Index: stable/11/sbin/geom/class/mirror/geom_mirror.c =================================================================== --- stable/11/sbin/geom/class/mirror/geom_mirror.c (revision 318751) +++ stable/11/sbin/geom/class/mirror/geom_mirror.c (revision 318752) @@ -1,487 +1,497 @@ /*- * Copyright (c) 2004-2009 Pawel Jakub Dawidek * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include uint32_t lib_version = G_LIB_VERSION; uint32_t version = G_MIRROR_VERSION; #define GMIRROR_BALANCE "load" #define GMIRROR_SLICE "4096" #define GMIRROR_PRIORITY "0" static void mirror_main(struct gctl_req *req, unsigned flags); static void mirror_activate(struct gctl_req *req); static void mirror_clear(struct gctl_req *req); static void mirror_dump(struct gctl_req *req); static void mirror_label(struct gctl_req *req); static void mirror_resize(struct gctl_req *req, unsigned flags); struct g_command class_commands[] = { { "activate", G_FLAG_VERBOSE, mirror_main, G_NULL_OPTS, "[-v] name prov ..." }, { "clear", G_FLAG_VERBOSE, mirror_main, G_NULL_OPTS, "[-v] prov ..." }, { "configure", G_FLAG_VERBOSE, NULL, { { 'a', "autosync", NULL, G_TYPE_BOOL }, { 'b', "balance", "", G_TYPE_STRING }, { 'd', "dynamic", NULL, G_TYPE_BOOL }, { 'f', "failsync", NULL, G_TYPE_BOOL }, { 'F', "nofailsync", NULL, G_TYPE_BOOL }, { 'h', "hardcode", NULL, G_TYPE_BOOL }, { 'n', "noautosync", NULL, G_TYPE_BOOL }, { 'p', "priority", "-1", G_TYPE_NUMBER }, { 's', "slice", "-1", G_TYPE_NUMBER }, G_OPT_SENTINEL }, "[-adfFhnv] [-b balance] [-s slice] name\n" "[-v] -p priority name prov" }, + { "create", G_FLAG_VERBOSE, NULL, + { + { 'b', "balance", GMIRROR_BALANCE, G_TYPE_STRING }, + { 'F', "nofailsync", NULL, G_TYPE_BOOL }, + { 'n', "noautosync", NULL, G_TYPE_BOOL }, + { 's', "slice", GMIRROR_SLICE, G_TYPE_NUMBER }, + G_OPT_SENTINEL + }, + "[-Fnv] [-b balance] [-s slice] name prov ..." + }, { "deactivate", G_FLAG_VERBOSE, NULL, G_NULL_OPTS, "[-v] name prov ..." }, { "destroy", G_FLAG_VERBOSE, NULL, { { 'f', "force", NULL, G_TYPE_BOOL }, G_OPT_SENTINEL }, "[-fv] name ..." }, { "dump", 0, mirror_main, G_NULL_OPTS, "prov ..." }, { "forget", G_FLAG_VERBOSE, NULL, G_NULL_OPTS, "name ..." }, { "label", G_FLAG_VERBOSE, mirror_main, { { 'b', "balance", GMIRROR_BALANCE, G_TYPE_STRING }, { 'F', "nofailsync", NULL, G_TYPE_BOOL }, { 'h', "hardcode", NULL, G_TYPE_BOOL }, { 'n', "noautosync", NULL, G_TYPE_BOOL }, { 's', "slice", GMIRROR_SLICE, G_TYPE_NUMBER }, G_OPT_SENTINEL }, "[-Fhnv] [-b balance] [-s slice] name prov ..." }, { "insert", G_FLAG_VERBOSE, NULL, { { 'h', "hardcode", NULL, G_TYPE_BOOL }, { 'i', "inactive", NULL, G_TYPE_BOOL }, { 'p', "priority", GMIRROR_PRIORITY, G_TYPE_NUMBER }, G_OPT_SENTINEL }, "[-hiv] [-p priority] name prov ..." }, { "rebuild", G_FLAG_VERBOSE, NULL, G_NULL_OPTS, "[-v] name prov ..." }, { "remove", G_FLAG_VERBOSE, NULL, G_NULL_OPTS, "[-v] name prov ..." }, { "resize", G_FLAG_VERBOSE, mirror_resize, { { 's', "size", "*", G_TYPE_STRING }, G_OPT_SENTINEL }, "[-s size] [-v] name" }, { "stop", G_FLAG_VERBOSE, NULL, { { 'f', "force", NULL, G_TYPE_BOOL }, G_OPT_SENTINEL }, "[-fv] name ..." }, G_CMD_SENTINEL }; static int verbose = 0; static void mirror_main(struct gctl_req *req, unsigned flags) { const char *name; if ((flags & G_FLAG_VERBOSE) != 0) verbose = 1; name = gctl_get_ascii(req, "verb"); if (name == NULL) { gctl_error(req, "No '%s' argument.", "verb"); return; } if (strcmp(name, "label") == 0) mirror_label(req); else if (strcmp(name, "clear") == 0) mirror_clear(req); else if (strcmp(name, "dump") == 0) mirror_dump(req); else if (strcmp(name, "activate") == 0) mirror_activate(req); else gctl_error(req, "Unknown command: %s.", name); } static void mirror_label(struct gctl_req *req) { struct g_mirror_metadata md; u_char sector[512]; const char *str; unsigned sectorsize; off_t mediasize; intmax_t val; int error, i, nargs, bal, hardcode; nargs = gctl_get_int(req, "nargs"); if (nargs < 2) { gctl_error(req, "Too few arguments."); return; } strlcpy(md.md_magic, G_MIRROR_MAGIC, sizeof(md.md_magic)); md.md_version = G_MIRROR_VERSION; str = gctl_get_ascii(req, "arg0"); strlcpy(md.md_name, str, sizeof(md.md_name)); md.md_mid = arc4random(); md.md_all = nargs - 1; md.md_mflags = 0; md.md_dflags = 0; md.md_genid = 0; md.md_syncid = 1; md.md_sync_offset = 0; val = gctl_get_intmax(req, "slice"); md.md_slice = val; str = gctl_get_ascii(req, "balance"); bal = balance_id(str); if (bal == -1) { gctl_error(req, "Invalid balance algorithm."); return; } md.md_balance = bal; if (gctl_get_int(req, "noautosync")) md.md_mflags |= G_MIRROR_DEVICE_FLAG_NOAUTOSYNC; if (gctl_get_int(req, "nofailsync")) md.md_mflags |= G_MIRROR_DEVICE_FLAG_NOFAILSYNC; hardcode = gctl_get_int(req, "hardcode"); /* * Calculate sectorsize by finding least common multiple from * sectorsizes of every disk and find the smallest mediasize. */ mediasize = 0; sectorsize = 0; for (i = 1; i < nargs; i++) { unsigned ssize; off_t msize; str = gctl_get_ascii(req, "arg%d", i); msize = g_get_mediasize(str); ssize = g_get_sectorsize(str); if (msize == 0 || ssize == 0) { gctl_error(req, "Can't get informations about %s: %s.", str, strerror(errno)); return; } msize -= ssize; if (mediasize == 0 || (mediasize > 0 && msize < mediasize)) mediasize = msize; if (sectorsize == 0) sectorsize = ssize; else sectorsize = g_lcm(sectorsize, ssize); } md.md_mediasize = mediasize; md.md_sectorsize = sectorsize; md.md_mediasize -= (md.md_mediasize % md.md_sectorsize); /* * Clear last sector first, to spoil all components if device exists. */ for (i = 1; i < nargs; i++) { str = gctl_get_ascii(req, "arg%d", i); error = g_metadata_clear(str, NULL); if (error != 0) { gctl_error(req, "Can't store metadata on %s: %s.", str, strerror(error)); return; } } /* * Ok, store metadata (use disk number as priority). */ for (i = 1; i < nargs; i++) { str = gctl_get_ascii(req, "arg%d", i); md.md_did = arc4random(); md.md_priority = i - 1; md.md_provsize = g_get_mediasize(str); assert(md.md_provsize != 0); if (!hardcode) bzero(md.md_provider, sizeof(md.md_provider)); else { if (strncmp(str, _PATH_DEV, sizeof(_PATH_DEV) - 1) == 0) str += sizeof(_PATH_DEV) - 1; strlcpy(md.md_provider, str, sizeof(md.md_provider)); } mirror_metadata_encode(&md, sector); error = g_metadata_store(str, sector, sizeof(sector)); if (error != 0) { fprintf(stderr, "Can't store metadata on %s: %s.\n", str, strerror(error)); gctl_error(req, "Not fully done."); continue; } if (verbose) printf("Metadata value stored on %s.\n", str); } } static void mirror_clear(struct gctl_req *req) { const char *name; int error, i, nargs; nargs = gctl_get_int(req, "nargs"); if (nargs < 1) { gctl_error(req, "Too few arguments."); return; } for (i = 0; i < nargs; i++) { name = gctl_get_ascii(req, "arg%d", i); error = g_metadata_clear(name, G_MIRROR_MAGIC); if (error != 0) { fprintf(stderr, "Can't clear metadata on %s: %s.\n", name, strerror(error)); gctl_error(req, "Not fully done."); continue; } if (verbose) printf("Metadata cleared on %s.\n", name); } } static void mirror_dump(struct gctl_req *req) { struct g_mirror_metadata md, tmpmd; const char *name; int error, i, nargs; nargs = gctl_get_int(req, "nargs"); if (nargs < 1) { gctl_error(req, "Too few arguments."); return; } for (i = 0; i < nargs; i++) { name = gctl_get_ascii(req, "arg%d", i); error = g_metadata_read(name, (u_char *)&tmpmd, sizeof(tmpmd), G_MIRROR_MAGIC); if (error != 0) { fprintf(stderr, "Can't read metadata from %s: %s.\n", name, strerror(error)); gctl_error(req, "Not fully done."); continue; } if (mirror_metadata_decode((u_char *)&tmpmd, &md) != 0) { fprintf(stderr, "MD5 hash mismatch for %s, skipping.\n", name); gctl_error(req, "Not fully done."); continue; } printf("Metadata on %s:\n", name); mirror_metadata_dump(&md); printf("\n"); } } static void mirror_activate(struct gctl_req *req) { struct g_mirror_metadata md, tmpmd; const char *name, *path; int error, i, nargs; nargs = gctl_get_int(req, "nargs"); if (nargs < 2) { gctl_error(req, "Too few arguments."); return; } name = gctl_get_ascii(req, "arg0"); for (i = 1; i < nargs; i++) { path = gctl_get_ascii(req, "arg%d", i); error = g_metadata_read(path, (u_char *)&tmpmd, sizeof(tmpmd), G_MIRROR_MAGIC); if (error != 0) { fprintf(stderr, "Cannot read metadata from %s: %s.\n", path, strerror(error)); gctl_error(req, "Not fully done."); continue; } if (mirror_metadata_decode((u_char *)&tmpmd, &md) != 0) { fprintf(stderr, "MD5 hash mismatch for provider %s, skipping.\n", path); gctl_error(req, "Not fully done."); continue; } if (strcmp(md.md_name, name) != 0) { fprintf(stderr, "Provider %s is not the mirror %s component.\n", path, name); gctl_error(req, "Not fully done."); continue; } md.md_dflags &= ~G_MIRROR_DISK_FLAG_INACTIVE; mirror_metadata_encode(&md, (u_char *)&tmpmd); error = g_metadata_store(path, (u_char *)&tmpmd, sizeof(tmpmd)); if (error != 0) { fprintf(stderr, "Cannot write metadata from %s: %s.\n", path, strerror(error)); gctl_error(req, "Not fully done."); continue; } if (verbose) printf("Provider %s activated.\n", path); } } static struct gclass * find_class(struct gmesh *mesh, const char *name) { struct gclass *classp; LIST_FOREACH(classp, &mesh->lg_class, lg_class) { if (strcmp(classp->lg_name, name) == 0) return (classp); } return (NULL); } static struct ggeom * find_geom(struct gclass *classp, const char *name) { struct ggeom *gp; LIST_FOREACH(gp, &classp->lg_geom, lg_geom) { if (strcmp(gp->lg_name, name) == 0) return (gp); } return (NULL); } static void mirror_resize(struct gctl_req *req, unsigned flags __unused) { struct gmesh mesh; struct gclass *classp; struct ggeom *gp; struct gprovider *pp; struct gconsumer *cp; off_t size; int error, nargs; const char *name; char ssize[30]; nargs = gctl_get_int(req, "nargs"); if (nargs < 1) { gctl_error(req, "Too few arguments."); return; } error = geom_gettree(&mesh); if (error) errc(EXIT_FAILURE, error, "Cannot get GEOM tree"); name = gctl_get_ascii(req, "class"); if (name == NULL) abort(); classp = find_class(&mesh, name); if (classp == NULL) errx(EXIT_FAILURE, "Class %s not found.", name); name = gctl_get_ascii(req, "arg0"); if (name == NULL) abort(); gp = find_geom(classp, name); if (gp == NULL) errx(EXIT_FAILURE, "No such geom: %s.", name); pp = LIST_FIRST(&gp->lg_provider); if (pp == NULL) errx(EXIT_FAILURE, "Provider of geom %s not found.", name); size = pp->lg_mediasize; name = gctl_get_ascii(req, "size"); if (name == NULL) errx(EXIT_FAILURE, "The size is not specified."); if (*name == '*') { #define CSZ(c) ((c)->lg_provider->lg_mediasize - \ (c)->lg_provider->lg_sectorsize) /* Find the maximum possible size */ LIST_FOREACH(cp, &gp->lg_consumer, lg_consumer) { if (CSZ(cp) > size) size = CSZ(cp); } LIST_FOREACH(cp, &gp->lg_consumer, lg_consumer) { if (CSZ(cp) < size) size = CSZ(cp); } #undef CSZ if (size == pp->lg_mediasize) errx(EXIT_FAILURE, "Cannot expand provider %s\n", pp->lg_name); } else { error = g_parse_lba(name, pp->lg_sectorsize, &size); if (error) errc(EXIT_FAILURE, error, "Invalid size param"); size *= pp->lg_sectorsize; } snprintf(ssize, sizeof(ssize), "%ju", (uintmax_t)size); gctl_change_param(req, "size", -1, ssize); geom_deletetree(&mesh); gctl_issue(req); } Index: stable/11/sbin/geom/class/mirror/gmirror.8 =================================================================== --- stable/11/sbin/geom/class/mirror/gmirror.8 (revision 318751) +++ stable/11/sbin/geom/class/mirror/gmirror.8 (revision 318752) @@ -1,387 +1,400 @@ .\" Copyright (c) 2004-2009 Pawel Jakub Dawidek .\" All rights reserved. .\" .\" Redistribution and use in source and binary forms, with or without .\" modification, are permitted provided that the following conditions .\" are met: .\" 1. Redistributions of source code must retain the above copyright .\" notice, this list of conditions and the following disclaimer. .\" 2. Redistributions in binary form must reproduce the above copyright .\" notice, this list of conditions and the following disclaimer in the .\" documentation and/or other materials provided with the distribution. .\" .\" THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND .\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE .\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE .\" ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE .\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL .\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS .\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) .\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT .\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY .\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF .\" SUCH DAMAGE. .\" .\" $FreeBSD$ .\" -.Dd December 27, 2013 +.Dd November 27, 2016 .Dt GMIRROR 8 .Os .Sh NAME .Nm gmirror .Nd "control utility for mirrored devices" .Sh SYNOPSIS .Nm .Cm label .Op Fl Fhnv .Op Fl b Ar balance .Op Fl s Ar slice .Ar name .Ar prov ... .Nm .Cm clear .Op Fl v .Ar prov ... .Nm +.Cm create +.Op Fl Fnv +.Op Fl b Ar balance +.Op Fl s Ar slice +.Ar name +.Ar prov ... +.Nm .Cm configure .Op Fl adfFhnv .Op Fl b Ar balance .Op Fl s Ar slice .Ar name .Nm .Cm configure .Op Fl v .Fl p Ar priority .Ar name .Ar prov .Nm .Cm rebuild .Op Fl v .Ar name .Ar prov ... .Nm .Cm resize .Op Fl v .Op Fl s Ar size .Ar name .Nm .Cm insert .Op Fl hiv .Op Fl p Ar priority .Ar name .Ar prov ... .Nm .Cm remove .Op Fl v .Ar name .Ar prov ... .Nm .Cm activate .Op Fl v .Ar name .Ar prov ... .Nm .Cm deactivate .Op Fl v .Ar name .Ar prov ... .Nm .Cm destroy .Op Fl fv .Ar name ... .Nm .Cm forget .Op Fl v .Ar name ... .Nm .Cm stop .Op Fl fv .Ar name ... .Nm .Cm dump .Ar prov ... .Nm .Cm list .Nm .Cm status .Nm .Cm load .Nm .Cm unload .Sh DESCRIPTION The .Nm utility is used for mirror (RAID1) configurations. After a mirror's creation, all components are detected and configured automatically. All operations like failure detection, stale component detection, rebuild of stale components, etc.\& are also done automatically. The .Nm utility uses on-disk metadata (stored in the provider's last sector) to store all needed information. Since the last sector is used for this purpose, it is possible to place a root file system on a mirror. .Pp The first argument to .Nm indicates an action to be performed: .Bl -tag -width ".Cm deactivate" .It Cm label Create a mirror. The order of components is important, because a component's priority is based on its position (starting from 0 to 255). The component with the biggest priority is used by the .Cm prefer balance algorithm and is also used as a master component when resynchronization is needed, e.g.\& after a power failure when the device was open for writing. .Pp Additional options include: .Bl -tag -width ".Fl b Ar balance" .It Fl b Ar balance Specifies balance algorithm to use, one of: .Bl -tag -width ".Cm round-robin" .It Cm load Read from the component with the lowest load. This is the default balance algorithm. .It Cm prefer Read from the component with the biggest priority. .It Cm round-robin Use round-robin algorithm when choosing component to read. .It Cm split Split read requests, which are bigger than or equal to slice size on N pieces, where N is the number of active components. .El .It Fl F Do not synchronize after a power failure or system crash. Assumes device is in consistent state. .It Fl h Hardcode providers' names in metadata. .It Fl n Turn off autosynchronization of stale components. .It Fl s Ar slice When using the .Cm split balance algorithm and an I/O READ request is bigger than or equal to this value, the I/O request will be split into N pieces, where N is the number of active components. Defaults to 4096 bytes. .El .It Cm clear Clear metadata on the given providers. +.It Cm create +Similar to +.Cm label, +but creates mirror without storing on-disk metadata in last sector. +This special "manual" operation mode assumes some external control to manage +mirror detection after reboot, device hot-plug and other external events. .It Cm configure Configure the given device. .Pp Additional options include: .Bl -tag -width ".Fl p Ar priority" .It Fl a Turn on autosynchronization of stale components. .It Fl b Ar balance Specifies balance algorithm to use. .It Fl d Do not hardcode providers' names in metadata. .It Fl f Synchronize device after a power failure or system crash. .It Fl F Do not synchronize after a power failure or system crash. Assumes device is in consistent state. .It Fl h Hardcode providers' names in metadata. .It Fl n Turn off autosynchronization of stale components. .It Fl p Ar priority Specifies priority for the given component .Ar prov . .It Fl s Ar slice Specifies slice size for .Cm split balance algorithm. .El .It Cm rebuild Rebuild the given mirror components forcibly. If autosynchronization was not turned off for the given device, this command should be unnecessary. .It Cm resize Change the size of the given mirror. .Pp Additional options include: .Bl -tag -width ".Fl s Ar size" .It Fl s Ar size New size of the mirror is expressed in logical block numbers. This option can be omitted, then it will be automatically calculated to maximum available size. .El .It Cm insert Add the given component(s) to the existing mirror. .Pp Additional options include: .Bl -tag -width ".Fl p Ar priority" .It Fl h Hardcode providers' names in metadata. .It Fl i Mark component(s) as inactive immediately after insertion. .It Fl p Ar priority Specifies priority of the given component(s). .El .It Cm remove Remove the given component(s) from the mirror and clear metadata on it. .It Cm activate Activate the given component(s), which were marked as inactive before. .It Cm deactivate Mark the given component(s) as inactive, so it will not be automatically connected to the mirror. .It Cm destroy Stop the given mirror and clear metadata on all its components. .Pp Additional options include: .Bl -tag -width ".Fl f" .It Fl f Stop the given mirror even if it is opened. .El .It Cm forget Forget about components which are not connected. This command is useful when a disk has failed and cannot be reconnected, preventing the .Cm remove command from being used to remove it. .It Cm stop Stop the given mirror. .Pp Additional options include: .Bl -tag -width ".Fl f" .It Fl f Stop the given mirror even if it is opened. .El .It Cm dump Dump metadata stored on the given providers. .It Cm list See .Xr geom 8 . .It Cm status See .Xr geom 8 . .It Cm load See .Xr geom 8 . .It Cm unload See .Xr geom 8 . .El .Pp Additional options include: .Bl -tag -width ".Fl v" .It Fl v Be more verbose. .El .Sh EXIT STATUS Exit status is 0 on success, and 1 if the command fails. .Sh EXAMPLES Use 3 disks to setup a mirror. Choose split balance algorithm, split only requests which are bigger than or equal to 2kB. Create file system, mount it, then unmount it and stop device: .Bd -literal -offset indent gmirror label -v -b split -s 2048 data da0 da1 da2 newfs /dev/mirror/data mount /dev/mirror/data /mnt \&... umount /mnt gmirror stop data gmirror unload .Ed .Pp Create a mirror on disk with valid data (note that the last sector of the disk will be overwritten). Add another disk to this mirror, so it will be synchronized with existing disk: .Bd -literal -offset indent gmirror label -v -b round-robin data da0 gmirror insert data da1 .Ed .Pp Create a mirror, but do not use automatic synchronization feature. Add another disk and rebuild it: .Bd -literal -offset indent gmirror label -v -n -b load data da0 da1 gmirror insert data da2 gmirror rebuild data da2 .Ed .Pp One disk failed. Replace it with a brand new one: .Bd -literal -offset indent gmirror forget data gmirror insert data da1 .Ed .Pp Create a mirror, deactivate one component, do the backup and connect it again. It will not be resynchronized, if there is no need to do so (there were no writes in the meantime): .Bd -literal -offset indent gmirror label data da0 da1 gmirror deactivate data da1 dd if=/dev/da1 of=/backup/data.img bs=1m gmirror activate data da1 .Ed .Sh NOTES Doing kernel dumps to .Nm providers is possible, but some conditions have to be met. First of all, a kernel dump will go only to one component and .Nm always chooses the component with the highest priority. Reading a dump from the mirror on boot will only work if the .Cm prefer balance algorithm is used (that way .Nm will read only from the component with the highest priority). If you use a different balance algorithm, you should add: .Bd -literal -offset indent gmirror configure -b prefer data .Ed .Pp to the .Pa /etc/rc.early script and: .Bd -literal -offset indent gmirror configure -b round-robin data .Ed .Pp to the .Pa /etc/rc.local script. The decision which component to choose for dumping is made when .Xr dumpon 8 is called. If on the next boot a component with a higher priority will be available, the prefer algorithm will choose to read from it and .Xr savecore 8 will find nothing. If on the next boot a component with the highest priority will be synchronized, the prefer balance algorithm will read from the next one, thus will find nothing there. .Sh SEE ALSO .Xr geom 4 , .Xr dumpon 8 , .Xr geom 8 , .Xr gvinum 8 , .Xr mount 8 , .Xr newfs 8 , .Xr savecore 8 , .Xr umount 8 .Sh HISTORY The .Nm utility appeared in .Fx 5.3 . .Sh AUTHORS .An Pawel Jakub Dawidek Aq Mt pjd@FreeBSD.org .Sh BUGS There should be a way to change a component's priority inside a running mirror. .Pp There should be a section with an implementation description. .Pp Documentation for sysctls .Va kern.geom.mirror.* is missing. Index: stable/11/sys/geom/mirror/g_mirror.c =================================================================== --- stable/11/sys/geom/mirror/g_mirror.c (revision 318751) +++ stable/11/sys/geom/mirror/g_mirror.c (revision 318752) @@ -1,3375 +1,3396 @@ /*- * Copyright (c) 2004-2006 Pawel Jakub Dawidek * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include FEATURE(geom_mirror, "GEOM mirroring support"); static MALLOC_DEFINE(M_MIRROR, "mirror_data", "GEOM_MIRROR Data"); SYSCTL_DECL(_kern_geom); static SYSCTL_NODE(_kern_geom, OID_AUTO, mirror, CTLFLAG_RW, 0, "GEOM_MIRROR stuff"); u_int g_mirror_debug = 0; SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, debug, CTLFLAG_RWTUN, &g_mirror_debug, 0, "Debug level"); static u_int g_mirror_timeout = 4; SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, timeout, CTLFLAG_RWTUN, &g_mirror_timeout, 0, "Time to wait on all mirror components"); static u_int g_mirror_idletime = 5; SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, idletime, CTLFLAG_RWTUN, &g_mirror_idletime, 0, "Mark components as clean when idling"); static u_int g_mirror_disconnect_on_failure = 1; SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, disconnect_on_failure, CTLFLAG_RWTUN, &g_mirror_disconnect_on_failure, 0, "Disconnect component on I/O failure."); static u_int g_mirror_syncreqs = 2; SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, sync_requests, CTLFLAG_RDTUN, &g_mirror_syncreqs, 0, "Parallel synchronization I/O requests."); #define MSLEEP(ident, mtx, priority, wmesg, timeout) do { \ G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ } while (0) static eventhandler_tag g_mirror_post_sync = NULL; static int g_mirror_shutdown = 0; static g_ctl_destroy_geom_t g_mirror_destroy_geom; static g_taste_t g_mirror_taste; static g_init_t g_mirror_init; static g_fini_t g_mirror_fini; static g_provgone_t g_mirror_providergone; static g_resize_t g_mirror_resize; struct g_class g_mirror_class = { .name = G_MIRROR_CLASS_NAME, .version = G_VERSION, .ctlreq = g_mirror_config, .taste = g_mirror_taste, .destroy_geom = g_mirror_destroy_geom, .init = g_mirror_init, .fini = g_mirror_fini, .providergone = g_mirror_providergone, .resize = g_mirror_resize }; static void g_mirror_destroy_provider(struct g_mirror_softc *sc); static int g_mirror_update_disk(struct g_mirror_disk *disk, u_int state); static void g_mirror_update_device(struct g_mirror_softc *sc, bool force); static void g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); static void g_mirror_sync_stop(struct g_mirror_disk *disk, int type); static void g_mirror_register_request(struct bio *bp); static void g_mirror_sync_release(struct g_mirror_softc *sc); static const char * g_mirror_disk_state2str(int state) { switch (state) { case G_MIRROR_DISK_STATE_NONE: return ("NONE"); case G_MIRROR_DISK_STATE_NEW: return ("NEW"); case G_MIRROR_DISK_STATE_ACTIVE: return ("ACTIVE"); case G_MIRROR_DISK_STATE_STALE: return ("STALE"); case G_MIRROR_DISK_STATE_SYNCHRONIZING: return ("SYNCHRONIZING"); case G_MIRROR_DISK_STATE_DISCONNECTED: return ("DISCONNECTED"); case G_MIRROR_DISK_STATE_DESTROY: return ("DESTROY"); default: return ("INVALID"); } } static const char * g_mirror_device_state2str(int state) { switch (state) { case G_MIRROR_DEVICE_STATE_STARTING: return ("STARTING"); case G_MIRROR_DEVICE_STATE_RUNNING: return ("RUNNING"); default: return ("INVALID"); } } static const char * g_mirror_get_diskname(struct g_mirror_disk *disk) { if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) return ("[unknown]"); return (disk->d_name); } /* * --- Events handling functions --- * Events in geom_mirror are used to maintain disks and device status * from one thread to simplify locking. */ static void g_mirror_event_free(struct g_mirror_event *ep) { free(ep, M_MIRROR); } int g_mirror_event_send(void *arg, int state, int flags) { struct g_mirror_softc *sc; struct g_mirror_disk *disk; struct g_mirror_event *ep; int error; ep = malloc(sizeof(*ep), M_MIRROR, M_WAITOK); G_MIRROR_DEBUG(4, "%s: Sending event %p.", __func__, ep); if ((flags & G_MIRROR_EVENT_DEVICE) != 0) { disk = NULL; sc = arg; } else { disk = arg; sc = disk->d_softc; } ep->e_disk = disk; ep->e_state = state; ep->e_flags = flags; ep->e_error = 0; mtx_lock(&sc->sc_events_mtx); TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); mtx_unlock(&sc->sc_events_mtx); G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); mtx_lock(&sc->sc_queue_mtx); wakeup(sc); mtx_unlock(&sc->sc_queue_mtx); if ((flags & G_MIRROR_EVENT_DONTWAIT) != 0) return (0); sx_assert(&sc->sc_lock, SX_XLOCKED); G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, ep); sx_xunlock(&sc->sc_lock); while ((ep->e_flags & G_MIRROR_EVENT_DONE) == 0) { mtx_lock(&sc->sc_events_mtx); MSLEEP(ep, &sc->sc_events_mtx, PRIBIO | PDROP, "m:event", hz * 5); } error = ep->e_error; g_mirror_event_free(ep); sx_xlock(&sc->sc_lock); return (error); } static struct g_mirror_event * g_mirror_event_get(struct g_mirror_softc *sc) { struct g_mirror_event *ep; mtx_lock(&sc->sc_events_mtx); ep = TAILQ_FIRST(&sc->sc_events); mtx_unlock(&sc->sc_events_mtx); return (ep); } static void g_mirror_event_remove(struct g_mirror_softc *sc, struct g_mirror_event *ep) { mtx_lock(&sc->sc_events_mtx); TAILQ_REMOVE(&sc->sc_events, ep, e_next); mtx_unlock(&sc->sc_events_mtx); } static void g_mirror_event_cancel(struct g_mirror_disk *disk) { struct g_mirror_softc *sc; struct g_mirror_event *ep, *tmpep; sc = disk->d_softc; sx_assert(&sc->sc_lock, SX_XLOCKED); mtx_lock(&sc->sc_events_mtx); TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) continue; if (ep->e_disk != disk) continue; TAILQ_REMOVE(&sc->sc_events, ep, e_next); if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) g_mirror_event_free(ep); else { ep->e_error = ECANCELED; wakeup(ep); } } mtx_unlock(&sc->sc_events_mtx); } /* * Return the number of disks in given state. * If state is equal to -1, count all connected disks. */ u_int g_mirror_ndisks(struct g_mirror_softc *sc, int state) { struct g_mirror_disk *disk; u_int n = 0; sx_assert(&sc->sc_lock, SX_LOCKED); LIST_FOREACH(disk, &sc->sc_disks, d_next) { if (state == -1 || disk->d_state == state) n++; } return (n); } /* * Find a disk in mirror by its disk ID. */ static struct g_mirror_disk * g_mirror_id2disk(struct g_mirror_softc *sc, uint32_t id) { struct g_mirror_disk *disk; sx_assert(&sc->sc_lock, SX_XLOCKED); LIST_FOREACH(disk, &sc->sc_disks, d_next) { if (disk->d_id == id) return (disk); } return (NULL); } static u_int g_mirror_nrequests(struct g_mirror_softc *sc, struct g_consumer *cp) { struct bio *bp; u_int nreqs = 0; mtx_lock(&sc->sc_queue_mtx); TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { if (bp->bio_from == cp) nreqs++; } mtx_unlock(&sc->sc_queue_mtx); return (nreqs); } static int g_mirror_is_busy(struct g_mirror_softc *sc, struct g_consumer *cp) { if (cp->index > 0) { G_MIRROR_DEBUG(2, "I/O requests for %s exist, can't destroy it now.", cp->provider->name); return (1); } if (g_mirror_nrequests(sc, cp) > 0) { G_MIRROR_DEBUG(2, "I/O requests for %s in queue, can't destroy it now.", cp->provider->name); return (1); } return (0); } static void g_mirror_destroy_consumer(void *arg, int flags __unused) { struct g_consumer *cp; g_topology_assert(); cp = arg; G_MIRROR_DEBUG(1, "Consumer %s destroyed.", cp->provider->name); g_detach(cp); g_destroy_consumer(cp); } static void g_mirror_kill_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) { struct g_provider *pp; int retaste_wait; g_topology_assert(); cp->private = NULL; if (g_mirror_is_busy(sc, cp)) return; pp = cp->provider; retaste_wait = 0; if (cp->acw == 1) { if ((pp->geom->flags & G_GEOM_WITHER) == 0) retaste_wait = 1; } G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", pp->name, -cp->acr, -cp->acw, -cp->ace, 0); if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) g_access(cp, -cp->acr, -cp->acw, -cp->ace); if (retaste_wait) { /* * After retaste event was send (inside g_access()), we can send * event to detach and destroy consumer. * A class, which has consumer to the given provider connected * will not receive retaste event for the provider. * This is the way how I ignore retaste events when I close * consumers opened for write: I detach and destroy consumer * after retaste event is sent. */ g_post_event(g_mirror_destroy_consumer, cp, M_WAITOK, NULL); return; } G_MIRROR_DEBUG(1, "Consumer %s destroyed.", pp->name); g_detach(cp); g_destroy_consumer(cp); } static int g_mirror_connect_disk(struct g_mirror_disk *disk, struct g_provider *pp) { struct g_consumer *cp; int error; g_topology_assert_not(); KASSERT(disk->d_consumer == NULL, ("Disk already connected (device %s).", disk->d_softc->sc_name)); g_topology_lock(); cp = g_new_consumer(disk->d_softc->sc_geom); cp->flags |= G_CF_DIRECT_RECEIVE; error = g_attach(cp, pp); if (error != 0) { g_destroy_consumer(cp); g_topology_unlock(); return (error); } error = g_access(cp, 1, 1, 1); if (error != 0) { g_detach(cp); g_destroy_consumer(cp); g_topology_unlock(); G_MIRROR_DEBUG(0, "Cannot open consumer %s (error=%d).", pp->name, error); return (error); } g_topology_unlock(); disk->d_consumer = cp; disk->d_consumer->private = disk; disk->d_consumer->index = 0; G_MIRROR_DEBUG(2, "Disk %s connected.", g_mirror_get_diskname(disk)); return (0); } static void g_mirror_disconnect_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) { g_topology_assert(); if (cp == NULL) return; if (cp->provider != NULL) g_mirror_kill_consumer(sc, cp); else g_destroy_consumer(cp); } /* * Initialize disk. This means allocate memory, create consumer, attach it * to the provider and open access (r1w1e1) to it. */ static struct g_mirror_disk * g_mirror_init_disk(struct g_mirror_softc *sc, struct g_provider *pp, struct g_mirror_metadata *md, int *errorp) { struct g_mirror_disk *disk; int i, error; disk = malloc(sizeof(*disk), M_MIRROR, M_NOWAIT | M_ZERO); if (disk == NULL) { error = ENOMEM; goto fail; } disk->d_softc = sc; error = g_mirror_connect_disk(disk, pp); if (error != 0) goto fail; disk->d_id = md->md_did; disk->d_state = G_MIRROR_DISK_STATE_NONE; disk->d_priority = md->md_priority; disk->d_flags = md->md_dflags; error = g_getattr("GEOM::candelete", disk->d_consumer, &i); if (error == 0 && i != 0) disk->d_flags |= G_MIRROR_DISK_FLAG_CANDELETE; if (md->md_provider[0] != '\0') disk->d_flags |= G_MIRROR_DISK_FLAG_HARDCODED; disk->d_sync.ds_consumer = NULL; disk->d_sync.ds_offset = md->md_sync_offset; disk->d_sync.ds_offset_done = md->md_sync_offset; disk->d_genid = md->md_genid; disk->d_sync.ds_syncid = md->md_syncid; if (errorp != NULL) *errorp = 0; return (disk); fail: if (errorp != NULL) *errorp = error; if (disk != NULL) free(disk, M_MIRROR); return (NULL); } static void g_mirror_destroy_disk(struct g_mirror_disk *disk) { struct g_mirror_softc *sc; g_topology_assert_not(); sc = disk->d_softc; sx_assert(&sc->sc_lock, SX_XLOCKED); LIST_REMOVE(disk, d_next); g_mirror_event_cancel(disk); if (sc->sc_hint == disk) sc->sc_hint = NULL; switch (disk->d_state) { case G_MIRROR_DISK_STATE_SYNCHRONIZING: g_mirror_sync_stop(disk, 1); /* FALLTHROUGH */ case G_MIRROR_DISK_STATE_NEW: case G_MIRROR_DISK_STATE_STALE: case G_MIRROR_DISK_STATE_ACTIVE: g_topology_lock(); g_mirror_disconnect_consumer(sc, disk->d_consumer); g_topology_unlock(); free(disk, M_MIRROR); break; default: KASSERT(0 == 1, ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state))); } } static void g_mirror_free_device(struct g_mirror_softc *sc) { mtx_destroy(&sc->sc_queue_mtx); mtx_destroy(&sc->sc_events_mtx); mtx_destroy(&sc->sc_done_mtx); sx_destroy(&sc->sc_lock); free(sc, M_MIRROR); } static void g_mirror_providergone(struct g_provider *pp) { struct g_mirror_softc *sc = pp->private; if ((--sc->sc_refcnt) == 0) g_mirror_free_device(sc); } static void g_mirror_destroy_device(struct g_mirror_softc *sc) { struct g_mirror_disk *disk; struct g_mirror_event *ep; struct g_geom *gp; struct g_consumer *cp, *tmpcp; g_topology_assert_not(); sx_assert(&sc->sc_lock, SX_XLOCKED); gp = sc->sc_geom; if (sc->sc_provider != NULL) g_mirror_destroy_provider(sc); for (disk = LIST_FIRST(&sc->sc_disks); disk != NULL; disk = LIST_FIRST(&sc->sc_disks)) { disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; g_mirror_update_metadata(disk); g_mirror_destroy_disk(disk); } while ((ep = g_mirror_event_get(sc)) != NULL) { g_mirror_event_remove(sc, ep); if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) g_mirror_event_free(ep); else { ep->e_error = ECANCELED; ep->e_flags |= G_MIRROR_EVENT_DONE; G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, ep); mtx_lock(&sc->sc_events_mtx); wakeup(ep); mtx_unlock(&sc->sc_events_mtx); } } callout_drain(&sc->sc_callout); g_topology_lock(); LIST_FOREACH_SAFE(cp, &sc->sc_sync.ds_geom->consumer, consumer, tmpcp) { g_mirror_disconnect_consumer(sc, cp); } g_wither_geom(sc->sc_sync.ds_geom, ENXIO); G_MIRROR_DEBUG(0, "Device %s destroyed.", gp->name); g_wither_geom(gp, ENXIO); sx_xunlock(&sc->sc_lock); if ((--sc->sc_refcnt) == 0) g_mirror_free_device(sc); g_topology_unlock(); } static void g_mirror_orphan(struct g_consumer *cp) { struct g_mirror_disk *disk; g_topology_assert(); disk = cp->private; if (disk == NULL) return; disk->d_softc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, G_MIRROR_EVENT_DONTWAIT); } /* * Function should return the next active disk on the list. * It is possible that it will be the same disk as given. * If there are no active disks on list, NULL is returned. */ static __inline struct g_mirror_disk * g_mirror_find_next(struct g_mirror_softc *sc, struct g_mirror_disk *disk) { struct g_mirror_disk *dp; for (dp = LIST_NEXT(disk, d_next); dp != disk; dp = LIST_NEXT(dp, d_next)) { if (dp == NULL) dp = LIST_FIRST(&sc->sc_disks); if (dp->d_state == G_MIRROR_DISK_STATE_ACTIVE) break; } if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) return (NULL); return (dp); } static struct g_mirror_disk * g_mirror_get_disk(struct g_mirror_softc *sc) { struct g_mirror_disk *disk; if (sc->sc_hint == NULL) { sc->sc_hint = LIST_FIRST(&sc->sc_disks); if (sc->sc_hint == NULL) return (NULL); } disk = sc->sc_hint; if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) { disk = g_mirror_find_next(sc, disk); if (disk == NULL) return (NULL); } sc->sc_hint = g_mirror_find_next(sc, disk); return (disk); } static int g_mirror_write_metadata(struct g_mirror_disk *disk, struct g_mirror_metadata *md) { struct g_mirror_softc *sc; struct g_consumer *cp; off_t offset, length; u_char *sector; int error = 0; g_topology_assert_not(); sc = disk->d_softc; sx_assert(&sc->sc_lock, SX_LOCKED); cp = disk->d_consumer; KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, ("Consumer %s closed? (r%dw%de%d).", cp->provider->name, cp->acr, cp->acw, cp->ace)); length = cp->provider->sectorsize; offset = cp->provider->mediasize - length; sector = malloc((size_t)length, M_MIRROR, M_WAITOK | M_ZERO); if (md != NULL && (sc->sc_flags & G_MIRROR_DEVICE_FLAG_WIPE) == 0) { /* * Handle the case, when the size of parent provider reduced. */ if (offset < md->md_mediasize) error = ENOSPC; else mirror_metadata_encode(md, sector); } if (error == 0) error = g_write_data(cp, offset, sector, length); free(sector, M_MIRROR); if (error != 0) { if ((disk->d_flags & G_MIRROR_DISK_FLAG_BROKEN) == 0) { disk->d_flags |= G_MIRROR_DISK_FLAG_BROKEN; G_MIRROR_DEBUG(0, "Cannot write metadata on %s " "(device=%s, error=%d).", g_mirror_get_diskname(disk), sc->sc_name, error); } else { G_MIRROR_DEBUG(1, "Cannot write metadata on %s " "(device=%s, error=%d).", g_mirror_get_diskname(disk), sc->sc_name, error); } if (g_mirror_disconnect_on_failure && g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 1) { sc->sc_bump_id |= G_MIRROR_BUMP_GENID; g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, G_MIRROR_EVENT_DONTWAIT); } } return (error); } static int g_mirror_clear_metadata(struct g_mirror_disk *disk) { int error; g_topology_assert_not(); sx_assert(&disk->d_softc->sc_lock, SX_LOCKED); + if (disk->d_softc->sc_type != G_MIRROR_TYPE_AUTOMATIC) + return (0); error = g_mirror_write_metadata(disk, NULL); if (error == 0) { G_MIRROR_DEBUG(2, "Metadata on %s cleared.", g_mirror_get_diskname(disk)); } else { G_MIRROR_DEBUG(0, "Cannot clear metadata on disk %s (error=%d).", g_mirror_get_diskname(disk), error); } return (error); } void g_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk, struct g_mirror_metadata *md) { strlcpy(md->md_magic, G_MIRROR_MAGIC, sizeof(md->md_magic)); md->md_version = G_MIRROR_VERSION; strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); md->md_mid = sc->sc_id; md->md_all = sc->sc_ndisks; md->md_slice = sc->sc_slice; md->md_balance = sc->sc_balance; md->md_genid = sc->sc_genid; md->md_mediasize = sc->sc_mediasize; md->md_sectorsize = sc->sc_sectorsize; md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK); bzero(md->md_provider, sizeof(md->md_provider)); if (disk == NULL) { md->md_did = arc4random(); md->md_priority = 0; md->md_syncid = 0; md->md_dflags = 0; md->md_sync_offset = 0; md->md_provsize = 0; } else { md->md_did = disk->d_id; md->md_priority = disk->d_priority; md->md_syncid = disk->d_sync.ds_syncid; md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK); if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) md->md_sync_offset = disk->d_sync.ds_offset_done; else md->md_sync_offset = 0; if ((disk->d_flags & G_MIRROR_DISK_FLAG_HARDCODED) != 0) { strlcpy(md->md_provider, disk->d_consumer->provider->name, sizeof(md->md_provider)); } md->md_provsize = disk->d_consumer->provider->mediasize; } } void g_mirror_update_metadata(struct g_mirror_disk *disk) { struct g_mirror_softc *sc; struct g_mirror_metadata md; int error; g_topology_assert_not(); sc = disk->d_softc; sx_assert(&sc->sc_lock, SX_LOCKED); + if (sc->sc_type != G_MIRROR_TYPE_AUTOMATIC) + return; if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WIPE) == 0) g_mirror_fill_metadata(sc, disk, &md); error = g_mirror_write_metadata(disk, &md); if (error == 0) { G_MIRROR_DEBUG(2, "Metadata on %s updated.", g_mirror_get_diskname(disk)); } else { G_MIRROR_DEBUG(0, "Cannot update metadata on disk %s (error=%d).", g_mirror_get_diskname(disk), error); } } static void g_mirror_bump_syncid(struct g_mirror_softc *sc) { struct g_mirror_disk *disk; g_topology_assert_not(); sx_assert(&sc->sc_lock, SX_XLOCKED); KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, ("%s called with no active disks (device=%s).", __func__, sc->sc_name)); sc->sc_syncid++; G_MIRROR_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, sc->sc_syncid); LIST_FOREACH(disk, &sc->sc_disks, d_next) { if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { disk->d_sync.ds_syncid = sc->sc_syncid; g_mirror_update_metadata(disk); } } } static void g_mirror_bump_genid(struct g_mirror_softc *sc) { struct g_mirror_disk *disk; g_topology_assert_not(); sx_assert(&sc->sc_lock, SX_XLOCKED); KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, ("%s called with no active disks (device=%s).", __func__, sc->sc_name)); sc->sc_genid++; G_MIRROR_DEBUG(1, "Device %s: genid bumped to %u.", sc->sc_name, sc->sc_genid); LIST_FOREACH(disk, &sc->sc_disks, d_next) { if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { disk->d_genid = sc->sc_genid; g_mirror_update_metadata(disk); } } } static int g_mirror_idle(struct g_mirror_softc *sc, int acw) { struct g_mirror_disk *disk; int timeout; g_topology_assert_not(); sx_assert(&sc->sc_lock, SX_XLOCKED); if (sc->sc_provider == NULL) return (0); if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) return (0); if (sc->sc_idle) return (0); if (sc->sc_writes > 0) return (0); if (acw > 0 || (acw == -1 && sc->sc_provider->acw > 0)) { timeout = g_mirror_idletime - (time_uptime - sc->sc_last_write); if (!g_mirror_shutdown && timeout > 0) return (timeout); } sc->sc_idle = 1; LIST_FOREACH(disk, &sc->sc_disks, d_next) { if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) continue; G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", g_mirror_get_diskname(disk), sc->sc_name); disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; g_mirror_update_metadata(disk); } return (0); } static void g_mirror_unidle(struct g_mirror_softc *sc) { struct g_mirror_disk *disk; g_topology_assert_not(); sx_assert(&sc->sc_lock, SX_XLOCKED); if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) return; sc->sc_idle = 0; sc->sc_last_write = time_uptime; LIST_FOREACH(disk, &sc->sc_disks, d_next) { if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) continue; G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", g_mirror_get_diskname(disk), sc->sc_name); disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; g_mirror_update_metadata(disk); } } static void g_mirror_flush_done(struct bio *bp) { struct g_mirror_softc *sc; struct bio *pbp; pbp = bp->bio_parent; sc = pbp->bio_to->private; mtx_lock(&sc->sc_done_mtx); if (pbp->bio_error == 0) pbp->bio_error = bp->bio_error; pbp->bio_completed += bp->bio_completed; pbp->bio_inbed++; if (pbp->bio_children == pbp->bio_inbed) { mtx_unlock(&sc->sc_done_mtx); g_io_deliver(pbp, pbp->bio_error); } else mtx_unlock(&sc->sc_done_mtx); g_destroy_bio(bp); } static void g_mirror_done(struct bio *bp) { struct g_mirror_softc *sc; sc = bp->bio_from->geom->softc; bp->bio_cflags = G_MIRROR_BIO_FLAG_REGULAR; mtx_lock(&sc->sc_queue_mtx); bioq_insert_tail(&sc->sc_queue, bp); mtx_unlock(&sc->sc_queue_mtx); wakeup(sc); } static void g_mirror_regular_request(struct bio *bp) { struct g_mirror_softc *sc; struct g_mirror_disk *disk; struct bio *pbp; g_topology_assert_not(); pbp = bp->bio_parent; sc = pbp->bio_to->private; bp->bio_from->index--; if (bp->bio_cmd == BIO_WRITE) sc->sc_writes--; disk = bp->bio_from->private; if (disk == NULL) { g_topology_lock(); g_mirror_kill_consumer(sc, bp->bio_from); g_topology_unlock(); } pbp->bio_inbed++; KASSERT(pbp->bio_inbed <= pbp->bio_children, ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, pbp->bio_children)); if (bp->bio_error == 0 && pbp->bio_error == 0) { G_MIRROR_LOGREQ(3, bp, "Request delivered."); g_destroy_bio(bp); if (pbp->bio_children == pbp->bio_inbed) { G_MIRROR_LOGREQ(3, pbp, "Request delivered."); pbp->bio_completed = pbp->bio_length; if (pbp->bio_cmd == BIO_WRITE || pbp->bio_cmd == BIO_DELETE) { bioq_remove(&sc->sc_inflight, pbp); /* Release delayed sync requests if possible. */ g_mirror_sync_release(sc); } g_io_deliver(pbp, pbp->bio_error); } return; } else if (bp->bio_error != 0) { if (pbp->bio_error == 0) pbp->bio_error = bp->bio_error; if (disk != NULL) { if ((disk->d_flags & G_MIRROR_DISK_FLAG_BROKEN) == 0) { disk->d_flags |= G_MIRROR_DISK_FLAG_BROKEN; G_MIRROR_LOGREQ(0, bp, "Request failed (error=%d).", bp->bio_error); } else { G_MIRROR_LOGREQ(1, bp, "Request failed (error=%d).", bp->bio_error); } if (g_mirror_disconnect_on_failure && g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 1) { sc->sc_bump_id |= G_MIRROR_BUMP_GENID; g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, G_MIRROR_EVENT_DONTWAIT); } } switch (pbp->bio_cmd) { case BIO_DELETE: case BIO_WRITE: pbp->bio_inbed--; pbp->bio_children--; break; } } g_destroy_bio(bp); switch (pbp->bio_cmd) { case BIO_READ: if (pbp->bio_inbed < pbp->bio_children) break; if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 1) g_io_deliver(pbp, pbp->bio_error); else { pbp->bio_error = 0; mtx_lock(&sc->sc_queue_mtx); bioq_insert_tail(&sc->sc_queue, pbp); mtx_unlock(&sc->sc_queue_mtx); G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); wakeup(sc); } break; case BIO_DELETE: case BIO_WRITE: if (pbp->bio_children == 0) { /* * All requests failed. */ } else if (pbp->bio_inbed < pbp->bio_children) { /* Do nothing. */ break; } else if (pbp->bio_children == pbp->bio_inbed) { /* Some requests succeeded. */ pbp->bio_error = 0; pbp->bio_completed = pbp->bio_length; } bioq_remove(&sc->sc_inflight, pbp); /* Release delayed sync requests if possible. */ g_mirror_sync_release(sc); g_io_deliver(pbp, pbp->bio_error); break; default: KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); break; } } static void g_mirror_sync_done(struct bio *bp) { struct g_mirror_softc *sc; G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); sc = bp->bio_from->geom->softc; bp->bio_cflags = G_MIRROR_BIO_FLAG_SYNC; mtx_lock(&sc->sc_queue_mtx); bioq_insert_tail(&sc->sc_queue, bp); mtx_unlock(&sc->sc_queue_mtx); wakeup(sc); } static void g_mirror_candelete(struct bio *bp) { struct g_mirror_softc *sc; struct g_mirror_disk *disk; int *val; sc = bp->bio_to->private; LIST_FOREACH(disk, &sc->sc_disks, d_next) { if (disk->d_flags & G_MIRROR_DISK_FLAG_CANDELETE) break; } val = (int *)bp->bio_data; *val = (disk != NULL); g_io_deliver(bp, 0); } static void g_mirror_kernel_dump(struct bio *bp) { struct g_mirror_softc *sc; struct g_mirror_disk *disk; struct bio *cbp; struct g_kerneldump *gkd; /* * We configure dumping to the first component, because this component * will be used for reading with 'prefer' balance algorithm. * If the component with the highest priority is currently disconnected * we will not be able to read the dump after the reboot if it will be * connected and synchronized later. Can we do something better? */ sc = bp->bio_to->private; disk = LIST_FIRST(&sc->sc_disks); gkd = (struct g_kerneldump *)bp->bio_data; if (gkd->length > bp->bio_to->mediasize) gkd->length = bp->bio_to->mediasize; cbp = g_clone_bio(bp); if (cbp == NULL) { g_io_deliver(bp, ENOMEM); return; } cbp->bio_done = g_std_done; g_io_request(cbp, disk->d_consumer); G_MIRROR_DEBUG(1, "Kernel dump will go to %s.", g_mirror_get_diskname(disk)); } static void g_mirror_flush(struct g_mirror_softc *sc, struct bio *bp) { struct bio_queue_head queue; struct g_mirror_disk *disk; struct g_consumer *cp; struct bio *cbp; bioq_init(&queue); LIST_FOREACH(disk, &sc->sc_disks, d_next) { if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) continue; cbp = g_clone_bio(bp); if (cbp == NULL) { while ((cbp = bioq_takefirst(&queue)) != NULL) g_destroy_bio(cbp); if (bp->bio_error == 0) bp->bio_error = ENOMEM; g_io_deliver(bp, bp->bio_error); return; } bioq_insert_tail(&queue, cbp); cbp->bio_done = g_mirror_flush_done; cbp->bio_caller1 = disk; cbp->bio_to = disk->d_consumer->provider; } while ((cbp = bioq_takefirst(&queue)) != NULL) { G_MIRROR_LOGREQ(3, cbp, "Sending request."); disk = cbp->bio_caller1; cbp->bio_caller1 = NULL; cp = disk->d_consumer; KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, cp->acw, cp->ace)); g_io_request(cbp, disk->d_consumer); } } static void g_mirror_start(struct bio *bp) { struct g_mirror_softc *sc; sc = bp->bio_to->private; /* * If sc == NULL or there are no valid disks, provider's error * should be set and g_mirror_start() should not be called at all. */ KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, ("Provider's error should be set (error=%d)(mirror=%s).", bp->bio_to->error, bp->bio_to->name)); G_MIRROR_LOGREQ(3, bp, "Request received."); switch (bp->bio_cmd) { case BIO_READ: case BIO_WRITE: case BIO_DELETE: break; case BIO_FLUSH: g_mirror_flush(sc, bp); return; case BIO_GETATTR: if (!strcmp(bp->bio_attribute, "GEOM::candelete")) { g_mirror_candelete(bp); return; } else if (strcmp("GEOM::kerneldump", bp->bio_attribute) == 0) { g_mirror_kernel_dump(bp); return; } /* FALLTHROUGH */ default: g_io_deliver(bp, EOPNOTSUPP); return; } mtx_lock(&sc->sc_queue_mtx); bioq_insert_tail(&sc->sc_queue, bp); mtx_unlock(&sc->sc_queue_mtx); G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); wakeup(sc); } /* * Return TRUE if the given request is colliding with a in-progress * synchronization request. */ static int g_mirror_sync_collision(struct g_mirror_softc *sc, struct bio *bp) { struct g_mirror_disk *disk; struct bio *sbp; off_t rstart, rend, sstart, send; u_int i; if (sc->sc_sync.ds_ndisks == 0) return (0); rstart = bp->bio_offset; rend = bp->bio_offset + bp->bio_length; LIST_FOREACH(disk, &sc->sc_disks, d_next) { if (disk->d_state != G_MIRROR_DISK_STATE_SYNCHRONIZING) continue; for (i = 0; i < g_mirror_syncreqs; i++) { sbp = disk->d_sync.ds_bios[i]; if (sbp == NULL) continue; sstart = sbp->bio_offset; send = sbp->bio_offset + sbp->bio_length; if (rend > sstart && rstart < send) return (1); } } return (0); } /* * Return TRUE if the given sync request is colliding with a in-progress regular * request. */ static int g_mirror_regular_collision(struct g_mirror_softc *sc, struct bio *sbp) { off_t rstart, rend, sstart, send; struct bio *bp; if (sc->sc_sync.ds_ndisks == 0) return (0); sstart = sbp->bio_offset; send = sbp->bio_offset + sbp->bio_length; TAILQ_FOREACH(bp, &sc->sc_inflight.queue, bio_queue) { rstart = bp->bio_offset; rend = bp->bio_offset + bp->bio_length; if (rend > sstart && rstart < send) return (1); } return (0); } /* * Puts request onto delayed queue. */ static void g_mirror_regular_delay(struct g_mirror_softc *sc, struct bio *bp) { G_MIRROR_LOGREQ(2, bp, "Delaying request."); bioq_insert_head(&sc->sc_regular_delayed, bp); } /* * Puts synchronization request onto delayed queue. */ static void g_mirror_sync_delay(struct g_mirror_softc *sc, struct bio *bp) { G_MIRROR_LOGREQ(2, bp, "Delaying synchronization request."); bioq_insert_tail(&sc->sc_sync_delayed, bp); } /* * Releases delayed regular requests which don't collide anymore with sync * requests. */ static void g_mirror_regular_release(struct g_mirror_softc *sc) { struct bio *bp, *bp2; TAILQ_FOREACH_SAFE(bp, &sc->sc_regular_delayed.queue, bio_queue, bp2) { if (g_mirror_sync_collision(sc, bp)) continue; bioq_remove(&sc->sc_regular_delayed, bp); G_MIRROR_LOGREQ(2, bp, "Releasing delayed request (%p).", bp); mtx_lock(&sc->sc_queue_mtx); bioq_insert_head(&sc->sc_queue, bp); #if 0 /* * wakeup() is not needed, because this function is called from * the worker thread. */ wakeup(&sc->sc_queue); #endif mtx_unlock(&sc->sc_queue_mtx); } } /* * Releases delayed sync requests which don't collide anymore with regular * requests. */ static void g_mirror_sync_release(struct g_mirror_softc *sc) { struct bio *bp, *bp2; TAILQ_FOREACH_SAFE(bp, &sc->sc_sync_delayed.queue, bio_queue, bp2) { if (g_mirror_regular_collision(sc, bp)) continue; bioq_remove(&sc->sc_sync_delayed, bp); G_MIRROR_LOGREQ(2, bp, "Releasing delayed synchronization request."); g_io_request(bp, bp->bio_from); } } /* * Handle synchronization requests. * Every synchronization request is two-steps process: first, READ request is * send to active provider and then WRITE request (with read data) to the provider * being synchronized. When WRITE is finished, new synchronization request is * send. */ static void g_mirror_sync_request(struct bio *bp) { struct g_mirror_softc *sc; struct g_mirror_disk *disk; bp->bio_from->index--; sc = bp->bio_from->geom->softc; disk = bp->bio_from->private; if (disk == NULL) { sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ g_topology_lock(); g_mirror_kill_consumer(sc, bp->bio_from); g_topology_unlock(); free(bp->bio_data, M_MIRROR); g_destroy_bio(bp); sx_xlock(&sc->sc_lock); return; } /* * Synchronization request. */ switch (bp->bio_cmd) { case BIO_READ: { struct g_consumer *cp; if (bp->bio_error != 0) { G_MIRROR_LOGREQ(0, bp, "Synchronization request failed (error=%d).", bp->bio_error); g_destroy_bio(bp); return; } G_MIRROR_LOGREQ(3, bp, "Synchronization request half-finished."); bp->bio_cmd = BIO_WRITE; bp->bio_cflags = 0; cp = disk->d_consumer; KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, cp->acw, cp->ace)); cp->index++; g_io_request(bp, cp); return; } case BIO_WRITE: { struct g_mirror_disk_sync *sync; off_t offset; void *data; int i; if (bp->bio_error != 0) { G_MIRROR_LOGREQ(0, bp, "Synchronization request failed (error=%d).", bp->bio_error); g_destroy_bio(bp); sc->sc_bump_id |= G_MIRROR_BUMP_GENID; g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, G_MIRROR_EVENT_DONTWAIT); return; } G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); sync = &disk->d_sync; if (sync->ds_offset >= sc->sc_mediasize || sync->ds_consumer == NULL || (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { /* Don't send more synchronization requests. */ sync->ds_inflight--; if (sync->ds_bios != NULL) { i = (int)(uintptr_t)bp->bio_caller1; sync->ds_bios[i] = NULL; } free(bp->bio_data, M_MIRROR); g_destroy_bio(bp); if (sync->ds_inflight > 0) return; if (sync->ds_consumer == NULL || (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { return; } /* Disk up-to-date, activate it. */ g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, G_MIRROR_EVENT_DONTWAIT); return; } /* Send next synchronization request. */ data = bp->bio_data; g_reset_bio(bp); bp->bio_cmd = BIO_READ; bp->bio_offset = sync->ds_offset; bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); sync->ds_offset += bp->bio_length; bp->bio_done = g_mirror_sync_done; bp->bio_data = data; bp->bio_from = sync->ds_consumer; bp->bio_to = sc->sc_provider; G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); sync->ds_consumer->index++; /* * Delay the request if it is colliding with a regular request. */ if (g_mirror_regular_collision(sc, bp)) g_mirror_sync_delay(sc, bp); else g_io_request(bp, sync->ds_consumer); /* Release delayed requests if possible. */ g_mirror_regular_release(sc); /* Find the smallest offset */ offset = sc->sc_mediasize; for (i = 0; i < g_mirror_syncreqs; i++) { bp = sync->ds_bios[i]; if (bp->bio_offset < offset) offset = bp->bio_offset; } if (sync->ds_offset_done + (MAXPHYS * 100) < offset) { /* Update offset_done on every 100 blocks. */ sync->ds_offset_done = offset; g_mirror_update_metadata(disk); } return; } default: KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", bp->bio_cmd, sc->sc_name)); break; } } static void g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) { struct g_mirror_disk *disk; struct g_consumer *cp; struct bio *cbp; LIST_FOREACH(disk, &sc->sc_disks, d_next) { if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) break; } if (disk == NULL) { if (bp->bio_error == 0) bp->bio_error = ENXIO; g_io_deliver(bp, bp->bio_error); return; } cbp = g_clone_bio(bp); if (cbp == NULL) { if (bp->bio_error == 0) bp->bio_error = ENOMEM; g_io_deliver(bp, bp->bio_error); return; } /* * Fill in the component buf structure. */ cp = disk->d_consumer; cbp->bio_done = g_mirror_done; cbp->bio_to = cp->provider; G_MIRROR_LOGREQ(3, cbp, "Sending request."); KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, cp->acw, cp->ace)); cp->index++; g_io_request(cbp, cp); } static void g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) { struct g_mirror_disk *disk; struct g_consumer *cp; struct bio *cbp; disk = g_mirror_get_disk(sc); if (disk == NULL) { if (bp->bio_error == 0) bp->bio_error = ENXIO; g_io_deliver(bp, bp->bio_error); return; } cbp = g_clone_bio(bp); if (cbp == NULL) { if (bp->bio_error == 0) bp->bio_error = ENOMEM; g_io_deliver(bp, bp->bio_error); return; } /* * Fill in the component buf structure. */ cp = disk->d_consumer; cbp->bio_done = g_mirror_done; cbp->bio_to = cp->provider; G_MIRROR_LOGREQ(3, cbp, "Sending request."); KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, cp->acw, cp->ace)); cp->index++; g_io_request(cbp, cp); } #define TRACK_SIZE (1 * 1024 * 1024) #define LOAD_SCALE 256 #define ABS(x) (((x) >= 0) ? (x) : (-(x))) static void g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) { struct g_mirror_disk *disk, *dp; struct g_consumer *cp; struct bio *cbp; int prio, best; /* Find a disk with the smallest load. */ disk = NULL; best = INT_MAX; LIST_FOREACH(dp, &sc->sc_disks, d_next) { if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) continue; prio = dp->load; /* If disk head is precisely in position - highly prefer it. */ if (dp->d_last_offset == bp->bio_offset) prio -= 2 * LOAD_SCALE; else /* If disk head is close to position - prefer it. */ if (ABS(dp->d_last_offset - bp->bio_offset) < TRACK_SIZE) prio -= 1 * LOAD_SCALE; if (prio <= best) { disk = dp; best = prio; } } KASSERT(disk != NULL, ("NULL disk for %s.", sc->sc_name)); cbp = g_clone_bio(bp); if (cbp == NULL) { if (bp->bio_error == 0) bp->bio_error = ENOMEM; g_io_deliver(bp, bp->bio_error); return; } /* * Fill in the component buf structure. */ cp = disk->d_consumer; cbp->bio_done = g_mirror_done; cbp->bio_to = cp->provider; G_MIRROR_LOGREQ(3, cbp, "Sending request."); KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, cp->acw, cp->ace)); cp->index++; /* Remember last head position */ disk->d_last_offset = bp->bio_offset + bp->bio_length; /* Update loads. */ LIST_FOREACH(dp, &sc->sc_disks, d_next) { dp->load = (dp->d_consumer->index * LOAD_SCALE + dp->load * 7) / 8; } g_io_request(cbp, cp); } static void g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) { struct bio_queue_head queue; struct g_mirror_disk *disk; struct g_consumer *cp; struct bio *cbp; off_t left, mod, offset, slice; u_char *data; u_int ndisks; if (bp->bio_length <= sc->sc_slice) { g_mirror_request_round_robin(sc, bp); return; } ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); slice = bp->bio_length / ndisks; mod = slice % sc->sc_provider->sectorsize; if (mod != 0) slice += sc->sc_provider->sectorsize - mod; /* * Allocate all bios before sending any request, so we can * return ENOMEM in nice and clean way. */ left = bp->bio_length; offset = bp->bio_offset; data = bp->bio_data; bioq_init(&queue); LIST_FOREACH(disk, &sc->sc_disks, d_next) { if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) continue; cbp = g_clone_bio(bp); if (cbp == NULL) { while ((cbp = bioq_takefirst(&queue)) != NULL) g_destroy_bio(cbp); if (bp->bio_error == 0) bp->bio_error = ENOMEM; g_io_deliver(bp, bp->bio_error); return; } bioq_insert_tail(&queue, cbp); cbp->bio_done = g_mirror_done; cbp->bio_caller1 = disk; cbp->bio_to = disk->d_consumer->provider; cbp->bio_offset = offset; cbp->bio_data = data; cbp->bio_length = MIN(left, slice); left -= cbp->bio_length; if (left == 0) break; offset += cbp->bio_length; data += cbp->bio_length; } while ((cbp = bioq_takefirst(&queue)) != NULL) { G_MIRROR_LOGREQ(3, cbp, "Sending request."); disk = cbp->bio_caller1; cbp->bio_caller1 = NULL; cp = disk->d_consumer; KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, cp->acw, cp->ace)); disk->d_consumer->index++; g_io_request(cbp, disk->d_consumer); } } static void g_mirror_register_request(struct bio *bp) { struct g_mirror_softc *sc; sc = bp->bio_to->private; switch (bp->bio_cmd) { case BIO_READ: switch (sc->sc_balance) { case G_MIRROR_BALANCE_LOAD: g_mirror_request_load(sc, bp); break; case G_MIRROR_BALANCE_PREFER: g_mirror_request_prefer(sc, bp); break; case G_MIRROR_BALANCE_ROUND_ROBIN: g_mirror_request_round_robin(sc, bp); break; case G_MIRROR_BALANCE_SPLIT: g_mirror_request_split(sc, bp); break; } return; case BIO_WRITE: case BIO_DELETE: { struct g_mirror_disk *disk; struct g_mirror_disk_sync *sync; struct bio_queue_head queue; struct g_consumer *cp; struct bio *cbp; /* * Delay the request if it is colliding with a synchronization * request. */ if (g_mirror_sync_collision(sc, bp)) { g_mirror_regular_delay(sc, bp); return; } if (sc->sc_idle) g_mirror_unidle(sc); else sc->sc_last_write = time_uptime; /* * Allocate all bios before sending any request, so we can * return ENOMEM in nice and clean way. */ bioq_init(&queue); LIST_FOREACH(disk, &sc->sc_disks, d_next) { sync = &disk->d_sync; switch (disk->d_state) { case G_MIRROR_DISK_STATE_ACTIVE: break; case G_MIRROR_DISK_STATE_SYNCHRONIZING: if (bp->bio_offset >= sync->ds_offset) continue; break; default: continue; } if (bp->bio_cmd == BIO_DELETE && (disk->d_flags & G_MIRROR_DISK_FLAG_CANDELETE) == 0) continue; cbp = g_clone_bio(bp); if (cbp == NULL) { while ((cbp = bioq_takefirst(&queue)) != NULL) g_destroy_bio(cbp); if (bp->bio_error == 0) bp->bio_error = ENOMEM; g_io_deliver(bp, bp->bio_error); return; } bioq_insert_tail(&queue, cbp); cbp->bio_done = g_mirror_done; cp = disk->d_consumer; cbp->bio_caller1 = cp; cbp->bio_to = cp->provider; KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, cp->acw, cp->ace)); } if (bioq_first(&queue) == NULL) { g_io_deliver(bp, EOPNOTSUPP); return; } while ((cbp = bioq_takefirst(&queue)) != NULL) { G_MIRROR_LOGREQ(3, cbp, "Sending request."); cp = cbp->bio_caller1; cbp->bio_caller1 = NULL; cp->index++; sc->sc_writes++; g_io_request(cbp, cp); } /* * Put request onto inflight queue, so we can check if new * synchronization requests don't collide with it. */ bioq_insert_tail(&sc->sc_inflight, bp); /* * Bump syncid on first write. */ if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) { sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; g_mirror_bump_syncid(sc); } return; } default: KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", bp->bio_cmd, sc->sc_name)); break; } } static int g_mirror_can_destroy(struct g_mirror_softc *sc) { struct g_geom *gp; struct g_consumer *cp; g_topology_assert(); gp = sc->sc_geom; if (gp->softc == NULL) return (1); if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_TASTING) != 0) return (0); LIST_FOREACH(cp, &gp->consumer, consumer) { if (g_mirror_is_busy(sc, cp)) return (0); } gp = sc->sc_sync.ds_geom; LIST_FOREACH(cp, &gp->consumer, consumer) { if (g_mirror_is_busy(sc, cp)) return (0); } G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.", sc->sc_name); return (1); } static int g_mirror_try_destroy(struct g_mirror_softc *sc) { if (sc->sc_rootmount != NULL) { G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, sc->sc_rootmount); root_mount_rel(sc->sc_rootmount); sc->sc_rootmount = NULL; } g_topology_lock(); if (!g_mirror_can_destroy(sc)) { g_topology_unlock(); return (0); } sc->sc_geom->softc = NULL; sc->sc_sync.ds_geom->softc = NULL; if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WAIT) != 0) { g_topology_unlock(); G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, &sc->sc_worker); /* Unlock sc_lock here, as it can be destroyed after wakeup. */ sx_xunlock(&sc->sc_lock); wakeup(&sc->sc_worker); sc->sc_worker = NULL; } else { g_topology_unlock(); g_mirror_destroy_device(sc); } return (1); } /* * Worker thread. */ static void g_mirror_worker(void *arg) { struct g_mirror_softc *sc; struct g_mirror_event *ep; struct bio *bp; int timeout; sc = arg; thread_lock(curthread); sched_prio(curthread, PRIBIO); thread_unlock(curthread); sx_xlock(&sc->sc_lock); for (;;) { G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); /* * First take a look at events. * This is important to handle events before any I/O requests. */ ep = g_mirror_event_get(sc); if (ep != NULL) { g_mirror_event_remove(sc, ep); if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { /* Update only device status. */ G_MIRROR_DEBUG(3, "Running event for device %s.", sc->sc_name); ep->e_error = 0; g_mirror_update_device(sc, true); } else { /* Update disk status. */ G_MIRROR_DEBUG(3, "Running event for disk %s.", g_mirror_get_diskname(ep->e_disk)); ep->e_error = g_mirror_update_disk(ep->e_disk, ep->e_state); if (ep->e_error == 0) g_mirror_update_device(sc, false); } if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { KASSERT(ep->e_error == 0, ("Error cannot be handled.")); g_mirror_event_free(ep); } else { ep->e_flags |= G_MIRROR_EVENT_DONE; G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, ep); mtx_lock(&sc->sc_events_mtx); wakeup(ep); mtx_unlock(&sc->sc_events_mtx); } if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { if (g_mirror_try_destroy(sc)) { curthread->td_pflags &= ~TDP_GEOM; G_MIRROR_DEBUG(1, "Thread exiting."); kproc_exit(0); } } G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); continue; } /* * Check if we can mark array as CLEAN and if we can't take * how much seconds should we wait. */ timeout = g_mirror_idle(sc, -1); /* * Now I/O requests. */ /* Get first request from the queue. */ mtx_lock(&sc->sc_queue_mtx); bp = bioq_takefirst(&sc->sc_queue); if (bp == NULL) { if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { mtx_unlock(&sc->sc_queue_mtx); if (g_mirror_try_destroy(sc)) { curthread->td_pflags &= ~TDP_GEOM; G_MIRROR_DEBUG(1, "Thread exiting."); kproc_exit(0); } mtx_lock(&sc->sc_queue_mtx); if (bioq_first(&sc->sc_queue) != NULL) { mtx_unlock(&sc->sc_queue_mtx); continue; } } sx_xunlock(&sc->sc_lock); /* * XXX: We can miss an event here, because an event * can be added without sx-device-lock and without * mtx-queue-lock. Maybe I should just stop using * dedicated mutex for events synchronization and * stick with the queue lock? * The event will hang here until next I/O request * or next event is received. */ MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w1", timeout * hz); sx_xlock(&sc->sc_lock); G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); continue; } mtx_unlock(&sc->sc_queue_mtx); if (bp->bio_from->geom == sc->sc_sync.ds_geom && (bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { g_mirror_sync_request(bp); /* READ */ } else if (bp->bio_to != sc->sc_provider) { if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) g_mirror_regular_request(bp); else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) g_mirror_sync_request(bp); /* WRITE */ else { KASSERT(0, ("Invalid request cflags=0x%hx to=%s.", bp->bio_cflags, bp->bio_to->name)); } } else { g_mirror_register_request(bp); } G_MIRROR_DEBUG(5, "%s: I'm here 9.", __func__); } } static void g_mirror_update_idle(struct g_mirror_softc *sc, struct g_mirror_disk *disk) { sx_assert(&sc->sc_lock, SX_LOCKED); if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) return; if (!sc->sc_idle && (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", g_mirror_get_diskname(disk), sc->sc_name); disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; } else if (sc->sc_idle && (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", g_mirror_get_diskname(disk), sc->sc_name); disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; } } static void g_mirror_sync_start(struct g_mirror_disk *disk) { struct g_mirror_softc *sc; struct g_consumer *cp; struct bio *bp; int error, i; g_topology_assert_not(); sc = disk->d_softc; sx_assert(&sc->sc_lock, SX_LOCKED); KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, ("Disk %s is not marked for synchronization.", g_mirror_get_diskname(disk))); KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, ("Device not in RUNNING state (%s, %u).", sc->sc_name, sc->sc_state)); sx_xunlock(&sc->sc_lock); g_topology_lock(); cp = g_new_consumer(sc->sc_sync.ds_geom); cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE; error = g_attach(cp, sc->sc_provider); KASSERT(error == 0, ("Cannot attach to %s (error=%d).", sc->sc_name, error)); error = g_access(cp, 1, 0, 0); KASSERT(error == 0, ("Cannot open %s (error=%d).", sc->sc_name, error)); g_topology_unlock(); sx_xlock(&sc->sc_lock); G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, g_mirror_get_diskname(disk)); if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) == 0) disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; KASSERT(disk->d_sync.ds_consumer == NULL, ("Sync consumer already exists (device=%s, disk=%s).", sc->sc_name, g_mirror_get_diskname(disk))); disk->d_sync.ds_consumer = cp; disk->d_sync.ds_consumer->private = disk; disk->d_sync.ds_consumer->index = 0; /* * Allocate memory for synchronization bios and initialize them. */ disk->d_sync.ds_bios = malloc(sizeof(struct bio *) * g_mirror_syncreqs, M_MIRROR, M_WAITOK); for (i = 0; i < g_mirror_syncreqs; i++) { bp = g_alloc_bio(); disk->d_sync.ds_bios[i] = bp; bp->bio_parent = NULL; bp->bio_cmd = BIO_READ; bp->bio_data = malloc(MAXPHYS, M_MIRROR, M_WAITOK); bp->bio_cflags = 0; bp->bio_offset = disk->d_sync.ds_offset; bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); disk->d_sync.ds_offset += bp->bio_length; bp->bio_done = g_mirror_sync_done; bp->bio_from = disk->d_sync.ds_consumer; bp->bio_to = sc->sc_provider; bp->bio_caller1 = (void *)(uintptr_t)i; } /* Increase the number of disks in SYNCHRONIZING state. */ sc->sc_sync.ds_ndisks++; /* Set the number of in-flight synchronization requests. */ disk->d_sync.ds_inflight = g_mirror_syncreqs; /* * Fire off first synchronization requests. */ for (i = 0; i < g_mirror_syncreqs; i++) { bp = disk->d_sync.ds_bios[i]; G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); disk->d_sync.ds_consumer->index++; /* * Delay the request if it is colliding with a regular request. */ if (g_mirror_regular_collision(sc, bp)) g_mirror_sync_delay(sc, bp); else g_io_request(bp, disk->d_sync.ds_consumer); } } /* * Stop synchronization process. * type: 0 - synchronization finished * 1 - synchronization stopped */ static void g_mirror_sync_stop(struct g_mirror_disk *disk, int type) { struct g_mirror_softc *sc; struct g_consumer *cp; g_topology_assert_not(); sc = disk->d_softc; sx_assert(&sc->sc_lock, SX_LOCKED); KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state))); if (disk->d_sync.ds_consumer == NULL) return; if (type == 0) { G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", sc->sc_name, g_mirror_get_diskname(disk)); } else /* if (type == 1) */ { G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", sc->sc_name, g_mirror_get_diskname(disk)); } free(disk->d_sync.ds_bios, M_MIRROR); disk->d_sync.ds_bios = NULL; cp = disk->d_sync.ds_consumer; disk->d_sync.ds_consumer = NULL; disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; sc->sc_sync.ds_ndisks--; sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ g_topology_lock(); g_mirror_kill_consumer(sc, cp); g_topology_unlock(); sx_xlock(&sc->sc_lock); } static void g_mirror_launch_provider(struct g_mirror_softc *sc) { struct g_mirror_disk *disk; struct g_provider *pp, *dp; sx_assert(&sc->sc_lock, SX_LOCKED); g_topology_lock(); pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); pp->flags |= G_PF_DIRECT_RECEIVE; pp->mediasize = sc->sc_mediasize; pp->sectorsize = sc->sc_sectorsize; pp->stripesize = 0; pp->stripeoffset = 0; /* Splitting of unmapped BIO's could work but isn't implemented now */ if (sc->sc_balance != G_MIRROR_BALANCE_SPLIT) pp->flags |= G_PF_ACCEPT_UNMAPPED; LIST_FOREACH(disk, &sc->sc_disks, d_next) { if (disk->d_consumer && disk->d_consumer->provider) { dp = disk->d_consumer->provider; if (dp->stripesize > pp->stripesize) { pp->stripesize = dp->stripesize; pp->stripeoffset = dp->stripeoffset; } /* A provider underneath us doesn't support unmapped */ if ((dp->flags & G_PF_ACCEPT_UNMAPPED) == 0) { G_MIRROR_DEBUG(0, "Cancelling unmapped " "because of %s.", dp->name); pp->flags &= ~G_PF_ACCEPT_UNMAPPED; } } } pp->private = sc; sc->sc_refcnt++; sc->sc_provider = pp; g_error_provider(pp, 0); g_topology_unlock(); G_MIRROR_DEBUG(0, "Device %s launched (%u/%u).", pp->name, g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE), sc->sc_ndisks); LIST_FOREACH(disk, &sc->sc_disks, d_next) { if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) g_mirror_sync_start(disk); } } static void g_mirror_destroy_provider(struct g_mirror_softc *sc) { struct g_mirror_disk *disk; struct bio *bp; g_topology_assert_not(); KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", sc->sc_name)); g_topology_lock(); g_error_provider(sc->sc_provider, ENXIO); mtx_lock(&sc->sc_queue_mtx); while ((bp = bioq_takefirst(&sc->sc_queue)) != NULL) { /* * Abort any pending I/O that wasn't generated by us. * Synchronization requests and requests destined for individual * mirror components can be destroyed immediately. */ if (bp->bio_to == sc->sc_provider && bp->bio_from->geom != sc->sc_sync.ds_geom) { g_io_deliver(bp, ENXIO); } else { if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) free(bp->bio_data, M_MIRROR); g_destroy_bio(bp); } } mtx_unlock(&sc->sc_queue_mtx); g_wither_provider(sc->sc_provider, ENXIO); sc->sc_provider = NULL; G_MIRROR_DEBUG(0, "Device %s: provider destroyed.", sc->sc_name); g_topology_unlock(); LIST_FOREACH(disk, &sc->sc_disks, d_next) { if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) g_mirror_sync_stop(disk, 1); } } static void g_mirror_go(void *arg) { struct g_mirror_softc *sc; sc = arg; G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); g_mirror_event_send(sc, 0, G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); } static u_int g_mirror_determine_state(struct g_mirror_disk *disk) { struct g_mirror_softc *sc; u_int state; sc = disk->d_softc; if (sc->sc_syncid == disk->d_sync.ds_syncid) { if ((disk->d_flags & G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { /* Disk does not need synchronization. */ state = G_MIRROR_DISK_STATE_ACTIVE; } else { if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { /* * We can start synchronization from * the stored offset. */ state = G_MIRROR_DISK_STATE_SYNCHRONIZING; } else { state = G_MIRROR_DISK_STATE_STALE; } } } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { /* * Reset all synchronization data for this disk, * because if it even was synchronized, it was * synchronized to disks with different syncid. */ disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; disk->d_sync.ds_offset = 0; disk->d_sync.ds_offset_done = 0; disk->d_sync.ds_syncid = sc->sc_syncid; if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { state = G_MIRROR_DISK_STATE_SYNCHRONIZING; } else { state = G_MIRROR_DISK_STATE_STALE; } } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { /* * Not good, NOT GOOD! * It means that mirror was started on stale disks * and more fresh disk just arrive. * If there were writes, mirror is broken, sorry. * I think the best choice here is don't touch * this disk and inform the user loudly. */ G_MIRROR_DEBUG(0, "Device %s was started before the freshest " "disk (%s) arrives!! It will not be connected to the " "running device.", sc->sc_name, g_mirror_get_diskname(disk)); g_mirror_destroy_disk(disk); state = G_MIRROR_DISK_STATE_NONE; /* Return immediately, because disk was destroyed. */ return (state); } G_MIRROR_DEBUG(3, "State for %s disk: %s.", g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); return (state); } /* * Update device state. */ static void g_mirror_update_device(struct g_mirror_softc *sc, bool force) { struct g_mirror_disk *disk; u_int state; sx_assert(&sc->sc_lock, SX_XLOCKED); switch (sc->sc_state) { case G_MIRROR_DEVICE_STATE_STARTING: { struct g_mirror_disk *pdisk, *tdisk; u_int dirty, ndisks, genid, syncid; KASSERT(sc->sc_provider == NULL, ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); /* * Are we ready? We are, if all disks are connected or * if we have any disks and 'force' is true. */ ndisks = g_mirror_ndisks(sc, -1); if (sc->sc_ndisks == ndisks || (force && ndisks > 0)) { ; } else if (ndisks == 0) { /* * Disks went down in starting phase, so destroy * device. */ callout_drain(&sc->sc_callout); sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, sc->sc_rootmount); root_mount_rel(sc->sc_rootmount); sc->sc_rootmount = NULL; return; } else { return; } /* * Activate all disks with the biggest syncid. */ if (force) { /* * If 'force' is true, we have been called due to * timeout, so don't bother canceling timeout. */ ndisks = 0; LIST_FOREACH(disk, &sc->sc_disks, d_next) { if ((disk->d_flags & G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { ndisks++; } } if (ndisks == 0) { /* No valid disks found, destroy device. */ sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, sc->sc_rootmount); root_mount_rel(sc->sc_rootmount); sc->sc_rootmount = NULL; return; } } else { /* Cancel timeout. */ callout_drain(&sc->sc_callout); } /* * Find the biggest genid. */ genid = 0; LIST_FOREACH(disk, &sc->sc_disks, d_next) { if (disk->d_genid > genid) genid = disk->d_genid; } sc->sc_genid = genid; /* * Remove all disks without the biggest genid. */ LIST_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tdisk) { if (disk->d_genid < genid) { G_MIRROR_DEBUG(0, "Component %s (device %s) broken, skipping.", g_mirror_get_diskname(disk), sc->sc_name); g_mirror_destroy_disk(disk); } } /* * Find the biggest syncid. */ syncid = 0; LIST_FOREACH(disk, &sc->sc_disks, d_next) { if (disk->d_sync.ds_syncid > syncid) syncid = disk->d_sync.ds_syncid; } /* * Here we need to look for dirty disks and if all disks * with the biggest syncid are dirty, we have to choose * one with the biggest priority and rebuild the rest. */ /* * Find the number of dirty disks with the biggest syncid. * Find the number of disks with the biggest syncid. * While here, find a disk with the biggest priority. */ dirty = ndisks = 0; pdisk = NULL; LIST_FOREACH(disk, &sc->sc_disks, d_next) { if (disk->d_sync.ds_syncid != syncid) continue; if ((disk->d_flags & G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { continue; } ndisks++; if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { dirty++; if (pdisk == NULL || pdisk->d_priority < disk->d_priority) { pdisk = disk; } } } if (dirty == 0) { /* No dirty disks at all, great. */ } else if (dirty == ndisks) { /* * Force synchronization for all dirty disks except one * with the biggest priority. */ KASSERT(pdisk != NULL, ("pdisk == NULL")); G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " "master disk for synchronization.", g_mirror_get_diskname(pdisk), sc->sc_name); LIST_FOREACH(disk, &sc->sc_disks, d_next) { if (disk->d_sync.ds_syncid != syncid) continue; if ((disk->d_flags & G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { continue; } KASSERT((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0, ("Disk %s isn't marked as dirty.", g_mirror_get_diskname(disk))); /* Skip the disk with the biggest priority. */ if (disk == pdisk) continue; disk->d_sync.ds_syncid = 0; } } else if (dirty < ndisks) { /* * Force synchronization for all dirty disks. * We have some non-dirty disks. */ LIST_FOREACH(disk, &sc->sc_disks, d_next) { if (disk->d_sync.ds_syncid != syncid) continue; if ((disk->d_flags & G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { continue; } if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { continue; } disk->d_sync.ds_syncid = 0; } } /* Reset hint. */ sc->sc_hint = NULL; sc->sc_syncid = syncid; if (force) { /* Remember to bump syncid on first write. */ sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; } state = G_MIRROR_DEVICE_STATE_RUNNING; G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", sc->sc_name, g_mirror_device_state2str(sc->sc_state), g_mirror_device_state2str(state)); sc->sc_state = state; LIST_FOREACH(disk, &sc->sc_disks, d_next) { state = g_mirror_determine_state(disk); g_mirror_event_send(disk, state, G_MIRROR_EVENT_DONTWAIT); if (state == G_MIRROR_DISK_STATE_STALE) sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; } break; } case G_MIRROR_DEVICE_STATE_RUNNING: if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 && g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { /* * No active disks or no disks at all, * so destroy device. */ if (sc->sc_provider != NULL) g_mirror_destroy_provider(sc); sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; break; } else if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0 && g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { /* * We have active disks, launch provider if it doesn't * exist. */ if (sc->sc_provider == NULL) g_mirror_launch_provider(sc); if (sc->sc_rootmount != NULL) { G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, sc->sc_rootmount); root_mount_rel(sc->sc_rootmount); sc->sc_rootmount = NULL; } } /* * Genid should be bumped immediately, so do it here. */ if ((sc->sc_bump_id & G_MIRROR_BUMP_GENID) != 0) { sc->sc_bump_id &= ~G_MIRROR_BUMP_GENID; g_mirror_bump_genid(sc); } break; default: KASSERT(1 == 0, ("Wrong device state (%s, %s).", sc->sc_name, g_mirror_device_state2str(sc->sc_state))); break; } } /* * Update disk state and device state if needed. */ #define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ "Disk %s state changed from %s to %s (device %s).", \ g_mirror_get_diskname(disk), \ g_mirror_disk_state2str(disk->d_state), \ g_mirror_disk_state2str(state), sc->sc_name) static int g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) { struct g_mirror_softc *sc; sc = disk->d_softc; sx_assert(&sc->sc_lock, SX_XLOCKED); again: G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), g_mirror_disk_state2str(state)); switch (state) { case G_MIRROR_DISK_STATE_NEW: /* * Possible scenarios: * 1. New disk arrive. */ /* Previous state should be NONE. */ KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state))); DISK_STATE_CHANGED(); disk->d_state = state; if (LIST_EMPTY(&sc->sc_disks)) LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); else { struct g_mirror_disk *dp; LIST_FOREACH(dp, &sc->sc_disks, d_next) { if (disk->d_priority >= dp->d_priority) { LIST_INSERT_BEFORE(dp, disk, d_next); dp = NULL; break; } if (LIST_NEXT(dp, d_next) == NULL) break; } if (dp != NULL) LIST_INSERT_AFTER(dp, disk, d_next); } G_MIRROR_DEBUG(1, "Device %s: provider %s detected.", sc->sc_name, g_mirror_get_diskname(disk)); if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) break; KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, g_mirror_device_state2str(sc->sc_state), g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state))); state = g_mirror_determine_state(disk); if (state != G_MIRROR_DISK_STATE_NONE) goto again; break; case G_MIRROR_DISK_STATE_ACTIVE: /* * Possible scenarios: * 1. New disk does not need synchronization. * 2. Synchronization process finished successfully. */ KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, g_mirror_device_state2str(sc->sc_state), g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state))); /* Previous state should be NEW or SYNCHRONIZING. */ KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state))); DISK_STATE_CHANGED(); if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; g_mirror_sync_stop(disk, 0); } disk->d_state = state; disk->d_sync.ds_offset = 0; disk->d_sync.ds_offset_done = 0; g_mirror_update_idle(sc, disk); g_mirror_update_metadata(disk); G_MIRROR_DEBUG(1, "Device %s: provider %s activated.", sc->sc_name, g_mirror_get_diskname(disk)); break; case G_MIRROR_DISK_STATE_STALE: /* * Possible scenarios: * 1. Stale disk was connected. */ /* Previous state should be NEW. */ KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state))); KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, g_mirror_device_state2str(sc->sc_state), g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state))); /* * STALE state is only possible if device is marked * NOAUTOSYNC. */ KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, g_mirror_device_state2str(sc->sc_state), g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state))); DISK_STATE_CHANGED(); disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; disk->d_state = state; g_mirror_update_metadata(disk); G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", sc->sc_name, g_mirror_get_diskname(disk)); break; case G_MIRROR_DISK_STATE_SYNCHRONIZING: /* * Possible scenarios: * 1. Disk which needs synchronization was connected. */ /* Previous state should be NEW. */ KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state))); KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, g_mirror_device_state2str(sc->sc_state), g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state))); DISK_STATE_CHANGED(); if (disk->d_state == G_MIRROR_DISK_STATE_NEW) disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; disk->d_state = state; if (sc->sc_provider != NULL) { g_mirror_sync_start(disk); g_mirror_update_metadata(disk); } break; case G_MIRROR_DISK_STATE_DISCONNECTED: /* * Possible scenarios: * 1. Device wasn't running yet, but disk disappear. * 2. Disk was active and disapppear. * 3. Disk disappear during synchronization process. */ if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { /* * Previous state should be ACTIVE, STALE or * SYNCHRONIZING. */ KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || disk->d_state == G_MIRROR_DISK_STATE_STALE || disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state))); } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { /* Previous state should be NEW. */ KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state))); /* * Reset bumping syncid if disk disappeared in STARTING * state. */ if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; #ifdef INVARIANTS } else { KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, g_mirror_device_state2str(sc->sc_state), g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state))); #endif } DISK_STATE_CHANGED(); G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", sc->sc_name, g_mirror_get_diskname(disk)); g_mirror_destroy_disk(disk); break; case G_MIRROR_DISK_STATE_DESTROY: { int error; error = g_mirror_clear_metadata(disk); if (error != 0) { G_MIRROR_DEBUG(0, "Device %s: failed to clear metadata on %s: %d.", sc->sc_name, g_mirror_get_diskname(disk), error); break; } DISK_STATE_CHANGED(); G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", sc->sc_name, g_mirror_get_diskname(disk)); g_mirror_destroy_disk(disk); sc->sc_ndisks--; LIST_FOREACH(disk, &sc->sc_disks, d_next) { g_mirror_update_metadata(disk); } break; } default: KASSERT(1 == 0, ("Unknown state (%u).", state)); break; } return (0); } #undef DISK_STATE_CHANGED int g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) { struct g_provider *pp; u_char *buf; int error; g_topology_assert(); error = g_access(cp, 1, 0, 0); if (error != 0) return (error); pp = cp->provider; g_topology_unlock(); /* Metadata are stored on last sector. */ buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, &error); g_topology_lock(); g_access(cp, -1, 0, 0); if (buf == NULL) { G_MIRROR_DEBUG(1, "Cannot read metadata from %s (error=%d).", cp->provider->name, error); return (error); } /* Decode metadata. */ error = mirror_metadata_decode(buf, md); g_free(buf); if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) return (EINVAL); if (md->md_version > G_MIRROR_VERSION) { G_MIRROR_DEBUG(0, "Kernel module is too old to handle metadata from %s.", cp->provider->name); return (EINVAL); } if (error != 0) { G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", cp->provider->name); return (error); } return (0); } static int g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, struct g_mirror_metadata *md) { if (g_mirror_id2disk(sc, md->md_did) != NULL) { G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", pp->name, md->md_did); return (EEXIST); } if (md->md_all != sc->sc_ndisks) { G_MIRROR_DEBUG(1, "Invalid '%s' field on disk %s (device %s), skipping.", "md_all", pp->name, sc->sc_name); return (EINVAL); } if (md->md_slice != sc->sc_slice) { G_MIRROR_DEBUG(1, "Invalid '%s' field on disk %s (device %s), skipping.", "md_slice", pp->name, sc->sc_name); return (EINVAL); } if (md->md_balance != sc->sc_balance) { G_MIRROR_DEBUG(1, "Invalid '%s' field on disk %s (device %s), skipping.", "md_balance", pp->name, sc->sc_name); return (EINVAL); } #if 0 if (md->md_mediasize != sc->sc_mediasize) { G_MIRROR_DEBUG(1, "Invalid '%s' field on disk %s (device %s), skipping.", "md_mediasize", pp->name, sc->sc_name); return (EINVAL); } #endif if (sc->sc_mediasize > pp->mediasize) { G_MIRROR_DEBUG(1, "Invalid size of disk %s (device %s), skipping.", pp->name, sc->sc_name); return (EINVAL); } if (md->md_sectorsize != sc->sc_sectorsize) { G_MIRROR_DEBUG(1, "Invalid '%s' field on disk %s (device %s), skipping.", "md_sectorsize", pp->name, sc->sc_name); return (EINVAL); } if ((sc->sc_sectorsize % pp->sectorsize) != 0) { G_MIRROR_DEBUG(1, "Invalid sector size of disk %s (device %s), skipping.", pp->name, sc->sc_name); return (EINVAL); } if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { G_MIRROR_DEBUG(1, "Invalid device flags on disk %s (device %s), skipping.", pp->name, sc->sc_name); return (EINVAL); } if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { G_MIRROR_DEBUG(1, "Invalid disk flags on disk %s (device %s), skipping.", pp->name, sc->sc_name); return (EINVAL); } return (0); } int g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, struct g_mirror_metadata *md) { struct g_mirror_disk *disk; int error; g_topology_assert_not(); G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); error = g_mirror_check_metadata(sc, pp, md); if (error != 0) return (error); if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING && md->md_genid < sc->sc_genid) { G_MIRROR_DEBUG(0, "Component %s (device %s) broken, skipping.", pp->name, sc->sc_name); return (EINVAL); } disk = g_mirror_init_disk(sc, pp, md, &error); if (disk == NULL) return (error); error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, G_MIRROR_EVENT_WAIT); if (error != 0) return (error); if (md->md_version < G_MIRROR_VERSION) { G_MIRROR_DEBUG(0, "Upgrading metadata on %s (v%d->v%d).", pp->name, md->md_version, G_MIRROR_VERSION); g_mirror_update_metadata(disk); } return (0); } static void g_mirror_destroy_delayed(void *arg, int flag) { struct g_mirror_softc *sc; int error; if (flag == EV_CANCEL) { G_MIRROR_DEBUG(1, "Destroying canceled."); return; } sc = arg; g_topology_unlock(); sx_xlock(&sc->sc_lock); KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) == 0, ("DESTROY flag set on %s.", sc->sc_name)); KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROYING) != 0, ("DESTROYING flag not set on %s.", sc->sc_name)); G_MIRROR_DEBUG(1, "Destroying %s (delayed).", sc->sc_name); error = g_mirror_destroy(sc, G_MIRROR_DESTROY_SOFT); if (error != 0) { G_MIRROR_DEBUG(0, "Cannot destroy %s (error=%d).", sc->sc_name, error); sx_xunlock(&sc->sc_lock); } g_topology_lock(); } static int g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) { struct g_mirror_softc *sc; int error = 0; g_topology_assert(); G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, acw, ace); sc = pp->private; KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name)); g_topology_unlock(); sx_xlock(&sc->sc_lock); if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0 || (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROYING) != 0 || LIST_EMPTY(&sc->sc_disks)) { if (acr > 0 || acw > 0 || ace > 0) error = ENXIO; goto end; } sc->sc_provider_open += acr + acw + ace; if (pp->acw + acw == 0) g_mirror_idle(sc, 0); if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROYING) != 0 && sc->sc_provider_open == 0) g_post_event(g_mirror_destroy_delayed, sc, M_WAITOK, sc, NULL); end: sx_xunlock(&sc->sc_lock); g_topology_lock(); return (error); } -static struct g_geom * -g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md) +struct g_geom * +g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md, + u_int type) { struct g_mirror_softc *sc; struct g_geom *gp; int error, timeout; g_topology_assert(); G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, md->md_mid); /* One disk is minimum. */ if (md->md_all < 1) return (NULL); /* * Action geom. */ gp = g_new_geomf(mp, "%s", md->md_name); sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); gp->start = g_mirror_start; gp->orphan = g_mirror_orphan; gp->access = g_mirror_access; gp->dumpconf = g_mirror_dumpconf; + sc->sc_type = type; sc->sc_id = md->md_mid; sc->sc_slice = md->md_slice; sc->sc_balance = md->md_balance; sc->sc_mediasize = md->md_mediasize; sc->sc_sectorsize = md->md_sectorsize; sc->sc_ndisks = md->md_all; sc->sc_flags = md->md_mflags; sc->sc_bump_id = 0; sc->sc_idle = 1; sc->sc_last_write = time_uptime; sc->sc_writes = 0; sc->sc_refcnt = 1; sx_init(&sc->sc_lock, "gmirror:lock"); bioq_init(&sc->sc_queue); mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); bioq_init(&sc->sc_regular_delayed); bioq_init(&sc->sc_inflight); bioq_init(&sc->sc_sync_delayed); LIST_INIT(&sc->sc_disks); TAILQ_INIT(&sc->sc_events); mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); callout_init(&sc->sc_callout, 1); mtx_init(&sc->sc_done_mtx, "gmirror:done", NULL, MTX_DEF); sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; gp->softc = sc; sc->sc_geom = gp; sc->sc_provider = NULL; sc->sc_provider_open = 0; /* * Synchronization geom. */ gp = g_new_geomf(mp, "%s.sync", md->md_name); gp->softc = sc; gp->orphan = g_mirror_orphan; sc->sc_sync.ds_geom = gp; sc->sc_sync.ds_ndisks = 0; error = kproc_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, "g_mirror %s", md->md_name); if (error != 0) { G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", sc->sc_name); g_destroy_geom(sc->sc_sync.ds_geom); g_destroy_geom(sc->sc_geom); g_mirror_free_device(sc); return (NULL); } G_MIRROR_DEBUG(1, "Device %s created (%u components, id=%u).", sc->sc_name, sc->sc_ndisks, sc->sc_id); sc->sc_rootmount = root_mount_hold("GMIRROR"); G_MIRROR_DEBUG(1, "root_mount_hold %p", sc->sc_rootmount); /* * Run timeout. */ timeout = g_mirror_timeout * hz; callout_reset(&sc->sc_callout, timeout, g_mirror_go, sc); return (sc->sc_geom); } int g_mirror_destroy(struct g_mirror_softc *sc, int how) { struct g_mirror_disk *disk; g_topology_assert_not(); sx_assert(&sc->sc_lock, SX_XLOCKED); if (sc->sc_provider_open != 0) { switch (how) { case G_MIRROR_DESTROY_SOFT: G_MIRROR_DEBUG(1, "Device %s is still open (%d).", sc->sc_name, sc->sc_provider_open); return (EBUSY); case G_MIRROR_DESTROY_DELAYED: G_MIRROR_DEBUG(1, "Device %s will be destroyed on last close.", sc->sc_name); LIST_FOREACH(disk, &sc->sc_disks, d_next) { if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { g_mirror_sync_stop(disk, 1); } } sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROYING; return (EBUSY); case G_MIRROR_DESTROY_HARD: G_MIRROR_DEBUG(1, "Device %s is still open, so it " "can't be definitely removed.", sc->sc_name); } } g_topology_lock(); if (sc->sc_geom->softc == NULL) { g_topology_unlock(); return (0); } sc->sc_geom->softc = NULL; sc->sc_sync.ds_geom->softc = NULL; g_topology_unlock(); sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; sc->sc_flags |= G_MIRROR_DEVICE_FLAG_WAIT; G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); sx_xunlock(&sc->sc_lock); mtx_lock(&sc->sc_queue_mtx); wakeup(sc); mtx_unlock(&sc->sc_queue_mtx); G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); while (sc->sc_worker != NULL) tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); sx_xlock(&sc->sc_lock); g_mirror_destroy_device(sc); return (0); } static void g_mirror_taste_orphan(struct g_consumer *cp) { KASSERT(1 == 0, ("%s called while tasting %s.", __func__, cp->provider->name)); } static struct g_geom * g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) { struct g_mirror_metadata md; struct g_mirror_softc *sc; struct g_consumer *cp; struct g_geom *gp; int error; g_topology_assert(); g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); gp = g_new_geomf(mp, "mirror:taste"); /* * This orphan function should be never called. */ gp->orphan = g_mirror_taste_orphan; cp = g_new_consumer(gp); g_attach(cp, pp); error = g_mirror_read_metadata(cp, &md); g_detach(cp); g_destroy_consumer(cp); g_destroy_geom(gp); if (error != 0) return (NULL); gp = NULL; if (md.md_provider[0] != '\0' && !g_compare_names(md.md_provider, pp->name)) return (NULL); if (md.md_provsize != 0 && md.md_provsize != pp->mediasize) return (NULL); if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { G_MIRROR_DEBUG(0, "Device %s: provider %s marked as inactive, skipping.", md.md_name, pp->name); return (NULL); } if (g_mirror_debug >= 2) mirror_metadata_dump(&md); /* * Let's check if device already exists. */ sc = NULL; LIST_FOREACH(gp, &mp->geom, geom) { sc = gp->softc; if (sc == NULL) continue; + if (sc->sc_type != G_MIRROR_TYPE_AUTOMATIC) + continue; if (sc->sc_sync.ds_geom == gp) continue; if (strcmp(md.md_name, sc->sc_name) != 0) continue; if (md.md_mid != sc->sc_id) { G_MIRROR_DEBUG(0, "Device %s already configured.", sc->sc_name); return (NULL); } break; } if (gp == NULL) { - gp = g_mirror_create(mp, &md); + gp = g_mirror_create(mp, &md, G_MIRROR_TYPE_AUTOMATIC); if (gp == NULL) { G_MIRROR_DEBUG(0, "Cannot create device %s.", md.md_name); return (NULL); } sc = gp->softc; } G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); g_topology_unlock(); sx_xlock(&sc->sc_lock); sc->sc_flags |= G_MIRROR_DEVICE_FLAG_TASTING; error = g_mirror_add_disk(sc, pp, &md); if (error != 0) { G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", pp->name, gp->name, error); if (LIST_EMPTY(&sc->sc_disks)) { g_cancel_event(sc); g_mirror_destroy(sc, G_MIRROR_DESTROY_HARD); g_topology_lock(); return (NULL); } gp = NULL; } sc->sc_flags &= ~G_MIRROR_DEVICE_FLAG_TASTING; if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { g_mirror_destroy(sc, G_MIRROR_DESTROY_HARD); g_topology_lock(); return (NULL); } sx_xunlock(&sc->sc_lock); g_topology_lock(); return (gp); } static void g_mirror_resize(struct g_consumer *cp) { struct g_mirror_disk *disk; g_topology_assert(); g_trace(G_T_TOPOLOGY, "%s(%s)", __func__, cp->provider->name); disk = cp->private; if (disk == NULL) return; g_topology_unlock(); g_mirror_update_metadata(disk); g_topology_lock(); } static int g_mirror_destroy_geom(struct gctl_req *req __unused, struct g_class *mp __unused, struct g_geom *gp) { struct g_mirror_softc *sc; int error; g_topology_unlock(); sc = gp->softc; sx_xlock(&sc->sc_lock); g_cancel_event(sc); error = g_mirror_destroy(gp->softc, G_MIRROR_DESTROY_SOFT); if (error != 0) sx_xunlock(&sc->sc_lock); g_topology_lock(); return (error); } static void g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp) { struct g_mirror_softc *sc; g_topology_assert(); sc = gp->softc; if (sc == NULL) return; /* Skip synchronization geom. */ if (gp == sc->sc_sync.ds_geom) return; if (pp != NULL) { /* Nothing here. */ } else if (cp != NULL) { struct g_mirror_disk *disk; disk = cp->private; if (disk == NULL) return; g_topology_unlock(); sx_xlock(&sc->sc_lock); sbuf_printf(sb, "%s%u\n", indent, (u_int)disk->d_id); if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { sbuf_printf(sb, "%s", indent); if (disk->d_sync.ds_offset == 0) sbuf_printf(sb, "0%%"); else { sbuf_printf(sb, "%u%%", (u_int)((disk->d_sync.ds_offset * 100) / sc->sc_provider->mediasize)); } sbuf_printf(sb, "\n"); if (disk->d_sync.ds_offset > 0) { sbuf_printf(sb, "%s%jd" "\n", indent, (intmax_t)disk->d_sync.ds_offset); } } sbuf_printf(sb, "%s%u\n", indent, disk->d_sync.ds_syncid); sbuf_printf(sb, "%s%u\n", indent, disk->d_genid); sbuf_printf(sb, "%s", indent); if (disk->d_flags == 0) sbuf_printf(sb, "NONE"); else { int first = 1; #define ADD_FLAG(flag, name) do { \ if ((disk->d_flags & (flag)) != 0) { \ if (!first) \ sbuf_printf(sb, ", "); \ else \ first = 0; \ sbuf_printf(sb, name); \ } \ } while (0) ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY"); ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED"); ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE"); ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING, "SYNCHRONIZING"); ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); ADD_FLAG(G_MIRROR_DISK_FLAG_BROKEN, "BROKEN"); #undef ADD_FLAG } sbuf_printf(sb, "\n"); sbuf_printf(sb, "%s%u\n", indent, disk->d_priority); sbuf_printf(sb, "%s%s\n", indent, g_mirror_disk_state2str(disk->d_state)); sx_xunlock(&sc->sc_lock); g_topology_lock(); } else { g_topology_unlock(); sx_xlock(&sc->sc_lock); + sbuf_printf(sb, "%s", indent); + switch (sc->sc_type) { + case G_MIRROR_TYPE_AUTOMATIC: + sbuf_printf(sb, "AUTOMATIC"); + break; + case G_MIRROR_TYPE_MANUAL: + sbuf_printf(sb, "MANUAL"); + break; + default: + sbuf_printf(sb, "UNKNOWN"); + break; + } + sbuf_printf(sb, "\n"); sbuf_printf(sb, "%s%u\n", indent, (u_int)sc->sc_id); sbuf_printf(sb, "%s%u\n", indent, sc->sc_syncid); sbuf_printf(sb, "%s%u\n", indent, sc->sc_genid); sbuf_printf(sb, "%s", indent); if (sc->sc_flags == 0) sbuf_printf(sb, "NONE"); else { int first = 1; #define ADD_FLAG(flag, name) do { \ if ((sc->sc_flags & (flag)) != 0) { \ if (!first) \ sbuf_printf(sb, ", "); \ else \ first = 0; \ sbuf_printf(sb, name); \ } \ } while (0) ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOFAILSYNC, "NOFAILSYNC"); ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); #undef ADD_FLAG } sbuf_printf(sb, "\n"); sbuf_printf(sb, "%s%u\n", indent, (u_int)sc->sc_slice); sbuf_printf(sb, "%s%s\n", indent, balance_name(sc->sc_balance)); sbuf_printf(sb, "%s%u\n", indent, sc->sc_ndisks); sbuf_printf(sb, "%s", indent); if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) sbuf_printf(sb, "%s", "STARTING"); else if (sc->sc_ndisks == g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE)) sbuf_printf(sb, "%s", "COMPLETE"); else sbuf_printf(sb, "%s", "DEGRADED"); sbuf_printf(sb, "\n"); sx_xunlock(&sc->sc_lock); g_topology_lock(); } } static void g_mirror_shutdown_post_sync(void *arg, int howto) { struct g_class *mp; struct g_geom *gp, *gp2; struct g_mirror_softc *sc; int error; if (panicstr != NULL) return; mp = arg; g_topology_lock(); g_mirror_shutdown = 1; LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { if ((sc = gp->softc) == NULL) continue; /* Skip synchronization geom. */ if (gp == sc->sc_sync.ds_geom) continue; g_topology_unlock(); sx_xlock(&sc->sc_lock); g_mirror_idle(sc, -1); g_cancel_event(sc); error = g_mirror_destroy(sc, G_MIRROR_DESTROY_DELAYED); if (error != 0) sx_xunlock(&sc->sc_lock); g_topology_lock(); } g_topology_unlock(); } static void g_mirror_init(struct g_class *mp) { g_mirror_post_sync = EVENTHANDLER_REGISTER(shutdown_post_sync, g_mirror_shutdown_post_sync, mp, SHUTDOWN_PRI_FIRST); if (g_mirror_post_sync == NULL) G_MIRROR_DEBUG(0, "Warning! Cannot register shutdown event."); } static void g_mirror_fini(struct g_class *mp) { if (g_mirror_post_sync != NULL) EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_mirror_post_sync); } DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); Index: stable/11/sys/geom/mirror/g_mirror.h =================================================================== --- stable/11/sys/geom/mirror/g_mirror.h (revision 318751) +++ stable/11/sys/geom/mirror/g_mirror.h (revision 318752) @@ -1,500 +1,508 @@ /*- * Copyright (c) 2004-2006 Pawel Jakub Dawidek * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #ifndef _G_MIRROR_H_ #define _G_MIRROR_H_ #include #include #define G_MIRROR_CLASS_NAME "MIRROR" #define G_MIRROR_MAGIC "GEOM::MIRROR" /* * Version history: * 0 - Initial version number. * 1 - Added 'prefer' balance algorithm. * 2 - Added md_genid field to metadata. * 3 - Added md_provsize field to metadata. * 4 - Added 'no failure synchronization' flag. */ #define G_MIRROR_VERSION 4 #define G_MIRROR_BALANCE_NONE 0 #define G_MIRROR_BALANCE_ROUND_ROBIN 1 #define G_MIRROR_BALANCE_LOAD 2 #define G_MIRROR_BALANCE_SPLIT 3 #define G_MIRROR_BALANCE_PREFER 4 #define G_MIRROR_BALANCE_MIN G_MIRROR_BALANCE_NONE #define G_MIRROR_BALANCE_MAX G_MIRROR_BALANCE_PREFER #define G_MIRROR_DISK_FLAG_DIRTY 0x0000000000000001ULL #define G_MIRROR_DISK_FLAG_SYNCHRONIZING 0x0000000000000002ULL #define G_MIRROR_DISK_FLAG_FORCE_SYNC 0x0000000000000004ULL #define G_MIRROR_DISK_FLAG_INACTIVE 0x0000000000000008ULL #define G_MIRROR_DISK_FLAG_HARDCODED 0x0000000000000010ULL #define G_MIRROR_DISK_FLAG_BROKEN 0x0000000000000020ULL #define G_MIRROR_DISK_FLAG_CANDELETE 0x0000000000000040ULL #define G_MIRROR_DISK_FLAG_MASK (G_MIRROR_DISK_FLAG_DIRTY | \ G_MIRROR_DISK_FLAG_SYNCHRONIZING | \ G_MIRROR_DISK_FLAG_FORCE_SYNC | \ G_MIRROR_DISK_FLAG_INACTIVE | \ G_MIRROR_DISK_FLAG_CANDELETE) #define G_MIRROR_DEVICE_FLAG_NOAUTOSYNC 0x0000000000000001ULL #define G_MIRROR_DEVICE_FLAG_NOFAILSYNC 0x0000000000000002ULL #define G_MIRROR_DEVICE_FLAG_MASK (G_MIRROR_DEVICE_FLAG_NOAUTOSYNC | \ G_MIRROR_DEVICE_FLAG_NOFAILSYNC) #ifdef _KERNEL extern u_int g_mirror_debug; #define G_MIRROR_DEBUG(lvl, ...) do { \ if (g_mirror_debug >= (lvl)) { \ printf("GEOM_MIRROR"); \ if (g_mirror_debug > 0) \ printf("[%u]", lvl); \ printf(": "); \ printf(__VA_ARGS__); \ printf("\n"); \ } \ } while (0) #define G_MIRROR_LOGREQ(lvl, bp, ...) do { \ if (g_mirror_debug >= (lvl)) { \ printf("GEOM_MIRROR"); \ if (g_mirror_debug > 0) \ printf("[%u]", lvl); \ printf(": "); \ printf(__VA_ARGS__); \ printf(" "); \ g_print_bio(bp); \ printf("\n"); \ } \ } while (0) #define G_MIRROR_BIO_FLAG_REGULAR 0x01 #define G_MIRROR_BIO_FLAG_SYNC 0x02 /* * Informations needed for synchronization. */ struct g_mirror_disk_sync { struct g_consumer *ds_consumer; /* Consumer connected to our mirror. */ off_t ds_offset; /* Offset of next request to send. */ off_t ds_offset_done; /* Offset of already synchronized region. */ u_int ds_syncid; /* Disk's synchronization ID. */ u_int ds_inflight; /* Number of in-flight sync requests. */ struct bio **ds_bios; /* BIOs for synchronization I/O. */ }; /* * Informations needed for synchronization. */ struct g_mirror_device_sync { struct g_geom *ds_geom; /* Synchronization geom. */ u_int ds_ndisks; /* Number of disks in SYNCHRONIZING state. */ }; #define G_MIRROR_DISK_STATE_NONE 0 #define G_MIRROR_DISK_STATE_NEW 1 #define G_MIRROR_DISK_STATE_ACTIVE 2 #define G_MIRROR_DISK_STATE_STALE 3 #define G_MIRROR_DISK_STATE_SYNCHRONIZING 4 #define G_MIRROR_DISK_STATE_DISCONNECTED 5 #define G_MIRROR_DISK_STATE_DESTROY 6 struct g_mirror_disk { uint32_t d_id; /* Disk ID. */ struct g_consumer *d_consumer; /* Consumer. */ struct g_mirror_softc *d_softc; /* Back-pointer to softc. */ int d_state; /* Disk state. */ u_int d_priority; /* Disk priority. */ u_int load; /* Averaged queue length */ off_t d_last_offset; /* Last read offset */ uint64_t d_flags; /* Additional flags. */ u_int d_genid; /* Disk's generation ID. */ struct g_mirror_disk_sync d_sync;/* Sync information. */ LIST_ENTRY(g_mirror_disk) d_next; }; #define d_name d_consumer->provider->name #define G_MIRROR_EVENT_DONTWAIT 0x1 #define G_MIRROR_EVENT_WAIT 0x2 #define G_MIRROR_EVENT_DEVICE 0x4 #define G_MIRROR_EVENT_DONE 0x8 struct g_mirror_event { struct g_mirror_disk *e_disk; int e_state; int e_flags; int e_error; TAILQ_ENTRY(g_mirror_event) e_next; }; #define G_MIRROR_DEVICE_FLAG_DESTROY 0x0100000000000000ULL #define G_MIRROR_DEVICE_FLAG_WAIT 0x0200000000000000ULL #define G_MIRROR_DEVICE_FLAG_DESTROYING 0x0400000000000000ULL #define G_MIRROR_DEVICE_FLAG_TASTING 0x0800000000000000ULL #define G_MIRROR_DEVICE_FLAG_WIPE 0x1000000000000000ULL #define G_MIRROR_DEVICE_STATE_STARTING 0 #define G_MIRROR_DEVICE_STATE_RUNNING 1 +#define G_MIRROR_TYPE_MANUAL 0 +#define G_MIRROR_TYPE_AUTOMATIC 1 + /* Bump syncid on first write. */ #define G_MIRROR_BUMP_SYNCID 0x1 /* Bump genid immediately. */ #define G_MIRROR_BUMP_GENID 0x2 struct g_mirror_softc { + u_int sc_type; /* Device type (manual/automatic). */ u_int sc_state; /* Device state. */ uint32_t sc_slice; /* Slice size. */ uint8_t sc_balance; /* Balance algorithm. */ uint64_t sc_mediasize; /* Device size. */ uint32_t sc_sectorsize; /* Sector size. */ uint64_t sc_flags; /* Additional flags. */ struct g_geom *sc_geom; struct g_provider *sc_provider; int sc_provider_open; uint32_t sc_id; /* Mirror unique ID. */ struct sx sc_lock; struct bio_queue_head sc_queue; struct mtx sc_queue_mtx; struct proc *sc_worker; struct bio_queue_head sc_regular_delayed; /* Delayed I/O requests due collision with sync requests. */ struct bio_queue_head sc_inflight; /* In-flight regular write requests. */ struct bio_queue_head sc_sync_delayed; /* Delayed sync requests due collision with regular requests. */ LIST_HEAD(, g_mirror_disk) sc_disks; u_int sc_ndisks; /* Number of disks. */ struct g_mirror_disk *sc_hint; u_int sc_genid; /* Generation ID. */ u_int sc_syncid; /* Synchronization ID. */ int sc_bump_id; struct g_mirror_device_sync sc_sync; int sc_idle; /* DIRTY flags removed. */ time_t sc_last_write; u_int sc_writes; u_int sc_refcnt; /* Number of softc references */ TAILQ_HEAD(, g_mirror_event) sc_events; struct mtx sc_events_mtx; struct callout sc_callout; struct root_hold_token *sc_rootmount; struct mtx sc_done_mtx; }; #define sc_name sc_geom->name +struct g_mirror_metadata; + u_int g_mirror_ndisks(struct g_mirror_softc *sc, int state); +struct g_geom * g_mirror_create(struct g_class *mp, + const struct g_mirror_metadata *md, u_int type); #define G_MIRROR_DESTROY_SOFT 0 #define G_MIRROR_DESTROY_DELAYED 1 #define G_MIRROR_DESTROY_HARD 2 int g_mirror_destroy(struct g_mirror_softc *sc, int how); int g_mirror_event_send(void *arg, int state, int flags); struct g_mirror_metadata; int g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, struct g_mirror_metadata *md); int g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md); void g_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk, struct g_mirror_metadata *md); void g_mirror_update_metadata(struct g_mirror_disk *disk); g_ctl_req_t g_mirror_config; #endif /* _KERNEL */ struct g_mirror_metadata { char md_magic[16]; /* Magic value. */ uint32_t md_version; /* Version number. */ char md_name[16]; /* Mirror name. */ uint32_t md_mid; /* Mirror unique ID. */ uint32_t md_did; /* Disk unique ID. */ uint8_t md_all; /* Number of disks in mirror. */ uint32_t md_genid; /* Generation ID. */ uint32_t md_syncid; /* Synchronization ID. */ uint8_t md_priority; /* Disk priority. */ uint32_t md_slice; /* Slice size. */ uint8_t md_balance; /* Balance type. */ uint64_t md_mediasize; /* Size of the smallest disk in mirror. */ uint32_t md_sectorsize; /* Sector size. */ uint64_t md_sync_offset; /* Synchronized offset. */ uint64_t md_mflags; /* Additional mirror flags. */ uint64_t md_dflags; /* Additional disk flags. */ char md_provider[16]; /* Hardcoded provider. */ uint64_t md_provsize; /* Provider's size. */ u_char md_hash[16]; /* MD5 hash. */ }; static __inline void mirror_metadata_encode(struct g_mirror_metadata *md, u_char *data) { MD5_CTX ctx; bcopy(md->md_magic, data, 16); le32enc(data + 16, md->md_version); bcopy(md->md_name, data + 20, 16); le32enc(data + 36, md->md_mid); le32enc(data + 40, md->md_did); *(data + 44) = md->md_all; le32enc(data + 45, md->md_genid); le32enc(data + 49, md->md_syncid); *(data + 53) = md->md_priority; le32enc(data + 54, md->md_slice); *(data + 58) = md->md_balance; le64enc(data + 59, md->md_mediasize); le32enc(data + 67, md->md_sectorsize); le64enc(data + 71, md->md_sync_offset); le64enc(data + 79, md->md_mflags); le64enc(data + 87, md->md_dflags); bcopy(md->md_provider, data + 95, 16); le64enc(data + 111, md->md_provsize); MD5Init(&ctx); MD5Update(&ctx, data, 119); MD5Final(md->md_hash, &ctx); bcopy(md->md_hash, data + 119, 16); } static __inline int mirror_metadata_decode_v0v1(const u_char *data, struct g_mirror_metadata *md) { MD5_CTX ctx; bcopy(data + 20, md->md_name, 16); md->md_mid = le32dec(data + 36); md->md_did = le32dec(data + 40); md->md_all = *(data + 44); md->md_syncid = le32dec(data + 45); md->md_priority = *(data + 49); md->md_slice = le32dec(data + 50); md->md_balance = *(data + 54); md->md_mediasize = le64dec(data + 55); md->md_sectorsize = le32dec(data + 63); md->md_sync_offset = le64dec(data + 67); md->md_mflags = le64dec(data + 75); md->md_dflags = le64dec(data + 83); bcopy(data + 91, md->md_provider, 16); bcopy(data + 107, md->md_hash, 16); MD5Init(&ctx); MD5Update(&ctx, data, 107); MD5Final(md->md_hash, &ctx); if (bcmp(md->md_hash, data + 107, 16) != 0) return (EINVAL); /* New fields. */ md->md_genid = 0; md->md_provsize = 0; return (0); } static __inline int mirror_metadata_decode_v2(const u_char *data, struct g_mirror_metadata *md) { MD5_CTX ctx; bcopy(data + 20, md->md_name, 16); md->md_mid = le32dec(data + 36); md->md_did = le32dec(data + 40); md->md_all = *(data + 44); md->md_genid = le32dec(data + 45); md->md_syncid = le32dec(data + 49); md->md_priority = *(data + 53); md->md_slice = le32dec(data + 54); md->md_balance = *(data + 58); md->md_mediasize = le64dec(data + 59); md->md_sectorsize = le32dec(data + 67); md->md_sync_offset = le64dec(data + 71); md->md_mflags = le64dec(data + 79); md->md_dflags = le64dec(data + 87); bcopy(data + 95, md->md_provider, 16); bcopy(data + 111, md->md_hash, 16); MD5Init(&ctx); MD5Update(&ctx, data, 111); MD5Final(md->md_hash, &ctx); if (bcmp(md->md_hash, data + 111, 16) != 0) return (EINVAL); /* New fields. */ md->md_provsize = 0; return (0); } static __inline int mirror_metadata_decode_v3v4(const u_char *data, struct g_mirror_metadata *md) { MD5_CTX ctx; bcopy(data + 20, md->md_name, 16); md->md_mid = le32dec(data + 36); md->md_did = le32dec(data + 40); md->md_all = *(data + 44); md->md_genid = le32dec(data + 45); md->md_syncid = le32dec(data + 49); md->md_priority = *(data + 53); md->md_slice = le32dec(data + 54); md->md_balance = *(data + 58); md->md_mediasize = le64dec(data + 59); md->md_sectorsize = le32dec(data + 67); md->md_sync_offset = le64dec(data + 71); md->md_mflags = le64dec(data + 79); md->md_dflags = le64dec(data + 87); bcopy(data + 95, md->md_provider, 16); md->md_provsize = le64dec(data + 111); bcopy(data + 119, md->md_hash, 16); MD5Init(&ctx); MD5Update(&ctx, data, 119); MD5Final(md->md_hash, &ctx); if (bcmp(md->md_hash, data + 119, 16) != 0) return (EINVAL); return (0); } static __inline int mirror_metadata_decode(const u_char *data, struct g_mirror_metadata *md) { int error; bcopy(data, md->md_magic, 16); md->md_version = le32dec(data + 16); switch (md->md_version) { case 0: case 1: error = mirror_metadata_decode_v0v1(data, md); break; case 2: error = mirror_metadata_decode_v2(data, md); break; case 3: case 4: error = mirror_metadata_decode_v3v4(data, md); break; default: error = EINVAL; break; } return (error); } static __inline const char * balance_name(u_int balance) { static const char *algorithms[] = { [G_MIRROR_BALANCE_NONE] = "none", [G_MIRROR_BALANCE_ROUND_ROBIN] = "round-robin", [G_MIRROR_BALANCE_LOAD] = "load", [G_MIRROR_BALANCE_SPLIT] = "split", [G_MIRROR_BALANCE_PREFER] = "prefer", [G_MIRROR_BALANCE_MAX + 1] = "unknown" }; if (balance > G_MIRROR_BALANCE_MAX) balance = G_MIRROR_BALANCE_MAX + 1; return (algorithms[balance]); } static __inline int balance_id(const char *name) { static const char *algorithms[] = { [G_MIRROR_BALANCE_NONE] = "none", [G_MIRROR_BALANCE_ROUND_ROBIN] = "round-robin", [G_MIRROR_BALANCE_LOAD] = "load", [G_MIRROR_BALANCE_SPLIT] = "split", [G_MIRROR_BALANCE_PREFER] = "prefer" }; int n; for (n = G_MIRROR_BALANCE_MIN; n <= G_MIRROR_BALANCE_MAX; n++) { if (strcmp(name, algorithms[n]) == 0) return (n); } return (-1); } static __inline void mirror_metadata_dump(const struct g_mirror_metadata *md) { static const char hex[] = "0123456789abcdef"; char hash[16 * 2 + 1]; u_int i; printf(" magic: %s\n", md->md_magic); printf(" version: %u\n", (u_int)md->md_version); printf(" name: %s\n", md->md_name); printf(" mid: %u\n", (u_int)md->md_mid); printf(" did: %u\n", (u_int)md->md_did); printf(" all: %u\n", (u_int)md->md_all); printf(" genid: %u\n", (u_int)md->md_genid); printf(" syncid: %u\n", (u_int)md->md_syncid); printf(" priority: %u\n", (u_int)md->md_priority); printf(" slice: %u\n", (u_int)md->md_slice); printf(" balance: %s\n", balance_name((u_int)md->md_balance)); printf(" mediasize: %jd\n", (intmax_t)md->md_mediasize); printf("sectorsize: %u\n", (u_int)md->md_sectorsize); printf("syncoffset: %jd\n", (intmax_t)md->md_sync_offset); printf(" mflags:"); if (md->md_mflags == 0) printf(" NONE"); else { if ((md->md_mflags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) printf(" NOFAILSYNC"); if ((md->md_mflags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0) printf(" NOAUTOSYNC"); } printf("\n"); printf(" dflags:"); if (md->md_dflags == 0) printf(" NONE"); else { if ((md->md_dflags & G_MIRROR_DISK_FLAG_DIRTY) != 0) printf(" DIRTY"); if ((md->md_dflags & G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) printf(" SYNCHRONIZING"); if ((md->md_dflags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) printf(" FORCE_SYNC"); if ((md->md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) printf(" INACTIVE"); } printf("\n"); printf("hcprovider: %s\n", md->md_provider); printf(" provsize: %ju\n", (uintmax_t)md->md_provsize); bzero(hash, sizeof(hash)); for (i = 0; i < 16; i++) { hash[i * 2] = hex[md->md_hash[i] >> 4]; hash[i * 2 + 1] = hex[md->md_hash[i] & 0x0f]; } printf(" MD5 hash: %s\n", hash); } #endif /* !_G_MIRROR_H_ */ Index: stable/11/sys/geom/mirror/g_mirror_ctl.c =================================================================== --- stable/11/sys/geom/mirror/g_mirror_ctl.c (revision 318751) +++ stable/11/sys/geom/mirror/g_mirror_ctl.c (revision 318752) @@ -1,894 +1,1095 @@ /*- * Copyright (c) 2004-2009 Pawel Jakub Dawidek * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include +#include #include #include #include #include #include #include #include #include #include #include static struct g_mirror_softc * g_mirror_find_device(struct g_class *mp, const char *name) { struct g_mirror_softc *sc; struct g_geom *gp; g_topology_lock(); LIST_FOREACH(gp, &mp->geom, geom) { sc = gp->softc; if (sc == NULL) continue; if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) continue; if (strcmp(gp->name, name) == 0 || strcmp(sc->sc_name, name) == 0) { g_topology_unlock(); sx_xlock(&sc->sc_lock); return (sc); } } g_topology_unlock(); return (NULL); } static struct g_mirror_disk * g_mirror_find_disk(struct g_mirror_softc *sc, const char *name) { struct g_mirror_disk *disk; sx_assert(&sc->sc_lock, SX_XLOCKED); if (strncmp(name, "/dev/", 5) == 0) name += 5; LIST_FOREACH(disk, &sc->sc_disks, d_next) { if (disk->d_consumer == NULL) continue; if (disk->d_consumer->provider == NULL) continue; if (strcmp(disk->d_consumer->provider->name, name) == 0) return (disk); } return (NULL); } static void g_mirror_ctl_configure(struct gctl_req *req, struct g_class *mp) { struct g_mirror_softc *sc; struct g_mirror_disk *disk; const char *name, *balancep, *prov; intmax_t *slicep, *priority; uint32_t slice; uint8_t balance; int *autosync, *noautosync, *failsync, *nofailsync, *hardcode, *dynamic; int *nargs, do_sync = 0, dirty = 1, do_priority = 0; nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs)); if (nargs == NULL) { gctl_error(req, "No '%s' argument.", "nargs"); return; } if (*nargs != 1 && *nargs != 2) { gctl_error(req, "Invalid number of arguments."); return; } name = gctl_get_asciiparam(req, "arg0"); if (name == NULL) { gctl_error(req, "No 'arg%u' argument.", 0); return; } balancep = gctl_get_asciiparam(req, "balance"); if (balancep == NULL) { gctl_error(req, "No '%s' argument.", "balance"); return; } autosync = gctl_get_paraml(req, "autosync", sizeof(*autosync)); if (autosync == NULL) { gctl_error(req, "No '%s' argument.", "autosync"); return; } noautosync = gctl_get_paraml(req, "noautosync", sizeof(*noautosync)); if (noautosync == NULL) { gctl_error(req, "No '%s' argument.", "noautosync"); return; } failsync = gctl_get_paraml(req, "failsync", sizeof(*failsync)); if (failsync == NULL) { gctl_error(req, "No '%s' argument.", "failsync"); return; } nofailsync = gctl_get_paraml(req, "nofailsync", sizeof(*nofailsync)); if (nofailsync == NULL) { gctl_error(req, "No '%s' argument.", "nofailsync"); return; } hardcode = gctl_get_paraml(req, "hardcode", sizeof(*hardcode)); if (hardcode == NULL) { gctl_error(req, "No '%s' argument.", "hardcode"); return; } dynamic = gctl_get_paraml(req, "dynamic", sizeof(*dynamic)); if (dynamic == NULL) { gctl_error(req, "No '%s' argument.", "dynamic"); return; } priority = gctl_get_paraml(req, "priority", sizeof(*priority)); if (priority == NULL) { gctl_error(req, "No '%s' argument.", "priority"); return; } if (*priority < -1 || *priority > 255) { gctl_error(req, "Priority range is 0 to 255, %jd given", *priority); return; } /* * Since we have a priority, we also need a provider now. * Note: be WARNS safe, by always assigning prov and only throw an * error if *priority != -1. */ prov = gctl_get_asciiparam(req, "arg1"); if (*priority > -1) { if (prov == NULL) { gctl_error(req, "Priority needs a disk name"); return; } do_priority = 1; } if (*autosync && *noautosync) { gctl_error(req, "'%s' and '%s' specified.", "autosync", "noautosync"); return; } if (*failsync && *nofailsync) { gctl_error(req, "'%s' and '%s' specified.", "failsync", "nofailsync"); return; } if (*hardcode && *dynamic) { gctl_error(req, "'%s' and '%s' specified.", "hardcode", "dynamic"); return; } sc = g_mirror_find_device(mp, name); if (sc == NULL) { gctl_error(req, "No such device: %s.", name); return; } if (*balancep == '\0') balance = sc->sc_balance; else { if (balance_id(balancep) == -1) { gctl_error(req, "Invalid balance algorithm."); sx_xunlock(&sc->sc_lock); return; } balance = balance_id(balancep); } slicep = gctl_get_paraml(req, "slice", sizeof(*slicep)); if (slicep == NULL) { gctl_error(req, "No '%s' argument.", "slice"); sx_xunlock(&sc->sc_lock); return; } if (*slicep == -1) slice = sc->sc_slice; else slice = *slicep; /* Enforce usage() of -p not allowing any other options. */ if (do_priority && (*autosync || *noautosync || *failsync || *nofailsync || *hardcode || *dynamic || *slicep != -1 || *balancep != '\0')) { sx_xunlock(&sc->sc_lock); gctl_error(req, "only -p accepted when setting priority"); return; } if (sc->sc_balance == balance && sc->sc_slice == slice && !*autosync && !*noautosync && !*failsync && !*nofailsync && !*hardcode && !*dynamic && !do_priority) { sx_xunlock(&sc->sc_lock); gctl_error(req, "Nothing has changed."); return; } if ((!do_priority && *nargs != 1) || (do_priority && *nargs != 2)) { sx_xunlock(&sc->sc_lock); gctl_error(req, "Invalid number of arguments."); return; } if (g_mirror_ndisks(sc, -1) < sc->sc_ndisks) { sx_xunlock(&sc->sc_lock); gctl_error(req, "Not all disks connected. Try 'forget' command " "first."); return; } sc->sc_balance = balance; sc->sc_slice = slice; if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0) { if (*autosync) { sc->sc_flags &= ~G_MIRROR_DEVICE_FLAG_NOAUTOSYNC; do_sync = 1; } } else { if (*noautosync) sc->sc_flags |= G_MIRROR_DEVICE_FLAG_NOAUTOSYNC; } if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) { if (*failsync) sc->sc_flags &= ~G_MIRROR_DEVICE_FLAG_NOFAILSYNC; } else { if (*nofailsync) { sc->sc_flags |= G_MIRROR_DEVICE_FLAG_NOFAILSYNC; dirty = 0; } } LIST_FOREACH(disk, &sc->sc_disks, d_next) { /* * Handle priority first, since we only need one disk, do one * operation on it and then we're done. No need to check other * flags, as usage doesn't allow it. */ if (do_priority) { if (strcmp(disk->d_name, prov) == 0) { if (disk->d_priority == *priority) gctl_error(req, "Nothing has changed."); else { disk->d_priority = *priority; g_mirror_update_metadata(disk); } break; } continue; } if (do_sync) { if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; } if (*hardcode) disk->d_flags |= G_MIRROR_DISK_FLAG_HARDCODED; else if (*dynamic) disk->d_flags &= ~G_MIRROR_DISK_FLAG_HARDCODED; if (!dirty) disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; g_mirror_update_metadata(disk); if (do_sync) { if (disk->d_state == G_MIRROR_DISK_STATE_STALE) { g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, G_MIRROR_EVENT_DONTWAIT); } } } sx_xunlock(&sc->sc_lock); } static void +g_mirror_create_orphan(struct g_consumer *cp) +{ + + KASSERT(1 == 0, ("%s called while creating %s.", __func__, + cp->provider->name)); +} + +static void +g_mirror_ctl_create(struct gctl_req *req, struct g_class *mp) +{ + struct g_mirror_metadata md; + struct g_geom *gp; + struct g_consumer *cp; + struct g_provider *pp; + struct g_mirror_softc *sc; + struct sbuf *sb; + const char *name; + char param[16]; + int *nargs; + intmax_t *val; + int *ival; + const char *sval; + int bal; + unsigned attached, no, sectorsize; + off_t mediasize; + + nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs)); + if (nargs == NULL) { + gctl_error(req, "No '%s' argument.", "nargs"); + return; + } + if (*nargs <= 2) { + gctl_error(req, "Too few arguments."); + return; + } + + strlcpy(md.md_magic, G_MIRROR_MAGIC, sizeof(md.md_magic)); + md.md_version = G_MIRROR_VERSION; + name = gctl_get_asciiparam(req, "arg0"); + if (name == NULL) { + gctl_error(req, "No 'arg%u' argument.", 0); + return; + } + strlcpy(md.md_name, name, sizeof(md.md_name)); + md.md_mid = arc4random(); + md.md_all = *nargs - 1; + md.md_genid = 0; + md.md_syncid = 1; + md.md_sync_offset = 0; + val = gctl_get_paraml(req, "slice", sizeof(*val)); + if (val == NULL) { + gctl_error(req, "No slice argument."); + return; + } + md.md_slice = *val; + sval = gctl_get_asciiparam(req, "balance"); + if (sval == NULL) { + gctl_error(req, "No balance argument."); + return; + } + bal = balance_id(sval); + if (bal < 0) { + gctl_error(req, "Invalid balance algorithm."); + return; + } + md.md_balance = bal; + md.md_mflags = 0; + md.md_dflags = 0; + ival = gctl_get_paraml(req, "noautosync", sizeof(*ival)); + if (ival != NULL && *ival) + md.md_mflags |= G_MIRROR_DEVICE_FLAG_NOAUTOSYNC; + ival = gctl_get_paraml(req, "nofailsync", sizeof(*ival)); + if (ival != NULL && *ival) + md.md_mflags |= G_MIRROR_DEVICE_FLAG_NOFAILSYNC; + /* These fields not used in manual mode. */ + bzero(md.md_provider, sizeof(md.md_provider)); + md.md_provsize = 0; + + g_topology_lock(); + mediasize = OFF_MAX; + sectorsize = 0; + gp = g_new_geomf(mp, "%s", md.md_name); + gp->orphan = g_mirror_create_orphan; + cp = g_new_consumer(gp); + for (no = 1; no < *nargs; no++) { + snprintf(param, sizeof(param), "arg%u", no); + name = gctl_get_asciiparam(req, param); + if (name == NULL) { + gctl_error(req, "No 'arg%u' argument.", no); +err: + g_destroy_consumer(cp); + g_destroy_geom(gp); + g_topology_unlock(); + return; + } + if (strncmp(name, "/dev/", strlen("/dev/")) == 0) + name += strlen("/dev/"); + pp = g_provider_by_name(name); + if (pp == NULL) { + G_MIRROR_DEBUG(1, "Disk %s is invalid.", name); + gctl_error(req, "Disk %s is invalid.", name); + goto err; + } + g_attach(cp, pp); + if (g_access(cp, 1, 0, 0) != 0) { + G_MIRROR_DEBUG(1, "Can't open disk %s.", name); + gctl_error(req, "Can't open disk %s.", name); +err2: + g_detach(cp); + goto err; + } + if (pp->mediasize == 0 || pp->sectorsize == 0) { + G_MIRROR_DEBUG(1, "Disk %s has no media.", name); + gctl_error(req, "Disk %s has no media.", name); + g_access(cp, -1, 0, 0); + goto err2; + } + if (pp->mediasize < mediasize) + mediasize = pp->mediasize; + if (pp->sectorsize > sectorsize) + sectorsize = pp->sectorsize; + g_access(cp, -1, 0, 0); + g_detach(cp); + } + g_destroy_consumer(cp); + g_destroy_geom(gp); + md.md_mediasize = mediasize; + md.md_sectorsize = sectorsize; + md.md_mediasize -= (md.md_mediasize % md.md_sectorsize); + + gp = g_mirror_create(mp, &md, G_MIRROR_TYPE_MANUAL); + if (gp == NULL) { + gctl_error(req, "Can't create %s.", md.md_name); + g_topology_unlock(); + return; + } + + sc = gp->softc; + g_topology_unlock(); + sx_xlock(&sc->sc_lock); + sc->sc_flags |= G_MIRROR_DEVICE_FLAG_TASTING; + sb = sbuf_new_auto(); + sbuf_printf(sb, "Can't attach disk(s) to %s:", gp->name); + for (attached = 0, no = 1; no < *nargs; no++) { + snprintf(param, sizeof(param), "arg%u", no); + name = gctl_get_asciiparam(req, param); + if (strncmp(name, "/dev/", strlen("/dev/")) == 0) + name += strlen("/dev/"); + pp = g_provider_by_name(name); + if (pp == NULL) { + G_MIRROR_DEBUG(1, "Provider %s disappear?!", name); + sbuf_printf(sb, " %s", name); + continue; + } + md.md_did = arc4random(); + md.md_priority = no - 1; + if (g_mirror_add_disk(sc, pp, &md) != 0) { + G_MIRROR_DEBUG(1, "Disk %u (%s) not attached to %s.", + no, pp->name, gp->name); + sbuf_printf(sb, " %s", pp->name); + continue; + } + attached++; + } + sbuf_finish(sb); + sc->sc_flags &= ~G_MIRROR_DEVICE_FLAG_TASTING; + if (md.md_all != attached || + (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { + g_mirror_destroy(gp->softc, G_MIRROR_DESTROY_HARD); + gctl_error(req, "%s", sbuf_data(sb)); + } else + sx_xunlock(&sc->sc_lock); + sbuf_delete(sb); +} + +static void g_mirror_ctl_rebuild(struct gctl_req *req, struct g_class *mp) { struct g_mirror_metadata md; struct g_mirror_softc *sc; struct g_mirror_disk *disk; struct g_provider *pp; const char *name; char param[16]; int error, *nargs; u_int i; nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs)); if (nargs == NULL) { gctl_error(req, "No '%s' argument.", "nargs"); return; } if (*nargs < 2) { gctl_error(req, "Too few arguments."); return; } name = gctl_get_asciiparam(req, "arg0"); if (name == NULL) { gctl_error(req, "No 'arg%u' argument.", 0); return; } sc = g_mirror_find_device(mp, name); if (sc == NULL) { gctl_error(req, "No such device: %s.", name); return; } for (i = 1; i < (u_int)*nargs; i++) { snprintf(param, sizeof(param), "arg%u", i); name = gctl_get_asciiparam(req, param); if (name == NULL) { gctl_error(req, "No 'arg%u' argument.", i); continue; } disk = g_mirror_find_disk(sc, name); if (disk == NULL) { gctl_error(req, "No such provider: %s.", name); continue; } if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 1 && disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) { /* * This is the last active disk. There will be nothing * to rebuild it from, so deny this request. */ gctl_error(req, "Provider %s is the last active provider in %s.", name, sc->sc_geom->name); break; } /* * Do rebuild by resetting syncid, disconnecting the disk and * connecting it again. */ disk->d_sync.ds_syncid = 0; if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0) disk->d_flags |= G_MIRROR_DISK_FLAG_FORCE_SYNC; g_mirror_update_metadata(disk); pp = disk->d_consumer->provider; g_topology_lock(); error = g_mirror_read_metadata(disk->d_consumer, &md); g_topology_unlock(); g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, G_MIRROR_EVENT_WAIT); if (error != 0) { gctl_error(req, "Cannot read metadata from %s.", pp->name); continue; } error = g_mirror_add_disk(sc, pp, &md); if (error != 0) { gctl_error(req, "Cannot reconnect component %s.", pp->name); continue; } } sx_xunlock(&sc->sc_lock); } static void g_mirror_ctl_insert(struct gctl_req *req, struct g_class *mp) { struct g_mirror_softc *sc; struct g_mirror_disk *disk; struct g_mirror_metadata md; struct g_provider *pp; struct g_consumer *cp; intmax_t *priority; const char *name; char param[16]; u_char *sector; u_int i, n; int error, *nargs, *hardcode, *inactive; struct { struct g_provider *provider; struct g_consumer *consumer; } *disks; + off_t mdsize; nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs)); if (nargs == NULL) { gctl_error(req, "No '%s' argument.", "nargs"); return; } if (*nargs < 2) { gctl_error(req, "Too few arguments."); return; } priority = gctl_get_paraml(req, "priority", sizeof(*priority)); if (priority == NULL) { gctl_error(req, "No '%s' argument.", "priority"); return; } inactive = gctl_get_paraml(req, "inactive", sizeof(*inactive)); if (inactive == NULL) { gctl_error(req, "No '%s' argument.", "inactive"); return; } hardcode = gctl_get_paraml(req, "hardcode", sizeof(*hardcode)); if (hardcode == NULL) { gctl_error(req, "No '%s' argument.", "hardcode"); return; } name = gctl_get_asciiparam(req, "arg0"); if (name == NULL) { gctl_error(req, "No 'arg%u' argument.", 0); return; } sc = g_mirror_find_device(mp, name); if (sc == NULL) { gctl_error(req, "No such device: %s.", name); return; } if (g_mirror_ndisks(sc, -1) < sc->sc_ndisks) { gctl_error(req, "Not all disks connected."); sx_xunlock(&sc->sc_lock); return; } disks = g_malloc(sizeof(*disks) * (*nargs), M_WAITOK | M_ZERO); g_topology_lock(); for (i = 1, n = 0; i < (u_int)*nargs; i++) { snprintf(param, sizeof(param), "arg%u", i); name = gctl_get_asciiparam(req, param); if (name == NULL) { gctl_error(req, "No 'arg%u' argument.", i); continue; } if (g_mirror_find_disk(sc, name) != NULL) { gctl_error(req, "Provider %s already inserted.", name); continue; } if (strncmp(name, "/dev/", 5) == 0) name += 5; pp = g_provider_by_name(name); if (pp == NULL) { gctl_error(req, "Unknown provider %s.", name); continue; } - if (sc->sc_provider->mediasize > - pp->mediasize - pp->sectorsize) { - gctl_error(req, "Provider %s too small.", name); - continue; - } - if ((sc->sc_provider->sectorsize % pp->sectorsize) != 0) { - gctl_error(req, "Invalid sectorsize of provider %s.", - name); - continue; - } cp = g_new_consumer(sc->sc_geom); if (g_attach(cp, pp) != 0) { g_destroy_consumer(cp); gctl_error(req, "Cannot attach to provider %s.", name); continue; } if (g_access(cp, 0, 1, 1) != 0) { + gctl_error(req, "Cannot access provider %s.", name); +err: g_detach(cp); g_destroy_consumer(cp); - gctl_error(req, "Cannot access provider %s.", name); continue; } + mdsize = (sc->sc_type == G_MIRROR_TYPE_AUTOMATIC) ? + pp->sectorsize : 0; + if (sc->sc_provider->mediasize > pp->mediasize - mdsize) { + gctl_error(req, "Provider %s too small.", name); +err2: + g_access(cp, 0, -1, -1); + goto err; + } + if ((sc->sc_provider->sectorsize % pp->sectorsize) != 0) { + gctl_error(req, "Invalid sectorsize of provider %s.", + name); + goto err2; + } + if (sc->sc_type != G_MIRROR_TYPE_AUTOMATIC) { + g_access(cp, 0, -1, -1); + g_detach(cp); + g_destroy_consumer(cp); + g_topology_unlock(); + sc->sc_ndisks++; + g_mirror_fill_metadata(sc, NULL, &md); + md.md_priority = *priority; + if (*inactive) + md.md_dflags |= G_MIRROR_DISK_FLAG_INACTIVE; + if (g_mirror_add_disk(sc, pp, &md) != 0) { + sc->sc_ndisks--; + gctl_error(req, "Disk %s not inserted.", name); + } + g_topology_lock(); + continue; + } disks[n].provider = pp; disks[n].consumer = cp; n++; } if (n == 0) { g_topology_unlock(); sx_xunlock(&sc->sc_lock); g_free(disks); return; } sc->sc_ndisks += n; again: for (i = 0; i < n; i++) { if (disks[i].consumer == NULL) continue; g_mirror_fill_metadata(sc, NULL, &md); md.md_priority = *priority; if (*inactive) md.md_dflags |= G_MIRROR_DISK_FLAG_INACTIVE; pp = disks[i].provider; if (*hardcode) { strlcpy(md.md_provider, pp->name, sizeof(md.md_provider)); } else { bzero(md.md_provider, sizeof(md.md_provider)); } md.md_provsize = pp->mediasize; sector = g_malloc(pp->sectorsize, M_WAITOK); mirror_metadata_encode(&md, sector); error = g_write_data(disks[i].consumer, pp->mediasize - pp->sectorsize, sector, pp->sectorsize); g_free(sector); if (error != 0) { gctl_error(req, "Cannot store metadata on %s.", pp->name); g_access(disks[i].consumer, 0, -1, -1); g_detach(disks[i].consumer); g_destroy_consumer(disks[i].consumer); disks[i].consumer = NULL; disks[i].provider = NULL; sc->sc_ndisks--; goto again; } } g_topology_unlock(); if (i == 0) { /* All writes failed. */ sx_xunlock(&sc->sc_lock); g_free(disks); return; } LIST_FOREACH(disk, &sc->sc_disks, d_next) { g_mirror_update_metadata(disk); } /* * Release provider and wait for retaste. */ g_topology_lock(); for (i = 0; i < n; i++) { if (disks[i].consumer == NULL) continue; g_access(disks[i].consumer, 0, -1, -1); g_detach(disks[i].consumer); g_destroy_consumer(disks[i].consumer); } g_topology_unlock(); sx_xunlock(&sc->sc_lock); g_free(disks); } static void g_mirror_ctl_remove(struct gctl_req *req, struct g_class *mp) { struct g_mirror_softc *sc; struct g_mirror_disk *disk; const char *name; char param[16]; int *nargs; u_int i, active; nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs)); if (nargs == NULL) { gctl_error(req, "No '%s' argument.", "nargs"); return; } if (*nargs < 2) { gctl_error(req, "Too few arguments."); return; } name = gctl_get_asciiparam(req, "arg0"); if (name == NULL) { gctl_error(req, "No 'arg%u' argument.", 0); return; } sc = g_mirror_find_device(mp, name); if (sc == NULL) { gctl_error(req, "No such device: %s.", name); return; } if (g_mirror_ndisks(sc, -1) < sc->sc_ndisks) { sx_xunlock(&sc->sc_lock); gctl_error(req, "Not all disks connected. Try 'forget' command " "first."); return; } active = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); for (i = 1; i < (u_int)*nargs; i++) { snprintf(param, sizeof(param), "arg%u", i); name = gctl_get_asciiparam(req, param); if (name == NULL) { gctl_error(req, "No 'arg%u' argument.", i); continue; } disk = g_mirror_find_disk(sc, name); if (disk == NULL) { gctl_error(req, "No such provider: %s.", name); continue; } if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) { if (active > 1) active--; else { gctl_error(req, "%s: Can't remove the last " "ACTIVE component %s.", sc->sc_geom->name, name); continue; } } g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DESTROY, G_MIRROR_EVENT_DONTWAIT); } sx_xunlock(&sc->sc_lock); } static void g_mirror_ctl_resize(struct gctl_req *req, struct g_class *mp) { struct g_mirror_softc *sc; struct g_mirror_disk *disk; uint64_t mediasize; const char *name, *s; char *x; int *nargs; nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs)); if (nargs == NULL) { gctl_error(req, "No '%s' argument.", "nargs"); return; } if (*nargs != 1) { gctl_error(req, "Missing device."); return; } name = gctl_get_asciiparam(req, "arg0"); if (name == NULL) { gctl_error(req, "No 'arg%u' argument.", 0); return; } s = gctl_get_asciiparam(req, "size"); if (s == NULL) { gctl_error(req, "No '%s' argument.", "size"); return; } mediasize = strtouq(s, &x, 0); if (*x != '\0' || mediasize == 0) { gctl_error(req, "Invalid '%s' argument.", "size"); return; } sc = g_mirror_find_device(mp, name); if (sc == NULL) { gctl_error(req, "No such device: %s.", name); return; } /* Deny shrinking of an opened provider */ if ((g_debugflags & 16) == 0 && sc->sc_provider_open > 0) { if (sc->sc_mediasize > mediasize) { gctl_error(req, "Device %s is busy.", sc->sc_provider->name); sx_xunlock(&sc->sc_lock); return; } } LIST_FOREACH(disk, &sc->sc_disks, d_next) { if (mediasize > disk->d_consumer->provider->mediasize - disk->d_consumer->provider->sectorsize) { gctl_error(req, "Provider %s is too small.", disk->d_name); sx_xunlock(&sc->sc_lock); return; } } /* Update the size. */ sc->sc_mediasize = mediasize; LIST_FOREACH(disk, &sc->sc_disks, d_next) { g_mirror_update_metadata(disk); } g_topology_lock(); g_resize_provider(sc->sc_provider, mediasize); g_topology_unlock(); sx_xunlock(&sc->sc_lock); } static void g_mirror_ctl_deactivate(struct gctl_req *req, struct g_class *mp) { struct g_mirror_softc *sc; struct g_mirror_disk *disk; const char *name; char param[16]; int *nargs; u_int i, active; nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs)); if (nargs == NULL) { gctl_error(req, "No '%s' argument.", "nargs"); return; } if (*nargs < 2) { gctl_error(req, "Too few arguments."); return; } name = gctl_get_asciiparam(req, "arg0"); if (name == NULL) { gctl_error(req, "No 'arg%u' argument.", 0); return; } sc = g_mirror_find_device(mp, name); if (sc == NULL) { gctl_error(req, "No such device: %s.", name); return; } active = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); for (i = 1; i < (u_int)*nargs; i++) { snprintf(param, sizeof(param), "arg%u", i); name = gctl_get_asciiparam(req, param); if (name == NULL) { gctl_error(req, "No 'arg%u' argument.", i); continue; } disk = g_mirror_find_disk(sc, name); if (disk == NULL) { gctl_error(req, "No such provider: %s.", name); continue; } if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) { if (active > 1) active--; else { gctl_error(req, "%s: Can't deactivate the " "last ACTIVE component %s.", sc->sc_geom->name, name); continue; } } disk->d_flags |= G_MIRROR_DISK_FLAG_INACTIVE; disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; g_mirror_update_metadata(disk); sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, G_MIRROR_EVENT_DONTWAIT); } sx_xunlock(&sc->sc_lock); } static void g_mirror_ctl_forget(struct gctl_req *req, struct g_class *mp) { struct g_mirror_softc *sc; struct g_mirror_disk *disk; const char *name; char param[16]; int *nargs; u_int i; nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs)); if (nargs == NULL) { gctl_error(req, "No '%s' argument.", "nargs"); return; } if (*nargs < 1) { gctl_error(req, "Missing device(s)."); return; } for (i = 0; i < (u_int)*nargs; i++) { snprintf(param, sizeof(param), "arg%u", i); name = gctl_get_asciiparam(req, param); if (name == NULL) { gctl_error(req, "No 'arg%u' argument.", i); return; } sc = g_mirror_find_device(mp, name); if (sc == NULL) { gctl_error(req, "No such device: %s.", name); return; } if (g_mirror_ndisks(sc, -1) == sc->sc_ndisks) { sx_xunlock(&sc->sc_lock); G_MIRROR_DEBUG(1, "All disks connected in %s, skipping.", sc->sc_name); continue; } sc->sc_ndisks = g_mirror_ndisks(sc, -1); LIST_FOREACH(disk, &sc->sc_disks, d_next) { g_mirror_update_metadata(disk); } sx_xunlock(&sc->sc_lock); } } static void g_mirror_ctl_stop(struct gctl_req *req, struct g_class *mp, int wipe) { struct g_mirror_softc *sc; int *force, *nargs, error; const char *name; char param[16]; u_int i; int how; nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs)); if (nargs == NULL) { gctl_error(req, "No '%s' argument.", "nargs"); return; } if (*nargs < 1) { gctl_error(req, "Missing device(s)."); return; } force = gctl_get_paraml(req, "force", sizeof(*force)); if (force == NULL) { gctl_error(req, "No '%s' argument.", "force"); return; } if (*force) how = G_MIRROR_DESTROY_HARD; else how = G_MIRROR_DESTROY_SOFT; for (i = 0; i < (u_int)*nargs; i++) { snprintf(param, sizeof(param), "arg%u", i); name = gctl_get_asciiparam(req, param); if (name == NULL) { gctl_error(req, "No 'arg%u' argument.", i); return; } sc = g_mirror_find_device(mp, name); if (sc == NULL) { gctl_error(req, "No such device: %s.", name); return; } g_cancel_event(sc); if (wipe) sc->sc_flags |= G_MIRROR_DEVICE_FLAG_WIPE; error = g_mirror_destroy(sc, how); if (error != 0) { gctl_error(req, "Cannot destroy device %s (error=%d).", sc->sc_geom->name, error); if (wipe) sc->sc_flags &= ~G_MIRROR_DEVICE_FLAG_WIPE; sx_xunlock(&sc->sc_lock); return; } /* No need to unlock, because lock is already dead. */ } } void g_mirror_config(struct gctl_req *req, struct g_class *mp, const char *verb) { uint32_t *version; g_topology_assert(); version = gctl_get_paraml(req, "version", sizeof(*version)); if (version == NULL) { gctl_error(req, "No '%s' argument.", "version"); return; } if (*version != G_MIRROR_VERSION) { gctl_error(req, "Userland and kernel parts are out of sync."); return; } g_topology_unlock(); if (strcmp(verb, "configure") == 0) g_mirror_ctl_configure(req, mp); + else if (strcmp(verb, "create") == 0) + g_mirror_ctl_create(req, mp); else if (strcmp(verb, "rebuild") == 0) g_mirror_ctl_rebuild(req, mp); else if (strcmp(verb, "insert") == 0) g_mirror_ctl_insert(req, mp); else if (strcmp(verb, "remove") == 0) g_mirror_ctl_remove(req, mp); else if (strcmp(verb, "resize") == 0) g_mirror_ctl_resize(req, mp); else if (strcmp(verb, "deactivate") == 0) g_mirror_ctl_deactivate(req, mp); else if (strcmp(verb, "forget") == 0) g_mirror_ctl_forget(req, mp); else if (strcmp(verb, "stop") == 0) g_mirror_ctl_stop(req, mp, 0); else if (strcmp(verb, "destroy") == 0) g_mirror_ctl_stop(req, mp, 1); else gctl_error(req, "Unknown verb."); g_topology_lock(); } Index: stable/11 =================================================================== --- stable/11 (revision 318751) +++ stable/11 (revision 318752) Property changes on: stable/11 ___________________________________________________________________ Modified: svn:mergeinfo ## -0,0 +0,1 ## Merged /head:r309321