chiark / gitweb /
Merge branch 'master' into journal
authorLennart Poettering <lennart@poettering.net>
Tue, 8 Nov 2011 18:44:16 +0000 (19:44 +0100)
committerLennart Poettering <lennart@poettering.net>
Tue, 8 Nov 2011 18:44:16 +0000 (19:44 +0100)
34 files changed:
.gitignore
Makefile.am
TODO
po/.gitignore
src/acl-util.c [new file with mode: 0644]
src/acl-util.h [new file with mode: 0644]
src/automount.c
src/cgroup-show.c
src/journal/Makefile [new symlink]
src/journal/journal-def.h [new file with mode: 0644]
src/journal/journal-file.c [new file with mode: 0644]
src/journal/journal-file.h [new file with mode: 0644]
src/journal/journalctl.c [new file with mode: 0644]
src/journal/journald.c [new file with mode: 0644]
src/journal/lookup3.c [new file with mode: 0644]
src/journal/lookup3.h [new file with mode: 0644]
src/journal/sd-journal.c [new file with mode: 0644]
src/journal/sd-journal.h [new file with mode: 0644]
src/journal/test-journal.c [new file with mode: 0644]
src/kmsg-syslogd.c
src/loginctl.c
src/logind-acl.c
src/machine-id-setup.c
src/macro.h
src/manager.c
src/pam-module.c
src/sd-id128.c [new file with mode: 0644]
src/sd-id128.h [new file with mode: 0644]
src/stdout-syslog-bridge.c
src/systemctl.c
src/test-id128.c [new file with mode: 0644]
src/util.c
src/util.h
tmpfiles.d/Makefile [new symlink]

index f4a8a45b4a10923b5341f3e728cbc3cd628afc12..265801ff5fc71a801f3f8b03aa609f8332d41fdf 100644 (file)
@@ -1,3 +1,7 @@
+systemd-journalctl
+systemd-journald
+test-id128
+test-journal
 test-install
 org.freedesktop.hostname1.xml
 org.freedesktop.locale1.xml
@@ -91,7 +95,7 @@ install-sh
 missing
 stamp-*
 *.stamp
-Makefile
+/Makefile
 ltmain.sh
 *.tar.bz2
 *.tar.gz
index d30d626ebfa3ef34c6d73ae0983c9dfe92c97795..58b3a63517fe7aca3ffdcc51031ab0e99b3d0445 100644 (file)
@@ -141,7 +141,8 @@ rootbin_PROGRAMS = \
        systemd-ask-password \
        systemd-tty-ask-password-agent \
        systemd-tmpfiles \
-       systemd-machine-id-setup
+       systemd-machine-id-setup \
+        systemd-journalctl
 
 bin_PROGRAMS = \
        systemd-cgls \
@@ -180,7 +181,8 @@ rootlibexec_PROGRAMS = \
        systemd-detect-virt \
        systemd-sysctl \
         systemd-logind \
-        systemd-uaccess
+        systemd-uaccess \
+        systemd-journald
 
 if ENABLE_BINFMT
 rootlibexec_PROGRAMS += \
@@ -232,7 +234,9 @@ noinst_PROGRAMS = \
        test-env-replace \
        test-strv \
         test-login \
-        test-install
+        test-install \
+        test-id128 \
+        test-journal
 
 if HAVE_PAM
 pamlib_LTLIBRARIES = \
@@ -701,7 +705,8 @@ libsystemd_core_la_SOURCES = \
         src/dbus-common.c \
         src/sd-daemon.c \
         src/install.c \
-        src/cgroup-attr.c
+        src/cgroup-attr.c \
+        src/sd-id128.c
 
 nodist_libsystemd_core_la_SOURCES = \
         src/load-fragment-gperf.c \
@@ -963,6 +968,60 @@ test_install_CFLAGS = \
 test_install_LDADD = \
        libsystemd-basic.la
 
+test_id128_SOURCES = \
+       src/test-id128.c \
+        src/sd-id128.c
+
+test_id128_CFLAGS = \
+       $(AM_CFLAGS)
+
+test_id128_LDADD = \
+       libsystemd-basic.la
+
+test_journal_SOURCES = \
+       src/journal/test-journal.c \
+       src/journal/sd-journal.c \
+        src/journal/journal-file.c \
+        src/journal/lookup3.c \
+        src/sd-id128.c
+
+test_journal_CFLAGS = \
+       $(AM_CFLAGS)
+
+test_journal_LDADD = \
+       libsystemd-basic.la
+
+systemd_journald_SOURCES = \
+       src/journal/journald.c \
+       src/journal/sd-journal.c \
+        src/journal/journal-file.c \
+        src/journal/lookup3.c \
+        src/sd-id128.c \
+        src/acl-util.c \
+        src/cgroup-util.c
+
+systemd_journald_CFLAGS = \
+       $(AM_CFLAGS) \
+        $(ACL_CFLAGS)
+
+systemd_journald_LDADD = \
+       libsystemd-basic.la \
+        libsystemd-daemon.la \
+        $(ACL_LIBS)
+
+systemd_journalctl_SOURCES = \
+       src/journal/journalctl.c \
+       src/journal/sd-journal.c \
+        src/journal/journal-file.c \
+        src/journal/lookup3.c \
+        src/sd-id128.c
+
+systemd_journalctl_CFLAGS = \
+       $(AM_CFLAGS)
+
+systemd_journalctl_LDADD = \
+       libsystemd-basic.la
+
 systemd_stdout_syslog_bridge_SOURCES = \
        src/stdout-syslog-bridge.c \
        src/tcpwrap.c
@@ -1107,10 +1166,12 @@ systemd_uaccess_SOURCES = \
 
 if HAVE_ACL
 systemd_logind_SOURCES += \
-       src/logind-acl.c
+       src/logind-acl.c \
+        src/acl-util.c
 
 systemd_uaccess_SOURCES += \
-       src/logind-acl.c
+       src/logind-acl.c \
+        src/acl-util.c
 endif
 
 systemd_uaccess_CFLAGS = \
@@ -1158,7 +1219,8 @@ systemd_tmpfiles_LDADD = \
 
 systemd_machine_id_setup_SOURCES = \
        src/machine-id-setup.c \
-       src/machine-id-main.c
+       src/machine-id-main.c \
+        src/sd-id128.c
 
 systemd_machine_id_setup_CFLAGS = \
        $(AM_CFLAGS)
diff --git a/TODO b/TODO
index 5303497164f70d1402a732a2f552a4e7b1a1c1fc..8c3034e079c734fdc3a14824927065526980545a 100644 (file)
--- a/TODO
+++ b/TODO
@@ -17,8 +17,12 @@ Bugfixes:
 
 * make polkit checks async
 
+* properly handle .mount unit state tracking when two mount points are stacked one on top of another on the exact same mount point.
+
 Features:
 
+* check utf8 everywhere
+
 * when an instanced service exits, remove its parent cgroup too if possible.
 
 * as Tom Gundersen pointed out there's a always a dep loop if people use crypto file systems with random keys
index 251edd4c81bce9aca1c2610763d08028dda1a310..ee1215654d7c15a1d2ad0c59e1b934e0164bf90c 100644 (file)
@@ -1,3 +1,4 @@
 POTFILES
 Makefile.in.in
 .intltool-merge-cache
+Makefile
diff --git a/src/acl-util.c b/src/acl-util.c
new file mode 100644 (file)
index 0000000..a2a9f9a
--- /dev/null
@@ -0,0 +1,68 @@
+/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
+
+/***
+  This file is part of systemd.
+
+  Copyright 2011 Lennart Poettering
+
+  systemd is free software; you can redistribute it and/or modify it
+  under the terms of the GNU General Public License as published by
+  the Free Software Foundation; either version 2 of the License, or
+  (at your option) any later version.
+
+  systemd is distributed in the hope that it will be useful, but
+  WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+  General Public License for more details.
+
+  You should have received a copy of the GNU General Public License
+  along with systemd; If not, see <http://www.gnu.org/licenses/>.
+***/
+
+#include <assert.h>
+#include <sys/acl.h>
+#include <acl/libacl.h>
+#include <errno.h>
+#include <stdbool.h>
+
+#include "acl-util.h"
+
+int acl_find_uid(acl_t acl, uid_t uid, acl_entry_t *entry) {
+        acl_entry_t i;
+        int found;
+
+        assert(acl);
+        assert(entry);
+
+        for (found = acl_get_entry(acl, ACL_FIRST_ENTRY, &i);
+             found > 0;
+             found = acl_get_entry(acl, ACL_NEXT_ENTRY, &i)) {
+
+                acl_tag_t tag;
+                uid_t *u;
+                bool b;
+
+                if (acl_get_tag_type(i, &tag) < 0)
+                        return -errno;
+
+                if (tag != ACL_USER)
+                        continue;
+
+                u = acl_get_qualifier(i);
+                if (!u)
+                        return -errno;
+
+                b = *u == uid;
+                acl_free(u);
+
+                if (b) {
+                        *entry = i;
+                        return 1;
+                }
+        }
+
+        if (found < 0)
+                return -errno;
+
+        return 0;
+}
diff --git a/src/acl-util.h b/src/acl-util.h
new file mode 100644 (file)
index 0000000..798ce43
--- /dev/null
@@ -0,0 +1,27 @@
+/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
+
+#ifndef fooaclutilhfoo
+#define fooaclutilhfoo
+
+/***
+  This file is part of systemd.
+
+  Copyright 2011 Lennart Poettering
+
+  systemd is free software; you can redistribute it and/or modify it
+  under the terms of the GNU General Public License as published by
+  the Free Software Foundation; either version 2 of the License, or
+  (at your option) any later version.
+
+  systemd is distributed in the hope that it will be useful, but
+  WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+  General Public License for more details.
+
+  You should have received a copy of the GNU General Public License
+  along with systemd; If not, see <http://www.gnu.org/licenses/>.
+***/
+
+int acl_find_uid(acl_t acl, uid_t uid, acl_entry_t *entry);
+
+#endif
index 29b807de515a1fbc76dae742e23f9c126f122fe7..6cf3c311b3995e3c0114060d9c067fb09399c1be 100644 (file)
@@ -770,7 +770,7 @@ static void automount_fd_event(Unit *u, int fd, uint32_t events, Watch *w) {
                 if (packet.v5_packet.pid > 0) {
                         char *p = NULL;
 
-                        get_process_name(packet.v5_packet.pid, &p);
+                        get_process_comm(packet.v5_packet.pid, &p);
                         log_debug("Got direct mount request for %s, triggered by %lu (%s)", packet.v5_packet.name, (unsigned long) packet.v5_packet.pid, strna(p));
                         free(p);
 
index bc9c2163291d128504b0624564ffc6443cf534be..03c942c66e23bd318b47bfbd9727cd7781539ea4 100644 (file)
@@ -133,7 +133,7 @@ static int show_cgroup_one_by_path(const char *path, const char *prefix, unsigne
                 for (i = 0; i < n; i++) {
                         char *t = NULL;
 
-                        get_process_cmdline(pids[i], n_columns, &t);
+                        get_process_cmdline(pids[i], n_columns, true, &t);
 
                         printf("%s%s %*lu %s\n",
                                prefix,
diff --git a/src/journal/Makefile b/src/journal/Makefile
new file mode 120000 (symlink)
index 0000000..d0b0e8e
--- /dev/null
@@ -0,0 +1 @@
+../Makefile
\ No newline at end of file
diff --git a/src/journal/journal-def.h b/src/journal/journal-def.h
new file mode 100644 (file)
index 0000000..1a63ca1
--- /dev/null
@@ -0,0 +1,156 @@
+/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
+
+#ifndef foojournaldefhfoo
+#define foojournaldefhfoo
+
+/***
+  This file is part of systemd.
+
+  Copyright 2011 Lennart Poettering
+
+  systemd is free software; you can redistribute it and/or modify it
+  under the terms of the GNU General Public License as published by
+  the Free Software Foundation; either version 2 of the License, or
+  (at your option) any later version.
+
+  systemd is distributed in the hope that it will be useful, but
+  WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+  General Public License for more details.
+
+  You should have received a copy of the GNU General Public License
+  along with systemd; If not, see <http://www.gnu.org/licenses/>.
+***/
+
+#include <inttypes.h>
+
+#include "macro.h"
+#include "sd-id128.h"
+
+typedef struct Header Header;
+typedef struct ObjectHeader ObjectHeader;
+typedef union Object Object;
+typedef struct DataObject DataObject;
+typedef struct FieldObject FieldObject;
+typedef struct EntryObject EntryObject;
+typedef struct HashTableObject HashTableObject;
+typedef struct EntryArrayObject EntryArrayObject;
+typedef struct EntryItem EntryItem;
+typedef struct HashItem HashItem;
+
+/* Object types */
+enum {
+        OBJECT_UNUSED,
+        OBJECT_DATA,
+        OBJECT_FIELD,
+        OBJECT_ENTRY,
+        OBJECT_DATA_HASH_TABLE,
+        OBJECT_FIELD_HASH_TABLE,
+        OBJECT_ENTRY_ARRAY,
+        _OBJECT_TYPE_MAX
+};
+
+_packed_ struct ObjectHeader {
+        uint8_t type;
+        uint8_t reserved[7];
+        uint64_t size;
+        uint8_t payload[];
+};
+
+_packed_ struct DataObject {
+        ObjectHeader object;
+        uint64_t hash;
+        uint64_t next_hash_offset;
+        uint64_t next_field_offset;
+        uint64_t entry_offset; /* the first array entry we store inline */
+        uint64_t entry_array_offset;
+        uint64_t n_entries;
+        uint8_t payload[];
+};
+
+_packed_ struct FieldObject {
+        ObjectHeader object;
+        uint64_t hash;
+        uint64_t next_hash_offset;
+        uint64_t head_data_offset;
+        uint64_t tail_data_offset;
+        uint8_t payload[];
+};
+
+_packed_ struct EntryItem {
+        uint64_t object_offset;
+        uint64_t hash;
+};
+
+_packed_ struct EntryObject {
+        ObjectHeader object;
+        uint64_t seqnum;
+        uint64_t realtime;
+        uint64_t monotonic;
+        sd_id128_t boot_id;
+        uint64_t xor_hash;
+        EntryItem items[];
+};
+
+_packed_ struct HashItem {
+        uint64_t head_hash_offset;
+        uint64_t tail_hash_offset;
+};
+
+_packed_ struct HashTableObject {
+        ObjectHeader object;
+        HashItem items[];
+};
+
+_packed_ struct EntryArrayObject {
+        ObjectHeader object;
+        uint64_t next_entry_array_offset;
+        uint64_t items[];
+};
+
+union Object {
+        ObjectHeader object;
+        DataObject data;
+        FieldObject field;
+        EntryObject entry;
+        HashTableObject hash_table;
+        EntryArrayObject entry_array;
+};
+
+enum {
+        STATE_OFFLINE,
+        STATE_ONLINE,
+        STATE_ARCHIVED
+};
+
+_packed_ struct Header {
+        uint8_t signature[8]; /* "LPKSHHRH" */
+        uint32_t compatible_flags;
+        uint32_t incompatible_flags;
+        uint8_t state;
+        uint8_t reserved[7];
+        sd_id128_t file_id;
+        sd_id128_t machine_id;
+        sd_id128_t boot_id;
+        sd_id128_t seqnum_id;
+        uint64_t arena_offset;
+        uint64_t arena_size;
+        uint64_t arena_max_size;  /* obsolete */
+        uint64_t arena_min_size;  /* obsolete */
+        uint64_t arena_keep_free; /* obsolete */
+        uint64_t data_hash_table_offset;     /* for looking up data objects */
+        uint64_t data_hash_table_size;
+        uint64_t field_hash_table_offset;     /* for looking up field objects */
+        uint64_t field_hash_table_size;
+        uint64_t tail_object_offset;
+        uint64_t n_objects;
+        uint64_t n_entries;
+        uint64_t seqnum;
+        uint64_t first_seqnum;
+        uint64_t entry_array_offset;
+        uint64_t head_entry_realtime;
+        uint64_t tail_entry_realtime;
+        uint64_t tail_entry_monotonic;
+};
+
+#endif
diff --git a/src/journal/journal-file.c b/src/journal/journal-file.c
new file mode 100644 (file)
index 0000000..427631d
--- /dev/null
@@ -0,0 +1,1903 @@
+/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
+
+/***
+  This file is part of systemd.
+
+  Copyright 2011 Lennart Poettering
+
+  systemd is free software; you can redistribute it and/or modify it
+  under the terms of the GNU General Public License as published by
+  the Free Software Foundation; either version 2 of the License, or
+  (at your option) any later version.
+
+  systemd is distributed in the hope that it will be useful, but
+  WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+  General Public License for more details.
+
+  You should have received a copy of the GNU General Public License
+  along with systemd; If not, see <http://www.gnu.org/licenses/>.
+***/
+
+#include <sys/mman.h>
+#include <errno.h>
+#include <sys/uio.h>
+#include <unistd.h>
+#include <sys/statvfs.h>
+#include <fcntl.h>
+#include <stddef.h>
+
+#include "journal-def.h"
+#include "journal-file.h"
+#include "lookup3.h"
+
+#define DEFAULT_ARENA_MAX_SIZE (16ULL*1024ULL*1024ULL*1024ULL)
+#define DEFAULT_ARENA_MIN_SIZE (256ULL*1024ULL)
+#define DEFAULT_ARENA_KEEP_FREE (1ULL*1024ULL*1024ULL)
+
+#define DEFAULT_MAX_USE (16ULL*1024ULL*1024ULL*16ULL)
+
+#define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*16ULL)
+#define DEFAULT_FIELD_HASH_TABLE_SIZE (2047ULL*16ULL)
+
+#define DEFAULT_WINDOW_SIZE (128ULL*1024ULL*1024ULL)
+
+static const char signature[] = { 'L', 'P', 'K', 'S', 'H', 'H', 'R', 'H' };
+
+#define ALIGN64(x) (((x) + 7ULL) & ~7ULL)
+
+void journal_file_close(JournalFile *f) {
+        int t;
+
+        assert(f);
+
+        if (f->header && f->writable)
+                f->header->state = STATE_OFFLINE;
+
+
+        for (t = 0; t < _WINDOW_MAX; t++)
+                if (f->windows[t].ptr)
+                        munmap(f->windows[t].ptr, f->windows[t].size);
+
+        if (f->fd >= 0)
+                close_nointr_nofail(f->fd);
+
+        free(f->path);
+        free(f);
+}
+
+static int journal_file_init_header(JournalFile *f, JournalFile *template) {
+        Header h;
+        ssize_t k;
+        int r;
+
+        assert(f);
+
+        zero(h);
+        memcpy(h.signature, signature, 8);
+        h.arena_offset = htole64(ALIGN64(sizeof(h)));
+        h.arena_max_size = htole64(DEFAULT_ARENA_MAX_SIZE);
+        h.arena_min_size = htole64(DEFAULT_ARENA_MIN_SIZE);
+        h.arena_keep_free = htole64(DEFAULT_ARENA_KEEP_FREE);
+
+        r = sd_id128_randomize(&h.file_id);
+        if (r < 0)
+                return r;
+
+        if (template) {
+                h.seqnum_id = template->header->seqnum_id;
+                h.seqnum = template->header->seqnum;
+        } else
+                h.seqnum_id = h.file_id;
+
+        k = pwrite(f->fd, &h, sizeof(h), 0);
+        if (k < 0)
+                return -errno;
+
+        if (k != sizeof(h))
+                return -EIO;
+
+        return 0;
+}
+
+static int journal_file_refresh_header(JournalFile *f) {
+        int r;
+        sd_id128_t boot_id;
+
+        assert(f);
+
+        r = sd_id128_get_machine(&f->header->machine_id);
+        if (r < 0)
+                return r;
+
+        r = sd_id128_get_boot(&boot_id);
+        if (r < 0)
+                return r;
+
+        if (sd_id128_equal(boot_id, f->header->boot_id))
+                f->tail_entry_monotonic_valid = true;
+
+        f->header->boot_id = boot_id;
+
+        f->header->state = STATE_ONLINE;
+        return 0;
+}
+
+static int journal_file_verify_header(JournalFile *f) {
+        assert(f);
+
+        if (memcmp(f->header, signature, 8))
+                return -EBADMSG;
+
+        if (f->header->incompatible_flags != 0)
+                return -EPROTONOSUPPORT;
+
+        if ((uint64_t) f->last_stat.st_size < (le64toh(f->header->arena_offset) + le64toh(f->header->arena_size)))
+                return -ENODATA;
+
+        if (f->writable) {
+                uint32_t state;
+                sd_id128_t machine_id;
+                int r;
+
+                r = sd_id128_get_machine(&machine_id);
+                if (r < 0)
+                        return r;
+
+                if (!sd_id128_equal(machine_id, f->header->machine_id))
+                        return -EHOSTDOWN;
+
+                state = f->header->state;
+
+                if (state == STATE_ONLINE)
+                        log_debug("Journal file %s is already online. Assuming unclean closing. Ignoring.", f->path);
+                else if (state == STATE_ARCHIVED)
+                        return -ESHUTDOWN;
+                else if (state != STATE_OFFLINE)
+                        log_debug("Journal file %s has unknown state %u. Ignoring.", f->path, state);
+        }
+
+        return 0;
+}
+
+static int journal_file_allocate(JournalFile *f, uint64_t offset, uint64_t size) {
+        uint64_t asize;
+        uint64_t old_size, new_size;
+
+        assert(f);
+
+        if (offset < le64toh(f->header->arena_offset))
+                return -EINVAL;
+
+        new_size = PAGE_ALIGN(offset + size);
+
+        /* We assume that this file is not sparse, and we know that
+         * for sure, since we always call posix_fallocate()
+         * ourselves */
+
+        old_size =
+                le64toh(f->header->arena_offset) +
+                le64toh(f->header->arena_size);
+
+        if (old_size >= new_size)
+                return 0;
+
+        asize = new_size - le64toh(f->header->arena_offset);
+
+        if (asize > le64toh(f->header->arena_min_size)) {
+                struct statvfs svfs;
+
+                if (fstatvfs(f->fd, &svfs) >= 0) {
+                        uint64_t available;
+
+                        available = svfs.f_bfree * svfs.f_bsize;
+
+                        if (available >= f->header->arena_keep_free)
+                                available -= f->header->arena_keep_free;
+                        else
+                                available = 0;
+
+                        if (new_size - old_size > available)
+                                return -E2BIG;
+                }
+        }
+
+        if (asize > le64toh(f->header->arena_max_size))
+                return -E2BIG;
+
+        if (posix_fallocate(f->fd, old_size, new_size - old_size) < 0)
+                return -errno;
+
+        if (fstat(f->fd, &f->last_stat) < 0)
+                return -errno;
+
+        f->header->arena_size = htole64(asize);
+
+        return 0;
+}
+
+static int journal_file_map(
+                JournalFile *f,
+                uint64_t offset,
+                uint64_t size,
+                void **_window,
+                uint64_t *_woffset,
+                uint64_t *_wsize,
+                void **ret) {
+
+        uint64_t woffset, wsize;
+        void *window;
+
+        assert(f);
+        assert(size > 0);
+        assert(ret);
+
+        woffset = offset & ~((uint64_t) page_size() - 1ULL);
+        wsize = size + (offset - woffset);
+        wsize = PAGE_ALIGN(wsize);
+
+        window = mmap(NULL, wsize, f->prot, MAP_SHARED, f->fd, woffset);
+        if (window == MAP_FAILED)
+                return -errno;
+
+        if (_window)
+                *_window = window;
+
+        if (_woffset)
+                *_woffset = woffset;
+
+        if (_wsize)
+                *_wsize = wsize;
+
+        *ret = (uint8_t*) window + (offset - woffset);
+
+        return 0;
+}
+
+static int journal_file_move_to(JournalFile *f, int wt, uint64_t offset, uint64_t size, void **ret) {
+        void *p;
+        uint64_t delta;
+        int r;
+        Window *w;
+
+        assert(f);
+        assert(ret);
+        assert(wt >= 0);
+        assert(wt < _WINDOW_MAX);
+
+        w = f->windows + wt;
+
+        if (_likely_(w->ptr &&
+                     w->offset <= offset &&
+                     w->offset + w->size >= offset + size)) {
+
+                *ret = (uint8_t*) w->ptr + (offset - w->offset);
+                return 0;
+        }
+
+        if (w->ptr) {
+                if (munmap(w->ptr, w->size) < 0)
+                        return -errno;
+
+                w->ptr = NULL;
+                w->size = w->offset = 0;
+        }
+
+        if (size < DEFAULT_WINDOW_SIZE) {
+                /* If the default window size is larger then what was
+                 * asked for extend the mapping a bit in the hope to
+                 * minimize needed remappings later on. We add half
+                 * the window space before and half behind the
+                 * requested mapping */
+
+                delta = PAGE_ALIGN((DEFAULT_WINDOW_SIZE - size) / 2);
+
+                if (offset < delta)
+                        delta = offset;
+
+                offset -= delta;
+                size += (DEFAULT_WINDOW_SIZE - delta);
+        } else
+                delta = 0;
+
+        r = journal_file_map(f,
+                             offset, size,
+                             &w->ptr, &w->offset, &w->size,
+                             &p);
+
+        if (r < 0)
+                return r;
+
+        *ret = (uint8_t*) p + delta;
+        return 0;
+}
+
+static bool verify_hash(Object *o) {
+        uint64_t h1, h2;
+
+        assert(o);
+
+        if (o->object.type == OBJECT_DATA) {
+                h1 = le64toh(o->data.hash);
+                h2 = hash64(o->data.payload, le64toh(o->object.size) - offsetof(Object, data.payload));
+        } else if (o->object.type == OBJECT_FIELD) {
+                h1 = le64toh(o->field.hash);
+                h2 = hash64(o->field.payload, le64toh(o->object.size) - offsetof(Object, field.payload));
+        } else
+                return true;
+
+        return h1 == h2;
+}
+
+int journal_file_move_to_object(JournalFile *f, int type, uint64_t offset, Object **ret) {
+        int r;
+        void *t;
+        Object *o;
+        uint64_t s;
+
+        assert(f);
+        assert(ret);
+        assert(type < _OBJECT_TYPE_MAX);
+
+        r = journal_file_move_to(f, type >= 0 ? type : WINDOW_UNKNOWN, offset, sizeof(ObjectHeader), &t);
+        if (r < 0)
+                return r;
+
+        o = (Object*) t;
+        s = le64toh(o->object.size);
+
+        if (s < sizeof(ObjectHeader))
+                return -EBADMSG;
+
+        if (type >= 0 && o->object.type != type)
+                return -EBADMSG;
+
+        if (s > sizeof(ObjectHeader)) {
+                r = journal_file_move_to(f, o->object.type, offset, s, &t);
+                if (r < 0)
+                        return r;
+
+                o = (Object*) t;
+        }
+
+        if (!verify_hash(o))
+                return -EBADMSG;
+
+        *ret = o;
+        return 0;
+}
+
+static uint64_t journal_file_seqnum(JournalFile *f, uint64_t *seqnum) {
+        uint64_t r;
+
+        assert(f);
+
+        r = le64toh(f->header->seqnum) + 1;
+
+        if (seqnum) {
+                /* If an external seqnum counter was passed, we update
+                 * both the local and the external one, and set it to
+                 * the maximum of both */
+
+                if (*seqnum + 1 > r)
+                        r = *seqnum + 1;
+
+                *seqnum = r;
+        }
+
+        f->header->seqnum = htole64(r);
+
+        if (f->header->first_seqnum == 0)
+                f->header->first_seqnum = htole64(r);
+
+        return r;
+}
+
+static int journal_file_append_object(JournalFile *f, int type, uint64_t size, Object **ret, uint64_t *offset) {
+        int r;
+        uint64_t p;
+        Object *tail, *o;
+        void *t;
+
+        assert(f);
+        assert(size >= sizeof(ObjectHeader));
+        assert(offset);
+        assert(ret);
+
+        p = le64toh(f->header->tail_object_offset);
+        if (p == 0)
+                p = le64toh(f->header->arena_offset);
+        else {
+                r = journal_file_move_to_object(f, -1, p, &tail);
+                if (r < 0)
+                        return r;
+
+                p += ALIGN64(le64toh(tail->object.size));
+        }
+
+        r = journal_file_allocate(f, p, size);
+        if (r < 0)
+                return r;
+
+        r = journal_file_move_to(f, type, p, size, &t);
+        if (r < 0)
+                return r;
+
+        o = (Object*) t;
+
+        zero(o->object);
+        o->object.type = type;
+        o->object.size = htole64(size);
+
+        f->header->tail_object_offset = htole64(p);
+        f->header->n_objects = htole64(le64toh(f->header->n_objects) + 1);
+
+        *ret = o;
+        *offset = p;
+
+        return 0;
+}
+
+static int journal_file_setup_data_hash_table(JournalFile *f) {
+        uint64_t s, p;
+        Object *o;
+        int r;
+
+        assert(f);
+
+        s = DEFAULT_DATA_HASH_TABLE_SIZE;
+        r = journal_file_append_object(f,
+                                       OBJECT_DATA_HASH_TABLE,
+                                       offsetof(Object, hash_table.items) + s,
+                                       &o, &p);
+        if (r < 0)
+                return r;
+
+        memset(o->hash_table.items, 0, s);
+
+        f->header->data_hash_table_offset = htole64(p + offsetof(Object, hash_table.items));
+        f->header->data_hash_table_size = htole64(s);
+
+        return 0;
+}
+
+static int journal_file_setup_field_hash_table(JournalFile *f) {
+        uint64_t s, p;
+        Object *o;
+        int r;
+
+        assert(f);
+
+        s = DEFAULT_FIELD_HASH_TABLE_SIZE;
+        r = journal_file_append_object(f,
+                                       OBJECT_FIELD_HASH_TABLE,
+                                       offsetof(Object, hash_table.items) + s,
+                                       &o, &p);
+        if (r < 0)
+                return r;
+
+        memset(o->hash_table.items, 0, s);
+
+        f->header->field_hash_table_offset = htole64(p + offsetof(Object, hash_table.items));
+        f->header->field_hash_table_size = htole64(s);
+
+        return 0;
+}
+
+static int journal_file_map_data_hash_table(JournalFile *f) {
+        uint64_t s, p;
+        void *t;
+        int r;
+
+        assert(f);
+
+        p = le64toh(f->header->data_hash_table_offset);
+        s = le64toh(f->header->data_hash_table_size);
+
+        r = journal_file_move_to(f,
+                                 WINDOW_DATA_HASH_TABLE,
+                                 p, s,
+                                 &t);
+        if (r < 0)
+                return r;
+
+        f->data_hash_table = t;
+        return 0;
+}
+
+static int journal_file_map_field_hash_table(JournalFile *f) {
+        uint64_t s, p;
+        void *t;
+        int r;
+
+        assert(f);
+
+        p = le64toh(f->header->field_hash_table_offset);
+        s = le64toh(f->header->field_hash_table_size);
+
+        r = journal_file_move_to(f,
+                                 WINDOW_FIELD_HASH_TABLE,
+                                 p, s,
+                                 &t);
+        if (r < 0)
+                return r;
+
+        f->field_hash_table = t;
+        return 0;
+}
+
+static int journal_file_link_data(JournalFile *f, Object *o, uint64_t offset, uint64_t hash) {
+        uint64_t p, h;
+        int r;
+
+        assert(f);
+        assert(o);
+        assert(offset > 0);
+        assert(o->object.type == OBJECT_DATA);
+
+        o->data.next_hash_offset = o->data.next_field_offset = 0;
+        o->data.entry_offset = o->data.entry_array_offset = 0;
+        o->data.n_entries = 0;
+
+        h = hash % (le64toh(f->header->data_hash_table_size) / sizeof(HashItem));
+        p = le64toh(f->data_hash_table[h].head_hash_offset);
+        if (p == 0) {
+                /* Only entry in the hash table is easy */
+                f->data_hash_table[h].head_hash_offset = htole64(offset);
+        } else {
+                /* Temporarily move back to the previous data object,
+                 * to patch in pointer */
+
+                r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
+                if (r < 0)
+                        return r;
+
+                o->data.next_hash_offset = htole64(offset);
+
+                r = journal_file_move_to_object(f, OBJECT_DATA, offset, &o);
+                if (r < 0)
+                        return r;
+        }
+
+        f->data_hash_table[h].tail_hash_offset = htole64(offset);
+
+        return 0;
+}
+
+int journal_file_find_data_object_with_hash(
+                JournalFile *f,
+                const void *data, uint64_t size, uint64_t hash,
+                Object **ret, uint64_t *offset) {
+        uint64_t p, osize, h;
+        int r;
+
+        assert(f);
+        assert(data || size == 0);
+
+        osize = offsetof(Object, data.payload) + size;
+
+        h = hash % (le64toh(f->header->data_hash_table_size) / sizeof(HashItem));
+        p = le64toh(f->data_hash_table[h].head_hash_offset);
+
+        while (p > 0) {
+                Object *o;
+
+                r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
+                if (r < 0)
+                        return r;
+
+                if (le64toh(o->object.size) == osize &&
+                    memcmp(o->data.payload, data, size) == 0) {
+
+                        if (le64toh(o->data.hash) != hash)
+                                return -EBADMSG;
+
+                        if (ret)
+                                *ret = o;
+
+                        if (offset)
+                                *offset = p;
+
+                        return 1;
+                }
+
+                p = le64toh(o->data.next_hash_offset);
+        }
+
+        return 0;
+}
+
+int journal_file_find_data_object(
+                JournalFile *f,
+                const void *data, uint64_t size,
+                Object **ret, uint64_t *offset) {
+
+        uint64_t hash;
+
+        assert(f);
+        assert(data || size == 0);
+
+        hash = hash64(data, size);
+
+        return journal_file_find_data_object_with_hash(f,
+                                                       data, size, hash,
+                                                       ret, offset);
+}
+
+static int journal_file_append_data(JournalFile *f, const void *data, uint64_t size, Object **ret, uint64_t *offset) {
+        uint64_t hash, p;
+        uint64_t osize;
+        Object *o;
+        int r;
+
+        assert(f);
+        assert(data || size == 0);
+
+        hash = hash64(data, size);
+
+        r = journal_file_find_data_object_with_hash(f, data, size, hash, &o, &p);
+        if (r < 0)
+                return r;
+        else if (r > 0) {
+
+                if (ret)
+                        *ret = o;
+
+                if (offset)
+                        *offset = p;
+
+                return 0;
+        }
+
+        osize = offsetof(Object, data.payload) + size;
+        r = journal_file_append_object(f, OBJECT_DATA, osize, &o, &p);
+        if (r < 0)
+                return r;
+
+        o->data.hash = htole64(hash);
+        memcpy(o->data.payload, data, size);
+
+        r = journal_file_link_data(f, o, p, hash);
+        if (r < 0)
+                return r;
+
+        if (ret)
+                *ret = o;
+
+        if (offset)
+                *offset = p;
+
+        return 0;
+}
+
+uint64_t journal_file_entry_n_items(Object *o) {
+        assert(o);
+        assert(o->object.type == htole64(OBJECT_ENTRY));
+
+        return (le64toh(o->object.size) - offsetof(Object, entry.items)) / sizeof(EntryItem);
+}
+
+static uint64_t journal_file_entry_array_n_items(Object *o) {
+        assert(o);
+        assert(o->object.type == htole64(OBJECT_ENTRY_ARRAY));
+
+        return (le64toh(o->object.size) - offsetof(Object, entry_array.items)) / sizeof(uint64_t);
+}
+
+static int link_entry_into_array(JournalFile *f,
+                                 uint64_t *first,
+                                 uint64_t *idx,
+                                 uint64_t p) {
+        int r;
+        uint64_t n = 0, ap = 0, q, i, a, hidx;
+        Object *o;
+
+        assert(f);
+        assert(first);
+        assert(idx);
+        assert(p > 0);
+
+        a = le64toh(*first);
+        i = hidx = le64toh(*idx);
+        while (a > 0) {
+
+                r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
+                if (r < 0)
+                        return r;
+
+                n = journal_file_entry_array_n_items(o);
+                if (i < n) {
+                        o->entry_array.items[i] = htole64(p);
+                        *idx = htole64(hidx + 1);
+                        return 0;
+                }
+
+                i -= n;
+                ap = a;
+                a = le64toh(o->entry_array.next_entry_array_offset);
+        }
+
+        if (hidx > n)
+                n = (hidx+1) * 2;
+        else
+                n = n * 2;
+
+        if (n < 4)
+                n = 4;
+
+        r = journal_file_append_object(f, OBJECT_ENTRY_ARRAY,
+                                       offsetof(Object, entry_array.items) + n * sizeof(uint64_t),
+                                       &o, &q);
+        if (r < 0)
+                return r;
+
+        o->entry_array.items[i] = htole64(p);
+
+        if (ap == 0)
+                *first = q;
+        else {
+                r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, ap, &o);
+                if (r < 0)
+                        return r;
+
+                o->entry_array.next_entry_array_offset = htole64(q);
+        }
+
+        *idx = htole64(hidx + 1);
+
+        return 0;
+}
+
+static int link_entry_into_array_plus_one(JournalFile *f,
+                                          uint64_t *extra,
+                                          uint64_t *first,
+                                          uint64_t *idx,
+                                          uint64_t p) {
+
+        int r;
+
+        assert(f);
+        assert(extra);
+        assert(first);
+        assert(idx);
+        assert(p > 0);
+
+        if (*idx == 0)
+                *extra = htole64(p);
+        else {
+                uint64_t i;
+
+                i = le64toh(*idx) - 1;
+                r = link_entry_into_array(f, first, &i, p);
+                if (r < 0)
+                        return r;
+        }
+
+        *idx = htole64(le64toh(*idx) + 1);
+        return 0;
+}
+
+static int journal_file_link_entry_item(JournalFile *f, Object *o, uint64_t offset, uint64_t i) {
+        uint64_t p;
+        int r;
+        assert(f);
+        assert(o);
+        assert(offset > 0);
+
+        p = le64toh(o->entry.items[i].object_offset);
+        if (p == 0)
+                return -EINVAL;
+
+        r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
+        if (r < 0)
+                return r;
+
+        return link_entry_into_array_plus_one(f,
+                                              &o->data.entry_offset,
+                                              &o->data.entry_array_offset,
+                                              &o->data.n_entries,
+                                              offset);
+}
+
+static int journal_file_link_entry(JournalFile *f, Object *o, uint64_t offset) {
+        uint64_t n, i;
+        int r;
+
+        assert(f);
+        assert(o);
+        assert(offset > 0);
+        assert(o->object.type == OBJECT_ENTRY);
+
+        /* Link up the entry itself */
+        r = link_entry_into_array(f,
+                                  &f->header->entry_array_offset,
+                                  &f->header->n_entries,
+                                  offset);
+        if (r < 0)
+                return r;
+
+        log_error("%s %lu", f->path, (unsigned long) f->header->n_entries);
+
+        if (f->header->head_entry_realtime == 0)
+                f->header->head_entry_realtime = o->entry.realtime;
+
+        f->header->tail_entry_realtime = o->entry.realtime;
+        f->header->tail_entry_monotonic = o->entry.monotonic;
+
+        f->tail_entry_monotonic_valid = true;
+
+        /* Link up the items */
+        n = journal_file_entry_n_items(o);
+        for (i = 0; i < n; i++) {
+                r = journal_file_link_entry_item(f, o, offset, i);
+                if (r < 0)
+                        return r;
+        }
+
+        return 0;
+}
+
+static int journal_file_append_entry_internal(
+                JournalFile *f,
+                const dual_timestamp *ts,
+                uint64_t xor_hash,
+                const EntryItem items[], unsigned n_items,
+                uint64_t *seqnum,
+                Object **ret, uint64_t *offset) {
+        uint64_t np;
+        uint64_t osize;
+        Object *o;
+        int r;
+
+        assert(f);
+        assert(items || n_items == 0);
+        assert(ts);
+
+        osize = offsetof(Object, entry.items) + (n_items * sizeof(EntryItem));
+
+        r = journal_file_append_object(f, OBJECT_ENTRY, osize, &o, &np);
+        if (r < 0)
+                return r;
+
+        o->entry.seqnum = htole64(journal_file_seqnum(f, seqnum));
+        memcpy(o->entry.items, items, n_items * sizeof(EntryItem));
+        o->entry.realtime = htole64(ts->realtime);
+        o->entry.monotonic = htole64(ts->monotonic);
+        o->entry.xor_hash = htole64(xor_hash);
+        o->entry.boot_id = f->header->boot_id;
+
+        r = journal_file_link_entry(f, o, np);
+        if (r < 0)
+                return r;
+
+        if (ret)
+                *ret = o;
+
+        if (offset)
+                *offset = np;
+
+        return 0;
+}
+
+int journal_file_append_entry(JournalFile *f, const dual_timestamp *ts, const struct iovec iovec[], unsigned n_iovec, uint64_t *seqnum, Object **ret, uint64_t *offset) {
+        unsigned i;
+        EntryItem *items;
+        int r;
+        uint64_t xor_hash = 0;
+        struct dual_timestamp _ts;
+
+        assert(f);
+        assert(iovec || n_iovec == 0);
+
+        if (!f->writable)
+                return -EPERM;
+
+        if (!ts) {
+                dual_timestamp_get(&_ts);
+                ts = &_ts;
+        }
+
+        if (f->tail_entry_monotonic_valid &&
+            ts->monotonic < le64toh(f->header->tail_entry_monotonic))
+                return -EINVAL;
+
+        if (ts->realtime < le64toh(f->header->tail_entry_realtime))
+                return -EINVAL;
+
+        items = new(EntryItem, n_iovec);
+        if (!items)
+                return -ENOMEM;
+
+        for (i = 0; i < n_iovec; i++) {
+                uint64_t p;
+                Object *o;
+
+                r = journal_file_append_data(f, iovec[i].iov_base, iovec[i].iov_len, &o, &p);
+                if (r < 0)
+                        goto finish;
+
+                xor_hash ^= le64toh(o->data.hash);
+                items[i].object_offset = htole64(p);
+                items[i].hash = o->data.hash;
+        }
+
+        r = journal_file_append_entry_internal(f, ts, xor_hash, items, n_iovec, seqnum, ret, offset);
+
+finish:
+        free(items);
+
+        return r;
+}
+
+static int generic_array_get(JournalFile *f,
+                             uint64_t first,
+                             uint64_t i,
+                             Object **ret, uint64_t *offset) {
+
+        Object *o;
+        uint64_t p, a;
+        int r;
+
+        assert(f);
+
+        a = first;
+        while (a > 0) {
+                uint64_t n;
+
+                r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
+                if (r < 0)
+                        return r;
+
+                n = journal_file_entry_array_n_items(o);
+                if (i < n) {
+                        p = le64toh(o->entry_array.items[i]);
+                        break;
+                }
+
+                i -= n;
+                a = le64toh(o->entry_array.next_entry_array_offset);
+        }
+
+        if (a <= 0 || p <= 0)
+                return 0;
+
+        r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
+        if (r < 0)
+                return r;
+
+        if (ret)
+                *ret = o;
+
+        if (offset)
+                *offset = p;
+
+        return 1;
+}
+
+static int generic_array_get_plus_one(JournalFile *f,
+                                      uint64_t extra,
+                                      uint64_t first,
+                                      uint64_t i,
+                                      Object **ret, uint64_t *offset) {
+
+        Object *o;
+
+        assert(f);
+
+        if (i == 0) {
+                int r;
+
+                r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, &o);
+                if (r < 0)
+                        return r;
+
+                if (ret)
+                        *ret = o;
+
+                if (offset)
+                        *offset = extra;
+
+                return 1;
+        }
+
+        return generic_array_get(f, first, i-1, ret, offset);
+}
+
+enum {
+        TEST_FOUND,
+        TEST_LEFT,
+        TEST_RIGHT
+};
+
+static int generic_array_bisect(JournalFile *f,
+                                uint64_t first,
+                                uint64_t n,
+                                uint64_t needle,
+                                int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle),
+                                direction_t direction,
+                                Object **ret,
+                                uint64_t *offset,
+                                uint64_t *idx) {
+
+        uint64_t a, p, t = 0, i = 0, last_p = 0;
+        bool subtract_one = false;
+        Object *o, *array = NULL;
+        int r;
+
+        assert(f);
+        assert(test_object);
+
+        a = first;
+        while (a > 0) {
+                uint64_t left, right, k, lp;
+
+                r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &array);
+                if (r < 0)
+                        return r;
+
+                k = journal_file_entry_array_n_items(array);
+                right = MIN(k, n);
+                if (right <= 0)
+                        return 0;
+
+                i = right - 1;
+                lp = p = le64toh(array->entry_array.items[i]);
+                if (p <= 0)
+                        return -EBADMSG;
+
+                r = test_object(f, p, needle);
+                if (r < 0)
+                        return r;
+
+                if (r == TEST_FOUND)
+                        r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
+
+                if (r == TEST_RIGHT) {
+                        left = 0;
+                        right -= 1;
+                        for (;;) {
+                                if (left == right) {
+                                        if (direction == DIRECTION_UP)
+                                                subtract_one = true;
+
+                                        i = left;
+                                        goto found;
+                                }
+
+                                assert(left < right);
+
+                                i = (left + right) / 2;
+                                p = le64toh(array->entry_array.items[i]);
+                                if (p <= 0)
+                                        return -EBADMSG;
+
+                                r = test_object(f, p, needle);
+                                if (r < 0)
+                                        return r;
+
+                                if (r == TEST_FOUND)
+                                        r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
+
+                                if (r == TEST_RIGHT)
+                                        right = i;
+                                else
+                                        left = i + 1;
+                        }
+                }
+
+                if (k > n)
+                        return 0;
+
+                last_p = lp;
+
+                n -= k;
+                t += k;
+                a = le64toh(array->entry_array.next_entry_array_offset);
+        }
+
+        return 0;
+
+found:
+        if (subtract_one && t == 0 && i == 0)
+                return 0;
+
+        if (subtract_one && i == 0)
+                p = last_p;
+        else if (subtract_one)
+                p = le64toh(array->entry_array.items[i-1]);
+        else
+                p = le64toh(array->entry_array.items[i]);
+
+        r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
+        if (r < 0)
+                return r;
+
+        if (ret)
+                *ret = o;
+
+        if (offset)
+                *offset = p;
+
+        if (idx)
+                *idx = t + i - (subtract_one ? 1 : 0);
+
+        return 1;
+}
+
+static int generic_array_bisect_plus_one(JournalFile *f,
+                                         uint64_t extra,
+                                         uint64_t first,
+                                         uint64_t n,
+                                         uint64_t needle,
+                                         int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle),
+                                         direction_t direction,
+                                         Object **ret,
+                                         uint64_t *offset,
+                                         uint64_t *idx) {
+
+        int r;
+
+        assert(f);
+        assert(test_object);
+
+        if (n <= 0)
+                return 0;
+
+        /* This bisects the array in object 'first', but first checks
+         * an extra  */
+
+        r = test_object(f, extra, needle);
+        if (r < 0)
+                return r;
+        else if (r == TEST_FOUND) {
+                Object *o;
+
+                r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, &o);
+                if (r < 0)
+                        return r;
+
+                if (ret)
+                        *ret = o;
+
+                if (offset)
+                        *offset = extra;
+        } else if (r == TEST_RIGHT)
+                return 0;
+
+        r = generic_array_bisect(f, first, n-1, needle, test_object, direction, ret, offset, idx);
+
+        if (r > 0)
+                (*idx) ++;
+
+        return r;
+}
+
+static int test_object_seqnum(JournalFile *f, uint64_t p, uint64_t needle) {
+        Object *o;
+        int r;
+
+        assert(f);
+        assert(p > 0);
+
+        r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
+        if (r < 0)
+                return r;
+
+        if (le64toh(o->entry.seqnum) == needle)
+                return TEST_FOUND;
+        else if (le64toh(o->entry.seqnum) < needle)
+                return TEST_LEFT;
+        else
+                return TEST_RIGHT;
+}
+
+int journal_file_move_to_entry_by_seqnum(
+                JournalFile *f,
+                uint64_t seqnum,
+                direction_t direction,
+                Object **ret,
+                uint64_t *offset) {
+
+        return generic_array_bisect(f,
+                                    le64toh(f->header->entry_array_offset),
+                                    le64toh(f->header->n_entries),
+                                    seqnum,
+                                    test_object_seqnum,
+                                    direction,
+                                    ret, offset, NULL);
+}
+
+static int test_object_realtime(JournalFile *f, uint64_t p, uint64_t needle) {
+        Object *o;
+        int r;
+
+        assert(f);
+        assert(p > 0);
+
+        r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
+        if (r < 0)
+                return r;
+
+        if (le64toh(o->entry.realtime) == needle)
+                return TEST_FOUND;
+        else if (le64toh(o->entry.realtime) < needle)
+                return TEST_LEFT;
+        else
+                return TEST_RIGHT;
+}
+
+int journal_file_move_to_entry_by_realtime(
+                JournalFile *f,
+                uint64_t realtime,
+                direction_t direction,
+                Object **ret,
+                uint64_t *offset) {
+
+        return generic_array_bisect(f,
+                                    le64toh(f->header->entry_array_offset),
+                                    le64toh(f->header->n_entries),
+                                    realtime,
+                                    test_object_realtime,
+                                    direction,
+                                    ret, offset, NULL);
+}
+
+static int test_object_monotonic(JournalFile *f, uint64_t p, uint64_t needle) {
+        Object *o;
+        int r;
+
+        assert(f);
+        assert(p > 0);
+
+        r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
+        if (r < 0)
+                return r;
+
+        if (le64toh(o->entry.monotonic) == needle)
+                return TEST_FOUND;
+        else if (le64toh(o->entry.monotonic) < needle)
+                return TEST_LEFT;
+        else
+                return TEST_RIGHT;
+}
+
+int journal_file_move_to_entry_by_monotonic(
+                JournalFile *f,
+                sd_id128_t boot_id,
+                uint64_t monotonic,
+                direction_t direction,
+                Object **ret,
+                uint64_t *offset) {
+
+        char t[8+32+1] = "_BOOT_ID=";
+        Object *o;
+        int r;
+
+        sd_id128_to_string(boot_id, t + 8);
+
+        r = journal_file_find_data_object(f, t, strlen(t), &o, NULL);
+        if (r < 0)
+                return r;
+        else if (r == 0)
+                return -ENOENT;
+
+        return generic_array_bisect_plus_one(f,
+                                             le64toh(o->data.entry_offset),
+                                             le64toh(o->data.entry_array_offset),
+                                             le64toh(o->data.n_entries),
+                                             monotonic,
+                                             test_object_monotonic,
+                                             direction,
+                                             ret, offset, NULL);
+}
+
+static int test_object_offset(JournalFile *f, uint64_t p, uint64_t needle) {
+        assert(f);
+        assert(p > 0);
+
+        if (p == needle)
+                return TEST_FOUND;
+        else if (p < needle)
+                return TEST_LEFT;
+        else
+                return TEST_RIGHT;
+}
+
+int journal_file_next_entry(
+                JournalFile *f,
+                Object *o, uint64_t p,
+                direction_t direction,
+                Object **ret, uint64_t *offset) {
+
+        uint64_t i, n;
+        int r;
+
+        assert(f);
+        assert(p > 0 || !o);
+
+        n = le64toh(f->header->n_entries);
+        if (n <= 0)
+                return 0;
+
+        if (!o)
+                i = direction == DIRECTION_DOWN ? 0 : n - 1;
+        else {
+                if (o->object.type != OBJECT_ENTRY)
+                        return -EINVAL;
+
+                r = generic_array_bisect(f,
+                                         le64toh(f->header->entry_array_offset),
+                                         le64toh(f->header->n_entries),
+                                         p,
+                                         test_object_offset,
+                                         DIRECTION_DOWN,
+                                         NULL, NULL,
+                                         &i);
+                if (r <= 0)
+                        return r;
+
+                if (direction == DIRECTION_DOWN) {
+                        if (i >= n - 1)
+                                return 0;
+
+                        i++;
+                } else {
+                        if (i <= 0)
+                                return 0;
+
+                        i--;
+                }
+        }
+
+        /* And jump to it */
+        return generic_array_get(f,
+                                 le64toh(f->header->entry_array_offset),
+                                 i,
+                                 ret, offset);
+}
+
+int journal_file_skip_entry(
+                JournalFile *f,
+                Object *o, uint64_t p,
+                int64_t skip,
+                Object **ret, uint64_t *offset) {
+
+        uint64_t i, n;
+        int r;
+
+        assert(f);
+        assert(o);
+        assert(p > 0);
+
+        if (o->object.type != OBJECT_ENTRY)
+                return -EINVAL;
+
+        r = generic_array_bisect(f,
+                                 le64toh(f->header->entry_array_offset),
+                                 le64toh(f->header->n_entries),
+                                 p,
+                                 test_object_offset,
+                                 DIRECTION_DOWN,
+                                 NULL, NULL,
+                                 &i);
+        if (r <= 0)
+                return r;
+
+        /* Calculate new index */
+        if (skip < 0) {
+                if ((uint64_t) -skip >= i)
+                        i = 0;
+                else
+                        i = i - (uint64_t) -skip;
+        } else
+                i  += (uint64_t) skip;
+
+        n = le64toh(f->header->n_entries);
+        if (n <= 0)
+                return -EBADMSG;
+
+        if (i >= n)
+                i = n-1;
+
+        return generic_array_get(f,
+                                 le64toh(f->header->entry_array_offset),
+                                 i,
+                                 ret, offset);
+}
+
+int journal_file_next_entry_for_data(
+                JournalFile *f,
+                Object *o, uint64_t p,
+                uint64_t data_offset,
+                direction_t direction,
+                Object **ret, uint64_t *offset) {
+
+        uint64_t n, i;
+        int r;
+        Object *d;
+
+        assert(f);
+        assert(p > 0 || !o);
+
+        r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
+        if (r <= 0)
+                return r;
+
+        n = le64toh(d->data.n_entries);
+        if (n <= 0)
+                return n;
+
+        if (!o)
+                i = direction == DIRECTION_DOWN ? 0 : n - 1;
+        else {
+                if (o->object.type != OBJECT_ENTRY)
+                        return -EINVAL;
+
+                r = generic_array_bisect_plus_one(f,
+                                                  le64toh(d->data.entry_offset),
+                                                  le64toh(d->data.entry_array_offset),
+                                                  le64toh(d->data.n_entries),
+                                                  p,
+                                                  test_object_offset,
+                                                  DIRECTION_DOWN,
+                                                  NULL, NULL,
+                                                  &i);
+
+                if (r <= 0)
+                        return r;
+
+                if (direction == DIRECTION_DOWN) {
+                        if (i >= n - 1)
+                                return 0;
+
+                        i++;
+                } else {
+                        if (i <= 0)
+                                return 0;
+
+                        i--;
+                }
+
+        }
+
+        return generic_array_get_plus_one(f,
+                                          le64toh(d->data.entry_offset),
+                                          le64toh(d->data.entry_array_offset),
+                                          i,
+                                          ret, offset);
+}
+
+int journal_file_move_to_entry_by_seqnum_for_data(
+                JournalFile *f,
+                uint64_t data_offset,
+                uint64_t seqnum,
+                direction_t direction,
+                Object **ret, uint64_t *offset) {
+
+        Object *d;
+        int r;
+
+        r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
+        if (r <= 0)
+                return r;
+
+        return generic_array_bisect_plus_one(f,
+                                             le64toh(d->data.entry_offset),
+                                             le64toh(d->data.entry_array_offset),
+                                             le64toh(d->data.n_entries),
+                                             seqnum,
+                                             test_object_seqnum,
+                                             direction,
+                                             ret, offset, NULL);
+}
+
+int journal_file_move_to_entry_by_realtime_for_data(
+                JournalFile *f,
+                uint64_t data_offset,
+                uint64_t realtime,
+                direction_t direction,
+                Object **ret, uint64_t *offset) {
+
+        Object *d;
+        int r;
+
+        r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
+        if (r <= 0)
+                return r;
+
+        return generic_array_bisect_plus_one(f,
+                                             le64toh(d->data.entry_offset),
+                                             le64toh(d->data.entry_array_offset),
+                                             le64toh(d->data.n_entries),
+                                             realtime,
+                                             test_object_realtime,
+                                             direction,
+                                             ret, offset, NULL);
+}
+
+void journal_file_dump(JournalFile *f) {
+        char a[33], b[33], c[33];
+        Object *o;
+        int r;
+        uint64_t p;
+
+        assert(f);
+
+        printf("File Path: %s\n"
+               "File ID: %s\n"
+               "Machine ID: %s\n"
+               "Boot ID: %s\n"
+               "Arena size: %llu\n"
+               "Objects: %lu\n"
+               "Entries: %lu\n",
+               f->path,
+               sd_id128_to_string(f->header->file_id, a),
+               sd_id128_to_string(f->header->machine_id, b),
+               sd_id128_to_string(f->header->boot_id, c),
+               (unsigned long long) le64toh(f->header->arena_size),
+               (unsigned long) le64toh(f->header->n_objects),
+               (unsigned long) le64toh(f->header->n_entries));
+
+        p = le64toh(f->header->arena_offset);
+        while (p != 0) {
+                r = journal_file_move_to_object(f, -1, p, &o);
+                if (r < 0)
+                        goto fail;
+
+                switch (o->object.type) {
+
+                case OBJECT_UNUSED:
+                        printf("Type: OBJECT_UNUSED\n");
+                        break;
+
+                case OBJECT_DATA:
+                        printf("Type: OBJECT_DATA\n");
+                        break;
+
+                case OBJECT_ENTRY:
+                        printf("Type: OBJECT_ENTRY %llu %llu %llu\n",
+                               (unsigned long long) le64toh(o->entry.seqnum),
+                               (unsigned long long) le64toh(o->entry.monotonic),
+                               (unsigned long long) le64toh(o->entry.realtime));
+                        break;
+
+                case OBJECT_FIELD_HASH_TABLE:
+                        printf("Type: OBJECT_FIELD_HASH_TABLE\n");
+                        break;
+
+                case OBJECT_DATA_HASH_TABLE:
+                        printf("Type: OBJECT_DATA_HASH_TABLE\n");
+                        break;
+
+                case OBJECT_ENTRY_ARRAY:
+                        printf("Type: OBJECT_ENTRY_ARRAY\n");
+                        break;
+                }
+
+                if (p == le64toh(f->header->tail_object_offset))
+                        p = 0;
+                else
+                        p = p + ALIGN64(le64toh(o->object.size));
+        }
+
+        return;
+fail:
+        log_error("File corrupt");
+}
+
+int journal_file_open(
+                const char *fname,
+                int flags,
+                mode_t mode,
+                JournalFile *template,
+                JournalFile **ret) {
+
+        JournalFile *f;
+        int r;
+        bool newly_created = false;
+
+        assert(fname);
+
+        if ((flags & O_ACCMODE) != O_RDONLY &&
+            (flags & O_ACCMODE) != O_RDWR)
+                return -EINVAL;
+
+        f = new0(JournalFile, 1);
+        if (!f)
+                return -ENOMEM;
+
+        f->fd = -1;
+        f->flags = flags;
+        f->mode = mode;
+        f->writable = (flags & O_ACCMODE) != O_RDONLY;
+        f->prot = prot_from_flags(flags);
+
+        f->path = strdup(fname);
+        if (!f->path) {
+                r = -ENOMEM;
+                goto fail;
+        }
+
+        f->fd = open(f->path, f->flags|O_CLOEXEC, f->mode);
+        if (f->fd < 0) {
+                r = -errno;
+                goto fail;
+        }
+
+        if (fstat(f->fd, &f->last_stat) < 0) {
+                r = -errno;
+                goto fail;
+        }
+
+        if (f->last_stat.st_size == 0 && f->writable) {
+                newly_created = true;
+
+                r = journal_file_init_header(f, template);
+                if (r < 0)
+                        goto fail;
+
+                if (fstat(f->fd, &f->last_stat) < 0) {
+                        r = -errno;
+                        goto fail;
+                }
+        }
+
+        if (f->last_stat.st_size < (off_t) sizeof(Header)) {
+                r = -EIO;
+                goto fail;
+        }
+
+        f->header = mmap(NULL, PAGE_ALIGN(sizeof(Header)), prot_from_flags(flags), MAP_SHARED, f->fd, 0);
+        if (f->header == MAP_FAILED) {
+                f->header = NULL;
+                r = -errno;
+                goto fail;
+        }
+
+        if (!newly_created) {
+                r = journal_file_verify_header(f);
+                if (r < 0)
+                        goto fail;
+        }
+
+        if (f->writable) {
+                r = journal_file_refresh_header(f);
+                if (r < 0)
+                        goto fail;
+        }
+
+        if (newly_created) {
+
+                r = journal_file_setup_field_hash_table(f);
+                if (r < 0)
+                        goto fail;
+
+                r = journal_file_setup_data_hash_table(f);
+                if (r < 0)
+                        goto fail;
+        }
+
+        r = journal_file_map_field_hash_table(f);
+        if (r < 0)
+                goto fail;
+
+        r = journal_file_map_data_hash_table(f);
+        if (r < 0)
+                goto fail;
+
+        if (ret)
+                *ret = f;
+
+        return 0;
+
+fail:
+        journal_file_close(f);
+
+        return r;
+}
+
+int journal_file_rotate(JournalFile **f) {
+        char *p;
+        size_t l;
+        JournalFile *old_file, *new_file = NULL;
+        int r;
+
+        assert(f);
+        assert(*f);
+
+        old_file = *f;
+
+        if (!old_file->writable)
+                return -EINVAL;
+
+        if (!endswith(old_file->path, ".journal"))
+                return -EINVAL;
+
+        l = strlen(old_file->path);
+
+        p = new(char, l + 1 + 16 + 1 + 32 + 1 + 16 + 1);
+        if (!p)
+                return -ENOMEM;
+
+        memcpy(p, old_file->path, l - 8);
+        p[l-8] = '@';
+        sd_id128_to_string(old_file->header->seqnum_id, p + l - 8 + 1);
+        snprintf(p + l - 8 + 1 + 32, 1 + 16 + 1 + 16 + 8 + 1,
+                 "-%016llx-%016llx.journal",
+                 (unsigned long long) le64toh((*f)->header->seqnum),
+                 (unsigned long long) le64toh((*f)->header->tail_entry_realtime));
+
+        r = rename(old_file->path, p);
+        free(p);
+
+        if (r < 0)
+                return -errno;
+
+        old_file->header->state = le32toh(STATE_ARCHIVED);
+
+        r = journal_file_open(old_file->path, old_file->flags, old_file->mode, old_file, &new_file);
+        journal_file_close(old_file);
+
+        *f = new_file;
+        return r;
+}
+
+struct vacuum_info {
+        off_t usage;
+        char *filename;
+
+        uint64_t realtime;
+        sd_id128_t seqnum_id;
+        uint64_t seqnum;
+};
+
+static int vacuum_compare(const void *_a, const void *_b) {
+        const struct vacuum_info *a, *b;
+
+        a = _a;
+        b = _b;
+
+        if (sd_id128_equal(a->seqnum_id, b->seqnum_id)) {
+                if (a->seqnum < b->seqnum)
+                        return -1;
+                else if (a->seqnum > b->seqnum)
+                        return 1;
+                else
+                        return 0;
+        }
+
+        if (a->realtime < b->realtime)
+                return -1;
+        else if (a->realtime > b->realtime)
+                return 1;
+        else
+                return memcmp(&a->seqnum_id, &b->seqnum_id, 16);
+}
+
+int journal_directory_vacuum(const char *directory, uint64_t max_use, uint64_t min_free) {
+        DIR *d;
+        int r = 0;
+        struct vacuum_info *list = NULL;
+        unsigned n_list = 0, n_allocated = 0, i;
+        uint64_t sum = 0;
+
+        assert(directory);
+
+        if (max_use <= 0)
+                max_use = DEFAULT_MAX_USE;
+
+        d = opendir(directory);
+        if (!d)
+                return -errno;
+
+        for (;;) {
+                int k;
+                struct dirent buf, *de;
+                size_t q;
+                struct stat st;
+                char *p;
+                unsigned long long seqnum, realtime;
+                sd_id128_t seqnum_id;
+
+                k = readdir_r(d, &buf, &de);
+                if (k != 0) {
+                        r = -k;
+                        goto finish;
+                }
+
+                if (!de)
+                        break;
+
+                if (!dirent_is_file_with_suffix(de, ".journal"))
+                        continue;
+
+                q = strlen(de->d_name);
+
+                if (q < 1 + 32 + 1 + 16 + 1 + 16 + 8)
+                        continue;
+
+                if (de->d_name[q-8-16-1] != '-' ||
+                    de->d_name[q-8-16-1-16-1] != '-' ||
+                    de->d_name[q-8-16-1-16-1-32-1] != '@')
+                        continue;
+
+                if (fstatat(dirfd(d), de->d_name, &st, AT_SYMLINK_NOFOLLOW) < 0)
+                        continue;
+
+                if (!S_ISREG(st.st_mode))
+                        continue;
+
+                p = strdup(de->d_name);
+                if (!p) {
+                        r = -ENOMEM;
+                        goto finish;
+                }
+
+                de->d_name[q-8-16-1-16-1] = 0;
+                if (sd_id128_from_string(de->d_name + q-8-16-1-16-1-32, &seqnum_id) < 0) {
+                        free(p);
+                        continue;
+                }
+
+                if (sscanf(de->d_name + q-8-16-1-16, "%16llx-%16llx.journal", &seqnum, &realtime) != 2) {
+                        free(p);
+                        continue;
+                }
+
+                if (n_list >= n_allocated) {
+                        struct vacuum_info *j;
+
+                        n_allocated = MAX(n_allocated * 2U, 8U);
+                        j = realloc(list, n_allocated * sizeof(struct vacuum_info));
+                        if (!j) {
+                                free(p);
+                                r = -ENOMEM;
+                                goto finish;
+                        }
+
+                        list = j;
+                }
+
+                list[n_list].filename = p;
+                list[n_list].usage = (uint64_t) st.st_blksize * (uint64_t) st.st_blocks;
+                list[n_list].seqnum = seqnum;
+                list[n_list].realtime = realtime;
+                list[n_list].seqnum_id = seqnum_id;
+
+                sum += list[n_list].usage;
+
+                n_list ++;
+        }
+
+        qsort(list, n_list, sizeof(struct vacuum_info), vacuum_compare);
+
+        for(i = 0; i < n_list; i++) {
+                struct statvfs ss;
+
+                if (fstatvfs(dirfd(d), &ss) < 0) {
+                        r = -errno;
+                        goto finish;
+                }
+
+                if (sum <= max_use &&
+                    (uint64_t) ss.f_bavail * (uint64_t) ss.f_bsize >= min_free)
+                        break;
+
+                if (unlinkat(dirfd(d), list[i].filename, 0) >= 0) {
+                        log_debug("Deleted archived journal %s/%s.", directory, list[i].filename);
+                        sum -= list[i].usage;
+                } else if (errno != ENOENT)
+                        log_warning("Failed to delete %s/%s: %m", directory, list[i].filename);
+        }
+
+finish:
+        for (i = 0; i < n_list; i++)
+                free(list[i].filename);
+
+        free(list);
+
+        if (d)
+                closedir(d);
+
+        return r;
+}
diff --git a/src/journal/journal-file.h b/src/journal/journal-file.h
new file mode 100644 (file)
index 0000000..664f917
--- /dev/null
@@ -0,0 +1,102 @@
+/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
+
+#ifndef foojournalfilehfoo
+#define foojournalfilehfoo
+
+/***
+  This file is part of systemd.
+
+  Copyright 2011 Lennart Poettering
+
+  systemd is free software; you can redistribute it and/or modify it
+  under the terms of the GNU General Public License as published by
+  the Free Software Foundation; either version 2 of the License, or
+  (at your option) any later version.
+
+  systemd is distributed in the hope that it will be useful, but
+  WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+  General Public License for more details.
+
+  You should have received a copy of the GNU General Public License
+  along with systemd; If not, see <http://www.gnu.org/licenses/>.
+***/
+
+#include <inttypes.h>
+
+#include "journal-def.h"
+#include "util.h"
+#include "sd-id128.h"
+
+typedef struct Window {
+        void *ptr;
+        uint64_t offset;
+        uint64_t size;
+} Window;
+
+enum {
+        WINDOW_UNKNOWN = OBJECT_UNUSED,
+        WINDOW_DATA = OBJECT_DATA,
+        WINDOW_ENTRY = OBJECT_ENTRY,
+        WINDOW_DATA_HASH_TABLE = OBJECT_DATA_HASH_TABLE,
+        WINDOW_FIELD_HASH_TABLE = OBJECT_FIELD_HASH_TABLE,
+        WINDOW_ENTRY_ARRAY = OBJECT_ENTRY_ARRAY,
+        WINDOW_HEADER,
+        _WINDOW_MAX
+};
+
+typedef struct JournalFile {
+        int fd;
+        char *path;
+        struct stat last_stat;
+        mode_t mode;
+        int flags;
+        int prot;
+        bool writable;
+        bool tail_entry_monotonic_valid;
+
+        Header *header;
+        HashItem *data_hash_table;
+        HashItem *field_hash_table;
+
+        Window windows[_WINDOW_MAX];
+
+        uint64_t current_offset;
+} JournalFile;
+
+typedef enum direction {
+        DIRECTION_UP,
+        DIRECTION_DOWN
+} direction_t;
+
+int journal_file_open(const char *fname, int flags, mode_t mode, JournalFile *template, JournalFile **ret);
+void journal_file_close(JournalFile *j);
+
+int journal_file_move_to_object(JournalFile *f, int type, uint64_t offset, Object **ret);
+
+uint64_t journal_file_entry_n_items(Object *o);
+
+int journal_file_append_entry(JournalFile *f, const dual_timestamp *ts, const struct iovec iovec[], unsigned n_iovec, uint64_t *seqno, Object **ret, uint64_t *offset);
+
+int journal_file_find_data_object(JournalFile *f, const void *data, uint64_t size, Object **ret, uint64_t *offset);
+int journal_file_find_data_object_with_hash(JournalFile *f, const void *data, uint64_t size, uint64_t hash, Object **ret, uint64_t *offset);
+
+int journal_file_next_entry(JournalFile *f, Object *o, uint64_t p, direction_t direction, Object **ret, uint64_t *offset);
+int journal_file_skip_entry(JournalFile *f, Object *o, uint64_t p, int64_t skip, Object **ret, uint64_t *offset);
+
+int journal_file_next_entry_for_data(JournalFile *f, Object *o, uint64_t p, uint64_t data_offset, direction_t direction, Object **ret, uint64_t *offset);
+
+int journal_file_move_to_entry_by_seqnum(JournalFile *f, uint64_t seqnum, direction_t direction, Object **ret, uint64_t *offset);
+int journal_file_move_to_entry_by_realtime(JournalFile *f, uint64_t realtime, direction_t direction, Object **ret, uint64_t *offset);
+int journal_file_move_to_entry_by_monotonic(JournalFile *f, sd_id128_t boot_id, uint64_t monotonic, direction_t direction, Object **ret, uint64_t *offset);
+
+int journal_file_move_to_entry_by_seqnum_for_data(JournalFile *f, uint64_t data_offset, uint64_t seqnum, direction_t direction, Object **ret, uint64_t *offset);
+int journal_file_move_to_entry_by_realtime_for_data(JournalFile *f, uint64_t data_offset, uint64_t realtime, direction_t direction, Object **ret, uint64_t *offset);
+
+void journal_file_dump(JournalFile *f);
+
+int journal_file_rotate(JournalFile **f);
+
+int journal_directory_vacuum(const char *directory, uint64_t max_use, uint64_t min_free);
+
+#endif
diff --git a/src/journal/journalctl.c b/src/journal/journalctl.c
new file mode 100644 (file)
index 0000000..9220efd
--- /dev/null
@@ -0,0 +1,90 @@
+/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
+
+/***
+  This file is part of systemd.
+
+  Copyright 2011 Lennart Poettering
+
+  systemd is free software; you can redistribute it and/or modify it
+  under the terms of the GNU General Public License as published by
+  the Free Software Foundation; either version 2 of the License, or
+  (at your option) any later version.
+
+  systemd is distributed in the hope that it will be useful, but
+  WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+  General Public License for more details.
+
+  You should have received a copy of the GNU General Public License
+  along with systemd; If not, see <http://www.gnu.org/licenses/>.
+***/
+
+#include <fcntl.h>
+#include <errno.h>
+#include <stddef.h>
+#include <string.h>
+#include <stdio.h>
+#include <unistd.h>
+#include <stdlib.h>
+
+#include "sd-journal.h"
+#include "log.h"
+
+int main(int argc, char *argv[]) {
+        int r, i;
+        sd_journal *j = NULL;
+
+        log_set_max_level(LOG_DEBUG);
+        log_set_target(LOG_TARGET_CONSOLE);
+
+        log_parse_environment();
+        log_open();
+
+        r = sd_journal_open(&j);
+        if (r < 0) {
+                log_error("Failed to open journal: %s", strerror(-r));
+                goto finish;
+        }
+
+        for (i = 1; i < argc; i++) {
+                r = sd_journal_add_match(j, argv[i], strlen(argv[i]));
+                if (r < 0) {
+                        log_error("Failed to add match: %s", strerror(-r));
+                        goto finish;
+                }
+        }
+
+        SD_JOURNAL_FOREACH(j) {
+
+                const void *data;
+                size_t length;
+                char *cursor;
+                uint64_t realtime = 0, monotonic = 0;
+
+                r = sd_journal_get_cursor(j, &cursor);
+                if (r < 0) {
+                        log_error("Failed to get cursor: %s", strerror(-r));
+                        goto finish;
+                }
+
+                printf("entry: %s\n", cursor);
+                free(cursor);
+
+                sd_journal_get_realtime_usec(j, &realtime);
+                sd_journal_get_monotonic_usec(j, &monotonic, NULL);
+                printf("realtime: %llu\n"
+                       "monotonic: %llu\n",
+                       (unsigned long long) realtime,
+                       (unsigned long long) monotonic);
+
+                SD_JOURNAL_FOREACH_DATA(j, data, length)
+                        printf("\t%.*s\n", (int) length, (const char*) data);
+
+        }
+
+finish:
+        if (j)
+                sd_journal_close(j);
+
+        return r < 0 ? EXIT_FAILURE : EXIT_SUCCESS;
+}
diff --git a/src/journal/journald.c b/src/journal/journald.c
new file mode 100644 (file)
index 0000000..89d8bee
--- /dev/null
@@ -0,0 +1,622 @@
+/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
+
+/***
+  This file is part of systemd.
+
+  Copyright 2011 Lennart Poettering
+
+  systemd is free software; you can redistribute it and/or modify it
+  under the terms of the GNU General Public License as published by
+  the Free Software Foundation; either version 2 of the License, or
+  (at your option) any later version.
+
+  systemd is distributed in the hope that it will be useful, but
+  WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+  General Public License for more details.
+
+  You should have received a copy of the GNU General Public License
+  along with systemd; If not, see <http://www.gnu.org/licenses/>.
+***/
+
+#include <sys/epoll.h>
+#include <sys/socket.h>
+#include <errno.h>
+#include <sys/signalfd.h>
+#include <unistd.h>
+#include <fcntl.h>
+#include <sys/acl.h>
+#include <acl/libacl.h>
+
+#include "hashmap.h"
+#include "journal-file.h"
+#include "sd-daemon.h"
+#include "socket-util.h"
+#include "acl-util.h"
+#include "cgroup-util.h"
+
+typedef struct Server {
+        int syslog_fd;
+        int epoll_fd;
+        int signal_fd;
+
+        JournalFile *runtime_journal;
+        JournalFile *system_journal;
+        Hashmap *user_journals;
+
+        uint64_t seqnum;
+} Server;
+
+static void fix_perms(JournalFile *f, uid_t uid) {
+        acl_t acl;
+        acl_entry_t entry;
+        acl_permset_t permset;
+        int r;
+
+        assert(f);
+
+        r = fchmod_and_fchown(f->fd, 0640, 0, 0);
+        if (r < 0)
+                log_warning("Failed to fix access mode/rights on %s, ignoring: %s", f->path, strerror(-r));
+
+        if (uid <= 0)
+                return;
+
+        acl = acl_get_fd(f->fd);
+        if (!acl) {
+                log_warning("Failed to read ACL on %s, ignoring: %m", f->path);
+                return;
+        }
+
+        r = acl_find_uid(acl, uid, &entry);
+        if (r <= 0) {
+
+                if (acl_create_entry(&acl, &entry) < 0 ||
+                    acl_set_tag_type(entry, ACL_USER) < 0 ||
+                    acl_set_qualifier(entry, &uid) < 0) {
+                        log_warning("Failed to patch ACL on %s, ignoring: %m", f->path);
+                        goto finish;
+                }
+        }
+
+        if (acl_get_permset(entry, &permset) < 0 ||
+            acl_add_perm(permset, ACL_READ) < 0 ||
+            acl_calc_mask(&acl) < 0) {
+                log_warning("Failed to patch ACL on %s, ignoring: %m", f->path);
+                goto finish;
+        }
+
+        if (acl_set_fd(f->fd, acl) < 0)
+                log_warning("Failed to set ACL on %s, ignoring: %m", f->path);
+
+finish:
+        acl_free(acl);
+}
+
+static JournalFile* find_journal(Server *s, uid_t uid) {
+        char *p;
+        int r;
+        JournalFile *f;
+        char ids[33];
+        sd_id128_t machine;
+
+        assert(s);
+
+        /* We split up user logs only on /var, not on /run */
+        if (!s->system_journal)
+                return s->runtime_journal;
+
+        if (uid <= 0)
+                return s->system_journal;
+
+        r = sd_id128_get_machine(&machine);
+        if (r < 0)
+                return s->system_journal;
+
+        f = hashmap_get(s->user_journals, UINT32_TO_PTR(uid));
+        if (f)
+                return f;
+
+        if (asprintf(&p, "/var/log/journal/%s/user-%lu.journal", sd_id128_to_string(machine, ids), (unsigned long) uid) < 0)
+                return s->system_journal;
+
+        r = journal_file_open(p, O_RDWR|O_CREAT, 0640, s->system_journal, &f);
+        free(p);
+
+        if (r < 0)
+                return s->system_journal;
+
+        fix_perms(f, uid);
+
+        r = hashmap_put(s->user_journals, UINT32_TO_PTR(uid), f);
+        if (r < 0) {
+                journal_file_close(f);
+                return s->system_journal;
+        }
+
+        return f;
+}
+
+static void process_message(Server *s, const char *buf, struct ucred *ucred, struct timeval *tv) {
+        char *message = NULL, *pid = NULL, *uid = NULL, *gid = NULL,
+                *source_time = NULL, *boot_id = NULL, *machine_id = NULL,
+                *comm = NULL, *cmdline = NULL, *hostname = NULL,
+                *audit_session = NULL, *audit_loginuid = NULL,
+                *syslog_priority = NULL, *syslog_facility = NULL,
+                *exe = NULL, *cgroup = NULL;
+        struct iovec iovec[17];
+        unsigned n = 0;
+        char idbuf[33];
+        sd_id128_t id;
+        int r;
+        char *t;
+        int priority = LOG_USER | LOG_INFO;
+        uid_t loginuid = 0, realuid = 0;
+        JournalFile *f;
+
+        parse_syslog_priority((char**) &buf, &priority);
+        skip_syslog_date((char**) &buf);
+
+        if (asprintf(&syslog_priority, "PRIORITY=%i", priority & LOG_PRIMASK) >= 0)
+                IOVEC_SET_STRING(iovec[n++], syslog_priority);
+
+        if (asprintf(&syslog_facility, "SYSLOG_FACILITY=%i", LOG_FAC(priority)) >= 0)
+                IOVEC_SET_STRING(iovec[n++], syslog_facility);
+
+        message = strappend("MESSAGE=", buf);
+        if (message)
+                IOVEC_SET_STRING(iovec[n++], message);
+
+        if (ucred) {
+                uint32_t session;
+                char *path;
+
+                realuid = ucred->uid;
+
+                if (asprintf(&pid, "_PID=%lu", (unsigned long) ucred->pid) >= 0)
+                        IOVEC_SET_STRING(iovec[n++], pid);
+
+                if (asprintf(&uid, "_UID=%lu", (unsigned long) ucred->uid) >= 0)
+                        IOVEC_SET_STRING(iovec[n++], uid);
+
+                if (asprintf(&gid, "_GID=%lu", (unsigned long) ucred->gid) >= 0)
+                        IOVEC_SET_STRING(iovec[n++], gid);
+
+                r = get_process_comm(ucred->pid, &t);
+                if (r >= 0) {
+                        comm = strappend("_COMM=", t);
+                        if (comm)
+                                IOVEC_SET_STRING(iovec[n++], comm);
+                        free(t);
+                }
+
+                r = get_process_exe(ucred->pid, &t);
+                if (r >= 0) {
+                        exe = strappend("_EXE=", t);
+                        if (comm)
+                                IOVEC_SET_STRING(iovec[n++], exe);
+                        free(t);
+                }
+
+                r = get_process_cmdline(ucred->pid, LINE_MAX, false, &t);
+                if (r >= 0) {
+                        cmdline = strappend("_CMDLINE=", t);
+                        if (cmdline)
+                                IOVEC_SET_STRING(iovec[n++], cmdline);
+                        free(t);
+                }
+
+                r = audit_session_from_pid(ucred->pid, &session);
+                if (r >= 0)
+                        if (asprintf(&audit_session, "_AUDIT_SESSION=%lu", (unsigned long) session) >= 0)
+                                IOVEC_SET_STRING(iovec[n++], audit_session);
+
+                r = audit_loginuid_from_pid(ucred->pid, &loginuid);
+                if (r >= 0)
+                        if (asprintf(&audit_loginuid, "_AUDIT_LOGINUID=%lu", (unsigned long) loginuid) >= 0)
+                                IOVEC_SET_STRING(iovec[n++], audit_loginuid);
+
+                r = cg_get_by_pid(SYSTEMD_CGROUP_CONTROLLER, ucred->pid, &path);
+                if (r >= 0) {
+                        cgroup = strappend("_SYSTEMD_CGROUP=", path);
+                        if (cgroup)
+                                IOVEC_SET_STRING(iovec[n++], cgroup);
+                        free(path);
+                }
+        }
+
+        if (tv) {
+                if (asprintf(&source_time, "_SOURCE_REALTIME_TIMESTAMP=%llu",
+                             (unsigned long long) timeval_load(tv)) >= 0)
+                        IOVEC_SET_STRING(iovec[n++], source_time);
+        }
+
+        /* Note that strictly speaking storing the boot id here is
+         * redundant since the entry includes this in-line
+         * anyway. However, we need this indexed, too. */
+        r = sd_id128_get_boot(&id);
+        if (r >= 0)
+                if (asprintf(&boot_id, "_BOOT_ID=%s", sd_id128_to_string(id, idbuf)) >= 0)
+                        IOVEC_SET_STRING(iovec[n++], boot_id);
+
+        r = sd_id128_get_machine(&id);
+        if (r >= 0)
+                if (asprintf(&machine_id, "_MACHINE_ID=%s", sd_id128_to_string(id, idbuf)) >= 0)
+                        IOVEC_SET_STRING(iovec[n++], machine_id);
+
+        t = gethostname_malloc();
+        if (t) {
+                hostname = strappend("_HOSTNAME=", t);
+                if (hostname)
+                        IOVEC_SET_STRING(iovec[n++], hostname);
+                free(t);
+        }
+
+        f = find_journal(s, realuid == 0 ? 0 : loginuid);
+        if (!f)
+                log_warning("Dropping message, as we can't find a place to store the data.");
+        else {
+                r = journal_file_append_entry(f, NULL, iovec, n, &s->seqnum, NULL, NULL);
+
+                if (r < 0)
+                        log_error("Failed to write entry, ignoring: %s", strerror(-r));
+        }
+
+        free(message);
+        free(pid);
+        free(uid);
+        free(gid);
+        free(comm);
+        free(exe);
+        free(cmdline);
+        free(source_time);
+        free(boot_id);
+        free(machine_id);
+        free(hostname);
+        free(audit_session);
+        free(audit_loginuid);
+        free(syslog_facility);
+        free(syslog_priority);
+        free(cgroup);
+}
+
+static int process_event(Server *s, struct epoll_event *ev) {
+        assert(s);
+
+        if (ev->events != EPOLLIN) {
+                log_info("Got invalid event from epoll.");
+                return -EIO;
+        }
+
+        if (ev->data.fd == s->signal_fd) {
+                struct signalfd_siginfo sfsi;
+                ssize_t n;
+
+                n = read(s->signal_fd, &sfsi, sizeof(sfsi));
+                if (n != sizeof(sfsi)) {
+
+                        if (n >= 0)
+                                return -EIO;
+
+                        if (errno == EINTR || errno == EAGAIN)
+                                return 0;
+
+                        return -errno;
+                }
+
+                log_debug("Received SIG%s", signal_to_string(sfsi.ssi_signo));
+                return 0;
+
+        }
+
+        if (ev->data.fd == s->syslog_fd) {
+                for (;;) {
+                        char buf[LINE_MAX+1];
+                        struct msghdr msghdr;
+                        struct iovec iovec;
+                        struct ucred *ucred = NULL;
+                        struct timeval *tv = NULL;
+                        struct cmsghdr *cmsg;
+                        union {
+                                struct cmsghdr cmsghdr;
+                                uint8_t buf[CMSG_SPACE(sizeof(struct ucred)) +
+                                            CMSG_SPACE(sizeof(struct timeval))];
+                        } control;
+                        ssize_t n;
+                        char *e;
+
+                        zero(iovec);
+                        iovec.iov_base = buf;
+                        iovec.iov_len = sizeof(buf)-1;
+
+                        zero(control);
+                        zero(msghdr);
+                        msghdr.msg_iov = &iovec;
+                        msghdr.msg_iovlen = 1;
+                        msghdr.msg_control = &control;
+                        msghdr.msg_controllen = sizeof(control);
+
+                        n = recvmsg(ev->data.fd, &msghdr, MSG_DONTWAIT);
+                        if (n < 0) {
+
+                                if (errno == EINTR || errno == EAGAIN)
+                                        return 1;
+
+                                log_error("recvmsg() failed: %m");
+                                return -errno;
+                        }
+
+                        for (cmsg = CMSG_FIRSTHDR(&msghdr); cmsg; cmsg = CMSG_NXTHDR(&msghdr, cmsg)) {
+
+                                if (cmsg->cmsg_level == SOL_SOCKET &&
+                                    cmsg->cmsg_type == SCM_CREDENTIALS &&
+                                    cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred)))
+                                        ucred = (struct ucred*) CMSG_DATA(cmsg);
+                                else if (cmsg->cmsg_level == SOL_SOCKET &&
+                                         cmsg->cmsg_type == SO_TIMESTAMP &&
+                                         cmsg->cmsg_len == CMSG_LEN(sizeof(struct timeval)))
+                                        tv = (struct timeval*) CMSG_DATA(cmsg);
+                        }
+
+                        e = memchr(buf, '\n', n);
+                        if (e)
+                                *e = 0;
+                        else
+                                buf[n] = 0;
+
+                        process_message(s, strstrip(buf), ucred, tv);
+                }
+
+                return 1;
+        }
+
+        log_error("Unknown event.");
+        return 0;
+}
+
+static int system_journal_open(Server *s) {
+        int r;
+        char *fn;
+        sd_id128_t machine;
+        char ids[33];
+
+        r = sd_id128_get_machine(&machine);
+        if (r < 0)
+                return r;
+
+        /* First try to create the machine path, but not the prefix */
+        fn = strappend("/var/log/journal/", sd_id128_to_string(machine, ids));
+        if (!fn)
+                return -ENOMEM;
+        (void) mkdir(fn, 0755);
+        free(fn);
+
+        /* The create the system journal file */
+        fn = join("/var/log/journal/", ids, "/system.journal", NULL);
+        if (!fn)
+                return -ENOMEM;
+
+        r = journal_file_open(fn, O_RDWR|O_CREAT, 0640, NULL, &s->system_journal);
+        free(fn);
+
+        if (r >= 0) {
+                fix_perms(s->system_journal, 0);
+                return r;
+        }
+
+        if (r < 0 && r != -ENOENT) {
+                log_error("Failed to open system journal: %s", strerror(-r));
+                return r;
+        }
+
+        /* /var didn't work, so try /run, but this time we
+         * create the prefix too */
+        fn = join("/run/log/journal/", ids, "/system.journal", NULL);
+        if (!fn)
+                return -ENOMEM;
+
+        (void) mkdir_parents(fn, 0755);
+        r = journal_file_open(fn, O_RDWR|O_CREAT, 0640, NULL, &s->runtime_journal);
+        free(fn);
+
+        if (r < 0) {
+                log_error("Failed to open runtime journal: %s", strerror(-r));
+                return r;
+        }
+
+        fix_perms(s->runtime_journal, 0);
+        return r;
+}
+
+static int server_init(Server *s) {
+        int n, one, r;
+        struct epoll_event ev;
+        sigset_t mask;
+
+        assert(s);
+
+        zero(*s);
+        s->syslog_fd = s->signal_fd = -1;
+
+        s->epoll_fd = epoll_create1(EPOLL_CLOEXEC);
+        if (s->epoll_fd < 0) {
+                log_error("Failed to create epoll object: %m");
+                return -errno;
+        }
+
+        n = sd_listen_fds(true);
+        if (n < 0) {
+                log_error("Failed to read listening file descriptors from environment: %s", strerror(-n));
+                return n;
+        }
+
+        if (n > 1) {
+                log_error("Too many file descriptors passed.");
+                return -EINVAL;
+        }
+
+        if (n == 1)
+                s->syslog_fd = SD_LISTEN_FDS_START;
+        else {
+                union sockaddr_union sa;
+
+                s->syslog_fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0);
+                if (s->syslog_fd < 0) {
+                        log_error("socket() failed: %m");
+                        return -errno;
+                }
+
+                zero(sa);
+                sa.un.sun_family = AF_UNIX;
+                strncpy(sa.un.sun_path, "/run/systemd/syslog", sizeof(sa.un.sun_path));
+
+                unlink(sa.un.sun_path);
+
+                r = bind(s->syslog_fd, &sa.sa, sizeof(sa.un));
+                if (r < 0) {
+                        log_error("bind() failed: %m");
+                        return -errno;
+                }
+
+                chmod(sa.un.sun_path, 0666);
+        }
+
+        one = 1;
+        r = setsockopt(s->syslog_fd, SOL_SOCKET, SO_PASSCRED, &one, sizeof(one));
+        if (r < 0) {
+                log_error("SO_PASSCRED failed: %m");
+                return -errno;
+        }
+
+        one = 1;
+        r = setsockopt(s->syslog_fd, SOL_SOCKET, SO_TIMESTAMP, &one, sizeof(one));
+        if (r < 0) {
+                log_error("SO_TIMESTAMP failed: %m");
+                return -errno;
+        }
+
+        zero(ev);
+        ev.events = EPOLLIN;
+        ev.data.fd = s->syslog_fd;
+        if (epoll_ctl(s->epoll_fd, EPOLL_CTL_ADD, s->syslog_fd, &ev) < 0) {
+                log_error("Failed to add server fd to epoll object: %m");
+                return -errno;
+        }
+
+        s->user_journals = hashmap_new(trivial_hash_func, trivial_compare_func);
+        if (!s->user_journals) {
+                log_error("Out of memory.");
+                return -ENOMEM;
+        }
+
+        r = system_journal_open(s);
+        if (r < 0)
+                return r;
+
+        assert_se(sigemptyset(&mask) == 0);
+        sigset_add_many(&mask, SIGINT, SIGTERM, -1);
+        assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
+
+        s->signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
+        if (s->signal_fd < 0) {
+                log_error("signalfd(): %m");
+                return -errno;
+        }
+
+        zero(ev);
+        ev.events = EPOLLIN;
+        ev.data.fd = s->signal_fd;
+
+        if (epoll_ctl(s->epoll_fd, EPOLL_CTL_ADD, s->signal_fd, &ev) < 0) {
+                log_error("epoll_ctl(): %m");
+                return -errno;
+        }
+
+        return 0;
+}
+
+static void server_done(Server *s) {
+        JournalFile *f;
+        assert(s);
+
+        if (s->system_journal)
+                journal_file_close(s->system_journal);
+
+        if (s->runtime_journal)
+                journal_file_close(s->runtime_journal);
+
+        while ((f = hashmap_steal_first(s->user_journals)))
+                journal_file_close(f);
+
+        hashmap_free(s->user_journals);
+
+        if (s->epoll_fd >= 0)
+                close_nointr_nofail(s->epoll_fd);
+
+        if (s->signal_fd >= 0)
+                close_nointr_nofail(s->signal_fd);
+
+        if (s->syslog_fd >= 0)
+                close_nointr_nofail(s->syslog_fd);
+}
+
+int main(int argc, char *argv[]) {
+        Server server;
+        int r;
+
+        /* if (getppid() != 1) { */
+        /*         log_error("This program should be invoked by init only."); */
+        /*         return EXIT_FAILURE; */
+        /* } */
+
+        if (argc > 1) {
+                log_error("This program does not take arguments.");
+                return EXIT_FAILURE;
+        }
+
+        log_set_target(LOG_TARGET_CONSOLE);
+        log_parse_environment();
+        log_open();
+
+        umask(0022);
+
+        r = server_init(&server);
+        if (r < 0)
+                goto finish;
+
+        log_debug("systemd-journald running as pid %lu", (unsigned long) getpid());
+
+        sd_notify(false,
+                  "READY=1\n"
+                  "STATUS=Processing messages...");
+#
+        for (;;) {
+                struct epoll_event event;
+
+                r = epoll_wait(server.epoll_fd, &event, 1, -1);
+                if (r < 0) {
+
+                        if (errno == EINTR)
+                                continue;
+
+                        log_error("epoll_wait() failed: %m");
+                        r = -errno;
+                        goto finish;
+                } else if (r == 0)
+                        break;
+
+                r = process_event(&server, &event);
+                if (r < 0)
+                        goto finish;
+                else if (r == 0)
+                        break;
+        }
+
+finish:
+        sd_notify(false,
+                  "STATUS=Shutting down...");
+
+        server_done(&server);
+
+        return r < 0 ? EXIT_FAILURE : EXIT_SUCCESS;
+}
diff --git a/src/journal/lookup3.c b/src/journal/lookup3.c
new file mode 100644 (file)
index 0000000..b90093a
--- /dev/null
@@ -0,0 +1,1003 @@
+/* Slightly modified by Lennart Poettering, to avoid name clashes, and
+ * unexport a few functions. */
+
+#include "lookup3.h"
+
+/*
+-------------------------------------------------------------------------------
+lookup3.c, by Bob Jenkins, May 2006, Public Domain.
+
+These are functions for producing 32-bit hashes for hash table lookup.
+hashword(), hashlittle(), hashlittle2(), hashbig(), mix(), and final()
+are externally useful functions.  Routines to test the hash are included
+if SELF_TEST is defined.  You can use this free for any purpose.  It's in
+the public domain.  It has no warranty.
+
+You probably want to use hashlittle().  hashlittle() and hashbig()
+hash byte arrays.  hashlittle() is is faster than hashbig() on
+little-endian machines.  Intel and AMD are little-endian machines.
+On second thought, you probably want hashlittle2(), which is identical to
+hashlittle() except it returns two 32-bit hashes for the price of one.
+You could implement hashbig2() if you wanted but I haven't bothered here.
+
+If you want to find a hash of, say, exactly 7 integers, do
+  a = i1;  b = i2;  c = i3;
+  mix(a,b,c);
+  a += i4; b += i5; c += i6;
+  mix(a,b,c);
+  a += i7;
+  final(a,b,c);
+then use c as the hash value.  If you have a variable length array of
+4-byte integers to hash, use hashword().  If you have a byte array (like
+a character string), use hashlittle().  If you have several byte arrays, or
+a mix of things, see the comments above hashlittle().
+
+Why is this so big?  I read 12 bytes at a time into 3 4-byte integers,
+then mix those integers.  This is fast (you can do a lot more thorough
+mixing with 12*3 instructions on 3 integers than you can with 3 instructions
+on 1 byte), but shoehorning those bytes into integers efficiently is messy.
+-------------------------------------------------------------------------------
+*/
+/* #define SELF_TEST 1 */
+
+#include <stdio.h>      /* defines printf for tests */
+#include <time.h>       /* defines time_t for timings in the test */
+#include <stdint.h>     /* defines uint32_t etc */
+#include <sys/param.h>  /* attempt to define endianness */
+#ifdef linux
+# include <endian.h>    /* attempt to define endianness */
+#endif
+
+/*
+ * My best guess at if you are big-endian or little-endian.  This may
+ * need adjustment.
+ */
+#if (defined(__BYTE_ORDER) && defined(__LITTLE_ENDIAN) && \
+     __BYTE_ORDER == __LITTLE_ENDIAN) || \
+    (defined(i386) || defined(__i386__) || defined(__i486__) || \
+     defined(__i586__) || defined(__i686__) || defined(vax) || defined(MIPSEL))
+# define HASH_LITTLE_ENDIAN 1
+# define HASH_BIG_ENDIAN 0
+#elif (defined(__BYTE_ORDER) && defined(__BIG_ENDIAN) && \
+       __BYTE_ORDER == __BIG_ENDIAN) || \
+      (defined(sparc) || defined(POWERPC) || defined(mc68000) || defined(sel))
+# define HASH_LITTLE_ENDIAN 0
+# define HASH_BIG_ENDIAN 1
+#else
+# define HASH_LITTLE_ENDIAN 0
+# define HASH_BIG_ENDIAN 0
+#endif
+
+#define hashsize(n) ((uint32_t)1<<(n))
+#define hashmask(n) (hashsize(n)-1)
+#define rot(x,k) (((x)<<(k)) | ((x)>>(32-(k))))
+
+/*
+-------------------------------------------------------------------------------
+mix -- mix 3 32-bit values reversibly.
+
+This is reversible, so any information in (a,b,c) before mix() is
+still in (a,b,c) after mix().
+
+If four pairs of (a,b,c) inputs are run through mix(), or through
+mix() in reverse, there are at least 32 bits of the output that
+are sometimes the same for one pair and different for another pair.
+This was tested for:
+* pairs that differed by one bit, by two bits, in any combination
+  of top bits of (a,b,c), or in any combination of bottom bits of
+  (a,b,c).
+* "differ" is defined as +, -, ^, or ~^.  For + and -, I transformed
+  the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
+  is commonly produced by subtraction) look like a single 1-bit
+  difference.
+* the base values were pseudorandom, all zero but one bit set, or
+  all zero plus a counter that starts at zero.
+
+Some k values for my "a-=c; a^=rot(c,k); c+=b;" arrangement that
+satisfy this are
+    4  6  8 16 19  4
+    9 15  3 18 27 15
+   14  9  3  7 17  3
+Well, "9 15 3 18 27 15" didn't quite get 32 bits diffing
+for "differ" defined as + with a one-bit base and a two-bit delta.  I
+used http://burtleburtle.net/bob/hash/avalanche.html to choose
+the operations, constants, and arrangements of the variables.
+
+This does not achieve avalanche.  There are input bits of (a,b,c)
+that fail to affect some output bits of (a,b,c), especially of a.  The
+most thoroughly mixed value is c, but it doesn't really even achieve
+avalanche in c.
+
+This allows some parallelism.  Read-after-writes are good at doubling
+the number of bits affected, so the goal of mixing pulls in the opposite
+direction as the goal of parallelism.  I did what I could.  Rotates
+seem to cost as much as shifts on every machine I could lay my hands
+on, and rotates are much kinder to the top and bottom bits, so I used
+rotates.
+-------------------------------------------------------------------------------
+*/
+#define mix(a,b,c) \
+{ \
+  a -= c;  a ^= rot(c, 4);  c += b; \
+  b -= a;  b ^= rot(a, 6);  a += c; \
+  c -= b;  c ^= rot(b, 8);  b += a; \
+  a -= c;  a ^= rot(c,16);  c += b; \
+  b -= a;  b ^= rot(a,19);  a += c; \
+  c -= b;  c ^= rot(b, 4);  b += a; \
+}
+
+/*
+-------------------------------------------------------------------------------
+final -- final mixing of 3 32-bit values (a,b,c) into c
+
+Pairs of (a,b,c) values differing in only a few bits will usually
+produce values of c that look totally different.  This was tested for
+* pairs that differed by one bit, by two bits, in any combination
+  of top bits of (a,b,c), or in any combination of bottom bits of
+  (a,b,c).
+* "differ" is defined as +, -, ^, or ~^.  For + and -, I transformed
+  the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
+  is commonly produced by subtraction) look like a single 1-bit
+  difference.
+* the base values were pseudorandom, all zero but one bit set, or
+  all zero plus a counter that starts at zero.
+
+These constants passed:
+ 14 11 25 16 4 14 24
+ 12 14 25 16 4 14 24
+and these came close:
+  4  8 15 26 3 22 24
+ 10  8 15 26 3 22 24
+ 11  8 15 26 3 22 24
+-------------------------------------------------------------------------------
+*/
+#define final(a,b,c) \
+{ \
+  c ^= b; c -= rot(b,14); \
+  a ^= c; a -= rot(c,11); \
+  b ^= a; b -= rot(a,25); \
+  c ^= b; c -= rot(b,16); \
+  a ^= c; a -= rot(c,4);  \
+  b ^= a; b -= rot(a,14); \
+  c ^= b; c -= rot(b,24); \
+}
+
+/*
+--------------------------------------------------------------------
+ This works on all machines.  To be useful, it requires
+ -- that the key be an array of uint32_t's, and
+ -- that the length be the number of uint32_t's in the key
+
+ The function hashword() is identical to hashlittle() on little-endian
+ machines, and identical to hashbig() on big-endian machines,
+ except that the length has to be measured in uint32_ts rather than in
+ bytes.  hashlittle() is more complicated than hashword() only because
+ hashlittle() has to dance around fitting the key bytes into registers.
+--------------------------------------------------------------------
+*/
+uint32_t jenkins_hashword(
+const uint32_t *k,                   /* the key, an array of uint32_t values */
+size_t          length,               /* the length of the key, in uint32_ts */
+uint32_t        initval)         /* the previous hash, or an arbitrary value */
+{
+  uint32_t a,b,c;
+
+  /* Set up the internal state */
+  a = b = c = 0xdeadbeef + (((uint32_t)length)<<2) + initval;
+
+  /*------------------------------------------------- handle most of the key */
+  while (length > 3)
+  {
+    a += k[0];
+    b += k[1];
+    c += k[2];
+    mix(a,b,c);
+    length -= 3;
+    k += 3;
+  }
+
+  /*------------------------------------------- handle the last 3 uint32_t's */
+  switch(length)                     /* all the case statements fall through */
+  {
+  case 3 : c+=k[2];
+  case 2 : b+=k[1];
+  case 1 : a+=k[0];
+    final(a,b,c);
+  case 0:     /* case 0: nothing left to add */
+    break;
+  }
+  /*------------------------------------------------------ report the result */
+  return c;
+}
+
+
+/*
+--------------------------------------------------------------------
+hashword2() -- same as hashword(), but take two seeds and return two
+32-bit values.  pc and pb must both be nonnull, and *pc and *pb must
+both be initialized with seeds.  If you pass in (*pb)==0, the output
+(*pc) will be the same as the return value from hashword().
+--------------------------------------------------------------------
+*/
+void jenkins_hashword2 (
+const uint32_t *k,                   /* the key, an array of uint32_t values */
+size_t          length,               /* the length of the key, in uint32_ts */
+uint32_t       *pc,                      /* IN: seed OUT: primary hash value */
+uint32_t       *pb)               /* IN: more seed OUT: secondary hash value */
+{
+  uint32_t a,b,c;
+
+  /* Set up the internal state */
+  a = b = c = 0xdeadbeef + ((uint32_t)(length<<2)) + *pc;
+  c += *pb;
+
+  /*------------------------------------------------- handle most of the key */
+  while (length > 3)
+  {
+    a += k[0];
+    b += k[1];
+    c += k[2];
+    mix(a,b,c);
+    length -= 3;
+    k += 3;
+  }
+
+  /*------------------------------------------- handle the last 3 uint32_t's */
+  switch(length)                     /* all the case statements fall through */
+  {
+  case 3 : c+=k[2];
+  case 2 : b+=k[1];
+  case 1 : a+=k[0];
+    final(a,b,c);
+  case 0:     /* case 0: nothing left to add */
+    break;
+  }
+  /*------------------------------------------------------ report the result */
+  *pc=c; *pb=b;
+}
+
+
+/*
+-------------------------------------------------------------------------------
+hashlittle() -- hash a variable-length key into a 32-bit value
+  k       : the key (the unaligned variable-length array of bytes)
+  length  : the length of the key, counting by bytes
+  initval : can be any 4-byte value
+Returns a 32-bit value.  Every bit of the key affects every bit of
+the return value.  Two keys differing by one or two bits will have
+totally different hash values.
+
+The best hash table sizes are powers of 2.  There is no need to do
+mod a prime (mod is sooo slow!).  If you need less than 32 bits,
+use a bitmask.  For example, if you need only 10 bits, do
+  h = (h & hashmask(10));
+In which case, the hash table should have hashsize(10) elements.
+
+If you are hashing n strings (uint8_t **)k, do it like this:
+  for (i=0, h=0; i<n; ++i) h = hashlittle( k[i], len[i], h);
+
+By Bob Jenkins, 2006.  bob_jenkins@burtleburtle.net.  You may use this
+code any way you wish, private, educational, or commercial.  It's free.
+
+Use for hash table lookup, or anything where one collision in 2^^32 is
+acceptable.  Do NOT use for cryptographic purposes.
+-------------------------------------------------------------------------------
+*/
+
+uint32_t jenkins_hashlittle( const void *key, size_t length, uint32_t initval)
+{
+  uint32_t a,b,c;                                          /* internal state */
+  union { const void *ptr; size_t i; } u;     /* needed for Mac Powerbook G4 */
+
+  /* Set up the internal state */
+  a = b = c = 0xdeadbeef + ((uint32_t)length) + initval;
+
+  u.ptr = key;
+  if (HASH_LITTLE_ENDIAN && ((u.i & 0x3) == 0)) {
+    const uint32_t *k = (const uint32_t *)key;         /* read 32-bit chunks */
+
+    /*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */
+    while (length > 12)
+    {
+      a += k[0];
+      b += k[1];
+      c += k[2];
+      mix(a,b,c);
+      length -= 12;
+      k += 3;
+    }
+
+    /*----------------------------- handle the last (probably partial) block */
+    /*
+     * "k[2]&0xffffff" actually reads beyond the end of the string, but
+     * then masks off the part it's not allowed to read.  Because the
+     * string is aligned, the masked-off tail is in the same word as the
+     * rest of the string.  Every machine with memory protection I've seen
+     * does it on word boundaries, so is OK with this.  But VALGRIND will
+     * still catch it and complain.  The masking trick does make the hash
+     * noticably faster for short strings (like English words).
+     */
+#ifndef VALGRIND
+
+    switch(length)
+    {
+    case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
+    case 11: c+=k[2]&0xffffff; b+=k[1]; a+=k[0]; break;
+    case 10: c+=k[2]&0xffff; b+=k[1]; a+=k[0]; break;
+    case 9 : c+=k[2]&0xff; b+=k[1]; a+=k[0]; break;
+    case 8 : b+=k[1]; a+=k[0]; break;
+    case 7 : b+=k[1]&0xffffff; a+=k[0]; break;
+    case 6 : b+=k[1]&0xffff; a+=k[0]; break;
+    case 5 : b+=k[1]&0xff; a+=k[0]; break;
+    case 4 : a+=k[0]; break;
+    case 3 : a+=k[0]&0xffffff; break;
+    case 2 : a+=k[0]&0xffff; break;
+    case 1 : a+=k[0]&0xff; break;
+    case 0 : return c;              /* zero length strings require no mixing */
+    }
+
+#else /* make valgrind happy */
+
+    k8 = (const uint8_t *)k;
+    switch(length)
+    {
+    case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
+    case 11: c+=((uint32_t)k8[10])<<16;  /* fall through */
+    case 10: c+=((uint32_t)k8[9])<<8;    /* fall through */
+    case 9 : c+=k8[8];                   /* fall through */
+    case 8 : b+=k[1]; a+=k[0]; break;
+    case 7 : b+=((uint32_t)k8[6])<<16;   /* fall through */
+    case 6 : b+=((uint32_t)k8[5])<<8;    /* fall through */
+    case 5 : b+=k8[4];                   /* fall through */
+    case 4 : a+=k[0]; break;
+    case 3 : a+=((uint32_t)k8[2])<<16;   /* fall through */
+    case 2 : a+=((uint32_t)k8[1])<<8;    /* fall through */
+    case 1 : a+=k8[0]; break;
+    case 0 : return c;
+    }
+
+#endif /* !valgrind */
+
+  } else if (HASH_LITTLE_ENDIAN && ((u.i & 0x1) == 0)) {
+    const uint16_t *k = (const uint16_t *)key;         /* read 16-bit chunks */
+    const uint8_t  *k8;
+
+    /*--------------- all but last block: aligned reads and different mixing */
+    while (length > 12)
+    {
+      a += k[0] + (((uint32_t)k[1])<<16);
+      b += k[2] + (((uint32_t)k[3])<<16);
+      c += k[4] + (((uint32_t)k[5])<<16);
+      mix(a,b,c);
+      length -= 12;
+      k += 6;
+    }
+
+    /*----------------------------- handle the last (probably partial) block */
+    k8 = (const uint8_t *)k;
+    switch(length)
+    {
+    case 12: c+=k[4]+(((uint32_t)k[5])<<16);
+             b+=k[2]+(((uint32_t)k[3])<<16);
+             a+=k[0]+(((uint32_t)k[1])<<16);
+             break;
+    case 11: c+=((uint32_t)k8[10])<<16;     /* fall through */
+    case 10: c+=k[4];
+             b+=k[2]+(((uint32_t)k[3])<<16);
+             a+=k[0]+(((uint32_t)k[1])<<16);
+             break;
+    case 9 : c+=k8[8];                      /* fall through */
+    case 8 : b+=k[2]+(((uint32_t)k[3])<<16);
+             a+=k[0]+(((uint32_t)k[1])<<16);
+             break;
+    case 7 : b+=((uint32_t)k8[6])<<16;      /* fall through */
+    case 6 : b+=k[2];
+             a+=k[0]+(((uint32_t)k[1])<<16);
+             break;
+    case 5 : b+=k8[4];                      /* fall through */
+    case 4 : a+=k[0]+(((uint32_t)k[1])<<16);
+             break;
+    case 3 : a+=((uint32_t)k8[2])<<16;      /* fall through */
+    case 2 : a+=k[0];
+             break;
+    case 1 : a+=k8[0];
+             break;
+    case 0 : return c;                     /* zero length requires no mixing */
+    }
+
+  } else {                        /* need to read the key one byte at a time */
+    const uint8_t *k = (const uint8_t *)key;
+
+    /*--------------- all but the last block: affect some 32 bits of (a,b,c) */
+    while (length > 12)
+    {
+      a += k[0];
+      a += ((uint32_t)k[1])<<8;
+      a += ((uint32_t)k[2])<<16;
+      a += ((uint32_t)k[3])<<24;
+      b += k[4];
+      b += ((uint32_t)k[5])<<8;
+      b += ((uint32_t)k[6])<<16;
+      b += ((uint32_t)k[7])<<24;
+      c += k[8];
+      c += ((uint32_t)k[9])<<8;
+      c += ((uint32_t)k[10])<<16;
+      c += ((uint32_t)k[11])<<24;
+      mix(a,b,c);
+      length -= 12;
+      k += 12;
+    }
+
+    /*-------------------------------- last block: affect all 32 bits of (c) */
+    switch(length)                   /* all the case statements fall through */
+    {
+    case 12: c+=((uint32_t)k[11])<<24;
+    case 11: c+=((uint32_t)k[10])<<16;
+    case 10: c+=((uint32_t)k[9])<<8;
+    case 9 : c+=k[8];
+    case 8 : b+=((uint32_t)k[7])<<24;
+    case 7 : b+=((uint32_t)k[6])<<16;
+    case 6 : b+=((uint32_t)k[5])<<8;
+    case 5 : b+=k[4];
+    case 4 : a+=((uint32_t)k[3])<<24;
+    case 3 : a+=((uint32_t)k[2])<<16;
+    case 2 : a+=((uint32_t)k[1])<<8;
+    case 1 : a+=k[0];
+             break;
+    case 0 : return c;
+    }
+  }
+
+  final(a,b,c);
+  return c;
+}
+
+
+/*
+ * hashlittle2: return 2 32-bit hash values
+ *
+ * This is identical to hashlittle(), except it returns two 32-bit hash
+ * values instead of just one.  This is good enough for hash table
+ * lookup with 2^^64 buckets, or if you want a second hash if you're not
+ * happy with the first, or if you want a probably-unique 64-bit ID for
+ * the key.  *pc is better mixed than *pb, so use *pc first.  If you want
+ * a 64-bit value do something like "*pc + (((uint64_t)*pb)<<32)".
+ */
+void jenkins_hashlittle2(
+  const void *key,       /* the key to hash */
+  size_t      length,    /* length of the key */
+  uint32_t   *pc,        /* IN: primary initval, OUT: primary hash */
+  uint32_t   *pb)        /* IN: secondary initval, OUT: secondary hash */
+{
+  uint32_t a,b,c;                                          /* internal state */
+  union { const void *ptr; size_t i; } u;     /* needed for Mac Powerbook G4 */
+
+  /* Set up the internal state */
+  a = b = c = 0xdeadbeef + ((uint32_t)length) + *pc;
+  c += *pb;
+
+  u.ptr = key;
+  if (HASH_LITTLE_ENDIAN && ((u.i & 0x3) == 0)) {
+    const uint32_t *k = (const uint32_t *)key;         /* read 32-bit chunks */
+
+    /*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */
+    while (length > 12)
+    {
+      a += k[0];
+      b += k[1];
+      c += k[2];
+      mix(a,b,c);
+      length -= 12;
+      k += 3;
+    }
+
+    /*----------------------------- handle the last (probably partial) block */
+    /*
+     * "k[2]&0xffffff" actually reads beyond the end of the string, but
+     * then masks off the part it's not allowed to read.  Because the
+     * string is aligned, the masked-off tail is in the same word as the
+     * rest of the string.  Every machine with memory protection I've seen
+     * does it on word boundaries, so is OK with this.  But VALGRIND will
+     * still catch it and complain.  The masking trick does make the hash
+     * noticably faster for short strings (like English words).
+     */
+#ifndef VALGRIND
+
+    switch(length)
+    {
+    case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
+    case 11: c+=k[2]&0xffffff; b+=k[1]; a+=k[0]; break;
+    case 10: c+=k[2]&0xffff; b+=k[1]; a+=k[0]; break;
+    case 9 : c+=k[2]&0xff; b+=k[1]; a+=k[0]; break;
+    case 8 : b+=k[1]; a+=k[0]; break;
+    case 7 : b+=k[1]&0xffffff; a+=k[0]; break;
+    case 6 : b+=k[1]&0xffff; a+=k[0]; break;
+    case 5 : b+=k[1]&0xff; a+=k[0]; break;
+    case 4 : a+=k[0]; break;
+    case 3 : a+=k[0]&0xffffff; break;
+    case 2 : a+=k[0]&0xffff; break;
+    case 1 : a+=k[0]&0xff; break;
+    case 0 : *pc=c; *pb=b; return;  /* zero length strings require no mixing */
+    }
+
+#else /* make valgrind happy */
+
+    k8 = (const uint8_t *)k;
+    switch(length)
+    {
+    case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
+    case 11: c+=((uint32_t)k8[10])<<16;  /* fall through */
+    case 10: c+=((uint32_t)k8[9])<<8;    /* fall through */
+    case 9 : c+=k8[8];                   /* fall through */
+    case 8 : b+=k[1]; a+=k[0]; break;
+    case 7 : b+=((uint32_t)k8[6])<<16;   /* fall through */
+    case 6 : b+=((uint32_t)k8[5])<<8;    /* fall through */
+    case 5 : b+=k8[4];                   /* fall through */
+    case 4 : a+=k[0]; break;
+    case 3 : a+=((uint32_t)k8[2])<<16;   /* fall through */
+    case 2 : a+=((uint32_t)k8[1])<<8;    /* fall through */
+    case 1 : a+=k8[0]; break;
+    case 0 : *pc=c; *pb=b; return;  /* zero length strings require no mixing */
+    }
+
+#endif /* !valgrind */
+
+  } else if (HASH_LITTLE_ENDIAN && ((u.i & 0x1) == 0)) {
+    const uint16_t *k = (const uint16_t *)key;         /* read 16-bit chunks */
+    const uint8_t  *k8;
+
+    /*--------------- all but last block: aligned reads and different mixing */
+    while (length > 12)
+    {
+      a += k[0] + (((uint32_t)k[1])<<16);
+      b += k[2] + (((uint32_t)k[3])<<16);
+      c += k[4] + (((uint32_t)k[5])<<16);
+      mix(a,b,c);
+      length -= 12;
+      k += 6;
+    }
+
+    /*----------------------------- handle the last (probably partial) block */
+    k8 = (const uint8_t *)k;
+    switch(length)
+    {
+    case 12: c+=k[4]+(((uint32_t)k[5])<<16);
+             b+=k[2]+(((uint32_t)k[3])<<16);
+             a+=k[0]+(((uint32_t)k[1])<<16);
+             break;
+    case 11: c+=((uint32_t)k8[10])<<16;     /* fall through */
+    case 10: c+=k[4];
+             b+=k[2]+(((uint32_t)k[3])<<16);
+             a+=k[0]+(((uint32_t)k[1])<<16);
+             break;
+    case 9 : c+=k8[8];                      /* fall through */
+    case 8 : b+=k[2]+(((uint32_t)k[3])<<16);
+             a+=k[0]+(((uint32_t)k[1])<<16);
+             break;
+    case 7 : b+=((uint32_t)k8[6])<<16;      /* fall through */
+    case 6 : b+=k[2];
+             a+=k[0]+(((uint32_t)k[1])<<16);
+             break;
+    case 5 : b+=k8[4];                      /* fall through */
+    case 4 : a+=k[0]+(((uint32_t)k[1])<<16);
+             break;
+    case 3 : a+=((uint32_t)k8[2])<<16;      /* fall through */
+    case 2 : a+=k[0];
+             break;
+    case 1 : a+=k8[0];
+             break;
+    case 0 : *pc=c; *pb=b; return;  /* zero length strings require no mixing */
+    }
+
+  } else {                        /* need to read the key one byte at a time */
+    const uint8_t *k = (const uint8_t *)key;
+
+    /*--------------- all but the last block: affect some 32 bits of (a,b,c) */
+    while (length > 12)
+    {
+      a += k[0];
+      a += ((uint32_t)k[1])<<8;
+      a += ((uint32_t)k[2])<<16;
+      a += ((uint32_t)k[3])<<24;
+      b += k[4];
+      b += ((uint32_t)k[5])<<8;
+      b += ((uint32_t)k[6])<<16;
+      b += ((uint32_t)k[7])<<24;
+      c += k[8];
+      c += ((uint32_t)k[9])<<8;
+      c += ((uint32_t)k[10])<<16;
+      c += ((uint32_t)k[11])<<24;
+      mix(a,b,c);
+      length -= 12;
+      k += 12;
+    }
+
+    /*-------------------------------- last block: affect all 32 bits of (c) */
+    switch(length)                   /* all the case statements fall through */
+    {
+    case 12: c+=((uint32_t)k[11])<<24;
+    case 11: c+=((uint32_t)k[10])<<16;
+    case 10: c+=((uint32_t)k[9])<<8;
+    case 9 : c+=k[8];
+    case 8 : b+=((uint32_t)k[7])<<24;
+    case 7 : b+=((uint32_t)k[6])<<16;
+    case 6 : b+=((uint32_t)k[5])<<8;
+    case 5 : b+=k[4];
+    case 4 : a+=((uint32_t)k[3])<<24;
+    case 3 : a+=((uint32_t)k[2])<<16;
+    case 2 : a+=((uint32_t)k[1])<<8;
+    case 1 : a+=k[0];
+             break;
+    case 0 : *pc=c; *pb=b; return;  /* zero length strings require no mixing */
+    }
+  }
+
+  final(a,b,c);
+  *pc=c; *pb=b;
+}
+
+
+
+/*
+ * hashbig():
+ * This is the same as hashword() on big-endian machines.  It is different
+ * from hashlittle() on all machines.  hashbig() takes advantage of
+ * big-endian byte ordering.
+ */
+uint32_t jenkins_hashbig( const void *key, size_t length, uint32_t initval)
+{
+  uint32_t a,b,c;
+  union { const void *ptr; size_t i; } u; /* to cast key to (size_t) happily */
+
+  /* Set up the internal state */
+  a = b = c = 0xdeadbeef + ((uint32_t)length) + initval;
+
+  u.ptr = key;
+  if (HASH_BIG_ENDIAN && ((u.i & 0x3) == 0)) {
+    const uint32_t *k = (const uint32_t *)key;         /* read 32-bit chunks */
+
+    /*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */
+    while (length > 12)
+    {
+      a += k[0];
+      b += k[1];
+      c += k[2];
+      mix(a,b,c);
+      length -= 12;
+      k += 3;
+    }
+
+    /*----------------------------- handle the last (probably partial) block */
+    /*
+     * "k[2]<<8" actually reads beyond the end of the string, but
+     * then shifts out the part it's not allowed to read.  Because the
+     * string is aligned, the illegal read is in the same word as the
+     * rest of the string.  Every machine with memory protection I've seen
+     * does it on word boundaries, so is OK with this.  But VALGRIND will
+     * still catch it and complain.  The masking trick does make the hash
+     * noticably faster for short strings (like English words).
+     */
+#ifndef VALGRIND
+
+    switch(length)
+    {
+    case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
+    case 11: c+=k[2]&0xffffff00; b+=k[1]; a+=k[0]; break;
+    case 10: c+=k[2]&0xffff0000; b+=k[1]; a+=k[0]; break;
+    case 9 : c+=k[2]&0xff000000; b+=k[1]; a+=k[0]; break;
+    case 8 : b+=k[1]; a+=k[0]; break;
+    case 7 : b+=k[1]&0xffffff00; a+=k[0]; break;
+    case 6 : b+=k[1]&0xffff0000; a+=k[0]; break;
+    case 5 : b+=k[1]&0xff000000; a+=k[0]; break;
+    case 4 : a+=k[0]; break;
+    case 3 : a+=k[0]&0xffffff00; break;
+    case 2 : a+=k[0]&0xffff0000; break;
+    case 1 : a+=k[0]&0xff000000; break;
+    case 0 : return c;              /* zero length strings require no mixing */
+    }
+
+#else  /* make valgrind happy */
+
+    k8 = (const uint8_t *)k;
+    switch(length)                   /* all the case statements fall through */
+    {
+    case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
+    case 11: c+=((uint32_t)k8[10])<<8;  /* fall through */
+    case 10: c+=((uint32_t)k8[9])<<16;  /* fall through */
+    case 9 : c+=((uint32_t)k8[8])<<24;  /* fall through */
+    case 8 : b+=k[1]; a+=k[0]; break;
+    case 7 : b+=((uint32_t)k8[6])<<8;   /* fall through */
+    case 6 : b+=((uint32_t)k8[5])<<16;  /* fall through */
+    case 5 : b+=((uint32_t)k8[4])<<24;  /* fall through */
+    case 4 : a+=k[0]; break;
+    case 3 : a+=((uint32_t)k8[2])<<8;   /* fall through */
+    case 2 : a+=((uint32_t)k8[1])<<16;  /* fall through */
+    case 1 : a+=((uint32_t)k8[0])<<24; break;
+    case 0 : return c;
+    }
+
+#endif /* !VALGRIND */
+
+  } else {                        /* need to read the key one byte at a time */
+    const uint8_t *k = (const uint8_t *)key;
+
+    /*--------------- all but the last block: affect some 32 bits of (a,b,c) */
+    while (length > 12)
+    {
+      a += ((uint32_t)k[0])<<24;
+      a += ((uint32_t)k[1])<<16;
+      a += ((uint32_t)k[2])<<8;
+      a += ((uint32_t)k[3]);
+      b += ((uint32_t)k[4])<<24;
+      b += ((uint32_t)k[5])<<16;
+      b += ((uint32_t)k[6])<<8;
+      b += ((uint32_t)k[7]);
+      c += ((uint32_t)k[8])<<24;
+      c += ((uint32_t)k[9])<<16;
+      c += ((uint32_t)k[10])<<8;
+      c += ((uint32_t)k[11]);
+      mix(a,b,c);
+      length -= 12;
+      k += 12;
+    }
+
+    /*-------------------------------- last block: affect all 32 bits of (c) */
+    switch(length)                   /* all the case statements fall through */
+    {
+    case 12: c+=k[11];
+    case 11: c+=((uint32_t)k[10])<<8;
+    case 10: c+=((uint32_t)k[9])<<16;
+    case 9 : c+=((uint32_t)k[8])<<24;
+    case 8 : b+=k[7];
+    case 7 : b+=((uint32_t)k[6])<<8;
+    case 6 : b+=((uint32_t)k[5])<<16;
+    case 5 : b+=((uint32_t)k[4])<<24;
+    case 4 : a+=k[3];
+    case 3 : a+=((uint32_t)k[2])<<8;
+    case 2 : a+=((uint32_t)k[1])<<16;
+    case 1 : a+=((uint32_t)k[0])<<24;
+             break;
+    case 0 : return c;
+    }
+  }
+
+  final(a,b,c);
+  return c;
+}
+
+
+#ifdef SELF_TEST
+
+/* used for timings */
+void driver1()
+{
+  uint8_t buf[256];
+  uint32_t i;
+  uint32_t h=0;
+  time_t a,z;
+
+  time(&a);
+  for (i=0; i<256; ++i) buf[i] = 'x';
+  for (i=0; i<1; ++i)
+  {
+    h = hashlittle(&buf[0],1,h);
+  }
+  time(&z);
+  if (z-a > 0) printf("time %d %.8x\n", z-a, h);
+}
+
+/* check that every input bit changes every output bit half the time */
+#define HASHSTATE 1
+#define HASHLEN   1
+#define MAXPAIR 60
+#define MAXLEN  70
+void driver2()
+{
+  uint8_t qa[MAXLEN+1], qb[MAXLEN+2], *a = &qa[0], *b = &qb[1];
+  uint32_t c[HASHSTATE], d[HASHSTATE], i=0, j=0, k, l, m=0, z;
+  uint32_t e[HASHSTATE],f[HASHSTATE],g[HASHSTATE],h[HASHSTATE];
+  uint32_t x[HASHSTATE],y[HASHSTATE];
+  uint32_t hlen;
+
+  printf("No more than %d trials should ever be needed \n",MAXPAIR/2);
+  for (hlen=0; hlen < MAXLEN; ++hlen)
+  {
+    z=0;
+    for (i=0; i<hlen; ++i)  /*----------------------- for each input byte, */
+    {
+      for (j=0; j<8; ++j)   /*------------------------ for each input bit, */
+      {
+        for (m=1; m<8; ++m) /*------------ for serveral possible initvals, */
+        {
+          for (l=0; l<HASHSTATE; ++l)
+            e[l]=f[l]=g[l]=h[l]=x[l]=y[l]=~((uint32_t)0);
+
+          /*---- check that every output bit is affected by that input bit */
+          for (k=0; k<MAXPAIR; k+=2)
+          {
+            uint32_t finished=1;
+            /* keys have one bit different */
+            for (l=0; l<hlen+1; ++l) {a[l] = b[l] = (uint8_t)0;}
+            /* have a and b be two keys differing in only one bit */
+            a[i] ^= (k<<j);
+            a[i] ^= (k>>(8-j));
+             c[0] = hashlittle(a, hlen, m);
+            b[i] ^= ((k+1)<<j);
+            b[i] ^= ((k+1)>>(8-j));
+             d[0] = hashlittle(b, hlen, m);
+            /* check every bit is 1, 0, set, and not set at least once */
+            for (l=0; l<HASHSTATE; ++l)
+            {
+              e[l] &= (c[l]^d[l]);
+              f[l] &= ~(c[l]^d[l]);
+              g[l] &= c[l];
+              h[l] &= ~c[l];
+              x[l] &= d[l];
+              y[l] &= ~d[l];
+              if (e[l]|f[l]|g[l]|h[l]|x[l]|y[l]) finished=0;
+            }
+            if (finished) break;
+          }
+          if (k>z) z=k;
+          if (k==MAXPAIR)
+          {
+             printf("Some bit didn't change: ");
+             printf("%.8x %.8x %.8x %.8x %.8x %.8x  ",
+                    e[0],f[0],g[0],h[0],x[0],y[0]);
+             printf("i %d j %d m %d len %d\n", i, j, m, hlen);
+          }
+          if (z==MAXPAIR) goto done;
+        }
+      }
+    }
+   done:
+    if (z < MAXPAIR)
+    {
+      printf("Mix success  %2d bytes  %2d initvals  ",i,m);
+      printf("required  %d  trials\n", z/2);
+    }
+  }
+  printf("\n");
+}
+
+/* Check for reading beyond the end of the buffer and alignment problems */
+void driver3()
+{
+  uint8_t buf[MAXLEN+20], *b;
+  uint32_t len;
+  uint8_t q[] = "This is the time for all good men to come to the aid of their country...";
+  uint32_t h;
+  uint8_t qq[] = "xThis is the time for all good men to come to the aid of their country...";
+  uint32_t i;
+  uint8_t qqq[] = "xxThis is the time for all good men to come to the aid of their country...";
+  uint32_t j;
+  uint8_t qqqq[] = "xxxThis is the time for all good men to come to the aid of their country...";
+  uint32_t ref,x,y;
+  uint8_t *p;
+
+  printf("Endianness.  These lines should all be the same (for values filled in):\n");
+  printf("%.8x                            %.8x                            %.8x\n",
+         hashword((const uint32_t *)q, (sizeof(q)-1)/4, 13),
+         hashword((const uint32_t *)q, (sizeof(q)-5)/4, 13),
+         hashword((const uint32_t *)q, (sizeof(q)-9)/4, 13));
+  p = q;
+  printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n",
+         hashlittle(p, sizeof(q)-1, 13), hashlittle(p, sizeof(q)-2, 13),
+         hashlittle(p, sizeof(q)-3, 13), hashlittle(p, sizeof(q)-4, 13),
+         hashlittle(p, sizeof(q)-5, 13), hashlittle(p, sizeof(q)-6, 13),
+         hashlittle(p, sizeof(q)-7, 13), hashlittle(p, sizeof(q)-8, 13),
+         hashlittle(p, sizeof(q)-9, 13), hashlittle(p, sizeof(q)-10, 13),
+         hashlittle(p, sizeof(q)-11, 13), hashlittle(p, sizeof(q)-12, 13));
+  p = &qq[1];
+  printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n",
+         hashlittle(p, sizeof(q)-1, 13), hashlittle(p, sizeof(q)-2, 13),
+         hashlittle(p, sizeof(q)-3, 13), hashlittle(p, sizeof(q)-4, 13),
+         hashlittle(p, sizeof(q)-5, 13), hashlittle(p, sizeof(q)-6, 13),
+         hashlittle(p, sizeof(q)-7, 13), hashlittle(p, sizeof(q)-8, 13),
+         hashlittle(p, sizeof(q)-9, 13), hashlittle(p, sizeof(q)-10, 13),
+         hashlittle(p, sizeof(q)-11, 13), hashlittle(p, sizeof(q)-12, 13));
+  p = &qqq[2];
+  printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n",
+         hashlittle(p, sizeof(q)-1, 13), hashlittle(p, sizeof(q)-2, 13),
+         hashlittle(p, sizeof(q)-3, 13), hashlittle(p, sizeof(q)-4, 13),
+         hashlittle(p, sizeof(q)-5, 13), hashlittle(p, sizeof(q)-6, 13),
+         hashlittle(p, sizeof(q)-7, 13), hashlittle(p, sizeof(q)-8, 13),
+         hashlittle(p, sizeof(q)-9, 13), hashlittle(p, sizeof(q)-10, 13),
+         hashlittle(p, sizeof(q)-11, 13), hashlittle(p, sizeof(q)-12, 13));
+  p = &qqqq[3];
+  printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n",
+         hashlittle(p, sizeof(q)-1, 13), hashlittle(p, sizeof(q)-2, 13),
+         hashlittle(p, sizeof(q)-3, 13), hashlittle(p, sizeof(q)-4, 13),
+         hashlittle(p, sizeof(q)-5, 13), hashlittle(p, sizeof(q)-6, 13),
+         hashlittle(p, sizeof(q)-7, 13), hashlittle(p, sizeof(q)-8, 13),
+         hashlittle(p, sizeof(q)-9, 13), hashlittle(p, sizeof(q)-10, 13),
+         hashlittle(p, sizeof(q)-11, 13), hashlittle(p, sizeof(q)-12, 13));
+  printf("\n");
+
+  /* check that hashlittle2 and hashlittle produce the same results */
+  i=47; j=0;
+  hashlittle2(q, sizeof(q), &i, &j);
+  if (hashlittle(q, sizeof(q), 47) != i)
+    printf("hashlittle2 and hashlittle mismatch\n");
+
+  /* check that hashword2 and hashword produce the same results */
+  len = 0xdeadbeef;
+  i=47, j=0;
+  hashword2(&len, 1, &i, &j);
+  if (hashword(&len, 1, 47) != i)
+    printf("hashword2 and hashword mismatch %x %x\n",
+           i, hashword(&len, 1, 47));
+
+  /* check hashlittle doesn't read before or after the ends of the string */
+  for (h=0, b=buf+1; h<8; ++h, ++b)
+  {
+    for (i=0; i<MAXLEN; ++i)
+    {
+      len = i;
+      for (j=0; j<i; ++j) *(b+j)=0;
+
+      /* these should all be equal */
+      ref = hashlittle(b, len, (uint32_t)1);
+      *(b+i)=(uint8_t)~0;
+      *(b-1)=(uint8_t)~0;
+      x = hashlittle(b, len, (uint32_t)1);
+      y = hashlittle(b, len, (uint32_t)1);
+      if ((ref != x) || (ref != y))
+      {
+        printf("alignment error: %.8x %.8x %.8x %d %d\n",ref,x,y,
+               h, i);
+      }
+    }
+  }
+}
+
+/* check for problems with nulls */
+ void driver4()
+{
+  uint8_t buf[1];
+  uint32_t h,i,state[HASHSTATE];
+
+
+  buf[0] = ~0;
+  for (i=0; i<HASHSTATE; ++i) state[i] = 1;
+  printf("These should all be different\n");
+  for (i=0, h=0; i<8; ++i)
+  {
+    h = hashlittle(buf, 0, h);
+    printf("%2ld  0-byte strings, hash is  %.8x\n", i, h);
+  }
+}
+
+void driver5()
+{
+  uint32_t b,c;
+  b=0, c=0, hashlittle2("", 0, &c, &b);
+  printf("hash is %.8lx %.8lx\n", c, b);   /* deadbeef deadbeef */
+  b=0xdeadbeef, c=0, hashlittle2("", 0, &c, &b);
+  printf("hash is %.8lx %.8lx\n", c, b);   /* bd5b7dde deadbeef */
+  b=0xdeadbeef, c=0xdeadbeef, hashlittle2("", 0, &c, &b);
+  printf("hash is %.8lx %.8lx\n", c, b);   /* 9c093ccd bd5b7dde */
+  b=0, c=0, hashlittle2("Four score and seven years ago", 30, &c, &b);
+  printf("hash is %.8lx %.8lx\n", c, b);   /* 17770551 ce7226e6 */
+  b=1, c=0, hashlittle2("Four score and seven years ago", 30, &c, &b);
+  printf("hash is %.8lx %.8lx\n", c, b);   /* e3607cae bd371de4 */
+  b=0, c=1, hashlittle2("Four score and seven years ago", 30, &c, &b);
+  printf("hash is %.8lx %.8lx\n", c, b);   /* cd628161 6cbea4b3 */
+  c = hashlittle("Four score and seven years ago", 30, 0);
+  printf("hash is %.8lx\n", c);   /* 17770551 */
+  c = hashlittle("Four score and seven years ago", 30, 1);
+  printf("hash is %.8lx\n", c);   /* cd628161 */
+}
+
+
+int main()
+{
+  driver1();   /* test that the key is hashed: used for timings */
+  driver2();   /* test that whole key is hashed thoroughly */
+  driver3();   /* test that nothing but the key is hashed */
+  driver4();   /* test hashing multiple buffers (all buffers are null) */
+  driver5();   /* test the hash against known vectors */
+  return 1;
+}
+
+#endif  /* SELF_TEST */
diff --git a/src/journal/lookup3.h b/src/journal/lookup3.h
new file mode 100644 (file)
index 0000000..31cc2f5
--- /dev/null
@@ -0,0 +1,25 @@
+/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
+
+#ifndef foolookup3hfoo
+#define foolookup3hfoo
+
+#include <inttypes.h>
+#include <sys/types.h>
+
+uint32_t jenkins_hashword(const uint32_t *k, size_t length, uint32_t initval);
+void jenkins_hashword2(const uint32_t *k, size_t length, uint32_t *pc, uint32_t *pb);
+
+uint32_t jenkins_hashlittle(const void *key, size_t length, uint32_t initval);
+void jenkins_hashlittle2(const void *key, size_t length, uint32_t *pc, uint32_t *pb);
+
+uint32_t jenkins_hashbig(const void *key, size_t length, uint32_t initval);
+
+static inline uint64_t hash64(const void *data, size_t length) {
+        uint32_t a = 0, b = 0;
+
+        jenkins_hashlittle2(data, length, &a, &b);
+
+        return ((uint64_t) a << 32ULL) | (uint64_t) b;
+}
+
+#endif
diff --git a/src/journal/sd-journal.c b/src/journal/sd-journal.c
new file mode 100644 (file)
index 0000000..85c5717
--- /dev/null
@@ -0,0 +1,1271 @@
+/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
+
+/***
+  This file is part of systemd.
+
+  Copyright 2011 Lennart Poettering
+
+  systemd is free software; you can redistribute it and/or modify it
+  under the terms of the GNU General Public License as published by
+  the Free Software Foundation; either version 2 of the License, or
+  (at your option) any later version.
+
+  systemd is distributed in the hope that it will be useful, but
+  WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+  General Public License for more details.
+
+  You should have received a copy of the GNU General Public License
+  along with systemd; If not, see <http://www.gnu.org/licenses/>.
+***/
+
+#include <errno.h>
+#include <fcntl.h>
+#include <stddef.h>
+
+#include "sd-journal.h"
+#include "journal-def.h"
+#include "journal-file.h"
+#include "hashmap.h"
+#include "list.h"
+#include "lookup3.h"
+
+typedef struct Match Match;
+
+struct Match {
+        char *data;
+        size_t size;
+        uint64_t le_hash;
+
+        LIST_FIELDS(Match, matches);
+};
+
+typedef enum location_type {
+        LOCATION_HEAD,
+        LOCATION_TAIL,
+        LOCATION_DISCRETE
+} location_type_t;
+
+typedef struct Location {
+        location_type_t type;
+
+        uint64_t seqnum;
+        sd_id128_t seqnum_id;
+        bool seqnum_set;
+
+        uint64_t realtime;
+        bool realtime_set;
+
+        uint64_t monotonic;
+        sd_id128_t boot_id;
+        bool monotonic_set;
+
+        uint64_t xor_hash;
+        bool xor_hash_set;
+} Location;
+
+struct sd_journal {
+        Hashmap *files;
+
+        Location current_location;
+        JournalFile *current_file;
+        uint64_t current_field;
+
+        LIST_HEAD(Match, matches);
+        unsigned n_matches;
+};
+
+static void detach_location(sd_journal *j) {
+        Iterator i;
+        JournalFile *f;
+
+        assert(j);
+
+        j->current_file = NULL;
+        j->current_field = 0;
+
+        HASHMAP_FOREACH(f, j->files, i)
+                f->current_offset = 0;
+}
+
+static void reset_location(sd_journal *j) {
+        assert(j);
+
+        detach_location(j);
+        zero(j->current_location);
+}
+
+static void init_location(Location *l, JournalFile *f, Object *o) {
+        assert(l);
+        assert(f);
+        assert(o->object.type == OBJECT_ENTRY);
+
+        l->type = LOCATION_DISCRETE;
+        l->seqnum = le64toh(o->entry.seqnum);
+        l->seqnum_id = f->header->seqnum_id;
+        l->realtime = le64toh(o->entry.realtime);
+        l->monotonic = le64toh(o->entry.monotonic);
+        l->boot_id = le64toh(o->entry.boot_id);
+        l->xor_hash = le64toh(o->entry.xor_hash);
+
+        l->seqnum_set = l->realtime_set = l->monotonic_set = l->xor_hash_set = true;
+}
+
+static void set_location(sd_journal *j, JournalFile *f, Object *o, uint64_t offset) {
+        assert(j);
+        assert(f);
+        assert(o);
+
+        init_location(&j->current_location, f, o);
+
+        j->current_file = f;
+        j->current_field = 0;
+
+        f->current_offset = offset;
+}
+
+static int same_field(const void *_a, size_t s, const void *_b, size_t t) {
+        const uint8_t *a = _a, *b = _b;
+        size_t j;
+        bool a_good = false, b_good = false, different = false;
+
+        for (j = 0; j < s && j < t; j++) {
+
+                if (a[j] == '=')
+                        a_good = true;
+                if (b[j] == '=')
+                        b_good = true;
+                if (a[j] != b[j])
+                        different = true;
+
+                if (a_good && b_good)
+                        return different ? 0 : 1;
+        }
+
+        return -EINVAL;
+}
+
+int sd_journal_add_match(sd_journal *j, const void *data, size_t size) {
+        Match *m, *after = NULL;
+        uint64_t le_hash;
+
+        assert(j);
+
+        if (size <= 0)
+                return -EINVAL;
+
+        assert(data);
+
+        le_hash = htole64(hash64(data, size));
+
+        LIST_FOREACH(matches, m, j->matches) {
+                int r;
+
+                if (m->le_hash == le_hash &&
+                    m->size == size &&
+                    memcmp(m->data, data, size) == 0)
+                        return 0;
+
+                r = same_field(data, size, m->data, m->size);
+                if (r < 0)
+                        return r;
+                else if (r > 0)
+                        after = m;
+        }
+
+        m = new0(Match, 1);
+        if (!m)
+                return -ENOMEM;
+
+        m->size = size;
+
+        m->data = malloc(m->size);
+        if (!m->data) {
+                free(m);
+                return -ENOMEM;
+        }
+
+        memcpy(m->data, data, size);
+        m->le_hash = le_hash;
+
+        /* Matches for the same fields we order adjacent to each
+         * other */
+        LIST_INSERT_AFTER(Match, matches, j->matches, after, m);
+        j->n_matches ++;
+
+        detach_location(j);
+
+        return 0;
+}
+
+void sd_journal_flush_matches(sd_journal *j) {
+        assert(j);
+
+        while (j->matches) {
+                Match *m = j->matches;
+
+                LIST_REMOVE(Match, matches, j->matches, m);
+                free(m->data);
+                free(m);
+        }
+
+        j->n_matches = 0;
+
+        detach_location(j);
+}
+
+static int compare_order(JournalFile *af, Object *ao,
+                         JournalFile *bf, Object *bo) {
+
+        uint64_t a, b;
+
+        assert(af);
+        assert(ao);
+        assert(bf);
+        assert(bo);
+
+        /* We operate on two different files here, hence we can access
+         * two objects at the same time, which we normally can't.
+         *
+         * If contents and timestamps match, these entries are
+         * identical, even if the seqnum does not match */
+
+        if (sd_id128_equal(ao->entry.boot_id, bo->entry.boot_id) &&
+            ao->entry.monotonic == bo->entry.monotonic &&
+            ao->entry.realtime == bo->entry.realtime &&
+            ao->entry.xor_hash == bo->entry.xor_hash)
+                return 0;
+
+        if (sd_id128_equal(af->header->seqnum_id, bf->header->seqnum_id)) {
+
+                /* If this is from the same seqnum source, compare
+                 * seqnums */
+                a = le64toh(ao->entry.seqnum);
+                b = le64toh(bo->entry.seqnum);
+
+                if (a < b)
+                        return -1;
+                if (a > b)
+                        return 1;
+
+                /* Wow! This is weird, different data but the same
+                 * seqnums? Something is borked, but let's make the
+                 * best of it and compare by time. */
+        }
+
+        if (sd_id128_equal(ao->entry.boot_id, bo->entry.boot_id)) {
+
+                /* If the boot id matches compare monotonic time */
+                a = le64toh(ao->entry.monotonic);
+                b = le64toh(bo->entry.monotonic);
+
+                if (a < b)
+                        return -1;
+                if (a > b)
+                        return 1;
+        }
+
+        /* Otherwise compare UTC time */
+        a = le64toh(ao->entry.realtime);
+        b = le64toh(ao->entry.realtime);
+
+        if (a < b)
+                return -1;
+        if (a > b)
+                return 1;
+
+        /* Finally, compare by contents */
+        a = le64toh(ao->entry.xor_hash);
+        b = le64toh(ao->entry.xor_hash);
+
+        if (a < b)
+                return -1;
+        if (a > b)
+                return 1;
+
+        return 0;
+}
+
+static int compare_with_location(JournalFile *af, Object *ao, Location *l) {
+        uint64_t a;
+
+        assert(af);
+        assert(ao);
+        assert(l);
+        assert(l->type == LOCATION_DISCRETE);
+
+        if (l->monotonic_set &&
+            sd_id128_equal(ao->entry.boot_id, l->boot_id) &&
+            l->realtime_set &&
+            le64toh(ao->entry.realtime) == l->realtime &&
+            l->xor_hash_set &&
+            le64toh(ao->entry.xor_hash) == l->xor_hash)
+                return 0;
+
+        if (l->seqnum_set &&
+            sd_id128_equal(af->header->seqnum_id, l->seqnum_id)) {
+
+                a = le64toh(ao->entry.seqnum);
+
+                if (a < l->seqnum)
+                        return -1;
+                if (a > l->seqnum)
+                        return 1;
+        }
+
+        if (l->monotonic_set &&
+            sd_id128_equal(ao->entry.boot_id, l->boot_id)) {
+
+                a = le64toh(ao->entry.monotonic);
+
+                if (a < l->monotonic)
+                        return -1;
+                if (a > l->monotonic)
+                        return 1;
+        }
+
+        if (l->realtime_set) {
+
+                a = le64toh(ao->entry.realtime);
+
+                if (a < l->realtime)
+                        return -1;
+                if (a > l->realtime)
+                        return 1;
+        }
+
+        if (l->xor_hash_set) {
+                a = le64toh(ao->entry.xor_hash);
+
+                if (a < l->xor_hash)
+                        return -1;
+                if (a > l->xor_hash)
+                        return 1;
+        }
+
+        return 0;
+}
+
+static int find_location(sd_journal *j, JournalFile *f, direction_t direction, Object **ret, uint64_t *offset) {
+        Object *o = NULL;
+        uint64_t p = 0;
+        int r;
+
+        assert(j);
+
+        if (!j->matches) {
+                /* No matches is simple */
+
+                if (j->current_location.type == LOCATION_HEAD)
+                        r = journal_file_next_entry(f, NULL, 0, DIRECTION_DOWN, &o, &p);
+                else if (j->current_location.type == LOCATION_TAIL)
+                        r = journal_file_next_entry(f, NULL, 0, DIRECTION_UP, &o, &p);
+                else if (j->current_location.seqnum_set &&
+                         sd_id128_equal(j->current_location.seqnum_id, f->header->seqnum_id))
+                        r = journal_file_move_to_entry_by_seqnum(f, j->current_location.seqnum, direction, &o, &p);
+                else if (j->current_location.monotonic_set)
+                        r = journal_file_move_to_entry_by_monotonic(f, j->current_location.boot_id, j->current_location.monotonic, direction, &o, &p);
+                else if (j->current_location.realtime_set)
+                        r = journal_file_move_to_entry_by_realtime(f, j->current_location.realtime, direction, &o, &p);
+                else
+                        r = journal_file_next_entry(f, NULL, 0, direction, &o, &p);
+
+                if (r <= 0)
+                        return r;
+
+        } else  {
+                Match *m, *term_match = NULL;
+                Object *to = NULL;
+                uint64_t tp = 0;
+
+                /* We have matches, first, let's jump to the monotonic
+                 * position if we have any, since it implies a
+                 * match. */
+
+                if (j->current_location.type == LOCATION_DISCRETE &&
+                    j->current_location.monotonic_set) {
+
+                        r = journal_file_move_to_entry_by_monotonic(f, j->current_location.boot_id, j->current_location.monotonic, direction, &o, &p);
+                        if (r <= 0)
+                                return r;
+                }
+
+                LIST_FOREACH(matches, m, j->matches) {
+                        Object *c, *d;
+                        uint64_t cp, dp;
+
+                        r = journal_file_find_data_object_with_hash(f, m->data, m->size, m->le_hash, &d, &dp);
+                        if (r <= 0)
+                                return r;
+
+                        if (j->current_location.type == LOCATION_HEAD)
+                                r = journal_file_next_entry_for_data(f, NULL, 0, dp, DIRECTION_DOWN, &c, &cp);
+                        else if (j->current_location.type == LOCATION_TAIL)
+                                r = journal_file_next_entry_for_data(f, NULL, 0, dp, DIRECTION_UP, &c, &cp);
+                        else if (j->current_location.seqnum_set &&
+                                 sd_id128_equal(j->current_location.seqnum_id, f->header->seqnum_id))
+                                r = journal_file_move_to_entry_by_seqnum_for_data(f, dp, j->current_location.seqnum, direction, &c, &cp);
+                        else if (j->current_location.realtime_set)
+                                r = journal_file_move_to_entry_by_realtime_for_data(f, dp, j->current_location.realtime, direction, &c, &cp);
+                        else
+                                r = journal_file_next_entry_for_data(f, NULL, 0, dp, direction, &c, &cp);
+
+                        if (!term_match) {
+                                term_match = m;
+
+                                if (r > 0) {
+                                        to = c;
+                                        tp = cp;
+                                }
+                        } else if (same_field(term_match->data, term_match->size, m->data, m->size)) {
+
+                                /* Same field as previous match... */
+                                if (r > 0) {
+
+                                        /* Find the earliest of the OR matches */
+
+                                        if (!to ||
+                                            (direction == DIRECTION_DOWN && cp < tp) ||
+                                            (direction == DIRECTION_UP && cp > tp)) {
+                                                to = c;
+                                                tp = tp;
+                                        }
+
+                                }
+
+                        } else {
+
+                                /* Previous term is finished, did anything match? */
+                                if (!to)
+                                        return 0;
+
+                                /* Find the last of the AND matches */
+                                if (!o ||
+                                    (direction == DIRECTION_DOWN && tp > p) ||
+                                    (direction == DIRECTION_UP && tp < p)) {
+                                        o = to;
+                                        p = tp;
+                                }
+
+                                term_match = m;
+
+                                if (r > 0) {
+                                        to = c;
+                                        tp = cp;
+                                } else {
+                                        to = NULL;
+                                        tp = 0;
+                                }
+                        }
+                }
+
+                /* Last term is finished, did anything match? */
+                if (!to)
+                        return 0;
+
+                if (!o ||
+                    (direction == DIRECTION_DOWN && tp > p) ||
+                    (direction == DIRECTION_UP && tp < p)) {
+                        o = to;
+                        p = tp;
+                }
+
+                if (!o)
+                        return 0;
+        }
+
+        if (ret)
+                *ret = o;
+
+        if (offset)
+                *offset = p;
+
+        return 1;
+}
+
+static int next_with_matches(sd_journal *j, JournalFile *f, direction_t direction, Object **ret, uint64_t *offset) {
+        int r;
+        uint64_t cp;
+        Object *c;
+
+        assert(j);
+        assert(f);
+        assert(ret);
+        assert(offset);
+
+        c = *ret;
+        cp = *offset;
+
+        if (!j->matches) {
+                /* No matches is easy */
+
+                r = journal_file_next_entry(f, c, cp, direction, &c, &cp);
+                if (r <= 0)
+                        return r;
+
+                if (ret)
+                        *ret = c;
+                if (offset)
+                        *offset = cp;
+                return 1;
+        }
+
+        /* So there are matches we have to adhere to, let's find the
+         * first entry that matches all of them */
+
+        for (;;) {
+                uint64_t np, n;
+                bool found, term_result = false;
+                Match *m, *term_match = NULL;
+
+                n = journal_file_entry_n_items(c);
+
+                /* Make sure we don't match the entry we are starting
+                 * from. */
+                found = cp > *offset;
+
+                np = 0;
+                LIST_FOREACH(matches, m, j->matches) {
+                        uint64_t q, k;
+
+                        /* Let's check if this is the beginning of a
+                         * new term, i.e. has a different field prefix
+                         * as the preceeding match. */
+                        if (!term_match) {
+                                term_match = m;
+                                term_result = false;
+                        } else if (!same_field(term_match->data, term_match->size, m->data, m->size)) {
+                                if (!term_result)
+                                        found = false;
+
+                                term_match = m;
+                                term_result = false;
+                        }
+
+                        for (k = 0; k < n; k++)
+                                if (c->entry.items[k].hash == m->le_hash)
+                                        break;
+
+                        if (k >= n) {
+                                /* Hmm, didn't find any field that
+                                 * matched this rule, so ignore this
+                                 * match. Go on with next match */
+                                continue;
+                        }
+
+                        term_result = true;
+
+                        /* Hmm, so, this field matched, let's remember
+                         * where we'd have to try next, in case the other
+                         * matches are not OK */
+
+                        r = journal_file_next_entry_for_data(f, c, cp, le64toh(c->entry.items[k].object_offset), direction, NULL, &q);
+                        if (r > 0) {
+
+                                if (direction == DIRECTION_DOWN) {
+                                        if (q > np)
+                                                np = q;
+                                } else {
+                                        if (np == 0 || q < np)
+                                                np = q;
+                                }
+                        }
+                }
+
+                /* Check the last term */
+                if (term_match && term_result)
+                        found = true;
+
+                /* Did this entry match against all matches? */
+                if (found) {
+                        if (ret)
+                                *ret = c;
+                        if (offset)
+                                *offset = cp;
+                        return 1;
+                }
+
+                /* Did we find a subsequent entry? */
+                if (np == 0)
+                        return 0;
+
+                /* Hmm, ok, this entry only matched partially, so
+                 * let's try another one */
+                cp = np;
+        }
+}
+
+static int next_beyond_location(sd_journal *j, JournalFile *f, direction_t direction, Object **ret, uint64_t *offset) {
+        Object *c;
+        uint64_t cp;
+        int compare_value, r;
+
+        assert(j);
+        assert(f);
+
+        if (f->current_offset > 0) {
+                r = journal_file_move_to_object(f, OBJECT_ENTRY, f->current_offset, &c);
+                if (r < 0)
+                        return r;
+
+                cp = f->current_offset;
+
+                r = next_with_matches(j, f, direction, &c, &cp);
+                if (r <= 0)
+                        return r;
+
+                compare_value = 1;
+        } else {
+                r = find_location(j, f, direction, &c, &cp);
+                if (r <= 0)
+                        return r;
+
+                compare_value = 0;
+        }
+
+        for (;;) {
+                bool found;
+
+                if (j->current_location.type == LOCATION_DISCRETE) {
+                        int k;
+
+                        k = compare_with_location(f, c, &j->current_location);
+                        if (direction == DIRECTION_DOWN)
+                                found = k >= compare_value;
+                        else
+                                found = k <= -compare_value;
+                } else
+                        found = true;
+
+                if (found) {
+                        if (ret)
+                                *ret = c;
+                        if (offset)
+                                *offset = cp;
+                        return 1;
+                }
+
+                r = next_with_matches(j, f, direction, &c, &cp);
+                if (r <= 0)
+                        return r;
+        }
+}
+
+static int real_journal_next(sd_journal *j, direction_t direction) {
+        JournalFile *f, *new_current = NULL;
+        Iterator i;
+        int r;
+        uint64_t new_offset = 0;
+        Object *new_entry = NULL;
+
+        assert(j);
+
+        HASHMAP_FOREACH(f, j->files, i) {
+                Object *o;
+                uint64_t p;
+                bool found;
+
+                r = next_beyond_location(j, f, direction, &o, &p);
+                if (r < 0)
+                        return r;
+                else if (r == 0)
+                        continue;
+
+                if (!new_current)
+                        found = true;
+                else {
+                        int k;
+
+                        k = compare_order(f, o, new_current, new_entry);
+
+                        if (direction == DIRECTION_DOWN)
+                                found = k < 0;
+                        else
+                                found = k > 0;
+                }
+
+                if (found) {
+                        new_current = f;
+                        new_entry = o;
+                        new_offset = p;
+                }
+        }
+
+        if (!new_current)
+                return 0;
+
+        set_location(j, new_current, new_entry, new_offset);
+
+        return 1;
+}
+
+int sd_journal_next(sd_journal *j) {
+        return real_journal_next(j, DIRECTION_DOWN);
+}
+
+int sd_journal_previous(sd_journal *j) {
+        return real_journal_next(j, DIRECTION_UP);
+}
+
+int sd_journal_next_skip(sd_journal *j, uint64_t skip) {
+        int c = 0, r;
+
+        assert(j);
+
+        while (skip > 0) {
+                r = sd_journal_next(j);
+                if (r < 0)
+                        return r;
+
+                if (r == 0)
+                        return c;
+
+                skip--;
+                c++;
+        }
+
+        return c;
+}
+
+int sd_journal_previous_skip(sd_journal *j, uint64_t skip) {
+        int c = 0, r;
+
+        assert(j);
+
+        while (skip > 0) {
+                r = sd_journal_previous(j);
+                if (r < 0)
+                        return r;
+
+                if (r == 0)
+                        return c;
+
+                skip--;
+                c++;
+        }
+
+        return 1;
+}
+
+int sd_journal_get_cursor(sd_journal *j, char **cursor) {
+        Object *o;
+        int r;
+        char bid[33], sid[33];
+
+        assert(j);
+        assert(cursor);
+
+        if (!j->current_file || j->current_file->current_offset <= 0)
+                return -EADDRNOTAVAIL;
+
+        r = journal_file_move_to_object(j->current_file, OBJECT_ENTRY, j->current_file->current_offset, &o);
+        if (r < 0)
+                return r;
+
+        sd_id128_to_string(j->current_file->header->seqnum_id, sid);
+        sd_id128_to_string(o->entry.boot_id, bid);
+
+        if (asprintf(cursor,
+                     "s=%s;i=%llx;b=%s;m=%llx;t=%llx;x=%llx;p=%s",
+                     sid, (unsigned long long) le64toh(o->entry.seqnum),
+                     bid, (unsigned long long) le64toh(o->entry.monotonic),
+                     (unsigned long long) le64toh(o->entry.realtime),
+                     (unsigned long long) le64toh(o->entry.xor_hash),
+                     file_name_from_path(j->current_file->path)) < 0)
+                return -ENOMEM;
+
+        return 1;
+}
+
+int sd_journal_seek_cursor(sd_journal *j, const char *cursor) {
+        char *w;
+        size_t l;
+        char *state;
+        unsigned long long seqnum, monotonic, realtime, xor_hash;
+        bool
+                seqnum_id_set = false,
+                seqnum_set = false,
+                boot_id_set = false,
+                monotonic_set = false,
+                realtime_set = false,
+                xor_hash_set = false;
+        sd_id128_t seqnum_id, boot_id;
+
+        assert(j);
+        assert(cursor);
+
+        FOREACH_WORD_SEPARATOR(w, l, cursor, ";", state) {
+                char *item;
+                int k = 0;
+
+                if (l < 2 || w[1] != '=')
+                        return -EINVAL;
+
+                item = strndup(w, l);
+                if (!item)
+                        return -ENOMEM;
+
+                switch (w[0]) {
+
+                case 's':
+                        seqnum_id_set = true;
+                        k = sd_id128_from_string(w+2, &seqnum_id);
+                        break;
+
+                case 'i':
+                        seqnum_set = true;
+                        if (sscanf(w+2, "%llx", &seqnum) != 1)
+                                k = -EINVAL;
+                        break;
+
+                case 'b':
+                        boot_id_set = true;
+                        k = sd_id128_from_string(w+2, &boot_id);
+                        break;
+
+                case 'm':
+                        monotonic_set = true;
+                        if (sscanf(w+2, "%llx", &monotonic) != 1)
+                                k = -EINVAL;
+                        break;
+
+                case 't':
+                        realtime_set = true;
+                        if (sscanf(w+2, "%llx", &realtime) != 1)
+                                k = -EINVAL;
+                        break;
+
+                case 'x':
+                        xor_hash_set = true;
+                        if (sscanf(w+2, "%llx", &xor_hash) != 1)
+                                k = -EINVAL;
+                        break;
+                }
+
+                free(item);
+
+                if (k < 0)
+                        return k;
+        }
+
+        if ((!seqnum_set || !seqnum_id_set) &&
+            (!monotonic_set || !boot_id_set) &&
+            !realtime_set)
+                return -EINVAL;
+
+        reset_location(j);
+
+        j->current_location.type = LOCATION_DISCRETE;
+
+        if (realtime_set) {
+                j->current_location.realtime = (uint64_t) realtime;
+                j->current_location.realtime_set = true;
+        }
+
+        if (seqnum_set && seqnum_id_set) {
+                j->current_location.seqnum = (uint64_t) seqnum;
+                j->current_location.seqnum_id = seqnum_id;
+                j->current_location.seqnum_set = true;
+        }
+
+        if (monotonic_set && boot_id_set) {
+                j->current_location.monotonic = (uint64_t) monotonic;
+                j->current_location.boot_id = boot_id;
+                j->current_location.monotonic_set = true;
+        }
+
+        if (xor_hash_set) {
+                j->current_location.xor_hash = (uint64_t) xor_hash;
+                j->current_location.xor_hash_set = true;
+        }
+
+        return 0;
+}
+
+int sd_journal_seek_monotonic_usec(sd_journal *j, sd_id128_t boot_id, uint64_t usec) {
+        assert(j);
+
+        reset_location(j);
+        j->current_location.type = LOCATION_DISCRETE;
+        j->current_location.boot_id = boot_id;
+        j->current_location.monotonic = usec;
+        j->current_location.monotonic_set = true;
+
+        return 0;
+}
+
+int sd_journal_seek_realtime_usec(sd_journal *j, uint64_t usec) {
+        assert(j);
+
+        reset_location(j);
+        j->current_location.type = LOCATION_DISCRETE;
+        j->current_location.realtime = usec;
+        j->current_location.realtime_set = true;
+
+        return 0;
+}
+
+int sd_journal_seek_head(sd_journal *j) {
+        assert(j);
+
+        reset_location(j);
+        j->current_location.type = LOCATION_HEAD;
+
+        return 0;
+}
+
+int sd_journal_seek_tail(sd_journal *j) {
+        assert(j);
+
+        reset_location(j);
+        j->current_location.type = LOCATION_TAIL;
+
+        return 0;
+}
+
+static int add_file(sd_journal *j, const char *prefix, const char *dir, const char *filename) {
+        char *fn;
+        int r;
+        JournalFile *f;
+
+        assert(j);
+        assert(prefix);
+        assert(filename);
+
+        if (dir)
+                fn = join(prefix, "/", dir, "/", filename, NULL);
+        else
+                fn = join(prefix, "/", filename, NULL);
+
+        if (!fn)
+                return -ENOMEM;
+
+        r = journal_file_open(fn, O_RDONLY, 0, NULL, &f);
+        free(fn);
+
+        if (r < 0) {
+                if (errno == ENOENT)
+                        return 0;
+
+                return r;
+        }
+
+        journal_file_dump(f);
+
+
+        r = hashmap_put(j->files, f->path, f);
+        if (r < 0) {
+                journal_file_close(f);
+                return r;
+        }
+
+        return 0;
+}
+
+static int add_directory(sd_journal *j, const char *prefix, const char *dir) {
+        char *fn;
+        int r;
+        DIR *d;
+
+        assert(j);
+        assert(prefix);
+        assert(dir);
+
+        fn = join(prefix, "/", dir, NULL);
+        if (!fn)
+                return -ENOMEM;
+
+        d = opendir(fn);
+        free(fn);
+
+        if (!d) {
+                if (errno == ENOENT)
+                        return 0;
+
+                return -errno;
+        }
+
+        for (;;) {
+                struct dirent buf, *de;
+
+                r = readdir_r(d, &buf, &de);
+                if (r != 0 || !de)
+                        break;
+
+                if (!dirent_is_file_with_suffix(de, ".journal"))
+                        continue;
+
+                r = add_file(j, prefix, dir, de->d_name);
+                if (r < 0)
+                        log_debug("Failed to add file %s/%s/%s: %s", prefix, dir, de->d_name, strerror(-r));
+        }
+
+        closedir(d);
+
+        return 0;
+}
+
+int sd_journal_open(sd_journal **ret) {
+        sd_journal *j;
+        const char *p;
+        const char search_paths[] =
+                "/run/log/journal\0"
+                "/var/log/journal\0";
+        int r;
+
+        assert(ret);
+
+        j = new0(sd_journal, 1);
+        if (!j)
+                return -ENOMEM;
+
+        j->files = hashmap_new(string_hash_func, string_compare_func);
+        if (!j->files) {
+                r = -ENOMEM;
+                goto fail;
+        }
+
+        /* We ignore most errors here, since the idea is to only open
+         * what's actually accessible, and ignore the rest. */
+
+        NULSTR_FOREACH(p, search_paths) {
+                DIR *d;
+
+                d = opendir(p);
+                if (!d) {
+                        if (errno != ENOENT)
+                                log_debug("Failed to open %s: %m", p);
+                        continue;
+                }
+
+                for (;;) {
+                        struct dirent buf, *de;
+                        sd_id128_t id;
+
+                        r = readdir_r(d, &buf, &de);
+                        if (r != 0 || !de)
+                                break;
+
+                        if (dirent_is_file_with_suffix(de, ".journal")) {
+                                r = add_file(j, p, NULL, de->d_name);
+                                if (r < 0)
+                                        log_debug("Failed to add file %s/%s: %s", p, de->d_name, strerror(-r));
+
+                        } else if ((de->d_type == DT_DIR || de->d_type == DT_UNKNOWN) &&
+                                   sd_id128_from_string(de->d_name, &id) >= 0) {
+
+                                r = add_directory(j, p, de->d_name);
+                                if (r < 0)
+                                        log_debug("Failed to add directory %s/%s: %s", p, de->d_name, strerror(-r));
+                        }
+                }
+
+                closedir(d);
+        }
+
+        *ret = j;
+        return 0;
+
+fail:
+        sd_journal_close(j);
+
+        return r;
+};
+
+void sd_journal_close(sd_journal *j) {
+        assert(j);
+
+        if (j->files) {
+                JournalFile *f;
+
+                while ((f = hashmap_steal_first(j->files)))
+                        journal_file_close(f);
+
+                hashmap_free(j->files);
+        }
+
+        sd_journal_flush_matches(j);
+
+        free(j);
+}
+
+int sd_journal_get_realtime_usec(sd_journal *j, uint64_t *ret) {
+        Object *o;
+        JournalFile *f;
+        int r;
+
+        assert(j);
+        assert(ret);
+
+        f = j->current_file;
+        if (!f)
+                return -EADDRNOTAVAIL;
+
+        if (f->current_offset <= 0)
+                return -EADDRNOTAVAIL;
+
+        r = journal_file_move_to_object(f, OBJECT_ENTRY, f->current_offset, &o);
+        if (r < 0)
+                return r;
+
+        *ret = le64toh(o->entry.realtime);
+        return 0;
+}
+
+int sd_journal_get_monotonic_usec(sd_journal *j, uint64_t *ret, sd_id128_t *ret_boot_id) {
+        Object *o;
+        JournalFile *f;
+        int r;
+        sd_id128_t id;
+
+        assert(j);
+        assert(ret);
+
+        f = j->current_file;
+        if (!f)
+                return -EADDRNOTAVAIL;
+
+        if (f->current_offset <= 0)
+                return -EADDRNOTAVAIL;
+
+        r = journal_file_move_to_object(f, OBJECT_ENTRY, f->current_offset, &o);
+        if (r < 0)
+                return r;
+
+        if (ret_boot_id)
+                *ret_boot_id = o->entry.boot_id;
+        else {
+                r = sd_id128_get_boot(&id);
+                if (r < 0)
+                        return r;
+
+                if (!sd_id128_equal(id, o->entry.boot_id))
+                        return -ENOENT;
+        }
+
+        *ret = le64toh(o->entry.monotonic);
+        return 0;
+}
+
+int sd_journal_get_data(sd_journal *j, const char *field, const void **data, size_t *size) {
+        JournalFile *f;
+        uint64_t i, n;
+        size_t field_length;
+        int r;
+        Object *o;
+
+        assert(j);
+        assert(field);
+        assert(data);
+        assert(size);
+
+        if (isempty(field) || strchr(field, '='))
+                return -EINVAL;
+
+        f = j->current_file;
+        if (!f)
+                return -EADDRNOTAVAIL;
+
+        if (f->current_offset <= 0)
+                return -EADDRNOTAVAIL;
+
+        r = journal_file_move_to_object(f, OBJECT_ENTRY, f->current_offset, &o);
+        if (r < 0)
+                return r;
+
+        field_length = strlen(field);
+
+        n = journal_file_entry_n_items(o);
+        for (i = 0; i < n; i++) {
+                uint64_t p, l, le_hash;
+                size_t t;
+
+                p = le64toh(o->entry.items[i].object_offset);
+                le_hash = o->entry.items[j->current_field].hash;
+                r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
+                if (r < 0)
+                        return r;
+
+                if (le_hash != o->data.hash)
+                        return -EBADMSG;
+
+                l = le64toh(o->object.size) - offsetof(Object, data.payload);
+
+                if (l >= field_length+1 &&
+                    memcmp(o->data.payload, field, field_length) == 0 &&
+                    o->data.payload[field_length] == '=') {
+
+                        t = (size_t) l;
+
+                        if ((uint64_t) t != l)
+                                return -E2BIG;
+
+                        *data = o->data.payload;
+                        *size = t;
+
+                        return 0;
+                }
+
+                r = journal_file_move_to_object(f, OBJECT_ENTRY, f->current_offset, &o);
+                if (r < 0)
+                        return r;
+        }
+
+        return -ENOENT;
+}
+
+int sd_journal_enumerate_data(sd_journal *j, const void **data, size_t *size) {
+        JournalFile *f;
+        uint64_t p, l, n, le_hash;
+        int r;
+        Object *o;
+        size_t t;
+
+        assert(j);
+        assert(data);
+        assert(size);
+
+        f = j->current_file;
+        if (!f)
+                return -EADDRNOTAVAIL;
+
+        if (f->current_offset <= 0)
+                return -EADDRNOTAVAIL;
+
+        r = journal_file_move_to_object(f, OBJECT_ENTRY, f->current_offset, &o);
+        if (r < 0)
+                return r;
+
+        n = journal_file_entry_n_items(o);
+        if (j->current_field >= n)
+                return 0;
+
+        p = le64toh(o->entry.items[j->current_field].object_offset);
+        le_hash = o->entry.items[j->current_field].hash;
+        r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
+        if (r < 0)
+                return r;
+
+        if (le_hash != o->data.hash)
+                return -EBADMSG;
+
+        l = le64toh(o->object.size) - offsetof(Object, data.payload);
+        t = (size_t) l;
+
+        /* We can't read objects larger than 4G on a 32bit machine */
+        if ((uint64_t) t != l)
+                return -E2BIG;
+
+        *data = o->data.payload;
+        *size = t;
+
+        j->current_field ++;
+
+        return 1;
+}
+
+void sd_journal_restart_data(sd_journal *j) {
+        assert(j);
+
+        j->current_field = 0;
+}
diff --git a/src/journal/sd-journal.h b/src/journal/sd-journal.h
new file mode 100644 (file)
index 0000000..8d7e314
--- /dev/null
@@ -0,0 +1,102 @@
+/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
+
+#ifndef foojournalhfoo
+#define foojournalhfoo
+
+/***
+  This file is part of systemd.
+
+  Copyright 2011 Lennart Poettering
+
+  systemd is free software; you can redistribute it and/or modify it
+  under the terms of the GNU General Public License as published by
+  the Free Software Foundation; either version 2 of the License, or
+  (at your option) any later version.
+
+  systemd is distributed in the hope that it will be useful, but
+  WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+  General Public License for more details.
+
+  You should have received a copy of the GNU General Public License
+  along with systemd; If not, see <http://www.gnu.org/licenses/>.
+***/
+
+#include <inttypes.h>
+#include <sys/types.h>
+
+#include "sd-id128.h"
+
+/* TODO:
+ *
+ *   - check LE/BE conversion for 8bit, 16bit, 32bit values
+ *   - implement inotify usage on client
+ *   - implement audit gateway
+ *   - implement native gateway
+ *   - implement stdout gateway
+ *   - extend hash tables table as we go
+ *   - accelerate looking for "all hostnames" and suchlike.
+ *   - throttling
+ *   - enforce limit on open journal files in journald and journalctl
+ *   - cryptographic hash
+ *   - fix space reservation logic
+ *   - comm, argv can be manipulated, should it be _COMM=, _CMDLINE= or COMM=, CMDLINE=?
+ */
+
+typedef struct sd_journal sd_journal;
+
+int sd_journal_open(sd_journal **ret);
+void sd_journal_close(sd_journal *j);
+
+int sd_journal_previous(sd_journal *j);
+int sd_journal_next(sd_journal *j);
+
+int sd_journal_previous_skip(sd_journal *j, uint64_t skip);
+int sd_journal_next_skip(sd_journal *j, uint64_t skip);
+
+int sd_journal_get_realtime_usec(sd_journal *j, uint64_t *ret);
+int sd_journal_get_monotonic_usec(sd_journal *j, uint64_t *ret, sd_id128_t *ret_boot_id);
+int sd_journal_get_data(sd_journal *j, const char *field, const void **data, size_t *l);
+int sd_journal_enumerate_data(sd_journal *j, const void **data, size_t *l);
+void sd_journal_restart_data(sd_journal *j);
+
+int sd_journal_add_match(sd_journal *j, const void *data, size_t size);
+void sd_journal_flush_matches(sd_journal *j);
+
+int sd_journal_seek_head(sd_journal *j);
+int sd_journal_seek_tail(sd_journal *j);
+int sd_journal_seek_monotonic_usec(sd_journal *j, sd_id128_t boot_id, uint64_t usec);
+int sd_journal_seek_realtime_usec(sd_journal *j, uint64_t usec);
+int sd_journal_seek_cursor(sd_journal *j, const char *cursor);
+
+int sd_journal_get_cursor(sd_journal *j, char **cursor);
+
+int sd_journal_query_unique(sd_journal *j, const char *field);      /* missing */
+int sd_journal_enumerate_unique(sd_journal *j, const void **data, size_t *l); /* missing */
+void sd_journal_restart_unique(sd_journal *j);                      /* missing */
+
+enum {
+        SD_JOURNAL_NOP,
+        SD_JOURNAL_APPEND,
+        SD_JOURNAL_INVALIDATE_ADD,
+        SD_JOURNAL_INVALIDATE_REMOVE
+};
+
+int sd_journal_get_fd(sd_journal *j);                              /* missing */
+int sd_journal_process(sd_journal *j);                             /* missing */
+
+#define SD_JOURNAL_FOREACH(j)                                           \
+        if (sd_journal_seek_head(j) >= 0)                               \
+                while (sd_journal_next(j) > 0)                          \
+
+#define SD_JOURNAL_FOREACH_BACKWARDS(j)                                 \
+        if (sd_journal_seek_tail(j) >= 0)                               \
+                while (sd_journal_previous(j) > 0)                      \
+
+#define SD_JOURNAL_FOREACH_DATA(j, data, l)                             \
+        for (sd_journal_restart_data(j); sd_journal_enumerate_data((j), &(data), &(l)) > 0; )
+
+#define SD_JOURNAL_FOREACH_UNIQUE(j, data, l)                           \
+        for (sd_journal_restart_unique(j); sd_journal_enumerate_data((j), &(data), &(l)) > 0; )
+
+#endif
diff --git a/src/journal/test-journal.c b/src/journal/test-journal.c
new file mode 100644 (file)
index 0000000..45ced12
--- /dev/null
@@ -0,0 +1,116 @@
+/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
+
+/***
+  This file is part of systemd.
+
+  Copyright 2011 Lennart Poettering
+
+  systemd is free software; you can redistribute it and/or modify it
+  under the terms of the GNU General Public License as published by
+  the Free Software Foundation; either version 2 of the License, or
+  (at your option) any later version.
+
+  systemd is distributed in the hope that it will be useful, but
+  WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+  General Public License for more details.
+
+  You should have received a copy of the GNU General Public License
+  along with systemd; If not, see <http://www.gnu.org/licenses/>.
+***/
+
+#include <fcntl.h>
+#include <unistd.h>
+
+#include "journal-file.h"
+#include "log.h"
+
+int main(int argc, char *argv[]) {
+        dual_timestamp ts;
+        JournalFile *f;
+        struct iovec iovec;
+        static const char test[] = "test", test2[] = "test2";
+        Object *o;
+        uint64_t p;
+
+        log_set_max_level(LOG_DEBUG);
+
+        unlink("test.journal");
+
+        assert_se(journal_file_open("test.journal", O_RDWR|O_CREAT, 0666, NULL, &f) == 0);
+
+        dual_timestamp_get(&ts);
+
+        iovec.iov_base = (void*) test;
+        iovec.iov_len = strlen(test);
+        assert_se(journal_file_append_entry(f, &ts, &iovec, 1, NULL, NULL, NULL) == 0);
+
+        iovec.iov_base = (void*) test2;
+        iovec.iov_len = strlen(test2);
+        assert_se(journal_file_append_entry(f, &ts, &iovec, 1, NULL, NULL, NULL) == 0);
+
+        iovec.iov_base = (void*) test;
+        iovec.iov_len = strlen(test);
+        assert_se(journal_file_append_entry(f, &ts, &iovec, 1, NULL, NULL, NULL) == 0);
+
+        journal_file_dump(f);
+
+        assert(journal_file_next_entry(f, NULL, 0, DIRECTION_DOWN, &o, &p) == 1);
+        assert(le64toh(o->entry.seqnum) == 1);
+
+        assert(journal_file_next_entry(f, o, p, DIRECTION_DOWN, &o, &p) == 1);
+        assert(le64toh(o->entry.seqnum) == 2);
+
+        assert(journal_file_next_entry(f, o, p, DIRECTION_DOWN, &o, &p) == 1);
+        assert(le64toh(o->entry.seqnum) == 3);
+
+        assert(journal_file_next_entry(f, o, p, DIRECTION_DOWN, &o, &p) == 0);
+
+        assert(journal_file_next_entry(f, NULL, 0, DIRECTION_DOWN, &o, &p) == 1);
+        assert(le64toh(o->entry.seqnum) == 1);
+
+        assert(journal_file_skip_entry(f, o, p, 2, &o, &p) == 1);
+        assert(le64toh(o->entry.seqnum) == 3);
+
+        assert(journal_file_skip_entry(f, o, p, -2, &o, &p) == 1);
+        assert(le64toh(o->entry.seqnum) == 1);
+
+        assert(journal_file_skip_entry(f, o, p, -2, &o, &p) == 1);
+        assert(le64toh(o->entry.seqnum) == 1);
+
+        assert(journal_file_find_data_object(f, test, strlen(test), NULL, &p) == 1);
+        assert(journal_file_next_entry_for_data(f, NULL, 0, p, DIRECTION_DOWN, &o, NULL) == 1);
+        assert(le64toh(o->entry.seqnum) == 1);
+
+        assert(journal_file_next_entry_for_data(f, NULL, 0, p, DIRECTION_UP, &o, NULL) == 1);
+        assert(le64toh(o->entry.seqnum) == 3);
+
+        assert(journal_file_find_data_object(f, test2, strlen(test2), NULL, &p) == 1);
+        assert(journal_file_next_entry_for_data(f, NULL, 0, p, DIRECTION_UP, &o, NULL) == 1);
+        assert(le64toh(o->entry.seqnum) == 2);
+
+        assert(journal_file_next_entry_for_data(f, NULL, 0, p, DIRECTION_DOWN, &o, NULL) == 1);
+        assert(le64toh(o->entry.seqnum) == 2);
+
+        assert(journal_file_find_data_object(f, "quux", 4, NULL, &p) == 0);
+
+        assert(journal_file_move_to_entry_by_seqnum(f, 1, DIRECTION_DOWN, &o, NULL) == 1);
+        assert(le64toh(o->entry.seqnum) == 1);
+
+        assert(journal_file_move_to_entry_by_seqnum(f, 3, DIRECTION_DOWN, &o, NULL) == 1);
+        assert(le64toh(o->entry.seqnum) == 3);
+
+        assert(journal_file_move_to_entry_by_seqnum(f, 2, DIRECTION_DOWN, &o, NULL) == 1);
+        assert(le64toh(o->entry.seqnum) == 2);
+
+        assert(journal_file_move_to_entry_by_seqnum(f, 10, DIRECTION_DOWN, &o, NULL) == 0);
+
+        journal_file_rotate(&f);
+        journal_file_rotate(&f);
+
+        journal_file_close(f);
+
+        journal_directory_vacuum(".", 3000000, 0);
+
+        return 0;
+}
index 0901a0e49bffb47965918f9f38fbaa6d11dad9cb..70cc0730eeed7c608502595518ca35d44d1d36a4 100644 (file)
@@ -65,45 +65,53 @@ static void server_done(Server *s) {
                 fdset_free(s->syslog_fds);
 }
 
-static int server_init(Server *s, unsigned n_sockets) {
-        int r;
-        unsigned i;
+static int server_init(Server *s) {
+        int i, r, n;
         struct epoll_event ev;
         sigset_t mask;
 
         assert(s);
-        assert(n_sockets > 0);
 
         zero(*s);
-
         s->kmsg_fd = s->signal_fd = -1;
 
-        if ((s->epoll_fd = epoll_create1(EPOLL_CLOEXEC)) < 0) {
-                r = -errno;
-                log_error("Failed to create epoll object: %s", strerror(errno));
-                goto fail;
+        s->epoll_fd = epoll_create1(EPOLL_CLOEXEC);
+        if (s->epoll_fd < 0) {
+                log_error("Failed to create epoll object: %m");
+                return -errno;
+        }
+
+        s->syslog_fds = fdset_new();
+        if (!s->syslog_fds) {
+                log_error("Failed to allocate file descriptor set: %s", strerror(ENOMEM));
+                return -ENOMEM;
         }
 
-        if (!(s->syslog_fds = fdset_new())) {
-                r = -ENOMEM;
-                log_error("Failed to allocate file descriptor set: %s", strerror(errno));
-                goto fail;
+        n = sd_listen_fds(true);
+        if (n < 0) {
+                log_error("Failed to read listening file descriptors from environment: %s", strerror(-n));
+                return n;
+        }
+
+        if (n <= 0 || n > SERVER_FD_MAX) {
+                log_error("No or too many file descriptors passed.");
+                return -EINVAL;
         }
 
-        for (i = 0; i < n_sockets; i++) {
+        for (i = 0; i < n; i++) {
                 int fd, one = 1;
 
                 fd = SD_LISTEN_FDS_START+i;
 
-                if ((r = sd_is_socket(fd, AF_UNSPEC, SOCK_DGRAM, -1)) < 0) {
+                r = sd_is_socket(fd, AF_UNSPEC, SOCK_DGRAM, -1);
+                if (r < 0) {
                         log_error("Failed to determine file descriptor type: %s", strerror(-r));
-                        goto fail;
+                        return r;
                 }
 
                 if (!r) {
                         log_error("Wrong file descriptor type.");
-                        r = -EINVAL;
-                        goto fail;
+                        return -EINVAL;
                 }
 
                 if (setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &one, sizeof(one)) < 0)
@@ -113,18 +121,19 @@ static int server_init(Server *s, unsigned n_sockets) {
                 ev.events = EPOLLIN;
                 ev.data.fd = fd;
                 if (epoll_ctl(s->epoll_fd, EPOLL_CTL_ADD, fd, &ev) < 0) {
-                        r = -errno;
-                        log_error("Failed to add server fd to epoll object: %s", strerror(errno));
-                        goto fail;
+                        log_error("Failed to add server fd to epoll object: %m");
+                        return -errno;
                 }
 
-                if ((r = fdset_put(s->syslog_fds, fd)) < 0) {
+                r = fdset_put(s->syslog_fds, fd);
+                if (r < 0) {
                         log_error("Failed to store file descriptor in set: %s", strerror(-r));
-                        goto fail;
+                        return r;
                 }
         }
 
-        if ((s->kmsg_fd = open("/dev/kmsg", O_WRONLY|O_NOCTTY|O_CLOEXEC)) < 0) {
+        s->kmsg_fd = open("/dev/kmsg", O_WRONLY|O_NOCTTY|O_CLOEXEC);
+        if (s->kmsg_fd < 0) {
                 log_error("Failed to open /dev/kmsg for logging: %m");
                 return -errno;
         }
@@ -133,7 +142,8 @@ static int server_init(Server *s, unsigned n_sockets) {
         sigset_add_many(&mask, SIGINT, SIGTERM, -1);
         assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
 
-        if ((s->signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC)) < 0) {
+        s->signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
+        if (s->signal_fd < 0) {
                 log_error("signalfd(): %m");
                 return -errno;
         }
@@ -148,80 +158,6 @@ static int server_init(Server *s, unsigned n_sockets) {
         }
 
         return 0;
-
-fail:
-        server_done(s);
-        return r;
-}
-
-static void skip_date(const char **buf) {
-        enum {
-                LETTER,
-                SPACE,
-                NUMBER,
-                SPACE_OR_NUMBER,
-                COLON
-        } sequence[] = {
-                LETTER, LETTER, LETTER,
-                SPACE,
-                SPACE_OR_NUMBER, NUMBER,
-                SPACE,
-                SPACE_OR_NUMBER, NUMBER,
-                COLON,
-                SPACE_OR_NUMBER, NUMBER,
-                COLON,
-                SPACE_OR_NUMBER, NUMBER,
-                SPACE
-        };
-
-        const char *p;
-        unsigned i;
-
-        assert(buf);
-        assert(*buf);
-
-        p = *buf;
-
-        for (i = 0; i < ELEMENTSOF(sequence); i++, p++) {
-
-                if (!*p)
-                        return;
-
-                switch (sequence[i]) {
-
-                case SPACE:
-                        if (*p != ' ')
-                                return;
-                        break;
-
-                case SPACE_OR_NUMBER:
-                        if (*p == ' ')
-                                break;
-
-                        /* fall through */
-
-                case NUMBER:
-                        if (*p < '0' || *p > '9')
-                                return;
-
-                        break;
-
-                case LETTER:
-                        if (!(*p >= 'A' && *p <= 'Z') &&
-                            !(*p >= 'a' && *p <= 'z'))
-                                return;
-
-                        break;
-
-                case COLON:
-                        if (*p != ':')
-                                return;
-                        break;
-
-                }
-        }
-
-        *buf = p;
 }
 
 static int read_process(const char **buf, struct iovec *iovec) {
@@ -266,28 +202,6 @@ static int read_process(const char **buf, struct iovec *iovec) {
         return 1;
 }
 
-static void skip_pid(const char **buf) {
-        const char *p;
-
-        assert(buf);
-        assert(*buf);
-
-        p = *buf;
-
-        if (*p != '[')
-                return;
-
-        p++;
-        p += strspn(p, "0123456789");
-
-        if (*p != ']')
-                return;
-
-        p++;
-
-        *buf = p;
-}
-
 static int write_message(Server *s, const char *buf, struct ucred *ucred) {
         ssize_t k;
         char priority[6], pid[16];
@@ -314,14 +228,14 @@ static int write_message(Server *s, const char *buf, struct ucred *ucred) {
         IOVEC_SET_STRING(iovec[i++], priority);
 
         /* Second, skip date */
-        skip_date(&buf);
+        skip_syslog_date((char**) &buf);
 
         /* Then, add process if set */
         if (read_process(&buf, &iovec[i]) > 0)
                 i++;
         else if (ucred &&
                  ucred->pid > 0 &&
-                 get_process_name(ucred->pid, &process) >= 0)
+                 get_process_comm(ucred->pid, &process) >= 0)
                 IOVEC_SET_STRING(iovec[i++], process);
 
         /* Skip the stored PID if we have a better one */
@@ -330,7 +244,7 @@ static int write_message(Server *s, const char *buf, struct ucred *ucred) {
                 char_array_0(pid);
                 IOVEC_SET_STRING(iovec[i++], pid);
 
-                skip_pid(&buf);
+                skip_syslog_pid((char**) &buf);
 
                 if (*buf == ':')
                         buf++;
@@ -368,7 +282,8 @@ static int process_event(Server *s, struct epoll_event *ev) {
                 struct signalfd_siginfo sfsi;
                 ssize_t n;
 
-                if ((n = read(s->signal_fd, &sfsi, sizeof(sfsi))) != sizeof(sfsi)) {
+                n = read(s->signal_fd, &sfsi, sizeof(sfsi));
+                if (n != sizeof(sfsi)) {
 
                         if (n >= 0)
                                 return -EIO;
@@ -407,7 +322,8 @@ static int process_event(Server *s, struct epoll_event *ev) {
                         msghdr.msg_control = &control;
                         msghdr.msg_controllen = sizeof(control);
 
-                        if ((n = recvmsg(ev->data.fd, &msghdr, MSG_DONTWAIT)) < 0) {
+                        n = recvmsg(ev->data.fd, &msghdr, MSG_DONTWAIT);
+                        if (n < 0) {
 
                                 if (errno == EINTR || errno == EAGAIN)
                                         return 1;
@@ -424,12 +340,14 @@ static int process_event(Server *s, struct epoll_event *ev) {
                         else
                                 ucred = NULL;
 
-                        if ((e = memchr(buf, '\n', n)))
+                        e = memchr(buf, '\n', n);
+                        if (e)
                                 *e = 0;
                         else
                                 buf[n] = 0;
 
-                        if ((k = write_message(s, strstrip(buf), ucred)) < 0)
+                        k = write_message(s, strstrip(buf), ucred);
+                        if (k < 0)
                                 return k;
                 }
         }
@@ -439,7 +357,7 @@ static int process_event(Server *s, struct epoll_event *ev) {
 
 int main(int argc, char *argv[]) {
         Server server;
-        int r = EXIT_FAILURE, n;
+        int r;
 
         if (getppid() != 1) {
                 log_error("This program should be invoked by init only.");
@@ -457,18 +375,9 @@ int main(int argc, char *argv[]) {
 
         umask(0022);
 
-        if ((n = sd_listen_fds(true)) < 0) {
-                log_error("Failed to read listening file descriptors from environment: %s", strerror(-r));
-                return EXIT_FAILURE;
-        }
-
-        if (n <= 0 || n > SERVER_FD_MAX) {
-                log_error("No or too many file descriptors passed.");
-                return EXIT_FAILURE;
-        }
-
-        if (server_init(&server, (unsigned) n) < 0)
-                return EXIT_FAILURE;
+        r = server_init(&server);
+        if (r < 0)
+                goto finish;
 
         log_debug("systemd-kmsg-syslogd running as pid %lu", (unsigned long) getpid());
 
@@ -478,36 +387,33 @@ int main(int argc, char *argv[]) {
 
         for (;;) {
                 struct epoll_event event;
-                int k;
 
-                if ((k = epoll_wait(server.epoll_fd, &event, 1, -1)) < 0) {
+                r = epoll_wait(server.epoll_fd, &event, 1, -1);
+                if (r < 0) {
 
                         if (errno == EINTR)
                                 continue;
 
                         log_error("epoll_wait() failed: %m");
-                        goto fail;
-                }
-
-                if (k <= 0)
+                        r = -errno;
+                        goto finish;
+                } else if (r == 0)
                         break;
 
-                if ((k = process_event(&server, &event)) < 0)
-                        goto fail;
-
-                if (k == 0)
+                r = process_event(&server, &event);
+                if (r < 0)
+                        goto finish;
+                else if (r == 0)
                         break;
         }
 
-        r = EXIT_SUCCESS;
-
         log_debug("systemd-kmsg-syslogd stopped as pid %lu", (unsigned long) getpid());
 
-fail:
+finish:
         sd_notify(false,
                   "STATUS=Shutting down...");
 
         server_done(&server);
 
-        return r;
+        return r < 0 ? EXIT_FAILURE : EXIT_SUCCESS;
 }
index 89762b66b01cf8f9592ba417313d4ec6eaccc2fe..1be47c8dde9b8a3340c251a947ebbf49d5a3a2f6 100644 (file)
@@ -393,7 +393,7 @@ static void print_session_status_info(SessionStatusInfo *i) {
 
                 printf("\t  Leader: %u", (unsigned) i->leader);
 
-                get_process_name(i->leader, &t);
+                get_process_comm(i->leader, &t);
                 if (t) {
                         printf(" (%s)", t);
                         free(t);
index 7a06b501d45fbe420caae814abf40f54347e3011..eb8a48d191f03189df4ed1a243dc3d0352811780 100644 (file)
 
 #include "logind-acl.h"
 #include "util.h"
-
-static int find_acl(acl_t acl, uid_t uid, acl_entry_t *entry) {
-        acl_entry_t i;
-        int found;
-
-        assert(acl);
-        assert(entry);
-
-        for (found = acl_get_entry(acl, ACL_FIRST_ENTRY, &i);
-             found > 0;
-             found = acl_get_entry(acl, ACL_NEXT_ENTRY, &i)) {
-
-                acl_tag_t tag;
-                uid_t *u;
-                bool b;
-
-                if (acl_get_tag_type(i, &tag) < 0)
-                        return -errno;
-
-                if (tag != ACL_USER)
-                        continue;
-
-                u = acl_get_qualifier(i);
-                if (!u)
-                        return -errno;
-
-                b = *u == uid;
-                acl_free(u);
-
-                if (b) {
-                        *entry = i;
-                        return 1;
-                }
-        }
-
-        if (found < 0)
-                return -errno;
-
-        return 0;
-}
+#include "acl-util.h"
 
 static int flush_acl(acl_t acl) {
         acl_entry_t i;
@@ -125,7 +86,7 @@ int devnode_acl(const char *path,
         } else if (del && old_uid > 0) {
                 acl_entry_t entry;
 
-                r = find_acl(acl, old_uid, &entry);
+                r = acl_find_uid(acl, old_uid, &entry);
                 if (r < 0)
                         goto finish;
 
@@ -144,7 +105,7 @@ int devnode_acl(const char *path,
                 acl_permset_t permset;
                 int rd, wt;
 
-                r = find_acl(acl, new_uid, &entry);
+                r = acl_find_uid(acl, new_uid, &entry);
                 if (r < 0)
                         goto finish;
 
index 519521fe6768f9de96fffb8bc44fc314a54991f7..9b25b10438b8702b8c239442a1c20bc414928396 100644 (file)
 #include "macro.h"
 #include "util.h"
 #include "log.h"
-
-static void make_v4_uuid(unsigned char *id) {
-        /* Stolen from generate_random_uuid() of drivers/char/random.c
-         * in the kernel sources */
-
-        /* Set UUID version to 4 --- truly random generation */
-        id[6] = (id[6] & 0x0F) | 0x40;
-
-        /* Set the UUID variant to DCE */
-        id[8] = (id[8] & 0x3F) | 0x80;
-}
+#include "sd-id128.h"
 
 static int generate(char id[34]) {
-        int fd;
-        unsigned char buf[16], *p;
+        int fd, r;
+        unsigned char *p;
+        sd_id128_t buf;
         char *q;
         ssize_t k;
 
@@ -68,26 +59,13 @@ static int generate(char id[34]) {
         }
 
         /* If that didn't work, generate a random machine id */
-        fd = open("/dev/urandom", O_RDONLY|O_CLOEXEC|O_NOCTTY);
-        if (fd < 0) {
-                log_error("Failed to open /dev/urandom: %m");
-                return -errno;
-        }
-
-        k = loop_read(fd, buf, sizeof(buf), false);
-        close_nointr_nofail(fd);
-
-        if (k != sizeof(buf)) {
-                log_error("Failed to read /dev/urandom: %s", strerror(k < 0 ? -k : EIO));
-                return k < 0 ? (int) k : -EIO;
+        r = sd_id128_randomize(&buf);
+        if (r < 0) {
+                log_error("Failed to open /dev/urandom: %s", strerror(-r));
+                return r;
         }
 
-        /* Turn this into a valid v4 UUID, to be nice. Note that we
-         * only guarantee this for newly generated UUIDs, not for
-         * pre-existing ones.*/
-        make_v4_uuid(buf);
-
-        for (p = buf, q = id; p < buf + sizeof(buf); p++, q += 2) {
+        for (p = buf.bytes, q = id; p < buf.bytes + sizeof(buf); p++, q += 2) {
                 q[0] = hexchar(*p >> 4);
                 q[1] = hexchar(*p & 15);
         }
index e7a4d2cde1940e531b113a0231a52c517bbf006d..3f30aa78920b4dad34e554d07fd6e8a42fef7760 100644 (file)
@@ -149,7 +149,7 @@ static inline size_t ALIGN_TO(size_t l, size_t ali) {
                 char *_s = (char *)(s);         \
                 _i->iov_base = _s;              \
                 _i->iov_len = strlen(_s);       \
-        } while(false);
+        } while(false)
 
 static inline size_t IOVEC_TOTAL_SIZE(const struct iovec *i, unsigned n) {
         unsigned j;
index 111167a8c2a19147af66dfec669a72de2b2f444a..f8cbcfcc98ff8fa7faf1f25dadf1ec7fa908e5f5 100644 (file)
@@ -2032,7 +2032,7 @@ static int manager_dispatch_sigchld(Manager *m) {
                 if (si.si_code == CLD_EXITED || si.si_code == CLD_KILLED || si.si_code == CLD_DUMPED) {
                         char *name = NULL;
 
-                        get_process_name(si.si_pid, &name);
+                        get_process_comm(si.si_pid, &name);
                         log_debug("Got SIGCHLD for process %lu (%s)", (unsigned long) si.si_pid, strna(name));
                         free(name);
                 }
@@ -2117,7 +2117,7 @@ static int manager_process_signal_fd(Manager *m) {
                 if (sfsi.ssi_pid > 0) {
                         char *p = NULL;
 
-                        get_process_name(sfsi.ssi_pid, &p);
+                        get_process_comm(sfsi.ssi_pid, &p);
 
                         log_debug("Received SIG%s from PID %lu (%s).",
                                   signal_to_string(sfsi.ssi_signo),
index dd05f93d426ade6ddeea76a5b0d0cc5b3868fca4..78f9b30d5bffa8db425508cf5360570571d72374 100644 (file)
@@ -163,42 +163,24 @@ static int get_user_data(
 
         const char *username = NULL;
         struct passwd *pw = NULL;
+        uid_t uid;
         int r;
-        bool have_loginuid = false;
-        char *s;
 
         assert(handle);
         assert(ret_username);
         assert(ret_pw);
 
-        if (have_effective_cap(CAP_AUDIT_CONTROL) > 0) {
-                /* Only use audit login uid if we are executed with
-                 * sufficient capabilities so that pam_loginuid could
-                 * do its job. If we are lacking the CAP_AUDIT_CONTROL
-                 * capabality we most likely are being run in a
-                 * container and /proc/self/loginuid is useless since
-                 * it probably contains a uid of the host system. */
-
-                if (read_one_line_file("/proc/self/loginuid", &s) >= 0) {
-                        uid_t uid;
-
-                        r = parse_uid(s, &uid);
-                        free(s);
-
-                        if (r >= 0 && uid != (uint32_t) -1) {
-                                have_loginuid = true;
-                                pw = pam_modutil_getpwuid(handle, uid);
-                        }
-                }
-        }
-
-        if (!have_loginuid) {
-                if ((r = pam_get_user(handle, &username, NULL)) != PAM_SUCCESS) {
+        r = audit_loginuid_from_pid(0, &uid);
+        if (r >= 0)
+                pw = pam_modutil_getpwuid(handle, uid);
+        else {
+                r = pam_get_user(handle, &username, NULL);
+                if (r != PAM_SUCCESS) {
                         pam_syslog(handle, LOG_ERR, "Failed to get user name.");
                         return r;
                 }
 
-                if (!username || !*username) {
+                if (isempty(username)) {
                         pam_syslog(handle, LOG_ERR, "User name not valid.");
                         return PAM_AUTH_ERR;
                 }
diff --git a/src/sd-id128.c b/src/sd-id128.c
new file mode 100644 (file)
index 0000000..f5e0432
--- /dev/null
@@ -0,0 +1,210 @@
+/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
+
+/***
+  This file is part of systemd.
+
+  Copyright 2011 Lennart Poettering
+
+  systemd is free software; you can redistribute it and/or modify it
+  under the terms of the GNU General Public License as published by
+  the Free Software Foundation; either version 2 of the License, or
+  (at your option) any later version.
+
+  systemd is distributed in the hope that it will be useful, but
+  WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+  General Public License for more details.
+
+  You should have received a copy of the GNU General Public License
+  along with systemd; If not, see <http://www.gnu.org/licenses/>.
+***/
+
+#include <errno.h>
+#include <fcntl.h>
+#include <unistd.h>
+
+#include "sd-id128.h"
+#include "util.h"
+#include "macro.h"
+
+char *sd_id128_to_string(sd_id128_t id, char s[33]) {
+        unsigned n;
+
+        assert(s);
+
+        for (n = 0; n < 16; n++) {
+                s[n*2] = hexchar(id.bytes[n] >> 4);
+                s[n*2+1] = hexchar(id.bytes[n] & 0xF);
+        }
+
+        s[32] = 0;
+
+        return s;
+}
+
+int sd_id128_from_string(const char s[33], sd_id128_t *ret) {
+        unsigned n;
+        sd_id128_t t;
+
+        assert(s);
+        assert(ret);
+
+        for (n = 0; n < 16; n++) {
+                int a, b;
+
+                a = unhexchar(s[n*2]);
+                if (a < 0)
+                        return -EINVAL;
+
+                b = unhexchar(s[n*2+1]);
+                if (b < 0)
+                        return -EINVAL;
+
+                t.bytes[n] = (a << 4) | b;
+        }
+
+        if (s[32] != 0)
+                return -EINVAL;
+
+        *ret = t;
+        return 0;
+}
+
+sd_id128_t sd_id128_make_v4_uuid(sd_id128_t id) {
+        /* Stolen from generate_random_uuid() of drivers/char/random.c
+         * in the kernel sources */
+
+        /* Set UUID version to 4 --- truly random generation */
+        id.bytes[6] = (id.bytes[6] & 0x0F) | 0x40;
+
+        /* Set the UUID variant to DCE */
+        id.bytes[8] = (id.bytes[8] & 0x3F) | 0x80;
+
+        return id;
+}
+
+int sd_id128_get_machine(sd_id128_t *ret) {
+        static __thread sd_id128_t saved_machine_id;
+        static __thread bool saved_machine_id_valid = false;
+        int fd;
+        char buf[32];
+        ssize_t k;
+        unsigned j;
+        sd_id128_t t;
+
+        if (saved_machine_id_valid) {
+                *ret = saved_machine_id;
+                return 0;
+        }
+
+        fd = open("/etc/machine-id", O_RDONLY|O_CLOEXEC|O_NOCTTY);
+        if (fd < 0)
+                return -errno;
+
+        k = loop_read(fd, buf, 32, false);
+        close_nointr_nofail(fd);
+
+        if (k < 0)
+                return (int) k;
+
+        if (k < 32)
+                return -EIO;
+
+        for (j = 0; j < 16; j++) {
+                int a, b;
+
+                a = unhexchar(buf[j*2]);
+                b = unhexchar(buf[j*2+1]);
+
+                if (a < 0 || b < 0)
+                        return -EIO;
+
+                t.bytes[j] = a << 4 | b;
+        }
+
+        saved_machine_id = t;
+        saved_machine_id_valid = true;
+
+        *ret = t;
+        return 0;
+}
+
+int sd_id128_get_boot(sd_id128_t *ret) {
+        static __thread sd_id128_t saved_boot_id;
+        static __thread bool saved_boot_id_valid = false;
+        int fd;
+        char buf[36];
+        ssize_t k;
+        unsigned j;
+        sd_id128_t t;
+        char *p;
+
+        if (saved_boot_id_valid) {
+                *ret = saved_boot_id;
+                return 0;
+        }
+
+        fd = open("/proc/sys/kernel/random/boot_id", O_RDONLY|O_CLOEXEC|O_NOCTTY);
+        if (fd < 0)
+                return -errno;
+
+        k = loop_read(fd, buf, 36, false);
+        close_nointr_nofail(fd);
+
+        if (k < 0)
+                return (int) k;
+
+        if (k < 36)
+                return -EIO;
+
+        for (j = 0, p = buf; j < 16; j++) {
+                int a, b;
+
+                if (*p == '-')
+                        p++;
+
+                a = unhexchar(p[0]);
+                b = unhexchar(p[1]);
+
+                if (a < 0 || b < 0)
+                        return -EIO;
+
+                t.bytes[j] = a << 4 | b;
+
+                p += 2;
+        }
+
+        saved_boot_id = t;
+        saved_boot_id_valid = true;
+
+        *ret = t;
+        return 0;
+}
+
+int sd_id128_randomize(sd_id128_t *ret) {
+        int fd;
+        ssize_t k;
+        sd_id128_t t;
+
+        assert(ret);
+
+        fd = open("/dev/urandom", O_RDONLY|O_CLOEXEC|O_NOCTTY);
+        if (fd < 0)
+                return -errno;
+
+        k = loop_read(fd, &t, 16, false);
+        close_nointr_nofail(fd);
+
+        if (k < 0)
+                return (int) k;
+
+        if (k < 16)
+                return -EIO;
+
+        /* Turn this into a valid v4 UUID, to be nice. Note that we
+         * only guarantee this for newly generated UUIDs, not for
+         * pre-existing ones.*/
+
+        *ret = sd_id128_make_v4_uuid(t);
+        return 0;
+}
diff --git a/src/sd-id128.h b/src/sd-id128.h
new file mode 100644 (file)
index 0000000..bfae78b
--- /dev/null
@@ -0,0 +1,56 @@
+/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
+
+#ifndef fooid128hfoo
+#define fooid128hfoo
+
+/***
+  This file is part of systemd.
+
+  Copyright 2011 Lennart Poettering
+
+  systemd is free software; you can redistribute it and/or modify it
+  under the terms of the GNU General Public License as published by
+  the Free Software Foundation; either version 2 of the License, or
+  (at your option) any later version.
+
+  systemd is distributed in the hope that it will be useful, but
+  WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+  General Public License for more details.
+
+  You should have received a copy of the GNU General Public License
+  along with systemd; If not, see <http://www.gnu.org/licenses/>.
+***/
+
+#include <inttypes.h>
+#include <stdbool.h>
+#include <string.h>
+
+typedef union sd_id128 sd_id128_t;
+
+union sd_id128 {
+        uint8_t bytes[16];
+        uint64_t qwords[2];
+};
+
+char *sd_id128_to_string(sd_id128_t id, char s[33]);
+
+int sd_id128_from_string(const char s[33], sd_id128_t *ret);
+
+int sd_id128_randomize(sd_id128_t *ret);
+
+sd_id128_t sd_id128_make_v4_uuid(sd_id128_t id);
+
+int sd_id128_get_machine(sd_id128_t *ret);
+
+int sd_id128_get_boot(sd_id128_t *ret);
+
+#define SD_ID128_MAKE(v0, v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11, v12, v13, v14, v15) \
+        ((sd_id128_t) { .bytes = { 0x##v0, 0x##v1, 0x##v2, 0x##v3, 0x##v4, 0x##v5, 0x##v6, 0x##v7, \
+                                   0x##v8, 0x##v9, 0x##v10, 0x##v11, 0x##v12, 0x##v13, 0x##v14, 0x##v15 }})
+
+static inline bool sd_id128_equal(sd_id128_t a, sd_id128_t b) {
+        return memcmp(&a, &b, 16) == 0;
+}
+
+#endif
index 9a0408819ec964a1626cadcd2090268065c8323f..d50df22c88acf93d9770d938375eab82be62e4f4 100644 (file)
@@ -649,7 +649,8 @@ int main(int argc, char *argv[]) {
 
         umask(0022);
 
-        if ((n = sd_listen_fds(true)) < 0) {
+        n = sd_listen_fds(true);
+        if (n < 0) {
                 log_error("Failed to read listening file descriptors from environment: %s", strerror(-r));
                 return EXIT_FAILURE;
         }
index 175159d68fb42fc87380d72263d198bc9f3eadeb..18074402e57a11af47433943bbc132a72e22af7a 100644 (file)
@@ -2201,7 +2201,7 @@ static void print_status_info(UnitStatusInfo *i) {
 
                         if (i->running) {
                                 char *t = NULL;
-                                get_process_name(i->main_pid, &t);
+                                get_process_comm(i->main_pid, &t);
                                 if (t) {
                                         printf(" (%s)", t);
                                         free(t);
@@ -2235,7 +2235,7 @@ static void print_status_info(UnitStatusInfo *i) {
 
                         printf(" Control: %u", (unsigned) i->control_pid);
 
-                        get_process_name(i->control_pid, &t);
+                        get_process_comm(i->control_pid, &t);
                         if (t) {
                                 printf(" (%s)", t);
                                 free(t);
diff --git a/src/test-id128.c b/src/test-id128.c
new file mode 100644 (file)
index 0000000..6c3928d
--- /dev/null
@@ -0,0 +1,49 @@
+/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
+
+/***
+  This file is part of systemd.
+
+  Copyright 2011 Lennart Poettering
+
+  systemd is free software; you can redistribute it and/or modify it
+  under the terms of the GNU General Public License as published by
+  the Free Software Foundation; either version 2 of the License, or
+  (at your option) any later version.
+
+  systemd is distributed in the hope that it will be useful, but
+  WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+  General Public License for more details.
+
+  You should have received a copy of the GNU General Public License
+  along with systemd; If not, see <http://www.gnu.org/licenses/>.
+***/
+
+#include <string.h>
+
+#include "sd-id128.h"
+#include "util.h"
+#include "macro.h"
+
+#define ID128_WALDI SD_ID128_MAKE(01, 02, 03, 04, 05, 06, 07, 08, 09, 0a, 0b, 0c, 0d, 0e, 0f, 10)
+
+int main(int argc, char *argv[]) {
+        sd_id128_t id, id2;
+        char t[33];
+
+        assert_se(sd_id128_randomize(&id) == 0);
+        printf("random: %s\n", sd_id128_to_string(id, t));
+
+        assert_se(sd_id128_from_string(t, &id2) == 0);
+        assert_se(sd_id128_equal(id, id2));
+
+        assert_se(sd_id128_get_machine(&id) == 0);
+        printf("machine: %s\n", sd_id128_to_string(id, t));
+
+        assert_se(sd_id128_get_boot(&id) == 0);
+        printf("boot: %s\n", sd_id128_to_string(id, t));
+
+        printf("waldi: %s\n", sd_id128_to_string(ID128_WALDI, t));
+
+        return 0;
+}
index e93e6f6cf5c962e90ac7b8a4a29cc8cfe00a80ce..e5b5e53f7e1c8a5c19cc1e75539b5c2cf96eab3d 100644 (file)
@@ -55,6 +55,7 @@
 #include <linux/rtc.h>
 #include <glob.h>
 #include <grp.h>
+#include <sys/mman.h>
 
 #include "macro.h"
 #include "util.h"
@@ -73,7 +74,7 @@ size_t page_size(void) {
         static __thread size_t pgsz = 0;
         long r;
 
-        if (_likely_(pgsz))
+        if (_likely_(pgsz > 0))
                 return pgsz;
 
         assert_se((r = sysconf(_SC_PAGESIZE)) > 0);
@@ -993,46 +994,51 @@ char *truncate_nl(char *s) {
         return s;
 }
 
-int get_process_name(pid_t pid, char **name) {
-        char *p;
+int get_process_comm(pid_t pid, char **name) {
         int r;
 
-        assert(pid >= 1);
         assert(name);
 
-        if (asprintf(&p, "/proc/%lu/comm", (unsigned long) pid) < 0)
-                return -ENOMEM;
-
-        r = read_one_line_file(p, name);
-        free(p);
+        if (pid == 0)
+                r = read_one_line_file("/proc/self/comm", name);
+        else {
+                char *p;
+                if (asprintf(&p, "/proc/%lu/comm", (unsigned long) pid) < 0)
+                        return -ENOMEM;
 
-        if (r < 0)
-                return r;
+                r = read_one_line_file(p, name);
+                free(p);
+        }
 
-        return 0;
+        return r;
 }
 
-int get_process_cmdline(pid_t pid, size_t max_length, char **line) {
-        char *p, *r, *k;
+int get_process_cmdline(pid_t pid, size_t max_length, bool comm_fallback, char **line) {
+        char *r, *k;
         int c;
         bool space = false;
         size_t left;
         FILE *f;
 
-        assert(pid >= 1);
         assert(max_length > 0);
         assert(line);
 
-        if (asprintf(&p, "/proc/%lu/cmdline", (unsigned long) pid) < 0)
-                return -ENOMEM;
+        if (pid == 0)
+                f = fopen("/proc/self/cmdline", "re");
+        else {
+                char *p;
+                if (asprintf(&p, "/proc/%lu/cmdline", (unsigned long) pid) < 0)
+                        return -ENOMEM;
 
-        f = fopen(p, "re");
-        free(p);
+                f = fopen(p, "re");
+                free(p);
+        }
 
         if (!f)
                 return -errno;
 
-        if (!(r = new(char, max_length))) {
+        r = new(char, max_length);
+        if (!r) {
                 fclose(f);
                 return -ENOMEM;
         }
@@ -1076,13 +1082,17 @@ int get_process_cmdline(pid_t pid, size_t max_length, char **line) {
 
                 free(r);
 
-                if ((h = get_process_name(pid, &t)) < 0)
+                if (!comm_fallback)
+                        return -ENOENT;
+
+                h = get_process_comm(pid, &t);
+                if (h < 0)
                         return h;
 
-                h = asprintf(&r, "[%s]", t);
+                r = join("[", t, "]", NULL);
                 free(t);
 
-                if (h < 0)
+                if (!r)
                         return -ENOMEM;
         }
 
@@ -1090,6 +1100,25 @@ int get_process_cmdline(pid_t pid, size_t max_length, char **line) {
         return 0;
 }
 
+int get_process_exe(pid_t pid, char **name) {
+        int r;
+
+        assert(name);
+
+        if (pid == 0)
+                r = readlink_malloc("/proc/self/exe", name);
+        else {
+                char *p;
+                if (asprintf(&p, "/proc/%lu/exe", (unsigned long) pid) < 0)
+                        return -ENOMEM;
+
+                r = readlink_malloc(p, name);
+                free(p);
+        }
+
+        return r;
+}
+
 char *strnappend(const char *s, const char *suffix, size_t b) {
         size_t a;
         char *r;
@@ -3503,6 +3532,22 @@ int chmod_and_chown(const char *path, mode_t mode, uid_t uid, gid_t gid) {
         return 0;
 }
 
+int fchmod_and_fchown(int fd, mode_t mode, uid_t uid, gid_t gid) {
+        assert(fd >= 0);
+
+        /* Under the assumption that we are running privileged we
+         * first change the access mode and only then hand out
+         * ownership to avoid a window where access is too open. */
+
+        if (fchmod(fd, mode) < 0)
+                return -errno;
+
+        if (fchown(fd, uid, gid) < 0)
+                return -errno;
+
+        return 0;
+}
+
 cpu_set_t* cpu_set_malloc(unsigned *ncpus) {
         cpu_set_t *r;
         unsigned n = 1024;
@@ -4270,7 +4315,7 @@ const char *default_term_for_tty(const char *tty) {
         return term;
 }
 
-bool dirent_is_file(struct dirent *de) {
+bool dirent_is_file(const struct dirent *de) {
         assert(de);
 
         if (ignore_file(de->d_name))
@@ -4284,6 +4329,15 @@ bool dirent_is_file(struct dirent *de) {
         return true;
 }
 
+bool dirent_is_file_with_suffix(const struct dirent *de, const char *suffix) {
+        assert(de);
+
+        if (!dirent_is_file(de))
+                return false;
+
+        return endswith(de->d_name, suffix);
+}
+
 void execute_directory(const char *directory, DIR *d, char *argv[]) {
         DIR *_d = NULL;
         struct dirent *de;
@@ -4456,6 +4510,98 @@ void parse_syslog_priority(char **p, int *priority) {
         *p += k;
 }
 
+void skip_syslog_pid(char **buf) {
+        char *p;
+
+        assert(buf);
+        assert(*buf);
+
+        p = *buf;
+
+        if (*p != '[')
+                return;
+
+        p++;
+        p += strspn(p, "0123456789");
+
+        if (*p != ']')
+                return;
+
+        p++;
+
+        *buf = p;
+}
+
+void skip_syslog_date(char **buf) {
+        enum {
+                LETTER,
+                SPACE,
+                NUMBER,
+                SPACE_OR_NUMBER,
+                COLON
+        } sequence[] = {
+                LETTER, LETTER, LETTER,
+                SPACE,
+                SPACE_OR_NUMBER, NUMBER,
+                SPACE,
+                SPACE_OR_NUMBER, NUMBER,
+                COLON,
+                SPACE_OR_NUMBER, NUMBER,
+                COLON,
+                SPACE_OR_NUMBER, NUMBER,
+                SPACE
+        };
+
+        char *p;
+        unsigned i;
+
+        assert(buf);
+        assert(*buf);
+
+        p = *buf;
+
+        for (i = 0; i < ELEMENTSOF(sequence); i++, p++) {
+
+                if (!*p)
+                        return;
+
+                switch (sequence[i]) {
+
+                case SPACE:
+                        if (*p != ' ')
+                                return;
+                        break;
+
+                case SPACE_OR_NUMBER:
+                        if (*p == ' ')
+                                break;
+
+                        /* fall through */
+
+                case NUMBER:
+                        if (*p < '0' || *p > '9')
+                                return;
+
+                        break;
+
+                case LETTER:
+                        if (!(*p >= 'A' && *p <= 'Z') &&
+                            !(*p >= 'a' && *p <= 'z'))
+                                return;
+
+                        break;
+
+                case COLON:
+                        if (*p != ':')
+                                return;
+                        break;
+
+                }
+        }
+
+        *buf = p;
+}
+
 int have_effective_cap(int value) {
         cap_t cap;
         cap_flag_value_t fv;
@@ -4675,21 +4821,6 @@ int vt_disallocate(const char *name) {
         return 0;
 }
 
-
-static int file_is_conf(const struct dirent *d, const char *suffix) {
-        assert(d);
-
-        if (ignore_file(d->d_name))
-                return 0;
-
-        if (d->d_type != DT_REG &&
-            d->d_type != DT_LNK &&
-            d->d_type != DT_UNKNOWN)
-                return 0;
-
-        return endswith(d->d_name, suffix);
-}
-
 static int files_add(Hashmap *h, const char *path, const char *suffix) {
         DIR *dir;
         struct dirent buffer, *de;
@@ -4715,7 +4846,7 @@ static int files_add(Hashmap *h, const char *path, const char *suffix) {
                 if (!de)
                         break;
 
-                if (!file_is_conf(de, suffix))
+                if (!dirent_is_file_with_suffix(de, suffix))
                         continue;
 
                 if (asprintf(&p, "%s/%s", path, de->d_name) < 0) {
@@ -5066,21 +5197,27 @@ int symlink_or_copy_atomic(const char *from, const char *to) {
 }
 
 int audit_session_from_pid(pid_t pid, uint32_t *id) {
-        char *p, *s;
+        char *s;
         uint32_t u;
         int r;
 
-        assert(pid >= 1);
         assert(id);
 
         if (have_effective_cap(CAP_AUDIT_CONTROL) <= 0)
                 return -ENOENT;
 
-        if (asprintf(&p, "/proc/%lu/sessionid", (unsigned long) pid) < 0)
-                return -ENOMEM;
+        if (pid == 0)
+                r = read_one_line_file("/proc/self/sessionid", &s);
+        else {
+                char *p;
+
+                if (asprintf(&p, "/proc/%lu/sessionid", (unsigned long) pid) < 0)
+                        return -ENOMEM;
+
+                r = read_one_line_file(p, &s);
+                free(p);
+        }
 
-        r = read_one_line_file(p, &s);
-        free(p);
         if (r < 0)
                 return r;
 
@@ -5097,6 +5234,51 @@ int audit_session_from_pid(pid_t pid, uint32_t *id) {
         return 0;
 }
 
+int audit_loginuid_from_pid(pid_t pid, uid_t *uid) {
+        char *s;
+        uid_t u;
+        int r;
+
+        assert(uid);
+
+        /* Only use audit login uid if we are executed with sufficient
+         * capabilities so that pam_loginuid could do its job. If we
+         * are lacking the CAP_AUDIT_CONTROL capabality we most likely
+         * are being run in a container and /proc/self/loginuid is
+         * useless since it probably contains a uid of the host
+         * system. */
+
+        if (have_effective_cap(CAP_AUDIT_CONTROL) <= 0)
+                return -ENOENT;
+
+        if (pid == 0)
+                r = read_one_line_file("/proc/self/loginuid", &s);
+        else {
+                char *p;
+
+                if (asprintf(&p, "/proc/%lu/loginuid", (unsigned long) pid) < 0)
+                        return -ENOMEM;
+
+                r = read_one_line_file(p, &s);
+                free(p);
+        }
+
+        if (r < 0)
+                return r;
+
+        r = parse_uid(s, &u);
+        free(s);
+
+        if (r < 0)
+                return r;
+
+        if (u == (uid_t) -1)
+                return -ENOENT;
+
+        *uid = (uid_t) u;
+        return 0;
+}
+
 bool display_is_local(const char *display) {
         assert(display);
 
@@ -5704,6 +5886,24 @@ int strdup_or_null(const char *a, char **b) {
         return 0;
 }
 
+int prot_from_flags(int flags) {
+
+        switch (flags & O_ACCMODE) {
+
+        case O_RDONLY:
+                return PROT_READ;
+
+        case O_WRONLY:
+                return PROT_WRITE;
+
+        case O_RDWR:
+                return PROT_READ|PROT_WRITE;
+
+        default:
+                return -EINVAL;
+        }
+}
+
 unsigned long cap_last_cap(void) {
         static __thread unsigned long saved;
         static __thread bool valid = false;
index a71a297eab6b56740efe070d3473dd3118b183f3..1db82f83e0a4ba7d2b2588d1c996507dbc99ed83 100644 (file)
@@ -248,8 +248,9 @@ int parent_of_path(const char *path, char **parent);
 
 int rmdir_parents(const char *path, const char *stop);
 
-int get_process_name(pid_t pid, char **name);
-int get_process_cmdline(pid_t pid, size_t max_length, char **line);
+int get_process_comm(pid_t pid, char **name);
+int get_process_cmdline(pid_t pid, size_t max_length, bool comm_fallback, char **line);
+int get_process_exe(pid_t pid, char **name);
 
 char hexchar(int x);
 int unhexchar(char c);
@@ -274,7 +275,9 @@ bool path_equal(const char *a, const char *b);
 
 char *ascii_strlower(char *path);
 
-bool dirent_is_file(struct dirent *de);
+bool dirent_is_file(const struct dirent *de);
+bool dirent_is_file_with_suffix(const struct dirent *de, const char *suffix);
+
 bool ignore_file(const char *filename);
 
 bool chars_intersect(const char *a, const char *b);
@@ -363,6 +366,7 @@ int get_ctty_devnr(pid_t pid, dev_t *d);
 int get_ctty(pid_t, dev_t *_devnr, char **r);
 
 int chmod_and_chown(const char *path, mode_t mode, uid_t uid, gid_t gid);
+int fchmod_and_fchown(int fd, mode_t mode, uid_t uid, gid_t gid);
 
 int rm_rf(const char *path, bool only_dirs, bool delete_root, bool honour_sticky);
 
@@ -415,6 +419,8 @@ bool nulstr_contains(const char*nulstr, const char *needle);
 bool plymouth_running(void);
 
 void parse_syslog_priority(char **p, int *priority);
+void skip_syslog_pid(char **buf);
+void skip_syslog_date(char **buf);
 
 int have_effective_cap(int value);
 
@@ -443,6 +449,7 @@ int hwclock_get_time(struct tm *tm);
 int hwclock_set_time(const struct tm *tm);
 
 int audit_session_from_pid(pid_t pid, uint32_t *id);
+int audit_loginuid_from_pid(pid_t pid, uid_t *uid);
 
 bool display_is_local(const char *display);
 int socket_from_display(const char *display, char **path);
@@ -506,6 +513,8 @@ extern char **saved_argv;
 
 bool kexec_loaded(void);
 
+int prot_from_flags(int flags);
+
 unsigned long cap_last_cap(void);
 
 #endif
diff --git a/tmpfiles.d/Makefile b/tmpfiles.d/Makefile
new file mode 120000 (symlink)
index 0000000..bd10475
--- /dev/null
@@ -0,0 +1 @@
+../src/Makefile
\ No newline at end of file