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 bool 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")) {
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);
237 ret = mount(dev, mntpt, MT_STR(&mop->mo_ldd), 0,
238 (mop->mo_mountopts == NULL) ?
239 "errors=remount-ro" : mop->mo_mountopts);
241 fprintf(stderr, "%s: Unable to mount %s: %s\n",
242 progname, dev, strerror(errno));
244 if (errno == ENODEV) {
245 fprintf(stderr, "Is the %s module available?\n",
246 MT_STR(&mop->mo_ldd));
251 /* Set up initial directories */
252 sprintf(filepnm, "%s/%s", mntpt, MOUNT_CONFIGS_DIR);
253 ret = mkdir(filepnm, 0777);
254 if ((ret != 0) && (errno != EEXIST)) {
255 fprintf(stderr, "%s: Can't make configs dir %s (%s)\n",
256 progname, filepnm, strerror(errno));
258 } else if (errno == EEXIST) {
262 /* Save the persistent mount data into a file. Lustre must pre-read
263 this file to get the real mount options. */
264 vprint("Writing %s\n", MOUNT_DATA_FILE);
265 sprintf(filepnm, "%s/%s", mntpt, MOUNT_DATA_FILE);
266 filep = fopen(filepnm, "w");
268 fprintf(stderr, "%s: Unable to create %s file: %s\n",
269 progname, filepnm, strerror(errno));
272 num = fwrite(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
273 if (num < 1 && ferror(filep)) {
274 fprintf(stderr, "%s: Unable to write to file (%s): %s\n",
275 progname, filepnm, strerror(errno));
279 fsync(filep->_fileno);
289 static int readcmd(char *cmd, char *buf, int len)
294 fp = popen(cmd, "r");
298 red = fread(buf, 1, len, fp);
301 /* strip trailing newline */
302 if (buf[red - 1] == '\n')
305 return (red == 0) ? -ENOENT : 0;
308 int ldiskfs_read_ldd(char *dev, struct lustre_disk_data *mo_ldd)
310 char tmpdir[] = "/tmp/dirXXXXXX";
315 int cmdsz = sizeof(cmd);
317 /* Make a temporary directory to hold Lustre data files. */
318 if (!mkdtemp(tmpdir)) {
319 fprintf(stderr, "%s: Can't create temporary directory %s: %s\n",
320 progname, tmpdir, strerror(errno));
324 /* TODO: it's worth observing the get_mountdata() function that is
325 in mount_utils.c for getting the mountdata out of the
328 /* Construct debugfs command line. */
329 snprintf(cmd, cmdsz, "%s -c -R 'dump /%s %s/mountdata' '%s'",
330 DEBUGFS, MOUNT_DATA_FILE, tmpdir, dev);
332 ret = run_command(cmd, cmdsz);
334 verrprint("%s: Unable to dump %s dir (%d)\n",
335 progname, MOUNT_CONFIGS_DIR, ret);
337 sprintf(filepnm, "%s/mountdata", tmpdir);
338 filep = fopen(filepnm, "r");
341 vprint("Reading %s\n", MOUNT_DATA_FILE);
342 num_read = fread(mo_ldd, sizeof(*mo_ldd), 1, filep);
343 if (num_read < 1 && ferror(filep)) {
344 fprintf(stderr, "%s: Unable to read from file %s: %s\n",
345 progname, filepnm, strerror(errno));
350 snprintf(cmd, cmdsz, "rm -rf %s", tmpdir);
351 run_command(cmd, cmdsz);
353 verrprint("Failed to read old data (%d)\n", ret);
355 /* As long as we at least have the label, we're good to go */
356 snprintf(cmd, sizeof(cmd), E2LABEL" %s", dev);
357 ret = readcmd(cmd, mo_ldd->ldd_svname, sizeof(mo_ldd->ldd_svname) - 1);
362 int ldiskfs_erase_ldd(struct mkfs_opts *mop, char *param)
367 void ldiskfs_print_ldd_params(struct mkfs_opts *mop)
369 printf("Parameters:%s\n", mop->mo_ldd.ldd_params);
372 /* Display the need for the latest e2fsprogs to be installed. make_backfs
373 * indicates if the caller is make_lustre_backfs() or not. */
374 static void disp_old_e2fsprogs_msg(const char *feature, int make_backfs)
376 static int msg_displayed;
379 fprintf(stderr, "WARNING: %s does not support %s "
380 "feature.\n\n", E2FSPROGS, feature);
386 fprintf(stderr, "WARNING: The %s package currently installed on "
387 "your system does not support \"%s\" feature.\n",
389 #if !(HAVE_LDISKFSPROGS)
390 fprintf(stderr, "Please install the latest version of e2fsprogs from\n"
391 "https://downloads.whamcloud.com/public/e2fsprogs/latest/\n"
392 "to enable this feature.\n");
395 fprintf(stderr, "Feature will not be enabled until %s"
396 "is updated and '%s -O %s %%{device}' "
397 "is run.\n\n", E2FSPROGS, TUNE2FS, feature);
400 /* Check whether the file exists in the device */
401 static int file_in_dev(char *file_name, char *dev_name)
404 char debugfs_cmd[256];
405 unsigned int inode_num;
408 /* Construct debugfs command line. */
409 snprintf(debugfs_cmd, sizeof(debugfs_cmd),
410 "%s -c -R 'stat %s' '%s' 2>&1 | egrep '(Inode|unsupported)'",
411 DEBUGFS, file_name, dev_name);
413 fp = popen(debugfs_cmd, "r");
415 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
419 if (fscanf(fp, "Inode: %u", &inode_num) == 1) { /* exist */
423 i = fread(debugfs_cmd, 1, sizeof(debugfs_cmd) - 1, fp);
426 fprintf(stderr, "%s", debugfs_cmd);
427 if (strstr(debugfs_cmd, "unsupported feature")) {
428 disp_old_e2fsprogs_msg("an unknown", 0);
437 /* Check whether the device has already been used with lustre */
438 int ldiskfs_is_lustre(char *dev, unsigned *mount_type)
442 ret = file_in_dev(MOUNT_DATA_FILE, dev);
444 /* in the -1 case, 'extents' means IS a lustre target */
445 *mount_type = LDD_MT_LDISKFS;
449 ret = file_in_dev(LAST_RCVD, dev);
451 *mount_type = LDD_MT_LDISKFS;
458 /* Check if a certain feature is supported by e2fsprogs.
459 * Firstly we try to use "debugfs supported_features" command to check if
460 * the feature is supported. If this fails we try to set this feature with
461 * mke2fs to check for its support. */
462 static bool is_e2fsprogs_feature_supp(const char *feature)
464 static char supp_features[4096] = "";
467 char imgname[] = "/tmp/test-img-XXXXXX";
471 if (supp_features[0] == '\0') {
472 snprintf(cmd, sizeof(cmd), "%s -c -R supported_features 2>&1",
475 /* Using popen() instead of run_command() since debugfs does
476 * not return proper error code if command is not supported */
477 fp = popen(cmd, "r");
479 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
481 ret = fread(supp_features, 1,
482 sizeof(supp_features) - 1, fp);
483 supp_features[ret] = '\0';
488 if (strstr(supp_features,
489 strncmp(feature, "-O ", 3) ? feature : feature + 3))
492 if ((fd = mkstemp(imgname)) < 0)
497 snprintf(cmd, sizeof(cmd), "%s -F %s %s 100 >/dev/null 2>&1",
498 MKE2FS, feature, imgname);
499 /* run_command() displays the output of mke2fs when it fails for
500 * some feature, so use system() directly */
508 * append_unique: append @key or @key=@val pair to @buf only if @key does not
510 * @buf: buffer to hold @key or @key=@val
511 * @prefix: prefix string before @key
513 * @val: value string if it's a @key=@val pair
515 static void append_unique(char *buf, char *prefix, char *key, char *val,
524 anchor = end = strstr(buf, key);
525 /* try to find exact match string in @buf */
526 while (end && *end != '\0' && *end != ',' && *end != ' ' && *end != '=')
529 if (anchor == NULL || strlen(key) != len ||
530 strncmp(anchor, key, len) != 0) {
532 strscat(buf, prefix, maxbuflen);
534 strscat(buf, key, maxbuflen);
536 strscat(buf, "=\"", maxbuflen);
537 strscat(buf, val, maxbuflen);
538 strscat(buf, "\"", maxbuflen);
543 static int enable_default_ext4_features(struct mkfs_opts *mop, char *anchor,
544 size_t maxbuflen, bool user_spec)
546 unsigned long long blocks = mop->mo_device_kb / mop->mo_blocksize_kb;
547 bool enable_64bit = false;
549 /* Enable large block addresses if the LUN is over 2^32 blocks. */
550 if (blocks > 0xffffffffULL && is_e2fsprogs_feature_supp("-O 64bit"))
553 if (IS_OST(&mop->mo_ldd)) {
554 append_unique(anchor, user_spec ? "," : " -O ",
555 "extents", NULL, maxbuflen);
556 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
557 } else if (IS_MDT(&mop->mo_ldd)) {
558 append_unique(anchor, user_spec ? "," : " -O ",
559 "dirdata", NULL, maxbuflen);
560 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
562 append_unique(anchor, ",", "extents", NULL, maxbuflen);
564 append_unique(anchor, ",", "^extents", NULL, maxbuflen);
566 append_unique(anchor, user_spec ? "," : " -O ",
567 "uninit_bg", NULL, maxbuflen);
570 /* Multiple mount protection enabled only if failover node specified */
571 if (mop->mo_flags & MO_FAILOVER) {
572 if (is_e2fsprogs_feature_supp("-O mmp"))
573 append_unique(anchor, ",", "mmp", NULL, maxbuflen);
575 disp_old_e2fsprogs_msg("mmp", 1);
578 /* Allow more than 65000 subdirectories */
579 if (is_e2fsprogs_feature_supp("-O dir_nlink"))
580 append_unique(anchor, ",", "dir_nlink", NULL, maxbuflen);
582 /* The following options are only valid for ext4-based ldiskfs.
583 * If --backfstype=ext3 is specified, do not enable them. */
584 if (mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3)
587 /* Enable quota by default */
588 if (is_e2fsprogs_feature_supp("-O quota")) {
589 append_unique(anchor, ",", "quota", NULL, maxbuflen);
592 fprintf(stderr, "\"-O quota\" must be supported by "
593 "e2fsprogs, please upgrade your e2fsprogs.\n");
597 /* Allow files larger than 2TB. Also needs LU-16, but not harmful. */
598 if (is_e2fsprogs_feature_supp("-O huge_file"))
599 append_unique(anchor, ",", "huge_file", NULL, maxbuflen);
602 append_unique(anchor, ",", "64bit", NULL, maxbuflen);
604 if (blocks >= 0x1000000000 && is_e2fsprogs_feature_supp("-O meta_bg"))
605 append_unique(anchor, ",", "meta_bg", NULL, maxbuflen);
607 if (enable_64bit || strstr(mop->mo_mkfsopts, "meta_bg"))
608 append_unique(anchor, ",", "^resize_inode", NULL, maxbuflen);
610 /* Allow xattrs larger than one block, stored in a separate inode */
611 if (IS_MDT(&mop->mo_ldd) && is_e2fsprogs_feature_supp("-O ea_inode"))
612 append_unique(anchor, ",", "ea_inode", NULL, maxbuflen);
614 /* Allow more than 10M directory entries */
615 if (IS_MDT(&mop->mo_ldd) && is_e2fsprogs_feature_supp("-O large_dir"))
616 append_unique(anchor, ",", "large_dir", NULL, maxbuflen);
618 /* Cluster inode/block bitmaps and inode table for more efficient IO.
619 * Align the flex groups on a 1MB boundary for better performance. */
620 /* This -O feature needs to go last, since it adds the "-G" option. */
621 if (is_e2fsprogs_feature_supp("-O flex_bg")) {
624 append_unique(anchor, ",", "flex_bg", NULL, maxbuflen);
626 if (IS_OST(&mop->mo_ldd) &&
627 strstr(mop->mo_mkfsopts, "-G") == NULL) {
628 snprintf(tmp_buf, sizeof(tmp_buf), " -G %u",
629 1024 / mop->mo_blocksize_kb);
630 strscat(anchor, tmp_buf, maxbuflen);
633 /* Don't add any more "-O" options here, see last comment above */
638 * moveopts_to_end: find the option string, move remaining strings to
639 * where option string starts, and append the option
641 * @start: where the option string starts before the move
642 * RETURN: where the option string starts after the move
644 static char *moveopts_to_end(char *start)
650 /* skip whitespace before options */
655 /* find end of option characters */
656 while (*end != ' ' && *end != '\0')
660 if (len >= sizeof(save))
661 len = sizeof(save) - 1;
664 strncpy(save, start, len);
667 /* move remaining options up front */
669 memmove(start, end, strlen(end));
670 *(start + strlen(end)) = '\0';
672 /* append the specified options */
673 if (*(start + strlen(start) - 1) != ' ')
675 idx = start + strlen(start);
681 /* Build fs according to type */
682 int ldiskfs_make_lustre(struct mkfs_opts *mop)
684 char mkfs_cmd[PATH_MAX];
688 int ret = 0, ext_opts = 0;
689 bool enable_64bit = false;
693 mop->mo_blocksize_kb = 4;
695 start = strstr(mop->mo_mkfsopts, "-b");
700 blocksize = strtol(start + 2, &end, 0);
701 if (end && (*end == 'k' || *end == 'K'))
703 /* EXT4_MIN_BLOCK_SIZE || EXT4_MAX_BLOCK_SIZE */
704 if (blocksize < 1024 || blocksize > 65536) {
706 "%s: blocksize %lu not in 1024-65536 bytes, normally 4096 bytes\n",
707 progname, blocksize);
711 if ((blocksize & (blocksize - 1)) != 0) {
713 "%s: blocksize %lu not a power-of-two value\n",
714 progname, blocksize);
717 mop->mo_blocksize_kb = blocksize >> 10;
720 if (!(mop->mo_flags & MO_IS_LOOP)) {
721 __u64 device_kb = get_device_size(mop->mo_device);
726 /* Compare to real size */
727 if (mop->mo_device_kb == 0 || device_kb < mop->mo_device_kb)
728 mop->mo_device_kb = device_kb;
731 if (mop->mo_device_kb != 0) {
734 if (mop->mo_device_kb < 32384) {
735 fprintf(stderr, "%s: size of filesystem must be larger "
736 "than 32MB, but is set to %lldKB\n",
737 progname, (long long)mop->mo_device_kb);
740 block_count = mop->mo_device_kb / mop->mo_blocksize_kb;
741 if (block_count > 0xffffffffULL) {
742 /* If the LUN size is just over 2^32 blocks, limit the
743 * filesystem size to 2^32-1 blocks to avoid problems
744 * with ldiskfs/mkfs not handling this well. b=22906
746 if (block_count < 0x100002000ULL)
748 0xffffffffULL * mop->mo_blocksize_kb;
754 if ((mop->mo_ldd.ldd_mount_type != LDD_MT_EXT3) &&
755 (mop->mo_ldd.ldd_mount_type != LDD_MT_LDISKFS) &&
756 (mop->mo_ldd.ldd_mount_type != LDD_MT_LDISKFS2)) {
757 fprintf(stderr, "%s: unsupported fs type: %d (%s)\n",
758 progname, mop->mo_ldd.ldd_mount_type,
759 MT_STR(&mop->mo_ldd));
764 /* Journal size in MB */
765 if (strstr(mop->mo_mkfsopts, "-J") == NULL &&
766 mop->mo_device_kb > 1024 * 1024) {
767 /* Choose our own default journal size */
768 long journal_mb = 0, max_mb;
770 /* cap journal size at 4GB for MDT, leave at 1GB for OSTs */
771 if (IS_MDT(&mop->mo_ldd))
773 else if (IS_OST(&mop->mo_ldd))
775 else /* Use mke2fs default size for MGS */
778 /* Use at most 4% of device for journal */
779 journal_mb = mop->mo_device_kb * 4 / (1024 * 100);
780 if (journal_mb > max_mb)
784 snprintf(buf, sizeof(buf), " -J size=%ld", journal_mb);
785 strscat(mop->mo_mkfsopts, buf,
786 sizeof(mop->mo_mkfsopts));
791 * The inode size is constituted by following elements
792 * (assuming all files are in composite layout and has
795 * ldiskfs inode size: 160
796 * MDT extended attributes size, including:
797 * ext4_xattr_header: 32
798 * LOV EA size: 32(lov_comp_md_v1) +
799 * 3 * 40(lov_comp_md_entry_v1) +
800 * 3 * 32(lov_mds_md) +
801 * stripes * 24(lov_ost_data) +
802 * 16(xattr_entry) + 4("lov")
803 * LMA EA size: 24(lustre_mdt_attrs) +
804 * 16(xattr_entry) + 4("lma")
805 * SOM EA size: 24(lustre_som_attrs) +
806 * 16(xattr_entry) + 4("som")
807 * link EA size: 24(link_ea_header) + 18(link_ea_entry) +
808 * 16(filename) + 16(xattr_entry) + 4("link")
809 * and some margin for 4-byte alignment, ACLs and other EAs.
811 * If we say the average filename length is about 32 bytes,
812 * the calculation looks like:
813 * 160 + 32 + (32+3*(40+32)+24*stripes+20) + (24+20) + (24+20) +
814 * (24+20) + (~42+16+20) + other <= 512*2^m, {m=0,1,2,3}
816 if (strstr(mop->mo_mkfsopts, "-I") == NULL) {
817 if (IS_MDT(&mop->mo_ldd)) {
818 if (mop->mo_stripe_count > 59)
819 inode_size = 512; /* bz 7241 */
820 /* see also "-i" below for EA blocks */
821 else if (mop->mo_stripe_count > 16)
825 } else if (IS_OST(&mop->mo_ldd)) {
826 /* We store MDS FID and necessary composite
827 * layout information in the OST object EA:
828 * ldiskfs inode size: 160
829 * OST extended attributes size, including:
830 * ext4_xattr_header: 32
831 * LMA EA size: 24(lustre_mdt_attrs) +
832 * 16(xattr_entry) + 4("lma")
833 * FID EA size: 52(filter_fid) +
834 * 16(xattr_entry) + 4("fid")
835 * 160 + 32 + (24+20) + (52+20) = 308
840 if (inode_size > 0) {
841 snprintf(buf, sizeof(buf), " -I %ld", inode_size);
842 strscat(mop->mo_mkfsopts, buf,
843 sizeof(mop->mo_mkfsopts));
847 /* Bytes_per_inode: disk size / num inodes */
848 if (strstr(mop->mo_mkfsopts, "-i") == NULL &&
849 strstr(mop->mo_mkfsopts, "-N") == NULL) {
850 long bytes_per_inode = 0;
852 /* Allocate more inodes on MDT devices. There is
853 * no data stored on the MDT, and very little extra
854 * metadata beyond the inode. It could go down as
855 * low as 1024 bytes, but this is conservative.
856 * Account for external EA blocks for wide striping.
858 if (IS_MDT(&mop->mo_ldd)) {
859 bytes_per_inode = inode_size + 1536;
861 if (mop->mo_stripe_count > 59) {
862 int extra = mop->mo_stripe_count * 24;
864 extra = ((extra - 1) | 4095) + 1;
865 bytes_per_inode += extra;
869 /* Allocate fewer inodes on large OST devices. Most
870 * filesystems can be much more aggressive than even
871 * this, but it is impossible to know in advance.
873 if (IS_OST(&mop->mo_ldd)) {
874 /* OST > 16TB assume average file size 1MB */
875 if (mop->mo_device_kb > (16ULL << 30))
876 bytes_per_inode = 1024 * 1024;
877 /* OST > 4TB assume average file size 512kB */
878 else if (mop->mo_device_kb > (4ULL << 30))
879 bytes_per_inode = 512 * 1024;
880 /* OST > 1TB assume average file size 256kB */
881 else if (mop->mo_device_kb > (1ULL << 30))
882 bytes_per_inode = 256 * 1024;
883 /* OST > 10GB assume average file size 64kB,
884 * plus a bit so that inodes will fit into a
885 * 256x flex_bg without overflowing.
887 else if (mop->mo_device_kb > (10ULL << 20))
888 bytes_per_inode = 69905;
891 if (bytes_per_inode > 0) {
892 snprintf(buf, sizeof(buf), " -i %ld", bytes_per_inode);
893 strscat(mop->mo_mkfsopts, buf,
894 sizeof(mop->mo_mkfsopts));
895 mop->mo_inode_size = bytes_per_inode;
900 strscat(mop->mo_mkfsopts, " -q", sizeof(mop->mo_mkfsopts));
902 /* start handle -O mkfs options */
903 start = strstr(mop->mo_mkfsopts, "-O");
905 if (strstr(start + 2, "-O") != NULL) {
907 "%s: don't specify multiple -O options\n",
911 start = moveopts_to_end(start);
912 maxbuflen = sizeof(mop->mo_mkfsopts) -
913 (start - mop->mo_mkfsopts) - strlen(start);
914 ret = enable_default_ext4_features(mop, start, maxbuflen, 1);
916 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts);
917 maxbuflen = sizeof(mop->mo_mkfsopts) - strlen(mop->mo_mkfsopts);
918 ret = enable_default_ext4_features(mop, start, maxbuflen, 0);
922 /* end handle -O mkfs options */
924 /* start handle -E mkfs options */
925 start = strstr(mop->mo_mkfsopts, "-E");
927 if (strstr(start + 2, "-E") != NULL) {
929 "%s: don't specify multiple -E options\n",
933 start = moveopts_to_end(start);
934 maxbuflen = sizeof(mop->mo_mkfsopts) -
935 (start - mop->mo_mkfsopts) - strlen(start);
938 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts);
939 maxbuflen = sizeof(mop->mo_mkfsopts) - strlen(mop->mo_mkfsopts);
942 /* In order to align the filesystem metadata on 1MB boundaries,
943 * give a resize value that will reserve a power-of-two group
944 * descriptor blocks, but leave one block for the superblock.
945 * Only useful for filesystems with < 2^32 blocks due to resize
948 if (!enable_64bit && strstr(mop->mo_mkfsopts, "meta_bg") == NULL &&
949 IS_OST(&mop->mo_ldd) && mop->mo_device_kb > 100 * 1024) {
950 unsigned int group_blocks = mop->mo_blocksize_kb * 8192;
951 unsigned int desc_per_block = mop->mo_blocksize_kb * 1024 / 32;
952 unsigned int resize_blks;
954 resize_blks = (1ULL<<32) - desc_per_block*group_blocks;
955 snprintf(buf, sizeof(buf), "%u", resize_blks);
956 append_unique(start, ext_opts ? "," : " -E ",
957 "resize", buf, maxbuflen);
961 /* Avoid zeroing out the full journal - speeds up mkfs */
962 if (is_e2fsprogs_feature_supp("-E lazy_journal_init"))
963 append_unique(start, ext_opts ? "," : " -E ",
964 "lazy_journal_init", NULL, maxbuflen);
965 /* end handle -E mkfs options */
967 /* Allow reformat of full devices (as opposed to partitions).
968 * We already checked for mounted dev.
970 strscat(mop->mo_mkfsopts, " -F", sizeof(mop->mo_mkfsopts));
972 snprintf(mkfs_cmd, sizeof(mkfs_cmd), "%s -j -b %d -L %s ", MKE2FS,
973 mop->mo_blocksize_kb * 1024, mop->mo_ldd.ldd_svname);
975 /* For loop device format the dev, not the filename */
976 dev = mop->mo_device;
977 if (mop->mo_flags & MO_IS_LOOP)
978 dev = mop->mo_loopdev;
980 vprint("formatting backing filesystem %s on %s\n",
981 MT_STR(&mop->mo_ldd), dev);
982 vprint("\ttarget name %s\n", mop->mo_ldd.ldd_svname);
983 vprint("\tkilobytes %llu\n", mop->mo_device_kb);
984 vprint("\toptions %s\n", mop->mo_mkfsopts);
986 /* mkfs_cmd's trailing space is important! */
987 strscat(mkfs_cmd, mop->mo_mkfsopts, sizeof(mkfs_cmd));
988 strscat(mkfs_cmd, " ", sizeof(mkfs_cmd));
989 strscat(mkfs_cmd, dev, sizeof(mkfs_cmd));
990 if (mop->mo_device_kb != 0) {
991 snprintf(buf, sizeof(buf), " %lluk",
992 (unsigned long long)mop->mo_device_kb);
993 strscat(mkfs_cmd, buf, sizeof(mkfs_cmd));
996 vprint("mkfs_cmd = %s\n", mkfs_cmd);
997 ret = run_command(mkfs_cmd, sizeof(mkfs_cmd));
1000 fprintf(stderr, "Unable to build fs %s (%d)\n", dev, ret);
1005 int ldiskfs_prepare_lustre(struct mkfs_opts *mop,
1006 char *wanted_mountopts, size_t len)
1008 struct lustre_disk_data *ldd = &mop->mo_ldd;
1011 /* Set MO_IS_LOOP to indicate a loopback device is needed */
1012 ret = is_block(mop->mo_device);
1015 } else if (ret == 0) {
1016 mop->mo_flags |= MO_IS_LOOP;
1019 if (IS_MDT(ldd) || IS_MGS(ldd))
1020 strscat(wanted_mountopts, ",user_xattr", len);
1025 int ldiskfs_fix_mountopts(struct mkfs_opts *mop, char *mountopts, size_t len)
1027 if (strstr(mountopts, "errors=") == NULL)
1028 strscat(mountopts, ",errors=remount-ro", len);
1033 static int read_file(const char *path, char *buf, int size)
1037 fd = fopen(path, "r");
1041 if (fgets(buf, size, fd) == NULL) {
1042 fprintf(stderr, "reading from %s: %s", path, strerror(errno));
1048 /* strip trailing newline */
1050 if (buf[size - 1] == '\n')
1051 buf[size - 1] = '\0';
1056 static int write_file(const char *path, const char *buf)
1060 fd = open(path, O_WRONLY);
1064 rc = write(fd, buf, strlen(buf));
1067 return rc < 0 ? errno : 0;
1070 static int tune_md_stripe_cache_size(const char *sys_path,
1071 struct mount_opts *mop)
1073 char path[PATH_MAX];
1074 unsigned long old_stripe_cache_size;
1075 unsigned long new_stripe_cache_size;
1076 char buf[3 * sizeof(old_stripe_cache_size) + 2];
1079 if (mop->mo_md_stripe_cache_size <= 0)
1082 new_stripe_cache_size = mop->mo_md_stripe_cache_size;
1084 snprintf(path, sizeof(path), "%s/%s", sys_path, STRIPE_CACHE_SIZE);
1085 rc = read_file(path, buf, sizeof(buf));
1088 fprintf(stderr, "warning: cannot read '%s': %s\n",
1089 path, strerror(errno));
1093 old_stripe_cache_size = strtoul(buf, NULL, 0);
1094 if (old_stripe_cache_size == 0 || old_stripe_cache_size == ULONG_MAX)
1097 if (new_stripe_cache_size <= old_stripe_cache_size)
1100 snprintf(buf, sizeof(buf), "%lu", new_stripe_cache_size);
1101 rc = write_file(path, buf);
1104 fprintf(stderr, "warning: cannot write '%s': %s\n",
1105 path, strerror(errno));
1112 static int tune_max_sectors_kb(const char *sys_path, struct mount_opts *mop)
1114 char path[PATH_MAX];
1115 unsigned long max_hw_sectors_kb;
1116 unsigned long old_max_sectors_kb;
1117 unsigned long new_max_sectors_kb;
1118 char buf[3 * sizeof(old_max_sectors_kb) + 2];
1121 if (mop->mo_max_sectors_kb >= 0) {
1122 new_max_sectors_kb = mop->mo_max_sectors_kb;
1123 goto have_new_max_sectors_kb;
1126 snprintf(path, sizeof(path), "%s/%s", sys_path, MAX_HW_SECTORS_KB_PATH);
1127 rc = read_file(path, buf, sizeof(buf));
1129 /* No MAX_HW_SECTORS_KB_PATH isn't necessary an
1130 * error for some devices. */
1134 max_hw_sectors_kb = strtoul(buf, NULL, 0);
1135 if (max_hw_sectors_kb == 0 || max_hw_sectors_kb == ULLONG_MAX) {
1136 /* No digits at all or something weird. */
1140 new_max_sectors_kb = max_hw_sectors_kb;
1142 /* Don't increase IO request size limit past 16MB. It is
1143 * about PTLRPC_MAX_BRW_SIZE, but that isn't in a public
1144 * header. Note that even though the block layer allows
1145 * larger values, setting max_sectors_kb = 32768 causes
1146 * crashes (LU-6974). */
1147 if (new_max_sectors_kb > 16 * 1024)
1148 new_max_sectors_kb = 16 * 1024;
1150 have_new_max_sectors_kb:
1151 snprintf(path, sizeof(path), "%s/%s", sys_path, MAX_SECTORS_KB_PATH);
1152 rc = read_file(path, buf, sizeof(buf));
1154 /* No MAX_SECTORS_KB_PATH isn't necessary an error for
1159 old_max_sectors_kb = strtoul(buf, NULL, 0);
1160 if (old_max_sectors_kb == 0 || old_max_sectors_kb == ULLONG_MAX) {
1161 /* No digits at all or something weird. */
1165 if (new_max_sectors_kb <= old_max_sectors_kb)
1168 snprintf(buf, sizeof(buf), "%lu", new_max_sectors_kb);
1169 rc = write_file(path, buf);
1172 fprintf(stderr, "warning: cannot write '%s': %s\n",
1173 path, strerror(errno));
1177 fprintf(stderr, "%s: increased '%s' from %lu to %lu\n",
1178 progname, path, old_max_sectors_kb, new_max_sectors_kb);
1183 static int tune_block_dev_scheduler(const char *sys_path, const char *new_sched)
1185 char path[PATH_MAX];
1191 /* Before setting the scheduler, we need to check to see if
1192 * it's already set to "noop". If it is then we don't want to
1193 * override that setting. If it's set to anything other than
1194 * "noop" then set the scheduler to what has been passed
1197 snprintf(path, sizeof(path), "%s/%s", sys_path, SCHEDULER_PATH);
1198 rc = read_file(path, buf, sizeof(buf));
1201 fprintf(stderr, "%s: cannot read '%s': %s\n",
1202 progname, path, strerror(errno));
1207 /* The expected format of buf: noop anticipatory deadline [cfq] */
1208 s = strchr(buf, '[');
1209 e = strchr(buf, ']');
1211 /* If the format is not what we expect then be safe and error out. */
1212 if (s == NULL || e == NULL || !(s < e)) {
1215 "%s: cannot parse scheduler options for '%s'\n",
1224 if (strcmp(old_sched, "noop") == 0 ||
1225 strcmp(old_sched, "deadline") == 0 ||
1226 strcmp(old_sched, "mq-deadline") == 0 ||
1227 strstr(old_sched, new_sched) == 0)
1230 rc = write_file(path, new_sched);
1234 "%s: cannot set scheduler on '%s': %s\n",
1235 progname, path, strerror(errno));
1239 fprintf(stderr, "%s: changed scheduler of '%s' from %s to %s\n",
1240 progname, path, old_sched, new_sched);
1245 static int tune_block_dev(const char *src, struct mount_opts *mop);
1247 static int tune_block_dev_slaves(const char *sys_path, struct mount_opts *mop)
1249 char slaves_path[PATH_MAX];
1254 snprintf(slaves_path, sizeof(slaves_path), "%s/slaves", sys_path);
1255 slaves_dir = opendir(slaves_path);
1256 if (slaves_dir == NULL) {
1257 if (errno == ENOENT)
1263 while ((d = readdir(slaves_dir)) != NULL) {
1264 char path[PATH_MAX * 2];
1267 if (d->d_type != DT_LNK)
1270 snprintf(path, sizeof(path), "/dev/%s", d->d_name);
1271 rc2 = tune_block_dev(path, mop);
1276 closedir(slaves_dir);
1281 /* This is to tune the kernel for good SCSI performance.
1282 * For that we set the value of /sys/block/{dev}/queue/max_sectors_kb
1283 * to the value of /sys/block/{dev}/queue/max_hw_sectors_kb */
1284 static int tune_block_dev(const char *src, struct mount_opts *mop)
1287 char sys_path[PATH_MAX];
1288 char partition_path[PATH_MAX + sizeof("partition")];
1289 char *real_sys_path = NULL;
1293 * Don't apply block device tuning for MDT or MGT devices,
1294 * since we don't need huge IO sizes to get good performance
1296 if (!IS_OST(&mop->mo_ldd))
1302 rc = stat(src, &st);
1305 fprintf(stderr, "warning: cannot stat '%s': %s\n",
1306 src, strerror(errno));
1310 if (!S_ISBLK(st.st_mode))
1313 if (major(st.st_rdev) == LOOP_MAJOR)
1316 snprintf(sys_path, sizeof(sys_path), "/sys/dev/block/%u:%u",
1317 major(st.st_rdev), minor(st.st_rdev));
1319 snprintf(partition_path, sizeof(partition_path), "%s/partition",
1322 rc = access(partition_path, F_OK);
1324 if (errno == ENOENT)
1325 goto have_whole_dev;
1328 fprintf(stderr, "warning: cannot access '%s': %s\n",
1329 partition_path, strerror(errno));
1334 snprintf(sys_path, sizeof(sys_path), "/sys/dev/block/%u:%u/..",
1335 major(st.st_rdev), minor(st.st_rdev));
1338 /* Since we recurse on slave devices we resolve the sys_path to
1339 * avoid path buffer overflows. */
1340 real_sys_path = realpath(sys_path, NULL);
1341 if (real_sys_path == NULL) {
1344 "warning: cannot resolve '%s': %s\n",
1345 sys_path, strerror(errno));
1350 if (major(st.st_rdev) == MD_MAJOR) {
1351 rc = tune_md_stripe_cache_size(real_sys_path, mop);
1353 /* Ignore errors from tune_max_sectors_kb() and
1354 * tune_scheduler(). The worst that will happen is a block
1355 * device with an "incorrect" scheduler. */
1356 tune_max_sectors_kb(real_sys_path, mop);
1357 tune_block_dev_scheduler(real_sys_path, DEFAULT_SCHEDULER);
1359 /* If device is multipath device then tune its slave
1361 rc = tune_block_dev_slaves(real_sys_path, mop);
1365 free(real_sys_path);
1370 int ldiskfs_tune_lustre(char *dev, struct mount_opts *mop)
1372 return tune_block_dev(dev, mop);
1375 int ldiskfs_label_lustre(struct mount_opts *mop)
1377 char label_cmd[PATH_MAX];
1380 snprintf(label_cmd, sizeof(label_cmd),
1381 TUNE2FS" -f -L '%s' '%s' >/dev/null 2>&1",
1382 mop->mo_ldd.ldd_svname, mop->mo_source);
1383 rc = run_command(label_cmd, sizeof(label_cmd));
1388 int ldiskfs_rename_fsname(struct mkfs_opts *mop, const char *oldname)
1390 struct mount_opts opts;
1391 struct lustre_disk_data *ldd = &mop->mo_ldd;
1392 char mntpt[] = "/tmp/mntXXXXXX";
1396 /* Change the filesystem label. */
1398 opts.mo_source = mop->mo_device;
1399 ret = ldiskfs_label_lustre(&opts);
1403 fprintf(stderr, "Can't change filesystem label: %s\n",
1408 /* Mount this device temporarily in order to write these files */
1409 if (mkdtemp(mntpt) == NULL) {
1414 fprintf(stderr, "Can't create temp mount point %s: %s\n",
1415 mntpt, strerror(ret));
1421 * Append file context to mount options if SE Linux is enabled
1423 if (is_selinux_enabled() > 0)
1424 append_context_for_mount(mntpt, mop);
1427 if (mop->mo_flags & MO_IS_LOOP)
1428 dev = mop->mo_loopdev;
1430 dev = mop->mo_device;
1431 ret = mount(dev, mntpt, MT_STR(ldd), 0, ldd->ldd_mount_opts);
1435 fprintf(stderr, "Unable to mount %s: %s\n",
1436 dev, strerror(ret));
1438 fprintf(stderr, "Is the %s module available?\n",
1443 ret = lustre_rename_fsname(mop, mntpt, oldname);
1451 /* Enable quota accounting */
1452 int ldiskfs_enable_quota(struct mkfs_opts *mop)
1456 int cmdsz = sizeof(cmd), ret;
1458 if (!is_e2fsprogs_feature_supp("-O quota")) {
1459 fprintf(stderr, "%s: \"-O quota\" is is not supported by "
1460 "current e2fsprogs\n", progname);
1464 dev = mop->mo_device;
1465 if (mop->mo_flags & MO_IS_LOOP)
1466 dev = mop->mo_loopdev;
1468 /* Quota feature is already enabled? */
1469 if (is_feature_enabled("quota", dev)) {
1470 vprint("Quota feature is already enabled.\n");
1474 /* Turn on quota feature by "tune2fs -O quota" */
1475 snprintf(cmd, cmdsz, "%s -O quota %s", TUNE2FS, dev);
1476 ret = run_command(cmd, cmdsz);
1478 fprintf(stderr, "command:%s (%d)", cmd, ret);
1483 int ldiskfs_init(void)
1485 /* Required because full path to DEBUGFS is not specified */
1486 setenv("PATH", "/bin:/sbin:/usr/bin:/usr/sbin:/usr/local/bin", 0);
1491 void ldiskfs_fini(void)
1497 struct module_backfs_ops ldiskfs_ops = {
1498 .init = ldiskfs_init,
1499 .fini = ldiskfs_fini,
1500 .read_ldd = ldiskfs_read_ldd,
1501 .write_ldd = ldiskfs_write_ldd,
1502 .erase_ldd = ldiskfs_erase_ldd,
1503 .print_ldd_params = ldiskfs_print_ldd_params,
1504 .is_lustre = ldiskfs_is_lustre,
1505 .make_lustre = ldiskfs_make_lustre,
1506 .prepare_lustre = ldiskfs_prepare_lustre,
1507 .fix_mountopts = ldiskfs_fix_mountopts,
1508 .tune_lustre = ldiskfs_tune_lustre,
1509 .label_lustre = ldiskfs_label_lustre,
1510 .enable_quota = ldiskfs_enable_quota,
1511 .rename_fsname = ldiskfs_rename_fsname,
1513 #endif /* PLUGIN_DIR */