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
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.
17 * Copyright (c) 2016, 2018 by Delphix. All rights reserved.
21 * ZFS Channel Programs (ZCP)
23 * The ZCP interface allows various ZFS commands and operations ZFS
24 * administrative operations (e.g. creating and destroying snapshots, typically
25 * performed via an ioctl to /dev/zfs by the zfs(8) command and
26 * libzfs/libzfs_core) to be run * programmatically as a Lua script. A ZCP
27 * script is run as a dsl_sync_task and fully executed during one transaction
28 * group sync. This ensures that no other changes can be written concurrently
29 * with a running Lua script. Combining multiple calls to the exposed ZFS
30 * functions into one script gives a number of benefits:
32 * 1. Atomicity. For some compound or iterative operations, it's useful to be
33 * able to guarantee that the state of a pool has not changed between calls to
36 * 2. Performance. If a large number of changes need to be made (e.g. deleting
37 * many filesystems), there can be a significant performance penalty as a
38 * result of the need to wait for a transaction group sync to pass for every
39 * single operation. When expressed as a single ZCP script, all these changes
40 * can be performed at once in one txg sync.
42 * A modified version of the Lua 5.2 interpreter is used to run channel program
43 * scripts. The Lua 5.2 manual can be found at:
45 * http://www.lua.org/manual/5.2/
47 * If being run by a user (via an ioctl syscall), executing a ZCP script
48 * requires root privileges in the global zone.
50 * Scripts are passed to zcp_eval() as a string, then run in a synctask by
51 * zcp_eval_sync(). Arguments can be passed into the Lua script as an nvlist,
52 * which will be converted to a Lua table. Similarly, values returned from
53 * a ZCP script will be converted to an nvlist. See zcp_lua_to_nvlist_impl()
54 * for details on exact allowed types and conversion.
56 * ZFS functionality is exposed to a ZCP script as a library of function calls.
57 * These calls are sorted into submodules, such as zfs.list and zfs.sync, for
58 * iterators and synctasks, respectively. Each of these submodules resides in
59 * its own source file, with a zcp_*_info structure describing each library
60 * call in the submodule.
62 * Error handling in ZCP scripts is handled by a number of different methods
65 * 1. Memory and time limits are in place to prevent a channel program from
66 * consuming excessive system or running forever. If one of these limits is
67 * hit, the channel program will be stopped immediately and return from
68 * zcp_eval() with an error code. No attempt will be made to roll back or undo
69 * any changes made by the channel program before the error occurred.
70 * Consumers invoking zcp_eval() from elsewhere in the kernel may pass a time
71 * limit of 0, disabling the time limit.
73 * 2. Internal Lua errors can occur as a result of a syntax error, calling a
74 * library function with incorrect arguments, invoking the error() function,
75 * failing an assert(), or other runtime errors. In these cases the channel
76 * program will stop executing and return from zcp_eval() with an error code.
77 * In place of a return value, an error message will also be returned in the
78 * 'result' nvlist containing information about the error. No attempt will be
79 * made to roll back or undo any changes made by the channel program before the
82 * 3. If an error occurs inside a ZFS library call which returns an error code,
83 * the error is returned to the Lua script to be handled as desired.
85 * In the first two cases, Lua's error-throwing mechanism is used, which
86 * longjumps out of the script execution with luaL_error() and returns with the
89 * See zfs-program(8) for more information on high level usage.
92 #include <sys/lua/lua.h>
93 #include <sys/lua/lualib.h>
94 #include <sys/lua/lauxlib.h>
96 #include <sys/dsl_prop.h>
97 #include <sys/dsl_synctask.h>
98 #include <sys/dsl_dataset.h>
100 #include <sys/zcp_iter.h>
101 #include <sys/zcp_prop.h>
102 #include <sys/zcp_global.h>
103 #include <sys/zvol.h>
106 #define KM_NORMALPRI 0
109 #define ZCP_NVLIST_MAX_DEPTH 20
111 static const uint64_t zfs_lua_check_instrlimit_interval
= 100;
112 uint64_t zfs_lua_max_instrlimit
= ZCP_MAX_INSTRLIMIT
;
113 uint64_t zfs_lua_max_memlimit
= ZCP_MAX_MEMLIMIT
;
116 * Forward declarations for mutually recursive functions
118 static int zcp_nvpair_value_to_lua(lua_State
*, nvpair_t
*, char *, int);
119 static int zcp_lua_to_nvlist_impl(lua_State
*, int, nvlist_t
*, const char *,
123 * The outer-most error callback handler for use with lua_pcall(). On
124 * error Lua will call this callback with a single argument that
125 * represents the error value. In most cases this will be a string
126 * containing an error message, but channel programs can use Lua's
127 * error() function to return arbitrary objects as errors. This callback
128 * returns (on the Lua stack) the original error object along with a traceback.
130 * Fatal Lua errors can occur while resources are held, so we also call any
131 * registered cleanup function here.
134 zcp_error_handler(lua_State
*state
)
140 VERIFY3U(1, ==, lua_gettop(state
));
141 msg
= lua_tostring(state
, 1);
142 luaL_traceback(state
, state
, msg
, 1);
147 zcp_argerror(lua_State
*state
, int narg
, const char *msg
, ...)
151 va_start(alist
, msg
);
152 const char *buf
= lua_pushvfstring(state
, msg
, alist
);
155 return (luaL_argerror(state
, narg
, buf
));
159 * Install a new cleanup function, which will be invoked with the given
160 * opaque argument if a fatal error causes the Lua interpreter to longjump out
161 * of a function call.
163 * If an error occurs, the cleanup function will be invoked exactly once and
166 * Returns the registered cleanup handler so the caller can deregister it
167 * if no error occurs.
169 zcp_cleanup_handler_t
*
170 zcp_register_cleanup(lua_State
*state
, zcp_cleanup_t cleanfunc
, void *cleanarg
)
172 zcp_run_info_t
*ri
= zcp_run_info(state
);
174 zcp_cleanup_handler_t
*zch
= kmem_alloc(sizeof (*zch
), KM_SLEEP
);
175 zch
->zch_cleanup_func
= cleanfunc
;
176 zch
->zch_cleanup_arg
= cleanarg
;
177 list_insert_head(&ri
->zri_cleanup_handlers
, zch
);
183 zcp_deregister_cleanup(lua_State
*state
, zcp_cleanup_handler_t
*zch
)
185 zcp_run_info_t
*ri
= zcp_run_info(state
);
186 list_remove(&ri
->zri_cleanup_handlers
, zch
);
187 kmem_free(zch
, sizeof (*zch
));
191 * Execute the currently registered cleanup handlers then free them and
192 * destroy the handler list.
195 zcp_cleanup(lua_State
*state
)
197 zcp_run_info_t
*ri
= zcp_run_info(state
);
199 for (zcp_cleanup_handler_t
*zch
=
200 list_remove_head(&ri
->zri_cleanup_handlers
); zch
!= NULL
;
201 zch
= list_remove_head(&ri
->zri_cleanup_handlers
)) {
202 zch
->zch_cleanup_func(zch
->zch_cleanup_arg
);
203 kmem_free(zch
, sizeof (*zch
));
208 * Convert the lua table at the given index on the Lua stack to an nvlist
211 * If the table can not be converted for any reason, NULL is returned and
212 * an error message is pushed onto the Lua stack.
215 zcp_table_to_nvlist(lua_State
*state
, int index
, int depth
)
219 * Converting a Lua table to an nvlist with key uniqueness checking is
220 * O(n^2) in the number of keys in the nvlist, which can take a long
221 * time when we return a large table from a channel program.
222 * Furthermore, Lua's table interface *almost* guarantees unique keys
223 * on its own (details below). Therefore, we don't use fnvlist_alloc()
224 * here to avoid the built-in uniqueness checking.
226 * The *almost* is because it's possible to have key collisions between
227 * e.g. the string "1" and the number 1, or the string "true" and the
228 * boolean true, so we explicitly check that when we're looking at a
229 * key which is an integer / boolean or a string that can be parsed as
230 * one of those types. In the worst case this could still devolve into
231 * O(n^2), so we only start doing these checks on boolean/integer keys
232 * once we've seen a string key which fits this weird usage pattern.
234 * Ultimately, we still want callers to know that the keys in this
235 * nvlist are unique, so before we return this we set the nvlist's
236 * flags to reflect that.
238 VERIFY0(nvlist_alloc(&nvl
, 0, KM_SLEEP
));
241 * Push an empty stack slot where lua_next() will store each
245 boolean_t saw_str_could_collide
= B_FALSE
;
246 while (lua_next(state
, index
) != 0) {
248 * The next key-value pair from the table at index is
249 * now on the stack, with the key at stack slot -2 and
250 * the value at slot -1.
254 const char *key
= NULL
;
255 boolean_t key_could_collide
= B_FALSE
;
257 switch (lua_type(state
, -2)) {
259 key
= lua_tostring(state
, -2);
261 /* check if this could collide with a number or bool */
264 if ((sscanf(key
, "%lld%n", &tmp
, &parselen
) > 0 &&
265 parselen
== strlen(key
)) ||
266 strcmp(key
, "true") == 0 ||
267 strcmp(key
, "false") == 0) {
268 key_could_collide
= B_TRUE
;
269 saw_str_could_collide
= B_TRUE
;
273 key
= (lua_toboolean(state
, -2) == B_TRUE
?
275 if (saw_str_could_collide
) {
276 key_could_collide
= B_TRUE
;
280 (void) snprintf(buf
, sizeof (buf
), "%lld",
281 (longlong_t
)lua_tonumber(state
, -2));
284 if (saw_str_could_collide
) {
285 key_could_collide
= B_TRUE
;
290 (void) lua_pushfstring(state
, "Invalid key "
291 "type '%s' in table",
292 lua_typename(state
, lua_type(state
, -2)));
296 * Check for type-mismatched key collisions, and throw an error.
298 if (key_could_collide
&& nvlist_exists(nvl
, key
)) {
300 (void) lua_pushfstring(state
, "Collision of "
301 "key '%s' in table", key
);
305 * Recursively convert the table value and insert into
306 * the new nvlist with the parsed key. To prevent
307 * stack overflow on circular or heavily nested tables,
308 * we track the current nvlist depth.
310 if (depth
>= ZCP_NVLIST_MAX_DEPTH
) {
312 (void) lua_pushfstring(state
, "Maximum table "
313 "depth (%d) exceeded for table",
314 ZCP_NVLIST_MAX_DEPTH
);
317 err
= zcp_lua_to_nvlist_impl(state
, -1, nvl
, key
,
322 * Error message has been pushed to the lua
323 * stack by the recursive call.
328 * Pop the value pushed by lua_next().
334 * Mark the nvlist as having unique keys. This is a little ugly, but we
335 * ensured above that there are no duplicate keys in the nvlist.
337 nvl
->nvl_nvflag
|= NV_UNIQUE_NAME
;
343 * Convert a value from the given index into the lua stack to an nvpair, adding
344 * it to an nvlist with the given key.
346 * Values are converted as follows:
351 * nil -> boolean (no value)
353 * Lua tables are converted to nvlists and then inserted. The table's keys
354 * are converted to strings then used as keys in the nvlist to store each table
355 * element. Keys are converted as follows:
357 * string -> no change
359 * boolean -> "true" | "false"
362 * In the case of a key collision, an error is thrown.
364 * If an error is encountered, a nonzero error code is returned, and an error
365 * string will be pushed onto the Lua stack.
368 zcp_lua_to_nvlist_impl(lua_State
*state
, int index
, nvlist_t
*nvl
,
369 const char *key
, int depth
)
372 * Verify that we have enough remaining space in the lua stack to parse
373 * a key-value pair and push an error.
375 if (!lua_checkstack(state
, 3)) {
376 (void) lua_pushstring(state
, "Lua stack overflow");
380 index
= lua_absindex(state
, index
);
382 switch (lua_type(state
, index
)) {
384 fnvlist_add_boolean(nvl
, key
);
387 fnvlist_add_boolean_value(nvl
, key
,
388 lua_toboolean(state
, index
));
391 fnvlist_add_int64(nvl
, key
, lua_tonumber(state
, index
));
394 fnvlist_add_string(nvl
, key
, lua_tostring(state
, index
));
397 nvlist_t
*value_nvl
= zcp_table_to_nvlist(state
, index
, depth
);
398 if (value_nvl
== NULL
)
399 return (SET_ERROR(EINVAL
));
401 fnvlist_add_nvlist(nvl
, key
, value_nvl
);
402 fnvlist_free(value_nvl
);
406 (void) lua_pushfstring(state
,
407 "Invalid value type '%s' for key '%s'",
408 lua_typename(state
, lua_type(state
, index
)), key
);
409 return (SET_ERROR(EINVAL
));
416 * Convert a lua value to an nvpair, adding it to an nvlist with the given key.
419 zcp_lua_to_nvlist(lua_State
*state
, int index
, nvlist_t
*nvl
, const char *key
)
422 * On error, zcp_lua_to_nvlist_impl pushes an error string onto the Lua
423 * stack before returning with a nonzero error code. If an error is
424 * returned, throw a fatal lua error with the given string.
426 if (zcp_lua_to_nvlist_impl(state
, index
, nvl
, key
, 0) != 0)
427 (void) lua_error(state
);
431 zcp_lua_to_nvlist_helper(lua_State
*state
)
433 nvlist_t
*nv
= (nvlist_t
*)lua_touserdata(state
, 2);
434 const char *key
= (const char *)lua_touserdata(state
, 1);
435 zcp_lua_to_nvlist(state
, 3, nv
, key
);
440 zcp_convert_return_values(lua_State
*state
, nvlist_t
*nvl
,
441 const char *key
, int *result
)
444 VERIFY3U(1, ==, lua_gettop(state
));
445 lua_pushcfunction(state
, zcp_lua_to_nvlist_helper
);
446 lua_pushlightuserdata(state
, (char *)key
);
447 lua_pushlightuserdata(state
, nvl
);
448 lua_pushvalue(state
, 1);
449 lua_remove(state
, 1);
450 err
= lua_pcall(state
, 3, 0, 0); /* zcp_lua_to_nvlist_helper */
452 zcp_lua_to_nvlist(state
, 1, nvl
, ZCP_RET_ERROR
);
453 *result
= SET_ERROR(ECHRNG
);
458 * Push a Lua table representing nvl onto the stack. If it can't be
459 * converted, return EINVAL, fill in errbuf, and push nothing. errbuf may
460 * be specified as NULL, in which case no error string will be output.
462 * Most nvlists are converted as simple key->value Lua tables, but we make
463 * an exception for the case where all nvlist entries are BOOLEANs (a string
464 * key without a value). In Lua, a table key pointing to a value of Nil
465 * (no value) is equivalent to the key not existing, so a BOOLEAN nvlist
466 * entry can't be directly converted to a Lua table entry. Nvlists of entirely
467 * BOOLEAN entries are frequently used to pass around lists of datasets, so for
468 * convenience we check for this case, and convert it to a simple Lua array of
472 zcp_nvlist_to_lua(lua_State
*state
, nvlist_t
*nvl
,
473 char *errbuf
, int errbuf_len
)
477 boolean_t has_values
= B_FALSE
;
479 * If the list doesn't have any values, just convert it to a string
482 for (pair
= nvlist_next_nvpair(nvl
, NULL
);
483 pair
!= NULL
; pair
= nvlist_next_nvpair(nvl
, pair
)) {
484 if (nvpair_type(pair
) != DATA_TYPE_BOOLEAN
) {
491 for (pair
= nvlist_next_nvpair(nvl
, NULL
);
492 pair
!= NULL
; pair
= nvlist_next_nvpair(nvl
, pair
)) {
493 (void) lua_pushinteger(state
, i
);
494 (void) lua_pushstring(state
, nvpair_name(pair
));
495 (void) lua_settable(state
, -3);
499 for (pair
= nvlist_next_nvpair(nvl
, NULL
);
500 pair
!= NULL
; pair
= nvlist_next_nvpair(nvl
, pair
)) {
501 int err
= zcp_nvpair_value_to_lua(state
, pair
,
507 (void) lua_setfield(state
, -2, nvpair_name(pair
));
514 * Push a Lua object representing the value of "pair" onto the stack.
516 * Only understands boolean_value, string, int64, nvlist,
517 * string_array, and int64_array type values. For other
518 * types, returns EINVAL, fills in errbuf, and pushes nothing.
521 zcp_nvpair_value_to_lua(lua_State
*state
, nvpair_t
*pair
,
522 char *errbuf
, int errbuf_len
)
531 switch (nvpair_type(pair
)) {
532 case DATA_TYPE_BOOLEAN_VALUE
:
533 (void) lua_pushboolean(state
,
534 fnvpair_value_boolean_value(pair
));
536 case DATA_TYPE_STRING
:
537 (void) lua_pushstring(state
, fnvpair_value_string(pair
));
539 case DATA_TYPE_INT64
:
540 (void) lua_pushinteger(state
, fnvpair_value_int64(pair
));
542 case DATA_TYPE_NVLIST
:
543 err
= zcp_nvlist_to_lua(state
,
544 fnvpair_value_nvlist(pair
), errbuf
, errbuf_len
);
546 case DATA_TYPE_STRING_ARRAY
: {
549 (void) nvpair_value_string_array(pair
, &strarr
, &nelem
);
551 for (int i
= 0; i
< nelem
; i
++) {
552 (void) lua_pushinteger(state
, i
+ 1);
553 (void) lua_pushstring(state
, strarr
[i
]);
554 (void) lua_settable(state
, -3);
558 case DATA_TYPE_UINT64_ARRAY
: {
561 (void) nvpair_value_uint64_array(pair
, &intarr
, &nelem
);
563 for (int i
= 0; i
< nelem
; i
++) {
564 (void) lua_pushinteger(state
, i
+ 1);
565 (void) lua_pushinteger(state
, intarr
[i
]);
566 (void) lua_settable(state
, -3);
570 case DATA_TYPE_INT64_ARRAY
: {
573 (void) nvpair_value_int64_array(pair
, &intarr
, &nelem
);
575 for (int i
= 0; i
< nelem
; i
++) {
576 (void) lua_pushinteger(state
, i
+ 1);
577 (void) lua_pushinteger(state
, intarr
[i
]);
578 (void) lua_settable(state
, -3);
583 if (errbuf
!= NULL
) {
584 (void) snprintf(errbuf
, errbuf_len
,
585 "Unhandled nvpair type %d for key '%s'",
586 nvpair_type(pair
), nvpair_name(pair
));
588 return (SET_ERROR(EINVAL
));
595 zcp_dataset_hold_error(lua_State
*state
, dsl_pool_t
*dp
, const char *dsname
,
598 if (error
== ENOENT
) {
599 (void) zcp_argerror(state
, 1, "no such dataset '%s'", dsname
);
600 return (0); /* not reached; zcp_argerror will longjmp */
601 } else if (error
== EXDEV
) {
602 (void) zcp_argerror(state
, 1,
603 "dataset '%s' is not in the target pool '%s'",
604 dsname
, spa_name(dp
->dp_spa
));
605 return (0); /* not reached; zcp_argerror will longjmp */
606 } else if (error
== EIO
) {
607 (void) luaL_error(state
,
608 "I/O error while accessing dataset '%s'", dsname
);
609 return (0); /* not reached; luaL_error will longjmp */
610 } else if (error
!= 0) {
611 (void) luaL_error(state
,
612 "unexpected error %d while accessing dataset '%s'",
614 return (0); /* not reached; luaL_error will longjmp */
620 * Note: will longjmp (via lua_error()) on error.
621 * Assumes that the dsname is argument #1 (for error reporting purposes).
624 zcp_dataset_hold(lua_State
*state
, dsl_pool_t
*dp
, const char *dsname
,
628 int error
= dsl_dataset_hold(dp
, dsname
, tag
, &ds
);
629 (void) zcp_dataset_hold_error(state
, dp
, dsname
, error
);
633 static int zcp_debug(lua_State
*);
634 static const zcp_lib_info_t zcp_debug_info
= {
638 { .za_name
= "debug string", .za_lua_type
= LUA_TSTRING
},
647 zcp_debug(lua_State
*state
)
649 const char *dbgstring
;
650 zcp_run_info_t
*ri
= zcp_run_info(state
);
651 const zcp_lib_info_t
*libinfo
= &zcp_debug_info
;
653 zcp_parse_args(state
, libinfo
->name
, libinfo
->pargs
, libinfo
->kwargs
);
655 dbgstring
= lua_tostring(state
, 1);
657 zfs_dbgmsg("txg %lld ZCP: %s", (longlong_t
)ri
->zri_tx
->tx_txg
,
663 static int zcp_exists(lua_State
*);
664 static const zcp_lib_info_t zcp_exists_info
= {
668 { .za_name
= "dataset", .za_lua_type
= LUA_TSTRING
},
677 zcp_exists(lua_State
*state
)
679 zcp_run_info_t
*ri
= zcp_run_info(state
);
680 dsl_pool_t
*dp
= ri
->zri_pool
;
681 const zcp_lib_info_t
*libinfo
= &zcp_exists_info
;
683 zcp_parse_args(state
, libinfo
->name
, libinfo
->pargs
, libinfo
->kwargs
);
685 const char *dsname
= lua_tostring(state
, 1);
688 int error
= dsl_dataset_hold(dp
, dsname
, FTAG
, &ds
);
690 dsl_dataset_rele(ds
, FTAG
);
691 lua_pushboolean(state
, B_TRUE
);
692 } else if (error
== ENOENT
) {
693 lua_pushboolean(state
, B_FALSE
);
694 } else if (error
== EXDEV
) {
695 return (luaL_error(state
, "dataset '%s' is not in the "
696 "target pool", dsname
));
697 } else if (error
== EIO
) {
698 return (luaL_error(state
, "I/O error opening dataset '%s'",
700 } else if (error
!= 0) {
701 return (luaL_error(state
, "unexpected error %d", error
));
708 * Allocate/realloc/free a buffer for the lua interpreter.
710 * When nsize is 0, behaves as free() and returns NULL.
712 * If ptr is NULL, behaves as malloc() and returns an allocated buffer of size
715 * Otherwise, behaves as realloc(), changing the allocation from osize to nsize.
716 * Shrinking the buffer size never fails.
718 * The original allocated buffer size is stored as a uint64 at the beginning of
719 * the buffer to avoid actually reallocating when shrinking a buffer, since lua
720 * requires that this operation never fail.
723 zcp_lua_alloc(void *ud
, void *ptr
, size_t osize
, size_t nsize
)
725 zcp_alloc_arg_t
*allocargs
= ud
;
729 int64_t *allocbuf
= (int64_t *)ptr
- 1;
730 int64_t allocsize
= *allocbuf
;
731 ASSERT3S(allocsize
, >, 0);
732 ASSERT3S(allocargs
->aa_alloc_remaining
+ allocsize
, <=,
733 allocargs
->aa_alloc_limit
);
734 allocargs
->aa_alloc_remaining
+= allocsize
;
735 vmem_free(allocbuf
, allocsize
);
738 } else if (ptr
== NULL
) {
740 int64_t allocsize
= nsize
+ sizeof (int64_t);
742 if (!allocargs
->aa_must_succeed
&&
744 allocsize
> allocargs
->aa_alloc_remaining
)) {
748 allocbuf
= vmem_alloc(allocsize
, KM_SLEEP
);
749 allocargs
->aa_alloc_remaining
-= allocsize
;
751 *allocbuf
= allocsize
;
752 return (allocbuf
+ 1);
753 } else if (nsize
<= osize
) {
755 * If shrinking the buffer, lua requires that the reallocation
760 ASSERT3U(nsize
, >, osize
);
762 uint64_t *luabuf
= zcp_lua_alloc(ud
, NULL
, 0, nsize
);
763 if (luabuf
== NULL
) {
766 (void) memcpy(luabuf
, ptr
, osize
);
767 VERIFY3P(zcp_lua_alloc(ud
, ptr
, osize
, 0), ==, NULL
);
773 zcp_lua_counthook(lua_State
*state
, lua_Debug
*ar
)
776 lua_getfield(state
, LUA_REGISTRYINDEX
, ZCP_RUN_INFO_KEY
);
777 zcp_run_info_t
*ri
= lua_touserdata(state
, -1);
780 * Check if we were canceled while waiting for the
781 * txg to sync or from our open context thread
783 if (ri
->zri_canceled
|| (!ri
->zri_sync
&& issig())) {
784 ri
->zri_canceled
= B_TRUE
;
785 (void) lua_pushstring(state
, "Channel program was canceled.");
786 (void) lua_error(state
);
791 * Check how many instructions the channel program has
792 * executed so far, and compare against the limit.
794 ri
->zri_curinstrs
+= zfs_lua_check_instrlimit_interval
;
795 if (ri
->zri_maxinstrs
!= 0 && ri
->zri_curinstrs
> ri
->zri_maxinstrs
) {
796 ri
->zri_timed_out
= B_TRUE
;
797 (void) lua_pushstring(state
,
798 "Channel program timed out.");
799 (void) lua_error(state
);
805 zcp_panic_cb(lua_State
*state
)
807 panic("unprotected error in call to Lua API (%s)\n",
808 lua_tostring(state
, -1));
813 zcp_eval_impl(dmu_tx_t
*tx
, zcp_run_info_t
*ri
)
816 lua_State
*state
= ri
->zri_state
;
818 VERIFY3U(3, ==, lua_gettop(state
));
820 /* finish initializing our runtime state */
821 ri
->zri_pool
= dmu_tx_pool(tx
);
823 list_create(&ri
->zri_cleanup_handlers
, sizeof (zcp_cleanup_handler_t
),
824 offsetof(zcp_cleanup_handler_t
, zch_node
));
827 * Store the zcp_run_info_t struct for this run in the Lua registry.
828 * Registry entries are not directly accessible by the Lua scripts but
829 * can be accessed by our callbacks.
831 lua_pushlightuserdata(state
, ri
);
832 lua_setfield(state
, LUA_REGISTRYINDEX
, ZCP_RUN_INFO_KEY
);
833 VERIFY3U(3, ==, lua_gettop(state
));
836 * Tell the Lua interpreter to call our handler every count
837 * instructions. Channel programs that execute too many instructions
838 * should die with ETIME.
840 (void) lua_sethook(state
, zcp_lua_counthook
, LUA_MASKCOUNT
,
841 zfs_lua_check_instrlimit_interval
);
844 * Tell the Lua memory allocator to stop using KM_SLEEP before handing
845 * off control to the channel program. Channel programs that use too
846 * much memory should die with ENOSPC.
848 ri
->zri_allocargs
->aa_must_succeed
= B_FALSE
;
851 * Call the Lua function that open-context passed us. This pops the
852 * function and its input from the stack and pushes any return
855 err
= lua_pcall(state
, 1, LUA_MULTRET
, 1);
858 * Let Lua use KM_SLEEP while we interpret the return values.
860 ri
->zri_allocargs
->aa_must_succeed
= B_TRUE
;
863 * Remove the error handler callback from the stack. At this point,
864 * there shouldn't be any cleanup handler registered in the handler
865 * list (zri_cleanup_handlers), regardless of whether it ran or not.
867 list_destroy(&ri
->zri_cleanup_handlers
);
868 lua_remove(state
, 1);
873 * Lua supports returning multiple values in a single return
874 * statement. Return values will have been pushed onto the
879 * To simplify the process of retrieving a return value from a
880 * channel program, we disallow returning more than one value
881 * to ZFS from the Lua script, yielding a singleton return
882 * nvlist of the form { "return": Return value 1 }.
884 int return_count
= lua_gettop(state
);
886 if (return_count
== 1) {
888 zcp_convert_return_values(state
, ri
->zri_outnvl
,
889 ZCP_RET_RETURN
, &ri
->zri_result
);
890 } else if (return_count
> 1) {
891 ri
->zri_result
= SET_ERROR(ECHRNG
);
892 lua_settop(state
, 0);
893 (void) lua_pushfstring(state
, "Multiple return "
894 "values not supported");
895 zcp_convert_return_values(state
, ri
->zri_outnvl
,
896 ZCP_RET_ERROR
, &ri
->zri_result
);
903 * The channel program encountered a fatal error within the
904 * script, such as failing an assertion, or calling a function
905 * with incompatible arguments. The error value and the
906 * traceback generated by zcp_error_handler() should be on the
909 VERIFY3U(1, ==, lua_gettop(state
));
910 if (ri
->zri_timed_out
) {
911 ri
->zri_result
= SET_ERROR(ETIME
);
912 } else if (ri
->zri_canceled
) {
913 ri
->zri_result
= SET_ERROR(EINTR
);
915 ri
->zri_result
= SET_ERROR(ECHRNG
);
918 zcp_convert_return_values(state
, ri
->zri_outnvl
,
919 ZCP_RET_ERROR
, &ri
->zri_result
);
921 if (ri
->zri_result
== ETIME
&& ri
->zri_outnvl
!= NULL
) {
922 (void) nvlist_add_uint64(ri
->zri_outnvl
,
923 ZCP_ARG_INSTRLIMIT
, ri
->zri_curinstrs
);
929 * The channel program encountered a fatal error within the
930 * script, and we encountered another error while trying to
931 * compute the traceback in zcp_error_handler(). We can only
932 * return the error message.
934 VERIFY3U(1, ==, lua_gettop(state
));
935 if (ri
->zri_timed_out
) {
936 ri
->zri_result
= SET_ERROR(ETIME
);
937 } else if (ri
->zri_canceled
) {
938 ri
->zri_result
= SET_ERROR(EINTR
);
940 ri
->zri_result
= SET_ERROR(ECHRNG
);
943 zcp_convert_return_values(state
, ri
->zri_outnvl
,
944 ZCP_RET_ERROR
, &ri
->zri_result
);
949 * Lua ran out of memory while running the channel program.
950 * There's not much we can do.
952 ri
->zri_result
= SET_ERROR(ENOSPC
);
960 zcp_pool_error(zcp_run_info_t
*ri
, const char *poolname
, int error
)
962 ri
->zri_result
= SET_ERROR(ECHRNG
);
963 lua_settop(ri
->zri_state
, 0);
964 (void) lua_pushfstring(ri
->zri_state
, "Could not open pool: %s "
965 "errno: %d", poolname
, error
);
966 zcp_convert_return_values(ri
->zri_state
, ri
->zri_outnvl
,
967 ZCP_RET_ERROR
, &ri
->zri_result
);
972 * This callback is called when txg_wait_synced_sig encountered a signal.
973 * The txg_wait_synced_sig will continue to wait for the txg to complete
974 * after calling this callback.
977 zcp_eval_sig(void *arg
, dmu_tx_t
*tx
)
980 zcp_run_info_t
*ri
= arg
;
982 ri
->zri_canceled
= B_TRUE
;
986 zcp_eval_sync(void *arg
, dmu_tx_t
*tx
)
988 zcp_run_info_t
*ri
= arg
;
991 * Open context should have setup the stack to contain:
992 * 1: Error handler callback
993 * 2: Script to run (converted to a Lua function)
994 * 3: nvlist input to function (converted to Lua table or nil)
996 VERIFY3U(3, ==, lua_gettop(ri
->zri_state
));
998 zcp_eval_impl(tx
, ri
);
1002 zcp_eval_open(zcp_run_info_t
*ri
, const char *poolname
)
1009 * See comment from the same assertion in zcp_eval_sync().
1011 VERIFY3U(3, ==, lua_gettop(ri
->zri_state
));
1013 error
= dsl_pool_hold(poolname
, FTAG
, &dp
);
1015 zcp_pool_error(ri
, poolname
, error
);
1020 * As we are running in open-context, we have no transaction associated
1021 * with the channel program. At the same time, functions from the
1022 * zfs.check submodule need to be associated with a transaction as
1023 * they are basically dry-runs of their counterparts in the zfs.sync
1024 * submodule. These functions should be able to run in open-context.
1025 * Therefore we create a new transaction that we later abort once
1026 * the channel program has been evaluated.
1028 tx
= dmu_tx_create_dd(dp
->dp_mos_dir
);
1030 zcp_eval_impl(tx
, ri
);
1034 dsl_pool_rele(dp
, FTAG
);
1038 zcp_eval(const char *poolname
, const char *program
, boolean_t sync
,
1039 uint64_t instrlimit
, uint64_t memlimit
, nvpair_t
*nvarg
, nvlist_t
*outnvl
)
1043 zcp_run_info_t runinfo
;
1045 if (instrlimit
> zfs_lua_max_instrlimit
)
1046 return (SET_ERROR(EINVAL
));
1047 if (memlimit
== 0 || memlimit
> zfs_lua_max_memlimit
)
1048 return (SET_ERROR(EINVAL
));
1050 zcp_alloc_arg_t allocargs
= {
1051 .aa_must_succeed
= B_TRUE
,
1052 .aa_alloc_remaining
= (int64_t)memlimit
,
1053 .aa_alloc_limit
= (int64_t)memlimit
,
1057 * Creates a Lua state with a memory allocator that uses KM_SLEEP.
1058 * This should never fail.
1060 state
= lua_newstate(zcp_lua_alloc
, &allocargs
);
1061 VERIFY(state
!= NULL
);
1062 (void) lua_atpanic(state
, zcp_panic_cb
);
1065 * Load core Lua libraries we want access to.
1067 VERIFY3U(1, ==, luaopen_base(state
));
1069 VERIFY3U(1, ==, luaopen_coroutine(state
));
1070 lua_setglobal(state
, LUA_COLIBNAME
);
1071 VERIFY0(lua_gettop(state
));
1072 VERIFY3U(1, ==, luaopen_string(state
));
1073 lua_setglobal(state
, LUA_STRLIBNAME
);
1074 VERIFY0(lua_gettop(state
));
1075 VERIFY3U(1, ==, luaopen_table(state
));
1076 lua_setglobal(state
, LUA_TABLIBNAME
);
1077 VERIFY0(lua_gettop(state
));
1080 * Load globally visible variables such as errno aliases.
1082 zcp_load_globals(state
);
1083 VERIFY0(lua_gettop(state
));
1086 * Load ZFS-specific modules.
1088 lua_newtable(state
);
1089 VERIFY3U(1, ==, zcp_load_list_lib(state
));
1090 lua_setfield(state
, -2, "list");
1091 VERIFY3U(1, ==, zcp_load_synctask_lib(state
, B_FALSE
));
1092 lua_setfield(state
, -2, "check");
1093 VERIFY3U(1, ==, zcp_load_synctask_lib(state
, B_TRUE
));
1094 lua_setfield(state
, -2, "sync");
1095 VERIFY3U(1, ==, zcp_load_get_lib(state
));
1096 lua_pushcclosure(state
, zcp_debug_info
.func
, 0);
1097 lua_setfield(state
, -2, zcp_debug_info
.name
);
1098 lua_pushcclosure(state
, zcp_exists_info
.func
, 0);
1099 lua_setfield(state
, -2, zcp_exists_info
.name
);
1100 lua_setglobal(state
, "zfs");
1101 VERIFY0(lua_gettop(state
));
1104 * Push the error-callback that calculates Lua stack traces on
1105 * unexpected failures.
1107 lua_pushcfunction(state
, zcp_error_handler
);
1108 VERIFY3U(1, ==, lua_gettop(state
));
1111 * Load the actual script as a function onto the stack as text ("t").
1112 * The only valid error condition is a syntax error in the script.
1113 * ERRMEM should not be possible because our allocator is using
1114 * KM_SLEEP. ERRGCMM should not be possible because we have not added
1115 * any objects with __gc metamethods to the interpreter that could
1118 err
= luaL_loadbufferx(state
, program
, strlen(program
),
1119 "channel program", "t");
1120 if (err
== LUA_ERRSYNTAX
) {
1121 fnvlist_add_string(outnvl
, ZCP_RET_ERROR
,
1122 lua_tostring(state
, -1));
1124 return (SET_ERROR(EINVAL
));
1127 VERIFY3U(2, ==, lua_gettop(state
));
1130 * Convert the input nvlist to a Lua object and put it on top of the
1134 err
= zcp_nvpair_value_to_lua(state
, nvarg
,
1135 errmsg
, sizeof (errmsg
));
1137 fnvlist_add_string(outnvl
, ZCP_RET_ERROR
, errmsg
);
1139 return (SET_ERROR(EINVAL
));
1141 VERIFY3U(3, ==, lua_gettop(state
));
1143 runinfo
.zri_state
= state
;
1144 runinfo
.zri_allocargs
= &allocargs
;
1145 runinfo
.zri_outnvl
= outnvl
;
1146 runinfo
.zri_result
= 0;
1147 runinfo
.zri_cred
= CRED();
1148 runinfo
.zri_proc
= curproc
;
1149 runinfo
.zri_timed_out
= B_FALSE
;
1150 runinfo
.zri_canceled
= B_FALSE
;
1151 runinfo
.zri_sync
= sync
;
1152 runinfo
.zri_space_used
= 0;
1153 runinfo
.zri_curinstrs
= 0;
1154 runinfo
.zri_maxinstrs
= instrlimit
;
1155 runinfo
.zri_new_zvols
= fnvlist_alloc();
1158 err
= dsl_sync_task_sig(poolname
, NULL
, zcp_eval_sync
,
1159 zcp_eval_sig
, &runinfo
, 0, ZFS_SPACE_CHECK_ZCP_EVAL
);
1161 zcp_pool_error(&runinfo
, poolname
, err
);
1163 zcp_eval_open(&runinfo
, poolname
);
1168 * Create device minor nodes for any new zvols.
1170 for (nvpair_t
*pair
= nvlist_next_nvpair(runinfo
.zri_new_zvols
, NULL
);
1172 pair
= nvlist_next_nvpair(runinfo
.zri_new_zvols
, pair
)) {
1173 zvol_create_minor(nvpair_name(pair
));
1175 fnvlist_free(runinfo
.zri_new_zvols
);
1177 return (runinfo
.zri_result
);
1181 * Retrieve metadata about the currently running channel program.
1184 zcp_run_info(lua_State
*state
)
1188 lua_getfield(state
, LUA_REGISTRYINDEX
, ZCP_RUN_INFO_KEY
);
1189 ri
= lua_touserdata(state
, -1);
1198 * The Lua language allows methods to be called with any number
1199 * of arguments of any type. When calling back into ZFS we need to sanitize
1200 * arguments from channel programs to make sure unexpected arguments or
1201 * arguments of the wrong type result in clear error messages. To do this
1202 * in a uniform way all callbacks from channel programs should use the
1203 * zcp_parse_args() function to interpret inputs.
1205 * Positional vs Keyword Arguments
1206 * ===============================
1208 * Every callback function takes a fixed set of required positional arguments
1209 * and optional keyword arguments. For example, the destroy function takes
1210 * a single positional string argument (the name of the dataset to destroy)
1211 * and an optional "defer" keyword boolean argument. When calling lua functions
1212 * with parentheses, only positional arguments can be used:
1214 * zfs.sync.snapshot("rpool@snap")
1216 * To use keyword arguments functions should be called with a single argument
1217 * that is a lua table containing mappings of integer -> positional arguments
1218 * and string -> keyword arguments:
1220 * zfs.sync.snapshot({1="rpool@snap", defer=true})
1222 * The lua language allows curly braces to be used in place of parenthesis as
1223 * syntactic sugar for this calling convention:
1225 * zfs.sync.snapshot{"rpool@snap", defer=true}
1229 * Throw an error and print the given arguments. If there are too many
1230 * arguments to fit in the output buffer, only the error format string is
1234 zcp_args_error(lua_State
*state
, const char *fname
, const zcp_arg_t
*pargs
,
1235 const zcp_arg_t
*kwargs
, const char *fmt
, ...)
1239 size_t len
= sizeof (errmsg
);
1243 va_start(argp
, fmt
);
1244 VERIFY3U(len
, >, vsnprintf(errmsg
, len
, fmt
, argp
));
1248 * Calculate the total length of the final string, including extra
1249 * formatting characters. If the argument dump would be too large,
1250 * only print the error string.
1252 msglen
= strlen(errmsg
);
1253 msglen
+= strlen(fname
) + 4; /* : + {} + null terminator */
1254 for (i
= 0; pargs
[i
].za_name
!= NULL
; i
++) {
1255 msglen
+= strlen(pargs
[i
].za_name
);
1256 msglen
+= strlen(lua_typename(state
, pargs
[i
].za_lua_type
));
1257 if (pargs
[i
+ 1].za_name
!= NULL
|| kwargs
[0].za_name
!= NULL
)
1258 msglen
+= 5; /* < + ( + )> + , */
1260 msglen
+= 4; /* < + ( + )> */
1262 for (i
= 0; kwargs
[i
].za_name
!= NULL
; i
++) {
1263 msglen
+= strlen(kwargs
[i
].za_name
);
1264 msglen
+= strlen(lua_typename(state
, kwargs
[i
].za_lua_type
));
1265 if (kwargs
[i
+ 1].za_name
!= NULL
)
1266 msglen
+= 4; /* =( + ) + , */
1268 msglen
+= 3; /* =( + ) */
1272 (void) luaL_error(state
, errmsg
);
1274 VERIFY3U(len
, >, strlcat(errmsg
, ": ", len
));
1275 VERIFY3U(len
, >, strlcat(errmsg
, fname
, len
));
1276 VERIFY3U(len
, >, strlcat(errmsg
, "{", len
));
1277 for (i
= 0; pargs
[i
].za_name
!= NULL
; i
++) {
1278 VERIFY3U(len
, >, strlcat(errmsg
, "<", len
));
1279 VERIFY3U(len
, >, strlcat(errmsg
, pargs
[i
].za_name
, len
));
1280 VERIFY3U(len
, >, strlcat(errmsg
, "(", len
));
1281 VERIFY3U(len
, >, strlcat(errmsg
,
1282 lua_typename(state
, pargs
[i
].za_lua_type
), len
));
1283 VERIFY3U(len
, >, strlcat(errmsg
, ")>", len
));
1284 if (pargs
[i
+ 1].za_name
!= NULL
|| kwargs
[0].za_name
!= NULL
) {
1285 VERIFY3U(len
, >, strlcat(errmsg
, ", ", len
));
1288 for (i
= 0; kwargs
[i
].za_name
!= NULL
; i
++) {
1289 VERIFY3U(len
, >, strlcat(errmsg
, kwargs
[i
].za_name
, len
));
1290 VERIFY3U(len
, >, strlcat(errmsg
, "=(", len
));
1291 VERIFY3U(len
, >, strlcat(errmsg
,
1292 lua_typename(state
, kwargs
[i
].za_lua_type
), len
));
1293 VERIFY3U(len
, >, strlcat(errmsg
, ")", len
));
1294 if (kwargs
[i
+ 1].za_name
!= NULL
) {
1295 VERIFY3U(len
, >, strlcat(errmsg
, ", ", len
));
1298 VERIFY3U(len
, >, strlcat(errmsg
, "}", len
));
1300 (void) luaL_error(state
, errmsg
);
1301 panic("unreachable code");
1305 zcp_parse_table_args(lua_State
*state
, const char *fname
,
1306 const zcp_arg_t
*pargs
, const zcp_arg_t
*kwargs
)
1311 for (i
= 0; pargs
[i
].za_name
!= NULL
; i
++) {
1313 * Check the table for this positional argument, leaving it
1314 * on the top of the stack once we finish validating it.
1316 lua_pushinteger(state
, i
+ 1);
1317 lua_gettable(state
, 1);
1319 type
= lua_type(state
, -1);
1320 if (type
== LUA_TNIL
) {
1321 zcp_args_error(state
, fname
, pargs
, kwargs
,
1322 "too few arguments");
1323 panic("unreachable code");
1324 } else if (type
!= pargs
[i
].za_lua_type
) {
1325 zcp_args_error(state
, fname
, pargs
, kwargs
,
1326 "arg %d wrong type (is '%s', expected '%s')",
1327 i
+ 1, lua_typename(state
, type
),
1328 lua_typename(state
, pargs
[i
].za_lua_type
));
1329 panic("unreachable code");
1333 * Remove the positional argument from the table.
1335 lua_pushinteger(state
, i
+ 1);
1337 lua_settable(state
, 1);
1340 for (i
= 0; kwargs
[i
].za_name
!= NULL
; i
++) {
1342 * Check the table for this keyword argument, which may be
1343 * nil if it was omitted. Leave the value on the top of
1344 * the stack after validating it.
1346 lua_getfield(state
, 1, kwargs
[i
].za_name
);
1348 type
= lua_type(state
, -1);
1349 if (type
!= LUA_TNIL
&& type
!= kwargs
[i
].za_lua_type
) {
1350 zcp_args_error(state
, fname
, pargs
, kwargs
,
1351 "kwarg '%s' wrong type (is '%s', expected '%s')",
1352 kwargs
[i
].za_name
, lua_typename(state
, type
),
1353 lua_typename(state
, kwargs
[i
].za_lua_type
));
1354 panic("unreachable code");
1358 * Remove the keyword argument from the table.
1361 lua_setfield(state
, 1, kwargs
[i
].za_name
);
1365 * Any entries remaining in the table are invalid inputs, print
1366 * an error message based on what the entry is.
1369 if (lua_next(state
, 1)) {
1370 if (lua_isnumber(state
, -2) && lua_tointeger(state
, -2) > 0) {
1371 zcp_args_error(state
, fname
, pargs
, kwargs
,
1372 "too many positional arguments");
1373 } else if (lua_isstring(state
, -2)) {
1374 zcp_args_error(state
, fname
, pargs
, kwargs
,
1375 "invalid kwarg '%s'", lua_tostring(state
, -2));
1377 zcp_args_error(state
, fname
, pargs
, kwargs
,
1378 "kwarg keys must be strings");
1380 panic("unreachable code");
1383 lua_remove(state
, 1);
1387 zcp_parse_pos_args(lua_State
*state
, const char *fname
, const zcp_arg_t
*pargs
,
1388 const zcp_arg_t
*kwargs
)
1393 for (i
= 0; pargs
[i
].za_name
!= NULL
; i
++) {
1394 type
= lua_type(state
, i
+ 1);
1395 if (type
== LUA_TNONE
) {
1396 zcp_args_error(state
, fname
, pargs
, kwargs
,
1397 "too few arguments");
1398 panic("unreachable code");
1399 } else if (type
!= pargs
[i
].za_lua_type
) {
1400 zcp_args_error(state
, fname
, pargs
, kwargs
,
1401 "arg %d wrong type (is '%s', expected '%s')",
1402 i
+ 1, lua_typename(state
, type
),
1403 lua_typename(state
, pargs
[i
].za_lua_type
));
1404 panic("unreachable code");
1407 if (lua_gettop(state
) != i
) {
1408 zcp_args_error(state
, fname
, pargs
, kwargs
,
1409 "too many positional arguments");
1410 panic("unreachable code");
1413 for (i
= 0; kwargs
[i
].za_name
!= NULL
; i
++) {
1419 * Checks the current Lua stack against an expected set of positional and
1420 * keyword arguments. If the stack does not match the expected arguments
1421 * aborts the current channel program with a useful error message, otherwise
1422 * it re-arranges the stack so that it contains the positional arguments
1423 * followed by the keyword argument values in declaration order. Any missing
1424 * keyword argument will be represented by a nil value on the stack.
1426 * If the stack contains exactly one argument of type LUA_TTABLE the curly
1427 * braces calling convention is assumed, otherwise the stack is parsed for
1428 * positional arguments only.
1430 * This function should be used by every function callback. It should be called
1431 * before the callback manipulates the Lua stack as it assumes the stack
1432 * represents the function arguments.
1435 zcp_parse_args(lua_State
*state
, const char *fname
, const zcp_arg_t
*pargs
,
1436 const zcp_arg_t
*kwargs
)
1438 if (lua_gettop(state
) == 1 && lua_istable(state
, 1)) {
1439 zcp_parse_table_args(state
, fname
, pargs
, kwargs
);
1441 zcp_parse_pos_args(state
, fname
, pargs
, kwargs
);
1445 ZFS_MODULE_PARAM(zfs_lua
, zfs_lua_
, max_instrlimit
, U64
, ZMOD_RW
,
1446 "Max instruction limit that can be specified for a channel program");
1448 ZFS_MODULE_PARAM(zfs_lua
, zfs_lua_
, max_memlimit
, U64
, ZMOD_RW
,
1449 "Max memory limit that can be specified for a channel program");