chiark / gitweb /
util: replace close_nointr_nofail() by a more useful safe_close()
[elogind.git] / src / journal / sd-journal.c
index 09ce949d3bd21a1f5e6d9be65f69e2af504f8274..7587211506f5f02d5503dc1c730201ebf88450aa 100644 (file)
 #include <stddef.h>
 #include <unistd.h>
 #include <sys/inotify.h>
+#include <sys/poll.h>
+#include <sys/vfs.h>
+#include <linux/magic.h>
 
 #include "sd-journal.h"
 #include "journal-def.h"
 #include "journal-file.h"
 #include "hashmap.h"
 #include "list.h"
+#include "strv.h"
 #include "path-util.h"
 #include "lookup3.h"
 #include "compress.h"
 #include "journal-internal.h"
+#include "missing.h"
+#include "catalog.h"
+#include "replace-var.h"
+#include "fileio.h"
 
 #define JOURNAL_FILES_MAX 1024
 
+#define JOURNAL_FILES_RECHECK_USEC (2 * USEC_PER_SEC)
+
+#define REPLACE_VAR_MAX 256
+
+#define DEFAULT_DATA_THRESHOLD (64*1024)
+
+static void remove_file_real(sd_journal *j, JournalFile *f);
+
+static bool journal_pid_changed(sd_journal *j) {
+        assert(j);
+
+        /* We don't support people creating a journal object and
+         * keeping it around over a fork(). Let's complain. */
+
+        return j->original_pid != getpid();
+}
+
+/* We return an error here only if we didn't manage to
+   memorize the real error. */
+static int set_put_error(sd_journal *j, int r) {
+        int k;
+
+        if (r >= 0)
+                return r;
+
+        k = set_ensure_allocated(&j->errors, trivial_hash_func, trivial_compare_func);
+        if (k < 0)
+                return k;
+
+        return set_put(j->errors, INT_TO_PTR(r));
+}
+
 static void detach_location(sd_journal *j) {
         Iterator i;
         JournalFile *f;
@@ -57,12 +97,13 @@ static void reset_location(sd_journal *j) {
         zero(j->current_location);
 }
 
-static void init_location(Location *l, JournalFile *f, Object *o) {
+static void init_location(Location *l, LocationType type, JournalFile *f, Object *o) {
         assert(l);
+        assert(type == LOCATION_DISCRETE || type == LOCATION_SEEK);
         assert(f);
         assert(o->object.type == OBJECT_ENTRY);
 
-        l->type = LOCATION_DISCRETE;
+        l->type = type;
         l->seqnum = le64toh(o->entry.seqnum);
         l->seqnum_id = f->header->seqnum_id;
         l->realtime = le64toh(o->entry.realtime);
@@ -73,124 +114,326 @@ static void init_location(Location *l, JournalFile *f, Object *o) {
         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) {
+static void set_location(sd_journal *j, LocationType type, JournalFile *f, Object *o,
+                         direction_t direction, uint64_t offset) {
         assert(j);
+        assert(type == LOCATION_DISCRETE || type == LOCATION_SEEK);
         assert(f);
         assert(o);
 
-        init_location(&j->current_location, f, o);
+        init_location(&j->current_location, type, f, o);
 
         j->current_file = f;
         j->current_field = 0;
 
+        f->last_direction = direction;
         f->current_offset = offset;
 }
 
-static int same_field(const void *_a, size_t s, const void *_b, size_t t) {
+static int match_is_valid(const void *data, size_t size) {
+        const char *b, *p;
+
+        assert(data);
+
+        if (size < 2)
+                return false;
+
+        if (startswith(data, "__"))
+                return false;
+
+        b = data;
+        for (p = b; p < b + size; p++) {
+
+                if (*p == '=')
+                        return p > b;
+
+                if (*p == '_')
+                        continue;
+
+                if (*p >= 'A' && *p <= 'Z')
+                        continue;
+
+                if (*p >= '0' && *p <= '9')
+                        continue;
+
+                return false;
+        }
+
+        return false;
+}
+
+static bool 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;
+                        return false;
 
-                if (a_good && b_good)
-                        return different ? 0 : 1;
+                if (a[j] == '=')
+                        return true;
+        }
+
+        assert_not_reached("\"=\" not found");
+}
+
+static Match *match_new(Match *p, MatchType t) {
+        Match *m;
+
+        m = new0(Match, 1);
+        if (!m)
+                return NULL;
+
+        m->type = t;
+
+        if (p) {
+                m->parent = p;
+                LIST_PREPEND(matches, p->matches, m);
         }
 
-        return -EINVAL;
+        return m;
+}
+
+static void match_free(Match *m) {
+        assert(m);
+
+        while (m->matches)
+                match_free(m->matches);
+
+        if (m->parent)
+                LIST_REMOVE(matches, m->parent->matches, m);
+
+        free(m->data);
+        free(m);
+}
+
+static void match_free_if_empty(Match *m) {
+        if (!m || m->matches)
+                return;
+
+        match_free(m);
 }
 
 _public_ int sd_journal_add_match(sd_journal *j, const void *data, size_t size) {
-        Match *m, *after = NULL;
+        Match *l3, *l4, *add_here = NULL, *m;
         le64_t le_hash;
 
-        if (!j)
-                return -EINVAL;
-        if (!data)
-                return -EINVAL;
-        if (size <= 0)
-                return -EINVAL;
+        assert_return(j, -EINVAL);
+        assert_return(!journal_pid_changed(j), -ECHILD);
+        assert_return(data, -EINVAL);
+
+        if (size == 0)
+                size = strlen(data);
+
+        assert_return(match_is_valid(data, size), -EINVAL);
+
+        /* level 0: AND term
+         * level 1: OR terms
+         * level 2: AND terms
+         * level 3: OR terms
+         * level 4: concrete matches */
+
+        if (!j->level0) {
+                j->level0 = match_new(NULL, MATCH_AND_TERM);
+                if (!j->level0)
+                        return -ENOMEM;
+        }
+
+        if (!j->level1) {
+                j->level1 = match_new(j->level0, MATCH_OR_TERM);
+                if (!j->level1)
+                        return -ENOMEM;
+        }
+
+        if (!j->level2) {
+                j->level2 = match_new(j->level1, MATCH_AND_TERM);
+                if (!j->level2)
+                        return -ENOMEM;
+        }
 
-        /* FIXME: iterating with multiple matches is currently
-         * broken */
-        if (j->matches)
-                return -ENOTSUP;
+        assert(j->level0->type == MATCH_AND_TERM);
+        assert(j->level1->type == MATCH_OR_TERM);
+        assert(j->level2->type == MATCH_AND_TERM);
 
         le_hash = htole64(hash64(data, size));
 
-        LIST_FOREACH(matches, m, j->matches) {
-                int r;
+        LIST_FOREACH(matches, l3, j->level2->matches) {
+                assert(l3->type == MATCH_OR_TERM);
 
-                if (m->le_hash == le_hash &&
-                    m->size == size &&
-                    memcmp(m->data, data, size) == 0)
-                        return 0;
+                LIST_FOREACH(matches, l4, l3->matches) {
+                        assert(l4->type == MATCH_DISCRETE);
 
-                r = same_field(data, size, m->data, m->size);
-                if (r < 0)
-                        return r;
-                else if (r > 0)
-                        after = m;
+                        /* Exactly the same match already? Then ignore
+                         * this addition */
+                        if (l4->le_hash == le_hash &&
+                            l4->size == size &&
+                            memcmp(l4->data, data, size) == 0)
+                                return 0;
+
+                        /* Same field? Then let's add this to this OR term */
+                        if (same_field(data, size, l4->data, l4->size)) {
+                                add_here = l3;
+                                break;
+                        }
+                }
+
+                if (add_here)
+                        break;
         }
 
-        m = new0(Match, 1);
+        if (!add_here) {
+                add_here = match_new(j->level2, MATCH_OR_TERM);
+                if (!add_here)
+                        goto fail;
+        }
+
+        m = match_new(add_here, MATCH_DISCRETE);
         if (!m)
-                return -ENOMEM;
+                goto fail;
 
+        m->le_hash = le_hash;
         m->size = size;
+        m->data = memdup(data, size);
+        if (!m->data)
+                goto fail;
 
-        m->data = malloc(m->size);
-        if (!m->data) {
-                free(m);
-                return -ENOMEM;
-        }
+        detach_location(j);
 
-        memcpy(m->data, data, size);
-        m->le_hash = le_hash;
+        return 0;
+
+fail:
+        match_free_if_empty(add_here);
+        match_free_if_empty(j->level2);
+        match_free_if_empty(j->level1);
+        match_free_if_empty(j->level0);
+
+        return -ENOMEM;
+}
 
-        /* Matches for the same fields we order adjacent to each
-         * other */
-        LIST_INSERT_AFTER(Match, matches, j->matches, after, m);
-        j->n_matches ++;
+_public_ int sd_journal_add_conjunction(sd_journal *j) {
+        assert_return(j, -EINVAL);
+        assert_return(!journal_pid_changed(j), -ECHILD);
 
-        detach_location(j);
+        if (!j->level0)
+                return 0;
+
+        if (!j->level1)
+                return 0;
+
+        if (!j->level1->matches)
+                return 0;
+
+        j->level1 = NULL;
+        j->level2 = NULL;
+
+        return 0;
+}
+
+_public_ int sd_journal_add_disjunction(sd_journal *j) {
+        assert_return(j, -EINVAL);
+        assert_return(!journal_pid_changed(j), -ECHILD);
+
+        if (!j->level0)
+                return 0;
+
+        if (!j->level1)
+                return 0;
+
+        if (!j->level2)
+                return 0;
 
+        if (!j->level2->matches)
+                return 0;
+
+        j->level2 = NULL;
         return 0;
 }
 
+static char *match_make_string(Match *m) {
+        char *p, *r;
+        Match *i;
+        bool enclose = false;
+
+        if (!m)
+                return strdup("none");
+
+        if (m->type == MATCH_DISCRETE)
+                return strndup(m->data, m->size);
+
+        p = NULL;
+        LIST_FOREACH(matches, i, m->matches) {
+                char *t, *k;
+
+                t = match_make_string(i);
+                if (!t) {
+                        free(p);
+                        return NULL;
+                }
+
+                if (p) {
+                        k = strjoin(p, m->type == MATCH_OR_TERM ? " OR " : " AND ", t, NULL);
+                        free(p);
+                        free(t);
+
+                        if (!k)
+                                return NULL;
+
+                        p = k;
+
+                        enclose = true;
+                } else
+                        p = t;
+        }
+
+        if (enclose) {
+                r = strjoin("(", p, ")", NULL);
+                free(p);
+                return r;
+        }
+
+        return p;
+}
+
+char *journal_make_match_string(sd_journal *j) {
+        assert(j);
+
+        return match_make_string(j->level0);
+}
+
 _public_ void sd_journal_flush_matches(sd_journal *j) {
         if (!j)
                 return;
 
-        while (j->matches) {
-                Match *m = j->matches;
-
-                LIST_REMOVE(Match, matches, j->matches, m);
-                free(m->data);
-                free(m);
-        }
+        if (j->level0)
+                match_free(j->level0);
 
-        j->n_matches = 0;
+        j->level0 = j->level1 = j->level2 = NULL;
 
         detach_location(j);
 }
 
-static int compare_order(JournalFile *af, Object *ao,
-                         JournalFile *bf, Object *bo) {
+static int compare_entry_order(JournalFile *af, Object *_ao,
+                               JournalFile *bf, uint64_t bp) {
 
         uint64_t a, b;
+        Object *ao, *bo;
+        int r;
 
         assert(af);
-        assert(ao);
         assert(bf);
-        assert(bo);
+        assert(_ao);
+
+        /* The mmap cache might invalidate the object from the first
+         * file if we look at the one from the second file. Hence
+         * temporarily copy the header of the first one, and look at
+         * that only. */
+        ao = alloca(offsetof(EntryObject, items));
+        memcpy(ao, _ao, offsetof(EntryObject, items));
+
+        r = journal_file_move_to_object(bf, OBJECT_ENTRY, bp, &bo);
+        if (r < 0)
+                return strcmp(af->path, bf->path);
 
         /* We operate on two different files here, hence we can access
          * two objects at the same time, which we normally can't.
@@ -223,7 +466,7 @@ static int compare_order(JournalFile *af, Object *ao,
 
         if (sd_id128_equal(ao->entry.boot_id, bo->entry.boot_id)) {
 
-                /* If the boot id matches compare monotonic time */
+                /* If the boot id matches, compare monotonic time */
                 a = le64toh(ao->entry.monotonic);
                 b = le64toh(bo->entry.monotonic);
 
@@ -233,9 +476,9 @@ static int compare_order(JournalFile *af, Object *ao,
                         return 1;
         }
 
-        /* Otherwise compare UTC time */
+        /* Otherwise, compare UTC time */
         a = le64toh(ao->entry.realtime);
-        b = le64toh(ao->entry.realtime);
+        b = le64toh(bo->entry.realtime);
 
         if (a < b)
                 return -1;
@@ -244,7 +487,7 @@ static int compare_order(JournalFile *af, Object *ao,
 
         /* Finally, compare by contents */
         a = le64toh(ao->entry.xor_hash);
-        b = le64toh(ao->entry.xor_hash);
+        b = le64toh(bo->entry.xor_hash);
 
         if (a < b)
                 return -1;
@@ -254,13 +497,13 @@ static int compare_order(JournalFile *af, Object *ao,
         return 0;
 }
 
-static int compare_with_location(JournalFile *af, Object *ao, Location *l) {
+_pure_ 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);
+        assert(l->type == LOCATION_DISCRETE || l->type == LOCATION_SEEK);
 
         if (l->monotonic_set &&
             sd_id128_equal(ao->entry.boot_id, l->boot_id) &&
@@ -314,314 +557,291 @@ static int compare_with_location(JournalFile *af, Object *ao, Location *l) {
         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;
+static int next_for_match(
+                sd_journal *j,
+                Match *m,
+                JournalFile *f,
+                uint64_t after_offset,
+                direction_t direction,
+                Object **ret,
+                uint64_t *offset) {
+
         int r;
+        uint64_t np = 0;
+        Object *n;
 
         assert(j);
+        assert(m);
+        assert(f);
 
-        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);
-
-                        if (r == -ENOENT) {
-                                /* boot id unknown in this file */
-                                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);
-                        }
-                } 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 (m->type == MATCH_DISCRETE) {
+                uint64_t dp;
 
+                r = journal_file_find_data_object_with_hash(f, m->data, m->size, le64toh(m->le_hash), NULL, &dp);
                 if (r <= 0)
                         return r;
 
-        } else  {
-                Match *m, *term_match = NULL;
-                Object *to = NULL;
-                uint64_t tp = 0;
+                return journal_file_move_to_entry_by_offset_for_data(f, dp, after_offset, direction, ret, offset);
 
-                /* We have matches, first, let's jump to the monotonic
-                 * position if we have any, since it implies a
-                 * match. */
+        } else if (m->type == MATCH_OR_TERM) {
+                Match *i;
 
-                if (j->current_location.type == LOCATION_DISCRETE &&
-                    j->current_location.monotonic_set) {
+                /* Find the earliest match beyond after_offset */
 
-                        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 == -ENOENT ? 0 : 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, le64toh(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);
+                LIST_FOREACH(matches, i, m->matches) {
+                        uint64_t cp;
 
+                        r = next_for_match(j, i, f, after_offset, direction, NULL, &cp);
                         if (r < 0)
                                 return r;
+                        else if (r > 0) {
+                                if (np == 0 || (direction == DIRECTION_DOWN ? cp < np : cp > np))
+                                        np = 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) {
+                if (np == 0)
+                        return 0;
 
-                                        /* Find the earliest of the OR matches */
+        } else if (m->type == MATCH_AND_TERM) {
+                Match *i, *last_moved;
 
-                                        if (!to ||
-                                            (direction == DIRECTION_DOWN && cp < tp) ||
-                                            (direction == DIRECTION_UP && cp > tp)) {
-                                                to = c;
-                                                tp = cp;
-                                        }
+                /* Always jump to the next matching entry and repeat
+                 * this until we find an offset that matches for all
+                 * matches. */
 
-                                }
+                if (!m->matches)
+                        return 0;
 
-                        } else {
+                r = next_for_match(j, m->matches, f, after_offset, direction, NULL, &np);
+                if (r <= 0)
+                        return r;
 
-                                /* Previous term is finished, did anything match? */
-                                if (!to)
-                                        return 0;
+                assert(direction == DIRECTION_DOWN ? np >= after_offset : np <= after_offset);
+                last_moved = m->matches;
 
-                                /* Find the last of the AND matches */
-                                if (!o ||
-                                    (direction == DIRECTION_DOWN && tp > p) ||
-                                    (direction == DIRECTION_UP && tp < p)) {
-                                        o = to;
-                                        p = tp;
-                                }
+                LIST_LOOP_BUT_ONE(matches, i, m->matches, last_moved) {
+                        uint64_t cp;
 
-                                term_match = m;
+                        r = next_for_match(j, i, f, np, direction, NULL, &cp);
+                        if (r <= 0)
+                                return r;
 
-                                if (r > 0) {
-                                        to = c;
-                                        tp = cp;
-                                } else {
-                                        to = NULL;
-                                        tp = 0;
-                                }
+                        assert(direction == DIRECTION_DOWN ? cp >= np : cp <= np);
+                        if (direction == DIRECTION_DOWN ? cp > np : cp < np) {
+                                np = cp;
+                                last_moved = i;
                         }
                 }
+        }
 
-                /* 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;
-                }
+        assert(np > 0);
 
-                if (!o)
-                        return 0;
-        }
+        r = journal_file_move_to_object(f, OBJECT_ENTRY, np, &n);
+        if (r < 0)
+                return r;
 
         if (ret)
-                *ret = o;
-
+                *ret = n;
         if (offset)
-                *offset = p;
+                *offset = np;
 
         return 1;
 }
 
-static int next_with_matches(sd_journal *j, JournalFile *f, direction_t direction, Object **ret, uint64_t *offset) {
+static int find_location_for_match(
+                sd_journal *j,
+                Match *m,
+                JournalFile *f,
+                direction_t direction,
+                Object **ret,
+                uint64_t *offset) {
+
         int r;
-        uint64_t cp;
-        Object *c;
 
         assert(j);
+        assert(m);
         assert(f);
-        assert(ret);
-        assert(offset);
-
-        c = *ret;
-        cp = *offset;
 
-        if (!j->matches) {
-                /* No matches is easy */
+        if (m->type == MATCH_DISCRETE) {
+                uint64_t dp;
 
-                r = journal_file_next_entry(f, c, cp, direction, &c, &cp);
+                r = journal_file_find_data_object_with_hash(f, m->data, m->size, le64toh(m->le_hash), NULL, &dp);
                 if (r <= 0)
                         return r;
 
-                if (ret)
-                        *ret = c;
-                if (offset)
-                        *offset = cp;
-                return 1;
-        }
+                /* FIXME: missing: find by monotonic */
 
-        /* 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;
-                Object *npo = 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;
-                        Object *qo = NULL;
-
-                        /* 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;
-                        }
+                if (j->current_location.type == LOCATION_HEAD)
+                        return journal_file_next_entry_for_data(f, NULL, 0, dp, DIRECTION_DOWN, ret, offset);
+                if (j->current_location.type == LOCATION_TAIL)
+                        return journal_file_next_entry_for_data(f, NULL, 0, dp, DIRECTION_UP, ret, offset);
+                if (j->current_location.seqnum_set && sd_id128_equal(j->current_location.seqnum_id, f->header->seqnum_id))
+                        return journal_file_move_to_entry_by_seqnum_for_data(f, dp, j->current_location.seqnum, direction, ret, offset);
+                if (j->current_location.monotonic_set) {
+                        r = journal_file_move_to_entry_by_monotonic_for_data(f, dp, j->current_location.boot_id, j->current_location.monotonic, direction, ret, offset);
+                        if (r != -ENOENT)
+                                return r;
+                }
+                if (j->current_location.realtime_set)
+                        return journal_file_move_to_entry_by_realtime_for_data(f, dp, j->current_location.realtime, direction, ret, offset);
 
-                        for (k = 0; k < n; k++)
-                                if (c->entry.items[k].hash == m->le_hash)
-                                        break;
+                return journal_file_next_entry_for_data(f, NULL, 0, dp, direction, ret, offset);
 
-                        if (k >= n) {
-                                /* Hmm, didn't find any field that
-                                 * matched this rule, so ignore this
-                                 * match. Go on with next match */
-                                continue;
-                        }
+        } else if (m->type == MATCH_OR_TERM) {
+                uint64_t np = 0;
+                Object *n;
+                Match *i;
 
-                        term_result = true;
+                /* Find the earliest match */
 
-                        /* Hmm, so, this field matched, let's remember
-                         * where we'd have to try next, in case the other
-                         * matches are not OK */
+                LIST_FOREACH(matches, i, m->matches) {
+                        uint64_t cp;
 
-                        r = journal_file_next_entry_for_data(f, c, cp, le64toh(c->entry.items[k].object_offset), direction, &qo, &q);
-                        /* This pointer is invalidated if the window was
-                         * remapped. May need to re-fetch it later */
-                        c = NULL;
+                        r = find_location_for_match(j, i, f, direction, NULL, &cp);
                         if (r < 0)
                                 return r;
-
-                        if (r > 0) {
-
-                                if (direction == DIRECTION_DOWN) {
-                                        if (q > np) {
-                                                np = q;
-                                                npo = qo;
-                                        }
-                                } else {
-                                        if (np == 0 || q < np) {
-                                                np = q;
-                                                npo = qo;
-                                        }
-                                }
+                        else if (r > 0) {
+                                if (np == 0 || (direction == DIRECTION_DOWN ? np > cp : np < cp))
+                                        np = cp;
                         }
                 }
 
-                /* Check the last term */
-                if (term_match && !term_result)
-                        found = false;
+                if (np == 0)
+                        return 0;
+
+                r = journal_file_move_to_object(f, OBJECT_ENTRY, np, &n);
+                if (r < 0)
+                        return r;
 
-                /* Did this entry match against all matches? */
-                if (found) {
-                        if (ret) {
-                                if (c == NULL) {
-                                        /* Re-fetch the entry */
-                                        r = journal_file_move_to_object(f, OBJECT_ENTRY, cp, &c);
-                                        if (r < 0)
-                                                return r;
-                                }
-                                *ret = c;
-                        }
-                        if (offset)
-                                *offset = cp;
-                        return 1;
-                }
+                if (ret)
+                        *ret = n;
+                if (offset)
+                        *offset = np;
 
-                /* Did we find a subsequent entry? */
-                if (np == 0)
+                return 1;
+
+        } else {
+                Match *i;
+                uint64_t np = 0;
+
+                assert(m->type == MATCH_AND_TERM);
+
+                /* First jump to the last match, and then find the
+                 * next one where all matches match */
+
+                if (!m->matches)
                         return 0;
 
-                /* Hmm, ok, this entry only matched partially, so
-                 * let's try another one */
-                cp = np;
-                c = npo;
+                LIST_FOREACH(matches, i, m->matches) {
+                        uint64_t cp;
+
+                        r = find_location_for_match(j, i, f, direction, NULL, &cp);
+                        if (r <= 0)
+                                return r;
+
+                        if (np == 0 || (direction == DIRECTION_DOWN ? cp > np : cp < np))
+                                np = cp;
+                }
+
+                return next_for_match(j, m, f, np, direction, ret, offset);
         }
 }
 
-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;
+static int find_location_with_matches(
+                sd_journal *j,
+                JournalFile *f,
+                direction_t direction,
+                Object **ret,
+                uint64_t *offset) {
+
+        int r;
 
         assert(j);
         assert(f);
+        assert(ret);
+        assert(offset);
 
-        if (f->current_offset > 0) {
-                cp = f->current_offset;
+        if (!j->level0) {
+                /* No matches is simple */
 
-                r = journal_file_move_to_object(f, OBJECT_ENTRY, cp, &c);
-                if (r < 0)
-                        return r;
+                if (j->current_location.type == LOCATION_HEAD)
+                        return journal_file_next_entry(f, NULL, 0, DIRECTION_DOWN, ret, offset);
+                if (j->current_location.type == LOCATION_TAIL)
+                        return journal_file_next_entry(f, NULL, 0, DIRECTION_UP, ret, offset);
+                if (j->current_location.seqnum_set && sd_id128_equal(j->current_location.seqnum_id, f->header->seqnum_id))
+                        return journal_file_move_to_entry_by_seqnum(f, j->current_location.seqnum, direction, ret, offset);
+                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, ret, offset);
+                        if (r != -ENOENT)
+                                return r;
+                }
+                if (j->current_location.realtime_set)
+                        return journal_file_move_to_entry_by_realtime(f, j->current_location.realtime, direction, ret, offset);
 
-                r = next_with_matches(j, f, direction, &c, &cp);
-                if (r <= 0)
-                        return r;
+                return journal_file_next_entry(f, NULL, 0, direction, ret, offset);
+        } else
+                return find_location_for_match(j, j->level0, f, direction, ret, offset);
+}
 
-                compare_value = 1;
-        } else {
-                r = find_location(j, f, direction, &c, &cp);
-                if (r <= 0)
-                        return r;
+static int next_with_matches(
+                sd_journal *j,
+                JournalFile *f,
+                direction_t direction,
+                Object **ret,
+                uint64_t *offset) {
 
-                compare_value = 0;
+        Object *c;
+        uint64_t cp;
+
+        assert(j);
+        assert(f);
+        assert(ret);
+        assert(offset);
+
+        c = *ret;
+        cp = *offset;
+
+        /* No matches is easy. We simple advance the file
+         * pointer by one. */
+        if (!j->level0)
+                return journal_file_next_entry(f, c, cp, direction, ret, offset);
+
+        /* If we have a match then we look for the next matching entry
+         * with an offset at least one step larger */
+        return next_for_match(j, j->level0, f, direction == DIRECTION_DOWN ? cp+1 : cp-1, direction, ret, offset);
+}
+
+static int next_beyond_location(sd_journal *j, JournalFile *f, direction_t direction, Object **ret, uint64_t *offset) {
+        Object *c;
+        uint64_t cp;
+        int r;
+
+        assert(j);
+        assert(f);
+
+        if (f->last_direction == direction && f->current_offset > 0) {
+                cp = f->current_offset;
+
+                r = journal_file_move_to_object(f, OBJECT_ENTRY, cp, &c);
+                if (r < 0)
+                        return r;
+
+                r = next_with_matches(j, f, direction, &c, &cp);
+                if (r <= 0)
+                        return r;
+        } else {
+                r = find_location_with_matches(j, f, direction, &c, &cp);
+                if (r <= 0)
+                        return r;
         }
 
+        /* OK, we found the spot, now let's advance until an entry
+         * that is actually different from what we were previously
+         * looking at. This is necessary to handle entries which exist
+         * in two (or more) journal files, and which shall all be
+         * suppressed but one. */
+
         for (;;) {
                 bool found;
 
@@ -630,9 +850,9 @@ static int next_beyond_location(sd_journal *j, JournalFile *f, direction_t direc
 
                         k = compare_with_location(f, c, &j->current_location);
                         if (direction == DIRECTION_DOWN)
-                                found = k >= compare_value;
+                                found = k > 0;
                         else
-                                found = k <= -compare_value;
+                                found = k < 0;
                 } else
                         found = true;
 
@@ -651,50 +871,51 @@ static int next_beyond_location(sd_journal *j, JournalFile *f, direction_t direc
 }
 
 static int real_journal_next(sd_journal *j, direction_t direction) {
-        JournalFile *f, *new_current = NULL;
+        JournalFile *f, *new_file = NULL;
+        uint64_t new_offset = 0;
+        uint64_t p = 0;
         Iterator i;
+        Object *o;
         int r;
-        uint64_t new_offset = 0;
-        Object *new_entry = NULL;
 
-        if (!j)
-                return -EINVAL;
+        assert_return(j, -EINVAL);
+        assert_return(!journal_pid_changed(j), -ECHILD);
 
         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)
+                if (r < 0) {
+                        log_debug("Can't iterate through %s, ignoring: %s", f->path, strerror(-r));
+                        remove_file_real(j, f);
+                        continue;
+                } else if (r == 0)
                         continue;
 
-                if (!new_current)
+                if (!new_file)
                         found = true;
                 else {
                         int k;
 
-                        k = compare_order(f, o, new_current, new_entry);
+                        k = compare_entry_order(f, o, new_file, new_offset);
 
-                        if (direction == DIRECTION_DOWN)
-                                found = k < 0;
-                        else
-                                found = k > 0;
+                        found = direction == DIRECTION_DOWN ? k < 0 : k > 0;
                 }
 
                 if (found) {
-                        new_current = f;
-                        new_entry = o;
+                        new_file = f;
                         new_offset = p;
                 }
         }
 
-        if (!new_current)
+        if (!new_file)
                 return 0;
 
-        set_location(j, new_current, new_entry, new_offset);
+        r = journal_file_move_to_object(new_file, OBJECT_ENTRY, new_offset, &o);
+        if (r < 0)
+                return r;
+
+        set_location(j, LOCATION_DISCRETE, new_file, o, direction, new_offset);
 
         return 1;
 }
@@ -710,8 +931,8 @@ _public_ int sd_journal_previous(sd_journal *j) {
 static int real_journal_next_skip(sd_journal *j, direction_t direction, uint64_t skip) {
         int c = 0, r;
 
-        if (!j)
-                return -EINVAL;
+        assert_return(j, -EINVAL);
+        assert_return(!journal_pid_changed(j), -ECHILD);
 
         if (skip == 0) {
                 /* If this is not a discrete skip, then at least
@@ -750,10 +971,9 @@ _public_ int sd_journal_get_cursor(sd_journal *j, char **cursor) {
         int r;
         char bid[33], sid[33];
 
-        if (!j)
-                return -EINVAL;
-        if (!cursor)
-                return -EINVAL;
+        assert_return(j, -EINVAL);
+        assert_return(!journal_pid_changed(j), -ECHILD);
+        assert_return(cursor, -EINVAL);
 
         if (!j->current_file || j->current_file->current_offset <= 0)
                 return -EADDRNOTAVAIL;
@@ -766,21 +986,19 @@ _public_ int sd_journal_get_cursor(sd_journal *j, char **cursor) {
         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),
-                     path_get_file_name(j->current_file->path)) < 0)
+                     "s=%s;i=%"PRIx64";b=%s;m=%"PRIx64";t=%"PRIx64";x=%"PRIx64,
+                     sid, le64toh(o->entry.seqnum),
+                     bid, le64toh(o->entry.monotonic),
+                     le64toh(o->entry.realtime),
+                     le64toh(o->entry.xor_hash)) < 0)
                 return -ENOMEM;
 
-        return 1;
+        return 0;
 }
 
 _public_ int sd_journal_seek_cursor(sd_journal *j, const char *cursor) {
-        char *w;
+        char *w, *state;
         size_t l;
-        char *state;
         unsigned long long seqnum, monotonic, realtime, xor_hash;
         bool
                 seqnum_id_set = false,
@@ -791,10 +1009,9 @@ _public_ int sd_journal_seek_cursor(sd_journal *j, const char *cursor) {
                 xor_hash_set = false;
         sd_id128_t seqnum_id, boot_id;
 
-        if (!j)
-                return -EINVAL;
-        if (!cursor)
-                return -EINVAL;
+        assert_return(j, -EINVAL);
+        assert_return(!journal_pid_changed(j), -ECHILD);
+        assert_return(!isempty(cursor), -EINVAL);
 
         FOREACH_WORD_SEPARATOR(w, l, cursor, ";", state) {
                 char *item;
@@ -811,35 +1028,35 @@ _public_ int sd_journal_seek_cursor(sd_journal *j, const char *cursor) {
 
                 case 's':
                         seqnum_id_set = true;
-                        k = sd_id128_from_string(w+2, &seqnum_id);
+                        k = sd_id128_from_string(item+2, &seqnum_id);
                         break;
 
                 case 'i':
                         seqnum_set = true;
-                        if (sscanf(w+2, "%llx", &seqnum) != 1)
+                        if (sscanf(item+2, "%llx", &seqnum) != 1)
                                 k = -EINVAL;
                         break;
 
                 case 'b':
                         boot_id_set = true;
-                        k = sd_id128_from_string(w+2, &boot_id);
+                        k = sd_id128_from_string(item+2, &boot_id);
                         break;
 
                 case 'm':
                         monotonic_set = true;
-                        if (sscanf(w+2, "%llx", &monotonic) != 1)
+                        if (sscanf(item+2, "%llx", &monotonic) != 1)
                                 k = -EINVAL;
                         break;
 
                 case 't':
                         realtime_set = true;
-                        if (sscanf(w+2, "%llx", &realtime) != 1)
+                        if (sscanf(item+2, "%llx", &realtime) != 1)
                                 k = -EINVAL;
                         break;
 
                 case 'x':
                         xor_hash_set = true;
-                        if (sscanf(w+2, "%llx", &xor_hash) != 1)
+                        if (sscanf(item+2, "%llx", &xor_hash) != 1)
                                 k = -EINVAL;
                         break;
                 }
@@ -857,7 +1074,7 @@ _public_ int sd_journal_seek_cursor(sd_journal *j, const char *cursor) {
 
         reset_location(j);
 
-        j->current_location.type = LOCATION_DISCRETE;
+        j->current_location.type = LOCATION_SEEK;
 
         if (realtime_set) {
                 j->current_location.realtime = (uint64_t) realtime;
@@ -884,12 +1101,94 @@ _public_ int sd_journal_seek_cursor(sd_journal *j, const char *cursor) {
         return 0;
 }
 
+_public_ int sd_journal_test_cursor(sd_journal *j, const char *cursor) {
+        int r;
+        char *w, *state;
+        size_t l;
+        Object *o;
+
+        assert_return(j, -EINVAL);
+        assert_return(!journal_pid_changed(j), -ECHILD);
+        assert_return(!isempty(cursor), -EINVAL);
+
+        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;
+
+        FOREACH_WORD_SEPARATOR(w, l, cursor, ";", state) {
+                _cleanup_free_ char *item = NULL;
+                sd_id128_t id;
+                unsigned long long ll;
+                int k = 0;
+
+                if (l < 2 || w[1] != '=')
+                        return -EINVAL;
+
+                item = strndup(w, l);
+                if (!item)
+                        return -ENOMEM;
+
+                switch (w[0]) {
+
+                case 's':
+                        k = sd_id128_from_string(item+2, &id);
+                        if (k < 0)
+                                return k;
+                        if (!sd_id128_equal(id, j->current_file->header->seqnum_id))
+                                return 0;
+                        break;
+
+                case 'i':
+                        if (sscanf(item+2, "%llx", &ll) != 1)
+                                return -EINVAL;
+                        if (ll != le64toh(o->entry.seqnum))
+                                return 0;
+                        break;
+
+                case 'b':
+                        k = sd_id128_from_string(item+2, &id);
+                        if (k < 0)
+                                return k;
+                        if (!sd_id128_equal(id, o->entry.boot_id))
+                                return 0;
+                        break;
+
+                case 'm':
+                        if (sscanf(item+2, "%llx", &ll) != 1)
+                                return -EINVAL;
+                        if (ll != le64toh(o->entry.monotonic))
+                                return 0;
+                        break;
+
+                case 't':
+                        if (sscanf(item+2, "%llx", &ll) != 1)
+                                return -EINVAL;
+                        if (ll != le64toh(o->entry.realtime))
+                                return 0;
+                        break;
+
+                case 'x':
+                        if (sscanf(item+2, "%llx", &ll) != 1)
+                                return -EINVAL;
+                        if (ll != le64toh(o->entry.xor_hash))
+                                return 0;
+                        break;
+                }
+        }
+
+        return 1;
+}
+
+
 _public_ int sd_journal_seek_monotonic_usec(sd_journal *j, sd_id128_t boot_id, uint64_t usec) {
-        if (!j)
-                return -EINVAL;
+        assert_return(j, -EINVAL);
+        assert_return(!journal_pid_changed(j), -ECHILD);
 
         reset_location(j);
-        j->current_location.type = LOCATION_DISCRETE;
+        j->current_location.type = LOCATION_SEEK;
         j->current_location.boot_id = boot_id;
         j->current_location.monotonic = usec;
         j->current_location.monotonic_set = true;
@@ -898,11 +1197,11 @@ _public_ int sd_journal_seek_monotonic_usec(sd_journal *j, sd_id128_t boot_id, u
 }
 
 _public_ int sd_journal_seek_realtime_usec(sd_journal *j, uint64_t usec) {
-        if (!j)
-                return -EINVAL;
+        assert_return(j, -EINVAL);
+        assert_return(!journal_pid_changed(j), -ECHILD);
 
         reset_location(j);
-        j->current_location.type = LOCATION_DISCRETE;
+        j->current_location.type = LOCATION_SEEK;
         j->current_location.realtime = usec;
         j->current_location.realtime_set = true;
 
@@ -910,8 +1209,8 @@ _public_ int sd_journal_seek_realtime_usec(sd_journal *j, uint64_t usec) {
 }
 
 _public_ int sd_journal_seek_head(sd_journal *j) {
-        if (!j)
-                return -EINVAL;
+        assert_return(j, -EINVAL);
+        assert_return(!journal_pid_changed(j), -ECHILD);
 
         reset_location(j);
         j->current_location.type = LOCATION_HEAD;
@@ -920,8 +1219,8 @@ _public_ int sd_journal_seek_head(sd_journal *j) {
 }
 
 _public_ int sd_journal_seek_tail(sd_journal *j) {
-        if (!j)
-                return -EINVAL;
+        assert_return(j, -EINVAL);
+        assert_return(!journal_pid_changed(j), -ECHILD);
 
         reset_location(j);
         j->current_location.type = LOCATION_TAIL;
@@ -929,48 +1228,80 @@ _public_ int sd_journal_seek_tail(sd_journal *j) {
         return 0;
 }
 
-static int add_file(sd_journal *j, const char *prefix, const char *dir, const char *filename) {
-        char *fn;
-        int r;
-        JournalFile *f;
+static void check_network(sd_journal *j, int fd) {
+        struct statfs sfs;
 
         assert(j);
-        assert(prefix);
-        assert(filename);
 
-        if ((j->flags & SD_JOURNAL_SYSTEM_ONLY) &&
-            !startswith(filename, "system.journal"))
-                return 0;
+        if (j->on_network)
+                return;
 
-        if (dir)
-                fn = join(prefix, "/", dir, "/", filename, NULL);
-        else
-                fn = join(prefix, "/", filename, NULL);
+        if (fstatfs(fd, &sfs) < 0)
+                return;
 
-        if (!fn)
-                return -ENOMEM;
+        j->on_network =
+                F_TYPE_EQUAL(sfs.f_type, CIFS_MAGIC_NUMBER) ||
+                F_TYPE_EQUAL(sfs.f_type, CODA_SUPER_MAGIC) ||
+                F_TYPE_EQUAL(sfs.f_type, NCP_SUPER_MAGIC) ||
+                F_TYPE_EQUAL(sfs.f_type, NFS_SUPER_MAGIC) ||
+                F_TYPE_EQUAL(sfs.f_type, SMB_SUPER_MAGIC);
+}
 
-        if (hashmap_get(j->files, fn)) {
-                free(fn);
-                return 0;
-        }
+static bool file_has_type_prefix(const char *prefix, const char *filename) {
+        const char *full, *tilded, *atted;
 
-        if (hashmap_size(j->files) >= JOURNAL_FILES_MAX) {
-                log_debug("Too many open journal files, not adding %s, ignoring.", fn);
-                free(fn);
-                return 0;
+        full = strappenda(prefix, ".journal");
+        tilded = strappenda(full, "~");
+        atted = strappenda(prefix, "@");
+
+        return streq(filename, full) ||
+               streq(filename, tilded) ||
+               startswith(filename, atted);
+}
+
+static bool file_type_wanted(int flags, const char *filename) {
+        if (!endswith(filename, ".journal") && !endswith(filename, ".journal~"))
+                return false;
+
+        /* no flags set â†’ every type is OK */
+        if (!(flags & (SD_JOURNAL_SYSTEM | SD_JOURNAL_CURRENT_USER)))
+                return true;
+
+        if (flags & SD_JOURNAL_SYSTEM && file_has_type_prefix("system", filename))
+                return true;
+
+        if (flags & SD_JOURNAL_CURRENT_USER) {
+                char prefix[5 + DECIMAL_STR_MAX(uid_t) + 1];
+
+                assert_se(snprintf(prefix, sizeof(prefix), "user-%lu", (unsigned long) getuid())
+                          < (int) sizeof(prefix));
+
+                if (file_has_type_prefix(prefix, filename))
+                        return true;
         }
 
-        r = journal_file_open(fn, O_RDONLY, 0, NULL, &f);
-        free(fn);
+        return false;
+}
 
-        if (r < 0) {
-                if (errno == ENOENT)
-                        return 0;
+static int add_any_file(sd_journal *j, const char *path) {
+        JournalFile *f = NULL;
+        int r;
 
-                return r;
+        assert(j);
+        assert(path);
+
+        if (hashmap_get(j->files, path))
+                return 0;
+
+        if (hashmap_size(j->files) >= JOURNAL_FILES_MAX) {
+                log_warning("Too many open journal files, not adding %s.", path);
+                return set_put_error(j, -ETOOMANYREFS);
         }
 
+        r = journal_file_open(path, O_RDONLY, 0, false, false, NULL, j->mmap, NULL, &f);
+        if (r < 0)
+                return r;
+
         /* journal_file_dump(f); */
 
         r = hashmap_put(j->files, f->path, f);
@@ -979,246 +1310,414 @@ static int add_file(sd_journal *j, const char *prefix, const char *dir, const ch
                 return r;
         }
 
-        log_debug("File %s got added.", f->path);
+        log_debug("File %s added.", f->path);
+
+        check_network(j, f->fd);
+
+        j->current_invalidate_counter ++;
 
         return 0;
 }
 
-static int remove_file(sd_journal *j, const char *prefix, const char *dir, const char *filename) {
-        char *fn;
-        JournalFile *f;
+static int add_file(sd_journal *j, const char *prefix, const char *filename) {
+        _cleanup_free_ char *path = NULL;
+        int r;
 
         assert(j);
         assert(prefix);
         assert(filename);
 
-        if (dir)
-                fn = join(prefix, "/", dir, "/", filename, NULL);
-        else
-                fn = join(prefix, "/", filename, NULL);
+        if (j->no_new_files ||
+            !file_type_wanted(j->flags, filename))
+                return 0;
 
-        if (!fn)
+        path = strjoin(prefix, "/", filename, NULL);
+        if (!path)
                 return -ENOMEM;
 
-        f = hashmap_get(j->files, fn);
-        free(fn);
+        r = add_any_file(j, path);
+        if (r == -ENOENT)
+                return 0;
+        return 0;
+}
+
+static int remove_file(sd_journal *j, const char *prefix, const char *filename) {
+        _cleanup_free_ char *path;
+        JournalFile *f;
+
+        assert(j);
+        assert(prefix);
+        assert(filename);
+
+        path = strjoin(prefix, "/", filename, NULL);
+        if (!path)
+                return -ENOMEM;
 
+        f = hashmap_get(j->files, path);
         if (!f)
                 return 0;
 
+        remove_file_real(j, f);
+        return 0;
+}
+
+static void remove_file_real(sd_journal *j, JournalFile *f) {
+        assert(j);
+        assert(f);
+
         hashmap_remove(j->files, f->path);
+
+        log_debug("File %s removed.", f->path);
+
+        if (j->current_file == f) {
+                j->current_file = NULL;
+                j->current_field = 0;
+        }
+
+        if (j->unique_file == f) {
+                j->unique_file = NULL;
+                j->unique_offset = 0;
+        }
+
         journal_file_close(f);
 
-        log_debug("File %s got removed.", f->path);
-        return 0;
+        j->current_invalidate_counter ++;
 }
 
-static int add_directory(sd_journal *j, const char *prefix, const char *dir) {
-        char *fn;
+static int add_directory(sd_journal *j, const char *prefix, const char *dirname) {
+        _cleanup_free_ char *path = NULL;
         int r;
-        DIR *d;
-        int wd;
+        _cleanup_closedir_ DIR *d = NULL;
         sd_id128_t id, mid;
+        Directory *m;
 
         assert(j);
         assert(prefix);
-        assert(dir);
+        assert(dirname);
+
+        log_debug("Considering %s/%s.", prefix, dirname);
 
         if ((j->flags & SD_JOURNAL_LOCAL_ONLY) &&
-            (sd_id128_from_string(dir, &id) < 0 ||
+            (sd_id128_from_string(dirname, &id) < 0 ||
              sd_id128_get_machine(&mid) < 0 ||
-             !sd_id128_equal(id, mid)))
+             !(sd_id128_equal(id, mid) || path_startswith(prefix, "/run"))))
             return 0;
 
-        fn = join(prefix, "/", dir, NULL);
-        if (!fn)
+        path = strjoin(prefix, "/", dirname, NULL);
+        if (!path)
                 return -ENOMEM;
 
-        d = opendir(fn);
-
+        d = opendir(path);
         if (!d) {
-                free(fn);
+                log_debug("Failed to open %s: %m", path);
                 if (errno == ENOENT)
                         return 0;
-
                 return -errno;
         }
 
-        wd = inotify_add_watch(j->inotify_fd, fn,
-                               IN_CREATE|IN_MOVED_TO|IN_MODIFY|IN_ATTRIB|IN_DELETE|
-                               IN_DELETE_SELF|IN_MOVE_SELF|IN_UNMOUNT|
-                               IN_DONT_FOLLOW|IN_ONLYDIR);
-        if (wd > 0) {
-                if (hashmap_put(j->inotify_wd_dirs, INT_TO_PTR(wd), fn) < 0)
-                        inotify_rm_watch(j->inotify_fd, wd);
-                else
-                        fn = NULL;
-        }
+        m = hashmap_get(j->directories_by_path, path);
+        if (!m) {
+                m = new0(Directory, 1);
+                if (!m)
+                        return -ENOMEM;
 
-        free(fn);
+                m->is_root = false;
+                m->path = path;
 
-        for (;;) {
-                struct dirent buf, *de;
+                if (hashmap_put(j->directories_by_path, m->path, m) < 0) {
+                        free(m);
+                        return -ENOMEM;
+                }
 
-                r = readdir_r(d, &buf, &de);
-                if (r != 0 || !de)
-                        break;
+                path = NULL; /* avoid freeing in cleanup */
+                j->current_invalidate_counter ++;
 
-                if (!dirent_is_file_with_suffix(de, ".journal"))
-                        continue;
+                log_debug("Directory %s added.", m->path);
 
-                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));
+        } else if (m->is_root)
+                return 0;
+
+        if (m->wd <= 0 && j->inotify_fd >= 0) {
+
+                m->wd = inotify_add_watch(j->inotify_fd, m->path,
+                                          IN_CREATE|IN_MOVED_TO|IN_MODIFY|IN_ATTRIB|IN_DELETE|
+                                          IN_DELETE_SELF|IN_MOVE_SELF|IN_UNMOUNT|IN_MOVED_FROM|
+                                          IN_ONLYDIR);
+
+                if (m->wd > 0 && hashmap_put(j->directories_by_wd, INT_TO_PTR(m->wd), m) < 0)
+                        inotify_rm_watch(j->inotify_fd, m->wd);
         }
 
-        closedir(d);
+        for (;;) {
+                struct dirent *de;
+
+                errno = 0;
+                de = readdir(d);
+                if (!de && errno != 0) {
+                        r = -errno;
+                        log_debug("Failed to read directory %s: %s",
+                                  m->path, strerror(errno));
+                        return r;
+                }
+                if (!de)
+                        break;
+
+                if (dirent_is_file_with_suffix(de, ".journal") ||
+                    dirent_is_file_with_suffix(de, ".journal~")) {
+                        r = add_file(j, m->path, de->d_name);
+                        if (r < 0) {
+                                log_debug("Failed to add file %s/%s: %s",
+                                          m->path, de->d_name, strerror(-r));
+                                r = set_put_error(j, r);
+                                if (r < 0)
+                                        return r;
+                        }
+                }
+        }
 
-        log_debug("Directory %s/%s got added.", prefix, dir);
+        check_network(j, dirfd(d));
 
         return 0;
 }
 
-static void remove_directory_wd(sd_journal *j, int wd) {
-        char *p;
+static int add_root_directory(sd_journal *j, const char *p) {
+        _cleanup_closedir_ DIR *d = NULL;
+        Directory *m;
+        int r;
 
         assert(j);
-        assert(wd > 0);
+        assert(p);
 
-        if (j->inotify_fd >= 0)
-                inotify_rm_watch(j->inotify_fd, wd);
+        if ((j->flags & SD_JOURNAL_RUNTIME_ONLY) &&
+            !path_startswith(p, "/run"))
+                return -EINVAL;
 
-        p = hashmap_remove(j->inotify_wd_dirs, INT_TO_PTR(wd));
+        if (j->prefix)
+                p = strappenda(j->prefix, p);
 
-        if (p) {
-                log_debug("Directory %s got removed.", p);
-                free(p);
-        }
-}
+        d = opendir(p);
+        if (!d)
+                return -errno;
 
-static void add_root_wd(sd_journal *j, const char *p) {
-        int wd;
-        char *k;
+        m = hashmap_get(j->directories_by_path, p);
+        if (!m) {
+                m = new0(Directory, 1);
+                if (!m)
+                        return -ENOMEM;
 
-        assert(j);
-        assert(p);
+                m->is_root = true;
+                m->path = strdup(p);
+                if (!m->path) {
+                        free(m);
+                        return -ENOMEM;
+                }
 
-        wd = inotify_add_watch(j->inotify_fd, p,
-                               IN_CREATE|IN_MOVED_TO|IN_MODIFY|IN_ATTRIB|IN_DELETE|
-                               IN_DONT_FOLLOW|IN_ONLYDIR);
-        if (wd <= 0)
-                return;
+                if (hashmap_put(j->directories_by_path, m->path, m) < 0) {
+                        free(m->path);
+                        free(m);
+                        return -ENOMEM;
+                }
 
-        k = strdup(p);
-        if (!k || hashmap_put(j->inotify_wd_roots, INT_TO_PTR(wd), k) < 0) {
-                inotify_rm_watch(j->inotify_fd, wd);
-                free(k);
-        }
-}
+                j->current_invalidate_counter ++;
 
-static void remove_root_wd(sd_journal *j, int wd) {
-        char *p;
+                log_debug("Root directory %s added.", m->path);
 
-        assert(j);
-        assert(wd > 0);
+        } else if (!m->is_root)
+                return 0;
 
-        if (j->inotify_fd >= 0)
-                inotify_rm_watch(j->inotify_fd, wd);
+        if (m->wd <= 0 && j->inotify_fd >= 0) {
 
-        p = hashmap_remove(j->inotify_wd_roots, INT_TO_PTR(wd));
+                m->wd = inotify_add_watch(j->inotify_fd, m->path,
+                                          IN_CREATE|IN_MOVED_TO|IN_MODIFY|IN_ATTRIB|IN_DELETE|
+                                          IN_ONLYDIR);
 
-        if (p) {
-                log_debug("Root %s got removed.", p);
-                free(p);
+                if (m->wd > 0 && hashmap_put(j->directories_by_wd, INT_TO_PTR(m->wd), m) < 0)
+                        inotify_rm_watch(j->inotify_fd, m->wd);
         }
-}
 
-_public_ int sd_journal_open(sd_journal **ret, int flags) {
-        sd_journal *j;
-        const char *p;
-        const char search_paths[] =
-                "/run/log/journal\0"
-                "/var/log/journal\0";
-        int r;
+        if (j->no_new_files)
+                return 0;
 
-        if (!ret)
-                return -EINVAL;
+        for (;;) {
+                struct dirent *de;
+                sd_id128_t id;
 
-        if (flags & ~(SD_JOURNAL_LOCAL_ONLY|
-                      SD_JOURNAL_RUNTIME_ONLY|
-                      SD_JOURNAL_SYSTEM_ONLY))
-                return -EINVAL;
-
-        j = new0(sd_journal, 1);
-        if (!j)
-                return -ENOMEM;
+                errno = 0;
+                de = readdir(d);
+                if (!de && errno != 0) {
+                        r = -errno;
+                        log_debug("Failed to read directory %s: %s",
+                                  m->path, strerror(errno));
+                        return r;
+                }
+                if (!de)
+                        break;
 
-        j->flags = flags;
+                if (dirent_is_file_with_suffix(de, ".journal") ||
+                    dirent_is_file_with_suffix(de, ".journal~")) {
+                        r = add_file(j, m->path, de->d_name);
+                        if (r < 0) {
+                                log_debug("Failed to add file %s/%s: %s",
+                                          m->path, de->d_name, strerror(-r));
+                                r = set_put_error(j, r);
+                                if (r < 0)
+                                        return r;
+                        }
+                } else if ((de->d_type == DT_DIR || de->d_type == DT_LNK || de->d_type == DT_UNKNOWN) &&
+                           sd_id128_from_string(de->d_name, &id) >= 0) {
 
-        j->inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC);
-        if (j->inotify_fd < 0) {
-                r = -errno;
-                goto fail;
+                        r = add_directory(j, m->path, de->d_name);
+                        if (r < 0)
+                                log_debug("Failed to add directory %s/%s: %s", m->path, de->d_name, strerror(-r));
+                }
         }
 
-        j->files = hashmap_new(string_hash_func, string_compare_func);
-        if (!j->files) {
-                r = -ENOMEM;
-                goto fail;
-        }
+        check_network(j, dirfd(d));
+
+        return 0;
+}
+
+static int remove_directory(sd_journal *j, Directory *d) {
+        assert(j);
 
-        j->inotify_wd_dirs = hashmap_new(trivial_hash_func, trivial_compare_func);
-        j->inotify_wd_roots = hashmap_new(trivial_hash_func, trivial_compare_func);
+        if (d->wd > 0) {
+                hashmap_remove(j->directories_by_wd, INT_TO_PTR(d->wd));
 
-        if (!j->inotify_wd_dirs || !j->inotify_wd_roots) {
-                r = -ENOMEM;
-                goto fail;
+                if (j->inotify_fd >= 0)
+                        inotify_rm_watch(j->inotify_fd, d->wd);
         }
 
+        hashmap_remove(j->directories_by_path, d->path);
+
+        if (d->is_root)
+                log_debug("Root directory %s removed.", d->path);
+        else
+                log_debug("Directory %s removed.", d->path);
+
+        free(d->path);
+        free(d);
+
+        return 0;
+}
+
+static int add_search_paths(sd_journal *j) {
+        int r;
+        const char search_paths[] =
+                "/run/log/journal\0"
+                "/var/log/journal\0";
+        const char *p;
+
+        assert(j);
+
         /* 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;
+                r = add_root_directory(j, p);
+                if (r < 0 && r != -ENOENT) {
+                        r = set_put_error(j, r);
+                        if (r < 0)
+                                return r;
+                }
+        }
 
-                if ((flags & SD_JOURNAL_RUNTIME_ONLY) &&
-                    !path_startswith(p, "/run"))
-                        continue;
+        return 0;
+}
 
-                d = opendir(p);
-                if (!d) {
-                        if (errno != ENOENT)
-                                log_debug("Failed to open %s: %m", p);
-                        continue;
+static int add_current_paths(sd_journal *j) {
+        Iterator i;
+        JournalFile *f;
+
+        assert(j);
+        assert(j->no_new_files);
+
+        /* Simply adds all directories for files we have open as
+         * "root" directories. We don't expect errors here, so we
+         * treat them as fatal. */
+
+        HASHMAP_FOREACH(f, j->files, i) {
+                _cleanup_free_ char *dir;
+                int r;
+
+                dir = dirname_malloc(f->path);
+                if (!dir)
+                        return -ENOMEM;
+
+                r = add_root_directory(j, dir);
+                if (r < 0) {
+                        set_put_error(j, r);
+                        return r;
                 }
+        }
 
-                add_root_wd(j, p);
+        return 0;
+}
 
-                for (;;) {
-                        struct dirent buf, *de;
-                        sd_id128_t id;
 
-                        r = readdir_r(d, &buf, &de);
-                        if (r != 0 || !de)
-                                break;
+static int allocate_inotify(sd_journal *j) {
+        assert(j);
 
-                        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));
+        if (j->inotify_fd < 0) {
+                j->inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC);
+                if (j->inotify_fd < 0)
+                        return -errno;
+        }
 
-                        } else if ((de->d_type == DT_DIR || de->d_type == DT_UNKNOWN) &&
-                                   sd_id128_from_string(de->d_name, &id) >= 0) {
+        if (!j->directories_by_wd) {
+                j->directories_by_wd = hashmap_new(trivial_hash_func, trivial_compare_func);
+                if (!j->directories_by_wd)
+                        return -ENOMEM;
+        }
 
-                                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));
-                        }
-                }
+        return 0;
+}
+
+static sd_journal *journal_new(int flags, const char *path) {
+        sd_journal *j;
+
+        j = new0(sd_journal, 1);
+        if (!j)
+                return NULL;
 
-                closedir(d);
+        j->original_pid = getpid();
+        j->inotify_fd = -1;
+        j->flags = flags;
+        j->data_threshold = DEFAULT_DATA_THRESHOLD;
+
+        if (path) {
+                j->path = strdup(path);
+                if (!j->path)
+                        goto fail;
         }
 
+        j->files = hashmap_new(string_hash_func, string_compare_func);
+        j->directories_by_path = hashmap_new(string_hash_func, string_compare_func);
+        j->mmap = mmap_cache_new();
+        if (!j->files || !j->directories_by_path || !j->mmap)
+                goto fail;
+
+        return j;
+
+fail:
+        sd_journal_close(j);
+        return NULL;
+}
+
+_public_ int sd_journal_open(sd_journal **ret, int flags) {
+        sd_journal *j;
+        int r;
+
+        assert_return(ret, -EINVAL);
+        assert_return((flags & ~(SD_JOURNAL_LOCAL_ONLY|SD_JOURNAL_RUNTIME_ONLY|SD_JOURNAL_SYSTEM|SD_JOURNAL_CURRENT_USER)) == 0, -EINVAL);
+
+        j = journal_new(flags, NULL);
+        if (!j)
+                return -ENOMEM;
+
+        r = add_search_paths(j);
+        if (r < 0)
+                goto fail;
+
         *ret = j;
         return 0;
 
@@ -1226,44 +1725,142 @@ fail:
         sd_journal_close(j);
 
         return r;
-};
+}
 
-_public_ void sd_journal_close(sd_journal *j) {
+_public_ int sd_journal_open_container(sd_journal **ret, const char *machine, int flags) {
+        _cleanup_free_ char *root = NULL, *class = NULL;
+        sd_journal *j;
+        char *p;
+        int r;
+
+        assert_return(machine, -EINVAL);
+        assert_return(ret, -EINVAL);
+        assert_return((flags & ~(SD_JOURNAL_LOCAL_ONLY|SD_JOURNAL_SYSTEM)) == 0, -EINVAL);
+        assert_return(filename_is_safe(machine), -EINVAL);
+
+        p = strappenda("/run/systemd/machines/", machine);
+        r = parse_env_file(p, NEWLINE, "ROOT", &root, "CLASS", &class, NULL);
+        if (r == -ENOENT)
+                return -EHOSTDOWN;
+        if (r < 0)
+                return r;
+        if (!root)
+                return -ENODATA;
+
+        if (!streq_ptr(class, "container"))
+                return -EIO;
+
+        j = journal_new(flags, NULL);
         if (!j)
-                return;
+                return -ENOMEM;
 
-        if (j->inotify_wd_dirs) {
-                void *k;
+        j->prefix = root;
+        root = NULL;
+
+        r = add_search_paths(j);
+        if (r < 0)
+                goto fail;
 
-                while ((k = hashmap_first_key(j->inotify_wd_dirs)))
-                        remove_directory_wd(j, PTR_TO_INT(k));
+        *ret = j;
+        return 0;
 
-                hashmap_free(j->inotify_wd_dirs);
-        }
+fail:
+        sd_journal_close(j);
+        return r;
+}
 
-        if (j->inotify_wd_roots) {
-                void *k;
+_public_ int sd_journal_open_directory(sd_journal **ret, const char *path, int flags) {
+        sd_journal *j;
+        int r;
 
-                while ((k = hashmap_first_key(j->inotify_wd_roots)))
-                        remove_root_wd(j, PTR_TO_INT(k));
+        assert_return(ret, -EINVAL);
+        assert_return(path, -EINVAL);
+        assert_return(flags == 0, -EINVAL);
+
+        j = journal_new(flags, path);
+        if (!j)
+                return -ENOMEM;
 
-                hashmap_free(j->inotify_wd_roots);
+        r = add_root_directory(j, path);
+        if (r < 0) {
+                set_put_error(j, r);
+                goto fail;
         }
 
-        if (j->files) {
-                JournalFile *f;
+        *ret = j;
+        return 0;
+
+fail:
+        sd_journal_close(j);
+
+        return r;
+}
+
+_public_ int sd_journal_open_files(sd_journal **ret, const char **paths, int flags) {
+        sd_journal *j;
+        const char **path;
+        int r;
 
-                while ((f = hashmap_steal_first(j->files)))
-                        journal_file_close(f);
+        assert_return(ret, -EINVAL);
+        assert_return(flags == 0, -EINVAL);
+
+        j = journal_new(flags, NULL);
+        if (!j)
+                return -ENOMEM;
 
-                hashmap_free(j->files);
+        STRV_FOREACH(path, paths) {
+                r = add_any_file(j, *path);
+                if (r < 0) {
+                        log_error("Failed to open %s: %s", *path, strerror(-r));
+                        goto fail;
+                }
         }
 
+        j->no_new_files = true;
+
+        *ret = j;
+        return 0;
+
+fail:
+        sd_journal_close(j);
+
+        return r;
+}
+
+_public_ void sd_journal_close(sd_journal *j) {
+        Directory *d;
+        JournalFile *f;
+
+        if (!j)
+                return;
+
         sd_journal_flush_matches(j);
 
-        if (j->inotify_fd >= 0)
-                close_nointr_nofail(j->inotify_fd);
+        while ((f = hashmap_steal_first(j->files)))
+                journal_file_close(f);
+
+        hashmap_free(j->files);
+
+        while ((d = hashmap_first(j->directories_by_path)))
+                remove_directory(j, d);
+
+        while ((d = hashmap_first(j->directories_by_wd)))
+                remove_directory(j, d);
+
+        hashmap_free(j->directories_by_path);
+        hashmap_free(j->directories_by_wd);
 
+        safe_close(j->inotify_fd);
+
+        if (j->mmap) {
+                log_debug("mmap cache statistics: %u hit, %u miss", mmap_cache_get_hit(j->mmap), mmap_cache_get_missed(j->mmap));
+                mmap_cache_unref(j->mmap);
+        }
+
+        free(j->path);
+        free(j->prefix);
+        free(j->unique_field);
+        set_free(j->errors);
         free(j);
 }
 
@@ -1272,10 +1869,9 @@ _public_ int sd_journal_get_realtime_usec(sd_journal *j, uint64_t *ret) {
         JournalFile *f;
         int r;
 
-        if (!j)
-                return -EINVAL;
-        if (!ret)
-                return -EINVAL;
+        assert_return(j, -EINVAL);
+        assert_return(!journal_pid_changed(j), -ECHILD);
+        assert_return(ret, -EINVAL);
 
         f = j->current_file;
         if (!f)
@@ -1298,10 +1894,8 @@ _public_ int sd_journal_get_monotonic_usec(sd_journal *j, uint64_t *ret, sd_id12
         int r;
         sd_id128_t id;
 
-        if (!j)
-                return -EINVAL;
-        if (!ret)
-                return -EINVAL;
+        assert_return(j, -EINVAL);
+        assert_return(!journal_pid_changed(j), -ECHILD);
 
         f = j->current_file;
         if (!f)
@@ -1325,10 +1919,40 @@ _public_ int sd_journal_get_monotonic_usec(sd_journal *j, uint64_t *ret, sd_id12
                         return -ESTALE;
         }
 
-        *ret = le64toh(o->entry.monotonic);
+        if (ret)
+                *ret = le64toh(o->entry.monotonic);
+
         return 0;
 }
 
+static bool field_is_valid(const char *field) {
+        const char *p;
+
+        assert(field);
+
+        if (isempty(field))
+                return false;
+
+        if (startswith(field, "__"))
+                return false;
+
+        for (p = field; *p; p++) {
+
+                if (*p == '_')
+                        continue;
+
+                if (*p >= 'A' && *p <= 'Z')
+                        continue;
+
+                if (*p >= '0' && *p <= '9')
+                        continue;
+
+                return false;
+        }
+
+        return true;
+}
+
 _public_ int sd_journal_get_data(sd_journal *j, const char *field, const void **data, size_t *size) {
         JournalFile *f;
         uint64_t i, n;
@@ -1336,17 +1960,12 @@ _public_ int sd_journal_get_data(sd_journal *j, const char *field, const void **
         int r;
         Object *o;
 
-        if (!j)
-                return -EINVAL;
-        if (!field)
-                return -EINVAL;
-        if (!data)
-                return -EINVAL;
-        if (!size)
-                return -EINVAL;
-
-        if (isempty(field) || strchr(field, '='))
-                return -EINVAL;
+        assert_return(j, -EINVAL);
+        assert_return(!journal_pid_changed(j), -ECHILD);
+        assert_return(field, -EINVAL);
+        assert_return(data, -EINVAL);
+        assert_return(size, -EINVAL);
+        assert_return(field_is_valid(field), -EINVAL);
 
         f = j->current_file;
         if (!f)
@@ -1388,7 +2007,8 @@ _public_ int sd_journal_get_data(sd_journal *j, const char *field, const void **
                                 uint64_t rsize;
 
                                 if (!uncompress_blob(o->data.payload, l,
-                                                     &f->compress_buffer, &f->compress_buffer_size, &rsize))
+                                                     &f->compress_buffer, &f->compress_buffer_size, &rsize,
+                                                     j->data_threshold))
                                         return -EBADMSG;
 
                                 *data = f->compress_buffer;
@@ -1423,20 +2043,48 @@ _public_ int sd_journal_get_data(sd_journal *j, const char *field, const void **
         return -ENOENT;
 }
 
+static int return_data(sd_journal *j, JournalFile *f, Object *o, const void **data, size_t *size) {
+        size_t t;
+        uint64_t l;
+
+        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;
+
+        if (o->object.flags & OBJECT_COMPRESSED) {
+#ifdef HAVE_XZ
+                uint64_t rsize;
+
+                if (!uncompress_blob(o->data.payload, l, &f->compress_buffer, &f->compress_buffer_size, &rsize, j->data_threshold))
+                        return -EBADMSG;
+
+                *data = f->compress_buffer;
+                *size = (size_t) rsize;
+#else
+                return -EPROTONOSUPPORT;
+#endif
+        } else {
+                *data = o->data.payload;
+                *size = t;
+        }
+
+        return 0;
+}
+
 _public_ int sd_journal_enumerate_data(sd_journal *j, const void **data, size_t *size) {
         JournalFile *f;
-        uint64_t p, l, n;
+        uint64_t p, n;
         le64_t le_hash;
         int r;
         Object *o;
-        size_t t;
 
-        if (!j)
-                return -EINVAL;
-        if (!data)
-                return -EINVAL;
-        if (!size)
-                return -EINVAL;
+        assert_return(j, -EINVAL);
+        assert_return(!journal_pid_changed(j), -ECHILD);
+        assert_return(data, -EINVAL);
+        assert_return(size, -EINVAL);
 
         f = j->current_file;
         if (!f)
@@ -1462,115 +2110,137 @@ _public_ int sd_journal_enumerate_data(sd_journal *j, const void **data, size_t
         if (le_hash != o->data.hash)
                 return -EBADMSG;
 
-        l = le64toh(o->object.size) - offsetof(Object, data.payload);
-        t = (size_t) l;
+        r = return_data(j, f, o, data, size);
+        if (r < 0)
+                return r;
 
-        /* We can't read objects larger than 4G on a 32bit machine */
-        if ((uint64_t) t != l)
-                return -E2BIG;
+        j->current_field ++;
 
-        if (o->object.flags & OBJECT_COMPRESSED) {
-#ifdef HAVE_XZ
-                uint64_t rsize;
+        return 1;
+}
 
-                if (!uncompress_blob(o->data.payload, l, &f->compress_buffer, &f->compress_buffer_size, &rsize))
-                        return -EBADMSG;
+_public_ void sd_journal_restart_data(sd_journal *j) {
+        if (!j)
+                return;
 
-                *data = f->compress_buffer;
-                *size = (size_t) rsize;
-#else
-                return -EPROTONOSUPPORT;
-#endif
-        } else {
-                *data = o->data.payload;
-                *size = t;
-        }
+        j->current_field = 0;
+}
 
-        j->current_field ++;
+_public_ int sd_journal_get_fd(sd_journal *j) {
+        int r;
 
-        return 1;
-}
+        assert_return(j, -EINVAL);
+        assert_return(!journal_pid_changed(j), -ECHILD);
 
-_public_ void sd_journal_restart_data(sd_journal *j) {
-        if (!j)
-                return;
+        if (j->inotify_fd >= 0)
+                return j->inotify_fd;
 
-        j->current_field = 0;
-}
+        r = allocate_inotify(j);
+        if (r < 0)
+                return r;
 
-_public_ int sd_journal_get_fd(sd_journal *j) {
-        if (!j)
-                return -EINVAL;
+        /* Iterate through all dirs again, to add them to the
+         * inotify */
+        if (j->no_new_files)
+                r = add_current_paths(j);
+        else if (j->path)
+                r = add_root_directory(j, j->path);
+        else
+                r = add_search_paths(j);
+        if (r < 0)
+                return r;
 
         return j->inotify_fd;
 }
 
-static void process_inotify_event(sd_journal *j, struct inotify_event *e) {
-        char *p;
-        int r;
+_public_ int sd_journal_get_events(sd_journal *j) {
+        int fd;
 
-        assert(j);
-        assert(e);
+        assert_return(j, -EINVAL);
+        assert_return(!journal_pid_changed(j), -ECHILD);
 
-        /* Is this a subdirectory we watch? */
-        p = hashmap_get(j->inotify_wd_dirs, INT_TO_PTR(e->wd));
-        if (p) {
+        fd = sd_journal_get_fd(j);
+        if (fd < 0)
+                return fd;
 
-                if (!(e->mask & IN_ISDIR) && e->len > 0 && endswith(e->name, ".journal")) {
+        return POLLIN;
+}
 
-                        /* Event for a journal file */
+_public_ int sd_journal_get_timeout(sd_journal *j, uint64_t *timeout_usec) {
+        int fd;
 
-                        if (e->mask & (IN_CREATE|IN_MOVED_TO|IN_MODIFY|IN_ATTRIB)) {
-                                r = add_file(j, p, NULL, e->name);
-                                if (r < 0)
-                                        log_debug("Failed to add file %s/%s: %s", p, e->name, strerror(-r));
-                        } else if (e->mask & (IN_DELETE|IN_UNMOUNT)) {
+        assert_return(j, -EINVAL);
+        assert_return(!journal_pid_changed(j), -ECHILD);
+        assert_return(timeout_usec, -EINVAL);
 
-                                r = remove_file(j, p, NULL, e->name);
-                                if (r < 0)
-                                        log_debug("Failed to remove file %s/%s: %s", p, e->name, strerror(-r));
-                        }
+        fd = sd_journal_get_fd(j);
+        if (fd < 0)
+                return fd;
 
-                } else if (e->len == 0) {
+        if (!j->on_network) {
+                *timeout_usec = (uint64_t) -1;
+                return 0;
+        }
 
-                        /* Event for the directory itself */
+        /* If we are on the network we need to regularly check for
+         * changes manually */
 
-                        if (e->mask & (IN_DELETE_SELF|IN_MOVE_SELF|IN_UNMOUNT))
-                                remove_directory_wd(j, e->wd);
-                }
+        *timeout_usec = j->last_process_usec + JOURNAL_FILES_RECHECK_USEC;
+        return 1;
+}
 
-                return;
-        }
+static void process_inotify_event(sd_journal *j, struct inotify_event *e) {
+        Directory *d;
+        int r;
 
-        /* Must be the root directory then? */
-        p = hashmap_get(j->inotify_wd_roots, INT_TO_PTR(e->wd));
-        if (p) {
+        assert(j);
+        assert(e);
+
+        /* Is this a subdirectory we watch? */
+        d = hashmap_get(j->directories_by_wd, INT_TO_PTR(e->wd));
+        if (d) {
                 sd_id128_t id;
 
-                if (!(e->mask & IN_ISDIR) && e->len > 0 && endswith(e->name, ".journal")) {
+                if (!(e->mask & IN_ISDIR) && e->len > 0 &&
+                    (endswith(e->name, ".journal") ||
+                     endswith(e->name, ".journal~"))) {
 
                         /* Event for a journal file */
 
                         if (e->mask & (IN_CREATE|IN_MOVED_TO|IN_MODIFY|IN_ATTRIB)) {
-                                r = add_file(j, p, NULL, e->name);
+                                r = add_file(j, d->path, e->name);
+                                if (r < 0) {
+                                        log_debug("Failed to add file %s/%s: %s",
+                                                  d->path, e->name, strerror(-r));
+                                        set_put_error(j, r);
+                                }
+
+                        } else if (e->mask & (IN_DELETE|IN_MOVED_FROM|IN_UNMOUNT)) {
+
+                                r = remove_file(j, d->path, e->name);
                                 if (r < 0)
-                                        log_debug("Failed to add file %s/%s: %s", p, e->name, strerror(-r));
-                        } else if (e->mask & (IN_DELETE|IN_UNMOUNT)) {
+                                        log_debug("Failed to remove file %s/%s: %s", d->path, e->name, strerror(-r));
+                        }
+
+                } else if (!d->is_root && e->len == 0) {
+
+                        /* Event for a subdirectory */
 
-                                r = remove_file(j, p, NULL, e->name);
+                        if (e->mask & (IN_DELETE_SELF|IN_MOVE_SELF|IN_UNMOUNT)) {
+                                r = remove_directory(j, d);
                                 if (r < 0)
-                                        log_debug("Failed to remove file %s/%s: %s", p, e->name, strerror(-r));
+                                        log_debug("Failed to remove directory %s: %s", d->path, strerror(-r));
                         }
 
-                } else if ((e->mask & IN_ISDIR) && e->len > 0 && sd_id128_from_string(e->name, &id) >= 0) {
 
-                        /* Event for subdirectory */
+                } else if (d->is_root && (e->mask & IN_ISDIR) && e->len > 0 && sd_id128_from_string(e->name, &id) >= 0) {
 
-                        if (e->mask & (IN_CREATE|IN_MOVED_TO|IN_MODIFY|IN_ATTRIB)) {
+                        /* Event for root directory */
 
-                                r = add_directory(j, p, e->name);
+                        if (e->mask & (IN_CREATE|IN_MOVED_TO|IN_MODIFY|IN_ATTRIB)) {
+                                r = add_directory(j, d->path, e->name);
                                 if (r < 0)
-                                        log_debug("Failed to add directory %s/%s: %s", p, e->name, strerror(-r));
+                                        log_debug("Failed to add directory %s/%s: %s", d->path, e->name, strerror(-r));
                         }
                 }
 
@@ -1583,11 +2253,25 @@ static void process_inotify_event(sd_journal *j, struct inotify_event *e) {
         log_warning("Unknown inotify event.");
 }
 
+static int determine_change(sd_journal *j) {
+        bool b;
+
+        assert(j);
+
+        b = j->current_invalidate_counter != j->last_invalidate_counter;
+        j->last_invalidate_counter = j->current_invalidate_counter;
+
+        return b ? SD_JOURNAL_INVALIDATE : SD_JOURNAL_APPEND;
+}
+
 _public_ int sd_journal_process(sd_journal *j) {
-        uint8_t buffer[sizeof(struct inotify_event) + FILENAME_MAX];
+        uint8_t buffer[sizeof(struct inotify_event) + FILENAME_MAX] _alignas_(struct inotify_event);
+        bool got_something = false;
 
-        if (!j)
-                return -EINVAL;
+        assert_return(j, -EINVAL);
+        assert_return(!journal_pid_changed(j), -ECHILD);
+
+        j->last_process_usec = now(CLOCK_MONOTONIC);
 
         for (;;) {
                 struct inotify_event *e;
@@ -1595,12 +2279,14 @@ _public_ int sd_journal_process(sd_journal *j) {
 
                 l = read(j->inotify_fd, buffer, sizeof(buffer));
                 if (l < 0) {
-                        if (errno == EINTR || errno == EAGAIN)
-                                return 0;
+                        if (errno == EAGAIN || errno == EINTR)
+                                return got_something ? determine_change(j) : SD_JOURNAL_NOP;
 
                         return -errno;
                 }
 
+                got_something = true;
+
                 e = (struct inotify_event*) buffer;
                 while (l > 0) {
                         size_t step;
@@ -1616,27 +2302,397 @@ _public_ int sd_journal_process(sd_journal *j) {
         }
 }
 
-/* _public_ int sd_journal_query_unique(sd_journal *j, const char *field) { */
-/*         if (!j) */
-/*                 return -EINVAL; */
-/*         if (!field) */
-/*                 return -EINVAL; */
+_public_ int sd_journal_wait(sd_journal *j, uint64_t timeout_usec) {
+        int r;
+        uint64_t t;
+
+        assert_return(j, -EINVAL);
+        assert_return(!journal_pid_changed(j), -ECHILD);
+
+        if (j->inotify_fd < 0) {
+
+                /* This is the first invocation, hence create the
+                 * inotify watch */
+                r = sd_journal_get_fd(j);
+                if (r < 0)
+                        return r;
+
+                /* The journal might have changed since the context
+                 * object was created and we weren't watching before,
+                 * hence don't wait for anything, and return
+                 * immediately. */
+                return determine_change(j);
+        }
+
+        r = sd_journal_get_timeout(j, &t);
+        if (r < 0)
+                return r;
+
+        if (t != (uint64_t) -1) {
+                usec_t n;
+
+                n = now(CLOCK_MONOTONIC);
+                t = t > n ? t - n : 0;
+
+                if (timeout_usec == (uint64_t) -1 || timeout_usec > t)
+                        timeout_usec = t;
+        }
+
+        do {
+                r = fd_wait_for_event(j->inotify_fd, POLLIN, timeout_usec);
+        } while (r == -EINTR);
+
+        if (r < 0)
+                return r;
+
+        return sd_journal_process(j);
+}
+
+_public_ int sd_journal_get_cutoff_realtime_usec(sd_journal *j, uint64_t *from, uint64_t *to) {
+        Iterator i;
+        JournalFile *f;
+        bool first = true;
+        uint64_t fmin = 0, tmax = 0;
+        int r;
 
-/*         return -ENOTSUP; */
-/* } */
+        assert_return(j, -EINVAL);
+        assert_return(!journal_pid_changed(j), -ECHILD);
+        assert_return(from || to, -EINVAL);
+        assert_return(from != to, -EINVAL);
 
-/* _public_ int sd_journal_enumerate_unique(sd_journal *j, const void **data, size_t *l) { */
-/*         if (!j) */
-/*                 return -EINVAL; */
-/*         if (!data) */
-/*                 return -EINVAL; */
-/*         if (!l) */
-/*                 return -EINVAL; */
+        HASHMAP_FOREACH(f, j->files, i) {
+                usec_t fr, t;
 
-/*         return -ENOTSUP; */
-/* } */
+                r = journal_file_get_cutoff_realtime_usec(f, &fr, &t);
+                if (r == -ENOENT)
+                        continue;
+                if (r < 0)
+                        return r;
+                if (r == 0)
+                        continue;
+
+                if (first) {
+                        fmin = fr;
+                        tmax = t;
+                        first = false;
+                } else {
+                        fmin = MIN(fr, fmin);
+                        tmax = MAX(t, tmax);
+                }
+        }
+
+        if (from)
+                *from = fmin;
+        if (to)
+                *to = tmax;
+
+        return first ? 0 : 1;
+}
+
+_public_ int sd_journal_get_cutoff_monotonic_usec(sd_journal *j, sd_id128_t boot_id, uint64_t *from, uint64_t *to) {
+        Iterator i;
+        JournalFile *f;
+        bool first = true;
+        int r;
+
+        assert_return(j, -EINVAL);
+        assert_return(!journal_pid_changed(j), -ECHILD);
+        assert_return(from || to, -EINVAL);
+        assert_return(from != to, -EINVAL);
+
+        HASHMAP_FOREACH(f, j->files, i) {
+                usec_t fr, t;
+
+                r = journal_file_get_cutoff_monotonic_usec(f, boot_id, &fr, &t);
+                if (r == -ENOENT)
+                        continue;
+                if (r < 0)
+                        return r;
+                if (r == 0)
+                        continue;
+
+                if (first) {
+                        if (from)
+                                *from = fr;
+                        if (to)
+                                *to = t;
+                        first = false;
+                } else {
+                        if (from)
+                                *from = MIN(fr, *from);
+                        if (to)
+                                *to = MAX(t, *to);
+                }
+        }
+
+        return first ? 0 : 1;
+}
+
+void journal_print_header(sd_journal *j) {
+        Iterator i;
+        JournalFile *f;
+        bool newline = false;
+
+        assert(j);
 
-/* _public_ void sd_journal_restart_unique(sd_journal *j) { */
-/*         if (!j) */
-/*                 return; */
-/* } */
+        HASHMAP_FOREACH(f, j->files, i) {
+                if (newline)
+                        putchar('\n');
+                else
+                        newline = true;
+
+                journal_file_print_header(f);
+        }
+}
+
+_public_ int sd_journal_get_usage(sd_journal *j, uint64_t *bytes) {
+        Iterator i;
+        JournalFile *f;
+        uint64_t sum = 0;
+
+        assert_return(j, -EINVAL);
+        assert_return(!journal_pid_changed(j), -ECHILD);
+        assert_return(bytes, -EINVAL);
+
+        HASHMAP_FOREACH(f, j->files, i) {
+                struct stat st;
+
+                if (fstat(f->fd, &st) < 0)
+                        return -errno;
+
+                sum += (uint64_t) st.st_blocks * 512ULL;
+        }
+
+        *bytes = sum;
+        return 0;
+}
+
+_public_ int sd_journal_query_unique(sd_journal *j, const char *field) {
+        char *f;
+
+        assert_return(j, -EINVAL);
+        assert_return(!journal_pid_changed(j), -ECHILD);
+        assert_return(!isempty(field), -EINVAL);
+        assert_return(field_is_valid(field), -EINVAL);
+
+        f = strdup(field);
+        if (!f)
+                return -ENOMEM;
+
+        free(j->unique_field);
+        j->unique_field = f;
+        j->unique_file = NULL;
+        j->unique_offset = 0;
+
+        return 0;
+}
+
+_public_ int sd_journal_enumerate_unique(sd_journal *j, const void **data, size_t *l) {
+        size_t k;
+
+        assert_return(j, -EINVAL);
+        assert_return(!journal_pid_changed(j), -ECHILD);
+        assert_return(data, -EINVAL);
+        assert_return(l, -EINVAL);
+        assert_return(j->unique_field, -EINVAL);
+
+        k = strlen(j->unique_field);
+
+        if (!j->unique_file) {
+                j->unique_file = hashmap_first(j->files);
+                if (!j->unique_file)
+                        return 0;
+                j->unique_offset = 0;
+        }
+
+        for (;;) {
+                JournalFile *of;
+                Iterator i;
+                Object *o;
+                const void *odata;
+                size_t ol;
+                bool found;
+                int r;
+
+                /* Proceed to next data object in the field's linked list */
+                if (j->unique_offset == 0) {
+                        r = journal_file_find_field_object(j->unique_file, j->unique_field, k, &o, NULL);
+                        if (r < 0)
+                                return r;
+
+                        j->unique_offset = r > 0 ? le64toh(o->field.head_data_offset) : 0;
+                } else {
+                        r = journal_file_move_to_object(j->unique_file, OBJECT_DATA, j->unique_offset, &o);
+                        if (r < 0)
+                                return r;
+
+                        j->unique_offset = le64toh(o->data.next_field_offset);
+                }
+
+                /* We reached the end of the list? Then start again, with the next file */
+                if (j->unique_offset == 0) {
+                        JournalFile *n;
+
+                        n = hashmap_next(j->files, j->unique_file->path);
+                        if (!n)
+                                return 0;
+
+                        j->unique_file = n;
+                        continue;
+                }
+
+                /* We do not use the type context here, but 0 instead,
+                 * so that we can look at this data object at the same
+                 * time as one on another file */
+                r = journal_file_move_to_object(j->unique_file, 0, j->unique_offset, &o);
+                if (r < 0)
+                        return r;
+
+                /* Let's do the type check by hand, since we used 0 context above. */
+                if (o->object.type != OBJECT_DATA) {
+                        log_error("%s:offset " OFSfmt ": object has type %d, expected %d",
+                                  j->unique_file->path, j->unique_offset,
+                                  o->object.type, OBJECT_DATA);
+                        return -EBADMSG;
+                }
+
+                r = journal_file_object_keep(j->unique_file, o, j->unique_offset);
+                if (r < 0)
+                        return r;
+
+                r = return_data(j, j->unique_file, o, &odata, &ol);
+                if (r < 0)
+                        return r;
+
+                /* OK, now let's see if we already returned this data
+                 * object by checking if it exists in the earlier
+                 * traversed files. */
+                found = false;
+                HASHMAP_FOREACH(of, j->files, i) {
+                        Object *oo;
+                        uint64_t op;
+
+                        if (of == j->unique_file)
+                                break;
+
+                        /* Skip this file it didn't have any fields
+                         * indexed */
+                        if (JOURNAL_HEADER_CONTAINS(of->header, n_fields) &&
+                            le64toh(of->header->n_fields) <= 0)
+                                continue;
+
+                        r = journal_file_find_data_object_with_hash(of, odata, ol, le64toh(o->data.hash), &oo, &op);
+                        if (r < 0)
+                                return r;
+
+                        if (r > 0)
+                                found = true;
+                }
+
+                if (found)
+                        continue;
+
+                r = journal_file_object_release(j->unique_file, o, j->unique_offset);
+                if (r < 0)
+                        return r;
+
+                r = return_data(j, j->unique_file, o, data, l);
+                if (r < 0)
+                        return r;
+
+                return 1;
+        }
+}
+
+_public_ void sd_journal_restart_unique(sd_journal *j) {
+        if (!j)
+                return;
+
+        j->unique_file = NULL;
+        j->unique_offset = 0;
+}
+
+_public_ int sd_journal_reliable_fd(sd_journal *j) {
+        assert_return(j, -EINVAL);
+        assert_return(!journal_pid_changed(j), -ECHILD);
+
+        return !j->on_network;
+}
+
+static char *lookup_field(const char *field, void *userdata) {
+        sd_journal *j = userdata;
+        const void *data;
+        size_t size, d;
+        int r;
+
+        assert(field);
+        assert(j);
+
+        r = sd_journal_get_data(j, field, &data, &size);
+        if (r < 0 ||
+            size > REPLACE_VAR_MAX)
+                return strdup(field);
+
+        d = strlen(field) + 1;
+
+        return strndup((const char*) data + d, size - d);
+}
+
+_public_ int sd_journal_get_catalog(sd_journal *j, char **ret) {
+        const void *data;
+        size_t size;
+        sd_id128_t id;
+        _cleanup_free_ char *text = NULL, *cid = NULL;
+        char *t;
+        int r;
+
+        assert_return(j, -EINVAL);
+        assert_return(!journal_pid_changed(j), -ECHILD);
+        assert_return(ret, -EINVAL);
+
+        r = sd_journal_get_data(j, "MESSAGE_ID", &data, &size);
+        if (r < 0)
+                return r;
+
+        cid = strndup((const char*) data + 11, size - 11);
+        if (!cid)
+                return -ENOMEM;
+
+        r = sd_id128_from_string(cid, &id);
+        if (r < 0)
+                return r;
+
+        r = catalog_get(CATALOG_DATABASE, id, &text);
+        if (r < 0)
+                return r;
+
+        t = replace_var(text, lookup_field, j);
+        if (!t)
+                return -ENOMEM;
+
+        *ret = t;
+        return 0;
+}
+
+_public_ int sd_journal_get_catalog_for_message_id(sd_id128_t id, char **ret) {
+        assert_return(ret, -EINVAL);
+
+        return catalog_get(CATALOG_DATABASE, id, ret);
+}
+
+_public_ int sd_journal_set_data_threshold(sd_journal *j, size_t sz) {
+        assert_return(j, -EINVAL);
+        assert_return(!journal_pid_changed(j), -ECHILD);
+
+        j->data_threshold = sz;
+        return 0;
+}
+
+_public_ int sd_journal_get_data_threshold(sd_journal *j, size_t *sz) {
+        assert_return(j, -EINVAL);
+        assert_return(!journal_pid_changed(j), -ECHILD);
+        assert_return(sz, -EINVAL);
+
+        *sz = j->data_threshold;
+        return 0;
+}