dcache: Translating dentry into pathname without taking rename_lock
As long as you're removing locks from prepend_name and complicating its innards, I notice that each and every call site follows it by prepending "/". How about moving that into prepend_name as well?
Also, if you happen to feel like it, you can delete the slash flag and replace it with "bptr != *buffer".
Another small tweak would be to the global_root part of the code. You could move the is_mounted(vfsmnt) test up, and combine the tail of that code path with the regular exit. All you have to do is change the !slash test to:
if (error >= 0 && bptr == *buffer) { /* Root directory */
if (--blen < 0)
error = -ENAMETOOLONG;
else
*--bptr = '/';
}This modified form is no more code than an inlined copy of prepend(), so we haven't actually slowed the fast path, but it avoids corrupting the return value of 0/1/2 if possible.
Quoted 19 lines · George Spelvin
As long as you're removing locks from
prepend_nameand complicating its innards, I notice that each and every call site follows it by prepending "/". How about moving that intoprepend_nameas well?Also, if you happen to feel like it, you can delete the slash flag and replace it with "bptr != *buffer".
Another small tweak would be to the
global_rootpart of the code. You could move theis_mounted(vfsmnt)test up, and combine the tail of that code path with the regular exit. All you have to do is change the !slash test to:if (error>= 0&& bptr == *buffer) { /* Root directory */ if (--blen< 0) error = -ENAMETOOLONG; else *--bptr = '/'; }
Quoted 3 lines
This modified form is no more code than an inlined copy of
prepend(), so we haven't actually slowed the fast path, but it avoids corrupting the return value of 0/1/2 if possible.
Thank for the suggestions. I will implement them in my v2 patch.
-Longman