xen/gntdev: Accommodate VMA splitting

Prior to this commit, the gntdev driver code did not handle the
following scenario correctly with paravirtualized (PV) Xen domains:

* User process sets up a gntdev mapping composed of two grant mappings
(i.e., two pages shared by another Xen domain).
* User process munmap()s one of the pages.
* User process munmap()s the remaining page.
* User process exits.

In the scenario above, the user process would cause the kernel to log
the following messages in dmesg for the first munmap(), and the second
munmap() call would result in similar log messages:

BUG: Bad page map in process doublemap.test pte:... pmd:...
page:0000000057c97bff refcount:1 mapcount:-1 \
mapping:0000000000000000 index:0x0 pfn:...
...
page dumped because: bad pte
...
file:gntdev fault:0x0 mmap:gntdev_mmap [xen_gntdev] readpage:0x0
...
Call Trace:
<TASK>
dump_stack_lvl+0x46/0x5e
print_bad_pte.cold+0x66/0xb6
unmap_page_range+0x7e5/0xdc0
unmap_vmas+0x78/0xf0
unmap_region+0xa8/0x110
__do_munmap+0x1ea/0x4e0
__vm_munmap+0x75/0x120
__x64_sys_munmap+0x28/0x40
do_syscall_64+0x38/0x90
entry_SYSCALL_64_after_hwframe+0x61/0xcb
...

For each munmap() call, the Xen hypervisor (if built with CONFIG_DEBUG)
would print out the following and trigger a general protection fault in
the affected Xen PV domain:

(XEN) d0v... Attempt to implicitly unmap d0's grant PTE ...
(XEN) d0v... Attempt to implicitly unmap d0's grant PTE ...

As of this writing, gntdev_grant_map structure's vma field (referred to
as map->vma below) is mainly used for checking the start and end
addresses of mappings. However, with split VMAs, these may change, and
there could be more than one VMA associated with a gntdev mapping.
Hence, remove the use of map->vma and rely on map->pages_vm_start for
the original start address and on (map->count << PAGE_SHIFT) for the
original mapping size. Let the invalidate() and find_special_page()
hooks use these.

Also, given that there can be multiple VMAs associated with a gntdev
mapping, move the "mmu_interval_notifier_remove(&map->notifier)" call to
the end of gntdev_put_map, so that the MMU notifier is only removed
after the closing of the last remaining VMA.

Finally, use an atomic to prevent inadvertent gntdev mapping re-use,
instead of using the map->live_grants atomic counter and/or the map->vma
pointer (the latter of which is now removed). This prevents the
userspace from mmap()'ing (with MAP_FIXED) a gntdev mapping over the
same address range as a previously set up gntdev mapping. This scenario
can be summarized with the following call-trace, which was valid prior
to this commit:

mmap
gntdev_mmap
mmap (repeat mmap with MAP_FIXED over the same address range)
gntdev_invalidate
unmap_grant_pages (sets 'being_removed' entries to true)
gnttab_unmap_refs_async
unmap_single_vma
gntdev_mmap (maps the shared pages again)
munmap
gntdev_invalidate
unmap_grant_pages
(no-op because 'being_removed' entries are true)
unmap_single_vma (For PV domains, Xen reports that a granted page
is being unmapped and triggers a general protection fault in the
affected domain, if Xen was built with CONFIG_DEBUG)

The fix for this last scenario could be worth its own commit, but we
opted for a single commit, because removing the gntdev_grant_map
structure's vma field requires guarding the entry to gntdev_mmap(), and
the live_grants atomic counter is not sufficient on its own to prevent
the mmap() over a pre-existing mapping.

Link: https://github.com/QubesOS/qubes-issues/issues/7631
Fixes: ab31523c2fca ("xen/gntdev: allow usermode to map granted pages")
Cc: stable@vger.kernel.org
Signed-off-by: M. Vefa Bicakci <m.v.b@runbox.com>
Reviewed-by: Juergen Gross <jgross@suse.com>
Link: https://lore.kernel.org/r/20221002222006.2077-3-m.v.b@runbox.com
Signed-off-by: Juergen Gross <jgross@suse.com>

xen/gntdev: Avoid blocking in unmap_grant_pages()

unmap_grant_pages() currently waits for the pages to no longer be used.
In https://github.com/QubesOS/qubes-issues/issues/7481, this lead to a
deadlock against i915: i915 was waiting for gntdev's MMU notifier to
finish, while gntdev was waiting for i915 to free its pages. I also
believe this is responsible for various deadlocks I have experienced in
the past.

Avoid these problems by making unmap_grant_pages async. This requires
making it return void, as any errors will not be available when the
function returns. Fortunately, the only use of the return value is a
WARN_ON(), which can be replaced by a WARN_ON when the error is
detected. Additionally, a failed call will not prevent further calls
from being made, but this is harmless.

Because unmap_grant_pages is now async, the grant handle will be sent to
INVALID_GRANT_HANDLE too late to prevent multiple unmaps of the same
handle. Instead, a separate bool array is allocated for this purpose.
This wastes memory, but stuffing this information in padding bytes is
too fragile. Furthermore, it is necessary to grab a reference to the
map before making the asynchronous call, and release the reference when
the call returns.

It is also necessary to guard against reentrancy in gntdev_map_put(),
and to handle the case where userspace tries to map a mapping whose
contents have not all been freed yet.

Fixes: 745282256c75 ("xen/gntdev: safely unmap grants in case they are still in use")
Cc: stable@vger.kernel.org
Signed-off-by: Demi Marie Obenour <demi@invisiblethingslab.com>
Reviewed-by: Juergen Gross <jgross@suse.com>
Link: https://lore.kernel.org/r/20220622022726.2538-1-demi@invisiblethingslab.com
Signed-off-by: Juergen Gross <jgross@suse.com>

xen: Use evtchn_type_t as a type for event channels

Make event channel functions pass event channel port using
evtchn_port_t type. It eliminates signed <-> unsigned conversion.

Signed-off-by: Yan Yankovskyi <yyankovskyi@gmail.com>
Reviewed-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Link: https://lore.kernel.org/r/20200323152343.GA28422@kbp1-lhp-F74019
Signed-off-by: Juergen Gross <jgross@suse.com>

xen/gntdev: replace global limit of mapped pages by limit per call

Today there is a global limit of pages mapped via /dev/xen/gntdev set
to 1 million pages per default. There is no reason why that limit is
existing, as total number of grant mappings is limited by the
hypervisor anyway and preferring kernel mappings over userspace ones
doesn't make sense. It should be noted that the gntdev device is
usable by root only.

Additionally checking of that limit is fragile, as the number of pages
to map via one call is specified in a 32-bit unsigned variable which
isn't tested to stay within reasonable limits (the only test is the
value to be <= zero, which basically excludes only calls without any
mapping requested). So trying to map e.g. 0xffff0000 pages while
already nearly 1000000 pages are mapped will effectively lower the
global number of mapped pages such that a parallel call mapping a
reasonable amount of pages can succeed in spite of the global limit
being violated.

So drop the global limit and introduce per call limit instead. This
per call limit (default: 65536 grant mappings) protects against
allocating insane large arrays in the kernel for doing a hypercall
which will fail anyway in case a user is e.g. trying to map billions
of pages.

Signed-off-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Oleksandr Andrushchenko <oleksandr_andrushchenko@epam.com>
Reviewed-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Signed-off-by: Juergen Gross <jgross@suse.com>

xen/gntdev: use mmu_interval_notifier_insert

gntdev simply wants to monitor a specific VMA for any notifier events,
this can be done straightforwardly using mmu_interval_notifier_insert()
over the VMA's VA range.

The notifier should be attached until the original VMA is destroyed.

It is unclear if any of this is even sane, but at least a lot of duplicate
code is removed.

Link: https://lore.kernel.org/r/20191112202231.3856-15-jgg@ziepe.ca
Reviewed-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

xen/gntdev: Add initial support for dma-buf UAPI

Add UAPI and IOCTLs for dma-buf grant device driver extension:
the extension allows userspace processes and kernel modules to
use Xen backed dma-buf implementation. With this extension grant
references to the pages of an imported dma-buf can be exported
for other domain use and grant references coming from a foreign
domain can be converted into a local dma-buf for local export.
Implement basic initialization and stubs for Xen DMA buffers'
support.

Signed-off-by: Oleksandr Andrushchenko <oleksandr_andrushchenko@epam.com>
Reviewed-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>

xen/gntdev: Make private routines/structures accessible

This is in preparation for adding support of DMA buffer
functionality: make map/unmap related code and structures, used
privately by gntdev, ready for dma-buf extension, which will re-use
these. Rename corresponding structures as those become non-private
to gntdev now.

Signed-off-by: Oleksandr Andrushchenko <oleksandr_andrushchenko@epam.com>
Reviewed-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>