License along with this library; if not, see <http://www.gnu.org/licenses/>.
*/
#include "private.h"
+#include <assert.h>
#include <ccan/likely/likely.h>
void tdb_munmap(struct tdb_context *tdb)
static int tdb_oob(struct tdb_context *tdb, tdb_off_t len, bool probe)
{
struct stat st;
+ int ret;
+
+ /* FIXME: We can't hold pointers during this: we could unmap! */
+ /* (We currently do this in traverse!) */
+// assert(!tdb->direct_access || tdb_has_expansion_lock(tdb));
+
if (len <= tdb->map_size)
return 0;
if (tdb->flags & TDB_INTERNAL) {
return -1;
}
- if (fstat(tdb->fd, &st) == -1) {
+ if (tdb_lock_expand(tdb, F_RDLCK) != 0)
+ return -1;
+
+ ret = fstat(tdb->fd, &st);
+
+ tdb_unlock_expand(tdb, F_RDLCK);
+
+ if (ret == -1) {
tdb->ecode = TDB_ERR_IO;
return -1;
}
/* Unmap, update size, remap */
tdb_munmap(tdb);
+
tdb->map_size = st.st_size;
tdb_mmap(tdb);
return 0;
/* 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)
return 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);
- return NULL;
- }
- return tdb_convert(tdb, pad, len);
+ return tdb_read_convert(tdb, off, pad, len) == -1 ? NULL : pad;
}
/* Endian conversion: we only ever deal with 8 byte quantities */
void *tdb_convert(const struct tdb_context *tdb, void *buf, tdb_len_t size)
{
- if (unlikely((tdb->flags & TDB_CONVERT))) {
+ if (unlikely((tdb->flags & TDB_CONVERT)) && buf) {
uint64_t i, *p = (uint64_t *)buf;
for (i = 0; i < size / 8; i++)
p[i] = bswap_64(p[i]);
uint64_t tdb_find_nonzero_off(struct tdb_context *tdb, tdb_off_t off,
uint64_t num)
{
- uint64_t i, *val;
- bool alloc = false;
+ uint64_t i;
+ const uint64_t *val;
- val = tdb_direct(tdb, off, num * sizeof(tdb_off_t));
- if (!unlikely(val)) {
- val = tdb_alloc_read(tdb, off, num * sizeof(tdb_off_t));
- if (!val)
- return num;
- alloc = true;
- }
+ /* Zero vs non-zero is the same unconverted: minor optimization. */
+ val = tdb_access_read(tdb, off, num * sizeof(tdb_off_t), false);
+ if (!val)
+ return num;
for (i = 0; i < num; i++) {
if (val[i])
break;
}
- if (unlikely(alloc))
- free(val);
+ tdb_access_release(tdb, val);
return i;
}
uint64_t tdb_find_zero_off(struct tdb_context *tdb, tdb_off_t off,
uint64_t num)
{
- uint64_t i, *val;
- bool alloc = false;
+ uint64_t i;
+ const uint64_t *val;
- val = tdb_direct(tdb, off, num * sizeof(tdb_off_t));
- if (!unlikely(val)) {
- val = tdb_alloc_read(tdb, off, num * sizeof(tdb_off_t));
- if (!val)
- return num;
- alloc = true;
- }
+ /* Zero vs non-zero is the same unconverted: minor optimization. */
+ val = tdb_access_read(tdb, off, num * sizeof(tdb_off_t), false);
+ if (!val)
+ return num;
for (i = 0; i < num; i++) {
if (!val[i])
break;
}
- if (unlikely(alloc))
- free(val);
+ tdb_access_release(tdb, val);
return i;
}
-static int fill(struct tdb_context *tdb,
- const void *buf, size_t size,
- tdb_off_t off, tdb_len_t len)
-{
- while (len) {
- size_t n = len > size ? size : len;
-
- if (!tdb_pwrite_all(tdb->fd, buf, n, off)) {
- tdb->ecode = TDB_ERR_IO;
- tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
- "fill write failed: giving up!\n");
- return -1;
- }
- len -= n;
- off += n;
- }
- return 0;
-}
-
int zero_out(struct tdb_context *tdb, tdb_off_t off, tdb_len_t len)
{
+ char buf[8192] = { 0 };
void *p = tdb_direct(tdb, off, len);
if (p) {
memset(p, 0, len);
return 0;
- } else {
- char buf[8192] = { 0 };
- return fill(tdb, buf, sizeof(buf), len, off);
}
+ while (len) {
+ unsigned todo = len < sizeof(buf) ? len : sizeof(buf);
+ if (tdb->methods->write(tdb, off, buf, todo) == -1)
+ return -1;
+ len -= todo;
+ off += todo;
+ }
+ return 0;
}
tdb_off_t tdb_read_off(struct tdb_context *tdb, tdb_off_t off)
{
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;
}
}
int tdb_write_convert(struct tdb_context *tdb, tdb_off_t off,
- void *rec, size_t len)
+ const void *rec, size_t len)
{
- return tdb->methods->write(tdb, off, tdb_convert(tdb, rec, len), len);
+ int ret;
+ if (unlikely((tdb->flags & TDB_CONVERT))) {
+ void *conv = malloc(len);
+ if (!conv) {
+ tdb->ecode = TDB_ERR_OOM;
+ tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
+ "tdb_write: no memory converting %zu bytes\n",
+ len);
+ return -1;
+ }
+ memcpy(conv, rec, len);
+ ret = tdb->methods->write(tdb, off,
+ tdb_convert(tdb, conv, len), len);
+ free(conv);
+ } else
+ ret = tdb->methods->write(tdb, off, rec, len);
+
+ return ret;
}
int tdb_read_convert(struct tdb_context *tdb, tdb_off_t off,
return tdb_write_convert(tdb, off, &val, sizeof(val));
}
-/* read a lump of data, allocating the space for it */
-void *tdb_alloc_read(struct tdb_context *tdb, tdb_off_t offset, tdb_len_t len)
+static void *_tdb_alloc_read(struct tdb_context *tdb, tdb_off_t offset,
+ tdb_len_t len, unsigned int prefix)
{
void *buf;
/* some systems don't like zero length malloc */
- buf = malloc(len ? len : 1);
+ buf = malloc(prefix + len ? prefix + len : 1);
if (unlikely(!buf)) {
tdb->ecode = TDB_ERR_OOM;
tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
"tdb_alloc_read malloc failed len=%lld\n",
- (long long)len);
- } else if (unlikely(tdb->methods->read(tdb, offset, buf, len))) {
+ (long long)prefix + len);
+ } else if (unlikely(tdb->methods->read(tdb, offset, buf+prefix, len))) {
free(buf);
buf = NULL;
}
return buf;
}
-uint64_t hash_record(struct tdb_context *tdb, tdb_off_t off)
+/* read a lump of data, allocating the space for it */
+void *tdb_alloc_read(struct tdb_context *tdb, tdb_off_t offset, tdb_len_t len)
{
- struct tdb_used_record pad, *r;
- void *key;
- uint64_t klen, hash;
-
- r = tdb_get(tdb, off, &pad, sizeof(*r));
- if (!r)
- /* FIXME */
- return 0;
+ return _tdb_alloc_read(tdb, offset, len, 0);
+}
- klen = rec_key_length(r);
- key = tdb_direct(tdb, off + sizeof(*r), klen);
- if (likely(key))
- return tdb_hash(tdb, key, klen);
+static int fill(struct tdb_context *tdb,
+ const void *buf, size_t size,
+ tdb_off_t off, tdb_len_t len)
+{
+ while (len) {
+ size_t n = len > size ? size : len;
- key = tdb_alloc_read(tdb, off + sizeof(*r), klen);
- if (unlikely(!key))
- return 0;
- hash = tdb_hash(tdb, key, klen);
- free(key);
- return hash;
-}
-
-/* Give a piece of tdb data to a parser */
-int tdb_parse_data(struct tdb_context *tdb, TDB_DATA key,
- tdb_off_t offset, tdb_len_t len,
- int (*parser)(TDB_DATA key, TDB_DATA data,
- void *private_data),
- void *private_data)
-{
- TDB_DATA data;
- int result;
- bool allocated = false;
-
- data.dsize = len;
- data.dptr = tdb_direct(tdb, offset, len);
- if (unlikely(!data.dptr)) {
- if (!(data.dptr = tdb_alloc_read(tdb, offset, len))) {
+ if (!tdb_pwrite_all(tdb->fd, buf, n, off)) {
+ tdb->ecode = TDB_ERR_IO;
+ tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
+ "fill write failed: giving up!\n");
return -1;
}
- allocated = true;
+ len -= n;
+ off += n;
}
- result = parser(key, data, private_data);
- if (unlikely(allocated))
- free(data.dptr);
- return result;
+ return 0;
}
/* 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;
}
+/* This is only neded for tdb_access_commit, but used everywhere to simplify. */
+struct tdb_access_hdr {
+ tdb_off_t off;
+ tdb_len_t len;
+ bool convert;
+};
+
const void *tdb_access_read(struct tdb_context *tdb,
- tdb_off_t off, tdb_len_t len)
+ tdb_off_t off, tdb_len_t len, bool convert)
{
- const void *ret = tdb_direct(tdb, off, len);
+ const void *ret = NULL;
+
+ if (likely(!(tdb->flags & TDB_CONVERT)))
+ ret = tdb_direct(tdb, off, len);
+
+ if (!ret) {
+ struct tdb_access_hdr *hdr;
+ hdr = _tdb_alloc_read(tdb, off, len, sizeof(*hdr));
+ if (hdr) {
+ ret = hdr + 1;
+ if (convert)
+ tdb_convert(tdb, (void *)ret, len);
+ }
+ } else
+ tdb->direct_access++;
+
+ return ret;
+}
+
+void *tdb_access_write(struct tdb_context *tdb,
+ tdb_off_t off, tdb_len_t len, bool convert)
+{
+ void *ret = NULL;
+
+ if (likely(!(tdb->flags & TDB_CONVERT)))
+ ret = tdb_direct(tdb, off, len);
+
+ if (!ret) {
+ struct tdb_access_hdr *hdr;
+ hdr = _tdb_alloc_read(tdb, off, len, sizeof(*hdr));
+ if (hdr) {
+ hdr->off = off;
+ hdr->len = len;
+ hdr->convert = convert;
+ ret = hdr + 1;
+ if (convert)
+ tdb_convert(tdb, (void *)ret, len);
+ }
+ } else
+ tdb->direct_access++;
- if (!ret)
- ret = tdb_alloc_read(tdb, off, len);
return ret;
}
if (!tdb->map_ptr
|| (char *)p < (char *)tdb->map_ptr
|| (char *)p >= (char *)tdb->map_ptr + tdb->map_size)
- free((void *)p);
+ free((struct tdb_access_hdr *)p - 1);
+ else
+ tdb->direct_access--;
+}
+
+int tdb_access_commit(struct tdb_context *tdb, void *p)
+{
+ int ret = 0;
+
+ if (!tdb->map_ptr
+ || (char *)p < (char *)tdb->map_ptr
+ || (char *)p >= (char *)tdb->map_ptr + tdb->map_size) {
+ struct tdb_access_hdr *hdr;
+
+ hdr = (struct tdb_access_hdr *)p - 1;
+ if (hdr->convert)
+ ret = tdb_write_convert(tdb, hdr->off, p, hdr->len);
+ else
+ ret = tdb_write(tdb, hdr->off, p, hdr->len);
+ free(hdr);
+ } else
+ tdb->direct_access--;
+
+ return ret;
}
#if 0
(*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)
{