*
* You should have received a copy of the GNU General Public License
* version 2 along with this program; If not, see
- * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
- *
- * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
- * CA 95054 USA or visit www.sun.com if you need additional information or
- * have any questions.
+ * http://www.gnu.org/licenses/gpl-2.0.html
*
* GPL HEADER END
*/
* Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2011, 2012, Whamcloud, Inc.
+ * Copyright (c) 2011, 2016, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#ifndef _GNU_SOURCE
#define _GNU_SOURCE
#endif
+#include <getopt.h>
+#include <mntent.h>
+#include <stdbool.h>
#include <stdlib.h>
#include <stdio.h>
#include <unistd.h>
#include <fcntl.h>
#include <errno.h>
#include <string.h>
-#include "obdctl.h"
+#include <sys/mount.h>
+#include <sys/param.h>
#include <lustre_ver.h>
-#include <glob.h>
#include <ctype.h>
#include <limits.h>
+#include <lnet/nidstr.h>
+#include <libcfs/util/string.h>
+
+#include "obdctl.h"
#include "mount_utils.h"
-#define MAX_HW_SECTORS_KB_PATH "queue/max_hw_sectors_kb"
-#define MAX_SECTORS_KB_PATH "queue/max_sectors_kb"
-#define STRIPE_CACHE_SIZE "md/stripe_cache_size"
-#define MAXOPT 4096
+#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 10, 53, 0)
+/*
+ * LU-1783
+ * We only #include a kernel level include file here because
+ * important MS_ flag #defines are missing from the SLES version
+ * of sys/mount.h
+ * In the future if SLES updates sys/mount.h to have a more complete
+ * set of flag #defines we should stop including linux/fs.h
+ */
+#if !defined(MS_RDONLY)
+#include <linux/fs.h>
+#endif
+#endif
+
+#ifdef HAVE_LIBMOUNT
+# define WITH_LIBMOUNT "(libmount)"
+#else
+# define WITH_LIBMOUNT ""
+#endif
+
#define MAX_RETRIES 99
-int verbose = 0;
-int md_stripe_cache_size = 16384;
-char *progname = NULL;
+int verbose;
+int version;
+char *progname;
void usage(FILE *out)
{
- fprintf(out, "%s v"LUSTRE_VERSION_STRING"\n", progname);
- fprintf(out, "\nThis mount helper should only be invoked via the "
- "mount (8) command,\ne.g. mount -t lustre dev dir\n\n");
- fprintf(out, "usage: %s [-fhnv] [-o <mntopt>] <device> <mountpt>\n",
- progname);
- fprintf(out,
- "\t<device>: the disk device, or for a client:\n"
- "\t\t<mgmtnid>[:<altmgtnid>...]:/<filesystem>-client\n"
- "\t<filesystem>: name of the Lustre filesystem (e.g. lustre1)\n"
- "\t<mountpt>: filesystem mountpoint (e.g. /mnt/lustre)\n"
- "\t-f|--fake: fake mount (updates /etc/mtab)\n"
- "\t-o force|--force: force mount even if already in /etc/mtab\n"
- "\t-h|--help: print this usage message\n"
- "\t-n|--nomtab: do not update /etc/mtab after mount\n"
- "\t-v|--verbose: print verbose config settings\n"
- "\t<mntopt>: one or more comma separated of:\n"
- "\t\t(no)flock,(no)user_xattr,(no)acl\n"
- "\t\tabort_recov: abort server recovery handling\n"
- "\t\tnosvc: only start MGC/MGS obds\n"
- "\t\tnomgs: only start target obds, using existing MGS\n"
- "\t\texclude=<ostname>[:<ostname>] : colon-separated list of "
- "inactive OSTs (e.g. lustre-OST0001)\n"
- "\t\tretry=<num>: number of times mount is retried by client\n"
- "\t\tmd_stripe_cache_size=<num>: set the raid stripe cache "
- "size for the underlying raid if present\n"
- );
- exit((out != stdout) ? EINVAL : 0);
+ fprintf(out, "\nThis mount helper should only be invoked via the "
+ "mount (8) command,\ne.g. mount -t lustre dev dir\n\n");
+ fprintf(out, "usage: %s [-fhnvV] [-o <mntopt>] <device> <mountpt>\n",
+ progname);
+ fprintf(out,
+ "\t<device>: the disk device, or for a client:\n"
+ "\t\t<mgsnid>[:<altmgsnid>...]:/<filesystem>[/<subdir>]\n"
+ "\t<filesystem>: name of the Lustre filesystem (e.g. lustre1)\n"
+ "\t<mountpt>: filesystem mountpoint (e.g. /mnt/lustre)\n"
+ "\t-f|--fake: fake mount (updates /etc/mtab)\n"
+ "\t-o force|--force: force mount even if already in /etc/mtab\n"
+ "\t-h|--help: print this usage message\n"
+ "\t-n|--nomtab: do not update /etc/mtab after mount\n"
+ "\t-v|--verbose: print verbose config settings\n"
+ "\t-V|--version: output build version of the utility and exit\n"
+ "\t<mntopt>: one or more comma separated of:\n"
+ "\t\t(no)flock,(no)user_xattr,(no)acl\n"
+ "\t\tabort_recov: abort server recovery handling\n"
+ "\t\tnosvc: only start MGC/MGS obds\n"
+ "\t\tnomgs: only start target obds, using existing MGS\n"
+ "\t\tnoscrub: NOT auto start OI scrub unless start explicitly\n"
+ "\t\tskip_lfsck: NOT auto resume the paused/crashed LFSCK\n"
+ "\t\texclude=<ostname>[:<ostname>] : colon-separated list of "
+ "inactive OSTs (e.g. lustre-OST0001)\n"
+#ifdef HAVE_GSS
+ "\t\tskpath=<file|directory>: Path to a file or directory of"
+ "key configuration files to load into the kernel keyring\n"
+#endif
+ "\t\tretry=<num>: number of times mount is retried by client\n"
+ "\t\tmd_stripe_cache_size=<num>: set the raid stripe cache "
+ "size for the underlying raid if present\n");
+ exit((out != stdout) ? EINVAL : 0);
}
/* Get rid of symbolic hostnames for tcp, since kernel can't do lookups */
return 0;
}
-static void append_option(char *options, const char *one)
+static int append_option(char *options, size_t options_len,
+ const char *param, const char *value)
{
- if (*options)
- strcat(options, ",");
- strcat(options, one);
+ int rc;
+ if (options[0] != '\0') {
+ rc = strlcat(options, ",", options_len);
+ if (rc >= options_len)
+ goto out_err;
+ }
+ rc = strlcat(options, param, options_len);
+ if (rc >= options_len)
+ goto out_err;
+ if (value != NULL) {
+ rc = strlcat(options, value, options_len);
+ if (rc >= options_len)
+ goto out_err;
+ }
+ return 0;
+out_err:
+ fprintf(stderr, "error: mount options %s%s too long\n", param, value);
+ return E2BIG;
}
/* Replace options with subset of Lustre-specific options, and
fill in mount flags */
-int parse_options(struct mount_opts *mop, char *orig_options, int *flagp)
+int parse_options(struct mount_opts *mop, char *orig_options,
+ int *flagp, size_t options_len)
{
- char *options, *opt, *nextopt, *arg, *val;
-
- options = calloc(strlen(orig_options) + 1, 1);
- *flagp = 0;
- nextopt = orig_options;
- while ((opt = strsep(&nextopt, ","))) {
- if (!*opt)
- /* empty option */
- continue;
-
- /* Handle retries in a slightly different
- * manner */
- arg = opt;
- val = strchr(opt, '=');
- /* please note that some ldiskfs mount options are also in the form
- * of param=value. We should pay attention not to remove those
- * mount options, see bug 22097. */
- if (val && strncmp(arg, "md_stripe_cache_size", 20) == 0) {
- md_stripe_cache_size = atoi(val + 1);
- } else if (val && strncmp(arg, "retry", 5) == 0) {
+ char *options, *opt, *nextopt, *arg, *val;
+ int rc = 0;
+
+ options = calloc(strlen(orig_options) + 1, 1);
+ *flagp = 0;
+ nextopt = orig_options;
+ while ((opt = strsep(&nextopt, ","))) {
+ if (!*opt)
+ /* empty option */
+ continue;
+
+ /* Handle retries in a slightly different
+ * manner */
+ arg = opt;
+ val = strchr(opt, '=');
+ /* please note that some ldiskfs mount options are also in
+ * the form of param=value. We should pay attention not to
+ * remove those mount options, see bug 22097. */
+ if (val && strncmp(arg, "max_sectors_kb", 14) == 0) {
+ mop->mo_max_sectors_kb = atoi(val + 1);
+ } else if (val &&
+ strncmp(arg, "md_stripe_cache_size", 20) == 0) {
+ mop->mo_md_stripe_cache_size = atoi(val + 1);
+ } else if (val && strncmp(arg, "retry", 5) == 0) {
mop->mo_retry = atoi(val + 1);
if (mop->mo_retry > MAX_RETRIES)
mop->mo_retry = MAX_RETRIES;
else if (mop->mo_retry < 0)
mop->mo_retry = 0;
- } else if (val && strncmp(arg, "mgssec", 6) == 0) {
- append_option(options, opt);
- } else if (strcmp(opt, "force") == 0) {
- //XXX special check for 'force' option
+ } else if (val && strncmp(arg, "mgssec", 6) == 0) {
+ rc = append_option(options, options_len, opt, NULL);
+ if (rc != 0)
+ goto out_options;
+ } else if (strncmp(arg, "nosvc", 5) == 0) {
+ mop->mo_nosvc = 1;
+ rc = append_option(options, options_len, opt, NULL);
+ if (rc != 0)
+ goto out_options;
+ } else if (strcmp(opt, "force") == 0) {
+ /* XXX special check for 'force' option */
++mop->mo_force;
printf("force: %d\n", mop->mo_force);
- } else if (parse_one_option(opt, flagp) == 0) {
- /* pass this on as an option */
- append_option(options, opt);
- }
- }
+#ifdef HAVE_GSS
+ } else if (val && strncmp(opt, "skpath=", 7) == 0) {
+ if (strlen(val) + 1 >= sizeof(mop->mo_skpath)) {
+ fprintf(stderr,
+ "%s: shared key path too long\n",
+ progname);
+ free(options);
+ return -1;
+ }
+ strncpy(mop->mo_skpath, val + 1, strlen(val + 1));
+#endif
+ } else if (parse_one_option(opt, flagp) == 0) {
+ /* pass this on as an option */
+ rc = append_option(options, options_len, opt, NULL);
+ if (rc != 0)
+ goto out_options;
+ }
+ }
#ifdef MS_STRICTATIME
- /* set strictatime to default if NOATIME or RELATIME
- not given explicit */
- if (!(*flagp & (MS_NOATIME | MS_RELATIME)))
- *flagp |= MS_STRICTATIME;
+#if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(3, 2, 53, 0)
+ /*
+ * LU-1783
+ * In the future when upstream fixes land in all supported kernels
+ * we should stop forcing MS_STRICTATIME in lustre mounts.
+ * We override the kernel level default of MS_RELATIME for now
+ * due to a kernel vfs level bug in atime updates that fails
+ * to reset timestamps from the future.
+ */
+#warn "remove MS_STRICTATIME override if kernel updates atime from the future"
#endif
- strcpy(orig_options, options);
- free(options);
- return 0;
-}
-
-
-int read_file(char *path, char *buf, int size)
-{
- FILE *fd;
-
- fd = fopen(path, "r");
- if (fd == NULL)
- return errno;
+ /* set strictatime to default if NOATIME or RELATIME
+ not given explicit */
+ if (!(*flagp & (MS_NOATIME | MS_RELATIME)))
+ *flagp |= MS_STRICTATIME;
+#endif
+ strcpy(orig_options, options);
- /* should not ignore fgets(3)'s return value */
- if (!fgets(buf, size, fd)) {
- fprintf(stderr, "reading from %s: %s", path, strerror(errno));
- fclose(fd);
- return 1;
- }
- fclose(fd);
- return 0;
+out_options:
+ free(options);
+ return rc;
}
-int write_file(char *path, char *buf)
+/* Add mgsnids from ldd params */
+static int add_mgsnids(struct mount_opts *mop, char *options,
+ const char *params, size_t options_len)
{
- FILE *fd;
-
- fd = fopen(path, "w");
- if (fd == NULL)
- return errno;
+ char *ptr = (char *)params;
+ char tmp, *sep;
+ int rc = 0;
+
+ while ((ptr = strstr(ptr, PARAM_MGSNODE)) != NULL) {
+ sep = strchr(ptr, ' ');
+ if (sep != NULL) {
+ tmp = *sep;
+ *sep = '\0';
+ }
+ rc = append_option(options, options_len, ptr, NULL);
+ if (rc != 0)
+ goto out;
+ mop->mo_have_mgsnid++;
+ if (sep) {
+ *sep = tmp;
+ ptr = sep;
+ } else {
+ break;
+ }
+ }
- fputs(buf, fd);
- fclose(fd);
- return 0;
+out:
+ return rc;
}
-/* This is to tune the kernel for good SCSI performance.
- * For that we set the value of /sys/block/{dev}/queue/max_sectors_kb
- * to the value of /sys/block/{dev}/queue/max_hw_sectors_kb */
-int set_blockdev_tunables(char *source, int fan_out)
+static int clear_update_ondisk(char *source, struct lustre_disk_data *ldd)
{
- glob_t glob_info = { 0 };
- struct stat stat_buf;
- char *chk_major, *chk_minor;
- char *savept = NULL, *dev;
- char *ret_path;
- char buf[PATH_MAX] = {'\0'}, path[PATH_MAX] = {'\0'};
- char real_path[PATH_MAX] = {'\0'};
- int i, rc = 0;
- int major, minor;
-
- if (!source)
- return -EINVAL;
-
- ret_path = realpath(source, real_path);
- if (ret_path == NULL) {
- if (verbose)
- fprintf(stderr, "warning: %s: cannot resolve: %s\n",
- source, strerror(errno));
- return -EINVAL;
- }
-
- if (strncmp(real_path, "/dev/loop", 9) == 0)
- return 0;
-
- if ((real_path[0] != '/') && (strpbrk(real_path, ",:") != NULL))
- return 0;
-
- snprintf(path, sizeof(path), "/sys/block%s", real_path + 4);
- if (access(path, X_OK) == 0)
- goto set_params;
-
- /* The name of the device say 'X' specified in /dev/X may not
- * match any entry under /sys/block/. In that case we need to
- * match the major/minor number to find the entry under
- * sys/block corresponding to /dev/X */
-
- /* Don't chop tail digit on /dev/mapper/xxx, LU-478 */
- if (strncmp(real_path, "/dev/mapper", 11) != 0) {
- dev = real_path + strlen(real_path);
- while (--dev > real_path && isdigit(*dev))
- *dev = 0;
-
- if (strncmp(real_path, "/dev/md_", 8) == 0)
- *dev = 0;
- }
-
- rc = stat(real_path, &stat_buf);
- if (rc) {
- if (verbose)
- fprintf(stderr, "warning: %s, device %s stat failed\n",
- strerror(errno), real_path);
- return rc;
- }
-
- major = major(stat_buf.st_rdev);
- minor = minor(stat_buf.st_rdev);
- rc = glob("/sys/block/*", GLOB_NOSORT, NULL, &glob_info);
- if (rc) {
- if (verbose)
- fprintf(stderr, "warning: failed to read entries under "
- "/sys/block\n");
- globfree(&glob_info);
- return rc;
- }
-
- for (i = 0; i < glob_info.gl_pathc; i++){
- snprintf(path, sizeof(path), "%s/dev", glob_info.gl_pathv[i]);
-
- rc = read_file(path, buf, sizeof(buf));
- if (rc)
- continue;
-
- if (buf[strlen(buf) - 1] == '\n')
- buf[strlen(buf) - 1] = '\0';
-
- chk_major = strtok_r(buf, ":", &savept);
- chk_minor = savept;
- if (major == atoi(chk_major) &&minor == atoi(chk_minor))
- break;
- }
-
- if (i == glob_info.gl_pathc) {
- if (verbose)
- fprintf(stderr,"warning: device %s does not match any "
- "entry under /sys/block\n", real_path);
- globfree(&glob_info);
- return -EINVAL;
- }
-
- /* Chop off "/dev" from path we found */
- path[strlen(glob_info.gl_pathv[i])] = '\0';
- globfree(&glob_info);
-
-set_params:
- if (strncmp(real_path, "/dev/md", 7) == 0) {
- snprintf(real_path, sizeof(real_path), "%s/%s", path,
- STRIPE_CACHE_SIZE);
-
- rc = read_file(real_path, buf, sizeof(buf));
- if (rc) {
- if (verbose)
- fprintf(stderr, "warning: opening %s: %s\n",
- real_path, strerror(errno));
- return 0;
- }
-
- if (atoi(buf) >= md_stripe_cache_size)
- return 0;
-
- if (strlen(buf) - 1 > 0) {
- snprintf(buf, sizeof(buf), "%d", md_stripe_cache_size);
- rc = write_file(real_path, buf);
- if (rc && verbose)
- fprintf(stderr, "warning: opening %s: %s\n",
- real_path, strerror(errno));
- }
- /* Return since raid and disk tunables are different */
- return rc;
- }
-
- snprintf(real_path, sizeof(real_path), "%s/%s", path,
- MAX_HW_SECTORS_KB_PATH);
- rc = read_file(real_path, buf, sizeof(buf));
- if (rc) {
- if (verbose)
- fprintf(stderr, "warning: opening %s: %s\n",
- real_path, strerror(errno));
- /* No MAX_HW_SECTORS_KB_PATH isn't necessary an
- * error for some device. */
- rc = 0;
- }
+ char wanted_mountopts[512] = "";
+ struct mkfs_opts mkop;
+ int ret;
+ int ret2;
+
+ memset(&mkop, 0, sizeof(mkop));
+ mkop.mo_ldd = *ldd;
+ mkop.mo_ldd.ldd_flags &= ~LDD_F_UPDATE;
+ mkop.mo_flags = MO_NOHOSTID_CHECK; /* Ignore missing hostid */
+ if (strlen(source) > sizeof(mkop.mo_device)-1) {
+ fatal();
+ fprintf(stderr, "Device name too long: %s\n", source);
+ return -E2BIG;
+ }
+ strncpy(mkop.mo_device, source, sizeof(mkop.mo_device));
+
+ ret = osd_prepare_lustre(&mkop,
+ wanted_mountopts, sizeof(wanted_mountopts));
+ if (ret) {
+ fatal();
+ fprintf(stderr, "Can't prepare device %s: %s\n",
+ source, strerror(ret));
+ return ret;
+ }
- if (strlen(buf) - 1 > 0) {
- snprintf(real_path, sizeof(real_path), "%s/%s", path,
- MAX_SECTORS_KB_PATH);
- rc = write_file(real_path, buf);
- if (rc) {
- if (verbose)
- fprintf(stderr, "warning: writing to %s: %s\n",
- real_path, strerror(errno));
- /* No MAX_SECTORS_KB_PATH isn't necessary an
- * error for some device. */
- rc = 0;
- }
- }
+ /* Create the loopback file */
+ if (mkop.mo_flags & MO_IS_LOOP) {
+ ret = access(mkop.mo_device, F_OK);
+ if (ret) {
+ ret = errno;
+ fatal();
+ fprintf(stderr, "Can't access device %s: %s\n",
+ source, strerror(ret));
+ return ret;
+ }
- if (fan_out) {
- char *slave = NULL;
- glob_info.gl_pathc = 0;
- glob_info.gl_offs = 0;
- /* if device is multipath device, tune its slave devices */
- snprintf(real_path, sizeof(real_path), "%s/slaves/*", path);
- rc = glob(real_path, GLOB_NOSORT, NULL, &glob_info);
-
- for (i = 0; rc == 0 && i < glob_info.gl_pathc; i++){
- slave = basename(glob_info.gl_pathv[i]);
- snprintf(real_path, sizeof(real_path), "/dev/%s", slave);
- rc = set_blockdev_tunables(real_path, 0);
- }
+ ret = loop_setup(&mkop);
+ if (ret) {
+ fatal();
+ fprintf(stderr, "Loop device setup for %s failed: %s\n",
+ mkop.mo_device, strerror(ret));
+ return ret;
+ }
+ }
+ ret = osd_write_ldd(&mkop);
+ if (ret != 0) {
+ fatal();
+ fprintf(stderr, "failed to write local files: %s\n",
+ strerror(ret));
+ }
- if (rc == GLOB_NOMATCH) {
- /* no slave device is not an error */
- rc = 0;
- } else if (rc && verbose) {
- if (slave == NULL) {
- fprintf(stderr, "warning: %s, failed to read"
- " entries under %s/slaves\n",
- strerror(errno), path);
- } else {
- fprintf(stderr, "unable to set tunables for"
- " slave device %s (slave would be"
- " unable to handle IO request from"
- " master %s)\n",
- real_path, source);
- }
- }
- globfree(&glob_info);
- }
+ ret2 = loop_cleanup(&mkop);
+ if (ret == 0)
+ ret = ret2;
- return rc;
+ return ret;
}
-static int parse_ldd(char *source, struct mount_opts *mop, char *options)
+static int parse_ldd(char *source, struct mount_opts *mop,
+ char *options, size_t options_len)
{
struct lustre_disk_data *ldd = &mop->mo_ldd;
- int rc;
+ char *cur, *start;
+ int rc = 0;
rc = osd_is_lustre(source, &ldd->ldd_mount_type);
if (rc == 0) {
return ENODEV;
}
- return 0;
+ rc = osd_read_ldd(source, ldd);
+ if (rc) {
+ fprintf(stderr, "%s: %s failed to read permanent mount"
+ " data: %s\n", progname, source,
+ rc >= 0 ? strerror(rc) : "");
+ return rc;
+ }
+
+ if ((IS_MDT(ldd) || IS_OST(ldd)) &&
+ (ldd->ldd_flags & LDD_F_NEED_INDEX)) {
+ fprintf(stderr, "%s: %s has no index assigned "
+ "(probably formatted with old mkfs)\n",
+ progname, source);
+ return EINVAL;
+ }
+
+ if (ldd->ldd_flags & LDD_F_UPGRADE14) {
+ fprintf(stderr, "%s: we cannot upgrade %s from this (very old) "
+ "Lustre version\n", progname, source);
+ return EINVAL;
+ }
+
+ if (ldd->ldd_flags & LDD_F_UPDATE)
+ clear_update_ondisk(source, ldd);
+
+ /* Since we never rewrite ldd, ignore temp flags */
+ ldd->ldd_flags &= ~(LDD_F_VIRGIN | LDD_F_WRITECONF);
+
+ /* svname of the form lustre:OST1234 means never registered */
+ rc = strlen(ldd->ldd_svname);
+ if (strcmp(ldd->ldd_svname, "MGS") != 0) {
+ if (rc < 8) {
+ fprintf(stderr, "%s: invalid name '%s'\n",
+ progname, ldd->ldd_svname);
+ return EINVAL;
+ } else if (ldd->ldd_svname[rc - 8] == ':') {
+ ldd->ldd_svname[rc - 8] = '-';
+ ldd->ldd_flags |= LDD_F_VIRGIN;
+ } else if (ldd->ldd_svname[rc - 8] == '=') {
+ ldd->ldd_svname[rc - 8] = '-';
+ ldd->ldd_flags |= LDD_F_WRITECONF;
+ }
+ }
+ /* backend osd type */
+ rc = append_option(options, options_len, "osd=",
+ mt_type(ldd->ldd_mount_type));
+ if (rc != 0)
+ return rc;
+
+ rc = append_option(options, options_len, ldd->ldd_mount_opts, NULL);
+ if (rc != 0)
+ return rc;
+
+ if (!mop->mo_have_mgsnid) {
+ /* Only use disk data if mount -o mgsnode=nid wasn't
+ * specified */
+ if (ldd->ldd_flags & LDD_F_SV_TYPE_MGS) {
+ rc = append_option(options, options_len, "mgs", NULL);
+ if (rc != 0)
+ return rc;
+ mop->mo_have_mgsnid++;
+ } else {
+ if (add_mgsnids(mop, options, ldd->ldd_params,
+ options_len))
+ return E2BIG;
+ }
+ }
+ /* Better have an mgsnid by now */
+ if (!mop->mo_have_mgsnid) {
+ fprintf(stderr, "%s: missing option mgsnode=<nid>\n",
+ progname);
+ return EINVAL;
+ }
+
+ if (ldd->ldd_flags & LDD_F_VIRGIN) {
+ rc = append_option(options, options_len, "virgin", NULL);
+ if (rc != 0)
+ return rc;
+ }
+ if (ldd->ldd_flags & LDD_F_UPDATE) {
+ rc = append_option(options, options_len, "update", NULL);
+ if (rc != 0)
+ return rc;
+ }
+ if (ldd->ldd_flags & LDD_F_WRITECONF) {
+ rc = append_option(options, options_len, "writeconf", NULL);
+ if (rc != 0)
+ return rc;
+ }
+ if (ldd->ldd_flags & LDD_F_NO_PRIMNODE) {
+ rc = append_option(options, options_len, "noprimnode", NULL);
+ if (rc != 0)
+ return rc;
+ }
+
+ /* prefix every lustre parameter with param= so that in-kernel
+ * mount can recognize them properly and send to MGS at registration */
+ start = ldd->ldd_params;
+ while (start && *start != '\0') {
+ while (*start == ' ') start++;
+ if (*start == '\0')
+ break;
+ cur = start;
+ start = strchr(cur, ' ');
+ if (start) {
+ *start = '\0';
+ start++;
+ }
+ rc = append_option(options, options_len, "param=", cur);
+ if (rc != 0)
+ return rc;
+ }
+
+ /* svname must be last option */
+ rc = append_option(options, options_len, "svname=", ldd->ldd_svname);
+
+ return rc;
}
static void set_defaults(struct mount_opts *mop)
mop->mo_have_mgsnid = 0;
mop->mo_md_stripe_cache_size = 16384;
mop->mo_orig_options = "";
+ mop->mo_nosvc = 0;
+ mop->mo_max_sectors_kb = -1;
}
static int parse_opts(int argc, char *const argv[], struct mount_opts *mop)
{"nomtab", 0, 0, 'n'},
{"options", 1, 0, 'o'},
{"verbose", 0, 0, 'v'},
+ {"version", 0, 0, 'V'},
{0, 0, 0, 0}
};
char real_path[PATH_MAX] = {'\0'};
char *ptr;
int opt, rc;
- while ((opt = getopt_long(argc, argv, "fhno:v",
+ while ((opt = getopt_long(argc, argv, "fhno:vV",
long_opt, NULL)) != EOF){
switch (opt) {
case 1:
case 'v':
++verbose;
break;
+ case 'V':
+ ++version;
+ fprintf(stdout, "%s %s %s\n", progname,
+ LUSTRE_VERSION_STRING, WITH_LIBMOUNT);
+ return 0;
default:
fprintf(stderr, "%s: unknown option '%c'\n",
progname, opt);
* symbolic link for instance
*/
if (realpath(mop->mo_usource, real_path) != NULL) {
- mop->mo_usource = strdup(real_path);
-
ptr = strrchr(real_path, '/');
if (ptr && strncmp(ptr, "/dm-", 4) == 0 && isdigit(*(ptr + 4))) {
snprintf(path, sizeof(path), "/sys/block/%s/dm/name", ptr+1);
fclose(f);
}
}
+ mop->mo_usource = strdup(real_path);
}
- mop->mo_source = convert_hostnames(mop->mo_usource);
- if (!mop->mo_source) {
- usage(stderr);
+ ptr = strstr(mop->mo_usource, ":/");
+ if (ptr != NULL) {
+ mop->mo_source = convert_hostnames(mop->mo_usource);
+ if (!mop->mo_source)
+ usage(stderr);
+ } else {
+ mop->mo_source = strdup(mop->mo_usource);
}
if (realpath(argv[optind + 1], mop->mo_target) == NULL) {
return 0;
}
+/* change label from <fsname>:<index> to
+ * <fsname>-<index> to indicate the device has
+ * been registered. only if the label is
+ * supposed to be changed and target service
+ * is supposed to start */
+static void label_lustre(struct mount_opts *mop)
+{
+ if (mop->mo_nosvc)
+ return;
+
+ if (mop->mo_ldd.ldd_flags & (LDD_F_VIRGIN | LDD_F_WRITECONF)) {
+ (void)osd_label_lustre(mop);
+ } else {
+ struct lustre_disk_data ldd;
+ int rc;
+
+ /* device label could be changed after journal recovery,
+ * it should also be relabeled for mount has succeeded. */
+ memset(&ldd, 0, sizeof(ldd));
+ rc = osd_read_ldd(mop->mo_source, &ldd);
+ if (rc == 0) {
+ rc = strlen(ldd.ldd_svname);
+ if (rc >= 8 && ldd.ldd_svname[rc - 8] != '-')
+ (void)osd_label_lustre(mop);
+ }
+ }
+}
+
int main(int argc, char *const argv[])
{
struct mount_opts mop;
char *options;
- int i, rc, flags;
+ int i, flags;
+ int rc;
+ bool client;
+ size_t maxopt_len;
+ size_t g_pagesize;
progname = strrchr(argv[0], '/');
progname = progname ? progname + 1 : argv[0];
set_defaults(&mop);
- rc = osd_init();
- if (rc)
+ g_pagesize = sysconf(_SC_PAGESIZE);
+ if (g_pagesize == -1) {
+ rc = errno;
+ printf("error: %d failed to get page size.\n", rc);
return rc;
+ }
+ maxopt_len = MIN(g_pagesize, 64 * 1024);
rc = parse_opts(argc, argv, &mop);
- if (rc)
+ if (rc || version)
return rc;
- if (verbose) {
- for (i = 0; i < argc; i++)
- printf("arg[%d] = %s\n", i, argv[i]);
+ if (verbose) {
+ for (i = 0; i < argc; i++)
+ printf("arg[%d] = %s\n", i, argv[i]);
printf("source = %s (%s), target = %s\n", mop.mo_usource,
mop.mo_source, mop.mo_target);
printf("options = %s\n", mop.mo_orig_options);
- }
+ }
+
+ options = malloc(maxopt_len);
+ if (options == NULL) {
+ fprintf(stderr, "can't allocate memory for options\n");
+ rc = ENOMEM;
+ goto out_mo_source;
+ }
+
+ if (strlen(mop.mo_orig_options) >= maxopt_len) {
+ fprintf(stderr, "error: mount options too long\n");
+ rc = E2BIG;
+ goto out_options;
+ }
- options = malloc(MAXOPT);
- if (options == NULL) {
- fprintf(stderr, "can't allocate memory for options\n");
- return -1;
- }
strcpy(options, mop.mo_orig_options);
- rc = parse_options(&mop, options, &flags);
+ rc = parse_options(&mop, options, &flags, maxopt_len);
if (rc) {
fprintf(stderr, "%s: can't parse options: %s\n",
progname, options);
- return(EINVAL);
+ rc = EINVAL;
+ goto out_options;
}
if (!mop.mo_force) {
fprintf(stderr, "%s: according to %s %s is "
"already mounted on %s\n", progname, MOUNTED,
mop.mo_usource, mop.mo_target);
- return(EEXIST);
+ rc = EEXIST;
+ goto out_options;
}
if (!rc && (flags & MS_REMOUNT)) {
fprintf(stderr, "%s: according to %s %s is "
"not already mounted on %s\n", progname, MOUNTED,
mop.mo_usource, mop.mo_target);
- return(ENOENT);
+ rc = ENOENT;
+ goto out_options;
}
}
if (flags & MS_REMOUNT)
mop.mo_nomtab++;
rc = access(mop.mo_target, F_OK);
- if (rc) {
- rc = errno;
+ if (rc) {
+ rc = errno;
fprintf(stderr, "%s: %s inaccessible: %s\n", progname,
mop.mo_target, strerror(errno));
- return rc;
- }
+ goto out_options;
+ }
- if (!strstr(mop.mo_usource, ":/")) {
- rc = parse_ldd(mop.mo_source, &mop, options);
+ client = (strstr(mop.mo_usource, ":/") != NULL);
+ if (!client) {
+ rc = osd_init();
if (rc)
- return rc;
+ goto out_options;
+
+ rc = parse_ldd(mop.mo_source, &mop, options, maxopt_len);
+ if (rc)
+ goto out_osd;
}
- /* In Linux 2.4, the target device doesn't get passed to any of our
- functions. So we'll stick it on the end of the options. */
- append_option(options, "device=");
- strcat(options, mop.mo_source);
+ /* In Linux 2.4, the target device doesn't get passed to any of our
+ functions. So we'll stick it on the end of the options. */
+ rc = append_option(options, maxopt_len, "device=", mop.mo_source);
+ if (rc != 0)
+ goto out_osd;
- if (verbose)
- printf("mounting device %s at %s, flags=%#x options=%s\n",
+ if (verbose)
+ printf("mounting device %s at %s, flags=%#x options=%s\n",
mop.mo_source, mop.mo_target, flags, options);
- if (!strstr(mop.mo_usource, ":/") && set_blockdev_tunables(mop.mo_source, 1)) {
- if (verbose)
- fprintf(stderr, "%s: unable to set tunables for %s"
- " (may cause reduced IO performance)\n",
- argv[0], mop.mo_source);
- }
+ if (!client && osd_tune_lustre(mop.mo_source, &mop)) {
+ if (verbose)
+ fprintf(stderr, "%s: unable to set tunables for %s"
+ " (may cause reduced IO performance)\n",
+ argv[0], mop.mo_source);
+ }
+
+#ifdef HAVE_GSS
+ if (mop.mo_skpath[0] != '\0') {
+ /* Treat shared key failures as fatal */
+ rc = load_shared_keys(&mop);
+ if (rc) {
+ fprintf(stderr,
+ "%s: Error loading shared keys: %s\n",
+ progname, strerror(rc));
+ goto out_osd;
+ }
+ }
+#endif /* HAVE_GSS */
if (!mop.mo_fake) {
- /* flags and target get to lustre_get_sb, but not
- lustre_fill_super. Lustre ignores the flags, but mount
- does not. */
- for (i = 0, rc = -EAGAIN; i <= mop.mo_retry && rc != 0; i++) {
+ /* flags and target get to lustre_get_sb(), but not
+ * lustre_fill_super(). Lustre ignores the flags, but mount
+ * does not. */
+ for (i = 0, rc = -EAGAIN; i <= mop.mo_retry && rc != 0; i++) {
rc = mount(mop.mo_source, mop.mo_target, "lustre",
flags, (void *)options);
- if (rc) {
+ if (rc == 0) {
+ if (!client)
+ label_lustre(&mop);
+ } else {
if (verbose) {
fprintf(stderr, "%s: mount %s at %s "
"failed: %s retries left: "
mop.mo_retry - i);
}
- if (mop.mo_retry) {
- sleep(1 << max((i/2), 5));
- }
- else {
- rc = errno;
- }
- }
- }
- }
+ if (mop.mo_retry) {
+ int limit = i/2 > 5 ? i/2 : 5;
+
+ sleep(1 << limit);
+ } else {
+ rc = errno;
+ }
+ }
+ }
+ }
if (rc) {
char *cli;
fprintf(stderr, "%s: mount %s at %s failed: %s\n", progname,
mop.mo_usource, mop.mo_target, strerror(errno));
+ if (errno == EBUSY)
+ fprintf(stderr, "Is the backend filesystem mounted?\n"
+ "Check /etc/mtab and /proc/mounts\n");
if (errno == ENODEV)
fprintf(stderr, "Are the lustre modules loaded?\n"
"Check /etc/modprobe.conf and "
rc = WEXITSTATUS(ret);
}
- } else if (!mop.mo_nomtab) {
- rc = update_mtab_entry(mop.mo_usource, mop.mo_target, "lustre",
- mop.mo_orig_options, 0,0,0);
- }
+ } else {
+ /* Deal with utab just for client. Note that we ignore
+ * the return value here since it is not worth to fail
+ * mount by prevent some rare cases */
+ if (strstr(mop.mo_usource, ":/") != NULL)
+ update_utab_entry(&mop);
+ if (!mop.mo_nomtab) {
+ rc = update_mtab_entry(mop.mo_usource, mop.mo_target,
+ "lustre", mop.mo_orig_options,
+ 0, 0, 0);
+ }
+ }
+
+out_osd:
+ if (!client)
+ osd_fini();
+out_options:
free(options);
+
+out_mo_source:
/* mo_usource should be freed, but we can rely on the kernel */
free(mop.mo_source);
-
- osd_fini();
-
- return rc;
+ return rc;
}