Skip to contentSkip to conversation

NOMMU: Fix MAP_PRIVATE mmap() of objects where the data can be mapped directly

PATCH In review David Howells 5 participants 10 messages Sep 24, 2009 – Sep 25, 2009

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 NOMMU: Fix MAP_PRIVATE mmap() of objects where the data can be mapped directly

Fix MAP_PRIVATE mmap() of files and devices where the data in the backing store might be mapped directly. Use the BDI_CAP_MAP_DIRECT capability flag to govern whether or not we should be trying to map a file directly. This can be used to determine whether or not a region has been filled in at the point where we call do_mmap_shared() or do_mmap_private().

The BDI_CAP_MAP_DIRECT capability flag is cleared by validate_mmap_request() if there's any reason we can't use it. It's also cleared in do_mmap_pgoff() if f_op->get_unmapped_area() fails.

Without this fix, attempting to run a program from a RomFS image on a non-mappable MTD partition results in a BUG as the kernel attempts XIP, and this can be caught in gdb:

Program received signal SIGABRT, Aborted.
0xc005dce8 in add_nommu_region (region=<value optimized out>) at mm/nommu.c:547 (gdb) bt

#0  0xc005dce8 in add_nommu_region (region=<value optimized out>) at mm/nommu.c:547
#1  0xc005f168 in do_mmap_pgoff (file=0xc31a6620, addr=<value optimized out>, len=3808, prot=3, flags=6146, pgoff=0) at mm/nommu.c:1373
#2  0xc00a96b8 in elf_fdpic_map_file (params=0xc33fbbec, file=0xc31a6620, mm=0xc31bef60, what=0xc0213144 "executable") at mm.h:1145
#3  0xc00aa8b4 in load_elf_fdpic_binary (bprm=0xc316cb00, regs=<value optimized out>) at fs/binfmt_elf_fdpic.c:343
#4  0xc006b588 in search_binary_handler (bprm=0x6, regs=0xc33fbce0) at fs/exec.c:1234
#5  0xc006c648 in do_execve (filename=<value optimized out>, argv=0xc3ad14cc, envp=0xc3ad1460, regs=0xc33fbce0) at fs/exec.c:1356
#6  0xc0008cf0 in sys_execve (name=<value optimized out>, argv=0xc3ad14cc, envp=0xc3ad1460) at arch/frv/kernel/process.c:263
#7  0xc00075dc in __syscall_call () at arch/frv/kernel/entry.S:897

Note that this fix does the following commit differently:

	commit a190887b58c32d19c2eee007c5eb8faa970a69ba
	Author: David Howells <[email protected]>
	Date:   Sat Sep 5 11:17:07 2009 -0700
	nommu: fix error handling in do_mmap_pgoff()
Cc Pekka Enberg <[email protected]>
Cc Paul Mundt <[email protected]>
Cc Mel Gorman <[email protected]>
Cc Greg Ungerer <[email protected]>

---

 mm/nommu.c |   34 ++++++++++++----------------------
 1 files changed, 12 insertions(+), 22 deletions(-)
mm/nommu.c +12 -22
index c459aec..cc24d9f 100644
@@ -1074,7 +1074,7 @@ static int do_mmap_shared_file(struct vm_area_struct *vma)
 	ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
 	if (ret == 0) {
 		vma->vm_region->vm_top = vma->vm_region->vm_end;
-		return ret;
+		return 0;
 	}
 	if (ret != -ENOSYS)
 		return ret;
@@ -1091,7 +1091,8 @@ static int do_mmap_shared_file(struct vm_area_struct *vma)
  */
 static int do_mmap_private(struct vm_area_struct *vma,
 			   struct vm_region *region,
-			   unsigned long len)
+			   unsigned long len,
+			   unsigned long capabilities)
 {
 	struct page *pages;
 	unsigned long total, point, n, rlen;
@@ -1102,13 +1103,13 @@ static int do_mmap_private(struct vm_area_struct *vma,
 	 * shared mappings on devices or memory
 	 * - VM_MAYSHARE will be set if it may attempt to share
 	 */
-	if (vma->vm_file) {
+	if (capabilities & BDI_CAP_MAP_DIRECT) {
 		ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
 		if (ret == 0) {
 			/* shouldn't return success if we're not sharing */
 			BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
 			vma->vm_region->vm_top = vma->vm_region->vm_end;
-			return ret;
+			return 0;
 		}
 		if (ret != -ENOSYS)
 			return ret;
@@ -1346,7 +1347,7 @@ unsigned long do_mmap_pgoff(struct file *file,
 		 * - this is the hook for quasi-memory character devices to
 		 *   tell us the location of a shared mapping
 		 */
-		if (file && file->f_op->get_unmapped_area) {
+		if (capabilities & BDI_CAP_MAP_DIRECT) {
 			addr = file->f_op->get_unmapped_area(file, addr, len,
 							     pgoff, flags);
 			if (IS_ERR((void *) addr)) {
@@ -1370,15 +1371,17 @@ unsigned long do_mmap_pgoff(struct file *file,
 	}
 
 	vma->vm_region = region;
-	add_nommu_region(region);
 
-	/* set up the mapping */
+	/* set up the mapping
+	 * - the region is filled in if BDI_CAP_MAP_DIRECT is still set
+	 */
 	if (file && vma->vm_flags & VM_SHARED)
 		ret = do_mmap_shared_file(vma);
 	else
-		ret = do_mmap_private(vma, region, len);
+		ret = do_mmap_private(vma, region, len, capabilities);
 	if (ret < 0)
-		goto error_put_region;
+		goto error_just_free;
+	add_nommu_region(region);
 
 	/* okay... we have a mapping; now we have to register it */
 	result = vma->vm_start;
@@ -1396,19 +1399,6 @@ share:
 	kleave(" = %lx", result);
 	return result;
 
-error_put_region:
-	__put_nommu_region(region);
-	if (vma) {
-		if (vma->vm_file) {
-			fput(vma->vm_file);
-			if (vma->vm_flags & VM_EXECUTABLE)
-				removed_exe_file_vma(vma->vm_mm);
-		}
-		kmem_cache_free(vm_area_cachep, vma);
-	}
-	kleave(" = %d [pr]", ret);
-	return ret;
-
 error_just_free:
 	up_write(&nommu_region_sem);
 error:
Acked-by on v1 1/1
#1 lore raw

Hi David,

Quoted 15 lines · David Howells

Fix MAP_PRIVATE mmap() of files and devices where the data in the backing store might be mapped directly.  Use the BDI_CAP_MAP_DIRECT capability flag to govern whether or not we should be trying to map a file directly.  This can be used to determine whether or not a region has been filled in at the point where we call do_mmap_shared() or do_mmap_private().

The BDI_CAP_MAP_DIRECT capability flag is cleared by validate_mmap_request() if there's any reason we can't use it.  It's also cleared in do_mmap_pgoff() if f_op->get_unmapped_area() fails.

Without this fix, attempting to run a program from a RomFS image on a non-mappable MTD partition results in a BUG as the kernel attempts XIP, and this can be caught in gdb:

Quoted 11 lines

Program received signal SIGABRT, Aborted.
0xc005dce8 in add_nommu_region (region=<value optimized out>) at mm/nommu.c:547
(gdb) bt
#0  0xc005dce8 in add_nommu_region (region=<value optimized out>) at mm/nommu.c:547
#1  0xc005f168 in do_mmap_pgoff (file=0xc31a6620, addr=<value optimized out>, len=3808, prot=3, flags=6146, pgoff=0) at mm/nommu.c:1373 #2  0xc00a96b8 in elf_fdpic_map_file (params=0xc33fbbec, file=0xc31a6620, mm=0xc31bef60, what=0xc0213144 "executable") at mm.h:1145 #3  0xc00aa8b4 in load_elf_fdpic_binary (bprm=0xc316cb00, regs=<value optimized out>) at fs/binfmt_elf_fdpic.c:343 #4  0xc006b588 in search_binary_handler (bprm=0x6, regs=0xc33fbce0) at fs/exec.c:1234
#5  0xc006c648 in do_execve (filename=<value optimized out>, argv=0xc3ad14cc, envp=0xc3ad1460, regs=0xc33fbce0) at fs/exec.c:1356 #6  0xc0008cf0 in sys_execve (name=<value optimized out>, argv=0xc3ad14cc, envp=0xc3ad1460) at arch/frv/kernel/process.c:263 #7  0xc00075dc in __syscall_call () at arch/frv/kernel/entry.S:897

I have no idea about the BDI_CAP_MAP_DIRECT bits but...

Quoted 6 lines

Note that this fix does the following commit differently:

       commit a190887b58c32d19c2eee007c5eb8faa970a69ba
       Author: David Howells <[email protected]>
       Date:   Sat Sep 5 11:17:07 2009 -0700
       nommu: fix error handling in do_mmap_pgoff()

...changes to this part of the code look like nice cleanup to me.

maintainer
#2 lore raw
In reply to David Howells
Quoted 10 lines · On Thu, 24 Sep 2009 15:13:10 +0100 David Howells <[email protected]> wrote:

Fix MAP_PRIVATE mmap() of files and devices where the data in the backing store might be mapped directly. Use the BDI_CAP_MAP_DIRECT capability flag to govern whether or not we should be trying to map a file directly. This can be used to determine whether or not a region has been filled in at the point where we call do_mmap_shared() or do_mmap_private().

The BDI_CAP_MAP_DIRECT capability flag is cleared by validate_mmap_request() if there's any reason we can't use it. It's also cleared in do_mmap_pgoff() if f_op->get_unmapped_area() fails.

The patch seems -stable-worthy but I see no cc:stable here?

Quoted 6 lines

Note that this fix does the following commit differently:

	commit a190887b58c32d19c2eee007c5eb8faa970a69ba
	Author: David Howells <[email protected]>
	Date:   Sat Sep 5 11:17:07 2009 -0700
	nommu: fix error handling in do_mmap_pgoff()

I don't understand what you're saying here.

Quoted 6 lines
Cc Pekka Enberg <[email protected]>
Cc Paul Mundt <[email protected]>
Cc Mel Gorman <[email protected]>
Cc Greg Ungerer <[email protected]>
#3 lore raw
Quoted 8 lines · Andrew Morton
Quoted 6 lines

Note that this fix does the following commit differently:

	commit a190887b58c32d19c2eee007c5eb8faa970a69ba
	Author: David Howells <[email protected]>
	Date:   Sat Sep 5 11:17:07 2009 -0700
	nommu: fix error handling in do_mmap_pgoff()

I don't understand what you're saying here.

Sorry, I was attempting to point out that this appears to revert some of that patch - whilst this is true, it then implements the fix differently.

David

maintainer
#4 lore raw
Quoted 9 lines · On Fri, 25 Sep 2009 01:39:17 +0100 David Howells <[email protected]> wrote:
Quoted 7 lines · Andrew Morton
Quoted 6 lines

Note that this fix does the following commit differently:

	commit a190887b58c32d19c2eee007c5eb8faa970a69ba
	Author: David Howells <[email protected]>
	Date:   Sat Sep 5 11:17:07 2009 -0700
	nommu: fix error handling in do_mmap_pgoff()
Quoted 5 lines
Quoted 1 line

I don't understand what you're saying here.

Sorry, I was attempting to point out that this appears to revert some of that patch - whilst this is true, it then implements the fix differently.

OK.

And what are your opinions on the -stable desirability?

#5 lore raw
Quoted 1 line · Andrew Morton

And what are your opinions on the -stable desirability?

I think that it would be a good idea to add it to stable.

David

Acked-by on v1 1/1
#6 lore raw
In reply to David Howells
Quoted 35 lines · David Howells

Fix MAP_PRIVATE mmap() of files and devices where the data in the backing store might be mapped directly. Use the BDI_CAP_MAP_DIRECT capability flag to govern whether or not we should be trying to map a file directly. This can be used to determine whether or not a region has been filled in at the point where we call do_mmap_shared() or do_mmap_private().

The BDI_CAP_MAP_DIRECT capability flag is cleared by validate_mmap_request() if there's any reason we can't use it. It's also cleared in do_mmap_pgoff() if f_op->get_unmapped_area() fails.

Without this fix, attempting to run a program from a RomFS image on a non-mappable MTD partition results in a BUG as the kernel attempts XIP, and this can be caught in gdb:

Program received signal SIGABRT, Aborted.
0xc005dce8 in add_nommu_region (region=<value optimized out>) at mm/nommu.c:547 (gdb) bt

#0  0xc005dce8 in add_nommu_region (region=<value optimized out>) at mm/nommu.c:547
#1  0xc005f168 in do_mmap_pgoff (file=0xc31a6620, addr=<value optimized out>, len=3808, prot=3, flags=6146, pgoff=0) at mm/nommu.c:1373
#2  0xc00a96b8 in elf_fdpic_map_file (params=0xc33fbbec, file=0xc31a6620, mm=0xc31bef60, what=0xc0213144 "executable") at mm.h:1145
#3  0xc00aa8b4 in load_elf_fdpic_binary (bprm=0xc316cb00, regs=<value optimized out>) at fs/binfmt_elf_fdpic.c:343
#4  0xc006b588 in search_binary_handler (bprm=0x6, regs=0xc33fbce0) at fs/exec.c:1234
#5  0xc006c648 in do_execve (filename=<value optimized out>, argv=0xc3ad14cc, envp=0xc3ad1460, regs=0xc33fbce0) at fs/exec.c:1356
#6  0xc0008cf0 in sys_execve (name=<value optimized out>, argv=0xc3ad14cc, envp=0xc3ad1460) at arch/frv/kernel/process.c:263
#7  0xc00075dc in __syscall_call () at arch/frv/kernel/entry.S:897

Note that this fix does the following commit differently:

	commit a190887b58c32d19c2eee007c5eb8faa970a69ba
	Author: David Howells <[email protected]>
	Date:   Sat Sep 5 11:17:07 2009 -0700
	nommu: fix error handling in do_mmap_pgoff()
Quoted 6 lines
Cc Pekka Enberg <[email protected]>
Cc Paul Mundt <[email protected]>
Cc Mel Gorman <[email protected]>
Cc Greg Ungerer <[email protected]>
#7 lore raw
In reply to David Howells
Quoted 71 lines · David Howells

Fix MAP_PRIVATE mmap() of files and devices where the data in the backing store might be mapped directly.  Use the BDI_CAP_MAP_DIRECT capability flag to govern whether or not we should be trying to map a file directly.  This can be used to determine whether or not a region has been filled in at the point where we call do_mmap_shared() or do_mmap_private().

The BDI_CAP_MAP_DIRECT capability flag is cleared by validate_mmap_request() if there's any reason we can't use it.  It's also cleared in do_mmap_pgoff() if f_op->get_unmapped_area() fails.

Without this fix, attempting to run a program from a RomFS image on a non-mappable MTD partition results in a BUG as the kernel attempts XIP, and this can be caught in gdb:

Program received signal SIGABRT, Aborted.
0xc005dce8 in add_nommu_region (region=<value optimized out>) at mm/nommu.c:547
(gdb) bt
#0  0xc005dce8 in add_nommu_region (region=<value optimized out>) at mm/nommu.c:547
#1  0xc005f168 in do_mmap_pgoff (file=0xc31a6620, addr=<value optimized out>, len=3808, prot=3, flags=6146, pgoff=0) at mm/nommu.c:1373 #2  0xc00a96b8 in elf_fdpic_map_file (params=0xc33fbbec, file=0xc31a6620, mm=0xc31bef60, what=0xc0213144 "executable") at mm.h:1145 #3  0xc00aa8b4 in load_elf_fdpic_binary (bprm=0xc316cb00, regs=<value optimized out>) at fs/binfmt_elf_fdpic.c:343 #4  0xc006b588 in search_binary_handler (bprm=0x6, regs=0xc33fbce0) at fs/exec.c:1234
#5  0xc006c648 in do_execve (filename=<value optimized out>, argv=0xc3ad14cc, envp=0xc3ad1460, regs=0xc33fbce0) at fs/exec.c:1356 #6  0xc0008cf0 in sys_execve (name=<value optimized out>, argv=0xc3ad14cc, envp=0xc3ad1460) at arch/frv/kernel/process.c:263 #7  0xc00075dc in __syscall_call () at arch/frv/kernel/entry.S:897

Note that this fix does the following commit differently:

       commit a190887b58c32d19c2eee007c5eb8faa970a69ba
       Author: David Howells <[email protected]>
       Date:   Sat Sep 5 11:17:07 2009 -0700
       nommu: fix error handling in do_mmap_pgoff()

Cc Pekka Enberg <[email protected]>
Cc Paul Mundt <[email protected]>
Cc Mel Gorman <[email protected]>
Cc Greg Ungerer <[email protected]>

---

 mm/nommu.c |   34 ++++++++++++----------------------

 1 files changed, 12 insertions(+), 22 deletions(-)
mm/nommu.c +3 -2
index c459aec..cc24d9f 100644
@@ -1074,7 +1074,7 @@ static int do_mmap_shared_file(struct vm_area_struct *vma)
        ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
        if (ret == 0) {
                vma->vm_region->vm_top = vma->vm_region->vm_end;
-               return ret;
+               return 0;
        }
        if (ret != -ENOSYS)
                return ret;
@@ -1091,7 +1091,8 @@ static int do_mmap_shared_file(struct vm_area_struct *vma)
  */
  static int do_mmap_private(struct vm_area_struct *vma,
                           struct vm_region *region,
-                          unsigned long len)
+                          unsigned long len,
+                          unsigned long capabilities)
  {
        struct page *pages;
        unsigned long total, point, n, rlen;

mm/nommu.c static int do_mmap_private(struct vm_area_struct *vma, · line 1106

Quoted 5 lines
mm/nommu.c +1 -1
@@ -1102,13 +1103,13 @@ static int do_mmap_private(struct vm_area_struct *vma,
         * shared mappings on devices or memory
         * - VM_MAYSHARE will be set if it may attempt to share
         */
-       if (vma->vm_file) {
+       if (capabilities & BDI_CAP_MAP_DIRECT) {

This will breaks many drivers, e.g. some frame-buffer drivers, on NOMMU system. Because they don't have get_unmapped_area().
These drivers depend on it's mmap() to return the frame-buffer base address.

-Graff

Quoted 63 lines
mm/nommu.c +5 -4
                ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
                if (ret == 0) {
                        /* shouldn't return success if we're not sharing */
                        BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
                        vma->vm_region->vm_top = vma->vm_region->vm_end;
-                       return ret;
+                       return 0;
                }
                if (ret != -ENOSYS)
                        return ret;
@@ -1346,7 +1347,7 @@ unsigned long do_mmap_pgoff(struct file *file,
                 * - this is the hook for quasi-memory character devices to
                 *   tell us the location of a shared mapping
                 */
-               if (file && file->f_op->get_unmapped_area) {
+               if (capabilities & BDI_CAP_MAP_DIRECT) {
                        addr = file->f_op->get_unmapped_area(file, addr, len,
                                                             pgoff, flags);
                        if (IS_ERR((void *) addr)) {
@@ -1370,15 +1371,17 @@ unsigned long do_mmap_pgoff(struct file *file,
        }
 
        vma->vm_region = region;
-       add_nommu_region(region);
 
-       /* set up the mapping */
+       /* set up the mapping
+        * - the region is filled in if BDI_CAP_MAP_DIRECT is still set
+        */
        if (file && vma->vm_flags & VM_SHARED)
                ret = do_mmap_shared_file(vma);

       else
-               ret = do_mmap_private(vma, region, len);
+               ret = do_mmap_private(vma, region, len, capabilities); if (ret < 0)
-               goto error_put_region;
+               goto error_just_free;
+       add_nommu_region(region);

       /* okay... we have a mapping; now we have to register it */ result = vma->vm_start;

mm/nommu.c +0 -13
@@ -1396,19 +1399,6 @@ share:
        kleave(" = %lx", result);
        return result;
 
-error_put_region:
-       __put_nommu_region(region);
-       if (vma) {
-               if (vma->vm_file) {
-                       fput(vma->vm_file);
-                       if (vma->vm_flags & VM_EXECUTABLE)
-                               removed_exe_file_vma(vma->vm_mm);
-               }
-               kmem_cache_free(vm_area_cachep, vma);
-       }
-       kleave(" = %d [pr]", ret);
-       return ret;
-

 error_just_free:
       up_write(&nommu_region_sem);
 error:

Signature · 1 line
-Graff
#8 lore raw
Quoted 5 lines · graff yang
Quoted 1 line

+       if (capabilities & BDI_CAP_MAP_DIRECT) {

This will breaks many drivers, e.g. some frame-buffer drivers, on NOMMU system. Because they don't have get_unmapped_area(). These drivers depend on it's mmap() to return the frame-buffer base address.

Then they won't work on NOMMU. Read Documentation/nommu-mmap.txt:

============================================
PROVIDING SHAREABLE CHARACTER DEVICE SUPPORT
============================================

To provide shareable character device support, a driver must provide a file->f_op->get_unmapped_area() operation. The mmap() routines will call this to get a proposed address for the mapping. This may return an error if it doesn't wish to honour the mapping because it's too long, at a weird offset, under some unsupported combination of flags or whatever.

The chardev driver doesn't provide the virtual address through its mmap() fop in MMU-mode either - that's provided by do_mmap_pgoff().

David

#9 lore raw
Quoted 21 lines · David Howells
Quoted 5 lines · graff yang
Quoted 1 line

+       if (capabilities & BDI_CAP_MAP_DIRECT) {

This will breaks many drivers, e.g. some frame-buffer drivers, on NOMMU system.  Because they don't have get_unmapped_area().  These drivers depend on it's mmap() to return the frame-buffer base address.

Then they won't work on NOMMU.  Read Documentation/nommu-mmap.txt:

       ============================================
       PROVIDING SHAREABLE CHARACTER DEVICE SUPPORT
       ============================================

       To provide shareable character device support, a driver must provide a file->f_op->get_unmapped_area() operation. The mmap() routines will call this to get a proposed address for the mapping. This may return an error if it doesn't wish to honour the mapping because it's too long, at a weird offset, under some unsupported combination of flags or whatever.

Quoted 5 lines

The chardev driver doesn't provide the virtual address through its mmap() fop in MMU-mode either - that's provided by do_mmap_pgoff().

David

OK, I think do_mmap_pgoff() now looks nice with your patch. Merely,
at least the alsa/fb driver
will crash due to the vm_region changes.
The alsa guys are fixing this issue:
http://mailman.alsa-project.org/pipermail/alsa-devel/2009-March/015802.html

Is there any other things be affected?

Signature · 1 line
-Graff