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, 2013, 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"
53 #include <lustre_ver.h>
56 #include <lnet/nidstr.h>
57 #include <lustre/lustre_idl.h>
58 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 10, 53, 0)
61 * We only #include a kernel level include file here because
62 * important MS_ flag #defines are missing from the SLES version
64 * In the future if SLES updates sys/mount.h to have a more complete
65 * set of flag #defines we should stop including linux/fs.h
67 #if !defined(MS_RDONLY)
73 #define MAX_RETRIES 99
81 fprintf(out, "\nThis mount helper should only be invoked via the "
82 "mount (8) command,\ne.g. mount -t lustre dev dir\n\n");
83 fprintf(out, "usage: %s [-fhnvV] [-o <mntopt>] <device> <mountpt>\n",
86 "\t<device>: the disk device, or for a client:\n"
87 "\t\t<mgmtnid>[:<altmgtnid>...]:/<filesystem>-client\n"
88 "\t<filesystem>: name of the Lustre filesystem (e.g. lustre1)\n"
89 "\t<mountpt>: filesystem mountpoint (e.g. /mnt/lustre)\n"
90 "\t-f|--fake: fake mount (updates /etc/mtab)\n"
91 "\t-o force|--force: force mount even if already in /etc/mtab\n"
92 "\t-h|--help: print this usage message\n"
93 "\t-n|--nomtab: do not update /etc/mtab after mount\n"
94 "\t-v|--verbose: print verbose config settings\n"
95 "\t-V|--version: output build version of the utility and exit\n"
96 "\t<mntopt>: one or more comma separated of:\n"
97 "\t\t(no)flock,(no)user_xattr,(no)acl\n"
98 "\t\tabort_recov: abort server recovery handling\n"
99 "\t\tnosvc: only start MGC/MGS obds\n"
100 "\t\tnomgs: only start target obds, using existing MGS\n"
101 "\t\texclude=<ostname>[:<ostname>] : colon-separated list of "
102 "inactive OSTs (e.g. lustre-OST0001)\n"
103 "\t\tretry=<num>: number of times mount is retried by client\n"
104 "\t\tmd_stripe_cache_size=<num>: set the raid stripe cache "
105 "size for the underlying raid if present\n");
106 exit((out != stdout) ? EINVAL : 0);
109 /* Get rid of symbolic hostnames for tcp, since kernel can't do lookups */
110 #define MAXNIDSTR 1024
111 static char *convert_hostnames(char *s1)
113 char *converted, *s2 = 0, *c;
115 int left = MAXNIDSTR;
118 converted = malloc(left);
119 if (converted == NULL) {
120 fprintf(stderr, "out of memory: needed %d bytes\n",
125 while ((left > 0) && (*s1 != '/')) {
126 s2 = strpbrk(s1, ",:");
131 nid = libcfs_str2nid(s1);
132 *s2 = sep; /* back to original string */
133 if (nid == LNET_NID_ANY)
135 c += snprintf(c, left, "%s%c", libcfs_nid2str(nid), sep);
136 left = converted + MAXNIDSTR - c;
139 snprintf(c, left, "%s", s1);
142 fprintf(stderr, "%s: Can't parse NID '%s'\n", progname, s1);
147 /*****************************************************************************
149 * This part was cribbed from util-linux/mount/mount.c. There was no clear
150 * license information, but many other files in the package are identified as
151 * GNU GPL, so it's a pretty safe bet that was their intent.
153 ****************************************************************************/
155 const char *opt; /* option name */
156 int inv; /* true if flag value should be inverted */
157 int mask; /* flag mask value */
160 static const struct opt_map opt_map[] = {
161 /*"optname", inv,ms_mask */
162 /* These flags are parsed by mount, not lustre */
163 { "defaults", 0, 0 }, /* default options */
164 { "remount", 0, MS_REMOUNT}, /* remount with different options */
165 { "rw", 1, MS_RDONLY }, /* read-write */
166 { "ro", 0, MS_RDONLY }, /* read-only */
167 { "exec", 1, MS_NOEXEC }, /* permit execution of binaries */
168 { "noexec", 0, MS_NOEXEC }, /* don't execute binaries */
169 { "suid", 1, MS_NOSUID }, /* honor suid executables */
170 { "nosuid", 0, MS_NOSUID }, /* don't honor suid executables */
171 { "dev", 1, MS_NODEV }, /* interpret device files */
172 { "nodev", 0, MS_NODEV }, /* don't interpret devices */
173 { "sync", 0, MS_SYNCHRONOUS}, /* synchronous I/O */
174 { "async", 1, MS_SYNCHRONOUS}, /* asynchronous I/O */
175 { "atime", 1, MS_NOATIME }, /* set file access time on read */
176 { "noatime", 0, MS_NOATIME }, /* do not set file access time on read */
178 { "diratime", 1, MS_NODIRATIME }, /* set file access time on read */
179 { "nodiratime",0,MS_NODIRATIME }, /* do not set file access time on read */
182 { "relatime", 0, MS_RELATIME }, /* set file access time on read */
183 { "norelatime",1,MS_RELATIME }, /* do not set file access time on read */
185 #ifdef MS_STRICTATIME
186 { "strictatime",0,MS_STRICTATIME }, /* update access time strictly */
188 { "auto", 0, 0 }, /* Can be mounted using -a */
189 { "noauto", 0, 0 }, /* Can only be mounted explicitly */
190 { "nousers", 1, 0 }, /* Forbid ordinary user to mount */
191 { "nouser", 1, 0 }, /* Forbid ordinary user to mount */
192 { "noowner", 1, 0 }, /* Device owner has no special privs */
193 { "_netdev", 0, 0 }, /* Device accessible only via network */
197 /****************************************************************************/
199 /* 1 = don't pass on to lustre
200 0 = pass on to lustre */
201 static int parse_one_option(const char *check, int *flagp)
203 const struct opt_map *opt;
205 for (opt = &opt_map[0]; opt->opt != NULL; opt++) {
206 if (strncmp(check, opt->opt, strlen(opt->opt)) == 0) {
209 *flagp &= ~(opt->mask);
216 /* Assume any unknown options are valid and pass them on. The mount
217 will fail if lmd_parse, ll_options or ldiskfs doesn't recognize it.*/
221 static void append_option(char *options, const char *one)
224 strcat(options, ",");
225 strcat(options, one);
228 /* Replace options with subset of Lustre-specific options, and
229 fill in mount flags */
230 int parse_options(struct mount_opts *mop, char *orig_options, int *flagp)
232 char *options, *opt, *nextopt, *arg, *val;
234 options = calloc(strlen(orig_options) + 1, 1);
236 nextopt = orig_options;
237 while ((opt = strsep(&nextopt, ","))) {
242 /* Handle retries in a slightly different
245 val = strchr(opt, '=');
246 /* please note that some ldiskfs mount options are also in the form
247 * of param=value. We should pay attention not to remove those
248 * mount options, see bug 22097. */
249 if (val && strncmp(arg, "md_stripe_cache_size", 20) == 0) {
250 mop->mo_md_stripe_cache_size = atoi(val + 1);
251 } else if (val && strncmp(arg, "retry", 5) == 0) {
252 mop->mo_retry = atoi(val + 1);
253 if (mop->mo_retry > MAX_RETRIES)
254 mop->mo_retry = MAX_RETRIES;
255 else if (mop->mo_retry < 0)
257 } else if (val && strncmp(arg, "mgssec", 6) == 0) {
258 append_option(options, opt);
259 } else if (strncmp(arg, "nosvc", 5) == 0) {
261 append_option(options, opt);
262 } else if (strcmp(opt, "force") == 0) {
263 /* XXX special check for 'force' option */
265 printf("force: %d\n", mop->mo_force);
266 } else if (parse_one_option(opt, flagp) == 0) {
267 /* pass this on as an option */
268 append_option(options, opt);
271 #ifdef MS_STRICTATIME
272 #if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(3, 2, 53, 0)
275 * In the future when upstream fixes land in all supported kernels
276 * we should stop forcing MS_STRICTATIME in lustre mounts.
277 * We override the kernel level default of MS_RELATIME for now
278 * due to a kernel vfs level bug in atime updates that fails
279 * to reset timestamps from the future.
281 #warn "remove MS_STRICTATIME override if kernel updates atime from the future"
283 /* set strictatime to default if NOATIME or RELATIME
284 not given explicit */
285 if (!(*flagp & (MS_NOATIME | MS_RELATIME)))
286 *flagp |= MS_STRICTATIME;
288 strcpy(orig_options, options);
294 /* Add mgsnids from ldd params */
295 static int add_mgsnids(struct mount_opts *mop, char *options,
298 char *ptr = (char *)params;
301 while ((ptr = strstr(ptr, PARAM_MGSNODE)) != NULL) {
302 sep = strchr(ptr, ' ');
307 append_option(options, ptr);
308 mop->mo_have_mgsnid++;
320 static int clear_update_ondisk(char *source, struct lustre_disk_data *ldd)
322 char always_mountopts[512] = "";
323 char default_mountopts[512] = "";
324 struct mkfs_opts mkop;
328 memset(&mkop, 0, sizeof(mkop));
330 mkop.mo_ldd.ldd_flags &= ~LDD_F_UPDATE;
331 if (strlen(source) > sizeof(mkop.mo_device)-1) {
333 fprintf(stderr, "Device name too long: %s\n", source);
336 strncpy(mkop.mo_device, source, sizeof(mkop.mo_device));
338 ret = osd_prepare_lustre(&mkop,
339 default_mountopts, sizeof(default_mountopts),
340 always_mountopts, sizeof(always_mountopts));
343 fprintf(stderr, "Can't prepare device %s: %s\n",
344 source, strerror(ret));
348 /* Create the loopback file */
349 if (mkop.mo_flags & MO_IS_LOOP) {
350 ret = access(mkop.mo_device, F_OK);
354 fprintf(stderr, "Can't access device %s: %s\n",
355 source, strerror(ret));
359 ret = loop_setup(&mkop);
362 fprintf(stderr, "Loop device setup for %s failed: %s\n",
363 mkop.mo_device, strerror(ret));
367 ret = osd_write_ldd(&mkop);
370 fprintf(stderr, "failed to write local files: %s\n",
374 ret2 = loop_cleanup(&mkop);
381 static int parse_ldd(char *source, struct mount_opts *mop, char *options)
383 struct lustre_disk_data *ldd = &mop->mo_ldd;
387 rc = osd_is_lustre(source, &ldd->ldd_mount_type);
389 fprintf(stderr, "%s: %s has not been formatted with mkfs.lustre"
390 " or the backend filesystem type is not supported by "
391 "this tool\n", progname, source);
395 rc = osd_read_ldd(source, ldd);
397 fprintf(stderr, "%s: %s failed to read permanent mount"
398 " data: %s\n", progname, source,
399 rc >= 0 ? strerror(rc) : "");
403 if ((IS_MDT(ldd) || IS_OST(ldd)) &&
404 (ldd->ldd_flags & LDD_F_NEED_INDEX)) {
405 fprintf(stderr, "%s: %s has no index assigned "
406 "(probably formatted with old mkfs)\n",
411 if (ldd->ldd_flags & LDD_F_UPGRADE14) {
412 fprintf(stderr, "%s: we cannot upgrade %s from this (very old) "
413 "Lustre version\n", progname, source);
417 if (ldd->ldd_flags & LDD_F_UPDATE)
418 clear_update_ondisk(source, ldd);
420 /* Since we never rewrite ldd, ignore temp flags */
421 ldd->ldd_flags &= ~(LDD_F_VIRGIN | LDD_F_WRITECONF);
423 /* svname of the form lustre:OST1234 means never registered */
424 rc = strlen(ldd->ldd_svname);
425 if (ldd->ldd_svname[rc - 8] == ':') {
426 ldd->ldd_svname[rc - 8] = '-';
427 ldd->ldd_flags |= LDD_F_VIRGIN;
428 } else if (ldd->ldd_svname[rc - 8] == '=') {
429 ldd->ldd_svname[rc - 8] = '-';
430 ldd->ldd_flags |= LDD_F_WRITECONF;
433 /* backend osd type */
434 append_option(options, "osd=");
435 strcat(options, mt_type(ldd->ldd_mount_type));
437 append_option(options, ldd->ldd_mount_opts);
439 if (!mop->mo_have_mgsnid) {
440 /* Only use disk data if mount -o mgsnode=nid wasn't
442 if (ldd->ldd_flags & LDD_F_SV_TYPE_MGS) {
443 append_option(options, "mgs");
444 mop->mo_have_mgsnid++;
446 add_mgsnids(mop, options, ldd->ldd_params);
449 /* Better have an mgsnid by now */
450 if (!mop->mo_have_mgsnid) {
451 fprintf(stderr, "%s: missing option mgsnode=<nid>\n",
456 if (ldd->ldd_flags & LDD_F_VIRGIN)
457 append_option(options, "virgin");
458 if (ldd->ldd_flags & LDD_F_UPDATE)
459 append_option(options, "update");
460 if (ldd->ldd_flags & LDD_F_WRITECONF)
461 append_option(options, "writeconf");
462 if (ldd->ldd_flags & LDD_F_NO_PRIMNODE)
463 append_option(options, "noprimnode");
465 /* prefix every lustre parameter with param= so that in-kernel
466 * mount can recognize them properly and send to MGS at registration */
467 start = ldd->ldd_params;
468 while (start && *start != '\0') {
469 while (*start == ' ') start++;
473 start = strchr(cur, ' ');
478 append_option(options, "param=");
479 strcat(options, cur);
482 /* svname must be last option */
483 append_option(options, "svname=");
484 strcat(options, ldd->ldd_svname);
489 static void set_defaults(struct mount_opts *mop)
491 memset(mop, 0, sizeof(*mop));
492 mop->mo_usource = NULL;
493 mop->mo_source = NULL;
498 mop->mo_have_mgsnid = 0;
499 mop->mo_md_stripe_cache_size = 16384;
500 mop->mo_orig_options = "";
504 static int parse_opts(int argc, char *const argv[], struct mount_opts *mop)
506 static struct option long_opt[] = {
510 {"nomtab", 0, 0, 'n'},
511 {"options", 1, 0, 'o'},
512 {"verbose", 0, 0, 'v'},
513 {"version", 0, 0, 'V'},
516 char real_path[PATH_MAX] = {'\0'};
518 char path[256], name[256];
523 while ((opt = getopt_long(argc, argv, "fhno:vV",
524 long_opt, NULL)) != EOF){
528 printf("force: %d\n", mop->mo_force);
532 printf("fake: %d\n", mop->mo_fake);
539 printf("nomtab: %d\n", mop->mo_nomtab);
542 mop->mo_orig_options = optarg;
549 fprintf(stdout, "%s %s\n", progname,
550 LUSTRE_VERSION_STRING);
553 fprintf(stderr, "%s: unknown option '%c'\n",
560 if (optind + 2 > argc) {
561 fprintf(stderr, "%s: too few arguments\n", progname);
565 mop->mo_usource = argv[optind];
566 if (!mop->mo_usource) {
571 * Try to get the real path to the device, in case it is a
572 * symbolic link for instance
574 if (realpath(mop->mo_usource, real_path) != NULL) {
575 ptr = strrchr(real_path, '/');
576 if (ptr && strncmp(ptr, "/dm-", 4) == 0 && isdigit(*(ptr + 4))) {
577 snprintf(path, sizeof(path), "/sys/block/%s/dm/name", ptr+1);
578 if ((f = fopen(path, "r"))) {
579 /* read "<name>\n" from sysfs */
580 if (fgets(name, sizeof(name), f) && (sz = strlen(name)) > 1) {
582 snprintf(real_path, sizeof(real_path), "/dev/mapper/%s", name);
587 mop->mo_usource = strdup(real_path);
590 ptr = strstr(mop->mo_usource, ":/");
592 mop->mo_source = convert_hostnames(mop->mo_usource);
596 mop->mo_source = strdup(mop->mo_usource);
599 if (realpath(argv[optind + 1], mop->mo_target) == NULL) {
601 fprintf(stderr, "warning: %s: cannot resolve: %s\n",
602 argv[optind + 1], strerror(errno));
609 int main(int argc, char *const argv[])
611 struct mount_opts mop;
615 progname = strrchr(argv[0], '/');
616 progname = progname ? progname + 1 : argv[0];
624 rc = parse_opts(argc, argv, &mop);
629 for (i = 0; i < argc; i++)
630 printf("arg[%d] = %s\n", i, argv[i]);
631 printf("source = %s (%s), target = %s\n", mop.mo_usource,
632 mop.mo_source, mop.mo_target);
633 printf("options = %s\n", mop.mo_orig_options);
636 options = malloc(MAXOPT);
637 if (options == NULL) {
638 fprintf(stderr, "can't allocate memory for options\n");
641 strcpy(options, mop.mo_orig_options);
642 rc = parse_options(&mop, options, &flags);
644 fprintf(stderr, "%s: can't parse options: %s\n",
650 rc = check_mtab_entry(mop.mo_usource, mop.mo_source,
651 mop.mo_target, "lustre");
652 if (rc && !(flags & MS_REMOUNT)) {
653 fprintf(stderr, "%s: according to %s %s is "
654 "already mounted on %s\n", progname, MOUNTED,
655 mop.mo_usource, mop.mo_target);
658 if (!rc && (flags & MS_REMOUNT)) {
659 fprintf(stderr, "%s: according to %s %s is "
660 "not already mounted on %s\n", progname, MOUNTED,
661 mop.mo_usource, mop.mo_target);
665 if (flags & MS_REMOUNT)
668 rc = access(mop.mo_target, F_OK);
671 fprintf(stderr, "%s: %s inaccessible: %s\n", progname,
672 mop.mo_target, strerror(errno));
676 if (!strstr(mop.mo_usource, ":/")) {
677 rc = parse_ldd(mop.mo_source, &mop, options);
682 /* In Linux 2.4, the target device doesn't get passed to any of our
683 functions. So we'll stick it on the end of the options. */
684 append_option(options, "device=");
685 strcat(options, mop.mo_source);
688 printf("mounting device %s at %s, flags=%#x options=%s\n",
689 mop.mo_source, mop.mo_target, flags, options);
691 if (!strstr(mop.mo_usource, ":/") &&
692 osd_tune_lustre(mop.mo_source, &mop)) {
694 fprintf(stderr, "%s: unable to set tunables for %s"
695 " (may cause reduced IO performance)\n",
696 argv[0], mop.mo_source);
700 /* flags and target get to lustre_get_sb, but not
701 lustre_fill_super. Lustre ignores the flags, but mount
703 for (i = 0, rc = -EAGAIN; i <= mop.mo_retry && rc != 0; i++) {
704 rc = mount(mop.mo_source, mop.mo_target, "lustre",
705 flags, (void *)options);
707 /* change label from <fsname>:<index> to
708 * <fsname>-<index> to indicate the device has
709 * been registered. only if the label is
710 * supposed to be changed and target service
711 * is supposed to start */
712 if (mop.mo_ldd.ldd_flags &
713 (LDD_F_VIRGIN | LDD_F_WRITECONF)) {
714 if (mop.mo_nosvc == 0)
715 (void)osd_label_lustre(&mop);
719 fprintf(stderr, "%s: mount %s at %s "
720 "failed: %s retries left: "
721 "%d\n", basename(progname),
722 mop.mo_usource, mop.mo_target,
728 sleep(1 << max((i/2), 5));
742 cli = strrchr(mop.mo_usource, ':');
743 if (cli && (strlen(cli) > 2))
748 fprintf(stderr, "%s: mount %s at %s failed: %s\n", progname,
749 mop.mo_usource, mop.mo_target, strerror(errno));
751 fprintf(stderr, "Is the backend filesystem mounted?\n"
752 "Check /etc/mtab and /proc/mounts\n");
754 fprintf(stderr, "Are the lustre modules loaded?\n"
755 "Check /etc/modprobe.conf and "
756 "/proc/filesystems\n");
757 if (errno == ENOTBLK)
758 fprintf(stderr, "Do you need -o loop?\n");
759 if (errno == ENOMEDIUM)
761 "This filesystem needs at least 1 OST\n");
762 if (errno == ENOENT) {
763 fprintf(stderr, "Is the MGS specification correct?\n");
764 fprintf(stderr, "Is the filesystem name correct?\n");
765 fprintf(stderr, "If upgrading, is the copied client log"
766 " valid? (see upgrade docs)\n");
768 if (errno == EALREADY)
769 fprintf(stderr, "The target service is already running."
770 " (%s)\n", mop.mo_usource);
772 fprintf(stderr, "The target service failed to start "
773 "(bad config log?) (%s). "
774 "See /var/log/messages.\n", mop.mo_usource);
776 fprintf(stderr, "Is the MGS running?\n");
777 if (errno == EADDRINUSE)
778 fprintf(stderr, "The target service's index is already "
779 "in use. (%s)\n", mop.mo_usource);
780 if (errno == EINVAL) {
781 fprintf(stderr, "This may have multiple causes.\n");
783 fprintf(stderr, "Is '%s' the correct filesystem"
785 fprintf(stderr, "Are the mount options correct?\n");
786 fprintf(stderr, "Check the syslog for more info.\n");
789 /* May as well try to clean up loop devs */
790 if (strncmp(mop.mo_usource, "/dev/loop", 9) == 0) {
793 sprintf(cmd, "/sbin/losetup -d %s", mop.mo_usource);
794 if ((ret = system(cmd)) < 0)
797 rc = WEXITSTATUS(ret);
800 } else if (!mop.mo_nomtab) {
801 rc = update_mtab_entry(mop.mo_usource, mop.mo_target, "lustre",
802 mop.mo_orig_options, 0,0,0);
806 /* mo_usource should be freed, but we can rely on the kernel */