bpf_mem_alloc() doesn't support zero-sized allocation, so removing these
tests from test_bpf_ma test. After the removal, there will no definition
for bin_data_8, so remove 8 from data_sizes array and adjust the index
of data_btf_ids array in all test cases accordingly.
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
At present, bpf memory allocator uses check_obj_size() to ensure that
ksize() of allocated pointer is equal with the unit_size of used
bpf_mem_cache. Its purpose is to prevent bpf_mem_free() from selecting
a bpf_mem_cache which has different unit_size compared with the
bpf_mem_cache used for allocation. But as reported by lkp, the return
value of ksize() or kmalloc_size_roundup() may change due to slab merge
and it will lead to the warning report in check_obj_size().
The reported warning happened as follows:
(1) in bpf_mem_cache_adjust_size(), kmalloc_size_roundup(96) returns the
object_size of kmalloc-96 instead of kmalloc-cg-96. The object_size of
kmalloc-96 is 96, so size_index for 96 is not adjusted accordingly.
(2) the object_size of kmalloc-cg-96 is adjust from 96 to 128 due to
slab merge in __kmem_cache_alias(). For SLAB, SLAB_HWCACHE_ALIGN is
enabled by default for kmalloc slab, so align is 64 and size is 128 for
kmalloc-cg-96. SLUB has a similar merge logic, but its object_size will
not be changed, because its align is 8 under x86-64.
(3) when unit_alloc() does kmalloc_node(96, __GFP_ACCOUNT, node),
ksize() returns 128 instead of 96 for the returned pointer.
(4) the warning in check_obj_size() is triggered.
Considering the slab merge can happen in anytime (e.g, a slab created in
a new module), the following case is also possible: during the
initialization of bpf_global_ma, there is no slab merge and ksize() for
a 96-bytes object returns 96. But after that a new slab created by a
kernel module is merged to kmalloc-cg-96 and the object_size of
kmalloc-cg-96 is adjust from 96 to 128 (which is possible for x86-64 +
CONFIG_SLAB, because its alignment requirement is 64 for 96-bytes slab).
So soon or later, when bpf_global_ma frees a 96-byte-sized pointer
which is allocated from bpf_mem_cache with unit_size=96, bpf_mem_free()
will free the pointer through a bpf_mem_cache in which unit_size is 128,
because the return value of ksize() changes. The warning for the
mismatch will be triggered again.
A feasible fix is introducing similar APIs compared with ksize() and
kmalloc_size_roundup() to return the actually-allocated size instead of
size which may change due to slab merge, but it will introduce
unnecessary dependency on the implementation details of mm subsystem.
As for now the pointer of bpf_mem_cache is saved in the 8-bytes area
(or 4-bytes under 32-bit host) above the returned pointer, using
unit_size in the saved bpf_mem_cache to select the target cache instead
of inferring the size from the pointer itself. Beside no extra
dependency on mm subsystem, the performance for bpf_mem_free_rcu() is
also improved as shown below.
Before applying the patch, the performances of bpf_mem_alloc() and
bpf_mem_free_rcu() on 8-CPUs VM with one producer are as follows:
kmalloc : alloc 11.69 ± 0.28M/s free 29.58 ± 0.93M/s
percpu : alloc 14.11 ± 0.52M/s free 14.29 ± 0.99M/s
After apply the patch, the performance for bpf_mem_free_rcu() increases
9% and 146% for kmalloc memory and per-cpu memory respectively:
kmalloc: alloc 11.01 ± 0.03M/s free 32.42 ± 0.48M/s
percpu: alloc 12.84 ± 0.12M/s free 35.24 ± 0.23M/s
After the fixes, there is no need to adjust size_index to fix the
mismatch between allocation and free, so remove it as well. Also return
NULL instead of ZERO_SIZE_PTR for zero-sized alloc in bpf_mem_alloc(),
because there is no bpf_mem_cache pointer saved above ZERO_SIZE_PTR.
Fixes: 9077fc228f ("bpf: Use kmalloc_size_roundup() to adjust size_index")
Reported-by: kernel test robot <[email protected]>
Closes: https://lore.kernel.org/bpf/[email protected]
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
When there is bpf_list_head or bpf_rb_root field in map value, the free
of map btf and the free of map value may run concurrently and there may
be use-after-free problem, so add two test cases to demonstrate it. And
the use-after-free problem can been easily reproduced by using bpf_next
tree and a KASAN-enabled kernel.
The first test case tests the racing between the free of map btf and the
free of array map. It constructs the racing by releasing the array map in
the end after other ref-counter of map btf has been released. To delay
the free of array map and make it be invoked after btf_free_rcu() is
invoked, it stresses system_unbound_wq by closing multiple percpu array
maps before it closes the array map.
The second case tests the racing between the free of map btf and the
free of inner map. Beside using the similar method as the first one
does, it uses bpf_map_delete_elem() to delete the inner map and to defer
the release of inner map after one RCU grace period.
The reason for using two skeletons is to prevent the release of outer
map and inner map in map_in_map_btf.c interfering the release of bpf
map in normal_map_btf.c.
Signed-off-by: Hou Tao <[email protected]>
Signed-off-by: Daniel Borkmann <[email protected]>
Acked-by: Yonghong Song <[email protected]>
Link: https://lore.kernel.org/bpf/[email protected]
If an abnormally huge cnt is used for multi-kprobes attachment, the
following warning will be reported:
------------[ cut here ]------------
WARNING: CPU: 1 PID: 392 at mm/util.c:632 kvmalloc_node+0xd9/0xe0
Modules linked in: bpf_testmod(O)
CPU: 1 PID: 392 Comm: test_progs Tainted: G ...... 6.7.0-rc3+ #32
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996)
......
RIP: 0010:kvmalloc_node+0xd9/0xe0
? __warn+0x89/0x150
? kvmalloc_node+0xd9/0xe0
bpf_kprobe_multi_link_attach+0x87/0x670
__sys_bpf+0x2a28/0x2bc0
__x64_sys_bpf+0x1a/0x30
do_syscall_64+0x36/0xb0
entry_SYSCALL_64_after_hwframe+0x6e/0x76
RIP: 0033:0x7fbe067f0e0d
......
</TASK>
---[ end trace 0000000000000000 ]---
So add a test to ensure the warning is fixed.
Signed-off-by: Hou Tao <[email protected]>
Signed-off-by: Daniel Borkmann <[email protected]>
Acked-by: Jiri Olsa <[email protected]>
Acked-by: Andrii Nakryiko <[email protected]>
Link: https://lore.kernel.org/bpf/[email protected]
Since libbpf v1.0, libbpf doesn't return error code embedded into the
pointer iteself, libbpf_get_error() is deprecated and it is basically
the same as using -errno directly.
So replace the invocations of libbpf_get_error() by -errno in
kprobe_multi_test. For libbpf_get_error() in test_attach_api_fails(),
saving -errno before invoking ASSERT_xx() macros just in case that
errno is overwritten by these macros. However, the invocation of
libbpf_get_error() in get_syms() should be kept intact, because
hashmap__new() still returns a pointer with embedded error code.
Signed-off-by: Hou Tao <[email protected]>
Signed-off-by: Daniel Borkmann <[email protected]>
Acked-by: Andrii Nakryiko <[email protected]>
Link: https://lore.kernel.org/bpf/[email protected]
If an abnormally huge cnt is used for multi-uprobes attachment, the
following warning will be reported:
------------[ cut here ]------------
WARNING: CPU: 7 PID: 406 at mm/util.c:632 kvmalloc_node+0xd9/0xe0
Modules linked in: bpf_testmod(O)
CPU: 7 PID: 406 Comm: test_progs Tainted: G ...... 6.7.0-rc3+ #32
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996) ......
RIP: 0010:kvmalloc_node+0xd9/0xe0
......
Call Trace:
<TASK>
? __warn+0x89/0x150
? kvmalloc_node+0xd9/0xe0
bpf_uprobe_multi_link_attach+0x14a/0x480
__sys_bpf+0x14a9/0x2bc0
do_syscall_64+0x36/0xb0
entry_SYSCALL_64_after_hwframe+0x6e/0x76
......
</TASK>
---[ end trace 0000000000000000 ]---
So add a test to ensure the warning is fixed.
Signed-off-by: Hou Tao <[email protected]>
Signed-off-by: Daniel Borkmann <[email protected]>
Acked-by: Jiri Olsa <[email protected]>
Acked-by: Andrii Nakryiko <[email protected]>
Link: https://lore.kernel.org/bpf/[email protected]
An abnormally big cnt may also be assigned to kprobe_multi.cnt when
attaching multiple kprobes. It will trigger the following warning in
kvmalloc_node():
if (unlikely(size > INT_MAX)) {
WARN_ON_ONCE(!(flags & __GFP_NOWARN));
return NULL;
}
Fix the warning by limiting the maximal number of kprobes in
bpf_kprobe_multi_link_attach(). If the number of kprobes is greater than
MAX_KPROBE_MULTI_CNT, the attachment will fail and return -E2BIG.
Fixes: 0dcac27254 ("bpf: Add multi kprobe link")
Signed-off-by: Hou Tao <[email protected]>
Signed-off-by: Daniel Borkmann <[email protected]>
Acked-by: Jiri Olsa <[email protected]>
Acked-by: Andrii Nakryiko <[email protected]>
Link: https://lore.kernel.org/bpf/[email protected]
rcu_read_lock() is no longer held when invoking bpf_event_entry_gen()
which is called by perf_event_fd_array_get_ptr(), so using GFP_KERNEL
instead of GFP_ATOMIC to reduce the possibility of failures due to
out-of-memory.
Acked-by: Yonghong Song <[email protected]>
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
There is no rcu-read-lock requirement for ops->map_fd_get_ptr() or
ops->map_fd_put_ptr(), so doesn't use rcu-read-lock for these two
callbacks.
For bpf_fd_array_map_update_elem(), accessing array->ptrs doesn't need
rcu-read-lock because array->ptrs must still be allocated. For
bpf_fd_htab_map_update_elem(), htab_map_update_elem() only requires
rcu-read-lock to be held to avoid the WARN_ON_ONCE(), so only use
rcu_read_lock() during the invocation of htab_map_update_elem().
Acked-by: Yonghong Song <[email protected]>
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
Since commit 638e4b825d ("bpf: Allows per-cpu maps and map-in-map in
sleepable programs"), sleepable BPF program can also use map-in-map, but
maybe_wait_bpf_programs() doesn't handle it accordingly. The main reason
is that using synchronize_rcu_tasks_trace() to wait for the completions
of these sleepable BPF programs may incur a very long delay and
userspace may think it is hung, so the wait for sleepable BPF programs
is skipped. Update the comments in maybe_wait_bpf_programs() to reflect
the reason.
Signed-off-by: Hou Tao <[email protected]>
Acked-by: Yonghong Song <[email protected]>
Acked-by: John Fastabend <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
generic_map_{delete,update}_batch() doesn't set uattr->batch.count as
zero before it tries to allocate memory for key. If the memory
allocation fails, the value of uattr->batch.count will be incorrect.
Fix it by setting uattr->batch.count as zero beore batched update or
deletion.
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
There is no need to call maybe_wait_bpf_programs() if update or deletion
operation fails. So only call maybe_wait_bpf_programs() if update or
deletion operation succeeds.
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
When doing batched lookup and deletion operations on htab of maps,
maybe_wait_bpf_programs() is needed to ensure all programs don't use the
inner map after the bpf syscall returns.
Instead of adding the wait in __htab_map_lookup_and_delete_batch(),
adding the wait in bpf_map_do_batch() and also removing the calling of
maybe_wait_bpf_programs() from generic_map_{delete,update}_batch().
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
Just like commit 9087c6ff8d ("bpf: Call maybe_wait_bpf_programs() only
once from generic_map_delete_batch()"), there is also no need to call
maybe_wait_bpf_programs() for each update in batched update, so only
call it once in generic_map_update_batch().
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
Both map_lookup_elem() and generic_map_lookup_batch() use
bpf_map_copy_value() to lookup and copy the value, and there is no
update operation in bpf_map_copy_value(), so just remove the invocation
of maybe_wait_bpf_programs() from it.
Fixes: 15c14a3dca ("bpf: Add bpf_map_{value_size, update_value, map_copy_value} functions")
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
Syscall program is running with rcu_read_lock_trace being held, so if
bpf_map_update_elem() or bpf_map_delete_elem() invokes
synchronize_rcu_tasks_trace() when operating on an outer map, there will
be dead-lock, so add a test to guarantee that it is dead-lock free.
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
Add test cases to test the race between the destroy of inner map due to
map-in-map update and the access of inner map in bpf program. The
following 4 combinations are added:
(1) array map in map array + bpf program
(2) array map in map array + sleepable bpf program
(3) array map in map htab + bpf program
(4) array map in map htab + sleepable bpf program
Before applying the fixes, when running `./test_prog -a map_in_map`, the
following error was reported:
==================================================================
BUG: KASAN: slab-use-after-free in array_map_update_elem+0x48/0x3e0
Read of size 4 at addr ffff888114f33824 by task test_progs/1858
CPU: 1 PID: 1858 Comm: test_progs Tainted: G O 6.6.0+ #7
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996) ......
Call Trace:
<TASK>
dump_stack_lvl+0x4a/0x90
print_report+0xd2/0x620
kasan_report+0xd1/0x110
__asan_load4+0x81/0xa0
array_map_update_elem+0x48/0x3e0
bpf_prog_be94a9f26772f5b7_access_map_in_array+0xe6/0xf6
trace_call_bpf+0x1aa/0x580
kprobe_perf_func+0xdd/0x430
kprobe_dispatcher+0xa0/0xb0
kprobe_ftrace_handler+0x18b/0x2e0
0xffffffffc02280f7
RIP: 0010:__x64_sys_getpgid+0x1/0x30
......
</TASK>
Allocated by task 1857:
kasan_save_stack+0x26/0x50
kasan_set_track+0x25/0x40
kasan_save_alloc_info+0x1e/0x30
__kasan_kmalloc+0x98/0xa0
__kmalloc_node+0x6a/0x150
__bpf_map_area_alloc+0x141/0x170
bpf_map_area_alloc+0x10/0x20
array_map_alloc+0x11f/0x310
map_create+0x28a/0xb40
__sys_bpf+0x753/0x37c0
__x64_sys_bpf+0x44/0x60
do_syscall_64+0x36/0xb0
entry_SYSCALL_64_after_hwframe+0x6e/0x76
Freed by task 11:
kasan_save_stack+0x26/0x50
kasan_set_track+0x25/0x40
kasan_save_free_info+0x2b/0x50
__kasan_slab_free+0x113/0x190
slab_free_freelist_hook+0xd7/0x1e0
__kmem_cache_free+0x170/0x260
kfree+0x9b/0x160
kvfree+0x2d/0x40
bpf_map_area_free+0xe/0x20
array_map_free+0x120/0x2c0
bpf_map_free_deferred+0xd7/0x1e0
process_one_work+0x462/0x990
worker_thread+0x370/0x670
kthread+0x1b0/0x200
ret_from_fork+0x3a/0x70
ret_from_fork_asm+0x1b/0x30
Last potentially related work creation:
kasan_save_stack+0x26/0x50
__kasan_record_aux_stack+0x94/0xb0
kasan_record_aux_stack_noalloc+0xb/0x20
__queue_work+0x331/0x950
queue_work_on+0x75/0x80
bpf_map_put+0xfa/0x160
bpf_map_fd_put_ptr+0xe/0x20
bpf_fd_array_map_update_elem+0x174/0x1b0
bpf_map_update_value+0x2b7/0x4a0
__sys_bpf+0x2551/0x37c0
__x64_sys_bpf+0x44/0x60
do_syscall_64+0x36/0xb0
entry_SYSCALL_64_after_hwframe+0x6e/0x76
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
When removing the inner map from the outer map, the inner map will be
freed after one RCU grace period and one RCU tasks trace grace
period, so it is certain that the bpf program, which may access the
inner map, has exited before the inner map is freed.
However there is no need to wait for one RCU tasks trace grace period if
the outer map is only accessed by non-sleepable program. So adding
sleepable_refcnt in bpf_map and increasing sleepable_refcnt when adding
the outer map into env->used_maps for sleepable program. Although the
max number of bpf program is INT_MAX - 1, the number of bpf programs
which are being loaded may be greater than INT_MAX, so using atomic64_t
instead of atomic_t for sleepable_refcnt. When removing the inner map
from the outer map, using sleepable_refcnt to decide whether or not a
RCU tasks trace grace period is needed before freeing the inner map.
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
When updating or deleting an inner map in map array or map htab, the map
may still be accessed by non-sleepable program or sleepable program.
However bpf_map_fd_put_ptr() decreases the ref-counter of the inner map
directly through bpf_map_put(), if the ref-counter is the last one
(which is true for most cases), the inner map will be freed by
ops->map_free() in a kworker. But for now, most .map_free() callbacks
don't use synchronize_rcu() or its variants to wait for the elapse of a
RCU grace period, so after the invocation of ops->map_free completes,
the bpf program which is accessing the inner map may incur
use-after-free problem.
Fix the free of inner map by invoking bpf_map_free_deferred() after both
one RCU grace period and one tasks trace RCU grace period if the inner
map has been removed from the outer map before. The deferment is
accomplished by using call_rcu() or call_rcu_tasks_trace() when
releasing the last ref-counter of bpf map. The newly-added rcu_head
field in bpf_map shares the same storage space with work field to
reduce the size of bpf_map.
Fixes: bba1dc0b55 ("bpf: Remove redundant synchronize_rcu.")
Fixes: 638e4b825d ("bpf: Allows per-cpu maps and map-in-map in sleepable programs")
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
Both map deletion operation, map release and map free operation use
fd_array_map_delete_elem() to remove the element from fd array and
need_defer is always true in fd_array_map_delete_elem(). For the map
deletion operation and map release operation, need_defer=true is
necessary, because the bpf program, which accesses the element in fd
array, may still alive. However for map free operation, it is certain
that the bpf program which owns the fd array has already been exited, so
setting need_defer as false is appropriate for map free operation.
So fix it by adding need_defer parameter to bpf_fd_array_map_clear() and
adding a new helper __fd_array_map_delete_elem() to handle the map
deletion, map release and map free operations correspondingly.
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
map is the pointer of outer map, and need_defer needs some explanation.
need_defer tells the implementation to defer the reference release of
the passed element and ensure that the element is still alive before
the bpf program, which may manipulate it, exits.
The following three cases will invoke map_fd_put_ptr() and different
need_defer values will be passed to these callers:
1) release the reference of the old element in the map during map update
or map deletion. The release must be deferred, otherwise the bpf
program may incur use-after-free problem, so need_defer needs to be
true.
2) release the reference of the to-be-added element in the error path of
map update. The to-be-added element is not visible to any bpf
program, so it is OK to pass false for need_defer parameter.
3) release the references of all elements in the map during map release.
Any bpf program which has access to the map must have been exited and
released, so need_defer=false will be OK.
These two parameters will be used by the following patches to fix the
potential use-after-free problem for map-in-map.
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
bpf_mem_cache_alloc_flags() may call __alloc() directly when there is no
free object in free list, but it doesn't initialize the allocation hint
for the returned pointer. It may lead to bad memory dereference when
freeing the pointer, so fix it by initializing the allocation hint.
Fixes: 822fb26bdb ("bpf: Add a hint to allocated objects.")
Signed-off-by: Hou Tao <[email protected]>
Acked-by: Yonghong Song <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
Export map_update_retriable() to make it usable for other map_test
cases. These cases may only need retry for specific errno, so add
a new callback parameter to let map_update_retriable() decide whether or
not the errno is retriable.
Signed-off-by: Hou Tao <[email protected]>
Signed-off-by: Andrii Nakryiko <[email protected]>
Link: https://lore.kernel.org/bpf/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
When updating per-cpu map in map_percpu_stats test, patch_map_thread()
only passes 4-bytes-sized value to bpf_map_update_elem(). The expected
size of the value is 8 * num_possible_cpus(), so fix it by passing a
value with enough-size for per-cpu map update.
Signed-off-by: Hou Tao <[email protected]>
Signed-off-by: Andrii Nakryiko <[email protected]>
Link: https://lore.kernel.org/bpf/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
When there are concurrent uref release and bpf timer init operations,
the following sequence diagram is possible. It will break the guarantee
provided by bpf_timer: bpf_timer will still be alive after userspace
application releases or unpins the map. It also will lead to kmemleak
for old kernel version which doesn't release bpf_timer when map is
released.
bpf program X:
bpf_timer_init()
lock timer->lock
read timer->timer as NULL
read map->usercnt != 0
process Y:
close(map_fd)
// put last uref
bpf_map_put_uref()
atomic_dec_and_test(map->usercnt)
array_map_free_timers()
bpf_timer_cancel_and_free()
// just return
read timer->timer is NULL
t = bpf_map_kmalloc_node()
timer->timer = t
unlock timer->lock
Fix the problem by checking map->usercnt after timer->timer is assigned,
so when there are concurrent uref release and bpf timer init, either
bpf_timer_cancel_and_free() from uref release reads a no-NULL timer
or the newly-added atomic64_read() returns a zero usercnt.
Because atomic_dec_and_test(map->usercnt) and READ_ONCE(timer->timer)
in bpf_timer_cancel_and_free() are not protected by a lock, so add
a memory barrier to guarantee the order between map->usercnt and
timer->timer. Also use WRITE_ONCE(timer->timer, x) to match the lockless
read of timer->timer in bpf_timer_cancel_and_free().
Reported-by: Hsin-Wei Hung <[email protected]>
Closes: https://lore.kernel.org/bpf/CABcoxUaT2k9hWsS1tNgXyoU3E-=PuOgMn737qK984fbFmfYixQ@mail.gmail.com
Fixes: b00628b1c7 ("bpf: Introduce bpf timers.")
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
There are two possible mismatched alloc and free cases in BPF memory
allocator:
1) allocate from cache X but free by cache Y with a different unit_size
2) allocate from per-cpu cache but free by kmalloc cache or vice versa
So add more WARN_ON_ONCE checks in free_bulk() and __free_by_rcu() to
spot these mismatched alloc and free early.
Signed-off-by: Hou Tao <[email protected]>
Signed-off-by: Daniel Borkmann <[email protected]>
Link: https://lore.kernel.org/bpf/[email protected]
Add the following 3 test cases for bpf memory allocator:
1) Do allocation in bpf program and free through map free
2) Do batch per-cpu allocation and per-cpu free in bpf program
3) Do per-cpu allocation in bpf program and free through map free
For per-cpu allocation, because per-cpu allocation can not refill timely
sometimes, so test 2) and test 3) consider it is OK for
bpf_percpu_obj_new_impl() to return NULL.
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
The following warning was reported when running "./test_progs -t
test_bpf_ma/percpu_free_through_map_free":
------------[ cut here ]------------
WARNING: CPU: 1 PID: 68 at kernel/bpf/memalloc.c:342
CPU: 1 PID: 68 Comm: kworker/u16:2 Not tainted 6.6.0-rc2+ #222
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996)
Workqueue: events_unbound bpf_map_free_deferred
RIP: 0010:bpf_mem_refill+0x21c/0x2a0
......
Call Trace:
<IRQ>
? bpf_mem_refill+0x21c/0x2a0
irq_work_single+0x27/0x70
irq_work_run_list+0x2a/0x40
irq_work_run+0x18/0x40
__sysvec_irq_work+0x1c/0xc0
sysvec_irq_work+0x73/0x90
</IRQ>
<TASK>
asm_sysvec_irq_work+0x1b/0x20
RIP: 0010:unit_free+0x50/0x80
......
bpf_mem_free+0x46/0x60
__bpf_obj_drop_impl+0x40/0x90
bpf_obj_free_fields+0x17d/0x1a0
array_map_free+0x6b/0x170
bpf_map_free_deferred+0x54/0xa0
process_scheduled_works+0xba/0x370
worker_thread+0x16d/0x2e0
kthread+0x105/0x140
ret_from_fork+0x39/0x60
ret_from_fork_asm+0x1b/0x30
</TASK>
---[ end trace 0000000000000000 ]---
The reason is simple: __bpf_obj_drop_impl() does not know the freeing
field is a per-cpu pointer and it uses bpf_global_ma to free the
pointer. Because bpf_global_ma is not a per-cpu allocator, so ksize() is
used to select the corresponding cache. The bpf_mem_cache with 16-bytes
unit_size will always be selected to do the unmatched free and it will
trigger the warning in free_bulk() eventually.
Because per-cpu kptr doesn't support list or rb-tree now, so fix the
problem by only checking whether or not the type of kptr is per-cpu in
bpf_obj_free_fields(), and using bpf_global_percpu_ma to these kptrs.
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
For bpf_global_percpu_ma, the pointer passed to bpf_mem_free_rcu() is
allocated by kmalloc() and its size is fixed (16-bytes on x86-64). So
no matter which cache allocates the dynamic per-cpu area, on x86-64
cache[2] will always be used to free the per-cpu area.
Fix the unbalance by checking whether the bpf memory allocator is
per-cpu or not and use pcpu_alloc_size() instead of ksize() to
find the correct cache for per-cpu free.
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
Introduce pcpu_alloc_size() to get the size of the dynamic per-cpu
area. It will be used by bpf memory allocator in the following patches.
BPF memory allocator maintains per-cpu area caches for multiple area
sizes and its free API only has the to-be-freed per-cpu pointer, so it
needs the size of dynamic per-cpu area to select the corresponding cache
when bpf program frees the dynamic per-cpu pointer.
Acked-by: Dennis Zhou <[email protected]>
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
There is no need to acquire pcpu_lock for pcpu_chunk_addr_search():
1) both pcpu_first_chunk & pcpu_reserved_chunk must have been
initialized before the invocation of free_percpu().
2) The dynamically-created chunk must be valid before the per-cpu
pointers allocated from it are freed.
So acquire pcpu_lock() after the invocation of pcpu_chunk_addr_search().
Acked-by: Dennis Zhou <[email protected]>
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
Commit d52b59315b ("bpf: Adjust size_index according to the value of
KMALLOC_MIN_SIZE") uses KMALLOC_MIN_SIZE to adjust size_index, but as
reported by Nathan, the adjustment is not enough, because
__kmalloc_minalign() also decides the minimal alignment of slab object
as shown in new_kmalloc_cache() and its value may be greater than
KMALLOC_MIN_SIZE (e.g., 64 bytes vs 8 bytes under a riscv QEMU VM).
Instead of invoking __kmalloc_minalign() in bpf subsystem to find the
maximal alignment, just using kmalloc_size_roundup() directly to get the
corresponding slab object size for each allocation size. If these two
sizes are unmatched, adjust size_index to select a bpf_mem_cache with
unit_size equal to the object_size of the underlying slab cache for the
allocation size.
Fixes: 822fb26bdb ("bpf: Add a hint to allocated objects.")
Reported-by: Nathan Chancellor <[email protected]>
Closes: https://lore.kernel.org/bpf/[email protected]/
Signed-off-by: Hou Tao <[email protected]>
Tested-by: Emil Renner Berthing <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
Add a test to test all possible and valid allocation size for bpf
memory allocator. For each possible allocation size, the test uses
the following two steps to test the alloc and free path:
1) allocate N (N > high_watermark) objects to trigger the refill
executed in irq_work.
2) free N objects to trigger the freeing executed in irq_work.
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
Add extra check in bpf_mem_alloc_init() to ensure the unit_size of
bpf_mem_cache is matched with the object_size of underlying slab cache.
If these two sizes are unmatched, print a warning once and return
-EINVAL in bpf_mem_alloc_init(), so the mismatch can be found early and
the potential issue can be prevented.
Suggested-by: Alexei Starovoitov <[email protected]>
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
When the unit_size of a bpf_mem_cache is unmatched with the object_size
of the underlying slab cache, the bpf_mem_cache will not be used, and
the allocation will be redirected to a bpf_mem_cache with a bigger
unit_size instead, so there is no need to prefill for these
unused bpf_mem_caches.
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
The following warning was reported when running "./test_progs -a
link_api -a linked_list" on a RISC-V QEMU VM:
------------[ cut here ]------------
WARNING: CPU: 3 PID: 261 at kernel/bpf/memalloc.c:342 bpf_mem_refill
Modules linked in: bpf_testmod(OE)
CPU: 3 PID: 261 Comm: test_progs- ... 6.5.0-rc5-01743-gdcb152bb8328 #2
Hardware name: riscv-virtio,qemu (DT)
epc : bpf_mem_refill+0x1fc/0x206
ra : irq_work_single+0x68/0x70
epc : ffffffff801b1bc4 ra : ffffffff8015fe84 sp : ff2000000001be20
gp : ffffffff82d26138 tp : ff6000008477a800 t0 : 0000000000046600
t1 : ffffffff812b6ddc t2 : 0000000000000000 s0 : ff2000000001be70
s1 : ff5ffffffffe8998 a0 : ff5ffffffffe8998 a1 : ff600003fef4b000
a2 : 000000000000003f a3 : ffffffff80008250 a4 : 0000000000000060
a5 : 0000000000000080 a6 : 0000000000000000 a7 : 0000000000735049
s2 : ff5ffffffffe8998 s3 : 0000000000000022 s4 : 0000000000001000
s5 : 0000000000000007 s6 : ff5ffffffffe8570 s7 : ffffffff82d6bd30
s8 : 000000000000003f s9 : ffffffff82d2c5e8 s10: 000000000000ffff
s11: ffffffff82d2c5d8 t3 : ffffffff81ea8f28 t4 : 0000000000000000
t5 : ff6000008fd28278 t6 : 0000000000040000
[<ffffffff801b1bc4>] bpf_mem_refill+0x1fc/0x206
[<ffffffff8015fe84>] irq_work_single+0x68/0x70
[<ffffffff8015feb4>] irq_work_run_list+0x28/0x36
[<ffffffff8015fefa>] irq_work_run+0x38/0x66
[<ffffffff8000828a>] handle_IPI+0x3a/0xb4
[<ffffffff800a5c3a>] handle_percpu_devid_irq+0xa4/0x1f8
[<ffffffff8009fafa>] generic_handle_domain_irq+0x28/0x36
[<ffffffff800ae570>] ipi_mux_process+0xac/0xfa
[<ffffffff8000a8ea>] sbi_ipi_handle+0x2e/0x88
[<ffffffff8009fafa>] generic_handle_domain_irq+0x28/0x36
[<ffffffff807ee70e>] riscv_intc_irq+0x36/0x4e
[<ffffffff812b5d3a>] handle_riscv_irq+0x54/0x86
[<ffffffff812b6904>] do_irq+0x66/0x98
---[ end trace 0000000000000000 ]---
The warning is due to WARN_ON_ONCE(tgt->unit_size != c->unit_size) in
free_bulk(). The direct reason is that a object is allocated and
freed by bpf_mem_caches with different unit_size.
The root cause is that KMALLOC_MIN_SIZE is 64 and there is no 96-bytes
slab cache in the specific VM. When linked_list test allocates a
72-bytes object through bpf_obj_new(), bpf_global_ma will allocate it
from a bpf_mem_cache with 96-bytes unit_size, but this bpf_mem_cache is
backed by 128-bytes slab cache. When the object is freed, bpf_mem_free()
uses ksize() to choose the corresponding bpf_mem_cache. Because the
object is allocated from 128-bytes slab cache, ksize() returns 128,
bpf_mem_free() chooses a 128-bytes bpf_mem_cache to free the object and
triggers the warning.
A similar warning will also be reported when using CONFIG_SLAB instead
of CONFIG_SLUB in a x86-64 kernel. Because CONFIG_SLUB defines
KMALLOC_MIN_SIZE as 8 but CONFIG_SLAB defines KMALLOC_MIN_SIZE as 32.
An alternative fix is to use kmalloc_size_round() in bpf_mem_alloc() to
choose a bpf_mem_cache which has the same unit_size with the backing
slab cache, but it may introduce performance degradation, so fix the
warning by adjusting the indexes in size_index according to the value of
KMALLOC_MIN_SIZE just like setup_kmalloc_cache_index_table() does.
Fixes: 822fb26bdb ("bpf: Add a hint to allocated objects.")
Reported-by: Björn Töpel <[email protected]>
Closes: https://lore.kernel.org/bpf/[email protected]
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
The test case creates 4 threads and then pins these 4 threads in CPU 0.
These 4 threads will run different bpf program through
bpf_prog_test_run_opts() and these bpf program will use bpf_obj_new()
and bpf_obj_drop() to allocate and free local kptrs concurrently.
Under preemptible kernel, bpf_obj_new() and bpf_obj_drop() may preempt
each other, bpf_obj_new() may return NULL and the test will fail before
applying these fixes as shown below:
test_preempted_bpf_ma_op:PASS:open_and_load 0 nsec
test_preempted_bpf_ma_op:PASS:attach 0 nsec
test_preempted_bpf_ma_op:PASS:no test prog 0 nsec
test_preempted_bpf_ma_op:PASS:no test prog 0 nsec
test_preempted_bpf_ma_op:PASS:no test prog 0 nsec
test_preempted_bpf_ma_op:PASS:no test prog 0 nsec
test_preempted_bpf_ma_op:PASS:pthread_create 0 nsec
test_preempted_bpf_ma_op:PASS:pthread_create 0 nsec
test_preempted_bpf_ma_op:PASS:pthread_create 0 nsec
test_preempted_bpf_ma_op:PASS:pthread_create 0 nsec
test_preempted_bpf_ma_op:PASS:run prog err 0 nsec
test_preempted_bpf_ma_op:PASS:run prog err 0 nsec
test_preempted_bpf_ma_op:PASS:run prog err 0 nsec
test_preempted_bpf_ma_op:PASS:run prog err 0 nsec
test_preempted_bpf_ma_op:FAIL:ENOMEM unexpected ENOMEM: got TRUE
#168 preempted_bpf_ma_op:FAIL
Summary: 0/0 PASSED, 0 SKIPPED, 1 FAILED
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
Both unit_free() and unit_free_rcu() invoke irq_work_raise() to free
freed objects back to slab and the invocation may also be preempted by
unit_alloc() and unit_alloc() may return NULL unexpectedly as shown in
the following case:
task A task B
unit_free()
// high_watermark = 48
// free_cnt = 49 after free
irq_work_raise()
// mark irq work as IRQ_WORK_PENDING
irq_work_claim()
// task B preempts task A
unit_alloc()
// free_cnt = 48 after alloc
// does unit_alloc() 32-times
......
// free_cnt = 16
unit_alloc()
// free_cnt = 15 after alloc
// irq work is already PENDING,
// so just return
irq_work_raise()
// does unit_alloc() 15-times
......
// free_cnt = 0
unit_alloc()
// free_cnt = 0 before alloc
return NULL
Fix it by enabling IRQ after irq_work_raise() completes.
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
When doing stress test for qp-trie, bpf_mem_alloc() returned NULL
unexpectedly because all qp-trie operations were initiated from
bpf syscalls and there was still available free memory. bpf_obj_new()
has the same problem as shown by the following selftest.
The failure is due to the preemption. irq_work_raise() will invoke
irq_work_claim() first to mark the irq work as pending and then inovke
__irq_work_queue_local() to raise an IPI. So when the current task
which is invoking irq_work_raise() is preempted by other task,
unit_alloc() may return NULL for preemption task as shown below:
task A task B
unit_alloc()
// low_watermark = 32
// free_cnt = 31 after alloc
irq_work_raise()
// mark irq work as IRQ_WORK_PENDING
irq_work_claim()
// task B preempts task A
unit_alloc()
// free_cnt = 30 after alloc
// irq work is already PENDING,
// so just return
irq_work_raise()
// does unit_alloc() 30-times
......
unit_alloc()
// free_cnt = 0 before alloc
return NULL
Fix it by enabling IRQ after irq_work_raise() completes. An alternative
fix is using preempt_{disable|enable}_notrace() pair, but it may have
extra overhead. Another feasible fix is to only disable preemption or
IRQ before invoking irq_work_queue() and enable preemption or IRQ after
the invocation completes, but it can't handle the case when
c->low_watermark is 1.
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
After synchronous_rcu(), both the dettached XDP program and
xdp_do_flush() are completed, and the only user of bpf_cpu_map_entry
will be cpu_map_kthread_run(), so instead of calling
__cpu_map_entry_replace() to stop kthread and cleanup entry after a RCU
grace period, do these things directly.
Signed-off-by: Hou Tao <[email protected]>
Reviewed-by: Toke Høiland-Jørgensen <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
As for now __cpu_map_entry_replace() uses call_rcu() to wait for the
inflight xdp program to exit the RCU read critical section, and then
launch kworker cpu_map_kthread_stop() to call kthread_stop() to flush
all pending xdp frames or skbs.
But it is unnecessary to use rcu_barrier() in cpu_map_kthread_stop() to
wait for the completion of __cpu_map_entry_free(), because rcu_barrier()
will wait for all pending RCU callbacks and cpu_map_kthread_stop() only
needs to wait for the completion of a specific __cpu_map_entry_free().
So use queue_rcu_work() to replace call_rcu(), schedule_work() and
rcu_barrier(). queue_rcu_work() will queue a __cpu_map_entry_free()
kworker after a RCU grace period. Because __cpu_map_entry_free() is
running in a kworker context, so it is OK to do all of these freeing
procedures include kthread_stop() in it.
After the update, there is no need to do reference-counting for
bpf_cpu_map_entry, because bpf_cpu_map_entry is freed directly in
__cpu_map_entry_free(), so just remove it.
Signed-off-by: Hou Tao <[email protected]>
Reviewed-by: Toke Høiland-Jørgensen <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
The following warning was reported when running xdp_redirect_cpu with
both skb-mode and stress-mode enabled:
------------[ cut here ]------------
Incorrect XDP memory type (-2128176192) usage
WARNING: CPU: 7 PID: 1442 at net/core/xdp.c:405
Modules linked in:
CPU: 7 PID: 1442 Comm: kworker/7:0 Tainted: G 6.5.0-rc2+ #1
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996)
Workqueue: events __cpu_map_entry_free
RIP: 0010:__xdp_return+0x1e4/0x4a0
......
Call Trace:
<TASK>
? show_regs+0x65/0x70
? __warn+0xa5/0x240
? __xdp_return+0x1e4/0x4a0
......
xdp_return_frame+0x4d/0x150
__cpu_map_entry_free+0xf9/0x230
process_one_work+0x6b0/0xb80
worker_thread+0x96/0x720
kthread+0x1a5/0x1f0
ret_from_fork+0x3a/0x70
ret_from_fork_asm+0x1b/0x30
</TASK>
The reason for the warning is twofold. One is due to the kthread
cpu_map_kthread_run() is stopped prematurely. Another one is
__cpu_map_ring_cleanup() doesn't handle skb mode and treats skbs in
ptr_ring as XDP frames.
Prematurely-stopped kthread will be fixed by the preceding patch and
ptr_ring will be empty when __cpu_map_ring_cleanup() is called. But
as the comments in __cpu_map_ring_cleanup() said, handling and freeing
skbs in ptr_ring as well to "catch any broken behaviour gracefully".
Fixes: 11941f8a85 ("bpf: cpumap: Implement generic cpumap")
Signed-off-by: Hou Tao <[email protected]>
Acked-by: Jesper Dangaard Brouer <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Martin KaFai Lau <[email protected]>
The following warning was reported when running stress-mode enabled
xdp_redirect_cpu with some RT threads:
------------[ cut here ]------------
WARNING: CPU: 4 PID: 65 at kernel/bpf/cpumap.c:135
CPU: 4 PID: 65 Comm: kworker/4:1 Not tainted 6.5.0-rc2+ #1
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996)
Workqueue: events cpu_map_kthread_stop
RIP: 0010:put_cpu_map_entry+0xda/0x220
......
Call Trace:
<TASK>
? show_regs+0x65/0x70
? __warn+0xa5/0x240
......
? put_cpu_map_entry+0xda/0x220
cpu_map_kthread_stop+0x41/0x60
process_one_work+0x6b0/0xb80
worker_thread+0x96/0x720
kthread+0x1a5/0x1f0
ret_from_fork+0x3a/0x70
ret_from_fork_asm+0x1b/0x30
</TASK>
The root cause is the same as commit 4369016497 ("bpf: cpumap: Fix memory
leak in cpu_map_update_elem"). The kthread is stopped prematurely by
kthread_stop() in cpu_map_kthread_stop(), and kthread() doesn't call
cpu_map_kthread_run() at all but XDP program has already queued some
frames or skbs into ptr_ring. So when __cpu_map_ring_cleanup() checks
the ptr_ring, it will find it was not emptied and report a warning.
An alternative fix is to use __cpu_map_ring_cleanup() to drop these
pending frames or skbs when kthread_stop() returns -EINTR, but it may
confuse the user, because these frames or skbs have been handled
correctly by XDP program. So instead of dropping these frames or skbs,
just make sure the per-cpu kthread is running before
__cpu_map_entry_alloc() returns.
After apply the fix, the error handle for kthread_stop() will be
unnecessary because it will always return 0, so just remove it.
Fixes: 6710e11269 ("bpf: introduce new bpf cpu map type BPF_MAP_TYPE_CPUMAP")
Signed-off-by: Hou Tao <[email protected]>
Reviewed-by: Pu Lehui <[email protected]>
Acked-by: Jesper Dangaard Brouer <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Martin KaFai Lau <[email protected]>
When doing mkfs.xfs on a pmem device, the following warning was
reported:
------------[ cut here ]------------
WARNING: CPU: 2 PID: 384 at block/blk-core.c:751 submit_bio_noacct
Modules linked in:
CPU: 2 PID: 384 Comm: mkfs.xfs Not tainted 6.4.0-rc7+ #154
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996)
RIP: 0010:submit_bio_noacct+0x340/0x520
......
Call Trace:
<TASK>
? submit_bio_noacct+0xd5/0x520
submit_bio+0x37/0x60
async_pmem_flush+0x79/0xa0
nvdimm_flush+0x17/0x40
pmem_submit_bio+0x370/0x390
__submit_bio+0xbc/0x190
submit_bio_noacct_nocheck+0x14d/0x370
submit_bio_noacct+0x1ef/0x520
submit_bio+0x55/0x60
submit_bio_wait+0x5a/0xc0
blkdev_issue_flush+0x44/0x60
The root cause is that submit_bio_noacct() needs bio_op() is either
WRITE or ZONE_APPEND for flush bio and async_pmem_flush() doesn't assign
REQ_OP_WRITE when allocating flush bio, so submit_bio_noacct just fail
the flush bio.
Simply fix it by adding the missing REQ_OP_WRITE for flush bio. And we
could fix the flush order issue and do flush optimization later.
Cc: [email protected] # 6.3+
Fixes: b4a6bb3a67 ("block: add a sanity check for non-write flush/fua bios")
Reviewed-by: Christoph Hellwig <[email protected]>
Reviewed-by: Chaitanya Kulkarni <[email protected]>
Reviewed-by: Pankaj Gupta <[email protected]>
Tested-by: Pankaj Gupta <[email protected]>
Signed-off-by: Hou Tao <[email protected]>
Signed-off-by: Dave Jiang <[email protected]>
The benchmark could be used to compare the performance of hash map
operations and the memory usage between different flavors of bpf memory
allocator (e.g., no bpf ma vs bpf ma vs reuse-after-gp bpf ma). It also
could be used to check the performance improvement or the memory saving
provided by optimization.
The benchmark creates a non-preallocated hash map which uses bpf memory
allocator and shows the operation performance and the memory usage of
the hash map under different use cases:
(1) overwrite
Each CPU overwrites nonoverlapping part of hash map. When each CPU
completes overwriting of 64 elements in hash map, it increases the
op_count.
(2) batch_add_batch_del
Each CPU adds then deletes nonoverlapping part of hash map in batch.
When each CPU adds and deletes 64 elements in hash map, it increases
the op_count twice.
(3) add_del_on_diff_cpu
Each two-CPUs pair adds and deletes nonoverlapping part of map
cooperatively. When each CPU adds or deletes 64 elements in hash map,
it will increase the op_count.
The following is the benchmark results when comparing between different
flavors of bpf memory allocator. These tests are conducted on a KVM guest
with 8 CPUs and 16 GB memory. The command line below is used to do all
the following benchmarks:
./bench htab-mem --use-case $name ${OPTS} -w3 -d10 -a -p8
These results show that preallocated hash map has both better performance
and smaller memory footprint.
(1) non-preallocated + no bpf memory allocator (v6.0.19)
use kmalloc() + call_rcu
overwrite per-prod-op: 11.24 ± 0.07k/s, avg mem: 82.64 ± 26.32MiB, peak mem: 119.18MiB
batch_add_batch_del per-prod-op: 18.45 ± 0.10k/s, avg mem: 50.47 ± 14.51MiB, peak mem: 94.96MiB
add_del_on_diff_cpu per-prod-op: 14.50 ± 0.03k/s, avg mem: 4.64 ± 0.73MiB, peak mem: 7.20MiB
(2) preallocated
OPTS=--preallocated
overwrite per-prod-op: 191.42 ± 0.09k/s, avg mem: 1.24 ± 0.00MiB, peak mem: 1.49MiB
batch_add_batch_del per-prod-op: 221.83 ± 0.17k/s, avg mem: 1.23 ± 0.00MiB, peak mem: 1.49MiB
add_del_on_diff_cpu per-prod-op: 39.66 ± 0.31k/s, avg mem: 1.47 ± 0.13MiB, peak mem: 1.75MiB
(3) normal bpf memory allocator
overwrite per-prod-op: 126.59 ± 0.02k/s, avg mem: 2.26 ± 0.00MiB, peak mem: 2.74MiB
batch_add_batch_del per-prod-op: 83.37 ± 0.20k/s, avg mem: 2.14 ± 0.17MiB, peak mem: 2.74MiB
add_del_on_diff_cpu per-prod-op: 21.25 ± 0.24k/s, avg mem: 17.50 ± 3.32MiB, peak mem: 28.87MiB
Acked-by: John Fastabend <[email protected]>
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
The theoretical maximum size of ring buffer is about 64GB, but now the
size of ring buffer is specified by max_entries in bpf_attr and its
maximum value is (4GB - 1), and it won't be possible for overflow.
So just remove the unnecessary size check in ringbuf_map_alloc() but
keep the comments for possible extension in future.
Reported-by: Dan Carpenter <[email protected]>
Signed-off-by: Hou Tao <[email protected]>
Signed-off-by: Daniel Borkmann <[email protected]>
Closes: https://lore.kernel.org/bpf/[email protected]
Link: https://lore.kernel.org/bpf/[email protected]
Considering that only bench_ringbufs.c supports consumer, just set the
default value of consumer_cnt as 0. After that, update the validity
check of consumer_cnt, remove unused consumer_thread code snippets and
set consumer_cnt as 1 in run_bench_ringbufs.sh accordingly.
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
When using option -a without --prod-affinity or --cons-affinity, if the
number of producers and consumers is greater than the number of online
CPUs, the benchmark will fail to run as shown below:
$ getconf _NPROCESSORS_ONLN
8
$ ./bench bpf-loop -a -p9
Setting up benchmark 'bpf-loop'...
setting affinity to CPU #8 failed: -22
Fix it by returning the remainder of next_cpu divided by the number of
online CPUs in next_cpu().
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
Since commit a78418e6a0 ("block: Always initialize bio IO priority on
submit"), bio->bi_ioprio will never be IOPRIO_CLASS_NONE when calling
blkcg_set_ioprio(), so there will be no way to promote the io-priority
of one cgroup to IOPRIO_CLASS_RT, because bi_ioprio will always be
greater than or equals to IOPRIO_CLASS_RT.
It seems possible to call blkcg_set_ioprio() first then try to
initialize bi_ioprio later in bio_set_ioprio(), but this doesn't work
for bio in which bi_ioprio is already initialized (e.g., direct-io), so
introduce a new promote-to-rt policy to promote the iopriority of bio to
IOPRIO_CLASS_RT if the ioprio is not already RT.
For none-to-rt policy, although it doesn't work now, but considering
that its purpose was also to override the io-priority to RT and allowing
for a smoother transition, just keep it and treat it as an alias of
the promote-to-rt policy.
Acked-by: Tejun Heo <[email protected]>
Reviewed-by: Bart Van Assche <[email protected]>
Reviewed-by: Chaitanya Kulkarni <[email protected]>
Reviewed-by: Jan Kara <[email protected]>
Signed-off-by: Hou Tao <[email protected]>
Reviewed-by: Bagas Sanjaya <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Jens Axboe <[email protected]>
The size of bpf_cpumask is fixed, so there is no need to allocate many
bpf_mem_caches for bpf_cpumask_ma, just one bpf_mem_cache is enough.
Also add comments for bpf_mem_alloc_init() in bpf_mem_alloc.h to prevent
future miuse.
Signed-off-by: Hou Tao <[email protected]>
Acked-by: Jiri Olsa <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
__hash_remove() removes hash_cell with _hash_lock locked, so acquiring
_hash_lock can guarantee no-NULL hc returned from dm_get_mdptr() must
have not been removed and hc->md must still be md.
__hash_remove() also acquires dm_hash_cells_mutex before setting mdptr
as NULL. So in dm_copy_name_and_uuid(), after acquiring
dm_hash_cells_mutex and ensuring returned hc is not NULL, the returned
hc must still be alive and hc->md must still be md.
Remove the unnecessary hc->md != md checks when using dm_get_mdptr()
with _hash_lock or dm_hash_cells_mutex acquired.
Signed-off-by: Hou Tao <[email protected]>
Signed-off-by: Mike Snitzer <[email protected]>
The reinitialization of spin-lock in map value after immediate reuse may
corrupt lookup with BPF_F_LOCK flag and result in hard lock-up, so add
one test case to demonstrate the problem.
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
Currently the freed element in bpf memory allocator may be immediately
reused, for htab map the reuse will reinitialize special fields in map
value (e.g., bpf_spin_lock), but lookup procedure may still access
these special fields, and it may lead to hard-lockup as shown below:
NMI backtrace for cpu 16
CPU: 16 PID: 2574 Comm: htab.bin Tainted: G L 6.1.0+ #1
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996),
RIP: 0010:queued_spin_lock_slowpath+0x283/0x2c0
......
Call Trace:
<TASK>
copy_map_value_locked+0xb7/0x170
bpf_map_copy_value+0x113/0x3c0
__sys_bpf+0x1c67/0x2780
__x64_sys_bpf+0x1c/0x20
do_syscall_64+0x30/0x60
entry_SYSCALL_64_after_hwframe+0x46/0xb0
......
</TASK>
For htab map, just like the preallocated case, these is no need to
initialize these special fields in map value again once these fields
have been initialized. For preallocated htab map, these fields are
initialized through __GFP_ZERO in bpf_map_area_alloc(), so do the
similar thing for non-preallocated htab in bpf memory allocator. And
there is no need to use __GFP_ZERO for per-cpu bpf memory allocator,
because __alloc_percpu_gfp() does it implicitly.
Fixes: 0fd7c5d433 ("bpf: Optimize call_rcu in non-preallocated hash map.")
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
Don't update recovery_cp when curr_resync is MD_RESYNC_ACTIVE, otherwise
md may skip the resync of the first 3 sectors if the resync procedure is
interrupted before the first calling of ->sync_request() as shown below:
md_do_sync thread control thread
// setup resync
mddev->recovery_cp = 0
j = 0
mddev->curr_resync = MD_RESYNC_ACTIVE
// e.g., set array as idle
set_bit(MD_RECOVERY_INTR, &&mddev_recovery)
// resync loop
// check INTR before calling sync_request
!test_bit(MD_RECOVERY_INTR, &mddev->recovery
// resync interrupted
// update recovery_cp from 0 to 3
// the resync of three 3 sectors will be skipped
mddev->recovery_cp = 3
Fixes: eac58d08d4 ("md: Use enum for overloaded magic numbers used by mddev->curr_resync")
Cc: [email protected] # 6.0+
Signed-off-by: Hou Tao <[email protected]>
Reviewed-by: Logan Gunthorpe <[email protected]>
Signed-off-by: Song Liu <[email protected]>
fscache_create_volume_work() uses wake_up_bit() to wake up the processes
which are waiting for the completion of volume creation. According to
comments in wake_up_bit() and waitqueue_active(), an extra smp_mb() is
needed to guarantee the memory order between FSCACHE_VOLUME_CREATING
flag and waitqueue_active() before invoking wake_up_bit().
Fixing it by using clear_and_wake_up_bit() to add the missing memory
barrier.
Reviewed-by: Jingbo Xu <[email protected]>
Signed-off-by: Hou Tao <[email protected]>
Signed-off-by: David Howells <[email protected]>
Reviewed-by: Jeff Layton <[email protected]>
Link: https://lore.kernel.org/r/[email protected]/ # v3
The freeing of relinquished volume will wake up the pending volume
acquisition by using wake_up_bit(), however it is mismatched with
wait_var_event() used in fscache_wait_on_volume_collision() and it will
never wake up the waiter in the wait-queue because these two functions
operate on different wait-queues.
According to the implementation in fscache_wait_on_volume_collision(),
if the wake-up of pending acquisition is delayed longer than 20 seconds
(e.g., due to the delay of on-demand fd closing), the first
wait_var_event_timeout() will timeout and the following wait_var_event()
will hang forever as shown below:
FS-Cache: Potential volume collision new=00000024 old=00000022
......
INFO: task mount:1148 blocked for more than 122 seconds.
Not tainted 6.1.0-rc6+ #1
task:mount state:D stack:0 pid:1148 ppid:1
Call Trace:
<TASK>
__schedule+0x2f6/0xb80
schedule+0x67/0xe0
fscache_wait_on_volume_collision.cold+0x80/0x82
__fscache_acquire_volume+0x40d/0x4e0
erofs_fscache_register_volume+0x51/0xe0 [erofs]
erofs_fscache_register_fs+0x19c/0x240 [erofs]
erofs_fc_fill_super+0x746/0xaf0 [erofs]
vfs_get_super+0x7d/0x100
get_tree_nodev+0x16/0x20
erofs_fc_get_tree+0x20/0x30 [erofs]
vfs_get_tree+0x24/0xb0
path_mount+0x2fa/0xa90
do_mount+0x7c/0xa0
__x64_sys_mount+0x8b/0xe0
do_syscall_64+0x30/0x60
entry_SYSCALL_64_after_hwframe+0x46/0xb0
Considering that wake_up_bit() is more selective, so fix it by using
wait_on_bit() instead of wait_var_event() to wait for the freeing of
relinquished volume. In addition because waitqueue_active() is used in
wake_up_bit() and clear_bit() doesn't imply any memory barrier, use
clear_and_wake_up_bit() to add the missing memory barrier between
cursor->flags and waitqueue_active().
Fixes: 62ab633523 ("fscache: Implement volume registration")
Reviewed-by: Jingbo Xu <[email protected]>
Signed-off-by: Hou Tao <[email protected]>
Signed-off-by: David Howells <[email protected]>
Reviewed-by: Jeff Layton <[email protected]>
Link: https://lore.kernel.org/r/[email protected]/ # v3
There are warnings reported from resolve_btfids when building vmlinux
with CONFIG_SECURITY_NETWORK disabled:
WARN: resolve_btfids: unresolved symbol bpf_lsm_sk_free_security
WARN: resolve_btfids: unresolved symbol bpf_lsm_sk_alloc_security
So only define BTF IDs for these LSM hooks when CONFIG_SECURITY_NETWORK
is enabled.
Fixes: c0c852dd18 ("bpf: Do not mark certain LSM hook arguments as trusted")
Signed-off-by: Hou Tao <[email protected]>
Signed-off-by: Daniel Borkmann <[email protected]>
Acked-by: Yonghong Song <[email protected]>
Link: https://lore.kernel.org/bpf/[email protected]
If there are pending rcu callback, free_mem_alloc() will use
rcu_barrier_tasks_trace() and rcu_barrier() to wait for the pending
__free_rcu_tasks_trace() and __free_rcu() callback.
If rcu_trace_implies_rcu_gp() is true, there will be no pending
__free_rcu(), so it will be OK to skip rcu_barrier() as well.
Acked-by: Yonghong Song <[email protected]>
Acked-by: Paul E. McKenney <[email protected]>
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
When there are batched freeing operations on a specific CPU, part of
the freed elements ((high_watermark - lower_watermark) / 2 + 1) will be
indirectly moved into waiting_for_gp list through free_by_rcu list.
After call_rcu_in_progress becomes false again, the remaining elements
in free_by_rcu list will be moved to waiting_for_gp list by the next
invocation of free_bulk(). However if the expiration of RCU tasks trace
grace period is relatively slow, none element in free_by_rcu list will
be moved.
So instead of invoking __alloc_percpu_gfp() or kmalloc_node() to
allocate a new object, in alloc_bulk() just check whether or not there is
freed element in free_by_rcu list and reuse it if available.
Acked-by: Yonghong Song <[email protected]>
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
When shared domain is enabled, doing mount twice with the same fsid and
domain_id will trigger sysfs warning as shown below:
sysfs: cannot create duplicate filename '/fs/erofs/d0,meta.bin'
CPU: 15 PID: 1051 Comm: mount Not tainted 6.1.0-rc6+ #1
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996)
Call Trace:
<TASK>
dump_stack_lvl+0x38/0x49
dump_stack+0x10/0x12
sysfs_warn_dup.cold+0x17/0x27
sysfs_create_dir_ns+0xb8/0xd0
kobject_add_internal+0xb1/0x240
kobject_init_and_add+0x71/0xa0
erofs_register_sysfs+0x89/0x110
erofs_fc_fill_super+0x98c/0xaf0
vfs_get_super+0x7d/0x100
get_tree_nodev+0x16/0x20
erofs_fc_get_tree+0x20/0x30
vfs_get_tree+0x24/0xb0
path_mount+0x2fa/0xa90
do_mount+0x7c/0xa0
__x64_sys_mount+0x8b/0xe0
do_syscall_64+0x30/0x60
entry_SYSCALL_64_after_hwframe+0x46/0xb0
The reason is erofs_fscache_register_cookie() doesn't guarantee the primary
data blob (aka fsid) is unique in the shared domain and
erofs_register_sysfs() invoked by the second mount will fail due to the
duplicated fsid in the shared domain and report warning.
It would be better to check the uniqueness of fsid before doing
erofs_register_sysfs(), so adding a new flags parameter for
erofs_fscache_register_cookie() and doing the uniqueness check if
EROFS_REG_COOKIE_NEED_NOEXIST is enabled.
After the patch, the error in dmesg for the duplicated mount would be:
erofs: ...: erofs_domain_register_cookie: XX already exists in domain YY
Reviewed-by: Jia Zhu <[email protected]>
Reviewed-by: Jingbo Xu <[email protected]>
Reviewed-by: Chao Yu <[email protected]>
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Fixes: 7d41963759 ("erofs: Support sharing cookies in the same domain")
Signed-off-by: Gao Xiang <[email protected]>
The test closes both iterator link fd and cgroup fd, and removes the
cgroup file to make a dead cgroup before reading from cgroup iterator.
It also uses kern_sync_rcu() and usleep() to wait for the release of
start cgroup. If the start cgroup is not pinned by cgroup iterator,
reading from iterator fd will trigger use-after-free.
Signed-off-by: Hou Tao <[email protected]>
Signed-off-by: Daniel Borkmann <[email protected]>
Acked-by: Yonghong Song <[email protected]>
Acked-by: Hao Luo <[email protected]>
Link: https://lore.kernel.org/bpf/[email protected]
bpf_iter_attach_cgroup() has already acquired an extra reference for the
start cgroup, but the reference may be released if the iterator link fd
is closed after the creation of iterator fd, and it may lead to
user-after-free problem when reading the iterator fd.
An alternative fix is pinning iterator link when opening iterator,
but it will make iterator link being still visible after the close of
iterator link fd and the behavior is different with other link types, so
just fixing it by acquiring another reference for the start cgroup.
Fixes: d4ccaf58a8 ("bpf: Introduce cgroup iter")
Signed-off-by: Hou Tao <[email protected]>
Signed-off-by: Daniel Borkmann <[email protected]>
Acked-by: Yonghong Song <[email protected]>
Link: https://lore.kernel.org/bpf/[email protected]
The top two bits of size are used as busy and discard flags, so reject
the reservation that has any of these special bits in the size. With the
addition of validity check, these is also no need to check whether or
not total_size is overflowed.
Signed-off-by: Hou Tao <[email protected]>
Signed-off-by: Andrii Nakryiko <[email protected]>
Link: https://lore.kernel.org/bpf/[email protected]
Similar with the overflow problem on ringbuf mmap, in user_ringbuf_map()
2 * max_entries may overflow u32 when mapping writeable region.
Fixing it by casting the size of writable mmap region into a __u64 and
checking whether or not there will be overflow during mmap.
Fixes: b66ccae01f ("bpf: Add libbpf logic for user-space ring buffer")
Signed-off-by: Hou Tao <[email protected]>
Signed-off-by: Andrii Nakryiko <[email protected]>
Link: https://lore.kernel.org/bpf/[email protected]
The maximum size of ringbuf is 2GB on x86-64 host, so 2 * max_entries
will overflow u32 when mapping producer page and data pages. Only
casting max_entries to size_t is not enough, because for 32-bits
application on 64-bits kernel the size of read-only mmap region
also could overflow size_t.
So fixing it by casting the size of read-only mmap region into a __u64
and checking whether or not there will be overflow during mmap.
Fixes: bf99c936f9 ("libbpf: Add BPF ring buffer support")
Signed-off-by: Hou Tao <[email protected]>
Signed-off-by: Andrii Nakryiko <[email protected]>
Link: https://lore.kernel.org/bpf/[email protected]
Using page size as max_entries when probing ring buffer map, else the
probe may fail on host with 64KB page size (e.g., an ARM64 host).
After the fix, the output of "bpftool feature" on above host will be
correct.
Before :
eBPF map_type ringbuf is NOT available
eBPF map_type user_ringbuf is NOT available
After :
eBPF map_type ringbuf is available
eBPF map_type user_ringbuf is available
Signed-off-by: Hou Tao <[email protected]>
Signed-off-by: Andrii Nakryiko <[email protected]>
Link: https://lore.kernel.org/bpf/[email protected]
Currently bpf_map_do_batch() first invokes fdget(batch.map_fd) to get
the target map file, then it invokes generic_map_update_batch() to do
batch update. generic_map_update_batch() will get the target map file
by using fdget(batch.map_fd) again and pass it to bpf_map_update_value().
The problem is map file returned by the second fdget() may be NULL or a
totally different file compared by map file in bpf_map_do_batch(). The
reason is that the first fdget() only guarantees the liveness of struct
file instead of file descriptor and the file description may be released
by concurrent close() through pick_file().
It doesn't incur any problem as for now, because maps with batch update
support don't use map file in .map_fd_get_ptr() ops. But it is better to
fix the potential access of an invalid map file.
Using __bpf_map_get() again in generic_map_update_batch() can not fix
the problem, because batch.map_fd may be closed and reopened, and the
returned map file may be different with map file got in
bpf_map_do_batch(), so just passing the map file directly to
.map_update_batch() in bpf_map_do_batch().
Signed-off-by: Hou Tao <[email protected]>
Signed-off-by: Daniel Borkmann <[email protected]>
Acked-by: Yonghong Song <[email protected]>
Link: https://lore.kernel.org/bpf/[email protected]
Since commit bfea9a8574 ("bpf: Add name to struct bpf_ksym"), when
reporting subprog ksymbol to perf, prog name instead of subprog name is
used. The backtrace of bpf program with subprogs will be incorrect as
shown below:
ffffffffc02deace bpf_prog_e44a3057dcb151f8_overwrite+0x66
ffffffffc02de9f7 bpf_prog_e44a3057dcb151f8_overwrite+0x9f
ffffffffa71d8d4e trace_call_bpf+0xce
ffffffffa71c2938 perf_call_bpf_enter.isra.0+0x48
overwrite is the entry program and it invokes the overwrite_htab subprog
through bpf_loop, but in above backtrace, overwrite program just jumps
inside itself.
Fixing it by using subprog name when reporting subprog ksymbol. After
the fix, the output of perf script will be correct as shown below:
ffffffffc031aad2 bpf_prog_37c0bec7d7c764a4_overwrite_htab+0x66
ffffffffc031a9e7 bpf_prog_c7eb827ef4f23e71_overwrite+0x9f
ffffffffa3dd8d4e trace_call_bpf+0xce
ffffffffa3dc2938 perf_call_bpf_enter.isra.0+0x48
Fixes: bfea9a8574 ("bpf: Add name to struct bpf_ksym")
Signed-off-by: Hou Tao <[email protected]>
Signed-off-by: Daniel Borkmann <[email protected]>
Acked-by: Jiri Olsa <[email protected]>
Link: https://lore.kernel.org/bpf/[email protected]
Except for waiting_for_gp list, there are no concurrent operations on
free_by_rcu, free_llist and free_llist_extra lists, so use
__llist_del_all() instead of llist_del_all(). waiting_for_gp list can be
deleted by RCU callback concurrently, so still use llist_del_all().
Acked-by: Stanislav Fomichev <[email protected]>
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
A busy irq work is an unfinished irq work and it can be either in the
pending state or in the running state. When destroying bpf memory
allocator, refill_work may be busy for PREEMPT_RT kernel in which irq
work is invoked in a per-CPU RT-kthread. It is also possible for kernel
with arch_irq_work_has_interrupt() being false (e.g. 1-cpu arm32 host or
mips) and irq work is inovked in timer interrupt.
The busy refill_work leads to various issues. The obvious one is that
there will be concurrent operations on free_by_rcu and free_list between
irq work and memory draining. Another one is call_rcu_in_progress will
not be reliable for the checking of pending RCU callback because
do_call_rcu() may have not been invoked by irq work yet. The other is
there will be use-after-free if irq work is freed before the callback
of irq work is invoked as shown below:
BUG: kernel NULL pointer dereference, address: 0000000000000000
#PF: supervisor instruction fetch in kernel mode
#PF: error_code(0x0010) - not-present page
PGD 12ab94067 P4D 12ab94067 PUD 1796b4067 PMD 0
Oops: 0010 [#1] PREEMPT_RT SMP
CPU: 5 PID: 64 Comm: irq_work/5 Not tainted 6.0.0-rt11+ #1
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996)
RIP: 0010:0x0
Code: Unable to access opcode bytes at 0xffffffffffffffd6.
RSP: 0018:ffffadc080293e78 EFLAGS: 00010286
RAX: 0000000000000000 RBX: ffffcdc07fb6a388 RCX: ffffa05000a2e000
RDX: ffffa05000a2e000 RSI: ffffffff96cc9827 RDI: ffffcdc07fb6a388
......
Call Trace:
<TASK>
irq_work_single+0x24/0x60
irq_work_run_list+0x24/0x30
run_irq_workd+0x23/0x30
smpboot_thread_fn+0x203/0x300
kthread+0x126/0x150
ret_from_fork+0x1f/0x30
</TASK>
Considering the ease of concurrency handling, no overhead for
irq_work_sync() under non-PREEMPT_RT kernel and has-irq-work-interrupt
kernel and the short wait time used for irq_work_sync() under PREEMPT_RT
(When running two test_maps on PREEMPT_RT kernel and 72-cpus host, the
max wait time is about 8ms and the 99th percentile is 10us), just using
irq_work_sync() to wait for busy refill_work to complete before memory
draining and memory freeing.
Fixes: 7c8199e24f ("bpf: Introduce any context BPF specific memory allocator.")
Acked-by: Stanislav Fomichev <[email protected]>
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
To support both sleepable and normal uprobe bpf program, the freeing of
trace program array chains a RCU-tasks-trace grace period and a normal
RCU grace period one after the other.
With the introduction of rcu_trace_implies_rcu_gp(),
__bpf_prog_array_free_sleepable_cb() can check whether or not a normal
RCU grace period has also passed after a RCU-tasks-trace grace period
has passed. If it is true, it is safe to invoke kfree() directly.
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
Local storage map is accessible for both sleepable and non-sleepable bpf
program, and its memory is freed by using both call_rcu_tasks_trace() and
kfree_rcu() to wait for both RCU-tasks-trace grace period and RCU grace
period to pass.
With the introduction of rcu_trace_implies_rcu_gp(), both
bpf_selem_free_rcu() and bpf_local_storage_free_rcu() can check whether
or not a normal RCU grace period has also passed after a RCU-tasks-trace
grace period has passed. If it is true, it is safe to call kfree()
directly.
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
The memory free logic in bpf memory allocator chains a RCU Tasks Trace
grace period and a normal RCU grace period one after the other, so it
can ensure that both sleepable and non-sleepable programs have finished.
With the introduction of rcu_trace_implies_rcu_gp(),
__free_rcu_tasks_trace() can check whether or not a normal RCU grace
period has also passed after a RCU Tasks Trace grace period has passed.
If it is true, freeing these elements directly, else freeing through
call_rcu().
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
SYS_pidfd_open may be undefined for old glibc, so using sys_pidfd_open()
helper defined in task_local_storage_helpers.h instead to fix potential
build failure.
And according to commit 7615d9e178 ("arch: wire-up pidfd_open()"), the
syscall number of pidfd_open is always 434 except for alpha architure,
so update the definition of __NR_pidfd_open accordingly.
Signed-off-by: Hou Tao <[email protected]>
Signed-off-by: Andrii Nakryiko <[email protected]>
Link: https://lore.kernel.org/bpf/[email protected]
Free the created fd or allocated bpf_object after test case succeeds,
else there will be resource leaks.
Spotted by using address sanitizer and checking the content of
/proc/$pid/fd directory.
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Martin KaFai Lau <[email protected]>
For a non-preallocated hash map on RT kernel, regular spinlock instead
of raw spinlock is used for bucket lock. The reason is that on RT kernel
memory allocation is forbidden under atomic context and regular spinlock
is sleepable under RT.
Now hash map has been fully converted to use bpf_map_alloc, and there
will be no synchronous memory allocation for non-preallocated hash map,
so it is safe to always use raw spinlock for bucket lock on RT. So
removing the usage of htab_use_raw_lock() and updating the comments
accordingly.
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
llnode could be NULL if there are new allocations after the checking of
c-free_cnt > c->high_watermark in bpf_mem_refill() and before the
calling of __llist_del_first() in free_bulk (e.g. a PREEMPT_RT kernel
or allocation in NMI context). And it will incur oops as shown below:
BUG: kernel NULL pointer dereference, address: 0000000000000000
#PF: supervisor write access in kernel mode
#PF: error_code(0x0002) - not-present page
PGD 0 P4D 0
Oops: 0002 [#1] PREEMPT_RT SMP
CPU: 39 PID: 373 Comm: irq_work/39 Tainted: G W 6.0.0-rc6-rt9+ #1
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996)
RIP: 0010:bpf_mem_refill+0x66/0x130
......
Call Trace:
<TASK>
irq_work_single+0x24/0x60
irq_work_run_list+0x24/0x30
run_irq_workd+0x18/0x20
smpboot_thread_fn+0x13f/0x2c0
kthread+0x121/0x140
? kthread_complete_and_exit+0x20/0x20
ret_from_fork+0x1f/0x30
</TASK>
Simply fixing it by checking whether or not llnode is NULL in free_bulk().
Fixes: 8d5a8011b3 ("bpf: Batch call_rcu callbacks instead of SLAB_TYPESAFE_BY_RCU.")
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>
Add test result message when test_task_storage_map_stress_lookup()
succeeds or is skipped. The test case can be skipped due to the choose
of preemption model in kernel config, so export skips in test_maps.c and
increase it when needed.
The following is the output of test_maps when the test case succeeds or
is skipped:
test_task_storage_map_stress_lookup:PASS
test_maps: OK, 0 SKIPPED
test_task_storage_map_stress_lookup SKIP (no CONFIG_PREEMPT)
test_maps: OK, 1 SKIPPED
Fixes: 73b97bc78b ("selftests/bpf: Test concurrent updates on bpf_task_storage_busy")
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Martin KaFai Lau <[email protected]>
When CONFIG_SECURITY_NETWORK is disabled, there will be build warnings
from resolve_btfids:
WARN: resolve_btfids: unresolved symbol bpf_lsm_socket_socketpair
......
WARN: resolve_btfids: unresolved symbol bpf_lsm_inet_conn_established
Fixing it by wrapping these BTF ID definitions by CONFIG_SECURITY_NETWORK.
Fixes: 69fd337a97 ("bpf: per-cgroup lsm flavor")
Fixes: 9113d7e48e ("bpf: expose bpf_{g,s}etsockopt to lsm cgroup")
Signed-off-by: Hou Tao <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Martin KaFai Lau <[email protected]>
Under full preemptible kernel, task local storage lookup operations on
the same CPU may update per-cpu bpf_task_storage_busy concurrently. If
the update of bpf_task_storage_busy is not preemption safe, the final
value of bpf_task_storage_busy may become not-zero forever and
bpf_task_storage_trylock() will always fail. So add a test case to
ensure the update of bpf_task_storage_busy is preemption safe.
Will skip the test case when CONFIG_PREEMPT is disabled, and it can only
reproduce the problem probabilistically. By increasing
TASK_STORAGE_MAP_NR_LOOP and running it under ARM64 VM with 4-cpus, it
takes about four rounds to reproduce:
> test_maps is modified to only run test_task_storage_map_stress_lookup()
$ export TASK_STORAGE_MAP_NR_THREAD=256
$ export TASK_STORAGE_MAP_NR_LOOP=81920
$ export TASK_STORAGE_MAP_PIN_CPU=1
$ time ./test_maps
test_task_storage_map_stress_lookup(135):FAIL:bad bpf_task_storage_busy got -2
real 0m24.743s
user 0m6.772s
sys 0m17.966s
Signed-off-by: Hou Tao <[email protected]>
Acked-by: Alexei Starovoitov <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Martin KaFai Lau <[email protected]>
Both __this_cpu_inc_return() and __this_cpu_dec() are not preemption
safe and now migrate_disable() doesn't disable preemption, so the update
of prog-active is not atomic and in theory under fully preemptible kernel
recurisve prevention may do not work.
Fixing by using the preemption-safe and IRQ-safe variants.
Fixes: ca06f55b90 ("bpf: Add per-program recursion prevention mechanism")
Signed-off-by: Hou Tao <[email protected]>
Acked-by: Alexei Starovoitov <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Martin KaFai Lau <[email protected]>
Now migrate_disable() does not disable preemption and under some
architectures (e.g. arm64) __this_cpu_{inc|dec|inc_return} are neither
preemption-safe nor IRQ-safe, so for fully preemptible kernel concurrent
lookups or updates on the same task local storage and on the same CPU
may make bpf_task_storage_busy be imbalanced, and
bpf_task_storage_trylock() on the specific cpu will always fail.
Fixing it by using this_cpu_{inc|dec|inc_return} when manipulating
bpf_task_storage_busy.
Fixes: bc235cdb42 ("bpf: Prevent deadlock from recursive bpf_task_storage_[get|delete]")
Signed-off-by: Hou Tao <[email protected]>
Acked-by: Alexei Starovoitov <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Martin KaFai Lau <[email protected]>
Hou Tao