1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240
//! Rust API wrapping the `ibverbs` RDMA library. //! //! `libibverbs` is a library that allows userspace processes to use RDMA "verbs" to perform //! high-throughput, low-latency network operations for both Infiniband (according to the //! Infiniband specifications) and iWarp (iWARP verbs specifications). It handles the control path //! of creating, modifying, querying and destroying resources such as Protection Domains, //! Completion Queues, Queue-Pairs, Shared Receive Queues, Address Handles, and Memory Regions. It //! also handles sending and receiving data posted to QPs and SRQs, and getting completions from //! CQs using polling and completions events. //! //! A good place to start is to look at the programs in [`examples/`](examples/), and the upstream //! [C examples]. You can test RDMA programs on modern Linux kernels even without specialized RDMA //! hardware by using [SoftRoCE][soft]. //! //! # For the detail-oriented //! //! The control path is implemented through system calls to the `uverbs` kernel module, which //! further calls the low-level HW driver. The data path is implemented through calls made to //! low-level HW library which, in most cases, interacts directly with the HW provides kernel and //! network stack bypass (saving context/mode switches) along with zero copy and an asynchronous //! I/O model. //! //! iWARP ethernet NICs support RDMA over hardware-offloaded TCP/IP, while InfiniBand is a general //! high-throughput, low-latency networking technology. InfiniBand host channel adapters (HCAs) and //! iWARP NICs commonly support direct hardware access from userspace (kernel bypass), and //! `libibverbs` supports this when available. //! //! For more information on RDMA verbs, see the [InfiniBand Architecture Specification][infini] //! vol. 1, especially chapter 11, and the RDMA Consortium's [RDMA Protocol Verbs //! Specification][RFC5040]. See also the upstream [`libibverbs/verbs.h`] file for the original C //! definitions, as well as the manpages for the `ibv_*` methods. //! //! # Library dependency //! //! `libibverbs` is usually available as a free-standing [library package]. It [used to be][1] //! self-contained, but has recently been adopted into [`rdma-core`]. `cargo` will automatically //! build the necessary library files and place them in `vendor/rdma-core/build/lib`. If a //! system-wide installation is not available, those library files can be used instead by copying //! them to `/usr/lib`, or by adding that path to the dynamic linking search path. //! //! # Thread safety //! //! All interfaces are `Sync` and `Send` since the underlying ibverbs API [is thread safe][safe]. //! //! # Documentation //! //! Much of the documentation of this crate borrows heavily from the excellent posts over at //! [RDMAmojo]. If you are going to be working a lot with ibverbs, chances are you will want to //! head over there. In particular, [this overview post][1] may be a good place to start. //! //! [`rdma-core`]: https://github.com/linux-rdma/rdma-core //! [`libibverbs/verbs.h`]: https://github.com/linux-rdma/rdma-core/blob/master/libibverbs/verbs.h //! [library package]: https://launchpad.net/ubuntu/+source/libibverbs //! [C examples]: https://github.com/linux-rdma/rdma-core/tree/master/libibverbs/examples //! [1]: https://git.kernel.org/pub/scm/libs/infiniband/libibverbs.git/about/ //! [infini]: http://www.infinibandta.org/content/pages.php?pg=technology_public_specification //! [RFC5040]: https://tools.ietf.org/html/rfc5040 //! [safe]: http://www.rdmamojo.com/2013/07/26/libibverbs-thread-safe-level/ //! [soft]: https://github.com/SoftRoCE/rxe-dev/wiki/rxe-dev:-Home //! [RDMAmojo]: http://www.rdmamojo.com/ //! [1]: http://www.rdmamojo.com/2012/05/18/libibverbs/ #![deny(missing_docs)] // avoid warnings about RDMAmojo, iWARP, InfiniBand, etc. not being in backticks #![cfg_attr(feature = "cargo-clippy", allow(doc_markdown))] use std::ffi::CStr; use std::io; use std::marker::PhantomData; use std::mem; use std::os::raw::c_void; use std::ptr; const PORT_NUM: u8 = 1; /// Direct access to low-level libverbs FFI. #[allow(non_upper_case_globals)] #[allow(non_camel_case_types)] #[allow(non_snake_case)] #[allow(missing_docs)] pub mod ffi; pub use ffi::ibv_qp_type; pub use ffi::ibv_wc; pub use ffi::ibv_wc_opcode; pub use ffi::ibv_wc_status; /// Access flags for use with `QueuePair` and `MemoryRegion`. pub use ffi::ibv_access_flags; /// Because `std::slice::SliceIndex` is still unstable, we follow @alexcrichton's suggestion in /// https://github.com/rust-lang/rust/issues/35729 and implement it ourselves. mod sliceindex; /// Get list of available RDMA devices. /// /// # Errors /// /// - `EPERM`: Permission denied. /// - `ENOMEM`: Insufficient memory to complete the operation. /// - `ENOSYS`: No kernel support for RDMA. pub fn devices() -> io::Result<DeviceList> { let mut n = 0i32; let devices = unsafe { ffi::ibv_get_device_list(&mut n as *mut _) }; if devices.is_null() { return Err(io::Error::last_os_error()); } let devices = unsafe { use std::slice; slice::from_raw_parts_mut(devices, n as usize) }; Ok(DeviceList(devices)) } /// List of available RDMA devices. pub struct DeviceList(&'static mut [*mut ffi::ibv_device]); unsafe impl Sync for DeviceList {} unsafe impl Send for DeviceList {} impl Drop for DeviceList { fn drop(&mut self) { unsafe { ffi::ibv_free_device_list(self.0.as_mut_ptr()) }; } } impl DeviceList { /// Returns an iterator over all found devices. pub fn iter(&self) -> DeviceListIter { DeviceListIter { list: self, i: 0 } } /// Returns the number of devices. pub fn len(&self) -> usize { self.0.len() } /// Returns `true` if there are any devices. pub fn is_empty(&self) -> bool { self.0.is_empty() } /// Returns the device at the given `index`, or `None` if out of bounds. pub fn get(&self, index: usize) -> Option<Device> { self.0.get(index).map(|d| d.into()) } } impl<'a> IntoIterator for &'a DeviceList { type Item = <DeviceListIter<'a> as Iterator>::Item; type IntoIter = DeviceListIter<'a>; fn into_iter(self) -> Self::IntoIter { DeviceListIter { list: self, i: 0 } } } /// Iterator over a `DeviceList`. pub struct DeviceListIter<'iter> { list: &'iter DeviceList, i: usize, } impl<'iter> Iterator for DeviceListIter<'iter> { type Item = Device<'iter>; fn next(&mut self) -> Option<Self::Item> { let e = self.list.0.get(self.i); if e.is_some() { self.i += 1; } e.map(|e| e.into()) } } /// An RDMA device. pub struct Device<'devlist>(&'devlist *mut ffi::ibv_device); unsafe impl<'devlist> Sync for Device<'devlist> {} unsafe impl<'devlist> Send for Device<'devlist> {} impl<'d> From<&'d *mut ffi::ibv_device> for Device<'d> { fn from(d: &'d *mut ffi::ibv_device) -> Self { Device(d) } } impl<'devlist> Device<'devlist> { /// Opens an RMDA device and creates a context for further use. /// /// This context will later be used to query its resources or for creating resources. /// /// Unlike what the verb name suggests, it doesn't actually open the device. This device was /// opened by the kernel low-level driver and may be used by other user/kernel level code. This /// verb only opens a context to allow user level applications to use it. /// /// # Errors /// /// - `EINVAL`: `PORT_NUM` is invalid (from `ibv_query_port_attr`). /// - `ENOMEM`: Out of memory (from `ibv_query_port_attr`). /// - `EMFILE`: Too many files are opened by this process (from `ibv_query_gid`). /// - Other: the device is not in `ACTIVE` or `ARMED` state. pub fn open(&self) -> io::Result<Context> { Context::with_device(*self.0) } /// Returns a string of the name, which is associated with this RDMA device. /// /// This name is unique within a specific machine (the same name cannot be assigned to more /// than one device). However, this name isn't unique across an InfiniBand fabric (this name /// can be found in different machines). /// /// When there are more than one RDMA devices in a computer, changing the device location in /// the computer (i.e. in the PCI bus) may result a change in the names associated with the /// devices. In order to distinguish between the device, it is recommended using the device /// GUID, returned by `Device::guid`. /// /// The name is composed from: /// /// - a *prefix* which describes the RDMA device vendor and model /// - `cxgb3` - Chelsio Communications, T3 RDMA family /// - `cxgb4` - Chelsio Communications, T4 RDMA family /// - `ehca` - IBM, eHCA family /// - `ipathverbs` - QLogic /// - `mlx4` - Mellanox Technologies, ConnectX family /// - `mthca` - Mellanox Technologies, InfiniHost family /// - `nes` - Intel, Intel-NE family /// - an *index* that helps to differentiate between several devices from the same vendor and /// family in the same computer pub fn name(&self) -> Option<&'devlist CStr> { let name_ptr = unsafe { ffi::ibv_get_device_name(*self.0) }; if name_ptr.is_null() { None } else { Some(unsafe { CStr::from_ptr(name_ptr) }) } } /// Returns the Global Unique IDentifier (GUID) of this RDMA device. /// /// This GUID, that was assigned to this device by its vendor during the manufacturing, is /// unique and can be used as an identifier to an RDMA device. /// /// From the prefix of the RDMA device GUID, one can know who is the vendor of that device /// using the [IEEE OUI](http://standards.ieee.org/develop/regauth/oui/oui.txt). /// /// # Errors /// /// - `EMFILE`: Too many files are opened by this process. pub fn guid(&self) -> io::Result<u64> { let guid = unsafe { ffi::ibv_get_device_guid(*self.0) }; if guid == 0 { Err(io::Error::last_os_error()) } else { Ok(guid) } } } /// An RDMA context bound to a device. pub struct Context { ctx: *mut ffi::ibv_context, port_attr: ffi::ibv_port_attr, gid: ffi::ibv_gid, } unsafe impl Sync for Context {} unsafe impl Send for Context {} impl Context { /// Opens a context for the given device, and queries its port and gid. fn with_device(dev: *mut ffi::ibv_device) -> io::Result<Context> { assert!(!dev.is_null()); let ctx = unsafe { ffi::ibv_open_device(dev) }; if ctx.is_null() { return Err(io::Error::new( io::ErrorKind::Other, "failed to open device".to_string(), )); } // TODO: from http://www.rdmamojo.com/2012/07/21/ibv_query_port/ // // Most of the port attributes, returned by ibv_query_port(), aren't constant and may be // changed, mainly by the SM (in InfiniBand), or by the Hardware. It is highly // recommended avoiding saving the result of this query, or to flush them when a new SM // (re)configures the subnet. // let mut port_attr = ffi::ibv_port_attr::default(); let errno = unsafe { ffi::ibv_query_port(ctx, PORT_NUM, &mut port_attr as *mut _) }; if errno != 0 { return Err(io::Error::from_raw_os_error(errno)); } // From http://www.rdmamojo.com/2012/08/02/ibv_query_gid/: // // The content of the GID table is valid only when the port_attr.state is either // IBV_PORT_ARMED or IBV_PORT_ACTIVE. For other states of the port, the value of the GID // table is indeterminate. // match port_attr.state { ffi::ibv_port_state::IBV_PORT_ACTIVE | ffi::ibv_port_state::IBV_PORT_ARMED => {} _ => { return Err(io::Error::new( io::ErrorKind::Other, "port is not ACTIVE or ARMED".to_string(), )); } } let mut gid = ffi::ibv_gid::default(); let ok = unsafe { ffi::ibv_query_gid(ctx, PORT_NUM, 0, &mut gid as *mut _) }; if ok != 0 { return Err(io::Error::last_os_error()); } Ok(Context { ctx, port_attr, gid, }) } /// Create a completion queue (CQ). /// /// When an outstanding Work Request, within a Send or Receive Queue, is completed, a Work /// Completion is being added to the CQ of that Work Queue. This Work Completion indicates that /// the outstanding Work Request has been completed (and no longer considered outstanding) and /// provides details on it (status, direction, opcode, etc.). /// /// A single CQ can be shared for sending, receiving, and sharing across multiple QPs. The Work /// Completion holds the information to specify the QP number and the Queue (Send or Receive) /// that it came from. /// /// `min_cq_entries` defines the minimum size of the CQ. The actual created size can be equal /// or higher than this value. `id` is an opaque identifier that is echoed by /// `CompletionQueue::poll`. /// /// # Errors /// /// - `EINVAL`: Invalid `min_cq_entries` (must be `1 <= cqe <= dev_cap.max_cqe`). /// - `ENOMEM`: Not enough resources to complete this operation. pub fn create_cq(&self, min_cq_entries: i32, id: isize) -> io::Result<CompletionQueue> { let cq = unsafe { ffi::ibv_create_cq( self.ctx, min_cq_entries, ptr::null::<c_void>().offset(id) as *mut _, ptr::null::<c_void>() as *mut _, 0, ) }; if cq.is_null() { Err(io::Error::last_os_error()) } else { Ok(CompletionQueue { _phantom: PhantomData, cq: cq, }) } } /// Allocate a protection domain (PDs) for the device's context. /// /// The created PD will be used primarily to create `QueuePair`s and `MemoryRegion`s. /// /// A protection domain is a means of protection, and helps you create a group of object that /// can work together. If several objects were created using PD1, and others were created using /// PD2, working with objects from group1 together with objects from group2 will not work. pub fn alloc_pd(&self) -> Result<ProtectionDomain, ()> { let pd = unsafe { ffi::ibv_alloc_pd(self.ctx) }; if pd.is_null() { Err(()) } else { Ok(ProtectionDomain { ctx: self, pd }) } } } impl Drop for Context { fn drop(&mut self) { let ok = unsafe { ffi::ibv_close_device(self.ctx) }; assert_eq!(ok, 0); } } /// A completion queue that allows subscribing to the completion of queued sends and receives. pub struct CompletionQueue<'ctx> { _phantom: PhantomData<&'ctx ()>, cq: *mut ffi::ibv_cq, } unsafe impl<'a> Send for CompletionQueue<'a> {} unsafe impl<'a> Sync for CompletionQueue<'a> {} impl<'ctx> CompletionQueue<'ctx> { /// Poll for (possibly multiple) work completions. /// /// A Work Completion indicates that a Work Request in a Work Queue, and all of the outstanding /// unsignaled Work Requests that posted to that Work Queue, associated with this CQ have /// completed. Any Receive Requests, signaled Send Requests and Send Requests that ended with /// an error will generate Work Completions. /// /// When a Work Request ends, a Work Completion is added to the tail of the CQ that this Work /// Queue is associated with. `poll` checks if Work Completions are present in a CQ, and pop /// them from the head of the CQ in the order they entered it (FIFO) into `completions`. After /// a Work Completion was popped from a CQ, it cannot be returned to it. `poll` returns the /// subset of `completions` that successfully completed. If the returned slice has fewer /// elements than the provided `completions` slice, the CQ was emptied. /// /// Not all attributes of the completed `ibv_wc`'s are always valid. If the completion status /// is not `IBV_WC_SUCCESS`, only the following attributes are valid: `wr_id`, `status`, /// `qp_num`, and `vendor_err`. /// /// Note that `poll` does not block or cause a context switch. This is why RDMA technologies /// can achieve very low latency (below 1 µs). #[inline] pub fn poll<'c>( &self, completions: &'c mut [ffi::ibv_wc], ) -> Result<&'c mut [ffi::ibv_wc], ()> { // TODO: from http://www.rdmamojo.com/2013/02/15/ibv_poll_cq/ // // One should consume Work Completions at a rate that prevents the CQ from being overrun // (hold more Work Completions than the CQ size). In case of an CQ overrun, the async // event `IBV_EVENT_CQ_ERR` will be triggered, and the CQ cannot be used anymore. // let ctx: *mut ffi::ibv_context = unsafe { &*self.cq }.context; let ops = &mut unsafe { &mut *ctx }.ops; let n = unsafe { ops.poll_cq.as_mut().unwrap()( self.cq, completions.len() as i32, completions.as_mut_ptr(), ) }; if n < 0 { Err(()) } else { Ok(&mut completions[0..n as usize]) } } } impl<'a> Drop for CompletionQueue<'a> { fn drop(&mut self) { let errno = unsafe { ffi::ibv_destroy_cq(self.cq) }; if errno != 0 { let e = io::Error::from_raw_os_error(errno); panic!("{}", e); } } } /// An unconfigured `QueuePair`. /// /// A `QueuePairBuilder` is used to configure a `QueuePair` before it is allocated and initialized. /// To construct one, use `ProtectionDomain::create_qp`. See also [RDMAmojo] for many more details. /// /// [RDMAmojo]: http://www.rdmamojo.com/2013/01/12/ibv_modify_qp/ pub struct QueuePairBuilder<'res> { ctx: isize, pd: &'res ProtectionDomain<'res>, send: &'res CompletionQueue<'res>, max_send_wr: u32, recv: &'res CompletionQueue<'res>, max_recv_wr: u32, max_send_sge: u32, max_recv_sge: u32, max_inline_data: u32, qp_type: ffi::ibv_qp_type::Type, // carried along to handshake phase access: ffi::ibv_access_flags, timeout: u8, retry_count: u8, rnr_retry: u8, min_rnr_timer: u8, } impl<'res> QueuePairBuilder<'res> { /// Prepare a new `QueuePair` builder. /// /// `max_send_wr` is the maximum number of outstanding Work Requests that can be posted to the /// Send Queue in that Queue Pair. Value must be in `[0..dev_cap.max_qp_wr]`. There may be RDMA /// devices that for specific transport types may support less outstanding Work Requests than /// the maximum reported value. /// /// Similarly, `max_recv_wr` is the maximum number of outstanding Work Requests that can be /// posted to the Receive Queue in that Queue Pair. Value must be in `[0..dev_cap.max_qp_wr]`. /// There may be RDMA devices that for specific transport types may support less outstanding /// Work Requests than the maximum reported value. This value is ignored if the Queue Pair is /// associated with an SRQ fn new<'scq, 'rcq, 'pd>( pd: &'pd ProtectionDomain, send: &'scq CompletionQueue, max_send_wr: u32, recv: &'rcq CompletionQueue, max_recv_wr: u32, qp_type: ffi::ibv_qp_type::Type, ) -> QueuePairBuilder<'res> where 'scq: 'res, 'rcq: 'res, 'pd: 'res, { QueuePairBuilder { ctx: 0, pd: pd, send, max_send_wr, recv, max_recv_wr, max_send_sge: 1, max_recv_sge: 1, max_inline_data: 0, qp_type, access: ffi::ibv_access_flags::IBV_ACCESS_LOCAL_WRITE, min_rnr_timer: 16, retry_count: 6, rnr_retry: 6, timeout: 4, } } /// Set the access flags for the new `QueuePair`. /// /// Defaults to `IBV_ACCESS_LOCAL_WRITE`. pub fn set_access(&mut self, access: ffi::ibv_access_flags) -> &mut Self { self.access = access; self } /// Set the access flags of the new `QueuePair` such that it allows remote reads and writes. pub fn allow_remote_rw(&mut self) -> &mut Self { self.access = self.access | ffi::ibv_access_flags::IBV_ACCESS_REMOTE_WRITE | ffi::ibv_access_flags::IBV_ACCESS_REMOTE_READ; self } /// Sets the minimum RNR NAK Timer Field Value for the new `QueuePair`. /// /// Defaults to 16 (2.56 ms delay). /// Relevant only for RC QPs. /// /// When an incoming message to this QP should consume a Work Request from the Receive Queue, /// but no Work Request is outstanding on that Queue, the QP will send an RNR NAK packet to /// the initiator. It does not affect RNR NAKs sent for other reasons. The value must be one of /// the following values: /// /// - 0 - 655.36 ms delay /// - 1 - 0.01 ms delay /// - 2 - 0.02 ms delay /// - 3 - 0.03 ms delay /// - 4 - 0.04 ms delay /// - 5 - 0.06 ms delay /// - 6 - 0.08 ms delay /// - 7 - 0.12 ms delay /// - 8 - 0.16 ms delay /// - 9 - 0.24 ms delay /// - 10 - 0.32 ms delay /// - 11 - 0.48 ms delay /// - 12 - 0.64 ms delay /// - 13 - 0.96 ms delay /// - 14 - 1.28 ms delay /// - 15 - 1.92 ms delay /// - 16 - 2.56 ms delay /// - 17 - 3.84 ms delay /// - 18 - 5.12 ms delay /// - 19 - 7.68 ms delay /// - 20 - 10.24 ms delay /// - 21 - 15.36 ms delay /// - 22 - 20.48 ms delay /// - 23 - 30.72 ms delay /// - 24 - 40.96 ms delay /// - 25 - 61.44 ms delay /// - 26 - 81.92 ms delay /// - 27 - 122.88 ms delay /// - 28 - 163.84 ms delay /// - 29 - 245.76 ms delay /// - 30 - 327.68 ms delay /// - 31 - 491.52 ms delay pub fn set_min_rnr_timer(&mut self, timer: u8) -> &mut Self { self.min_rnr_timer = timer; self } /// Sets the minimum timeout that the new `QueuePair` waits for ACK/NACK from remote QP before /// retransmitting the packet. /// /// Defaults to 4 (65.536µs). /// Relevant only to RC QPs. /// /// The value zero is special value that waits an infinite time for the ACK/NACK (useful /// for debugging). This means that if any packet in a message is being lost and no ACK or NACK /// is being sent, no retry will ever occur and the QP will just stop sending data. /// /// For any other value of timeout, the time calculation is `4.096*2^timeout`µs, giving: /// /// - 0 - infinite /// - 1 - 8.192 µs /// - 2 - 16.384 µs /// - 3 - 32.768 µs /// - 4 - 65.536 µs /// - 5 - 131.072 µs /// - 6 - 262.144 µs /// - 7 - 524.288 µs /// - 8 - 1.048 ms /// - 9 - 2.097 ms /// - 10 - 4.194 ms /// - 11 - 8.388 ms /// - 12 - 16.777 ms /// - 13 - 33.554 ms /// - 14 - 67.108 ms /// - 15 - 134.217 ms /// - 16 - 268.435 ms /// - 17 - 536.870 ms /// - 18 - 1.07 s /// - 19 - 2.14 s /// - 20 - 4.29 s /// - 21 - 8.58 s /// - 22 - 17.1 s /// - 23 - 34.3 s /// - 24 - 68.7 s /// - 25 - 137 s /// - 26 - 275 s /// - 27 - 550 s /// - 28 - 1100 s /// - 29 - 2200 s /// - 30 - 4400 s /// - 31 - 8800 s pub fn set_timeout(&mut self, timeout: u8) -> &mut Self { self.timeout = timeout; self } /// Sets the total number of times that the new `QueuePair` will try to resend the packets /// before reporting an error because the remote side doesn't answer in the primary path. /// /// This 3 bit value defaults to 6. /// /// # Panics /// /// Panics if a count higher than 7 is given. pub fn set_retry_count(&mut self, count: u8) -> &mut Self { assert!(count <= 7); self.retry_count = count; self } /// Sets the total number of times that the new `QueuePair` will try to resend the packets when /// an RNR NACK was sent by the remote QP before reporting an error. /// /// This 3 bit value defaults to 6. The value 7 is special and specify to retry sending the /// message indefinitely when a RNR Nack is being sent by remote side. /// /// # Panics /// /// Panics if a limit higher than 7 is given. pub fn set_rnr_retry(&mut self, n: u8) -> &mut Self { assert!(n <= 7); self.rnr_retry = n; self } /// Set the opaque context value for the new `QueuePair`. /// /// Defaults to 0. pub fn set_context(&mut self, ctx: isize) -> &mut Self { self.ctx = ctx; self } /// Create a new `QueuePair` from this builder template. /// /// The returned `QueuePair` is associated with the builder's `ProtectionDomain`. /// /// This method will fail if asked to create QP of a type other than `IBV_QPT_RC` or /// `IBV_QPT_UD` associated with an SRQ. /// /// # Errors /// /// - `EINVAL`: Invalid `ProtectionDomain`, sending or receiving `Context`, or invalid value /// provided in `max_send_wr`, `max_recv_wr`, or in `max_inline_data`. /// - `ENOMEM`: Not enough resources to complete this operation. /// - `ENOSYS`: QP with this Transport Service Type isn't supported by this RDMA device. /// - `EPERM`: Not enough permissions to create a QP with this Transport Service Type. pub fn build(&self) -> io::Result<PreparedQueuePair<'res>> { let mut attr = ffi::ibv_qp_init_attr { qp_context: unsafe { ptr::null::<c_void>().offset(self.ctx) } as *mut _, send_cq: self.send.cq as *const _ as *mut _, recv_cq: self.recv.cq as *const _ as *mut _, srq: ptr::null::<ffi::ibv_srq>() as *mut _, cap: ffi::ibv_qp_cap { max_send_wr: self.max_send_wr, max_recv_wr: self.max_recv_wr, max_send_sge: self.max_send_sge, max_recv_sge: self.max_recv_sge, max_inline_data: self.max_inline_data, }, qp_type: self.qp_type, sq_sig_all: 0, }; let qp = unsafe { ffi::ibv_create_qp(self.pd.pd, &mut attr as *mut _) }; if qp.is_null() { Err(io::Error::last_os_error()) } else { Ok(PreparedQueuePair { ctx: self.pd.ctx, qp: qp, access: self.access, timeout: self.timeout, retry_count: self.retry_count, rnr_retry: self.rnr_retry, min_rnr_timer: self.min_rnr_timer, }) } } } /// An allocated but uninitialized `QueuePair`. /// /// Specifically, this `QueuePair` has been allocated with `ibv_create_qp`, but has not yet been /// initialized with calls to `ibv_modify_qp`. /// /// To complete the construction of the `QueuePair`, you will need to obtain the /// `QueuePairEndpoint` of the remote end (by using `PreparedQueuePair::endpoint`), and then call /// `PreparedQueuePair::handshake` on both sides with the other side's `QueuePairEndpoint`: /// /// ```rust,ignore /// // on host 1 /// let pqp: PreparedQueuePair = ...; /// let host1end = pqp.endpoint(); /// host2.send(host1end); /// let host2end = host2.recv(); /// let qp = pqp.handshake(host2end); /// /// // on host 2 /// let pqp: PreparedQueuePair = ...; /// let host2end = pqp.endpoint(); /// host1.send(host2end); /// let host1end = host1.recv(); /// let qp = pqp.handshake(host1end); /// ``` pub struct PreparedQueuePair<'res> { ctx: &'res Context, qp: *mut ffi::ibv_qp, // carried from builder access: ffi::ibv_access_flags, min_rnr_timer: u8, timeout: u8, retry_count: u8, rnr_retry: u8, } /// An identifier for the network endpoint of a `QueuePair`. /// /// Internally, this contains the `QueuePair`'s `qp_num`, as well as the context's `lid` and `gid`. #[derive(Copy, Clone)] pub struct QueuePairEndpoint { num: u32, lid: u16, gid: ffi::ibv_gid, } impl<'res> PreparedQueuePair<'res> { /// Get the network endpoint for this `QueuePair`. /// /// This endpoint will need to be communicated to the `QueuePair` on the remote end. pub fn endpoint(&self) -> QueuePairEndpoint { let num = unsafe { &*self.qp }.qp_num; QueuePairEndpoint { num, lid: self.ctx.port_attr.lid, gid: self.ctx.gid, } } /// Set up the `QueuePair` such that it is ready to exchange packets with a remote `QueuePair`. /// /// Internally, this uses `ibv_modify_qp` to mark the `QueuePair` as initialized /// (`IBV_QPS_INIT`), ready to receive (`IBV_QPS_RTR`), and ready to send (`IBV_QPS_RTS`). /// Further discussion of the protocol can be found on [RDMAmojo]. /// /// The handshake also sets the following parameters, which are currently not configurable: /// /// # Examples /// /// ```text,ignore /// port_num = PORT_NUM; /// pkey_index = 0; /// rq_psn = 0; /// sq_psn = 0; /// /// max_dest_rd_atomic = 1; /// max_rd_atomic = 1; /// /// ah_attr.sl = 0; /// ah_attr.is_global = 1; /// ah_attr.src_path_bits = 0; /// ah_attr.grh.hop_limit = 0xff; /// ``` /// /// # Errors /// /// - `EINVAL`: Invalid value provided in `attr` or in `attr_mask`. /// - `ENOMEM`: Not enough resources to complete this operation. /// /// [RDMAmojo]: http://www.rdmamojo.com/2014/01/18/connecting-queue-pairs/ pub fn handshake(self, remote: QueuePairEndpoint) -> io::Result<QueuePair<'res>> { // init and associate with port let mut attr = ffi::ibv_qp_attr::default(); attr.qp_state = ffi::ibv_qp_state::IBV_QPS_INIT; attr.qp_access_flags = self.access.0; attr.pkey_index = 0; attr.port_num = PORT_NUM; let mask = ffi::ibv_qp_attr_mask::IBV_QP_STATE | ffi::ibv_qp_attr_mask::IBV_QP_PKEY_INDEX | ffi::ibv_qp_attr_mask::IBV_QP_PORT | ffi::ibv_qp_attr_mask::IBV_QP_ACCESS_FLAGS; let errno = unsafe { ffi::ibv_modify_qp(self.qp, &mut attr as *mut _, mask.0 as i32) }; if errno != 0 { return Err(io::Error::from_raw_os_error(errno)); } // set ready to receive let mut attr = ffi::ibv_qp_attr::default(); attr.qp_state = ffi::ibv_qp_state::IBV_QPS_RTR; attr.path_mtu = self.ctx.port_attr.active_mtu; attr.dest_qp_num = remote.num; attr.rq_psn = 0; attr.max_dest_rd_atomic = 1; attr.min_rnr_timer = self.min_rnr_timer; attr.ah_attr.is_global = 1; attr.ah_attr.dlid = remote.lid; attr.ah_attr.sl = 0; attr.ah_attr.src_path_bits = 0; attr.ah_attr.port_num = PORT_NUM; attr.ah_attr.grh.dgid = remote.gid; attr.ah_attr.grh.hop_limit = 0xff; let mask = ffi::ibv_qp_attr_mask::IBV_QP_STATE | ffi::ibv_qp_attr_mask::IBV_QP_AV | ffi::ibv_qp_attr_mask::IBV_QP_PATH_MTU | ffi::ibv_qp_attr_mask::IBV_QP_DEST_QPN | ffi::ibv_qp_attr_mask::IBV_QP_RQ_PSN | ffi::ibv_qp_attr_mask::IBV_QP_MAX_DEST_RD_ATOMIC | ffi::ibv_qp_attr_mask::IBV_QP_MIN_RNR_TIMER; let errno = unsafe { ffi::ibv_modify_qp(self.qp, &mut attr as *mut _, mask.0 as i32) }; if errno != 0 { return Err(io::Error::from_raw_os_error(errno)); } // set ready to send let mut attr = ffi::ibv_qp_attr::default(); attr.qp_state = ffi::ibv_qp_state::IBV_QPS_RTS; attr.timeout = self.timeout; attr.retry_cnt = self.retry_count; attr.sq_psn = 0; attr.rnr_retry = self.rnr_retry; attr.max_rd_atomic = 1; let mask = ffi::ibv_qp_attr_mask::IBV_QP_STATE | ffi::ibv_qp_attr_mask::IBV_QP_TIMEOUT | ffi::ibv_qp_attr_mask::IBV_QP_RETRY_CNT | ffi::ibv_qp_attr_mask::IBV_QP_SQ_PSN | ffi::ibv_qp_attr_mask::IBV_QP_RNR_RETRY | ffi::ibv_qp_attr_mask::IBV_QP_MAX_QP_RD_ATOMIC; let errno = unsafe { ffi::ibv_modify_qp(self.qp, &mut attr as *mut _, mask.0 as i32) }; if errno != 0 { return Err(io::Error::from_raw_os_error(errno)); } Ok(QueuePair { _phantom: PhantomData, qp: self.qp, }) } } /// A memory region that has been registered for use with RDMA. pub struct MemoryRegion<T> { mr: *mut ffi::ibv_mr, data: Vec<T>, } unsafe impl<T> Send for MemoryRegion<T> {} unsafe impl<T> Sync for MemoryRegion<T> {} use std::ops::{Deref, DerefMut}; impl<T> Deref for MemoryRegion<T> { type Target = [T]; fn deref(&self) -> &Self::Target { &self.data[..] } } impl<T> DerefMut for MemoryRegion<T> { fn deref_mut(&mut self) -> &mut Self::Target { &mut self.data[..] } } impl<T> MemoryRegion<T> { /// Get the remote authentication key used to allow direct remote access to this memory region. pub fn rkey(&self) -> RemoteKey { RemoteKey(unsafe { &*self.mr }.rkey) } } /// A key that authorizes direct memory access to a memory region. pub struct RemoteKey(u32); impl<T> Drop for MemoryRegion<T> { fn drop(&mut self) { let errno = unsafe { ffi::ibv_dereg_mr(self.mr) }; if errno != 0 { let e = io::Error::from_raw_os_error(errno); panic!("{}", e); } } } /// A protection domain for a device's context. pub struct ProtectionDomain<'ctx> { ctx: &'ctx Context, pd: *mut ffi::ibv_pd, } unsafe impl<'a> Sync for ProtectionDomain<'a> {} unsafe impl<'a> Send for ProtectionDomain<'a> {} impl<'ctx> ProtectionDomain<'ctx> { /// Creates a queue pair builder associated with this protection domain. /// /// `send` and `recv` are the device `Context` to associate with the send and receive queues /// respectively. `send` and `recv` may refer to the same `Context`. /// /// `qp_type` indicates the requested Transport Service Type of this QP: /// /// - `IBV_QPT_RC`: Reliable Connection /// - `IBV_QPT_UC`: Unreliable Connection /// - `IBV_QPT_UD`: Unreliable Datagram /// /// Note that both this protection domain, *and* both provided completion queues, must outlive /// the resulting `QueuePair`. pub fn create_qp<'pd, 'scq, 'rcq, 'res>( &'pd self, send: &'scq CompletionQueue, recv: &'rcq CompletionQueue, qp_type: ffi::ibv_qp_type::Type, ) -> QueuePairBuilder<'res> where 'scq: 'res, 'rcq: 'res, 'pd: 'res, { QueuePairBuilder::new(self, send, 1, recv, 1, qp_type) } /// Allocates and registers a Memory Region (MR) associated with this `ProtectionDomain`. /// /// This process allows the RDMA device to read and write data to the allocated memory. Only /// registered memory can be sent from and received to by `QueuePair`s. Performing this /// registration takes some time, so performing memory registration isn't recommended in the /// data path, when fast response is required. /// /// Every successful registration will result with a MR which has unique (within a specific /// RDMA device) `lkey` and `rkey` values. These keys must be communicated to the other end's /// `QueuePair` for direct memory access. /// /// The maximum size of the block that can be registered is limited to /// `device_attr.max_mr_size`. There isn't any way to know what is the total size of memory /// that can be registered for a specific device. /// /// `allocate` currently sets the following permissions for each new `MemoryRegion`: /// /// - `IBV_ACCESS_LOCAL_WRITE`: Enables Local Write Access /// - `IBV_ACCESS_REMOTE_WRITE`: Enables Remote Write Access /// - `IBV_ACCESS_REMOTE_READ`: Enables Remote Read Access /// - `IBV_ACCESS_REMOTE_ATOMIC`: Enables Remote Atomic Operation Access (if supported) /// /// Local read access is always enabled for the MR. /// /// # Panics /// /// Panics if the size of the memory region zero bytes, which can occur either if `n` is 0, or /// if `mem::size_of::<T>()` is 0. /// /// # Errors /// /// - `EINVAL`: Invalid access value. /// - `ENOMEM`: Not enough resources (either in operating system or in RDMA device) to /// complete this operation. pub fn allocate<T: Sized + Copy + Default>(&self, n: usize) -> io::Result<MemoryRegion<T>> { assert!(n > 0); assert!(mem::size_of::<T>() > 0); let mut data = Vec::with_capacity(n); data.resize(n, T::default()); let access = ffi::ibv_access_flags::IBV_ACCESS_LOCAL_WRITE | ffi::ibv_access_flags::IBV_ACCESS_REMOTE_WRITE | ffi::ibv_access_flags::IBV_ACCESS_REMOTE_READ | ffi::ibv_access_flags::IBV_ACCESS_REMOTE_ATOMIC; let mr = unsafe { ffi::ibv_reg_mr( self.pd, data.as_mut_ptr() as *mut _, n * mem::size_of::<T>(), access.0 as i32, ) }; // TODO // ibv_reg_mr() returns a pointer to the registered MR, or NULL if the request fails. // The local key (L_Key) field lkey is used as the lkey field of struct ibv_sge when // posting buffers with ibv_post_* verbs, and the the remote key (R_Key) field rkey is // used by remote processes to perform Atomic and RDMA operations. The remote process // places this rkey as the rkey field of struct ibv_send_wr passed to the ibv_post_send // function. if mr.is_null() { Err(io::Error::last_os_error()) } else { Ok(MemoryRegion { mr, data }) } } } impl<'a> Drop for ProtectionDomain<'a> { fn drop(&mut self) { let errno = unsafe { ffi::ibv_dealloc_pd(self.pd) }; if errno != 0 { let e = io::Error::from_raw_os_error(errno); panic!("{}", e); } } } /// A fully initialized and ready `QueuePair`. /// /// A queue pair is the actual object that sends and receives data in the RDMA architecture /// (something like a socket). It's not exactly like a socket, however. A socket is an abstraction, /// which is maintained by the network stack and doesn't have a physical resource behind it. A QP /// is a resource of an RDMA device and a QP number can be used by one process at the same time /// (similar to a socket that is associated with a specific TCP or UDP port number) pub struct QueuePair<'res> { _phantom: PhantomData<&'res ()>, qp: *mut ffi::ibv_qp, } unsafe impl<'a> Send for QueuePair<'a> {} unsafe impl<'a> Sync for QueuePair<'a> {} impl<'res> QueuePair<'res> { /// Posts a linked list of Work Requests (WRs) to the Send Queue of this Queue Pair. /// /// Generates a HW-specific Send Request for the memory at `mr[range]`, and adds it to the tail /// of the Queue Pair's Send Queue without performing any context switch. The RDMA device will /// handle it (later) in asynchronous way. If there is a failure in one of the WRs because the /// Send Queue is full or one of the attributes in the WR is bad, it stops immediately and /// return the pointer to that WR. /// /// `wr_id` is a 64 bits value associated with this WR. If a Work Completion will be generated /// when this Work Request ends, it will contain this value. /// /// Internally, the memory at `mr[range]` will be sent as a single `ibv_send_wr` using /// `IBV_WR_SEND`. The send has `IBV_SEND_SIGNALED` set, so a work completion will also be /// triggered as a result of this send. /// /// See also [RDMAmojo's `ibv_post_send` documentation][1]. /// /// # Safety /// /// The memory region can only be safely reused or dropped after the request is fully executed /// and a work completion has been retrieved from the corresponding completion queue (i.e., /// until `CompletionQueue::poll` returns a completion for this send). /// /// # Errors /// /// - `EINVAL`: Invalid value provided in the Work Request. /// - `ENOMEM`: Send Queue is full or not enough resources to complete this operation. /// - `EFAULT`: Invalid value provided in `QueuePair`. /// /// [1]: http://www.rdmamojo.com/2013/01/26/ibv_post_send/ #[inline] pub unsafe fn post_send<T, R>( &mut self, mr: &mut MemoryRegion<T>, range: R, wr_id: u64, ) -> io::Result<()> where R: sliceindex::SliceIndex<[T], Output = [T]>, { let range = range.index(mr); let mut sge = ffi::ibv_sge { addr: range.as_ptr() as u64, length: (mem::size_of::<T>() * range.len()) as u32, lkey: (&*mr.mr).lkey, }; let mut wr = ffi::ibv_send_wr { wr_id: wr_id, next: ptr::null::<ffi::ibv_send_wr>() as *mut _, sg_list: &mut sge as *mut _, num_sge: 1, opcode: ffi::ibv_wr_opcode::IBV_WR_SEND, send_flags: ffi::ibv_send_flags::IBV_SEND_SIGNALED.0, wr: Default::default(), qp_type: Default::default(), __bindgen_anon_1: Default::default(), __bindgen_anon_2: Default::default(), }; let mut bad_wr: *mut ffi::ibv_send_wr = ptr::null::<ffi::ibv_send_wr>() as *mut _; // TODO: // // ibv_post_send() posts the linked list of work requests (WRs) starting with wr to the // send queue of the queue pair qp. It stops processing WRs from this list at the first // failure (that can be detected immediately while requests are being posted), and // returns this failing WR through bad_wr. // // The user should not alter or destroy AHs associated with WRs until request is fully // executed and a work completion has been retrieved from the corresponding completion // queue (CQ) to avoid unexpected behavior. // // ... However, if the IBV_SEND_INLINE flag was set, the buffer can be reused // immediately after the call returns. let ctx = (&*self.qp).context; let ops = &mut (&mut *ctx).ops; let errno = ops.post_send.as_mut().unwrap()(self.qp, &mut wr as *mut _, &mut bad_wr as *mut _); if errno != 0 { Err(io::Error::from_raw_os_error(errno)) } else { Ok(()) } } /// Posts a linked list of Work Requests (WRs) to the Receive Queue of this Queue Pair. /// /// Generates a HW-specific Receive Request out of it and add it to the tail of the Queue /// Pair's Receive Queue without performing any context switch. The RDMA device will take one /// of those Work Requests as soon as an incoming opcode to that QP will consume a Receive /// Request (RR). If there is a failure in one of the WRs because the Receive Queue is full or /// one of the attributes in the WR is bad, it stops immediately and return the pointer to that /// WR. /// /// `wr_id` is a 64 bits value associated with this WR. When a Work Completion is generated /// when this Work Request ends, it will contain this value. /// /// Internally, the memory at `mr[range]` will be received into as a single `ibv_recv_wr`. /// /// See also [DDMAmojo's `ibv_post_recv` documentation][1]. /// /// # Safety /// /// The memory region can only be safely reused or dropped after the request is fully executed /// and a work completion has been retrieved from the corresponding completion queue (i.e., /// until `CompletionQueue::poll` returns a completion for this receive). /// /// # Errors /// /// - `EINVAL`: Invalid value provided in the Work Request. /// - `ENOMEM`: Receive Queue is full or not enough resources to complete this operation. /// - `EFAULT`: Invalid value provided in `QueuePair`. /// /// [1]: http://www.rdmamojo.com/2013/02/02/ibv_post_recv/ #[inline] pub unsafe fn post_receive<T, R>( &mut self, mr: &mut MemoryRegion<T>, range: R, wr_id: u64, ) -> io::Result<()> where R: sliceindex::SliceIndex<[T], Output = [T]>, { let range = range.index(mr); let mut sge = ffi::ibv_sge { addr: range.as_ptr() as u64, length: (mem::size_of::<T>() * range.len()) as u32, lkey: (&*mr.mr).lkey, }; let mut wr = ffi::ibv_recv_wr { wr_id: wr_id, next: ptr::null::<ffi::ibv_send_wr>() as *mut _, sg_list: &mut sge as *mut _, num_sge: 1, }; let mut bad_wr: *mut ffi::ibv_recv_wr = ptr::null::<ffi::ibv_recv_wr>() as *mut _; // TODO: // // If the QP qp is associated with a shared receive queue, you must use the function // ibv_post_srq_recv(), and not ibv_post_recv(), since the QP's own receive queue will not // be used. // // If a WR is being posted to a UD QP, the Global Routing Header (GRH) of the incoming // message will be placed in the first 40 bytes of the buffer(s) in the scatter list. If no // GRH is present in the incoming message, then the first bytes will be undefined. This // means that in all cases, the actual data of the incoming message will start at an offset // of 40 bytes into the buffer(s) in the scatter list. let ctx = (&*self.qp).context; let ops = &mut (&mut *ctx).ops; let errno = ops.post_recv.as_mut().unwrap()(self.qp, &mut wr as *mut _, &mut bad_wr as *mut _); if errno != 0 { Err(io::Error::from_raw_os_error(errno)) } else { Ok(()) } } } impl<'a> Drop for QueuePair<'a> { fn drop(&mut self) { // TODO: ibv_destroy_qp() fails if the QP is attached to a multicast group. let errno = unsafe { ffi::ibv_destroy_qp(self.qp) }; if errno != 0 { let e = io::Error::from_raw_os_error(errno); panic!("{}", e); } } }