#41 closed defect (fixed)
libc: incorrect __libc_back_fsUnixAttribsGet() usage caused st_dev and st_ino to be 0 in some stat()s and chmod & fchmod to set MODE EA without type
| Reported by: | bird | Owned by: | bird |
|---|---|---|---|
| Priority: | normal | Milestone: | libc-0.6.1 |
| Component: | libc-backend | Version: | 0.6 |
| Severity: | normal | Keywords: | unixeas stat |
| Cc: |
Description (last modified by )
we're not checking if we actually read any UnixEAs in __libc_back_fsUnixAttribsGet().
Change History (3)
comment:1 Changed 18 years ago by
| Description: | modified (diff) |
|---|---|
| Resolution: | → fixed |
| Status: | new → closed |
comment:2 Changed 18 years ago by
Added extra validations to the file mode mask to reject invalid ones.
comment:3 Changed 18 years ago by
| Summary: | libc: st_dev and st_ino isn't set if the file has some EAs but not UnixEAs. → libc: incorrect __libc_back_fsUnixAttribsGet() usage caused st_dev and st_ino to be 0 in some stat()s and chmod & fchmod to set MODE EA without type |
|---|
Just updated the title to reflect the problems actually solved.
As a side note, there was actually a fchmod() smoke test which was set to 'expected failure' because of the MODE EA screwup.
Note: See
TracTickets for help on using
tickets.
That's not quite right. the fstat backend is delaying the dev/ino calc until after getting unix attribs and won't do it unless the unix attrib call fails. I've added code to generate ino/dev when they are not present on a successful return from the attribute getter. I've also changed the two set file mode backend functions to use a different API for querying the file-mode mask, both for speed and so we don't add the MODE attribute unless it's already there (i.e. LIBC created the file). That'll cause less worries for now (it'll be changed in 0.7, probably).