From 34c10968cbe3b5591b3c0ce225b8694edd9709d0 Mon Sep 17 00:00:00 2001 From: Lennart Poettering Date: Wed, 18 Jun 2014 22:02:18 +0200 Subject: [PATCH] coredump: optionally store coredumps on disk, not in the journal Introduce a new configuration file /etc/systemd/coredump.conf to configure when to place coredumps in the journal and when on disk. Since the coredumps are quite large, default to storing them only on disk. --- Makefile.am | 8 + src/journal/coredump.c | 443 ++++++++++++++++++++++++++-------- src/journal/coredump.conf | 14 ++ src/journal/journald-server.c | 2 +- src/shared/acl-util.h | 8 + src/shared/sleep-config.c | 4 +- src/sleep/sleep.c | 1 + 7 files changed, 382 insertions(+), 98 deletions(-) create mode 100644 src/journal/coredump.conf diff --git a/Makefile.am b/Makefile.am index 82145c676..7c20e3384 100644 --- a/Makefile.am +++ b/Makefile.am @@ -3689,6 +3689,14 @@ systemd_coredump_LDADD = \ rootlibexec_PROGRAMS += \ systemd-coredump +dist_pkgsysconf_DATA += \ + src/journal/coredump.conf + +if HAVE_ACL +systemd_coredump_LDADD += \ + libsystemd-acl.la +endif + systemd_coredumpctl_SOURCES = \ src/journal/coredumpctl.c diff --git a/src/journal/coredump.c b/src/journal/coredump.c index 74b1dbd01..a396491a4 100644 --- a/src/journal/coredump.c +++ b/src/journal/coredump.c @@ -23,12 +23,11 @@ #include #include #include +#include +#include #include - -#ifdef HAVE_LOGIND #include -#endif #include "log.h" #include "util.h" @@ -37,13 +36,28 @@ #include "special.h" #include "cgroup-util.h" #include "journald-native.h" +#include "conf-parser.h" +#include "copy.h" + +#ifdef HAVE_ACL +#include +#include "acl-util.h" +#endif + +/* The maximum size up to which we process coredumps */ +#define PROCESS_SIZE_MAX ((off_t) (2LLU*1024LLU*1024LLU*1024LLU)) + +/* The maximum size up to which we leave the coredump around on + * disk */ +#define EXTERNAL_SIZE_MAX PROCESS_SIZE_MAX + +/* The maximum size up to which we store the coredump in the + * journal */ +#define JOURNAL_SIZE_MAX ((size_t) (767LU*1024LU*1024LU)) -/* Few programs have less than 3MiB resident */ -#define COREDUMP_MIN_START (3*1024*1024u) /* Make sure to not make this larger than the maximum journal entry * size. See ENTRY_SIZE_MAX in journald-native.c. */ -#define COREDUMP_MAX (767*1024*1024u) -assert_cc(COREDUMP_MAX <= ENTRY_SIZE_MAX); +assert_cc(JOURNAL_SIZE_MAX <= ENTRY_SIZE_MAX); enum { ARG_PID = 1, @@ -55,44 +69,272 @@ enum { _ARG_MAX }; -static int divert_coredump(void) { - _cleanup_fclose_ FILE *f = NULL; +typedef enum CoredumpStorage { + COREDUMP_STORAGE_NONE, + COREDUMP_STORAGE_EXTERNAL, + COREDUMP_STORAGE_JOURNAL, + COREDUMP_STORAGE_BOTH, + _COREDUMP_STORAGE_MAX, + _COREDUMP_STORAGE_INVALID = -1 +} CoredumpStorage; + +static CoredumpStorage arg_storage = COREDUMP_STORAGE_EXTERNAL; +static off_t arg_process_size_max = PROCESS_SIZE_MAX; +static off_t arg_external_size_max = EXTERNAL_SIZE_MAX; +static size_t arg_journal_size_max = JOURNAL_SIZE_MAX; + +static const char* const coredump_storage_table[_COREDUMP_STORAGE_MAX] = { + [COREDUMP_STORAGE_NONE] = "none", + [COREDUMP_STORAGE_EXTERNAL] = "external", + [COREDUMP_STORAGE_JOURNAL] = "journal", + [COREDUMP_STORAGE_BOTH] = "both", +}; + +DEFINE_PRIVATE_STRING_TABLE_LOOKUP(coredump_storage, CoredumpStorage); +static DEFINE_CONFIG_PARSE_ENUM(config_parse_coredump_storage, coredump_storage, CoredumpStorage, "Failed to parse storage setting"); + +static int parse_config(void) { + + static const ConfigTableItem items[] = { + { "Coredump", "ProcessSizeMax", config_parse_iec_off, 0, &arg_process_size_max }, + { "Coredump", "ExternalSizeMax", config_parse_iec_off, 0, &arg_external_size_max }, + { "Coredump", "JournalSizeMax", config_parse_iec_size, 0, &arg_journal_size_max }, + { "Coredump", "Storage", config_parse_coredump_storage, 0, &arg_storage }, + {} + }; + + return config_parse( + NULL, + "/etc/systemd/coredump.conf", + NULL, + "Coredump\0", + config_item_table_lookup, + (void*) items, + false, + false, + NULL); +} + +static int fix_acl(int fd, uid_t uid) { + +#ifdef HAVE_ACL + _cleanup_(acl_freep) acl_t acl = NULL; + acl_entry_t entry; + acl_permset_t permset; + + if (uid <= SYSTEM_UID_MAX) + return 0; + + /* Make sure normal users can read (but not write or delete) + * their own coredumps */ + + acl = acl_get_fd(fd); + if (!acl) { + log_error("Failed to get ACL: %m"); + return -errno; + } + + if (acl_create_entry(&acl, &entry) < 0 || + acl_set_tag_type(entry, ACL_USER) < 0 || + acl_set_qualifier(entry, &uid) < 0) { + log_error("Failed to patch ACL: %m"); + return -errno; + } + + if (acl_get_permset(entry, &permset) < 0 || + acl_add_perm(permset, ACL_READ) < 0 || + calc_acl_mask_if_needed(&acl) < 0) { + log_warning("Failed to patch ACL: %m"); + return -errno; + } + + if (acl_set_fd(fd, acl) < 0) { + log_error("Failed to apply ACL: %m"); + return -errno; + } +#endif + + return 0; +} + +static int fix_xattr(int fd, char *argv[]) { + int r = 0; + + /* Attach some metadate to coredumps via extended + * attributes. Just because we can. */ + + if (!isempty(argv[ARG_PID])) + if (fsetxattr(fd, "user.coredump.pid", argv[ARG_PID], strlen(argv[ARG_PID]), XATTR_CREATE) < 0) + r = -errno; + + if (!isempty(argv[ARG_UID])) + if (fsetxattr(fd, "user.coredump.uid", argv[ARG_UID], strlen(argv[ARG_UID]), XATTR_CREATE) < 0) + r = -errno; + + if (!isempty(argv[ARG_GID])) + if (fsetxattr(fd, "user.coredump.gid", argv[ARG_GID], strlen(argv[ARG_GID]), XATTR_CREATE) < 0) + r = -errno; + + if (!isempty(argv[ARG_SIGNAL])) + if (fsetxattr(fd, "user.coredump.signal", argv[ARG_SIGNAL], strlen(argv[ARG_SIGNAL]), XATTR_CREATE) < 0) + r = -errno; + + if (!isempty(argv[ARG_TIMESTAMP])) + if (fsetxattr(fd, "user.coredump.timestamp", argv[ARG_TIMESTAMP], strlen(argv[ARG_TIMESTAMP]), XATTR_CREATE) < 0) + r = -errno; - log_info("Detected coredump of the journal daemon itself, diverting coredump to /var/lib/systemd/coredump/."); + if (!isempty(argv[ARG_COMM])) + if (fsetxattr(fd, "user.coredump.comm", argv[ARG_COMM], strlen(argv[ARG_COMM]), XATTR_CREATE) < 0) + r = -errno; + + return r; +} + +#define filename_escape(s) xescape((s), "./") + +static int save_external_coredump(char **argv, uid_t uid, char **ret_filename, int *ret_fd, off_t *ret_size) { + _cleanup_free_ char *p = NULL, *t = NULL, *c = NULL, *fn = NULL, *tmp = NULL; + _cleanup_close_ int fd = -1; + sd_id128_t boot; + struct stat st; + int r; + + assert(argv); + assert(ret_filename); + assert(ret_fd); + assert(ret_size); + + c = filename_escape(argv[ARG_COMM]); + if (!c) + return log_oom(); + + p = filename_escape(argv[ARG_PID]); + if (!p) + return log_oom(); + + t = filename_escape(argv[ARG_TIMESTAMP]); + if (!t) + return log_oom(); + + r = sd_id128_get_boot(&boot); + if (r < 0) { + log_error("Failed to determine boot ID: %s", strerror(-r)); + return r; + } + + r = asprintf(&fn, + "/var/lib/systemd/coredump/core.%s." SD_ID128_FORMAT_STR ".%s.%s000000", + c, + SD_ID128_FORMAT_VAL(boot), + p, + t); + if (r < 0) + return log_oom(); + + tmp = tempfn_random(fn); + if (!tmp) + return log_oom(); mkdir_p_label("/var/lib/systemd/coredump", 0755); - f = fopen("/var/lib/systemd/coredump/core.systemd-journald", "we"); - if (!f) { + fd = open(tmp, O_CREAT|O_EXCL|O_RDWR|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW, 0640); + if (fd < 0) { log_error("Failed to create coredump file: %m"); return -errno; } - for (;;) { - uint8_t buffer[4096]; - size_t l, q; + r = copy_bytes(STDIN_FILENO, fd); + if (r < 0) { + log_error("Failed to dump coredump to file: %s", strerror(-r)); + goto fail; + } - l = fread(buffer, 1, sizeof(buffer), stdin); - if (l <= 0) { - if (ferror(f)) { - log_error("Failed to read coredump: %m"); - return -errno; - } + /* Ignore errors on these */ + fchmod(fd, 0640); + fix_acl(fd, uid); + fix_xattr(fd, argv); - break; - } + if (fsync(fd) < 0) { + log_error("Failed to sync coredump: %m"); + r = -errno; + goto fail; + } - q = fwrite(buffer, 1, l, f); - if (q != l) { - log_error("Failed to write coredump: %m"); - return -errno; - } + if (fstat(fd, &st) < 0) { + log_error("Failed to fstat coredump: %m"); + r = -errno; + goto fail; + } + + if (rename(tmp, fn) < 0) { + log_error("Failed to rename coredump: %m"); + r = -errno; + goto fail; + } + + *ret_filename = fn; + *ret_fd = fd; + *ret_size = st.st_size; + + fn = NULL; + fd = -1; + + return 0; + +fail: + unlink_noerrno(tmp); + return r; +} + +static int allocate_journal_field(int fd, size_t size, char **ret, size_t *ret_size) { + _cleanup_free_ char *field = NULL; + ssize_t n; + + assert(ret); + assert(ret_size); + + if (lseek(fd, 0, SEEK_SET) == (off_t) -1) { + log_warning("Failed to seek: %m"); + return -errno; } - fflush(f); + field = malloc(9 + size); + if (!field) { + log_warning("Failed to allocate memory fore coredump, coredump will not be stored."); + return -ENOMEM; + } + + memcpy(field, "COREDUMP=", 9); + + n = read(fd, field + 9, size); + if (n < 0) { + log_error("Failed to read core data: %s", strerror(-n)); + return (int) n; + } + if ((size_t) n < size) { + log_error("Core data too short."); + return -EIO; + } + + *ret = field; + *ret_size = size + 9; + + field = NULL; + + return 0; +} - if (ferror(f)) { - log_error("Failed to write coredump: %m"); +static int maybe_remove_external_coredump(const char *filename, off_t size) { + + if (!filename) + return 0; + + if (IN_SET(arg_storage, COREDUMP_STORAGE_EXTERNAL, COREDUMP_STORAGE_BOTH) && + size <= arg_external_size_max) + return 0; + + if (unlink(filename) < 0) { + log_error("Failed to unlink %s: %m", filename); return -errno; } @@ -100,47 +342,80 @@ static int divert_coredump(void) { } int main(int argc, char* argv[]) { + + _cleanup_free_ char *core_pid = NULL, *core_uid = NULL, *core_gid = NULL, *core_signal = NULL, + *core_timestamp = NULL, *core_comm = NULL, *core_exe = NULL, *core_unit = NULL, + *core_session = NULL, *core_message = NULL, *core_cmdline = NULL, *coredump_data = NULL, + *coredump_filename = NULL; + + _cleanup_close_ int coredump_fd = -1; + + struct iovec iovec[14]; + off_t coredump_size; int r, j = 0; - char *t; - ssize_t n; pid_t pid; uid_t uid; gid_t gid; - struct iovec iovec[14]; - size_t coredump_bufsize = 0, coredump_size = 0; - _cleanup_free_ char *core_pid = NULL, *core_uid = NULL, *core_gid = NULL, *core_signal = NULL, - *core_timestamp = NULL, *core_comm = NULL, *core_exe = NULL, *core_unit = NULL, - *core_session = NULL, *core_message = NULL, *core_cmdline = NULL, *coredump_data = NULL; + char *t; + /* Make sure we never enter a loop */ prctl(PR_SET_DUMPABLE, 0); - if (argc != _ARG_MAX) { - log_set_target(LOG_TARGET_JOURNAL_OR_KMSG); - log_open(); + /* First, log to a safe place, since we don't know what + * crashed and it might be journald which we'd rather not log + * to then. */ + log_set_target(LOG_TARGET_KMSG); + log_set_max_level(LOG_DEBUG); + log_open(); + if (argc != _ARG_MAX) { log_error("Invalid number of arguments passed from kernel."); r = -EINVAL; goto finish; } - r = parse_pid(argv[ARG_PID], &pid); + /* Ignore all parse errors */ + parse_config(); + log_debug("Selected storage '%s'.", coredump_storage_to_string(arg_storage)); + + r = parse_uid(argv[ARG_UID], &uid); if (r < 0) { - log_set_target(LOG_TARGET_JOURNAL_OR_KMSG); - log_open(); + log_error("Failed to parse UID."); + goto finish; + } + r = parse_pid(argv[ARG_PID], &pid); + if (r < 0) { log_error("Failed to parse PID."); goto finish; } + r = parse_gid(argv[ARG_GID], &gid); + if (r < 0) { + log_error("Failed to parse GID."); + goto finish; + } + if (cg_pid_get_unit(pid, &t) >= 0) { if (streq(t, SPECIAL_JOURNALD_SERVICE)) { - /* Make sure we don't make use of the journal, - * if it's the journal which is crashing */ - log_set_target(LOG_TARGET_KMSG); - log_open(); - r = divert_coredump(); + /* If we are journald, we cut things short, + * don't write to the journal, but still + * create a coredump. */ + + if (arg_storage != COREDUMP_STORAGE_NONE) + arg_storage = COREDUMP_STORAGE_EXTERNAL; + + r = save_external_coredump(argv, uid, &coredump_filename, &coredump_fd, &coredump_size); + if (r < 0) + goto finish; + + r = maybe_remove_external_coredump(coredump_filename, coredump_size); + if (r < 0) + goto finish; + + log_info("Detected coredump of the journal daemon itself, diverted to %s.", coredump_filename); goto finish; } @@ -151,23 +426,11 @@ int main(int argc, char* argv[]) { if (core_unit) IOVEC_SET_STRING(iovec[j++], core_unit); - /* OK, now we know it's not the journal, hence make use of - * it */ + /* OK, now we know it's not the journal, hence we can make use + * of it now. */ log_set_target(LOG_TARGET_JOURNAL_OR_KMSG); log_open(); - r = parse_uid(argv[ARG_UID], &uid); - if (r < 0) { - log_error("Failed to parse UID."); - goto finish; - } - - r = parse_gid(argv[ARG_GID], &gid); - if (r < 0) { - log_error("Failed to parse GID."); - goto finish; - } - core_pid = strappend("COREDUMP_PID=", argv[ARG_PID]); if (core_pid) IOVEC_SET_STRING(iovec[j++], core_pid); @@ -188,7 +451,6 @@ int main(int argc, char* argv[]) { if (core_comm) IOVEC_SET_STRING(iovec[j++], core_comm); -#ifdef HAVE_LOGIND if (sd_pid_get_session(pid, &t) >= 0) { core_session = strappend("COREDUMP_SESSION=", t); free(t); @@ -197,8 +459,6 @@ int main(int argc, char* argv[]) { IOVEC_SET_STRING(iovec[j++], core_session); } -#endif - if (get_process_exe(pid, &t) >= 0) { core_exe = strappend("COREDUMP_EXE=", t); free(t); @@ -226,12 +486,22 @@ int main(int argc, char* argv[]) { if (core_message) IOVEC_SET_STRING(iovec[j++], core_message); + /* Always stream the coredump to disk, if that's possible */ + r = save_external_coredump(argv, uid, &coredump_filename, &coredump_fd, &coredump_size); + if (r < 0) + goto finish; + + /* If we don't want to keep the coredump on disk, remove it + * now, as later on we will lack the privileges for it. */ + r = maybe_remove_external_coredump(coredump_filename, coredump_size); + if (r < 0) + goto finish; + /* Now, let's drop privileges to become the user who owns the * segfaulted process and allocate the coredump memory under * his uid. This also ensures that the credentials journald * will see are the ones of the coredumping user, thus making * sure the user himself gets access to the core dump. */ - if (setresgid(gid, gid, gid) < 0 || setresuid(uid, uid, uid) < 0) { log_error("Failed to drop privileges: %m"); @@ -239,40 +509,21 @@ int main(int argc, char* argv[]) { goto finish; } - for (;;) { - if (!GREEDY_REALLOC(coredump_data, coredump_bufsize, - MAX(coredump_size + 1, COREDUMP_MIN_START/2))) { - log_warning("Failed to allocate memory for core, core will not be stored."); - goto finalize; - } + /* Optionally store the entire coredump in the journal */ + if (IN_SET(arg_storage, COREDUMP_STORAGE_JOURNAL, COREDUMP_STORAGE_BOTH) && + coredump_size <= (off_t) arg_journal_size_max) { + size_t sz; - if (coredump_size == 0) { - memcpy(coredump_data, "COREDUMP=", 9); - coredump_size = 9; - } + /* Store the coredump itself in the journal */ - n = loop_read(STDIN_FILENO, coredump_data + coredump_size, - coredump_bufsize - coredump_size, false); - if (n < 0) { - log_error("Failed to read core data: %s", strerror(-n)); - r = (int) n; - goto finish; - } else if (n == 0) - break; - - coredump_size += n; - - if (coredump_size > COREDUMP_MAX) { - log_error("Core too large, core will not be stored."); - goto finalize; + r = allocate_journal_field(coredump_fd, (size_t) coredump_size, &coredump_data, &sz); + if (r >= 0) { + iovec[j].iov_base = coredump_data; + iovec[j].iov_len = sz; + j++; } } - iovec[j].iov_base = coredump_data; - iovec[j].iov_len = coredump_size; - j++; - -finalize: r = sd_journal_sendv(iovec, j); if (r < 0) log_error("Failed to log coredump: %s", strerror(-r)); diff --git a/src/journal/coredump.conf b/src/journal/coredump.conf new file mode 100644 index 000000000..53e471ed7 --- /dev/null +++ b/src/journal/coredump.conf @@ -0,0 +1,14 @@ +# This file is part of systemd. +# +# systemd is free software; you can redistribute it and/or modify it +# under the terms of the GNU Lesser General Public License as published by +# the Free Software Foundation; either version 2.1 of the License, or +# (at your option) any later version. +# +# See coredump.conf(5) for details + +[Coredump] +#Storage=external +#ProcessSizeMax=2G +#ExternalSizeMax=2G +#JournalSizeMax=767M diff --git a/src/journal/journald-server.c b/src/journal/journald-server.c index eda5dcf7e..39a1a07ed 100644 --- a/src/journal/journald-server.c +++ b/src/journal/journald-server.c @@ -205,7 +205,7 @@ void server_fix_perms(Server *s, JournalFile *f, uid_t uid) { log_warning("Failed to fix access mode on %s, ignoring: %s", f->path, strerror(-r)); #ifdef HAVE_ACL - if (uid <= 0) + if (uid <= SYSTEM_UID_MAX) return; acl = acl_get_fd(f->fd); diff --git a/src/shared/acl-util.h b/src/shared/acl-util.h index 36ef490d7..a753ad14f 100644 --- a/src/shared/acl-util.h +++ b/src/shared/acl-util.h @@ -26,3 +26,11 @@ int acl_find_uid(acl_t acl, uid_t uid, acl_entry_t *entry); int calc_acl_mask_if_needed(acl_t *acl_p); int search_acl_groups(char*** dst, const char* path, bool* belong); + +static inline void acl_freep(acl_t *acl) { + + if (!*acl) + return; + + acl_free(*acl); +} diff --git a/src/shared/sleep-config.c b/src/shared/sleep-config.c index cf1cd400e..1972cdbb3 100644 --- a/src/shared/sleep-config.c +++ b/src/shared/sleep-config.c @@ -31,6 +31,7 @@ #define USE(x, y) do{ (x) = (y); (y) = NULL; } while(0) int parse_sleep_config(const char *verb, char ***_modes, char ***_states) { + _cleanup_strv_free_ char **suspend_mode = NULL, **suspend_state = NULL, **hibernate_mode = NULL, **hibernate_state = NULL, @@ -44,7 +45,8 @@ int parse_sleep_config(const char *verb, char ***_modes, char ***_states) { { "Sleep", "HibernateState", config_parse_strv, 0, &hibernate_state }, { "Sleep", "HybridSleepMode", config_parse_strv, 0, &hybrid_mode }, { "Sleep", "HybridSleepState", config_parse_strv, 0, &hybrid_state }, - {}}; + {} + }; int r; FILE _cleanup_fclose_ *f; diff --git a/src/sleep/sleep.c b/src/sleep/sleep.c index 94bcb2944..118d10c23 100644 --- a/src/sleep/sleep.c +++ b/src/sleep/sleep.c @@ -135,6 +135,7 @@ static int execute(char **modes, char **states) { } static int help(void) { + printf("%s COMMAND\n\n" "Suspend the system, hibernate the system, or both.\n\n" "Commands:\n" -- 2.30.2