Skip to contentSkip to conversation

linux-user/mmap.c: fix integer underflow in target_mremap

PATCH Applied Jonathan Marler 4 participants 8 messages May 2, 2020 – May 28, 2020

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 linux-user/mmap.c: fix integer underflow in target_mremap

This code path in mmap occurs when a page size is decreased with mremap. When a section of pages is shrunk, qemu calls mmap_reserve on the pages that were released. However, it has the diff operation reversed, subtracting the larger old_size from the smaller new_size. Instead, it should be subtracting the smaller new_size from the larger old_size. You can also see in the previous line of the change that this mmap_reserve call only occurs when old_size > new_size.

---
 linux-user/mmap.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
linux-user/mmap.c +1 -1
index e378033797..caab62909e 100644
@@ -708,7 +708,7 @@ abi_long target_mremap(abi_ulong old_addr, abi_ulong old_size,
         if (prot == 0) {
             host_addr = mremap(g2h(old_addr), old_size, new_size, flags);
             if (host_addr != MAP_FAILED && reserved_va && old_size > new_size) {
-                mmap_reserve(old_addr + old_size, new_size - old_size);
+                mmap_reserve(old_addr + old_size, old_size - new_size);
             }
         } else {
             errno = ENOMEM;
Signature · 1 line
2.23.1
#1 lore raw

FYI, I applied this patch to the qemu build that zig uses to run non-native tests (
https://github.com/ziglang/qemu-static/blob/master/patch/mremap-underflow.diff
)

After applying it, my new code that calls mremap now passes, whereas before the fix I was getting a segfault.

Quoted 34 lines · Jonathan Marler

This code path in mmap occurs when a page size is decreased with mremap. When a section of pages is shrunk, qemu calls mmap_reserve on the pages that were released. However, it has the diff operation reversed, subtracting the larger old_size from the smaller new_size. Instead, it should be subtracting the smaller new_size from the larger old_size. You can also see in the previous line of the change that this mmap_reserve call only occurs when old_size > new_size.

---
 linux-user/mmap.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
linux-user/mmap.c +1 -1
index e378033797..caab62909e 100644
@@ -708,7 +708,7 @@ abi_long target_mremap(abi_ulong old_addr, abi_ulong
 old_size,
         if (prot == 0) {
             host_addr = mremap(g2h(old_addr), old_size, new_size, flags);
             if (host_addr != MAP_FAILED && reserved_va && old_size >
 new_size) {
-                mmap_reserve(old_addr + old_size, new_size - old_size);
+                mmap_reserve(old_addr + old_size, old_size - new_size);
             }
         } else {
             errno = ENOMEM;

--
2.23.1

#2 lore raw

Been a couple weeks, checking to see if anyone has looked at this.

Quoted 45 lines · Jonathan Marler

FYI, I applied this patch to the qemu build that zig uses to run non-native tests (
https://github.com/ziglang/qemu-static/blob/master/patch/mremap-underflow.diff
)

After applying it, my new code that calls mremap now passes, whereas before the fix I was getting a segfault.

Quoted 34 lines · Jonathan Marler

This code path in mmap occurs when a page size is decreased with mremap. When a section of pages is shrunk, qemu calls mmap_reserve on the pages that were released. However, it has the diff operation reversed, subtracting the larger old_size from the smaller new_size. Instead, it should be subtracting the smaller new_size from the larger old_size. You can also see in the previous line of the change that this mmap_reserve call only occurs when old_size > new_size.

---
 linux-user/mmap.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
linux-user/mmap.c +1 -1
index e378033797..caab62909e 100644
@@ -708,7 +708,7 @@ abi_long target_mremap(abi_ulong old_addr, abi_ulong
 old_size,
         if (prot == 0) {
             host_addr = mremap(g2h(old_addr), old_size, new_size, flags);
             if (host_addr != MAP_FAILED && reserved_va && old_size >
 new_size) {
-                mmap_reserve(old_addr + old_size, new_size - old_size);
+                mmap_reserve(old_addr + old_size, old_size - new_size);
             }
         } else {
             errno = ENOMEM;

--
2.23.1

#3 lore raw

Been a few more days. Not sure how often I should be pinging. If this is too much to ping every few days let me know.

Quoted 51 lines · Jonathan Marler

Been a couple weeks, checking to see if anyone has looked at this.

Quoted 46 lines · Jonathan Marler

FYI, I applied this patch to the qemu build that zig uses to run non-native tests (
https://github.com/ziglang/qemu-static/blob/master/patch/mremap-underflow.diff
)

After applying it, my new code that calls mremap now passes, whereas before the fix I was getting a segfault.

Quoted 35 lines · Jonathan Marler

This code path in mmap occurs when a page size is decreased with mremap. When a section of pages is shrunk, qemu calls mmap_reserve on the pages that were released. However, it has the diff operation reversed, subtracting the larger old_size from the smaller new_size. Instead, it should be subtracting the smaller new_size from the larger old_size. You can also see in the previous line of the change that this mmap_reserve call only occurs when old_size > new_size.

---
 linux-user/mmap.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
linux-user/mmap.c +1 -1
index e378033797..caab62909e 100644
@@ -708,7 +708,7 @@ abi_long target_mremap(abi_ulong old_addr, abi_ulong
 old_size,
         if (prot == 0) {
             host_addr = mremap(g2h(old_addr), old_size, new_size,
 flags);
             if (host_addr != MAP_FAILED && reserved_va && old_size >
 new_size) {
-                mmap_reserve(old_addr + old_size, new_size - old_size);
+                mmap_reserve(old_addr + old_size, old_size - new_size);
             }
         } else {
             errno = ENOMEM;

--
2.23.1

#4 lore raw

Hi Jonathan,
thanks for the patch!

CCing Riku and Laurent.

Quoted 2 lines · Jonathan Marler

Been a few more days. Not sure how often I should be pinging. If this is too much to ping every few days let me know.

Is not too much, but next time is better to CC the maintainers. You can use 'scripts/get_maintainer.pl' to get the list of maintainers and reviewers.

Please take a look at https://wiki.qemu.org/Contribute/SubmitAPatch

Quoted 14 lines
Quoted 10 lines · Jonathan Marler

Been a couple weeks, checking to see if anyone has looked at this.

Quoted 5 lines · Jonathan Marler

FYI, I applied this patch to the qemu build that zig uses to run non-native tests (
https://github.com/ziglang/qemu-static/blob/master/patch/mremap-underflow.diff
)

Quoted 7 lines
Quoted 7 lines
Quoted 7 lines

After applying it, my new code that calls mremap now passes, whereas before the fix I was getting a segfault.

Quoted 1 line · Jonathan Marler

should be "Buglink: https://bugs.launchpad.net/bugs/1876373"

Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines

This code path in mmap occurs when a page size is decreased with mremap. When a section of pages is shrunk, qemu calls mmap_reserve on the pages that were released. However, it has the diff operation reversed, subtracting the larger old_size from the smaller new_size. Instead, it should be subtracting the smaller new_size from the larger old_size. You can also see in the previous line of the change that this mmap_reserve call only occurs when old_size > new_size.

Please break the lines of the commit message (max 76 charactes per line):
https://wiki.qemu.org/Contribute/SubmitAPatch#Write_a_meaningful_commit_message

Thanks,
Stefano

Quoted 26 lines
Quoted 26 lines
Quoted 26 lines
Quoted 26 lines
---
 linux-user/mmap.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
linux-user/mmap.c +1 -1
index e378033797..caab62909e 100644
@@ -708,7 +708,7 @@ abi_long target_mremap(abi_ulong old_addr, abi_ulong
 old_size,
         if (prot == 0) {
             host_addr = mremap(g2h(old_addr), old_size, new_size,
 flags);
             if (host_addr != MAP_FAILED && reserved_va && old_size >
 new_size) {
-                mmap_reserve(old_addr + old_size, new_size - old_size);
+                mmap_reserve(old_addr + old_size, old_size - new_size);
             }
         } else {
             errno = ENOMEM;

--
2.23.1

#5 lore raw

Hi Jonathan.

Quoted 8 lines · Stefano Garzarella

Hi Jonathan,
thanks for the patch!

CCing Riku and Laurent.

Quoted 2 lines · Jonathan Marler

Been a few more days. Not sure how often I should be pinging. If this is too much to ping every few days let me know.

Pinging every week is fine. The problem here, as noticed by Stefano, is you forgot to Cc the maintainers, so they surely missed your patch.

Last time Riku sent an email to qemu-devel was more than 2 years ago, letling Laurent second him, then went MIA:
https://www.mail-archive.com/[email protected]/msg507843.html

I'd say count 1 week starting today.

Regards,

Phil.

Quoted 74 lines

Is not too much, but next time is better to CC the maintainers. You can use 'scripts/get_maintainer.pl' to get the list of maintainers and reviewers.

Please take a look at https://wiki.qemu.org/Contribute/SubmitAPatch

Quoted 14 lines
Quoted 10 lines · Jonathan Marler

Been a couple weeks, checking to see if anyone has looked at this.

Quoted 5 lines · Jonathan Marler

FYI, I applied this patch to the qemu build that zig uses to run non-native tests (
https://github.com/ziglang/qemu-static/blob/master/patch/mremap-underflow.diff
)

Quoted 7 lines
Quoted 7 lines
Quoted 7 lines

After applying it, my new code that calls mremap now passes, whereas before the fix I was getting a segfault.

Quoted 1 line · Jonathan Marler

should be "Buglink: https://bugs.launchpad.net/bugs/1876373"

Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines

This code path in mmap occurs when a page size is decreased with mremap. When a section of pages is shrunk, qemu calls mmap_reserve on the pages that were released. However, it has the diff operation reversed, subtracting the larger old_size from the smaller new_size. Instead, it should be subtracting the smaller new_size from the larger old_size. You can also see in the previous line of the change that this mmap_reserve call only occurs when old_size > new_size.

Please break the lines of the commit message (max 76 charactes per line):
https://wiki.qemu.org/Contribute/SubmitAPatch#Write_a_meaningful_commit_message

Thanks,
Stefano

Quoted 26 lines
Quoted 26 lines
Quoted 26 lines
Quoted 26 lines
---
  linux-user/mmap.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
linux-user/mmap.c +1 -1
index e378033797..caab62909e 100644
@@ -708,7 +708,7 @@ abi_long target_mremap(abi_ulong old_addr, abi_ulong
 old_size,
          if (prot == 0) {
              host_addr = mremap(g2h(old_addr), old_size, new_size,
 flags);
              if (host_addr != MAP_FAILED && reserved_va && old_size >
 new_size) {
-                mmap_reserve(old_addr + old_size, new_size - old_size);
+                mmap_reserve(old_addr + old_size, old_size - new_size);
              }
          } else {
              errno = ENOMEM;

--
2.23.1

Reviewed-by on v1 1/1
#6 lore raw
In reply to Jonathan Marler

Le 02/05/2020 à 18:12, Jonathan Marler a écrit :

Quoted 15 lines

This code path in mmap occurs when a page size is decreased with mremap. When a section of pages is shrunk, qemu calls mmap_reserve on the pages that were released. However, it has the diff operation reversed, subtracting the larger old_size from the smaller new_size. Instead, it should be subtracting the smaller new_size from the larger old_size. You can also see in the previous line of the change that this mmap_reserve call only occurs when old_size > new_size.

---
 linux-user/mmap.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
linux-user/mmap.c +0 -0
index e378033797..caab62909e 100644

linux-user/mmap.c abi_long target_mremap(abi_ulong old_addr, abi_ulong old_size, · line 714

Quoted 8 lines
linux-user/mmap.c +1 -1
@@ -708,7 +708,7 @@ abi_long target_mremap(abi_ulong old_addr, abi_ulong old_size,
         if (prot == 0) {
             host_addr = mremap(g2h(old_addr), old_size, new_size, flags);
             if (host_addr != MAP_FAILED && reserved_va && old_size > new_size) {
-                mmap_reserve(old_addr + old_size, new_size - old_size);
+                mmap_reserve(old_addr + old_size, old_size - new_size);
             }
         } else {
             errno = ENOMEM;
#7 lore raw
In reply to Jonathan Marler

Le 02/05/2020 à 18:12, Jonathan Marler a écrit :

Quoted 15 lines

This code path in mmap occurs when a page size is decreased with mremap. When a section of pages is shrunk, qemu calls mmap_reserve on the pages that were released. However, it has the diff operation reversed, subtracting the larger old_size from the smaller new_size. Instead, it should be subtracting the smaller new_size from the larger old_size. You can also see in the previous line of the change that this mmap_reserve call only occurs when old_size > new_size.

---
 linux-user/mmap.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
linux-user/mmap.c +0 -0
index e378033797..caab62909e 100644

linux-user/mmap.c abi_long target_mremap(abi_ulong old_addr, abi_ulong old_size, · line 714

Quoted 8 lines
linux-user/mmap.c +1 -1
@@ -708,7 +708,7 @@ abi_long target_mremap(abi_ulong old_addr, abi_ulong old_size,
         if (prot == 0) {
             host_addr = mremap(g2h(old_addr), old_size, new_size, flags);
             if (host_addr != MAP_FAILED && reserved_va && old_size > new_size) {
-                mmap_reserve(old_addr + old_size, new_size - old_size);
+                mmap_reserve(old_addr + old_size, old_size - new_size);
             }
         } else {
             errno = ENOMEM;

Applied to my linux-user branch.

Thanks,
Laurent