/* Either make a copy into pad and return that, or return ptr into mmap. */
/* Note: pad has to be a real object, so we can't get here if len
* overflows size_t */
-/* FIXME: Transaction */
void *tdb_get(struct tdb_context *tdb, tdb_off_t off, void *pad, size_t len)
{
- ssize_t r;
-
if (likely(!(tdb->flags & TDB_CONVERT))) {
void *ret = tdb_direct(tdb, off, len);
if (ret)
if (unlikely(tdb_oob(tdb, off + len, false) == -1))
return NULL;
- r = pread(tdb->fd, pad, len, off);
- if (r != (ssize_t)len) {
- /* Ensure ecode is set for log fn. */
- tdb->ecode = TDB_ERR_IO;
- tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
- "tdb_read failed at %llu "
- "len=%lld ret=%lld (%s) map_size=%lld\n",
- (long long)off, (long long)len,
- (long long)r, strerror(errno),
- (long long)tdb->map_size);
+ if (tdb->methods->read(tdb, off, pad, len) == -1)
return NULL;
- }
return tdb_convert(tdb, pad, len);
}
return 0;
} else {
char buf[8192] = { 0 };
- return fill(tdb, buf, sizeof(buf), len, off);
+ return fill(tdb, buf, sizeof(buf), off, len);
}
}
{
tdb_off_t pad, *ret;
- ret = tdb_get(tdb, off, &pad, sizeof(ret));
+ ret = tdb_get(tdb, off, &pad, sizeof(pad));
if (!ret) {
return TDB_OFF_ERR;
}
bool tdb_pwrite_all(int fd, const void *buf, size_t len, tdb_off_t off)
{
while (len) {
- size_t ret;
+ ssize_t ret;
ret = pwrite(fd, buf, len, off);
if (ret < 0)
return false;
errno = ENOSPC;
return false;
}
- buf += ret;
+ buf = (char *)buf + ret;
+ off += ret;
+ len -= ret;
+ }
+ return true;
+}
+
+/* Even on files, we can get partial reads due to signals. */
+bool tdb_pread_all(int fd, void *buf, size_t len, tdb_off_t off)
+{
+ while (len) {
+ ssize_t ret;
+ ret = pread(fd, buf, len, off);
+ if (ret < 0)
+ return false;
+ if (ret == 0) {
+ /* ETOOSHORT? */
+ errno = EWOULDBLOCK;
+ return false;
+ }
+ buf = (char *)buf + ret;
off += ret;
len -= ret;
}
return true;
}
+bool tdb_read_all(int fd, void *buf, size_t len)
+{
+ while (len) {
+ ssize_t ret;
+ ret = read(fd, buf, len);
+ if (ret < 0)
+ return false;
+ if (ret == 0) {
+ /* ETOOSHORT? */
+ errno = EWOULDBLOCK;
+ return false;
+ }
+ buf = (char *)buf + ret;
+ len -= ret;
+ }
+ return true;
+}
+
/* write a lump of data at a specified offset */
static int tdb_write(struct tdb_context *tdb, tdb_off_t off,
const void *buf, tdb_len_t len)
if (tdb->map_ptr) {
memcpy(buf, off + (char *)tdb->map_ptr, len);
} else {
- ssize_t ret = pread(tdb->fd, buf, len, off);
- if (ret != (ssize_t)len) {
+ if (!tdb_pread_all(tdb->fd, buf, len, off)) {
/* Ensure ecode is set for log fn. */
tdb->ecode = TDB_ERR_IO;
tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
"tdb_read failed at %lld "
- "len=%lld ret=%lld (%s) map_size=%lld\n",
+ "len=%lld (%s) map_size=%lld\n",
(long long)off, (long long)len,
- (long long)ret, strerror(errno),
+ strerror(errno),
(long long)tdb->map_size);
return -1;
}
void *key;
uint64_t klen, hash;
- r = tdb_get(tdb, off, &pad, sizeof(*r));
+ r = tdb_get(tdb, off, &pad, sizeof(pad));
if (!r)
/* FIXME */
return 0;
klen = rec_key_length(r);
- key = tdb_direct(tdb, off + sizeof(*r), klen);
+ key = tdb_direct(tdb, off + sizeof(pad), klen);
if (likely(key))
return tdb_hash(tdb, key, klen);
- key = tdb_alloc_read(tdb, off + sizeof(*r), klen);
+ key = tdb_alloc_read(tdb, off + sizeof(pad), klen);
if (unlikely(!key))
return 0;
hash = tdb_hash(tdb, key, klen);
/* expand a file. we prefer to use ftruncate, as that is what posix
says to use for mmap expansion */
-static int tdb_expand_file(struct tdb_context *tdb,
- tdb_len_t size, tdb_len_t addition)
+static int tdb_expand_file(struct tdb_context *tdb, tdb_len_t addition)
{
char buf[8192];
return -1;
}
- /* If this fails, we try to fill anyway. */
- if (ftruncate(tdb->fd, size+addition))
- ;
+ if (tdb->flags & TDB_INTERNAL) {
+ char *new = realloc(tdb->map_ptr, tdb->map_size + addition);
+ if (!new) {
+ tdb->ecode = TDB_ERR_OOM;
+ return -1;
+ }
+ tdb->map_ptr = new;
+ tdb->map_size += addition;
+ } else {
+ /* Unmap before trying to write; old TDB claimed OpenBSD had
+ * problem with this otherwise. */
+ tdb_munmap(tdb);
+
+ /* If this fails, we try to fill anyway. */
+ if (ftruncate(tdb->fd, tdb->map_size + addition))
+ ;
- /* now fill the file with something. This ensures that the
- file isn't sparse, which would be very bad if we ran out of
- disk. This must be done with write, not via mmap */
- memset(buf, 0x43, sizeof(buf));
- return fill(tdb, buf, sizeof(buf), addition, size);
+ /* now fill the file with something. This ensures that the
+ file isn't sparse, which would be very bad if we ran out of
+ disk. This must be done with write, not via mmap */
+ memset(buf, 0x43, sizeof(buf));
+ if (fill(tdb, buf, sizeof(buf), tdb->map_size, addition) == -1)
+ return -1;
+ tdb->map_size += addition;
+ tdb_mmap(tdb);
+ }
+ return 0;
}
const void *tdb_access_read(struct tdb_context *tdb,
(*chain) = h;
}
-
-/* expand the database by expanding the underlying file and doing the
- mmap again if necessary */
-int tdb_expand(struct tdb_context *tdb)
-{
- struct tdb_record rec;
- tdb_off_t offset, new_size;
-
- /* We have to lock every hash bucket and every free list. */
- do {
-
-
- if (tdb_lock(tdb, -1, F_WRLCK) == -1) {
- TDB_LOG((tdb, TDB_DEBUG_ERROR, "lock failed in tdb_expand\n"));
- return -1;
- }
-
- /* must know about any previous expansions by another process */
- tdb->methods->tdb_oob(tdb, tdb->map_size + 1, 1);
-
- /* always make room for at least 100 more records, and at
- least 25% more space. Round the database up to a multiple
- of the page size */
- new_size = MAX(tdb->map_size + size*100, tdb->map_size * 1.25);
- size = TDB_ALIGN(new_size, tdb->page_size) - tdb->map_size;
-
- if (!(tdb->flags & TDB_INTERNAL))
- tdb_munmap(tdb);
-
- /*
- * We must ensure the file is unmapped before doing this
- * to ensure consistency with systems like OpenBSD where
- * writes and mmaps are not consistent.
- */
-
- /* expand the file itself */
- if (!(tdb->flags & TDB_INTERNAL)) {
- if (tdb->methods->tdb_expand_file(tdb, tdb->map_size, size) != 0)
- goto fail;
- }
-
- tdb->map_size += size;
-
- if (tdb->flags & TDB_INTERNAL) {
- char *new_map_ptr = (char *)realloc(tdb->map_ptr,
- tdb->map_size);
- if (!new_map_ptr) {
- tdb->map_size -= size;
- goto fail;
- }
- tdb->map_ptr = new_map_ptr;
- } else {
- /*
- * We must ensure the file is remapped before adding the space
- * to ensure consistency with systems like OpenBSD where
- * writes and mmaps are not consistent.
- */
-
- /* We're ok if the mmap fails as we'll fallback to read/write */
- tdb_mmap(tdb);
- }
-
- /* form a new freelist record */
- memset(&rec,'\0',sizeof(rec));
- rec.rec_len = size - sizeof(rec);
-
- /* link it into the free list */
- offset = tdb->map_size - size;
- if (tdb_free(tdb, offset, &rec) == -1)
- goto fail;
-
- tdb_unlock(tdb, -1, F_WRLCK);
- return 0;
- fail:
- tdb_unlock(tdb, -1, F_WRLCK);
- return -1;
-}
-
/* read/write a tdb_off_t */
int tdb_ofs_read(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d)
{