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 */
60 #include <sys/types.h>
62 #include <sys/mount.h>
63 #include <sys/utsname.h>
71 #include <linux/fs.h> /* for BLKGETSIZE64 */
73 #include <linux/version.h>
74 #include <lustre_disk.h>
75 #include <lustre_param.h>
76 #include <lnet/lnetctl.h>
77 #include <lustre_ver.h>
80 #include <selinux/selinux.h>
83 #include "mount_utils.h"
85 #define MAX_HW_SECTORS_KB_PATH "queue/max_hw_sectors_kb"
86 #define MAX_SECTORS_KB_PATH "queue/max_sectors_kb"
87 #define SCHEDULER_PATH "queue/scheduler"
88 #define STRIPE_CACHE_SIZE "md/stripe_cache_size"
90 #define DEFAULT_SCHEDULER "deadline"
92 extern char *progname;
94 #define L_BLOCK_SIZE 4096
95 /* keep it less than LL_FID_NAMELEN */
96 #define DUMMY_FILE_NAME_LEN 25
97 #define EXT3_DIRENT_SIZE DUMMY_FILE_NAME_LEN
99 static void append_unique(char *buf, char *prefix, char *key, char *val,
103 * Concatenate context of the temporary mount point if selinux is enabled
106 static void append_context_for_mount(char *mntpt, struct mkfs_opts *mop)
108 security_context_t fcontext;
110 if (getfilecon(mntpt, &fcontext) < 0) {
111 /* Continuing with default behaviour */
112 fprintf(stderr, "%s: Get file context failed : %s\n",
113 progname, strerror(errno));
117 if (fcontext != NULL) {
118 append_unique(mop->mo_ldd.ldd_mount_opts,
119 ",", "context", fcontext,
120 sizeof(mop->mo_ldd.ldd_mount_opts));
126 /* return canonicalized absolute pathname, even if the target file does not
127 * exist, unlike realpath */
128 static char *absolute_path(char *devname)
130 char buf[PATH_MAX + 1] = "";
135 path = malloc(sizeof(buf));
139 if (devname[0] != '/') {
140 if (getcwd(buf, sizeof(buf) - 1) == NULL) {
144 len = snprintf(path, sizeof(buf), "%s/%s", buf, devname);
145 if (len >= sizeof(buf)) {
150 len = snprintf(path, sizeof(buf), "%s", devname);
151 if (len >= sizeof(buf)) {
157 /* truncate filename before calling realpath */
158 ptr = strrchr(path, '/');
164 if (buf != realpath(path, buf)) {
168 /* add the filename back */
169 len = snprintf(path, PATH_MAX, "%s/%s", buf, ptr+1);
170 if (len >= PATH_MAX) {
177 /* Determine if a device is a block device (as opposed to a file) */
178 static int is_block(char *devname)
184 devpath = absolute_path(devname);
185 if (devpath == NULL) {
186 fprintf(stderr, "%s: failed to resolve path to %s\n",
191 ret = access(devname, F_OK);
193 if (strncmp(devpath, "/dev/", 5) == 0) {
194 /* nobody sane wants to create a loopback file under
195 * /dev. Let's just report the device doesn't exist */
196 fprintf(stderr, "%s: %s apparently does not exist\n",
204 ret = stat(devpath, &st);
206 fprintf(stderr, "%s: cannot stat %s\n", progname, devpath);
209 ret = S_ISBLK(st.st_mode);
215 static int is_feature_enabled(const char *feature, const char *devpath)
219 char enabled_features[4096] = "";
222 snprintf(cmd, sizeof(cmd), "%s -R features %s 2>&1",
225 /* Using popen() instead of run_command() since debugfs does
226 * not return proper error code if command is not supported */
227 fp = popen(cmd, "r");
229 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
233 ret = fread(enabled_features, 1, sizeof(enabled_features) - 1, fp);
234 enabled_features[ret] = '\0';
237 if (strstr(enabled_features, feature))
242 /* Write the server config files */
243 int ldiskfs_write_ldd(struct mkfs_opts *mop)
245 char mntpt[] = "/tmp/mntXXXXXX";
252 /* Mount this device temporarily in order to write these files */
253 if (!mkdtemp(mntpt)) {
254 fprintf(stderr, "%s: Can't create temp mount point %s: %s\n",
255 progname, mntpt, strerror(errno));
260 * Append file context to mount options if SE Linux is enabled
263 if (is_selinux_enabled() > 0)
264 append_context_for_mount(mntpt, mop);
267 dev = mop->mo_device;
268 if (mop->mo_flags & MO_IS_LOOP)
269 dev = mop->mo_loopdev;
271 ret = mount(dev, mntpt, MT_STR(&mop->mo_ldd), 0,
272 (mop->mo_mountopts == NULL) ?
273 "errors=remount-ro" : mop->mo_mountopts);
275 fprintf(stderr, "%s: Unable to mount %s: %s\n",
276 progname, dev, strerror(errno));
278 if (errno == ENODEV) {
279 fprintf(stderr, "Is the %s module available?\n",
280 MT_STR(&mop->mo_ldd));
285 /* Set up initial directories */
286 sprintf(filepnm, "%s/%s", mntpt, MOUNT_CONFIGS_DIR);
287 ret = mkdir(filepnm, 0777);
288 if ((ret != 0) && (errno != EEXIST)) {
289 fprintf(stderr, "%s: Can't make configs dir %s (%s)\n",
290 progname, filepnm, strerror(errno));
292 } else if (errno == EEXIST) {
296 /* Save the persistent mount data into a file. Lustre must pre-read
297 this file to get the real mount options. */
298 vprint("Writing %s\n", MOUNT_DATA_FILE);
299 sprintf(filepnm, "%s/%s", mntpt, MOUNT_DATA_FILE);
300 filep = fopen(filepnm, "w");
302 fprintf(stderr, "%s: Unable to create %s file: %s\n",
303 progname, filepnm, strerror(errno));
306 num = fwrite(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
307 if (num < 1 && ferror(filep)) {
308 fprintf(stderr, "%s: Unable to write to file (%s): %s\n",
309 progname, filepnm, strerror(errno));
322 static int readcmd(char *cmd, char *buf, int len)
327 fp = popen(cmd, "r");
331 red = fread(buf, 1, len, fp);
334 /* strip trailing newline */
335 if (buf[red - 1] == '\n')
338 return (red == 0) ? -ENOENT : 0;
341 int ldiskfs_read_ldd(char *dev, struct lustre_disk_data *mo_ldd)
343 char tmpdir[] = "/tmp/dirXXXXXX";
348 int cmdsz = sizeof(cmd);
350 /* Make a temporary directory to hold Lustre data files. */
351 if (!mkdtemp(tmpdir)) {
352 fprintf(stderr, "%s: Can't create temporary directory %s: %s\n",
353 progname, tmpdir, strerror(errno));
357 /* TODO: it's worth observing the get_mountdata() function that is
358 in mount_utils.c for getting the mountdata out of the
361 /* Construct debugfs command line. */
362 snprintf(cmd, cmdsz, "%s -c -R 'dump /%s %s/mountdata' '%s'",
363 DEBUGFS, MOUNT_DATA_FILE, tmpdir, dev);
365 ret = run_command(cmd, cmdsz);
367 verrprint("%s: Unable to dump %s dir (%d)\n",
368 progname, MOUNT_CONFIGS_DIR, ret);
370 sprintf(filepnm, "%s/mountdata", tmpdir);
371 filep = fopen(filepnm, "r");
374 vprint("Reading %s\n", MOUNT_DATA_FILE);
375 num_read = fread(mo_ldd, sizeof(*mo_ldd), 1, filep);
376 if (num_read < 1 && ferror(filep)) {
377 fprintf(stderr, "%s: Unable to read from file %s: %s\n",
378 progname, filepnm, strerror(errno));
383 snprintf(cmd, cmdsz, "rm -rf %s", tmpdir);
384 run_command(cmd, cmdsz);
386 verrprint("Failed to read old data (%d)\n", ret);
388 /* As long as we at least have the label, we're good to go */
389 snprintf(cmd, sizeof(cmd), E2LABEL" %s", dev);
390 ret = readcmd(cmd, mo_ldd->ldd_svname, sizeof(mo_ldd->ldd_svname) - 1);
396 /* Display the need for the latest e2fsprogs to be installed. make_backfs
397 * indicates if the caller is make_lustre_backfs() or not. */
398 static void disp_old_e2fsprogs_msg(const char *feature, int make_backfs)
400 static int msg_displayed;
403 fprintf(stderr, "WARNING: %s does not support %s "
404 "feature.\n\n", E2FSPROGS, feature);
410 fprintf(stderr, "WARNING: The %s package currently installed on "
411 "your system does not support \"%s\" feature.\n",
413 #if !(HAVE_LDISKFSPROGS)
414 fprintf(stderr, "Please install the latest version of e2fsprogs from\n"
415 "https://downloads.hpdd.intel.com/public/e2fsprogs/latest/\n"
416 "to enable this feature.\n");
419 fprintf(stderr, "Feature will not be enabled until %s"
420 "is updated and '%s -O %s %%{device}' "
421 "is run.\n\n", E2FSPROGS, TUNE2FS, feature);
424 /* Check whether the file exists in the device */
425 static int file_in_dev(char *file_name, char *dev_name)
428 char debugfs_cmd[256];
429 unsigned int inode_num;
432 /* Construct debugfs command line. */
433 snprintf(debugfs_cmd, sizeof(debugfs_cmd),
434 "%s -c -R 'stat %s' '%s' 2>&1 | egrep '(Inode|unsupported)'",
435 DEBUGFS, file_name, dev_name);
437 fp = popen(debugfs_cmd, "r");
439 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
443 if (fscanf(fp, "Inode: %u", &inode_num) == 1) { /* exist */
447 i = fread(debugfs_cmd, 1, sizeof(debugfs_cmd) - 1, fp);
450 fprintf(stderr, "%s", debugfs_cmd);
451 if (strstr(debugfs_cmd, "unsupported feature")) {
452 disp_old_e2fsprogs_msg("an unknown", 0);
461 /* Check whether the device has already been used with lustre */
462 int ldiskfs_is_lustre(char *dev, unsigned *mount_type)
466 ret = file_in_dev(MOUNT_DATA_FILE, dev);
468 /* in the -1 case, 'extents' means IS a lustre target */
469 *mount_type = LDD_MT_LDISKFS;
473 ret = file_in_dev(LAST_RCVD, dev);
475 *mount_type = LDD_MT_LDISKFS;
482 /* Check if a certain feature is supported by e2fsprogs.
483 * Firstly we try to use "debugfs supported_features" command to check if
484 * the feature is supported. If this fails we try to set this feature with
485 * mke2fs to check for its support. */
486 static int is_e2fsprogs_feature_supp(const char *feature)
488 static char supp_features[4096] = "";
491 char imgname[] = "/tmp/test-img-XXXXXX";
495 if (supp_features[0] == '\0') {
496 snprintf(cmd, sizeof(cmd), "%s -c -R supported_features 2>&1",
499 /* Using popen() instead of run_command() since debugfs does
500 * not return proper error code if command is not supported */
501 fp = popen(cmd, "r");
503 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
506 ret = fread(supp_features, 1, sizeof(supp_features) - 1, fp);
507 supp_features[ret] = '\0';
510 if (ret > 0 && strstr(supp_features,
511 strncmp(feature, "-O ", 3) ? feature : feature+3))
514 if ((fd = mkstemp(imgname)) < 0)
519 snprintf(cmd, sizeof(cmd), "%s -F %s %s 100 >/dev/null 2>&1",
520 MKE2FS, feature, imgname);
521 /* run_command() displays the output of mke2fs when it fails for
522 * some feature, so use system() directly */
531 * append_unique: append @key or @key=@val pair to @buf only if @key does not
533 * @buf: buffer to hold @key or @key=@val
534 * @prefix: prefix string before @key
536 * @val: value string if it's a @key=@val pair
538 static void append_unique(char *buf, char *prefix, char *key, char *val,
547 anchor = end = strstr(buf, key);
548 /* try to find exact match string in @buf */
549 while (end && *end != '\0' && *end != ',' && *end != ' ' && *end != '=')
552 if (anchor == NULL || strlen(key) != len ||
553 strncmp(anchor, key, len) != 0) {
555 strscat(buf, prefix, maxbuflen);
557 strscat(buf, key, maxbuflen);
559 strscat(buf, "=\"", maxbuflen);
560 strscat(buf, val, maxbuflen);
561 strscat(buf, "\"", maxbuflen);
566 static int enable_default_ext4_features(struct mkfs_opts *mop, char *anchor,
567 size_t maxbuflen, int user_spec)
569 if (IS_OST(&mop->mo_ldd)) {
570 append_unique(anchor, user_spec ? "," : " -O ",
571 "extents", NULL, maxbuflen);
572 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
573 } else if (IS_MDT(&mop->mo_ldd)) {
574 append_unique(anchor, user_spec ? "," : " -O ",
575 "dirdata", NULL, maxbuflen);
576 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
577 append_unique(anchor, ",", "^extents", NULL, maxbuflen);
579 append_unique(anchor, user_spec ? "," : " -O ",
580 "uninit_bg", NULL, maxbuflen);
583 /* Multiple mount protection enabled only if failover node specified */
584 if (mop->mo_flags & MO_FAILOVER) {
585 if (is_e2fsprogs_feature_supp("-O mmp") == 0)
586 append_unique(anchor, ",", "mmp", NULL, maxbuflen);
588 disp_old_e2fsprogs_msg("mmp", 1);
591 /* Allow more than 65000 subdirectories */
592 if (is_e2fsprogs_feature_supp("-O dir_nlink") == 0)
593 append_unique(anchor, ",", "dir_nlink", NULL, maxbuflen);
595 /* The following options are only valid for ext4-based ldiskfs.
596 * If --backfstype=ext3 is specified, do not enable them. */
597 if (mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3)
600 /* Enable quota by default */
601 if (is_e2fsprogs_feature_supp("-O quota") == 0) {
602 append_unique(anchor, ",", "quota", NULL, maxbuflen);
605 fprintf(stderr, "\"-O quota\" must be supported by "
606 "e2fsprogs, please upgrade your e2fsprogs.\n");
610 /* Allow files larger than 2TB. Also needs LU-16, but not harmful. */
611 if (is_e2fsprogs_feature_supp("-O huge_file") == 0)
612 append_unique(anchor, ",", "huge_file", NULL, maxbuflen);
614 /* Enable large block addresses if the LUN is over 2^32 blocks. */
615 if (mop->mo_device_kb / (L_BLOCK_SIZE >> 10) >= 0x100002000ULL &&
616 is_e2fsprogs_feature_supp("-O 64bit") == 0)
617 append_unique(anchor, ",", "64bit", NULL, maxbuflen);
619 /* Cluster inode/block bitmaps and inode table for more efficient IO.
620 * Align the flex groups on a 1MB boundary for better performance. */
621 /* This -O feature needs to go last, since it adds the "-G" option. */
622 if (is_e2fsprogs_feature_supp("-O flex_bg") == 0) {
625 append_unique(anchor, ",", "flex_bg", NULL, maxbuflen);
627 if (IS_OST(&mop->mo_ldd) &&
628 strstr(mop->mo_mkfsopts, "-G") == NULL) {
629 snprintf(tmp_buf, sizeof(tmp_buf), " -G %u",
630 (1 << 20) / L_BLOCK_SIZE);
631 strscat(anchor, tmp_buf, maxbuflen);
634 /* Don't add any more "-O" options here, see last comment above */
639 * moveopts_to_end: find the option string, move remaining strings to
640 * where option string starts, and append the option
642 * @start: where the option string starts before the move
643 * RETURN: where the option string starts after the move
645 static char *moveopts_to_end(char *start)
651 /* skip whitespace before options */
656 /* find end of option characters */
657 while (*end != ' ' && *end != '\0')
661 if (len >= sizeof(save))
662 len = sizeof(save) - 1;
665 strncpy(save, start, len);
668 /* move remaining options up front */
670 memmove(start, end, strlen(end));
671 *(start + strlen(end)) = '\0';
673 /* append the specified options */
674 if (*(start + strlen(start) - 1) != ' ')
676 idx = start + strlen(start);
682 /* Build fs according to type */
683 int ldiskfs_make_lustre(struct mkfs_opts *mop)
685 __u64 device_kb = mop->mo_device_kb, block_count = 0;
686 char mkfs_cmd[PATH_MAX];
690 int ret = 0, ext_opts = 0;
693 if (!(mop->mo_flags & MO_IS_LOOP)) {
694 mop->mo_device_kb = get_device_size(mop->mo_device);
696 if (mop->mo_device_kb == 0)
699 /* Compare to real size */
700 if (device_kb == 0 || device_kb > mop->mo_device_kb)
701 device_kb = mop->mo_device_kb;
703 mop->mo_device_kb = device_kb;
706 if (mop->mo_device_kb != 0) {
707 if (mop->mo_device_kb < 32384) {
708 fprintf(stderr, "%s: size of filesystem must be larger "
709 "than 32MB, but is set to %lldKB\n",
710 progname, (long long)mop->mo_device_kb);
713 block_count = mop->mo_device_kb / (L_BLOCK_SIZE >> 10);
714 /* If the LUN size is just over 2^32 blocks, limit the
715 * filesystem size to 2^32-1 blocks to avoid problems with
716 * ldiskfs/mkfs not handling this size. Bug 22906 */
717 if (block_count > 0xffffffffULL && block_count < 0x100002000ULL)
718 block_count = 0xffffffffULL;
721 if ((mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3) ||
722 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS) ||
723 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS2)) {
726 /* Journal size in MB */
727 if (strstr(mop->mo_mkfsopts, "-J") == NULL &&
728 device_kb > 1024 * 1024) {
729 /* Choose our own default journal size */
730 long journal_mb = 0, max_mb;
732 /* cap journal size at 4GB for MDT,
733 * leave it at 400MB for OSTs. */
734 if (IS_MDT(&mop->mo_ldd))
736 else if (IS_OST(&mop->mo_ldd))
738 else /* Use mke2fs default size for MGS */
741 /* Use at most 4% of device for journal */
742 journal_mb = device_kb * 4 / (1024 * 100);
743 if (journal_mb > max_mb)
747 sprintf(buf, " -J size=%ld", journal_mb);
748 strscat(mop->mo_mkfsopts, buf,
749 sizeof(mop->mo_mkfsopts));
753 /* Inode size includes:
754 * ldiskfs inode size: 156
755 * extended attributes size, including:
756 * ext4_xattr_header: 32
757 * LOV EA size: 32(lov_mds_md) +
758 * stripes * 24(lov_ost_data) +
759 * 16(xattr_entry) + 3(lov)
760 * LMA EA size: 24(lustre_mdt_attrs) +
761 * 16(xattr_entry) + 3(lma)
762 * link EA size: 24(link_ea_header) + 18(link_ea_entry) +
763 * (filename) + 16(xattr_entry) + 4(link)
764 * and some margin for 4-byte alignment, ACLs and other EAs.
766 * If we say the average filename length is about 32 bytes,
767 * the calculation looks like:
768 * 156 + 32 + (32+24*N+19) + (24+19) + (24+18+~32+20) + other <=
769 * 512*2^m, {m=0,1,2,3}
771 if (strstr(mop->mo_mkfsopts, "-I") == NULL) {
772 if (IS_MDT(&mop->mo_ldd)) {
773 if (mop->mo_stripe_count > 69)
774 inode_size = 512; /* bz 7241 */
775 /* see also "-i" below for EA blocks */
776 else if (mop->mo_stripe_count > 26)
778 else if (mop->mo_stripe_count > 5)
782 } else if (IS_OST(&mop->mo_ldd)) {
783 /* We store MDS FID and OST objid in EA on OST
784 * we need to make inode bigger as well. */
788 if (inode_size > 0) {
789 sprintf(buf, " -I %ld", inode_size);
790 strscat(mop->mo_mkfsopts, buf,
791 sizeof(mop->mo_mkfsopts));
795 /* Bytes_per_inode: disk size / num inodes */
796 if (strstr(mop->mo_mkfsopts, "-i") == NULL &&
797 strstr(mop->mo_mkfsopts, "-N") == NULL) {
798 long bytes_per_inode = 0;
800 /* Allocate more inodes on MDT devices. There is
801 * no data stored on the MDT, and very little extra
802 * metadata beyond the inode. It could go down as
803 * low as 1024 bytes, but this is conservative.
804 * Account for external EA blocks for wide striping. */
805 if (IS_MDT(&mop->mo_ldd)) {
806 bytes_per_inode = inode_size + 1536;
808 if (mop->mo_stripe_count > 69) {
809 int extra = mop->mo_stripe_count * 24;
810 extra = ((extra - 1) | 4095) + 1;
811 bytes_per_inode += extra;
815 /* Allocate fewer inodes on large OST devices. Most
816 * filesystems can be much more aggressive than even
817 * this, but it is impossible to know in advance. */
818 if (IS_OST(&mop->mo_ldd)) {
819 /* OST > 16TB assume average file size 1MB */
820 if (device_kb > (16ULL << 30))
821 bytes_per_inode = 1024 * 1024;
822 /* OST > 4TB assume average file size 512kB */
823 else if (device_kb > (4ULL << 30))
824 bytes_per_inode = 512 * 1024;
825 /* OST > 1TB assume average file size 256kB */
826 else if (device_kb > (1ULL << 30))
827 bytes_per_inode = 256 * 1024;
828 /* OST > 10GB assume average file size 64kB,
829 * plus a bit so that inodes will fit into a
830 * 256x flex_bg without overflowing */
831 else if (device_kb > (10ULL << 20))
832 bytes_per_inode = 69905;
835 if (bytes_per_inode > 0) {
836 sprintf(buf, " -i %ld", bytes_per_inode);
837 strscat(mop->mo_mkfsopts, buf,
838 sizeof(mop->mo_mkfsopts));
843 strscat(mop->mo_mkfsopts, " -q",
844 sizeof(mop->mo_mkfsopts));
847 /* start handle -O mkfs options */
848 if ((start = strstr(mop->mo_mkfsopts, "-O")) != NULL) {
849 if (strstr(start + 2, "-O") != NULL) {
851 "%s: don't specify multiple -O options\n",
855 start = moveopts_to_end(start);
856 maxbuflen = sizeof(mop->mo_mkfsopts) -
857 (start - mop->mo_mkfsopts) - strlen(start);
858 ret = enable_default_ext4_features(mop, start, maxbuflen, 1);
860 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts),
861 maxbuflen = sizeof(mop->mo_mkfsopts) -
862 strlen(mop->mo_mkfsopts);
863 ret = enable_default_ext4_features(mop, start, maxbuflen, 0);
867 /* end handle -O mkfs options */
869 /* start handle -E mkfs options */
870 if ((start = strstr(mop->mo_mkfsopts, "-E")) != NULL) {
871 if (strstr(start + 2, "-E") != NULL) {
873 "%s: don't specify multiple -E options\n",
877 start = moveopts_to_end(start);
878 maxbuflen = sizeof(mop->mo_mkfsopts) -
879 (start - mop->mo_mkfsopts) - strlen(start);
882 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts);
883 maxbuflen = sizeof(mop->mo_mkfsopts) -
884 strlen(mop->mo_mkfsopts);
887 /* In order to align the filesystem metadata on 1MB boundaries,
888 * give a resize value that will reserve a power-of-two group
889 * descriptor blocks, but leave one block for the superblock.
890 * Only useful for filesystems with < 2^32 blocks due to resize
892 if (strstr(mop->mo_mkfsopts, "meta_bg") == NULL &&
893 IS_OST(&mop->mo_ldd) && mop->mo_device_kb > 100 * 1024 &&
894 mop->mo_device_kb * 1024 / L_BLOCK_SIZE <= 0xffffffffULL) {
895 unsigned group_blocks = L_BLOCK_SIZE * 8;
896 unsigned desc_per_block = L_BLOCK_SIZE / 32;
897 unsigned resize_blks;
899 resize_blks = (1ULL<<32) - desc_per_block*group_blocks;
900 snprintf(buf, sizeof(buf), "%u", resize_blks);
901 append_unique(start, ext_opts ? "," : " -E ",
902 "resize", buf, maxbuflen);
906 /* Avoid zeroing out the full journal - speeds up mkfs */
907 if (is_e2fsprogs_feature_supp("-E lazy_journal_init") == 0)
908 append_unique(start, ext_opts ? "," : " -E ",
909 "lazy_journal_init", NULL, maxbuflen);
910 /* end handle -E mkfs options */
912 /* Allow reformat of full devices (as opposed to
913 partitions.) We already checked for mounted dev. */
914 strscat(mop->mo_mkfsopts, " -F", sizeof(mop->mo_mkfsopts));
916 snprintf(mkfs_cmd, sizeof(mkfs_cmd),
917 "%s -j -b %d -L %s ", MKE2FS, L_BLOCK_SIZE,
918 mop->mo_ldd.ldd_svname);
920 fprintf(stderr,"%s: unsupported fs type: %d (%s)\n",
921 progname, mop->mo_ldd.ldd_mount_type,
922 MT_STR(&mop->mo_ldd));
926 /* For loop device format the dev, not the filename */
927 dev = mop->mo_device;
928 if (mop->mo_flags & MO_IS_LOOP)
929 dev = mop->mo_loopdev;
931 vprint("formatting backing filesystem %s on %s\n",
932 MT_STR(&mop->mo_ldd), dev);
933 vprint("\ttarget name %s\n", mop->mo_ldd.ldd_svname);
934 vprint("\t4k blocks %ju\n", (uintmax_t)block_count);
935 vprint("\toptions %s\n", mop->mo_mkfsopts);
937 /* mkfs_cmd's trailing space is important! */
938 strscat(mkfs_cmd, mop->mo_mkfsopts, sizeof(mkfs_cmd));
939 strscat(mkfs_cmd, " ", sizeof(mkfs_cmd));
940 strscat(mkfs_cmd, dev, sizeof(mkfs_cmd));
941 if (block_count != 0) {
942 snprintf(buf, sizeof(buf), " %ju",
943 (uintmax_t)block_count);
944 strscat(mkfs_cmd, buf, sizeof(mkfs_cmd));
947 vprint("mkfs_cmd = %s\n", mkfs_cmd);
948 ret = run_command(mkfs_cmd, sizeof(mkfs_cmd));
951 fprintf(stderr, "Unable to build fs %s (%d)\n", dev, ret);
956 int ldiskfs_prepare_lustre(struct mkfs_opts *mop,
957 char *wanted_mountopts, size_t len)
959 struct lustre_disk_data *ldd = &mop->mo_ldd;
962 /* Set MO_IS_LOOP to indicate a loopback device is needed */
963 ret = is_block(mop->mo_device);
966 } else if (ret == 0) {
967 mop->mo_flags |= MO_IS_LOOP;
970 if (IS_MDT(ldd) || IS_MGS(ldd))
971 strscat(wanted_mountopts, ",user_xattr", len);
976 int ldiskfs_fix_mountopts(struct mkfs_opts *mop, char *mountopts, size_t len)
978 if (strstr(mountopts, "errors=") == NULL)
979 strscat(mountopts, ",errors=remount-ro", len);
984 static int read_file(const char *path, char *buf, int size)
988 fd = fopen(path, "r");
992 if (fgets(buf, size, fd) == NULL) {
993 fprintf(stderr, "reading from %s: %s", path, strerror(errno));
999 /* strip trailing newline */
1001 if (buf[size - 1] == '\n')
1002 buf[size - 1] = '\0';
1007 static int write_file(const char *path, const char *buf)
1011 fd = open(path, O_WRONLY);
1015 rc = write(fd, buf, strlen(buf));
1018 return rc < 0 ? errno : 0;
1021 static int set_blockdev_scheduler(const char *path, const char *scheduler)
1023 char buf[PATH_MAX], *s, *e, orig_sched[50];
1026 /* Before setting the scheduler, we need to check to see if it's
1027 * already set to "noop". If it is, we don't want to override
1028 * that setting. If it's set to anything other than "noop", set
1029 * the scheduler to what has been passed in. */
1031 rc = read_file(path, buf, sizeof(buf));
1034 fprintf(stderr, "%s: cannot open '%s': %s\n",
1035 progname, path, strerror(errno));
1039 /* The expected format of buf: noop anticipatory deadline [cfq] */
1040 s = strchr(buf, '[');
1041 e = strchr(buf, ']');
1043 /* If the format is not what we expect. Play it safe and error out. */
1044 if (s == NULL || e == NULL) {
1046 fprintf(stderr, "%s: cannot parse scheduler "
1047 "options for '%s'\n", progname, path);
1051 snprintf(orig_sched, e - s, "%s", s + 1);
1053 if (strcmp(orig_sched, "noop") == 0 ||
1054 strcmp(orig_sched, scheduler) == 0)
1057 rc = write_file(path, scheduler);
1060 fprintf(stderr, "%s: cannot set scheduler on "
1061 "'%s': %s\n", progname, path,
1065 fprintf(stderr, "%s: change scheduler of %s from %s to %s\n",
1066 progname, path, orig_sched, scheduler);
1072 /* This is to tune the kernel for good SCSI performance.
1073 * For that we set the value of /sys/block/{dev}/queue/max_sectors_kb
1074 * to the value of /sys/block/{dev}/queue/max_hw_sectors_kb */
1075 static int set_blockdev_tunables(char *source, struct mount_opts *mop)
1077 glob_t glob_info = { 0 };
1078 struct stat stat_buf;
1079 char *chk_major, *chk_minor;
1080 char *savept = NULL, *dev;
1082 char buf[PATH_MAX] = {'\0'}, path[PATH_MAX] = {'\0'};
1083 char real_path[PATH_MAX] = {'\0'};
1091 ret_path = realpath(source, real_path);
1092 if (ret_path == NULL) {
1094 fprintf(stderr, "warning: %s: cannot resolve: %s\n",
1095 source, strerror(errno));
1099 if (strncmp(real_path, "/dev/loop", 9) == 0)
1102 if ((real_path[0] != '/') && (strpbrk(real_path, ",:") != NULL))
1105 snprintf(path, sizeof(path), "/sys/block%s", real_path + 4);
1106 if (access(path, X_OK) == 0)
1109 /* The name of the device say 'X' specified in /dev/X may not
1110 * match any entry under /sys/block/. In that case we need to
1111 * match the major/minor number to find the entry under
1112 * sys/block corresponding to /dev/X */
1114 /* Don't chop tail digit on /dev/mapper/xxx, LU-478 */
1115 if (strncmp(real_path, "/dev/mapper", 11) != 0) {
1116 dev = real_path + strlen(real_path);
1117 while (--dev > real_path && isdigit(*dev))
1120 if (strncmp(real_path, "/dev/md_", 8) == 0)
1124 rc = stat(real_path, &stat_buf);
1127 fprintf(stderr, "warning: %s, device %s stat failed\n",
1128 strerror(errno), real_path);
1132 major = major(stat_buf.st_rdev);
1133 minor = minor(stat_buf.st_rdev);
1134 rc = glob("/sys/block/*", GLOB_NOSORT, NULL, &glob_info);
1137 fprintf(stderr, "warning: failed to read entries under "
1139 globfree(&glob_info);
1143 for (i = 0; i < glob_info.gl_pathc; i++){
1144 snprintf(path, sizeof(path), "%s/dev", glob_info.gl_pathv[i]);
1146 rc = read_file(path, buf, sizeof(buf));
1150 if (buf[strlen(buf) - 1] == '\n')
1151 buf[strlen(buf) - 1] = '\0';
1153 chk_major = strtok_r(buf, ":", &savept);
1155 if (chk_major != NULL && major == atoi(chk_major) &&
1156 chk_minor != NULL && minor == atoi(chk_minor))
1160 if (i == glob_info.gl_pathc) {
1162 fprintf(stderr,"warning: device %s does not match any "
1163 "entry under /sys/block\n", real_path);
1164 globfree(&glob_info);
1168 /* Chop off "/dev" from path we found */
1169 path[strlen(glob_info.gl_pathv[i])] = '\0';
1170 globfree(&glob_info);
1173 if (strncmp(real_path, "/dev/md", 7) == 0) {
1174 snprintf(real_path, sizeof(real_path), "%s/%s", path,
1177 rc = read_file(real_path, buf, sizeof(buf));
1180 fprintf(stderr, "warning: opening %s: %s\n",
1181 real_path, strerror(errno));
1185 if (atoi(buf) >= mop->mo_md_stripe_cache_size)
1188 if (strlen(buf) - 1 > 0) {
1189 snprintf(buf, sizeof(buf), "%d",
1190 mop->mo_md_stripe_cache_size);
1191 rc = write_file(real_path, buf);
1192 if (rc != 0 && verbose)
1193 fprintf(stderr, "warning: opening %s: %s\n",
1194 real_path, strerror(errno));
1196 /* Return since raid and disk tunables are different */
1200 snprintf(real_path, sizeof(real_path), "%s/%s", path,
1201 MAX_HW_SECTORS_KB_PATH);
1202 rc = read_file(real_path, buf, sizeof(buf));
1205 fprintf(stderr, "warning: opening %s: %s\n",
1206 real_path, strerror(errno));
1207 /* No MAX_HW_SECTORS_KB_PATH isn't necessary an
1208 * error for some device. */
1212 if (strlen(buf) - 1 > 0) {
1213 char oldbuf[32] = "", *end = NULL;
1214 unsigned long long oldval, newval;
1216 snprintf(real_path, sizeof(real_path), "%s/%s", path,
1217 MAX_SECTORS_KB_PATH);
1218 rc = read_file(real_path, oldbuf, sizeof(oldbuf));
1219 /* Only set new parameter if different from the old one. */
1220 if (rc != 0 || strcmp(oldbuf, buf) == 0) {
1221 /* No MAX_SECTORS_KB_PATH isn't necessary an
1222 * error for some device. */
1226 newval = strtoull(buf, &end, 0);
1227 if (newval == 0 || newval == ULLONG_MAX || end == buf)
1230 /* Don't increase IO request size limit past 16MB. It is about
1231 * PTLRPC_MAX_BRW_SIZE, but that isn't in a public header.
1232 * Note that even though the block layer allows larger values,
1233 * setting max_sectors_kb = 32768 causes crashes (LU-6974). */
1234 if (newval > 16 * 1024) {
1236 snprintf(buf, sizeof(buf), "%llu", newval);
1239 oldval = strtoull(oldbuf, &end, 0);
1240 /* Don't shrink the current limit. */
1241 if (oldval != ULLONG_MAX && newval <= oldval)
1244 rc = write_file(real_path, buf);
1247 fprintf(stderr, "warning: writing to %s: %s\n",
1248 real_path, strerror(errno));
1249 /* No MAX_SECTORS_KB_PATH isn't necessary an
1250 * error for some device. */
1253 fprintf(stderr, "%s: increased %s from %s to %s\n",
1254 progname, real_path, oldbuf, buf);
1258 /* Purposely ignore errors reported from set_blockdev_scheduler.
1259 * The worst that will happen is a block device with an "incorrect"
1261 snprintf(real_path, sizeof(real_path), "%s/%s", path, SCHEDULER_PATH);
1262 set_blockdev_scheduler(real_path, DEFAULT_SCHEDULER);
1264 /* if device is multipath device, tune its slave devices */
1265 glob_info.gl_pathc = 0;
1266 glob_info.gl_offs = 0;
1267 snprintf(real_path, sizeof(real_path), "%s/slaves/*", path);
1268 rc = glob(real_path, GLOB_NOSORT, NULL, &glob_info);
1270 for (i = 0; rc == 0 && i < glob_info.gl_pathc; i++) {
1271 slave = basename(glob_info.gl_pathv[i]);
1272 snprintf(real_path, sizeof(real_path), "/dev/%s", slave);
1273 rc = set_blockdev_tunables(real_path, mop);
1276 if (rc == GLOB_NOMATCH) {
1277 /* no slave device is not an error */
1279 } else if (rc && verbose) {
1280 if (slave == NULL) {
1281 fprintf(stderr, "warning: %s, failed to read"
1282 " entries under %s/slaves\n",
1283 strerror(errno), path);
1285 fprintf(stderr, "unable to set tunables for"
1286 " slave device %s (slave would be"
1287 " unable to handle IO request from"
1292 globfree(&glob_info);
1297 int ldiskfs_tune_lustre(char *dev, struct mount_opts *mop)
1299 return set_blockdev_tunables(dev, mop);
1302 int ldiskfs_label_lustre(struct mount_opts *mop)
1304 char label_cmd[PATH_MAX];
1307 snprintf(label_cmd, sizeof(label_cmd),
1308 TUNE2FS" -f -L '%s' '%s' >/dev/null 2>&1",
1309 mop->mo_ldd.ldd_svname, mop->mo_source);
1310 rc = run_command(label_cmd, sizeof(label_cmd));
1315 /* Enable quota accounting */
1316 int ldiskfs_enable_quota(struct mkfs_opts *mop)
1320 int cmdsz = sizeof(cmd), ret;
1322 if (is_e2fsprogs_feature_supp("-O quota") != 0) {
1323 fprintf(stderr, "%s: \"-O quota\" is is not supported by "
1324 "current e2fsprogs\n", progname);
1328 dev = mop->mo_device;
1329 if (mop->mo_flags & MO_IS_LOOP)
1330 dev = mop->mo_loopdev;
1332 /* Quota feature is already enabled? */
1333 if (is_feature_enabled("quota", dev)) {
1334 vprint("Quota feature is already enabled.\n");
1338 /* Turn on quota feature by "tune2fs -O quota" */
1339 snprintf(cmd, cmdsz, "%s -O quota %s", TUNE2FS, dev);
1340 ret = run_command(cmd, cmdsz);
1342 fprintf(stderr, "command:%s (%d)", cmd, ret);
1347 int ldiskfs_init(void)
1349 /* Required because full path to DEBUGFS is not specified */
1350 setenv("PATH", "/bin:/sbin:/usr/bin:/usr/sbin:/usr/local/bin", 0);
1355 void ldiskfs_fini(void)