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, 2013, 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>
84 #include <selinux/selinux.h>
87 #define MAX_HW_SECTORS_KB_PATH "queue/max_hw_sectors_kb"
88 #define MAX_SECTORS_KB_PATH "queue/max_sectors_kb"
89 #define SCHEDULER_PATH "queue/scheduler"
90 #define STRIPE_CACHE_SIZE "md/stripe_cache_size"
92 #define DEFAULT_SCHEDULER "deadline"
94 extern char *progname;
96 #define L_BLOCK_SIZE 4096
97 /* keep it less than LL_FID_NAMELEN */
98 #define DUMMY_FILE_NAME_LEN 25
99 #define EXT3_DIRENT_SIZE DUMMY_FILE_NAME_LEN
102 * Concatenate context of the temporary mount point iff selinux is enabled
105 void append_context_for_mount(char *mntpt, struct mkfs_opts *mop)
107 security_context_t fcontext;
109 if (getfilecon(mntpt, &fcontext) < 0) {
110 /* Continuing with default behaviour */
111 fprintf(stderr, "%s: Get file context failed : %s\n",
112 progname, strerror(errno));
116 if (fcontext != NULL) {
117 strcat(mop->mo_ldd.ldd_mount_opts, ",context=");
118 strcat(mop->mo_ldd.ldd_mount_opts, fcontext);
124 /* Write the server config files */
125 int ldiskfs_write_ldd(struct mkfs_opts *mop)
127 char mntpt[] = "/tmp/mntXXXXXX";
134 /* Mount this device temporarily in order to write these files */
135 if (!mkdtemp(mntpt)) {
136 fprintf(stderr, "%s: Can't create temp mount point %s: %s\n",
137 progname, mntpt, strerror(errno));
142 * Append file context to mount options if SE Linux is enabled
145 if (is_selinux_enabled() > 0)
146 append_context_for_mount(mntpt, mop);
149 dev = mop->mo_device;
150 if (mop->mo_flags & MO_IS_LOOP)
151 dev = mop->mo_loopdev;
153 ret = mount(dev, mntpt, MT_STR(&mop->mo_ldd), 0,
154 mop->mo_ldd.ldd_mount_opts);
156 fprintf(stderr, "%s: Unable to mount %s: %s\n",
157 progname, dev, strerror(errno));
159 if (errno == ENODEV) {
160 fprintf(stderr, "Is the %s module available?\n",
161 MT_STR(&mop->mo_ldd));
166 /* Set up initial directories */
167 sprintf(filepnm, "%s/%s", mntpt, MOUNT_CONFIGS_DIR);
168 ret = mkdir(filepnm, 0777);
169 if ((ret != 0) && (errno != EEXIST)) {
170 fprintf(stderr, "%s: Can't make configs dir %s (%s)\n",
171 progname, filepnm, strerror(errno));
173 } else if (errno == EEXIST) {
177 /* Save the persistent mount data into a file. Lustre must pre-read
178 this file to get the real mount options. */
179 vprint("Writing %s\n", MOUNT_DATA_FILE);
180 sprintf(filepnm, "%s/%s", mntpt, MOUNT_DATA_FILE);
181 filep = fopen(filepnm, "w");
183 fprintf(stderr, "%s: Unable to create %s file: %s\n",
184 progname, filepnm, strerror(errno));
187 num = fwrite(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
188 if (num < 1 && ferror(filep)) {
189 fprintf(stderr, "%s: Unable to write to file (%s): %s\n",
190 progname, filepnm, strerror(errno));
203 static int readcmd(char *cmd, char *buf, int len)
208 fp = popen(cmd, "r");
212 red = fread(buf, 1, len, fp);
215 /* strip trailing newline */
216 if (buf[red - 1] == '\n')
219 return (red == 0) ? -ENOENT : 0;
222 int ldiskfs_read_ldd(char *dev, struct lustre_disk_data *mo_ldd)
224 char tmpdir[] = "/tmp/dirXXXXXX";
229 int cmdsz = sizeof(cmd);
231 /* Make a temporary directory to hold Lustre data files. */
232 if (!mkdtemp(tmpdir)) {
233 fprintf(stderr, "%s: Can't create temporary directory %s: %s\n",
234 progname, tmpdir, strerror(errno));
238 /* TODO: it's worth observing the get_mountdata() function that is
239 in mount_utils.c for getting the mountdata out of the
242 /* Construct debugfs command line. */
243 snprintf(cmd, cmdsz, "%s -c -R 'dump /%s %s/mountdata' '%s'",
244 DEBUGFS, MOUNT_DATA_FILE, tmpdir, dev);
246 ret = run_command(cmd, cmdsz);
248 verrprint("%s: Unable to dump %s dir (%d)\n",
249 progname, MOUNT_CONFIGS_DIR, ret);
251 sprintf(filepnm, "%s/mountdata", tmpdir);
252 filep = fopen(filepnm, "r");
255 vprint("Reading %s\n", MOUNT_DATA_FILE);
256 num_read = fread(mo_ldd, sizeof(*mo_ldd), 1, filep);
257 if (num_read < 1 && ferror(filep)) {
258 fprintf(stderr, "%s: Unable to read from file %s: %s\n",
259 progname, filepnm, strerror(errno));
264 snprintf(cmd, cmdsz, "rm -rf %s", tmpdir);
265 run_command(cmd, cmdsz);
267 verrprint("Failed to read old data (%d)\n", ret);
269 /* As long as we at least have the label, we're good to go */
270 snprintf(cmd, sizeof(cmd), E2LABEL" %s", dev);
271 ret = readcmd(cmd, mo_ldd->ldd_svname, sizeof(mo_ldd->ldd_svname) - 1);
277 /* Display the need for the latest e2fsprogs to be installed. make_backfs
278 * indicates if the caller is make_lustre_backfs() or not. */
279 void disp_old_e2fsprogs_msg(const char *feature, int make_backfs)
281 static int msg_displayed;
284 fprintf(stderr, "WARNING: %s does not support %s "
285 "feature.\n\n", E2FSPROGS, feature);
291 fprintf(stderr, "WARNING: The %s package currently installed on "
292 "your system does not support \"%s\" feature.\n",
294 #if !(HAVE_LDISKFSPROGS)
295 fprintf(stderr, "Please install the latest version of e2fsprogs from\n"
296 "http://downloads.whamcloud.com/public/e2fsprogs/latest/\n"
297 "to enable this feature.\n");
300 fprintf(stderr, "Feature will not be enabled until %s"
301 "is updated and '%s -O %s %%{device}' "
302 "is run.\n\n", E2FSPROGS, TUNE2FS, feature);
305 /* Check whether the file exists in the device */
306 static int file_in_dev(char *file_name, char *dev_name)
309 char debugfs_cmd[256];
310 unsigned int inode_num;
313 /* Construct debugfs command line. */
314 snprintf(debugfs_cmd, sizeof(debugfs_cmd),
315 "%s -c -R 'stat %s' '%s' 2>&1 | egrep '(Inode|unsupported)'",
316 DEBUGFS, file_name, dev_name);
318 fp = popen(debugfs_cmd, "r");
320 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
324 if (fscanf(fp, "Inode: %u", &inode_num) == 1) { /* exist */
328 i = fread(debugfs_cmd, 1, sizeof(debugfs_cmd) - 1, fp);
331 fprintf(stderr, "%s", debugfs_cmd);
332 if (strstr(debugfs_cmd, "unsupported feature")) {
333 disp_old_e2fsprogs_msg("an unknown", 0);
342 /* Check whether the device has already been used with lustre */
343 int ldiskfs_is_lustre(char *dev, unsigned *mount_type)
347 ret = file_in_dev(MOUNT_DATA_FILE, dev);
349 /* in the -1 case, 'extents' means IS a lustre target */
350 *mount_type = LDD_MT_LDISKFS;
354 ret = file_in_dev(LAST_RCVD, dev);
356 *mount_type = LDD_MT_LDISKFS;
363 /* Check if a certain feature is supported by e2fsprogs.
364 * Firstly we try to use "debugfs supported_features" command to check if
365 * the feature is supported. If this fails we try to set this feature with
366 * mke2fs to check for its support. */
367 static int is_e2fsprogs_feature_supp(const char *feature)
369 static char supp_features[4096] = "";
372 char imgname[] = "/tmp/test-img-XXXXXX";
376 if (supp_features[0] == '\0') {
377 snprintf(cmd, sizeof(cmd), "%s -c -R supported_features 2>&1",
380 /* Using popen() instead of run_command() since debugfs does
381 * not return proper error code if command is not supported */
382 fp = popen(cmd, "r");
384 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
387 ret = fread(supp_features, 1, sizeof(supp_features) - 1, fp);
388 supp_features[ret] = '\0';
391 if (ret > 0 && strstr(supp_features,
392 strncmp(feature, "-O ", 3) ? feature : feature+3))
395 if ((fd = mkstemp(imgname)) < 0)
400 snprintf(cmd, sizeof(cmd), "%s -F %s %s 100 >/dev/null 2>&1",
401 MKE2FS, feature, imgname);
402 /* run_command() displays the output of mke2fs when it fails for
403 * some feature, so use system() directly */
412 * append_unique: append @key or @key=@val pair to @buf only if @key does not
414 * @buf: buffer to hold @key or @key=@val
415 * @prefix: prefix string before @key
417 * @val: value string if it's a @key=@val pair
419 static void append_unique(char *buf, char *prefix, char *key, char *val,
428 anchor = end = strstr(buf, key);
429 /* try to find exact match string in @buf */
430 while (end && *end != '\0' && *end != ',' && *end != ' ' && *end != '=')
433 if (anchor == NULL || strlen(key) != len ||
434 strncmp(anchor, key, len) != 0) {
436 strscat(buf, prefix, maxbuflen);
438 strscat(buf, key, maxbuflen);
440 strscat(buf, "=", maxbuflen);
441 strscat(buf, val, maxbuflen);
446 static int enable_default_ext4_features(struct mkfs_opts *mop, char *anchor,
447 size_t maxbuflen, int user_spec)
449 if (IS_OST(&mop->mo_ldd)) {
450 append_unique(anchor, user_spec ? "," : " -O ",
451 "extents", NULL, sizeof(mop->mo_mkfsopts));
452 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
453 } else if (IS_MDT(&mop->mo_ldd)) {
454 append_unique(anchor, user_spec ? "," : " -O ",
455 "dirdata", NULL, maxbuflen);
456 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
457 append_unique(anchor, ",", "^extents", NULL, maxbuflen);
459 append_unique(anchor, user_spec ? "," : " -O ",
460 "uninit_bg", NULL, maxbuflen);
463 /* Multiple mount protection enabled only if failover node specified */
464 if (mop->mo_flags & MO_FAILOVER) {
465 if (is_e2fsprogs_feature_supp("-O mmp") == 0)
466 append_unique(anchor, ",", "mmp", NULL, maxbuflen);
468 disp_old_e2fsprogs_msg("mmp", 1);
471 /* Allow more than 65000 subdirectories */
472 if (is_e2fsprogs_feature_supp("-O dir_nlink") == 0)
473 append_unique(anchor, ",", "dir_nlink", NULL, maxbuflen);
475 /* The following options are only valid for ext4-based ldiskfs.
476 * If --backfstype=ext3 is specified, do not enable them. */
477 if (mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3)
480 /* Enable quota by default */
481 if (is_e2fsprogs_feature_supp("-O quota") == 0) {
482 append_unique(anchor, ",", "quota", NULL, maxbuflen);
485 fprintf(stderr, "\"-O quota\" must be supported by "
486 "e2fsprogs, please upgrade your e2fsprogs.\n");
490 /* Allow files larger than 2TB. Also needs LU-16, but not harmful. */
491 if (is_e2fsprogs_feature_supp("-O huge_file") == 0)
492 append_unique(anchor, ",", "huge_file", NULL, maxbuflen);
494 /* Enable large block addresses if the LUN is over 2^32 blocks. */
495 if (mop->mo_device_kb / (L_BLOCK_SIZE >> 10) >= 0x100002000ULL &&
496 is_e2fsprogs_feature_supp("-O 64bit") == 0)
497 append_unique(anchor, ",", "64bit", NULL, maxbuflen);
499 /* Cluster inode/block bitmaps and inode table for more efficient IO.
500 * Align the flex groups on a 1MB boundary for better performance. */
501 /* This -O feature needs to go last, since it adds the "-G" option. */
502 if (is_e2fsprogs_feature_supp("-O flex_bg") == 0) {
505 append_unique(anchor, ",", "flex_bg", NULL, maxbuflen);
507 if (IS_OST(&mop->mo_ldd)) {
508 snprintf(tmp_buf, sizeof(tmp_buf), " -G %u",
509 (1 << 20) / L_BLOCK_SIZE);
510 strscat(anchor, tmp_buf, maxbuflen);
513 /* Don't add any more "-O" options here, see last comment above */
518 * moveopts_to_end: find the option string, move remaining strings to
519 * where option string starts, and append the option
521 * @start: where the option string starts before the move
522 * RETURN: where the option string starts after the move
524 static char *moveopts_to_end(char *start)
529 /* skip whitespace before options */
534 /* find end of option characters */
535 while (*end != ' ' && *end != '\0')
539 strncpy(save, start, end - start);
540 save[end - start] = '\0';
542 /* move remaining options up front */
544 memmove(start, end, strlen(end));
545 *(start + strlen(end)) = '\0';
547 /* append the specified options */
548 if (*(start + strlen(start) - 1) != ' ')
550 idx = start + strlen(start);
556 /* Build fs according to type */
557 int ldiskfs_make_lustre(struct mkfs_opts *mop)
559 __u64 device_kb = mop->mo_device_kb, block_count = 0;
560 char mkfs_cmd[PATH_MAX];
564 int ret = 0, ext_opts = 0;
567 if (!(mop->mo_flags & MO_IS_LOOP)) {
568 mop->mo_device_kb = get_device_size(mop->mo_device);
570 if (mop->mo_device_kb == 0)
573 /* Compare to real size */
574 if (device_kb == 0 || device_kb > mop->mo_device_kb)
575 device_kb = mop->mo_device_kb;
577 mop->mo_device_kb = device_kb;
580 if (mop->mo_device_kb != 0) {
581 if (mop->mo_device_kb < 8096) {
582 fprintf(stderr, "%s: size of filesystem must be larger "
583 "than 8MB, but is set to %lldKB\n",
584 progname, (long long)mop->mo_device_kb);
587 block_count = mop->mo_device_kb / (L_BLOCK_SIZE >> 10);
588 /* If the LUN size is just over 2^32 blocks, limit the
589 * filesystem size to 2^32-1 blocks to avoid problems with
590 * ldiskfs/mkfs not handling this size. Bug 22906 */
591 if (block_count > 0xffffffffULL && block_count < 0x100002000ULL)
592 block_count = 0xffffffffULL;
595 if ((mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3) ||
596 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS) ||
597 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS2)) {
600 /* Journal size in MB */
601 if (strstr(mop->mo_mkfsopts, "-J") == NULL &&
602 device_kb > 1024 * 1024) {
603 /* Choose our own default journal size */
604 long journal_mb = 0, max_mb;
606 /* cap journal size at 4GB for MDT,
607 * leave it at 400MB for OSTs. */
608 if (IS_MDT(&mop->mo_ldd))
610 else if (IS_OST(&mop->mo_ldd))
612 else /* Use mke2fs default size for MGS */
615 /* Use at most 4% of device for journal */
616 journal_mb = device_kb * 4 / (1024 * 100);
617 if (journal_mb > max_mb)
621 sprintf(buf, " -J size=%ld", journal_mb);
622 strscat(mop->mo_mkfsopts, buf,
623 sizeof(mop->mo_mkfsopts));
627 /* Inode size (for extended attributes). The LOV EA size is
628 * 32 (EA hdr) + 32 (lov_mds_md) + stripes * 24 (lov_ost_data),
629 * and we want some margin above that for ACLs, other EAs... */
630 if (strstr(mop->mo_mkfsopts, "-I") == NULL) {
631 if (IS_MDT(&mop->mo_ldd)) {
632 if (mop->mo_stripe_count > 72)
633 inode_size = 512; /* bz 7241 */
634 /* see also "-i" below for EA blocks */
635 else if (mop->mo_stripe_count > 32)
637 else if (mop->mo_stripe_count > 10)
641 } else if (IS_OST(&mop->mo_ldd)) {
642 /* We store MDS FID and OST objid in EA on OST
643 * we need to make inode bigger as well. */
647 if (inode_size > 0) {
648 sprintf(buf, " -I %ld", inode_size);
649 strscat(mop->mo_mkfsopts, buf,
650 sizeof(mop->mo_mkfsopts));
654 /* Bytes_per_inode: disk size / num inodes */
655 if (strstr(mop->mo_mkfsopts, "-i") == NULL &&
656 strstr(mop->mo_mkfsopts, "-N") == NULL) {
657 long bytes_per_inode = 0;
659 /* Allocate more inodes on MDT devices. There is
660 * no data stored on the MDT, and very little extra
661 * metadata beyond the inode. It could go down as
662 * low as 1024 bytes, but this is conservative.
663 * Account for external EA blocks for wide striping. */
664 if (IS_MDT(&mop->mo_ldd)) {
665 bytes_per_inode = inode_size + 1536;
667 if (mop->mo_stripe_count > 72) {
668 int extra = mop->mo_stripe_count * 24;
669 extra = ((extra - 1) | 4095) + 1;
670 bytes_per_inode += extra;
674 /* Allocate fewer inodes on large OST devices. Most
675 * filesystems can be much more aggressive than even
676 * this, but it is impossible to know in advance. */
677 if (IS_OST(&mop->mo_ldd)) {
678 /* OST > 16TB assume average file size 1MB */
679 if (device_kb > (16ULL << 30))
680 bytes_per_inode = 1024 * 1024;
681 /* OST > 4TB assume average file size 512kB */
682 else if (device_kb > (4ULL << 30))
683 bytes_per_inode = 512 * 1024;
684 /* OST > 1TB assume average file size 256kB */
685 else if (device_kb > (1ULL << 30))
686 bytes_per_inode = 256 * 1024;
687 /* OST > 10GB assume average file size 64kB,
688 * plus a bit so that inodes will fit into a
689 * 256x flex_bg without overflowing */
690 else if (device_kb > (10ULL << 20))
691 bytes_per_inode = 69905;
694 if (bytes_per_inode > 0) {
695 sprintf(buf, " -i %ld", bytes_per_inode);
696 strscat(mop->mo_mkfsopts, buf,
697 sizeof(mop->mo_mkfsopts));
702 strscat(mop->mo_mkfsopts, " -q",
703 sizeof(mop->mo_mkfsopts));
706 /* start handle -O mkfs options */
707 if ((start = strstr(mop->mo_mkfsopts, "-O")) != NULL) {
708 if (strstr(start + 2, "-O") != NULL) {
710 "%s: don't specify multiple -O options\n",
714 start = moveopts_to_end(start);
715 maxbuflen = sizeof(mop->mo_mkfsopts) -
716 (start - mop->mo_mkfsopts) - strlen(start);
717 ret = enable_default_ext4_features(mop, start, maxbuflen, 1);
719 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts),
720 maxbuflen = sizeof(mop->mo_mkfsopts) -
721 strlen(mop->mo_mkfsopts);
722 ret = enable_default_ext4_features(mop, start, maxbuflen, 0);
726 /* end handle -O mkfs options */
728 /* start handle -E mkfs options */
729 if ((start = strstr(mop->mo_mkfsopts, "-E")) != NULL) {
730 if (strstr(start + 2, "-E") != NULL) {
732 "%s: don't specify multiple -E options\n",
736 start = moveopts_to_end(start);
737 maxbuflen = sizeof(mop->mo_mkfsopts) -
738 (start - mop->mo_mkfsopts) - strlen(start);
741 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts);
742 maxbuflen = sizeof(mop->mo_mkfsopts) -
743 strlen(mop->mo_mkfsopts);
746 /* In order to align the filesystem metadata on 1MB boundaries,
747 * give a resize value that will reserve a power-of-two group
748 * descriptor blocks, but leave one block for the superblock.
749 * Only useful for filesystems with < 2^32 blocks due to resize
751 if (IS_OST(&mop->mo_ldd) && mop->mo_device_kb > 100 * 1024 &&
752 mop->mo_device_kb * 1024 / L_BLOCK_SIZE <= 0xffffffffULL) {
753 unsigned group_blocks = L_BLOCK_SIZE * 8;
754 unsigned desc_per_block = L_BLOCK_SIZE / 32;
755 unsigned resize_blks;
757 resize_blks = (1ULL<<32) - desc_per_block*group_blocks;
758 snprintf(buf, sizeof(buf), "%u", resize_blks);
759 append_unique(start, ext_opts ? "," : " -E ",
760 "resize", buf, maxbuflen);
764 /* Avoid zeroing out the full journal - speeds up mkfs */
765 if (is_e2fsprogs_feature_supp("-E lazy_journal_init") == 0)
766 append_unique(start, ext_opts ? "," : " -E ",
767 "lazy_journal_init", NULL, maxbuflen);
768 /* end handle -E mkfs options */
770 /* Allow reformat of full devices (as opposed to
771 partitions.) We already checked for mounted dev. */
772 strscat(mop->mo_mkfsopts, " -F", sizeof(mop->mo_mkfsopts));
774 snprintf(mkfs_cmd, sizeof(mkfs_cmd),
775 "%s -j -b %d -L %s ", MKE2FS, L_BLOCK_SIZE,
776 mop->mo_ldd.ldd_svname);
778 fprintf(stderr,"%s: unsupported fs type: %d (%s)\n",
779 progname, mop->mo_ldd.ldd_mount_type,
780 MT_STR(&mop->mo_ldd));
784 /* For loop device format the dev, not the filename */
785 dev = mop->mo_device;
786 if (mop->mo_flags & MO_IS_LOOP)
787 dev = mop->mo_loopdev;
789 vprint("formatting backing filesystem %s on %s\n",
790 MT_STR(&mop->mo_ldd), dev);
791 vprint("\ttarget name %s\n", mop->mo_ldd.ldd_svname);
792 vprint("\t4k blocks "LPU64"\n", block_count);
793 vprint("\toptions %s\n", mop->mo_mkfsopts);
795 /* mkfs_cmd's trailing space is important! */
796 strscat(mkfs_cmd, mop->mo_mkfsopts, sizeof(mkfs_cmd));
797 strscat(mkfs_cmd, " ", sizeof(mkfs_cmd));
798 strscat(mkfs_cmd, dev, sizeof(mkfs_cmd));
799 if (block_count != 0) {
800 sprintf(buf, " "LPU64, block_count);
801 strscat(mkfs_cmd, buf, sizeof(mkfs_cmd));
804 vprint("mkfs_cmd = %s\n", mkfs_cmd);
805 ret = run_command(mkfs_cmd, sizeof(mkfs_cmd));
808 fprintf(stderr, "Unable to build fs %s (%d)\n", dev, ret);
813 int ldiskfs_prepare_lustre(struct mkfs_opts *mop,
814 char *default_mountopts, int default_len,
815 char *always_mountopts, int always_len)
817 struct lustre_disk_data *ldd = &mop->mo_ldd;
820 /* Set MO_IS_LOOP to indicate a loopback device is needed */
821 ret = is_block(mop->mo_device);
824 } else if (ret == 0) {
825 mop->mo_flags |= MO_IS_LOOP;
828 strscat(default_mountopts, ",errors=remount-ro", default_len);
829 if (IS_MDT(ldd) || IS_MGS(ldd))
830 strscat(always_mountopts, ",user_xattr", always_len);
835 int read_file(const char *path, char *buf, int size)
839 fd = fopen(path, "r");
843 /* should not ignore fgets(3)'s return value */
844 if (!fgets(buf, size, fd)) {
845 fprintf(stderr, "reading from %s: %s", path, strerror(errno));
853 int write_file(const char *path, const char *buf)
857 fd = fopen(path, "w");
866 int set_blockdev_scheduler(const char *path, const char *scheduler)
868 char buf[PATH_MAX], *c;
871 /* Before setting the scheduler, we need to check to see if it's
872 * already set to "noop". If it is, we don't want to override
873 * that setting. If it's set to anything other than "noop", set
874 * the scheduler to what has been passed in. */
876 rc = read_file(path, buf, sizeof(buf));
879 fprintf(stderr, "%s: cannot open '%s': %s\n",
880 progname, path, strerror(errno));
884 /* The expected format of buf: noop anticipatory deadline [cfq] */
885 c = strchr(buf, '[');
887 /* If c is NULL, the format is not what we expect. Play it safe
891 fprintf(stderr, "%s: cannot parse scheduler "
892 "options for '%s'\n", progname, path);
896 if (strncmp(c+1, "noop", 4) == 0)
899 rc = write_file(path, scheduler);
902 fprintf(stderr, "%s: cannot set scheduler on "
903 "'%s': %s\n", progname, path,
911 /* This is to tune the kernel for good SCSI performance.
912 * For that we set the value of /sys/block/{dev}/queue/max_sectors_kb
913 * to the value of /sys/block/{dev}/queue/max_hw_sectors_kb */
914 int set_blockdev_tunables(char *source, struct mount_opts *mop)
916 glob_t glob_info = { 0 };
917 struct stat stat_buf;
918 char *chk_major, *chk_minor;
919 char *savept = NULL, *dev;
921 char buf[PATH_MAX] = {'\0'}, path[PATH_MAX] = {'\0'};
922 char real_path[PATH_MAX] = {'\0'};
930 ret_path = realpath(source, real_path);
931 if (ret_path == NULL) {
933 fprintf(stderr, "warning: %s: cannot resolve: %s\n",
934 source, strerror(errno));
938 if (strncmp(real_path, "/dev/loop", 9) == 0)
941 if ((real_path[0] != '/') && (strpbrk(real_path, ",:") != NULL))
944 snprintf(path, sizeof(path), "/sys/block%s", real_path + 4);
945 if (access(path, X_OK) == 0)
948 /* The name of the device say 'X' specified in /dev/X may not
949 * match any entry under /sys/block/. In that case we need to
950 * match the major/minor number to find the entry under
951 * sys/block corresponding to /dev/X */
953 /* Don't chop tail digit on /dev/mapper/xxx, LU-478 */
954 if (strncmp(real_path, "/dev/mapper", 11) != 0) {
955 dev = real_path + strlen(real_path);
956 while (--dev > real_path && isdigit(*dev))
959 if (strncmp(real_path, "/dev/md_", 8) == 0)
963 rc = stat(real_path, &stat_buf);
966 fprintf(stderr, "warning: %s, device %s stat failed\n",
967 strerror(errno), real_path);
971 major = major(stat_buf.st_rdev);
972 minor = minor(stat_buf.st_rdev);
973 rc = glob("/sys/block/*", GLOB_NOSORT, NULL, &glob_info);
976 fprintf(stderr, "warning: failed to read entries under "
978 globfree(&glob_info);
982 for (i = 0; i < glob_info.gl_pathc; i++){
983 snprintf(path, sizeof(path), "%s/dev", glob_info.gl_pathv[i]);
985 rc = read_file(path, buf, sizeof(buf));
989 if (buf[strlen(buf) - 1] == '\n')
990 buf[strlen(buf) - 1] = '\0';
992 chk_major = strtok_r(buf, ":", &savept);
994 if (chk_major != NULL && major == atoi(chk_major) &&
995 chk_minor != NULL && minor == atoi(chk_minor))
999 if (i == glob_info.gl_pathc) {
1001 fprintf(stderr,"warning: device %s does not match any "
1002 "entry under /sys/block\n", real_path);
1003 globfree(&glob_info);
1007 /* Chop off "/dev" from path we found */
1008 path[strlen(glob_info.gl_pathv[i])] = '\0';
1009 globfree(&glob_info);
1012 if (strncmp(real_path, "/dev/md", 7) == 0) {
1013 snprintf(real_path, sizeof(real_path), "%s/%s", path,
1016 rc = read_file(real_path, buf, sizeof(buf));
1019 fprintf(stderr, "warning: opening %s: %s\n",
1020 real_path, strerror(errno));
1024 if (atoi(buf) >= mop->mo_md_stripe_cache_size)
1027 if (strlen(buf) - 1 > 0) {
1028 snprintf(buf, sizeof(buf), "%d",
1029 mop->mo_md_stripe_cache_size);
1030 rc = write_file(real_path, buf);
1032 fprintf(stderr, "warning: opening %s: %s\n",
1033 real_path, strerror(errno));
1035 /* Return since raid and disk tunables are different */
1039 snprintf(real_path, sizeof(real_path), "%s/%s", path,
1040 MAX_HW_SECTORS_KB_PATH);
1041 rc = read_file(real_path, buf, sizeof(buf));
1044 fprintf(stderr, "warning: opening %s: %s\n",
1045 real_path, strerror(errno));
1046 /* No MAX_HW_SECTORS_KB_PATH isn't necessary an
1047 * error for some device. */
1051 if (strlen(buf) - 1 > 0) {
1052 snprintf(real_path, sizeof(real_path), "%s/%s", path,
1053 MAX_SECTORS_KB_PATH);
1054 rc = write_file(real_path, buf);
1057 fprintf(stderr, "warning: writing to %s: %s\n",
1058 real_path, strerror(errno));
1059 /* No MAX_SECTORS_KB_PATH isn't necessary an
1060 * error for some device. */
1065 /* Purposely ignore errors reported from set_blockdev_scheduler.
1066 * The worst that will happen is a block device with an "incorrect"
1068 snprintf(real_path, sizeof(real_path), "%s/%s", path, SCHEDULER_PATH);
1069 set_blockdev_scheduler(real_path, DEFAULT_SCHEDULER);
1071 /* if device is multipath device, tune its slave devices */
1072 glob_info.gl_pathc = 0;
1073 glob_info.gl_offs = 0;
1074 snprintf(real_path, sizeof(real_path), "%s/slaves/*", path);
1075 rc = glob(real_path, GLOB_NOSORT, NULL, &glob_info);
1077 for (i = 0; rc == 0 && i < glob_info.gl_pathc; i++) {
1078 slave = basename(glob_info.gl_pathv[i]);
1079 snprintf(real_path, sizeof(real_path), "/dev/%s", slave);
1080 rc = set_blockdev_tunables(real_path, mop);
1083 if (rc == GLOB_NOMATCH) {
1084 /* no slave device is not an error */
1086 } else if (rc && verbose) {
1087 if (slave == NULL) {
1088 fprintf(stderr, "warning: %s, failed to read"
1089 " entries under %s/slaves\n",
1090 strerror(errno), path);
1092 fprintf(stderr, "unable to set tunables for"
1093 " slave device %s (slave would be"
1094 " unable to handle IO request from"
1099 globfree(&glob_info);
1104 int ldiskfs_tune_lustre(char *dev, struct mount_opts *mop)
1106 return set_blockdev_tunables(dev, mop);
1109 int ldiskfs_label_lustre(struct mount_opts *mop)
1111 char label_cmd[PATH_MAX];
1114 snprintf(label_cmd, sizeof(label_cmd),
1115 TUNE2FS" -f -L '%s' '%s' >/dev/null 2>&1",
1116 mop->mo_ldd.ldd_svname, mop->mo_source);
1117 rc = run_command(label_cmd, sizeof(label_cmd));
1122 /* return canonicalized absolute pathname, even if the target file does not
1123 * exist, unlike realpath */
1124 static char *absolute_path(char *devname)
1126 char buf[PATH_MAX + 1];
1130 path = malloc(PATH_MAX + 1);
1134 if (devname[0] != '/') {
1135 if (getcwd(buf, sizeof(buf) - 1) == NULL) {
1140 if (strlen(devname) > sizeof(buf)-strlen(buf)-1) {
1144 strncat(buf, devname, sizeof(buf)-strlen(buf)-1);
1146 if (strlen(devname) > sizeof(buf)-1) {
1150 strncpy(buf, devname, sizeof(buf));
1152 /* truncate filename before calling realpath */
1153 ptr = strrchr(buf, '/');
1159 if (path != realpath(buf, path)) {
1163 /* add the filename back */
1165 strcat(path, ptr + 1);
1169 /* Determine if a device is a block device (as opposed to a file) */
1170 int is_block(char* devname)
1176 devpath = absolute_path(devname);
1177 if (devpath == NULL) {
1178 fprintf(stderr, "%s: failed to resolve path to %s\n",
1183 ret = access(devname, F_OK);
1185 if (strncmp(devpath, "/dev/", 5) == 0) {
1186 /* nobody sane wants to create a loopback file under
1187 * /dev. Let's just report the device doesn't exist */
1188 fprintf(stderr, "%s: %s apparently does not exist\n",
1196 ret = stat(devpath, &st);
1198 fprintf(stderr, "%s: cannot stat %s\n", progname, devpath);
1201 ret = S_ISBLK(st.st_mode);
1207 static int is_feature_enabled(const char *feature, const char *devpath)
1211 char enabled_features[4096] = "";
1214 snprintf(cmd, sizeof(cmd), "%s -R features %s 2>&1",
1217 /* Using popen() instead of run_command() since debugfs does
1218 * not return proper error code if command is not supported */
1219 fp = popen(cmd, "r");
1221 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
1225 ret = fread(enabled_features, 1, sizeof(enabled_features) - 1, fp);
1226 enabled_features[ret] = '\0';
1229 if (strstr(enabled_features, feature))
1234 /* Enable quota accounting */
1235 int ldiskfs_enable_quota(struct mkfs_opts *mop)
1239 int cmdsz = sizeof(cmd), ret;
1241 if (is_e2fsprogs_feature_supp("-O quota") != 0) {
1242 fprintf(stderr, "%s: \"-O quota\" is is not supported by "
1243 "current e2fsprogs\n", progname);
1247 dev = mop->mo_device;
1248 if (mop->mo_flags & MO_IS_LOOP)
1249 dev = mop->mo_loopdev;
1251 /* Quota feature is already enabled? */
1252 if (is_feature_enabled("quota", dev)) {
1253 vprint("Quota feature is already enabled.\n");
1257 /* Turn on quota feature by "tune2fs -O quota" */
1258 snprintf(cmd, cmdsz, "%s -O quota %s", TUNE2FS, dev);
1259 ret = run_command(cmd, cmdsz);
1261 fprintf(stderr, "command:%s (%d)", cmd, ret);
1266 int ldiskfs_init(void)
1268 /* Required because full path to DEBUGFS is not specified */
1269 setenv("PATH", "/bin:/sbin:/usr/bin:/usr/sbin:/usr/local/bin", 0);
1274 void ldiskfs_fini(void)