X-Git-Url: http://www.chiark.greenend.org.uk/ucgi/~ianmdlvl/git?p=elogind.git;a=blobdiff_plain;f=src%2Fjournal%2Fjournal-file.c;h=718dc5d6eaa13f73e77ce8bb5f2106d9a1aa19ed;hp=a110a0090f33c217ca6e55354475adbcd3893733;hb=64825d3c589cd8742887f30acde8c57eceac2001;hpb=162566a4a12c35e1e86e35ced1748354f7ec935e

diff --git a/src/journal/journal-file.c b/src/journal/journal-file.c
index a110a0090..718dc5d6e 100644
--- a/src/journal/journal-file.c
+++ b/src/journal/journal-file.c
@@ -32,8 +32,8 @@
 #include "lookup3.h"
 #include "compress.h"
 
-#define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*16ULL)
-#define DEFAULT_FIELD_HASH_TABLE_SIZE (2047ULL*16ULL)
+#define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*sizeof(HashItem))
+#define DEFAULT_FIELD_HASH_TABLE_SIZE (333ULL*sizeof(HashItem))
 
 #define DEFAULT_WINDOW_SIZE (8ULL*1024ULL*1024ULL)
 
@@ -58,17 +58,24 @@
  * size */
 #define DEFAULT_KEEP_FREE (1024ULL*1024ULL)                    /* 1 MB */
 
-static const char signature[] = { 'L', 'P', 'K', 'S', 'H', 'H', 'R', 'H' };
+/* n_data was the first entry we added after the initial file format design */
+#define HEADER_SIZE_MIN ALIGN64(offsetof(Header, n_data))
 
 #define ALIGN64(x) (((x) + 7ULL) & ~7ULL)
 
+#define JOURNAL_HEADER_CONTAINS(h, field) \
+        (le64toh((h)->header_size) >= offsetof(Header, field) + sizeof((h)->field))
+
+static const char signature[] = { 'L', 'P', 'K', 'S', 'H', 'H', 'R', 'H' };
+
 void journal_file_close(JournalFile *f) {
         int t;
 
         assert(f);
 
         if (f->header) {
-                if (f->writable)
+                /* Mark the file offline. Don't override the archived state if it already is set */
+                if (f->writable && f->header->state == STATE_ONLINE)
                         f->header->state = STATE_OFFLINE;
 
                 munmap(f->header, PAGE_ALIGN(sizeof(Header)));
@@ -107,7 +114,7 @@ static int journal_file_init_header(JournalFile *f, JournalFile *template) {
 
         if (template) {
                 h.seqnum_id = template->header->seqnum_id;
-                h.seqnum = template->header->seqnum;
+                h.tail_seqnum = template->header->tail_seqnum;
         } else
                 h.seqnum_id = h.file_id;
 
@@ -161,7 +168,8 @@ static int journal_file_verify_header(JournalFile *f) {
                 return -EPROTONOSUPPORT;
 #endif
 
-        if (f->header->header_size != htole64(ALIGN64(sizeof(*(f->header)))))
+        /* The first addition was n_data, so check that we are at least this large */
+        if (le64toh(f->header->header_size) < HEADER_SIZE_MIN)
                 return -EBADMSG;
 
         if ((uint64_t) f->last_stat.st_size < (le64toh(f->header->header_size) + le64toh(f->header->arena_size)))
@@ -181,13 +189,15 @@ static int journal_file_verify_header(JournalFile *f) {
 
                 state = f->header->state;
 
-                if (state == STATE_ONLINE)
-                        log_debug("Journal file %s is already online. Assuming unclean closing. Ignoring.", f->path);
-                        /* FIXME: immediately rotate */
-                else if (state == STATE_ARCHIVED)
+                if (state == STATE_ONLINE) {
+                        log_debug("Journal file %s is already online. Assuming unclean closing.", f->path);
+                        return -EBUSY;
+                } else if (state == STATE_ARCHIVED)
                         return -ESHUTDOWN;
-                else if (state != STATE_OFFLINE)
-                        log_debug("Journal file %s has unknown state %u. Ignoring.", f->path, state);
+                else if (state != STATE_OFFLINE) {
+                        log_debug("Journal file %s has unknown state %u.", f->path, state);
+                        return -EBUSY;
+                }
         }
 
         return 0;
@@ -427,7 +437,7 @@ static uint64_t journal_file_seqnum(JournalFile *f, uint64_t *seqnum) {
 
         assert(f);
 
-        r = le64toh(f->header->seqnum) + 1;
+        r = le64toh(f->header->tail_seqnum) + 1;
 
         if (seqnum) {
                 /* If an external seqnum counter was passed, we update
@@ -440,10 +450,10 @@ static uint64_t journal_file_seqnum(JournalFile *f, uint64_t *seqnum) {
                 *seqnum = r;
         }
 
-        f->header->seqnum = htole64(r);
+        f->header->tail_seqnum = htole64(r);
 
-        if (f->header->first_seqnum == 0)
-                f->header->first_seqnum = htole64(r);
+        if (f->header->head_seqnum == 0)
+                f->header->head_seqnum = htole64(r);
 
         return r;
 }
@@ -500,7 +510,17 @@ static int journal_file_setup_data_hash_table(JournalFile *f) {
 
         assert(f);
 
-        s = DEFAULT_DATA_HASH_TABLE_SIZE;
+        /* We estimate that we need 1 hash table entry per 768 of
+           journal file and we want to make sure we never get beyond
+           75% fill level. Calculate the hash table size for the
+           maximum file size based on these metrics. */
+
+        s = (f->metrics.max_size * 4 / 768 / 3) * sizeof(HashItem);
+        if (s < DEFAULT_DATA_HASH_TABLE_SIZE)
+                s = DEFAULT_DATA_HASH_TABLE_SIZE;
+
+        log_info("Reserving %llu entries in hash table.", (unsigned long long) (s / sizeof(HashItem)));
+
         r = journal_file_append_object(f,
                                        OBJECT_DATA_HASH_TABLE,
                                        offsetof(Object, hash_table.items) + s,
@@ -614,6 +634,9 @@ static int journal_file_link_data(JournalFile *f, Object *o, uint64_t offset, ui
 
         f->data_hash_table[h].tail_hash_offset = htole64(offset);
 
+        if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
+                f->header->n_data = htole64(le64toh(f->header->n_data) + 1);
+
         return 0;
 }
 
@@ -765,7 +788,7 @@ static int journal_file_append_data(
         }
 #endif
 
-        if (!compressed)
+        if (!compressed && size > 0)
                 memcpy(o->data.payload, data, size);
 
         r = journal_file_link_data(f, o, p, hash);
@@ -1034,7 +1057,8 @@ int journal_file_append_entry(JournalFile *f, const dual_timestamp *ts, const st
             ts->monotonic < le64toh(f->header->tail_entry_monotonic))
                 return -EINVAL;
 
-        items = alloca(sizeof(EntryItem) * n_iovec);
+        /* alloca() can't take 0, hence let's allocate at least one */
+        items = alloca(sizeof(EntryItem) * MAX(1, n_iovec));
 
         for (i = 0; i < n_iovec; i++) {
                 uint64_t p;
@@ -1809,27 +1833,13 @@ int journal_file_move_to_entry_by_realtime_for_data(
 }
 
 void journal_file_dump(JournalFile *f) {
-        char a[33], b[33], c[33];
         Object *o;
         int r;
         uint64_t p;
 
         assert(f);
 
-        printf("File Path: %s\n"
-               "File ID: %s\n"
-               "Machine ID: %s\n"
-               "Boot ID: %s\n"
-               "Arena size: %llu\n"
-               "Objects: %lu\n"
-               "Entries: %lu\n",
-               f->path,
-               sd_id128_to_string(f->header->file_id, a),
-               sd_id128_to_string(f->header->machine_id, b),
-               sd_id128_to_string(f->header->boot_id, c),
-               (unsigned long long) le64toh(f->header->arena_size),
-               (unsigned long) le64toh(f->header->n_objects),
-               (unsigned long) le64toh(f->header->n_entries));
+        journal_file_print_header(f);
 
         p = le64toh(f->header->header_size);
         while (p != 0) {
@@ -1885,10 +1895,73 @@ fail:
         log_error("File corrupt");
 }
 
+void journal_file_print_header(JournalFile *f) {
+        char a[33], b[33], c[33];
+        char x[FORMAT_TIMESTAMP_MAX], y[FORMAT_TIMESTAMP_MAX];
+
+        assert(f);
+
+        printf("File Path: %s\n"
+               "File ID: %s\n"
+               "Machine ID: %s\n"
+               "Boot ID: %s\n"
+               "Sequential Number ID: %s\n"
+               "State: %s\n"
+               "Compatible Flags:%s%s\n"
+               "Incompatible Flags:%s%s\n"
+               "Header size: %llu\n"
+               "Arena size: %llu\n"
+               "Data Hash Table Size: %llu\n"
+               "Field Hash Table Size: %llu\n"
+               "Objects: %llu\n"
+               "Entry Objects: %llu\n"
+               "Rotate Suggested: %s\n"
+               "Head Sequential Number: %llu\n"
+               "Tail Sequential Number: %llu\n"
+               "Head Realtime Timestamp: %s\n"
+               "Tail Realtime Timestamp: %s\n",
+               f->path,
+               sd_id128_to_string(f->header->file_id, a),
+               sd_id128_to_string(f->header->machine_id, b),
+               sd_id128_to_string(f->header->boot_id, c),
+               sd_id128_to_string(f->header->seqnum_id, c),
+               f->header->state == STATE_OFFLINE ? "offline" :
+               f->header->state == STATE_ONLINE ? "online" :
+               f->header->state == STATE_ARCHIVED ? "archived" : "unknown",
+               (f->header->compatible_flags & HEADER_COMPATIBLE_SIGNED) ? " SIGNED" : "",
+               (f->header->compatible_flags & ~HEADER_COMPATIBLE_SIGNED) ? " ???" : "",
+               (f->header->incompatible_flags & HEADER_INCOMPATIBLE_COMPRESSED) ? " COMPRESSED" : "",
+               (f->header->incompatible_flags & ~HEADER_INCOMPATIBLE_COMPRESSED) ? " ???" : "",
+               (unsigned long long) le64toh(f->header->header_size),
+               (unsigned long long) le64toh(f->header->arena_size),
+               (unsigned long long) le64toh(f->header->data_hash_table_size) / sizeof(HashItem),
+               (unsigned long long) le64toh(f->header->field_hash_table_size) / sizeof(HashItem),
+               (unsigned long long) le64toh(f->header->n_objects),
+               (unsigned long long) le64toh(f->header->n_entries),
+               yes_no(journal_file_rotate_suggested(f)),
+               (unsigned long long) le64toh(f->header->head_seqnum),
+               (unsigned long long) le64toh(f->header->tail_seqnum),
+               format_timestamp(x, sizeof(x), le64toh(f->header->head_entry_realtime)),
+               format_timestamp(y, sizeof(y), le64toh(f->header->tail_entry_realtime)));
+
+        if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
+                printf("Data Objects: %llu\n"
+                       "Data Hash Table Fill: %.1f%%\n",
+                       (unsigned long long) le64toh(f->header->n_data),
+                       100.0 * (double) le64toh(f->header->n_data) / ((double) (le64toh(f->header->data_hash_table_size) / sizeof(HashItem))));
+
+        if (JOURNAL_HEADER_CONTAINS(f->header, n_fields))
+                printf("Field Objects: %llu\n"
+                       "Field Hash Table Fill: %.1f%%\n",
+                       (unsigned long long) le64toh(f->header->n_fields),
+                       100.0 * (double) le64toh(f->header->n_fields) / ((double) (le64toh(f->header->field_hash_table_size) / sizeof(HashItem))));
+}
+
 int journal_file_open(
                 const char *fname,
                 int flags,
                 mode_t mode,
+                JournalMetrics *metrics,
                 JournalFile *template,
                 JournalFile **ret) {
 
@@ -1915,10 +1988,8 @@ int journal_file_open(
         f->writable = (flags & O_ACCMODE) != O_RDONLY;
         f->prot = prot_from_flags(flags);
 
-        if (template) {
-                f->metrics = template->metrics;
+        if (template)
                 f->compress = template->compress;
-        }
 
         f->path = strdup(fname);
         if (!f->path) {
@@ -1950,7 +2021,7 @@ int journal_file_open(
                 }
         }
 
-        if (f->last_stat.st_size < (off_t) sizeof(Header)) {
+        if (f->last_stat.st_size < (off_t) HEADER_SIZE_MIN) {
                 r = -EIO;
                 goto fail;
         }
@@ -1969,6 +2040,12 @@ int journal_file_open(
         }
 
         if (f->writable) {
+                if (metrics) {
+                        journal_default_metrics(metrics, f->fd);
+                        f->metrics = *metrics;
+                } else if (template)
+                        f->metrics = template->metrics;
+
                 r = journal_file_refresh_header(f);
                 if (r < 0)
                         goto fail;
@@ -2032,7 +2109,7 @@ int journal_file_rotate(JournalFile **f) {
         sd_id128_to_string(old_file->header->seqnum_id, p + l - 8 + 1);
         snprintf(p + l - 8 + 1 + 32, 1 + 16 + 1 + 16 + 8 + 1,
                  "-%016llx-%016llx.journal",
-                 (unsigned long long) le64toh((*f)->header->seqnum),
+                 (unsigned long long) le64toh((*f)->header->tail_seqnum),
                  (unsigned long long) le64toh((*f)->header->tail_entry_realtime));
 
         r = rename(old_file->path, p);
@@ -2043,7 +2120,7 @@ int journal_file_rotate(JournalFile **f) {
 
         old_file->header->state = STATE_ARCHIVED;
 
-        r = journal_file_open(old_file->path, old_file->flags, old_file->mode, old_file, &new_file);
+        r = journal_file_open(old_file->path, old_file->flags, old_file->mode, NULL, old_file, &new_file);
         journal_file_close(old_file);
 
         *f = new_file;
@@ -2054,6 +2131,7 @@ int journal_file_open_reliably(
                 const char *fname,
                 int flags,
                 mode_t mode,
+                JournalMetrics *metrics,
                 JournalFile *template,
                 JournalFile **ret) {
 
@@ -2061,11 +2139,13 @@ int journal_file_open_reliably(
         size_t l;
         char *p;
 
-        r = journal_file_open(fname, flags, mode, template, ret);
+        r = journal_file_open(fname, flags, mode, metrics, template, ret);
         if (r != -EBADMSG && /* corrupted */
             r != -ENODATA && /* truncated */
             r != -EHOSTDOWN && /* other machine */
-            r != -EPROTONOSUPPORT) /* incompatible feature */
+            r != -EPROTONOSUPPORT && /* incompatible feature */
+            r != -EBUSY && /* unclean shutdown */
+            r != -ESHUTDOWN /* already archived */)
                 return r;
 
         if ((flags & O_ACCMODE) == O_RDONLY)
@@ -2088,9 +2168,9 @@ int journal_file_open_reliably(
         if (r < 0)
                 return -errno;
 
-        log_warning("File %s corrupted, renaming and replacing.", fname);
+        log_warning("File %s corrupted or uncleanly shut down, renaming and replacing.", fname);
 
-        return journal_file_open(fname, flags, mode, template, ret);
+        return journal_file_open(fname, flags, mode, metrics, template, ret);
 }
 
 struct vacuum_info {
@@ -2257,7 +2337,8 @@ int journal_directory_vacuum(const char *directory, uint64_t max_use, uint64_t m
                 n_list ++;
         }
 
-        qsort(list, n_list, sizeof(struct vacuum_info), vacuum_compare);
+        if (n_list > 0)
+                qsort(list, n_list, sizeof(struct vacuum_info), vacuum_compare);
 
         for(i = 0; i < n_list; i++) {
                 struct statvfs ss;
@@ -2510,3 +2591,44 @@ int journal_file_get_cutoff_monotonic_usec(JournalFile *f, sd_id128_t boot_id, u
 
         return 1;
 }
+
+bool journal_file_rotate_suggested(JournalFile *f) {
+        assert(f);
+
+        /* If we gained new header fields we gained new features,
+         * hence suggest a rotation */
+        if (le64toh(f->header->header_size) < sizeof(Header)) {
+                log_debug("%s uses an outdated header, suggesting rotation.", f->path);
+                return true;
+        }
+
+        /* Let's check if the hash tables grew over a certain fill
+         * level (75%, borrowing this value from Java's hash table
+         * implementation), and if so suggest a rotation. To calculate
+         * the fill level we need the n_data field, which only exists
+         * in newer versions. */
+
+        if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
+                if (le64toh(f->header->n_data) * 4ULL > (le64toh(f->header->data_hash_table_size) / sizeof(HashItem)) * 3ULL) {
+                        log_debug("Data hash table of %s has a fill level at %.1f (%llu of %llu items, %llu file size, %llu bytes per hash table item), suggesting rotation.",
+                                  f->path,
+                                  100.0 * (double) le64toh(f->header->n_data) / ((double) (le64toh(f->header->data_hash_table_size) / sizeof(HashItem))),
+                                  (unsigned long long) le64toh(f->header->n_data),
+                                  (unsigned long long) (le64toh(f->header->data_hash_table_size) / sizeof(HashItem)),
+                                  (unsigned long long) (f->last_stat.st_size),
+                                  (unsigned long long) (f->last_stat.st_size / le64toh(f->header->n_data)));
+                        return true;
+                }
+
+        if (JOURNAL_HEADER_CONTAINS(f->header, n_fields))
+                if (le64toh(f->header->n_fields) * 4ULL > (le64toh(f->header->field_hash_table_size) / sizeof(HashItem)) * 3ULL) {
+                        log_debug("Field hash table of %s has a fill level at %.1f (%llu of %llu items), suggesting rotation.",
+                                  f->path,
+                                  100.0 * (double) le64toh(f->header->n_fields) / ((double) (le64toh(f->header->field_hash_table_size) / sizeof(HashItem))),
+                                  (unsigned long long) le64toh(f->header->n_fields),
+                                  (unsigned long long) (le64toh(f->header->field_hash_table_size) / sizeof(HashItem)));
+                        return true;
+                }
+
+        return false;
+}