Re: [PATCH v1 7/7] fs/proc/kcore: use page_offline_(freeze|unfreeze)

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On 02.05.21 08:34, Mike Rapoport wrote:
On Thu, Apr 29, 2021 at 02:25:19PM +0200, David Hildenbrand wrote:
Let's properly synchronize with drivers that set PageOffline(). Unfreeze
every now and then, so drivers that want to set PageOffline() can make
progress.

Signed-off-by: David Hildenbrand <david@xxxxxxxxxx>
---
  fs/proc/kcore.c | 15 +++++++++++++++
  1 file changed, 15 insertions(+)

diff --git a/fs/proc/kcore.c b/fs/proc/kcore.c
index 92ff1e4436cb..3d7531f47389 100644
--- a/fs/proc/kcore.c
+++ b/fs/proc/kcore.c
@@ -311,6 +311,7 @@ static void append_kcore_note(char *notes, size_t *i, const char *name,
  static ssize_t
  read_kcore(struct file *file, char __user *buffer, size_t buflen, loff_t *fpos)
  {
+	size_t page_offline_frozen = 0;
  	char *buf = file->private_data;
  	size_t phdrs_offset, notes_offset, data_offset;
  	size_t phdrs_len, notes_len;
@@ -509,6 +510,18 @@ read_kcore(struct file *file, char __user *buffer, size_t buflen, loff_t *fpos)
  			pfn = __pa(start) >> PAGE_SHIFT;
  			page = pfn_to_online_page(pfn);

Can't this race with page offlining for the first time we get here?


To clarify, we have three types of offline pages in the kernel ...

a) Pages part of an offline memory section; the memap is stale and not trustworthy. pfn_to_online_page() checks that. We *can* protect against memory offlining using get_online_mems()/put_online_mems(), but usually avoid doing so as the race window is very small (and a problem all over the kernel we basically never hit) and locking is rather expensive. In the future, we might switch to rcu to handle that more efficiently and avoiding these possible races.

b) PageOffline(): logically offline pages contained in an online memory section with a sane memmap. virtio-mem calls these pages "fake offline"; something like a "temporary" memory hole. The new mechanism I propose will be used to handle synchronization as races can be more severe, e.g., when reading actual page content here.

c) Soft offline pages: hwpoisoned pages that are not actually harmful yet, but could become harmful in the future. So we better try to remove the page from the page allcoator and try to migrate away existing users.


So page_offline_* handle "b) PageOffline()" only. There is a tiny race between pfn_to_online_page(pfn) and looking at the memmap as we have in many cases already throughout the kernel, to be tackled in the future.


(A better name for PageOffline() might make sense; PageSoftOffline() would be catchy but interferes with c). PageLogicallyOffline() is ugly; PageFakeOffline() might do)

+			/*
+			 * Don't race against drivers that set PageOffline()
+			 * and expect no further page access.
+			 */
+			if (page_offline_frozen == MAX_ORDER_NR_PAGES) {
+				page_offline_unfreeze();
+				page_offline_frozen = 0;
+				cond_resched();
+			}
+			if (!page_offline_frozen++)
+				page_offline_freeze();
+

Don't we need to freeze before doing pfn_to_online_page()?

See my explanation above. Thanks!

--
Thanks,

David / dhildenb




[Index of Archives]     [Linux Ext4 Filesystem]     [Union Filesystem]     [Filesystem Testing]     [Ceph Users]     [Ecryptfs]     [AutoFS]     [Kernel Newbies]     [Share Photos]     [Security]     [Netfilter]     [Bugtraq]     [Yosemite News]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux Cachefs]     [Reiser Filesystem]     [Linux RAID]     [Samba]     [Device Mapper]     [CEPH Development]

  Powered by Linux