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/types.h>
70 #include <linux/version.h>
71 #include <linux/lnet/lnetctl.h>
72 #include <linux/lustre/lustre_ver.h>
73 #include <libcfs/util/string.h>
76 #include <selinux/selinux.h>
79 #include "mount_utils.h"
81 #define MAX_HW_SECTORS_KB_PATH "queue/max_hw_sectors_kb"
82 #define MAX_SECTORS_KB_PATH "queue/max_sectors_kb"
83 #define SCHEDULER_PATH "queue/scheduler"
84 #define STRIPE_CACHE_SIZE "md/stripe_cache_size"
86 #define DEFAULT_SCHEDULER "deadline"
88 extern char *progname;
90 #define L_BLOCK_SIZE 4096
91 /* keep it less than LL_FID_NAMELEN */
92 #define DUMMY_FILE_NAME_LEN 25
93 #define EXT3_DIRENT_SIZE DUMMY_FILE_NAME_LEN
95 static void append_unique(char *buf, char *prefix, char *key, char *val,
97 static int is_e2fsprogs_feature_supp(const char *feature);
98 static void disp_old_e2fsprogs_msg(const char *feature, int make_backfs);
101 * Concatenate context of the temporary mount point if selinux is enabled
104 static void append_context_for_mount(char *mntpt, struct mkfs_opts *mop)
106 security_context_t fcontext;
108 if (getfilecon(mntpt, &fcontext) < 0) {
109 /* Continuing with default behaviour */
110 fprintf(stderr, "%s: Get file context failed : %s\n",
111 progname, strerror(errno));
115 if (fcontext != NULL) {
116 append_unique(mop->mo_ldd.ldd_mount_opts,
117 ",", "context", fcontext,
118 sizeof(mop->mo_ldd.ldd_mount_opts));
124 /* Determine if a device is a block device (as opposed to a file) */
125 static int is_block(char *devname)
131 ret = cfs_abs_path(devname, &devpath);
133 fprintf(stderr, "%s: failed to resolve path '%s': %s\n",
134 progname, devname, strerror(-ret));
138 ret = access(devname, F_OK);
140 if (strncmp(devpath, "/dev/", 5) == 0) {
141 /* nobody sane wants to create a loopback file under
142 * /dev. Let's just report the device doesn't exist */
143 fprintf(stderr, "%s: %s apparently does not exist\n",
151 ret = stat(devpath, &st);
153 fprintf(stderr, "%s: cannot stat %s\n", progname, devpath);
156 ret = S_ISBLK(st.st_mode);
162 static int is_feature_enabled(const char *feature, const char *devpath)
166 char enabled_features[4096] = "";
169 snprintf(cmd, sizeof(cmd), "%s -c -R features %s 2>&1",
172 /* Using popen() instead of run_command() since debugfs does
173 * not return proper error code if command is not supported */
174 fp = popen(cmd, "r");
176 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
180 ret = fread(enabled_features, 1, sizeof(enabled_features) - 1, fp);
181 enabled_features[ret] = '\0';
184 if (strstr(enabled_features, feature))
189 /* Write the server config files */
190 int ldiskfs_write_ldd(struct mkfs_opts *mop)
192 char mntpt[] = "/tmp/mntXXXXXX";
199 /* Mount this device temporarily in order to write these files */
200 if (!mkdtemp(mntpt)) {
201 fprintf(stderr, "%s: Can't create temp mount point %s: %s\n",
202 progname, mntpt, strerror(errno));
207 * Append file context to mount options if SE Linux is enabled
210 if (is_selinux_enabled() > 0)
211 append_context_for_mount(mntpt, mop);
214 dev = mop->mo_device;
215 if (mop->mo_flags & MO_IS_LOOP)
216 dev = mop->mo_loopdev;
218 /* Multiple mount protection enabled if failover node specified */
219 if (mop->mo_flags & MO_FAILOVER &&
220 !is_feature_enabled("mmp", dev)) {
221 if (is_e2fsprogs_feature_supp("-O mmp") == 0) {
222 char *command = filepnm;
224 snprintf(command, sizeof(filepnm),
225 TUNE2FS" -O mmp '%s' >/dev/null 2>&1", dev);
226 ret = run_command(command, sizeof(filepnm));
229 "%s: Unable to set 'mmp' on %s: %d\n",
232 disp_old_e2fsprogs_msg("mmp", 1);
235 ret = mount(dev, mntpt, MT_STR(&mop->mo_ldd), 0,
236 (mop->mo_mountopts == NULL) ?
237 "errors=remount-ro" : mop->mo_mountopts);
239 fprintf(stderr, "%s: Unable to mount %s: %s\n",
240 progname, dev, strerror(errno));
242 if (errno == ENODEV) {
243 fprintf(stderr, "Is the %s module available?\n",
244 MT_STR(&mop->mo_ldd));
249 /* Set up initial directories */
250 sprintf(filepnm, "%s/%s", mntpt, MOUNT_CONFIGS_DIR);
251 ret = mkdir(filepnm, 0777);
252 if ((ret != 0) && (errno != EEXIST)) {
253 fprintf(stderr, "%s: Can't make configs dir %s (%s)\n",
254 progname, filepnm, strerror(errno));
256 } else if (errno == EEXIST) {
260 /* Save the persistent mount data into a file. Lustre must pre-read
261 this file to get the real mount options. */
262 vprint("Writing %s\n", MOUNT_DATA_FILE);
263 sprintf(filepnm, "%s/%s", mntpt, MOUNT_DATA_FILE);
264 filep = fopen(filepnm, "w");
266 fprintf(stderr, "%s: Unable to create %s file: %s\n",
267 progname, filepnm, strerror(errno));
270 num = fwrite(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
271 if (num < 1 && ferror(filep)) {
272 fprintf(stderr, "%s: Unable to write to file (%s): %s\n",
273 progname, filepnm, strerror(errno));
277 fsync(filep->_fileno);
287 static int readcmd(char *cmd, char *buf, int len)
292 fp = popen(cmd, "r");
296 red = fread(buf, 1, len, fp);
299 /* strip trailing newline */
300 if (buf[red - 1] == '\n')
303 return (red == 0) ? -ENOENT : 0;
306 int ldiskfs_read_ldd(char *dev, struct lustre_disk_data *mo_ldd)
308 char tmpdir[] = "/tmp/dirXXXXXX";
313 int cmdsz = sizeof(cmd);
315 /* Make a temporary directory to hold Lustre data files. */
316 if (!mkdtemp(tmpdir)) {
317 fprintf(stderr, "%s: Can't create temporary directory %s: %s\n",
318 progname, tmpdir, strerror(errno));
322 /* TODO: it's worth observing the get_mountdata() function that is
323 in mount_utils.c for getting the mountdata out of the
326 /* Construct debugfs command line. */
327 snprintf(cmd, cmdsz, "%s -c -R 'dump /%s %s/mountdata' '%s'",
328 DEBUGFS, MOUNT_DATA_FILE, tmpdir, dev);
330 ret = run_command(cmd, cmdsz);
332 verrprint("%s: Unable to dump %s dir (%d)\n",
333 progname, MOUNT_CONFIGS_DIR, ret);
335 sprintf(filepnm, "%s/mountdata", tmpdir);
336 filep = fopen(filepnm, "r");
339 vprint("Reading %s\n", MOUNT_DATA_FILE);
340 num_read = fread(mo_ldd, sizeof(*mo_ldd), 1, filep);
341 if (num_read < 1 && ferror(filep)) {
342 fprintf(stderr, "%s: Unable to read from file %s: %s\n",
343 progname, filepnm, strerror(errno));
348 snprintf(cmd, cmdsz, "rm -rf %s", tmpdir);
349 run_command(cmd, cmdsz);
351 verrprint("Failed to read old data (%d)\n", ret);
353 /* As long as we at least have the label, we're good to go */
354 snprintf(cmd, sizeof(cmd), E2LABEL" %s", dev);
355 ret = readcmd(cmd, mo_ldd->ldd_svname, sizeof(mo_ldd->ldd_svname) - 1);
360 int ldiskfs_erase_ldd(struct mkfs_opts *mop, char *param)
365 void ldiskfs_print_ldd_params(struct mkfs_opts *mop)
367 printf("Parameters:%s\n", mop->mo_ldd.ldd_params);
370 /* Display the need for the latest e2fsprogs to be installed. make_backfs
371 * indicates if the caller is make_lustre_backfs() or not. */
372 static void disp_old_e2fsprogs_msg(const char *feature, int make_backfs)
374 static int msg_displayed;
377 fprintf(stderr, "WARNING: %s does not support %s "
378 "feature.\n\n", E2FSPROGS, feature);
384 fprintf(stderr, "WARNING: The %s package currently installed on "
385 "your system does not support \"%s\" feature.\n",
387 #if !(HAVE_LDISKFSPROGS)
388 fprintf(stderr, "Please install the latest version of e2fsprogs from\n"
389 "https://downloads.hpdd.intel.com/public/e2fsprogs/latest/\n"
390 "to enable this feature.\n");
393 fprintf(stderr, "Feature will not be enabled until %s"
394 "is updated and '%s -O %s %%{device}' "
395 "is run.\n\n", E2FSPROGS, TUNE2FS, feature);
398 /* Check whether the file exists in the device */
399 static int file_in_dev(char *file_name, char *dev_name)
402 char debugfs_cmd[256];
403 unsigned int inode_num;
406 /* Construct debugfs command line. */
407 snprintf(debugfs_cmd, sizeof(debugfs_cmd),
408 "%s -c -R 'stat %s' '%s' 2>&1 | egrep '(Inode|unsupported)'",
409 DEBUGFS, file_name, dev_name);
411 fp = popen(debugfs_cmd, "r");
413 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
417 if (fscanf(fp, "Inode: %u", &inode_num) == 1) { /* exist */
421 i = fread(debugfs_cmd, 1, sizeof(debugfs_cmd) - 1, fp);
424 fprintf(stderr, "%s", debugfs_cmd);
425 if (strstr(debugfs_cmd, "unsupported feature")) {
426 disp_old_e2fsprogs_msg("an unknown", 0);
435 /* Check whether the device has already been used with lustre */
436 int ldiskfs_is_lustre(char *dev, unsigned *mount_type)
440 ret = file_in_dev(MOUNT_DATA_FILE, dev);
442 /* in the -1 case, 'extents' means IS a lustre target */
443 *mount_type = LDD_MT_LDISKFS;
447 ret = file_in_dev(LAST_RCVD, dev);
449 *mount_type = LDD_MT_LDISKFS;
456 /* Check if a certain feature is supported by e2fsprogs.
457 * Firstly we try to use "debugfs supported_features" command to check if
458 * the feature is supported. If this fails we try to set this feature with
459 * mke2fs to check for its support. */
460 static int is_e2fsprogs_feature_supp(const char *feature)
462 static char supp_features[4096] = "";
465 char imgname[] = "/tmp/test-img-XXXXXX";
469 if (supp_features[0] == '\0') {
470 snprintf(cmd, sizeof(cmd), "%s -c -R supported_features 2>&1",
473 /* Using popen() instead of run_command() since debugfs does
474 * not return proper error code if command is not supported */
475 fp = popen(cmd, "r");
477 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
480 ret = fread(supp_features, 1, sizeof(supp_features) - 1, fp);
481 supp_features[ret] = '\0';
484 if (ret > 0 && strstr(supp_features,
485 strncmp(feature, "-O ", 3) ? feature : feature+3))
488 if ((fd = mkstemp(imgname)) < 0)
493 snprintf(cmd, sizeof(cmd), "%s -F %s %s 100 >/dev/null 2>&1",
494 MKE2FS, feature, imgname);
495 /* run_command() displays the output of mke2fs when it fails for
496 * some feature, so use system() directly */
505 * append_unique: append @key or @key=@val pair to @buf only if @key does not
507 * @buf: buffer to hold @key or @key=@val
508 * @prefix: prefix string before @key
510 * @val: value string if it's a @key=@val pair
512 static void append_unique(char *buf, char *prefix, char *key, char *val,
521 anchor = end = strstr(buf, key);
522 /* try to find exact match string in @buf */
523 while (end && *end != '\0' && *end != ',' && *end != ' ' && *end != '=')
526 if (anchor == NULL || strlen(key) != len ||
527 strncmp(anchor, key, len) != 0) {
529 strscat(buf, prefix, maxbuflen);
531 strscat(buf, key, maxbuflen);
533 strscat(buf, "=\"", maxbuflen);
534 strscat(buf, val, maxbuflen);
535 strscat(buf, "\"", maxbuflen);
540 static int enable_default_ext4_features(struct mkfs_opts *mop, char *anchor,
541 size_t maxbuflen, int user_spec)
543 if (IS_OST(&mop->mo_ldd)) {
544 append_unique(anchor, user_spec ? "," : " -O ",
545 "extents", NULL, maxbuflen);
546 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
547 } else if (IS_MDT(&mop->mo_ldd)) {
548 append_unique(anchor, user_spec ? "," : " -O ",
549 "dirdata", NULL, maxbuflen);
550 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
551 append_unique(anchor, ",", "^extents", NULL, maxbuflen);
553 append_unique(anchor, user_spec ? "," : " -O ",
554 "uninit_bg", NULL, maxbuflen);
557 /* Multiple mount protection enabled only if failover node specified */
558 if (mop->mo_flags & MO_FAILOVER) {
559 if (is_e2fsprogs_feature_supp("-O mmp") == 0)
560 append_unique(anchor, ",", "mmp", NULL, maxbuflen);
562 disp_old_e2fsprogs_msg("mmp", 1);
565 /* Allow more than 65000 subdirectories */
566 if (is_e2fsprogs_feature_supp("-O dir_nlink") == 0)
567 append_unique(anchor, ",", "dir_nlink", NULL, maxbuflen);
569 /* The following options are only valid for ext4-based ldiskfs.
570 * If --backfstype=ext3 is specified, do not enable them. */
571 if (mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3)
574 /* Enable quota by default */
575 if (is_e2fsprogs_feature_supp("-O quota") == 0) {
576 append_unique(anchor, ",", "quota", NULL, maxbuflen);
579 fprintf(stderr, "\"-O quota\" must be supported by "
580 "e2fsprogs, please upgrade your e2fsprogs.\n");
584 /* Allow files larger than 2TB. Also needs LU-16, but not harmful. */
585 if (is_e2fsprogs_feature_supp("-O huge_file") == 0)
586 append_unique(anchor, ",", "huge_file", NULL, maxbuflen);
588 /* Enable large block addresses if the LUN is over 2^32 blocks. */
589 if (mop->mo_device_kb / (L_BLOCK_SIZE >> 10) >= 0x100002000ULL &&
590 is_e2fsprogs_feature_supp("-O 64bit") == 0)
591 append_unique(anchor, ",", "64bit", NULL, maxbuflen);
593 /* Cluster inode/block bitmaps and inode table for more efficient IO.
594 * Align the flex groups on a 1MB boundary for better performance. */
595 /* This -O feature needs to go last, since it adds the "-G" option. */
596 if (is_e2fsprogs_feature_supp("-O flex_bg") == 0) {
599 append_unique(anchor, ",", "flex_bg", NULL, maxbuflen);
601 if (IS_OST(&mop->mo_ldd) &&
602 strstr(mop->mo_mkfsopts, "-G") == NULL) {
603 snprintf(tmp_buf, sizeof(tmp_buf), " -G %u",
604 (1 << 20) / L_BLOCK_SIZE);
605 strscat(anchor, tmp_buf, maxbuflen);
608 /* Don't add any more "-O" options here, see last comment above */
613 * moveopts_to_end: find the option string, move remaining strings to
614 * where option string starts, and append the option
616 * @start: where the option string starts before the move
617 * RETURN: where the option string starts after the move
619 static char *moveopts_to_end(char *start)
625 /* skip whitespace before options */
630 /* find end of option characters */
631 while (*end != ' ' && *end != '\0')
635 if (len >= sizeof(save))
636 len = sizeof(save) - 1;
639 strncpy(save, start, len);
642 /* move remaining options up front */
644 memmove(start, end, strlen(end));
645 *(start + strlen(end)) = '\0';
647 /* append the specified options */
648 if (*(start + strlen(start) - 1) != ' ')
650 idx = start + strlen(start);
656 /* Build fs according to type */
657 int ldiskfs_make_lustre(struct mkfs_opts *mop)
659 __u64 device_kb = mop->mo_device_kb, block_count = 0;
660 char mkfs_cmd[PATH_MAX];
664 int ret = 0, ext_opts = 0;
667 if (!(mop->mo_flags & MO_IS_LOOP)) {
668 mop->mo_device_kb = get_device_size(mop->mo_device);
670 if (mop->mo_device_kb == 0)
673 /* Compare to real size */
674 if (device_kb == 0 || device_kb > mop->mo_device_kb)
675 device_kb = mop->mo_device_kb;
677 mop->mo_device_kb = device_kb;
680 if (mop->mo_device_kb != 0) {
681 if (mop->mo_device_kb < 32384) {
682 fprintf(stderr, "%s: size of filesystem must be larger "
683 "than 32MB, but is set to %lldKB\n",
684 progname, (long long)mop->mo_device_kb);
687 block_count = mop->mo_device_kb / (L_BLOCK_SIZE >> 10);
688 /* If the LUN size is just over 2^32 blocks, limit the
689 * filesystem size to 2^32-1 blocks to avoid problems with
690 * ldiskfs/mkfs not handling this size. Bug 22906 */
691 if (block_count > 0xffffffffULL && block_count < 0x100002000ULL)
692 block_count = 0xffffffffULL;
695 if ((mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3) ||
696 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS) ||
697 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS2)) {
700 /* Journal size in MB */
701 if (strstr(mop->mo_mkfsopts, "-J") == NULL &&
702 device_kb > 1024 * 1024) {
703 /* Choose our own default journal size */
704 long journal_mb = 0, max_mb;
706 /* cap journal size at 4GB for MDT,
707 * leave it at 400MB for OSTs. */
708 if (IS_MDT(&mop->mo_ldd))
710 else if (IS_OST(&mop->mo_ldd))
712 else /* Use mke2fs default size for MGS */
715 /* Use at most 4% of device for journal */
716 journal_mb = device_kb * 4 / (1024 * 100);
717 if (journal_mb > max_mb)
721 sprintf(buf, " -J size=%ld", journal_mb);
722 strscat(mop->mo_mkfsopts, buf,
723 sizeof(mop->mo_mkfsopts));
728 * The inode size is constituted by following elements
729 * (assuming all files are in composite layout and has
732 * ldiskfs inode size: 156
733 * extended attributes size, including:
734 * ext4_xattr_header: 32
735 * LOV EA size: 32(lov_comp_md_v1) +
736 * 3 * 40(lov_comp_md_entry_v1) +
737 * 3 * 32(lov_mds_md) +
738 * stripes * 24(lov_ost_data) +
739 * 16(xattr_entry) + 3(lov)
740 * LMA EA size: 24(lustre_mdt_attrs) +
741 * 16(xattr_entry) + 3(lma)
742 * link EA size: 24(link_ea_header) + 18(link_ea_entry) +
743 * (filename) + 16(xattr_entry) + 4(link)
744 * and some margin for 4-byte alignment, ACLs and other EAs.
746 * If we say the average filename length is about 32 bytes,
747 * the calculation looks like:
748 * 156 + 32 + (32+3*(40 + 32)+24*N+19) + (24+19) +
749 * (24+18+~32+20) + other <= 512*2^m, {m=0,1,2,3}
751 if (strstr(mop->mo_mkfsopts, "-I") == NULL) {
752 if (IS_MDT(&mop->mo_ldd)) {
753 if (mop->mo_stripe_count > 59)
754 inode_size = 512; /* bz 7241 */
755 /* see also "-i" below for EA blocks */
756 else if (mop->mo_stripe_count > 16)
760 } else if (IS_OST(&mop->mo_ldd)) {
761 /* We store MDS FID and necessary composite
762 * layout information in the OST object EA. */
766 if (inode_size > 0) {
767 sprintf(buf, " -I %ld", inode_size);
768 strscat(mop->mo_mkfsopts, buf,
769 sizeof(mop->mo_mkfsopts));
773 /* Bytes_per_inode: disk size / num inodes */
774 if (strstr(mop->mo_mkfsopts, "-i") == NULL &&
775 strstr(mop->mo_mkfsopts, "-N") == NULL) {
776 long bytes_per_inode = 0;
778 /* Allocate more inodes on MDT devices. There is
779 * no data stored on the MDT, and very little extra
780 * metadata beyond the inode. It could go down as
781 * low as 1024 bytes, but this is conservative.
782 * Account for external EA blocks for wide striping. */
783 if (IS_MDT(&mop->mo_ldd)) {
784 bytes_per_inode = inode_size + 1536;
786 if (mop->mo_stripe_count > 59) {
787 int extra = mop->mo_stripe_count * 24;
788 extra = ((extra - 1) | 4095) + 1;
789 bytes_per_inode += extra;
793 /* Allocate fewer inodes on large OST devices. Most
794 * filesystems can be much more aggressive than even
795 * this, but it is impossible to know in advance. */
796 if (IS_OST(&mop->mo_ldd)) {
797 /* OST > 16TB assume average file size 1MB */
798 if (device_kb > (16ULL << 30))
799 bytes_per_inode = 1024 * 1024;
800 /* OST > 4TB assume average file size 512kB */
801 else if (device_kb > (4ULL << 30))
802 bytes_per_inode = 512 * 1024;
803 /* OST > 1TB assume average file size 256kB */
804 else if (device_kb > (1ULL << 30))
805 bytes_per_inode = 256 * 1024;
806 /* OST > 10GB assume average file size 64kB,
807 * plus a bit so that inodes will fit into a
808 * 256x flex_bg without overflowing */
809 else if (device_kb > (10ULL << 20))
810 bytes_per_inode = 69905;
813 if (bytes_per_inode > 0) {
814 sprintf(buf, " -i %ld", bytes_per_inode);
815 strscat(mop->mo_mkfsopts, buf,
816 sizeof(mop->mo_mkfsopts));
821 strscat(mop->mo_mkfsopts, " -q",
822 sizeof(mop->mo_mkfsopts));
825 /* start handle -O mkfs options */
826 if ((start = strstr(mop->mo_mkfsopts, "-O")) != NULL) {
827 if (strstr(start + 2, "-O") != NULL) {
829 "%s: don't specify multiple -O options\n",
833 start = moveopts_to_end(start);
834 maxbuflen = sizeof(mop->mo_mkfsopts) -
835 (start - mop->mo_mkfsopts) - strlen(start);
836 ret = enable_default_ext4_features(mop, start, maxbuflen, 1);
838 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts),
839 maxbuflen = sizeof(mop->mo_mkfsopts) -
840 strlen(mop->mo_mkfsopts);
841 ret = enable_default_ext4_features(mop, start, maxbuflen, 0);
845 /* end handle -O mkfs options */
847 /* start handle -E mkfs options */
848 if ((start = strstr(mop->mo_mkfsopts, "-E")) != NULL) {
849 if (strstr(start + 2, "-E") != NULL) {
851 "%s: don't specify multiple -E options\n",
855 start = moveopts_to_end(start);
856 maxbuflen = sizeof(mop->mo_mkfsopts) -
857 (start - mop->mo_mkfsopts) - strlen(start);
860 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts);
861 maxbuflen = sizeof(mop->mo_mkfsopts) -
862 strlen(mop->mo_mkfsopts);
865 /* In order to align the filesystem metadata on 1MB boundaries,
866 * give a resize value that will reserve a power-of-two group
867 * descriptor blocks, but leave one block for the superblock.
868 * Only useful for filesystems with < 2^32 blocks due to resize
870 if (strstr(mop->mo_mkfsopts, "meta_bg") == NULL &&
871 IS_OST(&mop->mo_ldd) && mop->mo_device_kb > 100 * 1024 &&
872 mop->mo_device_kb * 1024 / L_BLOCK_SIZE <= 0xffffffffULL) {
873 unsigned group_blocks = L_BLOCK_SIZE * 8;
874 unsigned desc_per_block = L_BLOCK_SIZE / 32;
875 unsigned resize_blks;
877 resize_blks = (1ULL<<32) - desc_per_block*group_blocks;
878 snprintf(buf, sizeof(buf), "%u", resize_blks);
879 append_unique(start, ext_opts ? "," : " -E ",
880 "resize", buf, maxbuflen);
884 /* Avoid zeroing out the full journal - speeds up mkfs */
885 if (is_e2fsprogs_feature_supp("-E lazy_journal_init") == 0)
886 append_unique(start, ext_opts ? "," : " -E ",
887 "lazy_journal_init", NULL, maxbuflen);
888 /* end handle -E mkfs options */
890 /* Allow reformat of full devices (as opposed to
891 partitions.) We already checked for mounted dev. */
892 strscat(mop->mo_mkfsopts, " -F", sizeof(mop->mo_mkfsopts));
894 snprintf(mkfs_cmd, sizeof(mkfs_cmd),
895 "%s -j -b %d -L %s ", MKE2FS, L_BLOCK_SIZE,
896 mop->mo_ldd.ldd_svname);
898 fprintf(stderr,"%s: unsupported fs type: %d (%s)\n",
899 progname, mop->mo_ldd.ldd_mount_type,
900 MT_STR(&mop->mo_ldd));
904 /* For loop device format the dev, not the filename */
905 dev = mop->mo_device;
906 if (mop->mo_flags & MO_IS_LOOP)
907 dev = mop->mo_loopdev;
909 vprint("formatting backing filesystem %s on %s\n",
910 MT_STR(&mop->mo_ldd), dev);
911 vprint("\ttarget name %s\n", mop->mo_ldd.ldd_svname);
912 vprint("\t4k blocks %ju\n", (uintmax_t)block_count);
913 vprint("\toptions %s\n", mop->mo_mkfsopts);
915 /* mkfs_cmd's trailing space is important! */
916 strscat(mkfs_cmd, mop->mo_mkfsopts, sizeof(mkfs_cmd));
917 strscat(mkfs_cmd, " ", sizeof(mkfs_cmd));
918 strscat(mkfs_cmd, dev, sizeof(mkfs_cmd));
919 if (block_count != 0) {
920 snprintf(buf, sizeof(buf), " %ju",
921 (uintmax_t)block_count);
922 strscat(mkfs_cmd, buf, sizeof(mkfs_cmd));
925 vprint("mkfs_cmd = %s\n", mkfs_cmd);
926 ret = run_command(mkfs_cmd, sizeof(mkfs_cmd));
929 fprintf(stderr, "Unable to build fs %s (%d)\n", dev, ret);
934 int ldiskfs_prepare_lustre(struct mkfs_opts *mop,
935 char *wanted_mountopts, size_t len)
937 struct lustre_disk_data *ldd = &mop->mo_ldd;
940 /* Set MO_IS_LOOP to indicate a loopback device is needed */
941 ret = is_block(mop->mo_device);
944 } else if (ret == 0) {
945 mop->mo_flags |= MO_IS_LOOP;
948 if (IS_MDT(ldd) || IS_MGS(ldd))
949 strscat(wanted_mountopts, ",user_xattr", len);
954 int ldiskfs_fix_mountopts(struct mkfs_opts *mop, char *mountopts, size_t len)
956 if (strstr(mountopts, "errors=") == NULL)
957 strscat(mountopts, ",errors=remount-ro", len);
962 static int read_file(const char *path, char *buf, int size)
966 fd = fopen(path, "r");
970 if (fgets(buf, size, fd) == NULL) {
971 fprintf(stderr, "reading from %s: %s", path, strerror(errno));
977 /* strip trailing newline */
979 if (buf[size - 1] == '\n')
980 buf[size - 1] = '\0';
985 static int write_file(const char *path, const char *buf)
989 fd = open(path, O_WRONLY);
993 rc = write(fd, buf, strlen(buf));
996 return rc < 0 ? errno : 0;
999 static int set_blockdev_scheduler(const char *path, const char *scheduler)
1001 char buf[PATH_MAX], *s, *e, orig_sched[50];
1004 /* Before setting the scheduler, we need to check to see if it's
1005 * already set to "noop". If it is, we don't want to override
1006 * that setting. If it's set to anything other than "noop", set
1007 * the scheduler to what has been passed in. */
1009 rc = read_file(path, buf, sizeof(buf));
1012 fprintf(stderr, "%s: cannot open '%s': %s\n",
1013 progname, path, strerror(errno));
1017 /* The expected format of buf: noop anticipatory deadline [cfq] */
1018 s = strchr(buf, '[');
1019 e = strchr(buf, ']');
1021 /* If the format is not what we expect. Play it safe and error out. */
1022 if (s == NULL || e == NULL) {
1024 fprintf(stderr, "%s: cannot parse scheduler "
1025 "options for '%s'\n", progname, path);
1029 snprintf(orig_sched, e - s, "%s", s + 1);
1031 if (strcmp(orig_sched, "noop") == 0 ||
1032 strcmp(orig_sched, scheduler) == 0)
1035 rc = write_file(path, scheduler);
1038 fprintf(stderr, "%s: cannot set scheduler on "
1039 "'%s': %s\n", progname, path,
1043 fprintf(stderr, "%s: change scheduler of %s from %s to %s\n",
1044 progname, path, orig_sched, scheduler);
1050 /* This is to tune the kernel for good SCSI performance.
1051 * For that we set the value of /sys/block/{dev}/queue/max_sectors_kb
1052 * to the value of /sys/block/{dev}/queue/max_hw_sectors_kb */
1053 static int set_blockdev_tunables(char *source, struct mount_opts *mop)
1055 glob_t glob_info = { 0 };
1056 struct stat stat_buf;
1057 char *chk_major, *chk_minor;
1058 char *savept = NULL, *dev;
1060 char buf[PATH_MAX] = {'\0'}, path[PATH_MAX] = {'\0'};
1061 char real_path[PATH_MAX] = {'\0'};
1069 ret_path = realpath(source, real_path);
1070 if (ret_path == NULL) {
1072 fprintf(stderr, "warning: %s: cannot resolve: %s\n",
1073 source, strerror(errno));
1077 if (strncmp(real_path, "/dev/loop", 9) == 0)
1080 if ((real_path[0] != '/') && (strpbrk(real_path, ",:") != NULL))
1083 snprintf(path, sizeof(path), "/sys/block%s", real_path + 4);
1084 if (access(path, X_OK) == 0)
1087 /* The name of the device say 'X' specified in /dev/X may not
1088 * match any entry under /sys/block/. In that case we need to
1089 * match the major/minor number to find the entry under
1090 * sys/block corresponding to /dev/X */
1092 /* Don't chop tail digit on /dev/mapper/xxx, LU-478 */
1093 if (strncmp(real_path, "/dev/mapper", 11) != 0) {
1094 dev = real_path + strlen(real_path);
1095 while (--dev > real_path && isdigit(*dev))
1098 if (strncmp(real_path, "/dev/md", 7) == 0 && dev[0] == 'p')
1102 rc = stat(real_path, &stat_buf);
1105 fprintf(stderr, "warning: %s, device %s stat failed\n",
1106 strerror(errno), real_path);
1110 major = major(stat_buf.st_rdev);
1111 minor = minor(stat_buf.st_rdev);
1112 rc = glob("/sys/block/*", GLOB_NOSORT, NULL, &glob_info);
1115 fprintf(stderr, "warning: failed to read entries under "
1117 globfree(&glob_info);
1121 for (i = 0; i < glob_info.gl_pathc; i++){
1122 snprintf(path, sizeof(path), "%s/dev", glob_info.gl_pathv[i]);
1124 rc = read_file(path, buf, sizeof(buf));
1128 if (buf[strlen(buf) - 1] == '\n')
1129 buf[strlen(buf) - 1] = '\0';
1131 chk_major = strtok_r(buf, ":", &savept);
1133 if (chk_major != NULL && major == atoi(chk_major) &&
1134 chk_minor != NULL && minor == atoi(chk_minor))
1138 if (i == glob_info.gl_pathc) {
1140 fprintf(stderr,"warning: device %s does not match any "
1141 "entry under /sys/block\n", real_path);
1142 globfree(&glob_info);
1146 /* Chop off "/dev" from path we found */
1147 path[strlen(glob_info.gl_pathv[i])] = '\0';
1148 globfree(&glob_info);
1151 if (strncmp(real_path, "/dev/md", 7) == 0) {
1152 snprintf(real_path, sizeof(real_path), "%s/%s", path,
1155 rc = read_file(real_path, buf, sizeof(buf));
1158 fprintf(stderr, "warning: opening %s: %s\n",
1159 real_path, strerror(errno));
1163 if (atoi(buf) >= mop->mo_md_stripe_cache_size)
1166 if (strlen(buf) - 1 > 0) {
1167 snprintf(buf, sizeof(buf), "%d",
1168 mop->mo_md_stripe_cache_size);
1169 rc = write_file(real_path, buf);
1170 if (rc != 0 && verbose)
1171 fprintf(stderr, "warning: opening %s: %s\n",
1172 real_path, strerror(errno));
1174 /* Return since raid and disk tunables are different */
1178 if (mop->mo_max_sectors_kb >= 0) {
1179 snprintf(buf, sizeof(buf), "%d", mop->mo_max_sectors_kb);
1181 snprintf(real_path, sizeof(real_path), "%s/%s", path,
1182 MAX_HW_SECTORS_KB_PATH);
1183 rc = read_file(real_path, buf, sizeof(buf));
1186 fprintf(stderr, "warning: opening %s: %s\n",
1187 real_path, strerror(errno));
1188 /* No MAX_HW_SECTORS_KB_PATH isn't necessary an
1189 * error for some device. */
1194 if (strlen(buf) - 1 > 0) {
1195 char oldbuf[32] = "", *end = NULL;
1196 unsigned long long oldval, newval;
1198 snprintf(real_path, sizeof(real_path), "%s/%s", path,
1199 MAX_SECTORS_KB_PATH);
1200 rc = read_file(real_path, oldbuf, sizeof(oldbuf));
1201 /* Only set new parameter if different from the old one. */
1202 if (rc != 0 || strcmp(oldbuf, buf) == 0) {
1203 /* No MAX_SECTORS_KB_PATH isn't necessary an
1204 * error for some device. */
1208 newval = strtoull(buf, &end, 0);
1209 if (newval == 0 || newval == ULLONG_MAX || end == buf)
1212 /* Don't increase IO request size limit past 16MB. It is about
1213 * PTLRPC_MAX_BRW_SIZE, but that isn't in a public header.
1214 * Note that even though the block layer allows larger values,
1215 * setting max_sectors_kb = 32768 causes crashes (LU-6974). */
1216 if (mop->mo_max_sectors_kb < 0 && newval > 16 * 1024) {
1218 snprintf(buf, sizeof(buf), "%llu", newval);
1221 oldval = strtoull(oldbuf, &end, 0);
1222 /* Don't shrink the current limit. */
1223 if (mop->mo_max_sectors_kb < 0 && oldval != ULLONG_MAX &&
1227 rc = write_file(real_path, buf);
1230 fprintf(stderr, "warning: writing to %s: %s\n",
1231 real_path, strerror(errno));
1232 /* No MAX_SECTORS_KB_PATH isn't necessary an
1233 * error for some device. */
1236 fprintf(stderr, "%s: increased %s from %s to %s\n",
1237 progname, real_path, oldbuf, buf);
1241 /* Purposely ignore errors reported from set_blockdev_scheduler.
1242 * The worst that will happen is a block device with an "incorrect"
1244 snprintf(real_path, sizeof(real_path), "%s/%s", path, SCHEDULER_PATH);
1245 set_blockdev_scheduler(real_path, DEFAULT_SCHEDULER);
1247 /* if device is multipath device, tune its slave devices */
1248 glob_info.gl_pathc = 0;
1249 glob_info.gl_offs = 0;
1250 snprintf(real_path, sizeof(real_path), "%s/slaves/*", path);
1251 rc = glob(real_path, GLOB_NOSORT, NULL, &glob_info);
1253 for (i = 0; rc == 0 && i < glob_info.gl_pathc; i++) {
1254 slave = basename(glob_info.gl_pathv[i]);
1255 snprintf(real_path, sizeof(real_path), "/dev/%s", slave);
1256 rc = set_blockdev_tunables(real_path, mop);
1259 if (rc == GLOB_NOMATCH) {
1260 /* no slave device is not an error */
1262 } else if (rc && verbose) {
1263 if (slave == NULL) {
1264 fprintf(stderr, "warning: %s, failed to read"
1265 " entries under %s/slaves\n",
1266 strerror(errno), path);
1268 fprintf(stderr, "unable to set tunables for"
1269 " slave device %s (slave would be"
1270 " unable to handle IO request from"
1275 globfree(&glob_info);
1280 int ldiskfs_tune_lustre(char *dev, struct mount_opts *mop)
1282 return set_blockdev_tunables(dev, mop);
1285 int ldiskfs_label_lustre(struct mount_opts *mop)
1287 char label_cmd[PATH_MAX];
1290 snprintf(label_cmd, sizeof(label_cmd),
1291 TUNE2FS" -f -L '%s' '%s' >/dev/null 2>&1",
1292 mop->mo_ldd.ldd_svname, mop->mo_source);
1293 rc = run_command(label_cmd, sizeof(label_cmd));
1298 int ldiskfs_rename_fsname(struct mkfs_opts *mop, const char *oldname)
1300 struct mount_opts opts;
1301 struct lustre_disk_data *ldd = &mop->mo_ldd;
1302 char mntpt[] = "/tmp/mntXXXXXX";
1306 /* Change the filesystem label. */
1308 opts.mo_source = mop->mo_device;
1309 ret = ldiskfs_label_lustre(&opts);
1313 fprintf(stderr, "Can't change filesystem label: %s\n",
1318 /* Mount this device temporarily in order to write these files */
1319 if (mkdtemp(mntpt) == NULL) {
1324 fprintf(stderr, "Can't create temp mount point %s: %s\n",
1325 mntpt, strerror(ret));
1331 * Append file context to mount options if SE Linux is enabled
1333 if (is_selinux_enabled() > 0)
1334 append_context_for_mount(mntpt, mop);
1337 if (mop->mo_flags & MO_IS_LOOP)
1338 dev = mop->mo_loopdev;
1340 dev = mop->mo_device;
1341 ret = mount(dev, mntpt, MT_STR(ldd), 0, ldd->ldd_mount_opts);
1345 fprintf(stderr, "Unable to mount %s: %s\n",
1346 dev, strerror(ret));
1348 fprintf(stderr, "Is the %s module available?\n",
1353 ret = lustre_rename_fsname(mop, mntpt, oldname);
1361 /* Enable quota accounting */
1362 int ldiskfs_enable_quota(struct mkfs_opts *mop)
1366 int cmdsz = sizeof(cmd), ret;
1368 if (is_e2fsprogs_feature_supp("-O quota") != 0) {
1369 fprintf(stderr, "%s: \"-O quota\" is is not supported by "
1370 "current e2fsprogs\n", progname);
1374 dev = mop->mo_device;
1375 if (mop->mo_flags & MO_IS_LOOP)
1376 dev = mop->mo_loopdev;
1378 /* Quota feature is already enabled? */
1379 if (is_feature_enabled("quota", dev)) {
1380 vprint("Quota feature is already enabled.\n");
1384 /* Turn on quota feature by "tune2fs -O quota" */
1385 snprintf(cmd, cmdsz, "%s -O quota %s", TUNE2FS, dev);
1386 ret = run_command(cmd, cmdsz);
1388 fprintf(stderr, "command:%s (%d)", cmd, ret);
1393 int ldiskfs_init(void)
1395 /* Required because full path to DEBUGFS is not specified */
1396 setenv("PATH", "/bin:/sbin:/usr/bin:/usr/sbin:/usr/local/bin", 0);
1401 void ldiskfs_fini(void)
1407 struct module_backfs_ops ldiskfs_ops = {
1408 .init = ldiskfs_init,
1409 .fini = ldiskfs_fini,
1410 .read_ldd = ldiskfs_read_ldd,
1411 .write_ldd = ldiskfs_write_ldd,
1412 .erase_ldd = ldiskfs_erase_ldd,
1413 .print_ldd_params = ldiskfs_print_ldd_params,
1414 .is_lustre = ldiskfs_is_lustre,
1415 .make_lustre = ldiskfs_make_lustre,
1416 .prepare_lustre = ldiskfs_prepare_lustre,
1417 .fix_mountopts = ldiskfs_fix_mountopts,
1418 .tune_lustre = ldiskfs_tune_lustre,
1419 .label_lustre = ldiskfs_label_lustre,
1420 .enable_quota = ldiskfs_enable_quota,
1421 .rename_fsname = ldiskfs_rename_fsname,
1423 #endif /* PLUGIN_DIR */