Date: Sun, 21 Jan 2018 19:26:38 +0000 (UTC) From: Alexander Motin <mav@FreeBSD.org> To: src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-vendor@freebsd.org Subject: svn commit: r328220 - vendor-sys/illumos/dist/uts/common/fs/zfs vendor-sys/illumos/dist/uts/common/fs/zfs/sys vendor-sys/illumos/dist/uts/common/sys/fs vendor/illumos/dist/cmd/zfs vendor/illumos/di... Message-ID: <201801211926.w0LJQcF0099064@repo.freebsd.org>
next in thread | raw e-mail | index | archive | help
Author: mav Date: Sun Jan 21 19:26:38 2018 New Revision: 328220 URL: https://svnweb.freebsd.org/changeset/base/328220 Log: 8677 Open-Context Channel Programs illumos/illumos-gate@a3b2868063897ff0083dea538f55f9873eec981f https://www.illumos.org/issues/8677 We want to be able to run channel programs outside of synching context. This would greatly improve performance of channel program that just gather information, as we won't have to wait for synching context anymore. This feature should introduce the following: - A new command line flag in "zfs program" to specify our intention to run in open context. - A new flag/option within the channel program ioctl which selects the context. - Appropriate error handling whenever we try a channel program in open-context that contains zfs.sync* expressions. - Documentation for the new feature in the manual pages. Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: Chris Williamson <chris.williamson@delphix.com> Reviewed by: Pavel Zakharov <pavel.zakharov@delphix.com> Approved by: Robert Mustacchi <rm@joyent.com> Author: Serapheim Dimitropoulos <serapheim@delphix.com> Modified: vendor-sys/illumos/dist/uts/common/fs/zfs/dsl_destroy.c vendor-sys/illumos/dist/uts/common/fs/zfs/sys/zcp.h vendor-sys/illumos/dist/uts/common/fs/zfs/zcp.c vendor-sys/illumos/dist/uts/common/fs/zfs/zcp_synctask.c vendor-sys/illumos/dist/uts/common/fs/zfs/zfs_ioctl.c vendor-sys/illumos/dist/uts/common/sys/fs/zfs.h Changes in other areas also in this revision: Modified: vendor/illumos/dist/cmd/zfs/zfs_main.c vendor/illumos/dist/lib/libzfs/common/libzfs_dataset.c vendor/illumos/dist/lib/libzfs_core/common/libzfs_core.c vendor/illumos/dist/lib/libzfs_core/common/libzfs_core.h vendor/illumos/dist/man/man1m/zfs-program.1m vendor/illumos/dist/man/man1m/zfs.1m Modified: vendor-sys/illumos/dist/uts/common/fs/zfs/dsl_destroy.c ============================================================================== --- vendor-sys/illumos/dist/uts/common/fs/zfs/dsl_destroy.c Sun Jan 21 17:31:31 2018 (r328219) +++ vendor-sys/illumos/dist/uts/common/fs/zfs/dsl_destroy.c Sun Jan 21 19:26:38 2018 (r328220) @@ -542,6 +542,7 @@ dsl_destroy_snapshots_nvl(nvlist_t *snaps, boolean_t d nvlist_t *result = fnvlist_alloc(); int error = zcp_eval(nvpair_name(nvlist_next_nvpair(snaps, NULL)), program, + B_TRUE, 0, zfs_lua_max_memlimit, nvlist_next_nvpair(wrapper, NULL), result); Modified: vendor-sys/illumos/dist/uts/common/fs/zfs/sys/zcp.h ============================================================================== --- vendor-sys/illumos/dist/uts/common/fs/zfs/sys/zcp.h Sun Jan 21 17:31:31 2018 (r328219) +++ vendor-sys/illumos/dist/uts/common/fs/zfs/sys/zcp.h Sun Jan 21 19:26:38 2018 (r328220) @@ -38,20 +38,25 @@ extern uint64_t zfs_lua_max_memlimit; int zcp_argerror(lua_State *, int, const char *, ...); -int zcp_eval(const char *, const char *, uint64_t, uint64_t, nvpair_t *, - nvlist_t *); +int zcp_eval(const char *, const char *, boolean_t, uint64_t, uint64_t, + nvpair_t *, nvlist_t *); int zcp_load_list_lib(lua_State *); int zcp_load_synctask_lib(lua_State *, boolean_t); typedef void (zcp_cleanup_t)(void *); +typedef struct zcp_cleanup_handler { + zcp_cleanup_t *zch_cleanup_func; + void *zch_cleanup_arg; + list_node_t zch_node; +} zcp_cleanup_handler_t; typedef struct zcp_run_info { dsl_pool_t *zri_pool; /* - * An estimate of the total ammount of space consumed by all + * An estimate of the total amount of space consumed by all * synctasks we have successfully performed so far in this * channel program. Used to generate ENOSPC errors for syncfuncs. */ @@ -89,16 +94,21 @@ typedef struct zcp_run_info { boolean_t zri_timed_out; /* - * The currently registered cleanup function, which will be called - * with the stored argument if a fatal error occurs. + * Boolean indicating whether or not we are running in syncing + * context. */ - zcp_cleanup_t *zri_cleanup; - void *zri_cleanup_arg; + boolean_t zri_sync; + + /* + * List of currently registered cleanup handlers, which will be + * triggered in the event of a fatal error. + */ + list_t zri_cleanup_handlers; } zcp_run_info_t; zcp_run_info_t *zcp_run_info(lua_State *); -void zcp_register_cleanup(lua_State *, zcp_cleanup_t, void *); -void zcp_clear_cleanup(lua_State *); +zcp_cleanup_handler_t *zcp_register_cleanup(lua_State *, zcp_cleanup_t, void *); +void zcp_deregister_cleanup(lua_State *, zcp_cleanup_handler_t *); void zcp_cleanup(lua_State *); /* Modified: vendor-sys/illumos/dist/uts/common/fs/zfs/zcp.c ============================================================================== --- vendor-sys/illumos/dist/uts/common/fs/zfs/zcp.c Sun Jan 21 17:31:31 2018 (r328219) +++ vendor-sys/illumos/dist/uts/common/fs/zfs/zcp.c Sun Jan 21 19:26:38 2018 (r328220) @@ -130,13 +130,6 @@ typedef struct zcp_eval_arg { uint64_t ea_instrlimit; } zcp_eval_arg_t; -/*ARGSUSED*/ -static int -zcp_eval_check(void *arg, dmu_tx_t *tx) -{ - return (0); -} - /* * The outer-most error callback handler for use with lua_pcall(). On * error Lua will call this callback with a single argument that @@ -180,41 +173,45 @@ zcp_argerror(lua_State *state, int narg, const char *m * * If an error occurs, the cleanup function will be invoked exactly once and * then unreigstered. + * + * Returns the registered cleanup handler so the caller can deregister it + * if no error occurs. */ -void +zcp_cleanup_handler_t * zcp_register_cleanup(lua_State *state, zcp_cleanup_t cleanfunc, void *cleanarg) { zcp_run_info_t *ri = zcp_run_info(state); - /* - * A cleanup function should always be explicitly removed before - * installing a new one to avoid accidental clobbering. - */ - ASSERT3P(ri->zri_cleanup, ==, NULL); - ri->zri_cleanup = cleanfunc; - ri->zri_cleanup_arg = cleanarg; + zcp_cleanup_handler_t *zch = kmem_alloc(sizeof (*zch), KM_SLEEP); + zch->zch_cleanup_func = cleanfunc; + zch->zch_cleanup_arg = cleanarg; + list_insert_head(&ri->zri_cleanup_handlers, zch); + + return (zch); } void -zcp_clear_cleanup(lua_State *state) +zcp_deregister_cleanup(lua_State *state, zcp_cleanup_handler_t *zch) { zcp_run_info_t *ri = zcp_run_info(state); - - ri->zri_cleanup = NULL; - ri->zri_cleanup_arg = NULL; + list_remove(&ri->zri_cleanup_handlers, zch); + kmem_free(zch, sizeof (*zch)); } /* - * If it exists, execute the currently set cleanup function then unregister it. + * Execute the currently registered cleanup handlers then free them and + * destroy the handler list. */ void zcp_cleanup(lua_State *state) { zcp_run_info_t *ri = zcp_run_info(state); - if (ri->zri_cleanup != NULL) { - ri->zri_cleanup(ri->zri_cleanup_arg); - zcp_clear_cleanup(state); + for (zcp_cleanup_handler_t *zch = + list_remove_head(&ri->zri_cleanup_handlers); zch != NULL; + zch = list_remove_head(&ri->zri_cleanup_handlers)) { + zch->zch_cleanup_func(zch->zch_cleanup_arg); + kmem_free(zch, sizeof (*zch)); } } @@ -815,19 +812,12 @@ zcp_panic_cb(lua_State *state) } static void -zcp_eval_sync(void *arg, dmu_tx_t *tx) +zcp_eval_impl(dmu_tx_t *tx, boolean_t sync, zcp_eval_arg_t *evalargs) { int err; zcp_run_info_t ri; - zcp_eval_arg_t *evalargs = arg; lua_State *state = evalargs->ea_state; - /* - * Open context should have setup the stack to contain: - * 1: Error handler callback - * 2: Script to run (converted to a Lua function) - * 3: nvlist input to function (converted to Lua table or nil) - */ VERIFY3U(3, ==, lua_gettop(state)); /* @@ -840,8 +830,9 @@ zcp_eval_sync(void *arg, dmu_tx_t *tx) ri.zri_cred = evalargs->ea_cred; ri.zri_tx = tx; ri.zri_timed_out = B_FALSE; - ri.zri_cleanup = NULL; - ri.zri_cleanup_arg = NULL; + ri.zri_sync = sync; + list_create(&ri.zri_cleanup_handlers, sizeof (zcp_cleanup_handler_t), + offsetof(zcp_cleanup_handler_t, zch_node)); ri.zri_curinstrs = 0; ri.zri_maxinstrs = evalargs->ea_instrlimit; @@ -878,10 +869,10 @@ zcp_eval_sync(void *arg, dmu_tx_t *tx) /* * Remove the error handler callback from the stack. At this point, - * if there is a cleanup function registered, then it was registered - * but never run or removed, which should never occur. + * there shouldn't be any cleanup handler registered in the handler + * list (zri_cleanup_handlers), regardless of whether it ran or not. */ - ASSERT3P(ri.zri_cleanup, ==, NULL); + list_destroy(&ri.zri_cleanup_handlers); lua_remove(state, 1); switch (err) { @@ -963,9 +954,73 @@ zcp_eval_sync(void *arg, dmu_tx_t *tx) } } +static void +zcp_pool_error(zcp_eval_arg_t *evalargs, const char *poolname) +{ + evalargs->ea_result = SET_ERROR(ECHRNG); + (void) lua_pushfstring(evalargs->ea_state, "Could not open pool: %s", + poolname); + zcp_convert_return_values(evalargs->ea_state, evalargs->ea_outnvl, + ZCP_RET_ERROR, evalargs); + +} + +static void +zcp_eval_sync(void *arg, dmu_tx_t *tx) +{ + zcp_eval_arg_t *evalargs = arg; + + /* + * Open context should have setup the stack to contain: + * 1: Error handler callback + * 2: Script to run (converted to a Lua function) + * 3: nvlist input to function (converted to Lua table or nil) + */ + VERIFY3U(3, ==, lua_gettop(evalargs->ea_state)); + + zcp_eval_impl(tx, B_TRUE, evalargs); +} + +static void +zcp_eval_open(zcp_eval_arg_t *evalargs, const char *poolname) +{ + + int error; + dsl_pool_t *dp; + dmu_tx_t *tx; + + /* + * See comment from the same assertion in zcp_eval_sync(). + */ + VERIFY3U(3, ==, lua_gettop(evalargs->ea_state)); + + error = dsl_pool_hold(poolname, FTAG, &dp); + if (error != 0) { + zcp_pool_error(evalargs, poolname); + return; + } + + /* + * As we are running in open-context, we have no transaction associated + * with the channel program. At the same time, functions from the + * zfs.check submodule need to be associated with a transaction as + * they are basically dry-runs of their counterparts in the zfs.sync + * submodule. These functions should be able to run in open-context. + * Therefore we create a new transaction that we later abort once + * the channel program has been evaluated. + */ + tx = dmu_tx_create_dd(dp->dp_mos_dir); + + zcp_eval_impl(tx, B_FALSE, evalargs); + + dmu_tx_abort(tx); + + dsl_pool_rele(dp, FTAG); +} + int -zcp_eval(const char *poolname, const char *program, uint64_t instrlimit, - uint64_t memlimit, nvpair_t *nvarg, nvlist_t *outnvl) +zcp_eval(const char *poolname, const char *program, boolean_t sync, + uint64_t instrlimit, uint64_t memlimit, nvpair_t *nvarg, nvlist_t *outnvl) { int err; lua_State *state; @@ -1076,9 +1131,14 @@ zcp_eval(const char *poolname, const char *program, ui evalargs.ea_outnvl = outnvl; evalargs.ea_result = 0; - VERIFY0(dsl_sync_task(poolname, zcp_eval_check, - zcp_eval_sync, &evalargs, 0, ZFS_SPACE_CHECK_NONE)); - + if (sync) { + err = dsl_sync_task(poolname, NULL, + zcp_eval_sync, &evalargs, 0, ZFS_SPACE_CHECK_NONE); + if (err != 0) + zcp_pool_error(&evalargs, poolname); + } else { + zcp_eval_open(&evalargs, poolname); + } lua_close(state); return (evalargs.ea_result); Modified: vendor-sys/illumos/dist/uts/common/fs/zfs/zcp_synctask.c ============================================================================== --- vendor-sys/illumos/dist/uts/common/fs/zfs/zcp_synctask.c Sun Jan 21 17:31:31 2018 (r328219) +++ vendor-sys/illumos/dist/uts/common/fs/zfs/zcp_synctask.c Sun Jan 21 19:26:38 2018 (r328220) @@ -55,6 +55,10 @@ typedef struct zcp_synctask_info { * * If 'sync' is false, executes a dry run and returns the error code. * + * If we are not running in syncing context and we are not doing a dry run + * (meaning we are running a zfs.sync function in open-context) then we + * return a Lua error. + * * This function also handles common fatal error cases for channel program * library functions. If a fatal error occurs, err_dsname will be the dataset * name reported in error messages, if supplied. @@ -70,6 +74,13 @@ zcp_sync_task(lua_State *state, dsl_checkfunc_t *check if (!sync) return (err); + if (!ri->zri_sync) { + return (luaL_error(state, "running functions from the zfs.sync " + "submodule requires passing sync=TRUE to " + "lzc_channel_program() (i.e. do not specify the \"-n\" " + "command line argument)")); + } + if (err == 0) { syncfunc(arg, ri->zri_tx); } else if (err == EIO) { @@ -234,6 +245,15 @@ zcp_synctask_snapshot(lua_State *state, boolean_t sync zcp_run_info_t *ri = zcp_run_info(state); /* + * On old pools, the ZIL must not be active when a snapshot is created, + * but we can't suspend the ZIL because we're already in syncing + * context. + */ + if (spa_version(ri->zri_pool->dp_spa) < SPA_VERSION_FAST_SNAP) { + return (ENOTSUP); + } + + /* * We only allow for a single snapshot rather than a list, so the * error list output is unnecessary. */ @@ -243,33 +263,23 @@ zcp_synctask_snapshot(lua_State *state, boolean_t sync ddsa.ddsa_snaps = fnvlist_alloc(); fnvlist_add_boolean(ddsa.ddsa_snaps, dsname); - /* - * On old pools, the ZIL must not be active when a snapshot is created, - * but we can't suspend the ZIL because we're already in syncing - * context. - */ - if (spa_version(ri->zri_pool->dp_spa) < SPA_VERSION_FAST_SNAP) { - return (ENOTSUP); - } + zcp_cleanup_handler_t *zch = zcp_register_cleanup(state, + (zcp_cleanup_t *)&fnvlist_free, ddsa.ddsa_snaps); err = zcp_sync_task(state, dsl_dataset_snapshot_check, dsl_dataset_snapshot_sync, &ddsa, sync, dsname); + zcp_deregister_cleanup(state, zch); fnvlist_free(ddsa.ddsa_snaps); return (err); } -void -zcp_synctask_wrapper_cleanup(void *arg) -{ - fnvlist_free(arg); -} - static int zcp_synctask_wrapper(lua_State *state) { int err; + zcp_cleanup_handler_t *zch; int num_ret = 1; nvlist_t *err_details = fnvlist_alloc(); @@ -277,7 +287,8 @@ zcp_synctask_wrapper(lua_State *state) * Make sure err_details is properly freed, even if a fatal error is * thrown during the synctask. */ - zcp_register_cleanup(state, &zcp_synctask_wrapper_cleanup, err_details); + zch = zcp_register_cleanup(state, + (zcp_cleanup_t *)&fnvlist_free, err_details); zcp_synctask_info_t *info = lua_touserdata(state, lua_upvalueindex(1)); boolean_t sync = lua_toboolean(state, lua_upvalueindex(2)); @@ -317,7 +328,7 @@ zcp_synctask_wrapper(lua_State *state) num_ret++; } - zcp_clear_cleanup(state); + zcp_deregister_cleanup(state, zch); fnvlist_free(err_details); return (num_ret); Modified: vendor-sys/illumos/dist/uts/common/fs/zfs/zfs_ioctl.c ============================================================================== --- vendor-sys/illumos/dist/uts/common/fs/zfs/zfs_ioctl.c Sun Jan 21 17:31:31 2018 (r328219) +++ vendor-sys/illumos/dist/uts/common/fs/zfs/zfs_ioctl.c Sun Jan 21 19:26:38 2018 (r328220) @@ -3606,11 +3606,15 @@ zfs_ioc_channel_program(const char *poolname, nvlist_t { char *program; uint64_t instrlimit, memlimit; + boolean_t sync_flag; nvpair_t *nvarg = NULL; if (0 != nvlist_lookup_string(innvl, ZCP_ARG_PROGRAM, &program)) { return (EINVAL); } + if (0 != nvlist_lookup_boolean_value(innvl, ZCP_ARG_SYNC, &sync_flag)) { + sync_flag = B_TRUE; + } if (0 != nvlist_lookup_uint64(innvl, ZCP_ARG_INSTRLIMIT, &instrlimit)) { instrlimit = ZCP_DEFAULT_INSTRLIMIT; } @@ -3626,7 +3630,7 @@ zfs_ioc_channel_program(const char *poolname, nvlist_t if (memlimit == 0 || memlimit > zfs_lua_max_memlimit) return (EINVAL); - return (zcp_eval(poolname, program, instrlimit, memlimit, + return (zcp_eval(poolname, program, sync_flag, instrlimit, memlimit, nvarg, outnvl)); } Modified: vendor-sys/illumos/dist/uts/common/sys/fs/zfs.h ============================================================================== --- vendor-sys/illumos/dist/uts/common/sys/fs/zfs.h Sun Jan 21 17:31:31 2018 (r328219) +++ vendor-sys/illumos/dist/uts/common/sys/fs/zfs.h Sun Jan 21 19:26:38 2018 (r328220) @@ -963,6 +963,7 @@ typedef enum { */ #define ZCP_ARG_PROGRAM "program" #define ZCP_ARG_ARGLIST "arg" +#define ZCP_ARG_SYNC "sync" #define ZCP_ARG_INSTRLIMIT "instrlimit" #define ZCP_ARG_MEMLIMIT "memlimit"
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201801211926.w0LJQcF0099064>