- /*
+ /*
Unix SMB/CIFS implementation.
trivial database library
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)
if (tdb->flags & TDB_NOMMAP)
return;
- tdb->map_ptr = mmap(NULL, tdb->map_size,
- PROT_READ|(tdb->read_only? 0:PROT_WRITE),
+ tdb->map_ptr = mmap(NULL, tdb->map_size, tdb->mmap_flags,
MAP_SHARED, tdb->fd, 0);
/*
*/
if (tdb->map_ptr == MAP_FAILED) {
tdb->map_ptr = NULL;
- tdb->log(tdb, TDB_DEBUG_WARNING, tdb->log_priv,
- "tdb_mmap failed for size %lld (%s)\n",
- (long long)tdb->map_size, strerror(errno));
+ tdb_logerr(tdb, TDB_SUCCESS, TDB_LOG_WARNING,
+ "tdb_mmap failed for size %lld (%s)",
+ (long long)tdb->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
+ if necessary
note that "len" is the minimum length needed for the db
*/
static int tdb_oob(struct tdb_context *tdb, tdb_off_t len, bool probe)
{
struct stat st;
+ enum TDB_ERROR ecode;
+
+ /* We can't hold pointers during this: we could unmap! */
+ assert(!tdb->direct_access
+ || (tdb->flags & TDB_NOLOCK)
+ || tdb_has_expansion_lock(tdb));
+
if (len <= tdb->map_size)
return 0;
if (tdb->flags & TDB_INTERNAL) {
if (!probe) {
- /* Ensure ecode is set for log fn. */
- tdb->ecode = TDB_ERR_IO;
- tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
+ tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
"tdb_oob len %lld beyond internal"
- " malloc size %lld\n",
+ " malloc size %lld",
(long long)len,
(long long)tdb->map_size);
}
return -1;
}
- if (fstat(tdb->fd, &st) == -1) {
- tdb->ecode = TDB_ERR_IO;
+ ecode = tdb_lock_expand(tdb, F_RDLCK);
+ if (ecode != TDB_SUCCESS) {
+ tdb->ecode = ecode;
+ return -1;
+ }
+
+ if (fstat(tdb->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);
return -1;
}
+ tdb_unlock_expand(tdb, F_RDLCK);
+
if (st.st_size < (size_t)len) {
if (!probe) {
- /* Ensure ecode is set for log fn. */
- tdb->ecode = TDB_ERR_IO;
- tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
- "tdb_oob len %lld beyond eof at %lld\n",
- (long long)len, (long long)st.st_size);
+ tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
+ "tdb_oob len %zu beyond eof at %zu",
+ (size_t)len, st.st_size);
}
return -1;
}
/* Unmap, update size, remap */
tdb_munmap(tdb);
+
tdb->map_size = st.st_size;
tdb_mmap(tdb);
return 0;
}
-static void *tdb_direct(struct tdb_context *tdb, tdb_off_t off, size_t len)
-{
- if (unlikely(!tdb->map_ptr))
- return NULL;
-
- /* FIXME: We can do a subset of this! */
- if (tdb->transaction)
- return NULL;
-
- if (unlikely(tdb_oob(tdb, off + len, true) == -1))
- return NULL;
- return (char *)tdb->map_ptr + off;
-}
-
-/* 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 */
-void *tdb_get(struct tdb_context *tdb, tdb_off_t off, void *pad, size_t len)
-{
- if (likely(!(tdb->flags & TDB_CONVERT))) {
- void *ret = tdb_direct(tdb, off, len);
- if (ret)
- return ret;
- }
- 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)
{
return buf;
}
-/* Return first non-zero offset in num offset array, or num. */
/* FIXME: Return the off? */
-uint64_t tdb_find_nonzero_off(struct tdb_context *tdb, tdb_off_t off,
- uint64_t num)
+uint64_t tdb_find_nonzero_off(struct tdb_context *tdb,
+ tdb_off_t base, uint64_t start, uint64_t end)
{
uint64_t i;
const uint64_t *val;
/* Zero vs non-zero is the same unconverted: minor optimization. */
- val = tdb_access_read(tdb, off, num * sizeof(tdb_off_t), false);
+ val = tdb_access_read(tdb, base + start * sizeof(tdb_off_t),
+ (end - start) * sizeof(tdb_off_t), false);
if (!val)
- return num;
+ return end;
- for (i = 0; i < num; i++) {
+ for (i = 0; i < (end - start); i++) {
if (val[i])
break;
}
tdb_access_release(tdb, val);
- return i;
+ return start + i;
}
/* Return first zero offset in num offset array, or num. */
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);
+ void *p = tdb->methods->direct(tdb, off, len, true);
+
+ assert(!tdb->read_only);
if (p) {
memset(p, 0, len);
return 0;
}
while (len) {
unsigned todo = len < sizeof(buf) ? len : sizeof(buf);
- if (tdb->methods->write(tdb, off, buf, todo) == -1)
+ if (tdb->methods->twrite(tdb, off, buf, todo) == -1)
return -1;
len -= todo;
off += todo;
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(pad));
- if (!ret) {
- return TDB_OFF_ERR;
- }
- return *ret;
-}
+ tdb_off_t ret;
-/* Even on files, we can get partial writes due to signals. */
-bool tdb_pwrite_all(int fd, const void *buf, size_t len, tdb_off_t off)
-{
- while (len) {
- ssize_t ret;
- ret = pwrite(fd, buf, len, off);
- if (ret < 0)
- return false;
- if (ret == 0) {
- errno = ENOSPC;
- return false;
- }
- 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;
+ if (likely(!(tdb->flags & TDB_CONVERT))) {
+ tdb_off_t *p = tdb->methods->direct(tdb, off, sizeof(*p),
+ false);
+ if (p)
+ return *p;
}
- 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;
+ if (tdb_read_convert(tdb, off, &ret, sizeof(ret)) == -1)
+ return TDB_OFF_ERR;
+ return ret;
}
/* write a lump of data at a specified offset */
-static int tdb_write(struct tdb_context *tdb, tdb_off_t off,
+static int tdb_write(struct tdb_context *tdb, tdb_off_t off,
const void *buf, tdb_len_t len)
{
- if (len == 0) {
- return 0;
- }
-
if (tdb->read_only) {
- tdb->ecode = TDB_ERR_RDONLY;
+ tdb_logerr(tdb, TDB_ERR_RDONLY, TDB_LOG_USE_ERROR,
+ "Write to read-only database");
return -1;
}
+ /* FIXME: Bogus optimization? */
+ if (len == 0) {
+ return 0;
+ }
+
if (tdb->methods->oob(tdb, off + len, 0) != 0)
return -1;
if (tdb->map_ptr) {
memcpy(off + (char *)tdb->map_ptr, buf, len);
} else {
- if (!tdb_pwrite_all(tdb->fd, buf, len, off)) {
- tdb->ecode = TDB_ERR_IO;
- tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
- "tdb_write failed at %llu len=%llu (%s)\n",
- off, len, strerror(errno));
+ ssize_t ret;
+ ret = pwrite(tdb->fd, buf, len, off);
+ if (ret < len) {
+ /* This shouldn't happen: we avoid sparse files. */
+ if (ret >= 0)
+ errno = ENOSPC;
+
+ tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
+ "tdb_write: %zi at %zu len=%zu (%s)",
+ ret, (size_t)off, (size_t)len,
+ strerror(errno));
return -1;
}
}
if (tdb->map_ptr) {
memcpy(buf, off + (char *)tdb->map_ptr, len);
} else {
- 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 (%s) map_size=%lld\n",
- (long long)off, (long long)len,
- strerror(errno),
- (long long)tdb->map_size);
+ ssize_t r = pread(tdb->fd, buf, len, off);
+ if (r != len) {
+ 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);
return -1;
}
}
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);
+ tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
+ "tdb_write: no memory converting"
+ " %zu bytes", len);
return -1;
}
memcpy(conv, rec, len);
- ret = tdb->methods->write(tdb, off,
- tdb_convert(tdb, conv, len), len);
+ ret = tdb->methods->twrite(tdb, off,
+ tdb_convert(tdb, conv, len), len);
free(conv);
} else
- ret = tdb->methods->write(tdb, off, rec, len);
+ ret = tdb->methods->twrite(tdb, off, rec, len);
return ret;
}
int tdb_read_convert(struct tdb_context *tdb, tdb_off_t off,
void *rec, size_t len)
{
- int ret = tdb->methods->read(tdb, off, rec, len);
+ int ret = tdb->methods->tread(tdb, off, rec, len);
tdb_convert(tdb, rec, len);
return ret;
}
int tdb_write_off(struct tdb_context *tdb, tdb_off_t off, tdb_off_t val)
{
+ if (tdb->read_only) {
+ tdb_logerr(tdb, TDB_ERR_RDONLY, TDB_LOG_USE_ERROR,
+ "Write to read-only database");
+ return -1;
+ }
+
+ if (likely(!(tdb->flags & TDB_CONVERT))) {
+ tdb_off_t *p = tdb->methods->direct(tdb, off, sizeof(*p),
+ true);
+ if (p) {
+ *p = val;
+ return 0;
+ }
+ }
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);
- 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))) {
+ buf = malloc(prefix + len ? prefix + len : 1);
+ if (!buf) {
+ tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_USE_ERROR,
+ "tdb_alloc_read malloc failed len=%zu",
+ (size_t)(prefix + len));
+ } else if (unlikely(tdb->methods->tread(tdb, offset, buf+prefix, len)
+ == -1)) {
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;
- const void *key;
- uint64_t klen, hash;
-
- r = tdb_get(tdb, off, &pad, sizeof(pad));
- if (!r)
- /* FIXME */
- return 0;
-
- klen = rec_key_length(r);
- key = tdb_access_read(tdb, off + sizeof(pad), klen, false);
- if (!key)
- return 0;
-
- hash = tdb_hash(tdb, key, klen);
- tdb_access_release(tdb, key);
- return hash;
+ return _tdb_alloc_read(tdb, offset, len, 0);
}
static int fill(struct tdb_context *tdb,
{
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");
+ ssize_t ret = pwrite(tdb->fd, buf, n, off);
+ if (ret < n) {
+ if (ret >= 0)
+ errno = ENOSPC;
+
+ tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
+ "fill failed: %zi at %zu len=%zu (%s)",
+ ret, (size_t)off, (size_t)len,
+ strerror(errno));
return -1;
}
len -= n;
char buf[8192];
if (tdb->read_only) {
- tdb->ecode = TDB_ERR_RDONLY;
+ tdb_logerr(tdb, TDB_ERR_RDONLY, TDB_LOG_USE_ERROR,
+ "Expand on read-only database");
return -1;
}
if (tdb->flags & TDB_INTERNAL) {
char *new = realloc(tdb->map_ptr, tdb->map_size + addition);
if (!new) {
- tdb->ecode = TDB_ERR_OOM;
+ tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
+ "No memory to expand database");
return -1;
}
tdb->map_ptr = new;
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)
+ if (0 || fill(tdb, buf, sizeof(buf), tdb->map_size, addition) == -1)
return -1;
tdb->map_size += addition;
tdb_mmap(tdb);
const void *ret = NULL;
if (likely(!(tdb->flags & TDB_CONVERT)))
- ret = tdb_direct(tdb, off, len);
+ ret = tdb->methods->direct(tdb, off, len, false);
if (!ret) {
- ret = tdb_alloc_read(tdb, off, len);
- if (convert)
- tdb_convert(tdb, (void *)ret, len);
- }
- return 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);
+ }
+ } else
+ tdb->direct_access++;
-void tdb_access_release(struct tdb_context *tdb, const void *p)
-{
- if (!tdb->map_ptr
- || (char *)p < (char *)tdb->map_ptr
- || (char *)p >= (char *)tdb->map_ptr + tdb->map_size)
- free((void *)p);
+ return ret;
}
-#if 0
-/* 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)
+void *tdb_access_write(struct tdb_context *tdb,
+ tdb_off_t off, tdb_len_t len, bool convert)
{
- if (len == 0) {
- return 0;
- }
+ void *ret = NULL;
- if (tdb->read_only || tdb->traverse_read) {
- tdb->ecode = TDB_ERR_RDONLY;
- return -1;
+ if (tdb->read_only) {
+ tdb_logerr(tdb, TDB_ERR_RDONLY, TDB_LOG_USE_ERROR,
+ "Write to read-only database");
+ return NULL;
}
- if (tdb->methods->tdb_oob(tdb, off + len, 0) != 0)
- return -1;
+ if (likely(!(tdb->flags & TDB_CONVERT)))
+ ret = tdb->methods->direct(tdb, off, len, true);
- if (tdb->map_ptr) {
- memcpy(off + (char *)tdb->map_ptr, buf, len);
- } else {
- ssize_t written = pwrite(tdb->fd, buf, len, off);
- if ((written != (ssize_t)len) && (written != -1)) {
- /* try once more */
- tdb->ecode = TDB_ERR_IO;
- TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_write: wrote only "
- "%d of %d bytes at %d, trying once more\n",
- (int)written, len, off));
- written = pwrite(tdb->fd, (const char *)buf+written,
- len-written,
- off+written);
- }
- if (written == -1) {
- /* Ensure ecode is set for log fn. */
- tdb->ecode = TDB_ERR_IO;
- TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_write failed at %d "
- "len=%d (%s)\n", off, len, strerror(errno)));
- return -1;
- } else if (written != (ssize_t)len) {
- tdb->ecode = TDB_ERR_IO;
- TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_write: failed to "
- "write %d bytes at %d in two attempts\n",
- len, off));
- return -1;
+ 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);
}
- }
- return 0;
-}
-
+ } else
+ tdb->direct_access++;
+ return ret;
+}
-/*
- do an unlocked scan of the hash table heads to find the next non-zero head. The value
- will then be confirmed with the lock held
-*/
-static void tdb_next_hash_chain(struct tdb_context *tdb, uint32_t *chain)
+static struct tdb_access_hdr **find_hdr(struct tdb_context *tdb, const void *p)
{
- uint32_t h = *chain;
- if (tdb->map_ptr) {
- for (;h < tdb->header.hash_size;h++) {
- if (0 != *(uint32_t *)(TDB_HASH_TOP(h) + (unsigned char *)tdb->map_ptr)) {
- break;
- }
- }
- } else {
- uint32_t off=0;
- for (;h < tdb->header.hash_size;h++) {
- if (tdb_ofs_read(tdb, TDB_HASH_TOP(h), &off) != 0 || off != 0) {
- break;
- }
- }
+ struct tdb_access_hdr **hp;
+
+ for (hp = &tdb->access; *hp; hp = &(*hp)->next) {
+ if (*hp + 1 == p)
+ return hp;
}
- (*chain) = h;
+ return NULL;
}
-/* read/write a tdb_off_t */
-int tdb_ofs_read(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d)
+void tdb_access_release(struct tdb_context *tdb, const void *p)
{
- return tdb->methods->tdb_read(tdb, offset, (char*)d, sizeof(*d), DOCONV());
+ struct tdb_access_hdr *hdr, **hp = find_hdr(tdb, p);
+
+ if (hp) {
+ hdr = *hp;
+ *hp = hdr->next;
+ free(hdr);
+ } else
+ tdb->direct_access--;
}
-int tdb_ofs_write(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d)
+int tdb_access_commit(struct tdb_context *tdb, void *p)
{
- tdb_off_t off = *d;
- return tdb->methods->tdb_write(tdb, offset, CONVERT(off), sizeof(*d));
-}
+ struct tdb_access_hdr *hdr, **hp = find_hdr(tdb, p);
+ int ret = 0;
+
+ if (hp) {
+ hdr = *hp;
+ if (hdr->convert)
+ ret = tdb_write_convert(tdb, hdr->off, p, hdr->len);
+ else
+ ret = tdb_write(tdb, hdr->off, p, hdr->len);
+ *hp = hdr->next;
+ free(hdr);
+ } else
+ tdb->direct_access--;
+ return ret;
+}
-/* read/write a record */
-int tdb_rec_read(struct tdb_context *tdb, tdb_off_t offset, struct tdb_record *rec)
+static void *tdb_direct(struct tdb_context *tdb, tdb_off_t off, size_t len,
+ bool write_mode)
{
- if (tdb->methods->tdb_read(tdb, offset, rec, sizeof(*rec),DOCONV()) == -1)
- return -1;
- if (TDB_BAD_MAGIC(rec)) {
- /* Ensure ecode is set for log fn. */
- tdb->ecode = TDB_ERR_CORRUPT;
- TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_rec_read bad magic 0x%x at offset=%d\n", rec->magic, offset));
- return -1;
- }
- return tdb->methods->tdb_oob(tdb, rec->next+sizeof(*rec), 0);
+ if (unlikely(!tdb->map_ptr))
+ return NULL;
+
+ if (unlikely(tdb_oob(tdb, off + len, true) == -1))
+ return NULL;
+ return (char *)tdb->map_ptr + off;
}
-int tdb_rec_write(struct tdb_context *tdb, tdb_off_t offset, struct tdb_record *rec)
+void add_stat_(struct tdb_context *tdb, uint64_t *s, size_t val)
{
- struct tdb_record r = *rec;
- return tdb->methods->tdb_write(tdb, offset, CONVERT(r), sizeof(r));
+ if ((uintptr_t)s < (uintptr_t)tdb->stats + tdb->stats->size)
+ *s += val;
}
-#endif
static const struct tdb_methods io_methods = {
tdb_read,
tdb_write,
tdb_oob,
tdb_expand_file,
+ tdb_direct,
};
/*
projects / ccan / blobdiff