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 (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
32 * Copyright (c) 2011, 2012, Whamcloud, Inc.
35 * This file is part of Lustre, http://www.lustre.org/
36 * Lustre is a trademark of Sun Microsystems, Inc.
38 * lustre/utils/mount_lustre.c
40 * Author: Robert Read <rread@clusterfs.com>
41 * Author: Nathan Rutman <nathan@clusterfs.com>
53 #include <sys/mount.h>
58 #include <lustre_ver.h>
62 #include "mount_utils.h"
64 #define MAX_HW_SECTORS_KB_PATH "queue/max_hw_sectors_kb"
65 #define MAX_SECTORS_KB_PATH "queue/max_sectors_kb"
66 #define STRIPE_CACHE_SIZE "md/stripe_cache_size"
67 #define MAX_RETRIES 99
74 int md_stripe_cache_size = 16384;
75 char *progname = NULL;
79 fprintf(out, "%s v"LUSTRE_VERSION_STRING"\n", progname);
80 fprintf(out, "\nThis mount helper should only be invoked via the "
81 "mount (8) command,\ne.g. mount -t lustre dev dir\n\n");
82 fprintf(out, "usage: %s [-fhnv] [-o <mntopt>] <device> <mountpt>\n",
85 "\t<device>: the disk device, or for a client:\n"
86 "\t\t<mgmtnid>[:<altmgtnid>...]:/<filesystem>-client\n"
87 "\t<filesystem>: name of the Lustre filesystem (e.g. lustre1)\n"
88 "\t<mountpt>: filesystem mountpoint (e.g. /mnt/lustre)\n"
89 "\t-f|--fake: fake mount (updates /etc/mtab)\n"
90 "\t-o force|--force: force mount even if already in /etc/mtab\n"
91 "\t-h|--help: print this usage message\n"
92 "\t-n|--nomtab: do not update /etc/mtab after mount\n"
93 "\t-v|--verbose: print verbose config settings\n"
94 "\t<mntopt>: one or more comma separated of:\n"
95 "\t\t(no)flock,(no)user_xattr,(no)acl\n"
96 "\t\tabort_recov: abort server recovery handling\n"
97 "\t\tnosvc: only start MGC/MGS obds\n"
98 "\t\tnomgs: only start target obds, using existing MGS\n"
99 "\t\texclude=<ostname>[:<ostname>] : colon-separated list of "
100 "inactive OSTs (e.g. lustre-OST0001)\n"
101 "\t\tretry=<num>: number of times mount is retried by client\n"
102 "\t\tmd_stripe_cache_size=<num>: set the raid stripe cache "
103 "size for the underlying raid if present\n"
105 exit((out != stdout) ? EINVAL : 0);
108 static int check_mtab_entry(char *spec1, char *spec2, char *mtpt, char *type)
113 fp = setmntent(MOUNTED, "r");
117 while ((mnt = getmntent(fp)) != NULL) {
118 if ((strcmp(mnt->mnt_fsname, spec1) == 0 ||
119 strcmp(mnt->mnt_fsname, spec2) == 0) &&
120 strcmp(mnt->mnt_dir, mtpt) == 0 &&
121 strcmp(mnt->mnt_type, type) == 0) {
132 update_mtab_entry(char *spec, char *mtpt, char *type, char *opts,
133 int flags, int freq, int pass)
139 mnt.mnt_fsname = spec;
142 mnt.mnt_opts = opts ? opts : "";
144 mnt.mnt_passno = pass;
146 fp = setmntent(MOUNTED, "a+");
148 fprintf(stderr, "%s: setmntent(%s): %s:",
149 progname, MOUNTED, strerror (errno));
152 if ((addmntent(fp, &mnt)) == 1) {
153 fprintf(stderr, "%s: addmntent: %s:",
154 progname, strerror (errno));
163 /* Get rid of symbolic hostnames for tcp, since kernel can't do lookups */
164 #define MAXNIDSTR 1024
165 static char *convert_hostnames(char *s1)
167 char *converted, *s2 = 0, *c;
169 int left = MAXNIDSTR;
172 converted = malloc(left);
173 if (converted == NULL) {
174 fprintf(stderr, "out of memory: needed %d bytes\n",
179 while ((left > 0) && (*s1 != '/')) {
180 s2 = strpbrk(s1, ",:");
185 nid = libcfs_str2nid(s1);
186 *s2 = sep; /* back to original string */
187 if (nid == LNET_NID_ANY)
189 c += snprintf(c, left, "%s%c", libcfs_nid2str(nid), sep);
190 left = converted + MAXNIDSTR - c;
193 snprintf(c, left, "%s", s1);
196 fprintf(stderr, "%s: Can't parse NID '%s'\n", progname, s1);
201 /*****************************************************************************
203 * This part was cribbed from util-linux/mount/mount.c. There was no clear
204 * license information, but many other files in the package are identified as
205 * GNU GPL, so it's a pretty safe bet that was their intent.
207 ****************************************************************************/
209 const char *opt; /* option name */
210 int inv; /* true if flag value should be inverted */
211 int mask; /* flag mask value */
214 static const struct opt_map opt_map[] = {
215 /*"optname", inv,ms_mask */
216 /* These flags are parsed by mount, not lustre */
217 { "defaults", 0, 0 }, /* default options */
218 { "remount", 0, MS_REMOUNT}, /* remount with different options */
219 { "rw", 1, MS_RDONLY }, /* read-write */
220 { "ro", 0, MS_RDONLY }, /* read-only */
221 { "exec", 1, MS_NOEXEC }, /* permit execution of binaries */
222 { "noexec", 0, MS_NOEXEC }, /* don't execute binaries */
223 { "suid", 1, MS_NOSUID }, /* honor suid executables */
224 { "nosuid", 0, MS_NOSUID }, /* don't honor suid executables */
225 { "dev", 1, MS_NODEV }, /* interpret device files */
226 { "nodev", 0, MS_NODEV }, /* don't interpret devices */
227 { "sync", 0, MS_SYNCHRONOUS}, /* synchronous I/O */
228 { "async", 1, MS_SYNCHRONOUS}, /* asynchronous I/O */
229 { "atime", 1, MS_NOATIME }, /* set file access time on read */
230 { "noatime", 0, MS_NOATIME }, /* do not set file access time on read */
232 { "diratime", 1, MS_NODIRATIME }, /* set file access time on read */
233 { "nodiratime",0,MS_NODIRATIME }, /* do not set file access time on read */
236 { "relatime", 0, MS_RELATIME }, /* set file access time on read */
237 { "norelatime",1,MS_RELATIME }, /* do not set file access time on read */
239 #ifdef MS_STRICTATIME
240 { "strictatime",0,MS_STRICTATIME }, /* update access time strictly */
242 { "auto", 0, 0 }, /* Can be mounted using -a */
243 { "noauto", 0, 0 }, /* Can only be mounted explicitly */
244 { "nousers", 1, 0 }, /* Forbid ordinary user to mount */
245 { "nouser", 1, 0 }, /* Forbid ordinary user to mount */
246 { "noowner", 1, 0 }, /* Device owner has no special privs */
247 { "_netdev", 0, 0 }, /* Device accessible only via network */
251 /****************************************************************************/
253 /* 1 = don't pass on to lustre
254 0 = pass on to lustre */
255 static int parse_one_option(const char *check, int *flagp)
257 const struct opt_map *opt;
259 for (opt = &opt_map[0]; opt->opt != NULL; opt++) {
260 if (strncmp(check, opt->opt, strlen(opt->opt)) == 0) {
263 *flagp &= ~(opt->mask);
270 /* Assume any unknown options are valid and pass them on. The mount
271 will fail if lmd_parse, ll_options or ldiskfs doesn't recognize it.*/
275 static void append_option(char *options, const char *one)
278 strcat(options, ",");
279 strcat(options, one);
282 /* Replace options with subset of Lustre-specific options, and
283 fill in mount flags */
284 int parse_options(char *orig_options, int *flagp)
286 char *options, *opt, *nextopt, *arg, *val;
288 options = calloc(strlen(orig_options) + 1, 1);
290 nextopt = orig_options;
291 while ((opt = strsep(&nextopt, ","))) {
296 /* Handle retries in a slightly different
299 val = strchr(opt, '=');
300 /* please note that some ldiskfs mount options are also in the form
301 * of param=value. We should pay attention not to remove those
302 * mount options, see bug 22097. */
303 if (val && strncmp(arg, "md_stripe_cache_size", 20) == 0) {
304 md_stripe_cache_size = atoi(val + 1);
305 } else if (val && strncmp(arg, "retry", 5) == 0) {
306 retry = atoi(val + 1);
307 if (retry > MAX_RETRIES)
311 } else if (val && strncmp(arg, "mgssec", 6) == 0) {
312 append_option(options, opt);
313 } else if (strcmp(opt, "force") == 0) {
314 //XXX special check for 'force' option
316 printf("force: %d\n", force);
317 } else if (parse_one_option(opt, flagp) == 0) {
318 /* pass this on as an option */
319 append_option(options, opt);
322 #ifdef MS_STRICTATIME
323 /* set strictatime to default if NOATIME or RELATIME
324 not given explicit */
325 if (!(*flagp & (MS_NOATIME | MS_RELATIME)))
326 *flagp |= MS_STRICTATIME;
328 strcpy(orig_options, options);
334 int read_file(char *path, char *buf, int size)
338 fd = fopen(path, "r");
342 /* should not ignore fgets(3)'s return value */
343 if (!fgets(buf, size, fd)) {
344 fprintf(stderr, "reading from %s: %s", path, strerror(errno));
352 int write_file(char *path, char *buf)
356 fd = fopen(path, "w");
365 /* This is to tune the kernel for good SCSI performance.
366 * For that we set the value of /sys/block/{dev}/queue/max_sectors_kb
367 * to the value of /sys/block/{dev}/queue/max_hw_sectors_kb */
368 int set_blockdev_tunables(char *source, int fan_out)
370 glob_t glob_info = { 0 };
371 struct stat stat_buf;
372 char *chk_major, *chk_minor;
375 char buf[PATH_MAX] = {'\0'}, path[PATH_MAX] = {'\0'};
376 char real_path[PATH_MAX] = {'\0'};
383 ret_path = realpath(source, real_path);
384 if (ret_path == NULL) {
386 fprintf(stderr, "warning: %s: cannot resolve: %s\n",
387 source, strerror(errno));
391 if (strncmp(real_path, "/dev/loop", 9) == 0)
394 if ((real_path[0] != '/') && (strpbrk(real_path, ",:") != NULL))
397 snprintf(path, sizeof(path), "/sys/block%s", real_path + 4);
398 if (access(path, X_OK) == 0)
401 /* The name of the device say 'X' specified in /dev/X may not
402 * match any entry under /sys/block/. In that case we need to
403 * match the major/minor number to find the entry under
404 * sys/block corresponding to /dev/X */
406 /* Don't chop tail digit on /dev/mapper/xxx, LU-478 */
407 if (strncmp(real_path, "/dev/mapper", 11) != 0) {
408 dev = real_path + strlen(real_path);
409 while (--dev > real_path && isdigit(*dev))
412 if (strncmp(real_path, "/dev/md_", 8) == 0)
416 rc = stat(real_path, &stat_buf);
419 fprintf(stderr, "warning: %s, device %s stat failed\n",
420 strerror(errno), real_path);
424 major = major(stat_buf.st_rdev);
425 minor = minor(stat_buf.st_rdev);
426 rc = glob("/sys/block/*", GLOB_NOSORT, NULL, &glob_info);
429 fprintf(stderr, "warning: failed to read entries under "
431 globfree(&glob_info);
435 for (i = 0; i < glob_info.gl_pathc; i++){
436 snprintf(path, sizeof(path), "%s/dev", glob_info.gl_pathv[i]);
438 rc = read_file(path, buf, sizeof(buf));
442 if (buf[strlen(buf) - 1] == '\n')
443 buf[strlen(buf) - 1] = '\0';
445 chk_major = strtok_r(buf, ":", &savept);
447 if (major == atoi(chk_major) &&minor == atoi(chk_minor))
451 if (i == glob_info.gl_pathc) {
453 fprintf(stderr,"warning: device %s does not match any "
454 "entry under /sys/block\n", real_path);
455 globfree(&glob_info);
459 /* Chop off "/dev" from path we found */
460 path[strlen(glob_info.gl_pathv[i])] = '\0';
461 globfree(&glob_info);
464 if (strncmp(real_path, "/dev/md", 7) == 0) {
465 snprintf(real_path, sizeof(real_path), "%s/%s", path,
468 rc = read_file(real_path, buf, sizeof(buf));
471 fprintf(stderr, "warning: opening %s: %s\n",
472 real_path, strerror(errno));
476 if (atoi(buf) >= md_stripe_cache_size)
479 if (strlen(buf) - 1 > 0) {
480 snprintf(buf, sizeof(buf), "%d", md_stripe_cache_size);
481 rc = write_file(real_path, buf);
483 fprintf(stderr, "warning: opening %s: %s\n",
484 real_path, strerror(errno));
486 /* Return since raid and disk tunables are different */
490 snprintf(real_path, sizeof(real_path), "%s/%s", path,
491 MAX_HW_SECTORS_KB_PATH);
492 rc = read_file(real_path, buf, sizeof(buf));
495 fprintf(stderr, "warning: opening %s: %s\n",
496 real_path, strerror(errno));
497 /* No MAX_HW_SECTORS_KB_PATH isn't necessary an
498 * error for some device. */
502 if (strlen(buf) - 1 > 0) {
503 snprintf(real_path, sizeof(real_path), "%s/%s", path,
504 MAX_SECTORS_KB_PATH);
505 rc = write_file(real_path, buf);
508 fprintf(stderr, "warning: writing to %s: %s\n",
509 real_path, strerror(errno));
510 /* No MAX_SECTORS_KB_PATH isn't necessary an
511 * error for some device. */
518 glob_info.gl_pathc = 0;
519 glob_info.gl_offs = 0;
520 /* if device is multipath device, tune its slave devices */
521 snprintf(real_path, sizeof(real_path), "%s/slaves/*", path);
522 rc = glob(real_path, GLOB_NOSORT, NULL, &glob_info);
524 for (i = 0; rc == 0 && i < glob_info.gl_pathc; i++){
525 slave = basename(glob_info.gl_pathv[i]);
526 snprintf(real_path, sizeof(real_path), "/dev/%s", slave);
527 rc = set_blockdev_tunables(real_path, 0);
530 if (rc == GLOB_NOMATCH) {
531 /* no slave device is not an error */
533 } else if (rc && verbose) {
535 fprintf(stderr, "warning: %s, failed to read"
536 " entries under %s/slaves\n",
537 strerror(errno), path);
539 fprintf(stderr, "unable to set tunables for"
540 " slave device %s (slave would be"
541 " unable to handle IO request from"
546 globfree(&glob_info);
552 int main(int argc, char *const argv[])
554 char default_options[] = "";
555 char *usource, *source, *ptr;
556 char target[PATH_MAX] = {'\0'};
557 char real_path[PATH_MAX] = {'\0'};
558 char path[256], name[256];
561 char *options, *optcopy, *orig_options = default_options;
562 int i, nargs = 3, opt, rc, flags, optlen;
563 static struct option long_opt[] = {
567 {"nomtab", 0, 0, 'n'},
568 {"options", 1, 0, 'o'},
569 {"verbose", 0, 0, 'v'},
573 progname = strrchr(argv[0], '/');
574 progname = progname ? progname + 1 : argv[0];
576 while ((opt = getopt_long(argc, argv, "fhno:v",
577 long_opt, NULL)) != EOF){
581 printf("force: %d\n", force);
586 printf("fake: %d\n", fake);
594 printf("nomtab: %d\n", nomtab);
598 orig_options = optarg;
606 fprintf(stderr, "%s: unknown option '%c'\n",
613 if (optind + 2 > argc) {
614 fprintf(stderr, "%s: too few arguments\n", progname);
618 usource = argv[optind];
624 * Try to get the real path to the device, in case it is a
625 * symbolic link for instance
627 if (realpath(usource, real_path) != NULL) {
630 ptr = strrchr(real_path, '/');
631 if (ptr && strncmp(ptr, "/dm-", 4) == 0 && isdigit(*(ptr + 4))) {
632 snprintf(path, sizeof(path), "/sys/block/%s/dm/name", ptr+1);
633 if ((f = fopen(path, "r"))) {
634 /* read "<name>\n" from sysfs */
635 if (fgets(name, sizeof(name), f) && (sz = strlen(name)) > 1) {
637 snprintf(real_path, sizeof(real_path), "/dev/mapper/%s", name);
644 source = convert_hostnames(usource);
649 if (realpath(argv[optind + 1], target) == NULL) {
651 fprintf(stderr, "warning: %s: cannot resolve: %s\n",
652 argv[optind + 1], strerror(errno));
657 for (i = 0; i < argc; i++)
658 printf("arg[%d] = %s\n", i, argv[i]);
659 printf("source = %s (%s), target = %s\n", usource, source,
661 printf("options = %s\n", orig_options);
664 options = malloc(strlen(orig_options) + 1);
665 if (options == NULL) {
666 fprintf(stderr, "can't allocate memory for options\n");
669 strcpy(options, orig_options);
670 rc = parse_options(options, &flags);
672 fprintf(stderr, "%s: can't parse options: %s\n",
678 rc = check_mtab_entry(usource, source, target, "lustre");
679 if (rc && !(flags & MS_REMOUNT)) {
680 fprintf(stderr, "%s: according to %s %s is "
681 "already mounted on %s\n",
682 progname, MOUNTED, usource, target);
685 if (!rc && (flags & MS_REMOUNT)) {
686 fprintf(stderr, "%s: according to %s %s is "
687 "not already mounted on %s\n",
688 progname, MOUNTED, usource, target);
692 if (flags & MS_REMOUNT)
695 rc = access(target, F_OK);
698 fprintf(stderr, "%s: %s inaccessible: %s\n", progname, target,
703 /* In Linux 2.4, the target device doesn't get passed to any of our
704 functions. So we'll stick it on the end of the options. */
705 optlen = strlen(options) + strlen(",device=") + strlen(source) + 1;
706 optcopy = malloc(optlen);
707 if (optcopy == NULL) {
708 fprintf(stderr, "can't allocate memory to optcopy\n");
711 strcpy(optcopy, options);
713 strcat(optcopy, ",");
714 strcat(optcopy, "device=");
715 strcat(optcopy, source);
718 printf("mounting device %s at %s, flags=%#x options=%s\n",
719 source, target, flags, optcopy);
721 if (!strstr(usource, ":/") && set_blockdev_tunables(source, 1)) {
723 fprintf(stderr, "%s: unable to set tunables for %s"
724 " (may cause reduced IO performance)\n",
729 /* flags and target get to lustre_get_sb, but not
730 lustre_fill_super. Lustre ignores the flags, but mount
732 for (i = 0, rc = -EAGAIN; i <= retry && rc != 0; i++) {
733 rc = mount(source, target, "lustre", flags,
737 fprintf(stderr, "%s: mount %s at %s "
738 "failed: %s retries left: "
739 "%d\n", basename(progname),
741 strerror(errno), retry-i);
745 sleep(1 << max((i/2), 5));
759 cli = strrchr(usource, ':');
760 if (cli && (strlen(cli) > 2))
765 fprintf(stderr, "%s: mount %s at %s failed: %s\n", progname,
766 usource, target, strerror(errno));
768 fprintf(stderr, "Are the lustre modules loaded?\n"
769 "Check /etc/modprobe.conf and "
770 "/proc/filesystems\n");
771 if (errno == ENOTBLK)
772 fprintf(stderr, "Do you need -o loop?\n");
773 if (errno == ENOMEDIUM)
775 "This filesystem needs at least 1 OST\n");
776 if (errno == ENOENT) {
777 fprintf(stderr, "Is the MGS specification correct?\n");
778 fprintf(stderr, "Is the filesystem name correct?\n");
779 fprintf(stderr, "If upgrading, is the copied client log"
780 " valid? (see upgrade docs)\n");
782 if (errno == EALREADY)
783 fprintf(stderr, "The target service is already running."
786 fprintf(stderr, "The target service failed to start "
787 "(bad config log?) (%s). "
788 "See /var/log/messages.\n", usource);
790 fprintf(stderr, "Is the MGS running?\n");
791 if (errno == EADDRINUSE)
792 fprintf(stderr, "The target service's index is already "
793 "in use. (%s)\n", usource);
794 if (errno == EINVAL) {
795 fprintf(stderr, "This may have multiple causes.\n");
797 fprintf(stderr, "Is '%s' the correct filesystem"
799 fprintf(stderr, "Are the mount options correct?\n");
800 fprintf(stderr, "Check the syslog for more info.\n");
803 /* May as well try to clean up loop devs */
804 if (strncmp(usource, "/dev/loop", 9) == 0) {
807 sprintf(cmd, "/sbin/losetup -d %s", usource);
808 if ((ret = system(cmd)) < 0)
811 rc = WEXITSTATUS(ret);
814 } else if (!nomtab) {
815 rc = update_mtab_entry(usource, target, "lustre", orig_options,