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"
118 "\t\tskpath=<file|directory>: Path to a file or directory of"
119 "key configuration files to load into the kernel keyring\n"
121 "\t\tretry=<num>: number of times mount is retried by client\n"
122 "\t\tmd_stripe_cache_size=<num>: set the raid stripe cache "
123 "size for the underlying raid if present\n");
124 exit((out != stdout) ? EINVAL : 0);
127 /* Get rid of symbolic hostnames for tcp, since kernel can't do lookups */
128 #define MAXNIDSTR 1024
129 static char *convert_hostnames(char *s1)
131 char *converted, *s2 = 0, *c;
133 int left = MAXNIDSTR;
136 converted = malloc(left);
137 if (converted == NULL) {
138 fprintf(stderr, "out of memory: needed %d bytes\n",
143 while ((left > 0) && (*s1 != '/')) {
144 s2 = strpbrk(s1, ",:");
149 nid = libcfs_str2nid(s1);
150 *s2 = sep; /* back to original string */
151 if (nid == LNET_NID_ANY)
153 c += snprintf(c, left, "%s%c", libcfs_nid2str(nid), sep);
154 left = converted + MAXNIDSTR - c;
157 snprintf(c, left, "%s", s1);
160 fprintf(stderr, "%s: Can't parse NID '%s'\n", progname, s1);
165 /*****************************************************************************
167 * This part was cribbed from util-linux/mount/mount.c. There was no clear
168 * license information, but many other files in the package are identified as
169 * GNU GPL, so it's a pretty safe bet that was their intent.
171 ****************************************************************************/
173 const char *opt; /* option name */
174 int inv; /* true if flag value should be inverted */
175 int mask; /* flag mask value */
178 static const struct opt_map opt_map[] = {
179 /*"optname", inv,ms_mask */
180 /* These flags are parsed by mount, not lustre */
181 { "defaults", 0, 0 }, /* default options */
182 { "remount", 0, MS_REMOUNT}, /* remount with different options */
183 { "rw", 1, MS_RDONLY }, /* read-write */
184 { "ro", 0, MS_RDONLY }, /* read-only */
185 { "exec", 1, MS_NOEXEC }, /* permit execution of binaries */
186 { "noexec", 0, MS_NOEXEC }, /* don't execute binaries */
187 { "suid", 1, MS_NOSUID }, /* honor suid executables */
188 { "nosuid", 0, MS_NOSUID }, /* don't honor suid executables */
189 { "dev", 1, MS_NODEV }, /* interpret device files */
190 { "nodev", 0, MS_NODEV }, /* don't interpret devices */
191 { "sync", 0, MS_SYNCHRONOUS}, /* synchronous I/O */
192 { "async", 1, MS_SYNCHRONOUS}, /* asynchronous I/O */
193 { "atime", 1, MS_NOATIME }, /* set file access time on read */
194 { "noatime", 0, MS_NOATIME }, /* do not set file access time on read */
196 { "diratime", 1, MS_NODIRATIME }, /* set file access time on read */
197 { "nodiratime",0,MS_NODIRATIME }, /* do not set file access time on read */
200 { "relatime", 0, MS_RELATIME }, /* set file access time on read */
201 { "norelatime",1,MS_RELATIME }, /* do not set file access time on read */
203 #ifdef MS_STRICTATIME
204 { "strictatime",0,MS_STRICTATIME }, /* update access time strictly */
206 { "auto", 0, 0 }, /* Can be mounted using -a */
207 { "noauto", 0, 0 }, /* Can only be mounted explicitly */
208 { "nousers", 1, 0 }, /* Forbid ordinary user to mount */
209 { "nouser", 1, 0 }, /* Forbid ordinary user to mount */
210 { "noowner", 1, 0 }, /* Device owner has no special privs */
211 { "_netdev", 0, 0 }, /* Device accessible only via network */
215 /****************************************************************************/
217 /* 1 = don't pass on to lustre
218 0 = pass on to lustre */
219 static int parse_one_option(const char *check, int *flagp)
221 const struct opt_map *opt;
223 for (opt = &opt_map[0]; opt->opt != NULL; opt++) {
224 if (strncmp(check, opt->opt, strlen(opt->opt)) == 0) {
227 *flagp &= ~(opt->mask);
234 /* Assume any unknown options are valid and pass them on. The mount
235 will fail if lmd_parse, ll_options or ldiskfs doesn't recognize it.*/
239 static void append_option(char *options, const char *one)
242 strcat(options, ",");
243 strcat(options, one);
246 /* Replace options with subset of Lustre-specific options, and
247 fill in mount flags */
248 int parse_options(struct mount_opts *mop, char *orig_options, int *flagp)
250 char *options, *opt, *nextopt, *arg, *val;
252 options = calloc(strlen(orig_options) + 1, 1);
254 nextopt = orig_options;
255 while ((opt = strsep(&nextopt, ","))) {
260 /* Handle retries in a slightly different
263 val = strchr(opt, '=');
264 /* please note that some ldiskfs mount options are also in the form
265 * of param=value. We should pay attention not to remove those
266 * mount options, see bug 22097. */
267 if (val && strncmp(arg, "md_stripe_cache_size", 20) == 0) {
268 mop->mo_md_stripe_cache_size = atoi(val + 1);
269 } else if (val && strncmp(arg, "retry", 5) == 0) {
270 mop->mo_retry = atoi(val + 1);
271 if (mop->mo_retry > MAX_RETRIES)
272 mop->mo_retry = MAX_RETRIES;
273 else if (mop->mo_retry < 0)
275 } else if (val && strncmp(arg, "mgssec", 6) == 0) {
276 append_option(options, opt);
277 } else if (strncmp(arg, "nosvc", 5) == 0) {
279 append_option(options, opt);
280 } else if (strcmp(opt, "force") == 0) {
281 /* XXX special check for 'force' option */
283 printf("force: %d\n", mop->mo_force);
285 } else if (val && strncmp(opt, "skpath=", 7) == 0) {
286 if (strlen(val) + 1 >= sizeof(mop->mo_skpath)) {
288 "%s: shared key path too long\n",
293 strncpy(mop->mo_skpath, val + 1, strlen(val + 1));
295 } else if (parse_one_option(opt, flagp) == 0) {
296 /* pass this on as an option */
297 append_option(options, opt);
300 #ifdef MS_STRICTATIME
301 #if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(3, 2, 53, 0)
304 * In the future when upstream fixes land in all supported kernels
305 * we should stop forcing MS_STRICTATIME in lustre mounts.
306 * We override the kernel level default of MS_RELATIME for now
307 * due to a kernel vfs level bug in atime updates that fails
308 * to reset timestamps from the future.
310 #warn "remove MS_STRICTATIME override if kernel updates atime from the future"
312 /* set strictatime to default if NOATIME or RELATIME
313 not given explicit */
314 if (!(*flagp & (MS_NOATIME | MS_RELATIME)))
315 *flagp |= MS_STRICTATIME;
317 strcpy(orig_options, options);
323 /* Add mgsnids from ldd params */
324 static int add_mgsnids(struct mount_opts *mop, char *options,
327 char *ptr = (char *)params;
330 while ((ptr = strstr(ptr, PARAM_MGSNODE)) != NULL) {
331 sep = strchr(ptr, ' ');
336 append_option(options, ptr);
337 mop->mo_have_mgsnid++;
349 static int clear_update_ondisk(char *source, struct lustre_disk_data *ldd)
351 char wanted_mountopts[512] = "";
352 struct mkfs_opts mkop;
356 memset(&mkop, 0, sizeof(mkop));
358 mkop.mo_ldd.ldd_flags &= ~LDD_F_UPDATE;
359 mkop.mo_flags = MO_NOHOSTID_CHECK; /* Ignore missing hostid */
360 if (strlen(source) > sizeof(mkop.mo_device)-1) {
362 fprintf(stderr, "Device name too long: %s\n", source);
365 strncpy(mkop.mo_device, source, sizeof(mkop.mo_device));
367 ret = osd_prepare_lustre(&mkop,
368 wanted_mountopts, sizeof(wanted_mountopts));
371 fprintf(stderr, "Can't prepare device %s: %s\n",
372 source, strerror(ret));
376 /* Create the loopback file */
377 if (mkop.mo_flags & MO_IS_LOOP) {
378 ret = access(mkop.mo_device, F_OK);
382 fprintf(stderr, "Can't access device %s: %s\n",
383 source, strerror(ret));
387 ret = loop_setup(&mkop);
390 fprintf(stderr, "Loop device setup for %s failed: %s\n",
391 mkop.mo_device, strerror(ret));
395 ret = osd_write_ldd(&mkop);
398 fprintf(stderr, "failed to write local files: %s\n",
402 ret2 = loop_cleanup(&mkop);
409 static int parse_ldd(char *source, struct mount_opts *mop, char *options)
411 struct lustre_disk_data *ldd = &mop->mo_ldd;
415 rc = osd_is_lustre(source, &ldd->ldd_mount_type);
417 fprintf(stderr, "%s: %s has not been formatted with mkfs.lustre"
418 " or the backend filesystem type is not supported by "
419 "this tool\n", progname, source);
423 rc = osd_read_ldd(source, ldd);
425 fprintf(stderr, "%s: %s failed to read permanent mount"
426 " data: %s\n", progname, source,
427 rc >= 0 ? strerror(rc) : "");
431 if ((IS_MDT(ldd) || IS_OST(ldd)) &&
432 (ldd->ldd_flags & LDD_F_NEED_INDEX)) {
433 fprintf(stderr, "%s: %s has no index assigned "
434 "(probably formatted with old mkfs)\n",
439 if (ldd->ldd_flags & LDD_F_UPGRADE14) {
440 fprintf(stderr, "%s: we cannot upgrade %s from this (very old) "
441 "Lustre version\n", progname, source);
445 if (ldd->ldd_flags & LDD_F_UPDATE)
446 clear_update_ondisk(source, ldd);
448 /* Since we never rewrite ldd, ignore temp flags */
449 ldd->ldd_flags &= ~(LDD_F_VIRGIN | LDD_F_WRITECONF);
451 /* svname of the form lustre:OST1234 means never registered */
452 rc = strlen(ldd->ldd_svname);
453 if (strcmp(ldd->ldd_svname, "MGS") != 0) {
455 fprintf(stderr, "%s: invalid name '%s'\n",
456 progname, ldd->ldd_svname);
458 } else if (ldd->ldd_svname[rc - 8] == ':') {
459 ldd->ldd_svname[rc - 8] = '-';
460 ldd->ldd_flags |= LDD_F_VIRGIN;
461 } else if (ldd->ldd_svname[rc - 8] == '=') {
462 ldd->ldd_svname[rc - 8] = '-';
463 ldd->ldd_flags |= LDD_F_WRITECONF;
466 /* backend osd type */
467 append_option(options, "osd=");
468 strcat(options, mt_type(ldd->ldd_mount_type));
470 append_option(options, ldd->ldd_mount_opts);
472 if (!mop->mo_have_mgsnid) {
473 /* Only use disk data if mount -o mgsnode=nid wasn't
475 if (ldd->ldd_flags & LDD_F_SV_TYPE_MGS) {
476 append_option(options, "mgs");
477 mop->mo_have_mgsnid++;
479 add_mgsnids(mop, options, ldd->ldd_params);
482 /* Better have an mgsnid by now */
483 if (!mop->mo_have_mgsnid) {
484 fprintf(stderr, "%s: missing option mgsnode=<nid>\n",
489 if (ldd->ldd_flags & LDD_F_VIRGIN)
490 append_option(options, "virgin");
491 if (ldd->ldd_flags & LDD_F_UPDATE)
492 append_option(options, "update");
493 if (ldd->ldd_flags & LDD_F_WRITECONF)
494 append_option(options, "writeconf");
495 if (ldd->ldd_flags & LDD_F_NO_PRIMNODE)
496 append_option(options, "noprimnode");
498 /* prefix every lustre parameter with param= so that in-kernel
499 * mount can recognize them properly and send to MGS at registration */
500 start = ldd->ldd_params;
501 while (start && *start != '\0') {
502 while (*start == ' ') start++;
506 start = strchr(cur, ' ');
511 append_option(options, "param=");
512 strcat(options, cur);
515 /* svname must be last option */
516 append_option(options, "svname=");
517 strcat(options, ldd->ldd_svname);
522 static void set_defaults(struct mount_opts *mop)
524 memset(mop, 0, sizeof(*mop));
525 mop->mo_usource = NULL;
526 mop->mo_source = NULL;
531 mop->mo_have_mgsnid = 0;
532 mop->mo_md_stripe_cache_size = 16384;
533 mop->mo_orig_options = "";
537 static int parse_opts(int argc, char *const argv[], struct mount_opts *mop)
539 static struct option long_opt[] = {
543 {"nomtab", 0, 0, 'n'},
544 {"options", 1, 0, 'o'},
545 {"verbose", 0, 0, 'v'},
546 {"version", 0, 0, 'V'},
549 char real_path[PATH_MAX] = {'\0'};
551 char path[256], name[256];
556 while ((opt = getopt_long(argc, argv, "fhno:vV",
557 long_opt, NULL)) != EOF){
561 printf("force: %d\n", mop->mo_force);
565 printf("fake: %d\n", mop->mo_fake);
572 printf("nomtab: %d\n", mop->mo_nomtab);
575 mop->mo_orig_options = optarg;
582 fprintf(stdout, "%s %s %s\n", progname,
583 LUSTRE_VERSION_STRING, WITH_LIBMOUNT);
586 fprintf(stderr, "%s: unknown option '%c'\n",
593 if (optind + 2 > argc) {
594 fprintf(stderr, "%s: too few arguments\n", progname);
598 mop->mo_usource = argv[optind];
599 if (!mop->mo_usource) {
604 * Try to get the real path to the device, in case it is a
605 * symbolic link for instance
607 if (realpath(mop->mo_usource, real_path) != NULL) {
608 ptr = strrchr(real_path, '/');
609 if (ptr && strncmp(ptr, "/dm-", 4) == 0 && isdigit(*(ptr + 4))) {
610 snprintf(path, sizeof(path), "/sys/block/%s/dm/name", ptr+1);
611 if ((f = fopen(path, "r"))) {
612 /* read "<name>\n" from sysfs */
613 if (fgets(name, sizeof(name), f) && (sz = strlen(name)) > 1) {
615 snprintf(real_path, sizeof(real_path), "/dev/mapper/%s", name);
620 mop->mo_usource = strdup(real_path);
623 ptr = strstr(mop->mo_usource, ":/");
625 mop->mo_source = convert_hostnames(mop->mo_usource);
629 mop->mo_source = strdup(mop->mo_usource);
632 if (realpath(argv[optind + 1], mop->mo_target) == NULL) {
634 fprintf(stderr, "warning: %s: cannot resolve: %s\n",
635 argv[optind + 1], strerror(errno));
642 int main(int argc, char *const argv[])
644 struct mount_opts mop;
649 progname = strrchr(argv[0], '/');
650 progname = progname ? progname + 1 : argv[0];
654 rc = parse_opts(argc, argv, &mop);
659 for (i = 0; i < argc; i++)
660 printf("arg[%d] = %s\n", i, argv[i]);
661 printf("source = %s (%s), target = %s\n", mop.mo_usource,
662 mop.mo_source, mop.mo_target);
663 printf("options = %s\n", mop.mo_orig_options);
666 options = malloc(MAXOPT);
667 if (options == NULL) {
668 fprintf(stderr, "can't allocate memory for options\n");
671 strcpy(options, mop.mo_orig_options);
672 rc = parse_options(&mop, options, &flags);
674 fprintf(stderr, "%s: can't parse options: %s\n",
680 rc = check_mtab_entry(mop.mo_usource, mop.mo_source,
681 mop.mo_target, "lustre");
682 if (rc && !(flags & MS_REMOUNT)) {
683 fprintf(stderr, "%s: according to %s %s is "
684 "already mounted on %s\n", progname, MOUNTED,
685 mop.mo_usource, mop.mo_target);
688 if (!rc && (flags & MS_REMOUNT)) {
689 fprintf(stderr, "%s: according to %s %s is "
690 "not already mounted on %s\n", progname, MOUNTED,
691 mop.mo_usource, mop.mo_target);
695 if (flags & MS_REMOUNT)
698 rc = access(mop.mo_target, F_OK);
701 fprintf(stderr, "%s: %s inaccessible: %s\n", progname,
702 mop.mo_target, strerror(errno));
706 client = (strstr(mop.mo_usource, ":/") != NULL);
712 rc = parse_ldd(mop.mo_source, &mop, options);
717 /* In Linux 2.4, the target device doesn't get passed to any of our
718 functions. So we'll stick it on the end of the options. */
719 append_option(options, "device=");
720 strcat(options, mop.mo_source);
723 printf("mounting device %s at %s, flags=%#x options=%s\n",
724 mop.mo_source, mop.mo_target, flags, options);
726 if (!client && osd_tune_lustre(mop.mo_source, &mop)) {
728 fprintf(stderr, "%s: unable to set tunables for %s"
729 " (may cause reduced IO performance)\n",
730 argv[0], mop.mo_source);
734 if (mop.mo_skpath[0] != '\0') {
735 /* Treat shared key failures as fatal */
736 rc = load_shared_keys(&mop);
739 "%s: Error loading shared keys: %s\n",
740 progname, strerror(rc));
747 /* flags and target get to lustre_get_sb(), but not
748 * lustre_fill_super(). Lustre ignores the flags, but mount
750 for (i = 0, rc = -EAGAIN; i <= mop.mo_retry && rc != 0; i++) {
751 rc = mount(mop.mo_source, mop.mo_target, "lustre",
752 flags, (void *)options);
754 /* change label from <fsname>:<index> to
755 * <fsname>-<index> to indicate the device has
756 * been registered. only if the label is
757 * supposed to be changed and target service
758 * is supposed to start */
759 if (mop.mo_ldd.ldd_flags &
760 (LDD_F_VIRGIN | LDD_F_WRITECONF)) {
761 if (mop.mo_nosvc == 0)
762 (void)osd_label_lustre(&mop);
766 fprintf(stderr, "%s: mount %s at %s "
767 "failed: %s retries left: "
768 "%d\n", basename(progname),
769 mop.mo_usource, mop.mo_target,
775 int limit = i/2 > 5 ? i/2 : 5;
790 cli = strrchr(mop.mo_usource, ':');
791 if (cli && (strlen(cli) > 2))
796 fprintf(stderr, "%s: mount %s at %s failed: %s\n", progname,
797 mop.mo_usource, mop.mo_target, strerror(errno));
799 fprintf(stderr, "Is the backend filesystem mounted?\n"
800 "Check /etc/mtab and /proc/mounts\n");
802 fprintf(stderr, "Are the lustre modules loaded?\n"
803 "Check /etc/modprobe.conf and "
804 "/proc/filesystems\n");
805 if (errno == ENOTBLK)
806 fprintf(stderr, "Do you need -o loop?\n");
807 if (errno == ENOMEDIUM)
809 "This filesystem needs at least 1 OST\n");
810 if (errno == ENOENT) {
811 fprintf(stderr, "Is the MGS specification correct?\n");
812 fprintf(stderr, "Is the filesystem name correct?\n");
813 fprintf(stderr, "If upgrading, is the copied client log"
814 " valid? (see upgrade docs)\n");
816 if (errno == EALREADY)
817 fprintf(stderr, "The target service is already running."
818 " (%s)\n", mop.mo_usource);
820 fprintf(stderr, "The target service failed to start "
821 "(bad config log?) (%s). "
822 "See /var/log/messages.\n", mop.mo_usource);
824 fprintf(stderr, "Is the MGS running?\n");
825 if (errno == EADDRINUSE)
826 fprintf(stderr, "The target service's index is already "
827 "in use. (%s)\n", mop.mo_usource);
828 if (errno == EINVAL) {
829 fprintf(stderr, "This may have multiple causes.\n");
831 fprintf(stderr, "Is '%s' the correct filesystem"
833 fprintf(stderr, "Are the mount options correct?\n");
834 fprintf(stderr, "Check the syslog for more info.\n");
837 /* May as well try to clean up loop devs */
838 if (strncmp(mop.mo_usource, "/dev/loop", 9) == 0) {
841 sprintf(cmd, "/sbin/losetup -d %s", mop.mo_usource);
842 if ((ret = system(cmd)) < 0)
845 rc = WEXITSTATUS(ret);
849 /* Deal with utab just for client. Note that we ignore
850 * the return value here since it is not worth to fail
851 * mount by prevent some rare cases */
852 if (strstr(mop.mo_usource, ":/") != NULL)
853 update_utab_entry(&mop);
854 if (!mop.mo_nomtab) {
855 rc = update_mtab_entry(mop.mo_usource, mop.mo_target,
856 "lustre", mop.mo_orig_options,
862 /* mo_usource should be freed, but we can rely on the kernel */