| Message ID | 1601292670-1616-1-git-send-email-alan.maguire@oracle.com |
|---|---|
| Headers | show |
| Series | bpf: add helpers to support BTF-based kernel data display | expand |
On Mon, Sep 28, 2020 at 4:33 AM Alan Maguire <alan.maguire@oracle.com> wrote: > > Default output for an sk_buff looks like this (zeroed fields > are omitted): > > (struct sk_buff){ > .transport_header = (__u16)65535, > .mac_header = (__u16)65535, > .end = (sk_buff_data_t)192, > .head = (unsigned char *)0x000000007524fd8b, > .data = (unsigned char *)0x000000007524fd8b, > .truesize = (unsigned int)768, > .users = (refcount_t){ > .refs = (atomic_t){ > .counter = (int)1, > }, > }, > } > > Flags can modify aspects of output format; see patch 3 > for more details. Applied. Thanks a lot.
On Mon, Sep 28, 2020 at 4:36 AM Alan Maguire <alan.maguire@oracle.com> wrote: > > Add a test verifying iterating over tasks and displaying BTF > representation of task_struct succeeds. > > Suggested-by: Alexei Starovoitov <alexei.starovoitov@gmail.com> > Signed-off-by: Alan Maguire <alan.maguire@oracle.com> > --- Hey Alan, These selftests rely on having struct btf_ptr and BTF_F_COMPACT (and other) flags to be present in vmlinux.h. While there is nothing wrong with that per se, it does break selftests builds on older kernels, because there struct btf_ptr isn't yet present in kernel: progs/netif_receive_skb.c:131:21: error: use of undeclared identifier 'BTF_F_NONAME' TEST_BTF(str, int, BTF_F_NONAME, "0", 0); ^ progs/netif_receive_skb.c:131:2: error: use of undeclared identifier 'BTF_F_COMPACT'; did you mean 'TT_COMPAT'? TEST_BTF(str, int, BTF_F_NONAME, "0", 0); ^ progs/netif_receive_skb.c:50:28: note: expanded from macro 'TEST_BTF' __u64 _hflags = _flags | BTF_F_COMPACT; \ ^ /data/users/andriin/linux/tools/testing/selftests/bpf/tools/include/vmlinux.h:330:2: note: 'TT_COMPAT' declared here TT_COMPAT = 2, ^ fatal error: too many errors emitted, stopping now [-ferror-limit=] progs/bpf_iter_task_btf.c:21:24: error: variable has incomplete type 'struct btf_ptr' static struct btf_ptr ptr = { }; ^ /data/users/andriin/linux/tools/testing/selftests/bpf/tools/include/bpf/bpf_helper_defs.h:33:8: note: forward declaration of 'struct btf_ptr' struct btf_ptr; We actually do build the very latest selftests against old kernels (4.9 and 5.5 at the moment) as part of libbpf CI, so it would be nice to fix this problem and keep selftests compilable. Do you mind following up with a change to define struct btf_ptr and those BTF_F_xxx flags explicitly for selftests only, similarly to how we do it for bpf_iter context structs? See progs/bpf_iter.h for examples. Thanks. > tools/testing/selftests/bpf/prog_tests/bpf_iter.c | 74 ++++++++++++++++++++++ > .../selftests/bpf/progs/bpf_iter_task_btf.c | 50 +++++++++++++++ > 2 files changed, 124 insertions(+) > create mode 100644 tools/testing/selftests/bpf/progs/bpf_iter_task_btf.c > [...]
On 9/28/20 1:31 PM, Alan Maguire wrote: > Tests verifying snprintf()ing of various data structures, > flags combinations using a tp_btf program. Tests are skipped > if __builtin_btf_type_id is not available to retrieve BTF > type ids. > > Signed-off-by: Alan Maguire <alan.maguire@oracle.com> [...] > +void test_snprintf_btf(void) > +{ > + struct netif_receive_skb *skel; > + struct netif_receive_skb__bss *bss; > + int err, duration = 0; > + > + skel = netif_receive_skb__open(); > + if (CHECK(!skel, "skel_open", "failed to open skeleton\n")) > + return; > + > + err = netif_receive_skb__load(skel); > + if (CHECK(err, "skel_load", "failed to load skeleton: %d\n", err)) > + goto cleanup; > + > + bss = skel->bss; > + > + err = netif_receive_skb__attach(skel); > + if (CHECK(err, "skel_attach", "skeleton attach failed: %d\n", err)) > + goto cleanup; > + > + /* generate receive event */ > + system("ping -c 1 127.0.0.1 > /dev/null"); This generates the following new warning when compiling BPF selftests: [...] EXT-OBJ [test_progs] cgroup_helpers.o EXT-OBJ [test_progs] trace_helpers.o EXT-OBJ [test_progs] network_helpers.o EXT-OBJ [test_progs] testing_helpers.o TEST-OBJ [test_progs] snprintf_btf.test.o /root/bpf-next/tools/testing/selftests/bpf/prog_tests/snprintf_btf.c: In function ‘test_snprintf_btf’: /root/bpf-next/tools/testing/selftests/bpf/prog_tests/snprintf_btf.c:30:2: warning: ignoring return value of ‘system’, declared with attribute warn_unused_result [-Wunused-result] system("ping -c 1 127.0.0.1 > /dev/null"); ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [...] Please fix, thx!
This series attempts to provide a simple way for BPF programs (and in future other consumers) to utilize BPF Type Format (BTF) information to display kernel data structures in-kernel. The use case this functionality is applied to here is to support a snprintf()-like helper to copy a BTF representation of kernel data to a string, and a BPF seq file helper to display BTF data for an iterator. There is already support in kernel/bpf/btf.c for "show" functionality; the changes here generalize that support from seq-file specific verifier display to the more generic case and add another specific use case; rather than seq_printf()ing the show data, it is copied to a supplied string using a snprintf()-like function. Other future consumers of the show functionality could include a bpf_printk_btf() function which printk()ed the data instead. Oops messaging in particular would be an interesting application for such functionality. The above potential use case hints at a potential reply to a reasonable objection that such typed display should be solved by tracing programs, where the in-kernel tracing records data and the userspace program prints it out. While this is certainly the recommended approach for most cases, I believe having an in-kernel mechanism would be valuable also. Critically in BPF programs it greatly simplifies debugging and tracing of such data to invoking a simple helper. One challenge raised in an earlier iteration of this work - where the BTF printing was implemented as a printk() format specifier - was that the amount of data printed per printk() was large, and other format specifiers were far simpler. Here we sidestep that concern by printing components of the BTF representation as we go for the seq file case, and in the string case the snprintf()-like operation is intended to be a basis for perf event or ringbuf output. The reasons for avoiding bpf_trace_printk are that 1. bpf_trace_printk() strings are restricted in size and cannot display anything beyond trivial data structures; and 2. bpf_trace_printk() is for debugging purposes only. As Alexei suggested, a bpf_trace_puts() helper could solve this in the future but it still would be limited by the 1000 byte limit for traced strings. Default output for an sk_buff looks like this (zeroed fields are omitted): (struct sk_buff){ .transport_header = (__u16)65535, .mac_header = (__u16)65535, .end = (sk_buff_data_t)192, .head = (unsigned char *)0x000000007524fd8b, .data = (unsigned char *)0x000000007524fd8b, .truesize = (unsigned int)768, .users = (refcount_t){ .refs = (atomic_t){ .counter = (int)1, }, }, } Flags can modify aspects of output format; see patch 3 for more details. Changes since v6: - Updated safe data size to 32, object name size to 80. This increases the number of safe copies done, but performance is not a key goal here. WRT name size the largest type name length in bpf-next according to "pahole -s" is 64 bytes, so that still gives room for additional type qualifiers, parens etc within the name limit (Alexei, patch 2) - Remove inlines and converted as many #defines to functions as was possible. In a few cases - btf_show_type_value[s]() specifically - I left these as macros as btf_show_type_value[s]() prepends and appends format strings to the format specifier (in order to include indentation, delimiters etc so a macro makes that simpler (Alexei, patch 2) - Handle btf_resolve_size() error in btf_show_obj_safe() (Alexei, patch 2) - Removed clang loop unroll in BTF snprintf test (Alexei) - switched to using bpf_core_type_id_kernel(type) as suggested by Andrii, and Alexei noted that __builtin_btf_type_id(,1) should be used (patch 4) - Added skip logic if __builtin_btf_type_id is not available (patches 4,8) - Bumped limits on bpf iters to support printing larger structures (Alexei, patch 5) - Updated overflow bpf_iter tests to reflect new iter max size (patch 6) - Updated seq helper to use type id only (Alexei, patch 7) - Updated BTF task iter test to use task struct instead of struct fs_struct since new limits allow a task_struct to be displayed (patch 8) - Fixed E2BIG handling in iter task (Alexei, patch 8) Changes since v5: - Moved btf print prepare into patch 3, type show seq with flags into patch 2 (Alexei, patches 2,3) - Fixed build bot warnings around static declarations and printf attributes - Renamed functions to snprintf_btf/seq_printf_btf (Alexei, patches 3-6) Changes since v4: - Changed approach from a BPF trace event-centric design to one utilizing a snprintf()-like helper and an iter helper (Alexei, patches 3,5) - Added tests to verify BTF output (patch 4) - Added support to tests for verifying BTF type_id-based display as well as type name via __builtin_btf_type_id (Andrii, patch 4). - Augmented task iter tests to cover the BTF-based seq helper. Because a task_struct's BTF-based representation would overflow the PAGE_SIZE limit on iterator data, the "struct fs_struct" (task->fs) is displayed for each task instead (Alexei, patch 6). Changes since v3: - Moved to RFC since the approach is different (and bpf-next is closed) - Rather than using a printk() format specifier as the means of invoking BTF-enabled display, a dedicated BPF helper is used. This solves the issue of printk() having to output large amounts of data using a complex mechanism such as BTF traversal, but still provides a way for the display of such data to be achieved via BPF programs. Future work could include a bpf_printk_btf() function to invoke display via printk() where the elements of a data structure are printk()ed one at a time. Thanks to Petr Mladek, Andy Shevchenko and Rasmus Villemoes who took time to look at the earlier printk() format-specifier-focused version of this and provided feedback clarifying the problems with that approach. - Added trace id to the bpf_trace_printk events as a means of separating output from standard bpf_trace_printk() events, ensuring it can be easily parsed by the reader. - Added bpf_trace_btf() helper tests which do simple verification of the various display options. Changes since v2: - Alexei and Yonghong suggested it would be good to use probe_kernel_read() on to-be-shown data to ensure safety during operation. Safe copy via probe_kernel_read() to a buffer object in "struct btf_show" is used to support this. A few different approaches were explored including dynamic allocation and per-cpu buffers. The downside of dynamic allocation is that it would be done during BPF program execution for bpf_trace_printk()s using %pT format specifiers. The problem with per-cpu buffers is we'd have to manage preemption and since the display of an object occurs over an extended period and in printk context where we'd rather not change preemption status, it seemed tricky to manage buffer safety while considering preemption. The approach of utilizing stack buffer space via the "struct btf_show" seemed like the simplest approach. The stack size of the associated functions which have a "struct btf_show" on their stack to support show operation (btf_type_snprintf_show() and btf_type_seq_show()) stays under 500 bytes. The compromise here is the safe buffer we use is small - 256 bytes - and as a result multiple probe_kernel_read()s are needed for larger objects. Most objects of interest are smaller than this (e.g. "struct sk_buff" is 224 bytes), and while task_struct is a notable exception at ~8K, performance is not the priority for BTF-based display. (Alexei and Yonghong, patch 2). - safe buffer use is the default behaviour (and is mandatory for BPF) but unsafe display - meaning no safe copy is done and we operate on the object itself - is supported via a 'u' option. - pointers are prefixed with 0x for clarity (Alexei, patch 2) - added additional comments and explanations around BTF show code, especially around determining whether objects such zeroed. Also tried to comment safe object scheme used. (Yonghong, patch 2) - added late_initcall() to initialize vmlinux BTF so that it would not have to be initialized during printk operation (Alexei, patch 5) - removed CONFIG_BTF_PRINTF config option as it is not needed; CONFIG_DEBUG_INFO_BTF can be used to gate test behaviour and determining behaviour of type-based printk can be done via retrieval of BTF data; if it's not there BTF was unavailable or broken (Alexei, patches 4,6) - fix bpf_trace_printk test to use vmlinux.h and globals via skeleton infrastructure, removing need for perf events (Andrii, patch 8) Changes since v1: - changed format to be more drgn-like, rendering indented type info along with type names by default (Alexei) - zeroed values are omitted (Arnaldo) by default unless the '0' modifier is specified (Alexei) - added an option to print pointer values without obfuscation. The reason to do this is the sysctls controlling pointer display are likely to be irrelevant in many if not most tracing contexts. Some questions on this in the outstanding questions section below... - reworked printk format specifer so that we no longer rely on format %pT<type> but instead use a struct * which contains type information (Rasmus). This simplifies the printk parsing, makes use more dynamic and also allows specification by BTF id as well as name. - removed incorrect patch which tried to fix dereferencing of resolved BTF info for vmlinux; instead we skip modifiers for the relevant case (array element type determination) (Alexei). - fixed issues with negative snprintf format length (Rasmus) - added test cases for various data structure formats; base types, typedefs, structs, etc. - tests now iterate through all typedef, enum, struct and unions defined for vmlinux BTF and render a version of the target dummy value which is either all zeros or all 0xff values; the idea is this exercises the "skip if zero" and "print everything" cases. - added support in BPF for using the %pT format specifier in bpf_trace_printk() - added BPF tests which ensure %pT format specifier use works (Alexei). Alan Maguire (8): bpf: provide function to get vmlinux BTF information bpf: move to generic BTF show support, apply it to seq files/strings bpf: add bpf_snprintf_btf helper selftests/bpf: add bpf_snprintf_btf helper tests bpf: bump iter seq size to support BTF representation of large data structures selftests/bpf: fix overflow tests to reflect iter size increase bpf: add bpf_seq_printf_btf helper selftests/bpf: add test for bpf_seq_printf_btf helper include/linux/bpf.h | 3 + include/linux/btf.h | 39 + include/uapi/linux/bpf.h | 76 ++ kernel/bpf/bpf_iter.c | 4 +- kernel/bpf/btf.c | 1007 ++++++++++++++++++-- kernel/bpf/core.c | 2 + kernel/bpf/helpers.c | 4 + kernel/bpf/verifier.c | 18 +- kernel/trace/bpf_trace.c | 98 ++ scripts/bpf_helpers_doc.py | 2 + tools/include/uapi/linux/bpf.h | 76 ++ tools/testing/selftests/bpf/prog_tests/bpf_iter.c | 88 +- .../selftests/bpf/prog_tests/snprintf_btf.c | 60 ++ .../selftests/bpf/progs/bpf_iter_task_btf.c | 50 + .../selftests/bpf/progs/netif_receive_skb.c | 249 +++++ 15 files changed, 1659 insertions(+), 117 deletions(-) create mode 100644 tools/testing/selftests/bpf/prog_tests/snprintf_btf.c create mode 100644 tools/testing/selftests/bpf/progs/bpf_iter_task_btf.c create mode 100644 tools/testing/selftests/bpf/progs/netif_receive_skb.c