kernel/linux.git/include/linux/bpf_local_storage.h, branch v6.19.11

bpf: Replace bpf memory allocator with kmalloc_nolock() in local storage

2025-11-19T00:20:25+00:00

Replace bpf memory allocator with kmalloc_nolock() to reduce memory wastage due to preallocation. In bpf_selem_free(), an selem now needs to wait for a RCU grace period before being freed when reuse_now == true. Therefore, rcu_barrier() should be always be called in bpf_local_storage_map_free(). In bpf_local_storage_free(), since smap->storage_ma is no longer needed to return the memory, the function is now independent from smap. Remove the outdated comment in bpf_local_storage_alloc(). We already free selem after an RCU grace period in bpf_local_storage_update() when bpf_local_storage_alloc() failed the cmpxchg since commit c0d63f309186 ("bpf: Add bpf_selem_free()"). Signed-off-by: Amery Hung Reviewed-by: Martin KaFai Lau Link: https://lore.kernel.org/r/20251114201329.3275875-5-ameryhung@gmail.com Signed-off-by: Alexei Starovoitov

bpf: Save memory alloction info in bpf_local_storage

2025-11-19T00:20:25+00:00

Save the memory allocation method used for bpf_local_storage in the struct explicitly so that we don't need to go through the hassle to find out the info. When a later patch replaces BPF memory allocator with kmalloc_noloc(), bpf_local_storage_free() will no longer need smap->storage_ma to return the memory and completely remove the dependency on smap in bpf_local_storage_free(). Signed-off-by: Amery Hung Reviewed-by: Martin KaFai Lau Link: https://lore.kernel.org/r/20251114201329.3275875-4-ameryhung@gmail.com Signed-off-by: Alexei Starovoitov

bpf: Remove smap argument from bpf_selem_free()

2025-11-19T00:20:25+00:00

Since selem already saves a pointer to smap, use it instead of an additional argument in bpf_selem_free(). This requires moving the SDATA(selem)->smap assignment from bpf_selem_link_map() to bpf_selem_alloc() since bpf_selem_free() may be called without the selem being linked to smap in bpf_local_storage_update(). Signed-off-by: Amery Hung Reviewed-by: Martin KaFai Lau Link: https://lore.kernel.org/r/20251114201329.3275875-3-ameryhung@gmail.com Signed-off-by: Alexei Starovoitov

bpf: Always charge/uncharge memory when allocating/unlinking storage elements

2025-11-19T00:20:25+00:00

Since commit a96a44aba556 ("bpf: bpf_sk_storage: Fix invalid wait context lockdep report"), {charge,uncharge}_mem are always true when allocating a bpf_local_storage_elem or unlinking a bpf_local_storage_elem from local storage, so drop these arguments. No functional change. Signed-off-by: Amery Hung Reviewed-by: Martin KaFai Lau Link: https://lore.kernel.org/r/20251114201329.3275875-2-ameryhung@gmail.com Signed-off-by: Alexei Starovoitov

bpf: Consistently use bpf_rcu_lock_held() everywhere

2025-10-15T10:26:12+00:00

We have many places which open-code what's now is bpf_rcu_lock_held() macro, so replace all those places with a clean and short macro invocation. For that, move bpf_rcu_lock_held() macro into include/linux/bpf.h. Signed-off-by: Andrii Nakryiko Signed-off-by: Daniel Borkmann Acked-by: Jiri Olsa Link: https://lore.kernel.org/bpf/20251014201403.4104511-1-andrii@kernel.org

bpf: Postpone bpf_selem_free() in bpf_selem_unlink_storage_nolock()

2024-10-24T17:25:59+00:00

In a later patch, bpf_selem_free() will call unpin_user_page() through bpf_obj_free_fields(). unpin_user_page() may take spin_lock. However, some bpf_selem_free() call paths have held a raw_spin_lock. Like this: raw_spin_lock_irqsave() bpf_selem_unlink_storage_nolock() bpf_selem_free() unpin_user_page() spin_lock() To avoid spinlock nested in raw_spinlock, bpf_selem_free() should be done after releasing the raw_spinlock. The "bool reuse_now" arg is replaced with "struct hlist_head *free_selem_list" in bpf_selem_unlink_storage_nolock(). The bpf_selem_unlink_storage_nolock() will append the to-be-free selem at the free_selem_list. The caller of bpf_selem_unlink_storage_nolock() will need to call the new bpf_selem_free_list(free_selem_list, reuse_now) to free the selem after releasing the raw_spinlock. Note that the selem->snode cannot be reused for linking to the free_selem_list because the selem->snode is protected by the raw_spinlock that we want to avoid holding. A new "struct hlist_node free_node;" is union-ized with the rcu_head. Only the first one successfully hlist_del_init_rcu(&selem->snode) will be able to use the free_node. After succeeding hlist_del_init_rcu(&selem->snode), the free_node and rcu_head usage is serialized such that they can share the 16 bytes in a union. Signed-off-by: Martin KaFai Lau Link: https://lore.kernel.org/r/20241023234759.860539-5-martin.lau@linux.dev Signed-off-by: Alexei Starovoitov

bpf: Add "bool swap_uptrs" arg to bpf_local_storage_update() and bpf_selem_alloc()

2024-10-24T17:25:59+00:00

In a later patch, the task local storage will only accept uptr from the syscall update_elem and will not accept uptr from the bpf prog. The reason is the bpf prog does not have a way to provide a valid user space address. bpf_local_storage_update() and bpf_selem_alloc() are used by both bpf prog bpf_task_storage_get(BPF_LOCAL_STORAGE_GET_F_CREATE) and bpf syscall update_elem. "bool swap_uptrs" arg is added to bpf_local_storage_update() and bpf_selem_alloc() to tell if it is called by the bpf prog or by the bpf syscall. When swap_uptrs==true, it is called by the syscall. The arg is named (swap_)uptrs because the later patch will swap the uptrs between the newly allocated selem and the user space provided map_value. It will make error handling easier in case map->ops->map_update_elem() fails and the caller can decide if it needs to unpin the uptr in the user space provided map_value or the bpf_local_storage_update() has already taken the uptr ownership and will take care of unpinning it also. Only swap_uptrs==false is passed now. The logic to handle the true case will be added in a later patch. Signed-off-by: Martin KaFai Lau Link: https://lore.kernel.org/r/20241023234759.860539-4-martin.lau@linux.dev Signed-off-by: Alexei Starovoitov

bpf: Allow compiler to inline most of bpf_local_storage_lookup()

2024-02-11T22:06:24+00:00

In various performance profiles of kernels with BPF programs attached, bpf_local_storage_lookup() appears as a significant portion of CPU cycles spent. To enable the compiler generate more optimal code, turn bpf_local_storage_lookup() into a static inline function, where only the cache insertion code path is outlined Notably, outlining cache insertion helps avoid bloating callers by duplicating setting up calls to raw_spin_{lock,unlock}_irqsave() (on architectures which do not inline spin_lock/unlock, such as x86), which would cause the compiler produce worse code by deciding to outline otherwise inlinable functions. The call overhead is neutral, because we make 2 calls either way: either calling raw_spin_lock_irqsave() and raw_spin_unlock_irqsave(); or call __bpf_local_storage_insert_cache(), which calls raw_spin_lock_irqsave(), followed by a tail-call to raw_spin_unlock_irqsave() where the compiler can perform TCO and (in optimized uninstrumented builds) turns it into a plain jump. The call to __bpf_local_storage_insert_cache() can be elided entirely if cacheit_lockit is a false constant expression. Based on results from './benchs/run_bench_local_storage.sh' (21 trials, reboot between each trial; x86 defconfig + BPF, clang 16) this produces improvements in throughput and latency in the majority of cases, with an average (geomean) improvement of 8%: +---- Hashmap Control -------------------- | | + num keys: 10 | : | | +-+ hashmap (control) sequential get +----------------------+---------------------- | +- hits throughput | 14.789 M ops/s | 14.745 M ops/s ( ~ ) | +- hits latency | 67.679 ns/op | 67.879 ns/op ( ~ ) | +- important_hits throughput | 14.789 M ops/s | 14.745 M ops/s ( ~ ) | | + num keys: 1000 | : | | +-+ hashmap (control) sequential get +----------------------+---------------------- | +- hits throughput | 12.233 M ops/s | 12.170 M ops/s ( ~ ) | +- hits latency | 81.754 ns/op | 82.185 ns/op ( ~ ) | +- important_hits throughput | 12.233 M ops/s | 12.170 M ops/s ( ~ ) | | + num keys: 10000 | : | | +-+ hashmap (control) sequential get +----------------------+---------------------- | +- hits throughput | 7.220 M ops/s | 7.204 M ops/s ( ~ ) | +- hits latency | 138.522 ns/op | 138.842 ns/op ( ~ ) | +- important_hits throughput | 7.220 M ops/s | 7.204 M ops/s ( ~ ) | | + num keys: 100000 | : | | +-+ hashmap (control) sequential get +----------------------+---------------------- | +- hits throughput | 5.061 M ops/s | 5.165 M ops/s (+2.1%) | +- hits latency | 198.483 ns/op | 194.270 ns/op (-2.1%) | +- important_hits throughput | 5.061 M ops/s | 5.165 M ops/s (+2.1%) | | + num keys: 4194304 | : | | +-+ hashmap (control) sequential get +----------------------+---------------------- | +- hits throughput | 2.864 M ops/s | 2.882 M ops/s ( ~ ) | +- hits latency | 365.220 ns/op | 361.418 ns/op (-1.0%) | +- important_hits throughput | 2.864 M ops/s | 2.882 M ops/s ( ~ ) | +---- Local Storage ---------------------- | | + num_maps: 1 | : | | +-+ local_storage cache sequential get +----------------------+---------------------- | +- hits throughput | 33.005 M ops/s | 39.068 M ops/s (+18.4%) | +- hits latency | 30.300 ns/op | 25.598 ns/op (-15.5%) | +- important_hits throughput | 33.005 M ops/s | 39.068 M ops/s (+18.4%) | : | : | | +-+ local_storage cache interleaved get +----------------------+---------------------- | +- hits throughput | 37.151 M ops/s | 44.926 M ops/s (+20.9%) | +- hits latency | 26.919 ns/op | 22.259 ns/op (-17.3%) | +- important_hits throughput | 37.151 M ops/s | 44.926 M ops/s (+20.9%) | | + num_maps: 10 | : | | +-+ local_storage cache sequential get +----------------------+---------------------- | +- hits throughput | 32.288 M ops/s | 38.099 M ops/s (+18.0%) | +- hits latency | 30.972 ns/op | 26.248 ns/op (-15.3%) | +- important_hits throughput | 3.229 M ops/s | 3.810 M ops/s (+18.0%) | : | : | | +-+ local_storage cache interleaved get +----------------------+---------------------- | +- hits throughput | 34.473 M ops/s | 41.145 M ops/s (+19.4%) | +- hits latency | 29.010 ns/op | 24.307 ns/op (-16.2%) | +- important_hits throughput | 12.312 M ops/s | 14.695 M ops/s (+19.4%) | | + num_maps: 16 | : | | +-+ local_storage cache sequential get +----------------------+---------------------- | +- hits throughput | 32.524 M ops/s | 38.341 M ops/s (+17.9%) | +- hits latency | 30.748 ns/op | 26.083 ns/op (-15.2%) | +- important_hits throughput | 2.033 M ops/s | 2.396 M ops/s (+17.9%) | : | : | | +-+ local_storage cache interleaved get +----------------------+---------------------- | +- hits throughput | 34.575 M ops/s | 41.338 M ops/s (+19.6%) | +- hits latency | 28.925 ns/op | 24.193 ns/op (-16.4%) | +- important_hits throughput | 11.001 M ops/s | 13.153 M ops/s (+19.6%) | | + num_maps: 17 | : | | +-+ local_storage cache sequential get +----------------------+---------------------- | +- hits throughput | 28.861 M ops/s | 32.756 M ops/s (+13.5%) | +- hits latency | 34.649 ns/op | 30.530 ns/op (-11.9%) | +- important_hits throughput | 1.700 M ops/s | 1.929 M ops/s (+13.5%) | : | : | | +-+ local_storage cache interleaved get +----------------------+---------------------- | +- hits throughput | 31.529 M ops/s | 36.110 M ops/s (+14.5%) | +- hits latency | 31.719 ns/op | 27.697 ns/op (-12.7%) | +- important_hits throughput | 9.598 M ops/s | 10.993 M ops/s (+14.5%) | | + num_maps: 24 | : | | +-+ local_storage cache sequential get +----------------------+---------------------- | +- hits throughput | 18.602 M ops/s | 19.937 M ops/s (+7.2%) | +- hits latency | 53.767 ns/op | 50.166 ns/op (-6.7%) | +- important_hits throughput | 0.776 M ops/s | 0.831 M ops/s (+7.2%) | : | : | | +-+ local_storage cache interleaved get +----------------------+---------------------- | +- hits throughput | 21.718 M ops/s | 23.332 M ops/s (+7.4%) | +- hits latency | 46.047 ns/op | 42.865 ns/op (-6.9%) | +- important_hits throughput | 6.110 M ops/s | 6.564 M ops/s (+7.4%) | | + num_maps: 32 | : | | +-+ local_storage cache sequential get +----------------------+---------------------- | +- hits throughput | 14.118 M ops/s | 14.626 M ops/s (+3.6%) | +- hits latency | 70.856 ns/op | 68.381 ns/op (-3.5%) | +- important_hits throughput | 0.442 M ops/s | 0.458 M ops/s (+3.6%) | : | : | | +-+ local_storage cache interleaved get +----------------------+---------------------- | +- hits throughput | 17.111 M ops/s | 17.906 M ops/s (+4.6%) | +- hits latency | 58.451 ns/op | 55.865 ns/op (-4.4%) | +- important_hits throughput | 4.776 M ops/s | 4.998 M ops/s (+4.6%) | | + num_maps: 100 | : | | +-+ local_storage cache sequential get +----------------------+---------------------- | +- hits throughput | 5.281 M ops/s | 5.528 M ops/s (+4.7%) | +- hits latency | 192.398 ns/op | 183.059 ns/op (-4.9%) | +- important_hits throughput | 0.053 M ops/s | 0.055 M ops/s (+4.9%) | : | : | | +-+ local_storage cache interleaved get +----------------------+---------------------- | +- hits throughput | 6.265 M ops/s | 6.498 M ops/s (+3.7%) | +- hits latency | 161.436 ns/op | 152.877 ns/op (-5.3%) | +- important_hits throughput | 1.636 M ops/s | 1.697 M ops/s (+3.7%) | | + num_maps: 1000 | : | | +-+ local_storage cache sequential get +----------------------+---------------------- | +- hits throughput | 0.355 M ops/s | 0.354 M ops/s ( ~ ) | +- hits latency | 2826.538 ns/op | 2827.139 ns/op ( ~ ) | +- important_hits throughput | 0.000 M ops/s | 0.000 M ops/s ( ~ ) | : | : | | +-+ local_storage cache interleaved get +----------------------+---------------------- | +- hits throughput | 0.404 M ops/s | 0.403 M ops/s ( ~ ) | +- hits latency | 2481.190 ns/op | 2487.555 ns/op ( ~ ) | +- important_hits throughput | 0.102 M ops/s | 0.101 M ops/s ( ~ ) The on_lookup test in {cgrp,task}_ls_recursion.c is removed because the bpf_local_storage_lookup is no longer traceable and adding tracepoint will make the compiler generate worse code: https://lore.kernel.org/bpf/ZcJmok64Xqv6l4ZS@elver.google.com/ Signed-off-by: Marco Elver Cc: Martin KaFai Lau Acked-by: Yonghong Song Link: https://lore.kernel.org/r/20240207122626.3508658-1-elver@google.com Signed-off-by: Martin KaFai Lau

bpf: Use bpf_mem_cache_alloc/free for bpf_local_storage

2023-03-26T02:52:52+00:00

This patch uses bpf_mem_cache_alloc/free for allocating and freeing bpf_local_storage for task and cgroup storage. The changes are similar to the previous patch. A few things that worth to mention for bpf_local_storage: The local_storage is freed when the last selem is deleted. Before deleting a selem from local_storage, it needs to retrieve the local_storage->smap because the bpf_selem_unlink_storage_nolock() may have set it to NULL. Note that local_storage->smap may have already been NULL when the selem created this local_storage has been removed. In this case, call_rcu will be used to free the local_storage. Also, the bpf_ma (true or false) value is needed before calling bpf_local_storage_free(). The bpf_ma can either be obtained from the local_storage->smap (if available) or any of its selem's smap. A new helper check_storage_bpf_ma() is added to obtain bpf_ma for a deleting bpf_local_storage. When bpf_local_storage_alloc getting a reused memory, all fields are either in the correct values or will be initialized. 'cache[]' must already be all NULLs. 'list' must be empty. Others will be initialized. Cc: Namhyung Kim Signed-off-by: Martin KaFai Lau Link: https://lore.kernel.org/r/20230322215246.1675516-4-martin.lau@linux.dev Signed-off-by: Alexei Starovoitov

bpf: Use bpf_mem_cache_alloc/free in bpf_local_storage_elem

2023-03-26T02:52:52+00:00

This patch uses bpf_mem_alloc for the task and cgroup local storage that the bpf prog can easily get a hold of the storage owner's PTR_TO_BTF_ID. eg. bpf_get_current_task_btf() can be used in some of the kmalloc code path which will cause deadlock/recursion. bpf_mem_cache_alloc is deadlock free and will solve a legit use case in [1]. For sk storage, its batch creation benchmark shows a few percent regression when the sk create/destroy batch size is larger than 32. The sk creation/destruction happens much more often and depends on external traffic. Considering it is hypothetical to be able to cause deadlock with sk storage, it can cross the bridge to use bpf_mem_alloc till a legit (ie. useful) use case comes up. For inode storage, bpf_local_storage_destroy() is called before waiting for a rcu gp and its memory cannot be reused immediately. inode stays with kmalloc/kfree after the rcu [or tasks_trace] gp. A 'bool bpf_ma' argument is added to bpf_local_storage_map_alloc(). Only task and cgroup storage have 'bpf_ma == true' which means to use bpf_mem_cache_alloc/free(). This patch only changes selem to use bpf_mem_alloc for task and cgroup. The next patch will change the local_storage to use bpf_mem_alloc also for task and cgroup. Here is some more details on the changes: * memory allocation: After bpf_mem_cache_alloc(), the SDATA(selem)->data is zero-ed because bpf_mem_cache_alloc() could return a reused selem. It is to keep the existing bpf_map_kzalloc() behavior. Only SDATA(selem)->data is zero-ed. SDATA(selem)->data is the visible part to the bpf prog. No need to use zero_map_value() to do the zeroing because bpf_selem_free(..., reuse_now = true) ensures no bpf prog is using the selem before returning the selem through bpf_mem_cache_free(). For the internal fields of selem, they will be initialized when linking to the new smap and the new local_storage. When 'bpf_ma == false', nothing changes in this patch. It will stay with the bpf_map_kzalloc(). * memory free: The bpf_selem_free() and bpf_selem_free_rcu() are modified to handle the bpf_ma == true case. For the common selem free path where its owner is also being destroyed, the mem is freed in bpf_local_storage_destroy(), the owner (task and cgroup) has gone through a rcu gp. The memory can be reused immediately, so bpf_local_storage_destroy() will call bpf_selem_free(..., reuse_now = true) which will do bpf_mem_cache_free() for immediate reuse consideration. An exception is the delete elem code path. The delete elem code path is called from the helper bpf_*_storage_delete() and the syscall bpf_map_delete_elem(). This path is an unusual case for local storage because the common use case is to have the local storage staying with its owner life time so that the bpf prog and the user space does not have to monitor the owner's destruction. For the delete elem path, the selem cannot be reused immediately because there could be bpf prog using it. It will call bpf_selem_free(..., reuse_now = false) and it will wait for a rcu tasks trace gp before freeing the elem. The rcu callback is changed to do bpf_mem_cache_raw_free() instead of kfree(). When 'bpf_ma == false', it should be the same as before. __bpf_selem_free() is added to do the kfree_rcu and call_tasks_trace_rcu(). A few words on the 'reuse_now == true'. When 'reuse_now == true', it is still racing with bpf_local_storage_map_free which is under rcu protection, so it still needs to wait for a rcu gp instead of kfree(). Otherwise, the selem may be reused by slab for a totally different struct while the bpf_local_storage_map_free() is still using it (as a rcu reader). For the inode case, there may be other rcu readers also. In short, when bpf_ma == false and reuse_now == true => vanilla rcu. [1]: https://lore.kernel.org/bpf/20221118190109.1512674-1-namhyung@kernel.org/ Cc: Namhyung Kim Signed-off-by: Martin KaFai Lau Link: https://lore.kernel.org/r/20230322215246.1675516-3-martin.lau@linux.dev Signed-off-by: Alexei Starovoitov