4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2012, 2015, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/utils/mount_utils_ldiskfs.c
38 * Author: Nathan Rutman <nathan@clusterfs.com>
41 /* This source file is compiled into both mkfs.lustre and tunefs.lustre */
45 #endif /* HAVE_CONFIG_H */
50 #include "mount_utils.h"
59 #include <sys/types.h>
61 #include <sys/mount.h>
62 #include <sys/utsname.h>
70 #include <linux/fs.h> /* for BLKGETSIZE64 */
72 #include <linux/version.h>
73 #include <lustre_disk.h>
74 #include <lustre_param.h>
75 #include <lnet/lnetctl.h>
76 #include <lustre_ver.h>
79 #include <selinux/selinux.h>
82 #define MAX_HW_SECTORS_KB_PATH "queue/max_hw_sectors_kb"
83 #define MAX_SECTORS_KB_PATH "queue/max_sectors_kb"
84 #define SCHEDULER_PATH "queue/scheduler"
85 #define STRIPE_CACHE_SIZE "md/stripe_cache_size"
87 #define DEFAULT_SCHEDULER "deadline"
89 extern char *progname;
91 #define L_BLOCK_SIZE 4096
92 /* keep it less than LL_FID_NAMELEN */
93 #define DUMMY_FILE_NAME_LEN 25
94 #define EXT3_DIRENT_SIZE DUMMY_FILE_NAME_LEN
97 * Concatenate context of the temporary mount point iff selinux is enabled
100 static void append_context_for_mount(char *mntpt, struct mkfs_opts *mop)
102 security_context_t fcontext;
104 if (getfilecon(mntpt, &fcontext) < 0) {
105 /* Continuing with default behaviour */
106 fprintf(stderr, "%s: Get file context failed : %s\n",
107 progname, strerror(errno));
111 if (fcontext != NULL) {
112 strcat(mop->mo_ldd.ldd_mount_opts, ",context=");
113 strcat(mop->mo_ldd.ldd_mount_opts, fcontext);
119 /* return canonicalized absolute pathname, even if the target file does not
120 * exist, unlike realpath */
121 static char *absolute_path(char *devname)
123 char buf[PATH_MAX + 1] = "";
128 path = malloc(sizeof(buf));
132 if (devname[0] != '/') {
133 if (getcwd(buf, sizeof(buf) - 1) == NULL) {
137 len = snprintf(path, sizeof(buf), "%s/%s", buf, devname);
138 if (len >= sizeof(buf)) {
143 len = snprintf(path, sizeof(buf), "%s", devname);
144 if (len >= sizeof(buf)) {
150 /* truncate filename before calling realpath */
151 ptr = strrchr(path, '/');
157 if (buf != realpath(path, buf)) {
161 /* add the filename back */
162 len = snprintf(path, PATH_MAX, "%s/%s", buf, ptr+1);
163 if (len >= PATH_MAX) {
170 /* Determine if a device is a block device (as opposed to a file) */
171 static int is_block(char *devname)
177 devpath = absolute_path(devname);
178 if (devpath == NULL) {
179 fprintf(stderr, "%s: failed to resolve path to %s\n",
184 ret = access(devname, F_OK);
186 if (strncmp(devpath, "/dev/", 5) == 0) {
187 /* nobody sane wants to create a loopback file under
188 * /dev. Let's just report the device doesn't exist */
189 fprintf(stderr, "%s: %s apparently does not exist\n",
197 ret = stat(devpath, &st);
199 fprintf(stderr, "%s: cannot stat %s\n", progname, devpath);
202 ret = S_ISBLK(st.st_mode);
208 static int is_feature_enabled(const char *feature, const char *devpath)
212 char enabled_features[4096] = "";
215 snprintf(cmd, sizeof(cmd), "%s -R features %s 2>&1",
218 /* Using popen() instead of run_command() since debugfs does
219 * not return proper error code if command is not supported */
220 fp = popen(cmd, "r");
222 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
226 ret = fread(enabled_features, 1, sizeof(enabled_features) - 1, fp);
227 enabled_features[ret] = '\0';
230 if (strstr(enabled_features, feature))
235 /* Write the server config files */
236 int ldiskfs_write_ldd(struct mkfs_opts *mop)
238 char mntpt[] = "/tmp/mntXXXXXX";
245 /* Mount this device temporarily in order to write these files */
246 if (!mkdtemp(mntpt)) {
247 fprintf(stderr, "%s: Can't create temp mount point %s: %s\n",
248 progname, mntpt, strerror(errno));
253 * Append file context to mount options if SE Linux is enabled
256 if (is_selinux_enabled() > 0)
257 append_context_for_mount(mntpt, mop);
260 dev = mop->mo_device;
261 if (mop->mo_flags & MO_IS_LOOP)
262 dev = mop->mo_loopdev;
264 ret = mount(dev, mntpt, MT_STR(&mop->mo_ldd), 0,
265 (mop->mo_mountopts == NULL) ?
266 "errors=remount-ro" : mop->mo_mountopts);
268 fprintf(stderr, "%s: Unable to mount %s: %s\n",
269 progname, dev, strerror(errno));
271 if (errno == ENODEV) {
272 fprintf(stderr, "Is the %s module available?\n",
273 MT_STR(&mop->mo_ldd));
278 /* Set up initial directories */
279 sprintf(filepnm, "%s/%s", mntpt, MOUNT_CONFIGS_DIR);
280 ret = mkdir(filepnm, 0777);
281 if ((ret != 0) && (errno != EEXIST)) {
282 fprintf(stderr, "%s: Can't make configs dir %s (%s)\n",
283 progname, filepnm, strerror(errno));
285 } else if (errno == EEXIST) {
289 /* Save the persistent mount data into a file. Lustre must pre-read
290 this file to get the real mount options. */
291 vprint("Writing %s\n", MOUNT_DATA_FILE);
292 sprintf(filepnm, "%s/%s", mntpt, MOUNT_DATA_FILE);
293 filep = fopen(filepnm, "w");
295 fprintf(stderr, "%s: Unable to create %s file: %s\n",
296 progname, filepnm, strerror(errno));
299 num = fwrite(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
300 if (num < 1 && ferror(filep)) {
301 fprintf(stderr, "%s: Unable to write to file (%s): %s\n",
302 progname, filepnm, strerror(errno));
315 static int readcmd(char *cmd, char *buf, int len)
320 fp = popen(cmd, "r");
324 red = fread(buf, 1, len, fp);
327 /* strip trailing newline */
328 if (buf[red - 1] == '\n')
331 return (red == 0) ? -ENOENT : 0;
334 int ldiskfs_read_ldd(char *dev, struct lustre_disk_data *mo_ldd)
336 char tmpdir[] = "/tmp/dirXXXXXX";
341 int cmdsz = sizeof(cmd);
343 /* Make a temporary directory to hold Lustre data files. */
344 if (!mkdtemp(tmpdir)) {
345 fprintf(stderr, "%s: Can't create temporary directory %s: %s\n",
346 progname, tmpdir, strerror(errno));
350 /* TODO: it's worth observing the get_mountdata() function that is
351 in mount_utils.c for getting the mountdata out of the
354 /* Construct debugfs command line. */
355 snprintf(cmd, cmdsz, "%s -c -R 'dump /%s %s/mountdata' '%s'",
356 DEBUGFS, MOUNT_DATA_FILE, tmpdir, dev);
358 ret = run_command(cmd, cmdsz);
360 verrprint("%s: Unable to dump %s dir (%d)\n",
361 progname, MOUNT_CONFIGS_DIR, ret);
363 sprintf(filepnm, "%s/mountdata", tmpdir);
364 filep = fopen(filepnm, "r");
367 vprint("Reading %s\n", MOUNT_DATA_FILE);
368 num_read = fread(mo_ldd, sizeof(*mo_ldd), 1, filep);
369 if (num_read < 1 && ferror(filep)) {
370 fprintf(stderr, "%s: Unable to read from file %s: %s\n",
371 progname, filepnm, strerror(errno));
376 snprintf(cmd, cmdsz, "rm -rf %s", tmpdir);
377 run_command(cmd, cmdsz);
379 verrprint("Failed to read old data (%d)\n", ret);
381 /* As long as we at least have the label, we're good to go */
382 snprintf(cmd, sizeof(cmd), E2LABEL" %s", dev);
383 ret = readcmd(cmd, mo_ldd->ldd_svname, sizeof(mo_ldd->ldd_svname) - 1);
389 /* Display the need for the latest e2fsprogs to be installed. make_backfs
390 * indicates if the caller is make_lustre_backfs() or not. */
391 static void disp_old_e2fsprogs_msg(const char *feature, int make_backfs)
393 static int msg_displayed;
396 fprintf(stderr, "WARNING: %s does not support %s "
397 "feature.\n\n", E2FSPROGS, feature);
403 fprintf(stderr, "WARNING: The %s package currently installed on "
404 "your system does not support \"%s\" feature.\n",
406 #if !(HAVE_LDISKFSPROGS)
407 fprintf(stderr, "Please install the latest version of e2fsprogs from\n"
408 "https://downloads.hpdd.intel.com/public/e2fsprogs/latest/\n"
409 "to enable this feature.\n");
412 fprintf(stderr, "Feature will not be enabled until %s"
413 "is updated and '%s -O %s %%{device}' "
414 "is run.\n\n", E2FSPROGS, TUNE2FS, feature);
417 /* Check whether the file exists in the device */
418 static int file_in_dev(char *file_name, char *dev_name)
421 char debugfs_cmd[256];
422 unsigned int inode_num;
425 /* Construct debugfs command line. */
426 snprintf(debugfs_cmd, sizeof(debugfs_cmd),
427 "%s -c -R 'stat %s' '%s' 2>&1 | egrep '(Inode|unsupported)'",
428 DEBUGFS, file_name, dev_name);
430 fp = popen(debugfs_cmd, "r");
432 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
436 if (fscanf(fp, "Inode: %u", &inode_num) == 1) { /* exist */
440 i = fread(debugfs_cmd, 1, sizeof(debugfs_cmd) - 1, fp);
443 fprintf(stderr, "%s", debugfs_cmd);
444 if (strstr(debugfs_cmd, "unsupported feature")) {
445 disp_old_e2fsprogs_msg("an unknown", 0);
454 /* Check whether the device has already been used with lustre */
455 int ldiskfs_is_lustre(char *dev, unsigned *mount_type)
459 ret = file_in_dev(MOUNT_DATA_FILE, dev);
461 /* in the -1 case, 'extents' means IS a lustre target */
462 *mount_type = LDD_MT_LDISKFS;
466 ret = file_in_dev(LAST_RCVD, dev);
468 *mount_type = LDD_MT_LDISKFS;
475 /* Check if a certain feature is supported by e2fsprogs.
476 * Firstly we try to use "debugfs supported_features" command to check if
477 * the feature is supported. If this fails we try to set this feature with
478 * mke2fs to check for its support. */
479 static int is_e2fsprogs_feature_supp(const char *feature)
481 static char supp_features[4096] = "";
484 char imgname[] = "/tmp/test-img-XXXXXX";
488 if (supp_features[0] == '\0') {
489 snprintf(cmd, sizeof(cmd), "%s -c -R supported_features 2>&1",
492 /* Using popen() instead of run_command() since debugfs does
493 * not return proper error code if command is not supported */
494 fp = popen(cmd, "r");
496 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
499 ret = fread(supp_features, 1, sizeof(supp_features) - 1, fp);
500 supp_features[ret] = '\0';
503 if (ret > 0 && strstr(supp_features,
504 strncmp(feature, "-O ", 3) ? feature : feature+3))
507 if ((fd = mkstemp(imgname)) < 0)
512 snprintf(cmd, sizeof(cmd), "%s -F %s %s 100 >/dev/null 2>&1",
513 MKE2FS, feature, imgname);
514 /* run_command() displays the output of mke2fs when it fails for
515 * some feature, so use system() directly */
524 * append_unique: append @key or @key=@val pair to @buf only if @key does not
526 * @buf: buffer to hold @key or @key=@val
527 * @prefix: prefix string before @key
529 * @val: value string if it's a @key=@val pair
531 static void append_unique(char *buf, char *prefix, char *key, char *val,
540 anchor = end = strstr(buf, key);
541 /* try to find exact match string in @buf */
542 while (end && *end != '\0' && *end != ',' && *end != ' ' && *end != '=')
545 if (anchor == NULL || strlen(key) != len ||
546 strncmp(anchor, key, len) != 0) {
548 strscat(buf, prefix, maxbuflen);
550 strscat(buf, key, maxbuflen);
552 strscat(buf, "=", maxbuflen);
553 strscat(buf, val, maxbuflen);
558 static int enable_default_ext4_features(struct mkfs_opts *mop, char *anchor,
559 size_t maxbuflen, int user_spec)
561 if (IS_OST(&mop->mo_ldd)) {
562 append_unique(anchor, user_spec ? "," : " -O ",
563 "extents", NULL, sizeof(mop->mo_mkfsopts));
564 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
565 } else if (IS_MDT(&mop->mo_ldd)) {
566 append_unique(anchor, user_spec ? "," : " -O ",
567 "dirdata", NULL, maxbuflen);
568 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
569 append_unique(anchor, ",", "^extents", NULL, maxbuflen);
571 append_unique(anchor, user_spec ? "," : " -O ",
572 "uninit_bg", NULL, maxbuflen);
575 /* Multiple mount protection enabled only if failover node specified */
576 if (mop->mo_flags & MO_FAILOVER) {
577 if (is_e2fsprogs_feature_supp("-O mmp") == 0)
578 append_unique(anchor, ",", "mmp", NULL, maxbuflen);
580 disp_old_e2fsprogs_msg("mmp", 1);
583 /* Allow more than 65000 subdirectories */
584 if (is_e2fsprogs_feature_supp("-O dir_nlink") == 0)
585 append_unique(anchor, ",", "dir_nlink", NULL, maxbuflen);
587 /* The following options are only valid for ext4-based ldiskfs.
588 * If --backfstype=ext3 is specified, do not enable them. */
589 if (mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3)
592 /* Enable quota by default */
593 if (is_e2fsprogs_feature_supp("-O quota") == 0) {
594 append_unique(anchor, ",", "quota", NULL, maxbuflen);
597 fprintf(stderr, "\"-O quota\" must be supported by "
598 "e2fsprogs, please upgrade your e2fsprogs.\n");
602 /* Allow files larger than 2TB. Also needs LU-16, but not harmful. */
603 if (is_e2fsprogs_feature_supp("-O huge_file") == 0)
604 append_unique(anchor, ",", "huge_file", NULL, maxbuflen);
606 /* Enable large block addresses if the LUN is over 2^32 blocks. */
607 if (mop->mo_device_kb / (L_BLOCK_SIZE >> 10) >= 0x100002000ULL &&
608 is_e2fsprogs_feature_supp("-O 64bit") == 0)
609 append_unique(anchor, ",", "64bit", NULL, maxbuflen);
611 /* Cluster inode/block bitmaps and inode table for more efficient IO.
612 * Align the flex groups on a 1MB boundary for better performance. */
613 /* This -O feature needs to go last, since it adds the "-G" option. */
614 if (is_e2fsprogs_feature_supp("-O flex_bg") == 0) {
617 append_unique(anchor, ",", "flex_bg", NULL, maxbuflen);
619 if (IS_OST(&mop->mo_ldd) &&
620 strstr(mop->mo_mkfsopts, "-G") == NULL) {
621 snprintf(tmp_buf, sizeof(tmp_buf), " -G %u",
622 (1 << 20) / L_BLOCK_SIZE);
623 strscat(anchor, tmp_buf, maxbuflen);
626 /* Don't add any more "-O" options here, see last comment above */
631 * moveopts_to_end: find the option string, move remaining strings to
632 * where option string starts, and append the option
634 * @start: where the option string starts before the move
635 * RETURN: where the option string starts after the move
637 static char *moveopts_to_end(char *start)
643 /* skip whitespace before options */
648 /* find end of option characters */
649 while (*end != ' ' && *end != '\0')
653 if (len >= sizeof(save))
654 len = sizeof(save) - 1;
657 strncpy(save, start, len);
660 /* move remaining options up front */
662 memmove(start, end, strlen(end));
663 *(start + strlen(end)) = '\0';
665 /* append the specified options */
666 if (*(start + strlen(start) - 1) != ' ')
668 idx = start + strlen(start);
674 /* Build fs according to type */
675 int ldiskfs_make_lustre(struct mkfs_opts *mop)
677 __u64 device_kb = mop->mo_device_kb, block_count = 0;
678 char mkfs_cmd[PATH_MAX];
682 int ret = 0, ext_opts = 0;
685 if (!(mop->mo_flags & MO_IS_LOOP)) {
686 mop->mo_device_kb = get_device_size(mop->mo_device);
688 if (mop->mo_device_kb == 0)
691 /* Compare to real size */
692 if (device_kb == 0 || device_kb > mop->mo_device_kb)
693 device_kb = mop->mo_device_kb;
695 mop->mo_device_kb = device_kb;
698 if (mop->mo_device_kb != 0) {
699 if (mop->mo_device_kb < 8096) {
700 fprintf(stderr, "%s: size of filesystem must be larger "
701 "than 8MB, but is set to %lldKB\n",
702 progname, (long long)mop->mo_device_kb);
705 block_count = mop->mo_device_kb / (L_BLOCK_SIZE >> 10);
706 /* If the LUN size is just over 2^32 blocks, limit the
707 * filesystem size to 2^32-1 blocks to avoid problems with
708 * ldiskfs/mkfs not handling this size. Bug 22906 */
709 if (block_count > 0xffffffffULL && block_count < 0x100002000ULL)
710 block_count = 0xffffffffULL;
713 if ((mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3) ||
714 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS) ||
715 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS2)) {
718 /* Journal size in MB */
719 if (strstr(mop->mo_mkfsopts, "-J") == NULL &&
720 device_kb > 1024 * 1024) {
721 /* Choose our own default journal size */
722 long journal_mb = 0, max_mb;
724 /* cap journal size at 4GB for MDT,
725 * leave it at 400MB for OSTs. */
726 if (IS_MDT(&mop->mo_ldd))
728 else if (IS_OST(&mop->mo_ldd))
730 else /* Use mke2fs default size for MGS */
733 /* Use at most 4% of device for journal */
734 journal_mb = device_kb * 4 / (1024 * 100);
735 if (journal_mb > max_mb)
739 sprintf(buf, " -J size=%ld", journal_mb);
740 strscat(mop->mo_mkfsopts, buf,
741 sizeof(mop->mo_mkfsopts));
745 /* Inode size includes:
746 * ldiskfs inode size: 156
747 * extended attributes size, including:
748 * ext4_xattr_header: 32
749 * LOV EA size: 32(lov_mds_md) +
750 * stripes * 24(lov_ost_data) +
751 * 16(xattr_entry) + 3(lov)
752 * LMA EA size: 24(lustre_mdt_attrs) +
753 * 16(xattr_entry) + 3(lma)
754 * link EA size: 24(link_ea_header) + 18(link_ea_entry) +
755 * (filename) + 16(xattr_entry) + 4(link)
756 * and some margin for 4-byte alignment, ACLs and other EAs.
758 * If we say the average filename length is about 32 bytes,
759 * the calculation looks like:
760 * 156 + 32 + (32+24*N+19) + (24+19) + (24+18+~32+20) + other <=
761 * 512*2^m, {m=0,1,2,3}
763 if (strstr(mop->mo_mkfsopts, "-I") == NULL) {
764 if (IS_MDT(&mop->mo_ldd)) {
765 if (mop->mo_stripe_count > 69)
766 inode_size = 512; /* bz 7241 */
767 /* see also "-i" below for EA blocks */
768 else if (mop->mo_stripe_count > 26)
770 else if (mop->mo_stripe_count > 5)
774 } else if (IS_OST(&mop->mo_ldd)) {
775 /* We store MDS FID and OST objid in EA on OST
776 * we need to make inode bigger as well. */
780 if (inode_size > 0) {
781 sprintf(buf, " -I %ld", inode_size);
782 strscat(mop->mo_mkfsopts, buf,
783 sizeof(mop->mo_mkfsopts));
787 /* Bytes_per_inode: disk size / num inodes */
788 if (strstr(mop->mo_mkfsopts, "-i") == NULL &&
789 strstr(mop->mo_mkfsopts, "-N") == NULL) {
790 long bytes_per_inode = 0;
792 /* Allocate more inodes on MDT devices. There is
793 * no data stored on the MDT, and very little extra
794 * metadata beyond the inode. It could go down as
795 * low as 1024 bytes, but this is conservative.
796 * Account for external EA blocks for wide striping. */
797 if (IS_MDT(&mop->mo_ldd)) {
798 bytes_per_inode = inode_size + 1536;
800 if (mop->mo_stripe_count > 69) {
801 int extra = mop->mo_stripe_count * 24;
802 extra = ((extra - 1) | 4095) + 1;
803 bytes_per_inode += extra;
807 /* Allocate fewer inodes on large OST devices. Most
808 * filesystems can be much more aggressive than even
809 * this, but it is impossible to know in advance. */
810 if (IS_OST(&mop->mo_ldd)) {
811 /* OST > 16TB assume average file size 1MB */
812 if (device_kb > (16ULL << 30))
813 bytes_per_inode = 1024 * 1024;
814 /* OST > 4TB assume average file size 512kB */
815 else if (device_kb > (4ULL << 30))
816 bytes_per_inode = 512 * 1024;
817 /* OST > 1TB assume average file size 256kB */
818 else if (device_kb > (1ULL << 30))
819 bytes_per_inode = 256 * 1024;
820 /* OST > 10GB assume average file size 64kB,
821 * plus a bit so that inodes will fit into a
822 * 256x flex_bg without overflowing */
823 else if (device_kb > (10ULL << 20))
824 bytes_per_inode = 69905;
827 if (bytes_per_inode > 0) {
828 sprintf(buf, " -i %ld", bytes_per_inode);
829 strscat(mop->mo_mkfsopts, buf,
830 sizeof(mop->mo_mkfsopts));
835 strscat(mop->mo_mkfsopts, " -q",
836 sizeof(mop->mo_mkfsopts));
839 /* start handle -O mkfs options */
840 if ((start = strstr(mop->mo_mkfsopts, "-O")) != NULL) {
841 if (strstr(start + 2, "-O") != NULL) {
843 "%s: don't specify multiple -O options\n",
847 start = moveopts_to_end(start);
848 maxbuflen = sizeof(mop->mo_mkfsopts) -
849 (start - mop->mo_mkfsopts) - strlen(start);
850 ret = enable_default_ext4_features(mop, start, maxbuflen, 1);
852 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts),
853 maxbuflen = sizeof(mop->mo_mkfsopts) -
854 strlen(mop->mo_mkfsopts);
855 ret = enable_default_ext4_features(mop, start, maxbuflen, 0);
859 /* end handle -O mkfs options */
861 /* start handle -E mkfs options */
862 if ((start = strstr(mop->mo_mkfsopts, "-E")) != NULL) {
863 if (strstr(start + 2, "-E") != NULL) {
865 "%s: don't specify multiple -E options\n",
869 start = moveopts_to_end(start);
870 maxbuflen = sizeof(mop->mo_mkfsopts) -
871 (start - mop->mo_mkfsopts) - strlen(start);
874 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts);
875 maxbuflen = sizeof(mop->mo_mkfsopts) -
876 strlen(mop->mo_mkfsopts);
879 /* In order to align the filesystem metadata on 1MB boundaries,
880 * give a resize value that will reserve a power-of-two group
881 * descriptor blocks, but leave one block for the superblock.
882 * Only useful for filesystems with < 2^32 blocks due to resize
884 if (IS_OST(&mop->mo_ldd) && mop->mo_device_kb > 100 * 1024 &&
885 mop->mo_device_kb * 1024 / L_BLOCK_SIZE <= 0xffffffffULL) {
886 unsigned group_blocks = L_BLOCK_SIZE * 8;
887 unsigned desc_per_block = L_BLOCK_SIZE / 32;
888 unsigned resize_blks;
890 resize_blks = (1ULL<<32) - desc_per_block*group_blocks;
891 snprintf(buf, sizeof(buf), "%u", resize_blks);
892 append_unique(start, ext_opts ? "," : " -E ",
893 "resize", buf, maxbuflen);
897 /* Avoid zeroing out the full journal - speeds up mkfs */
898 if (is_e2fsprogs_feature_supp("-E lazy_journal_init") == 0)
899 append_unique(start, ext_opts ? "," : " -E ",
900 "lazy_journal_init", NULL, maxbuflen);
901 /* end handle -E mkfs options */
903 /* Allow reformat of full devices (as opposed to
904 partitions.) We already checked for mounted dev. */
905 strscat(mop->mo_mkfsopts, " -F", sizeof(mop->mo_mkfsopts));
907 snprintf(mkfs_cmd, sizeof(mkfs_cmd),
908 "%s -j -b %d -L %s ", MKE2FS, L_BLOCK_SIZE,
909 mop->mo_ldd.ldd_svname);
911 fprintf(stderr,"%s: unsupported fs type: %d (%s)\n",
912 progname, mop->mo_ldd.ldd_mount_type,
913 MT_STR(&mop->mo_ldd));
917 /* For loop device format the dev, not the filename */
918 dev = mop->mo_device;
919 if (mop->mo_flags & MO_IS_LOOP)
920 dev = mop->mo_loopdev;
922 vprint("formatting backing filesystem %s on %s\n",
923 MT_STR(&mop->mo_ldd), dev);
924 vprint("\ttarget name %s\n", mop->mo_ldd.ldd_svname);
925 vprint("\t4k blocks "LPU64"\n", block_count);
926 vprint("\toptions %s\n", mop->mo_mkfsopts);
928 /* mkfs_cmd's trailing space is important! */
929 strscat(mkfs_cmd, mop->mo_mkfsopts, sizeof(mkfs_cmd));
930 strscat(mkfs_cmd, " ", sizeof(mkfs_cmd));
931 strscat(mkfs_cmd, dev, sizeof(mkfs_cmd));
932 if (block_count != 0) {
933 sprintf(buf, " "LPU64, block_count);
934 strscat(mkfs_cmd, buf, sizeof(mkfs_cmd));
937 vprint("mkfs_cmd = %s\n", mkfs_cmd);
938 ret = run_command(mkfs_cmd, sizeof(mkfs_cmd));
941 fprintf(stderr, "Unable to build fs %s (%d)\n", dev, ret);
946 int ldiskfs_prepare_lustre(struct mkfs_opts *mop,
947 char *default_mountopts, int default_len,
948 char *always_mountopts, int always_len)
950 struct lustre_disk_data *ldd = &mop->mo_ldd;
953 /* Set MO_IS_LOOP to indicate a loopback device is needed */
954 ret = is_block(mop->mo_device);
957 } else if (ret == 0) {
958 mop->mo_flags |= MO_IS_LOOP;
961 strscat(default_mountopts, ",errors=remount-ro", default_len);
962 if (IS_MDT(ldd) || IS_MGS(ldd))
963 strscat(always_mountopts, ",user_xattr", always_len);
968 static int read_file(const char *path, char *buf, int size)
972 fd = fopen(path, "r");
976 if (fgets(buf, size, fd) == NULL) {
977 fprintf(stderr, "reading from %s: %s", path, strerror(errno));
983 /* strip trailing newline */
985 if (buf[size - 1] == '\n')
986 buf[size - 1] = '\0';
991 static int write_file(const char *path, const char *buf)
995 fd = open(path, O_WRONLY);
999 rc = write(fd, buf, strlen(buf));
1002 return rc < 0 ? errno : 0;
1005 static int set_blockdev_scheduler(const char *path, const char *scheduler)
1007 char buf[PATH_MAX], *s, *e, orig_sched[50];
1010 /* Before setting the scheduler, we need to check to see if it's
1011 * already set to "noop". If it is, we don't want to override
1012 * that setting. If it's set to anything other than "noop", set
1013 * the scheduler to what has been passed in. */
1015 rc = read_file(path, buf, sizeof(buf));
1018 fprintf(stderr, "%s: cannot open '%s': %s\n",
1019 progname, path, strerror(errno));
1023 /* The expected format of buf: noop anticipatory deadline [cfq] */
1024 s = strchr(buf, '[');
1025 e = strchr(buf, ']');
1027 /* If the format is not what we expect. Play it safe and error out. */
1028 if (s == NULL || e == NULL) {
1030 fprintf(stderr, "%s: cannot parse scheduler "
1031 "options for '%s'\n", progname, path);
1035 snprintf(orig_sched, e - s, "%s", s + 1);
1037 if (strcmp(orig_sched, "noop") == 0 ||
1038 strcmp(orig_sched, scheduler) == 0)
1041 rc = write_file(path, scheduler);
1044 fprintf(stderr, "%s: cannot set scheduler on "
1045 "'%s': %s\n", progname, path,
1049 fprintf(stderr, "%s: change scheduler of %s from %s to %s\n",
1050 progname, path, orig_sched, scheduler);
1056 /* This is to tune the kernel for good SCSI performance.
1057 * For that we set the value of /sys/block/{dev}/queue/max_sectors_kb
1058 * to the value of /sys/block/{dev}/queue/max_hw_sectors_kb */
1059 static int set_blockdev_tunables(char *source, struct mount_opts *mop)
1061 glob_t glob_info = { 0 };
1062 struct stat stat_buf;
1063 char *chk_major, *chk_minor;
1064 char *savept = NULL, *dev;
1066 char buf[PATH_MAX] = {'\0'}, path[PATH_MAX] = {'\0'};
1067 char real_path[PATH_MAX] = {'\0'};
1075 ret_path = realpath(source, real_path);
1076 if (ret_path == NULL) {
1078 fprintf(stderr, "warning: %s: cannot resolve: %s\n",
1079 source, strerror(errno));
1083 if (strncmp(real_path, "/dev/loop", 9) == 0)
1086 if ((real_path[0] != '/') && (strpbrk(real_path, ",:") != NULL))
1089 snprintf(path, sizeof(path), "/sys/block%s", real_path + 4);
1090 if (access(path, X_OK) == 0)
1093 /* The name of the device say 'X' specified in /dev/X may not
1094 * match any entry under /sys/block/. In that case we need to
1095 * match the major/minor number to find the entry under
1096 * sys/block corresponding to /dev/X */
1098 /* Don't chop tail digit on /dev/mapper/xxx, LU-478 */
1099 if (strncmp(real_path, "/dev/mapper", 11) != 0) {
1100 dev = real_path + strlen(real_path);
1101 while (--dev > real_path && isdigit(*dev))
1104 if (strncmp(real_path, "/dev/md_", 8) == 0)
1108 rc = stat(real_path, &stat_buf);
1111 fprintf(stderr, "warning: %s, device %s stat failed\n",
1112 strerror(errno), real_path);
1116 major = major(stat_buf.st_rdev);
1117 minor = minor(stat_buf.st_rdev);
1118 rc = glob("/sys/block/*", GLOB_NOSORT, NULL, &glob_info);
1121 fprintf(stderr, "warning: failed to read entries under "
1123 globfree(&glob_info);
1127 for (i = 0; i < glob_info.gl_pathc; i++){
1128 snprintf(path, sizeof(path), "%s/dev", glob_info.gl_pathv[i]);
1130 rc = read_file(path, buf, sizeof(buf));
1134 if (buf[strlen(buf) - 1] == '\n')
1135 buf[strlen(buf) - 1] = '\0';
1137 chk_major = strtok_r(buf, ":", &savept);
1139 if (chk_major != NULL && major == atoi(chk_major) &&
1140 chk_minor != NULL && minor == atoi(chk_minor))
1144 if (i == glob_info.gl_pathc) {
1146 fprintf(stderr,"warning: device %s does not match any "
1147 "entry under /sys/block\n", real_path);
1148 globfree(&glob_info);
1152 /* Chop off "/dev" from path we found */
1153 path[strlen(glob_info.gl_pathv[i])] = '\0';
1154 globfree(&glob_info);
1157 if (strncmp(real_path, "/dev/md", 7) == 0) {
1158 snprintf(real_path, sizeof(real_path), "%s/%s", path,
1161 rc = read_file(real_path, buf, sizeof(buf));
1164 fprintf(stderr, "warning: opening %s: %s\n",
1165 real_path, strerror(errno));
1169 if (atoi(buf) >= mop->mo_md_stripe_cache_size)
1172 if (strlen(buf) - 1 > 0) {
1173 snprintf(buf, sizeof(buf), "%d",
1174 mop->mo_md_stripe_cache_size);
1175 rc = write_file(real_path, buf);
1176 if (rc != 0 && verbose)
1177 fprintf(stderr, "warning: opening %s: %s\n",
1178 real_path, strerror(errno));
1180 /* Return since raid and disk tunables are different */
1184 snprintf(real_path, sizeof(real_path), "%s/%s", path,
1185 MAX_HW_SECTORS_KB_PATH);
1186 rc = read_file(real_path, buf, sizeof(buf));
1189 fprintf(stderr, "warning: opening %s: %s\n",
1190 real_path, strerror(errno));
1191 /* No MAX_HW_SECTORS_KB_PATH isn't necessary an
1192 * error for some device. */
1196 if (strlen(buf) - 1 > 0) {
1197 char oldbuf[32] = "", *end = NULL;
1198 unsigned long long oldval, newval;
1200 snprintf(real_path, sizeof(real_path), "%s/%s", path,
1201 MAX_SECTORS_KB_PATH);
1202 rc = read_file(real_path, oldbuf, sizeof(oldbuf));
1203 /* Only set new parameter if different from the old one. */
1204 if (rc != 0 || strcmp(oldbuf, buf) == 0) {
1205 /* No MAX_SECTORS_KB_PATH isn't necessary an
1206 * error for some device. */
1210 newval = strtoull(buf, &end, 0);
1211 if (newval == 0 || newval == ULLONG_MAX || end == buf)
1214 /* Don't increase IO request size limit past 16MB. It is about
1215 * PTLRPC_MAX_BRW_SIZE, but that isn't in a public header.
1216 * Note that even though the block layer allows larger values,
1217 * setting max_sectors_kb = 32768 causes crashes (LU-6974). */
1218 if (newval > 16 * 1024) {
1220 snprintf(buf, sizeof(buf), "%llu", newval);
1223 oldval = strtoull(oldbuf, &end, 0);
1224 /* Don't shrink the current limit. */
1225 if (oldval != ULLONG_MAX && newval <= oldval)
1228 rc = write_file(real_path, buf);
1231 fprintf(stderr, "warning: writing to %s: %s\n",
1232 real_path, strerror(errno));
1233 /* No MAX_SECTORS_KB_PATH isn't necessary an
1234 * error for some device. */
1237 fprintf(stderr, "%s: increased %s from %s to %s\n",
1238 progname, real_path, oldbuf, buf);
1242 /* Purposely ignore errors reported from set_blockdev_scheduler.
1243 * The worst that will happen is a block device with an "incorrect"
1245 snprintf(real_path, sizeof(real_path), "%s/%s", path, SCHEDULER_PATH);
1246 set_blockdev_scheduler(real_path, DEFAULT_SCHEDULER);
1248 /* if device is multipath device, tune its slave devices */
1249 glob_info.gl_pathc = 0;
1250 glob_info.gl_offs = 0;
1251 snprintf(real_path, sizeof(real_path), "%s/slaves/*", path);
1252 rc = glob(real_path, GLOB_NOSORT, NULL, &glob_info);
1254 for (i = 0; rc == 0 && i < glob_info.gl_pathc; i++) {
1255 slave = basename(glob_info.gl_pathv[i]);
1256 snprintf(real_path, sizeof(real_path), "/dev/%s", slave);
1257 rc = set_blockdev_tunables(real_path, mop);
1260 if (rc == GLOB_NOMATCH) {
1261 /* no slave device is not an error */
1263 } else if (rc && verbose) {
1264 if (slave == NULL) {
1265 fprintf(stderr, "warning: %s, failed to read"
1266 " entries under %s/slaves\n",
1267 strerror(errno), path);
1269 fprintf(stderr, "unable to set tunables for"
1270 " slave device %s (slave would be"
1271 " unable to handle IO request from"
1276 globfree(&glob_info);
1281 int ldiskfs_tune_lustre(char *dev, struct mount_opts *mop)
1283 return set_blockdev_tunables(dev, mop);
1286 int ldiskfs_label_lustre(struct mount_opts *mop)
1288 char label_cmd[PATH_MAX];
1291 snprintf(label_cmd, sizeof(label_cmd),
1292 TUNE2FS" -f -L '%s' '%s' >/dev/null 2>&1",
1293 mop->mo_ldd.ldd_svname, mop->mo_source);
1294 rc = run_command(label_cmd, sizeof(label_cmd));
1299 /* Enable quota accounting */
1300 int ldiskfs_enable_quota(struct mkfs_opts *mop)
1304 int cmdsz = sizeof(cmd), ret;
1306 if (is_e2fsprogs_feature_supp("-O quota") != 0) {
1307 fprintf(stderr, "%s: \"-O quota\" is is not supported by "
1308 "current e2fsprogs\n", progname);
1312 dev = mop->mo_device;
1313 if (mop->mo_flags & MO_IS_LOOP)
1314 dev = mop->mo_loopdev;
1316 /* Quota feature is already enabled? */
1317 if (is_feature_enabled("quota", dev)) {
1318 vprint("Quota feature is already enabled.\n");
1322 /* Turn on quota feature by "tune2fs -O quota" */
1323 snprintf(cmd, cmdsz, "%s -O quota %s", TUNE2FS, dev);
1324 ret = run_command(cmd, cmdsz);
1326 fprintf(stderr, "command:%s (%d)", cmd, ret);
1331 int ldiskfs_init(void)
1333 /* Required because full path to DEBUGFS is not specified */
1334 setenv("PATH", "/bin:/sbin:/usr/bin:/usr/sbin:/usr/local/bin", 0);
1339 void ldiskfs_fini(void)