X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Futils%2Fmount_utils_ldiskfs.c;h=3e50ae7714f64bdf029cbc300eb913ecfd5481d0;hb=f41968bb9e82836a06ffcd5fd3ed6a04b9273545;hp=3815d7247ff34c72136341548c375ae057c7345a;hpb=21d716e6c16424d9deb646456758ebbaa9c70fec;p=fs%2Flustre-release.git diff --git a/lustre/utils/mount_utils_ldiskfs.c b/lustre/utils/mount_utils_ldiskfs.c index 3815d72..3e50ae7 100644 --- a/lustre/utils/mount_utils_ldiskfs.c +++ b/lustre/utils/mount_utils_ldiskfs.c @@ -27,7 +27,7 @@ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2012, 2013, Intel Corporation. + * Copyright (c) 2012, 2015, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -47,12 +47,13 @@ #ifndef _GNU_SOURCE #define _GNU_SOURCE #endif -#include "mount_utils.h" -#include +#include #include +#include +#include +#include #include #include -#include #include #include @@ -66,15 +67,10 @@ #include #include -#ifdef __linux__ -/* libcfs.h is not really needed here, but on SLES10/PPC, fs.h includes idr.h - * which requires BITS_PER_LONG to be defined */ -#include #ifndef BLKGETSIZE64 #include /* for BLKGETSIZE64 */ #endif #include -#endif #include #include #include @@ -84,6 +80,8 @@ #include #endif +#include "mount_utils.h" + #define MAX_HW_SECTORS_KB_PATH "queue/max_hw_sectors_kb" #define MAX_SECTORS_KB_PATH "queue/max_sectors_kb" #define SCHEDULER_PATH "queue/scheduler" @@ -98,8 +96,11 @@ extern char *progname; #define DUMMY_FILE_NAME_LEN 25 #define EXT3_DIRENT_SIZE DUMMY_FILE_NAME_LEN +static void append_unique(char *buf, char *prefix, char *key, char *val, + size_t maxbuflen); + /* - * Concatenate context of the temporary mount point iff selinux is enabled + * Concatenate context of the temporary mount point if selinux is enabled */ #ifdef HAVE_SELINUX static void append_context_for_mount(char *mntpt, struct mkfs_opts *mop) @@ -114,8 +115,9 @@ static void append_context_for_mount(char *mntpt, struct mkfs_opts *mop) } if (fcontext != NULL) { - strcat(mop->mo_ldd.ldd_mount_opts, ",context="); - strcat(mop->mo_ldd.ldd_mount_opts, fcontext); + append_unique(mop->mo_ldd.ldd_mount_opts, + ",", "context", fcontext, + sizeof(mop->mo_ldd.ldd_mount_opts)); freecon(fcontext); } } @@ -267,7 +269,8 @@ int ldiskfs_write_ldd(struct mkfs_opts *mop) dev = mop->mo_loopdev; ret = mount(dev, mntpt, MT_STR(&mop->mo_ldd), 0, - mop->mo_ldd.ldd_mount_opts); + (mop->mo_mountopts == NULL) ? + "errors=remount-ro" : mop->mo_mountopts); if (ret) { fprintf(stderr, "%s: Unable to mount %s: %s\n", progname, dev, strerror(errno)); @@ -409,7 +412,7 @@ static void disp_old_e2fsprogs_msg(const char *feature, int make_backfs) E2FSPROGS, feature); #if !(HAVE_LDISKFSPROGS) fprintf(stderr, "Please install the latest version of e2fsprogs from\n" - "http://downloads.whamcloud.com/public/e2fsprogs/latest/\n" + "https://downloads.hpdd.intel.com/public/e2fsprogs/latest/\n" "to enable this feature.\n"); #endif if (make_backfs) @@ -553,8 +556,9 @@ static void append_unique(char *buf, char *prefix, char *key, char *val, strscat(buf, key, maxbuflen); if (val != NULL) { - strscat(buf, "=", maxbuflen); + strscat(buf, "=\"", maxbuflen); strscat(buf, val, maxbuflen); + strscat(buf, "\"", maxbuflen); } } } @@ -564,7 +568,7 @@ static int enable_default_ext4_features(struct mkfs_opts *mop, char *anchor, { if (IS_OST(&mop->mo_ldd)) { append_unique(anchor, user_spec ? "," : " -O ", - "extents", NULL, sizeof(mop->mo_mkfsopts)); + "extents", NULL, maxbuflen); append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen); } else if (IS_MDT(&mop->mo_ldd)) { append_unique(anchor, user_spec ? "," : " -O ", @@ -620,7 +624,8 @@ static int enable_default_ext4_features(struct mkfs_opts *mop, char *anchor, append_unique(anchor, ",", "flex_bg", NULL, maxbuflen); - if (IS_OST(&mop->mo_ldd)) { + if (IS_OST(&mop->mo_ldd) && + strstr(mop->mo_mkfsopts, "-G") == NULL) { snprintf(tmp_buf, sizeof(tmp_buf), " -G %u", (1 << 20) / L_BLOCK_SIZE); strscat(anchor, tmp_buf, maxbuflen); @@ -699,9 +704,9 @@ int ldiskfs_make_lustre(struct mkfs_opts *mop) } if (mop->mo_device_kb != 0) { - if (mop->mo_device_kb < 8096) { + if (mop->mo_device_kb < 32384) { fprintf(stderr, "%s: size of filesystem must be larger " - "than 8MB, but is set to %lldKB\n", + "than 32MB, but is set to %lldKB\n", progname, (long long)mop->mo_device_kb); return EINVAL; } @@ -745,17 +750,32 @@ int ldiskfs_make_lustre(struct mkfs_opts *mop) } } - /* Inode size (for extended attributes). The LOV EA size is - * 32 (EA hdr) + 32 (lov_mds_md) + stripes * 24 (lov_ost_data), - * and we want some margin above that for ACLs, other EAs... */ + /* Inode size includes: + * ldiskfs inode size: 156 + * extended attributes size, including: + * ext4_xattr_header: 32 + * LOV EA size: 32(lov_mds_md) + + * stripes * 24(lov_ost_data) + + * 16(xattr_entry) + 3(lov) + * LMA EA size: 24(lustre_mdt_attrs) + + * 16(xattr_entry) + 3(lma) + * link EA size: 24(link_ea_header) + 18(link_ea_entry) + + * (filename) + 16(xattr_entry) + 4(link) + * and some margin for 4-byte alignment, ACLs and other EAs. + * + * If we say the average filename length is about 32 bytes, + * the calculation looks like: + * 156 + 32 + (32+24*N+19) + (24+19) + (24+18+~32+20) + other <= + * 512*2^m, {m=0,1,2,3} + */ if (strstr(mop->mo_mkfsopts, "-I") == NULL) { if (IS_MDT(&mop->mo_ldd)) { - if (mop->mo_stripe_count > 72) + if (mop->mo_stripe_count > 69) inode_size = 512; /* bz 7241 */ /* see also "-i" below for EA blocks */ - else if (mop->mo_stripe_count > 32) + else if (mop->mo_stripe_count > 26) inode_size = 2048; - else if (mop->mo_stripe_count > 10) + else if (mop->mo_stripe_count > 5) inode_size = 1024; else inode_size = 512; @@ -785,7 +805,7 @@ int ldiskfs_make_lustre(struct mkfs_opts *mop) if (IS_MDT(&mop->mo_ldd)) { bytes_per_inode = inode_size + 1536; - if (mop->mo_stripe_count > 72) { + if (mop->mo_stripe_count > 69) { int extra = mop->mo_stripe_count * 24; extra = ((extra - 1) | 4095) + 1; bytes_per_inode += extra; @@ -909,8 +929,8 @@ int ldiskfs_make_lustre(struct mkfs_opts *mop) vprint("formatting backing filesystem %s on %s\n", MT_STR(&mop->mo_ldd), dev); - vprint("\ttarget name %s\n", mop->mo_ldd.ldd_svname); - vprint("\t4k blocks "LPU64"\n", block_count); + vprint("\ttarget name %s\n", mop->mo_ldd.ldd_svname); + vprint("\t4k blocks %ju\n", (uintmax_t)block_count); vprint("\toptions %s\n", mop->mo_mkfsopts); /* mkfs_cmd's trailing space is important! */ @@ -918,7 +938,8 @@ int ldiskfs_make_lustre(struct mkfs_opts *mop) strscat(mkfs_cmd, " ", sizeof(mkfs_cmd)); strscat(mkfs_cmd, dev, sizeof(mkfs_cmd)); if (block_count != 0) { - sprintf(buf, " "LPU64, block_count); + snprintf(buf, sizeof(buf), " %ju", + (uintmax_t)block_count); strscat(mkfs_cmd, buf, sizeof(mkfs_cmd)); } @@ -932,8 +953,7 @@ int ldiskfs_make_lustre(struct mkfs_opts *mop) } int ldiskfs_prepare_lustre(struct mkfs_opts *mop, - char *default_mountopts, int default_len, - char *always_mountopts, int always_len) + char *wanted_mountopts, size_t len) { struct lustre_disk_data *ldd = &mop->mo_ldd; int ret; @@ -946,9 +966,16 @@ int ldiskfs_prepare_lustre(struct mkfs_opts *mop, mop->mo_flags |= MO_IS_LOOP; } - strscat(default_mountopts, ",errors=remount-ro", default_len); if (IS_MDT(ldd) || IS_MGS(ldd)) - strscat(always_mountopts, ",user_xattr", always_len); + strscat(wanted_mountopts, ",user_xattr", len); + + return 0; +} + +int ldiskfs_fix_mountopts(struct mkfs_opts *mop, char *mountopts, size_t len) +{ + if (strstr(mountopts, "errors=") == NULL) + strscat(mountopts, ",errors=remount-ro", len); return 0; } @@ -961,32 +988,38 @@ static int read_file(const char *path, char *buf, int size) if (fd == NULL) return errno; - /* should not ignore fgets(3)'s return value */ - if (!fgets(buf, size, fd)) { + if (fgets(buf, size, fd) == NULL) { fprintf(stderr, "reading from %s: %s", path, strerror(errno)); fclose(fd); return 1; } fclose(fd); + + /* strip trailing newline */ + size = strlen(buf); + if (buf[size - 1] == '\n') + buf[size - 1] = '\0'; + return 0; } static int write_file(const char *path, const char *buf) { - FILE *fd; + int fd, rc; - fd = fopen(path, "w"); - if (fd == NULL) + fd = open(path, O_WRONLY); + if (fd < 0) return errno; - fputs(buf, fd); - fclose(fd); - return 0; + rc = write(fd, buf, strlen(buf)); + close(fd); + + return rc < 0 ? errno : 0; } static int set_blockdev_scheduler(const char *path, const char *scheduler) { - char buf[PATH_MAX], *c; + char buf[PATH_MAX], *s, *e, orig_sched[50]; int rc; /* Before setting the scheduler, we need to check to see if it's @@ -1003,18 +1036,21 @@ static int set_blockdev_scheduler(const char *path, const char *scheduler) } /* The expected format of buf: noop anticipatory deadline [cfq] */ - c = strchr(buf, '['); + s = strchr(buf, '['); + e = strchr(buf, ']'); - /* If c is NULL, the format is not what we expect. Play it safe - * and error out. */ - if (c == NULL) { + /* If the format is not what we expect. Play it safe and error out. */ + if (s == NULL || e == NULL) { if (verbose) fprintf(stderr, "%s: cannot parse scheduler " "options for '%s'\n", progname, path); return -EINVAL; } - if (strncmp(c+1, "noop", 4) == 0) + snprintf(orig_sched, e - s, "%s", s + 1); + + if (strcmp(orig_sched, "noop") == 0 || + strcmp(orig_sched, scheduler) == 0) return 0; rc = write_file(path, scheduler); @@ -1024,6 +1060,9 @@ static int set_blockdev_scheduler(const char *path, const char *scheduler) "'%s': %s\n", progname, path, strerror(errno)); return rc; + } else { + fprintf(stderr, "%s: change scheduler of %s from %s to %s\n", + progname, path, orig_sched, scheduler); } return rc; @@ -1149,7 +1188,7 @@ set_params: snprintf(buf, sizeof(buf), "%d", mop->mo_md_stripe_cache_size); rc = write_file(real_path, buf); - if (rc && verbose) + if (rc != 0 && verbose) fprintf(stderr, "warning: opening %s: %s\n", real_path, strerror(errno)); } @@ -1166,23 +1205,55 @@ set_params: real_path, strerror(errno)); /* No MAX_HW_SECTORS_KB_PATH isn't necessary an * error for some device. */ - rc = 0; + goto subdevs; } if (strlen(buf) - 1 > 0) { + char oldbuf[32] = "", *end = NULL; + unsigned long long oldval, newval; + snprintf(real_path, sizeof(real_path), "%s/%s", path, MAX_SECTORS_KB_PATH); + rc = read_file(real_path, oldbuf, sizeof(oldbuf)); + /* Only set new parameter if different from the old one. */ + if (rc != 0 || strcmp(oldbuf, buf) == 0) { + /* No MAX_SECTORS_KB_PATH isn't necessary an + * error for some device. */ + goto subdevs; + } + + newval = strtoull(buf, &end, 0); + if (newval == 0 || newval == ULLONG_MAX || end == buf) + goto subdevs; + + /* Don't increase IO request size limit past 16MB. It is about + * PTLRPC_MAX_BRW_SIZE, but that isn't in a public header. + * Note that even though the block layer allows larger values, + * setting max_sectors_kb = 32768 causes crashes (LU-6974). */ + if (newval > 16 * 1024) { + newval = 16 * 1024; + snprintf(buf, sizeof(buf), "%llu", newval); + } + + oldval = strtoull(oldbuf, &end, 0); + /* Don't shrink the current limit. */ + if (oldval != ULLONG_MAX && newval <= oldval) + goto subdevs; + rc = write_file(real_path, buf); - if (rc) { + if (rc != 0) { if (verbose) fprintf(stderr, "warning: writing to %s: %s\n", real_path, strerror(errno)); /* No MAX_SECTORS_KB_PATH isn't necessary an * error for some device. */ - rc = 0; + goto subdevs; } + fprintf(stderr, "%s: increased %s from %s to %s\n", + progname, real_path, oldbuf, buf); } +subdevs: /* Purposely ignore errors reported from set_blockdev_scheduler. * The worst that will happen is a block device with an "incorrect" * scheduler. */