From 097150e1b787fe3c6ac4fa882f8b040e6d37288a Mon Sep 17 00:00:00 2001 From: Brian Behlendorf Date: Fri, 30 Jul 2010 13:48:28 -0700 Subject: [PATCH] Minor device creation/removal (User) With the update to onnv_141 how minor devices were created and removed for ZVOL was substantially changed. The updated system is much more tightly integrated with Solaris's /dev/ filesystem. This is great for Solaris but bad for Linux. This patch keeps as much of those changes as possible which as useful for Linux. But it also reverts back to use the old system of explicit minor node creation via a decicated ioctl(). This worked well in zfs-0.4.9 and it should continue to work well. --- lib/libzfs/include/libzfs_impl.h | 2 + lib/libzfs/libzfs_dataset.c | 321 +++++++++++++++++++++++++++++-- lib/libzfs/libzfs_pool.c | 2 +- lib/libzfs/libzfs_sendrecv.c | 12 +- 4 files changed, 318 insertions(+), 19 deletions(-) diff --git a/lib/libzfs/include/libzfs_impl.h b/lib/libzfs/include/libzfs_impl.h index 89c48c1c03..98e81cacf3 100644 --- a/lib/libzfs/include/libzfs_impl.h +++ b/lib/libzfs/include/libzfs_impl.h @@ -186,6 +186,8 @@ zfs_handle_t *make_dataset_handle(libzfs_handle_t *, const char *); int zpool_open_silent(libzfs_handle_t *, const char *, zpool_handle_t **); +int zvol_create_link(libzfs_handle_t *, const char *); +int zvol_remove_link(libzfs_handle_t *, const char *); boolean_t zpool_name_valid(libzfs_handle_t *, boolean_t, const char *); void namespace_clear(libzfs_handle_t *); diff --git a/lib/libzfs/libzfs_dataset.c b/lib/libzfs/libzfs_dataset.c index fdd9ad4939..064645a2a1 100644 --- a/lib/libzfs/libzfs_dataset.c +++ b/lib/libzfs/libzfs_dataset.c @@ -56,6 +56,7 @@ #include "libzfs_impl.h" #include "zfs_deleg.h" +static int zvol_create_link_common(libzfs_handle_t *, const char *, int); static int userquota_propname_decode(const char *propname, boolean_t zoned, zfs_userquota_prop_t *typep, char *domain, int domainlen, uint64_t *ridp); @@ -2845,6 +2846,18 @@ zfs_create(libzfs_handle_t *hdl, const char *path, zfs_type_t type, /* create the dataset */ ret = zfs_ioctl(hdl, ZFS_IOC_CREATE, &zc); + if (ret == 0 && type == ZFS_TYPE_VOLUME) { + ret = zvol_create_link(hdl, path); + if (ret) { + (void) zfs_standard_error(hdl, errno, + dgettext(TEXT_DOMAIN, + "Volume successfully created, but device links " + "were not created")); + zcmd_free_nvlists(&zc); + return (-1); + } + } + zcmd_free_nvlists(&zc); /* check for failure */ @@ -2907,6 +2920,9 @@ zfs_destroy(zfs_handle_t *zhp, boolean_t defer) (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name)); if (ZFS_IS_VOLUME(zhp)) { + if (zvol_remove_link(zhp->zfs_hdl, zhp->zfs_name) != 0) + return (-1); + zc.zc_objset_type = DMU_OST_ZVOL; } else { zc.zc_objset_type = DMU_OST_ZFS; @@ -2949,9 +2965,17 @@ zfs_check_snap_cb(zfs_handle_t *zhp, void *arg) zfs_close(szhp); } + if (zhp->zfs_type == ZFS_TYPE_VOLUME) { + (void) zvol_remove_link(zhp->zfs_hdl, name); + /* + * NB: this is simply a best-effort. We don't want to + * return an error, because then we wouldn't visit all + * the volumes. + */ + } + dd->closezhp = B_TRUE; - if (!dd->gotone) - rv = zfs_iter_filesystems(zhp, zfs_check_snap_cb, arg); + rv = zfs_iter_filesystems(zhp, zfs_check_snap_cb, arg); if (closezhp) zfs_close(zhp); return (rv); @@ -3086,11 +3110,70 @@ zfs_clone(zfs_handle_t *zhp, const char *target, nvlist_t *props) return (zfs_standard_error(zhp->zfs_hdl, errno, errbuf)); } + } else if (ZFS_IS_VOLUME(zhp)) { + ret = zvol_create_link(zhp->zfs_hdl, target); } return (ret); } +typedef struct promote_data { + char cb_mountpoint[MAXPATHLEN]; + const char *cb_target; + const char *cb_errbuf; + uint64_t cb_pivot_txg; +} promote_data_t; + +static int +promote_snap_cb(zfs_handle_t *zhp, void *data) +{ + promote_data_t *pd = data; + zfs_handle_t *szhp; + char snapname[MAXPATHLEN]; + int rv = 0; + + /* We don't care about snapshots after the pivot point */ + if (zfs_prop_get_int(zhp, ZFS_PROP_CREATETXG) > pd->cb_pivot_txg) { + zfs_close(zhp); + return (0); + } + + /* Remove the device link if it's a zvol. */ + if (ZFS_IS_VOLUME(zhp)) + (void) zvol_remove_link(zhp->zfs_hdl, zhp->zfs_name); + + /* Check for conflicting names */ + (void) strlcpy(snapname, pd->cb_target, sizeof (snapname)); + (void) strlcat(snapname, strchr(zhp->zfs_name, '@'), sizeof (snapname)); + szhp = make_dataset_handle(zhp->zfs_hdl, snapname); + if (szhp != NULL) { + zfs_close(szhp); + zfs_error_aux(zhp->zfs_hdl, dgettext(TEXT_DOMAIN, + "snapshot name '%s' from origin \n" + "conflicts with '%s' from target"), + zhp->zfs_name, snapname); + rv = zfs_error(zhp->zfs_hdl, EZFS_EXISTS, pd->cb_errbuf); + } + zfs_close(zhp); + return (rv); +} + +static int +promote_snap_done_cb(zfs_handle_t *zhp, void *data) +{ + promote_data_t *pd = data; + + /* We don't care about snapshots after the pivot point */ + if (zfs_prop_get_int(zhp, ZFS_PROP_CREATETXG) <= pd->cb_pivot_txg) { + /* Create the device link if it's a zvol. */ + if (ZFS_IS_VOLUME(zhp)) + (void) zvol_create_link(zhp->zfs_hdl, zhp->zfs_name); + } + + zfs_close(zhp); + return (0); +} + /* * Promotes the given clone fs to be the clone parent. */ @@ -3100,7 +3183,10 @@ zfs_promote(zfs_handle_t *zhp) libzfs_handle_t *hdl = zhp->zfs_hdl; zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 }; char parent[MAXPATHLEN]; + char *cp; int ret; + zfs_handle_t *pzhp; + promote_data_t pd; char errbuf[1024]; (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, @@ -3118,7 +3204,29 @@ zfs_promote(zfs_handle_t *zhp) "not a cloned filesystem")); return (zfs_error(hdl, EZFS_BADTYPE, errbuf)); } + cp = strchr(parent, '@'); + *cp = '\0'; + /* Walk the snapshots we will be moving */ + pzhp = zfs_open(hdl, zhp->zfs_dmustats.dds_origin, ZFS_TYPE_SNAPSHOT); + if (pzhp == NULL) + return (-1); + pd.cb_pivot_txg = zfs_prop_get_int(pzhp, ZFS_PROP_CREATETXG); + zfs_close(pzhp); + pd.cb_target = zhp->zfs_name; + pd.cb_errbuf = errbuf; + pzhp = zfs_open(hdl, parent, ZFS_TYPE_DATASET); + if (pzhp == NULL) + return (-1); + (void) zfs_prop_get(pzhp, ZFS_PROP_MOUNTPOINT, pd.cb_mountpoint, + sizeof (pd.cb_mountpoint), NULL, NULL, 0, FALSE); + ret = zfs_iter_snapshots(pzhp, promote_snap_cb, &pd); + if (ret != 0) { + zfs_close(pzhp); + return (-1); + } + + /* issue the ioctl */ (void) strlcpy(zc.zc_value, zhp->zfs_dmustats.dds_origin, sizeof (zc.zc_value)); (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name)); @@ -3127,9 +3235,16 @@ zfs_promote(zfs_handle_t *zhp) if (ret != 0) { int save_errno = errno; + (void) zfs_iter_snapshots(pzhp, promote_snap_done_cb, &pd); + zfs_close(pzhp); + switch (save_errno) { case EEXIST: - /* There is a conflicting snapshot name. */ + /* + * There is a conflicting snapshot name. We + * should have caught this above, but they could + * have renamed something in the mean time. + */ zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "conflicting snapshot '%s' from parent '%s'"), zc.zc_string, parent); @@ -3138,7 +3253,44 @@ zfs_promote(zfs_handle_t *zhp) default: return (zfs_standard_error(hdl, save_errno, errbuf)); } + } else { + (void) zfs_iter_snapshots(zhp, promote_snap_done_cb, &pd); } + + zfs_close(pzhp); + return (ret); +} + +struct createdata { + const char *cd_snapname; + int cd_ifexists; +}; + +static int +zfs_create_link_cb(zfs_handle_t *zhp, void *arg) +{ + struct createdata *cd = arg; + int ret; + + if (zhp->zfs_type == ZFS_TYPE_VOLUME) { + char name[MAXPATHLEN]; + + (void) strlcpy(name, zhp->zfs_name, sizeof (name)); + (void) strlcat(name, "@", sizeof (name)); + (void) strlcat(name, cd->cd_snapname, sizeof (name)); + (void) zvol_create_link_common(zhp->zfs_hdl, name, + cd->cd_ifexists); + /* + * NB: this is simply a best-effort. We don't want to + * return an error, because then we wouldn't visit all + * the volumes. + */ + } + + ret = zfs_iter_filesystems(zhp, zfs_create_link_cb, cd); + + zfs_close(zhp); + return (ret); } @@ -3202,11 +3354,31 @@ zfs_snapshot(libzfs_handle_t *hdl, const char *path, boolean_t recursive, * if it was recursive, the one that actually failed will be in * zc.zc_name. */ - if (ret != 0) { + if (ret != 0) (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, "cannot create snapshot '%s@%s'"), zc.zc_name, zc.zc_value); - (void) zfs_standard_error(hdl, errno, errbuf); + + if (ret == 0 && recursive) { + struct createdata cd; + + cd.cd_snapname = delim + 1; + cd.cd_ifexists = B_FALSE; + (void) zfs_iter_filesystems(zhp, zfs_create_link_cb, &cd); } + if (ret == 0 && zhp->zfs_type == ZFS_TYPE_VOLUME) { + ret = zvol_create_link(zhp->zfs_hdl, path); + if (ret != 0) { + (void) zfs_standard_error(hdl, errno, + dgettext(TEXT_DOMAIN, + "Volume successfully snapshotted, but device links " + "were not created")); + zfs_close(zhp); + return (-1); + } + } + + if (ret != 0) + (void) zfs_standard_error(hdl, errno, errbuf); zfs_close(zhp); @@ -3309,6 +3481,8 @@ zfs_rollback(zfs_handle_t *zhp, zfs_handle_t *snap, boolean_t force) */ if (zhp->zfs_type == ZFS_TYPE_VOLUME) { + if (zvol_remove_link(zhp->zfs_hdl, zhp->zfs_name) != 0) + return (-1); if (zfs_which_resv_prop(zhp, &resv_prop) < 0) return (-1); old_volsize = zfs_prop_get_int(zhp, ZFS_PROP_VOLSIZE); @@ -3346,6 +3520,10 @@ zfs_rollback(zfs_handle_t *zhp, zfs_handle_t *snap, boolean_t force) */ if ((zhp->zfs_type == ZFS_TYPE_VOLUME) && (zhp = make_dataset_handle(zhp->zfs_hdl, zhp->zfs_name))) { + if ((err = zvol_create_link(zhp->zfs_hdl, zhp->zfs_name))) { + zfs_close(zhp); + return (err); + } if (restore_resv) { new_volsize = zfs_prop_get_int(zhp, ZFS_PROP_VOLSIZE); if (old_volsize != new_volsize) @@ -3494,6 +3672,7 @@ zfs_rename(zfs_handle_t *zhp, const char *target, boolean_t recursive) } if (recursive) { + struct destroydata dd; parentname = zfs_strdup(zhp->zfs_hdl, zhp->zfs_name); if (parentname == NULL) { @@ -3508,6 +3687,15 @@ zfs_rename(zfs_handle_t *zhp, const char *target, boolean_t recursive) goto error; } + dd.snapname = delim + 1; + dd.gotone = B_FALSE; + dd.closezhp = B_TRUE; + + /* We remove any zvol links prior to renaming them */ + ret = zfs_iter_filesystems(zhrp, zfs_check_snap_cb, &dd); + if (ret) { + goto error; + } } else { if ((cl = changelist_gather(zhp, ZFS_PROP_NAME, 0, 0)) == NULL) return (-1); @@ -3556,10 +3744,27 @@ zfs_rename(zfs_handle_t *zhp, const char *target, boolean_t recursive) * On failure, we still want to remount any filesystems that * were previously mounted, so we don't alter the system state. */ - if (!recursive) + if (recursive) { + struct createdata cd; + + /* only create links for datasets that had existed */ + cd.cd_snapname = delim + 1; + cd.cd_ifexists = B_TRUE; + (void) zfs_iter_filesystems(zhrp, zfs_create_link_cb, + &cd); + } else { (void) changelist_postfix(cl); + } } else { - if (!recursive) { + if (recursive) { + struct createdata cd; + + /* only create links for datasets that had existed */ + cd.cd_snapname = strchr(target, '@') + 1; + cd.cd_ifexists = B_TRUE; + ret = zfs_iter_filesystems(zhrp, zfs_create_link_cb, + &cd); + } else { changelist_rename(cl, zfs_get_name(zhp), target); ret = changelist_postfix(cl); } @@ -3578,21 +3783,105 @@ error: return (ret); } +/* + * Given a zvol dataset, issue the ioctl to create the appropriate minor node, + * and wait briefly for udev to create the /dev link. + */ +int +zvol_create_link(libzfs_handle_t *hdl, const char *dataset) +{ + return (zvol_create_link_common(hdl, dataset, B_FALSE)); +} + +static int +zvol_create_link_common(libzfs_handle_t *hdl, const char *dataset, int ifexists) +{ + zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 }; + char path[MAXPATHLEN]; + int error; + + (void) strlcpy(zc.zc_name, dataset, sizeof (zc.zc_name)); + + /* + * Issue the appropriate ioctl. + */ + if (ioctl(hdl->libzfs_fd, ZFS_IOC_CREATE_MINOR, &zc) != 0) { + switch (errno) { + case EEXIST: + /* + * Silently ignore the case where the link already + * exists. This allows 'zfs volinit' to be run multiple + * times without errors. + */ + return (0); + + case ENOENT: + /* + * Dataset does not exist in the kernel. If we + * don't care (see zfs_rename), then ignore the + * error quietly. + */ + if (ifexists) { + return (0); + } + + /* FALLTHROUGH */ + + default: + return (zfs_standard_error_fmt(hdl, errno, + dgettext(TEXT_DOMAIN, "cannot create device links " + "for '%s'"), dataset)); + } + } + + /* + * Wait up to 10 seconds for udev to create the device. + */ + (void) snprintf(path, sizeof (path), "%s/%s", ZVOL_DIR, dataset); + error = zpool_label_disk_wait(path, 10000); + if (error) + (void) printf(gettext("%s may not be immediately " + "available\n"), path); + + return (0); +} + +/* + * Remove a minor node for the given zvol and the associated /dev links. + */ +int +zvol_remove_link(libzfs_handle_t *hdl, const char *dataset) +{ + zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 }; + + (void) strlcpy(zc.zc_name, dataset, sizeof (zc.zc_name)); + + if (ioctl(hdl->libzfs_fd, ZFS_IOC_REMOVE_MINOR, &zc) != 0) { + switch (errno) { + case ENXIO: + /* + * Silently ignore the case where the link no longer + * exists, so that 'zfs volfini' can be run multiple + * times without errors. + */ + return (0); + + default: + return (zfs_standard_error_fmt(hdl, errno, + dgettext(TEXT_DOMAIN, "cannot remove device " + "links for '%s'"), dataset)); + } + } + + return (0); +} + nvlist_t * zfs_get_user_props(zfs_handle_t *zhp) { return (zhp->zfs_user_props); } -nvlist_t * -zfs_get_recvd_props(zfs_handle_t *zhp) -{ - if (zhp->zfs_recvd_props == NULL) - if (get_recvd_props_ioctl(zhp) != 0) - return (NULL); - return (zhp->zfs_recvd_props); -} - /* * This function is used by 'zfs list' to determine the exact set of columns to * display, and their maximum widths. This does two main things: diff --git a/lib/libzfs/libzfs_pool.c b/lib/libzfs/libzfs_pool.c index 04a5cbad7a..b85ac7bf25 100644 --- a/lib/libzfs/libzfs_pool.c +++ b/lib/libzfs/libzfs_pool.c @@ -3577,7 +3577,7 @@ zpool_label_disk_check(char *path) (void) close(fd); return EIDRM; } - + efi_free(vtoc); (void) close(fd); return 0; diff --git a/lib/libzfs/libzfs_sendrecv.c b/lib/libzfs/libzfs_sendrecv.c index 10aafb8c25..efec76afd2 100644 --- a/lib/libzfs/libzfs_sendrecv.c +++ b/lib/libzfs/libzfs_sendrecv.c @@ -2559,6 +2559,12 @@ zfs_receive_one(libzfs_handle_t *hdl, int infd, const char *tosnap, return (-1); } } + if (!flags.dryrun && zhp->zfs_type == ZFS_TYPE_VOLUME && + zvol_remove_link(hdl, zhp->zfs_name) != 0) { + zfs_close(zhp); + zcmd_free_nvlists(&zc); + return (-1); + } zfs_close(zhp); } else { /* @@ -2752,7 +2758,6 @@ zfs_receive_one(libzfs_handle_t *hdl, int infd, const char *tosnap, * children of the target filesystem if we did a replication * receive (indicated by stream_avl being non-NULL). */ -#ifdef HAVE_ZPL cp = strchr(zc.zc_value, '@'); if (cp && (ioctl_err == 0 || !newfs)) { zfs_handle_t *h; @@ -2763,6 +2768,10 @@ zfs_receive_one(libzfs_handle_t *hdl, int infd, const char *tosnap, if (h != NULL) { if (h->zfs_type == ZFS_TYPE_VOLUME) { *cp = '@'; + err = zvol_create_link(hdl, h->zfs_name); + if (err == 0 && ioctl_err == 0) + err = zvol_create_link(hdl, + zc.zc_value); } else if (newfs || stream_avl) { /* * Track the first/top of hierarchy fs, @@ -2775,7 +2784,6 @@ zfs_receive_one(libzfs_handle_t *hdl, int infd, const char *tosnap, } *cp = '@'; } -#endif /* HAVE_ZPL */ if (clp) { err |= changelist_postfix(clp);