[PATCH] arch/cris: use get_user_pages_fast()

[PATCH] arch/cris: use get_user_pages_fast()

[Davidlohr Bueso]

Since we currently hold mmap_sem across both gup calls (and
nothing more), we can substitute it with two _fast()
alternatives and possibly avoid grabbing the lock.

This was found while adding mmap_sem wrappers, and was also
previously reported by Al: https://lkml.org/lkml/2017/11/17/777

Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
---
 arch/cris/arch-v32/drivers/cryptocop.c | 29 ++++++++++-------------------
 1 file changed, 10 insertions(+), 19 deletions(-)

diff --git a/arch/cris/arch-v32/drivers/cryptocop.c b/arch/cris/arch-v32/drivers/cryptocop.c
index d688fe117dca..76f8d3b1d39e 100644
--- a/arch/cris/arch-v32/drivers/cryptocop.c
+++ b/arch/cris/arch-v32/drivers/cryptocop.c
@@ -2717,37 +2717,28 @@ static int cryptocop_ioctl_process(struct inode *inode, struct file *filp, unsig
 		}
 	}
 
-	/* Acquire the mm page semaphore. */
-	down_read(&current->mm->mmap_sem);
-
-	err = get_user_pages((unsigned long int)(oper.indata + prev_ix),
-			     noinpages,
-			     0,  /* read access only for in data */
-			     inpages,
-			     NULL);
+	err = get_user_pages_fast((unsigned long int)(oper.indata + prev_ix),
+				  noinpages,
+				  0,  /* read access only for in data */
+				  inpages);
 
 	if (err < 0) {
-		up_read(&current->mm->mmap_sem);
 		nooutpages = noinpages = 0;
-		DEBUG_API(printk("cryptocop_ioctl_process: get_user_pages indata\n"));
+		DEBUG_API(printk("cryptocop_ioctl_process: get_user_pages_fast indata\n"));
 		goto error_cleanup;
 	}
 	noinpages = err;
 	if (oper.do_cipher){
-		err = get_user_pages((unsigned long int)oper.cipher_outdata,
-				     nooutpages,
-				     FOLL_WRITE, /* write access for out data */
-				     outpages,
-				     NULL);
-		up_read(&current->mm->mmap_sem);
+		err = get_user_pages_fast((unsigned long int)oper.cipher_outdata,
+					  nooutpages,
+					  FOLL_WRITE, /* write access for out data */
+					  outpages);
 		if (err < 0) {
 			nooutpages = 0;
-			DEBUG_API(printk("cryptocop_ioctl_process: get_user_pages outdata\n"));
+			DEBUG_API(printk("cryptocop_ioctl_process: get_user_pages_fast outdata\n"));
 			goto error_cleanup;
 		}
 		nooutpages = err;
-	} else {
-		up_read(&current->mm->mmap_sem);
 	}
 
 	/* Add 6 to nooutpages to make room for possibly inserted buffers for storing digest and
-- 
2.13.6

Re: [PATCH] arch/cris: use get_user_pages_fast()

[Al Viro]

On Sun, Jan 21, 2018 at 02:59:29PM -0800, Davidlohr Bueso wrote:
> Since we currently hold mmap_sem across both gup calls (and
> nothing more), we can substitute it with two _fast()
> alternatives and possibly avoid grabbing the lock.
> 
> This was found while adding mmap_sem wrappers, and was also
> previously reported by Al: https://lkml.org/lkml/2017/11/17/777

See commit 9a949e8ff92246c0753b2805c2a001cb991fffe5
Author: Al Viro <viro@zeniv.linux.org.uk>
Date:   Sat Nov 18 14:37:46 2017 -0500

    cris: switch to get_user_pages_fast()
    
    no point holding ->mmap_sem over both calls.
    
    Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>


in -next...

Re: [PATCH] arch/cris: use get_user_pages_fast()

[Davidlohr Bueso]

On Sun, 21 Jan 2018, Al Viro wrote:

>On Sun, Jan 21, 2018 at 02:59:29PM -0800, Davidlohr Bueso wrote:
>> Since we currently hold mmap_sem across both gup calls (and
>> nothing more), we can substitute it with two _fast()
>> alternatives and possibly avoid grabbing the lock.
>>
>> This was found while adding mmap_sem wrappers, and was also
>> previously reported by Al: https://lkml.org/lkml/2017/11/17/777
>
>See commit 9a949e8ff92246c0753b2805c2a001cb991fffe5
>Author: Al Viro <viro@zeniv.linux.org.uk>
>Date:   Sat Nov 18 14:37:46 2017 -0500
>
>    cris: switch to get_user_pages_fast()
>
>    no point holding ->mmap_sem over both calls.
>
>    Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

Serves me for developing on Linus' tree, cool.

Now, regarding the other two which are technically bugs (not
holding mmap_sem under gup), you mention I don't see those in
next, so I'm assuming you didn't get to them.

- mthca_map_user_db(): I came to the same conclusion and tempted
  to send out the same idea of unlocking the mutex for gup_fast()
  so we can take mmap_sem if needed.

- ia64's store_virtual_to_phys() is a sysfs attribute and
  holds no (other) locks. How about we just convert it as well?

diff --git a/arch/ia64/kernel/err_inject.c b/arch/ia64/kernel/err_inject.c
index 85bba43e7d5d..658a8e06a69b 100644
--- a/arch/ia64/kernel/err_inject.c
+++ b/arch/ia64/kernel/err_inject.c
@@ -142,7 +142,7 @@ store_virtual_to_phys(struct device *dev, struct device_attribute *attr,
 	u64 virt_addr=simple_strtoull(buf, NULL, 16);
 	int ret;
 
-	ret = get_user_pages(virt_addr, 1, FOLL_WRITE, NULL, NULL);
+	ret = get_user_pages_fast(virt_addr, 1, FOLL_WRITE, NULL);
 	if (ret<=0) {
 #ifdef ERR_INJ_DEBUG
 		printk("Virtual address %lx is not existing.\n",virt_addr);


Thanks,
Davidlohr