along with systemd; If not, see <http://www.gnu.org/licenses/>.
***/
-#include <sys/mman.h>
#include <errno.h>
-#include <sys/uio.h>
-#include <unistd.h>
-#include <sys/statvfs.h>
#include <fcntl.h>
+#include <linux/fs.h>
#include <stddef.h>
-#include <sys/xattr.h>
+#include <sys/mman.h>
+#include <sys/statvfs.h>
+#include <sys/uio.h>
+#include <unistd.h>
+#include "alloc-util.h"
+#include "btrfs-util.h"
+#include "chattr-util.h"
+#include "compress.h"
+#include "fd-util.h"
+#include "journal-authenticate.h"
#include "journal-def.h"
#include "journal-file.h"
-#include "journal-authenticate.h"
#include "lookup3.h"
-#include "compress.h"
-#include "fsprg.h"
+#include "parse-util.h"
+#include "random-util.h"
+#include "string-util.h"
+#include "xattr-util.h"
#define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*sizeof(HashItem))
#define DEFAULT_FIELD_HASH_TABLE_SIZE (333ULL*sizeof(HashItem))
#define DEFAULT_MAX_USE_LOWER (1ULL*1024ULL*1024ULL) /* 1 MiB */
#define DEFAULT_MAX_USE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
+/* This is the default minimal use limit, how much we'll use even if keep_free suggests otherwise. */
+#define DEFAULT_MIN_USE (1ULL*1024ULL*1024ULL) /* 1 MiB */
+
/* This is the upper bound if we deduce max_size from max_use */
#define DEFAULT_MAX_SIZE_UPPER (128ULL*1024ULL*1024ULL) /* 128 MiB */
* size */
#define DEFAULT_KEEP_FREE (1024ULL*1024ULL) /* 1 MB */
+/* This is the default maximum number of journal files to keep around. */
+#define DEFAULT_N_MAX_FILES (100)
+
/* n_data was the first entry we added after the initial file format design */
#define HEADER_SIZE_MIN ALIGN64(offsetof(Header, n_data))
/* How much to increase the journal file size at once each time we allocate something new. */
#define FILE_SIZE_INCREASE (8ULL*1024ULL*1024ULL) /* 8MB */
+/* Reread fstat() of the file for detecting deletions at least this often */
+#define LAST_STAT_REFRESH_USEC (5*USEC_PER_SEC)
+
+/* The mmap context to use for the header we pick as one above the last defined typed */
+#define CONTEXT_HEADER _OBJECT_TYPE_MAX
+
static int journal_file_set_online(JournalFile *f) {
assert(f);
if (!(f->fd >= 0 && f->header))
return -EINVAL;
+ if (mmap_cache_got_sigbus(f->mmap, f->fd))
+ return -EIO;
+
switch(f->header->state) {
case STATE_ONLINE:
return 0;
fsync(f->fd);
+ if (mmap_cache_got_sigbus(f->mmap, f->fd))
+ return -EIO;
+
f->header->state = STATE_OFFLINE;
+ if (mmap_cache_got_sigbus(f->mmap, f->fd))
+ return -EIO;
+
fsync(f->fd);
return 0;
}
-void journal_file_close(JournalFile *f) {
+JournalFile* journal_file_close(JournalFile *f) {
assert(f);
#ifdef HAVE_GCRYPT
journal_file_append_tag(f);
#endif
- /* Sync everything to disk, before we mark the file offline */
+ journal_file_set_offline(f);
+
if (f->mmap && f->fd >= 0)
mmap_cache_close_fd(f->mmap, f->fd);
- journal_file_set_offline(f);
+ if (f->fd >= 0 && f->defrag_on_close) {
+
+ /* Be friendly to btrfs: turn COW back on again now,
+ * and defragment the file. We won't write to the file
+ * ever again, hence remove all fragmentation, and
+ * reenable all the good bits COW usually provides
+ * (such as data checksumming). */
- if (f->header)
- munmap(f->header, PAGE_ALIGN(sizeof(Header)));
+ (void) chattr_fd(f->fd, 0, FS_NOCOW_FL);
+ (void) btrfs_defrag_fd(f->fd);
+ }
safe_close(f->fd);
free(f->path);
if (f->mmap)
mmap_cache_unref(f->mmap);
- hashmap_free_free(f->chain_cache);
+ ordered_hashmap_free_free(f->chain_cache);
-#ifdef HAVE_XZ
+#if defined(HAVE_XZ) || defined(HAVE_LZ4)
free(f->compress_buffer);
#endif
#ifdef HAVE_GCRYPT
if (f->fss_file)
munmap(f->fss_file, PAGE_ALIGN(f->fss_file_size));
- else if (f->fsprg_state)
+ else
free(f->fsprg_state);
free(f->fsprg_seed);
#endif
free(f);
+ return NULL;
}
static int journal_file_init_header(JournalFile *f, JournalFile *template) {
- Header h;
+ Header h = {};
ssize_t k;
int r;
assert(f);
- zero(h);
memcpy(h.signature, HEADER_SIGNATURE, 8);
h.header_size = htole64(ALIGN64(sizeof(h)));
- h.incompatible_flags =
- htole32(f->compress ? HEADER_INCOMPATIBLE_COMPRESSED : 0);
+ h.incompatible_flags |= htole32(
+ f->compress_xz * HEADER_INCOMPATIBLE_COMPRESSED_XZ |
+ f->compress_lz4 * HEADER_INCOMPATIBLE_COMPRESSED_LZ4);
- h.compatible_flags =
- htole32(f->seal ? HEADER_COMPATIBLE_SEALED : 0);
+ h.compatible_flags = htole32(
+ f->seal * HEADER_COMPATIBLE_SEALED);
r = sd_id128_randomize(&h.file_id);
if (r < 0)
}
static int journal_file_refresh_header(JournalFile *f) {
- int r;
sd_id128_t boot_id;
+ int r;
assert(f);
f->header->boot_id = boot_id;
- journal_file_set_online(f);
+ r = journal_file_set_online(f);
/* Sync the online state to disk */
fsync(f->fd);
- return 0;
+ return r;
}
static int journal_file_verify_header(JournalFile *f) {
+ uint32_t flags;
+
assert(f);
if (memcmp(f->header->signature, HEADER_SIGNATURE, 8))
/* In both read and write mode we refuse to open files with
* incompatible flags we don't know */
-#ifdef HAVE_XZ
- if ((le32toh(f->header->incompatible_flags) & ~HEADER_INCOMPATIBLE_COMPRESSED) != 0)
- return -EPROTONOSUPPORT;
-#else
- if (f->header->incompatible_flags != 0)
+ flags = le32toh(f->header->incompatible_flags);
+ if (flags & ~HEADER_INCOMPATIBLE_SUPPORTED) {
+ if (flags & ~HEADER_INCOMPATIBLE_ANY)
+ log_debug("Journal file %s has unknown incompatible flags %"PRIx32,
+ f->path, flags & ~HEADER_INCOMPATIBLE_ANY);
+ flags = (flags & HEADER_INCOMPATIBLE_ANY) & ~HEADER_INCOMPATIBLE_SUPPORTED;
+ if (flags)
+ log_debug("Journal file %s uses incompatible flags %"PRIx32
+ " disabled at compilation time.", f->path, flags);
return -EPROTONOSUPPORT;
-#endif
+ }
/* When open for writing we refuse to open files with
* compatible flags, too */
- if (f->writable) {
-#ifdef HAVE_GCRYPT
- if ((le32toh(f->header->compatible_flags) & ~HEADER_COMPATIBLE_SEALED) != 0)
- return -EPROTONOSUPPORT;
-#else
- if (f->header->compatible_flags != 0)
- return -EPROTONOSUPPORT;
-#endif
+ flags = le32toh(f->header->compatible_flags);
+ if (f->writable && (flags & ~HEADER_COMPATIBLE_SUPPORTED)) {
+ if (flags & ~HEADER_COMPATIBLE_ANY)
+ log_debug("Journal file %s has unknown compatible flags %"PRIx32,
+ f->path, flags & ~HEADER_COMPATIBLE_ANY);
+ flags = (flags & HEADER_COMPATIBLE_ANY) & ~HEADER_COMPATIBLE_SUPPORTED;
+ if (flags)
+ log_debug("Journal file %s uses compatible flags %"PRIx32
+ " disabled at compilation time.", f->path, flags);
+ return -EPROTONOSUPPORT;
}
if (f->header->state >= _STATE_MAX)
!VALID64(le64toh(f->header->entry_array_offset)))
return -ENODATA;
- if (le64toh(f->header->data_hash_table_offset) < le64toh(f->header->header_size) ||
- le64toh(f->header->field_hash_table_offset) < le64toh(f->header->header_size) ||
- le64toh(f->header->tail_object_offset) < le64toh(f->header->header_size) ||
- le64toh(f->header->entry_array_offset) < le64toh(f->header->header_size))
- return -ENODATA;
-
if (f->writable) {
uint8_t state;
sd_id128_t machine_id;
} else if (state == STATE_ARCHIVED)
return -ESHUTDOWN;
else if (state != STATE_OFFLINE) {
- log_debug("Journal file %s has unknown state %u.", f->path, state);
+ log_debug("Journal file %s has unknown state %i.", f->path, state);
return -EBUSY;
}
}
- f->compress = JOURNAL_HEADER_COMPRESSED(f->header);
+ f->compress_xz = JOURNAL_HEADER_COMPRESSED_XZ(f->header);
+ f->compress_lz4 = JOURNAL_HEADER_COMPRESSED_LZ4(f->header);
f->seal = JOURNAL_HEADER_SEALED(f->header);
return 0;
}
+static int journal_file_fstat(JournalFile *f) {
+ assert(f);
+ assert(f->fd >= 0);
+
+ if (fstat(f->fd, &f->last_stat) < 0)
+ return -errno;
+
+ f->last_stat_usec = now(CLOCK_MONOTONIC);
+
+ /* Refuse appending to files that are already deleted */
+ if (f->last_stat.st_nlink <= 0)
+ return -EIDRM;
+
+ return 0;
+}
+
static int journal_file_allocate(JournalFile *f, uint64_t offset, uint64_t size) {
uint64_t old_size, new_size;
int r;
* for sure, since we always call posix_fallocate()
* ourselves */
+ if (mmap_cache_got_sigbus(f->mmap, f->fd))
+ return -EIO;
+
old_size =
le64toh(f->header->header_size) +
le64toh(f->header->arena_size);
if (new_size < le64toh(f->header->header_size))
new_size = le64toh(f->header->header_size);
- if (new_size <= old_size)
- return 0;
+ if (new_size <= old_size) {
+
+ /* We already pre-allocated enough space, but before
+ * we write to it, let's check with fstat() if the
+ * file got deleted, in order make sure we don't throw
+ * away the data immediately. Don't check fstat() for
+ * all writes though, but only once ever 10s. */
+
+ if (f->last_stat_usec + LAST_STAT_REFRESH_USEC > now(CLOCK_MONOTONIC))
+ return 0;
+
+ return journal_file_fstat(f);
+ }
+
+ /* Allocate more space. */
if (f->metrics.max_size > 0 && new_size > f->metrics.max_size)
return -E2BIG;
if (fstatvfs(f->fd, &svfs) >= 0) {
uint64_t available;
- available = svfs.f_bfree * svfs.f_bsize;
-
- if (available >= f->metrics.keep_free)
- available -= f->metrics.keep_free;
- else
- available = 0;
+ available = LESS_BY((uint64_t) svfs.f_bfree * (uint64_t) svfs.f_bsize, f->metrics.keep_free);
if (new_size - old_size > available)
return -E2BIG;
if (r != 0)
return -r;
- if (fstat(f->fd, &f->last_stat) < 0)
- return -errno;
-
f->header->arena_size = htole64(new_size - le64toh(f->header->header_size));
- return 0;
+ return journal_file_fstat(f);
+}
+
+static unsigned type_to_context(ObjectType type) {
+ /* One context for each type, plus one catch-all for the rest */
+ assert_cc(_OBJECT_TYPE_MAX <= MMAP_CACHE_MAX_CONTEXTS);
+ assert_cc(CONTEXT_HEADER < MMAP_CACHE_MAX_CONTEXTS);
+ return type > OBJECT_UNUSED && type < _OBJECT_TYPE_MAX ? type : 0;
}
-static int journal_file_move_to(JournalFile *f, int context, bool keep_always, uint64_t offset, uint64_t size, void **ret) {
+static int journal_file_move_to(JournalFile *f, ObjectType type, bool keep_always, uint64_t offset, uint64_t size, void **ret) {
+ int r;
+
assert(f);
assert(ret);
/* 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)
+ r = journal_file_fstat(f);
+ if (r < 0)
+ return r;
+
+ if (offset + size > (uint64_t) f->last_stat.st_size)
return -EADDRNOTAVAIL;
}
- return mmap_cache_get(f->mmap, f->fd, f->prot, context, keep_always, offset, size, &f->last_stat, ret);
+ return mmap_cache_get(f->mmap, f->fd, f->prot, type_to_context(type), keep_always, offset, size, &f->last_stat, ret);
}
static uint64_t minimum_header_size(Object *o) {
return table[o->object.type];
}
-int journal_file_move_to_object(JournalFile *f, int type, uint64_t offset, Object **ret) {
+int journal_file_move_to_object(JournalFile *f, ObjectType type, uint64_t offset, Object **ret) {
int r;
void *t;
Object *o;
if (!VALID64(offset))
return -EFAULT;
-
- r = journal_file_move_to(f, type_to_context(type), false, offset, sizeof(ObjectHeader), &t);
+ r = journal_file_move_to(f, type, false, offset, sizeof(ObjectHeader), &t);
if (r < 0)
return r;
if (s < minimum_header_size(o))
return -EBADMSG;
- if (type > 0 && o->object.type != type)
+ if (type > OBJECT_UNUSED && o->object.type != type)
return -EBADMSG;
if (s > sizeof(ObjectHeader)) {
- r = journal_file_move_to(f, o->object.type, false, offset, s, &t);
+ r = journal_file_move_to(f, type, false, offset, s, &t);
if (r < 0)
return r;
return r;
}
-int journal_file_append_object(JournalFile *f, int type, uint64_t size, Object **ret, uint64_t *offset) {
+int journal_file_append_object(JournalFile *f, ObjectType type, uint64_t size, Object **ret, uint64_t *offset) {
int r;
uint64_t p;
Object *tail, *o;
void *t;
assert(f);
- assert(type > 0 && type < _OBJECT_TYPE_MAX);
+ assert(type > OBJECT_UNUSED && type < _OBJECT_TYPE_MAX);
assert(size >= sizeof(ObjectHeader));
assert(offset);
assert(ret);
if (p == 0)
p = le64toh(f->header->header_size);
else {
- r = journal_file_move_to_object(f, -1, p, &tail);
+ r = journal_file_move_to_object(f, OBJECT_UNUSED, p, &tail);
if (r < 0)
return r;
assert(f);
- /* 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. */
+ /* We estimate that we need 1 hash table entry per 768 bytes
+ 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)
return 0;
}
-static int journal_file_map_data_hash_table(JournalFile *f) {
+int journal_file_map_data_hash_table(JournalFile *f) {
uint64_t s, p;
void *t;
int r;
assert(f);
+ if (f->data_hash_table)
+ return 0;
+
p = le64toh(f->header->data_hash_table_offset);
s = le64toh(f->header->data_hash_table_size);
return 0;
}
-static int journal_file_map_field_hash_table(JournalFile *f) {
+int journal_file_map_field_hash_table(JournalFile *f) {
uint64_t s, p;
void *t;
int r;
assert(f);
+ if (f->field_hash_table)
+ return 0;
+
p = le64toh(f->header->field_hash_table_offset);
s = le64toh(f->header->field_hash_table_size);
uint64_t offset,
uint64_t hash) {
- uint64_t p, h;
+ uint64_t p, h, m;
int r;
assert(f);
if (o->object.type != OBJECT_FIELD)
return -EINVAL;
- /* This might alter the window we are looking at */
+ m = le64toh(f->header->field_hash_table_size) / sizeof(HashItem);
+ if (m <= 0)
+ return -EBADMSG;
+ /* 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));
+ h = hash % m;
p = le64toh(f->field_hash_table[h].tail_hash_offset);
if (p == 0)
f->field_hash_table[h].head_hash_offset = htole64(offset);
uint64_t offset,
uint64_t hash) {
- uint64_t p, h;
+ uint64_t p, h, m;
int r;
assert(f);
if (o->object.type != OBJECT_DATA)
return -EINVAL;
- /* This might alter the window we are looking at */
+ m = le64toh(f->header->data_hash_table_size) / sizeof(HashItem);
+ if (m <= 0)
+ return -EBADMSG;
+ /* 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;
- h = hash % (le64toh(f->header->data_hash_table_size) / sizeof(HashItem));
+ h = hash % m;
p = le64toh(f->data_hash_table[h].tail_hash_offset);
if (p == 0)
/* Only entry in the hash table is easy */
const void *field, uint64_t size, uint64_t hash,
Object **ret, uint64_t *offset) {
- uint64_t p, osize, h;
+ uint64_t p, osize, h, m;
int r;
assert(f);
assert(field && size > 0);
+ /* If the field hash table is empty, we can't find anything */
+ if (le64toh(f->header->field_hash_table_size) <= 0)
+ return 0;
+
+ /* Map the field hash table, if it isn't mapped yet. */
+ r = journal_file_map_field_hash_table(f);
+ if (r < 0)
+ return r;
+
osize = offsetof(Object, field.payload) + size;
- if (f->header->field_hash_table_size == 0)
+ m = le64toh(f->header->field_hash_table_size) / sizeof(HashItem);
+ if (m <= 0)
return -EBADMSG;
- h = hash % (le64toh(f->header->field_hash_table_size) / sizeof(HashItem));
+ h = hash % m;
p = le64toh(f->field_hash_table[h].head_hash_offset);
while (p > 0) {
const void *data, uint64_t size, uint64_t hash,
Object **ret, uint64_t *offset) {
- uint64_t p, osize, h;
+ uint64_t p, osize, h, m;
int r;
assert(f);
assert(data || size == 0);
+ /* If there's no data hash table, then there's no entry. */
+ if (le64toh(f->header->data_hash_table_size) <= 0)
+ return 0;
+
+ /* Map the data hash table, if it isn't mapped yet. */
+ r = journal_file_map_data_hash_table(f);
+ if (r < 0)
+ return r;
+
osize = offsetof(Object, data.payload) + size;
- if (f->header->data_hash_table_size == 0)
+ m = le64toh(f->header->data_hash_table_size) / sizeof(HashItem);
+ if (m <= 0)
return -EBADMSG;
- h = hash % (le64toh(f->header->data_hash_table_size) / sizeof(HashItem));
+ h = hash % m;
p = le64toh(f->data_hash_table[h].head_hash_offset);
while (p > 0) {
if (le64toh(o->data.hash) != hash)
goto next;
- if (o->object.flags & OBJECT_COMPRESSED) {
-#ifdef HAVE_XZ
- uint64_t l, rsize;
+ if (o->object.flags & OBJECT_COMPRESSION_MASK) {
+#if defined(HAVE_XZ) || defined(HAVE_LZ4)
+ uint64_t l;
+ size_t rsize = 0;
l = le64toh(o->object.size);
if (l <= offsetof(Object, data.payload))
l -= offsetof(Object, data.payload);
- if (!uncompress_blob(o->data.payload, l, &f->compress_buffer, &f->compress_buffer_size, &rsize, 0))
- return -EBADMSG;
+ r = decompress_blob(o->object.flags & OBJECT_COMPRESSION_MASK,
+ o->data.payload, l, &f->compress_buffer, &f->compress_buffer_size, &rsize, 0);
+ if (r < 0)
+ return r;
if (rsize == size &&
memcmp(f->compress_buffer, data, size) == 0) {
#else
return -EPROTONOSUPPORT;
#endif
-
} else if (le64toh(o->object.size) == osize &&
memcmp(o->data.payload, data, size) == 0) {
uint64_t hash, p;
uint64_t osize;
Object *o;
- int r;
- bool compressed = false;
+ int r, compression = 0;
const void *eq;
assert(f);
r = journal_file_find_data_object_with_hash(f, data, size, hash, &o, &p);
if (r < 0)
return r;
- else if (r > 0) {
+ if (r > 0) {
if (ret)
*ret = o;
o->data.hash = htole64(hash);
-#ifdef HAVE_XZ
- if (f->compress &&
- size >= COMPRESSION_SIZE_THRESHOLD) {
- uint64_t rsize;
+#if defined(HAVE_XZ) || defined(HAVE_LZ4)
+ if (JOURNAL_FILE_COMPRESS(f) && size >= COMPRESSION_SIZE_THRESHOLD) {
+ size_t rsize = 0;
- compressed = compress_blob(data, size, o->data.payload, &rsize);
+ compression = compress_blob(data, size, o->data.payload, &rsize);
- if (compressed) {
+ if (compression >= 0) {
o->object.size = htole64(offsetof(Object, data.payload) + rsize);
- o->object.flags |= OBJECT_COMPRESSED;
+ o->object.flags |= compression;
- log_debug("Compressed data object %"PRIu64" -> %"PRIu64, size, rsize);
- }
+ log_debug("Compressed data object %"PRIu64" -> %zu using %s",
+ size, rsize, object_compressed_to_string(compression));
+ } else
+ /* Compression didn't work, we don't really care why, let's continue without compression */
+ compression = 0;
}
#endif
- if (!compressed && size > 0)
+ if (compression == 0 && size > 0)
memcpy(o->data.payload, data, size);
r = journal_file_link_data(f, o, p, hash);
__sync_synchronize();
if (ftruncate(f->fd, f->last_stat.st_size) < 0)
- log_error("Failed to truncate file to its own size: %m");
+ log_error_errno(errno, "Failed to truncate file to its own size: %m");
}
static int entry_item_cmp(const void *_a, const void *_b) {
r = journal_file_append_entry_internal(f, ts, xor_hash, items, n_iovec, seqnum, ret, offset);
+ /* If the memory mapping triggered a SIGBUS then we return an
+ * IO error and ignore the error code passed down to us, since
+ * it is very likely just an effect of a nullified replacement
+ * mapping page */
+
+ if (mmap_cache_got_sigbus(f->mmap, f->fd))
+ r = -EIO;
+
journal_file_post_change(f);
return r;
} ChainCacheItem;
static void chain_cache_put(
- Hashmap *h,
+ OrderedHashmap *h,
ChainCacheItem *ci,
uint64_t first,
uint64_t array,
if (array == first)
return;
- if (hashmap_size(h) >= CHAIN_CACHE_MAX)
- ci = hashmap_steal_first(h);
- else {
+ if (ordered_hashmap_size(h) >= CHAIN_CACHE_MAX) {
+ ci = ordered_hashmap_steal_first(h);
+ assert(ci);
+ } else {
ci = new(ChainCacheItem, 1);
if (!ci)
return;
ci->first = first;
- if (hashmap_put(h, &ci->first, ci) < 0) {
+ if (ordered_hashmap_put(h, &ci->first, ci) < 0) {
free(ci);
return;
}
a = first;
/* Try the chain cache first */
- ci = hashmap_get(f->chain_cache, &first);
+ ci = ordered_hashmap_get(f->chain_cache, &first);
if (ci && i > ci->total) {
a = ci->array;
i -= ci->total;
/* Start with the first array in the chain */
a = first;
- ci = hashmap_get(f->chain_cache, &first);
+ ci = ordered_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
}
}
- if (k > n) {
+ if (k >= n) {
if (direction == DIRECTION_UP) {
i = n;
subtract_one = true;
return 1;
}
-
static int generic_array_bisect_plus_one(
JournalFile *f,
uint64_t extra,
return TEST_RIGHT;
}
-int journal_file_move_to_entry_by_offset(
- JournalFile *f,
- uint64_t p,
- direction_t direction,
- Object **ret,
- uint64_t *offset) {
-
- return generic_array_bisect(f,
- le64toh(f->header->entry_array_offset),
- le64toh(f->header->n_entries),
- p,
- test_object_offset,
- direction,
- ret, offset, NULL);
-}
-
-
static int test_object_seqnum(JournalFile *f, uint64_t p, uint64_t needle) {
Object *o;
int r;
return TEST_RIGHT;
}
-static inline int find_data_object_by_boot_id(
+static int find_data_object_by_boot_id(
JournalFile *f,
sd_id128_t boot_id,
Object **o,
uint64_t *b) {
+
char t[sizeof("_BOOT_ID=")-1 + 32 + 1] = "_BOOT_ID=";
sd_id128_to_string(boot_id, t + 9);
ret, offset, NULL);
}
+void journal_file_reset_location(JournalFile *f) {
+ f->location_type = LOCATION_HEAD;
+ f->current_offset = 0;
+ f->current_seqnum = 0;
+ f->current_realtime = 0;
+ f->current_monotonic = 0;
+ zero(f->current_boot_id);
+ f->current_xor_hash = 0;
+}
+
+void journal_file_save_location(JournalFile *f, Object *o, uint64_t offset) {
+ f->location_type = LOCATION_SEEK;
+ f->current_offset = offset;
+ f->current_seqnum = le64toh(o->entry.seqnum);
+ f->current_realtime = le64toh(o->entry.realtime);
+ f->current_monotonic = le64toh(o->entry.monotonic);
+ f->current_boot_id = o->entry.boot_id;
+ f->current_xor_hash = le64toh(o->entry.xor_hash);
+}
+
+int journal_file_compare_locations(JournalFile *af, JournalFile *bf) {
+ assert(af);
+ assert(bf);
+ assert(af->location_type == LOCATION_SEEK);
+ assert(bf->location_type == LOCATION_SEEK);
+
+ /* If contents and timestamps match, these entries are
+ * identical, even if the seqnum does not match */
+ if (sd_id128_equal(af->current_boot_id, bf->current_boot_id) &&
+ af->current_monotonic == bf->current_monotonic &&
+ af->current_realtime == bf->current_realtime &&
+ af->current_xor_hash == bf->current_xor_hash)
+ return 0;
+
+ if (sd_id128_equal(af->header->seqnum_id, bf->header->seqnum_id)) {
+
+ /* If this is from the same seqnum source, compare
+ * seqnums */
+ if (af->current_seqnum < bf->current_seqnum)
+ return -1;
+ if (af->current_seqnum > bf->current_seqnum)
+ return 1;
+
+ /* Wow! This is weird, different data but the same
+ * seqnums? Something is borked, but let's make the
+ * best of it and compare by time. */
+ }
+
+ if (sd_id128_equal(af->current_boot_id, bf->current_boot_id)) {
+
+ /* If the boot id matches, compare monotonic time */
+ if (af->current_monotonic < bf->current_monotonic)
+ return -1;
+ if (af->current_monotonic > bf->current_monotonic)
+ return 1;
+ }
+
+ /* Otherwise, compare UTC time */
+ if (af->current_realtime < bf->current_realtime)
+ return -1;
+ if (af->current_realtime > bf->current_realtime)
+ return 1;
+
+ /* Finally, compare by contents */
+ if (af->current_xor_hash < bf->current_xor_hash)
+ return -1;
+ if (af->current_xor_hash > bf->current_xor_hash)
+ return 1;
+
+ return 0;
+}
+
int journal_file_next_entry(
JournalFile *f,
- Object *o, uint64_t p,
+ uint64_t p,
direction_t direction,
Object **ret, uint64_t *offset) {
int r;
assert(f);
- assert(p > 0 || !o);
n = le64toh(f->header->n_entries);
if (n <= 0)
return 0;
- if (!o)
+ if (p == 0)
i = direction == DIRECTION_DOWN ? 0 : n - 1;
else {
- if (o->object.type != OBJECT_ENTRY)
- return -EINVAL;
-
r = generic_array_bisect(f,
le64toh(f->header->entry_array_offset),
le64toh(f->header->n_entries),
return 1;
}
-int journal_file_skip_entry(
- JournalFile *f,
- Object *o, uint64_t p,
- int64_t skip,
- Object **ret, uint64_t *offset) {
-
- uint64_t i, n;
- int r;
-
- assert(f);
- assert(o);
- assert(p > 0);
-
- if (o->object.type != OBJECT_ENTRY)
- return -EINVAL;
-
- r = generic_array_bisect(f,
- le64toh(f->header->entry_array_offset),
- le64toh(f->header->n_entries),
- p,
- test_object_offset,
- DIRECTION_DOWN,
- NULL, NULL,
- &i);
- if (r <= 0)
- return r;
-
- /* Calculate new index */
- if (skip < 0) {
- if ((uint64_t) -skip >= i)
- i = 0;
- else
- i = i - (uint64_t) -skip;
- } else
- i += (uint64_t) skip;
-
- n = le64toh(f->header->n_entries);
- if (n <= 0)
- return -EBADMSG;
-
- if (i >= n)
- i = n-1;
-
- return generic_array_get(f,
- le64toh(f->header->entry_array_offset),
- i,
- ret, offset);
-}
-
int journal_file_next_entry_for_data(
JournalFile *f,
Object *o, uint64_t p,
p = le64toh(f->header->header_size);
while (p != 0) {
- r = journal_file_move_to_object(f, -1, p, &o);
+ r = journal_file_move_to_object(f, OBJECT_UNUSED, p, &o);
if (r < 0)
goto fail;
break;
default:
- printf("Type: unknown (%u)\n", o->object.type);
+ printf("Type: unknown (%i)\n", o->object.type);
break;
}
- if (o->object.flags & OBJECT_COMPRESSED)
- printf("Flags: COMPRESSED\n");
+ if (o->object.flags & OBJECT_COMPRESSION_MASK)
+ printf("Flags: %s\n",
+ object_compressed_to_string(o->object.flags & OBJECT_COMPRESSION_MASK));
if (p == le64toh(f->header->tail_object_offset))
p = 0;
"Sequential Number ID: %s\n"
"State: %s\n"
"Compatible Flags:%s%s\n"
- "Incompatible Flags:%s%s\n"
+ "Incompatible Flags:%s%s%s\n"
"Header size: %"PRIu64"\n"
"Arena size: %"PRIu64"\n"
"Data Hash Table Size: %"PRIu64"\n"
f->header->state == STATE_ONLINE ? "ONLINE" :
f->header->state == STATE_ARCHIVED ? "ARCHIVED" : "UNKNOWN",
JOURNAL_HEADER_SEALED(f->header) ? " SEALED" : "",
- (le32toh(f->header->compatible_flags) & ~HEADER_COMPATIBLE_SEALED) ? " ???" : "",
- JOURNAL_HEADER_COMPRESSED(f->header) ? " COMPRESSED" : "",
- (le32toh(f->header->incompatible_flags) & ~HEADER_INCOMPATIBLE_COMPRESSED) ? " ???" : "",
+ (le32toh(f->header->compatible_flags) & ~HEADER_COMPATIBLE_ANY) ? " ???" : "",
+ JOURNAL_HEADER_COMPRESSED_XZ(f->header) ? " COMPRESSED-XZ" : "",
+ JOURNAL_HEADER_COMPRESSED_LZ4(f->header) ? " COMPRESSED-LZ4" : "",
+ (le32toh(f->header->incompatible_flags) & ~HEADER_INCOMPATIBLE_ANY) ? " ???" : "",
le64toh(f->header->header_size),
le64toh(f->header->arena_size),
le64toh(f->header->data_hash_table_size) / sizeof(HashItem),
le64toh(f->header->n_entry_arrays));
if (fstat(f->fd, &st) >= 0)
- printf("Disk usage: %s\n", format_bytes(bytes, sizeof(bytes), (off_t) st.st_blocks * 512ULL));
+ printf("Disk usage: %s\n", format_bytes(bytes, sizeof(bytes), (uint64_t) st.st_blocks * 512ULL));
+}
+
+static int journal_file_warn_btrfs(JournalFile *f) {
+ unsigned attrs;
+ int r;
+
+ assert(f);
+
+ /* Before we write anything, check if the COW logic is turned
+ * off on btrfs. Given our write pattern that is quite
+ * unfriendly to COW file systems this should greatly improve
+ * performance on COW file systems, such as btrfs, at the
+ * expense of data integrity features (which shouldn't be too
+ * bad, given that we do our own checksumming). */
+
+ r = btrfs_is_filesystem(f->fd);
+ if (r < 0)
+ return log_warning_errno(r, "Failed to determine if journal is on btrfs: %m");
+ if (!r)
+ return 0;
+
+ r = read_attr_fd(f->fd, &attrs);
+ if (r < 0)
+ return log_warning_errno(r, "Failed to read file attributes: %m");
+
+ if (attrs & FS_NOCOW_FL) {
+ log_debug("Detected btrfs file system with copy-on-write disabled, all is good.");
+ return 0;
+ }
+
+ log_notice("Creating journal file %s on a btrfs file system, and copy-on-write is enabled. "
+ "This is likely to slow down journal access substantially, please consider turning "
+ "off the copy-on-write file attribute on the journal directory, using chattr +C.", f->path);
+
+ return 1;
}
int journal_file_open(
JournalFile *template,
JournalFile **ret) {
+ bool newly_created = false;
JournalFile *f;
+ void *h;
int r;
- bool newly_created = false;
assert(fname);
assert(ret);
f->flags = flags;
f->prot = prot_from_flags(flags);
f->writable = (flags & O_ACCMODE) != O_RDONLY;
-#ifdef HAVE_XZ
- f->compress = compress;
+#if defined(HAVE_LZ4)
+ f->compress_lz4 = compress;
+#elif defined(HAVE_XZ)
+ f->compress_xz = compress;
#endif
#ifdef HAVE_GCRYPT
f->seal = seal;
goto fail;
}
- f->chain_cache = hashmap_new(uint64_hash_func, uint64_compare_func);
+ f->chain_cache = ordered_hashmap_new(&uint64_hash_ops);
if (!f->chain_cache) {
r = -ENOMEM;
goto fail;
goto fail;
}
- if (fstat(f->fd, &f->last_stat) < 0) {
- r = -errno;
+ r = journal_file_fstat(f);
+ if (r < 0)
goto fail;
- }
if (f->last_stat.st_size == 0 && f->writable) {
- uint64_t crtime;
+
+ (void) journal_file_warn_btrfs(f);
/* Let's attach the creation time to the journal file,
* so that the vacuuming code knows the age of this
* 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.*/
+ * 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);
+ fd_setcrtime(f->fd, 0);
#ifdef HAVE_GCRYPT
/* Try to load the FSPRG state, and if we can't, then
if (r < 0)
goto fail;
- if (fstat(f->fd, &f->last_stat) < 0) {
- r = -errno;
+ r = journal_file_fstat(f);
+ if (r < 0)
goto fail;
- }
newly_created = true;
}
if (f->last_stat.st_size < (off_t) HEADER_SIZE_MIN) {
- r = -EIO;
+ r = -ENODATA;
goto fail;
}
- f->header = mmap(NULL, PAGE_ALIGN(sizeof(Header)), prot_from_flags(flags), MAP_SHARED, f->fd, 0);
- if (f->header == MAP_FAILED) {
- f->header = NULL;
- r = -errno;
+ r = mmap_cache_get(f->mmap, f->fd, f->prot, CONTEXT_HEADER, true, 0, PAGE_ALIGN(sizeof(Header)), &f->last_stat, &h);
+ if (r < 0)
goto fail;
- }
+
+ f->header = h;
if (!newly_created) {
r = journal_file_verify_header(f);
#endif
}
- r = journal_file_map_field_hash_table(f);
- if (r < 0)
- goto fail;
-
- r = journal_file_map_data_hash_table(f);
- if (r < 0)
+ if (mmap_cache_got_sigbus(f->mmap, f->fd)) {
+ r = -EIO;
goto fail;
+ }
*ret = f;
return 0;
fail:
+ if (f->fd >= 0 && mmap_cache_got_sigbus(f->mmap, f->fd))
+ r = -EIO;
+
journal_file_close(f);
return r;
if (r < 0)
return -ENOMEM;
+ /* Try to rename the file to the archived version. If the file
+ * already was deleted, we'll get ENOENT, let's ignore that
+ * case. */
r = rename(old_file->path, p);
- if (r < 0)
+ if (r < 0 && errno != ENOENT)
return -errno;
old_file->header->state = STATE_ARCHIVED;
+ /* Currently, btrfs is not very good with out write patterns
+ * and fragments heavily. Let's defrag our journal files when
+ * we archive them */
+ old_file->defrag_on_close = true;
+
r = journal_file_open(old_file->path, old_file->flags, old_file->mode, compress, seal, NULL, old_file->mmap, old_file, &new_file);
journal_file_close(old_file);
size_t l;
_cleanup_free_ char *p = NULL;
- r = journal_file_open(fname, flags, mode, compress, seal,
- metrics, mmap_cache, template, ret);
- if (r != -EBADMSG && /* corrupted */
- r != -ENODATA && /* truncated */
- r != -EHOSTDOWN && /* other machine */
- r != -EPROTONOSUPPORT && /* incompatible feature */
- r != -EBUSY && /* unclean shutdown */
- r != -ESHUTDOWN /* already archived */)
+ r = journal_file_open(fname, flags, mode, compress, seal, metrics, mmap_cache, template, ret);
+ if (!IN_SET(r,
+ -EBADMSG, /* corrupted */
+ -ENODATA, /* truncated */
+ -EHOSTDOWN, /* other machine */
+ -EPROTONOSUPPORT, /* incompatible feature */
+ -EBUSY, /* unclean shutdown */
+ -ESHUTDOWN, /* already archived */
+ -EIO, /* IO error, including SIGBUS on mmap */
+ -EIDRM /* File has been deleted */))
return r;
if ((flags & O_ACCMODE) == O_RDONLY)
/* The file is corrupted. Rotate it away and try it again (but only once) */
l = strlen(fname);
- if (asprintf(&p, "%.*s@%016llx-%016" PRIx64 ".journal~",
+ if (asprintf(&p, "%.*s@%016"PRIx64 "-%016"PRIx64 ".journal~",
(int) l - 8, fname,
- (unsigned long long) now(CLOCK_REALTIME),
+ now(CLOCK_REALTIME),
random_u64()) < 0)
return -ENOMEM;
- r = rename(fname, p);
- if (r < 0)
+ if (rename(fname, p) < 0)
return -errno;
- log_warning("File %s corrupted or uncleanly shut down, renaming and replacing.", fname);
+ /* btrfs doesn't cope well with our write pattern and
+ * fragments heavily. Let's defrag all files we rotate */
+
+ (void) chattr_path(p, false, FS_NOCOW_FL);
+ (void) btrfs_defrag(p);
+
+ log_warning_errno(r, "File %s corrupted or uncleanly shut down, renaming and replacing.", fname);
- return journal_file_open(fname, flags, mode, compress, seal,
- metrics, mmap_cache, template, ret);
+ return journal_file_open(fname, flags, mode, compress, seal, metrics, mmap_cache, template, ret);
}
int journal_file_copy_entry(JournalFile *from, JournalFile *to, Object *o, uint64_t p, uint64_t *seqnum, Object **ret, uint64_t *offset) {
if ((uint64_t) t != l)
return -E2BIG;
- if (o->object.flags & OBJECT_COMPRESSED) {
-#ifdef HAVE_XZ
- uint64_t rsize;
+ if (o->object.flags & OBJECT_COMPRESSION_MASK) {
+#if defined(HAVE_XZ) || defined(HAVE_LZ4)
+ size_t rsize = 0;
- if (!uncompress_blob(o->data.payload, l, &from->compress_buffer, &from->compress_buffer_size, &rsize, 0))
- return -EBADMSG;
+ r = decompress_blob(o->object.flags & OBJECT_COMPRESSION_MASK,
+ o->data.payload, l, &from->compress_buffer, &from->compress_buffer_size, &rsize, 0);
+ if (r < 0)
+ return r;
data = from->compress_buffer;
l = rsize;
return r;
}
- return journal_file_append_entry_internal(to, &ts, xor_hash, items, n, seqnum, ret, offset);
+ r = journal_file_append_entry_internal(to, &ts, xor_hash, items, n, seqnum, ret, offset);
+
+ if (mmap_cache_got_sigbus(to->mmap, to->fd))
+ return -EIO;
+
+ return r;
+}
+
+void journal_reset_metrics(JournalMetrics *m) {
+ assert(m);
+
+ /* Set everything to "pick automatic values". */
+
+ *m = (JournalMetrics) {
+ .min_use = (uint64_t) -1,
+ .max_use = (uint64_t) -1,
+ .min_size = (uint64_t) -1,
+ .max_size = (uint64_t) -1,
+ .keep_free = (uint64_t) -1,
+ .n_max_files = (uint64_t) -1,
+ };
}
void journal_default_metrics(JournalMetrics *m, int fd) {
- uint64_t fs_size = 0;
+ char a[FORMAT_BYTES_MAX], b[FORMAT_BYTES_MAX], c[FORMAT_BYTES_MAX], d[FORMAT_BYTES_MAX], e[FORMAT_BYTES_MAX];
struct statvfs ss;
- char a[FORMAT_BYTES_MAX], b[FORMAT_BYTES_MAX], c[FORMAT_BYTES_MAX], d[FORMAT_BYTES_MAX];
+ uint64_t fs_size;
assert(m);
assert(fd >= 0);
if (fstatvfs(fd, &ss) >= 0)
fs_size = ss.f_frsize * ss.f_blocks;
+ else {
+ log_debug_errno(errno, "Failed to detremine disk size: %m");
+ fs_size = 0;
+ }
if (m->max_use == (uint64_t) -1) {
} else {
m->max_use = PAGE_ALIGN(m->max_use);
- if (m->max_use < JOURNAL_FILE_SIZE_MIN*2)
+ if (m->max_use != 0 && m->max_use < JOURNAL_FILE_SIZE_MIN*2)
m->max_use = JOURNAL_FILE_SIZE_MIN*2;
}
+ if (m->min_use == (uint64_t) -1)
+ m->min_use = DEFAULT_MIN_USE;
+
+ if (m->min_use > m->max_use)
+ m->min_use = m->max_use;
+
if (m->max_size == (uint64_t) -1) {
m->max_size = PAGE_ALIGN(m->max_use / 8); /* 8 chunks */
} 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 != 0) {
+ 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->max_use != 0 && 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;
if (m->min_size < JOURNAL_FILE_SIZE_MIN)
m->min_size = JOURNAL_FILE_SIZE_MIN;
- if (m->min_size > m->max_size)
+ if (m->max_size != 0 && m->min_size > m->max_size)
m->max_size = m->min_size;
}
m->keep_free = DEFAULT_KEEP_FREE;
}
- log_debug("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));
+ if (m->n_max_files == (uint64_t) -1)
+ m->n_max_files = DEFAULT_N_MAX_FILES;
+
+ log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64,
+ format_bytes(a, sizeof(a), m->min_use),
+ format_bytes(b, sizeof(b), m->max_use),
+ format_bytes(c, sizeof(c), m->max_size),
+ format_bytes(d, sizeof(d), m->min_size),
+ format_bytes(e, sizeof(e), m->keep_free),
+ m->n_max_files);
}
int journal_file_get_cutoff_realtime_usec(JournalFile *f, usec_t *from, usec_t *to) {
projects / thirdparty / systemd.git / blobdiff