4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2014, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/utils/mount_lustre.c
38 * Author: Robert Read <rread@clusterfs.com>
39 * Author: Nathan Rutman <nathan@clusterfs.com>
54 #include <sys/mount.h>
55 #include <lustre_ver.h>
58 #include <lnet/nidstr.h>
59 #include <lustre/lustre_idl.h>
62 #include "mount_utils.h"
64 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 10, 53, 0)
67 * We only #include a kernel level include file here because
68 * important MS_ flag #defines are missing from the SLES version
70 * In the future if SLES updates sys/mount.h to have a more complete
71 * set of flag #defines we should stop including linux/fs.h
73 #if !defined(MS_RDONLY)
79 # define WITH_LIBMOUNT "(libmount)"
81 # define WITH_LIBMOUNT ""
85 #define MAX_RETRIES 99
93 fprintf(out, "\nThis mount helper should only be invoked via the "
94 "mount (8) command,\ne.g. mount -t lustre dev dir\n\n");
95 fprintf(out, "usage: %s [-fhnvV] [-o <mntopt>] <device> <mountpt>\n",
98 "\t<device>: the disk device, or for a client:\n"
99 "\t\t<mgsnid>[:<altmgsnid>...]:/<filesystem>[/<subdir>]\n"
100 "\t<filesystem>: name of the Lustre filesystem (e.g. lustre1)\n"
101 "\t<mountpt>: filesystem mountpoint (e.g. /mnt/lustre)\n"
102 "\t-f|--fake: fake mount (updates /etc/mtab)\n"
103 "\t-o force|--force: force mount even if already in /etc/mtab\n"
104 "\t-h|--help: print this usage message\n"
105 "\t-n|--nomtab: do not update /etc/mtab after mount\n"
106 "\t-v|--verbose: print verbose config settings\n"
107 "\t-V|--version: output build version of the utility and exit\n"
108 "\t<mntopt>: one or more comma separated of:\n"
109 "\t\t(no)flock,(no)user_xattr,(no)acl\n"
110 "\t\tabort_recov: abort server recovery handling\n"
111 "\t\tnosvc: only start MGC/MGS obds\n"
112 "\t\tnomgs: only start target obds, using existing MGS\n"
113 "\t\tnoscrub: NOT auto start OI scrub unless start explicitly\n"
114 "\t\tskip_lfsck: NOT auto resume the paused/crashed LFSCK\n"
115 "\t\texclude=<ostname>[:<ostname>] : colon-separated list of "
116 "inactive OSTs (e.g. lustre-OST0001)\n"
117 "\t\tretry=<num>: number of times mount is retried by client\n"
118 "\t\tmd_stripe_cache_size=<num>: set the raid stripe cache "
119 "size for the underlying raid if present\n");
120 exit((out != stdout) ? EINVAL : 0);
123 /* Get rid of symbolic hostnames for tcp, since kernel can't do lookups */
124 #define MAXNIDSTR 1024
125 static char *convert_hostnames(char *s1)
127 char *converted, *s2 = 0, *c;
129 int left = MAXNIDSTR;
132 converted = malloc(left);
133 if (converted == NULL) {
134 fprintf(stderr, "out of memory: needed %d bytes\n",
139 while ((left > 0) && (*s1 != '/')) {
140 s2 = strpbrk(s1, ",:");
145 nid = libcfs_str2nid(s1);
146 *s2 = sep; /* back to original string */
147 if (nid == LNET_NID_ANY)
149 c += snprintf(c, left, "%s%c", libcfs_nid2str(nid), sep);
150 left = converted + MAXNIDSTR - c;
153 snprintf(c, left, "%s", s1);
156 fprintf(stderr, "%s: Can't parse NID '%s'\n", progname, s1);
161 /*****************************************************************************
163 * This part was cribbed from util-linux/mount/mount.c. There was no clear
164 * license information, but many other files in the package are identified as
165 * GNU GPL, so it's a pretty safe bet that was their intent.
167 ****************************************************************************/
169 const char *opt; /* option name */
170 int inv; /* true if flag value should be inverted */
171 int mask; /* flag mask value */
174 static const struct opt_map opt_map[] = {
175 /*"optname", inv,ms_mask */
176 /* These flags are parsed by mount, not lustre */
177 { "defaults", 0, 0 }, /* default options */
178 { "remount", 0, MS_REMOUNT}, /* remount with different options */
179 { "rw", 1, MS_RDONLY }, /* read-write */
180 { "ro", 0, MS_RDONLY }, /* read-only */
181 { "exec", 1, MS_NOEXEC }, /* permit execution of binaries */
182 { "noexec", 0, MS_NOEXEC }, /* don't execute binaries */
183 { "suid", 1, MS_NOSUID }, /* honor suid executables */
184 { "nosuid", 0, MS_NOSUID }, /* don't honor suid executables */
185 { "dev", 1, MS_NODEV }, /* interpret device files */
186 { "nodev", 0, MS_NODEV }, /* don't interpret devices */
187 { "sync", 0, MS_SYNCHRONOUS}, /* synchronous I/O */
188 { "async", 1, MS_SYNCHRONOUS}, /* asynchronous I/O */
189 { "atime", 1, MS_NOATIME }, /* set file access time on read */
190 { "noatime", 0, MS_NOATIME }, /* do not set file access time on read */
192 { "diratime", 1, MS_NODIRATIME }, /* set file access time on read */
193 { "nodiratime",0,MS_NODIRATIME }, /* do not set file access time on read */
196 { "relatime", 0, MS_RELATIME }, /* set file access time on read */
197 { "norelatime",1,MS_RELATIME }, /* do not set file access time on read */
199 #ifdef MS_STRICTATIME
200 { "strictatime",0,MS_STRICTATIME }, /* update access time strictly */
202 { "auto", 0, 0 }, /* Can be mounted using -a */
203 { "noauto", 0, 0 }, /* Can only be mounted explicitly */
204 { "nousers", 1, 0 }, /* Forbid ordinary user to mount */
205 { "nouser", 1, 0 }, /* Forbid ordinary user to mount */
206 { "noowner", 1, 0 }, /* Device owner has no special privs */
207 { "_netdev", 0, 0 }, /* Device accessible only via network */
211 /****************************************************************************/
213 /* 1 = don't pass on to lustre
214 0 = pass on to lustre */
215 static int parse_one_option(const char *check, int *flagp)
217 const struct opt_map *opt;
219 for (opt = &opt_map[0]; opt->opt != NULL; opt++) {
220 if (strncmp(check, opt->opt, strlen(opt->opt)) == 0) {
223 *flagp &= ~(opt->mask);
230 /* Assume any unknown options are valid and pass them on. The mount
231 will fail if lmd_parse, ll_options or ldiskfs doesn't recognize it.*/
235 static void append_option(char *options, const char *one)
238 strcat(options, ",");
239 strcat(options, one);
242 /* Replace options with subset of Lustre-specific options, and
243 fill in mount flags */
244 int parse_options(struct mount_opts *mop, char *orig_options, int *flagp)
246 char *options, *opt, *nextopt, *arg, *val;
248 options = calloc(strlen(orig_options) + 1, 1);
250 nextopt = orig_options;
251 while ((opt = strsep(&nextopt, ","))) {
256 /* Handle retries in a slightly different
259 val = strchr(opt, '=');
260 /* please note that some ldiskfs mount options are also in the form
261 * of param=value. We should pay attention not to remove those
262 * mount options, see bug 22097. */
263 if (val && strncmp(arg, "md_stripe_cache_size", 20) == 0) {
264 mop->mo_md_stripe_cache_size = atoi(val + 1);
265 } else if (val && strncmp(arg, "retry", 5) == 0) {
266 mop->mo_retry = atoi(val + 1);
267 if (mop->mo_retry > MAX_RETRIES)
268 mop->mo_retry = MAX_RETRIES;
269 else if (mop->mo_retry < 0)
271 } else if (val && strncmp(arg, "mgssec", 6) == 0) {
272 append_option(options, opt);
273 } else if (strncmp(arg, "nosvc", 5) == 0) {
275 append_option(options, opt);
276 } else if (strcmp(opt, "force") == 0) {
277 /* XXX special check for 'force' option */
279 printf("force: %d\n", mop->mo_force);
280 } else if (parse_one_option(opt, flagp) == 0) {
281 /* pass this on as an option */
282 append_option(options, opt);
285 #ifdef MS_STRICTATIME
286 #if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(3, 2, 53, 0)
289 * In the future when upstream fixes land in all supported kernels
290 * we should stop forcing MS_STRICTATIME in lustre mounts.
291 * We override the kernel level default of MS_RELATIME for now
292 * due to a kernel vfs level bug in atime updates that fails
293 * to reset timestamps from the future.
295 #warn "remove MS_STRICTATIME override if kernel updates atime from the future"
297 /* set strictatime to default if NOATIME or RELATIME
298 not given explicit */
299 if (!(*flagp & (MS_NOATIME | MS_RELATIME)))
300 *flagp |= MS_STRICTATIME;
302 strcpy(orig_options, options);
308 /* Add mgsnids from ldd params */
309 static int add_mgsnids(struct mount_opts *mop, char *options,
312 char *ptr = (char *)params;
315 while ((ptr = strstr(ptr, PARAM_MGSNODE)) != NULL) {
316 sep = strchr(ptr, ' ');
321 append_option(options, ptr);
322 mop->mo_have_mgsnid++;
334 static int clear_update_ondisk(char *source, struct lustre_disk_data *ldd)
336 char wanted_mountopts[512] = "";
337 struct mkfs_opts mkop;
341 memset(&mkop, 0, sizeof(mkop));
343 mkop.mo_ldd.ldd_flags &= ~LDD_F_UPDATE;
344 mkop.mo_flags = MO_NOHOSTID_CHECK; /* Ignore missing hostid */
345 if (strlen(source) > sizeof(mkop.mo_device)-1) {
347 fprintf(stderr, "Device name too long: %s\n", source);
350 strncpy(mkop.mo_device, source, sizeof(mkop.mo_device));
352 ret = osd_prepare_lustre(&mkop,
353 wanted_mountopts, sizeof(wanted_mountopts));
356 fprintf(stderr, "Can't prepare device %s: %s\n",
357 source, strerror(ret));
361 /* Create the loopback file */
362 if (mkop.mo_flags & MO_IS_LOOP) {
363 ret = access(mkop.mo_device, F_OK);
367 fprintf(stderr, "Can't access device %s: %s\n",
368 source, strerror(ret));
372 ret = loop_setup(&mkop);
375 fprintf(stderr, "Loop device setup for %s failed: %s\n",
376 mkop.mo_device, strerror(ret));
380 ret = osd_write_ldd(&mkop);
383 fprintf(stderr, "failed to write local files: %s\n",
387 ret2 = loop_cleanup(&mkop);
394 static int parse_ldd(char *source, struct mount_opts *mop, char *options)
396 struct lustre_disk_data *ldd = &mop->mo_ldd;
400 rc = osd_is_lustre(source, &ldd->ldd_mount_type);
402 fprintf(stderr, "%s: %s has not been formatted with mkfs.lustre"
403 " or the backend filesystem type is not supported by "
404 "this tool\n", progname, source);
408 rc = osd_read_ldd(source, ldd);
410 fprintf(stderr, "%s: %s failed to read permanent mount"
411 " data: %s\n", progname, source,
412 rc >= 0 ? strerror(rc) : "");
416 if ((IS_MDT(ldd) || IS_OST(ldd)) &&
417 (ldd->ldd_flags & LDD_F_NEED_INDEX)) {
418 fprintf(stderr, "%s: %s has no index assigned "
419 "(probably formatted with old mkfs)\n",
424 if (ldd->ldd_flags & LDD_F_UPGRADE14) {
425 fprintf(stderr, "%s: we cannot upgrade %s from this (very old) "
426 "Lustre version\n", progname, source);
430 if (ldd->ldd_flags & LDD_F_UPDATE)
431 clear_update_ondisk(source, ldd);
433 /* Since we never rewrite ldd, ignore temp flags */
434 ldd->ldd_flags &= ~(LDD_F_VIRGIN | LDD_F_WRITECONF);
436 /* svname of the form lustre:OST1234 means never registered */
437 rc = strlen(ldd->ldd_svname);
438 if (strcmp(ldd->ldd_svname, "MGS") != 0) {
440 fprintf(stderr, "%s: invalid name '%s'\n",
441 progname, ldd->ldd_svname);
443 } else if (ldd->ldd_svname[rc - 8] == ':') {
444 ldd->ldd_svname[rc - 8] = '-';
445 ldd->ldd_flags |= LDD_F_VIRGIN;
446 } else if (ldd->ldd_svname[rc - 8] == '=') {
447 ldd->ldd_svname[rc - 8] = '-';
448 ldd->ldd_flags |= LDD_F_WRITECONF;
451 /* backend osd type */
452 append_option(options, "osd=");
453 strcat(options, mt_type(ldd->ldd_mount_type));
455 append_option(options, ldd->ldd_mount_opts);
457 if (!mop->mo_have_mgsnid) {
458 /* Only use disk data if mount -o mgsnode=nid wasn't
460 if (ldd->ldd_flags & LDD_F_SV_TYPE_MGS) {
461 append_option(options, "mgs");
462 mop->mo_have_mgsnid++;
464 add_mgsnids(mop, options, ldd->ldd_params);
467 /* Better have an mgsnid by now */
468 if (!mop->mo_have_mgsnid) {
469 fprintf(stderr, "%s: missing option mgsnode=<nid>\n",
474 if (ldd->ldd_flags & LDD_F_VIRGIN)
475 append_option(options, "virgin");
476 if (ldd->ldd_flags & LDD_F_UPDATE)
477 append_option(options, "update");
478 if (ldd->ldd_flags & LDD_F_WRITECONF)
479 append_option(options, "writeconf");
480 if (ldd->ldd_flags & LDD_F_NO_PRIMNODE)
481 append_option(options, "noprimnode");
483 /* prefix every lustre parameter with param= so that in-kernel
484 * mount can recognize them properly and send to MGS at registration */
485 start = ldd->ldd_params;
486 while (start && *start != '\0') {
487 while (*start == ' ') start++;
491 start = strchr(cur, ' ');
496 append_option(options, "param=");
497 strcat(options, cur);
500 /* svname must be last option */
501 append_option(options, "svname=");
502 strcat(options, ldd->ldd_svname);
507 static void set_defaults(struct mount_opts *mop)
509 memset(mop, 0, sizeof(*mop));
510 mop->mo_usource = NULL;
511 mop->mo_source = NULL;
516 mop->mo_have_mgsnid = 0;
517 mop->mo_md_stripe_cache_size = 16384;
518 mop->mo_orig_options = "";
522 static int parse_opts(int argc, char *const argv[], struct mount_opts *mop)
524 static struct option long_opt[] = {
528 {"nomtab", 0, 0, 'n'},
529 {"options", 1, 0, 'o'},
530 {"verbose", 0, 0, 'v'},
531 {"version", 0, 0, 'V'},
534 char real_path[PATH_MAX] = {'\0'};
536 char path[256], name[256];
541 while ((opt = getopt_long(argc, argv, "fhno:vV",
542 long_opt, NULL)) != EOF){
546 printf("force: %d\n", mop->mo_force);
550 printf("fake: %d\n", mop->mo_fake);
557 printf("nomtab: %d\n", mop->mo_nomtab);
560 mop->mo_orig_options = optarg;
567 fprintf(stdout, "%s %s %s\n", progname,
568 LUSTRE_VERSION_STRING, WITH_LIBMOUNT);
571 fprintf(stderr, "%s: unknown option '%c'\n",
578 if (optind + 2 > argc) {
579 fprintf(stderr, "%s: too few arguments\n", progname);
583 mop->mo_usource = argv[optind];
584 if (!mop->mo_usource) {
589 * Try to get the real path to the device, in case it is a
590 * symbolic link for instance
592 if (realpath(mop->mo_usource, real_path) != NULL) {
593 ptr = strrchr(real_path, '/');
594 if (ptr && strncmp(ptr, "/dm-", 4) == 0 && isdigit(*(ptr + 4))) {
595 snprintf(path, sizeof(path), "/sys/block/%s/dm/name", ptr+1);
596 if ((f = fopen(path, "r"))) {
597 /* read "<name>\n" from sysfs */
598 if (fgets(name, sizeof(name), f) && (sz = strlen(name)) > 1) {
600 snprintf(real_path, sizeof(real_path), "/dev/mapper/%s", name);
605 mop->mo_usource = strdup(real_path);
608 ptr = strstr(mop->mo_usource, ":/");
610 mop->mo_source = convert_hostnames(mop->mo_usource);
614 mop->mo_source = strdup(mop->mo_usource);
617 if (realpath(argv[optind + 1], mop->mo_target) == NULL) {
619 fprintf(stderr, "warning: %s: cannot resolve: %s\n",
620 argv[optind + 1], strerror(errno));
627 int main(int argc, char *const argv[])
629 struct mount_opts mop;
633 progname = strrchr(argv[0], '/');
634 progname = progname ? progname + 1 : argv[0];
638 rc = parse_opts(argc, argv, &mop);
643 for (i = 0; i < argc; i++)
644 printf("arg[%d] = %s\n", i, argv[i]);
645 printf("source = %s (%s), target = %s\n", mop.mo_usource,
646 mop.mo_source, mop.mo_target);
647 printf("options = %s\n", mop.mo_orig_options);
650 options = malloc(MAXOPT);
651 if (options == NULL) {
652 fprintf(stderr, "can't allocate memory for options\n");
655 strcpy(options, mop.mo_orig_options);
656 rc = parse_options(&mop, options, &flags);
658 fprintf(stderr, "%s: can't parse options: %s\n",
664 rc = check_mtab_entry(mop.mo_usource, mop.mo_source,
665 mop.mo_target, "lustre");
666 if (rc && !(flags & MS_REMOUNT)) {
667 fprintf(stderr, "%s: according to %s %s is "
668 "already mounted on %s\n", progname, MOUNTED,
669 mop.mo_usource, mop.mo_target);
672 if (!rc && (flags & MS_REMOUNT)) {
673 fprintf(stderr, "%s: according to %s %s is "
674 "not already mounted on %s\n", progname, MOUNTED,
675 mop.mo_usource, mop.mo_target);
679 if (flags & MS_REMOUNT)
682 rc = access(mop.mo_target, F_OK);
685 fprintf(stderr, "%s: %s inaccessible: %s\n", progname,
686 mop.mo_target, strerror(errno));
690 if (strstr(mop.mo_usource, ":/") == NULL) {
695 rc = parse_ldd(mop.mo_source, &mop, options);
700 /* In Linux 2.4, the target device doesn't get passed to any of our
701 functions. So we'll stick it on the end of the options. */
702 append_option(options, "device=");
703 strcat(options, mop.mo_source);
706 printf("mounting device %s at %s, flags=%#x options=%s\n",
707 mop.mo_source, mop.mo_target, flags, options);
709 if (strstr(mop.mo_usource, ":/") == NULL &&
710 osd_tune_lustre(mop.mo_source, &mop)) {
712 fprintf(stderr, "%s: unable to set tunables for %s"
713 " (may cause reduced IO performance)\n",
714 argv[0], mop.mo_source);
718 /* flags and target get to lustre_get_sb(), but not
719 * lustre_fill_super(). Lustre ignores the flags, but mount
721 for (i = 0, rc = -EAGAIN; i <= mop.mo_retry && rc != 0; i++) {
722 rc = mount(mop.mo_source, mop.mo_target, "lustre",
723 flags, (void *)options);
725 /* change label from <fsname>:<index> to
726 * <fsname>-<index> to indicate the device has
727 * been registered. only if the label is
728 * supposed to be changed and target service
729 * is supposed to start */
730 if (mop.mo_ldd.ldd_flags &
731 (LDD_F_VIRGIN | LDD_F_WRITECONF)) {
732 if (mop.mo_nosvc == 0)
733 (void)osd_label_lustre(&mop);
737 fprintf(stderr, "%s: mount %s at %s "
738 "failed: %s retries left: "
739 "%d\n", basename(progname),
740 mop.mo_usource, mop.mo_target,
746 int limit = i/2 > 5 ? i/2 : 5;
761 cli = strrchr(mop.mo_usource, ':');
762 if (cli && (strlen(cli) > 2))
767 fprintf(stderr, "%s: mount %s at %s failed: %s\n", progname,
768 mop.mo_usource, mop.mo_target, strerror(errno));
770 fprintf(stderr, "Is the backend filesystem mounted?\n"
771 "Check /etc/mtab and /proc/mounts\n");
773 fprintf(stderr, "Are the lustre modules loaded?\n"
774 "Check /etc/modprobe.conf and "
775 "/proc/filesystems\n");
776 if (errno == ENOTBLK)
777 fprintf(stderr, "Do you need -o loop?\n");
778 if (errno == ENOMEDIUM)
780 "This filesystem needs at least 1 OST\n");
781 if (errno == ENOENT) {
782 fprintf(stderr, "Is the MGS specification correct?\n");
783 fprintf(stderr, "Is the filesystem name correct?\n");
784 fprintf(stderr, "If upgrading, is the copied client log"
785 " valid? (see upgrade docs)\n");
787 if (errno == EALREADY)
788 fprintf(stderr, "The target service is already running."
789 " (%s)\n", mop.mo_usource);
791 fprintf(stderr, "The target service failed to start "
792 "(bad config log?) (%s). "
793 "See /var/log/messages.\n", mop.mo_usource);
795 fprintf(stderr, "Is the MGS running?\n");
796 if (errno == EADDRINUSE)
797 fprintf(stderr, "The target service's index is already "
798 "in use. (%s)\n", mop.mo_usource);
799 if (errno == EINVAL) {
800 fprintf(stderr, "This may have multiple causes.\n");
802 fprintf(stderr, "Is '%s' the correct filesystem"
804 fprintf(stderr, "Are the mount options correct?\n");
805 fprintf(stderr, "Check the syslog for more info.\n");
808 /* May as well try to clean up loop devs */
809 if (strncmp(mop.mo_usource, "/dev/loop", 9) == 0) {
812 sprintf(cmd, "/sbin/losetup -d %s", mop.mo_usource);
813 if ((ret = system(cmd)) < 0)
816 rc = WEXITSTATUS(ret);
820 /* Deal with utab just for client. Note that we ignore
821 * the return value here since it is not worth to fail
822 * mount by prevent some rare cases */
823 if (strstr(mop.mo_usource, ":/") != NULL)
824 update_utab_entry(&mop);
825 if (!mop.mo_nomtab) {
826 rc = update_mtab_entry(mop.mo_usource, mop.mo_target,
827 "lustre", mop.mo_orig_options,
833 /* mo_usource should be freed, but we can rely on the kernel */
836 if (strstr(mop.mo_usource, ":/") == NULL)