Whamcloud - gitweb
LU-1581 utils: osd_make_lustre() wrapper
[fs/lustre-release.git] / lustre / utils / mount_lustre.c
index 28e6114..6d28857 100644 (file)
@@ -1,6 +1,4 @@
-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
- *
+/*
  * GPL HEADER START
  *
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  * GPL HEADER END
  */
 /*
- * Copyright  2008 Sun Microsystems, Inc. All rights reserved
+ * 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/
@@ -48,9 +48,6 @@
 #include <fcntl.h>
 #include <errno.h>
 #include <string.h>
-#include <sys/mount.h>
-#include <mntent.h>
-#include <getopt.h>
 #include "obdctl.h"
 #include <lustre_ver.h>
 #include <glob.h>
@@ -68,7 +65,7 @@ int          nomtab = 0;
 int          fake = 0;
 int          force = 0;
 int          retry = 0;
-int          md_stripe_cache_size = 2048;
+int          md_stripe_cache_size = 16384;
 char         *progname = NULL;
 
 void usage(FILE *out)
@@ -102,61 +99,6 @@ void usage(FILE *out)
         exit((out != stdout) ? EINVAL : 0);
 }
 
-static int check_mtab_entry(char *spec1, char *spec2, 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, spec1) == 0 ||
-                     strcmp(mnt->mnt_fsname, spec2) == 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)
@@ -233,6 +175,9 @@ static const struct opt_map opt_map[] = {
   { "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 */
@@ -304,7 +249,7 @@ int parse_options(char *orig_options, int *flagp)
                                 retry = 0;
                 } else if (val && strncmp(arg, "mgssec", 6) == 0) {
                         append_option(options, opt);
-                } else if (strncmp(opt, "force", 5) == 0) {
+                } else if (strcmp(opt, "force") == 0) {
                         //XXX special check for 'force' option
                         ++force;
                         printf("force: %d\n", force);
@@ -313,6 +258,12 @@ int parse_options(char *orig_options, int *flagp)
                         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;
@@ -353,12 +304,12 @@ 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_blockdev_tunables(char *source)
+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;
+        char *savept = NULL, *dev;
         char *ret_path;
         char buf[PATH_MAX] = {'\0'}, path[PATH_MAX] = {'\0'};
         char real_path[PATH_MAX] = {'\0'};
@@ -390,12 +341,16 @@ int set_blockdev_tunables(char *source)
          * 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 = real_path + strlen(real_path);
-        while (--dev > real_path && isdigit(*dev))
-                *dev = 0;
 
-        if (strncmp(real_path, "/dev/md_", 8) == 0)
-                *dev = 0;
+        /* 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) {
@@ -412,6 +367,7 @@ int set_blockdev_tunables(char *source)
                 if (verbose)
                         fprintf(stderr, "warning: failed to read entries under "
                                 "/sys/block\n");
+                globfree(&glob_info);
                 return rc;
         }
 
@@ -453,7 +409,7 @@ set_params:
                         if (verbose)
                                 fprintf(stderr, "warning: opening %s: %s\n",
                                         real_path, strerror(errno));
-                        return rc;
+                        return 0;
                 }
 
                 if (atoi(buf) >= md_stripe_cache_size)
@@ -477,24 +433,70 @@ set_params:
                 if (verbose)
                         fprintf(stderr, "warning: opening %s: %s\n",
                                 real_path, strerror(errno));
-                return rc;
+                /* No MAX_HW_SECTORS_KB_PATH isn't necessary an
+                 * error for some device. */
+                rc = 0;
         }
 
         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 && verbose)
-                        fprintf(stderr, "warning: writing to %s: %s\n",
-                                real_path, strerror(errno));
+                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;
 }
 
 int main(int argc, char *const argv[])
 {
         char default_options[] = "";
-        char *usource, *source, *target, *ptr;
+        char *usource, *source, *ptr;
+        char target[PATH_MAX] = {'\0'};
+        char real_path[PATH_MAX] = {'\0'};
+        char path[256], name[256];
+        FILE *f;
+        size_t sz;
         char *options, *optcopy, *orig_options = default_options;
         int i, nargs = 3, opt, rc, flags, optlen;
         static struct option long_opt[] = {
@@ -557,16 +559,37 @@ int main(int argc, char *const argv[])
                 usage(stderr);
         }
 
+        /**
+         * Try to get the real path to the device, in case it is a
+         * symbolic link for instance
+         */
+        if (realpath(usource, real_path) != NULL) {
+                usource = 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);
+                        }
+                }
+        }
+
         source = convert_hostnames(usource);
         if (!source) {
                 usage(stderr);
         }
 
-        target = argv[optind + 1];
-        ptr = target + strlen(target) - 1;
-        while ((ptr > target) && (*ptr == '/')) {
-                *ptr = 0;
-                ptr--;
+        if (realpath(argv[optind + 1], target) == NULL) {
+                rc = errno;
+                fprintf(stderr, "warning: %s: cannot resolve: %s\n",
+                        argv[optind + 1], strerror(errno));
+                return rc;
         }
 
         if (verbose) {
@@ -634,15 +657,13 @@ int main(int argc, char *const argv[])
                 printf("mounting device %s at %s, flags=%#x options=%s\n",
                        source, target, flags, optcopy);
 
-        if (!strstr(usource, ":/") && set_blockdev_tunables(source)) {
+        if (!strstr(usource, ":/") && set_blockdev_tunables(source, 1)) {
                 if (verbose)
                         fprintf(stderr, "%s: unable to set tunables for %s"
                                 " (may cause reduced IO performance)\n",
                                 argv[0], source);
         }
 
-        register_service_tags(usource, source, target);
-
         if (!fake) {
                 /* flags and target get to lustre_get_sb, but not
                    lustre_fill_super.  Lustre ignores the flags, but mount
@@ -684,9 +705,8 @@ int main(int argc, char *const argv[])
                         usource, 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)