* Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2011, Whamcloud, Inc.
+ * Copyright (c) 2012, 2015, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#ifndef _GNU_SOURCE
#define _GNU_SOURCE
#endif
+#include "mount_utils.h"
#include <stdlib.h>
#include <stdio.h>
#include <unistd.h>
#include <fcntl.h>
#include <stdarg.h>
#include <mntent.h>
+#include <glob.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <limits.h>
#include <ctype.h>
-#ifdef __linux__
-/* libcfs.h is not really needed here, but on SLES10/PPC, fs.h includes idr.h
- * which requires BITS_PER_LONG to be defined */
-#include <libcfs/libcfs.h>
#ifndef BLKGETSIZE64
#include <linux/fs.h> /* for BLKGETSIZE64 */
#endif
#include <linux/version.h>
-#endif
#include <lustre_disk.h>
#include <lustre_param.h>
#include <lnet/lnetctl.h>
#include <lustre_ver.h>
-#include "mount_utils.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 DUMMY_FILE_NAME_LEN 25
#define EXT3_DIRENT_SIZE DUMMY_FILE_NAME_LEN
+static void append_unique(char *buf, char *prefix, char *key, char *val,
+ size_t maxbuflen);
+
+/*
+ * Concatenate context of the temporary mount point if 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) {
+ append_unique(mop->mo_ldd.ldd_mount_opts,
+ ",", "context", fcontext,
+ sizeof(mop->mo_ldd.ldd_mount_opts));
+ 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 write_local_files(struct mkfs_opts *mop)
+int ldiskfs_write_ldd(struct mkfs_opts *mop)
{
char mntpt[] = "/tmp/mntXXXXXX";
char filepnm[128];
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;
ret = mount(dev, mntpt, MT_STR(&mop->mo_ldd), 0,
- mop->mo_ldd.ldd_mount_opts);
+ (mop->mo_mountopts == NULL) ?
+ "errors=remount-ro" : mop->mo_mountopts);
if (ret) {
fprintf(stderr, "%s: Unable to mount %s: %s\n",
progname, dev, strerror(errno));
if (num < 1 && ferror(filep)) {
fprintf(stderr, "%s: Unable to write to file (%s): %s\n",
progname, filepnm, strerror(errno));
+ fclose(filep);
goto out_umnt;
}
fclose(filep);
return ret;
}
-int read_local_files(struct mkfs_opts *mop)
+static int readcmd(char *cmd, char *buf, int len)
+{
+ FILE *fp;
+ int red;
+
+ fp = popen(cmd, "r");
+ if (!fp)
+ return errno;
+
+ red = fread(buf, 1, len, fp);
+ pclose(fp);
+
+ /* strip trailing newline */
+ if (buf[red - 1] == '\n')
+ buf[red - 1] = '\0';
+
+ return (red == 0) ? -ENOENT : 0;
+}
+
+int ldiskfs_read_ldd(char *dev, struct lustre_disk_data *mo_ldd)
{
char tmpdir[] = "/tmp/dirXXXXXX";
char cmd[PATH_MAX];
char filepnm[128];
- char *dev;
FILE *filep;
int ret = 0;
int cmdsz = sizeof(cmd);
return errno;
}
- dev = mop->mo_device;
-
/* TODO: it's worth observing the get_mountdata() function that is
in mount_utils.c for getting the mountdata out of the
filesystem */
if (filep) {
size_t num_read;
vprint("Reading %s\n", MOUNT_DATA_FILE);
- num_read = fread(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
+ num_read = fread(mo_ldd, sizeof(*mo_ldd), 1, filep);
if (num_read < 1 && ferror(filep)) {
fprintf(stderr, "%s: Unable to read from file %s: %s\n",
progname, filepnm, strerror(errno));
- goto out_close;
}
+ fclose(filep);
}
-out_close:
- fclose(filep);
snprintf(cmd, cmdsz, "rm -rf %s", tmpdir);
run_command(cmd, cmdsz);
if (ret)
verrprint("Failed to read old data (%d)\n", ret);
+
+ /* As long as we at least have the label, we're good to go */
+ snprintf(cmd, sizeof(cmd), E2LABEL" %s", dev);
+ ret = readcmd(cmd, mo_ldd->ldd_svname, sizeof(mo_ldd->ldd_svname) - 1);
+
return ret;
}
/* 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;
E2FSPROGS, feature);
#if !(HAVE_LDISKFSPROGS)
fprintf(stderr, "Please install the latest version of e2fsprogs from\n"
- "http://downloads.whamcloud.com/public/e2fsprogs/latest/\n"
+ "https://downloads.hpdd.intel.com/public/e2fsprogs/latest/\n"
"to enable this feature.\n");
#endif
if (make_backfs)
pclose(fp);
return 1;
}
- i = fread(debugfs_cmd, 1, sizeof(debugfs_cmd), fp);
+ i = fread(debugfs_cmd, 1, sizeof(debugfs_cmd) - 1, fp);
if (i) {
debugfs_cmd[i] = 0;
fprintf(stderr, "%s", debugfs_cmd);
fprintf(stderr, "%s: %s\n", progname, strerror(errno));
return 0;
}
- ret = fread(supp_features, 1, sizeof(supp_features), fp);
- fclose(fp);
+ ret = fread(supp_features, 1, sizeof(supp_features) - 1, fp);
+ supp_features[ret] = '\0';
+ pclose(fp);
}
if (ret > 0 && strstr(supp_features,
strncmp(feature, "-O ", 3) ? feature : feature+3))
strscat(buf, key, maxbuflen);
if (val != NULL) {
- strscat(buf, "=", maxbuflen);
+ strscat(buf, "=\"", maxbuflen);
strscat(buf, val, maxbuflen);
+ strscat(buf, "\"", maxbuflen);
}
}
}
-static void enable_default_ext4_features(struct mkfs_opts *mop, char *anchor,
- size_t maxbuflen, int user_spec)
+static int enable_default_ext4_features(struct mkfs_opts *mop, char *anchor,
+ size_t maxbuflen, int user_spec)
{
if (IS_OST(&mop->mo_ldd)) {
append_unique(anchor, user_spec ? "," : " -O ",
/* The following options are only valid for ext4-based ldiskfs.
* If --backfstype=ext3 is specified, do not enable them. */
if (mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3)
- return;
+ return 0;
+
+ /* Enable quota by default */
+ if (is_e2fsprogs_feature_supp("-O quota") == 0) {
+ append_unique(anchor, ",", "quota", NULL, maxbuflen);
+ } else {
+ fatal();
+ fprintf(stderr, "\"-O quota\" must be supported by "
+ "e2fsprogs, please upgrade your e2fsprogs.\n");
+ return EINVAL;
+ }
/* Allow files larger than 2TB. Also needs LU-16, but not harmful. */
if (is_e2fsprogs_feature_supp("-O huge_file") == 0)
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);
append_unique(anchor, ",", "flex_bg", NULL, maxbuflen);
- if (IS_OST(&mop->mo_ldd)) {
+ if (IS_OST(&mop->mo_ldd) &&
+ strstr(mop->mo_mkfsopts, "-G") == NULL) {
snprintf(tmp_buf, sizeof(tmp_buf), " -G %u",
(1 << 20) / L_BLOCK_SIZE);
strscat(anchor, tmp_buf, maxbuflen);
}
}
/* Don't add any more "-O" options here, see last comment above */
+ return 0;
}
/**
*/
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 make_lustre_backfs(struct mkfs_opts *mop)
+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 < 32384) {
fprintf(stderr, "%s: size of filesystem must be larger "
- "than 8MB, but is set to %lldKB\n",
- progname, (long long)mop->mo_device_sz);
+ "than 32MB, but is set to %lldKB\n",
+ 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));
}
}
- /* Inode size (for extended attributes). The LOV EA size is
- * 32 (EA hdr) + 32 (lov_mds_md) + stripes * 24 (lov_ost_data),
- * and we want some margin above that for ACLs, other EAs... */
+ /* Inode size includes:
+ * ldiskfs inode size: 156
+ * extended attributes size, including:
+ * ext4_xattr_header: 32
+ * LOV EA size: 32(lov_mds_md) +
+ * stripes * 24(lov_ost_data) +
+ * 16(xattr_entry) + 3(lov)
+ * LMA EA size: 24(lustre_mdt_attrs) +
+ * 16(xattr_entry) + 3(lma)
+ * link EA size: 24(link_ea_header) + 18(link_ea_entry) +
+ * (filename) + 16(xattr_entry) + 4(link)
+ * and some margin for 4-byte alignment, ACLs and other EAs.
+ *
+ * If we say the average filename length is about 32 bytes,
+ * the calculation looks like:
+ * 156 + 32 + (32+24*N+19) + (24+19) + (24+18+~32+20) + other <=
+ * 512*2^m, {m=0,1,2,3}
+ */
if (strstr(mop->mo_mkfsopts, "-I") == NULL) {
if (IS_MDT(&mop->mo_ldd)) {
- if (mop->mo_stripe_count > 72)
+ if (mop->mo_stripe_count > 69)
inode_size = 512; /* bz 7241 */
/* see also "-i" below for EA blocks */
- else if (mop->mo_stripe_count > 32)
+ else if (mop->mo_stripe_count > 26)
inode_size = 2048;
- else if (mop->mo_stripe_count > 10)
+ else if (mop->mo_stripe_count > 5)
inode_size = 1024;
else
inode_size = 512;
if (IS_MDT(&mop->mo_ldd)) {
bytes_per_inode = inode_size + 1536;
- if (mop->mo_stripe_count > 72) {
+ if (mop->mo_stripe_count > 69) {
int extra = mop->mo_stripe_count * 24;
extra = ((extra - 1) | 4095) + 1;
bytes_per_inode += extra;
* 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;
}
start = moveopts_to_end(start);
maxbuflen = sizeof(mop->mo_mkfsopts) -
(start - mop->mo_mkfsopts) - strlen(start);
- enable_default_ext4_features(mop, start, maxbuflen, 1);
+ ret = enable_default_ext4_features(mop, start, maxbuflen, 1);
} else {
start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts),
maxbuflen = sizeof(mop->mo_mkfsopts) -
strlen(mop->mo_mkfsopts);
- enable_default_ext4_features(mop, start, maxbuflen, 0);
+ ret = enable_default_ext4_features(mop, start, maxbuflen, 0);
}
+ if (ret)
+ return ret;
/* end handle -O mkfs options */
/* start handle -E mkfs options */
* 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;
vprint("formatting backing filesystem %s on %s\n",
MT_STR(&mop->mo_ldd), dev);
- vprint("\ttarget name %s\n", mop->mo_ldd.ldd_svname);
- vprint("\t4k blocks "LPU64"\n", block_count);
+ vprint("\ttarget name %s\n", mop->mo_ldd.ldd_svname);
+ vprint("\t4k blocks %ju\n", (uintmax_t)block_count);
vprint("\toptions %s\n", mop->mo_mkfsopts);
/* mkfs_cmd's trailing space is important! */
strscat(mkfs_cmd, " ", sizeof(mkfs_cmd));
strscat(mkfs_cmd, dev, sizeof(mkfs_cmd));
if (block_count != 0) {
- sprintf(buf, " "LPU64, block_count);
+ snprintf(buf, sizeof(buf), " %ju",
+ (uintmax_t)block_count);
strscat(mkfs_cmd, buf, sizeof(mkfs_cmd));
}
}
int ldiskfs_prepare_lustre(struct mkfs_opts *mop,
- char *default_mountopts, int default_len,
- char *always_mountopts, int always_len)
+ char *wanted_mountopts, size_t len)
{
struct lustre_disk_data *ldd = &mop->mo_ldd;
int ret;
mop->mo_flags |= MO_IS_LOOP;
}
- strscat(default_mountopts, ",errors=remount-ro", default_len);
if (IS_MDT(ldd) || IS_MGS(ldd))
- strscat(always_mountopts, ",user_xattr", always_len);
+ strscat(wanted_mountopts, ",user_xattr", len);
return 0;
}
-/* return canonicalized absolute pathname, even if the target file does not
- * exist, unlike realpath */
-static char *absolute_path(char *devname)
+int ldiskfs_fix_mountopts(struct mkfs_opts *mop, char *mountopts, size_t len)
{
- char buf[PATH_MAX + 1];
- char *path;
- char *ptr;
+ if (strstr(mountopts, "errors=") == NULL)
+ strscat(mountopts, ",errors=remount-ro", len);
- path = malloc(PATH_MAX + 1);
- if (path == NULL)
- return NULL;
+ return 0;
+}
- if (devname[0] != '/') {
- if (getcwd(buf, sizeof(buf) - 1) == NULL)
- return NULL;
- strcat(buf, "/");
- strcat(buf, devname);
- } else {
- strcpy(buf, devname);
+static int read_file(const char *path, char *buf, int size)
+{
+ FILE *fd;
+
+ fd = fopen(path, "r");
+ if (fd == NULL)
+ return errno;
+
+ if (fgets(buf, size, fd) == NULL) {
+ fprintf(stderr, "reading from %s: %s", path, strerror(errno));
+ fclose(fd);
+ return 1;
}
- /* truncate filename before calling realpath */
- ptr = strrchr(buf, '/');
- if (ptr == NULL) {
- free(path);
- return NULL;
+ fclose(fd);
+
+ /* strip trailing newline */
+ size = strlen(buf);
+ if (buf[size - 1] == '\n')
+ buf[size - 1] = '\0';
+
+ return 0;
+}
+
+static int write_file(const char *path, const char *buf)
+{
+ int fd, rc;
+
+ fd = open(path, O_WRONLY);
+ if (fd < 0)
+ return errno;
+
+ rc = write(fd, buf, strlen(buf));
+ close(fd);
+
+ return rc < 0 ? errno : 0;
+}
+
+static int set_blockdev_scheduler(const char *path, const char *scheduler)
+{
+ char buf[PATH_MAX], *s, *e, orig_sched[50];
+ 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;
}
- *ptr = '\0';
- if (path != realpath(buf, path)) {
- free(path);
- return NULL;
+
+ /* The expected format of buf: noop anticipatory deadline [cfq] */
+ s = strchr(buf, '[');
+ e = strchr(buf, ']');
+
+ /* If the format is not what we expect. Play it safe and error out. */
+ if (s == NULL || e == NULL) {
+ if (verbose)
+ fprintf(stderr, "%s: cannot parse scheduler "
+ "options for '%s'\n", progname, path);
+ return -EINVAL;
}
- /* add the filename back */
- strcat(path, "/");
- strcat(path, ptr + 1);
- return path;
+
+ snprintf(orig_sched, e - s, "%s", s + 1);
+
+ if (strcmp(orig_sched, "noop") == 0 ||
+ strcmp(orig_sched, scheduler) == 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;
+ } else {
+ fprintf(stderr, "%s: change scheduler of %s from %s to %s\n",
+ progname, path, orig_sched, scheduler);
+ }
+
+ return rc;
}
-/* Determine if a device is a block device (as opposed to a file) */
-int is_block(char* devname)
+/* 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 */
+static int set_blockdev_tunables(char *source, struct mount_opts *mop)
{
- struct stat st;
- int ret = 0;
- char *devpath;
+ glob_t glob_info = { 0 };
+ struct stat stat_buf;
+ char *chk_major, *chk_minor;
+ char *savept = NULL, *dev;
+ char *ret_path;
+ char buf[PATH_MAX] = {'\0'}, path[PATH_MAX] = {'\0'};
+ char real_path[PATH_MAX] = {'\0'};
+ int i, rc = 0;
+ int major, minor;
+ char *slave = NULL;
+
+ if (!source)
+ return -EINVAL;
+
+ ret_path = realpath(source, real_path);
+ if (ret_path == NULL) {
+ if (verbose)
+ fprintf(stderr, "warning: %s: cannot resolve: %s\n",
+ source, strerror(errno));
+ return -EINVAL;
+ }
- devpath = absolute_path(devname);
- if (devpath == NULL) {
- fprintf(stderr, "%s: failed to resolve path to %s\n",
- progname, devname);
- return -1;
+ if (strncmp(real_path, "/dev/loop", 9) == 0)
+ return 0;
+
+ if ((real_path[0] != '/') && (strpbrk(real_path, ",:") != NULL))
+ return 0;
+
+ 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 */
+
+ /* 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;
}
- 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;
+ rc = stat(real_path, &stat_buf);
+ if (rc) {
+ if (verbose)
+ fprintf(stderr, "warning: %s, device %s stat failed\n",
+ strerror(errno), real_path);
+ return rc;
+ }
+
+ major = major(stat_buf.st_rdev);
+ minor = minor(stat_buf.st_rdev);
+ rc = glob("/sys/block/*", GLOB_NOSORT, NULL, &glob_info);
+ if (rc) {
+ if (verbose)
+ fprintf(stderr, "warning: failed to read entries under "
+ "/sys/block\n");
+ globfree(&glob_info);
+ return rc;
+ }
+
+ for (i = 0; i < glob_info.gl_pathc; i++){
+ snprintf(path, sizeof(path), "%s/dev", glob_info.gl_pathv[i]);
+
+ rc = read_file(path, buf, sizeof(buf));
+ if (rc)
+ continue;
+
+ if (buf[strlen(buf) - 1] == '\n')
+ buf[strlen(buf) - 1] = '\0';
+
+ chk_major = strtok_r(buf, ":", &savept);
+ chk_minor = savept;
+ if (chk_major != NULL && major == atoi(chk_major) &&
+ chk_minor != NULL && minor == atoi(chk_minor))
+ break;
+ }
+
+ if (i == glob_info.gl_pathc) {
+ if (verbose)
+ fprintf(stderr,"warning: device %s does not match any "
+ "entry under /sys/block\n", real_path);
+ globfree(&glob_info);
+ return -EINVAL;
+ }
+
+ /* 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;
}
- ret = 0;
- goto out;
+
+ if (atoi(buf) >= mop->mo_md_stripe_cache_size)
+ return 0;
+
+ if (strlen(buf) - 1 > 0) {
+ snprintf(buf, sizeof(buf), "%d",
+ mop->mo_md_stripe_cache_size);
+ rc = write_file(real_path, buf);
+ if (rc != 0 && verbose)
+ fprintf(stderr, "warning: opening %s: %s\n",
+ real_path, strerror(errno));
+ }
+ /* Return since raid and disk tunables are different */
+ return rc;
}
- ret = stat(devpath, &st);
- if (ret != 0) {
- fprintf(stderr, "%s: cannot stat %s\n", progname, devpath);
- goto out;
+
+ snprintf(real_path, sizeof(real_path), "%s/%s", path,
+ MAX_HW_SECTORS_KB_PATH);
+ rc = read_file(real_path, buf, sizeof(buf));
+ if (rc) {
+ if (verbose)
+ fprintf(stderr, "warning: opening %s: %s\n",
+ real_path, strerror(errno));
+ /* No MAX_HW_SECTORS_KB_PATH isn't necessary an
+ * error for some device. */
+ goto subdevs;
}
- ret = S_ISBLK(st.st_mode);
-out:
- free(devpath);
+
+ if (strlen(buf) - 1 > 0) {
+ char oldbuf[32] = "", *end = NULL;
+ unsigned long long oldval, newval;
+
+ snprintf(real_path, sizeof(real_path), "%s/%s", path,
+ MAX_SECTORS_KB_PATH);
+ rc = read_file(real_path, oldbuf, sizeof(oldbuf));
+ /* Only set new parameter if different from the old one. */
+ if (rc != 0 || strcmp(oldbuf, buf) == 0) {
+ /* No MAX_SECTORS_KB_PATH isn't necessary an
+ * error for some device. */
+ goto subdevs;
+ }
+
+ newval = strtoull(buf, &end, 0);
+ if (newval == 0 || newval == ULLONG_MAX || end == buf)
+ goto subdevs;
+
+ /* Don't increase IO request size limit past 16MB. It is about
+ * PTLRPC_MAX_BRW_SIZE, but that isn't in a public header.
+ * Note that even though the block layer allows larger values,
+ * setting max_sectors_kb = 32768 causes crashes (LU-6974). */
+ if (newval > 16 * 1024) {
+ newval = 16 * 1024;
+ snprintf(buf, sizeof(buf), "%llu", newval);
+ }
+
+ oldval = strtoull(oldbuf, &end, 0);
+ /* Don't shrink the current limit. */
+ if (oldval != ULLONG_MAX && newval <= oldval)
+ goto subdevs;
+
+ rc = write_file(real_path, buf);
+ if (rc != 0) {
+ 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. */
+ goto subdevs;
+ }
+ fprintf(stderr, "%s: increased %s from %s to %s\n",
+ progname, real_path, oldbuf, buf);
+ }
+
+subdevs:
+ /* 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);
+ }
+ }
+ globfree(&glob_info);
+
+ return rc;
+}
+
+int ldiskfs_tune_lustre(char *dev, struct mount_opts *mop)
+{
+ return set_blockdev_tunables(dev, mop);
+}
+
+int ldiskfs_label_lustre(struct mount_opts *mop)
+{
+ char label_cmd[PATH_MAX];
+ int rc;
+
+ snprintf(label_cmd, sizeof(label_cmd),
+ TUNE2FS" -f -L '%s' '%s' >/dev/null 2>&1",
+ mop->mo_ldd.ldd_svname, mop->mo_source);
+ rc = run_command(label_cmd, sizeof(label_cmd));
+
+ return rc;
+}
+
+/* Enable quota accounting */
+int ldiskfs_enable_quota(struct mkfs_opts *mop)
+{
+ char *dev;
+ char cmd[512];
+ int cmdsz = sizeof(cmd), ret;
+
+ if (is_e2fsprogs_feature_supp("-O quota") != 0) {
+ fprintf(stderr, "%s: \"-O quota\" is is not supported by "
+ "current e2fsprogs\n", progname);
+ return EINVAL;
+ }
+
+ dev = mop->mo_device;
+ if (mop->mo_flags & MO_IS_LOOP)
+ dev = mop->mo_loopdev;
+
+ /* Quota feature is already enabled? */
+ if (is_feature_enabled("quota", dev)) {
+ vprint("Quota feature is already enabled.\n");
+ return 0;
+ }
+
+ /* Turn on quota feature by "tune2fs -O quota" */
+ snprintf(cmd, cmdsz, "%s -O quota %s", TUNE2FS, dev);
+ ret = run_command(cmd, cmdsz);
+ if (ret)
+ fprintf(stderr, "command:%s (%d)", cmd, ret);
+
return ret;
}
+int ldiskfs_init(void)
+{
+ /* Required because full path to DEBUGFS is not specified */
+ setenv("PATH", "/bin:/sbin:/usr/bin:/usr/sbin:/usr/local/bin", 0);
+
+ return 0;
+}
+
+void ldiskfs_fini(void)
+{
+ return;
+}
+