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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/utils/mount_utils_ldiskfs.c
34 * Author: Nathan Rutman <nathan@clusterfs.com>
37 /* This source file is compiled into both mkfs.lustre and tunefs.lustre */
41 #endif /* HAVE_CONFIG_H */
56 #include <sys/types.h>
58 #include <sys/mount.h>
59 #include <sys/utsname.h>
60 #include <sys/sysmacros.h>
68 #include <linux/fs.h> /* for BLKGETSIZE64 */
70 #include <linux/major.h>
71 #include <linux/types.h>
72 #include <linux/version.h>
73 #include <linux/lnet/lnetctl.h>
74 #include <linux/lustre/lustre_ver.h>
75 #include <libcfs/util/string.h>
78 #include <selinux/selinux.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 SCHEDULER_PATH "queue/scheduler"
86 #define STRIPE_CACHE_SIZE "md/stripe_cache_size"
88 #define DEFAULT_SCHEDULER "deadline"
90 extern char *progname;
92 /* keep it less than LL_FID_NAMELEN */
93 #define DUMMY_FILE_NAME_LEN 25
94 #define EXT3_DIRENT_SIZE DUMMY_FILE_NAME_LEN
96 static void append_unique(char *buf, char *prefix, char *key, char *val,
98 static int is_e2fsprogs_feature_supp(const char *feature);
99 static void disp_old_e2fsprogs_msg(const char *feature, int make_backfs);
102 * Concatenate context of the temporary mount point if selinux is enabled
105 static 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 append_unique(mop->mo_ldd.ldd_mount_opts,
118 ",", "context", fcontext,
119 sizeof(mop->mo_ldd.ldd_mount_opts));
125 /* Determine if a device is a block device (as opposed to a file) */
126 static int is_block(char *devname)
132 ret = cfs_abs_path(devname, &devpath);
134 fprintf(stderr, "%s: failed to resolve path '%s': %s\n",
135 progname, devname, strerror(-ret));
139 ret = access(devname, F_OK);
141 if (strncmp(devpath, "/dev/", 5) == 0) {
142 /* nobody sane wants to create a loopback file under
143 * /dev. Let's just report the device doesn't exist */
144 fprintf(stderr, "%s: %s apparently does not exist\n",
152 ret = stat(devpath, &st);
154 fprintf(stderr, "%s: cannot stat %s\n", progname, devpath);
157 ret = S_ISBLK(st.st_mode);
163 static int is_feature_enabled(const char *feature, const char *devpath)
167 char enabled_features[4096] = "";
170 snprintf(cmd, sizeof(cmd), "%s -c -R features %s 2>&1",
173 /* Using popen() instead of run_command() since debugfs does
174 * not return proper error code if command is not supported */
175 fp = popen(cmd, "r");
177 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
181 ret = fread(enabled_features, 1, sizeof(enabled_features) - 1, fp);
182 enabled_features[ret] = '\0';
185 if (strstr(enabled_features, feature))
190 /* Write the server config files */
191 int ldiskfs_write_ldd(struct mkfs_opts *mop)
193 char mntpt[] = "/tmp/mntXXXXXX";
200 /* Mount this device temporarily in order to write these files */
201 if (!mkdtemp(mntpt)) {
202 fprintf(stderr, "%s: Can't create temp mount point %s: %s\n",
203 progname, mntpt, strerror(errno));
208 * Append file context to mount options if SE Linux is enabled
211 if (is_selinux_enabled() > 0)
212 append_context_for_mount(mntpt, mop);
215 dev = mop->mo_device;
216 if (mop->mo_flags & MO_IS_LOOP)
217 dev = mop->mo_loopdev;
219 /* Multiple mount protection enabled if failover node specified */
220 if (mop->mo_flags & MO_FAILOVER &&
221 !is_feature_enabled("mmp", dev)) {
222 if (is_e2fsprogs_feature_supp("-O mmp") == 0) {
223 char *command = filepnm;
225 snprintf(command, sizeof(filepnm),
226 TUNE2FS" -O mmp '%s' >/dev/null 2>&1", dev);
227 ret = run_command(command, sizeof(filepnm));
230 "%s: Unable to set 'mmp' on %s: %d\n",
233 disp_old_e2fsprogs_msg("mmp", 1);
236 ret = mount(dev, mntpt, MT_STR(&mop->mo_ldd), 0,
237 (mop->mo_mountopts == NULL) ?
238 "errors=remount-ro" : mop->mo_mountopts);
240 fprintf(stderr, "%s: Unable to mount %s: %s\n",
241 progname, dev, strerror(errno));
243 if (errno == ENODEV) {
244 fprintf(stderr, "Is the %s module available?\n",
245 MT_STR(&mop->mo_ldd));
250 /* Set up initial directories */
251 sprintf(filepnm, "%s/%s", mntpt, MOUNT_CONFIGS_DIR);
252 ret = mkdir(filepnm, 0777);
253 if ((ret != 0) && (errno != EEXIST)) {
254 fprintf(stderr, "%s: Can't make configs dir %s (%s)\n",
255 progname, filepnm, strerror(errno));
257 } else if (errno == EEXIST) {
261 /* Save the persistent mount data into a file. Lustre must pre-read
262 this file to get the real mount options. */
263 vprint("Writing %s\n", MOUNT_DATA_FILE);
264 sprintf(filepnm, "%s/%s", mntpt, MOUNT_DATA_FILE);
265 filep = fopen(filepnm, "w");
267 fprintf(stderr, "%s: Unable to create %s file: %s\n",
268 progname, filepnm, strerror(errno));
271 num = fwrite(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
272 if (num < 1 && ferror(filep)) {
273 fprintf(stderr, "%s: Unable to write to file (%s): %s\n",
274 progname, filepnm, strerror(errno));
278 fsync(filep->_fileno);
288 static int readcmd(char *cmd, char *buf, int len)
293 fp = popen(cmd, "r");
297 red = fread(buf, 1, len, fp);
300 /* strip trailing newline */
301 if (buf[red - 1] == '\n')
304 return (red == 0) ? -ENOENT : 0;
307 int ldiskfs_read_ldd(char *dev, struct lustre_disk_data *mo_ldd)
309 char tmpdir[] = "/tmp/dirXXXXXX";
314 int cmdsz = sizeof(cmd);
316 /* Make a temporary directory to hold Lustre data files. */
317 if (!mkdtemp(tmpdir)) {
318 fprintf(stderr, "%s: Can't create temporary directory %s: %s\n",
319 progname, tmpdir, strerror(errno));
323 /* TODO: it's worth observing the get_mountdata() function that is
324 in mount_utils.c for getting the mountdata out of the
327 /* Construct debugfs command line. */
328 snprintf(cmd, cmdsz, "%s -c -R 'dump /%s %s/mountdata' '%s'",
329 DEBUGFS, MOUNT_DATA_FILE, tmpdir, dev);
331 ret = run_command(cmd, cmdsz);
333 verrprint("%s: Unable to dump %s dir (%d)\n",
334 progname, MOUNT_CONFIGS_DIR, ret);
336 sprintf(filepnm, "%s/mountdata", tmpdir);
337 filep = fopen(filepnm, "r");
340 vprint("Reading %s\n", MOUNT_DATA_FILE);
341 num_read = fread(mo_ldd, sizeof(*mo_ldd), 1, filep);
342 if (num_read < 1 && ferror(filep)) {
343 fprintf(stderr, "%s: Unable to read from file %s: %s\n",
344 progname, filepnm, strerror(errno));
349 snprintf(cmd, cmdsz, "rm -rf %s", tmpdir);
350 run_command(cmd, cmdsz);
352 verrprint("Failed to read old data (%d)\n", ret);
354 /* As long as we at least have the label, we're good to go */
355 snprintf(cmd, sizeof(cmd), E2LABEL" %s", dev);
356 ret = readcmd(cmd, mo_ldd->ldd_svname, sizeof(mo_ldd->ldd_svname) - 1);
361 int ldiskfs_erase_ldd(struct mkfs_opts *mop, char *param)
366 void ldiskfs_print_ldd_params(struct mkfs_opts *mop)
368 printf("Parameters:%s\n", mop->mo_ldd.ldd_params);
371 /* Display the need for the latest e2fsprogs to be installed. make_backfs
372 * indicates if the caller is make_lustre_backfs() or not. */
373 static void disp_old_e2fsprogs_msg(const char *feature, int make_backfs)
375 static int msg_displayed;
378 fprintf(stderr, "WARNING: %s does not support %s "
379 "feature.\n\n", E2FSPROGS, feature);
385 fprintf(stderr, "WARNING: The %s package currently installed on "
386 "your system does not support \"%s\" feature.\n",
388 #if !(HAVE_LDISKFSPROGS)
389 fprintf(stderr, "Please install the latest version of e2fsprogs from\n"
390 "https://downloads.whamcloud.com/public/e2fsprogs/latest/\n"
391 "to enable this feature.\n");
394 fprintf(stderr, "Feature will not be enabled until %s"
395 "is updated and '%s -O %s %%{device}' "
396 "is run.\n\n", E2FSPROGS, TUNE2FS, feature);
399 /* Check whether the file exists in the device */
400 static int file_in_dev(char *file_name, char *dev_name)
403 char debugfs_cmd[256];
404 unsigned int inode_num;
407 /* Construct debugfs command line. */
408 snprintf(debugfs_cmd, sizeof(debugfs_cmd),
409 "%s -c -R 'stat %s' '%s' 2>&1 | egrep '(Inode|unsupported)'",
410 DEBUGFS, file_name, dev_name);
412 fp = popen(debugfs_cmd, "r");
414 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
418 if (fscanf(fp, "Inode: %u", &inode_num) == 1) { /* exist */
422 i = fread(debugfs_cmd, 1, sizeof(debugfs_cmd) - 1, fp);
425 fprintf(stderr, "%s", debugfs_cmd);
426 if (strstr(debugfs_cmd, "unsupported feature")) {
427 disp_old_e2fsprogs_msg("an unknown", 0);
436 /* Check whether the device has already been used with lustre */
437 int ldiskfs_is_lustre(char *dev, unsigned *mount_type)
441 ret = file_in_dev(MOUNT_DATA_FILE, dev);
443 /* in the -1 case, 'extents' means IS a lustre target */
444 *mount_type = LDD_MT_LDISKFS;
448 ret = file_in_dev(LAST_RCVD, dev);
450 *mount_type = LDD_MT_LDISKFS;
457 /* Check if a certain feature is supported by e2fsprogs.
458 * Firstly we try to use "debugfs supported_features" command to check if
459 * the feature is supported. If this fails we try to set this feature with
460 * mke2fs to check for its support. */
461 static int is_e2fsprogs_feature_supp(const char *feature)
463 static char supp_features[4096] = "";
466 char imgname[] = "/tmp/test-img-XXXXXX";
470 if (supp_features[0] == '\0') {
471 snprintf(cmd, sizeof(cmd), "%s -c -R supported_features 2>&1",
474 /* Using popen() instead of run_command() since debugfs does
475 * not return proper error code if command is not supported */
476 fp = popen(cmd, "r");
478 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
481 ret = fread(supp_features, 1, sizeof(supp_features) - 1, fp);
482 supp_features[ret] = '\0';
485 if (ret > 0 && strstr(supp_features,
486 strncmp(feature, "-O ", 3) ? feature : feature+3))
489 if ((fd = mkstemp(imgname)) < 0)
494 snprintf(cmd, sizeof(cmd), "%s -F %s %s 100 >/dev/null 2>&1",
495 MKE2FS, feature, imgname);
496 /* run_command() displays the output of mke2fs when it fails for
497 * some feature, so use system() directly */
506 * append_unique: append @key or @key=@val pair to @buf only if @key does not
508 * @buf: buffer to hold @key or @key=@val
509 * @prefix: prefix string before @key
511 * @val: value string if it's a @key=@val pair
513 static void append_unique(char *buf, char *prefix, char *key, char *val,
522 anchor = end = strstr(buf, key);
523 /* try to find exact match string in @buf */
524 while (end && *end != '\0' && *end != ',' && *end != ' ' && *end != '=')
527 if (anchor == NULL || strlen(key) != len ||
528 strncmp(anchor, key, len) != 0) {
530 strscat(buf, prefix, maxbuflen);
532 strscat(buf, key, maxbuflen);
534 strscat(buf, "=\"", maxbuflen);
535 strscat(buf, val, maxbuflen);
536 strscat(buf, "\"", maxbuflen);
541 static int enable_default_ext4_features(struct mkfs_opts *mop, char *anchor,
542 size_t maxbuflen, int user_spec)
544 int enable_64bit = 0;
546 /* Enable large block addresses if the LUN is over 2^32 blocks. */
547 if (mop->mo_device_kb / mop->mo_blocksize_kb > 0xffffffffULL &&
548 is_e2fsprogs_feature_supp("-O 64bit") == 0)
551 if (IS_OST(&mop->mo_ldd)) {
552 append_unique(anchor, user_spec ? "," : " -O ",
553 "extents", NULL, maxbuflen);
554 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
555 } else if (IS_MDT(&mop->mo_ldd)) {
556 append_unique(anchor, user_spec ? "," : " -O ",
557 "dirdata", NULL, maxbuflen);
558 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
560 append_unique(anchor, ",", "extents", NULL, maxbuflen);
562 append_unique(anchor, ",", "^extents", NULL, maxbuflen);
564 append_unique(anchor, user_spec ? "," : " -O ",
565 "uninit_bg", NULL, maxbuflen);
568 /* Multiple mount protection enabled only if failover node specified */
569 if (mop->mo_flags & MO_FAILOVER) {
570 if (is_e2fsprogs_feature_supp("-O mmp") == 0)
571 append_unique(anchor, ",", "mmp", NULL, maxbuflen);
573 disp_old_e2fsprogs_msg("mmp", 1);
576 /* Allow more than 65000 subdirectories */
577 if (is_e2fsprogs_feature_supp("-O dir_nlink") == 0)
578 append_unique(anchor, ",", "dir_nlink", NULL, maxbuflen);
580 /* The following options are only valid for ext4-based ldiskfs.
581 * If --backfstype=ext3 is specified, do not enable them. */
582 if (mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3)
585 /* Enable quota by default */
586 if (is_e2fsprogs_feature_supp("-O quota") == 0) {
587 append_unique(anchor, ",", "quota", NULL, maxbuflen);
590 fprintf(stderr, "\"-O quota\" must be supported by "
591 "e2fsprogs, please upgrade your e2fsprogs.\n");
595 /* Allow files larger than 2TB. Also needs LU-16, but not harmful. */
596 if (is_e2fsprogs_feature_supp("-O huge_file") == 0)
597 append_unique(anchor, ",", "huge_file", NULL, maxbuflen);
600 append_unique(anchor, ",", "64bit", NULL, maxbuflen);
602 /* Cluster inode/block bitmaps and inode table for more efficient IO.
603 * Align the flex groups on a 1MB boundary for better performance. */
604 /* This -O feature needs to go last, since it adds the "-G" option. */
605 if (is_e2fsprogs_feature_supp("-O flex_bg") == 0) {
608 append_unique(anchor, ",", "flex_bg", NULL, maxbuflen);
610 if (IS_OST(&mop->mo_ldd) &&
611 strstr(mop->mo_mkfsopts, "-G") == NULL) {
612 snprintf(tmp_buf, sizeof(tmp_buf), " -G %u",
613 1024 / mop->mo_blocksize_kb);
614 strscat(anchor, tmp_buf, maxbuflen);
617 /* Don't add any more "-O" options here, see last comment above */
622 * moveopts_to_end: find the option string, move remaining strings to
623 * where option string starts, and append the option
625 * @start: where the option string starts before the move
626 * RETURN: where the option string starts after the move
628 static char *moveopts_to_end(char *start)
634 /* skip whitespace before options */
639 /* find end of option characters */
640 while (*end != ' ' && *end != '\0')
644 if (len >= sizeof(save))
645 len = sizeof(save) - 1;
648 strncpy(save, start, len);
651 /* move remaining options up front */
653 memmove(start, end, strlen(end));
654 *(start + strlen(end)) = '\0';
656 /* append the specified options */
657 if (*(start + strlen(start) - 1) != ' ')
659 idx = start + strlen(start);
665 /* Build fs according to type */
666 int ldiskfs_make_lustre(struct mkfs_opts *mop)
668 char mkfs_cmd[PATH_MAX];
672 int ret = 0, ext_opts = 0;
673 bool have_64bit = false;
676 mop->mo_blocksize_kb = 4;
678 start = strstr(mop->mo_mkfsopts, "-b");
683 blocksize = strtol(start + 2, &end, 0);
684 if (end && (*end == 'k' || *end == 'K'))
686 /* EXT4_MIN_BLOCK_SIZE || EXT4_MAX_BLOCK_SIZE */
687 if (blocksize < 1024 || blocksize > 65536) {
689 "%s: blocksize %lu not in 1024-65536 bytes, normally 4096 bytes\n",
690 progname, blocksize);
694 if ((blocksize & (blocksize - 1)) != 0) {
696 "%s: blocksize %lu not a power-of-two value\n",
697 progname, blocksize);
700 mop->mo_blocksize_kb = blocksize >> 10;
703 if (!(mop->mo_flags & MO_IS_LOOP)) {
704 __u64 device_kb = get_device_size(mop->mo_device);
709 /* Compare to real size */
710 if (mop->mo_device_kb == 0 || device_kb < mop->mo_device_kb)
711 mop->mo_device_kb = device_kb;
714 if (mop->mo_device_kb != 0) {
717 if (mop->mo_device_kb < 32384) {
718 fprintf(stderr, "%s: size of filesystem must be larger "
719 "than 32MB, but is set to %lldKB\n",
720 progname, (long long)mop->mo_device_kb);
723 block_count = mop->mo_device_kb / mop->mo_blocksize_kb;
724 if (block_count > 0xffffffffULL) {
725 /* If the LUN size is just over 2^32 blocks, limit the
726 * filesystem size to 2^32-1 blocks to avoid problems
727 * with ldiskfs/mkfs not handling this well. b=22906
729 if (block_count < 0x100002000ULL)
731 0xffffffffULL * mop->mo_blocksize_kb;
738 if ((mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3) ||
739 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS) ||
740 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS2)) {
743 /* Journal size in MB */
744 if (strstr(mop->mo_mkfsopts, "-J") == NULL &&
745 mop->mo_device_kb > 1024 * 1024) {
746 /* Choose our own default journal size */
747 long journal_mb = 0, max_mb;
749 /* cap journal size at 4GB for MDT,
750 * leave it at 400MB for OSTs. */
751 if (IS_MDT(&mop->mo_ldd))
753 else if (IS_OST(&mop->mo_ldd))
755 else /* Use mke2fs default size for MGS */
758 /* Use at most 4% of device for journal */
759 journal_mb = mop->mo_device_kb * 4 / (1024 * 100);
760 if (journal_mb > max_mb)
764 sprintf(buf, " -J size=%ld", journal_mb);
765 strscat(mop->mo_mkfsopts, buf,
766 sizeof(mop->mo_mkfsopts));
771 * The inode size is constituted by following elements
772 * (assuming all files are in composite layout and has
775 * ldiskfs inode size: 160
776 * MDT extended attributes size, including:
777 * ext4_xattr_header: 32
778 * LOV EA size: 32(lov_comp_md_v1) +
779 * 3 * 40(lov_comp_md_entry_v1) +
780 * 3 * 32(lov_mds_md) +
781 * stripes * 24(lov_ost_data) +
782 * 16(xattr_entry) + 4("lov")
783 * LMA EA size: 24(lustre_mdt_attrs) +
784 * 16(xattr_entry) + 4("lma")
785 * SOM EA size: 24(lustre_som_attrs) +
786 * 16(xattr_entry) + 4("som")
787 * link EA size: 24(link_ea_header) + 18(link_ea_entry) +
788 * 16(filename) + 16(xattr_entry) + 4("link")
789 * and some margin for 4-byte alignment, ACLs and other EAs.
791 * If we say the average filename length is about 32 bytes,
792 * the calculation looks like:
793 * 160 + 32 + (32+3*(40+32)+24*stripes+20) + (24+20) + (24+20) +
794 * (24+20) + (~42+16+20) + other <= 512*2^m, {m=0,1,2,3}
796 if (strstr(mop->mo_mkfsopts, "-I") == NULL) {
797 if (IS_MDT(&mop->mo_ldd)) {
798 if (mop->mo_stripe_count > 59)
799 inode_size = 512; /* bz 7241 */
800 /* see also "-i" below for EA blocks */
801 else if (mop->mo_stripe_count > 16)
805 } else if (IS_OST(&mop->mo_ldd)) {
806 /* We store MDS FID and necessary composite
807 * layout information in the OST object EA:
808 * ldiskfs inode size: 160
809 * OST extended attributes size, including:
810 * ext4_xattr_header: 32
811 * LMA EA size: 24(lustre_mdt_attrs) +
812 * 16(xattr_entry) + 4("lma")
813 * FID EA size: 52(filter_fid) +
814 * 16(xattr_entry) + 4("fid")
815 * 160 + 32 + (24+20) + (52+20) = 308
820 if (inode_size > 0) {
821 sprintf(buf, " -I %ld", inode_size);
822 strscat(mop->mo_mkfsopts, buf,
823 sizeof(mop->mo_mkfsopts));
827 /* Bytes_per_inode: disk size / num inodes */
828 if (strstr(mop->mo_mkfsopts, "-i") == NULL &&
829 strstr(mop->mo_mkfsopts, "-N") == NULL) {
830 long bytes_per_inode = 0;
832 /* Allocate more inodes on MDT devices. There is
833 * no data stored on the MDT, and very little extra
834 * metadata beyond the inode. It could go down as
835 * low as 1024 bytes, but this is conservative.
836 * Account for external EA blocks for wide striping. */
837 if (IS_MDT(&mop->mo_ldd)) {
838 bytes_per_inode = inode_size + 1536;
840 if (mop->mo_stripe_count > 59) {
841 int extra = mop->mo_stripe_count * 24;
842 extra = ((extra - 1) | 4095) + 1;
843 bytes_per_inode += extra;
847 /* Allocate fewer inodes on large OST devices. Most
848 * filesystems can be much more aggressive than even
849 * this, but it is impossible to know in advance. */
850 if (IS_OST(&mop->mo_ldd)) {
851 /* OST > 16TB assume average file size 1MB */
852 if (mop->mo_device_kb > (16ULL << 30))
853 bytes_per_inode = 1024 * 1024;
854 /* OST > 4TB assume average file size 512kB */
855 else if (mop->mo_device_kb > (4ULL << 30))
856 bytes_per_inode = 512 * 1024;
857 /* OST > 1TB assume average file size 256kB */
858 else if (mop->mo_device_kb > (1ULL << 30))
859 bytes_per_inode = 256 * 1024;
860 /* OST > 10GB assume average file size 64kB,
861 * plus a bit so that inodes will fit into a
862 * 256x flex_bg without overflowing */
863 else if (mop->mo_device_kb > (10ULL << 20))
864 bytes_per_inode = 69905;
867 if (bytes_per_inode > 0) {
868 sprintf(buf, " -i %ld", bytes_per_inode);
869 strscat(mop->mo_mkfsopts, buf,
870 sizeof(mop->mo_mkfsopts));
871 mop->mo_inode_size = bytes_per_inode;
876 strscat(mop->mo_mkfsopts, " -q",
877 sizeof(mop->mo_mkfsopts));
880 /* start handle -O mkfs options */
881 if ((start = strstr(mop->mo_mkfsopts, "-O")) != NULL) {
882 if (strstr(start + 2, "-O") != NULL) {
884 "%s: don't specify multiple -O options\n",
888 start = moveopts_to_end(start);
889 maxbuflen = sizeof(mop->mo_mkfsopts) -
890 (start - mop->mo_mkfsopts) - strlen(start);
891 ret = enable_default_ext4_features(mop, start,
894 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts),
895 maxbuflen = sizeof(mop->mo_mkfsopts) -
896 strlen(mop->mo_mkfsopts);
897 ret = enable_default_ext4_features(mop, start,
902 /* end handle -O mkfs options */
904 /* start handle -E mkfs options */
905 if ((start = strstr(mop->mo_mkfsopts, "-E")) != NULL) {
906 if (strstr(start + 2, "-E") != NULL) {
908 "%s: don't specify multiple -E options\n",
912 start = moveopts_to_end(start);
913 maxbuflen = sizeof(mop->mo_mkfsopts) -
914 (start - mop->mo_mkfsopts) - strlen(start);
917 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts);
918 maxbuflen = sizeof(mop->mo_mkfsopts) -
919 strlen(mop->mo_mkfsopts);
922 /* In order to align the filesystem metadata on 1MB boundaries,
923 * give a resize value that will reserve a power-of-two group
924 * descriptor blocks, but leave one block for the superblock.
925 * Only useful for filesystems with < 2^32 blocks due to resize
927 if (strstr(mop->mo_mkfsopts, "meta_bg") == NULL &&
928 IS_OST(&mop->mo_ldd) && mop->mo_device_kb > 100 * 1024 &&
930 unsigned int group_blocks = mop->mo_blocksize_kb * 8192;
931 unsigned int desc_per_block =
932 mop->mo_blocksize_kb * 1024 / 32;
933 unsigned int resize_blks;
935 resize_blks = (1ULL<<32) - desc_per_block*group_blocks;
936 snprintf(buf, sizeof(buf), "%u", resize_blks);
937 append_unique(start, ext_opts ? "," : " -E ",
938 "resize", buf, maxbuflen);
942 /* Avoid zeroing out the full journal - speeds up mkfs */
943 if (is_e2fsprogs_feature_supp("-E lazy_journal_init") == 0)
944 append_unique(start, ext_opts ? "," : " -E ",
945 "lazy_journal_init", NULL, maxbuflen);
946 /* end handle -E mkfs options */
948 /* Allow reformat of full devices (as opposed to
949 partitions.) We already checked for mounted dev. */
950 strscat(mop->mo_mkfsopts, " -F", sizeof(mop->mo_mkfsopts));
952 snprintf(mkfs_cmd, sizeof(mkfs_cmd),
953 "%s -j -b %d -L %s ", MKE2FS,
954 mop->mo_blocksize_kb * 1024, mop->mo_ldd.ldd_svname);
956 fprintf(stderr,"%s: unsupported fs type: %d (%s)\n",
957 progname, mop->mo_ldd.ldd_mount_type,
958 MT_STR(&mop->mo_ldd));
962 /* For loop device format the dev, not the filename */
963 dev = mop->mo_device;
964 if (mop->mo_flags & MO_IS_LOOP)
965 dev = mop->mo_loopdev;
967 vprint("formatting backing filesystem %s on %s\n",
968 MT_STR(&mop->mo_ldd), dev);
969 vprint("\ttarget name %s\n", mop->mo_ldd.ldd_svname);
970 vprint("\tkilobytes %llu\n", mop->mo_device_kb);
971 vprint("\toptions %s\n", mop->mo_mkfsopts);
973 /* mkfs_cmd's trailing space is important! */
974 strscat(mkfs_cmd, mop->mo_mkfsopts, sizeof(mkfs_cmd));
975 strscat(mkfs_cmd, " ", sizeof(mkfs_cmd));
976 strscat(mkfs_cmd, dev, sizeof(mkfs_cmd));
977 if (mop->mo_device_kb != 0) {
978 snprintf(buf, sizeof(buf), " %lluk",
979 (unsigned long long)mop->mo_device_kb);
980 strscat(mkfs_cmd, buf, sizeof(mkfs_cmd));
983 vprint("mkfs_cmd = %s\n", mkfs_cmd);
984 ret = run_command(mkfs_cmd, sizeof(mkfs_cmd));
987 fprintf(stderr, "Unable to build fs %s (%d)\n", dev, ret);
992 int ldiskfs_prepare_lustre(struct mkfs_opts *mop,
993 char *wanted_mountopts, size_t len)
995 struct lustre_disk_data *ldd = &mop->mo_ldd;
998 /* Set MO_IS_LOOP to indicate a loopback device is needed */
999 ret = is_block(mop->mo_device);
1002 } else if (ret == 0) {
1003 mop->mo_flags |= MO_IS_LOOP;
1006 if (IS_MDT(ldd) || IS_MGS(ldd))
1007 strscat(wanted_mountopts, ",user_xattr", len);
1012 int ldiskfs_fix_mountopts(struct mkfs_opts *mop, char *mountopts, size_t len)
1014 if (strstr(mountopts, "errors=") == NULL)
1015 strscat(mountopts, ",errors=remount-ro", len);
1020 static int read_file(const char *path, char *buf, int size)
1024 fd = fopen(path, "r");
1028 if (fgets(buf, size, fd) == NULL) {
1029 fprintf(stderr, "reading from %s: %s", path, strerror(errno));
1035 /* strip trailing newline */
1037 if (buf[size - 1] == '\n')
1038 buf[size - 1] = '\0';
1043 static int write_file(const char *path, const char *buf)
1047 fd = open(path, O_WRONLY);
1051 rc = write(fd, buf, strlen(buf));
1054 return rc < 0 ? errno : 0;
1057 static int tune_md_stripe_cache_size(const char *sys_path,
1058 struct mount_opts *mop)
1060 char path[PATH_MAX];
1061 unsigned long old_stripe_cache_size;
1062 unsigned long new_stripe_cache_size;
1063 char buf[3 * sizeof(old_stripe_cache_size) + 2];
1066 if (mop->mo_md_stripe_cache_size <= 0)
1069 new_stripe_cache_size = mop->mo_md_stripe_cache_size;
1071 snprintf(path, sizeof(path), "%s/%s", sys_path, STRIPE_CACHE_SIZE);
1072 rc = read_file(path, buf, sizeof(buf));
1075 fprintf(stderr, "warning: cannot read '%s': %s\n",
1076 path, strerror(errno));
1080 old_stripe_cache_size = strtoul(buf, NULL, 0);
1081 if (old_stripe_cache_size == 0 || old_stripe_cache_size == ULONG_MAX)
1084 if (new_stripe_cache_size <= old_stripe_cache_size)
1087 snprintf(buf, sizeof(buf), "%lu", new_stripe_cache_size);
1088 rc = write_file(path, buf);
1091 fprintf(stderr, "warning: cannot write '%s': %s\n",
1092 path, strerror(errno));
1099 static int tune_max_sectors_kb(const char *sys_path, struct mount_opts *mop)
1101 char path[PATH_MAX];
1102 unsigned long max_hw_sectors_kb;
1103 unsigned long old_max_sectors_kb;
1104 unsigned long new_max_sectors_kb;
1105 char buf[3 * sizeof(old_max_sectors_kb) + 2];
1108 if (mop->mo_max_sectors_kb >= 0) {
1109 new_max_sectors_kb = mop->mo_max_sectors_kb;
1110 goto have_new_max_sectors_kb;
1113 snprintf(path, sizeof(path), "%s/%s", sys_path, MAX_HW_SECTORS_KB_PATH);
1114 rc = read_file(path, buf, sizeof(buf));
1116 /* No MAX_HW_SECTORS_KB_PATH isn't necessary an
1117 * error for some devices. */
1121 max_hw_sectors_kb = strtoul(buf, NULL, 0);
1122 if (max_hw_sectors_kb == 0 || max_hw_sectors_kb == ULLONG_MAX) {
1123 /* No digits at all or something weird. */
1127 new_max_sectors_kb = max_hw_sectors_kb;
1129 /* Don't increase IO request size limit past 16MB. It is
1130 * about PTLRPC_MAX_BRW_SIZE, but that isn't in a public
1131 * header. Note that even though the block layer allows
1132 * larger values, setting max_sectors_kb = 32768 causes
1133 * crashes (LU-6974). */
1134 if (new_max_sectors_kb > 16 * 1024)
1135 new_max_sectors_kb = 16 * 1024;
1137 have_new_max_sectors_kb:
1138 snprintf(path, sizeof(path), "%s/%s", sys_path, MAX_SECTORS_KB_PATH);
1139 rc = read_file(path, buf, sizeof(buf));
1141 /* No MAX_SECTORS_KB_PATH isn't necessary an error for
1146 old_max_sectors_kb = strtoul(buf, NULL, 0);
1147 if (old_max_sectors_kb == 0 || old_max_sectors_kb == ULLONG_MAX) {
1148 /* No digits at all or something weird. */
1152 if (new_max_sectors_kb <= old_max_sectors_kb)
1155 snprintf(buf, sizeof(buf), "%lu", new_max_sectors_kb);
1156 rc = write_file(path, buf);
1159 fprintf(stderr, "warning: cannot write '%s': %s\n",
1160 path, strerror(errno));
1164 fprintf(stderr, "%s: increased '%s' from %lu to %lu\n",
1165 progname, path, old_max_sectors_kb, new_max_sectors_kb);
1170 static int tune_block_dev_scheduler(const char *sys_path, const char *new_sched)
1172 char path[PATH_MAX];
1178 /* Before setting the scheduler, we need to check to see if
1179 * it's already set to "noop". If it is then we don't want to
1180 * override that setting. If it's set to anything other than
1181 * "noop" then set the scheduler to what has been passed
1184 snprintf(path, sizeof(path), "%s/%s", sys_path, SCHEDULER_PATH);
1185 rc = read_file(path, buf, sizeof(buf));
1188 fprintf(stderr, "%s: cannot read '%s': %s\n",
1189 progname, path, strerror(errno));
1194 /* The expected format of buf: noop anticipatory deadline [cfq] */
1195 s = strchr(buf, '[');
1196 e = strchr(buf, ']');
1198 /* If the format is not what we expect then be safe and error out. */
1199 if (s == NULL || e == NULL || !(s < e)) {
1202 "%s: cannot parse scheduler options for '%s'\n",
1211 if (strcmp(old_sched, "noop") == 0 ||
1212 strcmp(old_sched, "deadline") == 0 ||
1213 strcmp(old_sched, "mq-deadline") == 0 ||
1214 strstr(old_sched, new_sched) == 0)
1217 rc = write_file(path, new_sched);
1221 "%s: cannot set scheduler on '%s': %s\n",
1222 progname, path, strerror(errno));
1226 fprintf(stderr, "%s: changed scheduler of '%s' from %s to %s\n",
1227 progname, path, old_sched, new_sched);
1232 static int tune_block_dev(const char *src, struct mount_opts *mop);
1234 static int tune_block_dev_slaves(const char *sys_path, struct mount_opts *mop)
1236 char slaves_path[PATH_MAX];
1241 snprintf(slaves_path, sizeof(slaves_path), "%s/slaves", sys_path);
1242 slaves_dir = opendir(slaves_path);
1243 if (slaves_dir == NULL) {
1244 if (errno == ENOENT)
1250 while ((d = readdir(slaves_dir)) != NULL) {
1251 char path[PATH_MAX * 2];
1254 if (d->d_type != DT_LNK)
1257 snprintf(path, sizeof(path), "/dev/%s", d->d_name);
1258 rc2 = tune_block_dev(path, mop);
1263 closedir(slaves_dir);
1268 /* This is to tune the kernel for good SCSI performance.
1269 * For that we set the value of /sys/block/{dev}/queue/max_sectors_kb
1270 * to the value of /sys/block/{dev}/queue/max_hw_sectors_kb */
1271 static int tune_block_dev(const char *src, struct mount_opts *mop)
1274 char sys_path[PATH_MAX];
1275 char partition_path[PATH_MAX + sizeof("partition")];
1276 char *real_sys_path = NULL;
1280 * Don't apply block device tuning for MDT or MGT devices,
1281 * since we don't need huge IO sizes to get good performance
1283 if (!IS_OST(&mop->mo_ldd))
1289 rc = stat(src, &st);
1292 fprintf(stderr, "warning: cannot stat '%s': %s\n",
1293 src, strerror(errno));
1297 if (!S_ISBLK(st.st_mode))
1300 if (major(st.st_rdev) == LOOP_MAJOR)
1303 snprintf(sys_path, sizeof(sys_path), "/sys/dev/block/%u:%u",
1304 major(st.st_rdev), minor(st.st_rdev));
1306 snprintf(partition_path, sizeof(partition_path), "%s/partition",
1309 rc = access(partition_path, F_OK);
1311 if (errno == ENOENT)
1312 goto have_whole_dev;
1315 fprintf(stderr, "warning: cannot access '%s': %s\n",
1316 partition_path, strerror(errno));
1321 snprintf(sys_path, sizeof(sys_path), "/sys/dev/block/%u:%u/..",
1322 major(st.st_rdev), minor(st.st_rdev));
1325 /* Since we recurse on slave devices we resolve the sys_path to
1326 * avoid path buffer overflows. */
1327 real_sys_path = realpath(sys_path, NULL);
1328 if (real_sys_path == NULL) {
1331 "warning: cannot resolve '%s': %s\n",
1332 sys_path, strerror(errno));
1337 if (major(st.st_rdev) == MD_MAJOR) {
1338 rc = tune_md_stripe_cache_size(real_sys_path, mop);
1340 /* Ignore errors from tune_max_sectors_kb() and
1341 * tune_scheduler(). The worst that will happen is a block
1342 * device with an "incorrect" scheduler. */
1343 tune_max_sectors_kb(real_sys_path, mop);
1344 tune_block_dev_scheduler(real_sys_path, DEFAULT_SCHEDULER);
1346 /* If device is multipath device then tune its slave
1348 rc = tune_block_dev_slaves(real_sys_path, mop);
1352 free(real_sys_path);
1357 int ldiskfs_tune_lustre(char *dev, struct mount_opts *mop)
1359 return tune_block_dev(dev, mop);
1362 int ldiskfs_label_lustre(struct mount_opts *mop)
1364 char label_cmd[PATH_MAX];
1367 snprintf(label_cmd, sizeof(label_cmd),
1368 TUNE2FS" -f -L '%s' '%s' >/dev/null 2>&1",
1369 mop->mo_ldd.ldd_svname, mop->mo_source);
1370 rc = run_command(label_cmd, sizeof(label_cmd));
1375 int ldiskfs_rename_fsname(struct mkfs_opts *mop, const char *oldname)
1377 struct mount_opts opts;
1378 struct lustre_disk_data *ldd = &mop->mo_ldd;
1379 char mntpt[] = "/tmp/mntXXXXXX";
1383 /* Change the filesystem label. */
1385 opts.mo_source = mop->mo_device;
1386 ret = ldiskfs_label_lustre(&opts);
1390 fprintf(stderr, "Can't change filesystem label: %s\n",
1395 /* Mount this device temporarily in order to write these files */
1396 if (mkdtemp(mntpt) == NULL) {
1401 fprintf(stderr, "Can't create temp mount point %s: %s\n",
1402 mntpt, strerror(ret));
1408 * Append file context to mount options if SE Linux is enabled
1410 if (is_selinux_enabled() > 0)
1411 append_context_for_mount(mntpt, mop);
1414 if (mop->mo_flags & MO_IS_LOOP)
1415 dev = mop->mo_loopdev;
1417 dev = mop->mo_device;
1418 ret = mount(dev, mntpt, MT_STR(ldd), 0, ldd->ldd_mount_opts);
1422 fprintf(stderr, "Unable to mount %s: %s\n",
1423 dev, strerror(ret));
1425 fprintf(stderr, "Is the %s module available?\n",
1430 ret = lustre_rename_fsname(mop, mntpt, oldname);
1438 /* Enable quota accounting */
1439 int ldiskfs_enable_quota(struct mkfs_opts *mop)
1443 int cmdsz = sizeof(cmd), ret;
1445 if (is_e2fsprogs_feature_supp("-O quota") != 0) {
1446 fprintf(stderr, "%s: \"-O quota\" is is not supported by "
1447 "current e2fsprogs\n", progname);
1451 dev = mop->mo_device;
1452 if (mop->mo_flags & MO_IS_LOOP)
1453 dev = mop->mo_loopdev;
1455 /* Quota feature is already enabled? */
1456 if (is_feature_enabled("quota", dev)) {
1457 vprint("Quota feature is already enabled.\n");
1461 /* Turn on quota feature by "tune2fs -O quota" */
1462 snprintf(cmd, cmdsz, "%s -O quota %s", TUNE2FS, dev);
1463 ret = run_command(cmd, cmdsz);
1465 fprintf(stderr, "command:%s (%d)", cmd, ret);
1470 int ldiskfs_init(void)
1472 /* Required because full path to DEBUGFS is not specified */
1473 setenv("PATH", "/bin:/sbin:/usr/bin:/usr/sbin:/usr/local/bin", 0);
1478 void ldiskfs_fini(void)
1484 struct module_backfs_ops ldiskfs_ops = {
1485 .init = ldiskfs_init,
1486 .fini = ldiskfs_fini,
1487 .read_ldd = ldiskfs_read_ldd,
1488 .write_ldd = ldiskfs_write_ldd,
1489 .erase_ldd = ldiskfs_erase_ldd,
1490 .print_ldd_params = ldiskfs_print_ldd_params,
1491 .is_lustre = ldiskfs_is_lustre,
1492 .make_lustre = ldiskfs_make_lustre,
1493 .prepare_lustre = ldiskfs_prepare_lustre,
1494 .fix_mountopts = ldiskfs_fix_mountopts,
1495 .tune_lustre = ldiskfs_tune_lustre,
1496 .label_lustre = ldiskfs_label_lustre,
1497 .enable_quota = ldiskfs_enable_quota,
1498 .rename_fsname = ldiskfs_rename_fsname,
1500 #endif /* PLUGIN_DIR */