1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 2002 Cluster File Systems, Inc.
5 * Author: Robert Read <rread@clusterfs.com>
6 * Author: Nathan Rutman <nathan@clusterfs.com>
8 * This file is part of Lustre, http://www.lustre.org.
10 * Lustre is free software; you can redistribute it and/or
11 * modify it under the terms of version 2 of the GNU General Public
12 * License as published by the Free Software Foundation.
14 * Lustre is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
19 * You should have received a copy of the GNU General Public License
20 * along with Lustre; if not, write to the Free Software
21 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
33 #include <sys/mount.h>
36 #include <sys/utsname.h>
38 #include <lustre_ver.h>
41 #define MAX_HW_SECTORS_KB_PATH "queue/max_hw_sectors_kb"
42 #define MAX_SECTORS_KB_PATH "queue/max_sectors_kb"
48 static char *progname = NULL;
52 fprintf(out, "%s v"LUSTRE_VERSION_STRING"\n", progname);
53 fprintf(out, "\nThis mount helper should only be invoked via the "
54 "mount (8) command,\ne.g. mount -t lustre dev dir\n\n");
55 fprintf(out, "usage: %s [-fhnv] [-o <mntopt>] <device> <mountpt>\n",
58 "\t<device>: the disk device, or for a client:\n"
59 "\t\t<mgmtnid>[:<altmgtnid>...]:/<filesystem>-client\n"
60 "\t<filesystem>: name of the Lustre filesystem (e.g. lustre1)\n"
61 "\t<mountpt>: filesystem mountpoint (e.g. /mnt/lustre)\n"
62 "\t-f|--fake: fake mount (updates /etc/mtab)\n"
63 "\t--force: force mount even if already in /etc/mtab\n"
64 "\t-h|--help: print this usage message\n"
65 "\t-n|--nomtab: do not update /etc/mtab after mount\n"
66 "\t-v|--verbose: print verbose config settings\n"
67 "\t<mntopt>: one or more comma separated of:\n"
68 "\t\t(no)flock,(no)user_xattr,(no)acl\n"
69 "\t\tnosvc: only start MGC/MGS obds\n"
70 "\t\texclude=<ostname>[:<ostname>] : colon-separated list of "
71 "inactive OSTs (e.g. lustre-OST0001)\n"
73 exit((out != stdout) ? EINVAL : 0);
76 static int check_mtab_entry(char *spec, char *mtpt, char *type)
81 fp = setmntent(MOUNTED, "r");
85 while ((mnt = getmntent(fp)) != NULL) {
86 if (strcmp(mnt->mnt_fsname, spec) == 0 &&
87 strcmp(mnt->mnt_dir, mtpt) == 0 &&
88 strcmp(mnt->mnt_type, type) == 0) {
99 update_mtab_entry(char *spec, char *mtpt, char *type, char *opts,
100 int flags, int freq, int pass)
106 mnt.mnt_fsname = spec;
109 mnt.mnt_opts = opts ? opts : "";
111 mnt.mnt_passno = pass;
113 fp = setmntent(MOUNTED, "a+");
115 fprintf(stderr, "%s: setmntent(%s): %s:",
116 progname, MOUNTED, strerror (errno));
119 if ((addmntent(fp, &mnt)) == 1) {
120 fprintf(stderr, "%s: addmntent: %s:",
121 progname, strerror (errno));
130 /* Get rid of symbolic hostnames for tcp, since kernel can't do lookups */
131 #define MAXNIDSTR 1024
132 static char *convert_hostnames(char *s1)
134 char *converted, *s2 = 0, *c;
136 int left = MAXNIDSTR;
139 converted = malloc(left);
141 while ((left > 0) && (*s1 != '/')) {
142 s2 = strpbrk(s1, ",:");
147 nid = libcfs_str2nid(s1);
148 *s2 = sep; /* back to original string */
149 if (nid == LNET_NID_ANY)
151 c += snprintf(c, left, "%s%c", libcfs_nid2str(nid), sep);
152 left = converted + MAXNIDSTR - c;
155 snprintf(c, left, "%s", s1);
158 fprintf(stderr, "%s: Can't parse NID '%s'\n", progname, s1);
163 /*****************************************************************************
165 * This part was cribbed from util-linux/mount/mount.c. There was no clear
166 * license information, but many other files in the package are identified as
167 * GNU GPL, so it's a pretty safe bet that was their intent.
169 ****************************************************************************/
171 const char *opt; /* option name */
172 int inv; /* true if flag value should be inverted */
173 int mask; /* flag mask value */
176 static const struct opt_map opt_map[] = {
177 /*"optname", inv,ms_mask */
178 /* These flags are parsed by mount, not lustre */
179 { "defaults", 0, 0 }, /* default options */
180 { "remount", 0, MS_REMOUNT}, /* remount with different options */
181 { "rw", 1, MS_RDONLY }, /* read-write */
182 { "ro", 0, MS_RDONLY }, /* read-only */
183 { "exec", 1, MS_NOEXEC }, /* permit execution of binaries */
184 { "noexec", 0, MS_NOEXEC }, /* don't execute binaries */
185 { "suid", 1, MS_NOSUID }, /* honor suid executables */
186 { "nosuid", 0, MS_NOSUID }, /* don't honor suid executables */
187 { "dev", 1, MS_NODEV }, /* interpret device files */
188 { "nodev", 0, MS_NODEV }, /* don't interpret devices */
189 { "async", 1, MS_SYNCHRONOUS}, /* asynchronous I/O */
190 { "auto", 0, 0 }, /* Can be mounted using -a */
191 { "noauto", 0, 0 }, /* Can only be mounted explicitly */
192 { "nousers", 1, 0 }, /* Forbid ordinary user to mount */
193 { "nouser", 1, 0 }, /* Forbid ordinary user to mount */
194 { "noowner", 1, 0 }, /* Device owner has no special privs */
195 { "_netdev", 0, 0 }, /* Device accessible only via network */
198 /****************************************************************************/
200 /* 1 = don't pass on to lustre
201 0 = pass on to lustre */
202 static int parse_one_option(const char *check, int *flagp)
204 const struct opt_map *opt;
206 for (opt = &opt_map[0]; opt->opt != NULL; opt++) {
207 if (strncmp(check, opt->opt, strlen(opt->opt)) == 0) {
210 *flagp &= ~(opt->mask);
217 /* Assume any unknown options are valid and pass them on. The mount
218 will fail if lmd_parse, ll_options or ldiskfs doesn't recognize it.*/
222 /* Replace options with subset of Lustre-specific options, and
223 fill in mount flags */
224 int parse_options(char *orig_options, int *flagp)
226 char *options, *opt, *nextopt;
228 options = calloc(strlen(orig_options) + 1, 1);
230 nextopt = orig_options;
231 while ((opt = strsep(&nextopt, ","))) {
235 if (parse_one_option(opt, flagp) == 0) {
236 /* pass this on as an option */
238 strcat(options, ",");
239 strcat(options, opt);
242 strcpy(orig_options, options);
248 int read_file(char *path, char *buf, int size)
252 fd = fopen(path, "r");
256 fgets(buf, size, fd);
261 int write_file(char *path, char *buf)
265 fd = fopen(path, "w");
274 /* This is to tune the kernel for good SCSI performance.
275 * For that we set the value of /sys/block/{dev}/queue/max_sectors_kb
276 * to the value of /sys/block/{dev}/queue/max_hw_sectors_kb */
277 int set_tunables(char *source, int src_len)
280 struct stat stat_buf;
281 char *chk_major, *chk_minor;
282 char *savept, *dev, *s2 = 0;
283 char buf[PATH_MAX], path[PATH_MAX];
290 if (strncmp(source, "/dev/loop", 9) == 0)
293 if ((*source != '/') && ((s2 = strpbrk(source, ",:")) != NULL))
296 dev = source + src_len - 1;
297 while (dev > source && (*dev != '/')) {
302 snprintf(path, sizeof(path), "/sys/block%s/%s", dev,
303 MAX_HW_SECTORS_KB_PATH);
304 rc = read_file(path, buf, sizeof(buf));
305 if (!rc && (strlen(buf)-1)) {
306 snprintf(path, sizeof(path), "/sys/block%s/%s", dev,
307 MAX_SECTORS_KB_PATH);
308 rc = write_file(path, buf);
310 fprintf(stderr, "warning: opening %s: %s\n",
311 path, strerror(errno));
314 } else if (rc == ENOENT) {
315 /* The name of the device say 'X' specified in /dev/X may not match
316 * any entry under /sys/block/. In that case we need to match
317 * the major/minor number to find the entry under sys/block
318 * corresponding to /dev/X */
320 dev = source + src_len - 1;
321 while (dev > source) {
327 rc = stat(dev, &stat_buf);
329 fprintf(stderr, "warning: %s, Stat failed for device %s\n",
330 strerror(errno), dev);
333 major = major(stat_buf.st_rdev);
334 minor = minor(stat_buf.st_rdev);
335 rc = glob("/sys/block/*", GLOB_NOSORT, NULL, &glob_info);
337 fprintf(stderr, "warning: failed to read entries under /sys/block\n");
341 for (i = 0; i < glob_info.gl_pathc; i++){
342 snprintf(path, sizeof(path), "%s/dev", glob_info.gl_pathv[i]);
344 rc = read_file(path, buf, sizeof(buf));
348 if (buf[strlen(buf)-1] == '\n')
349 buf[strlen(buf)-1] = '\0';
351 chk_major = strtok_r(buf, ":", &savept);
353 if (major == atoi(chk_major) && minor == atoi(chk_minor))
357 if (i == glob_info.gl_pathc) {
358 fprintf(stderr,"warning: the device %s, does not match any"
359 "of /sys/block entries\n", source);
363 snprintf(path, sizeof(path), "%s/%s", glob_info.gl_pathv[i],
364 MAX_HW_SECTORS_KB_PATH);
365 rc = read_file(path, buf, sizeof(buf));
367 fprintf(stderr, "warning: opening %s: %s\n",
368 path, strerror(errno));
369 if (!rc && ((strlen(buf)-1) > 0)) {
370 snprintf(path, sizeof(path), "%s/%s",
371 glob_info.gl_pathv[i], MAX_SECTORS_KB_PATH);
372 rc = write_file(path, buf);
374 fprintf(stderr, "warning: opening %s: %s\n",
375 path, strerror(errno));
383 int main(int argc, char *const argv[])
385 char default_options[] = "";
386 char *usource, *source, *target, *ptr;
387 char *options, *optcopy, *orig_options = default_options;
388 int i, nargs = 3, opt, rc, flags, optlen;
389 static struct option long_opt[] = {
393 {"nomtab", 0, 0, 'n'},
394 {"options", 1, 0, 'o'},
395 {"verbose", 0, 0, 'v'},
399 progname = strrchr(argv[0], '/');
400 progname = progname ? progname + 1 : argv[0];
402 while ((opt = getopt_long(argc, argv, "fhno:v",
403 long_opt, NULL)) != EOF){
407 printf("force: %d\n", force);
412 printf("fake: %d\n", fake);
420 printf("nomtab: %d\n", nomtab);
424 orig_options = optarg;
432 fprintf(stderr, "%s: unknown option '%c'\n",
439 if (optind + 2 > argc) {
440 fprintf(stderr, "%s: too few arguments\n", progname);
444 usource = argv[optind];
445 source = convert_hostnames(usource);
446 target = argv[optind + 1];
447 ptr = target + strlen(target) - 1;
448 while ((ptr > target) && (*ptr == '/')) {
453 if (!usource || !source) {
458 for (i = 0; i < argc; i++)
459 printf("arg[%d] = %s\n", i, argv[i]);
460 printf("source = %s (%s), target = %s\n", usource, source, target);
461 printf("options = %s\n", orig_options);
464 options = malloc(strlen(orig_options) + 1);
465 strcpy(options, orig_options);
466 rc = parse_options(options, &flags);
468 fprintf(stderr, "%s: can't parse options: %s\n",
474 rc = check_mtab_entry(usource, target, "lustre");
475 if (rc && !(flags & MS_REMOUNT)) {
476 fprintf(stderr, "%s: according to %s %s is "
477 "already mounted on %s\n",
478 progname, MOUNTED, usource, target);
481 if (!rc && (flags & MS_REMOUNT)) {
482 fprintf(stderr, "%s: according to %s %s is "
483 "not already mounted on %s\n",
484 progname, MOUNTED, usource, target);
488 if (flags & MS_REMOUNT)
491 rc = access(target, F_OK);
494 fprintf(stderr, "%s: %s inaccessible: %s\n", progname, target,
499 /* In Linux 2.4, the target device doesn't get passed to any of our
500 functions. So we'll stick it on the end of the options. */
501 optlen = strlen(options) + strlen(",device=") + strlen(source) + 1;
502 optcopy = malloc(optlen);
503 strcpy(optcopy, options);
505 strcat(optcopy, ",");
506 strcat(optcopy, "device=");
507 strcat(optcopy, source);
510 printf("mounting device %s at %s, flags=%#x options=%s\n",
511 source, target, flags, optcopy);
513 if (set_tunables(source, strlen(source)))
514 fprintf(stderr, "%s: unable to set tunables for %s"
515 " (may cause reduced IO performance)",
519 /* flags and target get to lustre_get_sb, but not
520 lustre_fill_super. Lustre ignores the flags, but mount
522 rc = mount(source, target, "lustre", flags, (void *)optcopy);
529 cli = strrchr(usource, ':');
530 if (cli && (strlen(cli) > 2))
535 fprintf(stderr, "%s: mount %s at %s failed: %s\n", progname,
536 usource, target, strerror(errno));
538 fprintf(stderr, "Are the lustre modules loaded?\n"
539 "Check /etc/modprobe.conf and /proc/filesystems"
540 "\nNote 'alias lustre llite' should be removed"
541 " from modprobe.conf\n");
542 if (errno == ENOTBLK)
543 fprintf(stderr, "Do you need -o loop?\n");
544 if (errno == ENOMEDIUM)
546 "This filesystem needs at least 1 OST\n");
547 if (errno == ENOENT) {
548 fprintf(stderr, "Is the MGS specification correct?\n");
549 fprintf(stderr, "Is the filesystem name correct?\n");
550 fprintf(stderr, "If upgrading, is the copied client log"
551 " valid? (see upgrade docs)\n");
553 if (errno == EALREADY)
554 fprintf(stderr, "The target service is already running."
557 fprintf(stderr, "The target service failed to start "
558 "(bad config log?) (%s). "
559 "See /var/log/messages.\n", usource);
561 fprintf(stderr, "Is the MGS running?\n");
562 if (errno == EADDRINUSE)
563 fprintf(stderr, "The target service's index is already "
564 "in use. (%s)\n", usource);
565 if (errno == EINVAL) {
566 fprintf(stderr, "This may have multiple causes.\n");
568 fprintf(stderr, "Is '%s' the correct filesystem"
570 fprintf(stderr, "Are the mount options correct?\n");
571 fprintf(stderr, "Check the syslog for more info.\n");
574 /* May as well try to clean up loop devs */
575 if (strncmp(usource, "/dev/loop", 9) == 0) {
577 sprintf(cmd, "/sbin/losetup -d %s", usource);
581 } else if (!nomtab) {
582 rc = update_mtab_entry(usource, target, "lustre", orig_options,