]>
Commit | Line | Data |
---|---|---|
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 | /* | |
22 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. | |
23 | * Copyright (c) 2013 by Delphix. All rights reserved. | |
24 | * Copyright 2011 Nexenta Systems, Inc. All rights reserved. | |
25 | * Copyright (c) 2013 Steven Hartland. All rights reserved. | |
26 | */ | |
27 | ||
28 | /* | |
29 | * The objective of this program is to provide a DMU/ZAP/SPA stress test | |
30 | * that runs entirely in userland, is easy to use, and easy to extend. | |
31 | * | |
32 | * The overall design of the ztest program is as follows: | |
33 | * | |
34 | * (1) For each major functional area (e.g. adding vdevs to a pool, | |
35 | * creating and destroying datasets, reading and writing objects, etc) | |
36 | * we have a simple routine to test that functionality. These | |
37 | * individual routines do not have to do anything "stressful". | |
38 | * | |
39 | * (2) We turn these simple functionality tests into a stress test by | |
40 | * running them all in parallel, with as many threads as desired, | |
41 | * and spread across as many datasets, objects, and vdevs as desired. | |
42 | * | |
43 | * (3) While all this is happening, we inject faults into the pool to | |
44 | * verify that self-healing data really works. | |
45 | * | |
46 | * (4) Every time we open a dataset, we change its checksum and compression | |
47 | * functions. Thus even individual objects vary from block to block | |
48 | * in which checksum they use and whether they're compressed. | |
49 | * | |
50 | * (5) To verify that we never lose on-disk consistency after a crash, | |
51 | * we run the entire test in a child of the main process. | |
52 | * At random times, the child self-immolates with a SIGKILL. | |
53 | * This is the software equivalent of pulling the power cord. | |
54 | * The parent then runs the test again, using the existing | |
55 | * storage pool, as many times as desired. If backwards compatability | |
56 | * testing is enabled ztest will sometimes run the "older" version | |
57 | * of ztest after a SIGKILL. | |
58 | * | |
59 | * (6) To verify that we don't have future leaks or temporal incursions, | |
60 | * many of the functional tests record the transaction group number | |
61 | * as part of their data. When reading old data, they verify that | |
62 | * the transaction group number is less than the current, open txg. | |
63 | * If you add a new test, please do this if applicable. | |
64 | * | |
65 | * (7) Threads are created with a reduced stack size, for sanity checking. | |
66 | * Therefore, it's important not to allocate huge buffers on the stack. | |
67 | * | |
68 | * When run with no arguments, ztest runs for about five minutes and | |
69 | * produces no output if successful. To get a little bit of information, | |
70 | * specify -V. To get more information, specify -VV, and so on. | |
71 | * | |
72 | * To turn this into an overnight stress test, use -T to specify run time. | |
73 | * | |
74 | * You can ask more more vdevs [-v], datasets [-d], or threads [-t] | |
75 | * to increase the pool capacity, fanout, and overall stress level. | |
76 | * | |
77 | * Use the -k option to set the desired frequency of kills. | |
78 | * | |
79 | * When ztest invokes itself it passes all relevant information through a | |
80 | * temporary file which is mmap-ed in the child process. This allows shared | |
81 | * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always | |
82 | * stored at offset 0 of this file and contains information on the size and | |
83 | * number of shared structures in the file. The information stored in this file | |
84 | * must remain backwards compatible with older versions of ztest so that | |
85 | * ztest can invoke them during backwards compatibility testing (-B). | |
86 | */ | |
87 | ||
88 | #include <sys/zfs_context.h> | |
89 | #include <sys/spa.h> | |
90 | #include <sys/dmu.h> | |
91 | #include <sys/txg.h> | |
92 | #include <sys/dbuf.h> | |
93 | #include <sys/zap.h> | |
94 | #include <sys/dmu_objset.h> | |
95 | #include <sys/poll.h> | |
96 | #include <sys/stat.h> | |
97 | #include <sys/time.h> | |
98 | #include <sys/wait.h> | |
99 | #include <sys/mman.h> | |
100 | #include <sys/resource.h> | |
101 | #include <sys/zio.h> | |
102 | #include <sys/zil.h> | |
103 | #include <sys/zil_impl.h> | |
104 | #include <sys/vdev_impl.h> | |
105 | #include <sys/vdev_file.h> | |
106 | #include <sys/spa_impl.h> | |
107 | #include <sys/metaslab_impl.h> | |
108 | #include <sys/dsl_prop.h> | |
109 | #include <sys/dsl_dataset.h> | |
110 | #include <sys/dsl_destroy.h> | |
111 | #include <sys/dsl_scan.h> | |
112 | #include <sys/zio_checksum.h> | |
113 | #include <sys/refcount.h> | |
114 | #include <sys/zfeature.h> | |
115 | #include <sys/dsl_userhold.h> | |
116 | #include <stdio.h> | |
117 | #include <stdio_ext.h> | |
118 | #include <stdlib.h> | |
119 | #include <unistd.h> | |
120 | #include <signal.h> | |
121 | #include <umem.h> | |
122 | #include <dlfcn.h> | |
123 | #include <ctype.h> | |
124 | #include <math.h> | |
125 | #include <sys/fs/zfs.h> | |
126 | #include <libnvpair.h> | |
127 | ||
128 | static int ztest_fd_data = -1; | |
129 | static int ztest_fd_rand = -1; | |
130 | ||
131 | typedef struct ztest_shared_hdr { | |
132 | uint64_t zh_hdr_size; | |
133 | uint64_t zh_opts_size; | |
134 | uint64_t zh_size; | |
135 | uint64_t zh_stats_size; | |
136 | uint64_t zh_stats_count; | |
137 | uint64_t zh_ds_size; | |
138 | uint64_t zh_ds_count; | |
139 | } ztest_shared_hdr_t; | |
140 | ||
141 | static ztest_shared_hdr_t *ztest_shared_hdr; | |
142 | ||
143 | typedef struct ztest_shared_opts { | |
144 | char zo_pool[MAXNAMELEN]; | |
145 | char zo_dir[MAXNAMELEN]; | |
146 | char zo_alt_ztest[MAXNAMELEN]; | |
147 | char zo_alt_libpath[MAXNAMELEN]; | |
148 | uint64_t zo_vdevs; | |
149 | uint64_t zo_vdevtime; | |
150 | size_t zo_vdev_size; | |
151 | int zo_ashift; | |
152 | int zo_mirrors; | |
153 | int zo_raidz; | |
154 | int zo_raidz_parity; | |
155 | int zo_datasets; | |
156 | int zo_threads; | |
157 | uint64_t zo_passtime; | |
158 | uint64_t zo_killrate; | |
159 | int zo_verbose; | |
160 | int zo_init; | |
161 | uint64_t zo_time; | |
162 | uint64_t zo_maxloops; | |
163 | uint64_t zo_metaslab_gang_bang; | |
164 | } ztest_shared_opts_t; | |
165 | ||
166 | static const ztest_shared_opts_t ztest_opts_defaults = { | |
167 | .zo_pool = { 'z', 't', 'e', 's', 't', '\0' }, | |
168 | .zo_dir = { '/', 't', 'm', 'p', '\0' }, | |
169 | .zo_alt_ztest = { '\0' }, | |
170 | .zo_alt_libpath = { '\0' }, | |
171 | .zo_vdevs = 5, | |
172 | .zo_ashift = SPA_MINBLOCKSHIFT, | |
173 | .zo_mirrors = 2, | |
174 | .zo_raidz = 4, | |
175 | .zo_raidz_parity = 1, | |
176 | .zo_vdev_size = SPA_MINDEVSIZE, | |
177 | .zo_datasets = 7, | |
178 | .zo_threads = 23, | |
179 | .zo_passtime = 60, /* 60 seconds */ | |
180 | .zo_killrate = 70, /* 70% kill rate */ | |
181 | .zo_verbose = 0, | |
182 | .zo_init = 1, | |
183 | .zo_time = 300, /* 5 minutes */ | |
184 | .zo_maxloops = 50, /* max loops during spa_freeze() */ | |
185 | .zo_metaslab_gang_bang = 32 << 10 | |
186 | }; | |
187 | ||
188 | extern uint64_t metaslab_gang_bang; | |
189 | extern uint64_t metaslab_df_alloc_threshold; | |
190 | ||
191 | static ztest_shared_opts_t *ztest_shared_opts; | |
192 | static ztest_shared_opts_t ztest_opts; | |
193 | ||
194 | typedef struct ztest_shared_ds { | |
195 | uint64_t zd_seq; | |
196 | } ztest_shared_ds_t; | |
197 | ||
198 | static ztest_shared_ds_t *ztest_shared_ds; | |
199 | #define ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d]) | |
200 | ||
201 | #define BT_MAGIC 0x123456789abcdefULL | |
202 | #define MAXFAULTS() \ | |
203 | (MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1) | |
204 | ||
205 | enum ztest_io_type { | |
206 | ZTEST_IO_WRITE_TAG, | |
207 | ZTEST_IO_WRITE_PATTERN, | |
208 | ZTEST_IO_WRITE_ZEROES, | |
209 | ZTEST_IO_TRUNCATE, | |
210 | ZTEST_IO_SETATTR, | |
211 | ZTEST_IO_REWRITE, | |
212 | ZTEST_IO_TYPES | |
213 | }; | |
214 | ||
215 | typedef struct ztest_block_tag { | |
216 | uint64_t bt_magic; | |
217 | uint64_t bt_objset; | |
218 | uint64_t bt_object; | |
219 | uint64_t bt_offset; | |
220 | uint64_t bt_gen; | |
221 | uint64_t bt_txg; | |
222 | uint64_t bt_crtxg; | |
223 | } ztest_block_tag_t; | |
224 | ||
225 | typedef struct bufwad { | |
226 | uint64_t bw_index; | |
227 | uint64_t bw_txg; | |
228 | uint64_t bw_data; | |
229 | } bufwad_t; | |
230 | ||
231 | /* | |
232 | * XXX -- fix zfs range locks to be generic so we can use them here. | |
233 | */ | |
234 | typedef enum { | |
235 | RL_READER, | |
236 | RL_WRITER, | |
237 | RL_APPEND | |
238 | } rl_type_t; | |
239 | ||
240 | typedef struct rll { | |
241 | void *rll_writer; | |
242 | int rll_readers; | |
243 | kmutex_t rll_lock; | |
244 | kcondvar_t rll_cv; | |
245 | } rll_t; | |
246 | ||
247 | typedef struct rl { | |
248 | uint64_t rl_object; | |
249 | uint64_t rl_offset; | |
250 | uint64_t rl_size; | |
251 | rll_t *rl_lock; | |
252 | } rl_t; | |
253 | ||
254 | #define ZTEST_RANGE_LOCKS 64 | |
255 | #define ZTEST_OBJECT_LOCKS 64 | |
256 | ||
257 | /* | |
258 | * Object descriptor. Used as a template for object lookup/create/remove. | |
259 | */ | |
260 | typedef struct ztest_od { | |
261 | uint64_t od_dir; | |
262 | uint64_t od_object; | |
263 | dmu_object_type_t od_type; | |
264 | dmu_object_type_t od_crtype; | |
265 | uint64_t od_blocksize; | |
266 | uint64_t od_crblocksize; | |
267 | uint64_t od_gen; | |
268 | uint64_t od_crgen; | |
269 | char od_name[MAXNAMELEN]; | |
270 | } ztest_od_t; | |
271 | ||
272 | /* | |
273 | * Per-dataset state. | |
274 | */ | |
275 | typedef struct ztest_ds { | |
276 | ztest_shared_ds_t *zd_shared; | |
277 | objset_t *zd_os; | |
278 | krwlock_t zd_zilog_lock; | |
279 | zilog_t *zd_zilog; | |
280 | ztest_od_t *zd_od; /* debugging aid */ | |
281 | char zd_name[MAXNAMELEN]; | |
282 | kmutex_t zd_dirobj_lock; | |
283 | rll_t zd_object_lock[ZTEST_OBJECT_LOCKS]; | |
284 | rll_t zd_range_lock[ZTEST_RANGE_LOCKS]; | |
285 | } ztest_ds_t; | |
286 | ||
287 | /* | |
288 | * Per-iteration state. | |
289 | */ | |
290 | typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id); | |
291 | ||
292 | typedef struct ztest_info { | |
293 | ztest_func_t *zi_func; /* test function */ | |
294 | uint64_t zi_iters; /* iterations per execution */ | |
295 | uint64_t *zi_interval; /* execute every <interval> seconds */ | |
296 | } ztest_info_t; | |
297 | ||
298 | typedef struct ztest_shared_callstate { | |
299 | uint64_t zc_count; /* per-pass count */ | |
300 | uint64_t zc_time; /* per-pass time */ | |
301 | uint64_t zc_next; /* next time to call this function */ | |
302 | } ztest_shared_callstate_t; | |
303 | ||
304 | static ztest_shared_callstate_t *ztest_shared_callstate; | |
305 | #define ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c]) | |
306 | ||
307 | /* | |
308 | * Note: these aren't static because we want dladdr() to work. | |
309 | */ | |
310 | ztest_func_t ztest_dmu_read_write; | |
311 | ztest_func_t ztest_dmu_write_parallel; | |
312 | ztest_func_t ztest_dmu_object_alloc_free; | |
313 | ztest_func_t ztest_dmu_commit_callbacks; | |
314 | ztest_func_t ztest_zap; | |
315 | ztest_func_t ztest_zap_parallel; | |
316 | ztest_func_t ztest_zil_commit; | |
317 | ztest_func_t ztest_zil_remount; | |
318 | ztest_func_t ztest_dmu_read_write_zcopy; | |
319 | ztest_func_t ztest_dmu_objset_create_destroy; | |
320 | ztest_func_t ztest_dmu_prealloc; | |
321 | ztest_func_t ztest_fzap; | |
322 | ztest_func_t ztest_dmu_snapshot_create_destroy; | |
323 | ztest_func_t ztest_dsl_prop_get_set; | |
324 | ztest_func_t ztest_spa_prop_get_set; | |
325 | ztest_func_t ztest_spa_create_destroy; | |
326 | ztest_func_t ztest_fault_inject; | |
327 | ztest_func_t ztest_ddt_repair; | |
328 | ztest_func_t ztest_dmu_snapshot_hold; | |
329 | ztest_func_t ztest_spa_rename; | |
330 | ztest_func_t ztest_scrub; | |
331 | ztest_func_t ztest_dsl_dataset_promote_busy; | |
332 | ztest_func_t ztest_vdev_attach_detach; | |
333 | ztest_func_t ztest_vdev_LUN_growth; | |
334 | ztest_func_t ztest_vdev_add_remove; | |
335 | ztest_func_t ztest_vdev_aux_add_remove; | |
336 | ztest_func_t ztest_split_pool; | |
337 | ztest_func_t ztest_reguid; | |
338 | ztest_func_t ztest_spa_upgrade; | |
339 | ||
340 | uint64_t zopt_always = 0ULL * NANOSEC; /* all the time */ | |
341 | uint64_t zopt_incessant = 1ULL * NANOSEC / 10; /* every 1/10 second */ | |
342 | uint64_t zopt_often = 1ULL * NANOSEC; /* every second */ | |
343 | uint64_t zopt_sometimes = 10ULL * NANOSEC; /* every 10 seconds */ | |
344 | uint64_t zopt_rarely = 60ULL * NANOSEC; /* every 60 seconds */ | |
345 | ||
346 | ztest_info_t ztest_info[] = { | |
347 | { ztest_dmu_read_write, 1, &zopt_always }, | |
348 | { ztest_dmu_write_parallel, 10, &zopt_always }, | |
349 | { ztest_dmu_object_alloc_free, 1, &zopt_always }, | |
350 | { ztest_dmu_commit_callbacks, 1, &zopt_always }, | |
351 | { ztest_zap, 30, &zopt_always }, | |
352 | { ztest_zap_parallel, 100, &zopt_always }, | |
353 | { ztest_split_pool, 1, &zopt_always }, | |
354 | { ztest_zil_commit, 1, &zopt_incessant }, | |
355 | { ztest_zil_remount, 1, &zopt_sometimes }, | |
356 | { ztest_dmu_read_write_zcopy, 1, &zopt_often }, | |
357 | { ztest_dmu_objset_create_destroy, 1, &zopt_often }, | |
358 | { ztest_dsl_prop_get_set, 1, &zopt_often }, | |
359 | { ztest_spa_prop_get_set, 1, &zopt_sometimes }, | |
360 | #if 0 | |
361 | { ztest_dmu_prealloc, 1, &zopt_sometimes }, | |
362 | #endif | |
363 | { ztest_fzap, 1, &zopt_sometimes }, | |
364 | { ztest_dmu_snapshot_create_destroy, 1, &zopt_sometimes }, | |
365 | { ztest_spa_create_destroy, 1, &zopt_sometimes }, | |
366 | { ztest_fault_inject, 1, &zopt_sometimes }, | |
367 | { ztest_ddt_repair, 1, &zopt_sometimes }, | |
368 | { ztest_dmu_snapshot_hold, 1, &zopt_sometimes }, | |
369 | { ztest_reguid, 1, &zopt_rarely }, | |
370 | { ztest_spa_rename, 1, &zopt_rarely }, | |
371 | { ztest_scrub, 1, &zopt_rarely }, | |
372 | { ztest_spa_upgrade, 1, &zopt_rarely }, | |
373 | { ztest_dsl_dataset_promote_busy, 1, &zopt_rarely }, | |
374 | { ztest_vdev_attach_detach, 1, &zopt_sometimes }, | |
375 | { ztest_vdev_LUN_growth, 1, &zopt_rarely }, | |
376 | { ztest_vdev_add_remove, 1, | |
377 | &ztest_opts.zo_vdevtime }, | |
378 | { ztest_vdev_aux_add_remove, 1, | |
379 | &ztest_opts.zo_vdevtime }, | |
380 | }; | |
381 | ||
382 | #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t)) | |
383 | ||
384 | /* | |
385 | * The following struct is used to hold a list of uncalled commit callbacks. | |
386 | * The callbacks are ordered by txg number. | |
387 | */ | |
388 | typedef struct ztest_cb_list { | |
389 | kmutex_t zcl_callbacks_lock; | |
390 | list_t zcl_callbacks; | |
391 | } ztest_cb_list_t; | |
392 | ||
393 | /* | |
394 | * Stuff we need to share writably between parent and child. | |
395 | */ | |
396 | typedef struct ztest_shared { | |
397 | boolean_t zs_do_init; | |
398 | hrtime_t zs_proc_start; | |
399 | hrtime_t zs_proc_stop; | |
400 | hrtime_t zs_thread_start; | |
401 | hrtime_t zs_thread_stop; | |
402 | hrtime_t zs_thread_kill; | |
403 | uint64_t zs_enospc_count; | |
404 | uint64_t zs_vdev_next_leaf; | |
405 | uint64_t zs_vdev_aux; | |
406 | uint64_t zs_alloc; | |
407 | uint64_t zs_space; | |
408 | uint64_t zs_splits; | |
409 | uint64_t zs_mirrors; | |
410 | uint64_t zs_metaslab_sz; | |
411 | uint64_t zs_metaslab_df_alloc_threshold; | |
412 | uint64_t zs_guid; | |
413 | } ztest_shared_t; | |
414 | ||
415 | #define ID_PARALLEL -1ULL | |
416 | ||
417 | static char ztest_dev_template[] = "%s/%s.%llua"; | |
418 | static char ztest_aux_template[] = "%s/%s.%s.%llu"; | |
419 | ztest_shared_t *ztest_shared; | |
420 | ||
421 | static spa_t *ztest_spa = NULL; | |
422 | static ztest_ds_t *ztest_ds; | |
423 | ||
424 | static kmutex_t ztest_vdev_lock; | |
425 | ||
426 | /* | |
427 | * The ztest_name_lock protects the pool and dataset namespace used by | |
428 | * the individual tests. To modify the namespace, consumers must grab | |
429 | * this lock as writer. Grabbing the lock as reader will ensure that the | |
430 | * namespace does not change while the lock is held. | |
431 | */ | |
432 | static krwlock_t ztest_name_lock; | |
433 | ||
434 | static boolean_t ztest_dump_core = B_TRUE; | |
435 | static boolean_t ztest_exiting; | |
436 | ||
437 | /* Global commit callback list */ | |
438 | static ztest_cb_list_t zcl; | |
439 | /* Commit cb delay */ | |
440 | static uint64_t zc_min_txg_delay = UINT64_MAX; | |
441 | static int zc_cb_counter = 0; | |
442 | ||
443 | /* | |
444 | * Minimum number of commit callbacks that need to be registered for us to check | |
445 | * whether the minimum txg delay is acceptable. | |
446 | */ | |
447 | #define ZTEST_COMMIT_CB_MIN_REG 100 | |
448 | ||
449 | /* | |
450 | * If a number of txgs equal to this threshold have been created after a commit | |
451 | * callback has been registered but not called, then we assume there is an | |
452 | * implementation bug. | |
453 | */ | |
454 | #define ZTEST_COMMIT_CB_THRESH (TXG_CONCURRENT_STATES + 1000) | |
455 | ||
456 | extern uint64_t metaslab_gang_bang; | |
457 | extern uint64_t metaslab_df_alloc_threshold; | |
458 | ||
459 | enum ztest_object { | |
460 | ZTEST_META_DNODE = 0, | |
461 | ZTEST_DIROBJ, | |
462 | ZTEST_OBJECTS | |
463 | }; | |
464 | ||
465 | static void usage(boolean_t) __NORETURN; | |
466 | ||
467 | /* | |
468 | * These libumem hooks provide a reasonable set of defaults for the allocator's | |
469 | * debugging facilities. | |
470 | */ | |
471 | const char * | |
472 | _umem_debug_init(void) | |
473 | { | |
474 | return ("default,verbose"); /* $UMEM_DEBUG setting */ | |
475 | } | |
476 | ||
477 | const char * | |
478 | _umem_logging_init(void) | |
479 | { | |
480 | return ("fail,contents"); /* $UMEM_LOGGING setting */ | |
481 | } | |
482 | ||
483 | #define FATAL_MSG_SZ 1024 | |
484 | ||
485 | char *fatal_msg; | |
486 | ||
487 | static void | |
488 | fatal(int do_perror, char *message, ...) | |
489 | { | |
490 | va_list args; | |
491 | int save_errno = errno; | |
492 | char *buf; | |
493 | ||
494 | (void) fflush(stdout); | |
495 | buf = umem_alloc(FATAL_MSG_SZ, UMEM_NOFAIL); | |
496 | ||
497 | va_start(args, message); | |
498 | (void) sprintf(buf, "ztest: "); | |
499 | /* LINTED */ | |
500 | (void) vsprintf(buf + strlen(buf), message, args); | |
501 | va_end(args); | |
502 | if (do_perror) { | |
503 | (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf), | |
504 | ": %s", strerror(save_errno)); | |
505 | } | |
506 | (void) fprintf(stderr, "%s\n", buf); | |
507 | fatal_msg = buf; /* to ease debugging */ | |
508 | if (ztest_dump_core) | |
509 | abort(); | |
510 | exit(3); | |
511 | } | |
512 | ||
513 | static int | |
514 | str2shift(const char *buf) | |
515 | { | |
516 | const char *ends = "BKMGTPEZ"; | |
517 | int i; | |
518 | ||
519 | if (buf[0] == '\0') | |
520 | return (0); | |
521 | for (i = 0; i < strlen(ends); i++) { | |
522 | if (toupper(buf[0]) == ends[i]) | |
523 | break; | |
524 | } | |
525 | if (i == strlen(ends)) { | |
526 | (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", | |
527 | buf); | |
528 | usage(B_FALSE); | |
529 | } | |
530 | if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) { | |
531 | return (10*i); | |
532 | } | |
533 | (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf); | |
534 | usage(B_FALSE); | |
535 | /* NOTREACHED */ | |
536 | } | |
537 | ||
538 | static uint64_t | |
539 | nicenumtoull(const char *buf) | |
540 | { | |
541 | char *end; | |
542 | uint64_t val; | |
543 | ||
544 | val = strtoull(buf, &end, 0); | |
545 | if (end == buf) { | |
546 | (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf); | |
547 | usage(B_FALSE); | |
548 | } else if (end[0] == '.') { | |
549 | double fval = strtod(buf, &end); | |
550 | fval *= pow(2, str2shift(end)); | |
551 | if (fval > UINT64_MAX) { | |
552 | (void) fprintf(stderr, "ztest: value too large: %s\n", | |
553 | buf); | |
554 | usage(B_FALSE); | |
555 | } | |
556 | val = (uint64_t)fval; | |
557 | } else { | |
558 | int shift = str2shift(end); | |
559 | if (shift >= 64 || (val << shift) >> shift != val) { | |
560 | (void) fprintf(stderr, "ztest: value too large: %s\n", | |
561 | buf); | |
562 | usage(B_FALSE); | |
563 | } | |
564 | val <<= shift; | |
565 | } | |
566 | return (val); | |
567 | } | |
568 | ||
569 | static void | |
570 | usage(boolean_t requested) | |
571 | { | |
572 | const ztest_shared_opts_t *zo = &ztest_opts_defaults; | |
573 | ||
574 | char nice_vdev_size[10]; | |
575 | char nice_gang_bang[10]; | |
576 | FILE *fp = requested ? stdout : stderr; | |
577 | ||
578 | nicenum(zo->zo_vdev_size, nice_vdev_size); | |
579 | nicenum(zo->zo_metaslab_gang_bang, nice_gang_bang); | |
580 | ||
581 | (void) fprintf(fp, "Usage: %s\n" | |
582 | "\t[-v vdevs (default: %llu)]\n" | |
583 | "\t[-s size_of_each_vdev (default: %s)]\n" | |
584 | "\t[-a alignment_shift (default: %d)] use 0 for random\n" | |
585 | "\t[-m mirror_copies (default: %d)]\n" | |
586 | "\t[-r raidz_disks (default: %d)]\n" | |
587 | "\t[-R raidz_parity (default: %d)]\n" | |
588 | "\t[-d datasets (default: %d)]\n" | |
589 | "\t[-t threads (default: %d)]\n" | |
590 | "\t[-g gang_block_threshold (default: %s)]\n" | |
591 | "\t[-i init_count (default: %d)] initialize pool i times\n" | |
592 | "\t[-k kill_percentage (default: %llu%%)]\n" | |
593 | "\t[-p pool_name (default: %s)]\n" | |
594 | "\t[-f dir (default: %s)] file directory for vdev files\n" | |
595 | "\t[-V] verbose (use multiple times for ever more blather)\n" | |
596 | "\t[-E] use existing pool instead of creating new one\n" | |
597 | "\t[-T time (default: %llu sec)] total run time\n" | |
598 | "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n" | |
599 | "\t[-P passtime (default: %llu sec)] time per pass\n" | |
600 | "\t[-B alt_ztest (default: <none>)] alternate ztest path\n" | |
601 | "\t[-h] (print help)\n" | |
602 | "", | |
603 | zo->zo_pool, | |
604 | (u_longlong_t)zo->zo_vdevs, /* -v */ | |
605 | nice_vdev_size, /* -s */ | |
606 | zo->zo_ashift, /* -a */ | |
607 | zo->zo_mirrors, /* -m */ | |
608 | zo->zo_raidz, /* -r */ | |
609 | zo->zo_raidz_parity, /* -R */ | |
610 | zo->zo_datasets, /* -d */ | |
611 | zo->zo_threads, /* -t */ | |
612 | nice_gang_bang, /* -g */ | |
613 | zo->zo_init, /* -i */ | |
614 | (u_longlong_t)zo->zo_killrate, /* -k */ | |
615 | zo->zo_pool, /* -p */ | |
616 | zo->zo_dir, /* -f */ | |
617 | (u_longlong_t)zo->zo_time, /* -T */ | |
618 | (u_longlong_t)zo->zo_maxloops, /* -F */ | |
619 | (u_longlong_t)zo->zo_passtime); | |
620 | exit(requested ? 0 : 1); | |
621 | } | |
622 | ||
623 | static void | |
624 | process_options(int argc, char **argv) | |
625 | { | |
626 | char *path; | |
627 | ztest_shared_opts_t *zo = &ztest_opts; | |
628 | ||
629 | int opt; | |
630 | uint64_t value; | |
631 | char altdir[MAXNAMELEN] = { 0 }; | |
632 | ||
633 | bcopy(&ztest_opts_defaults, zo, sizeof (*zo)); | |
634 | ||
635 | while ((opt = getopt(argc, argv, | |
636 | "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:")) != EOF) { | |
637 | value = 0; | |
638 | switch (opt) { | |
639 | case 'v': | |
640 | case 's': | |
641 | case 'a': | |
642 | case 'm': | |
643 | case 'r': | |
644 | case 'R': | |
645 | case 'd': | |
646 | case 't': | |
647 | case 'g': | |
648 | case 'i': | |
649 | case 'k': | |
650 | case 'T': | |
651 | case 'P': | |
652 | case 'F': | |
653 | value = nicenumtoull(optarg); | |
654 | } | |
655 | switch (opt) { | |
656 | case 'v': | |
657 | zo->zo_vdevs = value; | |
658 | break; | |
659 | case 's': | |
660 | zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value); | |
661 | break; | |
662 | case 'a': | |
663 | zo->zo_ashift = value; | |
664 | break; | |
665 | case 'm': | |
666 | zo->zo_mirrors = value; | |
667 | break; | |
668 | case 'r': | |
669 | zo->zo_raidz = MAX(1, value); | |
670 | break; | |
671 | case 'R': | |
672 | zo->zo_raidz_parity = MIN(MAX(value, 1), 3); | |
673 | break; | |
674 | case 'd': | |
675 | zo->zo_datasets = MAX(1, value); | |
676 | break; | |
677 | case 't': | |
678 | zo->zo_threads = MAX(1, value); | |
679 | break; | |
680 | case 'g': | |
681 | zo->zo_metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1, | |
682 | value); | |
683 | break; | |
684 | case 'i': | |
685 | zo->zo_init = value; | |
686 | break; | |
687 | case 'k': | |
688 | zo->zo_killrate = value; | |
689 | break; | |
690 | case 'p': | |
691 | (void) strlcpy(zo->zo_pool, optarg, | |
692 | sizeof (zo->zo_pool)); | |
693 | break; | |
694 | case 'f': | |
695 | path = realpath(optarg, NULL); | |
696 | if (path == NULL) { | |
697 | (void) fprintf(stderr, "error: %s: %s\n", | |
698 | optarg, strerror(errno)); | |
699 | usage(B_FALSE); | |
700 | } else { | |
701 | (void) strlcpy(zo->zo_dir, path, | |
702 | sizeof (zo->zo_dir)); | |
703 | } | |
704 | break; | |
705 | case 'V': | |
706 | zo->zo_verbose++; | |
707 | break; | |
708 | case 'E': | |
709 | zo->zo_init = 0; | |
710 | break; | |
711 | case 'T': | |
712 | zo->zo_time = value; | |
713 | break; | |
714 | case 'P': | |
715 | zo->zo_passtime = MAX(1, value); | |
716 | break; | |
717 | case 'F': | |
718 | zo->zo_maxloops = MAX(1, value); | |
719 | break; | |
720 | case 'B': | |
721 | (void) strlcpy(altdir, optarg, sizeof (altdir)); | |
722 | break; | |
723 | case 'h': | |
724 | usage(B_TRUE); | |
725 | break; | |
726 | case '?': | |
727 | default: | |
728 | usage(B_FALSE); | |
729 | break; | |
730 | } | |
731 | } | |
732 | ||
733 | zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1); | |
734 | ||
735 | zo->zo_vdevtime = | |
736 | (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs : | |
737 | UINT64_MAX >> 2); | |
738 | ||
739 | if (strlen(altdir) > 0) { | |
740 | char *cmd; | |
741 | char *realaltdir; | |
742 | char *bin; | |
743 | char *ztest; | |
744 | char *isa; | |
745 | int isalen; | |
746 | ||
747 | cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); | |
748 | realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); | |
749 | ||
750 | VERIFY(NULL != realpath(getexecname(), cmd)); | |
751 | if (0 != access(altdir, F_OK)) { | |
752 | ztest_dump_core = B_FALSE; | |
753 | fatal(B_TRUE, "invalid alternate ztest path: %s", | |
754 | altdir); | |
755 | } | |
756 | VERIFY(NULL != realpath(altdir, realaltdir)); | |
757 | ||
758 | /* | |
759 | * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest". | |
760 | * We want to extract <isa> to determine if we should use | |
761 | * 32 or 64 bit binaries. | |
762 | */ | |
763 | bin = strstr(cmd, "/usr/bin/"); | |
764 | ztest = strstr(bin, "/ztest"); | |
765 | isa = bin + 9; | |
766 | isalen = ztest - isa; | |
767 | (void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest), | |
768 | "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa); | |
769 | (void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath), | |
770 | "%s/usr/lib/%.*s", realaltdir, isalen, isa); | |
771 | ||
772 | if (0 != access(zo->zo_alt_ztest, X_OK)) { | |
773 | ztest_dump_core = B_FALSE; | |
774 | fatal(B_TRUE, "invalid alternate ztest: %s", | |
775 | zo->zo_alt_ztest); | |
776 | } else if (0 != access(zo->zo_alt_libpath, X_OK)) { | |
777 | ztest_dump_core = B_FALSE; | |
778 | fatal(B_TRUE, "invalid alternate lib directory %s", | |
779 | zo->zo_alt_libpath); | |
780 | } | |
781 | ||
782 | umem_free(cmd, MAXPATHLEN); | |
783 | umem_free(realaltdir, MAXPATHLEN); | |
784 | } | |
785 | } | |
786 | ||
787 | static void | |
788 | ztest_kill(ztest_shared_t *zs) | |
789 | { | |
790 | zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa)); | |
791 | zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa)); | |
792 | ||
793 | /* | |
794 | * Before we kill off ztest, make sure that the config is updated. | |
795 | * See comment above spa_config_sync(). | |
796 | */ | |
797 | mutex_enter(&spa_namespace_lock); | |
798 | spa_config_sync(ztest_spa, B_FALSE, B_FALSE); | |
799 | mutex_exit(&spa_namespace_lock); | |
800 | ||
801 | if (ztest_opts.zo_verbose >= 3) | |
802 | zfs_dbgmsg_print(FTAG); | |
803 | ||
804 | (void) kill(getpid(), SIGKILL); | |
805 | } | |
806 | ||
807 | static uint64_t | |
808 | ztest_random(uint64_t range) | |
809 | { | |
810 | uint64_t r; | |
811 | ||
812 | ASSERT3S(ztest_fd_rand, >=, 0); | |
813 | ||
814 | if (range == 0) | |
815 | return (0); | |
816 | ||
817 | if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r)) | |
818 | fatal(1, "short read from /dev/urandom"); | |
819 | ||
820 | return (r % range); | |
821 | } | |
822 | ||
823 | /* ARGSUSED */ | |
824 | static void | |
825 | ztest_record_enospc(const char *s) | |
826 | { | |
827 | ztest_shared->zs_enospc_count++; | |
828 | } | |
829 | ||
830 | static uint64_t | |
831 | ztest_get_ashift(void) | |
832 | { | |
833 | if (ztest_opts.zo_ashift == 0) | |
834 | return (SPA_MINBLOCKSHIFT + ztest_random(3)); | |
835 | return (ztest_opts.zo_ashift); | |
836 | } | |
837 | ||
838 | static nvlist_t * | |
839 | make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift) | |
840 | { | |
841 | char *pathbuf; | |
842 | uint64_t vdev; | |
843 | nvlist_t *file; | |
844 | ||
845 | pathbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); | |
846 | ||
847 | if (ashift == 0) | |
848 | ashift = ztest_get_ashift(); | |
849 | ||
850 | if (path == NULL) { | |
851 | path = pathbuf; | |
852 | ||
853 | if (aux != NULL) { | |
854 | vdev = ztest_shared->zs_vdev_aux; | |
855 | (void) snprintf(path, MAXPATHLEN, | |
856 | ztest_aux_template, ztest_opts.zo_dir, | |
857 | pool == NULL ? ztest_opts.zo_pool : pool, | |
858 | aux, vdev); | |
859 | } else { | |
860 | vdev = ztest_shared->zs_vdev_next_leaf++; | |
861 | (void) snprintf(path, MAXPATHLEN, | |
862 | ztest_dev_template, ztest_opts.zo_dir, | |
863 | pool == NULL ? ztest_opts.zo_pool : pool, vdev); | |
864 | } | |
865 | } | |
866 | ||
867 | if (size != 0) { | |
868 | int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666); | |
869 | if (fd == -1) | |
870 | fatal(1, "can't open %s", path); | |
871 | if (ftruncate(fd, size) != 0) | |
872 | fatal(1, "can't ftruncate %s", path); | |
873 | (void) close(fd); | |
874 | } | |
875 | ||
876 | VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0); | |
877 | VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0); | |
878 | VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0); | |
879 | VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0); | |
880 | umem_free(pathbuf, MAXPATHLEN); | |
881 | ||
882 | return (file); | |
883 | } | |
884 | ||
885 | static nvlist_t * | |
886 | make_vdev_raidz(char *path, char *aux, char *pool, size_t size, | |
887 | uint64_t ashift, int r) | |
888 | { | |
889 | nvlist_t *raidz, **child; | |
890 | int c; | |
891 | ||
892 | if (r < 2) | |
893 | return (make_vdev_file(path, aux, pool, size, ashift)); | |
894 | child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL); | |
895 | ||
896 | for (c = 0; c < r; c++) | |
897 | child[c] = make_vdev_file(path, aux, pool, size, ashift); | |
898 | ||
899 | VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0); | |
900 | VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE, | |
901 | VDEV_TYPE_RAIDZ) == 0); | |
902 | VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY, | |
903 | ztest_opts.zo_raidz_parity) == 0); | |
904 | VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN, | |
905 | child, r) == 0); | |
906 | ||
907 | for (c = 0; c < r; c++) | |
908 | nvlist_free(child[c]); | |
909 | ||
910 | umem_free(child, r * sizeof (nvlist_t *)); | |
911 | ||
912 | return (raidz); | |
913 | } | |
914 | ||
915 | static nvlist_t * | |
916 | make_vdev_mirror(char *path, char *aux, char *pool, size_t size, | |
917 | uint64_t ashift, int r, int m) | |
918 | { | |
919 | nvlist_t *mirror, **child; | |
920 | int c; | |
921 | ||
922 | if (m < 1) | |
923 | return (make_vdev_raidz(path, aux, pool, size, ashift, r)); | |
924 | ||
925 | child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL); | |
926 | ||
927 | for (c = 0; c < m; c++) | |
928 | child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r); | |
929 | ||
930 | VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0); | |
931 | VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE, | |
932 | VDEV_TYPE_MIRROR) == 0); | |
933 | VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN, | |
934 | child, m) == 0); | |
935 | ||
936 | for (c = 0; c < m; c++) | |
937 | nvlist_free(child[c]); | |
938 | ||
939 | umem_free(child, m * sizeof (nvlist_t *)); | |
940 | ||
941 | return (mirror); | |
942 | } | |
943 | ||
944 | static nvlist_t * | |
945 | make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift, | |
946 | int log, int r, int m, int t) | |
947 | { | |
948 | nvlist_t *root, **child; | |
949 | int c; | |
950 | ||
951 | ASSERT(t > 0); | |
952 | ||
953 | child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL); | |
954 | ||
955 | for (c = 0; c < t; c++) { | |
956 | child[c] = make_vdev_mirror(path, aux, pool, size, ashift, | |
957 | r, m); | |
958 | VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG, | |
959 | log) == 0); | |
960 | } | |
961 | ||
962 | VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0); | |
963 | VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0); | |
964 | VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN, | |
965 | child, t) == 0); | |
966 | ||
967 | for (c = 0; c < t; c++) | |
968 | nvlist_free(child[c]); | |
969 | ||
970 | umem_free(child, t * sizeof (nvlist_t *)); | |
971 | ||
972 | return (root); | |
973 | } | |
974 | ||
975 | /* | |
976 | * Find a random spa version. Returns back a random spa version in the | |
977 | * range [initial_version, SPA_VERSION_FEATURES]. | |
978 | */ | |
979 | static uint64_t | |
980 | ztest_random_spa_version(uint64_t initial_version) | |
981 | { | |
982 | uint64_t version = initial_version; | |
983 | ||
984 | if (version <= SPA_VERSION_BEFORE_FEATURES) { | |
985 | version = version + | |
986 | ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1); | |
987 | } | |
988 | ||
989 | if (version > SPA_VERSION_BEFORE_FEATURES) | |
990 | version = SPA_VERSION_FEATURES; | |
991 | ||
992 | ASSERT(SPA_VERSION_IS_SUPPORTED(version)); | |
993 | return (version); | |
994 | } | |
995 | ||
996 | static int | |
997 | ztest_random_blocksize(void) | |
998 | { | |
999 | return (1 << (SPA_MINBLOCKSHIFT + | |
1000 | ztest_random(SPA_MAXBLOCKSHIFT - SPA_MINBLOCKSHIFT + 1))); | |
1001 | } | |
1002 | ||
1003 | static int | |
1004 | ztest_random_ibshift(void) | |
1005 | { | |
1006 | return (DN_MIN_INDBLKSHIFT + | |
1007 | ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1)); | |
1008 | } | |
1009 | ||
1010 | static uint64_t | |
1011 | ztest_random_vdev_top(spa_t *spa, boolean_t log_ok) | |
1012 | { | |
1013 | uint64_t top; | |
1014 | vdev_t *rvd = spa->spa_root_vdev; | |
1015 | vdev_t *tvd; | |
1016 | ||
1017 | ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); | |
1018 | ||
1019 | do { | |
1020 | top = ztest_random(rvd->vdev_children); | |
1021 | tvd = rvd->vdev_child[top]; | |
1022 | } while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) || | |
1023 | tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL); | |
1024 | ||
1025 | return (top); | |
1026 | } | |
1027 | ||
1028 | static uint64_t | |
1029 | ztest_random_dsl_prop(zfs_prop_t prop) | |
1030 | { | |
1031 | uint64_t value; | |
1032 | ||
1033 | do { | |
1034 | value = zfs_prop_random_value(prop, ztest_random(-1ULL)); | |
1035 | } while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF); | |
1036 | ||
1037 | return (value); | |
1038 | } | |
1039 | ||
1040 | static int | |
1041 | ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value, | |
1042 | boolean_t inherit) | |
1043 | { | |
1044 | const char *propname = zfs_prop_to_name(prop); | |
1045 | const char *valname; | |
1046 | char *setpoint; | |
1047 | uint64_t curval; | |
1048 | int error; | |
1049 | ||
1050 | error = dsl_prop_set_int(osname, propname, | |
1051 | (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value); | |
1052 | ||
1053 | if (error == ENOSPC) { | |
1054 | ztest_record_enospc(FTAG); | |
1055 | return (error); | |
1056 | } | |
1057 | ASSERT0(error); | |
1058 | ||
1059 | setpoint = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); | |
1060 | VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint)); | |
1061 | ||
1062 | if (ztest_opts.zo_verbose >= 6) { | |
1063 | VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0); | |
1064 | (void) printf("%s %s = %s at '%s'\n", | |
1065 | osname, propname, valname, setpoint); | |
1066 | } | |
1067 | umem_free(setpoint, MAXPATHLEN); | |
1068 | ||
1069 | return (error); | |
1070 | } | |
1071 | ||
1072 | static int | |
1073 | ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value) | |
1074 | { | |
1075 | spa_t *spa = ztest_spa; | |
1076 | nvlist_t *props = NULL; | |
1077 | int error; | |
1078 | ||
1079 | VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0); | |
1080 | VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0); | |
1081 | ||
1082 | error = spa_prop_set(spa, props); | |
1083 | ||
1084 | nvlist_free(props); | |
1085 | ||
1086 | if (error == ENOSPC) { | |
1087 | ztest_record_enospc(FTAG); | |
1088 | return (error); | |
1089 | } | |
1090 | ASSERT0(error); | |
1091 | ||
1092 | return (error); | |
1093 | } | |
1094 | ||
1095 | static void | |
1096 | ztest_rll_init(rll_t *rll) | |
1097 | { | |
1098 | rll->rll_writer = NULL; | |
1099 | rll->rll_readers = 0; | |
1100 | mutex_init(&rll->rll_lock, NULL, MUTEX_DEFAULT, NULL); | |
1101 | cv_init(&rll->rll_cv, NULL, CV_DEFAULT, NULL); | |
1102 | } | |
1103 | ||
1104 | static void | |
1105 | ztest_rll_destroy(rll_t *rll) | |
1106 | { | |
1107 | ASSERT(rll->rll_writer == NULL); | |
1108 | ASSERT(rll->rll_readers == 0); | |
1109 | mutex_destroy(&rll->rll_lock); | |
1110 | cv_destroy(&rll->rll_cv); | |
1111 | } | |
1112 | ||
1113 | static void | |
1114 | ztest_rll_lock(rll_t *rll, rl_type_t type) | |
1115 | { | |
1116 | mutex_enter(&rll->rll_lock); | |
1117 | ||
1118 | if (type == RL_READER) { | |
1119 | while (rll->rll_writer != NULL) | |
1120 | (void) cv_wait(&rll->rll_cv, &rll->rll_lock); | |
1121 | rll->rll_readers++; | |
1122 | } else { | |
1123 | while (rll->rll_writer != NULL || rll->rll_readers) | |
1124 | (void) cv_wait(&rll->rll_cv, &rll->rll_lock); | |
1125 | rll->rll_writer = curthread; | |
1126 | } | |
1127 | ||
1128 | mutex_exit(&rll->rll_lock); | |
1129 | } | |
1130 | ||
1131 | static void | |
1132 | ztest_rll_unlock(rll_t *rll) | |
1133 | { | |
1134 | mutex_enter(&rll->rll_lock); | |
1135 | ||
1136 | if (rll->rll_writer) { | |
1137 | ASSERT(rll->rll_readers == 0); | |
1138 | rll->rll_writer = NULL; | |
1139 | } else { | |
1140 | ASSERT(rll->rll_readers != 0); | |
1141 | ASSERT(rll->rll_writer == NULL); | |
1142 | rll->rll_readers--; | |
1143 | } | |
1144 | ||
1145 | if (rll->rll_writer == NULL && rll->rll_readers == 0) | |
1146 | cv_broadcast(&rll->rll_cv); | |
1147 | ||
1148 | mutex_exit(&rll->rll_lock); | |
1149 | } | |
1150 | ||
1151 | static void | |
1152 | ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type) | |
1153 | { | |
1154 | rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)]; | |
1155 | ||
1156 | ztest_rll_lock(rll, type); | |
1157 | } | |
1158 | ||
1159 | static void | |
1160 | ztest_object_unlock(ztest_ds_t *zd, uint64_t object) | |
1161 | { | |
1162 | rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)]; | |
1163 | ||
1164 | ztest_rll_unlock(rll); | |
1165 | } | |
1166 | ||
1167 | static rl_t * | |
1168 | ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset, | |
1169 | uint64_t size, rl_type_t type) | |
1170 | { | |
1171 | uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1)); | |
1172 | rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)]; | |
1173 | rl_t *rl; | |
1174 | ||
1175 | rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL); | |
1176 | rl->rl_object = object; | |
1177 | rl->rl_offset = offset; | |
1178 | rl->rl_size = size; | |
1179 | rl->rl_lock = rll; | |
1180 | ||
1181 | ztest_rll_lock(rll, type); | |
1182 | ||
1183 | return (rl); | |
1184 | } | |
1185 | ||
1186 | static void | |
1187 | ztest_range_unlock(rl_t *rl) | |
1188 | { | |
1189 | rll_t *rll = rl->rl_lock; | |
1190 | ||
1191 | ztest_rll_unlock(rll); | |
1192 | ||
1193 | umem_free(rl, sizeof (*rl)); | |
1194 | } | |
1195 | ||
1196 | static void | |
1197 | ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os) | |
1198 | { | |
1199 | zd->zd_os = os; | |
1200 | zd->zd_zilog = dmu_objset_zil(os); | |
1201 | zd->zd_shared = szd; | |
1202 | dmu_objset_name(os, zd->zd_name); | |
1203 | int l; | |
1204 | ||
1205 | if (zd->zd_shared != NULL) | |
1206 | zd->zd_shared->zd_seq = 0; | |
1207 | ||
1208 | rw_init(&zd->zd_zilog_lock, NULL, RW_DEFAULT, NULL); | |
1209 | mutex_init(&zd->zd_dirobj_lock, NULL, MUTEX_DEFAULT, NULL); | |
1210 | ||
1211 | for (l = 0; l < ZTEST_OBJECT_LOCKS; l++) | |
1212 | ztest_rll_init(&zd->zd_object_lock[l]); | |
1213 | ||
1214 | for (l = 0; l < ZTEST_RANGE_LOCKS; l++) | |
1215 | ztest_rll_init(&zd->zd_range_lock[l]); | |
1216 | } | |
1217 | ||
1218 | static void | |
1219 | ztest_zd_fini(ztest_ds_t *zd) | |
1220 | { | |
1221 | int l; | |
1222 | ||
1223 | mutex_destroy(&zd->zd_dirobj_lock); | |
1224 | rw_destroy(&zd->zd_zilog_lock); | |
1225 | ||
1226 | for (l = 0; l < ZTEST_OBJECT_LOCKS; l++) | |
1227 | ztest_rll_destroy(&zd->zd_object_lock[l]); | |
1228 | ||
1229 | for (l = 0; l < ZTEST_RANGE_LOCKS; l++) | |
1230 | ztest_rll_destroy(&zd->zd_range_lock[l]); | |
1231 | } | |
1232 | ||
1233 | #define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT) | |
1234 | ||
1235 | static uint64_t | |
1236 | ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag) | |
1237 | { | |
1238 | uint64_t txg; | |
1239 | int error; | |
1240 | ||
1241 | /* | |
1242 | * Attempt to assign tx to some transaction group. | |
1243 | */ | |
1244 | error = dmu_tx_assign(tx, txg_how); | |
1245 | if (error) { | |
1246 | if (error == ERESTART) { | |
1247 | ASSERT(txg_how == TXG_NOWAIT); | |
1248 | dmu_tx_wait(tx); | |
1249 | } else { | |
1250 | ASSERT3U(error, ==, ENOSPC); | |
1251 | ztest_record_enospc(tag); | |
1252 | } | |
1253 | dmu_tx_abort(tx); | |
1254 | return (0); | |
1255 | } | |
1256 | txg = dmu_tx_get_txg(tx); | |
1257 | ASSERT(txg != 0); | |
1258 | return (txg); | |
1259 | } | |
1260 | ||
1261 | static void | |
1262 | ztest_pattern_set(void *buf, uint64_t size, uint64_t value) | |
1263 | { | |
1264 | uint64_t *ip = buf; | |
1265 | uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size); | |
1266 | ||
1267 | while (ip < ip_end) | |
1268 | *ip++ = value; | |
1269 | } | |
1270 | ||
1271 | #ifndef NDEBUG | |
1272 | static boolean_t | |
1273 | ztest_pattern_match(void *buf, uint64_t size, uint64_t value) | |
1274 | { | |
1275 | uint64_t *ip = buf; | |
1276 | uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size); | |
1277 | uint64_t diff = 0; | |
1278 | ||
1279 | while (ip < ip_end) | |
1280 | diff |= (value - *ip++); | |
1281 | ||
1282 | return (diff == 0); | |
1283 | } | |
1284 | #endif | |
1285 | ||
1286 | static void | |
1287 | ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object, | |
1288 | uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg) | |
1289 | { | |
1290 | bt->bt_magic = BT_MAGIC; | |
1291 | bt->bt_objset = dmu_objset_id(os); | |
1292 | bt->bt_object = object; | |
1293 | bt->bt_offset = offset; | |
1294 | bt->bt_gen = gen; | |
1295 | bt->bt_txg = txg; | |
1296 | bt->bt_crtxg = crtxg; | |
1297 | } | |
1298 | ||
1299 | static void | |
1300 | ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object, | |
1301 | uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg) | |
1302 | { | |
1303 | ASSERT(bt->bt_magic == BT_MAGIC); | |
1304 | ASSERT(bt->bt_objset == dmu_objset_id(os)); | |
1305 | ASSERT(bt->bt_object == object); | |
1306 | ASSERT(bt->bt_offset == offset); | |
1307 | ASSERT(bt->bt_gen <= gen); | |
1308 | ASSERT(bt->bt_txg <= txg); | |
1309 | ASSERT(bt->bt_crtxg == crtxg); | |
1310 | } | |
1311 | ||
1312 | static ztest_block_tag_t * | |
1313 | ztest_bt_bonus(dmu_buf_t *db) | |
1314 | { | |
1315 | dmu_object_info_t doi; | |
1316 | ztest_block_tag_t *bt; | |
1317 | ||
1318 | dmu_object_info_from_db(db, &doi); | |
1319 | ASSERT3U(doi.doi_bonus_size, <=, db->db_size); | |
1320 | ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt)); | |
1321 | bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt)); | |
1322 | ||
1323 | return (bt); | |
1324 | } | |
1325 | ||
1326 | /* | |
1327 | * ZIL logging ops | |
1328 | */ | |
1329 | ||
1330 | #define lrz_type lr_mode | |
1331 | #define lrz_blocksize lr_uid | |
1332 | #define lrz_ibshift lr_gid | |
1333 | #define lrz_bonustype lr_rdev | |
1334 | #define lrz_bonuslen lr_crtime[1] | |
1335 | ||
1336 | static void | |
1337 | ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr) | |
1338 | { | |
1339 | char *name = (void *)(lr + 1); /* name follows lr */ | |
1340 | size_t namesize = strlen(name) + 1; | |
1341 | itx_t *itx; | |
1342 | ||
1343 | if (zil_replaying(zd->zd_zilog, tx)) | |
1344 | return; | |
1345 | ||
1346 | itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize); | |
1347 | bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, | |
1348 | sizeof (*lr) + namesize - sizeof (lr_t)); | |
1349 | ||
1350 | zil_itx_assign(zd->zd_zilog, itx, tx); | |
1351 | } | |
1352 | ||
1353 | static void | |
1354 | ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object) | |
1355 | { | |
1356 | char *name = (void *)(lr + 1); /* name follows lr */ | |
1357 | size_t namesize = strlen(name) + 1; | |
1358 | itx_t *itx; | |
1359 | ||
1360 | if (zil_replaying(zd->zd_zilog, tx)) | |
1361 | return; | |
1362 | ||
1363 | itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize); | |
1364 | bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, | |
1365 | sizeof (*lr) + namesize - sizeof (lr_t)); | |
1366 | ||
1367 | itx->itx_oid = object; | |
1368 | zil_itx_assign(zd->zd_zilog, itx, tx); | |
1369 | } | |
1370 | ||
1371 | static void | |
1372 | ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr) | |
1373 | { | |
1374 | itx_t *itx; | |
1375 | itx_wr_state_t write_state = ztest_random(WR_NUM_STATES); | |
1376 | ||
1377 | if (zil_replaying(zd->zd_zilog, tx)) | |
1378 | return; | |
1379 | ||
1380 | if (lr->lr_length > ZIL_MAX_LOG_DATA) | |
1381 | write_state = WR_INDIRECT; | |
1382 | ||
1383 | itx = zil_itx_create(TX_WRITE, | |
1384 | sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0)); | |
1385 | ||
1386 | if (write_state == WR_COPIED && | |
1387 | dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length, | |
1388 | ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) { | |
1389 | zil_itx_destroy(itx); | |
1390 | itx = zil_itx_create(TX_WRITE, sizeof (*lr)); | |
1391 | write_state = WR_NEED_COPY; | |
1392 | } | |
1393 | itx->itx_private = zd; | |
1394 | itx->itx_wr_state = write_state; | |
1395 | itx->itx_sync = (ztest_random(8) == 0); | |
1396 | itx->itx_sod += (write_state == WR_NEED_COPY ? lr->lr_length : 0); | |
1397 | ||
1398 | bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, | |
1399 | sizeof (*lr) - sizeof (lr_t)); | |
1400 | ||
1401 | zil_itx_assign(zd->zd_zilog, itx, tx); | |
1402 | } | |
1403 | ||
1404 | static void | |
1405 | ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr) | |
1406 | { | |
1407 | itx_t *itx; | |
1408 | ||
1409 | if (zil_replaying(zd->zd_zilog, tx)) | |
1410 | return; | |
1411 | ||
1412 | itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr)); | |
1413 | bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, | |
1414 | sizeof (*lr) - sizeof (lr_t)); | |
1415 | ||
1416 | itx->itx_sync = B_FALSE; | |
1417 | zil_itx_assign(zd->zd_zilog, itx, tx); | |
1418 | } | |
1419 | ||
1420 | static void | |
1421 | ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr) | |
1422 | { | |
1423 | itx_t *itx; | |
1424 | ||
1425 | if (zil_replaying(zd->zd_zilog, tx)) | |
1426 | return; | |
1427 | ||
1428 | itx = zil_itx_create(TX_SETATTR, sizeof (*lr)); | |
1429 | bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, | |
1430 | sizeof (*lr) - sizeof (lr_t)); | |
1431 | ||
1432 | itx->itx_sync = B_FALSE; | |
1433 | zil_itx_assign(zd->zd_zilog, itx, tx); | |
1434 | } | |
1435 | ||
1436 | /* | |
1437 | * ZIL replay ops | |
1438 | */ | |
1439 | static int | |
1440 | ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap) | |
1441 | { | |
1442 | char *name = (void *)(lr + 1); /* name follows lr */ | |
1443 | objset_t *os = zd->zd_os; | |
1444 | ztest_block_tag_t *bbt; | |
1445 | dmu_buf_t *db; | |
1446 | dmu_tx_t *tx; | |
1447 | uint64_t txg; | |
1448 | int error = 0; | |
1449 | ||
1450 | if (byteswap) | |
1451 | byteswap_uint64_array(lr, sizeof (*lr)); | |
1452 | ||
1453 | ASSERT(lr->lr_doid == ZTEST_DIROBJ); | |
1454 | ASSERT(name[0] != '\0'); | |
1455 | ||
1456 | tx = dmu_tx_create(os); | |
1457 | ||
1458 | dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name); | |
1459 | ||
1460 | if (lr->lrz_type == DMU_OT_ZAP_OTHER) { | |
1461 | dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL); | |
1462 | } else { | |
1463 | dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); | |
1464 | } | |
1465 | ||
1466 | txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); | |
1467 | if (txg == 0) | |
1468 | return (ENOSPC); | |
1469 | ||
1470 | ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid); | |
1471 | ||
1472 | if (lr->lrz_type == DMU_OT_ZAP_OTHER) { | |
1473 | if (lr->lr_foid == 0) { | |
1474 | lr->lr_foid = zap_create(os, | |
1475 | lr->lrz_type, lr->lrz_bonustype, | |
1476 | lr->lrz_bonuslen, tx); | |
1477 | } else { | |
1478 | error = zap_create_claim(os, lr->lr_foid, | |
1479 | lr->lrz_type, lr->lrz_bonustype, | |
1480 | lr->lrz_bonuslen, tx); | |
1481 | } | |
1482 | } else { | |
1483 | if (lr->lr_foid == 0) { | |
1484 | lr->lr_foid = dmu_object_alloc(os, | |
1485 | lr->lrz_type, 0, lr->lrz_bonustype, | |
1486 | lr->lrz_bonuslen, tx); | |
1487 | } else { | |
1488 | error = dmu_object_claim(os, lr->lr_foid, | |
1489 | lr->lrz_type, 0, lr->lrz_bonustype, | |
1490 | lr->lrz_bonuslen, tx); | |
1491 | } | |
1492 | } | |
1493 | ||
1494 | if (error) { | |
1495 | ASSERT3U(error, ==, EEXIST); | |
1496 | ASSERT(zd->zd_zilog->zl_replay); | |
1497 | dmu_tx_commit(tx); | |
1498 | return (error); | |
1499 | } | |
1500 | ||
1501 | ASSERT(lr->lr_foid != 0); | |
1502 | ||
1503 | if (lr->lrz_type != DMU_OT_ZAP_OTHER) | |
1504 | VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid, | |
1505 | lr->lrz_blocksize, lr->lrz_ibshift, tx)); | |
1506 | ||
1507 | VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db)); | |
1508 | bbt = ztest_bt_bonus(db); | |
1509 | dmu_buf_will_dirty(db, tx); | |
1510 | ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg); | |
1511 | dmu_buf_rele(db, FTAG); | |
1512 | ||
1513 | VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1, | |
1514 | &lr->lr_foid, tx)); | |
1515 | ||
1516 | (void) ztest_log_create(zd, tx, lr); | |
1517 | ||
1518 | dmu_tx_commit(tx); | |
1519 | ||
1520 | return (0); | |
1521 | } | |
1522 | ||
1523 | static int | |
1524 | ztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap) | |
1525 | { | |
1526 | char *name = (void *)(lr + 1); /* name follows lr */ | |
1527 | objset_t *os = zd->zd_os; | |
1528 | dmu_object_info_t doi; | |
1529 | dmu_tx_t *tx; | |
1530 | uint64_t object, txg; | |
1531 | ||
1532 | if (byteswap) | |
1533 | byteswap_uint64_array(lr, sizeof (*lr)); | |
1534 | ||
1535 | ASSERT(lr->lr_doid == ZTEST_DIROBJ); | |
1536 | ASSERT(name[0] != '\0'); | |
1537 | ||
1538 | VERIFY3U(0, ==, | |
1539 | zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object)); | |
1540 | ASSERT(object != 0); | |
1541 | ||
1542 | ztest_object_lock(zd, object, RL_WRITER); | |
1543 | ||
1544 | VERIFY3U(0, ==, dmu_object_info(os, object, &doi)); | |
1545 | ||
1546 | tx = dmu_tx_create(os); | |
1547 | ||
1548 | dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name); | |
1549 | dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END); | |
1550 | ||
1551 | txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); | |
1552 | if (txg == 0) { | |
1553 | ztest_object_unlock(zd, object); | |
1554 | return (ENOSPC); | |
1555 | } | |
1556 | ||
1557 | if (doi.doi_type == DMU_OT_ZAP_OTHER) { | |
1558 | VERIFY3U(0, ==, zap_destroy(os, object, tx)); | |
1559 | } else { | |
1560 | VERIFY3U(0, ==, dmu_object_free(os, object, tx)); | |
1561 | } | |
1562 | ||
1563 | VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx)); | |
1564 | ||
1565 | (void) ztest_log_remove(zd, tx, lr, object); | |
1566 | ||
1567 | dmu_tx_commit(tx); | |
1568 | ||
1569 | ztest_object_unlock(zd, object); | |
1570 | ||
1571 | return (0); | |
1572 | } | |
1573 | ||
1574 | static int | |
1575 | ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap) | |
1576 | { | |
1577 | objset_t *os = zd->zd_os; | |
1578 | void *data = lr + 1; /* data follows lr */ | |
1579 | uint64_t offset, length; | |
1580 | ztest_block_tag_t *bt = data; | |
1581 | ztest_block_tag_t *bbt; | |
1582 | uint64_t gen, txg, lrtxg, crtxg; | |
1583 | dmu_object_info_t doi; | |
1584 | dmu_tx_t *tx; | |
1585 | dmu_buf_t *db; | |
1586 | arc_buf_t *abuf = NULL; | |
1587 | rl_t *rl; | |
1588 | ||
1589 | if (byteswap) | |
1590 | byteswap_uint64_array(lr, sizeof (*lr)); | |
1591 | ||
1592 | offset = lr->lr_offset; | |
1593 | length = lr->lr_length; | |
1594 | ||
1595 | /* If it's a dmu_sync() block, write the whole block */ | |
1596 | if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) { | |
1597 | uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr); | |
1598 | if (length < blocksize) { | |
1599 | offset -= offset % blocksize; | |
1600 | length = blocksize; | |
1601 | } | |
1602 | } | |
1603 | ||
1604 | if (bt->bt_magic == BSWAP_64(BT_MAGIC)) | |
1605 | byteswap_uint64_array(bt, sizeof (*bt)); | |
1606 | ||
1607 | if (bt->bt_magic != BT_MAGIC) | |
1608 | bt = NULL; | |
1609 | ||
1610 | ztest_object_lock(zd, lr->lr_foid, RL_READER); | |
1611 | rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER); | |
1612 | ||
1613 | VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db)); | |
1614 | ||
1615 | dmu_object_info_from_db(db, &doi); | |
1616 | ||
1617 | bbt = ztest_bt_bonus(db); | |
1618 | ASSERT3U(bbt->bt_magic, ==, BT_MAGIC); | |
1619 | gen = bbt->bt_gen; | |
1620 | crtxg = bbt->bt_crtxg; | |
1621 | lrtxg = lr->lr_common.lrc_txg; | |
1622 | ||
1623 | tx = dmu_tx_create(os); | |
1624 | ||
1625 | dmu_tx_hold_write(tx, lr->lr_foid, offset, length); | |
1626 | ||
1627 | if (ztest_random(8) == 0 && length == doi.doi_data_block_size && | |
1628 | P2PHASE(offset, length) == 0) | |
1629 | abuf = dmu_request_arcbuf(db, length); | |
1630 | ||
1631 | txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); | |
1632 | if (txg == 0) { | |
1633 | if (abuf != NULL) | |
1634 | dmu_return_arcbuf(abuf); | |
1635 | dmu_buf_rele(db, FTAG); | |
1636 | ztest_range_unlock(rl); | |
1637 | ztest_object_unlock(zd, lr->lr_foid); | |
1638 | return (ENOSPC); | |
1639 | } | |
1640 | ||
1641 | if (bt != NULL) { | |
1642 | /* | |
1643 | * Usually, verify the old data before writing new data -- | |
1644 | * but not always, because we also want to verify correct | |
1645 | * behavior when the data was not recently read into cache. | |
1646 | */ | |
1647 | ASSERT(offset % doi.doi_data_block_size == 0); | |
1648 | if (ztest_random(4) != 0) { | |
1649 | int prefetch = ztest_random(2) ? | |
1650 | DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH; | |
1651 | ztest_block_tag_t rbt; | |
1652 | ||
1653 | VERIFY(dmu_read(os, lr->lr_foid, offset, | |
1654 | sizeof (rbt), &rbt, prefetch) == 0); | |
1655 | if (rbt.bt_magic == BT_MAGIC) { | |
1656 | ztest_bt_verify(&rbt, os, lr->lr_foid, | |
1657 | offset, gen, txg, crtxg); | |
1658 | } | |
1659 | } | |
1660 | ||
1661 | /* | |
1662 | * Writes can appear to be newer than the bonus buffer because | |
1663 | * the ztest_get_data() callback does a dmu_read() of the | |
1664 | * open-context data, which may be different than the data | |
1665 | * as it was when the write was generated. | |
1666 | */ | |
1667 | if (zd->zd_zilog->zl_replay) { | |
1668 | ztest_bt_verify(bt, os, lr->lr_foid, offset, | |
1669 | MAX(gen, bt->bt_gen), MAX(txg, lrtxg), | |
1670 | bt->bt_crtxg); | |
1671 | } | |
1672 | ||
1673 | /* | |
1674 | * Set the bt's gen/txg to the bonus buffer's gen/txg | |
1675 | * so that all of the usual ASSERTs will work. | |
1676 | */ | |
1677 | ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg); | |
1678 | } | |
1679 | ||
1680 | if (abuf == NULL) { | |
1681 | dmu_write(os, lr->lr_foid, offset, length, data, tx); | |
1682 | } else { | |
1683 | bcopy(data, abuf->b_data, length); | |
1684 | dmu_assign_arcbuf(db, offset, abuf, tx); | |
1685 | } | |
1686 | ||
1687 | (void) ztest_log_write(zd, tx, lr); | |
1688 | ||
1689 | dmu_buf_rele(db, FTAG); | |
1690 | ||
1691 | dmu_tx_commit(tx); | |
1692 | ||
1693 | ztest_range_unlock(rl); | |
1694 | ztest_object_unlock(zd, lr->lr_foid); | |
1695 | ||
1696 | return (0); | |
1697 | } | |
1698 | ||
1699 | static int | |
1700 | ztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap) | |
1701 | { | |
1702 | objset_t *os = zd->zd_os; | |
1703 | dmu_tx_t *tx; | |
1704 | uint64_t txg; | |
1705 | rl_t *rl; | |
1706 | ||
1707 | if (byteswap) | |
1708 | byteswap_uint64_array(lr, sizeof (*lr)); | |
1709 | ||
1710 | ztest_object_lock(zd, lr->lr_foid, RL_READER); | |
1711 | rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length, | |
1712 | RL_WRITER); | |
1713 | ||
1714 | tx = dmu_tx_create(os); | |
1715 | ||
1716 | dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length); | |
1717 | ||
1718 | txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); | |
1719 | if (txg == 0) { | |
1720 | ztest_range_unlock(rl); | |
1721 | ztest_object_unlock(zd, lr->lr_foid); | |
1722 | return (ENOSPC); | |
1723 | } | |
1724 | ||
1725 | VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset, | |
1726 | lr->lr_length, tx) == 0); | |
1727 | ||
1728 | (void) ztest_log_truncate(zd, tx, lr); | |
1729 | ||
1730 | dmu_tx_commit(tx); | |
1731 | ||
1732 | ztest_range_unlock(rl); | |
1733 | ztest_object_unlock(zd, lr->lr_foid); | |
1734 | ||
1735 | return (0); | |
1736 | } | |
1737 | ||
1738 | static int | |
1739 | ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap) | |
1740 | { | |
1741 | objset_t *os = zd->zd_os; | |
1742 | dmu_tx_t *tx; | |
1743 | dmu_buf_t *db; | |
1744 | ztest_block_tag_t *bbt; | |
1745 | uint64_t txg, lrtxg, crtxg; | |
1746 | ||
1747 | if (byteswap) | |
1748 | byteswap_uint64_array(lr, sizeof (*lr)); | |
1749 | ||
1750 | ztest_object_lock(zd, lr->lr_foid, RL_WRITER); | |
1751 | ||
1752 | VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db)); | |
1753 | ||
1754 | tx = dmu_tx_create(os); | |
1755 | dmu_tx_hold_bonus(tx, lr->lr_foid); | |
1756 | ||
1757 | txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); | |
1758 | if (txg == 0) { | |
1759 | dmu_buf_rele(db, FTAG); | |
1760 | ztest_object_unlock(zd, lr->lr_foid); | |
1761 | return (ENOSPC); | |
1762 | } | |
1763 | ||
1764 | bbt = ztest_bt_bonus(db); | |
1765 | ASSERT3U(bbt->bt_magic, ==, BT_MAGIC); | |
1766 | crtxg = bbt->bt_crtxg; | |
1767 | lrtxg = lr->lr_common.lrc_txg; | |
1768 | ||
1769 | if (zd->zd_zilog->zl_replay) { | |
1770 | ASSERT(lr->lr_size != 0); | |
1771 | ASSERT(lr->lr_mode != 0); | |
1772 | ASSERT(lrtxg != 0); | |
1773 | } else { | |
1774 | /* | |
1775 | * Randomly change the size and increment the generation. | |
1776 | */ | |
1777 | lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) * | |
1778 | sizeof (*bbt); | |
1779 | lr->lr_mode = bbt->bt_gen + 1; | |
1780 | ASSERT(lrtxg == 0); | |
1781 | } | |
1782 | ||
1783 | /* | |
1784 | * Verify that the current bonus buffer is not newer than our txg. | |
1785 | */ | |
1786 | ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, | |
1787 | MAX(txg, lrtxg), crtxg); | |
1788 | ||
1789 | dmu_buf_will_dirty(db, tx); | |
1790 | ||
1791 | ASSERT3U(lr->lr_size, >=, sizeof (*bbt)); | |
1792 | ASSERT3U(lr->lr_size, <=, db->db_size); | |
1793 | VERIFY0(dmu_set_bonus(db, lr->lr_size, tx)); | |
1794 | bbt = ztest_bt_bonus(db); | |
1795 | ||
1796 | ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg); | |
1797 | ||
1798 | dmu_buf_rele(db, FTAG); | |
1799 | ||
1800 | (void) ztest_log_setattr(zd, tx, lr); | |
1801 | ||
1802 | dmu_tx_commit(tx); | |
1803 | ||
1804 | ztest_object_unlock(zd, lr->lr_foid); | |
1805 | ||
1806 | return (0); | |
1807 | } | |
1808 | ||
1809 | zil_replay_func_t ztest_replay_vector[TX_MAX_TYPE] = { | |
1810 | NULL, /* 0 no such transaction type */ | |
1811 | (zil_replay_func_t)ztest_replay_create, /* TX_CREATE */ | |
1812 | NULL, /* TX_MKDIR */ | |
1813 | NULL, /* TX_MKXATTR */ | |
1814 | NULL, /* TX_SYMLINK */ | |
1815 | (zil_replay_func_t)ztest_replay_remove, /* TX_REMOVE */ | |
1816 | NULL, /* TX_RMDIR */ | |
1817 | NULL, /* TX_LINK */ | |
1818 | NULL, /* TX_RENAME */ | |
1819 | (zil_replay_func_t)ztest_replay_write, /* TX_WRITE */ | |
1820 | (zil_replay_func_t)ztest_replay_truncate, /* TX_TRUNCATE */ | |
1821 | (zil_replay_func_t)ztest_replay_setattr, /* TX_SETATTR */ | |
1822 | NULL, /* TX_ACL */ | |
1823 | NULL, /* TX_CREATE_ACL */ | |
1824 | NULL, /* TX_CREATE_ATTR */ | |
1825 | NULL, /* TX_CREATE_ACL_ATTR */ | |
1826 | NULL, /* TX_MKDIR_ACL */ | |
1827 | NULL, /* TX_MKDIR_ATTR */ | |
1828 | NULL, /* TX_MKDIR_ACL_ATTR */ | |
1829 | NULL, /* TX_WRITE2 */ | |
1830 | }; | |
1831 | ||
1832 | /* | |
1833 | * ZIL get_data callbacks | |
1834 | */ | |
1835 | ||
1836 | static void | |
1837 | ztest_get_done(zgd_t *zgd, int error) | |
1838 | { | |
1839 | ztest_ds_t *zd = zgd->zgd_private; | |
1840 | uint64_t object = zgd->zgd_rl->rl_object; | |
1841 | ||
1842 | if (zgd->zgd_db) | |
1843 | dmu_buf_rele(zgd->zgd_db, zgd); | |
1844 | ||
1845 | ztest_range_unlock(zgd->zgd_rl); | |
1846 | ztest_object_unlock(zd, object); | |
1847 | ||
1848 | if (error == 0 && zgd->zgd_bp) | |
1849 | zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); | |
1850 | ||
1851 | umem_free(zgd, sizeof (*zgd)); | |
1852 | } | |
1853 | ||
1854 | static int | |
1855 | ztest_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) | |
1856 | { | |
1857 | ztest_ds_t *zd = arg; | |
1858 | objset_t *os = zd->zd_os; | |
1859 | uint64_t object = lr->lr_foid; | |
1860 | uint64_t offset = lr->lr_offset; | |
1861 | uint64_t size = lr->lr_length; | |
1862 | blkptr_t *bp = &lr->lr_blkptr; | |
1863 | uint64_t txg = lr->lr_common.lrc_txg; | |
1864 | uint64_t crtxg; | |
1865 | dmu_object_info_t doi; | |
1866 | dmu_buf_t *db; | |
1867 | zgd_t *zgd; | |
1868 | int error; | |
1869 | ||
1870 | ztest_object_lock(zd, object, RL_READER); | |
1871 | error = dmu_bonus_hold(os, object, FTAG, &db); | |
1872 | if (error) { | |
1873 | ztest_object_unlock(zd, object); | |
1874 | return (error); | |
1875 | } | |
1876 | ||
1877 | crtxg = ztest_bt_bonus(db)->bt_crtxg; | |
1878 | ||
1879 | if (crtxg == 0 || crtxg > txg) { | |
1880 | dmu_buf_rele(db, FTAG); | |
1881 | ztest_object_unlock(zd, object); | |
1882 | return (ENOENT); | |
1883 | } | |
1884 | ||
1885 | dmu_object_info_from_db(db, &doi); | |
1886 | dmu_buf_rele(db, FTAG); | |
1887 | db = NULL; | |
1888 | ||
1889 | zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL); | |
1890 | zgd->zgd_zilog = zd->zd_zilog; | |
1891 | zgd->zgd_private = zd; | |
1892 | ||
1893 | if (buf != NULL) { /* immediate write */ | |
1894 | zgd->zgd_rl = ztest_range_lock(zd, object, offset, size, | |
1895 | RL_READER); | |
1896 | ||
1897 | error = dmu_read(os, object, offset, size, buf, | |
1898 | DMU_READ_NO_PREFETCH); | |
1899 | ASSERT(error == 0); | |
1900 | } else { | |
1901 | size = doi.doi_data_block_size; | |
1902 | if (ISP2(size)) { | |
1903 | offset = P2ALIGN(offset, size); | |
1904 | } else { | |
1905 | ASSERT(offset < size); | |
1906 | offset = 0; | |
1907 | } | |
1908 | ||
1909 | zgd->zgd_rl = ztest_range_lock(zd, object, offset, size, | |
1910 | RL_READER); | |
1911 | ||
1912 | error = dmu_buf_hold(os, object, offset, zgd, &db, | |
1913 | DMU_READ_NO_PREFETCH); | |
1914 | ||
1915 | if (error == 0) { | |
1916 | blkptr_t *obp = dmu_buf_get_blkptr(db); | |
1917 | if (obp) { | |
1918 | ASSERT(BP_IS_HOLE(bp)); | |
1919 | *bp = *obp; | |
1920 | } | |
1921 | ||
1922 | zgd->zgd_db = db; | |
1923 | zgd->zgd_bp = bp; | |
1924 | ||
1925 | ASSERT(db->db_offset == offset); | |
1926 | ASSERT(db->db_size == size); | |
1927 | ||
1928 | error = dmu_sync(zio, lr->lr_common.lrc_txg, | |
1929 | ztest_get_done, zgd); | |
1930 | ||
1931 | if (error == 0) | |
1932 | return (0); | |
1933 | } | |
1934 | } | |
1935 | ||
1936 | ztest_get_done(zgd, error); | |
1937 | ||
1938 | return (error); | |
1939 | } | |
1940 | ||
1941 | static void * | |
1942 | ztest_lr_alloc(size_t lrsize, char *name) | |
1943 | { | |
1944 | char *lr; | |
1945 | size_t namesize = name ? strlen(name) + 1 : 0; | |
1946 | ||
1947 | lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL); | |
1948 | ||
1949 | if (name) | |
1950 | bcopy(name, lr + lrsize, namesize); | |
1951 | ||
1952 | return (lr); | |
1953 | } | |
1954 | ||
1955 | void | |
1956 | ztest_lr_free(void *lr, size_t lrsize, char *name) | |
1957 | { | |
1958 | size_t namesize = name ? strlen(name) + 1 : 0; | |
1959 | ||
1960 | umem_free(lr, lrsize + namesize); | |
1961 | } | |
1962 | ||
1963 | /* | |
1964 | * Lookup a bunch of objects. Returns the number of objects not found. | |
1965 | */ | |
1966 | static int | |
1967 | ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count) | |
1968 | { | |
1969 | int missing = 0; | |
1970 | int error; | |
1971 | int i; | |
1972 | ||
1973 | ASSERT(mutex_held(&zd->zd_dirobj_lock)); | |
1974 | ||
1975 | for (i = 0; i < count; i++, od++) { | |
1976 | od->od_object = 0; | |
1977 | error = zap_lookup(zd->zd_os, od->od_dir, od->od_name, | |
1978 | sizeof (uint64_t), 1, &od->od_object); | |
1979 | if (error) { | |
1980 | ASSERT(error == ENOENT); | |
1981 | ASSERT(od->od_object == 0); | |
1982 | missing++; | |
1983 | } else { | |
1984 | dmu_buf_t *db; | |
1985 | ztest_block_tag_t *bbt; | |
1986 | dmu_object_info_t doi; | |
1987 | ||
1988 | ASSERT(od->od_object != 0); | |
1989 | ASSERT(missing == 0); /* there should be no gaps */ | |
1990 | ||
1991 | ztest_object_lock(zd, od->od_object, RL_READER); | |
1992 | VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os, | |
1993 | od->od_object, FTAG, &db)); | |
1994 | dmu_object_info_from_db(db, &doi); | |
1995 | bbt = ztest_bt_bonus(db); | |
1996 | ASSERT3U(bbt->bt_magic, ==, BT_MAGIC); | |
1997 | od->od_type = doi.doi_type; | |
1998 | od->od_blocksize = doi.doi_data_block_size; | |
1999 | od->od_gen = bbt->bt_gen; | |
2000 | dmu_buf_rele(db, FTAG); | |
2001 | ztest_object_unlock(zd, od->od_object); | |
2002 | } | |
2003 | } | |
2004 | ||
2005 | return (missing); | |
2006 | } | |
2007 | ||
2008 | static int | |
2009 | ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count) | |
2010 | { | |
2011 | int missing = 0; | |
2012 | int i; | |
2013 | ||
2014 | ASSERT(mutex_held(&zd->zd_dirobj_lock)); | |
2015 | ||
2016 | for (i = 0; i < count; i++, od++) { | |
2017 | if (missing) { | |
2018 | od->od_object = 0; | |
2019 | missing++; | |
2020 | continue; | |
2021 | } | |
2022 | ||
2023 | lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name); | |
2024 | ||
2025 | lr->lr_doid = od->od_dir; | |
2026 | lr->lr_foid = 0; /* 0 to allocate, > 0 to claim */ | |
2027 | lr->lrz_type = od->od_crtype; | |
2028 | lr->lrz_blocksize = od->od_crblocksize; | |
2029 | lr->lrz_ibshift = ztest_random_ibshift(); | |
2030 | lr->lrz_bonustype = DMU_OT_UINT64_OTHER; | |
2031 | lr->lrz_bonuslen = dmu_bonus_max(); | |
2032 | lr->lr_gen = od->od_crgen; | |
2033 | lr->lr_crtime[0] = time(NULL); | |
2034 | ||
2035 | if (ztest_replay_create(zd, lr, B_FALSE) != 0) { | |
2036 | ASSERT(missing == 0); | |
2037 | od->od_object = 0; | |
2038 | missing++; | |
2039 | } else { | |
2040 | od->od_object = lr->lr_foid; | |
2041 | od->od_type = od->od_crtype; | |
2042 | od->od_blocksize = od->od_crblocksize; | |
2043 | od->od_gen = od->od_crgen; | |
2044 | ASSERT(od->od_object != 0); | |
2045 | } | |
2046 | ||
2047 | ztest_lr_free(lr, sizeof (*lr), od->od_name); | |
2048 | } | |
2049 | ||
2050 | return (missing); | |
2051 | } | |
2052 | ||
2053 | static int | |
2054 | ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count) | |
2055 | { | |
2056 | int missing = 0; | |
2057 | int error; | |
2058 | int i; | |
2059 | ||
2060 | ASSERT(mutex_held(&zd->zd_dirobj_lock)); | |
2061 | ||
2062 | od += count - 1; | |
2063 | ||
2064 | for (i = count - 1; i >= 0; i--, od--) { | |
2065 | if (missing) { | |
2066 | missing++; | |
2067 | continue; | |
2068 | } | |
2069 | ||
2070 | /* | |
2071 | * No object was found. | |
2072 | */ | |
2073 | if (od->od_object == 0) | |
2074 | continue; | |
2075 | ||
2076 | lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name); | |
2077 | ||
2078 | lr->lr_doid = od->od_dir; | |
2079 | ||
2080 | if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) { | |
2081 | ASSERT3U(error, ==, ENOSPC); | |
2082 | missing++; | |
2083 | } else { | |
2084 | od->od_object = 0; | |
2085 | } | |
2086 | ztest_lr_free(lr, sizeof (*lr), od->od_name); | |
2087 | } | |
2088 | ||
2089 | return (missing); | |
2090 | } | |
2091 | ||
2092 | static int | |
2093 | ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size, | |
2094 | void *data) | |
2095 | { | |
2096 | lr_write_t *lr; | |
2097 | int error; | |
2098 | ||
2099 | lr = ztest_lr_alloc(sizeof (*lr) + size, NULL); | |
2100 | ||
2101 | lr->lr_foid = object; | |
2102 | lr->lr_offset = offset; | |
2103 | lr->lr_length = size; | |
2104 | lr->lr_blkoff = 0; | |
2105 | BP_ZERO(&lr->lr_blkptr); | |
2106 | ||
2107 | bcopy(data, lr + 1, size); | |
2108 | ||
2109 | error = ztest_replay_write(zd, lr, B_FALSE); | |
2110 | ||
2111 | ztest_lr_free(lr, sizeof (*lr) + size, NULL); | |
2112 | ||
2113 | return (error); | |
2114 | } | |
2115 | ||
2116 | static int | |
2117 | ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size) | |
2118 | { | |
2119 | lr_truncate_t *lr; | |
2120 | int error; | |
2121 | ||
2122 | lr = ztest_lr_alloc(sizeof (*lr), NULL); | |
2123 | ||
2124 | lr->lr_foid = object; | |
2125 | lr->lr_offset = offset; | |
2126 | lr->lr_length = size; | |
2127 | ||
2128 | error = ztest_replay_truncate(zd, lr, B_FALSE); | |
2129 | ||
2130 | ztest_lr_free(lr, sizeof (*lr), NULL); | |
2131 | ||
2132 | return (error); | |
2133 | } | |
2134 | ||
2135 | static int | |
2136 | ztest_setattr(ztest_ds_t *zd, uint64_t object) | |
2137 | { | |
2138 | lr_setattr_t *lr; | |
2139 | int error; | |
2140 | ||
2141 | lr = ztest_lr_alloc(sizeof (*lr), NULL); | |
2142 | ||
2143 | lr->lr_foid = object; | |
2144 | lr->lr_size = 0; | |
2145 | lr->lr_mode = 0; | |
2146 | ||
2147 | error = ztest_replay_setattr(zd, lr, B_FALSE); | |
2148 | ||
2149 | ztest_lr_free(lr, sizeof (*lr), NULL); | |
2150 | ||
2151 | return (error); | |
2152 | } | |
2153 | ||
2154 | static void | |
2155 | ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size) | |
2156 | { | |
2157 | objset_t *os = zd->zd_os; | |
2158 | dmu_tx_t *tx; | |
2159 | uint64_t txg; | |
2160 | rl_t *rl; | |
2161 | ||
2162 | txg_wait_synced(dmu_objset_pool(os), 0); | |
2163 | ||
2164 | ztest_object_lock(zd, object, RL_READER); | |
2165 | rl = ztest_range_lock(zd, object, offset, size, RL_WRITER); | |
2166 | ||
2167 | tx = dmu_tx_create(os); | |
2168 | ||
2169 | dmu_tx_hold_write(tx, object, offset, size); | |
2170 | ||
2171 | txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); | |
2172 | ||
2173 | if (txg != 0) { | |
2174 | dmu_prealloc(os, object, offset, size, tx); | |
2175 | dmu_tx_commit(tx); | |
2176 | txg_wait_synced(dmu_objset_pool(os), txg); | |
2177 | } else { | |
2178 | (void) dmu_free_long_range(os, object, offset, size); | |
2179 | } | |
2180 | ||
2181 | ztest_range_unlock(rl); | |
2182 | ztest_object_unlock(zd, object); | |
2183 | } | |
2184 | ||
2185 | static void | |
2186 | ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset) | |
2187 | { | |
2188 | int err; | |
2189 | ztest_block_tag_t wbt; | |
2190 | dmu_object_info_t doi; | |
2191 | enum ztest_io_type io_type; | |
2192 | uint64_t blocksize; | |
2193 | void *data; | |
2194 | ||
2195 | VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0); | |
2196 | blocksize = doi.doi_data_block_size; | |
2197 | data = umem_alloc(blocksize, UMEM_NOFAIL); | |
2198 | ||
2199 | /* | |
2200 | * Pick an i/o type at random, biased toward writing block tags. | |
2201 | */ | |
2202 | io_type = ztest_random(ZTEST_IO_TYPES); | |
2203 | if (ztest_random(2) == 0) | |
2204 | io_type = ZTEST_IO_WRITE_TAG; | |
2205 | ||
2206 | (void) rw_enter(&zd->zd_zilog_lock, RW_READER); | |
2207 | ||
2208 | switch (io_type) { | |
2209 | ||
2210 | case ZTEST_IO_WRITE_TAG: | |
2211 | ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0); | |
2212 | (void) ztest_write(zd, object, offset, sizeof (wbt), &wbt); | |
2213 | break; | |
2214 | ||
2215 | case ZTEST_IO_WRITE_PATTERN: | |
2216 | (void) memset(data, 'a' + (object + offset) % 5, blocksize); | |
2217 | if (ztest_random(2) == 0) { | |
2218 | /* | |
2219 | * Induce fletcher2 collisions to ensure that | |
2220 | * zio_ddt_collision() detects and resolves them | |
2221 | * when using fletcher2-verify for deduplication. | |
2222 | */ | |
2223 | ((uint64_t *)data)[0] ^= 1ULL << 63; | |
2224 | ((uint64_t *)data)[4] ^= 1ULL << 63; | |
2225 | } | |
2226 | (void) ztest_write(zd, object, offset, blocksize, data); | |
2227 | break; | |
2228 | ||
2229 | case ZTEST_IO_WRITE_ZEROES: | |
2230 | bzero(data, blocksize); | |
2231 | (void) ztest_write(zd, object, offset, blocksize, data); | |
2232 | break; | |
2233 | ||
2234 | case ZTEST_IO_TRUNCATE: | |
2235 | (void) ztest_truncate(zd, object, offset, blocksize); | |
2236 | break; | |
2237 | ||
2238 | case ZTEST_IO_SETATTR: | |
2239 | (void) ztest_setattr(zd, object); | |
2240 | break; | |
2241 | default: | |
2242 | break; | |
2243 | ||
2244 | case ZTEST_IO_REWRITE: | |
2245 | (void) rw_enter(&ztest_name_lock, RW_READER); | |
2246 | err = ztest_dsl_prop_set_uint64(zd->zd_name, | |
2247 | ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa), | |
2248 | B_FALSE); | |
2249 | VERIFY(err == 0 || err == ENOSPC); | |
2250 | err = ztest_dsl_prop_set_uint64(zd->zd_name, | |
2251 | ZFS_PROP_COMPRESSION, | |
2252 | ztest_random_dsl_prop(ZFS_PROP_COMPRESSION), | |
2253 | B_FALSE); | |
2254 | VERIFY(err == 0 || err == ENOSPC); | |
2255 | (void) rw_exit(&ztest_name_lock); | |
2256 | ||
2257 | VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data, | |
2258 | DMU_READ_NO_PREFETCH)); | |
2259 | ||
2260 | (void) ztest_write(zd, object, offset, blocksize, data); | |
2261 | break; | |
2262 | } | |
2263 | ||
2264 | (void) rw_exit(&zd->zd_zilog_lock); | |
2265 | ||
2266 | umem_free(data, blocksize); | |
2267 | } | |
2268 | ||
2269 | /* | |
2270 | * Initialize an object description template. | |
2271 | */ | |
2272 | static void | |
2273 | ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index, | |
2274 | dmu_object_type_t type, uint64_t blocksize, uint64_t gen) | |
2275 | { | |
2276 | od->od_dir = ZTEST_DIROBJ; | |
2277 | od->od_object = 0; | |
2278 | ||
2279 | od->od_crtype = type; | |
2280 | od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize(); | |
2281 | od->od_crgen = gen; | |
2282 | ||
2283 | od->od_type = DMU_OT_NONE; | |
2284 | od->od_blocksize = 0; | |
2285 | od->od_gen = 0; | |
2286 | ||
2287 | (void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]", | |
2288 | tag, (longlong_t)id, (u_longlong_t)index); | |
2289 | } | |
2290 | ||
2291 | /* | |
2292 | * Lookup or create the objects for a test using the od template. | |
2293 | * If the objects do not all exist, or if 'remove' is specified, | |
2294 | * remove any existing objects and create new ones. Otherwise, | |
2295 | * use the existing objects. | |
2296 | */ | |
2297 | static int | |
2298 | ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove) | |
2299 | { | |
2300 | int count = size / sizeof (*od); | |
2301 | int rv = 0; | |
2302 | ||
2303 | mutex_enter(&zd->zd_dirobj_lock); | |
2304 | if ((ztest_lookup(zd, od, count) != 0 || remove) && | |
2305 | (ztest_remove(zd, od, count) != 0 || | |
2306 | ztest_create(zd, od, count) != 0)) | |
2307 | rv = -1; | |
2308 | zd->zd_od = od; | |
2309 | mutex_exit(&zd->zd_dirobj_lock); | |
2310 | ||
2311 | return (rv); | |
2312 | } | |
2313 | ||
2314 | /* ARGSUSED */ | |
2315 | void | |
2316 | ztest_zil_commit(ztest_ds_t *zd, uint64_t id) | |
2317 | { | |
2318 | zilog_t *zilog = zd->zd_zilog; | |
2319 | ||
2320 | (void) rw_enter(&zd->zd_zilog_lock, RW_READER); | |
2321 | ||
2322 | zil_commit(zilog, ztest_random(ZTEST_OBJECTS)); | |
2323 | ||
2324 | /* | |
2325 | * Remember the committed values in zd, which is in parent/child | |
2326 | * shared memory. If we die, the next iteration of ztest_run() | |
2327 | * will verify that the log really does contain this record. | |
2328 | */ | |
2329 | mutex_enter(&zilog->zl_lock); | |
2330 | ASSERT(zd->zd_shared != NULL); | |
2331 | ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq); | |
2332 | zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq; | |
2333 | mutex_exit(&zilog->zl_lock); | |
2334 | ||
2335 | (void) rw_exit(&zd->zd_zilog_lock); | |
2336 | } | |
2337 | ||
2338 | /* | |
2339 | * This function is designed to simulate the operations that occur during a | |
2340 | * mount/unmount operation. We hold the dataset across these operations in an | |
2341 | * attempt to expose any implicit assumptions about ZIL management. | |
2342 | */ | |
2343 | /* ARGSUSED */ | |
2344 | void | |
2345 | ztest_zil_remount(ztest_ds_t *zd, uint64_t id) | |
2346 | { | |
2347 | objset_t *os = zd->zd_os; | |
2348 | ||
2349 | /* | |
2350 | * We grab the zd_dirobj_lock to ensure that no other thread is | |
2351 | * updating the zil (i.e. adding in-memory log records) and the | |
2352 | * zd_zilog_lock to block any I/O. | |
2353 | */ | |
2354 | mutex_enter(&zd->zd_dirobj_lock); | |
2355 | (void) rw_enter(&zd->zd_zilog_lock, RW_WRITER); | |
2356 | ||
2357 | /* zfs_sb_teardown() */ | |
2358 | zil_close(zd->zd_zilog); | |
2359 | ||
2360 | /* zfsvfs_setup() */ | |
2361 | VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog); | |
2362 | zil_replay(os, zd, ztest_replay_vector); | |
2363 | ||
2364 | (void) rw_exit(&zd->zd_zilog_lock); | |
2365 | mutex_exit(&zd->zd_dirobj_lock); | |
2366 | } | |
2367 | ||
2368 | /* | |
2369 | * Verify that we can't destroy an active pool, create an existing pool, | |
2370 | * or create a pool with a bad vdev spec. | |
2371 | */ | |
2372 | /* ARGSUSED */ | |
2373 | void | |
2374 | ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id) | |
2375 | { | |
2376 | ztest_shared_opts_t *zo = &ztest_opts; | |
2377 | spa_t *spa; | |
2378 | nvlist_t *nvroot; | |
2379 | ||
2380 | /* | |
2381 | * Attempt to create using a bad file. | |
2382 | */ | |
2383 | nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1); | |
2384 | VERIFY3U(ENOENT, ==, | |
2385 | spa_create("ztest_bad_file", nvroot, NULL, NULL)); | |
2386 | nvlist_free(nvroot); | |
2387 | ||
2388 | /* | |
2389 | * Attempt to create using a bad mirror. | |
2390 | */ | |
2391 | nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1); | |
2392 | VERIFY3U(ENOENT, ==, | |
2393 | spa_create("ztest_bad_mirror", nvroot, NULL, NULL)); | |
2394 | nvlist_free(nvroot); | |
2395 | ||
2396 | /* | |
2397 | * Attempt to create an existing pool. It shouldn't matter | |
2398 | * what's in the nvroot; we should fail with EEXIST. | |
2399 | */ | |
2400 | (void) rw_enter(&ztest_name_lock, RW_READER); | |
2401 | nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1); | |
2402 | VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL)); | |
2403 | nvlist_free(nvroot); | |
2404 | VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG)); | |
2405 | VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool)); | |
2406 | spa_close(spa, FTAG); | |
2407 | ||
2408 | (void) rw_exit(&ztest_name_lock); | |
2409 | } | |
2410 | ||
2411 | /* ARGSUSED */ | |
2412 | void | |
2413 | ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id) | |
2414 | { | |
2415 | spa_t *spa; | |
2416 | uint64_t initial_version = SPA_VERSION_INITIAL; | |
2417 | uint64_t version, newversion; | |
2418 | nvlist_t *nvroot, *props; | |
2419 | char *name; | |
2420 | ||
2421 | mutex_enter(&ztest_vdev_lock); | |
2422 | name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool); | |
2423 | ||
2424 | /* | |
2425 | * Clean up from previous runs. | |
2426 | */ | |
2427 | (void) spa_destroy(name); | |
2428 | ||
2429 | nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0, | |
2430 | 0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1); | |
2431 | ||
2432 | /* | |
2433 | * If we're configuring a RAIDZ device then make sure that the | |
2434 | * the initial version is capable of supporting that feature. | |
2435 | */ | |
2436 | switch (ztest_opts.zo_raidz_parity) { | |
2437 | case 0: | |
2438 | case 1: | |
2439 | initial_version = SPA_VERSION_INITIAL; | |
2440 | break; | |
2441 | case 2: | |
2442 | initial_version = SPA_VERSION_RAIDZ2; | |
2443 | break; | |
2444 | case 3: | |
2445 | initial_version = SPA_VERSION_RAIDZ3; | |
2446 | break; | |
2447 | } | |
2448 | ||
2449 | /* | |
2450 | * Create a pool with a spa version that can be upgraded. Pick | |
2451 | * a value between initial_version and SPA_VERSION_BEFORE_FEATURES. | |
2452 | */ | |
2453 | do { | |
2454 | version = ztest_random_spa_version(initial_version); | |
2455 | } while (version > SPA_VERSION_BEFORE_FEATURES); | |
2456 | ||
2457 | props = fnvlist_alloc(); | |
2458 | fnvlist_add_uint64(props, | |
2459 | zpool_prop_to_name(ZPOOL_PROP_VERSION), version); | |
2460 | VERIFY3S(spa_create(name, nvroot, props, NULL), ==, 0); | |
2461 | fnvlist_free(nvroot); | |
2462 | fnvlist_free(props); | |
2463 | ||
2464 | VERIFY3S(spa_open(name, &spa, FTAG), ==, 0); | |
2465 | VERIFY3U(spa_version(spa), ==, version); | |
2466 | newversion = ztest_random_spa_version(version + 1); | |
2467 | ||
2468 | if (ztest_opts.zo_verbose >= 4) { | |
2469 | (void) printf("upgrading spa version from %llu to %llu\n", | |
2470 | (u_longlong_t)version, (u_longlong_t)newversion); | |
2471 | } | |
2472 | ||
2473 | spa_upgrade(spa, newversion); | |
2474 | VERIFY3U(spa_version(spa), >, version); | |
2475 | VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config, | |
2476 | zpool_prop_to_name(ZPOOL_PROP_VERSION))); | |
2477 | spa_close(spa, FTAG); | |
2478 | ||
2479 | strfree(name); | |
2480 | mutex_exit(&ztest_vdev_lock); | |
2481 | } | |
2482 | ||
2483 | static vdev_t * | |
2484 | vdev_lookup_by_path(vdev_t *vd, const char *path) | |
2485 | { | |
2486 | vdev_t *mvd; | |
2487 | int c; | |
2488 | ||
2489 | if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0) | |
2490 | return (vd); | |
2491 | ||
2492 | for (c = 0; c < vd->vdev_children; c++) | |
2493 | if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) != | |
2494 | NULL) | |
2495 | return (mvd); | |
2496 | ||
2497 | return (NULL); | |
2498 | } | |
2499 | ||
2500 | /* | |
2501 | * Find the first available hole which can be used as a top-level. | |
2502 | */ | |
2503 | int | |
2504 | find_vdev_hole(spa_t *spa) | |
2505 | { | |
2506 | vdev_t *rvd = spa->spa_root_vdev; | |
2507 | int c; | |
2508 | ||
2509 | ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV); | |
2510 | ||
2511 | for (c = 0; c < rvd->vdev_children; c++) { | |
2512 | vdev_t *cvd = rvd->vdev_child[c]; | |
2513 | ||
2514 | if (cvd->vdev_ishole) | |
2515 | break; | |
2516 | } | |
2517 | return (c); | |
2518 | } | |
2519 | ||
2520 | /* | |
2521 | * Verify that vdev_add() works as expected. | |
2522 | */ | |
2523 | /* ARGSUSED */ | |
2524 | void | |
2525 | ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id) | |
2526 | { | |
2527 | ztest_shared_t *zs = ztest_shared; | |
2528 | spa_t *spa = ztest_spa; | |
2529 | uint64_t leaves; | |
2530 | uint64_t guid; | |
2531 | nvlist_t *nvroot; | |
2532 | int error; | |
2533 | ||
2534 | mutex_enter(&ztest_vdev_lock); | |
2535 | leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz; | |
2536 | ||
2537 | spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); | |
2538 | ||
2539 | ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves; | |
2540 | ||
2541 | /* | |
2542 | * If we have slogs then remove them 1/4 of the time. | |
2543 | */ | |
2544 | if (spa_has_slogs(spa) && ztest_random(4) == 0) { | |
2545 | /* | |
2546 | * Grab the guid from the head of the log class rotor. | |
2547 | */ | |
2548 | guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid; | |
2549 | ||
2550 | spa_config_exit(spa, SCL_VDEV, FTAG); | |
2551 | ||
2552 | /* | |
2553 | * We have to grab the zs_name_lock as writer to | |
2554 | * prevent a race between removing a slog (dmu_objset_find) | |
2555 | * and destroying a dataset. Removing the slog will | |
2556 | * grab a reference on the dataset which may cause | |
2557 | * dsl_destroy_head() to fail with EBUSY thus | |
2558 | * leaving the dataset in an inconsistent state. | |
2559 | */ | |
2560 | rw_enter(&ztest_name_lock, RW_WRITER); | |
2561 | error = spa_vdev_remove(spa, guid, B_FALSE); | |
2562 | rw_exit(&ztest_name_lock); | |
2563 | ||
2564 | if (error && error != EEXIST) | |
2565 | fatal(0, "spa_vdev_remove() = %d", error); | |
2566 | } else { | |
2567 | spa_config_exit(spa, SCL_VDEV, FTAG); | |
2568 | ||
2569 | /* | |
2570 | * Make 1/4 of the devices be log devices. | |
2571 | */ | |
2572 | nvroot = make_vdev_root(NULL, NULL, NULL, | |
2573 | ztest_opts.zo_vdev_size, 0, | |
2574 | ztest_random(4) == 0, ztest_opts.zo_raidz, | |
2575 | zs->zs_mirrors, 1); | |
2576 | ||
2577 | error = spa_vdev_add(spa, nvroot); | |
2578 | nvlist_free(nvroot); | |
2579 | ||
2580 | if (error == ENOSPC) | |
2581 | ztest_record_enospc("spa_vdev_add"); | |
2582 | else if (error != 0) | |
2583 | fatal(0, "spa_vdev_add() = %d", error); | |
2584 | } | |
2585 | ||
2586 | mutex_exit(&ztest_vdev_lock); | |
2587 | } | |
2588 | ||
2589 | /* | |
2590 | * Verify that adding/removing aux devices (l2arc, hot spare) works as expected. | |
2591 | */ | |
2592 | /* ARGSUSED */ | |
2593 | void | |
2594 | ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id) | |
2595 | { | |
2596 | ztest_shared_t *zs = ztest_shared; | |
2597 | spa_t *spa = ztest_spa; | |
2598 | vdev_t *rvd = spa->spa_root_vdev; | |
2599 | spa_aux_vdev_t *sav; | |
2600 | char *aux; | |
2601 | char *path; | |
2602 | uint64_t guid = 0; | |
2603 | int error; | |
2604 | ||
2605 | path = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); | |
2606 | ||
2607 | if (ztest_random(2) == 0) { | |
2608 | sav = &spa->spa_spares; | |
2609 | aux = ZPOOL_CONFIG_SPARES; | |
2610 | } else { | |
2611 | sav = &spa->spa_l2cache; | |
2612 | aux = ZPOOL_CONFIG_L2CACHE; | |
2613 | } | |
2614 | ||
2615 | mutex_enter(&ztest_vdev_lock); | |
2616 | ||
2617 | spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); | |
2618 | ||
2619 | if (sav->sav_count != 0 && ztest_random(4) == 0) { | |
2620 | /* | |
2621 | * Pick a random device to remove. | |
2622 | */ | |
2623 | guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid; | |
2624 | } else { | |
2625 | /* | |
2626 | * Find an unused device we can add. | |
2627 | */ | |
2628 | zs->zs_vdev_aux = 0; | |
2629 | for (;;) { | |
2630 | int c; | |
2631 | (void) snprintf(path, MAXPATHLEN, ztest_aux_template, | |
2632 | ztest_opts.zo_dir, ztest_opts.zo_pool, aux, | |
2633 | zs->zs_vdev_aux); | |
2634 | for (c = 0; c < sav->sav_count; c++) | |
2635 | if (strcmp(sav->sav_vdevs[c]->vdev_path, | |
2636 | path) == 0) | |
2637 | break; | |
2638 | if (c == sav->sav_count && | |
2639 | vdev_lookup_by_path(rvd, path) == NULL) | |
2640 | break; | |
2641 | zs->zs_vdev_aux++; | |
2642 | } | |
2643 | } | |
2644 | ||
2645 | spa_config_exit(spa, SCL_VDEV, FTAG); | |
2646 | ||
2647 | if (guid == 0) { | |
2648 | /* | |
2649 | * Add a new device. | |
2650 | */ | |
2651 | nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL, | |
2652 | (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1); | |
2653 | error = spa_vdev_add(spa, nvroot); | |
2654 | if (error != 0) | |
2655 | fatal(0, "spa_vdev_add(%p) = %d", nvroot, error); | |
2656 | nvlist_free(nvroot); | |
2657 | } else { | |
2658 | /* | |
2659 | * Remove an existing device. Sometimes, dirty its | |
2660 | * vdev state first to make sure we handle removal | |
2661 | * of devices that have pending state changes. | |
2662 | */ | |
2663 | if (ztest_random(2) == 0) | |
2664 | (void) vdev_online(spa, guid, 0, NULL); | |
2665 | ||
2666 | error = spa_vdev_remove(spa, guid, B_FALSE); | |
2667 | if (error != 0 && error != EBUSY) | |
2668 | fatal(0, "spa_vdev_remove(%llu) = %d", guid, error); | |
2669 | } | |
2670 | ||
2671 | mutex_exit(&ztest_vdev_lock); | |
2672 | ||
2673 | umem_free(path, MAXPATHLEN); | |
2674 | } | |
2675 | ||
2676 | /* | |
2677 | * split a pool if it has mirror tlvdevs | |
2678 | */ | |
2679 | /* ARGSUSED */ | |
2680 | void | |
2681 | ztest_split_pool(ztest_ds_t *zd, uint64_t id) | |
2682 | { | |
2683 | ztest_shared_t *zs = ztest_shared; | |
2684 | spa_t *spa = ztest_spa; | |
2685 | vdev_t *rvd = spa->spa_root_vdev; | |
2686 | nvlist_t *tree, **child, *config, *split, **schild; | |
2687 | uint_t c, children, schildren = 0, lastlogid = 0; | |
2688 | int error = 0; | |
2689 | ||
2690 | mutex_enter(&ztest_vdev_lock); | |
2691 | ||
2692 | /* ensure we have a useable config; mirrors of raidz aren't supported */ | |
2693 | if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) { | |
2694 | mutex_exit(&ztest_vdev_lock); | |
2695 | return; | |
2696 | } | |
2697 | ||
2698 | /* clean up the old pool, if any */ | |
2699 | (void) spa_destroy("splitp"); | |
2700 | ||
2701 | spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); | |
2702 | ||
2703 | /* generate a config from the existing config */ | |
2704 | mutex_enter(&spa->spa_props_lock); | |
2705 | VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE, | |
2706 | &tree) == 0); | |
2707 | mutex_exit(&spa->spa_props_lock); | |
2708 | ||
2709 | VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child, | |
2710 | &children) == 0); | |
2711 | ||
2712 | schild = malloc(rvd->vdev_children * sizeof (nvlist_t *)); | |
2713 | for (c = 0; c < children; c++) { | |
2714 | vdev_t *tvd = rvd->vdev_child[c]; | |
2715 | nvlist_t **mchild; | |
2716 | uint_t mchildren; | |
2717 | ||
2718 | if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) { | |
2719 | VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME, | |
2720 | 0) == 0); | |
2721 | VERIFY(nvlist_add_string(schild[schildren], | |
2722 | ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0); | |
2723 | VERIFY(nvlist_add_uint64(schild[schildren], | |
2724 | ZPOOL_CONFIG_IS_HOLE, 1) == 0); | |
2725 | if (lastlogid == 0) | |
2726 | lastlogid = schildren; | |
2727 | ++schildren; | |
2728 | continue; | |
2729 | } | |
2730 | lastlogid = 0; | |
2731 | VERIFY(nvlist_lookup_nvlist_array(child[c], | |
2732 | ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0); | |
2733 | VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0); | |
2734 | } | |
2735 | ||
2736 | /* OK, create a config that can be used to split */ | |
2737 | VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0); | |
2738 | VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE, | |
2739 | VDEV_TYPE_ROOT) == 0); | |
2740 | VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild, | |
2741 | lastlogid != 0 ? lastlogid : schildren) == 0); | |
2742 | ||
2743 | VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0); | |
2744 | VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0); | |
2745 | ||
2746 | for (c = 0; c < schildren; c++) | |
2747 | nvlist_free(schild[c]); | |
2748 | free(schild); | |
2749 | nvlist_free(split); | |
2750 | ||
2751 | spa_config_exit(spa, SCL_VDEV, FTAG); | |
2752 | ||
2753 | (void) rw_enter(&ztest_name_lock, RW_WRITER); | |
2754 | error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE); | |
2755 | (void) rw_exit(&ztest_name_lock); | |
2756 | ||
2757 | nvlist_free(config); | |
2758 | ||
2759 | if (error == 0) { | |
2760 | (void) printf("successful split - results:\n"); | |
2761 | mutex_enter(&spa_namespace_lock); | |
2762 | show_pool_stats(spa); | |
2763 | show_pool_stats(spa_lookup("splitp")); | |
2764 | mutex_exit(&spa_namespace_lock); | |
2765 | ++zs->zs_splits; | |
2766 | --zs->zs_mirrors; | |
2767 | } | |
2768 | mutex_exit(&ztest_vdev_lock); | |
2769 | ||
2770 | } | |
2771 | ||
2772 | /* | |
2773 | * Verify that we can attach and detach devices. | |
2774 | */ | |
2775 | /* ARGSUSED */ | |
2776 | void | |
2777 | ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id) | |
2778 | { | |
2779 | ztest_shared_t *zs = ztest_shared; | |
2780 | spa_t *spa = ztest_spa; | |
2781 | spa_aux_vdev_t *sav = &spa->spa_spares; | |
2782 | vdev_t *rvd = spa->spa_root_vdev; | |
2783 | vdev_t *oldvd, *newvd, *pvd; | |
2784 | nvlist_t *root; | |
2785 | uint64_t leaves; | |
2786 | uint64_t leaf, top; | |
2787 | uint64_t ashift = ztest_get_ashift(); | |
2788 | uint64_t oldguid, pguid; | |
2789 | uint64_t oldsize, newsize; | |
2790 | char *oldpath, *newpath; | |
2791 | int replacing; | |
2792 | int oldvd_has_siblings = B_FALSE; | |
2793 | int newvd_is_spare = B_FALSE; | |
2794 | int oldvd_is_log; | |
2795 | int error, expected_error; | |
2796 | ||
2797 | oldpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); | |
2798 | newpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); | |
2799 | ||
2800 | mutex_enter(&ztest_vdev_lock); | |
2801 | leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz; | |
2802 | ||
2803 | spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); | |
2804 | ||
2805 | /* | |
2806 | * Decide whether to do an attach or a replace. | |
2807 | */ | |
2808 | replacing = ztest_random(2); | |
2809 | ||
2810 | /* | |
2811 | * Pick a random top-level vdev. | |
2812 | */ | |
2813 | top = ztest_random_vdev_top(spa, B_TRUE); | |
2814 | ||
2815 | /* | |
2816 | * Pick a random leaf within it. | |
2817 | */ | |
2818 | leaf = ztest_random(leaves); | |
2819 | ||
2820 | /* | |
2821 | * Locate this vdev. | |
2822 | */ | |
2823 | oldvd = rvd->vdev_child[top]; | |
2824 | if (zs->zs_mirrors >= 1) { | |
2825 | ASSERT(oldvd->vdev_ops == &vdev_mirror_ops); | |
2826 | ASSERT(oldvd->vdev_children >= zs->zs_mirrors); | |
2827 | oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz]; | |
2828 | } | |
2829 | if (ztest_opts.zo_raidz > 1) { | |
2830 | ASSERT(oldvd->vdev_ops == &vdev_raidz_ops); | |
2831 | ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz); | |
2832 | oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz]; | |
2833 | } | |
2834 | ||
2835 | /* | |
2836 | * If we're already doing an attach or replace, oldvd may be a | |
2837 | * mirror vdev -- in which case, pick a random child. | |
2838 | */ | |
2839 | while (oldvd->vdev_children != 0) { | |
2840 | oldvd_has_siblings = B_TRUE; | |
2841 | ASSERT(oldvd->vdev_children >= 2); | |
2842 | oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)]; | |
2843 | } | |
2844 | ||
2845 | oldguid = oldvd->vdev_guid; | |
2846 | oldsize = vdev_get_min_asize(oldvd); | |
2847 | oldvd_is_log = oldvd->vdev_top->vdev_islog; | |
2848 | (void) strcpy(oldpath, oldvd->vdev_path); | |
2849 | pvd = oldvd->vdev_parent; | |
2850 | pguid = pvd->vdev_guid; | |
2851 | ||
2852 | /* | |
2853 | * If oldvd has siblings, then half of the time, detach it. | |
2854 | */ | |
2855 | if (oldvd_has_siblings && ztest_random(2) == 0) { | |
2856 | spa_config_exit(spa, SCL_VDEV, FTAG); | |
2857 | error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE); | |
2858 | if (error != 0 && error != ENODEV && error != EBUSY && | |
2859 | error != ENOTSUP) | |
2860 | fatal(0, "detach (%s) returned %d", oldpath, error); | |
2861 | goto out; | |
2862 | } | |
2863 | ||
2864 | /* | |
2865 | * For the new vdev, choose with equal probability between the two | |
2866 | * standard paths (ending in either 'a' or 'b') or a random hot spare. | |
2867 | */ | |
2868 | if (sav->sav_count != 0 && ztest_random(3) == 0) { | |
2869 | newvd = sav->sav_vdevs[ztest_random(sav->sav_count)]; | |
2870 | newvd_is_spare = B_TRUE; | |
2871 | (void) strcpy(newpath, newvd->vdev_path); | |
2872 | } else { | |
2873 | (void) snprintf(newpath, MAXPATHLEN, ztest_dev_template, | |
2874 | ztest_opts.zo_dir, ztest_opts.zo_pool, | |
2875 | top * leaves + leaf); | |
2876 | if (ztest_random(2) == 0) | |
2877 | newpath[strlen(newpath) - 1] = 'b'; | |
2878 | newvd = vdev_lookup_by_path(rvd, newpath); | |
2879 | } | |
2880 | ||
2881 | if (newvd) { | |
2882 | newsize = vdev_get_min_asize(newvd); | |
2883 | } else { | |
2884 | /* | |
2885 | * Make newsize a little bigger or smaller than oldsize. | |
2886 | * If it's smaller, the attach should fail. | |
2887 | * If it's larger, and we're doing a replace, | |
2888 | * we should get dynamic LUN growth when we're done. | |
2889 | */ | |
2890 | newsize = 10 * oldsize / (9 + ztest_random(3)); | |
2891 | } | |
2892 | ||
2893 | /* | |
2894 | * If pvd is not a mirror or root, the attach should fail with ENOTSUP, | |
2895 | * unless it's a replace; in that case any non-replacing parent is OK. | |
2896 | * | |
2897 | * If newvd is already part of the pool, it should fail with EBUSY. | |
2898 | * | |
2899 | * If newvd is too small, it should fail with EOVERFLOW. | |
2900 | */ | |
2901 | if (pvd->vdev_ops != &vdev_mirror_ops && | |
2902 | pvd->vdev_ops != &vdev_root_ops && (!replacing || | |
2903 | pvd->vdev_ops == &vdev_replacing_ops || | |
2904 | pvd->vdev_ops == &vdev_spare_ops)) | |
2905 | expected_error = ENOTSUP; | |
2906 | else if (newvd_is_spare && (!replacing || oldvd_is_log)) | |
2907 | expected_error = ENOTSUP; | |
2908 | else if (newvd == oldvd) | |
2909 | expected_error = replacing ? 0 : EBUSY; | |
2910 | else if (vdev_lookup_by_path(rvd, newpath) != NULL) | |
2911 | expected_error = EBUSY; | |
2912 | else if (newsize < oldsize) | |
2913 | expected_error = EOVERFLOW; | |
2914 | else if (ashift > oldvd->vdev_top->vdev_ashift) | |
2915 | expected_error = EDOM; | |
2916 | else | |
2917 | expected_error = 0; | |
2918 | ||
2919 | spa_config_exit(spa, SCL_VDEV, FTAG); | |
2920 | ||
2921 | /* | |
2922 | * Build the nvlist describing newpath. | |
2923 | */ | |
2924 | root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0, | |
2925 | ashift, 0, 0, 0, 1); | |
2926 | ||
2927 | error = spa_vdev_attach(spa, oldguid, root, replacing); | |
2928 | ||
2929 | nvlist_free(root); | |
2930 | ||
2931 | /* | |
2932 | * If our parent was the replacing vdev, but the replace completed, | |
2933 | * then instead of failing with ENOTSUP we may either succeed, | |
2934 | * fail with ENODEV, or fail with EOVERFLOW. | |
2935 | */ | |
2936 | if (expected_error == ENOTSUP && | |
2937 | (error == 0 || error == ENODEV || error == EOVERFLOW)) | |
2938 | expected_error = error; | |
2939 | ||
2940 | /* | |
2941 | * If someone grew the LUN, the replacement may be too small. | |
2942 | */ | |
2943 | if (error == EOVERFLOW || error == EBUSY) | |
2944 | expected_error = error; | |
2945 | ||
2946 | /* XXX workaround 6690467 */ | |
2947 | if (error != expected_error && expected_error != EBUSY) { | |
2948 | fatal(0, "attach (%s %llu, %s %llu, %d) " | |
2949 | "returned %d, expected %d", | |
2950 | oldpath, oldsize, newpath, | |
2951 | newsize, replacing, error, expected_error); | |
2952 | } | |
2953 | out: | |
2954 | mutex_exit(&ztest_vdev_lock); | |
2955 | ||
2956 | umem_free(oldpath, MAXPATHLEN); | |
2957 | umem_free(newpath, MAXPATHLEN); | |
2958 | } | |
2959 | ||
2960 | /* | |
2961 | * Callback function which expands the physical size of the vdev. | |
2962 | */ | |
2963 | vdev_t * | |
2964 | grow_vdev(vdev_t *vd, void *arg) | |
2965 | { | |
2966 | ASSERTV(spa_t *spa = vd->vdev_spa); | |
2967 | size_t *newsize = arg; | |
2968 | size_t fsize; | |
2969 | int fd; | |
2970 | ||
2971 | ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE); | |
2972 | ASSERT(vd->vdev_ops->vdev_op_leaf); | |
2973 | ||
2974 | if ((fd = open(vd->vdev_path, O_RDWR)) == -1) | |
2975 | return (vd); | |
2976 | ||
2977 | fsize = lseek(fd, 0, SEEK_END); | |
2978 | VERIFY(ftruncate(fd, *newsize) == 0); | |
2979 | ||
2980 | if (ztest_opts.zo_verbose >= 6) { | |
2981 | (void) printf("%s grew from %lu to %lu bytes\n", | |
2982 | vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize); | |
2983 | } | |
2984 | (void) close(fd); | |
2985 | return (NULL); | |
2986 | } | |
2987 | ||
2988 | /* | |
2989 | * Callback function which expands a given vdev by calling vdev_online(). | |
2990 | */ | |
2991 | /* ARGSUSED */ | |
2992 | vdev_t * | |
2993 | online_vdev(vdev_t *vd, void *arg) | |
2994 | { | |
2995 | spa_t *spa = vd->vdev_spa; | |
2996 | vdev_t *tvd = vd->vdev_top; | |
2997 | uint64_t guid = vd->vdev_guid; | |
2998 | uint64_t generation = spa->spa_config_generation + 1; | |
2999 | vdev_state_t newstate = VDEV_STATE_UNKNOWN; | |
3000 | int error; | |
3001 | ||
3002 | ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE); | |
3003 | ASSERT(vd->vdev_ops->vdev_op_leaf); | |
3004 | ||
3005 | /* Calling vdev_online will initialize the new metaslabs */ | |
3006 | spa_config_exit(spa, SCL_STATE, spa); | |
3007 | error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate); | |
3008 | spa_config_enter(spa, SCL_STATE, spa, RW_READER); | |
3009 | ||
3010 | /* | |
3011 | * If vdev_online returned an error or the underlying vdev_open | |
3012 | * failed then we abort the expand. The only way to know that | |
3013 | * vdev_open fails is by checking the returned newstate. | |
3014 | */ | |
3015 | if (error || newstate != VDEV_STATE_HEALTHY) { | |
3016 | if (ztest_opts.zo_verbose >= 5) { | |
3017 | (void) printf("Unable to expand vdev, state %llu, " | |
3018 | "error %d\n", (u_longlong_t)newstate, error); | |
3019 | } | |
3020 | return (vd); | |
3021 | } | |
3022 | ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY); | |
3023 | ||
3024 | /* | |
3025 | * Since we dropped the lock we need to ensure that we're | |
3026 | * still talking to the original vdev. It's possible this | |
3027 | * vdev may have been detached/replaced while we were | |
3028 | * trying to online it. | |
3029 | */ | |
3030 | if (generation != spa->spa_config_generation) { | |
3031 | if (ztest_opts.zo_verbose >= 5) { | |
3032 | (void) printf("vdev configuration has changed, " | |
3033 | "guid %llu, state %llu, expected gen %llu, " | |
3034 | "got gen %llu\n", | |
3035 | (u_longlong_t)guid, | |
3036 | (u_longlong_t)tvd->vdev_state, | |
3037 | (u_longlong_t)generation, | |
3038 | (u_longlong_t)spa->spa_config_generation); | |
3039 | } | |
3040 | return (vd); | |
3041 | } | |
3042 | return (NULL); | |
3043 | } | |
3044 | ||
3045 | /* | |
3046 | * Traverse the vdev tree calling the supplied function. | |
3047 | * We continue to walk the tree until we either have walked all | |
3048 | * children or we receive a non-NULL return from the callback. | |
3049 | * If a NULL callback is passed, then we just return back the first | |
3050 | * leaf vdev we encounter. | |
3051 | */ | |
3052 | vdev_t * | |
3053 | vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg) | |
3054 | { | |
3055 | uint_t c; | |
3056 | ||
3057 | if (vd->vdev_ops->vdev_op_leaf) { | |
3058 | if (func == NULL) | |
3059 | return (vd); | |
3060 | else | |
3061 | return (func(vd, arg)); | |
3062 | } | |
3063 | ||
3064 | for (c = 0; c < vd->vdev_children; c++) { | |
3065 | vdev_t *cvd = vd->vdev_child[c]; | |
3066 | if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL) | |
3067 | return (cvd); | |
3068 | } | |
3069 | return (NULL); | |
3070 | } | |
3071 | ||
3072 | /* | |
3073 | * Verify that dynamic LUN growth works as expected. | |
3074 | */ | |
3075 | /* ARGSUSED */ | |
3076 | void | |
3077 | ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id) | |
3078 | { | |
3079 | spa_t *spa = ztest_spa; | |
3080 | vdev_t *vd, *tvd; | |
3081 | metaslab_class_t *mc; | |
3082 | metaslab_group_t *mg; | |
3083 | size_t psize, newsize; | |
3084 | uint64_t top; | |
3085 | uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count; | |
3086 | ||
3087 | mutex_enter(&ztest_vdev_lock); | |
3088 | spa_config_enter(spa, SCL_STATE, spa, RW_READER); | |
3089 | ||
3090 | top = ztest_random_vdev_top(spa, B_TRUE); | |
3091 | ||
3092 | tvd = spa->spa_root_vdev->vdev_child[top]; | |
3093 | mg = tvd->vdev_mg; | |
3094 | mc = mg->mg_class; | |
3095 | old_ms_count = tvd->vdev_ms_count; | |
3096 | old_class_space = metaslab_class_get_space(mc); | |
3097 | ||
3098 | /* | |
3099 | * Determine the size of the first leaf vdev associated with | |
3100 | * our top-level device. | |
3101 | */ | |
3102 | vd = vdev_walk_tree(tvd, NULL, NULL); | |
3103 | ASSERT3P(vd, !=, NULL); | |
3104 | ASSERT(vd->vdev_ops->vdev_op_leaf); | |
3105 | ||
3106 | psize = vd->vdev_psize; | |
3107 | ||
3108 | /* | |
3109 | * We only try to expand the vdev if it's healthy, less than 4x its | |
3110 | * original size, and it has a valid psize. | |
3111 | */ | |
3112 | if (tvd->vdev_state != VDEV_STATE_HEALTHY || | |
3113 | psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) { | |
3114 | spa_config_exit(spa, SCL_STATE, spa); | |
3115 | mutex_exit(&ztest_vdev_lock); | |
3116 | return; | |
3117 | } | |
3118 | ASSERT(psize > 0); | |
3119 | newsize = psize + psize / 8; | |
3120 | ASSERT3U(newsize, >, psize); | |
3121 | ||
3122 | if (ztest_opts.zo_verbose >= 6) { | |
3123 | (void) printf("Expanding LUN %s from %lu to %lu\n", | |
3124 | vd->vdev_path, (ulong_t)psize, (ulong_t)newsize); | |
3125 | } | |
3126 | ||
3127 | /* | |
3128 | * Growing the vdev is a two step process: | |
3129 | * 1). expand the physical size (i.e. relabel) | |
3130 | * 2). online the vdev to create the new metaslabs | |
3131 | */ | |
3132 | if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL || | |
3133 | vdev_walk_tree(tvd, online_vdev, NULL) != NULL || | |
3134 | tvd->vdev_state != VDEV_STATE_HEALTHY) { | |
3135 | if (ztest_opts.zo_verbose >= 5) { | |
3136 | (void) printf("Could not expand LUN because " | |
3137 | "the vdev configuration changed.\n"); | |
3138 | } | |
3139 | spa_config_exit(spa, SCL_STATE, spa); | |
3140 | mutex_exit(&ztest_vdev_lock); | |
3141 | return; | |
3142 | } | |
3143 | ||
3144 | spa_config_exit(spa, SCL_STATE, spa); | |
3145 | ||
3146 | /* | |
3147 | * Expanding the LUN will update the config asynchronously, | |
3148 | * thus we must wait for the async thread to complete any | |
3149 | * pending tasks before proceeding. | |
3150 | */ | |
3151 | for (;;) { | |
3152 | boolean_t done; | |
3153 | mutex_enter(&spa->spa_async_lock); | |
3154 | done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks); | |
3155 | mutex_exit(&spa->spa_async_lock); | |
3156 | if (done) | |
3157 | break; | |
3158 | txg_wait_synced(spa_get_dsl(spa), 0); | |
3159 | (void) poll(NULL, 0, 100); | |
3160 | } | |
3161 | ||
3162 | spa_config_enter(spa, SCL_STATE, spa, RW_READER); | |
3163 | ||
3164 | tvd = spa->spa_root_vdev->vdev_child[top]; | |
3165 | new_ms_count = tvd->vdev_ms_count; | |
3166 | new_class_space = metaslab_class_get_space(mc); | |
3167 | ||
3168 | if (tvd->vdev_mg != mg || mg->mg_class != mc) { | |
3169 | if (ztest_opts.zo_verbose >= 5) { | |
3170 | (void) printf("Could not verify LUN expansion due to " | |
3171 | "intervening vdev offline or remove.\n"); | |
3172 | } | |
3173 | spa_config_exit(spa, SCL_STATE, spa); | |
3174 | mutex_exit(&ztest_vdev_lock); | |
3175 | return; | |
3176 | } | |
3177 | ||
3178 | /* | |
3179 | * Make sure we were able to grow the vdev. | |
3180 | */ | |
3181 | if (new_ms_count <= old_ms_count) | |
3182 | fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n", | |
3183 | old_ms_count, new_ms_count); | |
3184 | ||
3185 | /* | |
3186 | * Make sure we were able to grow the pool. | |
3187 | */ | |
3188 | if (new_class_space <= old_class_space) | |
3189 | fatal(0, "LUN expansion failed: class_space %llu <= %llu\n", | |
3190 | old_class_space, new_class_space); | |
3191 | ||
3192 | if (ztest_opts.zo_verbose >= 5) { | |
3193 | char oldnumbuf[6], newnumbuf[6]; | |
3194 | ||
3195 | nicenum(old_class_space, oldnumbuf); | |
3196 | nicenum(new_class_space, newnumbuf); | |
3197 | (void) printf("%s grew from %s to %s\n", | |
3198 | spa->spa_name, oldnumbuf, newnumbuf); | |
3199 | } | |
3200 | ||
3201 | spa_config_exit(spa, SCL_STATE, spa); | |
3202 | mutex_exit(&ztest_vdev_lock); | |
3203 | } | |
3204 | ||
3205 | /* | |
3206 | * Verify that dmu_objset_{create,destroy,open,close} work as expected. | |
3207 | */ | |
3208 | /* ARGSUSED */ | |
3209 | static void | |
3210 | ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx) | |
3211 | { | |
3212 | /* | |
3213 | * Create the objects common to all ztest datasets. | |
3214 | */ | |
3215 | VERIFY(zap_create_claim(os, ZTEST_DIROBJ, | |
3216 | DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0); | |
3217 | } | |
3218 | ||
3219 | static int | |
3220 | ztest_dataset_create(char *dsname) | |
3221 | { | |
3222 | uint64_t zilset = ztest_random(100); | |
3223 | int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0, | |
3224 | ztest_objset_create_cb, NULL); | |
3225 | ||
3226 | if (err || zilset < 80) | |
3227 | return (err); | |
3228 | ||
3229 | if (ztest_opts.zo_verbose >= 5) | |
3230 | (void) printf("Setting dataset %s to sync always\n", dsname); | |
3231 | return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC, | |
3232 | ZFS_SYNC_ALWAYS, B_FALSE)); | |
3233 | } | |
3234 | ||
3235 | /* ARGSUSED */ | |
3236 | static int | |
3237 | ztest_objset_destroy_cb(const char *name, void *arg) | |
3238 | { | |
3239 | objset_t *os; | |
3240 | dmu_object_info_t doi; | |
3241 | int error; | |
3242 | ||
3243 | /* | |
3244 | * Verify that the dataset contains a directory object. | |
3245 | */ | |
3246 | VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, FTAG, &os)); | |
3247 | error = dmu_object_info(os, ZTEST_DIROBJ, &doi); | |
3248 | if (error != ENOENT) { | |
3249 | /* We could have crashed in the middle of destroying it */ | |
3250 | ASSERT0(error); | |
3251 | ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER); | |
3252 | ASSERT3S(doi.doi_physical_blocks_512, >=, 0); | |
3253 | } | |
3254 | dmu_objset_disown(os, FTAG); | |
3255 | ||
3256 | /* | |
3257 | * Destroy the dataset. | |
3258 | */ | |
3259 | if (strchr(name, '@') != NULL) { | |
3260 | VERIFY0(dsl_destroy_snapshot(name, B_FALSE)); | |
3261 | } else { | |
3262 | VERIFY0(dsl_destroy_head(name)); | |
3263 | } | |
3264 | return (0); | |
3265 | } | |
3266 | ||
3267 | static boolean_t | |
3268 | ztest_snapshot_create(char *osname, uint64_t id) | |
3269 | { | |
3270 | char snapname[MAXNAMELEN]; | |
3271 | int error; | |
3272 | ||
3273 | (void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id); | |
3274 | ||
3275 | error = dmu_objset_snapshot_one(osname, snapname); | |
3276 | if (error == ENOSPC) { | |
3277 | ztest_record_enospc(FTAG); | |
3278 | return (B_FALSE); | |
3279 | } | |
3280 | if (error != 0 && error != EEXIST) { | |
3281 | fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname, | |
3282 | snapname, error); | |
3283 | } | |
3284 | return (B_TRUE); | |
3285 | } | |
3286 | ||
3287 | static boolean_t | |
3288 | ztest_snapshot_destroy(char *osname, uint64_t id) | |
3289 | { | |
3290 | char snapname[MAXNAMELEN]; | |
3291 | int error; | |
3292 | ||
3293 | (void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname, | |
3294 | (u_longlong_t)id); | |
3295 | ||
3296 | error = dsl_destroy_snapshot(snapname, B_FALSE); | |
3297 | if (error != 0 && error != ENOENT) | |
3298 | fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error); | |
3299 | return (B_TRUE); | |
3300 | } | |
3301 | ||
3302 | /* ARGSUSED */ | |
3303 | void | |
3304 | ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id) | |
3305 | { | |
3306 | ztest_ds_t *zdtmp; | |
3307 | int iters; | |
3308 | int error; | |
3309 | objset_t *os, *os2; | |
3310 | char *name; | |
3311 | zilog_t *zilog; | |
3312 | int i; | |
3313 | ||
3314 | zdtmp = umem_alloc(sizeof (ztest_ds_t), UMEM_NOFAIL); | |
3315 | name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); | |
3316 | ||
3317 | (void) rw_enter(&ztest_name_lock, RW_READER); | |
3318 | ||
3319 | (void) snprintf(name, MAXNAMELEN, "%s/temp_%llu", | |
3320 | ztest_opts.zo_pool, (u_longlong_t)id); | |
3321 | ||
3322 | /* | |
3323 | * If this dataset exists from a previous run, process its replay log | |
3324 | * half of the time. If we don't replay it, then dsl_destroy_head() | |
3325 | * (invoked from ztest_objset_destroy_cb()) should just throw it away. | |
3326 | */ | |
3327 | if (ztest_random(2) == 0 && | |
3328 | dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) { | |
3329 | ztest_zd_init(zdtmp, NULL, os); | |
3330 | zil_replay(os, zdtmp, ztest_replay_vector); | |
3331 | ztest_zd_fini(zdtmp); | |
3332 | dmu_objset_disown(os, FTAG); | |
3333 | } | |
3334 | ||
3335 | /* | |
3336 | * There may be an old instance of the dataset we're about to | |
3337 | * create lying around from a previous run. If so, destroy it | |
3338 | * and all of its snapshots. | |
3339 | */ | |
3340 | (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL, | |
3341 | DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS); | |
3342 | ||
3343 | /* | |
3344 | * Verify that the destroyed dataset is no longer in the namespace. | |
3345 | */ | |
3346 | VERIFY3U(ENOENT, ==, dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, | |
3347 | FTAG, &os)); | |
3348 | ||
3349 | /* | |
3350 | * Verify that we can create a new dataset. | |
3351 | */ | |
3352 | error = ztest_dataset_create(name); | |
3353 | if (error) { | |
3354 | if (error == ENOSPC) { | |
3355 | ztest_record_enospc(FTAG); | |
3356 | goto out; | |
3357 | } | |
3358 | fatal(0, "dmu_objset_create(%s) = %d", name, error); | |
3359 | } | |
3360 | ||
3361 | VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os)); | |
3362 | ||
3363 | ztest_zd_init(zdtmp, NULL, os); | |
3364 | ||
3365 | /* | |
3366 | * Open the intent log for it. | |
3367 | */ | |
3368 | zilog = zil_open(os, ztest_get_data); | |
3369 | ||
3370 | /* | |
3371 | * Put some objects in there, do a little I/O to them, | |
3372 | * and randomly take a couple of snapshots along the way. | |
3373 | */ | |
3374 | iters = ztest_random(5); | |
3375 | for (i = 0; i < iters; i++) { | |
3376 | ztest_dmu_object_alloc_free(zdtmp, id); | |
3377 | if (ztest_random(iters) == 0) | |
3378 | (void) ztest_snapshot_create(name, i); | |
3379 | } | |
3380 | ||
3381 | /* | |
3382 | * Verify that we cannot create an existing dataset. | |
3383 | */ | |
3384 | VERIFY3U(EEXIST, ==, | |
3385 | dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL)); | |
3386 | ||
3387 | /* | |
3388 | * Verify that we can hold an objset that is also owned. | |
3389 | */ | |
3390 | VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2)); | |
3391 | dmu_objset_rele(os2, FTAG); | |
3392 | ||
3393 | /* | |
3394 | * Verify that we cannot own an objset that is already owned. | |
3395 | */ | |
3396 | VERIFY3U(EBUSY, ==, | |
3397 | dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2)); | |
3398 | ||
3399 | zil_close(zilog); | |
3400 | dmu_objset_disown(os, FTAG); | |
3401 | ztest_zd_fini(zdtmp); | |
3402 | out: | |
3403 | (void) rw_exit(&ztest_name_lock); | |
3404 | ||
3405 | umem_free(name, MAXNAMELEN); | |
3406 | umem_free(zdtmp, sizeof (ztest_ds_t)); | |
3407 | } | |
3408 | ||
3409 | /* | |
3410 | * Verify that dmu_snapshot_{create,destroy,open,close} work as expected. | |
3411 | */ | |
3412 | void | |
3413 | ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id) | |
3414 | { | |
3415 | (void) rw_enter(&ztest_name_lock, RW_READER); | |
3416 | (void) ztest_snapshot_destroy(zd->zd_name, id); | |
3417 | (void) ztest_snapshot_create(zd->zd_name, id); | |
3418 | (void) rw_exit(&ztest_name_lock); | |
3419 | } | |
3420 | ||
3421 | /* | |
3422 | * Cleanup non-standard snapshots and clones. | |
3423 | */ | |
3424 | void | |
3425 | ztest_dsl_dataset_cleanup(char *osname, uint64_t id) | |
3426 | { | |
3427 | char *snap1name; | |
3428 | char *clone1name; | |
3429 | char *snap2name; | |
3430 | char *clone2name; | |
3431 | char *snap3name; | |
3432 | int error; | |
3433 | ||
3434 | snap1name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); | |
3435 | clone1name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); | |
3436 | snap2name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); | |
3437 | clone2name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); | |
3438 | snap3name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); | |
3439 | ||
3440 | (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", | |
3441 | osname, (u_longlong_t)id); | |
3442 | (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", | |
3443 | osname, (u_longlong_t)id); | |
3444 | (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", | |
3445 | clone1name, (u_longlong_t)id); | |
3446 | (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", | |
3447 | osname, (u_longlong_t)id); | |
3448 | (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", | |
3449 | clone1name, (u_longlong_t)id); | |
3450 | ||
3451 | error = dsl_destroy_head(clone2name); | |
3452 | if (error && error != ENOENT) | |
3453 | fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error); | |
3454 | error = dsl_destroy_snapshot(snap3name, B_FALSE); | |
3455 | if (error && error != ENOENT) | |
3456 | fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error); | |
3457 | error = dsl_destroy_snapshot(snap2name, B_FALSE); | |
3458 | if (error && error != ENOENT) | |
3459 | fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error); | |
3460 | error = dsl_destroy_head(clone1name); | |
3461 | if (error && error != ENOENT) | |
3462 | fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error); | |
3463 | error = dsl_destroy_snapshot(snap1name, B_FALSE); | |
3464 | if (error && error != ENOENT) | |
3465 | fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error); | |
3466 | ||
3467 | umem_free(snap1name, MAXNAMELEN); | |
3468 | umem_free(clone1name, MAXNAMELEN); | |
3469 | umem_free(snap2name, MAXNAMELEN); | |
3470 | umem_free(clone2name, MAXNAMELEN); | |
3471 | umem_free(snap3name, MAXNAMELEN); | |
3472 | } | |
3473 | ||
3474 | /* | |
3475 | * Verify dsl_dataset_promote handles EBUSY | |
3476 | */ | |
3477 | void | |
3478 | ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id) | |
3479 | { | |
3480 | objset_t *os; | |
3481 | char *snap1name; | |
3482 | char *clone1name; | |
3483 | char *snap2name; | |
3484 | char *clone2name; | |
3485 | char *snap3name; | |
3486 | char *osname = zd->zd_name; | |
3487 | int error; | |
3488 | ||
3489 | snap1name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); | |
3490 | clone1name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); | |
3491 | snap2name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); | |
3492 | clone2name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); | |
3493 | snap3name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); | |
3494 | ||
3495 | (void) rw_enter(&ztest_name_lock, RW_READER); | |
3496 | ||
3497 | ztest_dsl_dataset_cleanup(osname, id); | |
3498 | ||
3499 | (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", | |
3500 | osname, (u_longlong_t)id); | |
3501 | (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", | |
3502 | osname, (u_longlong_t)id); | |
3503 | (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", | |
3504 | clone1name, (u_longlong_t)id); | |
3505 | (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", | |
3506 | osname, (u_longlong_t)id); | |
3507 | (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", | |
3508 | clone1name, (u_longlong_t)id); | |
3509 | ||
3510 | error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1); | |
3511 | if (error && error != EEXIST) { | |
3512 | if (error == ENOSPC) { | |
3513 | ztest_record_enospc(FTAG); | |
3514 | goto out; | |
3515 | } | |
3516 | fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error); | |
3517 | } | |
3518 | ||
3519 | error = dmu_objset_clone(clone1name, snap1name); | |
3520 | if (error) { | |
3521 | if (error == ENOSPC) { | |
3522 | ztest_record_enospc(FTAG); | |
3523 | goto out; | |
3524 | } | |
3525 | fatal(0, "dmu_objset_create(%s) = %d", clone1name, error); | |
3526 | } | |
3527 | ||
3528 | error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1); | |
3529 | if (error && error != EEXIST) { | |
3530 | if (error == ENOSPC) { | |
3531 | ztest_record_enospc(FTAG); | |
3532 | goto out; | |
3533 | } | |
3534 | fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error); | |
3535 | } | |
3536 | ||
3537 | error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1); | |
3538 | if (error && error != EEXIST) { | |
3539 | if (error == ENOSPC) { | |
3540 | ztest_record_enospc(FTAG); | |
3541 | goto out; | |
3542 | } | |
3543 | fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error); | |
3544 | } | |
3545 | ||
3546 | error = dmu_objset_clone(clone2name, snap3name); | |
3547 | if (error) { | |
3548 | if (error == ENOSPC) { | |
3549 | ztest_record_enospc(FTAG); | |
3550 | goto out; | |
3551 | } | |
3552 | fatal(0, "dmu_objset_create(%s) = %d", clone2name, error); | |
3553 | } | |
3554 | ||
3555 | error = dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, FTAG, &os); | |
3556 | if (error) | |
3557 | fatal(0, "dmu_objset_own(%s) = %d", snap2name, error); | |
3558 | error = dsl_dataset_promote(clone2name, NULL); | |
3559 | if (error != EBUSY) | |
3560 | fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name, | |
3561 | error); | |
3562 | dmu_objset_disown(os, FTAG); | |
3563 | ||
3564 | out: | |
3565 | ztest_dsl_dataset_cleanup(osname, id); | |
3566 | ||
3567 | (void) rw_exit(&ztest_name_lock); | |
3568 | ||
3569 | umem_free(snap1name, MAXNAMELEN); | |
3570 | umem_free(clone1name, MAXNAMELEN); | |
3571 | umem_free(snap2name, MAXNAMELEN); | |
3572 | umem_free(clone2name, MAXNAMELEN); | |
3573 | umem_free(snap3name, MAXNAMELEN); | |
3574 | } | |
3575 | ||
3576 | #undef OD_ARRAY_SIZE | |
3577 | #define OD_ARRAY_SIZE 4 | |
3578 | ||
3579 | /* | |
3580 | * Verify that dmu_object_{alloc,free} work as expected. | |
3581 | */ | |
3582 | void | |
3583 | ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id) | |
3584 | { | |
3585 | ztest_od_t *od; | |
3586 | int batchsize; | |
3587 | int size; | |
3588 | int b; | |
3589 | ||
3590 | size = sizeof (ztest_od_t) * OD_ARRAY_SIZE; | |
3591 | od = umem_alloc(size, UMEM_NOFAIL); | |
3592 | batchsize = OD_ARRAY_SIZE; | |
3593 | ||
3594 | for (b = 0; b < batchsize; b++) | |
3595 | ztest_od_init(od + b, id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0); | |
3596 | ||
3597 | /* | |
3598 | * Destroy the previous batch of objects, create a new batch, | |
3599 | * and do some I/O on the new objects. | |
3600 | */ | |
3601 | if (ztest_object_init(zd, od, size, B_TRUE) != 0) | |
3602 | return; | |
3603 | ||
3604 | while (ztest_random(4 * batchsize) != 0) | |
3605 | ztest_io(zd, od[ztest_random(batchsize)].od_object, | |
3606 | ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT); | |
3607 | ||
3608 | umem_free(od, size); | |
3609 | } | |
3610 | ||
3611 | #undef OD_ARRAY_SIZE | |
3612 | #define OD_ARRAY_SIZE 2 | |
3613 | ||
3614 | /* | |
3615 | * Verify that dmu_{read,write} work as expected. | |
3616 | */ | |
3617 | void | |
3618 | ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id) | |
3619 | { | |
3620 | int size; | |
3621 | ztest_od_t *od; | |
3622 | ||
3623 | objset_t *os = zd->zd_os; | |
3624 | size = sizeof (ztest_od_t) * OD_ARRAY_SIZE; | |
3625 | od = umem_alloc(size, UMEM_NOFAIL); | |
3626 | dmu_tx_t *tx; | |
3627 | int i, freeit, error; | |
3628 | uint64_t n, s, txg; | |
3629 | bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT; | |
3630 | uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize; | |
3631 | uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t); | |
3632 | uint64_t regions = 997; | |
3633 | uint64_t stride = 123456789ULL; | |
3634 | uint64_t width = 40; | |
3635 | int free_percent = 5; | |
3636 | ||
3637 | /* | |
3638 | * This test uses two objects, packobj and bigobj, that are always | |
3639 | * updated together (i.e. in the same tx) so that their contents are | |
3640 | * in sync and can be compared. Their contents relate to each other | |
3641 | * in a simple way: packobj is a dense array of 'bufwad' structures, | |
3642 | * while bigobj is a sparse array of the same bufwads. Specifically, | |
3643 | * for any index n, there are three bufwads that should be identical: | |
3644 | * | |
3645 | * packobj, at offset n * sizeof (bufwad_t) | |
3646 | * bigobj, at the head of the nth chunk | |
3647 | * bigobj, at the tail of the nth chunk | |
3648 | * | |
3649 | * The chunk size is arbitrary. It doesn't have to be a power of two, | |
3650 | * and it doesn't have any relation to the object blocksize. | |
3651 | * The only requirement is that it can hold at least two bufwads. | |
3652 | * | |
3653 | * Normally, we write the bufwad to each of these locations. | |
3654 | * However, free_percent of the time we instead write zeroes to | |
3655 | * packobj and perform a dmu_free_range() on bigobj. By comparing | |
3656 | * bigobj to packobj, we can verify that the DMU is correctly | |
3657 | * tracking which parts of an object are allocated and free, | |
3658 | * and that the contents of the allocated blocks are correct. | |
3659 | */ | |
3660 | ||
3661 | /* | |
3662 | * Read the directory info. If it's the first time, set things up. | |
3663 | */ | |
3664 | ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize); | |
3665 | ztest_od_init(od + 1, id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize); | |
3666 | ||
3667 | if (ztest_object_init(zd, od, size, B_FALSE) != 0) { | |
3668 | umem_free(od, size); | |
3669 | return; | |
3670 | } | |
3671 | ||
3672 | bigobj = od[0].od_object; | |
3673 | packobj = od[1].od_object; | |
3674 | chunksize = od[0].od_gen; | |
3675 | ASSERT(chunksize == od[1].od_gen); | |
3676 | ||
3677 | /* | |
3678 | * Prefetch a random chunk of the big object. | |
3679 | * Our aim here is to get some async reads in flight | |
3680 | * for blocks that we may free below; the DMU should | |
3681 | * handle this race correctly. | |
3682 | */ | |
3683 | n = ztest_random(regions) * stride + ztest_random(width); | |
3684 | s = 1 + ztest_random(2 * width - 1); | |
3685 | dmu_prefetch(os, bigobj, n * chunksize, s * chunksize); | |
3686 | ||
3687 | /* | |
3688 | * Pick a random index and compute the offsets into packobj and bigobj. | |
3689 | */ | |
3690 | n = ztest_random(regions) * stride + ztest_random(width); | |
3691 | s = 1 + ztest_random(width - 1); | |
3692 | ||
3693 | packoff = n * sizeof (bufwad_t); | |
3694 | packsize = s * sizeof (bufwad_t); | |
3695 | ||
3696 | bigoff = n * chunksize; | |
3697 | bigsize = s * chunksize; | |
3698 | ||
3699 | packbuf = umem_alloc(packsize, UMEM_NOFAIL); | |
3700 | bigbuf = umem_alloc(bigsize, UMEM_NOFAIL); | |
3701 | ||
3702 | /* | |
3703 | * free_percent of the time, free a range of bigobj rather than | |
3704 | * overwriting it. | |
3705 | */ | |
3706 | freeit = (ztest_random(100) < free_percent); | |
3707 | ||
3708 | /* | |
3709 | * Read the current contents of our objects. | |
3710 | */ | |
3711 | error = dmu_read(os, packobj, packoff, packsize, packbuf, | |
3712 | DMU_READ_PREFETCH); | |
3713 | ASSERT0(error); | |
3714 | error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf, | |
3715 | DMU_READ_PREFETCH); | |
3716 | ASSERT0(error); | |
3717 | ||
3718 | /* | |
3719 | * Get a tx for the mods to both packobj and bigobj. | |
3720 | */ | |
3721 | tx = dmu_tx_create(os); | |
3722 | ||
3723 | dmu_tx_hold_write(tx, packobj, packoff, packsize); | |
3724 | ||
3725 | if (freeit) | |
3726 | dmu_tx_hold_free(tx, bigobj, bigoff, bigsize); | |
3727 | else | |
3728 | dmu_tx_hold_write(tx, bigobj, bigoff, bigsize); | |
3729 | ||
3730 | /* This accounts for setting the checksum/compression. */ | |
3731 | dmu_tx_hold_bonus(tx, bigobj); | |
3732 | ||
3733 | txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); | |
3734 | if (txg == 0) { | |
3735 | umem_free(packbuf, packsize); | |
3736 | umem_free(bigbuf, bigsize); | |
3737 | umem_free(od, size); | |
3738 | return; | |
3739 | } | |
3740 | ||
3741 | dmu_object_set_checksum(os, bigobj, | |
3742 | (enum zio_checksum)ztest_random_dsl_prop(ZFS_PROP_CHECKSUM), tx); | |
3743 | ||
3744 | dmu_object_set_compress(os, bigobj, | |
3745 | (enum zio_compress)ztest_random_dsl_prop(ZFS_PROP_COMPRESSION), tx); | |
3746 | ||
3747 | /* | |
3748 | * For each index from n to n + s, verify that the existing bufwad | |
3749 | * in packobj matches the bufwads at the head and tail of the | |
3750 | * corresponding chunk in bigobj. Then update all three bufwads | |
3751 | * with the new values we want to write out. | |
3752 | */ | |
3753 | for (i = 0; i < s; i++) { | |
3754 | /* LINTED */ | |
3755 | pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t)); | |
3756 | /* LINTED */ | |
3757 | bigH = (bufwad_t *)((char *)bigbuf + i * chunksize); | |
3758 | /* LINTED */ | |
3759 | bigT = (bufwad_t *)((char *)bigH + chunksize) - 1; | |
3760 | ||
3761 | ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize); | |
3762 | ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize); | |
3763 | ||
3764 | if (pack->bw_txg > txg) | |
3765 | fatal(0, "future leak: got %llx, open txg is %llx", | |
3766 | pack->bw_txg, txg); | |
3767 | ||
3768 | if (pack->bw_data != 0 && pack->bw_index != n + i) | |
3769 | fatal(0, "wrong index: got %llx, wanted %llx+%llx", | |
3770 | pack->bw_index, n, i); | |
3771 | ||
3772 | if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0) | |
3773 | fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH); | |
3774 | ||
3775 | if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0) | |
3776 | fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT); | |
3777 | ||
3778 | if (freeit) { | |
3779 | bzero(pack, sizeof (bufwad_t)); | |
3780 | } else { | |
3781 | pack->bw_index = n + i; | |
3782 | pack->bw_txg = txg; | |
3783 | pack->bw_data = 1 + ztest_random(-2ULL); | |
3784 | } | |
3785 | *bigH = *pack; | |
3786 | *bigT = *pack; | |
3787 | } | |
3788 | ||
3789 | /* | |
3790 | * We've verified all the old bufwads, and made new ones. | |
3791 | * Now write them out. | |
3792 | */ | |
3793 | dmu_write(os, packobj, packoff, packsize, packbuf, tx); | |
3794 | ||
3795 | if (freeit) { | |
3796 | if (ztest_opts.zo_verbose >= 7) { | |
3797 | (void) printf("freeing offset %llx size %llx" | |
3798 | " txg %llx\n", | |
3799 | (u_longlong_t)bigoff, | |
3800 | (u_longlong_t)bigsize, | |
3801 | (u_longlong_t)txg); | |
3802 | } | |
3803 | VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx)); | |
3804 | } else { | |
3805 | if (ztest_opts.zo_verbose >= 7) { | |
3806 | (void) printf("writing offset %llx size %llx" | |
3807 | " txg %llx\n", | |
3808 | (u_longlong_t)bigoff, | |
3809 | (u_longlong_t)bigsize, | |
3810 | (u_longlong_t)txg); | |
3811 | } | |
3812 | dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx); | |
3813 | } | |
3814 | ||
3815 | dmu_tx_commit(tx); | |
3816 | ||
3817 | /* | |
3818 | * Sanity check the stuff we just wrote. | |
3819 | */ | |
3820 | { | |
3821 | void *packcheck = umem_alloc(packsize, UMEM_NOFAIL); | |
3822 | void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL); | |
3823 | ||
3824 | VERIFY(0 == dmu_read(os, packobj, packoff, | |
3825 | packsize, packcheck, DMU_READ_PREFETCH)); | |
3826 | VERIFY(0 == dmu_read(os, bigobj, bigoff, | |
3827 | bigsize, bigcheck, DMU_READ_PREFETCH)); | |
3828 | ||
3829 | ASSERT(bcmp(packbuf, packcheck, packsize) == 0); | |
3830 | ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0); | |
3831 | ||
3832 | umem_free(packcheck, packsize); | |
3833 | umem_free(bigcheck, bigsize); | |
3834 | } | |
3835 | ||
3836 | umem_free(packbuf, packsize); | |
3837 | umem_free(bigbuf, bigsize); | |
3838 | umem_free(od, size); | |
3839 | } | |
3840 | ||
3841 | void | |
3842 | compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf, | |
3843 | uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg) | |
3844 | { | |
3845 | uint64_t i; | |
3846 | bufwad_t *pack; | |
3847 | bufwad_t *bigH; | |
3848 | bufwad_t *bigT; | |
3849 | ||
3850 | /* | |
3851 | * For each index from n to n + s, verify that the existing bufwad | |
3852 | * in packobj matches the bufwads at the head and tail of the | |
3853 | * corresponding chunk in bigobj. Then update all three bufwads | |
3854 | * with the new values we want to write out. | |
3855 | */ | |
3856 | for (i = 0; i < s; i++) { | |
3857 | /* LINTED */ | |
3858 | pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t)); | |
3859 | /* LINTED */ | |
3860 | bigH = (bufwad_t *)((char *)bigbuf + i * chunksize); | |
3861 | /* LINTED */ | |
3862 | bigT = (bufwad_t *)((char *)bigH + chunksize) - 1; | |
3863 | ||
3864 | ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize); | |
3865 | ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize); | |
3866 | ||
3867 | if (pack->bw_txg > txg) | |
3868 | fatal(0, "future leak: got %llx, open txg is %llx", | |
3869 | pack->bw_txg, txg); | |
3870 | ||
3871 | if (pack->bw_data != 0 && pack->bw_index != n + i) | |
3872 | fatal(0, "wrong index: got %llx, wanted %llx+%llx", | |
3873 | pack->bw_index, n, i); | |
3874 | ||
3875 | if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0) | |
3876 | fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH); | |
3877 | ||
3878 | if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0) | |
3879 | fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT); | |
3880 | ||
3881 | pack->bw_index = n + i; | |
3882 | pack->bw_txg = txg; | |
3883 | pack->bw_data = 1 + ztest_random(-2ULL); | |
3884 | ||
3885 | *bigH = *pack; | |
3886 | *bigT = *pack; | |
3887 | } | |
3888 | } | |
3889 | ||
3890 | #undef OD_ARRAY_SIZE | |
3891 | #define OD_ARRAY_SIZE 2 | |
3892 | ||
3893 | void | |
3894 | ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id) | |
3895 | { | |
3896 | objset_t *os = zd->zd_os; | |
3897 | ztest_od_t *od; | |
3898 | dmu_tx_t *tx; | |
3899 | uint64_t i; | |
3900 | int error; | |
3901 | int size; | |
3902 | uint64_t n, s, txg; | |
3903 | bufwad_t *packbuf, *bigbuf; | |
3904 | uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize; | |
3905 | uint64_t blocksize = ztest_random_blocksize(); | |
3906 | uint64_t chunksize = blocksize; | |
3907 | uint64_t regions = 997; | |
3908 | uint64_t stride = 123456789ULL; | |
3909 | uint64_t width = 9; | |
3910 | dmu_buf_t *bonus_db; | |
3911 | arc_buf_t **bigbuf_arcbufs; | |
3912 | dmu_object_info_t doi; | |
3913 | ||
3914 | size = sizeof (ztest_od_t) * OD_ARRAY_SIZE; | |
3915 | od = umem_alloc(size, UMEM_NOFAIL); | |
3916 | ||
3917 | /* | |
3918 | * This test uses two objects, packobj and bigobj, that are always | |
3919 | * updated together (i.e. in the same tx) so that their contents are | |
3920 | * in sync and can be compared. Their contents relate to each other | |
3921 | * in a simple way: packobj is a dense array of 'bufwad' structures, | |
3922 | * while bigobj is a sparse array of the same bufwads. Specifically, | |
3923 | * for any index n, there are three bufwads that should be identical: | |
3924 | * | |
3925 | * packobj, at offset n * sizeof (bufwad_t) | |
3926 | * bigobj, at the head of the nth chunk | |
3927 | * bigobj, at the tail of the nth chunk | |
3928 | * | |
3929 | * The chunk size is set equal to bigobj block size so that | |
3930 | * dmu_assign_arcbuf() can be tested for object updates. | |
3931 | */ | |
3932 | ||
3933 | /* | |
3934 | * Read the directory info. If it's the first time, set things up. | |
3935 | */ | |
3936 | ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0); | |
3937 | ztest_od_init(od + 1, id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize); | |
3938 | ||
3939 | ||
3940 | if (ztest_object_init(zd, od, size, B_FALSE) != 0) { | |
3941 | umem_free(od, size); | |
3942 | return; | |
3943 | } | |
3944 | ||
3945 | bigobj = od[0].od_object; | |
3946 | packobj = od[1].od_object; | |
3947 | blocksize = od[0].od_blocksize; | |
3948 | chunksize = blocksize; | |
3949 | ASSERT(chunksize == od[1].od_gen); | |
3950 | ||
3951 | VERIFY(dmu_object_info(os, bigobj, &doi) == 0); | |
3952 | VERIFY(ISP2(doi.doi_data_block_size)); | |
3953 | VERIFY(chunksize == doi.doi_data_block_size); | |
3954 | VERIFY(chunksize >= 2 * sizeof (bufwad_t)); | |
3955 | ||
3956 | /* | |
3957 | * Pick a random index and compute the offsets into packobj and bigobj. | |
3958 | */ | |
3959 | n = ztest_random(regions) * stride + ztest_random(width); | |
3960 | s = 1 + ztest_random(width - 1); | |
3961 | ||
3962 | packoff = n * sizeof (bufwad_t); | |
3963 | packsize = s * sizeof (bufwad_t); | |
3964 | ||
3965 | bigoff = n * chunksize; | |
3966 | bigsize = s * chunksize; | |
3967 | ||
3968 | packbuf = umem_zalloc(packsize, UMEM_NOFAIL); | |
3969 | bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL); | |
3970 | ||
3971 | VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db)); | |
3972 | ||
3973 | bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL); | |
3974 | ||
3975 | /* | |
3976 | * Iteration 0 test zcopy for DB_UNCACHED dbufs. | |
3977 | * Iteration 1 test zcopy to already referenced dbufs. | |
3978 | * Iteration 2 test zcopy to dirty dbuf in the same txg. | |
3979 | * Iteration 3 test zcopy to dbuf dirty in previous txg. | |
3980 | * Iteration 4 test zcopy when dbuf is no longer dirty. | |
3981 | * Iteration 5 test zcopy when it can't be done. | |
3982 | * Iteration 6 one more zcopy write. | |
3983 | */ | |
3984 | for (i = 0; i < 7; i++) { | |
3985 | uint64_t j; | |
3986 | uint64_t off; | |
3987 | ||
3988 | /* | |
3989 | * In iteration 5 (i == 5) use arcbufs | |
3990 | * that don't match bigobj blksz to test | |
3991 | * dmu_assign_arcbuf() when it can't directly | |
3992 | * assign an arcbuf to a dbuf. | |
3993 | */ | |
3994 | for (j = 0; j < s; j++) { | |
3995 | if (i != 5) { | |
3996 | bigbuf_arcbufs[j] = | |
3997 | dmu_request_arcbuf(bonus_db, chunksize); | |
3998 | } else { | |
3999 | bigbuf_arcbufs[2 * j] = | |
4000 | dmu_request_arcbuf(bonus_db, chunksize / 2); | |
4001 | bigbuf_arcbufs[2 * j + 1] = | |
4002 | dmu_request_arcbuf(bonus_db, chunksize / 2); | |
4003 | } | |
4004 | } | |
4005 | ||
4006 | /* | |
4007 | * Get a tx for the mods to both packobj and bigobj. | |
4008 | */ | |
4009 | tx = dmu_tx_create(os); | |
4010 | ||
4011 | dmu_tx_hold_write(tx, packobj, packoff, packsize); | |
4012 | dmu_tx_hold_write(tx, bigobj, bigoff, bigsize); | |
4013 | ||
4014 | txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); | |
4015 | if (txg == 0) { | |
4016 | umem_free(packbuf, packsize); | |
4017 | umem_free(bigbuf, bigsize); | |
4018 | for (j = 0; j < s; j++) { | |
4019 | if (i != 5) { | |
4020 | dmu_return_arcbuf(bigbuf_arcbufs[j]); | |
4021 | } else { | |
4022 | dmu_return_arcbuf( | |
4023 | bigbuf_arcbufs[2 * j]); | |
4024 | dmu_return_arcbuf( | |
4025 | bigbuf_arcbufs[2 * j + 1]); | |
4026 | } | |
4027 | } | |
4028 | umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *)); | |
4029 | umem_free(od, size); | |
4030 | dmu_buf_rele(bonus_db, FTAG); | |
4031 | return; | |
4032 | } | |
4033 | ||
4034 | /* | |
4035 | * 50% of the time don't read objects in the 1st iteration to | |
4036 | * test dmu_assign_arcbuf() for the case when there're no | |
4037 | * existing dbufs for the specified offsets. | |
4038 | */ | |
4039 | if (i != 0 || ztest_random(2) != 0) { | |
4040 | error = dmu_read(os, packobj, packoff, | |
4041 | packsize, packbuf, DMU_READ_PREFETCH); | |
4042 | ASSERT0(error); | |
4043 | error = dmu_read(os, bigobj, bigoff, bigsize, | |
4044 | bigbuf, DMU_READ_PREFETCH); | |
4045 | ASSERT0(error); | |
4046 | } | |
4047 | compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize, | |
4048 | n, chunksize, txg); | |
4049 | ||
4050 | /* | |
4051 | * We've verified all the old bufwads, and made new ones. | |
4052 | * Now write them out. | |
4053 | */ | |
4054 | dmu_write(os, packobj, packoff, packsize, packbuf, tx); | |
4055 | if (ztest_opts.zo_verbose >= 7) { | |
4056 | (void) printf("writing offset %llx size %llx" | |
4057 | " txg %llx\n", | |
4058 | (u_longlong_t)bigoff, | |
4059 | (u_longlong_t)bigsize, | |
4060 | (u_longlong_t)txg); | |
4061 | } | |
4062 | for (off = bigoff, j = 0; j < s; j++, off += chunksize) { | |
4063 | dmu_buf_t *dbt; | |
4064 | if (i != 5) { | |
4065 | bcopy((caddr_t)bigbuf + (off - bigoff), | |
4066 | bigbuf_arcbufs[j]->b_data, chunksize); | |
4067 | } else { | |
4068 | bcopy((caddr_t)bigbuf + (off - bigoff), | |
4069 | bigbuf_arcbufs[2 * j]->b_data, | |
4070 | chunksize / 2); | |
4071 | bcopy((caddr_t)bigbuf + (off - bigoff) + | |
4072 | chunksize / 2, | |
4073 | bigbuf_arcbufs[2 * j + 1]->b_data, | |
4074 | chunksize / 2); | |
4075 | } | |
4076 | ||
4077 | if (i == 1) { | |
4078 | VERIFY(dmu_buf_hold(os, bigobj, off, | |
4079 | FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0); | |
4080 | } | |
4081 | if (i != 5) { | |
4082 | dmu_assign_arcbuf(bonus_db, off, | |
4083 | bigbuf_arcbufs[j], tx); | |
4084 | } else { | |
4085 | dmu_assign_arcbuf(bonus_db, off, | |
4086 | bigbuf_arcbufs[2 * j], tx); | |
4087 | dmu_assign_arcbuf(bonus_db, | |
4088 | off + chunksize / 2, | |
4089 | bigbuf_arcbufs[2 * j + 1], tx); | |
4090 | } | |
4091 | if (i == 1) { | |
4092 | dmu_buf_rele(dbt, FTAG); | |
4093 | } | |
4094 | } | |
4095 | dmu_tx_commit(tx); | |
4096 | ||
4097 | /* | |
4098 | * Sanity check the stuff we just wrote. | |
4099 | */ | |
4100 | { | |
4101 | void *packcheck = umem_alloc(packsize, UMEM_NOFAIL); | |
4102 | void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL); | |
4103 | ||
4104 | VERIFY(0 == dmu_read(os, packobj, packoff, | |
4105 | packsize, packcheck, DMU_READ_PREFETCH)); | |
4106 | VERIFY(0 == dmu_read(os, bigobj, bigoff, | |
4107 | bigsize, bigcheck, DMU_READ_PREFETCH)); | |
4108 | ||
4109 | ASSERT(bcmp(packbuf, packcheck, packsize) == 0); | |
4110 | ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0); | |
4111 | ||
4112 | umem_free(packcheck, packsize); | |
4113 | umem_free(bigcheck, bigsize); | |
4114 | } | |
4115 | if (i == 2) { | |
4116 | txg_wait_open(dmu_objset_pool(os), 0); | |
4117 | } else if (i == 3) { | |
4118 | txg_wait_synced(dmu_objset_pool(os), 0); | |
4119 | } | |
4120 | } | |
4121 | ||
4122 | dmu_buf_rele(bonus_db, FTAG); | |
4123 | umem_free(packbuf, packsize); | |
4124 | umem_free(bigbuf, bigsize); | |
4125 | umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *)); | |
4126 | umem_free(od, size); | |
4127 | } | |
4128 | ||
4129 | /* ARGSUSED */ | |
4130 | void | |
4131 | ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id) | |
4132 | { | |
4133 | ztest_od_t *od; | |
4134 | ||
4135 | od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL); | |
4136 | uint64_t offset = (1ULL << (ztest_random(20) + 43)) + | |
4137 | (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT); | |
4138 | ||
4139 | /* | |
4140 | * Have multiple threads write to large offsets in an object | |
4141 | * to verify that parallel writes to an object -- even to the | |
4142 | * same blocks within the object -- doesn't cause any trouble. | |
4143 | */ | |
4144 | ztest_od_init(od, ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0); | |
4145 | ||
4146 | if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) | |
4147 | return; | |
4148 | ||
4149 | while (ztest_random(10) != 0) | |
4150 | ztest_io(zd, od->od_object, offset); | |
4151 | ||
4152 | umem_free(od, sizeof (ztest_od_t)); | |
4153 | } | |
4154 | ||
4155 | void | |
4156 | ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id) | |
4157 | { | |
4158 | ztest_od_t *od; | |
4159 | uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) + | |
4160 | (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT); | |
4161 | uint64_t count = ztest_random(20) + 1; | |
4162 | uint64_t blocksize = ztest_random_blocksize(); | |
4163 | void *data; | |
4164 | ||
4165 | od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL); | |
4166 | ||
4167 | ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0); | |
4168 | ||
4169 | if (ztest_object_init(zd, od, sizeof (ztest_od_t), | |
4170 | !ztest_random(2)) != 0) { | |
4171 | umem_free(od, sizeof (ztest_od_t)); | |
4172 | return; | |
4173 | } | |
4174 | ||
4175 | if (ztest_truncate(zd, od->od_object, offset, count * blocksize) != 0) { | |
4176 | umem_free(od, sizeof (ztest_od_t)); | |
4177 | return; | |
4178 | } | |
4179 | ||
4180 | ztest_prealloc(zd, od->od_object, offset, count * blocksize); | |
4181 | ||
4182 | data = umem_zalloc(blocksize, UMEM_NOFAIL); | |
4183 | ||
4184 | while (ztest_random(count) != 0) { | |
4185 | uint64_t randoff = offset + (ztest_random(count) * blocksize); | |
4186 | if (ztest_write(zd, od->od_object, randoff, blocksize, | |
4187 | data) != 0) | |
4188 | break; | |
4189 | while (ztest_random(4) != 0) | |
4190 | ztest_io(zd, od->od_object, randoff); | |
4191 | } | |
4192 | ||
4193 | umem_free(data, blocksize); | |
4194 | umem_free(od, sizeof (ztest_od_t)); | |
4195 | } | |
4196 | ||
4197 | /* | |
4198 | * Verify that zap_{create,destroy,add,remove,update} work as expected. | |
4199 | */ | |
4200 | #define ZTEST_ZAP_MIN_INTS 1 | |
4201 | #define ZTEST_ZAP_MAX_INTS 4 | |
4202 | #define ZTEST_ZAP_MAX_PROPS 1000 | |
4203 | ||
4204 | void | |
4205 | ztest_zap(ztest_ds_t *zd, uint64_t id) | |
4206 | { | |
4207 | objset_t *os = zd->zd_os; | |
4208 | ztest_od_t *od; | |
4209 | uint64_t object; | |
4210 | uint64_t txg, last_txg; | |
4211 | uint64_t value[ZTEST_ZAP_MAX_INTS]; | |
4212 | uint64_t zl_ints, zl_intsize, prop; | |
4213 | int i, ints; | |
4214 | dmu_tx_t *tx; | |
4215 | char propname[100], txgname[100]; | |
4216 | int error; | |
4217 | char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" }; | |
4218 | ||
4219 | od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL); | |
4220 | ztest_od_init(od, id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0); | |
4221 | ||
4222 | if (ztest_object_init(zd, od, sizeof (ztest_od_t), | |
4223 | !ztest_random(2)) != 0) | |
4224 | goto out; | |
4225 | ||
4226 | object = od->od_object; | |
4227 | ||
4228 | /* | |
4229 | * Generate a known hash collision, and verify that | |
4230 | * we can lookup and remove both entries. | |
4231 | */ | |
4232 | tx = dmu_tx_create(os); | |
4233 | dmu_tx_hold_zap(tx, object, B_TRUE, NULL); | |
4234 | txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); | |
4235 | if (txg == 0) | |
4236 | goto out; | |
4237 | for (i = 0; i < 2; i++) { | |
4238 | value[i] = i; | |
4239 | VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t), | |
4240 | 1, &value[i], tx)); | |
4241 | } | |
4242 | for (i = 0; i < 2; i++) { | |
4243 | VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i], | |
4244 | sizeof (uint64_t), 1, &value[i], tx)); | |
4245 | VERIFY3U(0, ==, | |
4246 | zap_length(os, object, hc[i], &zl_intsize, &zl_ints)); | |
4247 | ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); | |
4248 | ASSERT3U(zl_ints, ==, 1); | |
4249 | } | |
4250 | for (i = 0; i < 2; i++) { | |
4251 | VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx)); | |
4252 | } | |
4253 | dmu_tx_commit(tx); | |
4254 | ||
4255 | /* | |
4256 | * Generate a buch of random entries. | |
4257 | */ | |
4258 | ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS); | |
4259 | ||
4260 | prop = ztest_random(ZTEST_ZAP_MAX_PROPS); | |
4261 | (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop); | |
4262 | (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop); | |
4263 | bzero(value, sizeof (value)); | |
4264 | last_txg = 0; | |
4265 | ||
4266 | /* | |
4267 | * If these zap entries already exist, validate their contents. | |
4268 | */ | |
4269 | error = zap_length(os, object, txgname, &zl_intsize, &zl_ints); | |
4270 | if (error == 0) { | |
4271 | ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); | |
4272 | ASSERT3U(zl_ints, ==, 1); | |
4273 | ||
4274 | VERIFY(zap_lookup(os, object, txgname, zl_intsize, | |
4275 | zl_ints, &last_txg) == 0); | |
4276 | ||
4277 | VERIFY(zap_length(os, object, propname, &zl_intsize, | |
4278 | &zl_ints) == 0); | |
4279 | ||
4280 | ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); | |
4281 | ASSERT3U(zl_ints, ==, ints); | |
4282 | ||
4283 | VERIFY(zap_lookup(os, object, propname, zl_intsize, | |
4284 | zl_ints, value) == 0); | |
4285 | ||
4286 | for (i = 0; i < ints; i++) { | |
4287 | ASSERT3U(value[i], ==, last_txg + object + i); | |
4288 | } | |
4289 | } else { | |
4290 | ASSERT3U(error, ==, ENOENT); | |
4291 | } | |
4292 | ||
4293 | /* | |
4294 | * Atomically update two entries in our zap object. | |
4295 | * The first is named txg_%llu, and contains the txg | |
4296 | * in which the property was last updated. The second | |
4297 | * is named prop_%llu, and the nth element of its value | |
4298 | * should be txg + object + n. | |
4299 | */ | |
4300 | tx = dmu_tx_create(os); | |
4301 | dmu_tx_hold_zap(tx, object, B_TRUE, NULL); | |
4302 | txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); | |
4303 | if (txg == 0) | |
4304 | goto out; | |
4305 | ||
4306 | if (last_txg > txg) | |
4307 | fatal(0, "zap future leak: old %llu new %llu", last_txg, txg); | |
4308 | ||
4309 | for (i = 0; i < ints; i++) | |
4310 | value[i] = txg + object + i; | |
4311 | ||
4312 | VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t), | |
4313 | 1, &txg, tx)); | |
4314 | VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t), | |
4315 | ints, value, tx)); | |
4316 | ||
4317 | dmu_tx_commit(tx); | |
4318 | ||
4319 | /* | |
4320 | * Remove a random pair of entries. | |
4321 | */ | |
4322 | prop = ztest_random(ZTEST_ZAP_MAX_PROPS); | |
4323 | (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop); | |
4324 | (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop); | |
4325 | ||
4326 | error = zap_length(os, object, txgname, &zl_intsize, &zl_ints); | |
4327 | ||
4328 | if (error == ENOENT) | |
4329 | goto out; | |
4330 | ||
4331 | ASSERT0(error); | |
4332 | ||
4333 | tx = dmu_tx_create(os); | |
4334 | dmu_tx_hold_zap(tx, object, B_TRUE, NULL); | |
4335 | txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); | |
4336 | if (txg == 0) | |
4337 | goto out; | |
4338 | VERIFY3U(0, ==, zap_remove(os, object, txgname, tx)); | |
4339 | VERIFY3U(0, ==, zap_remove(os, object, propname, tx)); | |
4340 | dmu_tx_commit(tx); | |
4341 | out: | |
4342 | umem_free(od, sizeof (ztest_od_t)); | |
4343 | } | |
4344 | ||
4345 | /* | |
4346 | * Testcase to test the upgrading of a microzap to fatzap. | |
4347 | */ | |
4348 | void | |
4349 | ztest_fzap(ztest_ds_t *zd, uint64_t id) | |
4350 | { | |
4351 | objset_t *os = zd->zd_os; | |
4352 | ztest_od_t *od; | |
4353 | uint64_t object, txg; | |
4354 | int i; | |
4355 | ||
4356 | od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL); | |
4357 | ztest_od_init(od, id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0); | |
4358 | ||
4359 | if (ztest_object_init(zd, od, sizeof (ztest_od_t), | |
4360 | !ztest_random(2)) != 0) | |
4361 | goto out; | |
4362 | object = od->od_object; | |
4363 | ||
4364 | /* | |
4365 | * Add entries to this ZAP and make sure it spills over | |
4366 | * and gets upgraded to a fatzap. Also, since we are adding | |
4367 | * 2050 entries we should see ptrtbl growth and leaf-block split. | |
4368 | */ | |
4369 | for (i = 0; i < 2050; i++) { | |
4370 | char name[MAXNAMELEN]; | |
4371 | uint64_t value = i; | |
4372 | dmu_tx_t *tx; | |
4373 | int error; | |
4374 | ||
4375 | (void) snprintf(name, sizeof (name), "fzap-%llu-%llu", | |
4376 | (u_longlong_t)id, (u_longlong_t)value); | |
4377 | ||
4378 | tx = dmu_tx_create(os); | |
4379 | dmu_tx_hold_zap(tx, object, B_TRUE, name); | |
4380 | txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); | |
4381 | if (txg == 0) | |
4382 | goto out; | |
4383 | error = zap_add(os, object, name, sizeof (uint64_t), 1, | |
4384 | &value, tx); | |
4385 | ASSERT(error == 0 || error == EEXIST); | |
4386 | dmu_tx_commit(tx); | |
4387 | } | |
4388 | out: | |
4389 | umem_free(od, sizeof (ztest_od_t)); | |
4390 | } | |
4391 | ||
4392 | /* ARGSUSED */ | |
4393 | void | |
4394 | ztest_zap_parallel(ztest_ds_t *zd, uint64_t id) | |
4395 | { | |
4396 | objset_t *os = zd->zd_os; | |
4397 | ztest_od_t *od; | |
4398 | uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc; | |
4399 | dmu_tx_t *tx; | |
4400 | int i, namelen, error; | |
4401 | int micro = ztest_random(2); | |
4402 | char name[20], string_value[20]; | |
4403 | void *data; | |
4404 | ||
4405 | od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL); | |
4406 | ztest_od_init(od, ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0); | |
4407 | ||
4408 | if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) { | |
4409 | umem_free(od, sizeof (ztest_od_t)); | |
4410 | return; | |
4411 | } | |
4412 | ||
4413 | object = od->od_object; | |
4414 | ||
4415 | /* | |
4416 | * Generate a random name of the form 'xxx.....' where each | |
4417 | * x is a random printable character and the dots are dots. | |
4418 | * There are 94 such characters, and the name length goes from | |
4419 | * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names. | |
4420 | */ | |
4421 | namelen = ztest_random(sizeof (name) - 5) + 5 + 1; | |
4422 | ||
4423 | for (i = 0; i < 3; i++) | |
4424 | name[i] = '!' + ztest_random('~' - '!' + 1); | |
4425 | for (; i < namelen - 1; i++) | |
4426 | name[i] = '.'; | |
4427 | name[i] = '\0'; | |
4428 | ||
4429 | if ((namelen & 1) || micro) { | |
4430 | wsize = sizeof (txg); | |
4431 | wc = 1; | |
4432 | data = &txg; | |
4433 | } else { | |
4434 | wsize = 1; | |
4435 | wc = namelen; | |
4436 | data = string_value; | |
4437 | } | |
4438 | ||
4439 | count = -1ULL; | |
4440 | VERIFY0(zap_count(os, object, &count)); | |
4441 | ASSERT(count != -1ULL); | |
4442 | ||
4443 | /* | |
4444 | * Select an operation: length, lookup, add, update, remove. | |
4445 | */ | |
4446 | i = ztest_random(5); | |
4447 | ||
4448 | if (i >= 2) { | |
4449 | tx = dmu_tx_create(os); | |
4450 | dmu_tx_hold_zap(tx, object, B_TRUE, NULL); | |
4451 | txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); | |
4452 | if (txg == 0) | |
4453 | return; | |
4454 | bcopy(name, string_value, namelen); | |
4455 | } else { | |
4456 | tx = NULL; | |
4457 | txg = 0; | |
4458 | bzero(string_value, namelen); | |
4459 | } | |
4460 | ||
4461 | switch (i) { | |
4462 | ||
4463 | case 0: | |
4464 | error = zap_length(os, object, name, &zl_wsize, &zl_wc); | |
4465 | if (error == 0) { | |
4466 | ASSERT3U(wsize, ==, zl_wsize); | |
4467 | ASSERT3U(wc, ==, zl_wc); | |
4468 | } else { | |
4469 | ASSERT3U(error, ==, ENOENT); | |
4470 | } | |
4471 | break; | |
4472 | ||
4473 | case 1: | |
4474 | error = zap_lookup(os, object, name, wsize, wc, data); | |
4475 | if (error == 0) { | |
4476 | if (data == string_value && | |
4477 | bcmp(name, data, namelen) != 0) | |
4478 | fatal(0, "name '%s' != val '%s' len %d", | |
4479 | name, data, namelen); | |
4480 | } else { | |
4481 | ASSERT3U(error, ==, ENOENT); | |
4482 | } | |
4483 | break; | |
4484 | ||
4485 | case 2: | |
4486 | error = zap_add(os, object, name, wsize, wc, data, tx); | |
4487 | ASSERT(error == 0 || error == EEXIST); | |
4488 | break; | |
4489 | ||
4490 | case 3: | |
4491 | VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0); | |
4492 | break; | |
4493 | ||
4494 | case 4: | |
4495 | error = zap_remove(os, object, name, tx); | |
4496 | ASSERT(error == 0 || error == ENOENT); | |
4497 | break; | |
4498 | } | |
4499 | ||
4500 | if (tx != NULL) | |
4501 | dmu_tx_commit(tx); | |
4502 | ||
4503 | umem_free(od, sizeof (ztest_od_t)); | |
4504 | } | |
4505 | ||
4506 | /* | |
4507 | * Commit callback data. | |
4508 | */ | |
4509 | typedef struct ztest_cb_data { | |
4510 | list_node_t zcd_node; | |
4511 | uint64_t zcd_txg; | |
4512 | int zcd_expected_err; | |
4513 | boolean_t zcd_added; | |
4514 | boolean_t zcd_called; | |
4515 | spa_t *zcd_spa; | |
4516 | } ztest_cb_data_t; | |
4517 | ||
4518 | /* This is the actual commit callback function */ | |
4519 | static void | |
4520 | ztest_commit_callback(void *arg, int error) | |
4521 | { | |
4522 | ztest_cb_data_t *data = arg; | |
4523 | uint64_t synced_txg; | |
4524 | ||
4525 | VERIFY(data != NULL); | |
4526 | VERIFY3S(data->zcd_expected_err, ==, error); | |
4527 | VERIFY(!data->zcd_called); | |
4528 | ||
4529 | synced_txg = spa_last_synced_txg(data->zcd_spa); | |
4530 | if (data->zcd_txg > synced_txg) | |
4531 | fatal(0, "commit callback of txg %" PRIu64 " called prematurely" | |
4532 | ", last synced txg = %" PRIu64 "\n", data->zcd_txg, | |
4533 | synced_txg); | |
4534 | ||
4535 | data->zcd_called = B_TRUE; | |
4536 | ||
4537 | if (error == ECANCELED) { | |
4538 | ASSERT0(data->zcd_txg); | |
4539 | ASSERT(!data->zcd_added); | |
4540 | ||
4541 | /* | |
4542 | * The private callback data should be destroyed here, but | |
4543 | * since we are going to check the zcd_called field after | |
4544 | * dmu_tx_abort(), we will destroy it there. | |
4545 | */ | |
4546 | return; | |
4547 | } | |
4548 | ||
4549 | ASSERT(data->zcd_added); | |
4550 | ASSERT3U(data->zcd_txg, !=, 0); | |
4551 | ||
4552 | (void) mutex_enter(&zcl.zcl_callbacks_lock); | |
4553 | ||
4554 | /* See if this cb was called more quickly */ | |
4555 | if ((synced_txg - data->zcd_txg) < zc_min_txg_delay) | |
4556 | zc_min_txg_delay = synced_txg - data->zcd_txg; | |
4557 | ||
4558 | /* Remove our callback from the list */ | |
4559 | list_remove(&zcl.zcl_callbacks, data); | |
4560 | ||
4561 | (void) mutex_exit(&zcl.zcl_callbacks_lock); | |
4562 | ||
4563 | umem_free(data, sizeof (ztest_cb_data_t)); | |
4564 | } | |
4565 | ||
4566 | /* Allocate and initialize callback data structure */ | |
4567 | static ztest_cb_data_t * | |
4568 | ztest_create_cb_data(objset_t *os, uint64_t txg) | |
4569 | { | |
4570 | ztest_cb_data_t *cb_data; | |
4571 | ||
4572 | cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL); | |
4573 | ||
4574 | cb_data->zcd_txg = txg; | |
4575 | cb_data->zcd_spa = dmu_objset_spa(os); | |
4576 | list_link_init(&cb_data->zcd_node); | |
4577 | ||
4578 | return (cb_data); | |
4579 | } | |
4580 | ||
4581 | /* | |
4582 | * Commit callback test. | |
4583 | */ | |
4584 | void | |
4585 | ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id) | |
4586 | { | |
4587 | objset_t *os = zd->zd_os; | |
4588 | ztest_od_t *od; | |
4589 | dmu_tx_t *tx; | |
4590 | ztest_cb_data_t *cb_data[3], *tmp_cb; | |
4591 | uint64_t old_txg, txg; | |
4592 | int i, error = 0; | |
4593 | ||
4594 | od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL); | |
4595 | ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0); | |
4596 | ||
4597 | if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) { | |
4598 | umem_free(od, sizeof (ztest_od_t)); | |
4599 | return; | |
4600 | } | |
4601 | ||
4602 | tx = dmu_tx_create(os); | |
4603 | ||
4604 | cb_data[0] = ztest_create_cb_data(os, 0); | |
4605 | dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]); | |
4606 | ||
4607 | dmu_tx_hold_write(tx, od->od_object, 0, sizeof (uint64_t)); | |
4608 | ||
4609 | /* Every once in a while, abort the transaction on purpose */ | |
4610 | if (ztest_random(100) == 0) | |
4611 | error = -1; | |
4612 | ||
4613 | if (!error) | |
4614 | error = dmu_tx_assign(tx, TXG_NOWAIT); | |
4615 | ||
4616 | txg = error ? 0 : dmu_tx_get_txg(tx); | |
4617 | ||
4618 | cb_data[0]->zcd_txg = txg; | |
4619 | cb_data[1] = ztest_create_cb_data(os, txg); | |
4620 | dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]); | |
4621 | ||
4622 | if (error) { | |
4623 | /* | |
4624 | * It's not a strict requirement to call the registered | |
4625 | * callbacks from inside dmu_tx_abort(), but that's what | |
4626 | * it's supposed to happen in the current implementation | |
4627 | * so we will check for that. | |
4628 | */ | |
4629 | for (i = 0; i < 2; i++) { | |
4630 | cb_data[i]->zcd_expected_err = ECANCELED; | |
4631 | VERIFY(!cb_data[i]->zcd_called); | |
4632 | } | |
4633 | ||
4634 | dmu_tx_abort(tx); | |
4635 | ||
4636 | for (i = 0; i < 2; i++) { | |
4637 | VERIFY(cb_data[i]->zcd_called); | |
4638 | umem_free(cb_data[i], sizeof (ztest_cb_data_t)); | |
4639 | } | |
4640 | ||
4641 | umem_free(od, sizeof (ztest_od_t)); | |
4642 | return; | |
4643 | } | |
4644 | ||
4645 | cb_data[2] = ztest_create_cb_data(os, txg); | |
4646 | dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]); | |
4647 | ||
4648 | /* | |
4649 | * Read existing data to make sure there isn't a future leak. | |
4650 | */ | |
4651 | VERIFY(0 == dmu_read(os, od->od_object, 0, sizeof (uint64_t), | |
4652 | &old_txg, DMU_READ_PREFETCH)); | |
4653 | ||
4654 | if (old_txg > txg) | |
4655 | fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64, | |
4656 | old_txg, txg); | |
4657 | ||
4658 | dmu_write(os, od->od_object, 0, sizeof (uint64_t), &txg, tx); | |
4659 | ||
4660 | (void) mutex_enter(&zcl.zcl_callbacks_lock); | |
4661 | ||
4662 | /* | |
4663 | * Since commit callbacks don't have any ordering requirement and since | |
4664 | * it is theoretically possible for a commit callback to be called | |
4665 | * after an arbitrary amount of time has elapsed since its txg has been | |
4666 | * synced, it is difficult to reliably determine whether a commit | |
4667 | * callback hasn't been called due to high load or due to a flawed | |
4668 | * implementation. | |
4669 | * | |
4670 | * In practice, we will assume that if after a certain number of txgs a | |
4671 | * commit callback hasn't been called, then most likely there's an | |
4672 | * implementation bug.. | |
4673 | */ | |
4674 | tmp_cb = list_head(&zcl.zcl_callbacks); | |
4675 | if (tmp_cb != NULL && | |
4676 | tmp_cb->zcd_txg + ZTEST_COMMIT_CB_THRESH < txg) { | |
4677 | fatal(0, "Commit callback threshold exceeded, oldest txg: %" | |
4678 | PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg); | |
4679 | } | |
4680 | ||
4681 | /* | |
4682 | * Let's find the place to insert our callbacks. | |
4683 | * | |
4684 | * Even though the list is ordered by txg, it is possible for the | |
4685 | * insertion point to not be the end because our txg may already be | |
4686 | * quiescing at this point and other callbacks in the open txg | |
4687 | * (from other objsets) may have sneaked in. | |
4688 | */ | |
4689 | tmp_cb = list_tail(&zcl.zcl_callbacks); | |
4690 | while (tmp_cb != NULL && tmp_cb->zcd_txg > txg) | |
4691 | tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb); | |
4692 | ||
4693 | /* Add the 3 callbacks to the list */ | |
4694 | for (i = 0; i < 3; i++) { | |
4695 | if (tmp_cb == NULL) | |
4696 | list_insert_head(&zcl.zcl_callbacks, cb_data[i]); | |
4697 | else | |
4698 | list_insert_after(&zcl.zcl_callbacks, tmp_cb, | |
4699 | cb_data[i]); | |
4700 | ||
4701 | cb_data[i]->zcd_added = B_TRUE; | |
4702 | VERIFY(!cb_data[i]->zcd_called); | |
4703 | ||
4704 | tmp_cb = cb_data[i]; | |
4705 | } | |
4706 | ||
4707 | zc_cb_counter += 3; | |
4708 | ||
4709 | (void) mutex_exit(&zcl.zcl_callbacks_lock); | |
4710 | ||
4711 | dmu_tx_commit(tx); | |
4712 | ||
4713 | umem_free(od, sizeof (ztest_od_t)); | |
4714 | } | |
4715 | ||
4716 | /* ARGSUSED */ | |
4717 | void | |
4718 | ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id) | |
4719 | { | |
4720 | zfs_prop_t proplist[] = { | |
4721 | ZFS_PROP_CHECKSUM, | |
4722 | ZFS_PROP_COMPRESSION, | |
4723 | ZFS_PROP_COPIES, | |
4724 | ZFS_PROP_DEDUP | |
4725 | }; | |
4726 | int p; | |
4727 | ||
4728 | (void) rw_enter(&ztest_name_lock, RW_READER); | |
4729 | ||
4730 | for (p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++) | |
4731 | (void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p], | |
4732 | ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2)); | |
4733 | ||
4734 | (void) rw_exit(&ztest_name_lock); | |
4735 | } | |
4736 | ||
4737 | /* ARGSUSED */ | |
4738 | void | |
4739 | ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id) | |
4740 | { | |
4741 | nvlist_t *props = NULL; | |
4742 | ||
4743 | (void) rw_enter(&ztest_name_lock, RW_READER); | |
4744 | ||
4745 | (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO, | |
4746 | ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN)); | |
4747 | ||
4748 | VERIFY0(spa_prop_get(ztest_spa, &props)); | |
4749 | ||
4750 | if (ztest_opts.zo_verbose >= 6) | |
4751 | dump_nvlist(props, 4); | |
4752 | ||
4753 | nvlist_free(props); | |
4754 | ||
4755 | (void) rw_exit(&ztest_name_lock); | |
4756 | } | |
4757 | ||
4758 | static int | |
4759 | user_release_one(const char *snapname, const char *holdname) | |
4760 | { | |
4761 | nvlist_t *snaps, *holds; | |
4762 | int error; | |
4763 | ||
4764 | snaps = fnvlist_alloc(); | |
4765 | holds = fnvlist_alloc(); | |
4766 | fnvlist_add_boolean(holds, holdname); | |
4767 | fnvlist_add_nvlist(snaps, snapname, holds); | |
4768 | fnvlist_free(holds); | |
4769 | error = dsl_dataset_user_release(snaps, NULL); | |
4770 | fnvlist_free(snaps); | |
4771 | return (error); | |
4772 | } | |
4773 | ||
4774 | /* | |
4775 | * Test snapshot hold/release and deferred destroy. | |
4776 | */ | |
4777 | void | |
4778 | ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id) | |
4779 | { | |
4780 | int error; | |
4781 | objset_t *os = zd->zd_os; | |
4782 | objset_t *origin; | |
4783 | char snapname[100]; | |
4784 | char fullname[100]; | |
4785 | char clonename[100]; | |
4786 | char tag[100]; | |
4787 | char osname[MAXNAMELEN]; | |
4788 | nvlist_t *holds; | |
4789 | ||
4790 | (void) rw_enter(&ztest_name_lock, RW_READER); | |
4791 | ||
4792 | dmu_objset_name(os, osname); | |
4793 | ||
4794 | (void) snprintf(snapname, sizeof (snapname), "sh1_%llu", | |
4795 | (u_longlong_t)id); | |
4796 | (void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname); | |
4797 | (void) snprintf(clonename, sizeof (clonename), | |
4798 | "%s/ch1_%llu", osname, (u_longlong_t)id); | |
4799 | (void) snprintf(tag, sizeof (tag), "tag_%llu", (u_longlong_t)id); | |
4800 | ||
4801 | /* | |
4802 | * Clean up from any previous run. | |
4803 | */ | |
4804 | error = dsl_destroy_head(clonename); | |
4805 | if (error != ENOENT) | |
4806 | ASSERT0(error); | |
4807 | error = user_release_one(fullname, tag); | |
4808 | if (error != ESRCH && error != ENOENT) | |
4809 | ASSERT0(error); | |
4810 | error = dsl_destroy_snapshot(fullname, B_FALSE); | |
4811 | if (error != ENOENT) | |
4812 | ASSERT0(error); | |
4813 | ||
4814 | /* | |
4815 | * Create snapshot, clone it, mark snap for deferred destroy, | |
4816 | * destroy clone, verify snap was also destroyed. | |
4817 | */ | |
4818 | error = dmu_objset_snapshot_one(osname, snapname); | |
4819 | if (error) { | |
4820 | if (error == ENOSPC) { | |
4821 | ztest_record_enospc("dmu_objset_snapshot"); | |
4822 | goto out; | |
4823 | } | |
4824 | fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error); | |
4825 | } | |
4826 | ||
4827 | error = dmu_objset_clone(clonename, fullname); | |
4828 | if (error) { | |
4829 | if (error == ENOSPC) { | |
4830 | ztest_record_enospc("dmu_objset_clone"); | |
4831 | goto out; | |
4832 | } | |
4833 | fatal(0, "dmu_objset_clone(%s) = %d", clonename, error); | |
4834 | } | |
4835 | ||
4836 | error = dsl_destroy_snapshot(fullname, B_TRUE); | |
4837 | if (error) { | |
4838 | fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d", | |
4839 | fullname, error); | |
4840 | } | |
4841 | ||
4842 | error = dsl_destroy_head(clonename); | |
4843 | if (error) | |
4844 | fatal(0, "dsl_destroy_head(%s) = %d", clonename, error); | |
4845 | ||
4846 | error = dmu_objset_hold(fullname, FTAG, &origin); | |
4847 | if (error != ENOENT) | |
4848 | fatal(0, "dmu_objset_hold(%s) = %d", fullname, error); | |
4849 | ||
4850 | /* | |
4851 | * Create snapshot, add temporary hold, verify that we can't | |
4852 | * destroy a held snapshot, mark for deferred destroy, | |
4853 | * release hold, verify snapshot was destroyed. | |
4854 | */ | |
4855 | error = dmu_objset_snapshot_one(osname, snapname); | |
4856 | if (error) { | |
4857 | if (error == ENOSPC) { | |
4858 | ztest_record_enospc("dmu_objset_snapshot"); | |
4859 | goto out; | |
4860 | } | |
4861 | fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error); | |
4862 | } | |
4863 | ||
4864 | holds = fnvlist_alloc(); | |
4865 | fnvlist_add_string(holds, fullname, tag); | |
4866 | error = dsl_dataset_user_hold(holds, 0, NULL); | |
4867 | fnvlist_free(holds); | |
4868 | ||
4869 | if (error) | |
4870 | fatal(0, "dsl_dataset_user_hold(%s)", fullname, tag); | |
4871 | ||
4872 | error = dsl_destroy_snapshot(fullname, B_FALSE); | |
4873 | if (error != EBUSY) { | |
4874 | fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d", | |
4875 | fullname, error); | |
4876 | } | |
4877 | ||
4878 | error = dsl_destroy_snapshot(fullname, B_TRUE); | |
4879 | if (error) { | |
4880 | fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d", | |
4881 | fullname, error); | |
4882 | } | |
4883 | ||
4884 | error = user_release_one(fullname, tag); | |
4885 | if (error) | |
4886 | fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error); | |
4887 | ||
4888 | VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT); | |
4889 | ||
4890 | out: | |
4891 | (void) rw_exit(&ztest_name_lock); | |
4892 | } | |
4893 | ||
4894 | /* | |
4895 | * Inject random faults into the on-disk data. | |
4896 | */ | |
4897 | /* ARGSUSED */ | |
4898 | void | |
4899 | ztest_fault_inject(ztest_ds_t *zd, uint64_t id) | |
4900 | { | |
4901 | ztest_shared_t *zs = ztest_shared; | |
4902 | spa_t *spa = ztest_spa; | |
4903 | int fd; | |
4904 | uint64_t offset; | |
4905 | uint64_t leaves; | |
4906 | uint64_t bad = 0x1990c0ffeedecadeull; | |
4907 | uint64_t top, leaf; | |
4908 | char *path0; | |
4909 | char *pathrand; | |
4910 | size_t fsize; | |
4911 | int bshift = SPA_MAXBLOCKSHIFT + 2; /* don't scrog all labels */ | |
4912 | int iters = 1000; | |
4913 | int maxfaults; | |
4914 | int mirror_save; | |
4915 | vdev_t *vd0 = NULL; | |
4916 | uint64_t guid0 = 0; | |
4917 | boolean_t islog = B_FALSE; | |
4918 | ||
4919 | path0 = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); | |
4920 | pathrand = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); | |
4921 | ||
4922 | mutex_enter(&ztest_vdev_lock); | |
4923 | maxfaults = MAXFAULTS(); | |
4924 | leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz; | |
4925 | mirror_save = zs->zs_mirrors; | |
4926 | mutex_exit(&ztest_vdev_lock); | |
4927 | ||
4928 | ASSERT(leaves >= 1); | |
4929 | ||
4930 | /* | |
4931 | * Grab the name lock as reader. There are some operations | |
4932 | * which don't like to have their vdevs changed while | |
4933 | * they are in progress (i.e. spa_change_guid). Those | |
4934 | * operations will have grabbed the name lock as writer. | |
4935 | */ | |
4936 | (void) rw_enter(&ztest_name_lock, RW_READER); | |
4937 | ||
4938 | /* | |
4939 | * We need SCL_STATE here because we're going to look at vd0->vdev_tsd. | |
4940 | */ | |
4941 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); | |
4942 | ||
4943 | if (ztest_random(2) == 0) { | |
4944 | /* | |
4945 | * Inject errors on a normal data device or slog device. | |
4946 | */ | |
4947 | top = ztest_random_vdev_top(spa, B_TRUE); | |
4948 | leaf = ztest_random(leaves) + zs->zs_splits; | |
4949 | ||
4950 | /* | |
4951 | * Generate paths to the first leaf in this top-level vdev, | |
4952 | * and to the random leaf we selected. We'll induce transient | |
4953 | * write failures and random online/offline activity on leaf 0, | |
4954 | * and we'll write random garbage to the randomly chosen leaf. | |
4955 | */ | |
4956 | (void) snprintf(path0, MAXPATHLEN, ztest_dev_template, | |
4957 | ztest_opts.zo_dir, ztest_opts.zo_pool, | |
4958 | top * leaves + zs->zs_splits); | |
4959 | (void) snprintf(pathrand, MAXPATHLEN, ztest_dev_template, | |
4960 | ztest_opts.zo_dir, ztest_opts.zo_pool, | |
4961 | top * leaves + leaf); | |
4962 | ||
4963 | vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0); | |
4964 | if (vd0 != NULL && vd0->vdev_top->vdev_islog) | |
4965 | islog = B_TRUE; | |
4966 | ||
4967 | /* | |
4968 | * If the top-level vdev needs to be resilvered | |
4969 | * then we only allow faults on the device that is | |
4970 | * resilvering. | |
4971 | */ | |
4972 | if (vd0 != NULL && maxfaults != 1 && | |
4973 | (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) || | |
4974 | vd0->vdev_resilver_txg != 0)) { | |
4975 | /* | |
4976 | * Make vd0 explicitly claim to be unreadable, | |
4977 | * or unwriteable, or reach behind its back | |
4978 | * and close the underlying fd. We can do this if | |
4979 | * maxfaults == 0 because we'll fail and reexecute, | |
4980 | * and we can do it if maxfaults >= 2 because we'll | |
4981 | * have enough redundancy. If maxfaults == 1, the | |
4982 | * combination of this with injection of random data | |
4983 | * corruption below exceeds the pool's fault tolerance. | |
4984 | */ | |
4985 | vdev_file_t *vf = vd0->vdev_tsd; | |
4986 | ||
4987 | if (vf != NULL && ztest_random(3) == 0) { | |
4988 | (void) close(vf->vf_vnode->v_fd); | |
4989 | vf->vf_vnode->v_fd = -1; | |
4990 | } else if (ztest_random(2) == 0) { | |
4991 | vd0->vdev_cant_read = B_TRUE; | |
4992 | } else { | |
4993 | vd0->vdev_cant_write = B_TRUE; | |
4994 | } | |
4995 | guid0 = vd0->vdev_guid; | |
4996 | } | |
4997 | } else { | |
4998 | /* | |
4999 | * Inject errors on an l2cache device. | |
5000 | */ | |
5001 | spa_aux_vdev_t *sav = &spa->spa_l2cache; | |
5002 | ||
5003 | if (sav->sav_count == 0) { | |
5004 | spa_config_exit(spa, SCL_STATE, FTAG); | |
5005 | (void) rw_exit(&ztest_name_lock); | |
5006 | goto out; | |
5007 | } | |
5008 | vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)]; | |
5009 | guid0 = vd0->vdev_guid; | |
5010 | (void) strcpy(path0, vd0->vdev_path); | |
5011 | (void) strcpy(pathrand, vd0->vdev_path); | |
5012 | ||
5013 | leaf = 0; | |
5014 | leaves = 1; | |
5015 | maxfaults = INT_MAX; /* no limit on cache devices */ | |
5016 | } | |
5017 | ||
5018 | spa_config_exit(spa, SCL_STATE, FTAG); | |
5019 | (void) rw_exit(&ztest_name_lock); | |
5020 | ||
5021 | /* | |
5022 | * If we can tolerate two or more faults, or we're dealing | |
5023 | * with a slog, randomly online/offline vd0. | |
5024 | */ | |
5025 | if ((maxfaults >= 2 || islog) && guid0 != 0) { | |
5026 | if (ztest_random(10) < 6) { | |
5027 | int flags = (ztest_random(2) == 0 ? | |
5028 | ZFS_OFFLINE_TEMPORARY : 0); | |
5029 | ||
5030 | /* | |
5031 | * We have to grab the zs_name_lock as writer to | |
5032 | * prevent a race between offlining a slog and | |
5033 | * destroying a dataset. Offlining the slog will | |
5034 | * grab a reference on the dataset which may cause | |
5035 | * dsl_destroy_head() to fail with EBUSY thus | |
5036 | * leaving the dataset in an inconsistent state. | |
5037 | */ | |
5038 | if (islog) | |
5039 | (void) rw_enter(&ztest_name_lock, | |
5040 | RW_WRITER); | |
5041 | ||
5042 | VERIFY(vdev_offline(spa, guid0, flags) != EBUSY); | |
5043 | ||
5044 | if (islog) | |
5045 | (void) rw_exit(&ztest_name_lock); | |
5046 | } else { | |
5047 | /* | |
5048 | * Ideally we would like to be able to randomly | |
5049 | * call vdev_[on|off]line without holding locks | |
5050 | * to force unpredictable failures but the side | |
5051 | * effects of vdev_[on|off]line prevent us from | |
5052 | * doing so. We grab the ztest_vdev_lock here to | |
5053 | * prevent a race between injection testing and | |
5054 | * aux_vdev removal. | |
5055 | */ | |
5056 | mutex_enter(&ztest_vdev_lock); | |
5057 | (void) vdev_online(spa, guid0, 0, NULL); | |
5058 | mutex_exit(&ztest_vdev_lock); | |
5059 | } | |
5060 | } | |
5061 | ||
5062 | if (maxfaults == 0) | |
5063 | goto out; | |
5064 | ||
5065 | /* | |
5066 | * We have at least single-fault tolerance, so inject data corruption. | |
5067 | */ | |
5068 | fd = open(pathrand, O_RDWR); | |
5069 | ||
5070 | if (fd == -1) /* we hit a gap in the device namespace */ | |
5071 | goto out; | |
5072 | ||
5073 | fsize = lseek(fd, 0, SEEK_END); | |
5074 | ||
5075 | while (--iters != 0) { | |
5076 | offset = ztest_random(fsize / (leaves << bshift)) * | |
5077 | (leaves << bshift) + (leaf << bshift) + | |
5078 | (ztest_random(1ULL << (bshift - 1)) & -8ULL); | |
5079 | ||
5080 | if (offset >= fsize) | |
5081 | continue; | |
5082 | ||
5083 | mutex_enter(&ztest_vdev_lock); | |
5084 | if (mirror_save != zs->zs_mirrors) { | |
5085 | mutex_exit(&ztest_vdev_lock); | |
5086 | (void) close(fd); | |
5087 | goto out; | |
5088 | } | |
5089 | ||
5090 | if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad)) | |
5091 | fatal(1, "can't inject bad word at 0x%llx in %s", | |
5092 | offset, pathrand); | |
5093 | ||
5094 | mutex_exit(&ztest_vdev_lock); | |
5095 | ||
5096 | if (ztest_opts.zo_verbose >= 7) | |
5097 | (void) printf("injected bad word into %s," | |
5098 | " offset 0x%llx\n", pathrand, (u_longlong_t)offset); | |
5099 | } | |
5100 | ||
5101 | (void) close(fd); | |
5102 | out: | |
5103 | umem_free(path0, MAXPATHLEN); | |
5104 | umem_free(pathrand, MAXPATHLEN); | |
5105 | } | |
5106 | ||
5107 | /* | |
5108 | * Verify that DDT repair works as expected. | |
5109 | */ | |
5110 | void | |
5111 | ztest_ddt_repair(ztest_ds_t *zd, uint64_t id) | |
5112 | { | |
5113 | ztest_shared_t *zs = ztest_shared; | |
5114 | spa_t *spa = ztest_spa; | |
5115 | objset_t *os = zd->zd_os; | |
5116 | ztest_od_t *od; | |
5117 | uint64_t object, blocksize, txg, pattern, psize; | |
5118 | enum zio_checksum checksum = spa_dedup_checksum(spa); | |
5119 | dmu_buf_t *db; | |
5120 | dmu_tx_t *tx; | |
5121 | void *buf; | |
5122 | blkptr_t blk; | |
5123 | int copies = 2 * ZIO_DEDUPDITTO_MIN; | |
5124 | int i; | |
5125 | ||
5126 | blocksize = ztest_random_blocksize(); | |
5127 | blocksize = MIN(blocksize, 2048); /* because we write so many */ | |
5128 | ||
5129 | od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL); | |
5130 | ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0); | |
5131 | ||
5132 | if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) { | |
5133 | umem_free(od, sizeof (ztest_od_t)); | |
5134 | return; | |
5135 | } | |
5136 | ||
5137 | /* | |
5138 | * Take the name lock as writer to prevent anyone else from changing | |
5139 | * the pool and dataset properies we need to maintain during this test. | |
5140 | */ | |
5141 | (void) rw_enter(&ztest_name_lock, RW_WRITER); | |
5142 | ||
5143 | if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum, | |
5144 | B_FALSE) != 0 || | |
5145 | ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1, | |
5146 | B_FALSE) != 0) { | |
5147 | (void) rw_exit(&ztest_name_lock); | |
5148 | umem_free(od, sizeof (ztest_od_t)); | |
5149 | return; | |
5150 | } | |
5151 | ||
5152 | object = od[0].od_object; | |
5153 | blocksize = od[0].od_blocksize; | |
5154 | pattern = zs->zs_guid ^ dmu_objset_fsid_guid(os); | |
5155 | ||
5156 | ASSERT(object != 0); | |
5157 | ||
5158 | tx = dmu_tx_create(os); | |
5159 | dmu_tx_hold_write(tx, object, 0, copies * blocksize); | |
5160 | txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); | |
5161 | if (txg == 0) { | |
5162 | (void) rw_exit(&ztest_name_lock); | |
5163 | umem_free(od, sizeof (ztest_od_t)); | |
5164 | return; | |
5165 | } | |
5166 | ||
5167 | /* | |
5168 | * Write all the copies of our block. | |
5169 | */ | |
5170 | for (i = 0; i < copies; i++) { | |
5171 | uint64_t offset = i * blocksize; | |
5172 | int error = dmu_buf_hold(os, object, offset, FTAG, &db, | |
5173 | DMU_READ_NO_PREFETCH); | |
5174 | if (error != 0) { | |
5175 | fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u", | |
5176 | os, (long long)object, (long long) offset, error); | |
5177 | } | |
5178 | ASSERT(db->db_offset == offset); | |
5179 | ASSERT(db->db_size == blocksize); | |
5180 | ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) || | |
5181 | ztest_pattern_match(db->db_data, db->db_size, 0ULL)); | |
5182 | dmu_buf_will_fill(db, tx); | |
5183 | ztest_pattern_set(db->db_data, db->db_size, pattern); | |
5184 | dmu_buf_rele(db, FTAG); | |
5185 | } | |
5186 | ||
5187 | dmu_tx_commit(tx); | |
5188 | txg_wait_synced(spa_get_dsl(spa), txg); | |
5189 | ||
5190 | /* | |
5191 | * Find out what block we got. | |
5192 | */ | |
5193 | VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db, | |
5194 | DMU_READ_NO_PREFETCH)); | |
5195 | blk = *((dmu_buf_impl_t *)db)->db_blkptr; | |
5196 | dmu_buf_rele(db, FTAG); | |
5197 | ||
5198 | /* | |
5199 | * Damage the block. Dedup-ditto will save us when we read it later. | |
5200 | */ | |
5201 | psize = BP_GET_PSIZE(&blk); | |
5202 | buf = zio_buf_alloc(psize); | |
5203 | ztest_pattern_set(buf, psize, ~pattern); | |
5204 | ||
5205 | (void) zio_wait(zio_rewrite(NULL, spa, 0, &blk, | |
5206 | buf, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE, | |
5207 | ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL)); | |
5208 | ||
5209 | zio_buf_free(buf, psize); | |
5210 | ||
5211 | (void) rw_exit(&ztest_name_lock); | |
5212 | umem_free(od, sizeof (ztest_od_t)); | |
5213 | } | |
5214 | ||
5215 | /* | |
5216 | * Scrub the pool. | |
5217 | */ | |
5218 | /* ARGSUSED */ | |
5219 | void | |
5220 | ztest_scrub(ztest_ds_t *zd, uint64_t id) | |
5221 | { | |
5222 | spa_t *spa = ztest_spa; | |
5223 | ||
5224 | (void) spa_scan(spa, POOL_SCAN_SCRUB); | |
5225 | (void) poll(NULL, 0, 100); /* wait a moment, then force a restart */ | |
5226 | (void) spa_scan(spa, POOL_SCAN_SCRUB); | |
5227 | } | |
5228 | ||
5229 | /* | |
5230 | * Change the guid for the pool. | |
5231 | */ | |
5232 | /* ARGSUSED */ | |
5233 | void | |
5234 | ztest_reguid(ztest_ds_t *zd, uint64_t id) | |
5235 | { | |
5236 | spa_t *spa = ztest_spa; | |
5237 | uint64_t orig, load; | |
5238 | int error; | |
5239 | ||
5240 | orig = spa_guid(spa); | |
5241 | load = spa_load_guid(spa); | |
5242 | ||
5243 | (void) rw_enter(&ztest_name_lock, RW_WRITER); | |
5244 | error = spa_change_guid(spa); | |
5245 | (void) rw_exit(&ztest_name_lock); | |
5246 | ||
5247 | if (error != 0) | |
5248 | return; | |
5249 | ||
5250 | if (ztest_opts.zo_verbose >= 4) { | |
5251 | (void) printf("Changed guid old %llu -> %llu\n", | |
5252 | (u_longlong_t)orig, (u_longlong_t)spa_guid(spa)); | |
5253 | } | |
5254 | ||
5255 | VERIFY3U(orig, !=, spa_guid(spa)); | |
5256 | VERIFY3U(load, ==, spa_load_guid(spa)); | |
5257 | } | |
5258 | ||
5259 | /* | |
5260 | * Rename the pool to a different name and then rename it back. | |
5261 | */ | |
5262 | /* ARGSUSED */ | |
5263 | void | |
5264 | ztest_spa_rename(ztest_ds_t *zd, uint64_t id) | |
5265 | { | |
5266 | char *oldname, *newname; | |
5267 | spa_t *spa; | |
5268 | ||
5269 | (void) rw_enter(&ztest_name_lock, RW_WRITER); | |
5270 | ||
5271 | oldname = ztest_opts.zo_pool; | |
5272 | newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL); | |
5273 | (void) strcpy(newname, oldname); | |
5274 | (void) strcat(newname, "_tmp"); | |
5275 | ||
5276 | /* | |
5277 | * Do the rename | |
5278 | */ | |
5279 | VERIFY3U(0, ==, spa_rename(oldname, newname)); | |
5280 | ||
5281 | /* | |
5282 | * Try to open it under the old name, which shouldn't exist | |
5283 | */ | |
5284 | VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG)); | |
5285 | ||
5286 | /* | |
5287 | * Open it under the new name and make sure it's still the same spa_t. | |
5288 | */ | |
5289 | VERIFY3U(0, ==, spa_open(newname, &spa, FTAG)); | |
5290 | ||
5291 | ASSERT(spa == ztest_spa); | |
5292 | spa_close(spa, FTAG); | |
5293 | ||
5294 | /* | |
5295 | * Rename it back to the original | |
5296 | */ | |
5297 | VERIFY3U(0, ==, spa_rename(newname, oldname)); | |
5298 | ||
5299 | /* | |
5300 | * Make sure it can still be opened | |
5301 | */ | |
5302 | VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG)); | |
5303 | ||
5304 | ASSERT(spa == ztest_spa); | |
5305 | spa_close(spa, FTAG); | |
5306 | ||
5307 | umem_free(newname, strlen(newname) + 1); | |
5308 | ||
5309 | (void) rw_exit(&ztest_name_lock); | |
5310 | } | |
5311 | ||
5312 | /* | |
5313 | * Verify pool integrity by running zdb. | |
5314 | */ | |
5315 | static void | |
5316 | ztest_run_zdb(char *pool) | |
5317 | { | |
5318 | int status; | |
5319 | char *bin; | |
5320 | char *zdb; | |
5321 | char *zbuf; | |
5322 | FILE *fp; | |
5323 | ||
5324 | bin = umem_alloc(MAXPATHLEN + MAXNAMELEN + 20, UMEM_NOFAIL); | |
5325 | zdb = umem_alloc(MAXPATHLEN + MAXNAMELEN + 20, UMEM_NOFAIL); | |
5326 | zbuf = umem_alloc(1024, UMEM_NOFAIL); | |
5327 | ||
5328 | VERIFY(realpath(getexecname(), bin) != NULL); | |
5329 | if (strncmp(bin, "/usr/sbin/ztest", 15) == 0) { | |
5330 | strcpy(bin, "/usr/sbin/zdb"); /* Installed */ | |
5331 | } else if (strncmp(bin, "/sbin/ztest", 11) == 0) { | |
5332 | strcpy(bin, "/sbin/zdb"); /* Installed */ | |
5333 | } else { | |
5334 | strstr(bin, "/ztest/")[0] = '\0'; /* In-tree */ | |
5335 | strcat(bin, "/zdb/zdb"); | |
5336 | } | |
5337 | ||
5338 | (void) sprintf(zdb, | |
5339 | "%s -bcc%s%s -U %s %s", | |
5340 | bin, | |
5341 | ztest_opts.zo_verbose >= 3 ? "s" : "", | |
5342 | ztest_opts.zo_verbose >= 4 ? "v" : "", | |
5343 | spa_config_path, | |
5344 | pool); | |
5345 | ||
5346 | if (ztest_opts.zo_verbose >= 5) | |
5347 | (void) printf("Executing %s\n", strstr(zdb, "zdb ")); | |
5348 | ||
5349 | fp = popen(zdb, "r"); | |
5350 | ||
5351 | while (fgets(zbuf, 1024, fp) != NULL) | |
5352 | if (ztest_opts.zo_verbose >= 3) | |
5353 | (void) printf("%s", zbuf); | |
5354 | ||
5355 | status = pclose(fp); | |
5356 | ||
5357 | if (status == 0) | |
5358 | goto out; | |
5359 | ||
5360 | ztest_dump_core = 0; | |
5361 | if (WIFEXITED(status)) | |
5362 | fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status)); | |
5363 | else | |
5364 | fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status)); | |
5365 | out: | |
5366 | umem_free(bin, MAXPATHLEN + MAXNAMELEN + 20); | |
5367 | umem_free(zdb, MAXPATHLEN + MAXNAMELEN + 20); | |
5368 | umem_free(zbuf, 1024); | |
5369 | } | |
5370 | ||
5371 | static void | |
5372 | ztest_walk_pool_directory(char *header) | |
5373 | { | |
5374 | spa_t *spa = NULL; | |
5375 | ||
5376 | if (ztest_opts.zo_verbose >= 6) | |
5377 | (void) printf("%s\n", header); | |
5378 | ||
5379 | mutex_enter(&spa_namespace_lock); | |
5380 | while ((spa = spa_next(spa)) != NULL) | |
5381 | if (ztest_opts.zo_verbose >= 6) | |
5382 | (void) printf("\t%s\n", spa_name(spa)); | |
5383 | mutex_exit(&spa_namespace_lock); | |
5384 | } | |
5385 | ||
5386 | static void | |
5387 | ztest_spa_import_export(char *oldname, char *newname) | |
5388 | { | |
5389 | nvlist_t *config, *newconfig; | |
5390 | uint64_t pool_guid; | |
5391 | spa_t *spa; | |
5392 | int error; | |
5393 | ||
5394 | if (ztest_opts.zo_verbose >= 4) { | |
5395 | (void) printf("import/export: old = %s, new = %s\n", | |
5396 | oldname, newname); | |
5397 | } | |
5398 | ||
5399 | /* | |
5400 | * Clean up from previous runs. | |
5401 | */ | |
5402 | (void) spa_destroy(newname); | |
5403 | ||
5404 | /* | |
5405 | * Get the pool's configuration and guid. | |
5406 | */ | |
5407 | VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG)); | |
5408 | ||
5409 | /* | |
5410 | * Kick off a scrub to tickle scrub/export races. | |
5411 | */ | |
5412 | if (ztest_random(2) == 0) | |
5413 | (void) spa_scan(spa, POOL_SCAN_SCRUB); | |
5414 | ||
5415 | pool_guid = spa_guid(spa); | |
5416 | spa_close(spa, FTAG); | |
5417 | ||
5418 | ztest_walk_pool_directory("pools before export"); | |
5419 | ||
5420 | /* | |
5421 | * Export it. | |
5422 | */ | |
5423 | VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE)); | |
5424 | ||
5425 | ztest_walk_pool_directory("pools after export"); | |
5426 | ||
5427 | /* | |
5428 | * Try to import it. | |
5429 | */ | |
5430 | newconfig = spa_tryimport(config); | |
5431 | ASSERT(newconfig != NULL); | |
5432 | nvlist_free(newconfig); | |
5433 | ||
5434 | /* | |
5435 | * Import it under the new name. | |
5436 | */ | |
5437 | error = spa_import(newname, config, NULL, 0); | |
5438 | if (error != 0) { | |
5439 | dump_nvlist(config, 0); | |
5440 | fatal(B_FALSE, "couldn't import pool %s as %s: error %u", | |
5441 | oldname, newname, error); | |
5442 | } | |
5443 | ||
5444 | ztest_walk_pool_directory("pools after import"); | |
5445 | ||
5446 | /* | |
5447 | * Try to import it again -- should fail with EEXIST. | |
5448 | */ | |
5449 | VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0)); | |
5450 | ||
5451 | /* | |
5452 | * Try to import it under a different name -- should fail with EEXIST. | |
5453 | */ | |
5454 | VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0)); | |
5455 | ||
5456 | /* | |
5457 | * Verify that the pool is no longer visible under the old name. | |
5458 | */ | |
5459 | VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG)); | |
5460 | ||
5461 | /* | |
5462 | * Verify that we can open and close the pool using the new name. | |
5463 | */ | |
5464 | VERIFY3U(0, ==, spa_open(newname, &spa, FTAG)); | |
5465 | ASSERT(pool_guid == spa_guid(spa)); | |
5466 | spa_close(spa, FTAG); | |
5467 | ||
5468 | nvlist_free(config); | |
5469 | } | |
5470 | ||
5471 | static void | |
5472 | ztest_resume(spa_t *spa) | |
5473 | { | |
5474 | if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6) | |
5475 | (void) printf("resuming from suspended state\n"); | |
5476 | spa_vdev_state_enter(spa, SCL_NONE); | |
5477 | vdev_clear(spa, NULL); | |
5478 | (void) spa_vdev_state_exit(spa, NULL, 0); | |
5479 | (void) zio_resume(spa); | |
5480 | } | |
5481 | ||
5482 | static void * | |
5483 | ztest_resume_thread(void *arg) | |
5484 | { | |
5485 | spa_t *spa = arg; | |
5486 | ||
5487 | while (!ztest_exiting) { | |
5488 | if (spa_suspended(spa)) | |
5489 | ztest_resume(spa); | |
5490 | (void) poll(NULL, 0, 100); | |
5491 | } | |
5492 | ||
5493 | thread_exit(); | |
5494 | ||
5495 | return (NULL); | |
5496 | } | |
5497 | ||
5498 | #define GRACE 300 | |
5499 | ||
5500 | #if 0 | |
5501 | static void | |
5502 | ztest_deadman_alarm(int sig) | |
5503 | { | |
5504 | fatal(0, "failed to complete within %d seconds of deadline", GRACE); | |
5505 | } | |
5506 | #endif | |
5507 | ||
5508 | static void | |
5509 | ztest_execute(int test, ztest_info_t *zi, uint64_t id) | |
5510 | { | |
5511 | ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets]; | |
5512 | ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test); | |
5513 | hrtime_t functime = gethrtime(); | |
5514 | int i; | |
5515 | ||
5516 | for (i = 0; i < zi->zi_iters; i++) | |
5517 | zi->zi_func(zd, id); | |
5518 | ||
5519 | functime = gethrtime() - functime; | |
5520 | ||
5521 | atomic_add_64(&zc->zc_count, 1); | |
5522 | atomic_add_64(&zc->zc_time, functime); | |
5523 | ||
5524 | if (ztest_opts.zo_verbose >= 4) { | |
5525 | Dl_info dli; | |
5526 | (void) dladdr((void *)zi->zi_func, &dli); | |
5527 | (void) printf("%6.2f sec in %s\n", | |
5528 | (double)functime / NANOSEC, dli.dli_sname); | |
5529 | } | |
5530 | } | |
5531 | ||
5532 | static void * | |
5533 | ztest_thread(void *arg) | |
5534 | { | |
5535 | int rand; | |
5536 | uint64_t id = (uintptr_t)arg; | |
5537 | ztest_shared_t *zs = ztest_shared; | |
5538 | uint64_t call_next; | |
5539 | hrtime_t now; | |
5540 | ztest_info_t *zi; | |
5541 | ztest_shared_callstate_t *zc; | |
5542 | ||
5543 | while ((now = gethrtime()) < zs->zs_thread_stop) { | |
5544 | /* | |
5545 | * See if it's time to force a crash. | |
5546 | */ | |
5547 | if (now > zs->zs_thread_kill) | |
5548 | ztest_kill(zs); | |
5549 | ||
5550 | /* | |
5551 | * If we're getting ENOSPC with some regularity, stop. | |
5552 | */ | |
5553 | if (zs->zs_enospc_count > 10) | |
5554 | break; | |
5555 | ||
5556 | /* | |
5557 | * Pick a random function to execute. | |
5558 | */ | |
5559 | rand = ztest_random(ZTEST_FUNCS); | |
5560 | zi = &ztest_info[rand]; | |
5561 | zc = ZTEST_GET_SHARED_CALLSTATE(rand); | |
5562 | call_next = zc->zc_next; | |
5563 | ||
5564 | if (now >= call_next && | |
5565 | atomic_cas_64(&zc->zc_next, call_next, call_next + | |
5566 | ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) { | |
5567 | ztest_execute(rand, zi, id); | |
5568 | } | |
5569 | } | |
5570 | ||
5571 | thread_exit(); | |
5572 | ||
5573 | return (NULL); | |
5574 | } | |
5575 | ||
5576 | static void | |
5577 | ztest_dataset_name(char *dsname, char *pool, int d) | |
5578 | { | |
5579 | (void) snprintf(dsname, MAXNAMELEN, "%s/ds_%d", pool, d); | |
5580 | } | |
5581 | ||
5582 | static void | |
5583 | ztest_dataset_destroy(int d) | |
5584 | { | |
5585 | char name[MAXNAMELEN]; | |
5586 | int t; | |
5587 | ||
5588 | ztest_dataset_name(name, ztest_opts.zo_pool, d); | |
5589 | ||
5590 | if (ztest_opts.zo_verbose >= 3) | |
5591 | (void) printf("Destroying %s to free up space\n", name); | |
5592 | ||
5593 | /* | |
5594 | * Cleanup any non-standard clones and snapshots. In general, | |
5595 | * ztest thread t operates on dataset (t % zopt_datasets), | |
5596 | * so there may be more than one thing to clean up. | |
5597 | */ | |
5598 | for (t = d; t < ztest_opts.zo_threads; | |
5599 | t += ztest_opts.zo_datasets) | |
5600 | ztest_dsl_dataset_cleanup(name, t); | |
5601 | ||
5602 | (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL, | |
5603 | DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN); | |
5604 | } | |
5605 | ||
5606 | static void | |
5607 | ztest_dataset_dirobj_verify(ztest_ds_t *zd) | |
5608 | { | |
5609 | uint64_t usedobjs, dirobjs, scratch; | |
5610 | ||
5611 | /* | |
5612 | * ZTEST_DIROBJ is the object directory for the entire dataset. | |
5613 | * Therefore, the number of objects in use should equal the | |
5614 | * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself. | |
5615 | * If not, we have an object leak. | |
5616 | * | |
5617 | * Note that we can only check this in ztest_dataset_open(), | |
5618 | * when the open-context and syncing-context values agree. | |
5619 | * That's because zap_count() returns the open-context value, | |
5620 | * while dmu_objset_space() returns the rootbp fill count. | |
5621 | */ | |
5622 | VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs)); | |
5623 | dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch); | |
5624 | ASSERT3U(dirobjs + 1, ==, usedobjs); | |
5625 | } | |
5626 | ||
5627 | static int | |
5628 | ztest_dataset_open(int d) | |
5629 | { | |
5630 | ztest_ds_t *zd = &ztest_ds[d]; | |
5631 | uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq; | |
5632 | objset_t *os; | |
5633 | zilog_t *zilog; | |
5634 | char name[MAXNAMELEN]; | |
5635 | int error; | |
5636 | ||
5637 | ztest_dataset_name(name, ztest_opts.zo_pool, d); | |
5638 | ||
5639 | (void) rw_enter(&ztest_name_lock, RW_READER); | |
5640 | ||
5641 | error = ztest_dataset_create(name); | |
5642 | if (error == ENOSPC) { | |
5643 | (void) rw_exit(&ztest_name_lock); | |
5644 | ztest_record_enospc(FTAG); | |
5645 | return (error); | |
5646 | } | |
5647 | ASSERT(error == 0 || error == EEXIST); | |
5648 | ||
5649 | VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, zd, &os)); | |
5650 | (void) rw_exit(&ztest_name_lock); | |
5651 | ||
5652 | ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os); | |
5653 | ||
5654 | zilog = zd->zd_zilog; | |
5655 | ||
5656 | if (zilog->zl_header->zh_claim_lr_seq != 0 && | |
5657 | zilog->zl_header->zh_claim_lr_seq < committed_seq) | |
5658 | fatal(0, "missing log records: claimed %llu < committed %llu", | |
5659 | zilog->zl_header->zh_claim_lr_seq, committed_seq); | |
5660 | ||
5661 | ztest_dataset_dirobj_verify(zd); | |
5662 | ||
5663 | zil_replay(os, zd, ztest_replay_vector); | |
5664 | ||
5665 | ztest_dataset_dirobj_verify(zd); | |
5666 | ||
5667 | if (ztest_opts.zo_verbose >= 6) | |
5668 | (void) printf("%s replay %llu blocks, %llu records, seq %llu\n", | |
5669 | zd->zd_name, | |
5670 | (u_longlong_t)zilog->zl_parse_blk_count, | |
5671 | (u_longlong_t)zilog->zl_parse_lr_count, | |
5672 | (u_longlong_t)zilog->zl_replaying_seq); | |
5673 | ||
5674 | zilog = zil_open(os, ztest_get_data); | |
5675 | ||
5676 | if (zilog->zl_replaying_seq != 0 && | |
5677 | zilog->zl_replaying_seq < committed_seq) | |
5678 | fatal(0, "missing log records: replayed %llu < committed %llu", | |
5679 | zilog->zl_replaying_seq, committed_seq); | |
5680 | ||
5681 | return (0); | |
5682 | } | |
5683 | ||
5684 | static void | |
5685 | ztest_dataset_close(int d) | |
5686 | { | |
5687 | ztest_ds_t *zd = &ztest_ds[d]; | |
5688 | ||
5689 | zil_close(zd->zd_zilog); | |
5690 | dmu_objset_disown(zd->zd_os, zd); | |
5691 | ||
5692 | ztest_zd_fini(zd); | |
5693 | } | |
5694 | ||
5695 | /* | |
5696 | * Kick off threads to run tests on all datasets in parallel. | |
5697 | */ | |
5698 | static void | |
5699 | ztest_run(ztest_shared_t *zs) | |
5700 | { | |
5701 | kt_did_t *tid; | |
5702 | spa_t *spa; | |
5703 | objset_t *os; | |
5704 | kthread_t *resume_thread; | |
5705 | uint64_t object; | |
5706 | int error; | |
5707 | int t, d; | |
5708 | ||
5709 | ztest_exiting = B_FALSE; | |
5710 | ||
5711 | /* | |
5712 | * Initialize parent/child shared state. | |
5713 | */ | |
5714 | mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL); | |
5715 | rw_init(&ztest_name_lock, NULL, RW_DEFAULT, NULL); | |
5716 | ||
5717 | zs->zs_thread_start = gethrtime(); | |
5718 | zs->zs_thread_stop = | |
5719 | zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC; | |
5720 | zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop); | |
5721 | zs->zs_thread_kill = zs->zs_thread_stop; | |
5722 | if (ztest_random(100) < ztest_opts.zo_killrate) { | |
5723 | zs->zs_thread_kill -= | |
5724 | ztest_random(ztest_opts.zo_passtime * NANOSEC); | |
5725 | } | |
5726 | ||
5727 | mutex_init(&zcl.zcl_callbacks_lock, NULL, MUTEX_DEFAULT, NULL); | |
5728 | ||
5729 | list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t), | |
5730 | offsetof(ztest_cb_data_t, zcd_node)); | |
5731 | ||
5732 | /* | |
5733 | * Open our pool. | |
5734 | */ | |
5735 | kernel_init(FREAD | FWRITE); | |
5736 | VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG)); | |
5737 | spa->spa_debug = B_TRUE; | |
5738 | ztest_spa = spa; | |
5739 | ||
5740 | VERIFY0(dmu_objset_own(ztest_opts.zo_pool, | |
5741 | DMU_OST_ANY, B_TRUE, FTAG, &os)); | |
5742 | zs->zs_guid = dmu_objset_fsid_guid(os); | |
5743 | dmu_objset_disown(os, FTAG); | |
5744 | ||
5745 | spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN; | |
5746 | ||
5747 | /* | |
5748 | * We don't expect the pool to suspend unless maxfaults == 0, | |
5749 | * in which case ztest_fault_inject() temporarily takes away | |
5750 | * the only valid replica. | |
5751 | */ | |
5752 | if (MAXFAULTS() == 0) | |
5753 | spa->spa_failmode = ZIO_FAILURE_MODE_WAIT; | |
5754 | else | |
5755 | spa->spa_failmode = ZIO_FAILURE_MODE_PANIC; | |
5756 | ||
5757 | /* | |
5758 | * Create a thread to periodically resume suspended I/O. | |
5759 | */ | |
5760 | VERIFY3P((resume_thread = zk_thread_create(NULL, 0, | |
5761 | (thread_func_t)ztest_resume_thread, spa, TS_RUN, NULL, 0, 0, | |
5762 | PTHREAD_CREATE_JOINABLE)), !=, NULL); | |
5763 | ||
5764 | #if 0 | |
5765 | /* | |
5766 | * Set a deadman alarm to abort() if we hang. | |
5767 | */ | |
5768 | signal(SIGALRM, ztest_deadman_alarm); | |
5769 | alarm((zs->zs_thread_stop - zs->zs_thread_start) / NANOSEC + GRACE); | |
5770 | #endif | |
5771 | ||
5772 | /* | |
5773 | * Verify that we can safely inquire about about any object, | |
5774 | * whether it's allocated or not. To make it interesting, | |
5775 | * we probe a 5-wide window around each power of two. | |
5776 | * This hits all edge cases, including zero and the max. | |
5777 | */ | |
5778 | for (t = 0; t < 64; t++) { | |
5779 | for (d = -5; d <= 5; d++) { | |
5780 | error = dmu_object_info(spa->spa_meta_objset, | |
5781 | (1ULL << t) + d, NULL); | |
5782 | ASSERT(error == 0 || error == ENOENT || | |
5783 | error == EINVAL); | |
5784 | } | |
5785 | } | |
5786 | ||
5787 | /* | |
5788 | * If we got any ENOSPC errors on the previous run, destroy something. | |
5789 | */ | |
5790 | if (zs->zs_enospc_count != 0) { | |
5791 | int d = ztest_random(ztest_opts.zo_datasets); | |
5792 | ztest_dataset_destroy(d); | |
5793 | } | |
5794 | zs->zs_enospc_count = 0; | |
5795 | ||
5796 | tid = umem_zalloc(ztest_opts.zo_threads * sizeof (kt_did_t), | |
5797 | UMEM_NOFAIL); | |
5798 | ||
5799 | if (ztest_opts.zo_verbose >= 4) | |
5800 | (void) printf("starting main threads...\n"); | |
5801 | ||
5802 | /* | |
5803 | * Kick off all the tests that run in parallel. | |
5804 | */ | |
5805 | for (t = 0; t < ztest_opts.zo_threads; t++) { | |
5806 | kthread_t *thread; | |
5807 | ||
5808 | if (t < ztest_opts.zo_datasets && | |
5809 | ztest_dataset_open(t) != 0) | |
5810 | return; | |
5811 | ||
5812 | VERIFY3P(thread = zk_thread_create(NULL, 0, | |
5813 | (thread_func_t)ztest_thread, | |
5814 | (void *)(uintptr_t)t, TS_RUN, NULL, 0, 0, | |
5815 | PTHREAD_CREATE_JOINABLE), !=, NULL); | |
5816 | tid[t] = thread->t_tid; | |
5817 | } | |
5818 | ||
5819 | /* | |
5820 | * Wait for all of the tests to complete. We go in reverse order | |
5821 | * so we don't close datasets while threads are still using them. | |
5822 | */ | |
5823 | for (t = ztest_opts.zo_threads - 1; t >= 0; t--) { | |
5824 | thread_join(tid[t]); | |
5825 | if (t < ztest_opts.zo_datasets) | |
5826 | ztest_dataset_close(t); | |
5827 | } | |
5828 | ||
5829 | txg_wait_synced(spa_get_dsl(spa), 0); | |
5830 | ||
5831 | zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa)); | |
5832 | zs->zs_space = metaslab_class_get_space(spa_normal_class(spa)); | |
5833 | ||
5834 | if (ztest_opts.zo_verbose >= 3) | |
5835 | zfs_dbgmsg_print(FTAG); | |
5836 | ||
5837 | umem_free(tid, ztest_opts.zo_threads * sizeof (kt_did_t)); | |
5838 | ||
5839 | /* Kill the resume thread */ | |
5840 | ztest_exiting = B_TRUE; | |
5841 | thread_join(resume_thread->t_tid); | |
5842 | ztest_resume(spa); | |
5843 | ||
5844 | /* | |
5845 | * Right before closing the pool, kick off a bunch of async I/O; | |
5846 | * spa_close() should wait for it to complete. | |
5847 | */ | |
5848 | for (object = 1; object < 50; object++) | |
5849 | dmu_prefetch(spa->spa_meta_objset, object, 0, 1ULL << 20); | |
5850 | ||
5851 | /* Verify that at least one commit cb was called in a timely fashion */ | |
5852 | if (zc_cb_counter >= ZTEST_COMMIT_CB_MIN_REG) | |
5853 | VERIFY0(zc_min_txg_delay); | |
5854 | ||
5855 | spa_close(spa, FTAG); | |
5856 | ||
5857 | /* | |
5858 | * Verify that we can loop over all pools. | |
5859 | */ | |
5860 | mutex_enter(&spa_namespace_lock); | |
5861 | for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa)) | |
5862 | if (ztest_opts.zo_verbose > 3) | |
5863 | (void) printf("spa_next: found %s\n", spa_name(spa)); | |
5864 | mutex_exit(&spa_namespace_lock); | |
5865 | ||
5866 | /* | |
5867 | * Verify that we can export the pool and reimport it under a | |
5868 | * different name. | |
5869 | */ | |
5870 | if (ztest_random(2) == 0) { | |
5871 | char name[MAXNAMELEN]; | |
5872 | (void) snprintf(name, MAXNAMELEN, "%s_import", | |
5873 | ztest_opts.zo_pool); | |
5874 | ztest_spa_import_export(ztest_opts.zo_pool, name); | |
5875 | ztest_spa_import_export(name, ztest_opts.zo_pool); | |
5876 | } | |
5877 | ||
5878 | kernel_fini(); | |
5879 | ||
5880 | list_destroy(&zcl.zcl_callbacks); | |
5881 | mutex_destroy(&zcl.zcl_callbacks_lock); | |
5882 | rw_destroy(&ztest_name_lock); | |
5883 | mutex_destroy(&ztest_vdev_lock); | |
5884 | } | |
5885 | ||
5886 | static void | |
5887 | ztest_freeze(void) | |
5888 | { | |
5889 | ztest_ds_t *zd = &ztest_ds[0]; | |
5890 | spa_t *spa; | |
5891 | int numloops = 0; | |
5892 | ||
5893 | if (ztest_opts.zo_verbose >= 3) | |
5894 | (void) printf("testing spa_freeze()...\n"); | |
5895 | ||
5896 | kernel_init(FREAD | FWRITE); | |
5897 | VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG)); | |
5898 | VERIFY3U(0, ==, ztest_dataset_open(0)); | |
5899 | spa->spa_debug = B_TRUE; | |
5900 | ztest_spa = spa; | |
5901 | ||
5902 | /* | |
5903 | * Force the first log block to be transactionally allocated. | |
5904 | * We have to do this before we freeze the pool -- otherwise | |
5905 | * the log chain won't be anchored. | |
5906 | */ | |
5907 | while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) { | |
5908 | ztest_dmu_object_alloc_free(zd, 0); | |
5909 | zil_commit(zd->zd_zilog, 0); | |
5910 | } | |
5911 | ||
5912 | txg_wait_synced(spa_get_dsl(spa), 0); | |
5913 | ||
5914 | /* | |
5915 | * Freeze the pool. This stops spa_sync() from doing anything, | |
5916 | * so that the only way to record changes from now on is the ZIL. | |
5917 | */ | |
5918 | spa_freeze(spa); | |
5919 | ||
5920 | /* | |
5921 | * Run tests that generate log records but don't alter the pool config | |
5922 | * or depend on DSL sync tasks (snapshots, objset create/destroy, etc). | |
5923 | * We do a txg_wait_synced() after each iteration to force the txg | |
5924 | * to increase well beyond the last synced value in the uberblock. | |
5925 | * The ZIL should be OK with that. | |
5926 | */ | |
5927 | while (ztest_random(10) != 0 && | |
5928 | numloops++ < ztest_opts.zo_maxloops) { | |
5929 | ztest_dmu_write_parallel(zd, 0); | |
5930 | ztest_dmu_object_alloc_free(zd, 0); | |
5931 | txg_wait_synced(spa_get_dsl(spa), 0); | |
5932 | } | |
5933 | ||
5934 | /* | |
5935 | * Commit all of the changes we just generated. | |
5936 | */ | |
5937 | zil_commit(zd->zd_zilog, 0); | |
5938 | txg_wait_synced(spa_get_dsl(spa), 0); | |
5939 | ||
5940 | /* | |
5941 | * Close our dataset and close the pool. | |
5942 | */ | |
5943 | ztest_dataset_close(0); | |
5944 | spa_close(spa, FTAG); | |
5945 | kernel_fini(); | |
5946 | ||
5947 | /* | |
5948 | * Open and close the pool and dataset to induce log replay. | |
5949 | */ | |
5950 | kernel_init(FREAD | FWRITE); | |
5951 | VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG)); | |
5952 | ASSERT(spa_freeze_txg(spa) == UINT64_MAX); | |
5953 | VERIFY3U(0, ==, ztest_dataset_open(0)); | |
5954 | ztest_dataset_close(0); | |
5955 | ||
5956 | spa->spa_debug = B_TRUE; | |
5957 | ztest_spa = spa; | |
5958 | txg_wait_synced(spa_get_dsl(spa), 0); | |
5959 | ztest_reguid(NULL, 0); | |
5960 | ||
5961 | spa_close(spa, FTAG); | |
5962 | kernel_fini(); | |
5963 | } | |
5964 | ||
5965 | void | |
5966 | print_time(hrtime_t t, char *timebuf) | |
5967 | { | |
5968 | hrtime_t s = t / NANOSEC; | |
5969 | hrtime_t m = s / 60; | |
5970 | hrtime_t h = m / 60; | |
5971 | hrtime_t d = h / 24; | |
5972 | ||
5973 | s -= m * 60; | |
5974 | m -= h * 60; | |
5975 | h -= d * 24; | |
5976 | ||
5977 | timebuf[0] = '\0'; | |
5978 | ||
5979 | if (d) | |
5980 | (void) sprintf(timebuf, | |
5981 | "%llud%02lluh%02llum%02llus", d, h, m, s); | |
5982 | else if (h) | |
5983 | (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s); | |
5984 | else if (m) | |
5985 | (void) sprintf(timebuf, "%llum%02llus", m, s); | |
5986 | else | |
5987 | (void) sprintf(timebuf, "%llus", s); | |
5988 | } | |
5989 | ||
5990 | static nvlist_t * | |
5991 | make_random_props(void) | |
5992 | { | |
5993 | nvlist_t *props; | |
5994 | ||
5995 | VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0); | |
5996 | if (ztest_random(2) == 0) | |
5997 | return (props); | |
5998 | VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0); | |
5999 | ||
6000 | return (props); | |
6001 | } | |
6002 | ||
6003 | /* | |
6004 | * Create a storage pool with the given name and initial vdev size. | |
6005 | * Then test spa_freeze() functionality. | |
6006 | */ | |
6007 | static void | |
6008 | ztest_init(ztest_shared_t *zs) | |
6009 | { | |
6010 | spa_t *spa; | |
6011 | nvlist_t *nvroot, *props; | |
6012 | int i; | |
6013 | ||
6014 | mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL); | |
6015 | rw_init(&ztest_name_lock, NULL, RW_DEFAULT, NULL); | |
6016 | ||
6017 | kernel_init(FREAD | FWRITE); | |
6018 | ||
6019 | /* | |
6020 | * Create the storage pool. | |
6021 | */ | |
6022 | (void) spa_destroy(ztest_opts.zo_pool); | |
6023 | ztest_shared->zs_vdev_next_leaf = 0; | |
6024 | zs->zs_splits = 0; | |
6025 | zs->zs_mirrors = ztest_opts.zo_mirrors; | |
6026 | nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0, | |
6027 | 0, ztest_opts.zo_raidz, zs->zs_mirrors, 1); | |
6028 | props = make_random_props(); | |
6029 | for (i = 0; i < SPA_FEATURES; i++) { | |
6030 | char *buf; | |
6031 | VERIFY3S(-1, !=, asprintf(&buf, "feature@%s", | |
6032 | spa_feature_table[i].fi_uname)); | |
6033 | VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0)); | |
6034 | free(buf); | |
6035 | } | |
6036 | VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props, NULL)); | |
6037 | nvlist_free(nvroot); | |
6038 | nvlist_free(props); | |
6039 | ||
6040 | VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG)); | |
6041 | zs->zs_metaslab_sz = | |
6042 | 1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift; | |
6043 | spa_close(spa, FTAG); | |
6044 | ||
6045 | kernel_fini(); | |
6046 | ||
6047 | ztest_run_zdb(ztest_opts.zo_pool); | |
6048 | ||
6049 | ztest_freeze(); | |
6050 | ||
6051 | ztest_run_zdb(ztest_opts.zo_pool); | |
6052 | ||
6053 | rw_destroy(&ztest_name_lock); | |
6054 | mutex_destroy(&ztest_vdev_lock); | |
6055 | } | |
6056 | ||
6057 | static void | |
6058 | setup_data_fd(void) | |
6059 | { | |
6060 | static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX"; | |
6061 | ||
6062 | ztest_fd_data = mkstemp(ztest_name_data); | |
6063 | ASSERT3S(ztest_fd_data, >=, 0); | |
6064 | (void) unlink(ztest_name_data); | |
6065 | } | |
6066 | ||
6067 | static int | |
6068 | shared_data_size(ztest_shared_hdr_t *hdr) | |
6069 | { | |
6070 | int size; | |
6071 | ||
6072 | size = hdr->zh_hdr_size; | |
6073 | size += hdr->zh_opts_size; | |
6074 | size += hdr->zh_size; | |
6075 | size += hdr->zh_stats_size * hdr->zh_stats_count; | |
6076 | size += hdr->zh_ds_size * hdr->zh_ds_count; | |
6077 | ||
6078 | return (size); | |
6079 | } | |
6080 | ||
6081 | static void | |
6082 | setup_hdr(void) | |
6083 | { | |
6084 | int size; | |
6085 | ztest_shared_hdr_t *hdr; | |
6086 | ||
6087 | hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()), | |
6088 | PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0); | |
6089 | VERIFY3P(hdr, !=, MAP_FAILED); | |
6090 | ||
6091 | VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t))); | |
6092 | ||
6093 | hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t); | |
6094 | hdr->zh_opts_size = sizeof (ztest_shared_opts_t); | |
6095 | hdr->zh_size = sizeof (ztest_shared_t); | |
6096 | hdr->zh_stats_size = sizeof (ztest_shared_callstate_t); | |
6097 | hdr->zh_stats_count = ZTEST_FUNCS; | |
6098 | hdr->zh_ds_size = sizeof (ztest_shared_ds_t); | |
6099 | hdr->zh_ds_count = ztest_opts.zo_datasets; | |
6100 | ||
6101 | size = shared_data_size(hdr); | |
6102 | VERIFY3U(0, ==, ftruncate(ztest_fd_data, size)); | |
6103 | ||
6104 | (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize())); | |
6105 | } | |
6106 | ||
6107 | static void | |
6108 | setup_data(void) | |
6109 | { | |
6110 | int size, offset; | |
6111 | ztest_shared_hdr_t *hdr; | |
6112 | uint8_t *buf; | |
6113 | ||
6114 | hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()), | |
6115 | PROT_READ, MAP_SHARED, ztest_fd_data, 0); | |
6116 | VERIFY3P(hdr, !=, MAP_FAILED); | |
6117 | ||
6118 | size = shared_data_size(hdr); | |
6119 | ||
6120 | (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize())); | |
6121 | hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()), | |
6122 | PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0); | |
6123 | VERIFY3P(hdr, !=, MAP_FAILED); | |
6124 | buf = (uint8_t *)hdr; | |
6125 | ||
6126 | offset = hdr->zh_hdr_size; | |
6127 | ztest_shared_opts = (void *)&buf[offset]; | |
6128 | offset += hdr->zh_opts_size; | |
6129 | ztest_shared = (void *)&buf[offset]; | |
6130 | offset += hdr->zh_size; | |
6131 | ztest_shared_callstate = (void *)&buf[offset]; | |
6132 | offset += hdr->zh_stats_size * hdr->zh_stats_count; | |
6133 | ztest_shared_ds = (void *)&buf[offset]; | |
6134 | } | |
6135 | ||
6136 | static boolean_t | |
6137 | exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp) | |
6138 | { | |
6139 | pid_t pid; | |
6140 | int status; | |
6141 | char *cmdbuf = NULL; | |
6142 | ||
6143 | pid = fork(); | |
6144 | ||
6145 | if (cmd == NULL) { | |
6146 | cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); | |
6147 | (void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN); | |
6148 | cmd = cmdbuf; | |
6149 | } | |
6150 | ||
6151 | if (pid == -1) | |
6152 | fatal(1, "fork failed"); | |
6153 | ||
6154 | if (pid == 0) { /* child */ | |
6155 | char *emptyargv[2] = { cmd, NULL }; | |
6156 | char fd_data_str[12]; | |
6157 | ||
6158 | struct rlimit rl = { 1024, 1024 }; | |
6159 | (void) setrlimit(RLIMIT_NOFILE, &rl); | |
6160 | ||
6161 | (void) close(ztest_fd_rand); | |
6162 | VERIFY(11 >= snprintf(fd_data_str, 12, "%d", ztest_fd_data)); | |
6163 | VERIFY(0 == setenv("ZTEST_FD_DATA", fd_data_str, 1)); | |
6164 | ||
6165 | (void) enable_extended_FILE_stdio(-1, -1); | |
6166 | if (libpath != NULL) | |
6167 | VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1)); | |
6168 | (void) execv(cmd, emptyargv); | |
6169 | ztest_dump_core = B_FALSE; | |
6170 | fatal(B_TRUE, "exec failed: %s", cmd); | |
6171 | } | |
6172 | ||
6173 | if (cmdbuf != NULL) { | |
6174 | umem_free(cmdbuf, MAXPATHLEN); | |
6175 | cmd = NULL; | |
6176 | } | |
6177 | ||
6178 | while (waitpid(pid, &status, 0) != pid) | |
6179 | continue; | |
6180 | if (statusp != NULL) | |
6181 | *statusp = status; | |
6182 | ||
6183 | if (WIFEXITED(status)) { | |
6184 | if (WEXITSTATUS(status) != 0) { | |
6185 | (void) fprintf(stderr, "child exited with code %d\n", | |
6186 | WEXITSTATUS(status)); | |
6187 | exit(2); | |
6188 | } | |
6189 | return (B_FALSE); | |
6190 | } else if (WIFSIGNALED(status)) { | |
6191 | if (!ignorekill || WTERMSIG(status) != SIGKILL) { | |
6192 | (void) fprintf(stderr, "child died with signal %d\n", | |
6193 | WTERMSIG(status)); | |
6194 | exit(3); | |
6195 | } | |
6196 | return (B_TRUE); | |
6197 | } else { | |
6198 | (void) fprintf(stderr, "something strange happened to child\n"); | |
6199 | exit(4); | |
6200 | /* NOTREACHED */ | |
6201 | } | |
6202 | } | |
6203 | ||
6204 | static void | |
6205 | ztest_run_init(void) | |
6206 | { | |
6207 | int i; | |
6208 | ||
6209 | ztest_shared_t *zs = ztest_shared; | |
6210 | ||
6211 | ASSERT(ztest_opts.zo_init != 0); | |
6212 | ||
6213 | /* | |
6214 | * Blow away any existing copy of zpool.cache | |
6215 | */ | |
6216 | (void) remove(spa_config_path); | |
6217 | ||
6218 | /* | |
6219 | * Create and initialize our storage pool. | |
6220 | */ | |
6221 | for (i = 1; i <= ztest_opts.zo_init; i++) { | |
6222 | bzero(zs, sizeof (ztest_shared_t)); | |
6223 | if (ztest_opts.zo_verbose >= 3 && | |
6224 | ztest_opts.zo_init != 1) { | |
6225 | (void) printf("ztest_init(), pass %d\n", i); | |
6226 | } | |
6227 | ztest_init(zs); | |
6228 | } | |
6229 | } | |
6230 | ||
6231 | int | |
6232 | main(int argc, char **argv) | |
6233 | { | |
6234 | int kills = 0; | |
6235 | int iters = 0; | |
6236 | int older = 0; | |
6237 | int newer = 0; | |
6238 | ztest_shared_t *zs; | |
6239 | ztest_info_t *zi; | |
6240 | ztest_shared_callstate_t *zc; | |
6241 | char timebuf[100]; | |
6242 | char numbuf[6]; | |
6243 | spa_t *spa; | |
6244 | char *cmd; | |
6245 | boolean_t hasalt; | |
6246 | int f; | |
6247 | char *fd_data_str = getenv("ZTEST_FD_DATA"); | |
6248 | ||
6249 | (void) setvbuf(stdout, NULL, _IOLBF, 0); | |
6250 | ||
6251 | dprintf_setup(&argc, argv); | |
6252 | ||
6253 | ztest_fd_rand = open("/dev/urandom", O_RDONLY); | |
6254 | ASSERT3S(ztest_fd_rand, >=, 0); | |
6255 | ||
6256 | if (!fd_data_str) { | |
6257 | process_options(argc, argv); | |
6258 | ||
6259 | setup_data_fd(); | |
6260 | setup_hdr(); | |
6261 | setup_data(); | |
6262 | bcopy(&ztest_opts, ztest_shared_opts, | |
6263 | sizeof (*ztest_shared_opts)); | |
6264 | } else { | |
6265 | ztest_fd_data = atoi(fd_data_str); | |
6266 | setup_data(); | |
6267 | bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts)); | |
6268 | } | |
6269 | ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count); | |
6270 | ||
6271 | /* Override location of zpool.cache */ | |
6272 | VERIFY(asprintf((char **)&spa_config_path, "%s/zpool.cache", | |
6273 | ztest_opts.zo_dir) != -1); | |
6274 | ||
6275 | ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t), | |
6276 | UMEM_NOFAIL); | |
6277 | zs = ztest_shared; | |
6278 | ||
6279 | if (fd_data_str) { | |
6280 | metaslab_gang_bang = ztest_opts.zo_metaslab_gang_bang; | |
6281 | metaslab_df_alloc_threshold = | |
6282 | zs->zs_metaslab_df_alloc_threshold; | |
6283 | ||
6284 | if (zs->zs_do_init) | |
6285 | ztest_run_init(); | |
6286 | else | |
6287 | ztest_run(zs); | |
6288 | exit(0); | |
6289 | } | |
6290 | ||
6291 | hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0); | |
6292 | ||
6293 | if (ztest_opts.zo_verbose >= 1) { | |
6294 | (void) printf("%llu vdevs, %d datasets, %d threads," | |
6295 | " %llu seconds...\n", | |
6296 | (u_longlong_t)ztest_opts.zo_vdevs, | |
6297 | ztest_opts.zo_datasets, | |
6298 | ztest_opts.zo_threads, | |
6299 | (u_longlong_t)ztest_opts.zo_time); | |
6300 | } | |
6301 | ||
6302 | cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); | |
6303 | (void) strlcpy(cmd, getexecname(), MAXNAMELEN); | |
6304 | ||
6305 | zs->zs_do_init = B_TRUE; | |
6306 | if (strlen(ztest_opts.zo_alt_ztest) != 0) { | |
6307 | if (ztest_opts.zo_verbose >= 1) { | |
6308 | (void) printf("Executing older ztest for " | |
6309 | "initialization: %s\n", ztest_opts.zo_alt_ztest); | |
6310 | } | |
6311 | VERIFY(!exec_child(ztest_opts.zo_alt_ztest, | |
6312 | ztest_opts.zo_alt_libpath, B_FALSE, NULL)); | |
6313 | } else { | |
6314 | VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL)); | |
6315 | } | |
6316 | zs->zs_do_init = B_FALSE; | |
6317 | ||
6318 | zs->zs_proc_start = gethrtime(); | |
6319 | zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC; | |
6320 | ||
6321 | for (f = 0; f < ZTEST_FUNCS; f++) { | |
6322 | zi = &ztest_info[f]; | |
6323 | zc = ZTEST_GET_SHARED_CALLSTATE(f); | |
6324 | if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop) | |
6325 | zc->zc_next = UINT64_MAX; | |
6326 | else | |
6327 | zc->zc_next = zs->zs_proc_start + | |
6328 | ztest_random(2 * zi->zi_interval[0] + 1); | |
6329 | } | |
6330 | ||
6331 | /* | |
6332 | * Run the tests in a loop. These tests include fault injection | |
6333 | * to verify that self-healing data works, and forced crashes | |
6334 | * to verify that we never lose on-disk consistency. | |
6335 | */ | |
6336 | while (gethrtime() < zs->zs_proc_stop) { | |
6337 | int status; | |
6338 | boolean_t killed; | |
6339 | ||
6340 | /* | |
6341 | * Initialize the workload counters for each function. | |
6342 | */ | |
6343 | for (f = 0; f < ZTEST_FUNCS; f++) { | |
6344 | zc = ZTEST_GET_SHARED_CALLSTATE(f); | |
6345 | zc->zc_count = 0; | |
6346 | zc->zc_time = 0; | |
6347 | } | |
6348 | ||
6349 | /* Set the allocation switch size */ | |
6350 | zs->zs_metaslab_df_alloc_threshold = | |
6351 | ztest_random(zs->zs_metaslab_sz / 4) + 1; | |
6352 | ||
6353 | if (!hasalt || ztest_random(2) == 0) { | |
6354 | if (hasalt && ztest_opts.zo_verbose >= 1) { | |
6355 | (void) printf("Executing newer ztest: %s\n", | |
6356 | cmd); | |
6357 | } | |
6358 | newer++; | |
6359 | killed = exec_child(cmd, NULL, B_TRUE, &status); | |
6360 | } else { | |
6361 | if (hasalt && ztest_opts.zo_verbose >= 1) { | |
6362 | (void) printf("Executing older ztest: %s\n", | |
6363 | ztest_opts.zo_alt_ztest); | |
6364 | } | |
6365 | older++; | |
6366 | killed = exec_child(ztest_opts.zo_alt_ztest, | |
6367 | ztest_opts.zo_alt_libpath, B_TRUE, &status); | |
6368 | } | |
6369 | ||
6370 | if (killed) | |
6371 | kills++; | |
6372 | iters++; | |
6373 | ||
6374 | if (ztest_opts.zo_verbose >= 1) { | |
6375 | hrtime_t now = gethrtime(); | |
6376 | ||
6377 | now = MIN(now, zs->zs_proc_stop); | |
6378 | print_time(zs->zs_proc_stop - now, timebuf); | |
6379 | nicenum(zs->zs_space, numbuf); | |
6380 | ||
6381 | (void) printf("Pass %3d, %8s, %3llu ENOSPC, " | |
6382 | "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n", | |
6383 | iters, | |
6384 | WIFEXITED(status) ? "Complete" : "SIGKILL", | |
6385 | (u_longlong_t)zs->zs_enospc_count, | |
6386 | 100.0 * zs->zs_alloc / zs->zs_space, | |
6387 | numbuf, | |
6388 | 100.0 * (now - zs->zs_proc_start) / | |
6389 | (ztest_opts.zo_time * NANOSEC), timebuf); | |
6390 | } | |
6391 | ||
6392 | if (ztest_opts.zo_verbose >= 2) { | |
6393 | (void) printf("\nWorkload summary:\n\n"); | |
6394 | (void) printf("%7s %9s %s\n", | |
6395 | "Calls", "Time", "Function"); | |
6396 | (void) printf("%7s %9s %s\n", | |
6397 | "-----", "----", "--------"); | |
6398 | for (f = 0; f < ZTEST_FUNCS; f++) { | |
6399 | Dl_info dli; | |
6400 | ||
6401 | zi = &ztest_info[f]; | |
6402 | zc = ZTEST_GET_SHARED_CALLSTATE(f); | |
6403 | print_time(zc->zc_time, timebuf); | |
6404 | (void) dladdr((void *)zi->zi_func, &dli); | |
6405 | (void) printf("%7llu %9s %s\n", | |
6406 | (u_longlong_t)zc->zc_count, timebuf, | |
6407 | dli.dli_sname); | |
6408 | } | |
6409 | (void) printf("\n"); | |
6410 | } | |
6411 | ||
6412 | /* | |
6413 | * It's possible that we killed a child during a rename test, | |
6414 | * in which case we'll have a 'ztest_tmp' pool lying around | |
6415 | * instead of 'ztest'. Do a blind rename in case this happened. | |
6416 | */ | |
6417 | kernel_init(FREAD); | |
6418 | if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) { | |
6419 | spa_close(spa, FTAG); | |
6420 | } else { | |
6421 | char tmpname[MAXNAMELEN]; | |
6422 | kernel_fini(); | |
6423 | kernel_init(FREAD | FWRITE); | |
6424 | (void) snprintf(tmpname, sizeof (tmpname), "%s_tmp", | |
6425 | ztest_opts.zo_pool); | |
6426 | (void) spa_rename(tmpname, ztest_opts.zo_pool); | |
6427 | } | |
6428 | kernel_fini(); | |
6429 | ||
6430 | ztest_run_zdb(ztest_opts.zo_pool); | |
6431 | } | |
6432 | ||
6433 | if (ztest_opts.zo_verbose >= 1) { | |
6434 | if (hasalt) { | |
6435 | (void) printf("%d runs of older ztest: %s\n", older, | |
6436 | ztest_opts.zo_alt_ztest); | |
6437 | (void) printf("%d runs of newer ztest: %s\n", newer, | |
6438 | cmd); | |
6439 | } | |
6440 | (void) printf("%d killed, %d completed, %.0f%% kill rate\n", | |
6441 | kills, iters - kills, (100.0 * kills) / MAX(1, iters)); | |
6442 | } | |
6443 | ||
6444 | umem_free(cmd, MAXNAMELEN); | |
6445 | ||
6446 | return (0); | |
6447 | } |