]>
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) 2011, 2015 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 compatibility | |
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/zfs_rlock.h> | |
105 | #include <sys/vdev_impl.h> | |
106 | #include <sys/vdev_file.h> | |
107 | #include <sys/spa_impl.h> | |
108 | #include <sys/metaslab_impl.h> | |
109 | #include <sys/dsl_prop.h> | |
110 | #include <sys/dsl_dataset.h> | |
111 | #include <sys/dsl_destroy.h> | |
112 | #include <sys/dsl_scan.h> | |
113 | #include <sys/zio_checksum.h> | |
114 | #include <sys/refcount.h> | |
115 | #include <sys/zfeature.h> | |
116 | #include <sys/dsl_userhold.h> | |
117 | #include <stdio.h> | |
118 | #include <stdio_ext.h> | |
119 | #include <stdlib.h> | |
120 | #include <unistd.h> | |
121 | #include <signal.h> | |
122 | #include <umem.h> | |
123 | #include <ctype.h> | |
124 | #include <math.h> | |
125 | #include <sys/fs/zfs.h> | |
126 | #include <zfs_fletcher.h> | |
127 | #include <libnvpair.h> | |
128 | #ifdef __GLIBC__ | |
129 | #include <execinfo.h> /* for backtrace() */ | |
130 | #endif | |
131 | ||
132 | static int ztest_fd_data = -1; | |
133 | static int ztest_fd_rand = -1; | |
134 | ||
135 | typedef struct ztest_shared_hdr { | |
136 | uint64_t zh_hdr_size; | |
137 | uint64_t zh_opts_size; | |
138 | uint64_t zh_size; | |
139 | uint64_t zh_stats_size; | |
140 | uint64_t zh_stats_count; | |
141 | uint64_t zh_ds_size; | |
142 | uint64_t zh_ds_count; | |
143 | } ztest_shared_hdr_t; | |
144 | ||
145 | static ztest_shared_hdr_t *ztest_shared_hdr; | |
146 | ||
147 | typedef struct ztest_shared_opts { | |
148 | char zo_pool[MAXNAMELEN]; | |
149 | char zo_dir[MAXNAMELEN]; | |
150 | char zo_alt_ztest[MAXNAMELEN]; | |
151 | char zo_alt_libpath[MAXNAMELEN]; | |
152 | uint64_t zo_vdevs; | |
153 | uint64_t zo_vdevtime; | |
154 | size_t zo_vdev_size; | |
155 | int zo_ashift; | |
156 | int zo_mirrors; | |
157 | int zo_raidz; | |
158 | int zo_raidz_parity; | |
159 | int zo_datasets; | |
160 | int zo_threads; | |
161 | uint64_t zo_passtime; | |
162 | uint64_t zo_killrate; | |
163 | int zo_verbose; | |
164 | int zo_init; | |
165 | uint64_t zo_time; | |
166 | uint64_t zo_maxloops; | |
167 | uint64_t zo_metaslab_gang_bang; | |
168 | } ztest_shared_opts_t; | |
169 | ||
170 | static const ztest_shared_opts_t ztest_opts_defaults = { | |
171 | .zo_pool = { 'z', 't', 'e', 's', 't', '\0' }, | |
172 | .zo_dir = { '/', 't', 'm', 'p', '\0' }, | |
173 | .zo_alt_ztest = { '\0' }, | |
174 | .zo_alt_libpath = { '\0' }, | |
175 | .zo_vdevs = 5, | |
176 | .zo_ashift = SPA_MINBLOCKSHIFT, | |
177 | .zo_mirrors = 2, | |
178 | .zo_raidz = 4, | |
179 | .zo_raidz_parity = 1, | |
180 | .zo_vdev_size = SPA_MINDEVSIZE * 2, | |
181 | .zo_datasets = 7, | |
182 | .zo_threads = 23, | |
183 | .zo_passtime = 60, /* 60 seconds */ | |
184 | .zo_killrate = 70, /* 70% kill rate */ | |
185 | .zo_verbose = 0, | |
186 | .zo_init = 1, | |
187 | .zo_time = 300, /* 5 minutes */ | |
188 | .zo_maxloops = 50, /* max loops during spa_freeze() */ | |
189 | .zo_metaslab_gang_bang = 32 << 10 | |
190 | }; | |
191 | ||
192 | extern uint64_t metaslab_gang_bang; | |
193 | extern uint64_t metaslab_df_alloc_threshold; | |
194 | extern int metaslab_preload_limit; | |
195 | ||
196 | static ztest_shared_opts_t *ztest_shared_opts; | |
197 | static ztest_shared_opts_t ztest_opts; | |
198 | ||
199 | typedef struct ztest_shared_ds { | |
200 | uint64_t zd_seq; | |
201 | } ztest_shared_ds_t; | |
202 | ||
203 | static ztest_shared_ds_t *ztest_shared_ds; | |
204 | #define ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d]) | |
205 | ||
206 | #define BT_MAGIC 0x123456789abcdefULL | |
207 | #define MAXFAULTS() \ | |
208 | (MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1) | |
209 | ||
210 | enum ztest_io_type { | |
211 | ZTEST_IO_WRITE_TAG, | |
212 | ZTEST_IO_WRITE_PATTERN, | |
213 | ZTEST_IO_WRITE_ZEROES, | |
214 | ZTEST_IO_TRUNCATE, | |
215 | ZTEST_IO_SETATTR, | |
216 | ZTEST_IO_REWRITE, | |
217 | ZTEST_IO_TYPES | |
218 | }; | |
219 | ||
220 | typedef struct ztest_block_tag { | |
221 | uint64_t bt_magic; | |
222 | uint64_t bt_objset; | |
223 | uint64_t bt_object; | |
224 | uint64_t bt_dnodesize; | |
225 | uint64_t bt_offset; | |
226 | uint64_t bt_gen; | |
227 | uint64_t bt_txg; | |
228 | uint64_t bt_crtxg; | |
229 | } ztest_block_tag_t; | |
230 | ||
231 | typedef struct bufwad { | |
232 | uint64_t bw_index; | |
233 | uint64_t bw_txg; | |
234 | uint64_t bw_data; | |
235 | } bufwad_t; | |
236 | ||
237 | typedef struct rll { | |
238 | void *rll_writer; | |
239 | int rll_readers; | |
240 | kmutex_t rll_lock; | |
241 | kcondvar_t rll_cv; | |
242 | } rll_t; | |
243 | ||
244 | typedef struct zll { | |
245 | list_t z_list; | |
246 | kmutex_t z_lock; | |
247 | } zll_t; | |
248 | ||
249 | #define ZTEST_RANGE_LOCKS 64 | |
250 | #define ZTEST_OBJECT_LOCKS 64 | |
251 | ||
252 | /* | |
253 | * Object descriptor. Used as a template for object lookup/create/remove. | |
254 | */ | |
255 | typedef struct ztest_od { | |
256 | uint64_t od_dir; | |
257 | uint64_t od_object; | |
258 | dmu_object_type_t od_type; | |
259 | dmu_object_type_t od_crtype; | |
260 | uint64_t od_blocksize; | |
261 | uint64_t od_crblocksize; | |
262 | uint64_t od_crdnodesize; | |
263 | uint64_t od_gen; | |
264 | uint64_t od_crgen; | |
265 | char od_name[MAXNAMELEN]; | |
266 | } ztest_od_t; | |
267 | ||
268 | /* | |
269 | * Per-dataset state. | |
270 | */ | |
271 | typedef struct ztest_ds { | |
272 | ztest_shared_ds_t *zd_shared; | |
273 | objset_t *zd_os; | |
274 | rwlock_t zd_zilog_lock; | |
275 | zilog_t *zd_zilog; | |
276 | ztest_od_t *zd_od; /* debugging aid */ | |
277 | char zd_name[MAXNAMELEN]; | |
278 | kmutex_t zd_dirobj_lock; | |
279 | rll_t zd_object_lock[ZTEST_OBJECT_LOCKS]; | |
280 | zll_t zd_range_lock[ZTEST_RANGE_LOCKS]; | |
281 | } ztest_ds_t; | |
282 | ||
283 | /* | |
284 | * Per-iteration state. | |
285 | */ | |
286 | typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id); | |
287 | ||
288 | typedef struct ztest_info { | |
289 | ztest_func_t *zi_func; /* test function */ | |
290 | uint64_t zi_iters; /* iterations per execution */ | |
291 | uint64_t *zi_interval; /* execute every <interval> seconds */ | |
292 | const char *zi_funcname; /* name of test function */ | |
293 | } ztest_info_t; | |
294 | ||
295 | typedef struct ztest_shared_callstate { | |
296 | uint64_t zc_count; /* per-pass count */ | |
297 | uint64_t zc_time; /* per-pass time */ | |
298 | uint64_t zc_next; /* next time to call this function */ | |
299 | } ztest_shared_callstate_t; | |
300 | ||
301 | static ztest_shared_callstate_t *ztest_shared_callstate; | |
302 | #define ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c]) | |
303 | ||
304 | ztest_func_t ztest_dmu_read_write; | |
305 | ztest_func_t ztest_dmu_write_parallel; | |
306 | ztest_func_t ztest_dmu_object_alloc_free; | |
307 | ztest_func_t ztest_dmu_commit_callbacks; | |
308 | ztest_func_t ztest_zap; | |
309 | ztest_func_t ztest_zap_parallel; | |
310 | ztest_func_t ztest_zil_commit; | |
311 | ztest_func_t ztest_zil_remount; | |
312 | ztest_func_t ztest_dmu_read_write_zcopy; | |
313 | ztest_func_t ztest_dmu_objset_create_destroy; | |
314 | ztest_func_t ztest_dmu_prealloc; | |
315 | ztest_func_t ztest_fzap; | |
316 | ztest_func_t ztest_dmu_snapshot_create_destroy; | |
317 | ztest_func_t ztest_dsl_prop_get_set; | |
318 | ztest_func_t ztest_spa_prop_get_set; | |
319 | ztest_func_t ztest_spa_create_destroy; | |
320 | ztest_func_t ztest_fault_inject; | |
321 | ztest_func_t ztest_ddt_repair; | |
322 | ztest_func_t ztest_dmu_snapshot_hold; | |
323 | ztest_func_t ztest_spa_rename; | |
324 | ztest_func_t ztest_scrub; | |
325 | ztest_func_t ztest_dsl_dataset_promote_busy; | |
326 | ztest_func_t ztest_vdev_attach_detach; | |
327 | ztest_func_t ztest_vdev_LUN_growth; | |
328 | ztest_func_t ztest_vdev_add_remove; | |
329 | ztest_func_t ztest_vdev_aux_add_remove; | |
330 | ztest_func_t ztest_split_pool; | |
331 | ztest_func_t ztest_reguid; | |
332 | ztest_func_t ztest_spa_upgrade; | |
333 | ztest_func_t ztest_fletcher; | |
334 | ztest_func_t ztest_verify_dnode_bt; | |
335 | ||
336 | uint64_t zopt_always = 0ULL * NANOSEC; /* all the time */ | |
337 | uint64_t zopt_incessant = 1ULL * NANOSEC / 10; /* every 1/10 second */ | |
338 | uint64_t zopt_often = 1ULL * NANOSEC; /* every second */ | |
339 | uint64_t zopt_sometimes = 10ULL * NANOSEC; /* every 10 seconds */ | |
340 | uint64_t zopt_rarely = 60ULL * NANOSEC; /* every 60 seconds */ | |
341 | ||
342 | #define ZTI_INIT(func, iters, interval) \ | |
343 | { .zi_func = (func), \ | |
344 | .zi_iters = (iters), \ | |
345 | .zi_interval = (interval), \ | |
346 | .zi_funcname = # func } | |
347 | ||
348 | ztest_info_t ztest_info[] = { | |
349 | ZTI_INIT(ztest_dmu_read_write, 1, &zopt_always), | |
350 | ZTI_INIT(ztest_dmu_write_parallel, 10, &zopt_always), | |
351 | ZTI_INIT(ztest_dmu_object_alloc_free, 1, &zopt_always), | |
352 | ZTI_INIT(ztest_dmu_commit_callbacks, 1, &zopt_always), | |
353 | ZTI_INIT(ztest_zap, 30, &zopt_always), | |
354 | ZTI_INIT(ztest_zap_parallel, 100, &zopt_always), | |
355 | ZTI_INIT(ztest_split_pool, 1, &zopt_always), | |
356 | ZTI_INIT(ztest_zil_commit, 1, &zopt_incessant), | |
357 | ZTI_INIT(ztest_zil_remount, 1, &zopt_sometimes), | |
358 | ZTI_INIT(ztest_dmu_read_write_zcopy, 1, &zopt_often), | |
359 | ZTI_INIT(ztest_dmu_objset_create_destroy, 1, &zopt_often), | |
360 | ZTI_INIT(ztest_dsl_prop_get_set, 1, &zopt_often), | |
361 | ZTI_INIT(ztest_spa_prop_get_set, 1, &zopt_sometimes), | |
362 | #if 0 | |
363 | ZTI_INIT(ztest_dmu_prealloc, 1, &zopt_sometimes), | |
364 | #endif | |
365 | ZTI_INIT(ztest_fzap, 1, &zopt_sometimes), | |
366 | ZTI_INIT(ztest_dmu_snapshot_create_destroy, 1, &zopt_sometimes), | |
367 | ZTI_INIT(ztest_spa_create_destroy, 1, &zopt_sometimes), | |
368 | ZTI_INIT(ztest_fault_inject, 1, &zopt_sometimes), | |
369 | ZTI_INIT(ztest_ddt_repair, 1, &zopt_sometimes), | |
370 | ZTI_INIT(ztest_dmu_snapshot_hold, 1, &zopt_sometimes), | |
371 | ZTI_INIT(ztest_reguid, 1, &zopt_rarely), | |
372 | ZTI_INIT(ztest_spa_rename, 1, &zopt_rarely), | |
373 | ZTI_INIT(ztest_scrub, 1, &zopt_rarely), | |
374 | ZTI_INIT(ztest_spa_upgrade, 1, &zopt_rarely), | |
375 | ZTI_INIT(ztest_dsl_dataset_promote_busy, 1, &zopt_rarely), | |
376 | ZTI_INIT(ztest_vdev_attach_detach, 1, &zopt_sometimes), | |
377 | ZTI_INIT(ztest_vdev_LUN_growth, 1, &zopt_rarely), | |
378 | ZTI_INIT(ztest_vdev_add_remove, 1, &ztest_opts.zo_vdevtime), | |
379 | ZTI_INIT(ztest_vdev_aux_add_remove, 1, &ztest_opts.zo_vdevtime), | |
380 | ZTI_INIT(ztest_fletcher, 1, &zopt_rarely), | |
381 | ZTI_INIT(ztest_verify_dnode_bt, 1, &zopt_sometimes), | |
382 | }; | |
383 | ||
384 | #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t)) | |
385 | ||
386 | /* | |
387 | * The following struct is used to hold a list of uncalled commit callbacks. | |
388 | * The callbacks are ordered by txg number. | |
389 | */ | |
390 | typedef struct ztest_cb_list { | |
391 | kmutex_t zcl_callbacks_lock; | |
392 | list_t zcl_callbacks; | |
393 | } ztest_cb_list_t; | |
394 | ||
395 | /* | |
396 | * Stuff we need to share writably between parent and child. | |
397 | */ | |
398 | typedef struct ztest_shared { | |
399 | boolean_t zs_do_init; | |
400 | hrtime_t zs_proc_start; | |
401 | hrtime_t zs_proc_stop; | |
402 | hrtime_t zs_thread_start; | |
403 | hrtime_t zs_thread_stop; | |
404 | hrtime_t zs_thread_kill; | |
405 | uint64_t zs_enospc_count; | |
406 | uint64_t zs_vdev_next_leaf; | |
407 | uint64_t zs_vdev_aux; | |
408 | uint64_t zs_alloc; | |
409 | uint64_t zs_space; | |
410 | uint64_t zs_splits; | |
411 | uint64_t zs_mirrors; | |
412 | uint64_t zs_metaslab_sz; | |
413 | uint64_t zs_metaslab_df_alloc_threshold; | |
414 | uint64_t zs_guid; | |
415 | } ztest_shared_t; | |
416 | ||
417 | #define ID_PARALLEL -1ULL | |
418 | ||
419 | static char ztest_dev_template[] = "%s/%s.%llua"; | |
420 | static char ztest_aux_template[] = "%s/%s.%s.%llu"; | |
421 | ztest_shared_t *ztest_shared; | |
422 | ||
423 | static spa_t *ztest_spa = NULL; | |
424 | static ztest_ds_t *ztest_ds; | |
425 | ||
426 | static kmutex_t ztest_vdev_lock; | |
427 | ||
428 | /* | |
429 | * The ztest_name_lock protects the pool and dataset namespace used by | |
430 | * the individual tests. To modify the namespace, consumers must grab | |
431 | * this lock as writer. Grabbing the lock as reader will ensure that the | |
432 | * namespace does not change while the lock is held. | |
433 | */ | |
434 | static rwlock_t ztest_name_lock; | |
435 | ||
436 | static boolean_t ztest_dump_core = B_TRUE; | |
437 | static boolean_t ztest_exiting; | |
438 | ||
439 | /* Global commit callback list */ | |
440 | static ztest_cb_list_t zcl; | |
441 | /* Commit cb delay */ | |
442 | static uint64_t zc_min_txg_delay = UINT64_MAX; | |
443 | static int zc_cb_counter = 0; | |
444 | ||
445 | /* | |
446 | * Minimum number of commit callbacks that need to be registered for us to check | |
447 | * whether the minimum txg delay is acceptable. | |
448 | */ | |
449 | #define ZTEST_COMMIT_CB_MIN_REG 100 | |
450 | ||
451 | /* | |
452 | * If a number of txgs equal to this threshold have been created after a commit | |
453 | * callback has been registered but not called, then we assume there is an | |
454 | * implementation bug. | |
455 | */ | |
456 | #define ZTEST_COMMIT_CB_THRESH (TXG_CONCURRENT_STATES + 1000) | |
457 | ||
458 | extern uint64_t metaslab_gang_bang; | |
459 | extern uint64_t metaslab_df_alloc_threshold; | |
460 | ||
461 | enum ztest_object { | |
462 | ZTEST_META_DNODE = 0, | |
463 | ZTEST_DIROBJ, | |
464 | ZTEST_OBJECTS | |
465 | }; | |
466 | ||
467 | static void usage(boolean_t) __NORETURN; | |
468 | ||
469 | /* | |
470 | * These libumem hooks provide a reasonable set of defaults for the allocator's | |
471 | * debugging facilities. | |
472 | */ | |
473 | const char * | |
474 | _umem_debug_init(void) | |
475 | { | |
476 | return ("default,verbose"); /* $UMEM_DEBUG setting */ | |
477 | } | |
478 | ||
479 | const char * | |
480 | _umem_logging_init(void) | |
481 | { | |
482 | return ("fail,contents"); /* $UMEM_LOGGING setting */ | |
483 | } | |
484 | ||
485 | #define BACKTRACE_SZ 100 | |
486 | ||
487 | static void sig_handler(int signo) | |
488 | { | |
489 | struct sigaction action; | |
490 | #ifdef __GLIBC__ /* backtrace() is a GNU extension */ | |
491 | int nptrs; | |
492 | void *buffer[BACKTRACE_SZ]; | |
493 | ||
494 | nptrs = backtrace(buffer, BACKTRACE_SZ); | |
495 | backtrace_symbols_fd(buffer, nptrs, STDERR_FILENO); | |
496 | #endif | |
497 | ||
498 | /* | |
499 | * Restore default action and re-raise signal so SIGSEGV and | |
500 | * SIGABRT can trigger a core dump. | |
501 | */ | |
502 | action.sa_handler = SIG_DFL; | |
503 | sigemptyset(&action.sa_mask); | |
504 | action.sa_flags = 0; | |
505 | (void) sigaction(signo, &action, NULL); | |
506 | raise(signo); | |
507 | } | |
508 | ||
509 | #define FATAL_MSG_SZ 1024 | |
510 | ||
511 | char *fatal_msg; | |
512 | ||
513 | static void | |
514 | fatal(int do_perror, char *message, ...) | |
515 | { | |
516 | va_list args; | |
517 | int save_errno = errno; | |
518 | char *buf; | |
519 | ||
520 | (void) fflush(stdout); | |
521 | buf = umem_alloc(FATAL_MSG_SZ, UMEM_NOFAIL); | |
522 | ||
523 | va_start(args, message); | |
524 | (void) sprintf(buf, "ztest: "); | |
525 | /* LINTED */ | |
526 | (void) vsprintf(buf + strlen(buf), message, args); | |
527 | va_end(args); | |
528 | if (do_perror) { | |
529 | (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf), | |
530 | ": %s", strerror(save_errno)); | |
531 | } | |
532 | (void) fprintf(stderr, "%s\n", buf); | |
533 | fatal_msg = buf; /* to ease debugging */ | |
534 | if (ztest_dump_core) | |
535 | abort(); | |
536 | exit(3); | |
537 | } | |
538 | ||
539 | static int | |
540 | str2shift(const char *buf) | |
541 | { | |
542 | const char *ends = "BKMGTPEZ"; | |
543 | int i; | |
544 | ||
545 | if (buf[0] == '\0') | |
546 | return (0); | |
547 | for (i = 0; i < strlen(ends); i++) { | |
548 | if (toupper(buf[0]) == ends[i]) | |
549 | break; | |
550 | } | |
551 | if (i == strlen(ends)) { | |
552 | (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", | |
553 | buf); | |
554 | usage(B_FALSE); | |
555 | } | |
556 | if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) { | |
557 | return (10*i); | |
558 | } | |
559 | (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf); | |
560 | usage(B_FALSE); | |
561 | /* NOTREACHED */ | |
562 | } | |
563 | ||
564 | static uint64_t | |
565 | nicenumtoull(const char *buf) | |
566 | { | |
567 | char *end; | |
568 | uint64_t val; | |
569 | ||
570 | val = strtoull(buf, &end, 0); | |
571 | if (end == buf) { | |
572 | (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf); | |
573 | usage(B_FALSE); | |
574 | } else if (end[0] == '.') { | |
575 | double fval = strtod(buf, &end); | |
576 | fval *= pow(2, str2shift(end)); | |
577 | if (fval > UINT64_MAX) { | |
578 | (void) fprintf(stderr, "ztest: value too large: %s\n", | |
579 | buf); | |
580 | usage(B_FALSE); | |
581 | } | |
582 | val = (uint64_t)fval; | |
583 | } else { | |
584 | int shift = str2shift(end); | |
585 | if (shift >= 64 || (val << shift) >> shift != val) { | |
586 | (void) fprintf(stderr, "ztest: value too large: %s\n", | |
587 | buf); | |
588 | usage(B_FALSE); | |
589 | } | |
590 | val <<= shift; | |
591 | } | |
592 | return (val); | |
593 | } | |
594 | ||
595 | static void | |
596 | usage(boolean_t requested) | |
597 | { | |
598 | const ztest_shared_opts_t *zo = &ztest_opts_defaults; | |
599 | ||
600 | char nice_vdev_size[10]; | |
601 | char nice_gang_bang[10]; | |
602 | FILE *fp = requested ? stdout : stderr; | |
603 | ||
604 | nicenum(zo->zo_vdev_size, nice_vdev_size); | |
605 | nicenum(zo->zo_metaslab_gang_bang, nice_gang_bang); | |
606 | ||
607 | (void) fprintf(fp, "Usage: %s\n" | |
608 | "\t[-v vdevs (default: %llu)]\n" | |
609 | "\t[-s size_of_each_vdev (default: %s)]\n" | |
610 | "\t[-a alignment_shift (default: %d)] use 0 for random\n" | |
611 | "\t[-m mirror_copies (default: %d)]\n" | |
612 | "\t[-r raidz_disks (default: %d)]\n" | |
613 | "\t[-R raidz_parity (default: %d)]\n" | |
614 | "\t[-d datasets (default: %d)]\n" | |
615 | "\t[-t threads (default: %d)]\n" | |
616 | "\t[-g gang_block_threshold (default: %s)]\n" | |
617 | "\t[-i init_count (default: %d)] initialize pool i times\n" | |
618 | "\t[-k kill_percentage (default: %llu%%)]\n" | |
619 | "\t[-p pool_name (default: %s)]\n" | |
620 | "\t[-f dir (default: %s)] file directory for vdev files\n" | |
621 | "\t[-V] verbose (use multiple times for ever more blather)\n" | |
622 | "\t[-E] use existing pool instead of creating new one\n" | |
623 | "\t[-T time (default: %llu sec)] total run time\n" | |
624 | "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n" | |
625 | "\t[-P passtime (default: %llu sec)] time per pass\n" | |
626 | "\t[-B alt_ztest (default: <none>)] alternate ztest path\n" | |
627 | "\t[-h] (print help)\n" | |
628 | "", | |
629 | zo->zo_pool, | |
630 | (u_longlong_t)zo->zo_vdevs, /* -v */ | |
631 | nice_vdev_size, /* -s */ | |
632 | zo->zo_ashift, /* -a */ | |
633 | zo->zo_mirrors, /* -m */ | |
634 | zo->zo_raidz, /* -r */ | |
635 | zo->zo_raidz_parity, /* -R */ | |
636 | zo->zo_datasets, /* -d */ | |
637 | zo->zo_threads, /* -t */ | |
638 | nice_gang_bang, /* -g */ | |
639 | zo->zo_init, /* -i */ | |
640 | (u_longlong_t)zo->zo_killrate, /* -k */ | |
641 | zo->zo_pool, /* -p */ | |
642 | zo->zo_dir, /* -f */ | |
643 | (u_longlong_t)zo->zo_time, /* -T */ | |
644 | (u_longlong_t)zo->zo_maxloops, /* -F */ | |
645 | (u_longlong_t)zo->zo_passtime); | |
646 | exit(requested ? 0 : 1); | |
647 | } | |
648 | ||
649 | static void | |
650 | process_options(int argc, char **argv) | |
651 | { | |
652 | char *path; | |
653 | ztest_shared_opts_t *zo = &ztest_opts; | |
654 | ||
655 | int opt; | |
656 | uint64_t value; | |
657 | char altdir[MAXNAMELEN] = { 0 }; | |
658 | ||
659 | bcopy(&ztest_opts_defaults, zo, sizeof (*zo)); | |
660 | ||
661 | while ((opt = getopt(argc, argv, | |
662 | "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:")) != EOF) { | |
663 | value = 0; | |
664 | switch (opt) { | |
665 | case 'v': | |
666 | case 's': | |
667 | case 'a': | |
668 | case 'm': | |
669 | case 'r': | |
670 | case 'R': | |
671 | case 'd': | |
672 | case 't': | |
673 | case 'g': | |
674 | case 'i': | |
675 | case 'k': | |
676 | case 'T': | |
677 | case 'P': | |
678 | case 'F': | |
679 | value = nicenumtoull(optarg); | |
680 | } | |
681 | switch (opt) { | |
682 | case 'v': | |
683 | zo->zo_vdevs = value; | |
684 | break; | |
685 | case 's': | |
686 | zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value); | |
687 | break; | |
688 | case 'a': | |
689 | zo->zo_ashift = value; | |
690 | break; | |
691 | case 'm': | |
692 | zo->zo_mirrors = value; | |
693 | break; | |
694 | case 'r': | |
695 | zo->zo_raidz = MAX(1, value); | |
696 | break; | |
697 | case 'R': | |
698 | zo->zo_raidz_parity = MIN(MAX(value, 1), 3); | |
699 | break; | |
700 | case 'd': | |
701 | zo->zo_datasets = MAX(1, value); | |
702 | break; | |
703 | case 't': | |
704 | zo->zo_threads = MAX(1, value); | |
705 | break; | |
706 | case 'g': | |
707 | zo->zo_metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1, | |
708 | value); | |
709 | break; | |
710 | case 'i': | |
711 | zo->zo_init = value; | |
712 | break; | |
713 | case 'k': | |
714 | zo->zo_killrate = value; | |
715 | break; | |
716 | case 'p': | |
717 | (void) strlcpy(zo->zo_pool, optarg, | |
718 | sizeof (zo->zo_pool)); | |
719 | break; | |
720 | case 'f': | |
721 | path = realpath(optarg, NULL); | |
722 | if (path == NULL) { | |
723 | (void) fprintf(stderr, "error: %s: %s\n", | |
724 | optarg, strerror(errno)); | |
725 | usage(B_FALSE); | |
726 | } else { | |
727 | (void) strlcpy(zo->zo_dir, path, | |
728 | sizeof (zo->zo_dir)); | |
729 | } | |
730 | break; | |
731 | case 'V': | |
732 | zo->zo_verbose++; | |
733 | break; | |
734 | case 'E': | |
735 | zo->zo_init = 0; | |
736 | break; | |
737 | case 'T': | |
738 | zo->zo_time = value; | |
739 | break; | |
740 | case 'P': | |
741 | zo->zo_passtime = MAX(1, value); | |
742 | break; | |
743 | case 'F': | |
744 | zo->zo_maxloops = MAX(1, value); | |
745 | break; | |
746 | case 'B': | |
747 | (void) strlcpy(altdir, optarg, sizeof (altdir)); | |
748 | break; | |
749 | case 'h': | |
750 | usage(B_TRUE); | |
751 | break; | |
752 | case '?': | |
753 | default: | |
754 | usage(B_FALSE); | |
755 | break; | |
756 | } | |
757 | } | |
758 | ||
759 | zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1); | |
760 | ||
761 | zo->zo_vdevtime = | |
762 | (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs : | |
763 | UINT64_MAX >> 2); | |
764 | ||
765 | if (strlen(altdir) > 0) { | |
766 | char *cmd; | |
767 | char *realaltdir; | |
768 | char *bin; | |
769 | char *ztest; | |
770 | char *isa; | |
771 | int isalen; | |
772 | ||
773 | cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); | |
774 | realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); | |
775 | ||
776 | VERIFY(NULL != realpath(getexecname(), cmd)); | |
777 | if (0 != access(altdir, F_OK)) { | |
778 | ztest_dump_core = B_FALSE; | |
779 | fatal(B_TRUE, "invalid alternate ztest path: %s", | |
780 | altdir); | |
781 | } | |
782 | VERIFY(NULL != realpath(altdir, realaltdir)); | |
783 | ||
784 | /* | |
785 | * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest". | |
786 | * We want to extract <isa> to determine if we should use | |
787 | * 32 or 64 bit binaries. | |
788 | */ | |
789 | bin = strstr(cmd, "/usr/bin/"); | |
790 | ztest = strstr(bin, "/ztest"); | |
791 | isa = bin + 9; | |
792 | isalen = ztest - isa; | |
793 | (void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest), | |
794 | "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa); | |
795 | (void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath), | |
796 | "%s/usr/lib/%.*s", realaltdir, isalen, isa); | |
797 | ||
798 | if (0 != access(zo->zo_alt_ztest, X_OK)) { | |
799 | ztest_dump_core = B_FALSE; | |
800 | fatal(B_TRUE, "invalid alternate ztest: %s", | |
801 | zo->zo_alt_ztest); | |
802 | } else if (0 != access(zo->zo_alt_libpath, X_OK)) { | |
803 | ztest_dump_core = B_FALSE; | |
804 | fatal(B_TRUE, "invalid alternate lib directory %s", | |
805 | zo->zo_alt_libpath); | |
806 | } | |
807 | ||
808 | umem_free(cmd, MAXPATHLEN); | |
809 | umem_free(realaltdir, MAXPATHLEN); | |
810 | } | |
811 | } | |
812 | ||
813 | static void | |
814 | ztest_kill(ztest_shared_t *zs) | |
815 | { | |
816 | zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa)); | |
817 | zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa)); | |
818 | ||
819 | /* | |
820 | * Before we kill off ztest, make sure that the config is updated. | |
821 | * See comment above spa_config_sync(). | |
822 | */ | |
823 | mutex_enter(&spa_namespace_lock); | |
824 | spa_config_sync(ztest_spa, B_FALSE, B_FALSE); | |
825 | mutex_exit(&spa_namespace_lock); | |
826 | ||
827 | (void) kill(getpid(), SIGKILL); | |
828 | } | |
829 | ||
830 | static uint64_t | |
831 | ztest_random(uint64_t range) | |
832 | { | |
833 | uint64_t r; | |
834 | ||
835 | ASSERT3S(ztest_fd_rand, >=, 0); | |
836 | ||
837 | if (range == 0) | |
838 | return (0); | |
839 | ||
840 | if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r)) | |
841 | fatal(1, "short read from /dev/urandom"); | |
842 | ||
843 | return (r % range); | |
844 | } | |
845 | ||
846 | /* ARGSUSED */ | |
847 | static void | |
848 | ztest_record_enospc(const char *s) | |
849 | { | |
850 | ztest_shared->zs_enospc_count++; | |
851 | } | |
852 | ||
853 | static uint64_t | |
854 | ztest_get_ashift(void) | |
855 | { | |
856 | if (ztest_opts.zo_ashift == 0) | |
857 | return (SPA_MINBLOCKSHIFT + ztest_random(5)); | |
858 | return (ztest_opts.zo_ashift); | |
859 | } | |
860 | ||
861 | static nvlist_t * | |
862 | make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift) | |
863 | { | |
864 | char *pathbuf; | |
865 | uint64_t vdev; | |
866 | nvlist_t *file; | |
867 | ||
868 | pathbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); | |
869 | ||
870 | if (ashift == 0) | |
871 | ashift = ztest_get_ashift(); | |
872 | ||
873 | if (path == NULL) { | |
874 | path = pathbuf; | |
875 | ||
876 | if (aux != NULL) { | |
877 | vdev = ztest_shared->zs_vdev_aux; | |
878 | (void) snprintf(path, MAXPATHLEN, | |
879 | ztest_aux_template, ztest_opts.zo_dir, | |
880 | pool == NULL ? ztest_opts.zo_pool : pool, | |
881 | aux, vdev); | |
882 | } else { | |
883 | vdev = ztest_shared->zs_vdev_next_leaf++; | |
884 | (void) snprintf(path, MAXPATHLEN, | |
885 | ztest_dev_template, ztest_opts.zo_dir, | |
886 | pool == NULL ? ztest_opts.zo_pool : pool, vdev); | |
887 | } | |
888 | } | |
889 | ||
890 | if (size != 0) { | |
891 | int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666); | |
892 | if (fd == -1) | |
893 | fatal(1, "can't open %s", path); | |
894 | if (ftruncate(fd, size) != 0) | |
895 | fatal(1, "can't ftruncate %s", path); | |
896 | (void) close(fd); | |
897 | } | |
898 | ||
899 | VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0); | |
900 | VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0); | |
901 | VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0); | |
902 | VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0); | |
903 | umem_free(pathbuf, MAXPATHLEN); | |
904 | ||
905 | return (file); | |
906 | } | |
907 | ||
908 | static nvlist_t * | |
909 | make_vdev_raidz(char *path, char *aux, char *pool, size_t size, | |
910 | uint64_t ashift, int r) | |
911 | { | |
912 | nvlist_t *raidz, **child; | |
913 | int c; | |
914 | ||
915 | if (r < 2) | |
916 | return (make_vdev_file(path, aux, pool, size, ashift)); | |
917 | child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL); | |
918 | ||
919 | for (c = 0; c < r; c++) | |
920 | child[c] = make_vdev_file(path, aux, pool, size, ashift); | |
921 | ||
922 | VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0); | |
923 | VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE, | |
924 | VDEV_TYPE_RAIDZ) == 0); | |
925 | VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY, | |
926 | ztest_opts.zo_raidz_parity) == 0); | |
927 | VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN, | |
928 | child, r) == 0); | |
929 | ||
930 | for (c = 0; c < r; c++) | |
931 | nvlist_free(child[c]); | |
932 | ||
933 | umem_free(child, r * sizeof (nvlist_t *)); | |
934 | ||
935 | return (raidz); | |
936 | } | |
937 | ||
938 | static nvlist_t * | |
939 | make_vdev_mirror(char *path, char *aux, char *pool, size_t size, | |
940 | uint64_t ashift, int r, int m) | |
941 | { | |
942 | nvlist_t *mirror, **child; | |
943 | int c; | |
944 | ||
945 | if (m < 1) | |
946 | return (make_vdev_raidz(path, aux, pool, size, ashift, r)); | |
947 | ||
948 | child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL); | |
949 | ||
950 | for (c = 0; c < m; c++) | |
951 | child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r); | |
952 | ||
953 | VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0); | |
954 | VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE, | |
955 | VDEV_TYPE_MIRROR) == 0); | |
956 | VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN, | |
957 | child, m) == 0); | |
958 | ||
959 | for (c = 0; c < m; c++) | |
960 | nvlist_free(child[c]); | |
961 | ||
962 | umem_free(child, m * sizeof (nvlist_t *)); | |
963 | ||
964 | return (mirror); | |
965 | } | |
966 | ||
967 | static nvlist_t * | |
968 | make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift, | |
969 | int log, int r, int m, int t) | |
970 | { | |
971 | nvlist_t *root, **child; | |
972 | int c; | |
973 | ||
974 | ASSERT(t > 0); | |
975 | ||
976 | child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL); | |
977 | ||
978 | for (c = 0; c < t; c++) { | |
979 | child[c] = make_vdev_mirror(path, aux, pool, size, ashift, | |
980 | r, m); | |
981 | VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG, | |
982 | log) == 0); | |
983 | } | |
984 | ||
985 | VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0); | |
986 | VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0); | |
987 | VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN, | |
988 | child, t) == 0); | |
989 | ||
990 | for (c = 0; c < t; c++) | |
991 | nvlist_free(child[c]); | |
992 | ||
993 | umem_free(child, t * sizeof (nvlist_t *)); | |
994 | ||
995 | return (root); | |
996 | } | |
997 | ||
998 | /* | |
999 | * Find a random spa version. Returns back a random spa version in the | |
1000 | * range [initial_version, SPA_VERSION_FEATURES]. | |
1001 | */ | |
1002 | static uint64_t | |
1003 | ztest_random_spa_version(uint64_t initial_version) | |
1004 | { | |
1005 | uint64_t version = initial_version; | |
1006 | ||
1007 | if (version <= SPA_VERSION_BEFORE_FEATURES) { | |
1008 | version = version + | |
1009 | ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1); | |
1010 | } | |
1011 | ||
1012 | if (version > SPA_VERSION_BEFORE_FEATURES) | |
1013 | version = SPA_VERSION_FEATURES; | |
1014 | ||
1015 | ASSERT(SPA_VERSION_IS_SUPPORTED(version)); | |
1016 | return (version); | |
1017 | } | |
1018 | ||
1019 | static int | |
1020 | ztest_random_blocksize(void) | |
1021 | { | |
1022 | /* | |
1023 | * Choose a block size >= the ashift. | |
1024 | * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks. | |
1025 | */ | |
1026 | int maxbs = SPA_OLD_MAXBLOCKSHIFT; | |
1027 | if (spa_maxblocksize(ztest_spa) == SPA_MAXBLOCKSIZE) | |
1028 | maxbs = 20; | |
1029 | uint64_t block_shift = | |
1030 | ztest_random(maxbs - ztest_spa->spa_max_ashift + 1); | |
1031 | return (1 << (SPA_MINBLOCKSHIFT + block_shift)); | |
1032 | } | |
1033 | ||
1034 | static int | |
1035 | ztest_random_dnodesize(void) | |
1036 | { | |
1037 | int slots; | |
1038 | int max_slots = spa_maxdnodesize(ztest_spa) >> DNODE_SHIFT; | |
1039 | ||
1040 | if (max_slots == DNODE_MIN_SLOTS) | |
1041 | return (DNODE_MIN_SIZE); | |
1042 | ||
1043 | /* | |
1044 | * Weight the random distribution more heavily toward smaller | |
1045 | * dnode sizes since that is more likely to reflect real-world | |
1046 | * usage. | |
1047 | */ | |
1048 | ASSERT3U(max_slots, >, 4); | |
1049 | switch (ztest_random(10)) { | |
1050 | case 0: | |
1051 | slots = 5 + ztest_random(max_slots - 4); | |
1052 | break; | |
1053 | case 1 ... 4: | |
1054 | slots = 2 + ztest_random(3); | |
1055 | break; | |
1056 | default: | |
1057 | slots = 1; | |
1058 | break; | |
1059 | } | |
1060 | ||
1061 | return (slots << DNODE_SHIFT); | |
1062 | } | |
1063 | ||
1064 | static int | |
1065 | ztest_random_ibshift(void) | |
1066 | { | |
1067 | return (DN_MIN_INDBLKSHIFT + | |
1068 | ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1)); | |
1069 | } | |
1070 | ||
1071 | static uint64_t | |
1072 | ztest_random_vdev_top(spa_t *spa, boolean_t log_ok) | |
1073 | { | |
1074 | uint64_t top; | |
1075 | vdev_t *rvd = spa->spa_root_vdev; | |
1076 | vdev_t *tvd; | |
1077 | ||
1078 | ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); | |
1079 | ||
1080 | do { | |
1081 | top = ztest_random(rvd->vdev_children); | |
1082 | tvd = rvd->vdev_child[top]; | |
1083 | } while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) || | |
1084 | tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL); | |
1085 | ||
1086 | return (top); | |
1087 | } | |
1088 | ||
1089 | static uint64_t | |
1090 | ztest_random_dsl_prop(zfs_prop_t prop) | |
1091 | { | |
1092 | uint64_t value; | |
1093 | ||
1094 | do { | |
1095 | value = zfs_prop_random_value(prop, ztest_random(-1ULL)); | |
1096 | } while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF); | |
1097 | ||
1098 | return (value); | |
1099 | } | |
1100 | ||
1101 | static int | |
1102 | ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value, | |
1103 | boolean_t inherit) | |
1104 | { | |
1105 | const char *propname = zfs_prop_to_name(prop); | |
1106 | const char *valname; | |
1107 | char *setpoint; | |
1108 | uint64_t curval; | |
1109 | int error; | |
1110 | ||
1111 | error = dsl_prop_set_int(osname, propname, | |
1112 | (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value); | |
1113 | ||
1114 | if (error == ENOSPC) { | |
1115 | ztest_record_enospc(FTAG); | |
1116 | return (error); | |
1117 | } | |
1118 | ASSERT0(error); | |
1119 | ||
1120 | setpoint = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); | |
1121 | VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint)); | |
1122 | ||
1123 | if (ztest_opts.zo_verbose >= 6) { | |
1124 | int err; | |
1125 | ||
1126 | err = zfs_prop_index_to_string(prop, curval, &valname); | |
1127 | if (err) | |
1128 | (void) printf("%s %s = %llu at '%s'\n", | |
1129 | osname, propname, (unsigned long long)curval, | |
1130 | setpoint); | |
1131 | else | |
1132 | (void) printf("%s %s = %s at '%s'\n", | |
1133 | osname, propname, valname, setpoint); | |
1134 | } | |
1135 | umem_free(setpoint, MAXPATHLEN); | |
1136 | ||
1137 | return (error); | |
1138 | } | |
1139 | ||
1140 | static int | |
1141 | ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value) | |
1142 | { | |
1143 | spa_t *spa = ztest_spa; | |
1144 | nvlist_t *props = NULL; | |
1145 | int error; | |
1146 | ||
1147 | VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0); | |
1148 | VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0); | |
1149 | ||
1150 | error = spa_prop_set(spa, props); | |
1151 | ||
1152 | nvlist_free(props); | |
1153 | ||
1154 | if (error == ENOSPC) { | |
1155 | ztest_record_enospc(FTAG); | |
1156 | return (error); | |
1157 | } | |
1158 | ASSERT0(error); | |
1159 | ||
1160 | return (error); | |
1161 | } | |
1162 | ||
1163 | ||
1164 | /* | |
1165 | * Object and range lock mechanics | |
1166 | */ | |
1167 | typedef struct { | |
1168 | list_node_t z_lnode; | |
1169 | refcount_t z_refcnt; | |
1170 | uint64_t z_object; | |
1171 | zfs_rlock_t z_range_lock; | |
1172 | } ztest_znode_t; | |
1173 | ||
1174 | typedef struct { | |
1175 | rl_t *z_rl; | |
1176 | ztest_znode_t *z_ztznode; | |
1177 | } ztest_zrl_t; | |
1178 | ||
1179 | static ztest_znode_t * | |
1180 | ztest_znode_init(uint64_t object) | |
1181 | { | |
1182 | ztest_znode_t *zp = umem_alloc(sizeof (*zp), UMEM_NOFAIL); | |
1183 | ||
1184 | list_link_init(&zp->z_lnode); | |
1185 | refcount_create(&zp->z_refcnt); | |
1186 | zp->z_object = object; | |
1187 | zfs_rlock_init(&zp->z_range_lock); | |
1188 | ||
1189 | return (zp); | |
1190 | } | |
1191 | ||
1192 | static void | |
1193 | ztest_znode_fini(ztest_znode_t *zp) | |
1194 | { | |
1195 | ASSERT(refcount_is_zero(&zp->z_refcnt)); | |
1196 | zfs_rlock_destroy(&zp->z_range_lock); | |
1197 | zp->z_object = 0; | |
1198 | refcount_destroy(&zp->z_refcnt); | |
1199 | list_link_init(&zp->z_lnode); | |
1200 | umem_free(zp, sizeof (*zp)); | |
1201 | } | |
1202 | ||
1203 | static void | |
1204 | ztest_zll_init(zll_t *zll) | |
1205 | { | |
1206 | mutex_init(&zll->z_lock, NULL, MUTEX_DEFAULT, NULL); | |
1207 | list_create(&zll->z_list, sizeof (ztest_znode_t), | |
1208 | offsetof(ztest_znode_t, z_lnode)); | |
1209 | } | |
1210 | ||
1211 | static void | |
1212 | ztest_zll_destroy(zll_t *zll) | |
1213 | { | |
1214 | list_destroy(&zll->z_list); | |
1215 | mutex_destroy(&zll->z_lock); | |
1216 | } | |
1217 | ||
1218 | #define RL_TAG "range_lock" | |
1219 | static ztest_znode_t * | |
1220 | ztest_znode_get(ztest_ds_t *zd, uint64_t object) | |
1221 | { | |
1222 | zll_t *zll = &zd->zd_range_lock[object & (ZTEST_OBJECT_LOCKS - 1)]; | |
1223 | ztest_znode_t *zp = NULL; | |
1224 | mutex_enter(&zll->z_lock); | |
1225 | for (zp = list_head(&zll->z_list); (zp); | |
1226 | zp = list_next(&zll->z_list, zp)) { | |
1227 | if (zp->z_object == object) { | |
1228 | refcount_add(&zp->z_refcnt, RL_TAG); | |
1229 | break; | |
1230 | } | |
1231 | } | |
1232 | if (zp == NULL) { | |
1233 | zp = ztest_znode_init(object); | |
1234 | refcount_add(&zp->z_refcnt, RL_TAG); | |
1235 | list_insert_head(&zll->z_list, zp); | |
1236 | } | |
1237 | mutex_exit(&zll->z_lock); | |
1238 | return (zp); | |
1239 | } | |
1240 | ||
1241 | static void | |
1242 | ztest_znode_put(ztest_ds_t *zd, ztest_znode_t *zp) | |
1243 | { | |
1244 | zll_t *zll = NULL; | |
1245 | ASSERT3U(zp->z_object, !=, 0); | |
1246 | zll = &zd->zd_range_lock[zp->z_object & (ZTEST_OBJECT_LOCKS - 1)]; | |
1247 | mutex_enter(&zll->z_lock); | |
1248 | refcount_remove(&zp->z_refcnt, RL_TAG); | |
1249 | if (refcount_is_zero(&zp->z_refcnt)) { | |
1250 | list_remove(&zll->z_list, zp); | |
1251 | ztest_znode_fini(zp); | |
1252 | } | |
1253 | mutex_exit(&zll->z_lock); | |
1254 | } | |
1255 | ||
1256 | ||
1257 | static void | |
1258 | ztest_rll_init(rll_t *rll) | |
1259 | { | |
1260 | rll->rll_writer = NULL; | |
1261 | rll->rll_readers = 0; | |
1262 | mutex_init(&rll->rll_lock, NULL, MUTEX_DEFAULT, NULL); | |
1263 | cv_init(&rll->rll_cv, NULL, CV_DEFAULT, NULL); | |
1264 | } | |
1265 | ||
1266 | static void | |
1267 | ztest_rll_destroy(rll_t *rll) | |
1268 | { | |
1269 | ASSERT(rll->rll_writer == NULL); | |
1270 | ASSERT(rll->rll_readers == 0); | |
1271 | mutex_destroy(&rll->rll_lock); | |
1272 | cv_destroy(&rll->rll_cv); | |
1273 | } | |
1274 | ||
1275 | static void | |
1276 | ztest_rll_lock(rll_t *rll, rl_type_t type) | |
1277 | { | |
1278 | mutex_enter(&rll->rll_lock); | |
1279 | ||
1280 | if (type == RL_READER) { | |
1281 | while (rll->rll_writer != NULL) | |
1282 | (void) cv_wait(&rll->rll_cv, &rll->rll_lock); | |
1283 | rll->rll_readers++; | |
1284 | } else { | |
1285 | while (rll->rll_writer != NULL || rll->rll_readers) | |
1286 | (void) cv_wait(&rll->rll_cv, &rll->rll_lock); | |
1287 | rll->rll_writer = curthread; | |
1288 | } | |
1289 | ||
1290 | mutex_exit(&rll->rll_lock); | |
1291 | } | |
1292 | ||
1293 | static void | |
1294 | ztest_rll_unlock(rll_t *rll) | |
1295 | { | |
1296 | mutex_enter(&rll->rll_lock); | |
1297 | ||
1298 | if (rll->rll_writer) { | |
1299 | ASSERT(rll->rll_readers == 0); | |
1300 | rll->rll_writer = NULL; | |
1301 | } else { | |
1302 | ASSERT(rll->rll_readers != 0); | |
1303 | ASSERT(rll->rll_writer == NULL); | |
1304 | rll->rll_readers--; | |
1305 | } | |
1306 | ||
1307 | if (rll->rll_writer == NULL && rll->rll_readers == 0) | |
1308 | cv_broadcast(&rll->rll_cv); | |
1309 | ||
1310 | mutex_exit(&rll->rll_lock); | |
1311 | } | |
1312 | ||
1313 | static void | |
1314 | ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type) | |
1315 | { | |
1316 | rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)]; | |
1317 | ||
1318 | ztest_rll_lock(rll, type); | |
1319 | } | |
1320 | ||
1321 | static void | |
1322 | ztest_object_unlock(ztest_ds_t *zd, uint64_t object) | |
1323 | { | |
1324 | rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)]; | |
1325 | ||
1326 | ztest_rll_unlock(rll); | |
1327 | } | |
1328 | ||
1329 | static ztest_zrl_t * | |
1330 | ztest_zrl_init(rl_t *rl, ztest_znode_t *zp) | |
1331 | { | |
1332 | ztest_zrl_t *zrl = umem_alloc(sizeof (*zrl), UMEM_NOFAIL); | |
1333 | zrl->z_rl = rl; | |
1334 | zrl->z_ztznode = zp; | |
1335 | return (zrl); | |
1336 | } | |
1337 | ||
1338 | static void | |
1339 | ztest_zrl_fini(ztest_zrl_t *zrl) | |
1340 | { | |
1341 | umem_free(zrl, sizeof (*zrl)); | |
1342 | } | |
1343 | ||
1344 | static ztest_zrl_t * | |
1345 | ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset, | |
1346 | uint64_t size, rl_type_t type) | |
1347 | { | |
1348 | ztest_znode_t *zp = ztest_znode_get(zd, object); | |
1349 | rl_t *rl = zfs_range_lock(&zp->z_range_lock, offset, | |
1350 | size, type); | |
1351 | return (ztest_zrl_init(rl, zp)); | |
1352 | } | |
1353 | ||
1354 | static void | |
1355 | ztest_range_unlock(ztest_ds_t *zd, ztest_zrl_t *zrl) | |
1356 | { | |
1357 | zfs_range_unlock(zrl->z_rl); | |
1358 | ztest_znode_put(zd, zrl->z_ztznode); | |
1359 | ztest_zrl_fini(zrl); | |
1360 | } | |
1361 | ||
1362 | static void | |
1363 | ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os) | |
1364 | { | |
1365 | zd->zd_os = os; | |
1366 | zd->zd_zilog = dmu_objset_zil(os); | |
1367 | zd->zd_shared = szd; | |
1368 | dmu_objset_name(os, zd->zd_name); | |
1369 | int l; | |
1370 | ||
1371 | if (zd->zd_shared != NULL) | |
1372 | zd->zd_shared->zd_seq = 0; | |
1373 | ||
1374 | VERIFY(rwlock_init(&zd->zd_zilog_lock, USYNC_THREAD, NULL) == 0); | |
1375 | mutex_init(&zd->zd_dirobj_lock, NULL, MUTEX_DEFAULT, NULL); | |
1376 | ||
1377 | for (l = 0; l < ZTEST_OBJECT_LOCKS; l++) | |
1378 | ztest_rll_init(&zd->zd_object_lock[l]); | |
1379 | ||
1380 | for (l = 0; l < ZTEST_RANGE_LOCKS; l++) | |
1381 | ztest_zll_init(&zd->zd_range_lock[l]); | |
1382 | } | |
1383 | ||
1384 | static void | |
1385 | ztest_zd_fini(ztest_ds_t *zd) | |
1386 | { | |
1387 | int l; | |
1388 | ||
1389 | mutex_destroy(&zd->zd_dirobj_lock); | |
1390 | (void) rwlock_destroy(&zd->zd_zilog_lock); | |
1391 | ||
1392 | for (l = 0; l < ZTEST_OBJECT_LOCKS; l++) | |
1393 | ztest_rll_destroy(&zd->zd_object_lock[l]); | |
1394 | ||
1395 | for (l = 0; l < ZTEST_RANGE_LOCKS; l++) | |
1396 | ztest_zll_destroy(&zd->zd_range_lock[l]); | |
1397 | } | |
1398 | ||
1399 | #define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT) | |
1400 | ||
1401 | static uint64_t | |
1402 | ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag) | |
1403 | { | |
1404 | uint64_t txg; | |
1405 | int error; | |
1406 | ||
1407 | /* | |
1408 | * Attempt to assign tx to some transaction group. | |
1409 | */ | |
1410 | error = dmu_tx_assign(tx, txg_how); | |
1411 | if (error) { | |
1412 | if (error == ERESTART) { | |
1413 | ASSERT(txg_how == TXG_NOWAIT); | |
1414 | dmu_tx_wait(tx); | |
1415 | } else { | |
1416 | ASSERT3U(error, ==, ENOSPC); | |
1417 | ztest_record_enospc(tag); | |
1418 | } | |
1419 | dmu_tx_abort(tx); | |
1420 | return (0); | |
1421 | } | |
1422 | txg = dmu_tx_get_txg(tx); | |
1423 | ASSERT(txg != 0); | |
1424 | return (txg); | |
1425 | } | |
1426 | ||
1427 | static void | |
1428 | ztest_pattern_set(void *buf, uint64_t size, uint64_t value) | |
1429 | { | |
1430 | uint64_t *ip = buf; | |
1431 | uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size); | |
1432 | ||
1433 | while (ip < ip_end) | |
1434 | *ip++ = value; | |
1435 | } | |
1436 | ||
1437 | #ifndef NDEBUG | |
1438 | static boolean_t | |
1439 | ztest_pattern_match(void *buf, uint64_t size, uint64_t value) | |
1440 | { | |
1441 | uint64_t *ip = buf; | |
1442 | uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size); | |
1443 | uint64_t diff = 0; | |
1444 | ||
1445 | while (ip < ip_end) | |
1446 | diff |= (value - *ip++); | |
1447 | ||
1448 | return (diff == 0); | |
1449 | } | |
1450 | #endif | |
1451 | ||
1452 | static void | |
1453 | ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object, | |
1454 | uint64_t dnodesize, uint64_t offset, uint64_t gen, uint64_t txg, | |
1455 | uint64_t crtxg) | |
1456 | { | |
1457 | bt->bt_magic = BT_MAGIC; | |
1458 | bt->bt_objset = dmu_objset_id(os); | |
1459 | bt->bt_object = object; | |
1460 | bt->bt_dnodesize = dnodesize; | |
1461 | bt->bt_offset = offset; | |
1462 | bt->bt_gen = gen; | |
1463 | bt->bt_txg = txg; | |
1464 | bt->bt_crtxg = crtxg; | |
1465 | } | |
1466 | ||
1467 | static void | |
1468 | ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object, | |
1469 | uint64_t dnodesize, uint64_t offset, uint64_t gen, uint64_t txg, | |
1470 | uint64_t crtxg) | |
1471 | { | |
1472 | ASSERT3U(bt->bt_magic, ==, BT_MAGIC); | |
1473 | ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os)); | |
1474 | ASSERT3U(bt->bt_object, ==, object); | |
1475 | ASSERT3U(bt->bt_dnodesize, ==, dnodesize); | |
1476 | ASSERT3U(bt->bt_offset, ==, offset); | |
1477 | ASSERT3U(bt->bt_gen, <=, gen); | |
1478 | ASSERT3U(bt->bt_txg, <=, txg); | |
1479 | ASSERT3U(bt->bt_crtxg, ==, crtxg); | |
1480 | } | |
1481 | ||
1482 | static ztest_block_tag_t * | |
1483 | ztest_bt_bonus(dmu_buf_t *db) | |
1484 | { | |
1485 | dmu_object_info_t doi; | |
1486 | ztest_block_tag_t *bt; | |
1487 | ||
1488 | dmu_object_info_from_db(db, &doi); | |
1489 | ASSERT3U(doi.doi_bonus_size, <=, db->db_size); | |
1490 | ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt)); | |
1491 | bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt)); | |
1492 | ||
1493 | return (bt); | |
1494 | } | |
1495 | ||
1496 | /* | |
1497 | * Generate a token to fill up unused bonus buffer space. Try to make | |
1498 | * it unique to the object, generation, and offset to verify that data | |
1499 | * is not getting overwritten by data from other dnodes. | |
1500 | */ | |
1501 | #define ZTEST_BONUS_FILL_TOKEN(obj, ds, gen, offset) \ | |
1502 | (((ds) << 48) | ((gen) << 32) | ((obj) << 8) | (offset)) | |
1503 | ||
1504 | /* | |
1505 | * Fill up the unused bonus buffer region before the block tag with a | |
1506 | * verifiable pattern. Filling the whole bonus area with non-zero data | |
1507 | * helps ensure that all dnode traversal code properly skips the | |
1508 | * interior regions of large dnodes. | |
1509 | */ | |
1510 | void | |
1511 | ztest_fill_unused_bonus(dmu_buf_t *db, void *end, uint64_t obj, | |
1512 | objset_t *os, uint64_t gen) | |
1513 | { | |
1514 | uint64_t *bonusp; | |
1515 | ||
1516 | ASSERT(IS_P2ALIGNED((char *)end - (char *)db->db_data, 8)); | |
1517 | ||
1518 | for (bonusp = db->db_data; bonusp < (uint64_t *)end; bonusp++) { | |
1519 | uint64_t token = ZTEST_BONUS_FILL_TOKEN(obj, dmu_objset_id(os), | |
1520 | gen, bonusp - (uint64_t *)db->db_data); | |
1521 | *bonusp = token; | |
1522 | } | |
1523 | } | |
1524 | ||
1525 | /* | |
1526 | * Verify that the unused area of a bonus buffer is filled with the | |
1527 | * expected tokens. | |
1528 | */ | |
1529 | void | |
1530 | ztest_verify_unused_bonus(dmu_buf_t *db, void *end, uint64_t obj, | |
1531 | objset_t *os, uint64_t gen) | |
1532 | { | |
1533 | uint64_t *bonusp; | |
1534 | ||
1535 | for (bonusp = db->db_data; bonusp < (uint64_t *)end; bonusp++) { | |
1536 | uint64_t token = ZTEST_BONUS_FILL_TOKEN(obj, dmu_objset_id(os), | |
1537 | gen, bonusp - (uint64_t *)db->db_data); | |
1538 | VERIFY3U(*bonusp, ==, token); | |
1539 | } | |
1540 | } | |
1541 | ||
1542 | /* | |
1543 | * ZIL logging ops | |
1544 | */ | |
1545 | ||
1546 | #define lrz_type lr_mode | |
1547 | #define lrz_blocksize lr_uid | |
1548 | #define lrz_ibshift lr_gid | |
1549 | #define lrz_bonustype lr_rdev | |
1550 | #define lrz_dnodesize lr_crtime[1] | |
1551 | ||
1552 | static void | |
1553 | ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr) | |
1554 | { | |
1555 | char *name = (void *)(lr + 1); /* name follows lr */ | |
1556 | size_t namesize = strlen(name) + 1; | |
1557 | itx_t *itx; | |
1558 | ||
1559 | if (zil_replaying(zd->zd_zilog, tx)) | |
1560 | return; | |
1561 | ||
1562 | itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize); | |
1563 | bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, | |
1564 | sizeof (*lr) + namesize - sizeof (lr_t)); | |
1565 | ||
1566 | zil_itx_assign(zd->zd_zilog, itx, tx); | |
1567 | } | |
1568 | ||
1569 | static void | |
1570 | ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object) | |
1571 | { | |
1572 | char *name = (void *)(lr + 1); /* name follows lr */ | |
1573 | size_t namesize = strlen(name) + 1; | |
1574 | itx_t *itx; | |
1575 | ||
1576 | if (zil_replaying(zd->zd_zilog, tx)) | |
1577 | return; | |
1578 | ||
1579 | itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize); | |
1580 | bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, | |
1581 | sizeof (*lr) + namesize - sizeof (lr_t)); | |
1582 | ||
1583 | itx->itx_oid = object; | |
1584 | zil_itx_assign(zd->zd_zilog, itx, tx); | |
1585 | } | |
1586 | ||
1587 | static void | |
1588 | ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr) | |
1589 | { | |
1590 | itx_t *itx; | |
1591 | itx_wr_state_t write_state = ztest_random(WR_NUM_STATES); | |
1592 | ||
1593 | if (zil_replaying(zd->zd_zilog, tx)) | |
1594 | return; | |
1595 | ||
1596 | if (lr->lr_length > ZIL_MAX_LOG_DATA) | |
1597 | write_state = WR_INDIRECT; | |
1598 | ||
1599 | itx = zil_itx_create(TX_WRITE, | |
1600 | sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0)); | |
1601 | ||
1602 | if (write_state == WR_COPIED && | |
1603 | dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length, | |
1604 | ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) { | |
1605 | zil_itx_destroy(itx); | |
1606 | itx = zil_itx_create(TX_WRITE, sizeof (*lr)); | |
1607 | write_state = WR_NEED_COPY; | |
1608 | } | |
1609 | itx->itx_private = zd; | |
1610 | itx->itx_wr_state = write_state; | |
1611 | itx->itx_sync = (ztest_random(8) == 0); | |
1612 | itx->itx_sod += (write_state == WR_NEED_COPY ? lr->lr_length : 0); | |
1613 | ||
1614 | bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, | |
1615 | sizeof (*lr) - sizeof (lr_t)); | |
1616 | ||
1617 | zil_itx_assign(zd->zd_zilog, itx, tx); | |
1618 | } | |
1619 | ||
1620 | static void | |
1621 | ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr) | |
1622 | { | |
1623 | itx_t *itx; | |
1624 | ||
1625 | if (zil_replaying(zd->zd_zilog, tx)) | |
1626 | return; | |
1627 | ||
1628 | itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr)); | |
1629 | bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, | |
1630 | sizeof (*lr) - sizeof (lr_t)); | |
1631 | ||
1632 | itx->itx_sync = B_FALSE; | |
1633 | zil_itx_assign(zd->zd_zilog, itx, tx); | |
1634 | } | |
1635 | ||
1636 | static void | |
1637 | ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr) | |
1638 | { | |
1639 | itx_t *itx; | |
1640 | ||
1641 | if (zil_replaying(zd->zd_zilog, tx)) | |
1642 | return; | |
1643 | ||
1644 | itx = zil_itx_create(TX_SETATTR, sizeof (*lr)); | |
1645 | bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, | |
1646 | sizeof (*lr) - sizeof (lr_t)); | |
1647 | ||
1648 | itx->itx_sync = B_FALSE; | |
1649 | zil_itx_assign(zd->zd_zilog, itx, tx); | |
1650 | } | |
1651 | ||
1652 | /* | |
1653 | * ZIL replay ops | |
1654 | */ | |
1655 | static int | |
1656 | ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap) | |
1657 | { | |
1658 | char *name = (void *)(lr + 1); /* name follows lr */ | |
1659 | objset_t *os = zd->zd_os; | |
1660 | ztest_block_tag_t *bbt; | |
1661 | dmu_buf_t *db; | |
1662 | dmu_tx_t *tx; | |
1663 | uint64_t txg; | |
1664 | int error = 0; | |
1665 | int bonuslen; | |
1666 | ||
1667 | if (byteswap) | |
1668 | byteswap_uint64_array(lr, sizeof (*lr)); | |
1669 | ||
1670 | ASSERT(lr->lr_doid == ZTEST_DIROBJ); | |
1671 | ASSERT(name[0] != '\0'); | |
1672 | ||
1673 | tx = dmu_tx_create(os); | |
1674 | ||
1675 | dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name); | |
1676 | ||
1677 | if (lr->lrz_type == DMU_OT_ZAP_OTHER) { | |
1678 | dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL); | |
1679 | } else { | |
1680 | dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); | |
1681 | } | |
1682 | ||
1683 | txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); | |
1684 | if (txg == 0) | |
1685 | return (ENOSPC); | |
1686 | ||
1687 | ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid); | |
1688 | bonuslen = DN_BONUS_SIZE(lr->lrz_dnodesize); | |
1689 | ||
1690 | if (lr->lrz_type == DMU_OT_ZAP_OTHER) { | |
1691 | if (lr->lr_foid == 0) { | |
1692 | lr->lr_foid = zap_create_dnsize(os, | |
1693 | lr->lrz_type, lr->lrz_bonustype, | |
1694 | bonuslen, lr->lrz_dnodesize, tx); | |
1695 | } else { | |
1696 | error = zap_create_claim_dnsize(os, lr->lr_foid, | |
1697 | lr->lrz_type, lr->lrz_bonustype, | |
1698 | bonuslen, lr->lrz_dnodesize, tx); | |
1699 | } | |
1700 | } else { | |
1701 | if (lr->lr_foid == 0) { | |
1702 | lr->lr_foid = dmu_object_alloc_dnsize(os, | |
1703 | lr->lrz_type, 0, lr->lrz_bonustype, | |
1704 | bonuslen, lr->lrz_dnodesize, tx); | |
1705 | } else { | |
1706 | error = dmu_object_claim_dnsize(os, lr->lr_foid, | |
1707 | lr->lrz_type, 0, lr->lrz_bonustype, | |
1708 | bonuslen, lr->lrz_dnodesize, tx); | |
1709 | } | |
1710 | } | |
1711 | ||
1712 | if (error) { | |
1713 | ASSERT3U(error, ==, EEXIST); | |
1714 | ASSERT(zd->zd_zilog->zl_replay); | |
1715 | dmu_tx_commit(tx); | |
1716 | return (error); | |
1717 | } | |
1718 | ||
1719 | ASSERT(lr->lr_foid != 0); | |
1720 | ||
1721 | if (lr->lrz_type != DMU_OT_ZAP_OTHER) | |
1722 | VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid, | |
1723 | lr->lrz_blocksize, lr->lrz_ibshift, tx)); | |
1724 | ||
1725 | VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db)); | |
1726 | bbt = ztest_bt_bonus(db); | |
1727 | dmu_buf_will_dirty(db, tx); | |
1728 | ztest_bt_generate(bbt, os, lr->lr_foid, lr->lrz_dnodesize, -1ULL, | |
1729 | lr->lr_gen, txg, txg); | |
1730 | ztest_fill_unused_bonus(db, bbt, lr->lr_foid, os, lr->lr_gen); | |
1731 | dmu_buf_rele(db, FTAG); | |
1732 | ||
1733 | VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1, | |
1734 | &lr->lr_foid, tx)); | |
1735 | ||
1736 | (void) ztest_log_create(zd, tx, lr); | |
1737 | ||
1738 | dmu_tx_commit(tx); | |
1739 | ||
1740 | return (0); | |
1741 | } | |
1742 | ||
1743 | static int | |
1744 | ztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap) | |
1745 | { | |
1746 | char *name = (void *)(lr + 1); /* name follows lr */ | |
1747 | objset_t *os = zd->zd_os; | |
1748 | dmu_object_info_t doi; | |
1749 | dmu_tx_t *tx; | |
1750 | uint64_t object, txg; | |
1751 | ||
1752 | if (byteswap) | |
1753 | byteswap_uint64_array(lr, sizeof (*lr)); | |
1754 | ||
1755 | ASSERT(lr->lr_doid == ZTEST_DIROBJ); | |
1756 | ASSERT(name[0] != '\0'); | |
1757 | ||
1758 | VERIFY3U(0, ==, | |
1759 | zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object)); | |
1760 | ASSERT(object != 0); | |
1761 | ||
1762 | ztest_object_lock(zd, object, RL_WRITER); | |
1763 | ||
1764 | VERIFY3U(0, ==, dmu_object_info(os, object, &doi)); | |
1765 | ||
1766 | tx = dmu_tx_create(os); | |
1767 | ||
1768 | dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name); | |
1769 | dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END); | |
1770 | ||
1771 | txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); | |
1772 | if (txg == 0) { | |
1773 | ztest_object_unlock(zd, object); | |
1774 | return (ENOSPC); | |
1775 | } | |
1776 | ||
1777 | if (doi.doi_type == DMU_OT_ZAP_OTHER) { | |
1778 | VERIFY3U(0, ==, zap_destroy(os, object, tx)); | |
1779 | } else { | |
1780 | VERIFY3U(0, ==, dmu_object_free(os, object, tx)); | |
1781 | } | |
1782 | ||
1783 | VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx)); | |
1784 | ||
1785 | (void) ztest_log_remove(zd, tx, lr, object); | |
1786 | ||
1787 | dmu_tx_commit(tx); | |
1788 | ||
1789 | ztest_object_unlock(zd, object); | |
1790 | ||
1791 | return (0); | |
1792 | } | |
1793 | ||
1794 | static int | |
1795 | ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap) | |
1796 | { | |
1797 | objset_t *os = zd->zd_os; | |
1798 | void *data = lr + 1; /* data follows lr */ | |
1799 | uint64_t offset, length; | |
1800 | ztest_block_tag_t *bt = data; | |
1801 | ztest_block_tag_t *bbt; | |
1802 | uint64_t gen, txg, lrtxg, crtxg; | |
1803 | dmu_object_info_t doi; | |
1804 | dmu_tx_t *tx; | |
1805 | dmu_buf_t *db; | |
1806 | arc_buf_t *abuf = NULL; | |
1807 | ztest_zrl_t *rl; | |
1808 | ||
1809 | if (byteswap) | |
1810 | byteswap_uint64_array(lr, sizeof (*lr)); | |
1811 | ||
1812 | offset = lr->lr_offset; | |
1813 | length = lr->lr_length; | |
1814 | ||
1815 | /* If it's a dmu_sync() block, write the whole block */ | |
1816 | if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) { | |
1817 | uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr); | |
1818 | if (length < blocksize) { | |
1819 | offset -= offset % blocksize; | |
1820 | length = blocksize; | |
1821 | } | |
1822 | } | |
1823 | ||
1824 | if (bt->bt_magic == BSWAP_64(BT_MAGIC)) | |
1825 | byteswap_uint64_array(bt, sizeof (*bt)); | |
1826 | ||
1827 | if (bt->bt_magic != BT_MAGIC) | |
1828 | bt = NULL; | |
1829 | ||
1830 | ztest_object_lock(zd, lr->lr_foid, RL_READER); | |
1831 | rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER); | |
1832 | ||
1833 | VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db)); | |
1834 | ||
1835 | dmu_object_info_from_db(db, &doi); | |
1836 | ||
1837 | bbt = ztest_bt_bonus(db); | |
1838 | ASSERT3U(bbt->bt_magic, ==, BT_MAGIC); | |
1839 | gen = bbt->bt_gen; | |
1840 | crtxg = bbt->bt_crtxg; | |
1841 | lrtxg = lr->lr_common.lrc_txg; | |
1842 | ||
1843 | tx = dmu_tx_create(os); | |
1844 | ||
1845 | dmu_tx_hold_write(tx, lr->lr_foid, offset, length); | |
1846 | ||
1847 | if (ztest_random(8) == 0 && length == doi.doi_data_block_size && | |
1848 | P2PHASE(offset, length) == 0) | |
1849 | abuf = dmu_request_arcbuf(db, length); | |
1850 | ||
1851 | txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); | |
1852 | if (txg == 0) { | |
1853 | if (abuf != NULL) | |
1854 | dmu_return_arcbuf(abuf); | |
1855 | dmu_buf_rele(db, FTAG); | |
1856 | ztest_range_unlock(zd, rl); | |
1857 | ztest_object_unlock(zd, lr->lr_foid); | |
1858 | return (ENOSPC); | |
1859 | } | |
1860 | ||
1861 | if (bt != NULL) { | |
1862 | /* | |
1863 | * Usually, verify the old data before writing new data -- | |
1864 | * but not always, because we also want to verify correct | |
1865 | * behavior when the data was not recently read into cache. | |
1866 | */ | |
1867 | ASSERT(offset % doi.doi_data_block_size == 0); | |
1868 | if (ztest_random(4) != 0) { | |
1869 | int prefetch = ztest_random(2) ? | |
1870 | DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH; | |
1871 | ztest_block_tag_t rbt; | |
1872 | ||
1873 | VERIFY(dmu_read(os, lr->lr_foid, offset, | |
1874 | sizeof (rbt), &rbt, prefetch) == 0); | |
1875 | if (rbt.bt_magic == BT_MAGIC) { | |
1876 | ztest_bt_verify(&rbt, os, lr->lr_foid, 0, | |
1877 | offset, gen, txg, crtxg); | |
1878 | } | |
1879 | } | |
1880 | ||
1881 | /* | |
1882 | * Writes can appear to be newer than the bonus buffer because | |
1883 | * the ztest_get_data() callback does a dmu_read() of the | |
1884 | * open-context data, which may be different than the data | |
1885 | * as it was when the write was generated. | |
1886 | */ | |
1887 | if (zd->zd_zilog->zl_replay) { | |
1888 | ztest_bt_verify(bt, os, lr->lr_foid, 0, offset, | |
1889 | MAX(gen, bt->bt_gen), MAX(txg, lrtxg), | |
1890 | bt->bt_crtxg); | |
1891 | } | |
1892 | ||
1893 | /* | |
1894 | * Set the bt's gen/txg to the bonus buffer's gen/txg | |
1895 | * so that all of the usual ASSERTs will work. | |
1896 | */ | |
1897 | ztest_bt_generate(bt, os, lr->lr_foid, 0, offset, gen, txg, | |
1898 | crtxg); | |
1899 | } | |
1900 | ||
1901 | if (abuf == NULL) { | |
1902 | dmu_write(os, lr->lr_foid, offset, length, data, tx); | |
1903 | } else { | |
1904 | bcopy(data, abuf->b_data, length); | |
1905 | dmu_assign_arcbuf(db, offset, abuf, tx); | |
1906 | } | |
1907 | ||
1908 | (void) ztest_log_write(zd, tx, lr); | |
1909 | ||
1910 | dmu_buf_rele(db, FTAG); | |
1911 | ||
1912 | dmu_tx_commit(tx); | |
1913 | ||
1914 | ztest_range_unlock(zd, rl); | |
1915 | ztest_object_unlock(zd, lr->lr_foid); | |
1916 | ||
1917 | return (0); | |
1918 | } | |
1919 | ||
1920 | static int | |
1921 | ztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap) | |
1922 | { | |
1923 | objset_t *os = zd->zd_os; | |
1924 | dmu_tx_t *tx; | |
1925 | uint64_t txg; | |
1926 | ztest_zrl_t *rl; | |
1927 | ||
1928 | if (byteswap) | |
1929 | byteswap_uint64_array(lr, sizeof (*lr)); | |
1930 | ||
1931 | ztest_object_lock(zd, lr->lr_foid, RL_READER); | |
1932 | rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length, | |
1933 | RL_WRITER); | |
1934 | ||
1935 | tx = dmu_tx_create(os); | |
1936 | ||
1937 | dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length); | |
1938 | ||
1939 | txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); | |
1940 | if (txg == 0) { | |
1941 | ztest_range_unlock(zd, rl); | |
1942 | ztest_object_unlock(zd, lr->lr_foid); | |
1943 | return (ENOSPC); | |
1944 | } | |
1945 | ||
1946 | VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset, | |
1947 | lr->lr_length, tx) == 0); | |
1948 | ||
1949 | (void) ztest_log_truncate(zd, tx, lr); | |
1950 | ||
1951 | dmu_tx_commit(tx); | |
1952 | ||
1953 | ztest_range_unlock(zd, rl); | |
1954 | ztest_object_unlock(zd, lr->lr_foid); | |
1955 | ||
1956 | return (0); | |
1957 | } | |
1958 | ||
1959 | static int | |
1960 | ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap) | |
1961 | { | |
1962 | objset_t *os = zd->zd_os; | |
1963 | dmu_tx_t *tx; | |
1964 | dmu_buf_t *db; | |
1965 | ztest_block_tag_t *bbt; | |
1966 | uint64_t txg, lrtxg, crtxg, dnodesize; | |
1967 | ||
1968 | if (byteswap) | |
1969 | byteswap_uint64_array(lr, sizeof (*lr)); | |
1970 | ||
1971 | ztest_object_lock(zd, lr->lr_foid, RL_WRITER); | |
1972 | ||
1973 | VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db)); | |
1974 | ||
1975 | tx = dmu_tx_create(os); | |
1976 | dmu_tx_hold_bonus(tx, lr->lr_foid); | |
1977 | ||
1978 | txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); | |
1979 | if (txg == 0) { | |
1980 | dmu_buf_rele(db, FTAG); | |
1981 | ztest_object_unlock(zd, lr->lr_foid); | |
1982 | return (ENOSPC); | |
1983 | } | |
1984 | ||
1985 | bbt = ztest_bt_bonus(db); | |
1986 | ASSERT3U(bbt->bt_magic, ==, BT_MAGIC); | |
1987 | crtxg = bbt->bt_crtxg; | |
1988 | lrtxg = lr->lr_common.lrc_txg; | |
1989 | dnodesize = bbt->bt_dnodesize; | |
1990 | ||
1991 | if (zd->zd_zilog->zl_replay) { | |
1992 | ASSERT(lr->lr_size != 0); | |
1993 | ASSERT(lr->lr_mode != 0); | |
1994 | ASSERT(lrtxg != 0); | |
1995 | } else { | |
1996 | /* | |
1997 | * Randomly change the size and increment the generation. | |
1998 | */ | |
1999 | lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) * | |
2000 | sizeof (*bbt); | |
2001 | lr->lr_mode = bbt->bt_gen + 1; | |
2002 | ASSERT(lrtxg == 0); | |
2003 | } | |
2004 | ||
2005 | /* | |
2006 | * Verify that the current bonus buffer is not newer than our txg. | |
2007 | */ | |
2008 | ztest_bt_verify(bbt, os, lr->lr_foid, dnodesize, -1ULL, lr->lr_mode, | |
2009 | MAX(txg, lrtxg), crtxg); | |
2010 | ||
2011 | dmu_buf_will_dirty(db, tx); | |
2012 | ||
2013 | ASSERT3U(lr->lr_size, >=, sizeof (*bbt)); | |
2014 | ASSERT3U(lr->lr_size, <=, db->db_size); | |
2015 | VERIFY0(dmu_set_bonus(db, lr->lr_size, tx)); | |
2016 | bbt = ztest_bt_bonus(db); | |
2017 | ||
2018 | ztest_bt_generate(bbt, os, lr->lr_foid, dnodesize, -1ULL, lr->lr_mode, | |
2019 | txg, crtxg); | |
2020 | ztest_fill_unused_bonus(db, bbt, lr->lr_foid, os, bbt->bt_gen); | |
2021 | dmu_buf_rele(db, FTAG); | |
2022 | ||
2023 | (void) ztest_log_setattr(zd, tx, lr); | |
2024 | ||
2025 | dmu_tx_commit(tx); | |
2026 | ||
2027 | ztest_object_unlock(zd, lr->lr_foid); | |
2028 | ||
2029 | return (0); | |
2030 | } | |
2031 | ||
2032 | zil_replay_func_t ztest_replay_vector[TX_MAX_TYPE] = { | |
2033 | NULL, /* 0 no such transaction type */ | |
2034 | (zil_replay_func_t)ztest_replay_create, /* TX_CREATE */ | |
2035 | NULL, /* TX_MKDIR */ | |
2036 | NULL, /* TX_MKXATTR */ | |
2037 | NULL, /* TX_SYMLINK */ | |
2038 | (zil_replay_func_t)ztest_replay_remove, /* TX_REMOVE */ | |
2039 | NULL, /* TX_RMDIR */ | |
2040 | NULL, /* TX_LINK */ | |
2041 | NULL, /* TX_RENAME */ | |
2042 | (zil_replay_func_t)ztest_replay_write, /* TX_WRITE */ | |
2043 | (zil_replay_func_t)ztest_replay_truncate, /* TX_TRUNCATE */ | |
2044 | (zil_replay_func_t)ztest_replay_setattr, /* TX_SETATTR */ | |
2045 | NULL, /* TX_ACL */ | |
2046 | NULL, /* TX_CREATE_ACL */ | |
2047 | NULL, /* TX_CREATE_ATTR */ | |
2048 | NULL, /* TX_CREATE_ACL_ATTR */ | |
2049 | NULL, /* TX_MKDIR_ACL */ | |
2050 | NULL, /* TX_MKDIR_ATTR */ | |
2051 | NULL, /* TX_MKDIR_ACL_ATTR */ | |
2052 | NULL, /* TX_WRITE2 */ | |
2053 | }; | |
2054 | ||
2055 | /* | |
2056 | * ZIL get_data callbacks | |
2057 | */ | |
2058 | typedef struct ztest_zgd_private { | |
2059 | ztest_ds_t *z_zd; | |
2060 | ztest_zrl_t *z_rl; | |
2061 | uint64_t z_object; | |
2062 | } ztest_zgd_private_t; | |
2063 | ||
2064 | static void | |
2065 | ztest_get_done(zgd_t *zgd, int error) | |
2066 | { | |
2067 | ztest_zgd_private_t *zzp = zgd->zgd_private; | |
2068 | ztest_ds_t *zd = zzp->z_zd; | |
2069 | uint64_t object = zzp->z_object; | |
2070 | ||
2071 | if (zgd->zgd_db) | |
2072 | dmu_buf_rele(zgd->zgd_db, zgd); | |
2073 | ||
2074 | ztest_range_unlock(zd, zzp->z_rl); | |
2075 | ztest_object_unlock(zd, object); | |
2076 | ||
2077 | if (error == 0 && zgd->zgd_bp) | |
2078 | zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); | |
2079 | ||
2080 | umem_free(zgd, sizeof (*zgd)); | |
2081 | umem_free(zzp, sizeof (*zzp)); | |
2082 | } | |
2083 | ||
2084 | static int | |
2085 | ztest_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) | |
2086 | { | |
2087 | ztest_ds_t *zd = arg; | |
2088 | objset_t *os = zd->zd_os; | |
2089 | uint64_t object = lr->lr_foid; | |
2090 | uint64_t offset = lr->lr_offset; | |
2091 | uint64_t size = lr->lr_length; | |
2092 | blkptr_t *bp = &lr->lr_blkptr; | |
2093 | uint64_t txg = lr->lr_common.lrc_txg; | |
2094 | uint64_t crtxg; | |
2095 | dmu_object_info_t doi; | |
2096 | dmu_buf_t *db; | |
2097 | zgd_t *zgd; | |
2098 | int error; | |
2099 | ztest_zgd_private_t *zgd_private; | |
2100 | ||
2101 | ztest_object_lock(zd, object, RL_READER); | |
2102 | error = dmu_bonus_hold(os, object, FTAG, &db); | |
2103 | if (error) { | |
2104 | ztest_object_unlock(zd, object); | |
2105 | return (error); | |
2106 | } | |
2107 | ||
2108 | crtxg = ztest_bt_bonus(db)->bt_crtxg; | |
2109 | ||
2110 | if (crtxg == 0 || crtxg > txg) { | |
2111 | dmu_buf_rele(db, FTAG); | |
2112 | ztest_object_unlock(zd, object); | |
2113 | return (ENOENT); | |
2114 | } | |
2115 | ||
2116 | dmu_object_info_from_db(db, &doi); | |
2117 | dmu_buf_rele(db, FTAG); | |
2118 | db = NULL; | |
2119 | ||
2120 | zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL); | |
2121 | zgd->zgd_zilog = zd->zd_zilog; | |
2122 | zgd_private = umem_zalloc(sizeof (ztest_zgd_private_t), UMEM_NOFAIL); | |
2123 | zgd_private->z_zd = zd; | |
2124 | zgd_private->z_object = object; | |
2125 | zgd->zgd_private = zgd_private; | |
2126 | ||
2127 | if (buf != NULL) { /* immediate write */ | |
2128 | zgd_private->z_rl = ztest_range_lock(zd, object, offset, size, | |
2129 | RL_READER); | |
2130 | ||
2131 | error = dmu_read(os, object, offset, size, buf, | |
2132 | DMU_READ_NO_PREFETCH); | |
2133 | ASSERT(error == 0); | |
2134 | } else { | |
2135 | size = doi.doi_data_block_size; | |
2136 | if (ISP2(size)) { | |
2137 | offset = P2ALIGN(offset, size); | |
2138 | } else { | |
2139 | ASSERT(offset < size); | |
2140 | offset = 0; | |
2141 | } | |
2142 | ||
2143 | zgd_private->z_rl = ztest_range_lock(zd, object, offset, size, | |
2144 | RL_READER); | |
2145 | ||
2146 | error = dmu_buf_hold(os, object, offset, zgd, &db, | |
2147 | DMU_READ_NO_PREFETCH); | |
2148 | ||
2149 | if (error == 0) { | |
2150 | blkptr_t *obp = dmu_buf_get_blkptr(db); | |
2151 | if (obp) { | |
2152 | ASSERT(BP_IS_HOLE(bp)); | |
2153 | *bp = *obp; | |
2154 | } | |
2155 | ||
2156 | zgd->zgd_db = db; | |
2157 | zgd->zgd_bp = bp; | |
2158 | ||
2159 | ASSERT(db->db_offset == offset); | |
2160 | ASSERT(db->db_size == size); | |
2161 | ||
2162 | error = dmu_sync(zio, lr->lr_common.lrc_txg, | |
2163 | ztest_get_done, zgd); | |
2164 | ||
2165 | if (error == 0) | |
2166 | return (0); | |
2167 | } | |
2168 | } | |
2169 | ||
2170 | ztest_get_done(zgd, error); | |
2171 | ||
2172 | return (error); | |
2173 | } | |
2174 | ||
2175 | static void * | |
2176 | ztest_lr_alloc(size_t lrsize, char *name) | |
2177 | { | |
2178 | char *lr; | |
2179 | size_t namesize = name ? strlen(name) + 1 : 0; | |
2180 | ||
2181 | lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL); | |
2182 | ||
2183 | if (name) | |
2184 | bcopy(name, lr + lrsize, namesize); | |
2185 | ||
2186 | return (lr); | |
2187 | } | |
2188 | ||
2189 | void | |
2190 | ztest_lr_free(void *lr, size_t lrsize, char *name) | |
2191 | { | |
2192 | size_t namesize = name ? strlen(name) + 1 : 0; | |
2193 | ||
2194 | umem_free(lr, lrsize + namesize); | |
2195 | } | |
2196 | ||
2197 | /* | |
2198 | * Lookup a bunch of objects. Returns the number of objects not found. | |
2199 | */ | |
2200 | static int | |
2201 | ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count) | |
2202 | { | |
2203 | int missing = 0; | |
2204 | int error; | |
2205 | int i; | |
2206 | ||
2207 | ASSERT(mutex_held(&zd->zd_dirobj_lock)); | |
2208 | ||
2209 | for (i = 0; i < count; i++, od++) { | |
2210 | od->od_object = 0; | |
2211 | error = zap_lookup(zd->zd_os, od->od_dir, od->od_name, | |
2212 | sizeof (uint64_t), 1, &od->od_object); | |
2213 | if (error) { | |
2214 | ASSERT(error == ENOENT); | |
2215 | ASSERT(od->od_object == 0); | |
2216 | missing++; | |
2217 | } else { | |
2218 | dmu_buf_t *db; | |
2219 | ztest_block_tag_t *bbt; | |
2220 | dmu_object_info_t doi; | |
2221 | ||
2222 | ASSERT(od->od_object != 0); | |
2223 | ASSERT(missing == 0); /* there should be no gaps */ | |
2224 | ||
2225 | ztest_object_lock(zd, od->od_object, RL_READER); | |
2226 | VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os, | |
2227 | od->od_object, FTAG, &db)); | |
2228 | dmu_object_info_from_db(db, &doi); | |
2229 | bbt = ztest_bt_bonus(db); | |
2230 | ASSERT3U(bbt->bt_magic, ==, BT_MAGIC); | |
2231 | od->od_type = doi.doi_type; | |
2232 | od->od_blocksize = doi.doi_data_block_size; | |
2233 | od->od_gen = bbt->bt_gen; | |
2234 | dmu_buf_rele(db, FTAG); | |
2235 | ztest_object_unlock(zd, od->od_object); | |
2236 | } | |
2237 | } | |
2238 | ||
2239 | return (missing); | |
2240 | } | |
2241 | ||
2242 | static int | |
2243 | ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count) | |
2244 | { | |
2245 | int missing = 0; | |
2246 | int i; | |
2247 | ||
2248 | ASSERT(mutex_held(&zd->zd_dirobj_lock)); | |
2249 | ||
2250 | for (i = 0; i < count; i++, od++) { | |
2251 | if (missing) { | |
2252 | od->od_object = 0; | |
2253 | missing++; | |
2254 | continue; | |
2255 | } | |
2256 | ||
2257 | lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name); | |
2258 | ||
2259 | lr->lr_doid = od->od_dir; | |
2260 | lr->lr_foid = 0; /* 0 to allocate, > 0 to claim */ | |
2261 | lr->lrz_type = od->od_crtype; | |
2262 | lr->lrz_blocksize = od->od_crblocksize; | |
2263 | lr->lrz_ibshift = ztest_random_ibshift(); | |
2264 | lr->lrz_bonustype = DMU_OT_UINT64_OTHER; | |
2265 | lr->lrz_dnodesize = od->od_crdnodesize; | |
2266 | lr->lr_gen = od->od_crgen; | |
2267 | lr->lr_crtime[0] = time(NULL); | |
2268 | ||
2269 | if (ztest_replay_create(zd, lr, B_FALSE) != 0) { | |
2270 | ASSERT(missing == 0); | |
2271 | od->od_object = 0; | |
2272 | missing++; | |
2273 | } else { | |
2274 | od->od_object = lr->lr_foid; | |
2275 | od->od_type = od->od_crtype; | |
2276 | od->od_blocksize = od->od_crblocksize; | |
2277 | od->od_gen = od->od_crgen; | |
2278 | ASSERT(od->od_object != 0); | |
2279 | } | |
2280 | ||
2281 | ztest_lr_free(lr, sizeof (*lr), od->od_name); | |
2282 | } | |
2283 | ||
2284 | return (missing); | |
2285 | } | |
2286 | ||
2287 | static int | |
2288 | ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count) | |
2289 | { | |
2290 | int missing = 0; | |
2291 | int error; | |
2292 | int i; | |
2293 | ||
2294 | ASSERT(mutex_held(&zd->zd_dirobj_lock)); | |
2295 | ||
2296 | od += count - 1; | |
2297 | ||
2298 | for (i = count - 1; i >= 0; i--, od--) { | |
2299 | if (missing) { | |
2300 | missing++; | |
2301 | continue; | |
2302 | } | |
2303 | ||
2304 | /* | |
2305 | * No object was found. | |
2306 | */ | |
2307 | if (od->od_object == 0) | |
2308 | continue; | |
2309 | ||
2310 | lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name); | |
2311 | ||
2312 | lr->lr_doid = od->od_dir; | |
2313 | ||
2314 | if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) { | |
2315 | ASSERT3U(error, ==, ENOSPC); | |
2316 | missing++; | |
2317 | } else { | |
2318 | od->od_object = 0; | |
2319 | } | |
2320 | ztest_lr_free(lr, sizeof (*lr), od->od_name); | |
2321 | } | |
2322 | ||
2323 | return (missing); | |
2324 | } | |
2325 | ||
2326 | static int | |
2327 | ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size, | |
2328 | void *data) | |
2329 | { | |
2330 | lr_write_t *lr; | |
2331 | int error; | |
2332 | ||
2333 | lr = ztest_lr_alloc(sizeof (*lr) + size, NULL); | |
2334 | ||
2335 | lr->lr_foid = object; | |
2336 | lr->lr_offset = offset; | |
2337 | lr->lr_length = size; | |
2338 | lr->lr_blkoff = 0; | |
2339 | BP_ZERO(&lr->lr_blkptr); | |
2340 | ||
2341 | bcopy(data, lr + 1, size); | |
2342 | ||
2343 | error = ztest_replay_write(zd, lr, B_FALSE); | |
2344 | ||
2345 | ztest_lr_free(lr, sizeof (*lr) + size, NULL); | |
2346 | ||
2347 | return (error); | |
2348 | } | |
2349 | ||
2350 | static int | |
2351 | ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size) | |
2352 | { | |
2353 | lr_truncate_t *lr; | |
2354 | int error; | |
2355 | ||
2356 | lr = ztest_lr_alloc(sizeof (*lr), NULL); | |
2357 | ||
2358 | lr->lr_foid = object; | |
2359 | lr->lr_offset = offset; | |
2360 | lr->lr_length = size; | |
2361 | ||
2362 | error = ztest_replay_truncate(zd, lr, B_FALSE); | |
2363 | ||
2364 | ztest_lr_free(lr, sizeof (*lr), NULL); | |
2365 | ||
2366 | return (error); | |
2367 | } | |
2368 | ||
2369 | static int | |
2370 | ztest_setattr(ztest_ds_t *zd, uint64_t object) | |
2371 | { | |
2372 | lr_setattr_t *lr; | |
2373 | int error; | |
2374 | ||
2375 | lr = ztest_lr_alloc(sizeof (*lr), NULL); | |
2376 | ||
2377 | lr->lr_foid = object; | |
2378 | lr->lr_size = 0; | |
2379 | lr->lr_mode = 0; | |
2380 | ||
2381 | error = ztest_replay_setattr(zd, lr, B_FALSE); | |
2382 | ||
2383 | ztest_lr_free(lr, sizeof (*lr), NULL); | |
2384 | ||
2385 | return (error); | |
2386 | } | |
2387 | ||
2388 | static void | |
2389 | ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size) | |
2390 | { | |
2391 | objset_t *os = zd->zd_os; | |
2392 | dmu_tx_t *tx; | |
2393 | uint64_t txg; | |
2394 | ztest_zrl_t *rl; | |
2395 | ||
2396 | txg_wait_synced(dmu_objset_pool(os), 0); | |
2397 | ||
2398 | ztest_object_lock(zd, object, RL_READER); | |
2399 | rl = ztest_range_lock(zd, object, offset, size, RL_WRITER); | |
2400 | ||
2401 | tx = dmu_tx_create(os); | |
2402 | ||
2403 | dmu_tx_hold_write(tx, object, offset, size); | |
2404 | ||
2405 | txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); | |
2406 | ||
2407 | if (txg != 0) { | |
2408 | dmu_prealloc(os, object, offset, size, tx); | |
2409 | dmu_tx_commit(tx); | |
2410 | txg_wait_synced(dmu_objset_pool(os), txg); | |
2411 | } else { | |
2412 | (void) dmu_free_long_range(os, object, offset, size); | |
2413 | } | |
2414 | ||
2415 | ztest_range_unlock(zd, rl); | |
2416 | ztest_object_unlock(zd, object); | |
2417 | } | |
2418 | ||
2419 | static void | |
2420 | ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset) | |
2421 | { | |
2422 | int err; | |
2423 | ztest_block_tag_t wbt; | |
2424 | dmu_object_info_t doi; | |
2425 | enum ztest_io_type io_type; | |
2426 | uint64_t blocksize; | |
2427 | void *data; | |
2428 | ||
2429 | VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0); | |
2430 | blocksize = doi.doi_data_block_size; | |
2431 | data = umem_alloc(blocksize, UMEM_NOFAIL); | |
2432 | ||
2433 | /* | |
2434 | * Pick an i/o type at random, biased toward writing block tags. | |
2435 | */ | |
2436 | io_type = ztest_random(ZTEST_IO_TYPES); | |
2437 | if (ztest_random(2) == 0) | |
2438 | io_type = ZTEST_IO_WRITE_TAG; | |
2439 | ||
2440 | (void) rw_rdlock(&zd->zd_zilog_lock); | |
2441 | ||
2442 | switch (io_type) { | |
2443 | ||
2444 | case ZTEST_IO_WRITE_TAG: | |
2445 | ztest_bt_generate(&wbt, zd->zd_os, object, doi.doi_dnodesize, | |
2446 | offset, 0, 0, 0); | |
2447 | (void) ztest_write(zd, object, offset, sizeof (wbt), &wbt); | |
2448 | break; | |
2449 | ||
2450 | case ZTEST_IO_WRITE_PATTERN: | |
2451 | (void) memset(data, 'a' + (object + offset) % 5, blocksize); | |
2452 | if (ztest_random(2) == 0) { | |
2453 | /* | |
2454 | * Induce fletcher2 collisions to ensure that | |
2455 | * zio_ddt_collision() detects and resolves them | |
2456 | * when using fletcher2-verify for deduplication. | |
2457 | */ | |
2458 | ((uint64_t *)data)[0] ^= 1ULL << 63; | |
2459 | ((uint64_t *)data)[4] ^= 1ULL << 63; | |
2460 | } | |
2461 | (void) ztest_write(zd, object, offset, blocksize, data); | |
2462 | break; | |
2463 | ||
2464 | case ZTEST_IO_WRITE_ZEROES: | |
2465 | bzero(data, blocksize); | |
2466 | (void) ztest_write(zd, object, offset, blocksize, data); | |
2467 | break; | |
2468 | ||
2469 | case ZTEST_IO_TRUNCATE: | |
2470 | (void) ztest_truncate(zd, object, offset, blocksize); | |
2471 | break; | |
2472 | ||
2473 | case ZTEST_IO_SETATTR: | |
2474 | (void) ztest_setattr(zd, object); | |
2475 | break; | |
2476 | default: | |
2477 | break; | |
2478 | ||
2479 | case ZTEST_IO_REWRITE: | |
2480 | (void) rw_rdlock(&ztest_name_lock); | |
2481 | err = ztest_dsl_prop_set_uint64(zd->zd_name, | |
2482 | ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa), | |
2483 | B_FALSE); | |
2484 | VERIFY(err == 0 || err == ENOSPC); | |
2485 | err = ztest_dsl_prop_set_uint64(zd->zd_name, | |
2486 | ZFS_PROP_COMPRESSION, | |
2487 | ztest_random_dsl_prop(ZFS_PROP_COMPRESSION), | |
2488 | B_FALSE); | |
2489 | VERIFY(err == 0 || err == ENOSPC); | |
2490 | (void) rw_unlock(&ztest_name_lock); | |
2491 | ||
2492 | VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data, | |
2493 | DMU_READ_NO_PREFETCH)); | |
2494 | ||
2495 | (void) ztest_write(zd, object, offset, blocksize, data); | |
2496 | break; | |
2497 | } | |
2498 | ||
2499 | (void) rw_unlock(&zd->zd_zilog_lock); | |
2500 | ||
2501 | umem_free(data, blocksize); | |
2502 | } | |
2503 | ||
2504 | /* | |
2505 | * Initialize an object description template. | |
2506 | */ | |
2507 | static void | |
2508 | ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index, | |
2509 | dmu_object_type_t type, uint64_t blocksize, uint64_t dnodesize, | |
2510 | uint64_t gen) | |
2511 | { | |
2512 | od->od_dir = ZTEST_DIROBJ; | |
2513 | od->od_object = 0; | |
2514 | ||
2515 | od->od_crtype = type; | |
2516 | od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize(); | |
2517 | od->od_crdnodesize = dnodesize ? dnodesize : ztest_random_dnodesize(); | |
2518 | od->od_crgen = gen; | |
2519 | ||
2520 | od->od_type = DMU_OT_NONE; | |
2521 | od->od_blocksize = 0; | |
2522 | od->od_gen = 0; | |
2523 | ||
2524 | (void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]", | |
2525 | tag, (longlong_t)id, (u_longlong_t)index); | |
2526 | } | |
2527 | ||
2528 | /* | |
2529 | * Lookup or create the objects for a test using the od template. | |
2530 | * If the objects do not all exist, or if 'remove' is specified, | |
2531 | * remove any existing objects and create new ones. Otherwise, | |
2532 | * use the existing objects. | |
2533 | */ | |
2534 | static int | |
2535 | ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove) | |
2536 | { | |
2537 | int count = size / sizeof (*od); | |
2538 | int rv = 0; | |
2539 | ||
2540 | mutex_enter(&zd->zd_dirobj_lock); | |
2541 | if ((ztest_lookup(zd, od, count) != 0 || remove) && | |
2542 | (ztest_remove(zd, od, count) != 0 || | |
2543 | ztest_create(zd, od, count) != 0)) | |
2544 | rv = -1; | |
2545 | zd->zd_od = od; | |
2546 | mutex_exit(&zd->zd_dirobj_lock); | |
2547 | ||
2548 | return (rv); | |
2549 | } | |
2550 | ||
2551 | /* ARGSUSED */ | |
2552 | void | |
2553 | ztest_zil_commit(ztest_ds_t *zd, uint64_t id) | |
2554 | { | |
2555 | zilog_t *zilog = zd->zd_zilog; | |
2556 | ||
2557 | (void) rw_rdlock(&zd->zd_zilog_lock); | |
2558 | ||
2559 | zil_commit(zilog, ztest_random(ZTEST_OBJECTS)); | |
2560 | ||
2561 | /* | |
2562 | * Remember the committed values in zd, which is in parent/child | |
2563 | * shared memory. If we die, the next iteration of ztest_run() | |
2564 | * will verify that the log really does contain this record. | |
2565 | */ | |
2566 | mutex_enter(&zilog->zl_lock); | |
2567 | ASSERT(zd->zd_shared != NULL); | |
2568 | ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq); | |
2569 | zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq; | |
2570 | mutex_exit(&zilog->zl_lock); | |
2571 | ||
2572 | (void) rw_unlock(&zd->zd_zilog_lock); | |
2573 | } | |
2574 | ||
2575 | /* | |
2576 | * This function is designed to simulate the operations that occur during a | |
2577 | * mount/unmount operation. We hold the dataset across these operations in an | |
2578 | * attempt to expose any implicit assumptions about ZIL management. | |
2579 | */ | |
2580 | /* ARGSUSED */ | |
2581 | void | |
2582 | ztest_zil_remount(ztest_ds_t *zd, uint64_t id) | |
2583 | { | |
2584 | objset_t *os = zd->zd_os; | |
2585 | ||
2586 | /* | |
2587 | * We grab the zd_dirobj_lock to ensure that no other thread is | |
2588 | * updating the zil (i.e. adding in-memory log records) and the | |
2589 | * zd_zilog_lock to block any I/O. | |
2590 | */ | |
2591 | mutex_enter(&zd->zd_dirobj_lock); | |
2592 | (void) rw_wrlock(&zd->zd_zilog_lock); | |
2593 | ||
2594 | /* zfs_sb_teardown() */ | |
2595 | zil_close(zd->zd_zilog); | |
2596 | ||
2597 | /* zfsvfs_setup() */ | |
2598 | VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog); | |
2599 | zil_replay(os, zd, ztest_replay_vector); | |
2600 | ||
2601 | (void) rw_unlock(&zd->zd_zilog_lock); | |
2602 | mutex_exit(&zd->zd_dirobj_lock); | |
2603 | } | |
2604 | ||
2605 | /* | |
2606 | * Verify that we can't destroy an active pool, create an existing pool, | |
2607 | * or create a pool with a bad vdev spec. | |
2608 | */ | |
2609 | /* ARGSUSED */ | |
2610 | void | |
2611 | ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id) | |
2612 | { | |
2613 | ztest_shared_opts_t *zo = &ztest_opts; | |
2614 | spa_t *spa; | |
2615 | nvlist_t *nvroot; | |
2616 | ||
2617 | /* | |
2618 | * Attempt to create using a bad file. | |
2619 | */ | |
2620 | nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1); | |
2621 | VERIFY3U(ENOENT, ==, | |
2622 | spa_create("ztest_bad_file", nvroot, NULL, NULL)); | |
2623 | nvlist_free(nvroot); | |
2624 | ||
2625 | /* | |
2626 | * Attempt to create using a bad mirror. | |
2627 | */ | |
2628 | nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1); | |
2629 | VERIFY3U(ENOENT, ==, | |
2630 | spa_create("ztest_bad_mirror", nvroot, NULL, NULL)); | |
2631 | nvlist_free(nvroot); | |
2632 | ||
2633 | /* | |
2634 | * Attempt to create an existing pool. It shouldn't matter | |
2635 | * what's in the nvroot; we should fail with EEXIST. | |
2636 | */ | |
2637 | (void) rw_rdlock(&ztest_name_lock); | |
2638 | nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1); | |
2639 | VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL)); | |
2640 | nvlist_free(nvroot); | |
2641 | VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG)); | |
2642 | VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool)); | |
2643 | spa_close(spa, FTAG); | |
2644 | ||
2645 | (void) rw_unlock(&ztest_name_lock); | |
2646 | } | |
2647 | ||
2648 | /* ARGSUSED */ | |
2649 | void | |
2650 | ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id) | |
2651 | { | |
2652 | spa_t *spa; | |
2653 | uint64_t initial_version = SPA_VERSION_INITIAL; | |
2654 | uint64_t version, newversion; | |
2655 | nvlist_t *nvroot, *props; | |
2656 | char *name; | |
2657 | ||
2658 | mutex_enter(&ztest_vdev_lock); | |
2659 | name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool); | |
2660 | ||
2661 | /* | |
2662 | * Clean up from previous runs. | |
2663 | */ | |
2664 | (void) spa_destroy(name); | |
2665 | ||
2666 | nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0, | |
2667 | 0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1); | |
2668 | ||
2669 | /* | |
2670 | * If we're configuring a RAIDZ device then make sure that the | |
2671 | * the initial version is capable of supporting that feature. | |
2672 | */ | |
2673 | switch (ztest_opts.zo_raidz_parity) { | |
2674 | case 0: | |
2675 | case 1: | |
2676 | initial_version = SPA_VERSION_INITIAL; | |
2677 | break; | |
2678 | case 2: | |
2679 | initial_version = SPA_VERSION_RAIDZ2; | |
2680 | break; | |
2681 | case 3: | |
2682 | initial_version = SPA_VERSION_RAIDZ3; | |
2683 | break; | |
2684 | } | |
2685 | ||
2686 | /* | |
2687 | * Create a pool with a spa version that can be upgraded. Pick | |
2688 | * a value between initial_version and SPA_VERSION_BEFORE_FEATURES. | |
2689 | */ | |
2690 | do { | |
2691 | version = ztest_random_spa_version(initial_version); | |
2692 | } while (version > SPA_VERSION_BEFORE_FEATURES); | |
2693 | ||
2694 | props = fnvlist_alloc(); | |
2695 | fnvlist_add_uint64(props, | |
2696 | zpool_prop_to_name(ZPOOL_PROP_VERSION), version); | |
2697 | VERIFY3S(spa_create(name, nvroot, props, NULL), ==, 0); | |
2698 | fnvlist_free(nvroot); | |
2699 | fnvlist_free(props); | |
2700 | ||
2701 | VERIFY3S(spa_open(name, &spa, FTAG), ==, 0); | |
2702 | VERIFY3U(spa_version(spa), ==, version); | |
2703 | newversion = ztest_random_spa_version(version + 1); | |
2704 | ||
2705 | if (ztest_opts.zo_verbose >= 4) { | |
2706 | (void) printf("upgrading spa version from %llu to %llu\n", | |
2707 | (u_longlong_t)version, (u_longlong_t)newversion); | |
2708 | } | |
2709 | ||
2710 | spa_upgrade(spa, newversion); | |
2711 | VERIFY3U(spa_version(spa), >, version); | |
2712 | VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config, | |
2713 | zpool_prop_to_name(ZPOOL_PROP_VERSION))); | |
2714 | spa_close(spa, FTAG); | |
2715 | ||
2716 | strfree(name); | |
2717 | mutex_exit(&ztest_vdev_lock); | |
2718 | } | |
2719 | ||
2720 | static vdev_t * | |
2721 | vdev_lookup_by_path(vdev_t *vd, const char *path) | |
2722 | { | |
2723 | vdev_t *mvd; | |
2724 | int c; | |
2725 | ||
2726 | if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0) | |
2727 | return (vd); | |
2728 | ||
2729 | for (c = 0; c < vd->vdev_children; c++) | |
2730 | if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) != | |
2731 | NULL) | |
2732 | return (mvd); | |
2733 | ||
2734 | return (NULL); | |
2735 | } | |
2736 | ||
2737 | /* | |
2738 | * Find the first available hole which can be used as a top-level. | |
2739 | */ | |
2740 | int | |
2741 | find_vdev_hole(spa_t *spa) | |
2742 | { | |
2743 | vdev_t *rvd = spa->spa_root_vdev; | |
2744 | int c; | |
2745 | ||
2746 | ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV); | |
2747 | ||
2748 | for (c = 0; c < rvd->vdev_children; c++) { | |
2749 | vdev_t *cvd = rvd->vdev_child[c]; | |
2750 | ||
2751 | if (cvd->vdev_ishole) | |
2752 | break; | |
2753 | } | |
2754 | return (c); | |
2755 | } | |
2756 | ||
2757 | /* | |
2758 | * Verify that vdev_add() works as expected. | |
2759 | */ | |
2760 | /* ARGSUSED */ | |
2761 | void | |
2762 | ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id) | |
2763 | { | |
2764 | ztest_shared_t *zs = ztest_shared; | |
2765 | spa_t *spa = ztest_spa; | |
2766 | uint64_t leaves; | |
2767 | uint64_t guid; | |
2768 | nvlist_t *nvroot; | |
2769 | int error; | |
2770 | ||
2771 | mutex_enter(&ztest_vdev_lock); | |
2772 | leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz; | |
2773 | ||
2774 | spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); | |
2775 | ||
2776 | ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves; | |
2777 | ||
2778 | /* | |
2779 | * If we have slogs then remove them 1/4 of the time. | |
2780 | */ | |
2781 | if (spa_has_slogs(spa) && ztest_random(4) == 0) { | |
2782 | /* | |
2783 | * Grab the guid from the head of the log class rotor. | |
2784 | */ | |
2785 | guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid; | |
2786 | ||
2787 | spa_config_exit(spa, SCL_VDEV, FTAG); | |
2788 | ||
2789 | /* | |
2790 | * We have to grab the zs_name_lock as writer to | |
2791 | * prevent a race between removing a slog (dmu_objset_find) | |
2792 | * and destroying a dataset. Removing the slog will | |
2793 | * grab a reference on the dataset which may cause | |
2794 | * dsl_destroy_head() to fail with EBUSY thus | |
2795 | * leaving the dataset in an inconsistent state. | |
2796 | */ | |
2797 | rw_wrlock(&ztest_name_lock); | |
2798 | error = spa_vdev_remove(spa, guid, B_FALSE); | |
2799 | rw_unlock(&ztest_name_lock); | |
2800 | ||
2801 | if (error && error != EEXIST) | |
2802 | fatal(0, "spa_vdev_remove() = %d", error); | |
2803 | } else { | |
2804 | spa_config_exit(spa, SCL_VDEV, FTAG); | |
2805 | ||
2806 | /* | |
2807 | * Make 1/4 of the devices be log devices. | |
2808 | */ | |
2809 | nvroot = make_vdev_root(NULL, NULL, NULL, | |
2810 | ztest_opts.zo_vdev_size, 0, | |
2811 | ztest_random(4) == 0, ztest_opts.zo_raidz, | |
2812 | zs->zs_mirrors, 1); | |
2813 | ||
2814 | error = spa_vdev_add(spa, nvroot); | |
2815 | nvlist_free(nvroot); | |
2816 | ||
2817 | if (error == ENOSPC) | |
2818 | ztest_record_enospc("spa_vdev_add"); | |
2819 | else if (error != 0) | |
2820 | fatal(0, "spa_vdev_add() = %d", error); | |
2821 | } | |
2822 | ||
2823 | mutex_exit(&ztest_vdev_lock); | |
2824 | } | |
2825 | ||
2826 | /* | |
2827 | * Verify that adding/removing aux devices (l2arc, hot spare) works as expected. | |
2828 | */ | |
2829 | /* ARGSUSED */ | |
2830 | void | |
2831 | ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id) | |
2832 | { | |
2833 | ztest_shared_t *zs = ztest_shared; | |
2834 | spa_t *spa = ztest_spa; | |
2835 | vdev_t *rvd = spa->spa_root_vdev; | |
2836 | spa_aux_vdev_t *sav; | |
2837 | char *aux; | |
2838 | char *path; | |
2839 | uint64_t guid = 0; | |
2840 | int error; | |
2841 | ||
2842 | path = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); | |
2843 | ||
2844 | if (ztest_random(2) == 0) { | |
2845 | sav = &spa->spa_spares; | |
2846 | aux = ZPOOL_CONFIG_SPARES; | |
2847 | } else { | |
2848 | sav = &spa->spa_l2cache; | |
2849 | aux = ZPOOL_CONFIG_L2CACHE; | |
2850 | } | |
2851 | ||
2852 | mutex_enter(&ztest_vdev_lock); | |
2853 | ||
2854 | spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); | |
2855 | ||
2856 | if (sav->sav_count != 0 && ztest_random(4) == 0) { | |
2857 | /* | |
2858 | * Pick a random device to remove. | |
2859 | */ | |
2860 | guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid; | |
2861 | } else { | |
2862 | /* | |
2863 | * Find an unused device we can add. | |
2864 | */ | |
2865 | zs->zs_vdev_aux = 0; | |
2866 | for (;;) { | |
2867 | int c; | |
2868 | (void) snprintf(path, MAXPATHLEN, ztest_aux_template, | |
2869 | ztest_opts.zo_dir, ztest_opts.zo_pool, aux, | |
2870 | zs->zs_vdev_aux); | |
2871 | for (c = 0; c < sav->sav_count; c++) | |
2872 | if (strcmp(sav->sav_vdevs[c]->vdev_path, | |
2873 | path) == 0) | |
2874 | break; | |
2875 | if (c == sav->sav_count && | |
2876 | vdev_lookup_by_path(rvd, path) == NULL) | |
2877 | break; | |
2878 | zs->zs_vdev_aux++; | |
2879 | } | |
2880 | } | |
2881 | ||
2882 | spa_config_exit(spa, SCL_VDEV, FTAG); | |
2883 | ||
2884 | if (guid == 0) { | |
2885 | /* | |
2886 | * Add a new device. | |
2887 | */ | |
2888 | nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL, | |
2889 | (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1); | |
2890 | error = spa_vdev_add(spa, nvroot); | |
2891 | if (error != 0) | |
2892 | fatal(0, "spa_vdev_add(%p) = %d", nvroot, error); | |
2893 | nvlist_free(nvroot); | |
2894 | } else { | |
2895 | /* | |
2896 | * Remove an existing device. Sometimes, dirty its | |
2897 | * vdev state first to make sure we handle removal | |
2898 | * of devices that have pending state changes. | |
2899 | */ | |
2900 | if (ztest_random(2) == 0) | |
2901 | (void) vdev_online(spa, guid, 0, NULL); | |
2902 | ||
2903 | error = spa_vdev_remove(spa, guid, B_FALSE); | |
2904 | if (error != 0 && error != EBUSY) | |
2905 | fatal(0, "spa_vdev_remove(%llu) = %d", guid, error); | |
2906 | } | |
2907 | ||
2908 | mutex_exit(&ztest_vdev_lock); | |
2909 | ||
2910 | umem_free(path, MAXPATHLEN); | |
2911 | } | |
2912 | ||
2913 | /* | |
2914 | * split a pool if it has mirror tlvdevs | |
2915 | */ | |
2916 | /* ARGSUSED */ | |
2917 | void | |
2918 | ztest_split_pool(ztest_ds_t *zd, uint64_t id) | |
2919 | { | |
2920 | ztest_shared_t *zs = ztest_shared; | |
2921 | spa_t *spa = ztest_spa; | |
2922 | vdev_t *rvd = spa->spa_root_vdev; | |
2923 | nvlist_t *tree, **child, *config, *split, **schild; | |
2924 | uint_t c, children, schildren = 0, lastlogid = 0; | |
2925 | int error = 0; | |
2926 | ||
2927 | mutex_enter(&ztest_vdev_lock); | |
2928 | ||
2929 | /* ensure we have a useable config; mirrors of raidz aren't supported */ | |
2930 | if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) { | |
2931 | mutex_exit(&ztest_vdev_lock); | |
2932 | return; | |
2933 | } | |
2934 | ||
2935 | /* clean up the old pool, if any */ | |
2936 | (void) spa_destroy("splitp"); | |
2937 | ||
2938 | spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); | |
2939 | ||
2940 | /* generate a config from the existing config */ | |
2941 | mutex_enter(&spa->spa_props_lock); | |
2942 | VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE, | |
2943 | &tree) == 0); | |
2944 | mutex_exit(&spa->spa_props_lock); | |
2945 | ||
2946 | VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child, | |
2947 | &children) == 0); | |
2948 | ||
2949 | schild = malloc(rvd->vdev_children * sizeof (nvlist_t *)); | |
2950 | for (c = 0; c < children; c++) { | |
2951 | vdev_t *tvd = rvd->vdev_child[c]; | |
2952 | nvlist_t **mchild; | |
2953 | uint_t mchildren; | |
2954 | ||
2955 | if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) { | |
2956 | VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME, | |
2957 | 0) == 0); | |
2958 | VERIFY(nvlist_add_string(schild[schildren], | |
2959 | ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0); | |
2960 | VERIFY(nvlist_add_uint64(schild[schildren], | |
2961 | ZPOOL_CONFIG_IS_HOLE, 1) == 0); | |
2962 | if (lastlogid == 0) | |
2963 | lastlogid = schildren; | |
2964 | ++schildren; | |
2965 | continue; | |
2966 | } | |
2967 | lastlogid = 0; | |
2968 | VERIFY(nvlist_lookup_nvlist_array(child[c], | |
2969 | ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0); | |
2970 | VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0); | |
2971 | } | |
2972 | ||
2973 | /* OK, create a config that can be used to split */ | |
2974 | VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0); | |
2975 | VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE, | |
2976 | VDEV_TYPE_ROOT) == 0); | |
2977 | VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild, | |
2978 | lastlogid != 0 ? lastlogid : schildren) == 0); | |
2979 | ||
2980 | VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0); | |
2981 | VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0); | |
2982 | ||
2983 | for (c = 0; c < schildren; c++) | |
2984 | nvlist_free(schild[c]); | |
2985 | free(schild); | |
2986 | nvlist_free(split); | |
2987 | ||
2988 | spa_config_exit(spa, SCL_VDEV, FTAG); | |
2989 | ||
2990 | (void) rw_wrlock(&ztest_name_lock); | |
2991 | error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE); | |
2992 | (void) rw_unlock(&ztest_name_lock); | |
2993 | ||
2994 | nvlist_free(config); | |
2995 | ||
2996 | if (error == 0) { | |
2997 | (void) printf("successful split - results:\n"); | |
2998 | mutex_enter(&spa_namespace_lock); | |
2999 | show_pool_stats(spa); | |
3000 | show_pool_stats(spa_lookup("splitp")); | |
3001 | mutex_exit(&spa_namespace_lock); | |
3002 | ++zs->zs_splits; | |
3003 | --zs->zs_mirrors; | |
3004 | } | |
3005 | mutex_exit(&ztest_vdev_lock); | |
3006 | ||
3007 | } | |
3008 | ||
3009 | /* | |
3010 | * Verify that we can attach and detach devices. | |
3011 | */ | |
3012 | /* ARGSUSED */ | |
3013 | void | |
3014 | ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id) | |
3015 | { | |
3016 | ztest_shared_t *zs = ztest_shared; | |
3017 | spa_t *spa = ztest_spa; | |
3018 | spa_aux_vdev_t *sav = &spa->spa_spares; | |
3019 | vdev_t *rvd = spa->spa_root_vdev; | |
3020 | vdev_t *oldvd, *newvd, *pvd; | |
3021 | nvlist_t *root; | |
3022 | uint64_t leaves; | |
3023 | uint64_t leaf, top; | |
3024 | uint64_t ashift = ztest_get_ashift(); | |
3025 | uint64_t oldguid, pguid; | |
3026 | uint64_t oldsize, newsize; | |
3027 | char *oldpath, *newpath; | |
3028 | int replacing; | |
3029 | int oldvd_has_siblings = B_FALSE; | |
3030 | int newvd_is_spare = B_FALSE; | |
3031 | int oldvd_is_log; | |
3032 | int error, expected_error; | |
3033 | ||
3034 | oldpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); | |
3035 | newpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); | |
3036 | ||
3037 | mutex_enter(&ztest_vdev_lock); | |
3038 | leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz; | |
3039 | ||
3040 | spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); | |
3041 | ||
3042 | /* | |
3043 | * Decide whether to do an attach or a replace. | |
3044 | */ | |
3045 | replacing = ztest_random(2); | |
3046 | ||
3047 | /* | |
3048 | * Pick a random top-level vdev. | |
3049 | */ | |
3050 | top = ztest_random_vdev_top(spa, B_TRUE); | |
3051 | ||
3052 | /* | |
3053 | * Pick a random leaf within it. | |
3054 | */ | |
3055 | leaf = ztest_random(leaves); | |
3056 | ||
3057 | /* | |
3058 | * Locate this vdev. | |
3059 | */ | |
3060 | oldvd = rvd->vdev_child[top]; | |
3061 | if (zs->zs_mirrors >= 1) { | |
3062 | ASSERT(oldvd->vdev_ops == &vdev_mirror_ops); | |
3063 | ASSERT(oldvd->vdev_children >= zs->zs_mirrors); | |
3064 | oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz]; | |
3065 | } | |
3066 | if (ztest_opts.zo_raidz > 1) { | |
3067 | ASSERT(oldvd->vdev_ops == &vdev_raidz_ops); | |
3068 | ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz); | |
3069 | oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz]; | |
3070 | } | |
3071 | ||
3072 | /* | |
3073 | * If we're already doing an attach or replace, oldvd may be a | |
3074 | * mirror vdev -- in which case, pick a random child. | |
3075 | */ | |
3076 | while (oldvd->vdev_children != 0) { | |
3077 | oldvd_has_siblings = B_TRUE; | |
3078 | ASSERT(oldvd->vdev_children >= 2); | |
3079 | oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)]; | |
3080 | } | |
3081 | ||
3082 | oldguid = oldvd->vdev_guid; | |
3083 | oldsize = vdev_get_min_asize(oldvd); | |
3084 | oldvd_is_log = oldvd->vdev_top->vdev_islog; | |
3085 | (void) strcpy(oldpath, oldvd->vdev_path); | |
3086 | pvd = oldvd->vdev_parent; | |
3087 | pguid = pvd->vdev_guid; | |
3088 | ||
3089 | /* | |
3090 | * If oldvd has siblings, then half of the time, detach it. | |
3091 | */ | |
3092 | if (oldvd_has_siblings && ztest_random(2) == 0) { | |
3093 | spa_config_exit(spa, SCL_VDEV, FTAG); | |
3094 | error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE); | |
3095 | if (error != 0 && error != ENODEV && error != EBUSY && | |
3096 | error != ENOTSUP) | |
3097 | fatal(0, "detach (%s) returned %d", oldpath, error); | |
3098 | goto out; | |
3099 | } | |
3100 | ||
3101 | /* | |
3102 | * For the new vdev, choose with equal probability between the two | |
3103 | * standard paths (ending in either 'a' or 'b') or a random hot spare. | |
3104 | */ | |
3105 | if (sav->sav_count != 0 && ztest_random(3) == 0) { | |
3106 | newvd = sav->sav_vdevs[ztest_random(sav->sav_count)]; | |
3107 | newvd_is_spare = B_TRUE; | |
3108 | (void) strcpy(newpath, newvd->vdev_path); | |
3109 | } else { | |
3110 | (void) snprintf(newpath, MAXPATHLEN, ztest_dev_template, | |
3111 | ztest_opts.zo_dir, ztest_opts.zo_pool, | |
3112 | top * leaves + leaf); | |
3113 | if (ztest_random(2) == 0) | |
3114 | newpath[strlen(newpath) - 1] = 'b'; | |
3115 | newvd = vdev_lookup_by_path(rvd, newpath); | |
3116 | } | |
3117 | ||
3118 | if (newvd) { | |
3119 | newsize = vdev_get_min_asize(newvd); | |
3120 | } else { | |
3121 | /* | |
3122 | * Make newsize a little bigger or smaller than oldsize. | |
3123 | * If it's smaller, the attach should fail. | |
3124 | * If it's larger, and we're doing a replace, | |
3125 | * we should get dynamic LUN growth when we're done. | |
3126 | */ | |
3127 | newsize = 10 * oldsize / (9 + ztest_random(3)); | |
3128 | } | |
3129 | ||
3130 | /* | |
3131 | * If pvd is not a mirror or root, the attach should fail with ENOTSUP, | |
3132 | * unless it's a replace; in that case any non-replacing parent is OK. | |
3133 | * | |
3134 | * If newvd is already part of the pool, it should fail with EBUSY. | |
3135 | * | |
3136 | * If newvd is too small, it should fail with EOVERFLOW. | |
3137 | */ | |
3138 | if (pvd->vdev_ops != &vdev_mirror_ops && | |
3139 | pvd->vdev_ops != &vdev_root_ops && (!replacing || | |
3140 | pvd->vdev_ops == &vdev_replacing_ops || | |
3141 | pvd->vdev_ops == &vdev_spare_ops)) | |
3142 | expected_error = ENOTSUP; | |
3143 | else if (newvd_is_spare && (!replacing || oldvd_is_log)) | |
3144 | expected_error = ENOTSUP; | |
3145 | else if (newvd == oldvd) | |
3146 | expected_error = replacing ? 0 : EBUSY; | |
3147 | else if (vdev_lookup_by_path(rvd, newpath) != NULL) | |
3148 | expected_error = EBUSY; | |
3149 | else if (newsize < oldsize) | |
3150 | expected_error = EOVERFLOW; | |
3151 | else if (ashift > oldvd->vdev_top->vdev_ashift) | |
3152 | expected_error = EDOM; | |
3153 | else | |
3154 | expected_error = 0; | |
3155 | ||
3156 | spa_config_exit(spa, SCL_VDEV, FTAG); | |
3157 | ||
3158 | /* | |
3159 | * Build the nvlist describing newpath. | |
3160 | */ | |
3161 | root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0, | |
3162 | ashift, 0, 0, 0, 1); | |
3163 | ||
3164 | error = spa_vdev_attach(spa, oldguid, root, replacing); | |
3165 | ||
3166 | nvlist_free(root); | |
3167 | ||
3168 | /* | |
3169 | * If our parent was the replacing vdev, but the replace completed, | |
3170 | * then instead of failing with ENOTSUP we may either succeed, | |
3171 | * fail with ENODEV, or fail with EOVERFLOW. | |
3172 | */ | |
3173 | if (expected_error == ENOTSUP && | |
3174 | (error == 0 || error == ENODEV || error == EOVERFLOW)) | |
3175 | expected_error = error; | |
3176 | ||
3177 | /* | |
3178 | * If someone grew the LUN, the replacement may be too small. | |
3179 | */ | |
3180 | if (error == EOVERFLOW || error == EBUSY) | |
3181 | expected_error = error; | |
3182 | ||
3183 | /* XXX workaround 6690467 */ | |
3184 | if (error != expected_error && expected_error != EBUSY) { | |
3185 | fatal(0, "attach (%s %llu, %s %llu, %d) " | |
3186 | "returned %d, expected %d", | |
3187 | oldpath, oldsize, newpath, | |
3188 | newsize, replacing, error, expected_error); | |
3189 | } | |
3190 | out: | |
3191 | mutex_exit(&ztest_vdev_lock); | |
3192 | ||
3193 | umem_free(oldpath, MAXPATHLEN); | |
3194 | umem_free(newpath, MAXPATHLEN); | |
3195 | } | |
3196 | ||
3197 | /* | |
3198 | * Callback function which expands the physical size of the vdev. | |
3199 | */ | |
3200 | vdev_t * | |
3201 | grow_vdev(vdev_t *vd, void *arg) | |
3202 | { | |
3203 | ASSERTV(spa_t *spa = vd->vdev_spa); | |
3204 | size_t *newsize = arg; | |
3205 | size_t fsize; | |
3206 | int fd; | |
3207 | ||
3208 | ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE); | |
3209 | ASSERT(vd->vdev_ops->vdev_op_leaf); | |
3210 | ||
3211 | if ((fd = open(vd->vdev_path, O_RDWR)) == -1) | |
3212 | return (vd); | |
3213 | ||
3214 | fsize = lseek(fd, 0, SEEK_END); | |
3215 | VERIFY(ftruncate(fd, *newsize) == 0); | |
3216 | ||
3217 | if (ztest_opts.zo_verbose >= 6) { | |
3218 | (void) printf("%s grew from %lu to %lu bytes\n", | |
3219 | vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize); | |
3220 | } | |
3221 | (void) close(fd); | |
3222 | return (NULL); | |
3223 | } | |
3224 | ||
3225 | /* | |
3226 | * Callback function which expands a given vdev by calling vdev_online(). | |
3227 | */ | |
3228 | /* ARGSUSED */ | |
3229 | vdev_t * | |
3230 | online_vdev(vdev_t *vd, void *arg) | |
3231 | { | |
3232 | spa_t *spa = vd->vdev_spa; | |
3233 | vdev_t *tvd = vd->vdev_top; | |
3234 | uint64_t guid = vd->vdev_guid; | |
3235 | uint64_t generation = spa->spa_config_generation + 1; | |
3236 | vdev_state_t newstate = VDEV_STATE_UNKNOWN; | |
3237 | int error; | |
3238 | ||
3239 | ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE); | |
3240 | ASSERT(vd->vdev_ops->vdev_op_leaf); | |
3241 | ||
3242 | /* Calling vdev_online will initialize the new metaslabs */ | |
3243 | spa_config_exit(spa, SCL_STATE, spa); | |
3244 | error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate); | |
3245 | spa_config_enter(spa, SCL_STATE, spa, RW_READER); | |
3246 | ||
3247 | /* | |
3248 | * If vdev_online returned an error or the underlying vdev_open | |
3249 | * failed then we abort the expand. The only way to know that | |
3250 | * vdev_open fails is by checking the returned newstate. | |
3251 | */ | |
3252 | if (error || newstate != VDEV_STATE_HEALTHY) { | |
3253 | if (ztest_opts.zo_verbose >= 5) { | |
3254 | (void) printf("Unable to expand vdev, state %llu, " | |
3255 | "error %d\n", (u_longlong_t)newstate, error); | |
3256 | } | |
3257 | return (vd); | |
3258 | } | |
3259 | ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY); | |
3260 | ||
3261 | /* | |
3262 | * Since we dropped the lock we need to ensure that we're | |
3263 | * still talking to the original vdev. It's possible this | |
3264 | * vdev may have been detached/replaced while we were | |
3265 | * trying to online it. | |
3266 | */ | |
3267 | if (generation != spa->spa_config_generation) { | |
3268 | if (ztest_opts.zo_verbose >= 5) { | |
3269 | (void) printf("vdev configuration has changed, " | |
3270 | "guid %llu, state %llu, expected gen %llu, " | |
3271 | "got gen %llu\n", | |
3272 | (u_longlong_t)guid, | |
3273 | (u_longlong_t)tvd->vdev_state, | |
3274 | (u_longlong_t)generation, | |
3275 | (u_longlong_t)spa->spa_config_generation); | |
3276 | } | |
3277 | return (vd); | |
3278 | } | |
3279 | return (NULL); | |
3280 | } | |
3281 | ||
3282 | /* | |
3283 | * Traverse the vdev tree calling the supplied function. | |
3284 | * We continue to walk the tree until we either have walked all | |
3285 | * children or we receive a non-NULL return from the callback. | |
3286 | * If a NULL callback is passed, then we just return back the first | |
3287 | * leaf vdev we encounter. | |
3288 | */ | |
3289 | vdev_t * | |
3290 | vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg) | |
3291 | { | |
3292 | uint_t c; | |
3293 | ||
3294 | if (vd->vdev_ops->vdev_op_leaf) { | |
3295 | if (func == NULL) | |
3296 | return (vd); | |
3297 | else | |
3298 | return (func(vd, arg)); | |
3299 | } | |
3300 | ||
3301 | for (c = 0; c < vd->vdev_children; c++) { | |
3302 | vdev_t *cvd = vd->vdev_child[c]; | |
3303 | if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL) | |
3304 | return (cvd); | |
3305 | } | |
3306 | return (NULL); | |
3307 | } | |
3308 | ||
3309 | /* | |
3310 | * Verify that dynamic LUN growth works as expected. | |
3311 | */ | |
3312 | /* ARGSUSED */ | |
3313 | void | |
3314 | ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id) | |
3315 | { | |
3316 | spa_t *spa = ztest_spa; | |
3317 | vdev_t *vd, *tvd; | |
3318 | metaslab_class_t *mc; | |
3319 | metaslab_group_t *mg; | |
3320 | size_t psize, newsize; | |
3321 | uint64_t top; | |
3322 | uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count; | |
3323 | ||
3324 | mutex_enter(&ztest_vdev_lock); | |
3325 | spa_config_enter(spa, SCL_STATE, spa, RW_READER); | |
3326 | ||
3327 | top = ztest_random_vdev_top(spa, B_TRUE); | |
3328 | ||
3329 | tvd = spa->spa_root_vdev->vdev_child[top]; | |
3330 | mg = tvd->vdev_mg; | |
3331 | mc = mg->mg_class; | |
3332 | old_ms_count = tvd->vdev_ms_count; | |
3333 | old_class_space = metaslab_class_get_space(mc); | |
3334 | ||
3335 | /* | |
3336 | * Determine the size of the first leaf vdev associated with | |
3337 | * our top-level device. | |
3338 | */ | |
3339 | vd = vdev_walk_tree(tvd, NULL, NULL); | |
3340 | ASSERT3P(vd, !=, NULL); | |
3341 | ASSERT(vd->vdev_ops->vdev_op_leaf); | |
3342 | ||
3343 | psize = vd->vdev_psize; | |
3344 | ||
3345 | /* | |
3346 | * We only try to expand the vdev if it's healthy, less than 4x its | |
3347 | * original size, and it has a valid psize. | |
3348 | */ | |
3349 | if (tvd->vdev_state != VDEV_STATE_HEALTHY || | |
3350 | psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) { | |
3351 | spa_config_exit(spa, SCL_STATE, spa); | |
3352 | mutex_exit(&ztest_vdev_lock); | |
3353 | return; | |
3354 | } | |
3355 | ASSERT(psize > 0); | |
3356 | newsize = psize + psize / 8; | |
3357 | ASSERT3U(newsize, >, psize); | |
3358 | ||
3359 | if (ztest_opts.zo_verbose >= 6) { | |
3360 | (void) printf("Expanding LUN %s from %lu to %lu\n", | |
3361 | vd->vdev_path, (ulong_t)psize, (ulong_t)newsize); | |
3362 | } | |
3363 | ||
3364 | /* | |
3365 | * Growing the vdev is a two step process: | |
3366 | * 1). expand the physical size (i.e. relabel) | |
3367 | * 2). online the vdev to create the new metaslabs | |
3368 | */ | |
3369 | if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL || | |
3370 | vdev_walk_tree(tvd, online_vdev, NULL) != NULL || | |
3371 | tvd->vdev_state != VDEV_STATE_HEALTHY) { | |
3372 | if (ztest_opts.zo_verbose >= 5) { | |
3373 | (void) printf("Could not expand LUN because " | |
3374 | "the vdev configuration changed.\n"); | |
3375 | } | |
3376 | spa_config_exit(spa, SCL_STATE, spa); | |
3377 | mutex_exit(&ztest_vdev_lock); | |
3378 | return; | |
3379 | } | |
3380 | ||
3381 | spa_config_exit(spa, SCL_STATE, spa); | |
3382 | ||
3383 | /* | |
3384 | * Expanding the LUN will update the config asynchronously, | |
3385 | * thus we must wait for the async thread to complete any | |
3386 | * pending tasks before proceeding. | |
3387 | */ | |
3388 | for (;;) { | |
3389 | boolean_t done; | |
3390 | mutex_enter(&spa->spa_async_lock); | |
3391 | done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks); | |
3392 | mutex_exit(&spa->spa_async_lock); | |
3393 | if (done) | |
3394 | break; | |
3395 | txg_wait_synced(spa_get_dsl(spa), 0); | |
3396 | (void) poll(NULL, 0, 100); | |
3397 | } | |
3398 | ||
3399 | spa_config_enter(spa, SCL_STATE, spa, RW_READER); | |
3400 | ||
3401 | tvd = spa->spa_root_vdev->vdev_child[top]; | |
3402 | new_ms_count = tvd->vdev_ms_count; | |
3403 | new_class_space = metaslab_class_get_space(mc); | |
3404 | ||
3405 | if (tvd->vdev_mg != mg || mg->mg_class != mc) { | |
3406 | if (ztest_opts.zo_verbose >= 5) { | |
3407 | (void) printf("Could not verify LUN expansion due to " | |
3408 | "intervening vdev offline or remove.\n"); | |
3409 | } | |
3410 | spa_config_exit(spa, SCL_STATE, spa); | |
3411 | mutex_exit(&ztest_vdev_lock); | |
3412 | return; | |
3413 | } | |
3414 | ||
3415 | /* | |
3416 | * Make sure we were able to grow the vdev. | |
3417 | */ | |
3418 | if (new_ms_count <= old_ms_count) | |
3419 | fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n", | |
3420 | old_ms_count, new_ms_count); | |
3421 | ||
3422 | /* | |
3423 | * Make sure we were able to grow the pool. | |
3424 | */ | |
3425 | if (new_class_space <= old_class_space) | |
3426 | fatal(0, "LUN expansion failed: class_space %llu <= %llu\n", | |
3427 | old_class_space, new_class_space); | |
3428 | ||
3429 | if (ztest_opts.zo_verbose >= 5) { | |
3430 | char oldnumbuf[6], newnumbuf[6]; | |
3431 | ||
3432 | nicenum(old_class_space, oldnumbuf); | |
3433 | nicenum(new_class_space, newnumbuf); | |
3434 | (void) printf("%s grew from %s to %s\n", | |
3435 | spa->spa_name, oldnumbuf, newnumbuf); | |
3436 | } | |
3437 | ||
3438 | spa_config_exit(spa, SCL_STATE, spa); | |
3439 | mutex_exit(&ztest_vdev_lock); | |
3440 | } | |
3441 | ||
3442 | /* | |
3443 | * Verify that dmu_objset_{create,destroy,open,close} work as expected. | |
3444 | */ | |
3445 | /* ARGSUSED */ | |
3446 | static void | |
3447 | ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx) | |
3448 | { | |
3449 | /* | |
3450 | * Create the objects common to all ztest datasets. | |
3451 | */ | |
3452 | VERIFY(zap_create_claim(os, ZTEST_DIROBJ, | |
3453 | DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0); | |
3454 | } | |
3455 | ||
3456 | static int | |
3457 | ztest_dataset_create(char *dsname) | |
3458 | { | |
3459 | uint64_t zilset = ztest_random(100); | |
3460 | int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0, | |
3461 | ztest_objset_create_cb, NULL); | |
3462 | ||
3463 | if (err || zilset < 80) | |
3464 | return (err); | |
3465 | ||
3466 | if (ztest_opts.zo_verbose >= 5) | |
3467 | (void) printf("Setting dataset %s to sync always\n", dsname); | |
3468 | return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC, | |
3469 | ZFS_SYNC_ALWAYS, B_FALSE)); | |
3470 | } | |
3471 | ||
3472 | /* ARGSUSED */ | |
3473 | static int | |
3474 | ztest_objset_destroy_cb(const char *name, void *arg) | |
3475 | { | |
3476 | objset_t *os; | |
3477 | dmu_object_info_t doi; | |
3478 | int error; | |
3479 | ||
3480 | /* | |
3481 | * Verify that the dataset contains a directory object. | |
3482 | */ | |
3483 | VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, FTAG, &os)); | |
3484 | error = dmu_object_info(os, ZTEST_DIROBJ, &doi); | |
3485 | if (error != ENOENT) { | |
3486 | /* We could have crashed in the middle of destroying it */ | |
3487 | ASSERT0(error); | |
3488 | ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER); | |
3489 | ASSERT3S(doi.doi_physical_blocks_512, >=, 0); | |
3490 | } | |
3491 | dmu_objset_disown(os, FTAG); | |
3492 | ||
3493 | /* | |
3494 | * Destroy the dataset. | |
3495 | */ | |
3496 | if (strchr(name, '@') != NULL) { | |
3497 | VERIFY0(dsl_destroy_snapshot(name, B_FALSE)); | |
3498 | } else { | |
3499 | VERIFY0(dsl_destroy_head(name)); | |
3500 | } | |
3501 | return (0); | |
3502 | } | |
3503 | ||
3504 | static boolean_t | |
3505 | ztest_snapshot_create(char *osname, uint64_t id) | |
3506 | { | |
3507 | char snapname[MAXNAMELEN]; | |
3508 | int error; | |
3509 | ||
3510 | (void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id); | |
3511 | ||
3512 | error = dmu_objset_snapshot_one(osname, snapname); | |
3513 | if (error == ENOSPC) { | |
3514 | ztest_record_enospc(FTAG); | |
3515 | return (B_FALSE); | |
3516 | } | |
3517 | if (error != 0 && error != EEXIST) { | |
3518 | fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname, | |
3519 | snapname, error); | |
3520 | } | |
3521 | return (B_TRUE); | |
3522 | } | |
3523 | ||
3524 | static boolean_t | |
3525 | ztest_snapshot_destroy(char *osname, uint64_t id) | |
3526 | { | |
3527 | char snapname[MAXNAMELEN]; | |
3528 | int error; | |
3529 | ||
3530 | (void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname, | |
3531 | (u_longlong_t)id); | |
3532 | ||
3533 | error = dsl_destroy_snapshot(snapname, B_FALSE); | |
3534 | if (error != 0 && error != ENOENT) | |
3535 | fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error); | |
3536 | return (B_TRUE); | |
3537 | } | |
3538 | ||
3539 | /* ARGSUSED */ | |
3540 | void | |
3541 | ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id) | |
3542 | { | |
3543 | ztest_ds_t *zdtmp; | |
3544 | int iters; | |
3545 | int error; | |
3546 | objset_t *os, *os2; | |
3547 | char *name; | |
3548 | zilog_t *zilog; | |
3549 | int i; | |
3550 | ||
3551 | zdtmp = umem_alloc(sizeof (ztest_ds_t), UMEM_NOFAIL); | |
3552 | name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); | |
3553 | ||
3554 | (void) rw_rdlock(&ztest_name_lock); | |
3555 | ||
3556 | (void) snprintf(name, MAXNAMELEN, "%s/temp_%llu", | |
3557 | ztest_opts.zo_pool, (u_longlong_t)id); | |
3558 | ||
3559 | /* | |
3560 | * If this dataset exists from a previous run, process its replay log | |
3561 | * half of the time. If we don't replay it, then dsl_destroy_head() | |
3562 | * (invoked from ztest_objset_destroy_cb()) should just throw it away. | |
3563 | */ | |
3564 | if (ztest_random(2) == 0 && | |
3565 | dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) { | |
3566 | ztest_zd_init(zdtmp, NULL, os); | |
3567 | zil_replay(os, zdtmp, ztest_replay_vector); | |
3568 | ztest_zd_fini(zdtmp); | |
3569 | dmu_objset_disown(os, FTAG); | |
3570 | } | |
3571 | ||
3572 | /* | |
3573 | * There may be an old instance of the dataset we're about to | |
3574 | * create lying around from a previous run. If so, destroy it | |
3575 | * and all of its snapshots. | |
3576 | */ | |
3577 | (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL, | |
3578 | DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS); | |
3579 | ||
3580 | /* | |
3581 | * Verify that the destroyed dataset is no longer in the namespace. | |
3582 | */ | |
3583 | VERIFY3U(ENOENT, ==, dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, | |
3584 | FTAG, &os)); | |
3585 | ||
3586 | /* | |
3587 | * Verify that we can create a new dataset. | |
3588 | */ | |
3589 | error = ztest_dataset_create(name); | |
3590 | if (error) { | |
3591 | if (error == ENOSPC) { | |
3592 | ztest_record_enospc(FTAG); | |
3593 | goto out; | |
3594 | } | |
3595 | fatal(0, "dmu_objset_create(%s) = %d", name, error); | |
3596 | } | |
3597 | ||
3598 | VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os)); | |
3599 | ||
3600 | ztest_zd_init(zdtmp, NULL, os); | |
3601 | ||
3602 | /* | |
3603 | * Open the intent log for it. | |
3604 | */ | |
3605 | zilog = zil_open(os, ztest_get_data); | |
3606 | ||
3607 | /* | |
3608 | * Put some objects in there, do a little I/O to them, | |
3609 | * and randomly take a couple of snapshots along the way. | |
3610 | */ | |
3611 | iters = ztest_random(5); | |
3612 | for (i = 0; i < iters; i++) { | |
3613 | ztest_dmu_object_alloc_free(zdtmp, id); | |
3614 | if (ztest_random(iters) == 0) | |
3615 | (void) ztest_snapshot_create(name, i); | |
3616 | } | |
3617 | ||
3618 | /* | |
3619 | * Verify that we cannot create an existing dataset. | |
3620 | */ | |
3621 | VERIFY3U(EEXIST, ==, | |
3622 | dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL)); | |
3623 | ||
3624 | /* | |
3625 | * Verify that we can hold an objset that is also owned. | |
3626 | */ | |
3627 | VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2)); | |
3628 | dmu_objset_rele(os2, FTAG); | |
3629 | ||
3630 | /* | |
3631 | * Verify that we cannot own an objset that is already owned. | |
3632 | */ | |
3633 | VERIFY3U(EBUSY, ==, | |
3634 | dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2)); | |
3635 | ||
3636 | zil_close(zilog); | |
3637 | dmu_objset_disown(os, FTAG); | |
3638 | ztest_zd_fini(zdtmp); | |
3639 | out: | |
3640 | (void) rw_unlock(&ztest_name_lock); | |
3641 | ||
3642 | umem_free(name, MAXNAMELEN); | |
3643 | umem_free(zdtmp, sizeof (ztest_ds_t)); | |
3644 | } | |
3645 | ||
3646 | /* | |
3647 | * Verify that dmu_snapshot_{create,destroy,open,close} work as expected. | |
3648 | */ | |
3649 | void | |
3650 | ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id) | |
3651 | { | |
3652 | (void) rw_rdlock(&ztest_name_lock); | |
3653 | (void) ztest_snapshot_destroy(zd->zd_name, id); | |
3654 | (void) ztest_snapshot_create(zd->zd_name, id); | |
3655 | (void) rw_unlock(&ztest_name_lock); | |
3656 | } | |
3657 | ||
3658 | /* | |
3659 | * Cleanup non-standard snapshots and clones. | |
3660 | */ | |
3661 | void | |
3662 | ztest_dsl_dataset_cleanup(char *osname, uint64_t id) | |
3663 | { | |
3664 | char *snap1name; | |
3665 | char *clone1name; | |
3666 | char *snap2name; | |
3667 | char *clone2name; | |
3668 | char *snap3name; | |
3669 | int error; | |
3670 | ||
3671 | snap1name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); | |
3672 | clone1name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); | |
3673 | snap2name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); | |
3674 | clone2name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); | |
3675 | snap3name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); | |
3676 | ||
3677 | (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", | |
3678 | osname, (u_longlong_t)id); | |
3679 | (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", | |
3680 | osname, (u_longlong_t)id); | |
3681 | (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", | |
3682 | clone1name, (u_longlong_t)id); | |
3683 | (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", | |
3684 | osname, (u_longlong_t)id); | |
3685 | (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", | |
3686 | clone1name, (u_longlong_t)id); | |
3687 | ||
3688 | error = dsl_destroy_head(clone2name); | |
3689 | if (error && error != ENOENT) | |
3690 | fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error); | |
3691 | error = dsl_destroy_snapshot(snap3name, B_FALSE); | |
3692 | if (error && error != ENOENT) | |
3693 | fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error); | |
3694 | error = dsl_destroy_snapshot(snap2name, B_FALSE); | |
3695 | if (error && error != ENOENT) | |
3696 | fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error); | |
3697 | error = dsl_destroy_head(clone1name); | |
3698 | if (error && error != ENOENT) | |
3699 | fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error); | |
3700 | error = dsl_destroy_snapshot(snap1name, B_FALSE); | |
3701 | if (error && error != ENOENT) | |
3702 | fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error); | |
3703 | ||
3704 | umem_free(snap1name, MAXNAMELEN); | |
3705 | umem_free(clone1name, MAXNAMELEN); | |
3706 | umem_free(snap2name, MAXNAMELEN); | |
3707 | umem_free(clone2name, MAXNAMELEN); | |
3708 | umem_free(snap3name, MAXNAMELEN); | |
3709 | } | |
3710 | ||
3711 | /* | |
3712 | * Verify dsl_dataset_promote handles EBUSY | |
3713 | */ | |
3714 | void | |
3715 | ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id) | |
3716 | { | |
3717 | objset_t *os; | |
3718 | char *snap1name; | |
3719 | char *clone1name; | |
3720 | char *snap2name; | |
3721 | char *clone2name; | |
3722 | char *snap3name; | |
3723 | char *osname = zd->zd_name; | |
3724 | int error; | |
3725 | ||
3726 | snap1name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); | |
3727 | clone1name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); | |
3728 | snap2name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); | |
3729 | clone2name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); | |
3730 | snap3name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); | |
3731 | ||
3732 | (void) rw_rdlock(&ztest_name_lock); | |
3733 | ||
3734 | ztest_dsl_dataset_cleanup(osname, id); | |
3735 | ||
3736 | (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", | |
3737 | osname, (u_longlong_t)id); | |
3738 | (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", | |
3739 | osname, (u_longlong_t)id); | |
3740 | (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", | |
3741 | clone1name, (u_longlong_t)id); | |
3742 | (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", | |
3743 | osname, (u_longlong_t)id); | |
3744 | (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", | |
3745 | clone1name, (u_longlong_t)id); | |
3746 | ||
3747 | error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1); | |
3748 | if (error && error != EEXIST) { | |
3749 | if (error == ENOSPC) { | |
3750 | ztest_record_enospc(FTAG); | |
3751 | goto out; | |
3752 | } | |
3753 | fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error); | |
3754 | } | |
3755 | ||
3756 | error = dmu_objset_clone(clone1name, snap1name); | |
3757 | if (error) { | |
3758 | if (error == ENOSPC) { | |
3759 | ztest_record_enospc(FTAG); | |
3760 | goto out; | |
3761 | } | |
3762 | fatal(0, "dmu_objset_create(%s) = %d", clone1name, error); | |
3763 | } | |
3764 | ||
3765 | error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1); | |
3766 | if (error && error != EEXIST) { | |
3767 | if (error == ENOSPC) { | |
3768 | ztest_record_enospc(FTAG); | |
3769 | goto out; | |
3770 | } | |
3771 | fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error); | |
3772 | } | |
3773 | ||
3774 | error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1); | |
3775 | if (error && error != EEXIST) { | |
3776 | if (error == ENOSPC) { | |
3777 | ztest_record_enospc(FTAG); | |
3778 | goto out; | |
3779 | } | |
3780 | fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error); | |
3781 | } | |
3782 | ||
3783 | error = dmu_objset_clone(clone2name, snap3name); | |
3784 | if (error) { | |
3785 | if (error == ENOSPC) { | |
3786 | ztest_record_enospc(FTAG); | |
3787 | goto out; | |
3788 | } | |
3789 | fatal(0, "dmu_objset_create(%s) = %d", clone2name, error); | |
3790 | } | |
3791 | ||
3792 | error = dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, FTAG, &os); | |
3793 | if (error) | |
3794 | fatal(0, "dmu_objset_own(%s) = %d", snap2name, error); | |
3795 | error = dsl_dataset_promote(clone2name, NULL); | |
3796 | if (error == ENOSPC) { | |
3797 | dmu_objset_disown(os, FTAG); | |
3798 | ztest_record_enospc(FTAG); | |
3799 | goto out; | |
3800 | } | |
3801 | if (error != EBUSY) | |
3802 | fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name, | |
3803 | error); | |
3804 | dmu_objset_disown(os, FTAG); | |
3805 | ||
3806 | out: | |
3807 | ztest_dsl_dataset_cleanup(osname, id); | |
3808 | ||
3809 | (void) rw_unlock(&ztest_name_lock); | |
3810 | ||
3811 | umem_free(snap1name, MAXNAMELEN); | |
3812 | umem_free(clone1name, MAXNAMELEN); | |
3813 | umem_free(snap2name, MAXNAMELEN); | |
3814 | umem_free(clone2name, MAXNAMELEN); | |
3815 | umem_free(snap3name, MAXNAMELEN); | |
3816 | } | |
3817 | ||
3818 | #undef OD_ARRAY_SIZE | |
3819 | #define OD_ARRAY_SIZE 4 | |
3820 | ||
3821 | /* | |
3822 | * Verify that dmu_object_{alloc,free} work as expected. | |
3823 | */ | |
3824 | void | |
3825 | ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id) | |
3826 | { | |
3827 | ztest_od_t *od; | |
3828 | int batchsize; | |
3829 | int size; | |
3830 | int b; | |
3831 | ||
3832 | size = sizeof (ztest_od_t) * OD_ARRAY_SIZE; | |
3833 | od = umem_alloc(size, UMEM_NOFAIL); | |
3834 | batchsize = OD_ARRAY_SIZE; | |
3835 | ||
3836 | for (b = 0; b < batchsize; b++) | |
3837 | ztest_od_init(od + b, id, FTAG, b, DMU_OT_UINT64_OTHER, | |
3838 | 0, 0, 0); | |
3839 | ||
3840 | /* | |
3841 | * Destroy the previous batch of objects, create a new batch, | |
3842 | * and do some I/O on the new objects. | |
3843 | */ | |
3844 | if (ztest_object_init(zd, od, size, B_TRUE) != 0) | |
3845 | return; | |
3846 | ||
3847 | while (ztest_random(4 * batchsize) != 0) | |
3848 | ztest_io(zd, od[ztest_random(batchsize)].od_object, | |
3849 | ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT); | |
3850 | ||
3851 | umem_free(od, size); | |
3852 | } | |
3853 | ||
3854 | #undef OD_ARRAY_SIZE | |
3855 | #define OD_ARRAY_SIZE 2 | |
3856 | ||
3857 | /* | |
3858 | * Verify that dmu_{read,write} work as expected. | |
3859 | */ | |
3860 | void | |
3861 | ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id) | |
3862 | { | |
3863 | int size; | |
3864 | ztest_od_t *od; | |
3865 | ||
3866 | objset_t *os = zd->zd_os; | |
3867 | size = sizeof (ztest_od_t) * OD_ARRAY_SIZE; | |
3868 | od = umem_alloc(size, UMEM_NOFAIL); | |
3869 | dmu_tx_t *tx; | |
3870 | int i, freeit, error; | |
3871 | uint64_t n, s, txg; | |
3872 | bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT; | |
3873 | uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize; | |
3874 | uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t); | |
3875 | uint64_t regions = 997; | |
3876 | uint64_t stride = 123456789ULL; | |
3877 | uint64_t width = 40; | |
3878 | int free_percent = 5; | |
3879 | ||
3880 | /* | |
3881 | * This test uses two objects, packobj and bigobj, that are always | |
3882 | * updated together (i.e. in the same tx) so that their contents are | |
3883 | * in sync and can be compared. Their contents relate to each other | |
3884 | * in a simple way: packobj is a dense array of 'bufwad' structures, | |
3885 | * while bigobj is a sparse array of the same bufwads. Specifically, | |
3886 | * for any index n, there are three bufwads that should be identical: | |
3887 | * | |
3888 | * packobj, at offset n * sizeof (bufwad_t) | |
3889 | * bigobj, at the head of the nth chunk | |
3890 | * bigobj, at the tail of the nth chunk | |
3891 | * | |
3892 | * The chunk size is arbitrary. It doesn't have to be a power of two, | |
3893 | * and it doesn't have any relation to the object blocksize. | |
3894 | * The only requirement is that it can hold at least two bufwads. | |
3895 | * | |
3896 | * Normally, we write the bufwad to each of these locations. | |
3897 | * However, free_percent of the time we instead write zeroes to | |
3898 | * packobj and perform a dmu_free_range() on bigobj. By comparing | |
3899 | * bigobj to packobj, we can verify that the DMU is correctly | |
3900 | * tracking which parts of an object are allocated and free, | |
3901 | * and that the contents of the allocated blocks are correct. | |
3902 | */ | |
3903 | ||
3904 | /* | |
3905 | * Read the directory info. If it's the first time, set things up. | |
3906 | */ | |
3907 | ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, chunksize); | |
3908 | ztest_od_init(od + 1, id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, 0, | |
3909 | chunksize); | |
3910 | ||
3911 | if (ztest_object_init(zd, od, size, B_FALSE) != 0) { | |
3912 | umem_free(od, size); | |
3913 | return; | |
3914 | } | |
3915 | ||
3916 | bigobj = od[0].od_object; | |
3917 | packobj = od[1].od_object; | |
3918 | chunksize = od[0].od_gen; | |
3919 | ASSERT(chunksize == od[1].od_gen); | |
3920 | ||
3921 | /* | |
3922 | * Prefetch a random chunk of the big object. | |
3923 | * Our aim here is to get some async reads in flight | |
3924 | * for blocks that we may free below; the DMU should | |
3925 | * handle this race correctly. | |
3926 | */ | |
3927 | n = ztest_random(regions) * stride + ztest_random(width); | |
3928 | s = 1 + ztest_random(2 * width - 1); | |
3929 | dmu_prefetch(os, bigobj, 0, n * chunksize, s * chunksize, | |
3930 | ZIO_PRIORITY_SYNC_READ); | |
3931 | ||
3932 | /* | |
3933 | * Pick a random index and compute the offsets into packobj and bigobj. | |
3934 | */ | |
3935 | n = ztest_random(regions) * stride + ztest_random(width); | |
3936 | s = 1 + ztest_random(width - 1); | |
3937 | ||
3938 | packoff = n * sizeof (bufwad_t); | |
3939 | packsize = s * sizeof (bufwad_t); | |
3940 | ||
3941 | bigoff = n * chunksize; | |
3942 | bigsize = s * chunksize; | |
3943 | ||
3944 | packbuf = umem_alloc(packsize, UMEM_NOFAIL); | |
3945 | bigbuf = umem_alloc(bigsize, UMEM_NOFAIL); | |
3946 | ||
3947 | /* | |
3948 | * free_percent of the time, free a range of bigobj rather than | |
3949 | * overwriting it. | |
3950 | */ | |
3951 | freeit = (ztest_random(100) < free_percent); | |
3952 | ||
3953 | /* | |
3954 | * Read the current contents of our objects. | |
3955 | */ | |
3956 | error = dmu_read(os, packobj, packoff, packsize, packbuf, | |
3957 | DMU_READ_PREFETCH); | |
3958 | ASSERT0(error); | |
3959 | error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf, | |
3960 | DMU_READ_PREFETCH); | |
3961 | ASSERT0(error); | |
3962 | ||
3963 | /* | |
3964 | * Get a tx for the mods to both packobj and bigobj. | |
3965 | */ | |
3966 | tx = dmu_tx_create(os); | |
3967 | ||
3968 | dmu_tx_hold_write(tx, packobj, packoff, packsize); | |
3969 | ||
3970 | if (freeit) | |
3971 | dmu_tx_hold_free(tx, bigobj, bigoff, bigsize); | |
3972 | else | |
3973 | dmu_tx_hold_write(tx, bigobj, bigoff, bigsize); | |
3974 | ||
3975 | /* This accounts for setting the checksum/compression. */ | |
3976 | dmu_tx_hold_bonus(tx, bigobj); | |
3977 | ||
3978 | txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); | |
3979 | if (txg == 0) { | |
3980 | umem_free(packbuf, packsize); | |
3981 | umem_free(bigbuf, bigsize); | |
3982 | umem_free(od, size); | |
3983 | return; | |
3984 | } | |
3985 | ||
3986 | enum zio_checksum cksum; | |
3987 | do { | |
3988 | cksum = (enum zio_checksum) | |
3989 | ztest_random_dsl_prop(ZFS_PROP_CHECKSUM); | |
3990 | } while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS); | |
3991 | dmu_object_set_checksum(os, bigobj, cksum, tx); | |
3992 | ||
3993 | enum zio_compress comp; | |
3994 | do { | |
3995 | comp = (enum zio_compress) | |
3996 | ztest_random_dsl_prop(ZFS_PROP_COMPRESSION); | |
3997 | } while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS); | |
3998 | dmu_object_set_compress(os, bigobj, comp, tx); | |
3999 | ||
4000 | /* | |
4001 | * For each index from n to n + s, verify that the existing bufwad | |
4002 | * in packobj matches the bufwads at the head and tail of the | |
4003 | * corresponding chunk in bigobj. Then update all three bufwads | |
4004 | * with the new values we want to write out. | |
4005 | */ | |
4006 | for (i = 0; i < s; i++) { | |
4007 | /* LINTED */ | |
4008 | pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t)); | |
4009 | /* LINTED */ | |
4010 | bigH = (bufwad_t *)((char *)bigbuf + i * chunksize); | |
4011 | /* LINTED */ | |
4012 | bigT = (bufwad_t *)((char *)bigH + chunksize) - 1; | |
4013 | ||
4014 | ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize); | |
4015 | ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize); | |
4016 | ||
4017 | if (pack->bw_txg > txg) | |
4018 | fatal(0, "future leak: got %llx, open txg is %llx", | |
4019 | pack->bw_txg, txg); | |
4020 | ||
4021 | if (pack->bw_data != 0 && pack->bw_index != n + i) | |
4022 | fatal(0, "wrong index: got %llx, wanted %llx+%llx", | |
4023 | pack->bw_index, n, i); | |
4024 | ||
4025 | if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0) | |
4026 | fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH); | |
4027 | ||
4028 | if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0) | |
4029 | fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT); | |
4030 | ||
4031 | if (freeit) { | |
4032 | bzero(pack, sizeof (bufwad_t)); | |
4033 | } else { | |
4034 | pack->bw_index = n + i; | |
4035 | pack->bw_txg = txg; | |
4036 | pack->bw_data = 1 + ztest_random(-2ULL); | |
4037 | } | |
4038 | *bigH = *pack; | |
4039 | *bigT = *pack; | |
4040 | } | |
4041 | ||
4042 | /* | |
4043 | * We've verified all the old bufwads, and made new ones. | |
4044 | * Now write them out. | |
4045 | */ | |
4046 | dmu_write(os, packobj, packoff, packsize, packbuf, tx); | |
4047 | ||
4048 | if (freeit) { | |
4049 | if (ztest_opts.zo_verbose >= 7) { | |
4050 | (void) printf("freeing offset %llx size %llx" | |
4051 | " txg %llx\n", | |
4052 | (u_longlong_t)bigoff, | |
4053 | (u_longlong_t)bigsize, | |
4054 | (u_longlong_t)txg); | |
4055 | } | |
4056 | VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx)); | |
4057 | } else { | |
4058 | if (ztest_opts.zo_verbose >= 7) { | |
4059 | (void) printf("writing offset %llx size %llx" | |
4060 | " txg %llx\n", | |
4061 | (u_longlong_t)bigoff, | |
4062 | (u_longlong_t)bigsize, | |
4063 | (u_longlong_t)txg); | |
4064 | } | |
4065 | dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx); | |
4066 | } | |
4067 | ||
4068 | dmu_tx_commit(tx); | |
4069 | ||
4070 | /* | |
4071 | * Sanity check the stuff we just wrote. | |
4072 | */ | |
4073 | { | |
4074 | void *packcheck = umem_alloc(packsize, UMEM_NOFAIL); | |
4075 | void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL); | |
4076 | ||
4077 | VERIFY(0 == dmu_read(os, packobj, packoff, | |
4078 | packsize, packcheck, DMU_READ_PREFETCH)); | |
4079 | VERIFY(0 == dmu_read(os, bigobj, bigoff, | |
4080 | bigsize, bigcheck, DMU_READ_PREFETCH)); | |
4081 | ||
4082 | ASSERT(bcmp(packbuf, packcheck, packsize) == 0); | |
4083 | ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0); | |
4084 | ||
4085 | umem_free(packcheck, packsize); | |
4086 | umem_free(bigcheck, bigsize); | |
4087 | } | |
4088 | ||
4089 | umem_free(packbuf, packsize); | |
4090 | umem_free(bigbuf, bigsize); | |
4091 | umem_free(od, size); | |
4092 | } | |
4093 | ||
4094 | void | |
4095 | compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf, | |
4096 | uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg) | |
4097 | { | |
4098 | uint64_t i; | |
4099 | bufwad_t *pack; | |
4100 | bufwad_t *bigH; | |
4101 | bufwad_t *bigT; | |
4102 | ||
4103 | /* | |
4104 | * For each index from n to n + s, verify that the existing bufwad | |
4105 | * in packobj matches the bufwads at the head and tail of the | |
4106 | * corresponding chunk in bigobj. Then update all three bufwads | |
4107 | * with the new values we want to write out. | |
4108 | */ | |
4109 | for (i = 0; i < s; i++) { | |
4110 | /* LINTED */ | |
4111 | pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t)); | |
4112 | /* LINTED */ | |
4113 | bigH = (bufwad_t *)((char *)bigbuf + i * chunksize); | |
4114 | /* LINTED */ | |
4115 | bigT = (bufwad_t *)((char *)bigH + chunksize) - 1; | |
4116 | ||
4117 | ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize); | |
4118 | ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize); | |
4119 | ||
4120 | if (pack->bw_txg > txg) | |
4121 | fatal(0, "future leak: got %llx, open txg is %llx", | |
4122 | pack->bw_txg, txg); | |
4123 | ||
4124 | if (pack->bw_data != 0 && pack->bw_index != n + i) | |
4125 | fatal(0, "wrong index: got %llx, wanted %llx+%llx", | |
4126 | pack->bw_index, n, i); | |
4127 | ||
4128 | if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0) | |
4129 | fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH); | |
4130 | ||
4131 | if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0) | |
4132 | fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT); | |
4133 | ||
4134 | pack->bw_index = n + i; | |
4135 | pack->bw_txg = txg; | |
4136 | pack->bw_data = 1 + ztest_random(-2ULL); | |
4137 | ||
4138 | *bigH = *pack; | |
4139 | *bigT = *pack; | |
4140 | } | |
4141 | } | |
4142 | ||
4143 | #undef OD_ARRAY_SIZE | |
4144 | #define OD_ARRAY_SIZE 2 | |
4145 | ||
4146 | void | |
4147 | ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id) | |
4148 | { | |
4149 | objset_t *os = zd->zd_os; | |
4150 | ztest_od_t *od; | |
4151 | dmu_tx_t *tx; | |
4152 | uint64_t i; | |
4153 | int error; | |
4154 | int size; | |
4155 | uint64_t n, s, txg; | |
4156 | bufwad_t *packbuf, *bigbuf; | |
4157 | uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize; | |
4158 | uint64_t blocksize = ztest_random_blocksize(); | |
4159 | uint64_t chunksize = blocksize; | |
4160 | uint64_t regions = 997; | |
4161 | uint64_t stride = 123456789ULL; | |
4162 | uint64_t width = 9; | |
4163 | dmu_buf_t *bonus_db; | |
4164 | arc_buf_t **bigbuf_arcbufs; | |
4165 | dmu_object_info_t doi; | |
4166 | ||
4167 | size = sizeof (ztest_od_t) * OD_ARRAY_SIZE; | |
4168 | od = umem_alloc(size, UMEM_NOFAIL); | |
4169 | ||
4170 | /* | |
4171 | * This test uses two objects, packobj and bigobj, that are always | |
4172 | * updated together (i.e. in the same tx) so that their contents are | |
4173 | * in sync and can be compared. Their contents relate to each other | |
4174 | * in a simple way: packobj is a dense array of 'bufwad' structures, | |
4175 | * while bigobj is a sparse array of the same bufwads. Specifically, | |
4176 | * for any index n, there are three bufwads that should be identical: | |
4177 | * | |
4178 | * packobj, at offset n * sizeof (bufwad_t) | |
4179 | * bigobj, at the head of the nth chunk | |
4180 | * bigobj, at the tail of the nth chunk | |
4181 | * | |
4182 | * The chunk size is set equal to bigobj block size so that | |
4183 | * dmu_assign_arcbuf() can be tested for object updates. | |
4184 | */ | |
4185 | ||
4186 | /* | |
4187 | * Read the directory info. If it's the first time, set things up. | |
4188 | */ | |
4189 | ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0, 0); | |
4190 | ztest_od_init(od + 1, id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, 0, | |
4191 | chunksize); | |
4192 | ||
4193 | ||
4194 | if (ztest_object_init(zd, od, size, B_FALSE) != 0) { | |
4195 | umem_free(od, size); | |
4196 | return; | |
4197 | } | |
4198 | ||
4199 | bigobj = od[0].od_object; | |
4200 | packobj = od[1].od_object; | |
4201 | blocksize = od[0].od_blocksize; | |
4202 | chunksize = blocksize; | |
4203 | ASSERT(chunksize == od[1].od_gen); | |
4204 | ||
4205 | VERIFY(dmu_object_info(os, bigobj, &doi) == 0); | |
4206 | VERIFY(ISP2(doi.doi_data_block_size)); | |
4207 | VERIFY(chunksize == doi.doi_data_block_size); | |
4208 | VERIFY(chunksize >= 2 * sizeof (bufwad_t)); | |
4209 | ||
4210 | /* | |
4211 | * Pick a random index and compute the offsets into packobj and bigobj. | |
4212 | */ | |
4213 | n = ztest_random(regions) * stride + ztest_random(width); | |
4214 | s = 1 + ztest_random(width - 1); | |
4215 | ||
4216 | packoff = n * sizeof (bufwad_t); | |
4217 | packsize = s * sizeof (bufwad_t); | |
4218 | ||
4219 | bigoff = n * chunksize; | |
4220 | bigsize = s * chunksize; | |
4221 | ||
4222 | packbuf = umem_zalloc(packsize, UMEM_NOFAIL); | |
4223 | bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL); | |
4224 | ||
4225 | VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db)); | |
4226 | ||
4227 | bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL); | |
4228 | ||
4229 | /* | |
4230 | * Iteration 0 test zcopy for DB_UNCACHED dbufs. | |
4231 | * Iteration 1 test zcopy to already referenced dbufs. | |
4232 | * Iteration 2 test zcopy to dirty dbuf in the same txg. | |
4233 | * Iteration 3 test zcopy to dbuf dirty in previous txg. | |
4234 | * Iteration 4 test zcopy when dbuf is no longer dirty. | |
4235 | * Iteration 5 test zcopy when it can't be done. | |
4236 | * Iteration 6 one more zcopy write. | |
4237 | */ | |
4238 | for (i = 0; i < 7; i++) { | |
4239 | uint64_t j; | |
4240 | uint64_t off; | |
4241 | ||
4242 | /* | |
4243 | * In iteration 5 (i == 5) use arcbufs | |
4244 | * that don't match bigobj blksz to test | |
4245 | * dmu_assign_arcbuf() when it can't directly | |
4246 | * assign an arcbuf to a dbuf. | |
4247 | */ | |
4248 | for (j = 0; j < s; j++) { | |
4249 | if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) { | |
4250 | bigbuf_arcbufs[j] = | |
4251 | dmu_request_arcbuf(bonus_db, chunksize); | |
4252 | } else { | |
4253 | bigbuf_arcbufs[2 * j] = | |
4254 | dmu_request_arcbuf(bonus_db, chunksize / 2); | |
4255 | bigbuf_arcbufs[2 * j + 1] = | |
4256 | dmu_request_arcbuf(bonus_db, chunksize / 2); | |
4257 | } | |
4258 | } | |
4259 | ||
4260 | /* | |
4261 | * Get a tx for the mods to both packobj and bigobj. | |
4262 | */ | |
4263 | tx = dmu_tx_create(os); | |
4264 | ||
4265 | dmu_tx_hold_write(tx, packobj, packoff, packsize); | |
4266 | dmu_tx_hold_write(tx, bigobj, bigoff, bigsize); | |
4267 | ||
4268 | txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); | |
4269 | if (txg == 0) { | |
4270 | umem_free(packbuf, packsize); | |
4271 | umem_free(bigbuf, bigsize); | |
4272 | for (j = 0; j < s; j++) { | |
4273 | if (i != 5 || | |
4274 | chunksize < (SPA_MINBLOCKSIZE * 2)) { | |
4275 | dmu_return_arcbuf(bigbuf_arcbufs[j]); | |
4276 | } else { | |
4277 | dmu_return_arcbuf( | |
4278 | bigbuf_arcbufs[2 * j]); | |
4279 | dmu_return_arcbuf( | |
4280 | bigbuf_arcbufs[2 * j + 1]); | |
4281 | } | |
4282 | } | |
4283 | umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *)); | |
4284 | umem_free(od, size); | |
4285 | dmu_buf_rele(bonus_db, FTAG); | |
4286 | return; | |
4287 | } | |
4288 | ||
4289 | /* | |
4290 | * 50% of the time don't read objects in the 1st iteration to | |
4291 | * test dmu_assign_arcbuf() for the case when there're no | |
4292 | * existing dbufs for the specified offsets. | |
4293 | */ | |
4294 | if (i != 0 || ztest_random(2) != 0) { | |
4295 | error = dmu_read(os, packobj, packoff, | |
4296 | packsize, packbuf, DMU_READ_PREFETCH); | |
4297 | ASSERT0(error); | |
4298 | error = dmu_read(os, bigobj, bigoff, bigsize, | |
4299 | bigbuf, DMU_READ_PREFETCH); | |
4300 | ASSERT0(error); | |
4301 | } | |
4302 | compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize, | |
4303 | n, chunksize, txg); | |
4304 | ||
4305 | /* | |
4306 | * We've verified all the old bufwads, and made new ones. | |
4307 | * Now write them out. | |
4308 | */ | |
4309 | dmu_write(os, packobj, packoff, packsize, packbuf, tx); | |
4310 | if (ztest_opts.zo_verbose >= 7) { | |
4311 | (void) printf("writing offset %llx size %llx" | |
4312 | " txg %llx\n", | |
4313 | (u_longlong_t)bigoff, | |
4314 | (u_longlong_t)bigsize, | |
4315 | (u_longlong_t)txg); | |
4316 | } | |
4317 | for (off = bigoff, j = 0; j < s; j++, off += chunksize) { | |
4318 | dmu_buf_t *dbt; | |
4319 | if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) { | |
4320 | bcopy((caddr_t)bigbuf + (off - bigoff), | |
4321 | bigbuf_arcbufs[j]->b_data, chunksize); | |
4322 | } else { | |
4323 | bcopy((caddr_t)bigbuf + (off - bigoff), | |
4324 | bigbuf_arcbufs[2 * j]->b_data, | |
4325 | chunksize / 2); | |
4326 | bcopy((caddr_t)bigbuf + (off - bigoff) + | |
4327 | chunksize / 2, | |
4328 | bigbuf_arcbufs[2 * j + 1]->b_data, | |
4329 | chunksize / 2); | |
4330 | } | |
4331 | ||
4332 | if (i == 1) { | |
4333 | VERIFY(dmu_buf_hold(os, bigobj, off, | |
4334 | FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0); | |
4335 | } | |
4336 | if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) { | |
4337 | dmu_assign_arcbuf(bonus_db, off, | |
4338 | bigbuf_arcbufs[j], tx); | |
4339 | } else { | |
4340 | dmu_assign_arcbuf(bonus_db, off, | |
4341 | bigbuf_arcbufs[2 * j], tx); | |
4342 | dmu_assign_arcbuf(bonus_db, | |
4343 | off + chunksize / 2, | |
4344 | bigbuf_arcbufs[2 * j + 1], tx); | |
4345 | } | |
4346 | if (i == 1) { | |
4347 | dmu_buf_rele(dbt, FTAG); | |
4348 | } | |
4349 | } | |
4350 | dmu_tx_commit(tx); | |
4351 | ||
4352 | /* | |
4353 | * Sanity check the stuff we just wrote. | |
4354 | */ | |
4355 | { | |
4356 | void *packcheck = umem_alloc(packsize, UMEM_NOFAIL); | |
4357 | void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL); | |
4358 | ||
4359 | VERIFY(0 == dmu_read(os, packobj, packoff, | |
4360 | packsize, packcheck, DMU_READ_PREFETCH)); | |
4361 | VERIFY(0 == dmu_read(os, bigobj, bigoff, | |
4362 | bigsize, bigcheck, DMU_READ_PREFETCH)); | |
4363 | ||
4364 | ASSERT(bcmp(packbuf, packcheck, packsize) == 0); | |
4365 | ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0); | |
4366 | ||
4367 | umem_free(packcheck, packsize); | |
4368 | umem_free(bigcheck, bigsize); | |
4369 | } | |
4370 | if (i == 2) { | |
4371 | txg_wait_open(dmu_objset_pool(os), 0); | |
4372 | } else if (i == 3) { | |
4373 | txg_wait_synced(dmu_objset_pool(os), 0); | |
4374 | } | |
4375 | } | |
4376 | ||
4377 | dmu_buf_rele(bonus_db, FTAG); | |
4378 | umem_free(packbuf, packsize); | |
4379 | umem_free(bigbuf, bigsize); | |
4380 | umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *)); | |
4381 | umem_free(od, size); | |
4382 | } | |
4383 | ||
4384 | /* ARGSUSED */ | |
4385 | void | |
4386 | ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id) | |
4387 | { | |
4388 | ztest_od_t *od; | |
4389 | ||
4390 | od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL); | |
4391 | uint64_t offset = (1ULL << (ztest_random(20) + 43)) + | |
4392 | (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT); | |
4393 | ||
4394 | /* | |
4395 | * Have multiple threads write to large offsets in an object | |
4396 | * to verify that parallel writes to an object -- even to the | |
4397 | * same blocks within the object -- doesn't cause any trouble. | |
4398 | */ | |
4399 | ztest_od_init(od, ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0); | |
4400 | ||
4401 | if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) | |
4402 | return; | |
4403 | ||
4404 | while (ztest_random(10) != 0) | |
4405 | ztest_io(zd, od->od_object, offset); | |
4406 | ||
4407 | umem_free(od, sizeof (ztest_od_t)); | |
4408 | } | |
4409 | ||
4410 | void | |
4411 | ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id) | |
4412 | { | |
4413 | ztest_od_t *od; | |
4414 | uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) + | |
4415 | (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT); | |
4416 | uint64_t count = ztest_random(20) + 1; | |
4417 | uint64_t blocksize = ztest_random_blocksize(); | |
4418 | void *data; | |
4419 | ||
4420 | od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL); | |
4421 | ||
4422 | ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0, 0); | |
4423 | ||
4424 | if (ztest_object_init(zd, od, sizeof (ztest_od_t), | |
4425 | !ztest_random(2)) != 0) { | |
4426 | umem_free(od, sizeof (ztest_od_t)); | |
4427 | return; | |
4428 | } | |
4429 | ||
4430 | if (ztest_truncate(zd, od->od_object, offset, count * blocksize) != 0) { | |
4431 | umem_free(od, sizeof (ztest_od_t)); | |
4432 | return; | |
4433 | } | |
4434 | ||
4435 | ztest_prealloc(zd, od->od_object, offset, count * blocksize); | |
4436 | ||
4437 | data = umem_zalloc(blocksize, UMEM_NOFAIL); | |
4438 | ||
4439 | while (ztest_random(count) != 0) { | |
4440 | uint64_t randoff = offset + (ztest_random(count) * blocksize); | |
4441 | if (ztest_write(zd, od->od_object, randoff, blocksize, | |
4442 | data) != 0) | |
4443 | break; | |
4444 | while (ztest_random(4) != 0) | |
4445 | ztest_io(zd, od->od_object, randoff); | |
4446 | } | |
4447 | ||
4448 | umem_free(data, blocksize); | |
4449 | umem_free(od, sizeof (ztest_od_t)); | |
4450 | } | |
4451 | ||
4452 | /* | |
4453 | * Verify that zap_{create,destroy,add,remove,update} work as expected. | |
4454 | */ | |
4455 | #define ZTEST_ZAP_MIN_INTS 1 | |
4456 | #define ZTEST_ZAP_MAX_INTS 4 | |
4457 | #define ZTEST_ZAP_MAX_PROPS 1000 | |
4458 | ||
4459 | void | |
4460 | ztest_zap(ztest_ds_t *zd, uint64_t id) | |
4461 | { | |
4462 | objset_t *os = zd->zd_os; | |
4463 | ztest_od_t *od; | |
4464 | uint64_t object; | |
4465 | uint64_t txg, last_txg; | |
4466 | uint64_t value[ZTEST_ZAP_MAX_INTS]; | |
4467 | uint64_t zl_ints, zl_intsize, prop; | |
4468 | int i, ints; | |
4469 | dmu_tx_t *tx; | |
4470 | char propname[100], txgname[100]; | |
4471 | int error; | |
4472 | char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" }; | |
4473 | ||
4474 | od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL); | |
4475 | ztest_od_init(od, id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0, 0); | |
4476 | ||
4477 | if (ztest_object_init(zd, od, sizeof (ztest_od_t), | |
4478 | !ztest_random(2)) != 0) | |
4479 | goto out; | |
4480 | ||
4481 | object = od->od_object; | |
4482 | ||
4483 | /* | |
4484 | * Generate a known hash collision, and verify that | |
4485 | * we can lookup and remove both entries. | |
4486 | */ | |
4487 | tx = dmu_tx_create(os); | |
4488 | dmu_tx_hold_zap(tx, object, B_TRUE, NULL); | |
4489 | txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); | |
4490 | if (txg == 0) | |
4491 | goto out; | |
4492 | for (i = 0; i < 2; i++) { | |
4493 | value[i] = i; | |
4494 | VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t), | |
4495 | 1, &value[i], tx)); | |
4496 | } | |
4497 | for (i = 0; i < 2; i++) { | |
4498 | VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i], | |
4499 | sizeof (uint64_t), 1, &value[i], tx)); | |
4500 | VERIFY3U(0, ==, | |
4501 | zap_length(os, object, hc[i], &zl_intsize, &zl_ints)); | |
4502 | ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); | |
4503 | ASSERT3U(zl_ints, ==, 1); | |
4504 | } | |
4505 | for (i = 0; i < 2; i++) { | |
4506 | VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx)); | |
4507 | } | |
4508 | dmu_tx_commit(tx); | |
4509 | ||
4510 | /* | |
4511 | * Generate a buch of random entries. | |
4512 | */ | |
4513 | ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS); | |
4514 | ||
4515 | prop = ztest_random(ZTEST_ZAP_MAX_PROPS); | |
4516 | (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop); | |
4517 | (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop); | |
4518 | bzero(value, sizeof (value)); | |
4519 | last_txg = 0; | |
4520 | ||
4521 | /* | |
4522 | * If these zap entries already exist, validate their contents. | |
4523 | */ | |
4524 | error = zap_length(os, object, txgname, &zl_intsize, &zl_ints); | |
4525 | if (error == 0) { | |
4526 | ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); | |
4527 | ASSERT3U(zl_ints, ==, 1); | |
4528 | ||
4529 | VERIFY(zap_lookup(os, object, txgname, zl_intsize, | |
4530 | zl_ints, &last_txg) == 0); | |
4531 | ||
4532 | VERIFY(zap_length(os, object, propname, &zl_intsize, | |
4533 | &zl_ints) == 0); | |
4534 | ||
4535 | ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); | |
4536 | ASSERT3U(zl_ints, ==, ints); | |
4537 | ||
4538 | VERIFY(zap_lookup(os, object, propname, zl_intsize, | |
4539 | zl_ints, value) == 0); | |
4540 | ||
4541 | for (i = 0; i < ints; i++) { | |
4542 | ASSERT3U(value[i], ==, last_txg + object + i); | |
4543 | } | |
4544 | } else { | |
4545 | ASSERT3U(error, ==, ENOENT); | |
4546 | } | |
4547 | ||
4548 | /* | |
4549 | * Atomically update two entries in our zap object. | |
4550 | * The first is named txg_%llu, and contains the txg | |
4551 | * in which the property was last updated. The second | |
4552 | * is named prop_%llu, and the nth element of its value | |
4553 | * should be txg + object + n. | |
4554 | */ | |
4555 | tx = dmu_tx_create(os); | |
4556 | dmu_tx_hold_zap(tx, object, B_TRUE, NULL); | |
4557 | txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); | |
4558 | if (txg == 0) | |
4559 | goto out; | |
4560 | ||
4561 | if (last_txg > txg) | |
4562 | fatal(0, "zap future leak: old %llu new %llu", last_txg, txg); | |
4563 | ||
4564 | for (i = 0; i < ints; i++) | |
4565 | value[i] = txg + object + i; | |
4566 | ||
4567 | VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t), | |
4568 | 1, &txg, tx)); | |
4569 | VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t), | |
4570 | ints, value, tx)); | |
4571 | ||
4572 | dmu_tx_commit(tx); | |
4573 | ||
4574 | /* | |
4575 | * Remove a random pair of entries. | |
4576 | */ | |
4577 | prop = ztest_random(ZTEST_ZAP_MAX_PROPS); | |
4578 | (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop); | |
4579 | (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop); | |
4580 | ||
4581 | error = zap_length(os, object, txgname, &zl_intsize, &zl_ints); | |
4582 | ||
4583 | if (error == ENOENT) | |
4584 | goto out; | |
4585 | ||
4586 | ASSERT0(error); | |
4587 | ||
4588 | tx = dmu_tx_create(os); | |
4589 | dmu_tx_hold_zap(tx, object, B_TRUE, NULL); | |
4590 | txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); | |
4591 | if (txg == 0) | |
4592 | goto out; | |
4593 | VERIFY3U(0, ==, zap_remove(os, object, txgname, tx)); | |
4594 | VERIFY3U(0, ==, zap_remove(os, object, propname, tx)); | |
4595 | dmu_tx_commit(tx); | |
4596 | out: | |
4597 | umem_free(od, sizeof (ztest_od_t)); | |
4598 | } | |
4599 | ||
4600 | /* | |
4601 | * Testcase to test the upgrading of a microzap to fatzap. | |
4602 | */ | |
4603 | void | |
4604 | ztest_fzap(ztest_ds_t *zd, uint64_t id) | |
4605 | { | |
4606 | objset_t *os = zd->zd_os; | |
4607 | ztest_od_t *od; | |
4608 | uint64_t object, txg; | |
4609 | int i; | |
4610 | ||
4611 | od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL); | |
4612 | ztest_od_init(od, id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0, 0); | |
4613 | ||
4614 | if (ztest_object_init(zd, od, sizeof (ztest_od_t), | |
4615 | !ztest_random(2)) != 0) | |
4616 | goto out; | |
4617 | object = od->od_object; | |
4618 | ||
4619 | /* | |
4620 | * Add entries to this ZAP and make sure it spills over | |
4621 | * and gets upgraded to a fatzap. Also, since we are adding | |
4622 | * 2050 entries we should see ptrtbl growth and leaf-block split. | |
4623 | */ | |
4624 | for (i = 0; i < 2050; i++) { | |
4625 | char name[MAXNAMELEN]; | |
4626 | uint64_t value = i; | |
4627 | dmu_tx_t *tx; | |
4628 | int error; | |
4629 | ||
4630 | (void) snprintf(name, sizeof (name), "fzap-%llu-%llu", | |
4631 | (u_longlong_t)id, (u_longlong_t)value); | |
4632 | ||
4633 | tx = dmu_tx_create(os); | |
4634 | dmu_tx_hold_zap(tx, object, B_TRUE, name); | |
4635 | txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); | |
4636 | if (txg == 0) | |
4637 | goto out; | |
4638 | error = zap_add(os, object, name, sizeof (uint64_t), 1, | |
4639 | &value, tx); | |
4640 | ASSERT(error == 0 || error == EEXIST); | |
4641 | dmu_tx_commit(tx); | |
4642 | } | |
4643 | out: | |
4644 | umem_free(od, sizeof (ztest_od_t)); | |
4645 | } | |
4646 | ||
4647 | /* ARGSUSED */ | |
4648 | void | |
4649 | ztest_zap_parallel(ztest_ds_t *zd, uint64_t id) | |
4650 | { | |
4651 | objset_t *os = zd->zd_os; | |
4652 | ztest_od_t *od; | |
4653 | uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc; | |
4654 | dmu_tx_t *tx; | |
4655 | int i, namelen, error; | |
4656 | int micro = ztest_random(2); | |
4657 | char name[20], string_value[20]; | |
4658 | void *data; | |
4659 | ||
4660 | od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL); | |
4661 | ztest_od_init(od, ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0, 0); | |
4662 | ||
4663 | if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) { | |
4664 | umem_free(od, sizeof (ztest_od_t)); | |
4665 | return; | |
4666 | } | |
4667 | ||
4668 | object = od->od_object; | |
4669 | ||
4670 | /* | |
4671 | * Generate a random name of the form 'xxx.....' where each | |
4672 | * x is a random printable character and the dots are dots. | |
4673 | * There are 94 such characters, and the name length goes from | |
4674 | * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names. | |
4675 | */ | |
4676 | namelen = ztest_random(sizeof (name) - 5) + 5 + 1; | |
4677 | ||
4678 | for (i = 0; i < 3; i++) | |
4679 | name[i] = '!' + ztest_random('~' - '!' + 1); | |
4680 | for (; i < namelen - 1; i++) | |
4681 | name[i] = '.'; | |
4682 | name[i] = '\0'; | |
4683 | ||
4684 | if ((namelen & 1) || micro) { | |
4685 | wsize = sizeof (txg); | |
4686 | wc = 1; | |
4687 | data = &txg; | |
4688 | } else { | |
4689 | wsize = 1; | |
4690 | wc = namelen; | |
4691 | data = string_value; | |
4692 | } | |
4693 | ||
4694 | count = -1ULL; | |
4695 | VERIFY0(zap_count(os, object, &count)); | |
4696 | ASSERT(count != -1ULL); | |
4697 | ||
4698 | /* | |
4699 | * Select an operation: length, lookup, add, update, remove. | |
4700 | */ | |
4701 | i = ztest_random(5); | |
4702 | ||
4703 | if (i >= 2) { | |
4704 | tx = dmu_tx_create(os); | |
4705 | dmu_tx_hold_zap(tx, object, B_TRUE, NULL); | |
4706 | txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); | |
4707 | if (txg == 0) | |
4708 | return; | |
4709 | bcopy(name, string_value, namelen); | |
4710 | } else { | |
4711 | tx = NULL; | |
4712 | txg = 0; | |
4713 | bzero(string_value, namelen); | |
4714 | } | |
4715 | ||
4716 | switch (i) { | |
4717 | ||
4718 | case 0: | |
4719 | error = zap_length(os, object, name, &zl_wsize, &zl_wc); | |
4720 | if (error == 0) { | |
4721 | ASSERT3U(wsize, ==, zl_wsize); | |
4722 | ASSERT3U(wc, ==, zl_wc); | |
4723 | } else { | |
4724 | ASSERT3U(error, ==, ENOENT); | |
4725 | } | |
4726 | break; | |
4727 | ||
4728 | case 1: | |
4729 | error = zap_lookup(os, object, name, wsize, wc, data); | |
4730 | if (error == 0) { | |
4731 | if (data == string_value && | |
4732 | bcmp(name, data, namelen) != 0) | |
4733 | fatal(0, "name '%s' != val '%s' len %d", | |
4734 | name, data, namelen); | |
4735 | } else { | |
4736 | ASSERT3U(error, ==, ENOENT); | |
4737 | } | |
4738 | break; | |
4739 | ||
4740 | case 2: | |
4741 | error = zap_add(os, object, name, wsize, wc, data, tx); | |
4742 | ASSERT(error == 0 || error == EEXIST); | |
4743 | break; | |
4744 | ||
4745 | case 3: | |
4746 | VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0); | |
4747 | break; | |
4748 | ||
4749 | case 4: | |
4750 | error = zap_remove(os, object, name, tx); | |
4751 | ASSERT(error == 0 || error == ENOENT); | |
4752 | break; | |
4753 | } | |
4754 | ||
4755 | if (tx != NULL) | |
4756 | dmu_tx_commit(tx); | |
4757 | ||
4758 | umem_free(od, sizeof (ztest_od_t)); | |
4759 | } | |
4760 | ||
4761 | /* | |
4762 | * Commit callback data. | |
4763 | */ | |
4764 | typedef struct ztest_cb_data { | |
4765 | list_node_t zcd_node; | |
4766 | uint64_t zcd_txg; | |
4767 | int zcd_expected_err; | |
4768 | boolean_t zcd_added; | |
4769 | boolean_t zcd_called; | |
4770 | spa_t *zcd_spa; | |
4771 | } ztest_cb_data_t; | |
4772 | ||
4773 | /* This is the actual commit callback function */ | |
4774 | static void | |
4775 | ztest_commit_callback(void *arg, int error) | |
4776 | { | |
4777 | ztest_cb_data_t *data = arg; | |
4778 | uint64_t synced_txg; | |
4779 | ||
4780 | VERIFY(data != NULL); | |
4781 | VERIFY3S(data->zcd_expected_err, ==, error); | |
4782 | VERIFY(!data->zcd_called); | |
4783 | ||
4784 | synced_txg = spa_last_synced_txg(data->zcd_spa); | |
4785 | if (data->zcd_txg > synced_txg) | |
4786 | fatal(0, "commit callback of txg %" PRIu64 " called prematurely" | |
4787 | ", last synced txg = %" PRIu64 "\n", data->zcd_txg, | |
4788 | synced_txg); | |
4789 | ||
4790 | data->zcd_called = B_TRUE; | |
4791 | ||
4792 | if (error == ECANCELED) { | |
4793 | ASSERT0(data->zcd_txg); | |
4794 | ASSERT(!data->zcd_added); | |
4795 | ||
4796 | /* | |
4797 | * The private callback data should be destroyed here, but | |
4798 | * since we are going to check the zcd_called field after | |
4799 | * dmu_tx_abort(), we will destroy it there. | |
4800 | */ | |
4801 | return; | |
4802 | } | |
4803 | ||
4804 | ASSERT(data->zcd_added); | |
4805 | ASSERT3U(data->zcd_txg, !=, 0); | |
4806 | ||
4807 | (void) mutex_enter(&zcl.zcl_callbacks_lock); | |
4808 | ||
4809 | /* See if this cb was called more quickly */ | |
4810 | if ((synced_txg - data->zcd_txg) < zc_min_txg_delay) | |
4811 | zc_min_txg_delay = synced_txg - data->zcd_txg; | |
4812 | ||
4813 | /* Remove our callback from the list */ | |
4814 | list_remove(&zcl.zcl_callbacks, data); | |
4815 | ||
4816 | (void) mutex_exit(&zcl.zcl_callbacks_lock); | |
4817 | ||
4818 | umem_free(data, sizeof (ztest_cb_data_t)); | |
4819 | } | |
4820 | ||
4821 | /* Allocate and initialize callback data structure */ | |
4822 | static ztest_cb_data_t * | |
4823 | ztest_create_cb_data(objset_t *os, uint64_t txg) | |
4824 | { | |
4825 | ztest_cb_data_t *cb_data; | |
4826 | ||
4827 | cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL); | |
4828 | ||
4829 | cb_data->zcd_txg = txg; | |
4830 | cb_data->zcd_spa = dmu_objset_spa(os); | |
4831 | list_link_init(&cb_data->zcd_node); | |
4832 | ||
4833 | return (cb_data); | |
4834 | } | |
4835 | ||
4836 | /* | |
4837 | * Commit callback test. | |
4838 | */ | |
4839 | void | |
4840 | ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id) | |
4841 | { | |
4842 | objset_t *os = zd->zd_os; | |
4843 | ztest_od_t *od; | |
4844 | dmu_tx_t *tx; | |
4845 | ztest_cb_data_t *cb_data[3], *tmp_cb; | |
4846 | uint64_t old_txg, txg; | |
4847 | int i, error = 0; | |
4848 | ||
4849 | od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL); | |
4850 | ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0); | |
4851 | ||
4852 | if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) { | |
4853 | umem_free(od, sizeof (ztest_od_t)); | |
4854 | return; | |
4855 | } | |
4856 | ||
4857 | tx = dmu_tx_create(os); | |
4858 | ||
4859 | cb_data[0] = ztest_create_cb_data(os, 0); | |
4860 | dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]); | |
4861 | ||
4862 | dmu_tx_hold_write(tx, od->od_object, 0, sizeof (uint64_t)); | |
4863 | ||
4864 | /* Every once in a while, abort the transaction on purpose */ | |
4865 | if (ztest_random(100) == 0) | |
4866 | error = -1; | |
4867 | ||
4868 | if (!error) | |
4869 | error = dmu_tx_assign(tx, TXG_NOWAIT); | |
4870 | ||
4871 | txg = error ? 0 : dmu_tx_get_txg(tx); | |
4872 | ||
4873 | cb_data[0]->zcd_txg = txg; | |
4874 | cb_data[1] = ztest_create_cb_data(os, txg); | |
4875 | dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]); | |
4876 | ||
4877 | if (error) { | |
4878 | /* | |
4879 | * It's not a strict requirement to call the registered | |
4880 | * callbacks from inside dmu_tx_abort(), but that's what | |
4881 | * it's supposed to happen in the current implementation | |
4882 | * so we will check for that. | |
4883 | */ | |
4884 | for (i = 0; i < 2; i++) { | |
4885 | cb_data[i]->zcd_expected_err = ECANCELED; | |
4886 | VERIFY(!cb_data[i]->zcd_called); | |
4887 | } | |
4888 | ||
4889 | dmu_tx_abort(tx); | |
4890 | ||
4891 | for (i = 0; i < 2; i++) { | |
4892 | VERIFY(cb_data[i]->zcd_called); | |
4893 | umem_free(cb_data[i], sizeof (ztest_cb_data_t)); | |
4894 | } | |
4895 | ||
4896 | umem_free(od, sizeof (ztest_od_t)); | |
4897 | return; | |
4898 | } | |
4899 | ||
4900 | cb_data[2] = ztest_create_cb_data(os, txg); | |
4901 | dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]); | |
4902 | ||
4903 | /* | |
4904 | * Read existing data to make sure there isn't a future leak. | |
4905 | */ | |
4906 | VERIFY(0 == dmu_read(os, od->od_object, 0, sizeof (uint64_t), | |
4907 | &old_txg, DMU_READ_PREFETCH)); | |
4908 | ||
4909 | if (old_txg > txg) | |
4910 | fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64, | |
4911 | old_txg, txg); | |
4912 | ||
4913 | dmu_write(os, od->od_object, 0, sizeof (uint64_t), &txg, tx); | |
4914 | ||
4915 | (void) mutex_enter(&zcl.zcl_callbacks_lock); | |
4916 | ||
4917 | /* | |
4918 | * Since commit callbacks don't have any ordering requirement and since | |
4919 | * it is theoretically possible for a commit callback to be called | |
4920 | * after an arbitrary amount of time has elapsed since its txg has been | |
4921 | * synced, it is difficult to reliably determine whether a commit | |
4922 | * callback hasn't been called due to high load or due to a flawed | |
4923 | * implementation. | |
4924 | * | |
4925 | * In practice, we will assume that if after a certain number of txgs a | |
4926 | * commit callback hasn't been called, then most likely there's an | |
4927 | * implementation bug.. | |
4928 | */ | |
4929 | tmp_cb = list_head(&zcl.zcl_callbacks); | |
4930 | if (tmp_cb != NULL && | |
4931 | tmp_cb->zcd_txg + ZTEST_COMMIT_CB_THRESH < txg) { | |
4932 | fatal(0, "Commit callback threshold exceeded, oldest txg: %" | |
4933 | PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg); | |
4934 | } | |
4935 | ||
4936 | /* | |
4937 | * Let's find the place to insert our callbacks. | |
4938 | * | |
4939 | * Even though the list is ordered by txg, it is possible for the | |
4940 | * insertion point to not be the end because our txg may already be | |
4941 | * quiescing at this point and other callbacks in the open txg | |
4942 | * (from other objsets) may have sneaked in. | |
4943 | */ | |
4944 | tmp_cb = list_tail(&zcl.zcl_callbacks); | |
4945 | while (tmp_cb != NULL && tmp_cb->zcd_txg > txg) | |
4946 | tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb); | |
4947 | ||
4948 | /* Add the 3 callbacks to the list */ | |
4949 | for (i = 0; i < 3; i++) { | |
4950 | if (tmp_cb == NULL) | |
4951 | list_insert_head(&zcl.zcl_callbacks, cb_data[i]); | |
4952 | else | |
4953 | list_insert_after(&zcl.zcl_callbacks, tmp_cb, | |
4954 | cb_data[i]); | |
4955 | ||
4956 | cb_data[i]->zcd_added = B_TRUE; | |
4957 | VERIFY(!cb_data[i]->zcd_called); | |
4958 | ||
4959 | tmp_cb = cb_data[i]; | |
4960 | } | |
4961 | ||
4962 | zc_cb_counter += 3; | |
4963 | ||
4964 | (void) mutex_exit(&zcl.zcl_callbacks_lock); | |
4965 | ||
4966 | dmu_tx_commit(tx); | |
4967 | ||
4968 | umem_free(od, sizeof (ztest_od_t)); | |
4969 | } | |
4970 | ||
4971 | /* | |
4972 | * Visit each object in the dataset. Verify that its properties | |
4973 | * are consistent what was stored in the block tag when it was created, | |
4974 | * and that its unused bonus buffer space has not been overwritten. | |
4975 | */ | |
4976 | void | |
4977 | ztest_verify_dnode_bt(ztest_ds_t *zd, uint64_t id) | |
4978 | { | |
4979 | objset_t *os = zd->zd_os; | |
4980 | uint64_t obj; | |
4981 | int err = 0; | |
4982 | ||
4983 | for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) { | |
4984 | ztest_block_tag_t *bt = NULL; | |
4985 | dmu_object_info_t doi; | |
4986 | dmu_buf_t *db; | |
4987 | ||
4988 | if (dmu_bonus_hold(os, obj, FTAG, &db) != 0) | |
4989 | continue; | |
4990 | ||
4991 | dmu_object_info_from_db(db, &doi); | |
4992 | if (doi.doi_bonus_size >= sizeof (*bt)) | |
4993 | bt = ztest_bt_bonus(db); | |
4994 | ||
4995 | if (bt && bt->bt_magic == BT_MAGIC) { | |
4996 | ztest_bt_verify(bt, os, obj, doi.doi_dnodesize, | |
4997 | bt->bt_offset, bt->bt_gen, bt->bt_txg, | |
4998 | bt->bt_crtxg); | |
4999 | ztest_verify_unused_bonus(db, bt, obj, os, bt->bt_gen); | |
5000 | } | |
5001 | ||
5002 | dmu_buf_rele(db, FTAG); | |
5003 | } | |
5004 | } | |
5005 | ||
5006 | /* ARGSUSED */ | |
5007 | void | |
5008 | ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id) | |
5009 | { | |
5010 | zfs_prop_t proplist[] = { | |
5011 | ZFS_PROP_CHECKSUM, | |
5012 | ZFS_PROP_COMPRESSION, | |
5013 | ZFS_PROP_COPIES, | |
5014 | ZFS_PROP_DEDUP | |
5015 | }; | |
5016 | int p; | |
5017 | ||
5018 | (void) rw_rdlock(&ztest_name_lock); | |
5019 | ||
5020 | for (p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++) | |
5021 | (void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p], | |
5022 | ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2)); | |
5023 | ||
5024 | VERIFY0(ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_RECORDSIZE, | |
5025 | ztest_random_blocksize(), (int)ztest_random(2))); | |
5026 | ||
5027 | (void) rw_unlock(&ztest_name_lock); | |
5028 | } | |
5029 | ||
5030 | /* ARGSUSED */ | |
5031 | void | |
5032 | ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id) | |
5033 | { | |
5034 | nvlist_t *props = NULL; | |
5035 | ||
5036 | (void) rw_rdlock(&ztest_name_lock); | |
5037 | ||
5038 | (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO, | |
5039 | ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN)); | |
5040 | ||
5041 | VERIFY0(spa_prop_get(ztest_spa, &props)); | |
5042 | ||
5043 | if (ztest_opts.zo_verbose >= 6) | |
5044 | dump_nvlist(props, 4); | |
5045 | ||
5046 | nvlist_free(props); | |
5047 | ||
5048 | (void) rw_unlock(&ztest_name_lock); | |
5049 | } | |
5050 | ||
5051 | static int | |
5052 | user_release_one(const char *snapname, const char *holdname) | |
5053 | { | |
5054 | nvlist_t *snaps, *holds; | |
5055 | int error; | |
5056 | ||
5057 | snaps = fnvlist_alloc(); | |
5058 | holds = fnvlist_alloc(); | |
5059 | fnvlist_add_boolean(holds, holdname); | |
5060 | fnvlist_add_nvlist(snaps, snapname, holds); | |
5061 | fnvlist_free(holds); | |
5062 | error = dsl_dataset_user_release(snaps, NULL); | |
5063 | fnvlist_free(snaps); | |
5064 | return (error); | |
5065 | } | |
5066 | ||
5067 | /* | |
5068 | * Test snapshot hold/release and deferred destroy. | |
5069 | */ | |
5070 | void | |
5071 | ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id) | |
5072 | { | |
5073 | int error; | |
5074 | objset_t *os = zd->zd_os; | |
5075 | objset_t *origin; | |
5076 | char snapname[100]; | |
5077 | char fullname[100]; | |
5078 | char clonename[100]; | |
5079 | char tag[100]; | |
5080 | char osname[MAXNAMELEN]; | |
5081 | nvlist_t *holds; | |
5082 | ||
5083 | (void) rw_rdlock(&ztest_name_lock); | |
5084 | ||
5085 | dmu_objset_name(os, osname); | |
5086 | ||
5087 | (void) snprintf(snapname, sizeof (snapname), "sh1_%llu", | |
5088 | (u_longlong_t)id); | |
5089 | (void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname); | |
5090 | (void) snprintf(clonename, sizeof (clonename), | |
5091 | "%s/ch1_%llu", osname, (u_longlong_t)id); | |
5092 | (void) snprintf(tag, sizeof (tag), "tag_%llu", (u_longlong_t)id); | |
5093 | ||
5094 | /* | |
5095 | * Clean up from any previous run. | |
5096 | */ | |
5097 | error = dsl_destroy_head(clonename); | |
5098 | if (error != ENOENT) | |
5099 | ASSERT0(error); | |
5100 | error = user_release_one(fullname, tag); | |
5101 | if (error != ESRCH && error != ENOENT) | |
5102 | ASSERT0(error); | |
5103 | error = dsl_destroy_snapshot(fullname, B_FALSE); | |
5104 | if (error != ENOENT) | |
5105 | ASSERT0(error); | |
5106 | ||
5107 | /* | |
5108 | * Create snapshot, clone it, mark snap for deferred destroy, | |
5109 | * destroy clone, verify snap was also destroyed. | |
5110 | */ | |
5111 | error = dmu_objset_snapshot_one(osname, snapname); | |
5112 | if (error) { | |
5113 | if (error == ENOSPC) { | |
5114 | ztest_record_enospc("dmu_objset_snapshot"); | |
5115 | goto out; | |
5116 | } | |
5117 | fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error); | |
5118 | } | |
5119 | ||
5120 | error = dmu_objset_clone(clonename, fullname); | |
5121 | if (error) { | |
5122 | if (error == ENOSPC) { | |
5123 | ztest_record_enospc("dmu_objset_clone"); | |
5124 | goto out; | |
5125 | } | |
5126 | fatal(0, "dmu_objset_clone(%s) = %d", clonename, error); | |
5127 | } | |
5128 | ||
5129 | error = dsl_destroy_snapshot(fullname, B_TRUE); | |
5130 | if (error) { | |
5131 | fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d", | |
5132 | fullname, error); | |
5133 | } | |
5134 | ||
5135 | error = dsl_destroy_head(clonename); | |
5136 | if (error) | |
5137 | fatal(0, "dsl_destroy_head(%s) = %d", clonename, error); | |
5138 | ||
5139 | error = dmu_objset_hold(fullname, FTAG, &origin); | |
5140 | if (error != ENOENT) | |
5141 | fatal(0, "dmu_objset_hold(%s) = %d", fullname, error); | |
5142 | ||
5143 | /* | |
5144 | * Create snapshot, add temporary hold, verify that we can't | |
5145 | * destroy a held snapshot, mark for deferred destroy, | |
5146 | * release hold, verify snapshot was destroyed. | |
5147 | */ | |
5148 | error = dmu_objset_snapshot_one(osname, snapname); | |
5149 | if (error) { | |
5150 | if (error == ENOSPC) { | |
5151 | ztest_record_enospc("dmu_objset_snapshot"); | |
5152 | goto out; | |
5153 | } | |
5154 | fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error); | |
5155 | } | |
5156 | ||
5157 | holds = fnvlist_alloc(); | |
5158 | fnvlist_add_string(holds, fullname, tag); | |
5159 | error = dsl_dataset_user_hold(holds, 0, NULL); | |
5160 | fnvlist_free(holds); | |
5161 | ||
5162 | if (error == ENOSPC) { | |
5163 | ztest_record_enospc("dsl_dataset_user_hold"); | |
5164 | goto out; | |
5165 | } else if (error) { | |
5166 | fatal(0, "dsl_dataset_user_hold(%s, %s) = %u", | |
5167 | fullname, tag, error); | |
5168 | } | |
5169 | ||
5170 | error = dsl_destroy_snapshot(fullname, B_FALSE); | |
5171 | if (error != EBUSY) { | |
5172 | fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d", | |
5173 | fullname, error); | |
5174 | } | |
5175 | ||
5176 | error = dsl_destroy_snapshot(fullname, B_TRUE); | |
5177 | if (error) { | |
5178 | fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d", | |
5179 | fullname, error); | |
5180 | } | |
5181 | ||
5182 | error = user_release_one(fullname, tag); | |
5183 | if (error) | |
5184 | fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error); | |
5185 | ||
5186 | VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT); | |
5187 | ||
5188 | out: | |
5189 | (void) rw_unlock(&ztest_name_lock); | |
5190 | } | |
5191 | ||
5192 | /* | |
5193 | * Inject random faults into the on-disk data. | |
5194 | */ | |
5195 | /* ARGSUSED */ | |
5196 | void | |
5197 | ztest_fault_inject(ztest_ds_t *zd, uint64_t id) | |
5198 | { | |
5199 | ztest_shared_t *zs = ztest_shared; | |
5200 | spa_t *spa = ztest_spa; | |
5201 | int fd; | |
5202 | uint64_t offset; | |
5203 | uint64_t leaves; | |
5204 | uint64_t bad = 0x1990c0ffeedecadeull; | |
5205 | uint64_t top, leaf; | |
5206 | char *path0; | |
5207 | char *pathrand; | |
5208 | size_t fsize; | |
5209 | int bshift = SPA_MAXBLOCKSHIFT + 2; /* don't scrog all labels */ | |
5210 | int iters = 1000; | |
5211 | int maxfaults; | |
5212 | int mirror_save; | |
5213 | vdev_t *vd0 = NULL; | |
5214 | uint64_t guid0 = 0; | |
5215 | boolean_t islog = B_FALSE; | |
5216 | ||
5217 | path0 = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); | |
5218 | pathrand = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); | |
5219 | ||
5220 | mutex_enter(&ztest_vdev_lock); | |
5221 | maxfaults = MAXFAULTS(); | |
5222 | leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz; | |
5223 | mirror_save = zs->zs_mirrors; | |
5224 | mutex_exit(&ztest_vdev_lock); | |
5225 | ||
5226 | ASSERT(leaves >= 1); | |
5227 | ||
5228 | /* | |
5229 | * Grab the name lock as reader. There are some operations | |
5230 | * which don't like to have their vdevs changed while | |
5231 | * they are in progress (i.e. spa_change_guid). Those | |
5232 | * operations will have grabbed the name lock as writer. | |
5233 | */ | |
5234 | (void) rw_rdlock(&ztest_name_lock); | |
5235 | ||
5236 | /* | |
5237 | * We need SCL_STATE here because we're going to look at vd0->vdev_tsd. | |
5238 | */ | |
5239 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); | |
5240 | ||
5241 | if (ztest_random(2) == 0) { | |
5242 | /* | |
5243 | * Inject errors on a normal data device or slog device. | |
5244 | */ | |
5245 | top = ztest_random_vdev_top(spa, B_TRUE); | |
5246 | leaf = ztest_random(leaves) + zs->zs_splits; | |
5247 | ||
5248 | /* | |
5249 | * Generate paths to the first leaf in this top-level vdev, | |
5250 | * and to the random leaf we selected. We'll induce transient | |
5251 | * write failures and random online/offline activity on leaf 0, | |
5252 | * and we'll write random garbage to the randomly chosen leaf. | |
5253 | */ | |
5254 | (void) snprintf(path0, MAXPATHLEN, ztest_dev_template, | |
5255 | ztest_opts.zo_dir, ztest_opts.zo_pool, | |
5256 | top * leaves + zs->zs_splits); | |
5257 | (void) snprintf(pathrand, MAXPATHLEN, ztest_dev_template, | |
5258 | ztest_opts.zo_dir, ztest_opts.zo_pool, | |
5259 | top * leaves + leaf); | |
5260 | ||
5261 | vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0); | |
5262 | if (vd0 != NULL && vd0->vdev_top->vdev_islog) | |
5263 | islog = B_TRUE; | |
5264 | ||
5265 | /* | |
5266 | * If the top-level vdev needs to be resilvered | |
5267 | * then we only allow faults on the device that is | |
5268 | * resilvering. | |
5269 | */ | |
5270 | if (vd0 != NULL && maxfaults != 1 && | |
5271 | (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) || | |
5272 | vd0->vdev_resilver_txg != 0)) { | |
5273 | /* | |
5274 | * Make vd0 explicitly claim to be unreadable, | |
5275 | * or unwriteable, or reach behind its back | |
5276 | * and close the underlying fd. We can do this if | |
5277 | * maxfaults == 0 because we'll fail and reexecute, | |
5278 | * and we can do it if maxfaults >= 2 because we'll | |
5279 | * have enough redundancy. If maxfaults == 1, the | |
5280 | * combination of this with injection of random data | |
5281 | * corruption below exceeds the pool's fault tolerance. | |
5282 | */ | |
5283 | vdev_file_t *vf = vd0->vdev_tsd; | |
5284 | ||
5285 | if (vf != NULL && ztest_random(3) == 0) { | |
5286 | (void) close(vf->vf_vnode->v_fd); | |
5287 | vf->vf_vnode->v_fd = -1; | |
5288 | } else if (ztest_random(2) == 0) { | |
5289 | vd0->vdev_cant_read = B_TRUE; | |
5290 | } else { | |
5291 | vd0->vdev_cant_write = B_TRUE; | |
5292 | } | |
5293 | guid0 = vd0->vdev_guid; | |
5294 | } | |
5295 | } else { | |
5296 | /* | |
5297 | * Inject errors on an l2cache device. | |
5298 | */ | |
5299 | spa_aux_vdev_t *sav = &spa->spa_l2cache; | |
5300 | ||
5301 | if (sav->sav_count == 0) { | |
5302 | spa_config_exit(spa, SCL_STATE, FTAG); | |
5303 | (void) rw_unlock(&ztest_name_lock); | |
5304 | goto out; | |
5305 | } | |
5306 | vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)]; | |
5307 | guid0 = vd0->vdev_guid; | |
5308 | (void) strcpy(path0, vd0->vdev_path); | |
5309 | (void) strcpy(pathrand, vd0->vdev_path); | |
5310 | ||
5311 | leaf = 0; | |
5312 | leaves = 1; | |
5313 | maxfaults = INT_MAX; /* no limit on cache devices */ | |
5314 | } | |
5315 | ||
5316 | spa_config_exit(spa, SCL_STATE, FTAG); | |
5317 | (void) rw_unlock(&ztest_name_lock); | |
5318 | ||
5319 | /* | |
5320 | * If we can tolerate two or more faults, or we're dealing | |
5321 | * with a slog, randomly online/offline vd0. | |
5322 | */ | |
5323 | if ((maxfaults >= 2 || islog) && guid0 != 0) { | |
5324 | if (ztest_random(10) < 6) { | |
5325 | int flags = (ztest_random(2) == 0 ? | |
5326 | ZFS_OFFLINE_TEMPORARY : 0); | |
5327 | ||
5328 | /* | |
5329 | * We have to grab the zs_name_lock as writer to | |
5330 | * prevent a race between offlining a slog and | |
5331 | * destroying a dataset. Offlining the slog will | |
5332 | * grab a reference on the dataset which may cause | |
5333 | * dsl_destroy_head() to fail with EBUSY thus | |
5334 | * leaving the dataset in an inconsistent state. | |
5335 | */ | |
5336 | if (islog) | |
5337 | (void) rw_wrlock(&ztest_name_lock); | |
5338 | ||
5339 | VERIFY(vdev_offline(spa, guid0, flags) != EBUSY); | |
5340 | ||
5341 | if (islog) | |
5342 | (void) rw_unlock(&ztest_name_lock); | |
5343 | } else { | |
5344 | /* | |
5345 | * Ideally we would like to be able to randomly | |
5346 | * call vdev_[on|off]line without holding locks | |
5347 | * to force unpredictable failures but the side | |
5348 | * effects of vdev_[on|off]line prevent us from | |
5349 | * doing so. We grab the ztest_vdev_lock here to | |
5350 | * prevent a race between injection testing and | |
5351 | * aux_vdev removal. | |
5352 | */ | |
5353 | mutex_enter(&ztest_vdev_lock); | |
5354 | (void) vdev_online(spa, guid0, 0, NULL); | |
5355 | mutex_exit(&ztest_vdev_lock); | |
5356 | } | |
5357 | } | |
5358 | ||
5359 | if (maxfaults == 0) | |
5360 | goto out; | |
5361 | ||
5362 | /* | |
5363 | * We have at least single-fault tolerance, so inject data corruption. | |
5364 | */ | |
5365 | fd = open(pathrand, O_RDWR); | |
5366 | ||
5367 | if (fd == -1) /* we hit a gap in the device namespace */ | |
5368 | goto out; | |
5369 | ||
5370 | fsize = lseek(fd, 0, SEEK_END); | |
5371 | ||
5372 | while (--iters != 0) { | |
5373 | /* | |
5374 | * The offset must be chosen carefully to ensure that | |
5375 | * we do not inject a given logical block with errors | |
5376 | * on two different leaf devices, because ZFS can not | |
5377 | * tolerate that (if maxfaults==1). | |
5378 | * | |
5379 | * We divide each leaf into chunks of size | |
5380 | * (# leaves * SPA_MAXBLOCKSIZE * 4). Within each chunk | |
5381 | * there is a series of ranges to which we can inject errors. | |
5382 | * Each range can accept errors on only a single leaf vdev. | |
5383 | * The error injection ranges are separated by ranges | |
5384 | * which we will not inject errors on any device (DMZs). | |
5385 | * Each DMZ must be large enough such that a single block | |
5386 | * can not straddle it, so that a single block can not be | |
5387 | * a target in two different injection ranges (on different | |
5388 | * leaf vdevs). | |
5389 | * | |
5390 | * For example, with 3 leaves, each chunk looks like: | |
5391 | * 0 to 32M: injection range for leaf 0 | |
5392 | * 32M to 64M: DMZ - no injection allowed | |
5393 | * 64M to 96M: injection range for leaf 1 | |
5394 | * 96M to 128M: DMZ - no injection allowed | |
5395 | * 128M to 160M: injection range for leaf 2 | |
5396 | * 160M to 192M: DMZ - no injection allowed | |
5397 | */ | |
5398 | offset = ztest_random(fsize / (leaves << bshift)) * | |
5399 | (leaves << bshift) + (leaf << bshift) + | |
5400 | (ztest_random(1ULL << (bshift - 1)) & -8ULL); | |
5401 | ||
5402 | if (offset >= fsize) | |
5403 | continue; | |
5404 | ||
5405 | mutex_enter(&ztest_vdev_lock); | |
5406 | if (mirror_save != zs->zs_mirrors) { | |
5407 | mutex_exit(&ztest_vdev_lock); | |
5408 | (void) close(fd); | |
5409 | goto out; | |
5410 | } | |
5411 | ||
5412 | if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad)) | |
5413 | fatal(1, "can't inject bad word at 0x%llx in %s", | |
5414 | offset, pathrand); | |
5415 | ||
5416 | mutex_exit(&ztest_vdev_lock); | |
5417 | ||
5418 | if (ztest_opts.zo_verbose >= 7) | |
5419 | (void) printf("injected bad word into %s," | |
5420 | " offset 0x%llx\n", pathrand, (u_longlong_t)offset); | |
5421 | } | |
5422 | ||
5423 | (void) close(fd); | |
5424 | out: | |
5425 | umem_free(path0, MAXPATHLEN); | |
5426 | umem_free(pathrand, MAXPATHLEN); | |
5427 | } | |
5428 | ||
5429 | /* | |
5430 | * Verify that DDT repair works as expected. | |
5431 | */ | |
5432 | void | |
5433 | ztest_ddt_repair(ztest_ds_t *zd, uint64_t id) | |
5434 | { | |
5435 | ztest_shared_t *zs = ztest_shared; | |
5436 | spa_t *spa = ztest_spa; | |
5437 | objset_t *os = zd->zd_os; | |
5438 | ztest_od_t *od; | |
5439 | uint64_t object, blocksize, txg, pattern, psize; | |
5440 | enum zio_checksum checksum = spa_dedup_checksum(spa); | |
5441 | dmu_buf_t *db; | |
5442 | dmu_tx_t *tx; | |
5443 | void *buf; | |
5444 | blkptr_t blk; | |
5445 | int copies = 2 * ZIO_DEDUPDITTO_MIN; | |
5446 | int i; | |
5447 | ||
5448 | blocksize = ztest_random_blocksize(); | |
5449 | blocksize = MIN(blocksize, 2048); /* because we write so many */ | |
5450 | ||
5451 | od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL); | |
5452 | ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0, 0); | |
5453 | ||
5454 | if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) { | |
5455 | umem_free(od, sizeof (ztest_od_t)); | |
5456 | return; | |
5457 | } | |
5458 | ||
5459 | /* | |
5460 | * Take the name lock as writer to prevent anyone else from changing | |
5461 | * the pool and dataset properies we need to maintain during this test. | |
5462 | */ | |
5463 | (void) rw_wrlock(&ztest_name_lock); | |
5464 | ||
5465 | if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum, | |
5466 | B_FALSE) != 0 || | |
5467 | ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1, | |
5468 | B_FALSE) != 0) { | |
5469 | (void) rw_unlock(&ztest_name_lock); | |
5470 | umem_free(od, sizeof (ztest_od_t)); | |
5471 | return; | |
5472 | } | |
5473 | ||
5474 | object = od[0].od_object; | |
5475 | blocksize = od[0].od_blocksize; | |
5476 | pattern = zs->zs_guid ^ dmu_objset_fsid_guid(os); | |
5477 | ||
5478 | ASSERT(object != 0); | |
5479 | ||
5480 | tx = dmu_tx_create(os); | |
5481 | dmu_tx_hold_write(tx, object, 0, copies * blocksize); | |
5482 | txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); | |
5483 | if (txg == 0) { | |
5484 | (void) rw_unlock(&ztest_name_lock); | |
5485 | umem_free(od, sizeof (ztest_od_t)); | |
5486 | return; | |
5487 | } | |
5488 | ||
5489 | /* | |
5490 | * Write all the copies of our block. | |
5491 | */ | |
5492 | for (i = 0; i < copies; i++) { | |
5493 | uint64_t offset = i * blocksize; | |
5494 | int error = dmu_buf_hold(os, object, offset, FTAG, &db, | |
5495 | DMU_READ_NO_PREFETCH); | |
5496 | if (error != 0) { | |
5497 | fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u", | |
5498 | os, (long long)object, (long long) offset, error); | |
5499 | } | |
5500 | ASSERT(db->db_offset == offset); | |
5501 | ASSERT(db->db_size == blocksize); | |
5502 | ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) || | |
5503 | ztest_pattern_match(db->db_data, db->db_size, 0ULL)); | |
5504 | dmu_buf_will_fill(db, tx); | |
5505 | ztest_pattern_set(db->db_data, db->db_size, pattern); | |
5506 | dmu_buf_rele(db, FTAG); | |
5507 | } | |
5508 | ||
5509 | dmu_tx_commit(tx); | |
5510 | txg_wait_synced(spa_get_dsl(spa), txg); | |
5511 | ||
5512 | /* | |
5513 | * Find out what block we got. | |
5514 | */ | |
5515 | VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db, | |
5516 | DMU_READ_NO_PREFETCH)); | |
5517 | blk = *((dmu_buf_impl_t *)db)->db_blkptr; | |
5518 | dmu_buf_rele(db, FTAG); | |
5519 | ||
5520 | /* | |
5521 | * Damage the block. Dedup-ditto will save us when we read it later. | |
5522 | */ | |
5523 | psize = BP_GET_PSIZE(&blk); | |
5524 | buf = zio_buf_alloc(psize); | |
5525 | ztest_pattern_set(buf, psize, ~pattern); | |
5526 | ||
5527 | (void) zio_wait(zio_rewrite(NULL, spa, 0, &blk, | |
5528 | buf, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE, | |
5529 | ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL)); | |
5530 | ||
5531 | zio_buf_free(buf, psize); | |
5532 | ||
5533 | (void) rw_unlock(&ztest_name_lock); | |
5534 | umem_free(od, sizeof (ztest_od_t)); | |
5535 | } | |
5536 | ||
5537 | /* | |
5538 | * Scrub the pool. | |
5539 | */ | |
5540 | /* ARGSUSED */ | |
5541 | void | |
5542 | ztest_scrub(ztest_ds_t *zd, uint64_t id) | |
5543 | { | |
5544 | spa_t *spa = ztest_spa; | |
5545 | ||
5546 | (void) spa_scan(spa, POOL_SCAN_SCRUB); | |
5547 | (void) poll(NULL, 0, 100); /* wait a moment, then force a restart */ | |
5548 | (void) spa_scan(spa, POOL_SCAN_SCRUB); | |
5549 | } | |
5550 | ||
5551 | /* | |
5552 | * Change the guid for the pool. | |
5553 | */ | |
5554 | /* ARGSUSED */ | |
5555 | void | |
5556 | ztest_reguid(ztest_ds_t *zd, uint64_t id) | |
5557 | { | |
5558 | spa_t *spa = ztest_spa; | |
5559 | uint64_t orig, load; | |
5560 | int error; | |
5561 | ||
5562 | orig = spa_guid(spa); | |
5563 | load = spa_load_guid(spa); | |
5564 | ||
5565 | (void) rw_wrlock(&ztest_name_lock); | |
5566 | error = spa_change_guid(spa); | |
5567 | (void) rw_unlock(&ztest_name_lock); | |
5568 | ||
5569 | if (error != 0) | |
5570 | return; | |
5571 | ||
5572 | if (ztest_opts.zo_verbose >= 4) { | |
5573 | (void) printf("Changed guid old %llu -> %llu\n", | |
5574 | (u_longlong_t)orig, (u_longlong_t)spa_guid(spa)); | |
5575 | } | |
5576 | ||
5577 | VERIFY3U(orig, !=, spa_guid(spa)); | |
5578 | VERIFY3U(load, ==, spa_load_guid(spa)); | |
5579 | } | |
5580 | ||
5581 | /* | |
5582 | * Rename the pool to a different name and then rename it back. | |
5583 | */ | |
5584 | /* ARGSUSED */ | |
5585 | void | |
5586 | ztest_spa_rename(ztest_ds_t *zd, uint64_t id) | |
5587 | { | |
5588 | char *oldname, *newname; | |
5589 | spa_t *spa; | |
5590 | ||
5591 | (void) rw_wrlock(&ztest_name_lock); | |
5592 | ||
5593 | oldname = ztest_opts.zo_pool; | |
5594 | newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL); | |
5595 | (void) strcpy(newname, oldname); | |
5596 | (void) strcat(newname, "_tmp"); | |
5597 | ||
5598 | /* | |
5599 | * Do the rename | |
5600 | */ | |
5601 | VERIFY3U(0, ==, spa_rename(oldname, newname)); | |
5602 | ||
5603 | /* | |
5604 | * Try to open it under the old name, which shouldn't exist | |
5605 | */ | |
5606 | VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG)); | |
5607 | ||
5608 | /* | |
5609 | * Open it under the new name and make sure it's still the same spa_t. | |
5610 | */ | |
5611 | VERIFY3U(0, ==, spa_open(newname, &spa, FTAG)); | |
5612 | ||
5613 | ASSERT(spa == ztest_spa); | |
5614 | spa_close(spa, FTAG); | |
5615 | ||
5616 | /* | |
5617 | * Rename it back to the original | |
5618 | */ | |
5619 | VERIFY3U(0, ==, spa_rename(newname, oldname)); | |
5620 | ||
5621 | /* | |
5622 | * Make sure it can still be opened | |
5623 | */ | |
5624 | VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG)); | |
5625 | ||
5626 | ASSERT(spa == ztest_spa); | |
5627 | spa_close(spa, FTAG); | |
5628 | ||
5629 | umem_free(newname, strlen(newname) + 1); | |
5630 | ||
5631 | (void) rw_unlock(&ztest_name_lock); | |
5632 | } | |
5633 | ||
5634 | void | |
5635 | ztest_fletcher(ztest_ds_t *zd, uint64_t id) | |
5636 | { | |
5637 | hrtime_t end = gethrtime() + NANOSEC; | |
5638 | ||
5639 | while (gethrtime() <= end) { | |
5640 | int run_count = 100; | |
5641 | void *buf; | |
5642 | uint32_t size; | |
5643 | int *ptr; | |
5644 | int i; | |
5645 | zio_cksum_t zc_ref; | |
5646 | zio_cksum_t zc_ref_byteswap; | |
5647 | ||
5648 | size = ztest_random_blocksize(); | |
5649 | buf = umem_alloc(size, UMEM_NOFAIL); | |
5650 | ||
5651 | for (i = 0, ptr = buf; i < size / sizeof (*ptr); i++, ptr++) | |
5652 | *ptr = ztest_random(UINT_MAX); | |
5653 | ||
5654 | VERIFY0(fletcher_4_impl_set("scalar")); | |
5655 | fletcher_4_native(buf, size, &zc_ref); | |
5656 | fletcher_4_byteswap(buf, size, &zc_ref_byteswap); | |
5657 | ||
5658 | VERIFY0(fletcher_4_impl_set("cycle")); | |
5659 | while (run_count-- > 0) { | |
5660 | zio_cksum_t zc; | |
5661 | zio_cksum_t zc_byteswap; | |
5662 | ||
5663 | fletcher_4_byteswap(buf, size, &zc_byteswap); | |
5664 | fletcher_4_native(buf, size, &zc); | |
5665 | ||
5666 | VERIFY0(bcmp(&zc, &zc_ref, sizeof (zc))); | |
5667 | VERIFY0(bcmp(&zc_byteswap, &zc_ref_byteswap, | |
5668 | sizeof (zc_byteswap))); | |
5669 | } | |
5670 | ||
5671 | umem_free(buf, size); | |
5672 | } | |
5673 | } | |
5674 | ||
5675 | static int | |
5676 | ztest_check_path(char *path) | |
5677 | { | |
5678 | struct stat s; | |
5679 | /* return true on success */ | |
5680 | return (!stat(path, &s)); | |
5681 | } | |
5682 | ||
5683 | static void | |
5684 | ztest_get_zdb_bin(char *bin, int len) | |
5685 | { | |
5686 | char *zdb_path; | |
5687 | /* | |
5688 | * Try to use ZDB_PATH and in-tree zdb path. If not successful, just | |
5689 | * let popen to search through PATH. | |
5690 | */ | |
5691 | if ((zdb_path = getenv("ZDB_PATH"))) { | |
5692 | strlcpy(bin, zdb_path, len); /* In env */ | |
5693 | if (!ztest_check_path(bin)) { | |
5694 | ztest_dump_core = 0; | |
5695 | fatal(1, "invalid ZDB_PATH '%s'", bin); | |
5696 | } | |
5697 | return; | |
5698 | } | |
5699 | ||
5700 | VERIFY(realpath(getexecname(), bin) != NULL); | |
5701 | if (strstr(bin, "/ztest/")) { | |
5702 | strstr(bin, "/ztest/")[0] = '\0'; /* In-tree */ | |
5703 | strcat(bin, "/zdb/zdb"); | |
5704 | if (ztest_check_path(bin)) | |
5705 | return; | |
5706 | } | |
5707 | strcpy(bin, "zdb"); | |
5708 | } | |
5709 | ||
5710 | /* | |
5711 | * Verify pool integrity by running zdb. | |
5712 | */ | |
5713 | static void | |
5714 | ztest_run_zdb(char *pool) | |
5715 | { | |
5716 | int status; | |
5717 | char *bin; | |
5718 | char *zdb; | |
5719 | char *zbuf; | |
5720 | const int len = MAXPATHLEN + MAXNAMELEN + 20; | |
5721 | FILE *fp; | |
5722 | ||
5723 | bin = umem_alloc(len, UMEM_NOFAIL); | |
5724 | zdb = umem_alloc(len, UMEM_NOFAIL); | |
5725 | zbuf = umem_alloc(1024, UMEM_NOFAIL); | |
5726 | ||
5727 | ztest_get_zdb_bin(bin, len); | |
5728 | ||
5729 | (void) sprintf(zdb, | |
5730 | "%s -bcc%s%s -d -U %s %s", | |
5731 | bin, | |
5732 | ztest_opts.zo_verbose >= 3 ? "s" : "", | |
5733 | ztest_opts.zo_verbose >= 4 ? "v" : "", | |
5734 | spa_config_path, | |
5735 | pool); | |
5736 | ||
5737 | if (ztest_opts.zo_verbose >= 5) | |
5738 | (void) printf("Executing %s\n", strstr(zdb, "zdb ")); | |
5739 | ||
5740 | fp = popen(zdb, "r"); | |
5741 | ||
5742 | while (fgets(zbuf, 1024, fp) != NULL) | |
5743 | if (ztest_opts.zo_verbose >= 3) | |
5744 | (void) printf("%s", zbuf); | |
5745 | ||
5746 | status = pclose(fp); | |
5747 | ||
5748 | if (status == 0) | |
5749 | goto out; | |
5750 | ||
5751 | ztest_dump_core = 0; | |
5752 | if (WIFEXITED(status)) | |
5753 | fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status)); | |
5754 | else | |
5755 | fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status)); | |
5756 | out: | |
5757 | umem_free(bin, len); | |
5758 | umem_free(zdb, len); | |
5759 | umem_free(zbuf, 1024); | |
5760 | } | |
5761 | ||
5762 | static void | |
5763 | ztest_walk_pool_directory(char *header) | |
5764 | { | |
5765 | spa_t *spa = NULL; | |
5766 | ||
5767 | if (ztest_opts.zo_verbose >= 6) | |
5768 | (void) printf("%s\n", header); | |
5769 | ||
5770 | mutex_enter(&spa_namespace_lock); | |
5771 | while ((spa = spa_next(spa)) != NULL) | |
5772 | if (ztest_opts.zo_verbose >= 6) | |
5773 | (void) printf("\t%s\n", spa_name(spa)); | |
5774 | mutex_exit(&spa_namespace_lock); | |
5775 | } | |
5776 | ||
5777 | static void | |
5778 | ztest_spa_import_export(char *oldname, char *newname) | |
5779 | { | |
5780 | nvlist_t *config, *newconfig; | |
5781 | uint64_t pool_guid; | |
5782 | spa_t *spa; | |
5783 | int error; | |
5784 | ||
5785 | if (ztest_opts.zo_verbose >= 4) { | |
5786 | (void) printf("import/export: old = %s, new = %s\n", | |
5787 | oldname, newname); | |
5788 | } | |
5789 | ||
5790 | /* | |
5791 | * Clean up from previous runs. | |
5792 | */ | |
5793 | (void) spa_destroy(newname); | |
5794 | ||
5795 | /* | |
5796 | * Get the pool's configuration and guid. | |
5797 | */ | |
5798 | VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG)); | |
5799 | ||
5800 | /* | |
5801 | * Kick off a scrub to tickle scrub/export races. | |
5802 | */ | |
5803 | if (ztest_random(2) == 0) | |
5804 | (void) spa_scan(spa, POOL_SCAN_SCRUB); | |
5805 | ||
5806 | pool_guid = spa_guid(spa); | |
5807 | spa_close(spa, FTAG); | |
5808 | ||
5809 | ztest_walk_pool_directory("pools before export"); | |
5810 | ||
5811 | /* | |
5812 | * Export it. | |
5813 | */ | |
5814 | VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE)); | |
5815 | ||
5816 | ztest_walk_pool_directory("pools after export"); | |
5817 | ||
5818 | /* | |
5819 | * Try to import it. | |
5820 | */ | |
5821 | newconfig = spa_tryimport(config); | |
5822 | ASSERT(newconfig != NULL); | |
5823 | nvlist_free(newconfig); | |
5824 | ||
5825 | /* | |
5826 | * Import it under the new name. | |
5827 | */ | |
5828 | error = spa_import(newname, config, NULL, 0); | |
5829 | if (error != 0) { | |
5830 | dump_nvlist(config, 0); | |
5831 | fatal(B_FALSE, "couldn't import pool %s as %s: error %u", | |
5832 | oldname, newname, error); | |
5833 | } | |
5834 | ||
5835 | ztest_walk_pool_directory("pools after import"); | |
5836 | ||
5837 | /* | |
5838 | * Try to import it again -- should fail with EEXIST. | |
5839 | */ | |
5840 | VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0)); | |
5841 | ||
5842 | /* | |
5843 | * Try to import it under a different name -- should fail with EEXIST. | |
5844 | */ | |
5845 | VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0)); | |
5846 | ||
5847 | /* | |
5848 | * Verify that the pool is no longer visible under the old name. | |
5849 | */ | |
5850 | VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG)); | |
5851 | ||
5852 | /* | |
5853 | * Verify that we can open and close the pool using the new name. | |
5854 | */ | |
5855 | VERIFY3U(0, ==, spa_open(newname, &spa, FTAG)); | |
5856 | ASSERT(pool_guid == spa_guid(spa)); | |
5857 | spa_close(spa, FTAG); | |
5858 | ||
5859 | nvlist_free(config); | |
5860 | } | |
5861 | ||
5862 | static void | |
5863 | ztest_resume(spa_t *spa) | |
5864 | { | |
5865 | if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6) | |
5866 | (void) printf("resuming from suspended state\n"); | |
5867 | spa_vdev_state_enter(spa, SCL_NONE); | |
5868 | vdev_clear(spa, NULL); | |
5869 | (void) spa_vdev_state_exit(spa, NULL, 0); | |
5870 | (void) zio_resume(spa); | |
5871 | } | |
5872 | ||
5873 | static void * | |
5874 | ztest_resume_thread(void *arg) | |
5875 | { | |
5876 | spa_t *spa = arg; | |
5877 | ||
5878 | while (!ztest_exiting) { | |
5879 | if (spa_suspended(spa)) | |
5880 | ztest_resume(spa); | |
5881 | (void) poll(NULL, 0, 100); | |
5882 | } | |
5883 | ||
5884 | thread_exit(); | |
5885 | ||
5886 | return (NULL); | |
5887 | } | |
5888 | ||
5889 | #define GRACE 300 | |
5890 | ||
5891 | #if 0 | |
5892 | static void | |
5893 | ztest_deadman_alarm(int sig) | |
5894 | { | |
5895 | fatal(0, "failed to complete within %d seconds of deadline", GRACE); | |
5896 | } | |
5897 | #endif | |
5898 | ||
5899 | static void | |
5900 | ztest_execute(int test, ztest_info_t *zi, uint64_t id) | |
5901 | { | |
5902 | ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets]; | |
5903 | ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test); | |
5904 | hrtime_t functime = gethrtime(); | |
5905 | int i; | |
5906 | ||
5907 | for (i = 0; i < zi->zi_iters; i++) | |
5908 | zi->zi_func(zd, id); | |
5909 | ||
5910 | functime = gethrtime() - functime; | |
5911 | ||
5912 | atomic_add_64(&zc->zc_count, 1); | |
5913 | atomic_add_64(&zc->zc_time, functime); | |
5914 | ||
5915 | if (ztest_opts.zo_verbose >= 4) | |
5916 | (void) printf("%6.2f sec in %s\n", | |
5917 | (double)functime / NANOSEC, zi->zi_funcname); | |
5918 | } | |
5919 | ||
5920 | static void * | |
5921 | ztest_thread(void *arg) | |
5922 | { | |
5923 | int rand; | |
5924 | uint64_t id = (uintptr_t)arg; | |
5925 | ztest_shared_t *zs = ztest_shared; | |
5926 | uint64_t call_next; | |
5927 | hrtime_t now; | |
5928 | ztest_info_t *zi; | |
5929 | ztest_shared_callstate_t *zc; | |
5930 | ||
5931 | while ((now = gethrtime()) < zs->zs_thread_stop) { | |
5932 | /* | |
5933 | * See if it's time to force a crash. | |
5934 | */ | |
5935 | if (now > zs->zs_thread_kill) | |
5936 | ztest_kill(zs); | |
5937 | ||
5938 | /* | |
5939 | * If we're getting ENOSPC with some regularity, stop. | |
5940 | */ | |
5941 | if (zs->zs_enospc_count > 10) | |
5942 | break; | |
5943 | ||
5944 | /* | |
5945 | * Pick a random function to execute. | |
5946 | */ | |
5947 | rand = ztest_random(ZTEST_FUNCS); | |
5948 | zi = &ztest_info[rand]; | |
5949 | zc = ZTEST_GET_SHARED_CALLSTATE(rand); | |
5950 | call_next = zc->zc_next; | |
5951 | ||
5952 | if (now >= call_next && | |
5953 | atomic_cas_64(&zc->zc_next, call_next, call_next + | |
5954 | ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) { | |
5955 | ztest_execute(rand, zi, id); | |
5956 | } | |
5957 | } | |
5958 | ||
5959 | thread_exit(); | |
5960 | ||
5961 | return (NULL); | |
5962 | } | |
5963 | ||
5964 | static void | |
5965 | ztest_dataset_name(char *dsname, char *pool, int d) | |
5966 | { | |
5967 | (void) snprintf(dsname, MAXNAMELEN, "%s/ds_%d", pool, d); | |
5968 | } | |
5969 | ||
5970 | static void | |
5971 | ztest_dataset_destroy(int d) | |
5972 | { | |
5973 | char name[MAXNAMELEN]; | |
5974 | int t; | |
5975 | ||
5976 | ztest_dataset_name(name, ztest_opts.zo_pool, d); | |
5977 | ||
5978 | if (ztest_opts.zo_verbose >= 3) | |
5979 | (void) printf("Destroying %s to free up space\n", name); | |
5980 | ||
5981 | /* | |
5982 | * Cleanup any non-standard clones and snapshots. In general, | |
5983 | * ztest thread t operates on dataset (t % zopt_datasets), | |
5984 | * so there may be more than one thing to clean up. | |
5985 | */ | |
5986 | for (t = d; t < ztest_opts.zo_threads; | |
5987 | t += ztest_opts.zo_datasets) | |
5988 | ztest_dsl_dataset_cleanup(name, t); | |
5989 | ||
5990 | (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL, | |
5991 | DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN); | |
5992 | } | |
5993 | ||
5994 | static void | |
5995 | ztest_dataset_dirobj_verify(ztest_ds_t *zd) | |
5996 | { | |
5997 | uint64_t usedobjs, dirobjs, scratch; | |
5998 | ||
5999 | /* | |
6000 | * ZTEST_DIROBJ is the object directory for the entire dataset. | |
6001 | * Therefore, the number of objects in use should equal the | |
6002 | * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself. | |
6003 | * If not, we have an object leak. | |
6004 | * | |
6005 | * Note that we can only check this in ztest_dataset_open(), | |
6006 | * when the open-context and syncing-context values agree. | |
6007 | * That's because zap_count() returns the open-context value, | |
6008 | * while dmu_objset_space() returns the rootbp fill count. | |
6009 | */ | |
6010 | VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs)); | |
6011 | dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch); | |
6012 | ASSERT3U(dirobjs + 1, ==, usedobjs); | |
6013 | } | |
6014 | ||
6015 | static int | |
6016 | ztest_dataset_open(int d) | |
6017 | { | |
6018 | ztest_ds_t *zd = &ztest_ds[d]; | |
6019 | uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq; | |
6020 | objset_t *os; | |
6021 | zilog_t *zilog; | |
6022 | char name[MAXNAMELEN]; | |
6023 | int error; | |
6024 | ||
6025 | ztest_dataset_name(name, ztest_opts.zo_pool, d); | |
6026 | ||
6027 | (void) rw_rdlock(&ztest_name_lock); | |
6028 | ||
6029 | error = ztest_dataset_create(name); | |
6030 | if (error == ENOSPC) { | |
6031 | (void) rw_unlock(&ztest_name_lock); | |
6032 | ztest_record_enospc(FTAG); | |
6033 | return (error); | |
6034 | } | |
6035 | ASSERT(error == 0 || error == EEXIST); | |
6036 | ||
6037 | VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, zd, &os)); | |
6038 | (void) rw_unlock(&ztest_name_lock); | |
6039 | ||
6040 | ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os); | |
6041 | ||
6042 | zilog = zd->zd_zilog; | |
6043 | ||
6044 | if (zilog->zl_header->zh_claim_lr_seq != 0 && | |
6045 | zilog->zl_header->zh_claim_lr_seq < committed_seq) | |
6046 | fatal(0, "missing log records: claimed %llu < committed %llu", | |
6047 | zilog->zl_header->zh_claim_lr_seq, committed_seq); | |
6048 | ||
6049 | ztest_dataset_dirobj_verify(zd); | |
6050 | ||
6051 | zil_replay(os, zd, ztest_replay_vector); | |
6052 | ||
6053 | ztest_dataset_dirobj_verify(zd); | |
6054 | ||
6055 | if (ztest_opts.zo_verbose >= 6) | |
6056 | (void) printf("%s replay %llu blocks, %llu records, seq %llu\n", | |
6057 | zd->zd_name, | |
6058 | (u_longlong_t)zilog->zl_parse_blk_count, | |
6059 | (u_longlong_t)zilog->zl_parse_lr_count, | |
6060 | (u_longlong_t)zilog->zl_replaying_seq); | |
6061 | ||
6062 | zilog = zil_open(os, ztest_get_data); | |
6063 | ||
6064 | if (zilog->zl_replaying_seq != 0 && | |
6065 | zilog->zl_replaying_seq < committed_seq) | |
6066 | fatal(0, "missing log records: replayed %llu < committed %llu", | |
6067 | zilog->zl_replaying_seq, committed_seq); | |
6068 | ||
6069 | return (0); | |
6070 | } | |
6071 | ||
6072 | static void | |
6073 | ztest_dataset_close(int d) | |
6074 | { | |
6075 | ztest_ds_t *zd = &ztest_ds[d]; | |
6076 | ||
6077 | zil_close(zd->zd_zilog); | |
6078 | dmu_objset_disown(zd->zd_os, zd); | |
6079 | ||
6080 | ztest_zd_fini(zd); | |
6081 | } | |
6082 | ||
6083 | /* | |
6084 | * Kick off threads to run tests on all datasets in parallel. | |
6085 | */ | |
6086 | static void | |
6087 | ztest_run(ztest_shared_t *zs) | |
6088 | { | |
6089 | kt_did_t *tid; | |
6090 | spa_t *spa; | |
6091 | objset_t *os; | |
6092 | kthread_t *resume_thread; | |
6093 | uint64_t object; | |
6094 | int error; | |
6095 | int t, d; | |
6096 | ||
6097 | ztest_exiting = B_FALSE; | |
6098 | ||
6099 | /* | |
6100 | * Initialize parent/child shared state. | |
6101 | */ | |
6102 | mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL); | |
6103 | VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0); | |
6104 | ||
6105 | zs->zs_thread_start = gethrtime(); | |
6106 | zs->zs_thread_stop = | |
6107 | zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC; | |
6108 | zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop); | |
6109 | zs->zs_thread_kill = zs->zs_thread_stop; | |
6110 | if (ztest_random(100) < ztest_opts.zo_killrate) { | |
6111 | zs->zs_thread_kill -= | |
6112 | ztest_random(ztest_opts.zo_passtime * NANOSEC); | |
6113 | } | |
6114 | ||
6115 | mutex_init(&zcl.zcl_callbacks_lock, NULL, MUTEX_DEFAULT, NULL); | |
6116 | ||
6117 | list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t), | |
6118 | offsetof(ztest_cb_data_t, zcd_node)); | |
6119 | ||
6120 | /* | |
6121 | * Open our pool. | |
6122 | */ | |
6123 | kernel_init(FREAD | FWRITE); | |
6124 | VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG)); | |
6125 | spa->spa_debug = B_TRUE; | |
6126 | metaslab_preload_limit = ztest_random(20) + 1; | |
6127 | ztest_spa = spa; | |
6128 | ||
6129 | VERIFY0(dmu_objset_own(ztest_opts.zo_pool, | |
6130 | DMU_OST_ANY, B_TRUE, FTAG, &os)); | |
6131 | zs->zs_guid = dmu_objset_fsid_guid(os); | |
6132 | dmu_objset_disown(os, FTAG); | |
6133 | ||
6134 | spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN; | |
6135 | ||
6136 | /* | |
6137 | * We don't expect the pool to suspend unless maxfaults == 0, | |
6138 | * in which case ztest_fault_inject() temporarily takes away | |
6139 | * the only valid replica. | |
6140 | */ | |
6141 | if (MAXFAULTS() == 0) | |
6142 | spa->spa_failmode = ZIO_FAILURE_MODE_WAIT; | |
6143 | else | |
6144 | spa->spa_failmode = ZIO_FAILURE_MODE_PANIC; | |
6145 | ||
6146 | /* | |
6147 | * Create a thread to periodically resume suspended I/O. | |
6148 | */ | |
6149 | VERIFY3P((resume_thread = zk_thread_create(NULL, 0, | |
6150 | (thread_func_t)ztest_resume_thread, spa, TS_RUN, NULL, 0, 0, | |
6151 | PTHREAD_CREATE_JOINABLE)), !=, NULL); | |
6152 | ||
6153 | #if 0 | |
6154 | /* | |
6155 | * Set a deadman alarm to abort() if we hang. | |
6156 | */ | |
6157 | signal(SIGALRM, ztest_deadman_alarm); | |
6158 | alarm((zs->zs_thread_stop - zs->zs_thread_start) / NANOSEC + GRACE); | |
6159 | #endif | |
6160 | ||
6161 | /* | |
6162 | * Verify that we can safely inquire about about any object, | |
6163 | * whether it's allocated or not. To make it interesting, | |
6164 | * we probe a 5-wide window around each power of two. | |
6165 | * This hits all edge cases, including zero and the max. | |
6166 | */ | |
6167 | for (t = 0; t < 64; t++) { | |
6168 | for (d = -5; d <= 5; d++) { | |
6169 | error = dmu_object_info(spa->spa_meta_objset, | |
6170 | (1ULL << t) + d, NULL); | |
6171 | ASSERT(error == 0 || error == ENOENT || | |
6172 | error == EINVAL); | |
6173 | } | |
6174 | } | |
6175 | ||
6176 | /* | |
6177 | * If we got any ENOSPC errors on the previous run, destroy something. | |
6178 | */ | |
6179 | if (zs->zs_enospc_count != 0) { | |
6180 | int d = ztest_random(ztest_opts.zo_datasets); | |
6181 | ztest_dataset_destroy(d); | |
6182 | } | |
6183 | zs->zs_enospc_count = 0; | |
6184 | ||
6185 | tid = umem_zalloc(ztest_opts.zo_threads * sizeof (kt_did_t), | |
6186 | UMEM_NOFAIL); | |
6187 | ||
6188 | if (ztest_opts.zo_verbose >= 4) | |
6189 | (void) printf("starting main threads...\n"); | |
6190 | ||
6191 | /* | |
6192 | * Kick off all the tests that run in parallel. | |
6193 | */ | |
6194 | for (t = 0; t < ztest_opts.zo_threads; t++) { | |
6195 | kthread_t *thread; | |
6196 | ||
6197 | if (t < ztest_opts.zo_datasets && | |
6198 | ztest_dataset_open(t) != 0) | |
6199 | return; | |
6200 | ||
6201 | VERIFY3P(thread = zk_thread_create(NULL, 0, | |
6202 | (thread_func_t)ztest_thread, | |
6203 | (void *)(uintptr_t)t, TS_RUN, NULL, 0, 0, | |
6204 | PTHREAD_CREATE_JOINABLE), !=, NULL); | |
6205 | tid[t] = thread->t_tid; | |
6206 | } | |
6207 | ||
6208 | /* | |
6209 | * Wait for all of the tests to complete. We go in reverse order | |
6210 | * so we don't close datasets while threads are still using them. | |
6211 | */ | |
6212 | for (t = ztest_opts.zo_threads - 1; t >= 0; t--) { | |
6213 | thread_join(tid[t]); | |
6214 | if (t < ztest_opts.zo_datasets) | |
6215 | ztest_dataset_close(t); | |
6216 | } | |
6217 | ||
6218 | txg_wait_synced(spa_get_dsl(spa), 0); | |
6219 | ||
6220 | zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa)); | |
6221 | zs->zs_space = metaslab_class_get_space(spa_normal_class(spa)); | |
6222 | ||
6223 | umem_free(tid, ztest_opts.zo_threads * sizeof (kt_did_t)); | |
6224 | ||
6225 | /* Kill the resume thread */ | |
6226 | ztest_exiting = B_TRUE; | |
6227 | thread_join(resume_thread->t_tid); | |
6228 | ztest_resume(spa); | |
6229 | ||
6230 | /* | |
6231 | * Right before closing the pool, kick off a bunch of async I/O; | |
6232 | * spa_close() should wait for it to complete. | |
6233 | */ | |
6234 | for (object = 1; object < 50; object++) { | |
6235 | dmu_prefetch(spa->spa_meta_objset, object, 0, 0, 1ULL << 20, | |
6236 | ZIO_PRIORITY_SYNC_READ); | |
6237 | } | |
6238 | ||
6239 | /* Verify that at least one commit cb was called in a timely fashion */ | |
6240 | if (zc_cb_counter >= ZTEST_COMMIT_CB_MIN_REG) | |
6241 | VERIFY0(zc_min_txg_delay); | |
6242 | ||
6243 | spa_close(spa, FTAG); | |
6244 | ||
6245 | /* | |
6246 | * Verify that we can loop over all pools. | |
6247 | */ | |
6248 | mutex_enter(&spa_namespace_lock); | |
6249 | for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa)) | |
6250 | if (ztest_opts.zo_verbose > 3) | |
6251 | (void) printf("spa_next: found %s\n", spa_name(spa)); | |
6252 | mutex_exit(&spa_namespace_lock); | |
6253 | ||
6254 | /* | |
6255 | * Verify that we can export the pool and reimport it under a | |
6256 | * different name. | |
6257 | */ | |
6258 | if (ztest_random(2) == 0) { | |
6259 | char name[MAXNAMELEN]; | |
6260 | (void) snprintf(name, MAXNAMELEN, "%s_import", | |
6261 | ztest_opts.zo_pool); | |
6262 | ztest_spa_import_export(ztest_opts.zo_pool, name); | |
6263 | ztest_spa_import_export(name, ztest_opts.zo_pool); | |
6264 | } | |
6265 | ||
6266 | kernel_fini(); | |
6267 | ||
6268 | list_destroy(&zcl.zcl_callbacks); | |
6269 | mutex_destroy(&zcl.zcl_callbacks_lock); | |
6270 | (void) rwlock_destroy(&ztest_name_lock); | |
6271 | mutex_destroy(&ztest_vdev_lock); | |
6272 | } | |
6273 | ||
6274 | static void | |
6275 | ztest_freeze(void) | |
6276 | { | |
6277 | ztest_ds_t *zd = &ztest_ds[0]; | |
6278 | spa_t *spa; | |
6279 | int numloops = 0; | |
6280 | ||
6281 | if (ztest_opts.zo_verbose >= 3) | |
6282 | (void) printf("testing spa_freeze()...\n"); | |
6283 | ||
6284 | kernel_init(FREAD | FWRITE); | |
6285 | VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG)); | |
6286 | VERIFY3U(0, ==, ztest_dataset_open(0)); | |
6287 | spa->spa_debug = B_TRUE; | |
6288 | ztest_spa = spa; | |
6289 | ||
6290 | /* | |
6291 | * Force the first log block to be transactionally allocated. | |
6292 | * We have to do this before we freeze the pool -- otherwise | |
6293 | * the log chain won't be anchored. | |
6294 | */ | |
6295 | while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) { | |
6296 | ztest_dmu_object_alloc_free(zd, 0); | |
6297 | zil_commit(zd->zd_zilog, 0); | |
6298 | } | |
6299 | ||
6300 | txg_wait_synced(spa_get_dsl(spa), 0); | |
6301 | ||
6302 | /* | |
6303 | * Freeze the pool. This stops spa_sync() from doing anything, | |
6304 | * so that the only way to record changes from now on is the ZIL. | |
6305 | */ | |
6306 | spa_freeze(spa); | |
6307 | ||
6308 | /* | |
6309 | * Because it is hard to predict how much space a write will actually | |
6310 | * require beforehand, we leave ourselves some fudge space to write over | |
6311 | * capacity. | |
6312 | */ | |
6313 | uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2; | |
6314 | ||
6315 | /* | |
6316 | * Run tests that generate log records but don't alter the pool config | |
6317 | * or depend on DSL sync tasks (snapshots, objset create/destroy, etc). | |
6318 | * We do a txg_wait_synced() after each iteration to force the txg | |
6319 | * to increase well beyond the last synced value in the uberblock. | |
6320 | * The ZIL should be OK with that. | |
6321 | * | |
6322 | * Run a random number of times less than zo_maxloops and ensure we do | |
6323 | * not run out of space on the pool. | |
6324 | */ | |
6325 | while (ztest_random(10) != 0 && | |
6326 | numloops++ < ztest_opts.zo_maxloops && | |
6327 | metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) { | |
6328 | ztest_od_t od; | |
6329 | ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0); | |
6330 | VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE)); | |
6331 | ztest_io(zd, od.od_object, | |
6332 | ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT); | |
6333 | txg_wait_synced(spa_get_dsl(spa), 0); | |
6334 | } | |
6335 | ||
6336 | /* | |
6337 | * Commit all of the changes we just generated. | |
6338 | */ | |
6339 | zil_commit(zd->zd_zilog, 0); | |
6340 | txg_wait_synced(spa_get_dsl(spa), 0); | |
6341 | ||
6342 | /* | |
6343 | * Close our dataset and close the pool. | |
6344 | */ | |
6345 | ztest_dataset_close(0); | |
6346 | spa_close(spa, FTAG); | |
6347 | kernel_fini(); | |
6348 | ||
6349 | /* | |
6350 | * Open and close the pool and dataset to induce log replay. | |
6351 | */ | |
6352 | kernel_init(FREAD | FWRITE); | |
6353 | VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG)); | |
6354 | ASSERT(spa_freeze_txg(spa) == UINT64_MAX); | |
6355 | VERIFY3U(0, ==, ztest_dataset_open(0)); | |
6356 | ztest_dataset_close(0); | |
6357 | ||
6358 | spa->spa_debug = B_TRUE; | |
6359 | ztest_spa = spa; | |
6360 | txg_wait_synced(spa_get_dsl(spa), 0); | |
6361 | ztest_reguid(NULL, 0); | |
6362 | ||
6363 | spa_close(spa, FTAG); | |
6364 | kernel_fini(); | |
6365 | } | |
6366 | ||
6367 | void | |
6368 | print_time(hrtime_t t, char *timebuf) | |
6369 | { | |
6370 | hrtime_t s = t / NANOSEC; | |
6371 | hrtime_t m = s / 60; | |
6372 | hrtime_t h = m / 60; | |
6373 | hrtime_t d = h / 24; | |
6374 | ||
6375 | s -= m * 60; | |
6376 | m -= h * 60; | |
6377 | h -= d * 24; | |
6378 | ||
6379 | timebuf[0] = '\0'; | |
6380 | ||
6381 | if (d) | |
6382 | (void) sprintf(timebuf, | |
6383 | "%llud%02lluh%02llum%02llus", d, h, m, s); | |
6384 | else if (h) | |
6385 | (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s); | |
6386 | else if (m) | |
6387 | (void) sprintf(timebuf, "%llum%02llus", m, s); | |
6388 | else | |
6389 | (void) sprintf(timebuf, "%llus", s); | |
6390 | } | |
6391 | ||
6392 | static nvlist_t * | |
6393 | make_random_props(void) | |
6394 | { | |
6395 | nvlist_t *props; | |
6396 | ||
6397 | VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0); | |
6398 | if (ztest_random(2) == 0) | |
6399 | return (props); | |
6400 | VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0); | |
6401 | ||
6402 | return (props); | |
6403 | } | |
6404 | ||
6405 | /* | |
6406 | * Create a storage pool with the given name and initial vdev size. | |
6407 | * Then test spa_freeze() functionality. | |
6408 | */ | |
6409 | static void | |
6410 | ztest_init(ztest_shared_t *zs) | |
6411 | { | |
6412 | spa_t *spa; | |
6413 | nvlist_t *nvroot, *props; | |
6414 | int i; | |
6415 | ||
6416 | mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL); | |
6417 | VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0); | |
6418 | ||
6419 | kernel_init(FREAD | FWRITE); | |
6420 | ||
6421 | /* | |
6422 | * Create the storage pool. | |
6423 | */ | |
6424 | (void) spa_destroy(ztest_opts.zo_pool); | |
6425 | ztest_shared->zs_vdev_next_leaf = 0; | |
6426 | zs->zs_splits = 0; | |
6427 | zs->zs_mirrors = ztest_opts.zo_mirrors; | |
6428 | nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0, | |
6429 | 0, ztest_opts.zo_raidz, zs->zs_mirrors, 1); | |
6430 | props = make_random_props(); | |
6431 | for (i = 0; i < SPA_FEATURES; i++) { | |
6432 | char *buf; | |
6433 | VERIFY3S(-1, !=, asprintf(&buf, "feature@%s", | |
6434 | spa_feature_table[i].fi_uname)); | |
6435 | VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0)); | |
6436 | free(buf); | |
6437 | } | |
6438 | VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props, NULL)); | |
6439 | nvlist_free(nvroot); | |
6440 | nvlist_free(props); | |
6441 | ||
6442 | VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG)); | |
6443 | zs->zs_metaslab_sz = | |
6444 | 1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift; | |
6445 | spa_close(spa, FTAG); | |
6446 | ||
6447 | kernel_fini(); | |
6448 | ||
6449 | ztest_run_zdb(ztest_opts.zo_pool); | |
6450 | ||
6451 | ztest_freeze(); | |
6452 | ||
6453 | ztest_run_zdb(ztest_opts.zo_pool); | |
6454 | ||
6455 | (void) rwlock_destroy(&ztest_name_lock); | |
6456 | mutex_destroy(&ztest_vdev_lock); | |
6457 | } | |
6458 | ||
6459 | static void | |
6460 | setup_data_fd(void) | |
6461 | { | |
6462 | static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX"; | |
6463 | ||
6464 | ztest_fd_data = mkstemp(ztest_name_data); | |
6465 | ASSERT3S(ztest_fd_data, >=, 0); | |
6466 | (void) unlink(ztest_name_data); | |
6467 | } | |
6468 | ||
6469 | static int | |
6470 | shared_data_size(ztest_shared_hdr_t *hdr) | |
6471 | { | |
6472 | int size; | |
6473 | ||
6474 | size = hdr->zh_hdr_size; | |
6475 | size += hdr->zh_opts_size; | |
6476 | size += hdr->zh_size; | |
6477 | size += hdr->zh_stats_size * hdr->zh_stats_count; | |
6478 | size += hdr->zh_ds_size * hdr->zh_ds_count; | |
6479 | ||
6480 | return (size); | |
6481 | } | |
6482 | ||
6483 | static void | |
6484 | setup_hdr(void) | |
6485 | { | |
6486 | int size; | |
6487 | ztest_shared_hdr_t *hdr; | |
6488 | ||
6489 | hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()), | |
6490 | PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0); | |
6491 | ASSERT(hdr != MAP_FAILED); | |
6492 | ||
6493 | VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t))); | |
6494 | ||
6495 | hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t); | |
6496 | hdr->zh_opts_size = sizeof (ztest_shared_opts_t); | |
6497 | hdr->zh_size = sizeof (ztest_shared_t); | |
6498 | hdr->zh_stats_size = sizeof (ztest_shared_callstate_t); | |
6499 | hdr->zh_stats_count = ZTEST_FUNCS; | |
6500 | hdr->zh_ds_size = sizeof (ztest_shared_ds_t); | |
6501 | hdr->zh_ds_count = ztest_opts.zo_datasets; | |
6502 | ||
6503 | size = shared_data_size(hdr); | |
6504 | VERIFY3U(0, ==, ftruncate(ztest_fd_data, size)); | |
6505 | ||
6506 | (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize())); | |
6507 | } | |
6508 | ||
6509 | static void | |
6510 | setup_data(void) | |
6511 | { | |
6512 | int size, offset; | |
6513 | ztest_shared_hdr_t *hdr; | |
6514 | uint8_t *buf; | |
6515 | ||
6516 | hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()), | |
6517 | PROT_READ, MAP_SHARED, ztest_fd_data, 0); | |
6518 | ASSERT(hdr != MAP_FAILED); | |
6519 | ||
6520 | size = shared_data_size(hdr); | |
6521 | ||
6522 | (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize())); | |
6523 | hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()), | |
6524 | PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0); | |
6525 | ASSERT(hdr != MAP_FAILED); | |
6526 | buf = (uint8_t *)hdr; | |
6527 | ||
6528 | offset = hdr->zh_hdr_size; | |
6529 | ztest_shared_opts = (void *)&buf[offset]; | |
6530 | offset += hdr->zh_opts_size; | |
6531 | ztest_shared = (void *)&buf[offset]; | |
6532 | offset += hdr->zh_size; | |
6533 | ztest_shared_callstate = (void *)&buf[offset]; | |
6534 | offset += hdr->zh_stats_size * hdr->zh_stats_count; | |
6535 | ztest_shared_ds = (void *)&buf[offset]; | |
6536 | } | |
6537 | ||
6538 | static boolean_t | |
6539 | exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp) | |
6540 | { | |
6541 | pid_t pid; | |
6542 | int status; | |
6543 | char *cmdbuf = NULL; | |
6544 | ||
6545 | pid = fork(); | |
6546 | ||
6547 | if (cmd == NULL) { | |
6548 | cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); | |
6549 | (void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN); | |
6550 | cmd = cmdbuf; | |
6551 | } | |
6552 | ||
6553 | if (pid == -1) | |
6554 | fatal(1, "fork failed"); | |
6555 | ||
6556 | if (pid == 0) { /* child */ | |
6557 | char *emptyargv[2] = { cmd, NULL }; | |
6558 | char fd_data_str[12]; | |
6559 | ||
6560 | struct rlimit rl = { 1024, 1024 }; | |
6561 | (void) setrlimit(RLIMIT_NOFILE, &rl); | |
6562 | ||
6563 | (void) close(ztest_fd_rand); | |
6564 | VERIFY(11 >= snprintf(fd_data_str, 12, "%d", ztest_fd_data)); | |
6565 | VERIFY(0 == setenv("ZTEST_FD_DATA", fd_data_str, 1)); | |
6566 | ||
6567 | (void) enable_extended_FILE_stdio(-1, -1); | |
6568 | if (libpath != NULL) | |
6569 | VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1)); | |
6570 | (void) execv(cmd, emptyargv); | |
6571 | ztest_dump_core = B_FALSE; | |
6572 | fatal(B_TRUE, "exec failed: %s", cmd); | |
6573 | } | |
6574 | ||
6575 | if (cmdbuf != NULL) { | |
6576 | umem_free(cmdbuf, MAXPATHLEN); | |
6577 | cmd = NULL; | |
6578 | } | |
6579 | ||
6580 | while (waitpid(pid, &status, 0) != pid) | |
6581 | continue; | |
6582 | if (statusp != NULL) | |
6583 | *statusp = status; | |
6584 | ||
6585 | if (WIFEXITED(status)) { | |
6586 | if (WEXITSTATUS(status) != 0) { | |
6587 | (void) fprintf(stderr, "child exited with code %d\n", | |
6588 | WEXITSTATUS(status)); | |
6589 | exit(2); | |
6590 | } | |
6591 | return (B_FALSE); | |
6592 | } else if (WIFSIGNALED(status)) { | |
6593 | if (!ignorekill || WTERMSIG(status) != SIGKILL) { | |
6594 | (void) fprintf(stderr, "child died with signal %d\n", | |
6595 | WTERMSIG(status)); | |
6596 | exit(3); | |
6597 | } | |
6598 | return (B_TRUE); | |
6599 | } else { | |
6600 | (void) fprintf(stderr, "something strange happened to child\n"); | |
6601 | exit(4); | |
6602 | /* NOTREACHED */ | |
6603 | } | |
6604 | } | |
6605 | ||
6606 | static void | |
6607 | ztest_run_init(void) | |
6608 | { | |
6609 | int i; | |
6610 | ||
6611 | ztest_shared_t *zs = ztest_shared; | |
6612 | ||
6613 | ASSERT(ztest_opts.zo_init != 0); | |
6614 | ||
6615 | /* | |
6616 | * Blow away any existing copy of zpool.cache | |
6617 | */ | |
6618 | (void) remove(spa_config_path); | |
6619 | ||
6620 | /* | |
6621 | * Create and initialize our storage pool. | |
6622 | */ | |
6623 | for (i = 1; i <= ztest_opts.zo_init; i++) { | |
6624 | bzero(zs, sizeof (ztest_shared_t)); | |
6625 | if (ztest_opts.zo_verbose >= 3 && | |
6626 | ztest_opts.zo_init != 1) { | |
6627 | (void) printf("ztest_init(), pass %d\n", i); | |
6628 | } | |
6629 | ztest_init(zs); | |
6630 | } | |
6631 | } | |
6632 | ||
6633 | int | |
6634 | main(int argc, char **argv) | |
6635 | { | |
6636 | int kills = 0; | |
6637 | int iters = 0; | |
6638 | int older = 0; | |
6639 | int newer = 0; | |
6640 | ztest_shared_t *zs; | |
6641 | ztest_info_t *zi; | |
6642 | ztest_shared_callstate_t *zc; | |
6643 | char timebuf[100]; | |
6644 | char numbuf[6]; | |
6645 | spa_t *spa; | |
6646 | char *cmd; | |
6647 | boolean_t hasalt; | |
6648 | int f; | |
6649 | char *fd_data_str = getenv("ZTEST_FD_DATA"); | |
6650 | struct sigaction action; | |
6651 | ||
6652 | (void) setvbuf(stdout, NULL, _IOLBF, 0); | |
6653 | ||
6654 | dprintf_setup(&argc, argv); | |
6655 | ||
6656 | action.sa_handler = sig_handler; | |
6657 | sigemptyset(&action.sa_mask); | |
6658 | action.sa_flags = 0; | |
6659 | ||
6660 | if (sigaction(SIGSEGV, &action, NULL) < 0) { | |
6661 | (void) fprintf(stderr, "ztest: cannot catch SIGSEGV: %s.\n", | |
6662 | strerror(errno)); | |
6663 | exit(EXIT_FAILURE); | |
6664 | } | |
6665 | ||
6666 | if (sigaction(SIGABRT, &action, NULL) < 0) { | |
6667 | (void) fprintf(stderr, "ztest: cannot catch SIGABRT: %s.\n", | |
6668 | strerror(errno)); | |
6669 | exit(EXIT_FAILURE); | |
6670 | } | |
6671 | ||
6672 | ztest_fd_rand = open("/dev/urandom", O_RDONLY); | |
6673 | ASSERT3S(ztest_fd_rand, >=, 0); | |
6674 | ||
6675 | if (!fd_data_str) { | |
6676 | process_options(argc, argv); | |
6677 | ||
6678 | setup_data_fd(); | |
6679 | setup_hdr(); | |
6680 | setup_data(); | |
6681 | bcopy(&ztest_opts, ztest_shared_opts, | |
6682 | sizeof (*ztest_shared_opts)); | |
6683 | } else { | |
6684 | ztest_fd_data = atoi(fd_data_str); | |
6685 | setup_data(); | |
6686 | bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts)); | |
6687 | } | |
6688 | ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count); | |
6689 | ||
6690 | /* Override location of zpool.cache */ | |
6691 | VERIFY(asprintf((char **)&spa_config_path, "%s/zpool.cache", | |
6692 | ztest_opts.zo_dir) != -1); | |
6693 | ||
6694 | ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t), | |
6695 | UMEM_NOFAIL); | |
6696 | zs = ztest_shared; | |
6697 | ||
6698 | if (fd_data_str) { | |
6699 | metaslab_gang_bang = ztest_opts.zo_metaslab_gang_bang; | |
6700 | metaslab_df_alloc_threshold = | |
6701 | zs->zs_metaslab_df_alloc_threshold; | |
6702 | ||
6703 | if (zs->zs_do_init) | |
6704 | ztest_run_init(); | |
6705 | else | |
6706 | ztest_run(zs); | |
6707 | exit(0); | |
6708 | } | |
6709 | ||
6710 | hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0); | |
6711 | ||
6712 | if (ztest_opts.zo_verbose >= 1) { | |
6713 | (void) printf("%llu vdevs, %d datasets, %d threads," | |
6714 | " %llu seconds...\n", | |
6715 | (u_longlong_t)ztest_opts.zo_vdevs, | |
6716 | ztest_opts.zo_datasets, | |
6717 | ztest_opts.zo_threads, | |
6718 | (u_longlong_t)ztest_opts.zo_time); | |
6719 | } | |
6720 | ||
6721 | cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); | |
6722 | (void) strlcpy(cmd, getexecname(), MAXNAMELEN); | |
6723 | ||
6724 | zs->zs_do_init = B_TRUE; | |
6725 | if (strlen(ztest_opts.zo_alt_ztest) != 0) { | |
6726 | if (ztest_opts.zo_verbose >= 1) { | |
6727 | (void) printf("Executing older ztest for " | |
6728 | "initialization: %s\n", ztest_opts.zo_alt_ztest); | |
6729 | } | |
6730 | VERIFY(!exec_child(ztest_opts.zo_alt_ztest, | |
6731 | ztest_opts.zo_alt_libpath, B_FALSE, NULL)); | |
6732 | } else { | |
6733 | VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL)); | |
6734 | } | |
6735 | zs->zs_do_init = B_FALSE; | |
6736 | ||
6737 | zs->zs_proc_start = gethrtime(); | |
6738 | zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC; | |
6739 | ||
6740 | for (f = 0; f < ZTEST_FUNCS; f++) { | |
6741 | zi = &ztest_info[f]; | |
6742 | zc = ZTEST_GET_SHARED_CALLSTATE(f); | |
6743 | if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop) | |
6744 | zc->zc_next = UINT64_MAX; | |
6745 | else | |
6746 | zc->zc_next = zs->zs_proc_start + | |
6747 | ztest_random(2 * zi->zi_interval[0] + 1); | |
6748 | } | |
6749 | ||
6750 | /* | |
6751 | * Run the tests in a loop. These tests include fault injection | |
6752 | * to verify that self-healing data works, and forced crashes | |
6753 | * to verify that we never lose on-disk consistency. | |
6754 | */ | |
6755 | while (gethrtime() < zs->zs_proc_stop) { | |
6756 | int status; | |
6757 | boolean_t killed; | |
6758 | ||
6759 | /* | |
6760 | * Initialize the workload counters for each function. | |
6761 | */ | |
6762 | for (f = 0; f < ZTEST_FUNCS; f++) { | |
6763 | zc = ZTEST_GET_SHARED_CALLSTATE(f); | |
6764 | zc->zc_count = 0; | |
6765 | zc->zc_time = 0; | |
6766 | } | |
6767 | ||
6768 | /* Set the allocation switch size */ | |
6769 | zs->zs_metaslab_df_alloc_threshold = | |
6770 | ztest_random(zs->zs_metaslab_sz / 4) + 1; | |
6771 | ||
6772 | if (!hasalt || ztest_random(2) == 0) { | |
6773 | if (hasalt && ztest_opts.zo_verbose >= 1) { | |
6774 | (void) printf("Executing newer ztest: %s\n", | |
6775 | cmd); | |
6776 | } | |
6777 | newer++; | |
6778 | killed = exec_child(cmd, NULL, B_TRUE, &status); | |
6779 | } else { | |
6780 | if (hasalt && ztest_opts.zo_verbose >= 1) { | |
6781 | (void) printf("Executing older ztest: %s\n", | |
6782 | ztest_opts.zo_alt_ztest); | |
6783 | } | |
6784 | older++; | |
6785 | killed = exec_child(ztest_opts.zo_alt_ztest, | |
6786 | ztest_opts.zo_alt_libpath, B_TRUE, &status); | |
6787 | } | |
6788 | ||
6789 | if (killed) | |
6790 | kills++; | |
6791 | iters++; | |
6792 | ||
6793 | if (ztest_opts.zo_verbose >= 1) { | |
6794 | hrtime_t now = gethrtime(); | |
6795 | ||
6796 | now = MIN(now, zs->zs_proc_stop); | |
6797 | print_time(zs->zs_proc_stop - now, timebuf); | |
6798 | nicenum(zs->zs_space, numbuf); | |
6799 | ||
6800 | (void) printf("Pass %3d, %8s, %3llu ENOSPC, " | |
6801 | "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n", | |
6802 | iters, | |
6803 | WIFEXITED(status) ? "Complete" : "SIGKILL", | |
6804 | (u_longlong_t)zs->zs_enospc_count, | |
6805 | 100.0 * zs->zs_alloc / zs->zs_space, | |
6806 | numbuf, | |
6807 | 100.0 * (now - zs->zs_proc_start) / | |
6808 | (ztest_opts.zo_time * NANOSEC), timebuf); | |
6809 | } | |
6810 | ||
6811 | if (ztest_opts.zo_verbose >= 2) { | |
6812 | (void) printf("\nWorkload summary:\n\n"); | |
6813 | (void) printf("%7s %9s %s\n", | |
6814 | "Calls", "Time", "Function"); | |
6815 | (void) printf("%7s %9s %s\n", | |
6816 | "-----", "----", "--------"); | |
6817 | for (f = 0; f < ZTEST_FUNCS; f++) { | |
6818 | zi = &ztest_info[f]; | |
6819 | zc = ZTEST_GET_SHARED_CALLSTATE(f); | |
6820 | print_time(zc->zc_time, timebuf); | |
6821 | (void) printf("%7llu %9s %s\n", | |
6822 | (u_longlong_t)zc->zc_count, timebuf, | |
6823 | zi->zi_funcname); | |
6824 | } | |
6825 | (void) printf("\n"); | |
6826 | } | |
6827 | ||
6828 | /* | |
6829 | * It's possible that we killed a child during a rename test, | |
6830 | * in which case we'll have a 'ztest_tmp' pool lying around | |
6831 | * instead of 'ztest'. Do a blind rename in case this happened. | |
6832 | */ | |
6833 | kernel_init(FREAD); | |
6834 | if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) { | |
6835 | spa_close(spa, FTAG); | |
6836 | } else { | |
6837 | char tmpname[MAXNAMELEN]; | |
6838 | kernel_fini(); | |
6839 | kernel_init(FREAD | FWRITE); | |
6840 | (void) snprintf(tmpname, sizeof (tmpname), "%s_tmp", | |
6841 | ztest_opts.zo_pool); | |
6842 | (void) spa_rename(tmpname, ztest_opts.zo_pool); | |
6843 | } | |
6844 | kernel_fini(); | |
6845 | ||
6846 | ztest_run_zdb(ztest_opts.zo_pool); | |
6847 | } | |
6848 | ||
6849 | if (ztest_opts.zo_verbose >= 1) { | |
6850 | if (hasalt) { | |
6851 | (void) printf("%d runs of older ztest: %s\n", older, | |
6852 | ztest_opts.zo_alt_ztest); | |
6853 | (void) printf("%d runs of newer ztest: %s\n", newer, | |
6854 | cmd); | |
6855 | } | |
6856 | (void) printf("%d killed, %d completed, %.0f%% kill rate\n", | |
6857 | kills, iters - kills, (100.0 * kills) / MAX(1, iters)); | |
6858 | } | |
6859 | ||
6860 | umem_free(cmd, MAXNAMELEN); | |
6861 | ||
6862 | return (0); | |
6863 | } |