| Message ID | 1582627315-21123-1-git-send-email-macpaul.lin@mediatek.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [v3] usb: gadget: f_fs: try to fix AIO issue under ARM 64 bit TAGGED mode | expand |
On Tue, 2020-02-25 at 18:41 +0800, Macpaul Lin wrote: > This issue was found when adbd trying to open functionfs with AIO mode. > Usually, we need to set "setprop sys.usb.ffs.aio_compat 0" to enable > adbd with AIO mode on Android. > > When adbd is opening functionfs, it will try to read 24 bytes at the > first read I/O control. If this reading has been failed, adbd will > try to send FUNCTIONFS_CLEAR_HALT to functionfs. When adbd is in AIO > mode, functionfs will be acted with asyncronized I/O path. After the > successful read transfer has been completed by gadget hardware, the > following series of functions will be called. > ffs_epfile_async_io_complete() -> ffs_user_copy_worker() -> > copy_to_iter() -> _copy_to_iter() -> copyout() -> > iterate_and_advance() -> iterate_iovec() > > Adding debug trace to these functions, it has been found that in > copyout(), access_ok() will check if the user space address is valid > to write. However if CONFIG_ARM64_TAGGED_ADDR_ABI is enabled, adbd > always passes user space address start with "0x3C" to gadget's AIO > blocks. This tagged address will cause access_ok() check always fail. > Which causes later calculation in iterate_iovec() turn zero. > Copyout() won't copy data to userspace since the length to be copied > "v.iov_len" will be zero. Finally leads ffs_copy_to_iter() always return > -EFAULT, causes adbd cannot open functionfs and send > FUNCTIONFS_CLEAR_HALT. > > Signed-off-by: Macpaul Lin <macpaul.lin@mediatek.com> > --- > Changes for v3: > - Fix misspelling in commit message. Could you say "thank you" to Peter for his comment and add "Cc: Peter Chen <peter.chen@nxp.com>" to this patch, please? > > Changes for v2: > - Fix build error for 32-bit load. An #if defined(CONFIG_ARM64) still need > for avoiding undeclared defines. > > drivers/usb/gadget/function/f_fs.c | 5 +++++ > 1 file changed, 5 insertions(+) > > diff --git a/drivers/usb/gadget/function/f_fs.c b/drivers/usb/gadget/function/f_fs.c > index ce1d023..728c260 100644 > --- a/drivers/usb/gadget/function/f_fs.c > +++ b/drivers/usb/gadget/function/f_fs.c > @@ -35,6 +35,7 @@ > #include <linux/mmu_context.h> > #include <linux/poll.h> > #include <linux/eventfd.h> > +#include <linux/thread_info.h> > > #include "u_fs.h" > #include "u_f.h" > @@ -826,6 +827,10 @@ static void ffs_user_copy_worker(struct work_struct *work) > if (io_data->read && ret > 0) { > mm_segment_t oldfs = get_fs(); > > +#if defined(CONFIG_ARM64) > + if (IS_ENABLED(CONFIG_ARM64_TAGGED_ADDR_ABI)) > + set_thread_flag(TIF_TAGGED_ADDR); > +#endif > set_fs(USER_DS); > use_mm(io_data->mm); > ret = ffs_copy_to_iter(io_data->buf, ret, &io_data->data);
On Tue, Feb 25, 2020 at 06:41:55PM +0800, Macpaul Lin wrote: > diff --git a/drivers/usb/gadget/function/f_fs.c b/drivers/usb/gadget/function/f_fs.c > index ce1d023..728c260 100644 > --- a/drivers/usb/gadget/function/f_fs.c > +++ b/drivers/usb/gadget/function/f_fs.c > @@ -35,6 +35,7 @@ > #include <linux/mmu_context.h> > #include <linux/poll.h> > #include <linux/eventfd.h> > +#include <linux/thread_info.h> > > #include "u_fs.h" > #include "u_f.h" > @@ -826,6 +827,10 @@ static void ffs_user_copy_worker(struct work_struct *work) > if (io_data->read && ret > 0) { > mm_segment_t oldfs = get_fs(); > > +#if defined(CONFIG_ARM64) > + if (IS_ENABLED(CONFIG_ARM64_TAGGED_ADDR_ABI)) > + set_thread_flag(TIF_TAGGED_ADDR); > +#endif > set_fs(USER_DS); > use_mm(io_data->mm); > ret = ffs_copy_to_iter(io_data->buf, ret, &io_data->data); I really don't think that's the correct fix. The TIF_TAGGED_ADDR is a per-thread property and not really compatible with use_mm(). We've had tagged pointers in arm64 user-space since day 0 and access_ok() would have prevented them, so this config is not something new. For some reason, adb now passes them to the kernel (presumably because user-space makes more use of them). If you have strong reasons not to fix it in adb, the next best thing may be to untag the addresses in the usb gadget driver.
diff --git a/drivers/usb/gadget/function/f_fs.c b/drivers/usb/gadget/function/f_fs.c index ce1d023..728c260 100644 --- a/drivers/usb/gadget/function/f_fs.c +++ b/drivers/usb/gadget/function/f_fs.c @@ -35,6 +35,7 @@ #include <linux/mmu_context.h> #include <linux/poll.h> #include <linux/eventfd.h> +#include <linux/thread_info.h> #include "u_fs.h" #include "u_f.h" @@ -826,6 +827,10 @@ static void ffs_user_copy_worker(struct work_struct *work) if (io_data->read && ret > 0) { mm_segment_t oldfs = get_fs(); +#if defined(CONFIG_ARM64) + if (IS_ENABLED(CONFIG_ARM64_TAGGED_ADDR_ABI)) + set_thread_flag(TIF_TAGGED_ADDR); +#endif set_fs(USER_DS); use_mm(io_data->mm); ret = ffs_copy_to_iter(io_data->buf, ret, &io_data->data);
This issue was found when adbd trying to open functionfs with AIO mode. Usually, we need to set "setprop sys.usb.ffs.aio_compat 0" to enable adbd with AIO mode on Android. When adbd is opening functionfs, it will try to read 24 bytes at the first read I/O control. If this reading has been failed, adbd will try to send FUNCTIONFS_CLEAR_HALT to functionfs. When adbd is in AIO mode, functionfs will be acted with asyncronized I/O path. After the successful read transfer has been completed by gadget hardware, the following series of functions will be called. ffs_epfile_async_io_complete() -> ffs_user_copy_worker() -> copy_to_iter() -> _copy_to_iter() -> copyout() -> iterate_and_advance() -> iterate_iovec() Adding debug trace to these functions, it has been found that in copyout(), access_ok() will check if the user space address is valid to write. However if CONFIG_ARM64_TAGGED_ADDR_ABI is enabled, adbd always passes user space address start with "0x3C" to gadget's AIO blocks. This tagged address will cause access_ok() check always fail. Which causes later calculation in iterate_iovec() turn zero. Copyout() won't copy data to userspace since the length to be copied "v.iov_len" will be zero. Finally leads ffs_copy_to_iter() always return -EFAULT, causes adbd cannot open functionfs and send FUNCTIONFS_CLEAR_HALT. Signed-off-by: Macpaul Lin <macpaul.lin@mediatek.com> --- Changes for v3: - Fix misspelling in commit message. Changes for v2: - Fix build error for 32-bit load. An #if defined(CONFIG_ARM64) still need for avoiding undeclared defines. drivers/usb/gadget/function/f_fs.c | 5 +++++ 1 file changed, 5 insertions(+)