1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
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>
51 #include <sys/mount.h>
55 #include <lustre_ver.h>
59 #include "mount_utils.h"
61 #define MAX_HW_SECTORS_KB_PATH "queue/max_hw_sectors_kb"
62 #define MAX_SECTORS_KB_PATH "queue/max_sectors_kb"
63 #define MAX_RETRIES 99
70 char *progname = NULL;
74 fprintf(out, "%s v"LUSTRE_VERSION_STRING"\n", progname);
75 fprintf(out, "\nThis mount helper should only be invoked via the "
76 "mount (8) command,\ne.g. mount -t lustre dev dir\n\n");
77 fprintf(out, "usage: %s [-fhnv] [-o <mntopt>] <device> <mountpt>\n",
80 "\t<device>: the disk device, or for a client:\n"
81 "\t\t<mgmtnid>[:<altmgtnid>...]:/<filesystem>-client\n"
82 "\t<filesystem>: name of the Lustre filesystem (e.g. lustre1)\n"
83 "\t<mountpt>: filesystem mountpoint (e.g. /mnt/lustre)\n"
84 "\t-f|--fake: fake mount (updates /etc/mtab)\n"
85 "\t--force: force mount even if already in /etc/mtab\n"
86 "\t-h|--help: print this usage message\n"
87 "\t-n|--nomtab: do not update /etc/mtab after mount\n"
88 "\t-v|--verbose: print verbose config settings\n"
89 "\t<mntopt>: one or more comma separated of:\n"
90 "\t\t(no)flock,(no)user_xattr,(no)acl\n"
91 "\t\tabort_recov: abort server recovery handling\n"
92 "\t\tnosvc: only start MGC/MGS obds\n"
93 "\t\tnomgs: only start target obds, using existing MGS\n"
94 "\t\texclude=<ostname>[:<ostname>] : colon-separated list of "
95 "inactive OSTs (e.g. lustre-OST0001)\n"
96 "\t\tretry=<num>: number of times mount is retried by client\n"
98 exit((out != stdout) ? EINVAL : 0);
101 static int check_mtab_entry(char *spec1, char *spec2, char *mtpt, char *type)
106 fp = setmntent(MOUNTED, "r");
110 while ((mnt = getmntent(fp)) != NULL) {
111 if ((strcmp(mnt->mnt_fsname, spec1) == 0 ||
112 strcmp(mnt->mnt_fsname, spec2) == 0) &&
113 strcmp(mnt->mnt_dir, mtpt) == 0 &&
114 strcmp(mnt->mnt_type, type) == 0) {
125 update_mtab_entry(char *spec, char *mtpt, char *type, char *opts,
126 int flags, int freq, int pass)
132 mnt.mnt_fsname = spec;
135 mnt.mnt_opts = opts ? opts : "";
137 mnt.mnt_passno = pass;
139 fp = setmntent(MOUNTED, "a+");
141 fprintf(stderr, "%s: setmntent(%s): %s:",
142 progname, MOUNTED, strerror (errno));
145 if ((addmntent(fp, &mnt)) == 1) {
146 fprintf(stderr, "%s: addmntent: %s:",
147 progname, strerror (errno));
156 /* Get rid of symbolic hostnames for tcp, since kernel can't do lookups */
157 #define MAXNIDSTR 1024
158 static char *convert_hostnames(char *s1)
160 char *converted, *s2 = 0, *c;
162 int left = MAXNIDSTR;
165 converted = malloc(left);
167 while ((left > 0) && (*s1 != '/')) {
168 s2 = strpbrk(s1, ",:");
173 nid = libcfs_str2nid(s1);
174 *s2 = sep; /* back to original string */
175 if (nid == LNET_NID_ANY)
177 c += snprintf(c, left, "%s%c", libcfs_nid2str(nid), sep);
178 left = converted + MAXNIDSTR - c;
181 snprintf(c, left, "%s", s1);
184 fprintf(stderr, "%s: Can't parse NID '%s'\n", progname, s1);
189 /*****************************************************************************
191 * This part was cribbed from util-linux/mount/mount.c. There was no clear
192 * license information, but many other files in the package are identified as
193 * GNU GPL, so it's a pretty safe bet that was their intent.
195 ****************************************************************************/
197 const char *opt; /* option name */
198 int inv; /* true if flag value should be inverted */
199 int mask; /* flag mask value */
202 static const struct opt_map opt_map[] = {
203 /*"optname", inv,ms_mask */
204 /* These flags are parsed by mount, not lustre */
205 { "defaults", 0, 0 }, /* default options */
206 { "remount", 0, MS_REMOUNT}, /* remount with different options */
207 { "rw", 1, MS_RDONLY }, /* read-write */
208 { "ro", 0, MS_RDONLY }, /* read-only */
209 { "exec", 1, MS_NOEXEC }, /* permit execution of binaries */
210 { "noexec", 0, MS_NOEXEC }, /* don't execute binaries */
211 { "suid", 1, MS_NOSUID }, /* honor suid executables */
212 { "nosuid", 0, MS_NOSUID }, /* don't honor suid executables */
213 { "dev", 1, MS_NODEV }, /* interpret device files */
214 { "nodev", 0, MS_NODEV }, /* don't interpret devices */
215 { "sync", 0, MS_SYNCHRONOUS}, /* synchronous I/O */
216 { "async", 1, MS_SYNCHRONOUS}, /* asynchronous I/O */
217 { "atime", 1, MS_NOATIME }, /* set file access time on read */
218 { "noatime", 0, MS_NOATIME }, /* do not set file access time on read */
220 { "diratime", 1, MS_NODIRATIME }, /* set file access time on read */
221 { "nodiratime",0,MS_NODIRATIME }, /* do not set file access time on read */
224 { "relatime", 0, MS_RELATIME }, /* set file access time on read */
225 { "norelatime",1,MS_RELATIME }, /* do not set file access time on read */
227 { "auto", 0, 0 }, /* Can be mounted using -a */
228 { "noauto", 0, 0 }, /* Can only be mounted explicitly */
229 { "nousers", 1, 0 }, /* Forbid ordinary user to mount */
230 { "nouser", 1, 0 }, /* Forbid ordinary user to mount */
231 { "noowner", 1, 0 }, /* Device owner has no special privs */
232 { "_netdev", 0, 0 }, /* Device accessible only via network */
236 /****************************************************************************/
238 /* 1 = don't pass on to lustre
239 0 = pass on to lustre */
240 static int parse_one_option(const char *check, int *flagp)
242 const struct opt_map *opt;
244 for (opt = &opt_map[0]; opt->opt != NULL; opt++) {
245 if (strncmp(check, opt->opt, strlen(opt->opt)) == 0) {
248 *flagp &= ~(opt->mask);
255 /* Assume any unknown options are valid and pass them on. The mount
256 will fail if lmd_parse, ll_options or ldiskfs doesn't recognize it.*/
260 /* Replace options with subset of Lustre-specific options, and
261 fill in mount flags */
262 int parse_options(char *orig_options, int *flagp)
264 char *options, *opt, *nextopt, *arg, *val;
266 options = calloc(strlen(orig_options) + 1, 1);
268 nextopt = orig_options;
269 while ((opt = strsep(&nextopt, ","))) {
274 /* Handle retries in a slightly different
277 val = strchr(opt, '=');
278 if (val != NULL && strncmp(arg, "retry", 5) == 0) {
279 retry = atoi(val + 1);
280 if (retry > MAX_RETRIES)
285 else if (parse_one_option(opt, flagp) == 0) {
286 /* pass this on as an option */
288 strcat(options, ",");
289 strcat(options, opt);
292 strcpy(orig_options, options);
298 int read_file(char *path, char *buf, int size)
302 fd = fopen(path, "r");
306 fgets(buf, size, fd);
311 int write_file(char *path, char *buf)
315 fd = fopen(path, "w");
324 /* This is to tune the kernel for good SCSI performance.
325 * For that we set the value of /sys/block/{dev}/queue/max_sectors_kb
326 * to the value of /sys/block/{dev}/queue/max_hw_sectors_kb */
327 int set_tunables(char *source, int src_len)
330 struct stat stat_buf;
331 char *chk_major, *chk_minor;
332 char *savept, *dev, *s2 = 0;
334 char buf[PATH_MAX] = {'\0'}, path[PATH_MAX] = {'\0'};
335 char real_path[PATH_MAX] = {'\0'};
342 ret_path = realpath(source, real_path);
343 if (ret_path == NULL) {
345 fprintf(stderr, "warning: %s: cannot resolve: %s\n",
346 source, strerror(errno));
350 src_len = sizeof(real_path);
352 if (strncmp(real_path, "/dev/loop", 9) == 0)
355 if ((real_path[0] != '/') && ((s2 = strpbrk(real_path, ",:")) != NULL))
358 dev = real_path + src_len - 1;
359 while (dev > real_path && (*dev != '/')) {
364 snprintf(path, sizeof(path), "/sys/block%s/%s", dev,
365 MAX_HW_SECTORS_KB_PATH);
366 rc = read_file(path, buf, sizeof(buf));
367 if (rc == 0 && (strlen(buf) - 1) > 0) {
368 snprintf(path, sizeof(path), "/sys/block%s/%s", dev,
369 MAX_SECTORS_KB_PATH);
370 rc = write_file(path, buf);
372 fprintf(stderr, "warning: opening %s: %s\n",
373 path, strerror(errno));
380 /* The name of the device say 'X' specified in /dev/X may not
381 * match any entry under /sys/block/. In that case we need to
382 * match the major/minor number to find the entry under
383 * sys/block corresponding to /dev/X */
384 dev = real_path + src_len - 1;
385 while (dev > real_path) {
391 rc = stat(dev, &stat_buf);
394 fprintf(stderr, "warning: %s, device %s stat failed\n",
395 strerror(errno), dev);
399 major = major(stat_buf.st_rdev);
400 minor = minor(stat_buf.st_rdev);
401 rc = glob("/sys/block/*", GLOB_NOSORT, NULL, &glob_info);
404 fprintf(stderr, "warning: failed to read entries under "
409 for (i = 0; i < glob_info.gl_pathc; i++){
410 snprintf(path, sizeof(path), "%s/dev", glob_info.gl_pathv[i]);
412 rc = read_file(path, buf, sizeof(buf));
416 if (buf[strlen(buf) - 1] == '\n')
417 buf[strlen(buf) - 1] = '\0';
419 chk_major = strtok_r(buf, ":", &savept);
421 if (major == atoi(chk_major) &&minor == atoi(chk_minor))
425 if (i == glob_info.gl_pathc) {
427 fprintf(stderr,"warning: device %s does not match any "
428 "entry under /sys/block\n", real_path);
433 snprintf(path, sizeof(path), "%s/%s", glob_info.gl_pathv[i],
434 MAX_HW_SECTORS_KB_PATH);
435 rc = read_file(path, buf, sizeof(buf));
438 fprintf(stderr, "warning: opening %s: %s\n",
439 path, strerror(errno));
443 if (strlen(buf) - 1 > 0) {
444 snprintf(path, sizeof(path), "%s/%s",
445 glob_info.gl_pathv[i], MAX_SECTORS_KB_PATH);
446 rc = write_file(path, buf);
448 fprintf(stderr, "warning: writing to %s: %s\n",
449 path, strerror(errno));
453 globfree(&glob_info);
457 int main(int argc, char *const argv[])
459 char default_options[] = "";
460 char *usource, *source, *target, *ptr;
461 char *options, *optcopy, *orig_options = default_options;
462 int i, nargs = 3, opt, rc, flags, optlen;
463 static struct option long_opt[] = {
467 {"nomtab", 0, 0, 'n'},
468 {"options", 1, 0, 'o'},
469 {"verbose", 0, 0, 'v'},
473 progname = strrchr(argv[0], '/');
474 progname = progname ? progname + 1 : argv[0];
476 while ((opt = getopt_long(argc, argv, "fhno:v",
477 long_opt, NULL)) != EOF){
481 printf("force: %d\n", force);
486 printf("fake: %d\n", fake);
494 printf("nomtab: %d\n", nomtab);
498 orig_options = optarg;
506 fprintf(stderr, "%s: unknown option '%c'\n",
513 if (optind + 2 > argc) {
514 fprintf(stderr, "%s: too few arguments\n", progname);
518 usource = argv[optind];
519 source = convert_hostnames(usource);
520 target = argv[optind + 1];
521 ptr = target + strlen(target) - 1;
522 while ((ptr > target) && (*ptr == '/')) {
527 if (!usource || !source) {
532 for (i = 0; i < argc; i++)
533 printf("arg[%d] = %s\n", i, argv[i]);
534 printf("source = %s (%s), target = %s\n", usource, source,
536 printf("options = %s\n", orig_options);
539 options = malloc(strlen(orig_options) + 1);
540 strcpy(options, orig_options);
541 rc = parse_options(options, &flags);
543 fprintf(stderr, "%s: can't parse options: %s\n",
549 rc = check_mtab_entry(usource, source, target, "lustre");
550 if (rc && !(flags & MS_REMOUNT)) {
551 fprintf(stderr, "%s: according to %s %s is "
552 "already mounted on %s\n",
553 progname, MOUNTED, usource, target);
556 if (!rc && (flags & MS_REMOUNT)) {
557 fprintf(stderr, "%s: according to %s %s is "
558 "not already mounted on %s\n",
559 progname, MOUNTED, usource, target);
563 if (flags & MS_REMOUNT)
566 rc = access(target, F_OK);
569 fprintf(stderr, "%s: %s inaccessible: %s\n", progname, target,
574 /* In Linux 2.4, the target device doesn't get passed to any of our
575 functions. So we'll stick it on the end of the options. */
576 optlen = strlen(options) + strlen(",device=") + strlen(source) + 1;
577 optcopy = malloc(optlen);
578 strcpy(optcopy, options);
580 strcat(optcopy, ",");
581 strcat(optcopy, "device=");
582 strcat(optcopy, source);
585 printf("mounting device %s at %s, flags=%#x options=%s\n",
586 source, target, flags, optcopy);
588 if (!strstr(usource, ":/") && set_tunables(source, strlen(source)) &&
590 fprintf(stderr, "%s: unable to set tunables for %s"
591 " (may cause reduced IO performance)\n",
594 register_service_tags(usource, source, target);
597 /* flags and target get to lustre_get_sb, but not
598 lustre_fill_super. Lustre ignores the flags, but mount
600 for (i = 0, rc = -EAGAIN; i <= retry && rc != 0; i++) {
601 rc = mount(source, target, "lustre", flags,
605 fprintf(stderr, "%s: mount %s at %s "
606 "failed: %s retries left: "
607 "%d\n", basename(progname),
609 strerror(errno), retry-i);
613 sleep(1 << max((i/2), 5));
627 cli = strrchr(usource, ':');
628 if (cli && (strlen(cli) > 2))
633 fprintf(stderr, "%s: mount %s at %s failed: %s\n", progname,
634 usource, target, strerror(errno));
636 fprintf(stderr, "Are the lustre modules loaded?\n"
637 "Check /etc/modprobe.conf and /proc/filesystems"
638 "\nNote 'alias lustre llite' should be removed"
639 " from modprobe.conf\n");
640 if (errno == ENOTBLK)
641 fprintf(stderr, "Do you need -o loop?\n");
642 if (errno == ENOMEDIUM)
644 "This filesystem needs at least 1 OST\n");
645 if (errno == ENOENT) {
646 fprintf(stderr, "Is the MGS specification correct?\n");
647 fprintf(stderr, "Is the filesystem name correct?\n");
648 fprintf(stderr, "If upgrading, is the copied client log"
649 " valid? (see upgrade docs)\n");
651 if (errno == EALREADY)
652 fprintf(stderr, "The target service is already running."
655 fprintf(stderr, "The target service failed to start "
656 "(bad config log?) (%s). "
657 "See /var/log/messages.\n", usource);
659 fprintf(stderr, "Is the MGS running?\n");
660 if (errno == EADDRINUSE)
661 fprintf(stderr, "The target service's index is already "
662 "in use. (%s)\n", usource);
663 if (errno == EINVAL) {
664 fprintf(stderr, "This may have multiple causes.\n");
666 fprintf(stderr, "Is '%s' the correct filesystem"
668 fprintf(stderr, "Are the mount options correct?\n");
669 fprintf(stderr, "Check the syslog for more info.\n");
672 /* May as well try to clean up loop devs */
673 if (strncmp(usource, "/dev/loop", 9) == 0) {
675 sprintf(cmd, "/sbin/losetup -d %s", usource);
679 } else if (!nomtab) {
680 rc = update_mtab_entry(usource, target, "lustre", orig_options,