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, 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>
70 /* libcfs.h is not really needed here, but on SLES10/PPC, fs.h includes idr.h
71 * which requires BITS_PER_LONG to be defined */
72 #include <libcfs/libcfs.h>
74 #include <linux/fs.h> /* for BLKGETSIZE64 */
76 #include <linux/version.h>
78 #include <lustre_disk.h>
79 #include <lustre_param.h>
80 #include <lnet/lnetctl.h>
81 #include <lustre_ver.h>
83 #define MAX_HW_SECTORS_KB_PATH "queue/max_hw_sectors_kb"
84 #define MAX_SECTORS_KB_PATH "queue/max_sectors_kb"
85 #define STRIPE_CACHE_SIZE "md/stripe_cache_size"
87 extern char *progname;
89 #define L_BLOCK_SIZE 4096
90 /* keep it less than LL_FID_NAMELEN */
91 #define DUMMY_FILE_NAME_LEN 25
92 #define EXT3_DIRENT_SIZE DUMMY_FILE_NAME_LEN
94 /* Write the server config files */
95 int ldiskfs_write_ldd(struct mkfs_opts *mop)
97 char mntpt[] = "/tmp/mntXXXXXX";
104 /* Mount this device temporarily in order to write these files */
105 if (!mkdtemp(mntpt)) {
106 fprintf(stderr, "%s: Can't create temp mount point %s: %s\n",
107 progname, mntpt, strerror(errno));
111 dev = mop->mo_device;
112 if (mop->mo_flags & MO_IS_LOOP)
113 dev = mop->mo_loopdev;
115 ret = mount(dev, mntpt, MT_STR(&mop->mo_ldd), 0,
116 mop->mo_ldd.ldd_mount_opts);
118 fprintf(stderr, "%s: Unable to mount %s: %s\n",
119 progname, dev, strerror(errno));
121 if (errno == ENODEV) {
122 fprintf(stderr, "Is the %s module available?\n",
123 MT_STR(&mop->mo_ldd));
128 /* Set up initial directories */
129 sprintf(filepnm, "%s/%s", mntpt, MOUNT_CONFIGS_DIR);
130 ret = mkdir(filepnm, 0777);
131 if ((ret != 0) && (errno != EEXIST)) {
132 fprintf(stderr, "%s: Can't make configs dir %s (%s)\n",
133 progname, filepnm, strerror(errno));
135 } else if (errno == EEXIST) {
139 /* Save the persistent mount data into a file. Lustre must pre-read
140 this file to get the real mount options. */
141 vprint("Writing %s\n", MOUNT_DATA_FILE);
142 sprintf(filepnm, "%s/%s", mntpt, MOUNT_DATA_FILE);
143 filep = fopen(filepnm, "w");
145 fprintf(stderr, "%s: Unable to create %s file: %s\n",
146 progname, filepnm, strerror(errno));
149 num = fwrite(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
150 if (num < 1 && ferror(filep)) {
151 fprintf(stderr, "%s: Unable to write to file (%s): %s\n",
152 progname, filepnm, strerror(errno));
165 static int readcmd(char *cmd, char *buf, int len)
170 fp = popen(cmd, "r");
174 red = fread(buf, 1, len, fp);
177 /* strip trailing newline */
178 if (buf[red - 1] == '\n')
181 return (red == 0) ? -ENOENT : 0;
184 int ldiskfs_read_ldd(char *dev, struct lustre_disk_data *mo_ldd)
186 char tmpdir[] = "/tmp/dirXXXXXX";
191 int cmdsz = sizeof(cmd);
193 /* Make a temporary directory to hold Lustre data files. */
194 if (!mkdtemp(tmpdir)) {
195 fprintf(stderr, "%s: Can't create temporary directory %s: %s\n",
196 progname, tmpdir, strerror(errno));
200 /* TODO: it's worth observing the get_mountdata() function that is
201 in mount_utils.c for getting the mountdata out of the
204 /* Construct debugfs command line. */
205 snprintf(cmd, cmdsz, "%s -c -R 'dump /%s %s/mountdata' '%s'",
206 DEBUGFS, MOUNT_DATA_FILE, tmpdir, dev);
208 ret = run_command(cmd, cmdsz);
210 verrprint("%s: Unable to dump %s dir (%d)\n",
211 progname, MOUNT_CONFIGS_DIR, ret);
213 sprintf(filepnm, "%s/mountdata", tmpdir);
214 filep = fopen(filepnm, "r");
217 vprint("Reading %s\n", MOUNT_DATA_FILE);
218 num_read = fread(mo_ldd, sizeof(*mo_ldd), 1, filep);
219 if (num_read < 1 && ferror(filep)) {
220 fprintf(stderr, "%s: Unable to read from file %s: %s\n",
221 progname, filepnm, strerror(errno));
226 snprintf(cmd, cmdsz, "rm -rf %s", tmpdir);
227 run_command(cmd, cmdsz);
229 verrprint("Failed to read old data (%d)\n", ret);
231 /* As long as we at least have the label, we're good to go */
232 snprintf(cmd, sizeof(cmd), E2LABEL" %s", dev);
233 ret = readcmd(cmd, mo_ldd->ldd_svname, sizeof(mo_ldd->ldd_svname) - 1);
239 /* Display the need for the latest e2fsprogs to be installed. make_backfs
240 * indicates if the caller is make_lustre_backfs() or not. */
241 void disp_old_e2fsprogs_msg(const char *feature, int make_backfs)
243 static int msg_displayed;
246 fprintf(stderr, "WARNING: %s does not support %s "
247 "feature.\n\n", E2FSPROGS, feature);
253 fprintf(stderr, "WARNING: The %s package currently installed on "
254 "your system does not support \"%s\" feature.\n",
256 #if !(HAVE_LDISKFSPROGS)
257 fprintf(stderr, "Please install the latest version of e2fsprogs from\n"
258 "http://downloads.whamcloud.com/public/e2fsprogs/latest/\n"
259 "to enable this feature.\n");
262 fprintf(stderr, "Feature will not be enabled until %s"
263 "is updated and '%s -O %s %%{device}' "
264 "is run.\n\n", E2FSPROGS, TUNE2FS, feature);
267 /* Check whether the file exists in the device */
268 static int file_in_dev(char *file_name, char *dev_name)
271 char debugfs_cmd[256];
272 unsigned int inode_num;
275 /* Construct debugfs command line. */
276 snprintf(debugfs_cmd, sizeof(debugfs_cmd),
277 "%s -c -R 'stat %s' '%s' 2>&1 | egrep '(Inode|unsupported)'",
278 DEBUGFS, file_name, dev_name);
280 fp = popen(debugfs_cmd, "r");
282 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
286 if (fscanf(fp, "Inode: %u", &inode_num) == 1) { /* exist */
290 i = fread(debugfs_cmd, 1, sizeof(debugfs_cmd) - 1, fp);
293 fprintf(stderr, "%s", debugfs_cmd);
294 if (strstr(debugfs_cmd, "unsupported feature")) {
295 disp_old_e2fsprogs_msg("an unknown", 0);
304 /* Check whether the device has already been used with lustre */
305 int ldiskfs_is_lustre(char *dev, unsigned *mount_type)
309 ret = file_in_dev(MOUNT_DATA_FILE, dev);
311 /* in the -1 case, 'extents' means IS a lustre target */
312 *mount_type = LDD_MT_LDISKFS;
316 ret = file_in_dev(LAST_RCVD, dev);
318 *mount_type = LDD_MT_LDISKFS;
325 /* Check if a certain feature is supported by e2fsprogs.
326 * Firstly we try to use "debugfs supported_features" command to check if
327 * the feature is supported. If this fails we try to set this feature with
328 * mke2fs to check for its support. */
329 static int is_e2fsprogs_feature_supp(const char *feature)
331 static char supp_features[4096] = "";
334 char imgname[] = "/tmp/test-img-XXXXXX";
338 if (supp_features[0] == '\0') {
339 snprintf(cmd, sizeof(cmd), "%s -c -R supported_features 2>&1",
342 /* Using popen() instead of run_command() since debugfs does
343 * not return proper error code if command is not supported */
344 fp = popen(cmd, "r");
346 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
349 ret = fread(supp_features, 1, sizeof(supp_features) - 1, fp);
350 supp_features[ret] = '\0';
353 if (ret > 0 && strstr(supp_features,
354 strncmp(feature, "-O ", 3) ? feature : feature+3))
357 if ((fd = mkstemp(imgname)) < 0)
362 snprintf(cmd, sizeof(cmd), "%s -F %s %s 100 >/dev/null 2>&1",
363 MKE2FS, feature, imgname);
364 /* run_command() displays the output of mke2fs when it fails for
365 * some feature, so use system() directly */
374 * append_unique: append @key or @key=@val pair to @buf only if @key does not
376 * @buf: buffer to hold @key or @key=@val
377 * @prefix: prefix string before @key
379 * @val: value string if it's a @key=@val pair
381 static void append_unique(char *buf, char *prefix, char *key, char *val,
390 anchor = end = strstr(buf, key);
391 /* try to find exact match string in @buf */
392 while (end && *end != '\0' && *end != ',' && *end != ' ' && *end != '=')
395 if (anchor == NULL || strlen(key) != len ||
396 strncmp(anchor, key, len) != 0) {
398 strscat(buf, prefix, maxbuflen);
400 strscat(buf, key, maxbuflen);
402 strscat(buf, "=", maxbuflen);
403 strscat(buf, val, maxbuflen);
408 static int enable_default_ext4_features(struct mkfs_opts *mop, char *anchor,
409 size_t maxbuflen, int user_spec)
411 if (IS_OST(&mop->mo_ldd)) {
412 append_unique(anchor, user_spec ? "," : " -O ",
413 "extents", NULL, sizeof(mop->mo_mkfsopts));
414 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
415 } else if (IS_MDT(&mop->mo_ldd)) {
416 append_unique(anchor, user_spec ? "," : " -O ",
417 "dirdata", NULL, maxbuflen);
418 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
419 append_unique(anchor, ",", "^extents", NULL, maxbuflen);
421 append_unique(anchor, user_spec ? "," : " -O ",
422 "uninit_bg", NULL, maxbuflen);
425 /* Multiple mount protection enabled only if failover node specified */
426 if (mop->mo_flags & MO_FAILOVER) {
427 if (is_e2fsprogs_feature_supp("-O mmp") == 0)
428 append_unique(anchor, ",", "mmp", NULL, maxbuflen);
430 disp_old_e2fsprogs_msg("mmp", 1);
433 /* Allow more than 65000 subdirectories */
434 if (is_e2fsprogs_feature_supp("-O dir_nlink") == 0)
435 append_unique(anchor, ",", "dir_nlink", NULL, maxbuflen);
437 /* The following options are only valid for ext4-based ldiskfs.
438 * If --backfstype=ext3 is specified, do not enable them. */
439 if (mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3)
442 /* Enable quota by default */
443 if (is_e2fsprogs_feature_supp("-O quota") == 0) {
444 append_unique(anchor, ",", "quota", NULL, maxbuflen);
447 fprintf(stderr, "\"-O quota\" must be supported by "
448 "e2fsprogs, please upgrade your e2fsprogs.\n");
452 /* Allow files larger than 2TB. Also needs LU-16, but not harmful. */
453 if (is_e2fsprogs_feature_supp("-O huge_file") == 0)
454 append_unique(anchor, ",", "huge_file", NULL, maxbuflen);
456 /* Enable large block addresses if the LUN is over 2^32 blocks. */
457 if (mop->mo_device_sz / (L_BLOCK_SIZE >> 10) >= 0x100002000ULL &&
458 is_e2fsprogs_feature_supp("-O 64bit") == 0)
459 append_unique(anchor, ",", "64bit", NULL, maxbuflen);
461 /* Cluster inode/block bitmaps and inode table for more efficient IO.
462 * Align the flex groups on a 1MB boundary for better performance. */
463 /* This -O feature needs to go last, since it adds the "-G" option. */
464 if (is_e2fsprogs_feature_supp("-O flex_bg") == 0) {
467 append_unique(anchor, ",", "flex_bg", NULL, maxbuflen);
469 if (IS_OST(&mop->mo_ldd)) {
470 snprintf(tmp_buf, sizeof(tmp_buf), " -G %u",
471 (1 << 20) / L_BLOCK_SIZE);
472 strscat(anchor, tmp_buf, maxbuflen);
475 /* Don't add any more "-O" options here, see last comment above */
480 * moveopts_to_end: find the option string, move remaining strings to
481 * where option string starts, and append the option
483 * @start: where the option string starts before the move
484 * RETURN: where the option string starts after the move
486 static char *moveopts_to_end(char *start)
491 /* skip whitespace before options */
496 /* find end of option characters */
497 while (*end != ' ' && *end != '\0')
501 strncpy(save, start, end - start);
502 save[end - start] = '\0';
504 /* move remaining options up front */
506 memmove(start, end, strlen(end));
507 *(start + strlen(end)) = '\0';
509 /* append the specified options */
510 if (*(start + strlen(start) - 1) != ' ')
512 idx = start + strlen(start);
518 /* Build fs according to type */
519 int ldiskfs_make_lustre(struct mkfs_opts *mop)
521 __u64 device_sz = mop->mo_device_sz, block_count = 0;
522 char mkfs_cmd[PATH_MAX];
526 int ret = 0, ext_opts = 0;
529 if (!(mop->mo_flags & MO_IS_LOOP)) {
530 mop->mo_device_sz = get_device_size(mop->mo_device);
532 if (mop->mo_device_sz == 0)
535 /* Compare to real size */
536 if (device_sz == 0 || device_sz > mop->mo_device_sz)
537 device_sz = mop->mo_device_sz;
539 mop->mo_device_sz = device_sz;
542 if (mop->mo_device_sz != 0) {
543 if (mop->mo_device_sz < 8096){
544 fprintf(stderr, "%s: size of filesystem must be larger "
545 "than 8MB, but is set to %lldKB\n",
546 progname, (long long)mop->mo_device_sz);
549 block_count = mop->mo_device_sz / (L_BLOCK_SIZE >> 10);
550 /* If the LUN size is just over 2^32 blocks, limit the
551 * filesystem size to 2^32-1 blocks to avoid problems with
552 * ldiskfs/mkfs not handling this size. Bug 22906 */
553 if (block_count > 0xffffffffULL && block_count < 0x100002000ULL)
554 block_count = 0xffffffffULL;
557 if ((mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3) ||
558 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS) ||
559 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS2)) {
562 /* Journal size in MB */
563 if (strstr(mop->mo_mkfsopts, "-J") == NULL) {
564 /* Choose our own default journal size */
565 long journal_sz = 0, max_sz;
566 if (device_sz > 1024 * 1024) /* 1GB */
567 journal_sz = (device_sz / 102400) * 4;
568 /* cap journal size at 1GB */
569 if (journal_sz > 1024L)
572 max_sz = (102400 * L_BLOCK_SIZE) >> 20; /* 400MB */
573 if (journal_sz > max_sz)
576 sprintf(buf, " -J size=%ld", journal_sz);
577 strscat(mop->mo_mkfsopts, buf,
578 sizeof(mop->mo_mkfsopts));
582 /* Inode size (for extended attributes). The LOV EA size is
583 * 32 (EA hdr) + 32 (lov_mds_md) + stripes * 24 (lov_ost_data),
584 * and we want some margin above that for ACLs, other EAs... */
585 if (strstr(mop->mo_mkfsopts, "-I") == NULL) {
586 if (IS_MDT(&mop->mo_ldd)) {
587 if (mop->mo_stripe_count > 72)
588 inode_size = 512; /* bz 7241 */
589 /* see also "-i" below for EA blocks */
590 else if (mop->mo_stripe_count > 32)
592 else if (mop->mo_stripe_count > 10)
596 } else if (IS_OST(&mop->mo_ldd)) {
597 /* We store MDS FID and OST objid in EA on OST
598 * we need to make inode bigger as well. */
602 if (inode_size > 0) {
603 sprintf(buf, " -I %ld", inode_size);
604 strscat(mop->mo_mkfsopts, buf,
605 sizeof(mop->mo_mkfsopts));
609 /* Bytes_per_inode: disk size / num inodes */
610 if (strstr(mop->mo_mkfsopts, "-i") == NULL &&
611 strstr(mop->mo_mkfsopts, "-N") == NULL) {
612 long bytes_per_inode = 0;
614 /* Allocate more inodes on MDT devices. There is
615 * no data stored on the MDT, and very little extra
616 * metadata beyond the inode. It could go down as
617 * low as 1024 bytes, but this is conservative.
618 * Account for external EA blocks for wide striping. */
619 if (IS_MDT(&mop->mo_ldd)) {
620 bytes_per_inode = inode_size + 1536;
622 if (mop->mo_stripe_count > 72) {
623 int extra = mop->mo_stripe_count * 24;
624 extra = ((extra - 1) | 4095) + 1;
625 bytes_per_inode += extra;
629 /* Allocate fewer inodes on large OST devices. Most
630 * filesystems can be much more aggressive than even
631 * this, but it is impossible to know in advance. */
632 if (IS_OST(&mop->mo_ldd)) {
633 /* OST > 16TB assume average file size 1MB */
634 if (device_sz > (16ULL << 30))
635 bytes_per_inode = 1024 * 1024;
636 /* OST > 4TB assume average file size 512kB */
637 else if (device_sz > (4ULL << 30))
638 bytes_per_inode = 512 * 1024;
639 /* OST > 1TB assume average file size 256kB */
640 else if (device_sz > (1ULL << 30))
641 bytes_per_inode = 256 * 1024;
642 /* OST > 10GB assume average file size 64kB,
643 * plus a bit so that inodes will fit into a
644 * 256x flex_bg without overflowing */
645 else if (device_sz > (10ULL << 20))
646 bytes_per_inode = 69905;
649 if (bytes_per_inode > 0) {
650 sprintf(buf, " -i %ld", bytes_per_inode);
651 strscat(mop->mo_mkfsopts, buf,
652 sizeof(mop->mo_mkfsopts));
657 strscat(mop->mo_mkfsopts, " -q",
658 sizeof(mop->mo_mkfsopts));
661 /* start handle -O mkfs options */
662 if ((start = strstr(mop->mo_mkfsopts, "-O")) != NULL) {
663 if (strstr(start + 2, "-O") != NULL) {
665 "%s: don't specify multiple -O options\n",
669 start = moveopts_to_end(start);
670 maxbuflen = sizeof(mop->mo_mkfsopts) -
671 (start - mop->mo_mkfsopts) - strlen(start);
672 ret = enable_default_ext4_features(mop, start, maxbuflen, 1);
674 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts),
675 maxbuflen = sizeof(mop->mo_mkfsopts) -
676 strlen(mop->mo_mkfsopts);
677 ret = enable_default_ext4_features(mop, start, maxbuflen, 0);
681 /* end handle -O mkfs options */
683 /* start handle -E mkfs options */
684 if ((start = strstr(mop->mo_mkfsopts, "-E")) != NULL) {
685 if (strstr(start + 2, "-E") != NULL) {
687 "%s: don't specify multiple -E options\n",
691 start = moveopts_to_end(start);
692 maxbuflen = sizeof(mop->mo_mkfsopts) -
693 (start - mop->mo_mkfsopts) - strlen(start);
696 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts);
697 maxbuflen = sizeof(mop->mo_mkfsopts) -
698 strlen(mop->mo_mkfsopts);
701 /* In order to align the filesystem metadata on 1MB boundaries,
702 * give a resize value that will reserve a power-of-two group
703 * descriptor blocks, but leave one block for the superblock.
704 * Only useful for filesystems with < 2^32 blocks due to resize
706 if (IS_OST(&mop->mo_ldd) && mop->mo_device_sz > 100 * 1024 &&
707 mop->mo_device_sz * 1024 / L_BLOCK_SIZE <= 0xffffffffULL) {
708 unsigned group_blocks = L_BLOCK_SIZE * 8;
709 unsigned desc_per_block = L_BLOCK_SIZE / 32;
710 unsigned resize_blks;
712 resize_blks = (1ULL<<32) - desc_per_block*group_blocks;
713 snprintf(buf, sizeof(buf), "%u", resize_blks);
714 append_unique(start, ext_opts ? "," : " -E ",
715 "resize", buf, maxbuflen);
719 /* Avoid zeroing out the full journal - speeds up mkfs */
720 if (is_e2fsprogs_feature_supp("-E lazy_journal_init") == 0)
721 append_unique(start, ext_opts ? "," : " -E ",
722 "lazy_journal_init", NULL, maxbuflen);
723 /* end handle -E mkfs options */
725 /* Allow reformat of full devices (as opposed to
726 partitions.) We already checked for mounted dev. */
727 strscat(mop->mo_mkfsopts, " -F", sizeof(mop->mo_mkfsopts));
729 snprintf(mkfs_cmd, sizeof(mkfs_cmd),
730 "%s -j -b %d -L %s ", MKE2FS, L_BLOCK_SIZE,
731 mop->mo_ldd.ldd_svname);
733 fprintf(stderr,"%s: unsupported fs type: %d (%s)\n",
734 progname, mop->mo_ldd.ldd_mount_type,
735 MT_STR(&mop->mo_ldd));
739 /* For loop device format the dev, not the filename */
740 dev = mop->mo_device;
741 if (mop->mo_flags & MO_IS_LOOP)
742 dev = mop->mo_loopdev;
744 vprint("formatting backing filesystem %s on %s\n",
745 MT_STR(&mop->mo_ldd), dev);
746 vprint("\ttarget name %s\n", mop->mo_ldd.ldd_svname);
747 vprint("\t4k blocks "LPU64"\n", block_count);
748 vprint("\toptions %s\n", mop->mo_mkfsopts);
750 /* mkfs_cmd's trailing space is important! */
751 strscat(mkfs_cmd, mop->mo_mkfsopts, sizeof(mkfs_cmd));
752 strscat(mkfs_cmd, " ", sizeof(mkfs_cmd));
753 strscat(mkfs_cmd, dev, sizeof(mkfs_cmd));
754 if (block_count != 0) {
755 sprintf(buf, " "LPU64, block_count);
756 strscat(mkfs_cmd, buf, sizeof(mkfs_cmd));
759 vprint("mkfs_cmd = %s\n", mkfs_cmd);
760 ret = run_command(mkfs_cmd, sizeof(mkfs_cmd));
763 fprintf(stderr, "Unable to build fs %s (%d)\n", dev, ret);
768 int ldiskfs_prepare_lustre(struct mkfs_opts *mop,
769 char *default_mountopts, int default_len,
770 char *always_mountopts, int always_len)
772 struct lustre_disk_data *ldd = &mop->mo_ldd;
775 /* Set MO_IS_LOOP to indicate a loopback device is needed */
776 ret = is_block(mop->mo_device);
779 } else if (ret == 0) {
780 mop->mo_flags |= MO_IS_LOOP;
783 strscat(default_mountopts, ",errors=remount-ro", default_len);
784 if (IS_MDT(ldd) || IS_MGS(ldd))
785 strscat(always_mountopts, ",user_xattr", always_len);
790 int read_file(char *path, char *buf, int size)
794 fd = fopen(path, "r");
798 /* should not ignore fgets(3)'s return value */
799 if (!fgets(buf, size, fd)) {
800 fprintf(stderr, "reading from %s: %s", path, strerror(errno));
808 int write_file(char *path, char *buf)
812 fd = fopen(path, "w");
821 /* This is to tune the kernel for good SCSI performance.
822 * For that we set the value of /sys/block/{dev}/queue/max_sectors_kb
823 * to the value of /sys/block/{dev}/queue/max_hw_sectors_kb */
824 int set_blockdev_tunables(char *source, struct mount_opts *mop, int fan_out)
826 glob_t glob_info = { 0 };
827 struct stat stat_buf;
828 char *chk_major, *chk_minor;
829 char *savept = NULL, *dev;
831 char buf[PATH_MAX] = {'\0'}, path[PATH_MAX] = {'\0'};
832 char real_path[PATH_MAX] = {'\0'};
839 ret_path = realpath(source, real_path);
840 if (ret_path == NULL) {
842 fprintf(stderr, "warning: %s: cannot resolve: %s\n",
843 source, strerror(errno));
847 if (strncmp(real_path, "/dev/loop", 9) == 0)
850 if ((real_path[0] != '/') && (strpbrk(real_path, ",:") != NULL))
853 snprintf(path, sizeof(path), "/sys/block%s", real_path + 4);
854 if (access(path, X_OK) == 0)
857 /* The name of the device say 'X' specified in /dev/X may not
858 * match any entry under /sys/block/. In that case we need to
859 * match the major/minor number to find the entry under
860 * sys/block corresponding to /dev/X */
862 /* Don't chop tail digit on /dev/mapper/xxx, LU-478 */
863 if (strncmp(real_path, "/dev/mapper", 11) != 0) {
864 dev = real_path + strlen(real_path);
865 while (--dev > real_path && isdigit(*dev))
868 if (strncmp(real_path, "/dev/md_", 8) == 0)
872 rc = stat(real_path, &stat_buf);
875 fprintf(stderr, "warning: %s, device %s stat failed\n",
876 strerror(errno), real_path);
880 major = major(stat_buf.st_rdev);
881 minor = minor(stat_buf.st_rdev);
882 rc = glob("/sys/block/*", GLOB_NOSORT, NULL, &glob_info);
885 fprintf(stderr, "warning: failed to read entries under "
887 globfree(&glob_info);
891 for (i = 0; i < glob_info.gl_pathc; i++){
892 snprintf(path, sizeof(path), "%s/dev", glob_info.gl_pathv[i]);
894 rc = read_file(path, buf, sizeof(buf));
898 if (buf[strlen(buf) - 1] == '\n')
899 buf[strlen(buf) - 1] = '\0';
901 chk_major = strtok_r(buf, ":", &savept);
903 if (major == atoi(chk_major) &&minor == atoi(chk_minor))
907 if (i == glob_info.gl_pathc) {
909 fprintf(stderr,"warning: device %s does not match any "
910 "entry under /sys/block\n", real_path);
911 globfree(&glob_info);
915 /* Chop off "/dev" from path we found */
916 path[strlen(glob_info.gl_pathv[i])] = '\0';
917 globfree(&glob_info);
920 if (strncmp(real_path, "/dev/md", 7) == 0) {
921 snprintf(real_path, sizeof(real_path), "%s/%s", path,
924 rc = read_file(real_path, buf, sizeof(buf));
927 fprintf(stderr, "warning: opening %s: %s\n",
928 real_path, strerror(errno));
932 if (atoi(buf) >= mop->mo_md_stripe_cache_size)
935 if (strlen(buf) - 1 > 0) {
936 snprintf(buf, sizeof(buf), "%d",
937 mop->mo_md_stripe_cache_size);
938 rc = write_file(real_path, buf);
940 fprintf(stderr, "warning: opening %s: %s\n",
941 real_path, strerror(errno));
943 /* Return since raid and disk tunables are different */
947 snprintf(real_path, sizeof(real_path), "%s/%s", path,
948 MAX_HW_SECTORS_KB_PATH);
949 rc = read_file(real_path, buf, sizeof(buf));
952 fprintf(stderr, "warning: opening %s: %s\n",
953 real_path, strerror(errno));
954 /* No MAX_HW_SECTORS_KB_PATH isn't necessary an
955 * error for some device. */
959 if (strlen(buf) - 1 > 0) {
960 snprintf(real_path, sizeof(real_path), "%s/%s", path,
961 MAX_SECTORS_KB_PATH);
962 rc = write_file(real_path, buf);
965 fprintf(stderr, "warning: writing to %s: %s\n",
966 real_path, strerror(errno));
967 /* No MAX_SECTORS_KB_PATH isn't necessary an
968 * error for some device. */
975 glob_info.gl_pathc = 0;
976 glob_info.gl_offs = 0;
977 /* if device is multipath device, tune its slave devices */
978 snprintf(real_path, sizeof(real_path), "%s/slaves/*", path);
979 rc = glob(real_path, GLOB_NOSORT, NULL, &glob_info);
981 for (i = 0; rc == 0 && i < glob_info.gl_pathc; i++){
982 slave = basename(glob_info.gl_pathv[i]);
983 snprintf(real_path, sizeof(real_path), "/dev/%s", slave);
984 rc = set_blockdev_tunables(real_path, mop, 0);
987 if (rc == GLOB_NOMATCH) {
988 /* no slave device is not an error */
990 } else if (rc && verbose) {
992 fprintf(stderr, "warning: %s, failed to read"
993 " entries under %s/slaves\n",
994 strerror(errno), path);
996 fprintf(stderr, "unable to set tunables for"
997 " slave device %s (slave would be"
998 " unable to handle IO request from"
1003 globfree(&glob_info);
1009 int ldiskfs_tune_lustre(char *dev, struct mount_opts *mop)
1011 return set_blockdev_tunables(dev, mop, 1);
1014 int ldiskfs_label_lustre(struct mount_opts *mop)
1016 char label_cmd[PATH_MAX];
1019 snprintf(label_cmd, sizeof(label_cmd),
1020 TUNE2FS" -f -L %s %s >/dev/null 2>&1",
1021 mop->mo_ldd.ldd_svname, mop->mo_source);
1022 rc = run_command(label_cmd, sizeof(label_cmd));
1027 /* return canonicalized absolute pathname, even if the target file does not
1028 * exist, unlike realpath */
1029 static char *absolute_path(char *devname)
1031 char buf[PATH_MAX + 1];
1035 path = malloc(PATH_MAX + 1);
1039 if (devname[0] != '/') {
1040 if (getcwd(buf, sizeof(buf) - 1) == NULL) {
1045 strcat(buf, devname);
1047 strcpy(buf, devname);
1049 /* truncate filename before calling realpath */
1050 ptr = strrchr(buf, '/');
1056 if (path != realpath(buf, path)) {
1060 /* add the filename back */
1062 strcat(path, ptr + 1);
1066 /* Determine if a device is a block device (as opposed to a file) */
1067 int is_block(char* devname)
1073 devpath = absolute_path(devname);
1074 if (devpath == NULL) {
1075 fprintf(stderr, "%s: failed to resolve path to %s\n",
1080 ret = access(devname, F_OK);
1082 if (strncmp(devpath, "/dev/", 5) == 0) {
1083 /* nobody sane wants to create a loopback file under
1084 * /dev. Let's just report the device doesn't exist */
1085 fprintf(stderr, "%s: %s apparently does not exist\n",
1093 ret = stat(devpath, &st);
1095 fprintf(stderr, "%s: cannot stat %s\n", progname, devpath);
1098 ret = S_ISBLK(st.st_mode);
1104 static int is_feature_enabled(const char *feature, const char *devpath)
1108 char enabled_features[4096] = "";
1111 snprintf(cmd, sizeof(cmd), "%s -R features %s 2>&1",
1114 /* Using popen() instead of run_command() since debugfs does
1115 * not return proper error code if command is not supported */
1116 fp = popen(cmd, "r");
1118 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
1122 ret = fread(enabled_features, 1, sizeof(enabled_features) - 1, fp);
1123 enabled_features[ret] = '\0';
1126 if (strstr(enabled_features, feature))
1131 /* Enable quota accounting */
1132 int ldiskfs_enable_quota(struct mkfs_opts *mop)
1136 int cmdsz = sizeof(cmd), ret;
1138 if (is_e2fsprogs_feature_supp("-O quota") != 0) {
1139 fprintf(stderr, "%s: \"-O quota\" is is not supported by "
1140 "current e2fsprogs\n", progname);
1144 dev = mop->mo_device;
1145 if (mop->mo_flags & MO_IS_LOOP)
1146 dev = mop->mo_loopdev;
1148 /* Quota feature is already enabled? */
1149 if (is_feature_enabled("quota", dev)) {
1150 vprint("Quota feature is already enabled.\n");
1154 /* Turn on quota feature by "tune2fs -O quota" */
1155 snprintf(cmd, cmdsz, "%s -O quota %s", TUNE2FS, dev);
1156 ret = run_command(cmd, cmdsz);
1158 fprintf(stderr, "command:%s (%d)", cmd, ret);
1163 int ldiskfs_init(void)
1165 /* Required because full path to DEBUGFS is not specified */
1166 setenv("PATH", "/bin:/sbin:/usr/bin:/usr/sbin:/usr/local/bin", 0);
1171 void ldiskfs_fini(void)