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
64 int md_stripe_cache_size = 16384;
65 char *progname = NULL;
69 fprintf(out, "%s v"LUSTRE_VERSION_STRING"\n", progname);
70 fprintf(out, "\nThis mount helper should only be invoked via the "
71 "mount (8) command,\ne.g. mount -t lustre dev dir\n\n");
72 fprintf(out, "usage: %s [-fhnv] [-o <mntopt>] <device> <mountpt>\n",
75 "\t<device>: the disk device, or for a client:\n"
76 "\t\t<mgmtnid>[:<altmgtnid>...]:/<filesystem>-client\n"
77 "\t<filesystem>: name of the Lustre filesystem (e.g. lustre1)\n"
78 "\t<mountpt>: filesystem mountpoint (e.g. /mnt/lustre)\n"
79 "\t-f|--fake: fake mount (updates /etc/mtab)\n"
80 "\t-o force|--force: force mount even if already in /etc/mtab\n"
81 "\t-h|--help: print this usage message\n"
82 "\t-n|--nomtab: do not update /etc/mtab after mount\n"
83 "\t-v|--verbose: print verbose config settings\n"
84 "\t<mntopt>: one or more comma separated of:\n"
85 "\t\t(no)flock,(no)user_xattr,(no)acl\n"
86 "\t\tabort_recov: abort server recovery handling\n"
87 "\t\tnosvc: only start MGC/MGS obds\n"
88 "\t\tnomgs: only start target obds, using existing MGS\n"
89 "\t\texclude=<ostname>[:<ostname>] : colon-separated list of "
90 "inactive OSTs (e.g. lustre-OST0001)\n"
91 "\t\tretry=<num>: number of times mount is retried by client\n"
92 "\t\tmd_stripe_cache_size=<num>: set the raid stripe cache "
93 "size for the underlying raid if present\n"
95 exit((out != stdout) ? EINVAL : 0);
98 /* Get rid of symbolic hostnames for tcp, since kernel can't do lookups */
99 #define MAXNIDSTR 1024
100 static char *convert_hostnames(char *s1)
102 char *converted, *s2 = 0, *c;
104 int left = MAXNIDSTR;
107 converted = malloc(left);
108 if (converted == NULL) {
109 fprintf(stderr, "out of memory: needed %d bytes\n",
114 while ((left > 0) && (*s1 != '/')) {
115 s2 = strpbrk(s1, ",:");
120 nid = libcfs_str2nid(s1);
121 *s2 = sep; /* back to original string */
122 if (nid == LNET_NID_ANY)
124 c += snprintf(c, left, "%s%c", libcfs_nid2str(nid), sep);
125 left = converted + MAXNIDSTR - c;
128 snprintf(c, left, "%s", s1);
131 fprintf(stderr, "%s: Can't parse NID '%s'\n", progname, s1);
136 /*****************************************************************************
138 * This part was cribbed from util-linux/mount/mount.c. There was no clear
139 * license information, but many other files in the package are identified as
140 * GNU GPL, so it's a pretty safe bet that was their intent.
142 ****************************************************************************/
144 const char *opt; /* option name */
145 int inv; /* true if flag value should be inverted */
146 int mask; /* flag mask value */
149 static const struct opt_map opt_map[] = {
150 /*"optname", inv,ms_mask */
151 /* These flags are parsed by mount, not lustre */
152 { "defaults", 0, 0 }, /* default options */
153 { "remount", 0, MS_REMOUNT}, /* remount with different options */
154 { "rw", 1, MS_RDONLY }, /* read-write */
155 { "ro", 0, MS_RDONLY }, /* read-only */
156 { "exec", 1, MS_NOEXEC }, /* permit execution of binaries */
157 { "noexec", 0, MS_NOEXEC }, /* don't execute binaries */
158 { "suid", 1, MS_NOSUID }, /* honor suid executables */
159 { "nosuid", 0, MS_NOSUID }, /* don't honor suid executables */
160 { "dev", 1, MS_NODEV }, /* interpret device files */
161 { "nodev", 0, MS_NODEV }, /* don't interpret devices */
162 { "sync", 0, MS_SYNCHRONOUS}, /* synchronous I/O */
163 { "async", 1, MS_SYNCHRONOUS}, /* asynchronous I/O */
164 { "atime", 1, MS_NOATIME }, /* set file access time on read */
165 { "noatime", 0, MS_NOATIME }, /* do not set file access time on read */
167 { "diratime", 1, MS_NODIRATIME }, /* set file access time on read */
168 { "nodiratime",0,MS_NODIRATIME }, /* do not set file access time on read */
171 { "relatime", 0, MS_RELATIME }, /* set file access time on read */
172 { "norelatime",1,MS_RELATIME }, /* do not set file access time on read */
174 #ifdef MS_STRICTATIME
175 { "strictatime",0,MS_STRICTATIME }, /* update access time strictly */
177 { "auto", 0, 0 }, /* Can be mounted using -a */
178 { "noauto", 0, 0 }, /* Can only be mounted explicitly */
179 { "nousers", 1, 0 }, /* Forbid ordinary user to mount */
180 { "nouser", 1, 0 }, /* Forbid ordinary user to mount */
181 { "noowner", 1, 0 }, /* Device owner has no special privs */
182 { "_netdev", 0, 0 }, /* Device accessible only via network */
186 /****************************************************************************/
188 /* 1 = don't pass on to lustre
189 0 = pass on to lustre */
190 static int parse_one_option(const char *check, int *flagp)
192 const struct opt_map *opt;
194 for (opt = &opt_map[0]; opt->opt != NULL; opt++) {
195 if (strncmp(check, opt->opt, strlen(opt->opt)) == 0) {
198 *flagp &= ~(opt->mask);
205 /* Assume any unknown options are valid and pass them on. The mount
206 will fail if lmd_parse, ll_options or ldiskfs doesn't recognize it.*/
210 static void append_option(char *options, const char *one)
213 strcat(options, ",");
214 strcat(options, one);
217 /* Replace options with subset of Lustre-specific options, and
218 fill in mount flags */
219 int parse_options(struct mount_opts *mop, char *orig_options, int *flagp)
221 char *options, *opt, *nextopt, *arg, *val;
223 options = calloc(strlen(orig_options) + 1, 1);
225 nextopt = orig_options;
226 while ((opt = strsep(&nextopt, ","))) {
231 /* Handle retries in a slightly different
234 val = strchr(opt, '=');
235 /* please note that some ldiskfs mount options are also in the form
236 * of param=value. We should pay attention not to remove those
237 * mount options, see bug 22097. */
238 if (val && strncmp(arg, "md_stripe_cache_size", 20) == 0) {
239 md_stripe_cache_size = atoi(val + 1);
240 } else if (val && strncmp(arg, "retry", 5) == 0) {
241 mop->mo_retry = atoi(val + 1);
242 if (mop->mo_retry > MAX_RETRIES)
243 mop->mo_retry = MAX_RETRIES;
244 else if (mop->mo_retry < 0)
246 } else if (val && strncmp(arg, "mgssec", 6) == 0) {
247 append_option(options, opt);
248 } else if (strcmp(opt, "force") == 0) {
249 //XXX special check for 'force' option
251 printf("force: %d\n", mop->mo_force);
252 } else if (parse_one_option(opt, flagp) == 0) {
253 /* pass this on as an option */
254 append_option(options, opt);
257 #ifdef MS_STRICTATIME
258 /* set strictatime to default if NOATIME or RELATIME
259 not given explicit */
260 if (!(*flagp & (MS_NOATIME | MS_RELATIME)))
261 *flagp |= MS_STRICTATIME;
263 strcpy(orig_options, options);
269 int read_file(char *path, char *buf, int size)
273 fd = fopen(path, "r");
277 /* should not ignore fgets(3)'s return value */
278 if (!fgets(buf, size, fd)) {
279 fprintf(stderr, "reading from %s: %s", path, strerror(errno));
287 int write_file(char *path, char *buf)
291 fd = fopen(path, "w");
300 /* This is to tune the kernel for good SCSI performance.
301 * For that we set the value of /sys/block/{dev}/queue/max_sectors_kb
302 * to the value of /sys/block/{dev}/queue/max_hw_sectors_kb */
303 int set_blockdev_tunables(char *source, int fan_out)
305 glob_t glob_info = { 0 };
306 struct stat stat_buf;
307 char *chk_major, *chk_minor;
308 char *savept = NULL, *dev;
310 char buf[PATH_MAX] = {'\0'}, path[PATH_MAX] = {'\0'};
311 char real_path[PATH_MAX] = {'\0'};
318 ret_path = realpath(source, real_path);
319 if (ret_path == NULL) {
321 fprintf(stderr, "warning: %s: cannot resolve: %s\n",
322 source, strerror(errno));
326 if (strncmp(real_path, "/dev/loop", 9) == 0)
329 if ((real_path[0] != '/') && (strpbrk(real_path, ",:") != NULL))
332 snprintf(path, sizeof(path), "/sys/block%s", real_path + 4);
333 if (access(path, X_OK) == 0)
336 /* The name of the device say 'X' specified in /dev/X may not
337 * match any entry under /sys/block/. In that case we need to
338 * match the major/minor number to find the entry under
339 * sys/block corresponding to /dev/X */
341 /* Don't chop tail digit on /dev/mapper/xxx, LU-478 */
342 if (strncmp(real_path, "/dev/mapper", 11) != 0) {
343 dev = real_path + strlen(real_path);
344 while (--dev > real_path && isdigit(*dev))
347 if (strncmp(real_path, "/dev/md_", 8) == 0)
351 rc = stat(real_path, &stat_buf);
354 fprintf(stderr, "warning: %s, device %s stat failed\n",
355 strerror(errno), real_path);
359 major = major(stat_buf.st_rdev);
360 minor = minor(stat_buf.st_rdev);
361 rc = glob("/sys/block/*", GLOB_NOSORT, NULL, &glob_info);
364 fprintf(stderr, "warning: failed to read entries under "
366 globfree(&glob_info);
370 for (i = 0; i < glob_info.gl_pathc; i++){
371 snprintf(path, sizeof(path), "%s/dev", glob_info.gl_pathv[i]);
373 rc = read_file(path, buf, sizeof(buf));
377 if (buf[strlen(buf) - 1] == '\n')
378 buf[strlen(buf) - 1] = '\0';
380 chk_major = strtok_r(buf, ":", &savept);
382 if (major == atoi(chk_major) &&minor == atoi(chk_minor))
386 if (i == glob_info.gl_pathc) {
388 fprintf(stderr,"warning: device %s does not match any "
389 "entry under /sys/block\n", real_path);
390 globfree(&glob_info);
394 /* Chop off "/dev" from path we found */
395 path[strlen(glob_info.gl_pathv[i])] = '\0';
396 globfree(&glob_info);
399 if (strncmp(real_path, "/dev/md", 7) == 0) {
400 snprintf(real_path, sizeof(real_path), "%s/%s", path,
403 rc = read_file(real_path, buf, sizeof(buf));
406 fprintf(stderr, "warning: opening %s: %s\n",
407 real_path, strerror(errno));
411 if (atoi(buf) >= md_stripe_cache_size)
414 if (strlen(buf) - 1 > 0) {
415 snprintf(buf, sizeof(buf), "%d", md_stripe_cache_size);
416 rc = write_file(real_path, buf);
418 fprintf(stderr, "warning: opening %s: %s\n",
419 real_path, strerror(errno));
421 /* Return since raid and disk tunables are different */
425 snprintf(real_path, sizeof(real_path), "%s/%s", path,
426 MAX_HW_SECTORS_KB_PATH);
427 rc = read_file(real_path, buf, sizeof(buf));
430 fprintf(stderr, "warning: opening %s: %s\n",
431 real_path, strerror(errno));
432 /* No MAX_HW_SECTORS_KB_PATH isn't necessary an
433 * error for some device. */
437 if (strlen(buf) - 1 > 0) {
438 snprintf(real_path, sizeof(real_path), "%s/%s", path,
439 MAX_SECTORS_KB_PATH);
440 rc = write_file(real_path, buf);
443 fprintf(stderr, "warning: writing to %s: %s\n",
444 real_path, strerror(errno));
445 /* No MAX_SECTORS_KB_PATH isn't necessary an
446 * error for some device. */
453 glob_info.gl_pathc = 0;
454 glob_info.gl_offs = 0;
455 /* if device is multipath device, tune its slave devices */
456 snprintf(real_path, sizeof(real_path), "%s/slaves/*", path);
457 rc = glob(real_path, GLOB_NOSORT, NULL, &glob_info);
459 for (i = 0; rc == 0 && i < glob_info.gl_pathc; i++){
460 slave = basename(glob_info.gl_pathv[i]);
461 snprintf(real_path, sizeof(real_path), "/dev/%s", slave);
462 rc = set_blockdev_tunables(real_path, 0);
465 if (rc == GLOB_NOMATCH) {
466 /* no slave device is not an error */
468 } else if (rc && verbose) {
470 fprintf(stderr, "warning: %s, failed to read"
471 " entries under %s/slaves\n",
472 strerror(errno), path);
474 fprintf(stderr, "unable to set tunables for"
475 " slave device %s (slave would be"
476 " unable to handle IO request from"
481 globfree(&glob_info);
487 static void set_defaults(struct mount_opts *mop)
489 memset(mop, 0, sizeof(*mop));
490 mop->mo_usource = NULL;
491 mop->mo_source = NULL;
496 mop->mo_have_mgsnid = 0;
497 mop->mo_md_stripe_cache_size = 16384;
498 mop->mo_orig_options = "";
501 static int parse_opts(int argc, char *const argv[], struct mount_opts *mop)
503 static struct option long_opt[] = {
507 {"nomtab", 0, 0, 'n'},
508 {"options", 1, 0, 'o'},
509 {"verbose", 0, 0, 'v'},
512 char real_path[PATH_MAX] = {'\0'};
514 char path[256], name[256];
519 while ((opt = getopt_long(argc, argv, "fhno:v",
520 long_opt, NULL)) != EOF){
524 printf("force: %d\n", mop->mo_force);
528 printf("fake: %d\n", mop->mo_fake);
535 printf("nomtab: %d\n", mop->mo_nomtab);
538 mop->mo_orig_options = optarg;
544 fprintf(stderr, "%s: unknown option '%c'\n",
551 if (optind + 2 > argc) {
552 fprintf(stderr, "%s: too few arguments\n", progname);
556 mop->mo_usource = argv[optind];
557 if (!mop->mo_usource) {
562 * Try to get the real path to the device, in case it is a
563 * symbolic link for instance
565 if (realpath(mop->mo_usource, real_path) != NULL) {
566 mop->mo_usource = strdup(real_path);
568 ptr = strrchr(real_path, '/');
569 if (ptr && strncmp(ptr, "/dm-", 4) == 0 && isdigit(*(ptr + 4))) {
570 snprintf(path, sizeof(path), "/sys/block/%s/dm/name", ptr+1);
571 if ((f = fopen(path, "r"))) {
572 /* read "<name>\n" from sysfs */
573 if (fgets(name, sizeof(name), f) && (sz = strlen(name)) > 1) {
575 snprintf(real_path, sizeof(real_path), "/dev/mapper/%s", name);
582 mop->mo_source = convert_hostnames(mop->mo_usource);
583 if (!mop->mo_source) {
587 if (realpath(argv[optind + 1], mop->mo_target) == NULL) {
589 fprintf(stderr, "warning: %s: cannot resolve: %s\n",
590 argv[optind + 1], strerror(errno));
597 int main(int argc, char *const argv[])
599 struct mount_opts mop;
600 char *options, *optcopy;
601 int i, rc, flags, optlen;
603 progname = strrchr(argv[0], '/');
604 progname = progname ? progname + 1 : argv[0];
608 rc = parse_opts(argc, argv, &mop);
613 for (i = 0; i < argc; i++)
614 printf("arg[%d] = %s\n", i, argv[i]);
615 printf("source = %s (%s), target = %s\n", mop.mo_usource,
616 mop.mo_source, mop.mo_target);
617 printf("options = %s\n", mop.mo_orig_options);
620 options = malloc(strlen(mop.mo_orig_options) + 1);
621 if (options == NULL) {
622 fprintf(stderr, "can't allocate memory for options\n");
625 strcpy(options, mop.mo_orig_options);
626 rc = parse_options(&mop, options, &flags);
628 fprintf(stderr, "%s: can't parse options: %s\n",
634 rc = check_mtab_entry(mop.mo_usource, mop.mo_source,
635 mop.mo_target, "lustre");
636 if (rc && !(flags & MS_REMOUNT)) {
637 fprintf(stderr, "%s: according to %s %s is "
638 "already mounted on %s\n", progname, MOUNTED,
639 mop.mo_usource, mop.mo_target);
642 if (!rc && (flags & MS_REMOUNT)) {
643 fprintf(stderr, "%s: according to %s %s is "
644 "not already mounted on %s\n", progname, MOUNTED,
645 mop.mo_usource, mop.mo_target);
649 if (flags & MS_REMOUNT)
652 rc = access(mop.mo_target, F_OK);
655 fprintf(stderr, "%s: %s inaccessible: %s\n", progname,
656 mop.mo_target, strerror(errno));
660 /* In Linux 2.4, the target device doesn't get passed to any of our
661 functions. So we'll stick it on the end of the options. */
662 optlen = strlen(options) + strlen(",device=") + strlen(mop.mo_source) + 1;
663 optcopy = malloc(optlen);
664 if (optcopy == NULL) {
665 fprintf(stderr, "can't allocate memory to optcopy\n");
668 strcpy(optcopy, options);
670 strcat(optcopy, ",");
671 strcat(optcopy, "device=");
672 strcat(optcopy, mop.mo_source);
675 printf("mounting device %s at %s, flags=%#x options=%s\n",
676 mop.mo_source, mop.mo_target, flags, optcopy);
678 if (!strstr(mop.mo_usource, ":/") && set_blockdev_tunables(mop.mo_source, 1)) {
680 fprintf(stderr, "%s: unable to set tunables for %s"
681 " (may cause reduced IO performance)\n",
682 argv[0], mop.mo_source);
686 /* flags and target get to lustre_get_sb, but not
687 lustre_fill_super. Lustre ignores the flags, but mount
689 for (i = 0, rc = -EAGAIN; i <= mop.mo_retry && rc != 0; i++) {
690 rc = mount(mop.mo_source, mop.mo_target, "lustre",
691 flags, (void *)optcopy);
694 fprintf(stderr, "%s: mount %s at %s "
695 "failed: %s retries left: "
696 "%d\n", basename(progname),
697 mop.mo_usource, mop.mo_target,
703 sleep(1 << max((i/2), 5));
717 cli = strrchr(mop.mo_usource, ':');
718 if (cli && (strlen(cli) > 2))
723 fprintf(stderr, "%s: mount %s at %s failed: %s\n", progname,
724 mop.mo_usource, mop.mo_target, strerror(errno));
726 fprintf(stderr, "Are the lustre modules loaded?\n"
727 "Check /etc/modprobe.conf and "
728 "/proc/filesystems\n");
729 if (errno == ENOTBLK)
730 fprintf(stderr, "Do you need -o loop?\n");
731 if (errno == ENOMEDIUM)
733 "This filesystem needs at least 1 OST\n");
734 if (errno == ENOENT) {
735 fprintf(stderr, "Is the MGS specification correct?\n");
736 fprintf(stderr, "Is the filesystem name correct?\n");
737 fprintf(stderr, "If upgrading, is the copied client log"
738 " valid? (see upgrade docs)\n");
740 if (errno == EALREADY)
741 fprintf(stderr, "The target service is already running."
742 " (%s)\n", mop.mo_usource);
744 fprintf(stderr, "The target service failed to start "
745 "(bad config log?) (%s). "
746 "See /var/log/messages.\n", mop.mo_usource);
748 fprintf(stderr, "Is the MGS running?\n");
749 if (errno == EADDRINUSE)
750 fprintf(stderr, "The target service's index is already "
751 "in use. (%s)\n", mop.mo_usource);
752 if (errno == EINVAL) {
753 fprintf(stderr, "This may have multiple causes.\n");
755 fprintf(stderr, "Is '%s' the correct filesystem"
757 fprintf(stderr, "Are the mount options correct?\n");
758 fprintf(stderr, "Check the syslog for more info.\n");
761 /* May as well try to clean up loop devs */
762 if (strncmp(mop.mo_usource, "/dev/loop", 9) == 0) {
765 sprintf(cmd, "/sbin/losetup -d %s", mop.mo_usource);
766 if ((ret = system(cmd)) < 0)
769 rc = WEXITSTATUS(ret);
772 } else if (!mop.mo_nomtab) {
773 rc = update_mtab_entry(mop.mo_usource, mop.mo_target, "lustre",
774 mop.mo_orig_options, 0,0,0);
778 /* mo_usource should be freed, but we can rely on the kernel */