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.
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 <util/sscanf.h>
106 #define KM_NORMALPRI 0
109 #define ZCP_NVLIST_MAX_DEPTH 20
111 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 *,
122 typedef struct zcp_alloc_arg {
123 boolean_t aa_must_succeed;
124 int64_t aa_alloc_remaining;
125 int64_t aa_alloc_limit;
128 typedef struct zcp_eval_arg {
130 zcp_alloc_arg_t *ea_allocargs;
134 uint64_t ea_instrlimit;
139 zcp_eval_check(void *arg, dmu_tx_t *tx)
145 * The outer-most error callback handler for use with lua_pcall(). On
146 * error Lua will call this callback with a single argument that
147 * represents the error value. In most cases this will be a string
148 * containing an error message, but channel programs can use Lua's
149 * error() function to return arbitrary objects as errors. This callback
150 * returns (on the Lua stack) the original error object along with a traceback.
152 * Fatal Lua errors can occur while resources are held, so we also call any
153 * registered cleanup function here.
156 zcp_error_handler(lua_State *state)
162 VERIFY3U(1, ==, lua_gettop(state));
163 msg = lua_tostring(state, 1);
164 luaL_traceback(state, state, msg, 1);
169 zcp_argerror(lua_State *state, int narg, const char *msg, ...)
173 va_start(alist, msg);
174 const char *buf = lua_pushvfstring(state, msg, alist);
177 return (luaL_argerror(state, narg, buf));
181 * Install a new cleanup function, which will be invoked with the given
182 * opaque argument if a fatal error causes the Lua interpreter to longjump out
183 * of a function call.
185 * If an error occurs, the cleanup function will be invoked exactly once and
189 zcp_register_cleanup(lua_State *state, zcp_cleanup_t cleanfunc, void *cleanarg)
191 zcp_run_info_t *ri = zcp_run_info(state);
193 * A cleanup function should always be explicitly removed before
194 * installing a new one to avoid accidental clobbering.
196 ASSERT3P(ri->zri_cleanup, ==, NULL);
198 ri->zri_cleanup = cleanfunc;
199 ri->zri_cleanup_arg = cleanarg;
203 zcp_clear_cleanup(lua_State *state)
205 zcp_run_info_t *ri = zcp_run_info(state);
207 ri->zri_cleanup = NULL;
208 ri->zri_cleanup_arg = NULL;
212 * If it exists, execute the currently set cleanup function then unregister it.
215 zcp_cleanup(lua_State *state)
217 zcp_run_info_t *ri = zcp_run_info(state);
219 if (ri->zri_cleanup != NULL) {
220 ri->zri_cleanup(ri->zri_cleanup_arg);
221 zcp_clear_cleanup(state);
226 * Convert the lua table at the given index on the Lua stack to an nvlist
229 * If the table can not be converted for any reason, NULL is returned and
230 * an error message is pushed onto the Lua stack.
233 zcp_table_to_nvlist(lua_State *state, int index, int depth)
237 * Converting a Lua table to an nvlist with key uniqueness checking is
238 * O(n^2) in the number of keys in the nvlist, which can take a long
239 * time when we return a large table from a channel program.
240 * Furthermore, Lua's table interface *almost* guarantees unique keys
241 * on its own (details below). Therefore, we don't use fnvlist_alloc()
242 * here to avoid the built-in uniqueness checking.
244 * The *almost* is because it's possible to have key collisions between
245 * e.g. the string "1" and the number 1, or the string "true" and the
246 * boolean true, so we explicitly check that when we're looking at a
247 * key which is an integer / boolean or a string that can be parsed as
248 * one of those types. In the worst case this could still devolve into
249 * O(n^2), so we only start doing these checks on boolean/integer keys
250 * once we've seen a string key which fits this weird usage pattern.
252 * Ultimately, we still want callers to know that the keys in this
253 * nvlist are unique, so before we return this we set the nvlist's
254 * flags to reflect that.
256 VERIFY0(nvlist_alloc(&nvl, 0, KM_SLEEP));
259 * Push an empty stack slot where lua_next() will store each
263 boolean_t saw_str_could_collide = B_FALSE;
264 while (lua_next(state, index) != 0) {
266 * The next key-value pair from the table at index is
267 * now on the stack, with the key at stack slot -2 and
268 * the value at slot -1.
272 const char *key = NULL;
273 boolean_t key_could_collide = B_FALSE;
275 switch (lua_type(state, -2)) {
277 key = lua_tostring(state, -2);
279 /* check if this could collide with a number or bool */
282 if ((sscanf(key, "%lld%n", &tmp, &parselen) > 0 &&
283 parselen == strlen(key)) ||
284 strcmp(key, "true") == 0 ||
285 strcmp(key, "false") == 0) {
286 key_could_collide = B_TRUE;
287 saw_str_could_collide = B_TRUE;
291 key = (lua_toboolean(state, -2) == B_TRUE ?
293 if (saw_str_could_collide) {
294 key_could_collide = B_TRUE;
298 VERIFY3U(sizeof (buf), >,
299 snprintf(buf, sizeof (buf), "%lld",
300 (longlong_t)lua_tonumber(state, -2)));
302 if (saw_str_could_collide) {
303 key_could_collide = B_TRUE;
308 (void) lua_pushfstring(state, "Invalid key "
309 "type '%s' in table",
310 lua_typename(state, lua_type(state, -2)));
314 * Check for type-mismatched key collisions, and throw an error.
316 if (key_could_collide && nvlist_exists(nvl, key)) {
318 (void) lua_pushfstring(state, "Collision of "
319 "key '%s' in table", key);
323 * Recursively convert the table value and insert into
324 * the new nvlist with the parsed key. To prevent
325 * stack overflow on circular or heavily nested tables,
326 * we track the current nvlist depth.
328 if (depth >= ZCP_NVLIST_MAX_DEPTH) {
330 (void) lua_pushfstring(state, "Maximum table "
331 "depth (%d) exceeded for table",
332 ZCP_NVLIST_MAX_DEPTH);
335 err = zcp_lua_to_nvlist_impl(state, -1, nvl, key,
340 * Error message has been pushed to the lua
341 * stack by the recursive call.
346 * Pop the value pushed by lua_next().
352 * Mark the nvlist as having unique keys. This is a little ugly, but we
353 * ensured above that there are no duplicate keys in the nvlist.
355 nvl->nvl_nvflag |= NV_UNIQUE_NAME;
361 * Convert a value from the given index into the lua stack to an nvpair, adding
362 * it to an nvlist with the given key.
364 * Values are converted as follows:
369 * nil -> boolean (no value)
371 * Lua tables are converted to nvlists and then inserted. The table's keys
372 * are converted to strings then used as keys in the nvlist to store each table
373 * element. Keys are converted as follows:
375 * string -> no change
377 * boolean -> "true" | "false"
380 * In the case of a key collision, an error is thrown.
382 * If an error is encountered, a nonzero error code is returned, and an error
383 * string will be pushed onto the Lua stack.
386 zcp_lua_to_nvlist_impl(lua_State *state, int index, nvlist_t *nvl,
387 const char *key, int depth)
390 * Verify that we have enough remaining space in the lua stack to parse
391 * a key-value pair and push an error.
393 if (!lua_checkstack(state, 3)) {
394 (void) lua_pushstring(state, "Lua stack overflow");
398 index = lua_absindex(state, index);
400 switch (lua_type(state, index)) {
402 fnvlist_add_boolean(nvl, key);
405 fnvlist_add_boolean_value(nvl, key,
406 lua_toboolean(state, index));
409 fnvlist_add_int64(nvl, key, lua_tonumber(state, index));
412 fnvlist_add_string(nvl, key, lua_tostring(state, index));
415 nvlist_t *value_nvl = zcp_table_to_nvlist(state, index, depth);
416 if (value_nvl == NULL)
419 fnvlist_add_nvlist(nvl, key, value_nvl);
420 fnvlist_free(value_nvl);
424 (void) lua_pushfstring(state,
425 "Invalid value type '%s' for key '%s'",
426 lua_typename(state, lua_type(state, index)), key);
434 * Convert a lua value to an nvpair, adding it to an nvlist with the given key.
437 zcp_lua_to_nvlist(lua_State *state, int index, nvlist_t *nvl, const char *key)
440 * On error, zcp_lua_to_nvlist_impl pushes an error string onto the Lua
441 * stack before returning with a nonzero error code. If an error is
442 * returned, throw a fatal lua error with the given string.
444 if (zcp_lua_to_nvlist_impl(state, index, nvl, key, 0) != 0)
445 (void) lua_error(state);
449 zcp_lua_to_nvlist_helper(lua_State *state)
451 nvlist_t *nv = (nvlist_t *)lua_touserdata(state, 2);
452 const char *key = (const char *)lua_touserdata(state, 1);
453 zcp_lua_to_nvlist(state, 3, nv, key);
458 zcp_convert_return_values(lua_State *state, nvlist_t *nvl,
459 const char *key, zcp_eval_arg_t *evalargs)
462 lua_pushcfunction(state, zcp_lua_to_nvlist_helper);
463 lua_pushlightuserdata(state, (char *)key);
464 lua_pushlightuserdata(state, nvl);
465 lua_pushvalue(state, 1);
466 lua_remove(state, 1);
467 err = lua_pcall(state, 3, 0, 0); /* zcp_lua_to_nvlist_helper */
469 zcp_lua_to_nvlist(state, 1, nvl, ZCP_RET_ERROR);
470 evalargs->ea_result = SET_ERROR(ECHRNG);
475 * Push a Lua table representing nvl onto the stack. If it can't be
476 * converted, return EINVAL, fill in errbuf, and push nothing. errbuf may
477 * be specified as NULL, in which case no error string will be output.
479 * Most nvlists are converted as simple key->value Lua tables, but we make
480 * an exception for the case where all nvlist entries are BOOLEANs (a string
481 * key without a value). In Lua, a table key pointing to a value of Nil
482 * (no value) is equivalent to the key not existing, so a BOOLEAN nvlist
483 * entry can't be directly converted to a Lua table entry. Nvlists of entirely
484 * BOOLEAN entries are frequently used to pass around lists of datasets, so for
485 * convenience we check for this case, and convert it to a simple Lua array of
489 zcp_nvlist_to_lua(lua_State *state, nvlist_t *nvl,
490 char *errbuf, int errbuf_len)
494 boolean_t has_values = B_FALSE;
496 * If the list doesn't have any values, just convert it to a string
499 for (pair = nvlist_next_nvpair(nvl, NULL);
500 pair != NULL; pair = nvlist_next_nvpair(nvl, pair)) {
501 if (nvpair_type(pair) != DATA_TYPE_BOOLEAN) {
508 for (pair = nvlist_next_nvpair(nvl, NULL);
509 pair != NULL; pair = nvlist_next_nvpair(nvl, pair)) {
510 (void) lua_pushinteger(state, i);
511 (void) lua_pushstring(state, nvpair_name(pair));
512 (void) lua_settable(state, -3);
516 for (pair = nvlist_next_nvpair(nvl, NULL);
517 pair != NULL; pair = nvlist_next_nvpair(nvl, pair)) {
518 int err = zcp_nvpair_value_to_lua(state, pair,
524 (void) lua_setfield(state, -2, nvpair_name(pair));
531 * Push a Lua object representing the value of "pair" onto the stack.
533 * Only understands boolean_value, string, int64, nvlist,
534 * string_array, and int64_array type values. For other
535 * types, returns EINVAL, fills in errbuf, and pushes nothing.
538 zcp_nvpair_value_to_lua(lua_State *state, nvpair_t *pair,
539 char *errbuf, int errbuf_len)
548 switch (nvpair_type(pair)) {
549 case DATA_TYPE_BOOLEAN_VALUE:
550 (void) lua_pushboolean(state,
551 fnvpair_value_boolean_value(pair));
553 case DATA_TYPE_STRING:
554 (void) lua_pushstring(state, fnvpair_value_string(pair));
556 case DATA_TYPE_INT64:
557 (void) lua_pushinteger(state, fnvpair_value_int64(pair));
559 case DATA_TYPE_NVLIST:
560 err = zcp_nvlist_to_lua(state,
561 fnvpair_value_nvlist(pair), errbuf, errbuf_len);
563 case DATA_TYPE_STRING_ARRAY: {
566 (void) nvpair_value_string_array(pair, &strarr, &nelem);
568 for (int i = 0; i < nelem; i++) {
569 (void) lua_pushinteger(state, i + 1);
570 (void) lua_pushstring(state, strarr[i]);
571 (void) lua_settable(state, -3);
575 case DATA_TYPE_UINT64_ARRAY: {
578 (void) nvpair_value_uint64_array(pair, &intarr, &nelem);
580 for (int i = 0; i < nelem; i++) {
581 (void) lua_pushinteger(state, i + 1);
582 (void) lua_pushinteger(state, intarr[i]);
583 (void) lua_settable(state, -3);
587 case DATA_TYPE_INT64_ARRAY: {
590 (void) nvpair_value_int64_array(pair, &intarr, &nelem);
592 for (int i = 0; i < nelem; i++) {
593 (void) lua_pushinteger(state, i + 1);
594 (void) lua_pushinteger(state, intarr[i]);
595 (void) lua_settable(state, -3);
600 if (errbuf != NULL) {
601 (void) snprintf(errbuf, errbuf_len,
602 "Unhandled nvpair type %d for key '%s'",
603 nvpair_type(pair), nvpair_name(pair));
612 zcp_dataset_hold_error(lua_State *state, dsl_pool_t *dp, const char *dsname,
615 if (error == ENOENT) {
616 (void) zcp_argerror(state, 1, "no such dataset '%s'", dsname);
617 return (0); /* not reached; zcp_argerror will longjmp */
618 } else if (error == EXDEV) {
619 (void) zcp_argerror(state, 1,
620 "dataset '%s' is not in the target pool '%s'",
621 dsname, spa_name(dp->dp_spa));
622 return (0); /* not reached; zcp_argerror will longjmp */
623 } else if (error == EIO) {
624 (void) luaL_error(state,
625 "I/O error while accessing dataset '%s'", dsname);
626 return (0); /* not reached; luaL_error will longjmp */
627 } else if (error != 0) {
628 (void) luaL_error(state,
629 "unexpected error %d while accessing dataset '%s'",
631 return (0); /* not reached; luaL_error will longjmp */
637 * Note: will longjmp (via lua_error()) on error.
638 * Assumes that the dsname is argument #1 (for error reporting purposes).
641 zcp_dataset_hold(lua_State *state, dsl_pool_t *dp, const char *dsname,
645 int error = dsl_dataset_hold(dp, dsname, tag, &ds);
646 (void) zcp_dataset_hold_error(state, dp, dsname, error);
650 static int zcp_debug(lua_State *);
651 static zcp_lib_info_t zcp_debug_info = {
655 { .za_name = "debug string", .za_lua_type = LUA_TSTRING},
664 zcp_debug(lua_State *state)
666 const char *dbgstring;
667 zcp_run_info_t *ri = zcp_run_info(state);
668 zcp_lib_info_t *libinfo = &zcp_debug_info;
670 zcp_parse_args(state, libinfo->name, libinfo->pargs, libinfo->kwargs);
672 dbgstring = lua_tostring(state, 1);
674 zfs_dbgmsg("txg %lld ZCP: %s", ri->zri_tx->tx_txg, dbgstring);
679 static int zcp_exists(lua_State *);
680 static zcp_lib_info_t zcp_exists_info = {
684 { .za_name = "dataset", .za_lua_type = LUA_TSTRING},
693 zcp_exists(lua_State *state)
695 zcp_run_info_t *ri = zcp_run_info(state);
696 dsl_pool_t *dp = ri->zri_pool;
697 zcp_lib_info_t *libinfo = &zcp_exists_info;
699 zcp_parse_args(state, libinfo->name, libinfo->pargs, libinfo->kwargs);
701 const char *dsname = lua_tostring(state, 1);
704 int error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
706 dsl_dataset_rele(ds, FTAG);
707 lua_pushboolean(state, B_TRUE);
708 } else if (error == ENOENT) {
709 lua_pushboolean(state, B_FALSE);
710 } else if (error == EXDEV) {
711 return (luaL_error(state, "dataset '%s' is not in the "
712 "target pool", dsname));
713 } else if (error == EIO) {
714 return (luaL_error(state, "I/O error opening dataset '%s'",
716 } else if (error != 0) {
717 return (luaL_error(state, "unexpected error %d", error));
724 * Allocate/realloc/free a buffer for the lua interpreter.
726 * When nsize is 0, behaves as free() and returns NULL.
728 * If ptr is NULL, behaves as malloc() and returns an allocated buffer of size
731 * Otherwise, behaves as realloc(), changing the allocation from osize to nsize.
732 * Shrinking the buffer size never fails.
734 * The original allocated buffer size is stored as a uint64 at the beginning of
735 * the buffer to avoid actually reallocating when shrinking a buffer, since lua
736 * requires that this operation never fail.
739 zcp_lua_alloc(void *ud, void *ptr, size_t osize, size_t nsize)
741 zcp_alloc_arg_t *allocargs = ud;
742 int flags = (allocargs->aa_must_succeed) ?
743 KM_SLEEP : (KM_NOSLEEP | KM_NORMALPRI);
747 int64_t *allocbuf = (int64_t *)ptr - 1;
748 int64_t allocsize = *allocbuf;
749 ASSERT3S(allocsize, >, 0);
750 ASSERT3S(allocargs->aa_alloc_remaining + allocsize, <=,
751 allocargs->aa_alloc_limit);
752 allocargs->aa_alloc_remaining += allocsize;
753 vmem_free(allocbuf, allocsize);
756 } else if (ptr == NULL) {
758 int64_t allocsize = nsize + sizeof (int64_t);
760 if (!allocargs->aa_must_succeed &&
762 allocsize > allocargs->aa_alloc_remaining)) {
766 allocbuf = vmem_alloc(allocsize, flags);
767 if (allocbuf == NULL) {
770 allocargs->aa_alloc_remaining -= allocsize;
772 *allocbuf = allocsize;
773 return (allocbuf + 1);
774 } else if (nsize <= osize) {
776 * If shrinking the buffer, lua requires that the reallocation
781 ASSERT3U(nsize, >, osize);
783 uint64_t *luabuf = zcp_lua_alloc(ud, NULL, 0, nsize);
784 if (luabuf == NULL) {
787 (void) memcpy(luabuf, ptr, osize);
788 VERIFY3P(zcp_lua_alloc(ud, ptr, osize, 0), ==, NULL);
795 zcp_lua_counthook(lua_State *state, lua_Debug *ar)
798 * If we're called, check how many instructions the channel program has
799 * executed so far, and compare against the limit.
801 lua_getfield(state, LUA_REGISTRYINDEX, ZCP_RUN_INFO_KEY);
802 zcp_run_info_t *ri = lua_touserdata(state, -1);
804 ri->zri_curinstrs += zfs_lua_check_instrlimit_interval;
805 if (ri->zri_maxinstrs != 0 && ri->zri_curinstrs > ri->zri_maxinstrs) {
806 ri->zri_timed_out = B_TRUE;
807 (void) lua_pushstring(state,
808 "Channel program timed out.");
809 (void) lua_error(state);
814 zcp_panic_cb(lua_State *state)
816 panic("unprotected error in call to Lua API (%s)\n",
817 lua_tostring(state, -1));
822 zcp_eval_sync(void *arg, dmu_tx_t *tx)
826 zcp_eval_arg_t *evalargs = arg;
827 lua_State *state = evalargs->ea_state;
830 * Open context should have setup the stack to contain:
831 * 1: Error handler callback
832 * 2: Script to run (converted to a Lua function)
833 * 3: nvlist input to function (converted to Lua table or nil)
835 VERIFY3U(3, ==, lua_gettop(state));
838 * Store the zcp_run_info_t struct for this run in the Lua registry.
839 * Registry entries are not directly accessible by the Lua scripts but
840 * can be accessed by our callbacks.
842 ri.zri_space_used = 0;
843 ri.zri_pool = dmu_tx_pool(tx);
844 ri.zri_cred = evalargs->ea_cred;
846 ri.zri_timed_out = B_FALSE;
847 ri.zri_cleanup = NULL;
848 ri.zri_cleanup_arg = NULL;
849 ri.zri_curinstrs = 0;
850 ri.zri_maxinstrs = evalargs->ea_instrlimit;
852 lua_pushlightuserdata(state, &ri);
853 lua_setfield(state, LUA_REGISTRYINDEX, ZCP_RUN_INFO_KEY);
854 VERIFY3U(3, ==, lua_gettop(state));
857 * Tell the Lua interpreter to call our handler every count
858 * instructions. Channel programs that execute too many instructions
859 * should die with ETIME.
861 (void) lua_sethook(state, zcp_lua_counthook, LUA_MASKCOUNT,
862 zfs_lua_check_instrlimit_interval);
865 * Tell the Lua memory allocator to stop using KM_SLEEP before handing
866 * off control to the channel program. Channel programs that use too
867 * much memory should die with ENOSPC.
869 evalargs->ea_allocargs->aa_must_succeed = B_FALSE;
872 * Call the Lua function that open-context passed us. This pops the
873 * function and its input from the stack and pushes any return
876 err = lua_pcall(state, 1, LUA_MULTRET, 1);
879 * Let Lua use KM_SLEEP while we interpret the return values.
881 evalargs->ea_allocargs->aa_must_succeed = B_TRUE;
884 * Remove the error handler callback from the stack. At this point,
885 * if there is a cleanup function registered, then it was registered
886 * but never run or removed, which should never occur.
888 ASSERT3P(ri.zri_cleanup, ==, NULL);
889 lua_remove(state, 1);
894 * Lua supports returning multiple values in a single return
895 * statement. Return values will have been pushed onto the
900 * To simplify the process of retrieving a return value from a
901 * channel program, we disallow returning more than one value
902 * to ZFS from the Lua script, yielding a singleton return
903 * nvlist of the form { "return": Return value 1 }.
905 int return_count = lua_gettop(state);
907 if (return_count == 1) {
908 evalargs->ea_result = 0;
909 zcp_convert_return_values(state, evalargs->ea_outnvl,
910 ZCP_RET_RETURN, evalargs);
911 } else if (return_count > 1) {
912 evalargs->ea_result = SET_ERROR(ECHRNG);
913 (void) lua_pushfstring(state, "Multiple return "
914 "values not supported");
915 zcp_convert_return_values(state, evalargs->ea_outnvl,
916 ZCP_RET_ERROR, evalargs);
923 * The channel program encountered a fatal error within the
924 * script, such as failing an assertion, or calling a function
925 * with incompatible arguments. The error value and the
926 * traceback generated by zcp_error_handler() should be on the
929 VERIFY3U(1, ==, lua_gettop(state));
930 if (ri.zri_timed_out) {
931 evalargs->ea_result = SET_ERROR(ETIME);
933 evalargs->ea_result = SET_ERROR(ECHRNG);
936 zcp_convert_return_values(state, evalargs->ea_outnvl,
937 ZCP_RET_ERROR, evalargs);
939 if (evalargs->ea_result == ETIME &&
940 evalargs->ea_outnvl != NULL) {
941 (void) nvlist_add_uint64(evalargs->ea_outnvl,
942 ZCP_ARG_INSTRLIMIT, ri.zri_curinstrs);
948 * The channel program encountered a fatal error within the
949 * script, and we encountered another error while trying to
950 * compute the traceback in zcp_error_handler(). We can only
951 * return the error message.
953 VERIFY3U(1, ==, lua_gettop(state));
954 if (ri.zri_timed_out) {
955 evalargs->ea_result = SET_ERROR(ETIME);
957 evalargs->ea_result = SET_ERROR(ECHRNG);
960 zcp_convert_return_values(state, evalargs->ea_outnvl,
961 ZCP_RET_ERROR, evalargs);
966 * Lua ran out of memory while running the channel program.
967 * There's not much we can do.
969 evalargs->ea_result = SET_ERROR(ENOSPC);
977 zcp_eval(const char *poolname, const char *program, uint64_t instrlimit,
978 uint64_t memlimit, nvpair_t *nvarg, nvlist_t *outnvl)
982 zcp_eval_arg_t evalargs;
984 if (instrlimit > zfs_lua_max_instrlimit)
985 return (SET_ERROR(EINVAL));
986 if (memlimit == 0 || memlimit > zfs_lua_max_memlimit)
987 return (SET_ERROR(EINVAL));
989 zcp_alloc_arg_t allocargs = {
990 .aa_must_succeed = B_TRUE,
991 .aa_alloc_remaining = (int64_t)memlimit,
992 .aa_alloc_limit = (int64_t)memlimit,
996 * Creates a Lua state with a memory allocator that uses KM_SLEEP.
997 * This should never fail.
999 state = lua_newstate(zcp_lua_alloc, &allocargs);
1000 VERIFY(state != NULL);
1001 (void) lua_atpanic(state, zcp_panic_cb);
1004 * Load core Lua libraries we want access to.
1006 VERIFY3U(1, ==, luaopen_base(state));
1008 VERIFY3U(1, ==, luaopen_coroutine(state));
1009 lua_setglobal(state, LUA_COLIBNAME);
1010 VERIFY0(lua_gettop(state));
1011 VERIFY3U(1, ==, luaopen_string(state));
1012 lua_setglobal(state, LUA_STRLIBNAME);
1013 VERIFY0(lua_gettop(state));
1014 VERIFY3U(1, ==, luaopen_table(state));
1015 lua_setglobal(state, LUA_TABLIBNAME);
1016 VERIFY0(lua_gettop(state));
1019 * Load globally visible variables such as errno aliases.
1021 zcp_load_globals(state);
1022 VERIFY0(lua_gettop(state));
1025 * Load ZFS-specific modules.
1027 lua_newtable(state);
1028 VERIFY3U(1, ==, zcp_load_list_lib(state));
1029 lua_setfield(state, -2, "list");
1030 VERIFY3U(1, ==, zcp_load_synctask_lib(state, B_FALSE));
1031 lua_setfield(state, -2, "check");
1032 VERIFY3U(1, ==, zcp_load_synctask_lib(state, B_TRUE));
1033 lua_setfield(state, -2, "sync");
1034 VERIFY3U(1, ==, zcp_load_get_lib(state));
1035 lua_pushcclosure(state, zcp_debug_info.func, 0);
1036 lua_setfield(state, -2, zcp_debug_info.name);
1037 lua_pushcclosure(state, zcp_exists_info.func, 0);
1038 lua_setfield(state, -2, zcp_exists_info.name);
1039 lua_setglobal(state, "zfs");
1040 VERIFY0(lua_gettop(state));
1043 * Push the error-callback that calculates Lua stack traces on
1044 * unexpected failures.
1046 lua_pushcfunction(state, zcp_error_handler);
1047 VERIFY3U(1, ==, lua_gettop(state));
1050 * Load the actual script as a function onto the stack as text ("t").
1051 * The only valid error condition is a syntax error in the script.
1052 * ERRMEM should not be possible because our allocator is using
1053 * KM_SLEEP. ERRGCMM should not be possible because we have not added
1054 * any objects with __gc metamethods to the interpreter that could
1057 err = luaL_loadbufferx(state, program, strlen(program),
1058 "channel program", "t");
1059 if (err == LUA_ERRSYNTAX) {
1060 fnvlist_add_string(outnvl, ZCP_RET_ERROR,
1061 lua_tostring(state, -1));
1063 return (SET_ERROR(EINVAL));
1066 VERIFY3U(2, ==, lua_gettop(state));
1069 * Convert the input nvlist to a Lua object and put it on top of the
1073 err = zcp_nvpair_value_to_lua(state, nvarg,
1074 errmsg, sizeof (errmsg));
1076 fnvlist_add_string(outnvl, ZCP_RET_ERROR, errmsg);
1078 return (SET_ERROR(EINVAL));
1080 VERIFY3U(3, ==, lua_gettop(state));
1082 evalargs.ea_state = state;
1083 evalargs.ea_allocargs = &allocargs;
1084 evalargs.ea_instrlimit = instrlimit;
1085 evalargs.ea_cred = CRED();
1086 evalargs.ea_outnvl = outnvl;
1087 evalargs.ea_result = 0;
1089 VERIFY0(dsl_sync_task(poolname, zcp_eval_check,
1090 zcp_eval_sync, &evalargs, 0, ZFS_SPACE_CHECK_NONE));
1094 return (evalargs.ea_result);
1098 * Retrieve metadata about the currently running channel program.
1101 zcp_run_info(lua_State *state)
1105 lua_getfield(state, LUA_REGISTRYINDEX, ZCP_RUN_INFO_KEY);
1106 ri = lua_touserdata(state, -1);
1115 * The Lua language allows methods to be called with any number
1116 * of arguments of any type. When calling back into ZFS we need to sanitize
1117 * arguments from channel programs to make sure unexpected arguments or
1118 * arguments of the wrong type result in clear error messages. To do this
1119 * in a uniform way all callbacks from channel programs should use the
1120 * zcp_parse_args() function to interpret inputs.
1122 * Positional vs Keyword Arguments
1123 * ===============================
1125 * Every callback function takes a fixed set of required positional arguments
1126 * and optional keyword arguments. For example, the destroy function takes
1127 * a single positional string argument (the name of the dataset to destroy)
1128 * and an optional "defer" keyword boolean argument. When calling lua functions
1129 * with parentheses, only positional arguments can be used:
1131 * zfs.sync.snapshot("rpool@snap")
1133 * To use keyword arguments functions should be called with a single argument
1134 * that is a lua table containing mappings of integer -> positional arguments
1135 * and string -> keyword arguments:
1137 * zfs.sync.snapshot({1="rpool@snap", defer=true})
1139 * The lua language allows curly braces to be used in place of parenthesis as
1140 * syntactic sugar for this calling convention:
1142 * zfs.sync.snapshot{"rpool@snap", defer=true}
1146 * Throw an error and print the given arguments. If there are too many
1147 * arguments to fit in the output buffer, only the error format string is
1151 zcp_args_error(lua_State *state, const char *fname, const zcp_arg_t *pargs,
1152 const zcp_arg_t *kwargs, const char *fmt, ...)
1156 size_t len = sizeof (errmsg);
1160 va_start(argp, fmt);
1161 VERIFY3U(len, >, vsnprintf(errmsg, len, fmt, argp));
1165 * Calculate the total length of the final string, including extra
1166 * formatting characters. If the argument dump would be too large,
1167 * only print the error string.
1169 msglen = strlen(errmsg);
1170 msglen += strlen(fname) + 4; /* : + {} + null terminator */
1171 for (i = 0; pargs[i].za_name != NULL; i++) {
1172 msglen += strlen(pargs[i].za_name);
1173 msglen += strlen(lua_typename(state, pargs[i].za_lua_type));
1174 if (pargs[i + 1].za_name != NULL || kwargs[0].za_name != NULL)
1175 msglen += 5; /* < + ( + )> + , */
1177 msglen += 4; /* < + ( + )> */
1179 for (i = 0; kwargs[i].za_name != NULL; i++) {
1180 msglen += strlen(kwargs[i].za_name);
1181 msglen += strlen(lua_typename(state, kwargs[i].za_lua_type));
1182 if (kwargs[i + 1].za_name != NULL)
1183 msglen += 4; /* =( + ) + , */
1185 msglen += 3; /* =( + ) */
1189 (void) luaL_error(state, errmsg);
1191 VERIFY3U(len, >, strlcat(errmsg, ": ", len));
1192 VERIFY3U(len, >, strlcat(errmsg, fname, len));
1193 VERIFY3U(len, >, strlcat(errmsg, "{", len));
1194 for (i = 0; pargs[i].za_name != NULL; i++) {
1195 VERIFY3U(len, >, strlcat(errmsg, "<", len));
1196 VERIFY3U(len, >, strlcat(errmsg, pargs[i].za_name, len));
1197 VERIFY3U(len, >, strlcat(errmsg, "(", len));
1198 VERIFY3U(len, >, strlcat(errmsg,
1199 lua_typename(state, pargs[i].za_lua_type), len));
1200 VERIFY3U(len, >, strlcat(errmsg, ")>", len));
1201 if (pargs[i + 1].za_name != NULL || kwargs[0].za_name != NULL) {
1202 VERIFY3U(len, >, strlcat(errmsg, ", ", len));
1205 for (i = 0; kwargs[i].za_name != NULL; i++) {
1206 VERIFY3U(len, >, strlcat(errmsg, kwargs[i].za_name, len));
1207 VERIFY3U(len, >, strlcat(errmsg, "=(", len));
1208 VERIFY3U(len, >, strlcat(errmsg,
1209 lua_typename(state, kwargs[i].za_lua_type), len));
1210 VERIFY3U(len, >, strlcat(errmsg, ")", len));
1211 if (kwargs[i + 1].za_name != NULL) {
1212 VERIFY3U(len, >, strlcat(errmsg, ", ", len));
1215 VERIFY3U(len, >, strlcat(errmsg, "}", len));
1217 (void) luaL_error(state, errmsg);
1218 panic("unreachable code");
1222 zcp_parse_table_args(lua_State *state, const char *fname,
1223 const zcp_arg_t *pargs, const zcp_arg_t *kwargs)
1228 for (i = 0; pargs[i].za_name != NULL; i++) {
1230 * Check the table for this positional argument, leaving it
1231 * on the top of the stack once we finish validating it.
1233 lua_pushinteger(state, i + 1);
1234 lua_gettable(state, 1);
1236 type = lua_type(state, -1);
1237 if (type == LUA_TNIL) {
1238 zcp_args_error(state, fname, pargs, kwargs,
1239 "too few arguments");
1240 panic("unreachable code");
1241 } else if (type != pargs[i].za_lua_type) {
1242 zcp_args_error(state, fname, pargs, kwargs,
1243 "arg %d wrong type (is '%s', expected '%s')",
1244 i + 1, lua_typename(state, type),
1245 lua_typename(state, pargs[i].za_lua_type));
1246 panic("unreachable code");
1250 * Remove the positional argument from the table.
1252 lua_pushinteger(state, i + 1);
1254 lua_settable(state, 1);
1257 for (i = 0; kwargs[i].za_name != NULL; i++) {
1259 * Check the table for this keyword argument, which may be
1260 * nil if it was omitted. Leave the value on the top of
1261 * the stack after validating it.
1263 lua_getfield(state, 1, kwargs[i].za_name);
1265 type = lua_type(state, -1);
1266 if (type != LUA_TNIL && type != kwargs[i].za_lua_type) {
1267 zcp_args_error(state, fname, pargs, kwargs,
1268 "kwarg '%s' wrong type (is '%s', expected '%s')",
1269 kwargs[i].za_name, lua_typename(state, type),
1270 lua_typename(state, kwargs[i].za_lua_type));
1271 panic("unreachable code");
1275 * Remove the keyword argument from the table.
1278 lua_setfield(state, 1, kwargs[i].za_name);
1282 * Any entries remaining in the table are invalid inputs, print
1283 * an error message based on what the entry is.
1286 if (lua_next(state, 1)) {
1287 if (lua_isnumber(state, -2) && lua_tointeger(state, -2) > 0) {
1288 zcp_args_error(state, fname, pargs, kwargs,
1289 "too many positional arguments");
1290 } else if (lua_isstring(state, -2)) {
1291 zcp_args_error(state, fname, pargs, kwargs,
1292 "invalid kwarg '%s'", lua_tostring(state, -2));
1294 zcp_args_error(state, fname, pargs, kwargs,
1295 "kwarg keys must be strings");
1297 panic("unreachable code");
1300 lua_remove(state, 1);
1304 zcp_parse_pos_args(lua_State *state, const char *fname, const zcp_arg_t *pargs,
1305 const zcp_arg_t *kwargs)
1310 for (i = 0; pargs[i].za_name != NULL; i++) {
1311 type = lua_type(state, i + 1);
1312 if (type == LUA_TNONE) {
1313 zcp_args_error(state, fname, pargs, kwargs,
1314 "too few arguments");
1315 panic("unreachable code");
1316 } else if (type != pargs[i].za_lua_type) {
1317 zcp_args_error(state, fname, pargs, kwargs,
1318 "arg %d wrong type (is '%s', expected '%s')",
1319 i + 1, lua_typename(state, type),
1320 lua_typename(state, pargs[i].za_lua_type));
1321 panic("unreachable code");
1324 if (lua_gettop(state) != i) {
1325 zcp_args_error(state, fname, pargs, kwargs,
1326 "too many positional arguments");
1327 panic("unreachable code");
1330 for (i = 0; kwargs[i].za_name != NULL; i++) {
1336 * Checks the current Lua stack against an expected set of positional and
1337 * keyword arguments. If the stack does not match the expected arguments
1338 * aborts the current channel program with a useful error message, otherwise
1339 * it re-arranges the stack so that it contains the positional arguments
1340 * followed by the keyword argument values in declaration order. Any missing
1341 * keyword argument will be represented by a nil value on the stack.
1343 * If the stack contains exactly one argument of type LUA_TTABLE the curly
1344 * braces calling convention is assumed, otherwise the stack is parsed for
1345 * positional arguments only.
1347 * This function should be used by every function callback. It should be called
1348 * before the callback manipulates the Lua stack as it assumes the stack
1349 * represents the function arguments.
1352 zcp_parse_args(lua_State *state, const char *fname, const zcp_arg_t *pargs,
1353 const zcp_arg_t *kwargs)
1355 if (lua_gettop(state) == 1 && lua_istable(state, 1)) {
1356 zcp_parse_table_args(state, fname, pargs, kwargs);
1358 zcp_parse_pos_args(state, fname, pargs, kwargs);