On 9/1/21 10:52 AM, 王贇 wrote: ]> On 2021/9/1 下午5:47, Pavel Begunkov wrote: >> On 9/1/21 10:39 AM, 王贇 wrote: >>> We observed panic: >>> BUG: kernel NULL pointer dereference, address:0000000000000028 >>> [skip] >>> Oops: 0000 [#1] SMP PTI >>> CPU: 1 PID: 737 Comm: a.out Not tainted 5.14.0+ #58 >>> Hardware name: Red Hat KVM, BIOS 0.5.1 01/01/2011 >>> RIP: 0010:vfs_fadvise+0x1e/0x80 >>> [skip] >>> Call Trace: >>> ? tctx_task_work+0x111/0x2a0 >>> io_issue_sqe+0x524/0x1b90 >> >> Most likely it was fixed yesterday. Can you try? >> https://git.kernel.dk/cgit/linux-block/log/?h=for-5.15/io_uring >> >> Or these two patches in particular >> >> https://git.kernel.dk/cgit/linux-block/commit/?h=for-5.15/io_uring&id=c6d3d9cbd659de8f2176b4e4721149c88ac096d4 >> https://git.kernel.dk/cgit/linux-block/commit/?h=for-5.15/io_uring&id=b8ce1b9d25ccf81e1bbabd45b963ed98b2222df8 > > Yup, it no longer panic :-) awesome, thanks > > Regards, > Michael Wang > >> >>> This is caused by io_wq_submit_work() calling io_issue_sqe() >>> on a failed fadvise request, and the io_init_req() return error >>> before initialize the file for it, lead into the panic when >>> vfs_fadvise() try to access 'req->file'. >>> >>> This patch add the missing check & handle for failed request >>> before calling io_issue_sqe(). >>> >>> Signed-off-by: Michael Wang <yun.wang@xxxxxxxxxxxxxxxxx> >>> --- >>> fs/io_uring.c | 8 ++++++-- >>> 1 file changed, 6 insertions(+), 2 deletions(-) >>> >>> diff --git a/fs/io_uring.c b/fs/io_uring.c >>> index 6f35b12..bfec7bf 100644 >>> --- a/fs/io_uring.c >>> +++ b/fs/io_uring.c >>> @@ -2214,7 +2214,8 @@ static void io_req_task_submit(struct io_kiocb *req, bool *locked) >>> >>> io_tw_lock(ctx, locked); >>> /* req->task == current here, checking PF_EXITING is safe */ >>> - if (likely(!(req->task->flags & PF_EXITING))) >>> + if (likely(!(req->task->flags & PF_EXITING) && >>> + !(req->flags & REQ_F_FAIL))) >>> __io_queue_sqe(req); >>> else >>> io_req_complete_failed(req, -EFAULT); >>> @@ -6704,7 +6705,10 @@ static void io_wq_submit_work(struct io_wq_work *work) >>> >>> if (!ret) { >>> do { >>> - ret = io_issue_sqe(req, 0); >>> + if (likely(!(req->flags & REQ_F_FAIL))) >>> + ret = io_issue_sqe(req, 0); >>> + else >>> + io_req_complete_failed(req, -EFAULT); >>> /* >>> * We can get EAGAIN for polled IO even though we're >>> * forcing a sync submission from here, since we can't >>> >> -- Pavel Begunkov
