On Wed, Nov 18, 2020 at 07:38:49AM -0800, Saranya Muruganandam wrote: > From: Li Xi <lixi@xxxxxxx> > > This patch only copy the fs to a new one when -m is enabled. > It doesn't actually start any thread. When pass1 test finishes, > the new fs is copied back to the original context. > > This patch handles the fs fields in dblist, inode_map and block_map > properly. > > Signed-off-by: Li Xi <lixi@xxxxxxx> > Signed-off-by: Wang Shilong <wshilong@xxxxxxx> > Signed-off-by: Saranya Muruganandam <saranyamohan@xxxxxxxxxx> I'm a bit surprised that we're not adding a ext2fs_clone_fs() function, but instead creating an e2fsck_pass1_copy_fs() function. Again, what's going to happen when we need to copy the fs structure for other passes. Also, just simply copying the fs structure seems dangerous; how are we going to know which allocated substructures were created before the fs structure was clonsed (and hence can't be safely freed in a copy of the fs structure), and which substructures were allocated *after* the fs structure is cloned, in which case they need to be freed when the cloned fs structuers are released? Perhaps the simplest thing to do is to assume that *everything* is cloned. If that results in too much memory consumed, maybe we can add a set of flags indicating which structures should *not* be freed for a particular fs structure, with the assumption that they will be freed when the top-level master fs structure is closed. There may be some potential traps with this, since in some cases a substructure can be released and re-allocated, in which case, if that substructure is shared by child fs structures, they could be left pointing at already freed copies of the data structures. Basically, even if this works (and the fact that I saw a crash with a data structure being double-freed in the f_multithread_ok test strongly suggesting that it isn't), I'd be worried about making sure that the resulting architecture is one which is robust and maintainable, and not leaving traps for future developers when they don't realize what the assumptions are about shared substructures. Which strongly suggests that this should be a first-class aspect of libext2fs and it should be clearly documented how sharing does or doesn't work. Cheers, - Ted
