| /* sim_ether.c: OS-dependent network routines | |
| ------------------------------------------------------------------------------ | |
| Copyright (c) 2002-2007, David T. Hittner | |
| Permission is hereby granted, free of charge, to any person obtaining a | |
| copy of this software and associated documentation files (the "Software"), | |
| to deal in the Software without restriction, including without limitation | |
| the rights to use, copy, modify, merge, publish, distribute, sublicense, | |
| and/or sell copies of the Software, and to permit persons to whom the | |
| Software is furnished to do so, subject to the following conditions: | |
| The above copyright notice and this permission notice shall be included in | |
| all copies or substantial portions of the Software. | |
| THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
| IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
| FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | |
| THE AUTHOR BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER | |
| IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN | |
| CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. | |
| Except as contained in this notice, the name of the author shall not be | |
| used in advertising or otherwise to promote the sale, use or other dealings | |
| in this Software without prior written authorization from the author. | |
| ------------------------------------------------------------------------------ | |
| This ethernet simulation is based on the PCAP and WinPcap packages. | |
| PCAP/WinPcap was chosen as the basis for network code since it is the most | |
| "universal" of the various network packages available. Using this style has | |
| allowed rapid network development for the major SIMH platforms. Developing | |
| a network package specifically for SIMH was rejected due to the time required; | |
| the advantage would be a more easily compiled and integrated code set. | |
| There are various problems associated with use of ethernet networking, which | |
| would be true regardless of the network package used, since there are no | |
| universally accepted networking methods. The most serious of these is getting | |
| the proper networking package loaded onto the system, since most environments | |
| do not come with the network interface packages loaded. | |
| The second most serious network issue relates to security. The network | |
| simulation needs to simulate operating system level functionality (packet | |
| driving). However, the host network programming interfaces tend to operate at | |
| the user level of functionality, so getting to the full functionality of | |
| the network interface usually requires that the person executing the | |
| network code be a privileged user of the host system. See the PCAP/WinPcap | |
| documentation for the appropriate host platform if unprivileged use of | |
| networking is needed - there may be known workarounds. | |
| Define one of the two macros below to enable networking: | |
| USE_NETWORK - Create statically linked network code | |
| USE_SHARED - Create dynamically linked network code | |
| ------------------------------------------------------------------------------ | |
| Supported/Tested Platforms: | |
| Windows(NT,2K,XP,2K3,Vista,Win7) WinPcap V3.0+ | |
| Linux libpcap at least 0.9 | |
| OpenBSD,FreeBSD,NetBSD libpcap at least 0.9 | |
| MAC OS/X libpcap at least 0.9 | |
| Solaris Sparc libpcap at least 0.9 | |
| Solaris Intel libpcap at least 0.9 | |
| AIX ?? | |
| HP/UX ?? | |
| Compaq Tru64 Unix ?? | |
| VMS Alpha/Itanium VMS only, needs VMS libpcap | |
| WinPcap is available from: | |
| http://winpcap.polito.it/ | |
| libpcap for VMS is available from: | |
| http://simh.trailing-edge.com/sources/vms-pcap.zip | |
| libpcap for other Unix platforms is available at: | |
| NOTE: As of the release of this version of sim_ether.c ALL current | |
| *nix platforms ship with a sufficiently new version of | |
| libpcap, and ALL provide a libpcap-dev package for developing | |
| libpcap based applications. The OS vendor supplied version | |
| of libpcap AND the libpcap-dev components are preferred for | |
| proper operation of both simh AND other applications on the | |
| host system which use libpcap. | |
| Current Version: http://www.tcpdump.org/daily/libpcap-current.tar.gz | |
| Released Version: http://www.tcpdump.org/release/ | |
| When necessary (see NOTE above about vendor supplied libpcap), | |
| we've gotten the tarball, unpacked, built and installed it with: | |
| gzip -dc libpcap-current.tar.gz | tar xvf - | |
| cd libpcap-directory-name | |
| ./configure | |
| make | |
| make install | |
| Note: The "make install" step generally will have to be done as root. | |
| This will install libpcap in /usr/local/lib and /usr/local/include | |
| The current simh makefile will do the right thing to locate and | |
| reference the OS provided libpcap or the one just installed. | |
| Note: Building for the platforms indicated above, with the indicated libpcap, | |
| should automatically leverage the appropriate mechanisms contained here. | |
| Things are structured so that it is likely to work for any other as yet | |
| untested platform. If it works for you, please let the author know so we | |
| can update the table above. If it doesn't work, then the following #define | |
| variables can influence the operation on an untested platform. | |
| USE_BPF - Determines if this code leverages a libpcap/WinPcap | |
| provided bpf packet filtering facility. All tested | |
| environments have bpf facilities that work the way we | |
| need them to. However a new one might not. undefine | |
| this variable to let this code do its own filtering. | |
| USE_SETNONBLOCK - Specifies whether the libpcap environment's non-blocking | |
| semantics are to be leveraged. This helps to manage the | |
| varying behaviours of the kernel packet facilities | |
| leveraged by libpcap. | |
| USE_READER_THREAD - Specifies that packet reading should be done in the | |
| context of a separate thread. The Posix threading | |
| APIs are used. This option is less efficient than the | |
| default non-threaded approach, but it exists since some | |
| platforms don't want to work with nonblocking libpcap | |
| semantics. OpenBSD and NetBSD either don't have pthread | |
| APIs available, or they are too buggy to be useful. | |
| Using the threaded approach may require special compile | |
| and/or link time switches (i.e. -lpthread or -pthread, | |
| etc.) Consult the documentation for your platform as | |
| needed. Although this may be 'less efficient' than the | |
| non-threaded approach, the efficiency is an overall system | |
| efficiency not necessarily a simulator efficiency. This | |
| means that work is removed from the thread executing | |
| simulated instructions so the simulated system will most | |
| likely run faster (given that modern host CPUs are | |
| multi-core and have someplace to do this work in parallel). | |
| MUST_DO_SELECT - Specifies that, when USE_READER_THREAD is active, | |
| select() should be used to determine when available | |
| packets are ready for reading. Otherwise, we depend | |
| on the libpcap/kernel packet timeout specified on | |
| pcap_open_live. If USE_READER_THREAD is not set, then | |
| MUST_DO_SELECT is irrelevant | |
| USE_TAP_NETWORK - Specifies that support for tap networking should be | |
| included. This can be leveraged, along with OS bridging | |
| capabilities to share a single LAN interface. This | |
| allows device names of the form tap:tap0 to be specified | |
| at open time. This functionality is only useful/needed | |
| on *nix platforms since native sharing of Windows NIC | |
| devices works with no external magic. | |
| USE_VDE_NETWORK - Specifies that support for vde networking should be | |
| included. This can be leveraged, along with OS bridging | |
| capabilities to share a single LAN interface. It also | |
| can allow a simulator to have useful networking | |
| functionality when running without root access. This | |
| allows device names of the form vde:/tmp/switch to be | |
| specified at open time. This functionality is only | |
| available on *nix platforms since the vde api isn't | |
| available on Windows. | |
| NEED_PCAP_SENDPACKET | |
| - Specifies that you are using an older version of libpcap | |
| which doesn't provide a pcap_sendpacket API. | |
| NOTE: Changing these defines is done in either sim_ether.h OR on the global | |
| compiler command line which builds all of the modules included in a | |
| simulator. | |
| ------------------------------------------------------------------------------ | |
| Modification history: | |
| 30-Mar-12 MP Added host NIC address determination on supported VMS platforms | |
| 01-Mar-12 MP Made host NIC address determination on *nix platforms more | |
| robust. | |
| 01-Mar-12 MP Added host NIC address determination work when building | |
| under Cygwin | |
| 01-Mar-12 AGN Add conditionals for Cygwin dynamic loading of wpcap.dll | |
| 01-Mar-12 AGN Specify the full /usr/lib for dlopen under Apple Mac OS X. | |
| 17-Nov-11 MP Added dynamic loading of libpcap on *nix platforms | |
| 30-Oct-11 MP Added support for vde (Virtual Distributed Ethernet) networking | |
| 29-Oct-11 MP Added support for integrated Tap networking interfaces on OSX | |
| 12-Aug-11 MP Cleaned up payload length determination | |
| Fixed race condition detecting reflections when threaded | |
| reading and writing is enabled | |
| 18-Apr-11 MP Fixed race condition with self loopback packets in | |
| multithreaded environments | |
| 09-Jan-11 MP Fixed missing crc data when USE_READER_THREAD is defined and | |
| crc's are needed (only the pdp11_xu) | |
| 16-Dec-10 MP added priority boost for read and write threads when | |
| USE_READER_THREAD does I/O in separate threads. This helps | |
| throughput since it allows these I/O bound threads to preempt | |
| the main thread (which is executing simulated instructions). | |
| 09-Dec-10 MP allowed more flexible parsing of MAC address strings | |
| 09-Dec-10 MP Added support to determine if network address conflicts exist | |
| 07-Dec-10 MP Reworked DECnet self detection to the more general approach | |
| of loopback self when a Physical Address is being set. | |
| 04-Dec-10 MP Changed eth_write to do nonblocking writes when | |
| USE_READER_THREAD is defined. | |
| 20-Aug-10 TVO Fix for Mac OSX 10.6 | |
| 17-Jun-10 MP Fixed bug in the AUTODIN II hash filtering. | |
| 14-Jun-10 MP Added support for integrated Tap networking interfaces on BSD | |
| platforms. | |
| 13-Jun-10 MP Added support for integrated Tap networking interfaces on Linux | |
| platforms. | |
| 31-May-10 MP Added support for more TOE (TCP Offload Engine) features for IPv4 | |
| network traffic from the host and/or from hosts on the LAN. These | |
| new TOE features are: LSO (Large Send Offload) and Jumbo packet | |
| fragmentation support. These features allow a simulated network | |
| device to support traffic when a host leverages a NIC's Large | |
| Send Offload capabilities to fregment and/or segment outgoing | |
| network traffic. Additionally a simulated network device can | |
| reasonably exist on a LAN which is configured to use Jumbo frames. | |
| 21-May-10 MP Added functionality to fixup IP header checksums to accomodate | |
| packets from a host with a NIC which has TOE (TCP Offload Engine) | |
| enabled which is expected to implement the checksum computations | |
| in hardware. Since we catch packets before they arrive at the | |
| NIC the expected checksum insertions haven't been performed yet. | |
| This processing is only done for packets sent from the hoat to | |
| the guest we're supporting. In general this will be a relatively | |
| small number of packets so it is done for all IP frame packets | |
| coming from the hoat to the guest. In order to make the | |
| determination of packets specifically arriving from the host we | |
| need to know the hardware MAC address of the host NIC. Currently | |
| determining a NIC's MAC address is relatively easy on Windows. | |
| The non-windows code works on linux and may work on other *nix | |
| platforms either as is or with slight modifications. The code, | |
| as implemented, only messes with this activity if the host | |
| interface MAC address can be determined. | |
| 20-May-10 MP Added general support to deal with receiving packets smaller | |
| than ETH_MIN_PACKET in length. These come from packets | |
| looped back by some bridging mechanism and need to be padded | |
| to the minimum frame size. A real NIC won't pass us any | |
| packets like that. This fix belongs here since this layer | |
| is responsible for interfacing to they physical layer | |
| devices, AND it belongs here to get CRC processing right. | |
| 05-Mar-08 MP Added optional multicast filtering support for doing | |
| LANCE style AUTODIN II based hashed filtering. | |
| 07-Feb-08 MP Added eth_show_dev to display ethernet state | |
| Changed the return value from eth_read to return whether | |
| or not a packet was read. No existing callers used or | |
| checked constant return value that previously was being | |
| supplied. | |
| 29-Jan-08 MP Added eth_set_async to provide a mechanism (when | |
| USE_READER_THREAD is enabled) to allow packet reception | |
| to dynamically update the simulator event queue and | |
| potentially avoid polling for I/O. This provides a minimal | |
| overhead (no polling) maximal responsiveness for network | |
| activities. | |
| 29-Jan-08 MP Properly sequenced activities in eth_close to avoid a race | |
| condition when USE_READER_THREAD is enabled. | |
| 25-Jan-08 MP Changed the following when USE_READER_THREAD is enabled: | |
| - Fixed bug when the simulated device doesn't need crc | |
| in packet data which is read. | |
| - Added call to pcap_setmintocopy to minimize packet | |
| delivery latencies. | |
| - Added ethq_destroy and used it to avoid a memory leak in | |
| eth_close. | |
| - Properly cleaned up pthread mutexes in eth_close. | |
| Migrated to using sim_os_ms_sleep for a delay instead of | |
| a call to select(). | |
| Fixed the bpf filter used when no traffic is to be matched. | |
| Reworked eth_add_packet_crc32 implementation to avoid an | |
| extra buffer copy while reading packets. | |
| Fixedup #ifdef's relating to USE_SHARED so that setting | |
| USE_SHARED or USE_NETWORK will build a working network | |
| environment. | |
| 23-Jan-08 MP Reworked eth_packet_trace and eth_packet_trace_ex to allow | |
| only output ethernet header+crc and provide a mechanism for | |
| the simulated device to display full packet data debugging. | |
| 17-May-07 DTH Fixed non-ethernet device removal loop (from Naoki Hamada) | |
| 15-May-07 DTH Added dynamic loading of wpcap.dll; | |
| Corrected exceed max index bug in ethX lookup | |
| 04-May-07 DTH Corrected failure to look up ethernet device names in | |
| the registry on Windows XP x64 | |
| 10-Jul-06 RMS Fixed linux conditionalization (from Chaskiel Grundman) | |
| 02-Jun-06 JDB Fixed compiler warning for incompatible sscanf parameter | |
| 15-Dec-05 DTH Patched eth_host_devices [remove non-ethernet devices] | |
| (from Mark Pizzolato and Galen Tackett, 08-Jun-05) | |
| Patched eth_open [tun fix](from Antal Ritter, 06-Oct-05) | |
| 30-Nov-05 DTH Added option to regenerate CRC on received packets; some | |
| ethernet devices need to pass it on to the simulation, and by | |
| the time libpcap/winpcap gets the packet, the host OS network | |
| layer has already stripped CRC out of the packet | |
| 01-Dec-04 DTH Added Windows user-defined adapter names (from Timothe Litt) | |
| 25-Mar-04 MP Revised comments and minor #defines to deal with updated | |
| libpcap which now provides pcap_sendpacket on all platforms. | |
| 04-Feb-04 MP Returned success/fail status from eth_write to support | |
| determining if the current libpcap connection can successfully | |
| write packets. | |
| Added threaded approach to reading packets since | |
| this works better on some platforms (solaris intel) than the | |
| inconsistently implemented non-blocking read approach. | |
| 04-Feb-04 DTH Converted ETH_DEBUG to sim_debug | |
| 13-Jan-04 MP tested and fixed on OpenBSD, NetBS and FreeBSD. | |
| 09-Jan-04 MP removed the BIOCSHDRCMPLT ioctl() for OS/X | |
| 05-Jan-04 DTH Added eth_mac_scan | |
| 30-Dec-03 DTH Cleaned up queue routines, added no network support message | |
| 26-Dec-03 DTH Added ethernet show and queue functions from pdp11_xq | |
| 15-Dec-03 MP polished generic libpcap support. | |
| 05-Dec-03 DTH Genericized eth_devices() and #ifdefs | |
| 03-Dec-03 MP Added Solaris support | |
| 02-Dec-03 DTH Corrected decnet fix to use reflection counting | |
| 01-Dec-03 DTH Added BPF source filtering and reflection counting | |
| 28-Nov-03 DTH Rewrote eth_devices using universal pcap_findalldevs() | |
| 25-Nov-03 DTH Verified DECNET_FIX, reversed ifdef to mainstream code | |
| 19-Nov-03 MP Fixed BPF functionality on Linux/BSD. | |
| 17-Nov-03 DTH Added xBSD simplification | |
| 14-Nov-03 DTH Added #ifdef DECNET_FIX for problematic duplicate detection code | |
| 13-Nov-03 DTH Merged in __FreeBSD__ support | |
| 21-Oct-03 MP Added enriched packet dumping for debugging | |
| 20-Oct-03 MP Added support for multiple ethernet devices on VMS | |
| 20-Sep-03 Ankan Add VMS support (Alpha only) | |
| 29-Sep-03 MP Changed separator character in eth_fmt_mac to be ":" to | |
| format ethernet addresses the way the BPF compile engine | |
| wants to see them. | |
| Added BPF support to filter packets | |
| Added missing printf in eth_close | |
| 07-Jun-03 MP Added WIN32 support for DECNET duplicate address detection. | |
| 06-Jun-03 MP Fixed formatting of Ethernet Protocol Type in eth_packet_trace | |
| 30-May-03 DTH Changed WIN32 to _WIN32 for consistency | |
| 07-Mar-03 MP Fixed Linux implementation of PacketGetAdapterNames to also | |
| work on Red Hat 6.2-sparc and Debian 3.0r1-sparc. | |
| 03-Mar-03 MP Changed logging to be consistent on stdout and sim_log | |
| 01-Feb-03 MP Changed type of local variables in eth_packet_trace to | |
| conform to the interface needs of eth_mac_fmt wich produces | |
| char data instead of unsigned char data. Suggested by the | |
| DECC compiler. | |
| 15-Jan-03 DTH Corrected PacketGetAdapterNames parameter2 datatype | |
| 26-Dec-02 DTH Merged Mark Pizzolato's enhancements with main source | |
| Added networking documentation | |
| Changed _DEBUG to ETH_DEBUG | |
| 20-Dec-02 MP Added display of packet CRC to the eth_packet_trace. | |
| This helps distinguish packets with identical lengths | |
| and protocols. | |
| 05-Dec-02 MP With the goal of draining the input buffer more rapidly | |
| changed eth_read to call pcap_dispatch repeatedly until | |
| either a timeout returns nothing or a packet allowed by | |
| the filter is seen. This more closely reflects how the | |
| pcap layer will work when the filtering is actually done | |
| by a bpf filter. | |
| 31-Oct-02 DTH Added USE_NETWORK conditional | |
| Reworked not attached test | |
| Added OpenBSD support (from Federico Schwindt) | |
| Added ethX detection simplification (from Megan Gentry) | |
| Removed sections of temporary code | |
| Added parameter validation | |
| 23-Oct-02 DTH Beta 5 released | |
| 22-Oct-02 DTH Added all_multicast and promiscuous support | |
| Fixed not attached behavior | |
| 21-Oct-02 DTH Added NetBSD support (from Jason Thorpe) | |
| Patched buffer size to make sure entire packet is read in | |
| Made 'ethX' check characters passed as well as length | |
| Corrected copyright again | |
| 16-Oct-02 DTH Beta 4 released | |
| Corrected copyright | |
| 09-Oct-02 DTH Beta 3 released | |
| Added pdp11 write acceleration (from Patrick Caulfield) | |
| 08-Oct-02 DTH Beta 2 released | |
| Integrated with 2.10-0p4 | |
| Added variable vector and copyrights | |
| 04-Oct-02 DTH Added linux support (from Patrick Caulfield) | |
| 03-Oct-02 DTH Beta version of xq/sim_ether released for SIMH 2.09-11 | |
| 24-Sep-02 DTH Finished eth_devices, eth_getname | |
| 18-Sep-02 DTH Callbacks implemented | |
| 13-Sep-02 DTH Basic packet read/write written | |
| 20-Aug-02 DTH Created Sim_Ether for O/S independant ethernet implementation | |
| ------------------------------------------------------------------------------ | |
| */ | |
| #include <ctype.h> | |
| #include "sim_ether.h" | |
| #include "sim_sock.h" | |
| extern FILE *sim_log; | |
| /*============================================================================*/ | |
| /* OS-independant ethernet routines */ | |
| /*============================================================================*/ | |
| t_stat eth_mac_scan (ETH_MAC* mac, char* strmac) | |
| { | |
| unsigned int a0, a1, a2, a3, a4, a5; | |
| const ETH_MAC zeros = {0,0,0,0,0,0}; | |
| const ETH_MAC ones = {0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF}; | |
| ETH_MAC newmac; | |
| if ((6 != sscanf(strmac, "%x:%x:%x:%x:%x:%x", &a0, &a1, &a2, &a3, &a4, &a5)) && | |
| (6 != sscanf(strmac, "%x.%x.%x.%x.%x.%x", &a0, &a1, &a2, &a3, &a4, &a5)) && | |
| (6 != sscanf(strmac, "%x-%x-%x-%x-%x-%x", &a0, &a1, &a2, &a3, &a4, &a5))) | |
| return SCPE_ARG; | |
| if ((a0 > 0xFF) || (a1 > 0xFF) || (a2 > 0xFF) || (a3 > 0xFF) || (a4 > 0xFF) || (a5 > 0xFF)) | |
| return SCPE_ARG; | |
| newmac[0] = a0; | |
| newmac[1] = a1; | |
| newmac[2] = a2; | |
| newmac[3] = a3; | |
| newmac[4] = a4; | |
| newmac[5] = a5; | |
| /* final check - mac cannot be broadcast or multicast address */ | |
| if (!memcmp(newmac, zeros, sizeof(ETH_MAC)) || /* broadcast */ | |
| !memcmp(newmac, ones, sizeof(ETH_MAC)) || /* broadcast */ | |
| (newmac[0] & 0x01) /* multicast */ | |
| ) | |
| return SCPE_ARG; | |
| /* new mac is OK, copy into passed mac */ | |
| memcpy (*mac, newmac, sizeof(ETH_MAC)); | |
| return SCPE_OK; | |
| } | |
| void eth_mac_fmt(ETH_MAC* mac, char* buff) | |
| { | |
| uint8* m = (uint8*) mac; | |
| sprintf(buff, "%02X:%02X:%02X:%02X:%02X:%02X", m[0], m[1], m[2], m[3], m[4], m[5]); | |
| return; | |
| } | |
| static const uint32 crcTable[256] = { | |
| 0x00000000, 0x77073096, 0xEE0E612C, 0x990951BA, 0x076DC419, 0x706AF48F, | |
| 0xE963A535, 0x9E6495A3, 0x0EDB8832, 0x79DCB8A4, 0xE0D5E91E, 0x97D2D988, | |
| 0x09B64C2B, 0x7EB17CBD, 0xE7B82D07, 0x90BF1D91, 0x1DB71064, 0x6AB020F2, | |
| 0xF3B97148, 0x84BE41DE, 0x1ADAD47D, 0x6DDDE4EB, 0xF4D4B551, 0x83D385C7, | |
| 0x136C9856, 0x646BA8C0, 0xFD62F97A, 0x8A65C9EC, 0x14015C4F, 0x63066CD9, | |
| 0xFA0F3D63, 0x8D080DF5, 0x3B6E20C8, 0x4C69105E, 0xD56041E4, 0xA2677172, | |
| 0x3C03E4D1, 0x4B04D447, 0xD20D85FD, 0xA50AB56B, 0x35B5A8FA, 0x42B2986C, | |
| 0xDBBBC9D6, 0xACBCF940, 0x32D86CE3, 0x45DF5C75, 0xDCD60DCF, 0xABD13D59, | |
| 0x26D930AC, 0x51DE003A, 0xC8D75180, 0xBFD06116, 0x21B4F4B5, 0x56B3C423, | |
| 0xCFBA9599, 0xB8BDA50F, 0x2802B89E, 0x5F058808, 0xC60CD9B2, 0xB10BE924, | |
| 0x2F6F7C87, 0x58684C11, 0xC1611DAB, 0xB6662D3D, 0x76DC4190, 0x01DB7106, | |
| 0x98D220BC, 0xEFD5102A, 0x71B18589, 0x06B6B51F, 0x9FBFE4A5, 0xE8B8D433, | |
| 0x7807C9A2, 0x0F00F934, 0x9609A88E, 0xE10E9818, 0x7F6A0DBB, 0x086D3D2D, | |
| 0x91646C97, 0xE6635C01, 0x6B6B51F4, 0x1C6C6162, 0x856530D8, 0xF262004E, | |
| 0x6C0695ED, 0x1B01A57B, 0x8208F4C1, 0xF50FC457, 0x65B0D9C6, 0x12B7E950, | |
| 0x8BBEB8EA, 0xFCB9887C, 0x62DD1DDF, 0x15DA2D49, 0x8CD37CF3, 0xFBD44C65, | |
| 0x4DB26158, 0x3AB551CE, 0xA3BC0074, 0xD4BB30E2, 0x4ADFA541, 0x3DD895D7, | |
| 0xA4D1C46D, 0xD3D6F4FB, 0x4369E96A, 0x346ED9FC, 0xAD678846, 0xDA60B8D0, | |
| 0x44042D73, 0x33031DE5, 0xAA0A4C5F, 0xDD0D7CC9, 0x5005713C, 0x270241AA, | |
| 0xBE0B1010, 0xC90C2086, 0x5768B525, 0x206F85B3, 0xB966D409, 0xCE61E49F, | |
| 0x5EDEF90E, 0x29D9C998, 0xB0D09822, 0xC7D7A8B4, 0x59B33D17, 0x2EB40D81, | |
| 0xB7BD5C3B, 0xC0BA6CAD, 0xEDB88320, 0x9ABFB3B6, 0x03B6E20C, 0x74B1D29A, | |
| 0xEAD54739, 0x9DD277AF, 0x04DB2615, 0x73DC1683, 0xE3630B12, 0x94643B84, | |
| 0x0D6D6A3E, 0x7A6A5AA8, 0xE40ECF0B, 0x9309FF9D, 0x0A00AE27, 0x7D079EB1, | |
| 0xF00F9344, 0x8708A3D2, 0x1E01F268, 0x6906C2FE, 0xF762575D, 0x806567CB, | |
| 0x196C3671, 0x6E6B06E7, 0xFED41B76, 0x89D32BE0, 0x10DA7A5A, 0x67DD4ACC, | |
| 0xF9B9DF6F, 0x8EBEEFF9, 0x17B7BE43, 0x60B08ED5, 0xD6D6A3E8, 0xA1D1937E, | |
| 0x38D8C2C4, 0x4FDFF252, 0xD1BB67F1, 0xA6BC5767, 0x3FB506DD, 0x48B2364B, | |
| 0xD80D2BDA, 0xAF0A1B4C, 0x36034AF6, 0x41047A60, 0xDF60EFC3, 0xA867DF55, | |
| 0x316E8EEF, 0x4669BE79, 0xCB61B38C, 0xBC66831A, 0x256FD2A0, 0x5268E236, | |
| 0xCC0C7795, 0xBB0B4703, 0x220216B9, 0x5505262F, 0xC5BA3BBE, 0xB2BD0B28, | |
| 0x2BB45A92, 0x5CB36A04, 0xC2D7FFA7, 0xB5D0CF31, 0x2CD99E8B, 0x5BDEAE1D, | |
| 0x9B64C2B0, 0xEC63F226, 0x756AA39C, 0x026D930A, 0x9C0906A9, 0xEB0E363F, | |
| 0x72076785, 0x05005713, 0x95BF4A82, 0xE2B87A14, 0x7BB12BAE, 0x0CB61B38, | |
| 0x92D28E9B, 0xE5D5BE0D, 0x7CDCEFB7, 0x0BDBDF21, 0x86D3D2D4, 0xF1D4E242, | |
| 0x68DDB3F8, 0x1FDA836E, 0x81BE16CD, 0xF6B9265B, 0x6FB077E1, 0x18B74777, | |
| 0x88085AE6, 0xFF0F6A70, 0x66063BCA, 0x11010B5C, 0x8F659EFF, 0xF862AE69, | |
| 0x616BFFD3, 0x166CCF45, 0xA00AE278, 0xD70DD2EE, 0x4E048354, 0x3903B3C2, | |
| 0xA7672661, 0xD06016F7, 0x4969474D, 0x3E6E77DB, 0xAED16A4A, 0xD9D65ADC, | |
| 0x40DF0B66, 0x37D83BF0, 0xA9BCAE53, 0xDEBB9EC5, 0x47B2CF7F, 0x30B5FFE9, | |
| 0xBDBDF21C, 0xCABAC28A, 0x53B39330, 0x24B4A3A6, 0xBAD03605, 0xCDD70693, | |
| 0x54DE5729, 0x23D967BF, 0xB3667A2E, 0xC4614AB8, 0x5D681B02, 0x2A6F2B94, | |
| 0xB40BBE37, 0xC30C8EA1, 0x5A05DF1B, 0x2D02EF8D | |
| }; | |
| uint32 eth_crc32(uint32 crc, const void* vbuf, size_t len) | |
| { | |
| const uint32 mask = 0xFFFFFFFF; | |
| const unsigned char* buf = (const unsigned char*)vbuf; | |
| crc ^= mask; | |
| while (len > 8) { | |
| crc = (crc >> 8) ^ crcTable[ (crc ^ (*buf++)) & 0xFF ]; | |
| crc = (crc >> 8) ^ crcTable[ (crc ^ (*buf++)) & 0xFF ]; | |
| crc = (crc >> 8) ^ crcTable[ (crc ^ (*buf++)) & 0xFF ]; | |
| crc = (crc >> 8) ^ crcTable[ (crc ^ (*buf++)) & 0xFF ]; | |
| crc = (crc >> 8) ^ crcTable[ (crc ^ (*buf++)) & 0xFF ]; | |
| crc = (crc >> 8) ^ crcTable[ (crc ^ (*buf++)) & 0xFF ]; | |
| crc = (crc >> 8) ^ crcTable[ (crc ^ (*buf++)) & 0xFF ]; | |
| crc = (crc >> 8) ^ crcTable[ (crc ^ (*buf++)) & 0xFF ]; | |
| len -= 8; | |
| } | |
| while (0 != len--) | |
| crc = (crc >> 8) ^ crcTable[ (crc ^ (*buf++)) & 0xFF ]; | |
| return(crc ^ mask); | |
| } | |
| int eth_get_packet_crc32_data(const uint8 *msg, int len, uint8 *crcdata) | |
| { | |
| int crc_len; | |
| if (len <= ETH_MAX_PACKET) { | |
| uint32 crc = eth_crc32(0, msg, len); /* calculate CRC */ | |
| uint32 ncrc = htonl(crc); /* CRC in network order */ | |
| int size = sizeof(ncrc); /* size of crc field */ | |
| memcpy(crcdata, &ncrc, size); /* append crc to packet */ | |
| crc_len = len + size; /* set packet crc length */ | |
| } else { | |
| crc_len = 0; /* appending crc would destroy packet */ | |
| } | |
| return crc_len; | |
| } | |
| int eth_add_packet_crc32(uint8 *msg, int len) | |
| { | |
| int crc_len; | |
| if (len <= ETH_MAX_PACKET) { | |
| crc_len = eth_get_packet_crc32_data(msg, len, &msg[len]);/* append crc to packet */ | |
| } else { | |
| crc_len = 0; /* appending crc would destroy packet */ | |
| } | |
| return crc_len; | |
| } | |
| void eth_setcrc(ETH_DEV* dev, int need_crc) | |
| { | |
| dev->need_crc = need_crc; | |
| } | |
| void eth_packet_trace_ex(ETH_DEV* dev, const uint8 *msg, int len, char* txt, int detail, uint32 reason) | |
| { | |
| if (dev->dptr->dctrl & reason) { | |
| char src[20]; | |
| char dst[20]; | |
| unsigned short* proto = (unsigned short*) &msg[12]; | |
| uint32 crc = eth_crc32(0, msg, len); | |
| eth_mac_fmt((ETH_MAC*)&msg[0], dst); | |
| eth_mac_fmt((ETH_MAC*)&msg[6], src); | |
| sim_debug(reason, dev->dptr, "%s dst: %s src: %s proto: 0x%04X len: %d crc: %X\n", | |
| txt, dst, src, ntohs(*proto), len, crc); | |
| if (detail) { | |
| int i, same, group, sidx, oidx; | |
| char outbuf[80], strbuf[18]; | |
| static char hex[] = "0123456789ABCDEF"; | |
| for (i=same=0; i<len; i += 16) { | |
| if ((i > 0) && (0 == memcmp(&msg[i], &msg[i-16], 16))) { | |
| ++same; | |
| continue; | |
| } | |
| if (same > 0) { | |
| sim_debug(reason, dev->dptr, "%04X thru %04X same as above\n", i-(16*same), i-1); | |
| same = 0; | |
| } | |
| group = (((len - i) > 16) ? 16 : (len - i)); | |
| for (sidx=oidx=0; sidx<group; ++sidx) { | |
| outbuf[oidx++] = ' '; | |
| outbuf[oidx++] = hex[(msg[i+sidx]>>4)&0xf]; | |
| outbuf[oidx++] = hex[msg[i+sidx]&0xf]; | |
| if (isprint(msg[i+sidx])) | |
| strbuf[sidx] = msg[i+sidx]; | |
| else | |
| strbuf[sidx] = '.'; | |
| } | |
| outbuf[oidx] = '\0'; | |
| strbuf[sidx] = '\0'; | |
| sim_debug(reason, dev->dptr, "%04X%-48s %s\n", i, outbuf, strbuf); | |
| } | |
| if (same > 0) { | |
| sim_debug(reason, dev->dptr, "%04X thru %04X same as above\n", i-(16*same), len-1); | |
| } | |
| } | |
| } | |
| } | |
| void eth_packet_trace(ETH_DEV* dev, const uint8 *msg, int len, char* txt) | |
| { | |
| eth_packet_trace_ex(dev, msg, len, txt, 0, dev->dbit); | |
| } | |
| char* eth_getname(int number, char* name) | |
| { | |
| ETH_LIST list[ETH_MAX_DEVICE]; | |
| int count = eth_devices(ETH_MAX_DEVICE, list); | |
| if ((number < 0) || (count <= number)) | |
| return NULL; | |
| strcpy(name, list[number].name); | |
| return name; | |
| } | |
| char* eth_getname_bydesc(char* desc, char* name) | |
| { | |
| ETH_LIST list[ETH_MAX_DEVICE]; | |
| int count = eth_devices(ETH_MAX_DEVICE, list); | |
| int i; | |
| size_t j=strlen(desc); | |
| for (i=0; i<count; i++) { | |
| int found = 1; | |
| size_t k = strlen(list[i].desc); | |
| if (j != k) continue; | |
| for (k=0; k<j; k++) | |
| if (tolower(list[i].desc[k]) != tolower(desc[k])) | |
| found = 0; | |
| if (found == 0) continue; | |
| /* found a case-insensitive description match */ | |
| strcpy(name, list[i].name); | |
| return name; | |
| } | |
| /* not found */ | |
| return NULL; | |
| } | |
| /* strncasecmp() is not available on all platforms */ | |
| int eth_strncasecmp(char* string1, char* string2, size_t len) | |
| { | |
| size_t i; | |
| unsigned char s1, s2; | |
| for (i=0; i<len; i++) { | |
| s1 = string1[i]; | |
| s2 = string2[i]; | |
| if (islower (s1)) s1 = toupper (s1); | |
| if (islower (s2)) s2 = toupper (s2); | |
| if (s1 < s2) | |
| return -1; | |
| if (s1 > s2) | |
| return 1; | |
| if (s1 == 0) return 0; | |
| } | |
| return 0; | |
| } | |
| char* eth_getname_byname(char* name, char* temp) | |
| { | |
| ETH_LIST list[ETH_MAX_DEVICE]; | |
| int count = eth_devices(ETH_MAX_DEVICE, list); | |
| size_t n; | |
| int i, found; | |
| found = 0; | |
| n = strlen(name); | |
| for (i=0; i<count && !found; i++) { | |
| if ((n == strlen(list[i].name)) && | |
| (eth_strncasecmp(name, list[i].name, n) == 0)) { | |
| found = 1; | |
| strcpy(temp, list[i].name); /* only case might be different */ | |
| } | |
| } | |
| return (found ? temp : NULL); | |
| } | |
| char* eth_getdesc_byname(char* name, char* temp) | |
| { | |
| ETH_LIST list[ETH_MAX_DEVICE]; | |
| int count = eth_devices(ETH_MAX_DEVICE, list); | |
| size_t n; | |
| int i, found; | |
| found = 0; | |
| n = strlen(name); | |
| for (i=0; i<count && !found; i++) { | |
| if ((n == strlen(list[i].name)) && | |
| (eth_strncasecmp(name, list[i].name, n) == 0)) { | |
| found = 1; | |
| strcpy(temp, list[i].desc); | |
| } | |
| } | |
| return (found ? temp : NULL); | |
| } | |
| void eth_zero(ETH_DEV* dev) | |
| { | |
| /* set all members to NULL OR 0 */ | |
| memset(dev, 0, sizeof(ETH_DEV)); | |
| dev->reflections = -1; /* not established yet */ | |
| } | |
| static ETH_DEV **eth_open_devices = NULL; | |
| static int eth_open_device_count = 0; | |
| #if defined (USE_NETWORK) || defined (USE_SHARED) | |
| static void _eth_add_to_open_list (ETH_DEV* dev) | |
| { | |
| eth_open_devices = realloc(eth_open_devices, (eth_open_device_count+1)*sizeof(*eth_open_devices)); | |
| eth_open_devices[eth_open_device_count++] = dev; | |
| } | |
| static void _eth_remove_from_open_list (ETH_DEV* dev) | |
| { | |
| int i, j; | |
| for (i=0; i<eth_open_device_count; ++i) | |
| if (eth_open_devices[i] == dev) { | |
| for (j=i+1; j<eth_open_device_count; ++j) | |
| eth_open_devices[j-1] = eth_open_devices[j]; | |
| --eth_open_device_count; | |
| break; | |
| } | |
| } | |
| #endif | |
| t_stat eth_show (FILE* st, UNIT* uptr, int32 val, void* desc) | |
| { | |
| ETH_LIST list[ETH_MAX_DEVICE]; | |
| int number = eth_devices(ETH_MAX_DEVICE, list); | |
| fprintf(st, "ETH devices:\n"); | |
| if (number == -1) | |
| fprintf(st, " network support not available in simulator\n"); | |
| else | |
| if (number == 0) | |
| fprintf(st, " no network devices are available\n"); | |
| else { | |
| size_t min, len; | |
| int i; | |
| for (i=0, min=0; i<number; i++) | |
| if ((len = strlen(list[i].name)) > min) min = len; | |
| for (i=0; i<number; i++) | |
| fprintf(st," eth%d\t%-*s (%s)\n", i, (int)min, list[i].name, list[i].desc); | |
| } | |
| if (eth_open_device_count) { | |
| int i; | |
| char desc[ETH_DEV_DESC_MAX], *d; | |
| fprintf(st,"Open ETH Devices:\n"); | |
| for (i=0; i<eth_open_device_count; i++) { | |
| d = eth_getdesc_byname(eth_open_devices[i]->name, desc); | |
| if (d) | |
| fprintf(st, " %-7s%s (%s)\n", eth_open_devices[i]->dptr->name, eth_open_devices[i]->dptr->units[0].filename, d); | |
| else | |
| fprintf(st, " %-7s%s\n", eth_open_devices[i]->dptr->name, eth_open_devices[i]->dptr->units[0].filename); | |
| } | |
| } | |
| if (eth_open_device_count) { | |
| int i; | |
| char desc[ETH_DEV_DESC_MAX], *d; | |
| fprintf(st,"Open ETH Devices:\n"); | |
| for (i=0; i<eth_open_device_count; i++) { | |
| d = eth_getdesc_byname(eth_open_devices[i]->name, desc); | |
| if (d) | |
| fprintf(st, " %-7s%s (%s)\n", eth_open_devices[i]->dptr->name, eth_open_devices[i]->name, d); | |
| else | |
| fprintf(st, " %-7s%s\n", eth_open_devices[i]->dptr->name, eth_open_devices[i]->name); | |
| } | |
| } | |
| return SCPE_OK; | |
| } | |
| t_stat eth_show_devices (FILE* st, DEVICE *dptr, UNIT* uptr, int32 val, char* desc) | |
| { | |
| return eth_show (st, uptr, val, desc); | |
| } | |
| t_stat ethq_init(ETH_QUE* que, int max) | |
| { | |
| /* create dynamic queue if it does not exist */ | |
| if (!que->item) { | |
| que->item = (struct eth_item *) calloc(max, sizeof(struct eth_item)); | |
| if (!que->item) { | |
| /* failed to allocate memory */ | |
| char* msg = "EthQ: failed to allocate dynamic queue[%d]\r\n"; | |
| printf(msg, max); | |
| if (sim_log) fprintf(sim_log, msg, max); | |
| return SCPE_MEM; | |
| }; | |
| que->max = max; | |
| }; | |
| ethq_clear(que); | |
| return SCPE_OK; | |
| } | |
| t_stat ethq_destroy(ETH_QUE* que) | |
| { | |
| /* release dynamic queue if it exists */ | |
| ethq_clear(que); | |
| que->max = 0; | |
| if (que->item) { | |
| free(que->item); | |
| que->item = NULL; | |
| }; | |
| return SCPE_OK; | |
| } | |
| void ethq_clear(ETH_QUE* que) | |
| { | |
| int i; | |
| /* free up any extended packets */ | |
| for (i=0; i<que->max; ++i) | |
| if (que->item[i].packet.oversize) { | |
| free (que->item[i].packet.oversize); | |
| que->item[i].packet.oversize = NULL; | |
| } | |
| /* clear packet array */ | |
| memset(que->item, 0, sizeof(struct eth_item) * que->max); | |
| /* clear rest of structure */ | |
| que->count = que->head = que->tail = 0; | |
| } | |
| void ethq_remove(ETH_QUE* que) | |
| { | |
| struct eth_item* item = &que->item[que->head]; | |
| if (que->count) { | |
| if (item->packet.oversize) | |
| free (item->packet.oversize); | |
| memset(item, 0, sizeof(struct eth_item)); | |
| if (++que->head == que->max) | |
| que->head = 0; | |
| que->count--; | |
| } | |
| } | |
| void ethq_insert_data(ETH_QUE* que, int32 type, const uint8 *data, int used, size_t len, size_t crc_len, const uint8 *crc_data, int32 status) | |
| { | |
| struct eth_item* item; | |
| /* if queue empty, set pointers to beginning */ | |
| if (!que->count) { | |
| que->head = 0; | |
| que->tail = -1; | |
| } | |
| /* find new tail of the circular queue */ | |
| if (++que->tail == que->max) | |
| que->tail = 0; | |
| if (++que->count > que->max) { | |
| que->count = que->max; | |
| /* lose oldest packet */ | |
| if (++que->head == que->max) | |
| que->head = 0; | |
| que->loss++; | |
| } | |
| if (que->count > que->high) | |
| que->high = que->count; | |
| /* set information in (new) tail item */ | |
| item = &que->item[que->tail]; | |
| item->type = type; | |
| item->packet.len = len; | |
| item->packet.used = used; | |
| item->packet.crc_len = crc_len; | |
| if (len <= sizeof (item->packet.msg)) { | |
| memcpy(item->packet.msg, data, ((len > crc_len) ? len : crc_len)); | |
| if (crc_data && (crc_len > len)) | |
| memcpy(&item->packet.msg[len], crc_data, ETH_CRC_SIZE); | |
| } | |
| else { | |
| item->packet.oversize = realloc (item->packet.oversize, ((len > crc_len) ? len : crc_len)); | |
| memcpy(item->packet.oversize, data, ((len > crc_len) ? len : crc_len)); | |
| if (crc_data && (crc_len > len)) | |
| memcpy(&item->packet.oversize[len], crc_data, ETH_CRC_SIZE); | |
| } | |
| item->packet.status = status; | |
| } | |
| void ethq_insert(ETH_QUE* que, int32 type, ETH_PACK* pack, int32 status) | |
| { | |
| ethq_insert_data(que, type, pack->oversize ? pack->oversize : pack->msg, pack->used, pack->len, pack->crc_len, NULL, status); | |
| } | |
| /*============================================================================*/ | |
| /* Non-implemented versions */ | |
| /*============================================================================*/ | |
| #if !defined (USE_NETWORK) && !defined (USE_SHARED) | |
| t_stat eth_open(ETH_DEV* dev, char* name, DEVICE* dptr, uint32 dbit) | |
| {return SCPE_NOFNC;} | |
| t_stat eth_close (ETH_DEV* dev) | |
| {return SCPE_NOFNC;} | |
| t_stat eth_check_address_conflict (ETH_DEV* dev, | |
| ETH_MAC* const mac) | |
| {return SCPE_NOFNC;} | |
| t_stat eth_set_async (ETH_DEV *dev, int latency) | |
| {return SCPE_NOFNC;} | |
| t_stat eth_clr_async (ETH_DEV *dev) | |
| {return SCPE_NOFNC;} | |
| t_stat eth_write (ETH_DEV* dev, ETH_PACK* packet, ETH_PCALLBACK routine) | |
| {return SCPE_NOFNC;} | |
| int eth_read (ETH_DEV* dev, ETH_PACK* packet, ETH_PCALLBACK routine) | |
| {return SCPE_NOFNC;} | |
| t_stat eth_filter (ETH_DEV* dev, int addr_count, ETH_MAC* const addresses, | |
| ETH_BOOL all_multicast, ETH_BOOL promiscuous) | |
| {return SCPE_NOFNC;} | |
| t_stat eth_filter_hash (ETH_DEV* dev, int addr_count, ETH_MAC* const addresses, | |
| ETH_BOOL all_multicast, ETH_BOOL promiscuous, ETH_MULTIHASH* const hash) | |
| {return SCPE_NOFNC;} | |
| int eth_devices (int max, ETH_LIST* dev) | |
| {return -1;} | |
| void eth_show_dev (FILE* st, ETH_DEV* dev) | |
| {} | |
| #else /* endif unimplemented */ | |
| /*============================================================================*/ | |
| /* WIN32, Linux, and xBSD routines use WinPcap and libpcap packages */ | |
| /* OpenVMS Alpha uses a WinPcap port and an associated execlet */ | |
| /*============================================================================*/ | |
| #if defined (xBSD) || defined(__APPLE__) | |
| #include <sys/ioctl.h> | |
| #include <net/bpf.h> | |
| #endif /* xBSD */ | |
| #include <pcap.h> | |
| #include <string.h> | |
| #ifdef USE_TAP_NETWORK | |
| #if defined(__linux) | |
| #include <sys/ioctl.h> | |
| #include <net/if.h> | |
| #include <linux/if_tun.h> | |
| #elif defined(USE_BSDTUNTAP) | |
| #include <sys/types.h> | |
| #include <net/if_types.h> | |
| #include <net/if.h> | |
| #else /* We don't know how to do this on the current platform */ | |
| #undef USE_TAP_NETWORK | |
| #endif | |
| #endif /* USE_TAP_NETWORK */ | |
| #ifdef USE_VDE_NETWORK | |
| #include <libvdeplug.h> | |
| #endif /* USE_VDE_NETWORK */ | |
| /* Allows windows to look up user-defined adapter names */ | |
| #if defined(_WIN32) | |
| #include <winreg.h> | |
| #endif | |
| #ifdef HAVE_DLOPEN | |
| #include <dlfcn.h> | |
| #endif | |
| #if defined(USE_SHARED) && (defined(_WIN32) || defined(HAVE_DLOPEN)) | |
| /* Dynamic DLL loading technique and modified source comes from | |
| Etherial/WireShark capture_pcap.c */ | |
| /* Dynamic DLL load variables */ | |
| #ifdef _WIN32 | |
| static HINSTANCE hLib = 0; /* handle to DLL */ | |
| #else | |
| static void *hLib = 0; /* handle to Library */ | |
| #endif | |
| static int lib_loaded = 0; /* 0=not loaded, 1=loaded, 2=library load failed, 3=Func load failed */ | |
| static char* lib_name = | |
| #if defined(_WIN32) || defined(__CYGWIN__) | |
| "wpcap.dll"; | |
| #elif defined(__APPLE__) | |
| "/usr/lib/libpcap.A.dylib"; | |
| #else | |
| #define __STR_QUOTE(tok) #tok | |
| #define __STR(tok) __STR_QUOTE(tok) | |
| "libpcap." __STR(HAVE_DLOPEN); | |
| #endif | |
| static char* no_pcap = | |
| #if defined(_WIN32) || defined(__CYGWIN__) | |
| "wpcap load failure"; | |
| #else | |
| "libpcap load failure"; | |
| #endif | |
| /* define pointers to pcap functions needed */ | |
| static void (*p_pcap_close) (pcap_t *); | |
| static int (*p_pcap_compile) (pcap_t *, struct bpf_program *, const char *, int, bpf_u_int32); | |
| static int (*p_pcap_datalink) (pcap_t *); | |
| static int (*p_pcap_dispatch) (pcap_t *, int, pcap_handler, u_char *); | |
| static int (*p_pcap_findalldevs) (pcap_if_t **, char *); | |
| static void (*p_pcap_freealldevs) (pcap_if_t *); | |
| static void (*p_pcap_freecode) (struct bpf_program *); | |
| static char* (*p_pcap_geterr) (pcap_t *); | |
| static int (*p_pcap_lookupnet) (const char *, bpf_u_int32 *, bpf_u_int32 *, char *); | |
| static pcap_t* (*p_pcap_open_live) (const char *, int, int, int, char *); | |
| #ifdef _WIN32 | |
| static int (*p_pcap_setmintocopy) (pcap_t* handle, int); | |
| static HANDLE (*p_pcap_getevent) (pcap_t *); | |
| #else | |
| #ifdef MUST_DO_SELECT | |
| static int (*p_pcap_get_selectable_fd) (pcap_t *); | |
| #endif | |
| static int (*p_pcap_fileno) (pcap_t *); | |
| #endif | |
| static int (*p_pcap_sendpacket) (pcap_t* handle, const u_char* msg, int len); | |
| static int (*p_pcap_setfilter) (pcap_t *, struct bpf_program *); | |
| static char* (*p_pcap_lib_version) (void); | |
| /* load function pointer from DLL */ | |
| typedef int (*_func)(); | |
| static void load_function(char* function, _func* func_ptr) { | |
| #ifdef _WIN32 | |
| *func_ptr = (_func)((size_t)GetProcAddress(hLib, function)); | |
| #else | |
| *func_ptr = (_func)((size_t)dlsym(hLib, function)); | |
| #endif | |
| if (*func_ptr == 0) { | |
| char* msg = "Eth: Failed to find function '%s' in %s\r\n"; | |
| printf (msg, function, lib_name); | |
| if (sim_log) fprintf (sim_log, msg, function, lib_name); | |
| lib_loaded = 3; | |
| } | |
| } | |
| /* load wpcap.dll as required */ | |
| int load_pcap(void) { | |
| switch(lib_loaded) { | |
| case 0: /* not loaded */ | |
| /* attempt to load DLL */ | |
| #ifdef _WIN32 | |
| hLib = LoadLibraryA(lib_name); | |
| #else | |
| hLib = dlopen(lib_name, RTLD_NOW); | |
| #endif | |
| if (hLib == 0) { | |
| /* failed to load DLL */ | |
| char* msg = "Eth: Failed to load %s\r\n"; | |
| char* msg2 = | |
| #ifdef _WIN32 | |
| "Eth: You must install WinPcap 4.x to use networking\r\n"; | |
| #else | |
| "Eth: You must install libpcap to use networking\r\n"; | |
| #endif | |
| printf (msg, lib_name); | |
| printf ("%s", msg2); | |
| if (sim_log) { | |
| fprintf (sim_log, msg, lib_name); | |
| fprintf (sim_log, "%s", msg2); | |
| } | |
| lib_loaded = 2; | |
| break; | |
| } else { | |
| /* library loaded OK */ | |
| lib_loaded = 1; | |
| } | |
| /* load required functions; sets dll_load=3 on error */ | |
| load_function("pcap_close", (_func *) &p_pcap_close); | |
| load_function("pcap_compile", (_func *) &p_pcap_compile); | |
| load_function("pcap_datalink", (_func *) &p_pcap_datalink); | |
| load_function("pcap_dispatch", (_func *) &p_pcap_dispatch); | |
| load_function("pcap_findalldevs", (_func *) &p_pcap_findalldevs); | |
| load_function("pcap_freealldevs", (_func *) &p_pcap_freealldevs); | |
| load_function("pcap_freecode", (_func *) &p_pcap_freecode); | |
| load_function("pcap_geterr", (_func *) &p_pcap_geterr); | |
| load_function("pcap_lookupnet", (_func *) &p_pcap_lookupnet); | |
| load_function("pcap_open_live", (_func *) &p_pcap_open_live); | |
| #ifdef _WIN32 | |
| load_function("pcap_setmintocopy", (_func *) &p_pcap_setmintocopy); | |
| load_function("pcap_getevent", (_func *) &p_pcap_getevent); | |
| #else | |
| #ifdef MUST_DO_SELECT | |
| load_function("pcap_get_selectable_fd", (_func *) &p_pcap_get_selectable_fd); | |
| #endif | |
| load_function("pcap_fileno", (_func *) &p_pcap_fileno); | |
| #endif | |
| load_function("pcap_sendpacket", (_func *) &p_pcap_sendpacket); | |
| load_function("pcap_setfilter", (_func *) &p_pcap_setfilter); | |
| load_function("pcap_lib_version", (_func *) &p_pcap_lib_version); | |
| if (lib_loaded == 1) { | |
| /* log successful load */ | |
| char* version = p_pcap_lib_version(); | |
| printf("%s\n", version); | |
| if (sim_log) | |
| fprintf(sim_log, "%s\n", version); | |
| } | |
| break; | |
| default: /* loaded or failed */ | |
| break; | |
| } | |
| return (lib_loaded == 1) ? 1 : 0; | |
| } | |
| /* define functions with dynamic revectoring */ | |
| void pcap_close(pcap_t* a) { | |
| if (load_pcap() != 0) { | |
| p_pcap_close(a); | |
| } | |
| } | |
| /* OpenBSD has an ancient declaration of pcap_compile which doesn't have a const in the bpf string argument */ | |
| #if defined (__OpenBSD__) | |
| int pcap_compile(pcap_t* a, struct bpf_program* b, char* c, int d, bpf_u_int32 e) { | |
| #else | |
| int pcap_compile(pcap_t* a, struct bpf_program* b, const char* c, int d, bpf_u_int32 e) { | |
| #endif | |
| if (load_pcap() != 0) { | |
| return p_pcap_compile(a, b, c, d, e); | |
| } else { | |
| return 0; | |
| } | |
| } | |
| int pcap_datalink(pcap_t* a) { | |
| if (load_pcap() != 0) { | |
| return p_pcap_datalink(a); | |
| } else { | |
| return 0; | |
| } | |
| } | |
| int pcap_dispatch(pcap_t* a, int b, pcap_handler c, u_char* d) { | |
| if (load_pcap() != 0) { | |
| return p_pcap_dispatch(a, b, c, d); | |
| } else { | |
| return 0; | |
| } | |
| } | |
| int pcap_findalldevs(pcap_if_t** a, char* b) { | |
| if (load_pcap() != 0) { | |
| return p_pcap_findalldevs(a, b); | |
| } else { | |
| *a = 0; | |
| strcpy(b, no_pcap); | |
| return -1; | |
| } | |
| } | |
| void pcap_freealldevs(pcap_if_t* a) { | |
| if (load_pcap() != 0) { | |
| p_pcap_freealldevs(a); | |
| } | |
| } | |
| void pcap_freecode(struct bpf_program* a) { | |
| if (load_pcap() != 0) { | |
| p_pcap_freecode(a); | |
| } | |
| } | |
| char* pcap_geterr(pcap_t* a) { | |
| if (load_pcap() != 0) { | |
| return p_pcap_geterr(a); | |
| } else { | |
| return (char*) 0; | |
| } | |
| } | |
| int pcap_lookupnet(const char* a, bpf_u_int32* b, bpf_u_int32* c, char* d) { | |
| if (load_pcap() != 0) { | |
| return p_pcap_lookupnet(a, b, c, d); | |
| } else { | |
| return 0; | |
| } | |
| } | |
| pcap_t* pcap_open_live(const char* a, int b, int c, int d, char* e) { | |
| if (load_pcap() != 0) { | |
| return p_pcap_open_live(a, b, c, d, e); | |
| } else { | |
| return (pcap_t*) 0; | |
| } | |
| } | |
| #ifdef _WIN32 | |
| int pcap_setmintocopy(pcap_t* a, int b) { | |
| if (load_pcap() != 0) { | |
| return p_pcap_setmintocopy(a, b); | |
| } else { | |
| return 0; | |
| } | |
| } | |
| HANDLE pcap_getevent(pcap_t* a) { | |
| if (load_pcap() != 0) { | |
| return p_pcap_getevent(a); | |
| } else { | |
| return (HANDLE) 0; | |
| } | |
| } | |
| #else | |
| #ifdef MUST_DO_SELECT | |
| int pcap_get_selectable_fd(pcap_t* a) { | |
| if (load_pcap() != 0) { | |
| return p_pcap_get_selectable_fd(a); | |
| } else { | |
| return 0; | |
| } | |
| } | |
| #endif | |
| int pcap_fileno(pcap_t * a) { | |
| if (load_pcap() != 0) { | |
| return p_pcap_fileno(a); | |
| } else { | |
| return 0; | |
| } | |
| } | |
| #endif | |
| int pcap_sendpacket(pcap_t* a, const u_char* b, int c) { | |
| if (load_pcap() != 0) { | |
| return p_pcap_sendpacket(a, b, c); | |
| } else { | |
| return 0; | |
| } | |
| } | |
| int pcap_setfilter(pcap_t* a, struct bpf_program* b) { | |
| if (load_pcap() != 0) { | |
| return p_pcap_setfilter(a, b); | |
| } else { | |
| return 0; | |
| } | |
| } | |
| #endif | |
| /* Some platforms have always had pcap_sendpacket */ | |
| #if defined(_WIN32) || defined(__VMS) | |
| #define HAS_PCAP_SENDPACKET 1 | |
| #else | |
| /* The latest libpcap and WinPcap all have pcap_sendpacket */ | |
| #if !defined (NEED_PCAP_SENDPACKET) | |
| #define HAS_PCAP_SENDPACKET 1 | |
| #endif | |
| #endif | |
| #if !defined (HAS_PCAP_SENDPACKET) | |
| /* libpcap has no function to write a packet, so we need to implement | |
| pcap_sendpacket() for compatibility with the WinPcap base code. | |
| Return value: 0=Success, -1=Failure */ | |
| int pcap_sendpacket(pcap_t* handle, const u_char* msg, int len) | |
| { | |
| #if defined (__linux) | |
| return (send(pcap_fileno(handle), msg, len, 0) == len)? 0 : -1; | |
| #else | |
| return (write(pcap_fileno(handle), msg, len) == len)? 0 : -1; | |
| #endif /* linux */ | |
| } | |
| #endif /* !HAS_PCAP_SENDPACKET */ | |
| #if defined(_WIN32) || defined(__CYGWIN__) | |
| /* extracted from WinPcap's Packet32.h */ | |
| struct _PACKET_OID_DATA { | |
| uint32 Oid; ///< OID code. See the Microsoft DDK documentation or the file ntddndis.h | |
| ///< for a complete list of valid codes. | |
| uint32 Length; ///< Length of the data field | |
| uint8 Data[1]; ///< variable-lenght field that contains the information passed to or received | |
| ///< from the adapter. | |
| }; | |
| typedef struct _PACKET_OID_DATA PACKET_OID_DATA, *PPACKET_OID_DATA; | |
| typedef void **LPADAPTER; | |
| #define OID_802_3_CURRENT_ADDRESS 0x01010102 /* Extracted from ntddmdis.h */ | |
| static int pcap_mac_if_win32(char *AdapterName, unsigned char MACAddress[6]) | |
| { | |
| LPADAPTER lpAdapter; | |
| PPACKET_OID_DATA OidData; | |
| int Status; | |
| int ReturnValue; | |
| #ifdef _WIN32 | |
| HMODULE hDll; /* handle to DLL */ | |
| #else | |
| static void *hDll = NULL; /* handle to Library */ | |
| typedef int BOOLEAN; | |
| #endif | |
| LPADAPTER (*p_PacketOpenAdapter)(char *AdapterName); | |
| void (*p_PacketCloseAdapter)(LPADAPTER lpAdapter); | |
| int (*p_PacketRequest)(LPADAPTER AdapterObject,BOOLEAN Set,PPACKET_OID_DATA OidData); | |
| #ifdef _WIN32 | |
| hDll = LoadLibraryA("packet.dll"); | |
| p_PacketOpenAdapter = (LPADAPTER (*)(char *AdapterName))GetProcAddress(hDll, "PacketOpenAdapter"); | |
| p_PacketCloseAdapter = (void (*)(LPADAPTER lpAdapter))GetProcAddress(hDll, "PacketCloseAdapter"); | |
| p_PacketRequest = (int (*)(LPADAPTER AdapterObject,BOOLEAN Set,PPACKET_OID_DATA OidData))GetProcAddress(hDll, "PacketRequest"); | |
| #else | |
| hDll = dlopen("packet.dll", RTLD_NOW); | |
| p_PacketOpenAdapter = (LPADAPTER (*)(char *AdapterName))dlsym(hDll, "PacketOpenAdapter"); | |
| p_PacketCloseAdapter = (void (*)(LPADAPTER lpAdapter))dlsym(hDll, "PacketCloseAdapter"); | |
| p_PacketRequest = (int (*)(LPADAPTER AdapterObject,BOOLEAN Set,PPACKET_OID_DATA OidData))dlsym(hDll, "PacketRequest"); | |
| #endif | |
| /* Open the selected adapter */ | |
| lpAdapter = p_PacketOpenAdapter(AdapterName); | |
| if (!lpAdapter || (*lpAdapter == (void *)-1)) { | |
| #ifdef _WIN32 | |
| FreeLibrary(hDll); | |
| #else | |
| dlclose(hDll); | |
| #endif | |
| return -1; | |
| } | |
| /* Allocate a buffer to get the MAC adress */ | |
| OidData = malloc(6 + sizeof(PACKET_OID_DATA)); | |
| if (OidData == NULL) { | |
| p_PacketCloseAdapter(lpAdapter); | |
| #ifdef _WIN32 | |
| FreeLibrary(hDll); | |
| #else | |
| dlclose(hDll); | |
| #endif | |
| return -1; | |
| } | |
| /* Retrieve the adapter MAC querying the NIC driver */ | |
| OidData->Oid = OID_802_3_CURRENT_ADDRESS; | |
| OidData->Length = 6; | |
| memset(OidData->Data, 0, 6); | |
| Status = p_PacketRequest(lpAdapter, FALSE, OidData); | |
| if(Status) { | |
| memcpy(MACAddress, OidData->Data, 6); | |
| ReturnValue = 0; | |
| } else | |
| ReturnValue = -1; | |
| free(OidData); | |
| p_PacketCloseAdapter(lpAdapter); | |
| #ifdef _WIN32 | |
| FreeLibrary(hDll); | |
| #else | |
| dlclose(hDll); | |
| #endif | |
| return ReturnValue; | |
| } | |
| #endif /* defined(_WIN32) || defined(__CYGWIN__) */ | |
| #if defined (__VMS) && !defined(__VAX) | |
| #include <descrip.h> | |
| #include <iodef.h> | |
| #include <ssdef.h> | |
| #include <starlet.h> | |
| #include <stdio.h> | |
| #include <stsdef.h> | |
| #include <nmadef.h> | |
| static int pcap_mac_if_vms(char *AdapterName, unsigned char MACAddress[6]) | |
| { | |
| char VMS_Device[16]; | |
| $DESCRIPTOR(Device, VMS_Device); | |
| unsigned short iosb[4]; | |
| unsigned short *w; | |
| unsigned char *pha = NULL; | |
| unsigned char *hwa = NULL; | |
| int tmpval; | |
| int status; | |
| unsigned short characteristics[512]; | |
| long chardesc[] = {sizeof(characteristics), (long)&characteristics}; | |
| unsigned short chan; | |
| #pragma member_alignment save | |
| #pragma nomember_alignment | |
| static struct { | |
| short fmt; | |
| long val_fmt; | |
| short pty; | |
| long val_pty; | |
| short pad; | |
| long val_pad; | |
| } setup = { | |
| NMA$C_PCLI_FMT, NMA$C_LINFM_ETH, | |
| NMA$C_PCLI_PTY, 0x0090, | |
| NMA$C_PCLI_PAD, NMA$C_STATE_OFF, | |
| }; | |
| #pragma member_alignment restore | |
| long setupdesc[] = {sizeof(setup), (long)&setup}; | |
| /* Convert Interface Name to VMS Device Name */ | |
| /* This is a name shuffle */ | |
| /* WE0 becomes EWA0: */ | |
| /* SE1 becomes ESB0: */ | |
| /* XE0 becomes EXA0: */ | |
| tmpval = (int)(AdapterName[2]-'0'); | |
| if ((tmpval < 0) || (tmpval > 25)) | |
| return -1; | |
| VMS_Device[0] = toupper(AdapterName[1]); | |
| VMS_Device[1] = toupper(AdapterName[0]); | |
| VMS_Device[2] = 'A' + tmpval; | |
| VMS_Device[3] = '0'; | |
| VMS_Device[4] = '\0'; | |
| VMS_Device[5] = '\0'; | |
| Device.dsc$w_length = strlen(VMS_Device); | |
| if (!$VMS_STATUS_SUCCESS( sys$assign (&Device, &chan, 0, 0, 0) )) | |
| return -1; | |
| status = sys$qiow (0, chan, IO$_SETMODE|IO$M_CTRL|IO$M_STARTUP, &iosb, 0, 0, | |
| 0, &setupdesc, 0, 0, 0, 0); | |
| if ((!$VMS_STATUS_SUCCESS(status)) || (!$VMS_STATUS_SUCCESS(iosb[0]))) { | |
| sys$dassgn(chan); | |
| return -1; | |
| } | |
| status = sys$qiow (0, chan, IO$_SENSEMODE|IO$M_CTRL, &iosb, 0, 0, | |
| 0, &chardesc, 0, 0, 0, 0); | |
| sys$dassgn(chan); | |
| if ((!$VMS_STATUS_SUCCESS(status)) || (!$VMS_STATUS_SUCCESS(iosb[0]))) | |
| return -1; | |
| for (w=characteristics; w < &characteristics[iosb[1]]; ) { | |
| if ((((*w)&0xFFF) == NMA$C_PCLI_HWA) && (6 == *(w+1))) | |
| hwa = (unsigned char *)(w + 2); | |
| if ((((*w)&0xFFF) == NMA$C_PCLI_PHA) && (6 == *(w+1))) | |
| pha = (unsigned char *)(w + 2); | |
| if (((*w)&0x1000) == 0) | |
| w += 3; /* Skip over Longword Parameter */ | |
| else | |
| w += (2 + ((1 + *(w+1))/2)); /* Skip over String Parameter */ | |
| } | |
| if (pha != NULL) /* Prefer Physical Address */ | |
| memcpy(MACAddress, pha, 6); | |
| else | |
| if (hwa != NULL) /* Fallback to Hardware Address */ | |
| memcpy(MACAddress, hwa, 6); | |
| else | |
| return -1; | |
| return 0; | |
| } | |
| #endif | |
| static void eth_get_nic_hw_addr(ETH_DEV* dev, char *devname) | |
| { | |
| memset(&dev->host_nic_phy_hw_addr, 0, sizeof(dev->host_nic_phy_hw_addr)); | |
| dev->have_host_nic_phy_addr = 0; | |
| #if defined(_WIN32) || defined(__CYGWIN__) | |
| if (!pcap_mac_if_win32(devname, dev->host_nic_phy_hw_addr)) | |
| dev->have_host_nic_phy_addr = 1; | |
| #elif defined (__VMS) && !defined(__VAX) | |
| if (!pcap_mac_if_vms(devname, dev->host_nic_phy_hw_addr)) | |
| dev->have_host_nic_phy_addr = 1; | |
| #elif !defined(__CYGWIN__) && !defined(__VMS) | |
| if (1) { | |
| char command[1024]; | |
| FILE *f; | |
| int i; | |
| char *patterns[] = { | |
| "grep [0-9a-fA-F][0-9a-fA-F]:[0-9a-fA-F][0-9a-fA-F]:[0-9a-fA-F][0-9a-fA-F]:[0-9a-fA-F][0-9a-fA-F]:[0-9a-fA-F][0-9a-fA-F]:[0-9a-fA-F][0-9a-fA-F]", | |
| "egrep [0-9a-fA-F]?[0-9a-fA-F]:[0-9a-fA-F]?[0-9a-fA-F]:[0-9a-fA-F]?[0-9a-fA-F]:[0-9a-fA-F]?[0-9a-fA-F]:[0-9a-fA-F]?[0-9a-fA-F]:[0-9a-fA-F]?[0-9a-fA-F]", | |
| NULL}; | |
| if (0 == strncmp("vde:", devname, 4)) | |
| return; | |
| memset(command, 0, sizeof(command)); | |
| for (i=0; patterns[i] && (0 == dev->have_host_nic_phy_addr); ++i) { | |
| snprintf(command, sizeof(command)-1, "ifconfig %s | %s >NIC.hwaddr", devname, patterns[i]); | |
| (void)system(command); | |
| if (NULL != (f = fopen("NIC.hwaddr", "r"))) { | |
| while (0 == dev->have_host_nic_phy_addr) { | |
| if (fgets(command, sizeof(command)-1, f)) { | |
| char *p1, *p2; | |
| p1 = strchr(command, ':'); | |
| while (p1) { | |
| p2 = strchr(p1+1, ':'); | |
| if (p2 <= p1+3) { | |
| unsigned int mac_bytes[6]; | |
| if (6 == sscanf(p1-2, "%02x:%02x:%02x:%02x:%02x:%02x", &mac_bytes[0], &mac_bytes[1], &mac_bytes[2], &mac_bytes[3], &mac_bytes[4], &mac_bytes[5])) { | |
| dev->host_nic_phy_hw_addr[0] = mac_bytes[0]; | |
| dev->host_nic_phy_hw_addr[1] = mac_bytes[1]; | |
| dev->host_nic_phy_hw_addr[2] = mac_bytes[2]; | |
| dev->host_nic_phy_hw_addr[3] = mac_bytes[3]; | |
| dev->host_nic_phy_hw_addr[4] = mac_bytes[4]; | |
| dev->host_nic_phy_hw_addr[5] = mac_bytes[5]; | |
| dev->have_host_nic_phy_addr = 1; | |
| } | |
| break; | |
| } | |
| p1 = p2; | |
| } | |
| } | |
| else | |
| break; | |
| } | |
| fclose(f); | |
| remove("NIC.hwaddr"); | |
| } | |
| } | |
| } | |
| #endif | |
| } | |
| /* Forward declarations */ | |
| static void | |
| _eth_callback(u_char* info, const struct pcap_pkthdr* header, const u_char* data); | |
| static t_stat | |
| _eth_write(ETH_DEV* dev, ETH_PACK* packet, ETH_PCALLBACK routine); | |
| #if defined (USE_READER_THREAD) | |
| #include <pthread.h> | |
| static void * | |
| _eth_reader(void *arg) | |
| { | |
| ETH_DEV* volatile dev = (ETH_DEV*)arg; | |
| int status = 0; | |
| int sched_policy; | |
| struct sched_param sched_priority; | |
| #if defined (_WIN32) | |
| HANDLE hWait = pcap_getevent ((pcap_t*)dev->handle); | |
| #else | |
| int sel_ret; | |
| int do_select = 0; | |
| int select_fd = 0; | |
| switch (dev->eth_api) { | |
| case ETH_API_PCAP: | |
| #if defined (MUST_DO_SELECT) | |
| do_select = 1; | |
| select_fd = pcap_get_selectable_fd((pcap_t *)dev->handle); | |
| #endif | |
| break; | |
| case ETH_API_TAP: | |
| case ETH_API_VDE: | |
| do_select = 1; | |
| select_fd = dev->fd_handle; | |
| break; | |
| } | |
| #endif | |
| sim_debug(dev->dbit, dev->dptr, "Reader Thread Starting\n"); | |
| /* Boost Priority for this I/O thread vs the CPU instruction execution | |
| thread which in general won't be readily yielding the processor when | |
| this thread needs to run */ | |
| pthread_getschedparam (pthread_self(), &sched_policy, &sched_priority); | |
| ++sched_priority.sched_priority; | |
| pthread_setschedparam (pthread_self(), sched_policy, &sched_priority); | |
| while (dev->handle) { | |
| #if defined (_WIN32) | |
| if (WAIT_OBJECT_0 == WaitForSingleObject (hWait, 250)) { | |
| #else | |
| fd_set setl; | |
| struct timeval timeout; | |
| if (do_select) { | |
| FD_ZERO(&setl); | |
| FD_SET(select_fd, &setl); | |
| timeout.tv_sec = 0; | |
| timeout.tv_usec = 250*1000; | |
| sel_ret = select(1+select_fd, &setl, NULL, NULL, &timeout); | |
| } | |
| else | |
| sel_ret = 1; | |
| if (sel_ret < 0 && errno != EINTR) break; | |
| if (sel_ret > 0) { | |
| #endif /* _WIN32 */ | |
| if (!dev->handle) | |
| break; | |
| /* dispatch read request queue available packets */ | |
| switch (dev->eth_api) { | |
| case ETH_API_PCAP: | |
| status = pcap_dispatch ((pcap_t*)dev->handle, -1, &_eth_callback, (u_char*)dev); | |
| break; | |
| #ifdef USE_TAP_NETWORK | |
| case ETH_API_TAP: | |
| if (1) { | |
| struct pcap_pkthdr header; | |
| int len; | |
| u_char buf[ETH_MAX_JUMBO_FRAME]; | |
| memset(&header, 0, sizeof(header)); | |
| len = read(dev->fd_handle, buf, sizeof(buf)); | |
| if (len > 0) { | |
| status = 1; | |
| header.caplen = header.len = len; | |
| _eth_callback((u_char *)dev, &header, buf); | |
| } | |
| else | |
| status = 0; | |
| } | |
| break; | |
| #endif /* USE_TAP_NETWORK */ | |
| #ifdef USE_VDE_NETWORK | |
| case ETH_API_VDE: | |
| if (1) { | |
| struct pcap_pkthdr header; | |
| int len; | |
| u_char buf[ETH_MAX_JUMBO_FRAME]; | |
| memset(&header, 0, sizeof(header)); | |
| len = vde_recv((VDECONN *)dev->handle, buf, sizeof(buf), 0); | |
| if (len > 0) { | |
| status = 1; | |
| header.caplen = header.len = len; | |
| _eth_callback((u_char *)dev, &header, buf); | |
| } | |
| else | |
| status = 0; | |
| } | |
| break; | |
| #endif /* USE_VDE_NETWORK */ | |
| } | |
| if ((status > 0) && (dev->asynch_io)) { | |
| int wakeup_needed; | |
| pthread_mutex_lock (&dev->lock); | |
| wakeup_needed = (dev->read_queue.count != 0); | |
| pthread_mutex_unlock (&dev->lock); | |
| if (wakeup_needed) { | |
| sim_debug(dev->dbit, dev->dptr, "Queueing automatic poll\n"); | |
| sim_activate_abs (dev->dptr->units, dev->asynch_io_latency); | |
| } | |
| } | |
| } | |
| } | |
| sim_debug(dev->dbit, dev->dptr, "Reader Thread Exiting\n"); | |
| return NULL; | |
| } | |
| static void * | |
| _eth_writer(void *arg) | |
| { | |
| ETH_DEV* volatile dev = (ETH_DEV*)arg; | |
| struct write_request *request; | |
| int sched_policy; | |
| struct sched_param sched_priority; | |
| /* Boost Priority for this I/O thread vs the CPU instruction execution | |
| thread which in general won't be readily yielding the processor when | |
| this thread needs to run */ | |
| pthread_getschedparam (pthread_self(), &sched_policy, &sched_priority); | |
| ++sched_priority.sched_priority; | |
| pthread_setschedparam (pthread_self(), sched_policy, &sched_priority); | |
| sim_debug(dev->dbit, dev->dptr, "Writer Thread Starting\n"); | |
| pthread_mutex_lock (&dev->writer_lock); | |
| while (dev->handle) { | |
| pthread_cond_wait (&dev->writer_cond, &dev->writer_lock); | |
| while (NULL != (request = dev->write_requests)) { | |
| /* Pull buffer off request list */ | |
| dev->write_requests = request->next; | |
| pthread_mutex_unlock (&dev->writer_lock); | |
| dev->write_status = _eth_write(dev, &request->packet, NULL); | |
| pthread_mutex_lock (&dev->writer_lock); | |
| /* Put buffer on free buffer list */ | |
| request->next = dev->write_buffers; | |
| dev->write_buffers = request; | |
| } | |
| } | |
| pthread_mutex_unlock (&dev->writer_lock); | |
| sim_debug(dev->dbit, dev->dptr, "Writer Thread Exiting\n"); | |
| return NULL; | |
| } | |
| #endif | |
| t_stat eth_set_async (ETH_DEV *dev, int latency) | |
| { | |
| #if !defined(USE_READER_THREAD) || !defined(SIM_ASYNCH_IO) | |
| char *msg = "Eth: can't operate asynchronously, must poll\r\n"; | |
| printf ("%s", msg); | |
| if (sim_log) fprintf (sim_log, "%s", msg); | |
| return SCPE_NOFNC; | |
| #else | |
| int wakeup_needed; | |
| dev->asynch_io = 1; | |
| dev->asynch_io_latency = latency; | |
| pthread_mutex_lock (&dev->lock); | |
| wakeup_needed = (dev->read_queue.count != 0); | |
| pthread_mutex_unlock (&dev->lock); | |
| if (wakeup_needed) { | |
| sim_debug(dev->dbit, dev->dptr, "Queueing automatic poll\n"); | |
| sim_activate_abs (dev->dptr->units, dev->asynch_io_latency); | |
| } | |
| #endif | |
| return SCPE_OK; | |
| } | |
| t_stat eth_clr_async (ETH_DEV *dev) | |
| { | |
| #if !defined(USE_READER_THREAD) || !defined(SIM_ASYNCH_IO) | |
| return SCPE_NOFNC; | |
| #else | |
| /* make sure device exists */ | |
| if (!dev) return SCPE_UNATT; | |
| dev->asynch_io = 0; | |
| return SCPE_OK; | |
| #endif | |
| } | |
| t_stat eth_open(ETH_DEV* dev, char* name, DEVICE* dptr, uint32 dbit) | |
| { | |
| int bufsz = (BUFSIZ < ETH_MAX_PACKET) ? ETH_MAX_PACKET : BUFSIZ; | |
| char errbuf[PCAP_ERRBUF_SIZE]; | |
| char temp[1024]; | |
| char* savname = name; | |
| int num; | |
| char* msg; | |
| if (bufsz < ETH_MAX_JUMBO_FRAME) | |
| bufsz = ETH_MAX_JUMBO_FRAME; /* Enable handling of jumbo frames */ | |
| /* initialize device */ | |
| eth_zero(dev); | |
| /* translate name of type "ethX" to real device name */ | |
| if ((strlen(name) == 4) | |
| && (tolower(name[0]) == 'e') | |
| && (tolower(name[1]) == 't') | |
| && (tolower(name[2]) == 'h') | |
| && isdigit(name[3]) | |
| ) { | |
| num = atoi(&name[3]); | |
| savname = eth_getname(num, temp); | |
| if (savname == NULL) /* didn't translate */ | |
| return SCPE_OPENERR; | |
| } | |
| else { | |
| /* are they trying to use device description? */ | |
| savname = eth_getname_bydesc(name, temp); | |
| if (savname == NULL) { /* didn't translate */ | |
| /* probably is not ethX and has no description */ | |
| savname = eth_getname_byname(name, temp); | |
| if (savname == NULL) /* didn't translate */ | |
| savname = name; | |
| } | |
| } | |
| /* attempt to connect device */ | |
| memset(errbuf, 0, sizeof(errbuf)); | |
| if (0 == strncmp("tap:", savname, 4)) { | |
| int tun = -1; /* TUN/TAP Socket */ | |
| int on = 1; | |
| #if defined(USE_TAP_NETWORK) | |
| if (!strcmp(savname, "tap:tapN")) { | |
| msg = "Eth: Must specify actual tap device name (i.e. tap:tap0)\r\n"; | |
| printf (msg, errbuf); | |
| if (sim_log) fprintf (sim_log, msg, errbuf); | |
| return SCPE_OPENERR; | |
| } | |
| #endif | |
| #if defined(__linux) && defined(USE_TAP_NETWORK) | |
| if ((tun = open("/dev/net/tun", O_RDWR)) >= 0) { | |
| struct ifreq ifr; /* Interface Requests */ | |
| memset(&ifr, 0, sizeof(ifr)); | |
| /* Set up interface flags */ | |
| strcpy(ifr.ifr_name, savname+4); | |
| ifr.ifr_flags = IFF_TAP|IFF_NO_PI; | |
| /* Send interface requests to TUN/TAP driver. */ | |
| if (ioctl(tun, TUNSETIFF, &ifr) >= 0) { | |
| if (ioctl(tun, FIONBIO, &on)) { | |
| strncpy(errbuf, strerror(errno), sizeof(errbuf)-1); | |
| close(tun); | |
| } | |
| else { | |
| dev->fd_handle = tun; | |
| strcpy(savname, ifr.ifr_name); | |
| } | |
| } | |
| else | |
| strncpy(errbuf, strerror(errno), sizeof(errbuf)-1); | |
| } | |
| else | |
| strncpy(errbuf, strerror(errno), sizeof(errbuf)-1); | |
| #elif defined(USE_BSDTUNTAP) && defined(USE_TAP_NETWORK) | |
| if (1) { | |
| char dev_name[64] = ""; | |
| snprintf(dev_name, sizeof(dev_name)-1, "/dev/%s", savname+4); | |
| dev_name[sizeof(dev_name)-1] = '\0'; | |
| if ((tun = open(dev_name, O_RDWR)) >= 0) { | |
| if (ioctl(tun, FIONBIO, &on)) { | |
| strncpy(errbuf, strerror(errno), sizeof(errbuf)-1); | |
| close(tun); | |
| } | |
| else { | |
| dev->fd_handle = tun; | |
| strcpy(savname, savname+4); | |
| } | |
| #if defined (__APPLE__) | |
| if (1) { | |
| struct ifreq ifr; | |
| int s; | |
| memset (&ifr, 0, sizeof(ifr)); | |
| ifr.ifr_addr.sa_family = AF_INET; | |
| strncpy(ifr.ifr_name, savname, sizeof(ifr.ifr_name)); | |
| if ((s = socket(AF_INET, SOCK_DGRAM, 0)) >= 0) { | |
| if (ioctl(s, SIOCGIFFLAGS, (caddr_t)&ifr) >= 0) { | |
| ifr.ifr_flags |= IFF_UP; | |
| if (ioctl(s, SIOCSIFFLAGS, (caddr_t)&ifr)) { | |
| strncpy(errbuf, strerror(errno), sizeof(errbuf)-1); | |
| close(tun); | |
| } | |
| } | |
| close(s); | |
| } | |
| } | |
| #endif | |
| } | |
| else | |
| strncpy(errbuf, strerror(errno), sizeof(errbuf)-1); | |
| } | |
| #else | |
| strncpy(errbuf, "No support for tap: devices", sizeof(errbuf)-1); | |
| #endif /* !defined(__linux) && !defined(USE_BSDTUNTAP) */ | |
| if (0 == errbuf[0]) { | |
| dev->eth_api = ETH_API_TAP; | |
| dev->handle = (void *)1; /* Flag used to indicated open */ | |
| } | |
| } | |
| else | |
| if (0 == strncmp("vde:", savname, 4)) { | |
| #if defined(USE_VDE_NETWORK) | |
| struct vde_open_args voa; | |
| memset(&voa, 0, sizeof(voa)); | |
| if (!strcmp(savname, "vde:vdedevice")) { | |
| msg = "Eth: Must specify actual vde device name (i.e. vde:/tmp/switch)\r\n"; | |
| printf (msg, errbuf); | |
| if (sim_log) fprintf (sim_log, msg, errbuf); | |
| return SCPE_OPENERR; | |
| } | |
| if (!(dev->handle = (void*) vde_open(savname+4, "simh", &voa))) | |
| strncpy(errbuf, strerror(errno), sizeof(errbuf)-1); | |
| else { | |
| dev->eth_api = ETH_API_VDE; | |
| dev->fd_handle = vde_datafd((VDECONN*)dev->handle); | |
| } | |
| #else | |
| strncpy(errbuf, "No support for vde: network devices", sizeof(errbuf)-1); | |
| #endif /* !defined(__linux) && !defined(USE_BSDTUNTAP) */ | |
| } | |
| else { | |
| dev->handle = (void*) pcap_open_live(savname, bufsz, ETH_PROMISC, PCAP_READ_TIMEOUT, errbuf); | |
| if (!dev->handle) { /* can't open device */ | |
| msg = "Eth: pcap_open_live error - %s\r\n"; | |
| printf (msg, errbuf); | |
| if (sim_log) fprintf (sim_log, msg, errbuf); | |
| return SCPE_OPENERR; | |
| } | |
| dev->eth_api = ETH_API_PCAP; | |
| } | |
| if (errbuf[0]) { | |
| msg = "Eth: open error - %s\r\n"; | |
| printf (msg, errbuf); | |
| if (sim_log) fprintf (sim_log, msg, errbuf); | |
| return SCPE_OPENERR; | |
| } | |
| msg = "Eth: opened OS device %s\r\n"; | |
| printf (msg, savname); | |
| if (sim_log) fprintf (sim_log, msg, savname); | |
| /* get the NIC's hardware MAC address */ | |
| eth_get_nic_hw_addr(dev, savname); | |
| /* save name of device */ | |
| dev->name = malloc(strlen(savname)+1); | |
| strcpy(dev->name, savname); | |
| /* save debugging information */ | |
| dev->dptr = dptr; | |
| dev->dbit = dbit; | |
| #if !defined(HAS_PCAP_SENDPACKET) && defined (xBSD) && !defined (__APPLE__) | |
| /* Tell the kernel that the header is fully-formed when it gets it. | |
| This is required in order to fake the src address. */ | |
| if (dev->eth_api == ETH_API_PCAP) { | |
| int one = 1; | |
| ioctl(pcap_fileno(dev->handle), BIOCSHDRCMPLT, &one); | |
| } | |
| #endif /* xBSD */ | |
| #if defined (USE_READER_THREAD) | |
| if (1) { | |
| pthread_attr_t attr; | |
| #if defined(_WIN32) | |
| pcap_setmintocopy (dev->handle, 0); | |
| #endif | |
| ethq_init (&dev->read_queue, 200); /* initialize FIFO queue */ | |
| pthread_mutex_init (&dev->lock, NULL); | |
| pthread_mutex_init (&dev->writer_lock, NULL); | |
| pthread_mutex_init (&dev->self_lock, NULL); | |
| pthread_cond_init (&dev->writer_cond, NULL); | |
| pthread_attr_init(&attr); | |
| pthread_attr_setscope(&attr, PTHREAD_SCOPE_SYSTEM); | |
| pthread_create (&dev->reader_thread, &attr, _eth_reader, (void *)dev); | |
| pthread_create (&dev->writer_thread, &attr, _eth_writer, (void *)dev); | |
| pthread_attr_destroy(&attr); | |
| } | |
| #else /* !defined (USE_READER_THREAD */ | |
| #ifdef USE_SETNONBLOCK | |
| /* set ethernet device non-blocking so pcap_dispatch() doesn't hang */ | |
| if ((dev->eth_api == ETH_API_PCAP) && (pcap_setnonblock (dev->handle, 1, errbuf) == -1)) { | |
| msg = "Eth: Failed to set non-blocking: %s\r\n"; | |
| printf (msg, errbuf); | |
| if (sim_log) fprintf (sim_log, msg, errbuf); | |
| } | |
| #endif | |
| #endif /* !defined (USE_READER_THREAD */ | |
| #if defined (__APPLE__) | |
| if (dev->eth_api == ETH_API_PCAP) { | |
| /* Deliver packets immediately, needed for OS X 10.6.2 and later | |
| * (Snow-Leopard). | |
| * See this thread on libpcap and Mac Os X 10.6 Snow Leopard on | |
| * the tcpdump mailinglist: http://seclists.org/tcpdump/2010/q1/110 | |
| */ | |
| int v = 1; | |
| ioctl(pcap_fileno(dev->handle), BIOCIMMEDIATE, &v); | |
| } | |
| #endif | |
| _eth_add_to_open_list (dev); | |
| return SCPE_OK; | |
| } | |
| t_stat eth_close(ETH_DEV* dev) | |
| { | |
| char* msg = "Eth: closed %s\r\n"; | |
| pcap_t *pcap; | |
| int pcap_fd; | |
| /* make sure device exists */ | |
| if (!dev) return SCPE_UNATT; | |
| /* close the device */ | |
| pcap_fd = dev->fd_handle; /* save handle to possibly close later */ | |
| pcap = (pcap_t *)dev->handle; | |
| dev->handle = NULL; | |
| dev->fd_handle = 0; | |
| dev->have_host_nic_phy_addr = 0; | |
| #if defined (USE_READER_THREAD) | |
| pthread_join (dev->reader_thread, NULL); | |
| pthread_mutex_destroy (&dev->lock); | |
| pthread_cond_signal (&dev->writer_cond); | |
| pthread_join (dev->writer_thread, NULL); | |
| pthread_mutex_destroy (&dev->self_lock); | |
| pthread_mutex_destroy (&dev->writer_lock); | |
| pthread_cond_destroy (&dev->writer_cond); | |
| if (1) { | |
| struct write_request *buffer; | |
| while (NULL != (buffer = dev->write_buffers)) { | |
| dev->write_buffers = buffer->next; | |
| free(buffer); | |
| } | |
| while (NULL != (buffer = dev->write_requests)) { | |
| dev->write_requests = buffer->next; | |
| free(buffer); | |
| } | |
| } | |
| ethq_destroy (&dev->read_queue); /* release FIFO queue */ | |
| #endif | |
| switch (dev->eth_api) { | |
| case ETH_API_PCAP: | |
| pcap_close(pcap); | |
| break; | |
| #ifdef USE_TAP_NETWORK | |
| case ETH_API_TAP: | |
| close(pcap_fd); | |
| break; | |
| #endif | |
| #ifdef USE_VDE_NETWORK | |
| case ETH_API_VDE: | |
| vde_close((VDECONN*)pcap); | |
| break; | |
| #endif | |
| } | |
| printf (msg, dev->name); | |
| if (sim_log) fprintf (sim_log, msg, dev->name); | |
| /* clean up the mess */ | |
| free(dev->name); | |
| eth_zero(dev); | |
| _eth_remove_from_open_list (dev); | |
| return SCPE_OK; | |
| } | |
| t_stat eth_check_address_conflict (ETH_DEV* dev, | |
| ETH_MAC* const mac) | |
| { | |
| ETH_PACK send, recv; | |
| t_stat status; | |
| int responses = 0; | |
| char mac_string[32]; | |
| eth_mac_fmt(mac, mac_string); | |
| sim_debug(dev->dbit, dev->dptr, "Determining Address Conflict for MAC address: %s\n", mac_string); | |
| /* The process of checking address conflicts is used in two ways: | |
| 1) to determine the behavior of the currently running packet | |
| delivery facility regarding whether it may receive copies | |
| of every packet sent (and how many). | |
| 2) to verify if a MAC address which this facility is planning | |
| to use as the source address of packets is already in use | |
| by some other node on the local network | |
| Case #1, doesn't require (and explicitly doesn't want) any | |
| interaction or response from other systems on the LAN so | |
| therefore no considerations regarding switch packet forwarding | |
| are important. Meanwhile, Case #2 does require responses from | |
| other components on the LAN to provide useful functionality. | |
| The original designers of this mechanism did this when essentially | |
| all LANs were single collision domains (i.e. ALL nodes which might | |
| be affected by an address conflict were physically present on a single | |
| Ethernet cable which might have been extended by a couple of repeaters). | |
| Since that time, essentially no networks are single collision domains. | |
| Thick and thinwire Ethernet cables donÂ’t exist and very few networks | |
| even have hubs. Today, essentially all LANs are deployed using one | |
| or more layers of network switches. In a switched LAN environment, the | |
| switches on the LAN ‘learn’ which ports on the LAN source traffic from | |
| which MAC addresses and then forward traffic destined for particular | |
| MAC address to the appropriate ports. If a particular MAC address is | |
| already in use somewhere on the LAN, then the switches ‘know’ where | |
| it is. The host based test using the loopback protocol is poorly | |
| designed to detect this condition. This test is performed by the host | |
| first changing the deviceÂ’s Physical MAC address to the address which | |
| is to be tested, and then sending a loopback packet FROM AND TO this | |
| MAC address with a loopback reply to be sent by a system which may be | |
| currently using the MAC address. If no reply is received, then the | |
| MAC address is presumed to be unused. The sending of this packet will | |
| result in its delivery to the right system since the switch port/MAC | |
| address tables know where to deliver packets destined to this MAC | |
| address, however the response it generates wonÂ’t be delivered to the | |
| system performing the test since the switches on the LAN wonÂ’t know | |
| about the local port being the right target for packets with this MAC | |
| address. A better test design to detect these conflicts would be for | |
| the testing system to send a loopback packet FROM the current physical | |
| MAC address (BEFORE changing it) TO the MAC address being tested with | |
| the loopback response coming to the current physical MAC address of | |
| the device. If a response is received, then the address is in use and | |
| the attempt to change the deviceÂ’s MAC address should fail. Since we | |
| canÂ’t change the software running in these simulators to implement this | |
| better conflict detection approach, we can still ‘do the right thing’ | |
| in the sim_ether layer. WeÂ’re already handling the loopback test | |
| packets specially since we always had to avoid receiving the packets | |
| which were being sent, but needed to allow for the incoming loopback | |
| packets to be properly dealt with. We can extend this current special | |
| handling to change outgoing ‘loopback to self’ packets to have source | |
| AND loopback destination addresses in the packets to be the host NICÂ’s | |
| physical address. The switch network will already know the correct | |
| MAC/port relationship for the host NICÂ’s physical address, so loopback | |
| response packets will be delivered as needed. | |
| Code in _eth_write and _eth_callback provide the special handling to | |
| perform the described loopback packet adjustments, and code in | |
| eth_filter_hash makes sure that the loopback response packets are received. | |
| */ | |
| /* build a loopback forward request packet */ | |
| memset (&send, 0, sizeof(ETH_PACK)); | |
| send.len = ETH_MIN_PACKET; /* minimum packet size */ | |
| memcpy(&send.msg[0], mac, sizeof(ETH_MAC)); /* target address */ | |
| memcpy(&send.msg[6], mac, sizeof(ETH_MAC)); /* source address */ | |
| send.msg[12] = 0x90; /* loopback packet type */ | |
| send.msg[14] = 0; /* Offset */ | |
| send.msg[16] = 2; /* Forward */ | |
| memcpy(&send.msg[18], mac, sizeof(ETH_MAC)); /* Forward Destination */ | |
| send.msg[24] = 1; /* Reply */ | |
| eth_filter(dev, 1, (ETH_MAC *)mac, 0, 0); | |
| /* send the packet */ | |
| status = _eth_write (dev, &send, NULL); | |
| if (status != SCPE_OK) { | |
| char *msg; | |
| msg = (dev->eth_api == ETH_API_PCAP) ? | |
| "Eth: Error Transmitting packet: %s\r\n" | |
| "You may need to run as root, or install a libpcap version\r\n" | |
| "which is at least 0.9 from your OS vendor or www.tcpdump.org\r\n" : | |
| "Eth: Error Transmitting packet: %s\r\n" | |
| "You may need to run as root.\r\n"; | |
| printf(msg, strerror(errno)); | |
| if (sim_log) fprintf (sim_log, msg, strerror(errno)); | |
| return status; | |
| } | |
| sim_os_ms_sleep (300); /* time for a conflicting host to respond */ | |
| /* empty the read queue and count the responses */ | |
| do { | |
| memset (&recv, 0, sizeof(ETH_PACK)); | |
| status = eth_read (dev, &recv, NULL); | |
| if (((0 == memcmp(send.msg+12, recv.msg+12, 2)) && /* Protocol Match */ | |
| (0 == memcmp(send.msg, recv.msg+6, 6)) && /* Source Match */ | |
| (0 == memcmp(send.msg+6, recv.msg, 6))) || /* Destination Match */ | |
| (0 == memcmp(send.msg, recv.msg, 14))) /* Packet Match (Reflection) */ | |
| responses++; | |
| } while (recv.len > 0); | |
| sim_debug(dev->dbit, dev->dptr, "Address Conflict = %d\n", responses); | |
| return responses; | |
| } | |
| t_stat eth_reflect(ETH_DEV* dev) | |
| { | |
| /* Test with an address no NIC should have. */ | |
| /* We do this to avoid reflections from the wire, */ | |
| /* in the event that a simulated NIC has a MAC address conflict. */ | |
| ETH_MAC mac = {0xfe,0xff,0xff,0xff,0xff,0xfe}; | |
| sim_debug(dev->dbit, dev->dptr, "Determining Reflections...\n"); | |
| dev->reflections = 0; | |
| dev->reflections = eth_check_address_conflict (dev, &mac); | |
| sim_debug(dev->dbit, dev->dptr, "Reflections = %d\n", dev->reflections); | |
| return dev->reflections; | |
| } | |
| static | |
| t_stat _eth_write(ETH_DEV* dev, ETH_PACK* packet, ETH_PCALLBACK routine) | |
| { | |
| int status = 1; /* default to failure */ | |
| /* make sure device exists */ | |
| if (!dev) return SCPE_UNATT; | |
| /* make sure packet exists */ | |
| if (!packet) return SCPE_ARG; | |
| /* make sure packet is acceptable length */ | |
| if ((packet->len >= ETH_MIN_PACKET) && (packet->len <= ETH_MAX_PACKET)) { | |
| int loopback_self_frame = LOOPBACK_SELF_FRAME(packet->msg, packet->msg); | |
| eth_packet_trace (dev, packet->msg, packet->len, "writing"); | |
| /* record sending of loopback packet (done before actual send to avoid race conditions with receiver) */ | |
| if (loopback_self_frame) { | |
| if (dev->have_host_nic_phy_addr) { | |
| memcpy(&packet->msg[6], dev->host_nic_phy_hw_addr, sizeof(ETH_MAC)); | |
| memcpy(&packet->msg[18], dev->host_nic_phy_hw_addr, sizeof(ETH_MAC)); | |
| } | |
| #ifdef USE_READER_THREAD | |
| pthread_mutex_lock (&dev->self_lock); | |
| #endif | |
| dev->loopback_self_sent += dev->reflections; | |
| dev->loopback_self_sent_total++; | |
| #ifdef USE_READER_THREAD | |
| pthread_mutex_unlock (&dev->self_lock); | |
| #endif | |
| } | |
| /* dispatch write request (synchronous; no need to save write info to dev) */ | |
| switch (dev->eth_api) { | |
| case ETH_API_PCAP: | |
| status = pcap_sendpacket((pcap_t*)dev->handle, (u_char*)packet->msg, packet->len); | |
| break; | |
| #ifdef USE_TAP_NETWORK | |
| case ETH_API_TAP: | |
| status = (((int)packet->len == write(dev->fd_handle, (void *)packet->msg, packet->len)) ? 0 : -1); | |
| break; | |
| #endif | |
| #ifdef USE_VDE_NETWORK | |
| case ETH_API_VDE: | |
| status = vde_send((VDECONN*)dev->handle, (void *)packet->msg, packet->len, 0); | |
| if ((status == (int)packet->len) || (status == 0)) | |
| status = 0; | |
| else | |
| if ((status == -1) && ((errno == EAGAIN) || (errno == EWOULDBLOCK))) | |
| status = 0; | |
| else | |
| status = 1; | |
| break; | |
| #endif | |
| } | |
| /* On error, correct loopback bookkeeping */ | |
| if ((status != 0) && loopback_self_frame) { | |
| #ifdef USE_READER_THREAD | |
| pthread_mutex_lock (&dev->self_lock); | |
| #endif | |
| dev->loopback_self_sent -= dev->reflections; | |
| dev->loopback_self_sent_total--; | |
| #ifdef USE_READER_THREAD | |
| pthread_mutex_unlock (&dev->self_lock); | |
| #endif | |
| } | |
| } /* if packet->len */ | |
| /* call optional write callback function */ | |
| if (routine) | |
| (routine)(status); | |
| return ((status == 0) ? SCPE_OK : SCPE_IOERR); | |
| } | |
| t_stat eth_write(ETH_DEV* dev, ETH_PACK* packet, ETH_PCALLBACK routine) | |
| { | |
| #ifdef USE_READER_THREAD | |
| struct write_request *request; | |
| int write_queue_size = 1; | |
| /* make sure device exists */ | |
| if (!dev) return SCPE_UNATT; | |
| /* Get a buffer */ | |
| pthread_mutex_lock (&dev->writer_lock); | |
| if (NULL != (request = dev->write_buffers)) | |
| dev->write_buffers = request->next; | |
| pthread_mutex_unlock (&dev->writer_lock); | |
| if (NULL == request) | |
| request = malloc(sizeof(*request)); | |
| /* Copy buffer contents */ | |
| request->packet.len = packet->len; | |
| request->packet.used = packet->used; | |
| request->packet.status = packet->status; | |
| request->packet.crc_len = packet->crc_len; | |
| memcpy(request->packet.msg, packet->msg, packet->len); | |
| /* Insert buffer at the end of the write list (to make sure that */ | |
| /* packets make it to the wire in the order they were presented here) */ | |
| pthread_mutex_lock (&dev->writer_lock); | |
| request->next = NULL; | |
| if (dev->write_requests) { | |
| struct write_request *last_request = dev->write_requests; | |
| ++write_queue_size; | |
| while (last_request->next) { | |
| last_request = last_request->next; | |
| ++write_queue_size; | |
| } | |
| last_request->next = request; | |
| } | |
| else | |
| dev->write_requests = request; | |
| if (write_queue_size > dev->write_queue_peak) | |
| dev->write_queue_peak = write_queue_size; | |
| pthread_mutex_unlock (&dev->writer_lock); | |
| /* Awaken writer thread to perform actual write */ | |
| pthread_cond_signal (&dev->writer_cond); | |
| /* Return with a status from some prior write */ | |
| if (routine) | |
| (routine)(dev->write_status); | |
| return dev->write_status; | |
| #else | |
| return _eth_write(dev, packet, routine); | |
| #endif | |
| } | |
| static int | |
| _eth_hash_lookup(ETH_MULTIHASH hash, const u_char* data) | |
| { | |
| int key = 0x3f & (eth_crc32(0, data, 6) >> 26); | |
| key ^= 0x3f; | |
| return (hash[key>>3] & (1 << (key&0x7))); | |
| } | |
| #if 0 | |
| static int | |
| _eth_hash_validate(ETH_MAC *MultiCastList, int count, ETH_MULTIHASH hash) | |
| { | |
| ETH_MULTIHASH lhash; | |
| int i; | |
| memset(lhash, 0, sizeof(lhash)); | |
| for (i=0; i<count; ++i) { | |
| int key = 0x3f & (eth_crc32(0, MultiCastList[i], 6) >> 26); | |
| key ^= 0x3F; | |
| printf("MAC: %02X:%02X:%02X:%02X:%02X:%02X Key: %X, Byte: %X, Val: %X\n", | |
| MultiCastList[i][0], MultiCastList[i][1], MultiCastList[i][2], MultiCastList[i][3], MultiCastList[i][4], MultiCastList[i][5], | |
| key, key>>3, (1 << (key&0x7))); | |
| lhash[key>>3] |= (1 << (key&0x7)); | |
| } | |
| if (memcmp(hash, lhash, sizeof(lhash))) { | |
| printf("Inconsistent Computed Hash:\n"); | |
| printf("Should be: %02X %02X %02X %02X %02X %02X %02X %02X\n", | |
| hash[0], hash[1], hash[2], hash[3], | |
| hash[4], hash[5], hash[6], hash[7]); | |
| printf("Was: %02X %02X %02X %02X %02X %02X %02X %02X\n", | |
| lhash[0], lhash[1], lhash[2], lhash[3], | |
| lhash[4], lhash[5], lhash[6], lhash[7]); | |
| } | |
| else { | |
| printf("Should be: %02X %02X %02X %02X %02X %02X %02X %02X\n", | |
| hash[0], hash[1], hash[2], hash[3], | |
| hash[4], hash[5], hash[6], hash[7]); | |
| printf("Was: %02X %02X %02X %02X %02X %02X %02X %02X\n", | |
| lhash[0], lhash[1], lhash[2], lhash[3], | |
| lhash[4], lhash[5], lhash[6], lhash[7]); | |
| } | |
| return 0; | |
| } | |
| static void | |
| _eth_test_multicast_hash() | |
| { | |
| ETH_MAC tMacs[] = { | |
| {0xAB, 0x00, 0x04, 0x01, 0xAC, 0x10}, | |
| {0xAB, 0x00, 0x00, 0x04, 0x00, 0x00}, | |
| {0x09, 0x00, 0x2B, 0x00, 0x00, 0x0F}, | |
| {0x09, 0x00, 0x2B, 0x02, 0x01, 0x04}, | |
| {0x09, 0x00, 0x2B, 0x02, 0x01, 0x07}, | |
| {0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF}, | |
| {0x01, 0x00, 0x5E, 0x00, 0x00, 0x01}}; | |
| ETH_MULTIHASH thash = {0x01, 0x40, 0x00, 0x00, 0x48, 0x88, 0x40, 0x00}; | |
| _eth_hash_validate(tMacs, sizeof(tMacs)/sizeof(tMacs[0]), thash); | |
| } | |
| #endif | |
| /* The IP header */ | |
| struct IPHeader { | |
| uint8 verhlen; /* Version & Header Length in dwords */ | |
| #define IP_HLEN(IP) (((IP)->verhlen&0xF)<<2) /* Header Length in Bytes */ | |
| #define IP_VERSION(IP) ((((IP)->verhlen)>>4)&0xF) /* IP Version */ | |
| uint8 tos; /* Type of service */ | |
| uint16 total_len; /* Length of the packet in dwords */ | |
| uint16 ident; /* unique identifier */ | |
| uint16 flags; /* Fragmentation Flags */ | |
| #define IP_DF_FLAG (0x4000) | |
| #define IP_MF_FLAG (0x2000) | |
| #define IP_OFFSET_MASK (0x1FFF) | |
| #define IP_FRAG_DF(IP) (ntohs(((IP)->flags))&IP_DF_FLAG) | |
| #define IP_FRAG_MF(IP) (ntohs(((IP)->flags))&IP_MF_FLAG) | |
| #define IP_FRAG_OFFSET(IP) (ntohs(((IP)->flags))&IP_OFFSET_MASK) | |
| uint8 ttl; /* Time to live */ | |
| uint8 proto; /* Protocol number (TCP, UDP etc) */ | |
| uint16 checksum; /* IP checksum */ | |
| uint32 source_ip; /* Source Address */ | |
| uint32 dest_ip; /* Destination Address */ | |
| }; | |
| /* ICMP header */ | |
| struct ICMPHeader { | |
| uint8 type; /* ICMP packet type */ | |
| uint8 code; /* Type sub code */ | |
| uint16 checksum; /* ICMP Checksum */ | |
| uint32 otherstuff[1];/* optional data */ | |
| }; | |
| struct UDPHeader { | |
| uint16 source_port; | |
| uint16 dest_port; | |
| uint16 length; /* The length of the entire UDP datagram, including both header and Data fields. */ | |
| uint16 checksum; | |
| }; | |
| struct TCPHeader { | |
| uint16 source_port; | |
| uint16 dest_port; | |
| uint32 sequence_number; | |
| uint32 acknowledgement_number; | |
| uint16 data_offset_and_flags; | |
| #define TCP_DATA_OFFSET(TCP) ((ntohs((TCP)->data_offset_and_flags)>>12)<<2) | |
| #define TCP_CWR_FLAG (0x80) | |
| #define TCP_ECR_FLAG (0x40) | |
| #define TCP_URG_FLAG (0x20) | |
| #define TCP_ACK_FLAG (0x10) | |
| #define TCP_PSH_FLAG (0x08) | |
| #define TCP_RST_FLAG (0x04) | |
| #define TCP_SYN_FLAG (0x02) | |
| #define TCP_FIN_FLAG (0x01) | |
| #define TCP_FLAGS_MASK (0xFFF) | |
| uint16 window; | |
| uint16 checksum; | |
| uint16 urgent; | |
| uint16 otherstuff[1]; /* The rest of the packet */ | |
| }; | |
| #ifndef IPPROTO_TCP | |
| #define IPPROTO_TCP 6 /* tcp */ | |
| #endif | |
| #ifndef IPPROTO_UDP | |
| #define IPPROTO_UDP 17 /* user datagram protocol */ | |
| #endif | |
| #ifndef IPPROTO_ICMP | |
| #define IPPROTO_ICMP 1 /* control message protocol */ | |
| #endif | |
| static uint16 | |
| ip_checksum(uint16 *buffer, int size) | |
| { | |
| unsigned long cksum = 0; | |
| /* Sum all the words together, adding the final byte if size is odd */ | |
| while (size > 1) { | |
| cksum += *buffer++; | |
| size -= sizeof(*buffer); | |
| } | |
| if (size) { | |
| uint16 endword; | |
| uint8 *endbytes = (uint8 *)&endword; | |
| endbytes[0] = *((uint8 *)buffer); | |
| endbytes[1] = 0; | |
| cksum += endword; | |
| } | |
| /* Do a little shuffling */ | |
| cksum = (cksum >> 16) + (cksum & 0xffff); | |
| cksum += (cksum >> 16); | |
| /* Return the bitwise complement of the resulting mishmash */ | |
| return (uint16)(~cksum); | |
| } | |
| static uint16 | |
| pseudo_checksum(uint16 len, uint16 proto, uint16 *src_addr, uint16 *dest_addr, uint8 *buff) | |
| { | |
| uint32 sum; | |
| /* Sum the data first */ | |
| sum = 0xffff&(~ip_checksum((uint16 *)buff, len)); | |
| /* add the pseudo header which contains the IP source and destinationn addresses */ | |
| sum += src_addr[0]; | |
| sum += src_addr[1]; | |
| sum += dest_addr[0]; | |
| sum += dest_addr[1]; | |
| /* and the protocol number and the length of the UDP packet */ | |
| sum = sum + htons(proto) + htons(len); | |
| /* Do a little shuffling */ | |
| sum = (sum >> 16) + (sum & 0xffff); | |
| sum += (sum >> 16); | |
| /* Return the bitwise complement of the resulting mishmash */ | |
| return (uint16)(~sum); | |
| } | |
| static void | |
| _eth_fix_ip_jumbo_offload(ETH_DEV* dev, const u_char* msg, int len) | |
| { | |
| unsigned short* proto = (unsigned short*) &msg[12]; | |
| struct IPHeader *IP; | |
| struct TCPHeader *TCP = NULL; | |
| struct UDPHeader *UDP; | |
| struct ICMPHeader *ICMP; | |
| uint16 orig_checksum; | |
| uint16 payload_len; | |
| uint16 mtu_payload; | |
| uint16 ip_flags; | |
| uint16 frag_offset; | |
| struct pcap_pkthdr header; | |
| uint16 orig_tcp_flags; | |
| /* Only interested in IP frames */ | |
| if (ntohs(*proto) != 0x0800) { | |
| ++dev->jumbo_dropped; /* Non IP Frames are dropped */ | |
| return; | |
| } | |
| IP = (struct IPHeader *)&msg[14]; | |
| if (IP_VERSION(IP) != 4) { | |
| ++dev->jumbo_dropped; /* Non IPv4 jumbo frames are dropped */ | |
| return; | |
| } | |
| if ((IP_HLEN(IP) > len) || (ntohs(IP->total_len) > len)) { | |
| ++dev->jumbo_dropped; /* Bogus header length frames are dropped */ | |
| return; | |
| } | |
| if (IP_FRAG_OFFSET(IP) || IP_FRAG_MF(IP)) { | |
| ++dev->jumbo_dropped; /* Previously fragmented, but currently jumbo sized frames are dropped */ | |
| return; | |
| } | |
| switch (IP->proto) { | |
| case IPPROTO_UDP: | |
| UDP = (struct UDPHeader *)(((char *)IP)+IP_HLEN(IP)); | |
| if (ntohs(UDP->length) > (len-IP_HLEN(IP))) { | |
| ++dev->jumbo_dropped; /* Bogus UDP packet length (packet contained length exceeds packet size) frames are dropped */ | |
| return; | |
| } | |
| if (UDP->checksum == 0) | |
| break; /* UDP Checksums are disabled */ | |
| orig_checksum = UDP->checksum; | |
| UDP->checksum = 0; | |
| UDP->checksum = pseudo_checksum(ntohs(UDP->length), IPPROTO_UDP, (uint16 *)(&IP->source_ip), (uint16 *)(&IP->dest_ip), (uint8 *)UDP); | |
| if (orig_checksum != UDP->checksum) | |
| eth_packet_trace (dev, msg, len, "reading jumbo UDP header Checksum Fixed"); | |
| break; | |
| case IPPROTO_ICMP: | |
| ICMP = (struct ICMPHeader *)(((char *)IP)+IP_HLEN(IP)); | |
| orig_checksum = ICMP->checksum; | |
| ICMP->checksum = 0; | |
| ICMP->checksum = ip_checksum((uint16 *)ICMP, ntohs(IP->total_len)-IP_HLEN(IP)); | |
| if (orig_checksum != ICMP->checksum) | |
| eth_packet_trace (dev, msg, len, "reading jumbo ICMP header Checksum Fixed"); | |
| break; | |
| case IPPROTO_TCP: | |
| TCP = (struct TCPHeader *)(((char *)IP)+IP_HLEN(IP)); | |
| if ((TCP_DATA_OFFSET(TCP) > (len-IP_HLEN(IP))) || (TCP_DATA_OFFSET(TCP) < 20)) { | |
| ++dev->jumbo_dropped; /* Bogus TCP packet header length (packet contained length exceeds packet size) frames are dropped */ | |
| return; | |
| } | |
| /* We don't do anything with the TCP checksum since we're going to resegment the TCP data below */ | |
| break; | |
| default: | |
| ++dev->jumbo_dropped; /* We onlt handle UDP, ICMP and TCP jumbo frames others are dropped */ | |
| return; | |
| } | |
| /* Reasonable Checksums are now in the jumbo packet, but we've got to actually */ | |
| /* deliver ONLY standard sized ethernet frames. Our job here is to now act as */ | |
| /* a router might have to and fragment these IPv4 frames as they are delivered */ | |
| /* into the virtual NIC. We do this by walking down the packet and dispatching */ | |
| /* a chunk at a time recomputing an appropriate header for each chunk. For */ | |
| /* datagram oriented protocols (UDP and ICMP) this is done by simple packet */ | |
| /* fragmentation. For TCP this is done by breaking large packets into separate */ | |
| /* TCP packets. */ | |
| memset(&header, 0, sizeof(header)); | |
| switch (IP->proto) { | |
| case IPPROTO_UDP: | |
| case IPPROTO_ICMP: | |
| ++dev->jumbo_fragmented; | |
| /* When we're performing LSO (Large Send Offload), we're given a | |
| 'template' header which may not include a value being populated | |
| in the IP header length (which is only 16 bits). | |
| We process as payload everything which isn't known header data. */ | |
| payload_len = len - (14 + IP_HLEN(IP)); | |
| mtu_payload = ETH_MIN_JUMBO_FRAME - (14 + IP_HLEN(IP)); | |
| frag_offset = 0; | |
| while (payload_len > 0) { | |
| ip_flags = frag_offset; | |
| if (payload_len > mtu_payload) { | |
| ip_flags |= IP_MF_FLAG; | |
| IP->total_len = htons(((mtu_payload>>3)<<3) + IP_HLEN(IP)); | |
| } | |
| else { | |
| IP->total_len = htons(payload_len + IP_HLEN(IP)); | |
| } | |
| IP->flags = htons(ip_flags); | |
| IP->checksum = 0; | |
| IP->checksum = ip_checksum((uint16 *)IP, IP_HLEN(IP)); | |
| header.caplen = header.len = 14 + ntohs(IP->total_len); | |
| eth_packet_trace (dev, ((u_char *)IP)-14, header.len, "reading Datagram fragment"); | |
| #if ETH_MIN_JUMBO_FRAME < ETH_MAX_PACKET | |
| if (1) { | |
| /* Debugging is easier if we read packets directly with pcap | |
| (i.e. we can use Wireshark to verify packet contents) | |
| we don't want to do this all the time for 2 reasons: | |
| 1) sending through pcap involves kernel transitions and | |
| 2) if the current system reflects sent packets, the | |
| recieving side will receive and process 2 copies of | |
| any packets sent this way. */ | |
| ETH_PACK pkt; | |
| memset(&pkt, 0, sizeof(pkt)); | |
| memcpy(pkt.msg, ((u_char *)IP)-14, header.len); | |
| pkt.len = header.len; | |
| _eth_write(dev, &pkt, NULL); | |
| } | |
| #else | |
| _eth_callback((u_char *)dev, &header, ((u_char *)IP)-14); | |
| #endif | |
| payload_len -= (ntohs(IP->total_len) - IP_HLEN(IP)); | |
| frag_offset += (ntohs(IP->total_len) - IP_HLEN(IP))>>3; | |
| if (payload_len > 0) { | |
| /* Move the MAC and IP headers down to just prior to the next payload segment */ | |
| memcpy(((u_char *)IP) + ntohs(IP->total_len) - (14 + IP_HLEN(IP)), ((u_char *)IP) - 14, 14 + IP_HLEN(IP)); | |
| IP = (struct IPHeader *)(((u_char *)IP) + ntohs(IP->total_len) - IP_HLEN(IP)); | |
| } | |
| } | |
| break; | |
| case IPPROTO_TCP: | |
| ++dev->jumbo_fragmented; | |
| eth_packet_trace_ex (dev, ((u_char *)IP)-14, len, "Fragmenting Jumbo TCP segment", 1, dev->dbit); | |
| TCP = (struct TCPHeader *)(((char *)IP)+IP_HLEN(IP)); | |
| orig_tcp_flags = ntohs(TCP->data_offset_and_flags); | |
| /* When we're performing LSO (Large Send Offload), we're given a | |
| 'template' header which may not include a value being populated | |
| in the IP header length (which is only 16 bits). | |
| We process as payload everything which isn't known header data. */ | |
| payload_len = len - (14 + IP_HLEN(IP) + TCP_DATA_OFFSET(TCP)); | |
| mtu_payload = ETH_MIN_JUMBO_FRAME - (14 + IP_HLEN(IP) + TCP_DATA_OFFSET(TCP)); | |
| while (payload_len > 0) { | |
| if (payload_len > mtu_payload) { | |
| TCP->data_offset_and_flags = htons(orig_tcp_flags&~(TCP_PSH_FLAG|TCP_FIN_FLAG|TCP_RST_FLAG)); | |
| IP->total_len = htons(mtu_payload + IP_HLEN(IP) + TCP_DATA_OFFSET(TCP)); | |
| } | |
| else { | |
| TCP->data_offset_and_flags = htons(orig_tcp_flags); | |
| IP->total_len = htons(payload_len + IP_HLEN(IP) + TCP_DATA_OFFSET(TCP)); | |
| } | |
| IP->checksum = 0; | |
| IP->checksum = ip_checksum((uint16 *)IP, IP_HLEN(IP)); | |
| TCP->checksum = 0; | |
| TCP->checksum = pseudo_checksum(ntohs(IP->total_len)-IP_HLEN(IP), IPPROTO_TCP, (uint16 *)(&IP->source_ip), (uint16 *)(&IP->dest_ip), (uint8 *)TCP); | |
| header.caplen = header.len = 14 + ntohs(IP->total_len); | |
| eth_packet_trace_ex (dev, ((u_char *)IP)-14, header.len, "reading TCP segment", 1, dev->dbit); | |
| #if ETH_MIN_JUMBO_FRAME < ETH_MAX_PACKET | |
| if (1) { | |
| /* Debugging is easier if we read packets directly with pcap | |
| (i.e. we can use Wireshark to verify packet contents) | |
| we don't want to do this all the time for 2 reasons: | |
| 1) sending through pcap involves kernel transitions and | |
| 2) if the current system reflects sent packets, the | |
| recieving side will receive and process 2 copies of | |
| any packets sent this way. */ | |
| ETH_PACK pkt; | |
| memset(&pkt, 0, sizeof(pkt)); | |
| memcpy(pkt.msg, ((u_char *)IP)-14, header.len); | |
| pkt.len = header.len; | |
| _eth_write(dev, &pkt, NULL); | |
| } | |
| #else | |
| _eth_callback((u_char *)dev, &header, ((u_char *)IP)-14); | |
| #endif | |
| payload_len -= (ntohs(IP->total_len) - (IP_HLEN(IP) + TCP_DATA_OFFSET(TCP))); | |
| if (payload_len > 0) { | |
| /* Move the MAC, IP and TCP headers down to just prior to the next payload segment */ | |
| memcpy(((u_char *)IP) + ntohs(IP->total_len) - (14 + IP_HLEN(IP) + TCP_DATA_OFFSET(TCP)), ((u_char *)IP) - 14, 14 + IP_HLEN(IP) + TCP_DATA_OFFSET(TCP)); | |
| IP = (struct IPHeader *)(((u_char *)IP) + ntohs(IP->total_len) - (IP_HLEN(IP) + TCP_DATA_OFFSET(TCP))); | |
| TCP = (struct TCPHeader *)(((char *)IP)+IP_HLEN(IP)); | |
| TCP->sequence_number = htonl(mtu_payload + ntohl(TCP->sequence_number)); | |
| } | |
| } | |
| break; | |
| } | |
| } | |
| static void | |
| _eth_fix_ip_xsum_offload(ETH_DEV* dev, u_char* msg, int len) | |
| { | |
| unsigned short* proto = (unsigned short*) &msg[12]; | |
| struct IPHeader *IP; | |
| struct TCPHeader *TCP; | |
| struct UDPHeader *UDP; | |
| struct ICMPHeader *ICMP; | |
| uint16 orig_checksum; | |
| /* Only need to process locally originated packets */ | |
| if ((!dev->have_host_nic_phy_addr) || (memcmp(msg+6, dev->host_nic_phy_hw_addr, 6))) | |
| return; | |
| /* Only interested in IP frames */ | |
| if (ntohs(*proto) != 0x0800) | |
| return; | |
| IP = (struct IPHeader *)&msg[14]; | |
| if (IP_VERSION(IP) != 4) | |
| return; /* Only interested in IPv4 frames */ | |
| if ((IP_HLEN(IP) > len) || (ntohs(IP->total_len) > len)) | |
| return; /* Bogus header length */ | |
| orig_checksum = IP->checksum; | |
| IP->checksum = 0; | |
| IP->checksum = ip_checksum((uint16 *)IP, IP_HLEN(IP)); | |
| if (orig_checksum != IP->checksum) | |
| eth_packet_trace (dev, msg, len, "reading IP header Checksum Fixed"); | |
| if (IP_FRAG_OFFSET(IP) || IP_FRAG_MF(IP)) | |
| return; /* Insufficient data to compute payload checksum */ | |
| switch (IP->proto) { | |
| case IPPROTO_UDP: | |
| UDP = (struct UDPHeader *)(((char *)IP)+IP_HLEN(IP)); | |
| if (ntohs(UDP->length) > (len-IP_HLEN(IP))) | |
| return; /* packet contained length exceeds packet size */ | |
| if (UDP->checksum == 0) | |
| return; /* UDP Checksums are disabled */ | |
| orig_checksum = UDP->checksum; | |
| UDP->checksum = 0; | |
| UDP->checksum = pseudo_checksum(ntohs(UDP->length), IPPROTO_UDP, (uint16 *)(&IP->source_ip), (uint16 *)(&IP->dest_ip), (uint8 *)UDP); | |
| if (orig_checksum != UDP->checksum) | |
| eth_packet_trace (dev, msg, len, "reading UDP header Checksum Fixed"); | |
| break; | |
| case IPPROTO_TCP: | |
| TCP = (struct TCPHeader *)(((char *)IP)+IP_HLEN(IP)); | |
| orig_checksum = TCP->checksum; | |
| TCP->checksum = 0; | |
| TCP->checksum = pseudo_checksum(ntohs(IP->total_len)-IP_HLEN(IP), IPPROTO_TCP, (uint16 *)(&IP->source_ip), (uint16 *)(&IP->dest_ip), (uint8 *)TCP); | |
| if (orig_checksum != TCP->checksum) | |
| eth_packet_trace (dev, msg, len, "reading TCP header Checksum Fixed"); | |
| break; | |
| case IPPROTO_ICMP: | |
| ICMP = (struct ICMPHeader *)(((char *)IP)+IP_HLEN(IP)); | |
| orig_checksum = ICMP->checksum; | |
| ICMP->checksum = 0; | |
| ICMP->checksum = ip_checksum((uint16 *)ICMP, ntohs(IP->total_len)-IP_HLEN(IP)); | |
| if (orig_checksum != ICMP->checksum) | |
| eth_packet_trace (dev, msg, len, "reading ICMP header Checksum Fixed"); | |
| break; | |
| } | |
| } | |
| static void | |
| _eth_callback(u_char* info, const struct pcap_pkthdr* header, const u_char* data) | |
| { | |
| ETH_DEV* dev = (ETH_DEV*) info; | |
| int to_me; | |
| int from_me = 0; | |
| int i; | |
| int bpf_used; | |
| if ((dev->have_host_nic_phy_addr) && | |
| (LOOPBACK_PHYSICAL_RESPONSE(dev->host_nic_phy_hw_addr, dev->physical_addr, data))) { | |
| u_char *datacopy = malloc(header->len); | |
| memcpy(datacopy, data, header->len); | |
| memcpy(datacopy, dev->physical_addr, sizeof(ETH_MAC)); | |
| memcpy(datacopy+18, dev->physical_addr, sizeof(ETH_MAC)); | |
| _eth_callback(info, header, datacopy); | |
| free(datacopy); | |
| return; | |
| } | |
| switch (dev->eth_api) { | |
| case ETH_API_PCAP: | |
| #ifdef USE_BPF | |
| bpf_used = 1; | |
| to_me = 1; | |
| /* AUTODIN II hash mode? */ | |
| if ((dev->hash_filter) && (data[0] & 0x01) && (!dev->promiscuous) && (!dev->all_multicast)) | |
| to_me = _eth_hash_lookup(dev->hash, data); | |
| break; | |
| #endif /* USE_BPF */ | |
| case ETH_API_TAP: | |
| case ETH_API_VDE: | |
| bpf_used = 0; | |
| to_me = 0; | |
| eth_packet_trace (dev, data, header->len, "received"); | |
| for (i = 0; i < dev->addr_count; i++) { | |
| if (memcmp(data, dev->filter_address[i], 6) == 0) to_me = 1; | |
| if (memcmp(&data[6], dev->filter_address[i], 6) == 0) from_me = 1; | |
| } | |
| /* all multicast mode? */ | |
| if (dev->all_multicast && (data[0] & 0x01)) to_me = 1; | |
| /* promiscuous mode? */ | |
| if (dev->promiscuous) to_me = 1; | |
| /* AUTODIN II hash mode? */ | |
| if ((dev->hash_filter) && (!to_me) && (data[0] & 0x01)) | |
| to_me = _eth_hash_lookup(dev->hash, data); | |
| break; | |
| default: | |
| bpf_used = to_me = 0; /* Should NEVER happen */ | |
| abort(); | |
| break; | |
| } | |
| /* detect reception of loopback packet to our physical address */ | |
| if ((LOOPBACK_SELF_FRAME(dev->physical_addr, data)) || | |
| (dev->have_host_nic_phy_addr && | |
| LOOPBACK_PHYSICAL_REFLECTION(dev->host_nic_phy_hw_addr, data))) { | |
| #ifdef USE_READER_THREAD | |
| pthread_mutex_lock (&dev->self_lock); | |
| #endif | |
| dev->loopback_self_rcvd_total++; | |
| /* lower reflection count - if already zero, pass it on */ | |
| if (dev->loopback_self_sent > 0) { | |
| eth_packet_trace (dev, data, header->len, "ignored"); | |
| dev->loopback_self_sent--; | |
| to_me = 0; | |
| } | |
| else | |
| if (!bpf_used) | |
| from_me = 0; | |
| #ifdef USE_READER_THREAD | |
| pthread_mutex_unlock (&dev->self_lock); | |
| #endif | |
| } | |
| if (bpf_used ? to_me : (to_me && !from_me)) { | |
| if (header->len > ETH_MIN_JUMBO_FRAME) { | |
| if (header->len <= header->caplen) /* Whole Frame captured? */ | |
| _eth_fix_ip_jumbo_offload(dev, data, header->len); | |
| else | |
| ++dev->jumbo_truncated; | |
| return; | |
| } | |
| #if defined (USE_READER_THREAD) | |
| if (1) { | |
| int crc_len = 0; | |
| uint8 crc_data[4]; | |
| uint32 len = header->len; | |
| u_char* moved_data = NULL; | |
| if (header->len < ETH_MIN_PACKET) { /* Pad runt packets before CRC append */ | |
| moved_data = malloc(ETH_MIN_PACKET); | |
| memcpy(moved_data, data, len); | |
| memset(moved_data + len, 0, ETH_MIN_PACKET-len); | |
| len = ETH_MIN_PACKET; | |
| data = moved_data; | |
| } | |
| /* If necessary, fix IP header checksums for packets originated locally */ | |
| /* but were presumed to be traversing a NIC which was going to handle that task */ | |
| /* This must be done before any needed CRC calculation */ | |
| _eth_fix_ip_xsum_offload(dev, (u_char*)data, len); | |
| if (dev->need_crc) | |
| crc_len = eth_get_packet_crc32_data(data, len, crc_data); | |
| eth_packet_trace (dev, data, len, "rcvqd"); | |
| pthread_mutex_lock (&dev->lock); | |
| ethq_insert_data(&dev->read_queue, 2, data, 0, len, crc_len, crc_data, 0); | |
| pthread_mutex_unlock (&dev->lock); | |
| free(moved_data); | |
| } | |
| #else /* !USE_READER_THREAD */ | |
| /* set data in passed read packet */ | |
| dev->read_packet->len = header->len; | |
| memcpy(dev->read_packet->msg, data, header->len); | |
| /* Handle runt case and pad with zeros. */ | |
| /* The real NIC won't hand us runts from the wire, BUT we may be getting */ | |
| /* some packets looped back before they actually traverse the wire */ | |
| /* (by an internal bridge device for instance) */ | |
| if (header->len < ETH_MIN_PACKET) { | |
| memset(&dev->read_packet->msg[header->len], 0, ETH_MIN_PACKET-header->len); | |
| dev->read_packet->len = ETH_MIN_PACKET; | |
| } | |
| /* If necessary, fix IP header checksums for packets originated by the local host */ | |
| /* but were presumed to be traversing a NIC which was going to handle that task */ | |
| /* This must be done before any needed CRC calculation */ | |
| _eth_fix_ip_xsum_offload(dev, dev->read_packet->msg, dev->read_packet->len); | |
| if (dev->need_crc) | |
| dev->read_packet->crc_len = eth_add_packet_crc32(dev->read_packet->msg, dev->read_packet->len); | |
| else | |
| dev->read_packet->crc_len = 0; | |
| eth_packet_trace (dev, dev->read_packet->msg, dev->read_packet->len, "reading"); | |
| /* call optional read callback function */ | |
| if (dev->read_callback) | |
| (dev->read_callback)(0); | |
| #endif | |
| } | |
| } | |
| int eth_read(ETH_DEV* dev, ETH_PACK* packet, ETH_PCALLBACK routine) | |
| { | |
| int status; | |
| /* make sure device exists */ | |
| if (!dev) return 0; | |
| /* make sure packet exists */ | |
| if (!packet) return 0; | |
| packet->len = 0; | |
| #if !defined (USE_READER_THREAD) | |
| /* set read packet */ | |
| dev->read_packet = packet; | |
| /* set optional callback routine */ | |
| dev->read_callback = routine; | |
| /* dispatch read request to either receive a filtered packet or timeout */ | |
| do { | |
| switch (dev->eth_api) { | |
| case ETH_API_PCAP: | |
| status = pcap_dispatch((pcap_t*)dev->handle, 1, &_eth_callback, (u_char*)dev); | |
| break; | |
| #ifdef USE_TAP_NETWORK | |
| case ETH_API_TAP: | |
| if (1) { | |
| struct pcap_pkthdr header; | |
| int len; | |
| u_char buf[ETH_MAX_JUMBO_FRAME]; | |
| memset(&header, 0, sizeof(header)); | |
| len = read(dev->fd_handle, buf, sizeof(buf)); | |
| if (len > 0) { | |
| status = 1; | |
| header.caplen = header.len = len; | |
| _eth_callback((u_char *)dev, &header, buf); | |
| } | |
| else | |
| status = 0; | |
| } | |
| break; | |
| #endif /* USE_TAP_NETWORK */ | |
| #ifdef USE_VDE_NETWORK | |
| case ETH_API_VDE: | |
| if (1) { | |
| struct pcap_pkthdr header; | |
| int len; | |
| u_char buf[ETH_MAX_JUMBO_FRAME]; | |
| memset(&header, 0, sizeof(header)); | |
| len = vde_recv((VDECONN*)dev->handle, buf, sizeof(buf), 0); | |
| if (len > 0) { | |
| status = 1; | |
| header.caplen = header.len = len; | |
| _eth_callback((u_char *)dev, &header, buf); | |
| } | |
| else | |
| status = 0; | |
| } | |
| break; | |
| #endif /* USE_VDE_NETWORK */ | |
| } | |
| } while ((status) && (0 == packet->len)); | |
| #else /* USE_READER_THREAD */ | |
| status = 0; | |
| pthread_mutex_lock (&dev->lock); | |
| if (dev->read_queue.count > 0) { | |
| ETH_ITEM* item = &dev->read_queue.item[dev->read_queue.head]; | |
| packet->len = item->packet.len; | |
| packet->crc_len = item->packet.crc_len; | |
| memcpy(packet->msg, item->packet.msg, ((packet->len > packet->crc_len) ? packet->len : packet->crc_len)); | |
| status = 1; | |
| ethq_remove(&dev->read_queue); | |
| } | |
| pthread_mutex_unlock (&dev->lock); | |
| if ((status) && (routine)) | |
| routine(0); | |
| #endif | |
| return status; | |
| } | |
| t_stat eth_filter(ETH_DEV* dev, int addr_count, ETH_MAC* const addresses, | |
| ETH_BOOL all_multicast, ETH_BOOL promiscuous) | |
| { | |
| return eth_filter_hash(dev, addr_count, addresses, | |
| all_multicast, promiscuous, | |
| NULL); | |
| } | |
| t_stat eth_filter_hash(ETH_DEV* dev, int addr_count, ETH_MAC* const addresses, | |
| ETH_BOOL all_multicast, ETH_BOOL promiscuous, | |
| ETH_MULTIHASH* const hash) | |
| { | |
| int i; | |
| bpf_u_int32 bpf_subnet, bpf_netmask; | |
| char buf[114+66*ETH_FILTER_MAX]; | |
| char errbuf[PCAP_ERRBUF_SIZE]; | |
| char mac[20]; | |
| char* buf2; | |
| t_stat status; | |
| #ifdef USE_BPF | |
| struct bpf_program bpf; | |
| char* msg; | |
| #endif | |
| /* make sure device exists */ | |
| if (!dev) return SCPE_UNATT; | |
| /* filter count OK? */ | |
| if ((addr_count < 0) || (addr_count > ETH_FILTER_MAX)) | |
| return SCPE_ARG; | |
| else | |
| if (!addresses) return SCPE_ARG; | |
| /* test reflections. This is done early in this routine since eth_reflect */ | |
| /* calls eth_filter recursively and thus changes the state of the device. */ | |
| if (dev->reflections == -1) | |
| status = eth_reflect(dev); | |
| /* set new filter addresses */ | |
| for (i = 0; i < addr_count; i++) | |
| memcpy(dev->filter_address[i], addresses[i], sizeof(ETH_MAC)); | |
| dev->addr_count = addr_count; | |
| /* store other flags */ | |
| dev->all_multicast = all_multicast; | |
| dev->promiscuous = promiscuous; | |
| /* store multicast hash data */ | |
| dev->hash_filter = (hash != NULL); | |
| if (hash) { | |
| memcpy(dev->hash, hash, sizeof(*hash)); | |
| sim_debug(dev->dbit, dev->dptr, "Multicast Hash: %02X-%02X-%02X-%02X-%02X-%02X-%02X-%02X\n", | |
| dev->hash[0], dev->hash[1], dev->hash[2], dev->hash[3], | |
| dev->hash[4], dev->hash[5], dev->hash[6], dev->hash[7]); | |
| } | |
| /* print out filter information if debugging */ | |
| if (dev->dptr->dctrl & dev->dbit) { | |
| sim_debug(dev->dbit, dev->dptr, "Filter Set\n"); | |
| for (i = 0; i < addr_count; i++) { | |
| char mac[20]; | |
| eth_mac_fmt(&dev->filter_address[i], mac); | |
| sim_debug(dev->dbit, dev->dptr, " Addr[%d]: %s\n", i, mac); | |
| } | |
| if (dev->all_multicast) { | |
| sim_debug(dev->dbit, dev->dptr, "All Multicast\n"); | |
| } | |
| if (dev->promiscuous) { | |
| sim_debug(dev->dbit, dev->dptr, "Promiscuous\n"); | |
| } | |
| } | |
| /* setup BPF filters and other fields to minimize packet delivery */ | |
| strcpy(buf, ""); | |
| /* construct destination filters - since the real ethernet interface was set | |
| into promiscuous mode by eth_open(), we need to filter out the packets that | |
| our simulated interface doesn't want. */ | |
| if (!dev->promiscuous) { | |
| for (i = 0; i < addr_count; i++) { | |
| eth_mac_fmt(&dev->filter_address[i], mac); | |
| if (!strstr(buf, mac)) /* eliminate duplicates */ | |
| sprintf(&buf[strlen(buf)], "%s(ether dst %s)", (*buf) ? " or " : "((", mac); | |
| } | |
| if (dev->all_multicast || dev->hash_filter) | |
| sprintf(&buf[strlen(buf)], "%s(ether multicast)", (*buf) ? " or " : "(("); | |
| if (strlen(buf) > 0) | |
| sprintf(&buf[strlen(buf)], ")"); | |
| } | |
| /* construct source filters - this prevents packets from being reflected back | |
| by systems where WinPcap and libpcap cause packet reflections. Note that | |
| some systems do not reflect packets at all. This *assumes* that the | |
| simulated NIC will not send out packets with multicast source fields. */ | |
| if ((addr_count > 0) && (dev->reflections > 0)) { | |
| if (strlen(buf) > 0) | |
| sprintf(&buf[strlen(buf)], " and "); | |
| sprintf (&buf[strlen(buf)], "not ("); | |
| buf2 = &buf[strlen(buf)]; | |
| for (i = 0; i < addr_count; i++) { | |
| if (dev->filter_address[i][0] & 0x01) continue; /* skip multicast addresses */ | |
| eth_mac_fmt(&dev->filter_address[i], mac); | |
| if (!strstr(buf2, mac)) /* eliminate duplicates */ | |
| sprintf(&buf2[strlen(buf2)], "%s(ether src %s)", (*buf2) ? " or " : "", mac); | |
| } | |
| sprintf (&buf[strlen(buf)], ")"); | |
| } | |
| if (strlen(buf) > 0) | |
| sprintf(&buf[strlen(buf)], ")"); | |
| /* When changing the Physical Address on a LAN interface, VMS sends out a | |
| loopback packet with the source and destination addresses set to the same | |
| value as the Physical Address which is being setup. This packet is | |
| designed to find and help diagnose MAC address conflicts (which also | |
| include DECnet address conflicts). Normally, this packet would not be | |
| seen by the sender, only by the other machine that has the same Physical | |
| Address (or possibly DECnet address). If the ethernet subsystem is | |
| reflecting packets, the network startup will fail to start if it sees the | |
| reflected packet, since it thinks another system is using this Physical | |
| Address (or DECnet address). We have to let these packets through, so | |
| that if another machine has the same Physical Address (or DECnet address) | |
| that we can detect it. Both eth_write() and _eth_callback() help by | |
| checking the reflection count - eth_write() adds the reflection count to | |
| dev->loopback_self_sent, and _eth_callback() check the value - if the | |
| dev->loopback_self_sent count is zero, then the packet has come from | |
| another machine with the same address, and needs to be passed on to the | |
| simulated machine. */ | |
| memset(dev->physical_addr, 0, sizeof(ETH_MAC)); | |
| dev->loopback_self_sent = 0; | |
| /* check for physical address in filters */ | |
| if ((addr_count) && (dev->reflections > 0)) { | |
| for (i = 0; i < addr_count; i++) { | |
| if (dev->filter_address[i][0]&1) | |
| continue; /* skip all multicast addresses */ | |
| eth_mac_fmt(&dev->filter_address[i], mac); | |
| if (strcmp(mac, "00:00:00:00:00:00") != 0) { | |
| memcpy(dev->physical_addr, &dev->filter_address[i], sizeof(ETH_MAC)); | |
| /* let packets through where dst and src are the same as our physical address */ | |
| sprintf (&buf[strlen(buf)], " or ((ether dst %s) and (ether src %s))", mac, mac); | |
| if (dev->have_host_nic_phy_addr) { | |
| eth_mac_fmt(&dev->host_nic_phy_hw_addr, mac); | |
| sprintf(&buf[strlen(buf)], "or ((ether dst %s) and (ether proto 0x9000))", mac); | |
| } | |
| break; | |
| } | |
| } | |
| } | |
| if ((0 == strlen(buf)) && (!dev->promiscuous)) /* Empty filter means match nothing */ | |
| strcpy(buf, "ether host fe:ff:ff:ff:ff:ff"); /* this should be a good match nothing filter */ | |
| sim_debug(dev->dbit, dev->dptr, "BPF string is: |%s|\n", buf); | |
| /* get netmask, which is a required argument for compiling. The value, | |
| in our case isn't actually interesting since the filters we generate | |
| aren't referencing IP fields, networks or values */ | |
| if ((dev->eth_api == ETH_API_PCAP) && (pcap_lookupnet(dev->name, &bpf_subnet, &bpf_netmask, errbuf)<0)) | |
| bpf_netmask = 0; | |
| #ifdef USE_BPF | |
| if (dev->eth_api == ETH_API_PCAP) { | |
| /* compile filter string */ | |
| if ((status = pcap_compile(dev->handle, &bpf, buf, 1, bpf_netmask)) < 0) { | |
| sprintf(errbuf, "%s", pcap_geterr(dev->handle)); | |
| msg = "Eth: pcap_compile error: %s\r\n"; | |
| printf(msg, errbuf); | |
| if (sim_log) fprintf (sim_log, msg, errbuf); | |
| sim_debug(dev->dbit, dev->dptr, "Eth: pcap_compile error: %s\n", errbuf); | |
| /* show erroneous BPF string */ | |
| msg = "Eth: BPF string is: |%s|\r\n"; | |
| printf (msg, buf); | |
| if (sim_log) fprintf (sim_log, msg, buf); | |
| } | |
| else { | |
| /* apply compiled filter string */ | |
| if ((status = pcap_setfilter(dev->handle, &bpf)) < 0) { | |
| sprintf(errbuf, "%s", pcap_geterr(dev->handle)); | |
| msg = "Eth: pcap_setfilter error: %s\r\n"; | |
| printf(msg, errbuf); | |
| if (sim_log) fprintf (sim_log, msg, errbuf); | |
| sim_debug(dev->dbit, dev->dptr, "Eth: pcap_setfilter error: %s\n", errbuf); | |
| } | |
| else { | |
| #ifdef USE_SETNONBLOCK | |
| /* set file non-blocking */ | |
| status = pcap_setnonblock (dev->handle, 1, errbuf); | |
| #endif /* USE_SETNONBLOCK */ | |
| } | |
| pcap_freecode(&bpf); | |
| } | |
| #ifdef USE_READER_THREAD | |
| pthread_mutex_lock (&dev->lock); | |
| ethq_clear (&dev->read_queue); /* Empty FIFO Queue when filter list changes */ | |
| pthread_mutex_unlock (&dev->lock); | |
| #endif | |
| } | |
| #endif /* USE_BPF */ | |
| return SCPE_OK; | |
| } | |
| /* | |
| The libpcap provided API pcap_findalldevs() on most platforms, will | |
| leverage the getifaddrs() API if it is available in preference to | |
| alternate platform specific methods of determining the interface list. | |
| A limitation of getifaddrs() is that it returns only interfaces which | |
| have associated addresses. This may not include all of the interesting | |
| interfaces that we are interested in since a host may have dedicated | |
| interfaces for a simulator, which is otherwise unused by the host. | |
| One could hand craft the the build of libpcap to specifically use | |
| alternate methods to implement pcap_findalldevs(). However, this can | |
| get tricky, and would then result in a sort of deviant libpcap. | |
| This routine exists to allow platform specific code to validate and/or | |
| extend the set of available interfaces to include any that are not | |
| returned by pcap_findalldevs. | |
| */ | |
| int eth_host_devices(int used, int max, ETH_LIST* list) | |
| { | |
| pcap_t* conn; | |
| int i, j, datalink; | |
| char errbuf[PCAP_ERRBUF_SIZE]; | |
| for (i=0; i<used; ++i) { | |
| /* Cull any non-ethernet interface types */ | |
| conn = pcap_open_live(list[i].name, ETH_MAX_PACKET, ETH_PROMISC, PCAP_READ_TIMEOUT, errbuf); | |
| if (NULL != conn) datalink = pcap_datalink(conn), pcap_close(conn); | |
| if ((NULL == conn) || (datalink != DLT_EN10MB)) { | |
| for (j=i; j<used-1; ++j) | |
| list[j] = list[j+1]; | |
| --used; | |
| --i; | |
| } | |
| } /* for */ | |
| #if defined(_WIN32) | |
| /* replace device description with user-defined adapter name (if defined) */ | |
| for (i=0; i<used; i++) { | |
| char regkey[2048]; | |
| unsigned char regval[2048]; | |
| LONG status; | |
| DWORD reglen, regtype; | |
| HKEY reghnd; | |
| /* These registry keys don't seem to exist for all devices, so we simply ignore errors. */ | |
| /* Windows XP x64 registry uses wide characters by default, | |
| so we force use of narrow characters by using the 'A'(ANSI) version of RegOpenKeyEx. | |
| This could cause some problems later, if this code is internationalized. Ideally, | |
| the pcap lookup will return wide characters, and we should use them to build a wide | |
| registry key, rather than hardcoding the string as we do here. */ | |
| if (list[i].name[strlen( "\\Device\\NPF_" )] == '{') { | |
| sprintf( regkey, "SYSTEM\\CurrentControlSet\\Control\\Network\\" | |
| "{4D36E972-E325-11CE-BFC1-08002BE10318}\\%s\\Connection", list[i].name+ | |
| strlen( "\\Device\\NPF_" ) ); | |
| if ((status = RegOpenKeyExA (HKEY_LOCAL_MACHINE, regkey, 0, KEY_QUERY_VALUE, ®hnd)) != ERROR_SUCCESS) | |
| continue; | |
| reglen = sizeof(regval); | |
| /* look for user-defined adapter name, bail if not found */ | |
| /* same comment about Windows XP x64 (above) using RegQueryValueEx */ | |
| if ((status = RegQueryValueExA (reghnd, "Name", NULL, ®type, regval, ®len)) != ERROR_SUCCESS) { | |
| RegCloseKey (reghnd); | |
| continue; | |
| } | |
| /* make sure value is the right type, bail if not acceptable */ | |
| if ((regtype != REG_SZ) || (reglen > sizeof(regval))) { | |
| RegCloseKey (reghnd); | |
| continue; | |
| } | |
| /* registry value seems OK, finish up and replace description */ | |
| RegCloseKey (reghnd ); | |
| sprintf (list[i].desc, "%s", regval); | |
| } | |
| } /* for */ | |
| #endif | |
| #ifdef USE_TAP_NETWORK | |
| if (used < max) { | |
| #if defined(__OpenBSD__) | |
| sprintf(list[used].name, "%s", "tap:tunN"); | |
| #else | |
| sprintf(list[used].name, "%s", "tap:tapN"); | |
| #endif | |
| sprintf(list[used].desc, "%s", "Integrated Tun/Tap support"); | |
| ++used; | |
| } | |
| #endif | |
| #ifdef USE_VDE_NETWORK | |
| if (used < max) { | |
| sprintf(list[used].name, "%s", "vde:device"); | |
| sprintf(list[used].desc, "%s", "Integrated VDE support"); | |
| ++used; | |
| } | |
| #endif | |
| return used; | |
| } | |
| int eth_devices(int max, ETH_LIST* list) | |
| { | |
| int i = 0; | |
| #ifndef DONT_USE_PCAP_FINDALLDEVS | |
| pcap_if_t* alldevs; | |
| pcap_if_t* dev; | |
| char errbuf[PCAP_ERRBUF_SIZE]; | |
| memset(list, 0, max*sizeof(*list)); | |
| errbuf[0] = '\0'; | |
| /* retrieve the device list */ | |
| if (pcap_findalldevs(&alldevs, errbuf) == -1) { | |
| char* msg = "Eth: error in pcap_findalldevs: %s\r\n"; | |
| printf (msg, errbuf); | |
| if (sim_log) fprintf (sim_log, msg, errbuf); | |
| } | |
| else { | |
| /* copy device list into the passed structure */ | |
| for (i=0, dev=alldevs; dev && (i < max); dev=dev->next, ++i) { | |
| if ((dev->flags & PCAP_IF_LOOPBACK) || (!strcmp("any", dev->name))) continue; | |
| strncpy(list[i].name, dev->name, sizeof(list[i].name)-1); | |
| if (dev->description) | |
| strncpy(list[i].desc, dev->description, sizeof(list[i].desc)-1); | |
| else | |
| strncpy(list[i].desc, "No description available", sizeof(list[i].desc)-1); | |
| } | |
| /* free device list */ | |
| pcap_freealldevs(alldevs); | |
| } | |
| #endif | |
| /* Add any host specific devices and/or validate those already found */ | |
| i = eth_host_devices(i, max, list); | |
| /* If no devices were found and an error message was left in the buffer, display it */ | |
| if ((i == 0) && (errbuf[0])) { | |
| char* msg = "Eth: pcap_findalldevs warning: %s\r\n"; | |
| printf (msg, errbuf); | |
| if (sim_log) fprintf (sim_log, msg, errbuf); | |
| } | |
| /* return device count */ | |
| return i; | |
| } | |
| void eth_show_dev (FILE *st, ETH_DEV* dev) | |
| { | |
| fprintf(st, "Ethernet Device:\n"); | |
| if (!dev) { | |
| fprintf(st, "-- Not Attached\n"); | |
| return; | |
| } | |
| fprintf(st, " Name: %s\n", dev->name); | |
| fprintf(st, " Reflections: %d\n", dev->reflections); | |
| fprintf(st, " Self Loopbacks Sent: %d\n", dev->loopback_self_sent_total); | |
| fprintf(st, " Self Loopbacks Rcvd: %d\n", dev->loopback_self_rcvd_total); | |
| if (dev->have_host_nic_phy_addr) { | |
| char hw_mac[20]; | |
| eth_mac_fmt(&dev->host_nic_phy_hw_addr, hw_mac); | |
| fprintf(st, " Host NIC Address: %s\n", hw_mac); | |
| } | |
| if (dev->jumbo_dropped) | |
| fprintf(st, " Jumbo Dropped: %d\n", dev->jumbo_dropped); | |
| if (dev->jumbo_fragmented) | |
| fprintf(st, " Jumbo Fragmented: %d\n", dev->jumbo_fragmented); | |
| if (dev->jumbo_truncated) | |
| fprintf(st, " Jumbo Truncated: %d\n", dev->jumbo_truncated); | |
| #if defined(USE_READER_THREAD) | |
| fprintf(st, " Asynch Interrupts: %s\n", dev->asynch_io?"Enabled":"Disabled"); | |
| if (dev->asynch_io) | |
| fprintf(st, " Interrupt Latency: %d uSec\n", dev->asynch_io_latency); | |
| fprintf(st, " Read Queue: Count: %d\n", dev->read_queue.count); | |
| fprintf(st, " Read Queue: High: %d\n", dev->read_queue.high); | |
| fprintf(st, " Read Queue: Loss: %d\n", dev->read_queue.loss); | |
| fprintf(st, " Peak Write Queue Size: %d\n", dev->write_queue_peak); | |
| #endif | |
| } | |
| #endif /* USE_NETWORK */ |