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 int ldiskfs_read_ldd(char *dev, struct lustre_disk_data *mo_ldd)
167 char tmpdir[] = "/tmp/dirXXXXXX";
172 int cmdsz = sizeof(cmd);
174 /* Make a temporary directory to hold Lustre data files. */
175 if (!mkdtemp(tmpdir)) {
176 fprintf(stderr, "%s: Can't create temporary directory %s: %s\n",
177 progname, tmpdir, strerror(errno));
181 /* TODO: it's worth observing the get_mountdata() function that is
182 in mount_utils.c for getting the mountdata out of the
185 /* Construct debugfs command line. */
186 snprintf(cmd, cmdsz, "%s -c -R 'dump /%s %s/mountdata' '%s'",
187 DEBUGFS, MOUNT_DATA_FILE, tmpdir, dev);
189 ret = run_command(cmd, cmdsz);
191 verrprint("%s: Unable to dump %s dir (%d)\n",
192 progname, MOUNT_CONFIGS_DIR, ret);
194 sprintf(filepnm, "%s/mountdata", tmpdir);
195 filep = fopen(filepnm, "r");
198 vprint("Reading %s\n", MOUNT_DATA_FILE);
199 num_read = fread(mo_ldd, sizeof(*mo_ldd), 1, filep);
200 if (num_read < 1 && ferror(filep)) {
201 fprintf(stderr, "%s: Unable to read from file %s: %s\n",
202 progname, filepnm, strerror(errno));
209 snprintf(cmd, cmdsz, "rm -rf %s", tmpdir);
210 run_command(cmd, cmdsz);
212 verrprint("Failed to read old data (%d)\n", ret);
217 /* Display the need for the latest e2fsprogs to be installed. make_backfs
218 * indicates if the caller is make_lustre_backfs() or not. */
219 void disp_old_e2fsprogs_msg(const char *feature, int make_backfs)
221 static int msg_displayed;
224 fprintf(stderr, "WARNING: %s does not support %s "
225 "feature.\n\n", E2FSPROGS, feature);
231 fprintf(stderr, "WARNING: The %s package currently installed on "
232 "your system does not support \"%s\" feature.\n",
234 #if !(HAVE_LDISKFSPROGS)
235 fprintf(stderr, "Please install the latest version of e2fsprogs from\n"
236 "http://downloads.whamcloud.com/public/e2fsprogs/latest/\n"
237 "to enable this feature.\n");
240 fprintf(stderr, "Feature will not be enabled until %s"
241 "is updated and '%s -O %s %%{device}' "
242 "is run.\n\n", E2FSPROGS, TUNE2FS, feature);
245 /* Check whether the file exists in the device */
246 static int file_in_dev(char *file_name, char *dev_name)
249 char debugfs_cmd[256];
250 unsigned int inode_num;
253 /* Construct debugfs command line. */
254 snprintf(debugfs_cmd, sizeof(debugfs_cmd),
255 "%s -c -R 'stat %s' '%s' 2>&1 | egrep '(Inode|unsupported)'",
256 DEBUGFS, file_name, dev_name);
258 fp = popen(debugfs_cmd, "r");
260 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
264 if (fscanf(fp, "Inode: %u", &inode_num) == 1) { /* exist */
268 i = fread(debugfs_cmd, 1, sizeof(debugfs_cmd), fp);
271 fprintf(stderr, "%s", debugfs_cmd);
272 if (strstr(debugfs_cmd, "unsupported feature")) {
273 disp_old_e2fsprogs_msg("an unknown", 0);
282 /* Check whether the device has already been used with lustre */
283 int ldiskfs_is_lustre(char *dev, unsigned *mount_type)
287 ret = file_in_dev(MOUNT_DATA_FILE, dev);
289 /* in the -1 case, 'extents' means IS a lustre target */
290 *mount_type = LDD_MT_LDISKFS;
294 ret = file_in_dev(LAST_RCVD, dev);
296 *mount_type = LDD_MT_LDISKFS;
303 /* Check if a certain feature is supported by e2fsprogs.
304 * Firstly we try to use "debugfs supported_features" command to check if
305 * the feature is supported. If this fails we try to set this feature with
306 * mke2fs to check for its support. */
307 static int is_e2fsprogs_feature_supp(const char *feature)
309 static char supp_features[4096] = "";
312 char imgname[] = "/tmp/test-img-XXXXXX";
316 if (supp_features[0] == '\0') {
317 snprintf(cmd, sizeof(cmd), "%s -c -R supported_features 2>&1",
320 /* Using popen() instead of run_command() since debugfs does
321 * not return proper error code if command is not supported */
322 fp = popen(cmd, "r");
324 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
327 ret = fread(supp_features, 1, sizeof(supp_features), fp);
330 if (ret > 0 && strstr(supp_features,
331 strncmp(feature, "-O ", 3) ? feature : feature+3))
334 if ((fd = mkstemp(imgname)) < 0)
339 snprintf(cmd, sizeof(cmd), "%s -F %s %s 100 >/dev/null 2>&1",
340 MKE2FS, feature, imgname);
341 /* run_command() displays the output of mke2fs when it fails for
342 * some feature, so use system() directly */
351 * append_unique: append @key or @key=@val pair to @buf only if @key does not
353 * @buf: buffer to hold @key or @key=@val
354 * @prefix: prefix string before @key
356 * @val: value string if it's a @key=@val pair
358 static void append_unique(char *buf, char *prefix, char *key, char *val,
367 anchor = end = strstr(buf, key);
368 /* try to find exact match string in @buf */
369 while (end && *end != '\0' && *end != ',' && *end != ' ' && *end != '=')
372 if (anchor == NULL || strlen(key) != len ||
373 strncmp(anchor, key, len) != 0) {
375 strscat(buf, prefix, maxbuflen);
377 strscat(buf, key, maxbuflen);
379 strscat(buf, "=", maxbuflen);
380 strscat(buf, val, maxbuflen);
385 static void enable_default_ext4_features(struct mkfs_opts *mop, char *anchor,
386 size_t maxbuflen, int user_spec)
388 if (IS_OST(&mop->mo_ldd)) {
389 append_unique(anchor, user_spec ? "," : " -O ",
390 "extents", NULL, sizeof(mop->mo_mkfsopts));
391 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
392 } else if (IS_MDT(&mop->mo_ldd)) {
393 append_unique(anchor, user_spec ? "," : " -O ",
394 "dirdata", NULL, maxbuflen);
395 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
396 append_unique(anchor, ",", "^extents", NULL, maxbuflen);
398 append_unique(anchor, user_spec ? "," : " -O ",
399 "uninit_bg", NULL, maxbuflen);
402 /* Multiple mount protection enabled only if failover node specified */
403 if (mop->mo_flags & MO_FAILOVER) {
404 if (is_e2fsprogs_feature_supp("-O mmp") == 0)
405 append_unique(anchor, ",", "mmp", NULL, maxbuflen);
407 disp_old_e2fsprogs_msg("mmp", 1);
410 /* Allow more than 65000 subdirectories */
411 if (is_e2fsprogs_feature_supp("-O dir_nlink") == 0)
412 append_unique(anchor, ",", "dir_nlink", NULL, maxbuflen);
414 /* The following options are only valid for ext4-based ldiskfs.
415 * If --backfstype=ext3 is specified, do not enable them. */
416 if (mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3)
419 /* Allow files larger than 2TB. Also needs LU-16, but not harmful. */
420 if (is_e2fsprogs_feature_supp("-O huge_file") == 0)
421 append_unique(anchor, ",", "huge_file", NULL, maxbuflen);
423 /* Enable large block addresses if the LUN is over 2^32 blocks. */
424 if (mop->mo_device_sz / (L_BLOCK_SIZE >> 10) >= 0x100002000ULL &&
425 is_e2fsprogs_feature_supp("-O 64bit") == 0)
426 append_unique(anchor, ",", "64bit", NULL, maxbuflen);
428 /* Cluster inode/block bitmaps and inode table for more efficient IO.
429 * Align the flex groups on a 1MB boundary for better performance. */
430 /* This -O feature needs to go last, since it adds the "-G" option. */
431 if (is_e2fsprogs_feature_supp("-O flex_bg") == 0) {
434 append_unique(anchor, ",", "flex_bg", NULL, maxbuflen);
436 if (IS_OST(&mop->mo_ldd)) {
437 snprintf(tmp_buf, sizeof(tmp_buf), " -G %u",
438 (1 << 20) / L_BLOCK_SIZE);
439 strscat(anchor, tmp_buf, maxbuflen);
442 /* Don't add any more "-O" options here, see last comment above */
446 * moveopts_to_end: find the option string, move remaining strings to
447 * where option string starts, and append the option
449 * @start: where the option string starts before the move
450 * RETURN: where the option string starts after the move
452 static char *moveopts_to_end(char *start)
457 /* skip whitespace before options */
462 /* find end of option characters */
463 while (*end != ' ' && *end != '\0')
467 strncpy(save, start, end - start);
468 save[end - start] = '\0';
470 /* move remaining options up front */
472 memmove(start, end, strlen(end));
473 *(start + strlen(end)) = '\0';
475 /* append the specified options */
476 if (*(start + strlen(start) - 1) != ' ')
478 idx = start + strlen(start);
484 /* Build fs according to type */
485 int ldiskfs_make_lustre(struct mkfs_opts *mop)
487 __u64 device_sz = mop->mo_device_sz, block_count = 0;
488 char mkfs_cmd[PATH_MAX];
492 int ret = 0, ext_opts = 0;
495 if (!(mop->mo_flags & MO_IS_LOOP)) {
496 mop->mo_device_sz = get_device_size(mop->mo_device);
498 if (mop->mo_device_sz == 0)
501 /* Compare to real size */
502 if (device_sz == 0 || device_sz > mop->mo_device_sz)
503 device_sz = mop->mo_device_sz;
505 mop->mo_device_sz = device_sz;
508 if (mop->mo_device_sz != 0) {
509 if (mop->mo_device_sz < 8096){
510 fprintf(stderr, "%s: size of filesystem must be larger "
511 "than 8MB, but is set to %lldKB\n",
512 progname, (long long)mop->mo_device_sz);
515 block_count = mop->mo_device_sz / (L_BLOCK_SIZE >> 10);
516 /* If the LUN size is just over 2^32 blocks, limit the
517 * filesystem size to 2^32-1 blocks to avoid problems with
518 * ldiskfs/mkfs not handling this size. Bug 22906 */
519 if (block_count > 0xffffffffULL && block_count < 0x100002000ULL)
520 block_count = 0xffffffffULL;
523 if ((mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3) ||
524 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS) ||
525 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS2)) {
528 /* Journal size in MB */
529 if (strstr(mop->mo_mkfsopts, "-J") == NULL) {
530 /* Choose our own default journal size */
531 long journal_sz = 0, max_sz;
532 if (device_sz > 1024 * 1024) /* 1GB */
533 journal_sz = (device_sz / 102400) * 4;
534 /* cap journal size at 1GB */
535 if (journal_sz > 1024L)
538 max_sz = (102400 * L_BLOCK_SIZE) >> 20; /* 400MB */
539 if (journal_sz > max_sz)
542 sprintf(buf, " -J size=%ld", journal_sz);
543 strscat(mop->mo_mkfsopts, buf,
544 sizeof(mop->mo_mkfsopts));
548 /* Inode size (for extended attributes). The LOV EA size is
549 * 32 (EA hdr) + 32 (lov_mds_md) + stripes * 24 (lov_ost_data),
550 * and we want some margin above that for ACLs, other EAs... */
551 if (strstr(mop->mo_mkfsopts, "-I") == NULL) {
552 if (IS_MDT(&mop->mo_ldd)) {
553 if (mop->mo_stripe_count > 72)
554 inode_size = 512; /* bz 7241 */
555 /* see also "-i" below for EA blocks */
556 else if (mop->mo_stripe_count > 32)
558 else if (mop->mo_stripe_count > 10)
562 } else if (IS_OST(&mop->mo_ldd)) {
563 /* We store MDS FID and OST objid in EA on OST
564 * we need to make inode bigger as well. */
568 if (inode_size > 0) {
569 sprintf(buf, " -I %ld", inode_size);
570 strscat(mop->mo_mkfsopts, buf,
571 sizeof(mop->mo_mkfsopts));
575 /* Bytes_per_inode: disk size / num inodes */
576 if (strstr(mop->mo_mkfsopts, "-i") == NULL &&
577 strstr(mop->mo_mkfsopts, "-N") == NULL) {
578 long bytes_per_inode = 0;
580 /* Allocate more inodes on MDT devices. There is
581 * no data stored on the MDT, and very little extra
582 * metadata beyond the inode. It could go down as
583 * low as 1024 bytes, but this is conservative.
584 * Account for external EA blocks for wide striping. */
585 if (IS_MDT(&mop->mo_ldd)) {
586 bytes_per_inode = inode_size + 1536;
588 if (mop->mo_stripe_count > 72) {
589 int extra = mop->mo_stripe_count * 24;
590 extra = ((extra - 1) | 4095) + 1;
591 bytes_per_inode += extra;
595 /* Allocate fewer inodes on large OST devices. Most
596 * filesystems can be much more aggressive than even
597 * this, but it is impossible to know in advance. */
598 if (IS_OST(&mop->mo_ldd)) {
599 /* OST > 16TB assume average file size 1MB */
600 if (device_sz > (16ULL << 30))
601 bytes_per_inode = 1024 * 1024;
602 /* OST > 4TB assume average file size 512kB */
603 else if (device_sz > (4ULL << 30))
604 bytes_per_inode = 512 * 1024;
605 /* OST > 1TB assume average file size 256kB */
606 else if (device_sz > (1ULL << 30))
607 bytes_per_inode = 256 * 1024;
608 /* OST > 10GB assume average file size 64kB,
609 * plus a bit so that inodes will fit into a
610 * 256x flex_bg without overflowing */
611 else if (device_sz > (10ULL << 20))
612 bytes_per_inode = 69905;
615 if (bytes_per_inode > 0) {
616 sprintf(buf, " -i %ld", bytes_per_inode);
617 strscat(mop->mo_mkfsopts, buf,
618 sizeof(mop->mo_mkfsopts));
623 strscat(mop->mo_mkfsopts, " -q",
624 sizeof(mop->mo_mkfsopts));
627 /* start handle -O mkfs options */
628 if ((start = strstr(mop->mo_mkfsopts, "-O")) != NULL) {
629 if (strstr(start + 2, "-O") != NULL) {
631 "%s: don't specify multiple -O options\n",
635 start = moveopts_to_end(start);
636 maxbuflen = sizeof(mop->mo_mkfsopts) -
637 (start - mop->mo_mkfsopts) - strlen(start);
638 enable_default_ext4_features(mop, start, maxbuflen, 1);
640 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts),
641 maxbuflen = sizeof(mop->mo_mkfsopts) -
642 strlen(mop->mo_mkfsopts);
643 enable_default_ext4_features(mop, start, maxbuflen, 0);
645 /* end handle -O mkfs options */
647 /* start handle -E mkfs options */
648 if ((start = strstr(mop->mo_mkfsopts, "-E")) != NULL) {
649 if (strstr(start + 2, "-E") != NULL) {
651 "%s: don't specify multiple -E options\n",
655 start = moveopts_to_end(start);
656 maxbuflen = sizeof(mop->mo_mkfsopts) -
657 (start - mop->mo_mkfsopts) - strlen(start);
660 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts);
661 maxbuflen = sizeof(mop->mo_mkfsopts) -
662 strlen(mop->mo_mkfsopts);
665 /* In order to align the filesystem metadata on 1MB boundaries,
666 * give a resize value that will reserve a power-of-two group
667 * descriptor blocks, but leave one block for the superblock.
668 * Only useful for filesystems with < 2^32 blocks due to resize
670 if (IS_OST(&mop->mo_ldd) && mop->mo_device_sz > 100 * 1024 &&
671 mop->mo_device_sz * 1024 / L_BLOCK_SIZE <= 0xffffffffULL) {
672 unsigned group_blocks = L_BLOCK_SIZE * 8;
673 unsigned desc_per_block = L_BLOCK_SIZE / 32;
674 unsigned resize_blks;
676 resize_blks = (1ULL<<32) - desc_per_block*group_blocks;
677 snprintf(buf, sizeof(buf), "%u", resize_blks);
678 append_unique(start, ext_opts ? "," : " -E ",
679 "resize", buf, maxbuflen);
683 /* Avoid zeroing out the full journal - speeds up mkfs */
684 if (is_e2fsprogs_feature_supp("-E lazy_journal_init") == 0)
685 append_unique(start, ext_opts ? "," : " -E ",
686 "lazy_journal_init", NULL, maxbuflen);
687 /* end handle -E mkfs options */
689 /* Allow reformat of full devices (as opposed to
690 partitions.) We already checked for mounted dev. */
691 strscat(mop->mo_mkfsopts, " -F", sizeof(mop->mo_mkfsopts));
693 snprintf(mkfs_cmd, sizeof(mkfs_cmd),
694 "%s -j -b %d -L %s ", MKE2FS, L_BLOCK_SIZE,
695 mop->mo_ldd.ldd_svname);
697 fprintf(stderr,"%s: unsupported fs type: %d (%s)\n",
698 progname, mop->mo_ldd.ldd_mount_type,
699 MT_STR(&mop->mo_ldd));
703 /* For loop device format the dev, not the filename */
704 dev = mop->mo_device;
705 if (mop->mo_flags & MO_IS_LOOP)
706 dev = mop->mo_loopdev;
708 vprint("formatting backing filesystem %s on %s\n",
709 MT_STR(&mop->mo_ldd), dev);
710 vprint("\ttarget name %s\n", mop->mo_ldd.ldd_svname);
711 vprint("\t4k blocks "LPU64"\n", block_count);
712 vprint("\toptions %s\n", mop->mo_mkfsopts);
714 /* mkfs_cmd's trailing space is important! */
715 strscat(mkfs_cmd, mop->mo_mkfsopts, sizeof(mkfs_cmd));
716 strscat(mkfs_cmd, " ", sizeof(mkfs_cmd));
717 strscat(mkfs_cmd, dev, sizeof(mkfs_cmd));
718 if (block_count != 0) {
719 sprintf(buf, " "LPU64, block_count);
720 strscat(mkfs_cmd, buf, sizeof(mkfs_cmd));
723 vprint("mkfs_cmd = %s\n", mkfs_cmd);
724 ret = run_command(mkfs_cmd, sizeof(mkfs_cmd));
727 fprintf(stderr, "Unable to build fs %s (%d)\n", dev, ret);
732 int ldiskfs_prepare_lustre(struct mkfs_opts *mop,
733 char *default_mountopts, int default_len,
734 char *always_mountopts, int always_len)
736 struct lustre_disk_data *ldd = &mop->mo_ldd;
739 /* Set MO_IS_LOOP to indicate a loopback device is needed */
740 ret = is_block(mop->mo_device);
743 } else if (ret == 0) {
744 mop->mo_flags |= MO_IS_LOOP;
747 strscat(default_mountopts, ",errors=remount-ro", default_len);
748 if (IS_MDT(ldd) || IS_MGS(ldd))
749 strscat(always_mountopts, ",user_xattr", always_len);
754 int read_file(char *path, char *buf, int size)
758 fd = fopen(path, "r");
762 /* should not ignore fgets(3)'s return value */
763 if (!fgets(buf, size, fd)) {
764 fprintf(stderr, "reading from %s: %s", path, strerror(errno));
772 int write_file(char *path, char *buf)
776 fd = fopen(path, "w");
785 /* This is to tune the kernel for good SCSI performance.
786 * For that we set the value of /sys/block/{dev}/queue/max_sectors_kb
787 * to the value of /sys/block/{dev}/queue/max_hw_sectors_kb */
788 int set_blockdev_tunables(char *source, struct mount_opts *mop, int fan_out)
790 glob_t glob_info = { 0 };
791 struct stat stat_buf;
792 char *chk_major, *chk_minor;
793 char *savept = NULL, *dev;
795 char buf[PATH_MAX] = {'\0'}, path[PATH_MAX] = {'\0'};
796 char real_path[PATH_MAX] = {'\0'};
803 ret_path = realpath(source, real_path);
804 if (ret_path == NULL) {
806 fprintf(stderr, "warning: %s: cannot resolve: %s\n",
807 source, strerror(errno));
811 if (strncmp(real_path, "/dev/loop", 9) == 0)
814 if ((real_path[0] != '/') && (strpbrk(real_path, ",:") != NULL))
817 snprintf(path, sizeof(path), "/sys/block%s", real_path + 4);
818 if (access(path, X_OK) == 0)
821 /* The name of the device say 'X' specified in /dev/X may not
822 * match any entry under /sys/block/. In that case we need to
823 * match the major/minor number to find the entry under
824 * sys/block corresponding to /dev/X */
826 /* Don't chop tail digit on /dev/mapper/xxx, LU-478 */
827 if (strncmp(real_path, "/dev/mapper", 11) != 0) {
828 dev = real_path + strlen(real_path);
829 while (--dev > real_path && isdigit(*dev))
832 if (strncmp(real_path, "/dev/md_", 8) == 0)
836 rc = stat(real_path, &stat_buf);
839 fprintf(stderr, "warning: %s, device %s stat failed\n",
840 strerror(errno), real_path);
844 major = major(stat_buf.st_rdev);
845 minor = minor(stat_buf.st_rdev);
846 rc = glob("/sys/block/*", GLOB_NOSORT, NULL, &glob_info);
849 fprintf(stderr, "warning: failed to read entries under "
851 globfree(&glob_info);
855 for (i = 0; i < glob_info.gl_pathc; i++){
856 snprintf(path, sizeof(path), "%s/dev", glob_info.gl_pathv[i]);
858 rc = read_file(path, buf, sizeof(buf));
862 if (buf[strlen(buf) - 1] == '\n')
863 buf[strlen(buf) - 1] = '\0';
865 chk_major = strtok_r(buf, ":", &savept);
867 if (major == atoi(chk_major) &&minor == atoi(chk_minor))
871 if (i == glob_info.gl_pathc) {
873 fprintf(stderr,"warning: device %s does not match any "
874 "entry under /sys/block\n", real_path);
875 globfree(&glob_info);
879 /* Chop off "/dev" from path we found */
880 path[strlen(glob_info.gl_pathv[i])] = '\0';
881 globfree(&glob_info);
884 if (strncmp(real_path, "/dev/md", 7) == 0) {
885 snprintf(real_path, sizeof(real_path), "%s/%s", path,
888 rc = read_file(real_path, buf, sizeof(buf));
891 fprintf(stderr, "warning: opening %s: %s\n",
892 real_path, strerror(errno));
896 if (atoi(buf) >= mop->mo_md_stripe_cache_size)
899 if (strlen(buf) - 1 > 0) {
900 snprintf(buf, sizeof(buf), "%d",
901 mop->mo_md_stripe_cache_size);
902 rc = write_file(real_path, buf);
904 fprintf(stderr, "warning: opening %s: %s\n",
905 real_path, strerror(errno));
907 /* Return since raid and disk tunables are different */
911 snprintf(real_path, sizeof(real_path), "%s/%s", path,
912 MAX_HW_SECTORS_KB_PATH);
913 rc = read_file(real_path, buf, sizeof(buf));
916 fprintf(stderr, "warning: opening %s: %s\n",
917 real_path, strerror(errno));
918 /* No MAX_HW_SECTORS_KB_PATH isn't necessary an
919 * error for some device. */
923 if (strlen(buf) - 1 > 0) {
924 snprintf(real_path, sizeof(real_path), "%s/%s", path,
925 MAX_SECTORS_KB_PATH);
926 rc = write_file(real_path, buf);
929 fprintf(stderr, "warning: writing to %s: %s\n",
930 real_path, strerror(errno));
931 /* No MAX_SECTORS_KB_PATH isn't necessary an
932 * error for some device. */
939 glob_info.gl_pathc = 0;
940 glob_info.gl_offs = 0;
941 /* if device is multipath device, tune its slave devices */
942 snprintf(real_path, sizeof(real_path), "%s/slaves/*", path);
943 rc = glob(real_path, GLOB_NOSORT, NULL, &glob_info);
945 for (i = 0; rc == 0 && i < glob_info.gl_pathc; i++){
946 slave = basename(glob_info.gl_pathv[i]);
947 snprintf(real_path, sizeof(real_path), "/dev/%s", slave);
948 rc = set_blockdev_tunables(real_path, mop, 0);
951 if (rc == GLOB_NOMATCH) {
952 /* no slave device is not an error */
954 } else if (rc && verbose) {
956 fprintf(stderr, "warning: %s, failed to read"
957 " entries under %s/slaves\n",
958 strerror(errno), path);
960 fprintf(stderr, "unable to set tunables for"
961 " slave device %s (slave would be"
962 " unable to handle IO request from"
967 globfree(&glob_info);
973 int ldiskfs_tune_lustre(char *dev, struct mount_opts *mop)
975 return set_blockdev_tunables(dev, mop, 1);
978 int ldiskfs_label_lustre(struct mount_opts *mop)
980 char label_cmd[PATH_MAX];
983 snprintf(label_cmd, sizeof(label_cmd), E2LABEL" %s %s",
984 mop->mo_source, mop->mo_ldd.ldd_svname);
985 rc = run_command(label_cmd, sizeof(label_cmd));
990 /* return canonicalized absolute pathname, even if the target file does not
991 * exist, unlike realpath */
992 static char *absolute_path(char *devname)
994 char buf[PATH_MAX + 1];
998 path = malloc(PATH_MAX + 1);
1002 if (devname[0] != '/') {
1003 if (getcwd(buf, sizeof(buf) - 1) == NULL) {
1008 strcat(buf, devname);
1010 strcpy(buf, devname);
1012 /* truncate filename before calling realpath */
1013 ptr = strrchr(buf, '/');
1019 if (path != realpath(buf, path)) {
1023 /* add the filename back */
1025 strcat(path, ptr + 1);
1029 /* Determine if a device is a block device (as opposed to a file) */
1030 int is_block(char* devname)
1036 devpath = absolute_path(devname);
1037 if (devpath == NULL) {
1038 fprintf(stderr, "%s: failed to resolve path to %s\n",
1043 ret = access(devname, F_OK);
1045 if (strncmp(devpath, "/dev/", 5) == 0) {
1046 /* nobody sane wants to create a loopback file under
1047 * /dev. Let's just report the device doesn't exist */
1048 fprintf(stderr, "%s: %s apparently does not exist\n",
1056 ret = stat(devpath, &st);
1058 fprintf(stderr, "%s: cannot stat %s\n", progname, devpath);
1061 ret = S_ISBLK(st.st_mode);
1067 int ldiskfs_init(void)
1069 /* Required because full path to DEBUGFS is not specified */
1070 setenv("PATH", "/bin:/sbin:/usr/bin:/usr/sbin:/usr/local/bin", 0);
1075 void ldiskfs_fini(void)