chunkd/be-fs.c | 145 +++++++++++++++++++++++++++++++++++++++++++++++++-------
chunkd/chunkd.h | 3 +
2 files changed, 131 insertions(+), 17 deletions(-)
commit 394109d5c2fc2d15d91c2d36eecd57594922c1b3
Author: Jeff Garzik <[email protected]>
Date: Sat Jul 17 01:05:15 2010 -0400
chunkd: Add checksum table to on-disk format, one sum per 64k of data
Signed-off-by: Jeff Garzik <[email protected]>
diff --git a/chunkd/be-fs.c b/chunkd/be-fs.c
index 0a81134..5955afa 100644
--- a/chunkd/be-fs.c
+++ b/chunkd/be-fs.c
@@ -49,14 +49,23 @@ struct fs_obj {
int in_fd;
char *in_fn;
+
+ size_t checked_bytes;
+ SHA_CTX checksum;
+ unsigned int csum_idx;
+ void *csum_tbl;
+ size_t csum_tbl_sz;
+
+ unsigned int n_blk;
};
struct be_fs_obj_hdr {
char magic[4];
uint32_t key_len;
uint64_t value_len;
+ uint32_t n_blk;
- char reserved[16];
+ char reserved[12];
unsigned char hash[CHD_CSUM_SZ];
char owner[128];
@@ -204,6 +213,8 @@ static struct fs_obj *fs_obj_alloc(void)
obj->out_fd = -1;
obj->in_fd = -1;
+ SHA1_Init(&obj->checksum);
+
return obj;
}
@@ -314,6 +325,17 @@ static bool key_valid(const void *key, size_t key_len)
return true;
}
+static unsigned int fs_blk_count(uint64_t data_len)
+{
+ uint64_t n_blk;
+
+ n_blk = data_len >> CHUNK_BLK_ORDER;
+ if (data_len & (CHUNK_BLK_SZ - 1))
+ n_blk++;
+
+ return (unsigned int) n_blk;
+}
+
struct backend_obj *fs_obj_new(uint32_t table_id,
const void *key, size_t key_len,
uint64_t data_len,
@@ -321,6 +343,7 @@ struct backend_obj *fs_obj_new(uint32_t table_id,
{
struct fs_obj *obj;
char *fn = NULL;
+ size_t csum_bytes;
enum chunk_errcode erc = che_InternalError;
off_t skip_len;
@@ -335,6 +358,13 @@ struct backend_obj *fs_obj_new(uint32_t table_id,
return NULL;
}
+ obj->n_blk = fs_blk_count(data_len);
+ csum_bytes = obj->n_blk * CHD_CSUM_SZ;
+ obj->csum_tbl = malloc(csum_bytes);
+ if (!obj->csum_tbl)
+ goto err_out;
+ obj->csum_tbl_sz = csum_bytes;
+
/* build local fs pathname */
fn = fs_obj_pathname(table_id, key, key_len);
if (!fn)
@@ -355,7 +385,7 @@ struct backend_obj *fs_obj_new(uint32_t table_id,
obj->out_fn = fn;
/* calculate size of front-of-file metadata area */
- skip_len = sizeof(struct be_fs_obj_hdr) + key_len;
+ skip_len = sizeof(struct be_fs_obj_hdr) + key_len + csum_bytes;
/* position file pointer where object data (as in, not metadata)
* will begin
@@ -393,7 +423,10 @@ struct backend_obj *fs_obj_open(uint32_t table_id, const
char *user,
struct be_fs_obj_hdr hdr;
ssize_t rrc;
uint64_t value_len, tmp64;
+ size_t csum_bytes;
enum chunk_errcode erc = che_InternalError;
+ struct iovec iov[2];
+ size_t total_rd_len;
if (!key_valid(key, key_len)) {
*err_code = che_InvalidKey;
@@ -453,23 +486,45 @@ struct backend_obj *fs_obj_open(uint32_t table_id, const
char *user,
goto err_out;
value_len = GUINT64_FROM_LE(hdr.value_len);
+ obj->n_blk = GUINT32_FROM_LE(hdr.n_blk);
+ csum_bytes = obj->n_blk * CHD_CSUM_SZ;
/* verify file size large enough to contain value */
- tmp64 = value_len + sizeof(hdr) + key_len;
+ tmp64 = value_len + sizeof(hdr) + key_len + csum_bytes;
if (G_UNLIKELY(st.st_size < tmp64)) {
applog(LOG_ERR, "obj(%s) size error, too small", obj->in_fn);
goto err_out;
}
+ /* verify expected size of checksum table */
+ if (G_UNLIKELY(fs_blk_count(value_len) != obj->n_blk)) {
+ applog(LOG_ERR, "obj(%s) unexpected blk count "
+ "(%u from val sz, %u from hdr)",
+ obj->in_fn, fs_blk_count(value_len), obj->n_blk);
+ goto err_out;
+ }
+
+ obj->csum_tbl = malloc(csum_bytes);
+ if (!obj->csum_tbl)
+ goto err_out;
+ obj->csum_tbl_sz = csum_bytes;
+
obj->bo.key = malloc(key_len);
obj->bo.key_len = key_len;
if (!obj->bo.key)
goto err_out;
- /* read object variable-length header */
- rrc = read(obj->in_fd, obj->bo.key, key_len);
- if ((rrc != key_len) || (memcmp(key, obj->bo.key, key_len))) {
- applog(LOG_ERR, "read hdr key obj(%s) failed: %s",
+ /* init additional header segment list */
+ iov[0].iov_base = obj->bo.key;
+ iov[0].iov_len = key_len;
+ iov[1].iov_base = obj->csum_tbl;
+ iov[1].iov_len = csum_bytes;
+ total_rd_len = iov[0].iov_len + iov[1].iov_len;
+
+ /* read additional header segments (key, checksum table) */
+ rrc = readv(obj->in_fd, iov, ARRAY_SIZE(iov));
+ if ((rrc != total_rd_len) || (memcmp(key, obj->bo.key, key_len))) {
+ applog(LOG_ERR, "read addnl hdrs(%s) failed: %s",
obj->in_fn,
(rrc < 0) ? strerror(errno) : "<unknown reasons>");
goto err_out;
@@ -512,6 +567,7 @@ void fs_obj_free(struct backend_obj *bo)
if (obj->in_fd >= 0)
close(obj->in_fd);
+ free(obj->csum_tbl);
free(obj);
}
@@ -528,19 +584,59 @@ ssize_t fs_obj_read(struct backend_obj *bo, void *ptr,
size_t len)
return rc;
}
+static void obj_flush_csum(struct backend_obj *bo)
+{
+ struct fs_obj *obj = bo->private;
+ unsigned char md[CHD_CSUM_SZ];
+
+ if (G_UNLIKELY(obj->csum_idx >= obj->n_blk)) {
+ applog(LOG_ERR, "BUG %s: cidx %u, n_blk %u",
+ __func__, obj->csum_idx, obj->n_blk);
+ return;
+ }
+
+ SHA1_Final(md, &obj->checksum);
+
+ memcpy(obj->csum_tbl + ((obj->csum_idx++) * CHD_CSUM_SZ),
+ md, CHD_CSUM_SZ);
+
+ obj->checked_bytes = 0;
+ SHA1_Init(&obj->checksum);
+}
+
ssize_t fs_obj_write(struct backend_obj *bo, const void *ptr, size_t len)
{
struct fs_obj *obj = bo->private;
- ssize_t rc;
+ ssize_t total_written = 0;
- rc = write(obj->out_fd, ptr, len);
- if (rc < 0)
- applog(LOG_ERR, "obj write(%s) failed: %s",
- obj->out_fn, strerror(errno));
- else
- obj->written_bytes += rc;
+ while (len > 0) {
+ size_t unchecked;
+ ssize_t wrc;
- return rc;
+ unchecked = CHUNK_BLK_SZ - obj->checked_bytes;
+
+ wrc = write(obj->out_fd, ptr, MIN(unchecked, len));
+ if (wrc < 0) {
+ applog(LOG_ERR, "obj write(%s) failed: %s",
+ obj->out_fn, strerror(errno));
+ return wrc;
+ break;
+ }
+
+ SHA1_Update(&obj->checksum, ptr, wrc);
+
+ total_written += wrc;
+ obj->written_bytes += wrc;
+ obj->checked_bytes += wrc;
+ ptr += wrc;
+ len -= wrc;
+
+ /* if at end of 64k block, update csum table with new csum */
+ if (obj->checked_bytes == CHUNK_BLK_SZ)
+ obj_flush_csum(bo);
+ }
+
+ return total_written;
}
bool fs_obj_write_commit(struct backend_obj *bo, const char *user,
@@ -550,7 +646,7 @@ bool fs_obj_write_commit(struct backend_obj *bo, const char
*user,
struct be_fs_obj_hdr hdr;
ssize_t wrc;
size_t total_wr_len;
- struct iovec iov[2];
+ struct iovec iov[3];
if (G_UNLIKELY(obj->bo.size != obj->written_bytes)) {
applog(LOG_ERR, "BUG(%s): size/written_bytes mismatch:
%llu/%llu",
@@ -566,6 +662,19 @@ bool fs_obj_write_commit(struct backend_obj *bo, const
char *user,
strncpy(hdr.owner, user, sizeof(hdr.owner));
hdr.key_len = GUINT32_TO_LE(bo->key_len);
hdr.value_len = GUINT64_TO_LE(obj->written_bytes);
+ hdr.n_blk = GUINT32_TO_LE(obj->n_blk);
+
+ /* update checksum table with final csum, if necessary */
+ if (obj->checked_bytes > 0)
+ obj_flush_csum(bo);
+
+ if (G_UNLIKELY(obj->csum_idx != obj->n_blk)) {
+ applog(LOG_ERR, "BUG(%s): csum_idx/n_blk mismatch: %u/%u",
+ obj->out_fn, obj->csum_idx, obj->n_blk);
+ return false;
+ }
+
+ obj->csum_idx = 0;
/* go back to beginning of file */
if (lseek(obj->out_fd, 0, SEEK_SET) < 0) {
@@ -579,7 +688,9 @@ bool fs_obj_write_commit(struct backend_obj *bo, const char
*user,
iov[0].iov_len = sizeof(hdr);
iov[1].iov_base = bo->key;
iov[1].iov_len = bo->key_len;
- total_wr_len = iov[0].iov_len + iov[1].iov_len;
+ iov[2].iov_base = obj->csum_tbl;
+ iov[2].iov_len = obj->csum_tbl_sz;
+ total_wr_len = iov[0].iov_len + iov[1].iov_len + iov[2].iov_len;
/* write object header segments */
wrc = writev(obj->out_fd, iov, ARRAY_SIZE(iov));
diff --git a/chunkd/chunkd.h b/chunkd/chunkd.h
index 716704b..5c34967 100644
--- a/chunkd/chunkd.h
+++ b/chunkd/chunkd.h
@@ -36,6 +36,9 @@
#endif
enum {
+ CHUNK_BLK_ORDER = 16, /* 64k blocks */
+ CHUNK_BLK_SZ = 1 << CHUNK_BLK_ORDER,
+
CLI_DATA_BUF_SZ = 16 * 1024,
CHD_TRASH_MAX = 1000,
--
To unsubscribe from this list: send the line "unsubscribe hail-devel" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
