linux-stable.git/arch/sparc, branch linux-4.9.y Linux kernel stable tree sparc: use fallback for random_get_entropy() instead of zero 2022-06-25T09:45:13+00:00 Jason A. Donenfeld Jason@zx2c4.com 2022-04-08T16:03:13+00:00 b4cdbb50d8926feb46833efbd35f45f7342c8ba6 commit ac9756c79797bb98972736b13cfb239fd2cffb79 upstream. In the event that random_get_entropy() can't access a cycle counter or similar, falling back to returning 0 is really not the best we can do. Instead, at least calling random_get_entropy_fallback() would be preferable, because that always needs to return _something_, even falling back to jiffies eventually. It's not as though random_get_entropy_fallback() is super high precision or guaranteed to be entropic, but basically anything that's not zero all the time is better than returning zero all the time. This is accomplished by just including the asm-generic code like on other architectures, which means we can get rid of the empty stub function here. Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Arnd Bergmann <arnd@arndb.de> Cc: David S. Miller <davem@davemloft.net> Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> 
commit ac9756c79797bb98972736b13cfb239fd2cffb79 upstream.

In the event that random_get_entropy() can't access a cycle counter or
similar, falling back to returning 0 is really not the best we can do.
Instead, at least calling random_get_entropy_fallback() would be
preferable, because that always needs to return _something_, even
falling back to jiffies eventually. It's not as though
random_get_entropy_fallback() is super high precision or guaranteed to
be entropic, but basically anything that's not zero all the time is
better than returning zero all the time.

This is accomplished by just including the asm-generic code like on
other architectures, which means we can get rid of the empty stub
function here.

Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
gup: document and work around "COW can break either way" issue 2022-01-27T07:47:42+00:00 Linus Torvalds torvalds@linux-foundation.org 2022-01-24T13:41:50+00:00 0c29640bdecad332b9e2b884217c159f4aeb2556 commit 9bbd42e79720122334226afad9ddcac1c3e6d373 upstream. Doing a "get_user_pages()" on a copy-on-write page for reading can be ambiguous: the page can be COW'ed at any time afterwards, and the direction of a COW event isn't defined. Yes, whoever writes to it will generally do the COW, but if the thread that did the get_user_pages() unmapped the page before the write (and that could happen due to memory pressure in addition to any outright action), the writer could also just take over the old page instead. End result: the get_user_pages() call might result in a page pointer that is no longer associated with the original VM, and is associated with - and controlled by - another VM having taken it over instead. So when doing a get_user_pages() on a COW mapping, the only really safe thing to do would be to break the COW when getting the page, even when only getting it for reading. At the same time, some users simply don't even care. For example, the perf code wants to look up the page not because it cares about the page, but because the code simply wants to look up the physical address of the access for informational purposes, and doesn't really care about races when a page might be unmapped and remapped elsewhere. This adds logic to force a COW event by setting FOLL_WRITE on any copy-on-write mapping when FOLL_GET (or FOLL_PIN) is used to get a page pointer as a result. The current semantics end up being: - __get_user_pages_fast(): no change. If you don't ask for a write, you won't break COW. You'd better know what you're doing. - get_user_pages_fast(): the fast-case "look it up in the page tables without anything getting mmap_sem" now refuses to follow a read-only page, since it might need COW breaking. Which happens in the slow path - the fast path doesn't know if the memory might be COW or not. - get_user_pages() (including the slow-path fallback for gup_fast()): for a COW mapping, turn on FOLL_WRITE for FOLL_GET/FOLL_PIN, with very similar semantics to FOLL_FORCE. If it turns out that we want finer granularity (ie "only break COW when it might actually matter" - things like the zero page are special and don't need to be broken) we might need to push these semantics deeper into the lookup fault path. So if people care enough, it's possible that we might end up adding a new internal FOLL_BREAK_COW flag to go with the internal FOLL_COW flag we already have for tracking "I had a COW". Alternatively, if it turns out that different callers might want to explicitly control the forced COW break behavior, we might even want to make such a flag visible to the users of get_user_pages() instead of using the above default semantics. But for now, this is mostly commentary on the issue (this commit message being a lot bigger than the patch, and that patch in turn is almost all comments), with that minimal "enable COW breaking early" logic using the existing FOLL_WRITE behavior. [ It might be worth noting that we've always had this ambiguity, and it could arguably be seen as a user-space issue. You only get private COW mappings that could break either way in situations where user space is doing cooperative things (ie fork() before an execve() etc), but it _is_ surprising and very subtle, and fork() is supposed to give you independent address spaces. So let's treat this as a kernel issue and make the semantics of get_user_pages() easier to understand. Note that obviously a true shared mapping will still get a page that can change under us, so this does _not_ mean that get_user_pages() somehow returns any "stable" page ] [surenb: backport notes Replaced (gup_flags | FOLL_WRITE) with write=1 in gup_pgd_range. Removed FOLL_PIN usage in should_force_cow_break since it's missing in the earlier kernels.] Reported-by: Jann Horn <jannh@google.com> Tested-by: Christoph Hellwig <hch@lst.de> Acked-by: Oleg Nesterov <oleg@redhat.com> Acked-by: Kirill Shutemov <kirill@shutemov.name> Acked-by: Jan Kara <jack@suse.cz> Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: Matthew Wilcox <willy@infradead.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> [surenb: backport to 4.19 kernel] Cc: stable@vger.kernel.org # 4.19.x Signed-off-by: Suren Baghdasaryan <surenb@google.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> [bwh: Backported to 4.9: - Generic get_user_pages_fast() calls __get_user_pages_fast() here, so make it pass write=1 - Various architectures have their own implementations of get_user_pages_fast(), so apply the corresponding change there - Adjust context] Signed-off-by: Ben Hutchings <ben@decadent.org.uk> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> 
commit 9bbd42e79720122334226afad9ddcac1c3e6d373 upstream.

Doing a "get_user_pages()" on a copy-on-write page for reading can be
ambiguous: the page can be COW'ed at any time afterwards, and the
direction of a COW event isn't defined.

Yes, whoever writes to it will generally do the COW, but if the thread
that did the get_user_pages() unmapped the page before the write (and
that could happen due to memory pressure in addition to any outright
action), the writer could also just take over the old page instead.

End result: the get_user_pages() call might result in a page pointer
that is no longer associated with the original VM, and is associated
with - and controlled by - another VM having taken it over instead.

So when doing a get_user_pages() on a COW mapping, the only really safe
thing to do would be to break the COW when getting the page, even when
only getting it for reading.

At the same time, some users simply don't even care.

For example, the perf code wants to look up the page not because it
cares about the page, but because the code simply wants to look up the
physical address of the access for informational purposes, and doesn't
really care about races when a page might be unmapped and remapped
elsewhere.

This adds logic to force a COW event by setting FOLL_WRITE on any
copy-on-write mapping when FOLL_GET (or FOLL_PIN) is used to get a page
pointer as a result.

The current semantics end up being:

 - __get_user_pages_fast(): no change. If you don't ask for a write,
   you won't break COW. You'd better know what you're doing.

 - get_user_pages_fast(): the fast-case "look it up in the page tables
   without anything getting mmap_sem" now refuses to follow a read-only
   page, since it might need COW breaking.  Which happens in the slow
   path - the fast path doesn't know if the memory might be COW or not.

 - get_user_pages() (including the slow-path fallback for gup_fast()):
   for a COW mapping, turn on FOLL_WRITE for FOLL_GET/FOLL_PIN, with
   very similar semantics to FOLL_FORCE.

If it turns out that we want finer granularity (ie "only break COW when
it might actually matter" - things like the zero page are special and
don't need to be broken) we might need to push these semantics deeper
into the lookup fault path.  So if people care enough, it's possible
that we might end up adding a new internal FOLL_BREAK_COW flag to go
with the internal FOLL_COW flag we already have for tracking "I had a
COW".

Alternatively, if it turns out that different callers might want to
explicitly control the forced COW break behavior, we might even want to
make such a flag visible to the users of get_user_pages() instead of
using the above default semantics.

But for now, this is mostly commentary on the issue (this commit message
being a lot bigger than the patch, and that patch in turn is almost all
comments), with that minimal "enable COW breaking early" logic using the
existing FOLL_WRITE behavior.

[ It might be worth noting that we've always had this ambiguity, and it
  could arguably be seen as a user-space issue.

  You only get private COW mappings that could break either way in
  situations where user space is doing cooperative things (ie fork()
  before an execve() etc), but it _is_ surprising and very subtle, and
  fork() is supposed to give you independent address spaces.

  So let's treat this as a kernel issue and make the semantics of
  get_user_pages() easier to understand. Note that obviously a true
  shared mapping will still get a page that can change under us, so this
  does _not_ mean that get_user_pages() somehow returns any "stable"
  page ]

[surenb: backport notes
	Replaced (gup_flags | FOLL_WRITE) with write=1 in gup_pgd_range.
	Removed FOLL_PIN usage in should_force_cow_break since it's missing in
	the earlier kernels.]

Reported-by: Jann Horn <jannh@google.com>
Tested-by: Christoph Hellwig <hch@lst.de>
Acked-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Kirill Shutemov <kirill@shutemov.name>
Acked-by: Jan Kara <jack@suse.cz>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
[surenb: backport to 4.19 kernel]
Cc: stable@vger.kernel.org # 4.19.x
Signed-off-by: Suren Baghdasaryan <surenb@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[bwh: Backported to 4.9:
 - Generic get_user_pages_fast() calls __get_user_pages_fast() here,
   so make it pass write=1
 - Various architectures have their own implementations of
   get_user_pages_fast(), so apply the corresponding change there
 - Adjust context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
sparc64: fix pci_iounmap() when CONFIG_PCI is not set 2021-10-09T11:25:59+00:00 Linus Torvalds torvalds@linux-foundation.org 2021-09-20T17:56:32+00:00 dc027c500af26867dbfdf1f8172f34d17ad02f55 [ Upstream commit d8b1e10a2b8efaf71d151aa756052fbf2f3b6d57 ] Guenter reported [1] that the pci_iounmap() changes remain problematic, with sparc64 allnoconfig and tinyconfig still not building due to the header file changes and confusion with the arch-specific pci_iounmap() implementation. I'm pretty convinced that sparc should just use GENERIC_IOMAP instead of doing its own thing, since it turns out that the sparc64 version of pci_iounmap() is somewhat buggy (see [2]). But in the meantime, this just fixes the build by avoiding the trivial re-definition of the empty case. Link: https://lore.kernel.org/lkml/20210920134424.GA346531@roeck-us.net/ [1] Link: https://lore.kernel.org/lkml/CAHk-=wgheheFx9myQyy5osh79BAazvmvYURAtub2gQtMvLrhqQ@mail.gmail.com/ [2] Reported-by: Guenter Roeck <linux@roeck-us.net> Cc: David Miller <davem@davemloft.net> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Sasha Levin <sashal@kernel.org> 
[ Upstream commit d8b1e10a2b8efaf71d151aa756052fbf2f3b6d57 ]

Guenter reported [1] that the pci_iounmap() changes remain problematic,
with sparc64 allnoconfig and tinyconfig still not building due to the
header file changes and confusion with the arch-specific pci_iounmap()
implementation.

I'm pretty convinced that sparc should just use GENERIC_IOMAP instead of
doing its own thing, since it turns out that the sparc64 version of
pci_iounmap() is somewhat buggy (see [2]).  But in the meantime, this
just fixes the build by avoiding the trivial re-definition of the empty
case.

Link: https://lore.kernel.org/lkml/20210920134424.GA346531@roeck-us.net/ [1]
Link: https://lore.kernel.org/lkml/CAHk-=wgheheFx9myQyy5osh79BAazvmvYURAtub2gQtMvLrhqQ@mail.gmail.com/ [2]
Reported-by: Guenter Roeck <linux@roeck-us.net>
Cc: David Miller <davem@davemloft.net>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
sparc: avoid stringop-overread errors 2021-10-06T08:23:40+00:00 Linus Torvalds torvalds@linux-foundation.org 2021-09-06T23:06:04+00:00 fde522989a3e9cf32e7d76338fa8af018aeb8cd5 [ Upstream commit fc7c028dcdbfe981bca75d2a7b95f363eb691ef3 ] The sparc mdesc code does pointer games with 'struct mdesc_hdr', but didn't describe to the compiler how that header is then followed by the data that the header describes. As a result, gcc is now unhappy since it does stricter pointer range tracking, and doesn't understand about how these things work. This results in various errors like: arch/sparc/kernel/mdesc.c: In function ‘mdesc_node_by_name’: arch/sparc/kernel/mdesc.c:647:22: error: ‘strcmp’ reading 1 or more bytes from a region of size 0 [-Werror=stringop-overread] 647 | if (!strcmp(names + ep[ret].name_offset, name)) | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ which are easily avoided by just describing 'struct mdesc_hdr' better, and making the node_block() helper function look into that unsized data[] that follows the header. This makes the sparc64 build happy again at least for my cross-compiler version (gcc version 11.2.1). Link: https://lore.kernel.org/lkml/CAHk-=wi4NW3NC0xWykkw=6LnjQD6D_rtRtxY9g8gQAJXtQMi8A@mail.gmail.com/ Cc: Guenter Roeck <linux@roeck-us.net> Cc: David S. Miller <davem@davemloft.net> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Sasha Levin <sashal@kernel.org> 
[ Upstream commit fc7c028dcdbfe981bca75d2a7b95f363eb691ef3 ]

The sparc mdesc code does pointer games with 'struct mdesc_hdr', but
didn't describe to the compiler how that header is then followed by the
data that the header describes.

As a result, gcc is now unhappy since it does stricter pointer range
tracking, and doesn't understand about how these things work.  This
results in various errors like:

    arch/sparc/kernel/mdesc.c: In function ‘mdesc_node_by_name’:
    arch/sparc/kernel/mdesc.c:647:22: error: ‘strcmp’ reading 1 or more bytes from a region of size 0 [-Werror=stringop-overread]
      647 |                 if (!strcmp(names + ep[ret].name_offset, name))
          |                      ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

which are easily avoided by just describing 'struct mdesc_hdr' better,
and making the node_block() helper function look into that unsized
data[] that follows the header.

This makes the sparc64 build happy again at least for my cross-compiler
version (gcc version 11.2.1).

Link: https://lore.kernel.org/lkml/CAHk-=wi4NW3NC0xWykkw=6LnjQD6D_rtRtxY9g8gQAJXtQMi8A@mail.gmail.com/
Cc: Guenter Roeck <linux@roeck-us.net>
Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
sparc32: fix a user-triggerable oops in clear_user() 2021-03-03T16:44:44+00:00 Al Viro viro@zeniv.linux.org.uk 2020-07-20T01:21:51+00:00 00d0241c7631c9344a283f2e6e676903d2a387f0 commit 7780918b36489f0b2f9a3749d7be00c2ceaec513 upstream. Back in 2.1.29 the clear_user() guts (__bzero()) had been merged with memset(). Unfortunately, while all exception handlers had been copied, one of the exception table entries got lost. As the result, clear_user() starting at 128*n bytes before the end of page and spanning between 8 and 127 bytes into the next page would oops when the second page is unmapped. It's trivial to reproduce - all it takes is main() { int fd = open("/dev/zero", O_RDONLY); char *p = mmap(NULL, 16384, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANON, -1, 0); munmap(p + 8192, 8192); read(fd, p + 8192 - 128, 192); } which had been oopsing since March 1997. Says something about the quality of test coverage... ;-/ And while today sparc32 port is nearly dead, back in '97 it had been very much alive; in fact, sparc64 had only been in mainline for 3 months by that point... Cc: stable@kernel.org Fixes: v2.1.29 Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> 
commit 7780918b36489f0b2f9a3749d7be00c2ceaec513 upstream.

Back in 2.1.29 the clear_user() guts (__bzero()) had been merged
with memset().  Unfortunately, while all exception handlers had been
copied, one of the exception table entries got lost.  As the result,
clear_user() starting at 128*n bytes before the end of page and
spanning between 8 and 127 bytes into the next page would oops when
the second page is unmapped.  It's trivial to reproduce - all
it takes is

main()
{
	int fd = open("/dev/zero", O_RDONLY);
	char *p = mmap(NULL, 16384, PROT_READ|PROT_WRITE,
			MAP_PRIVATE|MAP_ANON, -1, 0);
	munmap(p + 8192, 8192);
	read(fd, p + 8192 - 128, 192);
}

which had been oopsing since March 1997.  Says something about
the quality of test coverage... ;-/  And while today sparc32 port
is nearly dead, back in '97 it had been very much alive; in fact,
sparc64 had only been in mainline for 3 months by that point...

Cc: stable@kernel.org
Fixes: v2.1.29
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
sparc64: only select COMPAT_BINFMT_ELF if BINFMT_ELF is set 2021-03-03T16:44:39+00:00 Randy Dunlap rdunlap@infradead.org 2020-11-26T00:40:11+00:00 4086dff96c3a2b4315477c960b5bd4bee44233bd [ Upstream commit 80bddf5c93a99e11fc9faf7e4b575d01cecd45d3 ] Currently COMPAT on SPARC64 selects COMPAT_BINFMT_ELF unconditionally, even when BINFMT_ELF is not enabled. This causes a kconfig warning. Instead, just select COMPAT_BINFMT_ELF if BINFMT_ELF is enabled. This builds cleanly with no kconfig warnings. WARNING: unmet direct dependencies detected for COMPAT_BINFMT_ELF Depends on [n]: COMPAT [=y] && BINFMT_ELF [=n] Selected by [y]: - COMPAT [=y] && SPARC64 [=y] Fixes: 26b4c912185a ("sparc,sparc64: unify Kconfig files") Signed-off-by: Randy Dunlap <rdunlap@infradead.org> Cc: "David S. Miller" <davem@davemloft.net> Cc: sparclinux@vger.kernel.org Cc: Sam Ravnborg <sam@ravnborg.org> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Sasha Levin <sashal@kernel.org> 
[ Upstream commit 80bddf5c93a99e11fc9faf7e4b575d01cecd45d3 ]

Currently COMPAT on SPARC64 selects COMPAT_BINFMT_ELF unconditionally,
even when BINFMT_ELF is not enabled. This causes a kconfig warning.

Instead, just select COMPAT_BINFMT_ELF if BINFMT_ELF is enabled.
This builds cleanly with no kconfig warnings.

WARNING: unmet direct dependencies detected for COMPAT_BINFMT_ELF
  Depends on [n]: COMPAT [=y] && BINFMT_ELF [=n]
  Selected by [y]:
  - COMPAT [=y] && SPARC64 [=y]

Fixes: 26b4c912185a ("sparc,sparc64: unify Kconfig files")
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: sparclinux@vger.kernel.org
Cc: Sam Ravnborg <sam@ravnborg.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
sparc64: remove mm_cpumask clearing to fix kthread_use_mm race 2020-11-10T09:23:52+00:00 Nicholas Piggin npiggin@gmail.com 2020-09-14T04:52:18+00:00 0e6277eb590c8f6feff41822bf27bad8b71e6135 [ Upstream commit bafb056ce27940c9994ea905336aa8f27b4f7275 ] The de facto (and apparently uncommented) standard for using an mm had, thanks to this code in sparc if nothing else, been that you must have a reference on mm_users *and that reference must have been obtained with mmget()*, i.e., from a thread with a reference to mm_users that had used the mm. The introduction of mmget_not_zero() in commit d2005e3f41d4 ("userfaultfd: don't pin the user memory in userfaultfd_file_create()") allowed mm_count holders to aoperate on user mappings asynchronously from the actual threads using the mm, but they were not to load those mappings into their TLB (i.e., walking vmas and page tables is okay, kthread_use_mm() is not). io_uring 2b188cc1bb857 ("Add io_uring IO interface") added code which does a kthread_use_mm() from a mmget_not_zero() refcount. The problem with this is code which previously assumed mm == current->mm and mm->mm_users == 1 implies the mm will remain single-threaded at least until this thread creates another mm_users reference, has now broken. arch/sparc/kernel/smp_64.c: if (atomic_read(&mm->mm_users) == 1) { cpumask_copy(mm_cpumask(mm), cpumask_of(cpu)); goto local_flush_and_out; } vs fs/io_uring.c if (unlikely(!(ctx->flags & IORING_SETUP_SQPOLL) || !mmget_not_zero(ctx->sqo_mm))) return -EFAULT; kthread_use_mm(ctx->sqo_mm); mmget_not_zero() could come in right after the mm_users == 1 test, then kthread_use_mm() which sets its CPU in the mm_cpumask. That update could be lost if cpumask_copy() occurs afterward. I propose we fix this by allowing mmget_not_zero() to be a first-class reference, and not have this obscure undocumented and unchecked restriction. The basic fix for sparc64 is to remove its mm_cpumask clearing code. The optimisation could be effectively restored by sending IPIs to mm_cpumask members and having them remove themselves from mm_cpumask. This is more tricky so I leave it as an exercise for someone with a sparc64 SMP. powerpc has a (currently similarly broken) example. Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Acked-by: David S. Miller <davem@davemloft.net> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200914045219.3736466-4-npiggin@gmail.com Signed-off-by: Sasha Levin <sashal@kernel.org> 
[ Upstream commit bafb056ce27940c9994ea905336aa8f27b4f7275 ]

The de facto (and apparently uncommented) standard for using an mm had,
thanks to this code in sparc if nothing else, been that you must have a
reference on mm_users *and that reference must have been obtained with
mmget()*, i.e., from a thread with a reference to mm_users that had used
the mm.

The introduction of mmget_not_zero() in commit d2005e3f41d4
("userfaultfd: don't pin the user memory in userfaultfd_file_create()")
allowed mm_count holders to aoperate on user mappings asynchronously
from the actual threads using the mm, but they were not to load those
mappings into their TLB (i.e., walking vmas and page tables is okay,
kthread_use_mm() is not).

io_uring 2b188cc1bb857 ("Add io_uring IO interface") added code which
does a kthread_use_mm() from a mmget_not_zero() refcount.

The problem with this is code which previously assumed mm == current->mm
and mm->mm_users == 1 implies the mm will remain single-threaded at
least until this thread creates another mm_users reference, has now
broken.

arch/sparc/kernel/smp_64.c:

    if (atomic_read(&mm->mm_users) == 1) {
        cpumask_copy(mm_cpumask(mm), cpumask_of(cpu));
        goto local_flush_and_out;
    }

vs fs/io_uring.c

    if (unlikely(!(ctx->flags & IORING_SETUP_SQPOLL) ||
                 !mmget_not_zero(ctx->sqo_mm)))
        return -EFAULT;
    kthread_use_mm(ctx->sqo_mm);

mmget_not_zero() could come in right after the mm_users == 1 test, then
kthread_use_mm() which sets its CPU in the mm_cpumask. That update could
be lost if cpumask_copy() occurs afterward.

I propose we fix this by allowing mmget_not_zero() to be a first-class
reference, and not have this obscure undocumented and unchecked
restriction.

The basic fix for sparc64 is to remove its mm_cpumask clearing code. The
optimisation could be effectively restored by sending IPIs to mm_cpumask
members and having them remove themselves from mm_cpumask. This is more
tricky so I leave it as an exercise for someone with a sparc64 SMP.
powerpc has a (currently similarly broken) example.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200914045219.3736466-4-npiggin@gmail.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
fix a braino in "sparc32: fix register window handling in genregs32_[gs]et()" 2020-06-30T19:38:38+00:00 Al Viro viro@zeniv.linux.org.uk 2020-06-07T03:44:24+00:00 307fa7afd3f7f8852e637303cc265298c9463296 [ Upstream commit 9d964e1b82d8182184153b70174f445ea616f053 ] lost npc in PTRACE_SETREGSET, breaking PTRACE_SETREGS as well Fixes: cf51e129b968 "sparc32: fix register window handling in genregs32_[gs]et()" Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: Sasha Levin <sashal@kernel.org> 
[ Upstream commit 9d964e1b82d8182184153b70174f445ea616f053 ]

lost npc in PTRACE_SETREGSET, breaking PTRACE_SETREGS as well

Fixes: cf51e129b968 "sparc32: fix register window handling in genregs32_[gs]et()"
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Sasha Levin <sashal@kernel.org>
sparc64: fix misuses of access_process_vm() in genregs32_[sg]et() 2020-06-20T08:24:21+00:00 Al Viro viro@zeniv.linux.org.uk 2020-05-17T19:37:50+00:00 3f948036c7f39e08f8c7170e000e466e85579b7c commit 142cd25293f6a7ecbdff4fb0af17de6438d46433 upstream. We do need access_process_vm() to access the target's reg_window. However, access to caller's memory (storing the result in genregs32_get(), fetching the new values in case of genregs32_set()) should be done by normal uaccess primitives. Fixes: ad4f95764040 ([SPARC64]: Fix user accesses in regset code.) Cc: stable@kernel.org Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> 
commit 142cd25293f6a7ecbdff4fb0af17de6438d46433 upstream.

We do need access_process_vm() to access the target's reg_window.
However, access to caller's memory (storing the result in
genregs32_get(), fetching the new values in case of genregs32_set())
should be done by normal uaccess primitives.

Fixes: ad4f95764040 ([SPARC64]: Fix user accesses in regset code.)
Cc: stable@kernel.org
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
sparc32: fix register window handling in genregs32_[gs]et() 2020-06-20T08:24:21+00:00 Al Viro viro@zeniv.linux.org.uk 2020-05-17T16:20:40+00:00 9c90a2d5693eeecbbb80e328e802a27ef65da42e commit cf51e129b96847f969bfb8af1ee1516a01a70b39 upstream. It needs access_process_vm() if the traced process does not share mm with the caller. Solution is similar to what sparc64 does. Note that genregs32_set() is only ever called with pos being 0 or 32 * sizeof(u32) (the latter - as part of PTRACE_SETREGS handling). Cc: stable@kernel.org Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> 
commit cf51e129b96847f969bfb8af1ee1516a01a70b39 upstream.

It needs access_process_vm() if the traced process does not share
mm with the caller.  Solution is similar to what sparc64 does.
Note that genregs32_set() is only ever called with pos being 0
or 32 * sizeof(u32) (the latter - as part of PTRACE_SETREGS
handling).

Cc: stable@kernel.org
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>