#define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*16ULL)
#define DEFAULT_FIELD_HASH_TABLE_SIZE (2047ULL*16ULL)
-#define DEFAULT_WINDOW_SIZE (128ULL*1024ULL*1024ULL)
+#define DEFAULT_WINDOW_SIZE (8ULL*1024ULL*1024ULL)
-#define COMPRESSION_SIZE_THRESHOLD (64ULL)
+#define COMPRESSION_SIZE_THRESHOLD (512ULL)
+
+/* This is the minimum journal file size */
+#define JOURNAL_FILE_SIZE_MIN (64ULL*1024ULL) /* 64 KiB */
+
+/* These are the lower and upper bounds if we deduce the max_use value
+ * from the file system size */
+#define DEFAULT_MAX_USE_LOWER (1ULL*1024ULL*1024ULL) /* 1 MiB */
+#define DEFAULT_MAX_USE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
+
+/* This is the upper bound if we deduce max_size from max_use */
+#define DEFAULT_MAX_SIZE_UPPER (128ULL*1024ULL*1024ULL) /* 128 MiB */
+
+/* This is the upper bound if we deduce the keep_free value from the
+ * file system size */
+#define DEFAULT_KEEP_FREE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
+
+/* This is the keep_free value when we can't determine the system
+ * size */
+#define DEFAULT_KEEP_FREE (1024ULL*1024ULL) /* 1 MB */
static const char signature[] = { 'L', 'P', 'K', 'S', 'H', 'H', 'R', 'H' };
f->header->boot_id = boot_id;
f->header->state = STATE_ONLINE;
+
+ __sync_synchronize();
+
return 0;
}
return -ENODATA;
if (f->writable) {
- uint32_t state;
+ uint8_t state;
sd_id128_t machine_id;
int r;
if (fstat(f->fd, &f->last_stat) < 0)
return -errno;
- f->header->arena_size = new_size - htole64(f->header->arena_offset);
+ f->header->arena_size = htole64(new_size - le64toh(f->header->arena_offset));
return 0;
}
}
static int journal_file_move_to(JournalFile *f, int wt, uint64_t offset, uint64_t size, void **ret) {
- void *p;
+ void *p = NULL;
uint64_t delta;
int r;
Window *w;
assert(wt >= 0);
assert(wt < _WINDOW_MAX);
+ if (offset + size > (uint64_t) f->last_stat.st_size) {
+ /* Hmm, out of range? Let's refresh the fstat() data
+ * first, before we trust that check. */
+
+ if (fstat(f->fd, &f->last_stat) < 0 ||
+ offset + size > (uint64_t) f->last_stat.st_size)
+ return -EADDRNOTAVAIL;
+ }
+
w = f->windows + wt;
if (_likely_(w->ptr &&
* the window space before and half behind the
* requested mapping */
- delta = PAGE_ALIGN((DEFAULT_WINDOW_SIZE - size) / 2);
+ delta = (DEFAULT_WINDOW_SIZE - size) / 2;
- if (offset < delta)
+ if (delta > offset)
delta = offset;
offset -= delta;
- size += (DEFAULT_WINDOW_SIZE - delta);
+ size = DEFAULT_WINDOW_SIZE;
} else
delta = 0;
- if (offset > (uint64_t) f->last_stat.st_size)
- return -EADDRNOTAVAIL;
-
if (offset + size > (uint64_t) f->last_stat.st_size)
- size = PAGE_ALIGN((uint64_t) f->last_stat.st_size - offset);
+ size = (uint64_t) f->last_stat.st_size - offset;
if (size <= 0)
return -EADDRNOTAVAIL;
assert(offset > 0);
assert(o->object.type == OBJECT_DATA);
+ /* This might alter the window we are looking at */
+
o->data.next_hash_offset = o->data.next_field_offset = 0;
o->data.entry_offset = o->data.entry_array_offset = 0;
o->data.n_entries = 0;
/* Only entry in the hash table is easy */
f->data_hash_table[h].head_hash_offset = htole64(offset);
} else {
- /* Temporarily move back to the previous data object,
- * to patch in pointer */
+ /* Move back to the previous data object, to patch in
+ * pointer */
r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
if (r < 0)
return r;
o->data.next_hash_offset = htole64(offset);
-
- r = journal_file_move_to_object(f, OBJECT_DATA, offset, &o);
- if (r < 0)
- return r;
}
f->data_hash_table[h].tail_hash_offset = htole64(offset);
JournalFile *f,
const void *data, uint64_t size, uint64_t hash,
Object **ret, uint64_t *offset) {
+
uint64_t p, osize, h;
int r;
ret, offset);
}
-static int journal_file_append_data(JournalFile *f, const void *data, uint64_t size, Object **ret, uint64_t *offset) {
+static int journal_file_append_data(
+ JournalFile *f,
+ const void *data, uint64_t size,
+ Object **ret, uint64_t *offset) {
+
uint64_t hash, p;
uint64_t osize;
Object *o;
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_DATA, p, &o);
+ if (r < 0)
+ return r;
+
if (ret)
*ret = o;
uint64_t journal_file_entry_n_items(Object *o) {
assert(o);
- assert(o->object.type == htole64(OBJECT_ENTRY));
+ assert(o->object.type == OBJECT_ENTRY);
return (le64toh(o->object.size) - offsetof(Object, entry.items)) / sizeof(EntryItem);
}
static uint64_t journal_file_entry_array_n_items(Object *o) {
assert(o);
- assert(o->object.type == htole64(OBJECT_ENTRY_ARRAY));
+ assert(o->object.type == OBJECT_ENTRY_ARRAY);
return (le64toh(o->object.size) - offsetof(Object, entry_array.items)) / sizeof(uint64_t);
}
static int link_entry_into_array(JournalFile *f,
- uint64_t *first,
- uint64_t *idx,
+ le64_t *first,
+ le64_t *idx,
uint64_t p) {
int r;
uint64_t n = 0, ap = 0, q, i, a, hidx;
o->entry_array.items[i] = htole64(p);
if (ap == 0)
- *first = q;
+ *first = htole64(q);
else {
r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, ap, &o);
if (r < 0)
}
static int link_entry_into_array_plus_one(JournalFile *f,
- uint64_t *extra,
- uint64_t *first,
- uint64_t *idx,
+ le64_t *extra,
+ le64_t *first,
+ le64_t *idx,
uint64_t p) {
int r;
if (*idx == 0)
*extra = htole64(p);
else {
- uint64_t i;
+ le64_t i;
- i = le64toh(*idx) - 1;
+ i = htole64(le64toh(*idx) - 1);
r = link_entry_into_array(f, first, &i, p);
if (r < 0)
return r;
assert(offset > 0);
assert(o->object.type == OBJECT_ENTRY);
+ __sync_synchronize();
+
/* Link up the entry itself */
r = link_entry_into_array(f,
&f->header->entry_array_offset,
if (r < 0)
return r;
- log_error("=> %s seqnr=%lu n_entries=%lu", f->path, (unsigned long) o->entry.seqnum, (unsigned long) f->header->n_entries);
+ /* log_debug("=> %s seqnr=%lu n_entries=%lu", f->path, (unsigned long) o->entry.seqnum, (unsigned long) f->header->n_entries); */
if (f->header->head_entry_realtime == 0)
f->header->head_entry_realtime = o->entry.realtime;
ts->monotonic < le64toh(f->header->tail_entry_monotonic))
return -EINVAL;
- if (ts->realtime < le64toh(f->header->tail_entry_realtime))
- return -EINVAL;
-
items = alloca(sizeof(EntryItem) * n_iovec);
for (i = 0; i < n_iovec; i++) {
Object **ret, uint64_t *offset) {
Object *o;
- uint64_t p, a;
+ uint64_t p = 0, a;
int r;
assert(f);
(flags & O_ACCMODE) != O_RDWR)
return -EINVAL;
+ if (!endswith(fname, ".journal"))
+ return -EINVAL;
+
f = new0(JournalFile, 1);
if (!f)
return -ENOMEM;
f->writable = (flags & O_ACCMODE) != O_RDONLY;
f->prot = prot_from_flags(flags);
- f->metrics.max_size = DEFAULT_MAX_SIZE;
- f->metrics.min_size = DEFAULT_MIN_SIZE;
- f->metrics.keep_free = DEFAULT_KEEP_FREE;
+ if (template) {
+ f->metrics = template->metrics;
+ f->compress = template->compress;
+ }
f->path = strdup(fname);
if (!f->path) {
l = strlen(old_file->path);
- p = new(char, l + 1 + 16 + 1 + 32 + 1 + 16 + 1);
+ p = new(char, l + 1 + 32 + 1 + 16 + 1 + 16 + 1);
if (!p)
return -ENOMEM;
if (r < 0)
return -errno;
- old_file->header->state = le32toh(STATE_ARCHIVED);
+ old_file->header->state = STATE_ARCHIVED;
r = journal_file_open(old_file->path, old_file->flags, old_file->mode, old_file, &new_file);
journal_file_close(old_file);
return r;
}
+int journal_file_open_reliably(
+ const char *fname,
+ int flags,
+ mode_t mode,
+ JournalFile *template,
+ JournalFile **ret) {
+
+ int r;
+ size_t l;
+ char *p;
+
+ r = journal_file_open(fname, flags, mode, template, ret);
+ if (r != -EBADMSG && /* corrupted */
+ r != -ENODATA && /* truncated */
+ r != -EHOSTDOWN && /* other machine */
+ r != -EPROTONOSUPPORT) /* incompatible feature */
+ return r;
+
+ if ((flags & O_ACCMODE) == O_RDONLY)
+ return r;
+
+ if (!(flags & O_CREAT))
+ return r;
+
+ /* The file is corrupted. Rotate it away and try it again (but only once) */
+
+ l = strlen(fname);
+ if (asprintf(&p, "%.*s@%016llx-%016llx.journal~",
+ (int) (l-8), fname,
+ (unsigned long long) now(CLOCK_REALTIME),
+ random_ull()) < 0)
+ return -ENOMEM;
+
+ r = rename(fname, p);
+ free(p);
+ if (r < 0)
+ return -errno;
+
+ log_warning("File %s corrupted, renaming and replacing.", fname);
+
+ return journal_file_open(fname, flags, mode, template, ret);
+}
+
struct vacuum_info {
off_t usage;
char *filename;
uint64_t realtime;
sd_id128_t seqnum_id;
uint64_t seqnum;
+
+ bool have_seqnum;
};
static int vacuum_compare(const void *_a, const void *_b) {
a = _a;
b = _b;
- if (sd_id128_equal(a->seqnum_id, b->seqnum_id)) {
+ if (a->have_seqnum && b->have_seqnum &&
+ sd_id128_equal(a->seqnum_id, b->seqnum_id)) {
if (a->seqnum < b->seqnum)
return -1;
else if (a->seqnum > b->seqnum)
return -1;
else if (a->realtime > b->realtime)
return 1;
- else
+ else if (a->have_seqnum && b->have_seqnum)
return memcmp(&a->seqnum_id, &b->seqnum_id, 16);
+ else
+ return strcmp(a->filename, b->filename);
}
int journal_directory_vacuum(const char *directory, uint64_t max_use, uint64_t min_free) {
assert(directory);
if (max_use <= 0)
- max_use = DEFAULT_MAX_USE;
+ return 0;
d = opendir(directory);
if (!d)
char *p;
unsigned long long seqnum, realtime;
sd_id128_t seqnum_id;
+ bool have_seqnum;
k = readdir_r(d, &buf, &de);
if (k != 0) {
if (!de)
break;
- if (!dirent_is_file_with_suffix(de, ".journal"))
+ if (fstatat(dirfd(d), de->d_name, &st, AT_SYMLINK_NOFOLLOW) < 0)
+ continue;
+
+ if (!S_ISREG(st.st_mode))
continue;
q = strlen(de->d_name);
- if (q < 1 + 32 + 1 + 16 + 1 + 16 + 8)
- continue;
+ if (endswith(de->d_name, ".journal")) {
- if (de->d_name[q-8-16-1] != '-' ||
- de->d_name[q-8-16-1-16-1] != '-' ||
- de->d_name[q-8-16-1-16-1-32-1] != '@')
- continue;
+ /* Vacuum archived files */
- if (fstatat(dirfd(d), de->d_name, &st, AT_SYMLINK_NOFOLLOW) < 0)
- continue;
+ if (q < 1 + 32 + 1 + 16 + 1 + 16 + 8)
+ continue;
- if (!S_ISREG(st.st_mode))
- continue;
+ if (de->d_name[q-8-16-1] != '-' ||
+ de->d_name[q-8-16-1-16-1] != '-' ||
+ de->d_name[q-8-16-1-16-1-32-1] != '@')
+ continue;
- p = strdup(de->d_name);
- if (!p) {
- r = -ENOMEM;
- goto finish;
- }
+ p = strdup(de->d_name);
+ if (!p) {
+ r = -ENOMEM;
+ goto finish;
+ }
- de->d_name[q-8-16-1-16-1] = 0;
- if (sd_id128_from_string(de->d_name + q-8-16-1-16-1-32, &seqnum_id) < 0) {
- free(p);
- continue;
- }
+ de->d_name[q-8-16-1-16-1] = 0;
+ if (sd_id128_from_string(de->d_name + q-8-16-1-16-1-32, &seqnum_id) < 0) {
+ free(p);
+ continue;
+ }
+
+ if (sscanf(de->d_name + q-8-16-1-16, "%16llx-%16llx.journal", &seqnum, &realtime) != 2) {
+ free(p);
+ continue;
+ }
+
+ have_seqnum = true;
+
+ } else if (endswith(de->d_name, ".journal~")) {
+ unsigned long long tmp;
+
+ /* Vacuum corrupted files */
- if (sscanf(de->d_name + q-8-16-1-16, "%16llx-%16llx.journal", &seqnum, &realtime) != 2) {
- free(p);
+ if (q < 1 + 16 + 1 + 16 + 8 + 1)
+ continue;
+
+ if (de->d_name[q-1-8-16-1] != '-' ||
+ de->d_name[q-1-8-16-1-16-1] != '@')
+ continue;
+
+ p = strdup(de->d_name);
+ if (!p) {
+ r = -ENOMEM;
+ goto finish;
+ }
+
+ if (sscanf(de->d_name + q-1-8-16-1-16, "%16llx-%16llx.journal~", &realtime, &tmp) != 2) {
+ free(p);
+ continue;
+ }
+
+ have_seqnum = false;
+ } else
continue;
- }
if (n_list >= n_allocated) {
struct vacuum_info *j;
}
list[n_list].filename = p;
- list[n_list].usage = (uint64_t) st.st_blksize * (uint64_t) st.st_blocks;
+ list[n_list].usage = 512UL * (uint64_t) st.st_blocks;
list[n_list].seqnum = seqnum;
list[n_list].realtime = realtime;
list[n_list].seqnum_id = seqnum_id;
+ list[n_list].have_seqnum = have_seqnum;
sum += list[n_list].usage;
break;
if (unlinkat(dirfd(d), list[i].filename, 0) >= 0) {
- log_debug("Deleted archived journal %s/%s.", directory, list[i].filename);
+ log_info("Deleted archived journal %s/%s.", directory, list[i].filename);
sum -= list[i].usage;
} else if (errno != ENOENT)
log_warning("Failed to delete %s/%s: %m", directory, list[i].filename);
items = alloca(sizeof(EntryItem) * n);
for (i = 0; i < n; i++) {
- uint64_t le_hash, l, h;
+ uint64_t l, h;
+ le64_t le_hash;
size_t t;
void *data;
Object *u;
return journal_file_append_entry_internal(to, &ts, xor_hash, items, n, seqnum, ret, offset);
}
+
+void journal_default_metrics(JournalMetrics *m, int fd) {
+ uint64_t fs_size = 0;
+ struct statvfs ss;
+ char a[FORMAT_BYTES_MAX], b[FORMAT_BYTES_MAX], c[FORMAT_BYTES_MAX], d[FORMAT_BYTES_MAX];
+
+ assert(m);
+ assert(fd >= 0);
+
+ if (fstatvfs(fd, &ss) >= 0)
+ fs_size = ss.f_frsize * ss.f_blocks;
+
+ if (m->max_use == (uint64_t) -1) {
+
+ if (fs_size > 0) {
+ m->max_use = PAGE_ALIGN(fs_size / 10); /* 10% of file system size */
+
+ if (m->max_use > DEFAULT_MAX_USE_UPPER)
+ m->max_use = DEFAULT_MAX_USE_UPPER;
+
+ if (m->max_use < DEFAULT_MAX_USE_LOWER)
+ m->max_use = DEFAULT_MAX_USE_LOWER;
+ } else
+ m->max_use = DEFAULT_MAX_USE_LOWER;
+ } else {
+ m->max_use = PAGE_ALIGN(m->max_use);
+
+ if (m->max_use < JOURNAL_FILE_SIZE_MIN*2)
+ m->max_use = JOURNAL_FILE_SIZE_MIN*2;
+ }
+
+ if (m->max_size == (uint64_t) -1) {
+ m->max_size = PAGE_ALIGN(m->max_use / 8); /* 8 chunks */
+
+ if (m->max_size > DEFAULT_MAX_SIZE_UPPER)
+ m->max_size = DEFAULT_MAX_SIZE_UPPER;
+ } else
+ m->max_size = PAGE_ALIGN(m->max_size);
+
+ if (m->max_size < JOURNAL_FILE_SIZE_MIN)
+ m->max_size = JOURNAL_FILE_SIZE_MIN;
+
+ if (m->max_size*2 > m->max_use)
+ m->max_use = m->max_size*2;
+
+ if (m->min_size == (uint64_t) -1)
+ m->min_size = JOURNAL_FILE_SIZE_MIN;
+ else {
+ m->min_size = PAGE_ALIGN(m->min_size);
+
+ if (m->min_size < JOURNAL_FILE_SIZE_MIN)
+ m->min_size = JOURNAL_FILE_SIZE_MIN;
+
+ if (m->min_size > m->max_size)
+ m->max_size = m->min_size;
+ }
+
+ if (m->keep_free == (uint64_t) -1) {
+
+ if (fs_size > 0) {
+ m->keep_free = PAGE_ALIGN(fs_size / 20); /* 5% of file system size */
+
+ if (m->keep_free > DEFAULT_KEEP_FREE_UPPER)
+ m->keep_free = DEFAULT_KEEP_FREE_UPPER;
+
+ } else
+ m->keep_free = DEFAULT_KEEP_FREE;
+ }
+
+ log_info("Fixed max_use=%s max_size=%s min_size=%s keep_free=%s",
+ format_bytes(a, sizeof(a), m->max_use),
+ format_bytes(b, sizeof(b), m->max_size),
+ format_bytes(c, sizeof(c), m->min_size),
+ format_bytes(d, sizeof(d), m->keep_free));
+}
~ianmdlvl / elogind.git / blobdiff