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/major.h>
70 #include <linux/types.h>
71 #include <linux/version.h>
72 #include <linux/lnet/lnetctl.h>
73 #include <linux/lustre/lustre_ver.h>
74 #include <libcfs/util/string.h>
77 #include <selinux/selinux.h>
80 #include "mount_utils.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,
98 static int is_e2fsprogs_feature_supp(const char *feature);
99 static void disp_old_e2fsprogs_msg(const char *feature, int make_backfs);
102 * Concatenate context of the temporary mount point if selinux is enabled
105 static void append_context_for_mount(char *mntpt, struct mkfs_opts *mop)
107 security_context_t fcontext;
109 if (getfilecon(mntpt, &fcontext) < 0) {
110 /* Continuing with default behaviour */
111 fprintf(stderr, "%s: Get file context failed : %s\n",
112 progname, strerror(errno));
116 if (fcontext != NULL) {
117 append_unique(mop->mo_ldd.ldd_mount_opts,
118 ",", "context", fcontext,
119 sizeof(mop->mo_ldd.ldd_mount_opts));
125 /* Determine if a device is a block device (as opposed to a file) */
126 static int is_block(char *devname)
132 ret = cfs_abs_path(devname, &devpath);
134 fprintf(stderr, "%s: failed to resolve path '%s': %s\n",
135 progname, devname, strerror(-ret));
139 ret = access(devname, F_OK);
141 if (strncmp(devpath, "/dev/", 5) == 0) {
142 /* nobody sane wants to create a loopback file under
143 * /dev. Let's just report the device doesn't exist */
144 fprintf(stderr, "%s: %s apparently does not exist\n",
152 ret = stat(devpath, &st);
154 fprintf(stderr, "%s: cannot stat %s\n", progname, devpath);
157 ret = S_ISBLK(st.st_mode);
163 static int is_feature_enabled(const char *feature, const char *devpath)
167 char enabled_features[4096] = "";
170 snprintf(cmd, sizeof(cmd), "%s -c -R features %s 2>&1",
173 /* Using popen() instead of run_command() since debugfs does
174 * not return proper error code if command is not supported */
175 fp = popen(cmd, "r");
177 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
181 ret = fread(enabled_features, 1, sizeof(enabled_features) - 1, fp);
182 enabled_features[ret] = '\0';
185 if (strstr(enabled_features, feature))
190 /* Write the server config files */
191 int ldiskfs_write_ldd(struct mkfs_opts *mop)
193 char mntpt[] = "/tmp/mntXXXXXX";
200 /* Mount this device temporarily in order to write these files */
201 if (!mkdtemp(mntpt)) {
202 fprintf(stderr, "%s: Can't create temp mount point %s: %s\n",
203 progname, mntpt, strerror(errno));
208 * Append file context to mount options if SE Linux is enabled
211 if (is_selinux_enabled() > 0)
212 append_context_for_mount(mntpt, mop);
215 dev = mop->mo_device;
216 if (mop->mo_flags & MO_IS_LOOP)
217 dev = mop->mo_loopdev;
219 /* Multiple mount protection enabled if failover node specified */
220 if (mop->mo_flags & MO_FAILOVER &&
221 !is_feature_enabled("mmp", dev)) {
222 if (is_e2fsprogs_feature_supp("-O mmp") == 0) {
223 char *command = filepnm;
225 snprintf(command, sizeof(filepnm),
226 TUNE2FS" -O mmp '%s' >/dev/null 2>&1", dev);
227 ret = run_command(command, sizeof(filepnm));
230 "%s: Unable to set 'mmp' on %s: %d\n",
233 disp_old_e2fsprogs_msg("mmp", 1);
236 ret = mount(dev, mntpt, MT_STR(&mop->mo_ldd), 0,
237 (mop->mo_mountopts == NULL) ?
238 "errors=remount-ro" : mop->mo_mountopts);
240 fprintf(stderr, "%s: Unable to mount %s: %s\n",
241 progname, dev, strerror(errno));
243 if (errno == ENODEV) {
244 fprintf(stderr, "Is the %s module available?\n",
245 MT_STR(&mop->mo_ldd));
250 /* Set up initial directories */
251 sprintf(filepnm, "%s/%s", mntpt, MOUNT_CONFIGS_DIR);
252 ret = mkdir(filepnm, 0777);
253 if ((ret != 0) && (errno != EEXIST)) {
254 fprintf(stderr, "%s: Can't make configs dir %s (%s)\n",
255 progname, filepnm, strerror(errno));
257 } else if (errno == EEXIST) {
261 /* Save the persistent mount data into a file. Lustre must pre-read
262 this file to get the real mount options. */
263 vprint("Writing %s\n", MOUNT_DATA_FILE);
264 sprintf(filepnm, "%s/%s", mntpt, MOUNT_DATA_FILE);
265 filep = fopen(filepnm, "w");
267 fprintf(stderr, "%s: Unable to create %s file: %s\n",
268 progname, filepnm, strerror(errno));
271 num = fwrite(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
272 if (num < 1 && ferror(filep)) {
273 fprintf(stderr, "%s: Unable to write to file (%s): %s\n",
274 progname, filepnm, strerror(errno));
278 fsync(filep->_fileno);
288 static int readcmd(char *cmd, char *buf, int len)
293 fp = popen(cmd, "r");
297 red = fread(buf, 1, len, fp);
300 /* strip trailing newline */
301 if (buf[red - 1] == '\n')
304 return (red == 0) ? -ENOENT : 0;
307 int ldiskfs_read_ldd(char *dev, struct lustre_disk_data *mo_ldd)
309 char tmpdir[] = "/tmp/dirXXXXXX";
314 int cmdsz = sizeof(cmd);
316 /* Make a temporary directory to hold Lustre data files. */
317 if (!mkdtemp(tmpdir)) {
318 fprintf(stderr, "%s: Can't create temporary directory %s: %s\n",
319 progname, tmpdir, strerror(errno));
323 /* TODO: it's worth observing the get_mountdata() function that is
324 in mount_utils.c for getting the mountdata out of the
327 /* Construct debugfs command line. */
328 snprintf(cmd, cmdsz, "%s -c -R 'dump /%s %s/mountdata' '%s'",
329 DEBUGFS, MOUNT_DATA_FILE, tmpdir, dev);
331 ret = run_command(cmd, cmdsz);
333 verrprint("%s: Unable to dump %s dir (%d)\n",
334 progname, MOUNT_CONFIGS_DIR, ret);
336 sprintf(filepnm, "%s/mountdata", tmpdir);
337 filep = fopen(filepnm, "r");
340 vprint("Reading %s\n", MOUNT_DATA_FILE);
341 num_read = fread(mo_ldd, sizeof(*mo_ldd), 1, filep);
342 if (num_read < 1 && ferror(filep)) {
343 fprintf(stderr, "%s: Unable to read from file %s: %s\n",
344 progname, filepnm, strerror(errno));
349 snprintf(cmd, cmdsz, "rm -rf %s", tmpdir);
350 run_command(cmd, cmdsz);
352 verrprint("Failed to read old data (%d)\n", ret);
354 /* As long as we at least have the label, we're good to go */
355 snprintf(cmd, sizeof(cmd), E2LABEL" %s", dev);
356 ret = readcmd(cmd, mo_ldd->ldd_svname, sizeof(mo_ldd->ldd_svname) - 1);
361 int ldiskfs_erase_ldd(struct mkfs_opts *mop, char *param)
366 void ldiskfs_print_ldd_params(struct mkfs_opts *mop)
368 printf("Parameters:%s\n", mop->mo_ldd.ldd_params);
371 /* Display the need for the latest e2fsprogs to be installed. make_backfs
372 * indicates if the caller is make_lustre_backfs() or not. */
373 static void disp_old_e2fsprogs_msg(const char *feature, int make_backfs)
375 static int msg_displayed;
378 fprintf(stderr, "WARNING: %s does not support %s "
379 "feature.\n\n", E2FSPROGS, feature);
385 fprintf(stderr, "WARNING: The %s package currently installed on "
386 "your system does not support \"%s\" feature.\n",
388 #if !(HAVE_LDISKFSPROGS)
389 fprintf(stderr, "Please install the latest version of e2fsprogs from\n"
390 "https://downloads.hpdd.intel.com/public/e2fsprogs/latest/\n"
391 "to enable this feature.\n");
394 fprintf(stderr, "Feature will not be enabled until %s"
395 "is updated and '%s -O %s %%{device}' "
396 "is run.\n\n", E2FSPROGS, TUNE2FS, feature);
399 /* Check whether the file exists in the device */
400 static int file_in_dev(char *file_name, char *dev_name)
403 char debugfs_cmd[256];
404 unsigned int inode_num;
407 /* Construct debugfs command line. */
408 snprintf(debugfs_cmd, sizeof(debugfs_cmd),
409 "%s -c -R 'stat %s' '%s' 2>&1 | egrep '(Inode|unsupported)'",
410 DEBUGFS, file_name, dev_name);
412 fp = popen(debugfs_cmd, "r");
414 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
418 if (fscanf(fp, "Inode: %u", &inode_num) == 1) { /* exist */
422 i = fread(debugfs_cmd, 1, sizeof(debugfs_cmd) - 1, fp);
425 fprintf(stderr, "%s", debugfs_cmd);
426 if (strstr(debugfs_cmd, "unsupported feature")) {
427 disp_old_e2fsprogs_msg("an unknown", 0);
436 /* Check whether the device has already been used with lustre */
437 int ldiskfs_is_lustre(char *dev, unsigned *mount_type)
441 ret = file_in_dev(MOUNT_DATA_FILE, dev);
443 /* in the -1 case, 'extents' means IS a lustre target */
444 *mount_type = LDD_MT_LDISKFS;
448 ret = file_in_dev(LAST_RCVD, dev);
450 *mount_type = LDD_MT_LDISKFS;
457 /* Check if a certain feature is supported by e2fsprogs.
458 * Firstly we try to use "debugfs supported_features" command to check if
459 * the feature is supported. If this fails we try to set this feature with
460 * mke2fs to check for its support. */
461 static int is_e2fsprogs_feature_supp(const char *feature)
463 static char supp_features[4096] = "";
466 char imgname[] = "/tmp/test-img-XXXXXX";
470 if (supp_features[0] == '\0') {
471 snprintf(cmd, sizeof(cmd), "%s -c -R supported_features 2>&1",
474 /* Using popen() instead of run_command() since debugfs does
475 * not return proper error code if command is not supported */
476 fp = popen(cmd, "r");
478 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
481 ret = fread(supp_features, 1, sizeof(supp_features) - 1, fp);
482 supp_features[ret] = '\0';
485 if (ret > 0 && strstr(supp_features,
486 strncmp(feature, "-O ", 3) ? feature : feature+3))
489 if ((fd = mkstemp(imgname)) < 0)
494 snprintf(cmd, sizeof(cmd), "%s -F %s %s 100 >/dev/null 2>&1",
495 MKE2FS, feature, imgname);
496 /* run_command() displays the output of mke2fs when it fails for
497 * some feature, so use system() directly */
506 * append_unique: append @key or @key=@val pair to @buf only if @key does not
508 * @buf: buffer to hold @key or @key=@val
509 * @prefix: prefix string before @key
511 * @val: value string if it's a @key=@val pair
513 static void append_unique(char *buf, char *prefix, char *key, char *val,
522 anchor = end = strstr(buf, key);
523 /* try to find exact match string in @buf */
524 while (end && *end != '\0' && *end != ',' && *end != ' ' && *end != '=')
527 if (anchor == NULL || strlen(key) != len ||
528 strncmp(anchor, key, len) != 0) {
530 strscat(buf, prefix, maxbuflen);
532 strscat(buf, key, maxbuflen);
534 strscat(buf, "=\"", maxbuflen);
535 strscat(buf, val, maxbuflen);
536 strscat(buf, "\"", maxbuflen);
541 static int enable_default_ext4_features(struct mkfs_opts *mop, char *anchor,
542 size_t maxbuflen, int user_spec)
544 if (IS_OST(&mop->mo_ldd)) {
545 append_unique(anchor, user_spec ? "," : " -O ",
546 "extents", NULL, maxbuflen);
547 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
548 } else if (IS_MDT(&mop->mo_ldd)) {
549 append_unique(anchor, user_spec ? "," : " -O ",
550 "dirdata", NULL, maxbuflen);
551 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
552 append_unique(anchor, ",", "^extents", NULL, maxbuflen);
554 append_unique(anchor, user_spec ? "," : " -O ",
555 "uninit_bg", NULL, maxbuflen);
558 /* Multiple mount protection enabled only if failover node specified */
559 if (mop->mo_flags & MO_FAILOVER) {
560 if (is_e2fsprogs_feature_supp("-O mmp") == 0)
561 append_unique(anchor, ",", "mmp", NULL, maxbuflen);
563 disp_old_e2fsprogs_msg("mmp", 1);
566 /* Allow more than 65000 subdirectories */
567 if (is_e2fsprogs_feature_supp("-O dir_nlink") == 0)
568 append_unique(anchor, ",", "dir_nlink", NULL, maxbuflen);
570 /* The following options are only valid for ext4-based ldiskfs.
571 * If --backfstype=ext3 is specified, do not enable them. */
572 if (mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3)
575 /* Enable quota by default */
576 if (is_e2fsprogs_feature_supp("-O quota") == 0) {
577 append_unique(anchor, ",", "quota", NULL, maxbuflen);
580 fprintf(stderr, "\"-O quota\" must be supported by "
581 "e2fsprogs, please upgrade your e2fsprogs.\n");
585 /* Allow files larger than 2TB. Also needs LU-16, but not harmful. */
586 if (is_e2fsprogs_feature_supp("-O huge_file") == 0)
587 append_unique(anchor, ",", "huge_file", NULL, maxbuflen);
589 /* Enable large block addresses if the LUN is over 2^32 blocks. */
590 if (mop->mo_device_kb / (L_BLOCK_SIZE >> 10) >= 0x100002000ULL &&
591 is_e2fsprogs_feature_supp("-O 64bit") == 0)
592 append_unique(anchor, ",", "64bit", NULL, maxbuflen);
594 /* Cluster inode/block bitmaps and inode table for more efficient IO.
595 * Align the flex groups on a 1MB boundary for better performance. */
596 /* This -O feature needs to go last, since it adds the "-G" option. */
597 if (is_e2fsprogs_feature_supp("-O flex_bg") == 0) {
600 append_unique(anchor, ",", "flex_bg", NULL, maxbuflen);
602 if (IS_OST(&mop->mo_ldd) &&
603 strstr(mop->mo_mkfsopts, "-G") == NULL) {
604 snprintf(tmp_buf, sizeof(tmp_buf), " -G %u",
605 (1 << 20) / L_BLOCK_SIZE);
606 strscat(anchor, tmp_buf, maxbuflen);
609 /* Don't add any more "-O" options here, see last comment above */
614 * moveopts_to_end: find the option string, move remaining strings to
615 * where option string starts, and append the option
617 * @start: where the option string starts before the move
618 * RETURN: where the option string starts after the move
620 static char *moveopts_to_end(char *start)
626 /* skip whitespace before options */
631 /* find end of option characters */
632 while (*end != ' ' && *end != '\0')
636 if (len >= sizeof(save))
637 len = sizeof(save) - 1;
640 strncpy(save, start, len);
643 /* move remaining options up front */
645 memmove(start, end, strlen(end));
646 *(start + strlen(end)) = '\0';
648 /* append the specified options */
649 if (*(start + strlen(start) - 1) != ' ')
651 idx = start + strlen(start);
657 /* Build fs according to type */
658 int ldiskfs_make_lustre(struct mkfs_opts *mop)
660 __u64 device_kb = mop->mo_device_kb, block_count = 0;
661 char mkfs_cmd[PATH_MAX];
665 int ret = 0, ext_opts = 0;
668 if (!(mop->mo_flags & MO_IS_LOOP)) {
669 mop->mo_device_kb = get_device_size(mop->mo_device);
671 if (mop->mo_device_kb == 0)
674 /* Compare to real size */
675 if (device_kb == 0 || device_kb > mop->mo_device_kb)
676 device_kb = mop->mo_device_kb;
678 mop->mo_device_kb = device_kb;
681 if (mop->mo_device_kb != 0) {
682 if (mop->mo_device_kb < 32384) {
683 fprintf(stderr, "%s: size of filesystem must be larger "
684 "than 32MB, but is set to %lldKB\n",
685 progname, (long long)mop->mo_device_kb);
688 block_count = mop->mo_device_kb / (L_BLOCK_SIZE >> 10);
689 /* If the LUN size is just over 2^32 blocks, limit the
690 * filesystem size to 2^32-1 blocks to avoid problems with
691 * ldiskfs/mkfs not handling this size. Bug 22906 */
692 if (block_count > 0xffffffffULL && block_count < 0x100002000ULL)
693 block_count = 0xffffffffULL;
696 if ((mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3) ||
697 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS) ||
698 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS2)) {
701 /* Journal size in MB */
702 if (strstr(mop->mo_mkfsopts, "-J") == NULL &&
703 device_kb > 1024 * 1024) {
704 /* Choose our own default journal size */
705 long journal_mb = 0, max_mb;
707 /* cap journal size at 4GB for MDT,
708 * leave it at 400MB for OSTs. */
709 if (IS_MDT(&mop->mo_ldd))
711 else if (IS_OST(&mop->mo_ldd))
713 else /* Use mke2fs default size for MGS */
716 /* Use at most 4% of device for journal */
717 journal_mb = device_kb * 4 / (1024 * 100);
718 if (journal_mb > max_mb)
722 sprintf(buf, " -J size=%ld", journal_mb);
723 strscat(mop->mo_mkfsopts, buf,
724 sizeof(mop->mo_mkfsopts));
729 * The inode size is constituted by following elements
730 * (assuming all files are in composite layout and has
733 * ldiskfs inode size: 156
734 * extended attributes size, including:
735 * ext4_xattr_header: 32
736 * LOV EA size: 32(lov_comp_md_v1) +
737 * 3 * 40(lov_comp_md_entry_v1) +
738 * 3 * 32(lov_mds_md) +
739 * stripes * 24(lov_ost_data) +
740 * 16(xattr_entry) + 3(lov)
741 * LMA EA size: 24(lustre_mdt_attrs) +
742 * 16(xattr_entry) + 3(lma)
743 * link EA size: 24(link_ea_header) + 18(link_ea_entry) +
744 * (filename) + 16(xattr_entry) + 4(link)
745 * and some margin for 4-byte alignment, ACLs and other EAs.
747 * If we say the average filename length is about 32 bytes,
748 * the calculation looks like:
749 * 156 + 32 + (32+3*(40 + 32)+24*N+19) + (24+19) +
750 * (24+18+~32+20) + other <= 512*2^m, {m=0,1,2,3}
752 if (strstr(mop->mo_mkfsopts, "-I") == NULL) {
753 if (IS_MDT(&mop->mo_ldd)) {
754 if (mop->mo_stripe_count > 59)
755 inode_size = 512; /* bz 7241 */
756 /* see also "-i" below for EA blocks */
757 else if (mop->mo_stripe_count > 16)
761 } else if (IS_OST(&mop->mo_ldd)) {
762 /* We store MDS FID and necessary composite
763 * layout information in the OST object EA. */
767 if (inode_size > 0) {
768 sprintf(buf, " -I %ld", inode_size);
769 strscat(mop->mo_mkfsopts, buf,
770 sizeof(mop->mo_mkfsopts));
774 /* Bytes_per_inode: disk size / num inodes */
775 if (strstr(mop->mo_mkfsopts, "-i") == NULL &&
776 strstr(mop->mo_mkfsopts, "-N") == NULL) {
777 long bytes_per_inode = 0;
779 /* Allocate more inodes on MDT devices. There is
780 * no data stored on the MDT, and very little extra
781 * metadata beyond the inode. It could go down as
782 * low as 1024 bytes, but this is conservative.
783 * Account for external EA blocks for wide striping. */
784 if (IS_MDT(&mop->mo_ldd)) {
785 bytes_per_inode = inode_size + 1536;
787 if (mop->mo_stripe_count > 59) {
788 int extra = mop->mo_stripe_count * 24;
789 extra = ((extra - 1) | 4095) + 1;
790 bytes_per_inode += extra;
794 /* Allocate fewer inodes on large OST devices. Most
795 * filesystems can be much more aggressive than even
796 * this, but it is impossible to know in advance. */
797 if (IS_OST(&mop->mo_ldd)) {
798 /* OST > 16TB assume average file size 1MB */
799 if (device_kb > (16ULL << 30))
800 bytes_per_inode = 1024 * 1024;
801 /* OST > 4TB assume average file size 512kB */
802 else if (device_kb > (4ULL << 30))
803 bytes_per_inode = 512 * 1024;
804 /* OST > 1TB assume average file size 256kB */
805 else if (device_kb > (1ULL << 30))
806 bytes_per_inode = 256 * 1024;
807 /* OST > 10GB assume average file size 64kB,
808 * plus a bit so that inodes will fit into a
809 * 256x flex_bg without overflowing */
810 else if (device_kb > (10ULL << 20))
811 bytes_per_inode = 69905;
814 if (bytes_per_inode > 0) {
815 sprintf(buf, " -i %ld", bytes_per_inode);
816 strscat(mop->mo_mkfsopts, buf,
817 sizeof(mop->mo_mkfsopts));
822 strscat(mop->mo_mkfsopts, " -q",
823 sizeof(mop->mo_mkfsopts));
826 /* start handle -O mkfs options */
827 if ((start = strstr(mop->mo_mkfsopts, "-O")) != NULL) {
828 if (strstr(start + 2, "-O") != NULL) {
830 "%s: don't specify multiple -O options\n",
834 start = moveopts_to_end(start);
835 maxbuflen = sizeof(mop->mo_mkfsopts) -
836 (start - mop->mo_mkfsopts) - strlen(start);
837 ret = enable_default_ext4_features(mop, start, maxbuflen, 1);
839 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts),
840 maxbuflen = sizeof(mop->mo_mkfsopts) -
841 strlen(mop->mo_mkfsopts);
842 ret = enable_default_ext4_features(mop, start, maxbuflen, 0);
846 /* end handle -O mkfs options */
848 /* start handle -E mkfs options */
849 if ((start = strstr(mop->mo_mkfsopts, "-E")) != NULL) {
850 if (strstr(start + 2, "-E") != NULL) {
852 "%s: don't specify multiple -E options\n",
856 start = moveopts_to_end(start);
857 maxbuflen = sizeof(mop->mo_mkfsopts) -
858 (start - mop->mo_mkfsopts) - strlen(start);
861 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts);
862 maxbuflen = sizeof(mop->mo_mkfsopts) -
863 strlen(mop->mo_mkfsopts);
866 /* In order to align the filesystem metadata on 1MB boundaries,
867 * give a resize value that will reserve a power-of-two group
868 * descriptor blocks, but leave one block for the superblock.
869 * Only useful for filesystems with < 2^32 blocks due to resize
871 if (strstr(mop->mo_mkfsopts, "meta_bg") == NULL &&
872 IS_OST(&mop->mo_ldd) && mop->mo_device_kb > 100 * 1024 &&
873 mop->mo_device_kb * 1024 / L_BLOCK_SIZE <= 0xffffffffULL) {
874 unsigned group_blocks = L_BLOCK_SIZE * 8;
875 unsigned desc_per_block = L_BLOCK_SIZE / 32;
876 unsigned resize_blks;
878 resize_blks = (1ULL<<32) - desc_per_block*group_blocks;
879 snprintf(buf, sizeof(buf), "%u", resize_blks);
880 append_unique(start, ext_opts ? "," : " -E ",
881 "resize", buf, maxbuflen);
885 /* Avoid zeroing out the full journal - speeds up mkfs */
886 if (is_e2fsprogs_feature_supp("-E lazy_journal_init") == 0)
887 append_unique(start, ext_opts ? "," : " -E ",
888 "lazy_journal_init", NULL, maxbuflen);
889 /* end handle -E mkfs options */
891 /* Allow reformat of full devices (as opposed to
892 partitions.) We already checked for mounted dev. */
893 strscat(mop->mo_mkfsopts, " -F", sizeof(mop->mo_mkfsopts));
895 snprintf(mkfs_cmd, sizeof(mkfs_cmd),
896 "%s -j -b %d -L %s ", MKE2FS, L_BLOCK_SIZE,
897 mop->mo_ldd.ldd_svname);
899 fprintf(stderr,"%s: unsupported fs type: %d (%s)\n",
900 progname, mop->mo_ldd.ldd_mount_type,
901 MT_STR(&mop->mo_ldd));
905 /* For loop device format the dev, not the filename */
906 dev = mop->mo_device;
907 if (mop->mo_flags & MO_IS_LOOP)
908 dev = mop->mo_loopdev;
910 vprint("formatting backing filesystem %s on %s\n",
911 MT_STR(&mop->mo_ldd), dev);
912 vprint("\ttarget name %s\n", mop->mo_ldd.ldd_svname);
913 vprint("\t4k blocks %ju\n", (uintmax_t)block_count);
914 vprint("\toptions %s\n", mop->mo_mkfsopts);
916 /* mkfs_cmd's trailing space is important! */
917 strscat(mkfs_cmd, mop->mo_mkfsopts, sizeof(mkfs_cmd));
918 strscat(mkfs_cmd, " ", sizeof(mkfs_cmd));
919 strscat(mkfs_cmd, dev, sizeof(mkfs_cmd));
920 if (block_count != 0) {
921 snprintf(buf, sizeof(buf), " %ju",
922 (uintmax_t)block_count);
923 strscat(mkfs_cmd, buf, sizeof(mkfs_cmd));
926 vprint("mkfs_cmd = %s\n", mkfs_cmd);
927 ret = run_command(mkfs_cmd, sizeof(mkfs_cmd));
930 fprintf(stderr, "Unable to build fs %s (%d)\n", dev, ret);
935 int ldiskfs_prepare_lustre(struct mkfs_opts *mop,
936 char *wanted_mountopts, size_t len)
938 struct lustre_disk_data *ldd = &mop->mo_ldd;
941 /* Set MO_IS_LOOP to indicate a loopback device is needed */
942 ret = is_block(mop->mo_device);
945 } else if (ret == 0) {
946 mop->mo_flags |= MO_IS_LOOP;
949 if (IS_MDT(ldd) || IS_MGS(ldd))
950 strscat(wanted_mountopts, ",user_xattr", len);
955 int ldiskfs_fix_mountopts(struct mkfs_opts *mop, char *mountopts, size_t len)
957 if (strstr(mountopts, "errors=") == NULL)
958 strscat(mountopts, ",errors=remount-ro", len);
963 static int read_file(const char *path, char *buf, int size)
967 fd = fopen(path, "r");
971 if (fgets(buf, size, fd) == NULL) {
972 fprintf(stderr, "reading from %s: %s", path, strerror(errno));
978 /* strip trailing newline */
980 if (buf[size - 1] == '\n')
981 buf[size - 1] = '\0';
986 static int write_file(const char *path, const char *buf)
990 fd = open(path, O_WRONLY);
994 rc = write(fd, buf, strlen(buf));
997 return rc < 0 ? errno : 0;
1000 static int tune_md_stripe_cache_size(const char *sys_path,
1001 struct mount_opts *mop)
1003 char path[PATH_MAX];
1004 unsigned long old_stripe_cache_size;
1005 unsigned long new_stripe_cache_size;
1006 char buf[3 * sizeof(old_stripe_cache_size) + 2];
1009 if (mop->mo_md_stripe_cache_size <= 0)
1012 new_stripe_cache_size = mop->mo_md_stripe_cache_size;
1014 snprintf(path, sizeof(path), "%s/%s", sys_path, STRIPE_CACHE_SIZE);
1015 rc = read_file(path, buf, sizeof(buf));
1018 fprintf(stderr, "warning: cannot read '%s': %s\n",
1019 path, strerror(errno));
1023 old_stripe_cache_size = strtoul(buf, NULL, 0);
1024 if (old_stripe_cache_size == 0 || old_stripe_cache_size == ULONG_MAX)
1027 if (new_stripe_cache_size <= old_stripe_cache_size)
1030 snprintf(buf, sizeof(buf), "%lu", new_stripe_cache_size);
1031 rc = write_file(path, buf);
1034 fprintf(stderr, "warning: cannot write '%s': %s\n",
1035 path, strerror(errno));
1042 static int tune_max_sectors_kb(const char *sys_path, struct mount_opts *mop)
1044 char path[PATH_MAX];
1045 unsigned long max_hw_sectors_kb;
1046 unsigned long old_max_sectors_kb;
1047 unsigned long new_max_sectors_kb;
1048 char buf[3 * sizeof(old_max_sectors_kb) + 2];
1051 if (mop->mo_max_sectors_kb >= 0) {
1052 new_max_sectors_kb = mop->mo_max_sectors_kb;
1053 goto have_new_max_sectors_kb;
1056 snprintf(path, sizeof(path), "%s/%s", sys_path, MAX_HW_SECTORS_KB_PATH);
1057 rc = read_file(path, buf, sizeof(buf));
1059 /* No MAX_HW_SECTORS_KB_PATH isn't necessary an
1060 * error for some devices. */
1064 max_hw_sectors_kb = strtoul(buf, NULL, 0);
1065 if (max_hw_sectors_kb == 0 || max_hw_sectors_kb == ULLONG_MAX) {
1066 /* No digits at all or something weird. */
1070 new_max_sectors_kb = max_hw_sectors_kb;
1072 /* Don't increase IO request size limit past 16MB. It is
1073 * about PTLRPC_MAX_BRW_SIZE, but that isn't in a public
1074 * header. Note that even though the block layer allows
1075 * larger values, setting max_sectors_kb = 32768 causes
1076 * crashes (LU-6974). */
1077 if (new_max_sectors_kb > 16 * 1024)
1078 new_max_sectors_kb = 16 * 1024;
1080 have_new_max_sectors_kb:
1081 snprintf(path, sizeof(path), "%s/%s", sys_path, MAX_SECTORS_KB_PATH);
1082 rc = read_file(path, buf, sizeof(buf));
1084 /* No MAX_SECTORS_KB_PATH isn't necessary an error for
1089 old_max_sectors_kb = strtoul(buf, NULL, 0);
1090 if (old_max_sectors_kb == 0 || old_max_sectors_kb == ULLONG_MAX) {
1091 /* No digits at all or something weird. */
1095 if (new_max_sectors_kb <= old_max_sectors_kb)
1098 snprintf(buf, sizeof(buf), "%lu", new_max_sectors_kb);
1099 rc = write_file(path, buf);
1102 fprintf(stderr, "warning: cannot write '%s': %s\n",
1103 path, strerror(errno));
1107 fprintf(stderr, "%s: increased '%s' from %lu to %lu\n",
1108 progname, path, old_max_sectors_kb, new_max_sectors_kb);
1113 static int tune_block_dev_scheduler(const char *sys_path, const char *new_sched)
1115 char path[PATH_MAX];
1121 /* Before setting the scheduler, we need to check to see if
1122 * it's already set to "noop". If it is then we don't want to
1123 * override that setting. If it's set to anything other than
1124 * "noop" then set the scheduler to what has been passed
1127 snprintf(path, sizeof(path), "%s/%s", sys_path, SCHEDULER_PATH);
1128 rc = read_file(path, buf, sizeof(buf));
1131 fprintf(stderr, "%s: cannot read '%s': %s\n",
1132 progname, path, strerror(errno));
1137 /* The expected format of buf: noop anticipatory deadline [cfq] */
1138 s = strchr(buf, '[');
1139 e = strchr(buf, ']');
1141 /* If the format is not what we expect then be safe and error out. */
1142 if (s == NULL || e == NULL || !(s < e)) {
1145 "%s: cannot parse scheduler options for '%s'\n",
1154 if (strcmp(old_sched, "noop") == 0 ||
1155 strcmp(old_sched, new_sched) == 0)
1158 rc = write_file(path, new_sched);
1162 "%s: cannot set scheduler on '%s': %s\n",
1163 progname, path, strerror(errno));
1167 fprintf(stderr, "%s: changed scheduler of '%s' from %s to %s\n",
1168 progname, path, old_sched, new_sched);
1173 static int tune_block_dev(const char *src, struct mount_opts *mop);
1175 static int tune_block_dev_slaves(const char *sys_path, struct mount_opts *mop)
1177 char slaves_path[PATH_MAX];
1182 snprintf(slaves_path, sizeof(slaves_path), "%s/slaves", sys_path);
1183 slaves_dir = opendir(slaves_path);
1184 if (slaves_dir == NULL) {
1185 if (errno == ENOENT)
1191 while ((d = readdir(slaves_dir)) != NULL) {
1192 char path[PATH_MAX];
1195 if (d->d_type != DT_LNK)
1198 snprintf(path, sizeof(path), "%s/%s", slaves_path, d->d_name);
1199 rc2 = tune_block_dev(path, mop);
1204 closedir(slaves_dir);
1209 /* This is to tune the kernel for good SCSI performance.
1210 * For that we set the value of /sys/block/{dev}/queue/max_sectors_kb
1211 * to the value of /sys/block/{dev}/queue/max_hw_sectors_kb */
1212 static int tune_block_dev(const char *src, struct mount_opts *mop)
1215 char sys_path[PATH_MAX];
1216 char partition_path[PATH_MAX];
1217 char *real_sys_path = NULL;
1223 rc = stat(src, &st);
1226 fprintf(stderr, "warning: cannot stat '%s': %s\n",
1227 src, strerror(errno));
1231 if (!S_ISBLK(st.st_mode))
1234 if (major(st.st_rdev) == LOOP_MAJOR)
1237 snprintf(sys_path, sizeof(sys_path), "/sys/dev/block/%u:%u",
1238 major(st.st_rdev), minor(st.st_rdev));
1240 snprintf(partition_path, sizeof(partition_path), "%s/partition",
1243 rc = access(partition_path, F_OK);
1245 if (errno == ENOENT)
1246 goto have_whole_dev;
1250 "warning: cannot access '%s': %s\n",
1251 partition_path, strerror(errno));
1256 snprintf(sys_path, sizeof(sys_path), "/sys/dev/block/%u:%u/..",
1257 major(st.st_rdev), minor(st.st_rdev));
1260 /* Since we recurse on slave devices we resolve the sys_path to
1261 * avoid path buffer overflows. */
1262 real_sys_path = realpath(sys_path, NULL);
1263 if (real_sys_path == NULL) {
1266 "warning: cannot resolve '%s': %s\n",
1267 sys_path, strerror(errno));
1272 if (major(st.st_rdev) == MD_MAJOR) {
1273 rc = tune_md_stripe_cache_size(real_sys_path, mop);
1275 /* Ignore errors from tune_max_sectors_kb() and
1276 * tune_scheduler(). The worst that will happen is a block
1277 * device with an "incorrect" scheduler. */
1278 tune_max_sectors_kb(real_sys_path, mop);
1279 tune_block_dev_scheduler(real_sys_path, DEFAULT_SCHEDULER);
1281 /* If device is multipath device then tune its slave
1283 rc = tune_block_dev_slaves(real_sys_path, mop);
1287 free(real_sys_path);
1292 int ldiskfs_tune_lustre(char *dev, struct mount_opts *mop)
1294 return tune_block_dev(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 int ldiskfs_rename_fsname(struct mkfs_opts *mop, const char *oldname)
1312 struct mount_opts opts;
1313 struct lustre_disk_data *ldd = &mop->mo_ldd;
1314 char mntpt[] = "/tmp/mntXXXXXX";
1318 /* Change the filesystem label. */
1320 opts.mo_source = mop->mo_device;
1321 ret = ldiskfs_label_lustre(&opts);
1325 fprintf(stderr, "Can't change filesystem label: %s\n",
1330 /* Mount this device temporarily in order to write these files */
1331 if (mkdtemp(mntpt) == NULL) {
1336 fprintf(stderr, "Can't create temp mount point %s: %s\n",
1337 mntpt, strerror(ret));
1343 * Append file context to mount options if SE Linux is enabled
1345 if (is_selinux_enabled() > 0)
1346 append_context_for_mount(mntpt, mop);
1349 if (mop->mo_flags & MO_IS_LOOP)
1350 dev = mop->mo_loopdev;
1352 dev = mop->mo_device;
1353 ret = mount(dev, mntpt, MT_STR(ldd), 0, ldd->ldd_mount_opts);
1357 fprintf(stderr, "Unable to mount %s: %s\n",
1358 dev, strerror(ret));
1360 fprintf(stderr, "Is the %s module available?\n",
1365 ret = lustre_rename_fsname(mop, mntpt, oldname);
1373 /* Enable quota accounting */
1374 int ldiskfs_enable_quota(struct mkfs_opts *mop)
1378 int cmdsz = sizeof(cmd), ret;
1380 if (is_e2fsprogs_feature_supp("-O quota") != 0) {
1381 fprintf(stderr, "%s: \"-O quota\" is is not supported by "
1382 "current e2fsprogs\n", progname);
1386 dev = mop->mo_device;
1387 if (mop->mo_flags & MO_IS_LOOP)
1388 dev = mop->mo_loopdev;
1390 /* Quota feature is already enabled? */
1391 if (is_feature_enabled("quota", dev)) {
1392 vprint("Quota feature is already enabled.\n");
1396 /* Turn on quota feature by "tune2fs -O quota" */
1397 snprintf(cmd, cmdsz, "%s -O quota %s", TUNE2FS, dev);
1398 ret = run_command(cmd, cmdsz);
1400 fprintf(stderr, "command:%s (%d)", cmd, ret);
1405 int ldiskfs_init(void)
1407 /* Required because full path to DEBUGFS is not specified */
1408 setenv("PATH", "/bin:/sbin:/usr/bin:/usr/sbin:/usr/local/bin", 0);
1413 void ldiskfs_fini(void)
1419 struct module_backfs_ops ldiskfs_ops = {
1420 .init = ldiskfs_init,
1421 .fini = ldiskfs_fini,
1422 .read_ldd = ldiskfs_read_ldd,
1423 .write_ldd = ldiskfs_write_ldd,
1424 .erase_ldd = ldiskfs_erase_ldd,
1425 .print_ldd_params = ldiskfs_print_ldd_params,
1426 .is_lustre = ldiskfs_is_lustre,
1427 .make_lustre = ldiskfs_make_lustre,
1428 .prepare_lustre = ldiskfs_prepare_lustre,
1429 .fix_mountopts = ldiskfs_fix_mountopts,
1430 .tune_lustre = ldiskfs_tune_lustre,
1431 .label_lustre = ldiskfs_label_lustre,
1432 .enable_quota = ldiskfs_enable_quota,
1433 .rename_fsname = ldiskfs_rename_fsname,
1435 #endif /* PLUGIN_DIR */