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, 2017, 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>
60 #include <sys/sysmacros.h>
68 #include <linux/fs.h> /* for BLKGETSIZE64 */
70 #include <linux/major.h>
71 #include <linux/types.h>
72 #include <linux/version.h>
73 #include <linux/lnet/lnetctl.h>
74 #include <linux/lustre/lustre_ver.h>
75 #include <libcfs/util/string.h>
78 #include <selinux/selinux.h>
81 #include "mount_utils.h"
83 #define MAX_HW_SECTORS_KB_PATH "queue/max_hw_sectors_kb"
84 #define MAX_SECTORS_KB_PATH "queue/max_sectors_kb"
85 #define SCHEDULER_PATH "queue/scheduler"
86 #define STRIPE_CACHE_SIZE "md/stripe_cache_size"
88 #define DEFAULT_SCHEDULER "deadline"
90 extern char *progname;
92 #define L_BLOCK_SIZE 4096
93 /* keep it less than LL_FID_NAMELEN */
94 #define DUMMY_FILE_NAME_LEN 25
95 #define EXT3_DIRENT_SIZE DUMMY_FILE_NAME_LEN
97 static void append_unique(char *buf, char *prefix, char *key, char *val,
99 static int is_e2fsprogs_feature_supp(const char *feature);
100 static void disp_old_e2fsprogs_msg(const char *feature, int make_backfs);
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 /* Determine if a device is a block device (as opposed to a file) */
127 static int is_block(char *devname)
133 ret = cfs_abs_path(devname, &devpath);
135 fprintf(stderr, "%s: failed to resolve path '%s': %s\n",
136 progname, devname, strerror(-ret));
140 ret = access(devname, F_OK);
142 if (strncmp(devpath, "/dev/", 5) == 0) {
143 /* nobody sane wants to create a loopback file under
144 * /dev. Let's just report the device doesn't exist */
145 fprintf(stderr, "%s: %s apparently does not exist\n",
153 ret = stat(devpath, &st);
155 fprintf(stderr, "%s: cannot stat %s\n", progname, devpath);
158 ret = S_ISBLK(st.st_mode);
164 static int is_feature_enabled(const char *feature, const char *devpath)
168 char enabled_features[4096] = "";
171 snprintf(cmd, sizeof(cmd), "%s -c -R features %s 2>&1",
174 /* Using popen() instead of run_command() since debugfs does
175 * not return proper error code if command is not supported */
176 fp = popen(cmd, "r");
178 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
182 ret = fread(enabled_features, 1, sizeof(enabled_features) - 1, fp);
183 enabled_features[ret] = '\0';
186 if (strstr(enabled_features, feature))
191 /* Write the server config files */
192 int ldiskfs_write_ldd(struct mkfs_opts *mop)
194 char mntpt[] = "/tmp/mntXXXXXX";
201 /* Mount this device temporarily in order to write these files */
202 if (!mkdtemp(mntpt)) {
203 fprintf(stderr, "%s: Can't create temp mount point %s: %s\n",
204 progname, mntpt, strerror(errno));
209 * Append file context to mount options if SE Linux is enabled
212 if (is_selinux_enabled() > 0)
213 append_context_for_mount(mntpt, mop);
216 dev = mop->mo_device;
217 if (mop->mo_flags & MO_IS_LOOP)
218 dev = mop->mo_loopdev;
220 /* Multiple mount protection enabled if failover node specified */
221 if (mop->mo_flags & MO_FAILOVER &&
222 !is_feature_enabled("mmp", dev)) {
223 if (is_e2fsprogs_feature_supp("-O mmp") == 0) {
224 char *command = filepnm;
226 snprintf(command, sizeof(filepnm),
227 TUNE2FS" -O mmp '%s' >/dev/null 2>&1", dev);
228 ret = run_command(command, sizeof(filepnm));
231 "%s: Unable to set 'mmp' on %s: %d\n",
234 disp_old_e2fsprogs_msg("mmp", 1);
237 ret = mount(dev, mntpt, MT_STR(&mop->mo_ldd), 0,
238 (mop->mo_mountopts == NULL) ?
239 "errors=remount-ro" : mop->mo_mountopts);
241 fprintf(stderr, "%s: Unable to mount %s: %s\n",
242 progname, dev, strerror(errno));
244 if (errno == ENODEV) {
245 fprintf(stderr, "Is the %s module available?\n",
246 MT_STR(&mop->mo_ldd));
251 /* Set up initial directories */
252 sprintf(filepnm, "%s/%s", mntpt, MOUNT_CONFIGS_DIR);
253 ret = mkdir(filepnm, 0777);
254 if ((ret != 0) && (errno != EEXIST)) {
255 fprintf(stderr, "%s: Can't make configs dir %s (%s)\n",
256 progname, filepnm, strerror(errno));
258 } else if (errno == EEXIST) {
262 /* Save the persistent mount data into a file. Lustre must pre-read
263 this file to get the real mount options. */
264 vprint("Writing %s\n", MOUNT_DATA_FILE);
265 sprintf(filepnm, "%s/%s", mntpt, MOUNT_DATA_FILE);
266 filep = fopen(filepnm, "w");
268 fprintf(stderr, "%s: Unable to create %s file: %s\n",
269 progname, filepnm, strerror(errno));
272 num = fwrite(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
273 if (num < 1 && ferror(filep)) {
274 fprintf(stderr, "%s: Unable to write to file (%s): %s\n",
275 progname, filepnm, strerror(errno));
279 fsync(filep->_fileno);
289 static int readcmd(char *cmd, char *buf, int len)
294 fp = popen(cmd, "r");
298 red = fread(buf, 1, len, fp);
301 /* strip trailing newline */
302 if (buf[red - 1] == '\n')
305 return (red == 0) ? -ENOENT : 0;
308 int ldiskfs_read_ldd(char *dev, struct lustre_disk_data *mo_ldd)
310 char tmpdir[] = "/tmp/dirXXXXXX";
315 int cmdsz = sizeof(cmd);
317 /* Make a temporary directory to hold Lustre data files. */
318 if (!mkdtemp(tmpdir)) {
319 fprintf(stderr, "%s: Can't create temporary directory %s: %s\n",
320 progname, tmpdir, strerror(errno));
324 /* TODO: it's worth observing the get_mountdata() function that is
325 in mount_utils.c for getting the mountdata out of the
328 /* Construct debugfs command line. */
329 snprintf(cmd, cmdsz, "%s -c -R 'dump /%s %s/mountdata' '%s'",
330 DEBUGFS, MOUNT_DATA_FILE, tmpdir, dev);
332 ret = run_command(cmd, cmdsz);
334 verrprint("%s: Unable to dump %s dir (%d)\n",
335 progname, MOUNT_CONFIGS_DIR, ret);
337 sprintf(filepnm, "%s/mountdata", tmpdir);
338 filep = fopen(filepnm, "r");
341 vprint("Reading %s\n", MOUNT_DATA_FILE);
342 num_read = fread(mo_ldd, sizeof(*mo_ldd), 1, filep);
343 if (num_read < 1 && ferror(filep)) {
344 fprintf(stderr, "%s: Unable to read from file %s: %s\n",
345 progname, filepnm, strerror(errno));
350 snprintf(cmd, cmdsz, "rm -rf %s", tmpdir);
351 run_command(cmd, cmdsz);
353 verrprint("Failed to read old data (%d)\n", ret);
355 /* As long as we at least have the label, we're good to go */
356 snprintf(cmd, sizeof(cmd), E2LABEL" %s", dev);
357 ret = readcmd(cmd, mo_ldd->ldd_svname, sizeof(mo_ldd->ldd_svname) - 1);
362 int ldiskfs_erase_ldd(struct mkfs_opts *mop, char *param)
367 void ldiskfs_print_ldd_params(struct mkfs_opts *mop)
369 printf("Parameters:%s\n", mop->mo_ldd.ldd_params);
372 /* Display the need for the latest e2fsprogs to be installed. make_backfs
373 * indicates if the caller is make_lustre_backfs() or not. */
374 static void disp_old_e2fsprogs_msg(const char *feature, int make_backfs)
376 static int msg_displayed;
379 fprintf(stderr, "WARNING: %s does not support %s "
380 "feature.\n\n", E2FSPROGS, feature);
386 fprintf(stderr, "WARNING: The %s package currently installed on "
387 "your system does not support \"%s\" feature.\n",
389 #if !(HAVE_LDISKFSPROGS)
390 fprintf(stderr, "Please install the latest version of e2fsprogs from\n"
391 "https://downloads.whamcloud.com/public/e2fsprogs/latest/\n"
392 "to enable this feature.\n");
395 fprintf(stderr, "Feature will not be enabled until %s"
396 "is updated and '%s -O %s %%{device}' "
397 "is run.\n\n", E2FSPROGS, TUNE2FS, feature);
400 /* Check whether the file exists in the device */
401 static int file_in_dev(char *file_name, char *dev_name)
404 char debugfs_cmd[256];
405 unsigned int inode_num;
408 /* Construct debugfs command line. */
409 snprintf(debugfs_cmd, sizeof(debugfs_cmd),
410 "%s -c -R 'stat %s' '%s' 2>&1 | egrep '(Inode|unsupported)'",
411 DEBUGFS, file_name, dev_name);
413 fp = popen(debugfs_cmd, "r");
415 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
419 if (fscanf(fp, "Inode: %u", &inode_num) == 1) { /* exist */
423 i = fread(debugfs_cmd, 1, sizeof(debugfs_cmd) - 1, fp);
426 fprintf(stderr, "%s", debugfs_cmd);
427 if (strstr(debugfs_cmd, "unsupported feature")) {
428 disp_old_e2fsprogs_msg("an unknown", 0);
437 /* Check whether the device has already been used with lustre */
438 int ldiskfs_is_lustre(char *dev, unsigned *mount_type)
442 ret = file_in_dev(MOUNT_DATA_FILE, dev);
444 /* in the -1 case, 'extents' means IS a lustre target */
445 *mount_type = LDD_MT_LDISKFS;
449 ret = file_in_dev(LAST_RCVD, dev);
451 *mount_type = LDD_MT_LDISKFS;
458 /* Check if a certain feature is supported by e2fsprogs.
459 * Firstly we try to use "debugfs supported_features" command to check if
460 * the feature is supported. If this fails we try to set this feature with
461 * mke2fs to check for its support. */
462 static int is_e2fsprogs_feature_supp(const char *feature)
464 static char supp_features[4096] = "";
467 char imgname[] = "/tmp/test-img-XXXXXX";
471 if (supp_features[0] == '\0') {
472 snprintf(cmd, sizeof(cmd), "%s -c -R supported_features 2>&1",
475 /* Using popen() instead of run_command() since debugfs does
476 * not return proper error code if command is not supported */
477 fp = popen(cmd, "r");
479 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
482 ret = fread(supp_features, 1, sizeof(supp_features) - 1, fp);
483 supp_features[ret] = '\0';
486 if (ret > 0 && strstr(supp_features,
487 strncmp(feature, "-O ", 3) ? feature : feature+3))
490 if ((fd = mkstemp(imgname)) < 0)
495 snprintf(cmd, sizeof(cmd), "%s -F %s %s 100 >/dev/null 2>&1",
496 MKE2FS, feature, imgname);
497 /* run_command() displays the output of mke2fs when it fails for
498 * some feature, so use system() directly */
507 * append_unique: append @key or @key=@val pair to @buf only if @key does not
509 * @buf: buffer to hold @key or @key=@val
510 * @prefix: prefix string before @key
512 * @val: value string if it's a @key=@val pair
514 static void append_unique(char *buf, char *prefix, char *key, char *val,
523 anchor = end = strstr(buf, key);
524 /* try to find exact match string in @buf */
525 while (end && *end != '\0' && *end != ',' && *end != ' ' && *end != '=')
528 if (anchor == NULL || strlen(key) != len ||
529 strncmp(anchor, key, len) != 0) {
531 strscat(buf, prefix, maxbuflen);
533 strscat(buf, key, maxbuflen);
535 strscat(buf, "=\"", maxbuflen);
536 strscat(buf, val, maxbuflen);
537 strscat(buf, "\"", maxbuflen);
542 static int enable_default_ext4_features(struct mkfs_opts *mop, char *anchor,
543 size_t maxbuflen, int user_spec)
545 int enable_64bit = 0;
547 /* Enable large block addresses if the LUN is over 2^32 blocks. */
548 if ((mop->mo_device_kb / (L_BLOCK_SIZE >> 10) > UINT32_MAX) &&
549 is_e2fsprogs_feature_supp("-O 64bit") == 0)
552 if (IS_OST(&mop->mo_ldd)) {
553 append_unique(anchor, user_spec ? "," : " -O ",
554 "extents", NULL, maxbuflen);
555 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
556 } else if (IS_MDT(&mop->mo_ldd)) {
557 append_unique(anchor, user_spec ? "," : " -O ",
558 "dirdata", NULL, maxbuflen);
559 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
561 append_unique(anchor, ",", "extents", NULL, maxbuflen);
563 append_unique(anchor, ",", "^extents", NULL, maxbuflen);
565 append_unique(anchor, user_spec ? "," : " -O ",
566 "uninit_bg", NULL, maxbuflen);
569 /* Multiple mount protection enabled only if failover node specified */
570 if (mop->mo_flags & MO_FAILOVER) {
571 if (is_e2fsprogs_feature_supp("-O mmp") == 0)
572 append_unique(anchor, ",", "mmp", NULL, maxbuflen);
574 disp_old_e2fsprogs_msg("mmp", 1);
577 /* Allow more than 65000 subdirectories */
578 if (is_e2fsprogs_feature_supp("-O dir_nlink") == 0)
579 append_unique(anchor, ",", "dir_nlink", NULL, maxbuflen);
581 /* The following options are only valid for ext4-based ldiskfs.
582 * If --backfstype=ext3 is specified, do not enable them. */
583 if (mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3)
586 /* Enable quota by default */
587 if (is_e2fsprogs_feature_supp("-O quota") == 0) {
588 append_unique(anchor, ",", "quota", NULL, maxbuflen);
591 fprintf(stderr, "\"-O quota\" must be supported by "
592 "e2fsprogs, please upgrade your e2fsprogs.\n");
596 /* Allow files larger than 2TB. Also needs LU-16, but not harmful. */
597 if (is_e2fsprogs_feature_supp("-O huge_file") == 0)
598 append_unique(anchor, ",", "huge_file", NULL, maxbuflen);
601 append_unique(anchor, ",", "64bit", NULL, maxbuflen);
603 /* Cluster inode/block bitmaps and inode table for more efficient IO.
604 * Align the flex groups on a 1MB boundary for better performance. */
605 /* This -O feature needs to go last, since it adds the "-G" option. */
606 if (is_e2fsprogs_feature_supp("-O flex_bg") == 0) {
609 append_unique(anchor, ",", "flex_bg", NULL, maxbuflen);
611 if (IS_OST(&mop->mo_ldd) &&
612 strstr(mop->mo_mkfsopts, "-G") == NULL) {
613 snprintf(tmp_buf, sizeof(tmp_buf), " -G %u",
614 (1 << 20) / L_BLOCK_SIZE);
615 strscat(anchor, tmp_buf, maxbuflen);
618 /* Don't add any more "-O" options here, see last comment above */
623 * moveopts_to_end: find the option string, move remaining strings to
624 * where option string starts, and append the option
626 * @start: where the option string starts before the move
627 * RETURN: where the option string starts after the move
629 static char *moveopts_to_end(char *start)
635 /* skip whitespace before options */
640 /* find end of option characters */
641 while (*end != ' ' && *end != '\0')
645 if (len >= sizeof(save))
646 len = sizeof(save) - 1;
649 strncpy(save, start, len);
652 /* move remaining options up front */
654 memmove(start, end, strlen(end));
655 *(start + strlen(end)) = '\0';
657 /* append the specified options */
658 if (*(start + strlen(start) - 1) != ' ')
660 idx = start + strlen(start);
666 /* Build fs according to type */
667 int ldiskfs_make_lustre(struct mkfs_opts *mop)
669 __u64 device_kb = mop->mo_device_kb, block_count = 0;
670 char mkfs_cmd[PATH_MAX];
674 int ret = 0, ext_opts = 0;
677 if (!(mop->mo_flags & MO_IS_LOOP)) {
678 mop->mo_device_kb = get_device_size(mop->mo_device);
680 if (mop->mo_device_kb == 0)
683 /* Compare to real size */
684 if (device_kb == 0 || device_kb > mop->mo_device_kb)
685 device_kb = mop->mo_device_kb;
687 mop->mo_device_kb = device_kb;
690 if (mop->mo_device_kb != 0) {
691 if (mop->mo_device_kb < 32384) {
692 fprintf(stderr, "%s: size of filesystem must be larger "
693 "than 32MB, but is set to %lldKB\n",
694 progname, (long long)mop->mo_device_kb);
697 block_count = mop->mo_device_kb / (L_BLOCK_SIZE >> 10);
698 /* If the LUN size is just over 2^32 blocks, limit the
699 * filesystem size to 2^32-1 blocks to avoid problems with
700 * ldiskfs/mkfs not handling this size. Bug 22906 */
701 if (block_count > 0xffffffffULL && block_count < 0x100002000ULL)
702 block_count = 0xffffffffULL;
705 if ((mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3) ||
706 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS) ||
707 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS2)) {
710 /* Journal size in MB */
711 if (strstr(mop->mo_mkfsopts, "-J") == NULL &&
712 device_kb > 1024 * 1024) {
713 /* Choose our own default journal size */
714 long journal_mb = 0, max_mb;
716 /* cap journal size at 4GB for MDT,
717 * leave it at 400MB for OSTs. */
718 if (IS_MDT(&mop->mo_ldd))
720 else if (IS_OST(&mop->mo_ldd))
722 else /* Use mke2fs default size for MGS */
725 /* Use at most 4% of device for journal */
726 journal_mb = device_kb * 4 / (1024 * 100);
727 if (journal_mb > max_mb)
731 sprintf(buf, " -J size=%ld", journal_mb);
732 strscat(mop->mo_mkfsopts, buf,
733 sizeof(mop->mo_mkfsopts));
738 * The inode size is constituted by following elements
739 * (assuming all files are in composite layout and has
742 * ldiskfs inode size: 160
743 * MDT extended attributes size, including:
744 * ext4_xattr_header: 32
745 * LOV EA size: 32(lov_comp_md_v1) +
746 * 3 * 40(lov_comp_md_entry_v1) +
747 * 3 * 32(lov_mds_md) +
748 * stripes * 24(lov_ost_data) +
749 * 16(xattr_entry) + 4("lov")
750 * LMA EA size: 24(lustre_mdt_attrs) +
751 * 16(xattr_entry) + 4("lma")
752 * SOM EA size: 24(lustre_som_attrs) +
753 * 16(xattr_entry) + 4("som")
754 * link EA size: 24(link_ea_header) + 18(link_ea_entry) +
755 * 16(filename) + 16(xattr_entry) + 4("link")
756 * and some margin for 4-byte alignment, ACLs and other EAs.
758 * If we say the average filename length is about 32 bytes,
759 * the calculation looks like:
760 * 160 + 32 + (32+3*(40+32)+24*stripes+20) + (24+20) + (24+20) +
761 * (24+20) + (~42+16+20) + other <= 512*2^m, {m=0,1,2,3}
763 if (strstr(mop->mo_mkfsopts, "-I") == NULL) {
764 if (IS_MDT(&mop->mo_ldd)) {
765 if (mop->mo_stripe_count > 59)
766 inode_size = 512; /* bz 7241 */
767 /* see also "-i" below for EA blocks */
768 else if (mop->mo_stripe_count > 16)
772 } else if (IS_OST(&mop->mo_ldd)) {
773 /* We store MDS FID and necessary composite
774 * layout information in the OST object EA:
775 * ldiskfs inode size: 160
776 * OST extended attributes size, including:
777 * ext4_xattr_header: 32
778 * LMA EA size: 24(lustre_mdt_attrs) +
779 * 16(xattr_entry) + 4("lma")
780 * FID EA size: 52(filter_fid) +
781 * 16(xattr_entry) + 4("fid")
782 * 160 + 32 + (24+20) + (52+20) = 308
787 if (inode_size > 0) {
788 sprintf(buf, " -I %ld", inode_size);
789 strscat(mop->mo_mkfsopts, buf,
790 sizeof(mop->mo_mkfsopts));
794 /* Bytes_per_inode: disk size / num inodes */
795 if (strstr(mop->mo_mkfsopts, "-i") == NULL &&
796 strstr(mop->mo_mkfsopts, "-N") == NULL) {
797 long bytes_per_inode = 0;
799 /* Allocate more inodes on MDT devices. There is
800 * no data stored on the MDT, and very little extra
801 * metadata beyond the inode. It could go down as
802 * low as 1024 bytes, but this is conservative.
803 * Account for external EA blocks for wide striping. */
804 if (IS_MDT(&mop->mo_ldd)) {
805 bytes_per_inode = inode_size + 1536;
807 if (mop->mo_stripe_count > 59) {
808 int extra = mop->mo_stripe_count * 24;
809 extra = ((extra - 1) | 4095) + 1;
810 bytes_per_inode += extra;
814 /* Allocate fewer inodes on large OST devices. Most
815 * filesystems can be much more aggressive than even
816 * this, but it is impossible to know in advance. */
817 if (IS_OST(&mop->mo_ldd)) {
818 /* OST > 16TB assume average file size 1MB */
819 if (device_kb > (16ULL << 30))
820 bytes_per_inode = 1024 * 1024;
821 /* OST > 4TB assume average file size 512kB */
822 else if (device_kb > (4ULL << 30))
823 bytes_per_inode = 512 * 1024;
824 /* OST > 1TB assume average file size 256kB */
825 else if (device_kb > (1ULL << 30))
826 bytes_per_inode = 256 * 1024;
827 /* OST > 10GB assume average file size 64kB,
828 * plus a bit so that inodes will fit into a
829 * 256x flex_bg without overflowing */
830 else if (device_kb > (10ULL << 20))
831 bytes_per_inode = 69905;
834 if (bytes_per_inode > 0) {
835 sprintf(buf, " -i %ld", bytes_per_inode);
836 strscat(mop->mo_mkfsopts, buf,
837 sizeof(mop->mo_mkfsopts));
838 mop->mo_inode_size = bytes_per_inode;
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 tune_md_stripe_cache_size(const char *sys_path,
1022 struct mount_opts *mop)
1024 char path[PATH_MAX];
1025 unsigned long old_stripe_cache_size;
1026 unsigned long new_stripe_cache_size;
1027 char buf[3 * sizeof(old_stripe_cache_size) + 2];
1030 if (mop->mo_md_stripe_cache_size <= 0)
1033 new_stripe_cache_size = mop->mo_md_stripe_cache_size;
1035 snprintf(path, sizeof(path), "%s/%s", sys_path, STRIPE_CACHE_SIZE);
1036 rc = read_file(path, buf, sizeof(buf));
1039 fprintf(stderr, "warning: cannot read '%s': %s\n",
1040 path, strerror(errno));
1044 old_stripe_cache_size = strtoul(buf, NULL, 0);
1045 if (old_stripe_cache_size == 0 || old_stripe_cache_size == ULONG_MAX)
1048 if (new_stripe_cache_size <= old_stripe_cache_size)
1051 snprintf(buf, sizeof(buf), "%lu", new_stripe_cache_size);
1052 rc = write_file(path, buf);
1055 fprintf(stderr, "warning: cannot write '%s': %s\n",
1056 path, strerror(errno));
1063 static int tune_max_sectors_kb(const char *sys_path, struct mount_opts *mop)
1065 char path[PATH_MAX];
1066 unsigned long max_hw_sectors_kb;
1067 unsigned long old_max_sectors_kb;
1068 unsigned long new_max_sectors_kb;
1069 char buf[3 * sizeof(old_max_sectors_kb) + 2];
1072 if (mop->mo_max_sectors_kb >= 0) {
1073 new_max_sectors_kb = mop->mo_max_sectors_kb;
1074 goto have_new_max_sectors_kb;
1077 snprintf(path, sizeof(path), "%s/%s", sys_path, MAX_HW_SECTORS_KB_PATH);
1078 rc = read_file(path, buf, sizeof(buf));
1080 /* No MAX_HW_SECTORS_KB_PATH isn't necessary an
1081 * error for some devices. */
1085 max_hw_sectors_kb = strtoul(buf, NULL, 0);
1086 if (max_hw_sectors_kb == 0 || max_hw_sectors_kb == ULLONG_MAX) {
1087 /* No digits at all or something weird. */
1091 new_max_sectors_kb = max_hw_sectors_kb;
1093 /* Don't increase IO request size limit past 16MB. It is
1094 * about PTLRPC_MAX_BRW_SIZE, but that isn't in a public
1095 * header. Note that even though the block layer allows
1096 * larger values, setting max_sectors_kb = 32768 causes
1097 * crashes (LU-6974). */
1098 if (new_max_sectors_kb > 16 * 1024)
1099 new_max_sectors_kb = 16 * 1024;
1101 have_new_max_sectors_kb:
1102 snprintf(path, sizeof(path), "%s/%s", sys_path, MAX_SECTORS_KB_PATH);
1103 rc = read_file(path, buf, sizeof(buf));
1105 /* No MAX_SECTORS_KB_PATH isn't necessary an error for
1110 old_max_sectors_kb = strtoul(buf, NULL, 0);
1111 if (old_max_sectors_kb == 0 || old_max_sectors_kb == ULLONG_MAX) {
1112 /* No digits at all or something weird. */
1116 if (new_max_sectors_kb <= old_max_sectors_kb)
1119 snprintf(buf, sizeof(buf), "%lu", new_max_sectors_kb);
1120 rc = write_file(path, buf);
1123 fprintf(stderr, "warning: cannot write '%s': %s\n",
1124 path, strerror(errno));
1128 fprintf(stderr, "%s: increased '%s' from %lu to %lu\n",
1129 progname, path, old_max_sectors_kb, new_max_sectors_kb);
1134 static int tune_block_dev_scheduler(const char *sys_path, const char *new_sched)
1136 char path[PATH_MAX];
1142 /* Before setting the scheduler, we need to check to see if
1143 * it's already set to "noop". If it is then we don't want to
1144 * override that setting. If it's set to anything other than
1145 * "noop" then set the scheduler to what has been passed
1148 snprintf(path, sizeof(path), "%s/%s", sys_path, SCHEDULER_PATH);
1149 rc = read_file(path, buf, sizeof(buf));
1152 fprintf(stderr, "%s: cannot read '%s': %s\n",
1153 progname, path, strerror(errno));
1158 /* The expected format of buf: noop anticipatory deadline [cfq] */
1159 s = strchr(buf, '[');
1160 e = strchr(buf, ']');
1162 /* If the format is not what we expect then be safe and error out. */
1163 if (s == NULL || e == NULL || !(s < e)) {
1166 "%s: cannot parse scheduler options for '%s'\n",
1175 if (strcmp(old_sched, "noop") == 0 ||
1176 strcmp(old_sched, new_sched) == 0)
1179 rc = write_file(path, new_sched);
1183 "%s: cannot set scheduler on '%s': %s\n",
1184 progname, path, strerror(errno));
1188 fprintf(stderr, "%s: changed scheduler of '%s' from %s to %s\n",
1189 progname, path, old_sched, new_sched);
1194 static int tune_block_dev(const char *src, struct mount_opts *mop);
1196 static int tune_block_dev_slaves(const char *sys_path, struct mount_opts *mop)
1198 char slaves_path[PATH_MAX];
1203 snprintf(slaves_path, sizeof(slaves_path), "%s/slaves", sys_path);
1204 slaves_dir = opendir(slaves_path);
1205 if (slaves_dir == NULL) {
1206 if (errno == ENOENT)
1212 while ((d = readdir(slaves_dir)) != NULL) {
1213 char path[PATH_MAX];
1216 if (d->d_type != DT_LNK)
1219 snprintf(path, sizeof(path), "%s/%s", slaves_path, d->d_name);
1220 rc2 = tune_block_dev(path, mop);
1225 closedir(slaves_dir);
1230 /* This is to tune the kernel for good SCSI performance.
1231 * For that we set the value of /sys/block/{dev}/queue/max_sectors_kb
1232 * to the value of /sys/block/{dev}/queue/max_hw_sectors_kb */
1233 static int tune_block_dev(const char *src, struct mount_opts *mop)
1236 char sys_path[PATH_MAX];
1237 char partition_path[PATH_MAX];
1238 char *real_sys_path = NULL;
1244 rc = stat(src, &st);
1247 fprintf(stderr, "warning: cannot stat '%s': %s\n",
1248 src, strerror(errno));
1252 if (!S_ISBLK(st.st_mode))
1255 if (major(st.st_rdev) == LOOP_MAJOR)
1258 snprintf(sys_path, sizeof(sys_path), "/sys/dev/block/%u:%u",
1259 major(st.st_rdev), minor(st.st_rdev));
1261 snprintf(partition_path, sizeof(partition_path), "%s/partition",
1264 rc = access(partition_path, F_OK);
1266 if (errno == ENOENT)
1267 goto have_whole_dev;
1271 "warning: cannot access '%s': %s\n",
1272 partition_path, strerror(errno));
1277 snprintf(sys_path, sizeof(sys_path), "/sys/dev/block/%u:%u/..",
1278 major(st.st_rdev), minor(st.st_rdev));
1281 /* Since we recurse on slave devices we resolve the sys_path to
1282 * avoid path buffer overflows. */
1283 real_sys_path = realpath(sys_path, NULL);
1284 if (real_sys_path == NULL) {
1287 "warning: cannot resolve '%s': %s\n",
1288 sys_path, strerror(errno));
1293 if (major(st.st_rdev) == MD_MAJOR) {
1294 rc = tune_md_stripe_cache_size(real_sys_path, mop);
1296 /* Ignore errors from tune_max_sectors_kb() and
1297 * tune_scheduler(). The worst that will happen is a block
1298 * device with an "incorrect" scheduler. */
1299 tune_max_sectors_kb(real_sys_path, mop);
1300 tune_block_dev_scheduler(real_sys_path, DEFAULT_SCHEDULER);
1302 /* If device is multipath device then tune its slave
1304 rc = tune_block_dev_slaves(real_sys_path, mop);
1308 free(real_sys_path);
1313 int ldiskfs_tune_lustre(char *dev, struct mount_opts *mop)
1315 return tune_block_dev(dev, mop);
1318 int ldiskfs_label_lustre(struct mount_opts *mop)
1320 char label_cmd[PATH_MAX];
1323 snprintf(label_cmd, sizeof(label_cmd),
1324 TUNE2FS" -f -L '%s' '%s' >/dev/null 2>&1",
1325 mop->mo_ldd.ldd_svname, mop->mo_source);
1326 rc = run_command(label_cmd, sizeof(label_cmd));
1331 int ldiskfs_rename_fsname(struct mkfs_opts *mop, const char *oldname)
1333 struct mount_opts opts;
1334 struct lustre_disk_data *ldd = &mop->mo_ldd;
1335 char mntpt[] = "/tmp/mntXXXXXX";
1339 /* Change the filesystem label. */
1341 opts.mo_source = mop->mo_device;
1342 ret = ldiskfs_label_lustre(&opts);
1346 fprintf(stderr, "Can't change filesystem label: %s\n",
1351 /* Mount this device temporarily in order to write these files */
1352 if (mkdtemp(mntpt) == NULL) {
1357 fprintf(stderr, "Can't create temp mount point %s: %s\n",
1358 mntpt, strerror(ret));
1364 * Append file context to mount options if SE Linux is enabled
1366 if (is_selinux_enabled() > 0)
1367 append_context_for_mount(mntpt, mop);
1370 if (mop->mo_flags & MO_IS_LOOP)
1371 dev = mop->mo_loopdev;
1373 dev = mop->mo_device;
1374 ret = mount(dev, mntpt, MT_STR(ldd), 0, ldd->ldd_mount_opts);
1378 fprintf(stderr, "Unable to mount %s: %s\n",
1379 dev, strerror(ret));
1381 fprintf(stderr, "Is the %s module available?\n",
1386 ret = lustre_rename_fsname(mop, mntpt, oldname);
1394 /* Enable quota accounting */
1395 int ldiskfs_enable_quota(struct mkfs_opts *mop)
1399 int cmdsz = sizeof(cmd), ret;
1401 if (is_e2fsprogs_feature_supp("-O quota") != 0) {
1402 fprintf(stderr, "%s: \"-O quota\" is is not supported by "
1403 "current e2fsprogs\n", progname);
1407 dev = mop->mo_device;
1408 if (mop->mo_flags & MO_IS_LOOP)
1409 dev = mop->mo_loopdev;
1411 /* Quota feature is already enabled? */
1412 if (is_feature_enabled("quota", dev)) {
1413 vprint("Quota feature is already enabled.\n");
1417 /* Turn on quota feature by "tune2fs -O quota" */
1418 snprintf(cmd, cmdsz, "%s -O quota %s", TUNE2FS, dev);
1419 ret = run_command(cmd, cmdsz);
1421 fprintf(stderr, "command:%s (%d)", cmd, ret);
1426 int ldiskfs_init(void)
1428 /* Required because full path to DEBUGFS is not specified */
1429 setenv("PATH", "/bin:/sbin:/usr/bin:/usr/sbin:/usr/local/bin", 0);
1434 void ldiskfs_fini(void)
1440 struct module_backfs_ops ldiskfs_ops = {
1441 .init = ldiskfs_init,
1442 .fini = ldiskfs_fini,
1443 .read_ldd = ldiskfs_read_ldd,
1444 .write_ldd = ldiskfs_write_ldd,
1445 .erase_ldd = ldiskfs_erase_ldd,
1446 .print_ldd_params = ldiskfs_print_ldd_params,
1447 .is_lustre = ldiskfs_is_lustre,
1448 .make_lustre = ldiskfs_make_lustre,
1449 .prepare_lustre = ldiskfs_prepare_lustre,
1450 .fix_mountopts = ldiskfs_fix_mountopts,
1451 .tune_lustre = ldiskfs_tune_lustre,
1452 .label_lustre = ldiskfs_label_lustre,
1453 .enable_quota = ldiskfs_enable_quota,
1454 .rename_fsname = ldiskfs_rename_fsname,
1456 #endif /* PLUGIN_DIR */