except, again, POLLFREE and POLL_BUSY_LOOP.
With this, we finally get to the promised end result:
- POLL{IN,OUT,...} are plain integers and *not* in __poll_t, so any
stray instances of ->poll() still using those will be caught by
sparse.
- eventpoll.c and select.c warning-free wrt __poll_t
- no more kernel-side definitions of POLL... - userland ones are
visible through the entire kernel (and used pretty much only for
mangle/demangle)
- same behavior as after the first series (i.e. sparc et.al. epoll(2)
working correctly).
Signed-off-by: Al Viro <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The only place that has any business including asm/poll.h
is linux/poll.h. Fortunately, asm/poll.h had only been
included in 3 places beyond that one, and all of them
are trivial to switch to using linux/poll.h.
Signed-off-by: Al Viro <[email protected]>
Same story as with dev_ifsioc(), except that the last cases with non-trivial
conversions had been taken out in 2013...
Signed-off-by: Al Viro <[email protected]>
Once upon a time net/socket.c:dev_ifsioc() used to handle SIOCSHWTSTAMP and
SIOCSIFMAP. These have different native and compat layout, so the format
conversion had been needed. In 2009 these two cases had been taken out,
turning the rest into a convoluted way to calling sock_do_ioctl(). We copy
compat structure into native one, call sock_do_ioctl() on that and copy
the result back for the in/out ioctls. No layout transformation anywhere,
so we might as well just call sock_do_ioctl() and skip all the headache with
copying.
Signed-off-by: Al Viro <[email protected]>
Only two of dev_ioctl() callers may pass SIOCGIFCONF to it.
Separating that codepath from the rest of dev_ioctl() allows both
to simplify dev_ioctl() itself (all other cases work with struct ifreq *)
*and* seriously simplify the compat side of that beast: all it takes
is passing to inet_gifconf() an extra argument - the size of individual
records (sizeof(struct ifreq) or sizeof(struct compat_ifreq)). With
dev_ifconf() called directly from sock_do_ioctl()/compat_dev_ifconf()
that's easy to arrange.
As the result, compat side of SIOCGIFCONF doesn't need any
allocations, copy_in_user() back and forth, etc.
Reviewed-by: Christoph Hellwig <[email protected]>
Signed-off-by: Al Viro <[email protected]>
--EWB Added #ifdef CONFIG_X86_X32_ABI to arch/x86/kernel/signal_compat.c
Changed #ifdef CONFIG_X86_X32 to #ifdef CONFIG_X86_X32_ABI in
linux/compat.h
CONFIG_X86_X32 is set when the user requests X32 support.
CONFIG_X86_X32_ABI is set when the user requests X32 support
and the tool-chain has X32 allowing X32 support to be built.
Signed-off-by: Al Viro <[email protected]>
Signed-off-by: Eric W. Biederman <[email protected]>
... having taught the latter that si_errno and si_code might be
swapped.
Signed-off-by: Al Viro <[email protected]>
Signed-off-by: Eric W. Biederman <[email protected]>
Got broken by "make sock_alloc_file() do sock_release() on failures" -
cleanup after sock_map_fd() failure got pulled all the way into
sock_alloc_file(), but it used to serve the case when sock_map_fd()
failed *before* getting to sock_alloc_file() as well, and that got
lost. Trivial to fix, fortunately.
Fixes: 8e1611e235 (make sock_alloc_file() do sock_release() on failures)
Reported-by: Dmitry Vyukov <[email protected]>
Signed-off-by: Al Viro <[email protected]>
similar to memdup_user(), but does *not* guarantee that result will
be physically contiguous; use only in cases where that's not a requirement
and free it with kvfree().
Signed-off-by: Al Viro <[email protected]>
Instead of doing that upon each ipcns creation, we do that the first
time mq_open(2) or mqueue mount is done in an ipcns. What's more,
doing that allows to get rid of mount_ns() use - we can go with
considerably cheaper mount_nodev(), avoiding the loop over all
mqueue superblock instances; ipcns->mq_mnt is used to locate preexisting
instance in O(1) time instead of O(instances) mount_ns() would've
cost us.
Based upon the version by Giuseppe Scrivano <[email protected]>; I've
added handling of userland mqueue mounts (original had been broken in
that area) and added a switch to mount_nodev().
Signed-off-by: Al Viro <[email protected]>
Similar to vfs_create(), but with caller-supplied callback (and
argument for it) to be used instead of ->create().
Signed-off-by: Al Viro <[email protected]>
Descriptor table is a shared object; it's not a place where you can
stick temporary references to files, especially when we don't need
an opened file at all.
Cc: [email protected] # v4.14
Fixes: 98589a0998 ("netfilter: xt_bpf: Fix XT_BPF_MODE_FD_PINNED mode of 'xt_bpf_info_v1'")
Signed-off-by: Al Viro <[email protected]>
Saving access_ok() is not worth the trouble; yes, the callers of ->read()
and ->write() will have done the right checks, but it's much too long
(and varied) call chains to rely upon.
Signed-off-by: Al Viro <[email protected]>
memdup_user() checks it, so the only effect would be failing with
-EINVAL instead of -EFAULT in case when access_ok() is false.
However, the caller has already checked access_ok() itself (and
would have buggered off with -EFAULT), so the check is completely
pointless. Removing it both simplifies the only instance
of ->dsp_load() and allows to get rid of the check in caller -
its sole effect used to be in preventing a bogus error value
from access_ok() in the instance. Let memdup_user() do the
right thing instead...
Signed-off-by: Al Viro <[email protected]>
The following primitives are defined in linux/bitfield.h:
* u32 le32_get_bits(__le32 val, u32 field) extracts the contents of the
bitfield specified by @field in little-endian 32bit object @val and
converts it to host-endian.
* void le32p_replace_bits(__le32 *p, u32 v, u32 field) replaces
the contents of the bitfield specified by @field in little-endian
32bit object pointed to by @p with the value of @v. New value is
given in host-endian and stored as little-endian.
* __le32 le32_replace_bits(__le32 old, u32 v, u32 field) is equivalent to
({__le32 tmp = old; le32p_replace_bits(&tmp, v, field); tmp;})
In other words, instead of modifying an object in memory, it takes
the initial value and returns the modified one.
* __le32 le32_encode_bits(u32 v, u32 field) is equivalent to
le32_replace_bits(0, v, field). In other words, it returns a little-endian
32bit object with the bitfield specified by @field containing the
value of @v and all bits outside that bitfield being zero.
Such set of helpers is defined for each of little-, big- and host-endian
types; e.g. u64_get_bits(val, field) will return the contents of the bitfield
specified by @field in host-endian 64bit object @val, etc. Of course, for
host-endian no conversion is involved.
Fields to access are specified as GENMASK() values - an N-bit field
starting at bit #M is encoded as GENMASK(M + N - 1, M). Note that
bit numbers refer to endianness of the object we are working with -
e.g. GENMASK(11, 0) in __be16 refers to the second byte and the lower
4 bits of the first byte. In __le16 it would refer to the first byte
and the lower 4 bits of the second byte, etc.
Field specification must be a constant; __builtin_constant_p() doesn't
have to be true for it, but compiler must be able to evaluate it at
build time. If it cannot or if the value does not encode any bitfield,
the build will fail.
If the value being stored in a bitfield is a constant that does not fit
into that bitfield, a warning will be generated at compile time.
Signed-off-by: Al Viro <[email protected]>
This changes calling conventions (and simplifies the hell out
the callers). New rules: once struct socket had been passed
to sock_alloc_file(), it's been consumed either by struct file
or by sock_release() done by sock_alloc_file(). Either way
the caller should not do sock_release() after that point.
Reviewed-by: Eric Dumazet <[email protected]>
Signed-off-by: Al Viro <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
1) it's fput() or sock_release(), not both
2) don't do fd_install() until the last failure exit.
3) not a bug per se, but... don't attach socket to struct file
until it's set up.
Take reserving descriptor into the caller, move fd_install() to the
caller, sanitize failure exits and calling conventions.
Cc: [email protected] # v4.6+
Acked-by: Tom Herbert <[email protected]>
Signed-off-by: Al Viro <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
* no need for callback argument - it's always the same one
* fold __qp_memcpy_from_queue() into its only caller, get rid of dead code
* pass struct iov_iter * without casting to void *
* don't pass buf_size at all - it's always iov_iter_count(to)
Signed-off-by: Al Viro <[email protected]>
switch both of its users to qp_memcpy_from_queue_iov() - just
make it take iov_iter * instead of msghdr * and arrange for a
iov_iter for it in all cases.
Signed-off-by: Al Viro <[email protected]>
Both qp_dequeue_locked() and qp_enqueue_locked() use the
buf_size argument to decide how much would be there to copy;
in case of iovec- (== msghdr-)based primitives it's not
iov_size, it's msg_data_left(msg).
Signed-off-by: Al Viro <[email protected]>
here we do need to reinitialize ->msg_iter on each call - the
data in buffer is overwritten every time, not appended to.
Signed-off-by: Al Viro <[email protected]>
The only caller that doesn't pass true in it is get_user_pages() and
it passes NULL in locked. The only place where we check it is
if (notify_locked && lock_dropped && *locked)
and lock_dropped can become true only if we have locked != NULL.
In other words, the second part of condition will be false when
called by get_user_pages().
Just get rid of the argument and turn the condition into
if (lock_dropped && *locked)
Signed-off-by: Al Viro <[email protected]>
Equivalent transformation - the only place in __get_user_pages_locked()
where we look at notify_drop argument is
if (notify_drop && lock_dropped && *locked) {
up_read(&mm->mmap_sem);
*locked = 0;
}
in the very end. Changing notify_drop from false to true won't change
behaviour unless *locked is non-zero. The caller is
ret = __get_user_pages_locked(current, mm, start, nr_pages, pages, NULL,
&locked, false, gup_flags | FOLL_TOUCH);
if (locked)
up_read(&mm->mmap_sem);
so in that case the original kernel would have done up_read() right after
return from __get_user_pages_locked(), while the modified one would've done
it right before the return.
Signed-off-by: Al Viro <[email protected]>
two callers that do so don't need to bother - we'd already
masked it with epi->event.events, which
* couldn't have changed since we are holding ->mtx
* had been set to event->events
* is still equal to event->events, since *event is never
changed by anything.
Signed-off-by: Al Viro <[email protected]>
First of all, NULL ->poll() means "always POLLIN, always POLLOUT", not an error.
Furthermore, mixing -EREMOTEIO with POLL... masks and expecting it to do anything
good is insane - both are arch-dependent, to start with. Pass a pointer to
store the error value separately and make it return POLLERR in such case.
And ->poll() calling conventions do *not* include "return -Esomething". Never
had.
Signed-off-by: Al Viro <[email protected]>
Al Viro