diff --git a/cmd/zed/agents/zfs_mod.c b/cmd/zed/agents/zfs_mod.c index 8dd75e0bb4da..921edcc63341 100644 --- a/cmd/zed/agents/zfs_mod.c +++ b/cmd/zed/agents/zfs_mod.c @@ -141,6 +141,17 @@ zfs_unavail_pool(zpool_handle_t *zhp, void *data) return (0); } +/* + * Write an array of strings to the zed log + */ +static void lines_to_zed_log_msg(char **lines, int lines_cnt) +{ + int i; + for (i = 0; i < lines_cnt; i++) { + zed_log_msg(LOG_INFO, "%s", lines[i]); + } +} + /* * Two stage replace on Linux * since we get disk notifications @@ -195,6 +206,8 @@ zfs_process_add(zpool_handle_t *zhp, nvlist_t *vdev, boolean_t labeled) boolean_t is_mpath_wholedisk = B_FALSE; uint_t c; vdev_stat_t *vs; + char **lines = NULL; + int lines_cnt = 0; if (nvlist_lookup_string(vdev, ZPOOL_CONFIG_PATH, &path) != 0) return; @@ -377,6 +390,22 @@ zfs_process_add(zpool_handle_t *zhp, nvlist_t *vdev, boolean_t labeled) if (is_mpath_wholedisk) { /* Don't label device mapper or multipath disks. */ + zed_log_msg(LOG_INFO, + " it's a multipath wholedisk, don't label"); + if (zpool_prepare_disk(zhp, vdev, "autoreplace", &lines, + &lines_cnt) != 0) { + zed_log_msg(LOG_INFO, + " zpool_prepare_disk: could not " + "prepare '%s' (%s)", fullpath, + libzfs_error_description(g_zfshdl)); + if (lines_cnt > 0) { + zed_log_msg(LOG_INFO, + " zfs_prepare_disk output:"); + lines_to_zed_log_msg(lines, lines_cnt); + } + libzfs_free_str_array(lines, lines_cnt); + return; + } } else if (!labeled) { /* * we're auto-replacing a raw disk, so label it first @@ -399,10 +428,18 @@ zfs_process_add(zpool_handle_t *zhp, nvlist_t *vdev, boolean_t labeled) * If this is a request to label a whole disk, then attempt to * write out the label. */ - if (zpool_label_disk(g_zfshdl, zhp, leafname) != 0) { - zed_log_msg(LOG_INFO, " zpool_label_disk: could not " + if (zpool_prepare_and_label_disk(g_zfshdl, zhp, leafname, + vdev, "autoreplace", &lines, &lines_cnt) != 0) { + zed_log_msg(LOG_INFO, + " zpool_prepare_and_label_disk: could not " "label '%s' (%s)", leafname, libzfs_error_description(g_zfshdl)); + if (lines_cnt > 0) { + zed_log_msg(LOG_INFO, + " zfs_prepare_disk output:"); + lines_to_zed_log_msg(lines, lines_cnt); + } + libzfs_free_str_array(lines, lines_cnt); (void) zpool_vdev_online(zhp, fullpath, ZFS_ONLINE_FORCEFAULT, &newstate); @@ -457,6 +494,8 @@ zfs_process_add(zpool_handle_t *zhp, nvlist_t *vdev, boolean_t labeled) DEV_BYID_PATH, new_devid); } + libzfs_free_str_array(lines, lines_cnt); + /* * Construct the root vdev to pass to zpool_vdev_attach(). While adding * the entire vdev structure is harmless, we construct a reduced set of diff --git a/cmd/zpool/zpool_iter.c b/cmd/zpool/zpool_iter.c index abfa2b7f6b90..486b8a1ac2b5 100644 --- a/cmd/zpool/zpool_iter.c +++ b/cmd/zpool/zpool_iter.c @@ -439,39 +439,23 @@ static void vdev_run_cmd(vdev_cmd_data_t *data, char *cmd) { int rc; - char *argv[2] = {cmd, 0}; - char *env[5] = {"PATH=/bin:/sbin:/usr/bin:/usr/sbin", NULL, NULL, NULL, - NULL}; + char *argv[2] = {cmd}; + char **env; char **lines = NULL; int lines_cnt = 0; int i; - /* Setup our custom environment variables */ - rc = asprintf(&env[1], "VDEV_PATH=%s", - data->path ? data->path : ""); - if (rc == -1) { - env[1] = NULL; + env = zpool_vdev_script_alloc_env(data->pool, data->path, data->upath, + data->vdev_enc_sysfs_path, NULL, NULL); + if (env == NULL) goto out; - } - - rc = asprintf(&env[2], "VDEV_UPATH=%s", - data->upath ? data->upath : ""); - if (rc == -1) { - env[2] = NULL; - goto out; - } - - rc = asprintf(&env[3], "VDEV_ENC_SYSFS_PATH=%s", - data->vdev_enc_sysfs_path ? - data->vdev_enc_sysfs_path : ""); - if (rc == -1) { - env[3] = NULL; - goto out; - } /* Run the command */ rc = libzfs_run_process_get_stdout_nopath(cmd, argv, env, &lines, &lines_cnt); + + zpool_vdev_script_free_env(env); + if (rc != 0) goto out; @@ -483,10 +467,6 @@ vdev_run_cmd(vdev_cmd_data_t *data, char *cmd) out: if (lines != NULL) libzfs_free_str_array(lines, lines_cnt); - - /* Start with i = 1 since env[0] was statically allocated */ - for (i = 1; i < ARRAY_SIZE(env); i++) - free(env[i]); } /* diff --git a/cmd/zpool/zpool_util.h b/cmd/zpool/zpool_util.h index da75866f5145..9e9c9c525331 100644 --- a/cmd/zpool/zpool_util.h +++ b/cmd/zpool/zpool_util.h @@ -124,6 +124,10 @@ vdev_cmd_data_list_t *all_pools_for_each_vdev_run(int argc, char **argv, void free_vdev_cmd_data_list(vdev_cmd_data_list_t *vcdl); +void free_vdev_cmd_data(vdev_cmd_data_t *data); + +int vdev_run_cmd_simple(char *path, char *cmd); + int check_device(const char *path, boolean_t force, boolean_t isspare, boolean_t iswholedisk); boolean_t check_sector_size_database(char *path, int *sector_size); diff --git a/cmd/zpool/zpool_vdev.c b/cmd/zpool/zpool_vdev.c index 3d83da641ecb..84b38cf9b1bc 100644 --- a/cmd/zpool/zpool_vdev.c +++ b/cmd/zpool/zpool_vdev.c @@ -921,6 +921,15 @@ zero_label(char *path) return (0); } +static void +lines_to_stderr(char *lines[], int lines_cnt) +{ + int i; + for (i = 0; i < lines_cnt; i++) { + fprintf(stderr, "%s\n", lines[i]); + } +} + /* * Go through and find any whole disks in the vdev specification, labelling them * as appropriate. When constructing the vdev spec, we were unable to open this @@ -932,7 +941,7 @@ zero_label(char *path) * need to get the devid after we label the disk. */ static int -make_disks(zpool_handle_t *zhp, nvlist_t *nv) +make_disks(zpool_handle_t *zhp, nvlist_t *nv, boolean_t replacing) { nvlist_t **child; uint_t c, children; @@ -1017,6 +1026,8 @@ make_disks(zpool_handle_t *zhp, nvlist_t *nv) */ if (!is_exclusive && !is_spare(NULL, udevpath)) { char *devnode = strrchr(devpath, '/') + 1; + char **lines = NULL; + int lines_cnt = 0; ret = strncmp(udevpath, UDISK_ROOT, strlen(UDISK_ROOT)); if (ret == 0) { @@ -1028,9 +1039,27 @@ make_disks(zpool_handle_t *zhp, nvlist_t *nv) /* * When labeling a pool the raw device node name * is provided as it appears under /dev/. + * + * Note that 'zhp' will be NULL when we're creating a + * pool. */ - if (zpool_label_disk(g_zfs, zhp, devnode) == -1) + if (zpool_prepare_and_label_disk(g_zfs, zhp, devnode, + nv, zhp == NULL ? "create" : + replacing ? "replace" : "add", &lines, + &lines_cnt) != 0) { + (void) fprintf(stderr, + gettext( + "Error preparing/labeling disk.\n")); + if (lines_cnt > 0) { + (void) fprintf(stderr, + gettext("zfs_prepare_disk output:\n")); + lines_to_stderr(lines, lines_cnt); + } + + libzfs_free_str_array(lines, lines_cnt); return (-1); + } + libzfs_free_str_array(lines, lines_cnt); /* * Wait for udev to signal the device is available @@ -1067,19 +1096,19 @@ make_disks(zpool_handle_t *zhp, nvlist_t *nv) } for (c = 0; c < children; c++) - if ((ret = make_disks(zhp, child[c])) != 0) + if ((ret = make_disks(zhp, child[c], replacing)) != 0) return (ret); if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES, &child, &children) == 0) for (c = 0; c < children; c++) - if ((ret = make_disks(zhp, child[c])) != 0) + if ((ret = make_disks(zhp, child[c], replacing)) != 0) return (ret); if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE, &child, &children) == 0) for (c = 0; c < children; c++) - if ((ret = make_disks(zhp, child[c])) != 0) + if ((ret = make_disks(zhp, child[c], replacing)) != 0) return (ret); return (0); @@ -1740,7 +1769,7 @@ split_mirror_vdev(zpool_handle_t *zhp, char *newname, nvlist_t *props, return (NULL); } - if (!flags.dryrun && make_disks(zhp, newroot) != 0) { + if (!flags.dryrun && make_disks(zhp, newroot, B_FALSE) != 0) { nvlist_free(newroot); return (NULL); } @@ -1861,7 +1890,7 @@ make_root_vdev(zpool_handle_t *zhp, nvlist_t *props, int force, int check_rep, /* * Run through the vdev specification and label any whole disks found. */ - if (!dryrun && make_disks(zhp, newroot) != 0) { + if (!dryrun && make_disks(zhp, newroot, replacing) != 0) { nvlist_free(newroot); return (NULL); } diff --git a/config/Rules.am b/config/Rules.am index 3b24e3630102..8c484d18c2a0 100644 --- a/config/Rules.am +++ b/config/Rules.am @@ -41,6 +41,7 @@ AM_CPPFLAGS += -D_REENTRANT AM_CPPFLAGS += -D_FILE_OFFSET_BITS=64 AM_CPPFLAGS += -D_LARGEFILE64_SOURCE AM_CPPFLAGS += -DLIBEXECDIR=\"$(libexecdir)\" +AM_CPPFLAGS += -DZFSEXECDIR=\"$(zfsexecdir)\" AM_CPPFLAGS += -DRUNSTATEDIR=\"$(runstatedir)\" AM_CPPFLAGS += -DSBINDIR=\"$(sbindir)\" AM_CPPFLAGS += -DSYSCONFDIR=\"$(sysconfdir)\" diff --git a/include/libzfs.h b/include/libzfs.h index 214a188f9474..c7ebc52fe9fb 100644 --- a/include/libzfs.h +++ b/include/libzfs.h @@ -318,6 +318,15 @@ extern nvlist_t *zpool_find_vdev(zpool_handle_t *, const char *, boolean_t *, extern nvlist_t *zpool_find_vdev_by_physpath(zpool_handle_t *, const char *, boolean_t *, boolean_t *, boolean_t *); extern int zpool_label_disk(libzfs_handle_t *, zpool_handle_t *, const char *); +extern int zpool_prepare_disk(zpool_handle_t *zhp, nvlist_t *vdev_nv, + const char *prepare_str, char **lines[], int *lines_cnt); +extern int zpool_prepare_and_label_disk(libzfs_handle_t *hdl, + zpool_handle_t *, const char *, nvlist_t *vdev_nv, const char *prepare_str, + char **lines[], int *lines_cnt); +extern char ** zpool_vdev_script_alloc_env(const char *pool_name, + const char *vdev_path, const char *vdev_upath, + const char *vdev_enc_sysfs_path, const char *opt_key, const char *opt_val); +extern void zpool_vdev_script_free_env(char **env); extern uint64_t zpool_vdev_path_to_guid(zpool_handle_t *zhp, const char *path); const char *zpool_get_state_str(zpool_handle_t *); diff --git a/lib/libzfs/libzfs.abi b/lib/libzfs/libzfs.abi index 58c2d7635f11..7e8e5ae02220 100644 --- a/lib/libzfs/libzfs.abi +++ b/lib/libzfs/libzfs.abi @@ -333,6 +333,8 @@ + + @@ -380,6 +382,8 @@ + + diff --git a/lib/libzfs/libzfs_util.c b/lib/libzfs/libzfs_util.c index 7c4d310782eb..256ae9d1b778 100644 --- a/lib/libzfs/libzfs_util.c +++ b/lib/libzfs/libzfs_util.c @@ -2121,3 +2121,196 @@ printf_color(const char *color, char *format, ...) return (rc); } + +/* PATH + 5 env vars + a NULL entry = 7 */ +#define ZPOOL_VDEV_SCRIPT_ENV_COUNT 7 + +/* + * There's a few places where ZFS will call external scripts (like the script + * in zpool.d/ and `zfs_prepare_disk`). These scripts are called with a + * reduced $PATH, and some vdev specific environment vars set. This function + * will allocate an populate the environment variable array that is passed to + * these scripts. The user must free the arrays with zpool_vdev_free_env() when + * they are done. + * + * The following env vars will be set (but value could be blank): + * + * POOL_NAME + * VDEV_PATH + * VDEV_UPATH + * VDEV_ENC_SYSFS_PATH + * + * In addition, you can set an optional environment variable named 'opt_key' + * to 'opt_val' if you want. + * + * Returns allocated env[] array on success, NULL otherwise. + */ +char ** +zpool_vdev_script_alloc_env(const char *pool_name, + const char *vdev_path, const char *vdev_upath, + const char *vdev_enc_sysfs_path, const char *opt_key, const char *opt_val) +{ + char **env = NULL; + int rc; + + env = calloc(ZPOOL_VDEV_SCRIPT_ENV_COUNT, sizeof (*env)); + if (!env) + return (NULL); + + env[0] = strdup("PATH=/bin:/sbin:/usr/bin:/usr/sbin"); + if (!env[0]) + goto error; + + /* Setup our custom environment variables */ + rc = asprintf(&env[1], "POOL_NAME=%s", pool_name ? pool_name : ""); + if (rc == -1) { + env[1] = NULL; + goto error; + } + + rc = asprintf(&env[2], "VDEV_PATH=%s", vdev_path ? vdev_path : ""); + if (rc == -1) { + env[2] = NULL; + goto error; + } + + rc = asprintf(&env[3], "VDEV_UPATH=%s", vdev_upath ? vdev_upath : ""); + if (rc == -1) { + env[3] = NULL; + goto error; + } + + rc = asprintf(&env[4], "VDEV_ENC_SYSFS_PATH=%s", + vdev_enc_sysfs_path ? vdev_enc_sysfs_path : ""); + if (rc == -1) { + env[4] = NULL; + goto error; + } + + if (opt_key != NULL) { + rc = asprintf(&env[5], "%s=%s", opt_key, + opt_val ? opt_val : ""); + if (rc == -1) { + env[5] = NULL; + goto error; + } + } + + return (env); + +error: + for (int i = 0; i < ZPOOL_VDEV_SCRIPT_ENV_COUNT; i++) + free(env[i]); + + free(env); + + return (NULL); +} + +/* + * Free the env[] array that was allocated by zpool_vdev_script_alloc_env(). + */ +void +zpool_vdev_script_free_env(char **env) +{ + for (int i = 0; i < ZPOOL_VDEV_SCRIPT_ENV_COUNT; i++) + free(env[i]); + + free(env); +} + +/* + * Prepare a disk by (optionally) running a program before labeling the disk. + * This can be useful for installing disk firmware or doing some pre-flight + * checks on the disk before it becomes part of the pool. The program run is + * located at ZFSEXECDIR/zfs_prepare_disk + * (E.x: /usr/local/libexec/zfs/zfs_prepare_disk). + * + * Return 0 on success, non-zero on failure. + */ +int +zpool_prepare_disk(zpool_handle_t *zhp, nvlist_t *vdev_nv, + const char *prepare_str, char **lines[], int *lines_cnt) +{ + const char *script_path = ZFSEXECDIR "/zfs_prepare_disk"; + const char *pool_name; + int rc = 0; + + /* Path to script and a NULL entry */ + char *argv[2] = {(char *)script_path}; + char **env = NULL; + char *path = NULL, *enc_sysfs_path = NULL; + char *upath; + *lines_cnt = 0; + + if (access(script_path, X_OK) != 0) { + /* No script, nothing to do */ + return (0); + } + + (void) nvlist_lookup_string(vdev_nv, ZPOOL_CONFIG_PATH, &path); + (void) nvlist_lookup_string(vdev_nv, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH, + &enc_sysfs_path); + + upath = zfs_get_underlying_path(path); + pool_name = zhp ? zpool_get_name(zhp) : NULL; + + env = zpool_vdev_script_alloc_env(pool_name, path, upath, + enc_sysfs_path, "VDEV_PREPARE", prepare_str); + + free(upath); + + if (env == NULL) { + return (ENOMEM); + } + + rc = libzfs_run_process_get_stdout(script_path, argv, env, lines, + lines_cnt); + + zpool_vdev_script_free_env(env); + + return (rc); +} + +/* + * Optionally run a script and then label a disk. The script can be used to + * prepare a disk for inclusion into the pool. For example, it might update + * the disk's firmware or check its health. + * + * The 'name' provided is the short name, stripped of any leading + * /dev path, and is passed to zpool_label_disk. vdev_nv is the nvlist for + * the vdev. prepare_str is a string that gets passed as the VDEV_PREPARE + * env variable to the script. + * + * The following env vars are passed to the script: + * + * POOL_NAME: The pool name (blank during zpool create) + * VDEV_PREPARE: Reason why the disk is being prepared for inclusion: + * "create", "add", "replace", or "autoreplace" + * VDEV_PATH: Path to the disk + * VDEV_UPATH: One of the 'underlying paths' to the disk. This is + * useful for DM devices. + * VDEV_ENC_SYSFS_PATH: Path to the disk's enclosure sysfs path, if available. + * + * Note, some of these values can be blank. + * + * Return 0 on success, non-zero otherwise. + */ +int +zpool_prepare_and_label_disk(libzfs_handle_t *hdl, zpool_handle_t *zhp, + const char *name, nvlist_t *vdev_nv, const char *prepare_str, + char **lines[], int *lines_cnt) +{ + int rc; + char vdev_path[MAXPATHLEN]; + (void) snprintf(vdev_path, sizeof (vdev_path), "%s/%s", DISK_ROOT, + name); + + /* zhp will be NULL when creating a pool */ + rc = zpool_prepare_disk(zhp, vdev_nv, prepare_str, lines, lines_cnt); + if (rc != 0) + return (rc); + + rc = zpool_label_disk(hdl, zhp, name); + return (rc); +} diff --git a/man/Makefile.am b/man/Makefile.am index 64650c2b988a..2608461625d4 100644 --- a/man/Makefile.am +++ b/man/Makefile.am @@ -104,7 +104,8 @@ dist_man_MANS = \ nodist_man_MANS = \ man8/zed.8 \ - man8/zfs-mount-generator.8 + man8/zfs-mount-generator.8 \ + man8/zfs_prepare_disk.8 SUBSTFILES += $(nodist_man_MANS) diff --git a/man/man8/.gitignore b/man/man8/.gitignore index f2fc702147e9..a468f9cbf9d3 100644 --- a/man/man8/.gitignore +++ b/man/man8/.gitignore @@ -1,2 +1,3 @@ /zed.8 /zfs-mount-generator.8 +/zfs_prepare_disk.8 diff --git a/man/man8/zfs_prepare_disk.8.in b/man/man8/zfs_prepare_disk.8.in new file mode 100644 index 000000000000..2a741531e415 --- /dev/null +++ b/man/man8/zfs_prepare_disk.8.in @@ -0,0 +1,70 @@ +.\" +.\" Developed at Lawrence Livermore National Laboratory (LLNL-CODE-403049). +.\" Copyright (C) 2023 Lawrence Livermore National Security, LLC. +.\" Refer to the OpenZFS git commit log for authoritative copyright attribution. +.\" +.\" The contents of this file are subject to the terms of the +.\" Common Development and Distribution License Version 1.0 (CDDL-1.0). +.\" You can obtain a copy of the license from the top-level file +.\" "OPENSOLARIS.LICENSE" or at . +.\" You may not use this file except in compliance with the license. +.\" +.\" Developed at Lawrence Livermore National Laboratory (LLNL-CODE-403049) +.\" +.Dd August 30, 2023 +.Dt ZFS_PREPARE_DISK 8 +.Os +. +.Sh NAME +.Nm zfs_prepare_disk +.Nd special script that gets run before bringing a disk into a pool +.Sh DESCRIPTION +.Nm +is an optional script that gets called by libzfs before bringing a disk into a +pool. +It can be modified by the user to run whatever commands are necessary to prepare +a disk for inclusion into the pool. +For example, users can add lines to +.Nm zfs_prepare_disk +to do things like update the drive's firmware or check the drive's health. +.Nm zfs_prepare_disk +is optional and can be removed if not needed. +libzfs will look for the script at @zfsexecdir@/zfs_prepare_disk. +. +.Ss Properties +.Nm zfs_prepare_disk +will be passed the following environment variables: +.sp +.Bl -tag -compact -width "VDEV_ENC_SYSFS_PATH" +. +.It Nm POOL_NAME +.No Name of the pool +.It Nm VDEV_PATH +.No Path to the disk (like /dev/sda) +.It Nm VDEV_PREPARE +.No Reason why the disk is being prepared for inclusion +('create', 'add', 'replace', or 'autoreplace'). +This can be useful if you only want the script to be run under certain actions. +.It Nm VDEV_UPATH +.No Path to one of the underlying devices for the +disk. +For multipath this would return one of the /dev/sd* paths to the disk. +If the device is not a device mapper device, then +.Nm VDEV_UPATH +just returns the same value as +.Nm VDEV_PATH +.It Nm VDEV_ENC_SYSFS_PATH +.No Path to the disk's enclosure sysfs path, if available +.El +.Pp +Note that some of these variables may have a blank value. +.Nm POOL_NAME +is blank at pool creation time, for example. +.Sh ENVIRONMENT +.Nm zfs_prepare_disk +runs with a limited $PATH. +.Sh EXIT STATUS +.Nm zfs_prepare_disk +should return 0 on success, non-zero otherwise. +If non-zero is returned, the disk will not be included in the pool. +. diff --git a/scripts/Makefile.am b/scripts/Makefile.am index 047ae7eaca6d..f3c133a29ae8 100644 --- a/scripts/Makefile.am +++ b/scripts/Makefile.am @@ -9,6 +9,9 @@ dist_pkgdata_SCRIPTS = \ zloop.sh \ zfs-helpers.sh +dist_zfsexec_SCRIPTS = \ + zfs_prepare_disk + EXTRA_SCRIPTS = \ commitcheck.sh \ common.sh.in \ diff --git a/scripts/zfs_prepare_disk b/scripts/zfs_prepare_disk new file mode 100755 index 000000000000..02aa9f8a7728 --- /dev/null +++ b/scripts/zfs_prepare_disk @@ -0,0 +1,17 @@ +#!/bin/sh +# +# This is an optional helper script that is automatically called by libzfs +# before a disk is about to be added into the pool. It can be modified by +# the user to run whatever commands are necessary to prepare a disk for +# inclusion into the pool. For example, users can add lines to this +# script to do things like update the drive's firmware or check the drive's +# health. The script is optional and can be removed if it is not needed. +# +# See the zfs_prepare_disk(8) man page for details. +# +# Example: +# +# echo "Prepare disk $VDEV_PATH ($VDEV_UPATH) for $VDEV_PREPARE in $POOL_NAME" +# + +exit 0