int io_shuffle_sendfile(request * req) { int bytes_written; unsigned int bytes_to_write; if (req->method == M_HEAD) { return complete_response(req); } bytes_to_write = (req->ranges->stop - req->ranges->start) + 1; if (bytes_to_write > system_bufsize) bytes_to_write = system_bufsize; retrysendfile: if (bytes_to_write == 0) { /* shouldn't get here, but... */ bytes_written = 0; } else { bytes_written = sendfile(req->fd, req->data_fd, &(req->filepos), bytes_to_write); if (bytes_written < 0) { if (errno == EWOULDBLOCK || errno == EAGAIN) { return -1; /* request blocked at the pipe level, but keep going */ } else if (errno == EINTR) { goto retrysendfile; } else { req->status = DEAD; #ifdef QUIET_DISCONNECT if (0) #else if (errno != EPIPE && errno != ECONNRESET) #endif { log_error_doc(req); perror("sendfile write"); } } return 0; } /* bytes_written */ } /* bytes_to_write */ /* sendfile automatically updates req->filepos, * don't touch! * req->filepos += bytes_written; */ req->ranges->start += bytes_written; req->bytes_written += bytes_written; if (req->ranges->stop + 1 == req->ranges->start) { return complete_response(req); } return 1; }
int io_shuffle(request * req) { off_t bytes_to_read; off_t bytes_written, bytes_to_write; if (req->method == M_HEAD) { return complete_response(req); } /* FIXME: This function doesn't take into account req->filesize * when *reading* into the buffer. Grr. * June 09, 2004: jdn, I don't think it's a problem anymore, * because the ranges are verified against the filesize, * and we cap bytes_to_read at bytes_to_write. */ bytes_to_read = BUFFER_SIZE - req->buffer_end - 256; bytes_to_write = (req->ranges->stop - req->ranges->start) + 1; if (bytes_to_read > bytes_to_write) bytes_to_read = bytes_to_write; if (bytes_to_read > 0 && req->data_fd) { off_t bytes_read; off_t temp; temp = lseek(req->data_fd, req->ranges->start, SEEK_SET); if (temp < 0) { req->status = DEAD; log_error_doc(req); perror("ioshuffle lseek"); return 0; } restartread: bytes_read = read(req->data_fd, req->buffer + req->buffer_end, bytes_to_read); if (bytes_read == -1) { if (errno == EINTR) goto restartread; else if (errno == EWOULDBLOCK || errno == EAGAIN) { /* not a fatal error, don't worry about it */ /* buffer is empty, we're blocking on read! */ if (req->buffer_end - req->buffer_start == 0) return -1; } else { req->status = DEAD; log_error_doc(req); perror("ioshuffle read"); return 0; } } else if (bytes_read == 0) { /* eof, write rest of buffer */ close(req->data_fd); req->data_fd = 0; } else { req->buffer_end += bytes_read; req->ranges->start += bytes_read; if ((req->ranges->stop + 1 - req->ranges->start) == 0) { return complete_response(req); } } } bytes_to_write = req->buffer_end - req->buffer_start; if (bytes_to_write == 0) { if (req->data_fd == 0) return 0; /* done */ req->buffer_end = req->buffer_start = 0; return 1; } restartwrite: bytes_written = write(req->fd, req->buffer + req->buffer_start, bytes_to_write); if (bytes_written == -1) { if (errno == EWOULDBLOCK || errno == EAGAIN) return -1; /* request blocked at the pipe level, but keep going */ else if (errno == EINTR) goto restartwrite; else { req->status = DEAD; log_error_doc(req); perror("ioshuffle write"); return 0; } } else if (bytes_written == 0) { } req->buffer_start += bytes_written; req->bytes_written += bytes_written; if (bytes_to_write == bytes_written) { req->buffer_end = req->buffer_start = 0; } return 1; }
int io_shuffle_sendfile(request * req) { off_t sendfile_offset; off_t bytes_written; off_t bytes_to_write; if (req->method == M_HEAD) { return complete_response(req); } /* XXX trouble if range is exactly 4G on a 32-bit machine? */ bytes_to_write = (req->ranges->stop - req->ranges->start) + 1; if (bytes_to_write > system_bufsize) bytes_to_write = system_bufsize; retrysendfile: if (bytes_to_write == 0) { /* shouldn't get here, but... */ bytes_written = 0; } else { /* arg 3 of sendfile should have type "off_t *" * struct range element start has type "unsigned long" * Where POSIX got the idea that an offset into a file * should be signed, I'll never know. */ sendfile_offset = req->ranges->start; if (sendfile_offset < 0) { req->status = DEAD; log_error_doc(req); fprintf(stderr, "impossible offset (%lu) requested of sendfile\n", req->ranges->start); return 0; } bytes_written = sendfile(req->fd, req->data_fd, &sendfile_offset, bytes_to_write); if (sendfile_offset < 0) { req->status = DEAD; log_error_doc(req); fprintf(stderr, "bad craziness in sendfile offset, returned %ld\n", (long) sendfile_offset); return 0; } req->ranges->start = sendfile_offset; if (bytes_written < 0) { if (errno == ENOSYS) { return io_shuffle(req); } else if (errno == EWOULDBLOCK || errno == EAGAIN) { /* request blocked at the pipe level, but keep going */ return -1; } else if (errno == EINTR) { goto retrysendfile; } else { req->status = DEAD; #ifdef QUIET_DISCONNECT if (0) #else if (errno != EPIPE && errno != ECONNRESET) #endif { log_error_doc(req); perror("sendfile write"); } } return 0; } else if (bytes_written == 0) { /* not sure how to handle this. * For now, treat it like it is blocked. */ return -1; }/* bytes_written */ } /* bytes_to_write */ /* sendfile automatically updates req->ranges->start * don't touch! * req->ranges->start += bytes_written; */ req->bytes_written += bytes_written; if (req->ranges->stop + 1 <= req->ranges->start) { return complete_response(req); } return 1; }
int init_get(request * req) { int data_fd, saved_errno; struct stat statbuf; volatile off_t bytes_free; data_fd = open(req->pathname, O_RDONLY|O_LARGEFILE); saved_errno = errno; /* might not get used */ while (use_lang_rewrite && data_fd == -1 && errno == ENOENT) { /* We cannot open that file - Check whether we can rewrite it * to a different language suffix. We only support filenames * of the format: "foo.ll.html" as an alias for "foo.html". */ unsigned int len; len = strlen(req->pathname); if (len < 6 || strcmp (req->pathname + len - 5, ".html")) break; /* does not end in ".html" */ if (len > 8 && req->pathname[len-8] == '.' && req->pathname[len-7] >= 'a' && req->pathname[len-7] <= 'z' && req->pathname[len-6] >= 'a' && req->pathname[len-6] <= 'z') { /* The request was for a language dependent file. Strip * it and try the generic form. */ char save_name[8]; strcpy (save_name, req->pathname + len - 7); strcpy (req->pathname + len - 7, "html"); data_fd = open(req->pathname, O_RDONLY); if (data_fd == -1) strcpy (req->pathname + len - 7, save_name); break; } else if ( 0 ) { /* Fixme: Other items to try from the list of accepted_languages */ data_fd = open(req->pathname, O_RDONLY); } else break; } #ifdef GUNZIP if (data_fd == -1 && errno == ENOENT) { /* cannot open */ /* it's either a gunzipped file or a directory */ char gzip_pathname[MAX_PATH_LENGTH]; unsigned int len; len = strlen(req->pathname); if (len + 4 > sizeof(gzip_pathname)) { log_error_doc(req); fprintf(stderr, "Pathname + .gz too long! (%s)\n", req->pathname); send_r_bad_request(req); return 0; } memcpy(gzip_pathname, req->pathname, len); memcpy(gzip_pathname + len, ".gz", 3); gzip_pathname[len + 3] = '\0'; data_fd = open(gzip_pathname, O_RDONLY|O_LARGEFILE); if (data_fd != -1) { close(data_fd); req->response_status = R_REQUEST_OK; if (req->pathname) free(req->pathname); req->pathname = strdup(gzip_pathname); if (!req->pathname) { boa_perror(req, "strdup req->pathname for gzipped filename " __FILE__ ":" STR(__LINE__)); return 0; } if (req->http_version != HTTP09) { req_write(req, http_ver_string(req->http_version)); req_write(req, " 200 OK-GUNZIP" CRLF); print_http_headers(req); print_content_type(req); print_last_modified(req); req_write(req, CRLF); req_flush(req); } if (req->method == M_HEAD) return 0; return init_cgi(req); } } #endif if (data_fd == -1) { log_error_doc(req); errno = saved_errno; perror("document open"); if (saved_errno == ENOENT) send_r_not_found(req); else if (saved_errno == EACCES) send_r_forbidden(req); else send_r_bad_request(req); return 0; } #ifdef ACCESS_CONTROL if (!access_allow(req->pathname)) { send_r_forbidden(req); return 0; } #endif fstat(data_fd, &statbuf); if (S_ISDIR(statbuf.st_mode)) { /* directory */ close(data_fd); /* close dir */ if (req->pathname[strlen(req->pathname) - 1] != '/') { char buffer[3 * MAX_PATH_LENGTH + 128]; unsigned int len; #ifdef ALLOW_LOCAL_REDIRECT len = strlen(req->request_uri); if (len + 2 > sizeof(buffer)) { send_r_error(req); return 0; } memcpy(buffer, req->request_uri, len); buffer[len] = '/'; buffer[len+1] = '\0'; #else char *host = server_name; unsigned int l2; char *port = NULL; const char *prefix = hsts_header? "https://" : "http://"; static unsigned int l3 = 0; static unsigned int l4 = 0; if (l4 == 0) { l4 = strlen(prefix); } len = strlen(req->request_uri); if (!port && server_port != 80 && !no_redirect_port) { port = strdup(simple_itoa(server_port)); if (port == NULL) { errno = ENOMEM; boa_perror(req, "Unable to perform simple_itoa conversion on server port!"); return 0; } l3 = strlen(port); } /* l3 and l4 are done */ if (req->host) { /* only shows up in vhost mode */ /* what about the port? (in vhost_mode?) */ /* we don't currently report ports that differ * from out "bound" (listening) port, so we don't care */ host = req->host; } l2 = strlen(host); if (server_port != 80 && !no_redirect_port) { if (l4 + l2 + 1 + l3 + len + 1 > sizeof(buffer)) { errno = ENOMEM; boa_perror(req, "buffer not large enough for directory redirect"); return 0; } memcpy(buffer, prefix, l4); memcpy(buffer + l4, host, l2); buffer[l4 + l2] = ':'; memcpy(buffer + l4 + l2 + 1, port, l3); memcpy(buffer + l4 + l2 + 1 + l3, req->request_uri, len); buffer[l4 + l2 + 1 + l3 + len] = '/'; buffer[l4 + l2 + 1 + l3 + len + 1] = '\0'; } else { if (l4 + l2 + len + 1 > sizeof(buffer)) { errno = ENOMEM; boa_perror(req, "buffer not large enough for directory redirect"); return 0; } memcpy(buffer, prefix, l4); memcpy(buffer + l4, host, l2); memcpy(buffer + l4 + l2, req->request_uri, len); buffer[l4 + l2 + len] = '/'; buffer[l4 + l2 + len + 1] = '\0'; } #endif /* ALLOW LOCAL REDIRECT */ send_r_moved_perm(req, buffer); return 0; } data_fd = get_dir(req, &statbuf); /* updates statbuf */ if (data_fd < 0) /* couldn't do it */ return 0; /* errors reported by get_dir */ else if (data_fd == 0 || data_fd == 1) return data_fd; /* else, data_fd contains the fd of the file... */ } if (!S_ISREG(statbuf.st_mode)) { /* regular file */ log_error_doc(req); fprintf(stderr, "Resulting file is not a regular file.\n"); send_r_bad_request(req); close(data_fd); return 0; } /* If-UnModified-Since asks * is the file newer than date located in time_cval * yes -> return 412 * no -> return 200 * * If-Modified-Since asks * is the file date less than or same as the date located in time_cval * yes -> return 304 * no -> return 200 * * If-Unmodified-Since overrides If-Modified-Since */ /* if (req->headers[H_IF_UNMODIFIED_SINCE] && modified_since(&(statbuf.st_mtime), req->headers[H_IF_UNMODIFIED_SINCE])) { send_r_precondition_failed(req); return 0; } else */ if (req->if_modified_since && !modified_since(&(statbuf.st_mtime), req->if_modified_since)) { send_r_not_modified(req); close(data_fd); return 0; } req->filesize = statbuf.st_size; req->last_modified = statbuf.st_mtime; /* ignore if-range without range */ if (req->header_ifrange && !req->ranges) req->header_ifrange = NULL; /* we don't support it yet */ req->header_ifrange = NULL; /* parse ranges now */ /* we have to wait until req->filesize exists to fix them up */ /* fixup handles handles communicating with the client */ /* ranges_fixup logs as appropriate, and sends * send_r_invalid_range on error. */ if (req->filesize == 0) { if (req->http_version < HTTP11) { send_r_request_ok(req); close(data_fd); return 0; } send_r_no_content(req); close(data_fd); return 0; } if (req->ranges && !ranges_fixup(req)) { close(data_fd); return 0; } /* if no range has been set, use default range */ #if 0 DEBUG(DEBUG_RANGE) { log_error_time(); fprintf(stderr, "if-range: %s\time_cval: %d\tmtime: %d\n", req->header_ifrange, req->time_cval, statbuf->st_mtime); } #endif /* If the entity tag given in the If-Range header matches the current entity tag for the entity, then the server should provide the specified sub-range of the entity using a 206 (Partial content) response. If the entity tag does not match, then the server should return the entire entity using a 200 (OK) response. */ /* IF we have range data *and* no if-range or if-range matches... */ #ifdef MAX_FILE_MMAP if (req->filesize > MAX_FILE_MMAP) { req->data_fd = data_fd; req->status = IOSHUFFLE; } else #endif { /* NOTE: I (Jon Nelson) tried performing a read(2) * into the output buffer provided the file data would * fit, before mmapping, and if successful, writing that * and stopping there -- all to avoid the cost * of a mmap. Oddly, it was *slower* in benchmarks. */ req->mmap_entry_var = find_mmap(data_fd, &statbuf); if (req->mmap_entry_var == NULL) { req->data_fd = data_fd; req->status = IOSHUFFLE; } else { req->data_mem = req->mmap_entry_var->mmap; close(data_fd); /* close data file */ } } if (!req->ranges) { req->ranges = range_pool_pop(); req->ranges->start = 0; req->ranges->stop = -1; if (!ranges_fixup(req)) { return 0; } send_r_request_ok(req); } else { /* FIXME: support if-range header here, by the following logic: * if !req->header_ifrange || st_mtime > header_ifrange, * send_r_partial_content * else * reset-ranges, etc... */ if (!req->header_ifrange) { send_r_partial_content(req); } else { /* either no if-range or the if-range does not match */ ranges_reset(req); req->ranges = range_pool_pop(); req->ranges->start = 0; req->ranges->stop = -1; if (!ranges_fixup(req)) { return 0; } send_r_request_ok(req); } } if (req->method == M_HEAD) { return complete_response(req); } bytes_free = 0; if (req->data_mem) { /* things can really go tilt if req->buffer_end > BUFFER_SIZE, * but basically that can't happen */ /* We lose statbuf here, so make sure response has been sent */ bytes_free = BUFFER_SIZE - req->buffer_end; /* 256 bytes for the **trailing** headers */ /* bytes is now how much the buffer can hold * after the headers */ } if (req->data_mem && bytes_free > 256) { unsigned int want; Range *r; r = req->ranges; want = (r->stop - r->start) + 1; if (bytes_free > want) bytes_free = want; else { /* bytes_free <= want */ ; } if (setjmp(env) == 0) { handle_sigbus = 1; memcpy(req->buffer + req->buffer_end, req->data_mem + r->start, bytes_free); handle_sigbus = 0; /* OK, SIGBUS **after** this point is very bad! */ } else { /* sigbus! */ log_error_doc(req); reset_output_buffer(req); send_r_error(req); log_error("Got SIGBUS in memcpy\n"); return 0; } req->buffer_end += bytes_free; req->bytes_written += bytes_free; r->start += bytes_free; if (bytes_free == want) { /* this will fit due to the 256 extra bytes_free */ return complete_response(req); } } /* We lose statbuf here, so make sure response has been sent */ return 1; }
int process_get(request * req) { off_t bytes_written; volatile off_t bytes_to_write; if (req->method == M_HEAD) { return complete_response(req); } bytes_to_write = (req->ranges->stop - req->ranges->start) + 1; if (bytes_to_write > system_bufsize) bytes_to_write = system_bufsize; if (setjmp(env) == 0) { handle_sigbus = 1; bytes_written = write(req->fd, req->data_mem + req->ranges->start, bytes_to_write); handle_sigbus = 0; /* OK, SIGBUS **after** this point is very bad! */ } else { /* sigbus! */ req->status = DEAD; log_error_doc(req); fprintf(stderr, "%sGot SIGBUS in write(2)!\n", get_commonlog_time()); /* sending an error here is inappropriate * if we are here, the file is mmapped, and thus, * a content-length has been sent. If we send fewer bytes * the client knows there has been a problem. * We run the risk of accidentally sending the right number * of bytes (or a few too many) and the client * won't be the wiser. */ return 0; } if (bytes_written < 0) { if (errno == EWOULDBLOCK || errno == EAGAIN) return -1; /* request blocked at the pipe level, but keep going */ else { #ifdef QUIET_DISCONNECT if (errno != EPIPE) { #else if (1) { #endif log_error_doc(req); /* Can generate lots of log entries, */ perror("write"); /* OK to disable if your logs get too big */ } req->status = DEAD; return 0; } } req->bytes_written += bytes_written; req->ranges->start += bytes_written; if ((req->ranges->stop + 1 - req->ranges->start) == 0) { return complete_response(req); } return 1; /* more to do */ } /* * Name: get_dir * Description: Called from process_get if the request is a directory. * statbuf must describe directory on input, since we may need its * device, inode, and mtime. * statbuf is updated, since we may need to check mtimes of a cache. * returns: * -1 error * 0 cgi (either gunzip or auto-generated) * >0 file descriptor of file */ int get_dir(request * req, struct stat *statbuf) { char pathname_with_index[MAX_PATH_LENGTH]; int data_fd; if (directory_index) { /* look for index.html first?? */ unsigned int l1, l2; l1 = strlen(req->pathname); l2 = strlen(directory_index); #ifdef GUNZIP if (l1 + l2 + 3 + 1 > sizeof(pathname_with_index)) { /* for .gz */ #else if (l1 + l2 + 1 > sizeof(pathname_with_index)) { #endif errno = ENOMEM; boa_perror(req, "pathname_with_index not large enough for pathname + index"); return -1; } memcpy(pathname_with_index, req->pathname, l1); /* doesn't copy NUL */ memcpy(pathname_with_index + l1, directory_index, l2 + 1); /* does */ data_fd = open(pathname_with_index, O_RDONLY|O_LARGEFILE); if (data_fd != -1) { /* user's index file */ /* We have to assume that directory_index will fit, because * if it doesn't, well, that's a huge configuration problem. * this is only the 'index.html' pathname for mime type */ memcpy(req->request_uri, directory_index, l2 + 1); /* for mimetype */ fstat(data_fd, statbuf); return data_fd; } if (errno == EACCES) { send_r_forbidden(req); return -1; } else if (errno != ENOENT) { /* if there is an error *other* than EACCES or ENOENT */ send_r_not_found(req); return -1; } #ifdef GUNZIP /* if we are here, trying index.html didn't work * try index.html.gz */ strcat(pathname_with_index, ".gz"); data_fd = open(pathname_with_index, O_RDONLY|O_LARGEFILE); if (data_fd != -1) { /* user's index file */ close(data_fd); req->response_status = R_REQUEST_OK; SQUASH_KA(req); if (req->pathname) free(req->pathname); req->pathname = strdup(pathname_with_index); if (!req->pathname) { boa_perror(req, "strdup of pathname_with_index for .gz files " __FILE__ ":" STR(__LINE__)); return 0; } if (req->http_version != HTTP09) { req_write(req, http_ver_string(req->http_version)); req_write(req, " 200 OK-GUNZIP" CRLF); print_http_headers(req); print_last_modified(req); req_write(req, "Content-Type: "); req_write(req, get_mime_type(directory_index)); req_write(req, CRLF CRLF); req_flush(req); } if (req->method == M_HEAD) return 0; return init_cgi(req); } #endif } /* only here if index.html, index.html.gz don't exist */ if (dirmaker != NULL) { /* don't look for index.html... maybe automake? */ req->response_status = R_REQUEST_OK; SQUASH_KA(req); /* the indexer should take care of all headers */ if (req->http_version != HTTP09) { req_write(req, http_ver_string(req->http_version)); req_write(req, " 200 OK" CRLF); print_http_headers(req); print_last_modified(req); req_write(req, "Content-Type: text/html" CRLF CRLF); req_flush(req); } if (req->method == M_HEAD) return 0; return init_cgi(req); /* in this case, 0 means success */ } else if (cachedir) { return get_cachedir_file(req, statbuf); } else { /* neither index.html nor autogenerate are allowed */ send_r_forbidden(req); return -1; /* nothing worked */ } } static int get_cachedir_file(request * req, struct stat *statbuf) { char pathname_with_index[MAX_PATH_LENGTH]; int data_fd; time_t real_dir_mtime; real_dir_mtime = statbuf->st_mtime; /* the sizeof() doesn't need a -1 because snprintf will * include the NUL when calculating if the size is enough */ snprintf(pathname_with_index, sizeof(pathname_with_index), "%s/dir.%d." PRINTF_OFF_T_ARG, cachedir, (int) statbuf->st_dev, statbuf->st_ino); data_fd = open(pathname_with_index, O_RDONLY|O_LARGEFILE); if (data_fd != -1) { /* index cache */ fstat(data_fd, statbuf); if (statbuf->st_mtime > real_dir_mtime) { statbuf->st_mtime = real_dir_mtime; /* lie */ strcpy(req->request_uri, directory_index); /* for mimetype */ return data_fd; } close(data_fd); unlink(pathname_with_index); /* cache is stale, delete it */ } if (index_directory(req, pathname_with_index) == -1) return -1; data_fd = open(pathname_with_index, O_RDONLY|O_LARGEFILE); /* Last chance */ if (data_fd != -1) { strcpy(req->request_uri, directory_index); /* for mimetype */ fstat(data_fd, statbuf); statbuf->st_mtime = real_dir_mtime; /* lie */ return data_fd; } boa_perror(req, "re-opening dircache"); return -1; /* Nothing worked. */ }