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>
45 #include "mount_utils.h"
54 #include <sys/mount.h>
56 #include <lustre_ver.h>
59 #include <lnet/nidstr.h>
60 #include <lustre/lustre_idl.h>
61 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 10, 53, 0)
64 * We only #include a kernel level include file here because
65 * important MS_ flag #defines are missing from the SLES version
67 * In the future if SLES updates sys/mount.h to have a more complete
68 * set of flag #defines we should stop including linux/fs.h
70 #if !defined(MS_RDONLY)
76 #define MAX_RETRIES 99
84 fprintf(out, "\nThis mount helper should only be invoked via the "
85 "mount (8) command,\ne.g. mount -t lustre dev dir\n\n");
86 fprintf(out, "usage: %s [-fhnvV] [-o <mntopt>] <device> <mountpt>\n",
89 "\t<device>: the disk device, or for a client:\n"
90 "\t\t<mgmtnid>[:<altmgtnid>...]:/<filesystem>-client\n"
91 "\t<filesystem>: name of the Lustre filesystem (e.g. lustre1)\n"
92 "\t<mountpt>: filesystem mountpoint (e.g. /mnt/lustre)\n"
93 "\t-f|--fake: fake mount (updates /etc/mtab)\n"
94 "\t-o force|--force: force mount even if already in /etc/mtab\n"
95 "\t-h|--help: print this usage message\n"
96 "\t-n|--nomtab: do not update /etc/mtab after mount\n"
97 "\t-v|--verbose: print verbose config settings\n"
98 "\t-V|--version: output build version of the utility and exit\n"
99 "\t<mntopt>: one or more comma separated of:\n"
100 "\t\t(no)flock,(no)user_xattr,(no)acl\n"
101 "\t\tabort_recov: abort server recovery handling\n"
102 "\t\tnosvc: only start MGC/MGS obds\n"
103 "\t\tnomgs: only start target obds, using existing MGS\n"
104 "\t\texclude=<ostname>[:<ostname>] : colon-separated list of "
105 "inactive OSTs (e.g. lustre-OST0001)\n"
106 "\t\tretry=<num>: number of times mount is retried by client\n"
107 "\t\tmd_stripe_cache_size=<num>: set the raid stripe cache "
108 "size for the underlying raid if present\n");
109 exit((out != stdout) ? EINVAL : 0);
112 /* Get rid of symbolic hostnames for tcp, since kernel can't do lookups */
113 #define MAXNIDSTR 1024
114 static char *convert_hostnames(char *s1)
116 char *converted, *s2 = 0, *c;
118 int left = MAXNIDSTR;
121 converted = malloc(left);
122 if (converted == NULL) {
123 fprintf(stderr, "out of memory: needed %d bytes\n",
128 while ((left > 0) && (*s1 != '/')) {
129 s2 = strpbrk(s1, ",:");
134 nid = libcfs_str2nid(s1);
135 *s2 = sep; /* back to original string */
136 if (nid == LNET_NID_ANY)
138 c += snprintf(c, left, "%s%c", libcfs_nid2str(nid), sep);
139 left = converted + MAXNIDSTR - c;
142 snprintf(c, left, "%s", s1);
145 fprintf(stderr, "%s: Can't parse NID '%s'\n", progname, s1);
150 /*****************************************************************************
152 * This part was cribbed from util-linux/mount/mount.c. There was no clear
153 * license information, but many other files in the package are identified as
154 * GNU GPL, so it's a pretty safe bet that was their intent.
156 ****************************************************************************/
158 const char *opt; /* option name */
159 int inv; /* true if flag value should be inverted */
160 int mask; /* flag mask value */
163 static const struct opt_map opt_map[] = {
164 /*"optname", inv,ms_mask */
165 /* These flags are parsed by mount, not lustre */
166 { "defaults", 0, 0 }, /* default options */
167 { "remount", 0, MS_REMOUNT}, /* remount with different options */
168 { "rw", 1, MS_RDONLY }, /* read-write */
169 { "ro", 0, MS_RDONLY }, /* read-only */
170 { "exec", 1, MS_NOEXEC }, /* permit execution of binaries */
171 { "noexec", 0, MS_NOEXEC }, /* don't execute binaries */
172 { "suid", 1, MS_NOSUID }, /* honor suid executables */
173 { "nosuid", 0, MS_NOSUID }, /* don't honor suid executables */
174 { "dev", 1, MS_NODEV }, /* interpret device files */
175 { "nodev", 0, MS_NODEV }, /* don't interpret devices */
176 { "sync", 0, MS_SYNCHRONOUS}, /* synchronous I/O */
177 { "async", 1, MS_SYNCHRONOUS}, /* asynchronous I/O */
178 { "atime", 1, MS_NOATIME }, /* set file access time on read */
179 { "noatime", 0, MS_NOATIME }, /* do not set file access time on read */
181 { "diratime", 1, MS_NODIRATIME }, /* set file access time on read */
182 { "nodiratime",0,MS_NODIRATIME }, /* do not set file access time on read */
185 { "relatime", 0, MS_RELATIME }, /* set file access time on read */
186 { "norelatime",1,MS_RELATIME }, /* do not set file access time on read */
188 #ifdef MS_STRICTATIME
189 { "strictatime",0,MS_STRICTATIME }, /* update access time strictly */
191 { "auto", 0, 0 }, /* Can be mounted using -a */
192 { "noauto", 0, 0 }, /* Can only be mounted explicitly */
193 { "nousers", 1, 0 }, /* Forbid ordinary user to mount */
194 { "nouser", 1, 0 }, /* Forbid ordinary user to mount */
195 { "noowner", 1, 0 }, /* Device owner has no special privs */
196 { "_netdev", 0, 0 }, /* Device accessible only via network */
200 /****************************************************************************/
202 /* 1 = don't pass on to lustre
203 0 = pass on to lustre */
204 static int parse_one_option(const char *check, int *flagp)
206 const struct opt_map *opt;
208 for (opt = &opt_map[0]; opt->opt != NULL; opt++) {
209 if (strncmp(check, opt->opt, strlen(opt->opt)) == 0) {
212 *flagp &= ~(opt->mask);
219 /* Assume any unknown options are valid and pass them on. The mount
220 will fail if lmd_parse, ll_options or ldiskfs doesn't recognize it.*/
224 static void append_option(char *options, const char *one)
227 strcat(options, ",");
228 strcat(options, one);
231 /* Replace options with subset of Lustre-specific options, and
232 fill in mount flags */
233 int parse_options(struct mount_opts *mop, char *orig_options, int *flagp)
235 char *options, *opt, *nextopt, *arg, *val;
237 options = calloc(strlen(orig_options) + 1, 1);
239 nextopt = orig_options;
240 while ((opt = strsep(&nextopt, ","))) {
245 /* Handle retries in a slightly different
248 val = strchr(opt, '=');
249 /* please note that some ldiskfs mount options are also in the form
250 * of param=value. We should pay attention not to remove those
251 * mount options, see bug 22097. */
252 if (val && strncmp(arg, "md_stripe_cache_size", 20) == 0) {
253 mop->mo_md_stripe_cache_size = atoi(val + 1);
254 } else if (val && strncmp(arg, "retry", 5) == 0) {
255 mop->mo_retry = atoi(val + 1);
256 if (mop->mo_retry > MAX_RETRIES)
257 mop->mo_retry = MAX_RETRIES;
258 else if (mop->mo_retry < 0)
260 } else if (val && strncmp(arg, "mgssec", 6) == 0) {
261 append_option(options, opt);
262 } else if (strncmp(arg, "nosvc", 5) == 0) {
264 append_option(options, opt);
265 } else if (strcmp(opt, "force") == 0) {
266 /* XXX special check for 'force' option */
268 printf("force: %d\n", mop->mo_force);
269 } else if (parse_one_option(opt, flagp) == 0) {
270 /* pass this on as an option */
271 append_option(options, opt);
274 #ifdef MS_STRICTATIME
275 #if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(3, 2, 53, 0)
278 * In the future when upstream fixes land in all supported kernels
279 * we should stop forcing MS_STRICTATIME in lustre mounts.
280 * We override the kernel level default of MS_RELATIME for now
281 * due to a kernel vfs level bug in atime updates that fails
282 * to reset timestamps from the future.
284 #warn "remove MS_STRICTATIME override if kernel updates atime from the future"
286 /* set strictatime to default if NOATIME or RELATIME
287 not given explicit */
288 if (!(*flagp & (MS_NOATIME | MS_RELATIME)))
289 *flagp |= MS_STRICTATIME;
291 strcpy(orig_options, options);
297 /* Add mgsnids from ldd params */
298 static int add_mgsnids(struct mount_opts *mop, char *options,
301 char *ptr = (char *)params;
304 while ((ptr = strstr(ptr, PARAM_MGSNODE)) != NULL) {
305 sep = strchr(ptr, ' ');
310 append_option(options, ptr);
311 mop->mo_have_mgsnid++;
323 static int clear_update_ondisk(char *source, struct lustre_disk_data *ldd)
325 char always_mountopts[512] = "";
326 char default_mountopts[512] = "";
327 struct mkfs_opts mkop;
331 memset(&mkop, 0, sizeof(mkop));
333 mkop.mo_ldd.ldd_flags &= ~LDD_F_UPDATE;
334 if (strlen(source) > sizeof(mkop.mo_device)-1) {
336 fprintf(stderr, "Device name too long: %s\n", source);
339 strncpy(mkop.mo_device, source, sizeof(mkop.mo_device));
341 ret = osd_prepare_lustre(&mkop,
342 default_mountopts, sizeof(default_mountopts),
343 always_mountopts, sizeof(always_mountopts));
346 fprintf(stderr, "Can't prepare device %s: %s\n",
347 source, strerror(ret));
351 /* Create the loopback file */
352 if (mkop.mo_flags & MO_IS_LOOP) {
353 ret = access(mkop.mo_device, F_OK);
357 fprintf(stderr, "Can't access device %s: %s\n",
358 source, strerror(ret));
362 ret = loop_setup(&mkop);
365 fprintf(stderr, "Loop device setup for %s failed: %s\n",
366 mkop.mo_device, strerror(ret));
370 ret = osd_write_ldd(&mkop);
373 fprintf(stderr, "failed to write local files: %s\n",
377 ret2 = loop_cleanup(&mkop);
384 static int parse_ldd(char *source, struct mount_opts *mop, char *options)
386 struct lustre_disk_data *ldd = &mop->mo_ldd;
390 rc = osd_is_lustre(source, &ldd->ldd_mount_type);
392 fprintf(stderr, "%s: %s has not been formatted with mkfs.lustre"
393 " or the backend filesystem type is not supported by "
394 "this tool\n", progname, source);
398 rc = osd_read_ldd(source, ldd);
400 fprintf(stderr, "%s: %s failed to read permanent mount"
401 " data: %s\n", progname, source,
402 rc >= 0 ? strerror(rc) : "");
406 if ((IS_MDT(ldd) || IS_OST(ldd)) &&
407 (ldd->ldd_flags & LDD_F_NEED_INDEX)) {
408 fprintf(stderr, "%s: %s has no index assigned "
409 "(probably formatted with old mkfs)\n",
414 if (ldd->ldd_flags & LDD_F_UPGRADE14) {
415 fprintf(stderr, "%s: we cannot upgrade %s from this (very old) "
416 "Lustre version\n", progname, source);
420 if (ldd->ldd_flags & LDD_F_UPDATE)
421 clear_update_ondisk(source, ldd);
423 /* Since we never rewrite ldd, ignore temp flags */
424 ldd->ldd_flags &= ~(LDD_F_VIRGIN | LDD_F_WRITECONF);
426 /* svname of the form lustre:OST1234 means never registered */
427 rc = strlen(ldd->ldd_svname);
428 if (strcmp(ldd->ldd_svname, "MGS") != 0) {
430 fprintf(stderr, "%s: invalid name '%s'\n",
431 progname, ldd->ldd_svname);
433 } else if (ldd->ldd_svname[rc - 8] == ':') {
434 ldd->ldd_svname[rc - 8] = '-';
435 ldd->ldd_flags |= LDD_F_VIRGIN;
436 } else if (ldd->ldd_svname[rc - 8] == '=') {
437 ldd->ldd_svname[rc - 8] = '-';
438 ldd->ldd_flags |= LDD_F_WRITECONF;
441 /* backend osd type */
442 append_option(options, "osd=");
443 strcat(options, mt_type(ldd->ldd_mount_type));
445 append_option(options, ldd->ldd_mount_opts);
447 if (!mop->mo_have_mgsnid) {
448 /* Only use disk data if mount -o mgsnode=nid wasn't
450 if (ldd->ldd_flags & LDD_F_SV_TYPE_MGS) {
451 append_option(options, "mgs");
452 mop->mo_have_mgsnid++;
454 add_mgsnids(mop, options, ldd->ldd_params);
457 /* Better have an mgsnid by now */
458 if (!mop->mo_have_mgsnid) {
459 fprintf(stderr, "%s: missing option mgsnode=<nid>\n",
464 if (ldd->ldd_flags & LDD_F_VIRGIN)
465 append_option(options, "virgin");
466 if (ldd->ldd_flags & LDD_F_UPDATE)
467 append_option(options, "update");
468 if (ldd->ldd_flags & LDD_F_WRITECONF)
469 append_option(options, "writeconf");
470 if (ldd->ldd_flags & LDD_F_NO_PRIMNODE)
471 append_option(options, "noprimnode");
473 /* prefix every lustre parameter with param= so that in-kernel
474 * mount can recognize them properly and send to MGS at registration */
475 start = ldd->ldd_params;
476 while (start && *start != '\0') {
477 while (*start == ' ') start++;
481 start = strchr(cur, ' ');
486 append_option(options, "param=");
487 strcat(options, cur);
490 /* svname must be last option */
491 append_option(options, "svname=");
492 strcat(options, ldd->ldd_svname);
497 static void set_defaults(struct mount_opts *mop)
499 memset(mop, 0, sizeof(*mop));
500 mop->mo_usource = NULL;
501 mop->mo_source = NULL;
506 mop->mo_have_mgsnid = 0;
507 mop->mo_md_stripe_cache_size = 16384;
508 mop->mo_orig_options = "";
512 static int parse_opts(int argc, char *const argv[], struct mount_opts *mop)
514 static struct option long_opt[] = {
518 {"nomtab", 0, 0, 'n'},
519 {"options", 1, 0, 'o'},
520 {"verbose", 0, 0, 'v'},
521 {"version", 0, 0, 'V'},
524 char real_path[PATH_MAX] = {'\0'};
526 char path[256], name[256];
531 while ((opt = getopt_long(argc, argv, "fhno:vV",
532 long_opt, NULL)) != EOF){
536 printf("force: %d\n", mop->mo_force);
540 printf("fake: %d\n", mop->mo_fake);
547 printf("nomtab: %d\n", mop->mo_nomtab);
550 mop->mo_orig_options = optarg;
557 fprintf(stdout, "%s %s\n", progname,
558 LUSTRE_VERSION_STRING);
561 fprintf(stderr, "%s: unknown option '%c'\n",
568 if (optind + 2 > argc) {
569 fprintf(stderr, "%s: too few arguments\n", progname);
573 mop->mo_usource = argv[optind];
574 if (!mop->mo_usource) {
579 * Try to get the real path to the device, in case it is a
580 * symbolic link for instance
582 if (realpath(mop->mo_usource, real_path) != NULL) {
583 ptr = strrchr(real_path, '/');
584 if (ptr && strncmp(ptr, "/dm-", 4) == 0 && isdigit(*(ptr + 4))) {
585 snprintf(path, sizeof(path), "/sys/block/%s/dm/name", ptr+1);
586 if ((f = fopen(path, "r"))) {
587 /* read "<name>\n" from sysfs */
588 if (fgets(name, sizeof(name), f) && (sz = strlen(name)) > 1) {
590 snprintf(real_path, sizeof(real_path), "/dev/mapper/%s", name);
595 mop->mo_usource = strdup(real_path);
598 ptr = strstr(mop->mo_usource, ":/");
600 mop->mo_source = convert_hostnames(mop->mo_usource);
604 mop->mo_source = strdup(mop->mo_usource);
607 if (realpath(argv[optind + 1], mop->mo_target) == NULL) {
609 fprintf(stderr, "warning: %s: cannot resolve: %s\n",
610 argv[optind + 1], strerror(errno));
617 int main(int argc, char *const argv[])
619 struct mount_opts mop;
623 progname = strrchr(argv[0], '/');
624 progname = progname ? progname + 1 : argv[0];
632 rc = parse_opts(argc, argv, &mop);
637 for (i = 0; i < argc; i++)
638 printf("arg[%d] = %s\n", i, argv[i]);
639 printf("source = %s (%s), target = %s\n", mop.mo_usource,
640 mop.mo_source, mop.mo_target);
641 printf("options = %s\n", mop.mo_orig_options);
644 options = malloc(MAXOPT);
645 if (options == NULL) {
646 fprintf(stderr, "can't allocate memory for options\n");
649 strcpy(options, mop.mo_orig_options);
650 rc = parse_options(&mop, options, &flags);
652 fprintf(stderr, "%s: can't parse options: %s\n",
658 rc = check_mtab_entry(mop.mo_usource, mop.mo_source,
659 mop.mo_target, "lustre");
660 if (rc && !(flags & MS_REMOUNT)) {
661 fprintf(stderr, "%s: according to %s %s is "
662 "already mounted on %s\n", progname, MOUNTED,
663 mop.mo_usource, mop.mo_target);
666 if (!rc && (flags & MS_REMOUNT)) {
667 fprintf(stderr, "%s: according to %s %s is "
668 "not already mounted on %s\n", progname, MOUNTED,
669 mop.mo_usource, mop.mo_target);
673 if (flags & MS_REMOUNT)
676 rc = access(mop.mo_target, F_OK);
679 fprintf(stderr, "%s: %s inaccessible: %s\n", progname,
680 mop.mo_target, strerror(errno));
684 if (!strstr(mop.mo_usource, ":/")) {
685 rc = parse_ldd(mop.mo_source, &mop, options);
690 /* In Linux 2.4, the target device doesn't get passed to any of our
691 functions. So we'll stick it on the end of the options. */
692 append_option(options, "device=");
693 strcat(options, mop.mo_source);
696 printf("mounting device %s at %s, flags=%#x options=%s\n",
697 mop.mo_source, mop.mo_target, flags, options);
699 if (!strstr(mop.mo_usource, ":/") &&
700 osd_tune_lustre(mop.mo_source, &mop)) {
702 fprintf(stderr, "%s: unable to set tunables for %s"
703 " (may cause reduced IO performance)\n",
704 argv[0], mop.mo_source);
708 /* flags and target get to lustre_get_sb, but not
709 lustre_fill_super. Lustre ignores the flags, but mount
711 for (i = 0, rc = -EAGAIN; i <= mop.mo_retry && rc != 0; i++) {
712 rc = mount(mop.mo_source, mop.mo_target, "lustre",
713 flags, (void *)options);
715 /* change label from <fsname>:<index> to
716 * <fsname>-<index> to indicate the device has
717 * been registered. only if the label is
718 * supposed to be changed and target service
719 * is supposed to start */
720 if (mop.mo_ldd.ldd_flags &
721 (LDD_F_VIRGIN | LDD_F_WRITECONF)) {
722 if (mop.mo_nosvc == 0)
723 (void)osd_label_lustre(&mop);
727 fprintf(stderr, "%s: mount %s at %s "
728 "failed: %s retries left: "
729 "%d\n", basename(progname),
730 mop.mo_usource, mop.mo_target,
736 sleep(1 << max((i/2), 5));
750 cli = strrchr(mop.mo_usource, ':');
751 if (cli && (strlen(cli) > 2))
756 fprintf(stderr, "%s: mount %s at %s failed: %s\n", progname,
757 mop.mo_usource, mop.mo_target, strerror(errno));
759 fprintf(stderr, "Is the backend filesystem mounted?\n"
760 "Check /etc/mtab and /proc/mounts\n");
762 fprintf(stderr, "Are the lustre modules loaded?\n"
763 "Check /etc/modprobe.conf and "
764 "/proc/filesystems\n");
765 if (errno == ENOTBLK)
766 fprintf(stderr, "Do you need -o loop?\n");
767 if (errno == ENOMEDIUM)
769 "This filesystem needs at least 1 OST\n");
770 if (errno == ENOENT) {
771 fprintf(stderr, "Is the MGS specification correct?\n");
772 fprintf(stderr, "Is the filesystem name correct?\n");
773 fprintf(stderr, "If upgrading, is the copied client log"
774 " valid? (see upgrade docs)\n");
776 if (errno == EALREADY)
777 fprintf(stderr, "The target service is already running."
778 " (%s)\n", mop.mo_usource);
780 fprintf(stderr, "The target service failed to start "
781 "(bad config log?) (%s). "
782 "See /var/log/messages.\n", mop.mo_usource);
784 fprintf(stderr, "Is the MGS running?\n");
785 if (errno == EADDRINUSE)
786 fprintf(stderr, "The target service's index is already "
787 "in use. (%s)\n", mop.mo_usource);
788 if (errno == EINVAL) {
789 fprintf(stderr, "This may have multiple causes.\n");
791 fprintf(stderr, "Is '%s' the correct filesystem"
793 fprintf(stderr, "Are the mount options correct?\n");
794 fprintf(stderr, "Check the syslog for more info.\n");
797 /* May as well try to clean up loop devs */
798 if (strncmp(mop.mo_usource, "/dev/loop", 9) == 0) {
801 sprintf(cmd, "/sbin/losetup -d %s", mop.mo_usource);
802 if ((ret = system(cmd)) < 0)
805 rc = WEXITSTATUS(ret);
808 } else if (!mop.mo_nomtab) {
809 rc = update_mtab_entry(mop.mo_usource, mop.mo_target, "lustre",
810 mop.mo_orig_options, 0,0,0);
814 /* mo_usource should be freed, but we can rely on the kernel */