#include <assert.h>
#include <ccan/likely/likely.h>
-void tdb_munmap(struct tdb_context *tdb)
+void tdb_munmap(struct tdb_file *file)
{
- if (tdb->flags & TDB_INTERNAL)
+ if (file->fd == -1)
return;
- if (tdb->map_ptr) {
- munmap(tdb->map_ptr, tdb->map_size);
- tdb->map_ptr = NULL;
+ if (file->map_ptr) {
+ munmap(file->map_ptr, file->map_size);
+ file->map_ptr = NULL;
}
}
void tdb_mmap(struct tdb_context *tdb)
{
+ int mmap_flags;
+
if (tdb->flags & TDB_INTERNAL)
return;
if (tdb->flags & TDB_NOMMAP)
return;
- tdb->map_ptr = mmap(NULL, tdb->map_size, tdb->mmap_flags,
- MAP_SHARED, tdb->fd, 0);
+ if ((tdb->open_flags & O_ACCMODE) == O_RDONLY)
+ mmap_flags = PROT_READ;
+ else
+ mmap_flags = PROT_READ | PROT_WRITE;
+
+ /* size_t can be smaller than off_t. */
+ if ((size_t)tdb->file->map_size == tdb->file->map_size) {
+ tdb->file->map_ptr = mmap(NULL, tdb->file->map_size,
+ mmap_flags,
+ MAP_SHARED, tdb->file->fd, 0);
+ } else
+ tdb->file->map_ptr = MAP_FAILED;
/*
* NB. When mmap fails it returns MAP_FAILED *NOT* NULL !!!!
*/
- if (tdb->map_ptr == MAP_FAILED) {
- tdb->map_ptr = NULL;
+ if (tdb->file->map_ptr == MAP_FAILED) {
+ tdb->file->map_ptr = NULL;
tdb_logerr(tdb, TDB_SUCCESS, TDB_LOG_WARNING,
"tdb_mmap failed for size %lld (%s)",
- (long long)tdb->map_size, strerror(errno));
+ (long long)tdb->file->map_size, strerror(errno));
}
}
/* check for an out of bounds access - if it is out of bounds then
see if the database has been expanded by someone else and expand
if necessary
- note that "len" is the minimum length needed for the db
+ note that "len" is the minimum length needed for the db.
+
+ If probe is true, len being too large isn't a failure.
*/
static enum TDB_ERROR tdb_oob(struct tdb_context *tdb, tdb_off_t len,
bool probe)
enum TDB_ERROR ecode;
/* We can't hold pointers during this: we could unmap! */
- assert(!tdb->direct_access
+ assert(!tdb->tdb2.direct_access
|| (tdb->flags & TDB_NOLOCK)
|| tdb_has_expansion_lock(tdb));
- if (len <= tdb->map_size)
- return 0;
+ if (len <= tdb->file->map_size)
+ return TDB_SUCCESS;
if (tdb->flags & TDB_INTERNAL) {
- if (!probe) {
- tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
- "tdb_oob len %lld beyond internal"
- " malloc size %lld",
- (long long)len,
- (long long)tdb->map_size);
- }
+ if (probe)
+ return TDB_SUCCESS;
+
+ tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
+ "tdb_oob len %lld beyond internal"
+ " malloc size %lld",
+ (long long)len,
+ (long long)tdb->file->map_size);
return TDB_ERR_IO;
}
return ecode;
}
- if (fstat(tdb->fd, &st) != 0) {
+ if (fstat(tdb->file->fd, &st) != 0) {
tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
"Failed to fstat file: %s", strerror(errno));
tdb_unlock_expand(tdb, F_RDLCK);
tdb_unlock_expand(tdb, F_RDLCK);
if (st.st_size < (size_t)len) {
- if (!probe) {
- tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
- "tdb_oob len %zu beyond eof at %zu",
- (size_t)len, st.st_size);
- }
+ if (probe)
+ return TDB_SUCCESS;
+
+ tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
+ "tdb_oob len %zu beyond eof at %zu",
+ (size_t)len, st.st_size);
return TDB_ERR_IO;
}
/* Unmap, update size, remap */
- tdb_munmap(tdb);
+ tdb_munmap(tdb->file);
- tdb->map_size = st.st_size;
+ tdb->file->map_size = st.st_size;
tdb_mmap(tdb);
return TDB_SUCCESS;
}
/* Endian conversion: we only ever deal with 8 byte quantities */
void *tdb_convert(const struct tdb_context *tdb, void *buf, tdb_len_t size)
{
+ assert(size % 8 == 0);
if (unlikely((tdb->flags & TDB_CONVERT)) && buf) {
uint64_t i, *p = (uint64_t *)buf;
for (i = 0; i < size / 8; i++)
return buf;
}
+/* Return first non-zero offset in offset array, or end, or -ve error. */
/* FIXME: Return the off? */
uint64_t tdb_find_nonzero_off(struct tdb_context *tdb,
tdb_off_t base, uint64_t start, uint64_t end)
/* Zero vs non-zero is the same unconverted: minor optimization. */
val = tdb_access_read(tdb, base + start * sizeof(tdb_off_t),
(end - start) * sizeof(tdb_off_t), false);
- if (!val)
- return end;
+ if (TDB_PTR_IS_ERR(val)) {
+ return TDB_ERR_TO_OFF(TDB_PTR_ERR(val));
+ }
for (i = 0; i < (end - start); i++) {
if (val[i])
return start + i;
}
-/* Return first zero offset in num offset array, or num. */
+/* Return first zero offset in num offset array, or num, or -ve error. */
uint64_t tdb_find_zero_off(struct tdb_context *tdb, tdb_off_t off,
uint64_t num)
{
/* 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;
+ if (TDB_PTR_IS_ERR(val)) {
+ return TDB_ERR_TO_OFF(TDB_PTR_ERR(val));
+ }
for (i = 0; i < num; i++) {
if (!val[i])
enum TDB_ERROR zero_out(struct tdb_context *tdb, tdb_off_t off, tdb_len_t len)
{
char buf[8192] = { 0 };
- void *p = tdb->methods->direct(tdb, off, len, true);
+ void *p = tdb->tdb2.io->direct(tdb, off, len, true);
enum TDB_ERROR ecode = TDB_SUCCESS;
- assert(!tdb->read_only);
+ assert(!(tdb->flags & TDB_RDONLY));
+ if (TDB_PTR_IS_ERR(p)) {
+ return TDB_PTR_ERR(p);
+ }
if (p) {
memset(p, 0, len);
return ecode;
}
while (len) {
unsigned todo = len < sizeof(buf) ? len : sizeof(buf);
- ecode = tdb->methods->twrite(tdb, off, buf, todo);
+ ecode = tdb->tdb2.io->twrite(tdb, off, buf, todo);
if (ecode != TDB_SUCCESS) {
break;
}
enum TDB_ERROR ecode;
if (likely(!(tdb->flags & TDB_CONVERT))) {
- tdb_off_t *p = tdb->methods->direct(tdb, off, sizeof(*p),
+ tdb_off_t *p = tdb->tdb2.io->direct(tdb, off, sizeof(*p),
false);
+ if (TDB_PTR_IS_ERR(p)) {
+ return TDB_ERR_TO_OFF(TDB_PTR_ERR(p));
+ }
if (p)
return *p;
}
ecode = tdb_read_convert(tdb, off, &ret, sizeof(ret));
if (ecode != TDB_SUCCESS) {
- tdb->ecode = ecode;
- return TDB_OFF_ERR;
+ return TDB_ERR_TO_OFF(ecode);
}
return ret;
}
{
enum TDB_ERROR ecode;
- if (tdb->read_only) {
+ if (tdb->flags & TDB_RDONLY) {
return tdb_logerr(tdb, TDB_ERR_RDONLY, TDB_LOG_USE_ERROR,
"Write to read-only database");
}
- /* FIXME: Bogus optimization? */
- if (len == 0) {
- return TDB_SUCCESS;
- }
-
- ecode = tdb->methods->oob(tdb, off + len, 0);
+ ecode = tdb->tdb2.io->oob(tdb, off + len, false);
if (ecode != TDB_SUCCESS) {
return ecode;
}
- if (tdb->map_ptr) {
- memcpy(off + (char *)tdb->map_ptr, buf, len);
+ if (tdb->file->map_ptr) {
+ memcpy(off + (char *)tdb->file->map_ptr, buf, len);
} else {
ssize_t ret;
- ret = pwrite(tdb->fd, buf, len, off);
- if (ret < len) {
+ ret = pwrite(tdb->file->fd, buf, len, off);
+ if (ret != len) {
/* This shouldn't happen: we avoid sparse files. */
if (ret >= 0)
errno = ENOSPC;
{
enum TDB_ERROR ecode;
- ecode = tdb->methods->oob(tdb, off + len, 0);
+ ecode = tdb->tdb2.io->oob(tdb, off + len, false);
if (ecode != TDB_SUCCESS) {
return ecode;
}
- if (tdb->map_ptr) {
- memcpy(buf, off + (char *)tdb->map_ptr, len);
+ if (tdb->file->map_ptr) {
+ memcpy(buf, off + (char *)tdb->file->map_ptr, len);
} else {
- ssize_t r = pread(tdb->fd, buf, len, off);
+ ssize_t r = pread(tdb->file->fd, buf, len, off);
if (r != len) {
return tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
"tdb_read failed with %zi at %zu "
"len=%zu (%s) map_size=%zu",
r, (size_t)off, (size_t)len,
strerror(errno),
- (size_t)tdb->map_size);
+ (size_t)tdb->file->map_size);
}
}
return TDB_SUCCESS;
" %zu bytes", len);
}
memcpy(conv, rec, len);
- ecode = tdb->methods->twrite(tdb, off,
+ ecode = tdb->tdb2.io->twrite(tdb, off,
tdb_convert(tdb, conv, len), len);
free(conv);
} else {
- ecode = tdb->methods->twrite(tdb, off, rec, len);
+ ecode = tdb->tdb2.io->twrite(tdb, off, rec, len);
}
return ecode;
}
enum TDB_ERROR tdb_read_convert(struct tdb_context *tdb, tdb_off_t off,
void *rec, size_t len)
{
- enum TDB_ERROR ecode = tdb->methods->tread(tdb, off, rec, len);
+ enum TDB_ERROR ecode = tdb->tdb2.io->tread(tdb, off, rec, len);
tdb_convert(tdb, rec, len);
return ecode;
}
enum TDB_ERROR tdb_write_off(struct tdb_context *tdb,
tdb_off_t off, tdb_off_t val)
{
- if (tdb->read_only) {
+ if (tdb->flags & TDB_RDONLY) {
return tdb_logerr(tdb, TDB_ERR_RDONLY, TDB_LOG_USE_ERROR,
"Write to read-only database");
}
if (likely(!(tdb->flags & TDB_CONVERT))) {
- tdb_off_t *p = tdb->methods->direct(tdb, off, sizeof(*p),
+ tdb_off_t *p = tdb->tdb2.io->direct(tdb, off, sizeof(*p),
true);
+ if (TDB_PTR_IS_ERR(p)) {
+ return TDB_PTR_ERR(p);
+ }
if (p) {
*p = val;
return TDB_SUCCESS;
static void *_tdb_alloc_read(struct tdb_context *tdb, tdb_off_t offset,
tdb_len_t len, unsigned int prefix)
{
- void *buf;
+ unsigned char *buf;
enum TDB_ERROR ecode;
/* some systems don't like zero length malloc */
tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_USE_ERROR,
"tdb_alloc_read malloc failed len=%zu",
(size_t)(prefix + len));
+ return TDB_ERR_PTR(TDB_ERR_OOM);
} else {
- ecode = tdb->methods->tread(tdb, offset, buf+prefix, len);
+ ecode = tdb->tdb2.io->tread(tdb, offset, buf+prefix, len);
if (unlikely(ecode != TDB_SUCCESS)) {
- tdb->ecode = ecode;
free(buf);
- buf = NULL;
+ return TDB_ERR_PTR(ecode);
}
}
return buf;
{
while (len) {
size_t n = len > size ? size : len;
- ssize_t ret = pwrite(tdb->fd, buf, n, off);
- if (ret < n) {
+ ssize_t ret = pwrite(tdb->file->fd, buf, n, off);
+ if (ret != n) {
if (ret >= 0)
errno = ENOSPC;
char buf[8192];
enum TDB_ERROR ecode;
- if (tdb->read_only) {
+ if (tdb->flags & TDB_RDONLY) {
return tdb_logerr(tdb, TDB_ERR_RDONLY, TDB_LOG_USE_ERROR,
"Expand on read-only database");
}
if (tdb->flags & TDB_INTERNAL) {
- char *new = realloc(tdb->map_ptr, tdb->map_size + addition);
+ char *new = realloc(tdb->file->map_ptr,
+ tdb->file->map_size + addition);
if (!new) {
return tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
"No memory to expand database");
}
- tdb->map_ptr = new;
- tdb->map_size += addition;
+ tdb->file->map_ptr = new;
+ tdb->file->map_size += addition;
} else {
/* Unmap before trying to write; old TDB claimed OpenBSD had
* problem with this otherwise. */
- tdb_munmap(tdb);
+ tdb_munmap(tdb->file);
/* If this fails, we try to fill anyway. */
- if (ftruncate(tdb->fd, tdb->map_size + addition))
+ if (ftruncate(tdb->file->fd, tdb->file->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));
- ecode = fill(tdb, buf, sizeof(buf), tdb->map_size, addition);
+ ecode = fill(tdb, buf, sizeof(buf), tdb->file->map_size,
+ addition);
if (ecode != TDB_SUCCESS)
return ecode;
- tdb->map_size += addition;
+ tdb->file->map_size += addition;
tdb_mmap(tdb);
}
return TDB_SUCCESS;
const void *tdb_access_read(struct tdb_context *tdb,
tdb_off_t off, tdb_len_t len, bool convert)
{
- const void *ret = NULL;
+ void *ret = NULL;
- if (likely(!(tdb->flags & TDB_CONVERT)))
- ret = tdb->methods->direct(tdb, off, len, false);
+ if (likely(!(tdb->flags & TDB_CONVERT))) {
+ ret = tdb->tdb2.io->direct(tdb, off, len, false);
+ if (TDB_PTR_IS_ERR(ret)) {
+ return ret;
+ }
+ }
if (!ret) {
struct tdb_access_hdr *hdr;
hdr = _tdb_alloc_read(tdb, off, len, sizeof(*hdr));
- if (hdr) {
- hdr->next = tdb->access;
- tdb->access = hdr;
- ret = hdr + 1;
- if (convert)
- tdb_convert(tdb, (void *)ret, len);
+ if (TDB_PTR_IS_ERR(hdr)) {
+ return hdr;
+ }
+ hdr->next = tdb->tdb2.access;
+ tdb->tdb2.access = hdr;
+ ret = hdr + 1;
+ if (convert) {
+ tdb_convert(tdb, (void *)ret, len);
}
} else
- tdb->direct_access++;
+ tdb->tdb2.direct_access++;
return ret;
}
{
void *ret = NULL;
- if (tdb->read_only) {
+ if (tdb->flags & TDB_RDONLY) {
tdb_logerr(tdb, TDB_ERR_RDONLY, TDB_LOG_USE_ERROR,
"Write to read-only database");
- return NULL;
+ return TDB_ERR_PTR(TDB_ERR_RDONLY);
}
- if (likely(!(tdb->flags & TDB_CONVERT)))
- ret = tdb->methods->direct(tdb, off, len, true);
+ if (likely(!(tdb->flags & TDB_CONVERT))) {
+ ret = tdb->tdb2.io->direct(tdb, off, len, true);
+
+ if (TDB_PTR_IS_ERR(ret)) {
+ return ret;
+ }
+ }
if (!ret) {
struct tdb_access_hdr *hdr;
hdr = _tdb_alloc_read(tdb, off, len, sizeof(*hdr));
- if (hdr) {
- hdr->next = tdb->access;
- tdb->access = hdr;
- hdr->off = off;
- hdr->len = len;
- hdr->convert = convert;
- ret = hdr + 1;
- if (convert)
- tdb_convert(tdb, (void *)ret, len);
+ if (TDB_PTR_IS_ERR(hdr)) {
+ return hdr;
}
+ hdr->next = tdb->tdb2.access;
+ tdb->tdb2.access = 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++;
+ tdb->tdb2.direct_access++;
return ret;
}
{
struct tdb_access_hdr **hp;
- for (hp = &tdb->access; *hp; hp = &(*hp)->next) {
+ for (hp = &tdb->tdb2.access; *hp; hp = &(*hp)->next) {
if (*hp + 1 == p)
return hp;
}
*hp = hdr->next;
free(hdr);
} else
- tdb->direct_access--;
+ tdb->tdb2.direct_access--;
}
enum TDB_ERROR tdb_access_commit(struct tdb_context *tdb, void *p)
*hp = hdr->next;
free(hdr);
} else {
- tdb->direct_access--;
+ tdb->tdb2.direct_access--;
ecode = TDB_SUCCESS;
}
static void *tdb_direct(struct tdb_context *tdb, tdb_off_t off, size_t len,
bool write_mode)
{
- if (unlikely(!tdb->map_ptr))
- return NULL;
+ enum TDB_ERROR ecode;
- if (unlikely(tdb_oob(tdb, off + len, true) != TDB_SUCCESS))
+ if (unlikely(!tdb->file->map_ptr))
return NULL;
- return (char *)tdb->map_ptr + off;
+
+ ecode = tdb_oob(tdb, off + len, false);
+ if (unlikely(ecode != TDB_SUCCESS))
+ return TDB_ERR_PTR(ecode);
+ return (char *)tdb->file->map_ptr + off;
}
-void add_stat_(struct tdb_context *tdb, uint64_t *s, size_t val)
+void tdb_inc_seqnum(struct tdb_context *tdb)
{
- if ((uintptr_t)s < (uintptr_t)tdb->stats + tdb->stats->size)
- *s += val;
+ tdb_off_t seq;
+
+ if (tdb->flags & TDB_VERSION1) {
+ tdb1_increment_seqnum_nonblock(tdb);
+ return;
+ }
+
+ if (likely(!(tdb->flags & TDB_CONVERT))) {
+ int64_t *direct;
+
+ direct = tdb->tdb2.io->direct(tdb,
+ offsetof(struct tdb_header,
+ seqnum),
+ sizeof(*direct), true);
+ if (likely(direct)) {
+ /* Don't let it go negative, even briefly */
+ if (unlikely((*direct) + 1) < 0)
+ *direct = 0;
+ (*direct)++;
+ return;
+ }
+ }
+
+ seq = tdb_read_off(tdb, offsetof(struct tdb_header, seqnum));
+ if (!TDB_OFF_IS_ERR(seq)) {
+ seq++;
+ if (unlikely((int64_t)seq < 0))
+ seq = 0;
+ tdb_write_off(tdb, offsetof(struct tdb_header, seqnum), seq);
+ }
}
static const struct tdb_methods io_methods = {
*/
void tdb_io_init(struct tdb_context *tdb)
{
- tdb->methods = &io_methods;
+ tdb->tdb2.io = &io_methods;
}
projects / ccan / blobdiff