- 24 May, 2023 3 commits
-
-
Andrii Nakryiko authored
JP Kobryn says: ==================== Due to the way the datasec maps like bss, data, rodata are memory mapped, they cannot be resized with bpf_map__set_value_size() like non-datasec maps can. This series offers a way to allow the resizing of datasec maps, by having the mapped regions resized as needed and also adjusting associated BTF info if possible. The thought behind this is to allow for use cases where a given datasec needs to scale to for example the number of CPU's present. A bpf program can have a global array in a data section with an initial length and before loading the bpf program, the array length could be extended to match the CPU count. The selftests included in this series perform this scaling to an arbitrary value to demonstrate how it can work. ==================== Signed-off-by:Andrii Nakryiko <andrii@kernel.org>
-
JP Kobryn authored
This patch adds test coverage for resizing datasec maps. The first two subtests resize the bss and custom data sections. In both cases, an initial array (of length one) has its element set to one. After resizing the rest of the array is filled with ones as well. A BPF program is then run to sum the respective arrays and back on the userspace side the sum is checked to be equal to the number of elements. The third subtest attempts to perform resizing under conditions that will result in either the resize failing or the BTF info being cleared. Signed-off-by:
JP Kobryn <inwardvessel@gmail.com> Signed-off-by:
Stanislav Fomichev <sdf@google.com> Link: https://lore.kernel.org/bpf/20230524004537.18614-3-inwardvessel@gmail.com
-
JP Kobryn authored
This patch updates bpf_map__set_value_size() so that if the given map is memory mapped, it will attempt to resize the mapped region. Initial contents of the mapped region are preserved. BTF is not required, but after the mapping is resized an attempt is made to adjust the associated BTF information if the following criteria is met: - BTF info is present - the map is a datasec - the final variable in the datasec is an array ... the resulting BTF info will be updated so that the final array variable is associated with a new BTF array type sized to cover the requested size. Note that the initial resizing of the memory mapped region can succeed while the subsequent BTF adjustment can fail. In this case, BTF info is dropped from the map by clearing the key and value type. Signed-off-by:
-
- 23 May, 2023 7 commits
-
-
Andrii Nakryiko authored
Add a selftest demonstrating using detach-mounted BPF FS using new mount APIs, and pinning and getting BPF map using such mount. This demonstrates how something like container manager could setup BPF FS, pin and adjust all the necessary objects in it, all before exposing BPF FS to a particular mount namespace. Also add a few subtests validating all meaningful combinations of path_fd and pathname. We use mounted /sys/fs/bpf location for these. Signed-off-by:
Daniel Borkmann <daniel@iogearbox.net> Link: https://lore.kernel.org/bpf/20230523170013.728457-5-andrii@kernel.org
-
Andrii Nakryiko authored
Add path_fd support for bpf_obj_pin() and bpf_obj_get() operations (through their opts-based variants). This allows to take advantage of new kernel-side support for O_PATH-based pin/get location specification. Signed-off-by:
-
Andrii Nakryiko authored
Current UAPI of BPF_OBJ_PIN and BPF_OBJ_GET commands of bpf() syscall forces users to specify pinning location as a string-based absolute or relative (to current working directory) path. This has various implications related to security (e.g., symlink-based attacks), forces BPF FS to be exposed in the file system, which can cause races with other applications. One of the feedbacks we got from folks working with containers heavily was that inability to use purely FD-based location specification was an unfortunate limitation and hindrance for BPF_OBJ_PIN and BPF_OBJ_GET commands. This patch closes this oversight, adding path_fd field to BPF_OBJ_PIN and BPF_OBJ_GET UAPI, following conventions established by *at() syscalls for dirfd + pathname combinations. This now allows interesting possibilities like working with detached BPF FS mount (e.g., to perform multiple pinnings without running a risk of someone interfering with them), and generally making pinning/getting more secure and not prone to any races and/or security attacks. This is demonstrated by a selftest added in subsequent patch that takes advantage of new mount APIs (fsopen, fsconfig, fsmount) to demonstrate creating detached BPF FS mount, pinning, and then getting BPF map out of it, all while never exposing this private instance of BPF FS to outside worlds. Signed-off-by:
Christian Brauner <brauner@kernel.org> Link: https://lore.kernel.org/bpf/20230523170013.728457-4-andrii@kernel.org
-
Andrii Nakryiko authored
Bump libbpf.map to v1.3.0 to start a new libbpf version cycle. Signed-off-by:
-
Andrii Nakryiko authored
Do a sanity check whether provided file-to-be-pinned is actually a BPF object (prog, map, btf) before calling security_path_mknod LSM hook. If it's not, LSM hook doesn't have to be triggered, as the operation has no chance of succeeding anyways. Suggested-by:
-
Larysa Zaremba authored
Add ability to specify a network interface used to resolve XDP hints kfuncs when loading program through bpftool. Usage: bpftool prog load [...] xdpmeta_dev <ifname> Writing just 'dev <ifname>' instead of 'xdpmeta_dev' is a very probable mistake that results in not very descriptive errors, so 'bpftool prog load [...] dev <ifname>' syntax becomes deprecated, followed by 'bpftool map create [...] dev <ifname>' for consistency. Now, to offload program, execute: bpftool prog load [...] offload_dev <ifname> To offload map: bpftool map create [...] offload_dev <ifname> 'dev <ifname>' still performs offloading in the commands above, but now triggers a warning and is excluded from bash completion. 'xdpmeta_dev' and 'offload_dev' are mutually exclusive options, because 'xdpmeta_dev' basically makes a program device-bound without loading it onto the said device. For now, offloaded programs cannot use XDP hints [0], but if this changes, using 'offload_dev <ifname>' should cover this case. [0] https://lore.kernel.org/bpf/a5a636cc-5b03-686f-4be0-000383b05cfc@linux.devSigned-off-by:
Larysa Zaremba <larysa.zaremba@intel.com> Signed-off-by:
Quentin Monnet <quentin@isovalent.com> Link: https://lore.kernel.org/bpf/20230517160103.1088185-1-larysa.zaremba@intel.com
-
Lorenzo Bianconi authored
Introduce selftests to check xdp_feature support for bond driver. Signed-off-by:
Lorenzo Bianconi <lorenzo@kernel.org> Signed-off-by:
Jussi Maki <joamaki@gmail.com> Link: https://lore.kernel.org/bpf/64cb8f20e6491f5b971f8d3129335093c359aad7.1684329998.git.lorenzo@kernel.org
-
- 20 May, 2023 10 commits
-
-
Martin KaFai Lau authored
Aditi Ghag says: ==================== This patch set adds the capability to destroy sockets in BPF. We plan to use the capability in Cilium to force client sockets to reconnect when their remote load-balancing backends are deleted. The other use case is on-the-fly policy enforcement where existing socket connections prevented by policies need to be terminated. The use cases, and more details around the selected approach were presented at LPC 2022 - https://lpc.events/event/16/contributions/1358/. RFC discussion - https://lore.kernel.org/netdev/CABG=zsBEh-P4NXk23eBJw7eajB5YJeRS7oPXnTAzs=yob4EMoQ@mail.gmail.com/T/#u. v8 patch series - https://lore.kernel.org/bpf/20230517175359.527917-1-aditi.ghag@isovalent.com/ v9 highlights: Address review comments: Martin: - Rearranged the kfunc filter patch, and added the missing break statement. - Squashed the extended selftest/bpf patch. Yonghong: - Revised commit message for patch 1. (Below notes are same as v8 patch series that are still relevant. Refer to earlier patch series versions for other notes.) - I hit a snag while writing the kfunc where verifier complained about the `sock_common` type passed from TCP iterator. With kfuncs, there don't seem to be any options available to pass BTF type hints to the verifier (equivalent of `ARG_PTR_TO_BTF_ID_SOCK_COMMON`, as was the case with the helper). As a result, I changed the argument type of the sock_destory kfunc to `sock_common`. ==================== Signed-off-by:
Martin KaFai Lau <martin.lau@kernel.org>
-
Aditi Ghag authored
The test cases for destroying sockets mirror the intended usages of the bpf_sock_destroy kfunc using iterators. The destroy helpers set `ECONNABORTED` error code that we can validate in the test code with client sockets. But UDP sockets have an overriding error code from `disconnect()` called during abort, so the error code validation is only done for TCP sockets. The failure test cases validate that the `bpf_sock_destroy` kfunc is not allowed from program attach types other than BPF trace iterator, and such programs fail to load. Signed-off-by:
Aditi Ghag <aditi.ghag@isovalent.com> Link: https://lore.kernel.org/r/20230519225157.760788-10-aditi.ghag@isovalent.comSigned-off-by:
-
Aditi Ghag authored
The helper will be used to programmatically retrieve and pass ports in userspace and kernel selftest programs. Suggested-by:
-
Aditi Ghag authored
The socket destroy kfunc is used to forcefully terminate sockets from certain BPF contexts. We plan to use the capability in Cilium load-balancing to terminate client sockets that continue to connect to deleted backends. The other use case is on-the-fly policy enforcement where existing socket connections prevented by policies need to be forcefully terminated. The kfunc also allows terminating sockets that may or may not be actively sending traffic. The kfunc can currently be called only from BPF TCP and UDP iterators where users can filter, and terminate selected sockets. More specifically, it can only be called from BPF contexts that ensure socket locking in order to allow synchronous execution of protocol specific `diag_destroy` handlers. The previous commit that batches UDP sockets during iteration facilitated a synchronous invocation of the UDP destroy callback from BPF context by skipping socket locks in `udp_abort`. TCP iterator already supported batching of sockets being iterated. To that end, `tracing_iter_filter` callback filter is added so that verifier can restrict the kfunc to programs with `BPF_TRACE_ITER` attach type, and reject other programs. The kfunc takes `sock_common` type argument, even though it expects, and casts them to a `sock` pointer. This enables the verifier to allow the sock_destroy kfunc to be called for TCP with `sock_common` and UDP with `sock` structs. Furthermore, as `sock_common` only has a subset of certain fields of `sock`, casting pointer to the latter type might not always be safe for certain sockets like request sockets, but these have a special handling in the diag_destroy handlers. Additionally, the kfunc is defined with `KF_TRUSTED_ARGS` flag to avoid the cases where a `PTR_TO_BTF_ID` sk is obtained by following another pointer. eg. getting a sk pointer (may be even NULL) by following another sk pointer. The pointer socket argument passed in TCP and UDP iterators is tagged as `PTR_TRUSTED` in {tcp,udp}_reg_info. The TRUSTED arg changes are contributed by Martin KaFai Lau <martin.lau@kernel.org>. Signed-off-by: -
Aditi Ghag authored
This commit adds the ability to filter kfuncs to certain BPF program types. This is required to limit bpf_sock_destroy kfunc implemented in follow-up commits to programs with attach type 'BPF_TRACE_ITER'. The commit adds a callback filter to 'struct btf_kfunc_id_set'. The filter has access to the `bpf_prog` construct including its properties such as `expected_attached_type`. Signed-off-by:
-
Aditi Ghag authored
Batch UDP sockets from BPF iterator that allows for overlapping locking semantics in BPF/kernel helpers executed in BPF programs. This facilitates BPF socket destroy kfunc (introduced by follow-up patches) to execute from BPF iterator programs. Previously, BPF iterators acquired the sock lock and sockets hash table bucket lock while executing BPF programs. This prevented BPF helpers that again acquire these locks to be executed from BPF iterators. With the batching approach, we acquire a bucket lock, batch all the bucket sockets, and then release the bucket lock. This enables BPF or kernel helpers to skip sock locking when invoked in the supported BPF contexts. The batching logic is similar to the logic implemented in TCP iterator: https://lore.kernel.org/bpf/20210701200613.1036157-1-kafai@fb.com/. Suggested-by:
-
Aditi Ghag authored
This is a preparatory commit to remove the field. The field was previously shared between proc fs and BPF UDP socket iterators. As the follow-up commits will decouple the implementation for the iterators, remove the field. As for BPF socket iterator, filtering of sockets is exepected to be done in BPF programs. Suggested-by:
-
Aditi Ghag authored
This is a preparatory commit that encapsulates the logic to get udp table in iterator inside udp_get_table_afinfo, and renames the function to `udp_get_table_seq` accordingly. Suggested-by:
-
Aditi Ghag authored
This is a preparatory commit to refactor code that matches socket attributes in iterators to a helper function, and use it in the proc fs iterator. Signed-off-by:
-
Aditi Ghag authored
This is a preparatory commit to replace `lock_sock_fast` with `lock_sock`,and facilitate BPF programs executed from the TCP sockets iterator to be able to destroy TCP sockets using the bpf_sock_destroy kfunc (implemented in follow-up commits). Previously, BPF TCP iterator was acquiring the sock lock with BH disabled. This led to scenarios where the sockets hash table bucket lock can be acquired with BH enabled in some path versus disabled in other. In such situation, kernel issued a warning since it thinks that in the BH enabled path the same bucket lock *might* be acquired again in the softirq context (BH disabled), which will lead to a potential dead lock. Since bpf_sock_destroy also happens in a process context, the potential deadlock warning is likely a false alarm. Here is a snippet of annotated stack trace that motivated this change: ``` Possible interrupt unsafe locking scenario: CPU0 CPU1 ---- ---- lock(&h->lhash2[i].lock); local_bh_disable(); lock(&h->lhash2[i].lock); kernel imagined possible scenario: local_bh_disable(); /* Possible softirq */ lock(&h->lhash2[i].lock); *** Potential Deadlock *** process context: lock_acquire+0xcd/0x330 _raw_spin_lock+0x33/0x40 ------> Acquire (bucket) lhash2.lock with BH enabled __inet_hash+0x4b/0x210 inet_csk_listen_start+0xe6/0x100 inet_listen+0x95/0x1d0 __sys_listen+0x69/0xb0 __x64_sys_listen+0x14/0x20 do_syscall_64+0x3c/0x90 entry_SYSCALL_64_after_hwframe+0x72/0xdc bpf_sock_destroy run from iterator: lock_acquire+0xcd/0x330 _raw_spin_lock+0x33/0x40 ------> Acquire (bucket) lhash2.lock with BH disabled inet_unhash+0x9a/0x110 tcp_set_state+0x6a/0x210 tcp_abort+0x10d/0x200 bpf_prog_6793c5ca50c43c0d_iter_tcp6_server+0xa4/0xa9 bpf_iter_run_prog+0x1ff/0x340 ------> lock_sock_fast that acquires sock lock with BH disabled bpf_iter_tcp_seq_show+0xca/0x190 bpf_seq_read+0x177/0x450 ``` Also, Yonghong reported a deadlock for non-listening TCP sockets that this change resolves. Previously, `lock_sock_fast` held the sock spin lock with BH which was again being acquired in `tcp_abort`: ``` watchdog: BUG: soft lockup - CPU#0 stuck for 86s! [test_progs:2331] RIP: 0010:queued_spin_lock_slowpath+0xd8/0x500 Call Trace: <TASK> _raw_spin_lock+0x84/0x90 tcp_abort+0x13c/0x1f0 bpf_prog_88539c5453a9dd47_iter_tcp6_client+0x82/0x89 bpf_iter_run_prog+0x1aa/0x2c0 ? preempt_count_sub+0x1c/0xd0 ? from_kuid_munged+0x1c8/0x210 bpf_iter_tcp_seq_show+0x14e/0x1b0 bpf_seq_read+0x36c/0x6a0 bpf_iter_tcp_seq_show lock_sock_fast __lock_sock_fast spin_lock_bh(&sk->sk_lock.slock); /* * Fast path return with bottom halves disabled and * sock::sk_lock.slock held.* */ ... tcp_abort local_bh_disable(); spin_lock(&((sk)->sk_lock.slock)); // from bh_lock_sock(sk) ``` With the switch to `lock_sock`, it calls `spin_unlock_bh` before returning: ``` lock_sock lock_sock_nested spin_lock_bh(&sk->sk_lock.slock); : spin_unlock_bh(&sk->sk_lock.slock); ``` Acked-by:Yonghong Song <yhs@meta.com> Acked-by:
-
- 19 May, 2023 3 commits
-
-
Alexei Starovoitov authored
Yafang Shao says: ==================== The target_btf_id can help us understand which kernel function is linked by a tracing prog. The target_btf_id and target_obj_id have already been exposed to userspace, so we just need to show them. For some other link types like perf_event and kprobe_multi, it is not easy to find which functions are attached either. We may support ->fill_link_info for them in the future. v1->v2: - Skip showing them in the plain output for the old kernels. (Quentin) - Coding improvement. (Andrii) ==================== Signed-off-by:Alexei Starovoitov <ast@kernel.org>
-
Yafang Shao authored
The target_btf_id can help us understand which kernel function is linked by a tracing prog. The target_btf_id and target_obj_id have already been exposed to userspace, so we just need to show them. The result as follows, $ tools/bpf/bpftool/bpftool link show 2: tracing prog 13 prog_type tracing attach_type trace_fentry target_obj_id 1 target_btf_id 13964 pids trace(10673) $ tools/bpf/bpftool/bpftool link show -j [{"id":2,"type":"tracing","prog_id":13,"prog_type":"tracing","attach_type":"trace_fentry","target_obj_id":1,"target_btf_id":13964,"pids":[{"pid":10673,"comm":"trace"}]}] Signed-off-by:Yafang Shao <laoar.shao@gmail.com> Acked-by:
Song Liu <song@kernel.org> Link: https://lore.kernel.org/r/20230517103126.68372-3-laoar.shao@gmail.comSigned-off-by:
-
Yafang Shao authored
The target_btf_id can help us understand which kernel function is linked by a tracing prog. The target_btf_id and target_obj_id have already been exposed to userspace, so we just need to show them. The result as follows, $ cat /proc/10673/fdinfo/10 pos: 0 flags: 02000000 mnt_id: 15 ino: 2094 link_type: tracing link_id: 2 prog_tag: a04f5eef06a7f555 prog_id: 13 attach_type: 24 target_obj_id: 1 target_btf_id: 13964 Signed-off-by:
-
- 17 May, 2023 17 commits
-
-
Yonghong Song authored
Currently kernel kfunc bpf_dynptr_is_rdonly() has prototype ... __bpf_kfunc bool bpf_dynptr_is_rdonly(struct bpf_dynptr_kern *ptr) ... while selftests bpf_kfuncs.h has: extern int bpf_dynptr_is_rdonly(const struct bpf_dynptr *ptr) __ksym; Such a mismatch might cause problems although currently it is okay in selftests. Fix it to prevent future potential surprise. Signed-off-by:
Yonghong Song <yhs@fb.com> Signed-off-by:
-
Yonghong Song authored
With latest llvm17, dynptr/test_dynptr_is_null subtest failed in my testing VM. The failure log looks like below: All error logs: tester_init:PASS:tester_log_buf 0 nsec process_subtest:PASS:obj_open_mem 0 nsec process_subtest:PASS:Can't alloc specs array 0 nsec verify_success:PASS:dynptr_success__open 0 nsec verify_success:PASS:bpf_object__find_program_by_name 0 nsec verify_success:PASS:dynptr_success__load 0 nsec verify_success:PASS:bpf_program__attach 0 nsec verify_success:FAIL:err unexpected err: actual 4 != expected 0 #65/9 dynptr/test_dynptr_is_null:FAIL The error happens for bpf prog test_dynptr_is_null in dynptr_success.c: if (bpf_dynptr_is_null(&ptr2)) { err = 4; goto exit; } The bpf_dynptr_is_null(&ptr) unexpectedly returned a non-zero value and the control went to the error path. Digging further, I found the root cause is due to function signature difference between kernel and user space. In kernel, we have ... __bpf_kfunc bool bpf_dynptr_is_null(struct bpf_dynptr_kern *ptr) ... while in bpf_kfuncs.h we have: extern int bpf_dynptr_is_null(const struct bpf_dynptr *ptr) __ksym; The kernel bpf_dynptr_is_null disasm code: ffffffff812f1a90 <bpf_dynptr_is_null>: ffffffff812f1a90: f3 0f 1e fa endbr64 ffffffff812f1a94: 0f 1f 44 00 00 nopl (%rax,%rax) ffffffff812f1a99: 53 pushq %rbx ffffffff812f1a9a: 48 89 fb movq %rdi, %rbx ffffffff812f1a9d: e8 ae 29 17 00 callq 0xffffffff81464450 <__asan_load8_noabort> ffffffff812f1aa2: 48 83 3b 00 cmpq $0x0, (%rbx) ffffffff812f1aa6: 0f 94 c0 sete %al ffffffff812f1aa9: 5b popq %rbx ffffffff812f1aaa: c3 retq Note that only 1-byte register %al is set and the other 7-bytes are not touched. In bpf program, the asm code for the above bpf_dynptr_is_null(&ptr2): 266: 85 10 00 00 ff ff ff ff call -0x1 267: b4 01 00 00 04 00 00 00 w1 = 0x4 268: 16 00 03 00 00 00 00 00 if w0 == 0x0 goto +0x3 <LBB9_8> Basically, 4-byte subregister is tested. This might cause error as the value other than the lowest byte might not be 0. This patch fixed the issue by using the identical func prototype across kernel and selftest user space. The fixed bpf asm code: 267: 85 10 00 00 ff ff ff ff call -0x1 268: 54 00 00 00 01 00 00 00 w0 &= 0x1 269: b4 01 00 00 04 00 00 00 w1 = 0x4 270: 16 00 03 00 00 00 00 00 if w0 == 0x0 goto +0x3 <LBB9_8> Signed-off-by: -
Dave Thaler authored
Update the documentation regarding shift operations to explain the use of a mask, since otherwise shifting by a value out of range (like negative) is undefined. Signed-off-by:
Dave Thaler <dthaler@microsoft.com> Signed-off-by:
-
Pengcheng Yang authored
Currently, when using prog loadall and the pin path is a bpffs mountpoint, bpffs will be repeatedly mounted to the parent directory of the bpffs mountpoint path. For example, a `bpftool prog loadall test.o /sys/fs/bpf` will trigger this. Signed-off-by:
Pengcheng Yang <yangpc@wangsu.com> Signed-off-by:
-
Alexey Gladkov authored
The sign-file utility (from scripts/) is used in prog_tests/verify_pkcs7_sig.c, but the utility should not be called as a test. Executing this utility produces the following error: selftests: /linux/tools/testing/selftests/bpf: urandom_read ok 16 selftests: /linux/tools/testing/selftests/bpf: urandom_read selftests: /linux/tools/testing/selftests/bpf: sign-file not ok 17 selftests: /linux/tools/testing/selftests/bpf: sign-file # exit=2 Also, urandom_read is mistakenly used as a test. It does not lead to an error, but should be moved over to TEST_GEN_FILES as well. The empty TEST_CUSTOM_PROGS can then be removed. Fixes: fc975906 ("selftests/bpf: Add test for bpf_verify_pkcs7_signature() kfunc") Signed-off-by:
Alexey Gladkov <legion@kernel.org> Signed-off-by:
Roberto Sassu <roberto.sassu@huawei.com> Acked-by:
-
Andrii Nakryiko authored
It's trivial for user to trigger "verifier log line truncated" warning, as verifier has a fixed-sized buffer of 1024 bytes (as of now), and there are at least two pieces of user-provided information that can be output through this buffer, and both can be arbitrarily sized by user: - BTF names; - BTF.ext source code lines strings. Verifier log buffer should be properly sized for typical verifier state output. But it's sort-of expected that this buffer won't be long enough in some circumstances. So let's drop the check. In any case code will work correctly, at worst truncating a part of a single line output. Reported-by: syzbot+8b2a08dfbd25fd933d75@syzkaller.appspotmail.com Signed-off-by:
-
Alexei Starovoitov authored
Magnus Karlsson says: ==================== Prepare the AF_XDP selftests test framework code for the upcoming multi-buffer support in AF_XDP. This so that the multi-buffer patch set does not become way too large. In that upcoming patch set, we are only including the multi-buffer tests together with any framework code that depends on the new options bit introduced in the AF_XDP multi-buffer implementation itself. Currently, the test framework is based on the premise that a packet consists of a single fragment and thus occupies a single buffer and a single descriptor. Multi-buffer breaks this assumption, as that is the whole purpose of it. Now, a packet can consist of multiple buffers and therefore consume multiple descriptors. The patch set starts with some clean-ups and simplifications followed by patches that make sure that the current code works even when a packet occupies multiple buffers. The actual code for sending and receiving multi-buffer packets will be included in the AF_XDP multi-buffer patch set as it depends on a new bit being used in the options field of the descriptor. Patch set anatomy: 1: The XDP program was unnecessarily changed many times. Fixes this. 2: There is no reason to generate a full UDP/IPv4 packet as it is never used. Simplify the code by just generating a valid Ethernet frame. 3: Introduce a more complicated payload pattern that can detect fragments out of bounds in a multi-buffer packet and other errors found in single-fragment packets. 4: As a convenience, dump the content of the faulty packet at error. 5: To simplify the code, make the usage of the packet stream for Tx and Rx more similar. 6: Store the offset of the packet in the buffer in the struct pkt definition instead of the address in the umem itself and introduce a simple buffer allocator. The address only made sense when all packets consumed a single buffer. Now, we do not know beforehand how many buffers a packet will consume, so we instead just allocate a buffer from the allocator and specify the offset within that buffer. 7: Test for huge pages only once instead of before each test that needs it. 8: Populate the fill ring based on how many frags are needed for each packet. 9: Change the data generation code so it can generate data for multi-buffer packets too. 10: Adjust the packet pacing algorithm so that it can cope with multi-buffer packets. The pacing algorithm is present so that Tx does not send too many packets/frames to Rx that it starts to drop packets. That would ruin the tests. v1 -> v2: * Fixed spelling error in patch #6 [Simon] * Fixed compilation error with llvm in patch #7 [Daniel] Thanks: Magnus ==================== Signed-off-by: -
Magnus Karlsson authored
Modify the packet pacing algorithm so that it works with multi-buffer packets. This algorithm makes sure we do not send too many buffers to the receiving thread so that packets have to be dropped. The previous algorithm made the assumption that each packet only consumes one buffer, but that is not true anymore when multi-buffer support gets added. Instead, we find out what the largest packet size is in the packet stream and assume that each packet will consume this many buffers. This is conservative and overly cautious as there might be smaller packets in the stream that need fewer buffers per packet. But it keeps the algorithm simple. Also simplify it by removing the pthread conditional and just test if there is enough space in the Rx thread before trying to send one more batch. Also makes the tests run faster. Signed-off-by:
Magnus Karlsson <magnus.karlsson@intel.com> Link: https://lore.kernel.org/r/20230516103109.3066-11-magnus.karlsson@gmail.comSigned-off-by:
-
Magnus Karlsson authored
Add the ability to generate data in the packets that are correct for multi-buffer packets. The ethernet header should only go into the first fragment followed by data and the others should only have data. We also need to modify the pkt_dump function so that it knows what fragment has an ethernet header so it can print this. Signed-off-by:
-
Magnus Karlsson authored
Populate the fill ring based on the number of frags a packet needs. With multi-buffer support, a packet might require more than a single fragment/buffer, so the function xsk_populate_fill_ring() needs to consider how many buffers a packet will consume, and put that many buffers on the fill ring for each packet it should receive. As we are still not sending any multi-buffer packets, the function will only produce one buffer per packet at the moment. Signed-off-by:
-
Magnus Karlsson authored
Test for hugepages only once at the beginning of the execution of the whole test suite, instead of before each test that needs huge pages. These are the tests that use unaligned mode. As more unaligned tests will be added, so the current system just does not scale. With this change, there are now three possible outcomes of a test run: fail, pass, or skip. To simplify the handling of this, the function testapp_validate_traffic() now returns this value to the main loop. As this function is used by nearly all tests, it meant a small change to most of them. Signed-off-by:
-
Magnus Karlsson authored
Store the offset in struct pkt instead of the address. This is important since address is only meaningful in the context of a packet that is stored in a single umem buffer and thus a single Tx descriptor. If the packet, in contrast need to be represented by multiple buffers in the umem, storing the address makes no sense since the packet will consist of multiple buffers in the umem at various addresses. This change is in preparation for the upcoming multi-buffer support in AF_XDP and the corresponding tests. So instead of indicating the address, we instead indicate the offset of the packet in the first buffer. The actual address of the buffer is allocated from the umem with a new function called umem_alloc_buffer(). This also means we can get rid of the use_fill_for_addr flag as the addresses fed into the fill ring will always be the offset from the pkt specification in the packet stream plus the address of the allocated buffer from the umem. No special casing needed. Signed-off-by:
-
Magnus Karlsson authored
Convert the current variable rx_pkt_nb to an iterator that can be used for both Rx and Tx. This to simplify the code and making Tx more like Rx that already has this feature. Signed-off-by:
-
Magnus Karlsson authored
Dump the content of the packet when a test finds that packets are received out of order, the length is wrong, or some other packet error. Use the already existing pkt_dump function for this and call it when the above errors are detected. Get rid of the command line option for dumping packets as it is not useful to print out thousands of good packets followed by the faulty one you would like to see. Signed-off-by:
-
Magnus Karlsson authored
Add a varying payload pattern within the packet. Instead of having just a packet number that is the same for all words in a packet, make each word different in the packet. The upper 16-bits are set to the packet number and the lower 16-bits are the sequence number of the words in this packet. So the 3rd packet's 5th 32-bit word of data will contain the number (2<<32) | 4 as they are numbered from 0. This will make it easier to detect fragments that are out of order when starting to test multi-buffer support. The member payload in the packet is renamed pkt_nb to reflect that it is now only a pkt_nb, not the real payload as seen above. Signed-off-by:
-
Magnus Karlsson authored
Implement support for generating pkts with variable length. Before this patch, they were all 64 bytes, exception for some packets of zero length and some that were too large. This feature will be used to test multi-buffer support for which large packets are needed. The packets are also made simpler, just a valid Ethernet header followed by a sequence number. This so that it will become easier to implement packet generation when each packet consists of multiple fragments. There is also a maintenance burden associated with carrying all this code for generating proper UDP/IP packets, especially since they are not needed. Signed-off-by:
-
Magnus Karlsson authored
Do not change the XDP program for the Tx thread when not needed. It was erroneously compared to the XDP program for the Rx thread, which is always going to be different, which meant that the code made unnecessary switches to the same program it had before. This did not affect functionality, just performance. Signed-off-by:
-
