On Tue, Apr 17, 2018 at 11:29:07PM -0700, Song Liu wrote: > As Miklos reported and suggested: > > This pattern repeats two times in trace_uprobe.c and in > kernel/events/core.c as well: > > ret = kern_path(filename, LOOKUP_FOLLOW, &path); > if (ret) > goto fail_address_parse; > > inode = igrab(d_inode(path.dentry)); > path_put(&path); > > And it's wrong. You can only hold a reference to the inode if you > have an active ref to the superblock as well (which is normally > through path.mnt) or holding s_umount.
Oops. I must have snatched it from the uprobe code without thinking. > This way unmounting the containing filesystem while the tracepoint is > active will give you the "VFS: Busy inodes after unmount..." message > and a crash when the inode is finally put. > > Solution: store path instead of inode. > > This patch fixes the issue in kernel/event/core.c. > > NOTE: Based on my understanding, perf_addr_filter only supports intel_pt. Coresight too, but that's probably even further away from what you have. > However, my test system doesn't support address filtering (or I made a > mistake?). Therefore, I have NOT tested this patch. Check /sys/devices/intel_pt/caps/num_address_ranges, if it's non-zero, it's supported. > Could someone please help test it? Yes: Reviewed-and-tested-by: Alexander Shishkin <alexander.shish...@linux.intel.com> The subject line needs a little love to be more like other perf commits, but other than that, looks good. Thanks!
