]> git.ipfire.org Git - thirdparty/kernel/stable-queue.git/commitdiff
6.12-stable patches
authorGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Thu, 13 Mar 2025 16:09:19 +0000 (17:09 +0100)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Thu, 13 Mar 2025 16:09:19 +0000 (17:09 +0100)
added patches:
mm-fix-kernel-bug-when-userfaultfd_move-encounters-swapcache.patch
mm-slab-kvfree_rcu-switch-to-wq_mem_reclaim-wq.patch
series
userfaultfd-fix-pte-unmapping-stack-allocated-pte-copies.patch

queue-6.12/mm-fix-kernel-bug-when-userfaultfd_move-encounters-swapcache.patch [new file with mode: 0644]
queue-6.12/mm-slab-kvfree_rcu-switch-to-wq_mem_reclaim-wq.patch [new file with mode: 0644]
queue-6.12/series [new file with mode: 0644]
queue-6.12/userfaultfd-fix-pte-unmapping-stack-allocated-pte-copies.patch [new file with mode: 0644]

diff --git a/queue-6.12/mm-fix-kernel-bug-when-userfaultfd_move-encounters-swapcache.patch b/queue-6.12/mm-fix-kernel-bug-when-userfaultfd_move-encounters-swapcache.patch
new file mode 100644 (file)
index 0000000..0b0cbb1
--- /dev/null
@@ -0,0 +1,276 @@
+From c50f8e6053b0503375c2975bf47f182445aebb4c Mon Sep 17 00:00:00 2001
+From: Barry Song <v-songbaohua@oppo.com>
+Date: Wed, 26 Feb 2025 13:14:00 +1300
+Subject: mm: fix kernel BUG when userfaultfd_move encounters swapcache
+
+From: Barry Song <v-songbaohua@oppo.com>
+
+commit c50f8e6053b0503375c2975bf47f182445aebb4c upstream.
+
+userfaultfd_move() checks whether the PTE entry is present or a
+swap entry.
+
+- If the PTE entry is present, move_present_pte() handles folio
+  migration by setting:
+
+  src_folio->index = linear_page_index(dst_vma, dst_addr);
+
+- If the PTE entry is a swap entry, move_swap_pte() simply copies
+  the PTE to the new dst_addr.
+
+This approach is incorrect because, even if the PTE is a swap entry,
+it can still reference a folio that remains in the swap cache.
+
+This creates a race window between steps 2 and 4.
+ 1. add_to_swap: The folio is added to the swapcache.
+ 2. try_to_unmap: PTEs are converted to swap entries.
+ 3. pageout: The folio is written back.
+ 4. Swapcache is cleared.
+If userfaultfd_move() occurs in the window between steps 2 and 4,
+after the swap PTE has been moved to the destination, accessing the
+destination triggers do_swap_page(), which may locate the folio in
+the swapcache. However, since the folio's index has not been updated
+to match the destination VMA, do_swap_page() will detect a mismatch.
+
+This can result in two critical issues depending on the system
+configuration.
+
+If KSM is disabled, both small and large folios can trigger a BUG
+during the add_rmap operation due to:
+
+ page_pgoff(folio, page) != linear_page_index(vma, address)
+
+[   13.336953] page: refcount:6 mapcount:1 mapping:00000000f43db19c index:0xffffaf150 pfn:0x4667c
+[   13.337520] head: order:2 mapcount:1 entire_mapcount:0 nr_pages_mapped:1 pincount:0
+[   13.337716] memcg:ffff00000405f000
+[   13.337849] anon flags: 0x3fffc0000020459(locked|uptodate|dirty|owner_priv_1|head|swapbacked|node=0|zone=0|lastcpupid=0xffff)
+[   13.338630] raw: 03fffc0000020459 ffff80008507b538 ffff80008507b538 ffff000006260361
+[   13.338831] raw: 0000000ffffaf150 0000000000004000 0000000600000000 ffff00000405f000
+[   13.339031] head: 03fffc0000020459 ffff80008507b538 ffff80008507b538 ffff000006260361
+[   13.339204] head: 0000000ffffaf150 0000000000004000 0000000600000000 ffff00000405f000
+[   13.339375] head: 03fffc0000000202 fffffdffc0199f01 ffffffff00000000 0000000000000001
+[   13.339546] head: 0000000000000004 0000000000000000 00000000ffffffff 0000000000000000
+[   13.339736] page dumped because: VM_BUG_ON_PAGE(page_pgoff(folio, page) != linear_page_index(vma, address))
+[   13.340190] ------------[ cut here ]------------
+[   13.340316] kernel BUG at mm/rmap.c:1380!
+[   13.340683] Internal error: Oops - BUG: 00000000f2000800 [#1] PREEMPT SMP
+[   13.340969] Modules linked in:
+[   13.341257] CPU: 1 UID: 0 PID: 107 Comm: a.out Not tainted 6.14.0-rc3-gcf42737e247a-dirty #299
+[   13.341470] Hardware name: linux,dummy-virt (DT)
+[   13.341671] pstate: 60000005 (nZCv daif -PAN -UAO -TCO -DIT -SSBS BTYPE=--)
+[   13.341815] pc : __page_check_anon_rmap+0xa0/0xb0
+[   13.341920] lr : __page_check_anon_rmap+0xa0/0xb0
+[   13.342018] sp : ffff80008752bb20
+[   13.342093] x29: ffff80008752bb20 x28: fffffdffc0199f00 x27: 0000000000000001
+[   13.342404] x26: 0000000000000000 x25: 0000000000000001 x24: 0000000000000001
+[   13.342575] x23: 0000ffffaf0d0000 x22: 0000ffffaf0d0000 x21: fffffdffc0199f00
+[   13.342731] x20: fffffdffc0199f00 x19: ffff000006210700 x18: 00000000ffffffff
+[   13.342881] x17: 6c203d2120296567 x16: 6170202c6f696c6f x15: 662866666f67705f
+[   13.343033] x14: 6567617028454741 x13: 2929737365726464 x12: ffff800083728ab0
+[   13.343183] x11: ffff800082996bf8 x10: 0000000000000fd7 x9 : ffff80008011bc40
+[   13.343351] x8 : 0000000000017fe8 x7 : 00000000fffff000 x6 : ffff8000829eebf8
+[   13.343498] x5 : c0000000fffff000 x4 : 0000000000000000 x3 : 0000000000000000
+[   13.343645] x2 : 0000000000000000 x1 : ffff0000062db980 x0 : 000000000000005f
+[   13.343876] Call trace:
+[   13.344045]  __page_check_anon_rmap+0xa0/0xb0 (P)
+[   13.344234]  folio_add_anon_rmap_ptes+0x22c/0x320
+[   13.344333]  do_swap_page+0x1060/0x1400
+[   13.344417]  __handle_mm_fault+0x61c/0xbc8
+[   13.344504]  handle_mm_fault+0xd8/0x2e8
+[   13.344586]  do_page_fault+0x20c/0x770
+[   13.344673]  do_translation_fault+0xb4/0xf0
+[   13.344759]  do_mem_abort+0x48/0xa0
+[   13.344842]  el0_da+0x58/0x130
+[   13.344914]  el0t_64_sync_handler+0xc4/0x138
+[   13.345002]  el0t_64_sync+0x1ac/0x1b0
+[   13.345208] Code: aa1503e0 f000f801 910f6021 97ff5779 (d4210000)
+[   13.345504] ---[ end trace 0000000000000000 ]---
+[   13.345715] note: a.out[107] exited with irqs disabled
+[   13.345954] note: a.out[107] exited with preempt_count 2
+
+If KSM is enabled, Peter Xu also discovered that do_swap_page() may
+trigger an unexpected CoW operation for small folios because
+ksm_might_need_to_copy() allocates a new folio when the folio index
+does not match linear_page_index(vma, addr).
+
+This patch also checks the swapcache when handling swap entries. If a
+match is found in the swapcache, it processes it similarly to a present
+PTE.
+However, there are some differences. For example, the folio is no longer
+exclusive because folio_try_share_anon_rmap_pte() is performed during
+unmapping.
+Furthermore, in the case of swapcache, the folio has already been
+unmapped, eliminating the risk of concurrent rmap walks and removing the
+need to acquire src_folio's anon_vma or lock.
+
+Note that for large folios, in the swapcache handling path, we directly
+return -EBUSY since split_folio() will return -EBUSY regardless if
+the folio is under writeback or unmapped. This is not an urgent issue,
+so a follow-up patch may address it separately.
+
+[v-songbaohua@oppo.com: minor cleanup according to Peter Xu]
+  Link: https://lkml.kernel.org/r/20250226024411.47092-1-21cnbao@gmail.com
+Link: https://lkml.kernel.org/r/20250226001400.9129-1-21cnbao@gmail.com
+Fixes: adef440691ba ("userfaultfd: UFFDIO_MOVE uABI")
+Signed-off-by: Barry Song <v-songbaohua@oppo.com>
+Acked-by: Peter Xu <peterx@redhat.com>
+Reviewed-by: Suren Baghdasaryan <surenb@google.com>
+Cc: Andrea Arcangeli <aarcange@redhat.com>
+Cc: Al Viro <viro@zeniv.linux.org.uk>
+Cc: Axel Rasmussen <axelrasmussen@google.com>
+Cc: Brian Geffon <bgeffon@google.com>
+Cc: Christian Brauner <brauner@kernel.org>
+Cc: David Hildenbrand <david@redhat.com>
+Cc: Hugh Dickins <hughd@google.com>
+Cc: Jann Horn <jannh@google.com>
+Cc: Kalesh Singh <kaleshsingh@google.com>
+Cc: Liam R. Howlett <Liam.Howlett@oracle.com>
+Cc: Lokesh Gidra <lokeshgidra@google.com>
+Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
+Cc: Michal Hocko <mhocko@suse.com>
+Cc: Mike Rapoport (IBM) <rppt@kernel.org>
+Cc: Nicolas Geoffray <ngeoffray@google.com>
+Cc: Ryan Roberts <ryan.roberts@arm.com>
+Cc: Shuah Khan <shuah@kernel.org>
+Cc: ZhangPeng <zhangpeng362@huawei.com>
+Cc: Tangquan Zheng <zhengtangquan@oppo.com>
+Cc: <stable@vger.kernel.org>
+Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
+[ surenb: resolved merged conflict caused by the difference in
+  move_swap_pte() arguments ]
+Signed-off-by: Suren Baghdasaryan <surenb@google.com>
+Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
+---
+ mm/userfaultfd.c |   75 ++++++++++++++++++++++++++++++++++++++++++++++++-------
+ 1 file changed, 66 insertions(+), 9 deletions(-)
+
+--- a/mm/userfaultfd.c
++++ b/mm/userfaultfd.c
+@@ -18,6 +18,7 @@
+ #include <asm/tlbflush.h>
+ #include <asm/tlb.h>
+ #include "internal.h"
++#include "swap.h"
+ static __always_inline
+ bool validate_dst_vma(struct vm_area_struct *dst_vma, unsigned long dst_end)
+@@ -1067,15 +1068,13 @@ out:
+       return err;
+ }
+-static int move_swap_pte(struct mm_struct *mm,
++static int move_swap_pte(struct mm_struct *mm, struct vm_area_struct *dst_vma,
+                        unsigned long dst_addr, unsigned long src_addr,
+                        pte_t *dst_pte, pte_t *src_pte,
+                        pte_t orig_dst_pte, pte_t orig_src_pte,
+-                       spinlock_t *dst_ptl, spinlock_t *src_ptl)
++                       spinlock_t *dst_ptl, spinlock_t *src_ptl,
++                       struct folio *src_folio)
+ {
+-      if (!pte_swp_exclusive(orig_src_pte))
+-              return -EBUSY;
+-
+       double_pt_lock(dst_ptl, src_ptl);
+       if (!pte_same(ptep_get(src_pte), orig_src_pte) ||
+@@ -1084,6 +1083,16 @@ static int move_swap_pte(struct mm_struc
+               return -EAGAIN;
+       }
++      /*
++       * The src_folio resides in the swapcache, requiring an update to its
++       * index and mapping to align with the dst_vma, where a swap-in may
++       * occur and hit the swapcache after moving the PTE.
++       */
++      if (src_folio) {
++              folio_move_anon_rmap(src_folio, dst_vma);
++              src_folio->index = linear_page_index(dst_vma, dst_addr);
++      }
++
+       orig_src_pte = ptep_get_and_clear(mm, src_addr, src_pte);
+       set_pte_at(mm, dst_addr, dst_pte, orig_src_pte);
+       double_pt_unlock(dst_ptl, src_ptl);
+@@ -1130,6 +1139,7 @@ static int move_pages_pte(struct mm_stru
+                         __u64 mode)
+ {
+       swp_entry_t entry;
++      struct swap_info_struct *si = NULL;
+       pte_t orig_src_pte, orig_dst_pte;
+       pte_t src_folio_pte;
+       spinlock_t *src_ptl, *dst_ptl;
+@@ -1312,6 +1322,8 @@ retry:
+                                      orig_dst_pte, orig_src_pte,
+                                      dst_ptl, src_ptl, src_folio);
+       } else {
++              struct folio *folio = NULL;
++
+               entry = pte_to_swp_entry(orig_src_pte);
+               if (non_swap_entry(entry)) {
+                       if (is_migration_entry(entry)) {
+@@ -1325,10 +1337,53 @@ retry:
+                       goto out;
+               }
+-              err = move_swap_pte(mm, dst_addr, src_addr,
+-                                  dst_pte, src_pte,
+-                                  orig_dst_pte, orig_src_pte,
+-                                  dst_ptl, src_ptl);
++              if (!pte_swp_exclusive(orig_src_pte)) {
++                      err = -EBUSY;
++                      goto out;
++              }
++
++              si = get_swap_device(entry);
++              if (unlikely(!si)) {
++                      err = -EAGAIN;
++                      goto out;
++              }
++              /*
++               * Verify the existence of the swapcache. If present, the folio's
++               * index and mapping must be updated even when the PTE is a swap
++               * entry. The anon_vma lock is not taken during this process since
++               * the folio has already been unmapped, and the swap entry is
++               * exclusive, preventing rmap walks.
++               *
++               * For large folios, return -EBUSY immediately, as split_folio()
++               * also returns -EBUSY when attempting to split unmapped large
++               * folios in the swapcache. This issue needs to be resolved
++               * separately to allow proper handling.
++               */
++              if (!src_folio)
++                      folio = filemap_get_folio(swap_address_space(entry),
++                                      swap_cache_index(entry));
++              if (!IS_ERR_OR_NULL(folio)) {
++                      if (folio_test_large(folio)) {
++                              err = -EBUSY;
++                              folio_put(folio);
++                              goto out;
++                      }
++                      src_folio = folio;
++                      src_folio_pte = orig_src_pte;
++                      if (!folio_trylock(src_folio)) {
++                              pte_unmap(&orig_src_pte);
++                              pte_unmap(&orig_dst_pte);
++                              src_pte = dst_pte = NULL;
++                              put_swap_device(si);
++                              si = NULL;
++                              /* now we can block and wait */
++                              folio_lock(src_folio);
++                              goto retry;
++                      }
++              }
++              err = move_swap_pte(mm, dst_vma, dst_addr, src_addr, dst_pte, src_pte,
++                              orig_dst_pte, orig_src_pte,
++                              dst_ptl, src_ptl, src_folio);
+       }
+ out:
+@@ -1345,6 +1400,8 @@ out:
+       if (src_pte)
+               pte_unmap(src_pte);
+       mmu_notifier_invalidate_range_end(&range);
++      if (si)
++              put_swap_device(si);
+       return err;
+ }
diff --git a/queue-6.12/mm-slab-kvfree_rcu-switch-to-wq_mem_reclaim-wq.patch b/queue-6.12/mm-slab-kvfree_rcu-switch-to-wq_mem_reclaim-wq.patch
new file mode 100644 (file)
index 0000000..28fc48b
--- /dev/null
@@ -0,0 +1,142 @@
+From dfd3df31c9db752234d7d2e09bef2aeabb643ce4 Mon Sep 17 00:00:00 2001
+From: "Uladzislau Rezki (Sony)" <urezki@gmail.com>
+Date: Fri, 28 Feb 2025 13:13:56 +0100
+Subject: mm/slab/kvfree_rcu: Switch to WQ_MEM_RECLAIM wq
+
+From: Uladzislau Rezki (Sony) <urezki@gmail.com>
+
+commit dfd3df31c9db752234d7d2e09bef2aeabb643ce4 upstream.
+
+Currently kvfree_rcu() APIs use a system workqueue which is
+"system_unbound_wq" to driver RCU machinery to reclaim a memory.
+
+Recently, it has been noted that the following kernel warning can
+be observed:
+
+<snip>
+workqueue: WQ_MEM_RECLAIM nvme-wq:nvme_scan_work is flushing !WQ_MEM_RECLAIM events_unbound:kfree_rcu_work
+  WARNING: CPU: 21 PID: 330 at kernel/workqueue.c:3719 check_flush_dependency+0x112/0x120
+  Modules linked in: intel_uncore_frequency(E) intel_uncore_frequency_common(E) skx_edac(E) ...
+  CPU: 21 UID: 0 PID: 330 Comm: kworker/u144:6 Tainted: G            E      6.13.2-0_g925d379822da #1
+  Hardware name: Wiwynn Twin Lakes MP/Twin Lakes Passive MP, BIOS YMM20 02/01/2023
+  Workqueue: nvme-wq nvme_scan_work
+  RIP: 0010:check_flush_dependency+0x112/0x120
+  Code: 05 9a 40 14 02 01 48 81 c6 c0 00 00 00 48 8b 50 18 48 81 c7 c0 00 00 00 48 89 f9 48 ...
+  RSP: 0018:ffffc90000df7bd8 EFLAGS: 00010082
+  RAX: 000000000000006a RBX: ffffffff81622390 RCX: 0000000000000027
+  RDX: 00000000fffeffff RSI: 000000000057ffa8 RDI: ffff88907f960c88
+  RBP: 0000000000000000 R08: ffffffff83068e50 R09: 000000000002fffd
+  R10: 0000000000000004 R11: 0000000000000000 R12: ffff8881001a4400
+  R13: 0000000000000000 R14: ffff88907f420fb8 R15: 0000000000000000
+  FS:  0000000000000000(0000) GS:ffff88907f940000(0000) knlGS:0000000000000000
+  CR2: 00007f60c3001000 CR3: 000000107d010005 CR4: 00000000007726f0
+  PKRU: 55555554
+  Call Trace:
+   <TASK>
+   ? __warn+0xa4/0x140
+   ? check_flush_dependency+0x112/0x120
+   ? report_bug+0xe1/0x140
+   ? check_flush_dependency+0x112/0x120
+   ? handle_bug+0x5e/0x90
+   ? exc_invalid_op+0x16/0x40
+   ? asm_exc_invalid_op+0x16/0x20
+   ? timer_recalc_next_expiry+0x190/0x190
+   ? check_flush_dependency+0x112/0x120
+   ? check_flush_dependency+0x112/0x120
+   __flush_work.llvm.1643880146586177030+0x174/0x2c0
+   flush_rcu_work+0x28/0x30
+   kvfree_rcu_barrier+0x12f/0x160
+   kmem_cache_destroy+0x18/0x120
+   bioset_exit+0x10c/0x150
+   disk_release.llvm.6740012984264378178+0x61/0xd0
+   device_release+0x4f/0x90
+   kobject_put+0x95/0x180
+   nvme_put_ns+0x23/0xc0
+   nvme_remove_invalid_namespaces+0xb3/0xd0
+   nvme_scan_work+0x342/0x490
+   process_scheduled_works+0x1a2/0x370
+   worker_thread+0x2ff/0x390
+   ? pwq_release_workfn+0x1e0/0x1e0
+   kthread+0xb1/0xe0
+   ? __kthread_parkme+0x70/0x70
+   ret_from_fork+0x30/0x40
+   ? __kthread_parkme+0x70/0x70
+   ret_from_fork_asm+0x11/0x20
+   </TASK>
+  ---[ end trace 0000000000000000 ]---
+<snip>
+
+To address this switch to use of independent WQ_MEM_RECLAIM
+workqueue, so the rules are not violated from workqueue framework
+point of view.
+
+Apart of that, since kvfree_rcu() does reclaim memory it is worth
+to go with WQ_MEM_RECLAIM type of wq because it is designed for
+this purpose.
+
+Fixes: 6c6c47b063b5 ("mm, slab: call kvfree_rcu_barrier() from kmem_cache_destroy()"),
+Reported-by: Keith Busch <kbusch@kernel.org>
+Closes: https://lore.kernel.org/all/Z7iqJtCjHKfo8Kho@kbusch-mbp/
+Cc: stable@vger.kernel.org
+Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
+Reviewed-by: Joel Fernandes <joelagnelf@nvidia.com>
+Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
+Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
+Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
+---
+ kernel/rcu/tree.c |   14 ++++++++++----
+ 1 file changed, 10 insertions(+), 4 deletions(-)
+
+--- a/kernel/rcu/tree.c
++++ b/kernel/rcu/tree.c
+@@ -3191,6 +3191,8 @@ void call_rcu(struct rcu_head *head, rcu
+ }
+ EXPORT_SYMBOL_GPL(call_rcu);
++static struct workqueue_struct *rcu_reclaim_wq;
++
+ /* Maximum number of jiffies to wait before draining a batch. */
+ #define KFREE_DRAIN_JIFFIES (5 * HZ)
+ #define KFREE_N_BATCHES 2
+@@ -3519,10 +3521,10 @@ __schedule_delayed_monitor_work(struct k
+       if (delayed_work_pending(&krcp->monitor_work)) {
+               delay_left = krcp->monitor_work.timer.expires - jiffies;
+               if (delay < delay_left)
+-                      mod_delayed_work(system_unbound_wq, &krcp->monitor_work, delay);
++                      mod_delayed_work(rcu_reclaim_wq, &krcp->monitor_work, delay);
+               return;
+       }
+-      queue_delayed_work(system_unbound_wq, &krcp->monitor_work, delay);
++      queue_delayed_work(rcu_reclaim_wq, &krcp->monitor_work, delay);
+ }
+ static void
+@@ -3620,7 +3622,7 @@ kvfree_rcu_queue_batch(struct kfree_rcu_
+                       // "free channels", the batch can handle. Break
+                       // the loop since it is done with this CPU thus
+                       // queuing an RCU work is _always_ success here.
+-                      queued = queue_rcu_work(system_unbound_wq, &krwp->rcu_work);
++                      queued = queue_rcu_work(rcu_reclaim_wq, &krwp->rcu_work);
+                       WARN_ON_ONCE(!queued);
+                       break;
+               }
+@@ -3708,7 +3710,7 @@ run_page_cache_worker(struct kfree_rcu_c
+       if (rcu_scheduler_active == RCU_SCHEDULER_RUNNING &&
+                       !atomic_xchg(&krcp->work_in_progress, 1)) {
+               if (atomic_read(&krcp->backoff_page_cache_fill)) {
+-                      queue_delayed_work(system_unbound_wq,
++                      queue_delayed_work(rcu_reclaim_wq,
+                               &krcp->page_cache_work,
+                                       msecs_to_jiffies(rcu_delay_page_cache_fill_msec));
+               } else {
+@@ -5662,6 +5664,10 @@ static void __init kfree_rcu_batch_init(
+       int i, j;
+       struct shrinker *kfree_rcu_shrinker;
++      rcu_reclaim_wq = alloc_workqueue("kvfree_rcu_reclaim",
++                      WQ_UNBOUND | WQ_MEM_RECLAIM, 0);
++      WARN_ON(!rcu_reclaim_wq);
++
+       /* Clamp it to [0:100] seconds interval. */
+       if (rcu_delay_page_cache_fill_msec < 0 ||
+               rcu_delay_page_cache_fill_msec > 100 * MSEC_PER_SEC) {
diff --git a/queue-6.12/series b/queue-6.12/series
new file mode 100644 (file)
index 0000000..8ceb56d
--- /dev/null
@@ -0,0 +1,3 @@
+mm-slab-kvfree_rcu-switch-to-wq_mem_reclaim-wq.patch
+mm-fix-kernel-bug-when-userfaultfd_move-encounters-swapcache.patch
+userfaultfd-fix-pte-unmapping-stack-allocated-pte-copies.patch
diff --git a/queue-6.12/userfaultfd-fix-pte-unmapping-stack-allocated-pte-copies.patch b/queue-6.12/userfaultfd-fix-pte-unmapping-stack-allocated-pte-copies.patch
new file mode 100644 (file)
index 0000000..eabea4b
--- /dev/null
@@ -0,0 +1,95 @@
+From 927e926d72d9155fde3264459fe9bfd7b5e40d28 Mon Sep 17 00:00:00 2001
+From: Suren Baghdasaryan <surenb@google.com>
+Date: Wed, 26 Feb 2025 10:55:09 -0800
+Subject: userfaultfd: fix PTE unmapping stack-allocated PTE copies
+
+From: Suren Baghdasaryan <surenb@google.com>
+
+commit 927e926d72d9155fde3264459fe9bfd7b5e40d28 upstream.
+
+Current implementation of move_pages_pte() copies source and destination
+PTEs in order to detect concurrent changes to PTEs involved in the move.
+However these copies are also used to unmap the PTEs, which will fail if
+CONFIG_HIGHPTE is enabled because the copies are allocated on the stack.
+Fix this by using the actual PTEs which were kmap()ed.
+
+Link: https://lkml.kernel.org/r/20250226185510.2732648-3-surenb@google.com
+Fixes: adef440691ba ("userfaultfd: UFFDIO_MOVE uABI")
+Signed-off-by: Suren Baghdasaryan <surenb@google.com>
+Reported-by: Peter Xu <peterx@redhat.com>
+Reviewed-by: Peter Xu <peterx@redhat.com>
+Cc: Andrea Arcangeli <aarcange@redhat.com>
+Cc: Barry Song <21cnbao@gmail.com>
+Cc: Barry Song <v-songbaohua@oppo.com>
+Cc: David Hildenbrand <david@redhat.com>
+Cc: Hugh Dickins <hughd@google.com>
+Cc: Jann Horn <jannh@google.com>
+Cc: Kalesh Singh <kaleshsingh@google.com>
+Cc: Liam R. Howlett <Liam.Howlett@Oracle.com>
+Cc: Lokesh Gidra <lokeshgidra@google.com>
+Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
+Cc: Matthew Wilcow (Oracle) <willy@infradead.org>
+Cc: <stable@vger.kernel.org>
+Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
+Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
+---
+ mm/userfaultfd.c |   20 ++++++++++----------
+ 1 file changed, 10 insertions(+), 10 deletions(-)
+
+--- a/mm/userfaultfd.c
++++ b/mm/userfaultfd.c
+@@ -1265,8 +1265,8 @@ retry:
+                       spin_unlock(src_ptl);
+                       if (!locked) {
+-                              pte_unmap(&orig_src_pte);
+-                              pte_unmap(&orig_dst_pte);
++                              pte_unmap(src_pte);
++                              pte_unmap(dst_pte);
+                               src_pte = dst_pte = NULL;
+                               /* now we can block and wait */
+                               folio_lock(src_folio);
+@@ -1282,8 +1282,8 @@ retry:
+               /* at this point we have src_folio locked */
+               if (folio_test_large(src_folio)) {
+                       /* split_folio() can block */
+-                      pte_unmap(&orig_src_pte);
+-                      pte_unmap(&orig_dst_pte);
++                      pte_unmap(src_pte);
++                      pte_unmap(dst_pte);
+                       src_pte = dst_pte = NULL;
+                       err = split_folio(src_folio);
+                       if (err)
+@@ -1308,8 +1308,8 @@ retry:
+                               goto out;
+                       }
+                       if (!anon_vma_trylock_write(src_anon_vma)) {
+-                              pte_unmap(&orig_src_pte);
+-                              pte_unmap(&orig_dst_pte);
++                              pte_unmap(src_pte);
++                              pte_unmap(dst_pte);
+                               src_pte = dst_pte = NULL;
+                               /* now we can block and wait */
+                               anon_vma_lock_write(src_anon_vma);
+@@ -1327,8 +1327,8 @@ retry:
+               entry = pte_to_swp_entry(orig_src_pte);
+               if (non_swap_entry(entry)) {
+                       if (is_migration_entry(entry)) {
+-                              pte_unmap(&orig_src_pte);
+-                              pte_unmap(&orig_dst_pte);
++                              pte_unmap(src_pte);
++                              pte_unmap(dst_pte);
+                               src_pte = dst_pte = NULL;
+                               migration_entry_wait(mm, src_pmd, src_addr);
+                               err = -EAGAIN;
+@@ -1371,8 +1371,8 @@ retry:
+                       src_folio = folio;
+                       src_folio_pte = orig_src_pte;
+                       if (!folio_trylock(src_folio)) {
+-                              pte_unmap(&orig_src_pte);
+-                              pte_unmap(&orig_dst_pte);
++                              pte_unmap(src_pte);
++                              pte_unmap(dst_pte);
+                               src_pte = dst_pte = NULL;
+                               put_swap_device(si);
+                               si = NULL;