4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2012, 2014, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/utils/mount_utils_ldiskfs.c
38 * Author: Nathan Rutman <nathan@clusterfs.com>
41 /* This source file is compiled into both mkfs.lustre and tunefs.lustre */
45 #endif /* HAVE_CONFIG_H */
50 #include "mount_utils.h"
59 #include <sys/types.h>
61 #include <sys/mount.h>
62 #include <sys/utsname.h>
69 /* libcfs.h is not really needed here, but on SLES10/PPC, fs.h includes idr.h
70 * which requires BITS_PER_LONG to be defined */
71 #include <libcfs/libcfs.h>
73 #include <linux/fs.h> /* for BLKGETSIZE64 */
75 #include <linux/version.h>
76 #include <lustre_disk.h>
77 #include <lustre_param.h>
78 #include <lnet/lnetctl.h>
79 #include <lustre_ver.h>
82 #include <selinux/selinux.h>
85 #define MAX_HW_SECTORS_KB_PATH "queue/max_hw_sectors_kb"
86 #define MAX_SECTORS_KB_PATH "queue/max_sectors_kb"
87 #define SCHEDULER_PATH "queue/scheduler"
88 #define STRIPE_CACHE_SIZE "md/stripe_cache_size"
90 #define DEFAULT_SCHEDULER "deadline"
92 extern char *progname;
94 #define L_BLOCK_SIZE 4096
95 /* keep it less than LL_FID_NAMELEN */
96 #define DUMMY_FILE_NAME_LEN 25
97 #define EXT3_DIRENT_SIZE DUMMY_FILE_NAME_LEN
100 * Concatenate context of the temporary mount point iff selinux is enabled
103 static void append_context_for_mount(char *mntpt, struct mkfs_opts *mop)
105 security_context_t fcontext;
107 if (getfilecon(mntpt, &fcontext) < 0) {
108 /* Continuing with default behaviour */
109 fprintf(stderr, "%s: Get file context failed : %s\n",
110 progname, strerror(errno));
114 if (fcontext != NULL) {
115 strcat(mop->mo_ldd.ldd_mount_opts, ",context=");
116 strcat(mop->mo_ldd.ldd_mount_opts, fcontext);
122 /* return canonicalized absolute pathname, even if the target file does not
123 * exist, unlike realpath */
124 static char *absolute_path(char *devname)
126 char buf[PATH_MAX + 1] = "";
131 path = malloc(sizeof(buf));
135 if (devname[0] != '/') {
136 if (getcwd(buf, sizeof(buf) - 1) == NULL) {
140 len = snprintf(path, sizeof(buf), "%s/%s", buf, devname);
141 if (len >= sizeof(buf)) {
146 len = snprintf(path, sizeof(buf), "%s", devname);
147 if (len >= sizeof(buf)) {
153 /* truncate filename before calling realpath */
154 ptr = strrchr(path, '/');
160 if (buf != realpath(path, buf)) {
164 /* add the filename back */
165 len = snprintf(path, PATH_MAX, "%s/%s", buf, ptr+1);
166 if (len >= PATH_MAX) {
173 /* Determine if a device is a block device (as opposed to a file) */
174 static int is_block(char *devname)
180 devpath = absolute_path(devname);
181 if (devpath == NULL) {
182 fprintf(stderr, "%s: failed to resolve path to %s\n",
187 ret = access(devname, F_OK);
189 if (strncmp(devpath, "/dev/", 5) == 0) {
190 /* nobody sane wants to create a loopback file under
191 * /dev. Let's just report the device doesn't exist */
192 fprintf(stderr, "%s: %s apparently does not exist\n",
200 ret = stat(devpath, &st);
202 fprintf(stderr, "%s: cannot stat %s\n", progname, devpath);
205 ret = S_ISBLK(st.st_mode);
211 static int is_feature_enabled(const char *feature, const char *devpath)
215 char enabled_features[4096] = "";
218 snprintf(cmd, sizeof(cmd), "%s -R features %s 2>&1",
221 /* Using popen() instead of run_command() since debugfs does
222 * not return proper error code if command is not supported */
223 fp = popen(cmd, "r");
225 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
229 ret = fread(enabled_features, 1, sizeof(enabled_features) - 1, fp);
230 enabled_features[ret] = '\0';
233 if (strstr(enabled_features, feature))
238 /* Write the server config files */
239 int ldiskfs_write_ldd(struct mkfs_opts *mop)
241 char mntpt[] = "/tmp/mntXXXXXX";
248 /* Mount this device temporarily in order to write these files */
249 if (!mkdtemp(mntpt)) {
250 fprintf(stderr, "%s: Can't create temp mount point %s: %s\n",
251 progname, mntpt, strerror(errno));
256 * Append file context to mount options if SE Linux is enabled
259 if (is_selinux_enabled() > 0)
260 append_context_for_mount(mntpt, mop);
263 dev = mop->mo_device;
264 if (mop->mo_flags & MO_IS_LOOP)
265 dev = mop->mo_loopdev;
267 ret = mount(dev, mntpt, MT_STR(&mop->mo_ldd), 0,
268 mop->mo_ldd.ldd_mount_opts);
270 fprintf(stderr, "%s: Unable to mount %s: %s\n",
271 progname, dev, strerror(errno));
273 if (errno == ENODEV) {
274 fprintf(stderr, "Is the %s module available?\n",
275 MT_STR(&mop->mo_ldd));
280 /* Set up initial directories */
281 sprintf(filepnm, "%s/%s", mntpt, MOUNT_CONFIGS_DIR);
282 ret = mkdir(filepnm, 0777);
283 if ((ret != 0) && (errno != EEXIST)) {
284 fprintf(stderr, "%s: Can't make configs dir %s (%s)\n",
285 progname, filepnm, strerror(errno));
287 } else if (errno == EEXIST) {
291 /* Save the persistent mount data into a file. Lustre must pre-read
292 this file to get the real mount options. */
293 vprint("Writing %s\n", MOUNT_DATA_FILE);
294 sprintf(filepnm, "%s/%s", mntpt, MOUNT_DATA_FILE);
295 filep = fopen(filepnm, "w");
297 fprintf(stderr, "%s: Unable to create %s file: %s\n",
298 progname, filepnm, strerror(errno));
301 num = fwrite(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
302 if (num < 1 && ferror(filep)) {
303 fprintf(stderr, "%s: Unable to write to file (%s): %s\n",
304 progname, filepnm, strerror(errno));
317 static int readcmd(char *cmd, char *buf, int len)
322 fp = popen(cmd, "r");
326 red = fread(buf, 1, len, fp);
329 /* strip trailing newline */
330 if (buf[red - 1] == '\n')
333 return (red == 0) ? -ENOENT : 0;
336 int ldiskfs_read_ldd(char *dev, struct lustre_disk_data *mo_ldd)
338 char tmpdir[] = "/tmp/dirXXXXXX";
343 int cmdsz = sizeof(cmd);
345 /* Make a temporary directory to hold Lustre data files. */
346 if (!mkdtemp(tmpdir)) {
347 fprintf(stderr, "%s: Can't create temporary directory %s: %s\n",
348 progname, tmpdir, strerror(errno));
352 /* TODO: it's worth observing the get_mountdata() function that is
353 in mount_utils.c for getting the mountdata out of the
356 /* Construct debugfs command line. */
357 snprintf(cmd, cmdsz, "%s -c -R 'dump /%s %s/mountdata' '%s'",
358 DEBUGFS, MOUNT_DATA_FILE, tmpdir, dev);
360 ret = run_command(cmd, cmdsz);
362 verrprint("%s: Unable to dump %s dir (%d)\n",
363 progname, MOUNT_CONFIGS_DIR, ret);
365 sprintf(filepnm, "%s/mountdata", tmpdir);
366 filep = fopen(filepnm, "r");
369 vprint("Reading %s\n", MOUNT_DATA_FILE);
370 num_read = fread(mo_ldd, sizeof(*mo_ldd), 1, filep);
371 if (num_read < 1 && ferror(filep)) {
372 fprintf(stderr, "%s: Unable to read from file %s: %s\n",
373 progname, filepnm, strerror(errno));
378 snprintf(cmd, cmdsz, "rm -rf %s", tmpdir);
379 run_command(cmd, cmdsz);
381 verrprint("Failed to read old data (%d)\n", ret);
383 /* As long as we at least have the label, we're good to go */
384 snprintf(cmd, sizeof(cmd), E2LABEL" %s", dev);
385 ret = readcmd(cmd, mo_ldd->ldd_svname, sizeof(mo_ldd->ldd_svname) - 1);
391 /* Display the need for the latest e2fsprogs to be installed. make_backfs
392 * indicates if the caller is make_lustre_backfs() or not. */
393 static void disp_old_e2fsprogs_msg(const char *feature, int make_backfs)
395 static int msg_displayed;
398 fprintf(stderr, "WARNING: %s does not support %s "
399 "feature.\n\n", E2FSPROGS, feature);
405 fprintf(stderr, "WARNING: The %s package currently installed on "
406 "your system does not support \"%s\" feature.\n",
408 #if !(HAVE_LDISKFSPROGS)
409 fprintf(stderr, "Please install the latest version of e2fsprogs from\n"
410 "https://downloads.hpdd.intel.com/public/e2fsprogs/latest/\n"
411 "to enable this feature.\n");
414 fprintf(stderr, "Feature will not be enabled until %s"
415 "is updated and '%s -O %s %%{device}' "
416 "is run.\n\n", E2FSPROGS, TUNE2FS, feature);
419 /* Check whether the file exists in the device */
420 static int file_in_dev(char *file_name, char *dev_name)
423 char debugfs_cmd[256];
424 unsigned int inode_num;
427 /* Construct debugfs command line. */
428 snprintf(debugfs_cmd, sizeof(debugfs_cmd),
429 "%s -c -R 'stat %s' '%s' 2>&1 | egrep '(Inode|unsupported)'",
430 DEBUGFS, file_name, dev_name);
432 fp = popen(debugfs_cmd, "r");
434 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
438 if (fscanf(fp, "Inode: %u", &inode_num) == 1) { /* exist */
442 i = fread(debugfs_cmd, 1, sizeof(debugfs_cmd) - 1, fp);
445 fprintf(stderr, "%s", debugfs_cmd);
446 if (strstr(debugfs_cmd, "unsupported feature")) {
447 disp_old_e2fsprogs_msg("an unknown", 0);
456 /* Check whether the device has already been used with lustre */
457 int ldiskfs_is_lustre(char *dev, unsigned *mount_type)
461 ret = file_in_dev(MOUNT_DATA_FILE, dev);
463 /* in the -1 case, 'extents' means IS a lustre target */
464 *mount_type = LDD_MT_LDISKFS;
468 ret = file_in_dev(LAST_RCVD, dev);
470 *mount_type = LDD_MT_LDISKFS;
477 /* Check if a certain feature is supported by e2fsprogs.
478 * Firstly we try to use "debugfs supported_features" command to check if
479 * the feature is supported. If this fails we try to set this feature with
480 * mke2fs to check for its support. */
481 static int is_e2fsprogs_feature_supp(const char *feature)
483 static char supp_features[4096] = "";
486 char imgname[] = "/tmp/test-img-XXXXXX";
490 if (supp_features[0] == '\0') {
491 snprintf(cmd, sizeof(cmd), "%s -c -R supported_features 2>&1",
494 /* Using popen() instead of run_command() since debugfs does
495 * not return proper error code if command is not supported */
496 fp = popen(cmd, "r");
498 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
501 ret = fread(supp_features, 1, sizeof(supp_features) - 1, fp);
502 supp_features[ret] = '\0';
505 if (ret > 0 && strstr(supp_features,
506 strncmp(feature, "-O ", 3) ? feature : feature+3))
509 if ((fd = mkstemp(imgname)) < 0)
514 snprintf(cmd, sizeof(cmd), "%s -F %s %s 100 >/dev/null 2>&1",
515 MKE2FS, feature, imgname);
516 /* run_command() displays the output of mke2fs when it fails for
517 * some feature, so use system() directly */
526 * append_unique: append @key or @key=@val pair to @buf only if @key does not
528 * @buf: buffer to hold @key or @key=@val
529 * @prefix: prefix string before @key
531 * @val: value string if it's a @key=@val pair
533 static void append_unique(char *buf, char *prefix, char *key, char *val,
542 anchor = end = strstr(buf, key);
543 /* try to find exact match string in @buf */
544 while (end && *end != '\0' && *end != ',' && *end != ' ' && *end != '=')
547 if (anchor == NULL || strlen(key) != len ||
548 strncmp(anchor, key, len) != 0) {
550 strscat(buf, prefix, maxbuflen);
552 strscat(buf, key, maxbuflen);
554 strscat(buf, "=", maxbuflen);
555 strscat(buf, val, maxbuflen);
560 static int enable_default_ext4_features(struct mkfs_opts *mop, char *anchor,
561 size_t maxbuflen, int user_spec)
563 if (IS_OST(&mop->mo_ldd)) {
564 append_unique(anchor, user_spec ? "," : " -O ",
565 "extents", NULL, sizeof(mop->mo_mkfsopts));
566 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
567 } else if (IS_MDT(&mop->mo_ldd)) {
568 append_unique(anchor, user_spec ? "," : " -O ",
569 "dirdata", NULL, maxbuflen);
570 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
571 append_unique(anchor, ",", "^extents", NULL, maxbuflen);
573 append_unique(anchor, user_spec ? "," : " -O ",
574 "uninit_bg", NULL, maxbuflen);
577 /* Multiple mount protection enabled only if failover node specified */
578 if (mop->mo_flags & MO_FAILOVER) {
579 if (is_e2fsprogs_feature_supp("-O mmp") == 0)
580 append_unique(anchor, ",", "mmp", NULL, maxbuflen);
582 disp_old_e2fsprogs_msg("mmp", 1);
585 /* Allow more than 65000 subdirectories */
586 if (is_e2fsprogs_feature_supp("-O dir_nlink") == 0)
587 append_unique(anchor, ",", "dir_nlink", NULL, maxbuflen);
589 /* The following options are only valid for ext4-based ldiskfs.
590 * If --backfstype=ext3 is specified, do not enable them. */
591 if (mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3)
594 /* Enable quota by default */
595 if (is_e2fsprogs_feature_supp("-O quota") == 0) {
596 append_unique(anchor, ",", "quota", NULL, maxbuflen);
599 fprintf(stderr, "\"-O quota\" must be supported by "
600 "e2fsprogs, please upgrade your e2fsprogs.\n");
604 /* Allow files larger than 2TB. Also needs LU-16, but not harmful. */
605 if (is_e2fsprogs_feature_supp("-O huge_file") == 0)
606 append_unique(anchor, ",", "huge_file", NULL, maxbuflen);
608 /* Enable large block addresses if the LUN is over 2^32 blocks. */
609 if (mop->mo_device_kb / (L_BLOCK_SIZE >> 10) >= 0x100002000ULL &&
610 is_e2fsprogs_feature_supp("-O 64bit") == 0)
611 append_unique(anchor, ",", "64bit", NULL, maxbuflen);
613 /* Cluster inode/block bitmaps and inode table for more efficient IO.
614 * Align the flex groups on a 1MB boundary for better performance. */
615 /* This -O feature needs to go last, since it adds the "-G" option. */
616 if (is_e2fsprogs_feature_supp("-O flex_bg") == 0) {
619 append_unique(anchor, ",", "flex_bg", NULL, maxbuflen);
621 if (IS_OST(&mop->mo_ldd)) {
622 snprintf(tmp_buf, sizeof(tmp_buf), " -G %u",
623 (1 << 20) / L_BLOCK_SIZE);
624 strscat(anchor, tmp_buf, maxbuflen);
627 /* Don't add any more "-O" options here, see last comment above */
632 * moveopts_to_end: find the option string, move remaining strings to
633 * where option string starts, and append the option
635 * @start: where the option string starts before the move
636 * RETURN: where the option string starts after the move
638 static char *moveopts_to_end(char *start)
644 /* skip whitespace before options */
649 /* find end of option characters */
650 while (*end != ' ' && *end != '\0')
654 if (len >= sizeof(save))
655 len = sizeof(save) - 1;
658 strncpy(save, start, len);
661 /* move remaining options up front */
663 memmove(start, end, strlen(end));
664 *(start + strlen(end)) = '\0';
666 /* append the specified options */
667 if (*(start + strlen(start) - 1) != ' ')
669 idx = start + strlen(start);
675 /* Build fs according to type */
676 int ldiskfs_make_lustre(struct mkfs_opts *mop)
678 __u64 device_kb = mop->mo_device_kb, block_count = 0;
679 char mkfs_cmd[PATH_MAX];
683 int ret = 0, ext_opts = 0;
686 if (!(mop->mo_flags & MO_IS_LOOP)) {
687 mop->mo_device_kb = get_device_size(mop->mo_device);
689 if (mop->mo_device_kb == 0)
692 /* Compare to real size */
693 if (device_kb == 0 || device_kb > mop->mo_device_kb)
694 device_kb = mop->mo_device_kb;
696 mop->mo_device_kb = device_kb;
699 if (mop->mo_device_kb != 0) {
700 if (mop->mo_device_kb < 8096) {
701 fprintf(stderr, "%s: size of filesystem must be larger "
702 "than 8MB, but is set to %lldKB\n",
703 progname, (long long)mop->mo_device_kb);
706 block_count = mop->mo_device_kb / (L_BLOCK_SIZE >> 10);
707 /* If the LUN size is just over 2^32 blocks, limit the
708 * filesystem size to 2^32-1 blocks to avoid problems with
709 * ldiskfs/mkfs not handling this size. Bug 22906 */
710 if (block_count > 0xffffffffULL && block_count < 0x100002000ULL)
711 block_count = 0xffffffffULL;
714 if ((mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3) ||
715 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS) ||
716 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS2)) {
719 /* Journal size in MB */
720 if (strstr(mop->mo_mkfsopts, "-J") == NULL &&
721 device_kb > 1024 * 1024) {
722 /* Choose our own default journal size */
723 long journal_mb = 0, max_mb;
725 /* cap journal size at 4GB for MDT,
726 * leave it at 400MB for OSTs. */
727 if (IS_MDT(&mop->mo_ldd))
729 else if (IS_OST(&mop->mo_ldd))
731 else /* Use mke2fs default size for MGS */
734 /* Use at most 4% of device for journal */
735 journal_mb = device_kb * 4 / (1024 * 100);
736 if (journal_mb > max_mb)
740 sprintf(buf, " -J size=%ld", journal_mb);
741 strscat(mop->mo_mkfsopts, buf,
742 sizeof(mop->mo_mkfsopts));
746 /* Inode size (for extended attributes). The LOV EA size is
747 * 32 (EA hdr) + 32 (lov_mds_md) + stripes * 24 (lov_ost_data),
748 * and we want some margin above that for ACLs, other EAs... */
749 if (strstr(mop->mo_mkfsopts, "-I") == NULL) {
750 if (IS_MDT(&mop->mo_ldd)) {
751 if (mop->mo_stripe_count > 72)
752 inode_size = 512; /* bz 7241 */
753 /* see also "-i" below for EA blocks */
754 else if (mop->mo_stripe_count > 32)
756 else if (mop->mo_stripe_count > 10)
760 } else if (IS_OST(&mop->mo_ldd)) {
761 /* We store MDS FID and OST objid in EA on OST
762 * we need to make inode bigger as well. */
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 > 72) {
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 (IS_OST(&mop->mo_ldd) && mop->mo_device_kb > 100 * 1024 &&
871 mop->mo_device_kb * 1024 / L_BLOCK_SIZE <= 0xffffffffULL) {
872 unsigned group_blocks = L_BLOCK_SIZE * 8;
873 unsigned desc_per_block = L_BLOCK_SIZE / 32;
874 unsigned resize_blks;
876 resize_blks = (1ULL<<32) - desc_per_block*group_blocks;
877 snprintf(buf, sizeof(buf), "%u", resize_blks);
878 append_unique(start, ext_opts ? "," : " -E ",
879 "resize", buf, maxbuflen);
883 /* Avoid zeroing out the full journal - speeds up mkfs */
884 if (is_e2fsprogs_feature_supp("-E lazy_journal_init") == 0)
885 append_unique(start, ext_opts ? "," : " -E ",
886 "lazy_journal_init", NULL, maxbuflen);
887 /* end handle -E mkfs options */
889 /* Allow reformat of full devices (as opposed to
890 partitions.) We already checked for mounted dev. */
891 strscat(mop->mo_mkfsopts, " -F", sizeof(mop->mo_mkfsopts));
893 snprintf(mkfs_cmd, sizeof(mkfs_cmd),
894 "%s -j -b %d -L %s ", MKE2FS, L_BLOCK_SIZE,
895 mop->mo_ldd.ldd_svname);
897 fprintf(stderr,"%s: unsupported fs type: %d (%s)\n",
898 progname, mop->mo_ldd.ldd_mount_type,
899 MT_STR(&mop->mo_ldd));
903 /* For loop device format the dev, not the filename */
904 dev = mop->mo_device;
905 if (mop->mo_flags & MO_IS_LOOP)
906 dev = mop->mo_loopdev;
908 vprint("formatting backing filesystem %s on %s\n",
909 MT_STR(&mop->mo_ldd), dev);
910 vprint("\ttarget name %s\n", mop->mo_ldd.ldd_svname);
911 vprint("\t4k blocks "LPU64"\n", block_count);
912 vprint("\toptions %s\n", mop->mo_mkfsopts);
914 /* mkfs_cmd's trailing space is important! */
915 strscat(mkfs_cmd, mop->mo_mkfsopts, sizeof(mkfs_cmd));
916 strscat(mkfs_cmd, " ", sizeof(mkfs_cmd));
917 strscat(mkfs_cmd, dev, sizeof(mkfs_cmd));
918 if (block_count != 0) {
919 sprintf(buf, " "LPU64, block_count);
920 strscat(mkfs_cmd, buf, sizeof(mkfs_cmd));
923 vprint("mkfs_cmd = %s\n", mkfs_cmd);
924 ret = run_command(mkfs_cmd, sizeof(mkfs_cmd));
927 fprintf(stderr, "Unable to build fs %s (%d)\n", dev, ret);
932 int ldiskfs_prepare_lustre(struct mkfs_opts *mop,
933 char *default_mountopts, int default_len,
934 char *always_mountopts, int always_len)
936 struct lustre_disk_data *ldd = &mop->mo_ldd;
939 /* Set MO_IS_LOOP to indicate a loopback device is needed */
940 ret = is_block(mop->mo_device);
943 } else if (ret == 0) {
944 mop->mo_flags |= MO_IS_LOOP;
947 strscat(default_mountopts, ",errors=remount-ro", default_len);
948 if (IS_MDT(ldd) || IS_MGS(ldd))
949 strscat(always_mountopts, ",user_xattr", always_len);
954 static int read_file(const char *path, char *buf, int size)
958 fd = fopen(path, "r");
962 if (fgets(buf, size, fd) == NULL) {
963 fprintf(stderr, "reading from %s: %s", path, strerror(errno));
969 /* strip trailing newline */
971 if (buf[size - 1] == '\n')
972 buf[size - 1] = '\0';
977 static int write_file(const char *path, const char *buf)
981 fd = fopen(path, "w");
990 static int set_blockdev_scheduler(const char *path, const char *scheduler)
992 char buf[PATH_MAX], *s, *e, orig_sched[50];
995 /* Before setting the scheduler, we need to check to see if it's
996 * already set to "noop". If it is, we don't want to override
997 * that setting. If it's set to anything other than "noop", set
998 * the scheduler to what has been passed in. */
1000 rc = read_file(path, buf, sizeof(buf));
1003 fprintf(stderr, "%s: cannot open '%s': %s\n",
1004 progname, path, strerror(errno));
1008 /* The expected format of buf: noop anticipatory deadline [cfq] */
1009 s = strchr(buf, '[');
1010 e = strchr(buf, ']');
1012 /* If the format is not what we expect. Play it safe and error out. */
1013 if (s == NULL || e == NULL) {
1015 fprintf(stderr, "%s: cannot parse scheduler "
1016 "options for '%s'\n", progname, path);
1020 snprintf(orig_sched, e - s, "%s", s + 1);
1022 if (strcmp(orig_sched, "noop") == 0 ||
1023 strcmp(orig_sched, scheduler) == 0)
1026 rc = write_file(path, scheduler);
1029 fprintf(stderr, "%s: cannot set scheduler on "
1030 "'%s': %s\n", progname, path,
1034 fprintf(stderr, "%s: change scheduler of %s from %s to %s\n",
1035 progname, path, orig_sched, scheduler);
1041 /* This is to tune the kernel for good SCSI performance.
1042 * For that we set the value of /sys/block/{dev}/queue/max_sectors_kb
1043 * to the value of /sys/block/{dev}/queue/max_hw_sectors_kb */
1044 static int set_blockdev_tunables(char *source, struct mount_opts *mop)
1046 glob_t glob_info = { 0 };
1047 struct stat stat_buf;
1048 char *chk_major, *chk_minor;
1049 char *savept = NULL, *dev;
1051 char buf[PATH_MAX] = {'\0'}, path[PATH_MAX] = {'\0'};
1052 char real_path[PATH_MAX] = {'\0'};
1060 ret_path = realpath(source, real_path);
1061 if (ret_path == NULL) {
1063 fprintf(stderr, "warning: %s: cannot resolve: %s\n",
1064 source, strerror(errno));
1068 if (strncmp(real_path, "/dev/loop", 9) == 0)
1071 if ((real_path[0] != '/') && (strpbrk(real_path, ",:") != NULL))
1074 snprintf(path, sizeof(path), "/sys/block%s", real_path + 4);
1075 if (access(path, X_OK) == 0)
1078 /* The name of the device say 'X' specified in /dev/X may not
1079 * match any entry under /sys/block/. In that case we need to
1080 * match the major/minor number to find the entry under
1081 * sys/block corresponding to /dev/X */
1083 /* Don't chop tail digit on /dev/mapper/xxx, LU-478 */
1084 if (strncmp(real_path, "/dev/mapper", 11) != 0) {
1085 dev = real_path + strlen(real_path);
1086 while (--dev > real_path && isdigit(*dev))
1089 if (strncmp(real_path, "/dev/md_", 8) == 0)
1093 rc = stat(real_path, &stat_buf);
1096 fprintf(stderr, "warning: %s, device %s stat failed\n",
1097 strerror(errno), real_path);
1101 major = major(stat_buf.st_rdev);
1102 minor = minor(stat_buf.st_rdev);
1103 rc = glob("/sys/block/*", GLOB_NOSORT, NULL, &glob_info);
1106 fprintf(stderr, "warning: failed to read entries under "
1108 globfree(&glob_info);
1112 for (i = 0; i < glob_info.gl_pathc; i++){
1113 snprintf(path, sizeof(path), "%s/dev", glob_info.gl_pathv[i]);
1115 rc = read_file(path, buf, sizeof(buf));
1119 if (buf[strlen(buf) - 1] == '\n')
1120 buf[strlen(buf) - 1] = '\0';
1122 chk_major = strtok_r(buf, ":", &savept);
1124 if (chk_major != NULL && major == atoi(chk_major) &&
1125 chk_minor != NULL && minor == atoi(chk_minor))
1129 if (i == glob_info.gl_pathc) {
1131 fprintf(stderr,"warning: device %s does not match any "
1132 "entry under /sys/block\n", real_path);
1133 globfree(&glob_info);
1137 /* Chop off "/dev" from path we found */
1138 path[strlen(glob_info.gl_pathv[i])] = '\0';
1139 globfree(&glob_info);
1142 if (strncmp(real_path, "/dev/md", 7) == 0) {
1143 snprintf(real_path, sizeof(real_path), "%s/%s", path,
1146 rc = read_file(real_path, buf, sizeof(buf));
1149 fprintf(stderr, "warning: opening %s: %s\n",
1150 real_path, strerror(errno));
1154 if (atoi(buf) >= mop->mo_md_stripe_cache_size)
1157 if (strlen(buf) - 1 > 0) {
1158 snprintf(buf, sizeof(buf), "%d",
1159 mop->mo_md_stripe_cache_size);
1160 rc = write_file(real_path, buf);
1162 fprintf(stderr, "warning: opening %s: %s\n",
1163 real_path, strerror(errno));
1165 /* Return since raid and disk tunables are different */
1169 snprintf(real_path, sizeof(real_path), "%s/%s", path,
1170 MAX_HW_SECTORS_KB_PATH);
1171 rc = read_file(real_path, buf, sizeof(buf));
1174 fprintf(stderr, "warning: opening %s: %s\n",
1175 real_path, strerror(errno));
1176 /* No MAX_HW_SECTORS_KB_PATH isn't necessary an
1177 * error for some device. */
1181 if (strlen(buf) - 1 > 0) {
1182 char oldbuf[32] = "", *end = NULL;
1183 unsigned long long oldval, newval;
1185 snprintf(real_path, sizeof(real_path), "%s/%s", path,
1186 MAX_SECTORS_KB_PATH);
1187 rc = read_file(real_path, oldbuf, sizeof(oldbuf));
1188 /* Only set new parameter if different from the old one. */
1189 if (rc != 0 || strcmp(oldbuf, buf) == 0) {
1190 /* No MAX_SECTORS_KB_PATH isn't necessary an
1191 * error for some device. */
1195 newval = strtoull(buf, &end, 0);
1196 if (newval == 0 || newval == ULLONG_MAX || end == buf)
1199 /* Don't increase IO request size limit past 32MB. It is about
1200 * 2x PTLRPC_MAX_BRW_SIZE, but that isn't defined publicly. */
1201 if (newval > 32 * 1024)
1204 oldval = strtoull(oldbuf, &end, 0);
1205 /* Don't shrink the current limit. */
1206 if (oldval != ULLONG_MAX && newval <= oldval)
1209 rc = write_file(real_path, buf);
1212 fprintf(stderr, "warning: writing to %s: %s\n",
1213 real_path, strerror(errno));
1214 /* No MAX_SECTORS_KB_PATH isn't necessary an
1215 * error for some device. */
1218 fprintf(stderr, "%s: increased %s from %s to %s\n",
1219 progname, real_path, oldbuf, buf);
1223 /* Purposely ignore errors reported from set_blockdev_scheduler.
1224 * The worst that will happen is a block device with an "incorrect"
1226 snprintf(real_path, sizeof(real_path), "%s/%s", path, SCHEDULER_PATH);
1227 set_blockdev_scheduler(real_path, DEFAULT_SCHEDULER);
1229 /* if device is multipath device, tune its slave devices */
1230 glob_info.gl_pathc = 0;
1231 glob_info.gl_offs = 0;
1232 snprintf(real_path, sizeof(real_path), "%s/slaves/*", path);
1233 rc = glob(real_path, GLOB_NOSORT, NULL, &glob_info);
1235 for (i = 0; rc == 0 && i < glob_info.gl_pathc; i++) {
1236 slave = basename(glob_info.gl_pathv[i]);
1237 snprintf(real_path, sizeof(real_path), "/dev/%s", slave);
1238 rc = set_blockdev_tunables(real_path, mop);
1241 if (rc == GLOB_NOMATCH) {
1242 /* no slave device is not an error */
1244 } else if (rc && verbose) {
1245 if (slave == NULL) {
1246 fprintf(stderr, "warning: %s, failed to read"
1247 " entries under %s/slaves\n",
1248 strerror(errno), path);
1250 fprintf(stderr, "unable to set tunables for"
1251 " slave device %s (slave would be"
1252 " unable to handle IO request from"
1257 globfree(&glob_info);
1262 int ldiskfs_tune_lustre(char *dev, struct mount_opts *mop)
1264 return set_blockdev_tunables(dev, mop);
1267 int ldiskfs_label_lustre(struct mount_opts *mop)
1269 char label_cmd[PATH_MAX];
1272 snprintf(label_cmd, sizeof(label_cmd),
1273 TUNE2FS" -f -L '%s' '%s' >/dev/null 2>&1",
1274 mop->mo_ldd.ldd_svname, mop->mo_source);
1275 rc = run_command(label_cmd, sizeof(label_cmd));
1280 /* Enable quota accounting */
1281 int ldiskfs_enable_quota(struct mkfs_opts *mop)
1285 int cmdsz = sizeof(cmd), ret;
1287 if (is_e2fsprogs_feature_supp("-O quota") != 0) {
1288 fprintf(stderr, "%s: \"-O quota\" is is not supported by "
1289 "current e2fsprogs\n", progname);
1293 dev = mop->mo_device;
1294 if (mop->mo_flags & MO_IS_LOOP)
1295 dev = mop->mo_loopdev;
1297 /* Quota feature is already enabled? */
1298 if (is_feature_enabled("quota", dev)) {
1299 vprint("Quota feature is already enabled.\n");
1303 /* Turn on quota feature by "tune2fs -O quota" */
1304 snprintf(cmd, cmdsz, "%s -O quota %s", TUNE2FS, dev);
1305 ret = run_command(cmd, cmdsz);
1307 fprintf(stderr, "command:%s (%d)", cmd, ret);
1312 int ldiskfs_init(void)
1314 /* Required because full path to DEBUGFS is not specified */
1315 setenv("PATH", "/bin:/sbin:/usr/bin:/usr/sbin:/usr/local/bin", 0);
1320 void ldiskfs_fini(void)