/* skip up to length bytes in a chunkqueue, return number of bytes skipped */
goffset li_chunkqueue_skip(liChunkQueue *cq, goffset length) {
liChunk *c;
goffset bytes = 0;
goffset we_have;
while ( (NULL != (c = li_chunkqueue_first_chunk(cq))) && (0 == (we_have = li_chunk_length(c)) || length > 0) ) {
if (we_have <= length) {
/* skip (delete) complete chunk */
if (c->type == STRING_CHUNK) cqlimit_update(cq, - (goffset)c->data.str->len);
else if (c->type == MEM_CHUNK) cqlimit_update(cq, - (goffset)c->mem->len);
else if (c->type == BUFFER_CHUNK) cqlimit_update(cq, - (goffset)c->data.buffer.length);
chunk_free(cq, c);
bytes += we_have;
length -= we_have;
} else { /* skip first part of a chunk */
c->offset += length;
bytes += length;
length = 0;
}
}
cq->bytes_out += bytes;
cq->length -= bytes;
return bytes;
}
/* steal the first chunk from in and append it to out, return number of bytes stolen */
goffset li_chunkqueue_steal_chunk(liChunkQueue *out, liChunkQueue *in) {
liChunk *c;
goffset length;
GList *l = g_queue_pop_head_link(&in->queue);
if (!l) return 0;
g_queue_push_tail_link(&out->queue, l);
c = (liChunk*) l->data;
length = li_chunk_length(c);
in->bytes_out += length;
in->length -= length;
out->bytes_in += length;
out->length += length;
if (in->limit != out->limit) {
if (c->type == STRING_CHUNK) {
cqlimit_update(out, c->data.str->len);
cqlimit_update(in, - (goffset)c->data.str->len);
} else if (c->type == MEM_CHUNK) {
cqlimit_update(out, c->mem->len);
cqlimit_update(in, - (goffset)c->mem->len);
} else if (c->type == BUFFER_CHUNK) {
cqlimit_update(out, c->data.buffer.length);
cqlimit_update(in, - (goffset)c->data.buffer.length);
}
}
return length;
}
/* first chunk must be a STRING_CHUNK ! */
liNetworkStatus li_network_backend_writev(int fd, liChunkQueue *cq, goffset *write_max, GError **err) {
off_t we_have;
ssize_t r;
gboolean did_write_something = FALSE;
liChunkIter ci;
liChunk *c;
liNetworkStatus res = LI_NETWORK_STATUS_FATAL_ERROR;
GArray *chunks = g_array_sized_new(FALSE, TRUE, sizeof(struct iovec), UIO_MAXIOV);
if (0 == cq->length) goto cleanup; /* FATAL ERROR */
do {
ci = li_chunkqueue_iter(cq);
if (STRING_CHUNK != (c = li_chunkiter_chunk(ci))->type && MEM_CHUNK != c->type && BUFFER_CHUNK != c->type) {
res = did_write_something ? LI_NETWORK_STATUS_SUCCESS : LI_NETWORK_STATUS_FATAL_ERROR;
goto cleanup;
}
we_have = 0;
do {
guint i = chunks->len;
off_t len = li_chunk_length(c);
struct iovec *v;
g_array_set_size(chunks, i + 1);
v = &g_array_index(chunks, struct iovec, i);
if (c->type == STRING_CHUNK) {
v->iov_base = c->data.str->str + c->offset;
} else if (c->type == MEM_CHUNK) {
v->iov_base = c->mem->data + c->offset;
} else { /* if (c->type == BUFFER_CHUNK) */
v->iov_base = c->data.buffer.buffer->addr + c->data.buffer.offset + c->offset;
}
if (len > *write_max - we_have) len = *write_max - we_have;
v->iov_len = len;
we_have += len;
} while (we_have < *write_max &&
li_chunkiter_next(&ci) &&
(STRING_CHUNK == (c = li_chunkiter_chunk(ci))->type || MEM_CHUNK == c->type || BUFFER_CHUNK == c->type) &&
chunks->len < UIO_MAXIOV);
while (-1 == (r = writev(fd, (struct iovec*) chunks->data, chunks->len))) {
switch (errno) {
case EAGAIN:
#if EWOULDBLOCK != EAGAIN
case EWOULDBLOCK:
#endif
res = LI_NETWORK_STATUS_WAIT_FOR_EVENT;
goto cleanup;
case ECONNRESET:
case EPIPE:
case ETIMEDOUT:
res = LI_NETWORK_STATUS_CONNECTION_CLOSE;
goto cleanup;
case EINTR:
break; /* try again */
default:
g_set_error(err, LI_NETWORK_ERROR, 0, "li_network_backend_writev: oops, write to fd=%d failed: %s", fd, g_strerror(errno));
goto cleanup;
}
}
if (0 == r) {
res = LI_NETWORK_STATUS_WAIT_FOR_EVENT;
goto cleanup;
}
li_chunkqueue_skip(cq, r);
*write_max -= r;
if (r != we_have) {
res = LI_NETWORK_STATUS_WAIT_FOR_EVENT;
goto cleanup;
}
if (0 == cq->length) {
res = LI_NETWORK_STATUS_SUCCESS;
goto cleanup;
}
did_write_something = TRUE;
g_array_set_size(chunks, 0);
} while (*write_max > 0);
res = LI_NETWORK_STATUS_SUCCESS;
cleanup:
g_array_free(chunks, TRUE);
return res;
}
static void bod_handle_data(bod_state *state) {
liChunkQueue *out = state->stream.out;
liChunkQueue *in;
if (out->is_closed) {
li_stream_disconnect(&state->stream);
bod_close(state);
return;
}
in = (state->stream.source != NULL) ? state->stream.source->out : NULL;
if (NULL == in) goto out;
if (NULL == state->vr) {
li_chunkqueue_steal_all(out, in);
goto out;
}
while (in->length > 0) {
liChunk *c = li_chunkqueue_first_chunk(in);
liChunkIter ci;
off_t length, data_len;
char *data = NULL;
GError *err;
assert(UNUSED_CHUNK != c->type);
switch (c->type) {
case UNUSED_CHUNK:
/* shouldn't happen anyway, but stealing it is ok here too */
case FILE_CHUNK:
if (state->split_on_file_chunks) {
bod_close(state);
} else {
bod_flush(state);
}
li_chunkqueue_steal_chunk(out, in);
break;
case STRING_CHUNK:
case MEM_CHUNK:
case BUFFER_CHUNK:
if (!bod_open(state)) return;
length = li_chunk_length(c);
ci = li_chunkqueue_iter(in);
err = NULL;
if (LI_HANDLER_GO_ON != li_chunkiter_read(ci, 0, length, &data, &data_len, &err)) {
if (NULL != err) {
VR_ERROR(state->vr, "%s", err->message);
g_error_free(err);
}
bod_error(state);
return;
}
while ( data_len > 0 ) {
ssize_t r;
r = pwrite(state->tempfile->fd, data, data_len, state->write_pos);
if (r < 0) {
switch (errno) {
case EINTR: continue;
default: break;
}
VR_ERROR(state->vr, "pwrite failed: %s", g_strerror(errno));
bod_stop(state); /* write failures are not critical */
return;
}
data += r;
data_len -= r;
state->write_pos += r;
}
li_chunkqueue_skip(in, length);
break;
}
}
bod_autoflush(state);
out:
if (NULL == in || in->is_closed) {
out->is_closed = TRUE;
bod_close(state); /* close/flush ignores out->is_closed */
li_stream_notify(&state->stream); /* if no flush happened we still notify */
}
}
liHandlerResult li_filter_buffer_on_disk(liVRequest *vr, liChunkQueue *out, liChunkQueue *in, bod_state *state) {
UNUSED(vr);
if (out->is_closed) {
in->is_closed = TRUE;
li_chunkqueue_skip_all(in);
bod_close(state);
return LI_HANDLER_GO_ON;
}
while (in->length > 0) {
liChunk *c = li_chunkqueue_first_chunk(in);
liChunkIter ci;
off_t length, data_len;
char *data = NULL;
GError *err;
switch (c->type) {
case UNUSED_CHUNK: return LI_HANDLER_ERROR;
case FILE_CHUNK:
bod_flush(out, state);
if (state->split_on_file_chunks) {
bod_close(state);
}
li_chunkqueue_steal_chunk(out, in);
break;
case STRING_CHUNK:
case MEM_CHUNK:
case BUFFER_CHUNK:
if (!bod_open(vr, state)) return LI_HANDLER_ERROR;
length = li_chunk_length(c);
ci = li_chunkqueue_iter(in);
err = NULL;
if (LI_HANDLER_GO_ON != li_chunkiter_read(ci, 0, length, &data, &data_len, &err)) {
if (NULL != err) {
VR_ERROR(vr, "%s", err->message);
g_error_free(err);
}
return LI_HANDLER_ERROR;
}
while ( data_len > 0 ) {
ssize_t r;
r = pwrite(state->tempfile->fd, data, data_len, state->write_pos);
if (r < 0) {
switch (errno) {
case EINTR: continue;
default: break;
}
VR_ERROR(vr, "pwrite failed: %s", g_strerror(errno));
return LI_HANDLER_ERROR;
}
data += r;
data_len -= r;
state->write_pos += r;
}
li_chunkqueue_skip(in, length);
break;
}
}
bod_autoflush(out, state);
if (in->is_closed) {
bod_flush(out, state);
out->is_closed = TRUE;
bod_close(state);
return LI_HANDLER_GO_ON;
}
return LI_HANDLER_GO_ON;
}
/* get the data from a chunk; easy in case of a STRING_CHUNK,
* but needs to do io in case of FILE_CHUNK; the data is _not_ marked as "done"
* may return HANDLER_GO_ON, HANDLER_ERROR
*/
liHandlerResult li_chunkiter_read(liVRequest *vr, liChunkIter iter, off_t start, off_t length, char **data_start, off_t *data_len) {
liChunk *c = li_chunkiter_chunk(iter);
off_t we_have, our_start;
liHandlerResult res = LI_HANDLER_GO_ON;
if (!c) return LI_HANDLER_ERROR;
if (!data_start || !data_len) return LI_HANDLER_ERROR;
we_have = li_chunk_length(c) - start;
if (length > we_have) length = we_have;
if (length <= 0) return LI_HANDLER_ERROR;
switch (c->type) {
case UNUSED_CHUNK: return LI_HANDLER_ERROR;
case STRING_CHUNK:
*data_start = c->data.str->str + c->offset + start;
*data_len = length;
break;
case MEM_CHUNK:
*data_start = (char*) c->mem->data + c->offset + start;
*data_len = length;
break;
case FILE_CHUNK:
if (LI_HANDLER_GO_ON != (res = li_chunkfile_open(vr, c->data.file.file))) return res;
if (length > MAX_MMAP_CHUNK) length = MAX_MMAP_CHUNK;
if (!c->mem) {
c->mem = g_byte_array_sized_new(length);
} else {
g_byte_array_set_size(c->mem, length);
}
our_start = start + c->offset + c->data.file.start;
read_chunk:
if (-1 == (we_have = pread(c->data.file.file->fd, c->mem->data, length, our_start))) {
if (EINTR == errno) goto read_chunk;
VR_ERROR(vr, "pread failed for '%s' (fd = %i): %s",
GSTR_SAFE_STR(c->data.file.file->name), c->data.file.file->fd,
g_strerror(errno));
g_byte_array_free(c->mem, TRUE);
c->mem = NULL;
return LI_HANDLER_ERROR;
} else if (we_have != length) {
/* may return less than requested bytes due to signals */
/* CON_TRACE(srv, "read return unexpected number of bytes"); */
if (we_have == 0) {
VR_ERROR(vr, "pread returned 0 bytes for '%s' (fd = %i): unexpected end of file?",
GSTR_SAFE_STR(c->data.file.file->name), c->data.file.file->fd);
g_byte_array_free(c->mem, TRUE);
c->mem = NULL;
return LI_HANDLER_ERROR;
}
length = we_have;
g_byte_array_set_size(c->mem, length);
}
*data_start = (char*) c->mem->data;
*data_len = length;
break;
case BUFFER_CHUNK:
*data_start = (char*) c->data.buffer.buffer->addr + c->data.buffer.offset + c->offset + start;
*data_len = length;
break;
}
return LI_HANDLER_GO_ON;
}
/* steal up to length bytes from in and put them into out, return number of bytes stolen */
goffset li_chunkqueue_steal_len(liChunkQueue *out, liChunkQueue *in, goffset length) {
liChunk *c, *cnew;
GList* l;
goffset bytes = 0, meminbytes = 0, memoutbytes = 0;
goffset we_have;
while ( (NULL != (c = li_chunkqueue_first_chunk(in))) && length > 0 ) {
we_have = li_chunk_length(c);
if (!we_have) { /* remove empty chunks */
if (c->type == STRING_CHUNK) meminbytes -= c->data.str->len;
else if (c->type == MEM_CHUNK) meminbytes -= c->mem->len;
else if (c->type == BUFFER_CHUNK) meminbytes -= c->data.buffer.length;
chunk_free(in, c);
continue;
}
if (we_have <= length) { /* move complete chunk */
l = g_queue_pop_head_link(&in->queue);
g_queue_push_tail_link(&out->queue, l);
bytes += we_have;
if (c->type == STRING_CHUNK) {
meminbytes -= c->data.str->len;
memoutbytes += c->data.str->len;
} else if (c->type == MEM_CHUNK) {
meminbytes -= c->mem->len;
memoutbytes += c->mem->len;
} else if (c->type == BUFFER_CHUNK) {
meminbytes -= c->data.buffer.length;
memoutbytes += c->data.buffer.length;
}
length -= we_have;
} else { /* copy first part of a chunk */
cnew = chunk_new();
switch (c->type) {
case UNUSED_CHUNK: /* impossible, has length 0 */
/* remove "empty" chunks */
chunk_free(in, c);
chunk_free(NULL, cnew);
continue;
case STRING_CHUNK: /* change type to MEM_CHUNK, as we copy it anyway */
cnew->type = MEM_CHUNK;
cnew->mem = g_byte_array_sized_new(length);
g_byte_array_append(cnew->mem, (guint8*) c->data.str->str + c->offset, length);
memoutbytes += length;
break;
case MEM_CHUNK:
cnew->type = MEM_CHUNK;
cnew->mem = g_byte_array_sized_new(length);
g_byte_array_append(cnew->mem, (guint8*) c->mem->data + c->offset, length);
memoutbytes += length;
break;
case FILE_CHUNK:
cnew->type = FILE_CHUNK;
li_chunkfile_acquire(c->data.file.file);
cnew->data.file.file = c->data.file.file;
cnew->data.file.start = c->data.file.start + c->offset;
cnew->data.file.length = length;
break;
case BUFFER_CHUNK:
cnew->type = BUFFER_CHUNK;
li_buffer_acquire(c->data.buffer.buffer);
cnew->data.buffer.buffer = c->data.buffer.buffer;
cnew->data.buffer.offset = c->data.buffer.offset + c->offset;
cnew->data.buffer.length = length;
memoutbytes += length;
break;
}
c->offset += length;
bytes += length;
length = 0;
g_queue_push_tail_link(&out->queue, &cnew->cq_link);
}
}
in->bytes_out += bytes;
in->length -= bytes;
out->bytes_in += bytes;
out->length += bytes;
cqlimit_update(out, memoutbytes);
cqlimit_update(in, meminbytes);
return bytes;
}
/* same as li_chunkiter_read, but tries mmap() first and falls back to pread();
* as accessing mmap()-ed areas may result in SIGBUS, you have to handle that signal somehow.
*/
liHandlerResult li_chunkiter_read_mmap(liVRequest *vr, liChunkIter iter, off_t start, off_t length, char **data_start, off_t *data_len) {
liChunk *c = li_chunkiter_chunk(iter);
off_t we_want, we_have, our_start, our_offset;
liHandlerResult res = LI_HANDLER_GO_ON;
int mmap_errno = 0;
if (!c) return LI_HANDLER_ERROR;
if (!data_start || !data_len) return LI_HANDLER_ERROR;
we_have = li_chunk_length(c) - start;
if (length > we_have) length = we_have;
if (length <= 0) return LI_HANDLER_ERROR;
switch (c->type) {
case UNUSED_CHUNK: return LI_HANDLER_ERROR;
case STRING_CHUNK:
*data_start = c->data.str->str + c->offset + start;
*data_len = length;
break;
case MEM_CHUNK:
*data_start = (char*) c->mem->data + c->offset + start;
*data_len = length;
break;
case FILE_CHUNK:
if (LI_HANDLER_GO_ON != (res = li_chunkfile_open(vr, c->data.file.file))) return res;
if (length > MAX_MMAP_CHUNK) length = MAX_MMAP_CHUNK;
if ( !(c->data.file.mmap.data != MAP_FAILED || c->mem) /* no data present */
|| !( /* or in the wrong range */
(start + c->offset + c->data.file.start >= c->data.file.mmap.offset)
&& (start + c->offset + c->data.file.start + length <= c->data.file.mmap.offset + (ssize_t) c->data.file.mmap.length)) ) {
/* then find new range */
our_start = start + c->offset + c->data.file.start; /* "start" maps to this offset in the file */
our_offset = our_start % MMAP_CHUNK_ALIGN; /* offset for "start" in new mmap block */
our_start -= our_offset; /* file offset for mmap */
we_want = length + MAX_MMAP_CHUNK;
if (we_want > we_have) we_want = we_have;
we_want += our_offset;
if (MAP_FAILED != c->data.file.mmap.data) {
munmap(c->data.file.mmap.data, c->data.file.mmap.length);
c->data.file.mmap.data = MAP_FAILED;
}
c->data.file.mmap.offset = our_start;
c->data.file.mmap.length = we_want;
if (!c->mem) { /* mmap did not fail till now */
c->data.file.mmap.data = mmap(0, we_want, PROT_READ, MAP_SHARED, c->data.file.file->fd, our_start);
mmap_errno = errno;
}
if (MAP_FAILED == c->data.file.mmap.data) {
/* fallback to pread(...) */
if (!c->mem) {
c->mem = g_byte_array_sized_new(we_want);
} else {
g_byte_array_set_size(c->mem, we_want);
}
read_chunk:
if (-1 == (we_have = pread(c->data.file.file->fd, c->mem->data, we_want, our_start))) {
if (EINTR == errno) goto read_chunk;
/* prefer the error of the first syscall */
if (0 != mmap_errno) {
VR_ERROR(vr, "mmap failed for '%s' (fd = %i): %s",
GSTR_SAFE_STR(c->data.file.file->name), c->data.file.file->fd,
g_strerror(mmap_errno));
} else {
VR_ERROR(vr, "pread failed for '%s' (fd = %i): %s",
GSTR_SAFE_STR(c->data.file.file->name), c->data.file.file->fd,
g_strerror(errno));
}
g_byte_array_free(c->mem, TRUE);
c->mem = NULL;
return LI_HANDLER_ERROR;
} else if (we_have != we_want) {
/* may return less than requested bytes due to signals */
/* CON_TRACE(srv, "read return unexpected number of bytes"); */
we_want = we_have;
if (length > we_have) length = we_have;
c->data.file.mmap.length = we_want;
g_byte_array_set_size(c->mem, we_want);
}
} else {
#ifdef HAVE_MADVISE
/* don't advise files < 64Kb */
if (c->data.file.mmap.length > (64*1024) &&
0 != madvise(c->data.file.mmap.data, c->data.file.mmap.length, MADV_WILLNEED)) {
VR_ERROR(vr, "madvise failed for '%s' (fd = %i): %s",
GSTR_SAFE_STR(c->data.file.file->name), c->data.file.file->fd,
g_strerror(errno));
}
#endif
}
}
*data_start = (c->mem ? (char*) c->mem->data : c->data.file.mmap.data) + start + c->offset + c->data.file.start - c->data.file.mmap.offset;
*data_len = length;
break;
case BUFFER_CHUNK:
*data_start = (char*) c->data.buffer.buffer->addr + c->data.buffer.offset + c->offset + start;
*data_len = length;
break;
}
return LI_HANDLER_GO_ON;
}
