X-Git-Url: http://git.ozlabs.org/?p=ccan;a=blobdiff_plain;f=ccan%2Ftdb2%2Fio.c;h=aa79def1e38cc5b49b500d1b513eb9dc446fbbb7;hp=cdee88aa1fd57321220d121a5bfe381332dc9269;hb=1fe7f55b1efc5deefb20815a03b56c8ef0f6dc53;hpb=06e0037d97f5e1d83667ec40627cef862f3b7b85

diff --git a/ccan/tdb2/io.c b/ccan/tdb2/io.c
index cdee88aa..aa79def1 100644
--- a/ccan/tdb2/io.c
+++ b/ccan/tdb2/io.c
@@ -26,6 +26,7 @@
    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)
@@ -47,8 +48,7 @@ void tdb_mmap(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);
 
 	/*
@@ -70,6 +70,11 @@ void tdb_mmap(struct tdb_context *tdb)
 static int tdb_oob(struct tdb_context *tdb, tdb_off_t len, bool probe)
 {
 	struct stat st;
+	int ret;
+
+	/* We can't hold pointers during this: we could unmap! */
+	assert(!tdb->direct_access || tdb_has_expansion_lock(tdb));
+
 	if (len <= tdb->map_size)
 		return 0;
 	if (tdb->flags & TDB_INTERNAL) {
@@ -85,7 +90,14 @@ static int tdb_oob(struct tdb_context *tdb, tdb_off_t len, bool probe)
 		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;
 	}
@@ -103,6 +115,7 @@ static int tdb_oob(struct tdb_context *tdb, tdb_off_t len, bool probe)
 
 	/* Unmap, update size, remap */
 	tdb_munmap(tdb);
+
 	tdb->map_size = st.st_size;
 	tdb_mmap(tdb);
 	return 0;
@@ -132,19 +145,13 @@ void *tdb_get(struct tdb_context *tdb, tdb_off_t off, void *pad, size_t len)
 		if (ret)
 			return ret;
 	}
-
-	if (unlikely(tdb_oob(tdb, off + len, false) == -1))
-		return NULL;
-
-	if (tdb->methods->read(tdb, off, pad, len) == -1)
-		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]);
@@ -152,84 +159,63 @@ 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 i, *val;
-	bool alloc = false;
-
-	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;
-	}
+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;
 
-	for (i = 0; i < num; i++) {
+	/* 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;
+
+	for (i = 0; i < (end - start); i++) {
 		if (val[i])
 			break;
 	}
-	if (unlikely(alloc))
-		free(val);
-	return i;
+	tdb_access_release(tdb, val);
+	return start + i;
 }
 
 /* Return first zero offset in num offset array, or num. */
 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), off, len);
 	}
+	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)
@@ -357,9 +343,26 @@ static int tdb_read(struct tdb_context *tdb, tdb_off_t off, void *buf,
 }
 
 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,
@@ -375,72 +378,48 @@ int tdb_write_off(struct tdb_context *tdb, tdb_off_t off, tdb_off_t val)
 	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(pad));
-	if (!r)
-		/* FIXME */
-		return 0;
+	return _tdb_alloc_read(tdb, offset, len, 0);
+}
 
-	klen = rec_key_length(r);
-	key = tdb_direct(tdb, off + sizeof(pad), 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(pad), 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
@@ -483,13 +462,57 @@ static int tdb_expand_file(struct tdb_context *tdb, tdb_len_t addition)
 	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;
 }
 
@@ -498,7 +521,30 @@ 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);
+		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