fs/proc/page.c: don't access uninitialized memmaps in fs/proc/page.c
authorDavid Hildenbrand <david@redhat.com>
Sat, 19 Oct 2019 03:19:20 +0000 (20:19 -0700)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Tue, 29 Oct 2019 08:17:37 +0000 (09:17 +0100)
commit aad5f69bc161af489dbb5934868bd347282f0764 upstream.

There are three places where we access uninitialized memmaps, namely:
- /proc/kpagecount
- /proc/kpageflags
- /proc/kpagecgroup

We have initialized memmaps either when the section is online or when the
page was initialized to the ZONE_DEVICE.  Uninitialized memmaps contain
garbage and in the worst case trigger kernel BUGs, especially with
CONFIG_PAGE_POISONING.

For example, not onlining a DIMM during boot and calling /proc/kpagecount
with CONFIG_PAGE_POISONING:

  :/# cat /proc/kpagecount > tmp.test
  BUG: unable to handle page fault for address: fffffffffffffffe
  #PF: supervisor read access in kernel mode
  #PF: error_code(0x0000) - not-present page
  PGD 114616067 P4D 114616067 PUD 114618067 PMD 0
  Oops: 0000 [#1] SMP NOPTI
  CPU: 0 PID: 469 Comm: cat Not tainted 5.4.0-rc1-next-20191004+ #11
  Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.12.1-0-ga5cab58e9a3f-prebuilt.qemu.4
  RIP: 0010:kpagecount_read+0xce/0x1e0
  Code: e8 09 83 e0 3f 48 0f a3 02 73 2d 4c 89 e7 48 c1 e7 06 48 03 3d ab 51 01 01 74 1d 48 8b 57 08 480
  RSP: 0018:ffffa14e409b7e78 EFLAGS: 00010202
  RAX: fffffffffffffffe RBX: 0000000000020000 RCX: 0000000000000000
  RDX: 0000000000000001 RSI: 00007f76b5595000 RDI: fffff35645000000
  RBP: 00007f76b5595000 R08: 0000000000000001 R09: 0000000000000000
  R10: 0000000000000000 R11: 0000000000000000 R12: 0000000000140000
  R13: 0000000000020000 R14: 00007f76b5595000 R15: ffffa14e409b7f08
  FS:  00007f76b577d580(0000) GS:ffff8f41bd400000(0000) knlGS:0000000000000000
  CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
  CR2: fffffffffffffffe CR3: 0000000078960000 CR4: 00000000000006f0
  Call Trace:
   proc_reg_read+0x3c/0x60
   vfs_read+0xc5/0x180
   ksys_read+0x68/0xe0
   do_syscall_64+0x5c/0xa0
   entry_SYSCALL_64_after_hwframe+0x49/0xbe

For now, let's drop support for ZONE_DEVICE from the three pseudo files
in order to fix this.  To distinguish offline memory (with garbage
memmap) from ZONE_DEVICE memory with properly initialized memmaps, we
would have to check get_dev_pagemap() and pfn_zone_device_reserved()
right now.  The usage of both (especially, special casing devmem) is
frowned upon and needs to be reworked.

The fundamental issue we have is:

if (pfn_to_online_page(pfn)) {
/* memmap initialized */
} else if (pfn_valid(pfn)) {
/*
 * ???
 * a) offline memory. memmap garbage.
 * b) devmem: memmap initialized to ZONE_DEVICE.
 * c) devmem: reserved for driver. memmap garbage.
 * (d) devmem: memmap currently initializing - garbage)
 */
}

We'll leave the pfn_zone_device_reserved() check in stable_page_flags()
in place as that function is also used from memory failure.  We now no
longer dump information about pages that are not in use anymore -
offline.

Link: http://lkml.kernel.org/r/20191009142435.3975-2-david@redhat.com
Fixes: f1dd2cd13c4b ("mm, memory_hotplug: do not associate hotadded memory to zones until online") [visible after d0dc12e86b319]
Signed-off-by: David Hildenbrand <david@redhat.com>
Reported-by: Qian Cai <cai@lca.pw>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Toshiki Fukasawa <t-fukasawa@vx.jp.nec.com>
Cc: Pankaj gupta <pagupta@redhat.com>
Cc: Mike Rapoport <rppt@linux.vnet.ibm.com>
Cc: Anthony Yznaga <anthony.yznaga@oracle.com>
Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.ibm.com>
Cc: <stable@vger.kernel.org> [4.13+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
fs/proc/page.c

index 1491918a33c385394fcaf06ef07dcd7751832c82..0c952c2171185b2565e43b246612a512930e2d0f 100644 (file)
@@ -42,10 +42,12 @@ static ssize_t kpagecount_read(struct file *file, char __user *buf,
                return -EINVAL;
 
        while (count > 0) {
-               if (pfn_valid(pfn))
-                       ppage = pfn_to_page(pfn);
-               else
-                       ppage = NULL;
+               /*
+                * TODO: ZONE_DEVICE support requires to identify
+                * memmaps that were actually initialized.
+                */
+               ppage = pfn_to_online_page(pfn);
+
                if (!ppage || PageSlab(ppage))
                        pcount = 0;
                else
@@ -214,10 +216,11 @@ static ssize_t kpageflags_read(struct file *file, char __user *buf,
                return -EINVAL;
 
        while (count > 0) {
-               if (pfn_valid(pfn))
-                       ppage = pfn_to_page(pfn);
-               else
-                       ppage = NULL;
+               /*
+                * TODO: ZONE_DEVICE support requires to identify
+                * memmaps that were actually initialized.
+                */
+               ppage = pfn_to_online_page(pfn);
 
                if (put_user(stable_page_flags(ppage), out)) {
                        ret = -EFAULT;
@@ -259,10 +262,11 @@ static ssize_t kpagecgroup_read(struct file *file, char __user *buf,
                return -EINVAL;
 
        while (count > 0) {
-               if (pfn_valid(pfn))
-                       ppage = pfn_to_page(pfn);
-               else
-                       ppage = NULL;
+               /*
+                * TODO: ZONE_DEVICE support requires to identify
+                * memmaps that were actually initialized.
+                */
+               ppage = pfn_to_online_page(pfn);
 
                if (ppage)
                        ino = page_cgroup_ino(ppage);