We don't actually need to return an output parameter providing mm sequence
number, rather we can separate that out into another function -
__vma_raw_mm_seqnum() - and have any callers which need to obtain that
invoke that instead.
The access to the raw sequence number requires that we hold the exclusive
mmap lock such that we know we can't race vma_end_write_all(), so move the
assert to __vma_raw_mm_seqnum() to make this requirement clear.
Also while we're here, convert all of the VM_BUG_ON_VMA()'s to
VM_WARN_ON_ONCE_VMA()'s in line with the convention that we do not invoke
oopses when we can avoid it.
[lorenzo.stoakes@oracle.com: minor tweaks, per Vlastimil]
Link: https://lkml.kernel.org/r/3fa89c13-232d-4eee-86cc-96caa75c2c67@lucifer.local
Link: https://lkml.kernel.org/r/ef6c415c2d2c03f529dca124ccaed66bc2f60edc.1769198904.git.lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Reviewed-by: Suren Baghdasaryan <surenb@google.com>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Waiman Long <longman@redhat.com>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
It is confusing to have __vma_start_exclude_readers() return 0, 1 or an
error (but only when waiting for readers in TASK_KILLABLE state), and
having the return value be stored in a stack variable called 'locked'
is further confusion.
More generally, we are doing a lot of rather finnicky things during the
acquisition of a state in which readers are excluded and moving out of
this state, including tracking whether we are detached or not or
whether an error occurred.
We are implementing logic in __vma_start_exclude_readers() that
effectively acts as if 'if one caller calls us do X, if another then do
Y', which is very confusing from a control flow perspective.
Introducing the shared helper object state helps us avoid this, as we
can now handle the 'an error arose but we're detached' condition
correctly in both callers - a warning if not detaching, and treating
the situation as if no error arose in the case of a VMA detaching.
This also acts to help document what's going on and allows us to add
some more logical debug asserts.
Also update vma_mark_detached() to add a guard clause for the likely
'already detached' state (given we hold the mmap write lock), and add a
comment about ephemeral VMA read lock reference count increments to
clarify why we are entering/exiting an exclusive locked state here.
Finally, separate vma_mark_detached() into its fast-path component and
make it inline, then place the slow path for excluding readers in
mmap_lock.c.
No functional change intended.
[akpm@linux-foundation.org: fix function naming in comments, add comment per Vlastimil per Lorenzo]
Link: https://lkml.kernel.org/r/7d3084d596c84da10dd374130a5055deba6439c0.1769198904.git.lorenzo.stoakes@oracle.com
Link: https://lkml.kernel.org/r/7d3084d596c84da10dd374130a5055deba6439c0.1769198904.git.lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Reviewed-by: Suren Baghdasaryan <surenb@google.com>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Waiman Long <longman@redhat.com>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
These functions are very confusing indeed. 'Entering' a lock could be
interpreted as acquiring it, but this is not what these functions are
interacting with.
Equally they don't indicate at all what kind of lock we are 'entering' or
'exiting'. Finally they are misleading as we invoke these functions when
we already hold a write lock to detach a VMA.
These functions are explicitly simply 'entering' and 'exiting' a state in
which we hold the EXCLUSIVE lock in order that we can either mark the VMA
as being write-locked, or mark the VMA detached.
Rename the functions accordingly, and also update
__vma_end_exclude_readers() to return detached state with a __must_check
directive, as it is simply clumsy to pass an output pointer here to
detached state and inconsistent vs. __vma_start_exclude_readers().
Finally, remove the unnecessary 'inline' directives.
No functional change intended.
Link: https://lkml.kernel.org/r/33273be9389712347d69987c408ca7436f0c1b22.1769198904.git.lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Suren Baghdasaryan <surenb@google.com>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Cc: Waiman Long <longman@redhat.com>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
We're doing precisely the same thing that __vma_exit_locked() does, so
de-duplicate this code and keep the refcount primitive in one place.
No functional change intended.
Link: https://lkml.kernel.org/r/c9759b593f6a158e984fa87abe2c3cbd368ef825.1769198904.git.lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Reviewed-by: Suren Baghdasaryan <surenb@google.com>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Cc: Waiman Long <longman@redhat.com>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
The code is littered with inscrutable and duplicative lockdep
incantations, replace these with defines which explain what is going on
and add commentary to explain what we're doing.
If lockdep is disabled these become no-ops. We must use defines so
_RET_IP_ remains meaningful.
These are self-documenting and aid readability of the code.
Additionally, instead of using the confusing rwsem_*() form for something
that is emphatically not an rwsem, we instead explicitly use
lock_[acquired, release]_shared/exclusive() lockdep invocations since we
are doing something rather custom here and these make more sense to use.
No functional change intended.
Link: https://lkml.kernel.org/r/fdae72441949ecf3b4a0ed3510da803e881bb153.1769198904.git.lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Reviewed-by: Suren Baghdasaryan <surenb@google.com>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Cc: Waiman Long <longman@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
The is_vma_writer_only() function is misnamed - this isn't determining if
there is only a write lock, as it checks for the presence of the
VM_REFCNT_EXCLUDE_READERS_FLAG.
Really, it is checking to see whether readers are excluded, with a
possibility of a false positive in the case of a detachment (there we
expect the vma->vm_refcnt to eventually be set to
VM_REFCNT_EXCLUDE_READERS_FLAG, whereas for an attached VMA we expect it
to eventually be set to VM_REFCNT_EXCLUDE_READERS_FLAG + 1).
Rename the function accordingly.
Relatedly, we use a __refcount_dec_and_test() primitive directly in
vma_refcount_put(), using the old value to determine what the reference
count ought to be after the operation is complete (ignoring racing
reference count adjustments).
Wrap this into a __vma_refcount_put_return() function, which we can then
utilise in vma_mark_detached() and thus keep the refcount primitive usage
abstracted.
This function, as the name implies, returns the value after the reference
count has been updated.
This reduces duplication in the two invocations of this function.
Also adjust comments, removing duplicative comments covered elsewhere and
adding more to aid understanding.
No functional change intended.
Link: https://lkml.kernel.org/r/32053580bff460eb1092ef780b526cefeb748bad.1769198904.git.lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Suren Baghdasaryan <surenb@google.com>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Cc: Waiman Long <longman@redhat.com>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
The possible vma->vm_refcnt values are confusing and vague, explain in
detail what these can be in a comment describing the vma->vm_refcnt field
and reference this comment in various places that read/write this field.
No functional change intended.
[akpm@linux-foundation.org: fix typo, per Suren]
Link: https://lkml.kernel.org/r/d462e7678c6cc7461f94e5b26c776547d80a67e8.1769198904.git.lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Suren Baghdasaryan <surenb@google.com>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Cc: Waiman Long <longman@redhat.com>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Patch series "mm: add and use vma_assert_stabilised() helper", v4.
This series first introduces a series of refactorings, intended to
significantly improve readability and abstraction of the code.
Sometimes we wish to assert that a VMA is stable, that is - the VMA cannot
be changed underneath us. This will be the case if EITHER the VMA lock or
the mmap lock is held.
We already open-code this in two places - anon_vma_name() in mm/madvise.c
and vma_flag_set_atomic() in include/linux/mm.h.
This series adds vma_assert_stablised() which abstract this can be used in
these callsites instead.
This implementation uses lockdep where possible - that is VMA read locks -
which correctly track read lock acquisition/release via:
vma_start_read() ->
rwsem_acquire_read()
vma_start_read_locked() ->
vma_start_read_locked_nested() ->
rwsem_acquire_read()
And:
vma_end_read() ->
vma_refcount_put() ->
rwsem_release()
We don't track the VMA locks using lockdep for VMA write locks, however
these are predicated upon mmap write locks whose lockdep state we do
track, and additionally vma_assert_stabillised() asserts this check if VMA
read lock is not held, so we get lockdep coverage in this case also.
We also add extensive comments to describe what we're doing.
There's some tricky stuff around mmap locking and stabilisation races that
we have to be careful of that I describe in the patch introducing
vma_assert_stabilised().
This change also lays the foundation for future series to add this assert
in further places where we wish to make it clear that we rely upon a
stabilised VMA.
The motivation for this change was precisely this.
This patch (of 10):
The VMA_LOCK_OFFSET value encodes a flag which vma->vm_refcnt is set to in
order to indicate that a VMA is in the process of having VMA read-locks
excluded in __vma_enter_locked() (that is, first checking if there are any
VMA read locks held, and if there are, waiting on them to be released).
This happens when a VMA write lock is being established, or a VMA is being
marked detached and discovers that the VMA reference count is elevated due
to read-locks temporarily elevating the reference count only to discover a
VMA write lock is in place.
The naming does not convey any of this, so rename VMA_LOCK_OFFSET to
VM_REFCNT_EXCLUDE_READERS_FLAG (with a sensible new prefix to
differentiate from the newly introduced VMA_*_BIT flags).
Also rename VMA_REF_LIMIT to VM_REFCNT_LIMIT to make this consistent also.
Update comments to reflect this.
No functional change intended.
Link: https://lkml.kernel.org/r/817bd763e5fe35f23e01347996f9007e6eb88460.1769198904.git.lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Reviewed-by: Suren Baghdasaryan <surenb@google.com>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Cc: Waiman Long <longman@redhat.com>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
If we get a signal, we need to restore the vm_refcnt. We don't think that
the refcount can actually be decremented to zero here as it requires the
VMA to be detached, and the vma_mark_detached() uses TASK_UNINTERRUPTIBLE.
However, that's a bit subtle, so handle it as if the refcount was zero at
the start of this function.
Link: https://lkml.kernel.org/r/20251128040100.3022561-1-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reported-by: syzbot+5b19bad23ac7f44bf8b8@syzkaller.appspotmail.com
Fixes: 2197bb60f8 ("mm: add vma_start_write_killable()")
Reviewed-by: Suren Baghdasaryan <surenb@google.com>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Reviewed-by: Liam R. Howlett <Liam.Howlett@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Move the commentary on how __vma_enter_locked() behaves from the body of
__vma_start_write() to the head of __vma_enter_locked() and merge it with
the existing documentation. Also add a call to
mmap_assert_write_locked().
Link: https://lkml.kernel.org/r/20251119042639.3937024-1-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
The retry in lock_vma_under_rcu() drops the rcu read lock before
reacquiring the lock and trying again. This may cause a use-after-free if
the maple node the maple state was using was freed.
The maple state is protected by the rcu read lock. When the lock is
dropped, the state cannot be reused as it tracks pointers to objects that
may be freed during the time where the lock was not held.
Any time the rcu read lock is dropped, the maple state must be
invalidated. Resetting the address and state to MA_START is the safest
course of action, which will result in the next operation starting from
the top of the tree.
Prior to commit 0b16f8bed1 ("mm: change vma_start_read() to drop RCU
lock on failure"), vma_start_read() would drop rcu read lock and return
NULL, so the retry would not have happened. However, now that
vma_start_read() drops rcu read lock on failure followed by a retry, we
may end up using a freed maple tree node cached in the maple state.
[surenb@google.com: changelog alteration]
Link: https://lkml.kernel.org/r/CAJuCfpEWMD-Z1j=nPYHcQW4F7E2Wka09KTXzGv7VE7oW1S8hcw@mail.gmail.com
Link: https://lkml.kernel.org/r/20251111215605.1721380-1-Liam.Howlett@oracle.com
Signed-off-by: Liam R. Howlett <Liam.Howlett@oracle.com>
Fixes: 0b16f8bed1 ("mm: change vma_start_read() to drop RCU lock on failure")
Reported-by: syzbot+131f9eb2b5807573275c@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=131f9eb2b5807573275c
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Suren Baghdasaryan <surenb@google.com>
Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Jann Horn <jannh@google.com>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Patch series "vma_start_write_killable"", v2.
When we added the VMA lock, we made a major oversight in not adding a
killable variant. That can run us into trouble where a thread takes the
VMA lock for read (eg handling a page fault) and then goes out to lunch
for an hour (eg doing reclaim). Another thread tries to modify the VMA,
taking the mmap_lock for write, then attempts to lock the VMA for write.
That blocks on the first thread, and ensures that every other page fault
now tries to take the mmap_lock for read. Because everything's in an
uninterruptible sleep, we can't kill the task, which makes me angry.
This patchset just adds vma_start_write_killable() and converts one caller
to use it. Most users are somewhat tricky to convert, so expect follow-up
individual patches per call-site which need careful analysis to make sure
we've done proper cleanup.
This patch (of 2):
The vma can be held read-locked for a substantial period of time, eg if
memory allocation needs to go into reclaim. It's useful to be able to
send fatal signals to threads which are waiting for the write lock.
Link: https://lkml.kernel.org/r/20251110203204.1454057-1-willy@infradead.org
Link: https://lkml.kernel.org/r/20251110203204.1454057-2-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: Suren Baghdasaryan <surenb@google.com>
Reviewed-by: Liam R. Howlett <Liam.Howlett@oracle.com>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Chris Li <chriscli@google.com>
Cc: Jann Horn <jannh@google.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
vma_start_read() can drop and reacquire RCU lock in certain failure cases.
It's not apparent that the RCU session started by the caller of this
function might be interrupted when vma_start_read() fails to lock the vma.
This might become a source of subtle bugs and to prevent that we change
the locking rules for vma_start_read() to drop RCU read lock upon failure.
This way it's more obvious that RCU-protected objects are unsafe after
vma locking fails.
Link: https://lkml.kernel.org/r/20250804233349.1278678-2-surenb@google.com
Suggested-by: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Suren Baghdasaryan <surenb@google.com>
Tested-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Jann Horn <jannh@google.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Limit the scope of vma_start_read() as it is used only as a helper for
higher-level locking functions implemented inside mmap_lock.c and we are
about to introduce more complex RCU rules for this function. The change
is pure code refactoring and has no functional changes.
Link: https://lkml.kernel.org/r/20250804233349.1278678-1-surenb@google.com
Suggested-by: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Suren Baghdasaryan <surenb@google.com>
Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Jann Horn <jannh@google.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
By inducing delays in the right places, Jann Horn created a reproducer for
a hard to hit UAF issue that became possible after VMAs were allowed to be
recycled by adding SLAB_TYPESAFE_BY_RCU to their cache.
Race description is borrowed from Jann's discovery report:
lock_vma_under_rcu() looks up a VMA locklessly with mas_walk() under
rcu_read_lock(). At that point, the VMA may be concurrently freed, and it
can be recycled by another process. vma_start_read() then increments the
vma->vm_refcnt (if it is in an acceptable range), and if this succeeds,
vma_start_read() can return a recycled VMA.
In this scenario where the VMA has been recycled, lock_vma_under_rcu()
will then detect the mismatching ->vm_mm pointer and drop the VMA through
vma_end_read(), which calls vma_refcount_put(). vma_refcount_put() drops
the refcount and then calls rcuwait_wake_up() using a copy of vma->vm_mm.
This is wrong: It implicitly assumes that the caller is keeping the VMA's
mm alive, but in this scenario the caller has no relation to the VMA's mm,
so the rcuwait_wake_up() can cause UAF.
The diagram depicting the race:
T1 T2 T3
== == ==
lock_vma_under_rcu
mas_walk
<VMA gets removed from mm>
mmap
<the same VMA is reallocated>
vma_start_read
__refcount_inc_not_zero_limited_acquire
munmap
__vma_enter_locked
refcount_add_not_zero
vma_end_read
vma_refcount_put
__refcount_dec_and_test
rcuwait_wait_event
<finish operation>
rcuwait_wake_up [UAF]
Note that rcuwait_wait_event() in T3 does not block because refcount was
already dropped by T1. At this point T3 can exit and free the mm causing
UAF in T1.
To avoid this we move vma->vm_mm verification into vma_start_read() and
grab vma->vm_mm to stabilize it before vma_refcount_put() operation.
[surenb@google.com: v3]
Link: https://lkml.kernel.org/r/20250729145709.2731370-1-surenb@google.com
Link: https://lkml.kernel.org/r/20250728175355.2282375-1-surenb@google.com
Fixes: 3104138517 ("mm: make vma cache SLAB_TYPESAFE_BY_RCU")
Signed-off-by: Suren Baghdasaryan <surenb@google.com>
Reported-by: Jann Horn <jannh@google.com>
Closes: https://lore.kernel.org/all/CAG48ez0-deFbVH=E3jbkWx=X3uVbd8nWeo6kbJPQ0KoUD+m2tA@mail.gmail.com/
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Jann Horn <jannh@google.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
With maple_tree supporting vma tree traversal under RCU and per-vma locks,
/proc/pid/maps can be read while holding individual vma locks instead of
locking the entire address space.
A completely lockless approach (walking vma tree under RCU) would be quite
complex with the main issue being get_vma_name() using callbacks which
might not work correctly with a stable vma copy, requiring original
(unstable) vma - see special_mapping_name() for example.
When per-vma lock acquisition fails, we take the mmap_lock for reading,
lock the vma, release the mmap_lock and continue. This fallback to mmap
read lock guarantees the reader to make forward progress even during lock
contention. This will interfere with the writer but for a very short time
while we are acquiring the per-vma lock and only when there was contention
on the vma reader is interested in.
We shouldn't see a repeated fallback to mmap read locks in practice, as
this require a very unlikely series of lock contentions (for instance due
to repeated vma split operations). However even if this did somehow
happen, we would still progress.
One case requiring special handling is when a vma changes between the time
it was found and the time it got locked. A problematic case would be if a
vma got shrunk so that its vm_start moved higher in the address space and
a new vma was installed at the beginning:
reader found: |--------VMA A--------|
VMA is modified: |-VMA B-|----VMA A----|
reader locks modified VMA A
reader reports VMA A: | gap |----VMA A----|
This would result in reporting a gap in the address space that does not
exist. To prevent this we retry the lookup after locking the vma, however
we do that only when we identify a gap and detect that the address space
was changed after we found the vma.
This change is designed to reduce mmap_lock contention and prevent a
process reading /proc/pid/maps files (often a low priority task, such as
monitoring/data collection services) from blocking address space updates.
Note that this change has a userspace visible disadvantage: it allows for
sub-page data tearing as opposed to the previous mechanism where data
tearing could happen only between pages of generated output data. Since
current userspace considers data tearing between pages to be acceptable,
we assume is will be able to handle sub-page data tearing as well.
Link: https://lkml.kernel.org/r/20250719182854.3166724-7-surenb@google.com
Signed-off-by: Suren Baghdasaryan <surenb@google.com>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Andrii Nakryiko <andrii@kernel.org>
Cc: Christian Brauner <brauner@kernel.org>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: David Hildenbrand <david@redhat.com>
Cc: Jann Horn <jannh@google.com>
Cc: Jeongjun Park <aha310510@gmail.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Josef Bacik <josef@toxicpanda.com>
Cc: Kalesh Singh <kaleshsingh@google.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: "Paul E . McKenney" <paulmck@kernel.org>
Cc: Peter Xu <peterx@redhat.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Shuah Khan <shuah@kernel.org>
Cc: Thomas Weißschuh <linux@weissschuh.net>
Cc: T.J. Mercier <tjmercier@google.com>
Cc: Ye Bin <yebin10@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Currently the VMA and mmap locking logic is entangled in two of the most
overwrought files in mm - include/linux/mm.h and mm/memory.c. Separate
this logic out so we can more easily make changes and create an
appropriate MAINTAINERS entry that spans only the logic relating to
locking.
This should have no functional change. Care is taken to avoid dependency
loops, we must regrettably keep release_fault_lock() and
assert_fault_locked() in mm.h as a result due to the dependence on the
vm_fault type.
Additionally we must declare rcuwait_wake_up() manually to avoid a
dependency cycle on linux/rcuwait.h.
Additionally move the nommu implementatino of lock_mm_and_find_vma() to
mmap_lock.c so everything lock-related is in one place.
Link: https://lkml.kernel.org/r/bec6c8e29fa8de9267a811a10b1bdae355d67ed4.1744799282.git.lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Reviewed-by: Suren Baghdasaryan <surenb@google.com>
Reviewed-by: Liam R. Howlett <Liam.Howlett@oracle.com>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Cc: David Hildenbrand <david@redhat.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: "Paul E . McKenney" <paulmck@kernel.org>
Cc: SeongJae Park <sj@kernel.org>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
We are starting to deploy mmap_lock tracepoint monitoring across our
fleet and the early results showed that these tracepoints are consuming
significant amount of CPUs in kernfs_path_from_node when enabled.
It seems like the kernel is trying to resolve the cgroup path in the
fast path of the locking code path when the tracepoints are enabled. In
addition for some application their metrics are regressing when
monitoring is enabled.
The cgroup path resolution can be slow and should not be done in the
fast path. Most userspace tools, like bpftrace, provides functionality
to get the cgroup path from cgroup id, so let's just trace the cgroup
id and the users can use better tools to get the path in the slow path.
Link: https://lkml.kernel.org/r/20241125171617.113892-1-shakeel.butt@linux.dev
Signed-off-by: Shakeel Butt <shakeel.butt@linux.dev>
Reviewed-by: Yosry Ahmed <yosryahmed@google.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Roman Gushchin <roman.gushchin@linux.dev>
Reviewed-by: Axel Rasmussen <axelrasmussen@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Suren Baghdasaryan <surenb@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Since 7d6be67cfd ("mm: mmap_lock: replace get_memcg_path_buf() with
on-stack buffer") we use trace_mmap_lock_reg()/unreg() only to maintain an
atomic reg_refcount which is checked to avoid performing
get_mm_memcg_path() in case none of the tracepoints using it is enabled.
This can be achieved directly by putting all the work needed for the
tracepoint behind the trace_mmap_lock_##type##_enabled(), as suggested by
Documentation/trace/tracepoints.rst and with the following advantages:
- uses the tracepoint's static key instead of evaluating a branch
- the check tracepoint specific, not shared by all of them
- we can get rid of trace_mmap_lock_reg()/unreg() completely
Thus use the trace_..._enabled() check and remove unnecessary code.
Link: https://lkml.kernel.org/r/20241105113456.95066-2-vbabka@suse.cz
Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Axel Rasmussen <axelrasmussen@google.com>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Masami Hiramatsu <mhiramat@kernel.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Commit 2b5067a814 ("mm: mmap_lock: add tracepoints around lock
acquisition") introduced TRACE_MMAP_LOCK_EVENT() macro using
preempt_disable() in order to let get_mm_memcg_path() return a percpu
buffer exclusively used by normal, softirq, irq and NMI contexts
respectively.
Commit 832b507253 ("mm: mmap_lock: use local locks instead of disabling
preemption") replaced preempt_disable() with local_lock(&memcg_paths.lock)
based on an argument that preempt_disable() has to be avoided because
get_mm_memcg_path() might sleep if PREEMPT_RT=y.
But syzbot started reporting
inconsistent {HARDIRQ-ON-W} -> {IN-HARDIRQ-W} usage.
and
inconsistent {SOFTIRQ-ON-W} -> {IN-SOFTIRQ-W} usage.
messages, for local_lock() does not disable IRQ.
We could replace local_lock() with local_lock_irqsave() in order to
suppress these messages. But this patch instead replaces percpu buffers
with on-stack buffer, for the size of each buffer returned by
get_memcg_path_buf() is only 256 bytes which is tolerable for allocating
from current thread's kernel stack memory.
Link: https://lkml.kernel.org/r/ef22d289-eadb-4ed9-863b-fbc922b33d8d@I-love.SAKURA.ne.jp
Reported-by: syzbot <syzbot+40905bca570ae6784745@syzkaller.appspotmail.com>
Closes: https://syzkaller.appspot.com/bug?extid=40905bca570ae6784745
Fixes: 832b507253 ("mm: mmap_lock: use local locks instead of disabling preemption")
Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Reviewed-by: Axel Rasmussen <axelrasmussen@google.com>
Cc: Nicolas Saenz Julienne <nsaenzju@redhat.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Commit 832b507253 ("mm: mmap_lock: use local locks instead of
disabling preemption") fixed a bug by using local locks.
But commit d01079f3d0 ("mm/mmap_lock: remove dead code for
!CONFIG_TRACING configurations") changed those lines back to the
original version.
I guess it was introduced by fixing conflicts.
Link: https://lkml.kernel.org/r/20210720074228.76342-1-songmuchun@bytedance.com
Fixes: d01079f3d0 ("mm/mmap_lock: remove dead code for !CONFIG_TRACING configurations")
Signed-off-by: Muchun Song <songmuchun@bytedance.com>
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Reviewed-by: Yang Shi <shy828301@gmail.com>
Reviewed-by: Pankaj Gupta <pankaj.gupta@ionos.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
make W=1 generates the following warning in mmap_lock.c for allnoconfig
mm/mmap_lock.c:213:6: warning: no previous prototype for `__mmap_lock_do_trace_start_locking' [-Wmissing-prototypes]
void __mmap_lock_do_trace_start_locking(struct mm_struct *mm, bool write)
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
mm/mmap_lock.c:219:6: warning: no previous prototype for `__mmap_lock_do_trace_acquire_returned' [-Wmissing-prototypes]
void __mmap_lock_do_trace_acquire_returned(struct mm_struct *mm, bool write,
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
mm/mmap_lock.c:226:6: warning: no previous prototype for `__mmap_lock_do_trace_released' [-Wmissing-prototypes]
void __mmap_lock_do_trace_released(struct mm_struct *mm, bool write)
On !CONFIG_TRACING configurations, the code is dead so put it behind an
#ifdef.
[cuibixuan@huawei.com: fix warning when CONFIG_TRACING is not defined]
Link: https://lkml.kernel.org/r/20210531033426.74031-1-cuibixuan@huawei.com
Link: https://lkml.kernel.org/r/20210520084809.8576-13-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Signed-off-by: Bixuan Cui <cuibixuan@huawei.com>
Reviewed-by: Yang Shi <shy828301@gmail.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Dan Streetman <ddstreet@ieee.org>
Cc: David Hildenbrand <david@redhat.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mmap_lock will explicitly disable/enable preemption upon manipulating its
local CPU variables. This is to be expected, but in this case, it doesn't
play well with PREEMPT_RT. The preemption disabled code section also
takes a spin-lock. Spin-locks in RT systems will try to schedule, which
is exactly what we're trying to avoid.
To mitigate this, convert the explicit preemption handling to local_locks.
Which are RT aware, and will disable migration instead of preemption when
PREEMPT_RT=y.
The faulty call trace looks like the following:
__mmap_lock_do_trace_*()
preempt_disable()
get_mm_memcg_path()
cgroup_path()
kernfs_path_from_node()
spin_lock_irqsave() /* Scheduling while atomic! */
Link: https://lkml.kernel.org/r/20210604163506.2103900-1-nsaenzju@redhat.com
Fixes: 2b5067a814 ("mm: mmap_lock: add tracepoints around lock acquisition ")
Signed-off-by: Nicolas Saenz Julienne <nsaenzju@redhat.com>
Tested-by: Axel Rasmussen <axelrasmussen@google.com>
Reviewed-by: Axel Rasmussen <axelrasmussen@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The goal of these tracepoints is to be able to debug lock contention
issues. This lock is acquired on most (all?) mmap / munmap / page fault
operations, so a multi-threaded process which does a lot of these can
experience significant contention.
We trace just before we start acquisition, when the acquisition returns
(whether it succeeded or not), and when the lock is released (or
downgraded). The events are broken out by lock type (read / write).
The events are also broken out by memcg path. For container-based
workloads, users often think of several processes in a memcg as a single
logical "task", so collecting statistics at this level is useful.
The end goal is to get latency information. This isn't directly included
in the trace events. Instead, users are expected to compute the time
between "start locking" and "acquire returned", using e.g. synthetic
events or BPF. The benefit we get from this is simpler code.
Because we use tracepoint_enabled() to decide whether or not to trace,
this patch has effectively no overhead unless tracepoints are enabled at
runtime. If tracepoints are enabled, there is a performance impact, but
how much depends on exactly what e.g. the BPF program does.
[axelrasmussen@google.com: fix use-after-free race and css ref leak in tracepoints]
Link: https://lkml.kernel.org/r/20201130233504.3725241-1-axelrasmussen@google.com
[axelrasmussen@google.com: v3]
Link: https://lkml.kernel.org/r/20201207213358.573750-1-axelrasmussen@google.com
[rostedt@goodmis.org: in-depth examples of tracepoint_enabled() usage, and per-cpu-per-context buffer design]
Link: https://lkml.kernel.org/r/20201105211739.568279-2-axelrasmussen@google.com
Signed-off-by: Axel Rasmussen <axelrasmussen@google.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Michel Lespinasse <walken@google.com>
Cc: Daniel Jordan <daniel.m.jordan@oracle.com>
Cc: Jann Horn <jannh@google.com>
Cc: Chinwen Chang <chinwen.chang@mediatek.com>
Cc: Davidlohr Bueso <dbueso@suse.de>
Cc: David Rientjes <rientjes@google.com>
Cc: Laurent Dufour <ldufour@linux.ibm.com>
Cc: Yafang Shao <laoar.shao@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
