X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Futils%2Fmount_lustre.c;h=d5d897cc9223d7b22405c2a0e8745e71e1ad5df4;hb=09bf54263496fed984c4f150b0a3bafeb548d3f7;hp=28e6114de8f5dff98219a5f449c778d1d4a0b5fe;hpb=cf43f7782aad3229b2f281980fc6023fc76067c4;p=fs%2Flustre-release.git diff --git a/lustre/utils/mount_lustre.c b/lustre/utils/mount_lustre.c index 28e6114..d5d897c 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 @@ -64,11 +61,7 @@ #define MAX_RETRIES 99 int verbose = 0; -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 +95,6 @@ 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 +171,9 @@ 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 */ @@ -275,7 +216,7 @@ static void append_option(char *options, const char *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, *arg, *val; @@ -297,22 +238,28 @@ int parse_options(char *orig_options, int *flagp) 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; + 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 (strncmp(opt, "force", 5) == 0) { + } else if (strcmp(opt, "force") == 0) { //XXX special check for 'force' option - ++force; - printf("force: %d\n", force); + ++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 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 +300,12 @@ 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 +337,16 @@ 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 +363,7 @@ 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 +405,7 @@ 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) @@ -477,148 +429,237 @@ 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[]) +static void set_defaults(struct mount_opts *mop) { - 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; - } - } + 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 = ""; +} - if (optind + 2 > argc) { - fprintf(stderr, "%s: too few arguments\n", progname); - usage(stderr); - } +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 "\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; +} - usource = argv[optind]; - if (!usource) { - usage(stderr); - } +int main(int argc, char *const argv[]) +{ + struct mount_opts mop; + char *options, *optcopy; + int i, rc, flags, optlen; - source = convert_hostnames(usource); - if (!source) { - usage(stderr); - } + progname = strrchr(argv[0], '/'); + progname = progname ? progname + 1 : argv[0]; - target = argv[optind + 1]; - ptr = target + strlen(target) - 1; - while ((ptr > target) && (*ptr == '/')) { - *ptr = 0; - ptr--; - } + set_defaults(&mop); + + 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); + options = malloc(strlen(mop.mo_orig_options) + 1); if (options == NULL) { fprintf(stderr, "can't allocate memory for options\n"); return -1; } - strcpy(options, orig_options); - rc = parse_options(options, &flags); + 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, source, 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; } /* 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; + optlen = strlen(options) + strlen(",device=") + strlen(mop.mo_source) + 1; optcopy = malloc(optlen); if (optcopy == NULL) { fprintf(stderr, "can't allocate memory to optcopy\n"); @@ -628,38 +669,37 @@ int main(int argc, char *const argv[]) if (*optcopy) strcat(optcopy, ","); strcat(optcopy, "device="); - strcat(optcopy, source); + strcat(optcopy, 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, optcopy); - if (!strstr(usource, ":/") && set_blockdev_tunables(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], source); + argv[0], mop.mo_source); } - register_service_tags(usource, source, target); - - 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. */ - for (i = 0, rc = -EAGAIN; i <= retry && rc != 0; i++) { - 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 *)optcopy); if (rc) { if (verbose) { fprintf(stderr, "%s: mount %s at %s " "failed: %s retries left: " "%d\n", basename(progname), - usource, target, - strerror(errno), retry-i); + mop.mo_usource, mop.mo_target, + strerror(errno), + mop.mo_retry - i); } - if (retry) { + if (mop.mo_retry) { sleep(1 << max((i/2), 5)); } else { @@ -674,19 +714,18 @@ int main(int argc, char *const argv[]) 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) @@ -700,16 +739,16 @@ 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) @@ -720,22 +759,23 @@ 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]; int ret; - sprintf(cmd, "/sbin/losetup -d %s", usource); + 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); + /* mo_usource should be freed, but we can rely on the kernel */ + free(mop.mo_source); return rc; }