
428870ff BB |
5736 | zi->zi_call_next = zs->zs_proc_start + |
5737 | ztest_random(2 * zi->zi_interval[0] + 1); | |
34dc7c2f BB |
5738 | } |
5739 | ||
34dc7c2f BB |
5740 | /* |
5741 | * Run the tests in a loop. These tests include fault injection | |
5742 | * to verify that self-healing data works, and forced crashes | |
5743 | * to verify that we never lose on-disk consistency. | |
5744 | */ | |
428870ff | 5745 | while (gethrtime() < zs->zs_proc_stop) { |
34dc7c2f BB |
5746 | int status; |
5747 | pid_t pid; | |
34dc7c2f BB |
5748 | |
5749 | /* | |
5750 | * Initialize the workload counters for each function. | |
5751 | */ | |
d6320ddb | 5752 | for (f = 0; f < ZTEST_FUNCS; f++) { |
34dc7c2f | 5753 | zi = &zs->zs_info[f]; |
428870ff | 5754 | zi->zi_call_count = 0; |
34dc7c2f BB |
5755 | zi->zi_call_time = 0; |
5756 | } | |
5757 | ||
9babb374 BB |
5758 | /* Set the allocation switch size */ |
5759 | metaslab_df_alloc_threshold = ztest_random(metaslab_sz / 4) + 1; | |
5760 | ||
34dc7c2f BB |
5761 | pid = fork(); |
5762 | ||
5763 | if (pid == -1) | |
5764 | fatal(1, "fork failed"); | |
5765 | ||
5766 | if (pid == 0) { /* child */ | |
5767 | struct rlimit rl = { 1024, 1024 }; | |
5768 | (void) setrlimit(RLIMIT_NOFILE, &rl); | |
5769 | (void) enable_extended_FILE_stdio(-1, -1); | |
428870ff | 5770 | ztest_run(zs); |
34dc7c2f BB |
5771 | exit(0); |
5772 | } | |
5773 | ||
5774 | while (waitpid(pid, &status, 0) != pid) | |
5775 | continue; | |
5776 | ||
5777 | if (WIFEXITED(status)) { | |
5778 | if (WEXITSTATUS(status) != 0) { | |
5779 | (void) fprintf(stderr, | |
5780 | "child exited with code %d\n", | |
5781 | WEXITSTATUS(status)); | |
5782 | exit(2); | |
5783 | } | |
5784 | } else if (WIFSIGNALED(status)) { | |
5785 | if (WTERMSIG(status) != SIGKILL) { | |
5786 | (void) fprintf(stderr, | |
5787 | "child died with signal %d\n", | |
5788 | WTERMSIG(status)); | |
5789 | exit(3); | |
5790 | } | |
5791 | kills++; | |
5792 | } else { | |
5793 | (void) fprintf(stderr, "something strange happened " | |
5794 | "to child\n"); | |
5795 | exit(4); | |
5796 | } | |
5797 | ||
5798 | iters++; | |
5799 | ||
5800 | if (zopt_verbose >= 1) { | |
5801 | hrtime_t now = gethrtime(); | |
5802 | ||
428870ff BB |
5803 | now = MIN(now, zs->zs_proc_stop); |
5804 | print_time(zs->zs_proc_stop - now, timebuf); | |
34dc7c2f BB |
5805 | nicenum(zs->zs_space, numbuf); |
5806 | ||
5807 | (void) printf("Pass %3d, %8s, %3llu ENOSPC, " | |
5808 | "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n", | |
5809 | iters, | |
5810 | WIFEXITED(status) ? "Complete" : "SIGKILL", | |
5811 | (u_longlong_t)zs->zs_enospc_count, | |
5812 | 100.0 * zs->zs_alloc / zs->zs_space, | |
5813 | numbuf, | |
428870ff | 5814 | 100.0 * (now - zs->zs_proc_start) / |
34dc7c2f BB |
5815 | (zopt_time * NANOSEC), timebuf); |
5816 | } | |
5817 | ||
5818 | if (zopt_verbose >= 2) { | |
5819 | (void) printf("\nWorkload summary:\n\n"); | |
5820 | (void) printf("%7s %9s %s\n", | |
5821 | "Calls", "Time", "Function"); | |
5822 | (void) printf("%7s %9s %s\n", | |
5823 | "-----", "----", "--------"); | |
d6320ddb | 5824 | for (f = 0; f < ZTEST_FUNCS; f++) { |
34dc7c2f BB |
5825 | Dl_info dli; |
5826 | ||
5827 | zi = &zs->zs_info[f]; | |
5828 | print_time(zi->zi_call_time, timebuf); | |
5829 | (void) dladdr((void *)zi->zi_func, &dli); | |
5830 | (void) printf("%7llu %9s %s\n", | |
428870ff | 5831 | (u_longlong_t)zi->zi_call_count, timebuf, |
34dc7c2f BB |
5832 | dli.dli_sname); |
5833 | } | |
5834 | (void) printf("\n"); | |
5835 | } | |
5836 | ||
5837 | /* | |
428870ff BB |
5838 | * It's possible that we killed a child during a rename test, |
5839 | * in which case we'll have a 'ztest_tmp' pool lying around | |
5840 | * instead of 'ztest'. Do a blind rename in case this happened. | |
34dc7c2f | 5841 | */ |
428870ff BB |
5842 | kernel_init(FREAD); |
5843 | if (spa_open(zopt_pool, &spa, FTAG) == 0) { | |
5844 | spa_close(spa, FTAG); | |
5845 | } else { | |
5846 | char tmpname[MAXNAMELEN]; | |
5847 | kernel_fini(); | |
5848 | kernel_init(FREAD | FWRITE); | |
5849 | (void) snprintf(tmpname, sizeof (tmpname), "%s_tmp", | |
5850 | zopt_pool); | |
5851 | (void) spa_rename(tmpname, zopt_pool); | |
5852 | } | |
34dc7c2f | 5853 | kernel_fini(); |
34dc7c2f | 5854 | |
428870ff BB |
5855 | ztest_run_zdb(zopt_pool); |
5856 | } | |
34dc7c2f BB |
5857 | |
5858 | if (zopt_verbose >= 1) { | |
5859 | (void) printf("%d killed, %d completed, %.0f%% kill rate\n", | |
5860 | kills, iters - kills, (100.0 * kills) / MAX(1, iters)); | |
5861 | } | |
5862 | ||
5863 | return (0); | |
5864 | } |