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 Whamcloud, Inc.
36 * This file is part of Lustre, http://www.lustre.org/
37 * Lustre is a trademark of Sun Microsystems, Inc.
39 * lustre/utils/mount_lustre.c
41 * Author: Robert Read <rread@clusterfs.com>
42 * Author: Nathan Rutman <nathan@clusterfs.com>
54 #include <sys/mount.h>
59 #include <lustre_ver.h>
63 #include "mount_utils.h"
65 #define MAX_HW_SECTORS_KB_PATH "queue/max_hw_sectors_kb"
66 #define MAX_SECTORS_KB_PATH "queue/max_sectors_kb"
67 #define STRIPE_CACHE_SIZE "md/stripe_cache_size"
68 #define MAX_RETRIES 99
75 int md_stripe_cache_size = 16384;
76 char *progname = NULL;
80 fprintf(out, "%s v"LUSTRE_VERSION_STRING"\n", progname);
81 fprintf(out, "\nThis mount helper should only be invoked via the "
82 "mount (8) command,\ne.g. mount -t lustre dev dir\n\n");
83 fprintf(out, "usage: %s [-fhnv] [-o <mntopt>] <device> <mountpt>\n",
86 "\t<device>: the disk device, or for a client:\n"
87 "\t\t<mgmtnid>[:<altmgtnid>...]:/<filesystem>-client\n"
88 "\t<filesystem>: name of the Lustre filesystem (e.g. lustre1)\n"
89 "\t<mountpt>: filesystem mountpoint (e.g. /mnt/lustre)\n"
90 "\t-f|--fake: fake mount (updates /etc/mtab)\n"
91 "\t-o force|--force: force mount even if already in /etc/mtab\n"
92 "\t-h|--help: print this usage message\n"
93 "\t-n|--nomtab: do not update /etc/mtab after mount\n"
94 "\t-v|--verbose: print verbose config settings\n"
95 "\t<mntopt>: one or more comma separated of:\n"
96 "\t\t(no)flock,(no)user_xattr,(no)acl\n"
97 "\t\tabort_recov: abort server recovery handling\n"
98 "\t\tnosvc: only start MGC/MGS obds\n"
99 "\t\tnomgs: only start target obds, using existing MGS\n"
100 "\t\texclude=<ostname>[:<ostname>] : colon-separated list of "
101 "inactive OSTs (e.g. lustre-OST0001)\n"
102 "\t\tretry=<num>: number of times mount is retried by client\n"
103 "\t\tmd_stripe_cache_size=<num>: set the raid stripe cache "
104 "size for the underlying raid if present\n"
106 exit((out != stdout) ? EINVAL : 0);
109 static int check_mtab_entry(char *spec1, char *spec2, char *mtpt, char *type)
114 fp = setmntent(MOUNTED, "r");
118 while ((mnt = getmntent(fp)) != NULL) {
119 if ((strcmp(mnt->mnt_fsname, spec1) == 0 ||
120 strcmp(mnt->mnt_fsname, spec2) == 0) &&
121 strcmp(mnt->mnt_dir, mtpt) == 0 &&
122 strcmp(mnt->mnt_type, type) == 0) {
133 update_mtab_entry(char *spec, char *mtpt, char *type, char *opts,
134 int flags, int freq, int pass)
140 mnt.mnt_fsname = spec;
143 mnt.mnt_opts = opts ? opts : "";
145 mnt.mnt_passno = pass;
147 fp = setmntent(MOUNTED, "a+");
149 fprintf(stderr, "%s: setmntent(%s): %s:",
150 progname, MOUNTED, strerror (errno));
153 if ((addmntent(fp, &mnt)) == 1) {
154 fprintf(stderr, "%s: addmntent: %s:",
155 progname, strerror (errno));
164 /* Get rid of symbolic hostnames for tcp, since kernel can't do lookups */
165 #define MAXNIDSTR 1024
166 static char *convert_hostnames(char *s1)
168 char *converted, *s2 = 0, *c;
170 int left = MAXNIDSTR;
173 converted = malloc(left);
174 if (converted == NULL) {
175 fprintf(stderr, "out of memory: needed %d bytes\n",
180 while ((left > 0) && (*s1 != '/')) {
181 s2 = strpbrk(s1, ",:");
186 nid = libcfs_str2nid(s1);
187 *s2 = sep; /* back to original string */
188 if (nid == LNET_NID_ANY)
190 c += snprintf(c, left, "%s%c", libcfs_nid2str(nid), sep);
191 left = converted + MAXNIDSTR - c;
194 snprintf(c, left, "%s", s1);
197 fprintf(stderr, "%s: Can't parse NID '%s'\n", progname, s1);
202 /*****************************************************************************
204 * This part was cribbed from util-linux/mount/mount.c. There was no clear
205 * license information, but many other files in the package are identified as
206 * GNU GPL, so it's a pretty safe bet that was their intent.
208 ****************************************************************************/
210 const char *opt; /* option name */
211 int inv; /* true if flag value should be inverted */
212 int mask; /* flag mask value */
215 static const struct opt_map opt_map[] = {
216 /*"optname", inv,ms_mask */
217 /* These flags are parsed by mount, not lustre */
218 { "defaults", 0, 0 }, /* default options */
219 { "remount", 0, MS_REMOUNT}, /* remount with different options */
220 { "rw", 1, MS_RDONLY }, /* read-write */
221 { "ro", 0, MS_RDONLY }, /* read-only */
222 { "exec", 1, MS_NOEXEC }, /* permit execution of binaries */
223 { "noexec", 0, MS_NOEXEC }, /* don't execute binaries */
224 { "suid", 1, MS_NOSUID }, /* honor suid executables */
225 { "nosuid", 0, MS_NOSUID }, /* don't honor suid executables */
226 { "dev", 1, MS_NODEV }, /* interpret device files */
227 { "nodev", 0, MS_NODEV }, /* don't interpret devices */
228 { "sync", 0, MS_SYNCHRONOUS}, /* synchronous I/O */
229 { "async", 1, MS_SYNCHRONOUS}, /* asynchronous I/O */
230 { "atime", 1, MS_NOATIME }, /* set file access time on read */
231 { "noatime", 0, MS_NOATIME }, /* do not set file access time on read */
233 { "diratime", 1, MS_NODIRATIME }, /* set file access time on read */
234 { "nodiratime",0,MS_NODIRATIME }, /* do not set file access time on read */
237 { "relatime", 0, MS_RELATIME }, /* set file access time on read */
238 { "norelatime",1,MS_RELATIME }, /* do not set file access time on read */
240 #ifdef MS_STRICTATIME
241 { "strictatime",0,MS_STRICTATIME }, /* update access time strictly */
243 { "auto", 0, 0 }, /* Can be mounted using -a */
244 { "noauto", 0, 0 }, /* Can only be mounted explicitly */
245 { "nousers", 1, 0 }, /* Forbid ordinary user to mount */
246 { "nouser", 1, 0 }, /* Forbid ordinary user to mount */
247 { "noowner", 1, 0 }, /* Device owner has no special privs */
248 { "_netdev", 0, 0 }, /* Device accessible only via network */
252 /****************************************************************************/
254 /* 1 = don't pass on to lustre
255 0 = pass on to lustre */
256 static int parse_one_option(const char *check, int *flagp)
258 const struct opt_map *opt;
260 for (opt = &opt_map[0]; opt->opt != NULL; opt++) {
261 if (strncmp(check, opt->opt, strlen(opt->opt)) == 0) {
264 *flagp &= ~(opt->mask);
271 /* Assume any unknown options are valid and pass them on. The mount
272 will fail if lmd_parse, ll_options or ldiskfs doesn't recognize it.*/
276 static void append_option(char *options, const char *one)
279 strcat(options, ",");
280 strcat(options, one);
283 /* Replace options with subset of Lustre-specific options, and
284 fill in mount flags */
285 int parse_options(char *orig_options, int *flagp)
287 char *options, *opt, *nextopt, *arg, *val;
289 options = calloc(strlen(orig_options) + 1, 1);
291 nextopt = orig_options;
292 while ((opt = strsep(&nextopt, ","))) {
297 /* Handle retries in a slightly different
300 val = strchr(opt, '=');
301 /* please note that some ldiskfs mount options are also in the form
302 * of param=value. We should pay attention not to remove those
303 * mount options, see bug 22097. */
304 if (val && strncmp(arg, "md_stripe_cache_size", 20) == 0) {
305 md_stripe_cache_size = atoi(val + 1);
306 } else if (val && strncmp(arg, "retry", 5) == 0) {
307 retry = atoi(val + 1);
308 if (retry > MAX_RETRIES)
312 } else if (val && strncmp(arg, "mgssec", 6) == 0) {
313 append_option(options, opt);
314 } else if (strcmp(opt, "force") == 0) {
315 //XXX special check for 'force' option
317 printf("force: %d\n", force);
318 } else if (parse_one_option(opt, flagp) == 0) {
319 /* pass this on as an option */
320 append_option(options, opt);
323 #ifdef MS_STRICTATIME
324 /* set strictatime to default if NOATIME or RELATIME
325 not given explicit */
326 if (!(*flagp & (MS_NOATIME | MS_RELATIME)))
327 *flagp |= MS_STRICTATIME;
329 strcpy(orig_options, options);
335 int read_file(char *path, char *buf, int size)
339 fd = fopen(path, "r");
343 /* should not ignore fgets(3)'s return value */
344 if (!fgets(buf, size, fd)) {
345 fprintf(stderr, "reading from %s: %s", path, strerror(errno));
353 int write_file(char *path, char *buf)
357 fd = fopen(path, "w");
366 /* This is to tune the kernel for good SCSI performance.
367 * For that we set the value of /sys/block/{dev}/queue/max_sectors_kb
368 * to the value of /sys/block/{dev}/queue/max_hw_sectors_kb */
369 int set_blockdev_tunables(char *source, int fan_out)
371 glob_t glob_info = { 0 };
372 struct stat stat_buf;
373 char *chk_major, *chk_minor;
376 char buf[PATH_MAX] = {'\0'}, path[PATH_MAX] = {'\0'};
377 char real_path[PATH_MAX] = {'\0'};
384 ret_path = realpath(source, real_path);
385 if (ret_path == NULL) {
387 fprintf(stderr, "warning: %s: cannot resolve: %s\n",
388 source, strerror(errno));
392 if (strncmp(real_path, "/dev/loop", 9) == 0)
395 if ((real_path[0] != '/') && (strpbrk(real_path, ",:") != NULL))
398 snprintf(path, sizeof(path), "/sys/block%s", real_path + 4);
399 if (access(path, X_OK) == 0)
402 /* The name of the device say 'X' specified in /dev/X may not
403 * match any entry under /sys/block/. In that case we need to
404 * match the major/minor number to find the entry under
405 * sys/block corresponding to /dev/X */
407 /* Don't chop tail digit on /dev/mapper/xxx, LU-478 */
408 if (strncmp(real_path, "/dev/mapper", 11) != 0) {
409 dev = real_path + strlen(real_path);
410 while (--dev > real_path && isdigit(*dev))
413 if (strncmp(real_path, "/dev/md_", 8) == 0)
417 rc = stat(real_path, &stat_buf);
420 fprintf(stderr, "warning: %s, device %s stat failed\n",
421 strerror(errno), real_path);
425 major = major(stat_buf.st_rdev);
426 minor = minor(stat_buf.st_rdev);
427 rc = glob("/sys/block/*", GLOB_NOSORT, NULL, &glob_info);
430 fprintf(stderr, "warning: failed to read entries under "
432 globfree(&glob_info);
436 for (i = 0; i < glob_info.gl_pathc; i++){
437 snprintf(path, sizeof(path), "%s/dev", glob_info.gl_pathv[i]);
439 rc = read_file(path, buf, sizeof(buf));
443 if (buf[strlen(buf) - 1] == '\n')
444 buf[strlen(buf) - 1] = '\0';
446 chk_major = strtok_r(buf, ":", &savept);
448 if (major == atoi(chk_major) &&minor == atoi(chk_minor))
452 if (i == glob_info.gl_pathc) {
454 fprintf(stderr,"warning: device %s does not match any "
455 "entry under /sys/block\n", real_path);
456 globfree(&glob_info);
460 /* Chop off "/dev" from path we found */
461 path[strlen(glob_info.gl_pathv[i])] = '\0';
462 globfree(&glob_info);
465 if (strncmp(real_path, "/dev/md", 7) == 0) {
466 snprintf(real_path, sizeof(real_path), "%s/%s", path,
469 rc = read_file(real_path, buf, sizeof(buf));
472 fprintf(stderr, "warning: opening %s: %s\n",
473 real_path, strerror(errno));
477 if (atoi(buf) >= md_stripe_cache_size)
480 if (strlen(buf) - 1 > 0) {
481 snprintf(buf, sizeof(buf), "%d", md_stripe_cache_size);
482 rc = write_file(real_path, buf);
484 fprintf(stderr, "warning: opening %s: %s\n",
485 real_path, strerror(errno));
487 /* Return since raid and disk tunables are different */
491 snprintf(real_path, sizeof(real_path), "%s/%s", path,
492 MAX_HW_SECTORS_KB_PATH);
493 rc = read_file(real_path, buf, sizeof(buf));
496 fprintf(stderr, "warning: opening %s: %s\n",
497 real_path, strerror(errno));
498 /* No MAX_HW_SECTORS_KB_PATH isn't necessary an
499 * error for some device. */
503 if (strlen(buf) - 1 > 0) {
504 snprintf(real_path, sizeof(real_path), "%s/%s", path,
505 MAX_SECTORS_KB_PATH);
506 rc = write_file(real_path, buf);
509 fprintf(stderr, "warning: writing to %s: %s\n",
510 real_path, strerror(errno));
511 /* No MAX_SECTORS_KB_PATH isn't necessary an
512 * error for some device. */
519 glob_info.gl_pathc = 0;
520 glob_info.gl_offs = 0;
521 /* if device is multipath device, tune its slave devices */
522 snprintf(real_path, sizeof(real_path), "%s/slaves/*", path);
523 rc = glob(real_path, GLOB_NOSORT, NULL, &glob_info);
525 for (i = 0; rc == 0 && i < glob_info.gl_pathc; i++){
526 slave = basename(glob_info.gl_pathv[i]);
527 snprintf(real_path, sizeof(real_path), "/dev/%s", slave);
528 rc = set_blockdev_tunables(real_path, 0);
531 if (rc == GLOB_NOMATCH) {
532 /* no slave device is not an error */
534 } else if (rc && verbose) {
536 fprintf(stderr, "warning: %s, failed to read"
537 " entries under %s/slaves\n",
538 strerror(errno), path);
540 fprintf(stderr, "unable to set tunables for"
541 " slave device %s (slave would be"
542 " unable to handle IO request from"
547 globfree(&glob_info);
553 int main(int argc, char *const argv[])
555 char default_options[] = "";
556 char *usource, *source, *ptr;
557 char target[PATH_MAX] = {'\0'};
558 char real_path[PATH_MAX] = {'\0'};
559 char path[256], name[256];
562 char *options, *optcopy, *orig_options = default_options;
563 int i, nargs = 3, opt, rc, flags, optlen;
564 static struct option long_opt[] = {
568 {"nomtab", 0, 0, 'n'},
569 {"options", 1, 0, 'o'},
570 {"verbose", 0, 0, 'v'},
574 progname = strrchr(argv[0], '/');
575 progname = progname ? progname + 1 : argv[0];
577 while ((opt = getopt_long(argc, argv, "fhno:v",
578 long_opt, NULL)) != EOF){
582 printf("force: %d\n", force);
587 printf("fake: %d\n", fake);
595 printf("nomtab: %d\n", nomtab);
599 orig_options = optarg;
607 fprintf(stderr, "%s: unknown option '%c'\n",
614 if (optind + 2 > argc) {
615 fprintf(stderr, "%s: too few arguments\n", progname);
619 usource = argv[optind];
625 * Try to get the real path to the device, in case it is a
626 * symbolic link for instance
628 if (realpath(usource, real_path) != NULL) {
631 ptr = strrchr(real_path, '/');
632 if (ptr && strncmp(ptr, "/dm-", 4) == 0 && isdigit(*(ptr + 4))) {
633 snprintf(path, sizeof(path), "/sys/block/%s/dm/name", ptr+1);
634 if ((f = fopen(path, "r"))) {
635 /* read "<name>\n" from sysfs */
636 if (fgets(name, sizeof(name), f) && (sz = strlen(name)) > 1) {
638 snprintf(real_path, sizeof(real_path), "/dev/mapper/%s", name);
645 source = convert_hostnames(usource);
650 if (realpath(argv[optind + 1], target) == NULL) {
652 fprintf(stderr, "warning: %s: cannot resolve: %s\n",
653 argv[optind + 1], strerror(errno));
658 for (i = 0; i < argc; i++)
659 printf("arg[%d] = %s\n", i, argv[i]);
660 printf("source = %s (%s), target = %s\n", usource, source,
662 printf("options = %s\n", orig_options);
665 options = malloc(strlen(orig_options) + 1);
666 if (options == NULL) {
667 fprintf(stderr, "can't allocate memory for options\n");
670 strcpy(options, orig_options);
671 rc = parse_options(options, &flags);
673 fprintf(stderr, "%s: can't parse options: %s\n",
679 rc = check_mtab_entry(usource, source, target, "lustre");
680 if (rc && !(flags & MS_REMOUNT)) {
681 fprintf(stderr, "%s: according to %s %s is "
682 "already mounted on %s\n",
683 progname, MOUNTED, usource, target);
686 if (!rc && (flags & MS_REMOUNT)) {
687 fprintf(stderr, "%s: according to %s %s is "
688 "not already mounted on %s\n",
689 progname, MOUNTED, usource, target);
693 if (flags & MS_REMOUNT)
696 rc = access(target, F_OK);
699 fprintf(stderr, "%s: %s inaccessible: %s\n", progname, target,
704 /* In Linux 2.4, the target device doesn't get passed to any of our
705 functions. So we'll stick it on the end of the options. */
706 optlen = strlen(options) + strlen(",device=") + strlen(source) + 1;
707 optcopy = malloc(optlen);
708 if (optcopy == NULL) {
709 fprintf(stderr, "can't allocate memory to optcopy\n");
712 strcpy(optcopy, options);
714 strcat(optcopy, ",");
715 strcat(optcopy, "device=");
716 strcat(optcopy, source);
719 printf("mounting device %s at %s, flags=%#x options=%s\n",
720 source, target, flags, optcopy);
722 if (!strstr(usource, ":/") && set_blockdev_tunables(source, 1)) {
724 fprintf(stderr, "%s: unable to set tunables for %s"
725 " (may cause reduced IO performance)\n",
729 register_service_tags(usource, source, target);
732 /* flags and target get to lustre_get_sb, but not
733 lustre_fill_super. Lustre ignores the flags, but mount
735 for (i = 0, rc = -EAGAIN; i <= retry && rc != 0; i++) {
736 rc = mount(source, target, "lustre", flags,
740 fprintf(stderr, "%s: mount %s at %s "
741 "failed: %s retries left: "
742 "%d\n", basename(progname),
744 strerror(errno), retry-i);
748 sleep(1 << max((i/2), 5));
762 cli = strrchr(usource, ':');
763 if (cli && (strlen(cli) > 2))
768 fprintf(stderr, "%s: mount %s at %s failed: %s\n", progname,
769 usource, target, strerror(errno));
771 fprintf(stderr, "Are the lustre modules loaded?\n"
772 "Check /etc/modprobe.conf and "
773 "/proc/filesystems\n");
774 if (errno == ENOTBLK)
775 fprintf(stderr, "Do you need -o loop?\n");
776 if (errno == ENOMEDIUM)
778 "This filesystem needs at least 1 OST\n");
779 if (errno == ENOENT) {
780 fprintf(stderr, "Is the MGS specification correct?\n");
781 fprintf(stderr, "Is the filesystem name correct?\n");
782 fprintf(stderr, "If upgrading, is the copied client log"
783 " valid? (see upgrade docs)\n");
785 if (errno == EALREADY)
786 fprintf(stderr, "The target service is already running."
789 fprintf(stderr, "The target service failed to start "
790 "(bad config log?) (%s). "
791 "See /var/log/messages.\n", usource);
793 fprintf(stderr, "Is the MGS running?\n");
794 if (errno == EADDRINUSE)
795 fprintf(stderr, "The target service's index is already "
796 "in use. (%s)\n", usource);
797 if (errno == EINVAL) {
798 fprintf(stderr, "This may have multiple causes.\n");
800 fprintf(stderr, "Is '%s' the correct filesystem"
802 fprintf(stderr, "Are the mount options correct?\n");
803 fprintf(stderr, "Check the syslog for more info.\n");
806 /* May as well try to clean up loop devs */
807 if (strncmp(usource, "/dev/loop", 9) == 0) {
810 sprintf(cmd, "/sbin/losetup -d %s", usource);
811 if ((ret = system(cmd)) < 0)
814 rc = WEXITSTATUS(ret);
817 } else if (!nomtab) {
818 rc = update_mtab_entry(usource, target, "lustre", orig_options,