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 <lustre_param.h>
72 #include <lnet/lnetctl.h>
73 #include <lustre_ver.h>
76 #include <selinux/selinux.h>
79 #include "mount_utils.h"
81 #define MAX_HW_SECTORS_KB_PATH "queue/max_hw_sectors_kb"
82 #define MAX_SECTORS_KB_PATH "queue/max_sectors_kb"
83 #define SCHEDULER_PATH "queue/scheduler"
84 #define STRIPE_CACHE_SIZE "md/stripe_cache_size"
86 #define DEFAULT_SCHEDULER "deadline"
88 extern char *progname;
90 #define L_BLOCK_SIZE 4096
91 /* keep it less than LL_FID_NAMELEN */
92 #define DUMMY_FILE_NAME_LEN 25
93 #define EXT3_DIRENT_SIZE DUMMY_FILE_NAME_LEN
95 static void append_unique(char *buf, char *prefix, char *key, char *val,
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 /* return canonicalized absolute pathname, even if the target file does not
125 * exist, unlike realpath */
126 static char *absolute_path(char *devname)
128 char buf[PATH_MAX + 1] = "";
133 path = malloc(sizeof(buf));
137 if (devname[0] != '/') {
138 if (getcwd(buf, sizeof(buf) - 1) == NULL) {
142 len = snprintf(path, sizeof(buf), "%s/%s", buf, devname);
143 if (len >= sizeof(buf)) {
148 len = snprintf(path, sizeof(buf), "%s", devname);
149 if (len >= sizeof(buf)) {
155 /* truncate filename before calling realpath */
156 ptr = strrchr(path, '/');
162 if (buf != realpath(path, buf)) {
166 /* add the filename back */
167 len = snprintf(path, PATH_MAX, "%s/%s", buf, ptr+1);
168 if (len >= PATH_MAX) {
175 /* Determine if a device is a block device (as opposed to a file) */
176 static int is_block(char *devname)
182 devpath = absolute_path(devname);
183 if (devpath == NULL) {
184 fprintf(stderr, "%s: failed to resolve path to %s\n",
189 ret = access(devname, F_OK);
191 if (strncmp(devpath, "/dev/", 5) == 0) {
192 /* nobody sane wants to create a loopback file under
193 * /dev. Let's just report the device doesn't exist */
194 fprintf(stderr, "%s: %s apparently does not exist\n",
202 ret = stat(devpath, &st);
204 fprintf(stderr, "%s: cannot stat %s\n", progname, devpath);
207 ret = S_ISBLK(st.st_mode);
213 static int is_feature_enabled(const char *feature, const char *devpath)
217 char enabled_features[4096] = "";
220 snprintf(cmd, sizeof(cmd), "%s -c -R features %s 2>&1",
223 /* Using popen() instead of run_command() since debugfs does
224 * not return proper error code if command is not supported */
225 fp = popen(cmd, "r");
227 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
231 ret = fread(enabled_features, 1, sizeof(enabled_features) - 1, fp);
232 enabled_features[ret] = '\0';
235 if (strstr(enabled_features, feature))
240 /* Write the server config files */
241 int ldiskfs_write_ldd(struct mkfs_opts *mop)
243 char mntpt[] = "/tmp/mntXXXXXX";
250 /* Mount this device temporarily in order to write these files */
251 if (!mkdtemp(mntpt)) {
252 fprintf(stderr, "%s: Can't create temp mount point %s: %s\n",
253 progname, mntpt, strerror(errno));
258 * Append file context to mount options if SE Linux is enabled
261 if (is_selinux_enabled() > 0)
262 append_context_for_mount(mntpt, mop);
265 dev = mop->mo_device;
266 if (mop->mo_flags & MO_IS_LOOP)
267 dev = mop->mo_loopdev;
269 /* Multiple mount protection enabled if failover node specified */
270 if (mop->mo_flags & MO_FAILOVER &&
271 !is_feature_enabled("mmp", dev)) {
272 if (is_e2fsprogs_feature_supp("-O mmp") == 0) {
273 char *command = filepnm;
275 snprintf(command, sizeof(filepnm),
276 TUNE2FS" -O mmp '%s' >/dev/null 2>&1", dev);
277 ret = run_command(command, sizeof(filepnm));
280 "%s: Unable to set 'mmp' on %s: %d\n",
283 disp_old_e2fsprogs_msg("mmp", 1);
286 ret = mount(dev, mntpt, MT_STR(&mop->mo_ldd), 0,
287 (mop->mo_mountopts == NULL) ?
288 "errors=remount-ro" : mop->mo_mountopts);
290 fprintf(stderr, "%s: Unable to mount %s: %s\n",
291 progname, dev, strerror(errno));
293 if (errno == ENODEV) {
294 fprintf(stderr, "Is the %s module available?\n",
295 MT_STR(&mop->mo_ldd));
300 /* Set up initial directories */
301 sprintf(filepnm, "%s/%s", mntpt, MOUNT_CONFIGS_DIR);
302 ret = mkdir(filepnm, 0777);
303 if ((ret != 0) && (errno != EEXIST)) {
304 fprintf(stderr, "%s: Can't make configs dir %s (%s)\n",
305 progname, filepnm, strerror(errno));
307 } else if (errno == EEXIST) {
311 /* Save the persistent mount data into a file. Lustre must pre-read
312 this file to get the real mount options. */
313 vprint("Writing %s\n", MOUNT_DATA_FILE);
314 sprintf(filepnm, "%s/%s", mntpt, MOUNT_DATA_FILE);
315 filep = fopen(filepnm, "w");
317 fprintf(stderr, "%s: Unable to create %s file: %s\n",
318 progname, filepnm, strerror(errno));
321 num = fwrite(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
322 if (num < 1 && ferror(filep)) {
323 fprintf(stderr, "%s: Unable to write to file (%s): %s\n",
324 progname, filepnm, strerror(errno));
328 fsync(filep->_fileno);
338 static int readcmd(char *cmd, char *buf, int len)
343 fp = popen(cmd, "r");
347 red = fread(buf, 1, len, fp);
350 /* strip trailing newline */
351 if (buf[red - 1] == '\n')
354 return (red == 0) ? -ENOENT : 0;
357 int ldiskfs_read_ldd(char *dev, struct lustre_disk_data *mo_ldd)
359 char tmpdir[] = "/tmp/dirXXXXXX";
364 int cmdsz = sizeof(cmd);
366 /* Make a temporary directory to hold Lustre data files. */
367 if (!mkdtemp(tmpdir)) {
368 fprintf(stderr, "%s: Can't create temporary directory %s: %s\n",
369 progname, tmpdir, strerror(errno));
373 /* TODO: it's worth observing the get_mountdata() function that is
374 in mount_utils.c for getting the mountdata out of the
377 /* Construct debugfs command line. */
378 snprintf(cmd, cmdsz, "%s -c -R 'dump /%s %s/mountdata' '%s'",
379 DEBUGFS, MOUNT_DATA_FILE, tmpdir, dev);
381 ret = run_command(cmd, cmdsz);
383 verrprint("%s: Unable to dump %s dir (%d)\n",
384 progname, MOUNT_CONFIGS_DIR, ret);
386 sprintf(filepnm, "%s/mountdata", tmpdir);
387 filep = fopen(filepnm, "r");
390 vprint("Reading %s\n", MOUNT_DATA_FILE);
391 num_read = fread(mo_ldd, sizeof(*mo_ldd), 1, filep);
392 if (num_read < 1 && ferror(filep)) {
393 fprintf(stderr, "%s: Unable to read from file %s: %s\n",
394 progname, filepnm, strerror(errno));
399 snprintf(cmd, cmdsz, "rm -rf %s", tmpdir);
400 run_command(cmd, cmdsz);
402 verrprint("Failed to read old data (%d)\n", ret);
404 /* As long as we at least have the label, we're good to go */
405 snprintf(cmd, sizeof(cmd), E2LABEL" %s", dev);
406 ret = readcmd(cmd, mo_ldd->ldd_svname, sizeof(mo_ldd->ldd_svname) - 1);
411 int ldiskfs_erase_ldd(struct mkfs_opts *mop, char *param)
416 void ldiskfs_print_ldd_params(struct mkfs_opts *mop)
418 printf("Parameters:%s\n", mop->mo_ldd.ldd_params);
421 /* Display the need for the latest e2fsprogs to be installed. make_backfs
422 * indicates if the caller is make_lustre_backfs() or not. */
423 static void disp_old_e2fsprogs_msg(const char *feature, int make_backfs)
425 static int msg_displayed;
428 fprintf(stderr, "WARNING: %s does not support %s "
429 "feature.\n\n", E2FSPROGS, feature);
435 fprintf(stderr, "WARNING: The %s package currently installed on "
436 "your system does not support \"%s\" feature.\n",
438 #if !(HAVE_LDISKFSPROGS)
439 fprintf(stderr, "Please install the latest version of e2fsprogs from\n"
440 "https://downloads.hpdd.intel.com/public/e2fsprogs/latest/\n"
441 "to enable this feature.\n");
444 fprintf(stderr, "Feature will not be enabled until %s"
445 "is updated and '%s -O %s %%{device}' "
446 "is run.\n\n", E2FSPROGS, TUNE2FS, feature);
449 /* Check whether the file exists in the device */
450 static int file_in_dev(char *file_name, char *dev_name)
453 char debugfs_cmd[256];
454 unsigned int inode_num;
457 /* Construct debugfs command line. */
458 snprintf(debugfs_cmd, sizeof(debugfs_cmd),
459 "%s -c -R 'stat %s' '%s' 2>&1 | egrep '(Inode|unsupported)'",
460 DEBUGFS, file_name, dev_name);
462 fp = popen(debugfs_cmd, "r");
464 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
468 if (fscanf(fp, "Inode: %u", &inode_num) == 1) { /* exist */
472 i = fread(debugfs_cmd, 1, sizeof(debugfs_cmd) - 1, fp);
475 fprintf(stderr, "%s", debugfs_cmd);
476 if (strstr(debugfs_cmd, "unsupported feature")) {
477 disp_old_e2fsprogs_msg("an unknown", 0);
486 /* Check whether the device has already been used with lustre */
487 int ldiskfs_is_lustre(char *dev, unsigned *mount_type)
491 ret = file_in_dev(MOUNT_DATA_FILE, dev);
493 /* in the -1 case, 'extents' means IS a lustre target */
494 *mount_type = LDD_MT_LDISKFS;
498 ret = file_in_dev(LAST_RCVD, dev);
500 *mount_type = LDD_MT_LDISKFS;
507 /* Check if a certain feature is supported by e2fsprogs.
508 * Firstly we try to use "debugfs supported_features" command to check if
509 * the feature is supported. If this fails we try to set this feature with
510 * mke2fs to check for its support. */
511 static int is_e2fsprogs_feature_supp(const char *feature)
513 static char supp_features[4096] = "";
516 char imgname[] = "/tmp/test-img-XXXXXX";
520 if (supp_features[0] == '\0') {
521 snprintf(cmd, sizeof(cmd), "%s -c -R supported_features 2>&1",
524 /* Using popen() instead of run_command() since debugfs does
525 * not return proper error code if command is not supported */
526 fp = popen(cmd, "r");
528 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
531 ret = fread(supp_features, 1, sizeof(supp_features) - 1, fp);
532 supp_features[ret] = '\0';
535 if (ret > 0 && strstr(supp_features,
536 strncmp(feature, "-O ", 3) ? feature : feature+3))
539 if ((fd = mkstemp(imgname)) < 0)
544 snprintf(cmd, sizeof(cmd), "%s -F %s %s 100 >/dev/null 2>&1",
545 MKE2FS, feature, imgname);
546 /* run_command() displays the output of mke2fs when it fails for
547 * some feature, so use system() directly */
556 * append_unique: append @key or @key=@val pair to @buf only if @key does not
558 * @buf: buffer to hold @key or @key=@val
559 * @prefix: prefix string before @key
561 * @val: value string if it's a @key=@val pair
563 static void append_unique(char *buf, char *prefix, char *key, char *val,
572 anchor = end = strstr(buf, key);
573 /* try to find exact match string in @buf */
574 while (end && *end != '\0' && *end != ',' && *end != ' ' && *end != '=')
577 if (anchor == NULL || strlen(key) != len ||
578 strncmp(anchor, key, len) != 0) {
580 strscat(buf, prefix, maxbuflen);
582 strscat(buf, key, maxbuflen);
584 strscat(buf, "=\"", maxbuflen);
585 strscat(buf, val, maxbuflen);
586 strscat(buf, "\"", maxbuflen);
591 static int enable_default_ext4_features(struct mkfs_opts *mop, char *anchor,
592 size_t maxbuflen, int user_spec)
594 if (IS_OST(&mop->mo_ldd)) {
595 append_unique(anchor, user_spec ? "," : " -O ",
596 "extents", NULL, maxbuflen);
597 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
598 } else if (IS_MDT(&mop->mo_ldd)) {
599 append_unique(anchor, user_spec ? "," : " -O ",
600 "dirdata", NULL, maxbuflen);
601 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
602 append_unique(anchor, ",", "^extents", NULL, maxbuflen);
604 append_unique(anchor, user_spec ? "," : " -O ",
605 "uninit_bg", NULL, maxbuflen);
608 /* Multiple mount protection enabled only if failover node specified */
609 if (mop->mo_flags & MO_FAILOVER) {
610 if (is_e2fsprogs_feature_supp("-O mmp") == 0)
611 append_unique(anchor, ",", "mmp", NULL, maxbuflen);
613 disp_old_e2fsprogs_msg("mmp", 1);
616 /* Allow more than 65000 subdirectories */
617 if (is_e2fsprogs_feature_supp("-O dir_nlink") == 0)
618 append_unique(anchor, ",", "dir_nlink", NULL, maxbuflen);
620 /* The following options are only valid for ext4-based ldiskfs.
621 * If --backfstype=ext3 is specified, do not enable them. */
622 if (mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3)
625 /* Enable quota by default */
626 if (is_e2fsprogs_feature_supp("-O quota") == 0) {
627 append_unique(anchor, ",", "quota", NULL, maxbuflen);
630 fprintf(stderr, "\"-O quota\" must be supported by "
631 "e2fsprogs, please upgrade your e2fsprogs.\n");
635 /* Allow files larger than 2TB. Also needs LU-16, but not harmful. */
636 if (is_e2fsprogs_feature_supp("-O huge_file") == 0)
637 append_unique(anchor, ",", "huge_file", NULL, maxbuflen);
639 /* Enable large block addresses if the LUN is over 2^32 blocks. */
640 if (mop->mo_device_kb / (L_BLOCK_SIZE >> 10) >= 0x100002000ULL &&
641 is_e2fsprogs_feature_supp("-O 64bit") == 0)
642 append_unique(anchor, ",", "64bit", NULL, maxbuflen);
644 /* Cluster inode/block bitmaps and inode table for more efficient IO.
645 * Align the flex groups on a 1MB boundary for better performance. */
646 /* This -O feature needs to go last, since it adds the "-G" option. */
647 if (is_e2fsprogs_feature_supp("-O flex_bg") == 0) {
650 append_unique(anchor, ",", "flex_bg", NULL, maxbuflen);
652 if (IS_OST(&mop->mo_ldd) &&
653 strstr(mop->mo_mkfsopts, "-G") == NULL) {
654 snprintf(tmp_buf, sizeof(tmp_buf), " -G %u",
655 (1 << 20) / L_BLOCK_SIZE);
656 strscat(anchor, tmp_buf, maxbuflen);
659 /* Don't add any more "-O" options here, see last comment above */
664 * moveopts_to_end: find the option string, move remaining strings to
665 * where option string starts, and append the option
667 * @start: where the option string starts before the move
668 * RETURN: where the option string starts after the move
670 static char *moveopts_to_end(char *start)
676 /* skip whitespace before options */
681 /* find end of option characters */
682 while (*end != ' ' && *end != '\0')
686 if (len >= sizeof(save))
687 len = sizeof(save) - 1;
690 strncpy(save, start, len);
693 /* move remaining options up front */
695 memmove(start, end, strlen(end));
696 *(start + strlen(end)) = '\0';
698 /* append the specified options */
699 if (*(start + strlen(start) - 1) != ' ')
701 idx = start + strlen(start);
707 /* Build fs according to type */
708 int ldiskfs_make_lustre(struct mkfs_opts *mop)
710 __u64 device_kb = mop->mo_device_kb, block_count = 0;
711 char mkfs_cmd[PATH_MAX];
715 int ret = 0, ext_opts = 0;
718 if (!(mop->mo_flags & MO_IS_LOOP)) {
719 mop->mo_device_kb = get_device_size(mop->mo_device);
721 if (mop->mo_device_kb == 0)
724 /* Compare to real size */
725 if (device_kb == 0 || device_kb > mop->mo_device_kb)
726 device_kb = mop->mo_device_kb;
728 mop->mo_device_kb = device_kb;
731 if (mop->mo_device_kb != 0) {
732 if (mop->mo_device_kb < 32384) {
733 fprintf(stderr, "%s: size of filesystem must be larger "
734 "than 32MB, but is set to %lldKB\n",
735 progname, (long long)mop->mo_device_kb);
738 block_count = mop->mo_device_kb / (L_BLOCK_SIZE >> 10);
739 /* If the LUN size is just over 2^32 blocks, limit the
740 * filesystem size to 2^32-1 blocks to avoid problems with
741 * ldiskfs/mkfs not handling this size. Bug 22906 */
742 if (block_count > 0xffffffffULL && block_count < 0x100002000ULL)
743 block_count = 0xffffffffULL;
746 if ((mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3) ||
747 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS) ||
748 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS2)) {
751 /* Journal size in MB */
752 if (strstr(mop->mo_mkfsopts, "-J") == NULL &&
753 device_kb > 1024 * 1024) {
754 /* Choose our own default journal size */
755 long journal_mb = 0, max_mb;
757 /* cap journal size at 4GB for MDT,
758 * leave it at 400MB for OSTs. */
759 if (IS_MDT(&mop->mo_ldd))
761 else if (IS_OST(&mop->mo_ldd))
763 else /* Use mke2fs default size for MGS */
766 /* Use at most 4% of device for journal */
767 journal_mb = device_kb * 4 / (1024 * 100);
768 if (journal_mb > max_mb)
772 sprintf(buf, " -J size=%ld", journal_mb);
773 strscat(mop->mo_mkfsopts, buf,
774 sizeof(mop->mo_mkfsopts));
779 * The inode size is constituted by following elements
780 * (assuming all files are in composite layout and has
783 * ldiskfs inode size: 156
784 * extended attributes size, including:
785 * ext4_xattr_header: 32
786 * LOV EA size: 32(lov_comp_md_v1) +
787 * 3 * 40(lov_comp_md_entry_v1) +
788 * 3 * 32(lov_mds_md) +
789 * stripes * 24(lov_ost_data) +
790 * 16(xattr_entry) + 3(lov)
791 * LMA EA size: 24(lustre_mdt_attrs) +
792 * 16(xattr_entry) + 3(lma)
793 * link EA size: 24(link_ea_header) + 18(link_ea_entry) +
794 * (filename) + 16(xattr_entry) + 4(link)
795 * and some margin for 4-byte alignment, ACLs and other EAs.
797 * If we say the average filename length is about 32 bytes,
798 * the calculation looks like:
799 * 156 + 32 + (32+3*(40 + 32)+24*N+19) + (24+19) +
800 * (24+18+~32+20) + other <= 512*2^m, {m=0,1,2,3}
802 if (strstr(mop->mo_mkfsopts, "-I") == NULL) {
803 if (IS_MDT(&mop->mo_ldd)) {
804 if (mop->mo_stripe_count > 59)
805 inode_size = 512; /* bz 7241 */
806 /* see also "-i" below for EA blocks */
807 else if (mop->mo_stripe_count > 16)
811 } else if (IS_OST(&mop->mo_ldd)) {
812 /* We store MDS FID and necessary composite
813 * layout information in the OST object EA. */
817 if (inode_size > 0) {
818 sprintf(buf, " -I %ld", inode_size);
819 strscat(mop->mo_mkfsopts, buf,
820 sizeof(mop->mo_mkfsopts));
824 /* Bytes_per_inode: disk size / num inodes */
825 if (strstr(mop->mo_mkfsopts, "-i") == NULL &&
826 strstr(mop->mo_mkfsopts, "-N") == NULL) {
827 long bytes_per_inode = 0;
829 /* Allocate more inodes on MDT devices. There is
830 * no data stored on the MDT, and very little extra
831 * metadata beyond the inode. It could go down as
832 * low as 1024 bytes, but this is conservative.
833 * Account for external EA blocks for wide striping. */
834 if (IS_MDT(&mop->mo_ldd)) {
835 bytes_per_inode = inode_size + 1536;
837 if (mop->mo_stripe_count > 59) {
838 int extra = mop->mo_stripe_count * 24;
839 extra = ((extra - 1) | 4095) + 1;
840 bytes_per_inode += extra;
844 /* Allocate fewer inodes on large OST devices. Most
845 * filesystems can be much more aggressive than even
846 * this, but it is impossible to know in advance. */
847 if (IS_OST(&mop->mo_ldd)) {
848 /* OST > 16TB assume average file size 1MB */
849 if (device_kb > (16ULL << 30))
850 bytes_per_inode = 1024 * 1024;
851 /* OST > 4TB assume average file size 512kB */
852 else if (device_kb > (4ULL << 30))
853 bytes_per_inode = 512 * 1024;
854 /* OST > 1TB assume average file size 256kB */
855 else if (device_kb > (1ULL << 30))
856 bytes_per_inode = 256 * 1024;
857 /* OST > 10GB assume average file size 64kB,
858 * plus a bit so that inodes will fit into a
859 * 256x flex_bg without overflowing */
860 else if (device_kb > (10ULL << 20))
861 bytes_per_inode = 69905;
864 if (bytes_per_inode > 0) {
865 sprintf(buf, " -i %ld", bytes_per_inode);
866 strscat(mop->mo_mkfsopts, buf,
867 sizeof(mop->mo_mkfsopts));
872 strscat(mop->mo_mkfsopts, " -q",
873 sizeof(mop->mo_mkfsopts));
876 /* start handle -O mkfs options */
877 if ((start = strstr(mop->mo_mkfsopts, "-O")) != NULL) {
878 if (strstr(start + 2, "-O") != NULL) {
880 "%s: don't specify multiple -O options\n",
884 start = moveopts_to_end(start);
885 maxbuflen = sizeof(mop->mo_mkfsopts) -
886 (start - mop->mo_mkfsopts) - strlen(start);
887 ret = enable_default_ext4_features(mop, start, maxbuflen, 1);
889 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts),
890 maxbuflen = sizeof(mop->mo_mkfsopts) -
891 strlen(mop->mo_mkfsopts);
892 ret = enable_default_ext4_features(mop, start, maxbuflen, 0);
896 /* end handle -O mkfs options */
898 /* start handle -E mkfs options */
899 if ((start = strstr(mop->mo_mkfsopts, "-E")) != NULL) {
900 if (strstr(start + 2, "-E") != NULL) {
902 "%s: don't specify multiple -E options\n",
906 start = moveopts_to_end(start);
907 maxbuflen = sizeof(mop->mo_mkfsopts) -
908 (start - mop->mo_mkfsopts) - strlen(start);
911 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts);
912 maxbuflen = sizeof(mop->mo_mkfsopts) -
913 strlen(mop->mo_mkfsopts);
916 /* In order to align the filesystem metadata on 1MB boundaries,
917 * give a resize value that will reserve a power-of-two group
918 * descriptor blocks, but leave one block for the superblock.
919 * Only useful for filesystems with < 2^32 blocks due to resize
921 if (strstr(mop->mo_mkfsopts, "meta_bg") == NULL &&
922 IS_OST(&mop->mo_ldd) && mop->mo_device_kb > 100 * 1024 &&
923 mop->mo_device_kb * 1024 / L_BLOCK_SIZE <= 0xffffffffULL) {
924 unsigned group_blocks = L_BLOCK_SIZE * 8;
925 unsigned desc_per_block = L_BLOCK_SIZE / 32;
926 unsigned resize_blks;
928 resize_blks = (1ULL<<32) - desc_per_block*group_blocks;
929 snprintf(buf, sizeof(buf), "%u", resize_blks);
930 append_unique(start, ext_opts ? "," : " -E ",
931 "resize", buf, maxbuflen);
935 /* Avoid zeroing out the full journal - speeds up mkfs */
936 if (is_e2fsprogs_feature_supp("-E lazy_journal_init") == 0)
937 append_unique(start, ext_opts ? "," : " -E ",
938 "lazy_journal_init", NULL, maxbuflen);
939 /* end handle -E mkfs options */
941 /* Allow reformat of full devices (as opposed to
942 partitions.) We already checked for mounted dev. */
943 strscat(mop->mo_mkfsopts, " -F", sizeof(mop->mo_mkfsopts));
945 snprintf(mkfs_cmd, sizeof(mkfs_cmd),
946 "%s -j -b %d -L %s ", MKE2FS, L_BLOCK_SIZE,
947 mop->mo_ldd.ldd_svname);
949 fprintf(stderr,"%s: unsupported fs type: %d (%s)\n",
950 progname, mop->mo_ldd.ldd_mount_type,
951 MT_STR(&mop->mo_ldd));
955 /* For loop device format the dev, not the filename */
956 dev = mop->mo_device;
957 if (mop->mo_flags & MO_IS_LOOP)
958 dev = mop->mo_loopdev;
960 vprint("formatting backing filesystem %s on %s\n",
961 MT_STR(&mop->mo_ldd), dev);
962 vprint("\ttarget name %s\n", mop->mo_ldd.ldd_svname);
963 vprint("\t4k blocks %ju\n", (uintmax_t)block_count);
964 vprint("\toptions %s\n", mop->mo_mkfsopts);
966 /* mkfs_cmd's trailing space is important! */
967 strscat(mkfs_cmd, mop->mo_mkfsopts, sizeof(mkfs_cmd));
968 strscat(mkfs_cmd, " ", sizeof(mkfs_cmd));
969 strscat(mkfs_cmd, dev, sizeof(mkfs_cmd));
970 if (block_count != 0) {
971 snprintf(buf, sizeof(buf), " %ju",
972 (uintmax_t)block_count);
973 strscat(mkfs_cmd, buf, sizeof(mkfs_cmd));
976 vprint("mkfs_cmd = %s\n", mkfs_cmd);
977 ret = run_command(mkfs_cmd, sizeof(mkfs_cmd));
980 fprintf(stderr, "Unable to build fs %s (%d)\n", dev, ret);
985 int ldiskfs_prepare_lustre(struct mkfs_opts *mop,
986 char *wanted_mountopts, size_t len)
988 struct lustre_disk_data *ldd = &mop->mo_ldd;
991 /* Set MO_IS_LOOP to indicate a loopback device is needed */
992 ret = is_block(mop->mo_device);
995 } else if (ret == 0) {
996 mop->mo_flags |= MO_IS_LOOP;
999 if (IS_MDT(ldd) || IS_MGS(ldd))
1000 strscat(wanted_mountopts, ",user_xattr", len);
1005 int ldiskfs_fix_mountopts(struct mkfs_opts *mop, char *mountopts, size_t len)
1007 if (strstr(mountopts, "errors=") == NULL)
1008 strscat(mountopts, ",errors=remount-ro", len);
1013 static int read_file(const char *path, char *buf, int size)
1017 fd = fopen(path, "r");
1021 if (fgets(buf, size, fd) == NULL) {
1022 fprintf(stderr, "reading from %s: %s", path, strerror(errno));
1028 /* strip trailing newline */
1030 if (buf[size - 1] == '\n')
1031 buf[size - 1] = '\0';
1036 static int write_file(const char *path, const char *buf)
1040 fd = open(path, O_WRONLY);
1044 rc = write(fd, buf, strlen(buf));
1047 return rc < 0 ? errno : 0;
1050 static int set_blockdev_scheduler(const char *path, const char *scheduler)
1052 char buf[PATH_MAX], *s, *e, orig_sched[50];
1055 /* Before setting the scheduler, we need to check to see if it's
1056 * already set to "noop". If it is, we don't want to override
1057 * that setting. If it's set to anything other than "noop", set
1058 * the scheduler to what has been passed in. */
1060 rc = read_file(path, buf, sizeof(buf));
1063 fprintf(stderr, "%s: cannot open '%s': %s\n",
1064 progname, path, strerror(errno));
1068 /* The expected format of buf: noop anticipatory deadline [cfq] */
1069 s = strchr(buf, '[');
1070 e = strchr(buf, ']');
1072 /* If the format is not what we expect. Play it safe and error out. */
1073 if (s == NULL || e == NULL) {
1075 fprintf(stderr, "%s: cannot parse scheduler "
1076 "options for '%s'\n", progname, path);
1080 snprintf(orig_sched, e - s, "%s", s + 1);
1082 if (strcmp(orig_sched, "noop") == 0 ||
1083 strcmp(orig_sched, scheduler) == 0)
1086 rc = write_file(path, scheduler);
1089 fprintf(stderr, "%s: cannot set scheduler on "
1090 "'%s': %s\n", progname, path,
1094 fprintf(stderr, "%s: change scheduler of %s from %s to %s\n",
1095 progname, path, orig_sched, scheduler);
1101 /* This is to tune the kernel for good SCSI performance.
1102 * For that we set the value of /sys/block/{dev}/queue/max_sectors_kb
1103 * to the value of /sys/block/{dev}/queue/max_hw_sectors_kb */
1104 static int set_blockdev_tunables(char *source, struct mount_opts *mop)
1106 glob_t glob_info = { 0 };
1107 struct stat stat_buf;
1108 char *chk_major, *chk_minor;
1109 char *savept = NULL, *dev;
1111 char buf[PATH_MAX] = {'\0'}, path[PATH_MAX] = {'\0'};
1112 char real_path[PATH_MAX] = {'\0'};
1120 ret_path = realpath(source, real_path);
1121 if (ret_path == NULL) {
1123 fprintf(stderr, "warning: %s: cannot resolve: %s\n",
1124 source, strerror(errno));
1128 if (strncmp(real_path, "/dev/loop", 9) == 0)
1131 if ((real_path[0] != '/') && (strpbrk(real_path, ",:") != NULL))
1134 snprintf(path, sizeof(path), "/sys/block%s", real_path + 4);
1135 if (access(path, X_OK) == 0)
1138 /* The name of the device say 'X' specified in /dev/X may not
1139 * match any entry under /sys/block/. In that case we need to
1140 * match the major/minor number to find the entry under
1141 * sys/block corresponding to /dev/X */
1143 /* Don't chop tail digit on /dev/mapper/xxx, LU-478 */
1144 if (strncmp(real_path, "/dev/mapper", 11) != 0) {
1145 dev = real_path + strlen(real_path);
1146 while (--dev > real_path && isdigit(*dev))
1149 if (strncmp(real_path, "/dev/md", 7) == 0 && dev[0] == 'p')
1153 rc = stat(real_path, &stat_buf);
1156 fprintf(stderr, "warning: %s, device %s stat failed\n",
1157 strerror(errno), real_path);
1161 major = major(stat_buf.st_rdev);
1162 minor = minor(stat_buf.st_rdev);
1163 rc = glob("/sys/block/*", GLOB_NOSORT, NULL, &glob_info);
1166 fprintf(stderr, "warning: failed to read entries under "
1168 globfree(&glob_info);
1172 for (i = 0; i < glob_info.gl_pathc; i++){
1173 snprintf(path, sizeof(path), "%s/dev", glob_info.gl_pathv[i]);
1175 rc = read_file(path, buf, sizeof(buf));
1179 if (buf[strlen(buf) - 1] == '\n')
1180 buf[strlen(buf) - 1] = '\0';
1182 chk_major = strtok_r(buf, ":", &savept);
1184 if (chk_major != NULL && major == atoi(chk_major) &&
1185 chk_minor != NULL && minor == atoi(chk_minor))
1189 if (i == glob_info.gl_pathc) {
1191 fprintf(stderr,"warning: device %s does not match any "
1192 "entry under /sys/block\n", real_path);
1193 globfree(&glob_info);
1197 /* Chop off "/dev" from path we found */
1198 path[strlen(glob_info.gl_pathv[i])] = '\0';
1199 globfree(&glob_info);
1202 if (strncmp(real_path, "/dev/md", 7) == 0) {
1203 snprintf(real_path, sizeof(real_path), "%s/%s", path,
1206 rc = read_file(real_path, buf, sizeof(buf));
1209 fprintf(stderr, "warning: opening %s: %s\n",
1210 real_path, strerror(errno));
1214 if (atoi(buf) >= mop->mo_md_stripe_cache_size)
1217 if (strlen(buf) - 1 > 0) {
1218 snprintf(buf, sizeof(buf), "%d",
1219 mop->mo_md_stripe_cache_size);
1220 rc = write_file(real_path, buf);
1221 if (rc != 0 && verbose)
1222 fprintf(stderr, "warning: opening %s: %s\n",
1223 real_path, strerror(errno));
1225 /* Return since raid and disk tunables are different */
1229 if (mop->mo_max_sectors_kb >= 0) {
1230 snprintf(buf, sizeof(buf), "%d", mop->mo_max_sectors_kb);
1232 snprintf(real_path, sizeof(real_path), "%s/%s", path,
1233 MAX_HW_SECTORS_KB_PATH);
1234 rc = read_file(real_path, buf, sizeof(buf));
1237 fprintf(stderr, "warning: opening %s: %s\n",
1238 real_path, strerror(errno));
1239 /* No MAX_HW_SECTORS_KB_PATH isn't necessary an
1240 * error for some device. */
1245 if (strlen(buf) - 1 > 0) {
1246 char oldbuf[32] = "", *end = NULL;
1247 unsigned long long oldval, newval;
1249 snprintf(real_path, sizeof(real_path), "%s/%s", path,
1250 MAX_SECTORS_KB_PATH);
1251 rc = read_file(real_path, oldbuf, sizeof(oldbuf));
1252 /* Only set new parameter if different from the old one. */
1253 if (rc != 0 || strcmp(oldbuf, buf) == 0) {
1254 /* No MAX_SECTORS_KB_PATH isn't necessary an
1255 * error for some device. */
1259 newval = strtoull(buf, &end, 0);
1260 if (newval == 0 || newval == ULLONG_MAX || end == buf)
1263 /* Don't increase IO request size limit past 16MB. It is about
1264 * PTLRPC_MAX_BRW_SIZE, but that isn't in a public header.
1265 * Note that even though the block layer allows larger values,
1266 * setting max_sectors_kb = 32768 causes crashes (LU-6974). */
1267 if (mop->mo_max_sectors_kb < 0 && newval > 16 * 1024) {
1269 snprintf(buf, sizeof(buf), "%llu", newval);
1272 oldval = strtoull(oldbuf, &end, 0);
1273 /* Don't shrink the current limit. */
1274 if (mop->mo_max_sectors_kb < 0 && oldval != ULLONG_MAX &&
1278 rc = write_file(real_path, buf);
1281 fprintf(stderr, "warning: writing to %s: %s\n",
1282 real_path, strerror(errno));
1283 /* No MAX_SECTORS_KB_PATH isn't necessary an
1284 * error for some device. */
1287 fprintf(stderr, "%s: increased %s from %s to %s\n",
1288 progname, real_path, oldbuf, buf);
1292 /* Purposely ignore errors reported from set_blockdev_scheduler.
1293 * The worst that will happen is a block device with an "incorrect"
1295 snprintf(real_path, sizeof(real_path), "%s/%s", path, SCHEDULER_PATH);
1296 set_blockdev_scheduler(real_path, DEFAULT_SCHEDULER);
1298 /* if device is multipath device, tune its slave devices */
1299 glob_info.gl_pathc = 0;
1300 glob_info.gl_offs = 0;
1301 snprintf(real_path, sizeof(real_path), "%s/slaves/*", path);
1302 rc = glob(real_path, GLOB_NOSORT, NULL, &glob_info);
1304 for (i = 0; rc == 0 && i < glob_info.gl_pathc; i++) {
1305 slave = basename(glob_info.gl_pathv[i]);
1306 snprintf(real_path, sizeof(real_path), "/dev/%s", slave);
1307 rc = set_blockdev_tunables(real_path, mop);
1310 if (rc == GLOB_NOMATCH) {
1311 /* no slave device is not an error */
1313 } else if (rc && verbose) {
1314 if (slave == NULL) {
1315 fprintf(stderr, "warning: %s, failed to read"
1316 " entries under %s/slaves\n",
1317 strerror(errno), path);
1319 fprintf(stderr, "unable to set tunables for"
1320 " slave device %s (slave would be"
1321 " unable to handle IO request from"
1326 globfree(&glob_info);
1331 int ldiskfs_tune_lustre(char *dev, struct mount_opts *mop)
1333 return set_blockdev_tunables(dev, mop);
1336 int ldiskfs_label_lustre(struct mount_opts *mop)
1338 char label_cmd[PATH_MAX];
1341 snprintf(label_cmd, sizeof(label_cmd),
1342 TUNE2FS" -f -L '%s' '%s' >/dev/null 2>&1",
1343 mop->mo_ldd.ldd_svname, mop->mo_source);
1344 rc = run_command(label_cmd, sizeof(label_cmd));
1349 int ldiskfs_rename_fsname(struct mkfs_opts *mop, const char *oldname)
1351 struct mount_opts opts;
1352 struct lustre_disk_data *ldd = &mop->mo_ldd;
1353 char mntpt[] = "/tmp/mntXXXXXX";
1357 /* Change the filesystem label. */
1359 opts.mo_source = mop->mo_device;
1360 ret = ldiskfs_label_lustre(&opts);
1364 fprintf(stderr, "Can't change filesystem label: %s\n",
1369 /* Mount this device temporarily in order to write these files */
1370 if (mkdtemp(mntpt) == NULL) {
1375 fprintf(stderr, "Can't create temp mount point %s: %s\n",
1376 mntpt, strerror(ret));
1382 * Append file context to mount options if SE Linux is enabled
1384 if (is_selinux_enabled() > 0)
1385 append_context_for_mount(mntpt, mop);
1388 if (mop->mo_flags & MO_IS_LOOP)
1389 dev = mop->mo_loopdev;
1391 dev = mop->mo_device;
1392 ret = mount(dev, mntpt, MT_STR(ldd), 0, ldd->ldd_mount_opts);
1396 fprintf(stderr, "Unable to mount %s: %s\n",
1397 dev, strerror(ret));
1399 fprintf(stderr, "Is the %s module available?\n",
1404 ret = lustre_rename_fsname(mop, mntpt, oldname);
1412 /* Enable quota accounting */
1413 int ldiskfs_enable_quota(struct mkfs_opts *mop)
1417 int cmdsz = sizeof(cmd), ret;
1419 if (is_e2fsprogs_feature_supp("-O quota") != 0) {
1420 fprintf(stderr, "%s: \"-O quota\" is is not supported by "
1421 "current e2fsprogs\n", progname);
1425 dev = mop->mo_device;
1426 if (mop->mo_flags & MO_IS_LOOP)
1427 dev = mop->mo_loopdev;
1429 /* Quota feature is already enabled? */
1430 if (is_feature_enabled("quota", dev)) {
1431 vprint("Quota feature is already enabled.\n");
1435 /* Turn on quota feature by "tune2fs -O quota" */
1436 snprintf(cmd, cmdsz, "%s -O quota %s", TUNE2FS, dev);
1437 ret = run_command(cmd, cmdsz);
1439 fprintf(stderr, "command:%s (%d)", cmd, ret);
1444 int ldiskfs_init(void)
1446 /* Required because full path to DEBUGFS is not specified */
1447 setenv("PATH", "/bin:/sbin:/usr/bin:/usr/sbin:/usr/local/bin", 0);
1452 void ldiskfs_fini(void)