journal: implement parallel file traversal

6 files changed, 340 insertions, 167 deletions

diff --git a/src/journal/journal-file.c b/src/journal/journal-file.c
index 934c043aff..6c8d712dbd 100644
--- a/src/journal/journal-file.c
+++ b/src/journal/journal-file.c
@@ -751,8 +751,8 @@ static int journal_file_append_entry_internal(
         o->object.type = htole64(OBJECT_ENTRY);
         o->entry.seqnum = htole64(journal_file_seqnum(f));
         memcpy(o->entry.items, items, n_items * sizeof(EntryItem));
-        o->entry.realtime = ts ? htole64(ts->realtime) : 0;
-        o->entry.monotonic = ts ? htole64(ts->monotonic) : 0;
+        o->entry.realtime = htole64(ts ? ts->realtime : now(CLOCK_REALTIME));
+        o->entry.monotonic = htole64(ts ? ts->monotonic : now(CLOCK_MONOTONIC));
         o->entry.xor_hash = htole64(xor_hash);
         o->entry.boot_id = f->header->boot_id;
 
@@ -1072,7 +1072,10 @@ void journal_file_dump(JournalFile *f) {
                         break;
 
                 case OBJECT_ENTRY:
-                        printf("Type: OBJECT_ENTRY %llu\n", (unsigned long long) le64toh(o->entry.seqnum));
+                        printf("Type: OBJECT_ENTRY %llu %llu %llu\n",
+                               (unsigned long long) le64toh(o->entry.seqnum),
+                               (unsigned long long) le64toh(o->entry.monotonic),
+                               (unsigned long long) le64toh(o->entry.realtime));
                         break;
 
                 case OBJECT_HASH_TABLE:
diff --git a/src/journal/journal-file.h b/src/journal/journal-file.h
index 4665f4da38..c51504d11b 100644
--- a/src/journal/journal-file.h
+++ b/src/journal/journal-file.h
@@ -52,19 +52,9 @@ typedef struct JournalFile {
         uint64_t window_size;
 
         uint64_t current_offset;
+        uint64_t current_field;
 } JournalFile;
 
-typedef struct JournalCursor {
-        uint8_t version;
-        uint8_t reserved[7];
-        uint64_t seqnum;
-        sd_id128_t seqnum_id;
-        sd_id128_t boot_id;
-        uint64_t monotonic;
-        uint64_t realtime;
-        uint64_t xor_hash;
-} JournalCursor;
-
 int journal_file_open(const char *fname, int flags, mode_t mode, JournalFile *template, JournalFile **ret);
 
 void journal_file_close(JournalFile *j);
diff --git a/src/journal/journalctl.c b/src/journal/journalctl.c
index 4645f9ebb0..bb1f18a21f 100644
--- a/src/journal/journalctl.c
+++ b/src/journal/journalctl.c
@@ -22,104 +22,51 @@
 #include <fcntl.h>
 #include <errno.h>
 #include <stddef.h>
+#include <string.h>
+#include <stdio.h>
+#include <unistd.h>
+#include <stdlib.h>
 
-#include "journal-file.h"
-
-static int system_journal_open(JournalFile **f) {
-        int r;
-        char *fn;
-        sd_id128_t machine;
-        char ids[33];
-
-        assert(f);
-
-        r = sd_id128_get_machine(&machine);
-        if (r < 0)
-                return r;
-
-        fn = join("/var/log/journal/", sd_id128_to_string(machine, ids), "/system.journal", NULL);
-        if (!fn)
-                return -ENOMEM;
-
-        r = journal_file_open(fn, O_RDONLY, 0640, NULL, f);
-        free(fn);
-
-        if (r >= 0)
-                return r;
-
-        if (r < 0 && r != -ENOENT) {
-                log_error("Failed to open system journal: %s", strerror(-r));
-                return r;
-        }
-
-        fn = join("/run/log/journal/", ids, "/system.journal", NULL);
-        if (!fn)
-                return -ENOMEM;
-
-        r = journal_file_open(fn, O_RDONLY, 0640, NULL, f);
-        free(fn);
-
-        if (r < 0) {
-                log_error("Failed to open system journal: %s", strerror(-r));
-                return r;
-        }
-
-        return r;
-}
+#include "sd-journal.h"
+#include "log.h"
 
 int main(int argc, char *argv[]) {
         int r;
-        JournalFile *f;
-        Object *o = NULL;
+        sd_journal *j = NULL;
+
+        log_set_max_level(LOG_DEBUG);
+        log_set_target(LOG_TARGET_CONSOLE);
 
         log_parse_environment();
         log_open();
 
-        r = system_journal_open(&f);
+        r = sd_journal_open(&j);
         if (r < 0) {
                 log_error("Failed to open journal: %s", strerror(-r));
-                return EXIT_FAILURE;
+                goto finish;
         }
 
-        for (;;) {
-                uint64_t offset;
-                uint64_t n, i;
+        SD_JOURNAL_FOREACH(j) {
+                const void *data;
+                size_t length;
+                char *cursor;
 
-                r = journal_file_next_entry(f, o, &o, &offset);
+                r = sd_journal_get_cursor(j, &cursor);
                 if (r < 0) {
-                        log_error("Failed to read journal: %s", strerror(-r));
+                        log_error("Failed to get cursor: %s", strerror(-r));
                         goto finish;
                 }
 
-                if (r == 0)
-                        break;
+                printf("entry: %s\n", cursor);
+                free(cursor);
 
-                printf("entry: %llu\n", (unsigned long long) le64toh(o->entry.seqnum));
-
-                n = journal_file_entry_n_items(o);
-                for (i = 0; i < n; i++) {
-                        uint64_t p, l;
-
-                        p = le64toh(o->entry.items[i].object_offset);
-                        r = journal_file_move_to_object(f, p, OBJECT_DATA, &o);
-                        if (r < 0) {
-                                log_error("Failed to move to data: %s", strerror(-r));
-                                goto finish;
-                        }
-
-                        l = o->object.size - offsetof(Object, data.payload);
-                        printf("\t[%.*s]\n", (int) l, o->data.payload);
-
-                        r = journal_file_move_to_object(f, offset, OBJECT_ENTRY, &o);
-                        if (r < 0) {
-                                log_error("Failed to move back to entry: %s", strerror(-r));
-                                goto finish;
-                        }
-                }
+                SD_JOURNAL_FOREACH_FIELD(j, data, length)
+                        printf("\t%.*s\n", (int) length, (const char*) data);
         }
 
 finish:
-        journal_file_close(f);
+        if (j)
+                sd_journal_close(j);
 
-        return 0;
+        return r < 0 ? EXIT_FAILURE : EXIT_SUCCESS;
 }
diff --git a/src/journal/journald.c b/src/journal/journald.c
index 7a2b50b017..1143d81abc 100644
--- a/src/journal/journald.c
+++ b/src/journal/journald.c
@@ -95,6 +95,8 @@ static JournalFile* find_journal(Server *s, uid_t uid) {
         char *p;
         int r;
         JournalFile *f;
+        char ids[33];
+        sd_id128_t machine;
 
         assert(s);
 
@@ -105,11 +107,15 @@ static JournalFile* find_journal(Server *s, uid_t uid) {
         if (uid <= 0)
                 return s->system_journal;
 
+        r = sd_id128_get_machine(&machine);
+        if (r < 0)
+                return s->system_journal;
+
         f = hashmap_get(s->user_journals, UINT32_TO_PTR(uid));
         if (f)
                 return f;
 
-        if (asprintf(&p, "/var/log/journal/user-%lu.journal", (unsigned long) uid) < 0)
+        if (asprintf(&p, "/var/log/journal/%s/user-%lu.journal", sd_id128_to_string(machine, ids), (unsigned long) uid) < 0)
                 return s->system_journal;
 
         r = journal_file_open(p, O_RDWR|O_CREAT, 0640, NULL, &f);
@@ -401,16 +407,11 @@ static int system_journal_open(Server *s) {
 
         /* /var didn't work, so try /run, but this time we
          * create the prefix too */
-        fn = strappend("/run/log/journal/", ids);
-        if (!fn)
-                return -ENOMEM;
-        (void) mkdir_p(fn, 0755);
-        free(fn);
-
-        /* Then create the runtime journal file */
         fn = join("/run/log/journal/", ids, "/system.journal", NULL);
         if (!fn)
                 return -ENOMEM;
+
+        (void) mkdir_parents(fn, 0755);
         r = journal_file_open(fn, O_RDWR|O_CREAT, 0640, NULL, &s->runtime_journal);
         free(fn);
 
diff --git a/src/journal/sd-journal.c b/src/journal/sd-journal.c
index 9f5f1e858b..6f47520217 100644
--- a/src/journal/sd-journal.c
+++ b/src/journal/sd-journal.c
@@ -21,6 +21,7 @@
 
 #include <errno.h>
 #include <fcntl.h>
+#include <stddef.h>
 
 #include "sd-journal.h"
 #include "journal-def.h"
@@ -151,7 +152,8 @@ int sd_journal_next(sd_journal *j) {
 
         if (new_current) {
                 j->current_file = new_current;
-                f->current_offset = new_offset;
+                j->current_file->current_offset = new_offset;
+                j->current_file->current_field = 0;
                 return 1;
         }
 
@@ -193,59 +195,134 @@ int sd_journal_previous(sd_journal *j) {
 
         if (new_current) {
                 j->current_file = new_current;
-                f->current_offset = new_offset;
+                j->current_file->current_offset = new_offset;
+                j->current_file->current_field = 0;
                 return 1;
         }
 
         return 0;
 }
 
-int sd_journal_get_cursor(sd_journal *j, void **cursor, size_t *size) {
-        JournalCursor *c;
+int sd_journal_get_cursor(sd_journal *j, char **cursor) {
         Object *o;
         int r;
+        char bid[33], sid[33];
 
         assert(j);
         assert(cursor);
-        assert(size);
 
-        if (!j->current_file || !j->current_file->current_offset <= 0)
-                return 0;
+        if (!j->current_file || j->current_file->current_offset <= 0)
+                return -EADDRNOTAVAIL;
 
         r = journal_file_move_to_object(j->current_file, j->current_file->current_offset, OBJECT_ENTRY, &o);
         if (r < 0)
                 return r;
 
-        c = new0(JournalCursor, 1);
-        if (!c)
-                return -ENOMEM;
+        sd_id128_to_string(j->current_file->header->seqnum_id, sid);
+        sd_id128_to_string(o->entry.boot_id, bid);
 
-        c->version = 1;
-        c->seqnum = o->entry.seqnum;
-        c->seqnum_id = j->current_file->header->seqnum_id;
-        c->boot_id = o->entry.boot_id;
-        c->monotonic = o->entry.monotonic;
-        c->realtime = o->entry.realtime;
-        c->xor_hash = o->entry.xor_hash;
-
-        *cursor = c;
-        *size = sizeof(JournalCursor);
+        if (asprintf(cursor,
+                     "s=%s;i=%llx;b=%s;m=%llx;t=%llx;x=%llx;p=%s",
+                     sid, (unsigned long long) le64toh(o->entry.seqnum),
+                     bid, (unsigned long long) le64toh(o->entry.monotonic),
+                     (unsigned long long) le64toh(o->entry.realtime),
+                     (unsigned long long) le64toh(o->entry.xor_hash),
+                     file_name_from_path(j->current_file->path)) < 0)
+                return -ENOMEM;
 
         return 1;
 }
 
-int sd_journal_set_cursor(sd_journal *j, const void *cursor, size_t size) {
+int sd_journal_set_cursor(sd_journal *j, const char *cursor) {
         return -EINVAL;
 }
 
+static int add_file(sd_journal *j, const char *prefix, const char *dir, const char *filename) {
+        char *fn;
+        int r;
+        JournalFile *f;
+
+        assert(j);
+        assert(prefix);
+        assert(filename);
+
+        if (dir)
+                fn = join(prefix, "/", dir, "/", filename, NULL);
+        else
+                fn = join(prefix, "/", filename, NULL);
+
+        if (!fn)
+                return -ENOMEM;
+
+        r = journal_file_open(fn, O_RDONLY, 0, NULL, &f);
+        free(fn);
+
+        if (r < 0) {
+                if (errno == ENOENT)
+                        return 0;
+
+                return r;
+        }
+
+        r = hashmap_put(j->files, f->path, f);
+        if (r < 0) {
+                journal_file_close(f);
+                return r;
+        }
+
+        return 0;
+}
+
+static int add_directory(sd_journal *j, const char *prefix, const char *dir) {
+        char *fn;
+        int r;
+        DIR *d;
+
+        assert(j);
+        assert(prefix);
+        assert(dir);
+
+        fn = join(prefix, "/", dir, NULL);
+        if (!fn)
+                return -ENOMEM;
+
+        d = opendir(fn);
+        free(fn);
+
+        if (!d) {
+                if (errno == ENOENT)
+                        return 0;
+
+                return -errno;
+        }
+
+        for (;;) {
+                struct dirent buf, *de;
+
+                r = readdir_r(d, &buf, &de);
+                if (r != 0 || !de)
+                        break;
+
+                if (!dirent_is_file_with_suffix(de, ".journal"))
+                        continue;
+
+                r = add_file(j, prefix, dir, de->d_name);
+                if (r < 0)
+                        log_debug("Failed to add file %s/%s/%s: %s", prefix, dir, de->d_name, strerror(-r));
+        }
+
+        closedir(d);
+
+        return 0;
+}
+
 int sd_journal_open(sd_journal **ret) {
         sd_journal *j;
-        char *fn;
         const char *p;
-        int r = 0;
         const char search_paths[] =
                 "/run/log/journal\0"
                 "/var/log/journal\0";
+        int r;
 
         assert(ret);
 
@@ -254,64 +331,47 @@ int sd_journal_open(sd_journal **ret) {
                 return -ENOMEM;
 
         j->files = hashmap_new(string_hash_func, string_compare_func);
-        if (!j->files)
+        if (!j->files) {
+                r = -ENOMEM;
                 goto fail;
+        }
+
+        /* We ignore most errors here, since the idea is to only open
+         * what's actually accessible, and ignore the rest. */
 
         NULSTR_FOREACH(p, search_paths) {
                 DIR *d;
 
                 d = opendir(p);
                 if (!d) {
-                        if (errno != ENOENT && r == 0)
-                                r = -errno;
-
+                        if (errno != ENOENT)
+                                log_debug("Failed to open %s: %m", p);
                         continue;
                 }
 
                 for (;;) {
                         struct dirent buf, *de;
-                        int k;
-                        JournalFile *f;
-
-                        k = readdir_r(d, &buf, &de);
-                        if (k != 0) {
-                                if (r == 0)
-                                        r = -k;
-
-                                break;
-                        }
+                        sd_id128_t id;
 
-                        if (!de)
+                        r = readdir_r(d, &buf, &de);
+                        if (r != 0 || !de)
                                 break;
 
-                        if (!dirent_is_file_with_suffix(de, ".journal"))
-                                continue;
+                        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));
 
-                        fn = join(p, "/", de->d_name, NULL);
-                        if (!fn) {
-                                r = -ENOMEM;
-                                closedir(d);
-                                goto fail;
-                        }
-
-                        k = journal_file_open(fn, O_RDONLY, 0, NULL, &f);
-                        free(fn);
-
-                        if (k < 0) {
+                        } else if ((de->d_type == DT_DIR || de->d_type == DT_UNKNOWN) &&
+                                   sd_id128_from_string(de->d_name, &id) >= 0) {
 
-                                if (r == 0)
-                                        r = -k;
-                        } else {
-                                k = hashmap_put(j->files, f->path, f);
-                                if (k < 0) {
-                                        journal_file_close(f);
-                                        closedir(d);
-
-                                        r = k;
-                                        goto fail;
-                                }
+                                r = add_directory(j, p, de->d_name);
+                                if (r < 0)
+                                        log_debug("Failed to add directory %s/%s: %s", p, de->d_name, strerror(-r));
                         }
                 }
+
+                closedir(d);
         }
 
         *ret = j;
@@ -337,3 +397,165 @@ void sd_journal_close(sd_journal *j) {
 
         free(j);
 }
+
+int sd_journal_get_realtime_usec(sd_journal *j, uint64_t *ret) {
+        Object *o;
+        JournalFile *f;
+        int r;
+
+        assert(j);
+        assert(ret);
+
+        f = j->current_file;
+        if (!f)
+                return 0;
+
+        if (f->current_offset <= 0)
+                return 0;
+
+        r = journal_file_move_to_object(f, f->current_offset, OBJECT_ENTRY, &o);
+        if (r < 0)
+                return r;
+
+        *ret = le64toh(o->entry.realtime);
+        return 1;
+}
+
+int sd_journal_get_monotonic_usec(sd_journal *j, uint64_t *ret) {
+        Object *o;
+        JournalFile *f;
+        int r;
+        sd_id128_t id;
+
+        assert(j);
+        assert(ret);
+
+        f = j->current_file;
+        if (!f)
+                return 0;
+
+        if (f->current_offset <= 0)
+                return 0;
+
+        r = sd_id128_get_machine(&id);
+        if (r < 0)
+                return r;
+
+        r = journal_file_move_to_object(f, f->current_offset, OBJECT_ENTRY, &o);
+        if (r < 0)
+                return r;
+
+        if (!sd_id128_equal(id, o->entry.boot_id))
+                return 0;
+
+        *ret = le64toh(o->entry.monotonic);
+        return 1;
+
+}
+
+int sd_journal_get_field(sd_journal *j, const char *field, const void **data, size_t *size) {
+        JournalFile *f;
+        uint64_t i, n;
+        size_t field_length;
+        int r;
+        Object *o;
+
+        assert(j);
+        assert(field);
+        assert(data);
+        assert(size);
+
+        if (isempty(field) || strchr(field, '='))
+                return -EINVAL;
+
+        f = j->current_file;
+        if (!f)
+                return 0;
+
+        if (f->current_offset <= 0)
+                return 0;
+
+        r = journal_file_move_to_object(f, f->current_offset, OBJECT_ENTRY, &o);
+        if (r < 0)
+                return r;
+
+        field_length = strlen(field);
+
+        n = journal_file_entry_n_items(o);
+        for (i = 0; i < n; i++) {
+                uint64_t p, l;
+                size_t t;
+
+                p = le64toh(o->entry.items[i].object_offset);
+                r = journal_file_move_to_object(f, p, OBJECT_DATA, &o);
+                if (r < 0)
+                        return r;
+
+                l = le64toh(o->object.size) - offsetof(Object, data.payload);
+
+                if (field_length+1 > l)
+                        continue;
+
+                if (memcmp(o->data.payload, field, field_length) ||
+                    o->data.payload[field_length] != '=')
+                        continue;
+
+                t = (size_t) l;
+
+                if ((uint64_t) t != l)
+                        return -E2BIG;
+
+                *data = o->data.payload;
+                *size = t;
+
+                return 1;
+        }
+
+        return 0;
+}
+
+int sd_journal_iterate_fields(sd_journal *j, const void **data, size_t *size) {
+        JournalFile *f;
+        uint64_t p, l, n;
+        size_t t;
+        int r;
+        Object *o;
+
+        assert(j);
+        assert(data);
+        assert(size);
+
+        f = j->current_file;
+        if (!f)
+                return 0;
+
+        if (f->current_offset <= 0)
+                return 0;
+
+        r = journal_file_move_to_object(f, f->current_offset, OBJECT_ENTRY, &o);
+        if (r < 0)
+                return r;
+
+        n = journal_file_entry_n_items(o);
+        if (f->current_field >= n)
+                return 0;
+
+        p = le64toh(o->entry.items[f->current_field].object_offset);
+        r = journal_file_move_to_object(f, p, OBJECT_DATA, &o);
+        if (r < 0)
+                return r;
+
+        l = le64toh(o->object.size) - offsetof(Object, data.payload);
+        t = (size_t) l;
+
+        /* We can't read objects larger than 4G on a 32bit machine */
+        if ((uint64_t) t != l)
+                return -E2BIG;
+
+        *data = o->data.payload;
+        *size = t;
+
+        f->current_field ++;
+
+        return 1;
+}
diff --git a/src/journal/sd-journal.h b/src/journal/sd-journal.h
index 55e58601fe..bbfcda6f2a 100644
--- a/src/journal/sd-journal.h
+++ b/src/journal/sd-journal.h
@@ -27,13 +27,14 @@
 
 /* TODO:
  *
- *   - implement rotation
  *   - check LE/BE conversion for 8bit, 16bit, 32bit values
  *   - implement parallel traversal
  *   - implement inotify usage on client
  *   - implement audit gateway
  *   - implement native gateway
+ *   - implement stdout gateway
  *   - extend hash table/bisect table as we go
+ *   - accelerate looking for "all hostnames" and suchlike.
  */
 
 typedef struct sd_journal sd_journal;
@@ -44,10 +45,10 @@ void sd_journal_close(sd_journal *j);
 int sd_journal_previous(sd_journal *j);
 int sd_journal_next(sd_journal *j);
 
-int sd_journal_get(sd_journal *j, const char *field, const void **data, size_t *size);
-int sd_journal_get_seqnum(sd_journal *j, uint64_t *ret);
 int sd_journal_get_realtime_usec(sd_journal *j, uint64_t *ret);
 int sd_journal_get_monotonic_usec(sd_journal *j, uint64_t *ret);
+int sd_journal_get_field(sd_journal *j, const char *field, const void **data, size_t *l);
+int sd_journal_iterate_fields(sd_journal *j, const void **data, size_t *l);
 
 int sd_journal_add_match(sd_journal *j, const char *field, const void *data, size_t size);
 void sd_journal_flush_matches(sd_journal *j);
@@ -59,9 +60,18 @@ int sd_journal_seek_seqnum(sd_journal *j, uint64_t seqnum);
 int sd_journal_seek_monotonic_usec(sd_journal *j, uint64_t usec);
 int sd_journal_seek_realtime_usec(sd_journal *j, uint64_t usec);
 
-int sd_journal_get_cursor(sd_journal *j, void **cursor, size_t *size);
-int sd_journal_set_cursor(sd_journal *j, const void *cursor, size_t size);
+int sd_journal_get_cursor(sd_journal *j, char **cursor);
+int sd_journal_set_cursor(sd_journal *j, const char *cursor);
 
 int sd_journal_get_fd(sd_journal *j);
 
+#define SD_JOURNAL_FOREACH(j)                   \
+        while (sd_journal_next(j) > 0)
+
+#define SD_JOURNAL_FOREACH_BACKWARDS(j)         \
+        while (sd_journal_previous(j) > 0)
+
+#define SD_JOURNAL_FOREACH_FIELD(j, data, l)                            \
+        while (sd_journal_iterate_fields((j), &(data), &(l)) > 0)
+
 #endif