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>
83 #define MAX_HW_SECTORS_KB_PATH "queue/max_hw_sectors_kb"
84 #define MAX_SECTORS_KB_PATH "queue/max_sectors_kb"
85 #define SCHEDULER_PATH "queue/scheduler"
86 #define STRIPE_CACHE_SIZE "md/stripe_cache_size"
88 #define DEFAULT_SCHEDULER "deadline"
90 extern char *progname;
92 #define L_BLOCK_SIZE 4096
93 /* keep it less than LL_FID_NAMELEN */
94 #define DUMMY_FILE_NAME_LEN 25
95 #define EXT3_DIRENT_SIZE DUMMY_FILE_NAME_LEN
97 /* Write the server config files */
98 int ldiskfs_write_ldd(struct mkfs_opts *mop)
100 char mntpt[] = "/tmp/mntXXXXXX";
107 /* Mount this device temporarily in order to write these files */
108 if (!mkdtemp(mntpt)) {
109 fprintf(stderr, "%s: Can't create temp mount point %s: %s\n",
110 progname, mntpt, strerror(errno));
114 dev = mop->mo_device;
115 if (mop->mo_flags & MO_IS_LOOP)
116 dev = mop->mo_loopdev;
118 ret = mount(dev, mntpt, MT_STR(&mop->mo_ldd), 0,
119 mop->mo_ldd.ldd_mount_opts);
121 fprintf(stderr, "%s: Unable to mount %s: %s\n",
122 progname, dev, strerror(errno));
124 if (errno == ENODEV) {
125 fprintf(stderr, "Is the %s module available?\n",
126 MT_STR(&mop->mo_ldd));
131 /* Set up initial directories */
132 sprintf(filepnm, "%s/%s", mntpt, MOUNT_CONFIGS_DIR);
133 ret = mkdir(filepnm, 0777);
134 if ((ret != 0) && (errno != EEXIST)) {
135 fprintf(stderr, "%s: Can't make configs dir %s (%s)\n",
136 progname, filepnm, strerror(errno));
138 } else if (errno == EEXIST) {
142 /* Save the persistent mount data into a file. Lustre must pre-read
143 this file to get the real mount options. */
144 vprint("Writing %s\n", MOUNT_DATA_FILE);
145 sprintf(filepnm, "%s/%s", mntpt, MOUNT_DATA_FILE);
146 filep = fopen(filepnm, "w");
148 fprintf(stderr, "%s: Unable to create %s file: %s\n",
149 progname, filepnm, strerror(errno));
152 num = fwrite(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
153 if (num < 1 && ferror(filep)) {
154 fprintf(stderr, "%s: Unable to write to file (%s): %s\n",
155 progname, filepnm, strerror(errno));
168 static int readcmd(char *cmd, char *buf, int len)
173 fp = popen(cmd, "r");
177 red = fread(buf, 1, len, fp);
180 /* strip trailing newline */
181 if (buf[red - 1] == '\n')
184 return (red == 0) ? -ENOENT : 0;
187 int ldiskfs_read_ldd(char *dev, struct lustre_disk_data *mo_ldd)
189 char tmpdir[] = "/tmp/dirXXXXXX";
194 int cmdsz = sizeof(cmd);
196 /* Make a temporary directory to hold Lustre data files. */
197 if (!mkdtemp(tmpdir)) {
198 fprintf(stderr, "%s: Can't create temporary directory %s: %s\n",
199 progname, tmpdir, strerror(errno));
203 /* TODO: it's worth observing the get_mountdata() function that is
204 in mount_utils.c for getting the mountdata out of the
207 /* Construct debugfs command line. */
208 snprintf(cmd, cmdsz, "%s -c -R 'dump /%s %s/mountdata' '%s'",
209 DEBUGFS, MOUNT_DATA_FILE, tmpdir, dev);
211 ret = run_command(cmd, cmdsz);
213 verrprint("%s: Unable to dump %s dir (%d)\n",
214 progname, MOUNT_CONFIGS_DIR, ret);
216 sprintf(filepnm, "%s/mountdata", tmpdir);
217 filep = fopen(filepnm, "r");
220 vprint("Reading %s\n", MOUNT_DATA_FILE);
221 num_read = fread(mo_ldd, sizeof(*mo_ldd), 1, filep);
222 if (num_read < 1 && ferror(filep)) {
223 fprintf(stderr, "%s: Unable to read from file %s: %s\n",
224 progname, filepnm, strerror(errno));
229 snprintf(cmd, cmdsz, "rm -rf %s", tmpdir);
230 run_command(cmd, cmdsz);
232 verrprint("Failed to read old data (%d)\n", ret);
234 /* As long as we at least have the label, we're good to go */
235 snprintf(cmd, sizeof(cmd), E2LABEL" %s", dev);
236 ret = readcmd(cmd, mo_ldd->ldd_svname, sizeof(mo_ldd->ldd_svname) - 1);
242 /* Display the need for the latest e2fsprogs to be installed. make_backfs
243 * indicates if the caller is make_lustre_backfs() or not. */
244 void disp_old_e2fsprogs_msg(const char *feature, int make_backfs)
246 static int msg_displayed;
249 fprintf(stderr, "WARNING: %s does not support %s "
250 "feature.\n\n", E2FSPROGS, feature);
256 fprintf(stderr, "WARNING: The %s package currently installed on "
257 "your system does not support \"%s\" feature.\n",
259 #if !(HAVE_LDISKFSPROGS)
260 fprintf(stderr, "Please install the latest version of e2fsprogs from\n"
261 "http://downloads.whamcloud.com/public/e2fsprogs/latest/\n"
262 "to enable this feature.\n");
265 fprintf(stderr, "Feature will not be enabled until %s"
266 "is updated and '%s -O %s %%{device}' "
267 "is run.\n\n", E2FSPROGS, TUNE2FS, feature);
270 /* Check whether the file exists in the device */
271 static int file_in_dev(char *file_name, char *dev_name)
274 char debugfs_cmd[256];
275 unsigned int inode_num;
278 /* Construct debugfs command line. */
279 snprintf(debugfs_cmd, sizeof(debugfs_cmd),
280 "%s -c -R 'stat %s' '%s' 2>&1 | egrep '(Inode|unsupported)'",
281 DEBUGFS, file_name, dev_name);
283 fp = popen(debugfs_cmd, "r");
285 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
289 if (fscanf(fp, "Inode: %u", &inode_num) == 1) { /* exist */
293 i = fread(debugfs_cmd, 1, sizeof(debugfs_cmd) - 1, fp);
296 fprintf(stderr, "%s", debugfs_cmd);
297 if (strstr(debugfs_cmd, "unsupported feature")) {
298 disp_old_e2fsprogs_msg("an unknown", 0);
307 /* Check whether the device has already been used with lustre */
308 int ldiskfs_is_lustre(char *dev, unsigned *mount_type)
312 ret = file_in_dev(MOUNT_DATA_FILE, dev);
314 /* in the -1 case, 'extents' means IS a lustre target */
315 *mount_type = LDD_MT_LDISKFS;
319 ret = file_in_dev(LAST_RCVD, dev);
321 *mount_type = LDD_MT_LDISKFS;
328 /* Check if a certain feature is supported by e2fsprogs.
329 * Firstly we try to use "debugfs supported_features" command to check if
330 * the feature is supported. If this fails we try to set this feature with
331 * mke2fs to check for its support. */
332 static int is_e2fsprogs_feature_supp(const char *feature)
334 static char supp_features[4096] = "";
337 char imgname[] = "/tmp/test-img-XXXXXX";
341 if (supp_features[0] == '\0') {
342 snprintf(cmd, sizeof(cmd), "%s -c -R supported_features 2>&1",
345 /* Using popen() instead of run_command() since debugfs does
346 * not return proper error code if command is not supported */
347 fp = popen(cmd, "r");
349 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
352 ret = fread(supp_features, 1, sizeof(supp_features) - 1, fp);
353 supp_features[ret] = '\0';
356 if (ret > 0 && strstr(supp_features,
357 strncmp(feature, "-O ", 3) ? feature : feature+3))
360 if ((fd = mkstemp(imgname)) < 0)
365 snprintf(cmd, sizeof(cmd), "%s -F %s %s 100 >/dev/null 2>&1",
366 MKE2FS, feature, imgname);
367 /* run_command() displays the output of mke2fs when it fails for
368 * some feature, so use system() directly */
377 * append_unique: append @key or @key=@val pair to @buf only if @key does not
379 * @buf: buffer to hold @key or @key=@val
380 * @prefix: prefix string before @key
382 * @val: value string if it's a @key=@val pair
384 static void append_unique(char *buf, char *prefix, char *key, char *val,
393 anchor = end = strstr(buf, key);
394 /* try to find exact match string in @buf */
395 while (end && *end != '\0' && *end != ',' && *end != ' ' && *end != '=')
398 if (anchor == NULL || strlen(key) != len ||
399 strncmp(anchor, key, len) != 0) {
401 strscat(buf, prefix, maxbuflen);
403 strscat(buf, key, maxbuflen);
405 strscat(buf, "=", maxbuflen);
406 strscat(buf, val, maxbuflen);
411 static int enable_default_ext4_features(struct mkfs_opts *mop, char *anchor,
412 size_t maxbuflen, int user_spec)
414 if (IS_OST(&mop->mo_ldd)) {
415 append_unique(anchor, user_spec ? "," : " -O ",
416 "extents", NULL, sizeof(mop->mo_mkfsopts));
417 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
418 } else if (IS_MDT(&mop->mo_ldd)) {
419 append_unique(anchor, user_spec ? "," : " -O ",
420 "dirdata", NULL, maxbuflen);
421 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
422 append_unique(anchor, ",", "^extents", NULL, maxbuflen);
424 append_unique(anchor, user_spec ? "," : " -O ",
425 "uninit_bg", NULL, maxbuflen);
428 /* Multiple mount protection enabled only if failover node specified */
429 if (mop->mo_flags & MO_FAILOVER) {
430 if (is_e2fsprogs_feature_supp("-O mmp") == 0)
431 append_unique(anchor, ",", "mmp", NULL, maxbuflen);
433 disp_old_e2fsprogs_msg("mmp", 1);
436 /* Allow more than 65000 subdirectories */
437 if (is_e2fsprogs_feature_supp("-O dir_nlink") == 0)
438 append_unique(anchor, ",", "dir_nlink", NULL, maxbuflen);
440 /* The following options are only valid for ext4-based ldiskfs.
441 * If --backfstype=ext3 is specified, do not enable them. */
442 if (mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3)
445 /* Enable quota by default */
446 if (is_e2fsprogs_feature_supp("-O quota") == 0) {
447 append_unique(anchor, ",", "quota", NULL, maxbuflen);
450 fprintf(stderr, "\"-O quota\" must be supported by "
451 "e2fsprogs, please upgrade your e2fsprogs.\n");
455 /* Allow files larger than 2TB. Also needs LU-16, but not harmful. */
456 if (is_e2fsprogs_feature_supp("-O huge_file") == 0)
457 append_unique(anchor, ",", "huge_file", NULL, maxbuflen);
459 /* Enable large block addresses if the LUN is over 2^32 blocks. */
460 if (mop->mo_device_sz / (L_BLOCK_SIZE >> 10) >= 0x100002000ULL &&
461 is_e2fsprogs_feature_supp("-O 64bit") == 0)
462 append_unique(anchor, ",", "64bit", NULL, maxbuflen);
464 /* Cluster inode/block bitmaps and inode table for more efficient IO.
465 * Align the flex groups on a 1MB boundary for better performance. */
466 /* This -O feature needs to go last, since it adds the "-G" option. */
467 if (is_e2fsprogs_feature_supp("-O flex_bg") == 0) {
470 append_unique(anchor, ",", "flex_bg", NULL, maxbuflen);
472 if (IS_OST(&mop->mo_ldd)) {
473 snprintf(tmp_buf, sizeof(tmp_buf), " -G %u",
474 (1 << 20) / L_BLOCK_SIZE);
475 strscat(anchor, tmp_buf, maxbuflen);
478 /* Don't add any more "-O" options here, see last comment above */
483 * moveopts_to_end: find the option string, move remaining strings to
484 * where option string starts, and append the option
486 * @start: where the option string starts before the move
487 * RETURN: where the option string starts after the move
489 static char *moveopts_to_end(char *start)
494 /* skip whitespace before options */
499 /* find end of option characters */
500 while (*end != ' ' && *end != '\0')
504 strncpy(save, start, end - start);
505 save[end - start] = '\0';
507 /* move remaining options up front */
509 memmove(start, end, strlen(end));
510 *(start + strlen(end)) = '\0';
512 /* append the specified options */
513 if (*(start + strlen(start) - 1) != ' ')
515 idx = start + strlen(start);
521 /* Build fs according to type */
522 int ldiskfs_make_lustre(struct mkfs_opts *mop)
524 __u64 device_sz = mop->mo_device_sz, block_count = 0;
525 char mkfs_cmd[PATH_MAX];
529 int ret = 0, ext_opts = 0;
532 if (!(mop->mo_flags & MO_IS_LOOP)) {
533 mop->mo_device_sz = get_device_size(mop->mo_device);
535 if (mop->mo_device_sz == 0)
538 /* Compare to real size */
539 if (device_sz == 0 || device_sz > mop->mo_device_sz)
540 device_sz = mop->mo_device_sz;
542 mop->mo_device_sz = device_sz;
545 if (mop->mo_device_sz != 0) {
546 if (mop->mo_device_sz < 8096){
547 fprintf(stderr, "%s: size of filesystem must be larger "
548 "than 8MB, but is set to %lldKB\n",
549 progname, (long long)mop->mo_device_sz);
552 block_count = mop->mo_device_sz / (L_BLOCK_SIZE >> 10);
553 /* If the LUN size is just over 2^32 blocks, limit the
554 * filesystem size to 2^32-1 blocks to avoid problems with
555 * ldiskfs/mkfs not handling this size. Bug 22906 */
556 if (block_count > 0xffffffffULL && block_count < 0x100002000ULL)
557 block_count = 0xffffffffULL;
560 if ((mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3) ||
561 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS) ||
562 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS2)) {
565 /* Journal size in MB */
566 if (strstr(mop->mo_mkfsopts, "-J") == NULL) {
567 /* Choose our own default journal size */
568 long journal_sz = 0, max_sz;
569 if (device_sz > 1024 * 1024) /* 1GB */
570 journal_sz = (device_sz / 102400) * 4;
571 /* cap journal size at 1GB */
572 if (journal_sz > 1024L)
575 max_sz = (102400 * L_BLOCK_SIZE) >> 20; /* 400MB */
576 if (journal_sz > max_sz)
579 sprintf(buf, " -J size=%ld", journal_sz);
580 strscat(mop->mo_mkfsopts, buf,
581 sizeof(mop->mo_mkfsopts));
585 /* Inode size (for extended attributes). The LOV EA size is
586 * 32 (EA hdr) + 32 (lov_mds_md) + stripes * 24 (lov_ost_data),
587 * and we want some margin above that for ACLs, other EAs... */
588 if (strstr(mop->mo_mkfsopts, "-I") == NULL) {
589 if (IS_MDT(&mop->mo_ldd)) {
590 if (mop->mo_stripe_count > 72)
591 inode_size = 512; /* bz 7241 */
592 /* see also "-i" below for EA blocks */
593 else if (mop->mo_stripe_count > 32)
595 else if (mop->mo_stripe_count > 10)
599 } else if (IS_OST(&mop->mo_ldd)) {
600 /* We store MDS FID and OST objid in EA on OST
601 * we need to make inode bigger as well. */
605 if (inode_size > 0) {
606 sprintf(buf, " -I %ld", inode_size);
607 strscat(mop->mo_mkfsopts, buf,
608 sizeof(mop->mo_mkfsopts));
612 /* Bytes_per_inode: disk size / num inodes */
613 if (strstr(mop->mo_mkfsopts, "-i") == NULL &&
614 strstr(mop->mo_mkfsopts, "-N") == NULL) {
615 long bytes_per_inode = 0;
617 /* Allocate more inodes on MDT devices. There is
618 * no data stored on the MDT, and very little extra
619 * metadata beyond the inode. It could go down as
620 * low as 1024 bytes, but this is conservative.
621 * Account for external EA blocks for wide striping. */
622 if (IS_MDT(&mop->mo_ldd)) {
623 bytes_per_inode = inode_size + 1536;
625 if (mop->mo_stripe_count > 72) {
626 int extra = mop->mo_stripe_count * 24;
627 extra = ((extra - 1) | 4095) + 1;
628 bytes_per_inode += extra;
632 /* Allocate fewer inodes on large OST devices. Most
633 * filesystems can be much more aggressive than even
634 * this, but it is impossible to know in advance. */
635 if (IS_OST(&mop->mo_ldd)) {
636 /* OST > 16TB assume average file size 1MB */
637 if (device_sz > (16ULL << 30))
638 bytes_per_inode = 1024 * 1024;
639 /* OST > 4TB assume average file size 512kB */
640 else if (device_sz > (4ULL << 30))
641 bytes_per_inode = 512 * 1024;
642 /* OST > 1TB assume average file size 256kB */
643 else if (device_sz > (1ULL << 30))
644 bytes_per_inode = 256 * 1024;
645 /* OST > 10GB assume average file size 64kB,
646 * plus a bit so that inodes will fit into a
647 * 256x flex_bg without overflowing */
648 else if (device_sz > (10ULL << 20))
649 bytes_per_inode = 69905;
652 if (bytes_per_inode > 0) {
653 sprintf(buf, " -i %ld", bytes_per_inode);
654 strscat(mop->mo_mkfsopts, buf,
655 sizeof(mop->mo_mkfsopts));
660 strscat(mop->mo_mkfsopts, " -q",
661 sizeof(mop->mo_mkfsopts));
664 /* start handle -O mkfs options */
665 if ((start = strstr(mop->mo_mkfsopts, "-O")) != NULL) {
666 if (strstr(start + 2, "-O") != NULL) {
668 "%s: don't specify multiple -O options\n",
672 start = moveopts_to_end(start);
673 maxbuflen = sizeof(mop->mo_mkfsopts) -
674 (start - mop->mo_mkfsopts) - strlen(start);
675 ret = enable_default_ext4_features(mop, start, maxbuflen, 1);
677 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts),
678 maxbuflen = sizeof(mop->mo_mkfsopts) -
679 strlen(mop->mo_mkfsopts);
680 ret = enable_default_ext4_features(mop, start, maxbuflen, 0);
684 /* end handle -O mkfs options */
686 /* start handle -E mkfs options */
687 if ((start = strstr(mop->mo_mkfsopts, "-E")) != NULL) {
688 if (strstr(start + 2, "-E") != NULL) {
690 "%s: don't specify multiple -E options\n",
694 start = moveopts_to_end(start);
695 maxbuflen = sizeof(mop->mo_mkfsopts) -
696 (start - mop->mo_mkfsopts) - strlen(start);
699 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts);
700 maxbuflen = sizeof(mop->mo_mkfsopts) -
701 strlen(mop->mo_mkfsopts);
704 /* In order to align the filesystem metadata on 1MB boundaries,
705 * give a resize value that will reserve a power-of-two group
706 * descriptor blocks, but leave one block for the superblock.
707 * Only useful for filesystems with < 2^32 blocks due to resize
709 if (IS_OST(&mop->mo_ldd) && mop->mo_device_sz > 100 * 1024 &&
710 mop->mo_device_sz * 1024 / L_BLOCK_SIZE <= 0xffffffffULL) {
711 unsigned group_blocks = L_BLOCK_SIZE * 8;
712 unsigned desc_per_block = L_BLOCK_SIZE / 32;
713 unsigned resize_blks;
715 resize_blks = (1ULL<<32) - desc_per_block*group_blocks;
716 snprintf(buf, sizeof(buf), "%u", resize_blks);
717 append_unique(start, ext_opts ? "," : " -E ",
718 "resize", buf, maxbuflen);
722 /* Avoid zeroing out the full journal - speeds up mkfs */
723 if (is_e2fsprogs_feature_supp("-E lazy_journal_init") == 0)
724 append_unique(start, ext_opts ? "," : " -E ",
725 "lazy_journal_init", NULL, maxbuflen);
726 /* end handle -E mkfs options */
728 /* Allow reformat of full devices (as opposed to
729 partitions.) We already checked for mounted dev. */
730 strscat(mop->mo_mkfsopts, " -F", sizeof(mop->mo_mkfsopts));
732 snprintf(mkfs_cmd, sizeof(mkfs_cmd),
733 "%s -j -b %d -L %s ", MKE2FS, L_BLOCK_SIZE,
734 mop->mo_ldd.ldd_svname);
736 fprintf(stderr,"%s: unsupported fs type: %d (%s)\n",
737 progname, mop->mo_ldd.ldd_mount_type,
738 MT_STR(&mop->mo_ldd));
742 /* For loop device format the dev, not the filename */
743 dev = mop->mo_device;
744 if (mop->mo_flags & MO_IS_LOOP)
745 dev = mop->mo_loopdev;
747 vprint("formatting backing filesystem %s on %s\n",
748 MT_STR(&mop->mo_ldd), dev);
749 vprint("\ttarget name %s\n", mop->mo_ldd.ldd_svname);
750 vprint("\t4k blocks "LPU64"\n", block_count);
751 vprint("\toptions %s\n", mop->mo_mkfsopts);
753 /* mkfs_cmd's trailing space is important! */
754 strscat(mkfs_cmd, mop->mo_mkfsopts, sizeof(mkfs_cmd));
755 strscat(mkfs_cmd, " ", sizeof(mkfs_cmd));
756 strscat(mkfs_cmd, dev, sizeof(mkfs_cmd));
757 if (block_count != 0) {
758 sprintf(buf, " "LPU64, block_count);
759 strscat(mkfs_cmd, buf, sizeof(mkfs_cmd));
762 vprint("mkfs_cmd = %s\n", mkfs_cmd);
763 ret = run_command(mkfs_cmd, sizeof(mkfs_cmd));
766 fprintf(stderr, "Unable to build fs %s (%d)\n", dev, ret);
771 int ldiskfs_prepare_lustre(struct mkfs_opts *mop,
772 char *default_mountopts, int default_len,
773 char *always_mountopts, int always_len)
775 struct lustre_disk_data *ldd = &mop->mo_ldd;
778 /* Set MO_IS_LOOP to indicate a loopback device is needed */
779 ret = is_block(mop->mo_device);
782 } else if (ret == 0) {
783 mop->mo_flags |= MO_IS_LOOP;
786 strscat(default_mountopts, ",errors=remount-ro", default_len);
787 if (IS_MDT(ldd) || IS_MGS(ldd))
788 strscat(always_mountopts, ",user_xattr", always_len);
793 int read_file(const char *path, char *buf, int size)
797 fd = fopen(path, "r");
801 /* should not ignore fgets(3)'s return value */
802 if (!fgets(buf, size, fd)) {
803 fprintf(stderr, "reading from %s: %s", path, strerror(errno));
811 int write_file(const char *path, const char *buf)
815 fd = fopen(path, "w");
824 int set_blockdev_scheduler(const char *path, const char *scheduler)
826 char buf[PATH_MAX], *c;
829 /* Before setting the scheduler, we need to check to see if it's
830 * already set to "noop". If it is, we don't want to override
831 * that setting. If it's set to anything other than "noop", set
832 * the scheduler to what has been passed in. */
834 rc = read_file(path, buf, sizeof(buf));
837 fprintf(stderr, "%s: cannot open '%s': %s\n",
838 progname, path, strerror(errno));
842 /* The expected format of buf: noop anticipatory deadline [cfq] */
843 c = strchr(buf, '[');
845 /* If c is NULL, the format is not what we expect. Play it safe
849 fprintf(stderr, "%s: cannot parse scheduler "
850 "options for '%s'\n", progname, path);
854 if (strncmp(c+1, "noop", 4) == 0)
857 rc = write_file(path, scheduler);
860 fprintf(stderr, "%s: cannot set scheduler on "
861 "'%s': %s\n", progname, path,
869 /* This is to tune the kernel for good SCSI performance.
870 * For that we set the value of /sys/block/{dev}/queue/max_sectors_kb
871 * to the value of /sys/block/{dev}/queue/max_hw_sectors_kb */
872 int set_blockdev_tunables(char *source, struct mount_opts *mop, int fan_out)
874 glob_t glob_info = { 0 };
875 struct stat stat_buf;
876 char *chk_major, *chk_minor;
877 char *savept = NULL, *dev;
879 char buf[PATH_MAX] = {'\0'}, path[PATH_MAX] = {'\0'};
880 char real_path[PATH_MAX] = {'\0'};
887 ret_path = realpath(source, real_path);
888 if (ret_path == NULL) {
890 fprintf(stderr, "warning: %s: cannot resolve: %s\n",
891 source, strerror(errno));
895 if (strncmp(real_path, "/dev/loop", 9) == 0)
898 if ((real_path[0] != '/') && (strpbrk(real_path, ",:") != NULL))
901 snprintf(path, sizeof(path), "/sys/block%s", real_path + 4);
902 if (access(path, X_OK) == 0)
905 /* The name of the device say 'X' specified in /dev/X may not
906 * match any entry under /sys/block/. In that case we need to
907 * match the major/minor number to find the entry under
908 * sys/block corresponding to /dev/X */
910 /* Don't chop tail digit on /dev/mapper/xxx, LU-478 */
911 if (strncmp(real_path, "/dev/mapper", 11) != 0) {
912 dev = real_path + strlen(real_path);
913 while (--dev > real_path && isdigit(*dev))
916 if (strncmp(real_path, "/dev/md_", 8) == 0)
920 rc = stat(real_path, &stat_buf);
923 fprintf(stderr, "warning: %s, device %s stat failed\n",
924 strerror(errno), real_path);
928 major = major(stat_buf.st_rdev);
929 minor = minor(stat_buf.st_rdev);
930 rc = glob("/sys/block/*", GLOB_NOSORT, NULL, &glob_info);
933 fprintf(stderr, "warning: failed to read entries under "
935 globfree(&glob_info);
939 for (i = 0; i < glob_info.gl_pathc; i++){
940 snprintf(path, sizeof(path), "%s/dev", glob_info.gl_pathv[i]);
942 rc = read_file(path, buf, sizeof(buf));
946 if (buf[strlen(buf) - 1] == '\n')
947 buf[strlen(buf) - 1] = '\0';
949 chk_major = strtok_r(buf, ":", &savept);
951 if (major == atoi(chk_major) &&minor == atoi(chk_minor))
955 if (i == glob_info.gl_pathc) {
957 fprintf(stderr,"warning: device %s does not match any "
958 "entry under /sys/block\n", real_path);
959 globfree(&glob_info);
963 /* Chop off "/dev" from path we found */
964 path[strlen(glob_info.gl_pathv[i])] = '\0';
965 globfree(&glob_info);
968 if (strncmp(real_path, "/dev/md", 7) == 0) {
969 snprintf(real_path, sizeof(real_path), "%s/%s", path,
972 rc = read_file(real_path, buf, sizeof(buf));
975 fprintf(stderr, "warning: opening %s: %s\n",
976 real_path, strerror(errno));
980 if (atoi(buf) >= mop->mo_md_stripe_cache_size)
983 if (strlen(buf) - 1 > 0) {
984 snprintf(buf, sizeof(buf), "%d",
985 mop->mo_md_stripe_cache_size);
986 rc = write_file(real_path, buf);
988 fprintf(stderr, "warning: opening %s: %s\n",
989 real_path, strerror(errno));
991 /* Return since raid and disk tunables are different */
995 snprintf(real_path, sizeof(real_path), "%s/%s", path,
996 MAX_HW_SECTORS_KB_PATH);
997 rc = read_file(real_path, buf, sizeof(buf));
1000 fprintf(stderr, "warning: opening %s: %s\n",
1001 real_path, strerror(errno));
1002 /* No MAX_HW_SECTORS_KB_PATH isn't necessary an
1003 * error for some device. */
1007 if (strlen(buf) - 1 > 0) {
1008 snprintf(real_path, sizeof(real_path), "%s/%s", path,
1009 MAX_SECTORS_KB_PATH);
1010 rc = write_file(real_path, buf);
1013 fprintf(stderr, "warning: writing to %s: %s\n",
1014 real_path, strerror(errno));
1015 /* No MAX_SECTORS_KB_PATH isn't necessary an
1016 * error for some device. */
1021 /* Purposely ignore errors reported from set_blockdev_scheduler.
1022 * The worst that will happen is a block device with an "incorrect"
1024 snprintf(real_path, sizeof(real_path), "%s/%s", path, SCHEDULER_PATH);
1025 set_blockdev_scheduler(real_path, DEFAULT_SCHEDULER);
1029 glob_info.gl_pathc = 0;
1030 glob_info.gl_offs = 0;
1031 /* if device is multipath device, tune its slave devices */
1032 snprintf(real_path, sizeof(real_path), "%s/slaves/*", path);
1033 rc = glob(real_path, GLOB_NOSORT, NULL, &glob_info);
1035 for (i = 0; rc == 0 && i < glob_info.gl_pathc; i++){
1036 slave = basename(glob_info.gl_pathv[i]);
1037 snprintf(real_path, sizeof(real_path), "/dev/%s", slave);
1038 rc = set_blockdev_tunables(real_path, mop, 0);
1041 if (rc == GLOB_NOMATCH) {
1042 /* no slave device is not an error */
1044 } else if (rc && verbose) {
1045 if (slave == NULL) {
1046 fprintf(stderr, "warning: %s, failed to read"
1047 " entries under %s/slaves\n",
1048 strerror(errno), path);
1050 fprintf(stderr, "unable to set tunables for"
1051 " slave device %s (slave would be"
1052 " unable to handle IO request from"
1057 globfree(&glob_info);
1063 int ldiskfs_tune_lustre(char *dev, struct mount_opts *mop)
1065 return set_blockdev_tunables(dev, mop, 1);
1068 int ldiskfs_label_lustre(struct mount_opts *mop)
1070 char label_cmd[PATH_MAX];
1073 snprintf(label_cmd, sizeof(label_cmd),
1074 TUNE2FS" -f -L '%s' '%s' >/dev/null 2>&1",
1075 mop->mo_ldd.ldd_svname, mop->mo_source);
1076 rc = run_command(label_cmd, sizeof(label_cmd));
1081 /* return canonicalized absolute pathname, even if the target file does not
1082 * exist, unlike realpath */
1083 static char *absolute_path(char *devname)
1085 char buf[PATH_MAX + 1];
1089 path = malloc(PATH_MAX + 1);
1093 if (devname[0] != '/') {
1094 if (getcwd(buf, sizeof(buf) - 1) == NULL) {
1099 strcat(buf, devname);
1101 strcpy(buf, devname);
1103 /* truncate filename before calling realpath */
1104 ptr = strrchr(buf, '/');
1110 if (path != realpath(buf, path)) {
1114 /* add the filename back */
1116 strcat(path, ptr + 1);
1120 /* Determine if a device is a block device (as opposed to a file) */
1121 int is_block(char* devname)
1127 devpath = absolute_path(devname);
1128 if (devpath == NULL) {
1129 fprintf(stderr, "%s: failed to resolve path to %s\n",
1134 ret = access(devname, F_OK);
1136 if (strncmp(devpath, "/dev/", 5) == 0) {
1137 /* nobody sane wants to create a loopback file under
1138 * /dev. Let's just report the device doesn't exist */
1139 fprintf(stderr, "%s: %s apparently does not exist\n",
1147 ret = stat(devpath, &st);
1149 fprintf(stderr, "%s: cannot stat %s\n", progname, devpath);
1152 ret = S_ISBLK(st.st_mode);
1158 static int is_feature_enabled(const char *feature, const char *devpath)
1162 char enabled_features[4096] = "";
1165 snprintf(cmd, sizeof(cmd), "%s -R features %s 2>&1",
1168 /* Using popen() instead of run_command() since debugfs does
1169 * not return proper error code if command is not supported */
1170 fp = popen(cmd, "r");
1172 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
1176 ret = fread(enabled_features, 1, sizeof(enabled_features) - 1, fp);
1177 enabled_features[ret] = '\0';
1180 if (strstr(enabled_features, feature))
1185 /* Enable quota accounting */
1186 int ldiskfs_enable_quota(struct mkfs_opts *mop)
1190 int cmdsz = sizeof(cmd), ret;
1192 if (is_e2fsprogs_feature_supp("-O quota") != 0) {
1193 fprintf(stderr, "%s: \"-O quota\" is is not supported by "
1194 "current e2fsprogs\n", progname);
1198 dev = mop->mo_device;
1199 if (mop->mo_flags & MO_IS_LOOP)
1200 dev = mop->mo_loopdev;
1202 /* Quota feature is already enabled? */
1203 if (is_feature_enabled("quota", dev)) {
1204 vprint("Quota feature is already enabled.\n");
1208 /* Turn on quota feature by "tune2fs -O quota" */
1209 snprintf(cmd, cmdsz, "%s -O quota %s", TUNE2FS, dev);
1210 ret = run_command(cmd, cmdsz);
1212 fprintf(stderr, "command:%s (%d)", cmd, ret);
1217 int ldiskfs_init(void)
1219 /* Required because full path to DEBUGFS is not specified */
1220 setenv("PATH", "/bin:/sbin:/usr/bin:/usr/sbin:/usr/local/bin", 0);
1225 void ldiskfs_fini(void)