X-Git-Url: https://www.chiark.greenend.org.uk/ucgi/~ianmdlvl/git?a=blobdiff_plain;f=src%2Fjournal%2Fjournal-file.c;h=6c9deacf2ce1c26ccbe94e9b7fcaff070b86961b;hb=a4bcff5ba36115495994e9f9ba66074471de76ab;hp=e778e1c67c19a5f798a6bdd4291557e6272eeeba;hpb=c53158818d8cdaf46b3f1b5299b9bda118a1043f;p=elogind.git

diff --git a/src/journal/journal-file.c b/src/journal/journal-file.c
index e778e1c67..6c9deacf2 100644
--- a/src/journal/journal-file.c
+++ b/src/journal/journal-file.c
@@ -27,6 +27,10 @@
 #include <fcntl.h>
 #include <stddef.h>
 
+#ifdef HAVE_XATTR
+#include <attr/xattr.h>
+#endif
+
 #include "journal-def.h"
 #include "journal-file.h"
 #include "journal-authenticate.h"
@@ -61,6 +65,9 @@
 /* n_data was the first entry we added after the initial file format design */
 #define HEADER_SIZE_MIN ALIGN64(offsetof(Header, n_data))
 
+/* How many entries to keep in the entry array chain cache at max */
+#define CHAIN_CACHE_MAX 20
+
 void journal_file_close(JournalFile *f) {
         assert(f);
 
@@ -93,6 +100,8 @@ void journal_file_close(JournalFile *f) {
         if (f->mmap)
                 mmap_cache_unref(f->mmap);
 
+        hashmap_free_free(f->chain_cache);
+
 #ifdef HAVE_XZ
         free(f->compress_buffer);
 #endif
@@ -260,8 +269,7 @@ static int journal_file_verify_header(JournalFile *f) {
 
         f->compress = JOURNAL_HEADER_COMPRESSED(f->header);
 
-        if (f->writable)
-                f->seal = JOURNAL_HEADER_SEALED(f->header);
+        f->seal = JOURNAL_HEADER_SEALED(f->header);
 
         return 0;
 }
@@ -396,7 +404,7 @@ int journal_file_move_to_object(JournalFile *f, int type, uint64_t offset, Objec
         if (s < minimum_header_size(o))
                 return -EBADMSG;
 
-        if (type >= 0 && o->object.type != type)
+        if (type > 0 && o->object.type != type)
                 return -EBADMSG;
 
         if (s > sizeof(ObjectHeader)) {
@@ -523,6 +531,9 @@ static int journal_file_setup_field_hash_table(JournalFile *f) {
 
         assert(f);
 
+        /* We use a fixed size hash table for the fields as this
+         * number should grow very slowly only */
+
         s = DEFAULT_FIELD_HASH_TABLE_SIZE;
         r = journal_file_append_object(f,
                                        OBJECT_FIELD_HASH_TABLE,
@@ -583,7 +594,52 @@ static int journal_file_map_field_hash_table(JournalFile *f) {
         return 0;
 }
 
-static int journal_file_link_data(JournalFile *f, Object *o, uint64_t offset, uint64_t hash) {
+static int journal_file_link_field(
+                JournalFile *f,
+                Object *o,
+                uint64_t offset,
+                uint64_t hash) {
+
+        uint64_t p, h;
+        int r;
+
+        assert(f);
+        assert(o);
+        assert(offset > 0);
+
+        if (o->object.type != OBJECT_FIELD)
+                return -EINVAL;
+
+        /* This might alter the window we are looking at */
+
+        o->field.next_hash_offset = o->field.head_data_offset = 0;
+
+        h = hash % (le64toh(f->header->field_hash_table_size) / sizeof(HashItem));
+        p = le64toh(f->field_hash_table[h].tail_hash_offset);
+        if (p == 0)
+                f->field_hash_table[h].head_hash_offset = htole64(offset);
+        else {
+                r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o);
+                if (r < 0)
+                        return r;
+
+                o->field.next_hash_offset = htole64(offset);
+        }
+
+        f->field_hash_table[h].tail_hash_offset = htole64(offset);
+
+        if (JOURNAL_HEADER_CONTAINS(f->header, n_fields))
+                f->header->n_fields = htole64(le64toh(f->header->n_fields) + 1);
+
+        return 0;
+}
+
+static int journal_file_link_data(
+                JournalFile *f,
+                Object *o,
+                uint64_t offset,
+                uint64_t hash) {
+
         uint64_t p, h;
         int r;
 
@@ -602,10 +658,10 @@ static int journal_file_link_data(JournalFile *f, Object *o, uint64_t offset, ui
 
         h = hash % (le64toh(f->header->data_hash_table_size) / sizeof(HashItem));
         p = le64toh(f->data_hash_table[h].tail_hash_offset);
-        if (p == 0) {
+        if (p == 0)
                 /* Only entry in the hash table is easy */
                 f->data_hash_table[h].head_hash_offset = htole64(offset);
-        } else {
+        else {
                 /* Move back to the previous data object, to patch in
                  * pointer */
 
@@ -624,6 +680,67 @@ static int journal_file_link_data(JournalFile *f, Object *o, uint64_t offset, ui
         return 0;
 }
 
+int journal_file_find_field_object_with_hash(
+                JournalFile *f,
+                const void *field, uint64_t size, uint64_t hash,
+                Object **ret, uint64_t *offset) {
+
+        uint64_t p, osize, h;
+        int r;
+
+        assert(f);
+        assert(field && size > 0);
+
+        osize = offsetof(Object, field.payload) + size;
+
+        if (f->header->field_hash_table_size == 0)
+                return -EBADMSG;
+
+        h = hash % (le64toh(f->header->field_hash_table_size) / sizeof(HashItem));
+        p = le64toh(f->field_hash_table[h].head_hash_offset);
+
+        while (p > 0) {
+                Object *o;
+
+                r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o);
+                if (r < 0)
+                        return r;
+
+                if (le64toh(o->field.hash) == hash &&
+                    le64toh(o->object.size) == osize &&
+                    memcmp(o->field.payload, field, size) == 0) {
+
+                        if (ret)
+                                *ret = o;
+                        if (offset)
+                                *offset = p;
+
+                        return 1;
+                }
+
+                p = le64toh(o->field.next_hash_offset);
+        }
+
+        return 0;
+}
+
+int journal_file_find_field_object(
+                JournalFile *f,
+                const void *field, uint64_t size,
+                Object **ret, uint64_t *offset) {
+
+        uint64_t hash;
+
+        assert(f);
+        assert(field && size > 0);
+
+        hash = hash64(field, size);
+
+        return journal_file_find_field_object_with_hash(f,
+                                                        field, size, hash,
+                                                        ret, offset);
+}
+
 int journal_file_find_data_object_with_hash(
                 JournalFile *f,
                 const void *data, uint64_t size, uint64_t hash,
@@ -717,6 +834,66 @@ int journal_file_find_data_object(
                                                        ret, offset);
 }
 
+static int journal_file_append_field(
+                JournalFile *f,
+                const void *field, uint64_t size,
+                Object **ret, uint64_t *offset) {
+
+        uint64_t hash, p;
+        uint64_t osize;
+        Object *o;
+        int r;
+
+        assert(f);
+        assert(field && size > 0);
+
+        hash = hash64(field, size);
+
+        r = journal_file_find_field_object_with_hash(f, field, size, hash, &o, &p);
+        if (r < 0)
+                return r;
+        else if (r > 0) {
+
+                if (ret)
+                        *ret = o;
+
+                if (offset)
+                        *offset = p;
+
+                return 0;
+        }
+
+        osize = offsetof(Object, field.payload) + size;
+        r = journal_file_append_object(f, OBJECT_FIELD, osize, &o, &p);
+
+        o->field.hash = htole64(hash);
+        memcpy(o->field.payload, field, size);
+
+        r = journal_file_link_field(f, o, p, hash);
+        if (r < 0)
+                return r;
+
+        /* The linking might have altered the window, so let's
+         * refresh our pointer */
+        r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o);
+        if (r < 0)
+                return r;
+
+#ifdef HAVE_GCRYPT
+        r = journal_file_hmac_put_object(f, OBJECT_FIELD, o, p);
+        if (r < 0)
+                return r;
+#endif
+
+        if (ret)
+                *ret = o;
+
+        if (offset)
+                *offset = p;
+
+        return 0;
+}
+
 static int journal_file_append_data(
                 JournalFile *f,
                 const void *data, uint64_t size,
@@ -727,6 +904,7 @@ static int journal_file_append_data(
         Object *o;
         int r;
         bool compressed = false;
+        const void *eq;
 
         assert(f);
         assert(data || size == 0);
@@ -783,6 +961,21 @@ static int journal_file_append_data(
         if (r < 0)
                 return r;
 
+        eq = memchr(data, '=', size);
+        if (eq && eq > data) {
+                uint64_t fp;
+                Object *fo;
+
+                /* Create field object ... */
+                r = journal_file_append_field(f, data, (uint8_t*) eq - (uint8_t*) data, &fo, &fp);
+                if (r < 0)
+                        return r;
+
+                /* ... and link it in. */
+                o->data.next_field_offset = fo->field.head_data_offset;
+                fo->field.head_data_offset = le64toh(p);
+        }
+
 #ifdef HAVE_GCRYPT
         r = journal_file_hmac_put_object(f, OBJECT_DATA, o, p);
         if (r < 0)
@@ -1054,6 +1247,16 @@ void journal_file_post_change(JournalFile *f) {
                 log_error("Failed to truncate file to its own size: %m");
 }
 
+static int entry_item_cmp(const void *_a, const void *_b) {
+        const EntryItem *a = _a, *b = _b;
+
+        if (le64toh(a->object_offset) < le64toh(b->object_offset))
+                return -1;
+        if (le64toh(a->object_offset) > le64toh(b->object_offset))
+                return 1;
+        return 0;
+}
+
 int journal_file_append_entry(JournalFile *f, const dual_timestamp *ts, const struct iovec iovec[], unsigned n_iovec, uint64_t *seqnum, Object **ret, uint64_t *offset) {
         unsigned i;
         EntryItem *items;
@@ -1098,6 +1301,10 @@ int journal_file_append_entry(JournalFile *f, const dual_timestamp *ts, const st
                 items[i].hash = o->data.hash;
         }
 
+        /* Order by the position on disk, in order to improve seek
+         * times for rotating media. */
+        qsort(items, n_iovec, sizeof(EntryItem), entry_item_cmp);
+
         r = journal_file_append_entry_internal(f, ts, xor_hash, items, n_iovec, seqnum, ret, offset);
 
         journal_file_post_change(f);
@@ -1105,37 +1312,89 @@ int journal_file_append_entry(JournalFile *f, const dual_timestamp *ts, const st
         return r;
 }
 
+typedef struct ChainCacheItem {
+        uint64_t first; /* the array at the begin of the chain */
+        uint64_t array; /* the cached array */
+        uint64_t begin; /* the first item in the cached array */
+        uint64_t total; /* the total number of items in all arrays before this one in the chain */
+} ChainCacheItem;
+
+static void chain_cache_put(
+                Hashmap *h,
+                ChainCacheItem *ci,
+                uint64_t first,
+                uint64_t array,
+                uint64_t begin,
+                uint64_t total) {
+
+        if (!ci) {
+                if (hashmap_size(h) >= CHAIN_CACHE_MAX)
+                        ci = hashmap_steal_first(h);
+                else {
+                        ci = new(ChainCacheItem, 1);
+                        if (!ci)
+                                return;
+                }
+
+                ci->first = first;
+
+                if (hashmap_put(h, &ci->first, ci) < 0) {
+                        free(ci);
+                        return;
+                }
+        } else
+                assert(ci->first == first);
+
+        ci->array = array;
+        ci->begin = begin;
+        ci->total = total;
+}
+
 static int generic_array_get(JournalFile *f,
                              uint64_t first,
                              uint64_t i,
                              Object **ret, uint64_t *offset) {
 
         Object *o;
-        uint64_t p = 0, a;
+        uint64_t p = 0, a, t = 0;
         int r;
+        ChainCacheItem *ci;
 
         assert(f);
 
         a = first;
+
+        /* Try the chain cache first */
+        ci = hashmap_get(f->chain_cache, &first);
+        if (ci && i > ci->total) {
+                a = ci->array;
+                i -= ci->total;
+                t = ci->total;
+        }
+
         while (a > 0) {
-                uint64_t n;
+                uint64_t k;
 
                 r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
                 if (r < 0)
                         return r;
 
-                n = journal_file_entry_array_n_items(o);
-                if (i < n) {
+                k = journal_file_entry_array_n_items(o);
+                if (i < k) {
                         p = le64toh(o->entry_array.items[i]);
-                        break;
+                        goto found;
                 }
 
-                i -= n;
+                i -= k;
+                t += k;
                 a = le64toh(o->entry_array.next_entry_array_offset);
         }
 
-        if (a <= 0 || p <= 0)
-                return 0;
+        return 0;
+
+found:
+        /* Let's cache this item for the next invocation */
+        chain_cache_put(f->chain_cache, ci, first, a, o->entry_array.items[0], t);
 
         r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
         if (r < 0)
@@ -1199,11 +1458,38 @@ static int generic_array_bisect(JournalFile *f,
         bool subtract_one = false;
         Object *o, *array = NULL;
         int r;
+        ChainCacheItem *ci;
 
         assert(f);
         assert(test_object);
 
+        /* Start with the first array in the chain */
         a = first;
+
+        ci = hashmap_get(f->chain_cache, &first);
+        if (ci && n > ci->total) {
+                /* Ah, we have iterated this bisection array chain
+                 * previously! Let's see if we can skip ahead in the
+                 * chain, as far as the last time. But we can't jump
+                 * backwards in the chain, so let's check that
+                 * first. */
+
+                r = test_object(f, ci->begin, needle);
+                if (r < 0)
+                        return r;
+
+                if (r == TEST_LEFT) {
+                        /* OK, what we are looking for is right of th
+                         * begin of this EntryArray, so let's jump
+                         * straight to previously cached array in the
+                         * chain */
+
+                        a = ci->array;
+                        n -= ci->total;
+                        t = ci->total;
+                }
+        }
+
         while (a > 0) {
                 uint64_t left, right, k, lp;
 
@@ -1284,6 +1570,9 @@ found:
         if (subtract_one && t == 0 && i == 0)
                 return 0;
 
+        /* Let's cache this item for the next invocation */
+        chain_cache_put(f->chain_cache, ci, first, a, array->entry_array.items[0], t);
+
         if (subtract_one && i == 0)
                 p = last_p;
         else if (subtract_one)
@@ -1882,6 +2171,10 @@ void journal_file_dump(JournalFile *f) {
                         printf("Type: OBJECT_DATA\n");
                         break;
 
+                case OBJECT_FIELD:
+                        printf("Type: OBJECT_FIELD\n");
+                        break;
+
                 case OBJECT_ENTRY:
                         printf("Type: OBJECT_ENTRY seqnum=%llu monotonic=%llu realtime=%llu\n",
                                (unsigned long long) le64toh(o->entry.seqnum),
@@ -1906,6 +2199,10 @@ void journal_file_dump(JournalFile *f) {
                                (unsigned long long) le64toh(o->tag.seqnum),
                                (unsigned long long) le64toh(o->tag.epoch));
                         break;
+
+                default:
+                        printf("Type: unknown (%u)\n", o->object.type);
+                        break;
                 }
 
                 if (o->object.flags & OBJECT_COMPRESSED)
@@ -1965,7 +2262,7 @@ void journal_file_print_header(JournalFile *f) {
                (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),
-               yes_no(journal_file_rotate_suggested(f)),
+               yes_no(journal_file_rotate_suggested(f, 0)),
                (unsigned long long) le64toh(f->header->head_entry_seqnum),
                (unsigned long long) le64toh(f->header->tail_entry_seqnum),
                format_timestamp(x, sizeof(x), le64toh(f->header->head_entry_realtime)),
@@ -2012,6 +2309,7 @@ int journal_file_open(
         bool newly_created = false;
 
         assert(fname);
+        assert(ret);
 
         if ((flags & O_ACCMODE) != O_RDONLY &&
             (flags & O_ACCMODE) != O_RDWR)
@@ -2054,6 +2352,12 @@ int journal_file_open(
                 goto fail;
         }
 
+        f->chain_cache = hashmap_new(uint64_hash_func, uint64_compare_func);
+        if (!f->chain_cache) {
+                r = -ENOMEM;
+                goto fail;
+        }
+
         f->fd = open(f->path, f->flags|O_CLOEXEC, f->mode);
         if (f->fd < 0) {
                 r = -errno;
@@ -2066,7 +2370,22 @@ int journal_file_open(
         }
 
         if (f->last_stat.st_size == 0 && f->writable) {
-                newly_created = true;
+#ifdef HAVE_XATTR
+                uint64_t crtime;
+
+                /* Let's attach the creation time to the journal file,
+                 * so that the vacuuming code knows the age of this
+                 * file even if the file might end up corrupted one
+                 * day... Ideally we'd just use the creation time many
+                 * file systems maintain for each file, but there is
+                 * currently no usable API to query this, hence let's
+                 * emulate this via extended attributes. If extended
+                 * attributes are not supported we'll just skip this,
+                 * and rely solely on mtime/atime/ctime of the file.*/
+
+                crtime = htole64((uint64_t) now(CLOCK_REALTIME));
+                fsetxattr(f->fd, "user.crtime_usec", &crtime, sizeof(crtime), XATTR_CREATE);
+#endif
 
 #ifdef HAVE_GCRYPT
                 /* Try to load the FSPRG state, and if we can't, then
@@ -2086,6 +2405,8 @@ int journal_file_open(
                         r = -errno;
                         goto fail;
                 }
+
+                newly_created = true;
         }
 
         if (f->last_stat.st_size < (off_t) HEADER_SIZE_MIN) {
@@ -2156,9 +2477,7 @@ int journal_file_open(
         if (r < 0)
                 goto fail;
 
-        if (ret)
-                *ret = f;
-
+        *ret = f;
         return 0;
 
 fail:
@@ -2195,8 +2514,8 @@ int journal_file_rotate(JournalFile **f, bool compress, bool seal) {
         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->tail_entry_seqnum),
-                 (unsigned long long) le64toh((*f)->header->tail_entry_realtime));
+                 (unsigned long long) le64toh((*f)->header->head_entry_seqnum),
+                 (unsigned long long) le64toh((*f)->header->head_entry_realtime));
 
         r = rename(old_file->path, p);
         free(p);
@@ -2489,7 +2808,7 @@ int journal_file_get_cutoff_monotonic_usec(JournalFile *f, sd_id128_t boot_id, u
         return 1;
 }
 
-bool journal_file_rotate_suggested(JournalFile *f) {
+bool journal_file_rotate_suggested(JournalFile *f, usec_t max_file_usec) {
         assert(f);
 
         /* If we gained new header fields we gained new features,
@@ -2527,5 +2846,22 @@ bool journal_file_rotate_suggested(JournalFile *f) {
                         return true;
                 }
 
+        /* Are the data objects properly indexed by field objects? */
+        if (JOURNAL_HEADER_CONTAINS(f->header, n_data) &&
+            JOURNAL_HEADER_CONTAINS(f->header, n_fields) &&
+            le64toh(f->header->n_data) > 0 &&
+            le64toh(f->header->n_fields) == 0)
+                return true;
+
+        if (max_file_usec > 0) {
+                usec_t t, h;
+
+                h = le64toh(f->header->head_entry_realtime);
+                t = now(CLOCK_REALTIME);
+
+                if (h > 0 && t > h + max_file_usec)
+                        return true;
+        }
+
         return false;
 }