Skip to contentSkip to conversation

proc: use untagged_addr() for pagemap_read addresses

PATCH RESEND Superseded Miles Chen 4 participants 9 messages Nov 23, 2020 – Nov 27, 2020

Subsystems (4)

v1 of 3 Superseded Latest is v3 →

lore
Share & tools
lore

Miles Chen

author
#0 lore raw

PATCH RESEND 1/1 proc: use untagged_addr() for pagemap_read addresses

When we try to visit the pagemap of a tagged userspace pointer, we find that the start_vaddr is not correct because of the tag.
To fix it, we should untag the usespace pointers in pagemap_read().

I tested with 5.10-rc4 and the issue remains.

My test code is baed on [1]:

A userspace pointer which has been tagged by 0xb4: 0xb400007662f541c8

=== userspace program ===

uint64 OsLayer::VirtualToPhysical(void *vaddr) {
	uint64 frame, paddr, pfnmask, pagemask;
	int pagesize = sysconf(_SC_PAGESIZE);
	off64_t off = ((uintptr_t)vaddr) / pagesize * 8; // off = 0xb400007662f541c8 / pagesize * 8 = 0x5a00003b317aa0
	int fd = open(kPagemapPath, O_RDONLY);
	...

	if (lseek64(fd, off, SEEK_SET) != off || read(fd, &frame, 8) != 8) {
		int err = errno;
		string errtxt = ErrorString(err);
		if (fd >= 0)
			close(fd);
		return 0;
	}

...
}

=== kernel fs/proc/task_mmu.c ===

static ssize_t pagemap_read(struct file *file, char __user *buf,
size_t count, loff_t *ppos)

{
	...
	src = *ppos;
	svpfn = src / PM_ENTRY_BYTES; // svpfn == 0xb400007662f54
	start_vaddr = svpfn << PAGE_SHIFT; // start_vaddr == 0xb400007662f54000
	end_vaddr = mm->task_size;

	/* watch out for wraparound */
	// svpfn == 0xb400007662f54
	// (mm->task_size >> PAGE) == 0x8000000
	if (svpfn > mm->task_size >> PAGE_SHIFT) // the condition is true because of the tag 0xb4
		start_vaddr = end_vaddr;

	ret = 0;
	while (count && (start_vaddr < end_vaddr)) { // we cannot visit correct entry because start_vaddr is set to end_vaddr
		int len;
		unsigned long end;
		...
	}
	...
}

[1] https://github.com/stressapptest/stressapptest/blob/master/src/os.cc#L158

---
 fs/proc/task_mmu.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
fs/proc/task_mmu.c +2 -2
index 217aa2705d5d..e9a70f7ee515 100644
@@ -1599,11 +1599,11 @@ static ssize_t pagemap_read(struct file *file, char __user *buf,
 
 	src = *ppos;
 	svpfn = src / PM_ENTRY_BYTES;
-	start_vaddr = svpfn << PAGE_SHIFT;
+	start_vaddr = untagged_addr(svpfn << PAGE_SHIFT);
 	end_vaddr = mm->task_size;
 
 	/* watch out for wraparound */
-	if (svpfn > mm->task_size >> PAGE_SHIFT)
+	if (start_vaddr > mm->task_size)
 		start_vaddr = end_vaddr;
 
 	/*
Signature · 1 line
2.18.0
#1 lore raw
Quoted 9 lines · Miles Chen

When we try to visit the pagemap of a tagged userspace pointer, we find that the start_vaddr is not correct because of the tag.
To fix it, we should untag the usespace pointers in pagemap_read().

I tested with 5.10-rc4 and the issue remains.

My test code is baed on [1]:

A userspace pointer which has been tagged by 0xb4: 0xb400007662f541c8

Sigh this patch is buggy.

Quoted 57 lines

=== userspace program ===

uint64 OsLayer::VirtualToPhysical(void *vaddr) {
	uint64 frame, paddr, pfnmask, pagemask;
	int pagesize = sysconf(_SC_PAGESIZE);
	off64_t off = ((uintptr_t)vaddr) / pagesize * 8; // off = 0xb400007662f541c8 / pagesize * 8 = 0x5a00003b317aa0
	int fd = open(kPagemapPath, O_RDONLY);
	...

	if (lseek64(fd, off, SEEK_SET) != off || read(fd, &frame, 8) != 8) {
		int err = errno;
		string errtxt = ErrorString(err);
		if (fd >= 0)
			close(fd);
		return 0;
	}

...
}

=== kernel fs/proc/task_mmu.c ===

static ssize_t pagemap_read(struct file *file, char __user *buf,
size_t count, loff_t *ppos)

{
	...
	src = *ppos;
	svpfn = src / PM_ENTRY_BYTES; // svpfn == 0xb400007662f54
	start_vaddr = svpfn << PAGE_SHIFT; // start_vaddr == 0xb400007662f54000
	end_vaddr = mm->task_size;

	/* watch out for wraparound */
	// svpfn == 0xb400007662f54
	// (mm->task_size >> PAGE) == 0x8000000
	if (svpfn > mm->task_size >> PAGE_SHIFT) // the condition is true because of the tag 0xb4
		start_vaddr = end_vaddr;

	ret = 0;
	while (count && (start_vaddr < end_vaddr)) { // we cannot visit correct entry because start_vaddr is set to end_vaddr
		int len;
		unsigned long end;
		...
	}
	...
}

[1] https://github.com/stressapptest/stressapptest/blob/master/src/os.cc#L158

---
 fs/proc/task_mmu.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
fs/proc/task_mmu.c +0 -0
index 217aa2705d5d..e9a70f7ee515 100644

fs/proc/task_mmu.c static ssize_t pagemap_read(struct file *file, char __user *buf, · line 1603

Quoted 6 lines
fs/proc/task_mmu.c +1 -1
@@ -1599,11 +1599,11 @@ static ssize_t pagemap_read(struct file *file, char __user *buf,
 
 	src = *ppos;
 	svpfn = src / PM_ENTRY_BYTES;
 
-	start_vaddr = svpfn << PAGE_SHIFT;
+	start_vaddr = untagged_addr(svpfn << PAGE_SHIFT);
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

Arguably the line above is safe, but unfortunately it has the possibility of suffering from overflow.

fs/proc/task_mmu.c static ssize_t pagemap_read(struct file *file, char __user *buf, · line 1607

Quoted 6 lines
fs/proc/task_mmu.c +1 -1
 	end_vaddr = mm->task_size;
 
 	/* watch out for wraparound */
-	if (svpfn > mm->task_size >> PAGE_SHIFT)
+	if (start_vaddr > mm->task_size)
 		start_vaddr = end_vaddr;

Overflow handling you are removing here.

fs/proc/task_mmu.c static ssize_t pagemap_read(struct file *file, char __user *buf, · line 1609

Quoted 2 lines
fs/proc/task_mmu.c +0 -0
 	/*

I suspect the proper way to handle this is to move the test for overflow earlier so the code looks something like:

	end_vaddr = mm->task_size;

	src = *ppos;
	svpfn = src / PM_ENTRY_BYTES;

	/* watch out for wraparound */
        start_vaddr = end_vaddr;
	if (svpfn < (ULONG_MAX >> PAGE_SHIFT))
        	start_vaddr = untagged_addr(svpfn << PAGE_SHIFT);

	/* Ensure the address is inside the task */
	if (start_vaddr > mm->task_size)
        	start_vaddr = end_vaddr;

Eric

_______________________________________________
Linux-mediatek mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-mediatek

author
#2 lore raw
Quoted 98 lines · Eric W. Biederman
Quoted 9 lines · Miles Chen

When we try to visit the pagemap of a tagged userspace pointer, we find that the start_vaddr is not correct because of the tag.
To fix it, we should untag the usespace pointers in pagemap_read().

I tested with 5.10-rc4 and the issue remains.

My test code is baed on [1]:

A userspace pointer which has been tagged by 0xb4: 0xb400007662f541c8

Sigh this patch is buggy.

Quoted 57 lines

=== userspace program ===

uint64 OsLayer::VirtualToPhysical(void *vaddr) {
	uint64 frame, paddr, pfnmask, pagemask;
	int pagesize = sysconf(_SC_PAGESIZE);
	off64_t off = ((uintptr_t)vaddr) / pagesize * 8; // off = 0xb400007662f541c8 / pagesize * 8 = 0x5a00003b317aa0
	int fd = open(kPagemapPath, O_RDONLY);
	...

	if (lseek64(fd, off, SEEK_SET) != off || read(fd, &frame, 8) != 8) {
		int err = errno;
		string errtxt = ErrorString(err);
		if (fd >= 0)
			close(fd);
		return 0;
	}

...
}

=== kernel fs/proc/task_mmu.c ===

static ssize_t pagemap_read(struct file *file, char __user *buf,
size_t count, loff_t *ppos)

{
	...
	src = *ppos;
	svpfn = src / PM_ENTRY_BYTES; // svpfn == 0xb400007662f54
	start_vaddr = svpfn << PAGE_SHIFT; // start_vaddr == 0xb400007662f54000
	end_vaddr = mm->task_size;

	/* watch out for wraparound */
	// svpfn == 0xb400007662f54
	// (mm->task_size >> PAGE) == 0x8000000
	if (svpfn > mm->task_size >> PAGE_SHIFT) // the condition is true because of the tag 0xb4
		start_vaddr = end_vaddr;

	ret = 0;
	while (count && (start_vaddr < end_vaddr)) { // we cannot visit correct entry because start_vaddr is set to end_vaddr
		int len;
		unsigned long end;
		...
	}
	...
}

[1] https://github.com/stressapptest/stressapptest/blob/master/src/os.cc#L158

---
 fs/proc/task_mmu.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
fs/proc/task_mmu.c +0 -0
index 217aa2705d5d..e9a70f7ee515 100644

fs/proc/task_mmu.c static ssize_t pagemap_read(struct file *file, char __user *buf, · line 1603

Quoted 6 lines
fs/proc/task_mmu.c +1 -1
@@ -1599,11 +1599,11 @@ static ssize_t pagemap_read(struct file *file, char __user *buf,
 
 	src = *ppos;
 	svpfn = src / PM_ENTRY_BYTES;
 
-	start_vaddr = svpfn << PAGE_SHIFT;
+	start_vaddr = untagged_addr(svpfn << PAGE_SHIFT);
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

Arguably the line above is safe, but unfortunately it has the possibility of suffering from overflow.

fs/proc/task_mmu.c static ssize_t pagemap_read(struct file *file, char __user *buf, · line 1607

Quoted 6 lines
fs/proc/task_mmu.c +1 -1
 	end_vaddr = mm->task_size;
 
 	/* watch out for wraparound */
-	if (svpfn > mm->task_size >> PAGE_SHIFT)
+	if (start_vaddr > mm->task_size)
 		start_vaddr = end_vaddr;

Overflow handling you are removing here.

fs/proc/task_mmu.c static ssize_t pagemap_read(struct file *file, char __user *buf, · line 1609

Quoted 2 lines
fs/proc/task_mmu.c +0 -0
 	/*

I suspect the proper way to handle this is to move the test for overflow earlier so the code looks something like:

Quoted 13 lines
	end_vaddr = mm->task_size;

	src = *ppos;
	svpfn = src / PM_ENTRY_BYTES;

	/* watch out for wraparound */
        start_vaddr = end_vaddr;
	if (svpfn < (ULONG_MAX >> PAGE_SHIFT))
        	start_vaddr = untagged_addr(svpfn << PAGE_SHIFT);

	/* Ensure the address is inside the task */
	if (start_vaddr > mm->task_size)
        	start_vaddr = end_vaddr;

Thanks for the comment, I will fix that in patch v2.

Miles

Quoted 3 lines

Eric

_______________________________________________
Linux-mediatek mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-mediatek

#3 lore raw
In reply to Miles Chen
Quoted 87 lines

-----Original Message-----
From: Miles Chen [mailto:[email protected]]
Sent: Monday, November 23, 2020 7:39 PM
To: Alexey Dobriyan <[email protected]>; Andrew Morton <[email protected]>
Cc: [email protected]; [email protected]; [email protected]; [email protected]; Miles Chen <[email protected]>
Subject: [RESEND PATCH v1] proc: use untagged_addr() for pagemap_read addresses

When we try to visit the pagemap of a tagged userspace pointer, we find that the start_vaddr is not correct because of the tag.
To fix it, we should untag the usespace pointers in pagemap_read().

I tested with 5.10-rc4 and the issue remains.

My test code is baed on [1]:

A userspace pointer which has been tagged by 0xb4: 0xb400007662f541c8

=== userspace program ===

uint64 OsLayer::VirtualToPhysical(void *vaddr) {
	uint64 frame, paddr, pfnmask, pagemask;
	int pagesize = sysconf(_SC_PAGESIZE);
	off64_t off = ((uintptr_t)vaddr) / pagesize * 8; // off =
0xb400007662f541c8 / pagesize * 8 = 0x5a00003b317aa0
	int fd = open(kPagemapPath, O_RDONLY);
	...

	if (lseek64(fd, off, SEEK_SET) != off || read(fd, &frame, 8) != 8) {
		int err = errno;
		string errtxt = ErrorString(err);
		if (fd >= 0)
			close(fd);
		return 0;
	}

...
}

=== kernel fs/proc/task_mmu.c ===

static ssize_t pagemap_read(struct file *file, char __user *buf,
size_t count, loff_t *ppos)

{
	...
	src = *ppos;
	svpfn = src / PM_ENTRY_BYTES; // svpfn == 0xb400007662f54
	start_vaddr = svpfn << PAGE_SHIFT; // start_vaddr ==
0xb400007662f54000
	end_vaddr = mm->task_size;

	/* watch out for wraparound */
	// svpfn == 0xb400007662f54
	// (mm->task_size >> PAGE) == 0x8000000
	if (svpfn > mm->task_size >> PAGE_SHIFT) // the condition is true because
of the tag 0xb4
		start_vaddr = end_vaddr;

	ret = 0;
	while (count && (start_vaddr < end_vaddr)) { // we cannot visit correct
entry because start_vaddr is set to end_vaddr
		int len;
		unsigned long end;
		...
	}
	...
}

[1]
https://github.com/stressapptest/stressapptest/blob/master/src/os.cc#L158

---
 fs/proc/task_mmu.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
fs/proc/task_mmu.c +0 -0
index 217aa2705d5d..e9a70f7ee515 100644
@@ -1599,11 +1599,11 @@ static ssize_t pagemap_read(struct file *file, char
 __user *buf,
 

fs/proc/task_mmu.c static ssize_t pagemap_read(struct file *file, char · line 1608

Quoted 10 lines
fs/proc/task_mmu.c +2 -2
 	src = *ppos;
 	svpfn = src / PM_ENTRY_BYTES;
-	start_vaddr = svpfn << PAGE_SHIFT;
+	start_vaddr = untagged_addr(svpfn << PAGE_SHIFT);
 	end_vaddr = mm->task_size;
 
 	/* watch out for wraparound */
-	if (svpfn > mm->task_size >> PAGE_SHIFT)
+	if (start_vaddr > mm->task_size)
 		start_vaddr = end_vaddr;

Wouldn't the untag be done by the user reading pagemap file? With this patch, even users pass an illegal address, for example, users put a tag on a virtual address which hasn't really a tag, they will still get the right pagemap.

Quoted 4 lines
fs/proc/task_mmu.c +0 -1
 	/*
--

2.18.0

Thanks
Barry

_______________________________________________
Linux-mediatek mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-mediatek

author
#4 lore raw
Quoted 92 lines · Song Bao Hua (Barry Song)
Quoted 92 lines

-----Original Message-----
From: Miles Chen [mailto:[email protected]]
Sent: Monday, November 23, 2020 7:39 PM
To: Alexey Dobriyan <[email protected]>; Andrew Morton <[email protected]>
Cc: [email protected]; [email protected]; [email protected]; [email protected]; Miles Chen <[email protected]>
Subject: [RESEND PATCH v1] proc: use untagged_addr() for pagemap_read addresses

When we try to visit the pagemap of a tagged userspace pointer, we find that the start_vaddr is not correct because of the tag.
To fix it, we should untag the usespace pointers in pagemap_read().

I tested with 5.10-rc4 and the issue remains.

My test code is baed on [1]:

A userspace pointer which has been tagged by 0xb4: 0xb400007662f541c8

=== userspace program ===

uint64 OsLayer::VirtualToPhysical(void *vaddr) {
	uint64 frame, paddr, pfnmask, pagemask;
	int pagesize = sysconf(_SC_PAGESIZE);
	off64_t off = ((uintptr_t)vaddr) / pagesize * 8; // off =
0xb400007662f541c8 / pagesize * 8 = 0x5a00003b317aa0
	int fd = open(kPagemapPath, O_RDONLY);
	...

	if (lseek64(fd, off, SEEK_SET) != off || read(fd, &frame, 8) != 8) {
		int err = errno;
		string errtxt = ErrorString(err);
		if (fd >= 0)
			close(fd);
		return 0;
	}

...
}

=== kernel fs/proc/task_mmu.c ===

static ssize_t pagemap_read(struct file *file, char __user *buf,
size_t count, loff_t *ppos)

{
	...
	src = *ppos;
	svpfn = src / PM_ENTRY_BYTES; // svpfn == 0xb400007662f54
	start_vaddr = svpfn << PAGE_SHIFT; // start_vaddr ==
0xb400007662f54000
	end_vaddr = mm->task_size;

	/* watch out for wraparound */
	// svpfn == 0xb400007662f54
	// (mm->task_size >> PAGE) == 0x8000000
	if (svpfn > mm->task_size >> PAGE_SHIFT) // the condition is true because
of the tag 0xb4
		start_vaddr = end_vaddr;

	ret = 0;
	while (count && (start_vaddr < end_vaddr)) { // we cannot visit correct
entry because start_vaddr is set to end_vaddr
		int len;
		unsigned long end;
		...
	}
	...
}

[1]
https://github.com/stressapptest/stressapptest/blob/master/src/os.cc#L158

---
 fs/proc/task_mmu.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
fs/proc/task_mmu.c +1 -1
index 217aa2705d5d..e9a70f7ee515 100644
@@ -1599,11 +1599,11 @@ static ssize_t pagemap_read(struct file *file, char
 __user *buf,
 
 	src = *ppos;
 	svpfn = src / PM_ENTRY_BYTES;
-	start_vaddr = svpfn << PAGE_SHIFT;
+	start_vaddr = untagged_addr(svpfn << PAGE_SHIFT);
 	end_vaddr = mm->task_size;
 
Quoted 10 lines

fs/proc/task_mmu.c static ssize_t pagemap_read(struct file *file, char __user *buf, · line 1607

Quoted 4 lines
fs/proc/task_mmu.c +1 -1
 	/* watch out for wraparound */
-	if (svpfn > mm->task_size >> PAGE_SHIFT)
+	if (start_vaddr > mm->task_size)
 		start_vaddr = end_vaddr;

Wouldn't the untag be done by the user reading pagemap file? With this patch, even users pass an illegal address, for example, users put a tag on a virtual address which hasn't really a tag, they will still get the right pagemap.

I run arm64 devices.
If the tagged pointer is enabled, the ARM top-byte Ignore is also enabled. It means the top-byte is always be ignored.

I think the kernel should not break existing userspace program, so it's better to handle the tagged user pointer in kernel.

grep 'untag' mm -RnH to see existing examples.

thanks,
Miles

Quoted 8 lines
Quoted 4 lines
fs/proc/task_mmu.c +0 -1
 	/*
--

2.18.0

Thanks
Barry

#5 lore raw
Quoted 120 lines

-----Original Message-----
From: Miles Chen [mailto:[email protected]]
Sent: Thursday, November 26, 2020 8:49 PM
To: Song Bao Hua (Barry Song) <[email protected]>
Cc: Alexey Dobriyan <[email protected]>; Andrew Morton <[email protected]>; [email protected]; [email protected]; [email protected]; [email protected]
Subject: RE: [RESEND PATCH v1] proc: use untagged_addr() for pagemap_read addresses

Quoted 71 lines · Song Bao Hua (Barry Song)
Quoted 71 lines

-----Original Message-----
From: Miles Chen [mailto:[email protected]]
Sent: Monday, November 23, 2020 7:39 PM
To: Alexey Dobriyan <[email protected]>; Andrew Morton <[email protected]>
Cc: [email protected]; [email protected]; [email protected]; [email protected]; Miles Chen <[email protected]>
Subject: [RESEND PATCH v1] proc: use untagged_addr() for pagemap_read addresses

When we try to visit the pagemap of a tagged userspace pointer, we find that the start_vaddr is not correct because of the tag.
To fix it, we should untag the usespace pointers in pagemap_read().

I tested with 5.10-rc4 and the issue remains.

My test code is baed on [1]:

A userspace pointer which has been tagged by 0xb4: 0xb400007662f541c8

=== userspace program ===

uint64 OsLayer::VirtualToPhysical(void *vaddr) {
	uint64 frame, paddr, pfnmask, pagemask;
	int pagesize = sysconf(_SC_PAGESIZE);
	off64_t off = ((uintptr_t)vaddr) / pagesize * 8; // off =
0xb400007662f541c8 / pagesize * 8 = 0x5a00003b317aa0
	int fd = open(kPagemapPath, O_RDONLY);
	...

	if (lseek64(fd, off, SEEK_SET) != off || read(fd, &frame, 8) != 8) {
		int err = errno;
		string errtxt = ErrorString(err);
		if (fd >= 0)
			close(fd);
		return 0;
	}

...
}

=== kernel fs/proc/task_mmu.c ===

static ssize_t pagemap_read(struct file *file, char __user *buf,
size_t count, loff_t *ppos)

{
	...
	src = *ppos;
	svpfn = src / PM_ENTRY_BYTES; // svpfn == 0xb400007662f54
	start_vaddr = svpfn << PAGE_SHIFT; // start_vaddr ==
0xb400007662f54000
	end_vaddr = mm->task_size;

	/* watch out for wraparound */
	// svpfn == 0xb400007662f54
	// (mm->task_size >> PAGE) == 0x8000000
	if (svpfn > mm->task_size >> PAGE_SHIFT) // the condition is true because
of the tag 0xb4
		start_vaddr = end_vaddr;

	ret = 0;
	while (count && (start_vaddr < end_vaddr)) { // we cannot visit correct
entry because start_vaddr is set to end_vaddr
		int len;
		unsigned long end;
		...
	}
	...
}
Quoted 2 lines
Quoted 2 lines

[1]

https://github.com/stressapptest/stressapptest/blob/master/src/os.cc#L158

Quoted 11 lines

fs/proc/task_mmu.c static ssize_t pagemap_read(struct file *file,

Quoted 11 lines
---
 fs/proc/task_mmu.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
fs/proc/task_mmu.c +0 -0
index 217aa2705d5d..e9a70f7ee515 100644
@@ -1599,11 +1599,11 @@ static ssize_t pagemap_read(struct file *file,

char

Quoted 8 lines
Quoted 8 lines
fs/proc/task_mmu.c +1 -1
 __user *buf,
 
 	src = *ppos;
 	svpfn = src / PM_ENTRY_BYTES;
-	start_vaddr = svpfn << PAGE_SHIFT;
+	start_vaddr = untagged_addr(svpfn << PAGE_SHIFT);
 	end_vaddr = mm->task_size;
 
Quoted 10 lines

fs/proc/task_mmu.c static ssize_t pagemap_read(struct file *file, char __user *buf, · line 1607

Quoted 4 lines
fs/proc/task_mmu.c +1 -1
 	/* watch out for wraparound */
-	if (svpfn > mm->task_size >> PAGE_SHIFT)
+	if (start_vaddr > mm->task_size)
 		start_vaddr = end_vaddr;

Wouldn't the untag be done by the user reading pagemap file? With this patch, even users pass an illegal address, for example, users put a tag on a virtual address which hasn't really a tag, they will still get the right pagemap.

Quoted 8 lines

I run arm64 devices.
If the tagged pointer is enabled, the ARM top-byte Ignore is also enabled. It means the top-byte is always be ignored.

I think the kernel should not break existing userspace program, so it's better to handle the tagged user pointer in kernel.

grep 'untag' mm -RnH to see existing examples.

I see your point. But the existing examples such as gup etc are receiving an address from userspace.

pagemap isn't getting an address, it is getting a file offset which should be figured out by userspace correctly. While we read a file, we should make sure the offset is in the range of the file length.

Quoted 10 lines

thanks,
Miles

Quoted 5 lines
Quoted 4 lines
fs/proc/task_mmu.c +0 -1
 	/*
--

2.18.0

Thanks
Barry
_______________________________________________
Linux-mediatek mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-mediatek

#6 lore raw
In reply to Miles Chen
Quoted 144 lines

-----Original Message-----
From: Song Bao Hua (Barry Song)
Sent: Thursday, November 26, 2020 8:53 PM
To: 'Miles Chen' <[email protected]>
Cc: Alexey Dobriyan <[email protected]>; Andrew Morton <[email protected]>; [email protected]; [email protected]; [email protected]; [email protected]
Subject: RE: [RESEND PATCH v1] proc: use untagged_addr() for pagemap_read addresses

Quoted 9 lines

-----Original Message-----
From: Miles Chen [mailto:[email protected]]
Sent: Thursday, November 26, 2020 8:49 PM
To: Song Bao Hua (Barry Song) <[email protected]>
Cc: Alexey Dobriyan <[email protected]>; Andrew Morton <[email protected]>; [email protected]; [email protected]; [email protected]; [email protected]
Subject: RE: [RESEND PATCH v1] proc: use untagged_addr() for

pagemap_read

Quoted 24 lines

addresses

Quoted 21 lines · Song Bao Hua (Barry Song)
Quoted 21 lines

-----Original Message-----
From: Miles Chen [mailto:[email protected]]
Sent: Monday, November 23, 2020 7:39 PM
To: Alexey Dobriyan <[email protected]>; Andrew Morton <[email protected]>
Cc: [email protected]; [email protected]; [email protected]; [email protected]; Miles Chen <[email protected]>
Subject: [RESEND PATCH v1] proc: use untagged_addr() for pagemap_read addresses

When we try to visit the pagemap of a tagged userspace pointer, we find that the start_vaddr is not correct because of the tag.
To fix it, we should untag the usespace pointers in pagemap_read().

I tested with 5.10-rc4 and the issue remains.

My test code is baed on [1]:

A userspace pointer which has been tagged by 0xb4:

0xb400007662f541c8

Quoted 33 lines
Quoted 33 lines
Quoted 33 lines

=== userspace program ===

uint64 OsLayer::VirtualToPhysical(void *vaddr) {
	uint64 frame, paddr, pfnmask, pagemask;
	int pagesize = sysconf(_SC_PAGESIZE);
	off64_t off = ((uintptr_t)vaddr) / pagesize * 8; // off =
0xb400007662f541c8 / pagesize * 8 = 0x5a00003b317aa0
	int fd = open(kPagemapPath, O_RDONLY);
	...

	if (lseek64(fd, off, SEEK_SET) != off || read(fd, &frame, 8) != 8) {
		int err = errno;
		string errtxt = ErrorString(err);
		if (fd >= 0)
			close(fd);
		return 0;
	}

...
}

=== kernel fs/proc/task_mmu.c ===

static ssize_t pagemap_read(struct file *file, char __user *buf,
size_t count, loff_t *ppos)

{
	...
	src = *ppos;
	svpfn = src / PM_ENTRY_BYTES; // svpfn == 0xb400007662f54
	start_vaddr = svpfn << PAGE_SHIFT; // start_vaddr ==
0xb400007662f54000
	end_vaddr = mm->task_size;
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
	/* watch out for wraparound */
	// svpfn == 0xb400007662f54
	// (mm->task_size >> PAGE) == 0x8000000
	if (svpfn > mm->task_size >> PAGE_SHIFT) // the condition is true

because

Quoted 5 lines
Quoted 5 lines
Quoted 5 lines

of the tag 0xb4

		start_vaddr = end_vaddr;

	ret = 0;
	while (count && (start_vaddr < end_vaddr)) { // we cannot visit

correct

Quoted 50 lines
Quoted 10 lines
Quoted 10 lines

entry because start_vaddr is set to end_vaddr

		int len;
		unsigned long end;
		...
	}
	...
}

[1]

https://github.com/stressapptest/stressapptest/blob/master/src/os.cc#L158

Quoted 11 lines

fs/proc/task_mmu.c static ssize_t pagemap_read(struct file *file,

Quoted 11 lines
---
 fs/proc/task_mmu.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
fs/proc/task_mmu.c +0 -0
index 217aa2705d5d..e9a70f7ee515 100644
@@ -1599,11 +1599,11 @@ static ssize_t pagemap_read(struct file *file,

char

Quoted 8 lines
Quoted 8 lines
fs/proc/task_mmu.c +1 -1
 __user *buf,
 
 	src = *ppos;
 	svpfn = src / PM_ENTRY_BYTES;
-	start_vaddr = svpfn << PAGE_SHIFT;
+	start_vaddr = untagged_addr(svpfn << PAGE_SHIFT);
 	end_vaddr = mm->task_size;
 
Quoted 10 lines

fs/proc/task_mmu.c static ssize_t pagemap_read(struct file *file, char __user *buf, · line 1607

Quoted 4 lines
fs/proc/task_mmu.c +1 -1
 	/* watch out for wraparound */
-	if (svpfn > mm->task_size >> PAGE_SHIFT)
+	if (start_vaddr > mm->task_size)
 		start_vaddr = end_vaddr;

Wouldn't the untag be done by the user reading pagemap file? With this patch, even users pass an illegal address, for example, users put a tag on a virtual address which hasn't really a tag, they will still get the right pagemap.

I run arm64 devices.
If the tagged pointer is enabled, the ARM top-byte Ignore is also enabled. It means the top-byte is always be ignored.

I think the kernel should not break existing userspace program, so it's better to handle the tagged user pointer in kernel.

Quoted 8 lines
Quoted 1 line

grep 'untag' mm -RnH to see existing examples.

I see your point. But the existing examples such as gup etc are receiving an address from userspace.

pagemap isn't getting an address, it is getting a file offset which should be figured out by userspace correctly. While we read a file, we should make sure the offset is in the range of the file length.

BTW, personally I don't object to this patch. I just feel a little bit weird as the parameter in pagemap_read is an offset in one file rather than a virtual address.

if there are many user programs which are written without the awareness of tag while reading pagemap or if other people like this patch, it might make sense to have it in kernel :-)

Quoted 12 lines
Quoted 11 lines

thanks,
Miles

Quoted 5 lines
Quoted 4 lines
fs/proc/task_mmu.c +0 -1
 	/*
--

2.18.0

Thanks
Barry

#7 lore raw
In reply to Miles Chen

Hi Miles,

Could you please cc me and Andrey Konovalov on future versions of this patch (if any)?

Quoted 8 lines · Miles Chen

When we try to visit the pagemap of a tagged userspace pointer, we find that the start_vaddr is not correct because of the tag.
To fix it, we should untag the usespace pointers in pagemap_read().

I tested with 5.10-rc4 and the issue remains.

My test code is baed on [1]:

Quoted 8 lines

A userspace pointer which has been tagged by 0xb4: 0xb400007662f541c8

=== userspace program ===

uint64 OsLayer::VirtualToPhysical(void *vaddr) {
        uint64 frame, paddr, pfnmask, pagemask;
        int pagesize = sysconf(_SC_PAGESIZE);
        off64_t off = ((uintptr_t)vaddr) / pagesize * 8; // off = 0xb400007662f541c8 / pagesize * 8 = 0x5a00003b317aa0

Arguably, that's a user-space bug since tagged file offsets were never supported. In this case it's not even a tag at bit 56 as per the arm64 tagged address ABI but rather down to bit 47. You could say that the problem is caused by the C library (malloc()) or whoever created the tagged vaddr and passed it to this function. It's not a kernel regression as we've never supported it.

Now, pagemap is a special case where the offset is usually not generated as a classic file offset but rather derived by shifting a user virtual address. I guess we can make a concession for pagemap (only) and allow such offset with the tag at bit (56 - PAGE_SHIFT + 3).

Please fix the patch as per Eric's suggestion on avoiding the overflow. You should also add a Cc: stable v5.4- as that's when we enabled the tagged address ABI on arm64 and when it's more likely for the C library/malloc() to start generating such pointers.

If the problem is only limited to this test, I'd rather fix the user but I can't tell how widespread the /proc/pid/pagemap usage is.

Thanks.

Signature · 1 line
Catalin
author
#8 lore raw
Quoted 23 lines · Catalin Marinas

Hi Miles,

Could you please cc me and Andrey Konovalov on future versions of this patch (if any)?

Quoted 8 lines · Miles Chen

When we try to visit the pagemap of a tagged userspace pointer, we find that the start_vaddr is not correct because of the tag.
To fix it, we should untag the usespace pointers in pagemap_read().

I tested with 5.10-rc4 and the issue remains.

My test code is baed on [1]:

Quoted 8 lines

A userspace pointer which has been tagged by 0xb4: 0xb400007662f541c8

=== userspace program ===

uint64 OsLayer::VirtualToPhysical(void *vaddr) {
        uint64 frame, paddr, pfnmask, pagemask;
        int pagesize = sysconf(_SC_PAGESIZE);
        off64_t off = ((uintptr_t)vaddr) / pagesize * 8; // off = 0xb400007662f541c8 / pagesize * 8 = 0x5a00003b317aa0
Quoted 6 lines

Arguably, that's a user-space bug since tagged file offsets were never supported. In this case it's not even a tag at bit 56 as per the arm64 tagged address ABI but rather down to bit 47. You could say that the problem is caused by the C library (malloc()) or whoever created the tagged vaddr and passed it to this function. It's not a kernel regression as we've never supported it.

thanks for the explaination.

Quoted 11 lines

Now, pagemap is a special case where the offset is usually not generated as a classic file offset but rather derived by shifting a user virtual address. I guess we can make a concession for pagemap (only) and allow such offset with the tag at bit (56 - PAGE_SHIFT + 3).

Please fix the patch as per Eric's suggestion on avoiding the overflow. You should also add a Cc: stable v5.4- as that's when we enabled the tagged address ABI on arm64 and when it's more likely for the C library/malloc() to start generating such pointers.

Got it, thanks for your reviewing and suggestion. I will follow Eric's suggestion and submit patch v2 and cc stable v5.4-

Miles

Quoted 6 lines

If the problem is only limited to this test, I'd rather fix the user but I can't tell how widespread the /proc/pid/pagemap usage is.

Thanks.