Optimize RAIDZ expansion
[zfs.git] / module / zfs / zcp_synctask.c
blob058910054d97201b7cf9f038c82759bf6001c62a
1 /*
2 * CDDL HEADER START
4 * This file and its contents are supplied under the terms of the
5 * Common Development and Distribution License ("CDDL"), version 1.0.
6 * You may only use this file in accordance with the terms of version
7 * 1.0 of the CDDL.
9 * A full copy of the text of the CDDL should have accompanied this
10 * source. A copy of the CDDL is also available via the Internet at
11 * http://www.illumos.org/license/CDDL.
13 * CDDL HEADER END
17 * Copyright (c) 2016, 2017 by Delphix. All rights reserved.
18 * Copyright (c) 2019, 2020 by Christian Schwarz. All rights reserved.
19 * Copyright 2020 Joyent, Inc.
22 #include <sys/lua/lua.h>
23 #include <sys/lua/lauxlib.h>
25 #include <sys/zcp.h>
26 #include <sys/zcp_set.h>
27 #include <sys/dsl_dir.h>
28 #include <sys/dsl_pool.h>
29 #include <sys/dsl_prop.h>
30 #include <sys/dsl_synctask.h>
31 #include <sys/dsl_dataset.h>
32 #include <sys/dsl_bookmark.h>
33 #include <sys/dsl_destroy.h>
34 #include <sys/dmu_objset.h>
35 #include <sys/zfs_znode.h>
36 #include <sys/zfeature.h>
37 #include <sys/metaslab.h>
39 #define DST_AVG_BLKSHIFT 14
41 typedef struct zcp_inherit_prop_arg {
42 lua_State *zipa_state;
43 const char *zipa_prop;
44 dsl_props_set_arg_t zipa_dpsa;
45 } zcp_inherit_prop_arg_t;
47 typedef int (zcp_synctask_func_t)(lua_State *, boolean_t, nvlist_t *);
48 typedef struct zcp_synctask_info {
49 const char *name;
50 zcp_synctask_func_t *func;
51 const zcp_arg_t pargs[4];
52 const zcp_arg_t kwargs[2];
53 zfs_space_check_t space_check;
54 int blocks_modified;
55 } zcp_synctask_info_t;
57 static void
58 zcp_synctask_cleanup(void *arg)
60 fnvlist_free(arg);
64 * Generic synctask interface for channel program syncfuncs.
66 * To perform some action in syncing context, we'd generally call
67 * dsl_sync_task(), but since the Lua script is already running inside a
68 * synctask we need to leave out some actions (such as acquiring the config
69 * rwlock and performing space checks).
71 * If 'sync' is false, executes a dry run and returns the error code.
73 * If we are not running in syncing context and we are not doing a dry run
74 * (meaning we are running a zfs.sync function in open-context) then we
75 * return a Lua error.
77 * This function also handles common fatal error cases for channel program
78 * library functions. If a fatal error occurs, err_dsname will be the dataset
79 * name reported in error messages, if supplied.
81 static int
82 zcp_sync_task(lua_State *state, dsl_checkfunc_t *checkfunc,
83 dsl_syncfunc_t *syncfunc, void *arg, boolean_t sync, const char *err_dsname)
85 int err;
86 zcp_run_info_t *ri = zcp_run_info(state);
88 err = checkfunc(arg, ri->zri_tx);
89 if (!sync)
90 return (err);
92 if (!ri->zri_sync) {
93 return (luaL_error(state, "running functions from the zfs.sync "
94 "submodule requires passing sync=TRUE to "
95 "lzc_channel_program() (i.e. do not specify the \"-n\" "
96 "command line argument)"));
99 if (err == 0) {
100 syncfunc(arg, ri->zri_tx);
101 } else if (err == EIO) {
102 if (err_dsname != NULL) {
103 return (luaL_error(state,
104 "I/O error while accessing dataset '%s'",
105 err_dsname));
106 } else {
107 return (luaL_error(state,
108 "I/O error while accessing dataset."));
112 return (err);
116 static int zcp_synctask_destroy(lua_State *, boolean_t, nvlist_t *);
117 static const zcp_synctask_info_t zcp_synctask_destroy_info = {
118 .name = "destroy",
119 .func = zcp_synctask_destroy,
120 .pargs = {
121 {.za_name = "filesystem | snapshot", .za_lua_type = LUA_TSTRING },
122 {NULL, 0}
124 .kwargs = {
125 {.za_name = "defer", .za_lua_type = LUA_TBOOLEAN },
126 {NULL, 0}
128 .space_check = ZFS_SPACE_CHECK_DESTROY,
129 .blocks_modified = 0
132 static int
133 zcp_synctask_destroy(lua_State *state, boolean_t sync, nvlist_t *err_details)
135 (void) err_details;
136 int err;
137 const char *dsname = lua_tostring(state, 1);
139 boolean_t issnap = (strchr(dsname, '@') != NULL);
141 if (!issnap && !lua_isnil(state, 2)) {
142 return (luaL_error(state,
143 "'deferred' kwarg only supported for snapshots: %s",
144 dsname));
147 if (issnap) {
148 dsl_destroy_snapshot_arg_t ddsa = { 0 };
149 ddsa.ddsa_name = dsname;
150 if (!lua_isnil(state, 2)) {
151 ddsa.ddsa_defer = lua_toboolean(state, 2);
152 } else {
153 ddsa.ddsa_defer = B_FALSE;
156 err = zcp_sync_task(state, dsl_destroy_snapshot_check,
157 dsl_destroy_snapshot_sync, &ddsa, sync, dsname);
158 } else {
159 dsl_destroy_head_arg_t ddha = { 0 };
160 ddha.ddha_name = dsname;
162 err = zcp_sync_task(state, dsl_destroy_head_check,
163 dsl_destroy_head_sync, &ddha, sync, dsname);
166 return (err);
169 static int zcp_synctask_promote(lua_State *, boolean_t, nvlist_t *);
170 static const zcp_synctask_info_t zcp_synctask_promote_info = {
171 .name = "promote",
172 .func = zcp_synctask_promote,
173 .pargs = {
174 {.za_name = "clone", .za_lua_type = LUA_TSTRING },
175 {NULL, 0}
177 .kwargs = {
178 {NULL, 0}
180 .space_check = ZFS_SPACE_CHECK_RESERVED,
181 .blocks_modified = 3
184 static int
185 zcp_synctask_promote(lua_State *state, boolean_t sync, nvlist_t *err_details)
187 int err;
188 dsl_dataset_promote_arg_t ddpa = { 0 };
189 const char *dsname = lua_tostring(state, 1);
190 zcp_run_info_t *ri = zcp_run_info(state);
192 ddpa.ddpa_clonename = dsname;
193 ddpa.err_ds = err_details;
194 ddpa.cr = ri->zri_cred;
195 ddpa.proc = ri->zri_proc;
198 * If there was a snapshot name conflict, then err_ds will be filled
199 * with a list of conflicting snapshot names.
201 err = zcp_sync_task(state, dsl_dataset_promote_check,
202 dsl_dataset_promote_sync, &ddpa, sync, dsname);
204 return (err);
207 static int zcp_synctask_rollback(lua_State *, boolean_t, nvlist_t *err_details);
208 static const zcp_synctask_info_t zcp_synctask_rollback_info = {
209 .name = "rollback",
210 .func = zcp_synctask_rollback,
211 .space_check = ZFS_SPACE_CHECK_RESERVED,
212 .blocks_modified = 1,
213 .pargs = {
214 {.za_name = "filesystem", .za_lua_type = LUA_TSTRING },
215 {0, 0}
217 .kwargs = {
218 {0, 0}
222 static int
223 zcp_synctask_rollback(lua_State *state, boolean_t sync, nvlist_t *err_details)
225 int err;
226 const char *dsname = lua_tostring(state, 1);
227 dsl_dataset_rollback_arg_t ddra = { 0 };
229 ddra.ddra_fsname = dsname;
230 ddra.ddra_result = err_details;
232 err = zcp_sync_task(state, dsl_dataset_rollback_check,
233 dsl_dataset_rollback_sync, &ddra, sync, dsname);
235 return (err);
238 static int zcp_synctask_snapshot(lua_State *, boolean_t, nvlist_t *);
239 static const zcp_synctask_info_t zcp_synctask_snapshot_info = {
240 .name = "snapshot",
241 .func = zcp_synctask_snapshot,
242 .pargs = {
243 {.za_name = "filesystem@snapname | volume@snapname",
244 .za_lua_type = LUA_TSTRING },
245 {NULL, 0}
247 .kwargs = {
248 {NULL, 0}
250 .space_check = ZFS_SPACE_CHECK_NORMAL,
251 .blocks_modified = 3
254 static int
255 zcp_synctask_snapshot(lua_State *state, boolean_t sync, nvlist_t *err_details)
257 (void) err_details;
258 int err;
259 dsl_dataset_snapshot_arg_t ddsa = { 0 };
260 const char *dsname = lua_tostring(state, 1);
261 zcp_run_info_t *ri = zcp_run_info(state);
264 * On old pools, the ZIL must not be active when a snapshot is created,
265 * but we can't suspend the ZIL because we're already in syncing
266 * context.
268 if (spa_version(ri->zri_pool->dp_spa) < SPA_VERSION_FAST_SNAP) {
269 return (SET_ERROR(ENOTSUP));
273 * We only allow for a single snapshot rather than a list, so the
274 * error list output is unnecessary.
276 ddsa.ddsa_errors = NULL;
277 ddsa.ddsa_props = NULL;
278 ddsa.ddsa_cr = ri->zri_cred;
279 ddsa.ddsa_proc = ri->zri_proc;
280 ddsa.ddsa_snaps = fnvlist_alloc();
281 fnvlist_add_boolean(ddsa.ddsa_snaps, dsname);
283 zcp_cleanup_handler_t *zch = zcp_register_cleanup(state,
284 zcp_synctask_cleanup, ddsa.ddsa_snaps);
286 err = zcp_sync_task(state, dsl_dataset_snapshot_check,
287 dsl_dataset_snapshot_sync, &ddsa, sync, dsname);
289 if (err == 0) {
291 * We may need to create a new device minor node for this
292 * dataset (if it is a zvol and the "snapdev" property is set).
293 * Save it in the nvlist so that it can be processed in open
294 * context.
296 fnvlist_add_boolean(ri->zri_new_zvols, dsname);
299 zcp_deregister_cleanup(state, zch);
300 fnvlist_free(ddsa.ddsa_snaps);
302 return (err);
305 static int zcp_synctask_rename_snapshot(lua_State *, boolean_t, nvlist_t *);
306 static const zcp_synctask_info_t zcp_synctask_rename_snapshot_info = {
307 .name = "rename_snapshot",
308 .func = zcp_synctask_rename_snapshot,
309 .pargs = {
310 {.za_name = "filesystem | volume", .za_lua_type = LUA_TSTRING },
311 {.za_name = "oldsnapname", .za_lua_type = LUA_TSTRING },
312 {.za_name = "newsnapname", .za_lua_type = LUA_TSTRING },
313 {NULL, 0}
315 .space_check = ZFS_SPACE_CHECK_RESERVED,
316 .blocks_modified = 1
319 static int
320 zcp_synctask_rename_snapshot(lua_State *state, boolean_t sync,
321 nvlist_t *err_details)
323 (void) err_details;
324 int err;
325 const char *fsname = lua_tostring(state, 1);
326 const char *oldsnapname = lua_tostring(state, 2);
327 const char *newsnapname = lua_tostring(state, 3);
329 struct dsl_dataset_rename_snapshot_arg ddrsa = { 0 };
330 ddrsa.ddrsa_fsname = fsname;
331 ddrsa.ddrsa_oldsnapname = oldsnapname;
332 ddrsa.ddrsa_newsnapname = newsnapname;
333 ddrsa.ddrsa_recursive = B_FALSE;
335 err = zcp_sync_task(state, dsl_dataset_rename_snapshot_check,
336 dsl_dataset_rename_snapshot_sync, &ddrsa, sync, NULL);
338 return (err);
341 static int zcp_synctask_inherit_prop(lua_State *, boolean_t,
342 nvlist_t *err_details);
343 static const zcp_synctask_info_t zcp_synctask_inherit_prop_info = {
344 .name = "inherit",
345 .func = zcp_synctask_inherit_prop,
346 .space_check = ZFS_SPACE_CHECK_RESERVED,
347 .blocks_modified = 2, /* 2 * numprops */
348 .pargs = {
349 { .za_name = "dataset", .za_lua_type = LUA_TSTRING },
350 { .za_name = "property", .za_lua_type = LUA_TSTRING },
351 { NULL, 0 }
353 .kwargs = {
354 { NULL, 0 }
358 static int
359 zcp_synctask_inherit_prop_check(void *arg, dmu_tx_t *tx)
361 zcp_inherit_prop_arg_t *args = arg;
362 zfs_prop_t prop = zfs_name_to_prop(args->zipa_prop);
364 if (prop == ZPROP_USERPROP) {
365 if (zfs_prop_user(args->zipa_prop))
366 return (0);
368 return (EINVAL);
371 if (zfs_prop_readonly(prop))
372 return (EINVAL);
374 if (!zfs_prop_inheritable(prop))
375 return (EINVAL);
377 return (dsl_props_set_check(&args->zipa_dpsa, tx));
380 static void
381 zcp_synctask_inherit_prop_sync(void *arg, dmu_tx_t *tx)
383 zcp_inherit_prop_arg_t *args = arg;
384 dsl_props_set_arg_t *dpsa = &args->zipa_dpsa;
386 dsl_props_set_sync(dpsa, tx);
389 static int
390 zcp_synctask_inherit_prop(lua_State *state, boolean_t sync,
391 nvlist_t *err_details)
393 (void) err_details;
394 int err;
395 zcp_inherit_prop_arg_t zipa = { 0 };
396 dsl_props_set_arg_t *dpsa = &zipa.zipa_dpsa;
398 const char *dsname = lua_tostring(state, 1);
399 const char *prop = lua_tostring(state, 2);
401 zipa.zipa_state = state;
402 zipa.zipa_prop = prop;
403 dpsa->dpsa_dsname = dsname;
404 dpsa->dpsa_source = ZPROP_SRC_INHERITED;
405 dpsa->dpsa_props = fnvlist_alloc();
406 fnvlist_add_boolean(dpsa->dpsa_props, prop);
408 zcp_cleanup_handler_t *zch = zcp_register_cleanup(state,
409 zcp_synctask_cleanup, dpsa->dpsa_props);
411 err = zcp_sync_task(state, zcp_synctask_inherit_prop_check,
412 zcp_synctask_inherit_prop_sync, &zipa, sync, dsname);
414 zcp_deregister_cleanup(state, zch);
415 fnvlist_free(dpsa->dpsa_props);
417 return (err);
420 static int zcp_synctask_bookmark(lua_State *, boolean_t, nvlist_t *);
421 static const zcp_synctask_info_t zcp_synctask_bookmark_info = {
422 .name = "bookmark",
423 .func = zcp_synctask_bookmark,
424 .pargs = {
425 {.za_name = "snapshot | bookmark", .za_lua_type = LUA_TSTRING },
426 {.za_name = "bookmark", .za_lua_type = LUA_TSTRING },
427 {NULL, 0}
429 .kwargs = {
430 {NULL, 0}
432 .space_check = ZFS_SPACE_CHECK_NORMAL,
433 .blocks_modified = 1,
436 static int
437 zcp_synctask_bookmark(lua_State *state, boolean_t sync, nvlist_t *err_details)
439 (void) err_details;
440 int err;
441 const char *source = lua_tostring(state, 1);
442 const char *new = lua_tostring(state, 2);
444 nvlist_t *bmarks = fnvlist_alloc();
445 fnvlist_add_string(bmarks, new, source);
447 zcp_cleanup_handler_t *zch = zcp_register_cleanup(state,
448 zcp_synctask_cleanup, bmarks);
450 dsl_bookmark_create_arg_t dbca = {
451 .dbca_bmarks = bmarks,
452 .dbca_errors = NULL,
454 err = zcp_sync_task(state, dsl_bookmark_create_check,
455 dsl_bookmark_create_sync, &dbca, sync, source);
457 zcp_deregister_cleanup(state, zch);
458 fnvlist_free(bmarks);
460 return (err);
463 static int zcp_synctask_set_prop(lua_State *, boolean_t, nvlist_t *err_details);
464 static const zcp_synctask_info_t zcp_synctask_set_prop_info = {
465 .name = "set_prop",
466 .func = zcp_synctask_set_prop,
467 .space_check = ZFS_SPACE_CHECK_RESERVED,
468 .blocks_modified = 2,
469 .pargs = {
470 { .za_name = "dataset", .za_lua_type = LUA_TSTRING },
471 { .za_name = "property", .za_lua_type = LUA_TSTRING },
472 { .za_name = "value", .za_lua_type = LUA_TSTRING },
473 { NULL, 0 }
475 .kwargs = {
476 { NULL, 0 }
480 static int
481 zcp_synctask_set_prop(lua_State *state, boolean_t sync, nvlist_t *err_details)
483 (void) err_details;
484 int err;
485 zcp_set_prop_arg_t args = { 0 };
487 const char *dsname = lua_tostring(state, 1);
488 const char *prop = lua_tostring(state, 2);
489 const char *val = lua_tostring(state, 3);
491 args.state = state;
492 args.dsname = dsname;
493 args.prop = prop;
494 args.val = val;
496 err = zcp_sync_task(state, zcp_set_prop_check, zcp_set_prop_sync,
497 &args, sync, dsname);
499 return (err);
502 static int
503 zcp_synctask_wrapper(lua_State *state)
505 int err;
506 zcp_cleanup_handler_t *zch;
507 int num_ret = 1;
508 nvlist_t *err_details = fnvlist_alloc();
511 * Make sure err_details is properly freed, even if a fatal error is
512 * thrown during the synctask.
514 zch = zcp_register_cleanup(state, zcp_synctask_cleanup, err_details);
516 zcp_synctask_info_t *info = lua_touserdata(state, lua_upvalueindex(1));
517 boolean_t sync = lua_toboolean(state, lua_upvalueindex(2));
519 zcp_run_info_t *ri = zcp_run_info(state);
520 dsl_pool_t *dp = ri->zri_pool;
522 /* MOS space is triple-dittoed, so we multiply by 3. */
523 uint64_t funcspace =
524 ((uint64_t)info->blocks_modified << DST_AVG_BLKSHIFT) * 3;
526 zcp_parse_args(state, info->name, info->pargs, info->kwargs);
528 err = 0;
529 if (info->space_check != ZFS_SPACE_CHECK_NONE) {
530 uint64_t quota = dsl_pool_unreserved_space(dp,
531 info->space_check);
532 uint64_t used = dsl_dir_phys(dp->dp_root_dir)->dd_used_bytes +
533 ri->zri_space_used;
535 if (used + funcspace > quota) {
536 err = SET_ERROR(ENOSPC);
540 if (err == 0) {
541 err = info->func(state, sync, err_details);
544 if (err == 0) {
545 ri->zri_space_used += funcspace;
548 lua_pushnumber(state, (lua_Number)err);
549 if (fnvlist_num_pairs(err_details) > 0) {
550 (void) zcp_nvlist_to_lua(state, err_details, NULL, 0);
551 num_ret++;
554 zcp_deregister_cleanup(state, zch);
555 fnvlist_free(err_details);
557 return (num_ret);
561 zcp_load_synctask_lib(lua_State *state, boolean_t sync)
563 const zcp_synctask_info_t *zcp_synctask_funcs[] = {
564 &zcp_synctask_destroy_info,
565 &zcp_synctask_promote_info,
566 &zcp_synctask_rollback_info,
567 &zcp_synctask_snapshot_info,
568 &zcp_synctask_rename_snapshot_info,
569 &zcp_synctask_inherit_prop_info,
570 &zcp_synctask_bookmark_info,
571 &zcp_synctask_set_prop_info,
572 NULL
575 lua_newtable(state);
577 for (int i = 0; zcp_synctask_funcs[i] != NULL; i++) {
578 const zcp_synctask_info_t *info = zcp_synctask_funcs[i];
579 lua_pushlightuserdata(state, (void *)(uintptr_t)info);
580 lua_pushboolean(state, sync);
581 lua_pushcclosure(state, &zcp_synctask_wrapper, 2);
582 lua_setfield(state, -2, info->name);
585 return (1);