| Message ID | 1582011672-17189-1-git-send-email-macpaul.lin@mediatek.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | lib: iov_iter.c: fix a possible calculation error on remaining bytes | expand |
On Tue, Feb 18, 2020 at 03:41:12PM +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 > fisrt 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 > iterate_iovec(), the calculation result of n will be turned into zero. > n = wanted - n; /* 0 == n = 24 - 24; */ > Which causes copyout() won't copy data to userspace since the length > to be copied "v.iov_len" will be zero, which isn't correct. This also > leads ffs_copy_to_iter() always return -EFAULT. Finally adbd cannot > open functionfs and send FUNCTIONFS_CLEAR_HALT. > > Signed-off-by: Macpaul Lin <macpaul.lin@mediatek.com> > --- > lib/iov_iter.c | 3 ++- > 1 file changed, 2 insertions(+), 1 deletion(-) > > diff --git a/lib/iov_iter.c b/lib/iov_iter.c > index fb29c02c6a3c..f9334144e259 100644 > --- a/lib/iov_iter.c > +++ b/lib/iov_iter.c > @@ -36,7 +36,8 @@ > skip = __v.iov_len; \ > n -= __v.iov_len; \ > } \ > - n = wanted - n; \ > + if (n != wanted) \ > + n = wanted - n; \ > } First of all, nothing in that line can possibly *cause* copyout() to do anything - it's after the calls of step. What's more, this changes behaviour only when wanted would've been equal to n, doesn't it? Which translates into "no decrements of n have happened at all", i.e. "nothing has been copied". IOW, it's a consequence of no copyout, not the cause of such. You can make copy_to_iter() lie and pretend if has copied everything when it has copied nothing, but that won't change the underlying bug. So I'm afraid your debugging is not finished - you still need to find out what causes the copyout failures and/or BS iov_iter padded by caller.
On Tue, 2020-02-18 at 12:41 +0000, Al Viro wrote: > On Tue, Feb 18, 2020 at 03:41:12PM +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 > > fisrt 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 > > iterate_iovec(), the calculation result of n will be turned into zero. > > n = wanted - n; /* 0 == n = 24 - 24; */ > > Which causes copyout() won't copy data to userspace since the length > > to be copied "v.iov_len" will be zero, which isn't correct. This also > > leads ffs_copy_to_iter() always return -EFAULT. Finally adbd cannot > > open functionfs and send FUNCTIONFS_CLEAR_HALT. > > > > Signed-off-by: Macpaul Lin <macpaul.lin@mediatek.com> > > --- > > lib/iov_iter.c | 3 ++- > > 1 file changed, 2 insertions(+), 1 deletion(-) > > > > diff --git a/lib/iov_iter.c b/lib/iov_iter.c > > index fb29c02c6a3c..f9334144e259 100644 > > --- a/lib/iov_iter.c > > +++ b/lib/iov_iter.c > > @@ -36,7 +36,8 @@ > > skip = __v.iov_len; \ > > n -= __v.iov_len; \ > > } \ > > - n = wanted - n; \ > > + if (n != wanted) \ > > + n = wanted - n; \ > > } > > First of all, nothing in that line can possibly *cause* > copyout() to do anything - it's after the calls of step. What's > more, this changes behaviour only when wanted would've been equal to > n, doesn't it? Which translates into "no decrements of n have > happened at all", i.e. "nothing has been copied". IOW, it's > a consequence of no copyout, not the cause of such. You can > make copy_to_iter() lie and pretend if has copied everything > when it has copied nothing, but that won't change the underlying > bug. > > So I'm afraid your debugging is not finished - you > still need to find out what causes the copyout failures and/or > BS iov_iter padded by caller. Thanks for your explanation. After these days, I've found this issue only happened on arm 64-bit system, while arm 32-bit system works well. After more debugging, it looks like arm 64 tagged ABI related issue. I've update a new patch here: usb: gadget: f_fs: try to fix AIO issue under ARM 64 bit TAGGED mode https://lkml.org/lkml/2020/2/23/16. Thanks. Macpaul Lin.
diff --git a/lib/iov_iter.c b/lib/iov_iter.c index fb29c02c6a3c..f9334144e259 100644 --- a/lib/iov_iter.c +++ b/lib/iov_iter.c @@ -36,7 +36,8 @@ skip = __v.iov_len; \ n -= __v.iov_len; \ } \ - n = wanted - n; \ + if (n != wanted) \ + n = wanted - n; \ } #define iterate_kvec(i, n, __v, __p, skip, STEP) { \
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 fisrt 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 iterate_iovec(), the calculation result of n will be turned into zero. n = wanted - n; /* 0 == n = 24 - 24; */ Which causes copyout() won't copy data to userspace since the length to be copied "v.iov_len" will be zero, which isn't correct. This also leads ffs_copy_to_iter() always return -EFAULT. Finally adbd cannot open functionfs and send FUNCTIONFS_CLEAR_HALT. Signed-off-by: Macpaul Lin <macpaul.lin@mediatek.com> --- lib/iov_iter.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)