Skip to contentSkip to conversation

hugetlbfs: don't delete error page from pagecache

PATCH Applied James Houghton 5 participants 13 messages Oct 18, 2022 – Nov 7, 2022

Subsystems (3)

v1 of 2 Applied 1/1 Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 hugetlbfs: don't delete error page from pagecache

This change is very similar to the change that was made for shmem [1], and it solves the same problem but for HugeTLBFS instead.

Currently, when poison is found in a HugeTLB page, the page is removed from the page cache. That means that attempting to map or read that hugepage in the future will result in a new hugepage being allocated instead of notifying the user that the page was poisoned. As [1] states, this is effectively memory corruption.

The fix is to leave the page in the page cache. If the user attempts to use a poisoned HugeTLB page with a syscall, the syscall will fail with EIO, the same error code that shmem uses. For attempts to map the page, the thread will get a BUS_MCEERR_AR SIGBUS.

[1]: commit a76054266661 ("mm: shmem: don't truncate page if memory failure happens")

---
 fs/hugetlbfs/inode.c | 13 ++++++-------
 mm/hugetlb.c         |  4 ++++
 mm/memory-failure.c  |  5 ++++-
 3 files changed, 14 insertions(+), 8 deletions(-)
fs/hugetlbfs/inode.c +6 -7
index fef5165b73a5..7f836f8f9db1 100644
@@ -328,6 +328,12 @@ static ssize_t hugetlbfs_read_iter(struct kiocb *iocb, struct iov_iter *to)
 		} else {
 			unlock_page(page);
 
+			if (PageHWPoison(page)) {
+				put_page(page);
+				retval = -EIO;
+				break;
+			}
+
 			/*
 			 * We have the page, copy it to user space buffer.
 			 */
@@ -1111,13 +1117,6 @@ static int hugetlbfs_migrate_folio(struct address_space *mapping,
 static int hugetlbfs_error_remove_page(struct address_space *mapping,
 				struct page *page)
 {
-	struct inode *inode = mapping->host;
-	pgoff_t index = page->index;
-
-	hugetlb_delete_from_page_cache(page_folio(page));
-	if (unlikely(hugetlb_unreserve_pages(inode, index, index + 1, 1)))
-		hugetlb_fix_reserve_counts(inode);
-
 	return 0;
 }
 
mm/hugetlb.c +4 -0
index 97896165fd3f..5120a9ccbf5b 100644
@@ -6101,6 +6101,10 @@ int hugetlb_mcopy_atomic_pte(struct mm_struct *dst_mm,
 
 	ptl = huge_pte_lock(h, dst_mm, dst_pte);
 
+	ret = -EIO;
+	if (PageHWPoison(page))
+		goto out_release_unlock;
+
 	/*
 	 * We allow to overwrite a pte marker: consider when both MISSING|WP
 	 * registered, we firstly wr-protect a none pte which has no page cache
mm/memory-failure.c +4 -1
index 145bb561ddb3..bead6bccc7f2 100644
@@ -1080,6 +1080,7 @@ static int me_huge_page(struct page_state *ps, struct page *p)
 	int res;
 	struct page *hpage = compound_head(p);
 	struct address_space *mapping;
+	bool extra_pins = false;
 
 	if (!PageHuge(hpage))
 		return MF_DELAYED;
@@ -1087,6 +1088,8 @@ static int me_huge_page(struct page_state *ps, struct page *p)
 	mapping = page_mapping(hpage);
 	if (mapping) {
 		res = truncate_error_page(hpage, page_to_pfn(p), mapping);
+		/* The page is kept in page cache. */
+		extra_pins = true;
 		unlock_page(hpage);
 	} else {
 		unlock_page(hpage);
@@ -1104,7 +1107,7 @@ static int me_huge_page(struct page_state *ps, struct page *p)
 		}
 	}
 
-	if (has_extra_refcount(ps, p, false))
+	if (has_extra_refcount(ps, p, extra_pins))
 		res = MF_FAILED;
 
 	return res;
Signature · 1 line
2.38.0.413.g74048e4d9e-goog
#1 lore raw
Quoted 10 lines · James Houghton

This change is very similar to the change that was made for shmem [1], and it solves the same problem but for HugeTLBFS instead.

Currently, when poison is found in a HugeTLB page, the page is removed from the page cache. That means that attempting to map or read that hugepage in the future will result in a new hugepage being allocated instead of notifying the user that the page was poisoned. As [1] states, this is effectively memory corruption.

Quoted 8 lines

The fix is to leave the page in the page cache. If the user attempts to use a poisoned HugeTLB page with a syscall, the syscall will fail with EIO, the same error code that shmem uses. For attempts to map the page, the thread will get a BUS_MCEERR_AR SIGBUS.

[1]: commit a76054266661 ("mm: shmem: don't truncate page if memory failure happens")

Thanks for the patch. Yes, we should do the same thing for hugetlbfs. When I was working on shmem I did look into hugetlbfs too. But the problem is we actually make the whole hugetlb page unavailable even though just one 4K sub page is hwpoisoned. It may be fine to 2M hugetlb page, but a lot of memory may be a huge waste for 1G hugetlb page, particular for the page fault path.

So I discussed this with Mike offline last year, and I was told Google was working on PTE mapped hugetlb page. That should be able to solve the problem. And we'd like to have the high-granularity hugetlb mapping support as the predecessor.

There were some other details, but I can't remember all of them, I have to refresh my memory by rereading the email discussions...

Quoted 85 lines
---
 fs/hugetlbfs/inode.c | 13 ++++++-------
 mm/hugetlb.c         |  4 ++++
 mm/memory-failure.c  |  5 ++++-
 3 files changed, 14 insertions(+), 8 deletions(-)
fs/hugetlbfs/inode.c +6 -7
index fef5165b73a5..7f836f8f9db1 100644
@@ -328,6 +328,12 @@ static ssize_t hugetlbfs_read_iter(struct kiocb *iocb, struct iov_iter *to)
                } else {
                        unlock_page(page);
 
+                       if (PageHWPoison(page)) {
+                               put_page(page);
+                               retval = -EIO;
+                               break;
+                       }
+
                        /*
                         * We have the page, copy it to user space buffer.
                         */
@@ -1111,13 +1117,6 @@ static int hugetlbfs_migrate_folio(struct address_space *mapping,
 static int hugetlbfs_error_remove_page(struct address_space *mapping,
                                struct page *page)
 {
-       struct inode *inode = mapping->host;
-       pgoff_t index = page->index;
-
-       hugetlb_delete_from_page_cache(page_folio(page));
-       if (unlikely(hugetlb_unreserve_pages(inode, index, index + 1, 1)))
-               hugetlb_fix_reserve_counts(inode);
-
        return 0;
 }
 
mm/hugetlb.c +4 -0
index 97896165fd3f..5120a9ccbf5b 100644
@@ -6101,6 +6101,10 @@ int hugetlb_mcopy_atomic_pte(struct mm_struct *dst_mm,
 
        ptl = huge_pte_lock(h, dst_mm, dst_pte);
 
+       ret = -EIO;
+       if (PageHWPoison(page))
+               goto out_release_unlock;
+
        /*
         * We allow to overwrite a pte marker: consider when both MISSING|WP
         * registered, we firstly wr-protect a none pte which has no page cache
mm/memory-failure.c +4 -2
index 145bb561ddb3..bead6bccc7f2 100644
@@ -1080,6 +1080,7 @@ static int me_huge_page(struct page_state *ps, struct page *p)
        int res;
        struct page *hpage = compound_head(p);
        struct address_space *mapping;
+       bool extra_pins = false;
 
        if (!PageHuge(hpage))
                return MF_DELAYED;
@@ -1087,6 +1088,8 @@ static int me_huge_page(struct page_state *ps, struct page *p)
        mapping = page_mapping(hpage);
        if (mapping) {
                res = truncate_error_page(hpage, page_to_pfn(p), mapping);
+               /* The page is kept in page cache. */
+               extra_pins = true;
                unlock_page(hpage);
        } else {
                unlock_page(hpage);
@@ -1104,7 +1107,7 @@ static int me_huge_page(struct page_state *ps, struct page *p)
                }
        }
 
-       if (has_extra_refcount(ps, p, false))
+       if (has_extra_refcount(ps, p, extra_pins))
                res = MF_FAILED;
 
        return res;
--

2.38.0.413.g74048e4d9e-goog

#2 lore raw
Quoted 19 lines · Yang Shi
Quoted 17 lines · James Houghton

This change is very similar to the change that was made for shmem [1], and it solves the same problem but for HugeTLBFS instead.

Currently, when poison is found in a HugeTLB page, the page is removed from the page cache. That means that attempting to map or read that hugepage in the future will result in a new hugepage being allocated instead of notifying the user that the page was poisoned. As [1] states, this is effectively memory corruption.

The fix is to leave the page in the page cache. If the user attempts to use a poisoned HugeTLB page with a syscall, the syscall will fail with EIO, the same error code that shmem uses. For attempts to map the page, the thread will get a BUS_MCEERR_AR SIGBUS.

[1]: commit a76054266661 ("mm: shmem: don't truncate page if memory failure happens")

Quoted 8 lines
Quoted 1 line

Thanks for the patch. Yes, we should do the same thing for hugetlbfs. When I was working on shmem I did look into hugetlbfs too. But the problem is we actually make the whole hugetlb page unavailable even though just one 4K sub page is hwpoisoned. It may be fine to 2M hugetlb page, but a lot of memory may be a huge waste for 1G hugetlb page, particular for the page fault path.

Right -- it is wasted until a hole is punched or the file is truncated. Although we're wasting the rest of the hugepage for a little longer with this patch, I think it's worth it to have correct behavior.

Quoted 8 lines

So I discussed this with Mike offline last year, and I was told Google was working on PTE mapped hugetlb page. That should be able to solve the problem. And we'd like to have the high-granularity hugetlb mapping support as the predecessor.

There were some other details, but I can't remember all of them, I have to refresh my memory by rereading the email discussions...

Yes! I am working on this. :) I will send up a series in the coming weeks that implements basic support for high-granularity mapping (HGM). This patch is required for hwpoison semantics to work properly for high-granularity mapping (and, as the patch states, for shared HugeTLB mappings generally). For HGM, if we partially map a hugepage and find poison, faulting on the unmapped bits of it will allocate a new hugepage. By keeping the poisoned page in the pagecache, we correctly give userspace a SIGBUS. I didn't mention this in the commit description because I think this patch is correct on its own.

I haven't implemented PAGE_SIZE poisoning of HugeTLB pages yet, but high-granularity mapping unblocks this work. Hopefully that will be ready in the coming months. :)

- James Houghton

Quoted 86 lines
Quoted 86 lines
---
 fs/hugetlbfs/inode.c | 13 ++++++-------
 mm/hugetlb.c         |  4 ++++
 mm/memory-failure.c  |  5 ++++-
 3 files changed, 14 insertions(+), 8 deletions(-)
fs/hugetlbfs/inode.c +6 -7
index fef5165b73a5..7f836f8f9db1 100644
@@ -328,6 +328,12 @@ static ssize_t hugetlbfs_read_iter(struct kiocb *iocb, struct iov_iter *to)
                } else {
                        unlock_page(page);
 
+                       if (PageHWPoison(page)) {
+                               put_page(page);
+                               retval = -EIO;
+                               break;
+                       }
+
                        /*
                         * We have the page, copy it to user space buffer.
                         */
@@ -1111,13 +1117,6 @@ static int hugetlbfs_migrate_folio(struct address_space *mapping,
 static int hugetlbfs_error_remove_page(struct address_space *mapping,
                                struct page *page)
 {
-       struct inode *inode = mapping->host;
-       pgoff_t index = page->index;
-
-       hugetlb_delete_from_page_cache(page_folio(page));
-       if (unlikely(hugetlb_unreserve_pages(inode, index, index + 1, 1)))
-               hugetlb_fix_reserve_counts(inode);
-
        return 0;
 }
 
mm/hugetlb.c +4 -0
index 97896165fd3f..5120a9ccbf5b 100644
@@ -6101,6 +6101,10 @@ int hugetlb_mcopy_atomic_pte(struct mm_struct *dst_mm,
 
        ptl = huge_pte_lock(h, dst_mm, dst_pte);
 
+       ret = -EIO;
+       if (PageHWPoison(page))
+               goto out_release_unlock;
+
        /*
         * We allow to overwrite a pte marker: consider when both MISSING|WP
         * registered, we firstly wr-protect a none pte which has no page cache
mm/memory-failure.c +4 -2
index 145bb561ddb3..bead6bccc7f2 100644
@@ -1080,6 +1080,7 @@ static int me_huge_page(struct page_state *ps, struct page *p)
        int res;
        struct page *hpage = compound_head(p);
        struct address_space *mapping;
+       bool extra_pins = false;
 
        if (!PageHuge(hpage))
                return MF_DELAYED;
@@ -1087,6 +1088,8 @@ static int me_huge_page(struct page_state *ps, struct page *p)
        mapping = page_mapping(hpage);
        if (mapping) {
                res = truncate_error_page(hpage, page_to_pfn(p), mapping);
+               /* The page is kept in page cache. */
+               extra_pins = true;
                unlock_page(hpage);
        } else {
                unlock_page(hpage);
@@ -1104,7 +1107,7 @@ static int me_huge_page(struct page_state *ps, struct page *p)
                }
        }
 
-       if (has_extra_refcount(ps, p, false))
+       if (has_extra_refcount(ps, p, extra_pins))
                res = MF_FAILED;
 
        return res;
--

2.38.0.413.g74048e4d9e-goog

#3 lore raw
Quoted 44 lines · James Houghton
Quoted 19 lines · Yang Shi
Quoted 17 lines · James Houghton

This change is very similar to the change that was made for shmem [1], and it solves the same problem but for HugeTLBFS instead.

Currently, when poison is found in a HugeTLB page, the page is removed from the page cache. That means that attempting to map or read that hugepage in the future will result in a new hugepage being allocated instead of notifying the user that the page was poisoned. As [1] states, this is effectively memory corruption.

The fix is to leave the page in the page cache. If the user attempts to use a poisoned HugeTLB page with a syscall, the syscall will fail with EIO, the same error code that shmem uses. For attempts to map the page, the thread will get a BUS_MCEERR_AR SIGBUS.

[1]: commit a76054266661 ("mm: shmem: don't truncate page if memory failure happens")

Quoted 8 lines
Quoted 1 line

Thanks for the patch. Yes, we should do the same thing for hugetlbfs. When I was working on shmem I did look into hugetlbfs too. But the problem is we actually make the whole hugetlb page unavailable even though just one 4K sub page is hwpoisoned. It may be fine to 2M hugetlb page, but a lot of memory may be a huge waste for 1G hugetlb page, particular for the page fault path.

Right -- it is wasted until a hole is punched or the file is truncated. Although we're wasting the rest of the hugepage for a little longer with this patch, I think it's worth it to have correct behavior.

Quoted 8 lines

So I discussed this with Mike offline last year, and I was told Google was working on PTE mapped hugetlb page. That should be able to solve the problem. And we'd like to have the high-granularity hugetlb mapping support as the predecessor.

There were some other details, but I can't remember all of them, I have to refresh my memory by rereading the email discussions...

Quoted 9 lines

Yes! I am working on this. :) I will send up a series in the coming weeks that implements basic support for high-granularity mapping (HGM). This patch is required for hwpoison semantics to work properly for high-granularity mapping (and, as the patch states, for shared HugeTLB mappings generally). For HGM, if we partially map a hugepage and find poison, faulting on the unmapped bits of it will allocate a new hugepage. By keeping the poisoned page in the pagecache, we correctly give userspace a SIGBUS. I didn't mention this in the commit description because I think this patch is correct on its own.

I don't mean it is not correct. I'm not sure which one (this patch or HGM) would go first. But it sounds like you thought this patch should be the predecessor of HGM since it could deliver userspace the correct signal?

Quoted 93 lines

I haven't implemented PAGE_SIZE poisoning of HugeTLB pages yet, but high-granularity mapping unblocks this work. Hopefully that will be ready in the coming months. :)

- James Houghton

Quoted 86 lines
Quoted 86 lines
---
 fs/hugetlbfs/inode.c | 13 ++++++-------
 mm/hugetlb.c         |  4 ++++
 mm/memory-failure.c  |  5 ++++-
 3 files changed, 14 insertions(+), 8 deletions(-)
fs/hugetlbfs/inode.c +6 -7
index fef5165b73a5..7f836f8f9db1 100644
@@ -328,6 +328,12 @@ static ssize_t hugetlbfs_read_iter(struct kiocb *iocb, struct iov_iter *to)
                } else {
                        unlock_page(page);
 
+                       if (PageHWPoison(page)) {
+                               put_page(page);
+                               retval = -EIO;
+                               break;
+                       }
+
                        /*
                         * We have the page, copy it to user space buffer.
                         */
@@ -1111,13 +1117,6 @@ static int hugetlbfs_migrate_folio(struct address_space *mapping,
 static int hugetlbfs_error_remove_page(struct address_space *mapping,
                                struct page *page)
 {
-       struct inode *inode = mapping->host;
-       pgoff_t index = page->index;
-
-       hugetlb_delete_from_page_cache(page_folio(page));
-       if (unlikely(hugetlb_unreserve_pages(inode, index, index + 1, 1)))
-               hugetlb_fix_reserve_counts(inode);
-
        return 0;
 }
 
mm/hugetlb.c +4 -0
index 97896165fd3f..5120a9ccbf5b 100644
@@ -6101,6 +6101,10 @@ int hugetlb_mcopy_atomic_pte(struct mm_struct *dst_mm,
 
        ptl = huge_pte_lock(h, dst_mm, dst_pte);
 
+       ret = -EIO;
+       if (PageHWPoison(page))
+               goto out_release_unlock;
+
        /*
         * We allow to overwrite a pte marker: consider when both MISSING|WP
         * registered, we firstly wr-protect a none pte which has no page cache
mm/memory-failure.c +4 -2
index 145bb561ddb3..bead6bccc7f2 100644
@@ -1080,6 +1080,7 @@ static int me_huge_page(struct page_state *ps, struct page *p)
        int res;
        struct page *hpage = compound_head(p);
        struct address_space *mapping;
+       bool extra_pins = false;
 
        if (!PageHuge(hpage))
                return MF_DELAYED;
@@ -1087,6 +1088,8 @@ static int me_huge_page(struct page_state *ps, struct page *p)
        mapping = page_mapping(hpage);
        if (mapping) {
                res = truncate_error_page(hpage, page_to_pfn(p), mapping);
+               /* The page is kept in page cache. */
+               extra_pins = true;
                unlock_page(hpage);
        } else {
                unlock_page(hpage);
@@ -1104,7 +1107,7 @@ static int me_huge_page(struct page_state *ps, struct page *p)
                }
        }
 
-       if (has_extra_refcount(ps, p, false))
+       if (has_extra_refcount(ps, p, extra_pins))
                res = MF_FAILED;
 
        return res;
--

2.38.0.413.g74048e4d9e-goog

#4 lore raw
In reply to Yang Shi
Quoted 18 lines · Yang Shi
Quoted 17 lines · James Houghton

This change is very similar to the change that was made for shmem [1], and it solves the same problem but for HugeTLBFS instead.

Currently, when poison is found in a HugeTLB page, the page is removed from the page cache. That means that attempting to map or read that hugepage in the future will result in a new hugepage being allocated instead of notifying the user that the page was poisoned. As [1] states, this is effectively memory corruption.

The fix is to leave the page in the page cache. If the user attempts to use a poisoned HugeTLB page with a syscall, the syscall will fail with EIO, the same error code that shmem uses. For attempts to map the page, the thread will get a BUS_MCEERR_AR SIGBUS.

[1]: commit a76054266661 ("mm: shmem: don't truncate page if memory failure happens")

Quoted 8 lines
Quoted 1 line

Thanks for the patch. Yes, we should do the same thing for hugetlbfs. When I was working on shmem I did look into hugetlbfs too. But the problem is we actually make the whole hugetlb page unavailable even though just one 4K sub page is hwpoisoned. It may be fine to 2M hugetlb page, but a lot of memory may be a huge waste for 1G hugetlb page, particular for the page fault path.

One thing that complicated this a bit is the vmemmap optimizations for hugetlb. However, I believe Naoya may have addressed this recently.

Quoted 4 lines

So I discussed this with Mike offline last year, and I was told Google was working on PTE mapped hugetlb page. That should be able to solve the problem. And we'd like to have the high-granularity hugetlb mapping support as the predecessor.

Yes, I went back in my notes and noticed it had been one year. No offense intended to James and his great work on HGM. However, in hindsight we should have fixed this in some way without waiting for a HGM based.

Quoted 2 lines

There were some other details, but I can't remember all of them, I have to refresh my memory by rereading the email discussions...

I think the complicating factor was vmemmap optimization. As mentioned above, this may have already been addressed by Naoya in patches to indicate which sub-page(s) had the actual error.

As Yang Shi notes, this patch makes the entire hugetlb page inaccessible. With some work, we could allow reads to everything but the sub-page with error. However, this should be much easier with HGM. And, we could potentially even do page faults everywhere but the sub-page with error.

I still think it may be better to wait for HGM instead of trying to do read access to all but sub-page with error now. But, entirely open to other opinions.

I plan to do a review of this patch a little later.

Signature · 1 line
Mike Kravetz
#5 lore raw
Quoted 48 lines · Mike Kravetz
Quoted 18 lines · Yang Shi
Quoted 17 lines · James Houghton

This change is very similar to the change that was made for shmem [1], and it solves the same problem but for HugeTLBFS instead.

Currently, when poison is found in a HugeTLB page, the page is removed from the page cache. That means that attempting to map or read that hugepage in the future will result in a new hugepage being allocated instead of notifying the user that the page was poisoned. As [1] states, this is effectively memory corruption.

The fix is to leave the page in the page cache. If the user attempts to use a poisoned HugeTLB page with a syscall, the syscall will fail with EIO, the same error code that shmem uses. For attempts to map the page, the thread will get a BUS_MCEERR_AR SIGBUS.

[1]: commit a76054266661 ("mm: shmem: don't truncate page if memory failure happens")

Quoted 8 lines
Quoted 1 line

Thanks for the patch. Yes, we should do the same thing for hugetlbfs. When I was working on shmem I did look into hugetlbfs too. But the problem is we actually make the whole hugetlb page unavailable even though just one 4K sub page is hwpoisoned. It may be fine to 2M hugetlb page, but a lot of memory may be a huge waste for 1G hugetlb page, particular for the page fault path.

One thing that complicated this a bit is the vmemmap optimizations for hugetlb. However, I believe Naoya may have addressed this recently.

Quoted 4 lines

So I discussed this with Mike offline last year, and I was told Google was working on PTE mapped hugetlb page. That should be able to solve the problem. And we'd like to have the high-granularity hugetlb mapping support as the predecessor.

Yes, I went back in my notes and noticed it had been one year. No offense intended to James and his great work on HGM. However, in hindsight we should have fixed this in some way without waiting for a HGM based.

Quoted 2 lines

There were some other details, but I can't remember all of them, I have to refresh my memory by rereading the email discussions...

I think the complicating factor was vmemmap optimization. As mentioned above, this may have already been addressed by Naoya in patches to indicate which sub-page(s) had the actual error.

Quoted 8 lines

As Yang Shi notes, this patch makes the entire hugetlb page inaccessible. With some work, we could allow reads to everything but the sub-page with error. However, this should be much easier with HGM. And, we could potentially even do page faults everywhere but the sub-page with error.

I still think it may be better to wait for HGM instead of trying to do read access to all but sub-page with error now. But, entirely open to other opinions.

I have no strong preference about which goes first.

Quoted 4 lines

I plan to do a review of this patch a little later. --
Mike Kravetz

Reviewed-by on v1 1/1
#6 lore raw
In reply to James Houghton
Quoted 14 lines · James Houghton

This change is very similar to the change that was made for shmem [1], and it solves the same problem but for HugeTLBFS instead.

Currently, when poison is found in a HugeTLB page, the page is removed from the page cache. That means that attempting to map or read that hugepage in the future will result in a new hugepage being allocated instead of notifying the user that the page was poisoned. As [1] states, this is effectively memory corruption.

The fix is to leave the page in the page cache. If the user attempts to use a poisoned HugeTLB page with a syscall, the syscall will fail with EIO, the same error code that shmem uses. For attempts to map the page, the thread will get a BUS_MCEERR_AR SIGBUS.

Quoted 8 lines

[1]: commit a76054266661 ("mm: shmem: don't truncate page if memory failure happens")

---
 fs/hugetlbfs/inode.c | 13 ++++++-------
 mm/hugetlb.c         |  4 ++++
 mm/memory-failure.c  |  5 ++++-
 3 files changed, 14 insertions(+), 8 deletions(-)

Thanks James!

Code looks correct. One observation below, but I do not suggest changing.

Quoted 60 lines
fs/hugetlbfs/inode.c +6 -7
index fef5165b73a5..7f836f8f9db1 100644
@@ -328,6 +328,12 @@ static ssize_t hugetlbfs_read_iter(struct kiocb *iocb, struct iov_iter *to)
 		} else {
 			unlock_page(page);
 
+			if (PageHWPoison(page)) {
+				put_page(page);
+				retval = -EIO;
+				break;
+			}
+
 			/*
 			 * We have the page, copy it to user space buffer.
 			 */
@@ -1111,13 +1117,6 @@ static int hugetlbfs_migrate_folio(struct address_space *mapping,
 static int hugetlbfs_error_remove_page(struct address_space *mapping,
 				struct page *page)
 {
-	struct inode *inode = mapping->host;
-	pgoff_t index = page->index;
-
-	hugetlb_delete_from_page_cache(page_folio(page));
-	if (unlikely(hugetlb_unreserve_pages(inode, index, index + 1, 1)))
-		hugetlb_fix_reserve_counts(inode);
-
 	return 0;
 }
 
mm/hugetlb.c +4 -0
index 97896165fd3f..5120a9ccbf5b 100644
@@ -6101,6 +6101,10 @@ int hugetlb_mcopy_atomic_pte(struct mm_struct *dst_mm,
 
 	ptl = huge_pte_lock(h, dst_mm, dst_pte);
 
+	ret = -EIO;
+	if (PageHWPoison(page))
+		goto out_release_unlock;
+
 	/*
 	 * We allow to overwrite a pte marker: consider when both MISSING|WP
 	 * registered, we firstly wr-protect a none pte which has no page cache
mm/memory-failure.c +1 -0
index 145bb561ddb3..bead6bccc7f2 100644
@@ -1080,6 +1080,7 @@ static int me_huge_page(struct page_state *ps, struct page *p)
 	int res;
 	struct page *hpage = compound_head(p);
 	struct address_space *mapping;
+	bool extra_pins = false;
 
 	if (!PageHuge(hpage))
 		return MF_DELAYED;

mm/memory-failure.c static int me_huge_page(struct page_state *ps, struct page *p) · line 1091

Quoted 4 lines
mm/memory-failure.c +1 -0
@@ -1087,6 +1088,8 @@ static int me_huge_page(struct page_state *ps, struct page *p)
 	mapping = page_mapping(hpage);
 	if (mapping) {
 		res = truncate_error_page(hpage, page_to_pfn(p), mapping);
+		/* The page is kept in page cache. */

That looks a bit silly as we are know truncate_error_page() is a noop and call it anyway. I suppose we could just set 'res = MF_RECOVERED'. However, it really should be left as is in case more non-hugetlb functionality is added to truncate_error_page() in the future.

Signature · 1 line
Mike Kravetz
Quoted 16 lines
mm/memory-failure.c +2 -1
+		extra_pins = true;
 		unlock_page(hpage);
 	} else {
 		unlock_page(hpage);
@@ -1104,7 +1107,7 @@ static int me_huge_page(struct page_state *ps, struct page *p)
 		}
 	}
 
-	if (has_extra_refcount(ps, p, false))
+	if (has_extra_refcount(ps, p, extra_pins))
 		res = MF_FAILED;
 
 	return res;
Signature · 1 line
2.38.0.413.g74048e4d9e-goog
Tested-by Reviewed-by on v1 1/1
#7 lore raw
In reply to James Houghton
Quoted 9 lines · James Houghton

This change is very similar to the change that was made for shmem [1], and it solves the same problem but for HugeTLBFS instead.

Currently, when poison is found in a HugeTLB page, the page is removed from the page cache. That means that attempting to map or read that hugepage in the future will result in a new hugepage being allocated instead of notifying the user that the page was poisoned. As [1] states, this is effectively memory corruption.

Quoted 8 lines

The fix is to leave the page in the page cache. If the user attempts to use a poisoned HugeTLB page with a syscall, the syscall will fail with EIO, the same error code that shmem uses. For attempts to map the page, the thread will get a BUS_MCEERR_AR SIGBUS.

[1]: commit a76054266661 ("mm: shmem: don't truncate page if memory failure happens")

I did some testing and found no issue. So I agree with this patch. Thank you very much.

As for whether to go with HGM patchset or not, I have no strong opinion. As you stated in another email this patch is correct without HGM patch, so it's OK to me to make this merged first.

Thanks,
Naoya Horiguchi

#8 lore raw
Quoted 22 lines · HORIGUCHI NAOYA(堀口 直也)
Quoted 9 lines · James Houghton

This change is very similar to the change that was made for shmem [1], and it solves the same problem but for HugeTLBFS instead.

Currently, when poison is found in a HugeTLB page, the page is removed from the page cache. That means that attempting to map or read that hugepage in the future will result in a new hugepage being allocated instead of notifying the user that the page was poisoned. As [1] states, this is effectively memory corruption.

Quoted 8 lines

The fix is to leave the page in the page cache. If the user attempts to use a poisoned HugeTLB page with a syscall, the syscall will fail with EIO, the same error code that shmem uses. For attempts to map the page, the thread will get a BUS_MCEERR_AR SIGBUS.

[1]: commit a76054266661 ("mm: shmem: don't truncate page if memory failure happens")

I did some testing and found no issue. So I agree with this patch. Thank you very much.

Quoted 6 lines

As for whether to go with HGM patchset or not, I have no strong opinion. As you stated in another email this patch is correct without HGM patch, so it's OK to me to make this merged first.

Thanks Naoya.

This is a late thought, but ...
Should this patch and Yang Shi's shmem patch be backported to stable releases? Both address potential data corruption/loss, so it certainly seems like stable material.

Signature · 1 line
Mike Kravetz
#9 lore raw
Quoted 30 lines · Mike Kravetz
Quoted 22 lines · HORIGUCHI NAOYA(堀口 直也)
Quoted 9 lines · James Houghton

This change is very similar to the change that was made for shmem [1], and it solves the same problem but for HugeTLBFS instead.

Currently, when poison is found in a HugeTLB page, the page is removed from the page cache. That means that attempting to map or read that hugepage in the future will result in a new hugepage being allocated instead of notifying the user that the page was poisoned. As [1] states, this is effectively memory corruption.

Quoted 8 lines

The fix is to leave the page in the page cache. If the user attempts to use a poisoned HugeTLB page with a syscall, the syscall will fail with EIO, the same error code that shmem uses. For attempts to map the page, the thread will get a BUS_MCEERR_AR SIGBUS.

[1]: commit a76054266661 ("mm: shmem: don't truncate page if memory failure happens")

I did some testing and found no issue. So I agree with this patch. Thank you very much.

Quoted 6 lines

As for whether to go with HGM patchset or not, I have no strong opinion. As you stated in another email this patch is correct without HGM patch, so it's OK to me to make this merged first.

Quoted 6 lines

Thanks Naoya.

This is a late thought, but ...
Should this patch and Yang Shi's shmem patch be backported to stable releases? Both address potential data corruption/loss, so it certainly seems like stable material.

Yes, I agree that backporting these could be helpful.

So I think that I'll try to backport commit a7605426666 and its dependencies to 5.15 (and older LTS if possible). For this patch, just adding "Cc: stable" should be enough for now.

Thanks,
Naoya Horiguchi

#10 lore raw
Quoted 38 lines · HORIGUCHI NAOYA(堀口 直也)
Quoted 30 lines · Mike Kravetz
Quoted 22 lines · HORIGUCHI NAOYA(堀口 直也)
Quoted 9 lines · James Houghton

This change is very similar to the change that was made for shmem [1], and it solves the same problem but for HugeTLBFS instead.

Currently, when poison is found in a HugeTLB page, the page is removed from the page cache. That means that attempting to map or read that hugepage in the future will result in a new hugepage being allocated instead of notifying the user that the page was poisoned. As [1] states, this is effectively memory corruption.

Quoted 8 lines

The fix is to leave the page in the page cache. If the user attempts to use a poisoned HugeTLB page with a syscall, the syscall will fail with EIO, the same error code that shmem uses. For attempts to map the page, the thread will get a BUS_MCEERR_AR SIGBUS.

[1]: commit a76054266661 ("mm: shmem: don't truncate page if memory failure happens")

I did some testing and found no issue. So I agree with this patch. Thank you very much.

Quoted 6 lines

As for whether to go with HGM patchset or not, I have no strong opinion. As you stated in another email this patch is correct without HGM patch, so it's OK to me to make this merged first.

Quoted 6 lines

Thanks Naoya.

This is a late thought, but ...
Should this patch and Yang Shi's shmem patch be backported to stable releases? Both address potential data corruption/loss, so it certainly seems like stable material.

Quoted 5 lines

Yes, I agree that backporting these could be helpful.

So I think that I'll try to backport commit a7605426666 and its dependencies to 5.15 (and older LTS if possible). For this patch, just adding "Cc: stable" should be enough for now.

We have now actually seen users impacted by this issue. I suggest adding the following tags to this patch.

Fixes 78bb920344b8 ("mm: hwpoison: dissolve in-use hugepage in unrecoverable memory error")
Signature · 1 line
Mike Kravetz
maintainer
#11 lore raw
Quoted 5 lines · On Mon, 7 Nov 2022 08:55:43 -0800 Mike Kravetz <[email protected]> wrote:

We have now actually seen users impacted by this issue. I suggest adding the following tags to this patch.

Fixes 78bb920344b8 ("mm: hwpoison: dissolve in-use hugepage in unrecoverable memory error")

I've made those changes.

#12 lore raw
In reply to HORIGUCHI NAOYA(堀口 直也)
Quoted 32 lines · HORIGUCHI NAOYA(堀口 直也)
Quoted 30 lines · Mike Kravetz
Quoted 28 lines · HORIGUCHI NAOYA(堀口 直也)
Quoted 9 lines · James Houghton

This change is very similar to the change that was made for shmem [1], and it solves the same problem but for HugeTLBFS instead.

Currently, when poison is found in a HugeTLB page, the page is removed from the page cache. That means that attempting to map or read that hugepage in the future will result in a new hugepage being allocated instead of notifying the user that the page was poisoned. As [1] states, this is effectively memory corruption.

Quoted 8 lines

The fix is to leave the page in the page cache. If the user attempts to use a poisoned HugeTLB page with a syscall, the syscall will fail with EIO, the same error code that shmem uses. For attempts to map the page, the thread will get a BUS_MCEERR_AR SIGBUS.

[1]: commit a76054266661 ("mm: shmem: don't truncate page if memory failure happens")

I did some testing and found no issue. So I agree with this patch. Thank you very much.

As for whether to go with HGM patchset or not, I have no strong opinion. As you stated in another email this patch is correct without HGM patch, so it's OK to me to make this merged first.

Quoted 8 lines
Quoted 6 lines

Thanks Naoya.

This is a late thought, but ...
Should this patch and Yang Shi's shmem patch be backported to stable releases? Both address potential data corruption/loss, so it certainly seems like stable material.

Yes, I agree that backporting these could be helpful.

+1

Quoted 4 lines

So I think that I'll try to backport commit a7605426666 and its dependencies to 5.15 (and older LTS if possible). For this patch, just adding "Cc: stable" should be enough for now.

Thanks, Naoya.

Quoted 3 lines

Thanks,
Naoya Horiguchi