Daniel Borkmann <daniel@xxxxxxxxxxxxx> writes: > On 6/17/21 11:27 PM, Toke Høiland-Jørgensen wrote: >> XDP_REDIRECT works by a three-step process: the bpf_redirect() and >> bpf_redirect_map() helpers will lookup the target of the redirect and store >> it (along with some other metadata) in a per-CPU struct bpf_redirect_info. >> Next, when the program returns the XDP_REDIRECT return code, the driver >> will call xdp_do_redirect() which will use the information thus stored to >> actually enqueue the frame into a bulk queue structure (that differs >> slightly by map type, but shares the same principle). Finally, before >> exiting its NAPI poll loop, the driver will call xdp_do_flush(), which will >> flush all the different bulk queues, thus completing the redirect. >> >> Pointers to the map entries will be kept around for this whole sequence of >> steps, protected by RCU. However, there is no top-level rcu_read_lock() in >> the core code; instead drivers add their own rcu_read_lock() around the XDP >> portions of the code, but somewhat inconsistently as Martin discovered[0]. >> However, things still work because everything happens inside a single NAPI >> poll sequence, which means it's between a pair of calls to >> local_bh_disable()/local_bh_enable(). So Paul suggested[1] that we could >> document this intention by using rcu_dereference_check() with >> rcu_read_lock_bh_held() as a second parameter, thus allowing sparse and >> lockdep to verify that everything is done correctly. >> >> This patch does just that: we add an __rcu annotation to the map entry >> pointers and remove the various comments explaining the NAPI poll assurance >> strewn through devmap.c in favour of a longer explanation in filter.c. The >> goal is to have one coherent documentation of the entire flow, and rely on >> the RCU annotations as a "standard" way of communicating the flow in the >> map code (which can additionally be understood by sparse and lockdep). >> >> The RCU annotation replacements result in a fairly straight-forward >> replacement where READ_ONCE() becomes rcu_dereference_check(), WRITE_ONCE() >> becomes rcu_assign_pointer() and xchg() and cmpxchg() gets wrapped in the >> proper constructs to cast the pointer back and forth between __rcu and >> __kernel address space (for the benefit of sparse). The one complication is >> that xskmap has a few constructions where double-pointers are passed back >> and forth; these simply all gain __rcu annotations, and only the final >> reference/dereference to the inner-most pointer gets changed. >> >> With this, everything can be run through sparse without eliciting >> complaints, and lockdep can verify correctness even without the use of >> rcu_read_lock() in the drivers. Subsequent patches will clean these up from >> the drivers. >> >> [0] https://lore.kernel.org/bpf/20210415173551.7ma4slcbqeyiba2r@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxx/ >> [1] https://lore.kernel.org/bpf/20210419165837.GA975577@paulmck-ThinkPad-P17-Gen-1/ >> >> Signed-off-by: Toke Høiland-Jørgensen <toke@xxxxxxxxxx> >> --- >> include/net/xdp_sock.h | 2 +- >> kernel/bpf/cpumap.c | 13 +++++++---- >> kernel/bpf/devmap.c | 49 ++++++++++++++++++------------------------ >> net/core/filter.c | 28 ++++++++++++++++++++++++ >> net/xdp/xsk.c | 4 ++-- >> net/xdp/xsk.h | 4 ++-- >> net/xdp/xskmap.c | 29 ++++++++++++++----------- >> 7 files changed, 80 insertions(+), 49 deletions(-) > [...] >> __dev_map_entry_free); >> diff --git a/net/core/filter.c b/net/core/filter.c >> index caa88955562e..0b7db5c70385 100644 >> --- a/net/core/filter.c >> +++ b/net/core/filter.c >> @@ -3922,6 +3922,34 @@ static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = { >> .arg2_type = ARG_ANYTHING, >> }; >> >> +/* XDP_REDIRECT works by a three-step process, implemented in the functions >> + * below: >> + * >> + * 1. The bpf_redirect() and bpf_redirect_map() helpers will lookup the target >> + * of the redirect and store it (along with some other metadata) in a per-CPU >> + * struct bpf_redirect_info. >> + * >> + * 2. When the program returns the XDP_REDIRECT return code, the driver will >> + * call xdp_do_redirect() which will use the information in struct >> + * bpf_redirect_info to actually enqueue the frame into a map type-specific >> + * bulk queue structure. >> + * >> + * 3. Before exiting its NAPI poll loop, the driver will call xdp_do_flush(), >> + * which will flush all the different bulk queues, thus completing the >> + * redirect. >> + * >> + * Pointers to the map entries will be kept around for this whole sequence of >> + * steps, protected by RCU. However, there is no top-level rcu_read_lock() in >> + * the core code; instead, the RCU protection relies on everything happening >> + * inside a single NAPI poll sequence, which means it's between a pair of calls >> + * to local_bh_disable()/local_bh_enable(). >> + * >> + * The map entries are marked as __rcu and the map code makes sure to >> + * dereference those pointers with rcu_dereference_check() in a way that works >> + * for both sections that to hold an rcu_read_lock() and sections that are >> + * called from NAPI without a separate rcu_read_lock(). The code below does not >> + * use RCU annotations, but relies on those in the map code. > > One more follow-up question related to tc BPF: given we do use rcu_read_lock_bh() > in case of sch_handle_egress(), could we also remove the rcu_read_lock() pair > from cls_bpf_classify() then? I believe so, yeah. Patch 2 in this series should even make lockdep stop complaining about it :) I can add a patch removing the rcu_read_lock() from cls_bpf in the next version. > It would also be great if this scenario in general could be placed > under the Documentation/RCU/whatisRCU.rst as an example, so we could > refer to the official doc on this, too, if Paul is good with this. I'll take a look and see if I can find a way to fit it in there... > Could you also update the RCU comment in bpf_prog_run_xdp()? Or > alternatively move all the below driver comments in there as a single > location? > > /* This code is invoked within a single NAPI poll cycle and thus under > * local_bh_disable(), which provides the needed RCU protection. > */ Sure, can do. And yeah, I do agree that moving the comment in there makes more sense than scattering it over all the drivers, even if that means I have to go back and edit all the drivers again :P -Toke
