#include <lnet/lnetctl.h>
#include <lustre_ver.h>
+#ifdef HAVE_SELINUX
+#include <selinux/selinux.h>
+#endif
+
#define MAX_HW_SECTORS_KB_PATH "queue/max_hw_sectors_kb"
#define MAX_SECTORS_KB_PATH "queue/max_sectors_kb"
+#define SCHEDULER_PATH "queue/scheduler"
#define STRIPE_CACHE_SIZE "md/stripe_cache_size"
+#define DEFAULT_SCHEDULER "deadline"
+
extern char *progname;
#define L_BLOCK_SIZE 4096
#define DUMMY_FILE_NAME_LEN 25
#define EXT3_DIRENT_SIZE DUMMY_FILE_NAME_LEN
+/*
+ * Concatenate context of the temporary mount point iff selinux is enabled
+ */
+#ifdef HAVE_SELINUX
+static void append_context_for_mount(char *mntpt, struct mkfs_opts *mop)
+{
+ security_context_t fcontext;
+
+ if (getfilecon(mntpt, &fcontext) < 0) {
+ /* Continuing with default behaviour */
+ fprintf(stderr, "%s: Get file context failed : %s\n",
+ progname, strerror(errno));
+ return;
+ }
+
+ if (fcontext != NULL) {
+ strcat(mop->mo_ldd.ldd_mount_opts, ",context=");
+ strcat(mop->mo_ldd.ldd_mount_opts, fcontext);
+ freecon(fcontext);
+ }
+}
+#endif
+
+/* return canonicalized absolute pathname, even if the target file does not
+ * exist, unlike realpath */
+static char *absolute_path(char *devname)
+{
+ char buf[PATH_MAX + 1] = "";
+ char *path;
+ char *ptr;
+ int len;
+
+ path = malloc(sizeof(buf));
+ if (path == NULL)
+ return NULL;
+
+ if (devname[0] != '/') {
+ if (getcwd(buf, sizeof(buf) - 1) == NULL) {
+ free(path);
+ return NULL;
+ }
+ len = snprintf(path, sizeof(buf), "%s/%s", buf, devname);
+ if (len >= sizeof(buf)) {
+ free(path);
+ return NULL;
+ }
+ } else {
+ len = snprintf(path, sizeof(buf), "%s", devname);
+ if (len >= sizeof(buf)) {
+ free(path);
+ return NULL;
+ }
+ }
+
+ /* truncate filename before calling realpath */
+ ptr = strrchr(path, '/');
+ if (ptr == NULL) {
+ free(path);
+ return NULL;
+ }
+ *ptr = '\0';
+ if (buf != realpath(path, buf)) {
+ free(path);
+ return NULL;
+ }
+ /* add the filename back */
+ len = snprintf(path, PATH_MAX, "%s/%s", buf, ptr+1);
+ if (len >= PATH_MAX) {
+ free(path);
+ return NULL;
+ }
+ return path;
+}
+
+/* Determine if a device is a block device (as opposed to a file) */
+static int is_block(char *devname)
+{
+ struct stat st;
+ int ret = 0;
+ char *devpath;
+
+ devpath = absolute_path(devname);
+ if (devpath == NULL) {
+ fprintf(stderr, "%s: failed to resolve path to %s\n",
+ progname, devname);
+ return -1;
+ }
+
+ ret = access(devname, F_OK);
+ if (ret != 0) {
+ if (strncmp(devpath, "/dev/", 5) == 0) {
+ /* nobody sane wants to create a loopback file under
+ * /dev. Let's just report the device doesn't exist */
+ fprintf(stderr, "%s: %s apparently does not exist\n",
+ progname, devpath);
+ ret = -1;
+ goto out;
+ }
+ ret = 0;
+ goto out;
+ }
+ ret = stat(devpath, &st);
+ if (ret != 0) {
+ fprintf(stderr, "%s: cannot stat %s\n", progname, devpath);
+ goto out;
+ }
+ ret = S_ISBLK(st.st_mode);
+out:
+ free(devpath);
+ return ret;
+}
+
+static int is_feature_enabled(const char *feature, const char *devpath)
+{
+ char cmd[PATH_MAX];
+ FILE *fp;
+ char enabled_features[4096] = "";
+ int ret = 1;
+
+ snprintf(cmd, sizeof(cmd), "%s -R features %s 2>&1",
+ DEBUGFS, devpath);
+
+ /* Using popen() instead of run_command() since debugfs does
+ * not return proper error code if command is not supported */
+ fp = popen(cmd, "r");
+ if (!fp) {
+ fprintf(stderr, "%s: %s\n", progname, strerror(errno));
+ return 0;
+ }
+
+ ret = fread(enabled_features, 1, sizeof(enabled_features) - 1, fp);
+ enabled_features[ret] = '\0';
+ pclose(fp);
+
+ if (strstr(enabled_features, feature))
+ return 1;
+ return 0;
+}
+
/* Write the server config files */
int ldiskfs_write_ldd(struct mkfs_opts *mop)
{
return errno;
}
+ /*
+ * Append file context to mount options if SE Linux is enabled
+ */
+ #ifdef HAVE_SELINUX
+ if (is_selinux_enabled() > 0)
+ append_context_for_mount(mntpt, mop);
+ #endif
+
dev = mop->mo_device;
if (mop->mo_flags & MO_IS_LOOP)
dev = mop->mo_loopdev;
/* Display the need for the latest e2fsprogs to be installed. make_backfs
* indicates if the caller is make_lustre_backfs() or not. */
-void disp_old_e2fsprogs_msg(const char *feature, int make_backfs)
+static void disp_old_e2fsprogs_msg(const char *feature, int make_backfs)
{
static int msg_displayed;
}
ret = fread(supp_features, 1, sizeof(supp_features) - 1, fp);
supp_features[ret] = '\0';
- fclose(fp);
+ pclose(fp);
}
if (ret > 0 && strstr(supp_features,
strncmp(feature, "-O ", 3) ? feature : feature+3))
append_unique(anchor, ",", "huge_file", NULL, maxbuflen);
/* Enable large block addresses if the LUN is over 2^32 blocks. */
- if (mop->mo_device_sz / (L_BLOCK_SIZE >> 10) >= 0x100002000ULL &&
+ if (mop->mo_device_kb / (L_BLOCK_SIZE >> 10) >= 0x100002000ULL &&
is_e2fsprogs_feature_supp("-O 64bit") == 0)
append_unique(anchor, ",", "64bit", NULL, maxbuflen);
*/
static char *moveopts_to_end(char *start)
{
+ size_t len;
char save[512];
char *end, *idx;
while (*end != ' ' && *end != '\0')
++end;
+ len = end - start;
+ if (len >= sizeof(save))
+ len = sizeof(save) - 1;
+
/* save options */
- strncpy(save, start, end - start);
- save[end - start] = '\0';
+ strncpy(save, start, len);
+ save[len] = '\0';
/* move remaining options up front */
if (*end)
/* Build fs according to type */
int ldiskfs_make_lustre(struct mkfs_opts *mop)
{
- __u64 device_sz = mop->mo_device_sz, block_count = 0;
+ __u64 device_kb = mop->mo_device_kb, block_count = 0;
char mkfs_cmd[PATH_MAX];
char buf[64];
char *start;
size_t maxbuflen;
if (!(mop->mo_flags & MO_IS_LOOP)) {
- mop->mo_device_sz = get_device_size(mop->mo_device);
+ mop->mo_device_kb = get_device_size(mop->mo_device);
- if (mop->mo_device_sz == 0)
+ if (mop->mo_device_kb == 0)
return ENODEV;
/* Compare to real size */
- if (device_sz == 0 || device_sz > mop->mo_device_sz)
- device_sz = mop->mo_device_sz;
+ if (device_kb == 0 || device_kb > mop->mo_device_kb)
+ device_kb = mop->mo_device_kb;
else
- mop->mo_device_sz = device_sz;
+ mop->mo_device_kb = device_kb;
}
- if (mop->mo_device_sz != 0) {
- if (mop->mo_device_sz < 8096){
+ if (mop->mo_device_kb != 0) {
+ if (mop->mo_device_kb < 8096) {
fprintf(stderr, "%s: size of filesystem must be larger "
"than 8MB, but is set to %lldKB\n",
- progname, (long long)mop->mo_device_sz);
+ progname, (long long)mop->mo_device_kb);
return EINVAL;
}
- block_count = mop->mo_device_sz / (L_BLOCK_SIZE >> 10);
+ block_count = mop->mo_device_kb / (L_BLOCK_SIZE >> 10);
/* If the LUN size is just over 2^32 blocks, limit the
* filesystem size to 2^32-1 blocks to avoid problems with
* ldiskfs/mkfs not handling this size. Bug 22906 */
long inode_size = 0;
/* Journal size in MB */
- if (strstr(mop->mo_mkfsopts, "-J") == NULL) {
+ if (strstr(mop->mo_mkfsopts, "-J") == NULL &&
+ device_kb > 1024 * 1024) {
/* Choose our own default journal size */
- long journal_sz = 0, max_sz;
- if (device_sz > 1024 * 1024) /* 1GB */
- journal_sz = (device_sz / 102400) * 4;
- /* cap journal size at 1GB */
- if (journal_sz > 1024L)
- journal_sz = 1024L;
- /* man mkfs.ext3 */
- max_sz = (102400 * L_BLOCK_SIZE) >> 20; /* 400MB */
- if (journal_sz > max_sz)
- journal_sz = max_sz;
- if (journal_sz) {
- sprintf(buf, " -J size=%ld", journal_sz);
+ long journal_mb = 0, max_mb;
+
+ /* cap journal size at 4GB for MDT,
+ * leave it at 400MB for OSTs. */
+ if (IS_MDT(&mop->mo_ldd))
+ max_mb = 4096;
+ else if (IS_OST(&mop->mo_ldd))
+ max_mb = 400;
+ else /* Use mke2fs default size for MGS */
+ max_mb = 0;
+
+ /* Use at most 4% of device for journal */
+ journal_mb = device_kb * 4 / (1024 * 100);
+ if (journal_mb > max_mb)
+ journal_mb = max_mb;
+
+ if (journal_mb) {
+ sprintf(buf, " -J size=%ld", journal_mb);
strscat(mop->mo_mkfsopts, buf,
sizeof(mop->mo_mkfsopts));
}
* this, but it is impossible to know in advance. */
if (IS_OST(&mop->mo_ldd)) {
/* OST > 16TB assume average file size 1MB */
- if (device_sz > (16ULL << 30))
+ if (device_kb > (16ULL << 30))
bytes_per_inode = 1024 * 1024;
/* OST > 4TB assume average file size 512kB */
- else if (device_sz > (4ULL << 30))
+ else if (device_kb > (4ULL << 30))
bytes_per_inode = 512 * 1024;
/* OST > 1TB assume average file size 256kB */
- else if (device_sz > (1ULL << 30))
+ else if (device_kb > (1ULL << 30))
bytes_per_inode = 256 * 1024;
/* OST > 10GB assume average file size 64kB,
* plus a bit so that inodes will fit into a
* 256x flex_bg without overflowing */
- else if (device_sz > (10ULL << 20))
+ else if (device_kb > (10ULL << 20))
bytes_per_inode = 69905;
}
* descriptor blocks, but leave one block for the superblock.
* Only useful for filesystems with < 2^32 blocks due to resize
* limitations. */
- if (IS_OST(&mop->mo_ldd) && mop->mo_device_sz > 100 * 1024 &&
- mop->mo_device_sz * 1024 / L_BLOCK_SIZE <= 0xffffffffULL) {
+ if (IS_OST(&mop->mo_ldd) && mop->mo_device_kb > 100 * 1024 &&
+ mop->mo_device_kb * 1024 / L_BLOCK_SIZE <= 0xffffffffULL) {
unsigned group_blocks = L_BLOCK_SIZE * 8;
unsigned desc_per_block = L_BLOCK_SIZE / 32;
unsigned resize_blks;
return 0;
}
-int read_file(char *path, char *buf, int size)
+static int read_file(const char *path, char *buf, int size)
{
FILE *fd;
return 0;
}
-int write_file(char *path, char *buf)
+static int write_file(const char *path, const char *buf)
{
FILE *fd;
return 0;
}
+static int set_blockdev_scheduler(const char *path, const char *scheduler)
+{
+ char buf[PATH_MAX], *c;
+ int rc;
+
+ /* Before setting the scheduler, we need to check to see if it's
+ * already set to "noop". If it is, we don't want to override
+ * that setting. If it's set to anything other than "noop", set
+ * the scheduler to what has been passed in. */
+
+ rc = read_file(path, buf, sizeof(buf));
+ if (rc) {
+ if (verbose)
+ fprintf(stderr, "%s: cannot open '%s': %s\n",
+ progname, path, strerror(errno));
+ return rc;
+ }
+
+ /* The expected format of buf: noop anticipatory deadline [cfq] */
+ c = strchr(buf, '[');
+
+ /* If c is NULL, the format is not what we expect. Play it safe
+ * and error out. */
+ if (c == NULL) {
+ if (verbose)
+ fprintf(stderr, "%s: cannot parse scheduler "
+ "options for '%s'\n", progname, path);
+ return -EINVAL;
+ }
+
+ if (strncmp(c+1, "noop", 4) == 0)
+ return 0;
+
+ rc = write_file(path, scheduler);
+ if (rc) {
+ if (verbose)
+ fprintf(stderr, "%s: cannot set scheduler on "
+ "'%s': %s\n", progname, path,
+ strerror(errno));
+ return rc;
+ }
+
+ return rc;
+}
+
/* 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, struct mount_opts *mop, int fan_out)
+static int set_blockdev_tunables(char *source, struct mount_opts *mop)
{
glob_t glob_info = { 0 };
struct stat stat_buf;
char real_path[PATH_MAX] = {'\0'};
int i, rc = 0;
int major, minor;
+ char *slave = NULL;
if (!source)
return -EINVAL;
chk_major = strtok_r(buf, ":", &savept);
chk_minor = savept;
- if (major == atoi(chk_major) &&minor == atoi(chk_minor))
+ if (chk_major != NULL && major == atoi(chk_major) &&
+ chk_minor != NULL && minor == atoi(chk_minor))
break;
}
}
}
- 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, mop, 0);
- }
+ /* Purposely ignore errors reported from set_blockdev_scheduler.
+ * The worst that will happen is a block device with an "incorrect"
+ * scheduler. */
+ snprintf(real_path, sizeof(real_path), "%s/%s", path, SCHEDULER_PATH);
+ set_blockdev_scheduler(real_path, DEFAULT_SCHEDULER);
+
+ /* if device is multipath device, tune its slave devices */
+ glob_info.gl_pathc = 0;
+ glob_info.gl_offs = 0;
+ 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, mop);
+ }
- 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);
- }
+ 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);
}
+ globfree(&glob_info);
return rc;
}
int ldiskfs_tune_lustre(char *dev, struct mount_opts *mop)
{
- return set_blockdev_tunables(dev, mop, 1);
+ return set_blockdev_tunables(dev, mop);
}
int ldiskfs_label_lustre(struct mount_opts *mop)
return rc;
}
-/* return canonicalized absolute pathname, even if the target file does not
- * exist, unlike realpath */
-static char *absolute_path(char *devname)
-{
- char buf[PATH_MAX + 1];
- char *path;
- char *ptr;
-
- path = malloc(PATH_MAX + 1);
- if (path == NULL)
- return NULL;
-
- if (devname[0] != '/') {
- if (getcwd(buf, sizeof(buf) - 1) == NULL) {
- free(path);
- return NULL;
- }
- strcat(buf, "/");
- strcat(buf, devname);
- } else {
- strcpy(buf, devname);
- }
- /* truncate filename before calling realpath */
- ptr = strrchr(buf, '/');
- if (ptr == NULL) {
- free(path);
- return NULL;
- }
- *ptr = '\0';
- if (path != realpath(buf, path)) {
- free(path);
- return NULL;
- }
- /* add the filename back */
- strcat(path, "/");
- strcat(path, ptr + 1);
- return path;
-}
-
-/* Determine if a device is a block device (as opposed to a file) */
-int is_block(char* devname)
-{
- struct stat st;
- int ret = 0;
- char *devpath;
-
- devpath = absolute_path(devname);
- if (devpath == NULL) {
- fprintf(stderr, "%s: failed to resolve path to %s\n",
- progname, devname);
- return -1;
- }
-
- ret = access(devname, F_OK);
- if (ret != 0) {
- if (strncmp(devpath, "/dev/", 5) == 0) {
- /* nobody sane wants to create a loopback file under
- * /dev. Let's just report the device doesn't exist */
- fprintf(stderr, "%s: %s apparently does not exist\n",
- progname, devpath);
- ret = -1;
- goto out;
- }
- ret = 0;
- goto out;
- }
- ret = stat(devpath, &st);
- if (ret != 0) {
- fprintf(stderr, "%s: cannot stat %s\n", progname, devpath);
- goto out;
- }
- ret = S_ISBLK(st.st_mode);
-out:
- free(devpath);
- return ret;
-}
-
-static int is_feature_enabled(const char *feature, const char *devpath)
-{
- char cmd[PATH_MAX];
- FILE *fp;
- char enabled_features[4096] = "";
- int ret = 1;
-
- snprintf(cmd, sizeof(cmd), "%s -R features %s 2>&1",
- DEBUGFS, devpath);
-
- /* Using popen() instead of run_command() since debugfs does
- * not return proper error code if command is not supported */
- fp = popen(cmd, "r");
- if (!fp) {
- fprintf(stderr, "%s: %s\n", progname, strerror(errno));
- return 0;
- }
-
- ret = fread(enabled_features, 1, sizeof(enabled_features) - 1, fp);
- enabled_features[ret] = '\0';
- fclose(fp);
-
- if (strstr(enabled_features, feature))
- return 1;
- return 0;
-}
-
/* Enable quota accounting */
int ldiskfs_enable_quota(struct mkfs_opts *mop)
{