On Mon, Mar 16, 2020 at 10:15:39PM -0700, Sean Christopherson wrote:
> Drop the synchronize_srcu() from sgx_encl_mm_add() and replace it with a
> mm_list versioning concept to avoid deadlock when adding a mm during
> dup_mmap()/fork(), and to ensure copied PTEs are zapped.
>
> When dup_mmap() runs, it holds mmap_sem for write in both the old mm and
> new mm. Invoking synchronize_srcu() while holding mmap_sem of a mm that
> is already attached to the enclave will deadlock if the reclaimer is in
> the process of walking mm_list, as the reclaimer will try to acquire
> mmap_sem (of the old mm) while holding encl->srcu for read.
>
> INFO: task ksgxswapd:181 blocked for more than 120 seconds.
> ksgxswapd D 0 181 2 0x80004000
> Call Trace:
> __schedule+0x2db/0x700
> schedule+0x44/0xb0
> rwsem_down_read_slowpath+0x370/0x470
> down_read+0x95/0xa0
> sgx_reclaim_pages+0x1d2/0x7d0
> ksgxswapd+0x151/0x2e0
> kthread+0x120/0x140
> ret_from_fork+0x35/0x40
>
> INFO: task fork_consistenc:18824 blocked for more than 120 seconds.
> fork_consistenc D 0 18824 18786 0x00004320
> Call Trace:
> __schedule+0x2db/0x700
> schedule+0x44/0xb0
> schedule_timeout+0x205/0x300
> wait_for_completion+0xb7/0x140
> __synchronize_srcu.part.22+0x81/0xb0
> synchronize_srcu_expedited+0x27/0x30
> synchronize_srcu+0x57/0xe0
> sgx_encl_mm_add+0x12b/0x160
> sgx_vma_open+0x22/0x40
> dup_mm+0x521/0x580
> copy_process+0x1a56/0x1b50
> _do_fork+0x85/0x3a0
> __x64_sys_clone+0x8e/0xb0
> do_syscall_64+0x57/0x1b0
> entry_SYSCALL_64_after_hwframe+0x44/0xa9
>
> Furthermore, doing synchronize_srcu() in sgx_encl_mm_add() does not
> prevent the new mm from having stale PTEs pointing at the EPC page to be
> reclaimed. dup_mmap() calls vm_ops->open()/sgx_encl_mm_add() _after_
> PTEs are copied to the new mm, i.e. blocking fork() until reclaim zaps
> the old mm is pointless as the stale PTEs have already been created in
> the new mm.
>
> All other flows that walk mm_list can safely race with dup_mmap() or are
> protected by a different mechanism. Add comments to all srcu readers
> that don't check the list version to document why its ok for the flow to
> ignore the version.
>
> Note, synchronize_srcu() is still needed when removing a mm from an
> enclave, as the srcu readers must complete their walk before the mm can
> be freed. Removing a mm is never done while holding mmap_sem.
>
> Signed-off-by: Sean Christopherson <sean.j.christopherson@xxxxxxxxx>
> ---
> arch/x86/kernel/cpu/sgx/encl.c | 22 +++++++++++++++++++--
> arch/x86/kernel/cpu/sgx/encl.h | 1 +
> arch/x86/kernel/cpu/sgx/reclaim.c | 33 +++++++++++++++++++++++++++++++
> 3 files changed, 54 insertions(+), 2 deletions(-)
>
> diff --git a/arch/x86/kernel/cpu/sgx/encl.c b/arch/x86/kernel/cpu/sgx/encl.c
> index d6a19bdd1921..b9a7c56f7c25 100644
> --- a/arch/x86/kernel/cpu/sgx/encl.c
> +++ b/arch/x86/kernel/cpu/sgx/encl.c
> @@ -196,6 +196,12 @@ int sgx_encl_mm_add(struct sgx_encl *encl, struct mm_struct *mm)
> struct sgx_encl_mm *encl_mm;
> int ret;
>
> + /*
> + * This flow relies on mmap_sem to provide mutual exclusivity (for a
> + * given mm) to prevent duplicate instances of an encl_mm on the list.
> + */
> + lockdep_assert_held_write(&mm->mmap_sem);
> +
> if (atomic_read(&encl->flags) & SGX_ENCL_DEAD)
> return -EINVAL;
>
> @@ -223,10 +229,22 @@ int sgx_encl_mm_add(struct sgx_encl *encl, struct mm_struct *mm)
>
> spin_lock(&encl->mm_lock);
> list_add_rcu(&encl_mm->list, &encl->mm_list);
> + /*
> + * Ensure the mm is added to the list before updating the version.
> + * Pairs with the smp_rmb() in sgx_reclaimer_block().
> + */
> + smp_wmb();
> + encl->mm_list_version++;
> spin_unlock(&encl->mm_lock);
>
> - synchronize_srcu(&encl->srcu);
> -
> + /*
> + * DO NOT call synchronize_srcu()! When this is called via dup_mmap(),
> + * mmap_sem is held for write in both the old mm and new mm, and the
> + * reclaimer may be holding srcu for read while waiting on down_read()
> + * for the old mm's mmap_sem, i.e. synchronizing will deadlock.
> + * Incrementing the list version ensures readers that must not race
> + * with a mm being added will see the updated list.
> + */
> return 0;
> }
>
> diff --git a/arch/x86/kernel/cpu/sgx/encl.h b/arch/x86/kernel/cpu/sgx/encl.h
> index 44b353aa8866..f0f72e591244 100644
> --- a/arch/x86/kernel/cpu/sgx/encl.h
> +++ b/arch/x86/kernel/cpu/sgx/encl.h
> @@ -74,6 +74,7 @@ struct sgx_encl {
> struct mutex lock;
> struct list_head mm_list;
> spinlock_t mm_lock;
> + unsigned long mm_list_version;
> struct file *backing;
> struct kref refcount;
> struct srcu_struct srcu;
> diff --git a/arch/x86/kernel/cpu/sgx/reclaim.c b/arch/x86/kernel/cpu/sgx/reclaim.c
> index 39f0ddefbb79..3b4b849c5b2e 100644
> --- a/arch/x86/kernel/cpu/sgx/reclaim.c
> +++ b/arch/x86/kernel/cpu/sgx/reclaim.c
> @@ -155,6 +155,11 @@ static bool sgx_reclaimer_age(struct sgx_epc_page *epc_page)
> bool ret = true;
> int idx;
>
> + /*
> + * Note, this can race with sgx_encl_mm_add(), but worst case scenario
> + * a page will be reclaimed immediately after it's accessed in the new
> + * process/mm.
> + */
> idx = srcu_read_lock(&encl->srcu);
>
> list_for_each_entry_rcu(encl_mm, &encl->mm_list, list) {
> @@ -184,10 +189,20 @@ static void sgx_reclaimer_block(struct sgx_epc_page *epc_page)
> struct sgx_encl_page *page = epc_page->owner;
> unsigned long addr = SGX_ENCL_PAGE_ADDR(page);
> struct sgx_encl *encl = page->encl;
> + unsigned long mm_list_version;
> struct sgx_encl_mm *encl_mm;
> struct vm_area_struct *vma;
> int idx, ret;
>
> +retry:
> + mm_list_version = encl->mm_list_version;
> + /*
> + * Ensure the list version is read before walking the list to prevent
> + * beginning the walk with the old list using the new version. Pairs
> + * with the smp_wmb() in sgx_encl_mm_add().
> + */
> + smp_rmb();
> +
> idx = srcu_read_lock(&encl->srcu);
>
> list_for_each_entry_rcu(encl_mm, &encl->mm_list, list) {
> @@ -207,6 +222,19 @@ static void sgx_reclaimer_block(struct sgx_epc_page *epc_page)
>
> srcu_read_unlock(&encl->srcu, idx);
>
> + /*
> + * Redo the zapping if a mm was added to the list while zapping was in
> + * progress. dup_mmap() copies the PTEs for VM_PFNMAP VMAs, i.e. the
> + * new mm won't take a page fault and so won't see that the page is
> + * tagged RECLAIMED. Note, vm_ops->open()/sgx_encl_mm_add() is called
> + * _after_ PTEs are copied, and dup_mmap() holds the old mm's mmap_sem
> + * for write, so the version check is only needed to protect against
> + * dup_mmap() running after the list walk started but before the old
> + * mm's PTEs were zapped.
> + */
> + if (unlikely(encl->mm_list_version != mm_list_version))
> + goto retry;
> +
> mutex_lock(&encl->lock);
>
> if (!(atomic_read(&encl->flags) & SGX_ENCL_DEAD)) {
> @@ -250,6 +278,11 @@ static const cpumask_t *sgx_encl_ewb_cpumask(struct sgx_encl *encl)
> struct sgx_encl_mm *encl_mm;
> int idx;
>
> + /*
> + * Note, this can race with sgx_encl_mm_add(), but ETRACK has already
> + * been executed, so CPUs running in the new mm will enter the enclave
> + * in a different epoch.
> + */
> cpumask_clear(cpumask);
>
> idx = srcu_read_lock(&encl->srcu);
> --
> 2.24.1
>
Please recheck the remarks that I made about inline comments in the
source code.
/Jarkko.
