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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/utils/mount_utils_ldiskfs.c
34 * Author: Nathan Rutman <nathan@clusterfs.com>
37 /* This source file is compiled into both mkfs.lustre and tunefs.lustre */
41 #endif /* HAVE_CONFIG_H */
56 #include <sys/types.h>
58 #include <sys/mount.h>
59 #include <sys/utsname.h>
67 #include <linux/fs.h> /* for BLKGETSIZE64 */
69 #include <linux/version.h>
70 #include <lustre_disk.h>
71 #include <lustre_param.h>
72 #include <lnet/lnetctl.h>
73 #include <lustre_ver.h>
76 #include <selinux/selinux.h>
79 #include "mount_utils.h"
81 #define MAX_HW_SECTORS_KB_PATH "queue/max_hw_sectors_kb"
82 #define MAX_SECTORS_KB_PATH "queue/max_sectors_kb"
83 #define SCHEDULER_PATH "queue/scheduler"
84 #define STRIPE_CACHE_SIZE "md/stripe_cache_size"
86 #define DEFAULT_SCHEDULER "deadline"
88 extern char *progname;
90 #define L_BLOCK_SIZE 4096
91 /* keep it less than LL_FID_NAMELEN */
92 #define DUMMY_FILE_NAME_LEN 25
93 #define EXT3_DIRENT_SIZE DUMMY_FILE_NAME_LEN
95 static void append_unique(char *buf, char *prefix, char *key, char *val,
99 * Concatenate context of the temporary mount point if selinux is enabled
102 static void append_context_for_mount(char *mntpt, struct mkfs_opts *mop)
104 security_context_t fcontext;
106 if (getfilecon(mntpt, &fcontext) < 0) {
107 /* Continuing with default behaviour */
108 fprintf(stderr, "%s: Get file context failed : %s\n",
109 progname, strerror(errno));
113 if (fcontext != NULL) {
114 append_unique(mop->mo_ldd.ldd_mount_opts,
115 ",", "context", fcontext,
116 sizeof(mop->mo_ldd.ldd_mount_opts));
122 /* return canonicalized absolute pathname, even if the target file does not
123 * exist, unlike realpath */
124 static char *absolute_path(char *devname)
126 char buf[PATH_MAX + 1] = "";
131 path = malloc(sizeof(buf));
135 if (devname[0] != '/') {
136 if (getcwd(buf, sizeof(buf) - 1) == NULL) {
140 len = snprintf(path, sizeof(buf), "%s/%s", buf, devname);
141 if (len >= sizeof(buf)) {
146 len = snprintf(path, sizeof(buf), "%s", devname);
147 if (len >= sizeof(buf)) {
153 /* truncate filename before calling realpath */
154 ptr = strrchr(path, '/');
160 if (buf != realpath(path, buf)) {
164 /* add the filename back */
165 len = snprintf(path, PATH_MAX, "%s/%s", buf, ptr+1);
166 if (len >= PATH_MAX) {
173 /* Determine if a device is a block device (as opposed to a file) */
174 static int is_block(char *devname)
180 devpath = absolute_path(devname);
181 if (devpath == NULL) {
182 fprintf(stderr, "%s: failed to resolve path to %s\n",
187 ret = access(devname, F_OK);
189 if (strncmp(devpath, "/dev/", 5) == 0) {
190 /* nobody sane wants to create a loopback file under
191 * /dev. Let's just report the device doesn't exist */
192 fprintf(stderr, "%s: %s apparently does not exist\n",
200 ret = stat(devpath, &st);
202 fprintf(stderr, "%s: cannot stat %s\n", progname, devpath);
205 ret = S_ISBLK(st.st_mode);
211 static int is_feature_enabled(const char *feature, const char *devpath)
215 char enabled_features[4096] = "";
218 snprintf(cmd, sizeof(cmd), "%s -c -R features %s 2>&1",
221 /* Using popen() instead of run_command() since debugfs does
222 * not return proper error code if command is not supported */
223 fp = popen(cmd, "r");
225 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
229 ret = fread(enabled_features, 1, sizeof(enabled_features) - 1, fp);
230 enabled_features[ret] = '\0';
233 if (strstr(enabled_features, feature))
238 /* Write the server config files */
239 int ldiskfs_write_ldd(struct mkfs_opts *mop)
241 char mntpt[] = "/tmp/mntXXXXXX";
248 /* Mount this device temporarily in order to write these files */
249 if (!mkdtemp(mntpt)) {
250 fprintf(stderr, "%s: Can't create temp mount point %s: %s\n",
251 progname, mntpt, strerror(errno));
256 * Append file context to mount options if SE Linux is enabled
259 if (is_selinux_enabled() > 0)
260 append_context_for_mount(mntpt, mop);
263 dev = mop->mo_device;
264 if (mop->mo_flags & MO_IS_LOOP)
265 dev = mop->mo_loopdev;
267 ret = mount(dev, mntpt, MT_STR(&mop->mo_ldd), 0,
268 (mop->mo_mountopts == NULL) ?
269 "errors=remount-ro" : mop->mo_mountopts);
271 fprintf(stderr, "%s: Unable to mount %s: %s\n",
272 progname, dev, strerror(errno));
274 if (errno == ENODEV) {
275 fprintf(stderr, "Is the %s module available?\n",
276 MT_STR(&mop->mo_ldd));
281 /* Set up initial directories */
282 sprintf(filepnm, "%s/%s", mntpt, MOUNT_CONFIGS_DIR);
283 ret = mkdir(filepnm, 0777);
284 if ((ret != 0) && (errno != EEXIST)) {
285 fprintf(stderr, "%s: Can't make configs dir %s (%s)\n",
286 progname, filepnm, strerror(errno));
288 } else if (errno == EEXIST) {
292 /* Save the persistent mount data into a file. Lustre must pre-read
293 this file to get the real mount options. */
294 vprint("Writing %s\n", MOUNT_DATA_FILE);
295 sprintf(filepnm, "%s/%s", mntpt, MOUNT_DATA_FILE);
296 filep = fopen(filepnm, "w");
298 fprintf(stderr, "%s: Unable to create %s file: %s\n",
299 progname, filepnm, strerror(errno));
302 num = fwrite(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
303 if (num < 1 && ferror(filep)) {
304 fprintf(stderr, "%s: Unable to write to file (%s): %s\n",
305 progname, filepnm, strerror(errno));
309 fsync(filep->_fileno);
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);
392 int ldiskfs_erase_ldd(struct mkfs_opts *mop, char *param)
397 void ldiskfs_print_ldd_params(struct mkfs_opts *mop)
399 printf("Parameters:%s\n", mop->mo_ldd.ldd_params);
402 /* Display the need for the latest e2fsprogs to be installed. make_backfs
403 * indicates if the caller is make_lustre_backfs() or not. */
404 static void disp_old_e2fsprogs_msg(const char *feature, int make_backfs)
406 static int msg_displayed;
409 fprintf(stderr, "WARNING: %s does not support %s "
410 "feature.\n\n", E2FSPROGS, feature);
416 fprintf(stderr, "WARNING: The %s package currently installed on "
417 "your system does not support \"%s\" feature.\n",
419 #if !(HAVE_LDISKFSPROGS)
420 fprintf(stderr, "Please install the latest version of e2fsprogs from\n"
421 "https://downloads.hpdd.intel.com/public/e2fsprogs/latest/\n"
422 "to enable this feature.\n");
425 fprintf(stderr, "Feature will not be enabled until %s"
426 "is updated and '%s -O %s %%{device}' "
427 "is run.\n\n", E2FSPROGS, TUNE2FS, feature);
430 /* Check whether the file exists in the device */
431 static int file_in_dev(char *file_name, char *dev_name)
434 char debugfs_cmd[256];
435 unsigned int inode_num;
438 /* Construct debugfs command line. */
439 snprintf(debugfs_cmd, sizeof(debugfs_cmd),
440 "%s -c -R 'stat %s' '%s' 2>&1 | egrep '(Inode|unsupported)'",
441 DEBUGFS, file_name, dev_name);
443 fp = popen(debugfs_cmd, "r");
445 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
449 if (fscanf(fp, "Inode: %u", &inode_num) == 1) { /* exist */
453 i = fread(debugfs_cmd, 1, sizeof(debugfs_cmd) - 1, fp);
456 fprintf(stderr, "%s", debugfs_cmd);
457 if (strstr(debugfs_cmd, "unsupported feature")) {
458 disp_old_e2fsprogs_msg("an unknown", 0);
467 /* Check whether the device has already been used with lustre */
468 int ldiskfs_is_lustre(char *dev, unsigned *mount_type)
472 ret = file_in_dev(MOUNT_DATA_FILE, dev);
474 /* in the -1 case, 'extents' means IS a lustre target */
475 *mount_type = LDD_MT_LDISKFS;
479 ret = file_in_dev(LAST_RCVD, dev);
481 *mount_type = LDD_MT_LDISKFS;
488 /* Check if a certain feature is supported by e2fsprogs.
489 * Firstly we try to use "debugfs supported_features" command to check if
490 * the feature is supported. If this fails we try to set this feature with
491 * mke2fs to check for its support. */
492 static int is_e2fsprogs_feature_supp(const char *feature)
494 static char supp_features[4096] = "";
497 char imgname[] = "/tmp/test-img-XXXXXX";
501 if (supp_features[0] == '\0') {
502 snprintf(cmd, sizeof(cmd), "%s -c -R supported_features 2>&1",
505 /* Using popen() instead of run_command() since debugfs does
506 * not return proper error code if command is not supported */
507 fp = popen(cmd, "r");
509 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
512 ret = fread(supp_features, 1, sizeof(supp_features) - 1, fp);
513 supp_features[ret] = '\0';
516 if (ret > 0 && strstr(supp_features,
517 strncmp(feature, "-O ", 3) ? feature : feature+3))
520 if ((fd = mkstemp(imgname)) < 0)
525 snprintf(cmd, sizeof(cmd), "%s -F %s %s 100 >/dev/null 2>&1",
526 MKE2FS, feature, imgname);
527 /* run_command() displays the output of mke2fs when it fails for
528 * some feature, so use system() directly */
537 * append_unique: append @key or @key=@val pair to @buf only if @key does not
539 * @buf: buffer to hold @key or @key=@val
540 * @prefix: prefix string before @key
542 * @val: value string if it's a @key=@val pair
544 static void append_unique(char *buf, char *prefix, char *key, char *val,
553 anchor = end = strstr(buf, key);
554 /* try to find exact match string in @buf */
555 while (end && *end != '\0' && *end != ',' && *end != ' ' && *end != '=')
558 if (anchor == NULL || strlen(key) != len ||
559 strncmp(anchor, key, len) != 0) {
561 strscat(buf, prefix, maxbuflen);
563 strscat(buf, key, maxbuflen);
565 strscat(buf, "=\"", maxbuflen);
566 strscat(buf, val, maxbuflen);
567 strscat(buf, "\"", maxbuflen);
572 static int enable_default_ext4_features(struct mkfs_opts *mop, char *anchor,
573 size_t maxbuflen, int user_spec)
575 if (IS_OST(&mop->mo_ldd)) {
576 append_unique(anchor, user_spec ? "," : " -O ",
577 "extents", NULL, maxbuflen);
578 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
579 } else if (IS_MDT(&mop->mo_ldd)) {
580 append_unique(anchor, user_spec ? "," : " -O ",
581 "dirdata", NULL, maxbuflen);
582 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
583 append_unique(anchor, ",", "^extents", NULL, maxbuflen);
585 append_unique(anchor, user_spec ? "," : " -O ",
586 "uninit_bg", NULL, maxbuflen);
589 /* Multiple mount protection enabled only if failover node specified */
590 if (mop->mo_flags & MO_FAILOVER) {
591 if (is_e2fsprogs_feature_supp("-O mmp") == 0)
592 append_unique(anchor, ",", "mmp", NULL, maxbuflen);
594 disp_old_e2fsprogs_msg("mmp", 1);
597 /* Allow more than 65000 subdirectories */
598 if (is_e2fsprogs_feature_supp("-O dir_nlink") == 0)
599 append_unique(anchor, ",", "dir_nlink", NULL, maxbuflen);
601 /* The following options are only valid for ext4-based ldiskfs.
602 * If --backfstype=ext3 is specified, do not enable them. */
603 if (mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3)
606 /* Enable quota by default */
607 if (is_e2fsprogs_feature_supp("-O quota") == 0) {
608 append_unique(anchor, ",", "quota", NULL, maxbuflen);
611 fprintf(stderr, "\"-O quota\" must be supported by "
612 "e2fsprogs, please upgrade your e2fsprogs.\n");
616 /* Allow files larger than 2TB. Also needs LU-16, but not harmful. */
617 if (is_e2fsprogs_feature_supp("-O huge_file") == 0)
618 append_unique(anchor, ",", "huge_file", NULL, maxbuflen);
620 /* Enable large block addresses if the LUN is over 2^32 blocks. */
621 if (mop->mo_device_kb / (L_BLOCK_SIZE >> 10) >= 0x100002000ULL &&
622 is_e2fsprogs_feature_supp("-O 64bit") == 0)
623 append_unique(anchor, ",", "64bit", NULL, maxbuflen);
625 /* Cluster inode/block bitmaps and inode table for more efficient IO.
626 * Align the flex groups on a 1MB boundary for better performance. */
627 /* This -O feature needs to go last, since it adds the "-G" option. */
628 if (is_e2fsprogs_feature_supp("-O flex_bg") == 0) {
631 append_unique(anchor, ",", "flex_bg", NULL, maxbuflen);
633 if (IS_OST(&mop->mo_ldd) &&
634 strstr(mop->mo_mkfsopts, "-G") == NULL) {
635 snprintf(tmp_buf, sizeof(tmp_buf), " -G %u",
636 (1 << 20) / L_BLOCK_SIZE);
637 strscat(anchor, tmp_buf, maxbuflen);
640 /* Don't add any more "-O" options here, see last comment above */
645 * moveopts_to_end: find the option string, move remaining strings to
646 * where option string starts, and append the option
648 * @start: where the option string starts before the move
649 * RETURN: where the option string starts after the move
651 static char *moveopts_to_end(char *start)
657 /* skip whitespace before options */
662 /* find end of option characters */
663 while (*end != ' ' && *end != '\0')
667 if (len >= sizeof(save))
668 len = sizeof(save) - 1;
671 strncpy(save, start, len);
674 /* move remaining options up front */
676 memmove(start, end, strlen(end));
677 *(start + strlen(end)) = '\0';
679 /* append the specified options */
680 if (*(start + strlen(start) - 1) != ' ')
682 idx = start + strlen(start);
688 /* Build fs according to type */
689 int ldiskfs_make_lustre(struct mkfs_opts *mop)
691 __u64 device_kb = mop->mo_device_kb, block_count = 0;
692 char mkfs_cmd[PATH_MAX];
696 int ret = 0, ext_opts = 0;
699 if (!(mop->mo_flags & MO_IS_LOOP)) {
700 mop->mo_device_kb = get_device_size(mop->mo_device);
702 if (mop->mo_device_kb == 0)
705 /* Compare to real size */
706 if (device_kb == 0 || device_kb > mop->mo_device_kb)
707 device_kb = mop->mo_device_kb;
709 mop->mo_device_kb = device_kb;
712 if (mop->mo_device_kb != 0) {
713 if (mop->mo_device_kb < 32384) {
714 fprintf(stderr, "%s: size of filesystem must be larger "
715 "than 32MB, but is set to %lldKB\n",
716 progname, (long long)mop->mo_device_kb);
719 block_count = mop->mo_device_kb / (L_BLOCK_SIZE >> 10);
720 /* If the LUN size is just over 2^32 blocks, limit the
721 * filesystem size to 2^32-1 blocks to avoid problems with
722 * ldiskfs/mkfs not handling this size. Bug 22906 */
723 if (block_count > 0xffffffffULL && block_count < 0x100002000ULL)
724 block_count = 0xffffffffULL;
727 if ((mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3) ||
728 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS) ||
729 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS2)) {
732 /* Journal size in MB */
733 if (strstr(mop->mo_mkfsopts, "-J") == NULL &&
734 device_kb > 1024 * 1024) {
735 /* Choose our own default journal size */
736 long journal_mb = 0, max_mb;
738 /* cap journal size at 4GB for MDT,
739 * leave it at 400MB for OSTs. */
740 if (IS_MDT(&mop->mo_ldd))
742 else if (IS_OST(&mop->mo_ldd))
744 else /* Use mke2fs default size for MGS */
747 /* Use at most 4% of device for journal */
748 journal_mb = device_kb * 4 / (1024 * 100);
749 if (journal_mb > max_mb)
753 sprintf(buf, " -J size=%ld", journal_mb);
754 strscat(mop->mo_mkfsopts, buf,
755 sizeof(mop->mo_mkfsopts));
759 /* Inode size includes:
760 * ldiskfs inode size: 156
761 * extended attributes size, including:
762 * ext4_xattr_header: 32
763 * LOV EA size: 32(lov_mds_md) +
764 * stripes * 24(lov_ost_data) +
765 * 16(xattr_entry) + 3(lov)
766 * LMA EA size: 24(lustre_mdt_attrs) +
767 * 16(xattr_entry) + 3(lma)
768 * link EA size: 24(link_ea_header) + 18(link_ea_entry) +
769 * (filename) + 16(xattr_entry) + 4(link)
770 * and some margin for 4-byte alignment, ACLs and other EAs.
772 * If we say the average filename length is about 32 bytes,
773 * the calculation looks like:
774 * 156 + 32 + (32+24*N+19) + (24+19) + (24+18+~32+20) + other <=
775 * 512*2^m, {m=0,1,2,3}
777 if (strstr(mop->mo_mkfsopts, "-I") == NULL) {
778 if (IS_MDT(&mop->mo_ldd)) {
779 if (mop->mo_stripe_count > 69)
780 inode_size = 512; /* bz 7241 */
781 /* see also "-i" below for EA blocks */
782 else if (mop->mo_stripe_count > 26)
784 else if (mop->mo_stripe_count > 5)
788 } else if (IS_OST(&mop->mo_ldd)) {
789 /* We store MDS FID and OST objid in EA on OST
790 * we need to make inode bigger as well. */
794 if (inode_size > 0) {
795 sprintf(buf, " -I %ld", inode_size);
796 strscat(mop->mo_mkfsopts, buf,
797 sizeof(mop->mo_mkfsopts));
801 /* Bytes_per_inode: disk size / num inodes */
802 if (strstr(mop->mo_mkfsopts, "-i") == NULL &&
803 strstr(mop->mo_mkfsopts, "-N") == NULL) {
804 long bytes_per_inode = 0;
806 /* Allocate more inodes on MDT devices. There is
807 * no data stored on the MDT, and very little extra
808 * metadata beyond the inode. It could go down as
809 * low as 1024 bytes, but this is conservative.
810 * Account for external EA blocks for wide striping. */
811 if (IS_MDT(&mop->mo_ldd)) {
812 bytes_per_inode = inode_size + 1536;
814 if (mop->mo_stripe_count > 69) {
815 int extra = mop->mo_stripe_count * 24;
816 extra = ((extra - 1) | 4095) + 1;
817 bytes_per_inode += extra;
821 /* Allocate fewer inodes on large OST devices. Most
822 * filesystems can be much more aggressive than even
823 * this, but it is impossible to know in advance. */
824 if (IS_OST(&mop->mo_ldd)) {
825 /* OST > 16TB assume average file size 1MB */
826 if (device_kb > (16ULL << 30))
827 bytes_per_inode = 1024 * 1024;
828 /* OST > 4TB assume average file size 512kB */
829 else if (device_kb > (4ULL << 30))
830 bytes_per_inode = 512 * 1024;
831 /* OST > 1TB assume average file size 256kB */
832 else if (device_kb > (1ULL << 30))
833 bytes_per_inode = 256 * 1024;
834 /* OST > 10GB assume average file size 64kB,
835 * plus a bit so that inodes will fit into a
836 * 256x flex_bg without overflowing */
837 else if (device_kb > (10ULL << 20))
838 bytes_per_inode = 69905;
841 if (bytes_per_inode > 0) {
842 sprintf(buf, " -i %ld", bytes_per_inode);
843 strscat(mop->mo_mkfsopts, buf,
844 sizeof(mop->mo_mkfsopts));
849 strscat(mop->mo_mkfsopts, " -q",
850 sizeof(mop->mo_mkfsopts));
853 /* start handle -O mkfs options */
854 if ((start = strstr(mop->mo_mkfsopts, "-O")) != NULL) {
855 if (strstr(start + 2, "-O") != NULL) {
857 "%s: don't specify multiple -O options\n",
861 start = moveopts_to_end(start);
862 maxbuflen = sizeof(mop->mo_mkfsopts) -
863 (start - mop->mo_mkfsopts) - strlen(start);
864 ret = enable_default_ext4_features(mop, start, maxbuflen, 1);
866 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts),
867 maxbuflen = sizeof(mop->mo_mkfsopts) -
868 strlen(mop->mo_mkfsopts);
869 ret = enable_default_ext4_features(mop, start, maxbuflen, 0);
873 /* end handle -O mkfs options */
875 /* start handle -E mkfs options */
876 if ((start = strstr(mop->mo_mkfsopts, "-E")) != NULL) {
877 if (strstr(start + 2, "-E") != NULL) {
879 "%s: don't specify multiple -E options\n",
883 start = moveopts_to_end(start);
884 maxbuflen = sizeof(mop->mo_mkfsopts) -
885 (start - mop->mo_mkfsopts) - strlen(start);
888 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts);
889 maxbuflen = sizeof(mop->mo_mkfsopts) -
890 strlen(mop->mo_mkfsopts);
893 /* In order to align the filesystem metadata on 1MB boundaries,
894 * give a resize value that will reserve a power-of-two group
895 * descriptor blocks, but leave one block for the superblock.
896 * Only useful for filesystems with < 2^32 blocks due to resize
898 if (strstr(mop->mo_mkfsopts, "meta_bg") == NULL &&
899 IS_OST(&mop->mo_ldd) && mop->mo_device_kb > 100 * 1024 &&
900 mop->mo_device_kb * 1024 / L_BLOCK_SIZE <= 0xffffffffULL) {
901 unsigned group_blocks = L_BLOCK_SIZE * 8;
902 unsigned desc_per_block = L_BLOCK_SIZE / 32;
903 unsigned resize_blks;
905 resize_blks = (1ULL<<32) - desc_per_block*group_blocks;
906 snprintf(buf, sizeof(buf), "%u", resize_blks);
907 append_unique(start, ext_opts ? "," : " -E ",
908 "resize", buf, maxbuflen);
912 /* Avoid zeroing out the full journal - speeds up mkfs */
913 if (is_e2fsprogs_feature_supp("-E lazy_journal_init") == 0)
914 append_unique(start, ext_opts ? "," : " -E ",
915 "lazy_journal_init", NULL, maxbuflen);
916 /* end handle -E mkfs options */
918 /* Allow reformat of full devices (as opposed to
919 partitions.) We already checked for mounted dev. */
920 strscat(mop->mo_mkfsopts, " -F", sizeof(mop->mo_mkfsopts));
922 snprintf(mkfs_cmd, sizeof(mkfs_cmd),
923 "%s -j -b %d -L %s ", MKE2FS, L_BLOCK_SIZE,
924 mop->mo_ldd.ldd_svname);
926 fprintf(stderr,"%s: unsupported fs type: %d (%s)\n",
927 progname, mop->mo_ldd.ldd_mount_type,
928 MT_STR(&mop->mo_ldd));
932 /* For loop device format the dev, not the filename */
933 dev = mop->mo_device;
934 if (mop->mo_flags & MO_IS_LOOP)
935 dev = mop->mo_loopdev;
937 vprint("formatting backing filesystem %s on %s\n",
938 MT_STR(&mop->mo_ldd), dev);
939 vprint("\ttarget name %s\n", mop->mo_ldd.ldd_svname);
940 vprint("\t4k blocks %ju\n", (uintmax_t)block_count);
941 vprint("\toptions %s\n", mop->mo_mkfsopts);
943 /* mkfs_cmd's trailing space is important! */
944 strscat(mkfs_cmd, mop->mo_mkfsopts, sizeof(mkfs_cmd));
945 strscat(mkfs_cmd, " ", sizeof(mkfs_cmd));
946 strscat(mkfs_cmd, dev, sizeof(mkfs_cmd));
947 if (block_count != 0) {
948 snprintf(buf, sizeof(buf), " %ju",
949 (uintmax_t)block_count);
950 strscat(mkfs_cmd, buf, sizeof(mkfs_cmd));
953 vprint("mkfs_cmd = %s\n", mkfs_cmd);
954 ret = run_command(mkfs_cmd, sizeof(mkfs_cmd));
957 fprintf(stderr, "Unable to build fs %s (%d)\n", dev, ret);
962 int ldiskfs_prepare_lustre(struct mkfs_opts *mop,
963 char *wanted_mountopts, size_t len)
965 struct lustre_disk_data *ldd = &mop->mo_ldd;
968 /* Set MO_IS_LOOP to indicate a loopback device is needed */
969 ret = is_block(mop->mo_device);
972 } else if (ret == 0) {
973 mop->mo_flags |= MO_IS_LOOP;
976 if (IS_MDT(ldd) || IS_MGS(ldd))
977 strscat(wanted_mountopts, ",user_xattr", len);
982 int ldiskfs_fix_mountopts(struct mkfs_opts *mop, char *mountopts, size_t len)
984 if (strstr(mountopts, "errors=") == NULL)
985 strscat(mountopts, ",errors=remount-ro", len);
990 static int read_file(const char *path, char *buf, int size)
994 fd = fopen(path, "r");
998 if (fgets(buf, size, fd) == NULL) {
999 fprintf(stderr, "reading from %s: %s", path, strerror(errno));
1005 /* strip trailing newline */
1007 if (buf[size - 1] == '\n')
1008 buf[size - 1] = '\0';
1013 static int write_file(const char *path, const char *buf)
1017 fd = open(path, O_WRONLY);
1021 rc = write(fd, buf, strlen(buf));
1024 return rc < 0 ? errno : 0;
1027 static int set_blockdev_scheduler(const char *path, const char *scheduler)
1029 char buf[PATH_MAX], *s, *e, orig_sched[50];
1032 /* Before setting the scheduler, we need to check to see if it's
1033 * already set to "noop". If it is, we don't want to override
1034 * that setting. If it's set to anything other than "noop", set
1035 * the scheduler to what has been passed in. */
1037 rc = read_file(path, buf, sizeof(buf));
1040 fprintf(stderr, "%s: cannot open '%s': %s\n",
1041 progname, path, strerror(errno));
1045 /* The expected format of buf: noop anticipatory deadline [cfq] */
1046 s = strchr(buf, '[');
1047 e = strchr(buf, ']');
1049 /* If the format is not what we expect. Play it safe and error out. */
1050 if (s == NULL || e == NULL) {
1052 fprintf(stderr, "%s: cannot parse scheduler "
1053 "options for '%s'\n", progname, path);
1057 snprintf(orig_sched, e - s, "%s", s + 1);
1059 if (strcmp(orig_sched, "noop") == 0 ||
1060 strcmp(orig_sched, scheduler) == 0)
1063 rc = write_file(path, scheduler);
1066 fprintf(stderr, "%s: cannot set scheduler on "
1067 "'%s': %s\n", progname, path,
1071 fprintf(stderr, "%s: change scheduler of %s from %s to %s\n",
1072 progname, path, orig_sched, scheduler);
1078 /* This is to tune the kernel for good SCSI performance.
1079 * For that we set the value of /sys/block/{dev}/queue/max_sectors_kb
1080 * to the value of /sys/block/{dev}/queue/max_hw_sectors_kb */
1081 static int set_blockdev_tunables(char *source, struct mount_opts *mop)
1083 glob_t glob_info = { 0 };
1084 struct stat stat_buf;
1085 char *chk_major, *chk_minor;
1086 char *savept = NULL, *dev;
1088 char buf[PATH_MAX] = {'\0'}, path[PATH_MAX] = {'\0'};
1089 char real_path[PATH_MAX] = {'\0'};
1097 ret_path = realpath(source, real_path);
1098 if (ret_path == NULL) {
1100 fprintf(stderr, "warning: %s: cannot resolve: %s\n",
1101 source, strerror(errno));
1105 if (strncmp(real_path, "/dev/loop", 9) == 0)
1108 if ((real_path[0] != '/') && (strpbrk(real_path, ",:") != NULL))
1111 snprintf(path, sizeof(path), "/sys/block%s", real_path + 4);
1112 if (access(path, X_OK) == 0)
1115 /* The name of the device say 'X' specified in /dev/X may not
1116 * match any entry under /sys/block/. In that case we need to
1117 * match the major/minor number to find the entry under
1118 * sys/block corresponding to /dev/X */
1120 /* Don't chop tail digit on /dev/mapper/xxx, LU-478 */
1121 if (strncmp(real_path, "/dev/mapper", 11) != 0) {
1122 dev = real_path + strlen(real_path);
1123 while (--dev > real_path && isdigit(*dev))
1126 if (strncmp(real_path, "/dev/md_", 8) == 0)
1130 rc = stat(real_path, &stat_buf);
1133 fprintf(stderr, "warning: %s, device %s stat failed\n",
1134 strerror(errno), real_path);
1138 major = major(stat_buf.st_rdev);
1139 minor = minor(stat_buf.st_rdev);
1140 rc = glob("/sys/block/*", GLOB_NOSORT, NULL, &glob_info);
1143 fprintf(stderr, "warning: failed to read entries under "
1145 globfree(&glob_info);
1149 for (i = 0; i < glob_info.gl_pathc; i++){
1150 snprintf(path, sizeof(path), "%s/dev", glob_info.gl_pathv[i]);
1152 rc = read_file(path, buf, sizeof(buf));
1156 if (buf[strlen(buf) - 1] == '\n')
1157 buf[strlen(buf) - 1] = '\0';
1159 chk_major = strtok_r(buf, ":", &savept);
1161 if (chk_major != NULL && major == atoi(chk_major) &&
1162 chk_minor != NULL && minor == atoi(chk_minor))
1166 if (i == glob_info.gl_pathc) {
1168 fprintf(stderr,"warning: device %s does not match any "
1169 "entry under /sys/block\n", real_path);
1170 globfree(&glob_info);
1174 /* Chop off "/dev" from path we found */
1175 path[strlen(glob_info.gl_pathv[i])] = '\0';
1176 globfree(&glob_info);
1179 if (strncmp(real_path, "/dev/md", 7) == 0) {
1180 snprintf(real_path, sizeof(real_path), "%s/%s", path,
1183 rc = read_file(real_path, buf, sizeof(buf));
1186 fprintf(stderr, "warning: opening %s: %s\n",
1187 real_path, strerror(errno));
1191 if (atoi(buf) >= mop->mo_md_stripe_cache_size)
1194 if (strlen(buf) - 1 > 0) {
1195 snprintf(buf, sizeof(buf), "%d",
1196 mop->mo_md_stripe_cache_size);
1197 rc = write_file(real_path, buf);
1198 if (rc != 0 && verbose)
1199 fprintf(stderr, "warning: opening %s: %s\n",
1200 real_path, strerror(errno));
1202 /* Return since raid and disk tunables are different */
1206 if (mop->mo_max_sectors_kb >= 0) {
1207 snprintf(buf, sizeof(buf), "%d", mop->mo_max_sectors_kb);
1209 snprintf(real_path, sizeof(real_path), "%s/%s", path,
1210 MAX_HW_SECTORS_KB_PATH);
1211 rc = read_file(real_path, buf, sizeof(buf));
1214 fprintf(stderr, "warning: opening %s: %s\n",
1215 real_path, strerror(errno));
1216 /* No MAX_HW_SECTORS_KB_PATH isn't necessary an
1217 * error for some device. */
1222 if (strlen(buf) - 1 > 0) {
1223 char oldbuf[32] = "", *end = NULL;
1224 unsigned long long oldval, newval;
1226 snprintf(real_path, sizeof(real_path), "%s/%s", path,
1227 MAX_SECTORS_KB_PATH);
1228 rc = read_file(real_path, oldbuf, sizeof(oldbuf));
1229 /* Only set new parameter if different from the old one. */
1230 if (rc != 0 || strcmp(oldbuf, buf) == 0) {
1231 /* No MAX_SECTORS_KB_PATH isn't necessary an
1232 * error for some device. */
1236 newval = strtoull(buf, &end, 0);
1237 if (newval == 0 || newval == ULLONG_MAX || end == buf)
1240 /* Don't increase IO request size limit past 16MB. It is about
1241 * PTLRPC_MAX_BRW_SIZE, but that isn't in a public header.
1242 * Note that even though the block layer allows larger values,
1243 * setting max_sectors_kb = 32768 causes crashes (LU-6974). */
1244 if (mop->mo_max_sectors_kb < 0 && newval > 16 * 1024) {
1246 snprintf(buf, sizeof(buf), "%llu", newval);
1249 oldval = strtoull(oldbuf, &end, 0);
1250 /* Don't shrink the current limit. */
1251 if (mop->mo_max_sectors_kb < 0 && oldval != ULLONG_MAX &&
1255 rc = write_file(real_path, buf);
1258 fprintf(stderr, "warning: writing to %s: %s\n",
1259 real_path, strerror(errno));
1260 /* No MAX_SECTORS_KB_PATH isn't necessary an
1261 * error for some device. */
1264 fprintf(stderr, "%s: increased %s from %s to %s\n",
1265 progname, real_path, oldbuf, buf);
1269 /* Purposely ignore errors reported from set_blockdev_scheduler.
1270 * The worst that will happen is a block device with an "incorrect"
1272 snprintf(real_path, sizeof(real_path), "%s/%s", path, SCHEDULER_PATH);
1273 set_blockdev_scheduler(real_path, DEFAULT_SCHEDULER);
1275 /* if device is multipath device, tune its slave devices */
1276 glob_info.gl_pathc = 0;
1277 glob_info.gl_offs = 0;
1278 snprintf(real_path, sizeof(real_path), "%s/slaves/*", path);
1279 rc = glob(real_path, GLOB_NOSORT, NULL, &glob_info);
1281 for (i = 0; rc == 0 && i < glob_info.gl_pathc; i++) {
1282 slave = basename(glob_info.gl_pathv[i]);
1283 snprintf(real_path, sizeof(real_path), "/dev/%s", slave);
1284 rc = set_blockdev_tunables(real_path, mop);
1287 if (rc == GLOB_NOMATCH) {
1288 /* no slave device is not an error */
1290 } else if (rc && verbose) {
1291 if (slave == NULL) {
1292 fprintf(stderr, "warning: %s, failed to read"
1293 " entries under %s/slaves\n",
1294 strerror(errno), path);
1296 fprintf(stderr, "unable to set tunables for"
1297 " slave device %s (slave would be"
1298 " unable to handle IO request from"
1303 globfree(&glob_info);
1308 int ldiskfs_tune_lustre(char *dev, struct mount_opts *mop)
1310 return set_blockdev_tunables(dev, mop);
1313 int ldiskfs_label_lustre(struct mount_opts *mop)
1315 char label_cmd[PATH_MAX];
1318 snprintf(label_cmd, sizeof(label_cmd),
1319 TUNE2FS" -f -L '%s' '%s' >/dev/null 2>&1",
1320 mop->mo_ldd.ldd_svname, mop->mo_source);
1321 rc = run_command(label_cmd, sizeof(label_cmd));
1326 int ldiskfs_rename_fsname(struct mkfs_opts *mop, const char *oldname)
1328 struct mount_opts opts;
1329 struct lustre_disk_data *ldd = &mop->mo_ldd;
1330 char mntpt[] = "/tmp/mntXXXXXX";
1334 /* Change the filesystem label. */
1336 opts.mo_source = mop->mo_device;
1337 ret = ldiskfs_label_lustre(&opts);
1341 fprintf(stderr, "Can't change filesystem label: %s\n",
1346 /* Mount this device temporarily in order to write these files */
1347 if (mkdtemp(mntpt) == NULL) {
1352 fprintf(stderr, "Can't create temp mount point %s: %s\n",
1353 mntpt, strerror(ret));
1359 * Append file context to mount options if SE Linux is enabled
1361 if (is_selinux_enabled() > 0)
1362 append_context_for_mount(mntpt, mop);
1365 if (mop->mo_flags & MO_IS_LOOP)
1366 dev = mop->mo_loopdev;
1368 dev = mop->mo_device;
1369 ret = mount(dev, mntpt, MT_STR(ldd), 0, ldd->ldd_mount_opts);
1373 fprintf(stderr, "Unable to mount %s: %s\n",
1374 dev, strerror(ret));
1376 fprintf(stderr, "Is the %s module available?\n",
1381 ret = lustre_rename_fsname(mop, mntpt, oldname);
1389 /* Enable quota accounting */
1390 int ldiskfs_enable_quota(struct mkfs_opts *mop)
1394 int cmdsz = sizeof(cmd), ret;
1396 if (is_e2fsprogs_feature_supp("-O quota") != 0) {
1397 fprintf(stderr, "%s: \"-O quota\" is is not supported by "
1398 "current e2fsprogs\n", progname);
1402 dev = mop->mo_device;
1403 if (mop->mo_flags & MO_IS_LOOP)
1404 dev = mop->mo_loopdev;
1406 /* Quota feature is already enabled? */
1407 if (is_feature_enabled("quota", dev)) {
1408 vprint("Quota feature is already enabled.\n");
1412 /* Turn on quota feature by "tune2fs -O quota" */
1413 snprintf(cmd, cmdsz, "%s -O quota %s", TUNE2FS, dev);
1414 ret = run_command(cmd, cmdsz);
1416 fprintf(stderr, "command:%s (%d)", cmd, ret);
1421 int ldiskfs_init(void)
1423 /* Required because full path to DEBUGFS is not specified */
1424 setenv("PATH", "/bin:/sbin:/usr/bin:/usr/sbin:/usr/local/bin", 0);
1429 void ldiskfs_fini(void)