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>
77 #include "mount_utils.h"
79 #define MAX_HW_SECTORS_KB_PATH "queue/max_hw_sectors_kb"
80 #define MAX_SECTORS_KB_PATH "queue/max_sectors_kb"
81 #define SCHEDULER_PATH "queue/scheduler"
82 #define STRIPE_CACHE_SIZE "md/stripe_cache_size"
84 #define DEFAULT_SCHEDULER "deadline"
86 extern char *progname;
88 /* keep it less than LL_FID_NAMELEN */
89 #define DUMMY_FILE_NAME_LEN 25
90 #define EXT3_DIRENT_SIZE DUMMY_FILE_NAME_LEN
92 static void append_unique(char *buf, char *prefix, char *key, char *val,
94 static bool is_e2fsprogs_feature_supp(const char *feature);
95 static void disp_old_e2fsprogs_msg(const char *feature, int make_backfs);
97 /* Determine if a device is a block device (as opposed to a file) */
98 static int is_block(char *devname)
104 ret = cfs_abs_path(devname, &devpath);
106 fprintf(stderr, "%s: failed to resolve path '%s': %s\n",
107 progname, devname, strerror(-ret));
111 ret = access(devname, F_OK);
113 if (strncmp(devpath, "/dev/", 5) == 0) {
114 /* nobody sane wants to create a loopback file under
115 * /dev. Let's just report the device doesn't exist */
116 fprintf(stderr, "%s: %s apparently does not exist\n",
124 ret = stat(devpath, &st);
126 fprintf(stderr, "%s: cannot stat %s\n", progname, devpath);
129 ret = S_ISBLK(st.st_mode);
135 static int is_feature_enabled(const char *feature, const char *devpath)
139 char enabled_features[4096] = "";
142 snprintf(cmd, sizeof(cmd), "%s -c -R features %s 2>&1",
145 /* Using popen() instead of run_command() since debugfs does
146 * not return proper error code if command is not supported */
147 fp = popen(cmd, "r");
149 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
153 ret = fread(enabled_features, 1, sizeof(enabled_features) - 1, fp);
154 enabled_features[ret] = '\0';
157 if (strstr(enabled_features, feature))
162 /* Write the server config files */
163 int ldiskfs_write_ldd(struct mkfs_opts *mop)
165 char mntpt[] = "/tmp/mntXXXXXX";
172 /* Mount this device temporarily in order to write these files */
173 if (!mkdtemp(mntpt)) {
174 fprintf(stderr, "%s: Can't create temp mount point %s: %s\n",
175 progname, mntpt, strerror(errno));
179 dev = mop->mo_device;
180 if (mop->mo_flags & MO_IS_LOOP)
181 dev = mop->mo_loopdev;
183 /* Multiple mount protection enabled if failover node specified */
184 if (mop->mo_flags & MO_FAILOVER &&
185 !is_feature_enabled("mmp", dev)) {
186 if (is_e2fsprogs_feature_supp("-O mmp")) {
187 char *command = filepnm;
189 snprintf(command, sizeof(filepnm),
190 TUNE2FS" -O mmp '%s' >/dev/null 2>&1", dev);
191 ret = run_command(command, sizeof(filepnm));
194 "%s: Unable to set 'mmp' on %s: %d\n",
197 disp_old_e2fsprogs_msg("mmp", 1);
201 ret = mount(dev, mntpt, MT_STR(&mop->mo_ldd), 0,
202 (mop->mo_mountopts == NULL) ?
203 "errors=remount-ro" : mop->mo_mountopts);
205 fprintf(stderr, "%s: Unable to mount %s: %s\n",
206 progname, dev, strerror(errno));
208 if (errno == ENODEV) {
209 fprintf(stderr, "Is the %s module available?\n",
210 MT_STR(&mop->mo_ldd));
215 /* Set up initial directories */
216 sprintf(filepnm, "%s/%s", mntpt, MOUNT_CONFIGS_DIR);
217 ret = mkdir(filepnm, 0777);
218 if ((ret != 0) && (errno != EEXIST)) {
219 fprintf(stderr, "%s: Can't make configs dir %s (%s)\n",
220 progname, filepnm, strerror(errno));
222 } else if (errno == EEXIST) {
226 /* Save the persistent mount data into a file. Lustre must pre-read
227 this file to get the real mount options. */
228 vprint("Writing %s\n", MOUNT_DATA_FILE);
229 sprintf(filepnm, "%s/%s", mntpt, MOUNT_DATA_FILE);
230 filep = fopen(filepnm, "w");
232 fprintf(stderr, "%s: Unable to create %s file: %s\n",
233 progname, filepnm, strerror(errno));
236 num = fwrite(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
237 if (num < 1 && ferror(filep)) {
238 fprintf(stderr, "%s: Unable to write to file (%s): %s\n",
239 progname, filepnm, strerror(errno));
243 fsync(filep->_fileno);
253 static int readcmd(char *cmd, char *buf, int len)
258 fp = popen(cmd, "r");
262 red = fread(buf, 1, len, fp);
265 /* strip trailing newline */
266 if (buf[red - 1] == '\n')
269 return (red == 0) ? -ENOENT : 0;
272 int ldiskfs_read_ldd(char *dev, struct lustre_disk_data *mo_ldd)
274 char tmpdir[] = "/tmp/dirXXXXXX";
279 int cmdsz = sizeof(cmd);
281 /* Make a temporary directory to hold Lustre data files. */
282 if (!mkdtemp(tmpdir)) {
283 fprintf(stderr, "%s: Can't create temporary directory %s: %s\n",
284 progname, tmpdir, strerror(errno));
288 /* TODO: it's worth observing the get_mountdata() function that is
289 in mount_utils.c for getting the mountdata out of the
292 /* Construct debugfs command line. */
293 snprintf(cmd, cmdsz, "%s -c -R 'dump /%s %s/mountdata' '%s'",
294 DEBUGFS, MOUNT_DATA_FILE, tmpdir, dev);
296 ret = run_command(cmd, cmdsz);
298 verrprint("%s: Unable to dump %s dir (%d)\n",
299 progname, MOUNT_CONFIGS_DIR, ret);
301 sprintf(filepnm, "%s/mountdata", tmpdir);
302 filep = fopen(filepnm, "r");
305 vprint("Reading %s\n", MOUNT_DATA_FILE);
306 num_read = fread(mo_ldd, sizeof(*mo_ldd), 1, filep);
307 if (num_read < 1 && ferror(filep)) {
308 fprintf(stderr, "%s: Unable to read from file %s: %s\n",
309 progname, filepnm, strerror(errno));
314 snprintf(cmd, cmdsz, "rm -rf %s", tmpdir);
315 run_command(cmd, cmdsz);
317 verrprint("Failed to read old data (%d)\n", ret);
319 /* As long as we at least have the label, we're good to go */
320 snprintf(cmd, sizeof(cmd), E2LABEL" %s", dev);
321 ret = readcmd(cmd, mo_ldd->ldd_svname, sizeof(mo_ldd->ldd_svname) - 1);
326 int ldiskfs_erase_ldd(struct mkfs_opts *mop, char *param)
331 void ldiskfs_print_ldd_params(struct mkfs_opts *mop)
333 printf("Parameters:%s\n", mop->mo_ldd.ldd_params);
336 /* Display the need for the latest e2fsprogs to be installed. make_backfs
337 * indicates if the caller is make_lustre_backfs() or not. */
338 static void disp_old_e2fsprogs_msg(const char *feature, int make_backfs)
340 static int msg_displayed;
343 fprintf(stderr, "WARNING: %s does not support %s "
344 "feature.\n\n", E2FSPROGS, feature);
350 fprintf(stderr, "WARNING: The %s package currently installed on "
351 "your system does not support \"%s\" feature.\n",
353 #if !(HAVE_LDISKFSPROGS)
354 fprintf(stderr, "Please install the latest version of e2fsprogs from\n"
355 "https://downloads.whamcloud.com/public/e2fsprogs/latest/\n"
356 "to enable this feature.\n");
360 "Feature will not be enabled until %s is updated and '%s -O %s %%{device}' is run.\n\n",
361 E2FSPROGS, TUNE2FS, feature);
364 /* Check whether the file exists in the device */
365 static int file_in_dev(char *file_name, char *dev_name)
368 char debugfs_cmd[256];
369 unsigned int inode_num;
372 /* Construct debugfs command line. */
373 snprintf(debugfs_cmd, sizeof(debugfs_cmd),
374 "%s -c -R 'stat %s' '%s' 2>&1 | egrep '(Inode|unsupported)'",
375 DEBUGFS, file_name, dev_name);
377 fp = popen(debugfs_cmd, "r");
379 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
383 if (fscanf(fp, "Inode: %u", &inode_num) == 1) { /* exist */
387 i = fread(debugfs_cmd, 1, sizeof(debugfs_cmd) - 1, fp);
390 fprintf(stderr, "%s", debugfs_cmd);
391 if (strstr(debugfs_cmd, "unsupported feature")) {
392 disp_old_e2fsprogs_msg("an unknown", 0);
401 /* Check whether the device has already been used with lustre */
402 int ldiskfs_is_lustre(char *dev, unsigned *mount_type)
406 ret = file_in_dev(MOUNT_DATA_FILE, dev);
408 /* in the -1 case, 'extents' means IS a lustre target */
409 *mount_type = LDD_MT_LDISKFS;
413 ret = file_in_dev(LAST_RCVD, dev);
415 *mount_type = LDD_MT_LDISKFS;
422 /* Check if a certain feature is supported by e2fsprogs.
423 * Firstly we try to use "debugfs supported_features" command to check if
424 * the feature is supported. If this fails we try to set this feature with
425 * mke2fs to check for its support. */
426 static bool is_e2fsprogs_feature_supp(const char *feature)
428 static char supp_features[4096] = "";
431 char imgname[] = "/tmp/test-img-XXXXXX";
435 if (supp_features[0] == '\0') {
436 snprintf(cmd, sizeof(cmd), "%s -c -R supported_features 2>&1",
439 /* Using popen() instead of run_command() since debugfs does
440 * not return proper error code if command is not supported */
441 fp = popen(cmd, "r");
443 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
445 ret = fread(supp_features, 1,
446 sizeof(supp_features) - 1, fp);
447 supp_features[ret] = '\0';
452 if (strstr(supp_features,
453 strncmp(feature, "-O ", 3) ? feature : feature + 3))
456 if ((fd = mkstemp(imgname)) < 0)
461 snprintf(cmd, sizeof(cmd), "%s -F %s %s 100 >/dev/null 2>&1",
462 MKE2FS, feature, imgname);
463 /* run_command() displays the output of mke2fs when it fails for
464 * some feature, so use system() directly */
472 * append_unique: append @key or @key=@val pair to @buf only if @key does not
474 * @buf: buffer to hold @key or @key=@val
475 * @prefix: prefix string before @key
477 * @val: value string if it's a @key=@val pair
479 static void append_unique(char *buf, char *prefix, char *key, char *val,
488 anchor = end = strstr(buf, key);
489 /* try to find exact match string in @buf */
490 while (end && *end != '\0' && *end != ',' && *end != ' ' && *end != '=')
493 if (anchor == NULL || strlen(key) != len ||
494 strncmp(anchor, key, len) != 0) {
496 strscat(buf, prefix, maxbuflen);
498 strscat(buf, key, maxbuflen);
500 strscat(buf, "=\"", maxbuflen);
501 strscat(buf, val, maxbuflen);
502 strscat(buf, "\"", maxbuflen);
507 static int enable_default_ext4_features(struct mkfs_opts *mop, char *anchor,
508 size_t maxbuflen, bool user_spec)
510 unsigned long long blocks = mop->mo_device_kb / mop->mo_blocksize_kb;
511 bool enable_64bit = false;
513 /* Enable large block addresses if the LUN is over 2^32 blocks. */
514 if (blocks > 0xffffffffULL && is_e2fsprogs_feature_supp("-O 64bit"))
517 if (IS_OST(&mop->mo_ldd)) {
518 append_unique(anchor, user_spec ? "," : " -O ",
519 "extents", NULL, maxbuflen);
520 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
521 } else if (IS_MDT(&mop->mo_ldd)) {
522 append_unique(anchor, user_spec ? "," : " -O ",
523 "dirdata", NULL, maxbuflen);
524 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
526 append_unique(anchor, ",", "extents", NULL, maxbuflen);
528 append_unique(anchor, ",", "^extents", NULL, maxbuflen);
530 append_unique(anchor, user_spec ? "," : " -O ",
531 "uninit_bg", NULL, maxbuflen);
534 /* Multiple mount protection enabled only if failover node specified */
535 if (mop->mo_flags & MO_FAILOVER) {
536 if (is_e2fsprogs_feature_supp("-O mmp"))
537 append_unique(anchor, ",", "mmp", NULL, maxbuflen);
539 disp_old_e2fsprogs_msg("mmp", 1);
542 /* Allow more than 65000 subdirectories */
543 if (is_e2fsprogs_feature_supp("-O dir_nlink"))
544 append_unique(anchor, ",", "dir_nlink", NULL, maxbuflen);
546 /* The following options are only valid for ext4-based ldiskfs.
547 * If --backfstype=ext3 is specified, do not enable them. */
548 if (mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3)
551 /* Enable quota by default */
552 if (is_e2fsprogs_feature_supp("-O quota")) {
553 append_unique(anchor, ",", "quota", NULL, maxbuflen);
556 fprintf(stderr, "\"-O quota\" must be supported by "
557 "e2fsprogs, please upgrade your e2fsprogs.\n");
561 /* Allow files larger than 2TB. Also needs LU-16, but not harmful. */
562 if (is_e2fsprogs_feature_supp("-O huge_file"))
563 append_unique(anchor, ",", "huge_file", NULL, maxbuflen);
566 append_unique(anchor, ",", "64bit", NULL, maxbuflen);
568 if (blocks >= 0x1000000000 && is_e2fsprogs_feature_supp("-O meta_bg"))
569 append_unique(anchor, ",", "meta_bg", NULL, maxbuflen);
571 if (enable_64bit || strstr(mop->mo_mkfsopts, "meta_bg"))
572 append_unique(anchor, ",", "^resize_inode", NULL, maxbuflen);
574 /* Allow xattrs larger than one block, stored in a separate inode */
575 if (IS_MDT(&mop->mo_ldd) && is_e2fsprogs_feature_supp("-O ea_inode"))
576 append_unique(anchor, ",", "ea_inode", NULL, maxbuflen);
578 /* Allow more than 10M directory entries */
579 if (IS_MDT(&mop->mo_ldd) && is_e2fsprogs_feature_supp("-O large_dir"))
580 append_unique(anchor, ",", "large_dir", NULL, maxbuflen);
582 /* Cluster inode/block bitmaps and inode table for more efficient IO.
583 * Align the flex groups on a 1MB boundary for better performance. */
584 /* This -O feature needs to go last, since it adds the "-G" option. */
585 if (is_e2fsprogs_feature_supp("-O flex_bg")) {
588 append_unique(anchor, ",", "flex_bg", NULL, maxbuflen);
590 if (IS_OST(&mop->mo_ldd) &&
591 strstr(mop->mo_mkfsopts, "-G") == NULL) {
592 snprintf(tmp_buf, sizeof(tmp_buf), " -G %u",
593 1024 / mop->mo_blocksize_kb);
594 strscat(anchor, tmp_buf, maxbuflen);
597 /* Don't add any more "-O" options here, see last comment above */
602 * moveopts_to_end: find the option string, move remaining strings to
603 * where option string starts, and append the option
605 * @start: where the option string starts before the move
606 * RETURN: where the option string starts after the move
608 static char *moveopts_to_end(char *start)
614 /* skip whitespace before options */
619 /* find end of option characters */
620 while (*end != ' ' && *end != '\0')
624 if (len >= sizeof(save))
625 len = sizeof(save) - 1;
628 strncpy(save, start, len);
631 /* move remaining options up front */
633 memmove(start, end, strlen(end));
634 *(start + strlen(end)) = '\0';
636 /* append the specified options */
637 if (*(start + strlen(start) - 1) != ' ')
639 idx = start + strlen(start);
645 /* Build fs according to type */
646 int ldiskfs_make_lustre(struct mkfs_opts *mop)
648 char mkfs_cmd[PATH_MAX];
652 int ret = 0, ext_opts = 0;
653 bool enable_64bit = false;
657 mop->mo_blocksize_kb = 4;
659 start = strstr(mop->mo_mkfsopts, "-b");
664 blocksize = strtol(start + 2, &end, 0);
665 if (end && (*end == 'k' || *end == 'K'))
667 /* EXT4_MIN_BLOCK_SIZE || EXT4_MAX_BLOCK_SIZE */
668 if (blocksize < 1024 || blocksize > 65536) {
670 "%s: blocksize %lu not in 1024-65536 bytes, normally 4096 bytes\n",
671 progname, blocksize);
675 if ((blocksize & (blocksize - 1)) != 0) {
677 "%s: blocksize %lu not a power-of-two value\n",
678 progname, blocksize);
681 mop->mo_blocksize_kb = blocksize >> 10;
684 if (!(mop->mo_flags & MO_IS_LOOP)) {
685 __u64 device_kb = get_device_size(mop->mo_device);
690 /* Compare to real size */
691 if (mop->mo_device_kb == 0 || device_kb < mop->mo_device_kb)
692 mop->mo_device_kb = device_kb;
695 if (mop->mo_device_kb != 0) {
698 if (mop->mo_device_kb < 32384) {
699 fprintf(stderr, "%s: size of filesystem must be larger "
700 "than 32MB, but is set to %lldKB\n",
701 progname, (long long)mop->mo_device_kb);
704 block_count = mop->mo_device_kb / mop->mo_blocksize_kb;
705 if (block_count > 0xffffffffULL) {
706 /* If the LUN size is just over 2^32 blocks, limit the
707 * filesystem size to 2^32-1 blocks to avoid problems
708 * with ldiskfs/mkfs not handling this well. b=22906
710 if (block_count < 0x100002000ULL)
712 0xffffffffULL * mop->mo_blocksize_kb;
718 if ((mop->mo_ldd.ldd_mount_type != LDD_MT_EXT3) &&
719 (mop->mo_ldd.ldd_mount_type != LDD_MT_LDISKFS) &&
720 (mop->mo_ldd.ldd_mount_type != LDD_MT_LDISKFS2)) {
721 fprintf(stderr, "%s: unsupported fs type: %d (%s)\n",
722 progname, mop->mo_ldd.ldd_mount_type,
723 MT_STR(&mop->mo_ldd));
728 /* Journal size in MB */
729 if (strstr(mop->mo_mkfsopts, "-J") == NULL &&
730 mop->mo_device_kb > 1024 * 1024) {
731 /* Choose our own default journal size */
732 long journal_mb = 0, max_mb;
734 /* cap journal size at 4GB for MDT, leave at 1GB for OSTs */
735 if (IS_MDT(&mop->mo_ldd))
737 else if (IS_OST(&mop->mo_ldd))
739 else /* Use mke2fs default size for MGS */
742 /* Use at most 4% of device for journal */
743 journal_mb = mop->mo_device_kb * 4 / (1024 * 100);
744 if (journal_mb > max_mb)
748 snprintf(buf, sizeof(buf), " -J size=%ld", journal_mb);
749 strscat(mop->mo_mkfsopts, buf,
750 sizeof(mop->mo_mkfsopts));
755 * The inode size is constituted by following elements
756 * (assuming all files are in composite layout and has
759 * ldiskfs inode size: 160
760 * MDT extended attributes size, including:
761 * ext4_xattr_header: 32
762 * LOV EA size: 32(lov_comp_md_v1) +
763 * 3 * 40(lov_comp_md_entry_v1) +
764 * 3 * 32(lov_mds_md) +
765 * stripes * 24(lov_ost_data) +
766 * 16(xattr_entry) + 4("lov")
767 * LMA EA size: 24(lustre_mdt_attrs) +
768 * 16(xattr_entry) + 4("lma")
769 * SOM EA size: 24(lustre_som_attrs) +
770 * 16(xattr_entry) + 4("som")
771 * link EA size: 24(link_ea_header) + 18(link_ea_entry) +
772 * 16(filename) + 16(xattr_entry) + 4("link")
773 * and some margin for 4-byte alignment, ACLs and other EAs.
775 * If we say the average filename length is about 32 bytes,
776 * the calculation looks like:
777 * 160 + 32 + (32+3*(40+32)+24*stripes+20) + (24+20) + (24+20) +
778 * (24+20) + (~42+16+20) + other <= 512*2^m, {m=0,1,2,3}
780 if (strstr(mop->mo_mkfsopts, "-I") == NULL) {
781 if (IS_MDT(&mop->mo_ldd)) {
782 if (mop->mo_stripe_count > 59)
783 inode_size = 512; /* bz 7241 */
784 /* see also "-i" below for EA blocks */
785 else if (mop->mo_stripe_count > 16)
789 } else if (IS_OST(&mop->mo_ldd)) {
790 /* We store MDS FID and necessary composite
791 * layout information in the OST object EA:
792 * ldiskfs inode size: 160
793 * OST extended attributes size, including:
794 * ext4_xattr_header: 32
795 * LMA EA size: 24(lustre_mdt_attrs) +
796 * 16(xattr_entry) + 4("lma")
797 * FID EA size: 52(filter_fid) +
798 * 16(xattr_entry) + 4("fid")
799 * 160 + 32 + (24+20) + (52+20) = 308
804 if (inode_size > 0) {
805 snprintf(buf, sizeof(buf), " -I %ld", inode_size);
806 strscat(mop->mo_mkfsopts, buf,
807 sizeof(mop->mo_mkfsopts));
811 /* Bytes_per_inode: disk size / num inodes */
812 if (strstr(mop->mo_mkfsopts, "-i") == NULL &&
813 strstr(mop->mo_mkfsopts, "-N") == NULL) {
814 long bytes_per_inode = 0;
816 /* Allocate more inodes on MDT devices. There is
817 * no data stored on the MDT, and very little extra
818 * metadata beyond the inode. It could go down as
819 * low as 1024 bytes, but this is conservative.
820 * Account for external EA blocks for wide striping.
822 if (IS_MDT(&mop->mo_ldd)) {
823 bytes_per_inode = inode_size + 1536;
825 if (mop->mo_stripe_count > 59) {
826 int extra = mop->mo_stripe_count * 24;
828 extra = ((extra - 1) | 4095) + 1;
829 bytes_per_inode += extra;
833 /* Allocate fewer inodes on large OST devices. Most
834 * filesystems can be much more aggressive than even
835 * this, but it is impossible to know in advance.
837 if (IS_OST(&mop->mo_ldd)) {
838 /* OST > 16TB assume average file size 1MB */
839 if (mop->mo_device_kb > (16ULL << 30))
840 bytes_per_inode = 1024 * 1024;
841 /* OST > 4TB assume average file size 512kB */
842 else if (mop->mo_device_kb > (4ULL << 30))
843 bytes_per_inode = 512 * 1024;
844 /* OST > 1TB assume average file size 256kB */
845 else if (mop->mo_device_kb > (1ULL << 30))
846 bytes_per_inode = 256 * 1024;
847 /* OST > 10GB assume average file size 64kB,
848 * plus a bit so that inodes will fit into a
849 * 256x flex_bg without overflowing.
851 else if (mop->mo_device_kb > (10ULL << 20))
852 bytes_per_inode = 69905;
855 if (bytes_per_inode > 0) {
856 snprintf(buf, sizeof(buf), " -i %ld", bytes_per_inode);
857 strscat(mop->mo_mkfsopts, buf,
858 sizeof(mop->mo_mkfsopts));
859 mop->mo_inode_size = bytes_per_inode;
864 strscat(mop->mo_mkfsopts, " -q", sizeof(mop->mo_mkfsopts));
866 /* start handle -O mkfs options */
867 start = strstr(mop->mo_mkfsopts, "-O");
869 if (strstr(start + 2, "-O") != NULL) {
871 "%s: don't specify multiple -O options\n",
875 start = moveopts_to_end(start);
876 maxbuflen = sizeof(mop->mo_mkfsopts) -
877 (start - mop->mo_mkfsopts) - strlen(start);
878 ret = enable_default_ext4_features(mop, start, maxbuflen, 1);
880 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts);
881 maxbuflen = sizeof(mop->mo_mkfsopts) - strlen(mop->mo_mkfsopts);
882 ret = enable_default_ext4_features(mop, start, maxbuflen, 0);
886 /* end handle -O mkfs options */
888 /* start handle -E mkfs options */
889 start = strstr(mop->mo_mkfsopts, "-E");
891 if (strstr(start + 2, "-E") != NULL) {
893 "%s: don't specify multiple -E options\n",
897 start = moveopts_to_end(start);
898 maxbuflen = sizeof(mop->mo_mkfsopts) -
899 (start - mop->mo_mkfsopts) - strlen(start);
902 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts);
903 maxbuflen = sizeof(mop->mo_mkfsopts) - strlen(mop->mo_mkfsopts);
906 /* In order to align the filesystem metadata on 1MB boundaries,
907 * give a resize value that will reserve a power-of-two group
908 * descriptor blocks, but leave one block for the superblock.
909 * Only useful for filesystems with < 2^32 blocks due to resize
912 if (!enable_64bit && strstr(mop->mo_mkfsopts, "meta_bg") == NULL &&
913 IS_OST(&mop->mo_ldd) && mop->mo_device_kb > 100 * 1024) {
914 unsigned int group_blocks = mop->mo_blocksize_kb * 8192;
915 unsigned int desc_per_block = mop->mo_blocksize_kb * 1024 / 32;
916 unsigned int resize_blks;
918 resize_blks = (1ULL<<32) - desc_per_block*group_blocks;
919 snprintf(buf, sizeof(buf), "%u", resize_blks);
920 append_unique(start, ext_opts ? "," : " -E ",
921 "resize", buf, maxbuflen);
925 /* Avoid zeroing out the full journal - speeds up mkfs */
926 if (is_e2fsprogs_feature_supp("-E lazy_journal_init"))
927 append_unique(start, ext_opts ? "," : " -E ",
928 "lazy_journal_init", NULL, maxbuflen);
929 /* end handle -E mkfs options */
931 /* Allow reformat of full devices (as opposed to partitions).
932 * We already checked for mounted dev.
934 strscat(mop->mo_mkfsopts, " -F", sizeof(mop->mo_mkfsopts));
936 snprintf(mkfs_cmd, sizeof(mkfs_cmd), "%s -j -b %d -L %s ", MKE2FS,
937 mop->mo_blocksize_kb * 1024, mop->mo_ldd.ldd_svname);
939 /* For loop device format the dev, not the filename */
940 dev = mop->mo_device;
941 if (mop->mo_flags & MO_IS_LOOP)
942 dev = mop->mo_loopdev;
944 vprint("formatting backing filesystem %s on %s\n",
945 MT_STR(&mop->mo_ldd), dev);
946 vprint("\ttarget name %s\n", mop->mo_ldd.ldd_svname);
947 vprint("\tkilobytes %llu\n", mop->mo_device_kb);
948 vprint("\toptions %s\n", mop->mo_mkfsopts);
950 /* mkfs_cmd's trailing space is important! */
951 strscat(mkfs_cmd, mop->mo_mkfsopts, sizeof(mkfs_cmd));
952 strscat(mkfs_cmd, " ", sizeof(mkfs_cmd));
953 strscat(mkfs_cmd, dev, sizeof(mkfs_cmd));
954 if (mop->mo_device_kb != 0) {
955 snprintf(buf, sizeof(buf), " %lluk",
956 (unsigned long long)mop->mo_device_kb);
957 strscat(mkfs_cmd, buf, sizeof(mkfs_cmd));
960 vprint("mkfs_cmd = %s\n", mkfs_cmd);
961 ret = run_command(mkfs_cmd, sizeof(mkfs_cmd));
964 fprintf(stderr, "Unable to build fs %s (%d)\n", dev, ret);
969 int ldiskfs_prepare_lustre(struct mkfs_opts *mop,
970 char *wanted_mountopts, size_t len)
972 struct lustre_disk_data *ldd = &mop->mo_ldd;
975 /* Set MO_IS_LOOP to indicate a loopback device is needed */
976 ret = is_block(mop->mo_device);
979 } else if (ret == 0) {
980 mop->mo_flags |= MO_IS_LOOP;
983 if (IS_MDT(ldd) || IS_MGS(ldd))
984 strscat(wanted_mountopts, ",user_xattr", len);
989 int ldiskfs_fix_mountopts(struct mkfs_opts *mop, char *mountopts, size_t len)
991 if (strstr(mountopts, "errors=") == NULL)
992 strscat(mountopts, ",errors=remount-ro", len);
997 static int read_file(const char *path, char *buf, int size)
1001 fd = fopen(path, "r");
1005 if (fgets(buf, size, fd) == NULL) {
1006 fprintf(stderr, "reading from %s: %s", path, strerror(errno));
1012 /* strip trailing newline */
1014 if (buf[size - 1] == '\n')
1015 buf[size - 1] = '\0';
1020 static int write_file(const char *path, const char *buf)
1024 fd = open(path, O_WRONLY);
1028 rc = write(fd, buf, strlen(buf));
1031 return rc < 0 ? errno : 0;
1034 static int tune_md_stripe_cache_size(const char *sys_path,
1035 struct mount_opts *mop)
1037 char path[PATH_MAX];
1038 unsigned long old_stripe_cache_size;
1039 unsigned long new_stripe_cache_size;
1040 char buf[3 * sizeof(old_stripe_cache_size) + 2];
1043 if (mop->mo_md_stripe_cache_size <= 0)
1046 new_stripe_cache_size = mop->mo_md_stripe_cache_size;
1048 snprintf(path, sizeof(path), "%s/%s", sys_path, STRIPE_CACHE_SIZE);
1049 rc = read_file(path, buf, sizeof(buf));
1052 fprintf(stderr, "warning: cannot read '%s': %s\n",
1053 path, strerror(errno));
1057 old_stripe_cache_size = strtoul(buf, NULL, 0);
1058 if (old_stripe_cache_size == 0 || old_stripe_cache_size == ULONG_MAX)
1061 if (new_stripe_cache_size <= old_stripe_cache_size)
1064 snprintf(buf, sizeof(buf), "%lu", new_stripe_cache_size);
1065 rc = write_file(path, buf);
1068 fprintf(stderr, "warning: cannot write '%s': %s\n",
1069 path, strerror(errno));
1076 static int tune_max_sectors_kb(const char *sys_path, struct mount_opts *mop)
1078 char path[PATH_MAX];
1079 unsigned long max_hw_sectors_kb;
1080 unsigned long old_max_sectors_kb;
1081 unsigned long new_max_sectors_kb;
1082 char buf[3 * sizeof(old_max_sectors_kb) + 2];
1085 if (mop->mo_max_sectors_kb >= 0) {
1086 new_max_sectors_kb = mop->mo_max_sectors_kb;
1087 goto have_new_max_sectors_kb;
1090 snprintf(path, sizeof(path), "%s/%s", sys_path, MAX_HW_SECTORS_KB_PATH);
1091 rc = read_file(path, buf, sizeof(buf));
1093 /* No MAX_HW_SECTORS_KB_PATH isn't necessary an
1094 * error for some devices. */
1098 max_hw_sectors_kb = strtoul(buf, NULL, 0);
1099 if (max_hw_sectors_kb == 0 || max_hw_sectors_kb == ULLONG_MAX) {
1100 /* No digits at all or something weird. */
1104 new_max_sectors_kb = max_hw_sectors_kb;
1106 /* Don't increase IO request size limit past 16MB. It is
1107 * about PTLRPC_MAX_BRW_SIZE, but that isn't in a public
1108 * header. Note that even though the block layer allows
1109 * larger values, setting max_sectors_kb = 32768 causes
1110 * crashes (LU-6974). */
1111 if (new_max_sectors_kb > 16 * 1024)
1112 new_max_sectors_kb = 16 * 1024;
1114 have_new_max_sectors_kb:
1115 snprintf(path, sizeof(path), "%s/%s", sys_path, MAX_SECTORS_KB_PATH);
1116 rc = read_file(path, buf, sizeof(buf));
1118 /* No MAX_SECTORS_KB_PATH isn't necessary an error for
1123 old_max_sectors_kb = strtoul(buf, NULL, 0);
1124 if (old_max_sectors_kb == 0 || old_max_sectors_kb == ULLONG_MAX) {
1125 /* No digits at all or something weird. */
1129 if (new_max_sectors_kb <= old_max_sectors_kb)
1132 snprintf(buf, sizeof(buf), "%lu", new_max_sectors_kb);
1133 rc = write_file(path, buf);
1136 fprintf(stderr, "warning: cannot write '%s': %s\n",
1137 path, strerror(errno));
1141 fprintf(stderr, "%s: increased '%s' from %lu to %lu\n",
1142 progname, path, old_max_sectors_kb, new_max_sectors_kb);
1147 static int tune_block_dev_scheduler(const char *sys_path, const char *new_sched)
1149 char path[PATH_MAX];
1155 /* Before setting the scheduler, we need to check to see if
1156 * it's already set to "noop". If it is then we don't want to
1157 * override that setting. If it's set to anything other than
1158 * "noop" then set the scheduler to what has been passed
1161 snprintf(path, sizeof(path), "%s/%s", sys_path, SCHEDULER_PATH);
1162 rc = read_file(path, buf, sizeof(buf));
1165 fprintf(stderr, "%s: cannot read '%s': %s\n",
1166 progname, path, strerror(errno));
1171 /* The expected format of buf: noop anticipatory deadline [cfq] */
1172 s = strchr(buf, '[');
1173 e = strchr(buf, ']');
1175 /* If the format is not what we expect then be safe and error out. */
1176 if (s == NULL || e == NULL || !(s < e)) {
1179 "%s: cannot parse scheduler options for '%s'\n",
1188 if (strcmp(old_sched, "noop") == 0 ||
1189 strcmp(old_sched, "deadline") == 0 ||
1190 strcmp(old_sched, "mq-deadline") == 0 ||
1191 strstr(old_sched, new_sched) == 0)
1194 rc = write_file(path, new_sched);
1198 "%s: cannot set scheduler on '%s': %s\n",
1199 progname, path, strerror(errno));
1203 fprintf(stderr, "%s: changed scheduler of '%s' from %s to %s\n",
1204 progname, path, old_sched, new_sched);
1209 static int tune_block_dev(const char *src, struct mount_opts *mop);
1211 static int tune_block_dev_slaves(const char *sys_path, struct mount_opts *mop)
1213 char slaves_path[PATH_MAX];
1218 snprintf(slaves_path, sizeof(slaves_path), "%s/slaves", sys_path);
1219 slaves_dir = opendir(slaves_path);
1220 if (slaves_dir == NULL) {
1221 if (errno == ENOENT)
1227 while ((d = readdir(slaves_dir)) != NULL) {
1228 char path[PATH_MAX * 2];
1231 if (d->d_type != DT_LNK)
1234 snprintf(path, sizeof(path), "/dev/%s", d->d_name);
1235 rc2 = tune_block_dev(path, mop);
1240 closedir(slaves_dir);
1245 /* This is to tune the kernel for good SCSI performance.
1246 * For that we set the value of /sys/block/{dev}/queue/max_sectors_kb
1247 * to the value of /sys/block/{dev}/queue/max_hw_sectors_kb */
1248 static int tune_block_dev(const char *src, struct mount_opts *mop)
1251 char sys_path[PATH_MAX];
1252 char partition_path[PATH_MAX + sizeof("partition")];
1253 char *real_sys_path = NULL;
1257 * Don't apply block device tuning for MDT or MGT devices,
1258 * since we don't need huge IO sizes to get good performance
1260 if (!IS_OST(&mop->mo_ldd))
1266 rc = stat(src, &st);
1269 fprintf(stderr, "warning: cannot stat '%s': %s\n",
1270 src, strerror(errno));
1274 if (!S_ISBLK(st.st_mode))
1277 if (major(st.st_rdev) == LOOP_MAJOR)
1280 snprintf(sys_path, sizeof(sys_path), "/sys/dev/block/%u:%u",
1281 major(st.st_rdev), minor(st.st_rdev));
1283 snprintf(partition_path, sizeof(partition_path), "%s/partition",
1286 rc = access(partition_path, F_OK);
1288 if (errno == ENOENT)
1289 goto have_whole_dev;
1292 fprintf(stderr, "warning: cannot access '%s': %s\n",
1293 partition_path, strerror(errno));
1298 snprintf(sys_path, sizeof(sys_path), "/sys/dev/block/%u:%u/..",
1299 major(st.st_rdev), minor(st.st_rdev));
1302 /* Since we recurse on slave devices we resolve the sys_path to
1303 * avoid path buffer overflows. */
1304 real_sys_path = realpath(sys_path, NULL);
1305 if (real_sys_path == NULL) {
1308 "warning: cannot resolve '%s': %s\n",
1309 sys_path, strerror(errno));
1314 if (major(st.st_rdev) == MD_MAJOR) {
1315 rc = tune_md_stripe_cache_size(real_sys_path, mop);
1317 /* Ignore errors from tune_max_sectors_kb() and
1318 * tune_scheduler(). The worst that will happen is a block
1319 * device with an "incorrect" scheduler. */
1320 tune_max_sectors_kb(real_sys_path, mop);
1321 tune_block_dev_scheduler(real_sys_path, DEFAULT_SCHEDULER);
1323 /* If device is multipath device then tune its slave
1325 rc = tune_block_dev_slaves(real_sys_path, mop);
1329 free(real_sys_path);
1334 int ldiskfs_tune_lustre(char *dev, struct mount_opts *mop)
1336 return tune_block_dev(dev, mop);
1339 int ldiskfs_label_lustre(struct mount_opts *mop)
1341 char label_cmd[PATH_MAX];
1344 snprintf(label_cmd, sizeof(label_cmd),
1345 TUNE2FS" -f -L '%s' '%s' >/dev/null 2>&1",
1346 mop->mo_ldd.ldd_svname, mop->mo_source);
1347 rc = run_command(label_cmd, sizeof(label_cmd));
1352 int ldiskfs_rename_fsname(struct mkfs_opts *mop, const char *oldname)
1354 struct mount_opts opts;
1355 struct lustre_disk_data *ldd = &mop->mo_ldd;
1356 char mntpt[] = "/tmp/mntXXXXXX";
1360 /* Change the filesystem label. */
1362 opts.mo_source = mop->mo_device;
1363 ret = ldiskfs_label_lustre(&opts);
1367 fprintf(stderr, "Can't change filesystem label: %s\n",
1372 /* Mount this device temporarily in order to write these files */
1373 if (mkdtemp(mntpt) == NULL) {
1378 fprintf(stderr, "Can't create temp mount point %s: %s\n",
1379 mntpt, strerror(ret));
1383 if (mop->mo_flags & MO_IS_LOOP)
1384 dev = mop->mo_loopdev;
1386 dev = mop->mo_device;
1387 ret = mount(dev, mntpt, MT_STR(ldd), 0, ldd->ldd_mount_opts);
1391 fprintf(stderr, "Unable to mount %s: %s\n",
1392 dev, strerror(ret));
1394 fprintf(stderr, "Is the %s module available?\n",
1399 ret = lustre_rename_fsname(mop, mntpt, oldname);
1407 /* Enable quota accounting */
1408 int ldiskfs_enable_quota(struct mkfs_opts *mop)
1412 int cmdsz = sizeof(cmd), ret;
1414 if (!is_e2fsprogs_feature_supp("-O quota")) {
1415 fprintf(stderr, "%s: \"-O quota\" is is not supported by "
1416 "current e2fsprogs\n", progname);
1420 dev = mop->mo_device;
1421 if (mop->mo_flags & MO_IS_LOOP)
1422 dev = mop->mo_loopdev;
1424 /* Quota feature is already enabled? */
1425 if (is_feature_enabled("quota", dev)) {
1426 vprint("Quota feature is already enabled.\n");
1430 /* Turn on quota feature by "tune2fs -O quota" */
1431 snprintf(cmd, cmdsz, "%s -O quota %s", TUNE2FS, dev);
1432 ret = run_command(cmd, cmdsz);
1434 fprintf(stderr, "command:%s (%d)", cmd, ret);
1439 int ldiskfs_init(void)
1441 /* Required because full path to DEBUGFS is not specified */
1442 setenv("PATH", "/bin:/sbin:/usr/bin:/usr/sbin:/usr/local/bin", 0);
1447 void ldiskfs_fini(void)
1453 struct module_backfs_ops ldiskfs_ops = {
1454 .init = ldiskfs_init,
1455 .fini = ldiskfs_fini,
1456 .read_ldd = ldiskfs_read_ldd,
1457 .write_ldd = ldiskfs_write_ldd,
1458 .erase_ldd = ldiskfs_erase_ldd,
1459 .print_ldd_params = ldiskfs_print_ldd_params,
1460 .is_lustre = ldiskfs_is_lustre,
1461 .make_lustre = ldiskfs_make_lustre,
1462 .prepare_lustre = ldiskfs_prepare_lustre,
1463 .fix_mountopts = ldiskfs_fix_mountopts,
1464 .tune_lustre = ldiskfs_tune_lustre,
1465 .label_lustre = ldiskfs_label_lustre,
1466 .enable_quota = ldiskfs_enable_quota,
1467 .rename_fsname = ldiskfs_rename_fsname,
1469 #endif /* PLUGIN_DIR */