X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Futils%2Fmount_lustre.c;h=6d28857fecac52f446adf240143fae3deaa14e8d;hp=bbbe9ef4ba656f7309c9560895808772c921b686;hb=fd4cee4cc545679d2a00c2a20939564e36700631;hpb=8e7439f9d8646cf916248173fdd841b456b26eac diff --git a/lustre/utils/mount_lustre.c b/lustre/utils/mount_lustre.c index bbbe9ef..6d28857 100644 --- a/lustre/utils/mount_lustre.c +++ b/lustre/utils/mount_lustre.c @@ -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. @@ -26,8 +24,10 @@ * 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 #include #include -#include -#include -#include #include "obdctl.h" #include #include @@ -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 */ @@ -266,6 +211,13 @@ 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) @@ -284,27 +236,34 @@ int parse_options(char *orig_options, int *flagp) * manner */ arg = opt; val = strchr(opt, '='); - if (val != NULL) { - if (strncmp(arg, "md_stripe_cache_size", 20) == 0) { - md_stripe_cache_size = atoi(val + 1); - } else if (strncmp(arg, "retry", 5) == 0) { - retry = atoi(val + 1); - if (retry > MAX_RETRIES) - retry = MAX_RETRIES; - else if (retry < 0) - retry = 0; - } - } else if (strncmp(opt, "force", 5) == 0) { + /* 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) { + retry = atoi(val + 1); + if (retry > MAX_RETRIES) + retry = MAX_RETRIES; + else if (retry < 0) + 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 ++force; printf("force: %d\n", 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; @@ -319,7 +278,12 @@ 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; } @@ -340,12 +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'}; @@ -377,12 +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) { @@ -399,6 +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; } @@ -440,7 +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) @@ -464,24 +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[] = { @@ -544,16 +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 "\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) { @@ -621,15 +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 @@ -671,9 +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) @@ -709,8 +742,12 @@ int main(int argc, char *const argv[]) /* May as well try to clean up loop devs */ if (strncmp(usource, "/dev/loop", 9) == 0) { char cmd[256]; + int ret; sprintf(cmd, "/sbin/losetup -d %s", usource); - system(cmd); + if ((ret = system(cmd)) < 0) + rc = errno; + else if (ret > 0) + rc = WEXITSTATUS(ret); } } else if (!nomtab) {