it actually worked only when requested area ended on the page boundary...
Reported-by: Marco Grassi <[email protected]>
Signed-off-by: Al Viro <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
looking for duplicate ->iov_base makes sense only for
iovec-backed iterators; for kvec-backed ones it's pointless,
for bvec-backed ones it's pointless and broken on 32bit (we
walk through an array of struct bio_vec accessing them as if
they were struct iovec; works by accident on 64bit, but on
32bit it'll blow up) and for pipe-backed ones it's pointless
and ends up oopsing.
Signed-off-by: Al Viro <[email protected]>
by making sure we call iov_iter_advance() on original
iov_iter even if direct_IO (done on its copy) has returned 0.
It's a no-op for old iov_iter flavours and does the right thing
(== truncation of the stuff we'd allocated, but not filled) in
ITER_PIPE case. Failures (e.g. -EIO) get caught and dealt with
by cleanup in generic_file_read_iter().
Signed-off-by: Al Viro <[email protected]>
we only use iov_iter_get_pages_alloc() and iov_iter_advance() -
pages are filled by kernel_readv() via a kvec array (as we used
to do all along), so iov_iter here is used only as a way of
arranging for those pages to be in pipe.
Signed-off-by: Al Viro <[email protected]>
iov_iter variant for passing data into pipe. copy_to_iter()
copies data into page(s) it has allocated and stuffs them into
the pipe; copy_page_to_iter() stuffs there a reference to the
page given to it. Both will try to coalesce if possible.
iov_iter_zero() is similar to copy_to_iter(); iov_iter_get_pages()
and friends will do as copy_to_iter() would have and return the
pages where the data would've been copied. iov_iter_advance()
will truncate everything past the spot it has advanced to.
New primitive: iov_iter_pipe(), used for initializing those.
pipe should be locked all along.
Running out of space acts as fault would for iovec-backed ones;
in other words, giving it to ->read_iter() may result in short
read if the pipe overflows, or -EFAULT if it happens with nothing
copied there.
In other words, ->read_iter() on those acts pretty much like
->splice_read(). Moreover, all generic_file_splice_read() users,
as well as many other ->splice_read() instances can be switched
to that scheme - that'll happen in the next commit.
Signed-off-by: Al Viro <[email protected]>
since pipe_lock is the outermost now, we don't need to drop/regain
socket locks around the call of splice_to_pipe() from skb_splice_bits(),
which kills the need to have a socket-specific callback; we can just
call splice_to_pipe() and be done with that.
Signed-off-by: Al Viro <[email protected]>
single-buffer analogue of splice_to_pipe(); vmsplice_to_pipe() switched
to that, leaving splice_to_pipe() only for ->splice_read() instances
(and that only until they are converted as well).
Signed-off-by: Al Viro <[email protected]>
* splice_to_pipe() stops at pipe overflow and does *not* take pipe_lock
* ->splice_read() instances do the same
* vmsplice_to_pipe() and do_splice() (ultimate callers of splice_to_pipe())
arrange for waiting, looping, etc. themselves.
That should make pipe_lock the outermost one.
Unfortunately, existing rules for the amount passed by vmsplice_to_pipe()
and do_splice() are quite ugly _and_ userland code can be easily broken
by changing those. It's not even "no more than the maximal capacity of
this pipe" - it's "once we'd fed pipe->nr_buffers pages into the pipe,
leave instead of waiting".
Considering how poorly these rules are documented, let's try "wait for some
space to appear, unless given SPLICE_F_NONBLOCK, then push into pipe
and if we run into overflow, we are done".
Signed-off-by: Al Viro <[email protected]>
Make local filesystems treat a fault as shortened IO,
returning -EFAULT only if nothing had been transferred.
That's how everything else (NFS, FUSE, ceph, Lustre)
behaves.
Signed-off-by: Al Viro <[email protected]>
* the only remaining callers of "short" fault-ins are just as happy with generic
variants (both in lib/iov_iter.c); switch them to multipage variants, kill the
"short" ones
* rename the multipage variants to now available plain ones.
* get rid of compat macro defining iov_iter_fault_in_multipage_readable by
expanding it in its only user.
Signed-off-by: Al Viro <[email protected]>
Switching iov_iter fault-in to multipages variants has exposed an old
bug in underlying fault_in_multipages_...(); they break if the range
passed to them wraps around. Normally access_ok() done by callers will
prevent such (and it's a guaranteed EFAULT - ERR_PTR() values fall into
such a range and they should not point to any valid objects).
However, on architectures where userland and kernel live in different
MMU contexts (e.g. s390) access_ok() is a no-op and on those a range
with a wraparound can reach fault_in_multipages_...().
Since any wraparound means EFAULT there, the fix is trivial - turn
those
while (uaddr <= end)
...
into
if (unlikely(uaddr > end))
return -EFAULT;
do
...
while (uaddr <= end);
Reported-by: Jan Stancek <[email protected]>
Tested-by: Jan Stancek <[email protected]>
Cc: [email protected] # v3.5+
Signed-off-by: Al Viro <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
... it would get converted to regular if such had been attempted
Signed-off-by: Al Viro <[email protected]>
Signed-off-by: Jan Kara <[email protected]>
get_user_ex(x, ptr) should zero x on failure. It's not a lot of a leak
(at most we are leaking uninitialized 64bit value off the kernel stack,
and in a fairly constrained situation, at that), but the fix is trivial,
so...
Cc: [email protected]
Signed-off-by: Al Viro <[email protected]>
[ This sat in different branch from the uaccess fixes since mid-August ]
Signed-off-by: Linus Torvalds <[email protected]>
really ugly, but apparently avr32 compilers turns access_ok() into
something so bad that they want it in assembler. Left that way,
zeroing added in inline wrapper.
Cc: [email protected]
Signed-off-by: Al Viro <[email protected]>
It could be done in exception-handling bits in __get_user_b() et.al.,
but the surgery involved would take more knowledge of sh64 details
than I have or _want_ to have.
Cc: [email protected]
Signed-off-by: Al Viro <[email protected]>
* should zero on any failure
* __get_user() should use __copy_from_user(), not copy_from_user()
Cc: [email protected]
Signed-off-by: Al Viro <[email protected]>
should clear on access_ok() failures. Also remove the useless
range truncation logics.
Cc: [email protected]
Signed-off-by: Al Viro <[email protected]>
... that should zero on faults. Also remove the <censored> helpful
logics wrt range truncation copied from ppc32. Where it had ever
been needed only in case of copy_from_user() *and* had not been merged
into the mainline until a month after the need had disappeared.
A decade before openrisc went into mainline, I might add...
Cc: [email protected]
Signed-off-by: Al Viro <[email protected]>
It's -EFAULT, not -1 (and contrary to the comment in there,
__strnlen_user() can return 0 - on faults).
Cc: [email protected]
Acked-by: Richard Kuo <[email protected]>
Signed-off-by: Al Viro <[email protected]>
It should check access_ok(). Otherwise a bunch of places turn into
trivially exploitable rootholes.
Cc: [email protected]
Signed-off-by: Al Viro <[email protected]>
* copy_from_user() on access_ok() failure ought to zero the destination
* none of those primitives should skip the access_ok() check in case of
small constant size.
Cc: [email protected]
Acked-by: Jesper Nilsson <[email protected]>
Signed-off-by: Al Viro <[email protected]>
... in all cases, including the failing access_ok()
Note that some architectures using asm-generic/uaccess.h have
__copy_from_user() not zeroing the tail on failure halfway
through. This variant works either way.
Cc: [email protected]
Signed-off-by: Al Viro <[email protected]>
In v9fs_vfs_rename() we need to clone the parents' fids, not just
find them.
Spotted-by: Johannes Berg <[email protected]>
Signed-off-by: Al Viro <[email protected]>
The only place where we feed to __d_rehash() something other than
d_hash(dentry->d_name.hash) is __d_move(), where we give it d_hash
of another dentry. Postpone rehashing until we'd switched the
names and we are rid of that exception, along with the need to
keep _d_rehash() and __d_rehash() separate.
Signed-off-by: Al Viro <[email protected]>
Since 2006 we have ->i_bdev pinning bdev in question, so there's no
way to get to bdev ->evict_inode() while there's an aliasing inode
anywhere. In other words, the only place walking the list of aliases
is guaranteed to do it only when the list is empty...
Remove the detritus; it should've been done in "[PATCH] Fix a race
condition between ->i_mapping and iput()", but nobody had noticed it
back then.
Signed-off-by: Al Viro <[email protected]>
Al Viro