4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Whamcloud, Inc.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/utils/mount_lustre.c
38 * Author: Robert Read <rread@clusterfs.com>
39 * Author: Nathan Rutman <nathan@clusterfs.com>
52 #include <lustre_ver.h>
56 #include "mount_utils.h"
58 #define MAX_HW_SECTORS_KB_PATH "queue/max_hw_sectors_kb"
59 #define MAX_SECTORS_KB_PATH "queue/max_sectors_kb"
60 #define STRIPE_CACHE_SIZE "md/stripe_cache_size"
61 #define MAX_RETRIES 99
68 int md_stripe_cache_size = 16384;
69 char *progname = NULL;
73 fprintf(out, "%s v"LUSTRE_VERSION_STRING"\n", progname);
74 fprintf(out, "\nThis mount helper should only be invoked via the "
75 "mount (8) command,\ne.g. mount -t lustre dev dir\n\n");
76 fprintf(out, "usage: %s [-fhnv] [-o <mntopt>] <device> <mountpt>\n",
79 "\t<device>: the disk device, or for a client:\n"
80 "\t\t<mgmtnid>[:<altmgtnid>...]:/<filesystem>-client\n"
81 "\t<filesystem>: name of the Lustre filesystem (e.g. lustre1)\n"
82 "\t<mountpt>: filesystem mountpoint (e.g. /mnt/lustre)\n"
83 "\t-f|--fake: fake mount (updates /etc/mtab)\n"
84 "\t-o force|--force: force mount even if already in /etc/mtab\n"
85 "\t-h|--help: print this usage message\n"
86 "\t-n|--nomtab: do not update /etc/mtab after mount\n"
87 "\t-v|--verbose: print verbose config settings\n"
88 "\t<mntopt>: one or more comma separated of:\n"
89 "\t\t(no)flock,(no)user_xattr,(no)acl\n"
90 "\t\tabort_recov: abort server recovery handling\n"
91 "\t\tnosvc: only start MGC/MGS obds\n"
92 "\t\tnomgs: only start target obds, using existing MGS\n"
93 "\t\texclude=<ostname>[:<ostname>] : colon-separated list of "
94 "inactive OSTs (e.g. lustre-OST0001)\n"
95 "\t\tretry=<num>: number of times mount is retried by client\n"
96 "\t\tmd_stripe_cache_size=<num>: set the raid stripe cache "
97 "size for the underlying raid if present\n"
99 exit((out != stdout) ? EINVAL : 0);
102 /* Get rid of symbolic hostnames for tcp, since kernel can't do lookups */
103 #define MAXNIDSTR 1024
104 static char *convert_hostnames(char *s1)
106 char *converted, *s2 = 0, *c;
108 int left = MAXNIDSTR;
111 converted = malloc(left);
112 if (converted == NULL) {
113 fprintf(stderr, "out of memory: needed %d bytes\n",
118 while ((left > 0) && (*s1 != '/')) {
119 s2 = strpbrk(s1, ",:");
124 nid = libcfs_str2nid(s1);
125 *s2 = sep; /* back to original string */
126 if (nid == LNET_NID_ANY)
128 c += snprintf(c, left, "%s%c", libcfs_nid2str(nid), sep);
129 left = converted + MAXNIDSTR - c;
132 snprintf(c, left, "%s", s1);
135 fprintf(stderr, "%s: Can't parse NID '%s'\n", progname, s1);
140 /*****************************************************************************
142 * This part was cribbed from util-linux/mount/mount.c. There was no clear
143 * license information, but many other files in the package are identified as
144 * GNU GPL, so it's a pretty safe bet that was their intent.
146 ****************************************************************************/
148 const char *opt; /* option name */
149 int inv; /* true if flag value should be inverted */
150 int mask; /* flag mask value */
153 static const struct opt_map opt_map[] = {
154 /*"optname", inv,ms_mask */
155 /* These flags are parsed by mount, not lustre */
156 { "defaults", 0, 0 }, /* default options */
157 { "remount", 0, MS_REMOUNT}, /* remount with different options */
158 { "rw", 1, MS_RDONLY }, /* read-write */
159 { "ro", 0, MS_RDONLY }, /* read-only */
160 { "exec", 1, MS_NOEXEC }, /* permit execution of binaries */
161 { "noexec", 0, MS_NOEXEC }, /* don't execute binaries */
162 { "suid", 1, MS_NOSUID }, /* honor suid executables */
163 { "nosuid", 0, MS_NOSUID }, /* don't honor suid executables */
164 { "dev", 1, MS_NODEV }, /* interpret device files */
165 { "nodev", 0, MS_NODEV }, /* don't interpret devices */
166 { "sync", 0, MS_SYNCHRONOUS}, /* synchronous I/O */
167 { "async", 1, MS_SYNCHRONOUS}, /* asynchronous I/O */
168 { "atime", 1, MS_NOATIME }, /* set file access time on read */
169 { "noatime", 0, MS_NOATIME }, /* do not set file access time on read */
171 { "diratime", 1, MS_NODIRATIME }, /* set file access time on read */
172 { "nodiratime",0,MS_NODIRATIME }, /* do not set file access time on read */
175 { "relatime", 0, MS_RELATIME }, /* set file access time on read */
176 { "norelatime",1,MS_RELATIME }, /* do not set file access time on read */
178 #ifdef MS_STRICTATIME
179 { "strictatime",0,MS_STRICTATIME }, /* update access time strictly */
181 { "auto", 0, 0 }, /* Can be mounted using -a */
182 { "noauto", 0, 0 }, /* Can only be mounted explicitly */
183 { "nousers", 1, 0 }, /* Forbid ordinary user to mount */
184 { "nouser", 1, 0 }, /* Forbid ordinary user to mount */
185 { "noowner", 1, 0 }, /* Device owner has no special privs */
186 { "_netdev", 0, 0 }, /* Device accessible only via network */
190 /****************************************************************************/
192 /* 1 = don't pass on to lustre
193 0 = pass on to lustre */
194 static int parse_one_option(const char *check, int *flagp)
196 const struct opt_map *opt;
198 for (opt = &opt_map[0]; opt->opt != NULL; opt++) {
199 if (strncmp(check, opt->opt, strlen(opt->opt)) == 0) {
202 *flagp &= ~(opt->mask);
209 /* Assume any unknown options are valid and pass them on. The mount
210 will fail if lmd_parse, ll_options or ldiskfs doesn't recognize it.*/
214 static void append_option(char *options, const char *one)
217 strcat(options, ",");
218 strcat(options, one);
221 /* Replace options with subset of Lustre-specific options, and
222 fill in mount flags */
223 int parse_options(char *orig_options, int *flagp)
225 char *options, *opt, *nextopt, *arg, *val;
227 options = calloc(strlen(orig_options) + 1, 1);
229 nextopt = orig_options;
230 while ((opt = strsep(&nextopt, ","))) {
235 /* Handle retries in a slightly different
238 val = strchr(opt, '=');
239 /* please note that some ldiskfs mount options are also in the form
240 * of param=value. We should pay attention not to remove those
241 * mount options, see bug 22097. */
242 if (val && strncmp(arg, "md_stripe_cache_size", 20) == 0) {
243 md_stripe_cache_size = atoi(val + 1);
244 } else if (val && strncmp(arg, "retry", 5) == 0) {
245 retry = atoi(val + 1);
246 if (retry > MAX_RETRIES)
250 } else if (val && strncmp(arg, "mgssec", 6) == 0) {
251 append_option(options, opt);
252 } else if (strcmp(opt, "force") == 0) {
253 //XXX special check for 'force' option
255 printf("force: %d\n", force);
256 } else if (parse_one_option(opt, flagp) == 0) {
257 /* pass this on as an option */
258 append_option(options, opt);
261 #ifdef MS_STRICTATIME
262 /* set strictatime to default if NOATIME or RELATIME
263 not given explicit */
264 if (!(*flagp & (MS_NOATIME | MS_RELATIME)))
265 *flagp |= MS_STRICTATIME;
267 strcpy(orig_options, options);
273 int read_file(char *path, char *buf, int size)
277 fd = fopen(path, "r");
281 /* should not ignore fgets(3)'s return value */
282 if (!fgets(buf, size, fd)) {
283 fprintf(stderr, "reading from %s: %s", path, strerror(errno));
291 int write_file(char *path, char *buf)
295 fd = fopen(path, "w");
304 /* This is to tune the kernel for good SCSI performance.
305 * For that we set the value of /sys/block/{dev}/queue/max_sectors_kb
306 * to the value of /sys/block/{dev}/queue/max_hw_sectors_kb */
307 int set_blockdev_tunables(char *source, int fan_out)
309 glob_t glob_info = { 0 };
310 struct stat stat_buf;
311 char *chk_major, *chk_minor;
312 char *savept = NULL, *dev;
314 char buf[PATH_MAX] = {'\0'}, path[PATH_MAX] = {'\0'};
315 char real_path[PATH_MAX] = {'\0'};
322 ret_path = realpath(source, real_path);
323 if (ret_path == NULL) {
325 fprintf(stderr, "warning: %s: cannot resolve: %s\n",
326 source, strerror(errno));
330 if (strncmp(real_path, "/dev/loop", 9) == 0)
333 if ((real_path[0] != '/') && (strpbrk(real_path, ",:") != NULL))
336 snprintf(path, sizeof(path), "/sys/block%s", real_path + 4);
337 if (access(path, X_OK) == 0)
340 /* The name of the device say 'X' specified in /dev/X may not
341 * match any entry under /sys/block/. In that case we need to
342 * match the major/minor number to find the entry under
343 * sys/block corresponding to /dev/X */
345 /* Don't chop tail digit on /dev/mapper/xxx, LU-478 */
346 if (strncmp(real_path, "/dev/mapper", 11) != 0) {
347 dev = real_path + strlen(real_path);
348 while (--dev > real_path && isdigit(*dev))
351 if (strncmp(real_path, "/dev/md_", 8) == 0)
355 rc = stat(real_path, &stat_buf);
358 fprintf(stderr, "warning: %s, device %s stat failed\n",
359 strerror(errno), real_path);
363 major = major(stat_buf.st_rdev);
364 minor = minor(stat_buf.st_rdev);
365 rc = glob("/sys/block/*", GLOB_NOSORT, NULL, &glob_info);
368 fprintf(stderr, "warning: failed to read entries under "
370 globfree(&glob_info);
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);
394 globfree(&glob_info);
398 /* Chop off "/dev" from path we found */
399 path[strlen(glob_info.gl_pathv[i])] = '\0';
400 globfree(&glob_info);
403 if (strncmp(real_path, "/dev/md", 7) == 0) {
404 snprintf(real_path, sizeof(real_path), "%s/%s", path,
407 rc = read_file(real_path, buf, sizeof(buf));
410 fprintf(stderr, "warning: opening %s: %s\n",
411 real_path, strerror(errno));
415 if (atoi(buf) >= md_stripe_cache_size)
418 if (strlen(buf) - 1 > 0) {
419 snprintf(buf, sizeof(buf), "%d", md_stripe_cache_size);
420 rc = write_file(real_path, buf);
422 fprintf(stderr, "warning: opening %s: %s\n",
423 real_path, strerror(errno));
425 /* Return since raid and disk tunables are different */
429 snprintf(real_path, sizeof(real_path), "%s/%s", path,
430 MAX_HW_SECTORS_KB_PATH);
431 rc = read_file(real_path, buf, sizeof(buf));
434 fprintf(stderr, "warning: opening %s: %s\n",
435 real_path, strerror(errno));
436 /* No MAX_HW_SECTORS_KB_PATH isn't necessary an
437 * error for some device. */
441 if (strlen(buf) - 1 > 0) {
442 snprintf(real_path, sizeof(real_path), "%s/%s", path,
443 MAX_SECTORS_KB_PATH);
444 rc = write_file(real_path, buf);
447 fprintf(stderr, "warning: writing to %s: %s\n",
448 real_path, strerror(errno));
449 /* No MAX_SECTORS_KB_PATH isn't necessary an
450 * error for some device. */
457 glob_info.gl_pathc = 0;
458 glob_info.gl_offs = 0;
459 /* if device is multipath device, tune its slave devices */
460 snprintf(real_path, sizeof(real_path), "%s/slaves/*", path);
461 rc = glob(real_path, GLOB_NOSORT, NULL, &glob_info);
463 for (i = 0; rc == 0 && i < glob_info.gl_pathc; i++){
464 slave = basename(glob_info.gl_pathv[i]);
465 snprintf(real_path, sizeof(real_path), "/dev/%s", slave);
466 rc = set_blockdev_tunables(real_path, 0);
469 if (rc == GLOB_NOMATCH) {
470 /* no slave device is not an error */
472 } else if (rc && verbose) {
474 fprintf(stderr, "warning: %s, failed to read"
475 " entries under %s/slaves\n",
476 strerror(errno), path);
478 fprintf(stderr, "unable to set tunables for"
479 " slave device %s (slave would be"
480 " unable to handle IO request from"
485 globfree(&glob_info);
491 int main(int argc, char *const argv[])
493 char default_options[] = "";
494 char *usource, *source, *ptr;
495 char target[PATH_MAX] = {'\0'};
496 char real_path[PATH_MAX] = {'\0'};
497 char path[256], name[256];
500 char *options, *optcopy, *orig_options = default_options;
501 int i, nargs = 3, opt, rc, flags, optlen;
502 static struct option long_opt[] = {
506 {"nomtab", 0, 0, 'n'},
507 {"options", 1, 0, 'o'},
508 {"verbose", 0, 0, 'v'},
512 progname = strrchr(argv[0], '/');
513 progname = progname ? progname + 1 : argv[0];
515 while ((opt = getopt_long(argc, argv, "fhno:v",
516 long_opt, NULL)) != EOF){
520 printf("force: %d\n", force);
525 printf("fake: %d\n", fake);
533 printf("nomtab: %d\n", nomtab);
537 orig_options = optarg;
545 fprintf(stderr, "%s: unknown option '%c'\n",
552 if (optind + 2 > argc) {
553 fprintf(stderr, "%s: too few arguments\n", progname);
557 usource = argv[optind];
563 * Try to get the real path to the device, in case it is a
564 * symbolic link for instance
566 if (realpath(usource, real_path) != NULL) {
569 ptr = strrchr(real_path, '/');
570 if (ptr && strncmp(ptr, "/dm-", 4) == 0 && isdigit(*(ptr + 4))) {
571 snprintf(path, sizeof(path), "/sys/block/%s/dm/name", ptr+1);
572 if ((f = fopen(path, "r"))) {
573 /* read "<name>\n" from sysfs */
574 if (fgets(name, sizeof(name), f) && (sz = strlen(name)) > 1) {
576 snprintf(real_path, sizeof(real_path), "/dev/mapper/%s", name);
583 source = convert_hostnames(usource);
588 if (realpath(argv[optind + 1], target) == NULL) {
590 fprintf(stderr, "warning: %s: cannot resolve: %s\n",
591 argv[optind + 1], strerror(errno));
596 for (i = 0; i < argc; i++)
597 printf("arg[%d] = %s\n", i, argv[i]);
598 printf("source = %s (%s), target = %s\n", usource, source,
600 printf("options = %s\n", orig_options);
603 options = malloc(strlen(orig_options) + 1);
604 if (options == NULL) {
605 fprintf(stderr, "can't allocate memory for options\n");
608 strcpy(options, orig_options);
609 rc = parse_options(options, &flags);
611 fprintf(stderr, "%s: can't parse options: %s\n",
617 rc = check_mtab_entry(usource, source, target, "lustre");
618 if (rc && !(flags & MS_REMOUNT)) {
619 fprintf(stderr, "%s: according to %s %s is "
620 "already mounted on %s\n",
621 progname, MOUNTED, usource, target);
624 if (!rc && (flags & MS_REMOUNT)) {
625 fprintf(stderr, "%s: according to %s %s is "
626 "not already mounted on %s\n",
627 progname, MOUNTED, usource, target);
631 if (flags & MS_REMOUNT)
634 rc = access(target, F_OK);
637 fprintf(stderr, "%s: %s inaccessible: %s\n", progname, target,
642 /* In Linux 2.4, the target device doesn't get passed to any of our
643 functions. So we'll stick it on the end of the options. */
644 optlen = strlen(options) + strlen(",device=") + strlen(source) + 1;
645 optcopy = malloc(optlen);
646 if (optcopy == NULL) {
647 fprintf(stderr, "can't allocate memory to optcopy\n");
650 strcpy(optcopy, options);
652 strcat(optcopy, ",");
653 strcat(optcopy, "device=");
654 strcat(optcopy, source);
657 printf("mounting device %s at %s, flags=%#x options=%s\n",
658 source, target, flags, optcopy);
660 if (!strstr(usource, ":/") && set_blockdev_tunables(source, 1)) {
662 fprintf(stderr, "%s: unable to set tunables for %s"
663 " (may cause reduced IO performance)\n",
668 /* flags and target get to lustre_get_sb, but not
669 lustre_fill_super. Lustre ignores the flags, but mount
671 for (i = 0, rc = -EAGAIN; i <= retry && rc != 0; i++) {
672 rc = mount(source, target, "lustre", flags,
676 fprintf(stderr, "%s: mount %s at %s "
677 "failed: %s retries left: "
678 "%d\n", basename(progname),
680 strerror(errno), retry-i);
684 sleep(1 << max((i/2), 5));
698 cli = strrchr(usource, ':');
699 if (cli && (strlen(cli) > 2))
704 fprintf(stderr, "%s: mount %s at %s failed: %s\n", progname,
705 usource, target, strerror(errno));
707 fprintf(stderr, "Are the lustre modules loaded?\n"
708 "Check /etc/modprobe.conf and "
709 "/proc/filesystems\n");
710 if (errno == ENOTBLK)
711 fprintf(stderr, "Do you need -o loop?\n");
712 if (errno == ENOMEDIUM)
714 "This filesystem needs at least 1 OST\n");
715 if (errno == ENOENT) {
716 fprintf(stderr, "Is the MGS specification correct?\n");
717 fprintf(stderr, "Is the filesystem name correct?\n");
718 fprintf(stderr, "If upgrading, is the copied client log"
719 " valid? (see upgrade docs)\n");
721 if (errno == EALREADY)
722 fprintf(stderr, "The target service is already running."
725 fprintf(stderr, "The target service failed to start "
726 "(bad config log?) (%s). "
727 "See /var/log/messages.\n", usource);
729 fprintf(stderr, "Is the MGS running?\n");
730 if (errno == EADDRINUSE)
731 fprintf(stderr, "The target service's index is already "
732 "in use. (%s)\n", usource);
733 if (errno == EINVAL) {
734 fprintf(stderr, "This may have multiple causes.\n");
736 fprintf(stderr, "Is '%s' the correct filesystem"
738 fprintf(stderr, "Are the mount options correct?\n");
739 fprintf(stderr, "Check the syslog for more info.\n");
742 /* May as well try to clean up loop devs */
743 if (strncmp(usource, "/dev/loop", 9) == 0) {
746 sprintf(cmd, "/sbin/losetup -d %s", usource);
747 if ((ret = system(cmd)) < 0)
750 rc = WEXITSTATUS(ret);
753 } else if (!nomtab) {
754 rc = update_mtab_entry(usource, target, "lustre", orig_options,