There are two code locations that implement the SG_IO ioctl: the old
sg.c driver, and the generic scsi_ioctl helper that is in turn used by
multiple drivers.

To eradicate the old compat_ioctl conversion handler for the SG_IO
command, I implement a readable pair of put_sg_io_hdr() /get_sg_io_hdr()
helper functions that can be used for both compat and native mode,
and then I call this from both drivers.

For the iovec handling, there is already a compat_import_iovec() function
that can simply be called in place of import_iovec().

To avoid having to pass the compat/native state through multiple
indirections, I mark the SG_IO command itself as compatible in
fs/compat_ioctl.c and use in_compat_syscall() to figure out where
we are called from.

As a side-effect of this, the sg.c driver now also accepts the 32-bit
sg_io_hdr format in compat mode using the read/write interface, not
just ioctl. This should improve compatiblity with old 32-bit binaries,
but it would break if any application intentionally passes the 64-bit
data structure in compat mode here.

Steffen Maier helped debug an issue in an earlier version of this patch.

Cc: Steffen Maier <maier@linux.ibm.com>
Cc: linux-scsi@vger.kernel.org
Cc: Doug Gilbert <dgilbert@interlog.com>
Cc: "James E.J. Bottomley" <jejb@linux.ibm.com>
Cc: "Martin K. Petersen" <martin.petersen@oracle.com>
Signed-off-by: NArnd Bergmann <arnd@arndb.de>

All watchdog drivers implement the same set of ioctl commands, and
fortunately all of them are compatible between 32-bit and 64-bit
architectures.

Modern drivers always go through drivers/watchdog/wdt.c as an abstraction
layer, but older ones implement their own file_operations on a character
device for this.

Move the handling from fs/compat_ioctl.c into the individual drivers.

Note that most of the legacy drivers will never be used on 64-bit
hardware, because they are for an old 32-bit SoC implementation, but
doing them all at once is safer than trying to guess which ones do
or do not need the compat_ioctl handling.
Reviewed-by: NGuenter Roeck <linux@roeck-us.net>
Signed-off-by: NArnd Bergmann <arnd@arndb.de>

Remove the special case for FITRIM, and make file systems
handle that like all other ioctl commands with their own
handlers.

Cc: linux-ext4@vger.kernel.org
Cc: linux-f2fs-devel@lists.sourceforge.net
Cc: Mikulas Patocka <mikulas@artax.karlin.mff.cuni.cz>
Cc: linux-nilfs@vger.kernel.org
Cc: ocfs2-devel@oss.oracle.com
Signed-off-by: NArnd Bergmann <arnd@arndb.de>

Out of the four ioctl commands supported on gfs2, only FITRIM
works in compat mode.

Add a proper handler based on the ext4 implementation.

Fixes: 6ddc5c3d ("gfs2: getlabel support")
Reviewed-by: NBob Peterson <rpeterso@redhat.com>
Cc: Andreas Gruenbacher <agruenba@redhat.com>
Signed-off-by: NArnd Bergmann <arnd@arndb.de>

The last users are all gone, so let's remove the macro as well.
Signed-off-by: NArnd Bergmann <arnd@arndb.de>

Commit aa98aa31 ("md: move compat_ioctl handling into md.c")
already removed the COMPATIBLE_IOCTL() table entries and added
a complete implementation, but a few lines got left behind and
should also be removed here.

Cc: linux-raid@vger.kernel.org
Cc: Song Liu <song@kernel.org>
Signed-off-by: NArnd Bergmann <arnd@arndb.de>

The /dev/rawX implementation already handles these just fine, so
the entries in the table are not needed any more.
Signed-off-by: NArnd Bergmann <arnd@arndb.de>

The /proc/pci/ implementation already handles these just fine, so
the entries in the table are not needed any more.

Cc: linux-pci@vger.kernel.org
Cc: Bjorn Helgaas <bhelgaas@google.com>
Signed-off-by: NArnd Bergmann <arnd@arndb.de>

The joystick driver already handles these just fine, so
the entries in the table are not needed any more.

Cc: linux-input@vger.kernel.org
Cc: Dmitry Torokhov <dmitry.torokhov@gmail.com>
Signed-off-by: NArnd Bergmann <arnd@arndb.de>

These are all handled by the random driver, so instead of listing
each ioctl, we can use the generic compat_ptr_ioctl() helper.
Acked-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: NArnd Bergmann <arnd@arndb.de>

Since commit 07d106d0 ("vfs: fix up ENOIOCTLCMD error handling"),
we don't warn about unhandled compat-ioctl command code any more, but
just return the same error that a native file descriptor returns when
there is no handler.

This means the IGNORE_IOCTL() annotations are completely useless and
can all be removed. TIOCSTART/TIOCSTOP and KDGHWCLK/KDSHWCLK fall into
the same category, but for some reason were listed as COMPATIBLE_IOCTL().
Signed-off-by: NArnd Bergmann <arnd@arndb.de>

The SNDCTL_* and SOUND_* commands are the old OSS user interface.

I checked all the sound ioctl commands listed in fs/compat_ioctl.c
to see if we still need the translation handlers. Here is what I
found:

- sound/oss/ is (almost) gone from the kernel, this is what actually
  needed all the translations
- The ALSA emulation for OSS correctly handles all compat_ioctl
  commands already.
- sound/oss/dmasound/ is the last holdout of the original OSS code,
  this is only used on arch/m68k, which has no 64-bit mode and
  hence needs no compat handlers
- arch/um/drivers/hostaudio_kern.c may run in 64-bit mode with
  32-bit x86 user space underneath it. This rare corner case is
  the only one that still needs the compat handlers.

By adding a simple redirect of .compat_ioctl to .unlocked_ioctl in the
UML driver, we can remove all the COMPATIBLE_IOCTL() annotations without
a change in functionality. For completeness, I'm adding the same thing
to the dmasound file, knowing that it makes no difference.

The compat_ioctl list contains one comment about SNDCTL_DSP_MAPINBUF and
SNDCTL_DSP_MAPOUTBUF, which actually would need a translation handler
if implemented. However, the native implementation just returns -EINVAL,
so we don't care.
Reviewed-by: NTakashi Iwai <tiwai@suse.de>
Signed-off-by: NArnd Bergmann <arnd@arndb.de>

The two drivers implementing these both gained proper compat_ioctl()
handlers a long time ago with commits bb6c8d8f ("HID: hiddev:
Add 32bit ioctl compatibilty") and ae5e49c7 ("HID: hidraw: add
compatibility ioctl() for 32-bit applications."), so the lists in
fs/compat_ioctl.c are no longer used.

It appears that the lists were also incomplete, so the translation
didn't actually work correctly when it was still in use.

Remove them as cleanup.

Cc: linux-bluetooth@vger.kernel.org
Cc: Marcel Holtmann <marcel@holtmann.org>
Cc: Johan Hedberg <johan.hedberg@gmail.com>
Signed-off-by: NArnd Bergmann <arnd@arndb.de>

As of commit f0193d3e ("change semantics of ldisc ->compat_ioctl()"),
all hciuart ioctl commands are handled correctly in the driver, and we
never need to go through the table here.

Cc: linux-bluetooth@vger.kernel.org
Cc: Marcel Holtmann <marcel@holtmann.org>
Cc: Johan Hedberg <johan.hedberg@gmail.com>
Signed-off-by: NArnd Bergmann <arnd@arndb.de>

All these ioctl commands are compatible, so we can handle
them with a trivial wrapper in hci_sock.c and remove
the listing in fs/compat_ioctl.c.

A few of the commands pass integer arguments instead of
pointers, so for correctness skip the compat_ptr() conversion
here.
Acked-by: NMarcel Holtmann <marcel@holtmann.org>
Signed-off-by: NArnd Bergmann <arnd@arndb.de>

All these ioctl commands are compatible, so we can handle
them with a trivial wrapper in rfcomm/sock.c and remove
the listing in fs/compat_ioctl.c.
Acked-by: NMarcel Holtmann <marcel@holtmann.org>
Signed-off-by: NArnd Bergmann <arnd@arndb.de>

Neither the old isdn4linux interface nor the newer mISDN stack
ever had working 32-bit compat mode as far as I can tell.

However, the CAPI stack has some ioctl commands that are
correctly listed in fs/compat_ioctl.c.

We can trivially move all of those into the corresponding
file that implement the native handlers by adding a compat_ioctl
redirect to that.

I did notice that treating CAPI_MANUFACTURER_CMD() as compatible
is broken, so I'm also adding a handler for that, realizing that
in all likelyhood, nobody is ever going to call it.

Cc: Karsten Keil <isdn@linux-pingi.de>
Cc: netdev@vger.kernel.org
Cc: isdn4linux@listserv.isdn4linux.de
Signed-off-by: NArnd Bergmann <arnd@arndb.de>

These are two obscure ioctl commands, in a driver that only
has compatible commands, so just let the driver handle this
itself.
Acked-by: NBartlomiej Zolnierkiewicz <b.zolnierkie@samsung.com>
Signed-off-by: NArnd Bergmann <arnd@arndb.de>

MTIOCPOS and MTIOCGET are incompatible between 32-bit and 64-bit user
space, and traditionally have been translated in fs/compat_ioctl.c.

To get rid of that translation handler, move a corresponding
implementation into each of the four drivers implementing those commands.

The interesting part of that is now in a new linux/mtio.h header that
wraps the existing uapi/linux/mtio.h header and provides an abstraction
to let drivers handle both cases easily. Using an in_compat_syscall()
check, the caller does not have to keep track of whether this was
called through .unlocked_ioctl() or .compat_ioctl().
Acked-by: NHeiko Carstens <heiko.carstens@de.ibm.com>
Cc: "Kai Mäkisara" <Kai.Makisara@kolumbus.fi>
Cc: linux-scsi@vger.kernel.org
Cc: "James E.J. Bottomley" <jejb@linux.ibm.com>
Cc: "Martin K. Petersen" <martin.petersen@oracle.com>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: NArnd Bergmann <arnd@arndb.de>

A handful of drivers all have a trivial wrapper around their ioctl
handler, but don't call the compat_ptr() conversion function at the
moment. In practice this does not matter, since none of them are used
on the s390 architecture and for all other architectures, compat_ptr()
does not do anything, but using the new compat_ptr_ioctl()
helper makes it more correct in theory, and simplifies the code.

I checked that all ioctl handlers in these files are compatible
and take either pointer arguments or no argument.
Acked-by: NAl Viro <viro@zeniv.linux.org.uk>
Acked-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Acked-by: NAndrew Donnellan <andrew.donnellan@au1.ibm.com>
Acked-by: NFelipe Balbi <felipe.balbi@linux.intel.com>
Signed-off-by: NArnd Bergmann <arnd@arndb.de>

The .ioctl and .compat_ioctl file operations have the same prototype so
they can both point to the same function, which works great almost all
the time when all the commands are compatible.

One exception is the s390 architecture, where a compat pointer is only
31 bit wide, and converting it into a 64-bit pointer requires calling
compat_ptr(). Most drivers here will never run in s390, but since we now
have a generic helper for it, it's easy enough to use it consistently.

I double-checked all these drivers to ensure that all ioctl arguments
are used as pointers or are ignored, but are not interpreted as integer
values.
Acked-by: NJason Gunthorpe <jgg@mellanox.com>
Acked-by: NDaniel Vetter <daniel.vetter@ffwll.ch>
Acked-by: NMauro Carvalho Chehab <mchehab+samsung@kernel.org>
Acked-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Acked-by: NDavid Sterba <dsterba@suse.com>
Acked-by: NDarren Hart (VMware) <dvhart@infradead.org>
Acked-by: NJonathan Cameron <Jonathan.Cameron@huawei.com>
Acked-by: NBjorn Andersson <bjorn.andersson@linaro.org>
Acked-by: NDan Williams <dan.j.williams@intel.com>
Signed-off-by: NArnd Bergmann <arnd@arndb.de>

Each of these drivers has a copy of the same trivial helper function to
convert the pointer argument and then call the native ioctl handler.

We now have a generic implementation of that, so use it.
Acked-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Acked-by: NMichael S. Tsirkin <mst@redhat.com>
Acked-by: NDavid S. Miller <davem@davemloft.net>
Acked-by: NJarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
Reviewed-by: NJarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
Reviewed-by: NJason Gunthorpe <jgg@mellanox.com>
Reviewed-by: NJiri Kosina <jkosina@suse.cz>
Reviewed-by: NStefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: NCornelia Huck <cohuck@redhat.com>
Signed-off-by: NArnd Bergmann <arnd@arndb.de>

We no longer need the rtc compat handling to be in common code, now that
all drivers are either moved to the rtc-class framework, or (rarely)
exist in drivers/char for architectures without compat mode (m68k,
alpha and ia64, respectively).

I checked the list of ioctl commands in drivers, and the ones that are
not already handled are all compatible, again with the one exception of
m68k driver, which implements RTC_PLL_GET and RTC_PLL_SET, but has no
compat mode.

Unlike earlier versions of this patch, I'm now adding a separate
compat_ioctl handler that takes care of RTC_IRQP_READ32/RTC_IRQP_SET32
and treats all other commands as compatible, leaving the native
behavior unchanged.

The old conversion handler also deals with RTC_EPOCH_READ and
RTC_EPOCH_SET, which are not handled in rtc-dev.c but only in a single
device driver (rtc-vr41xx), so I'm adding the compat version in the same
place. I don't expect other drivers to need those commands in the future.
Acked-by: NAlexandre Belloni <alexandre.belloni@bootlin.com>
Reviewed-by: NBen Hutchings <ben.hutchings@codethink.co.uk>
Signed-off-by: NArnd Bergmann <arnd@arndb.de>
---
v4: handle RTC_EPOCH_SET32 in rtc_dev_compat_ioctl
v3: handle RTC_IRQP_READ32/RTC_IRQP_SET32 in rtc_dev_compat_ioctl
v2: merge compat handler into ioctl function to avoid the
    compat_alloc_user_space() roundtrip, based on feedback
    from Al Viro.

The ceph_ioctl function is used both for files and directories, but only
the files support doing that in 32-bit compat mode.

On the s390 architecture, there is also a problem with invalid 31-bit
pointers that need to be passed through compat_ptr().

Use the new compat_ptr_ioctl() to address both issues.

Note: When backporting this patch to stable kernels, "compat_ioctl:
add compat_ptr_ioctl()" is needed as well.
Reviewed-by: N"Yan, Zheng" <zyan@redhat.com>
Cc: stable@vger.kernel.org
Signed-off-by: NArnd Bergmann <arnd@arndb.de>

Handle ioctls that might be handled without reaching ->ioctl() in
native case on the top level there.  The counterpart of vfs_ioctl()
(i.e. calling ->unlock_ioctl(), etc.) left as-is; eventually
that would turn simply into the call of ->compat_ioctl(), but
that'll take more work.  Once that is done, we can move the
remains of compat_sys_ioctl() into fs/ioctl.c and finally bury
fs/compat_ioctl.c.
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>
Signed-off-by: NArnd Bergmann <arnd@arndb.de>

... and lose the ridiculous games with compat_alloc_user_space()
there.
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>
Signed-off-by: NArnd Bergmann <arnd@arndb.de>

casting to pointer to int, only to pass that to function that
takes pointer to void and uses it as pointer to structure is
really asking for trouble.

"Some pointer, I'm not sure what to" is spelled "void *",
not "int *"; use that.

And declare the functions we are passing that pointer to
as taking the pointer to what they really want to access.
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>
Signed-off-by: NArnd Bergmann <arnd@arndb.de>

... and hadn't for a long time.
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>
Signed-off-by: NArnd Bergmann <arnd@arndb.de>

it takes a pointer argument, regular file or no regular file
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>
Signed-off-by: NArnd Bergmann <arnd@arndb.de>

Unlike FICLONE, all of those take a pointer argument; they do need
compat_ptr() applied to arg.

Fixes: d79bdd52 ("vfs: wire up compat ioctl for CLONE/CLONE_RANGE")
Fixes: 54dbc151 ("vfs: hoist the btrfs deduplication ioctl to the vfs")
Fixes: ceac204e ("fs: make fiemap work from compat_ioctl")
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>
Signed-off-by: NArnd Bergmann <arnd@arndb.de>

Many drivers have ioctl() handlers that are completely compatible between
32-bit and 64-bit architectures, except for the argument that is passed
down from user space and may have to be passed through compat_ptr()
in order to become a valid 64-bit pointer.

Using ".compat_ptr = compat_ptr_ioctl" in file operations should let
us simplify a lot of those drivers to avoid #ifdef checks, and convert
additional drivers that don't have proper compat handling yet.

On most architectures, the compat_ptr_ioctl() just passes all arguments
to the corresponding ->ioctl handler. The exception is arch/s390, where
compat_ptr() clears the top bit of a 32-bit pointer value, so user space
pointers to the second 2GB alias the first 2GB, as is the case for native
32-bit s390 user space.

The compat_ptr_ioctl() function must therefore be used only with
ioctl functions that either ignore the argument or pass a pointer to a
compatible data type.

If any ioctl command handled by fops->unlocked_ioctl passes a plain
integer instead of a pointer, or any of the passed data types is
incompatible between 32-bit and 64-bit architectures, a proper handler
is required instead of compat_ptr_ioctl.
Signed-off-by: NArnd Bergmann <arnd@arndb.de>
---
v3: add a better description
v2: use compat_ptr_ioctl instead of generic_compat_ioctl_ptrarg,
as suggested by Al Viro

In commit 4ed28639 ("fs, elf: drop MAP_FIXED usage from elf_map") we
changed elf to use MAP_FIXED_NOREPLACE instead of MAP_FIXED for the
executable mappings.

Then, people reported that it broke some binaries that had overlapping
segments from the same file, and commit ad55eac7 ("elf: enforce
MAP_FIXED on overlaying elf segments") re-instated MAP_FIXED for some
overlaying elf segment cases.  But only some - despite the summary line
of that commit, it only did it when it also does a temporary brk vma for
one obvious overlapping case.

Now Russell King reports another overlapping case with old 32-bit x86
binaries, which doesn't trigger that limited case.  End result: we had
better just drop MAP_FIXED_NOREPLACE entirely, and go back to MAP_FIXED.

Yes, it's a sign of old binaries generated with old tool-chains, but we
do pride ourselves on not breaking existing setups.

This still leaves MAP_FIXED_NOREPLACE in place for the load_elf_interp()
and the old load_elf_library() use-cases, because nobody has reported
breakage for those. Yet.

Note that in all the cases seen so far, the overlapping elf sections
seem to be just re-mapping of the same executable with different section
attributes.  We could possibly introduce a new MAP_FIXED_NOFILECHANGE
flag or similar, which acts like NOREPLACE, but allows just remapping
the same executable file using different protection flags.

It's not clear that would make a huge difference to anything, but if
people really hate that "elf remaps over previous maps" behavior, maybe
at least a more limited form of remapping would alleviate some concerns.

Alternatively, we should take a look at our elf_map() logic to see if we
end up not mapping things properly the first time.

In the meantime, this is the minimal "don't do that then" patch while
people hopefully think about it more.
Reported-by: NRussell King <linux@armlinux.org.uk>
Fixes: 4ed28639 ("fs, elf: drop MAP_FIXED usage from elf_map")
Fixes: ad55eac7 ("elf: enforce  MAP_FIXED on overlaying elf segments")
Cc: Michal Hocko <mhocko@suse.com>
Cc: Kees Cook <keescook@chromium.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Pull dma-mapping regression fix from Christoph Hellwig:
 "Revert an incorret hunk from a patch that caused problems on various
  arm boards (Andrey Smirnov)"

* tag 'dma-mapping-5.4-1' of git://git.infradead.org/users/hch/dma-mapping:
  dma-mapping: fix false positive warnings in dma_common_free_remap()

Pull ARM SoC fixes from Olof Johansson:
 "A few fixes this time around:

   - Fixup of some clock specifications for DRA7 (device-tree fix)

   - Removal of some dead/legacy CPU OPP/PM code for OMAP that throws
     warnings at boot

   - A few more minor fixups for OMAPs, most around display

   - Enable STM32 QSPI as =y since their rootfs sometimes comes from
     there

   - Switch CONFIG_REMOTEPROC to =y since it went from tristate to bool

   - Fix of thermal zone definition for ux500 (5.4 regression)"

* tag 'armsoc-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/soc/soc:
  ARM: multi_v7_defconfig: Fix SPI_STM32_QSPI support
  ARM: dts: ux500: Fix up the CPU thermal zone
  arm64/ARM: configs: Change CONFIG_REMOTEPROC from m to y
  ARM: dts: am4372: Set memory bandwidth limit for DISPC
  ARM: OMAP2+: Fix warnings with broken omap2_set_init_voltage()
  ARM: OMAP2+: Add missing LCDC midlemode for am335x
  ARM: OMAP2+: Fix missing reset done flag for am3 and am43
  ARM: dts: Fix gpio0 flags for am335x-icev2
  ARM: omap2plus_defconfig: Enable more droid4 devices as loadable modules
  ARM: omap2plus_defconfig: Enable DRM_TI_TFP410
  DTS: ARM: gta04: introduce legacy spi-cs-high to make display work again
  ARM: dts: Fix wrong clocks for dra7 mcasp
  clk: ti: dra7: Fix mcasp8 clock bits

Merge tag 'kbuild-fixes-v5.4' of git://git.kernel.org/pub/scm/linux/kernel/git/masahiroy/linux-kbuild

Pull Kbuild fixes from Masahiro Yamada:

 - remove unneeded ar-option and KBUILD_ARFLAGS

 - remove long-deprecated SUBDIRS

 - fix modpost to suppress false-positive warnings for UML builds

 - fix namespace.pl to handle relative paths to ${objtree}, ${srctree}

 - make setlocalversion work for /bin/sh

 - make header archive reproducible

 - fix some Makefiles and documents

* tag 'kbuild-fixes-v5.4' of git://git.kernel.org/pub/scm/linux/kernel/git/masahiroy/linux-kbuild:
  kheaders: make headers archive reproducible
  kbuild: update compile-test header list for v5.4-rc2
  kbuild: two minor updates for Documentation/kbuild/modules.rst
  scripts/setlocalversion: clear local variable to make it work for sh
  namespace: fix namespace.pl script to support relative paths
  video/logo: do not generate unneeded logo C files
  video/logo: remove unneeded *.o pattern from clean-files
  integrity: remove pointless subdir-$(CONFIG_...)
  integrity: remove unneeded, broken attempt to add -fshort-wchar
  modpost: fix static EXPORT_SYMBOL warnings for UML build
  kbuild: correct formatting of header in kbuild module docs
  kbuild: remove SUBDIRS support
  kbuild: remove ar-option and KBUILD_ARFLAGS

Pull SCSI fixes from James Bottomley:
 "Twelve patches mostly small but obvious fixes or cosmetic but small
  updates"

* tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi:
  scsi: qla2xxx: Fix Nport ID display value
  scsi: qla2xxx: Fix N2N link up fail
  scsi: qla2xxx: Fix N2N link reset
  scsi: qla2xxx: Optimize NPIV tear down process
  scsi: qla2xxx: Fix stale mem access on driver unload
  scsi: qla2xxx: Fix unbound sleep in fcport delete path.
  scsi: qla2xxx: Silence fwdump template message
  scsi: hisi_sas: Make three functions static
  scsi: megaraid: disable device when probe failed after enabled device
  scsi: storvsc: setup 1:1 mapping between hardware queue and CPU queue
  scsi: qedf: Remove always false 'tmp_prio < 0' statement
  scsi: ufs: skip shutdown if hba is not powered
  scsi: bnx2fc: Handle scope bits when array returns BUSY or TSF

This makes getdents() and getdents64() do sanity checking on the
pathname that it gives to user space.  And to mitigate the performance
impact of that, it first cleans up the way it does the user copying, so
that the code avoids doing the SMAP/PAN updates between each part of the
dirent structure write.

I really wanted to do this during the merge window, but didn't have
time.  The conversion of filldir to unsafe_put_user() is something I've
had around for years now in a private branch, but the extra pathname
checking finally made me clean it up to the point where it is mergable.

It's worth noting that the filename validity checking really should be a
bit smarter: it would be much better to delay the error reporting until
the end of the readdir, so that non-corrupted filenames are still
returned.  But that involves bigger changes, so let's see if anybody
actually hits the corrupt directory entry case before worrying about it
further.

* branch 'readdir':
  Make filldir[64]() verify the directory entry filename is valid
  Convert filldir[64]() from __put_user() to unsafe_put_user()

This has been discussed several times, and now filesystem people are
talking about doing it individually at the filesystem layer, so head
that off at the pass and just do it in getdents{64}().

This is partially based on a patch by Jann Horn, but checks for NUL
bytes as well, and somewhat simplified.

There's also commentary about how it might be better if invalid names
due to filesystem corruption don't cause an immediate failure, but only
an error at the end of the readdir(), so that people can still see the
filenames that are ok.

There's also been discussion about just how much POSIX strictly speaking
requires this since it's about filesystem corruption.  It's really more
"protect user space from bad behavior" as pointed out by Jann.  But
since Eric Biederman looked up the POSIX wording, here it is for context:

 "From readdir:

   The readdir() function shall return a pointer to a structure
   representing the directory entry at the current position in the
   directory stream specified by the argument dirp, and position the
   directory stream at the next entry. It shall return a null pointer
   upon reaching the end of the directory stream. The structure dirent
   defined in the <dirent.h> header describes a directory entry.

  From definitions:

   3.129 Directory Entry (or Link)

   An object that associates a filename with a file. Several directory
   entries can associate names with the same file.

  ...

   3.169 Filename

   A name consisting of 1 to {NAME_MAX} bytes used to name a file. The
   characters composing the name may be selected from the set of all
   character values excluding the slash character and the null byte. The
   filenames dot and dot-dot have special meaning. A filename is
   sometimes referred to as a 'pathname component'."

Note that I didn't bother adding the checks to any legacy interfaces
that nobody uses.

Also note that if this ends up being noticeable as a performance
regression, we can fix that to do a much more optimized model that
checks for both NUL and '/' at the same time one word at a time.

We haven't really tended to optimize 'memchr()', and it only checks for
one pattern at a time anyway, and we really _should_ check for NUL too
(but see the comment about "soft errors" in the code about why it
currently only checks for '/')

See the CONFIG_DCACHE_WORD_ACCESS case of hash_name() for how the name
lookup code looks for pathname terminating characters in parallel.

Link: https://lore.kernel.org/lkml/20190118161440.220134-2-jannh@google.com/
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Jann Horn <jannh@google.com>
Cc: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

We really should avoid the "__{get,put}_user()" functions entirely,
because they can easily be mis-used and the original intent of being
used for simple direct user accesses no longer holds in a post-SMAP/PAN
world.

Manually optimizing away the user access range check makes no sense any
more, when the range check is generally much cheaper than the "enable
user accesses" code that the __{get,put}_user() functions still need.

So instead of __put_user(), use the unsafe_put_user() interface with
user_access_{begin,end}() that really does generate better code these
days, and which is generally a nicer interface.  Under some loads, the
multiple user writes that filldir() does are actually quite noticeable.

This also makes the dirent name copy use unsafe_put_user() with a couple
of macros.  We do not want to make function calls with SMAP/PAN
disabled, and the code this generates is quite good when the
architecture uses "asm goto" for unsafe_put_user() like x86 does.

Note that this doesn't bother with the legacy cases.  Nobody should use
them anyway, so performance doesn't really matter there.
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>