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) 2011, Whamcloud, Inc.
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 */
58 #include <sys/types.h>
60 #include <sys/mount.h>
61 #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>
77 #include <lustre_disk.h>
78 #include <lustre_param.h>
79 #include <lnet/lnetctl.h>
80 #include <lustre_ver.h>
81 #include "mount_utils.h"
83 #define MAX_HW_SECTORS_KB_PATH "queue/max_hw_sectors_kb"
84 #define MAX_SECTORS_KB_PATH "queue/max_sectors_kb"
85 #define 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));
228 snprintf(cmd, cmdsz, "rm -rf %s", tmpdir);
229 run_command(cmd, cmdsz);
231 verrprint("Failed to read old data (%d)\n", ret);
233 /* As long as we at least have the label, we're good to go */
234 snprintf(cmd, sizeof(cmd), E2LABEL" %s", dev);
235 ret = readcmd(cmd, mo_ldd->ldd_svname, sizeof(mo_ldd->ldd_svname) - 1);
241 /* Display the need for the latest e2fsprogs to be installed. make_backfs
242 * indicates if the caller is make_lustre_backfs() or not. */
243 void disp_old_e2fsprogs_msg(const char *feature, int make_backfs)
245 static int msg_displayed;
248 fprintf(stderr, "WARNING: %s does not support %s "
249 "feature.\n\n", E2FSPROGS, feature);
255 fprintf(stderr, "WARNING: The %s package currently installed on "
256 "your system does not support \"%s\" feature.\n",
258 #if !(HAVE_LDISKFSPROGS)
259 fprintf(stderr, "Please install the latest version of e2fsprogs from\n"
260 "http://downloads.whamcloud.com/public/e2fsprogs/latest/\n"
261 "to enable this feature.\n");
264 fprintf(stderr, "Feature will not be enabled until %s"
265 "is updated and '%s -O %s %%{device}' "
266 "is run.\n\n", E2FSPROGS, TUNE2FS, feature);
269 /* Check whether the file exists in the device */
270 static int file_in_dev(char *file_name, char *dev_name)
273 char debugfs_cmd[256];
274 unsigned int inode_num;
277 /* Construct debugfs command line. */
278 snprintf(debugfs_cmd, sizeof(debugfs_cmd),
279 "%s -c -R 'stat %s' '%s' 2>&1 | egrep '(Inode|unsupported)'",
280 DEBUGFS, file_name, dev_name);
282 fp = popen(debugfs_cmd, "r");
284 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
288 if (fscanf(fp, "Inode: %u", &inode_num) == 1) { /* exist */
292 i = fread(debugfs_cmd, 1, sizeof(debugfs_cmd), fp);
295 fprintf(stderr, "%s", debugfs_cmd);
296 if (strstr(debugfs_cmd, "unsupported feature")) {
297 disp_old_e2fsprogs_msg("an unknown", 0);
306 /* Check whether the device has already been used with lustre */
307 int ldiskfs_is_lustre(char *dev, unsigned *mount_type)
311 ret = file_in_dev(MOUNT_DATA_FILE, dev);
313 /* in the -1 case, 'extents' means IS a lustre target */
314 *mount_type = LDD_MT_LDISKFS;
318 ret = file_in_dev(LAST_RCVD, dev);
320 *mount_type = LDD_MT_LDISKFS;
327 /* Check if a certain feature is supported by e2fsprogs.
328 * Firstly we try to use "debugfs supported_features" command to check if
329 * the feature is supported. If this fails we try to set this feature with
330 * mke2fs to check for its support. */
331 static int is_e2fsprogs_feature_supp(const char *feature)
333 static char supp_features[4096] = "";
336 char imgname[] = "/tmp/test-img-XXXXXX";
340 if (supp_features[0] == '\0') {
341 snprintf(cmd, sizeof(cmd), "%s -c -R supported_features 2>&1",
344 /* Using popen() instead of run_command() since debugfs does
345 * not return proper error code if command is not supported */
346 fp = popen(cmd, "r");
348 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
351 ret = fread(supp_features, 1, sizeof(supp_features), fp);
354 if (ret > 0 && strstr(supp_features,
355 strncmp(feature, "-O ", 3) ? feature : feature+3))
358 if ((fd = mkstemp(imgname)) < 0)
363 snprintf(cmd, sizeof(cmd), "%s -F %s %s 100 >/dev/null 2>&1",
364 MKE2FS, feature, imgname);
365 /* run_command() displays the output of mke2fs when it fails for
366 * some feature, so use system() directly */
375 * append_unique: append @key or @key=@val pair to @buf only if @key does not
377 * @buf: buffer to hold @key or @key=@val
378 * @prefix: prefix string before @key
380 * @val: value string if it's a @key=@val pair
382 static void append_unique(char *buf, char *prefix, char *key, char *val,
391 anchor = end = strstr(buf, key);
392 /* try to find exact match string in @buf */
393 while (end && *end != '\0' && *end != ',' && *end != ' ' && *end != '=')
396 if (anchor == NULL || strlen(key) != len ||
397 strncmp(anchor, key, len) != 0) {
399 strscat(buf, prefix, maxbuflen);
401 strscat(buf, key, maxbuflen);
403 strscat(buf, "=", maxbuflen);
404 strscat(buf, val, maxbuflen);
409 static int enable_default_ext4_features(struct mkfs_opts *mop, char *anchor,
410 size_t maxbuflen, int user_spec)
412 if (IS_OST(&mop->mo_ldd)) {
413 append_unique(anchor, user_spec ? "," : " -O ",
414 "extents", NULL, sizeof(mop->mo_mkfsopts));
415 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
416 } else if (IS_MDT(&mop->mo_ldd)) {
417 append_unique(anchor, user_spec ? "," : " -O ",
418 "dirdata", NULL, maxbuflen);
419 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
420 append_unique(anchor, ",", "^extents", NULL, maxbuflen);
422 append_unique(anchor, user_spec ? "," : " -O ",
423 "uninit_bg", NULL, maxbuflen);
426 /* Multiple mount protection enabled only if failover node specified */
427 if (mop->mo_flags & MO_FAILOVER) {
428 if (is_e2fsprogs_feature_supp("-O mmp") == 0)
429 append_unique(anchor, ",", "mmp", NULL, maxbuflen);
431 disp_old_e2fsprogs_msg("mmp", 1);
434 /* Allow more than 65000 subdirectories */
435 if (is_e2fsprogs_feature_supp("-O dir_nlink") == 0)
436 append_unique(anchor, ",", "dir_nlink", NULL, maxbuflen);
438 /* The following options are only valid for ext4-based ldiskfs.
439 * If --backfstype=ext3 is specified, do not enable them. */
440 if (mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3)
443 /* Enable quota by default */
444 if (is_e2fsprogs_feature_supp("-O quota") == 0) {
445 append_unique(anchor, ",", "quota", NULL, maxbuflen);
448 fprintf(stderr, "\"-O quota\" must be supported by "
449 "e2fsprogs, please upgrade your e2fsprogs.\n");
453 /* Allow files larger than 2TB. Also needs LU-16, but not harmful. */
454 if (is_e2fsprogs_feature_supp("-O huge_file") == 0)
455 append_unique(anchor, ",", "huge_file", NULL, maxbuflen);
457 /* Enable large block addresses if the LUN is over 2^32 blocks. */
458 if (mop->mo_device_sz / (L_BLOCK_SIZE >> 10) >= 0x100002000ULL &&
459 is_e2fsprogs_feature_supp("-O 64bit") == 0)
460 append_unique(anchor, ",", "64bit", NULL, maxbuflen);
462 /* Cluster inode/block bitmaps and inode table for more efficient IO.
463 * Align the flex groups on a 1MB boundary for better performance. */
464 /* This -O feature needs to go last, since it adds the "-G" option. */
465 if (is_e2fsprogs_feature_supp("-O flex_bg") == 0) {
468 append_unique(anchor, ",", "flex_bg", NULL, maxbuflen);
470 if (IS_OST(&mop->mo_ldd)) {
471 snprintf(tmp_buf, sizeof(tmp_buf), " -G %u",
472 (1 << 20) / L_BLOCK_SIZE);
473 strscat(anchor, tmp_buf, maxbuflen);
476 /* Don't add any more "-O" options here, see last comment above */
481 * moveopts_to_end: find the option string, move remaining strings to
482 * where option string starts, and append the option
484 * @start: where the option string starts before the move
485 * RETURN: where the option string starts after the move
487 static char *moveopts_to_end(char *start)
492 /* skip whitespace before options */
497 /* find end of option characters */
498 while (*end != ' ' && *end != '\0')
502 strncpy(save, start, end - start);
503 save[end - start] = '\0';
505 /* move remaining options up front */
507 memmove(start, end, strlen(end));
508 *(start + strlen(end)) = '\0';
510 /* append the specified options */
511 if (*(start + strlen(start) - 1) != ' ')
513 idx = start + strlen(start);
519 /* Build fs according to type */
520 int ldiskfs_make_lustre(struct mkfs_opts *mop)
522 __u64 device_sz = mop->mo_device_sz, block_count = 0;
523 char mkfs_cmd[PATH_MAX];
527 int ret = 0, ext_opts = 0;
530 if (!(mop->mo_flags & MO_IS_LOOP)) {
531 mop->mo_device_sz = get_device_size(mop->mo_device);
533 if (mop->mo_device_sz == 0)
536 /* Compare to real size */
537 if (device_sz == 0 || device_sz > mop->mo_device_sz)
538 device_sz = mop->mo_device_sz;
540 mop->mo_device_sz = device_sz;
543 if (mop->mo_device_sz != 0) {
544 if (mop->mo_device_sz < 8096){
545 fprintf(stderr, "%s: size of filesystem must be larger "
546 "than 8MB, but is set to %lldKB\n",
547 progname, (long long)mop->mo_device_sz);
550 block_count = mop->mo_device_sz / (L_BLOCK_SIZE >> 10);
551 /* If the LUN size is just over 2^32 blocks, limit the
552 * filesystem size to 2^32-1 blocks to avoid problems with
553 * ldiskfs/mkfs not handling this size. Bug 22906 */
554 if (block_count > 0xffffffffULL && block_count < 0x100002000ULL)
555 block_count = 0xffffffffULL;
558 if ((mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3) ||
559 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS) ||
560 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS2)) {
563 /* Journal size in MB */
564 if (strstr(mop->mo_mkfsopts, "-J") == NULL) {
565 /* Choose our own default journal size */
566 long journal_sz = 0, max_sz;
567 if (device_sz > 1024 * 1024) /* 1GB */
568 journal_sz = (device_sz / 102400) * 4;
569 /* cap journal size at 1GB */
570 if (journal_sz > 1024L)
573 max_sz = (102400 * L_BLOCK_SIZE) >> 20; /* 400MB */
574 if (journal_sz > max_sz)
577 sprintf(buf, " -J size=%ld", journal_sz);
578 strscat(mop->mo_mkfsopts, buf,
579 sizeof(mop->mo_mkfsopts));
583 /* Inode size (for extended attributes). The LOV EA size is
584 * 32 (EA hdr) + 32 (lov_mds_md) + stripes * 24 (lov_ost_data),
585 * and we want some margin above that for ACLs, other EAs... */
586 if (strstr(mop->mo_mkfsopts, "-I") == NULL) {
587 if (IS_MDT(&mop->mo_ldd)) {
588 if (mop->mo_stripe_count > 72)
589 inode_size = 512; /* bz 7241 */
590 /* see also "-i" below for EA blocks */
591 else if (mop->mo_stripe_count > 32)
593 else if (mop->mo_stripe_count > 10)
597 } else if (IS_OST(&mop->mo_ldd)) {
598 /* We store MDS FID and OST objid in EA on OST
599 * we need to make inode bigger as well. */
603 if (inode_size > 0) {
604 sprintf(buf, " -I %ld", inode_size);
605 strscat(mop->mo_mkfsopts, buf,
606 sizeof(mop->mo_mkfsopts));
610 /* Bytes_per_inode: disk size / num inodes */
611 if (strstr(mop->mo_mkfsopts, "-i") == NULL &&
612 strstr(mop->mo_mkfsopts, "-N") == NULL) {
613 long bytes_per_inode = 0;
615 /* Allocate more inodes on MDT devices. There is
616 * no data stored on the MDT, and very little extra
617 * metadata beyond the inode. It could go down as
618 * low as 1024 bytes, but this is conservative.
619 * Account for external EA blocks for wide striping. */
620 if (IS_MDT(&mop->mo_ldd)) {
621 bytes_per_inode = inode_size + 1536;
623 if (mop->mo_stripe_count > 72) {
624 int extra = mop->mo_stripe_count * 24;
625 extra = ((extra - 1) | 4095) + 1;
626 bytes_per_inode += extra;
630 /* Allocate fewer inodes on large OST devices. Most
631 * filesystems can be much more aggressive than even
632 * this, but it is impossible to know in advance. */
633 if (IS_OST(&mop->mo_ldd)) {
634 /* OST > 16TB assume average file size 1MB */
635 if (device_sz > (16ULL << 30))
636 bytes_per_inode = 1024 * 1024;
637 /* OST > 4TB assume average file size 512kB */
638 else if (device_sz > (4ULL << 30))
639 bytes_per_inode = 512 * 1024;
640 /* OST > 1TB assume average file size 256kB */
641 else if (device_sz > (1ULL << 30))
642 bytes_per_inode = 256 * 1024;
643 /* OST > 10GB assume average file size 64kB,
644 * plus a bit so that inodes will fit into a
645 * 256x flex_bg without overflowing */
646 else if (device_sz > (10ULL << 20))
647 bytes_per_inode = 69905;
650 if (bytes_per_inode > 0) {
651 sprintf(buf, " -i %ld", bytes_per_inode);
652 strscat(mop->mo_mkfsopts, buf,
653 sizeof(mop->mo_mkfsopts));
658 strscat(mop->mo_mkfsopts, " -q",
659 sizeof(mop->mo_mkfsopts));
662 /* start handle -O mkfs options */
663 if ((start = strstr(mop->mo_mkfsopts, "-O")) != NULL) {
664 if (strstr(start + 2, "-O") != NULL) {
666 "%s: don't specify multiple -O options\n",
670 start = moveopts_to_end(start);
671 maxbuflen = sizeof(mop->mo_mkfsopts) -
672 (start - mop->mo_mkfsopts) - strlen(start);
673 ret = enable_default_ext4_features(mop, start, maxbuflen, 1);
675 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts),
676 maxbuflen = sizeof(mop->mo_mkfsopts) -
677 strlen(mop->mo_mkfsopts);
678 ret = enable_default_ext4_features(mop, start, maxbuflen, 0);
682 /* end handle -O mkfs options */
684 /* start handle -E mkfs options */
685 if ((start = strstr(mop->mo_mkfsopts, "-E")) != NULL) {
686 if (strstr(start + 2, "-E") != NULL) {
688 "%s: don't specify multiple -E options\n",
692 start = moveopts_to_end(start);
693 maxbuflen = sizeof(mop->mo_mkfsopts) -
694 (start - mop->mo_mkfsopts) - strlen(start);
697 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts);
698 maxbuflen = sizeof(mop->mo_mkfsopts) -
699 strlen(mop->mo_mkfsopts);
702 /* In order to align the filesystem metadata on 1MB boundaries,
703 * give a resize value that will reserve a power-of-two group
704 * descriptor blocks, but leave one block for the superblock.
705 * Only useful for filesystems with < 2^32 blocks due to resize
707 if (IS_OST(&mop->mo_ldd) && mop->mo_device_sz > 100 * 1024 &&
708 mop->mo_device_sz * 1024 / L_BLOCK_SIZE <= 0xffffffffULL) {
709 unsigned group_blocks = L_BLOCK_SIZE * 8;
710 unsigned desc_per_block = L_BLOCK_SIZE / 32;
711 unsigned resize_blks;
713 resize_blks = (1ULL<<32) - desc_per_block*group_blocks;
714 snprintf(buf, sizeof(buf), "%u", resize_blks);
715 append_unique(start, ext_opts ? "," : " -E ",
716 "resize", buf, maxbuflen);
720 /* Avoid zeroing out the full journal - speeds up mkfs */
721 if (is_e2fsprogs_feature_supp("-E lazy_journal_init") == 0)
722 append_unique(start, ext_opts ? "," : " -E ",
723 "lazy_journal_init", NULL, maxbuflen);
724 /* end handle -E mkfs options */
726 /* Allow reformat of full devices (as opposed to
727 partitions.) We already checked for mounted dev. */
728 strscat(mop->mo_mkfsopts, " -F", sizeof(mop->mo_mkfsopts));
730 snprintf(mkfs_cmd, sizeof(mkfs_cmd),
731 "%s -j -b %d -L %s ", MKE2FS, L_BLOCK_SIZE,
732 mop->mo_ldd.ldd_svname);
734 fprintf(stderr,"%s: unsupported fs type: %d (%s)\n",
735 progname, mop->mo_ldd.ldd_mount_type,
736 MT_STR(&mop->mo_ldd));
740 /* For loop device format the dev, not the filename */
741 dev = mop->mo_device;
742 if (mop->mo_flags & MO_IS_LOOP)
743 dev = mop->mo_loopdev;
745 vprint("formatting backing filesystem %s on %s\n",
746 MT_STR(&mop->mo_ldd), dev);
747 vprint("\ttarget name %s\n", mop->mo_ldd.ldd_svname);
748 vprint("\t4k blocks "LPU64"\n", block_count);
749 vprint("\toptions %s\n", mop->mo_mkfsopts);
751 /* mkfs_cmd's trailing space is important! */
752 strscat(mkfs_cmd, mop->mo_mkfsopts, sizeof(mkfs_cmd));
753 strscat(mkfs_cmd, " ", sizeof(mkfs_cmd));
754 strscat(mkfs_cmd, dev, sizeof(mkfs_cmd));
755 if (block_count != 0) {
756 sprintf(buf, " "LPU64, block_count);
757 strscat(mkfs_cmd, buf, sizeof(mkfs_cmd));
760 vprint("mkfs_cmd = %s\n", mkfs_cmd);
761 ret = run_command(mkfs_cmd, sizeof(mkfs_cmd));
764 fprintf(stderr, "Unable to build fs %s (%d)\n", dev, ret);
769 int ldiskfs_prepare_lustre(struct mkfs_opts *mop,
770 char *default_mountopts, int default_len,
771 char *always_mountopts, int always_len)
773 struct lustre_disk_data *ldd = &mop->mo_ldd;
776 /* Set MO_IS_LOOP to indicate a loopback device is needed */
777 ret = is_block(mop->mo_device);
780 } else if (ret == 0) {
781 mop->mo_flags |= MO_IS_LOOP;
784 strscat(default_mountopts, ",errors=remount-ro", default_len);
785 if (IS_MDT(ldd) || IS_MGS(ldd))
786 strscat(always_mountopts, ",user_xattr", always_len);
791 int read_file(char *path, char *buf, int size)
795 fd = fopen(path, "r");
799 /* should not ignore fgets(3)'s return value */
800 if (!fgets(buf, size, fd)) {
801 fprintf(stderr, "reading from %s: %s", path, strerror(errno));
809 int write_file(char *path, char *buf)
813 fd = fopen(path, "w");
822 /* This is to tune the kernel for good SCSI performance.
823 * For that we set the value of /sys/block/{dev}/queue/max_sectors_kb
824 * to the value of /sys/block/{dev}/queue/max_hw_sectors_kb */
825 int set_blockdev_tunables(char *source, struct mount_opts *mop, int fan_out)
827 glob_t glob_info = { 0 };
828 struct stat stat_buf;
829 char *chk_major, *chk_minor;
830 char *savept = NULL, *dev;
832 char buf[PATH_MAX] = {'\0'}, path[PATH_MAX] = {'\0'};
833 char real_path[PATH_MAX] = {'\0'};
840 ret_path = realpath(source, real_path);
841 if (ret_path == NULL) {
843 fprintf(stderr, "warning: %s: cannot resolve: %s\n",
844 source, strerror(errno));
848 if (strncmp(real_path, "/dev/loop", 9) == 0)
851 if ((real_path[0] != '/') && (strpbrk(real_path, ",:") != NULL))
854 snprintf(path, sizeof(path), "/sys/block%s", real_path + 4);
855 if (access(path, X_OK) == 0)
858 /* The name of the device say 'X' specified in /dev/X may not
859 * match any entry under /sys/block/. In that case we need to
860 * match the major/minor number to find the entry under
861 * sys/block corresponding to /dev/X */
863 /* Don't chop tail digit on /dev/mapper/xxx, LU-478 */
864 if (strncmp(real_path, "/dev/mapper", 11) != 0) {
865 dev = real_path + strlen(real_path);
866 while (--dev > real_path && isdigit(*dev))
869 if (strncmp(real_path, "/dev/md_", 8) == 0)
873 rc = stat(real_path, &stat_buf);
876 fprintf(stderr, "warning: %s, device %s stat failed\n",
877 strerror(errno), real_path);
881 major = major(stat_buf.st_rdev);
882 minor = minor(stat_buf.st_rdev);
883 rc = glob("/sys/block/*", GLOB_NOSORT, NULL, &glob_info);
886 fprintf(stderr, "warning: failed to read entries under "
888 globfree(&glob_info);
892 for (i = 0; i < glob_info.gl_pathc; i++){
893 snprintf(path, sizeof(path), "%s/dev", glob_info.gl_pathv[i]);
895 rc = read_file(path, buf, sizeof(buf));
899 if (buf[strlen(buf) - 1] == '\n')
900 buf[strlen(buf) - 1] = '\0';
902 chk_major = strtok_r(buf, ":", &savept);
904 if (major == atoi(chk_major) &&minor == atoi(chk_minor))
908 if (i == glob_info.gl_pathc) {
910 fprintf(stderr,"warning: device %s does not match any "
911 "entry under /sys/block\n", real_path);
912 globfree(&glob_info);
916 /* Chop off "/dev" from path we found */
917 path[strlen(glob_info.gl_pathv[i])] = '\0';
918 globfree(&glob_info);
921 if (strncmp(real_path, "/dev/md", 7) == 0) {
922 snprintf(real_path, sizeof(real_path), "%s/%s", path,
925 rc = read_file(real_path, buf, sizeof(buf));
928 fprintf(stderr, "warning: opening %s: %s\n",
929 real_path, strerror(errno));
933 if (atoi(buf) >= mop->mo_md_stripe_cache_size)
936 if (strlen(buf) - 1 > 0) {
937 snprintf(buf, sizeof(buf), "%d",
938 mop->mo_md_stripe_cache_size);
939 rc = write_file(real_path, buf);
941 fprintf(stderr, "warning: opening %s: %s\n",
942 real_path, strerror(errno));
944 /* Return since raid and disk tunables are different */
948 snprintf(real_path, sizeof(real_path), "%s/%s", path,
949 MAX_HW_SECTORS_KB_PATH);
950 rc = read_file(real_path, buf, sizeof(buf));
953 fprintf(stderr, "warning: opening %s: %s\n",
954 real_path, strerror(errno));
955 /* No MAX_HW_SECTORS_KB_PATH isn't necessary an
956 * error for some device. */
960 if (strlen(buf) - 1 > 0) {
961 snprintf(real_path, sizeof(real_path), "%s/%s", path,
962 MAX_SECTORS_KB_PATH);
963 rc = write_file(real_path, buf);
966 fprintf(stderr, "warning: writing to %s: %s\n",
967 real_path, strerror(errno));
968 /* No MAX_SECTORS_KB_PATH isn't necessary an
969 * error for some device. */
976 glob_info.gl_pathc = 0;
977 glob_info.gl_offs = 0;
978 /* if device is multipath device, tune its slave devices */
979 snprintf(real_path, sizeof(real_path), "%s/slaves/*", path);
980 rc = glob(real_path, GLOB_NOSORT, NULL, &glob_info);
982 for (i = 0; rc == 0 && i < glob_info.gl_pathc; i++){
983 slave = basename(glob_info.gl_pathv[i]);
984 snprintf(real_path, sizeof(real_path), "/dev/%s", slave);
985 rc = set_blockdev_tunables(real_path, mop, 0);
988 if (rc == GLOB_NOMATCH) {
989 /* no slave device is not an error */
991 } else if (rc && verbose) {
993 fprintf(stderr, "warning: %s, failed to read"
994 " entries under %s/slaves\n",
995 strerror(errno), path);
997 fprintf(stderr, "unable to set tunables for"
998 " slave device %s (slave would be"
999 " unable to handle IO request from"
1004 globfree(&glob_info);
1010 int ldiskfs_tune_lustre(char *dev, struct mount_opts *mop)
1012 return set_blockdev_tunables(dev, mop, 1);
1015 int ldiskfs_label_lustre(struct mount_opts *mop)
1017 char label_cmd[PATH_MAX];
1020 snprintf(label_cmd, sizeof(label_cmd), E2LABEL" %s %s",
1021 mop->mo_source, mop->mo_ldd.ldd_svname);
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 int ldiskfs_init(void)
1106 /* Required because full path to DEBUGFS is not specified */
1107 setenv("PATH", "/bin:/sbin:/usr/bin:/usr/sbin:/usr/local/bin", 0);
1112 void ldiskfs_fini(void)