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>
42 #define MAX_HW_SECTORS_KB_PATH "queue/max_hw_sectors_kb"
43 #define MAX_SECTORS_KB_PATH "queue/max_sectors_kb"
49 static char *progname = NULL;
53 fprintf(out, "%s v"LUSTRE_VERSION_STRING"\n", progname);
54 fprintf(out, "\nThis mount helper should only be invoked via the "
55 "mount (8) command,\ne.g. mount -t lustre dev dir\n\n");
56 fprintf(out, "usage: %s [-fhnv] [-o <mntopt>] <device> <mountpt>\n",
59 "\t<device>: the disk device, or for a client:\n"
60 "\t\t<mgmtnid>[:<altmgtnid>...]:/<filesystem>-client\n"
61 "\t<filesystem>: name of the Lustre filesystem (e.g. lustre1)\n"
62 "\t<mountpt>: filesystem mountpoint (e.g. /mnt/lustre)\n"
63 "\t-f|--fake: fake mount (updates /etc/mtab)\n"
64 "\t--force: force mount even if already in /etc/mtab\n"
65 "\t-h|--help: print this usage message\n"
66 "\t-n|--nomtab: do not update /etc/mtab after mount\n"
67 "\t-v|--verbose: print verbose config settings\n"
68 "\t<mntopt>: one or more comma separated of:\n"
69 "\t\t(no)flock,(no)user_xattr,(no)acl\n"
70 "\t\tnosvc: only start MGC/MGS obds\n"
71 "\t\texclude=<ostname>[:<ostname>] : colon-separated list of "
72 "inactive OSTs (e.g. lustre-OST0001)\n"
74 exit((out != stdout) ? EINVAL : 0);
77 static int check_mtab_entry(char *spec, char *mtpt, char *type)
82 fp = setmntent(MOUNTED, "r");
86 while ((mnt = getmntent(fp)) != NULL) {
87 if (strcmp(mnt->mnt_fsname, spec) == 0 &&
88 strcmp(mnt->mnt_dir, mtpt) == 0 &&
89 strcmp(mnt->mnt_type, type) == 0) {
100 update_mtab_entry(char *spec, char *mtpt, char *type, char *opts,
101 int flags, int freq, int pass)
107 mnt.mnt_fsname = spec;
110 mnt.mnt_opts = opts ? opts : "";
112 mnt.mnt_passno = pass;
114 fp = setmntent(MOUNTED, "a+");
116 fprintf(stderr, "%s: setmntent(%s): %s:",
117 progname, MOUNTED, strerror (errno));
120 if ((addmntent(fp, &mnt)) == 1) {
121 fprintf(stderr, "%s: addmntent: %s:",
122 progname, strerror (errno));
131 /* Get rid of symbolic hostnames for tcp, since kernel can't do lookups */
132 #define MAXNIDSTR 1024
133 static char *convert_hostnames(char *s1)
135 char *converted, *s2 = 0, *c;
137 int left = MAXNIDSTR;
140 converted = malloc(left);
142 while ((left > 0) && (*s1 != '/')) {
143 s2 = strpbrk(s1, ",:");
148 nid = libcfs_str2nid(s1);
149 *s2 = sep; /* back to original string */
150 if (nid == LNET_NID_ANY)
152 c += snprintf(c, left, "%s%c", libcfs_nid2str(nid), sep);
153 left = converted + MAXNIDSTR - c;
156 snprintf(c, left, "%s", s1);
159 fprintf(stderr, "%s: Can't parse NID '%s'\n", progname, s1);
164 /*****************************************************************************
166 * This part was cribbed from util-linux/mount/mount.c. There was no clear
167 * license information, but many other files in the package are identified as
168 * GNU GPL, so it's a pretty safe bet that was their intent.
170 ****************************************************************************/
172 const char *opt; /* option name */
173 int inv; /* true if flag value should be inverted */
174 int mask; /* flag mask value */
177 static const struct opt_map opt_map[] = {
178 /*"optname", inv,ms_mask */
179 /* These flags are parsed by mount, not lustre */
180 { "defaults", 0, 0 }, /* default options */
181 { "remount", 0, MS_REMOUNT}, /* remount with different options */
182 { "rw", 1, MS_RDONLY }, /* read-write */
183 { "ro", 0, MS_RDONLY }, /* read-only */
184 { "exec", 1, MS_NOEXEC }, /* permit execution of binaries */
185 { "noexec", 0, MS_NOEXEC }, /* don't execute binaries */
186 { "suid", 1, MS_NOSUID }, /* honor suid executables */
187 { "nosuid", 0, MS_NOSUID }, /* don't honor suid executables */
188 { "dev", 1, MS_NODEV }, /* interpret device files */
189 { "nodev", 0, MS_NODEV }, /* don't interpret devices */
190 { "sync", 0, MS_SYNCHRONOUS}, /* synchronous I/O */
191 { "async", 1, MS_SYNCHRONOUS}, /* asynchronous I/O */
192 { "auto", 0, 0 }, /* Can be mounted using -a */
193 { "noauto", 0, 0 }, /* Can only be mounted explicitly */
194 { "nousers", 1, 0 }, /* Forbid ordinary user to mount */
195 { "nouser", 1, 0 }, /* Forbid ordinary user to mount */
196 { "noowner", 1, 0 }, /* Device owner has no special privs */
197 { "_netdev", 0, 0 }, /* Device accessible only via network */
200 /****************************************************************************/
202 /* 1 = don't pass on to lustre
203 0 = pass on to lustre */
204 static int parse_one_option(const char *check, int *flagp)
206 const struct opt_map *opt;
208 for (opt = &opt_map[0]; opt->opt != NULL; opt++) {
209 if (strncmp(check, opt->opt, strlen(opt->opt)) == 0) {
212 *flagp &= ~(opt->mask);
219 /* Assume any unknown options are valid and pass them on. The mount
220 will fail if lmd_parse, ll_options or ldiskfs doesn't recognize it.*/
224 /* Replace options with subset of Lustre-specific options, and
225 fill in mount flags */
226 int parse_options(char *orig_options, int *flagp)
228 char *options, *opt, *nextopt;
230 options = calloc(strlen(orig_options) + 1, 1);
232 nextopt = orig_options;
233 while ((opt = strsep(&nextopt, ","))) {
237 if (parse_one_option(opt, flagp) == 0) {
238 /* pass this on as an option */
240 strcat(options, ",");
241 strcat(options, opt);
244 strcpy(orig_options, options);
250 int read_file(char *path, char *buf, int size)
254 fd = fopen(path, "r");
258 fgets(buf, size, fd);
263 int write_file(char *path, char *buf)
267 fd = fopen(path, "w");
276 /* This is to tune the kernel for good SCSI performance.
277 * For that we set the value of /sys/block/{dev}/queue/max_sectors_kb
278 * to the value of /sys/block/{dev}/queue/max_hw_sectors_kb */
279 int set_tunables(char *source, int src_len)
282 struct stat stat_buf;
283 char *chk_major, *chk_minor;
284 char *savept, *dev, *s2 = 0;
285 char buf[PATH_MAX], path[PATH_MAX];
292 if (strncmp(source, "/dev/loop", 9) == 0)
295 if ((*source != '/') && ((s2 = strpbrk(source, ",:")) != NULL))
298 dev = source + src_len - 1;
299 while (dev > source && (*dev != '/')) {
304 snprintf(path, sizeof(path), "/sys/block%s/%s", dev,
305 MAX_HW_SECTORS_KB_PATH);
306 rc = read_file(path, buf, sizeof(buf));
307 if (rc == 0 && (strlen(buf) - 1) > 0) {
308 snprintf(path, sizeof(path), "/sys/block%s/%s", dev,
309 MAX_SECTORS_KB_PATH);
310 rc = write_file(path, buf);
312 fprintf(stderr, "warning: opening %s: %s\n",
313 path, strerror(errno));
320 /* The name of the device say 'X' specified in /dev/X may not
321 * match any entry under /sys/block/. In that case we need to
322 * match the major/minor number to find the entry under
323 * sys/block corresponding to /dev/X */
324 dev = source + src_len - 1;
325 while (dev > source) {
331 rc = stat(dev, &stat_buf);
334 fprintf(stderr, "warning: %s, device %s stat failed\n",
335 strerror(errno), dev);
339 major = major(stat_buf.st_rdev);
340 minor = minor(stat_buf.st_rdev);
341 rc = glob("/sys/block/*", GLOB_NOSORT, NULL, &glob_info);
344 fprintf(stderr, "warning: failed to read entries under "
349 for (i = 0; i < glob_info.gl_pathc; i++){
350 snprintf(path, sizeof(path), "%s/dev", glob_info.gl_pathv[i]);
352 rc = read_file(path, buf, sizeof(buf));
356 if (buf[strlen(buf) - 1] == '\n')
357 buf[strlen(buf) - 1] = '\0';
359 chk_major = strtok_r(buf, ":", &savept);
361 if (major == atoi(chk_major) &&minor == atoi(chk_minor))
365 if (i == glob_info.gl_pathc) {
367 fprintf(stderr,"warning: device %s does not match any "
368 "entry under /sys/block\n", source);
372 snprintf(path, sizeof(path), "%s/%s", glob_info.gl_pathv[i],
373 MAX_HW_SECTORS_KB_PATH);
374 rc = read_file(path, buf, sizeof(buf));
377 fprintf(stderr, "warning: opening %s: %s\n",
378 path, strerror(errno));
382 if (strlen(buf) - 1 > 0) {
383 snprintf(path, sizeof(path), "%s/%s",
384 glob_info.gl_pathv[i], MAX_SECTORS_KB_PATH);
385 rc = write_file(path, buf);
387 fprintf(stderr, "warning: writing to %s: %s\n",
388 path, strerror(errno));
393 int main(int argc, char *const argv[])
395 char default_options[] = "";
396 char *usource, *source, *target, *ptr;
397 char *options, *optcopy, *orig_options = default_options;
398 int i, nargs = 3, opt, rc, flags, optlen;
399 static struct option long_opt[] = {
403 {"nomtab", 0, 0, 'n'},
404 {"options", 1, 0, 'o'},
405 {"verbose", 0, 0, 'v'},
409 progname = strrchr(argv[0], '/');
410 progname = progname ? progname + 1 : argv[0];
412 while ((opt = getopt_long(argc, argv, "fhno:v",
413 long_opt, NULL)) != EOF){
417 printf("force: %d\n", force);
422 printf("fake: %d\n", fake);
430 printf("nomtab: %d\n", nomtab);
434 orig_options = optarg;
442 fprintf(stderr, "%s: unknown option '%c'\n",
449 if (optind + 2 > argc) {
450 fprintf(stderr, "%s: too few arguments\n", progname);
454 usource = argv[optind];
455 source = convert_hostnames(usource);
456 target = argv[optind + 1];
457 ptr = target + strlen(target) - 1;
458 while ((ptr > target) && (*ptr == '/')) {
463 if (!usource || !source) {
468 for (i = 0; i < argc; i++)
469 printf("arg[%d] = %s\n", i, argv[i]);
470 printf("source = %s (%s), target = %s\n", usource, source, target);
471 printf("options = %s\n", orig_options);
474 options = malloc(strlen(orig_options) + 1);
475 strcpy(options, orig_options);
476 rc = parse_options(options, &flags);
478 fprintf(stderr, "%s: can't parse options: %s\n",
484 rc = check_mtab_entry(usource, target, "lustre");
485 if (rc && !(flags & MS_REMOUNT)) {
486 fprintf(stderr, "%s: according to %s %s is "
487 "already mounted on %s\n",
488 progname, MOUNTED, usource, target);
491 if (!rc && (flags & MS_REMOUNT)) {
492 fprintf(stderr, "%s: according to %s %s is "
493 "not already mounted on %s\n",
494 progname, MOUNTED, usource, target);
498 if (flags & MS_REMOUNT)
501 rc = access(target, F_OK);
504 fprintf(stderr, "%s: %s inaccessible: %s\n", progname, target,
509 /* In Linux 2.4, the target device doesn't get passed to any of our
510 functions. So we'll stick it on the end of the options. */
511 optlen = strlen(options) + strlen(",device=") + strlen(source) + 1;
512 optcopy = malloc(optlen);
513 strcpy(optcopy, options);
515 strcat(optcopy, ",");
516 strcat(optcopy, "device=");
517 strcat(optcopy, source);
520 printf("mounting device %s at %s, flags=%#x options=%s\n",
521 source, target, flags, optcopy);
523 if (set_tunables(source, strlen(source)) && verbose)
524 fprintf(stderr, "%s: unable to set tunables for %s"
525 " (may cause reduced IO performance)",
529 /* flags and target get to lustre_get_sb, but not
530 lustre_fill_super. Lustre ignores the flags, but mount
532 rc = mount(source, target, "lustre", flags, (void *)optcopy);
539 cli = strrchr(usource, ':');
540 if (cli && (strlen(cli) > 2))
545 fprintf(stderr, "%s: mount %s at %s failed: %s\n", progname,
546 usource, target, strerror(errno));
548 fprintf(stderr, "Are the lustre modules loaded?\n"
549 "Check /etc/modprobe.conf and /proc/filesystems"
550 "\nNote 'alias lustre llite' should be removed"
551 " from modprobe.conf\n");
552 if (errno == ENOTBLK)
553 fprintf(stderr, "Do you need -o loop?\n");
554 if (errno == ENOMEDIUM)
556 "This filesystem needs at least 1 OST\n");
557 if (errno == ENOENT) {
558 fprintf(stderr, "Is the MGS specification correct?\n");
559 fprintf(stderr, "Is the filesystem name correct?\n");
560 fprintf(stderr, "If upgrading, is the copied client log"
561 " valid? (see upgrade docs)\n");
563 if (errno == EALREADY)
564 fprintf(stderr, "The target service is already running."
567 fprintf(stderr, "The target service failed to start "
568 "(bad config log?) (%s). "
569 "See /var/log/messages.\n", usource);
571 fprintf(stderr, "Is the MGS running?\n");
572 if (errno == EADDRINUSE)
573 fprintf(stderr, "The target service's index is already "
574 "in use. (%s)\n", usource);
575 if (errno == EINVAL) {
576 fprintf(stderr, "This may have multiple causes.\n");
578 fprintf(stderr, "Is '%s' the correct filesystem"
580 fprintf(stderr, "Are the mount options correct?\n");
581 fprintf(stderr, "Check the syslog for more info.\n");
584 /* May as well try to clean up loop devs */
585 if (strncmp(usource, "/dev/loop", 9) == 0) {
587 sprintf(cmd, "/sbin/losetup -d %s", usource);
591 } else if (!nomtab) {
592 rc = update_mtab_entry(usource, target, "lustre", orig_options,