| Message ID | afe4deb020b781c76e9df8403a744f88a8725cd2.1575517685.git.ethercflow@gmail.com |
|---|---|
| State | Changes Requested |
| Delegated to: | BPF Maintainers |
| Headers | show |
| Series | bpf: adding get_file_path helper | expand |
On Wed, Dec 04, 2019 at 11:20:35PM -0500, Wenbo Zhang wrote: > > +BPF_CALL_3(bpf_get_file_path, char *, dst, u32, size, int, fd) > +{ > + struct file *f; > + char *p; > + int ret = -EBADF; > + > + /* Ensure we're in user context which is safe for the helper to > + * run. This helper has no business in a kthread. > + */ > + if (unlikely(in_interrupt() || > + current->flags & (PF_KTHREAD | PF_EXITING))) { > + ret = -EPERM; > + goto error; > + } > + > + /* Use fget_raw instead of fget to support O_PATH, and it doesn't > + * have any sleepable code, so it's ok to be here. > + */ > + f = fget_raw(fd); > + if (!f) > + goto error; > + > + /* For unmountable pseudo filesystem, it seems to have no meaning > + * to get their fake paths as they don't have path, and to be no > + * way to validate this function pointer can be always safe to call > + * in the current context. > + */ > + if (f->f_path.dentry->d_op && f->f_path.dentry->d_op->d_dname) { > + ret = -EINVAL; > + fput(f); > + goto error; > + } > + > + /* After filter unmountable pseudo filesytem, d_path won't call > + * dentry->d_op->d_name(), the normally path doesn't have any > + * sleepable code, and despite it uses the current macro to get > + * fs_struct (current->fs), we've already ensured we're in user > + * context, so it's ok to be here. > + */ > + p = d_path(&f->f_path, dst, size); Above 'if's are not enough to make sure that it won't dead lock. Allowing it in tracing_func_proto() means that it's available to kprobe too. Hence deadlock is possible. Please see previous email thread. This helper is safe in tracepoint+bpf only.
> On Sat, Nov 23, 2019 at 05:35:14AM +0000, Al Viro wrote: > > On Fri, Nov 22, 2019 at 09:19:21PM -0800, Alexei Starovoitov wrote: > > > > > hard to tell. It will be run out of bpf prog that attaches to kprobe or > > > tracepoint. What is the concern about locking? > > > d_path() doesn't take any locks and doesn't depend on any locks. Above 'if' > > > checks that plain d_path() is used and not some specilized callback with > > > unknown logic. > > > > It sure as hell does. It might end up taking rename_lock and/or mount_lock > > spinlock components. It'll try not to, but if the first pass ends up with > > seqlock mismatch, it will just grab the spinlock the second time around. > ohh. got it. I missed _or_lock() part in there. > The need_seqretry() logic is tricky. afaics there is no way for the checks > outside of prepend_path() to prevent spin_lock to happen. And adding a flag to > prepend_path() to return early if retry is needed is too ugly. So this helper > won't be safe to be run out of kprobe. But if we allow it for tracepoints only > it should be ok. I think. There are no tracepoints in inner guts of vfs and I > don't think they will ever be. So running in tracepoint->bpf_prog->d_path we > will be sure that rename_lock+mount_lock can be safely spinlocked. Am I missing > something? Hi Alexei, Would you please give me an example of a deadlock condition under kprobe+bpf? I'm not familiar with this detail and want to learn more. > Above 'if's are not enough to make sure that it won't dead lock. > Allowing it in tracing_func_proto() means that it's available to kprobe too. > Hence deadlock is possible. Please see previous email thread. > This helper is safe in tracepoint+bpf only. So I should move it to `tp_prog_prog_func_proto` and `raw_tp_prog_func_prog` right? Is raw tracepoint+bpf safe? Thank you. Alexei Starovoitov <alexei.starovoitov@gmail.com> 于2019年12月5日周四 下午3:19写道： > > On Wed, Dec 04, 2019 at 11:20:35PM -0500, Wenbo Zhang wrote: > > > > +BPF_CALL_3(bpf_get_file_path, char *, dst, u32, size, int, fd) > > +{ > > + struct file *f; > > + char *p; > > + int ret = -EBADF; > > + > > + /* Ensure we're in user context which is safe for the helper to > > + * run. This helper has no business in a kthread. > > + */ > > + if (unlikely(in_interrupt() || > > + current->flags & (PF_KTHREAD | PF_EXITING))) { > > + ret = -EPERM; > > + goto error; > > + } > > + > > + /* Use fget_raw instead of fget to support O_PATH, and it doesn't > > + * have any sleepable code, so it's ok to be here. > > + */ > > + f = fget_raw(fd); > > + if (!f) > > + goto error; > > + > > + /* For unmountable pseudo filesystem, it seems to have no meaning > > + * to get their fake paths as they don't have path, and to be no > > + * way to validate this function pointer can be always safe to call > > + * in the current context. > > + */ > > + if (f->f_path.dentry->d_op && f->f_path.dentry->d_op->d_dname) { > > + ret = -EINVAL; > > + fput(f); > > + goto error; > > + } > > + > > + /* After filter unmountable pseudo filesytem, d_path won't call > > + * dentry->d_op->d_name(), the normally path doesn't have any > > + * sleepable code, and despite it uses the current macro to get > > + * fs_struct (current->fs), we've already ensured we're in user > > + * context, so it's ok to be here. > > + */ > > + p = d_path(&f->f_path, dst, size); > > Above 'if's are not enough to make sure that it won't dead lock. > Allowing it in tracing_func_proto() means that it's available to kprobe too. > Hence deadlock is possible. Please see previous email thread. > This helper is safe in tracepoint+bpf only. >
diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h index dbbcf0b02970..71d9705df120 100644 --- a/include/uapi/linux/bpf.h +++ b/include/uapi/linux/bpf.h @@ -2821,6 +2821,32 @@ union bpf_attr { * Return * On success, the strictly positive length of the string, including * the trailing NUL character. On error, a negative value. + * + * int bpf_get_file_path(char *path, u32 size, int fd) + * Description + * Get **file** atrribute from the current task by *fd*, then call + * **d_path** to get it's absolute path and copy it as string into + * *path* of *size*. Notice the **path** don't support unmountable + * pseudo filesystems as they don't have path (eg: SOCKFS, PIPEFS). + * The *size* must be strictly positive. On success, the helper + * makes sure that the *path* is NUL-terminated, and the buffer + * could be: + * - a regular full path (include mountable fs eg: /proc, /sys) + * - a regular full path with "(deleted)" at the end. + * On failure, it is filled with zeroes. + * Return + * On success, returns the length of the copied string INCLUDING + * the trailing NUL. + * + * On failure, the returned value is one of the following: + * + * **-EPERM** if no permission to get the path (eg: in irq ctx). + * + * **-EBADF** if *fd* is invalid. + * + * **-EINVAL** if *fd* corresponds to a unmountable pseudo fs + * + * **-ENAMETOOLONG** if full path is longer than *size* */ #define __BPF_FUNC_MAPPER(FN) \ FN(unspec), \ @@ -2938,7 +2964,8 @@ union bpf_attr { FN(probe_read_user), \ FN(probe_read_kernel), \ FN(probe_read_user_str), \ - FN(probe_read_kernel_str), + FN(probe_read_kernel_str), \ + FN(get_file_path), /* integer value in 'imm' field of BPF_CALL instruction selects which helper * function eBPF program intends to call diff --git a/kernel/trace/bpf_trace.c b/kernel/trace/bpf_trace.c index ffc91d4935ac..16df8163d681 100644 --- a/kernel/trace/bpf_trace.c +++ b/kernel/trace/bpf_trace.c @@ -762,6 +762,72 @@ static const struct bpf_func_proto bpf_send_signal_proto = { .arg1_type = ARG_ANYTHING, }; +BPF_CALL_3(bpf_get_file_path, char *, dst, u32, size, int, fd) +{ + struct file *f; + char *p; + int ret = -EBADF; + + /* Ensure we're in user context which is safe for the helper to + * run. This helper has no business in a kthread. + */ + if (unlikely(in_interrupt() || + current->flags & (PF_KTHREAD | PF_EXITING))) { + ret = -EPERM; + goto error; + } + + /* Use fget_raw instead of fget to support O_PATH, and it doesn't + * have any sleepable code, so it's ok to be here. + */ + f = fget_raw(fd); + if (!f) + goto error; + + /* For unmountable pseudo filesystem, it seems to have no meaning + * to get their fake paths as they don't have path, and to be no + * way to validate this function pointer can be always safe to call + * in the current context. + */ + if (f->f_path.dentry->d_op && f->f_path.dentry->d_op->d_dname) { + ret = -EINVAL; + fput(f); + goto error; + } + + /* After filter unmountable pseudo filesytem, d_path won't call + * dentry->d_op->d_name(), the normally path doesn't have any + * sleepable code, and despite it uses the current macro to get + * fs_struct (current->fs), we've already ensured we're in user + * context, so it's ok to be here. + */ + p = d_path(&f->f_path, dst, size); + if (IS_ERR(p)) { + ret = PTR_ERR(p); + fput(f); + goto error; + } + + ret = strlen(p); + memmove(dst, p, ret); + dst[ret++] = '\0'; + fput(f); + return ret; + +error: + memset(dst, '0', size); + return ret; +} + +static const struct bpf_func_proto bpf_get_file_path_proto = { + .func = bpf_get_file_path, + .gpl_only = true, + .ret_type = RET_INTEGER, + .arg1_type = ARG_PTR_TO_UNINIT_MEM, + .arg2_type = ARG_CONST_SIZE, + .arg3_type = ARG_ANYTHING, +}; + static const struct bpf_func_proto * tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) { @@ -822,6 +888,8 @@ tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) #endif case BPF_FUNC_send_signal: return &bpf_send_signal_proto; + case BPF_FUNC_get_file_path: + return &bpf_get_file_path_proto; default: return NULL; } diff --git a/tools/include/uapi/linux/bpf.h b/tools/include/uapi/linux/bpf.h index dbbcf0b02970..71d9705df120 100644 --- a/tools/include/uapi/linux/bpf.h +++ b/tools/include/uapi/linux/bpf.h @@ -2821,6 +2821,32 @@ union bpf_attr { * Return * On success, the strictly positive length of the string, including * the trailing NUL character. On error, a negative value. + * + * int bpf_get_file_path(char *path, u32 size, int fd) + * Description + * Get **file** atrribute from the current task by *fd*, then call + * **d_path** to get it's absolute path and copy it as string into + * *path* of *size*. Notice the **path** don't support unmountable + * pseudo filesystems as they don't have path (eg: SOCKFS, PIPEFS). + * The *size* must be strictly positive. On success, the helper + * makes sure that the *path* is NUL-terminated, and the buffer + * could be: + * - a regular full path (include mountable fs eg: /proc, /sys) + * - a regular full path with "(deleted)" at the end. + * On failure, it is filled with zeroes. + * Return + * On success, returns the length of the copied string INCLUDING + * the trailing NUL. + * + * On failure, the returned value is one of the following: + * + * **-EPERM** if no permission to get the path (eg: in irq ctx). + * + * **-EBADF** if *fd* is invalid. + * + * **-EINVAL** if *fd* corresponds to a unmountable pseudo fs + * + * **-ENAMETOOLONG** if full path is longer than *size* */ #define __BPF_FUNC_MAPPER(FN) \ FN(unspec), \ @@ -2938,7 +2964,8 @@ union bpf_attr { FN(probe_read_user), \ FN(probe_read_kernel), \ FN(probe_read_user_str), \ - FN(probe_read_kernel_str), + FN(probe_read_kernel_str), \ + FN(get_file_path), /* integer value in 'imm' field of BPF_CALL instruction selects which helper * function eBPF program intends to call
When people want to identify which file system files are being opened, read, and written to, they can use this helper with file descriptor as input to achieve this goal. Other pseudo filesystems are also supported. This requirement is mainly discussed here: https://github.com/iovisor/bcc/issues/237 v10->v11: addressed Al and Alexei's feedback - fix missing fput() v9->v10: addressed Andrii's feedback - send this patch together with the patch selftests as one patch series v8->v9: - format helper description v7->v8: addressed Alexei's feedback - use fget_raw instead of fdget_raw, as fdget_raw is only used inside fs/ - ensure we're in user context which is safe fot the help to run - filter unmountable pseudo filesystem, because they don't have real path - supplement the description of this helper function v6->v7: - fix missing signed-off-by line v5->v6: addressed Andrii's feedback - avoid unnecessary goto end by having two explicit returns v4->v5: addressed Andrii and Daniel's feedback - rename bpf_fd2path to bpf_get_file_path to be consistent with other helper's names - when fdget_raw fails, set ret to -EBADF instead of -EINVAL - remove fdput from fdget_raw's error path - use IS_ERR instead of IS_ERR_OR_NULL as d_path ether returns a pointer into the buffer or an error code if the path was too long - modify the normal path's return value to return copied string length including NUL - update this helper description's Return bits. v3->v4: addressed Daniel's feedback - fix missing fdput() - move fd2path from kernel/bpf/trace.c to kernel/trace/bpf_trace.c - move fd2path's test code to another patch - add comment to explain why use fdget_raw instead of fdget v2->v3: addressed Yonghong's feedback - remove unnecessary LOCKDOWN_BPF_READ - refactor error handling section for enhanced readability - provide a test case in tools/testing/selftests/bpf v1->v2: addressed Daniel's feedback - fix backward compatibility - add this helper description - fix signed-off name Signed-off-by: Wenbo Zhang <ethercflow@gmail.com> --- include/uapi/linux/bpf.h | 29 ++++++++++++++- kernel/trace/bpf_trace.c | 68 ++++++++++++++++++++++++++++++++++ tools/include/uapi/linux/bpf.h | 29 ++++++++++++++- 3 files changed, 124 insertions(+), 2 deletions(-)