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>
37 #include <lustre_ver.h>
41 #include "mount_utils.h"
43 #define MAX_HW_SECTORS_KB_PATH "queue/max_hw_sectors_kb"
44 #define MAX_SECTORS_KB_PATH "queue/max_sectors_kb"
50 char *progname = NULL;
54 fprintf(out, "%s v"LUSTRE_VERSION_STRING"\n", progname);
55 fprintf(out, "\nThis mount helper should only be invoked via the "
56 "mount (8) command,\ne.g. mount -t lustre dev dir\n\n");
57 fprintf(out, "usage: %s [-fhnv] [-o <mntopt>] <device> <mountpt>\n",
60 "\t<device>: the disk device, or for a client:\n"
61 "\t\t<mgmtnid>[:<altmgtnid>...]:/<filesystem>-client\n"
62 "\t<filesystem>: name of the Lustre filesystem (e.g. lustre1)\n"
63 "\t<mountpt>: filesystem mountpoint (e.g. /mnt/lustre)\n"
64 "\t-f|--fake: fake mount (updates /etc/mtab)\n"
65 "\t--force: force mount even if already in /etc/mtab\n"
66 "\t-h|--help: print this usage message\n"
67 "\t-n|--nomtab: do not update /etc/mtab after mount\n"
68 "\t-v|--verbose: print verbose config settings\n"
69 "\t<mntopt>: one or more comma separated of:\n"
70 "\t\t(no)flock,(no)user_xattr,(no)acl\n"
71 "\t\tnosvc: only start MGC/MGS obds\n"
72 "\t\tnomgs: only start target obds, using existing MGS\n"
73 "\t\texclude=<ostname>[:<ostname>] : colon-separated list of "
74 "inactive OSTs (e.g. lustre-OST0001)\n"
76 exit((out != stdout) ? EINVAL : 0);
79 static int check_mtab_entry(char *spec, char *mtpt, char *type)
84 fp = setmntent(MOUNTED, "r");
88 while ((mnt = getmntent(fp)) != NULL) {
89 if (strcmp(mnt->mnt_fsname, spec) == 0 &&
90 strcmp(mnt->mnt_dir, mtpt) == 0 &&
91 strcmp(mnt->mnt_type, type) == 0) {
102 update_mtab_entry(char *spec, char *mtpt, char *type, char *opts,
103 int flags, int freq, int pass)
109 mnt.mnt_fsname = spec;
112 mnt.mnt_opts = opts ? opts : "";
114 mnt.mnt_passno = pass;
116 fp = setmntent(MOUNTED, "a+");
118 fprintf(stderr, "%s: setmntent(%s): %s:",
119 progname, MOUNTED, strerror (errno));
122 if ((addmntent(fp, &mnt)) == 1) {
123 fprintf(stderr, "%s: addmntent: %s:",
124 progname, strerror (errno));
133 /* Get rid of symbolic hostnames for tcp, since kernel can't do lookups */
134 #define MAXNIDSTR 1024
135 static char *convert_hostnames(char *s1)
137 char *converted, *s2 = 0, *c;
139 int left = MAXNIDSTR;
142 converted = malloc(left);
144 while ((left > 0) && (*s1 != '/')) {
145 s2 = strpbrk(s1, ",:");
150 nid = libcfs_str2nid(s1);
151 *s2 = sep; /* back to original string */
152 if (nid == LNET_NID_ANY)
154 c += snprintf(c, left, "%s%c", libcfs_nid2str(nid), sep);
155 left = converted + MAXNIDSTR - c;
158 snprintf(c, left, "%s", s1);
161 fprintf(stderr, "%s: Can't parse NID '%s'\n", progname, s1);
166 /*****************************************************************************
168 * This part was cribbed from util-linux/mount/mount.c. There was no clear
169 * license information, but many other files in the package are identified as
170 * GNU GPL, so it's a pretty safe bet that was their intent.
172 ****************************************************************************/
174 const char *opt; /* option name */
175 int inv; /* true if flag value should be inverted */
176 int mask; /* flag mask value */
179 static const struct opt_map opt_map[] = {
180 /*"optname", inv,ms_mask */
181 /* These flags are parsed by mount, not lustre */
182 { "defaults", 0, 0 }, /* default options */
183 { "remount", 0, MS_REMOUNT}, /* remount with different options */
184 { "rw", 1, MS_RDONLY }, /* read-write */
185 { "ro", 0, MS_RDONLY }, /* read-only */
186 { "exec", 1, MS_NOEXEC }, /* permit execution of binaries */
187 { "noexec", 0, MS_NOEXEC }, /* don't execute binaries */
188 { "suid", 1, MS_NOSUID }, /* honor suid executables */
189 { "nosuid", 0, MS_NOSUID }, /* don't honor suid executables */
190 { "dev", 1, MS_NODEV }, /* interpret device files */
191 { "nodev", 0, MS_NODEV }, /* don't interpret devices */
192 { "sync", 0, MS_SYNCHRONOUS}, /* synchronous I/O */
193 { "async", 1, MS_SYNCHRONOUS}, /* asynchronous I/O */
194 { "atime", 1, MS_NOATIME }, /* set file access time on read */
195 { "noatime", 0, MS_NOATIME }, /* do not set file access time on read */
197 { "diratime", 1, MS_NODIRATIME }, /* set file access time on read */
198 { "nodiratime",0,MS_NODIRATIME }, /* do not set file access time on read */
201 { "relatime", 0, MS_RELATIME }, /* set file access time on read */
202 { "norelatime",1,MS_RELATIME }, /* do not set file access time on read */
204 { "auto", 0, 0 }, /* Can be mounted using -a */
205 { "noauto", 0, 0 }, /* Can only be mounted explicitly */
206 { "nousers", 1, 0 }, /* Forbid ordinary user to mount */
207 { "nouser", 1, 0 }, /* Forbid ordinary user to mount */
208 { "noowner", 1, 0 }, /* Device owner has no special privs */
209 { "_netdev", 0, 0 }, /* Device accessible only via network */
213 /****************************************************************************/
215 /* 1 = don't pass on to lustre
216 0 = pass on to lustre */
217 static int parse_one_option(const char *check, int *flagp)
219 const struct opt_map *opt;
221 for (opt = &opt_map[0]; opt->opt != NULL; opt++) {
222 if (strncmp(check, opt->opt, strlen(opt->opt)) == 0) {
225 *flagp &= ~(opt->mask);
232 /* Assume any unknown options are valid and pass them on. The mount
233 will fail if lmd_parse, ll_options or ldiskfs doesn't recognize it.*/
237 /* Replace options with subset of Lustre-specific options, and
238 fill in mount flags */
239 int parse_options(char *orig_options, int *flagp)
241 char *options, *opt, *nextopt;
243 options = calloc(strlen(orig_options) + 1, 1);
245 nextopt = orig_options;
246 while ((opt = strsep(&nextopt, ","))) {
250 if (parse_one_option(opt, flagp) == 0) {
251 /* pass this on as an option */
253 strcat(options, ",");
254 strcat(options, opt);
257 strcpy(orig_options, options);
263 int read_file(char *path, char *buf, int size)
267 fd = fopen(path, "r");
271 fgets(buf, size, fd);
276 int write_file(char *path, char *buf)
280 fd = fopen(path, "w");
289 /* This is to tune the kernel for good SCSI performance.
290 * For that we set the value of /sys/block/{dev}/queue/max_sectors_kb
291 * to the value of /sys/block/{dev}/queue/max_hw_sectors_kb */
292 int set_tunables(char *source, int src_len)
295 struct stat stat_buf;
296 char *chk_major, *chk_minor;
297 char *savept, *dev, *s2 = 0;
299 char buf[PATH_MAX] = {'\0'}, path[PATH_MAX] = {'\0'};
300 char real_path[PATH_MAX] = {'\0'};
307 ret_path = realpath(source, real_path);
308 if (ret_path == NULL) {
310 fprintf(stderr, "warning: %s: cannot resolve: %s",
311 source, strerror(errno));
315 src_len = sizeof(real_path);
317 if (strncmp(real_path, "/dev/loop", 9) == 0)
320 if ((real_path[0] != '/') && ((s2 = strpbrk(real_path, ",:")) != NULL))
323 dev = real_path + src_len - 1;
324 while (dev > real_path && (*dev != '/')) {
329 snprintf(path, sizeof(path), "/sys/block%s/%s", dev,
330 MAX_HW_SECTORS_KB_PATH);
331 rc = read_file(path, buf, sizeof(buf));
332 if (rc == 0 && (strlen(buf) - 1) > 0) {
333 snprintf(path, sizeof(path), "/sys/block%s/%s", dev,
334 MAX_SECTORS_KB_PATH);
335 rc = write_file(path, buf);
337 fprintf(stderr, "warning: opening %s: %s\n",
338 path, strerror(errno));
345 /* The name of the device say 'X' specified in /dev/X may not
346 * match any entry under /sys/block/. In that case we need to
347 * match the major/minor number to find the entry under
348 * sys/block corresponding to /dev/X */
349 dev = real_path + src_len - 1;
350 while (dev > real_path) {
356 rc = stat(dev, &stat_buf);
359 fprintf(stderr, "warning: %s, device %s stat failed\n",
360 strerror(errno), dev);
364 major = major(stat_buf.st_rdev);
365 minor = minor(stat_buf.st_rdev);
366 rc = glob("/sys/block/*", GLOB_NOSORT, NULL, &glob_info);
369 fprintf(stderr, "warning: failed to read entries under "
374 for (i = 0; i < glob_info.gl_pathc; i++){
375 snprintf(path, sizeof(path), "%s/dev", glob_info.gl_pathv[i]);
377 rc = read_file(path, buf, sizeof(buf));
381 if (buf[strlen(buf) - 1] == '\n')
382 buf[strlen(buf) - 1] = '\0';
384 chk_major = strtok_r(buf, ":", &savept);
386 if (major == atoi(chk_major) &&minor == atoi(chk_minor))
390 if (i == glob_info.gl_pathc) {
392 fprintf(stderr,"warning: device %s does not match any "
393 "entry under /sys/block\n", real_path);
398 snprintf(path, sizeof(path), "%s/%s", glob_info.gl_pathv[i],
399 MAX_HW_SECTORS_KB_PATH);
400 rc = read_file(path, buf, sizeof(buf));
403 fprintf(stderr, "warning: opening %s: %s\n",
404 path, strerror(errno));
408 if (strlen(buf) - 1 > 0) {
409 snprintf(path, sizeof(path), "%s/%s",
410 glob_info.gl_pathv[i], MAX_SECTORS_KB_PATH);
411 rc = write_file(path, buf);
413 fprintf(stderr, "warning: writing to %s: %s\n",
414 path, strerror(errno));
418 globfree(&glob_info);
422 int main(int argc, char *const argv[])
424 char default_options[] = "";
425 char *usource, *source, *target, *ptr;
426 char *options, *optcopy, *orig_options = default_options;
427 int i, nargs = 3, opt, rc, flags, optlen;
428 static struct option long_opt[] = {
432 {"nomtab", 0, 0, 'n'},
433 {"options", 1, 0, 'o'},
434 {"verbose", 0, 0, 'v'},
438 progname = strrchr(argv[0], '/');
439 progname = progname ? progname + 1 : argv[0];
441 while ((opt = getopt_long(argc, argv, "fhno:v",
442 long_opt, NULL)) != EOF){
446 printf("force: %d\n", force);
451 printf("fake: %d\n", fake);
459 printf("nomtab: %d\n", nomtab);
463 orig_options = optarg;
471 fprintf(stderr, "%s: unknown option '%c'\n",
478 if (optind + 2 > argc) {
479 fprintf(stderr, "%s: too few arguments\n", progname);
483 usource = argv[optind];
484 source = convert_hostnames(usource);
485 target = argv[optind + 1];
486 ptr = target + strlen(target) - 1;
487 while ((ptr > target) && (*ptr == '/')) {
492 if (!usource || !source) {
497 for (i = 0; i < argc; i++)
498 printf("arg[%d] = %s\n", i, argv[i]);
499 printf("source = %s (%s), target = %s\n", usource, source,
501 printf("options = %s\n", orig_options);
504 options = malloc(strlen(orig_options) + 1);
505 strcpy(options, orig_options);
506 rc = parse_options(options, &flags);
508 fprintf(stderr, "%s: can't parse options: %s\n",
514 rc = check_mtab_entry(usource, target, "lustre");
515 if (rc && !(flags & MS_REMOUNT)) {
516 fprintf(stderr, "%s: according to %s %s is "
517 "already mounted on %s\n",
518 progname, MOUNTED, usource, target);
521 if (!rc && (flags & MS_REMOUNT)) {
522 fprintf(stderr, "%s: according to %s %s is "
523 "not already mounted on %s\n",
524 progname, MOUNTED, usource, target);
528 if (flags & MS_REMOUNT)
531 rc = access(target, F_OK);
534 fprintf(stderr, "%s: %s inaccessible: %s\n", progname, target,
539 /* In Linux 2.4, the target device doesn't get passed to any of our
540 functions. So we'll stick it on the end of the options. */
541 optlen = strlen(options) + strlen(",device=") + strlen(source) + 1;
542 optcopy = malloc(optlen);
543 strcpy(optcopy, options);
545 strcat(optcopy, ",");
546 strcat(optcopy, "device=");
547 strcat(optcopy, source);
550 printf("mounting device %s at %s, flags=%#x options=%s\n",
551 source, target, flags, optcopy);
553 if (set_tunables(source, strlen(source)) && verbose)
554 fprintf(stderr, "%s: unable to set tunables for %s"
555 " (may cause reduced IO performance)\n",
558 register_service_tags(usource, source, target);
561 /* flags and target get to lustre_get_sb, but not
562 lustre_fill_super. Lustre ignores the flags, but mount
564 rc = mount(source, target, "lustre", flags, (void *)optcopy);
571 cli = strrchr(usource, ':');
572 if (cli && (strlen(cli) > 2))
577 fprintf(stderr, "%s: mount %s at %s failed: %s\n", progname,
578 usource, target, strerror(errno));
580 fprintf(stderr, "Are the lustre modules loaded?\n"
581 "Check /etc/modprobe.conf and /proc/filesystems"
582 "\nNote 'alias lustre llite' should be removed"
583 " from modprobe.conf\n");
584 if (errno == ENOTBLK)
585 fprintf(stderr, "Do you need -o loop?\n");
586 if (errno == ENOMEDIUM)
588 "This filesystem needs at least 1 OST\n");
589 if (errno == ENOENT) {
590 fprintf(stderr, "Is the MGS specification correct?\n");
591 fprintf(stderr, "Is the filesystem name correct?\n");
592 fprintf(stderr, "If upgrading, is the copied client log"
593 " valid? (see upgrade docs)\n");
595 if (errno == EALREADY)
596 fprintf(stderr, "The target service is already running."
599 fprintf(stderr, "The target service failed to start "
600 "(bad config log?) (%s). "
601 "See /var/log/messages.\n", usource);
603 fprintf(stderr, "Is the MGS running?\n");
604 if (errno == EADDRINUSE)
605 fprintf(stderr, "The target service's index is already "
606 "in use. (%s)\n", usource);
607 if (errno == EINVAL) {
608 fprintf(stderr, "This may have multiple causes.\n");
610 fprintf(stderr, "Is '%s' the correct filesystem"
612 fprintf(stderr, "Are the mount options correct?\n");
613 fprintf(stderr, "Check the syslog for more info.\n");
616 /* May as well try to clean up loop devs */
617 if (strncmp(usource, "/dev/loop", 9) == 0) {
619 sprintf(cmd, "/sbin/losetup -d %s", usource);
623 } else if (!nomtab) {
624 rc = update_mtab_entry(usource, target, "lustre", orig_options,