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
96 static void append_unique(char *buf, char *prefix, char *key, char *val,
100 * Concatenate context of the temporary mount point if selinux is enabled
103 static void append_context_for_mount(char *mntpt, struct mkfs_opts *mop)
105 security_context_t fcontext;
107 if (getfilecon(mntpt, &fcontext) < 0) {
108 /* Continuing with default behaviour */
109 fprintf(stderr, "%s: Get file context failed : %s\n",
110 progname, strerror(errno));
114 if (fcontext != NULL) {
115 append_unique(mop->mo_ldd.ldd_mount_opts,
116 ",", "context", fcontext,
117 sizeof(mop->mo_ldd.ldd_mount_opts));
123 /* return canonicalized absolute pathname, even if the target file does not
124 * exist, unlike realpath */
125 static char *absolute_path(char *devname)
127 char buf[PATH_MAX + 1] = "";
132 path = malloc(sizeof(buf));
136 if (devname[0] != '/') {
137 if (getcwd(buf, sizeof(buf) - 1) == NULL) {
141 len = snprintf(path, sizeof(buf), "%s/%s", buf, devname);
142 if (len >= sizeof(buf)) {
147 len = snprintf(path, sizeof(buf), "%s", devname);
148 if (len >= sizeof(buf)) {
154 /* truncate filename before calling realpath */
155 ptr = strrchr(path, '/');
161 if (buf != realpath(path, buf)) {
165 /* add the filename back */
166 len = snprintf(path, PATH_MAX, "%s/%s", buf, ptr+1);
167 if (len >= PATH_MAX) {
174 /* Determine if a device is a block device (as opposed to a file) */
175 static int is_block(char *devname)
181 devpath = absolute_path(devname);
182 if (devpath == NULL) {
183 fprintf(stderr, "%s: failed to resolve path to %s\n",
188 ret = access(devname, F_OK);
190 if (strncmp(devpath, "/dev/", 5) == 0) {
191 /* nobody sane wants to create a loopback file under
192 * /dev. Let's just report the device doesn't exist */
193 fprintf(stderr, "%s: %s apparently does not exist\n",
201 ret = stat(devpath, &st);
203 fprintf(stderr, "%s: cannot stat %s\n", progname, devpath);
206 ret = S_ISBLK(st.st_mode);
212 static int is_feature_enabled(const char *feature, const char *devpath)
216 char enabled_features[4096] = "";
219 snprintf(cmd, sizeof(cmd), "%s -R features %s 2>&1",
222 /* Using popen() instead of run_command() since debugfs does
223 * not return proper error code if command is not supported */
224 fp = popen(cmd, "r");
226 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
230 ret = fread(enabled_features, 1, sizeof(enabled_features) - 1, fp);
231 enabled_features[ret] = '\0';
234 if (strstr(enabled_features, feature))
239 /* Write the server config files */
240 int ldiskfs_write_ldd(struct mkfs_opts *mop)
242 char mntpt[] = "/tmp/mntXXXXXX";
249 /* Mount this device temporarily in order to write these files */
250 if (!mkdtemp(mntpt)) {
251 fprintf(stderr, "%s: Can't create temp mount point %s: %s\n",
252 progname, mntpt, strerror(errno));
257 * Append file context to mount options if SE Linux is enabled
260 if (is_selinux_enabled() > 0)
261 append_context_for_mount(mntpt, mop);
264 dev = mop->mo_device;
265 if (mop->mo_flags & MO_IS_LOOP)
266 dev = mop->mo_loopdev;
268 ret = mount(dev, mntpt, MT_STR(&mop->mo_ldd), 0,
269 (mop->mo_mountopts == NULL) ?
270 "errors=remount-ro" : mop->mo_mountopts);
272 fprintf(stderr, "%s: Unable to mount %s: %s\n",
273 progname, dev, strerror(errno));
275 if (errno == ENODEV) {
276 fprintf(stderr, "Is the %s module available?\n",
277 MT_STR(&mop->mo_ldd));
282 /* Set up initial directories */
283 sprintf(filepnm, "%s/%s", mntpt, MOUNT_CONFIGS_DIR);
284 ret = mkdir(filepnm, 0777);
285 if ((ret != 0) && (errno != EEXIST)) {
286 fprintf(stderr, "%s: Can't make configs dir %s (%s)\n",
287 progname, filepnm, strerror(errno));
289 } else if (errno == EEXIST) {
293 /* Save the persistent mount data into a file. Lustre must pre-read
294 this file to get the real mount options. */
295 vprint("Writing %s\n", MOUNT_DATA_FILE);
296 sprintf(filepnm, "%s/%s", mntpt, MOUNT_DATA_FILE);
297 filep = fopen(filepnm, "w");
299 fprintf(stderr, "%s: Unable to create %s file: %s\n",
300 progname, filepnm, strerror(errno));
303 num = fwrite(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
304 if (num < 1 && ferror(filep)) {
305 fprintf(stderr, "%s: Unable to write to file (%s): %s\n",
306 progname, filepnm, strerror(errno));
319 static int readcmd(char *cmd, char *buf, int len)
324 fp = popen(cmd, "r");
328 red = fread(buf, 1, len, fp);
331 /* strip trailing newline */
332 if (buf[red - 1] == '\n')
335 return (red == 0) ? -ENOENT : 0;
338 int ldiskfs_read_ldd(char *dev, struct lustre_disk_data *mo_ldd)
340 char tmpdir[] = "/tmp/dirXXXXXX";
345 int cmdsz = sizeof(cmd);
347 /* Make a temporary directory to hold Lustre data files. */
348 if (!mkdtemp(tmpdir)) {
349 fprintf(stderr, "%s: Can't create temporary directory %s: %s\n",
350 progname, tmpdir, strerror(errno));
354 /* TODO: it's worth observing the get_mountdata() function that is
355 in mount_utils.c for getting the mountdata out of the
358 /* Construct debugfs command line. */
359 snprintf(cmd, cmdsz, "%s -c -R 'dump /%s %s/mountdata' '%s'",
360 DEBUGFS, MOUNT_DATA_FILE, tmpdir, dev);
362 ret = run_command(cmd, cmdsz);
364 verrprint("%s: Unable to dump %s dir (%d)\n",
365 progname, MOUNT_CONFIGS_DIR, ret);
367 sprintf(filepnm, "%s/mountdata", tmpdir);
368 filep = fopen(filepnm, "r");
371 vprint("Reading %s\n", MOUNT_DATA_FILE);
372 num_read = fread(mo_ldd, sizeof(*mo_ldd), 1, filep);
373 if (num_read < 1 && ferror(filep)) {
374 fprintf(stderr, "%s: Unable to read from file %s: %s\n",
375 progname, filepnm, strerror(errno));
380 snprintf(cmd, cmdsz, "rm -rf %s", tmpdir);
381 run_command(cmd, cmdsz);
383 verrprint("Failed to read old data (%d)\n", ret);
385 /* As long as we at least have the label, we're good to go */
386 snprintf(cmd, sizeof(cmd), E2LABEL" %s", dev);
387 ret = readcmd(cmd, mo_ldd->ldd_svname, sizeof(mo_ldd->ldd_svname) - 1);
393 /* Display the need for the latest e2fsprogs to be installed. make_backfs
394 * indicates if the caller is make_lustre_backfs() or not. */
395 static void disp_old_e2fsprogs_msg(const char *feature, int make_backfs)
397 static int msg_displayed;
400 fprintf(stderr, "WARNING: %s does not support %s "
401 "feature.\n\n", E2FSPROGS, feature);
407 fprintf(stderr, "WARNING: The %s package currently installed on "
408 "your system does not support \"%s\" feature.\n",
410 #if !(HAVE_LDISKFSPROGS)
411 fprintf(stderr, "Please install the latest version of e2fsprogs from\n"
412 "https://downloads.hpdd.intel.com/public/e2fsprogs/latest/\n"
413 "to enable this feature.\n");
416 fprintf(stderr, "Feature will not be enabled until %s"
417 "is updated and '%s -O %s %%{device}' "
418 "is run.\n\n", E2FSPROGS, TUNE2FS, feature);
421 /* Check whether the file exists in the device */
422 static int file_in_dev(char *file_name, char *dev_name)
425 char debugfs_cmd[256];
426 unsigned int inode_num;
429 /* Construct debugfs command line. */
430 snprintf(debugfs_cmd, sizeof(debugfs_cmd),
431 "%s -c -R 'stat %s' '%s' 2>&1 | egrep '(Inode|unsupported)'",
432 DEBUGFS, file_name, dev_name);
434 fp = popen(debugfs_cmd, "r");
436 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
440 if (fscanf(fp, "Inode: %u", &inode_num) == 1) { /* exist */
444 i = fread(debugfs_cmd, 1, sizeof(debugfs_cmd) - 1, fp);
447 fprintf(stderr, "%s", debugfs_cmd);
448 if (strstr(debugfs_cmd, "unsupported feature")) {
449 disp_old_e2fsprogs_msg("an unknown", 0);
458 /* Check whether the device has already been used with lustre */
459 int ldiskfs_is_lustre(char *dev, unsigned *mount_type)
463 ret = file_in_dev(MOUNT_DATA_FILE, dev);
465 /* in the -1 case, 'extents' means IS a lustre target */
466 *mount_type = LDD_MT_LDISKFS;
470 ret = file_in_dev(LAST_RCVD, dev);
472 *mount_type = LDD_MT_LDISKFS;
479 /* Check if a certain feature is supported by e2fsprogs.
480 * Firstly we try to use "debugfs supported_features" command to check if
481 * the feature is supported. If this fails we try to set this feature with
482 * mke2fs to check for its support. */
483 static int is_e2fsprogs_feature_supp(const char *feature)
485 static char supp_features[4096] = "";
488 char imgname[] = "/tmp/test-img-XXXXXX";
492 if (supp_features[0] == '\0') {
493 snprintf(cmd, sizeof(cmd), "%s -c -R supported_features 2>&1",
496 /* Using popen() instead of run_command() since debugfs does
497 * not return proper error code if command is not supported */
498 fp = popen(cmd, "r");
500 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
503 ret = fread(supp_features, 1, sizeof(supp_features) - 1, fp);
504 supp_features[ret] = '\0';
507 if (ret > 0 && strstr(supp_features,
508 strncmp(feature, "-O ", 3) ? feature : feature+3))
511 if ((fd = mkstemp(imgname)) < 0)
516 snprintf(cmd, sizeof(cmd), "%s -F %s %s 100 >/dev/null 2>&1",
517 MKE2FS, feature, imgname);
518 /* run_command() displays the output of mke2fs when it fails for
519 * some feature, so use system() directly */
528 * append_unique: append @key or @key=@val pair to @buf only if @key does not
530 * @buf: buffer to hold @key or @key=@val
531 * @prefix: prefix string before @key
533 * @val: value string if it's a @key=@val pair
535 static void append_unique(char *buf, char *prefix, char *key, char *val,
544 anchor = end = strstr(buf, key);
545 /* try to find exact match string in @buf */
546 while (end && *end != '\0' && *end != ',' && *end != ' ' && *end != '=')
549 if (anchor == NULL || strlen(key) != len ||
550 strncmp(anchor, key, len) != 0) {
552 strscat(buf, prefix, maxbuflen);
554 strscat(buf, key, maxbuflen);
556 strscat(buf, "=", maxbuflen);
557 strscat(buf, val, maxbuflen);
562 static int enable_default_ext4_features(struct mkfs_opts *mop, char *anchor,
563 size_t maxbuflen, int user_spec)
565 if (IS_OST(&mop->mo_ldd)) {
566 append_unique(anchor, user_spec ? "," : " -O ",
567 "extents", NULL, sizeof(mop->mo_mkfsopts));
568 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
569 } else if (IS_MDT(&mop->mo_ldd)) {
570 append_unique(anchor, user_spec ? "," : " -O ",
571 "dirdata", NULL, maxbuflen);
572 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
573 append_unique(anchor, ",", "^extents", NULL, maxbuflen);
575 append_unique(anchor, user_spec ? "," : " -O ",
576 "uninit_bg", NULL, maxbuflen);
579 /* Multiple mount protection enabled only if failover node specified */
580 if (mop->mo_flags & MO_FAILOVER) {
581 if (is_e2fsprogs_feature_supp("-O mmp") == 0)
582 append_unique(anchor, ",", "mmp", NULL, maxbuflen);
584 disp_old_e2fsprogs_msg("mmp", 1);
587 /* Allow more than 65000 subdirectories */
588 if (is_e2fsprogs_feature_supp("-O dir_nlink") == 0)
589 append_unique(anchor, ",", "dir_nlink", NULL, maxbuflen);
591 /* The following options are only valid for ext4-based ldiskfs.
592 * If --backfstype=ext3 is specified, do not enable them. */
593 if (mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3)
596 /* Enable quota by default */
597 if (is_e2fsprogs_feature_supp("-O quota") == 0) {
598 append_unique(anchor, ",", "quota", NULL, maxbuflen);
601 fprintf(stderr, "\"-O quota\" must be supported by "
602 "e2fsprogs, please upgrade your e2fsprogs.\n");
606 /* Allow files larger than 2TB. Also needs LU-16, but not harmful. */
607 if (is_e2fsprogs_feature_supp("-O huge_file") == 0)
608 append_unique(anchor, ",", "huge_file", NULL, maxbuflen);
610 /* Enable large block addresses if the LUN is over 2^32 blocks. */
611 if (mop->mo_device_kb / (L_BLOCK_SIZE >> 10) >= 0x100002000ULL &&
612 is_e2fsprogs_feature_supp("-O 64bit") == 0)
613 append_unique(anchor, ",", "64bit", NULL, maxbuflen);
615 /* Cluster inode/block bitmaps and inode table for more efficient IO.
616 * Align the flex groups on a 1MB boundary for better performance. */
617 /* This -O feature needs to go last, since it adds the "-G" option. */
618 if (is_e2fsprogs_feature_supp("-O flex_bg") == 0) {
621 append_unique(anchor, ",", "flex_bg", NULL, maxbuflen);
623 if (IS_OST(&mop->mo_ldd) &&
624 strstr(mop->mo_mkfsopts, "-G") == NULL) {
625 snprintf(tmp_buf, sizeof(tmp_buf), " -G %u",
626 (1 << 20) / L_BLOCK_SIZE);
627 strscat(anchor, tmp_buf, maxbuflen);
630 /* Don't add any more "-O" options here, see last comment above */
635 * moveopts_to_end: find the option string, move remaining strings to
636 * where option string starts, and append the option
638 * @start: where the option string starts before the move
639 * RETURN: where the option string starts after the move
641 static char *moveopts_to_end(char *start)
647 /* skip whitespace before options */
652 /* find end of option characters */
653 while (*end != ' ' && *end != '\0')
657 if (len >= sizeof(save))
658 len = sizeof(save) - 1;
661 strncpy(save, start, len);
664 /* move remaining options up front */
666 memmove(start, end, strlen(end));
667 *(start + strlen(end)) = '\0';
669 /* append the specified options */
670 if (*(start + strlen(start) - 1) != ' ')
672 idx = start + strlen(start);
678 /* Build fs according to type */
679 int ldiskfs_make_lustre(struct mkfs_opts *mop)
681 __u64 device_kb = mop->mo_device_kb, block_count = 0;
682 char mkfs_cmd[PATH_MAX];
686 int ret = 0, ext_opts = 0;
689 if (!(mop->mo_flags & MO_IS_LOOP)) {
690 mop->mo_device_kb = get_device_size(mop->mo_device);
692 if (mop->mo_device_kb == 0)
695 /* Compare to real size */
696 if (device_kb == 0 || device_kb > mop->mo_device_kb)
697 device_kb = mop->mo_device_kb;
699 mop->mo_device_kb = device_kb;
702 if (mop->mo_device_kb != 0) {
703 if (mop->mo_device_kb < 32384) {
704 fprintf(stderr, "%s: size of filesystem must be larger "
705 "than 32MB, but is set to %lldKB\n",
706 progname, (long long)mop->mo_device_kb);
709 block_count = mop->mo_device_kb / (L_BLOCK_SIZE >> 10);
710 /* If the LUN size is just over 2^32 blocks, limit the
711 * filesystem size to 2^32-1 blocks to avoid problems with
712 * ldiskfs/mkfs not handling this size. Bug 22906 */
713 if (block_count > 0xffffffffULL && block_count < 0x100002000ULL)
714 block_count = 0xffffffffULL;
717 if ((mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3) ||
718 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS) ||
719 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS2)) {
722 /* Journal size in MB */
723 if (strstr(mop->mo_mkfsopts, "-J") == NULL &&
724 device_kb > 1024 * 1024) {
725 /* Choose our own default journal size */
726 long journal_mb = 0, max_mb;
728 /* cap journal size at 4GB for MDT,
729 * leave it at 400MB for OSTs. */
730 if (IS_MDT(&mop->mo_ldd))
732 else if (IS_OST(&mop->mo_ldd))
734 else /* Use mke2fs default size for MGS */
737 /* Use at most 4% of device for journal */
738 journal_mb = device_kb * 4 / (1024 * 100);
739 if (journal_mb > max_mb)
743 sprintf(buf, " -J size=%ld", journal_mb);
744 strscat(mop->mo_mkfsopts, buf,
745 sizeof(mop->mo_mkfsopts));
749 /* Inode size includes:
750 * ldiskfs inode size: 156
751 * extended attributes size, including:
752 * ext4_xattr_header: 32
753 * LOV EA size: 32(lov_mds_md) +
754 * stripes * 24(lov_ost_data) +
755 * 16(xattr_entry) + 3(lov)
756 * LMA EA size: 24(lustre_mdt_attrs) +
757 * 16(xattr_entry) + 3(lma)
758 * link EA size: 24(link_ea_header) + 18(link_ea_entry) +
759 * (filename) + 16(xattr_entry) + 4(link)
760 * and some margin for 4-byte alignment, ACLs and other EAs.
762 * If we say the average filename length is about 32 bytes,
763 * the calculation looks like:
764 * 156 + 32 + (32+24*N+19) + (24+19) + (24+18+~32+20) + other <=
765 * 512*2^m, {m=0,1,2,3}
767 if (strstr(mop->mo_mkfsopts, "-I") == NULL) {
768 if (IS_MDT(&mop->mo_ldd)) {
769 if (mop->mo_stripe_count > 69)
770 inode_size = 512; /* bz 7241 */
771 /* see also "-i" below for EA blocks */
772 else if (mop->mo_stripe_count > 26)
774 else if (mop->mo_stripe_count > 5)
778 } else if (IS_OST(&mop->mo_ldd)) {
779 /* We store MDS FID and OST objid in EA on OST
780 * we need to make inode bigger as well. */
784 if (inode_size > 0) {
785 sprintf(buf, " -I %ld", inode_size);
786 strscat(mop->mo_mkfsopts, buf,
787 sizeof(mop->mo_mkfsopts));
791 /* Bytes_per_inode: disk size / num inodes */
792 if (strstr(mop->mo_mkfsopts, "-i") == NULL &&
793 strstr(mop->mo_mkfsopts, "-N") == NULL) {
794 long bytes_per_inode = 0;
796 /* Allocate more inodes on MDT devices. There is
797 * no data stored on the MDT, and very little extra
798 * metadata beyond the inode. It could go down as
799 * low as 1024 bytes, but this is conservative.
800 * Account for external EA blocks for wide striping. */
801 if (IS_MDT(&mop->mo_ldd)) {
802 bytes_per_inode = inode_size + 1536;
804 if (mop->mo_stripe_count > 69) {
805 int extra = mop->mo_stripe_count * 24;
806 extra = ((extra - 1) | 4095) + 1;
807 bytes_per_inode += extra;
811 /* Allocate fewer inodes on large OST devices. Most
812 * filesystems can be much more aggressive than even
813 * this, but it is impossible to know in advance. */
814 if (IS_OST(&mop->mo_ldd)) {
815 /* OST > 16TB assume average file size 1MB */
816 if (device_kb > (16ULL << 30))
817 bytes_per_inode = 1024 * 1024;
818 /* OST > 4TB assume average file size 512kB */
819 else if (device_kb > (4ULL << 30))
820 bytes_per_inode = 512 * 1024;
821 /* OST > 1TB assume average file size 256kB */
822 else if (device_kb > (1ULL << 30))
823 bytes_per_inode = 256 * 1024;
824 /* OST > 10GB assume average file size 64kB,
825 * plus a bit so that inodes will fit into a
826 * 256x flex_bg without overflowing */
827 else if (device_kb > (10ULL << 20))
828 bytes_per_inode = 69905;
831 if (bytes_per_inode > 0) {
832 sprintf(buf, " -i %ld", bytes_per_inode);
833 strscat(mop->mo_mkfsopts, buf,
834 sizeof(mop->mo_mkfsopts));
839 strscat(mop->mo_mkfsopts, " -q",
840 sizeof(mop->mo_mkfsopts));
843 /* start handle -O mkfs options */
844 if ((start = strstr(mop->mo_mkfsopts, "-O")) != NULL) {
845 if (strstr(start + 2, "-O") != NULL) {
847 "%s: don't specify multiple -O options\n",
851 start = moveopts_to_end(start);
852 maxbuflen = sizeof(mop->mo_mkfsopts) -
853 (start - mop->mo_mkfsopts) - strlen(start);
854 ret = enable_default_ext4_features(mop, start, maxbuflen, 1);
856 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts),
857 maxbuflen = sizeof(mop->mo_mkfsopts) -
858 strlen(mop->mo_mkfsopts);
859 ret = enable_default_ext4_features(mop, start, maxbuflen, 0);
863 /* end handle -O mkfs options */
865 /* start handle -E mkfs options */
866 if ((start = strstr(mop->mo_mkfsopts, "-E")) != NULL) {
867 if (strstr(start + 2, "-E") != NULL) {
869 "%s: don't specify multiple -E options\n",
873 start = moveopts_to_end(start);
874 maxbuflen = sizeof(mop->mo_mkfsopts) -
875 (start - mop->mo_mkfsopts) - strlen(start);
878 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts);
879 maxbuflen = sizeof(mop->mo_mkfsopts) -
880 strlen(mop->mo_mkfsopts);
883 /* In order to align the filesystem metadata on 1MB boundaries,
884 * give a resize value that will reserve a power-of-two group
885 * descriptor blocks, but leave one block for the superblock.
886 * Only useful for filesystems with < 2^32 blocks due to resize
888 if (IS_OST(&mop->mo_ldd) && mop->mo_device_kb > 100 * 1024 &&
889 mop->mo_device_kb * 1024 / L_BLOCK_SIZE <= 0xffffffffULL) {
890 unsigned group_blocks = L_BLOCK_SIZE * 8;
891 unsigned desc_per_block = L_BLOCK_SIZE / 32;
892 unsigned resize_blks;
894 resize_blks = (1ULL<<32) - desc_per_block*group_blocks;
895 snprintf(buf, sizeof(buf), "%u", resize_blks);
896 append_unique(start, ext_opts ? "," : " -E ",
897 "resize", buf, maxbuflen);
901 /* Avoid zeroing out the full journal - speeds up mkfs */
902 if (is_e2fsprogs_feature_supp("-E lazy_journal_init") == 0)
903 append_unique(start, ext_opts ? "," : " -E ",
904 "lazy_journal_init", NULL, maxbuflen);
905 /* end handle -E mkfs options */
907 /* Allow reformat of full devices (as opposed to
908 partitions.) We already checked for mounted dev. */
909 strscat(mop->mo_mkfsopts, " -F", sizeof(mop->mo_mkfsopts));
911 snprintf(mkfs_cmd, sizeof(mkfs_cmd),
912 "%s -j -b %d -L %s ", MKE2FS, L_BLOCK_SIZE,
913 mop->mo_ldd.ldd_svname);
915 fprintf(stderr,"%s: unsupported fs type: %d (%s)\n",
916 progname, mop->mo_ldd.ldd_mount_type,
917 MT_STR(&mop->mo_ldd));
921 /* For loop device format the dev, not the filename */
922 dev = mop->mo_device;
923 if (mop->mo_flags & MO_IS_LOOP)
924 dev = mop->mo_loopdev;
926 vprint("formatting backing filesystem %s on %s\n",
927 MT_STR(&mop->mo_ldd), dev);
928 vprint("\ttarget name %s\n", mop->mo_ldd.ldd_svname);
929 vprint("\t4k blocks %ju\n", (uintmax_t)block_count);
930 vprint("\toptions %s\n", mop->mo_mkfsopts);
932 /* mkfs_cmd's trailing space is important! */
933 strscat(mkfs_cmd, mop->mo_mkfsopts, sizeof(mkfs_cmd));
934 strscat(mkfs_cmd, " ", sizeof(mkfs_cmd));
935 strscat(mkfs_cmd, dev, sizeof(mkfs_cmd));
936 if (block_count != 0) {
937 snprintf(buf, sizeof(buf), " %ju",
938 (uintmax_t)block_count);
939 strscat(mkfs_cmd, buf, sizeof(mkfs_cmd));
942 vprint("mkfs_cmd = %s\n", mkfs_cmd);
943 ret = run_command(mkfs_cmd, sizeof(mkfs_cmd));
946 fprintf(stderr, "Unable to build fs %s (%d)\n", dev, ret);
951 int ldiskfs_prepare_lustre(struct mkfs_opts *mop,
952 char *wanted_mountopts, size_t len)
954 struct lustre_disk_data *ldd = &mop->mo_ldd;
957 /* Set MO_IS_LOOP to indicate a loopback device is needed */
958 ret = is_block(mop->mo_device);
961 } else if (ret == 0) {
962 mop->mo_flags |= MO_IS_LOOP;
965 if (IS_MDT(ldd) || IS_MGS(ldd))
966 strscat(wanted_mountopts, ",user_xattr", len);
971 int ldiskfs_fix_mountopts(struct mkfs_opts *mop, char *mountopts, size_t len)
973 if (strstr(mountopts, "errors=") == NULL)
974 strscat(mountopts, ",errors=remount-ro", len);
979 static int read_file(const char *path, char *buf, int size)
983 fd = fopen(path, "r");
987 if (fgets(buf, size, fd) == NULL) {
988 fprintf(stderr, "reading from %s: %s", path, strerror(errno));
994 /* strip trailing newline */
996 if (buf[size - 1] == '\n')
997 buf[size - 1] = '\0';
1002 static int write_file(const char *path, const char *buf)
1006 fd = open(path, O_WRONLY);
1010 rc = write(fd, buf, strlen(buf));
1013 return rc < 0 ? errno : 0;
1016 static int set_blockdev_scheduler(const char *path, const char *scheduler)
1018 char buf[PATH_MAX], *s, *e, orig_sched[50];
1021 /* Before setting the scheduler, we need to check to see if it's
1022 * already set to "noop". If it is, we don't want to override
1023 * that setting. If it's set to anything other than "noop", set
1024 * the scheduler to what has been passed in. */
1026 rc = read_file(path, buf, sizeof(buf));
1029 fprintf(stderr, "%s: cannot open '%s': %s\n",
1030 progname, path, strerror(errno));
1034 /* The expected format of buf: noop anticipatory deadline [cfq] */
1035 s = strchr(buf, '[');
1036 e = strchr(buf, ']');
1038 /* If the format is not what we expect. Play it safe and error out. */
1039 if (s == NULL || e == NULL) {
1041 fprintf(stderr, "%s: cannot parse scheduler "
1042 "options for '%s'\n", progname, path);
1046 snprintf(orig_sched, e - s, "%s", s + 1);
1048 if (strcmp(orig_sched, "noop") == 0 ||
1049 strcmp(orig_sched, scheduler) == 0)
1052 rc = write_file(path, scheduler);
1055 fprintf(stderr, "%s: cannot set scheduler on "
1056 "'%s': %s\n", progname, path,
1060 fprintf(stderr, "%s: change scheduler of %s from %s to %s\n",
1061 progname, path, orig_sched, scheduler);
1067 /* This is to tune the kernel for good SCSI performance.
1068 * For that we set the value of /sys/block/{dev}/queue/max_sectors_kb
1069 * to the value of /sys/block/{dev}/queue/max_hw_sectors_kb */
1070 static int set_blockdev_tunables(char *source, struct mount_opts *mop)
1072 glob_t glob_info = { 0 };
1073 struct stat stat_buf;
1074 char *chk_major, *chk_minor;
1075 char *savept = NULL, *dev;
1077 char buf[PATH_MAX] = {'\0'}, path[PATH_MAX] = {'\0'};
1078 char real_path[PATH_MAX] = {'\0'};
1086 ret_path = realpath(source, real_path);
1087 if (ret_path == NULL) {
1089 fprintf(stderr, "warning: %s: cannot resolve: %s\n",
1090 source, strerror(errno));
1094 if (strncmp(real_path, "/dev/loop", 9) == 0)
1097 if ((real_path[0] != '/') && (strpbrk(real_path, ",:") != NULL))
1100 snprintf(path, sizeof(path), "/sys/block%s", real_path + 4);
1101 if (access(path, X_OK) == 0)
1104 /* The name of the device say 'X' specified in /dev/X may not
1105 * match any entry under /sys/block/. In that case we need to
1106 * match the major/minor number to find the entry under
1107 * sys/block corresponding to /dev/X */
1109 /* Don't chop tail digit on /dev/mapper/xxx, LU-478 */
1110 if (strncmp(real_path, "/dev/mapper", 11) != 0) {
1111 dev = real_path + strlen(real_path);
1112 while (--dev > real_path && isdigit(*dev))
1115 if (strncmp(real_path, "/dev/md_", 8) == 0)
1119 rc = stat(real_path, &stat_buf);
1122 fprintf(stderr, "warning: %s, device %s stat failed\n",
1123 strerror(errno), real_path);
1127 major = major(stat_buf.st_rdev);
1128 minor = minor(stat_buf.st_rdev);
1129 rc = glob("/sys/block/*", GLOB_NOSORT, NULL, &glob_info);
1132 fprintf(stderr, "warning: failed to read entries under "
1134 globfree(&glob_info);
1138 for (i = 0; i < glob_info.gl_pathc; i++){
1139 snprintf(path, sizeof(path), "%s/dev", glob_info.gl_pathv[i]);
1141 rc = read_file(path, buf, sizeof(buf));
1145 if (buf[strlen(buf) - 1] == '\n')
1146 buf[strlen(buf) - 1] = '\0';
1148 chk_major = strtok_r(buf, ":", &savept);
1150 if (chk_major != NULL && major == atoi(chk_major) &&
1151 chk_minor != NULL && minor == atoi(chk_minor))
1155 if (i == glob_info.gl_pathc) {
1157 fprintf(stderr,"warning: device %s does not match any "
1158 "entry under /sys/block\n", real_path);
1159 globfree(&glob_info);
1163 /* Chop off "/dev" from path we found */
1164 path[strlen(glob_info.gl_pathv[i])] = '\0';
1165 globfree(&glob_info);
1168 if (strncmp(real_path, "/dev/md", 7) == 0) {
1169 snprintf(real_path, sizeof(real_path), "%s/%s", path,
1172 rc = read_file(real_path, buf, sizeof(buf));
1175 fprintf(stderr, "warning: opening %s: %s\n",
1176 real_path, strerror(errno));
1180 if (atoi(buf) >= mop->mo_md_stripe_cache_size)
1183 if (strlen(buf) - 1 > 0) {
1184 snprintf(buf, sizeof(buf), "%d",
1185 mop->mo_md_stripe_cache_size);
1186 rc = write_file(real_path, buf);
1187 if (rc != 0 && verbose)
1188 fprintf(stderr, "warning: opening %s: %s\n",
1189 real_path, strerror(errno));
1191 /* Return since raid and disk tunables are different */
1195 snprintf(real_path, sizeof(real_path), "%s/%s", path,
1196 MAX_HW_SECTORS_KB_PATH);
1197 rc = read_file(real_path, buf, sizeof(buf));
1200 fprintf(stderr, "warning: opening %s: %s\n",
1201 real_path, strerror(errno));
1202 /* No MAX_HW_SECTORS_KB_PATH isn't necessary an
1203 * error for some device. */
1207 if (strlen(buf) - 1 > 0) {
1208 char oldbuf[32] = "", *end = NULL;
1209 unsigned long long oldval, newval;
1211 snprintf(real_path, sizeof(real_path), "%s/%s", path,
1212 MAX_SECTORS_KB_PATH);
1213 rc = read_file(real_path, oldbuf, sizeof(oldbuf));
1214 /* Only set new parameter if different from the old one. */
1215 if (rc != 0 || strcmp(oldbuf, buf) == 0) {
1216 /* No MAX_SECTORS_KB_PATH isn't necessary an
1217 * error for some device. */
1221 newval = strtoull(buf, &end, 0);
1222 if (newval == 0 || newval == ULLONG_MAX || end == buf)
1225 /* Don't increase IO request size limit past 16MB. It is about
1226 * PTLRPC_MAX_BRW_SIZE, but that isn't in a public header.
1227 * Note that even though the block layer allows larger values,
1228 * setting max_sectors_kb = 32768 causes crashes (LU-6974). */
1229 if (newval > 16 * 1024) {
1231 snprintf(buf, sizeof(buf), "%llu", newval);
1234 oldval = strtoull(oldbuf, &end, 0);
1235 /* Don't shrink the current limit. */
1236 if (oldval != ULLONG_MAX && newval <= oldval)
1239 rc = write_file(real_path, buf);
1242 fprintf(stderr, "warning: writing to %s: %s\n",
1243 real_path, strerror(errno));
1244 /* No MAX_SECTORS_KB_PATH isn't necessary an
1245 * error for some device. */
1248 fprintf(stderr, "%s: increased %s from %s to %s\n",
1249 progname, real_path, oldbuf, buf);
1253 /* Purposely ignore errors reported from set_blockdev_scheduler.
1254 * The worst that will happen is a block device with an "incorrect"
1256 snprintf(real_path, sizeof(real_path), "%s/%s", path, SCHEDULER_PATH);
1257 set_blockdev_scheduler(real_path, DEFAULT_SCHEDULER);
1259 /* if device is multipath device, tune its slave devices */
1260 glob_info.gl_pathc = 0;
1261 glob_info.gl_offs = 0;
1262 snprintf(real_path, sizeof(real_path), "%s/slaves/*", path);
1263 rc = glob(real_path, GLOB_NOSORT, NULL, &glob_info);
1265 for (i = 0; rc == 0 && i < glob_info.gl_pathc; i++) {
1266 slave = basename(glob_info.gl_pathv[i]);
1267 snprintf(real_path, sizeof(real_path), "/dev/%s", slave);
1268 rc = set_blockdev_tunables(real_path, mop);
1271 if (rc == GLOB_NOMATCH) {
1272 /* no slave device is not an error */
1274 } else if (rc && verbose) {
1275 if (slave == NULL) {
1276 fprintf(stderr, "warning: %s, failed to read"
1277 " entries under %s/slaves\n",
1278 strerror(errno), path);
1280 fprintf(stderr, "unable to set tunables for"
1281 " slave device %s (slave would be"
1282 " unable to handle IO request from"
1287 globfree(&glob_info);
1292 int ldiskfs_tune_lustre(char *dev, struct mount_opts *mop)
1294 return set_blockdev_tunables(dev, mop);
1297 int ldiskfs_label_lustre(struct mount_opts *mop)
1299 char label_cmd[PATH_MAX];
1302 snprintf(label_cmd, sizeof(label_cmd),
1303 TUNE2FS" -f -L '%s' '%s' >/dev/null 2>&1",
1304 mop->mo_ldd.ldd_svname, mop->mo_source);
1305 rc = run_command(label_cmd, sizeof(label_cmd));
1310 /* Enable quota accounting */
1311 int ldiskfs_enable_quota(struct mkfs_opts *mop)
1315 int cmdsz = sizeof(cmd), ret;
1317 if (is_e2fsprogs_feature_supp("-O quota") != 0) {
1318 fprintf(stderr, "%s: \"-O quota\" is is not supported by "
1319 "current e2fsprogs\n", progname);
1323 dev = mop->mo_device;
1324 if (mop->mo_flags & MO_IS_LOOP)
1325 dev = mop->mo_loopdev;
1327 /* Quota feature is already enabled? */
1328 if (is_feature_enabled("quota", dev)) {
1329 vprint("Quota feature is already enabled.\n");
1333 /* Turn on quota feature by "tune2fs -O quota" */
1334 snprintf(cmd, cmdsz, "%s -O quota %s", TUNE2FS, dev);
1335 ret = run_command(cmd, cmdsz);
1337 fprintf(stderr, "command:%s (%d)", cmd, ret);
1342 int ldiskfs_init(void)
1344 /* Required because full path to DEBUGFS is not specified */
1345 setenv("PATH", "/bin:/sbin:/usr/bin:/usr/sbin:/usr/local/bin", 0);
1350 void ldiskfs_fini(void)