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, 2012, Whamcloud, Inc.
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"
53 #include <lustre_ver.h>
56 #if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(2, 10, 51, 0)
59 * We only #include a kernel level include file here because
60 * important MS_ flag #defines are missing from the SLES version
62 * In the future if SLES updates sys/mount.h to have a more complete
63 * set of flag #defines we should stop including linux/fs.h
65 #warn remove kernel include
71 #define MAX_RETRIES 99
74 char *progname = NULL;
78 fprintf(out, "%s v"LUSTRE_VERSION_STRING"\n", progname);
79 fprintf(out, "\nThis mount helper should only be invoked via the "
80 "mount (8) command,\ne.g. mount -t lustre dev dir\n\n");
81 fprintf(out, "usage: %s [-fhnv] [-o <mntopt>] <device> <mountpt>\n",
84 "\t<device>: the disk device, or for a client:\n"
85 "\t\t<mgmtnid>[:<altmgtnid>...]:/<filesystem>-client\n"
86 "\t<filesystem>: name of the Lustre filesystem (e.g. lustre1)\n"
87 "\t<mountpt>: filesystem mountpoint (e.g. /mnt/lustre)\n"
88 "\t-f|--fake: fake mount (updates /etc/mtab)\n"
89 "\t-o force|--force: force mount even if already in /etc/mtab\n"
90 "\t-h|--help: print this usage message\n"
91 "\t-n|--nomtab: do not update /etc/mtab after mount\n"
92 "\t-v|--verbose: print verbose config settings\n"
93 "\t<mntopt>: one or more comma separated of:\n"
94 "\t\t(no)flock,(no)user_xattr,(no)acl\n"
95 "\t\tabort_recov: abort server recovery handling\n"
96 "\t\tnosvc: only start MGC/MGS obds\n"
97 "\t\tnomgs: only start target obds, using existing MGS\n"
98 "\t\texclude=<ostname>[:<ostname>] : colon-separated list of "
99 "inactive OSTs (e.g. lustre-OST0001)\n"
100 "\t\tretry=<num>: number of times mount is retried by client\n"
101 "\t\tmd_stripe_cache_size=<num>: set the raid stripe cache "
102 "size for the underlying raid if present\n"
104 exit((out != stdout) ? EINVAL : 0);
107 /* Get rid of symbolic hostnames for tcp, since kernel can't do lookups */
108 #define MAXNIDSTR 1024
109 static char *convert_hostnames(char *s1)
111 char *converted, *s2 = 0, *c;
113 int left = MAXNIDSTR;
116 converted = malloc(left);
117 if (converted == NULL) {
118 fprintf(stderr, "out of memory: needed %d bytes\n",
123 while ((left > 0) && (*s1 != '/')) {
124 s2 = strpbrk(s1, ",:");
129 nid = libcfs_str2nid(s1);
130 *s2 = sep; /* back to original string */
131 if (nid == LNET_NID_ANY)
133 c += snprintf(c, left, "%s%c", libcfs_nid2str(nid), sep);
134 left = converted + MAXNIDSTR - c;
137 snprintf(c, left, "%s", s1);
140 fprintf(stderr, "%s: Can't parse NID '%s'\n", progname, s1);
145 /*****************************************************************************
147 * This part was cribbed from util-linux/mount/mount.c. There was no clear
148 * license information, but many other files in the package are identified as
149 * GNU GPL, so it's a pretty safe bet that was their intent.
151 ****************************************************************************/
153 const char *opt; /* option name */
154 int inv; /* true if flag value should be inverted */
155 int mask; /* flag mask value */
158 static const struct opt_map opt_map[] = {
159 /*"optname", inv,ms_mask */
160 /* These flags are parsed by mount, not lustre */
161 { "defaults", 0, 0 }, /* default options */
162 { "remount", 0, MS_REMOUNT}, /* remount with different options */
163 { "rw", 1, MS_RDONLY }, /* read-write */
164 { "ro", 0, MS_RDONLY }, /* read-only */
165 { "exec", 1, MS_NOEXEC }, /* permit execution of binaries */
166 { "noexec", 0, MS_NOEXEC }, /* don't execute binaries */
167 { "suid", 1, MS_NOSUID }, /* honor suid executables */
168 { "nosuid", 0, MS_NOSUID }, /* don't honor suid executables */
169 { "dev", 1, MS_NODEV }, /* interpret device files */
170 { "nodev", 0, MS_NODEV }, /* don't interpret devices */
171 { "sync", 0, MS_SYNCHRONOUS}, /* synchronous I/O */
172 { "async", 1, MS_SYNCHRONOUS}, /* asynchronous I/O */
173 { "atime", 1, MS_NOATIME }, /* set file access time on read */
174 { "noatime", 0, MS_NOATIME }, /* do not set file access time on read */
176 { "diratime", 1, MS_NODIRATIME }, /* set file access time on read */
177 { "nodiratime",0,MS_NODIRATIME }, /* do not set file access time on read */
180 { "relatime", 0, MS_RELATIME }, /* set file access time on read */
181 { "norelatime",1,MS_RELATIME }, /* do not set file access time on read */
183 #ifdef MS_STRICTATIME
184 { "strictatime",0,MS_STRICTATIME }, /* update access time strictly */
186 { "auto", 0, 0 }, /* Can be mounted using -a */
187 { "noauto", 0, 0 }, /* Can only be mounted explicitly */
188 { "nousers", 1, 0 }, /* Forbid ordinary user to mount */
189 { "nouser", 1, 0 }, /* Forbid ordinary user to mount */
190 { "noowner", 1, 0 }, /* Device owner has no special privs */
191 { "_netdev", 0, 0 }, /* Device accessible only via network */
195 /****************************************************************************/
197 /* 1 = don't pass on to lustre
198 0 = pass on to lustre */
199 static int parse_one_option(const char *check, int *flagp)
201 const struct opt_map *opt;
203 for (opt = &opt_map[0]; opt->opt != NULL; opt++) {
204 if (strncmp(check, opt->opt, strlen(opt->opt)) == 0) {
207 *flagp &= ~(opt->mask);
214 /* Assume any unknown options are valid and pass them on. The mount
215 will fail if lmd_parse, ll_options or ldiskfs doesn't recognize it.*/
219 static void append_option(char *options, const char *one)
222 strcat(options, ",");
223 strcat(options, one);
226 /* Replace options with subset of Lustre-specific options, and
227 fill in mount flags */
228 int parse_options(struct mount_opts *mop, char *orig_options, int *flagp)
230 char *options, *opt, *nextopt, *arg, *val;
232 options = calloc(strlen(orig_options) + 1, 1);
234 nextopt = orig_options;
235 while ((opt = strsep(&nextopt, ","))) {
240 /* Handle retries in a slightly different
243 val = strchr(opt, '=');
244 /* please note that some ldiskfs mount options are also in the form
245 * of param=value. We should pay attention not to remove those
246 * mount options, see bug 22097. */
247 if (val && strncmp(arg, "md_stripe_cache_size", 20) == 0) {
248 mop->mo_md_stripe_cache_size = atoi(val + 1);
249 } else if (val && strncmp(arg, "retry", 5) == 0) {
250 mop->mo_retry = atoi(val + 1);
251 if (mop->mo_retry > MAX_RETRIES)
252 mop->mo_retry = MAX_RETRIES;
253 else if (mop->mo_retry < 0)
255 } else if (val && strncmp(arg, "mgssec", 6) == 0) {
256 append_option(options, opt);
257 } else if (strncmp(arg, "nosvc", 5) == 0) {
259 append_option(options, opt);
260 } else if (strcmp(opt, "force") == 0) {
261 //XXX special check for 'force' option
263 printf("force: %d\n", mop->mo_force);
264 } else if (parse_one_option(opt, flagp) == 0) {
265 /* pass this on as an option */
266 append_option(options, opt);
269 #ifdef MS_STRICTATIME
270 #if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(2, 10, 51, 0)
273 * In the future when upstream fixes land in all supported kernels
274 * we should stop forcing MS_STRICTATIME in lustre mounts.
275 * We override the kernel level default of MS_RELATIME for now
276 * due to a kernel vfs level bug in atime updates that fails
277 * to reset timestamps from the future.
279 #warn remove MS_STRICTATIME override
281 /* set strictatime to default if NOATIME or RELATIME
282 not given explicit */
283 if (!(*flagp & (MS_NOATIME | MS_RELATIME)))
284 *flagp |= MS_STRICTATIME;
286 strcpy(orig_options, options);
291 /* Add mgsnids from ldd params */
292 static int add_mgsnids(struct mount_opts *mop, char *options,
295 char *ptr = (char *)params;
298 while ((ptr = strstr(ptr, PARAM_MGSNODE)) != NULL) {
299 sep = strchr(ptr, ' ');
304 append_option(options, ptr);
305 mop->mo_have_mgsnid++;
317 static int parse_ldd(char *source, struct mount_opts *mop, char *options)
319 struct lustre_disk_data *ldd = &mop->mo_ldd;
323 rc = osd_is_lustre(source, &ldd->ldd_mount_type);
325 fprintf(stderr, "%s: %s has not been formatted with mkfs.lustre"
326 " or the backend filesystem type is not supported by "
327 "this tool\n", progname, source);
331 rc = osd_read_ldd(source, ldd);
333 fprintf(stderr, "%s: %s failed to read permanent mount"
334 " data: %s\n", progname, source, strerror(rc));
338 if (ldd->ldd_flags & LDD_F_NEED_INDEX) {
339 fprintf(stderr, "%s: %s has no index assigned "
340 "(probably formatted with old mkfs)\n",
345 if (ldd->ldd_flags & LDD_F_UPGRADE14) {
346 fprintf(stderr, "%s: we cannot upgrade %s from this (very old) "
347 "Lustre version\n", progname, source);
351 /* Since we never rewrite ldd, ignore temp flags */
352 ldd->ldd_flags &= ~(LDD_F_VIRGIN | LDD_F_UPDATE | LDD_F_WRITECONF);
354 /* svname of the form lustre:OST1234 means never registered */
355 rc = strlen(ldd->ldd_svname);
356 if (ldd->ldd_svname[rc - 8] == ':') {
357 ldd->ldd_svname[rc - 8] = '-';
358 ldd->ldd_flags |= LDD_F_VIRGIN;
359 } else if (ldd->ldd_svname[rc - 8] == '=') {
360 ldd->ldd_svname[rc - 8] = '-';
361 ldd->ldd_flags |= LDD_F_WRITECONF;
364 /* backend osd type */
365 append_option(options, "osd=");
366 strcat(options, mt_type(ldd->ldd_mount_type));
368 append_option(options, ldd->ldd_mount_opts);
370 if (!mop->mo_have_mgsnid) {
371 /* Only use disk data if mount -o mgsnode=nid wasn't
373 if (ldd->ldd_flags & LDD_F_SV_TYPE_MGS) {
374 append_option(options, "mgs");
375 mop->mo_have_mgsnid++;
377 add_mgsnids(mop, options, ldd->ldd_params);
380 /* Better have an mgsnid by now */
381 if (!mop->mo_have_mgsnid) {
382 fprintf(stderr, "%s: missing option mgsnode=<nid>\n",
387 if (ldd->ldd_flags & LDD_F_VIRGIN)
388 append_option(options, "virgin");
389 if (ldd->ldd_flags & LDD_F_WRITECONF)
390 append_option(options, "writeconf");
391 if (ldd->ldd_flags & LDD_F_NO_PRIMNODE)
392 append_option(options, "noprimnode");
394 /* prefix every lustre parameter with param= so that in-kernel
395 * mount can recognize them properly and send to MGS at registration */
396 start = ldd->ldd_params;
397 while (start && *start != '\0') {
398 while (*start == ' ') start++;
402 start = strchr(cur, ' ');
407 append_option(options, "param=");
408 strcat(options, cur);
411 /* svname must be last option */
412 append_option(options, "svname=");
413 strcat(options, ldd->ldd_svname);
418 static void set_defaults(struct mount_opts *mop)
420 memset(mop, 0, sizeof(*mop));
421 mop->mo_usource = NULL;
422 mop->mo_source = NULL;
427 mop->mo_have_mgsnid = 0;
428 mop->mo_md_stripe_cache_size = 16384;
429 mop->mo_orig_options = "";
433 static int parse_opts(int argc, char *const argv[], struct mount_opts *mop)
435 static struct option long_opt[] = {
439 {"nomtab", 0, 0, 'n'},
440 {"options", 1, 0, 'o'},
441 {"verbose", 0, 0, 'v'},
444 char real_path[PATH_MAX] = {'\0'};
446 char path[256], name[256];
451 while ((opt = getopt_long(argc, argv, "fhno:v",
452 long_opt, NULL)) != EOF){
456 printf("force: %d\n", mop->mo_force);
460 printf("fake: %d\n", mop->mo_fake);
467 printf("nomtab: %d\n", mop->mo_nomtab);
470 mop->mo_orig_options = optarg;
476 fprintf(stderr, "%s: unknown option '%c'\n",
483 if (optind + 2 > argc) {
484 fprintf(stderr, "%s: too few arguments\n", progname);
488 mop->mo_usource = argv[optind];
489 if (!mop->mo_usource) {
494 * Try to get the real path to the device, in case it is a
495 * symbolic link for instance
497 if (realpath(mop->mo_usource, real_path) != NULL) {
498 ptr = strrchr(real_path, '/');
499 if (ptr && strncmp(ptr, "/dm-", 4) == 0 && isdigit(*(ptr + 4))) {
500 snprintf(path, sizeof(path), "/sys/block/%s/dm/name", ptr+1);
501 if ((f = fopen(path, "r"))) {
502 /* read "<name>\n" from sysfs */
503 if (fgets(name, sizeof(name), f) && (sz = strlen(name)) > 1) {
505 snprintf(real_path, sizeof(real_path), "/dev/mapper/%s", name);
510 mop->mo_usource = strdup(real_path);
513 ptr = strstr(mop->mo_usource, ":/");
515 mop->mo_source = convert_hostnames(mop->mo_usource);
519 mop->mo_source = strdup(mop->mo_usource);
522 if (realpath(argv[optind + 1], mop->mo_target) == NULL) {
524 fprintf(stderr, "warning: %s: cannot resolve: %s\n",
525 argv[optind + 1], strerror(errno));
532 int main(int argc, char *const argv[])
534 struct mount_opts mop;
538 progname = strrchr(argv[0], '/');
539 progname = progname ? progname + 1 : argv[0];
547 rc = parse_opts(argc, argv, &mop);
552 for (i = 0; i < argc; i++)
553 printf("arg[%d] = %s\n", i, argv[i]);
554 printf("source = %s (%s), target = %s\n", mop.mo_usource,
555 mop.mo_source, mop.mo_target);
556 printf("options = %s\n", mop.mo_orig_options);
559 options = malloc(MAXOPT);
560 if (options == NULL) {
561 fprintf(stderr, "can't allocate memory for options\n");
564 strcpy(options, mop.mo_orig_options);
565 rc = parse_options(&mop, options, &flags);
567 fprintf(stderr, "%s: can't parse options: %s\n",
573 rc = check_mtab_entry(mop.mo_usource, mop.mo_source,
574 mop.mo_target, "lustre");
575 if (rc && !(flags & MS_REMOUNT)) {
576 fprintf(stderr, "%s: according to %s %s is "
577 "already mounted on %s\n", progname, MOUNTED,
578 mop.mo_usource, mop.mo_target);
581 if (!rc && (flags & MS_REMOUNT)) {
582 fprintf(stderr, "%s: according to %s %s is "
583 "not already mounted on %s\n", progname, MOUNTED,
584 mop.mo_usource, mop.mo_target);
588 if (flags & MS_REMOUNT)
591 rc = access(mop.mo_target, F_OK);
594 fprintf(stderr, "%s: %s inaccessible: %s\n", progname,
595 mop.mo_target, strerror(errno));
599 if (!strstr(mop.mo_usource, ":/")) {
600 rc = parse_ldd(mop.mo_source, &mop, options);
605 /* In Linux 2.4, the target device doesn't get passed to any of our
606 functions. So we'll stick it on the end of the options. */
607 append_option(options, "device=");
608 strcat(options, mop.mo_source);
611 printf("mounting device %s at %s, flags=%#x options=%s\n",
612 mop.mo_source, mop.mo_target, flags, options);
614 if (!strstr(mop.mo_usource, ":/") &&
615 osd_tune_lustre(mop.mo_source, &mop)) {
617 fprintf(stderr, "%s: unable to set tunables for %s"
618 " (may cause reduced IO performance)\n",
619 argv[0], mop.mo_source);
623 /* flags and target get to lustre_get_sb, but not
624 lustre_fill_super. Lustre ignores the flags, but mount
626 for (i = 0, rc = -EAGAIN; i <= mop.mo_retry && rc != 0; i++) {
627 rc = mount(mop.mo_source, mop.mo_target, "lustre",
628 flags, (void *)options);
631 fprintf(stderr, "%s: mount %s at %s "
632 "failed: %s retries left: "
633 "%d\n", basename(progname),
634 mop.mo_usource, mop.mo_target,
640 sleep(1 << max((i/2), 5));
654 cli = strrchr(mop.mo_usource, ':');
655 if (cli && (strlen(cli) > 2))
660 fprintf(stderr, "%s: mount %s at %s failed: %s\n", progname,
661 mop.mo_usource, mop.mo_target, strerror(errno));
663 fprintf(stderr, "Is the backend filesystem mounted?\n"
664 "Check /etc/mtab and /proc/mounts\n");
666 fprintf(stderr, "Are the lustre modules loaded?\n"
667 "Check /etc/modprobe.conf and "
668 "/proc/filesystems\n");
669 if (errno == ENOTBLK)
670 fprintf(stderr, "Do you need -o loop?\n");
671 if (errno == ENOMEDIUM)
673 "This filesystem needs at least 1 OST\n");
674 if (errno == ENOENT) {
675 fprintf(stderr, "Is the MGS specification correct?\n");
676 fprintf(stderr, "Is the filesystem name correct?\n");
677 fprintf(stderr, "If upgrading, is the copied client log"
678 " valid? (see upgrade docs)\n");
680 if (errno == EALREADY)
681 fprintf(stderr, "The target service is already running."
682 " (%s)\n", mop.mo_usource);
684 fprintf(stderr, "The target service failed to start "
685 "(bad config log?) (%s). "
686 "See /var/log/messages.\n", mop.mo_usource);
688 fprintf(stderr, "Is the MGS running?\n");
689 if (errno == EADDRINUSE)
690 fprintf(stderr, "The target service's index is already "
691 "in use. (%s)\n", mop.mo_usource);
692 if (errno == EINVAL) {
693 fprintf(stderr, "This may have multiple causes.\n");
695 fprintf(stderr, "Is '%s' the correct filesystem"
697 fprintf(stderr, "Are the mount options correct?\n");
698 fprintf(stderr, "Check the syslog for more info.\n");
701 /* May as well try to clean up loop devs */
702 if (strncmp(mop.mo_usource, "/dev/loop", 9) == 0) {
705 sprintf(cmd, "/sbin/losetup -d %s", mop.mo_usource);
706 if ((ret = system(cmd)) < 0)
709 rc = WEXITSTATUS(ret);
712 } else if (!mop.mo_nomtab) {
713 rc = update_mtab_entry(mop.mo_usource, mop.mo_target, "lustre",
714 mop.mo_orig_options, 0,0,0);
716 /* change label from <fsname>:<index> to <fsname>-<index>
717 * to indicate the device has been registered.
718 * only if the label is supposed to be changed and
719 * target service is supposed to start */
720 if (mop.mo_ldd.ldd_flags & (LDD_F_VIRGIN | LDD_F_WRITECONF)) {
721 if (mop.mo_nosvc == 0 )
722 (void) osd_label_lustre(&mop);
727 /* mo_usource should be freed, but we can rely on the kernel */