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 */
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), 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), fp);
352 if (ret > 0 && strstr(supp_features,
353 strncmp(feature, "-O ", 3) ? feature : feature+3))
356 if ((fd = mkstemp(imgname)) < 0)
361 snprintf(cmd, sizeof(cmd), "%s -F %s %s 100 >/dev/null 2>&1",
362 MKE2FS, feature, imgname);
363 /* run_command() displays the output of mke2fs when it fails for
364 * some feature, so use system() directly */
373 * append_unique: append @key or @key=@val pair to @buf only if @key does not
375 * @buf: buffer to hold @key or @key=@val
376 * @prefix: prefix string before @key
378 * @val: value string if it's a @key=@val pair
380 static void append_unique(char *buf, char *prefix, char *key, char *val,
389 anchor = end = strstr(buf, key);
390 /* try to find exact match string in @buf */
391 while (end && *end != '\0' && *end != ',' && *end != ' ' && *end != '=')
394 if (anchor == NULL || strlen(key) != len ||
395 strncmp(anchor, key, len) != 0) {
397 strscat(buf, prefix, maxbuflen);
399 strscat(buf, key, maxbuflen);
401 strscat(buf, "=", maxbuflen);
402 strscat(buf, val, maxbuflen);
407 static int enable_default_ext4_features(struct mkfs_opts *mop, char *anchor,
408 size_t maxbuflen, int user_spec)
410 if (IS_OST(&mop->mo_ldd)) {
411 append_unique(anchor, user_spec ? "," : " -O ",
412 "extents", NULL, sizeof(mop->mo_mkfsopts));
413 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
414 } else if (IS_MDT(&mop->mo_ldd)) {
415 append_unique(anchor, user_spec ? "," : " -O ",
416 "dirdata", NULL, maxbuflen);
417 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
418 append_unique(anchor, ",", "^extents", NULL, maxbuflen);
420 append_unique(anchor, user_spec ? "," : " -O ",
421 "uninit_bg", NULL, maxbuflen);
424 /* Multiple mount protection enabled only if failover node specified */
425 if (mop->mo_flags & MO_FAILOVER) {
426 if (is_e2fsprogs_feature_supp("-O mmp") == 0)
427 append_unique(anchor, ",", "mmp", NULL, maxbuflen);
429 disp_old_e2fsprogs_msg("mmp", 1);
432 /* Allow more than 65000 subdirectories */
433 if (is_e2fsprogs_feature_supp("-O dir_nlink") == 0)
434 append_unique(anchor, ",", "dir_nlink", NULL, maxbuflen);
436 /* The following options are only valid for ext4-based ldiskfs.
437 * If --backfstype=ext3 is specified, do not enable them. */
438 if (mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3)
441 /* Enable quota by default */
442 if (is_e2fsprogs_feature_supp("-O quota") == 0) {
443 append_unique(anchor, ",", "quota", NULL, maxbuflen);
446 fprintf(stderr, "\"-O quota\" must be supported by "
447 "e2fsprogs, please upgrade your e2fsprogs.\n");
451 /* Allow files larger than 2TB. Also needs LU-16, but not harmful. */
452 if (is_e2fsprogs_feature_supp("-O huge_file") == 0)
453 append_unique(anchor, ",", "huge_file", NULL, maxbuflen);
455 /* Enable large block addresses if the LUN is over 2^32 blocks. */
456 if (mop->mo_device_sz / (L_BLOCK_SIZE >> 10) >= 0x100002000ULL &&
457 is_e2fsprogs_feature_supp("-O 64bit") == 0)
458 append_unique(anchor, ",", "64bit", NULL, maxbuflen);
460 /* Cluster inode/block bitmaps and inode table for more efficient IO.
461 * Align the flex groups on a 1MB boundary for better performance. */
462 /* This -O feature needs to go last, since it adds the "-G" option. */
463 if (is_e2fsprogs_feature_supp("-O flex_bg") == 0) {
466 append_unique(anchor, ",", "flex_bg", NULL, maxbuflen);
468 if (IS_OST(&mop->mo_ldd)) {
469 snprintf(tmp_buf, sizeof(tmp_buf), " -G %u",
470 (1 << 20) / L_BLOCK_SIZE);
471 strscat(anchor, tmp_buf, maxbuflen);
474 /* Don't add any more "-O" options here, see last comment above */
479 * moveopts_to_end: find the option string, move remaining strings to
480 * where option string starts, and append the option
482 * @start: where the option string starts before the move
483 * RETURN: where the option string starts after the move
485 static char *moveopts_to_end(char *start)
490 /* skip whitespace before options */
495 /* find end of option characters */
496 while (*end != ' ' && *end != '\0')
500 strncpy(save, start, end - start);
501 save[end - start] = '\0';
503 /* move remaining options up front */
505 memmove(start, end, strlen(end));
506 *(start + strlen(end)) = '\0';
508 /* append the specified options */
509 if (*(start + strlen(start) - 1) != ' ')
511 idx = start + strlen(start);
517 /* Build fs according to type */
518 int ldiskfs_make_lustre(struct mkfs_opts *mop)
520 __u64 device_sz = mop->mo_device_sz, block_count = 0;
521 char mkfs_cmd[PATH_MAX];
525 int ret = 0, ext_opts = 0;
528 if (!(mop->mo_flags & MO_IS_LOOP)) {
529 mop->mo_device_sz = get_device_size(mop->mo_device);
531 if (mop->mo_device_sz == 0)
534 /* Compare to real size */
535 if (device_sz == 0 || device_sz > mop->mo_device_sz)
536 device_sz = mop->mo_device_sz;
538 mop->mo_device_sz = device_sz;
541 if (mop->mo_device_sz != 0) {
542 if (mop->mo_device_sz < 8096){
543 fprintf(stderr, "%s: size of filesystem must be larger "
544 "than 8MB, but is set to %lldKB\n",
545 progname, (long long)mop->mo_device_sz);
548 block_count = mop->mo_device_sz / (L_BLOCK_SIZE >> 10);
549 /* If the LUN size is just over 2^32 blocks, limit the
550 * filesystem size to 2^32-1 blocks to avoid problems with
551 * ldiskfs/mkfs not handling this size. Bug 22906 */
552 if (block_count > 0xffffffffULL && block_count < 0x100002000ULL)
553 block_count = 0xffffffffULL;
556 if ((mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3) ||
557 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS) ||
558 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS2)) {
561 /* Journal size in MB */
562 if (strstr(mop->mo_mkfsopts, "-J") == NULL) {
563 /* Choose our own default journal size */
564 long journal_sz = 0, max_sz;
565 if (device_sz > 1024 * 1024) /* 1GB */
566 journal_sz = (device_sz / 102400) * 4;
567 /* cap journal size at 1GB */
568 if (journal_sz > 1024L)
571 max_sz = (102400 * L_BLOCK_SIZE) >> 20; /* 400MB */
572 if (journal_sz > max_sz)
575 sprintf(buf, " -J size=%ld", journal_sz);
576 strscat(mop->mo_mkfsopts, buf,
577 sizeof(mop->mo_mkfsopts));
581 /* Inode size (for extended attributes). The LOV EA size is
582 * 32 (EA hdr) + 32 (lov_mds_md) + stripes * 24 (lov_ost_data),
583 * and we want some margin above that for ACLs, other EAs... */
584 if (strstr(mop->mo_mkfsopts, "-I") == NULL) {
585 if (IS_MDT(&mop->mo_ldd)) {
586 if (mop->mo_stripe_count > 72)
587 inode_size = 512; /* bz 7241 */
588 /* see also "-i" below for EA blocks */
589 else if (mop->mo_stripe_count > 32)
591 else if (mop->mo_stripe_count > 10)
595 } else if (IS_OST(&mop->mo_ldd)) {
596 /* We store MDS FID and OST objid in EA on OST
597 * we need to make inode bigger as well. */
601 if (inode_size > 0) {
602 sprintf(buf, " -I %ld", inode_size);
603 strscat(mop->mo_mkfsopts, buf,
604 sizeof(mop->mo_mkfsopts));
608 /* Bytes_per_inode: disk size / num inodes */
609 if (strstr(mop->mo_mkfsopts, "-i") == NULL &&
610 strstr(mop->mo_mkfsopts, "-N") == NULL) {
611 long bytes_per_inode = 0;
613 /* Allocate more inodes on MDT devices. There is
614 * no data stored on the MDT, and very little extra
615 * metadata beyond the inode. It could go down as
616 * low as 1024 bytes, but this is conservative.
617 * Account for external EA blocks for wide striping. */
618 if (IS_MDT(&mop->mo_ldd)) {
619 bytes_per_inode = inode_size + 1536;
621 if (mop->mo_stripe_count > 72) {
622 int extra = mop->mo_stripe_count * 24;
623 extra = ((extra - 1) | 4095) + 1;
624 bytes_per_inode += extra;
628 /* Allocate fewer inodes on large OST devices. Most
629 * filesystems can be much more aggressive than even
630 * this, but it is impossible to know in advance. */
631 if (IS_OST(&mop->mo_ldd)) {
632 /* OST > 16TB assume average file size 1MB */
633 if (device_sz > (16ULL << 30))
634 bytes_per_inode = 1024 * 1024;
635 /* OST > 4TB assume average file size 512kB */
636 else if (device_sz > (4ULL << 30))
637 bytes_per_inode = 512 * 1024;
638 /* OST > 1TB assume average file size 256kB */
639 else if (device_sz > (1ULL << 30))
640 bytes_per_inode = 256 * 1024;
641 /* OST > 10GB assume average file size 64kB,
642 * plus a bit so that inodes will fit into a
643 * 256x flex_bg without overflowing */
644 else if (device_sz > (10ULL << 20))
645 bytes_per_inode = 69905;
648 if (bytes_per_inode > 0) {
649 sprintf(buf, " -i %ld", bytes_per_inode);
650 strscat(mop->mo_mkfsopts, buf,
651 sizeof(mop->mo_mkfsopts));
656 strscat(mop->mo_mkfsopts, " -q",
657 sizeof(mop->mo_mkfsopts));
660 /* start handle -O mkfs options */
661 if ((start = strstr(mop->mo_mkfsopts, "-O")) != NULL) {
662 if (strstr(start + 2, "-O") != NULL) {
664 "%s: don't specify multiple -O options\n",
668 start = moveopts_to_end(start);
669 maxbuflen = sizeof(mop->mo_mkfsopts) -
670 (start - mop->mo_mkfsopts) - strlen(start);
671 ret = enable_default_ext4_features(mop, start, maxbuflen, 1);
673 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts),
674 maxbuflen = sizeof(mop->mo_mkfsopts) -
675 strlen(mop->mo_mkfsopts);
676 ret = enable_default_ext4_features(mop, start, maxbuflen, 0);
680 /* end handle -O mkfs options */
682 /* start handle -E mkfs options */
683 if ((start = strstr(mop->mo_mkfsopts, "-E")) != NULL) {
684 if (strstr(start + 2, "-E") != NULL) {
686 "%s: don't specify multiple -E options\n",
690 start = moveopts_to_end(start);
691 maxbuflen = sizeof(mop->mo_mkfsopts) -
692 (start - mop->mo_mkfsopts) - strlen(start);
695 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts);
696 maxbuflen = sizeof(mop->mo_mkfsopts) -
697 strlen(mop->mo_mkfsopts);
700 /* In order to align the filesystem metadata on 1MB boundaries,
701 * give a resize value that will reserve a power-of-two group
702 * descriptor blocks, but leave one block for the superblock.
703 * Only useful for filesystems with < 2^32 blocks due to resize
705 if (IS_OST(&mop->mo_ldd) && mop->mo_device_sz > 100 * 1024 &&
706 mop->mo_device_sz * 1024 / L_BLOCK_SIZE <= 0xffffffffULL) {
707 unsigned group_blocks = L_BLOCK_SIZE * 8;
708 unsigned desc_per_block = L_BLOCK_SIZE / 32;
709 unsigned resize_blks;
711 resize_blks = (1ULL<<32) - desc_per_block*group_blocks;
712 snprintf(buf, sizeof(buf), "%u", resize_blks);
713 append_unique(start, ext_opts ? "," : " -E ",
714 "resize", buf, maxbuflen);
718 /* Avoid zeroing out the full journal - speeds up mkfs */
719 if (is_e2fsprogs_feature_supp("-E lazy_journal_init") == 0)
720 append_unique(start, ext_opts ? "," : " -E ",
721 "lazy_journal_init", NULL, maxbuflen);
722 /* end handle -E mkfs options */
724 /* Allow reformat of full devices (as opposed to
725 partitions.) We already checked for mounted dev. */
726 strscat(mop->mo_mkfsopts, " -F", sizeof(mop->mo_mkfsopts));
728 snprintf(mkfs_cmd, sizeof(mkfs_cmd),
729 "%s -j -b %d -L %s ", MKE2FS, L_BLOCK_SIZE,
730 mop->mo_ldd.ldd_svname);
732 fprintf(stderr,"%s: unsupported fs type: %d (%s)\n",
733 progname, mop->mo_ldd.ldd_mount_type,
734 MT_STR(&mop->mo_ldd));
738 /* For loop device format the dev, not the filename */
739 dev = mop->mo_device;
740 if (mop->mo_flags & MO_IS_LOOP)
741 dev = mop->mo_loopdev;
743 vprint("formatting backing filesystem %s on %s\n",
744 MT_STR(&mop->mo_ldd), dev);
745 vprint("\ttarget name %s\n", mop->mo_ldd.ldd_svname);
746 vprint("\t4k blocks "LPU64"\n", block_count);
747 vprint("\toptions %s\n", mop->mo_mkfsopts);
749 /* mkfs_cmd's trailing space is important! */
750 strscat(mkfs_cmd, mop->mo_mkfsopts, sizeof(mkfs_cmd));
751 strscat(mkfs_cmd, " ", sizeof(mkfs_cmd));
752 strscat(mkfs_cmd, dev, sizeof(mkfs_cmd));
753 if (block_count != 0) {
754 sprintf(buf, " "LPU64, block_count);
755 strscat(mkfs_cmd, buf, sizeof(mkfs_cmd));
758 vprint("mkfs_cmd = %s\n", mkfs_cmd);
759 ret = run_command(mkfs_cmd, sizeof(mkfs_cmd));
762 fprintf(stderr, "Unable to build fs %s (%d)\n", dev, ret);
767 int ldiskfs_prepare_lustre(struct mkfs_opts *mop,
768 char *default_mountopts, int default_len,
769 char *always_mountopts, int always_len)
771 struct lustre_disk_data *ldd = &mop->mo_ldd;
774 /* Set MO_IS_LOOP to indicate a loopback device is needed */
775 ret = is_block(mop->mo_device);
778 } else if (ret == 0) {
779 mop->mo_flags |= MO_IS_LOOP;
782 strscat(default_mountopts, ",errors=remount-ro", default_len);
783 if (IS_MDT(ldd) || IS_MGS(ldd))
784 strscat(always_mountopts, ",user_xattr", always_len);
789 int read_file(char *path, char *buf, int size)
793 fd = fopen(path, "r");
797 /* should not ignore fgets(3)'s return value */
798 if (!fgets(buf, size, fd)) {
799 fprintf(stderr, "reading from %s: %s", path, strerror(errno));
807 int write_file(char *path, char *buf)
811 fd = fopen(path, "w");
820 /* This is to tune the kernel for good SCSI performance.
821 * For that we set the value of /sys/block/{dev}/queue/max_sectors_kb
822 * to the value of /sys/block/{dev}/queue/max_hw_sectors_kb */
823 int set_blockdev_tunables(char *source, struct mount_opts *mop, int fan_out)
825 glob_t glob_info = { 0 };
826 struct stat stat_buf;
827 char *chk_major, *chk_minor;
828 char *savept = NULL, *dev;
830 char buf[PATH_MAX] = {'\0'}, path[PATH_MAX] = {'\0'};
831 char real_path[PATH_MAX] = {'\0'};
838 ret_path = realpath(source, real_path);
839 if (ret_path == NULL) {
841 fprintf(stderr, "warning: %s: cannot resolve: %s\n",
842 source, strerror(errno));
846 if (strncmp(real_path, "/dev/loop", 9) == 0)
849 if ((real_path[0] != '/') && (strpbrk(real_path, ",:") != NULL))
852 snprintf(path, sizeof(path), "/sys/block%s", real_path + 4);
853 if (access(path, X_OK) == 0)
856 /* The name of the device say 'X' specified in /dev/X may not
857 * match any entry under /sys/block/. In that case we need to
858 * match the major/minor number to find the entry under
859 * sys/block corresponding to /dev/X */
861 /* Don't chop tail digit on /dev/mapper/xxx, LU-478 */
862 if (strncmp(real_path, "/dev/mapper", 11) != 0) {
863 dev = real_path + strlen(real_path);
864 while (--dev > real_path && isdigit(*dev))
867 if (strncmp(real_path, "/dev/md_", 8) == 0)
871 rc = stat(real_path, &stat_buf);
874 fprintf(stderr, "warning: %s, device %s stat failed\n",
875 strerror(errno), real_path);
879 major = major(stat_buf.st_rdev);
880 minor = minor(stat_buf.st_rdev);
881 rc = glob("/sys/block/*", GLOB_NOSORT, NULL, &glob_info);
884 fprintf(stderr, "warning: failed to read entries under "
886 globfree(&glob_info);
890 for (i = 0; i < glob_info.gl_pathc; i++){
891 snprintf(path, sizeof(path), "%s/dev", glob_info.gl_pathv[i]);
893 rc = read_file(path, buf, sizeof(buf));
897 if (buf[strlen(buf) - 1] == '\n')
898 buf[strlen(buf) - 1] = '\0';
900 chk_major = strtok_r(buf, ":", &savept);
902 if (major == atoi(chk_major) &&minor == atoi(chk_minor))
906 if (i == glob_info.gl_pathc) {
908 fprintf(stderr,"warning: device %s does not match any "
909 "entry under /sys/block\n", real_path);
910 globfree(&glob_info);
914 /* Chop off "/dev" from path we found */
915 path[strlen(glob_info.gl_pathv[i])] = '\0';
916 globfree(&glob_info);
919 if (strncmp(real_path, "/dev/md", 7) == 0) {
920 snprintf(real_path, sizeof(real_path), "%s/%s", path,
923 rc = read_file(real_path, buf, sizeof(buf));
926 fprintf(stderr, "warning: opening %s: %s\n",
927 real_path, strerror(errno));
931 if (atoi(buf) >= mop->mo_md_stripe_cache_size)
934 if (strlen(buf) - 1 > 0) {
935 snprintf(buf, sizeof(buf), "%d",
936 mop->mo_md_stripe_cache_size);
937 rc = write_file(real_path, buf);
939 fprintf(stderr, "warning: opening %s: %s\n",
940 real_path, strerror(errno));
942 /* Return since raid and disk tunables are different */
946 snprintf(real_path, sizeof(real_path), "%s/%s", path,
947 MAX_HW_SECTORS_KB_PATH);
948 rc = read_file(real_path, buf, sizeof(buf));
951 fprintf(stderr, "warning: opening %s: %s\n",
952 real_path, strerror(errno));
953 /* No MAX_HW_SECTORS_KB_PATH isn't necessary an
954 * error for some device. */
958 if (strlen(buf) - 1 > 0) {
959 snprintf(real_path, sizeof(real_path), "%s/%s", path,
960 MAX_SECTORS_KB_PATH);
961 rc = write_file(real_path, buf);
964 fprintf(stderr, "warning: writing to %s: %s\n",
965 real_path, strerror(errno));
966 /* No MAX_SECTORS_KB_PATH isn't necessary an
967 * error for some device. */
974 glob_info.gl_pathc = 0;
975 glob_info.gl_offs = 0;
976 /* if device is multipath device, tune its slave devices */
977 snprintf(real_path, sizeof(real_path), "%s/slaves/*", path);
978 rc = glob(real_path, GLOB_NOSORT, NULL, &glob_info);
980 for (i = 0; rc == 0 && i < glob_info.gl_pathc; i++){
981 slave = basename(glob_info.gl_pathv[i]);
982 snprintf(real_path, sizeof(real_path), "/dev/%s", slave);
983 rc = set_blockdev_tunables(real_path, mop, 0);
986 if (rc == GLOB_NOMATCH) {
987 /* no slave device is not an error */
989 } else if (rc && verbose) {
991 fprintf(stderr, "warning: %s, failed to read"
992 " entries under %s/slaves\n",
993 strerror(errno), path);
995 fprintf(stderr, "unable to set tunables for"
996 " slave device %s (slave would be"
997 " unable to handle IO request from"
1002 globfree(&glob_info);
1008 int ldiskfs_tune_lustre(char *dev, struct mount_opts *mop)
1010 return set_blockdev_tunables(dev, mop, 1);
1013 int ldiskfs_label_lustre(struct mount_opts *mop)
1015 char label_cmd[PATH_MAX];
1018 snprintf(label_cmd, sizeof(label_cmd), E2LABEL" %s %s",
1019 mop->mo_source, mop->mo_ldd.ldd_svname);
1020 rc = run_command(label_cmd, sizeof(label_cmd));
1025 /* return canonicalized absolute pathname, even if the target file does not
1026 * exist, unlike realpath */
1027 static char *absolute_path(char *devname)
1029 char buf[PATH_MAX + 1];
1033 path = malloc(PATH_MAX + 1);
1037 if (devname[0] != '/') {
1038 if (getcwd(buf, sizeof(buf) - 1) == NULL) {
1043 strcat(buf, devname);
1045 strcpy(buf, devname);
1047 /* truncate filename before calling realpath */
1048 ptr = strrchr(buf, '/');
1054 if (path != realpath(buf, path)) {
1058 /* add the filename back */
1060 strcat(path, ptr + 1);
1064 /* Determine if a device is a block device (as opposed to a file) */
1065 int is_block(char* devname)
1071 devpath = absolute_path(devname);
1072 if (devpath == NULL) {
1073 fprintf(stderr, "%s: failed to resolve path to %s\n",
1078 ret = access(devname, F_OK);
1080 if (strncmp(devpath, "/dev/", 5) == 0) {
1081 /* nobody sane wants to create a loopback file under
1082 * /dev. Let's just report the device doesn't exist */
1083 fprintf(stderr, "%s: %s apparently does not exist\n",
1091 ret = stat(devpath, &st);
1093 fprintf(stderr, "%s: cannot stat %s\n", progname, devpath);
1096 ret = S_ISBLK(st.st_mode);
1102 static int is_feature_enabled(const char *feature, const char *devpath)
1106 char enabled_features[4096] = "";
1108 snprintf(cmd, sizeof(cmd), "%s -R features %s 2>&1",
1111 /* Using popen() instead of run_command() since debugfs does
1112 * not return proper error code if command is not supported */
1113 fp = popen(cmd, "r");
1115 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
1119 fread(enabled_features, 1, sizeof(enabled_features), fp);
1122 if (strstr(enabled_features, feature))
1127 /* Enable quota accounting */
1128 int ldiskfs_enable_quota(struct mkfs_opts *mop)
1132 int cmdsz = sizeof(cmd), ret;
1134 if (is_e2fsprogs_feature_supp("-O quota") != 0) {
1135 fprintf(stderr, "%s: \"-O quota\" is is not supported by "
1136 "current e2fsprogs\n", progname);
1140 dev = mop->mo_device;
1141 if (mop->mo_flags & MO_IS_LOOP)
1142 dev = mop->mo_loopdev;
1144 /* Quota feature is already enabled? */
1145 if (is_feature_enabled("quota", dev)) {
1146 vprint("Quota feature is already enabled.\n");
1150 /* Turn on quota feature by "tune2fs -O quota" */
1151 snprintf(cmd, cmdsz, "%s -O quota %s", TUNE2FS, dev);
1152 ret = run_command(cmd, cmdsz);
1154 fprintf(stderr, "command:%s (%d)", cmd, ret);
1159 int ldiskfs_init(void)
1161 /* Required because full path to DEBUGFS is not specified */
1162 setenv("PATH", "/bin:/sbin:/usr/bin:/usr/sbin:/usr/local/bin", 0);
1167 void ldiskfs_fini(void)