| Message ID | 20250205183839.395081-1-mjguzik@gmail.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | CONFIG_VFS_DEBUG at last | expand |
On Wed, Feb 5, 2025 at 7:38 PM Mateusz Guzik <mjguzik@gmail.com> wrote: > > This adds a super basic version just to get the mechanism going and > adds sample usage. > > The macro set is incomplete (e.g., lack of locking macros) and > dump_inode routine fails to dump any state yet, to be implemented(tm). > > I think despite the primitive state this is complete enough to start > sprinkling warns as necessary. > > Mateusz Guzik (3): > vfs: add initial support for CONFIG_VFS_DEBUG > vfs: catch invalid modes in may_open > vfs: use the new debug macros in inode_set_cached_link() > > fs/namei.c | 2 ++ > include/linux/fs.h | 16 +++---------- > include/linux/vfsdebug.h | 50 ++++++++++++++++++++++++++++++++++++++++ > lib/Kconfig.debug | 9 ++++++++ > 4 files changed, 64 insertions(+), 13 deletions(-) > create mode 100644 include/linux/vfsdebug.h > > -- > 2.43.0 > The produced warn is ugly as sin:\, for example for that bad size: [ 51.433206] VFS_WARN_ON_INODE(__builtin_choose_expr((sizeof(int) == sizeof(*(8 ? ((void *)((long)(__builtin_strlen(link)) * 0l)) : (int *)8))), __builtin_strlen(link), __fortify_strlen(link)) != linklen) failed for inode ff32f7c350c8aec8 maybe there is a way to work it around, the code is literally lifted out of mmdebug.h so they presumably have the same problem apart from that the assert in may_open is backwards, the code normally is not reached. anyhow I expect to send a v2, but will wait for feedback before I do
On Wed, Feb 05, 2025 at 07:57:07PM +0100, Mateusz Guzik wrote: > On Wed, Feb 5, 2025 at 7:38 PM Mateusz Guzik <mjguzik@gmail.com> wrote: > > > > This adds a super basic version just to get the mechanism going and > > adds sample usage. > > > > The macro set is incomplete (e.g., lack of locking macros) and > > dump_inode routine fails to dump any state yet, to be implemented(tm). > > > > I think despite the primitive state this is complete enough to start > > sprinkling warns as necessary. > > > > Mateusz Guzik (3): > > vfs: add initial support for CONFIG_VFS_DEBUG > > vfs: catch invalid modes in may_open > > vfs: use the new debug macros in inode_set_cached_link() > > > > fs/namei.c | 2 ++ > > include/linux/fs.h | 16 +++---------- > > include/linux/vfsdebug.h | 50 ++++++++++++++++++++++++++++++++++++++++ > > lib/Kconfig.debug | 9 ++++++++ > > 4 files changed, 64 insertions(+), 13 deletions(-) > > create mode 100644 include/linux/vfsdebug.h > > > > -- > > 2.43.0 > > > > The produced warn is ugly as sin:\, for example for that bad size: > [ 51.433206] VFS_WARN_ON_INODE(__builtin_choose_expr((sizeof(int) == > sizeof(*(8 ? ((void *)((long)(__builtin_strlen(link)) * 0l)) : (int > *)8))), __builtin_strlen(link), __fortify_strlen(link)) != linklen) > failed for inode ff32f7c350c8aec8 > > maybe there is a way to work it around, the code is literally lifted > out of mmdebug.h so they presumably have the same problem > > apart from that the assert in may_open is backwards, the code normally > is not reached. > > anyhow I expect to send a v2, but will wait for feedback before I do I think it is overdue that we do something like this. Being able to be more liberal with _meaningful_ asserts can help us in the long run. So I'm supportive.