* 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.
*/
/*
#include <errno.h>
#include <string.h>
#include <sys/mount.h>
+#include <linux/fs.h>
#include <mntent.h>
#include <getopt.h>
#include "obdctl.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 MAX_RETRIES 99
int verbose = 0;
int fake = 0;
int force = 0;
int retry = 0;
+int md_stripe_cache_size = 16384;
char *progname = NULL;
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);
}
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, ",:");
{ "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 */
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)
* manner */
arg = opt;
val = strchr(opt, '=');
- if (val != NULL && strncmp(arg, "retry", 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 (parse_one_option(opt, flagp) == 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;
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;
}
/* 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)
{
glob_t glob_info;
struct stat stat_buf;
char *chk_major, *chk_minor;
- char *savept, *dev, *s2 = 0;
+ char *savept, *dev;
char *ret_path;
char buf[PATH_MAX] = {'\0'}, path[PATH_MAX] = {'\0'};
char real_path[PATH_MAX] = {'\0'};
return -EINVAL;
}
- src_len = sizeof(real_path);
-
if (strncmp(real_path, "/dev/loop", 9) == 0)
return 0;
- if ((real_path[0] != '/') && ((s2 = strpbrk(real_path, ",:")) != NULL))
+ if ((real_path[0] != '/') && (strpbrk(real_path, ",:") != NULL))
return 0;
- dev = real_path + src_len - 1;
- while (dev > real_path && (*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 (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 = real_path + src_len - 1;
- while (dev > real_path) {
- if (isdigit(*dev))
- *dev = 0;
- dev--;
- }
+ dev = real_path + strlen(real_path);
+ while (--dev > real_path && isdigit(*dev))
+ *dev = 0;
- rc = stat(dev, &stat_buf);
+ 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), dev);
+ strerror(errno), real_path);
return rc;
}
if (verbose)
fprintf(stderr,"warning: device %s does not match any "
"entry under /sys/block\n", real_path);
- rc = -EINVAL;
- goto out;
+ 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));
- goto out;
+ real_path, strerror(errno));
+ return rc;
}
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);
+ 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",
- path, strerror(errno));
+ real_path, strerror(errno));
}
-
-out:
- 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[] = {
}
usource = argv[optind];
- source = convert_hostnames(usource);
- target = argv[optind + 1];
- ptr = target + strlen(target) - 1;
- while ((ptr > target) && (*ptr == '/')) {
- *ptr = 0;
- ptr--;
+ if (!usource) {
+ 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);
+ }
+ }
}
- if (!usource || !source) {
+ source = convert_hostnames(usource);
+ if (!source) {
usage(stderr);
}
+ 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) {
for (i = 0; i < argc; i++)
printf("arg[%d] = %s\n", i, argv[i]);
}
options = malloc(strlen(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);
if (rc) {
functions. So we'll stick it on the end of the options. */
optlen = strlen(options) + strlen(",device=") + strlen(source) + 1;
optcopy = malloc(optlen);
+ if (optcopy == NULL) {
+ fprintf(stderr, "can't allocate memory to optcopy\n");
+ return -1;
+ }
strcpy(optcopy, options);
if (*optcopy)
strcat(optcopy, ",");
printf("mounting device %s at %s, flags=%#x options=%s\n",
source, target, flags, optcopy);
- if (!strstr(usource, ":/") && set_tunables(source, strlen(source)) &&
- verbose)
- fprintf(stderr, "%s: unable to set tunables for %s"
+ if (!strstr(usource, ":/") && set_blockdev_tunables(source)) {
+ 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);
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)
/* 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) {
git://git.whamcloud.com / fs / lustre-release.git / blobdiff