when wstating a file, its directory should be updated to
reflect this change.
here is what the manpage states:
> The mtime field reflects the time of the last change of content
> (except when later changed by wstat). For a directory it is the
> time of the most recent remove, create, or wstat of a file in the
> directory.
with the -s flag, we should read 9P messages from
standard *INPUT* (fd 0) and write responses to
standard *OUTPUT* (fd 1).
before these servers where reading from fd 1,
assuming they where both the same files.
buffers which still have requests queued on them are not free!
we cannot chanedev() a buffer while it has still requests queued on it
and we canot just queue our request (having different address) on the
buffer while there are other requests before it, otherwise we would
create artificial block dependency that can cause deadlock.
it is possible for another getbuf() on buffer b to come in
before undelayreq() calls givebuf() on a buffer again. then
givebuf() would find b already busy and abort().
instead, we now handle what getbuf() did in givebuf() and
consider the Buf* argument to givebuf() as a hint only for
the case when we have to actually flush/read a block from
disk.
check parent directory permission *after* we determined
that the new name does not exist in the parent, so that
when the new name is the same as old name then no write
permission is required in the parent directory.
we used to do getbuf() with nodata flag so it only worked when
we where lucky and got the same in memory block back. this
is uncritical once you have reamed the filesystem, its just
that sometimes ream would fail with "ream successfull, then
hjfs: fsinit: file ./hjfs not found".
check for write premission in the parent directory
for open with ORCLOSE. honor CHFNOPERM not just in
chancreat(), pikeshedd the error handling. added
estrdup()/erealloc() that call sysfatal instead
of returning nil.
getdent() introduced a bug as the qid check fails in
the case when we walk to DUMPROOTQID, but get ROOTQID
in the directory instead.
newentry() getblk(..., GBCREATE) caused the whole
directory to get dumped until it hit a free slot.
we cannot do this because this changes the addresses
of Dentries of files but doesnt update the loctree.
this caused the bogus walks when we hit a different
directory than we expected.
what we do now is the following: newentry() was changed
to never dump anything. it will just read the directory
and return a the address of a free slot or create a
new block having space for one. chancreat() then makes
a loc in the loctree for the free slot, drops the dirent
buffer and calls willmodify() on the new slot. this will
dump the block containing the slot (and possible other
slots) and updates the loctree to the new version.
after that, chancreate() modifies the slot filling the
Dentry. there should be no race, as newentry() skips
slots that have a loc in the loctree.
theres one case where newentry() can dump the block
immidiately when it is called from dump. added new
parameter for that and documented in the comment.
createuserdir() was simplified by just calling chancreat().
to get arround the permission check, a new per channel
flag CHFNOPERM was added.
always check if the directory entry qid from the loc still
matches the one on disk before doing anything. helps
catching bugs and is basically equivalent to what cwfs does
with its checktag.
make a haveloc() check in newentry() to make sure we dont
allocate a file slot thats still in use, but deleted.
this seems to fix the NPROC>1 build problems.