Messages in this thread | | | Date | Thu, 17 Jan 2008 16:40:35 +0300 | From | "Anton Salikhmetov" <> | Subject | Re: [PATCH -v5 2/2] Updating ctime and mtime at syncing |
| |
2008/1/17, Miklos Szeredi <miklos@szeredi.hu>: > > 2008/1/17, Miklos Szeredi <miklos@szeredi.hu>: > > > > > 4. Recording the time was the file data changed > > > > > > > > > > Finally, I noticed yet another issue with the previous version of my patch. > > > > > Specifically, the time stamps were set to the current time of the moment > > > > > when syncing but not the write reference was being done. This led to the > > > > > following adverse effect on my development system: > > > > > > > > > > 1) a text file A was updated by process B; > > > > > 2) process B exits without calling any of the *sync() functions; > > > > > 3) vi editor opens the file A; > > > > > 4) file data synced, file times updated; > > > > > 5) vi is confused by "thinking" that the file was changed after 3). > > > > > > Updating the time in remove_vma() would fix this, no? > > > > We need to save modification time. Otherwise, updating time stamps > > will be confusing the vi editor. > > remove_vma() will be called when process B exits, so if the times are > updated there, and the flag is cleared, the times won't be updated > later. > > > > > > > > > All these changes to inode.c are unnecessary, I think. > > > > > > > > The first part is necessary to account for "remembering" the modification time. > > > > > > > > The second part is for handling block device files. I cannot see any other > > > > sane way to update file times for them. > > > > > > Use file_update_time(), which will do the right thing. It will in > > > fact do the same thing as write(2) on the device, which is really what > > > we want. > > > > > > Block devices being mapped for write through different device > > > nodes..., well, I don't think we really need to handle such weird > > > corner cases 100% acurately. > > > > The file_update_time() cannot be used for implementing > > the "auto-update" feature, because the sync() system call > > doesn't "know" about the file which was memory-mapped. > > I'm not sure this auto-updating is really needed (POSIX doesn't > mandate it).
Peter Shtaubach, author of the first solution for this bug, and Jacob Ostergaard, the reporter of this bug, insist the "auto-update" feature to be implemented.
> > At least split it out into a separate patch, so it can be considered > separately on it's own merit. > > I think doing the same with the page-table reprotecting in MS_ASYNC is > also a good idea. That will leave us with > > 1) a base patch: update time just from fsync() and remove_vma() > 2) update time on sync(2) as well > 3) update time on MS_ASYNC as well > > I'd happily ack the first one, which would solve the most serious > issues, but have some reservations about the other two. > > Miklos >
| |