#include "network_backends.h" #ifdef USE_LINUX_SENDFILE #include #include #include #include #include #include #include #include #include #include #include #include #include #include "network.h" #include "fdevent.h" #include "log.h" #include "stat_cache.h" int network_write_chunkqueue_linuxsendfile(server *srv, connection *con, int fd, chunkqueue *cq) { chunk *c; size_t chunks_written = 0; for(c = cq->first; c; c = c->next, chunks_written++) { int chunk_finished = 0; switch(c->type) { case MEM_CHUNK: { char * offset; size_t toSend; ssize_t r; size_t num_chunks, i; struct iovec chunks[UIO_MAXIOV]; chunk *tc; size_t num_bytes = 0; /* we can't send more then SSIZE_MAX bytes in one chunk */ /* build writev list * * 1. limit: num_chunks < UIO_MAXIOV * 2. limit: num_bytes < SSIZE_MAX */ for (num_chunks = 0, tc = c; tc && tc->type == MEM_CHUNK && num_chunks < UIO_MAXIOV; tc = tc->next, num_chunks++); for (tc = c, i = 0; i < num_chunks; tc = tc->next, i++) { if (tc->mem->used == 0) { chunks[i].iov_base = tc->mem->ptr; chunks[i].iov_len = 0; } else { offset = tc->mem->ptr + tc->offset; toSend = tc->mem->used - 1 - tc->offset; chunks[i].iov_base = offset; /* protect the return value of writev() */ if (toSend > SSIZE_MAX || num_bytes + toSend > SSIZE_MAX) { chunks[i].iov_len = SSIZE_MAX - num_bytes; num_chunks = i + 1; break; } else { chunks[i].iov_len = toSend; } num_bytes += toSend; } } if ((r = writev(fd, chunks, num_chunks)) < 0) { switch (errno) { case EAGAIN: case EINTR: r = 0; break; case EPIPE: case ECONNRESET: return -2; default: log_error_write(srv, __FILE__, __LINE__, "ssd", "writev failed:", strerror(errno), fd); return -1; } } /* check which chunks have been written */ cq->bytes_out += r; for(i = 0, tc = c; i < num_chunks; i++, tc = tc->next) { if (r >= (ssize_t)chunks[i].iov_len) { /* written */ r -= chunks[i].iov_len; tc->offset += chunks[i].iov_len; if (chunk_finished) { /* skip the chunks from further touches */ chunks_written++; c = c->next; } else { /* chunks_written + c = c->next is done in the for()*/ chunk_finished++; } } else { /* partially written */ tc->offset += r; chunk_finished = 0; break; } } break; } case FILE_CHUNK: { ssize_t r; off_t offset; size_t toSend; stat_cache_entry *sce = NULL; if (HANDLER_ERROR == stat_cache_get_entry(srv, con, c->file.name, &sce)) { log_error_write(srv, __FILE__, __LINE__, "sb", strerror(errno), c->file.name); return -1; } offset = c->file.start + c->offset; /* limit the toSend to 2^31-1 bytes in a chunk */ toSend = c->file.length - c->offset > ((1 << 30) - 1) ? ((1 << 30) - 1) : c->file.length - c->offset; if (offset > sce->st.st_size) { log_error_write(srv, __FILE__, __LINE__, "sb", "file was shrinked:", c->file.name); return -1; } /* open file if not already opened */ if (-1 == c->file.fd) { if (-1 == (c->file.fd = open(c->file.name->ptr, O_RDONLY))) { log_error_write(srv, __FILE__, __LINE__, "ss", "open failed: ", strerror(errno)); return -1; } #ifdef FD_CLOEXEC fcntl(c->file.fd, F_SETFD, FD_CLOEXEC); #endif } /* Linux sendfile() */ if (-1 == (r = sendfile(fd, c->file.fd, &offset, toSend))) { switch (errno) { case EAGAIN: case EINTR: r = 0; break; case EPIPE: case ECONNRESET: return -2; default: log_error_write(srv, __FILE__, __LINE__, "ssd", "sendfile failed:", strerror(errno), fd); return -1; } } c->offset += r; cq->bytes_out += r; if (c->offset == c->file.length) { chunk_finished = 1; /* chunk_free() / chunk_reset() will cleanup for us but it is a ok to be faster :) */ if (c->file.fd != -1) { close(c->file.fd); c->file.fd = -1; } } break; } default: log_error_write(srv, __FILE__, __LINE__, "ds", c, "type not known"); return -1; } if (!chunk_finished) { /* not finished yet */ break; } } return chunks_written; } #endif #if 0 network_linuxsendfile_init(void) { p->write = network_linuxsendfile_write_chunkset; } #endif