linux/kernel/bpf/task_iter.c

848 lines
20 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0-only
/* Copyright (c) 2020 Facebook */
#include <linux/init.h>
#include <linux/namei.h>
#include <linux/pid_namespace.h>
#include <linux/fs.h>
#include <linux/fdtable.h>
#include <linux/filter.h>
#include <linux/btf_ids.h>
#include "mmap_unlock_work.h"
static const char * const iter_task_type_names[] = {
"ALL",
"TID",
"PID",
};
struct bpf_iter_seq_task_common {
struct pid_namespace *ns;
enum bpf_iter_task_type type;
u32 pid;
u32 pid_visiting;
};
struct bpf_iter_seq_task_info {
/* The first field must be struct bpf_iter_seq_task_common.
* this is assumed by {init, fini}_seq_pidns() callback functions.
*/
struct bpf_iter_seq_task_common common;
u32 tid;
};
static struct task_struct *task_group_seq_get_next(struct bpf_iter_seq_task_common *common,
u32 *tid,
bool skip_if_dup_files)
{
struct task_struct *task;
struct pid *pid;
u32 saved_tid;
if (!*tid) {
/* The first time, the iterator calls this function. */
pid = find_pid_ns(common->pid, common->ns);
task = get_pid_task(pid, PIDTYPE_TGID);
if (!task)
return NULL;
*tid = common->pid;
common->pid_visiting = common->pid;
return task;
}
/* If the control returns to user space and comes back to the
* kernel again, *tid and common->pid_visiting should be the
* same for task_seq_start() to pick up the correct task.
*/
if (*tid == common->pid_visiting) {
pid = find_pid_ns(common->pid_visiting, common->ns);
task = get_pid_task(pid, PIDTYPE_PID);
return task;
}
task = find_task_by_pid_ns(common->pid_visiting, common->ns);
if (!task)
return NULL;
retry:
task = next_thread(task);
saved_tid = *tid;
*tid = __task_pid_nr_ns(task, PIDTYPE_PID, common->ns);
if (!*tid || *tid == common->pid) {
/* Run out of tasks of a process. The tasks of a
* thread_group are linked as circular linked list.
*/
*tid = saved_tid;
return NULL;
}
common->pid_visiting = *tid;
if (skip_if_dup_files && task->files == task->group_leader->files)
goto retry;
get_task_struct(task);
return task;
}
static struct task_struct *task_seq_get_next(struct bpf_iter_seq_task_common *common,
u32 *tid,
bool skip_if_dup_files)
{
struct task_struct *task = NULL;
struct pid *pid;
if (common->type == BPF_TASK_ITER_TID) {
if (*tid && *tid != common->pid)
return NULL;
rcu_read_lock();
pid = find_pid_ns(common->pid, common->ns);
if (pid) {
task = get_pid_task(pid, PIDTYPE_TGID);
*tid = common->pid;
}
rcu_read_unlock();
return task;
}
if (common->type == BPF_TASK_ITER_TGID) {
rcu_read_lock();
task = task_group_seq_get_next(common, tid, skip_if_dup_files);
rcu_read_unlock();
return task;
}
rcu_read_lock();
retry:
pid = find_ge_pid(*tid, common->ns);
if (pid) {
*tid = pid_nr_ns(pid, common->ns);
task = get_pid_task(pid, PIDTYPE_PID);
if (!task) {
++*tid;
goto retry;
} else if (skip_if_dup_files && !thread_group_leader(task) &&
task->files == task->group_leader->files) {
put_task_struct(task);
task = NULL;
++*tid;
goto retry;
}
}
rcu_read_unlock();
return task;
}
static void *task_seq_start(struct seq_file *seq, loff_t *pos)
{
struct bpf_iter_seq_task_info *info = seq->private;
struct task_struct *task;
task = task_seq_get_next(&info->common, &info->tid, false);
if (!task)
return NULL;
if (*pos == 0)
++*pos;
return task;
}
static void *task_seq_next(struct seq_file *seq, void *v, loff_t *pos)
{
struct bpf_iter_seq_task_info *info = seq->private;
struct task_struct *task;
++*pos;
++info->tid;
put_task_struct((struct task_struct *)v);
task = task_seq_get_next(&info->common, &info->tid, false);
if (!task)
return NULL;
return task;
}
struct bpf_iter__task {
__bpf_md_ptr(struct bpf_iter_meta *, meta);
__bpf_md_ptr(struct task_struct *, task);
};
DEFINE_BPF_ITER_FUNC(task, struct bpf_iter_meta *meta, struct task_struct *task)
static int __task_seq_show(struct seq_file *seq, struct task_struct *task,
bool in_stop)
{
struct bpf_iter_meta meta;
struct bpf_iter__task ctx;
struct bpf_prog *prog;
meta.seq = seq;
prog = bpf_iter_get_info(&meta, in_stop);
if (!prog)
return 0;
ctx.meta = &meta;
ctx.task = task;
return bpf_iter_run_prog(prog, &ctx);
}
static int task_seq_show(struct seq_file *seq, void *v)
{
return __task_seq_show(seq, v, false);
}
static void task_seq_stop(struct seq_file *seq, void *v)
{
if (!v)
(void)__task_seq_show(seq, v, true);
else
put_task_struct((struct task_struct *)v);
}
static int bpf_iter_attach_task(struct bpf_prog *prog,
union bpf_iter_link_info *linfo,
struct bpf_iter_aux_info *aux)
{
unsigned int flags;
struct pid *pid;
pid_t tgid;
if ((!!linfo->task.tid + !!linfo->task.pid + !!linfo->task.pid_fd) > 1)
return -EINVAL;
aux->task.type = BPF_TASK_ITER_ALL;
if (linfo->task.tid != 0) {
aux->task.type = BPF_TASK_ITER_TID;
aux->task.pid = linfo->task.tid;
}
if (linfo->task.pid != 0) {
aux->task.type = BPF_TASK_ITER_TGID;
aux->task.pid = linfo->task.pid;
}
if (linfo->task.pid_fd != 0) {
aux->task.type = BPF_TASK_ITER_TGID;
pid = pidfd_get_pid(linfo->task.pid_fd, &flags);
if (IS_ERR(pid))
return PTR_ERR(pid);
tgid = pid_nr_ns(pid, task_active_pid_ns(current));
aux->task.pid = tgid;
put_pid(pid);
}
return 0;
}
static const struct seq_operations task_seq_ops = {
.start = task_seq_start,
.next = task_seq_next,
.stop = task_seq_stop,
.show = task_seq_show,
};
struct bpf_iter_seq_task_file_info {
/* The first field must be struct bpf_iter_seq_task_common.
* this is assumed by {init, fini}_seq_pidns() callback functions.
*/
struct bpf_iter_seq_task_common common;
struct task_struct *task;
u32 tid;
u32 fd;
};
static struct file *
task_file_seq_get_next(struct bpf_iter_seq_task_file_info *info)
{
u32 saved_tid = info->tid;
struct task_struct *curr_task;
bpf/task_iter: In task_file_seq_get_next use task_lookup_next_fd_rcu When discussing[1] exec and posix file locks it was realized that none of the callers of get_files_struct fundamentally needed to call get_files_struct, and that by switching them to helper functions instead it will both simplify their code and remove unnecessary increments of files_struct.count. Those unnecessary increments can result in exec unnecessarily unsharing files_struct which breaking posix locks, and it can result in fget_light having to fallback to fget reducing system performance. Using task_lookup_next_fd_rcu simplifies task_file_seq_get_next, by moving the checking for the maximum file descritor into the generic code, and by remvoing the need for capturing and releasing a reference on files_struct. As the reference count of files_struct no longer needs to be maintained bpf_iter_seq_task_file_info can have it's files member removed and task_file_seq_get_next no longer needs it's fstruct argument. The curr_fd local variable does need to become unsigned to be used with fnext_task. As curr_fd is assigned from and assigned a u32 making curr_fd an unsigned int won't cause problems and might prevent them. [1] https://lkml.kernel.org/r/20180915160423.GA31461@redhat.com Suggested-by: Oleg Nesterov <oleg@redhat.com> v1: https://lkml.kernel.org/r/20200817220425.9389-11-ebiederm@xmission.com Link: https://lkml.kernel.org/r/20201120231441.29911-16-ebiederm@xmission.com Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
2020-11-20 23:14:33 +00:00
unsigned int curr_fd = info->fd;
/* If this function returns a non-NULL file object,
bpf/task_iter: In task_file_seq_get_next use task_lookup_next_fd_rcu When discussing[1] exec and posix file locks it was realized that none of the callers of get_files_struct fundamentally needed to call get_files_struct, and that by switching them to helper functions instead it will both simplify their code and remove unnecessary increments of files_struct.count. Those unnecessary increments can result in exec unnecessarily unsharing files_struct which breaking posix locks, and it can result in fget_light having to fallback to fget reducing system performance. Using task_lookup_next_fd_rcu simplifies task_file_seq_get_next, by moving the checking for the maximum file descritor into the generic code, and by remvoing the need for capturing and releasing a reference on files_struct. As the reference count of files_struct no longer needs to be maintained bpf_iter_seq_task_file_info can have it's files member removed and task_file_seq_get_next no longer needs it's fstruct argument. The curr_fd local variable does need to become unsigned to be used with fnext_task. As curr_fd is assigned from and assigned a u32 making curr_fd an unsigned int won't cause problems and might prevent them. [1] https://lkml.kernel.org/r/20180915160423.GA31461@redhat.com Suggested-by: Oleg Nesterov <oleg@redhat.com> v1: https://lkml.kernel.org/r/20200817220425.9389-11-ebiederm@xmission.com Link: https://lkml.kernel.org/r/20201120231441.29911-16-ebiederm@xmission.com Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
2020-11-20 23:14:33 +00:00
* it held a reference to the task/file.
* Otherwise, it does not hold any reference.
*/
again:
if (info->task) {
curr_task = info->task;
curr_fd = info->fd;
} else {
curr_task = task_seq_get_next(&info->common, &info->tid, true);
Merge git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf Daniel Borkmann says: ==================== pull-request: bpf 2020-12-28 The following pull-request contains BPF updates for your *net* tree. There is a small merge conflict between bpf tree commit 69ca310f3416 ("bpf: Save correct stopping point in file seq iteration") and net tree commit 66ed594409a1 ("bpf/task_iter: In task_file_seq_get_next use task_lookup_next_fd_rcu"). The get_files_struct() does not exist anymore in net, so take the hunk in HEAD and add the `info->tid = curr_tid` to the error path: [...] curr_task = task_seq_get_next(ns, &curr_tid, true); if (!curr_task) { info->task = NULL; info->tid = curr_tid; return NULL; } /* set info->task and info->tid */ [...] We've added 10 non-merge commits during the last 9 day(s) which contain a total of 11 files changed, 75 insertions(+), 20 deletions(-). The main changes are: 1) Various AF_XDP fixes such as fill/completion ring leak on failed bind and fixing a race in skb mode's backpressure mechanism, from Magnus Karlsson. 2) Fix latency spikes on lockdep enabled kernels by adding a rescheduling point to BPF hashtab initialization, from Eric Dumazet. 3) Fix a splat in task iterator by saving the correct stopping point in the seq file iteration, from Jonathan Lemon. 4) Fix BPF maps selftest by adding retries in case hashtab returns EBUSY errors on update/deletes, from Andrii Nakryiko. 5) Fix BPF selftest error reporting to something more user friendly if the vmlinux BTF cannot be found, from Kamal Mostafa. ==================== Signed-off-by: David S. Miller <davem@davemloft.net>
2020-12-28 23:20:48 +00:00
if (!curr_task) {
info->task = NULL;
return NULL;
}
/* set info->task */
info->task = curr_task;
if (saved_tid == info->tid)
curr_fd = info->fd;
else
curr_fd = 0;
}
rcu_read_lock();
bpf/task_iter: In task_file_seq_get_next use task_lookup_next_fd_rcu When discussing[1] exec and posix file locks it was realized that none of the callers of get_files_struct fundamentally needed to call get_files_struct, and that by switching them to helper functions instead it will both simplify their code and remove unnecessary increments of files_struct.count. Those unnecessary increments can result in exec unnecessarily unsharing files_struct which breaking posix locks, and it can result in fget_light having to fallback to fget reducing system performance. Using task_lookup_next_fd_rcu simplifies task_file_seq_get_next, by moving the checking for the maximum file descritor into the generic code, and by remvoing the need for capturing and releasing a reference on files_struct. As the reference count of files_struct no longer needs to be maintained bpf_iter_seq_task_file_info can have it's files member removed and task_file_seq_get_next no longer needs it's fstruct argument. The curr_fd local variable does need to become unsigned to be used with fnext_task. As curr_fd is assigned from and assigned a u32 making curr_fd an unsigned int won't cause problems and might prevent them. [1] https://lkml.kernel.org/r/20180915160423.GA31461@redhat.com Suggested-by: Oleg Nesterov <oleg@redhat.com> v1: https://lkml.kernel.org/r/20200817220425.9389-11-ebiederm@xmission.com Link: https://lkml.kernel.org/r/20201120231441.29911-16-ebiederm@xmission.com Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
2020-11-20 23:14:33 +00:00
for (;; curr_fd++) {
struct file *f;
bpf/task_iter: In task_file_seq_get_next use task_lookup_next_fd_rcu When discussing[1] exec and posix file locks it was realized that none of the callers of get_files_struct fundamentally needed to call get_files_struct, and that by switching them to helper functions instead it will both simplify their code and remove unnecessary increments of files_struct.count. Those unnecessary increments can result in exec unnecessarily unsharing files_struct which breaking posix locks, and it can result in fget_light having to fallback to fget reducing system performance. Using task_lookup_next_fd_rcu simplifies task_file_seq_get_next, by moving the checking for the maximum file descritor into the generic code, and by remvoing the need for capturing and releasing a reference on files_struct. As the reference count of files_struct no longer needs to be maintained bpf_iter_seq_task_file_info can have it's files member removed and task_file_seq_get_next no longer needs it's fstruct argument. The curr_fd local variable does need to become unsigned to be used with fnext_task. As curr_fd is assigned from and assigned a u32 making curr_fd an unsigned int won't cause problems and might prevent them. [1] https://lkml.kernel.org/r/20180915160423.GA31461@redhat.com Suggested-by: Oleg Nesterov <oleg@redhat.com> v1: https://lkml.kernel.org/r/20200817220425.9389-11-ebiederm@xmission.com Link: https://lkml.kernel.org/r/20201120231441.29911-16-ebiederm@xmission.com Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
2020-11-20 23:14:33 +00:00
f = task_lookup_next_fd_rcu(curr_task, &curr_fd);
if (!f)
bpf/task_iter: In task_file_seq_get_next use task_lookup_next_fd_rcu When discussing[1] exec and posix file locks it was realized that none of the callers of get_files_struct fundamentally needed to call get_files_struct, and that by switching them to helper functions instead it will both simplify their code and remove unnecessary increments of files_struct.count. Those unnecessary increments can result in exec unnecessarily unsharing files_struct which breaking posix locks, and it can result in fget_light having to fallback to fget reducing system performance. Using task_lookup_next_fd_rcu simplifies task_file_seq_get_next, by moving the checking for the maximum file descritor into the generic code, and by remvoing the need for capturing and releasing a reference on files_struct. As the reference count of files_struct no longer needs to be maintained bpf_iter_seq_task_file_info can have it's files member removed and task_file_seq_get_next no longer needs it's fstruct argument. The curr_fd local variable does need to become unsigned to be used with fnext_task. As curr_fd is assigned from and assigned a u32 making curr_fd an unsigned int won't cause problems and might prevent them. [1] https://lkml.kernel.org/r/20180915160423.GA31461@redhat.com Suggested-by: Oleg Nesterov <oleg@redhat.com> v1: https://lkml.kernel.org/r/20200817220425.9389-11-ebiederm@xmission.com Link: https://lkml.kernel.org/r/20201120231441.29911-16-ebiederm@xmission.com Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
2020-11-20 23:14:33 +00:00
break;
bpf: Use get_file_rcu() instead of get_file() for task_file iterator 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 eaaacd23910f ("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: eaaacd23910f ("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
2020-08-17 17:42:14 +00:00
if (!get_file_rcu(f))
continue;
/* set info->fd */
info->fd = curr_fd;
rcu_read_unlock();
return f;
}
/* the current task is done, go to the next task */
rcu_read_unlock();
put_task_struct(curr_task);
if (info->common.type == BPF_TASK_ITER_TID) {
info->task = NULL;
return NULL;
}
info->task = NULL;
info->fd = 0;
saved_tid = ++(info->tid);
goto again;
}
static void *task_file_seq_start(struct seq_file *seq, loff_t *pos)
{
struct bpf_iter_seq_task_file_info *info = seq->private;
struct file *file;
info->task = NULL;
file = task_file_seq_get_next(info);
if (file && *pos == 0)
++*pos;
return file;
}
static void *task_file_seq_next(struct seq_file *seq, void *v, loff_t *pos)
{
struct bpf_iter_seq_task_file_info *info = seq->private;
++*pos;
++info->fd;
fput((struct file *)v);
return task_file_seq_get_next(info);
}
struct bpf_iter__task_file {
__bpf_md_ptr(struct bpf_iter_meta *, meta);
__bpf_md_ptr(struct task_struct *, task);
u32 fd __aligned(8);
__bpf_md_ptr(struct file *, file);
};
DEFINE_BPF_ITER_FUNC(task_file, struct bpf_iter_meta *meta,
struct task_struct *task, u32 fd,
struct file *file)
static int __task_file_seq_show(struct seq_file *seq, struct file *file,
bool in_stop)
{
struct bpf_iter_seq_task_file_info *info = seq->private;
struct bpf_iter__task_file ctx;
struct bpf_iter_meta meta;
struct bpf_prog *prog;
meta.seq = seq;
prog = bpf_iter_get_info(&meta, in_stop);
if (!prog)
return 0;
ctx.meta = &meta;
ctx.task = info->task;
ctx.fd = info->fd;
ctx.file = file;
return bpf_iter_run_prog(prog, &ctx);
}
static int task_file_seq_show(struct seq_file *seq, void *v)
{
return __task_file_seq_show(seq, v, false);
}
static void task_file_seq_stop(struct seq_file *seq, void *v)
{
struct bpf_iter_seq_task_file_info *info = seq->private;
if (!v) {
(void)__task_file_seq_show(seq, v, true);
} else {
fput((struct file *)v);
put_task_struct(info->task);
info->task = NULL;
}
}
static int init_seq_pidns(void *priv_data, struct bpf_iter_aux_info *aux)
{
struct bpf_iter_seq_task_common *common = priv_data;
common->ns = get_pid_ns(task_active_pid_ns(current));
common->type = aux->task.type;
common->pid = aux->task.pid;
return 0;
}
static void fini_seq_pidns(void *priv_data)
{
struct bpf_iter_seq_task_common *common = priv_data;
put_pid_ns(common->ns);
}
static const struct seq_operations task_file_seq_ops = {
.start = task_file_seq_start,
.next = task_file_seq_next,
.stop = task_file_seq_stop,
.show = task_file_seq_show,
};
struct bpf_iter_seq_task_vma_info {
/* The first field must be struct bpf_iter_seq_task_common.
* this is assumed by {init, fini}_seq_pidns() callback functions.
*/
struct bpf_iter_seq_task_common common;
struct task_struct *task;
struct mm_struct *mm;
struct vm_area_struct *vma;
u32 tid;
unsigned long prev_vm_start;
unsigned long prev_vm_end;
};
enum bpf_task_vma_iter_find_op {
task_vma_iter_first_vma, /* use find_vma() with addr 0 */
task_vma_iter_next_vma, /* use vma_next() with curr_vma */
task_vma_iter_find_vma, /* use find_vma() to find next vma */
};
static struct vm_area_struct *
task_vma_seq_get_next(struct bpf_iter_seq_task_vma_info *info)
{
enum bpf_task_vma_iter_find_op op;
struct vm_area_struct *curr_vma;
struct task_struct *curr_task;
struct mm_struct *curr_mm;
u32 saved_tid = info->tid;
/* If this function returns a non-NULL vma, it holds a reference to
* the task_struct, holds a refcount on mm->mm_users, and holds
* read lock on vma->mm->mmap_lock.
* If this function returns NULL, it does not hold any reference or
* lock.
*/
if (info->task) {
curr_task = info->task;
curr_vma = info->vma;
curr_mm = info->mm;
/* In case of lock contention, drop mmap_lock to unblock
* the writer.
*
* After relock, call find(mm, prev_vm_end - 1) to find
* new vma to process.
*
* +------+------+-----------+
* | VMA1 | VMA2 | VMA3 |
* +------+------+-----------+
* | | | |
* 4k 8k 16k 400k
*
* For example, curr_vma == VMA2. Before unlock, we set
*
* prev_vm_start = 8k
* prev_vm_end = 16k
*
* There are a few cases:
*
* 1) VMA2 is freed, but VMA3 exists.
*
* find_vma() will return VMA3, just process VMA3.
*
* 2) VMA2 still exists.
*
* find_vma() will return VMA2, process VMA2->next.
*
* 3) no more vma in this mm.
*
* Process the next task.
*
* 4) find_vma() returns a different vma, VMA2'.
*
* 4.1) If VMA2 covers same range as VMA2', skip VMA2',
* because we already covered the range;
* 4.2) VMA2 and VMA2' covers different ranges, process
* VMA2'.
*/
if (mmap_lock_is_contended(curr_mm)) {
info->prev_vm_start = curr_vma->vm_start;
info->prev_vm_end = curr_vma->vm_end;
op = task_vma_iter_find_vma;
mmap_read_unlock(curr_mm);
if (mmap_read_lock_killable(curr_mm)) {
mmput(curr_mm);
goto finish;
}
} else {
op = task_vma_iter_next_vma;
}
} else {
again:
curr_task = task_seq_get_next(&info->common, &info->tid, true);
if (!curr_task) {
info->tid++;
goto finish;
}
if (saved_tid != info->tid) {
/* new task, process the first vma */
op = task_vma_iter_first_vma;
} else {
/* Found the same tid, which means the user space
* finished data in previous buffer and read more.
* We dropped mmap_lock before returning to user
* space, so it is necessary to use find_vma() to
* find the next vma to process.
*/
op = task_vma_iter_find_vma;
}
curr_mm = get_task_mm(curr_task);
if (!curr_mm)
goto next_task;
if (mmap_read_lock_killable(curr_mm)) {
mmput(curr_mm);
goto finish;
}
}
switch (op) {
case task_vma_iter_first_vma:
curr_vma = find_vma(curr_mm, 0);
break;
case task_vma_iter_next_vma:
curr_vma = find_vma(curr_mm, curr_vma->vm_end);
break;
case task_vma_iter_find_vma:
/* We dropped mmap_lock so it is necessary to use find_vma
* to find the next vma. This is similar to the mechanism
* in show_smaps_rollup().
*/
curr_vma = find_vma(curr_mm, info->prev_vm_end - 1);
/* case 1) and 4.2) above just use curr_vma */
/* check for case 2) or case 4.1) above */
if (curr_vma &&
curr_vma->vm_start == info->prev_vm_start &&
curr_vma->vm_end == info->prev_vm_end)
curr_vma = find_vma(curr_mm, curr_vma->vm_end);
break;
}
if (!curr_vma) {
/* case 3) above, or case 2) 4.1) with vma->next == NULL */
mmap_read_unlock(curr_mm);
mmput(curr_mm);
goto next_task;
}
info->task = curr_task;
info->vma = curr_vma;
info->mm = curr_mm;
return curr_vma;
next_task:
if (info->common.type == BPF_TASK_ITER_TID)
goto finish;
put_task_struct(curr_task);
info->task = NULL;
info->mm = NULL;
info->tid++;
goto again;
finish:
if (curr_task)
put_task_struct(curr_task);
info->task = NULL;
info->vma = NULL;
info->mm = NULL;
return NULL;
}
static void *task_vma_seq_start(struct seq_file *seq, loff_t *pos)
{
struct bpf_iter_seq_task_vma_info *info = seq->private;
struct vm_area_struct *vma;
vma = task_vma_seq_get_next(info);
if (vma && *pos == 0)
++*pos;
return vma;
}
static void *task_vma_seq_next(struct seq_file *seq, void *v, loff_t *pos)
{
struct bpf_iter_seq_task_vma_info *info = seq->private;
++*pos;
return task_vma_seq_get_next(info);
}
struct bpf_iter__task_vma {
__bpf_md_ptr(struct bpf_iter_meta *, meta);
__bpf_md_ptr(struct task_struct *, task);
__bpf_md_ptr(struct vm_area_struct *, vma);
};
DEFINE_BPF_ITER_FUNC(task_vma, struct bpf_iter_meta *meta,
struct task_struct *task, struct vm_area_struct *vma)
static int __task_vma_seq_show(struct seq_file *seq, bool in_stop)
{
struct bpf_iter_seq_task_vma_info *info = seq->private;
struct bpf_iter__task_vma ctx;
struct bpf_iter_meta meta;
struct bpf_prog *prog;
meta.seq = seq;
prog = bpf_iter_get_info(&meta, in_stop);
if (!prog)
return 0;
ctx.meta = &meta;
ctx.task = info->task;
ctx.vma = info->vma;
return bpf_iter_run_prog(prog, &ctx);
}
static int task_vma_seq_show(struct seq_file *seq, void *v)
{
return __task_vma_seq_show(seq, false);
}
static void task_vma_seq_stop(struct seq_file *seq, void *v)
{
struct bpf_iter_seq_task_vma_info *info = seq->private;
if (!v) {
(void)__task_vma_seq_show(seq, true);
} else {
/* info->vma has not been seen by the BPF program. If the
* user space reads more, task_vma_seq_get_next should
* return this vma again. Set prev_vm_start to ~0UL,
* so that we don't skip the vma returned by the next
* find_vma() (case task_vma_iter_find_vma in
* task_vma_seq_get_next()).
*/
info->prev_vm_start = ~0UL;
info->prev_vm_end = info->vma->vm_end;
mmap_read_unlock(info->mm);
mmput(info->mm);
info->mm = NULL;
put_task_struct(info->task);
info->task = NULL;
}
}
static const struct seq_operations task_vma_seq_ops = {
.start = task_vma_seq_start,
.next = task_vma_seq_next,
.stop = task_vma_seq_stop,
.show = task_vma_seq_show,
};
static const struct bpf_iter_seq_info task_seq_info = {
.seq_ops = &task_seq_ops,
.init_seq_private = init_seq_pidns,
.fini_seq_private = fini_seq_pidns,
.seq_priv_size = sizeof(struct bpf_iter_seq_task_info),
};
static int bpf_iter_fill_link_info(const struct bpf_iter_aux_info *aux, struct bpf_link_info *info)
{
switch (aux->task.type) {
case BPF_TASK_ITER_TID:
info->iter.task.tid = aux->task.pid;
break;
case BPF_TASK_ITER_TGID:
info->iter.task.pid = aux->task.pid;
break;
default:
break;
}
return 0;
}
static void bpf_iter_task_show_fdinfo(const struct bpf_iter_aux_info *aux, struct seq_file *seq)
{
seq_printf(seq, "task_type:\t%s\n", iter_task_type_names[aux->task.type]);
if (aux->task.type == BPF_TASK_ITER_TID)
seq_printf(seq, "tid:\t%u\n", aux->task.pid);
else if (aux->task.type == BPF_TASK_ITER_TGID)
seq_printf(seq, "pid:\t%u\n", aux->task.pid);
}
static struct bpf_iter_reg task_reg_info = {
.target = "task",
.attach_target = bpf_iter_attach_task,
bpf: Permit cond_resched for some iterators Commit e679654a704e ("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
2020-10-28 06:10:54 +00:00
.feature = BPF_ITER_RESCHED,
.ctx_arg_info_size = 1,
.ctx_arg_info = {
{ offsetof(struct bpf_iter__task, task),
PTR_TO_BTF_ID_OR_NULL },
},
.seq_info = &task_seq_info,
.fill_link_info = bpf_iter_fill_link_info,
.show_fdinfo = bpf_iter_task_show_fdinfo,
};
static const struct bpf_iter_seq_info task_file_seq_info = {
.seq_ops = &task_file_seq_ops,
.init_seq_private = init_seq_pidns,
.fini_seq_private = fini_seq_pidns,
.seq_priv_size = sizeof(struct bpf_iter_seq_task_file_info),
};
static struct bpf_iter_reg task_file_reg_info = {
.target = "task_file",
.attach_target = bpf_iter_attach_task,
bpf: Permit cond_resched for some iterators Commit e679654a704e ("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
2020-10-28 06:10:54 +00:00
.feature = BPF_ITER_RESCHED,
.ctx_arg_info_size = 2,
.ctx_arg_info = {
{ offsetof(struct bpf_iter__task_file, task),
PTR_TO_BTF_ID_OR_NULL },
{ offsetof(struct bpf_iter__task_file, file),
PTR_TO_BTF_ID_OR_NULL },
},
.seq_info = &task_file_seq_info,
.fill_link_info = bpf_iter_fill_link_info,
.show_fdinfo = bpf_iter_task_show_fdinfo,
};
static const struct bpf_iter_seq_info task_vma_seq_info = {
.seq_ops = &task_vma_seq_ops,
.init_seq_private = init_seq_pidns,
.fini_seq_private = fini_seq_pidns,
.seq_priv_size = sizeof(struct bpf_iter_seq_task_vma_info),
};
static struct bpf_iter_reg task_vma_reg_info = {
.target = "task_vma",
.attach_target = bpf_iter_attach_task,
.feature = BPF_ITER_RESCHED,
.ctx_arg_info_size = 2,
.ctx_arg_info = {
{ offsetof(struct bpf_iter__task_vma, task),
PTR_TO_BTF_ID_OR_NULL },
{ offsetof(struct bpf_iter__task_vma, vma),
PTR_TO_BTF_ID_OR_NULL },
},
.seq_info = &task_vma_seq_info,
.fill_link_info = bpf_iter_fill_link_info,
.show_fdinfo = bpf_iter_task_show_fdinfo,
};
BPF_CALL_5(bpf_find_vma, struct task_struct *, task, u64, start,
bpf_callback_t, callback_fn, void *, callback_ctx, u64, flags)
{
struct mmap_unlock_irq_work *work = NULL;
struct vm_area_struct *vma;
bool irq_work_busy = false;
struct mm_struct *mm;
int ret = -ENOENT;
if (flags)
return -EINVAL;
if (!task)
return -ENOENT;
mm = task->mm;
if (!mm)
return -ENOENT;
irq_work_busy = bpf_mmap_unlock_get_irq_work(&work);
if (irq_work_busy || !mmap_read_trylock(mm))
return -EBUSY;
vma = find_vma(mm, start);
if (vma && vma->vm_start <= start && vma->vm_end > start) {
callback_fn((u64)(long)task, (u64)(long)vma,
(u64)(long)callback_ctx, 0, 0);
ret = 0;
}
bpf_mmap_unlock_mm(work, mm);
return ret;
}
const struct bpf_func_proto bpf_find_vma_proto = {
.func = bpf_find_vma,
.ret_type = RET_INTEGER,
.arg1_type = ARG_PTR_TO_BTF_ID,
.arg1_btf_id = &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
.arg2_type = ARG_ANYTHING,
.arg3_type = ARG_PTR_TO_FUNC,
.arg4_type = ARG_PTR_TO_STACK_OR_NULL,
.arg5_type = ARG_ANYTHING,
};
DEFINE_PER_CPU(struct mmap_unlock_irq_work, mmap_unlock_work);
static void do_mmap_read_unlock(struct irq_work *entry)
{
struct mmap_unlock_irq_work *work;
if (WARN_ON_ONCE(IS_ENABLED(CONFIG_PREEMPT_RT)))
return;
work = container_of(entry, struct mmap_unlock_irq_work, irq_work);
mmap_read_unlock_non_owner(work->mm);
}
static int __init task_iter_init(void)
{
struct mmap_unlock_irq_work *work;
int ret, cpu;
for_each_possible_cpu(cpu) {
work = per_cpu_ptr(&mmap_unlock_work, cpu);
init_irq_work(&work->irq_work, do_mmap_read_unlock);
}
task_reg_info.ctx_arg_info[0].btf_id = btf_tracing_ids[BTF_TRACING_TYPE_TASK];
ret = bpf_iter_reg_target(&task_reg_info);
if (ret)
return ret;
task_file_reg_info.ctx_arg_info[0].btf_id = btf_tracing_ids[BTF_TRACING_TYPE_TASK];
task_file_reg_info.ctx_arg_info[1].btf_id = btf_tracing_ids[BTF_TRACING_TYPE_FILE];
ret = bpf_iter_reg_target(&task_file_reg_info);
if (ret)
return ret;
task_vma_reg_info.ctx_arg_info[0].btf_id = btf_tracing_ids[BTF_TRACING_TYPE_TASK];
task_vma_reg_info.ctx_arg_info[1].btf_id = btf_tracing_ids[BTF_TRACING_TYPE_VMA];
return bpf_iter_reg_target(&task_vma_reg_info);
}
late_initcall(task_iter_init);