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, 2017 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 occured.
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.
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 <util/sscanf.h>
105 #define ZCP_NVLIST_MAX_DEPTH 20
107 uint64_t zfs_lua_check_instrlimit_interval
= 100;
108 uint64_t zfs_lua_max_instrlimit
= ZCP_MAX_INSTRLIMIT
;
109 uint64_t zfs_lua_max_memlimit
= ZCP_MAX_MEMLIMIT
;
112 * Forward declarations for mutually recursive functions
114 static int zcp_nvpair_value_to_lua(lua_State
*, nvpair_t
*, char *, int);
115 static int zcp_lua_to_nvlist_impl(lua_State
*, int, nvlist_t
*, const char *,
118 typedef struct zcp_alloc_arg
{
119 boolean_t aa_must_succeed
;
120 int64_t aa_alloc_remaining
;
121 int64_t aa_alloc_limit
;
124 typedef struct zcp_eval_arg
{
126 zcp_alloc_arg_t
*ea_allocargs
;
130 uint64_t ea_instrlimit
;
134 * The outer-most error callback handler for use with lua_pcall(). On
135 * error Lua will call this callback with a single argument that
136 * represents the error value. In most cases this will be a string
137 * containing an error message, but channel programs can use Lua's
138 * error() function to return arbitrary objects as errors. This callback
139 * returns (on the Lua stack) the original error object along with a traceback.
141 * Fatal Lua errors can occur while resources are held, so we also call any
142 * registered cleanup function here.
145 zcp_error_handler(lua_State
*state
)
151 VERIFY3U(1, ==, lua_gettop(state
));
152 msg
= lua_tostring(state
, 1);
153 luaL_traceback(state
, state
, msg
, 1);
158 zcp_argerror(lua_State
*state
, int narg
, const char *msg
, ...)
162 va_start(alist
, msg
);
163 const char *buf
= lua_pushvfstring(state
, msg
, alist
);
166 return (luaL_argerror(state
, narg
, buf
));
170 * Install a new cleanup function, which will be invoked with the given
171 * opaque argument if a fatal error causes the Lua interpreter to longjump out
172 * of a function call.
174 * If an error occurs, the cleanup function will be invoked exactly once and
177 * Returns the registered cleanup handler so the caller can deregister it
178 * if no error occurs.
180 zcp_cleanup_handler_t
*
181 zcp_register_cleanup(lua_State
*state
, zcp_cleanup_t cleanfunc
, void *cleanarg
)
183 zcp_run_info_t
*ri
= zcp_run_info(state
);
185 zcp_cleanup_handler_t
*zch
= kmem_alloc(sizeof (*zch
), KM_SLEEP
);
186 zch
->zch_cleanup_func
= cleanfunc
;
187 zch
->zch_cleanup_arg
= cleanarg
;
188 list_insert_head(&ri
->zri_cleanup_handlers
, zch
);
194 zcp_deregister_cleanup(lua_State
*state
, zcp_cleanup_handler_t
*zch
)
196 zcp_run_info_t
*ri
= zcp_run_info(state
);
197 list_remove(&ri
->zri_cleanup_handlers
, zch
);
198 kmem_free(zch
, sizeof (*zch
));
202 * Execute the currently registered cleanup handlers then free them and
203 * destroy the handler list.
206 zcp_cleanup(lua_State
*state
)
208 zcp_run_info_t
*ri
= zcp_run_info(state
);
210 for (zcp_cleanup_handler_t
*zch
=
211 list_remove_head(&ri
->zri_cleanup_handlers
); zch
!= NULL
;
212 zch
= list_remove_head(&ri
->zri_cleanup_handlers
)) {
213 zch
->zch_cleanup_func(zch
->zch_cleanup_arg
);
214 kmem_free(zch
, sizeof (*zch
));
219 * Convert the lua table at the given index on the Lua stack to an nvlist
222 * If the table can not be converted for any reason, NULL is returned and
223 * an error message is pushed onto the Lua stack.
226 zcp_table_to_nvlist(lua_State
*state
, int index
, int depth
)
230 * Converting a Lua table to an nvlist with key uniqueness checking is
231 * O(n^2) in the number of keys in the nvlist, which can take a long
232 * time when we return a large table from a channel program.
233 * Furthermore, Lua's table interface *almost* guarantees unique keys
234 * on its own (details below). Therefore, we don't use fnvlist_alloc()
235 * here to avoid the built-in uniqueness checking.
237 * The *almost* is because it's possible to have key collisions between
238 * e.g. the string "1" and the number 1, or the string "true" and the
239 * boolean true, so we explicitly check that when we're looking at a
240 * key which is an integer / boolean or a string that can be parsed as
241 * one of those types. In the worst case this could still devolve into
242 * O(n^2), so we only start doing these checks on boolean/integer keys
243 * once we've seen a string key which fits this weird usage pattern.
245 * Ultimately, we still want callers to know that the keys in this
246 * nvlist are unique, so before we return this we set the nvlist's
247 * flags to reflect that.
249 VERIFY0(nvlist_alloc(&nvl
, 0, KM_SLEEP
));
252 * Push an empty stack slot where lua_next() will store each
256 boolean_t saw_str_could_collide
= B_FALSE
;
257 while (lua_next(state
, index
) != 0) {
259 * The next key-value pair from the table at index is
260 * now on the stack, with the key at stack slot -2 and
261 * the value at slot -1.
265 const char *key
= NULL
;
266 boolean_t key_could_collide
= B_FALSE
;
268 switch (lua_type(state
, -2)) {
270 key
= lua_tostring(state
, -2);
272 /* check if this could collide with a number or bool */
275 if ((sscanf(key
, "%lld%n", &tmp
, &parselen
) > 0 &&
276 parselen
== strlen(key
)) ||
277 strcmp(key
, "true") == 0 ||
278 strcmp(key
, "false") == 0) {
279 key_could_collide
= B_TRUE
;
280 saw_str_could_collide
= B_TRUE
;
284 key
= (lua_toboolean(state
, -2) == B_TRUE
?
286 if (saw_str_could_collide
) {
287 key_could_collide
= B_TRUE
;
291 VERIFY3U(sizeof (buf
), >,
292 snprintf(buf
, sizeof (buf
), "%lld",
293 (longlong_t
)lua_tonumber(state
, -2)));
295 if (saw_str_could_collide
) {
296 key_could_collide
= B_TRUE
;
301 (void) lua_pushfstring(state
, "Invalid key "
302 "type '%s' in table",
303 lua_typename(state
, lua_type(state
, -2)));
307 * Check for type-mismatched key collisions, and throw an error.
309 if (key_could_collide
&& nvlist_exists(nvl
, key
)) {
311 (void) lua_pushfstring(state
, "Collision of "
312 "key '%s' in table", key
);
316 * Recursively convert the table value and insert into
317 * the new nvlist with the parsed key. To prevent
318 * stack overflow on circular or heavily nested tables,
319 * we track the current nvlist depth.
321 if (depth
>= ZCP_NVLIST_MAX_DEPTH
) {
323 (void) lua_pushfstring(state
, "Maximum table "
324 "depth (%d) exceeded for table",
325 ZCP_NVLIST_MAX_DEPTH
);
328 err
= zcp_lua_to_nvlist_impl(state
, -1, nvl
, key
,
333 * Error message has been pushed to the lua
334 * stack by the recursive call.
339 * Pop the value pushed by lua_next().
345 * Mark the nvlist as having unique keys. This is a little ugly, but we
346 * ensured above that there are no duplicate keys in the nvlist.
348 nvl
->nvl_nvflag
|= NV_UNIQUE_NAME
;
354 * Convert a value from the given index into the lua stack to an nvpair, adding
355 * it to an nvlist with the given key.
357 * Values are converted as follows:
362 * nil -> boolean (no value)
364 * Lua tables are converted to nvlists and then inserted. The table's keys
365 * are converted to strings then used as keys in the nvlist to store each table
366 * element. Keys are converted as follows:
368 * string -> no change
370 * boolean -> "true" | "false"
373 * In the case of a key collision, an error is thrown.
375 * If an error is encountered, a nonzero error code is returned, and an error
376 * string will be pushed onto the Lua stack.
379 zcp_lua_to_nvlist_impl(lua_State
*state
, int index
, nvlist_t
*nvl
,
380 const char *key
, int depth
)
383 * Verify that we have enough remaining space in the lua stack to parse
384 * a key-value pair and push an error.
386 if (!lua_checkstack(state
, 3)) {
387 (void) lua_pushstring(state
, "Lua stack overflow");
391 index
= lua_absindex(state
, index
);
393 switch (lua_type(state
, index
)) {
395 fnvlist_add_boolean(nvl
, key
);
398 fnvlist_add_boolean_value(nvl
, key
,
399 lua_toboolean(state
, index
));
402 fnvlist_add_int64(nvl
, key
, lua_tonumber(state
, index
));
405 fnvlist_add_string(nvl
, key
, lua_tostring(state
, index
));
408 nvlist_t
*value_nvl
= zcp_table_to_nvlist(state
, index
, depth
);
409 if (value_nvl
== NULL
)
412 fnvlist_add_nvlist(nvl
, key
, value_nvl
);
413 fnvlist_free(value_nvl
);
417 (void) lua_pushfstring(state
,
418 "Invalid value type '%s' for key '%s'",
419 lua_typename(state
, lua_type(state
, index
)), key
);
427 * Convert a lua value to an nvpair, adding it to an nvlist with the given key.
430 zcp_lua_to_nvlist(lua_State
*state
, int index
, nvlist_t
*nvl
, const char *key
)
433 * On error, zcp_lua_to_nvlist_impl pushes an error string onto the Lua
434 * stack before returning with a nonzero error code. If an error is
435 * returned, throw a fatal lua error with the given string.
437 if (zcp_lua_to_nvlist_impl(state
, index
, nvl
, key
, 0) != 0)
438 (void) lua_error(state
);
442 zcp_lua_to_nvlist_helper(lua_State
*state
)
444 nvlist_t
*nv
= (nvlist_t
*)lua_touserdata(state
, 2);
445 const char *key
= (const char *)lua_touserdata(state
, 1);
446 zcp_lua_to_nvlist(state
, 3, nv
, key
);
451 zcp_convert_return_values(lua_State
*state
, nvlist_t
*nvl
,
452 const char *key
, zcp_eval_arg_t
*evalargs
)
455 lua_pushcfunction(state
, zcp_lua_to_nvlist_helper
);
456 lua_pushlightuserdata(state
, (char *)key
);
457 lua_pushlightuserdata(state
, nvl
);
458 lua_pushvalue(state
, 1);
459 lua_remove(state
, 1);
460 err
= lua_pcall(state
, 3, 0, 0); /* zcp_lua_to_nvlist_helper */
462 zcp_lua_to_nvlist(state
, 1, nvl
, ZCP_RET_ERROR
);
463 evalargs
->ea_result
= SET_ERROR(ECHRNG
);
468 * Push a Lua table representing nvl onto the stack. If it can't be
469 * converted, return EINVAL, fill in errbuf, and push nothing. errbuf may
470 * be specified as NULL, in which case no error string will be output.
472 * Most nvlists are converted as simple key->value Lua tables, but we make
473 * an exception for the case where all nvlist entries are BOOLEANs (a string
474 * key without a value). In Lua, a table key pointing to a value of Nil
475 * (no value) is equivalent to the key not existing, so a BOOLEAN nvlist
476 * entry can't be directly converted to a Lua table entry. Nvlists of entirely
477 * BOOLEAN entries are frequently used to pass around lists of datasets, so for
478 * convenience we check for this case, and convert it to a simple Lua array of
482 zcp_nvlist_to_lua(lua_State
*state
, nvlist_t
*nvl
,
483 char *errbuf
, int errbuf_len
)
487 boolean_t has_values
= B_FALSE
;
489 * If the list doesn't have any values, just convert it to a string
492 for (pair
= nvlist_next_nvpair(nvl
, NULL
);
493 pair
!= NULL
; pair
= nvlist_next_nvpair(nvl
, pair
)) {
494 if (nvpair_type(pair
) != DATA_TYPE_BOOLEAN
) {
501 for (pair
= nvlist_next_nvpair(nvl
, NULL
);
502 pair
!= NULL
; pair
= nvlist_next_nvpair(nvl
, pair
)) {
503 (void) lua_pushinteger(state
, i
);
504 (void) lua_pushstring(state
, nvpair_name(pair
));
505 (void) lua_settable(state
, -3);
509 for (pair
= nvlist_next_nvpair(nvl
, NULL
);
510 pair
!= NULL
; pair
= nvlist_next_nvpair(nvl
, pair
)) {
511 int err
= zcp_nvpair_value_to_lua(state
, pair
,
517 (void) lua_setfield(state
, -2, nvpair_name(pair
));
524 * Push a Lua object representing the value of "pair" onto the stack.
526 * Only understands boolean_value, string, int64, nvlist,
527 * string_array, and int64_array type values. For other
528 * types, returns EINVAL, fills in errbuf, and pushes nothing.
531 zcp_nvpair_value_to_lua(lua_State
*state
, nvpair_t
*pair
,
532 char *errbuf
, int errbuf_len
)
541 switch (nvpair_type(pair
)) {
542 case DATA_TYPE_BOOLEAN_VALUE
:
543 (void) lua_pushboolean(state
,
544 fnvpair_value_boolean_value(pair
));
546 case DATA_TYPE_STRING
:
547 (void) lua_pushstring(state
, fnvpair_value_string(pair
));
549 case DATA_TYPE_INT64
:
550 (void) lua_pushinteger(state
, fnvpair_value_int64(pair
));
552 case DATA_TYPE_NVLIST
:
553 err
= zcp_nvlist_to_lua(state
,
554 fnvpair_value_nvlist(pair
), errbuf
, errbuf_len
);
556 case DATA_TYPE_STRING_ARRAY
: {
559 (void) nvpair_value_string_array(pair
, &strarr
, &nelem
);
561 for (int i
= 0; i
< nelem
; i
++) {
562 (void) lua_pushinteger(state
, i
+ 1);
563 (void) lua_pushstring(state
, strarr
[i
]);
564 (void) lua_settable(state
, -3);
568 case DATA_TYPE_UINT64_ARRAY
: {
571 (void) nvpair_value_uint64_array(pair
, &intarr
, &nelem
);
573 for (int i
= 0; i
< nelem
; i
++) {
574 (void) lua_pushinteger(state
, i
+ 1);
575 (void) lua_pushinteger(state
, intarr
[i
]);
576 (void) lua_settable(state
, -3);
580 case DATA_TYPE_INT64_ARRAY
: {
583 (void) nvpair_value_int64_array(pair
, &intarr
, &nelem
);
585 for (int i
= 0; i
< nelem
; i
++) {
586 (void) lua_pushinteger(state
, i
+ 1);
587 (void) lua_pushinteger(state
, intarr
[i
]);
588 (void) lua_settable(state
, -3);
593 if (errbuf
!= NULL
) {
594 (void) snprintf(errbuf
, errbuf_len
,
595 "Unhandled nvpair type %d for key '%s'",
596 nvpair_type(pair
), nvpair_name(pair
));
605 zcp_dataset_hold_error(lua_State
*state
, dsl_pool_t
*dp
, const char *dsname
,
608 if (error
== ENOENT
) {
609 (void) zcp_argerror(state
, 1, "no such dataset '%s'", dsname
);
610 return (0); /* not reached; zcp_argerror will longjmp */
611 } else if (error
== EXDEV
) {
612 (void) zcp_argerror(state
, 1,
613 "dataset '%s' is not in the target pool '%s'",
614 dsname
, spa_name(dp
->dp_spa
));
615 return (0); /* not reached; zcp_argerror will longjmp */
616 } else if (error
== EIO
) {
617 (void) luaL_error(state
,
618 "I/O error while accessing dataset '%s'", dsname
);
619 return (0); /* not reached; luaL_error will longjmp */
620 } else if (error
!= 0) {
621 (void) luaL_error(state
,
622 "unexpected error %d while accessing dataset '%s'",
624 return (0); /* not reached; luaL_error will longjmp */
630 * Note: will longjmp (via lua_error()) on error.
631 * Assumes that the dsname is argument #1 (for error reporting purposes).
634 zcp_dataset_hold(lua_State
*state
, dsl_pool_t
*dp
, const char *dsname
,
638 int error
= dsl_dataset_hold(dp
, dsname
, tag
, &ds
);
639 (void) zcp_dataset_hold_error(state
, dp
, dsname
, error
);
643 static int zcp_debug(lua_State
*);
644 static zcp_lib_info_t zcp_debug_info
= {
648 { .za_name
= "debug string", .za_lua_type
= LUA_TSTRING
},
657 zcp_debug(lua_State
*state
)
659 const char *dbgstring
;
660 zcp_run_info_t
*ri
= zcp_run_info(state
);
661 zcp_lib_info_t
*libinfo
= &zcp_debug_info
;
663 zcp_parse_args(state
, libinfo
->name
, libinfo
->pargs
, libinfo
->kwargs
);
665 dbgstring
= lua_tostring(state
, 1);
667 zfs_dbgmsg("txg %lld ZCP: %s", ri
->zri_tx
->tx_txg
, dbgstring
);
672 static int zcp_exists(lua_State
*);
673 static zcp_lib_info_t zcp_exists_info
= {
677 { .za_name
= "dataset", .za_lua_type
= LUA_TSTRING
},
686 zcp_exists(lua_State
*state
)
688 zcp_run_info_t
*ri
= zcp_run_info(state
);
689 dsl_pool_t
*dp
= ri
->zri_pool
;
690 zcp_lib_info_t
*libinfo
= &zcp_exists_info
;
692 zcp_parse_args(state
, libinfo
->name
, libinfo
->pargs
, libinfo
->kwargs
);
694 const char *dsname
= lua_tostring(state
, 1);
697 int error
= dsl_dataset_hold(dp
, dsname
, FTAG
, &ds
);
699 dsl_dataset_rele(ds
, FTAG
);
700 lua_pushboolean(state
, B_TRUE
);
701 } else if (error
== ENOENT
) {
702 lua_pushboolean(state
, B_FALSE
);
703 } else if (error
== EXDEV
) {
704 return (luaL_error(state
, "dataset '%s' is not in the "
705 "target pool", dsname
));
706 } else if (error
== EIO
) {
707 return (luaL_error(state
, "I/O error opening dataset '%s'",
709 } else if (error
!= 0) {
710 return (luaL_error(state
, "unexpected error %d", error
));
717 * Allocate/realloc/free a buffer for the lua interpreter.
719 * When nsize is 0, behaves as free() and returns NULL.
721 * If ptr is NULL, behaves as malloc() and returns an allocated buffer of size
724 * Otherwise, behaves as realloc(), changing the allocation from osize to nsize.
725 * Shrinking the buffer size never fails.
727 * The original allocated buffer size is stored as a uint64 at the beginning of
728 * the buffer to avoid actually reallocating when shrinking a buffer, since lua
729 * requires that this operation never fail.
732 zcp_lua_alloc(void *ud
, void *ptr
, size_t osize
, size_t nsize
)
734 zcp_alloc_arg_t
*allocargs
= ud
;
735 int flags
= (allocargs
->aa_must_succeed
) ?
736 KM_SLEEP
: (KM_NOSLEEP
| KM_NORMALPRI
);
740 int64_t *allocbuf
= (int64_t *)ptr
- 1;
741 int64_t allocsize
= *allocbuf
;
742 ASSERT3S(allocsize
, >, 0);
743 ASSERT3S(allocargs
->aa_alloc_remaining
+ allocsize
, <=,
744 allocargs
->aa_alloc_limit
);
745 allocargs
->aa_alloc_remaining
+= allocsize
;
746 kmem_free(allocbuf
, allocsize
);
749 } else if (ptr
== NULL
) {
751 int64_t allocsize
= nsize
+ sizeof (int64_t);
753 if (!allocargs
->aa_must_succeed
&&
755 allocsize
> allocargs
->aa_alloc_remaining
)) {
759 allocbuf
= kmem_alloc(allocsize
, flags
);
760 if (allocbuf
== NULL
) {
763 allocargs
->aa_alloc_remaining
-= allocsize
;
765 *allocbuf
= allocsize
;
766 return (allocbuf
+ 1);
767 } else if (nsize
<= osize
) {
769 * If shrinking the buffer, lua requires that the reallocation
774 ASSERT3U(nsize
, >, osize
);
776 uint64_t *luabuf
= zcp_lua_alloc(ud
, NULL
, 0, nsize
);
777 if (luabuf
== NULL
) {
780 (void) memcpy(luabuf
, ptr
, osize
);
781 VERIFY3P(zcp_lua_alloc(ud
, ptr
, osize
, 0), ==, NULL
);
788 zcp_lua_counthook(lua_State
*state
, lua_Debug
*ar
)
791 * If we're called, check how many instructions the channel program has
792 * executed so far, and compare against the limit.
794 lua_getfield(state
, LUA_REGISTRYINDEX
, ZCP_RUN_INFO_KEY
);
795 zcp_run_info_t
*ri
= lua_touserdata(state
, -1);
797 ri
->zri_curinstrs
+= zfs_lua_check_instrlimit_interval
;
798 if (ri
->zri_maxinstrs
!= 0 && ri
->zri_curinstrs
> ri
->zri_maxinstrs
) {
799 ri
->zri_timed_out
= B_TRUE
;
800 (void) lua_pushstring(state
,
801 "Channel program timed out.");
802 (void) lua_error(state
);
807 zcp_panic_cb(lua_State
*state
)
809 panic("unprotected error in call to Lua API (%s)\n",
810 lua_tostring(state
, -1));
815 zcp_eval_impl(dmu_tx_t
*tx
, boolean_t sync
, zcp_eval_arg_t
*evalargs
)
819 lua_State
*state
= evalargs
->ea_state
;
821 VERIFY3U(3, ==, lua_gettop(state
));
824 * Store the zcp_run_info_t struct for this run in the Lua registry.
825 * Registry entries are not directly accessible by the Lua scripts but
826 * can be accessed by our callbacks.
828 ri
.zri_space_used
= 0;
829 ri
.zri_pool
= dmu_tx_pool(tx
);
830 ri
.zri_cred
= evalargs
->ea_cred
;
832 ri
.zri_timed_out
= B_FALSE
;
834 list_create(&ri
.zri_cleanup_handlers
, sizeof (zcp_cleanup_handler_t
),
835 offsetof(zcp_cleanup_handler_t
, zch_node
));
836 ri
.zri_curinstrs
= 0;
837 ri
.zri_maxinstrs
= evalargs
->ea_instrlimit
;
839 lua_pushlightuserdata(state
, &ri
);
840 lua_setfield(state
, LUA_REGISTRYINDEX
, ZCP_RUN_INFO_KEY
);
841 VERIFY3U(3, ==, lua_gettop(state
));
844 * Tell the Lua interpreter to call our handler every count
845 * instructions. Channel programs that execute too many instructions
846 * should die with ETIME.
848 (void) lua_sethook(state
, zcp_lua_counthook
, LUA_MASKCOUNT
,
849 zfs_lua_check_instrlimit_interval
);
852 * Tell the Lua memory allocator to stop using KM_SLEEP before handing
853 * off control to the channel program. Channel programs that use too
854 * much memory should die with ENOSPC.
856 evalargs
->ea_allocargs
->aa_must_succeed
= B_FALSE
;
859 * Call the Lua function that open-context passed us. This pops the
860 * function and its input from the stack and pushes any return
863 err
= lua_pcall(state
, 1, LUA_MULTRET
, 1);
866 * Let Lua use KM_SLEEP while we interpret the return values.
868 evalargs
->ea_allocargs
->aa_must_succeed
= B_TRUE
;
871 * Remove the error handler callback from the stack. At this point,
872 * there shouldn't be any cleanup handler registered in the handler
873 * list (zri_cleanup_handlers), regardless of whether it ran or not.
875 list_destroy(&ri
.zri_cleanup_handlers
);
876 lua_remove(state
, 1);
881 * Lua supports returning multiple values in a single return
882 * statement. Return values will have been pushed onto the
887 * To simplify the process of retrieving a return value from a
888 * channel program, we disallow returning more than one value
889 * to ZFS from the Lua script, yielding a singleton return
890 * nvlist of the form { "return": Return value 1 }.
892 int return_count
= lua_gettop(state
);
894 if (return_count
== 1) {
895 evalargs
->ea_result
= 0;
896 zcp_convert_return_values(state
, evalargs
->ea_outnvl
,
897 ZCP_RET_RETURN
, evalargs
);
898 } else if (return_count
> 1) {
899 evalargs
->ea_result
= SET_ERROR(ECHRNG
);
900 (void) lua_pushfstring(state
, "Multiple return "
901 "values not supported");
902 zcp_convert_return_values(state
, evalargs
->ea_outnvl
,
903 ZCP_RET_ERROR
, evalargs
);
910 * The channel program encountered a fatal error within the
911 * script, such as failing an assertion, or calling a function
912 * with incompatible arguments. The error value and the
913 * traceback generated by zcp_error_handler() should be on the
916 VERIFY3U(1, ==, lua_gettop(state
));
917 if (ri
.zri_timed_out
) {
918 evalargs
->ea_result
= SET_ERROR(ETIME
);
920 evalargs
->ea_result
= SET_ERROR(ECHRNG
);
923 zcp_convert_return_values(state
, evalargs
->ea_outnvl
,
924 ZCP_RET_ERROR
, evalargs
);
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 evalargs
->ea_result
= SET_ERROR(ETIME
);
938 evalargs
->ea_result
= SET_ERROR(ECHRNG
);
941 zcp_convert_return_values(state
, evalargs
->ea_outnvl
,
942 ZCP_RET_ERROR
, evalargs
);
947 * Lua ran out of memory while running the channel program.
948 * There's not much we can do.
950 evalargs
->ea_result
= SET_ERROR(ENOSPC
);
958 zcp_pool_error(zcp_eval_arg_t
*evalargs
, const char *poolname
)
960 evalargs
->ea_result
= SET_ERROR(ECHRNG
);
961 (void) lua_pushfstring(evalargs
->ea_state
, "Could not open pool: %s",
963 zcp_convert_return_values(evalargs
->ea_state
, evalargs
->ea_outnvl
,
964 ZCP_RET_ERROR
, evalargs
);
969 zcp_eval_sync(void *arg
, dmu_tx_t
*tx
)
971 zcp_eval_arg_t
*evalargs
= arg
;
974 * Open context should have setup the stack to contain:
975 * 1: Error handler callback
976 * 2: Script to run (converted to a Lua function)
977 * 3: nvlist input to function (converted to Lua table or nil)
979 VERIFY3U(3, ==, lua_gettop(evalargs
->ea_state
));
981 zcp_eval_impl(tx
, B_TRUE
, evalargs
);
985 zcp_eval_open(zcp_eval_arg_t
*evalargs
, const char *poolname
)
993 * See comment from the same assertion in zcp_eval_sync().
995 VERIFY3U(3, ==, lua_gettop(evalargs
->ea_state
));
997 error
= dsl_pool_hold(poolname
, FTAG
, &dp
);
999 zcp_pool_error(evalargs
, poolname
);
1004 * As we are running in open-context, we have no transaction associated
1005 * with the channel program. At the same time, functions from the
1006 * zfs.check submodule need to be associated with a transaction as
1007 * they are basically dry-runs of their counterparts in the zfs.sync
1008 * submodule. These functions should be able to run in open-context.
1009 * Therefore we create a new transaction that we later abort once
1010 * the channel program has been evaluated.
1012 tx
= dmu_tx_create_dd(dp
->dp_mos_dir
);
1014 zcp_eval_impl(tx
, B_FALSE
, evalargs
);
1018 dsl_pool_rele(dp
, FTAG
);
1022 zcp_eval(const char *poolname
, const char *program
, boolean_t sync
,
1023 uint64_t instrlimit
, uint64_t memlimit
, nvpair_t
*nvarg
, nvlist_t
*outnvl
)
1027 zcp_eval_arg_t evalargs
;
1029 if (instrlimit
> zfs_lua_max_instrlimit
)
1030 return (SET_ERROR(EINVAL
));
1031 if (memlimit
== 0 || memlimit
> zfs_lua_max_memlimit
)
1032 return (SET_ERROR(EINVAL
));
1034 zcp_alloc_arg_t allocargs
= {
1035 .aa_must_succeed
= B_TRUE
,
1036 .aa_alloc_remaining
= (int64_t)memlimit
,
1037 .aa_alloc_limit
= (int64_t)memlimit
,
1041 * Creates a Lua state with a memory allocator that uses KM_SLEEP.
1042 * This should never fail.
1044 state
= lua_newstate(zcp_lua_alloc
, &allocargs
);
1045 VERIFY(state
!= NULL
);
1046 (void) lua_atpanic(state
, zcp_panic_cb
);
1049 * Load core Lua libraries we want access to.
1051 VERIFY3U(1, ==, luaopen_base(state
));
1053 VERIFY3U(1, ==, luaopen_coroutine(state
));
1054 lua_setglobal(state
, LUA_COLIBNAME
);
1055 VERIFY0(lua_gettop(state
));
1056 VERIFY3U(1, ==, luaopen_string(state
));
1057 lua_setglobal(state
, LUA_STRLIBNAME
);
1058 VERIFY0(lua_gettop(state
));
1059 VERIFY3U(1, ==, luaopen_table(state
));
1060 lua_setglobal(state
, LUA_TABLIBNAME
);
1061 VERIFY0(lua_gettop(state
));
1064 * Load globally visible variables such as errno aliases.
1066 zcp_load_globals(state
);
1067 VERIFY0(lua_gettop(state
));
1070 * Load ZFS-specific modules.
1072 lua_newtable(state
);
1073 VERIFY3U(1, ==, zcp_load_list_lib(state
));
1074 lua_setfield(state
, -2, "list");
1075 VERIFY3U(1, ==, zcp_load_synctask_lib(state
, B_FALSE
));
1076 lua_setfield(state
, -2, "check");
1077 VERIFY3U(1, ==, zcp_load_synctask_lib(state
, B_TRUE
));
1078 lua_setfield(state
, -2, "sync");
1079 VERIFY3U(1, ==, zcp_load_get_lib(state
));
1080 lua_pushcclosure(state
, zcp_debug_info
.func
, 0);
1081 lua_setfield(state
, -2, zcp_debug_info
.name
);
1082 lua_pushcclosure(state
, zcp_exists_info
.func
, 0);
1083 lua_setfield(state
, -2, zcp_exists_info
.name
);
1084 lua_setglobal(state
, "zfs");
1085 VERIFY0(lua_gettop(state
));
1088 * Push the error-callback that calculates Lua stack traces on
1089 * unexpected failures.
1091 lua_pushcfunction(state
, zcp_error_handler
);
1092 VERIFY3U(1, ==, lua_gettop(state
));
1095 * Load the actual script as a function onto the stack as text ("t").
1096 * The only valid error condition is a syntax error in the script.
1097 * ERRMEM should not be possible because our allocator is using
1098 * KM_SLEEP. ERRGCMM should not be possible because we have not added
1099 * any objects with __gc metamethods to the interpreter that could
1102 err
= luaL_loadbufferx(state
, program
, strlen(program
),
1103 "channel program", "t");
1104 if (err
== LUA_ERRSYNTAX
) {
1105 fnvlist_add_string(outnvl
, ZCP_RET_ERROR
,
1106 lua_tostring(state
, -1));
1108 return (SET_ERROR(EINVAL
));
1111 VERIFY3U(2, ==, lua_gettop(state
));
1114 * Convert the input nvlist to a Lua object and put it on top of the
1118 err
= zcp_nvpair_value_to_lua(state
, nvarg
,
1119 errmsg
, sizeof (errmsg
));
1121 fnvlist_add_string(outnvl
, ZCP_RET_ERROR
, errmsg
);
1123 return (SET_ERROR(EINVAL
));
1125 VERIFY3U(3, ==, lua_gettop(state
));
1127 evalargs
.ea_state
= state
;
1128 evalargs
.ea_allocargs
= &allocargs
;
1129 evalargs
.ea_instrlimit
= instrlimit
;
1130 evalargs
.ea_cred
= CRED();
1131 evalargs
.ea_outnvl
= outnvl
;
1132 evalargs
.ea_result
= 0;
1135 err
= dsl_sync_task(poolname
, NULL
,
1136 zcp_eval_sync
, &evalargs
, 0, ZFS_SPACE_CHECK_ZCP_EVAL
);
1138 zcp_pool_error(&evalargs
, poolname
);
1140 zcp_eval_open(&evalargs
, poolname
);
1144 return (evalargs
.ea_result
);
1148 * Retrieve metadata about the currently running channel program.
1151 zcp_run_info(lua_State
*state
)
1155 lua_getfield(state
, LUA_REGISTRYINDEX
, ZCP_RUN_INFO_KEY
);
1156 ri
= lua_touserdata(state
, -1);
1165 * The Lua language allows methods to be called with any number
1166 * of arguments of any type. When calling back into ZFS we need to sanitize
1167 * arguments from channel programs to make sure unexpected arguments or
1168 * arguments of the wrong type result in clear error messages. To do this
1169 * in a uniform way all callbacks from channel programs should use the
1170 * zcp_parse_args() function to interpret inputs.
1172 * Positional vs Keyword Arguments
1173 * ===============================
1175 * Every callback function takes a fixed set of required positional arguments
1176 * and optional keyword arguments. For example, the destroy function takes
1177 * a single positional string argument (the name of the dataset to destroy)
1178 * and an optional "defer" keyword boolean argument. When calling lua functions
1179 * with parentheses, only positional arguments can be used:
1181 * zfs.sync.snapshot("rpool@snap")
1183 * To use keyword arguments functions should be called with a single argument
1184 * that is a lua table containing mappings of integer -> positional arguments
1185 * and string -> keyword arguments:
1187 * zfs.sync.snapshot({1="rpool@snap", defer=true})
1189 * The lua language allows curly braces to be used in place of parenthesis as
1190 * syntactic sugar for this calling convention:
1192 * zfs.sync.snapshot{"rpool@snap", defer=true}
1196 * Throw an error and print the given arguments. If there are too many
1197 * arguments to fit in the output buffer, only the error format string is
1201 zcp_args_error(lua_State
*state
, const char *fname
, const zcp_arg_t
*pargs
,
1202 const zcp_arg_t
*kwargs
, const char *fmt
, ...)
1206 size_t len
= sizeof (errmsg
);
1210 va_start(argp
, fmt
);
1211 VERIFY3U(len
, >, vsnprintf(errmsg
, len
, fmt
, argp
));
1215 * Calculate the total length of the final string, including extra
1216 * formatting characters. If the argument dump would be too large,
1217 * only print the error string.
1219 msglen
= strlen(errmsg
);
1220 msglen
+= strlen(fname
) + 4; /* : + {} + null terminator */
1221 for (i
= 0; pargs
[i
].za_name
!= NULL
; i
++) {
1222 msglen
+= strlen(pargs
[i
].za_name
);
1223 msglen
+= strlen(lua_typename(state
, pargs
[i
].za_lua_type
));
1224 if (pargs
[i
+ 1].za_name
!= NULL
|| kwargs
[0].za_name
!= NULL
)
1225 msglen
+= 5; /* < + ( + )> + , */
1227 msglen
+= 4; /* < + ( + )> */
1229 for (i
= 0; kwargs
[i
].za_name
!= NULL
; i
++) {
1230 msglen
+= strlen(kwargs
[i
].za_name
);
1231 msglen
+= strlen(lua_typename(state
, kwargs
[i
].za_lua_type
));
1232 if (kwargs
[i
+ 1].za_name
!= NULL
)
1233 msglen
+= 4; /* =( + ) + , */
1235 msglen
+= 3; /* =( + ) */
1239 (void) luaL_error(state
, errmsg
);
1241 VERIFY3U(len
, >, strlcat(errmsg
, ": ", len
));
1242 VERIFY3U(len
, >, strlcat(errmsg
, fname
, len
));
1243 VERIFY3U(len
, >, strlcat(errmsg
, "{", len
));
1244 for (i
= 0; pargs
[i
].za_name
!= NULL
; i
++) {
1245 VERIFY3U(len
, >, strlcat(errmsg
, "<", len
));
1246 VERIFY3U(len
, >, strlcat(errmsg
, pargs
[i
].za_name
, len
));
1247 VERIFY3U(len
, >, strlcat(errmsg
, "(", len
));
1248 VERIFY3U(len
, >, strlcat(errmsg
,
1249 lua_typename(state
, pargs
[i
].za_lua_type
), len
));
1250 VERIFY3U(len
, >, strlcat(errmsg
, ")>", len
));
1251 if (pargs
[i
+ 1].za_name
!= NULL
|| kwargs
[0].za_name
!= NULL
) {
1252 VERIFY3U(len
, >, strlcat(errmsg
, ", ", len
));
1255 for (i
= 0; kwargs
[i
].za_name
!= NULL
; i
++) {
1256 VERIFY3U(len
, >, strlcat(errmsg
, kwargs
[i
].za_name
, len
));
1257 VERIFY3U(len
, >, strlcat(errmsg
, "=(", len
));
1258 VERIFY3U(len
, >, strlcat(errmsg
,
1259 lua_typename(state
, kwargs
[i
].za_lua_type
), len
));
1260 VERIFY3U(len
, >, strlcat(errmsg
, ")", len
));
1261 if (kwargs
[i
+ 1].za_name
!= NULL
) {
1262 VERIFY3U(len
, >, strlcat(errmsg
, ", ", len
));
1265 VERIFY3U(len
, >, strlcat(errmsg
, "}", len
));
1267 (void) luaL_error(state
, errmsg
);
1268 panic("unreachable code");
1272 zcp_parse_table_args(lua_State
*state
, const char *fname
,
1273 const zcp_arg_t
*pargs
, const zcp_arg_t
*kwargs
)
1278 for (i
= 0; pargs
[i
].za_name
!= NULL
; i
++) {
1280 * Check the table for this positional argument, leaving it
1281 * on the top of the stack once we finish validating it.
1283 lua_pushinteger(state
, i
+ 1);
1284 lua_gettable(state
, 1);
1286 type
= lua_type(state
, -1);
1287 if (type
== LUA_TNIL
) {
1288 zcp_args_error(state
, fname
, pargs
, kwargs
,
1289 "too few arguments");
1290 panic("unreachable code");
1291 } else if (type
!= pargs
[i
].za_lua_type
) {
1292 zcp_args_error(state
, fname
, pargs
, kwargs
,
1293 "arg %d wrong type (is '%s', expected '%s')",
1294 i
+ 1, lua_typename(state
, type
),
1295 lua_typename(state
, pargs
[i
].za_lua_type
));
1296 panic("unreachable code");
1300 * Remove the positional argument from the table.
1302 lua_pushinteger(state
, i
+ 1);
1304 lua_settable(state
, 1);
1307 for (i
= 0; kwargs
[i
].za_name
!= NULL
; i
++) {
1309 * Check the table for this keyword argument, which may be
1310 * nil if it was omitted. Leave the value on the top of
1311 * the stack after validating it.
1313 lua_getfield(state
, 1, kwargs
[i
].za_name
);
1315 type
= lua_type(state
, -1);
1316 if (type
!= LUA_TNIL
&& type
!= kwargs
[i
].za_lua_type
) {
1317 zcp_args_error(state
, fname
, pargs
, kwargs
,
1318 "kwarg '%s' wrong type (is '%s', expected '%s')",
1319 kwargs
[i
].za_name
, lua_typename(state
, type
),
1320 lua_typename(state
, kwargs
[i
].za_lua_type
));
1321 panic("unreachable code");
1325 * Remove the keyword argument from the table.
1328 lua_setfield(state
, 1, kwargs
[i
].za_name
);
1332 * Any entries remaining in the table are invalid inputs, print
1333 * an error message based on what the entry is.
1336 if (lua_next(state
, 1)) {
1337 if (lua_isnumber(state
, -2) && lua_tointeger(state
, -2) > 0) {
1338 zcp_args_error(state
, fname
, pargs
, kwargs
,
1339 "too many positional arguments");
1340 } else if (lua_isstring(state
, -2)) {
1341 zcp_args_error(state
, fname
, pargs
, kwargs
,
1342 "invalid kwarg '%s'", lua_tostring(state
, -2));
1344 zcp_args_error(state
, fname
, pargs
, kwargs
,
1345 "kwarg keys must be strings");
1347 panic("unreachable code");
1350 lua_remove(state
, 1);
1354 zcp_parse_pos_args(lua_State
*state
, const char *fname
, const zcp_arg_t
*pargs
,
1355 const zcp_arg_t
*kwargs
)
1360 for (i
= 0; pargs
[i
].za_name
!= NULL
; i
++) {
1361 type
= lua_type(state
, i
+ 1);
1362 if (type
== LUA_TNONE
) {
1363 zcp_args_error(state
, fname
, pargs
, kwargs
,
1364 "too few arguments");
1365 panic("unreachable code");
1366 } else if (type
!= pargs
[i
].za_lua_type
) {
1367 zcp_args_error(state
, fname
, pargs
, kwargs
,
1368 "arg %d wrong type (is '%s', expected '%s')",
1369 i
+ 1, lua_typename(state
, type
),
1370 lua_typename(state
, pargs
[i
].za_lua_type
));
1371 panic("unreachable code");
1374 if (lua_gettop(state
) != i
) {
1375 zcp_args_error(state
, fname
, pargs
, kwargs
,
1376 "too many positional arguments");
1377 panic("unreachable code");
1380 for (i
= 0; kwargs
[i
].za_name
!= NULL
; i
++) {
1386 * Checks the current Lua stack against an expected set of positional and
1387 * keyword arguments. If the stack does not match the expected arguments
1388 * aborts the current channel program with a useful error message, otherwise
1389 * it re-arranges the stack so that it contains the positional arguments
1390 * followed by the keyword argument values in declaration order. Any missing
1391 * keyword argument will be represented by a nil value on the stack.
1393 * If the stack contains exactly one argument of type LUA_TTABLE the curly
1394 * braces calling convention is assumed, otherwise the stack is parsed for
1395 * positional arguments only.
1397 * This function should be used by every function callback. It should be called
1398 * before the callback manipulates the Lua stack as it assumes the stack
1399 * represents the function arguments.
1402 zcp_parse_args(lua_State
*state
, const char *fname
, const zcp_arg_t
*pargs
,
1403 const zcp_arg_t
*kwargs
)
1405 if (lua_gettop(state
) == 1 && lua_istable(state
, 1)) {
1406 zcp_parse_table_args(state
, fname
, pargs
, kwargs
);
1408 zcp_parse_pos_args(state
, fname
, pargs
, kwargs
);