Simplify task_file_seq_get_next() by removing two in/out arguments: task
and fstruct. Use info->task and info->files instead.
Signed-off-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20201120002833.2481110-1-songliubraving@fb.com
Commit e679654a70 ("bpf: Fix a rcu_sched stall issue with
bpf task/task_file iterator") tries to fix rcu stalls warning
which is caused by bpf task_file iterator when running
"bpftool prog".
rcu: INFO: rcu_sched self-detected stall on CPU
rcu: \x097-....: (20999 ticks this GP) idle=302/1/0x4000000000000000 softirq=1508852/1508852 fqs=4913
\x09(t=21031 jiffies g=2534773 q=179750)
NMI backtrace for cpu 7
CPU: 7 PID: 184195 Comm: bpftool Kdump: loaded Tainted: G W 5.8.0-00004-g68bfc7f8c1b4 #6
Hardware name: Quanta Twin Lakes MP/Twin Lakes Passive MP, BIOS F09_3A17 05/03/2019
Call Trace:
<IRQ>
dump_stack+0x57/0x70
nmi_cpu_backtrace.cold+0x14/0x53
? lapic_can_unplug_cpu.cold+0x39/0x39
nmi_trigger_cpumask_backtrace+0xb7/0xc7
rcu_dump_cpu_stacks+0xa2/0xd0
rcu_sched_clock_irq.cold+0x1ff/0x3d9
? tick_nohz_handler+0x100/0x100
update_process_times+0x5b/0x90
tick_sched_timer+0x5e/0xf0
__hrtimer_run_queues+0x12a/0x2a0
hrtimer_interrupt+0x10e/0x280
__sysvec_apic_timer_interrupt+0x51/0xe0
asm_call_on_stack+0xf/0x20
</IRQ>
sysvec_apic_timer_interrupt+0x6f/0x80
...
task_file_seq_next+0x52/0xa0
bpf_seq_read+0xb9/0x320
vfs_read+0x9d/0x180
ksys_read+0x5f/0xe0
do_syscall_64+0x38/0x60
entry_SYSCALL_64_after_hwframe+0x44/0xa9
The fix is to limit the number of bpf program runs to be
one million. This fixed the program in most cases. But
we also found under heavy load, which can increase the wallclock
time for bpf_seq_read(), the warning may still be possible.
For example, calling bpf_delay() in the "while" loop of
bpf_seq_read(), which will introduce artificial delay,
the warning will show up in my qemu run.
static unsigned q;
volatile unsigned *p = &q;
volatile unsigned long long ll;
static void bpf_delay(void)
{
int i, j;
for (i = 0; i < 10000; i++)
for (j = 0; j < 10000; j++)
ll += *p;
}
There are two ways to fix this issue. One is to reduce the above
one million threshold to say 100,000 and hopefully rcu warning will
not show up any more. Another is to introduce a target feature
which enables bpf_seq_read() calling cond_resched().
This patch took second approach as the first approach may cause
more -EAGAIN failures for read() syscalls. Note that not all bpf_iter
targets can permit cond_resched() in bpf_seq_read() as some, e.g.,
netlink seq iterator, rcu read lock critical section spans through
seq_ops->next() -> seq_ops->show() -> seq_ops->next().
For the kernel code with the above hack, "bpftool p" roughly takes
38 seconds to finish on my VM with 184 bpf program runs.
Using the following command, I am able to collect the number of
context switches:
perf stat -e context-switches -- ./bpftool p >& log
Without this patch,
69 context-switches
With this patch,
75 context-switches
This patch added additional 6 context switches, roughly every 6 seconds
to reschedule, to avoid lengthy no-rescheduling which may cause the
above RCU warnings.
Signed-off-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20201028061054.1411116-1-yhs@fb.com
Currently, task_file iterator iterates all files from all tasks.
This may potentially visit a lot of duplicated files if there are
many tasks sharing the same files, e.g., typical pthreads
where these pthreads and the main thread are sharing the same files.
This patch changed task_file iterator to skip a particular task
if that task shares the same files as its group_leader (the task
having the same tgid and also task->tgid == task->pid).
This will preserve the same result, visiting all files from all
tasks, and will reduce runtime cost significantl, e.g., if there are
a lot of pthreads and the process has a lot of open files.
Suggested-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Link: https://lore.kernel.org/bpf/20200902023112.1672792-1-yhs@fb.com
Currently when traversing all tasks, the next tid
is always increased by one. This may result in
visiting the same task multiple times in a
pid namespace.
This patch fixed the issue by seting the next
tid as pid_nr_ns(pid, ns) + 1, similar to
funciton next_tgid().
Signed-off-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Cc: Rik van Riel <riel@surriel.com>
Link: https://lore.kernel.org/bpf/20200818222310.2181500-1-yhs@fb.com
With latest `bpftool prog` command, we observed the following kernel
panic.
BUG: kernel NULL pointer dereference, address: 0000000000000000
#PF: supervisor instruction fetch in kernel mode
#PF: error_code(0x0010) - not-present page
PGD dfe894067 P4D dfe894067 PUD deb663067 PMD 0
Oops: 0010 [#1] SMP
CPU: 9 PID: 6023 ...
RIP: 0010:0x0
Code: Bad RIP value.
RSP: 0000:ffffc900002b8f18 EFLAGS: 00010286
RAX: ffff8883a405f400 RBX: ffff888e46a6bf00 RCX: 000000008020000c
RDX: 0000000000000000 RSI: 0000000000000001 RDI: ffff8883a405f400
RBP: ffff888e46a6bf50 R08: 0000000000000000 R09: ffffffff81129600
R10: ffff8883a405f300 R11: 0000160000000000 R12: 0000000000002710
R13: 000000e9494b690c R14: 0000000000000202 R15: 0000000000000009
FS: 00007fd9187fe700(0000) GS:ffff888e46a40000(0000) knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: ffffffffffffffd6 CR3: 0000000de5d33002 CR4: 0000000000360ee0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Call Trace:
<IRQ>
rcu_core+0x1a4/0x440
__do_softirq+0xd3/0x2c8
irq_exit+0x9d/0xa0
smp_apic_timer_interrupt+0x68/0x120
apic_timer_interrupt+0xf/0x20
</IRQ>
RIP: 0033:0x47ce80
Code: Bad RIP value.
RSP: 002b:00007fd9187fba40 EFLAGS: 00000206 ORIG_RAX: ffffffffffffff13
RAX: 0000000000000002 RBX: 00007fd931789160 RCX: 000000000000010c
RDX: 00007fd9308cdfb4 RSI: 00007fd9308cdfb4 RDI: 00007ffedd1ea0a8
RBP: 00007fd9187fbab0 R08: 000000000000000e R09: 000000000000002a
R10: 0000000000480210 R11: 00007fd9187fc570 R12: 00007fd9316cc400
R13: 0000000000000118 R14: 00007fd9308cdfb4 R15: 00007fd9317a9380
After further analysis, the bug is triggered by
Commit eaaacd2391 ("bpf: Add task and task/file iterator targets")
which introduced task_file bpf iterator, which traverses all open file
descriptors for all tasks in the current namespace.
The latest `bpftool prog` calls a task_file bpf program to traverse
all files in the system in order to associate processes with progs/maps, etc.
When traversing files for a given task, rcu read_lock is taken to
access all files in a file_struct. But it used get_file() to grab
a file, which is not right. It is possible file->f_count is 0 and
get_file() will unconditionally increase it.
Later put_file() may cause all kind of issues with the above
as one of sympotoms.
The failure can be reproduced with the following steps in a few seconds:
$ cat t.c
#include <stdio.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <unistd.h>
#define N 10000
int fd[N];
int main() {
int i;
for (i = 0; i < N; i++) {
fd[i] = open("./note.txt", 'r');
if (fd[i] < 0) {
fprintf(stderr, "failed\n");
return -1;
}
}
for (i = 0; i < N; i++)
close(fd[i]);
return 0;
}
$ gcc -O2 t.c
$ cat run.sh
#/bin/bash
for i in {1..100}
do
while true; do ./a.out; done &
done
$ ./run.sh
$ while true; do bpftool prog >& /dev/null; done
This patch used get_file_rcu() which only grabs a file if the
file->f_count is not zero. This is to ensure the file pointer
is always valid. The above reproducer did not fail for more
than 30 minutes.
Fixes: eaaacd2391 ("bpf: Add task and task/file iterator targets")
Suggested-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Link: https://lore.kernel.org/bpf/20200817174214.252601-1-yhs@fb.com
This patch refactored target bpf_iter_init_seq_priv_t callback
function to accept additional information. This will be needed
in later patches for map element targets since a particular
map should be passed to traverse elements for that particular
map. In the future, other information may be passed to target
as well, e.g., pid, cgroup id, etc. to customize the iterator.
Signed-off-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200723184110.590156-1-yhs@fb.com
There is no functionality change for this patch.
Struct bpf_iter_reg is used to register a bpf_iter target,
which includes information for both prog_load, link_create
and seq_file creation.
This patch puts fields related seq_file creation into
a different structure. This will be useful for map
elements iterator where one iterator covers different
map types and different map types may have different
seq_ops, init/fini private_data function and
private_data size.
Signed-off-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200723184109.590030-1-yhs@fb.com
Currently, the pos pointer in bpf iterator map/task/task_file
seq_ops->start() is always incremented.
This is incorrect. It should be increased only if
*pos is 0 (for SEQ_START_TOKEN) since these start()
function actually returns the first real object.
If *pos is not 0, it merely found the object
based on the state in seq->private, and not really
advancing the *pos. This patch fixed this issue
by only incrementing *pos if it is 0.
Note that the old *pos calculation, although not
correct, does not affect correctness of bpf_iter
as bpf_iter seq_file->read() does not support llseek.
This patch also renamed "mid" in bpf_map iterator
seq_file private data to "map_id" for better clarity.
Fixes: 6086d29def ("bpf: Add bpf_map iterator")
Fixes: eaaacd2391 ("bpf: Add task and task/file iterator targets")
Signed-off-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200722195156.4029817-1-yhs@fb.com
One additional field btf_id is added to struct
bpf_ctx_arg_aux to store the precomputed btf_ids.
The btf_id is computed at build time with
BTF_ID_LIST or BTF_ID_LIST_GLOBAL macro definitions.
All existing bpf iterators are changed to used
pre-compute btf_ids.
Signed-off-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200720163403.1393551-1-yhs@fb.com
task_seq_get_next might stop prematurely if get_pid_task() fails to get
task_struct. Failure to do so doesn't mean that there are no more tasks with
higher pids. Procfs's iteration algorithm (see next_tgid in fs/proc/base.c)
does a retry in such case. After this fix, instead of stopping prematurely
after about 300 tasks on my server, bpf_iter program now returns >4000, which
sounds much closer to reality.
Fixes: eaaacd2391 ("bpf: Add task and task/file iterator targets")
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20200514055137.1564581-1-andriin@fb.com
Commit b121b341e5 ("bpf: Add PTR_TO_BTF_ID_OR_NULL
support") adds a field btf_id_or_null_non0_off to
bpf_prog->aux structure to indicate that the
first ctx argument is PTR_TO_BTF_ID reg_type and
all others are PTR_TO_BTF_ID_OR_NULL.
This approach does not really scale if we have
other different reg types in the future, e.g.,
a pointer to a buffer.
This patch enables bpf_iter targets registering ctx argument
reg types which may be different from the default one.
For example, for pointers to structures, the default reg_type
is PTR_TO_BTF_ID for tracing program. The target can register
a particular pointer type as PTR_TO_BTF_ID_OR_NULL which can
be used by the verifier to enforce accesses.
Signed-off-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Link: https://lore.kernel.org/bpf/20200513180221.2949882-1-yhs@fb.com
Currently bpf_iter_reg_target takes parameters from target
and allocates memory to save them. This is really not
necessary, esp. in the future we may grow information
passed from targets to bpf_iter manager.
The patch refactors the code so target reg_info
becomes static and bpf_iter manager can just take
a reference to it.
Signed-off-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200513180219.2949605-1-yhs@fb.com
Only the tasks belonging to "current" pid namespace
are enumerated.
For task/file target, the bpf program will have access to
struct task_struct *task
u32 fd
struct file *file
where fd/file is an open file for the task.
Signed-off-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Link: https://lore.kernel.org/bpf/20200509175911.2476407-1-yhs@fb.com
Song Liu