X-Git-Url: https://git.ozlabs.org/?p=ccan;a=blobdiff_plain;f=ccan%2Ftdb2%2Fopen.c;h=e238d99248b6ed5813418661a9bb86f3c1f966da;hp=59b4667d1bb86320ef79f226e3de609888dc6939;hb=6f7cb26e589cea081e71c59801eae87178967861;hpb=fc44729c471bcdc7ab0410448d991a65e74cec3d
diff --git a/ccan/tdb2/open.c b/ccan/tdb2/open.c
index 59b4667d..e238d992 100644
--- a/ccan/tdb2/open.c
+++ b/ccan/tdb2/open.c
@@ -16,6 +16,7 @@
License along with this library; if not, see .
*/
#include "private.h"
+#include
#include
/* all tdbs, to detect double-opens (fcntl file don't nest!) */
@@ -95,6 +96,15 @@ static uint64_t random_number(struct tdb_context *tdb)
return ret;
}
+static void tdb2_context_init(struct tdb_context *tdb)
+{
+ /* Initialize the TDB2 fields here */
+ tdb_io_init(tdb);
+ tdb->tdb2.direct_access = 0;
+ tdb->tdb2.transaction = NULL;
+ tdb->tdb2.access = NULL;
+}
+
struct new_database {
struct tdb_header hdr;
struct tdb_freetable ftable;
@@ -125,6 +135,7 @@ static enum TDB_ERROR tdb_new_database(struct tdb_context *tdb,
newdb.hdr.recovery = 0;
newdb.hdr.features_used = newdb.hdr.features_offered = TDB_FEATURE_MASK;
newdb.hdr.seqnum = 0;
+ newdb.hdr.capabilities = 0;
memset(newdb.hdr.reserved, 0, sizeof(newdb.hdr.reserved));
/* Initial hashes are empty. */
memset(newdb.hdr.hashtable, 0, sizeof(newdb.hdr.hashtable));
@@ -195,6 +206,7 @@ static enum TDB_ERROR tdb_new_file(struct tdb_context *tdb)
tdb->file->lockrecs = NULL;
tdb->file->allrecord_lock.count = 0;
tdb->file->refcnt = 1;
+ tdb->file->map_ptr = NULL;
return TDB_SUCCESS;
}
@@ -347,6 +359,71 @@ void tdb_unset_attribute(struct tdb_context *tdb,
}
}
+static bool is_tdb1(struct tdb1_header *hdr, const void *buf, ssize_t rlen)
+{
+ /* This code assumes we've tried to read entire tdb1 header. */
+ BUILD_ASSERT(sizeof(*hdr) <= sizeof(struct tdb_header));
+
+ if (rlen < (ssize_t)sizeof(*hdr)) {
+ return false;
+ }
+
+ memcpy(hdr, buf, sizeof(*hdr));
+ if (strcmp(hdr->magic_food, TDB_MAGIC_FOOD) != 0)
+ return false;
+
+ return hdr->version == TDB1_VERSION
+ || hdr->version == TDB1_BYTEREV(TDB1_VERSION);
+}
+
+/* The top three bits of the capability tell us whether it matters. */
+enum TDB_ERROR unknown_capability(struct tdb_context *tdb, const char *caller,
+ tdb_off_t type)
+{
+ if (type & TDB_CAP_NOOPEN) {
+ return tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
+ "%s: file has unknown capability %llu",
+ caller, type & TDB_CAP_NOOPEN);
+ }
+
+ if ((type & TDB_CAP_NOWRITE) && !(tdb->flags & TDB_RDONLY)) {
+ return tdb_logerr(tdb, TDB_ERR_RDONLY, TDB_LOG_ERROR,
+ "%s: file has unknown capability %llu"
+ " (cannot write to it)",
+ caller, type & TDB_CAP_NOOPEN);
+ }
+
+ if (type & TDB_CAP_NOCHECK) {
+ tdb->flags |= TDB_CANT_CHECK;
+ }
+ return TDB_SUCCESS;
+}
+
+static enum TDB_ERROR capabilities_ok(struct tdb_context *tdb,
+ tdb_off_t capabilities)
+{
+ tdb_off_t off, next;
+ enum TDB_ERROR ecode = TDB_SUCCESS;
+ const struct tdb_capability *cap;
+
+ /* Check capability list. */
+ for (off = capabilities; off && ecode == TDB_SUCCESS; off = next) {
+ cap = tdb_access_read(tdb, off, sizeof(*cap), true);
+ if (TDB_PTR_IS_ERR(cap)) {
+ return TDB_PTR_ERR(cap);
+ }
+
+ switch (cap->type & TDB_CAP_TYPE_MASK) {
+ /* We don't understand any capabilities (yet). */
+ default:
+ ecode = unknown_capability(tdb, "tdb_open", cap->type);
+ }
+ next = cap->next;
+ tdb_access_release(tdb, cap);
+ }
+ return ecode;
+}
+
struct tdb_context *tdb_open(const char *name, int tdb_flags,
int open_flags, mode_t mode,
union tdb_attribute *attr)
@@ -360,6 +437,7 @@ struct tdb_context *tdb_open(const char *name, int tdb_flags,
struct tdb_header hdr;
struct tdb_attribute_seed *seed = NULL;
struct tdb_attribute_tdb1_hashsize *hsize_attr = NULL;
+ struct tdb_attribute_tdb1_max_dead *maxsize_attr = NULL;
tdb_bool_err berr;
enum TDB_ERROR ecode;
int openlock;
@@ -388,10 +466,6 @@ struct tdb_context *tdb_open(const char *name, int tdb_flags,
memset(&tdb->stats, 0, sizeof(tdb->stats));
tdb->stats.base.attr = TDB_ATTRIBUTE_STATS;
tdb->stats.size = sizeof(tdb->stats);
- tdb_io_init(tdb);
- tdb->tdb2.direct_access = 0;
- tdb->tdb2.transaction = NULL;
- tdb->tdb2.access = NULL;
while (attr) {
switch (attr->base.attr) {
@@ -409,6 +483,9 @@ struct tdb_context *tdb_open(const char *name, int tdb_flags,
case TDB_ATTRIBUTE_TDB1_HASHSIZE:
hsize_attr = &attr->tdb1_hashsize;
break;
+ case TDB_ATTRIBUTE_TDB1_MAX_DEAD:
+ maxsize_attr = &attr->tdb1_max_dead;
+ break;
default:
/* These are set as normal. */
ecode = tdb_set_attribute(tdb, attr);
@@ -420,7 +497,7 @@ struct tdb_context *tdb_open(const char *name, int tdb_flags,
if (tdb_flags & ~(TDB_INTERNAL | TDB_NOLOCK | TDB_NOMMAP | TDB_CONVERT
| TDB_NOSYNC | TDB_SEQNUM | TDB_ALLOW_NESTING
- | TDB_RDONLY)) {
+ | TDB_RDONLY | TDB_VERSION1)) {
ecode = tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
"tdb_open: unknown flags %u", tdb_flags);
goto fail;
@@ -486,13 +563,21 @@ struct tdb_context *tdb_open(const char *name, int tdb_flags,
goto fail;
}
tdb->file->fd = -1;
- ecode = tdb_new_database(tdb, seed, &hdr);
+ if (tdb->flags & TDB_VERSION1)
+ ecode = tdb1_new_database(tdb, hsize_attr, maxsize_attr);
+ else {
+ ecode = tdb_new_database(tdb, seed, &hdr);
+ if (ecode == TDB_SUCCESS) {
+ tdb_convert(tdb, &hdr.hash_seed,
+ sizeof(hdr.hash_seed));
+ tdb->hash_seed = hdr.hash_seed;
+ tdb2_context_init(tdb);
+ tdb_ftable_init(tdb);
+ }
+ }
if (ecode != TDB_SUCCESS) {
goto fail;
}
- tdb_convert(tdb, &hdr.hash_seed, sizeof(hdr.hash_seed));
- tdb->hash_seed = hdr.hash_seed;
- tdb_ftable_init(tdb);
return tdb;
}
@@ -534,7 +619,7 @@ struct tdb_context *tdb_open(const char *name, int tdb_flags,
tdb->file->device = st.st_dev;
tdb->file->inode = st.st_ino;
tdb->file->map_ptr = NULL;
- tdb->file->map_size = sizeof(struct tdb_header);
+ tdb->file->map_size = 0;
}
/* ensure there is only one process initialising at once */
@@ -558,6 +643,12 @@ struct tdb_context *tdb_open(const char *name, int tdb_flags,
/* If they used O_TRUNC, read will return 0. */
rlen = pread(tdb->file->fd, &hdr, sizeof(hdr), 0);
if (rlen == 0 && (open_flags & O_CREAT)) {
+ if (tdb->flags & TDB_VERSION1) {
+ ecode = tdb1_new_database(tdb, hsize_attr, maxsize_attr);
+ if (ecode != TDB_SUCCESS)
+ goto fail;
+ goto finished;
+ }
ecode = tdb_new_database(tdb, seed, &hdr);
if (ecode != TDB_SUCCESS) {
goto fail;
@@ -569,6 +660,12 @@ struct tdb_context *tdb_open(const char *name, int tdb_flags,
goto fail;
} else if (rlen < sizeof(hdr)
|| strcmp(hdr.magic_food, TDB_MAGIC_FOOD) != 0) {
+ if (is_tdb1(&tdb->tdb1.header, &hdr, rlen)) {
+ ecode = tdb1_open(tdb, maxsize_attr);
+ if (!ecode)
+ goto finished;
+ goto fail;
+ }
ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
"tdb_open: %s is not a tdb file", name);
goto fail;
@@ -578,6 +675,12 @@ struct tdb_context *tdb_open(const char *name, int tdb_flags,
if (hdr.version == bswap_64(TDB_VERSION))
tdb->flags |= TDB_CONVERT;
else {
+ if (is_tdb1(&tdb->tdb1.header, &hdr, rlen)) {
+ ecode = tdb1_open(tdb, maxsize_attr);
+ if (!ecode)
+ goto finished;
+ goto fail;
+ }
/* wrong version */
ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
"tdb_open:"
@@ -593,6 +696,13 @@ struct tdb_context *tdb_open(const char *name, int tdb_flags,
goto fail;
}
+ /* This is a version2 tdb. */
+ if (tdb->flags & TDB_VERSION1) {
+ tdb->flags &= ~TDB_VERSION1;
+ }
+
+ tdb2_context_init(tdb);
+
tdb_convert(tdb, &hdr, sizeof(hdr));
tdb->hash_seed = hdr.hash_seed;
hash_test = TDB_HASH_MAGIC;
@@ -606,6 +716,11 @@ struct tdb_context *tdb_open(const char *name, int tdb_flags,
goto fail;
}
+ ecode = capabilities_ok(tdb, hdr.capabilities);
+ if (ecode != TDB_SUCCESS) {
+ goto fail;
+ }
+
/* Clear any features we don't understand. */
if ((open_flags & O_ACCMODE) != O_RDONLY) {
hdr.features_used &= TDB_FEATURE_MASK;
@@ -617,49 +732,64 @@ struct tdb_context *tdb_open(const char *name, int tdb_flags,
goto fail;
}
+finished:
+ if (tdb->flags & TDB_VERSION1) {
+ /* if needed, run recovery */
+ if (tdb1_transaction_recover(tdb) == -1) {
+ ecode = tdb->last_error;
+ goto fail;
+ }
+ }
+
tdb_unlock_open(tdb, openlock);
- /* This make sure we have current map_size and mmap. */
- ecode = tdb->tdb2.io->oob(tdb, tdb->file->map_size + 1, true);
+ /* This makes sure we have current map_size and mmap. */
+ if (tdb->flags & TDB_VERSION1) {
+ ecode = tdb1_probe_length(tdb);
+ } else {
+ ecode = tdb->tdb2.io->oob(tdb, tdb->file->map_size, 1, true);
+ }
if (unlikely(ecode != TDB_SUCCESS))
goto fail;
- /* Now it's fully formed, recover if necessary. */
- berr = tdb_needs_recovery(tdb);
- if (unlikely(berr != false)) {
- if (berr < 0) {
- ecode = berr;
- goto fail;
+ if (!(tdb->flags & TDB_VERSION1)) {
+ /* Now it's fully formed, recover if necessary. */
+ berr = tdb_needs_recovery(tdb);
+ if (unlikely(berr != false)) {
+ if (berr < 0) {
+ ecode = TDB_OFF_TO_ERR(berr);
+ goto fail;
+ }
+ ecode = tdb_lock_and_recover(tdb);
+ if (ecode != TDB_SUCCESS) {
+ goto fail;
+ }
}
- ecode = tdb_lock_and_recover(tdb);
+
+ ecode = tdb_ftable_init(tdb);
if (ecode != TDB_SUCCESS) {
goto fail;
}
}
- ecode = tdb_ftable_init(tdb);
- if (ecode != TDB_SUCCESS) {
- goto fail;
- }
-
tdb->next = tdbs;
tdbs = tdb;
return tdb;
fail:
/* Map ecode to some logical errno. */
- switch (ecode) {
- case TDB_ERR_CORRUPT:
- case TDB_ERR_IO:
+ switch (TDB_ERR_TO_OFF(ecode)) {
+ case TDB_ERR_TO_OFF(TDB_ERR_CORRUPT):
+ case TDB_ERR_TO_OFF(TDB_ERR_IO):
saved_errno = EIO;
break;
- case TDB_ERR_LOCK:
+ case TDB_ERR_TO_OFF(TDB_ERR_LOCK):
saved_errno = EWOULDBLOCK;
break;
- case TDB_ERR_OOM:
+ case TDB_ERR_TO_OFF(TDB_ERR_OOM):
saved_errno = ENOMEM;
break;
- case TDB_ERR_EINVAL:
+ case TDB_ERR_TO_OFF(TDB_ERR_EINVAL):
saved_errno = EINVAL;
break;
default:
@@ -702,8 +832,14 @@ int tdb_close(struct tdb_context *tdb)
tdb_trace(tdb, "tdb_close");
- if (tdb->tdb2.transaction) {
- tdb_transaction_cancel(tdb);
+ if (tdb->flags & TDB_VERSION1) {
+ if (tdb->tdb1.transaction) {
+ tdb1_transaction_cancel(tdb);
+ }
+ } else {
+ if (tdb->tdb2.transaction) {
+ tdb_transaction_cancel(tdb);
+ }
}
if (tdb->file->map_ptr) {