Skip to contentSkip to conversation

riscv: fix locking violation in page fault handler

PATCH In review Andreas Schwab 3 participants 8 messages May 7, 2019 – May 29, 2019 RISC-V

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 riscv: fix locking violation in page fault handler

When a user mode process accesses an address in the vmalloc area do_page_fault tries to unlock the mmap semaphore when it isn't locked.

---
 arch/riscv/mm/fault.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
arch/riscv/mm/fault.c +2 -1
index 88401d5125bc..c51878e5a66a 100644
@@ -181,6 +181,7 @@ asmlinkage void do_page_fault(struct pt_regs *regs)
 	up_read(&mm->mmap_sem);
 	/* User mode accesses just cause a SIGSEGV */
 	if (user_mode(regs)) {
+bad_area_do_trap:
 		do_trap(regs, SIGSEGV, code, addr, tsk);
 		return;
 	}
@@ -230,7 +231,7 @@ asmlinkage void do_page_fault(struct pt_regs *regs)
 		int index;
 
 		if (user_mode(regs))
-			goto bad_area;
+			goto bad_area_do_trap;
 
 		/*
 		 * Synchronize this task's top level page-table
Signature · 7 lines
2.21.0


-- 
Andreas Schwab, SUSE Labs, [email protected]
GPG Key fingerprint = 0196 BAD8 1CE9 1970 F4BE  1748 E4D4 88E3 0EEA B9D7
"And now for something completely different."
#1 lore raw
Quoted 22 lines · Andreas Schwab

When a user mode process accesses an address in the vmalloc area do_page_fault tries to unlock the mmap semaphore when it isn't locked.

---
 arch/riscv/mm/fault.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
arch/riscv/mm/fault.c +1 -0
index 88401d5125bc..c51878e5a66a 100644
@@ -181,6 +181,7 @@ asmlinkage void do_page_fault(struct pt_regs *regs)
 	up_read(&mm->mmap_sem);
 	/* User mode accesses just cause a SIGSEGV */
 	if (user_mode(regs)) {
+bad_area_do_trap:
 		do_trap(regs, SIGSEGV, code, addr, tsk);
 		return;
 	}

arch/riscv/mm/fault.c asmlinkage void do_page_fault(struct pt_regs *regs) · line 237

Quoted 8 lines
arch/riscv/mm/fault.c +1 -1
@@ -230,7 +231,7 @@ asmlinkage void do_page_fault(struct pt_regs *regs)
 		int index;
 
 		if (user_mode(regs))
-			goto bad_area;
+			goto bad_area_do_trap;
 
 		/*
 		 * Synchronize this task's top level page-table

In this case I think it will be a lot cleaner if you just duplicated the do_trap call. On a slightly different note - is there any reason why do_page_fault is such a spaghetti mess? At the very least the code under no_context label could go into it's own function since it just kills the process and never returns? Furthermore the whole vmalloc_fault just cries for being factored out in a function, it's explicitly in it's own
block.

#2 lore raw
Quoted 3 lines · Nikolay Borisov

At the very least the code under
no_context label could go into it's own function since it just kills the process and never returns?

This is not true.

Andreas.

Signature · 3 lines
Andreas Schwab, SUSE Labs, [email protected]
GPG Key fingerprint = 0196 BAD8 1CE9 1970 F4BE  1748 E4D4 88E3 0EEA B9D7
"And now for something completely different."
#3 lore raw
Quoted 7 lines · Andreas Schwab
Quoted 3 lines · Nikolay Borisov

At the very least the code under
no_context label could go into it's own function since it just kills the process and never returns?

This is not true.

Be more specific, according to do_task_dead after the last __schedule is called the calling context is no more?

Quoted 3 lines

Andreas.

#4 lore raw
Quoted 11 lines · Nikolay Borisov
Quoted 7 lines · Andreas Schwab
Quoted 3 lines · Nikolay Borisov

At the very least the code under
no_context label could go into it's own function since it just kills the process and never returns?

This is not true.

Be more specific, according to do_task_dead after the last __schedule is called the calling context is no more?

	/* Are we prepared to handle this kernel fault? */
	if (fixup_exception(regs))
		return;

Andreas.

Signature · 3 lines
Andreas Schwab, SUSE Labs, [email protected]
GPG Key fingerprint = 0196 BAD8 1CE9 1970 F4BE  1748 E4D4 88E3 0EEA B9D7
"And now for something completely different."
Reviewed-by on v1 1/1
#5 lore raw
In reply to Andreas Schwab
Quoted 21 lines · On Tue, 07 May 2019 00:36:46 PDT (-0700), [email protected] wrote:

When a user mode process accesses an address in the vmalloc area do_page_fault tries to unlock the mmap semaphore when it isn't locked.

---
 arch/riscv/mm/fault.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
arch/riscv/mm/fault.c +1 -0
index 88401d5125bc..c51878e5a66a 100644
@@ -181,6 +181,7 @@ asmlinkage void do_page_fault(struct pt_regs *regs)
 	up_read(&mm->mmap_sem);
 	/* User mode accesses just cause a SIGSEGV */
 	if (user_mode(regs)) {
+bad_area_do_trap:
 		do_trap(regs, SIGSEGV, code, addr, tsk);
 		return;
 	}

arch/riscv/mm/fault.c asmlinkage void do_page_fault(struct pt_regs *regs) · line 237

Quoted 8 lines
arch/riscv/mm/fault.c +1 -1
@@ -230,7 +231,7 @@ asmlinkage void do_page_fault(struct pt_regs *regs)
 		int index;
 
 		if (user_mode(regs))
-			goto bad_area;
+			goto bad_area_do_trap;
 
 		/*
 		 * Synchronize this task's top level page-table

I got lost with all the gotos, I think something like this is cleaner

   diff --git a/arch/riscv/mm/fault.c b/arch/riscv/mm/fault.c
   index 26293bc053a8..cec8be9e2d6a 100644
   --- a/arch/riscv/mm/fault.c
   +++ b/arch/riscv/mm/fault.c
   @@ -229,8 +229,9 @@ asmlinkage void do_page_fault(struct pt_regs *regs)
                   pte_t *pte_k;
                   int index;

+ /* User mode accesses just cause a SIGSEGV */
if (user_mode(regs))
- goto bad_area;
+ return do_trap(regs, SIGSEGV, code, addr, tsk);

/*
* Synchronize this task's top level page-table

Unless anyone has a better idea?

Either way:

LMK if you, or anyone else, has a preference. I'm assuming this will go in through my tree, so I've picked up my version for now :)

#6 lore raw
Quoted 2 lines · Palmer Dabbelt

LMK if you, or anyone else, has a preference. I'm assuming this will go in through my tree, so I've picked up my version for now :)

You did?

Andreas.

Signature · 3 lines
Andreas Schwab, SUSE Labs, [email protected]
GPG Key fingerprint = 0196 BAD8 1CE9 1970 F4BE  1748 E4D4 88E3 0EEA B9D7
"And now for something completely different."
#7 lore raw
Quoted 6 lines · On Thu, 16 May 2019 00:42:01 PDT (-0700), [email protected] wrote:
Quoted 2 lines · Palmer Dabbelt

LMK if you, or anyone else, has a preference. I'm assuming this will go in through my tree, so I've picked up my version for now :)

You did?

It ended up landing in Linus' tree as 8fef9900d43f ("riscv: fix locking violation in page fault handler"), so it looks like I did manage avoid losing this one.

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