Whamcloud - gitweb
LU-1581 utils: mount to recognize backend type
[fs/lustre-release.git] / lustre / utils / mount_lustre.c
index 74d2910..e6b38ef 100644 (file)
@@ -1,51 +1,69 @@
-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
+/*
+ * GPL HEADER START
  *
- *  Copyright (C) 2002 Cluster File Systems, Inc.
- *   Author: Robert Read <rread@clusterfs.com>
- *   Author: Nathan Rutman <nathan@clusterfs.com>
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
- *   This file is part of Lustre, http://www.lustre.org.
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 only,
+ * as published by the Free Software Foundation.
  *
- *   Lustre is free software; you can redistribute it and/or
- *   modify it under the terms of version 2 of the GNU General Public
- *   License as published by the Free Software Foundation.
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License version 2 for more details (a copy is included
+ * in the LICENSE file that accompanied this code).
  *
- *   Lustre is distributed in the hope that it will be useful,
- *   but WITHOUT ANY WARRANTY; without even the implied warranty of
- *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
- *   GNU General Public License for more details.
+ * 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
  *
- *   You should have received a copy of the GNU General Public License
- *   along with Lustre; if not, write to the Free Software
- *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ * 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.
  *
+ * 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.
+ */
+/*
+ * This file is part of Lustre, http://www.lustre.org/
+ * Lustre is a trademark of Sun Microsystems, Inc.
+ *
+ * lustre/utils/mount_lustre.c
+ *
+ * Author: Robert Read <rread@clusterfs.com>
+ * Author: Nathan Rutman <nathan@clusterfs.com>
  */
 
-
+#ifndef _GNU_SOURCE
 #define _GNU_SOURCE
+#endif
 #include <stdlib.h>
 #include <stdio.h>
 #include <unistd.h>
 #include <fcntl.h>
 #include <errno.h>
 #include <string.h>
-#include <sys/mount.h>
-#include <mntent.h>
-#include <getopt.h>
-#include <sys/utsname.h>
 #include "obdctl.h"
 #include <lustre_ver.h>
 #include <glob.h>
+#include <ctype.h>
+#include <limits.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
+#define MAX_RETRIES 99
 
 int          verbose = 0;
-int          nomtab = 0;
-int          fake = 0;
-int          force = 0;
-static char *progname = NULL;
+int          md_stripe_cache_size = 16384;
+char         *progname = NULL;
 
 void usage(FILE *out)
 {
@@ -60,73 +78,24 @@ void usage(FILE *out)
                 "\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--force: force mount even if already in /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);
 }
 
-static int check_mtab_entry(char *spec, char *mtpt, char *type)
-{
-        FILE *fp;
-        struct mntent *mnt;
-
-        fp = setmntent(MOUNTED, "r");
-        if (fp == NULL)
-                return(0);
-
-        while ((mnt = getmntent(fp)) != NULL) {
-                if (strcmp(mnt->mnt_fsname, spec) == 0 &&
-                        strcmp(mnt->mnt_dir, mtpt) == 0 &&
-                        strcmp(mnt->mnt_type, type) == 0) {
-                        endmntent(fp);
-                        return(EEXIST);
-                }
-        }
-        endmntent(fp);
-
-        return(0);
-}
-
-static int
-update_mtab_entry(char *spec, char *mtpt, char *type, char *opts,
-                  int flags, int freq, int pass)
-{
-        FILE *fp;
-        struct mntent mnt;
-        int rc = 0;
-
-        mnt.mnt_fsname = spec;
-        mnt.mnt_dir = mtpt;
-        mnt.mnt_type = type;
-        mnt.mnt_opts = opts ? opts : "";
-        mnt.mnt_freq = freq;
-        mnt.mnt_passno = pass;
-
-        fp = setmntent(MOUNTED, "a+");
-        if (fp == NULL) {
-                fprintf(stderr, "%s: setmntent(%s): %s:",
-                        progname, MOUNTED, strerror (errno));
-                rc = 16;
-        } else {
-                if ((addmntent(fp, &mnt)) == 1) {
-                        fprintf(stderr, "%s: addmntent: %s:",
-                                progname, strerror (errno));
-                        rc = 16;
-                }
-                endmntent(fp);
-        }
-
-        return rc;
-}
-
 /* Get rid of symbolic hostnames for tcp, since kernel can't do lookups */
 #define MAXNIDSTR 1024
 static char *convert_hostnames(char *s1)
@@ -137,6 +106,11 @@ static char *convert_hostnames(char *s1)
         lnet_nid_t nid;
 
         converted = malloc(left);
+        if (converted == NULL) {
+                fprintf(stderr, "out of memory: needed %d bytes\n",
+                        MAXNIDSTR);
+                return NULL;
+        }
         c = converted;
         while ((left > 0) && (*s1 != '/')) {
                 s2 = strpbrk(s1, ",:");
@@ -186,13 +160,28 @@ static const struct opt_map opt_map[] = {
   { "nosuid",   0, MS_NOSUID },      /* don't honor suid executables */
   { "dev",      1, MS_NODEV  },      /* interpret device files  */
   { "nodev",    0, MS_NODEV  },      /* don't interpret devices */
+  { "sync",     0, MS_SYNCHRONOUS},  /* synchronous I/O */
   { "async",    1, MS_SYNCHRONOUS},  /* asynchronous I/O */
+  { "atime",    1, MS_NOATIME  },    /* set file access time on read */
+  { "noatime",  0, MS_NOATIME  },    /* do not set file access time on read */
+#ifdef MS_NODIRATIME
+  { "diratime", 1, MS_NODIRATIME },  /* set file access time on read */
+  { "nodiratime",0,MS_NODIRATIME },  /* do not set file access time on read */
+#endif
+#ifdef MS_RELATIME
+  { "relatime", 0, MS_RELATIME },  /* set file access time on read */
+  { "norelatime",1,MS_RELATIME },  /* do not set file access time on read */
+#endif
+#ifdef MS_STRICTATIME
+  { "strictatime",0,MS_STRICTATIME },  /* update access time strictly */
+#endif
   { "auto",     0, 0         },      /* Can be mounted using -a */
   { "noauto",   0, 0         },      /* Can only be mounted explicitly */
   { "nousers",  1, 0         },      /* Forbid ordinary user to mount */
   { "nouser",   1, 0         },      /* Forbid ordinary user to mount */
   { "noowner",  1, 0         },      /* Device owner has no special privs */
   { "_netdev",  0, 0         },      /* Device accessible only via network */
+  { "loop",     0, 0         },
   { NULL,       0, 0         }
 };
 /****************************************************************************/
@@ -219,11 +208,18 @@ static int parse_one_option(const char *check, int *flagp)
         return 0;
 }
 
+static void append_option(char *options, const char *one)
+{
+        if (*options)
+                strcat(options, ",");
+        strcat(options, one);
+}
+
 /* Replace options with subset of Lustre-specific options, and
    fill in mount flags */
-int parse_options(char *orig_options, int *flagp)
+int parse_options(struct mount_opts *mop, char *orig_options, int *flagp)
 {
-        char *options, *opt, *nextopt;
+        char *options, *opt, *nextopt, *arg, *val;
 
         options = calloc(strlen(orig_options) + 1, 1);
         *flagp = 0;
@@ -232,13 +228,39 @@ int parse_options(char *orig_options, int *flagp)
                 if (!*opt)
                         /* empty option */
                         continue;
-                if (parse_one_option(opt, flagp) == 0) {
+
+                /* 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) {
+                       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
+                       ++mop->mo_force;
+                       printf("force: %d\n", mop->mo_force);
+                } else if (parse_one_option(opt, flagp) == 0) {
                         /* pass this on as an option */
-                        if (*options)
-                                strcat(options, ",");
-                        strcat(options, opt);
+                        append_option(options, opt);
                 }
         }
+#ifdef MS_STRICTATIME
+                /* 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);
         free(options);
         return 0;
@@ -253,7 +275,12 @@ int read_file(char *path, char *buf, int size)
         if (fd == NULL)
                 return errno;
 
-        fgets(buf, size, fd);
+        /* 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;
 }
@@ -274,62 +301,59 @@ int write_file(char *path, char *buf)
 /* 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_tunables(char *source, int src_len)
+int set_blockdev_tunables(char *source, int fan_out)
 {
-        glob_t glob_info;
+        glob_t glob_info = { 0 };
         struct stat stat_buf;
         char *chk_major, *chk_minor;
-        char *savept, *dev, *s2 = 0;
-        char buf[PATH_MAX], path[PATH_MAX];
+        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;
 
-        if (strncmp(source, "/dev/loop", 9) == 0)
-                return 0;
+        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 ((*source != '/') && ((s2 = strpbrk(source, ",:")) != NULL))
+        if (strncmp(real_path, "/dev/loop", 9) == 0)
                 return 0;
 
-        dev = source + src_len - 1;
-        while (dev > source && (*dev != '/')) {
-                if (isdigit(*dev))
-                        *dev = 0;
-                dev--;
-        }
-        snprintf(path, sizeof(path), "/sys/block%s/%s", dev,
-                 MAX_HW_SECTORS_KB_PATH);
-        rc = read_file(path, buf, sizeof(buf));
-        if (rc == 0 && (strlen(buf) - 1) > 0) {
-                snprintf(path, sizeof(path), "/sys/block%s/%s", dev,
-                         MAX_SECTORS_KB_PATH);
-                rc = write_file(path, buf);
-                if (rc && verbose)
-                        fprintf(stderr, "warning: opening %s: %s\n",
-                                path, strerror(errno));
-                return rc;
-        }
+        if ((real_path[0] != '/') && (strpbrk(real_path, ",:") != NULL))
+                return 0;
 
-        if (rc != ENOENT)
-                return rc;
+        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 */
-        dev = source + src_len - 1;
-        while (dev > source) {
-                if (isdigit(*dev))
+
+        /* 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;
-                dev--;
         }
 
-        rc = stat(dev, &stat_buf);
+        rc = stat(real_path, &stat_buf);
         if (rc) {
-                fprintf(stderr, "warning: %s, device %s stat failed\n",
-                        strerror(errno), dev);
+                if (verbose)
+                        fprintf(stderr, "warning: %s, device %s stat failed\n",
+                                strerror(errno), real_path);
                 return rc;
         }
 
@@ -337,8 +361,10 @@ int set_tunables(char *source, int src_len)
         minor = minor(stat_buf.st_rdev);
         rc = glob("/sys/block/*", GLOB_NOSORT, NULL, &glob_info);
         if (rc) {
-                fprintf(stderr, "warning: failed to read entries under "
-                        "/sys/block\n");
+                if (verbose)
+                        fprintf(stderr, "warning: failed to read entries under "
+                                "/sys/block\n");
+                globfree(&glob_info);
                 return rc;
         }
 
@@ -361,190 +387,367 @@ int set_tunables(char *source, int src_len)
         if (i == glob_info.gl_pathc) {
                 if (verbose)
                         fprintf(stderr,"warning: device %s does not match any "
-                                "entry under /sys/block\n", source);
+                                "entry under /sys/block\n", real_path);
+                globfree(&glob_info);
                 return -EINVAL;
         }
 
-        snprintf(path, sizeof(path), "%s/%s", glob_info.gl_pathv[i],
+        /* 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(path, buf, sizeof(buf));
+        rc = read_file(real_path, buf, sizeof(buf));
         if (rc) {
                 if (verbose)
                         fprintf(stderr, "warning: opening %s: %s\n",
-                                path, strerror(errno));
-                return rc;
+                                real_path, strerror(errno));
+                /* No MAX_HW_SECTORS_KB_PATH isn't necessary an
+                 * error for some device. */
+                rc = 0;
         }
 
         if (strlen(buf) - 1 > 0) {
-                snprintf(path, sizeof(path), "%s/%s",
-                         glob_info.gl_pathv[i], MAX_SECTORS_KB_PATH);
-                rc = write_file(path, buf);
-                if (rc && verbose)
-                        fprintf(stderr, "warning: writing to %s: %s\n",
-                                path, strerror(errno));
+                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;
+                }
         }
+
+        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);
+                }
+
+                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);
+        }
+
         return rc;
 }
 
+static int parse_ldd(char *source, struct mount_opts *mop, char *options)
+{
+       struct lustre_disk_data *ldd = &mop->mo_ldd;
+       int rc;
+
+       rc = osd_is_lustre(source, &ldd->ldd_mount_type);
+       if (rc == 0) {
+               fprintf(stderr, "%s: %s has not been formatted with mkfs.lustre"
+                       " or the backend filesystem type is not supported by "
+                       "this tool\n", progname, source);
+               return ENODEV;
+       }
+
+       /* backend osd type */
+       append_option(options, "osd=");
+       strcat(options, mt_type(ldd->ldd_mount_type));
+
+       return 0;
+}
+
+static void set_defaults(struct mount_opts *mop)
+{
+       memset(mop, 0, sizeof(*mop));
+       mop->mo_usource = NULL;
+       mop->mo_source = NULL;
+       mop->mo_nomtab = 0;
+       mop->mo_fake = 0;
+       mop->mo_force = 0;
+       mop->mo_retry = 0;
+       mop->mo_have_mgsnid = 0;
+       mop->mo_md_stripe_cache_size = 16384;
+       mop->mo_orig_options = "";
+}
+
+static int parse_opts(int argc, char *const argv[], struct mount_opts *mop)
+{
+       static struct option long_opt[] = {
+               {"fake", 0, 0, 'f'},
+               {"force", 0, 0, 1},
+               {"help", 0, 0, 'h'},
+               {"nomtab", 0, 0, 'n'},
+               {"options", 1, 0, 'o'},
+               {"verbose", 0, 0, 'v'},
+               {0, 0, 0, 0}
+       };
+       char real_path[PATH_MAX] = {'\0'};
+       FILE *f;
+       char path[256], name[256];
+       size_t sz;
+       char *ptr;
+       int opt, rc;
+
+       while ((opt = getopt_long(argc, argv, "fhno:v",
+                                 long_opt, NULL)) != EOF){
+               switch (opt) {
+               case 1:
+                       ++mop->mo_force;
+                       printf("force: %d\n", mop->mo_force);
+                       break;
+               case 'f':
+                       ++mop->mo_fake;
+                       printf("fake: %d\n", mop->mo_fake);
+                       break;
+               case 'h':
+                       usage(stdout);
+                       break;
+               case 'n':
+                       ++mop->mo_nomtab;
+                       printf("nomtab: %d\n", mop->mo_nomtab);
+                       break;
+               case 'o':
+                       mop->mo_orig_options = optarg;
+                       break;
+               case 'v':
+                       ++verbose;
+                       break;
+               default:
+                       fprintf(stderr, "%s: unknown option '%c'\n",
+                                       progname, opt);
+                       usage(stderr);
+                       break;
+               }
+       }
+
+       if (optind + 2 > argc) {
+               fprintf(stderr, "%s: too few arguments\n", progname);
+               usage(stderr);
+       }
+
+       mop->mo_usource = argv[optind];
+       if (!mop->mo_usource) {
+               usage(stderr);
+       }
+
+       /**
+        * Try to get the real path to the device, in case it is a
+        * 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);
+                       if ((f = fopen(path, "r"))) {
+                               /* read "<name>\n" from sysfs */
+                               if (fgets(name, sizeof(name), f) && (sz = strlen(name)) > 1) {
+                                       name[sz - 1] = '\0';
+                                       snprintf(real_path, sizeof(real_path), "/dev/mapper/%s", name);
+                               }
+                               fclose(f);
+                       }
+               }
+       }
+
+       mop->mo_source = convert_hostnames(mop->mo_usource);
+       if (!mop->mo_source) {
+               usage(stderr);
+       }
+
+       if (realpath(argv[optind + 1], mop->mo_target) == NULL) {
+               rc = errno;
+               fprintf(stderr, "warning: %s: cannot resolve: %s\n",
+                               argv[optind + 1], strerror(errno));
+               return rc;
+       }
+
+       return 0;
+}
+
 int main(int argc, char *const argv[])
 {
-        char default_options[] = "";
-        char *usource, *source, *target, *ptr;
-        char *options, *optcopy, *orig_options = default_options;
-        int i, nargs = 3, opt, rc, flags, optlen;
-        static struct option long_opt[] = {
-                {"fake", 0, 0, 'f'},
-                {"force", 0, 0, 1},
-                {"help", 0, 0, 'h'},
-                {"nomtab", 0, 0, 'n'},
-                {"options", 1, 0, 'o'},
-                {"verbose", 0, 0, 'v'},
-                {0, 0, 0, 0}
-        };
-
-        progname = strrchr(argv[0], '/');
-        progname = progname ? progname + 1 : argv[0];
-
-        while ((opt = getopt_long(argc, argv, "fhno:v",
-                                  long_opt, NULL)) != EOF){
-                switch (opt) {
-                case 1:
-                        ++force;
-                        printf("force: %d\n", force);
-                        nargs++;
-                        break;
-                case 'f':
-                        ++fake;
-                        printf("fake: %d\n", fake);
-                        nargs++;
-                        break;
-                case 'h':
-                        usage(stdout);
-                        break;
-                case 'n':
-                        ++nomtab;
-                        printf("nomtab: %d\n", nomtab);
-                        nargs++;
-                        break;
-                case 'o':
-                        orig_options = optarg;
-                        nargs++;
-                        break;
-                case 'v':
-                        ++verbose;
-                        nargs++;
-                        break;
-                default:
-                        fprintf(stderr, "%s: unknown option '%c'\n",
-                                progname, opt);
-                        usage(stderr);
-                        break;
-                }
-        }
+       struct mount_opts mop;
+       char *options;
+       int i, rc, flags;
 
-        if (optind + 2 > argc) {
-                fprintf(stderr, "%s: too few arguments\n", progname);
-                usage(stderr);
-        }
+       progname = strrchr(argv[0], '/');
+       progname = progname ? progname + 1 : argv[0];
 
-        usource = argv[optind];
-        source = convert_hostnames(usource);
-        target = argv[optind + 1];
-        ptr = target + strlen(target) - 1;
-        while ((ptr > target) && (*ptr == '/')) {
-                *ptr = 0;
-                ptr--;
-        }
+       set_defaults(&mop);
 
-        if (!usource || !source) {
-                usage(stderr);
-        }
+       rc = osd_init();
+       if (rc)
+               return rc;
+
+       rc = parse_opts(argc, argv, &mop);
+       if (rc)
+               return rc;
 
         if (verbose) {
                 for (i = 0; i < argc; i++)
                         printf("arg[%d] = %s\n", i, argv[i]);
-                printf("source = %s (%s), target = %s\n", usource, source, target);
-                printf("options = %s\n", orig_options);
+               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(strlen(orig_options) + 1);
-        strcpy(options, orig_options);
-        rc = parse_options(options, &flags);
+       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);
         if (rc) {
                 fprintf(stderr, "%s: can't parse options: %s\n",
                         progname, options);
                 return(EINVAL);
         }
 
-        if (!force) {
-                rc = check_mtab_entry(usource, target, "lustre");
+       if (!mop.mo_force) {
+               rc = check_mtab_entry(mop.mo_usource, mop.mo_source,
+                                     mop.mo_target, "lustre");
                 if (rc && !(flags & MS_REMOUNT)) {
                         fprintf(stderr, "%s: according to %s %s is "
-                                "already mounted on %s\n",
-                                progname, MOUNTED, usource, target);
+                               "already mounted on %s\n", progname, MOUNTED,
+                               mop.mo_usource, mop.mo_target);
                         return(EEXIST);
                 }
                 if (!rc && (flags & MS_REMOUNT)) {
                         fprintf(stderr, "%s: according to %s %s is "
-                                "not already mounted on %s\n",
-                                progname, MOUNTED, usource, target);
+                               "not already mounted on %s\n", progname, MOUNTED,
+                               mop.mo_usource, mop.mo_target);
                         return(ENOENT);
                 }
         }
         if (flags & MS_REMOUNT)
-                nomtab++;
+               mop.mo_nomtab++;
 
-        rc = access(target, F_OK);
+       rc = access(mop.mo_target, F_OK);
         if (rc) {
                 rc = errno;
-                fprintf(stderr, "%s: %s inaccessible: %s\n", progname, target,
-                        strerror(errno));
+               fprintf(stderr, "%s: %s inaccessible: %s\n", progname,
+                       mop.mo_target, strerror(errno));
                 return rc;
         }
 
+       if (!strstr(mop.mo_usource, ":/")) {
+               rc = parse_ldd(mop.mo_source, &mop, options);
+               if (rc)
+                       return rc;
+       }
+
         /* 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. */
-        optlen = strlen(options) + strlen(",device=") + strlen(source) + 1;
-        optcopy = malloc(optlen);
-        strcpy(optcopy, options);
-        if (*optcopy)
-                strcat(optcopy, ",");
-        strcat(optcopy, "device=");
-        strcat(optcopy, source);
+       append_option(options, "device=");
+       strcat(options, mop.mo_source);
 
         if (verbose)
                 printf("mounting device %s at %s, flags=%#x options=%s\n",
-                       source, target, flags, optcopy);
+                      mop.mo_source, mop.mo_target, flags, options);
 
-        if (set_tunables(source, strlen(source)) && verbose)
-                fprintf(stderr, "%s: unable to set tunables for %s"
-                                " (may cause reduced IO performance)",
-                                argv[0], source);
+       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 (!fake)
+       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. */
-                rc = mount(source, target, "lustre", flags, (void *)optcopy);
+                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 (verbose) {
+                                        fprintf(stderr, "%s: mount %s at %s "
+                                                "failed: %s retries left: "
+                                                "%d\n", basename(progname),
+                                               mop.mo_usource, mop.mo_target,
+                                                strerror(errno),
+                                               mop.mo_retry - i);
+                                }
+
+                                if (mop.mo_retry) {
+                                        sleep(1 << max((i/2), 5));
+                                }
+                                else {
+                                        rc = errno;
+                                }
+                        }
+                }
+        }
 
         if (rc) {
                 char *cli;
 
                 rc = errno;
 
-                cli = strrchr(usource, ':');
+               cli = strrchr(mop.mo_usource, ':');
                 if (cli && (strlen(cli) > 2))
                         cli += 2;
                 else
                         cli = NULL;
 
                 fprintf(stderr, "%s: mount %s at %s failed: %s\n", progname,
-                        usource, target, strerror(errno));
+                       mop.mo_usource, mop.mo_target, strerror(errno));
                 if (errno == ENODEV)
                         fprintf(stderr, "Are the lustre modules loaded?\n"
-                                "Check /etc/modprobe.conf and /proc/filesystems"
-                                "\nNote 'alias lustre llite' should be removed"
-                                " from modprobe.conf\n");
+                                "Check /etc/modprobe.conf and "
+                                "/proc/filesystems\n");
                 if (errno == ENOTBLK)
                         fprintf(stderr, "Do you need -o loop?\n");
                 if (errno == ENOMEDIUM)
@@ -558,16 +761,16 @@ int main(int argc, char *const argv[])
                 }
                 if (errno == EALREADY)
                         fprintf(stderr, "The target service is already running."
-                                " (%s)\n", usource);
+                               " (%s)\n", mop.mo_usource);
                 if (errno == ENXIO)
                         fprintf(stderr, "The target service failed to start "
                                 "(bad config log?) (%s).  "
-                                "See /var/log/messages.\n", usource);
+                               "See /var/log/messages.\n", mop.mo_usource);
                 if (errno == EIO)
                         fprintf(stderr, "Is the MGS running?\n");
                 if (errno == EADDRINUSE)
                         fprintf(stderr, "The target service's index is already "
-                                "in use. (%s)\n", usource);
+                               "in use. (%s)\n", mop.mo_usource);
                 if (errno == EINVAL) {
                         fprintf(stderr, "This may have multiple causes.\n");
                         if (cli)
@@ -578,18 +781,26 @@ int main(int argc, char *const argv[])
                 }
 
                 /* May as well try to clean up loop devs */
-                if (strncmp(usource, "/dev/loop", 9) == 0) {
+               if (strncmp(mop.mo_usource, "/dev/loop", 9) == 0) {
                         char cmd[256];
-                        sprintf(cmd, "/sbin/losetup -d %s", usource);
-                        system(cmd);
+                        int ret;
+                       sprintf(cmd, "/sbin/losetup -d %s", mop.mo_usource);
+                        if ((ret = system(cmd)) < 0)
+                                rc = errno;
+                        else if (ret > 0)
+                                rc = WEXITSTATUS(ret);
                 }
 
-        } else if (!nomtab) {
-                rc = update_mtab_entry(usource, target, "lustre", orig_options,
-                                       0,0,0);
+       } else if (!mop.mo_nomtab) {
+               rc = update_mtab_entry(mop.mo_usource, mop.mo_target, "lustre",
+                                      mop.mo_orig_options, 0,0,0);
         }
 
-        free(optcopy);
-        free(source);
+       free(options);
+       /* mo_usource should be freed, but we can rely on the kernel */
+       free(mop.mo_source);
+
+       osd_fini();
+
         return rc;
 }